Professional Documents
Culture Documents
Handbook of X-Ray
Handbook of X-Ray
Handbook of X-Ray
X-Ray Spectrometry
Second Edition, Revised and Expanded
edited by
René E. Van Grieken
University of Antwerp
Antwerp, Belgium
Andrzej A. Markowicz
Vienna, Austria
First edition was published as Handbook of X-Ray Spectrometry: Methods and Techniques
Headquarters
Marcel Dekker, Inc.
270 Madison Avenue, New York, NY 10016
tel: 212-696-9000; fax: 212-685-4540
The publisher offers discounts on this book when ordered in bulk quantities. For more information,
write to Special Sales=Professional Marketing at the headquarters address above.
Neither this book nor any part may be reproduced or transmitted in any form or by any means,
electronic or mechanical, including photocopying, microfilming, and recording, or by any in-
formation storage and retrieval system, without permission in writing from the publisher.
The positive response to the first edition of Handbook of X-Ray Spectrometry: Methods
and Techniques and its commercial success have shown that in the early 1990s there was
a clear demand for an exhaustive book covering most of the specialized approaches in this
field. Therefore, some five years after the first edition appeared, the idea of publishing
a second edition emerged. In the meantime, remarkable and steady progress has been
made in both instrumental and methodological aspects of x-ray spectrometry. This
progress includes considerable improvements in the design and production technology of
detectors and in capillary optics applied for focusing the primary photon beam. The
advances in instrumentation, spectrum evaluation, and quantification have resulted in
improved analytical performance and in further extensions of the applicability range of x-
ray spectrometry. Consequently, most of the authors who contributed to the first edition
of this book enthusiastically accepted the invitation to update their chapters. The progress
made during the last decade is reflected well in the chapters of the second edition, which
were all considerably revised, updated, and expanded. A completely new chapter on mi-
crobeam x-ray fluorescence analysis has also been included.
Chapter 1 reviews the basic physics behind x-ray emission techniques, and refers to
extensive appendices for all the basic and generally applicable x-ray physics constants.
New analytical expressions have been introduced for the calculation of fundamental
parameters such as the fluorescence yield, incoherent scattering function, atomic form
factor, and total mass attenuation coefficient.
Chapter 2 outlines established and new instrumentation and discusses the perfor-
mances of wavelength-dispersive x-ray fluorescence (XRF) analysis, which, with probably
15,000 units in operation worldwide today, is still the workhorse of x-ray analysis. Its
applications include process control, materials analysis, metallurgy, mining, and almost
every other major branch of science. The additional material in this edition covers new
sources of excitation and comprehensive comparisons of the technical parameters of newly
produced wavelength-dispersive spectrometers.
Chapter 3 has been completely reconsidered, modified, and rewritten by a new au-
thor. The basic principles, background, and recent advances are described for the tube-
excited energy-dispersive mode, which is invoked so frequently in research on environ-
mental and biological samples. This chapter is based on a fresh look and follows
a completely different approach.
Scientists in recent years have been somewhat ambivalent regarding the role of x-ray
emission spectrometry in analytical chemistry. Whereas no radically new and stunning
developments have been seen, there has been remarkably steady progress, both instru-
mental and methodological, in the more conventional realms of x-ray fluorescence. For the
more specialized approaches—for example, x-ray emission induced by synchrotron ra-
diation, radioisotopes and polarized x-ray beams, and total-reflection x-ray fluorescence—
and for advanced spectrum analysis methods, exponential growth and=or increasing ac-
ceptance has occurred. Contrary to previous books on x-ray emission analysis, these latter
approaches make up a large portion of the present Handbook of X-Ray Spectrometry.
The major milestone developments that shaped the field of x-ray spectrometry and
now have widespread applications all took place more than twenty years ago. After
wavelength-dispersive x-ray spectrometry had been demonstrated and a high-vacuum
x-ray tube had been introduced by Coolidge in 1913, the prototype of the first modern
commercial x-ray spectrometer with a sealed x-ray tube was built by Friedmann and Birks
in 1948. The first electron microprobe was successfully developed in 1951 by Castaing,
who also outlined the fundamental concepts of quantitative analysis with it. The semi-
conductor or Si(Li) detector, which heralded the advent of energy-dispersive x-ray
fluorescence, was developed around 1965 at Lawrence Berkeley Laboratory. Accelerator-
based particle-induced x-ray emission analysis was developed just before 1970, mostly at
the University of Lund. The various popular matrix correction methods by Lucas-Tooth,
Traill and Lachance, Claisse and Quintin, Tertian, and several others, were all proposed in
the 1960s. One may thus wonder whether the more conventional types of x-ray fluores-
cence analysis have reached a state of saturation and consolidation, typical for a mature
and routinely applied analysis technique. Reviewing the state of the art and describing
recent progress for wavelength- and energy-dispersive x-ray fluorescence, electron and
heavy charged-particle-induced x-ray emission, quantification, and sample preparation
methods is the purpose of the remaining part of this book.
Chapter 1 reviews the basic physics behind the x-ray emission techniques, and refers
to the appendixes for all the basic and generally applicable x-ray physics constants.
Chapter 2 outlines established and new instrumentation and discusses the performances of
wavelength-dispersive x-ray fluorescence analysis, which, with probably 14,000 units in
operation worldwide today, is still the workhorse of x-ray analysis with applications in
a wide range of disciplines including process control, materials analysis, metallurgy,
1 X-ray Physics
Andrzej A. Markowicz
I. Introduction
II. History
III. General Features
IV. Emission of Continuous Radiation
V. Emission of Characteristic X-rays
VI. Interaction of Photons with Matter
VII. Intensity of Characteristic X-rays
VIII. IUPAC Notation for X-ray Spectroscopy
Appendixes
I. Critical Absorption Wavelengths and Critical Absorption Energies
II. Characteristic X-ray Wavelengths (Å) and Energies (keV)
III. Radiative Transition Probabilities
IV. Natural Widths of K and L Levels and Ka X-ray Lines (FWHM), in eV
V. Wavelengths of K Satellite Lines (Å) 56
VI. Fluorescence Yields and Coster–Kronig Transition Probabilities
VII. Coefficients for Calculating the Photoelectric Absorption
Cross Sections t (Barns=Atom) Via ln–ln Representation
VIII. Coefficients for Calculating the Incoherent Collision
Cross Sections sc (Barns=Atom) Via the ln–ln Representation
IX. Coefficients for Calculating the Coherent Scattering
Cross Sections sR (Barns=Atom) Via the ln–ln Representation
X. Parameters for Calculating the Total Mass Attenuation
Coefficients in the Energy Range 0.1–1000 keV [Via Eq. (78)]
XI. Total Mass Attenuation Coefficients for Low-Energy Ka Lines
XII. Correspondence Between Old Siegbahn and New IUPAC
Notation X-ray Diagram Lines
References
4 Spectrum Evaluation
Piet Van Espen
I. Introduction
II. Fundamental Aspects
III. Spectrum Processing Methods
IV. Continuum Estimation Methods
V. Simple Net Peak Area Determination
VI. Least-Squares Fitting Using Reference Spectra
VII. Least-Squares Fitting Using Analytical Functions
VIII. Methods Based on the Monte Carlo Technique
IX. The Least-Squares-Fitting Method
X. Computer Implementation of Various Algorithms
References
11 Microbeam XRF
Anders Rindby and Koen H. A. Janssens
I. Introduction and Historical Perspective
II. Theoretical Background
III. Instrumentation for Microbeam XRF
IV. Collection and Processing of m-XRF Data
V. Applications
References
Joachim Heckel, Ph.D. Spectro Analytical Instruments, GmbH & Co. KG, Kleve, Ger-
many
Klas G. Malmqvist, Ph.D. Lund University and Lund Institute of Technology, Lund,
Sweden
Andrzej A. Markowicz
Vienna, Austria
I. INTRODUCTION
In this introductory chapter, the basic concepts and processes of x-ray physics that relate
to x-ray spectrometry are presented. Special emphasis is on the emission of the continuum
and characteristic x-rays as well as on the interactions of photons with matter. In the
latter, only major processes of the interactions are covered in detail, and the cross sections
for different types of interactions and the fundamental parameters for other processes
involved in the emission of the characteristic x-rays are given by the analytical expressions
and=or in a tabulated form. Basic equations for the intensity of the characteristic x-rays
for the different modes of x-ray spectrometry are also presented (without derivation).
Detailed expressions relating the emitted intensity of the characteristic x-rays to the
concentration of the element in the specimen are discussed in the subsequent chapters of
this handbook dedicated to specific modes of x-ray spectrometry.
II. HISTORY
Continuous x-rays are produced when electrons, or other high-energy charged particles,
such as protons or a-particles, lose energy in passing through the Coulomb field of a
nucleus. In this interaction, the radiant energy (photons) lost by the electron is called
bremsstrahlung (from the German bremsen, to brake, and Strahlung, radiation; this term
sometimes designates the interaction itself). The emission of continuous x-rays finds a
simple explanation in terms of classic electromagnetic theory, because, according to this
theory, the acceleration of charged particles should be accompanied by the emission of
radiation. In the case of high-energy electrons striking a target, they must be rapidly
decelerated as they penetrate the material of the target, and such a high negative accel-
eration should produce a pulse of radiation.
The continuous x-ray spectrum generated by electrons in an x-ray tube is char-
acterized by a short-wavelength limit lmin , corresponding to the maximum energy of the
exciting electrons:
hc
lmin ¼ ð1Þ
eV0
where h is Planck’s constant, c is the velocity of light, e is the electron charge, and V0 is the
potential difference applied to the tube. This relation of the short-wavelength limit to the
applied potential is called the Duane–Hunt law.
The probability of radiative energy loss (bremsstrahlung) is roughly proportional to
q2 Z2 T=M20 , where q is the particle charge in units of the electron charge e, Z is the atomic
number of the target material, T is the particle kinetic energy, and M0 is the rest mass of
the particle. Because protons and heavier particles have large masses compared to the
electron mass, they radiate relatively little; for example, the intensity of continuous x-rays
generated by protons is about four orders of magnitude lower than that generated by
electrons.
The ratio of energy lost by bremsstrahlung to that lost by ionization can be
approximated by
2
m0 ZT
ð2Þ
M0 1600m0 c2
where m0 the rest mass of the electron.
A. Spectral Distribution
The continuous x-ray spectrum generated by electrons in an x-ray tube (thick target) is
characterized by the following features:
1. Short-wavelength limit, lmin [Eq. (1)]; below this wavelength, no radiation is
observed.
Copyright © 2002 Marcel Dekker, Inc.
2. Wavelength of maximum intensity lmax , approximately 1.5 times lmin ; however,
the relationship between lmax and lmin depends to some extent on voltage,
voltage waveform, and atomic number.
3. Total intensity nearly proportional to the square of the voltage and the first
power of the atomic number of the target material.
The most complete empirical work on the overall shape of the energy distribution
curve for a thick target has been of Kulenkampff (1922, 1933), who found the following
formula for the energy distribution;
IðvÞ dv ¼ i aZðv0 vÞ þ bZ2 dv ð3Þ
where IðnÞ dn is the intensity of the continuous x-rays within a frequency range
ðn; n þ dvÞ; i is the electron current striking the target, Z is the atomic number of the
target material, n0 is the cutoff frequency ð¼ c=lmin Þ above which the intensity is zero, and
a and b are constants independent of atomic number, voltage, and cutoff wavelength. The
second term in Eq. (3) is usually small compared to the first and is often neglected.
The total integrated intensity at all frequencies is
pffiffi AZ e dv;
2 2 5
IðvÞ dv ¼ 316p v < v0
3m0 V0 c3 ð7Þ
IðvÞ dv ¼ 0; v > v0
where A is the atomic mass of the target material. When the decrease in velocity of the
electrons in a thick target was taken into account by applying the Thomson–Whiddington
law (Dyson, 1973), Kramers found, for a thick target,
8pe2 h
IðvÞ dv ¼ pffiffiffi Zðv0 vÞ dv ð8Þ
3 3lm0 c3
where l is approximately 6. The efficiency of production of the x-rays calculated via
Kramers’ law is given by
50 for 37.8 kV to 65 for 16.4 kV. Figure 2 illustrates the intensity of the continuous
x-rays observed in the Al foil for different thicknesses as a function of the angle for a
voltage of 30 kV (Stephenson, 1957). The theoretical curve is from the theory of Scherzer
(1932). The continuous x-ray intensity drops to zero at 180 , and although it is not zero at
0 as the theory of Scherzer predicts, it can be seen from Figure 2 that for a thinner foil, a
lower intensity at 0 is obtained. Summarizing, it appears that the intensity of the con-
tinuous x-rays emitted by thin foils has a maximum at about 55 relative to the incident
electron beam and becomes zero at 180 .
The continuous radiation from thick targets is characterized by a much smaller
anisotropy than that from thin targets. This is because in thick targets the electrons are
rarely stopped in one collision and usually their directions have considerable variation.
The use of electromagnetic theory predicts a maximum energy at right angles to the in-
cident electron beam at low voltages, with the maximum moving slightly away from
perpendicularity toward the direction of the elctron beam as the voltage is increased. In
general, an increase in the anisotropy of the continuous x-rays from thick targets is ob-
served at the short-wavelength limit and for low-Z targets (Dyson, 1973).
Copyright © 2002 Marcel Dekker, Inc.
Figure 2 Intensity of continuous x-rays as a function of direction for different thicknesses of the
A1 target together with theoretical prediction. (From Stephenson, 1957.)
Continuous x-ray beams are partially polarized only from extremely thin targets; the
angular region of polarization is sharply peaked about the photon emission angle
y ¼ m0 c2 =E0 , where E0 is the energy of the primary electron beam. Electron scattering in
the target broadens the peak and shifts the maximum to larger angles. Polarization is
defined by (Kenney, 1966)
ds?ðy; E0 ; En Þ dskðy; E0 ; En Þ
Pðy; E0 ; En Þ ¼ ð11Þ
ds?ðy; E0 ; En Þ þ dskðy; E0 ; En Þ
where an electron of energy E0 radiates a photon of energy En at angle y; ds?ðy; E0 ; En Þ
and dskðy; E0 ; En Þ are the cross sections for generation of the continuous radiation with
the electric vector perpendicular (?) and parallel (k) to the plane defined by the incident
electron and the radiated photon, respectively. Polarization is difficult to observe, and only
thin, low-yield radiators give evidence for this effect. When the electron is relativistic
before and after the radiation, the electrical vector is most probably in the ? direction.
Practical thick-target Bremsstrahlung shows no polarization effects whatever (Dyson,
1973; Stephenson, 1957; Kenney, 1966).
ln U
Qi ¼ pe4 ns bs ð12Þ
UE2c;i
where U ¼ E=Ec;i is the overvoltage, defined as the ratio of the instantaneous energy of the
electron at each point of the trajectory to that required to ionize an atom of element i, Ec;i
is the critical excitation energy, and ns and bs are constants for a particular shell:
s ¼ K: ns ¼ 2; bs ¼ 0:35
s ¼ L: ns ¼ 8; bs ¼ 0:25
The cross section for ionization Qi is a strong function of the overvoltage, which shows a
maximum at U ffi 3–4 (Heinrich, 1981; Goldstein et al., 1981).
The probability (or cross section) of ionization of an inner atomic shell by a charged
particle is given by (Merzbacher and Lewis, 1958)
8pr20 q2 fs
ss ¼ ð13Þ
Z 4 Zs
where r0 is the classic radius of the electron equal to 2.81861015 m, q is the particle
charge, Z is the atomic number of the target material, fs is a factor depending on the wave
functions of the electrons for a particular shell, and Zs is a function of the energy of the
incident particles.
In the case of electromagnetic radiation (x or g), the ionization of an inner atomic
shell is a result of the photoelectric effect. This effect involves the disappearance of a ra-
diation photon and the photoelectric ejection of one electron from the absorbing atom,
leaving the atom in an excited level. The kinetic energy of the ejected photoelectron is
given by the difference between the photon energy hn and the atomic binding energy of the
electron Ec (critical excitation energy). Critical absorption wavelengths (Clark, 1963) re-
lated to the critical absorption energies (Burr, 1974) via the equation l(nm) ¼ 1.24=E(ke V)
are presented in Appendix I. The wavelenghts of K, L, M, and N absorption edges can also
be calculated by using simple empirical equations (Norrish and Tao, 1993).
For energies far from the absorption edge and in the nonrelativistic range, the cross
section tK for the ejection of an electron from the K shell is given by (Heitler, 1954)
pffiffiffi 7=2
32 2 2 Z5 m 0 c2
tK ¼ pr0 ð14Þ
3 ð137Þ4 hv
Equation (14) is not fully adequate in the neighborhood of an absorption edge; in this
case, Eq. (14) should be multiplied by a correction factor f(X ) (Stobbe, 1930):
1=2 4X arccot X
D e
fðXÞ ¼ 2p ð15Þ
hv 1 e2pX
where
Copyright © 2002 Marcel Dekker, Inc.
1=2
D
X¼ ð15aÞ
hv D
with
1 m 0 c2
D ffi ðZ 0:3Þ2 ð15bÞ
2 ð137Þ2
When the energy of the incident photon is of the order m0 c2 or greater, relativistic
cross sections for the photoelectric effect must be used (Sauter, 1931).
Figure 3 Commonly used terminology of energy levels and x-ray lines. (From Sandström, 1957.)
IðKa1 þ Ka2 Þ
pKa ¼ ð22Þ
IðKa1 þ Ka2 Þ þ IðKb1 þ Kb2 Þ
Wernisch (1985) proposed a useful expression for the calculation of the transition prob-
ability pKa for different elements:
8
< 1:052 4:39 104 Z2i ; 11 Zi 19
pKa;i ¼ 0:896 6:575 104 Zi ; 20 Zi 29 ð23Þ
:
1:0366 6:82 103 Zi þ 4:815 105 Z2i ; 30 Z2i 60
For the L level, split into three subshells, several electron transitions exist. The transition
probability pLa, defined as the fraction of the transitions resulting in La1 and La2 radiation
from the total of possible transitions into the L3 subshell, can be calculated by the ex-
pression (Wernisch, 1985)
8
< 0:944; 39 Zi 44
pLa;i ¼ 4:461 101 þ 5:493 102 Zi ð24Þ
:
7:717 104 Z2i þ 3:525 106 Z3i ; 45 Zi 82
Radiative transition probabilities for various K and L x-ray lines (West, 1982–83) are
presented in detail in Appendix III. The experimental results, together with the estimated
95% confidence limits, for the relative intensity ratios for K and L x-ray lines for selected
elements with atomic number from Z ¼ 14 to 92 have been reported by Stoev and Dlouhy
(1994). The values are in a good agreement with other published experimental data.
Because the electron vacancy created in an atom by charged particles or electro-
magnetic radiation has a certain lifetime, the atomic levels E1 and E2 [Eq. (16)] and the
characteristic x-ray lines are characterized by the so-called natural widths (Krause and
Oliver, 1979). The natural x-ray linewidths are the sums of the widths of the atomic levels
involved in the transitions. Semiempirical values of the natural widths of K, L1 , L2 and L3
levels, Ka1 and Ka2 x-ray lines for the elements 10 Z 110 are presented in Appendix
IV. Uncertainties in the level width values are from 3% to 10% for the K shell and from
8% to 30% for the L subshell. Uncertainties in the natural x-ray linewidth values are from
3% to 10% for Ka1;2 . In both cases, the largest uncertainties are for low-Z elements
(Krause and Oliver, 1979).
C. X-ray Satellites
A large number of x-ray lines have been reported that do not fit into the normal energy-
level diagram (Clark, 1955; Kawai and Gohshi, 1986). Most of the x-ray lines, called
satellites or nondiagram lines, are very weak and are of rather little consequence in ana-
lytical x-ray spectrometry. By analogy to the satellites in optical spectra, it was supposed
that the origin of the nondiagram x-ray lines is double or manyfold ionization of an atom
through electron impact. Following the ionization, a multiple electron transition results in
emission of a single photon of energy higher than that of the characteristic x-rays. The
majority of nondiagram lines originate from the dipole-allowed deexcitation of multiply
Copyright © 2002 Marcel Dekker, Inc.
ionized or excited states and are called multiple-ionization satellites. A line where the
initial state has two vacancies in the same shell, notably the K shell, is called a hypersa-
tellite. In practice, the most important nondiagram x-ray lines occur in the Ka series; they
are denoted as the Ka3 ;a4 doublet, and their origin is a double electron transition. The
probability of a multiple-electron transition resulting in the emission of satellite x-ray lines
is considerably higher for low-Z elements than for heavy and medium elements. For
instance, the intensity of the AlKa3 satellite line is roughly 10% of that of the AlKa1 ; a2
characteristic x-rays.
Appendix V tabulates wavelengths of the K satellite lines. A new class of satellites
that are inside the natural width of the parent lines was observed by Kawai and Gohshi
(1986). The origin of these satellites, called parasites or hidden satellites, is multiple
ionization in nonadjacent shells.
E. Auger Effect
It has already been stated that the excess of energy an atom possesses after removing one
electron from an inner shell by whatever means may be emitted as characteristic radiation.
Alternatively, however, an excited atom may return to a state of lower energy by ejecting
one of its own electrons from a less tightly bound state. The radiationless transition is
called the Auger effect, and the ejected electrons are called Auger electrons (Auger, 1925;
Burhop, 1952). Generally, the probability of the Auger effect increases with a decrease in
the difference of the corresponding energy states, and it is the highest for the low-Z
elements.
Because an excited atom already has one electron missing (e.g., in the K shell) and
another electron is ejected in an Auger process (e.g., from the L shell), the atom is left in a
doubly-ionized state in which two electrons are missing. This atom may return to its
normal state by single- or double-electron jumps with the emission of diagram or satellite
lines, respectively. Alternatively, another Auger process may occur in which a third
electron is ejected from the M shell.
The Auger effect also occurs after capture of a negative meson by an atom. As the
meson changes energy levels in approaching the nucleus, the energy released may be either
emitted as a photon or transferred directly to an electron that is emitted as a high-energy
Auger electron (in the keV range for hydrogen and the MeV range for heavy elements).
Measurements of the energy and intensity of the Auger electrons are applied ex-
tensively in surface physics studies (Auger electron spectroscopy).
F. FluorescenceYield
An important consequence of the Auger effect is that the actual number of x-ray photons
produced from an atom is less than expected, because a vacancy in a given shell might be
Copyright © 2002 Marcel Dekker, Inc.
filled through a nonradiative transition. The probability that a vacancy in an atomic shell
or subshell is filled through a radiative transition is called the fluorescence yield. The
application of this definition to the K shell of an atom is straightforward, and the fluor-
escence yield of the K shell is
IK
oK ¼ ð25Þ
nK
where IK is the total number of characteristic K x-ray photons emitted from a sample and
nK is the number of primary K shell vacancies.
The definition of the fluorescence yield of higher atomic shells is more complicated,
for the following two reasons:
1. Shells above the K shell consist of more than one subshell; the average
fluorescence yield depends on how the shells are ionized.
2. Coster–Kronig transitions occur, which are nonradiative transitions between the
subshells of an atomic shell having the same principal quantum number (Fink,
1974; Bambynek et al., 1972).
In case Coster–Kronig transitions are absent, the fluorescence yield of the ith sub-
shell of a shell, whose principal quantum number is indicated by XðX ¼ L; M; . . .Þ, is
given as
IX
i ¼
oX ð26Þ
i
nX
i
X
k
X ¼
o NX X
i oi ð27Þ
i¼1
where NX
i is the relative number of primary vacancies in the subshell i of shell X:
nX X
k
i ¼ Pk
NX ; i ¼1
NX ð28Þ
i
X
i¼1 ni i¼1
The summations in Eqs. (27) and (28) extend over all k subshells of shell X. For the de-
finition of the average fluorescence yield, the primary vacancy distribution must be fixed;
X generally is not
that is, Coster–Kronig transitions must be absent. It is noteworthy that o
a fundamental property of the atom, but depends both on the atomic subshell fluorescence
yields oX X
i and on the relative number of primary vacancies Ni characteristic of the method
used to ionize the atoms.
In the presence of Coster–Kronig transitions, which modify the primary vacancy
distribution by the transfer of ionization from one subshell with a given energy to a
subshell with less energy, the average fluorescence yields can be calculated by using two
alternative approaches. In the first, the average fluorescence yield o X is regarded as a
linear combination of the subshell fluorescence yields oX i with a vacancy distribution
modified by Coster–Kronig transitions:
X
k X
k
X ¼
o i oi ;
VX X
i >1
VX ð29Þ
i¼1 i¼1
1 ¼ N1
VX X
2 ¼ N2 þ f 12 N1
VX X X X
ð30Þ
VX
3 ¼ NX
3 þ fX X
23 N2 þ ðfX
13 þ fX
12 23 ÞN1
fX X
X is a linear combination of
In an alternative approach, the mean fluorescence yield o
the relative numbers of primary vacancies NX i :
X
k
X ¼
o NX X
i ni ð31Þ
i¼1
where nX i represents the total number of characteristic x-rays that result per primary
vacancy in the Xi subshell. The transformation relations between the coefficients nX i and
the subshell fluorescence yields oX i follow from Eqs. (29) through (31) and are given in
Fink (1974) and Bambynek et al. (1972).
Among the fluorescence yield oX X
i , the Auger yield ai , and the Coster–Kronig
X
transition probabilities fij , the following relationship must hold (Krause, 1979):
X
k
i þ ai þ
oX ij ¼ 1 ð32Þ
X
fX
i¼1
X
k
ax ¼ VX X
i ai ð33Þ
i¼1
The values of the K, L, and M shell fluorescence yields, the Coster–Kronig transition
probabilities, as well as the Auger yields are given in Appendix VI.
Although, in principle, the K shell fluorescence yield ok can be calculated theoreti-
cally, experimental data are applied in practice. The following semiempirical equation, due
to Burhop (1952), gives values correct to a few percent between Z ¼ 23 and Z ¼ 57 and
less accurate values outside these limits:
1=4
oK
¼ 6:4 102 þ 3:40 102 Z 1:03 106 Z3 ð34Þ
1 oK
The fluorescence yield for the K series can also be calculated from a different equation:
1=4
oK
¼ 0:217 þ 0:03318Z 1:14 106 Z3 ð35Þ
1 oK
which gives quite good agreement with the experimental values for almost all elements.
Based on a critical review of the available experimental and theoretical data, a recom-
mended set of oK ; o L ; and o
M values together with the following analytical expressions
were given by Hubbell (1989) and Hubbell et al. (1994):
Copyright © 2002 Marcel Dekker, Inc.
" #4 8 " #4 91
X
3 < X
3 =
oK ð1 Z 100Þ ¼ Ci Zi 1þ Ci Zi ð36Þ
i¼0
: i¼0
;
with
C0 ¼ 0:0370 0:0052
C1 ¼ 0:03112 0:00044
C2 ¼ ð5:44 0:11Þ 105
C3 ¼ ð1:25 0:07Þ 106
with
C0 ¼ 0:17765
C1 ¼ 0:00298937
C2 ¼ 8:91297 105
C3 ¼ 2:67184 107
For the N shell, the best source of o N data is a theoretical work of McGuire (1974)
which provides oN1 ; oN2 , and oN3 values for 25 elements over the range 38 Z 103
and oN4 ; oN5 , and oN6;7 values for 20 elements over the range 50 Z 103.
The average fluorescence yield o N can be calculated from (Hubbell, 1989)
X7
1
N ¼
o NN oN ð41Þ
i¼1
32 i i
Interactions of photons with matter, by which individual photons are removed or deflected
from a primary beam of x or g radiation, may be classified according to the following:
The kind of target, such as electrons, atoms or nuclei, with which the photon
interacts
The type of event, such as absorption, scattering, or pair production, that takes
place
These interactions are thought to be independent of the origin of the photon (nuclear
transition for g-rays versus electronic transition for x-rays); hence, we use the term
‘‘photon’’ to refer to both g- and x-rays here.
Possible interactions are summarized in Table 1 (Hubbell, 1969), where t is the total
photoelectric absorption cross section per atom (t ¼ tK þ tL þ ) and sR and sC are
Rayleigh and Compton collision cross sections, respectively.
The probability of each of these many competing independent processes can be
expressed as a collision cross section per atom, per electron, or per nucleus in the absorber.
The sum of all these cross sections, normalized to a per atom basis, is then the probability
stot that the incident photon will have an interaction of some kind while passing through a
very thin absorber that contains one atom per square centimeter of area normal to the
path of the incident photon:
stot ¼ t þ sR þ sC þ ð42Þ
The total collision cross section per atom stot , when multiplied by the number of
atoms per cubic centimeter of absorber, is then the linear attenuation coefficient m per
centimeter of travel in the absorber:
1 cm2 g N0 atoms
m ¼ stot r ð43Þ
cm atom cm3 A g
where r is the density of the medium and N0 is Avogadro’s number (6.0225261023
atoms=g atom). The mass attenuation coefficient m (cm2=g) is the ratio of the linear at-
tenuation coefficient and the density of the material.
It is worth mentioning that the absorption coefficient is a much more restricted
concept than the attenuation coefficient. Attenuation includes the purely elastic process in
which the photon is merely deflected and does not give up any of its initial energy to the
absorber; in this process, only a scattering coefficient is involved. In a photoelectric in-
teraction, the entire energy of the incident photon is absorbed by an atom of the medium.
Copyright © 2002 Marcel Dekker, Inc.
Table 1 Classification of Photon Interactions
Scattering
Interaction with atomic Photoelectric effecta Rayleigh scatteringa Compton scatteringa Two-photon Compton
electrons Z4 low energy sR Z2 sC Z scattering
t Z
Z5 high energy
Interaction with nucleus Nuclear photoelectric effect: Nuclear coherent Nuclear Compton
or bound nucleons reactions (g, n) (g, p), scattering scattering
photofission (g, g) (g, g0 )
Z Z2 Z
(E
10 MeV)
Interaction with electrical field 1. Electron–positron pair production Delbrück scattering
surrounding charged particles in field of nucleous, Z4
Z2
(E
1.02 MeV)
2. Electron–positron pair production
in electron field
Z
(E
2.04 MeV)
3. Nucleon–antinucleon pair
production
(E
3 GeV)
Interactions with mesons Photomeson production Coherent resonant
(E
150 MeV) scattering (g, g)
a
Major effects of photon attenuation in matter, which are of great importance in practical x-ray spectrometry.
Source: From Hubbell, 1969.
where I is the transmitted intensity, I0 is the incident intensity, and t is the absorber
thickness in centimeters.
If the absorber is a chemical compound or a mixture, its mass attenuation coefficient
can be approximately evaluated from the coefficients mi for the constituent elements
according to the weighted average:
X
n
m¼ Wi mi ð45Þ
i¼1
where Wi is the weight fraction of the ith element and n is the total number of the elements
in the absorber. The ‘‘mixture rule’’ [Eq. (45)] ignores changes in the atomic wave function
resulting from changes in the molecular, chemical, or crystalline environment of an atom.
Above 10 keV, errors from this approximation are expected to be less than a few percent
(except in the regions just above absorption edges), but at very low energies (10–100 eV),
errors of a factor of 2 can occur (Deslattes, 1969).
For situations more complicated than the narrow-beam geometry, the attenuation is
still basically exponential, but it is modified by two additional factors. The first of these,
sometimes called a geometry factor, depends on the source absorber geometry. The other
factor, often called the buildup factor, takes into account secondary photons produced in
the absorber, mainly as the result of one or more Compton scatters, which finally reach the
detector. The determination of the buildup factor, defined as the ratio of the observed
effect to the effect produced only by the primary radiation, constitutes a large part of g-ray
transport theory (Evans, 1963).
In subsequent sections, only major effects of photon attenuation are discussed
in detail.
A. Photoelectric Absorption
In the photoelectric absorption described partially in Sec. V.A, a photon disappears and
an electron is ejected from an atom. The K shell electrons, which are the most tightly
bound, are the most important for this effect in the energy region considered in XRS. If the
photon energy drops below the binding energy of a given shell, however, an electron from
that shell cannot be ejected. Hence, a plot of t versus photon energy exhibits the char-
acteristic ‘‘absorption edges.’’
The mass photoelectric absorption coefficient tN0 =A at the incident energy E (keV)
can approximately be calculated based on Walter’s equations (Compton and Allison,
1935):
Copyright © 2002 Marcel Dekker, Inc.
8
> 30:3Z3:94
< AE3 for E > EK
tN0
¼ 0:978Z ð46Þ
4:30
for EL1 < E < EK
A > AE3
: 0:78Z3:94
AE3
for EM1 < E <L3
Based on available experimental and theoretical information for approximately
10,000 combinations of Z and E covering 87 elements and the energy range 1 keV to
1 MeV, the following ln–ln polynomials for the photoeffect cross section tj have been fitted
in incident photon energy between each absorption-edge region (Hubbell et al., 1974):
P3
1;2;or
ln tj ¼ Aij fln½E ðkeVÞgi ð47Þ
i¼0
In this polynomial, the total photoeffect cross section tj represents one of the following
sums:
t1 ¼ tM þ tN þ tO þ or tN þ tO þ ;
EM1 < E < EL3 or E < EM5
ð48Þ
t2 ¼ tL þ tM þ tN þ tO þ ; EL1 < E < EK
t3 ¼ tK þ tL þ tM þ tN þ tO þ ; E > EK
The values of the fitted coefficients Aij for the ln–ln representation are given in Appendix
VII (McMaster et al., 1969). In multiple-edge regions (e.g., between L1 and L3 edge en-
ergies), the photoelectric absorption cross sections are also obtained via Eq. (47) by using
the following constant ‘‘jump ratios’’ j (t just above an absorption edge divided by t just
below that absorption edge):
j ¼ 1:16 for EL2 < E < EL1
j ¼ 1:64 ¼ 1:16 1:41 for EL3 < E < EL2
j ¼ 1:1 for EM2 < E < EM1
ð49Þ
j ¼ 1:21 ¼ 1:1 1:1 for EM3 < E < EM2
j ¼ 1:45 ¼ 1:1 1:1 1:2 for EM4 < E < EM3
j ¼ 2:18 ¼ 1:1 1:1 1:2 1:5 for EM5 < E < EM4
Simple expressions for calculating the values of the energies of all photoabsorption edges
are given in Sec. VI.D.
The experimental ratio of the total photoelectric absorption cross section t to the
K shell component tK can be fitted with an accuracy of 2–3% by the equation (Hubbell,
1969)
t
ffi 1 þ 0:01481 ln2 Z 0:000788 ln3 Z ð50Þ
tK
Based on the tables of McMaster et al. (1969), Poehn et al. (1985) found a useful
approximation for the calculation of the jump ratios (called also jump factors) for the
K shell ( jK ) and L3 subshell ( jL3 ):
B. Compton Scattering
Compton scattering (Compton, 1923a, 1923b) is the interaction of a photon with a free
electron that is considered to be at rest. The weak binding of electrons to atoms may be
neglected, provided the momentum transferred to the electron greatly exceeds the mo-
mentum of the electron in the bound state. Considering the conservation of momentum
and energy leads to the following equations:
hn0
hn ¼ ð52Þ
1 þ gð1 cos yÞ
gð1 cos yÞ
T ¼ hn0 hn ¼ hn0 ð53Þ
1 þ gð1 cos yÞ
1 y
tan j ¼ cot ð54Þ
1þg 2
hn0
g¼
m 0 c2
where hn0 and hn are the energies of the incident and scattered photon, respectively, y is the
angle between the photon directions of travel before and following a scattering interaction,
and T and f are the kinetic energy and scattering angle of the Compton recoil electron,
respectively.
For f ¼ 180 , Eqs. (52) and (53) reduce to
1
ðhnÞmin ¼ hn0 ð55Þ
1 þ 2g
and
2g
Tmax ¼ hn0 ð56Þ
1 þ 2g
The differential Klein–Nishina collision cross section dsKN =dO (defined as the ratio of the
number of photons scattered in a particular direction to the number of incident photons)
for unpolarized photons striking unbound, randomly oriented electrons is given by (Klein
and Nishina, 1929)
dsKN r20 hn 2 hn0 hn cm2
¼ þ sin2 y ð57Þ
dO 2 hn0 hn hn0 electron sr
For very small energies hn0 m0 c2 , the expression reduces to the classic Thompson
scattering cross section for electromagnetic radiation on an electron:
Again, at the low-energy limit, this cross section reduces to the classic Thomson cross
section:
8 cm2
sTh ¼ pr20 ¼ 0:6652 1024 ð62Þ
3 electron
The usual Klein–Nishina theory that assumes that the target electron is free and at
rest cannot be directly applicable in some cases. Departures from it occur at low energies
because of electron-binding effects and, at high energies, because of the possibility of
emission of an additional photon (double Compton effect) and radiative corrections as-
sociated with emission and reabsorption of virtual photons; these corrections are discussed
in the work of Hubbell (1969).
The total incoherent (Compton) collision cross section per atom sC, involving the
binding corrections by applying the so-called incoherent scattering function S(x, Z ), can
be calculated according to
Z1 (
1 2
sC ¼ r0 ½1 þ gð1 cos yÞ2
2
1
" # )
g2 ð1 cos yÞ2 cm2
1 þ cos y þ
2
ZSðx; ZÞ 2pdðcos yÞ ð65Þ
1 þ gð1 cos yÞ atom
where x ¼ sinðy=2Þ=l is the momentum transfer parameter and l is the photon wavelength
(in angstroms).
Copyright © 2002 Marcel Dekker, Inc.
The values of the incoherent scattering function S(x, Z ) and the incoherent collision
cross section sC are given by Hubbell et al. (1975). A useful combination of analytical
functions for calculating S(x, Z ) has recently been proposed by Szalóki (1996):
8
>
< X 3
s1 ðx; ZÞ ¼ di ½expðxÞ 1i ; 0 x xi
Sðx; ZÞ ¼
>
: i¼1
s2 ðx; ZÞ ¼ ½Z s1 ðx; ZÞ t2 g1 ðxÞ þ t2 g2 ðxÞ þ s1 ðx1 ; ZÞ x1 < x
ð66Þ
where
and
The parameters for the calculation of the S(x, Z ), including the critical value of x1, are
given by Szalóki (1996) for all elements (Z ¼ 1–100) and any values of x. The average
deviation between the calculated [Eq. (66)] and tabulated data (Hubbell et al., 1975) is
slightly above 1%.
The incoherent collision cross sections sC can also be calculated by using ln–ln
polynomials already defined by Eq. (47) (by simply substituting tj with sC and taking
i ¼ 3). The values of the fitted coefficients for the ln–ln representation for sC valid in the
photon energy range 1 keV to 1 MeV are given in Appendix VIII.
To complete this subsection, it is worth mentioning the Compton effect for polarized
radiation. The differential collision cross section ðdsKN =dOÞpp for the plane-polarized ra-
diation scattered by unoriented electrons has also been derived by Klein and Nishina. It
represents the probability that a photon, passing through a target containing one electron
per square centimeter, will be scattered at an angle y into a solid angle dO in a plane
making an angle b with respect to the plane containing the electrical vector of the incident
wave:
dsKN r20 hn 2 hn0 hn cm2
¼ þ 2 sin y cos b
2 2
ð67Þ
dO pp 2 hn0 hn hn0 electron sr
The cross section has its maximum value for b ¼ 90 , indicating that the photon and
electron tend to be scattered at right angles to the electrical vector of the incident
radiation.
The scattering of circularly polarized (cp) photons by electrons with spins aligned in
the direction of the incident photon is described by
2
dsKN 2 hn
¼ r0
dO cp hn0
hn0 hn hn0 hn cm2
þ sin2 y cos y ð68Þ
hn hn0 hn hn0 electron sr
The first term is the usual Klein–Nishina formula for unpolarized radiation. The þ sign
for the additional term applies to right circularly polarized photons.
Copyright © 2002 Marcel Dekker, Inc.
C. Rayleigh Scattering
Rayleigh scattering is a process by which photons are scattered by bound atomic electrons
and in which the atom is neither ionized nor excited. The incident photons are scattered
with unchanged frequency and with a definite phase relation between the incoming and
scattered waves. The intensity of the radiation scattered by an atom is determined by
summing the amplitudes of the radiation coherently scattered by each of the electrons
bound in the atom. It should be emphasized that, in Rayleigh scattering, the coherence
extends only over the Z electrons of individual atoms. The interference is always con-
structive, provided the phase change over the diameter of the atom is less than one-half a
wavelength; that is, whenever
4p y
ra sin <1 ð69Þ
l 2
dsR 1 2 cm2
¼ r0 ð1 þ cos2 yÞ j Fðx; ZÞj2 ð70Þ
dO 2 atom sr
Z1
sin½ð2p=lÞrs
Fðx; ZÞ ¼ rðrÞ4pr dr ð71Þ
ð2p=lÞrs
0
where rðrÞ is the total density, r is the distance from the nucleus, and s ¼ 2 sinðy=2Þ. The
atomic form factor has been calculated for Z < 26 using the Hartree electronic distribution
(Pirenne, 1946) and for Z > 26 using the Fermi–Thomas distribution (Compton and
Allison, 1935).
At high photon energies, Rayleigh scattering is confined to small angles; at low
energies, particularly for high-Z materials, the angular distribution of the Rayleigh-scat-
tered radiation is much broader. A useful simple criterion for judging the angular spread
of Rayleigh scattering is given by (Evans, 1958).
0:0133Z1=3
yR ¼ 2 arcsin ð72Þ
E ðMevÞ
where yR is the opening half-angle of a cone containing at least 75% of the Rayleigh-
scattered photons. In the forward direction, jFðx; ZÞj2 ¼ Z2 , so that Rayleigh scattering
becomes appreciable in magnitude and must be accounted for in any g- or x-ray scattering
experiments.
The total coherent (Rayleigh) scattering cross section per atom sR can be calcu-
lated from
Copyright © 2002 Marcel Dekker, Inc.
Z1
1
sR ¼ r20 ð1 þ cos2 yÞjFðx; ZÞj2 2p dðcos yÞ
2
1
ð73Þ
Z1 2
3 cm
¼ sTh ð1 þ cos2 yÞjFðx; ZÞj2 dðcos yÞ
8 atom
1
The values of the atomic form factor F(x, Z ) and the coherent scattering cross section sR
are given in the work of (Hubbell et al., 1975). Recently, Szalóki (1996) proposed a useful
combination of analytical functions to calculate F(x, Z):
8
> f11 ðx; ZÞ ¼ a expðb1 xÞ þ ðZ aÞ expðcxÞ; 0 x x1
>
> f ðx; ZÞ ¼ f ðx ; ZÞ expðb ðx xÞ;
< 12 11 1 2 1 x 1 x x2
F1 ðx; ZÞ ¼ f13 ðx; ZÞ ¼ f12 ðx2 ; ZÞ exp½b3 ðx2 xÞ; x 2 x x3 ð74Þ
>
> h ib4
>
: f14 ðx; ZÞ ¼ f13 ðx3 ; ZÞ x ;
x3 x 3 x x4
where 1 Z 7
8
< f21 ðx; ZÞ ¼ a expðb1 xÞ þ ðZ aÞ expðcxÞ; 0 x x1
F2 ðx; ZÞ ¼ f22 ðx; ZÞ ¼ f21 ðx1 ; ZÞ exp½b2 ðx1 xÞ; x 1 x x2 ð75Þ
:
f23 ðx; ZÞ ¼ f22 ðx2 ; ZÞ exp½b3 ðx2 xÞ; x 2 x x3
where 8 Z 100
The parameters for the calculation of F(x, Z ), including the critical values of x1, x2,
and x3, are given by Szalóki (1996) for all elements (Z ¼ 1–100) and the momentum
transfer x from 0 to 15 Å71. The average deviation between the calculated [Eqs. (74) and
(75)] and tabulated data (Hubbell et al., 1975) is less than 2%.
The simplest method for calculating the coherent scattering cross section sR consists
in applying the ln–ln representation [see Eq. (47) with sR instead of tj and i ¼ 3]. The
values of the fitted coefficients for ln–ln polynomials for calculating sR in the photon
energy range 1 keV to 1 MeV are given in Appendix IX.
Applicability range
i ri si ti ni %a Zmin Zmax
1 3 3 5
K 1.304610 2.633610 9.718610 4.144610 3.5 11 63
L1 4.5066101 1.5666102 7.5996104 1.7926105 2.2 28 83
L2 6.0186101 1.9646102 5.9356104 1.8436105 2.3 30 83
L3 3.3906101 4.9316102 2.3366103 1.8366106 1.9 30 83
M1 8.645 3.9776101 5.9636103 3.6246105 0.4 52 83
M2 7.499 3.4596101 5.2506103 3.2636105 0.4 55 83
M3 6.280 2.8316101 4.1176103 2.5056105 0.4 55 83
M4 4.778 2.1846101 3.3036103 2.1156105 0.4 60 83
M5 2.421 1.1726101 1.8456103 1.3976105 0.4 61 83
a
Standard deviation of calculated energies [Eq. (77)] relative to the energies from the tables of McMaster et al.
(1969).
Source: From Wernisch et al., 1984.
where T1(l) and T2(l) factors are related to the dipole and multipole transitions, respec-
tively. Although according to this equation a slight displacement of the peak position of
the Raman band is expected with the scattering angle, the peak position does not shift in
most experiments.
In general, x-ray Raman scattering gives information about the unoccupied states
above the Fermi level of the relevant solids. Moreover, this kind of inelastic scattering by
electrons in solids should sometimes be taken into account in x-ray spectrum evaluation.
This section provides some of the necessary background information for subsequent
chapters dealing with quantitative x-ray analysis. Derivation of any relationship between
excitation source intensity and measured characteristic x-rays is sometimes complex and is
presented in detail in many relevant books, such as those by Jenkins et al. (1981) and
Tertian and Claisse (1982) on XRF analysis and by Goldstein et al. (1981) and Heinrich
(1981) on EPXMA.
A. Photon Excitation
When continuous (polychromatic) radiation is used to excite the characteristic x-rays of
element i in a completely homogeneous sample of thickness T (cm) and when enhancement
effects are neglected, the intensity of the fluorescent radiation Ii (Ei) is described by
dO1 dO2 eðEi Þ
Ii ðEi Þ dO1 dO2 ¼
4p sin C1
Zmax
E
1 exp½rTðmðE0 Þ csc C1 þ mðEi Þ csc C2 Þ
ai ðE0 Þ I0 ðE0 Þ dE0 ð88Þ
mðE0 Þ csc C1 þ mðEi Þ csc C2
Ec;i
with
1
ai ðE0 Þ ¼ Wi t0i ðE0 Þoi pi 1 ð89Þ
ji
where dO1 and dO2 are the differential solid angles for the incident (primary) and emerging
(characteristic) radiation, respectively; eðEi Þ is the intrinsic detector efficiency for recording
a photon of energy Ei ; Ec;i and Emax are the critical absorption energy of element i and the
maximum energy in the excitation spectrum; r is the density of the specimen (in g=cm3);
C1 and C2 are the effective incidence and takeoff angles, respectively; mðE0 Þ and mðEi Þ are
the total mass attenuation coefficients (in cm2=g) for the whole specimen [Eq. (45)] at
Copyright © 2002 Marcel Dekker, Inc.
energies E0 and Ei , respectively; I0 ðE0 Þ dE0 is the number of incident photons per second
per steradian in the energy interval E0 to E0 þ dE0 ; Wi is the weight fraction of the ith
element; and t0i ðE0 Þ is the total photoelectric mass absorption coefficient for the ith element
at the energy E0 (in cm2=g).
Because large solid angles O1 and O2 are used for the excitation and characteristic
x-rays in practical spectrometers, Eq. (88) should also be integrated over these finite solid
angles. Such calculations can often be omitted; however, and for a given measurement
geometry, an experimentally determined geometry factor G can be applied.
As seen from Eq. (88), the intensity of characteristic x-rays is modified by the effects
of primary [mðE0 Þ] and secondary [mðEi Þ] absorption in the specimen; this is a major source
of the so-called matrix effects in XRF analysis.
If the excitation source is monochromatic (emits only one energy), Eq. (88) simplifies to
1. Thin-SampleTechnique
If the total mass per unit area of a given sample m ¼ rT is small, Eq. (90) simplifies to
G
Iithin ðEi Þ ¼ eðEi Þai ðE0 ÞI0 ðE0 Þm ð92Þ
sin C1
The relative error resulting from applying Eq. (92) instead of Eq. (90) does not exceed 5%
when the total mass per unit area satisfies the condition
0:1
m thin ð93Þ
mðE0 Þ csc C1 þ mðEi Þ csc C2
A major feature of the thin-sample technique is that the intensity of the characteristic
x-rays, Iithin , depends linearly on the concentration of the ith element (or on its mass per
unit area); it is equivalent to the fact that, in the thin-sample technique, matrix effects can
safely be neglected.
Copyright © 2002 Marcel Dekker, Inc.
2. Thick-SampleTechnique
By the term ‘‘thick sample,’’ we mean a sample whose mass per unit area (thickness) is
greater than the so-called saturation mass (thickness). The saturation thickness is defined
as a limiting value above which practically no further increase in the intensity of the
characteristic radiation is observed as the sample thickness is increased.
If the total mass per unit area of a given sample is sufficiently large, Eq. (90) sim-
plifies to
The relative error resulting from applying Eq. (94) instead of Eq. (90) does not exceed 1%
when the total mass per unit area satisfies the condition
4:61
m thick
ð95Þ
mðE0 Þ csc C1 þ mðEi Þ csc C2
B. Electron Excitation
For a thin foil (defined as a foil of thickness such that the beam electron undergoes only one
scattering act), the intensity of the characteristic x-rays Iitf (in photons per second) is given by
1
Iitf ¼ Ki I0 Qi oi N0 rTWi ð96Þ
Ai
where Ki is a factor depending on the measurement geometry and detection efficiency and
the other symbols have the following dimensions: I0 ¼ electrons=s, Qi ¼ ionizations per
electron for 1 atom=cm2, oi ¼ photons=ionization. N0 ¼ atoms=mol, 1=Ai ¼ mol=g,
r ¼ g=cm3, and T ¼ cm.
For a bulk target, Eq. (96) must be integrated over the electron path, taking into
account the loss of energy by the electron beam:
ZEc;i
Wi Qi ðEÞ
Iibulk ¼ I0 oi Ki N0 dE ð97Þ
Ai dE=dðrxÞ
E0
where dE=dðrxÞ is the electron energy loss per unit of distance traveled in a material, given
by the Bethe equation (Heinrich, 1981; Bethe and Ashkin, 1953):
dE 1 Xn
Wj Z j E
¼ 78; 500 ln 1:166 keV=g cm2 ð98Þ
dðrxÞ E j¼1 Aj Jj
where Jj is the mean ionization potential of element jðJj ¼ 9:76Zj þ 58:5Zj0:19 eVÞ
(Goldstein et al., 1981). Strictly speaking, the Bethe equation is only valid for the electron
energy E > 6:3Jj . Below this limit, the energy loss of the electrons should be described
either by the modification of Rao-Sahib and Wittry (1972) [Eq. (99)] or by the modi-
fication of Love et al. (l978) [Eq. (l00)]:
Copyright © 2002 Marcel Dekker, Inc.
dE 1 X n
Wj Z j
¼ 0:6236 105 pffiffiffiffi pffiffiffiffi keV=g cm2 ð99Þ
dðrxÞ A
E j¼1 j Jj
dE X n
Wj Z j 1
¼ 1=2
keV=g cm2 ð100Þ
dðrxÞ j¼1
A j j 1:18 10 ðE=Jj Þ
J 5 þ 1:47 106 E=Jj
However, to describe the intensity of characteristic x-rays emitted, three effects should be
considered additionally (Heinrich, 1981; Goldstein et al., 1981; Love and Scott, 1981):
1. Absorption of characteristic x-rays within the specimen
2. Electron backscattering
3. Secondary fluorescence by characteristic x-rays and=or bremsstrahlung con-
tinuum produced by an electron beam
C. Particle Excitation
For a thin, uniform, homogeneous target, the intensity of characteristic x-rays Itti is given
by the simple formula
1
I tti ¼ Ki I0 N0 ss;i ðE0 Þoi mi ð101Þ
Ai
where mi is the areal density of the element with atomic number Zi and atomic mass Ai .
For a thick, homogeneous target, the intensity of characteristic x-rays, Iithick , from
the element i of concentration Wi can be calculated by (Campbell and Cookson, 1984)
Ki oi N0 X
0
ss;i ðEÞTi ðEÞ
Iithick ¼ I 0 Wi dE ð102Þ
Ai E
SðEÞ
0
where S(E) is the stopping power and Ti ðEÞ is the photon attenuation factor. The latter is
given by
0 1
ZE
B sin C1 dE C
Ti ðEÞ ¼ exp@mðEi Þ A ð103Þ
sin C2 SðEÞ
E0
More details on various x-ray analytical techniques are provided in subsequent chapters.
The nomenclature commonly used in XRS to describe x-ray emission spectra was in-
troduced by M. Sieghahn in the l920s and is based on the relative intensity of lines from
different series. A new and more systematic notation for x-ray emission lines and ab-
sorption edges, based on the energy-level designation, was developed by the International
Union of Pure and Applied Chemistry (Jenkins et al. 1991). Because the new notation,
called the IUPAC notation, replaces the Siegbahn’s notation, some characteristic features
of the new nomenclature must be mentioned. The IUPAC notation prescribes Arabic
numerals for subscripts; the original notation uses Roman numerals (e.g., L2 and L3 ,
Copyright © 2002 Marcel Dekker, Inc.
instead of LII and LIII ). In the IUPAC notation, states with double or multiple vacancies
should be denoted by, for example, K2 ; KL1 , and Ln2;3 , which correspond to the electron
configurations 1s2 ; 1s1 2s1 , and 2pn , respectively. X-ray transitions and x-ray emission
diagram lines are denoted by the initial (placed first) and final x-ray levels separated by a
hyphen. To conform with the IUPAC notation of x-ray spectra, the hyphen separating the
initial and final state levels should also be introduced into the notation for Auger electron
emission process. The IUPAC notation is compared with the Sieghahn notation in Ap-
pendix XII.
1 H 918 0.014
2 He 504 0.025
3 Li 226.953 0.055
4 Be 106.9 0.116
5 B 64.6 0.192
6 C 43.767 0.283
7 N 31.052 0.399
8 O 23.367 0.531
9 F 18.05 0.687
10 Ne 14.19 0.874 258 0.048 564 0.022 564 0.022
11 Na 11.48 1.08 225 0.055 365 0.034 365 0.034
12 Mg 9.512 1.303 197 0.063 248 0.050 253 0.049
13 Al 7.951 1.559 143 0.087 170 0.073 172 0.072
14 Si 6.745 1.837 105 0.118 125 0.099 127 0.098
15 P 5.787 2.142 81.0 0.153 96.1 0.129 96.9 0.128
16 S 5.018 2.470 64.2 0.193 75.6 0.164 76.1 0.163
17 Cl 4.397 2.819 52.1 0.238 61.1 0.203 61.4 0.202
18 Ar 3.871 3.202 43.2 0.287 50.2 0.247 50.6 0.245
19 K 3.437 3.606 36.4 0.341 41.8 0.297 42.2 0.294
20 Ca 3.070 4.037 30.7 0.399 35.2 0.352 35.5 0.349
21 Sc 2.757 4.495 26.8 0.462 30.2 0.411 30.8 0.402
22 Ti 2.497 4.963 23.4 0.530 27.0 0.460 27.3 0.454
23 V 2.269 5.462 20.5 0.604 23.9 0.519 24.2 0.512
24 Cr 2.070 5.987 18.3 0.679 21.3 0.583 21.6 0.574
25 Mn 1.896 6.535 16.3 0.762 19.1 0.650 19.4 0.639
26 Fe 1.743 7.109 14.6 0.849 17.2 0.721 17.5 0.708
27 Co 1.608 7.707 13.3 0.929 15.6 0.794 15.9 0.779
28 Ni 1.488 8.329 12.22 1.015 14.2 0.871 14.5 0.853
29 Cu 1.380 8.978 11.27 1.100 13.0 0.953 13.3 0.933
Line a1,2 a1 a2 b1 b3 b2 b4
Approximate
intensity 150 100 50 15 5 <1
Li 3 230
B 4 113
B 5 67
C 6 44
N 7 31.603
O 8 23.707
F 9 18.307
Ne 10 14.615 14.460
Na 11 11.909 11.574 11.726
Mg 12 9.889 9.559 9.667
Al 13 8.339 8.338 8.341 7.960 8.059
Si 14 7.126 7.125 7.127 6.778
P 15 6.155 6.154 6.157 5.804
S 16 5.373 5.372 5.375 5.032
Cl 17 4.729 4.728 4.731 4.403
Ar 18 4.192 4.191 4.194 3.886
K 19 3.744 3.742 3.745 3.454
Ca 20 3.360 3.359 3.362 3.089
Sc 21 3.032 3.031 3.034 2.780
Ti 22 2.750 2.749 2.753 2.514
V 23 2.505 2.503 2.507 2.285
Cr 24 2.291 2.290 2.294 2.085
Mn 25 2.103 2.102 2.105 1.910
Fe 26 1.937 1.936 1.940 1.757
Co 27 1.791 1.789 1.793 1.621
Ni 28 1.659 1.658 1.661 1.500 1.489
Cu 29 1.542 1.540 1.544 1.392 1.393 1.381
Zn 30 1.437 1.435 1.439 1.296 1.284
Ga 31 1.341 1.340 1.344 1.207 1.208 1.196
Ge 32 1.256 1.255 1.258 1.129 1.129 1.117
As 33 1.177 1.175 1.179 1.057 1.058 1.045
Se 34 1.106 1.105 1.109 0.992 0.993 0.980
Br 35 1.041 1.040 1.044 0.933 0.933 0.921
Kr 36 0.981 0.980 0.984 0.879 0.879 0.866 0.866
Rb 37 0.927 0.926 0.930 0.829 0.830 0.817 0.815
Sr 38 0.877 0.875 0.880 0.783 0.784 0.771 0.770
Y 39 0.831 0.829 0.833 0.740 0.741 0.728 0.727
Zr 40 0.788 0.786 0.791 0.701 0.702 0.690 0.689
Nb 41 0.748 0.747 0.751 0.665 0.666 0.654 0.653
Mo 42 0.710 0.709 0.713 0.632 0.633 0.621 0.620
Tc 43 0.676 0.675 0.679 0.601 0.590
Ru 44 0.644 0.643 0.647 0.572 0.573 0.562 0.561
Rh 45 0.614 0.613 0.617 0.546 0.546 0.535 0.534
Pd 46 0.587 0.585 0.590 0.521 0.521 0.510
Line a1,2 a1 a2 b1 b3 b2 b4
Approximate
intensity 150 100 50 15 5 <1
Line a a1 a2 b1 b2 b3 b4 b5 b6 b7 b9
Approximate
intensity 110 100 10 50 20 6 4 1 <1 <1 <1
Na 11
Mg 12
Al 13
Si 14
P 15
S 16
Cl 17
Ar 18
K 19
Ca 20 36.393 36.022
Sc 21 31.393 31.072
Ti 22 27.445 27.074
V 23 24.309 23.898
Cr 24 21.713 21.323 19.429
Mn 25 19.489 19.158 17.757
Fe 26 17.602 17.290 15.742
Co 27 16.000 15.698 14.269
Ni 28 14.595 14.308 13.167
Cu 29 13.357 13.079 12.115
Zn 30 12.282 12.009 11.225
Ga 31 11.313 11.045
Ge 32 10.456 10.194
As 33 9.671 9.414 8.930
Se 34 8.990 8.735
Br 35 8.375 8.126
Kr 36
Rb 37 7.318 7.325 7.075 6.788 6.821 6.984
Sr 38 6.863 6.870 6.623 6.367 6.403 6.519
Y 39 6.449 6.456 6.211 5.983 6.018 6.094
Zr 40 6.070 6.077 5.836 5.586 5.632 5.668 5.710
Nb 41 5.725 5.732 5.492 5.238 5.310 5.346 5.361
Mo 42 5.406 5.414 5.176 4.923 5.013 5.048 5.048
Tc 43
Ru 44 4.846 4.854 4.620 4.372 4.487 4.532 4.487
Rh 45 4.597 4.605 4.374 4.130 4.253 4.289 4.242
Pd 46 4.368 4.376 4.146 3.909 4.034 4.071 4.016 3.792
Ag 47 4.154 4.162 3.935 3.703 3.834 3.870 3.808 3.605
Cd 48 3.956 3.965 3.739 3.514 3.644 3.681 3.614 3.430
In 49 3.752 3.781 3.555 3.339 3.470 3.507 3.436 3.268
Sn 50 3.600 3.609 3.385 3.175 3.306 3.344 3.270 3.155 3.115
Sb 51 3.439 3.448 3.226 3.023 3.152 3.190 3.115 3.005 2.973
Te 52 3.290 3.299 3.077 2.882 3.009 3.046 2.971 2.863 2.839
I 53 3.148 3.157 2.937 2.751 2.874 2.912 2.837 2.730 2.713
Xe 54
Cs 55 2.892 2.902 2.683 2.511 2.628 2.666 2.593 2.485 2.478
410
260
180
67.84 67.25
56.212 56.813
47.835 47.325
41.042 40.542
36.671 35.200
31.423 30.942
27.826 27.375
24.840 24.339
22.315 21.864
20.201 19.73
18.358 17.86
16.693 16.304
15.297 14.940
14.081 13.719
12.976 12.620
11.944 11.608
11.069 10.732
10.293 9.959
9.583 9.253
Line a a1 a2 b1 b2 b3 b4 b5 b6 b7 b9
Approximate
intensity 110 100 10 50 20 6 4 1 <1 <1 <1
0.716
0.776 0.838 0.844 0.694 0.682 0.675 0.649 0.717 0.673 0.680 1.167 0.908
0.730 0.653 0.642 0.635 0.611 0.675 0.632 0.640 1.115 0.855 1.011 1.080
0.708 0.634 0.624 0.617 0.594 0.655 0.613 1.091 0.830
0.687 0.615 0.605 0.598 0.577 0.635 0.595 0.601 1.067 0.806 0.904 1.035
0.597
0.579
0.562
Line a1 a2 b g l
K 19 680
Cu 29 170
Ru 44 26.85
Rh 45 25.00
Pd 46
Ag 47 21.80
Cd 48 20.46
In 49
Sn 50 17.94
Sb 51 16.92
Te 52 15.93
Ba 56 12.700
La 57 14.88 14.51 12.064
Ce 58 14.06 13.78 11.534 18.38
Pr 59 10.997
Nd 60 12.675 Band 10.504
Sm 62 Band Band 9.599
Eu 63 Band 10.744 9.211 14.22
Gd 64 Band 10.253 8.844 13.57
Tb 65 Band 9.792 8.485 12.98
Dy 66 Band 9.364 8.144 12.43
Ho 67 Band 8.965 7.865 11.86
Er 68 Band 8.593 7.545 11.37
Tm 69 8.460 8.246
Yb 70 8.139 8.155 7.909 7.023 10.48
Lu 71 7.840 7.600 6.761 10.07
Hf 72 7.539 7.546 7.304 6.543 9.69
Ta 73 7.251 7.258 7.022 6.312 9.32
W 74 6.983 6.990 6.756 6.088 8.96
Re 75 6.528 6.504 5.887 8.63
Os 76 6.490 6.267 5.681
Ir 77 6.261 6.275 6.037 5.501 8.02
Pt 78 6.046 6.057 5.828 5.320 7.74
Au 79 5.840 5.854 5.623 5.145 7.47
Hg 80 5.666 5.452
Tl 81 5.461 5.472 5.250 4.825 6.97
Pb 82 5.285 5.299 5.075 4.674 6.74
Bi 83 5.118 5.129 4.909 4.531 6.52
Th 90 4.138 4.151 3.942 3.679 5.24
Pa 91 4.022 4.035 3.827 3.577 5.08
U 92 3.910 3.924 3.715 3.480 4.95
Source: From Clark, 1963.
Atomic
number Element Kb2 Kb1 Ka1 Ka2 Lg1 Lb2 Lb1 La1 La2
3 Li 0.052
4 Be 0.110
5 B 0.185
6 C 0.282
7 N 0.392
8 C 0.523
9 F 0.677
10 Ne 0.851
11 Na 1.067 1.041
12 Mg 1.297 1.254
13 Al 1.553 1.487 1.486
14 Si 1.832 1.40 1.739
15 P 2.136 2.015 2.014
16 S 2.464 2.309 2.306
17 Cl 2.815 2.622 2.621
18 Ar 3.192 2.957 2.955
19 K 3.589 3.313 3.310
20 Ca 4.012 3.691 3.688 0.344 0.341
21 Sc 4.460 4.090 4.085 0.399 0.395
22 Ti 4.931 4.510 4.504 0.458 0.492
23 V 5.427 4.952 4.944 0.519 0.510
24 Cr 5.946 5.414 5.405 0.581 0.571
25 Mn 6.490 5.898 5.887 0.647 0.636
26 Fe 7.057 6.403 6.390 0.717 0.704
27 Co 7.647 6.930 6.915 0.790 0.775
28 Ni 8.328 8.264 7.477 7.460 0.866 0.849
29 Cu 8.976 8.904 8.047 8.027 0.943 0.928
30 Zn 9.657 9.571 8.638 8.615 1.032 1.009
31 Ga 10.365 10.263 9.251 9.234 1.122 1.096
32 Ge 11.100 10.981 9.885 9.854 1.216 1.166
33 As 11.863 11.725 10.543 10.507 1.517 1.282
34 Se 12.651 12.495 11.221 11.181 1.419 1.379
35 Br 13.465 13.290 11.923 11.877 1.526 1.480
36 Kr 14.313 14.112 12.648 12.597 1.638 1.587
37 Rb 15.184 14.960 13.394 13.335 1.752 1.694 1.692
38 Sr 16.083 15.834 14.164 14.097 1.872 1.806 1.805
39 Y 17.011 16.736 14.957 14.882 1.996 1.922 1.920
40 Zr 17.969 17.666 15.774 15.690 2.302 2.219 2.124 2.042 2.040
41 Nb 18.951 18.621 16.614 16.520 2.462 2.367 2.257 2.166 2.163
42 Mo 19.964 19.607 17.478 17.373 2.623 2.518 2.395 2.293 2.290
43 Tc 21.012 20.585 18.410 18.328 2.792 2.674 2.538 2.424 2.420
44 Ru 22.072 21.655 19.278 19.149 2.964 2.836 2.683 2.558 2.554
45 Rh 23.169 22.721 20.214 20.072 3.144 3.001 2.834 2.696 2.692
46 Pd 24.297 23.816 21.175 21.018 3.328 3.172 2.990 2.838 2.833
47 Ag 25.454 24.942 22.162 21.988 3.519 3.348 3.151 2.994 2.978
48 Cd 26.641 26.093 23.172 22.982 3.716 3.528 3.316 3.133 3.127
49 In 27.859 27.274 24.207 24.000 3.920 3.713 3.487 3.287 3.279
50 Sn 29.106 28.483 25.270 25.042 4.131 3.904 3.662 3.444 3.436
Atomic
number Element Kb2 Kb1 Ka1 Ka2 Lg1 Lb2 Lb1 La1 La2
Atomic e e
number Element Ka2 =Ka1 Kb3 =Kb1 ðKb1 þ Kb3 Þ=Ka1 Kb1 a =Ka1 Kb2 b =Ka1 Kbc =Kad
Atomic
number Element Lb3 Lb4 Lg3 Lg2
36 Kr 100 — 10.8 —
38 Sr 100 — 14.7 —
40 Zr 100 — 18.1 —
42 Mo 100 71.0 21.0 —
44 Ru 100 64.9 23.4 —
46 Pd 100 61.9 25.5 —
48 Cd 100 61.0 27.2 —
50 Sn 100 61.4 28.5 —
52 Te 100 62.6 29.5 —
54 Xe 100 64.1 30.3 —
56 Ba 100 65.5 30.8 —
58 Ce 100 66.7 31.2 —
60 Nd 100 67.5 31.4 —
62 Sm 100 69.7 31.5 —
64 Gd 100 68.0 31.5 —
66 Dy 100 67.8 31.4 —
68 Er 100 67.6 31.4 —
70 Yb 100 67.5 31.4 18.5
72 Hf 100 67.6 31.4 20.0
74 W 100 68.3 31.6 21.5
76 Os 100 69.6 31.9 23.2
78 Pt 100 71.7 32.3 25.0
80 Hg 100 74.8 33.0 26.9
82 Pb 100 78.8 34.0 28.8
84 Po 100 83.9 35.2 30.9
86 Rn 100 89.8 36.8 33.1
88 Ra 100 95.0 38.7 35.3
90 Th 100 102.0 41.0 37.7
92 U 100 110.0 43.8 40.2
94 Pu 100 120.0 47.5 44.0
Atomic
number Element Lb1 LZ Lg1 Lg6
30 Zn 100 12.3 — —
32 Ge 100 10.4 — —
34 Se 100 8.75 — —
36 Kr 100 7.40 — —
38 Sr 100 6.25 — —
40 Zr 100 5.25 0.91 —
42 Mo 100 4.35 6.71 —
44 Ru 100 3.63 10.6 —
46 Pd 100 3.0 13.1 —
48 Cd 100 2.6 14.5 —
50 Sn 100 2.35 15.3 —
52 Te 100 2.25 15.4 —
54 Xe 100 2.2 15.6 —
56 Ba 100 2.16 15.6 —
58 Ce 100 2.12 15.7 —
60 Nd 100 2.10 15.8 —
62 Sn 100 2.10 15.9 —
64 Gd 100 2.10 16.1 —
66 Dy 100 2.10 16.5 —
68 Er 100 2.10 17.0 —
70 Yb 100 2.12 17.6 —
72 Hf 100 2.13 18.4 —
74 W 100 2.16 19.2 0.375
76 Os 100 2.20 20.1 1.73
78 Pt 100 2.23 20.9 2.42
80 Hg 100 2.28 21.7 2.98
82 Pb 100 2.33 22.3 3.45
84 Po 100 2.40 22.8 3.88
86 Rn 100 2.45 23.2 4.29
88 Ra 100 2.50 23.4 4.74
90 Th 100 2.60 23.7 5.25
92 U 100 2.80 24.0 5.88
94 Pu 100 2.30 24.2 6.65
22 Ti 100 — — — — 40.37
24 Cr 100 — — — — 26.13
26 Fe 100 — — — — 15.35
28 Ni 100 — — — — 10.29
30 Zn 100 — — — — 7.56
32 Ge 100 — — — — 5.96
34 Se 100 — — — — 4.98
36 Kr 100 — — — — 4.36
38 Sr 100 — — — — 3.98
40 Zr 100 2.43 — — — 3.75
42 Mo 100 6.40 12.5 — — 3.65
44 Ru 100 9.55 12.2 — — 3.58
46 Pd 100 12.1 12.1 — — 3.55
48 Cd 100 13.9 11.9 — — 3.56
50 Sn 100 15.4 11.7 — — 5.59
52 Te 100 16.4 11.5 — — 3.62
54 Xe 100 17.2 11.3 — — 3.67
56 Ba 100 17.8 11.2 — — 3.73
58 Ce 100 18.2 11.1 — — 3.79
60 Nd 100 18.5 11.1 — — 3.86
62 Sm 100 18.8 11.0 — — 3.92
64 Gd 100 19.2 11.1 — — 3.99
66 Dy 100 19.6 11.1 — — 4.07
68 Er 100 20.0 11.2 — — 4.15
70 Yb 100 20.5 11.2 — — 4.23
72 Hf 100 21.2 11.3 — 1.15 4.32
74 W 100 21.9 11.3 0.242 1.28 4.42
76 Os 100 22.7 11.4 0.873 1.38 4.53
78 Pt 100 23.5 11.4 1.74 1.46 4.65
80 Hg 100 24.4 11.5 2.62 1.55 4.78
82 Pb 100 25.3 11.5 3.24 1.59 4.93
84 Po 100 26.2 11.4 3.85 1.65 5.09
86 Rn 100 26.8 11.4 4.28 1.70 5.27
88 Ra 100 27.3 11.3 4.69 1.75 5.46
90 Th 100 27.5 11.1 4.94 1.80 5.69
92 U 100 27.5 11.0 5.20 1.85 5.93
94 Pu 100 27.0 10.5 5.40 1.89 6.18
Source: From West, 1982–83.
Atomic
number Element aII aI a3 a3I a3II a4 a4I a5 a6 a7 b0 bI
6.753
5.800 5.792 5.712 5.691
5.028 5.023
4.400 4.395
3.882
3.449 3.441 3.412 3.404
3.087 3.082 3.054 3.048
2.772 2.749 2.744
2.506 2.489
2.277 2.262
2.078 2.066
1.895
1.749 1.742
1.614 1.607
1.494 1.487
1.391 1.380
1.295 1.282
1.207
1.128 1.116
1.042 1.044 1.064 1.050 1.054 1.047
0.9770 0.9786 0.9996 0.9854 0.9889
0.9177 0.9194 0.9403 0.9629 0.9297 0.9228
1.8745
0.8135 0.8155 0.8365 0.8234 0.8259 0.8193
0.7681 0.7699 0.7911 0.7778 0.7802 0.7736
0.7261 0.7218 0.7492 0.7362 0.7382 0.7315
0.6870 0.6890 0.7105 0.6973 0.6993 0.6924
0.6510 0.6531 0.6741 0.6619 0.6636 0.6568
0.6288 0.6303 0.6235 0.6256 0.6519
6 C 0.0009 45 Rh 0.81
7 N 0.0015 46 Pd 0.82
8 O 0.0022 47 Ag 0.83
10 Ne 0.0100 48 Cd 0.84
11 Na 0.020 49 In 0.85
12 Mg 0.030 50 Sn 0.86
13 Al 0.040 51 Sb 0.87
14 Si 0.055 52 Te 0.875
15 P 0.070 53 I 0.88
16 S 0.090 54 Xe 0.89
17 Cl 0.105 55 Cs 0.895
18 Ar 0.125 56 Ba 0.90
19 K 0.140 57 La 0.905
20 Sc 0.165 58 Ce 0.91
21 Ca 0.190 59 Pr 0.915
22 Ti 0.220 60 Nd 0.92
23 V 0.240 61 Pm 0.925
24 Cr 0.26 62 Sm 0.93
25 Mn 0.285 63 Eu 0.93
26 Fe 0.32 64 Gd 0.935
27 Co 0.345 65 Tb 0.94
28 Ni 0.375 66 Dy 0.94
29 Cu 0.41 67 Ho 0.945
30 Zn 0.435 68 Er 0.945
31 Ga 0.47 69 Tm 0.95
32 Ge 0.50 70 Yb 0.95
33 As 0.53 71 In 0.95
34 Se 0.565 72 Hf 0.955
35 Br 0.60 73 Ta 0.955
36 Kr 0.635 74 W 0.96
37 Rb 0.665 75 Re 0.96
38 Sr 0.685 76 Os 0.96
39 Y 0.71 77 Ir 0.96
40 Zr 0.72 78 Pd 0.965
41 Nb 0.755 79 Au 0.965
42 Mo 0.77 80 Hg 0.965
43 Tc 0.785 82 Pb 0.97
44 Ru 0.80 92 U 0.97
Source: From Birks, 1971b and Fink, 1974.
54 Xe 0.10 0.01
56 Ba 0.06 0.05 0.01
65 Tb 0.18 0.165 0.018 0.188 0.016
67 Ho 0.22 0.03
0.170 0.055 0.169 0.030
68 Er 0.21 0.03
0.185 0.060 0.172 0.032
70 Yb 0.20 0.02
0.188 0.011 0.183 0.011
71 Lu 0.22 0.03
0.251 0.035
72 Hf 0.22 0.03
0.228 0.025
73 Ta 0.25 0.02 0.27 0.01
0.257 0.013 0.25 0.03
0.191
0.228 0.013
0.254 0.025
74 W 0.207
0.272 0.037
75 Re 0.284 0.043
76 Os 0.290 0.030
77 Ir 0.244
0.262 0.036
78 Pt 0.331 0.021 0.262
0.31 0.04
0.317 0.029
0.291 0.018
79 Au 0.276
0.31 0.04
0.317 0.025
80 Hg 0.39 0.03 0.40 0.02
0.319 0.010 0.32 0.05
0.367 0.050
0.300 0.010
81 Tl 0.07 0.02 0.319 0.010 0.37 0.07
0.373 0.025 0.386 0.053
0.306 0.010
0.330 0.021
82 Pb 0.07 0.02 0.363 0.015 0.337
0.09 0.02 0.315 0.013
0.32
0.35 0.05
0.354 0.028
83 Bi 0.12 0.01 0.32 0.04 0.367
0.095 0.005 0.38 0.02 0.36
0.37 0.05
0.362 0.029
0.40 0.05
0.340 0.018
90 Th 0.42
0.517 0.042
91 Pa 0.46 0.05
92 U 0.44
0.500 0.040
96 Cm 0.28 0.06 0.552 0.032 0.515 0.034
0.55 0.02 0.63 0.02
Source: From Bambynek et al., 1972.
56 Ba 0.66 0.07
65 Tb 0.41 0.36 0.43 0.28 0.066 0.014
67 Ho 0.205 0.034
68 Er 0.255 0.025
70 Yb 0.142 0.009
73 Ta < 0.14 0.19 0.148 0.010
< 0.36 0.20 0.04
74 W 0.27 0.03
75 Re 0.30 0.04
77 Ir 0.46 0.06
78 Pt 0.50 0.05
79 Au 0.25 0.13 0.51 0.13 0.22
0.61 0.07
80 Hg 0.74 0.04 0.22 0.04
0.08 0.02
0.188 0.010
81 Tl 0.17 0.05 0.76 0.10 0.25 0.13
0.14 0.03 0.57 0.10 0.169 0.010
0.56 0.07 0.159 0.013
0.56 0.05
82 Pb 0.15 0.04 0.57 0.03 0.164 0.016
0.17 0.05 0.61 0.08 0.156 0.010
83 Bi 0.19 0.05 0.58 0.05 þ0.14
0.06
7 0.06
0.18 0.02 0.58 0.02 0.164
92 U 0.23 0.12
93 Np 0.10 0.04 0.55 0.09 þ0.05
0.02
70.02
94 Pu 0.22 0.08
0.24 0.08
96 Cm 0.038 0.022 0.68 0.04 0.188 0.019
Source: From Bambynek et al., 1972.
Atomic
number Element M
o oLMa oLMb o1 þ f12o2 ni oi
5 B 1.7E-03
6 C 2.8E-03
7 N 5.2E-03
8 O 8.3E-03
9 F 0.013
10 Ne 0.018
11 Na 0.023
12 Mg 0.030 2.9E-05 1.2E-03 1.2E-03 0.962 0.32 0.64
13 Al 0.039 2.6E-05 7.5E-04 7.5E-04 0.965 0.32 0.64
14 Si 0.050 3.0E-05 3.7E-04 3.8E-04 0.959 0.32 0.64
15 P 0.063 3.9E-05 3.1E-04 3.1E-04 0.951 0.32 0.63
16 S 0.078 7.4E-05 2.6E-04 2.6E-04 0.944 0.32 0.62
17 Cl 0.097 1.2E-04 2.4E-04 2.4E-04 0.939 0.32 0.62
18 Ar 0.118 1.8E-04 2.2E-04 2.2E-04 0.934 0.31 0.62
19 K 0.140 2.4E-04 2.7E-04 2.7E-04 0.929 0.31 0.62
20 Ca 0.163 3.1E-04 3.3E-04 3.3E-04 0.920 0.31 0.61
21 Sc 0.188 3.9E-04 8.4E-04 8.4E-04 0.912 0.31 0.60
22 Ti 0.214 4.7E-04 1.5E-03 1.5E-03 0.902 0.31 0.59
23 V 0.243 5.8E-04 2.6E-03 2.6E-03 0.894 0.31 0.58
24 Cr 0.275 7.1E-04 3.7E-03 3.7E-03 0.885 0.31 0.57
25 Mn 0.308 8.4E-04 5.0E-03 5.0E-03 0.877 0.30 0.58
26 Fe 0.340 1.0E-03 6.3E-03 6.3E-03 0.868 0.30 0.57
27 Co 0.373 1.2E-03 7.7E-03 7.7E-03 0.856 0.30 0.56
28 Ni 0.406 1.4E-03 8.6E-03 9.3E-03 0.847 0.30 0.55 0.028
29 Cu 0.440 1.6E-03 0.010 0.011 0.839 0.30 0.54 0.028
30 Zn 0.474 1.8E-03 0.011 0.012 0.831 0.29 0.54 0.026
31 Ga 0.507 2.1E-03 0.012 0.013 0.822 0.29 0.53 0.032
32 Ge 0.535 2.4E-03 0.013 0.015 0.815 0.28 0.53 0.050
33 As 0.562 2.8E-03 0.014 0.016 0.809 0.28 0.53 0.063
34 Se 0.589 3.2E-03 0.016 0.018 0.804 0.28 0.52 0.076
35 Br 0.618 3.6E-03 0.018 0.020 0.800 0.28 0.52 0.088
36 Kr 0.643 4.1E-03 0.020 0.022 0.797 0.27 0.52 0.100
37 Rb 0.667 4.6E-03 0.022 0.024 0.794 0.27 0.52 0.109
38 Sb 0.690 5.1E-03 0.024 0.026 0.790 0.27 0.52 0.117
39 Y 0.710 5.9E-03 0.026 0.028 0.785 0.26 0.52 0.126
40 Zr 0.730 6.8E-03 0.028 0.031 0.779 0.26 0.52 0.132
41 Nb 0.747 9.4E-03 0.031 0.034 0.713 0.10 0.61 0.137
42 Mo 0.765 0.010 0.034 0.037 0.712 0.10 0.61 0.141
43 Tc 0.780 0.011 0.037 0.040 0.711 0.10 0.61 0.144
44 Ru 0.794 0.012 0.040 0.043 0.709 0.10 0.61 0.148
45 Rh 0.808 0.013 0.043 0.046 0.705 0.10 0.60 0.150
46 Pd 0.820 0.014 0.047 0.049 0.700 0.10 0.60 0.151
47 Ag 0.831 0.016 0.051 0.052 0.694 0.10 0.59 0.153
48 Cd 0.843 0.018 0.056 0.056 0.688 0.10 0.59 0.155
49 In 0.853 0.020 0.061 0.060 0.681 0.10 0.59 0.157
Z aK a1 a2 a3 Z aK a1 a2 a3
Z aK a1 a2 a3 Z aK a1 a2 a3
Table 1 E > EK a
Atomic Atomic
number Element weight A0 A1 A2 A3
Atomic Atomic
number Element weight A0 A1 A2 A3
Atomic Atomic
number Element weight A0 A1 A2
Atomic Atomic
number Element weight A0 A1 A2
Table1 Parameters for Calculating the Total Mass Attenuation Coefficients for Photon Energies Above the K Absorption Edge (k1, k2, k3, k4) and for
Energies Between L1 and EK Absorption Edge (l1, l2, l3, l4); Z ¼ Atomic Number, Sy ¼ Symbol of Corresponding Element
Z Sy EL1 l1 l2 l3 l4 EK k1 k2 k3 k4
Z Sy EL1 l1 l2 l3 l4 EK k1 k2 k3 k4
61 Pm 7.4279 4.93318 2.62474 70.08002 70.02998 45.1840 5.92214 1.70715 70.44407 70.06326
62 Sm 7.7368 4.95712 2.61647 70.07625 70.02487 46.8342 5.92934 1.69904 70.44407 70.06312
63 Eu 8.0520 5.00931 2.61226 70.08680 70.03205 48.5190 5.94975 1.67827 70.44795 70.06335
64 Gd 8.3756 5.03443 2.60481 70.08912 70.03162 50.2391 5.92827 1.64215 70.45629 70.06402
65 Tb 8.7080 5.08055 2.60115 70.07597 70.02286 51.9957 5.95498 1.63471 70.45467 70.06359
66 Dy 9.0458 5.11495 2.59238 70.08230 70.02541 53.7885 5.95991 1.61100 70.46029 70.06408
67 Ho 9.3942 5.15537 2.58071 70.08885 70.02644 55.6177 5.94804 1.55826 70.47717 70.06595
68 Er 9.7513 5.19648 2.57148 70.09459 70.02802 57.4855 5.96408 1.54243 70.47949 70.06606
69 Tm 10.1157 5.23938 2.56466 70.09047 70.02454 59.3896 5.88779 1.43854 70.50808 70.06870
70 Yb 10.4864 5.26855 2.55541 70.10174 70.03012 61.3323 6.04546 1.57974 70.45878 70.06334
71 Lu 10.8704 5.31163 2.55490 70.09208 70.02484 63.3138 6.07395 1.57891 70.45631 70.06301
72 Hf 11.2707 5.34305 2.54268 70.10077 70.02734 65.3508 6.00771 1.48367 70.48598 70.06608
73 Ta 11.6815 5.37806 2.53595 70.09548 70.02359 67.4164 6.05159 1.50856 70.47273 70.06439
74 W 12.0998 5.41320 2.53128 70.10036 70.02631 69.5250 6.00250 1.43225 70.49552 70.06672
75 Re 12.5267 5.44913 2.52042 70.10577 70.02724 71.6764 6.03030 1.43738 70.48952 70.06580
76 Os 12.9680 5.47715 2.52666 70.09007 70.02154 73.8708 6.11400 1.50446 70.46496 70.06309
77 Ir 13.4185 5.50865 2.50428 70.10492 70.02495 76.1110 6.09889 1.45990 70.47717 70.06429
78 Pt 13.8799 5.54158 2.50551 70.09847 70.02303 78.3948 6.15378 1.53187 70.44176 70.05974
79 Au 14.3528 5.57564 2.48980 70.10793 70.02561 80.7249 6.15303 1.47494 70.46441 70.06253
80 Hg 14.8393 5.60041 2.47759 70.11241 70.02570 83.1023 6.19862 1.50676 70.45079 70.06097
81 Tl 15.3467 5.62415 2.46173 70.12251 70.02850 85.5304 6.17641 1.46912 70.46009 70.06183
82 Pb 15.8608 5.65349 2.45575 70.12044 70.02698 88.0045 6.17648 1.45115 70.46277 70.06196
83 Bi 16.3875 5.69400 2.47086 70.10349 70.02277 90.5259 6.30427 1.55736 70.42520 70.05792
84 Po 16.9393 5.73466 2.46840 70.09608 70.01973 93.1050 6.21512 1.43087 70.46426 70.06195
85 At 17.4930 5.76863 2.45600 70.10153 70.02099 95.7299 6.18945 1.38290 70.47598 70.06300
86 Rn 18.0490 5.74682 2.43686 70.11031 70.02272 98.4040 6.24492 1.47699 70.43939 70.05879
87 Fr 18.6390 5.78319 2.43220 70.10993 70.02259 101.1370 6.31471 1.51865 70.42408 70.05721
88 Ra 19.2367 5.80588 2.41932 70.11330 70.02255 103.9219 6.26760 1.45136 70.44373 70.05917
79 Au 3.4249 4.22770 2.52409 0.01426 70.04262 11.9187 5.125 2.690 13.7336 5.406 2.303 14.3528
80 Hg 3.5616 4.26093 2.51954 0.02092 70.04665 12.2839 5.156 2.643 14.2087 5.430 2.337 14.8393
81 Tl 3.7041 4.29302 2.51974 0.01369 70.04376 12.6575 5.184 2.630 14.6979 5.451 2.323 15.3467
82 Pb 3.8507 4.32934 2.51778 0.00460 70.03614 13.0352 5.220 2.644 15.2000 5.512 2.502 15.8608
83 Bi 3.9991 4.36991 2.52184 70.01213 70.02654 13.4186 5.257 2.642 15.7111 5.536 2.453 16.3875
84 Po 4.1494 4.41866 2.52158 70.00971 70.03298 13.8138 5.301 2.625 16.2443 5.555 2.387 16.9393
85 At 4.3170 4.46126 2.51394 0.00640 70.04593 14.2135 5.347 2.642 16.7847 5.574 2.335 17.4930
86 Rn 4.4820 4.45433 2.51411 70.00666 70.03871 14.6194 5.334 2.637 17.3371 5.535 2.281 18.0490
87 Fr 4.6520 4.49562 2.51108 70.01358 70.02823 15.0312 5.376 2.647 17.9065 5.650 2.494 18.6390
88 Ra 4.8220 4.52666 2.50733 0.00692 70.05016 15.4444 5.410 2.653 18.4843 5.644 2.411 19.2367
89 Ac 5.0020 4.56932 2.50607 70.01644 70.03124 15.8710 5.454 2.662 19.0832 5.761 2.594 19.8400
90 Th 5.1823 4.59153 2.50190 70.00815 70.03948 16.3003 5.490 2.696 19.6932 5.883 2.800 20.4721
91 Pa 5.3669 4.63772 2.49776 70.00936 70.03490 16.7331 5.525 2.666 20.3137 5.798 2.534 21.1046
92 U 5.5480 4.65216 2.49170 70.02703 70.01144 17.1663 5.541 2.672 20.9476 5.829 2.571 21.7574
Source: From Orlic et al., 1993. Reprinted with kind permission from Elsevier Science – NL, Sara Burgehartstraat 25, 1055 KV Amsterdam, The Netherlands.
E0 < E < Eedge EM5 < E < EM4 EM4 < E < EM3 EM3 < E < EM2 EM2 < E < EM1
Z E0 n1 n2 n3 n4 EM5 m42 m41 EM4 m32 m31 EM3 m22 m21 EM2 m12 m11 EM1
E0 < E < Eedge EM5 < E < EM4 EM4 < E < EM3 EM3 < E < EM2 EM2 < E < EM1
Z E0 n1 n2 n3 n4 EM5 m42 m41 EM4 m32 m31 EM3 m22 m21 EM2 m12 m11 EM1
69 0.1926 75.0355 9.4382 72.1146 0.1726 1.4677 63.572 726.240 1.5146 5.841 1.257 1.8845 3.889 2.370 2.0998 4.114 2.278 2.3068
70 0.1085 71.8380 6.2849 71.0736 0.0593 1.5278 54.494 722.411 1.5763 5.900 1.222 1.9498 3.916 2.373 2.1730 4.125 2.287 2.3981
71 0.1085 70.3276 4.8522 70.6175 0.0113 1.5885 57.769 724.531 1.6394 6.893 0.681 2.0236 3.876 2.422 2.2635 4.129 2.309 2.4912
72 0.1085 70.1868 4.8402 70.6304 0.0127 1.6517 55.928 724.196 1.7164 7.170 0.527 2.1076 3.916 2.420 2.3654 4.158 2.311 2.6009
73 0.1085 71.9405 6.5977 71.1795 0.0672 1.7351 53.589 723.567 1.7932 6.974 0.627 2.1940 3.957 2.420 2.4687 4.146 2.343 2.7080
74 1.1085 71.1407 5.9151 70.9671 0.0439 1.8092 50.662 722.588 1.8716 6.894 0.671 2.2810 4.019 2.408 2.5749 4.178 2.344 2.8196
75 0.1085 70.3842 5.1929 70.7225 0.0155 1.8829 30.913 712.446 1.9489 6.854 0.666 2.3673 4.063 2.405 2.6816 4.214 2.346 2.9317
76 0.1085 71.8806 6.8129 71.2684 0.0733 1.9601 54.605 725.802 2.0308 6.900 0.581 2.4572 4.032 2.445 2.7922 4.228 2.356 3.0485
77 0.1085 72.9368 7.9106 71.6078 0.1054 2.0404 48.393 722.958 2.1161 7.249 0.343 2.5507 4.126 2.410 2.9087 4.307 2.327 3.1737
78 0.1328 1.5113 3.4177 70.1106 70.0581 2.1216 45.233 721.695 2.2019 7.209 0.344 2.6454 4.166 2.408 3.0265 4.323 2.338 3.2960
79 0.1511 3.5698 1.1260 0.7499 70.1652 2.2057 43.502 721.199 2.2911 7.102 0.393 2.7430 4.199 2.414 3.1478 4.317 2.370 3.4249
80 0.2122 3.6427 1.0962 0.7581 70.1655 2.2949 40.458 719.895 2.3849 6.847 0.540 2.8471 4.187 2.445 3.2785 4.382 2.344 3.5616
81 0.1926 3.6296 1.1014 0.7912 70.1770 2.3893 31.751 715.013 2.4851 6.055 1.064 2.9566 4.191 2.467 3.4157 4.314 2.419 3.7041
82 0.1717 4.6393 70.1055 1.2868 70.2451 2.4840 20.254 78.092 2.5856 5.339 1.574 3.0664 4.259 2.443 3.5542 4.453 2.332 3.8507
83 0.1926 5.7697 71.3933 1.7810 70.3078 2.5796 9.171 71.075 2.6876 4.475 2.212 3.1769 4.277 2.459 3.6963 4.383 2.420 3.9991
84 0.2122 5.6131 71.3279 1.8314 70.3245 2.6839 70.200 5.161 2.7980 3.586 2.895 3.3019 4.327 2.459 3.8541 4.460 2.395 4.1494
85 0.3117 5.1757 70.6452 1.5552 70.2925 2.7867 74.893 8.501 2.9087 3.141 3.276 3.4260 4.372 2.457 4.0080 4.494 2.402 4.3170
86 0.3924 4.2011 0.6074 1.0490 70.2281 2.8924 77.059 10.172 3.0215 3.222 3.251 3.5280 4.382 2.442 4.1590 4.500 2.386 4.4820
87 0.3924 4.5198 0.1763 1.2693 70.2654 2.9999 72.615 7.198 3.1362 3.649 2.961 3.6638 4.405 2.458 4.3270 4.523 2.398 4.6520
88 0.3924 5.5170 71.1906 1.9034 70.3623 3.1049 0.627 4.924 3.2484 3.982 2.716 3.7918 4.391 2.501 4.4895 4.511 2.442 4.8220
89 0.3924 6.3317 72.2638 2.3875 70.4354 3.2190 2.581 3.529 3.3702 4.255 2.525 3.9098 4.460 2.475 4.6560 4.599 2.387 5.0020
90 0.2770 4.1761 0.8164 1.0031 70.2383 3.3320 3.793 2.621 3.4908 4.355 2.455 4.0461 4.495 2.462 4.8304 4.576 2.437 5.1823
91 0.2770 4.3146 0.8409 0.9488 70.2273 3.4418 3.639 2.790 3.6112 4.369 2.484 4.1738 4.526 2.477 5.0009 4.603 2.457 5.3669
92 0.2770 3.9954 1.2801 0.7662 70.2038 3.5517 3.662 2.788 3.7276 4.409 2.460 4.3034 4.527 2.492 5.1822 4.614 2.458 5.5480
Source: From Orlic et al., 1993. Reprinted with kind permission from Elsevier Science – NL, Sara Burgehartstraat 25, 1055 KV Amsterdam, The Netherlands.
B C N O F Ne Na Mg Al
Atomic 6.67 þ 1a 4.48 þ 1 3.16 þ 1 2.36 þ 1 1.83 þ 1 1.46 þ 1 1.19 þ 1 9.89 þ 0 8.34 þ 0
number Element 1.83 7 1 2.77 7 1 3.92 7 1 5.25 7 1 6.77 7 1 8.49 7 1 1.04 þ 0 1.25 þ 0 1.49 þ 0
B C N O F Ne Na Mg Al
Atomic 6.67 þ 1a 4.48 þ 1 3.16 þ 1 2.36 þ 1 1.83 þ 1 1.46 þ 1 1.19 þ 1 9.89 þ 0 8.34 þ 0
number Element 1.83 7 1 2.77 7 1 3.92 7 1 5.25 7 1 6.77 7 1 8.49 7 1 1.04 þ 0 1.25 þ 0 1.49 þ 0
B C N O F Ne Na Mg Al
Atomic 6.67 þ 1a 4.48 þ 1 3.16 þ 1 2.36 þ 1 1.83 þ 1 1.46 þ 1 1.19 þ 1 9.89 þ 0 8.34 þ 0
number Element 1.83 7 1 2.77 7 1 3.92 7 1 5.25 7 1 6.77 7 1 8.49 7 1 1.04 þ 0 1.25 þ 0 1.49 þ 0
Arndt UW, Brown PJ, Colliex C, Cowley JM, Creagh DC, Dolling G, Fink M, Freund AK,
Hilderbrandt R, Howie A, Hubbell JH, Gjønnes, Lander G, Lynch DF, McAuley WJ, Pynn R,
Ross AW, Rowe JM, Sears VF, Spence JCH, Steeds JW, Valvoda V, Willis BTM, Wilson
AJC, Zvyagin BB. In: Wilson AJC, ed. International Tables for Crystallography. Vol. C.
Mathematical, Physical and Chemical Tables. Dordrecht: Kluwer Academic Publishers, 1992,
p 162.
Auger P. J Phys 6:205, 1925.
Bambynek W, Crasemann B, Fink RW, Freund HU, Mark H, Swift CD, Price RE, Venugopala
Rao P. Rev Mod Phys 44:716, 1972; erratum: 46:853, 1974.
Behrens P. Trends Anal Chem 11:218, 1992a.
Behrens P. Trends Anal Chem 11:237, 1992b.
Bethe HA. In: Annalen der Physik, Band 5. Leipzig: Verlag von Johann Ambrosius Barth, 1930,
p 325.
Bethe HA, Ashkin J. Experimental Nuclear Physics, Vol. 1. New York: Wiley, 1953, p 252.
Birks LS. Electron Probe Microanalysis. New York: Wiley–Interscience, 1971a, p 51.
Birks LS. Electron Probe Microanalysis. New York: Wiley–Interscience, 1971b, p 172.
Burhop EHS. The Auger Effect and Other Radiationless Transitions, New York: Cambridge
University Press, 1952.
Burr A. In: Robinson JW, ed. Handbook of Spectroscopy, Vol. I. Cleveland, OH: CRC Press, 1974,
p 24.
Campbell JI, Cookson JA. Nucl Instrum Methods Phys Res B3:185, 1984.
Clark GL. Applied X-Rays, New York: McGraw-Hill, 1955, p 144.
Clark GL. In: Clark GL, ed. Encyclopedia of X-Rays and Gamma Rays. New York: Reinhold
Publishing, 1963, p 1124.
Compton AH. Phys Rev 21:15, 1923a.
Compton AH. Phys Rev 22:409, 1923b.
Compton AH, Allison SK. X-Rays in Theory and Experiment. New York: Van Nostrand, 1935,
p 537.
Condon EU. In: Condon EU, Odishaw H, eds. Handbook of Physics, Part 7. New York: McGraw-
Hill, 1958, p 7.
Davis B, Mitchell DP. Phys Rev 32:331, 1928.
Deslattes RD. Acta Crystallogr A25:89, 1969.
Dirac PAM. The Principles of Quantum Mechanics. 4th ed. Oxford: Clarendon Press=Oxford
University Press, 1958, p 136.
Dyson NA. X-rays in Atomic and Nuclear Physics. London: Longman Group, 1973, p 42.
Evans RD. In: Flügge S, ed. Encyclopedia of Physics, Vol. XXXIV. Berlin: Springer-Verlag, 1958,
p 218.
Evans RD. In: Gray DE, ed. American Institute of Physics Handbook. New York: McGraw-Hill,
1963, p 8.
Fink RW. In: Robinson, JW, ed. Handbook of Spectroscopy, Vol. I. Cleveland, OH: CRC Press,
1974, p 219.
Goldstein JI, Newbury DE, Echlin P, Joy DC, Fiori C, Lifshin E. Scanning Electron Microscopy
and X-ray Microanalysis. New York: Plenum Press, 1981, p 105.
Green M, Cosslett VE. Proc Phys Soc 78:1206, 1961.
Hanke W, Wernisch J, Pöhn C. X-ray Spectrom 14(1):43, 1985.
Hasnain SS, Ed. X-ray Absorption Fine Structure, Proceedings of the Conference Held in York,
August 1990. Chichester: Ellis Horwood, 1991.
Heinrich KFJ. Electron Beam X-Ray Microanalysis. New York: Van Nostrand–Reinhold, l98l,
p 234.
Heitler W. The Quantum Theory of Radiation. 3rd ed. London: Oxford University Press, 1954,
p 207.
Jozef A. Helsen
Catholic University of Leuven, Leuven, Belgium
Andrzej Kuczumow
Lublin Catholic University, Lublin, Poland
I. INTRODUCTION
Moseley’s law, formulated soon after the discovery of x-rays by Röntgen (Röntgen, 1896),
represented the direct and definite onset of the use of x-ray spectrometry in chemistry. The
first successful period was completed in the beginning of the 1920s by creating order in the
periodic table of the elements and by the discovery of the missing elements (Weebs and
Leicester, 1967). X-ray fluorescence (XRF) spectrometry has some unique features so that
it became an irreplaceable tool for the analyst. Not may techniques had such a brilliant
start to their analytical careers!
gThe first commercial instrument became available around 1940 and was derived
from the x-ray goniometer of a diffraction instrument. A reasonable estimation of the
number of wavelength-dispersive (WDXRF) instruments working today is 15000, of
which some 20% are multichannel spectrometers. In addition, about 3000 instruments
working in this mode are attached to the electron microprobes, mainly in Japan. Some
3000 energy-dispersive (EDXRF) instruments are working independently, and some
1500–2000 others are attached to electron and other [e.g., proton (Uda et al., 1987,
Folkmann and Frederiksen, 1990)] microprobes; a few hundred instruments are con-
nected to synchrotrons as excitation sources. About 300 instruments are working now in
total reflection (TRXRF) and several in grazing emission modes (Klockenkämper, 1997).
The average price of a new WDXPF instrument in 1998 is of the order of magnitude of
$250,000, $100,000 for EDXRF, and $200,000 for TRXRF (Claes et al., 1997) and this is
not expected to rise much in the coming years. The calculation of the total money in-
volved in the x-ray instrument market made hereafter was based on the 50% amortization
estimate per instrument. We consider in our account only these parts of Particle-induced
x-ray emission (PIXE), electron microscope, and synchrotron arrangements which can be
treated as x-ray attachments. An auxiliary market was created also around the main
equipment business, covering the spare parts, recent supplements, and service connected
with the maintenance. Then, the commercial value of the instruments running today may
be estimated at some US$4 billion (Markowicz and Van Grieken, 1986; Van Grieken
et al., 1986). An important body of scientific literature relates to x-ray spectrometry
where n is the frequency (in Hz), c is the speed of light (in m=s), l is the wavelength (in m),
k is the constant for a given series, and s is the screening constant. The relationship
Copyright © 2002 Marcel Dekker, Inc.
Figure 1 Phenomena accompanying the interaction of x-rays with matter: (a) different
applications for analytical purposes (abbreviations of methods coupled with the use of particular
phenomena are mentioned; (b) different versions according to the incidence and emergence angles
(I—total reflection XRF or PIXE; II—grazing emission XRF; III—grazing incidence–grazing
emission XRF).
and enhancement coefficients may be balanced such that both effects cancel precisely and
then the diagonal is approximated (curve III).
The effects just discussed are all energy (wavelength) dependent. This means that any
calculations converting intensities into concentration should include the attenuation and
higher-order fluorescence effects and also integrate over all energies present in the exciting
primary beam above the value of the absorption edge and over all fluorescence lines. This
is an ab initio approach for conversion of intensities into concentration implemented by
several authors. We mention here the articles by Gillam and Heal (1952), Sherman (1954,
1955), Shiraiwa and Fujino (1966, 1967), who in the 1950s and the 1960s proposed suitable
equations. Sparks (1976) and Li-Xing (1984) implemented small corrections and refine-
ments concerning the details of these expressions, and de Boer (1990) and de Boer and
Brouwer (1990) gave solutions to some difficult exponential integrals related to the en-
hancement. Important contributions to the numerical side of quantitative XRF are due to
Fernandez (1989), Fernandez and Molinari (1990), Rousseau et al. (1996), and Mantler
(1984). The first two authors prepared commercial, customer-oriented versions of the
numerical programs XRFPc and CiROU, CiLAC, CiLT, and CiREG, respectively, and
Mantler paved the way for the analyses of multilayers. It should be emphasized that
Fernandez solved the transport equations for x-ray beams, the very general and powerful
but difficult tool for the description of particle beams. Likewise, Gardner and Hawthorne
(1975) obtained similar results by Monte Carlo simulation of x-ray excitation. A specific
approach was presented by Dane et al. (1996). These authors solved the problem of
quantification in situations where either an incomplete data set is available or where a
reasonable assumption about composition and=or depth profile cannot be done. Then, the
idea of processing whole sets of solutions (populations or strings or generations) has been
introduced, instead of processing the particular concentration values in consecutive
iterations.
In the fundamental parameter methods, the incidence and takeoff angles are the
important parameters. As a rule, they are kept constant during the traditional run of
the analysis. However, there have been significant efforts to use the angle-resolved version
Copyright © 2002 Marcel Dekker, Inc.
of x-ray XRF spectrometers for depth profiling of the samples (Gries, 1992; Schmitt
et al., 1994).
For a complete treatment of equations deriving the line intensities in x-ray fluor-
escence spectrum from the elemental sample composition, the reader referred is to
Chapter 5.
Figure 6 Overlapping of CKa by higher-order x-rays of Ni, Fe, and Cr. (From Arai, 1987.)
Notation
l (uX) l (mÅ) Element Transition usuelle Ordre E (keV) n=R ðn=RÞ1=2
The main parts of a spectrometer can be represented in the simplest form by Figure 7. The
analyzing crystal is the central point of the wavelength-dispersive instrument. On the left-
hand side of the crystal, we find (1) the source of excitation, (2) the filters and devices for
shaping the exciting radiation (collimators and masks), and (3) the sample. On the right-
hand side, we find (4) devices for shaping the diffracted beam (collimators) and (5) the
detector. Signals from the detector are fed into the electronic circuitry where they are
shaped to be processed by the computer software for data analysis. This arrangement can
be reduced or made more complicated according to the demand. However, all possible
reductions lead to less flexible devices, and complication does not necessarily enhance the
quality of the instrument.
1. Alternative Con¢gurations
To overcome the absorption of the low-wavelength tail of the continuum, a windowless
configuration was considered. This is an obvious solution, but it requires that the whole
spectrometer with sample, collimators, crystals, and flow counter be evacuated to the low
pressure suitable for securing an acceptably long life for the tube filament and conducting
analyses in the required range of the soft x-rays.
Nordfors (1956) advocated a dual-anode tube with two anodes physically separated.
They were excited by deviating the electron beam to either of the two anodes (Fig. 11).
Lack of stability prevented this solution from gaining commercial interest. Probably, a
modified version of this idea was introduced in the series 2000 x-ray spectrometers of
Diano (dual target þ dual filament; see Section VI).
Tubes with exchangeable anodes are another possibility for solving the problem of
comfortable switching between different anodes. To the best of our knowledge, this
solution is not commercially available because of the tedious exchange operation and
difficulties in rigorous repeating of previous conditions of the tube action.
Using the sample as an anode is another solution, but then the application is
bound to conducting materials. In a modified form, this solution is implemented in
scanning electron microprobes. Because this ‘‘anode’’ cannot be adequately cooled, the
input power must be restricted to very low values, entailing low count rates and,
consequently, reducing the precision that can be obtained within reasonable counting
times.
Another possibility is to apply an existing 18-kW rotating anode x-ray tube gene-
rator. Sufficient intensity is obtained. Part of this gain can be sacrificed and Boehme (1987)
and Nichols et al. (1987) introduced the microdiffraction beam collimator into the beam
Copyright © 2002 Marcel Dekker, Inc.
Figure 9 (a) Principle of excitation in a side-window tube. (b) Principle of excitation in a dual-
anode x-ray tube (side window); Mo, W, or Au (for A) and Sc, Rh, or Cr (for B) are commercially
available anode components. The spectral feature of the tube spectrum depends on the energy of the
incident electrons.
derived from such a high-power tube. This collimates the beam to about 30 mm and allows
microfluorescence. Squeezing the beam to more or less the same diameter is possible when
applying the polycapillary lens, even more so with the single capillary. These solutions are
very restrictive for the beam intensity; thus, a reasonable device configuration includes an
energy-dispersive detector instead.
2. Optimization
Because the takeoff angle is a very important parameter, the surface of the anode can be
formed in steps in such a way that the yield of low-wavelength radiation is optimized. As is
clear from Figure 12, the step-shaped surface decreases the escape depths.
In a conventional EWT, the reduction of the anode to window distance t is limited
(Fig. 10). According to an old idea of Thordarson (1939), Botden et al. (1952) developed
an x-ray tube with a gold anode evaporated directly on the beryllium window. The tube
was used for surface radiotherapy. After many years, Phillips returned to Botden’s
construction and again reduced t to zero by sputtering the anode target element directly
on the beryllium window (Fig. 13). In a SWT, the anode is at ground potential, which
Copyright © 2002 Marcel Dekker, Inc.
Figure 10 Principle of end-window x-ray tubes.
Figure 11 X-ray tube with two separate anodes: (1) anodes: (2) filaments; (3) cooling;
(4) separating brass wall; (5) insulator. (After Norfords, 1956.)
considerably facilitates and simplifies the safety requirements. This geometry enhances
the intensity by an order of magnitude, allowing a much lower input power: 200 W for
an intensity equivalent to the earlier 3-kW EWT. As a result of the low input power, the
heat produced in the target by the electron beam is easily dissipated and no supple-
mentary water cooling is required. The tube construction is reduced and only the most
basic components are left. The commercial name for this development is T3 or TTT
(target transmission tube; Fig. 14c). It is used exclusively in simultaneous WDXRF
instruments, especially in the new version of the PW1660 (Philips Technical Materials
Simultaneous X-Ray Spectrometer System PW 1660) (see Section VI).
Copyright © 2002 Marcel Dekker, Inc.
Figure 13 Anode elements deposited on the beryllium window.
3. End-WindowTube
The essential parts of an EWT are displayed in Figure 14a. The ring-shaped cathode and
the cooling circuit of the anode are typical features of EWT construction. The latter
consists of a closed circuit filled with deionized water. Deionized water is necessary to
reduce the conductivity and enhance the safety because the circuit is at anode potential!
Electrons are diverted to the anode surface by electron-optics elements. Because the fila-
ment and window are at equal potential, no electrons hit the window. Heating is mainly by
radiation from the anode, and an external cooling circuit is provided, which may be
connected to normal water supplies because it is not in contact with a voltage source.
Although the construction just described is in a common use now, it cannot be
considered as an ideal source of x-rays. Recently, van Sprang and Bekkers (1998) and
Kuczumow et al. (2000) have discussed the deviations from the homogeneity in the end-
window x-ray tube output. The geometry of the excitation system, as described by an
angle supported on the tube axis – the sample point – the goniometer entrance axis,
deviates from the formal geometry (about 83 ). The ‘‘real,’’ not formal, geometry of the
device can be read out from the shift in the Compton-scatter energy. It creates the
special demands for all the constructors of microprobes and milliprobes. They must
avoid using the conventional end-window tube. There are also the consequences for the
normal analysis: The uniformity of the samples and spinning operation are essential for
getting reproducible results. Otherwise, particular spots on a surface of the sample can
be irradiated in a different manner.
4. Side-WindowTube
An expanded view of a SWT is shown in Figure 14b. The essential differences from an
EWT are the distance between cathode and anode, an earthed anode, and a single cooling
circuit for the anode only. This circuit may be directly connected to the normal water
supply because it is at ground potential. The geometric arrangement of the cathode
supports higher potentials of the cathode with respect to the anode. This is the basic
reason that tube potentials of 100 kV are allowed. Side-window tubes with fine-focus
anodes are used in many microprobe constructions. They are appreciated for their photon
output, easy to transform into a parallel, collimated beam. A consequence of the regulated
anode potentials is the possibility of use of dual anodes. The available dual-anode tubes
are summarized later in Table 10.
in collimators and crystals). The desparate need for more efficient excitation sources
emerges with many more subtle applications of x-rays: the microprobe constructions,
(micro)diffraction experiments with organic or biochemical materials, research on the fine
structure of the absorption edge, focusing of x-rays, or the application of the total
reflection, for example. If one does not have access to the synchrotron, no patience to wait
for the beam time or no money for paying great sums for simple analyses, then the tube
with a rotating anode seems to be the best solution.
Copyright © 2002 Marcel Dekker, Inc.
Figure 14 Continued
7. High-Voltage Generator
A high-voltage generator supplying up to 60 kV for an EWT or up to 100 kV for a SWT
and an output power of 3–4 kW (or much lower, 0.2 kW for TTT tubes) is required. A
special generator is needed if one wants the high-power rotating anode tube. The con-
ventional power supplies were very cumbersome because of the size of the transformer.
The new generation belongs to the so-called switched power supplies. Such a power supply
is basically a dc=dc converter in which a dc voltage is electronically switched at high
frequencies (several kilohertz) and fed into an inductance–capacitance network. The
output power is not continuously regulated but is controlled through pulse-width mod-
ulation. One possible scheme of a switched power supply is given in Figure 15. The mains
are rectified (a), high-frequency switched (b), transformed (c), and rectified and smoothed
(d). The output voltage is sensed (e) and compared to a reference voltage (f), and this
signal monitors the pulse-width modulator (g), which, in turn, commands the switching
circuit (b). In the high-frequency transformer, a ferrite core is used; this is a lightweight
component. The whole setup allows the size of the generator to be reduced to a fraction of
the volume of classical generators.
The whole system is generally monitored by microprocessor. Power switching of the
tube is done preferentially along the isowatt curve (constant input power), all under
microprocessor control. The system has a number of safety switches on flow and tem-
perature controllers of cooling water with microswitches on all panels, which when
Overvoltage U0 is the ratio of the initial electron energy E0 to the excitation energy of
a given shell Eex . The symbols a, b, and d are experimentally determined constants. The
intrinsic merit of this formula is its dependence on the physical parameter U0 and on Z [see
also the work by Murata and Shibahara (1981) with Monte Carlo estimations of the
penetration depth of electrons].
Recently, a series of papers on the numerical description of the tube output has been
published by Ebel et al. (1989) and Schoßmann et al. (1995, 1997). The solutions were
derived from the different depth distribution functions approximating the electron pene-
tration in anode materials.
Apart from all efforts concentrating on the numerical estimation of x-ray tube
emission, experimentally recorded spectra may always be used [collected in Gilfrich and
Birks (1968), Gilfrich et al. (1971), Gilfrich (1974), Brown et al. (1975), and Loomis and
Keith (1976)].
The same kind of calculation as for the spectral x-ray tube output is very helpful
for the description of the background in wavelength-dispersive XRF analysis. The
background is mainly determined by the scattered bremsstrahlung radiation from the
tube. The theoretical estimation of the continuous radiation should always be helpful
for solving background problems (Arai and Omote, 1988; Omote and Arai, 1989; Arai
and Shoji, 1990; Arai, 1991). Sometimes, the knowledge of background can be applied
for analytical aims (Kuczumow et al., 1992). In this case, a selected bremsstrahlung
channel provides quantitative (but not qualitative!) information about the sample
contents.
A similar approach has been applied in an article by Kuczumow et al. (1995) for the
extraction of the information included in the coherently scattered signal in the energy-
dispersive mode on the capillary version of x-ray spectrometry. A similar extraction of the
information from the Rayleigh signal in wavelength-dispersive XRF and from the selected
bremsstrahlung channel in the electron microprobe is possible as well (Kuczumow et al.,
1999). It concerns special samples (e.g., the biological ones) and the information obtained
is connected with the description of the matrix. It brings additional knowledge about the
density of matrix, which is sometimes of essential significance in data evaluation. The
articles just mentioned described the analysis of tree rings and other similar periodic
structures, where the variability of the density of the matrix was, by far, more important
than the variability in the chemical composition.
D. Dispersive Elements
Both the primary photon beam derived from the x-ray tube (or another source of radia-
tion) and the secondary or fluorescence beam derived from the sample can be mono-
chromatized. The monochromatization of the primary beam is not essential but can be
made for more efficient excitation of the selected elements by photons of chosen wave-
lengths or for radical simplification of calculation procedures (the fundamental parameter
calculation is then transformed to simple Lachance–Traill–Tertian correction). Mono-
chromatization may be compared to the action of a narrow bandpass filter by which only a
small band of the whole spectrum is transmitted. The width of the bandpass is related to
the spectral resolution. A number of parameters characterizes this process.
Figure 16 Two spectra from the same sample demonstrate how an auxilliary collimator reduces
line overlap when measuring heavy elements (a) without and (b) with auxillary collimator. (Courtesy
of Phillips Analytical. From Phillips Materials.)
1. Crystals
According to the orientation to the possible crystallographic planes of the crystal, different
values for d are preferentially diffracting and, coupled with it, a different resolution results
Copyright © 2002 Marcel Dekker, Inc.
for a given wavelength. A few values has been calculated for LiF in different orientations
and these are listed in Table 2. From the dispersion, it can be expected that the Ka line of
manganese can be separated from the chromium Kb by LiF (220), not by LiF (200). The
values in Table 2 are calculated for ideal crystals and a parallel beam. For real crystals, the
dispersion and resolving power are less favorable; some real values for a complete spec-
trometer are provided later. In Table 3, a list is presented for crystals currently available
for spectrometers: lithium fluoride (LiF), silicon (Si), germanium (Ge), pyrolytic graphite
(PG), indium antimonide (InSb), pentaerythritol (PE or PET), ethylenediamine-d-tartrate
(EDDT), ammonium dihydrogen phosphate (ADP), thallium hydrogen phosphate
(TlAP), and multilayers with their commercial designations. Important characteristics
other than the interplanar distances codetermine the ultimate usefulness of an analyzing
crystal: spectral resolution, mosaicity, reflectivity, stability, the thermal expansion coeffi-
cient, and the spectral range.
a. Resolution
The spectral resolution DE=E of the crystals oscillates about the value 102, with few
exceptions. The data for the resolution of different crystals are included later in
Figure 26a, a summarizing figure on resolution [see also Kuczumow and Helsen (1989)].
The spectral resolution for the crystals is, generally speaking, better than that of other
dispersive devices, such as detectors or filters, but the difference with respect to the
energy-dispersive spectrometers becomes less favorable for the short wavelengths. The
progress in superconducting tunnel junctions detectors (Finkbeiner et al., 1991; De
Korte, 1992; Le Grand et al., 1993) and microcalorimeter detectors (Lessyna et al., 1993;
McCammon et al., 1993; Silver et al., 1996) during the last decade shows that these
devices would have even better spectral resolution in the medium energy range (i.e.,
above 4–5 keV) than the crystals (see Fig. 26b). However, it is a question of compromise
among price, necessity to work at liquid-helium temperatures, the very value of spectral
resolution necessary for a given analysis, and the loss of intensity in crystals on one
hand and the poor count rates allowed in the newest detectors on the other hand, which
decides what kind of resolving device will be the future choice for commercial appli-
cations.
b. Mosaicity and Reflectivity
Real crystals have all kinds of imperfections, and mosaicity is one of them. Mosaicity
refers to the existence of ‘‘blocks’’ within the crystal with sizes of the order of magnitude
of 100 Å, which have slightly different orientations and lead to widening the diffracted
peaks. This happens to widely differing degress: topaz, EDDT, ADP, and gypsum exhibit
low mosaicity, quartz and LiF a little higher mosaicity, and PET a significantly higher
mosaicity. A result is a slight angular widening of the diffracted peak on one hand, but on
the other hand, the intensity of the peak is generally enhanced. Widening is smaller than
widening due to other geometric factors of the whole spectrometer and the loss can be
tolerated while compensated by the increase in intensity. Some mechanical treatments
enhance the mosaicity, and LiF crystals are easily improved in that way. The angular
intensity distribution of a diffracted peak, the rocking curve, can be determined by double-
crystal spectrometer. In Figure 17, the peak height of the represented curve obtained by a
high-resolution spectrometer is determined by atomic scatter factors and the space group
of the analyzing crystal, whereas the width at half-maximum is determined by the mo-
saicity of the analyzing crystal and the divergence of the incident and diffracted beams.
The surface under the curve is the integral reflection or, when the curve is normalized to
the intensity of the incident beam, the integral reflection coefficient. This coefficient is very
much dependent on mosaicity. Abraded LiF has a 10-fold increase in the reflection
coefficient compared to freshly cleaved LiF (36105 to 46104 rad), measured by using
CuKa radiation (Birks, 1969). LiF crystals have also the great advantage of low ab-
sorption by their constituting atoms. Topaz and quartz, otherwise good crystals, have
Copyright © 2002 Marcel Dekker, Inc.
Figure 17 Rocking curve that would be obtained from a real crystal-diffracting parallel
monochromatic radiation (rocking curves are actually measured on double-crystal spectrometers):
P ¼ the peak diffraction coefficient; R ¼ the integral reflection coefficient. (From Birks, 1969.)
poor reflectivity properties, whereas PET is very good in this respect. Reflection constants
as a function of wavelength are given in Figure 18.
c. Stability and Temperature
The mechanical stability of most crystals is satisfactory, but there are exceptions. Gypsum
can effloresce (especially in a high vacuum); PET has a tendency to change phases on aging
Figure18 Single-crystal integral reflection coefficients of graphite, LiF (200), LiF (220), and KAP.
[From Gilfrich et al., 1971). Reprinted by permission of Analytical Chemistry. Copyright #
American Chemical Society.]
that of conventional flat geometry). The curved crystal geometry is applied in simultaneous
instruments in which scanning is not used and an optimized arrangement is chosen for each
wavelength (element) implemented. Then, logarithmically curved crystals are used. In a
new version of the sequential instrument (Venus), Philips used the reverted channels taken
from the simultaneous instrument. Crystals applied in that construction are curved as in a
simultaneous device and their position is additionally adjusted by a screw to trap as much
intensity as possible. In general, during the last two decades, great progress took place in
the planning and construction of the optical elements based on curved reflecting or dif-
fracting surfaces of crystals and multilayers; most of the developments were in the syn-
chrotron version of x-ray spectroscopy (Ice and Sparks, 1984).
The curved crystal version of the x-ray microprobe is not the only known solution. The
progress in x-ray optics during last decades enabled the easy concentration of x-rays while
being transmitted through a single capillary (Rindby, 1986; Carpenter 1989; Thiel et al.,
1989) or even through a semilens based on the bundle of capillaries (Kumakhov and Ko-
marov, 1990; Yiming et al., 1994; Dagabov et al., 1995; Xunliang et al., 1995; X-Ray Optical
Systems Inc., 1997, 1998). This first possibility, due to the great loss of intensity resulting
from a poor interception of the primary beam by the narrow inlet of a single capillary, is
limited to the energy-dispersive version of x-ray spectrometry (Attaelmanan et al., 1994).
The second version is more interesting from the point of view of WDXRF. A properly
curved capillary bundle can serve as a kind of semifocusing lens with a well-defined
Copyright © 2002 Marcel Dekker, Inc.
Figure 21 Cross section of a typical multicapillary ‘‘lens’’: (a) focusing configuration, from wide
spot (divergent source) to another spot (semifocus); (b) configuration transforming the divergent
into the parallel beam. (Courtesy of X-Ray Optical Systems Inc., Albany, NY. and personally from
Dr. J. P. Bly.)
curvature and moderately-defined focus (Fig. 21). The intensity of the beam leaving the
‘‘lens’’ is, of course, lower than the intensity derived from the x-ray tube, but still sufficient
for many applications. Moreover, the primary collimator is removed in this system and the
gain from this removal is partially balancing the loss resulting from the bundle construction
and operation. One can estimate the focal distance f of such a lens as
df ¼ d0 þ 2 f ycr ð8Þ
where d is the spot size, d0 is the diameter of the capillary end, and ycr is the critical angle
of the total reflection. As a rule, the polycapillary array is introduced to concentrate the
primary beam, but it can also be used for the collection of secondary x-rays outgoing from
the sample, before reaching the curved crystals in nonconventional arrangements.
2. Multilayers
The limitation of XRF analysis in the region of soft x-rays is one of the most serious
limitations of crystal monochromators, although none of the ‘‘first principles’’ constituting
the background of XRF method puts this limitation explicitly. Moseley’s law determines
the rules for qualitative analysis of any element except hydrogen and helium, and the
Shiraiwa and Fujino equations govern quantitative analysis. Irresolvable problems oc-
curred until the late eighties because of troublesome absorption, lack of an adequate
dispersive device, and bad detection while passing to longer wavelengths. The concept of
multilayers or, strictly speaking, layered synthetic microstructures (LSMs) or multilayer
interference mirrors (MIMs) solved at least part of the problems.
Two approaches to the idea of a suitable analyzing device in the soft x-ray domain
have been applied from the very beginning (Underwood and Barbee, 1981). The first re-
sulted from consideration of Bragg’s equation [Eq. (2)]. This indicates that if any crystal of
a longer interplanar distance is found, it will be possible to disperse longer wavelengths in
the soft x-ray range. The potential candidates for long-period crystals or quasicrystals,
such as clay minerals of the chlorite (14 Å) or illite–montmorillonite (25–30 Å) type, are
not sufficiently ordered in this respect (Weaver and Pollard, 1973). Unfortunately, natural
crystals, irrespective of their different crystallographic planes, offer only limited possibi-
lities in this spectral range; specifically, their reflectivity is very low. The class of structures
called Langmuir–Blodgett films was more promising, but the reflectivity and thermal,
mechanical, and chemical stabilities were rather poor.
Copyright © 2002 Marcel Dekker, Inc.
A correct solution resulted from generalization of Bragg’s law, namely that dif-
fraction also occurs in media consisting of layers of different refraction coefficients—in
other words, in sites of different periodically changing electron densities. Such structures
may be formed in an artificial way. They may be treated as crystallike substances ordered
along the c axis (i.e., spacing d or 2d), but with uncontrolled arrangement within the
ab plane. The structure in the ab plane can be amorphous, provided that the optical and
mechanical properties are uniform in these directions.
The first effort of synthesis was made in 1940, but the Au–Cu multilayer of Du
Mond and Youtz (1940) did not survive more than a few days because of deterioration by
interdiffusion. The problems of thermal, radiative, diffusional, and chemical instabilities
of multilayers have recently been solved. Significant progress in this field has taken place
in the last three decades, hence some years before the application of multilayers in
WDXRF.
The second approach was the result of an increasing interest in x-ray optics. As a
branch of physics, x-ray optics is nearly as old as the discovery by Röntgen, but because of
the specific properties of x-rays, its progress has been slow. All existing substances exhibit
complex refractive indices:
m ¼ 1 d ib ð9Þ
where m is the refraction index, d is the so-called unit decreament, a real number, and the
imaginary term b is related to absorption. The refraction indices in the x-ray region are all
below but very close to 1. This makes all efforts of x-ray focusing by lenses in a traditional
sense an impractical task. Franks (1977) gives an example of a lens with a curvature radius
of 10 mm built of a material with d ¼ 5 105 . The focal length of this lens is 100 m. In
such a situation, only glancing, grazing, and diffracting features of materials have been of
practical interest for a long time. Otherwise, the grazing optics conditions for x-rays are
extremely severe. However, the need for mirrors, monochromators, and focusing devices
for the x-ray region is increasing rapidly in the fields of synchrotron radiation, plasma
research and diagnostics, x-ray microscopy, x-ray fluorescence analysis, diffraction, thermo-
nuclear fusion, x-ray laser, x-ray astronomy, and even x-ray waveguides. The invention of
multilayers defined this time as a complex set of particular layers consisting of materials of
periodically different refractive indices gave great impetus to progress in this branch of
optics. Some important properties of the materials used to produce the multilayers are
discussed in the next section.
a. Nature of Materials
In principle, three kinds of material are necessary for the synthesis of multilayers. First,
there must be a support (sometimes called a substrate), such as a piece of flat or curved
silica layer, with an ideally polished surface. This condition is very important because the
roughness of the surface is translated to the deposited layers, correctable almost ex-
clusively when one of the layers consists of amorphous materials such as carbon. After
years of progress in the field, when the roughness of the interfaces have probably reached
the limit of about 2 Å, the most important contribution in it is from the deviations of the
substrate from flatness (Kortright, 1996). Two sources of relatively inexpensive and pre-
cisely polished supports were found: silicon wafers as used in microelectronics (curved)
and glass or fused-silica optical elements for low-scatter mirrors (flat or curved). The
additional polishing is the last step in the preparation of the substrate for further manu-
facturing of the multilayer. An even more complicated role for the substrate is described
by Nicolosi et al. (1986) in whole coal analysis. The authors determined the contents of
Copyright © 2002 Marcel Dekker, Inc.
C, N, O, and F using multilayered microstructure PX2 and the contents of P, S, K, and Ca
by taking advantage of diffracting properties of the supporting wafer of silicon (111) on
which the multilayer was settled.
Heavy metals (Hf, Pb, W, Mo, Pt, Ir, Ru, Rh, Os, Ni, Co, Fe, and Cr) and light
elements or compounds [Al, B, B4C (Ovonic Synthetic Materials Co., 1987), BN, Be, C, Si,
SiO2, SiC, Ti and Sc—the last with the high reflection] may be used as materials for al-
ternating high and low optical densities, at the same time being strong and weak scatterers,
respectively. Sometimes, materials from the first group are called absorbers, and those
belonging to the second group are called spacers. The reactive materials and substances
with a low melting point are excluded from the list. Both components of the multilayer
should not share a common crystalographic structure, to avoid the epitaxial growth. The
selection rule for the materials can easily be understood from Figure 22 for boron. The
mass attenuation coefficients for the boron K line, as presented versus the atomic number
of the absorber, have several minima. One falls in the region of boron (weak self-ab-
sorption) and carbon, and the next very deep minimum is in the region of nio-
bium=molybdenum. It puts the constraints on the materials, which can be used for the
determination of boron. The first, a low-density material, can be carbon or even better
B4C; the second, a high-density material, should be niobium or molybdenum. Indeed,
Philips made the structure PX3, composed of Mo=B4C, with a period of 195 Å, adviced
for the determination of boron. However, when a new multilayer should be synthesized
Figure 22 Mass attenuation coefficient for the BKa line drawn against the possible absorbers in
the periodic table. Arrows show the deep minima in coefficient value. (Courtesy of Bruno Vrebos’
inspiration.)
b. Smoothness
This parameter is of great importance on each level, starting from the supporting
substrate to the last upper layer. All rough places, uneven layers, disturbances in in-
terfaces, interdiffusion, and chemical interaction regions or defects, act in a manner
analogous to the thermal motions in classical diffraction analysis. During synthesis, the
extent of roughness can vary in a different manner: In some cases, it tends to
smoothness, in others to further growth. The use of amorphous materials to form some
layers favors smoothing. The greater interplanar distances are, the more pronounced
the roughness is. On the other hand, the smoothness probably has its limits, even in
the case of multilayers with relatively small interplanar distances, with the Debye–
Wallner parameter s tending to reach a value of about 2 Å. It probably results from
the smoothness level obtained during the substrate production. Finally, it puts some
domination of the Debye–Wallner parameter over other possible factors influencing the
quality of multilayers in the region where the interplanar distance is not very long (say,
a few tens of angstroms).
where s is the width of a smooth transition layer and a is the propagation angle of
electromagnetic radiation, measured from the normal to the boundary between a medium
with one refractive index to another.
The question arises, however, whether the errors result from a rapid or from a mild
change of the optical features on the boundaries. Any roughness of a given layer decreases
its reflectivity (Payne and Clemens, 1993), but a random distribution of roughness may
bear higher reflectivities of the higher-order peaks. Moreover, many other physical factors
can influence the uniformity and expected layer characteristics: Both the support and the
particular monolayers can have different slopes, densities are subject to fluctuations, the
elements can diffuse, or some chemical elements can mutually react (e.g., W with C or Mo
with Si) (Rosen et al., 1993). It should be emphasized that despite of so many obstacles,
the magnitude of the roughness in such a totally artificial structure can be kept in the range
of an atom radius (Barbee, 1986).
The numerical description of multilayers is based on the matrix method (a matrix
expresses the features of a layer or interlayer region, and for the characteristics of a
multilayer, the product of matrices is written) or on the recursive use of layer–reflection–
transmission characteristics. Deviations of empirical results (mainly reflectivity measure-
ments) from the model predictions are associated with roughness. In this case, seff, the
roughness parameter, results from the best fit of empirical data to the model.
The selection of an adequate method of synthesis and subsequent control of pro-
duction and quality is a key question. Many methods were checked, and the aim of all
these processes was the deposition in which one atom follows another. The physical vapor
condensation, chemical vapor deposition, electrochemical covering, and sequential ad-
sorption of films are the most promising methods. However, for the time being, the first
method prevails. Two basic versions of physical vapor deposition are used—with thermal
and sputter sources. In the thermal source technique, the covering materials are evapo-
rated by electron beam or laser evaporation. The vapor source and samples are in a fixed
position. The time of evaporation is a parameter controlled by shutters or by pulsed laser
irradiation. An in situ monitoring system was used as a feedback system for controlling the
shutter. Spiller (1981) invented a system for thermal deposition. He used a soft x-ray re-
flectometer (a source plus a detector), placed in an evaporation chamber to register the
reflectivity of successive layers. The open loop is set according to the indications of
maximum reflectivity for a given layer.
It is possible by this method to obtain a multilayer with an area up to 25 cm2.
The rates of deposition vary in the range of 2–100 Å=s and must be carefully con-
trolled because the successive layers must be commensurate. Sometimes, especially
Copyright © 2002 Marcel Dekker, Inc.
when multilayers of large surface are produced, a special motion of the substrate in the
evaporation chamber may be necessary to guarantee the same evaporation rate for all
points of the surface. Other difficulties arise from this type of deposition: the high
vacuum required in the evaporation chamber (< 102 Pa), radiation heating of the
deposited film, and the problems with thermal sources (alloys may segregate during
melting).
The sputter source systems (Barbee, 1981) belong to another processing method. The
glow discharge, magnetron, electron cyclotron, ion beam, and triode devices may serve as
sputter sources. The plasma discharge is always the first step of the process. The plasma
flux is directed by the electrical potential difference to the cathode surface, where the ions
and atoms dislodge the maternal atoms. The secondary atoms and ions establish the
proper flux (Sigmund, 1981), which passes to the substrate to deposit a layer on it. Samples
are placed on a special rotatory table, appearing at fixed intervals under the secondary flux
courses. The rotation speed, precisely controlled, is the regulating factor for a designated
thickness of the layer. The purity of secondary fluxes and their energies also influence the
quality of the multilayer. The main errors in this method result from the ion and atom
inclusion into layer (there are high-energy tails in the fluxes) and from heating caused by
flux energy deposition in thin layers. However, substrate rotation may overcome many of
these defects. Sputter rates range from 1 to 20 Å=s; the covered surfaces may be quite
large. Today, the magnetron sputtering is the most common method, with the best results
obtained as far as the reflectivity of the products is concerned.
After presenting two basic methods of layer formation, some attention must be paid
to methods of control. As mentioned earlier, reflectivity estimations are of fundamental
importance in multilayer quality determination, both experimentally and in theory
(Spiller, 1981). A comparison of reflectivity predicted and obtained in an experiment is
instructive, allowing, at least, understanding the source of the roughness (Henke et al.,
1986; Luck et al., 1992). Different peak, absolute, and integrated reflectivities were mea-
sured with the use of both laboratory x-ray sources and synchrotron facilities. These re-
sults confirmed the layered structures of the synthesized materials; as a rule, reflectivity
was smaller than predicted, but the results show that the optical parameters estimated
until now (Henke et al., 1982, 1988) are reasonably correct, with the shorter the wave-
length concerned, the better.
The methods just discussed only allow the deduction of an indirect image of the
multilayer structure. Other methods give direct insight into the structure. Thus, the dif-
fraction data may be very useful, but they require large sample volumes. Progress in ca-
pillary version of the diffractometer (Bilderback et al., 1994), allowing the observation of
the samples as small as 56103 mm3, may bring a solution to the structural problems.
Very small samples are needed for obtaining the structural information by electron mi-
croscopy, especially in combination with a special technique of sample preparation, so-
called microcleavage (Lepêtre et al., 1986). Wedge-shaped strips of materials are placed in
the electron beam in such a manner that the planes of particular layers are parallel to the
beam direction. Both the transmission image and the diffraction pattern can be obtained.
The in-depth and lateral variations, the roughness, and the visual image of the structure
are in the field of observation. High-resolution electron microscopy can give even better
results, and electron energy loss spectroscopy (EELS) may provide the complementary
chemical information on the subsequent layers (Lepêtre et al., 1986).
Requirements for the application of multilayers in XRF may be summarized as
follows: They should satisfy Bragg’s law for the assumed wavelength domain within
permissible goniometer angles, have a high reflectivity and a good spectral resolution,
Copyright © 2002 Marcel Dekker, Inc.
suppress diffraction peaks of higher order, and do not absorb soft radiation too in-
tensively. Flat or curved multilayers can be used. Dedicated designs are described in the
literature for the application of curved multilayer optics (Van Eenbergen and Volbert,
1987; Gilfrich et al., 1982). The geometric requirements imposed on the curvature of
multilayers for XRF are modest in comparison with their applications in other fields of
x-ray optics. The increasing spectral distance between K lines of the neighboring elements
when going to lower-Z elements may require a dedicated multilayer for each element.
Sometimes, great differences in the mass attenuation coefficients for the lines of neigh-
bouring elements of interest exist for a given material—attenuator. Then, one can con-
struct a multilayer efficiently diffracting the radiation of one element while deeply
suppressing it for other elements. The dedicated multilayers are the preferable choice for
the simultaneous WDXRF instrument, with an optimized channel for each element.
Multilayers can be also applied in a special version of WDXRF analysis, called grazing
emission x-ray analysis (de Bokx and Urbach, 1995). Curved multilayers are also used in
electron microprobes (Ovonic Synthetic Materials Co., 1987).
From what has just been described, it may be concluded that in order to analyze a
useful range of soft x-rays, at least three different multilayers have to be selected. Such
combinations are currently available commercially (e.g., for example from Osmic, Philips,
Siemens, Rigaku, and Shimadzu) (Ovonic Synthetic Materials Co., 1987; Van Eenbergen
and Volbert, 1987; D. K. G. de Boer, personal communication, 1998):
PX1 2d ¼ 49:3 Å oxygen to magnesium
PX3 2d ¼ 195 Å boron and possibly for beryllium (Nicolosi et al., 1987)
PX4 2d ¼ 122 Å carbon
PX5 2d ¼ 112 Å nitrogen
PX6 2d ¼ 300 Å beryllium
and from the Ovonic Synthetic Materials Company (1987):
OV-040A 2d ¼ 40 Å fluorine to silicon
with multilayers changing 2d spacing every 20 Å up to
OV-140B 2d ¼ 140 Å carbon to oxygen
and also the MoB4C structure:
OV-H series 2d ¼ 244 Å beryllium to boron
up to
OV-300H 2d ¼ 300 Å
Probably, materials with even greater interplanar distances can also be supplied. However,
recent scientific (except the XRF field) interest in multilayers is centered, instead, more on
multilayers with periods of 10–20 Å for their excellent reflectivity or for their polarizing
abilities (Schäfers et al., 1998).
These materials are described in the current literature as completely thermally stable;
thus, they do not need temperature stabilization when used in x-ray spectrometers
(Nicolosi et al., 1986; Barbee, 1986; Van Eenbergen and Volbert, 1987). Some sensitivity
to the damaging influence of intense radiation [above 0.15 J=cm2 (Kohler et al., 1985)],
confirmed later in an article by Kortright et al. (1991) and by a comprehensive report by
MacGowan et al. (1993), may limit the application of the multilayer for mono-
chromatization of dense flux of x-rays, which usually happens in synchrotron, plasma,
Copyright © 2002 Marcel Dekker, Inc.
or laser research. Perhaps, the more important conclusion from the latter study is that,
after the impact of high power on multilayers, the amorphous carbon layer transforms
partially into a graphite form. It improves the reflectivity but greatly deteriorates the
stability and changes the interplanar distances. The phenomenon of ultrafast graphitiza-
tion ( 40 ps) can, in the future, be applied for the construction of x-ray switches. These
data can be compared with the results cited in a report by Barbee (1986). He claims that
during the sputtering process of multilayer synthesis, very high temporary doses of energy
can be deposited in the layers (power density ffi 105 W=cm3=s1), which involves a sig-
nificant increase in local temperatures (50–180 C).
The relative spectral resolution of dispersive devices is often expressed in the form of
the important parameter DE=E or Dl=l, which is the basis for their choice in XRF ap-
plications. Unfortunately, this is the weakest point of the multilayer. In a commercially
available specimen, the relative spectral resolution changes from about 0.025 to over 0.1
while passing from structures with 2d ¼ 40 Å to those with 2d ¼ 244 Å, respectively, for
different measured lines (Ovonic Synthetic Materials Co., 1987), and for the structure with
2d ¼ 75 Å, it changes from 0.025 to 0.04 in the wavelength range 7–75 Å (Henke et al.,
1988), which is about three times worse (Ovonic Synthetic Materials Co., 1987) or even
more (Henke et al., 1986) than for a good classical pseudocrystal. The comparison of
spectra of the same sample by using multilayers with different spacings, as in Figure 23, is
very instructive (Van Eerbergen and Volbert, 1987). Relatively broad internal bandpasses
of multilayers diminish the resolution of the whole spectrographic device, seriously in-
creasing the detection limits. This is of special importance to the analysis of the L or M
series, and avoids the efforts to search the chemical shifts in x-ray spectra [although see
Habulibaz et al. (1996), where chemical shift is investigated for the SiL spectrum, dif-
fracted on a multilayer with an interplanar distance 300 Å]. The relatively wide bandpass
of the multilayer can be used in a constructive way. As proven by Vrebos (personal
communication, 1997; see Fig. 24), the imminent width of the BKa peak is so great that
the contribution from the secondary beam collimation to the total width of the signal is
negligible. In that situation, the analyst can remove the fine or medium collimation or
Figure 23 Comparison of sensitivity and resolution for layered synthetic microstructures with 2d
values of 5, 12, and 16 nm from bottom to top, respectively. Increased sensitivity together with
decreased resolution on the drop in 2d is evident. (From Van Eenbergen and Volbert, 1987.)
even the collimation completely. The gain is double. It results from the greater intensity
given by the multilayer as such, as well as from the lack of the collimator, which is always
one of the most intensity-restrictive parts of the WD spectrometer. This kind of work with
a limited collimation is possible only in the region of soft x-rays, where the population of
lines is rare and in the absence of the very soft L and M lines of heavier elements.
The suppression of higher-order diffraction peaks gives better spectral purity and
may at least partially compensate for the poorer resolution of multilayers. The isolation of
the weak MgKa line from the dominant CaKa line is possible by the use of PX1 structure
for the analysis of cement (Fig. 25). This is an extreme example of the superiority of
multilayers over a classical crystal in particular cases (Nicolosi et al., 1986). Another great
advantage of multilayers is their great reflectivity. It should be added that the peak re-
flectivity is 3–10 times greater than that of corresponding classical crystals.
Multilayers were frequently used in the routine determination of light elements.
Many examples can be cited, including carbon (Nicolosi et al., 1986, 1987; Van Eenbergen
and Volbert, 1987; Philips Application Note No. 745) and boron determination (Van
Eenbergen and Volbert, 1987; Nicolosi et al., 1987; Adamson et al., 1991; Philips Appli-
cation Note No. 737 and No. 803; Uhlig and Müller, 1991; van Sprang and Bekkers,
1998), and beryllium detection (Nicolosi et al., 1987; Siemens Lab Report X-ray Analysis,
1994; Anonymous article, The Rigaku J., 1997). In the analysis, advocated by Rigaku, the
application of the high-power 4-kW generator was the additional factor, enhancing the
detection power for beryllium determination. The detection limit for that element was
found to be in the region of a few tenths of a percent. Determinations of light elements
were made using both flat optics sequential spectrometers and curved optics in simulta-
neous instruments (Van Eenbergen and Volbert, 1987). After overcoming the technical
difficulties, the results obtained from such analyses may seem quite trivial (Nicolosi et al.,
1987; Bonvin et al., 1995). However, not always are things going in that usual way. Van
Sprang and Bekkers, (1998) emphasized the number of problems, coupled with light-ele-
Copyright © 2002 Marcel Dekker, Inc.
Figure 25 2W scans for Mg in cement using TlAP and PX1. (From Nicolosi et al., 1986.)
ment determination. For the analysis of boron in glass, only the combination of methods
gave good results in a whole range of the detectable concentrations: The fundamental
parameter method using the fluorescent signals for the low B concentrations and the use of
the incoherently scattered radiation as a measure of the light-element fraction above 4%.
One should note some special features of x-ray spectrometry in the soft x-ray range: The
type of analysis changes from bulk to surface measurements; the influence of the chemical
state and the subsequent wavelength shifts on the analysis can be significant; the wave-
length of soft x-rays is comparable to the surface roughness; and the problem of the
surface quality becomes extremely important.
Considering the so-called total information depth of x-rays [see Sec.IV.G.1, Eq. (19)],
it can be calculated that in the case of iron determination in a steel sample, the effective
energy of exciting photons being 20 keV, the secondary x-rays arriving at the detector
emerge from depths up to 62 mm. A carbon signal, emitted by a coal sample excited in the
same manner, emerges from depths of only 8.3 mm. The problem looks even more dra-
matic if carbon is determined in a steel sample: In this case, signals emerge from depths as
small as 0.45 mm! This means that the analytical volume for iron is about 140 times greater
than that for carbon and the information on both constituents really refers to totally
differing volumes of the steel sample. Moreover, the method of excitation is also critical in
this case: The excitation of light elements is most efficient if made with relatively
soft x-rays, the energy of which does not substantially exceed the value of the absorption
edge. Excitation by use of hard radiation is not profitable in this case, as photons are
mainly stopped inside the sample. The most restricting stage is the very short escape path
of the characteristic photons for low-Z elements.
The effect of the chemical state can be observed during the analysis of some elements
using soft x-rays. The wavelengths of La1,2 lines of chromium and manganese are 0.573
and 0.637 Å, respectively, and should be analyzed in the region otherwise reserved for the
Copyright © 2002 Marcel Dekker, Inc.
K series of oxygen and fluorine, hence by the use of the PX-1, PX-4 or OV-120B pseudo-
crystals. The question is whether the limited spectral resolution of multilayers is sufficient
to show such subtle effects as chemical shifts and to separate the mentioned lines of metals
from the disturbing lines of oxygen or fluorine if present in the sample.
The wavelengths of characteristic x-rays in the K series are longer than atomic radii
starting from Ga (GaKa ¼ 1.337 Å; atomic radius of Ga ¼ 1.26 Å) and passing to the
lighter elements. Radiation with a wavelength comparable to the dimensions of an ob-
stacle always result in surface problems, with scattering and diffraction at the head. One
should take into account that even the most perfectly polished sample has a surface
roughness of the order of at least one atomic radius and that the best layers in multilayer
structures have a roughness, expressed as the effective roughness seff, on the order of one
to several atomic radii (Barbee, 1986; Kortright, 1996).
These latter remarks are not intended to weaken interest in multilayer optics and its
application in XRF, but only to evoke a special awareness of potential problems.
3. Spectral Resolution
Figure 26a shows the features of commonly used dispersive devise (Heinrich, 1981;
Caciuffo et al., 1987; Plotnikow and Pszenicznyj, 1973; Fitzgerald and Gantzel, 1971;
Burkhalter and Campbell, 1967; Gohshi et al., 1982a, 1982b; Potts et al., 1985; Sparks,
1980; Salem and Lee, 1976; Bent, 1970; Bandas et al., 1978; Gilfrich, 1987). The width of
the bandpass DE (expressed as FWHM whenever possible) is plotted versus the energy of
radiation E; the parameter is relative to the spectral resolution DE=E. A line of given
value of DE=E is called the isoresolving line. These lines divide the whole figure into two
zones of different resolutions (or bandpasses). Note that both the typical dispersing
devices and typical detectors are collected in Figure 26a. Only a few detecting devices
with no resolving power (the reciprocal of the relative spectral resolution) at all exist
(e.g., Geiger-Müller counter), and no detector exists with perfect, infinite resolving
power. Real dispersive devices exhibit very contrasting features, with the parameter
DE=E varying from almost 1 (scintillator counter or semiconductor photocathode) to
systems with DE=E as small as 104 [Ge (111) crystals and two- or three-crystal spec-
trometers]. Even better dispersive devices are needed, for example, for x-ray inelastic
scattering spectroscopy in which relative energy resolution near 107 is wanted. Indeed,
we can cite here some exemplary articles announcing the achievement of similar values:
1:14 107 at a level of 14:4 keV (Chumakov et al., 1996) or 2 108 for 25:7 keV
(Verbeni et al., 1996). XRF spectrometry has rather modest demands for resolution. We
see that for introducing the characteristics of the devices with spectral resolution on a
level of 108 in our diagram, the scale of Figure 26a should be prolonged by approxi-
mately four orders of magnitude at the bottom!
Another interesting conclusion can be drawn from Figure 26a. Nearly all dispersive
structures exhibit variable relative spectral resolution, but only KAP, topaz, and Ge
(111) crystals and, to some extent, also multilayer structures can be considered as iso-
resolving systems. The microcalorimeters show rather unusual behavior, compared with
other detectors. They have more or less the same absolute spectral resolution in the
whole useful range (2–20 keV). As we can see from the shape of the DK and DL curves,
nature does not demand isoresolving behavior from the dispersive structures. Rather,
these structures, with resolution curves parallel to the curves mentioned, would be op-
timal for spectrometric aims: The Si–Li detector seems to have the better features.
Generally speaking, a bandpass range of 101–103 is considered analytically useful.
Copyright © 2002 Marcel Dekker, Inc.
Figure 26 (a) Spectral resolution of different devices: DZ ¼ spectral difference between analogous
lines of adjacent elements; DW ¼ width of spectral line; Ka1 7 Ka2 ¼ difference between Ka1 and Ka2
lines; (b) the imminent features of x-ray spectra extracted from (a) for better characterization of the
analytical demands; (c) characteristics of new microcalorimeter detectors as compared with typical
Si–Li and LiF (200) dispersive structures.
Systems with bandpasses below 103 are used only in spectrophysical work. For making
a proper choice of the dispersive structure for study of chemical shifts, one has to take
into account the curves DW and Ka1 7 Ka2 (or even La1 7 La2) from Figure 26b. Figure
26b shows the spectral characteristics, extracted from Figure 26a, as a better demon-
stration of the immanent features of x-ray spectra. Depending on what matters in
analysis, the adequate choice of the resolving system should be made. The regions
occupied in the diagram by the most common [SiLi and crystal (e.g., LiF)] or most
promising (microcalorimeter) detectors are shown in Figure 26c for easier comparison
with data from Figure 26b. Figure 27 represents some example of the parallel analysis of
the same sample, based on spectra from the electron microprobe acquired by the
WDXRF system (TLAP), EDXRF system with Si(Li), and EDXRF system with the
Copyright © 2002 Marcel Dekker, Inc.
Figure 26 Continued
microcalorimeter. For the time being, some preponderance of the LiF detector is ob-
served for light element determinations (below 4 keV) (Wollman et al., 1997), but even
this was endangered when Wollman announced the energy resolution of 2.0þ=0.1 eV
at 1.5 keV level (Wollman et al., 2000).
Spectral resolution is only one parameter determining the selection of a spectro-
meter. Other parameters, however, such as speed of measurement, count rate capability,
geometrical collection efficiency, and ease of operation may affect the choice of the most
suitable instrument.
E. The Goniometer
The goniometer is basically a very simple construction and is the analog of a dif-
fractometer as used in x-ray diffraction (the crystallographic research preceded the x-ray
fluorescence method). The essential features are shown in Figure 7. The sample is posi-
tioned in front of the tube window. To account for possible heterogeneities, the sample is
Copyright © 2002 Marcel Dekker, Inc.
Figure 26 Continued
spun about its axis at 30 rotations per minute. The analyzer crystal is positioned in the
center of the diffractometer circle, rotates around an axis perpendicular to the plane of the
drawing, and passes through the macroscopic plane of the crystal. The detector moves
along the diffractometer circle supported by the goniometer arm. As imposed by the Bragg
condition, the crystal rotates over an angle 2W, realizing equal angles of incidence and
diffraction with respect to the crystal plane. The crystal is diffracting the incident fluor-
escent beam of the opening angle determined by the slit width and length of the collimator.
Rotation of crystal and detector are either mechanically coupled through gears with a
ratio of 1 : 2 or mechanically decoupled but moved by separate computer-controlled
stepping motors, but both arrangements are still implemented (cf. Tables 8–10 in Sec. VI).
The idea of mechanical decoupling and optical position control was first introduced by
Applied Research Laboratories (see Sec. VI). This technique offers a number of ad-
vantages:
Alignment of crystals is no longer necessary. They can be positioned on a known
spectral line and the offsets are memorized by the computer.
Copyright © 2002 Marcel Dekker, Inc.
Figure 27 (a) NIST microcalorimeter spectrum of AlGaAs (solid line) compared to parallelly
determined EDS spectrum [with commercial Si(Li) detector]; (b) WDS spectrum of the same sample
(with TLAP applied as a crystal), spectrum transformed from wavelength into energy representation.
Observe still a little better performance of the WDS spectrometer for the considered energy region
1–1.6 keV. (From Wollman et al., 1997.)
The detectors can be juxtaposed on the goniometer arm and brought into position as
required by the program. Tandem arrangement of flow and scintillation is no
longer necessary.
Slewing and scanning speed range is virtually unlimited.
Optical position control (Moiré fringes or optical encoder) allows very high angular
precision and accuracy.
Low-energy radiation is strongly attenuated by air. The Ka radiation of a pure
copper sample (sample 2’’, measured on a Philips 1410 sequential spectrometer, SWT
with chromium anode, 45 kV and 60 mA) produces 67800, 33200, and 2660 cps, respec-
tively, in vacuum and 48000, 16800, and 890 cps, respectively, in air. The results mentioned
are not easily repeated on recent computer-monitored instruments (e.g., Philips 2400)
which do not routinely allow working under air at normal pressure. To decrease the ab-
sorption of photons, all wide-range spectrometers operate in a controlled atmosphere
(vacuum or helium). The vacuum cabinet contains the entire spectrometer, except for some
instruments such as the scintillator detector. Attenuation of the high-energy radiation is
low, and to reduce the size of the vacuum chamber, the scintillation detector was and still
is placed by some producers outside the chamber. A long rectangular window of a me-
chanically strong low-density polymer (Mylar) along the diffractometer circle allows the
high-energy photons to reach the detector.
High angular precision and accuracy is required for positioning the goniometer arm;
therefore, severe mechanical constraints are imposed on its construction. Typical is an
Copyright © 2002 Marcel Dekker, Inc.
angular reproducibility of 0.001 or less and a mechanical resolution of a few thousands of
a degree or less (cf. Sec. VI). In most modern designs, 2W and W axes are decoupled.
Scanning (step size can be software selected) or measuring at preprogrammed positions
(e.g., element line, background left and right of the line for many elements) is possible.
Typical angular scan speeds (2W) are in the range from 0.1 to 180 =min and slewing speeds
can be up to 4800 =min. The slewing speed is the maximum rotation speed of the goni-
ometer arm for passing from one position to another without recording. The 2W range
covered is from 4 to 152 , with slight differences between manufacturers. Because of the
high slewing speeds and full automation, the analysis time is severely reduced compared to
the older intsruments and, when not too many elements have to measured, is often not
prohibitively long.
Jenkins et al. (1984) and Croke and Nicolosi (1987) announced a dual-channel se-
quential spectrometer. A twin primary collimator was installed consisting of an upper, fine
collimator and a lower, coarse collimator. The beam is ‘‘divided’’ into two parts, each
striking its own crystal. Two crystals are mounted close together but with a 15 inclination
difference. The diffracted beams are detected by two detectors 30 apart. This fixed angle
between both detectors, however, limits the benefits of a dual-channel instrument.
Two alternative constructions should be discussed here. Simultaneous or multi-
channel instruments are equipped with monochromators set by the manufacturer (or by the
customer) at a fixed position for the detection of a single element. Usually, curved crystal
optics are used, and the whole setup is optimized for that specific element with an ap-
propriate curved crystal, entrance and exit slits, and detector. The crystals are logarith-
mically or logarithmically and cylindrically curved as explained in Sec. III.D.1.e, see also
Figure 20. To give some more flexibility to multichannel instruments, one or two channels
can be equipped with a scanner. The optics are either focusing (e.g., Bruker) or nonfocusing
(e.g., Philips). The flat crystal arrangement is principally a reduced size of the normal
monochromator. The focusing arrangement is complicated by the fact that crystal and
detector must be moved to focus the diffracted beam on the detector at any position. The
scan range of 2W for the Bruker scanner is 30 –120 and 10 –100 for the Philips scanner.
Full details of typical construction differences are given in Sec. VI on instrumentation.
Two hybrid, difficult-to-classify spectrometers are discussed in this section:
MDX1000 (Oxford) and the Venus 100 (Philips). The so-called ‘‘flexi-channel’’ of
MDX1000 is an energy-dispersive spectrometer. The Venus 100 has only fixed channels,
but they are sequentially presented to the fluorescent beam. Moreover, manufacturers are
focusing more and more on dedicated hyphenated instruments combining x-ray fluores-
cence with diffraction measurements, such is an ARL 8600S and Philips CubiX.
F. Detectors
The objective of a detector is the transformation of photon energy into an electrical pulse.
Pulses are counted over a period of time, and the count rate, expressed in counts per
second or any other unit of time, is a measure of the intensity of the detected x-ray beam.
From the theoretical point of view, the interaction of photons with matter is explained by
the particle character of electromagnetic radiation, not by its wave nature. Four main
classes of detectors may be distinguished for use in x-ray spectrometry according to the
medium responsible for the energy transformation: gas detectors, scintillation detectors,
semiconductor detectors, and recently introduced detectors supported on superconductive
elements (tunnel junctions and microcalorimeters). They differ in the efficiency of detecting
photons of a given energy, determining the range of energy or wavelength for which they
Copyright © 2002 Marcel Dekker, Inc.
are suited. The width of distribution of pulse amplitudes as a result of impact of photons
of the same energy determines the spectral resolution. Until recently, the best performance
in energy discrimination has been obtained with semiconductor detectors and the full
width at half-maximum (FWHM) is for the typical XRF domain in the range of 130 eV.
This allows electronic pulse-height selection discriminating the energy of the impacting
photons (at the expense of count rate, a maximum of 30 kcps). In WD spectrometers,
pulse-height discrimination is required for removal of noise of low amplitude and of
higher-order peaks of other elements interfering with the analytical line. It is a dis-
advantage typical for diffraction spectrometers. High-resolution detectors are not required
in WD spectrometers because the interfering photons differ so substantially in energy that
moderate discrimination is sufficient (a few hundred electron volts).
High-resolution detectors constitute the heart of energy-dispersive devices and are
discussed extensively in Chapter 3. Now, it is obvious that new types of device, super-
conductive junction detectors and microcalorimeters, exhibit energy discrimination an
order of magnitude better. Of course, in the future they might be the obvious choice in
applications demanding better spectral resolution, but the availability of the super-
conductive detectors is still a problem, as well as the maximum count rates they can
manage and their routine maintenance and price.
Only gas-filled and scintillation detectors will be discussed here as the devices sup-
plementing crystal dispersive operation. Some attention will be paid also to the micro-
calorimeter counters, because the stormy period in their development and because of the
prospects for the construction of the future generation of XRF (see Sec. VII).
1. Gas-Filled Detectors
In this type of detector, the energy-exchange process occurs between photons and gas
atoms or molecules in a strong electric field. When a photon strikes an atom, there is a
given probability that the quantum of energy hn of the photon is imparted completely to
an orbital electron of the atom. As a result of this gain in energy, the electron emerges
from the atom with a kinetic energy of Ekin ¼ hn W (Einstein’s photoelectric equation).
W is the work function of the electron for leaving its orbital and is a specific constant of
the atom. In the case of argon,
hn
Ar ! Arþ þ e
The electron imparts its kinetic energy to other atoms, creating a series of electron–cation
pairs on its path through the gas in a number equal to Ekin divided by the energy needed to
expel an electron. Values for different gaseous and solid detector media are included in
He 27.8
Ne 27.4
Ar 26.4
Kr 22.8
Xe 20.8
NaI (Tl) 50
Si (Li) 3.8
Ge 3
constant flow of gas through the counter (Fig. 29a). This is called a flow-proportional
counter. The gas is Ar with CH4 at a concentration of 10%, and the flow is about
0.5 dm3=h. For detectors in simultaneous instruments, helium is also used, or 88%
He þ 12% CO2. For very specific applications, other combinations of gases may be
useful. The gas is stabilized for pressure, flow, and temperature. This is of significance
because a 1% change in gas pressure or temperature introduces a much more serious
6–7% change in pulse amplitude, which can discredit the results of the analysis (Short,
1991). The amplification factor in the proportional range is between 102 and 106. The
current pulse is conducted over a resistor by which it is converted into a voltage pulse
that is amplified, shaped, possibly discriminated in height, and counted. The count rate
is a maximum of 26106 cps. The operating voltages range from 1000 to 3000 V.
b. Sealed Proportional Detector
Basically, the construction is the same as in the flow-proportional detector but the win-
dows are thicker and no constant flow compensation for a leak is needed (Fig. 29b). The
gases used are Ne, Kr, and Xe with Al, Be, mica, Mylar, or polypropylene windows.
According to the element for which the radiation is to be detected, the most suitable
combination of crystal, gas, and windows is chosen; for example, a PET crystal with a Ne
detector with Al (for the determination of magnesium!) or a LiF (100) crystal with an Ar
or Kr detector and a Be window for the detection many of medium-Z elements. If the
filament is 80 mm thick, it is stiff and easy to install, has a long lifetime, and is easily
exchanged.
Because the proportional counter is often used in tandem with a scintillation
counter, two windows are present on both sides of the main body. The absorption
Copyright © 2002 Marcel Dekker, Inc.
probability of photons decreases with photon energy; the high-energy photons pass to the
scintillation counter through the proportional counter without being essentially absorbed.
The sealed detectors are applied in simultaneous instruments on their own and are not
used in tandem with a scintillation counter. There is no need for a tandem arrangement
because each channel is optimized for one element with the best-suited combination of
detector and crystal.
c. Pulse-Height Distribution
In the detector, a photon generates a number of ion pairs proportional to the pulse am-
plitude. The formation of ion pairs is subject to statistical fluctuations, which means that
the number of pairs formed oscillates around the most probable value. A record of pulses
and their amplitudes as they emerge from the detector as a function of time is represented
in Figure 30. When these pulses are collected according to amplitude, a distribution
emerges as shown on the right side of Figure 30. The width of the distribution is measured
at half-maximum and is designated FWHM. This determines the energy resolution of the
detector. It is small for semiconductor detectors, such as Si(Li) (about 130–150 eV for the
5.9-keV MnKa lines from a standard 55Fe source; see Chapter 3), about 900 eV for gas
proportional counters, and 3500 eV for a scintillation counter, all measured for the same
photon energy. Note that this distribution differs from a distribution in intensity for
a normal spectral line (intensity versus wavelength) because of natural and instrumental
line broadening.
d. Escape Peak
As mentioned in the beginning of this section, a very probable method of energy exchange
between a photon and an atom of the counter gas is the removal of an outer-shell electron.
The probability of removal of inner-shell electrons is not zero, however, and this phe-
nomenon happens in a number of interactions, in which case the specific K or L radiation
is emitted. Because the wavelength of this radiation is on the long-wavelength side of the
absorption edge, the mass attenuation coefficient is low and this characteristic photon has
a good chance of escaping from the detector. The remainder of the photon energy is
transferred to the electron of which the kinetic energy is generating a number of ion pairs
but necessarily in a smaller number as an electron having received the whole photon en-
ergy. Thus, a number of pulses is created, distributed around an amplitude proportional to
the energy of this electron. The absorption edge for ArK is 3203 eV; the energy of the
FeKa line is 6403 eV. Pulses centered around 6403 7 3203 ¼ 3200 eV are generated, and
Figure 30 Amplitude–time record of impulses from the detector (left) and transformation of it
into a pulse-height distribution (right).
2. Scintillation Detectors
The energy exchange occurs in this type of detector in a medium of higher density and
with high-Z elements in the matrix, namely a thallium-doped sodium iodide [NaI(Tl)],
and as can be expected, this detector is only efficient for high-energy photons (< 2 Å
or > 6 keV). The outer orbital electron for an iodide ion requires about 30 eV to be
knocked out. Hence, the originally ejected electron is imparted with almost all the
initial energy of the photon. The ejected electron dissipates its energy by promoting
valence-band electrons to an excited state 3 eV above ground level, an energy emitted
on deexcitation as a photon of 3 eV or l ¼ 410 nm. The intensity of the emitted light
pulse is proportional to the number of electrons excited by the x-ray photon
(ffi hv=3 eV). The pulses are detected by a photomultiplier in which the light pulse
produces a few photoelectrons from the cathode material (e.g., indium antimonide).
These electrons are accelerated inside the vacuum tube between the cathode and the
first intermediary anode, called a dynode (Fig. 31). They gain kinetic energy and
generate, in turn, a higher number of electrons from the dynode. This process is most
often repeated 10 times, resulting in a substantial multiplication of the original number
of electrons. The multiplication depends on the potential differences between the suc-
cessive dynodes (100–150 V), and the order of magnitude of this multiplication is 106,
which is called the amplification factor A. In general, the photomultiplier is a front-end
Copyright © 2002 Marcel Dekker, Inc.
Figure 31 Layout of a scintillation counter.
tube, and the iodide crystal is in close contact with the tube window. The crystal itself
has on its outer faces a reflective coating (but transparent for the photons above
6 keV) and a coating against moisture. In the external electronic circuit, the electrons
provoke a current pulse of which the amplitude is ultimately proportional to the energy
of the photon captured in the crystal. The conversion of the x-ray photons into a
current pulse occurs over different steps: the initial formation of a photoelectron
(I ! I0 þ e ), the quantified dissipation of the kinetic energy of this electron in 3-eV
steps, emission of 410-nm light photons, the production of photoelectrons from the
photocathode of the photomultiplier, and the multiplication of electrons. Although the
production of the number of light photons in the crystal (hn=3 eV) is higher than the
number of ion pairs created in a flow counter (hv=30 eV), the subsequent conversion
into electrons is inefficient and roughly 1 electron is produced for every 10 light
photons! The statistical errors of each process, particularly in the photomultiplier tube,
propagate into the error of the output current pulse, thus substantially widening the
distribution of pulse amplitudes and, consequently, reducing the spectral resolution.
The DE is approximately two to three times worse than for a flow-proportional counter
(see Fig. 26a), making a scintillation counter unsuitable for ED spectrometry. However,
the resolution is still sufficient for a discrimination of higher-order spectra.
Hardware for pulse amplifying, shaping, and counting are much the same as for the
proportional counters. Also, an escape peak is present, with its energy equal to the initial
photon energy minus the energy at the iodine K absorption edge.
As already discussed, in many sequential spectrometers, flow-proportional and
scintillation detectors are used in tandem, coupled in different ways. In another
arrangement, both detectors are shifted over 30 . A secondary collimator is placed in front
of the scintillation detector but inside the vacuum chamber; the scintillation counter
is outside the chamber, but close to the exit window. According to the manufacturer
(Rigaku), this allows a gain in count rate of 15%.
3. Alternative Con¢gurations
Some interesting modifications of the detector have been proposed in the article by Ebel
et al. (1983). These authors considered the possibility of reducing count losses between
Copyright © 2002 Marcel Dekker, Inc.
analyzer crystal and detector (see Fig. 32) by removing the secondary collimator and al-
lowing the detector to collect a divergent beam and, at the same time, by using a position-
sensitive wire detector (with an energy resolution of 100 eV for detected photons with
energies of about 10 keV). Position-sensitive constructions are popular in many physical
applications, unfortunately not in the XRF field. Potentially, it may lead to new solutions
for WDXRF.
Another interesting effort is the use of a scintillation gas instead of the scintillation
crystal (Kikkert, 1983; Palicarpo, 1978). This is the so-called gas-scintillation propor-
tional counter (GPSC). As is clear from Figure 26a, the resolution is substantially better
than that of its ‘‘parents.’’ The production of avalanches of ion pairs by the expelled
electrons is the same as described previously. The discharge currents at the electrodes are
not measured, however, but, rather, the scintillation pulses in the ultraviolet part of the
optical spectrum are detected by a photomultiplier as in other scintillation counters. An
essential difference with conventional gas proportional counters is that the space charge
during the passage of a photon is avoided. The freedom-of-charge-related phenomena
enable using higher counting rates (up to 2.56104 cps; Dos Santos et al., 1993). The
lifetime of a GPSC can be quite long, up to several years, if there is no gas leakage in the
meantime. The ability to use counters with large windows positively differentiates this
kind of construction from room-temperature solid-state detectors, such as HgI2. Their
pulse amplitude is also proportional to the energy of the incident photon. The counter
can be supplied with a digital rise-time discrimination analyzer to suppress the pulses
Figure 32 Geometry with position-sensitive detector: T ¼ x-ray tube; specimen with point P
irradiated by primary x-rays; crystal; mp ¼ mirror plane; M ¼ mirror point of P; PSD ¼ position-
sensitive detector; l ¼ distance between mirror point and detector wire. (From Ebel et al., 1983.)
Nm
Nr ¼ ð11Þ
1 N m td
where Nr is the real count rate (count=s or cps), Nm is the measured count rate (cps), and
td is the dead time (s).
If we allow a given level of discrepancy, say a%, then the following equation is valid:
a
¼ N r td ð12Þ
100 a
The real count rate and dead time are thus connected by a hyperbolic relation. If the
allowed value of discrepancy must be kept constant, the left-hand side of the hyperbolic
equation is constant.
However, this is only part of the problem. As already stated, a photon, arriving just
at the end of the dead time, is registered with reduced amplitude. It is only after the so-
called recovery time that a new photon is registered with normal amplitude.
An improved method for calculation of the dead time was proposed by Bonetto and
Riveros (1984). These authors used the well-known fact that the second-order peaks are
significantly less intensive than their first-order analogs, but the ratio of the intensities
should be constant. If it is not constant, the deviation results fully from dead-time losses in
the first-order signal. They derived the relation
N m1
¼ B AN m1 ð13Þ
N m2
where B ¼ Nr1 =Nr2 and A ¼ Btd . The indices 1 and 2 refer to the order of the spectrum.
A summary of the constituting parts of spectrometers and the wavelength ranges in
which they are used is given in Figure 33. Note that different excitation sources are used in
different branches of XRF, namely radioisotopes for EDXRF and tubes for WDXRF;
there are also examples of synchrotron radiation. The x-ray laser is mentioned, but its
appropriate version for XRF applications does not yet exist.
Copyright © 2002 Marcel Dekker, Inc.
Figure 33 Review of different elements serving the construction of a spectrometer: optimal range
(crossed area), feasible range (hatched area), and possible range (open area). (Adapted from Bertin,
1975.)
In this section, we review a number of items important to both qualitative and quantitative
analyses. The qualitative application of XRF spectrometry is obvious. In simple cases,
quite unequivocal assignments of spectral lines can be made based on a library of Ka or La
lines or, in case of doubt, when an overlap exists, Kb or other L or M lines are checked to
confirm or reject a line assignment. A few more complex situations are discussed.
Quantitative analysis may also be simple. In the low-concentration range, quasilinear
relationships are found between intensity and concentration. The same holds for de-
terminations of higher concentrations but in a very narrow range for samples with quite
comparable matrices. Interpolation on standard curves, obtained on suitable standard
samples, allows a quick and sufficiently accurate quantitative determination. Another
intensity–concentration relation, valid for binary samples, is based on the relative in-
tensities of Ka and Kb lines in the case that the absorption edge of another element is
located in between these two lines. In these boundary cases, matrix effects are negligible or
interbalanced but always exist. In normal practice, they compromise the conversion of
intensity into concentration in such a way that more or less complicated mathematical
methods must be used for obtaining any accurate quantitative determination. The ap-
proach by fundamental parameter programs or correction algorithms is dealt with in
Chapter 5. Between simple qualitative work and rigorous quantitative work, however, an
area of applications is situated in which knowledge of the exact concentration or the
complete qualitative composition is less important than the overall aspect of a spectrum
for recognizing classes of samples (i.e., pattern recognition). An introductory discussion of
this aspect is given in this section. Another point of interest in qualitative as well as to
quantitative work is the measurement of the background. A good estimation of the
background is necessary for obtaining net count rates, but decreasing the background
allows lowering the detection limits. The origin and magnitude of the background is
bound to the physical processes involved in the method of excitation and, thus, indirectly
to the type of spectrometer. This aspect is also treated in this section. Finally, the said
disturbing phenomena associated with the excitation of fluorescence radiation, as
bremsstrahlung and the scattered radiation, can be used to increase the total analytical
information and they also have some interesting applications.
A. Background
The background in WDXRF has four main sources: (1) the coherent and incoherent
scattering of source radiation, (2) the presence of characteristic radiation of other than
sample origin (materials of sample cup masks, sample cups, collimators, spectrometer
housing, for example), (3) the detector, and (4) environmental noise (i.e., resulting from
the threshold natural radiation of materials and even from a cosmic origin). The back-
ground from characteristic radiation from construction materials of the spectrometer can
be minimized by careful construction and geometric arrangement of the device. Scattering
from sample cup masks can efficiently be reduced by masks in front of the primary col-
limator. This contribution is especially great when small samples are analyzed. The beam
mask reduces the irradiated area and may totally eliminate scatter from sample cup masks.
Higher-order scatter increases with increasing wavelength. Pulse-height selection, as ex-
plained earlier, supresses counts from higher-order diffraction, but this technique is most
efficient in the long-wavelength region and is decreasingly efficient toward the short-wa-
velength end. The short-wavelengths contribution to the background can be reduced by
Copyright © 2002 Marcel Dekker, Inc.
installing adequate filters. A 100-mm titanium foil removes the characteristic lines of
chromium from a chromium anode tube, significantly improving the conditions for the
detection of Cr, Mn, Fe, Co, and Cu (Tertian and Claisse, 1982).
For Cr tube radiation and a LiF (200) analyzer crystal, removal of scattered chro-
mium radiation may be necessary when detecting the neighboring first-order lines. As
Table 5 shows, not very many common elements are interfering. For steel analysis, only
manganese really represents a problem. Because chromium is present in prevailing con-
centrations in the environment (tube anode!), manganese determination is also difficult
with an aluminum filter. The simplest solution in such a case is the use of a LiF (220)
crystal with better resolution in this region (but also with lower intensity). If tubes with
other targets are used, tables are similar to Table 5 but with even more complex inter-
ference if targets like Mo and Rh are applied. The other elements listed in Table 5 refer
mostly to less common elements and less intense lines. Other higher-order lines may in-
terfere as well if the concentration of elements, from which the lines are derived, is high.
For special cases, a convenient combination of tube targets, analyzer crystals, and filters
can almost always be selected. Let us look at an example for the rhodium target tube. In
the determination of palladium, PdKa1 overlaps with the Compton peak of RhKb
(Fig. 34). A molybdenum filter in the primary beam reduces this peak to a very small level,
allowing a perfect estimation of the location of the maximum and the determination of the
intensity of the palladium line.
Another efficient solution for background suppression is the use of linearly polarized
x-rays. An excellent example was given by Knöchel et al. (1983) and it is reproduced in
Figure 35. The analysis of nitrogen with a small concentration of xenon was made by a
synchrotron excited XRF. Some special geometries allowing the use of polarized x-rays
have been proposed by Wobrauschek and Aiginger (1980, 1983, 1985, 1986) for the more
conventional XRF spectrometers. For EDXRF systems, a special commercial polarizing
design is available (SPECTRO A.I. Gmbh). The subject is of great significance to syn-
chrotron-based XRF and discussed as such in Chapter 8.
A careful estimation of the background is apparently the simplest way to take the
background into account! A first approach, but with limited application, is measurement
on blanks—simple at first glance, but the adequate blank is often a rarity. Moreover, the
blank ought to have the same scattering properties. Intensive efforts for mathematical
modeling of background were performed but frequently for other purposes.
The most common method of background evaluation is the linear (or even non-
linear) interpolation from background countings at a suitable angle above and below the
peak position (suitable means not too far away for an easy linear interpolation, outside the
tailings of the peak, and not at the position of an adjacent peak). Consequently, a good
Table 5 Interfering Lines for CrKa and CrKb with LiF 200 crystal
spectral resolution of the system is essential because only in this case are peaks very
narrow and do not overlap, and the background level is easily reached in the intervals
between peaks.
Finally, we refer to a method based on the relationship of fluorescent and scattered
radiation intensities versus the value of attenuation coefficients. No measurements of back-
ground for the estimation of net line intensities is required in this case (Bougault et al., 1977).
If we cannot overcome the background problem, it would be advantageous to use the
background or scattered tube lines for increasing the information about the sample. There
are very interesting examples of such work and they will be shown in Sec. IV.C.
Copyright © 2002 Marcel Dekker, Inc.
Figure 35 Synchrotron-excited spectra of gaseous nitrogen with a small amount of xenon taken in
the direction of polarization and perpendicular to the orbit plane. Counts normalized to stored
electron current are integrated over the lifetime of the detection system. DORIS was operated at
3.3 GeV and about 55 mA. (From Knöchel et al., 1983. Reprinted by permission of the author and
Elsevier Science Publishers.)
B. Qualitative Spectrometry
Although the number of spectral lines is very limited compared to atomic emission
spectra in the ultraviolet or visible region, overlapping of lines in x-ray spectra does
exist. A classic example can be found in the works of Gentry et al. (1976), Sparks et al.
(1978), and Sparks (1980), and an example is reproduced in Figure 36. At energies at
which K lines of elements between palladium and cesium (about 22 and up to 30 keV)
can be present, additionally the tails of L lines of uranium and thorium are found, and
also the theoretically calculated M lines of the superheavy elements of Z ¼ 110 [in the
time which passed since 1976 this element has been discovered (Hofmann et al., 1994)]
and 126 should be expected. The erroneous attribution of some lines to these hy-
pothetical elements was the subject of extended research and long discussions (Sparks
et al., 1978; Sparks, 1980). In the software of modern spectrometers, special searching
blocks exist for the assignment of lines: The detected line is compared with data com-
piled in the program. After the initial assignment, other lines of the element are iden-
tified according to data from the library. The initial assignment is eventually rejected if
not confirmed by other lines (Fig. 37). Multichannel wavelength-dispersive instruments
are not suited for this type of work, except for elements implemented on the spectro-
meter (maximum 28 elements). Sequential instruments can cover all elements (from
beryllium on). The spectra can be displayed on the monitor screen of all modern
spectrometers, documented, memorized as files, and printed. Particular parts of the
spectrum can be scanned occasionally with higher resolution and sensitivity (step
scanning) according to the time available for the analysis and the information required.
Parts of particular interest can be displayed separately with a linear intensity axis or
rescaled to square root or logarithmic scales. The peaks can be identified on the display
by the angle 2W, by element name, either by wavelength or energy.
Copyright © 2002 Marcel Dekker, Inc.
Figure 36 Fluorescence spectrum from a monazite particle excited with 37-keV synchrotron
radiation (ORNL DATA) shows an improved signal-to-background ratio over that excited with
5.7-MeV protons (FSU DATA). Data points every 20 eV in ORNL spectrum and every 61.8 eV in
FSU spectrum. (From Sparks, 1980.)
Figure 37 Example of qualitative analysis by rejection. The suspected MoLa peak in the lower
spectrum was rejected after comparison with the standard containing molybdenum (upper
spectrum). (Courtesy of Phillips Analytical.)
The new term m0kC2 ðEi Þ was introduced in Eq. (16). Although dependent on the excitation
condition E and composition, this new term has the great advantage of being dependent
on the latter in a simple way. The possibility of deriving the Lachance–Traill equation
directly from the Shiraiwa and Fujino equation under the conditions applicable to Eq. (16)
is also well established (Kuczumow and Holland, 1989).
The next interesting case concerns scattered radiation as a source of analytical al-
gorithms. Similar to the derivation made by Kuczumow (1982), an expression on the basis
of coherently scattered radiation is obtained (Kuczumow, 1988):
X
Wi ¼ Rcoh
i þ Rcoh coh
i a ij b coh
ij Wj ð17Þ
j6¼i
where
mjC1 ðE0 Þ þ mjC2 ðE0 Þ miC1 ðE0 Þ miC2 ðE0 Þ
ij ¼
acoh ð17aÞ
miC1 ðE0 Þ þ miC2 ðE0 Þ
sj ðE0 Þ
coh
ij ¼
bcoh ð17bÞ
i ðE0 Þ
scoh
and Ricoh results from the comparison of a coherently scattered line from the sample and
from the pure element i and scoh is the mass coherent scatter coefficient (for the set of
incident and takeoff angles, as always in the conditions of the analysis). For Compton-
scattered radiation, analogous expressions are obtained:
X
Wi ¼ Rcom
i þ Rcom
i acom
ij bcom
ij Wj ð18Þ
j6¼i
Rcom
i is the ratio of the Compton peak intensity of the sample to that of the pure ele-
ment. Equations (17) and (18) may be called Lachance–Traill equations for scattered
radiation by their general appearance as well as by the way they have been derived.
These equations are less sensitive to the type of components and changes in composi-
tions than their analogs for fluorescent radiation, which otherwise is an advantage in
analyses of samples widely varying in composition. From assumption, they include only
the terms for strictly defined energy values, even if the analysis is carried out with a
polychromatic source! In Table 6, the coefficients a and b from Eqs. (15)–(18) are col-
lected, calculated for the K lines of La and Sm, the main components of samples con-
sisting of La2O3 þ Sm2O3 excited by 60-keV photons at the incidence and emergence
angles equal to 80 and 70 , respectively (the example is taken from EDXRF, but it is
also instructive here).
When applied to nonhomogeneous samples, also the a’s are proved to be dependent
not only on the composition and the energy but also on particle size for dispersions or
layer thickness for piles of layers of different composition (Helsen and Vrebos, 1984a,
1984b).
Kuczumow et al. (1992) have proved that parallely with the conventional analysis of
binary Au–Cu alloys by fluorescent lines (Fig. 38), it is possible to get similar information
using coherently or incoherently scattered tube radiation. Sometimes, the results are even
better than for the direct fluorescence analysis.
Still other possibilities of using the scattered lines exist. Kuczumow et al. (1995) have
proved, by a proper mathematical transformation of the scattered Rayleigh signal, that for
light, chemically homogeneous matrices (as biological materials, e.g., wood), this new
function of scattered radiation, called the corrected scattered intensity, strictly follows the
density changes in the analyzed material (Fig. 39). The use of Rayleigh signals eliminates
the necessity of transmission measurements and brings important auxiliary information
(e.g., about the density). Although the method just described has been applied up to now
for x-ray capillary-microprobe-type measurements only, it may be used generally (Luggar
and Gilboy, 1994).
La2O3 Sm2O3
D. Pattern Recognition
The spectrum of a sample of complex composition is characterized by a set of lines,
documented by their position and intensity. If we imagine an n-dimensional space with the
intensities on the coordinate axes, the set of intensities determines a point in this space.
Such an intensity space is convertible into an n-dimensional space of concentrations, and
the point in the intensity representation has its equivalent in the concentration space
(Klimasara and Berry, 1987). The aim of pattern recognition is to attribute a sample to a
given class. A class is in a discrete way a set of points of similar features in the intensity or
concentration space or, in a continuous way, a subspace inside a complete n-dimensional
space. This subspace is restrained by a closed hyperplane, which can be approximated by a
number of hyperplanes of smaller dimensionality. Inherent to all experimental determi-
nations, the determination of intensity is also bound to a certain degree of uncertainty.
This implies that a point, as such, cannot be determined, only a small subspace of limited
size, limited by the assumed confidence levels. Equally ill-defined for the same reasons is
the boundary hyperplane surrounding the subspace occupied by a class of samples. The
‘‘thickness’’ of this ‘‘space ring’’ surrounding the subspace is determined in relation to each
Copyright © 2002 Marcel Dekker, Inc.
Figure 39 Line scans from x-ray microprobe measurements that show the correspondence
between transmission (dotted) and corrected scattered (solid) data, and the sequence of tree rings of
Silesian Pinus silvestris. The linear scan was made using a 89-mm capillary and a step size of 89 mm.
(Modified from Kuczumow et al., 1995.)
1. Delimitation of Classes
If we consider classes as separate subspaces in n-dimensional configuration space, it is
obvious that each of these classes has its own boundary hyperplane or closed space ring.
If these hyperplanes can be determined in an analytical way, we are able to know the
boundaries of classes. In simple cases, a class can be delimited by construction of a low-
dimensional hyperplane that closely approximates the real boundary. The simplest ex-
ample is the trivial division of the whole configuration space on two categories; occupied
space–empty space (zero–one decision). It can be done by assuming the limiting surface
(‘‘wall’’) inbetween the mentioned subspaces. If classes are separated by empty spaces, the
differentiation of classes is even easier.
5. Examples
Samples may be classified into groups according to their origin. A primitive version of
such x-ray methods was demonstrated by Lagarde et al. (1984). They observed the
characteristic features of spectra of ceramics samples from western Africa and then put the
samples in order ‘‘by the naked eye’’ according to their origin. A classic reference here is
the work of Kowalski et al. (1972), which was followed by many other articles on arch-
eology. Kowalski recognized the origin of obsidian artifacts by studying a set of intensities
of 10 trace elements and projected the image of this set from 10-dimensional configuration
space into a 2-dimensional space (nonlinear mapping). Similar applications in geology are
obvious. Lo I Yin et al. (1989a, 1989b) proposed the application of the pattern-recogni-
tion-supported XRF to the initial selection and analysis of Mars Rover geological sam-
ples. They discovered on this occasion that it is not necessary to use sophisticated
detection systems. On the contrary, they applied probably the most primitive version of an
x-ray spectrometer with a proportional counter as detector. The resolving power of such a
detecting system belongs to the worst among the existing ones (compare Fig. 26a), but
even in these conditions, the results of the sample-selection procedure are good. In such
classification procedures, putting emphasis on advanced data analysis can be a substitute
for using a sophisticated high-resolution spectrometer for special analyses in difficult
conditions.
Copyright © 2002 Marcel Dekker, Inc.
If only a limited amount of data is available, samples can only be subdivided in
groups. Discrimination of the alloys 663, B-1900, and 1455 can be done on the basis of
their zirconium content (Kevex Material Analyzer Mode 6500 brochure). Nickel–cobalt
alloys were subdivided on the basis of a logical analysis of successive spectral responses: Is
the ratio ‘‘contents of elements A=contents of element B’’ greater than some assumed level
X: A=B > X (Fig. 40)? Complete analysis of the alloy was not necessary. The same is often
done when samples can be compared to standards [e.g., the ‘‘framing action’’ by software
(description of Siemens software)].
An important application of pattern recognition may be found in ‘‘two-stage
analysis.’’ For complex samples with nonlinear dependence of intensity on concentration,
the analysis can be broken down in steps with a smaller number of dimensions (lower
degree of complexity) or the samples may first be subdivided in classes before applying
correction algorithms. The appropriate sets of differential equations exist: they establish
the relation between the value of relative intensities [Claisse–Thinh systems (Claisse and
Thinh, 1979)] or between increments of intensities [Kloyber et al. (1980) and Kuczumow
(1984) systems] and increments of concentrations. These increments are calculated with
respect to a central standard that represents the origin of a local system of coordinates. In
Figure 40 Selective procedure of the KEVEX Material Analyzer Model 6500 based on successive
analysis of the element ratios. (Courtesy of KEVEX Corp.)
F. Detection Limits
The importance of the correct determination of detection limits has been acknowledged by
many chemists, especially those involved in environmental regulation. The accepted de-
finition is as follows: ‘‘The limit of detection is the lowest concentration level that can be
determined to be statistically significant from an analytical blank’’ (Nomenclature, 1978).
The limit of detection is expressed as a concentration cL or an amount qL and is derived
from the smallest measure IL, I being the instrument reading. How the data are treated in
the statistics is described in Chapter 5. For a thorough discussion we may refer to the
article by Long and Winefordner (1983). This article also contains interesting references to
articles of Kaiser, Boumans, and others, which are therefore not repeated here. For the
derivation of some simple statistical dependencies, the small, almost forgotten book of
Beers (1958) is instructive. Finally, we can recommend the excellent discussion of the
physical origin of background sources and detection limits in PIXE and synchrotron-ex-
cited XRF given by Sparks (1980), which is also valid for WDXRF.
Here, we only invoke that according to the International Union of Pure and Applied
Chemistry (IUPAC) recommendation, the characteristic signal is statistically attributed to
the element when it is elevated not less than 3s(s ¼ standard deviation) above the back-
ground level. It makes the confidence level equal to 99.86%. The detection limit is then
calculated for the standard deviation multiplicity k ¼ 3 and is expressed as a function of
intensity in a unit of time (R ¼ N=t) with t as the counting time and with the sensitivity m
expressed in cps=% or cps=ppm.
In Table 7 a series of values for cL are given for randomly chosen elements collected
from manufactures’ documents. No general list can be included because detection limits as
well as sensitivities depend heavily on matrix composition and on the experimental con-
ditions. Instead, this is intended as an indication of order of magnitude.
Each discussion of detection limits provides methods for deciding whether a signal
value IU for a unknown was greater than background reading IB with a given level of
confidence. Another important approach is advocated and applied by Clayton et al.
(1987). They state that ‘‘traditional techniques for determining detection limits have been
concerned only with providing protection against type I errors, or false positive conclu-
sions (i.e., reporting an analyte as present when it is not). They have not considered type II
errors, or false negative assertions (i.e., reporting an analyte as not present when it is).’’
In their article, they discuss a test of the null hypothesis c ¼ 0 versus alternatives that c > 0,
where c denotes the true but unknown concentration of an analyte in a medium and for
two models of calibration curves with known or with unknown parameters. The test is
applied to chromatographic data. Their theory can certainly be applied to XRF data, but
the attempt has not yet been reported.
G. Limits of XRF
Compared to ultraviolet (UV)–visible emission spectra, x-ray spectra are ‘‘oligoline’’ ones
and very convenient with respect to the resolution of the spectrometers. The UV-visible
Copyright © 2002 Marcel Dekker, Inc.
Table 7 Detection Limitsa
emission spectrum of iron contains some 2000 lines between 200 and 800 nm; its
complete x-ray counterpart has a dozen important lines between 1.757 Å (Kb1) and
0.615 Å (Lg)! For everyday work, however, only two lines, 1.936 and 1.940 Å (Ka1,2)
and 1.757 Å (Kb), play a role. The densely populated L series spectrum such as that of
gold has a set of about 20 lines between 0.9 and 1.5 Å, a spectrum that can be
managed by a spectrometer with a relative spectral resolution of 102. The prospects
for the x-ray spectrometric range should be bright, but what are the limits to this
method?
1. Application Limits
All elements from beryllium on can be detected and=or quantitatively determined. On the
high-Z element or low-l side, the spectral resolution may be a problem for pure
spectroscopic studies, but seldom or never for qualitative or quantitative applications.
Because the emitted energy of photons is high and the attenuation coefficients are low,
the critical depth is great and, consequently, the analyzed volume is very representative
of the composition of the bulk. The critical depth is the thickness dcrit of a layer parallel
to the surface, from which 99% of the intensity sensed by the detector is recruited.
Copyright © 2002 Marcel Dekker, Inc.
It is a function of the takeoff angle C2 , the linear absorption coefficient mr of the matrix,
and the density:
ln 1 II1t ln102 4:61
dcrit ¼ lim ! ð19Þ
mr=sinC2
mr=sinC2 m r=sinC2
where It is the intensity emitted by a thin foil and I 1 is the intensity emitted by an in-
finitely thick sample.
For long wavelengths, mr becomes high while dcrit and analyzed volume become
low. If the sampled volume is low, the analysis may be called a near-surface determi-
nation, not representative for the bulk (see discussion in Sec. III.D.2.d). In that sense,
the results of the determination of heavy and light elements in the sample including
different kinds of elements can be nonequivalent. The heavier elements are then de-
termined in the bulk, whereas the light ones are determined on the surface only. This can
lead to surprising results if the process of surface segregation is occurring. Moreover,
during recent density determinations (Tsuji and Wagatsuma, 1996) of coating layers by a
similar method, Grazing emission x-ray fluorescence, striking differences have been re-
vealed between surface and bulk densities; for example, the density of a 27-nm-thick
coating lead layer was equal to hardly 35% of the normal, the bulk value of the lead
density (Tsuji et al., 1997). It is a true limit by the force of nature, although it may be
exploited for near-surface determination of light elements. From the concept of critical
depth, it is obvious that coatings can be analyzed up to certain thicknesses depending on
the absorption (and enhancement!) characteristics of the coating and substrate; the
overall thickness range is from 0.01 mm to a few hundred micrometers for the normal
angle versions of XRF.
Apart from the analysis of macrosamples, WD spectrometers may also be fit to
electron microprobes, where they supplement the ED spectrometer for better spectral
resolution of light elements. Curved crystal optics is generally used because of optimum
efficiency (compare the quantum efficiency of a classical WD spectrometer ffi108, curved
crystal WD ffi 106, and ED spectrometer ffi104). WD spectrometers are also needed for
heavy-particle-induced emission; because of the presence of many satellite lines, good
resolution is required (Watson et al., 1977; Chapter 12).
The dynamic range of analysis covers about five orders of magnitude, from 103 to
100%. Below 103%, preconcentration procedures must be used.
The intensity and uniformity of exciting beams delivered by conventional end-win-
dow tubes were not sufficient to allow spot analysis. However, Boehme (1987) and Nichols
et al. (1987) published the results of two-dimensional mapping executed with small beams
from high-power sources (rotating anode tubes), allowing lateral resolution of about
30 mm. This offsets the boundaries to a new domain: high-spatial-resolution x-ray spec-
trometry. Still, the possibilities of the tube-excited, slit-based WDXRF microprobe are
relatively poor in comparison with an x-ray capillary-based EDXRF system or especially
proton and electron microprobes (Chapters 11–13). However, for many applications, the
lateral resolution mentioned earlier is quite sufficient and this technique is time-saving.
Moreover, the tube can be coupled with the polycapillary device. Note, that, for example,
the diameter of the spot obtained from the polycapillary lens is in a range 50–500 mm, but
the careful work with such a beam can give a somewhat better spatial resolution. In
Sec. VI.B.3, we will discuss the milliprobes, a new solution which seems to be the most
suitable for the wavelength-dispersive mode of operation.
Copyright © 2002 Marcel Dekker, Inc.
2. Limits of Precision and Accuracy
An analyst is always threatened by errors in the analysis outliers, random, and systematic
errors. When a given operation or measurement provokes an offset between the true and
measured values, a systematic error is said to be introduced. This offset may be due to a
wrong calibration of any kind or to a wrong standard. Such errors can only be detected by
analyzing a sample from the same batch with other techniques. Random errors and
outliers can be dealt with by statistical methods. One example of such a method is de-
scribed by Plesch (1981) and implemented in the software package Spectra 310 (Siemens).
A calibration with n standards is assumed to give a standard deviation. The standard
deviation ss is calculated according to
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X ðC0 Ci Þ2
ss ¼ i
ð20Þ
np
where Ci is the true weight fraction and C0i the result obtained; p is the number of para-
meters used in the fit: p ¼ 2 for a linear and p ¼ 3 for a parabolic fit. If the concentration Ca
of an element in one specimen differs considerably from C0a , the presence of an outlier is
suspected. The suspected value is eliminated, and a new standard deviation is calculated.
Ca is an outlier with a probability of P% according to the inequality (F-test) (S1=S2)2 > Fp.
The threshold values for Fp are listed in the current statistical literature. Application of the
F-test in fact checks whether a statistically significant difference exists between the stan-
dard deviation calculated with and without the suspected result. Possible causes of outliers
are numerous, including all kinds of human mistake, temporary instrument failure, and
heterogeneity of specimens. Recently, the numerical programs helping to segregate the
data on the ‘‘true’’ ones and outliers are available. They are based on the Grubbs test
(Grubbs, 1969; Rousseeuw and Leroy, 1987).
A measurement of fluorescent radiation is, as in any other experimental determi-
nation, subject to minimal error. The standard deviation of a radiation measurement is
equal to the square rootpofffiffiffiffi the number of accumulated counts N. The real standard de-
viation is greater than N because of errors of instrumental origin. Another source of
error is the conversion of intensity into concentration and, last but not least, inherent to
the sample, microheterogeneity, which is difficult to foresee and to quantify and is present
even when all normal precautions for careful sample preparation are taken.
Heterogeneities caused by sample preparation are dealt with elsewhere. These arti-
facts should be absent here. Segregation, however, which is the origin of what we call
microheterogeneity, is a bulk (or sometimes surface) phenomenon of crystallographic
origin: solid solutions of compounds or elements that are not perfectly soluble in each
other, segregating into multiphase components. The ‘‘degree’’ of heterogeneity in the
composition of the segregated phase as well as in the size of the phases depends on the
history of the solution (thermal, chemical, or mechanical treatment). It is not easily
mastered and is difficult to quantify. Its importance for XRF analysis has been ac-
knowledged by several authors; Examples are tin and lead in solders (Glade and Post,
1970), silicates in fused beads (Novosel-Radovic et al., 1984), and silicon in Si–Al alloys
(Michaelis and Kilday, 1962). The first report we are aware of is an article by Claisse
(1957) on the determination of FeS in a sulfur matrix. A systematic study has been devoted
to this phenomenon by Helsen and Vrebos (1984a, 1984b, 1986a, 1986b) and Vrebos and
Helsen (1983, 1985a, 1985b, 1986), both by Monte Carlo simulation on hypothetical
mixtures and by measurements on samples of known segregation. It was found by si-
mulation on a (hypothetical) dispersion of spherical particles of iron in a chromium matrix
Copyright © 2002 Marcel Dekker, Inc.
that even for a particle diameter of 1 mm, an error is introduced in the relative intensity of
FeKa of 6% (on the level of cFe ¼ 0.5) with respect to the perfectly homogeneous solution.
The effect was proved experimentally on alloys of Al–Si, Al–Mo, and Al–Zn. The extent of
the effect changes with time, as in fused beads. This was demonstrated by the determi-
nations of Novosel-Radovic et al. (1984) on fluxed silicates and also by Monte Carlo si-
mulations by Vrebos and Helsen (1985a, 1985b). Microsegregation imposes a true physical
limit to the accuracy obtainable by XRF spectrometry on quite a substantial number of
samples, including minerals and alloys.
3. High-Z Limit
We already insisted on the low-Z limit in Secs. III.D.1 and III.D.2, but what about the
high-energy end of the spectrum?
Commercially available generators and side-window tubes allow working at high
voltages up to 100 kV, bringing even the K line of bismuth into the potential analytical
spectrometric range. If a tungsten target is used, however, characteristic lines represent
only a minority of the integrated emitted intensity, thus, excitation is bound mostly to
continuous radiation, the maximum of which lies at 50 keV. Thus, only elements with K
absorption edges below this value can be efficiently excited (Eu, Z ¼ 63, and all elements
below). The above remark is even more valid for the spectrometers with end-window
tubes, most commonly used in present-day instruments, of which the high voltage is
limited to 60 kV. Possible excitation of K lines for all the elements from the periodic table
by the tubes operating in voltages up to 160 kV is bound to metal–ceramic low-power
tubes, cooperating with EDXRF systems and is outside the scope of this chapter. Thus,
the available voltage limitation is the first one.
If we return to Figure 26b and compare the spectral resolution of the LiF spectro-
meter to spectral distances between analogous lines in the K series, it is obvious that spectral
resolution is worse above 35 keV (for K lines of Pr, Z ¼ 59). This is a second limitation.
Most of goniometers have their optimal 2W range between 10 –15 and 70 : At the
higher 2W end, the angular dispersion becomes insufficient; at the lower 2W end, the in-
tensity becomes very low. If we choose LiF (420) with the shortest interplanar distance,
2d ¼ 1.802 Å, useful measurements can be made at 10.0 corresponding to photons of
39.6 keV (Sm, Z ¼ 62 in the first-order spectrum). This is a third limitation.
Studying Figure 26 further, we note that the spectral resolution of the LiF spec-
trometer is below the spectral resolution of the semiconductor Si–Li detector for photons
of energy exceeding 17 keV, which corresponds to MoKa, Z ¼ 42. This is the fourth and
most serious limitation in the K series. For analysis in the L series, there is no such
limitation because the whole energy range lies within the capabilities of wavelength-dis-
persive devices. However, using the new microcalorimetric detectors with extraordinary
and relatively constant spectral resolution, one can anticipate the preponderance of the
spectral resolution factor for wavelength-dispersive crystals only in a region below
4–5 keV, which means the Ka line of V, Z ¼ 23 (see Fig. 27). Then, the ability to make the
efficient trace analysis (because of higher count rates) and the availability of the device are
still the advantageous sides of application of the crystal dispersion in the future.
The wide voltage range of generators should not be used without some awareness of
inherent drawbacks. For example, the K absorption edge of Mo is 20 keV. Thus, each
x-ray tube with characteristic lines above 20 keV and working above twice this 20 kV for
the maximum of bremsstrahlung generation is quite good for the excitation of Mo. The
potential, however, should not exceed this value substantially much because the maximum
Copyright © 2002 Marcel Dekker, Inc.
of the white radiation would shift more and more away from the molybdenum absorption
edge and even the spectral characteristics do not improve. If we divide the mass coefficient
of photoelectron excitation by the sum of mass coefficients for Compton and coherent
scatters (this is a kind of measure of peak-to-background ratio), we obtain the values 76
and 18 for tube-excitation potentials of 40 and 100 kV, respectively! This is a substantial
loss in peak-to-background ratio while passing to higher voltages.
The discussion in this section shows that the advantage of high-energy photons
(generated by very high voltages in SWT) does not compensate for the drawbacks and this
probably explains why most manufacturers fit EWT tubes to the instruments. During the
last years, manufacturers rather resigned from high-voltage generators and tubes, even in
high-power 4-kW constructions. The use of a higher current is now a preferred option.
Thus, we expect that only spectrometers with a high voltage up to 60 kV will dominate the
market in the future. The effective spectral range above 30–40 keV should probably be left
for typical energy-dispersive detectors.
In the preceding paragraphs, we tried to show that although the spectrometers are
almost perfect, there are limits to precision and accuracy due to the nature of the sample
and to the mathematical conversion of intensity to concentration. These uncertainties are
generally greater than those introduced by instrumental parameters.
Figure 42 Chemical shift as a function of valence. The energy shift divided by the ionicity is
plotted (solid line). The dotted curve illustrates the uncorrected values of DE. (From Sumbaev, 1970.
Reprinted with permission of J. B. Adashko.)
Figure 43 Satellite peak intensity ratio Kb5 =Kb1 versus oxidation number. (From Asada et al.,
1975.)
Figure 46 Spectrometer with EWT and sample-surface-down position but with the addition of a fixed
channel. This channel is the similar to the ones used in simultaneous spectrometers. (Philips PW2404)
(Philips Technical Materials. Sequential X-Ray Spectrometer System PW 2404. Courtesy of Philips.)
realizes the simultaneous determination of more than one element by enclosing within the
vacuum chamber a complete second monochromator of their unique extremely compact
design as represented in Figure 47. It is characterized by the complete absence of gears and
a positioning system by counting Moiréfringes resulting from interference of two grating
systems.
A development of the last few years is the substitution of diaphragms by a small-
diameter aperture allowing analysis of an area as small as approximately 1 mm2 (see
Sec. VI.A.3 about milliprobes). An appropriate mechanism is moving the sample, which
allows element mapping. This feature is realized in the XRF-1700 or 1500 series of Shi-
madzu (Fig. 48). It is an instrument with sample-surface-up geometry and fitted with an
attenuator, reducing the intensity by a factor of 10 (for a major element determination);
the merit of this is that no tube power switching is required for major and minor element
determination.
Copyright © 2002 Marcel Dekker, Inc.
Figure 48 Sample-surface-up spectrometer, fitted with small-aperture collimator diaphragms and
an attenuator; all parts are enclosed in the vacuum chamber (Shimadzu XRF-1700). (Courtesy of
Shimadzu.)
Only one manufacturer still produces an instrument with SWT. A number of in-
struments of a former generation are still in use and the tubes are still available. Diano
Corporation continues the production of a spectrometer with sample-surface-up geometry
and SWT (cf. Table 10). The merits and disadvantages of EWT or SWT were previously
discussed. However, if a good homogeneity of the beam is required, a fine-focus SWT is by
far a better solution.
Sample holders of slightly different design for solids and liquids, for sample surface
up or down, are offered by the manufacturers. The holders for instruments with sample-
surface-up geometry are fitted with a spring mechanism to press the sample against the
sample mask, both for sequential and simultaneous instruments. The holders adapting
insert for liquids have a shape that can be recognized automatically and will not be ac-
cepted if the spectrometer is not under a He atmosphere (or eventually switches auto-
matically from vacuum to He).
2. Simultaneous Instruments
All simultaneous instruments are equipped with end-window tubes with very similar con-
struction features. Up to 30 monochromators may be arranged radially around the tube
head, with alternating high and low takeoff angles. One or more monochromators may be
replaced by programmable goniometers of the flat (a normal flat crystal goniometer of re-
duced size) or curved crystal type. The latter is small but limited in angular range; the former
has a large angular range but is large and occupies at least two positions otherwise taken by
fixed monochromators. As already mentioned, Philips offers a small, low-power target
transmission tube (TTT tube) used in the spectrometer PW 1660 (Philips Technical
Copyright © 2002 Marcel Dekker, Inc.
Materials. Simultaneous X-ray Spectrometer System PW 1660) or X0 CEM capable of si-
multaneous determination of only 14 fixed elements or 12 elements when equipped with a
programmable goniometer. In most instruments the samples are placed in the surface-up
position. Some instruments (e.g., Oxford and PW2600 from Philips) irradiate in the sample-
down position. The monochromators can be fitted to the outside of the vacuum cabinet, in
which case they need not be vacuum tight. Figure 49 shows a generally used geometry.
In the ARL sequential instruments, fixed channels (up to eight) or a complete second
monochromator can be added. In Philips’ sequential spectrometer (PW 2400), an optional
two fixed channels may be installed, making this instrument partly simultaneous. This is
made feasible by the use of the special x-ray front end tube, the model Super Sharp Tube.
3. Milliprobes
In an obvious effort toward the new fields of application, two companies introduced
options, which seem to be restricted up to now to the very special, laboratory-made in-
struments. Rigaku and Shimadzu independently constructed the analytical systems with
special masks: nonmovable (Rigaku) and scanning type (Shimadzu). In both versions, the
diameter of the opening is 1 mm. Coupled with a 4-kW generator and high-power tube, it
can carry out the analysis of the small pieces of a sample. In the Shimadzu version,
C. Intercomparison of Instruments
Only a limited number of manufacturers produce x-ray spectrometers. Hereafter, we will
discuss the instruments of seven manufacturers subdivided in three classes: sequential and
simultaneous instruments and a third class that we called hybrid spectrometers, because
they have characteristics belonging to both former classes. The series numbers together
with the names of the manufacturer are given in Table 8. In Tables 9–11, the numbers in
the top row of the third, fifth, and last column refer to the manufacturer and series number
of Table 8. The reader will find the actual addresses of the manufacturer’s headquarters at
the end of this subsection.
1. Sequential Instruments
In Table 9, a summary is given of the main characteristics of the current instruments as a
guide to the market for the interested reader and for those who are considering buying a
new instrument or replacing an older instrument.
A question mark in the table means that the data were not available in the
brochures. When nothing is marked, it means that the given item is not
applicable.
Manufacturer
1 2 3, 30 4, 40 5 6, 60
Generator
Power (kW) 3 3 3, 4 3 or 4 3 or 4 3 or 4
Voltage range=step ?–60=? (opt.: ?–70=?) 15–60=1 20–60=1 60=? 10–60=1 60=?
(kV=step)
Current range=step (mA) ?–100 0–80=0.1 10–125=1 100 or 150=? 5–150=1 100 or 140=?
Switch time (s) ? ? < 2,5 ? Fast —
Switch mode ? ? Isowatt curve ? — —
Stability D%/% mains 0.0001 0.0005 0.0005 0.0005 0.0001 0.0002
Source
X-ray tube: geometry, EWT, 3 SWT EWT, 3 EWT, 3 or 4 EWT, 3 or 4
load (kW)
Anode, type, Be window Rh, 3GN, 75 Cr, EA-75, 250 Rh, SS, 75 or 125 Rh Rh, AGRh66G, Rh
thickness (mm) 75 or 125
Option other anodes W, Mo, 125 W,Pt,W=Cr, Others on request Rh=Cr,Rh,=C, Mo, W, Au, Cr Rh=Cr, Rh=W
Pt=Cr,Mo=Cr Rh=W
Primary beam filters 3 None 5 1 or 5 10 4
number
Elements Cu, Al, Fe — Al, Cu Zr, Ti, Al, Cu Cu, Al Al, Ti, Ni, Cr
Sample
Sample surface Down Up Down Up Down Up
Sample turret: number 2 4 2 6 2 8 or 2
of positions
Sample=spectrometer seal y n n n y n
(yes, no)
Optics
Diaphragms: number of 3 — 4 6 4 5
positions
Collimators: selectable y, 4 1 y, 3 y, 2 or 3 y, 4 3
(yes, no), number
Manufacturer
1 2 3, 30 4, 40 5 6, 60
Installation
Instrument dimensions: 920*1080*760 1400*860*1346 1092*1410*920 1400*1420*1054 1070*1890*880 1350*1770*1080
H*W*D (mm)
Weight (kg) 450 590 580 750 700 780
Electricity supply (V, kVA) 220, 6 208–480, 10 200–240, 8, 5 200–220, 10 115–400, 8 200–220, 8 or 9
Cooling water: bar, max. > 2, 20, 0.5–5 > 1, 29, 19 > 3.5–8, 20, 6–11 ?, < 30, 3.5–11 4–6, —, 1.5 > 1.5–3, 30, 3–10
temp. ( C), flow (L=min)
Compressed air: bar, — — 4–5, ? None 6–10, 20 —
flow (L=min)
Manufacturer
1 2 3 4 5
Generator
Power (kW) 3 3 3 0.2 3
Voltage (kV), step (kV) ?–60=? ?–60=? 5–60=1 (a) 50 20–60=1
Current (mA), step (mA) ?–100=? ?–75=? 5–125=1 4 5–150=1
Stability D%/% mains 0.0001 0.0005 0.0005 0.0005 0.0001
Source
X-ray tube:
Load (kW) 3 3 3 TTT* Rh 66 G
Anodes Rh, W Rh, Cr, Mo Rh, other on Cr Rh, Mo, W,
request Au, Cr
Be windows (mm) 125, 75 125, 75 125, 75 125, 75
Primary beam filters 4 (a) 10 3
positions
Elements Cu, Al, Fe
Primary beam attenuators y, 14 y
(Yes, No)
Samples
Sample surface Up Up Down (b)
Size: diameter, height (mm) 75, 50 51, 40 51, 40 40, ? 52, 57
Sample changer (positions) 11, 32, or 8 (b) 6, 72, 300 8, 58, 100, 138
Monochromator (fixed
channels)
Takeoff angle ( ) ? 29, 44 29, 44 25, 39, 5 26, 24 (a)
Maximum number 30, 20, 11 (c) 28 (a) 28 14 28
Crystals Curved x Log. curved
Detectors Gas counter, sc Gas (Ar flow, Ne, fc, sc
Kr sealed)
Element range Be–U Be–U Be–U
Goniometer
Manufacturer
1 2, 20 3
Generator
Voltage (kV)=current 40–50=4=200 50=4=200 id.
(mA)=power (W)
Stability D%=%mains 0.007 0.0005 id.
Max. mains 210–250 180–264 id.
variation (V)
X-ray source
Tube Similar (a) Target transmission id.
Anode Rh Sc Cr
External Cooling None None id.
Sample
Sample surface up, down Down Down id.
Spinning (rpm) 30
Dimension (diameter, 48, 40 26–47, 40 id.
height in mm)
Sample changer ? Single, 10 Single
Spectrometer chamber Vac, He optional Vac, He Vac
Monochromator (fixed)
Element covered F–U S, S-bg., Zr, Ni, Al, Si, Na, Mg,
Ni-bg., Mn, Ti, Mn, K, Ca,
Mn-bg., V, V-bg P, S, Fe
Crystals Flat, curved optional Curved id.
Additional Flexi-channel None id.
spectrometers (EDXRF)
Dimensions 1090*800*800 id. 1200*750*960
(H*W*D in mm)
Weight (kg) 160 id. 180
3. Hybrid Instruments
The instrument from Oxford MDX1000 is classified as hybrid because it combines a WD fixed
channel with what is called a Flexi channel, an EDXRF unit. The x-ray tube is similar to the
target transmission tube, produced by Philips, but no other details are given in their brochures.
The instrument series Venus 100 from Philips are called hybrid because Philips uses
fixed channels, which are, however, switched sequentially in front of the sample. The in-
struments are delivered for specific purposes: The PW4110=10 is provided for the analysis
of industrial minerals, particularly for the ceramic industry, and the PW4100=20 and
Copyright © 2002 Marcel Dekker, Inc.
PW4120=20 for the determination of, for example, nickel, sulfur, and vanadium in pet-
roleum products. The element range can be adapted to specific purposes.
Both instruments from Oxford and Philips are simple, low-cost, but high-perfor-
mance devices which can be used in harsh industrial environments; they are simple to
install and are characterized by small dimensions, light weight, no water cooling, no
compressed air, and no severe requirements for the stability of the mains.
Hybrid instruments are presented in Table 11.
The time passed since the first edition of this handbook has been one of the most stormy
periods in research and applications of x-rays since their discovery! It is well known how
Copyright © 2002 Marcel Dekker, Inc.
much Röntgen was confused, although totally convinced, by the impossibility to de-
monstrate the electromagnetic nature of x-rays through reflection, diffraction, or inter-
ference experiments! For many years, it remained impossible and further progress was very
slow and tedious. Recent years have brought the real breakthrough. We know of three
recent realizations of x-ray lenses:
Two diffractive:
1. Zone plates (in other words, Fresnel diffracting structures), earlier restricted to
the very soft x-rays. They are very important for the construction of the x-ray
microscopes and for the observation of living biological objects in the so-called
‘‘water window.’’ But now, by combining the Fresnel diffraction patterns with
Bragg’s phenomenon, the special Bragg–Fresnel lens was constructed, well
suited for focusing hard synchrotron radiation (Aristov et al., 1986; Snigirev,
1995; Snigirev et al., 1995; Chevallier and Dhez, 1997).
2. Snigirev refractive lens (Snigirev et al., 1996; 1998; Elleaume, 1998; Hastings,
1996; Eskildsen, et al., 1998), consisting of one or two mutually perpendicular
arrays of holes drilled in low-Z dense material (Be, Al). Depending on the
number and shape of the holes, they can focus hard x-ray beams,
One version using total reflection:
1. Kumakhov polycapillary lenslike device, which might be somewhat poor in
focusing capability (spot is 50–500 mm wide) but very efficient with respect to
the fraction of the primary input power turned into a small spot on the analyzed
surface. Interesting products were announced by X-Ray Optical Systems Inc.,
Advertising Materials.
In parallel, great progress was realized in the construction of energy-dispersive de-
tectors with increasingly better spectral resolution. The most promising device seems to be
the microcalorimeter detector, a device with very good spectral resolution parameters.
Moreover, its spectral resolution is more or less constant over the whole spectral range
covered (1–20 keV). With such a device, the spectrum deconvolution problems, so char-
acteristic for devices with poor spectral resolution, should vanish. Potentially, it lowers the
detection limits, by narrowing the energy region in which the very sharp peak is placed.
Only a very restricted number of background counts is detected in that region, thus im-
proving the peak-to-background count rate ratio. However, the detection limit depends on
the maximum count rate allowed for these detectors, which is, at the moment, very re-
stricted. Recent discussions show (Ladbury, 1998) that intensive work is going on to
overcome this shortcoming in the nearest future. The aim is to get the device with an area
of 40 mm2, with count rates up to 105 cps.
Another progress took place in the construction of time-resolved excitation sources.
The classic radiation generator for the purpose is synchrotron. Probably much more
convenient, the ‘‘portable’’ and switchable version of a time-resolved x-ray source will be a
compact-flash source. Furthermore, the construction of the x-ray laser—an old dream!—is
also progressing, maybe slowly, but in a promising direction.
Collected together, all available information about this progress justifies hope on the
construction of new versions of existing instruments as well as the advent of new ones. It is
worth noting that in the present chapter we considered an XRF spectrometer mainly as a
device used for the determination of elemental composition. What is now discussed tes-
tifies that future instruments will give both compositional and spatial resolution (mi-
croprobe) and time resolution (kinetic probe). New kinds of lenses and normal incidence
Copyright © 2002 Marcel Dekker, Inc.
mirrors give an impetus to the construction of x-ray microscopes also in the hard part of
the x-ray spectrum. One can easily imagine a microscope (but one easily construct!)
working in the hard x-rays regime supported by two different x-ray lenses—one acting as
an ocular lens and one as an objective lens.
The new version of the microprobe could then efficiently concentrate x-rays (e.g., by
Kumakhov polycapillary) on the confined areas and then record the fluorescent radiation
with the excellent spectral resolution of a microcalorimeter as the energy-dispersive de-
tector with the whole spectrum completed in one operation. Such a device is now under
construction in the National Institute of Standards and Technology (Wollman et al.,
1997a, 1997b; Anonymous, 1997). The microcalorimeter detector, attached to a scanning
electron microscope, has, for the time being, a spectral resolution of 7.2 eV 0.4 eV,
calculated in the standard way for the MnKa line, or 2.0 eV for 1.5 keV (Wollman et al.,
2000). Count rates are below 150 cps, but it is, by far, not the last word in this matter.
A newcomer in the generation of possible applications of the classical WDXRF,
rather unexpectedly brought to the awareness of scientists, was that the extended fine
structure of fluorescent spectra emitted by solid samples can be observed (Hayashi et al.,
1997; Kawai et al., 1998). Up to now, this field was reserved for the absorption mode,
mostly in combination with synchrotron radiation. The discovery of the extended fine
structure in experiments performed under normal laboratory conditions opens a new field
in the study of local structure in condensed matter and makes this method available to
lower-budget laboratories.
ACKNOWLEDGMENTS
REFERENCES
Andrew T. Ellis
Oxford Instruments Analytical Ltd., High Wycombe, Buckinghamshire, England
I. INTRODUCTION
The scope of this chapter is laboratory and industrial x-ray fluorescence (XRF) analysis
systems in which x-ray tubes are used for excitation and energy-dispersive (ED) semi-
conductor detectors, as opposed to Bragg-diffraction (wavelength) dispersion devices,
are used. Such ED detectors directly measure the energy of the x-rays by collecting the
ionization produced in a suitable semiconductor material. The coverage of this chapter
specifically excludes the following topics, which are covered in other chapters: on-line
EDXRF systems (see Chapter 7), total-reflection XRF (TXRF) (see Chapter 9), polarized
beam XRF (see Chapter 10), and x-ray microfluorescence (XRMF) (see Chapter 11).
Early approaches to EDXRF used gas proportional counters or scintillation de-
tectors to determine directly the energy of the x-rays. Such systems were limited in their
application because of the inherently poor energy resolution, which precluded the se-
paration of characteristic x-rays of adjacent elements in the periodic table. The limited
actual energy resolution of such detectors led them to be known as nondispersive, but this
drawback has been effectively countered in some benchtop instruments through the use of
novel arrangements of primary and secondary beam filters (Ross, 1928; Kirkpatrick, 1939;
Kirkpatrick, 1944; Field, 1993). The real breakthrough in EDXRF came in the late 1960s
(Bertolini et al., 1965; Bowman et al., 1966; Elad and Nakamura, 1966; Aitken, 1968) with
the arrival of solid-state semiconductor diode detectors and their associated pulse pro-
cessing circuits. These detection systems were developed through the early 1970s, often in
electron microscopes, to the point at which practical x-ray spectrometry with an energy
resolution of 200 eV or less became possible (Frankel and Aitken, 1970; Landis et al.,
1971; Heath, 1972). Although the energy-resolution capability of semiconductor detectors
remained greatly inferior to that achieved by wavelength-dispersive (WD) XRF systems,
the increased efficiency inherent in the energy-dispersive method compensated in many
analytical applications and permitted the use of a multiplicity of experimental geometries
not practical with WDXRF. A wide variety of EDXRF analytical systems based on
radioisotopoe sources, x-ray tubes, charged-particle accelerators, microprobe electron
beams, and synchrotron light sources have been developed in recent years.
principal driver for best precision and detection limits is peak-to-background (P/B) ratio.
However, as mentioned earlier, the limited total counting capacity of the EDXRF system
and the analysis of multielement samples are added complications in deriving optimum
excitation conditions. In general, the kV governs the sensitivity and the primary beam
filter governs the background.
The energy distribution of the spectrum arriving at the sample governs the effec-
tiveness of excitation for element XRF lines. In order to excite x-ray fluorescence, it is
necessary to have incident x-ray energies above the absorption edge for the element’s line
series to be excited. In order to have adequate excitation, there needs to be a high intensity
of x-rays with energies higher than the edge and conventional wisdom has suggested that
the tube kV should be 1.52 times the absorption edge of interest. This overvoltage en-
sures that there is a substantial proportion of the x-ray tube output spectrum available for
excitation of the element line(s) of interest. Figure 3 shows the unfiltered x-ray tube
spectrum scattered from a thin-polymer-film sample when the x-ray tube is operated at
5, 10, and 15 kV. The observed scatter spectrum is a reasonable representation of the
spectrum striking the sample and clearly shows the following:
Copyright © 2002 Marcel Dekker, Inc.
Figure 3 Scattered, unfiltered excitation spectrum provided by a silver target x-ray tube operated
at (1) 5, (2) 10, and (3) 15 kV.
The high-energy cutoff at the operating kV. This is also known as the DuaneHunt
limit.
The bremsstrahlung ‘‘hump’’.
The intense characteristic AgL lines (around 3 keV) from the tube anode. The intensity
of the AgL lines increases with the kV (i.e., as the overpotential increases).
The tube L lines of Ag and Rh are particularly effective for exciting the light elements in
the range 12.5 keV. Once higher-energy lines (e.g., FeK at 6.4 keV) are sought, the L
lines provide no excitation, as they are lower in energy than the absorption edge (FeK at
7.11 keV) and the bremsstrahlung hump provides the excitation. Despite the highest in-
tensity of the bremsstrahlung being lower than AgL lines, it is the integrated excitation
spectrum, higher in energy than the absorption edge, which is important. When the x-ray
tube kV exceeds the energy of the K edge of the tube anode material, the characteristic K
lines will begin to dominate the excitation spectrum. This is shown in Figure 6, which
depicts the scattered excitation spectrum with the tube operated at 35 kV, at which the
AgK characteristic lines are strongly excited.
The basic sensitivity for analysis is governed by the kV applied to the x-ray tube and
an example of this is seen in Figure 4. The sample was a pressed pellet and the kV was set
in the range 616 kV using a vacuum path in the spectrometer and no primary beam filter.
At each kV, the rhodium-target x-ray tube current was adjusted to give 50% deadtime.
The K-series line intensities were used and the sensitivity for each element has been nor-
malized to the value at 10 kV. From the plots, the following are clear:
The Si intensity falls off as the kV increases. This is due to the increasing distance of
the tube main spectrum from the SiK absorption edge and to the increased relative
contribution of the other elements to the total counts in the spectrum. This latter
Copyright © 2002 Marcel Dekker, Inc.
Figure 4 Effect of kV on sensitivity for the major elements in a geological sample (see text for
details).
factor is important and derives from the limited total count rate for the whole
spectrum.
The FeK lines are not excited until at least 8 kV is applied. The FeK absorption edge
is at 7.111 keV and the maximum sensitivity for Fe is at approximately twice this
value.
The general rule of thumb in which the kV is set to twice the highest absorption-edge
energy is a reasonable place to start. However, as more elements are sought across a wider
energy range, there usually comes a point at which the ‘‘balance’’ of the spectrum is un-
favorable for the analytical problem being solved. When this happens, split the spectrum
into two or more areas of interest and apply a kV to best suit each of these element ranges.
In any event, as the kV and the number of elements increase, the importance of using a
primary beam filter and splitting the spectrum into optimized energy regions increases.
absorption coefficient for its own characteristic lines, as these lie immediately below the
absorption-edge jump. In the case of Ag, the AgKa1,2 lines are at 22.1 keV and the
absorption coefficient of Ag for this energy is only 14 cm2=g.
The effect of two silver primary filters, of different thickness, on the spectrum from a
silver anode x-ray tube operated at 35 kV is shown in Figure 6. The unfiltered 35-kV
excitation (curve 1) provides equally intense AgL and AgK characteristic lines and a broad
bremsstrahlung hump centered on 1215 keV. Any element in the 515 keV range will
have the benefit of excitation by both the AgK characteristic lines and the bremsstrahlung.
However, there exists the disadvantage of sitting on scattered background derived from
the radiation in the bremsstrahlung hump. With the thin Ag filter in place (curve 2), the
AgL characteristic lines are completely absorbed and the bremsstrahlung hump is reduced
to a low-energy tail on the AgK characteristic lines. The resulting excitation yields high
P=B ratios for lines in the energy range 412 keV. Use of the thicker Ag filter (curve 3),
reduces even further the residual bremsstrahlung below the AgK lines and provides
pseudomonochromatic radiation, which yields high P=B ratios for lines in the energy range
515 keV. The strongly filtered AgK line excitation provides excellent sensitivity and
background reduction and a clear, intense Compton-scatter peak, which is good for trace
analysis where a matrix correction based on Compton scatter is used. The same can be said
for anode and filter materials of Mo, Rh, and Pd, which also provide characteristic K lines
in this important region of the spectrum.
A specific and well-documented example of optimum selection of x-ray tube anode,
excitation kV, and primary beam filter can be seen in work by Potts et al. (1986). In that
work, a Co anode x-ray tube and an iron primary beam filter were used to remove the
Copyright © 2002 Marcel Dekker, Inc.
Figure 6 Scattered excitation spectrum provided by a silver target x-ray tube operated at 35 kV.
The plots show the effect of two thicknesses of Ag primary beam filter: (1) unfiltered; (2) thin Ag
filter; (3) thick Ag filter.
CoKb line which would otherwise strongly excite the Fe that typically dominates this part
of the spectrum in geological analysis.
Most modern EDXRF systems provide the ability to use more than one set of
conditions per analysis. Depending on the analytical needs, the kV and filter are chosen to
optimize the important parts of the spectrum and the data from spectra at each of these
conditions are consolidated to provide the best P=B values. In order to establish the op-
timum conditions and make the best use of analysis time, the best strategy is to select a
typical sample and then to make measurements with the different conditions available. Use
criteria such as detection limits to decide upon the optimum conditions for the required
analysis problem. Often, the instrument manufacturer’s recommendations can con-
veniently be taken as the starting point for any investigation.
D. X-ray Tubes
X-rays were first discovered over a century ago (see Michette and Pfauntsch, 1996) by
Röntgen (1896a, 1896b), and since that time, they have become used in a great variety of
applications. These applications encompass medical and dental x-ray radiology and
imaging, industrial and baggage inspection, industrial nondestructive testing (NDT),
industrial thickness gauging, and x-ray spectrometry in all its guises. In most, if not all,
of these fields, the convenience of the high-voltage x-ray tube has substantial benefits in
terms of convenience, safety, and availability of appropriate spectral and intensity output.
The modern x-ray tube used for EDXRF analysis is based on a design introduced by
Coolidge as long ago as 1913 (Coolidge, 1913). The basic design of x-ray tubes typically
used in EDXRF instrumentation consists of the following:
A radiation-shielded glass envelope
An independent tungsten filament with current control unit
A high-melting-point metallic anode
A high-voltage connection to the anode
A beryllium foil exit window
Some means of dissipating the heat generated
In order to operate, the x-ray tube requires a controllable, stable high-voltage power
supply capable of providing typically 560 kV. In addition, there is a low-voltage power
supply for providing current to heat the filament, thereby controlling the electron beam
current and output intensity. X-ray tubes used for EDXRF analysis are typically operated
with a power in the range of 1400 W. For direct excitation, the power output is typically,
no greater than 50 W. Reproducibility of analytical results is achieved for x-ray tube ex-
citation by operating for a measured amount of time at a fixed emission current (dc op-
eration). This approach depends on the reasonable assumption that the total output x-ray
flux is proportional to the number of electrons striking the anode.
In some types of x-ray tube, a control grid can be used to modulate or pulse the
emission current. This can be used to modulate the electron beam current to ensure
constant output of radiation from the tube in dc operation or it can be used to rapidly
Copyright © 2002 Marcel Dekker, Inc.
switch the electron beam on and off for pulsed operation. If the pulsing is synchronized
with x-ray detection, increased output countrate can be achieved by reducing pulse pileup
(see Sec. IV.C).
The key features of the x-ray tube used in EDXRF instruments are as follows:
Easily shielded and safe to operate
Compact and able to be placed close to the sample
Stable output over an extended period, typically months
Well-defined spectral output
High-voltage controllable in small increments, typically 1 kV, throughout the
operating range (typically 550 kV)
Beam current controllable in small increments, typically 10 mA, throughout the
operating range (typically 101000 mA).
The geometry and designs typically used and the characteristics of each are described in
the following sections. Further reading on the subject can be found in Skillicorn (1982,
1983).
1. Side-Window Geometry
Side-window geometry is the oldest and, perhaps, the most common x-ray tube geometry
used in EDXRF instruments and it is suitable for both high-kV and high-power operation.
Figure 7 shows the essential details of this design.
The geometric coupling of tube anode spot to sample is limited first by the distance
of the spot itself from the exit window and then, potentially, by the tube encapsulation
itself. Typically, these tubes are operated in the range 550 kV. They can dissipate 50 W
with the need only for forced-air cooling in order to keep the tube external body below a
temperature of, say, 55 C. It is important not to exceed the maximum temperature stated
by the x-ray tube manufacturer. Higher powers (up to 100 W or 120 W) normally need
some form of liquid cooling, which is typically achieved through the use of a closed-cycle
liquid heat exchanger.
The energy spectrum that is output from a side-window x-ray tube is governed by the
tube anode material, the angle of the anode, the kV applied to it, and the exit window’s
material and thickness. Tube anode materials are usually Rh, Ag, or W, although Ti, Cr,
Co, Cu, Mo, Pd, or Au is also available. The selection of anode material depends on the
intended application, but Rh, Ag, and Pd are generally the most popular, as they are
highly compatible with tube manufacture and operation. In addition, their K and L
characteristic emission lines are in regions of the energy spectrum where they offer high
sensitivity with minimum spectral interference. Dual-target x-ray tubes have been used in
attempts to maximize flexibility and performance. These efforts have met with limited
success in EDXRF, in part due to the spectral cross-contamination that is typically
experienced in the designs used to date. There has been, however, considerable success in
the use of dual-target x-ray tubes for WDXRF, where the problems of spectral con-
tamination have much less impact. Typically, the anode is cut at an angle of 20 35 from
normal to the incident electron beam and the exit window is usually made from beryllium
foil of thickness in the range 50250 mm.
The tube output spectrum can be measured or estimated and such information is
important in analytical correction procedures, particularly those using fundamental
parameter calculations for which accurate and computationally convenient tube output
spectrum models are required (see Chapter 2, Sec. I.B for further details.)
Copyright © 2002 Marcel Dekker, Inc.
Figure 7 Side-window x-ray tube.
2. End-Window Geometry
The design features of such an x-ray tube are shown in Figure 8. The same factors of
operation and tube spectrum apply to this type of tube as was applied to side-window
tubes. The tube spectrum is altered by the fact that the takeoff angle is normal to the anode
surface and so there is minimal self-absorption by the target material. This high takeoff
angle makes such a design particularly good for low-energy excitation. One of the im-
portant benefits of the end-window design is the very tight coupling of anode spot to
sample that can be achieved with a ‘‘sharp-nose’’ design. The spot-to window coupling can
be close with suitable electron optics (e.g., 12 mm in the Varian EG50 design). The use of a
high positive voltage on the target minimizes the effect which repelled electrons would
have of causing an unacceptable amount of exit-window heating. This reduced heating
allows thinner beryllium windows to be used, which further improves the low-energy
x-ray output.
3. TransmissionTarget Geometry
The design features of such an x-ray tube are shown in Figure 9. In this type of x-ray tube,
the geometry is basically end window, but in this embodiment, there is a metal foil target
parallel to and inside the beryllium exit window. The incident high-voltage electron beam
strikes and enters one side of the thin-film target and the resulting radiation is heavily
filtered as it passes through the foil and exits the far side of the target. This novel design
was initially developed (Jaklevic et al., 1972) in order to reduce the bremsstrahlung from
the x-ray tube, which would then reduce the scattered background in the spectrum and
thereby improve detection limits. The initial design used a molybdenum transmission
target of 0.12 mm thickness and, when operated at 42 kV and up to 25 W, was shown to
give an output spectrum consisting principally of characteristic MoK radiation. The
output spectrum was comparable to an 125I radioisotope source coupled to a molyb-
denum-foil secondary target.
This type of x-ray tube was made commercially available by WatkinsJohnson
(Hershyn, 1975a, 1975b). The spectrum output from this type of tube was investigated by
Zulliger and Stewart (1975), who calculated and measured spectra from Cr, Mo, and W
transmission target x-ray tubes. The targets were either thin (50 mm) metal foils or thin
films (510 mm) on a beryllium substrate of 0.250.5 mm thickness. The thin targets mean
that only low-power tubes (< 100 W typically) can be made successfully. These authors
found, as predicted, a substantial reduction in bremsstrahlung compared to the char-
acteristic lines and quoted detection limits for a number of applications. The Cr target was
found, not unsurprisingly, to be best for the light elements S and Ca. Although the energy
range covered was claimed to be wide, the intrinsic fixed filtration of this design makes it a
less valuable general excitation source compared to a side- or end-window tube with an
external, selectable filter. A novel, multiple-target transmission tube has even been de-
scribed (Kis-Varga, 1988) in which the five anode materials (Fe, Cu, Ge, Mo, and Ag) were
changed by a magnetic mechanism.
The intrinsic filtration of the low-energy bremsstrahlung makes this type of tube
much less suitable for excitation of low energy (i.e., < 2-keV) x-ray lines. Consequently,
their usage is limited to specific applications where the intrinsic filtration and potential for
close geometric coupling of the tube anode spot to the sample are major benefits. In fact,
in one design, the WatkinsJohnson tubes did not have a target closely coupled to the
exit window, which, when combined with the intrinsic lossess in the target, led to low
output flux from the tubes and, consequently, reduced sensitivity. Finally, these tubes are
not useful when operated at low voltages (Skillicorn, 1982) due to strong absorption in the
target. This makes them much less sensitive than solid-target, side- or end-window tubes
for selectively exciting the light elements in the presence of large amounts of heavier
elements.
Copyright © 2002 Marcel Dekker, Inc.
III. SEMICONDUCTOR DETECTORS
In this type of detector, the total ionization produced by each x-ray photon striking the
detector is converted to a voltage signal with amplitude proportional to the incident en-
ergy. Specially designed preamplification and processing electronics are employed to
maintain the linearity of the voltage signal with respect to the original charge pulse.
A multichannel analyzer accumulates an energy spectrum of the sequential events in a
histogram memory. Because the energy analysis does not depend in any way on the dif-
fraction or focusing of the x-rays by the detector, the geometry of the system is relatively
insensitive to the placement of the detector with respect to the sample. An exception to this
is in polarized excitation XRF (Chapter 10), where the orientation angle can be used to
suppress scattered background. This insensitivity to geometry and ability to place the
detector element close to the sample provides for a large solid angle and high geometric
efficiency. In addition, the mechanism by which the ionization signal is measured is not
restricted to a narrow energy region, thus allowing the simultaneous detection of x-rays
over a wide dynamic range of the emission spectrum. The principal advantages of EDXRF
derive from its capability for simultaneous detection, with high geometric efficiency, of
characteristic x-rays from multiple elements.
cases following the emission of Auger electrons or multiple low-energy x-rays and sub-
sequent re-absorption. The result of these multiple ionization process is the essentially
instantaneous production of a large number of free electronhole pairs in the sensitive
volume of the detector structure. These free-charge carriers form as a cloud from which
they are separated by the field gradient. The front detector electrical contact, being biased
at 500 V, for example, attracts the hole and repels the electrons, the latter negative
charge cloud is swept to the rear contact and converted, by means of the FET, to an
amplified voltage pulse. The number of charge carriers produced is directly proportional
to the energy of the x-ray photon incident on the detector. Thus, the number of charges
collected results in a voltage pulse whose magnitude is, in turn, directly proportional to the
energy of the detected photon.
B. Crystal Materials
The characteristics of energy-dispersive semiconductor x-ray spectrometers derive pri-
marily from the properties of the materials used to make the device itself. Although the
most common material for semiconductor detectors in EDXRF is silicon, devices based on
the use of germanium or compound semiconductors such as mercury(II) iodide (HgI2),
CdTe, CdZnTe, or GaAs have also been used to varying degrees. In particular, mercury
(II) iodide detectors (Swierkowski et al., 1974; Slapa et al., 1976; Ponpon and Sieskind,
1996) have been used successfully in both research and commercial products for many
years. Early devices with adequate performance at room temperature were flown on space
missions. More recently, devices have been used in hand-held EDXRF spectrometers
where the benefit of ambient temperature operation, with or without thermoelectric
(Peltier) cooling of the field-effect transistor (FET), have been successfully exploited.
Iwanczyk et al. (1996) compared the performance of HgI2, Si PIN photoiodide, and CdTe
detectors for this latter, important application. Large arrays of up to 100 HgI2 detectors
Copyright © 2002 Marcel Dekker, Inc.
have also been fabricated (Iwanczyk et al., 1995; Patt et al., 1995). Dabrowski et al. (1981)
have been particularly successful in designing, fabricating, and characterizing HgI2 de-
tectors throughout the past two decades and their work should be consulted for further
details of this important type of EDXRF detector. Although high-purity germanium
(HPGe) is widely used for detectors in electron microscopes as a result of its slightly better
resolution than Si(Li) detectors, the use of HPGe detectors is less popular in photon-
excited EDXRF spectrometry. One reason for this lack of acceptance may lie in the high
escape peak intensities (see Sec. III.E.1) which can strongly interfere in trace analysis.
Table 2 lists properties relevant to x-ray detection for Si, Ge, and HgI2, which are the
materials most commonly employed in commercial EDXRF spectrometers. A variety of
other compound semiconductor materials, such as GaAs, CdTe, and CdZnTe, have been
used for applications in which specific properties are particularly beneficial. These prop-
erties include the following:
Higher atomic number for increased photoelectric cross section
Larger band gap for lower thermal leakage at room temperature
A lower band gap corresponding to a smaller average energy for creation of
holeelectron pairs (Knoll, 1979), which offers improved energy resolution.
Recalling the direct relationship described earlier between photon energy and number of
charge carriers and, therefore, the detectors signal resulting from detection of a particular
x-ray photon, one can use the data from Table 2 in the following example for detection of
CaKa1 (KL3), which has an energy of 3.691 keV:
No. electronhole pairs produced ¼ 3691=3.86 ¼ 956
Charge on electron ¼ 1.661019 C
Therefore,
Charge from CaKa1 interaction ¼ 1.5361016 C
Assuming a feedback capacitance in the first stage amplification of 0.1 pF, the
resulting output voltage pulse is 1.53 mV.
From this, we can see the very small magnitude of the signals involved. This has led to very
great efforts having been made in the design of detector and signal processing systems to
minimize all sources of electronic noise which would otherwise interfere very strongly with
the signal being sought.
In addition to the conventional Si(Li) detector, Si PIN (Haselberger et al., 1996;
Cesareo et al., 1996) and Si drift detectors (Lechner et al., 1996; Bertuccio et al., 1996)
have also become available in recent years. The active area of these devices is generally
C. Cooling Systems
The need to minimize the electronic noise of the detection system is paramount for ob-
taining the best energy resolution. The main source of noise in the detector itself is leakage
current, which derives from the generation of charge carriers in the absence of x-rays
through, for example, thermal vibrations of the detector crystal lattice. Whatever the
cooling system used, the intention of it is to minimize leakage, and the lower the tem-
perature, the less leakage will occur. The boiling point of liquid nitrogen (77 K) is a typical
temperature used for the detector crystal, although the FET may require a different and
often higher temperature to minimize its noise contribution and, therefore, obtain opti-
mum detector resolution.
The need for reduced operating temperature results in the need for careful design in
the insulation from ambient temperatures of the detector crystal. Effective insulation is
typically achieved by enclosing the detector head (crystal and FET) assembly in a evac-
uated cryostat. In order for the low-energy x-rays (< 2 keV) to reach the detector crystal,
the vacuum is usually retained by a thin (typically 550 mm) beryllium entrance window
(Fig. 11). In recent years, the thinnest windows (5, 8, and 12 mm) have become available
with coatings which ensure that there is no leakage of He through the Be window. Any
helium getting through the window would degrade the high vacuum within the cryostat
and lead to increased temperature, leakage current, and deteriorating performance.
The original and perhaps simplest way of cooling a detector unit is to attach the
measurement head via a cold-finger assembly to a dewar vessel containing liquid nitrogen
(LN). The dewar itself is typically a complex technological component if it is to provide
high performance and long operational lifetime. The modern LN cryostat delivers opti-
mum resolution and the ability to be cycled through multiple warm-up=cooldown cycles
without degradation in performance. The inevitable and natural boiling-off from the
dewar of LN over a period (1 L per day is typical) means that the dewar needs regular
replenishment. In order to minimize the frequency of replenishing LN, a large dewar is
preferable, although this desire for a long LN holding time may lead to a rather bulky
system. LN volumes of 715 L are typical, although some large systems may use volumes
up to 25 L and some small, hand-held units may use 1 L or less. In any event, the fol-
lowing are important:
Use only liquid nitrogen: Liquid air, although delivering the required temperature
range, will fractionate, yielding O2, which is an explosion hazard in the chemical
laboratory.
Ensure that there is no ice or dirt in the LN used. The presence of ice or dirt in the
dewar can cause microphony, which can degrade energy resolution. Ice and dirt
can be eliminated by pouring the LN through a funnel equipped with a wire
screen (e.g., a motor fuel metal filter funnel).
Observe safe LN handling procedures. Avoid overfilling and spillages as liquid
nitrogen can cause embrittlement of polymers and cable insulation or result in
Copyright © 2002 Marcel Dekker, Inc.
Figure 11 Cross section of a groove-type Si(Li) detector mounted in its cryostat end cap.
D. Energy Resolution
The energy resolution of the semiconductor spectrometer system determines the ability of
a given system to resolve characteristic x-rays from multielement samples. It is normally
defined as the full width at half-maximum (FWHM) of the pulseheight distribution
measured for a monoenergetic x-ray at a specified energy. A convenient choice of x-ray
energy at which resolution is quoted is the weighted mean MnKa line at 5.895 keV, as this
emission line is readily available from 55Fe radioisotope sources. The MnK x-rays are of
sufficiently low energy that the contribution to the FWHM of the unresolved Ka1 þ Ka2
doublet (5.898 and 5.887 keV, respectively) and the intrinsic width of the emission lines
can be neglected. This is not true for higher-energy characteristic x-rays (e.g., MoK lines)
for which the separation of the Ka1 þ Ka2 doublet and the intrinsic width of the emission
line must be considered. Figure 12 shows a typical pulseheight spectrum of MnK x-rays
taken from an 55Fe source.
Typical state-of-the-art 30-mm2 Si(Li) detectors achieve a FWHM better than
140 eV at 5.9 keV, although values as low as 130 eV are achievable. However, this number
is but one indicator of the quality of an EDXRF detector system, and other factors such as
maximum count rate or the presence of background and artifacts may be more important
in many analytical applications. Peltier-cooled Si PIN detectors offer FWHM values
typically in the range 165220 eV, whereas Si drift detectors claim < 150 eV. The inter-
dependence of energy resolution and count rate is discussed below, although it is worth
noting at this point that the user can exert a large degree of control over this by making an
appropriate selection of pulse processor settings.
If one neglects the natural linewidth of the x-ray lines, the instrumental energy re-
solution of a semiconductor detector x-ray spectrometer is a function of two independent
factors. One of these is determined by the properties of the detector itself; the other is
dependent on the nature of the electronic pulse processing employed. In some systems,
a pulser or strobe signal may be injected into the measurement system in order to monitor
the resolution of the electronic system independent of any peak broadening due to the
detector itself.
The measured FWHM of the x-ray line (DETotal) is the sum, in quadrature, of the
contribution due solely to detector processes (DEDet) and that associated with the elec-
tronic pulse processing system (DEElec):
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
DETotal ¼ DE2Det þ DE2Elec ð1Þ
The DEDet component is determined by the statistics of the free-charge production process
occurring in the depleted volume of the diode. The average number of electronhole pairs
Copyright © 2002 Marcel Dekker, Inc.
Figure 12 Pulseheight spectrum of MnK x-rays taken with a 30-mm2 Si(Li) detector.
produced by an incident photon can be calculated as the total photon energy divided by the
mean energy required to produce a single electronhole pair (see Sec. III.B). If the fluc-
tuation in this average were governed by Poisson statistics, the standard deviation would be
rffiffiffiffi
pffiffiffi E
s¼ n¼ ð2Þ
e
In semiconductor devices, the details of the energy-loss process are such that the individual
events are not strictly independent and a departure from Poisson behavior is observed.
This departure is taken into account by the inclusion of the Fano factor in the expression
for the detector contribution to the FWHM:
rffiffiffiffiffiffiffi
pffiffiffiffiffiffi FE
s ¼ Fn ¼ ð3Þ
e
Taking
sðEÞ pffiffiffiffiffiffi
¼ Fn ð4Þ
E
and, rearranging,
pffiffiffiffiffiffiffiffi
DEDet ¼ 2:35 FeE ð5Þ
where e is the average energy required to produce a free electronhole pair, E is the energy
of the photon, F is the Fano factor (see Table 2), and the factor 2.35 converts the standard
deviation to FWHM for a Poisson distribution.
A typical value for the detector contribution to resolution, also known as the disper-
sion, is 120. Examination of the values of e and F listed in Table 2 shows that for an
Copyright © 2002 Marcel Dekker, Inc.
equivalent energy, the detector contribution to the resolution is 28% less for the case of Ge
compared to Si. This results in better energy resolution being obtained for HPGe detectors
and could be an important consideration in the choice of detector for certain applications.
However, this theoretical advantage in energy resolution of Ge over Si(Li) is mitigated
in many systems because of the dominance of electronic noise (e.g., from the FET). In ad-
dition, the very high GeK escape peak probability (see Sec. III.E.1) is likely to be a serious
limitation to the use of HPGe detectors in most EDXRF work. HPGe detectors are widely
used in electron microscopes, where energies < 10 keV are likely to be of most interest.
E. Spectrum Artifacts
The simplicity of the element line spectra encountered in XRF spectrometry is one of its
major benefits when compared with other atomic spectroscopy methods. However, a few
peaks can arise in the EDXRF spectrum from sources other than the elements in the
sample and it is wise to be aware of their nature. Unlike basic detector efficiency and
spectrum background, which are defined by the selection of detector window and the
detector manufacturing process, the end user can exert some control over the effects of
spectrum artifacts. From the spectra presented so far, it is clear that the bremsstrahlung
continuum and the Compton (and Rayleigh)-scatter peaks arise from the interaction of the
exciting radiation with the sample. These features are not regarded as artifacts.
1. Escape Peaks
For an incident x-ray with energy higher than the SiK absorption edge, the detection
process will involve the generation, through x-ray fluorescence of the detector material, of
SiK x-rays. The vast majority of these will themselves immediately be absorbed in the
detector volume and contribute to the overall charge collected for the original incident
x-ray photon. This is the normal mechanism of x-ray detection, which was described
earlier. There is, however, a finite probability that the SiKa x-rays produced will escape
from the detector volume and not contribute to the charge collected for the original
photon that was detected. The resulting energy detected will be reduced by 1.740 keV,
which is the energy of the SiKa x-ray that escaped. The greatest probability of escape arises
close to the front of the detector crystal from which SiK x-rays can more easily escape.
X-rays detected far within the volume will still generate SiK x-rays, but they will all be
absorbed before they can reach the outer surface of the crystal and escape. The probability
of generating the SiKb x-ray and of it escaping is extremely low and this effect can safely be
ignored. The result of the SiK escape process is a peak in the collected spectrum, which is
1.74 keV lower in energy than the parent peak. The Si escape peaks for TiKa and TiKb are
clearly seen, 1.74 keV below the TiK lines, in Figure 13 (indicated by 1 and 2, respectively)
(see also Table 3). In many detectors, the incomplete charge-collection tail on the low-
energy side of the parent peak will obscure the Si escape peak from TiKb.
The escape probability is highest for lines closest to the SiK absorption edge at
1.838 keV and may be calculated as described by Statham (1976) and Dyson (1974). The
intensity falls rapidly from around 3% of the parent peak at 23 keV to around 0.1% of
the parent peak at 10 keV. At energies above 10 keV, the effect is negligible, but below this
energy, it is important that spectrum-processing software takes account of the peak.
Typically, the processing software will make a correction in which the Si escape peak
is removed and the most comprehensive packages will reinstate the lost intensity to the
parent peak. If such packages are available, then make use of them before carrying out
Copyright © 2002 Marcel Dekker, Inc.
Figure13 Peak artifacts from a Ti spectrum collected by a Si(Li) detector (see Table 3 for details).
qualitative analysis in order to avoid classic errors such as the Si escape peaks from ZnK
being identified as Co!
In the case of HPGe, the escape probability for GeL lines is negligible in XRF and
can be ignored. The escape of the GeK lines has, however, a very high probability, and for
Se, the escape peak intensity is around 16% of the parent peak. This is due to the com-
bination of the high detector efficiency for GeK x-rays and the much greater depth within
the crystal from which the GeK lines can escape. There is further complication in that
there are many GeK lines and each of these lines will result in an escape peak. The
complexity and intensity of Ge escape peaks can easily be seen in the spectrum from a
molybdenum oxide sample, shown in Figure 14.
The principal GeK lines and their characteristic energies are as follows:
GeKa: 9.876 keV
GeKb1: 10.981 keV
GeKb2: 11.100 keV
The assignments of the numbered GeK escape peaks arising from the detection of the
MoK lines are listed in Table 4.
The energy and intensity of these GeK escape peaks interferes very strongly over a
wide and important region of the spectrum. The interference is considerable from the GeK
escape peaks of Nb and Mo in stainless steels and corrections for these intense escape
peaks are not sufficiently accurate to retain the intrinsic performance of EDXRF for what
would be a reliable and straightforward analysis using a Si(Li) detector. The severe lim-
itation posed by the Ge escape peaks in EDXRF is also demonstrated in Figure 15 (see
also Table 5) for a typical determination of minor and trace heavy metals in a low-Z
matrix. In this case, the complex GeK escape peaks from the Rh backscatter peaks
completely dominate the important part of the spectrum for trace analysis.
In conclusion, the Si escape peak is small and easily corrected, but the number and
high intensity of the GeK escape peaks severely limit the value of the HPGe detector for
tube-excited EDXRF. In the case of energy-dispersive (ED) microanalysis, however,
Table 4 GeK Escape Peak Assignments and Relative Intensity for Figure 14
the HPGe detector has benefits in that the intensity of the GeL escape peaks is negligible
and the energy resolution is better than that of Si(Li) detectors.
2. Sum Peaks
Sum peaks arise from a specific form of peak pileup (see Sec. IV.C) where two events from
high-intensity peaks arrive in the pulse processing electronics so close together in time that
the pileup inspector cannot recognize them as two events. The effect of this is for the signals to
be seen as one and for them to be registered at the energy that is the sum of the two. The sum
peaks of Ti are seen in Figure 13, in which the most intense peak (3) is from the sum of two
TiKa events. The smallest peak (5), from the sum of two TiKb events is only visible when the
parent peak is very intense. At first glance, peaks 3 and 4 may appear like a typical Ka, Kb
Peak Assignment
1 RhKa ComptonGeKa
2 RhKa RayleighGeKa
3 RhKb ComptonGeKa
4 RhKb RayleighGeKa
3. Diffraction Peaks
Whenever a crystalline sample is measured in an EDXRF spectrometer, there is the
possibility that the conditions for Bragg diffraction will be met. This is exacerbated by the
divergent geometry of the incident and detected beam and the variety of excitation con-
ditions available. The worst cases are where unfiltered primary radiation is used, as there
are then many energies and angles which increase the probability of meeting the Bragg
condition for a crystalline component in the sample. The more monochromatic the ex-
citing beam is, the less likely it is that the Bragg condition will be met. Figure 16 shows the
diffraction peak obtained from a boric acid pellet which was excited with a silver anode
x-ray tube operated at 5 kV with no primary beam filter and in an approximately 90
geometry.
The spectrum is overlaid with one from a sulfur pellet measured under exactly the
same conditions. It can be seen that the diffraction peak is broader than a K series line at
that energy and the low-energy tailing is more pronounced. There is some possibility for
misidentification, but the most likely problem arises in any spectrum processing not setup
to take account of this peak. In cases such as silicon wafers, where the crystal planes are
highly oriented, the sample can be tilted or rotated to minimize the often sharp diffraction
peaks that can be observed. Alternatively, effective use can be made of additional colli-
mation to constrain more closely the angles that can satisfy the Bragg condition or of
alternate excitation conditions, which change or reduce the incident energies available for
diffraction. Figure 17 shows a comparison of the spectra taken from the same two samples
under different excitation conditions.
The increased kV and use of an absorption filter completely change the energy
distribution of the excitation spectrum and eliminate the diffraction peak. This second set
of conditions is particularly effective for lines in this part of the spectrum and effectively
removes problems from diffraction peaks. Once a diffraction peak is identified as causing a
problem, a change of excitation conditions is often the best measure to adopt. The use of
monochromatic or polarized excitation or of secondary targets in Cartesian geometry will
also eliminate diffraction peaks.
Figure 17 Spectrum from boric acid, overlaid with one from a sulfur pellet, showing removal
of the diffraction peak seen in Figure 16. Conditions: Ag target x-ray tube, 15 kV, thin Al primary
beam filter.
associated with the surface layer of the semiconductor material itself. This layer is com-
prised of an evaporated metal contact used to form the rectifying Schottky barrier and a
thin surface layer (the dead layer) of inactive semiconductor material from which charge
cannot be collected. The thickness of evaporated metal, normally Au, can be determined by
direct measurement during the manufacturing process and is typically in the range of 10 nm.
The effective ‘‘dead layer’’ of semiconductor material is a more complicated parameter to
determine. Empirical studies have attempted to measure the absorption inherent in the
semiconductor surface layer. Such studies established that the effective thickness is de-
termined largely by the absorption length of the low-energy photons and the charge-
transport characteristics of the associated ionization products. In the simplest model, there
is competition from the rate of diffusion of the electronhole distribution, against the
gradient of the applied field. For distances near the entry contact, a part of the charge can
diffuse into the contact and be lost to the signal before the electrical field can sweep it to the
opposing electrode. This loss of charge can be interpreted in terms of an effective ‘‘window’’
thickness of typically 0.2 mm silicon equivalent. In addition, thin evaporated contacts are
deposited on the entry surface that can absorb incident low-energy photons.
Detailed experiments have indicated that the charge-collection efficiency for events
near the surface depends in a complex way on low-energy x-ray and charge-transport
properties (Llacer et al., 1977; Goulding, 1977). As a consequence of these entry-window-
related effects, the efficiency for low-energy photons can be reduced. The events lost from
the photopeak can then appear in a continuum background below the full-energy peak,
where they reduce detectability and interfere with spectral analysis. Similarly, secondary
electrons or photons originating from the photoelectric absorption events that occur in-
itially in the active volume can escape. This results in a collected charge pulse of reduced
Copyright © 2002 Marcel Dekker, Inc.
amplitude. An easily observed manifestation involves the observation of discrete peaks in
the spectrum. Continuous-loss processes involving electron escape can also occur, al-
though the probability is small. These and related mechanisms that reduce the amplitude
for a given event can lower the photopeak efficiency relative to the simple model described.
There have been studies in which the efficiencies of Si(Li) and Ge have been carefully
measured using calibrated sources of x-rays spanning the energy region of interest (Cohen,
1980; Szoghy et al., 1981; Campbell and McGhee, 1986). These indicate that the maximum
intrinsic photopeak efficiency is reduced by a few percent relative to the curves shown in
Figure 18 and is slightly higher at the upper energy cut off than calculated. These results
substantiate the overall validity of the simple photoelectric absorption model but de-
monstrate the limitations if precise results are required. For laboratory applications of
EDXRF, it is not necessary to explicitly determine the photoelectric efficiency function
because it is included in the overall calibration factor of the instrument. In any event, the
internal efficiency of the detector itself is defined by the design and manufacture of the
detector and this is out of the control of most users.
The efficiency curve shown in Figure 18 illustrates that for conventional EDXRF
measurements, the absorption of fluorescence x-rays in air and in the Be entry window
limits the accessible energy range to photons greater than approximately 2 keV. The ab-
sorption losses in the air path can be significantly reduced by the common practice of
employing a vacuum or He atmosphere. A vacuum path is preferable due to higher x-ray
transmission, and if helium is used, the beryllium entrance window must be able to
eliminate leakage into the cryostat with subsequent loss of its vacuum. Coated beryllium
windows are commonly used nowadays to minimize helium leakage into the cryostat.
The necessity of a beryllium window to maintain vacuum integrity between the
cryostat enclosure and atmospheric pressure does not constitute a serious limitation for
most analyses. However, there are applications in which the detection of x-rays with en-
ergies below 1 keV (e.g., for F) becomes necessary. The relatively high detection limits
achieved by EDXRF systems because of the poor energy resolution (compared with
WDXRF) combined with absorption by the beryllium window generally make EDXRF
the second choice for these very low-energy x-ray lines. A beryllium window thickness of
8 mm is typically used for very light-element analysis, although windows as thin as 5 mm
have been used. The use of extremely thin windows may lead to increasing concerns for
helium integrity and equipment cost.
High-strength thin windows made from low-atomic-number elements that are cap-
able of withstanding a full 1-atm-pressure differential have become commercially available
in the last decade. These include self-supporting 0.5-mm diamond polycrystalline films
and 0.25-mm windows composed of a vapor-deposited amorphous material consisting of
90% boron by weight with nitrogen and oxygen for the remainder. These windows
exhibit significant x-ray transmission for photons well below 1 keV and have found
widespread use in x-ray microanalysis.
G. Detector Background
Effects associated with the partial collection of the photoelectric signal by the detector
have a small effect on its efficiency. However, these incomplete charge-collection processes
can have consequences that are far more serious for analytical performance through their
effect on spectral background. The effect of incomplete charge collection of these events is
to produce tailing on the low-energy side of a peak. In addition, it produces a continuum
of events that appear as a shelf on the low-energy side of major peaks in the spectrum. This
Copyright © 2002 Marcel Dekker, Inc.
background tailing and shelf interfere with the measurement of lower-energy fluorescence
x-rays and is a significant factor in obtaining the lowest detection limits for the determi-
nation of trace quantities on the low-energy side of major peaks.
Studies designed to reduce this background have indicated that in addition to
fundamental x-ray and electron-energy-loss processes, a more significant background
resulting from incomplete charge collection from the detector active volume is normally
the dominant contribution (Goulding et al., 1972; Jaklevic and Goulding, 1972). This
process is an artifact of the detector operation in which the collection of the free charge
from the depleted volume is inhibited as a result of nonuniformities in the applied
electrical field. These nonuniformities are typically associated with edge effects at the
periphery of the cylindrical detector. The incomplete charge-collection background can
be reduced either by external collimators that prevent the incident radiation from in-
teracting in the periphery of the detector or by internal electronic collimation brought
about by the use of a guard-ring structure (Goulding et al., 1972). The tailing can
sometimes be improved by increasing the detector bias, but this is not an option
available to the majority of users. There has been much work on characterizing the
detector line shape and the effects on background and tailing of incomplete charge
collection (Campbell and Wang, 1991; Campbell et al., 1997; Heckel and Scholz, 1987;
Lepy et al., 1997). Such studies have led to a greater understanding of the processes
involved and to peak-fitting procedures that can take account of them. However, im-
provements in this form of background can only be achieved through improvements in
detector design and manufacturing processes.
The P=B ratio of a detector is a standard measurement of detector quality. The
measurement is made using a 55Fe radioisotope source to ensure that there is no other
contribution to background. The measurement cannot be made using x-ray tube excita-
tion, as it is extremely difficult to eliminate the residual bremsstrahlung continuum. The
peak is taken from the most intense channel in the MnKa peak and the background is
taken as the average intensity in the channels from 0.9 to 1.1 keV. A high-quality Si(Li)
detector will deliver a P=B ratio in excess of 10,000:1 and have minimal low-energy tailing.
particularly important. Early detectors used pulsed optical charge restoration, but newer
systems make use of integrated solid-state devices incorporating low-noise-junction FETs
to minimize electronic noise (Statham and Nashashibi, 1988; Lund et al., 1995, Lund et al.,
1996). However, the pulse processing must also achieve a very important goal of ampli-
fying the weak charge signal to a measurable level while suppressing random fluctuations
in the signal amplitude produced by thermal noise. This is achieved by generating a
carefully defined pulse shape in the main amplifier, which restricts the Fourier frequency
components in the final signal in such a way that the signal contributions are emphasized
relative to the noise fluctuations.
The most common pulse shapers employed in modern semiconductor spectro-
meters generate output pulses that are Gaussian, triangular, or cusp shaped (Fairstein
and Hahn, 1965; Kandiah et al., 1972; Landis et al., 1982). Each is capable of achieving
adequate energy resolution for EDXRF analysis; the differences are due mainly to the
effective time interval required for processing a pulse. Because the relative amplitude of
the noise contribution is a strong function of the characteristic time constant associated
with the pulse shaper, the difference between pulse shapes becomes important for ap-
plications in which high counting rates are important. A detailed analysis of the effects
of various pulse-shaping options on spectrometer performance is presented by Goulding
and Landis (1982). More recently, sufficiently fast, sampling analog-to-digital converters
(ADCs) have become available and these have resulted in the availability of digital pulse
processors (Warburton et al., 1988). These devices can implement the optimum noise-
reduction filters in conjunction with much higher count rates than conventional analog
pulse processors.
As pointed out below in the following subsection, however, it is not always desirable
to operate at the minimum-noise shaping time because an unacceptably large dead time
may result. Although EDXRF instruments are designed with a suitable compromise be-
tween energy resolution and count rate capabilities, an understanding of these trade-offs
on the part of the user is important for optimum usage of the instrument.
Copyright © 2002 Marcel Dekker, Inc.
B. Resolution and Count Rate
There exists a great deal of literature on the nature of the detection and pulse processing
systems and the great strides which have been made over the past 20 years to improve
resolution and increase count rate. For many years, the complete integration of the de-
tection and electronics systems within the instrumentation has meant that the individual
choice of and matching by the EDXRF user of detector, main amplifier, ADC unit, and
MCA is no longer required. Once the application is determined, the choice of system will
typically revolve around achievable analytical performance.
The user will typically be left with a relatively straightforward selection of the single
signal processing setting providing the required resolution and count rate
(see Fig. 20). There is only a single setting and, unfortunately, resolution and rate go in
opposite directions when this setting is changed. In one direction, the amplifier time
constant will increase, which gives better resolution (lower FWHM) at the expense of a
lower count rate into the MCA memory. In the opposite direction, resolution degrades but
count rate increases. Once the EDXRF system is chosen, this control is really the only way
in which the user can influence resolution and count rate.
It is a little like choosing a car where you select the type of vehicle most suited to
your use. That vehicle then has a gearing selection system that allows you some degree of
trade-off in control. Low gears give highest accuracy in maneuverability (i.e., resolution)
but lowest speed (i.e., count rate). Conversely, the highest gears give a high speed (rate)
at the expense of accurate maneuverability!
At one end of the range, best resolution is obtained with lowest total count rate,
whereas at the other, highest count rate is delivered with the worst energy resolution.
When severe peak overlap is the dominant limitation to analytical performance, the best
resolution should be selected. In this case, which is typical for the light elements, the
Figure 20 Plots of energy resolution (at 5.9 keV) and count rate for a Si(Li) detector.
C. Pulse Pileup
The count-rate limitations associated with a semiconductor spectrometer are an inherent
property associated with the finite pulse processing time required by the electronic shaping
network. When a random sequence of pulses is incident on the detection system, some of
the events cannot be processed without ambiguity. To appreciate the fundamental nature
of this limitation and its relationship to system performance, some elementary concepts of
electronic pulse processing must be considered.
Figure 21 illustrates the time sequence of pulses that occur at various stages in the
pulse processing chain. Trace A shows the output of the charge-integrating preamplifier.
The steps at times 1, 2, and 3 represent the charge integrals of discrete events. Traces B and
C are the outputs of the fast (short shaping time) amplifier and an associated discriminator
that serves as a timing marker for the individual events. The main shaping amplifier output
is shown as the Gaussian pulse shapes (in trace D). For each event, a total pulse processing
time (neglecting analog-to-digital conversion in the pulseheight analyzer) of td is re-
quired after the arrival of the pulse and before the system is ready to accept the next event.
Although the average counting rate detected by the system can be well below the
frequency defined by the reciprocal of the pulse width td, that the events are statistically
uncorrelated implies that the events are not uniformly distributed in time. There then
exists a probability that two pulses will occur within the same processing time interval.
This is illustrated by the overlap of pulses 2 and 3 in the trace.
At a low average counting rate, this pulse overlap is not a limiting factor. As the
average counting rate increases, however, a point is reached at which there is a significant
probability that a second event will occur before the first event has been fully analyzed.
If the two events occur within a time less than the shaping time of the amplifier, the charge
signals are indistinguishable and an erroneous ‘‘pileup’’ energy signal results.
Modern spectrometers employ some form of rejection circuitry to eliminate these
pileup events from the pulseheight spectrum. Typical systems rely on the inspection of
the fast discriminator output to determine if two pulses have occurred in rapid succes-
sion. Appropriate logic is then employed to gate the output of the processor to eliminate
the resultant ambiguous energy signals. This is shown in traces E and F of Figure 21, in
which the logic signal causes the output to be inhibited when the pulses overlapped to
produce an ambiguous pileup energy output. Because the fast shaper that generates the
discriminator output has inherently poorer energy resolution relative to the slow channel,
the pileup rejection circuit is limited in the minimum energy of pulses it can detect.
However, in typical tube-excited EDXRF, the major fraction of pulses occur from
scattered high-energy radiation and a pileup rejector system is an important feature in
the system design.
The pileup probability is obviously a function of the characteristic shaping time tp,
which, in turn, establishes the effective dead time, td, of the system. This probability is
independent of the details of the specific type of pileup rejection circuit used to eliminate
the ambiguous events. The number of events that experience pileup and are consequently
eliminated from the spectrum can be estimated for the case of a totally random arrival
Copyright © 2002 Marcel Dekker, Inc.
Figure 21 Time sequence of pulses during the processing of a series of x-ray events; (A) the output
of the preamplifier is represented as a series of voltage steps reflecting the integrated charge from the
individual events; (B) the output from a fast shaping amplifier that operates with a shorter time
constant than the main shaping system; (C) a fast discriminator timing pulse derived from B; (D) the
output of the main shaper, with the pileup of pulses 2 and 3 indicated; (E) the dead-time gates; and
(F) the final output with the ambiguous 2 and 3 pileup pulses rejected.
distribution. For a series of events randomly distributed in time with an average frequency
N0, the probability P that no pileup events occur within a characteristic time td after a
given event can be expressed as
P ¼ N0 eN0 td
From this expression, we can calculate the fraction of events transmitted through the
system. Figure 22 is a plot of nonpileup output rate versus input rate expressed in terms of
a characteristic pulse processing time.
The input rate for which the output rate is a maximum is seen to be the reciprocal of
the shaping time. The output rate at this point is reduced by a factor of 1=e. It should be
emphasized that this behavior is a fundamental consequence of random arrival statistics
and a finite measurement time. However, one should be aware that it is always possible to
reduce the characteristic processing time to achieve an increase in counting rate at some
sacrifice in energy resolution. It is also worth noting that, by working at too high a dead
Copyright © 2002 Marcel Dekker, Inc.
Figure 22 Nonpileup output rate (N0) as a function of input rate (N1) scaled as a function of
a characteristic shaping time. Maximum output is 0.37 of the input rate that is equal to the reciprocal
of the time constant.
time, the throughput curve inverts and the output rate begins to fall. Further increase
of the dead time will eventually lead to paralysis of the system and negligible output
countrate.
Although different algorithms can be used to correct for events lost to pileup, there
is no way to eliminate the effect through passive pulse processing (Gedcke, 1972; Hayes
et al., 1978; Statham, 1977). Because different manufacturers of x-ray equipment vary in
their approaches to pileup rejection, the only way to evaluate the throughput of the system
is using a variable intensity source of radiation. The input rate can be determined by
scaling the fast discriminator output, which is sometimes available from the instrument
data system. The output rate can be simultaneously measured in the pulseheight ana-
lyser. A plot of nonpileup output rate versus input rate can then be generated and com-
pared to the ideal case shown in Figure 22.
When the excitation source can be switched off in a time interval that is short com-
pared to the characteristic pulse processing time, it is possible to increase the average output
counting rate by eliminating the effects of pileup. Such pulsed excitation systems rely on the
ability to detect an event in the fast channel and to shut off the excitation before a second
pileup event can occur. In this mode of operation, the output rate can equal the input rate
up to the point at which the system is continuously busy. This method of pileup control has
been implemented using pulsed x-ray tubes (Jaklevic et al., 1976; Stewart et al., 1976),
although the increased complexity of the electron optics tends to lead to increased cost.
D. Dead-Time Correction
The presence of pulse pileup must be considered in any system designed for quantitative
measurements because it causes the efficiency with which pulses are processed to be rate
dependent. EDXRF analytical systems are designed to correct for this discrepancy
through a variety of approaches. The most straightforward involve the use of a live-time
Copyright © 2002 Marcel Dekker, Inc.
clock consisting of a gated oscillator and scaler. The oscillator clock is turned off when the
system is busy processing pulses. In this way, the duration of the measurement in terms of
a live-time interval is extended to correct for those periods when the system is incapable of
processing a pulse. Alternative methods rely on the direct measurement of the fast dis-
criminator pulses to keep track of those events that were missed. An additional correction
is added to compensate for those intervals in which pileup occurs (Bloomfield et al., 1983;
Hayes et al., 1978).
A simple empirical method used to correct for dead-time losses is a measurement of
the ratio of input to output counts as a function of input rate over the range of values
normally encountered. Subsequent analyses require that the input rate be measured for
each experiment. The live-time correction can then be applied using the previously de-
termined response function.
Because all live-time correction methods have some limitation on the range of
counting rates over which they can be used, it is important that methods be devised to
evaluate the precision with which such corrections are generated. A carefully prepared
series of standards of varying concentrations represents a direct approach. A potentially
more precise method involves the use of a single thin-film standard. Variable-mass targets
are placed behind the standard to vary the total counting rate over the range of interest.
If the variable mass target is chosen so that the variable intensity of scattered or fluor-
escence radiation does not induce fluorescence in the thin-film standard, the measured
intensity of fluorescence from the standard should be independent of total counting rate.
The ability of the dead-time correction system to compensate for pileup can then be
empirically evaluated. All modern EDXRF systems incorporate effective dead-time cor-
rection circuitry, which is likely to be at least as accurate as experiments designed to test
this standard feature.
V. SUMMARY
The technology that makes chemical analysis with EDXRF practical is based on the use of
semiconductor x-ray detectors and associated pulse processing and data acquisition sys-
tems. The present chapter attempted to explain to the analyst the basic concepts behind
the operation of these components and the manner in which they influence overall system
performance. The trade-offs one must make between such parameters as excitation con-
ditions, detector resolution, count rate, and other design variables determine how effec-
tively one can tailor a given instrument or experimental apparatus to a specific application.
Furthermore, a thorough understanding of the factors that limit performance should
enable one to implement experimental tests to determine the effectiveness of a particular
approach and evaluate various commercial options.
REFERENCES
I. INTRODUCTION
This chapter deals with (mathematical) procedures to extract relevant information from
acquired x-ray spectra. Smoothing of the spectrum results in a graph that can be easier
interpreted by the human observer. To determine which elements are present in a speci-
men, peak search methods are used. To obtain the analytically important net peak areas of
the fluorescence lines, a variety of methods, ranging from simple summation to sophisti-
cated least-squares-fitting procedures, are to the disposal of the spectroscopist.
Spectrum evaluation is a crucial step in x-ray spectrometry, as much as sample pre-
paration and quantification. As with any analytical technique, the performance of x-ray
fluorescence analysis is determined by the weakest step in the process. Spectrum evaluation
in energy-dispersive x-ray fluorescence analysis (EDXRF) is certainly more critical than in
wavelength-dispersive spectrometry (WDXRF) because of the relatively low resolution of
the solid-state detectors employed. The often-quoted inferior accuracy of EDXRF can, to a
large part, be attributed to errors associated with the evaluation of these spectra. As a
consequence of this, most of the published work in this field deals with ED spectrometry.
Although rate meters and=or strip-chart recorders have been employed in WD
spectrometry for a long time, the processing of ED spectra by means of computers has
always been more evident because of their inherent digital nature. Some of the techniques
to be discussed have their roots in g-ray spectrometry developed mainly in the sixties; for
others (notably the spectrum-fitting procedures), EDXRF has developed its own specia-
lized data processing methodology. The availability of inexpensive and fast personal
computers together with the implementation of mature spectrum evaluation packages on
these machines has brought sophisticated spectrum evaluation within the reach of each
x-ray spectrometry laboratory.
In this chapter, various methods for spectrum evaluation are discussed, with emphasis
on energy-dispersive x-ray spectra. Most of the methods are relevant for x-ray fluorescence,
particle-induced x-ray emission (PIXE), and electron beam x-ray analysis [electron probe x-
ray microanalysis (EPXMA), scanning electron microscopy – energy dispersive x-ray
analysis (SEM–EDX), and analytical electron microscopy (AEM)]. The principles of the
methods and their practical use are discussed. Least-squares fitting, which is of importance
not only for spectrum evaluation but also for qualification procedures, is discussed in detail
in Sec. IX. Section X presents computer implementations of the main algorithms.
The aims of spectrum evaluation is to extract analytically relevant information from ex-
perimental spectra. Obtaining this information is not straightforward because the spectral
data are always corrupted with measurement noise.
1. Characteristic Lines
The characteristic radiation of a particular x-ray line has a Lorentz distribution. Peak
profiles observed with a semiconductor detector are the convolution of this Lorentz dis-
tribution with nearly Gaussian detector response function, giving rise to what is known as
Voigt profile (Wilkinson, 1971). Because the Lorentz width is of the order of only 10 eV
for elements with atomic number below 50, whereas the width of the detector response
function is of the order of 160 eV, a Gauss function is an adequate approximation of the
line profile. Only for K lines of elements such as U and Th does the Lorentz contribution
become significant and need to be taken into account (Gunnink, 1977).
A more close inspection of the peak shape reveals some distinct tailing at the low
energy side of the peak and a shelf extending to zero energy. This is mainly due to
incomplete charge collection caused by detector imperfections (dead layer and regions of
low electric field) as discussed in Chapter 3. The effect is most pronounced for low
energy x-rays. For photons above 15 keV Compton scatter in the detector also con-
tributes to deviation from the Gaussian shape. The distortion caused by incomplete
charge collection has been described theoretically (Joy, 1985; Heckel and Scholz, 1987).
Various functions have been proposed to model the real peak shape more accurately
(Campbell et al., 1987).
The observed emission spectrum of an element is the result of many transitions, as
explained in Chapter 1. The resulting x-ray lines, including possible satellite lines, need to
be considered in the evaluation of an x-ray spectrum. A more detailed discussion on the
representation of the K and L spectra and the peak shape is given in Sec. VII.
2. Continuum
The continuum observed in x-ray spectra results from a variety of processes. The con-
tinuum in electron-induced x-ray spectra is almost completely due to the retardation of the
primary electrons (bremsstrahlung). The intensity distribution of the continuum radiation
emitted by the sample is in first approximation given by Kramer’s formula (Chapter 1).
Absorption in the detector windows and in the sample causes this continuous decreasing
function to fall off at low energies, giving rise to the typical continuum shape observed.
The attenuation of this bremsstrahlung by major elements in the sample also causes ab-
sorption edges to appear in these spectra. Continuum modeling for electron-induced x-ray
spectra has been studied in detail by a number of authors (Statham, 1976a, 1976b,
Smith et al., 1975; Sherry and Vander Sande, 1977; Bloomfield and Love, 1985).
For particle-induced x-ray emission, a similar continuum is observed also mainly due
to secondary electron bremsstrahlung. Other (nuclear) processes contribute here, making
it virtually impossible to derive a real physical model for the continuum. Special absorbers
placed between the sample and detector further alter the shape of the continuum.
Copyright © 2002 Marcel Dekker, Inc.
In x-ray fluorescence, the main source of the continuum is the coherent and incoherent
scattering of the excitation radiation by the sample. The shape can therefore become very
complex and depends both on the initial shape of the excitation spectrum and on the sample
composition. When white radiation is used for the excitation, the continuum is mainly ra-
diative and absorption edges can also be observed. With quasimonoenergetic excitation
(secondary target, radioisotope), the incomplete charge collection of the intense coherently
and incoherently scattered peaks is responsible for most of the continuum (see Chapter 3).
Also here, realistic physical models for the description of the continuum are not used.
The incomplete charge collection of intense fluorescence lines in the spectrum further
complicates the continuum. The cumulative effect of the incomplete charge collection of all
lines causes the apparent continuum at lower energies to be significantly higher that ex-
pected on the basis of the primary continuum generating processes.
3. Escape Peaks
Escape peaks result from the escape of SiK or GeK photons from the detector after
photoelectric absorption of the impinging x-ray photon near the edge regions of the de-
tector. The energy deposited in the detector by the incoming x-ray is diminished with the
energy of the escaping SiK or GeK photon. Typical examples of the interference due to Si
escape peaks are the interference of TiKa (4.51 keV) by the FeKa escape at 4.65 keV and
the interference of FeKa by the CuKa escape.
For a Si(Li) detector, the escape peaks is expected 1.742 keV (SiKa) below the parent
peak. Experimentally, it is observed that the energy difference is slightly but significantly
higher, 1.750 keV (Van Espen et al., 1980). Ge escape peaks are observed 9.876 (GeKa)
and 10.984 keV (GeKb) below the parent peak. The width of the escape peaks is smaller
than the width of the parent peak and corresponds to the spectrometer resolution at the
energy of the escape peak.
The escape fraction f is defined as the number of counts in the escape peak Ne di-
vided by the number of detected counts (escape þ parent). Assuming normal incidence to
the detector and escape only from the front surface, the following formula can be derived
for the escape fraction (Reed and Ware, 1972):
Ne 1 1 m m
f¼ ¼ oK 1 1 K ln 1 þ I ð4Þ
Np þ Ne 2 r mI mK
where mI and mK are the mass-attenuation coefficient of silicon for the impinging and the
SiK radiation, respectively, oK is the K fluorescence yield, and r is the K jump ratio of
silicon. Using 0.047 for the fluorescence yield and 10.8 for the jump ratio, the calculated
escape fraction is in very good agreement with the experimentally determined values for
impinging photons up to 15 keV (Van Espen et al., 1980). Equation (4) is also applicable
for estimating the escape fraction in Ge detectors, provided that the parameters related to
Si are substituted with these of Ge. Knowing the energy, width, and intensity of the escape
peak, corrections for its presence can be made in a straightforward manner.
N_ 12 ¼ 2tN_ 1 N_ 1 ð6Þ
with N_ 11 the count rate (counts=s) in a sum peak due to the coincidence of two x-rays with
the same energy, N_ 12 the count rate of a sum peak resulting from two x-rays with different
energy, and t the pulse-pair resolution time in seconds. Sum peaks are often found when a
few large peaks at lower energy dominate the spectrum. Typical examples are PIXE
spectra of biological and geological samples. The high count rate of the K and CaK lines
produces sum peaks that are easily observed in the high-energy region of the spectrum
where the continuum is low. It is important to note that the intensity of sum peaks is
count-rate dependent, they can be reduced and virtually eliminated by performing the
measurement with a lower primary beam intensity. A method for correcting for the
contribution of sum peaks in least-squares fitting has been proposed by Johansson (1982)
and is discussed further in Sec. VII.
6. Other Artifacts
A number of other features might appear in an x-ray spectrum and can cause problems
during the spectrum evaluation. In the K x-ray spectra of elements with atomic number
between 20 and 40, one can detect a peaklike structure with a rather poorly defined
maximum and a slowly declining tail (Van Espen et al., 1979a). This structure is due to the
KLL radiative Auger transition, which is an alternative decay mode of the K vacancy.
The maximum is observed at the energy of the KLL Auger electron transition energy.
The intensity of the radiative Auger structure varies from approximately 1% of the Ka line
for elements below Ca to 0.1% for elements above Zn. For chlorine and lower atomic
number elements, the radiative Auger band overlaps with the Ka peak. In most analytical
applications, this effect will not cause serious problems. The structure can be considered as
part of non-Gaussian peak tail.
The scattering of the excitation radiation in x-ray fluorescence is responsible for most
of the continuum observed in the spectrum. When characteristic lines are present in the
excitation spectrum, two peaks might be observed. The Rayleigh (coherently)-scattered
peak has a position and width as expected for a normal fluorescence line. The Compton
(incoherently scattered) peak is shifted to lower energies according to the well-known
Compton formula and is much broader than a normal characteristic line at that energy.
This broader structure, resulting from scattering over a range of angles and Doppler effect,
Copyright © 2002 Marcel Dekker, Inc.
is difficult to model analytically (Van Dyck and Van Grieken, 1983). The structure is
further complicated by multiple scattering phenomena in the sample (Vincze et al., 1999).
Apart from these commonly encountered scattering processes, it is possible to detect x-
ray Raman scattering (Van Espen et al., 1979b). Again, a bandlike structure is observed with
an energy maximum given by the incident photon energy minus the electron-binding energy.
The Raman effect is most prominently present when exciting elements with atomic number
Z 7 2 to Z þ 7 with the K radiation of element Z. In this case, Raman scattering occurs on L
electrons. For x-ray excitation energies between 15 and 25 keV, the Raman scattering on the
K electrons of Al to Cl can also be observed. Because of its high-energy edge, the effect may
appear as a peak in the spectrum with possible erroneous identification as a fluorescence line.
The intensity of the Raman band increases as the incident photon energy comes closer to the
binding energy of the electron. The observed intensity can amount to as much as 10% of the
L fluorescence intensity for the elements Rh to Cs when excitation with MoK x-rays is used.
When the excitation source is highly collimated diffraction peaks can be observed in
the x-ray fluorescence spectrum of crystalline materials. It is often difficult to deal with
these diffraction patterns. They can interfere with the fluorescence lines or even be mis-
interpreted as being fluorescence lines, giving rise to false identification of elements.
Spectrum processing refers to mathematical techniques that alter the outlook of the
spectral data. This is often done, using some digital filter, to reduce the noise, to locate
peaks, or to suppress the continuum. In this section, various methods of filtering are
discussed. Because of its relation to the frequency domain, the concept of Fourier trans-
formation is introduced first.
and gives an idea about the dominant frequencies in the spectrum. Because there are n
different nonzero real and imaginary coefficients, no information is lost by the Fourier
transform and the inverse transformation is always possible:
X
n1
j2pux
fðxÞ ¼ FðuÞ exp ð9Þ
u¼0
n
Figure 3 shows that power spectrum of the pulseheight distribution in Figure 2 (a single
Gaussian on a constant continuum). The frequency (inverse channel number) is defined as
u=n, with n ¼ 256 and u ¼ 0, . . . , n=2. The amplitude of the zero frequency jFð0Þj2 , which is
equal to the average of the spectrum, is not shown. The dominating low frequencies
originate from the continuum and from the Gaussian peak, whereas the higher frequencies
are caused mainly by the counting statistics. It is clear that if we eliminate those high
frequencies, we are reducing this noise. This can be done by multiplying the Fourier
transform with a suitable function:
GðuÞ ¼ FðuÞHðuÞ ð10Þ
An example of such a function is a high-frequency cutoff filter:
1; u ucrit
HðuÞ ¼ ð11Þ
0; u > ucrit
which sets the real and imaginary coefficients above a frequency ucrit to zero. If we apply
this filter to the Fourier transform of Figure 3 using ucrit ¼ 0.05 and then apply the inverse
Fourier transformation [Eq. (9)], the result as shown by the solid line in Figure 2 is ob-
tained. The peak shape is preserved, but most of the statistical fluctuations are eliminated.
Copyright © 2002 Marcel Dekker, Inc.
Figure 3 Fourier power spectrum of the pulseheight distribution shown in Figure 2.
If we would cut off at even lower frequencies, peak distortions at the top and at the base of
the peak would become more pronounced.
This Fourier filtering can also be done directly in the original data space. Indeed, the
convolution theorem says that multiplication in the Fourier space is equivalent to con-
volution in the original space:
GðuÞ ¼ FðuÞHðuÞ , fðxÞ hðxÞ ¼ gðxÞ ð12Þ
The convolute at data point x is defined as the sum of the products of the original data and
the filter centered around point x:
X
gðxÞ ¼ fðxÞ hðxÞ ¼ fðx x0 Þhðx0 Þ ð13Þ
x0
h(x) is called a digital filter and is the inverse Fourier transformation of H(u). In
general, this convolution or filtering of a spectrum yi with some weighing function is
expressed as
1 X
j¼m
yi ¼ hj yiþj ð14Þ
N j¼m
where hj are the convolution integers and N is a suitable normalization factor. The filter
width is given by 2m þ 1. Fourier filtering with the purpose to reduce or eliminate some
(high or low) frequency components in the spectrum can thus be implemented as a con-
volution of the original data with a digital filter. This convolution also alters the variance
of the original data. Applying the concept of error propagation, one finds that the variance
of the convoluted data is given by
Copyright © 2002 Marcel Dekker, Inc.
1 X m
s2y ¼ h2 yiþj ð15Þ
i N2 j¼m i
The fact that the detector response function changes with energy (becomes broader) and,
more importantly, the presence of noise prohibits the straightforward application of this
Fourier deconvolution technique. Indeed, in the presence of noise, the measured signal
must be presented by
yðxÞ ¼ fðxÞ hðxÞ þ nðxÞ ð17Þ
and its Fourier transform
YðuÞ ¼ FðuÞHðuÞ þ NðuÞ ð18Þ
or
YðuÞ NðuÞ
¼ FðuÞ þ ð19Þ
HðuÞ HðuÞ
At high frequencies, the response, H(u), goes to zero while N(u) is still significant, so that
the noise is emphasized in the inverse transformation. This clearly shows that the noise
(counting statistics) is the ultimate limitation for any spectrum processing and analysis
method.
A clear introduction to Fourier transformations related to signal processing can be
found in the work of Massart et al. (1998). Algorithms for Fourier transformation and
related topics are given in the work of Press et al. (1998). Detailed discussions on Fourier
deconvolution can be found in many textbooks (Jansson, 1984; Brook and Wynne, 1988).
Fourier deconvolution in x-ray spectrometry based on maximum a posteriori or maximum
entropy principles is discussed by several authors (Schwalbe and Trussell, 1981; Nunez
et al., 1988; Gertner et al., 1989). Gertner implemented this method for the analysis of real
x-ray spectra and compared the results with those obtained by simple peak fitting. The
problem that the deconvolution algorithms are limited to systems exhibiting translational
invariance was overcome by a transformation of the spectrum, so that the resolution
becomes independent of the energy.
B. Smoothing
pffiffiffi
Because of the uncertainty y on each channel content yi , fictitious maxima can occur
both on the continuum and on the slope of the characteristic peaks. Removal or sup-
pression of these fluctuations can be useful during the visual inspection of spectra (e.g., for
locating small peaks on a noisy continuum) and is used in most automatic peak search and
Copyright © 2002 Marcel Dekker, Inc.
continuum estimation procedures. Although smoothing can be useful in qualitative ana-
lysis, its use is not recommended as part of any quantitative spectrum evaluation.
Smoothing, although reducing the uncertainty in the data locally, redistributes the original
channel content over the neighboring channel, thus introducing distortion in the spectrum.
Accordingly, smoothing can provide a (small) improvement in the statistical precision
obtainable with simple peak integration but is of no advantage when used with least-
squares-fitting procedures in which assumptions about the peak shapes are made.
1. Moving Average
The most straightforward way of smoothing (any) fluctuating signal is to employ the
‘‘box-car’’ or moving-average technique. Starting from a measured spectrum y, a
smoothed spectrum y* is obtained by calculating the mean channel content around each
channel i:
1 X
þm
yi ¼ yi ¼ yiþj ð20Þ
2m þ 1 j¼m
This can be seen as a convolution [Eq. (14)] with all coefficients hj ¼ 1. The smoothing
effect obviously depends on the width of the filter, 2m þ 1. The operation being a simple
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
averaging, the standard deviation of the smoothed data is reduced by a factor 2m þ 1 in
regions where yi is nearly constant. On the other hand, such a filter introduces a con-
siderable amount of peak distortion. This distortion depends on the filter width-to-peak
width ratio. Figure 4 shows the peak distortion effects when a moving-average filter of
widths 9, 17, and 25 is applied to a peak with
P full width at half-maximum (FWHM) equal
to nine channels. Being a unit area filter hj =N ¼ 1 with N ¼ 2m þ 1, the total counts in
the peak is not affected in an appreciable way other than by rounding errors.
Figure 4 Distortion introduced by smoothing of a peak with a moving-average filter 9, 17, and 25
points wide. The FWHM of the original peak is nine channels.
yðiÞ ¼ a0 þ a1 ði i0 Þ þ a2 ði i0 Þ2 ð21Þ
can be made. Once we have determined the coefficients aj , the value of the polynomial at
the central channel i0 can be used as the smoothed value:
yi ¼ yði0 Þ ¼ a0 ð22Þ
This concept is schematically illustrated in Figure 6. By moving the central channel to the
right (from i0 to i0 þ 1), the next smoothed channel content can be calculated by repeating
the entire procedure.
Figure 5 Percent change in peak height and width introduced by filtering with a moving-average
and a Savitsky and Golay polynomial filter as a function of the filter width-to-peak FWHM ratio,
(2m þ 1)/FWHM.
At first sight, this smoothing method would require a least-squares fit for each
channel in the spectrum. However, the fact that the x values are equidistant allows us to
formulate the problem in such a way that the polynomial coefficients can be expressed as
simple linear combinations involving only yi values:
1 X
j¼m
ak ¼ Ck; j yiþk ð23Þ
Nk j¼m
This means that it is possible to implement the least-squares-fitting procedure more effi-
ciently as a convolution of the spectrum with a filter having appropriate weights. For this
second-order polynomial, the coefficients are given by
C0; j 3ð3m2 þ 3m 1 5j2 Þ
¼ ð24Þ
N0 ð2m 1Þð2m þ 1Þð2m þ 3Þ
Smoothing with a five-point second-degree polynomial ð2m þ 1 ¼ 5Þ thus becomes
1
yi ¼ a0 ¼ ð3yi2 þ 12yi1 þ 17yi þ 12yiþ1 3yiþ2 Þ ð25Þ
35
In general, for a polynomial of degree r fitted to 2m þ 1 points, this can be written as
1 X
j¼m
yi ¼ Crm; j yiþj ð26Þ
Nrm j¼m
where the convolution integers Crm;j and the normalization factors Nrm do not depend on
the data to be smoothed but are a function only of the polynomial degree r and on the filter
Copyright © 2002 Marcel Dekker, Inc.
half-width m. Table 1 lists the coefficients of polynomial smoothing filters with widths
between 5 and 25 points. The coefficients for a second- and a third-degree polynomial are
identical. In comparison with the moving-average filter of the same width, polynomial
smoothing filters are less effective in removing noise but have the advantage of causing less
peak distortion. The distortion effect as a function of the filter width-to-peak width ratio is
given in Figure 5. When the filter becomes much wider than the peak, the smoothed
spectrum features oscillations near the peak boundaries, as illustrated in Figure 7.
An interesting feature of this type of filters is that they can produce not only a
smoothed spectrum but also a smoothed first and second derivative of the spectrum. If we
differentiate Eq. (21) and take the value at the center position,
0 dyðiÞ
y ¼ ¼ a1 ð27Þ
di i¼i0
d 2 yðiÞ
y00 ¼ ¼ 2a2 ð28Þ
di2 i¼i0
it follows from Eq. (23) that the smoothed first and second derivative of the spectrum can
also be calculated by means of suitable convolution coefficients. For instance, for the first
derivative of a second-order polynomial, using five data points this becomes
1
y0
i ¼ ð2yi2 yi1 þ yiþ1 þ 2yiþ2 Þ ð29Þ
10
The corresponding convolution integers for the calculation of the smoothed first and
second derivative are listed in Tables 24. The use of the derivative spectra is illustrated in
next section dealing with peak search methods. The FORTRAN implementation of the
Savitsky and Golay filters is given in Sec. X.
Variations on these smoothing strategies, such as the use of variable-width filters are
reviewed by Yule (1967). The effect of repeated smoothing on the accuracy and precision
of peak area determination is discussed by Nielson (1978). A more comprehensive treat-
ment on polynomial smoothing can be found in Enke and Nieman (1976) and its
references.
Table 1 Savitsky and Golay Coefficients for Second- and Third-Degree (r ¼ 2 and r ¼ 3)
Pj¼m
Polynomial Smoothing Filter; yi ¼ ð1=Nrm Þ j¼m Crm; j yiþj , Filter Width ¼ 2m þ 1
j ðCrm; j ¼ Crm;j Þ
m Nrm 0 1 2 3 4 5 6 7 8 9 10 11 12
2 35 17 12 3
3 21 7 6 3 2
4 231 59 54 39 14 21
5 429 89 84 69 44 9 36
6 143 25 24 21 16 9 0 11
7 1105 167 162 147 122 87 42 13 78
8 323 43 42 39 34 27 18 7 6 21
9 2261 269 264 249 224 189 144 89 24 51 136
10 3059 329 324 309 284 249 204 149 84 9 76 171
11 805 79 78 75 70 63 54 43 30 15 2 21 42
12 5175 467 462 447 422 387 342 287 222 147 62 33 138 253
j ðCrm; j ¼ Crm;j Þ
m Nrm 0 1 2 3 4 5 6 7 8 9 10 11 12
2 10 0 1 2
3 28 0 1 2 3
4 60 0 1 2 3 4
5 110 0 1 2 3 4 5
6 182 0 1 2 3 4 5 6
7 280 0 1 2 3 4 5 6 7
8 408 0 1 2 3 4 5 6 7 8
9 570 0 1 2 3 4 5 6 7 8 9
10 770 0 1 2 3 4 5 6 7 8 9 10
11 1012 0 1 2 3 4 5 6 7 8 9 10 11
12 1300 0 1 2 3 4 5 6 7 8 9 10 11 12
j ðCrm; j ¼ Crm;j Þ
m Nrm 0 1 2 3 4 5 6 7 8 9 10 11 12
2 12 0 8 1
3 252 0 58 67 22
4 1,188 0 126 193 142 86
5 5,148 0 296 503 532 294 300
6 24,024 0 832 1,489 1,796 1,578 660 1,133
7 334,152 0 7,506 13,843 17,842 18,334 14,150 4,121 12,922
8 23,256 0 358 673 902 1,002 930 643 98 748
9 255,816 0 2816 5,363 7,372 8,574 8,700 7,481 4,648 68 6,936
10 3,634,092 0 29,592 56,881 79,564 95,338 101,900 96,947 78,176 43,284 10,032 84,075
11 197,340 0 1,222 2,365 3,350 4,098 4,530 4,567 4,130 3,140 1,518 815 3,938
12 1,776,060 0 8,558 16,649 23,806 29,562 33,450 35,003 33,754 29,236 20,982 8,525 8,602 30,866
j ðCrm; j ¼ Crm;j Þ
m Nrm 0 1 2 3 4 5 6 7 8 9 10 11 12
2 7 2 1 2
3 42 4 3 0 5
4 462 20 17 8 7 28
5 429 10 9 6 1 6 15
6 1,001 14 13 10 5 2 11 22
7 6,188 56 53 44 29 8 19 52 91
8 3,876 24 23 20 15 8 1 12 25 40
9 6,783 30 29 26 21 14 5 6 19 34 51
10 33,649 110 107 98 83 62 35 2 37 82 133 190
11 17,710 44 43 40 35 28 19 8 5 20 37 56 77
12 26,910 52 51 48 43 36 27 16 3 12 29 48 69 92
The rather heuristic algorithm is discussed in Section X, together with the computer im-
plementation. The effect of the filter is illustrated in Figure 8, where it is compared with the
other smoothing methods discussed.
constant and positive coefficients and two side lobes with constant and negative coefficients.
Convoluting an x-ray spectrum with this kind of filter yields spectra in which the continuum
is removed and peaks are easily locatable. They are similar to inverted second-derivative
spectra. An important representative of this group of filters is the ‘‘top-hat’’ filter, which has
a central window with an odd number of channels w and two side windows each n channels
wide. The value of the filter coefficients follows from the zero-area constraint:
8 1
< 2n ; n w2 k < w2
hk ¼ 1
; w2 k þ w2 ð30Þ
: w1
2n ; þ w2 < k w2 þ n
The filtered spectrum is obtained by the convolution of the original spectrum with this filter:
X
k¼nþw=2
yi ¼ hk yiþk ð31Þ
k¼nw=2
The effect of this filter on a typical spectrum is shown in Figure 10. The variance of the
filtered spectrum is obtained by simple error propagation:
X
k¼nþw=2
s2y ¼ h2k yiþk ð32Þ
i
k¼nw=2
If yi is significantly different from zero, a peak structure is found and the top of the peak
can approximately be located by searching for the maximum. Thus, i corresponds to the
position of a peak maximum in the original spectrum if
yi > rsyi ð33Þ
and
yi1 yi > yiþ1 ð34Þ
In Figure 11, the positive part of the filtered spectrum (w ¼ 9 and n ¼ 5) and the decision
line ðrsy Þ for r ¼ 1 and 4 are displayed.
If required, other peak features can be obtained from the filtered spectrum: The
distance between the two local minima is a measure of the width of the peak and the height
at the maximum is related to the net peak area.
Because the width and heights of the peaks in the filtered spectrum strongly depend
on the dimensions of the filter, it is important that its dimensions are matched to the width
of the peaks in the original spectrum. From considerations of peak detectability (signal-to-
noise ratio) and resolution (peak broadening), it follows that the optimum width of the
positive window w is equal to the FWHM of the peaks (Robertson et al., 1972). The width
of the negative side windows should be chosen as large as the curvature of the continuum
allows. A reasonable compromise between sensitivity to peak shapes and rejection of
continuum is reached when n equals FWHM=2 to FWHM=3. Typical values for the
sensitivity factor r are between 2 and 4. Higher values result in the loss of small peaks;
lower values will cause continuum fluctuations to be interpreted as peaks.
Other zero-area rectangular filters, variations to the top-hat filter, are also in use,
such as the square-wave filter with typical coefficient sequence 1, 1, 2, 2, 1, 1 (Philips
and Marlow 1976; McCullagh, 1982) and the symmetric square-wave filter with coeffi-
cients 1, 1, 1, 1 (Op De Beeck and Hoste 1975). A detailed account of the performance
of this filter is given by Op De Beeck and Hoste (1975). A method using a Gaussian
correlator function is discussed by Black (1969).
Copyright © 2002 Marcel Dekker, Inc.
Figure 11 Peak search using the positive part of the top-hat filtered spectrum and the decision
level for one and four times the standard deviation. The original spectrum is shown at the bottom.
Once the peak top is approximately located in the filtered spectrum, a more precise
maximum can be found by fitting a parabola over a few channels around the peak. For a
well-defined peak on a low continuum (or after continuum subtraction), the channel
content near the top of the peak can be approximated by a Gaussian:
" #
ðx mÞ2
yi h exp ð35Þ
2s2
a1
m¼ ð37Þ
2a2
1 yi1 yiþ1
m¼iþ ð40Þ
2 yi1 þ yiþ1 2yi
This method might be preferred for small peaks when the continuum cannot be
disregarded.
A FORTRAN implementation of a peak search algorithm is given in Section X.
Except for some special quantification procedures (e.g., the peak-to-background method
in electron microscopy), the relevant analytical information is found in the net peak areas
and continuum is considered a nuisance. There are, in principle, three ways to deal with
the continuum: (1) the continuum can be suppressed or eliminated by a suitable filter;
(2) the continuum can be estimated and subtracted from the spectrum prior to the esti-
mation of the net peak areas; and (3) the continuum can be estimated simultaneously with
the other features in the spectrum. The first approach is discussed in Section VI, where the
continuum is removed from spectra by applying a top-hat filter followed by linear least-
squares fit of the spectrum with a number of (also filtered) reference spectra. Least-squares
fit (linear or nonlinear) with analytical functions (Sec. VII) allows the simultaneous esti-
mation of continuum and peaks, providing a suitable mathematical function can be found
for the continuum. In this section, we discuss a number of procedures that aim to estimate
the continuum independently of the other features in the spectrum. Once estimated, this
continuum can be subtracted from the original spectrum and all methods for further
processing, ranging from simple peak integration to least-squares fitting can be applied.
Any continuum estimation procedure must fulfill two important requirements. First,
the method must be able to reliably estimate the continuum in all kinds of situations (e.g.,
small isolated peaks on a high continuum as well as in the proximity of a matrix line).
Second, to permit processing of a large number of spectra, the method needs to be nearly
free of user-adjustable parameters.
Although a number of useful continuum estimation procedures has been developed,
it must be realized that their accuracy in estimating the continuum is not optimal. In one
way or another, they rely on the difference in frequency response of the continuum
compared to other structures such as peaks, the former mainly consisting of low
Copyright © 2002 Marcel Dekker, Inc.
frequencies (slowly varying). Because the peaks also exhibit low frequencies at the peak
boundaries, it is difficult to control the method in such a way that it correctly discriminates
between peaks and continuum. This results in either a small underestimation or over-
estimation of the continuum, introducing potentially large relative errors for small peaks.
In this respect, the fitting of the continuum with analytical functions may provide more
optimal results (Vekemans et al., 1994). A considerable advantage of the methods dis-
cussed here is that they do not assume an explicit mathematical model of the continuum.
Constructing a detailed and accurate analytical model for the continuum based in physical
theory is nearly impossible except for some simple geometry and particular excitation
conditions. Most often, some polynomial type of function must be chosen when fitting a
portion of the spectrum with analytical functions.
A. Peak Stripping
These methods are based on the removal of rapidly varying structures in a spectrum by
comparing the channel content yi with the channel content of its neighbors. Clayton et al.
(1987) proposed a method which compares the content of channel i with the mean of its
two direct neighbors:
yi1 þ yiþ1
mi ¼ ð41Þ
2
If yi is smaller than mi, the content of channel i is replaced by the mean mi. If this trans-
formation is executed once of all channels, one can observe a slight reduction in the height
of the peaks while the rest of the spectrum remains virtually unchanged. By repeating this
procedure, the peaks are gradually ‘‘stripped’’ from the spectrum. Because the method
tends to connect local minima, it is very sensitive to local fluctuations in the continuum due
to counting statistics. This makes smoothing of the spectrum, as discussed in the previous
section, prior to the stripping process mandatory. Depending on the width of the peaks
after typically 1000 cycles, the stripping converges and a more or less smooth continuum
remains. To reduce the number of iterations, it might be advantageous to perform a log or
pffiffiffiffi
square root transformation to the data prior to the stripping: y0i ¼ logðyi þ 1Þ or y0i ¼ yi .
After the stripping, the continuum shape is obtained by applying the inverse transforma-
tion. A major disadvantage of this method is that after a number of cycles, the bases of
partially overlapping peaks are transformed into broad ‘‘humps,’’ which take much longer
to remove than isolated peaks. The method was originally applied to PIXE spectra but
proves to be generally applicable for pulseheight spectra.
In Figure 12, this method is applied to estimate the continuum of an x-ray spectrum
in the region between 1.6 and 13.0 keV. The spectrum results from a 200-mg=cm2 pellet of
NIST SRM Bovine Liver sample excited with the white spectrum of an Rh-anode x-ray
tube filtered through a thin Rh filter (Tracor Spectrace 5000). Because of the white tube
spectrum, a considerable continuum intensity was observed, increasing quite steeply in the
region above 10 keV. To calculate the continuum, the following algorithm was used:
(1) the square root of the original spectrum was taken; (2) these data were smoothed with a
10-point Savitsky and Golay filter; (3) a number of iterations were performed applying
Eq. (41) over the region of interest; (4) the square of each data point was taken (back
transformation) to obtain the final continuum shape. In Figure 12, the continuum after
10, 100, and, finally, 500 iterations is shown.
As a generalization of the above-discussed method, the average of two channels a
distance w away from i can be used:
Copyright © 2002 Marcel Dekker, Inc.
Figure 12 Continuum estimate after 10, 100, and 500 iterations obtained with simple iterative
peak stripping.
yiw þ yiþw
mi ¼ ð42Þ
2
Ryan et al. (1988) proposed using twice the FWHM of the spectrometer at channel i as the
value for w. They reported that only 24 passes were required to produce acceptable
continuum shapes
pffiffiffi in PIXE spectra. During the last eight cycles, w is progressively reduced
by the factor 2 to obtain a smooth continuum. To compress the dynamic range of the
spectrum, a double-log transformation of the spectrum, log[log( yi+1)+1], before the
actual stripping was proposed. In combination with the low statistics digital filter, this
procedure is called the SNIP algorithm (Statistical Nonlinear Iterative Peak clipping).
A variant of this procedure is implemented in the procedure ‘‘SNIPBG’’ given in Sec. X.
Instead of the double logarithmic, we employed a square root transformation, and a Sa-
vitsky and Golay smoothing is performed on the square root data. The width w is kept
constant over the entire spectrum. The value of w is also used as the width of the smoothing
filter. Using this implementation, the continuum of the above-discussed spectrum is cal-
culated and represented in Figure 13. The width was set 11 to channels approximately
corresponding to the FWHM of the peaks in the center of the spectrum and 24 iterations
were done. Apart from delivering a smoother continuum with smaller ‘‘humps,’’ this
method executes much faster than the original method proposed by Clayton.
In both in WDXRF and EDXRF, the concentration of the analyte is proportional to the
number of counts under the characteristic x-ray peak corrected for the continuum. At
constant resolution, this proportionality also exits for the net peak height. In EDXRF,
preference is given to the peak area. In WDXRF, the acquisition of the entire peak profile
is very time-consuming and the count rate is usually measured only at the peak maximum.
Table 5 Data on the Fit of the Continuum of the Spectrum Shown in Figure 14 Using Orthogonal
Polynomials
4 126 7 (3.98±0.20)106
5 199 16 (1.88±0.09)107
6 267 6 (7.66±0.28)1010
7 266 6 (3.66±2.03)1013
YBL and YBR are the values of the continuum at the channels iBL and iBR , left and right
from the peaks, respectively. These values are best estimated by averaging over a number
of channels:
1 X iBL1
NBL
YBL ¼ yi ¼ ð52Þ
nBL i¼iBL2 nBL
1 X
iBR1
NBR
YBR ¼ yi ¼ ð53Þ
nBR i¼iBR2
nBR
The number of channels in the continuum windows are nBL ¼ iBL2 iBL1 þ 1 and
nBR ¼ iBR2 iBR1 þ 1. The center position of the continuum windows (not necessarily an
integer number!) used in Eq. (51) are iBL ¼ ðiBL1 þ iBL2 Þ=2 and iBR ¼ ðiBR1 þ iBR2 Þ=2.
If both continuum windows have equal width, nBL ¼ nBR ¼ nB =2, and are positioned
symmetrically with respect to the peak window ðiP iBL ¼ iBR iP Þ a much simpler ex-
pression is obtained for the net peak area:
nP
NP ¼ NT ðNBL þ NBR Þ ð54Þ
nB
where NBL and NBR are the total counts in the left and right continuum windows, re-
spectively, each nB =2 channels wide, and nP equals the number of channels in the peak
window. Applying the principle of error propagation, the uncertainty in the net peak area
is then given by
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 ffi
nP
sNP ¼ N T þ ðNBL þ NBR Þ ð55Þ
nB
Various counting strategies can be considered and the effect on the precision can be esti-
mated using Eq. (57) (Bertin, 1970). In a ‘‘optimum fixed time’’ strategy, the minimum
Copyright © 2002 Marcel Dekker, Inc.
uncertainty is obtained when, for a total measurement time t ¼ tP þ tB ; tP and tB are chosen
in such a way that their ratio is equal to the square root of the peak-to-continuum ratio:
rffiffiffiffiffi
tP IP
¼ ð58Þ
tB IB
Under, these conditions, the uncertainty in the net intensity is given by
pffiffiffiffiffi pffiffiffiffiffi
IP þ IB
sI ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð59Þ
tP þ tB
In this section, two techniques based on linear least squares are discussed. The filter-fit
methods makes use of library spetra, measured or calculated spectra of pure compounds,
that are used to describe the spectra of complex samples. The other method is based on
partial least-squares (PLS) regression, a multivariate calibration technique. In this case, no
spectrum evaluation in the strict sense is performed, but, rather, relations between the
concentrations of the compounds in the samples and the entire spectrum are established.
In this way, quantitative analysis is possible without obtaining net peak areas of the
characteristic lines.
A. Filter-Fit Method
1. Theory
If a measured spectrum of an unknown sample can be described as a linear combination of
spectra of pure elements constituting the sample, then the following mathematical model
can be written:
X
m
ymod
i ¼ aj xji ð60Þ
j¼1
with ymod
i the content of channel i in the model spectrum and xji the content of channel i in
the jth reference spectrum. The coefficients aj are a measure of the contribution of pure
reference spectra to the unknown spectrum and can be used for quantitative analysis. The
values of the coefficients aj are obtained via multiple linear least-squares fitting, mini-
mizing the sum of the weighted squared differences between the measured spectrum and
the model:
" #2
Xn2
1 2
Xn2
1 Xm
w ¼2
½ yi yðiÞ ¼ yi aj xji ð61Þ
s2
i¼n1 i
s2
i¼n1 i j¼1
where yi and si are the channel content and the uncertainty of the measured spectrum,
respectively, and n1 and n2 are the limits of the fitting region. A detailed discussion of the
least-squares-fitting method is given in Sec. IX.
The assumption of linear additivity [Eq. (60)] holds normally reasonable well for the
characteristic lines in the spectrum, but not for the continuum. To apply this technique,
the continuum can be removed from the unknown spectrum and from the reference
spectra using one of the procedures described in Sec. IV before the actual least-squares fit.
Copyright © 2002 Marcel Dekker, Inc.
Another, frequently used approach is to apply a digital filter to both unknown and
reference spectra. This variant is known as the filter-fit method (Schamber, 1977; Statham,
1978; McCarthy and Schamber, 1981) and is discussed in some detail below.
By the discrete convolution of a spectrum with top-hat filter [Eqs. (30) and (31)], the
low-frequency component (i.e. the slowly varying continuum) is effectively suppressed as
discussed in Sec. III. Apart from removing the slowly varying continuum, a rather severe
distortion of the peaks is also introduced. If we apply this filter to both the unknown
spectrum and the reference spectra, the nonadditive continuum is removed and the same
type of peak distortion will be introduced in all spectra, allowing us to apply the method of
multiple linear least-squares fitting to the filtered spectra. Equation (61) then becomes
" #2
Xn2
1 X
m
w2 ¼ y0 aj x0ji ð62Þ
s0 2 i j¼1
i¼n1 i
Where y0i and x0ji are the filtered unknown and reference spectra, respectively, s0i 2 is the
variance of y0i given by
X
s02
i ¼ h2k yiþk ð63Þ
k
The least-squares estimates of the contribution of each reference spectrum is then given by
(see Sec. IX)
X
m
aj ¼ a1
jk bk ; j ¼ 1; . . . ; m ð64Þ
k¼1
with
Xn2
1 0 0
bj ¼ yx
02 i ji
ð65Þ
s
i¼n1 i
Xn2
1 0 0
ajk ¼ x x ð66Þ
s0 2 ki ji
i¼n1 i
The uncertainty in each coefficient aj is directly estimated from the error matrix:
s2aj ¼ a1
jj ð67Þ
Schamber (1977) suggested the following equation for the uncertainties, taking into ac-
count the effect of the filter:
nw 1
s2aj ¼ a ð68Þ
n þ w jj
Where w is the width of the central positive part of the filter and n is the width of the
negative wings. A measure of the goodness of fit is available through the reduced w2 value:
1
w2n ¼ w2 ð69Þ
ðn2 n1 þ 1Þ m
which is the w2 value of Eq. (62) divided by the number of points in the fit minus the
number of reference spectra. A value close to 1 means a good fit, indicating that the
reference spectra are capable of adequately describing the unknown spectrum.
Copyright © 2002 Marcel Dekker, Inc.
Most of the merits and the disadvantages of the filter-fit method can be deduced
directly from the mathematical derivation given in the preceding paragraphs. The most
interesting aspect of the filter-fit method is that it does not require any mathematical
model for the continuum and that, at least in principle, the shape of the peaks in the
unknown spectrum are exactly represented by the reference spectra. Reference spectra
should be acquired with good counting statistics, at least better than the unknown spec-
trum, because the least-squares method assumes that there is no error in the independent
variables x0ji . Reference spectra can be obtained from single-element standards. Only the
portion of the spectrum that contains peaks needs to be retained as reference in the fit.
Multielement standards can be used if the peaks of each element are well separated.
The reference spectra must provide an accurate model of the peak structure present
in the unknown spectrum. This requires that reference and unknown spectra are acquired
under strictly identical spectrometer conditions. Changes in resolution and, especially,
energy shifts can cause large systematic errors. The magnitude of this error depends on the
degree of peak overlap. Peak shifts of more than a few electron volts should be avoided,
which is readily possible with modern detector electronics. If shifts are observed over long
periods of operations of the spectrometer, careful recalibration of the spectrometer is
required or, better, the reference spectra should be acquired again. Also, peak shift and
peak broadening due to differences in count rate between standards and unknown must be
avoided. Differential absorption is another problem that might influence the accuracy of
the model. Because of the difference in x-ray attenuation in the reference and the un-
known, the Kb to Ka ratios might be different in the two spectra. This becomes especially
problematic if the Kb line is above and the Ka line below an absorption edge of a major
element of the unknown sample. The magnitude of the error depends on the peak overlap.
Careful selection of the samples to produce the reference spectra is therefore required.
The procedure requires that a reference spectrum be included for each element
present in the unknown. The method allows no mechanism to deal with sum peaks. Apart
from removing the continuum, the filter also has some smoothing effect on the spectrum
and causes the peak structure to be spread out over more channels. This is equivalent to
fitting a spectrum with a somewhat lower resolution than originally acquired. Therefore,
the precision and detection limits attainable with the filter-fit method are slightly worse
than optimal. The width of the filter is important in this respect. Schamber (1977) suggests
taking the width of the top of the filter equal to the FWHM resolution of the spectrometer,
u ¼ FWHM. The width of the wings can be taken as v ¼ u=2.
2. Application
The calculation procedure is quite simple and requires the following steps. The top-hat filter
is applied to the unknown spectrum and the m reference spectra [Eqs. (30) and (31)], and the
modified uncertainties are calculated using Eq. (63). Next, the vector b of length m and the
m m square matrix a are formed using Eqs. (65) and (66), summing over the part of the
spectrum one wants to analyze (n1 n2 ). Only the relevant part, such as the Ka or the Ka
plus the Kb, of the reference spectra needs to be retained; the rest of the filtered spectrum
can be set to zero. After calculating the inverse matrix a1 , the contribution of each re-
ference to the unknown and its uncertainty are calculated using Eqs. (64) and (67) or (68).
A computer implementation of the filter-fit method is given in Sec. X. The method
was used to analyze part of an x-ray spectrum of a polished NIST SRM 1103 brass sample
(Fig. 16A). The measurements were carried out using a Mo x-ray tube and a Zr secondary
target and filter assembly. Spectra of pure metals (Fe, Ni, Cu, Zn, and Pb) were acquired
Copyright © 2002 Marcel Dekker, Inc.
Figure 16 (A) Part of the x-ray spectrum of NIST SRM 1103 brass sample. (B) Top-hat-filtered
spectrum and result of fit using reference spectra.
under identical experimental conditions. A top-hat filter of width w ¼ 5 was used. Table 6
shows how the spectra were divided in regions of interest to produce the reference spectra.
Because considerable x-ray attenuation is present in brass, separate references were created
for the Ka and Kb of Cu and Zn. This was not done for Fe, Ni, and Pb because these
elements are only present as minor constituents in the brass sample. Figure 16B shows the
filtered brass spectrum and the resulting fit using the seven (filtered) reference spectra. The
region below CuKa is expanded 100 times, and the region above ZnKa is expanded 10
times. As can be seen, the agreement between the filtered brass spectrum and the fit is very
good. The reduced w2 value is 8.5. This high value is probably due to small peaks shifts in
the reference spectra compared to brass spectrum.
Table 7 compares the results of the filter fit with the results obtained by nonlinear
least-squares fitting using analytical functions (see Sec. VII). Although the w2 value of the
nonlinear fit is slightly better (2.7), one observes an excellent agreement between the two
methods for the analytical important data (i.e., intensity ratios). The uncertainties for
small peaks are slightly higher with filter-fit method, as explained previously.
The filter-fit method is fast and relatively easy to implement. It can produce reliable
results when the spectrometer calibration can be kept constant within a few electronvolts
and suitable standards for each element present in the sample are available. The method
Copyright © 2002 Marcel Dekker, Inc.
Table 6 Data on the Reference Spectra and the Unknown Spectrum Used
in the Filter-Fit Procedure (Fig. 16)
Fe 0.0053±0.0002 0.0047±0.0001 11
Ni 0.0019±0.0002 0.00201±0.00008 6
Cu 0.546±0.001 0.551±0.001 0.9
Zn 0.390±0.001 0.3912±0.0008 0.31
Pb 0.0310±0.0006 0.0308±0.0003 0.65
performs well when one has to deal with a difficult to model continuum. If information on
trace elements and major elements is required (very large peaks next to very small ones),
the method might not be optimal. This filter-fit method is frequently used to process x-ray
spectra obtained with electron microscopes (SEM-EDX), often in combination with a
ZAF or Phi-Rho-Z correction procedure.
TnA and UnA are the score matrices, the values of the AðA pÞ latent variables. P0Ap
and Q0Am are the loading matrices describing the relation between the latent variables
(T and U) and the original variables (X and Y). The number of latent variables A in the
model is of crucial importance and its optimum value must be found by cross-validation.
The matrices E and F contain the residuals, part of the original spectral data and the
concentration data, respectively, not accounted for when using A latent variables. The
inner relation is written as:
ua ¼ ba ta ; a ¼ 1; . . . ; A ð74Þ
from which the regression coefficients B can be obtained. This operation can be seen as a least-
squares fit between the X block and Y block scores. The final PLS model can thus be written as
Copyright © 2002 Marcel Dekker, Inc.
Y ¼ TBQ0 þ F ð75Þ
A graphical representation of the PLS model is given in Figure 17. In the normal PLS
algorithm Y is a vector of concentrations for one element and a separate model is build
for each element. If all Y-variables are predicted simultaneously, as in the case of
Equation (70), the PLS2 algorithm is used. This method performs better when the con-
centrations in the samples are highly correlated.
The quality of the calibration model can be judged by the root mean square error of
prediction (RMSEP):
" #1=2
1X n
2
RMSEP ¼ ð^
yi yi Þ ð76Þ
n i¼1
where y^i is the concentration in sample i predicted by the model yi is the true concentration
in the standard. To determine the optimum number of latent variables, the RMSEP is
calculated using PLS models with different numbers of latent variables A. The RMSEP
values are plotted against A and the value where a minimum or a plateau is reached is
taken. For small datasets, the calculation of the RMSEP is done using cross-validation.
When a large number of standards are available, they are split in a training set (ap-
proximately two-thirds of the samples) and a prediction set (approximately one-third of
the samples). The training set is used to build the PLS model and the RMSEP is calculated
based on the concentrations of the prediction set. Alternatively, for smaller calibration
sets, leave-one-out cross-validation can be used. Each sample is excluded once from the
dataset and predicted by the model built with the remaining samples. This is repeated until
all samples have been excluded once.
Geladi and Kowalski (1986) published a tutorial on PLS and its relation to other
regression techniques. A standard work on the PLS method is the book by Martens and
2. Application
The PLS method is illustrated with the analysis of aqueous solution containing Ni, Cu,
and As with concentration in the ranges 1850, 0.545, and 520 g=L, respectively,
whereas Zn, Fe, and Sb were present in more or less constant amounts. Spectra were
acquired for 1000 s from 5-mL solution using an EDXRF spectrometer equipped with a
Rh x-ray tube operating at 30 kV and a Si(Li) detector having 160 eV resolution at MnKa.
From the 22 samples, 10 were used to build the PLS mode and the remaining were
used as the prediction set. Table 8 gives the composition of the samples in the calibration
set. Figure 18 shows part of the spectrum between 5 and 14 keV of sample number 9. The
CuKa line is considerably interfered by the NiKb line. Absorption effects can be expected
because the NiK absorption edge (8.33 keV) is just above the CuKa line (8.04 keV) and the
AsK lines can cause secondary fluorescence of Cu and Ni.
A PLS 1 model is build for Cu. The X matrix consists of 451 x-ray intensities
(variables) between 5 and 14 keV (channels 250 to 700) of the 10 samples. The Y matrix
contains the Cu concentration of those samples.
In Figure 19, the RMSEP based on leave-one-out cross-validation is plotted as
function of the number of latent variables. A minimum error of 0.57 g=L is obtained for
three latent variables so that the PLS model with three latent variable is retained.
Figure 20 compares the true Cu concentration with the predicted concentration for the
calibration set. The PLS model predicts very accurately the Cu concentrations in the
range from 1 to 40 g=L. The so-called regression coefficients of the PLS model show
which variables (channels) are used in the model. They are plotted in Figure 21. As could
be expected, the Cu concentrations are predicted from the channel content corresponding
to the CuK lines. The influence of absorption and enhancement results in a small negative
contribution from the Ni peak and a small positive contribution from the As peaks, re-
spectively. The PLS model thus handles both the problem of spectral interference and
matrix effects. The Cu concentration predicted by the PLS model for the test set is given in
Table 9. Except for the two samples with the highest concentrations, the Cu concentration
Concentration (g=L)
Sample No. Ni Cu Zn
Figure 19 RMSEP values versus the number of latent variables for the prediction of the Cu
concentration.
in the ‘‘unknown’’ samples is very well estimated. The predicted of the test set is generally
somewhat worse that the prediction of the calibration set. To build an accurate model, a
large number of standards spanning the concentration range of interest for each element is
required. This is certainly the major drawback of PLS for its application in XRF,
Copyright © 2002 Marcel Dekker, Inc.
Figure 20 Comparison of ‘‘true’’ and predicted Cu concentrations for the samples in the PLS
calibration set.
Figure 21 Regression coefficients of the PLS model for Cu, showing which variables (channels in
the x-ray spectrum) are used to predict Cu.
Cu concentration (g=L)
especially when solid samples are considered. This problem can to some extend be
overcome by building a calibration set via Monte Carlo simulation. Just as with the filter-
fit method, standards and unknowns need to be measured under strictly identical spec-
trometer conditions. Changes in gain or resolution will cause systematic errors in the
calculated concentrations.
Swerts and Van Espen (1993) demonstrated the use of PLS for the determination of
S in graphite using a Rh x-ray tube excitation XRF equipped with a Si(Li)detector. Be-
cause of diffraction effects, least-squares fitting of the spectra was nearly impossible. Using
PLS, the sulfur content could be determined in a concentration range 260%, with an
accuracy of better than 5% relative standard deviation. Urbánski and Kowalska applied
the PLS method to the determination of Sr and Fe in powdered rock samples and to the
determination of the S and ash content in coal using a radioisotope XRF system equipped
with a low-resolution gas proportional counter. They also demonstrated the usefulness of
this method for the determination of the thickness and composition of Sn-Pb and Ni-Fe
layered structures (Urbánski and Kowalska, 1995). Molt and Schramm (1987) compared
principal components regression (PCR), PLS for the determination of S, exhibiting strong
interference from Mo, in aqueous and solid samples. The results were also compared with
quantitative analysis using the method developed by Lucas-Tooth and Price (1961).
Equally good results were obtained with all three methods. Similar results were obtained
by Lemberge and Van Espen for the determination of Ni, Cu, and As in liquid samples.
They demonstrated that taking the square root of the data improves the PLS model and
that the PLS method extracts information from the scattered excitation radiation to de-
scribe the matrix effects (Lemberge and Van Espen, 1999).
A widely used and certainly the most flexible procedure for evaluating complex x-ray
spectra is based on least-squares fitting of the spectral data with an analytical function.
The method is conceptually simple, but not trivial to implement and use.
Copyright © 2002 Marcel Dekker, Inc.
A. Concept
In this method, an algebraic function, including analytical importantly parameters such as
the net areas of the fluorescence lines, is used as a model for the measured spectrum. The
object function (w2) is defined as the weighted sum of squares of the differences between
this model y(i) and the measured spectrum yi over a region of the spectrum:
Xn2
1
w2 ¼ 2
½ yi yði; a1 ; . . . ; am Þ2 ð77Þ
i¼n1
si
where s2i is the variance of data point i, normally taken as s2i ¼ yi , and aj are the para-
meters of the model. The optimum values of the parameters are those for which w2 is
minimal. They can be found by setting the partial derivatives of w2 with respect to the
parameters to zero:
@w2
¼ 0; j ¼ 1; . . . ; m ð78Þ
@aj
If the model is linear in all the parameters aj, these equations result in set of m linear
equations in the m unknowns aj, which can be solved algebraically. This is known as linear
least-squares fitting. If the model is nonlinear in one or more of its parameters, a direct
solution is not possible and the optimum value of the parameters must be found itera-
tively. An initial value is given to the parameters and they are varied in some way until a
minimum for w2 is obtained. The latter is equivalent to searching for a minimum in the
m þ 1-dimensional w2 hypersurface. This is known as nonlinear least-squares fitting. The
selection of a suitable minimization algorithm is very important because it determines to a
large extent the performance of the method. A detailed discussion of linear and nonlinear
least-squares fitting is given in Sec. IX.
The most difficult problem to solve when applying this least-squares procedure is the
construction an analytical function that accurately describes the observed spectrum. The
model must be capable of describing accurately the spectral data in the fitting region. This
requires an appropriate model for the continuum, the characteristic lines of the elements
and for all other features present in the spectrum such as absorption edges, escape, and
sum peaks. Although the response function of the energy-dispersive detector is, to a very
good approximation, Gaussian, deviation from the Gaussian shape needs to be taken into
account. Failure to construct an accurate model will result in systematic errors, which
under certain conditions may lead to gross positive or negative errors in the estimated
peak areas. On the other hand, the fitting function should remain simple, with as few
parameters as possible. Especially for the nonlinear fitting, the location of the w2 minimum
becomes problematic when a large number of parameters is involved.
In general, the fitting model consists of two parts:
X
yðiÞ ¼ yB ðiÞ þ yP ðiÞ ð79Þ
P
where y(i) is calculated content of channel i and the first part describes the continuum and
the second part the contributions of all peaklike structures.
Because the fitting functions for both linear and nonlinear least-squares fitting have
many features in common, we treat the detailed description of the fitting function for the
most general case of nonlinear least squares. Moreover, if the least-squares fit is done
using the Marquardt algorithm, the linear least-squares fit is computationally a particular
case of the nonlinear least-squares fit. Programs based on this algorithm can perform
Copyright © 2002 Marcel Dekker, Inc.
linear and nonlinear fitting using the same computer code. A large part of the discussion
given here is based on the computer code AXIL, developed by the author for spectrum
fitting of photon-, electron-, and particle-induced x-ray spectra (Van Espen et al., 1977a,
1977b, 1979b, 1986).
1. Linear Polynomial
A linear polynomial of the type
2. Exponential Polynomial
A linear polynomial cannot be used to fit the continuum over the entire spectrum or to fit
regions of high positive or negative curvature. Higher curvature can be modeled by
functions of the type
3. Bremsstrahlung Continuum
The exponential polynomial is not suitable for describing the shape of the continuum
observed in electron- and particle-induced x-ray spectra, mainly because of the high
Copyright © 2002 Marcel Dekker, Inc.
curvature at the low-energy region of the spectrum. The continuum results from photons
emitted from the sample by the retardation of fast electrons. The slope of the emitted
continuum is essentially an exponentially decreasing function according to Kramer’s for-
mula. At low energies, the emitted photons are strongly absorbed by the detector windows
and by the sample. A suitable function to describe such a radiative continuum is an
exponential polynominal multiplied by the absorption characteristics of the spectrometer:
4. Continuum Removal
An alternative to an algebraic description of the continuum is to estimate the continuum
first using one of the procedures outlined in Sec. IV and to substract this continuum from
the measured spectrum before the actual least-squares fitting. To correctly implement the
least-squares fit after subtraction of the continuum, the weights 1=s2i [Eq. (77)] must be
adjusted. If y0i represents the spectral data after subtraction of the continuum,
y0i ¼ yi yB ðiÞ, the variance of y0i is given by s0i 2 ¼ s2i þ s2yB ðiÞ . A reasonable approximation
for s2yB ðiÞ is yB ðiÞ itself, so that the total variance becomes s0i 2 ¼ yi þ yB ðiÞ. If this adjust-
ment of the weights is not made, the uncertainty in the net peak areas are understimated,
especially for small peaks on a high continuum.
It is rather difficult for an inexperienced user to select the appropriate continuum
model for a given spectrum. The following might serve as a general guideline. For fitting
regions 23 keV wide, a linear polynomial continuum is often adequate. To fit large re-
gions of XRF spectra, the exponential polynomial provides the most accurate results, with
k typically, between 4 and 6. The same holds for the bremsstrahlung continuum of SEM-
EDX and PIXE spectra. The simplest method from the users’ point of view is the
continuum stripping, but this method does not provide optimum results. A slight under-
estimation or overestimation might occur, resulting in large relative errors in the area
determination of small peaks (Vekemans et al., 1995).
1. Single Gaussian
A Gaussian peak is characterized by three parameters: the position, width, and height or
area. It is desirable to describe the peak in terms of its area rather than its height because
the area is directly related to the number of x-ray photons detected, whereas the height
depends on the spectrometer resolution. The first approximation to the profile of a single
peak is then given by
Copyright © 2002 Marcel Dekker, Inc.
" #
A ðxi mÞ2
pffiffiffiffiffiffi exp ð83Þ
s 2p 2s2
where A is the peak area (counts), s is the width of the Gaussian expressed in channels,
and m ispthe location
ffiffiffiffiffiffiffiffiffiffi
ffi of the peak maximum. The often used FWHM is related to s by the
factor 2 2 ln 2 or FWHM ¼ 2.35s. In Eq. (83), the peak area is a linear parameter; the
width and position are nonlinear parameters. This implies that a nonlinear least-squares
procedure is required to find optimum values for the latter two parameters. Using a linear
least-squares method assumes that the position and width of the peak are know with high
accuracy from calibration.
To describe part of a measured spectrum, the fitting function must contain a number
of such functions, one for each peak. For 10 elements and 2 peaks (Ka and Kb) per element,
we would need to optimize 60 parameters. It is highly unlikely that such a nonlinear least-
squares fit will terminate successfully at the global minimum. To overcome this problem,
the fitting function can be written in a different way as shown in the next subsection.
with Ej the energy (in eV) of the x-ray line and s the peak width given by
NOISE 2
s ¼
2
þ3:58ðFANOÞEj ð86Þ
2:3548
In this equation, NOISE is the electronic contribution to the peak width (typically
80100 eV FWHM) with the factor 2.3548 to convert to s units, FANO is the Fano factor
( 0.114), and 3.85
ffi is the energy required to produce an electronhole pair in silicon. The
pffiffiffiffiffi
term GAIN=s 2p in Eq. (85) is required to normalize the Gaussian so that the sum over all
channels is unity.
For linear least-squares fitting, ZERO, GAIN, NOISE, and FANO are physically
meaningful constants. In the case of nonlinear least squares, they are parameters to be
refined during the fitting. The advantage of optimizing the energy and resolution calibra-
tion parameters rather than the position and width of each peak is a vast reduction of the
dimensionality of the problem. The nonlinear fit of 10 peaks would now involve 14 para-
meters compared to 30. Even more importantly, all information available in the spectrum is
used to estimate ZERO, GAIN, NOISE, and FANO and thus the positions and the widths
Copyright © 2002 Marcel Dekker, Inc.
of all peaks. Imagine a small, severely overlapping doublet with a well-defined peak on both
sides of this doublet. These two peaks will contribute most to the determination of the four
calibration parameters, virtually fixing the position and the width of the two peaks in the
doublet. As a consequence, their areas can be determined much more accurately.
Referring to our discussion on information content in Sec. II, we did not obtain this
extra performance for nothing. We have supplied extra information: the energy of the
peaks and the two calibration relations [Eqs. (84) and (86)]. Fitting with Eq. (85) requires
that the extra information we supply is indeed correct.
With modern electronics, the linearity of the energy calibration [Eq. (84)] holds very
well in regions above 2 keV. Fitting the entire spectrum, including the low-energy region,
might require more complex energy calibration functions. To fit PIXE spectra from 1 to
30 keV, Maenhaut and Vandenhaut (Maenhaut and Vandenhaut, 1986) suggested the
following function: i ¼ C1 þ C2 E þ C3 expðC4 EÞ.
The relation between the square of the peak width and the energy [Eq. (86)] is based
on theoretical considerations. The relation holds very well if the doublet splitting of the
x-ray lines is taken into account. The Ka1Ka2 separation increases from a negligible value
for Ca (3.5 eV) to nearly 100 eV for Mo. The observed peak shape of a K line is actually an
envelope of two peaks. This envelope can be represented rather well by a single Gaussian,
but failing to take this doublet splitting into account (i.e., fitting with a single Gaussian
where doublets are required) will result in peak widths that do not obey Eq. (86). To
illustrate this, the observed width of a number of Ka lines as function of the x-ray energy is
presented in Figure 22. The dotted line represents the width of the Ka doublet fitted as one
peak. The solid (straight) line shows the width of the individual lines in the doublet.
X
NP
yP ðiÞ ¼ A Rj Gði; Ej Þ ð87Þ
j¼1
where G are the Gaussians for the various lines with energy Ej and Rj the relative P in-
tensities of the lines. The summation runs over all lines in the group (NP) with Rj ¼ 1.
The transition probabilities of all lines originating from a vacancy in the same (sub)
shell (K, LI, LII, . . .) are constants, independent of the excitation. However, the relative in-
tensities depend on the absorption in the sample and in the detector windows. To take this
into account, the x-ray attenuation must be included in Eq. (87). The relative intensity ratios
are obtained by multiplying the transition probabilities with an absorption correction term:
Rj Ta ðEnj Þ
R0j ¼ P ð88Þ
j Rj Ta ðEnj Þ
Contributions from various subgroups (i.e., between K and L, between LI and LII, etc.)
depend on the type of excitation (photons, electrons, protons) and on the excitation en-
ergy. General values cannot be given and must be determined for the particular excitation
Copyright © 2002 Marcel Dekker, Inc.
Figure 22 FWHM of various Ka lines, fitted as single peak and as a Ka1Ka2 doublet.
4. Modified Gaussians
When fitting very large peaks with a Gaussian, the deviation from the pure Gaussian shape
becomes significant. In Figure 24, a MnK spectrum with 107 counts in the MnKa peak is
shown. One observes a tailing on the low-energy side of the peaks and a continuum that is
Copyright © 2002 Marcel Dekker, Inc.
Figure 23 Fit of a complex L line spectrum of tungsten. In total, 24 transitions, divided over the 3
L subshells, are required for the description on the spectrum.
Figure 24 MnK line spectrum with very good counting statistics. The difference from the
Gaussian response is obtained by subtracting all Gaussian peaks. From this, the peak-shape
correction is calculated.
Figure 25 Fit of the spectrum of a brass sample (NIST SRM 1106) (a) fitted with a simple
Gaussians and (b) fitted with Gaussians including tail and step functions.
where CðiÞ is the numerical peak-shape correction at channel i. Values in the table are
interpolated to account for the difference between the energy scale of the correction and
the actual energy calibration of the spectrum. Similar to the parameters of the non-
Gaussian analytical functions, the shape of the numeric correction seems to vary slowly
from one element to another. This allows us to interpolate the peak-shape correction for
all elements from a limited set of experimentally determined corrections.
A major disadvantage of this method is that it is quite difficult and laborious to
obtain good experimental peak-shape corrections. Although they are, in principle, de-
tector dependent, experience has proven that the same set of corrections can be used for
different detectors with reasonable success, proving the fundamental nature of the ob-
served non-Gaussian shape. Another disadvantage is that the peak-shape correction for
the Kß becomes underestimated if strong differential absorption takes place because the
peak-shape correction is only related to the area of the Ka peak. Also, it is nearly im-
possible to apply this method to the description of L-line spectra. A mayor advantage
however is the computational simplicity of the method and the fact that no extra para-
meters are required in the model.
5. Absorption Correction
The absorption correction term Ta ðEÞ, used in the Eqs. (82) and (88), includes the x-ray
attenuation in all layers and windows between the sample surface and the active area of the
detector. For high energetic photons also, the transparency of the detector crystal needs to be
taken into account. In x-ray fluorescence, the attenuation in the sample, causing additional
changes in the relative intensities, can also be considered, providing the sample composition
is known. The total correction term is thus composed of a number of contributions:
Ta ðEÞ ¼ TDet þ TPath þ TSample ð96Þ
The detector contribution for a Si(Li) detector is given by
Copyright © 2002 Marcel Dekker, Inc.
TDet ðEÞ ¼ emBe ðrdÞBe emAu ðrdÞAu emSi ðrdÞSi ð1 emSi ðrDÞSi Þ ð97Þ
where m, r, and d are the mass-attenuation coefficient, the density, and the thickness of the
Be window, the gold contact layer, and the silicon dead layer. In the last term, D is the
thickness of the detector crystal.
Any absorption in the path between the sample and the detector can be modeled in a
similar way. For an air path, the absorption is given by
Also, various polynomial type functions expressing the escape ratio as a function of the
energy of the parent peak are in use. The coefficients of the function can be determined by
least-squares fitting from experimental escape ratios. For spectra obtained with a Ge
detector, one needs to account in a similar way for both the GeKa and GeKb escape peak
for elements above arsenic.
The incorporation of the sum peaks in the fitting model is more complex. The
method discussed below was first implemented by Johansson in the HEX program
(Johansson, 1982). Initially, the spectrum is fitted without considering pileup peaks. The
peaks are then sorted according to their height and the n largest peaks are retained. Peaks
that differs less than 50 eV are combined to one peak. Using Eqs. (5) and (6), the relative
intensities of all possible nðn þ 1Þ=2 pileup peaks and their energies are calculated and the
m most intense are retained. Knowing the relative intensities and the energies of these m
pileup peaks, they can be included in the fitting model as one ‘‘pileup element.’’ In the next
iteration, the total peak area A of this pileup element is obtained. The construction of the
pileup element can be repeated during the next iterations as more reliable peak areas
become available. In Figure 26, part of an PIXE spectrum is shown fitted with and without
sum peaks included.
Figure 27 Effect of the use of a constraint on the shape of the w2 response surface. (Left: marginal
effect for a large peak; right: important contribution for a small peak.)
E. Examples
To illustrate the working of the nonlinear least-squares-fitting method, an artificial spec-
trum with CuK and ZnK lines is fitted with four Gaussian peaks on a constant continuum.
Using the Marquardt algorithm, the position, width, and area of each peak are de-
termined. The fitting function thus is
" #
X4
ði ajþ1 Þ2
yðiÞ ¼ a0 i þ aj exp ð109Þ
j¼1
2a2jþ2
with i the channel number (independent variable) and aj the parameters to be determined,
13 in total. In Table 10, the values of the parameters to generate the spectrum (true va-
lues), the initial guesses for the nonlinear parameters, and the final fitted values are given.
The initial guesses were deliberately taken rather far away from the true values. Figure 29
(top) shows the fitted spectrum after the first and second iterations. During the second
iteration, the Marquardt algorithm evolved into a gradient search, drastically changing the
position and width of the peaks. Even after five iterations, the calculated spectrum deviates
considerably from the ‘‘measured’’ spectrum, as can be seen from Figure 29 (bottom).
Figure 28 The effect of weighing the least-squares fit is shown on part of an PIXE spectrum with
a very small number of counts per channel.
Area (counts)
CuKa 100,000 0 100,134±321
CuKb 13,400 0 13,163±169
ZnKa 30,000 0 30,092±213
ZnKb 4,106 0 4138±83
Position (channel number)
CuKa 402.05 395 402.03±0.01
CuKb 445.25 450 445.35±0.06
ZnKa 431.55 435 431.59±0.04
ZnKb 478.60 485 478.68±0.09
Width (channels)
CuKa 3.913 3 3.91±0.01
CuKb 4.033 3 3.99±0.05
ZnKa 3.995 3 4.02±0.03
ZnKb 4.123 3 4.06±0.08
Finally, after 10 iterations, a perfect match between them is obtained, with a reduced w2
value of 0.96.
From Table 10, it is evident that the fit was quite successful, with all peak areas,
positions, and widths estimated correctly within the calculated standard deviation. One
observes that the uncertainties in the peak areas are approximately equal to the square
root of the peak area and that the position and the width of the peaks are estimated very
precisely (within 0.01 channel or 0.2 eV), especially for the larger peaks.
By observing how the iteration procedure changes the peak width and position
parameters, one can imagine that something might go wrong. Especially if the spectrum is
more complex, chances are high that the iteration stops in a false minimum or even drifts
away completely. In both cases, physically incorrect parameter estimates will be obtained.
In practice, it is course possible to give much better initial estimates for the peak position
and width parameters than used in this example.
In a next example, a complex spectrum (geological reference material JG1, excited
with MoK x-rays from a secondary target system) is evaluated using nonlinear least-
squares fitting. In Figure 30, the spectrum and the fit are shown together with the residuals
of the fit (see p. 298). Due to the large number of overlapping lines, the method used in the
first example (fitting the position and width of each peak independently) is not applicable
in this case. For the description of the spectrum from 1 to 18 keV, the fluorescence lines of
21 elements were used. Au, Hg, Pb, and Th were treated each as one L group
(L1 þ L2 þ L3 ) and Al, Si, Ti, V, Cr, Mn, Ni, Cu, Zn, Ga, Al, Br, Sr, Rb, Y as one K group
(Ka þ Kb); K, Ca, and Fe were fitted with individual Ka and Kb peaks. The coherently
scattered MoKa radiation was fitted with a Gaussian and the incoherent scattered MoKa
radiation is fitted with a Voigt profile. Including escape and sum peaks, this amounts to
well over 100 peak profiles. Step and tail functions [Eq. (89)] were included for all peaks,
using expressions to relate the step and tail heights and the tail width to the energy of the
peak. The continuum is described by an exponential function [Eq. (81)] with six para-
meters. The least-squares fit thus performed required the refinement of 37 parameters
Copyright © 2002 Marcel Dekker, Inc.
Figure 29 Artificial CuK and ZnK line spectrum fitted with a nonlinear least-squares procedures
to optimize peak area, position, and width. (Top: Fitted spectrum after first and second iteration;
bottom: after fifth and final iterations.)
(4 calibration parameters, 27 peak areas, 6 continuum parameters, 3 step and tail para-
meters, and 1 Voigt parameter). The minimum w2 value of 1.47 is obtained after 16
iterations. The residuals indicate an overall good fit with most of the residuals in the 73
to þ3 interval, without any systematic patterns. Its interesting to note that the fitted
continuum is well below the base of the peaks, especially in the region from channel 200 to
600. The continuum describes correctly the small scattered bremsstrahlung contribution
from the x-ray tube above 12 keV (channel 600) and the Compton scattering in the de-
tector at the low-energy side. Most of the apparent continuum in this secondary-target
EDXRF spectrum is due to incomplete charge collection and tailing phenomena of the
scattered Mo excitation radiation and the fluorescence lines.
are caused by discrepancies between the fitting model and the observed data and cause
inaccuracies in the net peak areas.
1. Error Estimate
Section IX explains how the least-squares-fitting method (linear as well as nonlinear)
allows the estimation of the uncertainties in the fitted parameters. These uncertainties
result from the propagation of the statistical fluctuations in the spectral data into the
parameters. Intuitively, one could come to the conclusion that the standard deviation of
the peak area should be close to the square root of the peak area. This is indeed the case
for a large, interference-free peak on a low continuum, but if the peak is sitting on a high
continuum and=or is seriously overlapped by another peak, the uncertainty in the esti-
mated peak area will be much larger.
A properly implemented least-squares method not only correctly estimates the net
peak areas but also their uncertainty, taking into account the continuum and the degree of
peak overlap, providing, of course, that the fitting model is capable to describe the
measured spectrum. The closer together the peaks are, the higher the uncertainties in the
two peak areas will become. (Theoretically, in the limit of complete overlap, the un-
certainty will be infinite and the area of the two peaks will have complete erratic values,
but their sum will still represent correctly the total net peak area of the two peaks; in
practice, the curvature matrix a will be singular so that the matrix inversion fails.)
Copyright © 2002 Marcel Dekker, Inc.
The uncertainties in the net peak areas can be used to decide if a peak is indeed
present in the spectrum and to calculate the detection limit. A peak area is statistically not
significant different from zero if its value is in the range ±3s. Any value above 3s gives
clear evidence that the peak is present; any value below 73s would indicate that there is
something wrong with the model because truly negative peak areas are physically mean-
ingless. Because the uncertainty in the net peak area includes the influence of continuum
and peak overlap, it can be used to calculate the a posteriori detection limits of the ele-
ments (peaks) present in the spectrum. Three situations can occur:
From the definition of the least-squares method, it follows that the w2 value [Eq. (75)]
estimates how well the model describes the data. The reduced w2 value, obtained by di-
viding w2 by the number of degrees of freedom;
1 1
w2n ¼ w2 ¼ w2 ð110Þ
n nm
has an expected value of 1 for a ‘‘perfect’’ fit. The number of degrees of freedom equals the
number of data points (n) minus the number of parameters (m) estimated during the fit.
Because w2 is also a random variable, the observed value will often be slightly larger or
smaller than 1. Actually, w2 follows (approximately) a chi-square distribution and the 90%
confidence interval is given by
1 X m
1
w2P ¼ ½yi yðiÞ2 ð112Þ
n1 n2 i¼n2 s2i
Although the w2 values gives an indication of the goodness of fit, visual inspection of
the fit is highly recommended. Because of the large dynamic range of the data, a plot of the
spectrum and the fit on a linear scale nearly always give the impression of a perfect fit. A
plot of the logarithm or the square root of the data is more appropriate. The best method
is to plot the residuals of the fit as in done in Figure 30. The residuals are defined as
yi yðiÞ
ri ¼ ð114Þ
si
It is the sum of the squares of these residuals that were minimized by the least-squares
method. Values in excess of þ3 or 73 and especially the presence of a pattern in the
residuals then indicate poorly fitted regions.
Monte Carlo techniques for the simulation of x-ray spectra are becoming more and more
popular, particularly because of the fast computers available today. These simulated
spectra are useful for studying the behavior and performance of various spectrum pro-
cessing methods. The Monte Carlo technique also can be used in quantitative analysis
procedures, as will be discussed in Sec. VIII. B.
Figure 32 Simulation of a ‘‘child’’ spectrum from a ‘‘parent.’’ (Top: original spectrum; middle:
the cumulative distribution; bottom: generated child.)
In contrast to the normal library least-squares method, this method has the advantage that
the library spectra are simulated for a composition close to the composition of the spec-
trum to analyzed, rather than measured from standards. This eliminates the necessity of
applying the top-hat filter and problems related to changes in Kb=Ka ratios, and again the
method combines spectrum evaluation with quantitative analysis.
Finally the ability to simulate x-ray spectra that agree very well with real measured
spectra opens the possibility to used them as ‘‘standards’’ for the quantitative analysis
based on partial least-squares regression (see Sec. VI.B). Indeed, as this method only
functions correctly if the PLS model is built using a large number of standards covering
the entire concentration domain, it seems advantageous to use simulated spectra for this.
All the inter-element interactions can be accounted for by the simulation and only a few
real standards are required to scale the simulated spectra.
Copyright © 2002 Marcel Dekker, Inc.
IX. THE LEAST-SQUARES-FITTING METHOD
The aim of the least-squares method is to obtain ‘‘optimal’’ values for the parameters of a
function that models the dependence of experimental data. The method has its roots in
statistics but is also considered part of numerical analysis. The least-squares parameter
estimate, also known as curve fitting, plays an important role in experimental science. In
x-ray fluorescence, it is used in many calibration procedures and it forms the basis of a
series of spectrum analysis techniques. In this section, an overview of the least-squares
method with emphasis on spectrum analysis is given.
Based on the type of fitting function, one makes a distinction between linear and
nonlinear least-squares fitting because one requires numerical techniques of different
complexity to solve the problem in the two cases. The linear least-squares method deals
with the fitting of functions that are linear in the parameters to be estimated. For this
problem, a direct algebra solution exists. If the fitting function is not linear in one or more
of the parameters, one uses nonlinear least-squares fitting and the solution can only be
found iteratively. A group of linear functions of general interest are the polynomials, the
straight line being the simplest case. The special case of orthogonal polynomials will also
be considered. If more than one independent variable ðx1i ; x2i ; . . . ; xmi Þ is associated with
each measurement of the dependent variable yi , one speaks of multivariate regression.
Spectrum analysis using the library function (e.g., filter-fit method) belongs to this cate-
gory. If analytical function (e.g., Gaussians) are fitted to a spectrum, the method of linear
or nonlinear least square is used, depending on whether nonlinear parameters, such as the
peak position and width, are determined.
A. Linear-Least Squares
Considered the problem of fitting experimental data with the following linear function:
y ¼ a1 X1 þ a2 X2 þ þ am Xm ð120Þ
This function covers all linear least-squares problems. If m ¼ 2; X1 ¼ 1, and X2 ¼ x, the
straight-line equation y ¼ a1 þ a2 x is obtained. For m > 2 and Xk ¼ xk1 , Eq. (120) is a
polynomial y ¼ a1 þ a2 x þ a3 x2 þ þ am xm1 to be fitted to the experimental data
points fxi ; yi ; si g i ¼ 1; . . . ; n. If X1 ; . . . ; Xm represents different independent variables, the
case of multiple linear regression is dealt with. Because of this generality, we will discuss
the linear least-squares method based on Eq. (120) in detail.
Assume a set of n experimental data points:
The minimum is found by setting the partial derivatives of w2 with respect to the para-
meters to zero:
Copyright © 2002 Marcel Dekker, Inc.
@w2 Xn
1
¼ 2 ðyi a1 X1 a2 X2 am Xm ÞXk ¼ 0; k ¼ 1; . . . ; m ð123Þ
@ak i¼1
s 2
i
Dropping the weights 1=s2i temporarily for clarity, we obtain a set of m simultaneous
equations in the m unknown ak :
X X X X
yi X 1 ¼ a1 X1 X1 þ a2 X2 X1 þ þ am Xm X1
X X X X
yi X 2 ¼ a1 X1 X2 þ a2 X2 X2 þ þ am Xm X2
.. ð124Þ
X X X. X
yi X m ¼ a1 X 1 X m þ a2 X2 Xm þ þ am Xm Xm
where the summations run over all experimental data points i. These equations are known
as the normal equations. The solution—the values of ak —can easily be found using matrix
algebra. Because two (column) matrices are equal if their corresponding elements are
equal, the set of equations can be in matrix form as
2X 3 2 X X X 3
yi X1 a1 X1 X1 þ a2 X 2 X 1 þ þ am Xm X1
6 X 7 6 X X X 7
6 yi X2 7 6 X1 X2 þ a2 X 2 X 2 þ þ am Xm X2 7
6 7 6 a1 7
6 .. 7¼6 .. 7 ð125Þ
6 7 6 7
4X . 5 4 X X . X 5
yi Xm a1 X 1 X m þ a2 X2 Xm þ þ am Xm Xm
The right-hand column matrix can be written as the product of a square matrix a and a
column matrix a:
2X 3 2 X X X 3
yi X1 X1 X1 X2 X1 Xm X1 2 a1 3
6X 7 6 X X X 76 7
6 yi X2 7 6 X2 X2 Xm X2 7
6 7 6 X1 X2 76 a2 7
6 . 7
6 7 ¼ 6 7 ð126Þ
6 . 7 6 .. 74 .. 7
6
4 X .. 5 4X X . X 5 5
yi Xm X1 Xm X2 Xm Xm Xm a2
or
b ¼ aa ð127Þ
This equation can be solved for a by premultiplying both sides of the equation with
the inverse matrix a1 ,
a1 b ¼ a1 aa ¼ Ia ð128Þ
or, I being the identity matrix,
a ¼ a1 b ð129Þ
Introducing the weights again, the elements of the matrices are given by
X n
1
bj ¼ y X ; j ¼ 1; . . . ; m
2 i j
ð130Þ
i¼1
si
Xn
1
ajk ¼ Xk Xj ; j ¼ 1; . . . ; m; k ¼ 1; . . . ; m ð131Þ
i¼1
s2i
and
Copyright © 2002 Marcel Dekker, Inc.
X
m
aj ¼ a1
jk bk ; j ¼ 1; . . . ; m ð132Þ
k¼1
where a1 1
jk are the elements of the inverse of the matrix a .
The uncertainty in the estimate of aj is due to the uncertainty of each measurement
multiplied by the effect that measurement has on aj :
Xn 2
@aj
s2aj ¼ s2i ð133Þ
i¼1
@yi
Because a1
jk is independent of yi , the partial derive is simply
@aj 1Xm
¼ 2 a1 Xk ðiÞ ð134Þ
@yi si k¼1 jk
This results in the simple statement that the variance (square of uncertainty) of a fitted
parameter aj is given by the diagonal element j of the inverse matrix a1 . The off-diagonal
elements are the covariances. For this reason, a1 is often called the error matrix. Simi-
larly, a is called the curvature matrix because the elements are a measure of the curvature
of the w2 hypersurface in the m-dimensional parameter space. It can easily be shown that
1 @ 2 w2 X 1
¼ Xk Xj ¼ ajk ð137Þ
2 @aj @ak i
s2i
If the uncertainties in the data points si are unknown and are the same for all data
points ðsi ¼ sÞ, these equations can still be used by setting the weights wi ¼ 1=s2i to 1.
Assuming the fitting model is correct, s can be estimated from the data:
1 X
s2i ¼ s2 s2 ¼ ðyi a1 X1 a2 X2 am Xm Þ2 ð138Þ
nm i
If the uncertainties in the data points are known, the reduced w2 value can be calculated as
a measure of the goodness of fit:
1 X 1 1
w2n ¼ ðyi a1 X1 a2 X2 am Xm Þ2 ¼ w2 ð140Þ
n m i s2i nm
The expected value of w2n is 1.0, but due to the random nature of the experimental data values
slightly smaller or greater than 1 will be observed even for a ‘‘perfect’’ fit. w2n follows a chi-
square distribution with n m degrees of freedom, and a 90% confidence interval is defined by
Copyright © 2002 Marcel Dekker, Inc.
w2 ðn; P ¼ 0:95Þ w2n w2 ðn; P ¼ 0:05Þ ð141Þ
where w2 ðn; PÞ is the (tabulated) critical value of the w2 distribution for n degrees of freedom
at a confidence level P. Observed w2n values outside this interval indicate a deviation
between the fit and the data that cannot be attribute to random statistical fluctuations.
The least-squares estimates of the coefficient Cj are determined by minimizing the weighted
sum of squares:
Xn
w2 ¼ wi ðyi yðiÞÞ2 ð143Þ
i¼1
which results in a set of m þ 1 normal equations in the m þ 1 unknown. Because the Pj ðxi Þ
are a set of orthogonal polynomials, they have the property that
Xn
wi Pj ðxi ÞPk ðxi Þ ¼ gk djk ð144Þ
i¼1
and
b0 ¼ 0 and P0 ðxi Þ ¼ 1 ð151Þ
Thus, an example of a first-order orthogonal polynomial is
C0 P0 þ C1 P1 ¼ C0 þ C1 ðxi a0 Þ ð152Þ
with
X
n .X
a0 ¼ w i xi wi ð153Þ
i¼1
In this part, we consider the fitting of a function that is nonlinear in one or more fitting
parameters. Examples of such functions are a decay curve,
whose minimum will be reached when the partial derivative with respect to the parameters are
zero; however, this will result in a set of m equations for which no general solution exists. The
other approach to the problem is then to consider w2 as a continuous function of the parameters
aj (i.e., w2 will take a certain value for each set of values of the parameters aj for a given dataset
fxi ; yi ; s2i gÞ. w2 thus forms a hypersurface in the m-dimensional space, formed by the fitting
parameter aj . This surface must be searched to locate the minimum of w2 . Once found, the
corresponding coordinate values of the axes are the optimum values of the fitting parameters.
The problem of nonlinear least-squares fitting is thus reduced to the problem of
finding the minimum of a function in an m-dimensional space. Any algorithm that per-
forms this task should operate according to the following:
1. Given some initial set of values for the parameters aini evaluate w2 :
w2old ¼ w2 ðaini Þ
2. Find a new set of values anew such that w2new < w2old .
3. Test the minimum of w2 value:
if w2new is the (true) minimum accept anew as the optimum values of the fit
else w2old ¼ w2new and repeat Step 2.
From the scheme, the iterative nature of the nonlinear least-squares fitting methods be-
comes evident. Moreover, it shows some other important aspects of the method: Initial
values are required to start the search, we need a procedure to obtain a new set of
parameters which preferably are such that w2 is decreasing, and we need to be sure that the
true minimum, not some local minimum, is finally reached.
A variety of algorithms has been proposed, ranging from brute-force mapping
procedures dividing the m-dimensional parameter space in small cells and evaluating w2 in
each point, to more subtle simplex search procedures (Fiori et al., 1981). The most
Copyright © 2002 Marcel Dekker, Inc.
important group of algorithms is nevertheless based on the evaluation of the curvature
matrix. The gradient method and the first-order expansion will be discussed briefly, as they
form the basis of the most widely used LeverbergMarquardt algorithm (Marquardt,
1963; Bevington and Robinson, 1992; Press et al., 1988).
a. The Gradient Method
Having a fitting function y ¼ yðx; aÞ and w2 defined as a function of the m parameters aj ,
Xn
1
w2 ¼ w2 ðaÞ ¼ ½y yðxi ; aÞ2
2 i
ð163Þ
i¼1
s i
where j is the unit vector along the axis j and the components of the gradient are given by
@w2 X 1 @y
¼ 2 ½yi yðxi ; aÞ ð165Þ
@aj i
2
si @aj
It is convenient to define
1 @w2
bj ¼ ð166Þ
2 @aj
The gradient gives the direction in which w2 increases most rapidly. A method to locate the
minimum can thus be developed on this basis. Given the current set of parameters aj , a
new set of parameters a0j is calculated (for all j simultaneously):
a0j ¼ aj þ Daj bj ð167Þ
which follows the direction of steepest descent and guarantees a decrease of w2 (at least if
the appropriate step sizes Daj are taken).
The gradient method works quite well away from the minimum, but near the
minimum, the gradient becomes very small (at the minimum, even zero). Fortunately, the
method discussed next behaves in the opposite way.
b. First-Order Expansion
If we write the fitting function yðxi ; aÞ as a first-order Taylor expansion of the parameters
aj around y0 ,
X @y0 ðx; aÞ
yðx; aÞ ¼ y0 ðx; aÞ þ daj ð168Þ
j
@aj
we obtain an (approximation) to the fitting function which is linear in the parameter in-
crements daj . y0 ðx; aÞ is the value of the fitting function for some initial set of parameter a.
Using this function, we can now express w2 as
" #2
X 1 X @y0 ðxi ; aÞ
w ¼
2
yi y0 ðxi ; aÞ daj ð169Þ
i
s2i j
@aj
and we can use the method of linear least squares to find the parameters daj so that w2 will
be minimal. We are thus fitting the difference y0i ¼ yi y0 ðxi ; aÞ with the derivatives as
Copyright © 2002 Marcel Dekker, Inc.
variables and the increments daj as unknowns. With reference to the section on linear
least-squares fitting [Eq. (122)],
@y0 ðxi Þ
Xj ¼ ð170Þ
@aj
and [Eq. (130) and (131)]
X n
1 @y0 ðxi Þ
bj ¼ 2 i
½y y0 ðxi Þ ð171Þ
i¼1
si @aj
X 1 @y0 ðxi Þ @y0 ðxi Þ
n
ajk ¼ ð172Þ
i¼1
s2i @aj @ak
At the minimum, the partial derivation of w2 with respect to the parameter ak will be zero:
@w2 @w20 X @ 2 w20
¼ þ dak ¼ 0 ð179Þ
@ak @ak j
@aj @ak
0 ajk ð1 þ lÞ; j ¼ k
ajk ¼ ð182Þ
ajk ; j 6¼ k
where ajk is given by Eq. (172) and the matrix equation to be solved for the increments daj is
X
bj ¼ a0jk dak ð183Þ
k
When l is very large (l
1), the diagonal elements of a dominate and Eq. (183) reduces to
bj a0jj dak ð184Þ
or
1 1 @w2
dak b ð185Þ
a0jj j
a0jj @ak
which is the gradient, scaled by a factor a0jj . On the other hand, for small values of
l (l 1), the solution is very close to first-order expansion.
The algorithm proceeds as follows:
1. Given some initial values of the parameters aj , evaluate w2 ¼ w2 ðaÞ and initialize
l ¼ 0.0001.
2. Compute b and a matrices using Eqs. (171) and (172).
3. Modify the diagonal elements a0jj ¼ ajj þ l and compute da.
4. If w2 ða þ daÞ
w2 ðaÞ
increase l by a factor of 10 and repeat Step 3;
If w2 ða þ daÞ < w2 ðaÞ
decrease l by a factor of 10
accept new parameters estimates a a þ da and repeat Step 2.
The algorithm thus performs two loops: the inner loop incrementing l until w2 starts to
decrease and the outer loop calculating successively better approximations to the optimum
values of the parameters. The outer loop can be stopped when w2 decreases by a negligible
absolute or relative amount.
Copyright © 2002 Marcel Dekker, Inc.
Once the minimum is reached, the diagonal elements are an estimate of the un-
certainty in the fitting parameters just as in the case of linear least squares:
s2aj ¼ a1
jj ð186Þ
which is equal to ða0jj Þ1 providing the scaling factor l is much smaller than 1.
In Sec. X, a number of computer programs for linear and nonlinear least-squares
fitting are given. Further information can be found in many textbook (Press et al., 1988).
The book by Bevington and Robinson (1992) contains a very clear and practical discus-
sion of the least-squares method.
In this section, a number of computer routines related to spectrum evaluation are listed.
The calculation routines are written in FORTRAN. Some example programs, calling this
FORTRAN routines, are written in C. The programs were tested using Microsoft FORTRAN
version 4.0A and C version 5.1. Most of the routines are written for clarity rather than
optimized for speed or minimum space requirement.
A. Smoothing
1. Savitsky and Golay Polynomial Smoothing
The subroutine SGSMTH calculates a smoothed spectrum using a second-degree poly-
nomial filter (see Sec. III.B.2).
For each channel i in the spectrum, two windows, one on each side of the channel of width
f6FWHM(Ei ) channels are considered. In both windows, the channel contents are
summed, yielding a left sum L and a right sum R. Both windows are subsequently reduced
in width until either the total sum S ¼ L þ yi þ R falls below some constant minimum M
or until two conditions are met:
pffiffiffiffi
1. S is less than a cutoff value N ¼ A yi , with A a constant.
2. The slope ðR þ 1Þ=ðL þ 1Þ lies between 1=r and r, with r a constant.
The minimum constant M sets the base degree of smoothing in a region of vanishing
counts. The first condition ensures that smoothing is confined to the low statistics region
of the spectrum; the second condition avoids the incorporation of the edges of the peaks in
the averaging.
When the above conditions are satisfied, the average S=(2f6FWHM þ 1) is adopted
as a smoothed channel count. The following parameters were found to yield good results
when treating PIXE spectra: f ¼ 1.5, A ¼ 75, M ¼ 10, and r ¼ 1.3.
B. Peak Search
The subroutine LOCPEAKS locates peaks in a spectrum using positive part of tophyhat
filter (see Sec III.C).
Input: Y Spectrum
NCHAN Number of channels in the spectrum
R Peak search sensitivity factor, typical 2 to 4
IWID Width of the filter, approx. equal to the FWHM of the peaks
MAXP Maximum number of peaks to locate (size of array IPOS)
Output: NPEAK Number of peaks found
IPOS Peak positions (channel number)
C. Continuum Estimation
1. Peak Stripping
The subroutine SNIPBG, a variant of the SNIP algorithm, calculates the continuum via peak
stripping (see Sec. IV.A).
INPUT: Y Spectrum
NCHAN Number of channels in the spectrum
ICH1,ICH2 First and last channels of region to calculate the continuum
FWHM Width parameter for smoothing and stripping algorithm, set it
to average FWHM of peaks in the spectrum, typical value 8.0
NITER Number of iterations of SNIP algorithm, typical 24
Output: YBACK Calculated continuum in the region ICH1ICH2
Comment: Uses subroutine SGSMTH
The routine calls ADJWEIG to adjust the weights. Further, the subroutine ORTPOL is
used to fit the polynomial. The iteration (adjustment of weights) stops when all coefficients
cj change less than one standard deviation or when the maximum number of iterations is
reached.
D. Filter-Fit method
The C program FILFIT is a test implementation of the filter-fit method (see Sec. VI). This
program simply coordinates all input and output, allocates the required memory, and calls
two FORTRAN routines that do the actual work. The subroutine TOPHAT returns the
Copyright © 2002 Marcel Dekker, Inc.
convolute of the spectrum with the top-hat filter or the weights (the inverse of the variance
of the filtered spectrum). The general-purpose subroutine LINREG is called to perform the
multiple linear least-squares fit. The output includes the reduced w2 value, the parameters
of the fit aj (which is an estimate of the ratio of the intensity in the analyzed spectrum to
the intensity in the standard for the considered X-ray lines), and their standard deviation.
The routine ‘‘GETSPEC’’ reads the spectral data and must be supplied by the user.
#include <stdio.h>
#include <malloc.h>
#include <float.h>
#include <math.h>
void fortran TOPHAT( );
void fortran LINREG( );
float spec[2048];
main( )
{
int nchan, first_ch_fit, last_ch_fit, width, ierr;
int i, first_ch_ref, last_ch_ref, ref, num_ref, num_points;
int filter_mode ¼ 0, weight_mode ¼ 1, ioff;
float meas_time, ref_meas_time, *scale_fac;
float *x, *xp, *y, *w, *yfit, *a, *sa, chi;
double *beta, *alpha;
char filename [64];
// input width of tophat filter
scanf(‘‘%hd’’, &width);
// input spectrum to fit and fitting region
scanf(‘‘%s’’, filename);
scanf(‘‘%hd %hd’’, &first_ch_fit, &last_ch_fit);
nchan ¼ Getspec(spec, filename, &meas_time);
num_points ¼ last_ch_fit 7 first_ch_fit þ 1;
// filter spectrum and store in y[ ]
y ¼ (float *)calloc(num_points, sizeof(float));
TOPHAT(spec, y, &nchan, &first_ch_fit, &last_ch_fit, &width, &filter_mode);
//calculate weights of fit and save in w[ ]
w ¼ (float *)calloc(num_points, sizeof(float));
TOPHAT(spec, w, &nchan, &first_ch_fit, &last_ch_fit, &width, &weight_mode);
//read reference spectra, filter and store in x[ ]
scanf(‘‘%hd’’, &num_ref);
scale_fac ¼ (float *)calloc(num_ref, sizeof(float));
x ¼ (float *)calloc(num_points*num_ref, sizeof(float));
for(ref ¼ 0; ref < num_ref; refþþ) {
scanf(‘‘%s’’, filename);
nchan ¼ Getspec(spec, filename, &ref_meas_time);
scale_fac[ref] ¼ ref_meas_time/meas_time;
scanf(‘‘%hd %hd’’, &first_ch_ref, &last_ch_ref);
if (first_ch_ref < first_ch_fit) first_ch_ref ¼ first_ch_fit;
if(last_ch_ref > last_ch_fit) last_ch_ref ¼ last_ch-fit;
ioff ¼ ref*num_points þ first_ch_ref 7 first_ch_fit;
xp ¼ x þ ioff;
TOPHAT(spec, xp, &nchan, &first_ch_ref, &last_ch_ref, &width, &filter_mode);
}
// Program NLRFIT
#include <stdio.h>
#include <malloc.h>
#include <float.h>
#include <math.h>
#define MAX_PERKS 10
#define MAX_CHAN 1024
void fortran MARQFIT();
float spec [MAX_CHAN];
// Fortran common block structure COMMON/FITFUN/NB, NP
struct common_block {short NB, NP;};
extern struct common_block fortran FITFUN;
main( )
{
char specfile[64];
int nchan, first_ch_fit, last_ch_fit, nb, np;
int i, j, n, num_points, num_param, ierr, max_iter;
float ini_pos[MAX_PEAKS], ini_wid[MAX_PEAKS];
float *x, *xp, *y, *w, *yfit, *a, *sa, chi, lamda, crit_dif;
float *b, *beta, *deriv, *alpha;
double *work;
// Input of parameters and spectral data
scanf(‘‘%s’’, specfile);
scanf(‘‘%hd %hd %hd %f’’, &first_ch_fit, &last_ch_fit, &max_iter, &crit_dif);
scanf(‘‘%hd %hd’’, &np, &nb);
for(i ¼ 0; i < np; i þ þ )
scanf(‘‘%f %f’’, &ini_pos[i], &ini_wid[i]);
Input: ISEED Set to any negative number to initialize the random sequence
Output: NRAND Normal distributed random deviate with zero mean and unit
variance.
G. Least-Squares Procedures
1. Linear Regression
Subroutine LINREG is a general-purpose (multiple) linear regression routine (see Sec. IX. A).
3. Nonlinear Regression
The subroutine MARQFIT performs nonlinear least-squares-fitting according to the Mar-
quardt algorithm (see Sec. IX.C).
The routine requires two user-supplied subroutines: FITFUNC to evaluate the fitting
function y(i) with the current set of parameters a and the DERFUNC to calculate the de-
rivatives of the fitting function with respect to the parameters.
SUBROUTINE MARQFIT (IERR, CHISQR, FLAMDA, CRIDIF, MAXITER,
> X, Y, W, YFIT, NPTS, A, SA, NTERMS,
> B, BETA, DERIV, ALFA, ARR)
INTEGER*2 IERR, NPTS, NTERMS
REAL*4 CHISQR, FLAMDA, CRIDIF
REAL*4 X(NPTS), Y(NPTS), W(NPTS), YFIT(NPTS)
REAL*4 A(NTERMS), SA(NTERMS)
REAL*4 B(1), BETA(1), DERIV(1), ALFA(1)
REAL*8 ARR(1)
PARAMETER (FLAMMAX ¼ 1E4, FLAMMIN ¼ 1E 7 6)
4. Matrix Inversion
Subroutine LMINV is a general-purpose routine to invert a symmetric matrix.
Input: ARR Upper triangle and diagonal of real symmetric matrix stored in
linear array, size ¼ N(N þ 1)=2.
N Order of matrix (number of columns)
Output: IERR Error status, IERR ¼ 0 inverse obtained, IERR ¼ 1 singular
matrix
REFERENCES
Johan L. de Vries*
Eindhoven, The Netherlands
Bruno A. R. Vrebos
Philips Analytical, Almelo, The Netherlands
I. INTRODUCTION
*Retired.
depth t below the surface, the remaining fraction of the intensity It(l0) is given by the
Lambert–Beer law:
It ðl0 Þ ¼ I0 ðl0 Þ exp½ms ðl0 Þrs t ðcsc c0 Þ ð2Þ
ms(l0) is the mass-attenuation coefficient of the specimen for photons with wavelength l0
(the subscript s refers to the specimen) and rs is the density of the specimen. Note that due
to the angle of incidence, the path length traveled is given by t csc c0 . The mass-attenuation
coefficient ms(l0) in this equation is thus calculated for the specimen; this is done by adding
the mass-attenuation coefficients for all elements j present in the specimen, each multiplied
with its mass fraction Wj:
X
n
ms ðl0 Þ ¼ ½mj ðl0 ÞWj ð3Þ
j¼1
where n is the total number of elements present in the specimen. The fraction of the in-
cident beam absorbed by the analyte i in the layer between t and (t þ dt) is given by
Wi mi ðl0 Þrs csc c0 dt ð4Þ
It is assumed here that the composition of the specimen is uniform throughout. In other
words, Wi and Wj are independent of the position of the layer (t, t þ dt) within the
specimen.
Only a fraction of the photons absorbed creates vacancies in the K shell; this fraction
is given by (riK 1)=riK, where riK is the absorption jump ratio of the K shell of element i.
The fraction of K vacancies emitting x-rays is given by the fluorescence yield oiK. The
fraction of Ka photons in the total of x-rays emitted for the analyte is given by the
transition probability fiKa. These factors can be combined in a factor designated Qi(l0, li):
riK 1
Qi ðl0 ; li Þ ¼ oi fiKa ð5Þ
riK
In the above derivation, it is assumed that the characteristic line of interest is a Ka. If
another line is considered, the relevant changes need to be made.
The characteristic photons thus generated are isotropically emitted in all directions,
without a preferential direction. Only a fraction is emitted toward the detector. If O is the
solid angle as viewed by the collimator – detector system, expressed in steradians, then the
fraction is given by O=4p. The angle O should be small enough so that the beam can be
considered to be a parallel beam leaving the specimen at a single well-defined angle c00
with the surface. The fraction of characteristic photons with wavelength li not absorbed
between the layer (t, t þ dt) and the surface is given by
Copyright © 2002 Marcel Dekker, Inc.
exp½ms ðli Þrs t csc c00 ð6Þ
All photons reaching the surface and propagating in the direction indicated by the
angle c00 are assumed to be detected. If the detector has absorbing elements (such as
windows) or detection efficiencies different from unity, then these can also be taken into
account. Also, the attenuation by the medium (e.g., air) between specimen and detector
can be calculated using similar expressions.
The intensity of element i, as excited by the incident beam with wavelength l0, is
labeled Pi(l0) (explicitly denoting the primary fluorescence effect) and is given by
Pi ðl0 Þ ¼ I0 ðl0 Þ exp½ms ðl0 Þrs t csc c0
O ð7Þ
mi ðl0 ÞWi Qi ðl0 ; li Þrs cscc0 dt exp½ms ðli Þrs t csc c00
4p
Combining factors leads to
O
Pi ðl0 Þ ¼ I0 ðl0 Þmi ðl0 ÞWi Qi ðl0 ; li Þ ðcsc c0 Þ
4p
exp½ðms ðl0 Þ csc c0 þ ms ðli Þ csc c00 Þrs trs dt ð8Þ
The contribution of all layers between the surface and the ‘‘bottom’’ of the specimen have to
be summed. This can be done by integrating the above expression over dt, from 0 (the surface)
to the ‘‘bottom.’’ In practice, for x-rays, the thickness of bulk specimens can be considered to
be infinite, so the upper limit is 1 . Note that t is always used in combination with rs, so the
integration will be done over rst. Taking all constant factors outside the integral, one obtains
O
Pi ðl0 Þ ¼ I0 ðl0 Þmi ðl0 ÞWi ðcsc c0 ÞQi ðl0 ; li Þ
4p
Z1
exp½ðms ðl0 Þ csc c0 þ ms ðli Þ csc c00 Þrs trs dt ð9Þ
0
Taking a ¼ (ms(l0) csc c0 þ ms(li) csc c00 ) and noting that rs dt ¼ d(rst), the following
expression is obtained for the primary fluorescence of the analyte i in the specimen s:
I0 ðl0 Þmi ðl0 ÞWi ðcsc c0 ÞQi ðl0 ; li ÞðO=4pÞ
Pi ðl0 Þ ¼ ð11Þ
ms ðl0 Þ csc c0 þ ms ðli Þ csc c00
Often, the element specific factors, given by Eq. (5), are combined with the instrument
specific factor O=4p. This leads to a simpler expression for the primary fluorescence:
mi ðl0 Þ
Pi ðl0 Þ ¼ Ki I0 ðl0 Þ Wi ð12Þ
ms ðl0 Þ þ Gms ðli Þ
where Ki ¼ ðO=4pÞQi ðl0 ; li Þ and G ¼ csc c00 =csc c0 ¼ sin c0 = sin c00 .
In the above derivation, the following assumptions have been made:
(a) The specimen is completely homogeneous.
(b) The specimen extends to infinity in three dimensions.
Copyright © 2002 Marcel Dekker, Inc.
(c) The primary rays are not scattered on their way to the layer dt.
(d) No enhancement effects occur.
(e) The characteristic radiation is not scattered on its way to the specimen surface.
The simplest case is excitation of Ka (or Kb) photons. For the characteristic photons
associated with the L lines, other effects, such as Coster–Kronig transitions and so forth
have to be taken into account when describing the fraction of absorbed primary photons
that give rise to characteristic photons. For a more detailed discussion on these effects,
refer to Chapter 1.
There are two criteria to be satisfied for a characteristic photon lj to be able to cause
secondary fluorescence:
1. It must be excited by the incident photon l0 (this means that the energy of the
incident photon must be higher than the energy of the absorption edge
associated with lj).
2. The energy of the photon lj must be higher than the energy of the absorption
edge for the analyte.
The derivation of the intensities of characteristic x-rays as a function of specimen
composition has first been done by Sherman (1955). The resulting equations, however,
were rather unwieldy. Equations (9) and (13) have been published later by Shiraiwa and
Fujino (1966) and Sparks (1976).
In practice, many computer programs only consider a few lines for enhancement,
and lines that have a low transition probability f are usually neglected. Data regarding
these transition probabilities for a line within a series can be found, for example, in
Appendix II of Chapter 1. The enhancement phenomenon will be more pronounced if the
x-rays of the enhancing elements are only slightly more energetic than the energy of the
absorption edge of the element i. The enhancement may contribute up to 40–50% of
the total fluorescent radiation Ii , especially where the concentration of the enhancing
elements is much greater than the concentration of the analyte. This applies even more for
the light elements, where the primary spectrum may not be very effective as the most
intense wavelengths are far away from the absorption edges of these light elements.
The effect of scattered radiation is generally ignored, although its contribution can
also be calculated (Pollai et al., 1971).
If the characteristic line of the analyte considered is one of its L lines, then it is
possible that other lines of the analyte itself enhance the characteristic line considered. The
energy of the K edge of La (atomic number Z ¼ 57) is 38.9 keV (see Chapter 1), so the K
lines of all elements with Z < 57 are excited if, for example, an x-ray tube is used at
Copyright © 2002 Marcel Dekker, Inc.
Table 1 Data for Selected Absorption Edges and Characteristic Line of Pb
K 88.04
L1 15.86 La1(L3–M5) 10.55
L2 15.20 Lb1(L2–M4) 12.61
L3 13.04 Lg1(L2–N4) 14.76
Source: Appendix I and Appendix II of Chapter 1 (this volume).
Also, tertiary fluorescence is possible, where the incident photon excites element k (pri-
mary fluorescence), whose radiation excites element j (secondary fluorescence), whose
radiation, in turn, excites element i, causing tertiary fluorescence. This contribution is
generally lower than 3% of the total fluorescence and is commonly ignored, as shown by
Shiraiwa and Fujino (1967, 1974).
3. Excitation by Continuous Spectra
If the incident beam is polychromatic rather than monochromatic, Eq. (15) needs to be
calculated for each wavelength. Wavelengths longer than the wavelength of the absorption
edge ledge,i of the analyte cannot excite fluorescence, so these need not to be considered. If
JðlÞ is the function representing the tube spectrum, then Eqs. (11) and (13) can still be
used, provided that I0 ðl0 Þ is replaced by JðlÞ, representing the intensity of the incident
spectrum at wavelength l:
lZedge; i lZedge; i " #
X
Ii ¼ Ii ðJðlÞÞ dl ¼ Pi ðlÞ þ Sij ðl; lj Þ dl ð16Þ
j
lmin lmin
This equation allows the calculation of the intensity of characteristic radiation of a given
analyte in an infinitely thick specimen. Equations such as Eqs. (11), (14), and (16) are often
referred to as fundamental parameter equations because they allow the calculation of the
intensity of fluorescent radiation as a function of the composition of the specimen (weight
fractions Wi ), the incident spectrum [JðlÞ], and the configuration of the spectrometer used
(c0 ; c00 , and O). All other variables used are fundamental constants, such as the mass-at-
tenuation coefficients for a given element at a given wavelength or its fluorescence yield
and so forth.
Copyright © 2002 Marcel Dekker, Inc.
C. Some Observations
The integration over t in Eq. (9) is taken from zero to infinity. It is obvious that the first
layers contribute more to the intensity of li than the more inward layers. Theoretically,
even at large values of t, a very minor contribution to the intensity is still to be expected.
Often, the (minimum) infinite depth is defined arbitrarily as that thickness t where the
contribution of the layer ðt; t þ dtÞ is 0.01% of that of the surface layer. In this case, it is
defined relative to the surface layer. Alternatively, it is defined as the thickness where the
contribution to the total intensity is less than 1%. In this case, it is relative to the total
intensity from a ‘‘truly infinitely’’ thick specimen. The value of the infinite thickness de-
pends on the value of the absorption coefficients and the density of the specimen. In
practice, it may vary from a few micrometers for heavy matrices and long wavelengths to
centimeters for short wavelengths and light matrices, as in solutions.
For a given element and a fixed geometry, an efficiency factor Cðl0 ; li Þ can be in-
troduced:
mi ðl0 Þ
Cðl0 ; li Þ ¼ Pn Pn ð17Þ
j¼1 Wj mj ðl0 Þ þ G j¼1 Wj mj ðli Þ
P
It is obvious that the terms Wj mj are the origin of nonlinear calibration lines, as
variations in Wj influence the value of the denominator. For a pure metal, this reduces to
mi ðl0 Þ
Cðl0 ; li Þ ¼ ð18Þ
mi ðl0 Þ þ Gmi ðli Þ
In the general case, the wavelengths in the primary spectrum close to the absorption
edge are the most effective in exciting the analyte i. The efficiency factor Cðl0 ; li Þ is thus a
combination of the absorption curve of analyte i as a function of l and the spectral dis-
tribution. In a first approximation, an ‘‘effective wavelength’’ le can be introduced, which
has the same effect of excitation of element i as the total primary spectrum. The exact value
of this le will be influenced by the characteristic tube lines if they are active in exciting i.
Otherwise, le can, in general, be assumed to have a value of approximately two-thirds of
the absorption edge ledge. Its actual value is, however, dependent on the chemical com-
position of the specimen. For instance, for Fe, the wavelength of the K edge ledge is
0.174 nm (7.11 keV) and an effective wavelength le of 0.116 nm is obtained using this rule
of thumb. In the ZnO–Fe2O3 system, le was found to vary from 0.130 nm for 100% Fe2O3
to 0.119 nm for 10% Fe2O3 in ZnO. The estimated value of 0.116 nm is in good agreement
with the experimental one for Fe2O3.
Another interesting case is the analysis of a heavy element i in a light matrix. In
the summation in the denominator of Eq. (12), the terms Wi mi ðl0 Þ and Wi mi ðli Þ are
then the most important. The other terms can, in first approximation, be neglected if
Wi is not small. However, that means that the terms Wi in the numerator and the
denominator cancel and the measured intensity becomes independent of Wi ; thus, the
analysis becomes impossible in this extreme case. A solution to this problem is found
by making the influence of the terms Wi mi ðlÞ in the denominator less dominating by
adding a large term Wa ma ðlÞ. This can be done be making Wa large (e.g., diluting) or
ma ðlÞ large, by adding heavy absorber. Equation (17) enables one to calculate before-
hand how large this term should be to eliminate fluctuations in the concentration of the
other elements.
In the derivation of the fluorescence of the analyte i in the preceding paragraphs, the
following simplifications were made:
Copyright © 2002 Marcel Dekker, Inc.
1. First, it was assumed that the primary rays follow a linear path to the layer dt at
depth t. However, the primary rays may also be scattered. In general, the loss in
intensity of the primary beam of photons due to scattering may be neglected.
These scattering effects become more important when the primary x-rays are
more energetic and the average atomic number of the matrix decreases. This
scatter may give a higher background in the secondary spectrum, thus leading to
a poorer precision of the analysis. On the other hand, the excitation efficiency
may be enhanced, as the primary rays ‘‘dwell longer’’ in the active layers, thus
having a higher probability to encounter atoms of element i. This effect may
overrule the increase in intensity of the background radiation. A case in hand is
the determination of Sn in oils which gives better results using the SnK lines at a
high x-ray tube voltage, than using the SnL lines at moderate voltages.
Incidentally, this scattering of the primary radiation makes it possible to check
the voltage over the x-ray tube. According to Bragg’s law, the intensity of the
primary spectrum is zero at an angle y0, given by
1 nlmin
y0 ¼ sin ð19Þ
2dcrystal
where 2dcrystal is the 2d spacing of the crystal used and n is an integer number.
lmin (in nm) is given by
1:24
lmin ¼ ð20Þ
V
where V is the voltage on the x-ray tube, expressed in kilovolts. In practice, a
lower value for V will be found when Compton scattering dominates over
Rayleigh scattering and thus lmin found has a value too low by the Compton
shift, which is about 0.024 nm for most spectrometers; the actual value depends
on the incidence and exit angle (see also Sec. V.C.1.)
2. The integral in Eq. (9) was taken from zero to infinity; further, it was assumed
that the specimen is completely homogeneous. This, of course, is never realized
in practice, as we are dealing with discrete atoms in chemical compounds. In
powders, the different compounds may have a tendency to cluster. The particles
will, in general, have different sizes and shapes. Putting the sample into solution,
either aqueous or solid (melt) may overcome this problem.
It was stated formerly that infinite thickness may vary from 20 mm to a few cen-
timeters. However, the most effective layers are much thinner. Thus, the number of dis-
crete particles actually contributing to the fluorescent radiation may be rather small.
B. Random Errors
1. Counting Statistics
If an x-ray measurement consisting of the determination of a number of counts N is re-
peated n times the results N1,N2, N3, . . . ,Nn would spread about the true value N0. If n is
large, the distribution of the measurements would follow a Gaussian distribution,
" #
1 ðN N0 Þ2
WðNÞ ¼ pffiffiffiffiffiffiffiffiffi exp ð24Þ
2pN 2N
pffiffiffiffiffiffiffiffiffiffiffiffi
provided N is also large. The standard deviation s of the distribution is equal to Nmean ,
again if n and N are large, where Nmean is the mean of n determinations. From the
properties of the Gaussian distribution, the following hold:
68.3% of all values N will be between N0s and N0 þ s.
95.4% of all values N will be between N02s and N0 þ 2s.
99.7% of all values N will be between N03s and N0 þ 3s.
pffiffiffiffi
ffiffiffiffi is a certain probability that the true result N0 will lie between N N
Similarly,pthere
and N þ N, assuming the same distribution for N and N0. Measurement results are
commonly expressed as a count rate (intensity per unit time) instead of an intensity,
which gives the number of counts collected in the counting interval. This allows an easier
comparison of measurements made with different counting times, but the measuring
time needs to be specified in order to be able to assess the counting statistical error.
Copyright © 2002 Marcel Dekker, Inc.
The determined concentration is dependent on the net count rate, which is the peak
count rate Rp minus the background count rate Rb. The total measuring time T equals
tp þ tb , where tp and tb are the measurement times for peak and background, respec-
tively. In modern equipment, there is no significant statistical error in the measurement
of t. We can thus assume that R follows the same Gaussian distribution as N with the
same relative standard deviation eN. eN is defined as
sN
eN ¼ ð25Þ
N
Hence,
pffiffiffiffi
N 1 1
eN ¼ ¼ pffiffiffiffi ¼ pffiffiffiffipffiffi ¼ eR ð26Þ
N N R t
and
pffiffiffiffi
R
sR ¼ eR R ¼ pffiffi ð27Þ
t
It is obvious that the relative counting error decreases as t increases.
When a net count rate has to be determined the peak, Rp and the background, Rb,
have to be measured; there are thus two independent variables. The standard deviation of
the net intensity, sd, is given by
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Rp Rb
sd ¼ s2p þ s2b ¼ þ ð28Þ
tp tb
2. Instrumental Errors
If the instrumental and counting uncertainties are random and independent variables, then
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
etot ¼ e2instr þ e2count ð40Þ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
einstr ¼ e2tot e2count ð41Þ
Although the counting error is influenced by the instrumental error, Eq. (41) is still a good
approximation. stot can be found from a series of repeated results of one measurement;
thus, scount is known. To check the instrumental instability, all the functions should be
measured separately. A radioactive source 55Fe, for example, can be used to check the
detector and electronic circuitry. The x-ray tube can be checked by repeating the mea-
surements with specimen and goniometer in a fixed position, eliminating errors stemming
from the mechanics of the spectrometer.
In another series of experiments, recycling between different angles checks the re-
producibility of the goniometer, whereas repositioning the specimen between measure-
ments checks the specimen holder, the reproducibility of the specimen loading mechanics,
and so forth. A comprehensive series of tests for wavelength-dispersive x-ray fluorescence
spectrometers is described in the Australian Standard 2563-1982 (1982). Sometimes the
error stot found is smaller than expected, or even smaller than scount . This may indicate that
an unexpected systematic error is involved, or there may be an uncorrected deadtime in the
equipments, (i.e., the true counting rate is higher than measured, which means that the
relative error is smaller).
3. Detection Limit
A characteristic line intensity decreases with decreasing concentration of the analyte and
finally disappears in the background noise. The true background intensity may be con-
stant, but the results of the measurements fluctuate around a mean value Rbmean . To be
significantly different from the background, a signal Rp must, although it is larger than
Rbmean , be distinguished from the spread in Rb . In other words, if we measure a signal Rp
larger than Rb and we assume the analyte is present, what is the probability that our
assumption is correct? If the results of the measurements are random and follow a
Gaussian distribution, then this probability is determined by sRb . If the measurement Rp is
higher than Rb þ 2sRb , then the probability that our assumption is correct is approxi-
mately 95% if a higher certainty is required (e.g., 99.7%), then Rp should be larger than
Rb þ 3sRb . Thus, the net intensity is 3sRb and the detection limit, DL, would be
3sRb
DL ¼ ð42Þ
M
where M is the sensitivity in counts per second per percent. So, the detection limit in the
above equation is the concentration corresponding to a net peak intensity of 3sRb .
However, in x-ray spectrometry, the background signal is specimen dependent and cannot
Copyright © 2002 Marcel Dekker, Inc.
be measured independently, as in radioactivity measurements. Hence, Rb has to be mea-
sured in an off-peak location in the spectrum. The resultffi Rb found in measuring time t s is
pffiffiffiffiffiffiffiffiffiffiffiffiffi
assumed to be Rbmean and sRb is assumed to be ðRb =tÞ. Thus, two measurements have to
be made: Rp and Rb , each in time t s. The detection limit thus becomes
pffiffiffi rffiffiffiffiffiffi
3 2 Rb
DL ¼ ð43Þ
M T
where T ¼ 2t. If we are satisfied with a 95% probability that our assumption is correct,
then
pffiffiffi rffiffiffiffiffiffi
2 2 Rb
DL ¼ ð44Þ
M T
which is roughly equal to
rffiffiffiffiffiffi
3 Rb
DL ¼ ð45Þ
M T
It is obvious that the detection limit decreases if the counting time increases. However, the
total error in Rb contains the instrumental error as well. Thus, there is no sense to increase
the counting time when the instrumental error dominates.
Ingham and Vrebos (1994) have shown that the detection limit can be improved by
carefully selecting a primary filter. If the application of such a primary beam filter re-
duces the intensity of the (scattered) continuum from the tube more than it affects the
sensitivity, the detection limit is improved. The loss in sensitivity M needs to be more
than compensated for by the reduction in background intensity; as from Eq. (45), the
detection limit is proportional to the square root of Rb and inversely proportional to the
sensitivity M.
6. Particle Statistics
Only a limited volume of the specimen can actually contribute to the fluorescent radiation.
As long as this active volume is the same in standards and actual specimens and the atomic
distribution is completely homogeneous, this poses no problem. However, the atoms are
bound into chemical compounds, forming finite particles with different chemical compo-
sitions. The analyte may only occur in particles with a certain chemical composition and
not in other particles. Then, only these specific particles can contribute to the fluorescent
radiation of the analyte i. Therefore, the count rate Ri measured depends on the number of
those particles present in the active volume, where, evidently, the first layers contribute
most of the fluorescent radiation.
Table 2 gives an indication of the penetration depth of radiation of various wa-
velengths into matrices with varying absorption power. It is evident that for most solid
specimens, the fluorescent radiation originates within 20 mm or less from the surface. To
get an idea of how many particles can actually contribute to the fluorescent intensity of
analyte i, let us assume that the irradiated area of 10 cm2 is covered with cubic particles
of 100 mm dimension in a random fashion. Assuming a filling factor of 0.8 and assuming
that the analyte i is only present in 10% of the particles, then 10 108 104 101
0:8 ¼ 8000 particles could be actually contributing. Assuming
pffiffiffiffiffiffiffiffiffiffi a Gaussian distribution,
then this number would have a standard deviation of 8000 ¼ 90 particles, or a relative
standard deviation of approximately 1.1%. If the concentration of analyte is only 1%,
then this relative standard deviation would be roughly 3.3%. In practice, these errors
might even be larger, as the radiation of the specimen is not homogeneous because the
primary spectrum originates in a rather small anode and passes through a large window
and is, thus, conically shaped. Spinning the specimen in its own plane during the ana-
lysis will reduce this error. Furthermore, the first layers are the most effective, having
Table 2 ‘‘Infinite’’ Thickness (in mm for Certain Analytical Lines as a Function of the Matrix)
Analytical line Fe base Mg base H2O solution Borate Borate La2O3 10%
C. Systematic Errors
1. Dead T|me
After an x-ray photon is detected in the counter and accompanying electronics, it takes a
certain time before the counting circuit is ready to accept the next photon. Any photon
entering the counter within this period, called the dead time of the counter circuit, is
simply not registered and is thus lost. This dead time is of the order of a few microseconds.
The counting losses are thus dependent on the actual count rate. The measured count rate
Rm is always lower than the true count rate RT . Their relation can be approximated by the
expression
Rm
RT ¼ ð47Þ
1 td Rm
where td is the dead time. For instance, if td ¼ 1 ms and Rm ¼ 105 counts per second, the
dead-time loss is approximately 10%. With modern equipment, very high count rates can
be handled, to reach sufficient precision in a short time. It is therefore necessary to reduce
these losses. In most wavelength-dispersive instruments, an automatic dead-time correc-
tion circuit is included. Energy-dispersive instruments, on the other hand, tend to collect
counts for the specified time; thus, the measurement takes longer because the total time
required in this case consists of the specified measuring time (‘‘lifetime’’) and the dead time.
2. Matrix Effects
The fluorescent intensity of the analyte i is, as discussed earlier, not only dependent on its
concentration but can also be strongly dependent on the composition of the specimen
itself. The primary rays will be absorbed and scattered and secondary fluorescence may
occur. All of these effects depend on the chemical composition of the specimen. The
importance of these effects depends on the concentration of these matrix elements and
their influence (e.g., their absorption of primary and secondary x-rays). These matrix ef-
fects may introduce large systematic errors when they are not properly accounted for, as
discussed in Sec. V.
2. Spectral Overlap
Two or more characteristic lines may not be completely separated from the analytical line.
This separation may be improved by using a crystal with better dispersion (e.g., a lower d
value). However, the choice must often be made between high intensity and high disper-
sion. If the disturbing line is due to a high-order crystal reflection, its influence may be
strongly reduced by the proper setting of the pulse-height selector. However, in some
cases, the escape peak of the interfering line may be within the pulse-height selector
window. For instance, the third-order reflection using a penta erythritol (PE) crystal, of
the characteristic tube lines of a Sc anode, scattered by the specimen will slightly interfere
with the analysis for Al in a light matrix, as the energy of the ScKa escape peak in an Ar-
filled gas detector is very close to the energy of the AlKa line.
Often, the overlap is due to a diagram line of an element of which another diagram
line is free of overlap. As, in general, two diagram lines of one element have a constant
intensity ratio, the measured intensity of the nonoverlapped line of the disturbing element
multiplied by a constant factor (experimentally determined) may be subtracted from the
measured intensity of the analytical line to give the characteristic intensity. Absorption
effects can, however, strongly influence the ratio. This is most clearly the case when there is
an absorption edge of a major element between the two diagram lines considered.
Copyright © 2002 Marcel Dekker, Inc.
3. Primary Radiation Scattered by the Specimen
Photons of all wavelengths present in the white spectrum of the incident beam are scat-
tered by the specimen, including the characteristic lines of the tube anode, giving rise to a
continuous background. If, however, the specimen consists of rather coarse grains, it may
happen that a crystallite is in a favorable position for Bragg diffraction for a wavelength of
the continuum; thus, a sharp peak will be found in the spectral analysis. The influence of
primary tube lines, coherently or incoherently scattered, may be eliminated by the proper
choice of the anode.
5. Satellite Lines
The common wavelength tables give only characteristic K, L, and M lines of most ele-
ments. The M lines of heavy elements may interfere with the K lines of light elements.
Nondiagram satellite lines may also occur, often giving rise to an unexpected background
level. One of the tables that includes such lines is provided by NIH (Bethesda, MD)
(Garbauskas and Goehner, 1983).* The lines that are most obvious to observe are some of
the satellites lines of Al, Si, and P with a wavelength-dispersive spectrometer. In Figure 3,
a spectrum over aluminum is shown, on which some of these lines have been annotated.
*Database made available through C. Fiori, National Institutes of Health, Bethesda, Maryland.
control specimen fall outside a predetermined range, then adequate measures must be
taken. The first step is usually to perform a correction for drift. DeGroot (1990) has de-
scribed how the use of statistical process control (SPC) can be beneficial in this respect.
The SPC charts that can be maintained in this way allow one to check the performance of
the spectrometer system.
C. Drift-Correction Monitors
Correction for drift can be made by measuring selected specimens for each of the ana-
lytes and calculate the ratio between the observed intensities and those obtained when the
calibration was performed. The ratios can then be applied as a correction to the slope
of the calibration graphs or, as is done more often, the measured intensities of the
unknowns are corrected for drift, prior to the conversion to concentration. If drift
correction is performed regularly (e.g., once a day), the drift between subsequent
measurements is very small and, thus, measurements with high precision are required,
otherwise the counting statistical error of the measurement would become dominant.
Drift correction should not even be applied unless the correction is significant. Because
the precision of an intensity measurement is determined by the number of counts col-
lected, drift-correction monitors are, ideally, specimens on which high count rates can be
obtained. Drift-correction monitors do not have to be specimens with a composition
similar to the unknowns.
As different components of the spectrometer (such as the x-ray tube) age, not only will
the sensitivity be affected (this is generally a downward trend), but in some cases, the
background will vary also. To correct for this change in background intensity, measure-
ments of the background must be performed. This poses a problem, inasmuch as the count
Copyright © 2002 Marcel Dekker, Inc.
rate on the background is usually low and thus measurements are not vary precise or they
require a long measuring time. However, the considerations on the counting statistical error
made earlier (Sec. III.B.1) offer some suggestions. First, the intensity of the background is
irrelevant when the intensity of the analyte peak is much higher. Obviously, this has to
apply for all specimens to be measured. If that is the case, then variations in the background
intensity are also negligible. When the background matters, it should already be measured
anyway and then drift correction on the background intensity is not required, as the net
count rate is corrected for drift. The only case that is not covered here is the contribution of
spectral contamination from the x-ray tube (e.g., Cu, W, Ag, Fe, etc.). In that case, the
background, including the contamination, must be measured on peak. This implies that
measurements must be done for specimens with zero analyte concentration. Again, two
cases can be distinguished: If the background including the contamination is not important
compared to the count rate observed, then no corrections are required. On the other hand,
if one of the analytes is present at low concentration, then the contribution to the back-
ground due to the contaminant must be checked periodically and taken into account.
After the drift correction is performed, a quality control specimen should be mea-
sured to verify the procedure.
D. Recalibration Standards
Sometimes (e.g., after a major maintenance on a spectrometer), drift correction does not
bring the quality control specimens back in line with the expectations. In those cases, the
calibration curve must be reconstructed. This can be done by measuring all the standard
specimens again and repeating the complete calibration procedure. Because calibrations
often use many standards and validating each calibration is required, this can be a time-
consuming process, even if the validation is limited to a quick visual inspection of the
calibration graphs. In such cases, a recalibration can be performed based on only a few
standards. The idea of recalibration is to reconstruct the calibration graph, without having
to measure all the standards again. This is done by selecting a few standard specimens for
each analyte and measuring these (the top and bottom point in Fig. 4). Subsequently,
when determining the parameters (such as slope and intercept) of the regression, for the
concentrations of these specimens in the calibration the certified values are no longer used,
but the values as found on the calibration line at the time of the original calibration (the
x-ray values) are. These x-ray values have been found based on all standard specimens
used, and the idea is to ‘‘fix’’ the calibration line again through these points. As a result,
the statistical data are now skewed, but the values for the slope and the intercept are very
close to the original ones; the small differences between old and new values are due to the
counting statistical errors in the measurements and these are also present when unknowns
are measured. The specimens used for recalibration have also been used for the calibra-
tion. The only requirement is that the recalibration specimens have count rates that are
different enough so that the determination of the slope is accurate enough. For each of the
calibration lines, the number of selected specimens must be at least the same as the number
of parameters to be determined. If the slope and intercept are determined, at least two
recalibration standards are required. However, if in this case three or four standards are
used, it is possible to detect gross counting artifacts (e.g., caused by mislabeled standards,
incorrect loading, etc.). The root mean square error or the correlation coefficient on the
calibration line (which can only be calculated if more specimens are used than parameters
determined) has no relationship with the accuracy of the analysis. In fact, a near-perfect
correlation should be obtained. As in the case of drift correction, it is recommended to
Copyright © 2002 Marcel Dekker, Inc.
Figure 4 The original calibration line, based on seven data points (h) can be reconstructed using
only two data points (in this case, top and bottom), with concentrations (x-ray values) modified to
those obtained on the calibration (j).
measure, after recalibration, the quality control specimen(s), as gross errors might thus be
identified before unknown specimen are analyzed.
E. Conclusion
Setting up a calibration to be used over extended periods of time requires considerable
amounts of work and preparation. It involves not only the selection and procurement
of standard specimens but also drift correction monitors and recalibration standards.
Also, the specimen preparation method is an essential part of the whole procedure.
The result, however, is the ability to produce quantitative results to a previously as-
sessed degree of accuracy and precision over extended periods of time with minimal
work, once the specimen preparation procedure is set up and the initial calibration is
performed.
Figure 5 Straight lines through the data points can be determined by minimizing either the sum of
the squares of the residuals DI or DW.
B. Matrix Effect
Applying Eq. (50) or (53) requires that all standards must be similar to the unknown in all
aspects considered: matrix effect, homogeneity, and so on. This would lead to the use of
standard specimens with a very limited concentration range. Such a requirement is in
disagreement with the observation that the variance on the slope factor is smaller with
increasing range: The use of standard specimens covering only a small concentration range
will lead to a calibration graph with large uncertainty on the slope and intercept. On the one
hand, this advocates the use of a set of standards with a wide range of concentrations; on
the other hand, the requirement of similarity in matrix effects tends to limit the range.
Obviously, a compromise must be made. Equation (50) is a simplification of the more
general equation describing the relationship among analyte concentration Wi , specimen
homogeneity Si , measured intensity Ii , and matrix effect Mi :
Wi ¼ Ki Ii Mi Si ð55Þ
The term ‘‘specimen homogeneity’’ also includes the grain size effect and the mineralogical
effect. These are notoriously difficult to treat mathematically; in fact, most methods de-
scribing the grain size effect rigorously assume, for example, the dispersed phase to be
perfect spheres of a given diameter or an arrangement of cubes (Bonetto and Riveros,
1985). Other methods allow more variability, but these also require a priori more in-
formation about the specimen, such as the composition of the individual granular phases,
Copyright © 2002 Marcel Dekker, Inc.
the average shape and size of the phases, and so forth (Hunter and Rhodes, 1972; Lubecki
et al., 1968; Holynska and Markowicz, 1981). The fact that the specimen homogeneity is as
yet not described by a single successful method is one of the reasons that Eq. (55) is
commonly reduced to
Wi ¼ K i I i Mi ð56Þ
Fortunately, by using adequate specimen preparation methods the effect of Si between
specimens (standards as well as unknowns) can be rendered constant. This constant factor
is then absorbed by the sensitivity Ki .
As a first approximation, the degree of variation in matrix effect between two spe-
cimens for a given analyte i can be estimated by calculating, for both compositions, the
following parameter:
mi
Ii ¼ Pi ð57Þ
ms ðl0 Þ þ Gms ðli Þ
where ms ðl0 Þ and ms ðli Þ are the mass-attenuation coefficients of the specimen considered
for wavelengths l0 and li , respectively, and G is the geometrical factor for both compo-
sitions. The relative difference between these expressions should not exceed a few percent,
otherwise the matrix effects become too important to ignore. If the specimens include flux
or a binding agent, then these must also be considered in Eq. (57).
With increasing range of concentrations, deviations from linearity will be observed
due to variations in matrix effects between specimens and standards. The analyst must
then resort to other methods to obtain accurate results.
Matrix effects are studied most easily by considering binary systems (i.e., specimens
with only two elements or compounds).
In the case of absorption (both primary and secondary absorption must be con-
sidered), three cases can be distinguished (see Fig. 6):
Enhancement will generally lead to a calibration graph like curve 4 in Figure 6. The
effect of enhancement is usually smaller than that of positive or negative absorption, as
indicated by the position, relative to curve 1, of curves 2 and 3 and curve 4.
It can be shown that the behavior of the calibration curves can be explained in terms
of attenuation coefficients only if absorption is the only matrix effect. Furthermore,
if monochromatic excitation is used, a single constant (calculated from attenuation
Copyright © 2002 Marcel Dekker, Inc.
Figure 6 Calibration curves for binaries. Curve 1: no net matrix-effect; curve 2: net absorption of
the analyte’s radiation by the matrix (positive absorption); curve 3: net absorption of the analyte’s
radiation by the analyte (negative absorption); curve 4: enhancement of the analyte’s radiation by
the matrix.
coefficients) suffices to express the effect of one element on the intensity of another. In the
following section, various methods to deal with matrix effects will be discussed.
characteristic lines. They are also shifted toward longer wavelengths by an amount Dl,
which is given by
Dl ¼ 0:00243ð1 cos cÞ ð61Þ
0 00
where c is the angle through which the radiation is scattered: c ¼ c þ c and Dl is ex-
pressed in nanometers.
In the spectrometer used for the recording of the spectra in Figure 8, c is 100 .
The Compton shift, Dl, is thus 0.0029 nm for this configuration. The maxima of the
Compton peaks (at 0.0576 nm and 0.0644 nm) in Figure 8 are in good agreement with
the theoretical values: 0.0546 nm þ 0.0029 nm ¼ 0.0575 nm and 0.0613 nm þ 0.0029 nm ¼
0.0642 nm, respectively.
The intensity of the Compton-scattered radiation is higher for shorter wavelengths
and for specimens consisting of elements with low atomic numbers. For a given wave-
length (e.g., the characteristic radiation of a tube line), the intensity of Compton scatter
decreases as the specimen consists of more and more elements with higher atomic
numbers. For specimens made up of oxides, the scattered intensity is usually so intense
that it can be measured with sufficient precision in a relatively short time. On the other
hand, for specimens made up predominantly of heavier elements, such as steel and even
more so for brasses and solders, the intensity of the scattered radiation is very low (see
Fig. 8) and counting the statistical error can preclude precise analysis in a reasonable
amount of time. The most common application where this approach (or a variant
thereof) is used is in the determination of trace elements in specimens of geological
origin. This is illustrated in Figure 9 for the determination of Sr in specimens of widely
Copyright © 2002 Marcel Dekker, Inc.
varying geological origin. In Figure 9a, the net count rate for SrKa is plotted against the
concentration for a large number of specimens. There is a considerable spread around
the calibration line established. The scatter is greatly reduced when the net count rate of
Figure 9 (a) Net count rate of SrKa as a function of Sr concentration for a large number of
specimens of varying geological origin. There is considerable spread around the calibration line.
(b) The ratio of the count rates of the SrKa radiation and the Compton-scattered tube line is plotted
against the concentration of Sr. The spread of the data points around the calibration line is now
much reduced compared to (a); this is especially the case for the point labeled A.
2. Internal Standard
In this method, an element is added to each specimen in a fixed proportion to the original
sample. This addition has to be made to the standard samples as well as to the unknowns.
The characteristic radiation of the element added should be similar to the characteristic
radiation of the analyte in terms of absorption and enhancement properties in the matrix
considered. Such an element is called an ‘‘added internal standard’’, or ‘‘internal stan-
dard’’ for short. In practice, the method works equally well if a pure element or a pure
compound is added, or if a solution with the internal standard element is used. If a so-
lution is used, care must be taken that the solution itself does not contain any elements
that are to be analyzed. The composition of the solution used as the additive must be
constant, otherwise it might affect the matrix effect. The intensity of the internal standard
is affected by matrix effects in much the same way as the intensity of the analyte, provided
there are no absorption edges (leading to difference in absorption) or characteristic lines
including scattered tube lines (leading to difference due to enhancement) between the two
wavelengths considered. Because
Wi ¼ K i I i Mi ð62Þ
for the analyte i and
Ws ¼ Ks Is Ms ð63Þ
for the internal standard s, the following ratio can be obtained by dividing Equation (62)
by Equation (63):
Ii
¼ Kis Wi ð64Þ
Is
where
K s Ms
Kis ¼ ð65Þ
K i Mi W s
Because the same amount of internal standard is added to all specimens, Ws is essentially a
constant and can be included in the constant Kis. It should be noted that Mi (and Ms) is
not a constant over the concentration range of interest (otherwise linear calibration would
suffice) but depends on the matrix elements. However, if both Mi and Ms vary in a similar
manner with the matrix elements, the ratio Mi=Ms is less sensitive to variation in the
matrix effect and, in practice, can be considered a constant. In practice, the constant Kis is
determined using linear regression. The main advantage of the internal standard method
over the scattered-radiation method is its ability to correct effectively for enhancement as
well as for absorption. It also corrects—at least partially—for variations in density of
pressed specimens. The requirement that the intensity of the characteristic radiation of
both the analyte and the internal standard element vary in the same manner with the
Copyright © 2002 Marcel Dekker, Inc.
matrix effects imposes that there should be no absorption edges and no characteristic
radiation from other elements between the measured line of the analyte and that of the
internal standard. Furthermore, ideally, the analyte should not enhance the internal
standard or vice versa. If Ka radiation is measured and if the atomic number of the analyte
is Z (with Z > 23), then, very often, the elements with atomic number Z 1 or Z þ 1 are
very good candidates. This assures that there are no K absorption edges and no K emission
lines of other elements between the two elements considered. The element with atomic
number Z is not enhanced by Ka radiation from an element with Z þ 1, but only by the
much weaker Kb radiation, whereas for elements with atomic number Z þ 2 and higher,
both Ka and Kb contribute to enhancement. In practice, some enhancement between the
internal standard element and the analyte or vice versa is allowed, as the concentration of
the internal standard is constant and the ratio is based on intensities. The absence of major
elements in the specimens with L absorption edges and emission lines, however, must be
checked for. The situations that must be avoided are (1) the case where a major line of a
major matrix element is between the absorption edges of the analyte and the internal
standard and (2) the case where a major absorption edge of a major matrix element is
situated between the measured characteristic lines of the analyte and the internal standard.
In the first case, the matrix element would enhance either the analyte or the internal
standard element, but not both; in the second case, the matrix element absorbs strongly
either the radiation from the analyte or the internal standard, but not both. In both of
these cases, varying concentrations of the matrix element will lead to variable and different
effects on the intensities of the analyte and of the internal standard, and the ratio used in
Eq. (64) will not compensate for such events.
The method, however, has some important limitations:
The specimen preparation is made more complicated and is more susceptible to
errors.
The addition of reagents and the requirement of homogeneity of the specimen tends
to limit the practical application of the method to the analysis of liquids and
fused specimens, although it sometimes finds application in the analysis of
pressed powders.
Although the rule Z þ 1 or Z 1 can serve as a rule of thumb, it is quite clear that for
samples where many elements are to be quantified, a suitable internal standard
cannot be found for every analyte element. Sometimes, more elements are used
in one internal standard solution to provide suitable internal standards for
more analytes.
Also, the fact that the internal standard method is easier to apply to liquids can
generate some problems. Heavier elements (e.g., Mo) are more difficult to
determine using this method, because liquid specimens are generally not of
infinite thickness for the K wavelengths of these heavier elements. In such cases,
the L line can be used, with an appropriate internal standard. The method will,
however, also provide some compensation for the effects of noninfinite
thickness, especially if the wavelength of the internal standard selected is very
similar to that of the analyte line.
Theoretically, L lines of a given element can be used as internal standards for K lines
of other elements and vice versa if these wavelengths are reasonably close to each other
and neither interfering lines nor edges occur between them. In principle, the method allows
the determination of one or two elements in a specimen without requiring analysis (or
knowledge) of the complete matrix.
Copyright © 2002 Marcel Dekker, Inc.
The range of concentration over which this method is suitable can be quite large, up
to 10–20 % in favorable situations, but the internal standard technique is most effective at
low concentrations (or high dilutions). The method finds, for instance, application in the
determination of Ni and V in petroleum products, where MnKa is used as an internal
standard (ISO, 1995). The internal standard method allows an accurate determination of
these elements in a much wider variety of petroleum products than the method based on
linear calibration.
Figure 10 Standard addition method. The net intensity is plotted versus weight fraction of the
element added to the sample and a ‘‘best-fit’’ line is determined. The intercept of that line with the
concentration axis is Wi.
4. Dilution Methods
Dilution methods can also eliminate or reduce the variation of the matrix effect, rather than
compensating for such variation. The dilution method can be explained using Eq. (17):
mi ðl0 Þ
Cðl0 ; li Þ ¼ Pn Pn ð17Þ
j¼1 Wj mj ðl0 Þ þ G j¼1 Wj mj ðli Þ
where ms ðl0 Þ and ms ðli Þ are the mass-attenuation coefficients of the specimen for the pri-
mary wavelength l0 and analyte wavelength li, respectively. Apparently, deviations from
linearity are due to variations in ms ðl0 Þ and=or ms ðli Þ. Enhancement is ignored at this
stage. If one adds, to the sample, D grams of a diluent (d ) for each gram of sample, the
denominator or Eq. (67) becomes
1 D
½m ðl0 Þ þ Gms ðli Þ þ ½m ðl0 Þ þ Gmd ðli Þ ð68Þ
1þD s 1þD d
If the term D=ð1 þ DÞ½md ðl0 Þ þ Gmd ðli Þ is much larger than 1=ð1 þ DÞ½ms ðl0 Þ þ Gms ðli Þ,
the factor Cðl0 ; li Þ becomes essentially a constant and variations due to varying matrix
effects between samples become negligible. This can be done in two ways.
(a) Making D=ð1 þ DÞ large by diluting each sample by adding a large, known
amount of a diluent.
Copyright © 2002 Marcel Dekker, Inc.
(b) Adding a smaller quantity of diluent than in the previous case, but with a much
larger value for md ðl0 Þ þ Gmd ðli Þ. This is called the technique of the heavy
absorber.
Both of these procedures, however, require the addition of reagents to the sample. This
can easily be done for dissolved samples, either in liquids or fused samples, but it is more
difficult for powdered samples (homogeneity!).
These methods do not eliminate the matrix effects completely, but reduce their
influence. On the other hand, they also reduce the line intensity of the analyte; thus, a
compromise must be sought.
Dilution methods also have the advantage of reducing the enhancement effect if one
uses a nonfluorescing diluent (e.g., H2O or Li2B4O7). In this case, the effect is reduced by
the fact that the concentrations of both the enhancing element and the analyte are reduced.
If the diluent contains elements whose characteristic x-rays can excite the analyte, as well
as some other matrix elements, then the contribution of those unknown quantities of
matrix elements to the total enhancement is reduced: The enhancement of the analyte by
the diluent would then be determining and can be considered to be constant.
This method allows the determination of all measurable elements in the sample, as
opposed to the standard addition method, where an addition must be made for each
element of interest.
D. Mathematical Methods
1. General
The term ‘‘mathematical methods’’ refers to those methods that calculate rather than
eliminate or measure the matrix effect.
Mathematical methods are independent of the specimen preparation in the sense that
specimen preparation is taken into account if the composition of the specimen presented to
the spectrometer has been changed (e.g., by fusion), but mathematical methods do not
prescribe the specimen preparation method as is done, for example, by the standard ad-
dition method. The actual calculation method used to convert intensities to concentrations
does not affect the choice of the specimen preparation method. The aim of the specimen
preparation is limited to the presentation to the spectrometer of a specimen that is
homogeneous (with respect to the XRF technique) and that has a well-defined, flat surface
representative for the bulk of the specimen. Mathematical methods usually require
knowledge of all elements in the standard specimens and allow determination of all mea-
surable elements in the unknowns. In practice, trace elements can be neglected in the cal-
culations for the analytes present at higher concentrations, as these trace compounds are
neither subject to an important (and variable) matrix effect nor do they contribute sig-
nificantly to the matrix effect of other elements. Their concentrations are often found by
straightforward linear regression. The mathematical methods are divided in two main ca-
tegories: the fundamental parameter method and the methods using influence coefficients.
where Pi ðlÞ is the contribution of the primary fluorescence caused by incident photons
with wavelength l, and Sij ðl; lj Þ is the contribution of the secondary fluorescence (en-
hancement) by characteristics photons lj which have been excited by primary photons l.
The summation in Eq. (16) or (69) is over all elements j that have characteristic lines that
can excite the analyte i. For each of these elements j, all characteristic lines must be
considered. This is quite simple if none of the L lines or M lines of element j can excite the
analyte. In that case, only the Ka and Kb lines are to be considered. If the L lines of an
element j are energetic enough for enhancement of the analyte, the sheer number of L lines
(e.g., W has more than 20 characteristic L lines that can be considered) would make the
calculation very time-consuming. Therefore, most programs consider only three to five L
lines for each element. A similar reasoning holds for the M lines.
Application of these formulas was originally limited to the prediction of intensities
for specimens with given composition. The application of those formulas for analysis was
not pursued until the 1960s. The method of analyzing specimens by fundamental para-
meter equations has been developed independently around the same time by Criss and
Birks (1968) and by Shiraiwa and Fujino (1966). Due to the large amount of calculations
involved (especially the integration of the incident spectrum and the calculation of the
contribution of enhancement), these programs initially ran on mainframes and mini
computers. A PC version of the same program was proposed by Criss in 1980 (Criss,
1980a).
The application of a fundamental parameter method for analyzing specimens con-
sists of two steps: calibration and analysis. Both steps will be discussed in more detail in
the following subsections.
b. Calibration
The fundamental parameter equation is used to predict the intensity of characteristic lines
for a composition identical to that of the standard used. If more than one standard spe-
cimen is used, the calculations are repeated for each of the standards. The calculations are
performed using the appropriate geometry (i.e., incidence and take-off angles are taken in
agreement with those of the spectrometer used) and the parameters determining the tube
spectrum (such as the anode material, voltage, thickness of beryllium window, and so on)
correspond to the ones used in the spectrometer for the measurements. The intensities
predicted are (almost always) net intensities, void of background, line overlap, crystal
fluorescence, and so on. Hence, the measured intensities must be corrected for such
spectral artifacts. These theoretically predicted intensities are then linked to the actually
measured ones. If only one standard specimen is used, the ratio between the measured
intensity and the calculated intensity is calculated. If more than one standard is used, the
net intensities obtained from the measurements are plotted versus the calculated intensities
and a straight line can be determined for each characteristic line measured. The slope of
Copyright © 2002 Marcel Dekker, Inc.
such a line is the proportionality factor between predicted (calculated) and measured in-
tensities. In general, this relationship will be determined more accurately if more standards
are used.
A special case of calibration ensues when one uses pure elements as standards. Di-
viding each of the measured (net) intensities of the corresponding pure element gives the
relative intensity. This relative intensity can be calculated directly by some fundamental
parameter programs. In fact, some programs use equations that express the intensity of
characteristic radiation directly in terms of relative intensity. The relative intensity in this
respect is thus defined as the intensity of the sample, divided by the intensity of the cor-
responding pure element (or compound, if the concentration of the analyte is defined in
compound concentration), under identical conditions for excitation and detection.
Basically, the calibration function, which is determined using the measured in-
tensities of the standards and the calculated intensities, accounts for instrument-related
factors only; matrix effects are accounted for by using the physical theory as described in
the fundamental parameter equation.
The instrument-related parameters for a wavelength-dispersive spectrometer are as
follows:
Collimation
Crystal reflectivity
Efficiency of detector(s)
Fraction of emergent beam, allowed into the detectors, after Bragg reflection (is also
dependent on Bragg angle)
For energy-dispersive spectrometers, the instrument-related parameters are collimation
and detector efficiency. The effect of the windows of the detector, the dead layer, and so
forth can also be taken into account.
c. Analysis
Step 1. For every unknown specimen, a first estimate of the composition is made.
There are several ways to obtain such a first estimate. They vary from using a simple, fixed
composition (e.g., equal to 100%, divided by the number of elements considered; in the
first estimate, all the concentrations of each of the elements are thus taken equal to one
another) to the composition derived from the measured intensities in combination with the
calibration curves. Using the calibration data, it is possible to estimate for each element
the intensity that would be obtained if the pure elements were measured. These numbers
are then used to divide the intensity of the corresponding element, measured in the un-
known specimen. The resulting fractions are scaled to 100% and used as the first estimate.
Step 2. For this estimate of composition, the theoretical intensities are calculated.
These are converted to measured intensities, using the calibration data, so that these two
sets of intensities for the same specimen can be compared.
Step 3. The next estimate of composition is obtained based on the difference between
the measured and calculated intensities. Again, there are different methods available:
1. The simplest method is based on linear interpolation. If, for a given element, the
measured intensity is 10% higher than the calculated intensity, the concentration
of that element is increased by 10%.
2. Rather than a linear relationship, some authors (Criss and Birks, 1968) use an
interpolation based on three points. This is done because the relationship
between concentration and intensity is usually non linear over a wider range.
If the specimen is a pseudobinary, hyperbolic relationships have proven to be
Copyright © 2002 Marcel Dekker, Inc.
better approximations. For more complex specimens, it still works out quite
well, because the concentrations of the other elements are considered fixed at this
stage. The hyperbolic equation requires a minimum of three points for its
parameters to be determined. The points requiring the least additional
calculation time are the following:
The origin (net intensity zero, at concentration zero).
The pure element, W ¼ 100%; the corresponding intensity has already been
calculated.
The point W ¼ current estimate, the intensity has already been calculated.
These three points allow an hyperbolic relation to be established around the
current estimated composition. From this curve and the measured
intensity, the new concentration estimate is derived. This approach is
repeated for every analyte element. This method usually provides a faster
convergence than using the simple straight line.
3. It is also possible to use gradient methods to determine the next composition
estimate. The formula for the first derivative, with respect to concentration, of
the fundamental parameter equation have been published (Shiraiwa and Fujino,
1968), but the calculation is cumbersome and time-consuming. Also, the
derivatives can be obtained by a finite-difference method, where the effect of a
small change in composition on the intensity is observed.
Step 4. The process, starting at Step 2, is now repeated until convergence is obtained.
Different convergence criteria exist. The calculation can be terminated if one of the fol-
lowing criteria is satisfied for all the elements (compounds) concerned:
1. The intensities, calculated in Step 2, do not change from one step to another, by
more than a present level.
2. The intensities, calculated in Step 2 agree, to within a preset level, with the
measured intensities
3. The compositions, calculated in Step 3, do not change from one step to another
by more than a present level (e.g., 0.0005 or 0.0001 by weight fraction).
One or more of these criteria might be incorporated in the program. These criteria,
however, are no guarantee that the final result is accurate to within the level, specified in
the convergence criteria. Furthermore, especially with convergence criteria based on
concentrations (such as criteria 3), it must be realized that a convergence criterion of
0.0005 is unacceptable when determining elements at levels below 0.0005.
d. Extensions to the Method
More complex scenarios are possible. The most common one includes the calculation
of a set of influence coefficients (based on theoretical calculations) to obtain a com-
position quickly, close to the final result. Next, the fundamental parameter method is
applied (Rousseau, 1984a). This should yield faster convergence in terms of compu-
tation time, because the calculation by influence factors of the preliminary composition
is very fast. This method reduces the number of evaluations of the fundamental param-
eter equation.
Also, the different programs available differ quite markedly in their treatment of
the intensities of the standards measured. Some programs use a weighting of the stan-
dards, stressing the standard(s) closest (in terms of intensity) to the unknown (Criss,
Copyright © 2002 Marcel Dekker, Inc.
1980a). Such programs use a different calibration for each unknown specimen. The
unknown specimen dictates which standards will be given a high weighting factor and
which standards will be used with less weighting. Other programs use all standards with
equal weighting.
e. Typical Results
Early results of the fundamental parameter method on stainless steels are given by Criss
and Birks (1968). The average relative differences between x-ray results and certified values
were about 3–4%. Later, Criss et al. (1978) reported accuracies of about 1.5% relative for
stainless steels, using a more accurate fundamental parameter program. Typical results for
tool steel alloys are given in Table 3. Often, the fundamental parameter method is con-
sidered to be less accurate than an influence coefficient algorithm. This is caused primarily
by the fact that the fundamental parameter method has extensively been used and
described as a method that allows quantitative analysis with only a few standards. This
is obviously an advantage, but it does not imply that fundamental parameter methods
cannot be used in combination with many standards similar to the unknown. As a matter
of fact, on several occasions and with a variety of matrices, the authors have obtained
results of analysis with an accuracy similar to that of influence coefficient algorithms when
using the same standards in both cases.
f. Factors Affecting Accuracy
The accuracy of the final results is determined by the following:
The measurement
The specimen preparation
The physical constants used in the fundamental parameter equation
The limited description of the physical processes that are considered in the
fundamental parameter equations
The standard and the calibration
In the following discussion, the effect of measurements and specimen preparation
will not be considered.
Table 3 Analysis of Seven Tool Steels with a Fundamental Parameter Program [XRF11, from
CRISS SOFTWARE, Largo, MD (Criss, 1980)]
Element Minimum conc. (%) Maximum conc. (%) Standard deviation (%)
Incidence and exit angles. The incidence angle in most wavelength-dispersive (WD) and
energy-dispersive (ED) spectrometers is, in fact, defined by a relatively wide cone with a different
intensity at the boundaries compared to the center. This incident cone is neglected and the
incident radiation is considered parallel, along a single, fixed direction. A similar observation
holds for the exit angle. The effect is far less pronounced if diffraction from a plane crystal
surface is used for dispersion, as is done, for example, in most sequential WD spectrometers.
This has been studied to some extent by Müller (1972). To our knowledge, none of the fun-
damental parameter programs available takes this effect into account. Its influence, however, is,
to some extent, compensated for by calibration with standard specimens.
Spectrum of incident beam. The spectrum of the incident beam from an x-ray tube
spectrum requires more attention. Parts of the primary spectrum might excite an element B
that, in turn, excites element A very efficiently. In such cases, this enhancement may make
the intensity of element A sensitive to small errors in the tube spectrum representation,
which would not be compensated for if the pure A was used for calibration. This can arise,
for example, in the analysis of silica–zirconia specimens with a Rh tube (Criss, 1980b).
Pure silica is relatively insensitive to the intensity of the characteristic K lines of Rh. In
combination with Zr, however, the situation is different. Indeed, the RhK lines are strongly
absorbed by Zr. Zr then emits K and L lines that enhance Si. As a result, the Si intensity is
more sensitive to the RhK lines in SiO2–ZrO2 mixtures than it is in pure SiO2. Tube spectra
have been calculated using, for example, the algorithm of Pella et al. (1985).
Mass-attenuation coefficients. There are several compilation of mass-attenuation
coefficients, published in the literature. A continuing effort to compile the most compre-
hensive table has been undertaken by the National Institute of Standards and Technology
(formerly National Bureau of Standards), Gaitherburg, MD.
When selecting a table of mass-attenuation coefficients for use in a fundamental
parameter program, the following question must be addressed: Does the table cover all the
analytical needs? (In practice, does it cover the complete range of interest from the longest
wavelength considered, down to the excitation potential of the tube?)
The analyst should be aware that the use of formulas to generate mass-attenuation
coefficients can lead to values that can be significantly different from the corresponding
table values.
Presently, for applications in XRF, the complications of McMaster et al. (1969),
Heinrich (1966), Leroux and Thinh (1977), or Veigele (1974) are most often used. A short
discussion on the agreement between some of these compilations has been presented by
Vrebos and Pella (1988). A more recent compilation has been published by de Boer (1989).
Fluorescence yields. A comprehensive reference to fluorescence yields, including
Coster–Kronig transitions, can be found in the work of Bambynek et al. (1972). (see
also Chapter 1, and Appendix VI).
Copyright © 2002 Marcel Dekker, Inc.
Absorption jump ratios. These can be derived from the tables of attenuation coef-
ficients.
Ratios of different fluorescent lines within a family. Data for the K spectra can be
found in the work by Venugopalo Rao et al. (1972) (see also Chapter 1).
Wavelengths of absorption edges and emission lines. A comprehensive table was
published by Bearden (1967) and is also presented in the appendices to Chapter 1. Because
attenuation coefficients are wavelength dependent, an error in a wavelength of any char-
acteristic line will automatically lead to a bias in the corresponding attenuation coefficients.
h. Limited Physical Processes Considered
The fundamental parameter equation [Eq. (16)] does not consider all physical processes in
the specimens. Three of the most obvious that are missing are described here.
Tertiary fluorescence. Although the formula for tertiary fluorescence has been de-
rived by, for example, Shiraiwa and Fujino (1966) and Pollai and Ebel (1971), it is not
included in most fundamental parameter programs. Usually, the tertiary fluorescence effect
is considered small enough to be negligible. Shiraiwa and Fujino (1967, 1974) have pre-
sented data showing a maximum contribution of tertiary fluorescence of about 3% relative
to the total intensity of Cr in Fe–Cr–Ni specimens. Therefore, even in Fe–Cr–Ni specimens
whose characteristic lines and absorption edges are ideally positioned relative to one an-
other to favor enhancement, the effect of tertiary fluorescence is quite limited. Higher-order
enhancement is also possible, but it is even less pronounced than tertiary fluorescence.
Scatter. Other processes not considered in most of the fundamental parameter
methods are coherent and incoherent scatter of both the primary spectrum and the
fluorescent lines. This is usually justified by pointing out that the photoelectric effect is, by
far, the major contribution to the total absorption. It is believed that the contribution by
scattered photons to the excitation of characteristic photons is negligible. However, in
some cases the scattered primary spectrum may have a considerable influence, as illu-
strated earlier in this chapter. The equations describing the contribution of scatter to
fluorescent intensity have been derived by Pollai et al. (1971). These equations have ob-
viously many similarities to those for secondary fluorescence.
Photoelectrons. The processes that are probably the must unknown in the funda-
mental parameter method are related to the contributions of the photoelectrons and of the
Auger electrons that are produced as a result of absorption of the primary and fluorescent
x-ray photons. These electrons have sufficient energy to excite other atoms and thus create
additional fluorescence. This is especially important in the case of low-atomic-number
elements, as has been described by Mantler (1993) and has been illustrated, for example,
by Kaufmann et al. (1994).
i. Standards and Calibration
The use of good standards (similar to the unknown) will almost always lead to more
accurate results, compared to a situation where the standards used have a widely different
composition from that of the unknown. This is because most of the uncertainties, caused
by inaccuracies in the physical constants, cancel. The degree of similarity between stan-
dards and unknown has an important effect on the accuracy of the analysis.
for a multielement specimen, with n being the total number of elements or compounds. In
most of the influence coefficient algorithms, one element is eliminated from the summation
{i.e., one influence coefficient is used when dealing with binaries, as in Eq. (70), and n1
coefficients deal with a specimen consisting of n compounds [Eq. (71)]}. In Eq. (71) this is
explicitly indicated by the j 6¼ e under the summation sign. The eliminated compound e can
be any of the ones present in the specimens; however, most authors eliminate the analyte.
The expression for Mi is then used as follows:
2 3
6 X
n 7
Wi ¼ R i 6
4 1 þ mij Wj 7
5 ð72Þ
j¼1
j6¼e
which links the relative intensity and the influence coefficients to the composition of the
specimen. The relative intensity Ri for a given analyte is defined as the ratio of the net
measured intensity Ii in the specimen and the intensity that would have been measured on
the pure analyte IðiÞ under identical conditions:
Ii
Ri ¼ ð73Þ
IðiÞ
In practice, the relative intensity is often derived indirectly from measurements on stan-
dards, and the pure element (or compound) is not required. Equation (56) can be rewritten
in terms of Ri :
Wi ¼ R i Mi ð74Þ
Ii M i
IðiÞ ¼ ð77Þ
Wi
where Mi is calculated and Ii is measured. Mi can be calculated using Eq. (75), where Ri is
calculated from theory for the standard specimen of known composition. Therefore, al-
though many influence coefficient algorithms are presented using the format of Eq. (72),
involving the relative intensity, there is no real need to perform measurements of the pure
analyte, as Eq. (72) can be written as
2 3
6 X
n 7
Wi ¼ K i I i 6
41 þ mij Wij 7
5 ð78Þ
j¼1
j6¼e
where Ki is then determined during the calibration phase. Furthermore, if the background
is not subtracted, Eq. (78) can be written as
Copyright © 2002 Marcel Dekker, Inc.
2 3
6 X
n 7
Wi ¼ Bi þ Ki Ii 6
41 þ mij Wij 7
5 ð79Þ
j¼1
j6¼e
where the summation covers all n elements (or compounds) in the specimen, except the
analyte itself. Hence, there are n1 terms in the summation. This is common to all cur-
rently used algorithms. Most of the algorithms, developed earlier [such as Sherman’s
(1953) and Beattie and Brissey’s (1954)], used to have n terms, rather than n1, for spe-
cimens with n elements.
Equations (80a), (80b), and (80c) are linear equations in the concentrations of the
elements WA , WB , and WC , respectively. Note that there are only two coefficients for each
analyte element. Consider, for example, the first equation of the set 80, namely Eq. (80a):
Element A is the analyte and its concentration is equal to the relative intensity RA , mul-
tiplied by the matrix correction factor (1 þ aAB WB þ aAC WC ). This matrix correction
factor has only two coefficients: one (aAB ) to describe the effect of element B on the in-
tensity of A and, similarly, one to describe the effect of element C on the intensity of A.
The value of the coefficient aAA , which would correct for the effect of A on its own in-
tensity (sometimes—but incorrectly—referred to as self-absorption) is zero. Similarly, aBB
and aCC are also zero. The effect of A on A, however, is taken into account, as will be
shown in the next subsection.
Calculation of the coefficients. Lachance and Traill also showed that the influence
coefficients, ij can be calculated for monochromatic excitation by photons with wave-
length 0 (assuming absorption only) from the expression
mj ðl0 Þ cscðc0 Þ þ mj ðli Þ cscðc00 Þ
aij ¼ 1 ð82Þ
mi ðl0 Þ cscðc0 Þ þ mi ðli Þ cscðc00 Þ
Wi ¼ Ri ½1 þ aij Wj ð83Þ
to
ðWi =Ri Þ 1
aij ¼ ð84Þ
Wj
yields an expression that can be used to obtain aij , based on the composition of the
binary and the relative intensity Ri . The drawbacks associated with this method are as
follows:
1. The calculation of Ri requires the measurements of the intensity on the pure i
(element or compound). This could lead to large errors if the intensity of i in the
binary is much lower than that of the pure, due to, for example, nonlinearity of
the detectors.
2. The pure elements (or compounds) are not always easily available or could be
unsuitable to present to the spectrometer (e.g., pure Na or Tl).
3. Equation (84) is very prone to error propagation when Wi is close to 1. The
numerator is then a difference between two quantities of similar magnitude, and
the denominator is then close to zero, magnifying the errors.
4. Also, the availability of suitable binary specimens can present problems: Some
alloys tend to segregate and homogeneous specimens are then difficult to
obtain.
The coefficients aij can also be calculated from theory: Calculate Ri for the binary
with composition ðWi ; Wj Þ rather than obtain it from measurements and substitute in Eq.
(84). This method eliminates drawbacks 1, 2, and 4. However a better method—without
the problem associated to error propagation—is to use Eq. (75) directly with the values for
Ri and Wi . Lachance (1988) has also presented methods to calculate the values of the
coefficients from theory.
Copyright © 2002 Marcel Dekker, Inc.
The Lachance–Traill algorithm assumes the following:
1. The influence coefficients can be treated as constants, independent of
concentration; this limits the concentration range in cases where the matrix
effects change considerably with composition.
2. The influence coefficients are invariant to the presence and nature of other
matrix elements. So aFeCr , determined for use in Fe–Cr–Ni ternary specimens, is
the same as aFeCr in Fe–Cr–Mo–W–Ta or Fe–Cr specimens.
b. The de Jongh Algorithm
Formulation. In 1973, de Jongh proposed an influence coefficient algorithm (de
Jongh, 1973), based on fundamental parameter calculations. The general formulation of
his equation is
" #
X
n
Wi ¼ E i R i 1 þ aij Wj ð85Þ
j¼1
j6¼e
Note that in order to obtain the concentration of elements A and B, the concentration of
C, WC , is not required. This is different from Lachance and Traill’s algorithm: In order to
calculate the concentration of A, using Eq. (80a), the concentrations of both B and C are
required. If the user is not really interested in element C (e.g., element C is iron in stainless
steels), Eq. (86c) need not be considered and the analysis of the ternary specimen can be
done by measuring RA and RB and solving Eqs. (86a) and (86b).
Calculation of the coefficients. de Jongh also presented a method to calculate the
coefficients from theory. The basis is an approximation of Wi =Ri by a Taylor series
around an ‘‘average composition’’:
Wi
¼ Ei þ di1 DW1 þ di2 DW2 þ þ din DWn ð87Þ
Ri
where Ei is a constant given by
Wi
Ei ¼ ð88Þ
Ri average
and
DWi ¼ Wi Wi;average ð89Þ
dij are the partial derivatives of Wi =Ri with respect to concentration:
@ðWi =Ri Þ
dij ¼ ð90Þ
@Wj
or
DWe ¼ DW1 DW2 DWn ð93Þ
one element e can be eliminated. The resulting equation is similar to Eq. (87), but has only
n1 terms:
Wi
¼ Ei þ bi1 DW1 þ bi2 DW2 þ þ bin DWn ð94Þ
Ri
with
bi1 ¼ di1 die ð95Þ
for all bij except bie , which is equal to zero. Equation (94) has n1 terms, but they are still
in DW, rather than W. Transformation of DW to W is done by substituting Eq. (89) in
Eq. (94):
Wi X n X n
¼ Ei bij Wj;average þ bij Wj ð96Þ
Ri j6¼e j6¼e
indicating that the weight fraction of the analyte is calculated from its relative intensity, a
matrix correction term, and the matrix correction term for the average composition (which
for a given composition is a constant). The value for Mi,average can be calculated using the
influence coefficients calculated and taking the composition (Wi, Wj) equal to the average
Copyright © 2002 Marcel Dekker, Inc.
Table 4 Analysis of Stainless Steels with Theoretical Influence Coefficients (de Jongh)
composition. This equation is very similar at first sight to the Lachance–Traill equation
[Eq. (81)], except for the term Mi,average.
Typical results. Tables with de Jongh’s coefficients have been used for a wide
variety of materials, including high-temperature alloys, brass, solders, cements, glasses,
and so forth. An example for stainless steels is given in Table 4. Results are shown for
Mn, Ni, and Cr only. For these analytes, the matrix effects are the most important.
Other elements, such as Si, P, S, and C, are also present at trace level. The coefficients are
calculated at a given composition [see Eq. (91)]. The practical range of concentration over
which these coefficients yield accurate results varies from 5% to 15% in alloys to the whole
range from 0% to 100% in fused oxide specimens.
Comparison between the algorithms of de Jongh and Lachance–Traill. The following
points can be noted:
1. The basis of the de Jongh algorithm is a Taylor series expansion, around an
average (or reference) composition. The values of the coefficients calculated
depend on this composition.
2. De Jongh can eliminate any element; Lachance and Traill eliminate the analyte
itself: aij is zero. De Jongh eliminates (i.e., fixes the coefficient to zero) the same
element for all analytes. Eliminating the base material (e.g., iron in steels) or the
loss on ignition (for beads) generally leads to smaller numerical values for the
coefficients and avoids the necessity to determine all elements.
3. De Jongh’s coefficients are calculated at a given reference composition. They are
composition dependent and take into account all elements present. A coefficient
aij represents the effect of element j on the element i in the presence of all other
elements: They are multielement coefficients rather than binary coefficients. This
is seen in Table 5, where the values of the coefficients aCrCr and aCrNi are shown
for several different specimens, but with identical concentrations for the analyte
Table 5 Values for Influence Coefficients aCrCr and aCrNi, Calculated According to the Algorithm
of de Jongh, for Selected Specimens
where the summation over j has n1 terms (all n elements, except i), and the summation
over k has (n2)=2 terms (all n elements, except the analyte i and element j; furthermore, if
aAj k is used, then aAkj is not). For a binary specimen, Eq. (99) reduces to
WA ¼ RA ½1 þ a0AB WB ð100Þ
with
a0AB ¼ aAB þ aABB WB ð101Þ
clearly showing that the influence coefficient a0AB varies linearly with composition (i.e.,
WB ). For binaries, WA ¼ 1 WB ; hence, Eq. (101) can also be rearranged to
a0AB ¼ aAB þ aABA WA ð102Þ
Equations (101) and (102) are, at least theoretically, identical. It has been shown, however,
that Eq. (102) is preferable to Eq. (101) if specimens with more than two elements (or
compounds) are analyzed (Lachance and Claisse, 1980). This will be discussed in more
detail in Sec. V.D.6. Note that the value of aAB in Eq. (101) is different from its value in
Eq. (102).
Cross-product coefficients. For a ternary specimen, the Claisse–Quintin algorithm
can be written
WA ¼ RA ½1 þ aAB WB þ aABB W2B þ aAC WC þ aACC W2C þ aABC WB WC ð103Þ
The terms
aAB WB þ aABB W2B
and
aAC WC þ aACC W2C
where only one coefficient is used for each interfering element. The coefficients Aij are used
for cases where absorption is the dominant effect. In this case, the coefficient Bik is taken
equal to zero. If, for a given analyte, all Bik coefficients are zero, Eq. (104) reduces to the
Lachance–Traill expression. When enhancement by element k dominates, a Bik coefficient
is used. The corresponding Aij coefficient is then taken equal to zero. Hence, the total
number of terms in both summations is n 1.
The correction factor for enhancement by element k can be rewritten as
Bik
aik ¼ ð105Þ
1 þ Wi
showing that aik varies with concentration in a nonlinear fashion. The algorithm is very
popular when analyzing stainless steels and steels in general.
Among the disadvantages of the Rasberry–Heinrich algorithm are the following:
aij2 Wm
mij ffi aBij ¼ aij1 þ ð106Þ
1 þ aij3 ð1 Wm Þ
with
Wm ¼ 1 Wi ð107Þ
Wm is the concentration of all matrix elements. It has been shown by Lachance and Claisse
(1980), as well as by Tertian (1976), that variable binary coefficients must be expressed in
terms of Wm (or 1Wi). For binary specimens, Eq. (106) can be rewritten using Wj for Wm
and Wi for (1Wm):
aij2 Wj
mij ffi aBij ¼ aij1 þ ð108Þ
1 þ aij3 Wi
For specimens with more than two compounds, however, the difference between Eqs. (106)
and (108) becomes clear. The value for the influence coefficient mij is approximated over
the complete concentration range for the binary by the function in Eq. (106), which relies
on three coefficients only. The excellent agreement between the true influence coefficient
mij and the approximation of Eq. (106) is shown in Figure 12 for Fe in FeNi (severe
enhancement) and for Fe in FeCr (pronounced absorption).
For multielement specimens, cross-product coeffcients aijk are used to correct for the
crossed effect, similar to Eq. (99). The general equation for a multielement specimen is
" #
X n X n X n
aij2 Wm
Wi ¼ R i 1 þ aij1 þ Wj þ aijk Wj Wk ð109Þ
j6¼i
1 þ aij3 ð1 Wm Þ j6¼i k6¼i;k>j
where the summation over j has n 1 terms (all n elements, except i) and the summation
over k has (n 2)=2 terms (all n elements, except the analyte i and element j; furthermore if
aijk is used, then aikj is not).
Vrebos and Helsen (1986) have published some data on this algorithm, clearly
showing the accuracy of the algorithm, using theoretically calculated intensities. The use of
Copyright © 2002 Marcel Dekker, Inc.
Table 6 Composition of the Specimens Used for the Calculations of the
Coefficients for Lachance’s Three-Coefficient Algorithm, in Weight Fraction
Specimen No. Wi Wj Wk
theoretically calculated intensities has the advantage that it avoids errors due to specimen
preparation and measurement errors associated with actual measured data. Pella and co-
workers (1986) have presented a comparison of the algorithm with several others and with
a fundamental parameter method using experimental data.
Calculation of the coefficients. The coefficients aij1 , aij2 , and aij3 are calculated using
fundamental parameters at three binaries i j. The cross-product coefficients are calculated
from a ternary. The compositions of the specimens concerned are listed in Table 6. The
‘‘specimens’’ referred to in Table 6 are hypothetical specimens. The intensities are calcu-
lated from fundamental parameters and require no actual measurements on real specimens.
Step 1. Calculate the relative intensity Ri for the first composition in Table 6. If the
analysis of interest has more than three elements, then the system is divided in combi-
nations of three elements i, j, k at a time. The analyte is element i, and j and k are two
interfering elements. If the system considered is with compound phases, such as oxides,
then the compositions in Table 6 are assumed to be for the oxides.
Step 2. Using Eq. (84), the corresponding influence coefficient aBij can be calculated.
Step 3. For this composition, Wm ¼ 1 Wi ¼ Wj ¼ 0:001, which is small enough to
be considered zero. Hence, Eq. (106) reduces to
aBij ¼ aij1 ð110Þ
aBij has been calculated in Step 2, so aij1 can be computed.
Step 4. Calculate the intensity for the second composition of Table 6 and use Eq. (84)
to calculate aBij . In most cases, this value will be different from the one found in Step 2
because the compositions involved are different.
Step 5. 1 Wm ¼ Wi ¼ 0:001 is small enough to be considered zero; hence, Eq. (106)
reduces to
aBij ¼ aij1 þ aij2 ð111Þ
aij1 and aBij are known so aij2 can be calculated.
Step 6. Calculate the intensity for the third composition of Table 6 and use Eq. (84)
to calculate aBij . In most cases, this value will be different from the one found in Step 2 or 4
because the compositions involved are different.
Step 7. Using Wm ¼ 1 Wi ¼ 0:5 ¼ Wi , Eq. (106) reduces to
aij2 ð0:5Þ
aBij ¼ aij1 þ ð112Þ
1 þ aij3 ð0:5Þ
SiKa(FeKa) SiKa(FeLa)
mation is shown in Figure 13 for the FeNi and the FeCr binaries. The agreement for the
straight line is obviously not as good as with the COLA algorithm, especially in those cases
where the value of the true influence coefficient varies markedly, as is the case for Fe in
FeCr (absorption). Equation (117) has been compared to the three-coefficient algorithm of
Lachance by Vrebos and Helsen (1986). They show that the accuracy is somewhat less
than for Lachance’s method, but for most practical purposes, the Rousseau algorithm
should give acceptable results.
Calculation of the coefficients. Rousseau has shown that the fundamental para-
meter equation can be rearranged to
" #
Xn
Wi ¼ R i 1 þ aij Wj ð118Þ
j6¼1
and he also proposed a method to calculate the a coefficients directly from fundamental
parameters, without calculating the intensity first (Rousseau, 1984a). As a matter of fact,
Rousseau first calculates the coefficients for a given composition and then calculates the
intensity, using Eq. (118). The coefficients in Eq. (116) are calculated in a way very similar
to the method described in Sec. V.D.5.d. The compositions involved are given in Table 8.
The ‘‘specimens’’ referred to in Table 8 are hypothetical specimens. The intensity is cal-
culated from fundamental parameters and requires no actual measurements on real spe-
cimens. For the first two binaries of Table 8, the influence coefficient is calculated [symbol
aij(0.20,0.80) and aij(0.80,0.20), respectively]. Then the corresponding values are sub-
stituted in Eq. (117):
aij ð0:20; 0:80Þ ¼ aij1 þ aij2 ð0:80Þ ð119aÞ
the relative intensity of FeKa as a function of the weight fraction of Fe in FeNiCr spe-
cimens. There is considerable spread of the intensity of FeKa, even for a constant weight
fraction of Fe. For specimens with a weight fraction of 0.10 Fe, the relative intensity of
FeKa varies between 0.036 and 0.16 (points marked 1 and 2 in Fig. 14). This is due to the
Copyright © 2002 Marcel Dekker, Inc.
Table 8 Composition of the Specimens Used for the Calculations of the
Coefficients for the Linear Approximation According to Rousseau’s Algorithm,
in Weight Fraction
Specimen No. Wi Wj Wk
rather different effect that Ni and Cr have on Fe:Cr is an absorber for FeKa radiation,
whereas the NiK radiation can enhance FeK radiation through the process of secondary
fluorescence (enhancement). For these specimens, the matrix effect MFe can be calculated
from Eq. (75). The total matrix effect on Fe, MFe(FeNiCr), in these specimens, at a fixed
Fe concentration of 0.1, for example, varies from 0.63 (for 0.1 Fe in FeNi, point 2) to 2.8
(for 0.1 Fe in FeCr, point 1).
Now, the problem is how to calculate the matrix effect in this case, based on in-
fluence coefficients. Assume a specimen with the following composition: WFe ¼ 0.1,
Figure 14 The relative intensity of FeKa as a function of the concentration of Fe, in the presence
of Ni and Cr. For every given weight fraction of Fe, the highest value of the intensity is obtained for
the binary system FeNi (enhancement), whereas the lowest intensity is for the binary FeCr
(absorption). The intermediate values are for ternary specimens, where the concentrations of Ni and
Cr vary in steps of 0.1 weight fraction. At a weight fraction of Fe ¼ 0.7, the four data points labeled
a, b, c, and d represent the following specimens (WFe, WNi, WCr): a ¼ (0.7, 0.0, 0.3), b ¼ (0.7, 0.1,
0.2), c ¼ (0.7, 0.2, 0.1), and d ¼ (0.7, 0.3, 0.0). Points labeled 1 and 2: see text. Experimental
conditions: W tube at 45 kV, 1-mm Be window, incidence and take-off angles 63 and 33 ,
respectively.
ij ¼ aij þ tijk Wk
aM ð126Þ
B
where tijk is a coefficient expressing the effect of element k on the influence coefficient aM
ij .
Similarly,
ik ¼ aik þ tikj Wj
aM ð127Þ
B
and is based on cross-product coefficients to correct for crossed effect introduced by the
use of binary influence coefficients. The use of the cross-product coefficients is not
mandated by the concentration range to be covered (the binary coefficients as calculated
by, for example, the algorithm of Lachance are more than adequate) but is a consequence
of the use of binary coefficients.
7. Application
In Secs.V.D.4 and V.D.5, several influence coefficient algorithms have been discussed.
Application of the resulting equations for calibration and analysis will be discussed here
and is equally valid for any of the influence coefficient algorithms.
a. Calibration
Step 1. It is assumed that the coefficients have been calculated from theory, for
example, using Eq. (84) or (97).
Copyright © 2002 Marcel Dekker, Inc.
Step 2. Calculate the matrix correction term [the square brackets in Eq. (81), (85),
(99), (104), and (109)] for all standard specimens and for a given analyte. The coefficients
are known (Step 1), and for standard specimens, all weight fractions Wi and Wj are
known.
Step 3. Plot the measured intensity of the analyte, multiplied by the corresponding
matrix correction term against analyte weight fraction. Then, determine the ‘‘best’’ line,
Wi ¼ Bi þ Ki Ii ½1 þ ð132Þ
by minimizing DWi (see Sec. V.A). Note that Eq. (132) is more general than Eq. (50),
which does not correct for matrix effects. This process is repeated for all analytes. Other
methods are also feasible. The most common variant is the one where
W
¼ Bi þ Ki Ii ð133Þ
½1 þ
is used. This is nearly equivalent to Eq. (132) but with brackets:
Wi ¼ ðBi þ Ki Ii Þ½1 þ ð134Þ
The term ðBi þ Ki Ii Þ is related directly to the relative intensity Ri. Corrections for line
overlap should only affect this term.
b. Analysis
For each of the analytes, a set of equations has to be solved for the unknown Wi, Wj and
so forth. If the matrix correction term used is the one according to Lachance and Trail
[Eq. (81)] or de Jongh [Eq. (85)], then the set of equations can be solved algebraically (n
linear equations with n unknowns for Lachance and Traill and n1 equations with n1
unknowns for de Jongh). Mostly, however, an iterative method is used. As a first estimate,
one can simply take the matrix correction term equal to 1. This yields a first estimate of the
composition Wi, Wj, and so on. This first estimate is used to calculate the matrix cor-
rection terms for all analytes. Subsequently, a new composition estimate can be obtained.
This process is repeated until none of the concentrations changed between subsequent
iterations by more than a preset quantity.
If the matrix correction is done using algorithms which use more than one coefficient
{e.g., Claisse and Quintin [Eq. (99)] or Rasberry and Heinrich [Eq. (104)]}, then the
equations are not linear in the unknown concentrations and an algebraic solution is not
possible. An iterative method, such as described earlier can be used.
where aij represents the influence coefficient of element j on the analyte i and ti is the time
(in s) required to accumulate a preset number of counts. The constants aij are determined
from measurements on specimens with known composition. Determination of the com-
position of an unknown involves the solving of the above set of linear equations
[Eq. (135)]. This set, however, is homogeneous: Its constant terms are all equal to zero. So,
only ratios among the unknown Wi can be obtained. In order to obtain the weight frac-
tions Wi, an extra equation is required. Sherman proposed using the sum of all the weight
fractions of all the elements (or components) in the specimen, which ideally, should be
equal to unity. For a ternary specimen,
WA þ WB þ WC ¼ 1 ð136Þ
Using Eq. (136), one of the equations in the set of Eqs. (135) can be eliminated. The
solution obtained, however, is not unique: For a ternary, any one of the three equations
can be eliminated. This yields three different combinations. Furthermore, any of the
three elements can be eliminated in each of the combinations. Hence, a total of 363 ¼ 9
different sets can be derived from Eqs. (135) and (136), and each of these sets will
generate different results. In general, the algorithm yields n2 different results for a system
with n elements or compounds. This is clearly undesirable, because it is hard to de-
termine which set will give the most accurate results. Another disadvantage is the fact
that the sum of the elements determined always equals unity, even if the most abundant
element has been neglected. Furthermore, the numerical values of the coefficients de-
pend, among other parameters such as geometry and excitation conditions, also on the
number of counts accumulated. Nonquantifiable parameters, such as reflectivity of the
diffracting crystal used in wavelength-dispersive spectrometers, or tube contamination
will also affect the value of the coefficients. The coefficients determined on a given
spectrometer cannot be used with another instrument; they are not transferable.
The other algorithms discussed use some form of a ratio method: The Lachance and
Traill algorithm, for example, uses relative intensities. The measurements are then done,
relative to a monitor; this reduces, or eliminates, the effect of such nonquantifiable
parameters.
Copyright © 2002 Marcel Dekker, Inc.
b. The Algorithm of Lucas-Tooth and Price
Lucas-Tooth and Price (1961) developed a correction algorithm, where the matrix effect
was corrected for, using intensity (rather than concentration) of the interfering elements.
The equation can be written as
" #
X n
Wi ¼ Bi þ Ii k0 þ kij Ij ð137Þ
j6¼i
where Bi is a background term and k0 and kij are the correction coefficients. A total of
n þ 1 coefficients have to be determined, requiring at least n þ 1 standards. Usually,
however, a much larger number of standards is used. The coefficients are then determined
by, for example, a least-squares method. The corrections for the effect of the matrix on the
analyte are done via the intensities of the interfering elements; their concentrations are not
required. The method assumes that the calibration curves of the interfering elements
themselves are all linear; the correction is done using intensities rather than concentra-
tions. The algorithm will, therefore, have a limited range. Its use will be limited to ap-
plications where only one or two elements are to be analyzed (it still involves
measurements of all interfering element intensities) and where a computer of limited
capabilities is used (although calculation of the coefficients involves much more compute
capabilities than the subsequent routine analysis of unknowns).
The advantages of the method are as follows:
The method is very fast, because the calculation of composition of the unknowns
requires no iteration.
Analysis of only one element is possible; this requires, however, the determination of
all relevant correction factors.
Very simple algorithm, requiring very little calculation.
c. Algorithms Based on Concentrations
Algorithms similar to Eq. (137) have been proposed, using corrections based on con-
centrations rather than intensities. The values of the coefficients were then to be derived
from multiple-regression analysis on a large suite of standards. The main aim was to
obtain correction factors that could be determined on one spectrometer and used, without
alteration, on another instrument. In practice, the coefficients still have to be adjusted
because of the intimate and inseparable entanglement of spectrometer-dependent factors
with matrix effects. Furthermore, compared to the algorithms based on intensities, some of
the advantages of the latter are not retained: A calibration for all elements present is now
required, calculation of the composition of unknowns requires iteration, and so forth.
In principle, methods based on theoretically calculated influence coefficients are re-
commended.
VI. CONCLUSION
Among the advantages of XRF analysis are the facts that the method is nondestructive
and allows direct analysis involving little or no specimen preparation. Analysis of major
and minor constituents requires correction for matrix effects of variable (from one spe-
cimen to another) magnitude. If the matrix varies appreciably from one specimen to the
next, then even the intensity of elements present at a trace level can be subject to matrix
effects and a correction is required. Several methods for matrix correction have been
Copyright © 2002 Marcel Dekker, Inc.
described. Each of these methods have their own advantages and disadvantages. These, by
themselves, do not generally lead to the selection of ‘‘best’’ method. The choice of the
method to use is also determined by the particular application.
From the previous sections, it may appear that the mathematical methods are more
powerful than the compensation methods. Yet, if only one or two elements at a trace level
in liquids have to be determined, compensation methods (either standard addition or the
use of an internal standard) can turn out to be better suited than, for examples, rigorous
fundamental parameter calculations. Compensation methods will correct for the effect of
an unknown, but constant, matrix. Also, they do not require the analysis of all con-
stituents in the specimen. The mathematical methods (fundamental parameters as well as
methods based on theoretical influence coefficients), on the other hand can handle cases in
which the matrix effect is more variable from one specimen to another. In this respect, they
appear to be more flexible than the compensation methods, but they do require more
knowledge of the complete matrix. All elements contributing significantly to the matrix
effect must be quantified (either by x-ray measurement or by another technique) even if the
determination of their concentrations is not required by the person who submits the
sample to the analyst. Once a particular algorithm is selected, it is customary to use for all
analytes. However, it must be stressed that this is not a requirement. There is only one
requirement for adequate matrix correction: Each analyte should be corrected adequately,
by whatever method.
If complete analysis (covering all major elements) is required, the analyst has the
choice between the fundamental parameter method and algorithms, based on influence
coefficients. Commonly, fundamental parameter methods are (or were) used in research
environments rather than for routine analysis in industry. This choice is more often made
on considerations such as the availability of the programs and computers than on dif-
ferences in analytical capabilities. Influence coefficient algorithms tend to be used in
combination with more standards compared to fundamental parameter methods, because
their structure and simple mathematical representation facilitates interpretation of the
data (establishing a relationship between concentration and intensity, corrected for matrix
effect). The final choice, however, has to be made by the analyst.
REFERENCES
Anderson CH, Mander JE, Leitner JW. Adv X-Ray Anal 17:214, 1974.
Australian Standard 2563-1982, Wavelength Dispersive X-ray Fluorescence Spectrometers—
Methods of Test for Determination of Precision. North Sydney, NSW: Standards Association
of Australia, 1982.
Bambynek W, Crasemann B, Fink RW, Freund HU, Mark H, Swift CD, Price RE, Venugopala
Rao P. Rev Mod Phys 44:716, 1972.
Bearden JA. Rev Mod Phys 39:78, 1967.
Beattie HJ, Brissey RM. Anal Chem 26:980, 1954.
Bonetto RD, Riveros JA. X-Ray Spectrom 14:2, 1985.
Claisse F, Quintin M. Can Spectrosc 12:129, 1967.
Criss JW. Adv X-Ray Anal 23:93, 1980a.
Criss JW. Adv X-Ray Anal 23:111, 1980b.
Criss JW, Birks LS. Anal Chem 40:1080, 1968.
Criss JW, Birks LS, Gilfrich JV. Anal Chem 50:33, 1978.
de Boer DKG. Spectrochim Acta 44B:1171, 1989.
de Jongh WK. X-Ray Spectrom 2:151, 1973.
Bertin EP. Principles and Practice of X-Ray Spectrometric Analysis. 2nd ed. New York: Plenum
Press, 1975.
Jenkins R, Gould RW, Gedcke D. Quantitative X-Ray Spectrometry. 2nd ed. New York: Marcel
Dekker, 1995.
Tertian R, Claisse F. Principles of Quantitative X-Ray Fluorescence Analysis. New York: Wiley,
1982.
Andrzej A. Markowicz
Vienna, Austria
René E. Van Grieken
University of Antwerp, Antwerp, Belgium
I. INTRODUCTION
In the absence of enhancement effects and assuming monochromatic excitation, the mass
per unit area of the element i; mi , for homogeneous intermediate-thickness samples, can be
calculated from [see Eq. (90) in Chapter 1]:
Ii ðEi Þ
mi ¼ Abcorr ð2Þ
Bi
where
GI0 ðE0 Þ"ðEi Þi0 ðE0 Þ!i pi ð1 1=ji Þ
Bi ¼ ð3Þ
sin 1
and Abcorr is the absorption correction factor given by
½ðE0 Þ csc 1 þ ðEi Þ csc 2 m
Abcorr ¼ ð4Þ
1 expf½ðE0 Þ csc 1 þ ðEi Þ csc 2 mg
[The symbols used in Eqs. (2)–(4) are explained in Chapter 1, Eqs. (88) and (89).]
The value of the constant Bi (sometimes called the sensitivity factor) can be de-
termined either experimentally as the slope of the straight calibration line for the ith
element obtained with thin homogeneous samples or semiempirically based on both the
experimentally determined GI0 ðE0 Þ value and the relevant fundamental parameters (Yap
et al., 1987; Markowicz et al., 1992a).
The absorption correction factor Abcorr represents the combined attenuation of
the primary and fluorescent radiations in the whole specimen and can be determined
individually for each sample by transmission experiments (Leroux and Mahmud, 1996;
Giauque et al., 1973). These are done by measuring the x-ray intensities with and without
the specimen from a thick multielement target located at a position adjacent to the back of
the specimen, as shown in Figure 1.
If ðIi ÞS ; ðIi ÞT ; and ðIi Þ0 are the intensities after background correction from the
sample alone, from the sample plus target, and from the target alone, respectively, then the
combined fraction of the exciting and fluorescent radiations transmitted through the total
sample thickness is expressed by
ðIi ÞT ðIi ÞS
expf½ðE0 Þ csc 1 þ ðEi Þ csc 2 mg ¼ H ð5Þ
ðIi Þ0
After a simple transformation, Eq. (2) can be rewritten as
Figure 1 Experimental procedure used in the emission–transmission method for the correction of
matrix absorption effects.
Figure 2 Curves of [m(E0) csc 1 þ (Ei) csc 2]m and mm(Ei) csc 2 values versus fluorescence
x-ray energy for an NBS SRM 1632 coal specimen. (From Giauque et al., 1979. Reprinted with
permission from Analytical Chemistry. Copyright American Chemical Society.)
where ðWCa Þn and ðWCa Þn1 are the values of the weight fractions for Ca obtained in the
subsequent iterations.
Copyright © 2002 Marcel Dekker, Inc.
Figure 3 Graphical explanation of the idea for the modified version of the E–T method: see text for
details. (From Markowicz and Haselberger, 1992. Reprinted with permission of Pergamon Press Ltd.)
A calculation procedure applied for the Ca absorption edge can easily be repeated
for any other absorption discontinuity at a lower energy and, finally, a total absorption
curve ln ln A versus ln E is obtained.
The importance of the additional correction for the discontinuities in the absorption
properties of the samples to be analyzed was demonstrated, among others, for the analysis
of coal samples. Neglecting the absorption edges correction for Fe and Ca at the con-
centration level of around 1% resulted in considerable errors of the analysis, up to 60%
for the determination of Ca (Markowicz and Haselberger, 1992).
Van Dyck et al. (1980) developed a correction method that allows calculations of the
absorption coefficients (and absorption correction factors as well) at any energy for in-
termediate-thickness samples, without additional measurements, by using the ratio of the
x-ray signals from a Zr wire positioned in front of the sample and from a Pd foil placed
behind the sample, both in a fixed geometry, as shown in Figure 4. The Zr wire provides an
external reference signal ðZrKÞ; which is applied for normalization of all measured
fluorescent intensities to reduce considerably the effect of variations in exciting x-ray tube
intensity and of dead-time losses.
The coefficients for higher energies are calculated with an iterative program from the
experimentally measured absorption coefficient at the PdL energy (2.9 keV), ðEPd Þ. In the
first step, the total attenuation coefficient at the PdL energy, caused exclusively by the low-
Z elements (e.g., Z < 17) in the sample that show no characteristic peak above 3.0 keV in
the spectrum, is calculated from the normalized measured intensities, ðIPd ÞT and ðIPd Þ0 and
Copyright © 2002 Marcel Dekker, Inc.
Figure 4 Measurement geometry applied in an automatic absorption correction method. (From
Van Dyck et al., 1980. Reprinted by permission of John Wiley & Sons, Ltd.)
the different characteristic peaks recorded in the spectrum. This attenuation coefficient due
to low-Z elements, low-Z ðEPd Þ, is given by (Van Dyck et al., 1980)
low-Z ðEPd Þ
Pn 0
ln½ðIPd Þ0 =ðIPd ÞT =m csc 2 ðE0 Þ csc 1 =csc 2 j¼1 Wj j ðEPd Þ
¼ P0 ð9Þ
1 nj¼1 Wj
where n0 is the number of characteristic peak in the XRF spectrum and j ðEPd Þ is the mass
attenuation coefficient for the PdL energy in the element j (McMaster et al., 1969), giving
rise to a characteristic peak in the XRF spectrum.
The weight fraction of the element j in the sample, Wj , is correlated with the recorded
characteristic peak through the sensitivity factor, routinely obtained by measuring thin
single or multielement standards. As a first approximation, the characteristic peak in-
tensities are not corrected for absorption. In a second step, based on the low-Z ðEPd Þ value,
the absorption coefficients of the low-Z matrix for other energies, low-Z ðEÞ, are calculated
quantitatively by assuming nearly parallel properties of the logarithmic absorption curves
ln ðEÞ versus ln E.
The total mass attenuation coefficient ðEÞ for the characteristic x-rays of an element
in the whole sample can now be calculated, taking into account the contributions j ðEÞ of
the high-Z elements:
!
Xn0 Xn0
ðEÞ ¼ 1 Wj low-Z ðEÞ þ Wj j ðEÞ ð10Þ
j¼1 j¼1
size effects, and a heterogeneous sample load is presented in the work of Van Dyck et al.
(1980). The grain size and sample heterogeneity effects induce inaccuracies on the ab-
sorption coefficient determinations that may well reach 20% for particulate samples, such
as intermediate-thickness deposits of geological materials. Thus, this approach has the
same limitations as all emission–transmission methods applied to heterogeneous samples.
where Ii ðEi Þ is the absolute uncertainty of the characteristic x-ray intensity Ii ðEi Þ (usually
provided by a computer routine applied for spectrum evaluation), Abcorr is the absolute
uncertainty of the absorption correction factor, and Bi is the absolute uncertainty of the
sensitivity (or calibration) factor Bi (usually provided by a computer calibration routine,
as a mean percent difference between the experimental and fitted values for the calibration
factors). The total relative uncertainty of the mass per unit area mi =mi is calculated from
" #1=2
mi Ii ðEi Þ 2 Abcorr 2 Bi 2
¼ þ þ ð12Þ
mi Ii ðEi Þ Abcorr Bi
where ½ðIi Þ0 and ½ðIi ÞT are the absolute uncertainities of the characteristic x-ray
intensities obtained from the target alone and from the sample plus target, respectively.
After a simple transformation, mi =mi can be given by
" #1=2
mi Ii ðEi Þ 2 ðA 1 ln AÞ2 2 Bi 2
¼ þ ðAÞ þ ð13Þ
mi Ii ðEi Þ ðA 1Þ2 A2 ðln AÞ2 Bi
It is obvious that the total relative uncertainty of the weight fraction of the ith element
Wi =Wi is equal to mi =mi .
Equations (13) and (12b) can be used for calculating the total uncertainities of
the emission–transmission method as well as the contribution from various sources of
the uncertainties ½Ii ðEi Þ; Abcorr ; Bi . It has been demonstrated that in the de-
termination of trace elements, the dominant contribution to the total uncertainty is
from the uncertainty of the peak-area calculation for the characteristic x-rays of the
element of interest. In this case, the contribution from the uncertainties of the ab-
sorption correction factor and sensitivity can practically be neglected. In the determi-
nation of major elements, the largest contribution to the total uncertainty is from the
uncertainty of the absorption correction factor and=or sensitivity; in this case, the
contribution of the uncertainty of the peak-area calculation can safely be neglected.
Such a detailed analysis of the contribution of various sources of uncertainties can
be useful when there is a need to identify the most critical point of the whole
analytical procedure and some reduction of the total uncertainty of the analysis is
required.
Another factor which influences the accuracy of the emission–transmission
method is the heterogeneity of the sample loading. Markowicz and Abdunnabi (1991)
derived general expressions describing the accuracy of the E–T method for various
types of sample loading heterogeneities (including incompletely loaded samples) within
a wide range of sample thicknesses and characteristics x-ray energies. The results,
confirmed by experiments, clearly show that inhomogeneities of sample loading may
critically affect the accuracy of the E–T method, in particular in the analysis of in-
termediate-thickness samples with strong inhomogeneities and large values for the
ðE0 Þðcsc 1 Þm and ðEi Þðcsc 2 Þm factors. The most critical influence of sample loading
heterogeneity on the accuracy of the analytical results is observed for incompletely
loaded samples of intermediate thickness. The method proposed in the work of
Markowicz and Abdunnabi (1991) gives a possibility for evaluating the errors arising
from some types of sample loading heterogeneities as well as for evaluating the range
of applicability of the simplest version of the E–T method (Markowicz et al., 1992b) if
a certain value for the total uncertainty Wi =Wi is accepted.
Copyright © 2002 Marcel Dekker, Inc.
III. ABSORPTION CORRECTION METHODS VIA SCATTERED PRIMARY
RADIATION
The use of scattered primary radiation in XRF analysis provides an alternative to the
common problem of matching standards of similar composition to samples to be analyzed.
The backscatter peaks are sometimes treated as fluorescent peaks from internal standards
because they suffer matrix absorption similar to that of fluorescent peaks and behave si-
milarly with instrumental variations. They also provide the only direct spectral measure of
the total or average matrix of geological, biological, or other materials containing large
quantities of light elements, such as carbon, nitrogen, and oxygen, usually not observed by
their characteristic x-ray peaks.
X-ray fluorescence matrix correction methods based on the use of scattered radiation
have mostly been applied in quantitative analysis of infinitely thick samples under a wide
variety of experimental conditions, including discrete and continuum primary radiation
sources and detection by both wavelength- and energy-dispersive systems (Andermann
and Kemp, 1958; Kalman and Heller, 1962; Taylor and Andermann, 1971, 1973; Leo-
nardo and Saitta, 1977; Livingstone, 1982; Kikkert, 1983; Markowicz, 1984). The scat-
tered-radiation methods utilize the incoherent (Compton) and=or coherent (Rayleigh)
scatter peaks from line excitation sources or the intense high-energy region from con-
tinuum sources.
where k0 is a constant for a given measurement geometry and detection efficiency, Com ðE0 Þ
is the Compton mass-scattering coefficient of the sample material for the primary radiation
of energy E0 ðcm2 =gÞ; ðECom Þ is the total mass-attenuation coefficient of the sample
material for Compton-scattered primary radiation of energy ECom ðcm2 =gÞ [see Eq. (52) in
Chapter 1]. Equation (14) is valid for monochromatic excitation.
Assuming that the relation of the atomic number to the mass number is constant for
every element to be found in the sample and that ðE0 Þ ffi ðECom Þ, the following sim-
plified formula for the intensity of Compton-scattered radiation from a multielement
sample can be obtained:
where ai ; ail ; and bi are constant coefficients obtained experimentally on the basis of
standard samples and n0 is the number of the elements to be determined, including the ith
element. As is seen from Eq. (16), the absorption matrix correction is carried out via the
intensities of characteristic x-rays of all elements determined and the intensity ICom, re-
flecting, for the most part, the variations in the composition of light matrix. To apply the
absorption matrix correction, Eq. (16), the total mass per unit area of the specimen must
be evaluated, for example, by sample weighing. This fluorescent–Compton correction
method can be used in the XRF analysis of homogeneous intermediate-thickness samples
in a limited range of mass per unit area (m < 10mthin).
where k00 is a constant for a given measurement geometry and detection efficiency for the
primary x-rays of energy E0 , and coh ðE0 Þ is coherent mass-scattering coefficient of the
sample material for the primary radiation (cm2=g) [see Eq. (73) in Chapter 1].
Bazan and Bonner (1976) showed, for the first time, a linear relation between the
effective absorption coefficient (defined as the sum of the sample absorption coefficients for
exciting and characteristic x-rays) and the ratio of incoherent to coherent scattering.
Copyright © 2002 Marcel Dekker, Inc.
However, the coefficients of the calibration line varied somewhat with the matrix, and this
hampered practical applications of this simple approach.
Markowicz (1979) found that, theoretically the sensitivity of the absorption cor-
rection via the incoherent=coherent scattered x-ray intensities ratio is better than that of
the absorption procedure involving each of the scattered radiations individually. For in-
termediate-thickness samples, in a limited range of rather small values of mass per unit
area, the intensities of the Compton-scattered radiation ICom and the coherent scattered
radiation Icoh are different functions of the total mass-attenuation coefficient of the
incident radiation ðE0 Þ; the intensity ICom is a linearly decreasing function and the
intensity Icoh appears to be a linearly increasing function of the ðE0 Þ. For a limited
range of ðE0 Þ values, the following simple expression can be used to evaluate ðE0 Þ
(Markowicz, 1979):
ðE0 Þ ¼ C1 þ C2 m þ C3 r þ C4 mr ð18Þ
where
ICom
r¼
Icoh
and C1–C4 are constants calculated by the least-squares fit on the basis of experimental
results for standard samples.
The values of the total mass-attenuation coefficient of the fluorescent radiation in a
whole sample, ðEi Þ, is obtained from the simple dependence of the ðEi Þ=ðE0 Þ ratio on
the values ðE0 Þ, calculated separately for each element to be determined. Finally, sample
weighing provides the value of mass per unit area, m, and the calculation of the absorption
correction factor Abcorr via Eq. (4) can be simply performed if, of course, the values of the
effective angles 1 and 2 are evaluated experimentally or theoretically. The applicability
of the proposed matrix correction method (Markowicz, 1979) involving both incoherent
and coherent scattered primary radiations is limited to XRF analysis of intermediate-
thickness samples of mass per unit area smaller than about 10mthin.
A backscattered fundamental parameters (BFP) method for quantitative XRF
analysis of intermediate samples of variable composition and thickness was developed by
Nielson (1977). The method utilizes thin-film multielement calibration of the spectrometer
and mathematical matrix correction in which the samples are modeled as a composite of
heavy elements, which are quantified through their characteristic radiation, and light
elements, estimated through the coherent and incoherent x-ray scatter peaks. Figure 6
schematically illustrates the basis for analyzing the heavy elements (Z > 13) and the light
elements (H, C, N, O, Na, and others), which must be estimated by the difference from the
scattered x-ray peaks. The BFP method utilizes coherently and incoherently scattered
x-rays to identify and estimate the quantities of two light elements representative of the
light-element portion of sample matrix. The quantities of the two light elements a and b
are estimated by solving for Wa and Wb in the simultaneous equations
X
n0
gIcoh Wj cohj ðE0 Þ ¼ Wa coha ðE0 Þ þ Wb cohb ðE0 Þ ð19Þ
j¼1
and
X
n0
hICom Wj Comj ðE0 Þ ¼ Wa Coma ðE0 Þ þ Wb Comb ðE0 Þ ð20Þ
j¼1
The heavy- and light-element concentrations are used in computing the absorption
correction factor Abcorr [Eq. (4)] and the enhancement correction factor [Eq. (91) in
Chapter 1]. Because the concentrations and corrections are interdependent, all calculations
are carried out by iteration [for more details see (Nielson, 1977)].
To improve the sensitivity of the determination of low-Z elements, Sanders et al.
(1983) extended the previously described BFP method. The new method utilizes the co-
herent and incoherent backscatter intensities to compute matrix corrections (Nielson and
Sanders, 1982) from the combined results of two separate energy-dispersive XRF
(EDXRF) data from different (TiK and ZrK) excitation sources. The Ti-excited spec-
trum allows a more sensitive determination of elements in the Al–Ca range.
The coherent=incoherent scatter ratio is also applied in an absorption correction
procedure developed by Van Dyck and Van Grieken (1980) for monochromatic x-ray
excitation. In this method, coherent and incoherent scattered radiations are used to cal-
culate, first, the effective mass of the sample and, second, the absorption coefficients for
x-rays of interest and, hence, the absorption correction factors. The effective thickness is
the sample thickness weighted at every point for the excitation-detection efficiency, in the
same way as the measured characteristic radiation is weighted.
Copyright © 2002 Marcel Dekker, Inc.
Assuming that the major elements of the sample do not differ too greatly in atomic
number, the effective thickness meff can be calculated from (Van Espen et al., 1979)
ICom
meff ¼ ð22Þ
fb0 Abcorr ðECom Þ½b0 Icoh Abcorr ðECom Þ=a0 ICom Abcorr ðEcoh Þb1 gða1 b1 Þ1
X
n
Scoh ¼ a0 Wj Zaj 1
j¼1
and
X
n
SCom ¼ b0 Wj Zbj 1 ð23Þ
j¼1
respectively. Abcorr(ECom ) and Abcorr(Ecoh ) are the absorption correction factors for the
incoherent and coherent scatter radiation, respectively, as defined in Eq. (4).
A reasonably accurate effective mass is obtained by modeling the sample as a
composite of high-Z elements, calculated from their characteristics peaks using Eq. (2),
and of a light matrix with mass per unit area, mlow-Z , evaluated from the coherent and
incoherent scatter peaks after subtraction of the high-Z element contribution. The method
for the determination of the effective thickness allows the analysis of samples of hetero-
geneous thickness and irregular shape.
More recently, Araújo et al. (1990) have developed a very similar procedure for
effective sample mass assessment of intermediate thickness samples via the coherently and
incoherently scattered radiation, as a first step in the automated matrix effect evaluation,
for the case of filtered polychromatic continuum excitation with a Rh tube. Validation
with geological standards and deposited slurries on Mylar foils and Nuclepore filters gave
very satisfactory results.
In the method proposed by Van Dyck and Van Grieken (1980), calculation of the
mass-attenuation coefficient for x-rays of interest is preceded by an evaluation of the
mass-attenuation coefficient m (2.956 keV) at the ArKa energy (2.956 keV). This energy is
preferred because it is at the lower end of the energy range that can safely be used in
conventional EDXRF analysis and because, when working under vacuum, it is situated
in a peak-free part of an XRF spectrum. The value of m (2.956 keV) is derived from the
ratio coherent to incoherent scatter intensities R, based on the relationship (see Fig. 7)
of the calculated mass-attenuation coefficient at 2.956 keV versus the measured R ratio:
energy in Fig. 5). Finally, the mass absorption coefficients for the different x-ray energies E
are calculated by adding the low-Z and high-Z absorption contributions [see the total
absorption cure, Eq. (10) and Fig. 5].
The proposed procedure for the automatic determination of the mass-attenuation
coefficient, based on the coherent=incoherent scatter ratio, has several obvious merits
compared with the emission–transmission absorption correction method. First, there is
no supplementary measurement or work needed [apart from experimentally obtaining
the semiempirical dependence of the (2.956 keV) on the R ratio (Fig. 7) and sensi-
tivity factors Bi , Eq. (2), on the basis of thin standard samples], because the additional
information on a sample composition is present in the spectrum itself. Second, because
the energies of coherently and incoherently scattered primary radiation, from which the
information is extracted, are higher than that of the incident radiation used in trans-
mission experiments, secondary effects (i.e., grain size effects, inhomogeneous thickness
of the sample, and irregular sample surface) are less important. A third positive point
Copyright © 2002 Marcel Dekker, Inc.
of the described procedure (Van Dyck and Van Grieken, 1980) is the independence,
within certain limits, of the analytical results on the mass of the specimen. Also, it is
worth emphasizing the capability of determining reasonably accurate mass absorption
coefficients when the mean atomic number of the sample varies drastically.
Most of the existing correction procedures in XRF analysis of intermediate-thickness
samples ignore the enhancement effect. It appears, however, that for some special cases,
the enhancement effect should be taken into account. Van Dyck et al. (1986) derived
theoretical formulas for secondary fluorescent x-ray intensities in medium-thickness
samples based on the Sherman equations. Their computer routine for enhancement cor-
rections was incorporated into an overall program for evaluation of x-ray spectra and
calculation of x-ray absorption correction factors from scatter peaks.
The methods, based on coherent and incoherent scatter radiation, developed by
Van Dyck and Van Grieken (1980) for overall matrix effect corrections in the case of
monochromatic excitation and by Araújo et al. (1990) for effective sample mass
determination in case of polychromatic radiation have later been expanded by He and
Van Espen (1991) into a general and versatile procedure for quantitative EDXRF with
polychromatic excitation. The comprehensive method uses the scattered radiation to es-
timate the composition and mass of the low-Z element matrix (with optimal use of the
fundamental parameter approach) and also uses the characteristic x-ray peaks to estimate
the concentrations of the higher-Z elements. An iterative process is then executed. Using
the initially estimated sample mass and composition, the weight fraction of each element is
calculated until the calculated composition converges. The absorption and enhancement
corrections are calculated from the latest iteration. Better sample mass, low-Z matrix
composition, and analyte concentrations are computed after this. This overall procedure is
repeated again until subsequent overall iterations do not yield significant differences. The
method is, in principle, applicable to samples of any thickness and composition and to
polychromatic excitation. The procedure has been implemented as part of the popular
software package AXIL-QXAS (Van Espen et al., 1997), which includes spectrum ac-
quisition and spectrum analysis and runs on a PC. Excellent quantitative results have been
obtained with it.
Wegrzynek et al. (1993) have recently developed a direct correction procedure for the
enhancement effect in intermediate thickness samples, which is based on the results of the
emission–transmission measurements and does not require any iterative calculations. The
enhancement term ENHi is calculated for the intermediate-thickness samples from
(Wegrzynek et al., 1993)
1
ENHi ¼ ½ðD þ 1ÞSUM1 þ DðSUM2 þ SUM3Þ ð25Þ
2i ðE0 Þ
with
1
ðE0 Þ ðEi Þ
D ¼ exp þ m 1 ð25aÞ
sin 1 sin 2
X
sin 1 ðE0 Þ=ðEq Þ
SUM1 ¼ i ðEq ÞKjq ðE0 ÞWj ln þ 1
j;q
ðE0 Þ sin 1
sin 2 ðEi Þ=ðEq Þ
þ ln þ 1 ð25bÞ
ðEi Þ sin 2
where pj is the transition probability for a given spectral line in an x-ray series of the
enhancing jth element. Also in this case, the emission–transmission measurements can be
used to support the enhancement effect correction. The enhancement factor is calculated
either for each spectral line of the enhancing element separately or for a weighted average
energy of the whole x-ray series. The differences in the results obtained by using the two
approaches are practically negligible (Markowicz et al., 1992b). Equation (26), although
very simple, can also be applied in order to (1) define the region of sample thickness where
the enhancement effect is negligible (or smaller than a certain predefined level) and (2) to
determine the minimum sample thickness for which the bulk enhancement correction can
already be used.
The fundamentals of the quantitative XRF analysis of granular specimens involve typical
problems of radiation physics (i.e., the interaction of photons with a specimen of finite
size).
where f ðE0 Þ and f ðEi Þ are the linear attenuation coefficients (cm1) for primary and
fluorescent radiation in fluorescent particles, respectively, and ar is the radiometric particle
diameter.
The radiometric diameter, introduced by Claisse and Samson (1962), represents the
mean geometrical path of x-rays through one particle. These and many other authors (Berry
et al., 1969; Lubecki et al., 1968; Hunter and Rhodes, 1972; Rhodes and Hunter, 1972) have
taken ar as simply equal to the volume of the grain divided by the particle area presented to
the radiation, averaged over all possible orientations of the grain. Hence, for spherical
particles, the radiometric diameter is equal to 0.67a (with a ¼ geometric diameter). One can
easily visualize, however, that such an approach assumes equal weighting of the con-
tribution from all possible radiative paths to the average. In view of the different absorption
effects themselves, this is untrue. Markowicz et al. (1980) introduced, instead of the com-
monly used radiometric diameter approach, the concept of an effective absorption-weighted
radiometric diameter for fluorescent radiation, depending on both the geometry and ab-
sorption effects, and provided a comparison of these two approaches for single spherical
particles for two excitation-detection geometries ( and /2). This alternative approach
allows quantitative evaluation of the discrepancies resulting from the concept ar ¼ particle
volume=average area. It was concluded that for the geometry, both approaches give
practically the same results (maximum relative differences amount only to 5%), but for /2
geometry, the radiometric diameter approach can safely be applied only for very small
particles and=or at relatively high energies of primary radiation (Markowicz et al., 1980).
For samples with a certain particle size distribution described by a function fðar Þ, the
heterogeneity factor Fi can be calculated according to the formula (Rhodes and Hunter,
1972)
aZr max
fðar Þð1 exp½ar ðf ðE0 Þ csc 1 þ f ðEi Þ csc 2 ÞÞ dar
Fi ¼ ð29Þ
ðf ðE0 Þ csc 1 þ f ðEi Þ csc 2 Þar
ar min
where ar min and ar max are the smallest and the largest radiometric particle diameters,
respectively, and
dVf
fðar Þ ¼ ð30Þ
ðVf Þt dar
where dVf is the volume of fluorescent particles having a size between ar and ðar þ dar Þ and
ðVf Þt is the total volume of the fluorescent particles.
The theoretical predictions given by Eqs. (28) and (29) have been compared with the
experimental results for the heterogeneity factors obtained for samples with some discrete
particle sizes and various particle size distributions (Holynska and Markowicz, 1981). For
granular samples of copper sulfide, a satisfactory agreement has been obtained; for samples
of iron oxide, some discrepancies due to agglomeration of the particles have been observed.
Copyright © 2002 Marcel Dekker, Inc.
An in-depth study of the influence of sample thickness, excitation energy, and
excitation-detection geometry on the particle size effects in XRF analysis of intermediate-
thickness samples was carried out and the results are presented in the work of Van Dyck
et al. (1985).
deviation Sr ; Figure 10 shows the theoretical relationship of ðKi2 Þmax with Sr for different
values of t0 . The families of curves shown in Figures 9 and 10 allow us to determine the
application limits of the particle size correction method [i.e., to determine ðKi2 Þmax in
various configurations]. In practice, however, it is more interesting to know the maximum
value of the radiometric particle diameter for which the particle size effect can still be
corrected. This can be determined from Eq. (32).
The value of t0opt can be determined from Figure 9 for a given value of the relative standard
deviation Sr . Thus, one can obtain the following inequality, which should be fulfilled by
the parameter t0 :
t0 t0opt ð38Þ
This means that the value of the energy of the exciting radiation E01 of the ‘‘correction’’
source should fulfill the inequality
E01
ðE01 Þopt ð39Þ
where ðE01 Þopt is the energy of the exciting radiation E01 for which t0 ¼ t0opt . Taking into
account that the efficiency of photoexcitation depends on the energy of the primary
Copyright © 2002 Marcel Dekker, Inc.
Figure 10 Theoretical relationship of the maximum value of the particle size correction factor
ðKi2 Þmax with the relative standard deviation Sr for different values of the parameter t0 : ðSr Þ0 is the
maximum acceptable value of the relative standard deviation Sr (see text for details). (From
Markowicz, 1983. Reprinted by permission of John Wiley & Sons, Ltd.)
radiation, the practical conclusion can be drawn that the values of both the parameter t0
and the energy of the exciting radiation E01 should be as close as possible to the values of
t0opt and ðE01 Þopt , respectively.
On the other hand, for a given pair of sources of primary radiation (i.e., for a given
value of the parameter t0 ), it is possible to estimate, using Figure 10, the maximum ac-
ceptable value ðSr Þ0 of the relative standard deviation of the ratio ðIi1 =Ii2 Þrel . In con-
sequence, the appropriate measurement times and=or activities of the sources of primary
radiation can be selected.
Although the correction method may look complicated, it is currently the only
real correction procedure dealing with the particle size effect in XRF analysis of granular
intermediate-thickness specimens. The idea of applying different excitation x-ray energies to
estimate particle size corrections was also exploited by Nielson and Rogers (1986).
Copyright © 2002 Marcel Dekker, Inc.
REFERENCES
Stanislaw Piorek*
Niton Corporation, Billerica, Massachusetts
I. INTRODUCTION
Radioisotope x-ray fluorescence (XRF) and x-ray preferential absorption (XRA) techni-
ques are used extensively for the analysis of materials, covering such diverse applications
as analysis of alloys, coal, environmental samples, paper, waste materials, and metalli-
ferous mineral ores and products (Rhodes, 1971; Rhodes and Rautala, 1983; Watt, 1978
and 1983; Watt and Steffner, 1985; Piorek 1997). Many of these analyses are undertaken in
the harsh environment of industrial plants and in the field. Some are continuous on-line
analyses of material being processed in industry, where instantaneous analysis information
is required for the control of rapidly changing processes.
Radioisotope x-ray analysis systems are often tailored to a specific but limited range
of applications. They are similar and often considerably less expensive than analysis
systems based on x-ray tubes, but these attributes are often gained at the expense of
flexibility of use for a wide range of applications.
Operators making analyses in the field or in industrial plants are usually less skilled
than those working in the laboratory with x-ray tube systems. Manufacturers of radio-
isotope x-ray analysis systems compensate for this by producing simple semiautomated or
fully automated systems whose output, calibrated for the specific application, is given
directly in terms of concentrations of elements required or in terms of a simple pass=fail-
type decision.
Radioisotope x-ray techniques are preferred to x-ray tube techniques when simpli-
city, ruggedness, reliability, and cost of equipment are important, when minimum size,
weight, and power consumption are necessary, when a very constant and predictable x-ray
output is required, when the use of high-energy x-rays is advantageous, and when short
x-ray path lengths are required to minimize the absorption of low-energy x-rays in air.
Also of significant analytical importance is the fact that the radioisotope excitation is
usually monoenergetic (monochromatic) as opposed to polyenergetic (polychromatic)
excitation characteristic for x-ray tubes.
The basic equations for x-ray analysis are given in Chapter 1. Some additional equations
used for XRF, XRA, and XRS analyses are presented here. The typical geometries (Watt
and Steffner, 1985; Jenkins et al., 1981 b) of the radioisotope source, sample, and detector
used are shown in Figures 1 and 2.
Figure 1 The three geometries for radioisotope-excited x-ray fluorescence analysis: (a) annular
source, (b) central source, and (c) side source. (From Jenkins et al., 1981b.)
A. Absorption of X-rays
The intensity, I, of a narrow beam of monoenergetic x- or g-rays transmitted through a
sample (already shown in Chapter 1) is given by
I ¼ I0 emrt ð1Þ
where
X
n
m¼ ðWi mi Þ ð2Þ
i¼1
I0 is the intensity of x-rays detected without the sample. m and t are the mass absorption
coefficient and path length of x-rays in the sample, respectively, r is the bulk density of the
sample, and mi and Wi are the mass absorption coefficient and weight fraction of the ith
element in the sample, respectively.
Equation (1) also holds for broad beams of x-rays when the cross section for pho-
toelectric absorption is much greater than that for Compton and coherent scattering [i.e.,
particularly for low-energy x-rays and high-atomic-number (Z) elements]. This assump-
tion can be verified by reference to Appendices VIVIII of Chapter 1.
Figure 3 Loss in energy of x-rays in Compton scattering at angles 90 , 120 , and 150 .
The I ’s are the x-ray intensities of the elements measured, kij are correction coefficients
determined by multivariable linear least-squares fitting, and Wi is concentration of the
analyte i. There are as many of this equations as there are analytes to be measured.
The L-TP model allows for the calibration of an x-ray analyzer with the suite of
calibration samples without the necessity of knowing assays for any element but the
Copyright © 2002 Marcel Dekker, Inc.
desired analyte. Due to its simplicity and ruggedness, this approach is often used in the
calibration of benchtop and portable x-ray analyzers; the analyzer measures x-ray in-
tensities of the important interfering elements in calibration samples and then develops a
calibration equation for the analyte(s) by multivariable, linear least-squares fitting.
lnðI0 =IÞ=rt mM
Wi ¼ ð11Þ
mi mM
where mi is the mass absorption coefficient of the x-rays in the analyte and mM is the ab-
sorption coefficient of the matrix, given by
X
n
mM ¼ ðmj Wj Þ ð12Þ
j6¼i
and the subscript j refers to the matrix elements and S Wj ¼ 1 Wi . The concentration of
the analyte can thus be determined if the product of the bulk density and thickness of the
sample is known and the mass absorption coefficients of the matrix elements are ap-
proximately equal or the composition of the matrix does not vary.
In practice, XRA analysis usually involves measurements of transmission of narrow
beams of x-ray, at two x-ray energies, through the sample (Watt and Steffner, 1985). This
is called dual-energy (x- or g-ray) transmission (DUET) analysis. The beams are usually
coincident and thus not differently affected by heterogeneity of the sample in the beam
path (Fig. 2).
From Eq. (11), the concentration Wi of the desired analyte is
ðm0M Rm00M Þ
Wi ¼ ð13Þ
ðm0i m0M Þ Rðm00i m00M Þ
where
lnðI0 =IÞ0
R¼ ð14Þ
lnðI0 =IÞ00
and
m0M þ ðm0i m0M ÞWi m0
R¼ ¼ ð15Þ
m00M þ ðm00i m00M ÞWi m00
The prime and double prime refer to the first and second x-ray energies, respectively. The
concentration is thus determined independently of the density and thickness of the sample
through which the coincident x-ray beams pass.
Copyright © 2002 Marcel Dekker, Inc.
The sensitivity of analysis is high when m0i
m0M and when rt is large. The analysis is
accurate when the ratio m0M =m00M is constant, independently of variations in composition of
the sample matrix. This ratio is approximately constant when the x-ray energies of the
transmitted x-ray are just above and below the K shell absorption edge energy of the
analyte and at higher energies when, at each x-ray energy, the mass absorption coefficients
of all matrix elements are about the same. In the latter case, the energy of the higher-
energy x-ray is usually chosen so that the mass absorption coefficients of the analyte and
matrix elements are the same, the transmission measurement thus determining the mass
per unit area of sample in the x-ray beam.
The uncertainty in determination of R [Eq. (14)] caused by counting statistics is
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 2
R ðdI=IÞ0 ðdI=IÞ00
dR ¼ þ ð16Þ
rt m0 m00
where
0 2 0 2 0 2
dI dI dI
¼ 00 þ 0 ð17Þ
I I0 I
and
00 2 00 2 00 2
dI dI0 dI
¼ 00 þ 00 ð18Þ
I I0 I
and where dI=I is the relative counting statistical uncertainty and m is the mass absorption
coefficient of the x-ray in the sample. The corresponding uncertainty in determining the
concentration of the analyte can be found by substituting R þ dR for R in Eq. (13).
The uncertainty in the determination of the concentration of the analyte, caused by
an increase in the concentration (dC) of one matrix element k replacing another matrix
element l, can be calculated by increasing the mass absorption coefficient of the sample
matrix by
mM ½new ¼ mM ½old þ ðmk ml ÞdC ð19Þ
and substituting the new mass absorption coefficient into Eq. (13).
These equations accurately predict all aspects of XRA analysis (Watt and Steffner,
1985) except when the sample is so highly heterogeneous that within the beam of x-rays,
there are significant differences in absorption of the x-rays.
The characteristics of radioisotope x-ray sources and detectors are described here. A full
understanding of the different characteristics of scintillation, proportional, and solid-state
detectors is essential because of the need to tailor their use to specific applications and to
environmental conditions in the field and in industrial plants.
A. Radioisotope Sources
There are only a few radioisotope sources that are used frequently for x-ray analysis; these are
listed with their most important characteristics in Table 1. Also included are two radio-
isotopes that emit high-energy x-rays and they are used most frequently with the x-ray
sources to correct for changes in sample mass per unit area, thickness, or bulk density.
The activity of radioisotopes is specified in terms of the rate of disintegration of the
radioactive atoms [i.e., decays per second, or becquerel (Bq)]. One becquerel (Bq), an SI
unit, is defined as one disintegration per second. The unit of becquerel replaces the non-SI
unit, the curie (Ci), which equals 3.761010 becquerel. Unfortunately, the unit of becquerel
is not very practical. For example, a typical, useful activity of 100 mCi has to be expressed
in gigabecquerels (GBq). A practical conversion relation between the two units is
100 mCi ¼ 3:7 GBq
The typical number of x- or g-rays emitted per disintegration by the given radioisotope is
listed in Table 1 so that the essential parameter of the radioisotope source, the number of
x-rays or g-rays emitted per second, can be calculated. The emission rate of radioisotope
decreases with time according to the law of natural decay, the number of radioisotope
atoms decaying from N0 to N after an elapsed time t being given by
N ¼ N0 e0:693t=T1=2 ð20Þ
where T1=2 is the so-called half-life of the radioisotope. The source decays to half of its
original emission rate during the time equal to its half-life. The radioisotope source is
usually replaced after one to two half-lives.
The physical size of radioisotope x-ray sources is small. Figure 4 shows the en-
capsulations of typical cylindrical and annular sources of 109Cd (Amersham, 1986).
Cylindrical sources used in portable analyzers are usually 8-mm-diameter by 5-mm-thick
Copyright © 2002 Marcel Dekker, Inc.
Table 1 Properties of Radioisotope Sources Used for XRF, XRA, and XRS Analysis and
Determination of Bulk Density r, Mass per Unit Area (rt), and Thickness t in X-ray Analysis
Dose at 1 m
Radio- Half-life X- or g-ray Photons per from 1 GBq Analytical
isotope (years) energy (keV) disintegration (27 mCi) (mSv=h) technique
55
Fe 2.7 MnK x-rays (5.9, 6.5) 0.28 —a XRF
238
Pu 88 UL x-rays (1330) 0.13 —a XRF
244
Cm 17.8 PuL x-rays (1421) 0.08 —a XRF
109
Cd 1.3 AgK x-rays (22, 25) 1.02 —a XRF
g-rays at 88 0.04
125
I 0.16 g-rays at 35 0.07 2.7 XRF
TeK x-rays (2732) 1.38
241
Am 433 g-rays at 59.5 0.36 3.6 XRF, XRA, XRS
153
Gd 0.66 EuK x-rays (4148) 1.10 27 XRA
g-rays at 97 0.30
g-rays at 103 0.20
57
Co 0.74 g-rays at 122 0.86 24 XRF, XRA, XRS
g-rays at 136 0.11
133
Ba 10.8 g-rays at 81 0.34 65 XRA, XRS,
g-rays at 276 0.07 rt
g-rays at 303 0.18
g-rays at 356 0.62
0.09
137
Cs 30.2 g-rays at 662 0.85 83.7 rt
a
It is difficult to assign a radiological protection meaning to the dose of low-energy x-rays.
capsules and are often referred to as ‘‘pellets.’’ When economics justifies it, a special source
may be designed for a particular type of the analyzer. A so-called ‘‘lollipop’’ source, de-
signed specifically for light-element analysis probes made by Metorex Int. is such an ex-
ample (Amersham IEC.600 series). It is made in a form of a flat copper ring, 1 mm thick,
15 mm in diameter, with a 8-mm opening. One side of this flat copper annulus is elec-
troplated with a 55Fe isotope, over which a few-micrometer-thick Ni flashing is applied.
Such source geometry allows for a very close coupling between sample and a proportional
detector window. Consequently, the quantitative analysis of light elements, down to A1, is
possible without the nuisance of helium purge or vacuum.
There are international codes for the safe use of radioisotopes, and a simple in-
troduction to radiation protection has been published (Martin and Harbison, 1986). Each
organization using radioactive substances is required to hold a license, issued in most
countries by a government health department or atomic energy authority. The Interna-
tional Commission on Radiological Protection (ICRP, 1985) recommends that, for
members of the public, it would be prudent to limit exposures to radiation on the basis of a
lifetime average annual dose of 1 millisievert (mSv). Table 1 lists the approximate, typical
dose rates at 1 m from each radioisotope source, assuming no absorption of the emitted
radiation within the source or by air. The x-ray dose is inversely proportional to the square
of the distance from the source. X-ray doses received during the operation of x- and g-ray
instrumentation and gauges are trivial compared with the maximum permitted doses be-
cause of the low x-ray output of radioisotope sources, careful design of operating tech-
niques, and x-ray shielding.
Copyright © 2002 Marcel Dekker, Inc.
109
Figure 4 Encapsulation of disk and annular Cd g-ray sources. Dimensions are in millimeters.
(From Amersham, 1986.)
Class
Test for
resistance to 1 2 3 4 5 6
Temperature No test 740 C (20 min.), 740 C (20 min.) 740 C(20 min.), 740 C (20 min.), 740 C (20 min.),
þ80 C (1 h) þ180 C (1 h)
þ 400 C (1 h) and
þ 600 C (1 h), þ 800 C (1 h)
thermal shock thermal shock from thermal shock from
from 400 C to 20 C 600 C to 20 C 800 C to 20 C
External No test 25 kPa absolute to 25 kPa absolute to 25 kPa absolute 25 kPa absolute 25 kPa absolute to
pressureb atmospheric pressure 2 MPa absolute to 7 MPa absolute to 70 MPa absolute 170 MPa absolute
Impactc No test 50 g from 1 m 200 g from 1 m 2 kg from 1 m 5 kg from 1 m 20 kg from 1 m
Vibrations No test 30 min, 25500 Hz at 30 min, 2550 Hz at 90 min, 2580 Hz
5gn peak amplitude 5gn peak amplitude; at 1.5 mm amplitude
5090 Hz at 0.635 mm peak-to-peak;
amplitude peak-to-peak; 802000 Hz at 20gn
90500 Hz at 10gn
Punctured No test 1 g from 1 m 10 g from 1 m 50 g from 1 m 300 g from 1 m 1 kg from 1 m
a
Details of the testing procedures are given in ISO.2919 and BS.5288. A further class X can be used when a special test procedure has been adopted.
b
External pressure 100 kPa ¼ 1 atm (approximate).
c
The source, positioned on a steel anvil, is struck by a steel hammer of the required weight; the hammer has a flat striking surface, 25 mm in diameter, with the edges
rounded.
d
The source, positioned on a hardened steel anvil, is struck by a hardened pin, 6 mm long and 3 mm diameter, with a hemispherical end, fixed to a hammer of the required
weight.
Note: The shaded cells of the table when read from top to bottom will correspond to class C33222, which is the minimum requirement for low-energy g-ray and x-ray sources
used in XRF instrumentation and g-ray gauges.
relative, and its is therefore more convenient and accurate to express their FWHMs in
units of energy rather than percent. Energy resolution, expressed as the full width at
half(peak)-maximum (FWHM) and shown for each type of detector as continuous line in
Figure 6, was calculated from equations given by Jenkins et al. (1981d). It is important to
remember that practical energy resolution for any detector is always worse than that
quoted by 10% to sometimes 20% relative. This is because the spectra measured from real
samples usually contain large amounts of backscattered radiation, and, additionally, the
intensities of the analytes may be very high.
As mentioned earlier, the energy resolution of solid-state detectors is superior to that
for proportional and scintillation detectors (see Fig. 26 in Chapter 2). Figure 7 shows the
calculated energy spectrum for the detection of 8-keV x-rays in each detector and also the
energies of the Ka x-rays in the 69-keV energy range. Figure 8 shows the difference in
energy of Ka x-rays between adjacent atomic number elements.
Table 3 compares the difference in Ka x-ray energies with the energy resolution for
aluminum, iron, and tin based on data given in Figures 6 and 8. Solid-state detectors are the
only detectors that can fully resolve Ka x-rays of adjacent Z elements. The factors affecting
their resolution are discussed in detail in Chapter 3 (Sec. III.D). Proportional detectors have
an energy resolution less than twice the energy difference in Ka x-rays of adjacent Z elements.
Hence, their energy-resolving power is useful even if there are adjacent Z elements in the
sample. Scintillation detectors (see Chapter 2, Sec. III.F.2) have such limited resolving power
that other techniques must be used to discriminate between adjacent Z elements, such as
balanced filters. This achieved, however, at the expense of some loss in sensitivity of analysis.
Figure 9 illustrates the resolving capabilities of detectors by showing calculated energy
spectra for identical concentrations of Fe and Ni in the same sample.
Copyright © 2002 Marcel Dekker, Inc.
Figure 6 X-ray energy resolution (FWHM) of scintillation, proportional, and solid-state
detectors. The continuous lines are calculated (data from Jenkins et al., 1981d), the silicon detector
results being based on a small detector (10 mm263 mm): The (diamonds) resolutions are for various
solid-state detectors (EG&G Ortec, 1986); (þ) typical and (6) best for the high-resolution
proportional detectors (Metorex, 1986); and (open circles) typical and (solid circles) best resolutions
for specific NaI scintillation detectors (Harshaw).
The background intensity, Ibgd , is measured in the same window as the intensity
of a given x-ray peak. The width of the window is usually set equal to the peak
FWHM. If a sample containing a certain percentage of the analyte is measured with a
detector of energy resolution E1 (where E1 ¼ FWHM), it will generate in its FWHM-
wide window a certain intensity, I1. If the detector resolution now degrades to, say, E2
(i.e., the analyte’s x-ray peak is now wider), the intensity I2 will still be equal to I1 (as
long as the measurement window is FWHM-wide). This is because the energy de-
posited by the photon in the detector is represented by the area of the whole peak.
Therefore, if the peak is wider, it has to be smaller, and if it is narrower (better re-
solution), it has to be taller—for the area of the peak to remain constant (see Fig. 5).
However, the background intensity will be larger in the case of a worse energy re-
solution, E2, than in the case of E1, because the integration window for the back-
ground intensity is wider in the case E2 than in the case E1. Then, it follows from
Eq. (19) that the DL for the E2 case will be larger (worse) than in the case of E1
energy resolution. Therefore, for the best DLs, it is of paramount importance to use
the detector with the best energy resolution.
The efficiency of the detector for registering the x-rays at high x-ray energies is
determined by the probability that the x-ray interacts with atoms in the sensitive volume of
the detector. The most efficient detectors are those with a high atomic number and a high
mass per unit area. For x-ray energies above about 40 keV, solid-state detectors made of
germanium are preferred to those made of silicon. Also, of the two detectors made of the
same material, the thicker one will have a better efficiency for high-energy x-rays. This is
why even the smallest Si(Li) detector (usually at least 3 mm thick) is superior to the
siliconp-i-n diode detector, which is usually not thicker than 0.5 mm. The efficiency of
Figure 8 The energy difference between the Ka x-rays of adjacent atomic number elements.
Figure 10 Calculated efficiencies of NaI scintillation detector and silicon and germanium solid-
state detectors used in XRF analysis.
C. Electronics
The electronics used with the various detectors are discussed in Chapters 2 (Sec. III.F) and
in 3 (Sec. IV) and are also covered in detail by Jenkins et al. (1981a). The limits to accuracy
and sensitivity of XRF analysis are usually determined by the limitations of the detector in
energy resolution, efficiency of detection, and maximum count rate, for example, rather
than of the electronics. With the excellent gain stabilization electronics now available and
the reliability of low-power, surface-mount components, it is rare that the electronics
system is a significant limiting factor even in the harsh environmental conditions of
industrial plants.
Copyright © 2002 Marcel Dekker, Inc.
IV. X-RAYAND g-RAY TECHNIQUES
The range of radioisotope x- and g-ray techniques used for analysis is far more extensive
than the range based on x-ray tube techniques. Almost all x-ray tube systems are based on
the high-energy resolution of wavelength-dispersive (the crystal spectrometer) or energy-
dispersive (the solid-state detector) devices. With this high resolving power, there is less
need to tailor a technique to the specific application. Radioisotope x-ray systems, espe-
cially those involving scintillation or proportional detectors, usually must be carefully
matched to the specific application. This disadvantage is more than compensated for by
such attributes as mechanical ruggedness, simplicity, and portability, which are so im-
portant in industrial and even more so in field applications.
The selection of the radioisotope source to analyze different elements depends on
many factors, including whether the energy of the radioisotope x- or g-rays is sufficient to
excite the element, the energies of the x-rays scattered by the sample, and the energy re-
solution of the detector. Figure 12 is an approximate guide and, although prepared for
proportional detectors (courtesy of Metorex Int., formerly Outokumpu Oy), can be used
for solid-state detectors and to a more limited extent for scintillation detectors. The iso-
topes of 238Pu and 244Cm, emitting x-rays similar in energy, can be used interchangeably
although curium is preferred for safety reasons. A 57Co isotope can be used with scintil-
lation detectors or germanium solid-state detectors for the K shell XRF analysis for high-
Z elements, such as uranium, gold, and lead.
This section reviews the radioisotope XRF techniques used with solid-state, pro-
portional, and scintillation detectors, x-ray preferential absorption techniques that are
normally based on the use of scintillation detectors, and x-ray scattering techniques that
are often based on use of scintillation detectors. An example of the application of each
technique is also given.
Figure13 The spectrum of x-rays from a copper, lead, and zinc ore sample excited by 238Pu x-rays
and detected by a Si(Li) solid-state detector.
proportional counter filled with a neonargon Penning mixture gas to a pressure of 7 bar
(Jarvinen and Sipila, 1984b). The detection limits are comparable to those obtained with a
silicon solid-state detector with x-ray tube excitation (Jarvinen and Sipila, 1984b). If other
elements in the atomic number range 2630 had been present, however, there would have
been incomplete resolution of the K x-rays emitted and, hence, poorer sensitivity of
analysis.
Figure 17 shows detection limits for the elements in water (Jarvinen and Sipila,
1984a; other data courtesy of Asoma Instruments), normalized to a common counting
time of 100 s. The low detection limits were obtained by careful choice of filling gas and
pressure in relation to the exciting x-ray energy used. The issue of proper match of the
detector with the energy of exciting radiation is much more important for gas-filled pro-
portional detectors than for the solid-state ones. This is because of the much greater
uniform detection efficiency of the solid-state detector for x-rays than that of the gas-filled
one. For most, except low-atomic-number elements, the detection limits were less than
10 mg=g, which is similar to the results for solid-state detectors (Fig. 15); however, they
were achieved in a time six times shorter than before. These low detection limits were
achieved using different proportional-counter gas fillings, and, as a consequence, the
technique is less flexible than that based on the solid-state detector. If adjacent Z elements
had been present, the detection limits for the proportional counter would have been
considerably worse, but for the solid-state detector, they would have been much less
changed.
Copyright © 2002 Marcel Dekker, Inc.
Figure 16 Spectrum of x-rays from a pure water sample and from water spiked with 100 mg/g of
iron and zinc as determined using a (Ne þ Ar)-filled proportional detector and excited by x-rays from
a 244Cm source. (From Jarvinen and Sipila, 1984b.)
1. Filters
Filters placed between the sample and detector (Fig. 18) increase the sensitivity of analysis
by filtering out a higher proportion of fluorescent x-rays of matrix elements than those of
the analyte. Zinc, for example, may be the analyte in samples also containing iron. Cal-
culations based on Eq. (1) and mass absorption coefficients [see Eq. (76) in Chapter 1]
show that a 27-mg=cm2 aluminum filter transmits 27% of the Ka x-rays of 8.6 keV but
only 4.5% of the iron Ka x-rays of 6.4 keV (Fig. 19); that is, it reduces the intensity of iron
K x-rays relative to the zinc K x-rays by a factor of 6. If the sample also contains lead,
however, about 55% of the lead La x-rays (10.5 keV) would be transmitted, twice that of
the zinc Ka x-rays. In this case (Fig. 19), an absorption-edge-type filter of copper (22.4 mg/
cm2) could be used to reduce the lead La x-ray transmission to only 1 % and also to reduce
the iron K x-ray peak. This selective enhancement of the zinc compared with the iron and
lead x-ray components partly compensates for the limitation of the poor resolution of the
scintillation detector. Although some iron and lead fluorescent x-rays will still be detected
within the pulse-height channel set about zinc Ka x-ray peak, in many applications this
component will have been sufficiently reduced to make the analysis possible.
If the measurement with one absorption-edge filter does not give sufficient selectivity
to the desired element’s fluorescent x-rays, balanced filters are used (Fig. 20). The in-
tensities of x-rays in the fluorescent x-ray channel are measured separately, first with one
filter and then with the other (with ‘‘up’’ and ‘‘down’’ filter). The atomic numbers of the
two filters are chosen so that their K shell absorption-edge energies are just above and
below the energy of the Ka x-rays of the analyte. The masses per unit area of the two filters
are chosen so that the product mrt [Eq. (1)] is practically identical for both filters, except
Figure 20 X-ray transmission through balanced filters of copper (6) and nickel (solid circle).
Count-rate measurements made first with one and then with the other filter are subtracted to give a
count rate proportional to the intensity of zinc Ka x-rays from a sample.
2. DirectExcitation Assemblies
The directexcitation technique (Fig. 18) is the most widely used of the three scintillation
detector assemblies. The energy of the radioisotope x-ray is usually chosen so that the
fluorescent and backscattered x-rays are resolved by energy analysis (Fig. 21). The in-
tensity of the backscattered x-rays is used to correct for the absorption of the analyte’s
fluorescent x-rays by the sample matrix [Eq. (8) or (9)]. The filter enhances the sensitivity
and selectivity of analysis.
Direct-excitation assemblies are used extensively in industry (e.g., in laboratory and
portable elemental analyzers (Rhodes, 1971) and in on-line (in-stream) analysis of mineral
slurries (Watt, 1983).
3. DetectorRadiator Assemblies
The detectorradiator assembly (Fig. 18) discriminated well against interfering x-rays of
energy just below that of the fluorescent x-rays of the measured element (Watt, 1972). The
Copyright © 2002 Marcel Dekker, Inc.
basis of this discrimination is that the atomic number of the radiator element can be
chosen so that, of the two x-ray components with nearly similar energies, only the higher
of the two has sufficient energy to excite the K x-rays of the radiator element. The detector
is shielded from the sample and, hence, sees only the x-rays emitted by the radiator. Ba-
lanced-radiator techniques, analogous to balanced-filter techniques, can also be used to
improve selectivity to the analyte if there is another matrix element, in this case emitting
fluorescent x-rays of energy higher than those of the analyte. The count rates obtained
with detectorradiator assemblies are about 5% of those obtained with direct-excitation
assemblies, using a source of the same activity, because of the additional excitation stage
of the radiator.
The intensity of higher-energy x-rays scattered by the sample can be measured si-
multaneously in the one assembly by use of a second radiator element of atomic number
considerably higher than that of the first radiator. The x-ray energies of the two com-
ponents are well resolved and similar to that shown in Figure 21.
Detectorradiator systems are much less widely used than balanced-filter techni-
ques. They are less versatile than balanced-filter systems. Applications include the de-
termination of lead in zinc concentrates, in which the zinc Ka x-rays (8.6 keV) from the
high concentration of zinc (e.g., 50 wt%) swamp the lead L x-rays (10.514.8 keV) from
the low concentration of lead (e.g., 0.5 wt%). A radiator of zinc (absorption-edge energy
of 9.66 keV) is excited by the lead L but not by the zinc K x-rays. This radiator technique
238
Figure 21 Typical spectrum of x-rays from copper ore slurry excited by a Pu source and
measured using a scintillation detector.
The choice of a radioisotope x-ray analysis technique for a specific application depends on
several interacting factors: the overall accuracy of the sampling and analysis required, the
time available to achieve this accuracy, technique available to obtain a sufficiently
representative sample of the material being analyzed, and the sample preparation
Copyright © 2002 Marcel Dekker, Inc.
requirements, such as grinding, pressing, and so forth. The influence of each of these
factors on the accuracy of XRF analysis is discussed here.
where
sinstr Instrumental component such as thermal drift
sstat Uncertainty due to statistics of counting
shet Uncertainty reflecting sample heterogeneity
scal Reflects uncertainties contributed by reference analysis and calibra-
tion fit
ssamp:plac: Uncertainty due to sample placement
smatrx Uncertainty associated with matrix effects
spart:size Uncertainty caused by varying particle size of analyzed material
s?? Any other potential, unidentified uncertainty
This formula assumes that all uncertainties included are random and normally distributed
and that systematic uncertainty (bias) is negligible. Under these assumptions, the stot can
be regarded as a good measure of the accuracy of analysis.
The first two uncertainty components, sinstr and sstat , are often treated together
as main constituents of instrument precision (repeatability), whereas all the other
uncertainties identified in Eq. (22) can be regarded as contributing to user or application
related uncertainty. Thus, Eq. (22) can be written as follows:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
stot ¼ s2instr:precis: þ s2applic: ð23Þ
where
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
sinstr:precis: ¼ s2instr þ s2stat ð24Þ
and
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
sapplic: ¼ s2het þ s2cal þ s2samp:plac: þ s2matrx þ s2part:size þ þ s2?? ð25Þ
Some of the uncertainties can be identified and easily controlled or minimized by the
analyst, such as uncertainty due to statistics of counting, or calibration uncertainty. For
example, the statistical uncertainty of counting, sstat , can be easily reduced by extending
Copyright © 2002 Marcel Dekker, Inc.
the measurement time, whereas the uncertainty associated with calibration, scal , can be
reduced by using more accurately analyzed calibration samples and by selection of a better
calibration equation.
Sample placement uncertainties, ssamp:plac: , can be reduced by always following the
same procedure for preparing and loading the sample into the measuring chamber of the
analyzer, or—partially—by spinning the sample during measurement.
Uncertainties due to matrix variability, smatrx , and particle size, spart:size , can be sig-
nificantly reduced by proper sample preparation techniques such as matrix dilution
(Jenkins et al., 1981a) and by grinding, pelletizing, or fusing the sample with spectrometric
flux, respectively.
Although all these uncertainty minimization techniques are valid and readily ap-
plicable in laboratory environments, they are an unattainable luxury in field and industrial
applications where grinding, fusing or any type of sample preparation is simply out of the
question.
This is why the accuracy of analysis in field and industrial environments tend to be
governed by the application-related uncertainties rather than by the usually much smaller
contribution from the instrument’s precision. It is also for this reason that the measure-
ment time may often be selected from the point of view of process throughput rather than
overall uncertainty of analysis; usually, good instrumental precision allows the analyst to
shorten the measurement time without sacrificing the former. Nevertheless, the mea-
surement time is always a compromise between the throughput of the process and ana-
lytical requirements.
In the coating industry, the time for the determination of the coating mass of tin
and zinc on steel must be less than a few seconds if the result is to be useful; hence
continuous analysis directly on the main coating line is essential. In mineral con-
centrators, rapid changes in the grade of ore entering the plant and the time taken for
the ore to pass through the plant (about 15 min) make it essential for the process
slurries to be analyzed within 5 min. This can be achieved by continuous analysis either
directly in stream or on slurries in sampling loops (bylines). In scrap yards, alloys must
be sorted into different types in short periods: otherwise the sorting operation is not
economical. The lower the price per pound of scrap, the faster the sorting has to be. In
this case, it is essential to have analysis equipment that is both portable and capable of
producing rapid results.
It is important to realize that XRF analysis does require some finite time period
within which a meaningful measurement can take place. The consequence of this fact is
that XRF analysis results in process control applications are, most of the time, the average
rather than point estimates of the process parameters such as coating thickness, sulfur
concentration of diesel fuel, and so forth.
VI. APPLICATIONS
Table 4 lists some important applications of radioisotope x-ray techniques based on XRF
equipment, usually referred to as laboratory or portable elemental analyzers. The analysis
techniques involve the use of scintillation, proportional, and semiconductor detectors.
These analyzers are in widespread use in many application areas and in many industries
(Rhodes, 1971; Rhodes and Rautala, 1983). The total number used worldwide probably is
about 10,000. Some major suppliers of radioisotope based instruments are listed in Table 5.
A photograph of a portable model of an x-ray analyzer for alloys, build around a silicon
p-i-n diode detector is shown in Figure 23 and a laboratory model based on gas-filled
proportional detectors is shown in Figure 24. Many solid-state detector systems are also in
routine use in laboratories.
Table 6 lists types of on-line analysis systems in routine use. These include systems
for the on-line analysis of mineral slurries, flowing powders, coal, coal slurries, paper,
determination of sulfur in oil and petroleum products, and analysis of coatings. Most of
the analysis systems tend to be based on scintillation and proportional detectors, but some
of the more recently developed systems also use solid-state detectors. Table 6 is also an
attempt to list some commercial suppliers of the on-line equipment. Example of the ap-
plication of these techniques are now discussed in details, with emphasis given to appli-
cations of industrial importance. Some examples illustrate the interaction of sampling and
sample presentation with the selection of the appropriate radioisotope technique.
A. Identification of Alloys
Historically, the first applications of portable XRF analyzers were in mining and pro-
specting. It was only with the advent of the on-board memory and microprocessors that the
portable XRF analyzers found wider acceptance and use for analytically more demanding
alloy identification and analysis. Since then, several thousand of these analyzers have been
sold, making alloy sorting and analysis a ‘‘flagship’’ application for portable XRF analy-
zers. These analyzers can assay the alloy and=or identify it by its grade or common name.
Rapid sorting of alloys is required in many areas of the metals industry, such as
smelting, fabrication, inventory and incoming material control, and the sorting of scrap
(Piorek and Rhodes, 1986; Berry, 1981; Piorek, 1989). Some common alloy groups include
nickel alloys, copper alloys, stainless and high-temperature steels, and carbon and chro-
miummolybdenum steels. Although 4050 elements are involved in the alloying process,
in any given alloy there are only 1020, and of these, only about 10 are required for the
identification of specific alloy.
Copyright © 2002 Marcel Dekker, Inc.
Table 4 Some Typical Applications of Radioisotope Based Laboratory and Portable
Elemental Analyzers
Alloy sorting and Low-alloy steels; stainless steels; nickel alloys; high-temperature alloys;
identification titanium, aluminum alloys; specialty alloys; metal scrap
Mining and mineral Copper, lead, zinc, tin, arsenic, molybdenum, nickel, iron, chromium,
bismuth, and uranium in commercial-grade ores, concentrates,
and tailings; titanium and iron in silica sand; silicon, potassium,
titanium, and iron in clays; phosphate rock
Pulp and paper Thickness of silicone coatings on paper and polymer membranes; calcium,
titanium, filler in paper
Environmental Soil screening for metals (Cr, Cu, Ni, Pb, Zn, As, Cd, Hg, Sb); hazardous
materials (e.g., lead, arsenic, chromium, or cadmium in waste sludge);
trace elements in wastewater discharge; metals in air particulates on
filters; chlorine (halogens) in waste oil; sulfur in diesel fuel
Fibers, films and Copper, zinc, tin, gold, silver and chromium plating thicknesses; metals
coatings in plating solutions; silver in photographic film; manganese coating
thickness on magnetic tape; titanium on glass; ruthenium on electrodes
Chemicals and Lead, titanium, and zinc in paint; sulfur, iron, alumina, silica, and calcium
process control in cement; vanadium in catalysts; palladium and gold coatings on silica
spheres uses as catalysts; zinc, chromium, nickel in plating baths
Plastics Calcium, lead, tin, and chlorine in PVC; zinc and bromine in polystyrene;
chlorine in urethane rubbers; bromine and chlorine in butyl rubbers;
silicon in polythene; TiO2 in nylon; bromine in Styrofoam
Agricultural Fertilizers (calcium, phosphates, potassium); copper, chromium, and arsenic
in wood preservatives and treated wood; bromine in almonds; ironzinc
ratio in meat for grading; minerals in cattle feed; titanium in fillers
Cosmetics Titanium, iron, lead in powders
Pharmaceutical Metals in vitamin pills; zinc in insulin
Petroleum products Lead, calcium, sulfur, vanadium, and chlorine in gasoline or oil;
sulfur in petroleum coke; sulfur and ash in coal; lubricating oils additives
The main requirements of analytical equipment for alloy identification are port-
ability, speed and reliability of identification, and an ability to be used by unskilled
operators. Balanced-filter techniques have two main disadvantages. Concentrations of at
least several elements must be determined; hence, separate measurements must be made
with several sets of balanced filters. The sensitivity of analysis is insufficient for the lower
concentrations of some specific elements in the alloys.
Piorek and Rhodes (1986) showed that by using XRF analysis based on a 111-MBq
(3 mCi) 109Cd source and a high-resolution proportional detector, many alloys can be
identified in one measurement. Measurement is first made to identify the alloy by group.
The spectrum of the unknown alloy is then compared with the key features of spectra of
known alloys in the group, which are stored in a memory chip in the equipment. This so-
called grade-identification mode of analysis is particularly useful because it does not
require the operator to be proficient in the specifics of alloy composition. In the grade-
identification mode, the analyzer uses a pattern-recognition algorithm to compare the
x-ray intensities of the measured sample with those of the stored references of alloy
standards. This approach is very fast, taking about 35 s to identify a single alloy. In
Copyright © 2002 Marcel Dekker, Inc.
Table 5 Manufacturers and Suppliers of Radioisotope-Based Laboratory and Portable
Elemental Analyzers
Figure 23 The Metorex Int. portable alloy analyzer based on high-resolution silicon p-i-n diode
detector, MetalMasterTM 2000. (Courtesy Metorex Int.)
comparison, assaying the alloy may take up to 4 min in order to maintain low measure-
ment uncertainty. Table 7 shows the identification results for different alloys by group,
obtained with the analyzer pictured in Figure 23. The probability of correct identification
is satisfactory for all alloys except the carbon steels. The results in Table 7 were generated
by measuring each alloy sample for 5 s for at least 10 times and comparing its spectral
features against the library of about 145 different alloys. The number of correct identifi-
cations ratioed to the total number of trials for a given alloy group was then recorded in
Table 7 as a measure of probability of correct identification. Some of the identification
failures are for alloys very close in composition, for which the main alloying elements
differ in concentrations by less than 1%. The most difficult identification is for carbon
steels in which the concentrations of alloying elements are very low in the presence of
almost 100% iron, and the difference in concentrations of the same elements between two
grades approaches the sensitivity of the XRF technique. Therefore, the XRF is not a
method of choice for identification of carbon steels, for which better techniques, such as
optical emission spectrometry, exist (Piorek et al., 1997).
By using a silicon p-i-n diode detector in place of a gas proportional one, it has been
possible to extend positive identification to such specific alloys as B1900 and B1900Hf,
which differ from each other by less than about 1% Hf, in the presence of several percent
of W and Ta and a balance of nickel.
Overall, the portable alloy analyzers offer a much simpler approach to identification
of alloys, with a reliability of identification as good as that for more complex techniques.
Despite the significant differences in design, the portable alloy analyzers exhibit si-
milar performance in the assay mode; that is, they offer measurement precisions ranging
from 0.01% to 0.5% absolute depending on the element, its concentration range, and the
Copyright © 2002 Marcel Dekker, Inc.
Table 6 On-line Analysis Systems Based on Radioisotope X-ray Sourcesa
matrix. Several references treat this subject in greater detail (Fookes et al., 1975; Jenkins
et al., 1981c, 1981d; Piorek and Rhodes, 1986; Piorek 1989; Spiegel and Horowitz, 1981;
Berry, 1981). Table 8 shows typical performance data for a contemporary portable XRF
alloy analyzer, such as illustrated in Figure 23.
Copyright © 2002 Marcel Dekker, Inc.
Table 7 Performance of a Portable X-ray Analyzer in Alloy Identification
Identification results
Alloy group Measured elements (% feasible)
Nickel and cobalt alloys Ti, Cr, Fe, Co, Ni, Cu, Nb, Mo, W, Hf, Ta 100
Copper alloys Mn, Fe, Ni, Cu, Zn, Pb, Sn 90100; 98100
Stainless steels and Ti, Cr, Mn, Fe, Co, Ni, Cu, Nb, Mo 90100; 100
high-temperature alloys
Cr/Mo steels Cr, Fe, Ni, Mo 90100; 100
Low-alloy steels Cr, Mn, Fe, Ni, Cu, Nb, Mo 6580; 90100
Titanium alloys Ti, V, Mn, Cr, Zr, Mo, Sn 95100;
Aluminum alloys Mn, Fe, Cr, Cu, Zn 90100; 95100
Note: If two results are given, the first refers to a gas-filled proportional detector, and the second to a solid state,
silicon p-i-n diode detector.
Source: From Piorex et al., 1997.
Low-alloy steels 0.01 0.04 0.1 0.25 0.25 0.10 0.05 0.10 0.006 0.15 0.15
0.02 0.1 0.2 0.50 0.25 0.15 0.01
Stainless, hi-temp steels 0.015 0.20 0.10 0.20 0.20 0.20 0.06 0.20 0.01 0.30 0.05
0.03 0.30 0.20 0.30 0.30 0.10 0.03 0.30
Ni/Co alloys 0.10 0.20 0.10 0.12 0.10 0.20 0.05 0.30 0.02 0.30 0.15
0.50 0.30 0.50 0.50 0.50 0.30 0.08
Cu alloys, brass/bronze 0.02 0.10 0.02 0.02 0.05 0.05 0.15 0.07 0.01 0.008 0.20
0.06 0.06 0.08 0.40 0.30 0.20 0.30
Aluminum alloys 0.02 0.05 0.10 0.05 0.05 0.04 0.05 0.06 0.003 0.005 0.01
0.20 0.10 0.005 0.20 0.02
Titanium alloys 0.40 0.10 0.10 0.06 0.05 0.05 0.02 0.02 0.008 0.005 0.01
Notes: The values listed are typical precision ranges in percent absolute for total assay time per sample not longer
than 60 s. The differences are due to the analyzer model and/or the radioisotope(s) used.
Precision data listed are obtained with a 185-MBq (5 mCi) 109Cd and 1.48-GBq (40 mCi) 55Fe sources.
Source: From Piorek et al., 1997.
is about 3 m=s, and the coal particle top size is in the range 10150 mm, depending on the
application. The process requirement for analysis time varies from 1 or 2 s for the fast
sorting of coal to 10 min for the steady-state control used in coal preparation plants.
Three x-ray or low-energy g-ray techniques have been developed: backscatter of
x-rays (H. Fraenkel, private communication, 1987), backscatter of low-energy g-rays
(Fauth et al., 1986; Cierpicz, 1986), and dual-energy g-ray transmission (Fookes et al.,
1983; Gravitis et al., 1987). Each relies on the photoelectric effect, which depends on
atomic number, and on the fact that ash (mainly SiO2 and Al2O3 with smaller con-
centrations of Fe2O3) has an effective atomic number, Z, greater than that of the coal
matter (carbon, hydrogen, nitrogen and oxygen).
The x-ray technique (H. Fraenkel, private communication, 1987) depends on the
scatter of approximately 17-keV x-rays from 238Pu (or 244Cm) source in the coal and, at
the same time, excitation of iron K x-rays in the coal to correct for the high absorption per
unit mass by Fe2O3 compared with Al2O3 and SiO2. Because the low-energy x-rays pe-
netrate only thin layers of coal, the coal is sampled from the conveyor, subsampled, and
ground to 5-mm top size particles, partially dried, and then presented in a moving stream
of controlled geometry to the radioisotope x-ray analysis system. This system compensates
for the effect of variations in Fe2O3 in the ash, a significant source of uncertainty in some
applications. However, it involves complex sampling, sample handling and processing,
and blockages occur when the coal is very wet.
The low-energy g-ray technique, using an 241Am (60 keV) source, depends on mea-
surement of the intensity of g-ray scattered from thick layers (< 15 cm) of coal. It was first
used on a high-throughput sample line (Fauth et al., 1986). Although coal must be
sampled, there is no need for the coal to be subsampled and crushed because of the high
Copyright © 2002 Marcel Dekker, Inc.
penetration of g-rays in the coal. The technique has been adapted for use on-line (Cierpicz,
1986), the analysis head unit riding on a raft that is spring loaded so that it is always
touching the top of the coal on the conveyor. Its use is restricted to conveyor speeds of less
than 2 m=s and to minimum thickness of 15 cm of coal on the conveyor compared to the
normal practice of 520 cm.
The dual-energy g-ray transmission technique (Fookes et al., 1983; Gravitis et al.,
1987) measures coal directly on the conveyor belt (Fig. 28), eliminating the need for
sampling the coal. The ash content is determined independently of vertical segregation
of coal on the belt, and if segregation across the belt occurs, the narrow beam of g-rays can
be made to scan across the belt to obtain a representative sample. The coal mass per unit
area in the g-ray beam must be < 3 or 4 g=cm2 to achieve sufficient sensitivity of analysis.
Variations of iron in the ash limits the accuracy of ash determination in some applications.
The choice of a suitable x- or g-ray analysis technique is highly influenced by the
complexities in sampling of the coal on the conveyor and the subsequent subsampling and
grinding. Radioisotope techniques that measure directly on-line are preferred to those
involving sampling, and if sampling is necessary, preference is given to those that minimize
sampling and sample presentation.
Dual-energy g-ray transmission is now the preferred technique for the on-line
determination of ash content of coal, except for applications in which unacceptable un-
certainties in ash are caused by variations in iron in the ash. In this case, a high-energy
g-ray technique, which is based on the pair production interaction and is much less sen-
sitive to variations in iron, is preferred. Its main disadvantages are cost and that it must
operate on a sample line (Sowerby, 1986).
routinely by one of two radioisotope x-ray techniques. One technique combines XRF
(55Fe and a proportional detector) and b-ray transmission. The other combines XRA,
using 22-keV silver K x-rays that are absorbed equally per unit weight by both carbon and
hydrogen, with a nonnuclear technique to determine the density of the oil. In Japan alone,
more than 100 of the former gauges and several hundred of the latter are used (Table 6).
More recently, this application received an application boost from regulatory agencies
such as U.S. Environmental Protection Agency. Since October of 1993, no on-highway
diesel fuel the United States can contain more than 500 mg=kg of sulfur. This regulation
forced sulfur monitoring for compliance at refineries, distributorships, and local suppliers.
A proportional detector with neon gas filling and a thin, 13-mm Be window, coupled with a
1.48-GBq (40 mCi) 55Fe isotope allows sulfur analysis in diesel fuel down to 50 mg=kg,
with a precision of 8 mg=kg (1s), in 300 s measurement time (Piorek and Piorek, 1993).
Copyright © 2002 Marcel Dekker, Inc.
Figure 29 Mass absorption coefficients of the most frequently used filler materials, water and
cellulose, relating to the manufacture of paper. (From Kelha et al., 1983.)
A factor of 5 better results may be obtained by using low-power x-ray tube in the place of
the isotope. The issue of sulfur measurement in petroleum products for compliance with
the EPA regulations has been addressed by American Society for Testing and Measure-
ment (ASTM), with two ASTM Standard Test Methods, D2622 and D4294. Similar
regulations for sulfur in petroleum products for on-highway use are being implemented in
Europe and Asia.
Spent engine oil is burned as fuel in industrial boilers and home heating furnaces.
Because chlorinated solvents have routinely been mixed with much of this oil, there is a
danger that hazardous levels of these compounds or their derivatives will be released into
the atmosphere. The U.S. Environmental Protection Agency has banned the sale of used
oil for fuel if the total halogen level (interpreted as total chlorine level) exceeds 1000 mg=g.
If the total halogen level is more than 4000 mg=g, such oil has to be considered as
Copyright © 2002 Marcel Dekker, Inc.
Figure 30 Measuring principle used in the on-line analysis of paper: (a) XRF, (b) XRA, and (c) b-ray absorption. (From Kelha et al.,
1983.)
VIII. CONCLUSIONS
Figure 32 Niton II, XRF Alloy Analyzer based on a silicon p-i-n diode. (Courtesy Niton Corp.)
ACKNOWLEDGMENTS
The author thanks the many scientists, organizations, and companies who generously
supplied information used in this chapter and to the following for permission to reprint
figures used in this chapter: Pergamon Press (Figs. 2, 14, 18, 21, 22, and 28); EG&G Ortec
(Fig. 1); The Radiochemical Centre, Amersham (Fig. 4); the International Atomic Energy
Agency (Fig. 7); The Analyst (Fig. 20); Outokumpu Oy (now Metorex Int.) (Figs. 11 and
23); IEEE (Fig. 16); Automatica (Fig. 25); Australian Mineral Development Laboratories
(Fig. 26); The Australasian Institute of Mining and Metallurgy (Fig. 27); and Acta
Polytechnica Scandinavia (Figs. 29 and 30).
REFERENCES
Keith W. Jones
Brookhaven National Laboratory, Upton, New York
I. INTRODUCTION
It has been known for almost 100 years that the acceleration of a charged particle will
result in the radiation of electromagnetic energy. The development of the betatron and
synchrotron electron accelerators 50 years later led to the experimental observation of
radiation from electrons circulating in a closed orbit (Elder et al., 1947) and to naming it
specifically synchrotron radiation after the accelerator used to produce it. The first re-
cognition of the unique properties of the synchrotron radiation by Tombulian and
Harman (1956) then brought about an explosive development of activity in constructing
improved sources for production of the radiation and using the radiation in experimental
science.
The original synchrotrons were designed for use in nuclear physics research. Later,
facilities were designed to optimize the conditions for production of x-rays. The compo-
nents of the new facilities include a source of electrons or positrons and an accelerator to
produce high-energy beams. This might be done by the use of a linear accelerator to
produce energies of around 100 MeV. These beams are then injected into a synchrotron
and boosted to energies in the GeV range. Finally, the beam is stored in the same accel-
erator used to attain the final energy or in a separate storage ring. An acceleration field in
the ring is used to supply energy to the beam to compensate for the radiation energy loss.
Copyright © 2002 Marcel Dekker, Inc.
The lifetime of the stored beam is many hours, so that, in practice, the synchrotron source
is almost starting to become similar to a standard x-ray tube in use.
The synchrotron-produced x-ray beams have unique properties that make them
desirable for use. They have a continuous energy distribution so that monoenergetic
beams can be produced over a wide range of energies. The photons are highly polarized in
the plane of the electron beam orbit, which is extremely important for background re-
duction in SRIXE-type experiments in particular. The x-rays are emitted in a continuous
band in the horizontal direction but are highly collimated in the vertical direction. It is
therefore possible to produce intense beams with little angular divergence. The source size
is small and, as a result, the production of intense beams of small area is feasible. The
synchrotron source is a pulsed source because of the nature of synchrotron-type accel-
erators. The x-rays are produced in narrow bursts, less than 1 ns in width, and have a time
between pulses of around 20 ns or more.
The main parameters of interest in defining the synchrotron source are as follows:
1. Magnitude of the stored electron=positron current. Typically, currents are in the
range from 100 to 1000 mA. The lifetimes of the stored beam are many hours.
The lifetimes for stored electrons at the National Synchrotron Light Source
(NSLS) at Brookhaven National Laboratory, (Upton, NY) have typical values
of around 24 h. At Laboratoire pour L’Utilisation de Radiation Electromagne-
tique (LURE), Orsay, France, where positron beams are used, the lifetimes are
even longer because the positive beam does not trap positively charged heavy
ions produced from the residual gases in the vacuum chamber.
2. Assuming that the size and angular divergence of the electron beam are not
important, the source brightness is defined as [in units of photons=s=mr2=(0.1%
bandwidth)]:
d 2I o
¼ 1:327 1013 E2 ðGeVÞiðAÞH2 ð1Þ
dy dc oc
where E is the energy of the electron beam (in GeV), i the current (in A), H2
(o=oc) is a function tabulated in Kim (1986), o is the photon angular frequency,
and oc is the critical frequency which splits the emitted power into halves and is
given by the expression 3g3c=2r, where g is the electron energy in units of the
electron rest mass, c is the velocity of light, and r is the radius of curvature of the
electron path. The angles y and c are the angles of emission in the plane of the
electron orbit and perpendicular to that plane, respectively.
3. The total photon emission is found by integrating over c and is given by [in units
of photons=s1=mr1=(0.1% bandwidth)]
dI o
¼ 2:457 1013 EðGeVÞ iðAÞ G1 ð2Þ
dy oc
For electron beams with nonzero emittance (finite area and angular divergence), it is
necessary to define another quantity, the brilliance, which is the number of photons
emitted into angular intervals dy and dc at angles y and c from an infinitesimal source
area (in units of photons=s1=mr2=mm2=(0.1% bandwidth)]. The values of brilliance
and brightness are important in evaluating the performance expected for focused and
collimated x-ray microscopes. Kim (1986) provides a detailed discussion of the emittance
effects, polarization, and performance of wiggler and undulator insertion devices.
Copyright © 2002 Marcel Dekker, Inc.
Values attained for these quantities at the 2.5-GeV x-ray ring of the NSLS are shown
in Figures 2–4. Its x-ray beams are typical of those produced by second-generation syn-
chrotron storage rings. The ring energy is high enough to produce x-rays over an energy
range sufficient to produce K x-rays from elements to about Z ¼ 40 with good efficiency
and L x-rays throughout the periodic table. Thus, it is highly suitable for use as the basis of
a system for x-ray microscopy-based SRIXE.
The brilliance obtained from two third-generation synchrotrons is shown in Figures 5
and 6 as a function of the photon energy for several types of devices. The brilliance
obtained from an undulator insertion device at the 7-GeV Argonne National Laboratory
Advanced Photon Source is shown in Figure 5. The brilliance for different bending
magnet, undulator, and wiggler devices at the 8-GeV Spring-8 facility in Japan is given in
Figure 6. It can be seen that the third-generation storage rings have gained approximately
three to four orders of magnitude in brilliance when compared to that found from bending
magnets at the NSLS. Similar gains are found for third-generation rings operating at lower
energies although the spectrum of x-rays is naturally affected by the energy of the stored
electron=positron beam. This increase translates to a gain in x-ray intensity that enables
new types of x-ray fluorescence experiments.
The high degree of linear polarization of the x-rays from the synchrotron source is a
major factor in making the synchrotron XRM a sensitive instrument. The physics describing
the interaction of polarized x-rays with matter is therefore an important topic. Hanson (1990)
Figure 2 The brilliance of the NSLS x-ray ring is plotted as a function of the x-ray energy
produced. The brilliance is important in defining limitations on production of x-ray images using
focusing optics. (From Jones KW, et al. Ultramicroscopy 24:313, 1988.)
has carried out an extensive examination of the scattering problem and given methods for
assessing particular geometries used in the XRM. A second point is the continuous x-ray
energy spectrum extending from energies in the infrared to hard x-rays with energies of several
hundred kiloelectron volts. The tunable energy is important for many types of experiments.
There are now many synchrotron facilities located around the world that are suitable for use
in various types of x-ray spectrometry measurement. They fall roughly into three classes.
First-generation synchrotrons were built primarily as high-energy physics machines
and were used secondarily for synchrotron radiation production. The Cornell High En-
ergy Synchrotron Source (CHESS) at Ithaca, NY is an example.
Second-generation synchrotrons were optimized as radiation sources, and as a result,
produce x-ray beams with superior brilliance and brightness characteristics. The two rings
of the Brookhaven NSLS fall in this category.
Finally, the third generation of synchrotrons is now being designed. They will be the
first sources intended to incorporate insertion devices, wigglers, and undulators in the design
phase. In some cases, the ring energy is increased to give better performance with the in-
sertion devices. Synchrotrons designed with these features will begin operation in the latter
part of the 1990s. The European Synchrotron Radiation Facility (ESRF) at Grenoble,
Copyright © 2002 Marcel Dekker, Inc.
Figure 4 The x-ray flux produced by the NSLS x-ray ring is plotted as a function of the x-ray
energy produced. The values are given after integration over the vertical opening angle of the beam.
The definition of the flux is given in Eq. (2). (From Jones KW, et al. Ultramicroscopy 24:313, 1988.)
France, the Super Photon Ring-9 GeV (SPring-8) in Kansai, Japan, and the Advanced
Photon Source (APS) at Argonne National Laboratory, Argonne, IL, illustrate this case.
A listing of a number of synchrotron laboratories producing high-energy x-ray
beams suitable for use in SRIXE is given in Tables 1–3 (Fuggle, 1990; Jackson, 1990;
Winick, 1989, 1990). The number of XRM beam lines is growing rapidly and their em-
ployment for research as a result is becoming widespread.
The apparatus that is used for x-ray microscopy measurements varies substantially from
laboratory to laboratory. A schematic diagram of the components of a very compre-
hensive system is shown in Figure 7. All the components are not necessarily used in a
specific instrument. The most important differences between systems lies in the treatment
of the incident beam. The simplest approach is to use the white beam (full-energy
spectrum) and a collimator. The more complex systems use focusing mirrors to collect
more photons and demagnify the beam and monochromators to produce monoenergetic
beams. At present, the performance of the various systems is quite comparable in terms of
spatial resolution and detection limits (DLs). Thus, a versatile and flexible approach to the
Copyright © 2002 Marcel Dekker, Inc.
Figure 5 The brilliance obtained from an undulator insertion device at the 7-GeV Argonne
National Laboratory Advanced Photon Source. (From Advanced Photon Source Research,
ANL=APL=TB-31 1(1):5, 1998. Argonne National Laboratory, managed and operated by The
University of Chicago for the U.S. Department of Energy under Contract No. W-31-109-ENG-38.)
Figure 6 The brilliance for different bending magnet, undulator, and wiggler devices at the 8-GeV
SPring-8 facility in Japan. (From Kitamura H. SPring-8 Annual Report 1994, JASRI, 1995, p 47.)
choice and design of the components is advisable. A number of the different instruments
are described briefly in the following subsections.
motors for producing variable-sized beams of size greater than about 10 mm. The col-
limation approach is ultimately limited by beam-spreading related to finite source and
pinhole dimensions and to diffraction.
The collimated XRM (CXRM) was first operated at the now defunct Cambridge
Electron Accelerator by Horowitz and Howell (1972). They used a pinhole made by the
evaporation of a thick gold layer around a 2-mm quartz fiber, followed by subsequent
etching away of the quartz. The x-ray energy was about 2.2 keV, and image contrast was
achieved using determination of the linear attenuation coefficient. The specimen was
moved past the incident beam, and fluorescent x-rays were detected with a proportional
counter. A spatial resolution of 2 mm was measured with this pioneering apparatus.
Later versions of the CXRM have been put into operation at Hamburg and
Brookhaven. Some of the details of the operations are given to illustrate some of the more
important operational details.
It is often best to maximize the photon flux on the sample by employing a white
beam of x-rays. The flux available at a point 9 m from a NSLS bending magnet source is
shown in Figure 8. The flux from this source, integrated over the entire energy spectrum, is
about 36108 photons=s=m2 with a 100-mA stored electron current.
For best operation of the CXRM, care must be taken in shaping the incident x-ray
spectrum using filters. The influence of the beam filters has been studied by the Hamburg
group (Knöchel et al., 1983). Their results are shown in Figure 9. By varying the effective
energy, it is possible to tailor the beam to give the best possible DL for a given atomic
number. Filters on the detector can be used in some cases to reduce the effect of a major
Copyright © 2002 Marcel Dekker, Inc.
Figure 7 A comprehensive synchrotron beam line designed for use as a x-ray microscope. Not all
the components would be utilized at a given time in practice. The rather varied uses of the system
require a flexible approach to the design of the equipment.
element. In experiments that examine the distribution of trace elements in bone, a filter of
polyimide can reduce the high rates caused by the calcium in the bone, but it will not have
a large effect on the x-rays from iron, copper, and zinc.
An equally critical task is the alignment of the energy-dispersive x-ray detector.
Kwiatek et al. (1990) have reported on this phase of the optimization procedure. The
importance of aligning the detector in the horizontal plane can be seen by reference to
Figure 10. In addition, the energy spectrum and degree of polarization change as a
function of the vertical distance from the plane of the electron orbit in the storage ring.
Figure 11 shows the relative photon flux for the two polarization states. Examination of
the curves shows that the alignment needs to be made to an accuracy of better than a few
hundred micrometers to get the best reduction of scattered background.
Results of the experimental background=peak ratio determination as a function of
vertical displacement are shown in Figure 12 for elements from calcium to zinc in a gelatin
matrix and for palladium in a pyrrhotite matrix. The dependence on scattering angle in the
Copyright © 2002 Marcel Dekker, Inc.
Figure 8 Energy distribution of photon flux produced by a bending magnet on the NSLS x-ray
ring at a electron current of 100 mA. (From Jones et al., 1990b.)
Figure 9 Values for DLs obtained for a white-light XRM are shown as a function of atomic
number. The change in the DLs as a function of filtering of the incident beam using aluminum filters
is an important feature of this type of arrangement. (From Knöchel et al., 1983.)
Figure 11 Polarization of NSLS x-ray beams is given as a function of distance from the plane of
the electron orbit for x-ray energies of 10, 20, and 30 keV. (From Hanson AL, et al. Nucl Instrum
Methods Phys Res B24=25:400, 1987.)
horizontal plane is shown in Figure 13. The details of the methods used to make the
alignments are given by Kwiatek et al. (1990).
The Brookhaven work has shown that it is possible to achieve spatial resolutions
below the 10-mm range using the CXRM. The spatial resolution of the instrument has
been demonstrated by scanning a thin-evaporated gold straight edge through the beam
and recording the intensity of the L x-rays as a function of distance. The resolution in
this case was 3.5 mm. Some control over the resolution for a given collimator is obtained by
inclining the collimator with respect to the incident beam.
The Brookhaven device is routinely used for trace element measurements with a spatial
resolution of less than 10 mm. The sensitivity and DLs obtained under these conditions have
been reported by Jones et al. (1990b) and their results are displayed in Figures 14 and 15.
Copyright © 2002 Marcel Dekker, Inc.
Figure 13 The background-to-signal ratio is plotted as a function of scattering angle in the
horizontal plane. The curve was obtained by displacement of the detector and the equivalent angular
range spanned was roughly 4 . (From Kwiaek et al., 1990.)
Figure 15 DLs measured for the BNL collimated XRM under the conditions described in the text
and for Fig. 14. (From Jones et al., 1990b.)
Fe (cps) Zn (cps)
Beam filter
thickness (mm A) Mb Tc M T
1. LURE (1987)
The equipment used for SRIXE at LURE has been described by Chevallier et al. (1987)
and Brissaud et al. (1989). The white beam from the storage ring passes through a
beryllium window and is incident on a curved crystal of mosaic pyrolytic graphite. The
size of the incident beam is 2 cm, 1 cm before the monochromator, which produces a
final focused beam of 14-keV photons with a size of 1 mm61 mm. The fluorescent x-rays
are detected with a Si(Li) detector collimated using a 2.8-mm aperture. The DLs
achieved are around 1 ppm for thick geological specimens and much less for organic
matrices. The probe has only been used for measurements, which do not require high
spatial resolution.
Figure 17 Schematic diagram of the focusing XRM in operation at the Daresbury XRM. (From
Van Langevelde et al., 1990c.)
3. SRS (1990)
The instrumentation at the Daresbury SRS has been developed by a group of collaborators
from the Free University in Amsterdam and Warwick University at Coventry. The latest
device uses a silicon crystals as both a monochromator and focusing device (Van Langevelde
et al., 1990a, 1990b, 1990c). A diagram of the instrument is shown in Figure 17. The crystal
Copyright © 2002 Marcel Dekker, Inc.
is a 100-mm-thick silicon crystal bent with a radius of 100 mm in the sagittal plane and
5740 mm in the meridional plane. The beam is focused to a spot size of about 15 mm620 mm
at an energy of 15 keV. The demagnification is 1000 in the horizontal plane and 15 in the
horizontal plane, with a photon flux increase of greater than 104. A flux of 15-keV photons
at the specimen of 104 photons=(S mm2 mA) is obtained. This is an impressive result when it
is remembered that the device is located about 80 m from the electron beam.
4. Novosibirsk (1989)
Baryshev et al. (1989) working at the Novosibirsk VEPP-3 synchrotron have used both
monoenergetic and white beams of x-rays. A single-crystal pyrolytic graphite mono-
chromator was used to produce monoenergetic beams with energies between 8 and 35 keV.
The spatial resolution was 60 mm for the monoenergetic beam and 30 mm for the white
beam. The DL for the monoenergetic beam was 10 ppm for elements from iron to
strontium for a 1–3-s run.
5. LBL (1988)
The group at the Lawrence Berkeley Laboratory has developed a focusing system based
on the Kirkpatrick–Baez geometry (Giauque et al., 1988; Kirkpatrick and Baez, 1948;
Thompson et al., 1987; Thompson et al., 1988; Underwood et al., 1988, Wu et al., 1990).
A schematic diagram of their system is shown in Figure 18. The system uses a parallel
beam of photons to produce an image which is demagnified by about a factor of 100 to
produce final images of a few micrometers. The use of multilayer coatings of tungsten
carbide on the mirrors gives a quasimonoenergetic beam with a bandwidth of about 10%
at 10 keV and a high throughput. Much of the experimental work has been carried out at
the NSLS in collaboration with the BNL group. It is thus possible to make comparisons in
performance between the CXRM and FXRM on the same storage ring.
Figure 18 Schematic diagram of the focusing XRM of the Berkeley group used at the BNL X26
beam line. (From Thompson et al., 1988.)
Figure 19 Schematic layout of K–B mirrors and a four-crystal channel-cut monochromator for
micrometer precision micro-x-ray absorption spectrometry and micro-diffraction developed by
Padmore et al. at the ALS. (From Warwick et al., 1998.)
Bajt, Jones, and Bilderback showed that SRIXE experiments can be successfully per-
formed with the capillaries. SRIXE experiments are made difficult by the large divergence
of the x-ray beam after it emerges from the capillary tip. In order to achieve the best
spatial resolution, it is necessary to place the sample within a few millimeters of
the capillary and at the same time arrange optimal detection geometry for the x-rays from
the sample at a 90 angle to the incident beam.
C. Experiments at a Distance
The effects of the remarkable increase in computer communications through the Internet
and World Wide Web are visible everywhere. In the past few years, there has been a
development of interest in making use of these technologies to help make facilities useful
to researchers at remote locations. One example has been the DOE 2000 initiative of the
U.S. Department of Energy (US Department of Energy, 1997). There are two components
of this initiative that are related to use of synchrotron beam lines. First, there is the
Copyright © 2002 Marcel Dekker, Inc.
concept of national collaboratories. These are ‘‘laboratories without walls that unite ex-
pertise, instruments, and computers, enabling scientists to carry out cooperative research
independent of geography.’’ The second is to set up pilot projects that are ‘‘virtual la-
boratories that give scientists the technology to collectively observe and attack problems
using combinations of ideas, methodologies, and instrumentation that do not exist at any
single location.’’
Examples of this idea exist for several types of instrumentation. An example of the
application to a synchrotron beam line is that of the ‘‘Spectro-Microscopy Collaboratory
at the Advanced Light Source: A Distributed Collaboratory Testbed’’ (Agarwal et al.,
1997). This prototype of a collaboratory is supposed to enable operation of equipment on
an undulator beam line at the ALS from the University of Wisconsin–Milwaukee. Some of
the topics of concern for the collaboratory are security mechanisms, data dissemination,
remote monitoring and control, safety mechanisms for beam line operation, resource
arbitration to decide location of control, video conferencing and remote viewing, network
needs, and shared electronic notebooks for data handling. Another example involves the
University of Florida and the APS (CAT Communicator, 1998). This project is supposed
to allow remote operation of a materials research beam line. The computer components
needed are similar to those listed for the LBL–Wisconsin demonstration.
Successful XRMs have been put into operation using both continuum and monoenergetic
synchrotron-produced x-rays. The continuum radiation is extremely convenient to use
because it is easy to construct a CXRM with a minimum of equipment and achieve ex-
cellent performance. Further, the broad-band excitation means that measurements can be
made for essentially all elements in the periodic table in a single exposure. Monoenergetic
radiation can be used at an energy optimized for production of the characteristic x-rays of
a given element, thus reducing radiation damage in organic materials. Counting-rate
limitations in energy-dispersive detectors are reduced because of the elimination of scat-
tered x-ray events. The energy can be tuned to eliminate interferences (e.g., lead–arsenic)
and to eliminate excitation of elements with Z higher than that of the element of interest.
Maps can also be constructed by subtraction of images obtained above and below ab-
sorption edges.
Successful XRMs have also been produced in collimated and focused modes em-
ploying either continuum or monoenergetic radiation. The brute-force-type collimated
continuum radiation microprobe employed at Brookhaven and Hamburg has been
comparable to the other types of probe in terms of spatial resolution and DL. The per-
formance of the Brookhaven instrument has been compared with the performance of the
LBL Kirkpatrick–Baez XRM operated on the same NSLS beam line (Giauque et al.,
1988). A comparison of results obtained with the CXRM positioned at 10 m from the
source with the FXRM at 20 m from the source is given by Rivers et al. (1992). Figure 22
shows the results of the comparison. The DL obtained with the FXRM is somewhat better
than for the CXRM, but the wider energy range of the CXRM is a substantial advantage
in some cases. The quality of the Kirkpatrick–Baez optics makes it feasible to dispense
with the use of a collimator placed close to the specimen, which is an advantage in the
design of the experiments in some cases.
For ultimate performance in terms of the spatial resolution, FXRM might be the
best choice because the construction of good collimators on the 1-mm2 size or smaller scale
Copyright © 2002 Marcel Dekker, Inc.
Figure 22 Sensitivity of the BNL collimated white-light XRM compared with the sensitivity
obtained with the Berkeley focused XRM. (From Rivers ML, et al., 1992.)
VI. QUANTITATION
Methods for making quantitative elemental determinations using x-ray fluorescence have
been developed over many years. These approaches are discussed for conventional x-ray
sources in Chapter 5 and 6. Some of the approaches used at synchrotron sources are given
here to show how the methods developed for use with conventional tubes have been used
with the new radiation source.
Copyright © 2002 Marcel Dekker, Inc.
Giauque et al. (1986) measured several U.S. National Bureau of Standards (NBS),
now National Institute of Standards and Technology (NIST), and the Japan National
Institute of Environmental Science (NIES) standard reference materials (SRM) at the 54-
pole wiggler beam line at the Stanford Synchrotron Radiation Laboratory. They referred
their measurements to a copper standard prepared by evaporation where the weight was
determined from gravimetric measurements. Multielement standards were prepared by
dissolving known weights of an element in an acidic solution. Monoenergetic radiation
was used for the work at energies of 10 and 18 keV. The beam size was defined by an
aperture 3 mm in size. The differences in sample mass were accounted for by normalizing
to the intensity of the scattered radiation (Giauque et al., 1986). The results obtained for
three NBS materials are shown in Table 5.
The work of Giauque et al. (1986) addressed quantitation in thin biological speci-
mens where matrix effects were negligible and where the Compton scattering could be used
for normalization of masses. It is necessary to extend this approach if thick specimens are
to be investigated. There are many geological experiments where this situation holds.
Methods for quantitation have been discussed by Brissaud et al. (1989) and by Lu et al.
(1989). Corrections are made for attenuation of incoming and fluorescent x-rays by the
sample matrix and by any filters employed as well as for secondary fluorescence. The
composition of the major elements is generally known for geological materials; hence,
concentrations can be referred to that of one of the major elements.
Brissaud et al. (1989) compared the SRIXE results with several standards. The re-
sults are shown in Table 6. The table gives the recommended concentration and the
SRIXE value. It can be seen that agreement is quite good. In this case, as noted earlier, a
comparatively large 1-mm beam was used. Lu et al. (1989) used a microprobe with a beam
size of 30 mm660 mm to analyze different specimens of feldspars. They compared their
SRIXE results with values obtained using an electron probe and atomic absorption
spectroscopy. The agreement with the electron probe was good, but the atomic absorption
values tended to be systematically lower than the SRIXE values. Results for the com-
parison with the electron probe are shown for iron and strontium in Figure 23.
For thin biological specimens, concentrations can be established by use of a sensi-
tivity curve such as the one displayed in Figure 14. Corrections for differences in thickness
can be made by normalizing to the intensity of the scattered incoherent peak if a mono-
energetic beam is used or to the continuum comprising both incoherent and coherent
scattering if white beam is used (Giauque et al., 1979).
Quantitation of SRIXE draws on many years of experience gained using tube-excited
x-ray fluorescence. The main difference between the x-ray sources occurs when SRIXE is
used with microbeams with dimensions on the micrometer scale. In this case, strict at-
tention must be given to the uniformity of the standards used and to the experience gained
in calibration of the electron microprobe (Hren et al., 1979) and proton microprobe
(Johansson and Campbell, 1988).
The related questions of sensitivites and DLs have been addressed by calculations based
on the known physical parameters of the XRM systems and by empirical determinations.
The detailed understanding of the x-ray production process using synchrotron radiation
is helpful in assessing the sensitivities and DLs that can be achieved in SXRM. The
results cited for the sensitivities and DLs should be taken as representative of the current
Copyright © 2002 Marcel Dekker, Inc.
Table 5 Results Determined for Three NBS Standard References Materials (mg=g)
Nonfat milk powder SRM 1549 diska Wheat flour SRM 1567 diska Rice flour SRM 1568 diska
51 mg=cm2 wtb 37 mg wtc 500 mg 60 mg=cm2 wtb 43 mg wtc 400 mg 60 mg=cm2 wtb 43 mg wtc 400 mg
K 16,900 3,000 17,800 2,000 1,360 40 1,220 130 1,120 20 1,360 160
Ca 13,000 500 12,000 800 190 10 174 10 140 20 158 14
Cr 0.0026 0.0007 < 0.6 < 0.3 < 0.4
Mn 0.26 0.06 0.33 0.12 8.5 0.5 8.2 1.8 20.1 0.4 22.1 2.8
Fe (2.1) 2.30 0.16 18.3 1.0 17.1 4.8 8.7 0.6 9.1 1.2
Ni 0.24 0.06 (0.18) 0.11 0.06 (0.16) 0.18 0.06
Cu 0.7 0.1 0.65 0.04 2.0 0.3 1.88 0.12 2.2 0.3 2.21 0.22
Zn 46.1 2.2 46.9 0.9 10.6 1.0 10.3 0.4 19.4 1.0 21.9 1.8
As (0.0019) < 0.05 (0.006) < 0.03 0.41 0.05 0.42 0.09
Se 0.11 0.01 0.09 0.04 1.1 0.2 0.92 0.06 0.4 0.1 0.38 0.04
Br (12) 12.1 0.2 (9) 8.5 1.4 (1) 1.19 0.17
Rb (11) 13.1 0.2 (1) 0.94 0.06 (7) 8.4 0.9
Sr 3.69 0.10 0.82 0.04 0.19 0.04
Hg 0.0008 0.0002 < 0.1 0.001 0.0008 < 0.06 0.0060 0.0007 < 0.08
Pb 0.019 0.003 < 0.1 0.020 0.010 < 0.1 0.045 0.010 0.10 0.09
a
Mass thickness of disks.
b
Weight of area scanned.
c
Recommended sample weight.
Source: From Giauque et al., 1986.
Element GST 16.6 21.7 GST 16.6 21.7 GST 16.6 21.7
situation. The actual values depend on the particular experimental conditions, syn-
chrotron ring currents, spatial resolutions, and so forth so that exact comparisons are not
terribly meaningful.
Sparks (1980), Gordon (1982), Gordon and Jones (1985), and Grodzins (1983) have
presented calculations of the minimum detection limits to be expected using the second-
generation synchrotron radiation sources such as the NSLS to produce x-ray microbeams.
Many experimental determinations have been made for the two quantities. The results
obtained by different groups using different types of XRMs are given here.
Figures 14 and 15 show the values obtained by Jones et al. (1990b) using the colli-
mated microprobe at the NSLS. Their values are relevant to thin specimens with an or-
ganic matrix. Values for the calculated DLs extrapolated from the work of Gordon and
Jones (1985) are included in Figure 15. It can be seen that the experimental values are in
the same range as the calculations although the experimental system was not the same as
the one considered theoretically. The sensitivities that are shown in Figure 14 indicate the
agreement between predicted (solid curve) and measured (data points) counting rates. The
curves were calculated from knowledge of the experimental arrangement, synchrotron
energy spectrum, and specimen parameters. The experimental points show the measured
values assuming the same specimen parameters.
Copyright © 2002 Marcel Dekker, Inc.
Figure 23 Comparison of concentrations for Sr and Fe in feldspar obtained using the NSLS
XRM with those obtained using an electron microprobe. The solid lines show the values expected for
exact agreement between the two methods. The good agreement validates the use of the XRM
method for geological analyses. (From Lu et al., 1989.)
Figure 24 Relative sensitivity for determination of elements from K to Zn obtained using the LBL
Kirkpatrick–Baez XRM at the NSLS X26 beam line. NIST (NBS) thin glass Standard Reference
Materials 1832 and 1833 were used. (From Giauque et al., 1988.)
Few experiments appear to have been devoted to measuring detection limits at the
third-generation sources. Results from the ESRF have been discussed in comparison to the
values found at the NSLS in the survey of beam lines given in Sec. IV.B. Improvements in
spatial resolution and=or detection limits can be generally assumed. These improvements
should be by about a factor of 100 when comparing the use of undulator beam lines at the
new facilities with their bending magnet performance.
Passage of a photon beam through a specimen results in energy deposition through the
photoelectric effect and Compton-scattering processes. This energy deposition results in a
breaking of molecular bonds as the secondary electrons produced lose their energy by
further ionization or scattering processes with other atoms. The effects have been ex-
amined in great detail for electron beams used in electron microscopy. Much less has been
done for the XRM. This is not surprising because the field is much younger and less ex-
tensively developed than is the field of electron microscopy. Another important reason is
that the photon beam fluences employed thus far have been substantially smaller than
those employed in the electron microscope, and as a result, the magnitude of beam-
induced damage has not been so important.
Biological and other organic materials are generally more susceptible to beam-
induced damage than are other materials. For this reason, the discussion is limited to these
materials. In the future, when more intense synchrotron x-ray sources are available, it will
be necessary to expand the list of materials considered.
Copyright © 2002 Marcel Dekker, Inc.
A qualitative calculation can also be done to illustrate the energy deposition for
photon beams of various energies. Assume that the photon flux used for a typical fluor-
escence experiment is 106 photons=(s mm2). Typical run times are 10 min or less. The
maximum photon fluence for current XRM instruments is thus about 66108 pho-
tons=mm2. It is interesting to note that these fluences are now starting to approach
the range of the fluences found in use of the electron microprobe. In order to estimate
the energy deposition, it is assumed that attenuation coefficients can be represented by the
photoelectric process attenuation coefficient only and that all of the energy of the pho-
toelectrons is absorbed in the volume considered. The Compton-scattering process is re-
latively small and can be neglected in a qualitative estimate of the dose. The results of the
calculation are shown for photon beams with energies less than 20 keV.
The x-ray dose needed to kill living biological systems has been examined in detail over
the years (Kirz and Sayre, 1980; Sayre et al., 1977, 1978; Slatkin et al., 1984; Themner et al.,
1990). A dose which exceeds 1 Gy is likely to cause serious damage to a biological cell or system.
Thus, there are limitations to the use of x-ray beams for the examination of living
systems, as there are, of course, for all other beams. The limitations depend on the ab-
sorbed dose. One way that this can be done is by measuring the linear attenuation coef-
ficients in CMT or in projection-radiography-type experiments. Spanne (1989) has
calculated the dose given to an object in obtaining an image with a signal-to-noise ratio of
5 for a water phantom with a contrasting detail with a diameter 0.005 of the phantom
diameter. The results of Spanne for objects with different diameters and compositions are
shown in Figure 26a and 26b. It can be seen that the dose is strongly energy dependent and
Figure 26 (a) Absorbed dose at the center of a circular water phantom for detection of an element
of fat with a signal-to-noise ratio of 5 as a function of photon energy. The cylinder diameters are
shown in the figure. (b) Absorbed dose at the center of 1-mm-diameter water phantoms for detection
of elements of fat, air and calcium with a signal-to-noise ratio of 5. The element diameter is 0.005 of
the phantom diameter or 5 mm. (From Spanne, 1989.)
There are now examples of the uses of the XRM in many different fields. A brief description
of the diverse applications is given here to illustrate the rapid development of the field and to
give an idea of the ways the XRM may be used in the future. The examples presented cover
Copyright © 2002 Marcel Dekker, Inc.
Figure 27 Morphology of red blood cells is shown as a function of the fluence of incident 15-keV
photons. The fluences for the different exposures are as follows: 0, 0.461019, 1.161019, 1.661019,
2.061019 and 2.461019 photons=cm2 for frames (a)–(f), respectively. (From Slatkin et al., 1984.)
Copyright © 2002 Marcel Dekker, Inc.
Figure 28 Mass loss observed for skin sample irradiated with a 66-keV beam of photons shown as
a function of the dose. The mass loss was determined by observation of the scattered radiation
intensity. (From Themner et al., 1990.)
mainly activities during the past 10 years. The total number of experiments performed
during this time is very large. The intent is to give an example of work covering a variety of
topics. There is no intent to give an encyclopedic review of all applications.
A. Archaeology
Brissaud et al. (1989) made low-resolution measurements on a number of different ma-
terials. One case involved examination of three Gallic coins using synchrotron radiation
and a comparison with the results obtained using proton-induced x-ray emission and
neutron-activation analysis. The x-ray and proton beams probe material close to the
surface of the coin, whereas the activation approach gives the bulk concentration. The
results given in Figure 29 show differences in the concentrations found with these methods
and also show that the activation approach is not applicable to all elements. This
straightforward example shows that the synchrotron can be used to good effect in studying
archaeological and other materials with spatial resolutions of the order of 1 mm. It is
however, a type of experiment that should be viewed as a bridge between the use of
conventional x-ray tube sources and the synchrotron sources with high brilliance.
bone are similar and that the effects on bone growth and mineralization caused by Pb
influences the distribution of both cations. This experiment shows that SRIXE can be used
as a valuable complement to more standard techniques used in biomedical investigations.
The effect of lead on the developing brain has been investigated by Cline et al. (1996)
and Jones et al. (1997). They studied the retinotectal system of frog tadpoles as a function
of exposure to various levels of lead during growth. The lead content in the optic tectum of
the exposed tadpoles was measured using SRIXE at the NSLS X26 beam line with white
light and a collimated beam. Lead levels of about 200 ppb were successfully detected. Use
of SRIXE was advantageous because it can be used with the very small amounts of sample
Copyright © 2002 Marcel Dekker, Inc.
Figure 31 (a) Scan over a thin section of the tibia diaphysis of a rat which had been treated with
gallium nitrate showing the distribution of the Ca, Zn, and Ga. Some of the variations come from
irregularities in specimen thickness. The spatial distribution is a function of the gallium nitrate
treatment of the rat. The step size was 9 mm. (b) Map of the distribution of gallium (left) and calcium
(right) in a fetal rat ulna bone after exposure to gallium (25 mM) in a culture medium. The light
regions have the highest elemental concentrations. A scanning electron micrograph (center) shows
the bone structure in the same region. The dark portions at the top of the figure are the noncalcified
cartilaginous portions of the bone, which accumulate little gallium. Most of the gallium is distributed
in metabolically active regions of the metaphysis and diaphysis. (From Bockman et al., 1990.)
material available. Atomic absorption spectroscopy could not be used for that reason. The
results of the overall experiment demonstrated the impact of lead on the developing brain
through the structure of the retinal axons. Further, work with the chelating agent 2,3-
dimercaptosuccinic acid showed that the effect of lead on the neuronal structure could be
reversed at a level superior to removal of the lead source alone. These conclusions are
relevant to lead-induced cognitive deficits in humans. The results demonstrate that SRIXE
can be a useful measurement technique in many different types of biomedical applications
where it is necessary to measure trace concentrations of an element in thin tissue sections
where sample size or handling preclude other conventional approaches. Lower detection
limits at undulator beam lines will increase the scope of this type of analytical experiment.
Osteoarthritic degeneration of articular cartilage is a degenerative disease of major
importance which can be brought about by senescence and trauma. It causes changes in
the cartilage and in the bone. The metacarpal joints of horses affected by osteoarthritis can
be used as an animal model for study of the disease. Rizzo et al. (1995) have studied 5-mm-
thick cartilage slices obtained from normal and arthritic horses using x-ray microbeams
(10 mm in size) at the X26A beam line at the NSLS to map the spatial distribution of S, Ca,
and Zn along a line through the section. It was found that there were differences in the
Copyright © 2002 Marcel Dekker, Inc.
Figure 31
results for the normal and arthritic tissue for all three elements. It is argued that the zinc
may be contained in the metalloenzyme alkaline phosphatase. This result may be of use in
understanding the metabolic processes of importance in osteoarthritis.
determine wood density and elemental composition. Figure 33 shows results obtained for
the distribution of Mn and the density in the region analyzed. The area examined is at the
boundary between winter and onset of growth in the spring. Measurements at the cellular
level make it possible to examine pathways for accumulation of toxic elements from the
environment. The experiments at the NSLS have been aimed at looking at accumulations
of metals over a long-term period at trees growing in close proximity to a metal smelter
(Martin et al., 1998). The focus of this work was to attempt to determine the relationship
between the time-varying environmental conditions and the composition of the wood. The
results obtained using SRIXE and secondary ion mass spectroscopy (SIMS) to examine
thick wood specimens showed that the distributions were very heterogeneous and that
growth cycles were not easily distinguished from the trace element concentrations.
In geology, many of the experiments used thin sections of rocks for studies of specific
types of minerals in a heterogeneous matrix. Examples of this type of application are
the study of zoned carbonate gangue cements found in Tennessee (Kopp et al., 1990).
The XRM measurements were used in an attempt to interpret the effects of trace amounts
Copyright © 2002 Marcel Dekker, Inc.
Figure 33 Results obtained for the distribution of Mn and the density in the region analyzed in
the ESRF experiments were carried out using high-resolution beams formed by capillary focusing to
determine wood density and elemental composition. (Rindby A, Engström, From ESRF
Highlights—1996=1997, p 80.)
spatial resolution and improved detection limits feasible with new XRM equipment will
make it possible to substantially improve this type of investigation.
X. TOMOGRAPHY
Several groups have done CMT work with synchrotrons. Flannery et al. (1987)
developed a third-generation type system at the NSLS. They used a x-ray magnification
system with a scintillation detector coupled to an image intensifier to produce images with
a claimed resolution down to 1 mm. A similar approach was used by Bonse et al. (1986)
and Kinney et al. (1988) at SSRL. Spanne and Rivers (1987) demonstrated at a first-
generation system at the NSLS X26 beam line. Later work with the apparatus has pro-
duced images with a spatial resolution of 1 mm61 mm and a slice thickness of 5 mm, quite
comparable to the results from the third-generation devices. More recently, a third-gen-
eration apparatus has also been put into operation at BNL (Dowd et al., 1998). It can be
used for analyses with voxel sizes as small as 2.7 mm3.
The first-generation approach takes longer to produce images but has the advantage
that beam scattering effects in the sample are eliminated and SRIXE measurements can be
performed to produce elemental maps. In the third-generation systems, elemental maps
are made by subtracting images taken above and below the absorption edge of interest.
The DL for such an approach is about 0.1%.
Spanne (1990) has carried out a pilot study with the aim of evaluating the potential
for mapping of light elements at the cellular level in the rat sciatic nerve using fluorescence
CMT. A comparison of the mean free path for characteristic x-rays from potassium and
typical sciatic nerve sizes shows that it is feasible to make corrections for the attenuation of
the potassium K x-rays in the nerve. The computed emission tomogram of the distribution
of potassium in the epineurium of a rat sciatic nerve given in Figure 36 illustrates this
point. Note that the short escape depth for the potassium x-rays that are observed
Copyright © 2002 Marcel Dekker, Inc.
Figure 36 Computed emission tomogram showing the distribution in the epineurium of a freeze-
dried rat sciatic nerve. The pixel size was 3 mm63 mm and a slice thickness of 5 mm. The matrix size
was 1756175 pixels. (From Jones et al., 1990a.)
necessitates special measures during the reconstruction of the image. Saubermann (per-
sonal communication, 1989) points out that fluorescence CMT makes the studies of ele-
mental distributions in unsectioned samples possible. Examination of unsectioned samples
also makes in vitro analysis of sections of nerves several millimeters long feasible.
Copyright © 2002 Marcel Dekker, Inc.
Longitudinal distributions of elements can then be conveniently studied by scanning at
different heights, and it is even possible to go back to a previously mapped region for a
more detailed examination if steep concentration gradients are discovered. Longitudinal
concentration gradients have been demonstrated in the rat sciatic nerve, although with a
very poor longitudinal resolution and may be of significance in nerve injury (LoPachin
et al., 1988; LoPachin et al., 1990).
Exxon (Auzerais et al., 1996) and BNL (Spanne et al., 1994, Coles et al., 1996, 1998a,
1998b) groups have investigated topics related to the microgeometry of sandstones in
parallel experiments at the NSLS. In addition to measuring properties of the sandstones
such as porosity, premeability, tortuosity, and connectivity, attention was given to the
displacement of oil by water and modeling of flow through the experimentally determined
structures. A structure of Fontainebleau sandstone measured in the experiment of Spanne
et al. (1994) is shown in Figure 37. The structure was analyzed to give a theoretical re-
presentation of the structure that could be used for predicting fluid flow at larger sizes scales.
Feng et al. (1999) have examined the structure of micrometeorites using the BNL
equipment. Previous examinations of the particle have shown that there are void spaces
and small nuggets of Pt in the interior of some of these particles. This information was
gained by laborious sequential sectioning of particles with a diameter of about 400 mm.
The use of CMT to gain the same information represents a major step forward in the
analytical technique for the field. A representation of a Pt nugget observed in a micro-
meteorite is displayed in Figure 38. The results show that it will be possible to system-
atically examine large numbers of specimens to obtain significant information about the
nuggets and about the history of the particle as it passes through the Earth’s atmosphere.
Several groups at the ESRF and NSLS have been concerned with the development of
improved methods for obtaining contrast between materials with similar x-ray absorption
coefficients (see, for example, Raven et al., 1997). Buffiere et al. (1998) have applied the
phase-contrast technique, based on diffraction patterns produced at discontinuities in the
sample, to measure damage in metal matrix composites. This was an ingenious experiment
that investigated the development of cracks in the composite as a function of tensile forces
applied to the specimen in situ. The experiment revealed the cracking of SiC particles,
decohesion between sample phases, and propagation of pores.
The development of CMT at synchrotrons has been very rapid over the past 10 years
from the standpoint of instrumentation. Applications have also been developed, but
perhaps at a slower pace. This will change in the future, with an increase in the avail-
ability of the technique and improvements that can be made with the increasing use
of third-generation synchrotron undulators. It is apparent that both absorption and
emission approaches are needed and that both approaches are required at any synchrotron
XRM facility. The emission technique will be greatly assisted by the use of undulator
beam lines to obtain higher counting rates.
Extended x-ray absorption fine structure and the related XANES have been widely applied
to give information on the chemical state of elements in many different materials (Winick
and Doniach, 1980; Koningsberger and Prinz, 1988). Many of the experiments that have
been carried out have used relatively large x-ray beams and thick specimens to make
absorption measurements on a timescale of minutes possible. This approach is not useful
for elemental concentrations less than about 0.1–1%. The use of EXAFS and XANES at
lower concentrations can be achieved by use of fluorescent x-ray detection. Cramer et al.,
Copyright © 2002 Marcel Dekker, Inc.
Figure 37 View of the pore space of Fontainebleau sandstone obtained by CMT. The pore
structure is shown as white and the rock as black; that is, the pore structure is opaque and the rock
transparent. (From Spanne et al., 1994.)
(1988) have developed at 13-element Si(Li) x-ray detector which gives a high effective
detection efficiency for this type of application. Other workers have developed means of
acquiring EXAFS spectra on a millisecond timescale (Tolentino et al., 1990) and have
made measurements of chemical state on a finer spatial scale (Iida et al., 1989). The future
development of x-ray microscopy can thus be seen to include the use of SRIXE with
EXAFS and XANES and the development of new techniques to make it possible to work
with improved DLs and beam sizes at the micrometer level.
The NSLS X26 group has employed a simple channel-cut silicon monochromator
with an energy resolution of about 1.1 eV for several demonstration XANES experiments.
The beam was first defined with a four-jawed aperture whose size could be adjusted using
computer-controlled stepping motor drivers. This was followed by the monochromator
placed about 10 cm upstream from the target. The beam moved vertically on the target by
about 60 mm during the scan. This was not important for the resolutions used, but could
be easily compensated for by a correlated motion of the target to keep the same spot under
the beam. The first test used a thick specimen of NIST SRM 1570 spinach leaves, which
contained 550 ppm iron. The iron x-rays were detected with the XRM equipment de-
scribed earlier. The work was done with a beam size of 2 mm2. The results of a scan of
a pure iron specimen (beam size 200 mm6200 mm) and the spinach leaves are shown in
Figure 39. The spectra agree well with the results of a similar scan done on the NSLS X19
beam line using the 13-element Si(Li) detector. Extrapolation from these initial values
showed that work with beam sizes down to 100 mm and perhaps lower was feasible with
the existing equipment for this particular target.
Copyright © 2002 Marcel Dekker, Inc.
Figure 38 CMT three-dimensional image of the interior of type-I deep-sea spherule with parts of
the surface cut away. The compact bright spot in the upper right is a platinum group metal nugget.
The light mass in the foreground represents a hole in the interior of the spherule. (From Feng et al.,
1999.)
A more stringent demonstration is the use of thinner specimens. For this purpose,
measurements were made on the chromium contained in olivine and pyroxene components
in a 30-mm section of lunar mare basalt 15555 from Apollo 15 and in a 10-mm section of a
rat kidney. The lunar basalt study was undertaken because the oxidation state of the
chromium could shed light on conditions existing at the time of the formation of the
mineral studied (Sutton et al., 1991). The rat kidney measurement was needed to cast light
on nephrotoxic effects resulting from environmental exposures. It is hypothesized that the
oxidation state of the chromium changed during its passage from lungs to kidneys with
related implications for health effects.
The basalt specimen contained chromium at a level of about 1000 ppm and the
kidney at a level of about 50 ppm. Figure 40 shows the XANES spectrum for chromium in
pyroxene and olivine contained in the lunar basalt taken with a beam resolution of
200 mm6200 mm. Figure 41 shows the spectrum obtained for the rat kidney, but in this
case with a 1-mm61-mm beam size. The beam was positioned over the medulla portion of
the kidney. The XANES spectra show that chromium exists primarily in the 2þ and 3þ
states in the lunar olivines but as the 3þ state in the kidney.
Copyright © 2002 Marcel Dekker, Inc.
Figure 39 Fluorescent XANES spectrum obtained for Fe at the 550-pm level in a thick NIST
SRM 1570 (spinach) sample compared to the spectrum obtained under the same conditions for an
iron foil.
relate to speciation of contaminants in soils and sediments (Bertsch et al., 1994; Bajt et al.,
1993; Sutton et al., 1994; Tokunaga et al. 1997, 1998). Ilman et al. (personal commu-
nication, 1996) have investigated the chemical state of Cr and As in wood treated with
chromated copper–arsenate and the role of Mn in fungal diseases of wood. The work,
taken as a whole, serves as an actual demonstration of the concept mentioned in Sec. IX
for definition of a field of Molecular Environmental Science (Robinson, 1997).
There are several examples of the way that these experiments can be extended at third-
generation synchrotron facilities. Sarret et al. (1998) used a bending magnet beam line at
the ESRF to do experiments with a spatial resolution of 300 mm6300 mm for examination
of trace elements at concentrations as low as 100 ppm. These parameters are similar to
those at the NSLS X26 beam line. Measurements were made to study uptake mechanisms
of Pb and Zn in lichens. Lichens were of interest because they are commonly used as a
biomonitor to assess environmental pollution. Analysis of these spectra made it possible to
reach an understanding of how Pb and Zn are transported and absorbed in the plant.
A demonstration experiment similar in intent to the work at the NSLS and ESRF
has been performed at the ALS (Warwick et al., 1998). In this example, measurements
were made with a much higher spatial resolution. The oxidation state of Cr in a soil sample
was determined by first scanning a region of 60 mm680 mm. One region of high Cr con-
centration was found and chosen for examination with XANES. The spectrum obtained
Copyright © 2002 Marcel Dekker, Inc.
Figure 41 Fluorescent XANES spectrum obtained for Cr at the 50-ppm level contained in a thin
section of rat kidney. The spatial resolution was 1 mm61 mm. The spectrum shows that there is little
Cr VI present in this portion of the kidney.
from this region with an area of a few square micrometers indicated that the Cr was in
both Cr(III) and Cr(VI) oxidation states. The results of this experiment are shown in
Figure 42.
Another example is given by the work of Cai et al. (1998) at the APS. They de-
scribe a beam line using a undulator source and zone plate focusing to obtain a beam
size of less than 0.25 mm and a photon flux of 561010 photons=s at a bandwidth 0.01%.
It is possible to do x-ray diffraction, x-ray fluorescence, and computed microtomography
Copyright © 2002 Marcel Dekker, Inc.
Figure 42 Elemental mapping and chemical speciation of Cr in a soil specimen on the micrometer-
size scale. Measurements performed by R. Reeder et al. (Dept. of Geosciences, SUNY Stony Brook)
at the ALS. (From Warwick et al., 1998.)
simultaneously on this beam line. Data are presented for spatial distributions of Mn, Fe,
Cu, and Zn in a root, hydrated P. lanceolata, infected by a fungus, mycorrhizal fungus
G. mosseae. A XANES spectrum for Mn obtained from a location in the root at a spatial
resolution of 1 mm63 mm is shown in Figure 43.
The examples of work that has been done at the NSLS, ESRF, ALS, and APS on
soils and metal transport in plants during the past decade illustrates the usefulness of the
synchrotron XRM for geological and environmental work. A similar collection of ex-
periments can be assembled to illustrate other applications of SRIXE=XANES=EXAFS
using both second- and third-generation facilities. It is clear that the scientific applications
of these techniques will increase dramatically in the future.
What does the future hold for synchrotron XRM? Several different developments can be
predicted for the first decade of the next millenium, 2000–2010, with some confidence.
First, continuation of the present types of experiments will go on at an expanded rate
as the virtues of the method become better known. The applications will benefit from enhance-
ments in the facilities, which will bring spatial resolutions to 1 mm2 for two-dimensional
maps or to 1 mm3 for CMT. Relatively minor improvements to the existing systems are
needed to achieve this level of resolution. Detection-level (DL) values should also improve
somewhat with the introduction of improvements in the x-ray detector systems, again
using known techniques. There will be a further merging of XRM techniques with those
developed for EXAFS and XANES.
Second, development of new focusing methods will begin and this will lead to the
routine achievement of spatial resolutions of the order of 0.1 mm by 2000–2005. An ex-
ample is the production of zone plates suitable for focusing of 8-keV x-rays by Bionta et al.
(1990) and the application of capillary focusing by Bilderback et al. (1994a,b,c).
Copyright © 2002 Marcel Dekker, Inc.
Figure 43 A XANES spectrum for Mn obtained from a location in the root at a spatial resolution of
1 mm63 mm. (Cai et al., 1998. Argonne National Laboratory, managed and operated by The
University of Chicago for the U.S. Department of Energy under Contract No. W-31-109-ENG-38.)
Figure 44 The spatial resolution (expressed as the beam area) obtained with high-energy x-ray
microscopes is plotted as a function of time. The best spatial resolutions in 1990 are around 1–4 mm2.
If improvements had continued at the same rate, resolutions would have been close to atomic
dimensions around the year 2000. This did not happen.
ACKNOWLEDGMENTS
Ade H, Kirz J, Hulbert SL, Johnson ED, Anderson E, Kern D. Appl Phys Lett 56:1841, 1990.
Agarwal D. An Overview of the Spectro-Microscopy Collaboratory at the Advanced Light Source:
A Distributed Collaboratory Testbed. March 3, 1997; http://www-itg.lbl.gov/BL7 Collab
Altwicker ER, Schonberg JS, Konduri NV, Milligan MS, Hazard Waste Hazard Mater 7:73,
1990.
Auzerais FM, Dunsmuir J, Ferréol BB, Martys N, Olson J, Ramakrishnan TS, Rothman DH,
Schwartz LM. Geophys Res Lett 23:705, 1996.
Bajt S, Clark SB, Sutton SR, Rivers ML, Smith JV. Anal Chem 65:1800, 1993.
Baryshev VB, Gavrilov NG, Daryin AV, Zolotarev KV, Kulipanov GN, Mezentsev NA, Terekhov
YaV. Rev Sci Instrum 60:2456, 1989.
Bertsch PM, Hunter DB, Sutton SR, Bajt S, Rivers ML. Environ Sci Technol 28:980, 1994.
Bilderback DH, Rev Sci Instrum 66:2058, 1995b.
Bilderback DH, Hoffman SA, Thiel DJ. Science 263:201, 1994a.
Bilderback DH, Hoffman SA, Thiel DJ. Synchrotr Radiat News 7:27, 1994c.
Bilderback DH, Pahl R, Freeland R. Chess Newslett 41, 1995a.
Bilderback DH, Thiel DJ, Pahl R, Brister KE. J Synchrotr Radiat 1:37, 1994b.
Bionta RM, Ables E, Clamp O, Edwards OD, Gabriele PC, Miller K, Ott LL, Skulina KM, Tilley R,
Viada T. Opt Eng 29:576, 1990.
Bockman RS, Repo MA, Warrell RP, Jr, Pounds JG, Schidlovsky G, Gordon BM, Jones KW. Proc
Natl Acad Sci USA 87:4149, 1990.
Bonse U, Johnson Q, Nichols M, Nusshardt R, Krasmaki S, Kinney JH. Nucl Instrum Methods
Phys Res A245:644, 1986.
Brissaud I, Wang JX, Chevallier PJ. Radioanal Nucl Chem 131:399, 1989.
Buffiere J-Y, Maire E, Cloetens P, Peix G, Salomé M, Baruchel J. ESRF Newslett 30: 20–21, April
1998.
Caiz, Yun W, Pratt ST, Miller RM, Gluskin E, Hunter DB, Jarstfer AG, Kemner KM, Lai B, Lee
H-R, Legnini DG, Rodrigues W, Smith CI. Advanced Photon Source Research, ANL/APS/
TB-31, 1:5, 1998.
CAT Communicator 8. Distance Links: Working Remotely at the APS. CAT Communicator 8, May
1998, 1–2, 1998.
Chen JR, Chao ECT, Minkin JA, Back JM, Jones KW, Rivers ML, Sutton SR. Nucl Instrum
Methods Phys Res B49:533, 1990.
Chevallier P, Jehanno C, Maurette M, Sutton SR, Wang J. J Geophys Res 192:E649, 1987.
Cline HT, Witte S, Jones KW. Proc Natl Acad Sci USA 93:9915, 1996.
Coles ME, Hazlett RD, Muegge EL, Jones KW, Andrews B, Dowd B, Siddons P, Peskin A,
Spanne P, Soll WE. Proceedings of 1996 Annual Technical Conference and Exhibition,
Denver, Colorado, 1996, Paper SPE 36531, p 413.
Coles ME, Hazlett RD, Spanne P, Soll WE, Muegge EL, Jones KW. J Petrol Sci Eng 19:55, 1998a.
Coles ME, Hazlett RD, Muegge EL, Jones KW, Andrews B, Dowd B, Siddons P, Peskin A,
Spanne P, Soll WE. SPE Reservoir Eval Eng 288, August 1998b.
Cornacchia M. Synchrotr Radiat News 11:28, 1998.
Cramer SP, Tench O, Yocum M, George GN. Nucl Instrum Methods Phys Res A266:586, 1988.
Delaney JS, Bajt S, Sutton SR, Dyar MD. In: Dyar MD, MacCammon C, Schaefer MW, eds.
Mineral Spectroscopy, A Tribute to Roger G. Burns, Special Publication No. 5. St. Louis: The
Geochemical Society, 1996, p 165.
Delaney JS, Dyar MD, Sutton SR, Bajt S. Geology 26:139, 1998a.
Delaney JS, Sutton SR, Dyar MD. Lunar Planet Sci XXIX, #1241, March 1998b.
Dowd BA, Andrews AB, Marr RB, Siddons DP, Jones KW, Peskin AM. 47th Annual Denver X-ray
Conference, Colorado Springs, CO, 1998.
Elder FR, Gurenwitsch AM, Langmuir RV, Pollock HC. Phys Rev 71:829, 1947.
I. INTRODUCTION
The phenomenon of total reflection of x-rays had been discovered by Compton (1923). He
found that the reflectivity of a flat target strongly increased below a critical angle of only
0.1 . In 1971, Yoneda and Horiuchi (1971) first took advantage of this effect for x-ray
fluorescence (XRF). They proposed the analysis of a small amount of material deposited
on a flat totally reflecting support. This idea was subsequently implemented in the so-
called total-reflection x-ray fluorescence (TXRF) analysis which has spread out worldwide.
It is now a recognized analytical tool with high sensitivity and low detection limits, down
to the femtogram range. TXRF is basically an energy-dispersive technique in a special
excitation geometry. This geometry is achieved by adjusting the sample carrier inclined
under angles of about 1 mrad (0.06 ) to the primary beam. The incident beam thus im-
pinges at angles below the critical angle of (external) total reflection for x-rays. Usually the
liquid sample, with a volume of only 1–100 mL, is pipetted in the center of the reflector. As
a result of the drying process where the liquid part of the sample is evaporated, the residue
is irregularly distributed on the reflector (within an area of a few millimeters in diameter),
forming a very thin sample. (See Figure 1.)
Total-reflection XRF offers the following advantages:
If the adjusted incident angle of the primary radiation is below the critical angle,
almost 100% of the incident photons are totally reflected, so the primary
radiation scarcely penetrates into the reflector and the background contribu-
tion from scattering on the substrate is drastically reduced.
The sample is excited by both the direct and the reflected beam, which results in a
doubling of the fluorescent intensity.
Due to the geometry, it is possible to position the detector close to the surface of the
reflector where, in the center, the sample is located. This results in a large solid
angle for the detection of the fluorescence signal.
Consequently, the peak-to-background ratio in TXRF is high and the detection limits are
drastically improved by several orders of magnitude as compared to conventional XRF.
The very distinct angular dependence of the characteristic x-ray intensities close to
the total-reflection regime can be used to investigate surface impurities, thin near-surface
layers, and even molecules absorbed on flat surfaces. The analysis of the composition,
thickness, and density of layers as well as the nondestructive in-depth examination of
concentration profiles in the range of 1–500 nm are possible.
Demands for TXRF are as follows:
Higher photon flux of excitation sources as compared to standard XRF
Small source size (typically 10 mm640 mm for x-ray tubes for TXRF)
Low divergence (in one direction below 1 mrad).
As most widely spread excitation sources, high-power Mo x-ray diffraction tubes are
in use. Depending on the chemical elements of interest, other anodes might be preferred.
The insertion of a spectral modification device in the beam path of the primary radiation
improves the background; that is, a multilayer monochromator suppresses in the ideal case
all photons, except the ones with an energy range which covers the most intense char-
acteristic line of the anode material.
It was demonstrated that the optimum excitation source for TXRF is a synchrotron
storage ring. The routine determination of light elements (below Si) with TXRF is still a
challenging task: Both the excitation as well as the detection of the fluorescence radiation
of interest are difficult. In this context, several spectrometers were constructed, based on
ultrathin entrance window and windowless detectors as well as synchrotron radiation and
new prototype x-ray tubes applied for excitation.
Typical applications of TXRF include analysis of drinking water, river water, rain-
water, seawater, wastewater, body fluids, tissue, purity of chemicals (acids, bases, solvents,
etc.), oils and greases, aerosols, fly ash, soils, forensic and art-historical objects, study of
thin layers and depth profiles, and so forth.
The theoretical fundamentals of TXRF can be deduced in a way analogous to that applied
in light optics. One has to consider the interaction of an electromagnetic wave that hits the
plane boundary between vacuum and a medium described by its refraction index n, which
takes into account both scattering and absorption and can be calculated by quantum
mechanics:
Copyright © 2002 Marcel Dekker, Inc.
n ¼ 1 d ib ð1Þ
It can also be expressed as function of atomic scattering factors f1 and f2 (see Fig. 2), which
are tabulated by atomic number dependent on the incident x-ray energy (Henke et al., 1993):
e 2 l2 r
n ¼ 1 NA ð f1 þ if2 Þ ð2Þ
m0 c2 2p A
where NA is Avogadro’s number, l is the wavelength of the incident radiation, r is the
density of the medium, A is the atomic mass, e and m0 are the electric charges and rest
mass of an electron, respectively, and c is the velocity of light.
By comparing Eqs. (1) and (2), we obtain
r0 r
d ¼ NA l2 f1 ð3Þ
2p A
where r0 is the classical electron radius:
e2
r0 ¼ ð4Þ
m0 c 2
The real part of the complex refraction index (1 7 d) is slightly smaller than unity with
d ¼ 10 7 5–10 7 6 in the energy range of x-rays. The minus sign reflects the fact that the
bound electrons follow the exciting photons in the opposite phase. Consequently the re-
fraction index for x-rays is always smaller than 1 (except sometimes for the energies close
to absorption edges). The imaginary part b (usually smaller than d) is a measure of the
absorption and is related to the photoelectric mass absorption coefficient tm :
Figure 2 f1 and f2 for silicon as a function of the incident radiation energy. At the absorption edge
energy of 1838 eV, discontinuities in the scattering and absorption behavior are found. For energies
far above this energy, f1 approaches asymptotically the value of its atomic number Z ¼ 14.
For x-rays, the real part of nreflector is slightly less than 1, whereas the refraction index for
vacuum is 1 (in good approximation also valid for air). Consequently, the reflector is the
optically thinner medium than vacuum (air), and the reflected beam will be deflected
toward the boundary, which is completely different as compared to usual light optics. Due
to the fact that the angles for which the total reflection effect is observed for x-rays are
small, sine functions can be replaced by their argument ðsin j0 ! j0 Þ, and cosine by
1 j20 =2.
The electromagnetic waves for the incoming, reflected, and refracted beams can be
defined in the usual way for light optics, and Fresnel’s formulas will give the ratios of, for
example, the reflected ðER Þ and incoming ðE0 Þ amplitudes:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ER j 0 j20 2d 2ib
¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð7Þ
E0 j þ j2 2d 2ib
0 0
For exact calculations, the polarization state of the primary radiation and the propagation
of the refracted beam as the so-called inhomogeneous wave have to be considered. It
appears, however, that the effects for MoKa primary radiation and a quartz reflector are
negligible and can be ignored in the calculation of reflection (R ) and transmission (T )
coefficients as well as of refraction angles jT :
Figure 3 The incident (I0), the reflected (IR), and the refracted (transmitted, IT) beams at the
interface of two media. The refraction index of the medium from the radiation comes (usually air or
vacuum) is greater than the one of the reflector.
j0
x0 ¼ ð10bÞ
jcrit
b
y¼ ð10cÞ
d
The reflection and transmission coefficients fulfill the following condition (as a direct
consequence of energy conservation):
RþT1 ð11Þ
The refraction angle jT, given by
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi
pffiffiffi h i
2
jT ¼ d x20 1 þy2 þ x20 1 ð12Þ
is a function of the incidence angle and coincides with it for the angles well above the
critical angle. The refraction angle is very small, but does not vanish for an (hypothetical)
incidence parallel to the interface ðj0 ¼ 0Þ (see Fig. 4).
Figure 4 The refraction angle jT as a function of the incident angle j0 for MoKa and a quartz
reflector. The critical angle for this configuration is at 1.8 mrad. Even for very small incident angles
close to 0 , the refraction angle is 5 mrad.
Figure 5 Calculated reflectivities of several reflector materials for MoKa radiation. The values of
the critical angles jcrit for C, Si, Ge, Nb, and Ta (1.75, 1.8, 2.5, 3.1, and 4.2 mrad, respectively) are
indicated by circles.
Figure 6 The interference (standing wave) zone between the incident (I0) and reflected (IR) plane
waves with wavelength l shows nodes and antinodes with a period of D.
Figure 7 Relative intensities of the standing waves for the incident angles of 0.9, 1.8, (jcrit), and
3.6 mrad (primary radiation: MoKa, silicon substrate). The standing-wave effect above the
interface increases with decreasing incident angle. The dependence of the distance between maxima
D on the incident angle [see Eq. (16)] is also seen. Inside the medium, the refraction angle is
responsible for the intensity decrease in the direction normal to the surface; the propagation is very
shallow below the surface for small incident angles.
The attenuation of the refracted beam inside the substrate on the way to the location
where an interaction takes place may not be neglected. However, the
attenuation from this point to the surface toward the detector can usually be
ignored.
For the incident angles larger than the critical angle, the geometry factor is responsible for
the increase in the spectral background, whereas the sudden reduction for the smaller
angles originates from the almost perfect total reflection; the reflection coefficient is nearly
1 in this range. (See Figure 9.)
III. INSTRUMENTATION
A. Excitation Sources for TXRF
The usual excitation source for TXRF is a high-power diffraction x-ray tube with a Mo
anode with an electrical power of 2–3 kW. This type of x-ray tube is also available with Cr,
Cu, Ag, and W targets. The line focus of the anode has to be used so that the emitted
brilliance is in correlation with the slit collimation necessary to produce a narrow beam
with the divergence less than the critical angles involved. A higher photon flux on the
sample can be achieved by using rotating anodes, which can stand up to 18 kW. In all
cases, the focal size of the electron beam on the anode is a line with the dimensions of
0.468 mm2 (fine focus) or 0.4612 mm2 (long fine focus). The emission of the x-rays is
observed under the angle of 6 to the anode surface, so that the width of the focus is
reduced optically by the projection with sin 6 ( ¼ 0.1) to 0.04 mm.
Copyright © 2002 Marcel Dekker, Inc.
Figure 9 The angular dependence of the background intensity for TXRF measurements (MoKa
primary radiation under total reflection on a quartz reflector; the critical angle is 1.8 mrad). The
dashed line represents the case where no total reflection is observed.
Monochromators also can modify the primary radiation and they are usually set to
the energy of the most intense characteristic line of the anode material. For a Mo-anode
x-ray tube MoKa or for a W-anode x-ray tube WLb are selected, but a part of the
continuum can be monochromatized as well. Commonly used crystal monochromators
Copyright © 2002 Marcel Dekker, Inc.
Figure 12 (a) Scattered spectrum of a Mo x-ray tube (40 kV acceleration voltage) as used in a
TXRF module spectrometer; (b) scattered spectrum, under the identical conditions as above, but
with a cutoff reflector inserted—adjusted to suppress primary photons between 20 and 40 keV.
(c) scattered spectrum, under the identical conditions as in (a) and (b), but with the cutoff reflector
slightly de-adjusted. As a consequence, the MoKb line is also suppressed.
have the disadvantage of a very narrow energy band transmitted (usually in the range of
few electron volts), whereas synthetic multilayer structures are characterized by higher
DE=E and reflectivities of up to 75% for premium quality materials. (See Fig. 15.)
The excitation of an element is the most efficient when the energy of the exciting
radiation is close to its absorption edge. The search for anode materials for special ap-
plications has led to the development of Al- or Si-anode x-ray tubes which are suitable for
the determination of elements below Si, such as F, Na, Mg, and Al.
A spectrometer has been developed with a variable double multilayer pair used as a
tunable monochromator combined with an x-ray tube with an anode made of a homo-
geneous alloy of molybdenum and tungsten (Knoth et al., 1997). Three discretely ad-
justable excitation energies (9.7, 17.5, and 35 keV) are supplied by this combination and
cover a much wider range of the elements than the detectable with a single spectrometer.
The detection limits were determined to be 0.4 pg for Ni using WLb excitation, 0.6 pg
for Pb with MoKa excitation, and 8 pg for Cd (determined by using its K lines) with the
35-keV bandpass excitation.
The applicability of three configurations of curved multilayer mirrors as mono-
chromatizing and focusing elements has been studied theoretically (Stoev et al., 1998) and
the predicted theoretical results were compared with those obtained with a prototype
spectrometer (tunable between 9.7 and 17.5 keV, with the above-mentioned W=Mo alloy
anode) for the element Ni (Knoth et al., 1999).
Currently, synchrotron radiation is the most brilliant photon beam for XRF and it
has almost ideal features for TXRF. It is several orders of magnitude brighter than the
output of x-ray tubes and is naturally collimated and linearly polarized in the plane of
the orbit of the electrons of positrons. The spectral distribution is continuous; therefore,
Copyright © 2002 Marcel Dekker, Inc.
Figure13 Spectrum of a multielement standard (25 ng of each element) prepared as residue sample
on a quartz sample reflector (well-adjusted cutoff reflector). Conditions: TXRF module, Mo tube,
50 kV, 40 mA, 100 s measuring time. SSr and DLSr are the sensitivity factor and detection limit for Sr,
respectively.
B. Sample Reflectors
For the trace analysis of granular residues, a carrier with high reflectivity that serves as a
totally reflecting sample support is required. Therefore, the mean roughness should be in
the range of only a few nanometers and the overall flatness should typically be less than
l=20 (l ¼ 589 nm, the mean wavelength of the visible light). Furthermore, reflectors
should be free of impurities so that the blank spectrum should be free from con-
tamination peaks and the carrier material must not have fluorescence peaks in the
spectral region of interest. In addition, the carrier material must be chemically inert (also
against strong chemicals, which are often used for the sample preparation), easy to be
cleaned for repeated use. They should be commercially available and inexpensive.
Typically, they are disk shaped with a 30-mm diameter and a thickness of 3–5 mm. The
carrier materials commonly used are quartz glass in the synthetic form as Synsil, or
Suprasil, fused silica, and elemental Si. Heavier elements like Ti, Nb, and Ta were
Copyright © 2002 Marcel Dekker, Inc.
Figure 14 Experimentally determined sensitivities (total) for the K lines of several elements for a
TXRF spectrometer with a cutoff filter. The excitation can be split (theoretically) into the
contributions from the characteristic MoK lines (MoKa Kb) and bremsstrahlung (continuum). The
ratio of the two sensitivity factors depends, among others, on the applied high voltage (in this case
45 kV). Zr and Nb (Z ¼ 40, 41) have the K absorption edges already above the energy of MoKa.
recently tested and showed a good performance. Boron nitride, glassy carbon, Plexiglas,
and Perspex (for single use) are suitable (e.g., for the determination of light elements).
All components of a TXRF spectrometer need to be adjusted (at least once) by
translation and rotation stages and tilters, either manually, or by using remote controllers.
To visualize the effect of total reflection, one can use a proper charge-coupled detector
(CCD) camera where the x-rays are falling directly on the chip and become visible on a
monitor. In a more traditional approach, the x-rays can be observed on a ZnS screen.
C. Detectors
Total-reflection XRF is an energy-dispersive XRF method (see also Chapter 3); the
radiation is measured mainly by Si(Li)-detectors. A good detector offers a high-energy
resolution [Full width at half-maximum (FWHM) in the range of 140 eV at 5.89 keV],
intrinsic efficiency close to 1 for the x-ray lines of interest, symmetric peak shapes, and low
contribution to the background. Primarily, incomplete charge collection at the electrodes
leads to low-energy tailing. The detector escape effect creates escape peaks and thus an
increased background in certain spectral regions. An inherent advantage of semiconductor
detectors is the possibility of bringing the detector crystal very close to the sample, which
results in a large solid angle. Light elements emit fluorescent lines in the range from 100
to 1000 eV. The usually used Be entrance window would completely absorb them, so new
window materials, offering better transmission characteristics, are used instead.
Copyright © 2002 Marcel Dekker, Inc.
Figure15 Two spectra of a sample with a total mass of 700 ng, containing Mg, Cl, Sc, and so forth
excited with a Cr tube, obtained with a cutoff reflector inserted into the beam path (Cut-off), as a
spectrum-modifying device, and a multilayer monochromator (ML). The spectral background and
the characteristic lines are smaller for the latter case; also the scattered CrKb line is suppressed.
In 1971, Yoneda and Horiuchi (1971) published an article with the first experimental
data from an energy-dispersive x-ray fluorescence measurement with a Ge detector in
Copyright © 2002 Marcel Dekker, Inc.
total-reflection geometry. The sample was a dried spot of 100 ng of Ni from a water
solution of NiCl2 on an ‘‘optical flat.’’ In 1974, Aiginger and Wobrauschek (1974)
published results using a Si(Li) detector and a Suprasil reflector where 5 mL of Cr salts in
aqueous solution were dried and analyzed with a Cu-anode x-ray tube. A more detailed
publication (Wobrauschek and Aiginger, 1975) contained theoretical considerations,
details of the setup, calibration curves, steps for the quantification, and detection limits
of 4 ng for the element Cr. In 1977, Knoth et al. (1977) presented their first prototype of
a total-reflection XRF spectrometer and analytical results for a blood serum analysis.
From that time on, several groups worldwide participated in the further development of
this technique. From 1986 on, regular biannual TXRF user meetings have been held
(Proceedings of the TXRF Conference, 1989, 1991, 1993, 1995, 1997). The first book
dealing exclusively with TXRF was written by Klockenkämper (1997).
Figure 16 Twofold excitation of the sample for TXRF by the primary beam (1) and the reflected
beam (2), for a case where the reflector is long enough so that beam (2) is observed.
Figure 17 Spectrum of an aerosol sample, which was collected in a Batelle impactor (stage 1)
directly on a polycarbonate sample carrier. On top 30 ng of Sc were pipetted as internal standard
element (Cr anode, 30 kV, 30 mA, multilayer monochromator, 500 s measuring time).
Preparation
Chemical
Freeze- matrix Open Pressure Freeze-
Sample Drying drying separation digestion Ashing Suspension Solution digestion cutting
Rain, river water 0.1–3 ng=mL 20–100 pg=mL 3–20 pg=mL 1–3 ng=mL
Blood, serum Digestion: 20–80 ng=mL 40–220 ng=mL
2–30 ng=mL
Air dust, ash, 5–200 mg=g 10–100 mg=g 0.1–3 mg=g
aerosols
Air dust on filter 0.6–20 ng=cm2 0.2–6 ng=cm2
Suspended matter 3–25 mg=g 10–100 mg=g
Sediment 10–100 mg=g 15–300 mg=g
Powdered 1–10 mg=g 0.2–2 mg=g
biomaterial
Fine roots 1–10 mg=g Digestion:
0.1–1 mg=g
High-purity acids 5–50 pg=mL
Tissue, foodstuff, 0.5–5 mg=g
biomaterial
Mineral oil 1–15 mg=g
Mussel, fish 0.1–1 mg=g
High-purity water 1 pg=mL
Source: Data from Klockenkämper, 1997.
D. Some Applications
Three main advantages characterize TXRF: simultaneous multielement capability, low
detection limits for many elements, and small sample volume. Additional advantages are
the absence of matrix effects, easy calibration, fast analysis, and comparatively low costs.
Table 2 gives an overview of various kinds of sample that have been already analyzed with
TXRF. Generally all kinds of aqueous or acidic liquids where the liquid matrix can be
evaporated, leaving a small amount on a sample reflector, can be analyzed. Oils, alcohols,
whole blood, and blood serum can be analyzed after special treatment.
Typical for TXRF are water samples (Fig. 19). Pure water like rainwater or tap
water can be directly analyzed (Holynska et al., 1998; Barreiros et al., 1997); for riverwater
or seawater as well as wastewater, sample preparation is usually required to remove
Environment
Water: Rain water, river water, sea water, drinking water, waste water
Air: Aerosols, airborne particles, dust, fly ash
Soil: Sediments, sewage sludge
Plant material: Algae, hay, leaves, lichen, moss, needles, roots, wood
Foodstuff: Fish, flour, fruits, crab, mussel, mushrooms, nuts, vegetables, wine, tea
Various: Coal, peat
Medicine=biology=pharmacology
Body fluids: Blood, serum, urine, amniotic fluid
Tissue: Hair, kidney, liver, lung, nails, stomach, colon
Various: Enzymes, polysaccharides, glucose, proteins, cosmetics, biofilms
Industrial=technical
Surface analysis: Water surfaces
Implanted ions
Thin films
Oil: Crude oil, fuel oil, grease
Chemicals: Acids, bases, salts, solvents
Fusion=fission research: Transmutational elements Al þ Cu, iodine in water
Geology=mineralogy
Ores, rocks, minerals, rare earth elements
Fine arts=archeology=forensic
Pigments, paintings, varnish
Bronzes, pottery, jewelry
Textile fibres, glass, cognac, dollar bills, gunshot residue, drugs, tapes, sperm, finger-prints
suspended matter or the salt content. Detection limits are in the low nanograms per
milliliter range. To reduce the detection limits down to the 10-pg=ml level, freeze-drying
with additional leaching of the residue with HNO3 has to be applied. To remove the salt
matrix from seawater, complexation, chromatographic adsorption, and subsequent elu-
tion have to be performed. This method was developed by Prange et al. (1985). Suspended
matter separated from filtrates can be analyzed by digesting the filter with HNO3 (Prange
et al., 1993). For the analysis of sediments, microwave digestion is recommended
(Koopmann and Prange, 1991).
Environmental monitoring can also be done by analyzing appropriate plants like
moss or lichen (biomonitoring). The analysis of plants requires a proper sample pre-
paration. After cleaning, shredding, freeze-drying, and finally pulverizing, the powder is
mixed with HNO3 and digested in a Teflon bomb.
Vegetable oil is recommended to be diluted with toluene and an oil-based standard
should be added as internal standard (Reus, 1991). After pipetting an aliquot on the
sample reflector, it is heated up to evaporate the volatile parts of the matrix. Detection
limits are in the range of 3–20 ng=g. Also, low-temperature oxygen plasma ashing is
suitable; the resulting residue has to be dissolved in HNO3.
Both depletion and accumulation of trace elements can influence the biological
functions of human beings. Whole blood and blood serum, amniotic fluid, organ tissue,
hair, and dental plaque have been analyzed by TXRF. Detection limits down to 20 ng=mL
for body fluids after microwave digestion (Prange et al., 1989) were reported.
Quality control of ultrapure reagents, like acids, bases and solvents can be performed
by using TXRF (Prange et al., 1991). Crude oils (Ojeda et al., 1993), lubricating oils
(Bilbrey et al., 1987), motor oils (Freitag et al., 1989), and diesel fuel (Yap et al., 1988)
have also been analyzed. Light oils can be diluted with chloroform or toluene, the volatile
Copyright © 2002 Marcel Dekker, Inc.
matrix then removed by evaporation, and detection limits in the range of micrograms per
milliliter are achievable. If detection limits of nanograms per milliliter are required, low-
temperature oxygen plasma ashing is necessary. For forensic applications the micro-
analytical capability of TXRF is appreciated, because microsamples are usually available.
In recent years, several specific problems have been investigated, including analysis of hair
samples, glass particles, tape fragments, drug powder, semen traces, gunshot residues, and
textile fibers (Ninomiya et al., 1995; Prange et al., 1995). Oil paints were investigated by
analyzing the pigments, which are characterized by a limited number of major elements.
Cotton tips can be used to remove an amount of about 1 mg from the paint and deposit it
on a sample reflector (Klockenkämper et al., 1993).
E. Quantification
One of the inherent advantages of TXRF is the fact that the sample forms a thin film on
the sample reflector; thus, no matrix effects have to be considered and the so-called thin-
film approximation is applicable (both absorption and enhancement effects can be neg-
lected). The intensity of the fluorescent radiation for, the Ka line, for example, of an
element i with concentration ci in the sample (with mass m ) is then given by
ZEmax
m i
IðEiKa Þ ¼ I0 ðEÞG s ðEÞci f ðEiKa ÞeðEiKa Þ dE ð20Þ
sin j Ka
E¼Eiabs
where Eabs is the energy of the absorption edge of element i, Emax is the maximum energy of
the excitation spectrum, EKa i
is the energy of the Ka line of element i, I0 ðEÞ is the spectral
distribution of the exciting radiation, G is the geometry factor, siKa is the fluorescence cross
section for the K shell of element i, fðEKa i
Þ is the absorption factor for the fluorescence
radiation between the sample and the detector, and eðEKa i
Þ is the relative detector efficiency
i
for the energy EKa . It is assumed that the sample is always completely irradiated by the
primary radiation.
In this special case, the relation between concentration and fluorescence intensity is
linear. The so called sensitivity [counts=(second)(sample mass)] can be defined:
Ii
Si ¼ ð21Þ
ci m
and Si depends only on fundamental parameters and the measuring conditions, which
usually can be assumed to be constant. When an element St is used as the internal standard
the relative sensitivity (sensitivity factors) for elements i defined as
R Emax
fðEiKa ÞeðEKa
i
Þ E¼E i I0 ðEÞsiKa ðEÞ dE
Srel ¼
i
R Emax
abs
ð22Þ
Ka ÞeðEKa Þ E¼ESt I0 ðEÞsKa ðEÞ dE
fðESt St St
abs
can be established experimentally with artificially prepared standards (Fig. 20) or calcu-
lated theoretically. The determination of the concentration ci of the element i in an un-
known sample spiked with the same internal standard element is then simple:
Ii 1
ci ¼ cSt ð23Þ
ISt Sirel
There is a linear correlation between intensity Ii and concentration ci . The addition of
one element as the internal standard of known concentration to the sample is necessary
Copyright © 2002 Marcel Dekker, Inc.
Figure 20 Fitted calibration curve (relative sensitivities) for K lines (Atominstitut TXRF module,
Mo anode) with Ga as internal standard element. The circles represent the measured elements
actually used for the calibration (standard solutions with various concentrations). Other elements in
a sample can also be determined by interpolation (in this case, S, K, Mn, Br, and Rb).
primarily because of the inhomogeneous excitation due to the standing-wave field above
the sample reflector surface, where the sample is positioned. Also, geometric and vol-
umetric errors can be canceled.
A sample is ‘‘thin’’ if its thickness does not exceed the critical thickness
(Klockenkämper, 1997), which is about 4 mm for organic tissue, 0.7 mm for mineral
powders, and 0.01 mm for metallic smears. Under the assumption that the matrix ab-
sorption for the analyte differs only slightly from that of the internal standard element,
these values can generally be higher by a factor of 10–400. For the calculation of these
values, the standing-wave field was not taken into account. This effect and the sample self-
absorption can lead to contradictory requirements for the sample thickness (de Boer,
1991a).
Figure 21 shows a comparison of detection limits for various analytical methods in
the trace element range, following the work of Klockenkämper (1997). Inductively coupled
plasma–mass spectrometry (ICP–MS) and INAA are more macro than micro methods. In
comparison, TXRF requires only very small sample volumes (in the range of microliters).
ICP–MS provides lower detection limits, but both spectral interferences and matrix effects
make quantification more complex.
V. SURFACE ANALYSIS
Figure 22 Wafer mapping for the element Na of 7610 spots (1-mm scan steps, relative
coordinates). Synchrotron radiation excitation (SSRL, Beam line III-4), measuring time for each
pixel: 10 s.
Figure 23 Typical components of a TXRF spectrometer for wafer analysis with all necessary
degrees of freedom for surface mapping (x and y movement), height control (z movement), and
rotation around the x axis (angle of incidence). The special detector and the evacuable sample
chamber allow also the detection of low-Z elements.
A. Quantification
For the case of granular residues on a substrate (particulate type), which is equivalent to
chemical analysis by using TXRF, the intensity above the critical angle is constant because
the thin, small ‘‘sample’’ is completely excited. The intensity doubles at the critical angle in
a steplike fashion and remains at the twofold value down to very small angles due to total
reflection,
Iiparticle ðj0 Þ ¼ kparticle I0 ci ½1 þ Rðj0 Þ ð24Þ
following the angular behavior of the reflection coefficient Rðj0 Þ. This intensity is pro-
portional to the primary intensity I0 and the interesting area-related concentration ci. For
incidence angles j0 below jcrit , Eq. (24) can be simplified to
Iiparticle ffi 2 kparticle I0 ci ð24aÞ
For buried layers (film like) in a substrate, the intensity far above the critical angle becomes
constant. The asymptotic behavior of the intensities for the particulate and filmlike con-
taminations can even be equal if both concentration values are the same and the appro-
priate scaling factors kparticle and kfilm are chosen. However, the intensity for the buried
layers steadily increases with decreasing angle and can reach (theoretically) the fourfold
value at the critical angle. For the smaller incidence angles, the intensity is going to zero,
according to
j
Iifilm ðj0 Þ ¼ kfilm I0 ci ½1 Rðj0 Þ 0
zP
j0
¼ k I0 ci Tðj0 Þ
film
rtm ð25Þ
jT ðj0 Þ
A. Depth Profiles
The fluorescence intensity Ii ðj0 Þ of an element i, implanted in a flat smooth substrate, with
a depth profile ci ðzÞ showing a variation of its concentration over depth z (normal to the
surface), is a function of the incident angle j0 :
Z1
j0 m1
Ii ðj0 Þ / Tðj0 Þ ci ðzÞ exp þ m2 rz r dz ð27Þ
jT ðj0 Þ jT ðj0 Þ
z¼0
which is the generalized form of Eq. (26). The mass-attenuation coefficients m1 and m2
describe the absorption of the exciting and the fluorescence radiation of the element i,
respectively; r is the density of the substrate. The transmission coefficient Tðj0 Þ takes into
account the refracted part of the incoming radiation and jT is the refraction angle inside
the substrate. Examples are given for Ni (Fig. 25) and Al (Fig 26) depth profiles.
In practice, the general shape of a depth profile has to be given and some parameters
can be fitted (Kregsamer et al., 1999) or different functions can be tested (Weisbrod et al.,
1991; Schwenke et al., 1997; Mori et al., 1997), although, theoretically, a Laplace trans-
formation is possible.
B. Thin Films
The composition, density, and thickness of single as well as periodic and nonperiodic
multiple films can be analyzed (Weisbrod et al., 1991; Schwenke et al., 1992). The fluor-
escence intensity Ii ðj0 Þ of an element i (with concentration ci ) embedded in a single thin
film with thickness d on top of a flat smooth substrate, is a function of the incident angle
j0 and is described by (de Boer, 1991a, b)
The electric fields ET (transmitted by the film surface) and ER (reflected from the substrate)
are derived from Fresnel’s equations and are complex quantities. The wavelength of the
exciting radiation is l. The formation of standing wave above a reflecting surface is al-
ready taken into account by this formalism. An example is given for a Ni thin film in
Figure 27.
The layer(s) and the substrate must be homogeneous, flat, and smooth, with perfectly
sharp interfaces. With some efforts, even a certain roughness of the interfaces can be taken
into account in the calculations (van den Hoogenhof and de Boer, 1993; Nevot and Croce,
1980). A review on grazing incidence X-ray spectrometry and reflectometry had been given
by Stoev and Sakurai (1999).
There are different strategies to improve the detection limits, either by increasing the
sensitivity, by reduction of the spectral background, or extending the measuring time
(which, however, is limited for practical reasons).
One approach to reducing the spectral background it to use total-reflection geo-
metry. A further possibility for reducing the background is the use of polarized primary
radiation (Ryon and Zahrt, 1993; Aiginger and Wobrauschek, 1974). Due to the ani-
sotropy in the intensity of scattered radiation, it is advantageous to place a detector in
such a position that almost only the isotropic emission of the fluorescence signal is
detected. It is possible to combine TXRF with polarized primary radiation excitation.
Moreover, the use of monochromatic primary radiation improves the background
conditions, because only photons with one energy are scattered (incomplete charge
Copyright © 2002 Marcel Dekker, Inc.
Figure 27 Calculated NiKa fluorescence intensities as a function of the incidence angle for Ni
films on quartz substrate, excited by MoKa (normalized to values at 10 mrad). For a thickness
around 1 nm, the maximum of the distribution is found at the critical angle of quartz (1.8 mrad). For
around 25 nm film thickness, the maximum position coincides with the critical angle of nickel
(3.6 mrad).
collection still causes a background). Tunable monochromators enable the exciting en-
ergy to be adjusted slightly above the absorption edge of the element of interest (‘‘se-
lective excitation’’). Sometimes, it is even possible to set this energy below the absorption
edge of a matrix element with high concentration, with the drawback, however, of an
increased background due to Raman scattering if it is too close to this edge.
Highly intensive broad-band photon sources realized in the synchrotron radiation
facilities can increase the sensitivity by orders of magnitude as compared to conventional
x-ray tubes. Its outstanding properties offer new possibilities for TXRF. The intense
primary beam with a continuous spectral distribution from infrared to high-energy pho-
tons, the linear polarization in the orbit plane, the small source size, and its natural col-
limation are features best suited for the excitation in total-reflection geometry. For
optimum excitation conditions, the spectral distribution can be modified by using elements
like cutoff mirrors, crystal and multilayer monochromators and filters, and so forth
(Wobrauschek et al., 1997a, b).
Multilayer monochromators are well suited for synchrotron radiation-excited XRF
(Bilderback, 1982; Pianetta and Barbee, 1988). In comparison to crystal monochromators,
they offer a larger bandwidth (DE=E 0.01), which leads to higher photon flux on the
sample. The combination of TXRF with synchrotron radiation can be achieved by several
geometrical arrangements for the entrance slit, sample reflector, and detector. Figure 28
shows three possibilities. For the uppermost geometry (Fig. 28a), the polarization effect is
fully utilized by positioning the detector axis in the plane of the orbit. Scattered radiation
is not emitted in that direction. The beam efficiently and homogeneously illuminates the
Copyright © 2002 Marcel Dekker, Inc.
Figure 28 Three possible geometrical arrangements of synchrotron radiation-excited TXRF.
Figure 29 Experimental setup for TXRF experiments at HASYLAB, utilizing the radiation from
the DORIS III positron storage ring. Several slits and diaphragms and beam stoppers are used to
prevent the count-rate saturation of the detection system by direct and scattered radiation from
primary beam, transmitted photons through the monochromator and so on.
The absorption coefficient drops steadily as the energy E increases above an element’s
absorption edge. Therefore, (diffraction) x-ray tubes with standard anode materials (Sc,
Cr, Cu, etc.) are poor exciters for light elements, as their characteristic emissions are far
above the absorption edges of the elements of interest. To improve the sensitivity for
light elements, it is necessary to use either an x-ray tube which emits intensive char-
acteristic radiation with an energy as close as possible to the absorption edge of the
interesting element, or synchrotron radiation with its continuous spectral range down to
low energies.
The spectrometer used for the experiments described in Sec. VIII is characterized
in detail in the works of Streli et al. (1993a, b). A special Ge(HP) detector which meets
all the requirements for low-energy detection is used. It has a thin entrance window
(diamond window, 0.4 mm thick, transmission for oxygen Ka ¼ 85%), an ion-implanted
contact layer, a thin dead layer, and a low electronic noise contribution (FWHM at
5.9 keV ¼ 125 eV). Measurements were performed under vacuum conditions to reduce
absorption. The radiation of a Cr x-ray tube (1.3 kW) was monochromatized with a
W=C multilayer. Details are described in the work of Streli et al. (1995).
To improve the excitation conditions, a windowless x-ray tube was used. Recently,
various anode materials were tested after computational optimization of the takeoff
angle of the x-ray tube to obtain a high photon flux on the sample. The calculation takes
into account the required brilliance and the self-attenuation of the emitted photons in
the anode. Al, Si (Fig. 32), and Mo were tested experimentally as anode materials.
Si offers the advantage of a high t value for Al, but a low one for Si itself, which is often
the major constituent of the reflector substrate. For the analysis of impurities on
Si-wafer surfaces, this is a great advantage, because the high fluorescence intensity of
Copyright © 2002 Marcel Dekker, Inc.
Figure 32 Spectrum of 4.5 ng of Na, Mg, and Al each. The sample was measured in an evacuated
TXRF spectrometer with a Si-anode windowless x-ray tube (20 kV, 25 mA, 200 s measuring time),
cutoff reflector, and a Ge(HP) detector with a diamond window.
Si is reduced (Streli et al., 1997a, b). Detection limits obtained differ very much with the
excitation conditions. The standard Cr x-ray tube provides about 90 pg for Mg, with the
windowless Si-anode tube, 7 pg were obtained.
Synchrotron radiation as an excitation source offers the advantage of an intensive,
naturally collimated x-ray beam of a wide spectral range, also in the low-energy region; it
is, therefore, the ideal source for light element’s determination by TXRF. Experiments
have been performed at SSRL, Beamline III-4. This bending magnet beam line is equipped
with a horizontally deflecting torroidal Au mirror to cut off photons with an energy higher
than 3 keV. A differential pumping system offers the possibility of connecting the mea-
suring chamber directly, without any window. Various filters and a double multilayer
monochromator can be inserted into the beam path. The evacuable sample chamber was
connected to the monochromator chamber with a flexible coupling and an interlock valve
that closes the main valve upstream if the pressure exceeds 10 7 3 mbar. Due to the down-
looking design of the used detector, the advantage of the linear polarization of synchro-
tron radiation for further background suppression could not be utilized (Streli et al., 1994,
1997a, b). In order to adapt the spectral distribution of the synchrotron radiation for the
analysis of Si-wafer surfaces, a 12-mm-thick Si filter was inserted into the beam path,
leading to ‘‘quasimonochromatic’’ radiation with a bandwidth of about 400 eV ( just below
the absorption edge of Si).
A promising application of low-Z TXRF is the quality control of Si-wafer sur-
faces. For the semiconductor industry, Al and Na are of special interest, because they
influence the production yield of ICs negatively. It is important to have an analytical
method sensitive enough for the determination of ultratraces of these elements. Droplet
samples of 100 pg of Mg had been prepared on Si-wafer surfaces and analyzed with
synchrotron radiation excitation. The best results were obtained for the ‘‘quasimono-
chromatic’’ mode with detection limits of 60 fg for Mg (Fig. 33) (Streli et al., 1997a, b).
Copyright © 2002 Marcel Dekker, Inc.
Figure 33 Spectrum of a 100-pg Mg sample on a Si wafer. Measured at SSRL, Beamline III-4
with 12-mm-thick Si filter inserted into the primary beam path (ring current 54 mA, measuring time
100 s). The extrapolated detection limits are 60 fg. Na and Al were contaminations.
Depth profile and thin-film analysis of low-Z elements, usually not detectable by
conventional instruments, can be performed by synchrotron radiation excitation. An
angular scan for a thin film of carbon (25 nm) on silicon demonstrates the power of
synchrotron-radiation-excitation XRF in total-reflection geometry—even for low-energy
fluorescent lines (Fig. 34).
Currently, there is only one TXRF spectrometer commercially available for the
determination of Na and Al, with a rotating W-anode x-ray tube (excitation by WM lines);
detection limits of 861010 and 361011 atoms=cm2 for Na and Al, respectively, were ob-
tained (Fukuda et al., 1997).
REFERENCES
Joachim Heckel
Spectro Analytical Instruments, GmbH & Co. KG, Kleve, Germany
Richard W. Ryon
Lawrence Livermore National Laboratory, Livermore, California
I. INTRODUCTION
In this chapter, we explore the sources of spectral background and show how polarization
effects can yield improved detection thresholds. In 1963, Champion and Whittem pointed
out that the detection of Co traces in aqueous solution can be improved significantly
due to polarization of the background in a wavelength-dispersive x-ray fluorescence
(WDXRF) setup. We see that the stationary arrangement of components used in energy-
dispersive x-ray fluorescence (EDXRF) is ideally suited for geometrical configurations that
exploit polarization phenomena to reduce background and thereby improve signal-to-
noise ratios. The ability of simultaneously measure a wide range of elements is one of the
greatest advantages of EDXRF. This advantage is strongly reduced when the count-rate
limitation of the ED detection electronics is taken into consideration. This is due to si-
multaneous recording of the entire primary source radiation scattered on the specimen and
is especially true for examinations on samples with light matrices. Thus, the main purpose
of this chapter is to show that the multiple-element character of EDXRF can be ad-
vantageously utilized through the application of polarized x-rays.
Observed spectral background is caused by several interactions of radiation with
system components, the specimen, and the detector. A principal cause of background is the
scatter of source radiation by the specimen into the detector. The scattered radiation adds
directly to the background under analyte lines when broad-band primary radiation is used to
excite fluorescence. Even when monochromatic radiation is used for excitation, the scatter of
this primary radiation adds indirectly to the background because of incomplete charge
collection in the detector. In addition, primary radiation carries with it fluorescence from the
x-ray tube anode contaminants and collimator materials, and the scattered radiation causes
fluorescence of collimator materials between the specimen and the detector, and the gold
contact layer on the surface of the detector. Scattered radiation also causes low-energy
background due to the residual electron kinetic energy when Compton scatter from the
detector itself occurs. All of these sources of spectral noise and interference can be reduced by
minimizing the scatter of source radiation into the detector by using polarized radiation to
excite fluorescence. The most important advantage of this technique in trace element analysis
II. THEORY
A. Scattering of Nonpolarized Radiation
A decade after the discovery of x-rays by Roentgen in 1895, Barkla (1905) demonstrated
that this newly discovered radiation could be polarized by scattering, thus supporting the
hypothesis that x-rays are electromagnetic radiation (i.e., their wave nature). A few years
later, Friedrich et al. (1913) demonstrated x-ray diffraction, which also substantiated the
electromagnetic hypothesis.
Thomson and Thomson (1933) found, based on the classical electromagnetic theory,
that the intensity of radiation scattered by a free electron is
2 2
1 e
IS ¼ I0 2 sin2 a ð5Þ
r m0 c2
where I0 is the intensity of the incident beam at the x-ray tube window, IS is the intensity
of the scattered beam at distance r, e is the charge of the electron ð1:6 1019 CÞ, m0 is the
rest mass of the electron ð9:11 1031 kgÞ, c is the speed of light ð3:00 108 m=sÞ, r is the
distance to the point of observation (m), and a is the angle between the scattering electron
~
S (Pointing vector) and the direction of acceleration of the electron. The acceleration
vector is perpendicular to the direction of propagation of the incident radiation ~ S0 and
parallel to the electric field vector ~E0 of the incident radiation. If ~
S0 is in the z direction
and ~S is in the xz plane, the situation is as shown in Figure 1. For standard x-ray sources,
such as x-ray tubes, ~ E0 of the photons incident on the scattering electron is random in
direction (but always perpendicular to the direction of propagation) so that on the
average
hE20;x i ¼ hE20;y i ¼ 12E20 ð6Þ
To obtain the scattered intensity at a general scattering angle Y in the xz plane, it is
simply necessary to decompose the incident ~ E0 into components and sum their individual
intensity contributions. Now, ~ E0;y causes the electron to oscillate in the y direction so that
the angle a ¼ p=2 and sin2 a ¼ 1. ~ E0;x causes the electron to vibrate in the x direction,
so the angle a ¼ p=2 þ Y and sin2 a ¼ cos2 Y, where Y is the indicated scattering angle.
Combining these considerations of the dependencies of ~ E0;x and ~ E0;y on the scattering
angle a and recalling that the intensity of a wave is equal to the squares of its amplitude
ðI ¼ E20 ¼ E20;x þ E20;y Þ, the following result is obtained from Eqs. (5) and (6):
2 2
1 e
IS ¼ I0 2 ð1 þ cos2 YÞ ð7Þ
2r m0 c2
From small-angle solid geometry, the quantity r2 in Eq. (7) is equal to dA=dO, where dA is
the surface area of the expanding wave front of the solid angle dO at a distance r from the
point of scatter. Therefore,
IS r2 dse
dA ¼ e ð1 þ cos2 YÞ dO ¼ dO ð8Þ
I0 2 dO
If the intensity of the scattered radiation is summed over the surface area dA, the fraction
of the incident photon flux, which is scattered by the electron into the solid angle dO in the
direction Y, where re ¼ e2 =m0 c2 ¼ 2:82 1015 m is defined as the classical electron radius
and dse =dO as the differential cross section for scattering by a single electron,
dse r2e
¼ ð1 þ cos2 YÞ ð9Þ
dO 2
This type of scattering is often referred to as Rayleigh scattering but is more appropriately
called Thomson Scattering. Note that, relative to the core atomic electrons, the valence
electrons are ‘‘free’’ in the Thomson sense. Corrections for electrons that are neither core
nor valence require the atomic structure factor.
Copyright © 2002 Marcel Dekker, Inc.
Compton scattering is often taken to have the same cross section as Thomson
scattering, but Thomson cross sections are only low-energy approximations. At higher
energies, the inclusion of quantum mechanical considerations leads to the Klein–Nishina
formulation (Klein and Nishina, 1928), presented in Chapter 1, which is repeated here:
!
dsKN r2e 1 þ cos2 Y g2 ð1 cos YÞ2
¼ 1þ ð10Þ
dO 2 ½1 þ gð1 cos YÞ2 ð1 þ cos2 YÞ½1 þ gð1 cos YÞ
where g is the incident photon energy=m0 c2 ¼ 1:96 103 incident photon energy (keV).
Equation (10) presents the collisional cross section, which is also the approach of most
other authors. Because of the decrease in energy upon Compton scattering, the energy
scattered in a particular direction as a fraction of the incident intensity (Evans, 1958)
is needed. Thus, the differential Klein–Nishina scattering cross section is dsSKN =dO ¼
ðn0 =n0 ÞðdsKN =dOÞ. The wavelength shift due to Compton scattering is l0 l ¼ hð1
cos YÞm0 c, converting to frequency by the relationship ln ¼ c; n0 = n ¼ 1=½1þ gð1 cos YÞ
is obtained.
The fractional background reduction, even at a scattering angle of exactly p=2, is not
quite zero because of the Klein–Nishina limitation. The limitation to achievable polar-
ization is significant when exciting K-line fluorescence of heavy elements (Strittmatter,
1982). As g approaches zero in Eq. (10), the Klein–Nishina cross section approaches the
classical Thomson value in Eq. (9). The transition to description of the scattering on
the electron shell of an atom with the atomic number, Z, is carried out by introduction of
the scattering amplitude. If the small Klein–Nishina effect at lower energies is ignored,
it follows that unpolarized incident electromagnetic radiation scattered through an angle
of Y ¼ p=2 is linear ð~ S ¼ S~ex Þ polarized, with only Ey surviving.
where hE20;x i and hE20;y i represent the mean intensity components I0;x and I0;y , respectively,
for all of the observed photons. The dependency of the coherent and incoherent scattering
cross sections for polarized radiation is exploited when using EDPXRF. The angle b (see
Fig. 1 with b ¼ p=2 a) between the electric field vector of the incident radiation ~ E0 and
the electric field vector of the scattered radiation ~ E is introduced to describe the polar-
ization dependence of the scattering cross section. The following is obtained from Eq. (5)
(Hanson, 1986):
dse
¼ r2e cos2 b ð12Þ
dO
The polarization dependency of scattering is determined using cos2 b. This means that the
component E of the electric field vector of the scattered photon is proportional to the
E0 onto the plane which is perpendicular to the direction of propagation ~
projection of ~ S of
the scattered photon
E2 E20 cos2 b ð13Þ
ð~ E; ~
E0 ; ~ SÞ ¼ 0
E0 ~
ð~ S0 Þ ¼ 0 ð14Þ
E ~
ð~ SÞ ¼ 0
The first equation in Eq. (14) is also valid for Compton scattering, as g approaches zero
in Eq. (10). Using Eqs. (13) and (14), the mean electric field vector is defined as
hE2 i ¼ hE2x i þ hE2y i þ hE2z i as a function of E0;x ; E0;y ; j, and d (see Fig. 1) (Brumme,
1990):
When a series approximation of Eq. (17) is conducted around dmax ¼ 0 for a scattering
angle Y ¼ p=2 to the second order in dmax , the following is obtained:
r2e
sO ðP0 ; dmax Þ ¼ pð1 P0 Þd2max ð18Þ
2
drawing on Eq. (18) for the derivation of Eq. (2). For a polarization P0 ¼ 1, which ac-
cording to Eq. (11) is equivalent to ~E0 ¼ E0;x~
ex , and a scattering angle of Y ¼ p=2; there is
no scattering in the x direction. With P0 ¼ 1, which is equivalent to ~ E0 ¼ E0;y~
ey ; scat-
tering in the x direction reaches its maximum. In addition, it follows from Eq. (18) that the
radiation scattered in the solid-angle component O ¼ 2pð1 cos dmax Þ is proportional to
the square of the scattering angle dmax . The scattering angle dmax is defined by the colli-
mator system.
The polarization P of the scattered radiation measured in the xz plane according to
Figure 1 is
When scatter in the solid angle O is also considered, in accordance with the above
observations the following is obtained:
Copyright © 2002 Marcel Dekker, Inc.
d2max
P1 ð20Þ
1 P0
In the ideal case of scatter exactly around p=2 ðdmax ¼ 0Þ, the polarization P of the
scattered radiation is independent of the primary polarization and has a value of unity.
The electric field vector has exactly one component in the y direction. For greater scat-
tering angles (i.e. greater collimator diameters), the polarization is reduced as described in
Eq. (20). The primary polarization has a smaller influence than dmax . The best values of P
are achieved, as expected, with a primary polarization P0 ¼ 1ð~ E0 ¼ E0;y~
ey Þ.
C. Cartesian Geometry
In Barkla’s experiment (Compton and Hagenow, 1924) a source of x-rays is incident on a
scatterer S1, as shown in Figure 2. The scattered rays from S1 intercepted by S2 are nearly
linearly polarized, as the angle S0–S1–S2 is p=2 plane. The electric field vector is perpendi-
cular to the S0–S1–S2 plane. Radiation from S1 scattered by S2 into Iz also undergoes p=2
scattering, but it is in the same plane as the first scattering and no further annihilation takes
place. However, scattering of the radiation from S1 by S2 into Iy (also through p=2) anni-
hilates all remaining radiation. Thus, Iz receives radiation, by Iy does not. This discussion is
idealized; the measured value Iy=Iz depends on the tightness of collimation of the paths S0 to
S1, S1 to S2, S2 to Iz, and S2 to Iy, and also on the materials and thickness of S1 and S2.
For the experimental determination of the degree of polarization (Brumme et al.,
1990) according to Figure 2, it is necessary to realize that the value Peff calculated with
Eq. (21) represents an effective polarization.
Figure 2 Schematic geometry for polarized beam XRF (PXRF). S0 is the x-ray source, S1 is the
polarizer, S2 is the specimen, and I y is the intensity at the detector. I z is the intensity at an alternate,
in plane position for the detector, which does not eliminate source radiation.
Proof-of-principle experiments in the early 1970s (Dzubay et al., 1974; Howell et al., 1975),
using single collimators to define the three orthogonal beams (Fig. 2), demonstrated im-
proved signal-to-noise ratios, but at severely reduced intensities. A few years later, it was
demonstrated (Ryon, 1977) that the way to maximize analyte intensities while minimizing
background is to open the apertures of the collimators until maximum systems throughput
is achieved while maintaining the orthogonal beam geometry. Such nonzero beam
divergence causes a small decrease in the degree of polarization, which is more than
compensated for by the increase in x-ray flux. Detection limits are improved in comparison
"
Figure 3 Comparison of excitation methods. The sample is an oil standard which contains 21
elements (e.g., Ca, Ti, V, Cr, Mn, Fe, Ni, Cu, Zn, Mo, Ag, Cd, Sn, Ba, and Pb) with concentrations
of 30 mg=g. The spectra (log scale) are (top) nonpolarized, direct excitation by radiation from
rhodium anode x-ray tube, (middle) molybdenum secondary excitation, and (bottom) polarized
excitation by scatter from a HOPG=Al2 O3 target. The measuring time amounts to 200 s for each
spectrum. The counting rate was first adjusted by increasing the tube current to its maximum value
and then increasing the collimator aperture to give a total count rate approximately 60,000 cps using
a Si(Li) detector with an active area of 25 mm2 and a energy resolution of 166 eV for MnKa. In all
cases, a Rh end-window tube was used.
A. Scattering Material
The scattering effectiveness, e, of a planar target, S1 (see Fig. 2) is defined as
sS ðEÞ rmðEÞd
eðEÞ ¼ 1 exp ð22Þ
mðEÞ sin 45
where mðEÞ is the mass-attenuation coefficient of the target material for the x-ray energy E
(in cm2=g), sS is the scattering coefficient (coherent þ incoherent) for the target material
for a scattering angle of p=2 and an x-ray energy E (in cm2=g), r is the density of the target
(in g=cm3), and d is the thickness of the target (in cm).
The quotient s=m of scatter and absorption must be a maximum for the scattering
target material. Materials with a very small average atomic number are especially suited
for this. They have a relatively small mass-attenuation coefficient (m < 1 cm3=g) in the
energy ranges of interest, so that the thickness of the scattering target must be increased in
order to obtain a sufficiently high scattering intensity according to Eq. (22). Thick scat-
tering targets enable great variations in the scattering angle, which logically leads to poor
polarization (Table 1). Additionally, the probability of multiple-scattering processes in-
creases by increasing the thickness of the scattering target and, with it, the depolarization
(see Table 1). Polarization losses have been calculated by Zahrt and Ryon (1981).
Table 1 Scattering Target Effectiveness e and Polarization Calculated with Monte Carlo
Simulations Without ðP0 Þ and With ðPMS Þ Consideration of Multiple-Scattering Processes in the
Target as a Function of the Thickness, dT , of a Planar Carbon Target (E ¼ 19:6 keVÞ
Table 3 Influence of the Dimensions of a Quadratic Anode Focal Spot on the Polarization
(Target Thickness: 1 mm)
E. Multiple-Layer Scatterers
As later indicated (see Table 4), if a single element is determined, a monochromatic source
gives the lowest detection limits. However, a polychromatic source is desirable when
multielement analysis is performed. If polychromatic radiation is to be polarized by
scattering, high-Z scatterers absorb the low-energy portion of the incident spectrum,
whereas low-Z scatterers do not interact effectively with the high-energy components of
the spectrum (Zahrt and Ryon, 1986). In both cases, this limits the intensity over a broad
range of energies. In both of the examined energy ranges, E < 25 keV and
25 < E < 60 keV, B4C is well suited as the polarizer. However, optimized polarization
systems for both energy ranges differ from each other. Compared with B4C target, slight
improvements can be achieved using a combination of HOPG (highly oriented pyrolytic
graphite) on Al2O3. An HOPG target (d ¼ 1:2 mm) optimized for the energy range
8 < E < 25 keV is inefficient in the upper energy range, due to its low density
(r ¼ 2:2 g=cm3). For the collimator diameters of 5 mm used here, the Bragg reflection is
negligible (Kanngießer et al., 1992). When the target is glued to an Al2O3 substrate,
Copyright © 2002 Marcel Dekker, Inc.
Table 4 Detection Limits (n.d. ¼ Not Detectable) for Some Elements in a Base Oil Using Direct
Excitation, Monochromatic Excitation Using a Mo Secondary Target, and Direct Excitation
with Linearly Polarized X-rays
the polarization decreases imperceptibly due to multiple scattering in the lower energy
range E < 25 keV. However, in the energy range E < 25 keV, the Al2O3 substrate improves
the scattering efficiency at the cost of polarization. On balance, such a multiple-layer
scatterer is useful for determining a wide range of elements simultaneously with very good
detection limits and sensitivities.
F. Applications
Based on Eqs. (3) and (4), Barkla targets are especially suited for multiple-trace-element
analysis in a light matrix. Typical applications include traces in organic substances
[polymers, oils, pharmaceuticals, biological materials (Heckel, 1995), etc.], traces in sili-
cates [rocks (Heckel et al., 1991), soils (Heckel et al., 1992), sludges, cinders, etc.], as well
as traces in Mg and Al alloys.
In a first application, detection limits and sensitivities for traces in an oil matrix are
compared among direct excitation, secondary target excitation, and polarization excita-
tion (see Fig. 3). As mentioned several times earlier, the count-rate limitation of the de-
tection system is the deciding factor for optimization of the polarization system. It must
also be noted that the electronic noise and, with it, the energy resolution of the spectro-
meter are functions of the pulse peaking time of the pulse-processing electronics and,
through it, of the maximum possible pulse throughput rate. For multiple-element analysis
in the energy range between 1 and 50 keV, the energy resolution for MnKa should be
better than 180 eV. For the spectra displayed in Figure 3, a detection system with an
energy resolution of 166 eV at an input count rate of about 60,000 cps was used. It must be
noted that no x-ray filters are used to keep the goal of a multielement analysis in a short
measuring time. Especially in the case of polarized or direct excitation, the detection limits
of a limited number of elements can be improved by using x-ray filters significantly. The
detection limits, DLs (see Chapter 2; confidence level: 99.86%), for the excitation types
displayed in Figure 3 are summarized in Table 4; the analytical sensitivities, are
summarized in Table 5. A comparison of the numerical values in Table 4 indicates that
Copyright © 2002 Marcel Dekker, Inc.
Table 5 Analytical Sensitivity for Some Elements in a Base Oil Using Direct Excitation,
Monochromatic Excitation Using a Mo Secondary Target, and Direct Excitation with Linearly
Polarized X-rays
the detection limits determined with excitation using polarized x-rays are, on average five
times better than those determined with direct excitation. An improvement of only a factor
of 2.4 was achieved for the element Mo. In contrast to direct excitation using RhK ra-
diation, the Compton scattered RhKa line is unable to excite Mo. The detection limits for
the elements 46 < Z < 60 are worse than those for the elements 23 < Z < 41, because
these elements are only excited by bremsstrahlung. An improvement is possible when a
voltage > 50 kV is used. The Rh tube is operated with a power of 175 W for polarization.
If a 2-kW tube is used, the effective polarization can be improved from Peff 82% for the
Compton-scattered RhKa line to Peff 87% for the same detector input count rate
through reduction of the collimator diameter. This clear increase in polarization results in
a 40% improvement of the detection limits displayed for polarization in Table 4.
If one compares the values for polarization equipment with those obtained with Mo
secondary targets, an improvement in the detection limits of a factor up to 2.5 is obtained
for all of the elements with absorption edges that are only slightly below the energy of the
MoKa line (e.g., Pb in Table 4). Similar improvements are expected for the elements Sr and
Zr. Detection limits of the polarization setup become better compared to those obtained
with Mo secondary target if the energy difference between absorption edge of the elements
and the MoKa line becomes greater. Of course, not all of the elements can be determined
with the Mo secondary target. This reflects an additional advantage of Cartesian excita-
tion geometry. A change in the targets is sufficient to switch from a multiple-element
application with a polarizer to a ‘‘single-element application’’ when analyzing single
elements, such as Pb in oil, or a couple of elements with absorption edges that are only
slightly below the energy of the characteristic radiation of the secondary target. Changing
of the targets is computer controlled in commercially available instruments.
A comparison of the numerical values in Table 5 shows that, on the average, a
factor-of-4 improvement in the sensitivity is obtained using the polarization unit as
compared to direct excitation. The sensitivities are also improved by a factor of between 2
and 7 compared to those of the Mo secondary target.
Copyright © 2002 Marcel Dekker, Inc.
If the detection limits given in Table 4 are not sufficient for the elements with atomic
numbers in the range 45 < Z < 61, for example, then an improvement can be obtained by
optimizing the polarization unit for the respective energy range. For example, the de-
termination of Cd, Sn, and Sb with a concentration range C > 0.1 mg=g in rocks represents
an interesting geological application. From the above discussion, it follows that an ad-
ditional improvement in the peak-to-background ratio can be achieved using an x-ray
filter. The result of the optimization is displayed in Figure 4 for the spectrum of the
standard granite AC-E. As can be seen in Figure 4, the peak-to-background ratios for Ag
and Cd are sufficient for the determination of the elements in sub-ppm concentration
range. The root-mean-square deviation for the Cd calibration [see Eq. (28)], which is based
on 22 international rock standards, is 0.2 mg=g. The DLs are summarized in Table 6.
In contrast to the analysis of oil, which has been discussed earlier, when analyzing
geological samples with input count rates of 60,000 cps or more, it is necessary to take
pileup peaks into consideration at energies E < 25 keV. This is because the concentration
of the elements K, Ca, Ti, and Fe may amount to % levels. A pileup peak from FeKa þ
FeKa at 12.8 keV near the PbLb line at 12.6 keV and near the ThLa line at 12.95 keV or a
pileup peak from FeKa þ CaKa at 10.09 keV near the HgLa line at 9.99 keV are typical
examples. These analytical problems can be contained using filters between the target and
the specimen or between the specimen and the detector, by reducing the input count rate
(see Fig. 5) or by improving the pileup rejection. With the given analytical conditions,
detection limits at sub-ppm levels were obtained for most of the trace elements in geo-
logical specimens (see Tables 6 and 7).
Because the iron content in many geological specimens amounts to the % levels,
an increase in the background in the energy region E < 6.4 keV occurs due to incomplete
charge collection in the Si(Li) detector. This background worsens the detection limits,
particularly for V, Cr, and Mn. With a Co secondary target and an Fe filter between the
target and the specimen, only the CoKa line excites the specimen. This line is not able to
excite Fe. The detection limits of the elements V, Cr, and Mn are improved by using the
Co secondary target in an Fe-rich matrix (e.g. Fe ores). The secondary target is again
successfully combined with polarizers.
Table 7 Detection Limits for Various Elements in Rocks Using Excitation with Linearly Polarized
X-rays
use in a polarized-beam XRF experiment should have several characteristics. The first
criterion is a material that has a reflection of high diffraction power (integral reflectivity
> 104 rad) at 2y near p=2 for the characteristic wavelength of the anode material used. In
addition, the material should be stable, must be free of contamination with characteristic
lines in the energy region 1 < E < 15 keV, easy to orient and cut to the desired orientation,
obtainable at reasonable cost, and perhaps nontoxic.
The orthogonal triaxial geometry for Bragg-diffraction polarization is similar to that
for Barkla scattering, shown in Figure 2. What is important here is the physics of the
diffraction process and the parameters of the diffraction crystal. There are two theories of
x-ray diffraction; the wave kinematics (K theory) and the dynamic (D theory). The K
theory is valid for small crystals or mosaic blocks. The D theory must be used when
absorption and=or the interaction between incident and diffracted beams becomes im-
portant. According to the theory of Zachariasen (1967), the integrated reflectivity Ri of an
ideal mosaic crystal in the symmetrical Bragg case has the following form:
2
r0 1 þ cos2 2yB 1 sin2 yB
Ri ¼ jFhkl j2 l3 exp 2B ð23Þ
V 2 sin 2yB 2m0 l2
where
Ri Integrated intensity compared to the incident beam
r0 Classical electron radius
yB Bragg diffraction angle
Copyright © 2002 Marcel Dekker, Inc.
PN
Fhkl n¼1 fn exp½2piðhun þ kvn þ lwn Þ
structure factor of the lattice plane hkl
un ; vn ; wn Fractional coordinates ¼ 1=axn ; 1=byn ; 1=czn of the nth
atom in the unit cell
a; b; c Unit cell lengths
fn Atomic scattering factor
V Unit-cell volume
m0 Linear absorption coefficient
B Thermal Debye parameter at room temperature (B ¼ B0 K þ B293 K )
Figure 6 Excitation spectrum of a flat HOPG crystal in comparison with graphite with a Bragg
angle of 43.2 (distance, tube crystal: 26 mm; distance, crystal detector: 28 mm; collimator diameter:
1 mm each; measurement in air; tube voltage: 30 kV). Logarithmic scale!
where Ri,n is the integral reflectivity of the nth reflection, n ¼ {1, 2, 3}, oj is the fluorescent
yield of the element j, excited with reflection n, and rj is the jump ratio for the element j.
When observing the K values for typical elements from the three excitation regions,
for example, Si, Ca, and Fe, similar values are found: K1;Si K2;Ca K3;Fe ; that is, the
element specific sensitivities for the given excitations are similar. The very small fluores-
cence yield for the light elements, such as Al and Si, is compensated for by higher integral
reflectivity.
B. Curved Crystals
To increase the intensity at the sample, one may bend (Johann) or bend and grind (Jo-
hansson) the crystals (see Chapter 2). With a flat crystal in a narrow-beam geometry, the
angular acceptance is essentially the rocking curve, generally
0.01 . With a bent crystal,
an angular acceptance of 5 –10 is readily conceivable, with an increase in intensity
proportional to the size of the fan beam. Aiginger and Wobrauschek (1981) report an
intensity gain of 3.1 using Cu(113)–CuKa in Johann geometry compared to the flat crystal.
Calculations for Johann and Johannson geometries have been performed by Zahrt (1983,
1986). Although the calculations used the estimated values for a mosaic distribution
function and its full width at half-maximum, the reported gain is also calculated. Beckhoff
and Kanngießer (personal communication, 1996) calculated the intensity of the reflected
RhL radiation on a spherically bent HOPG(002) crystal with a 32 mm radius of curvature
(C1 ¼ C2 ¼ C3 ¼ 12 mm according to Fig. 2) for various focal spot sizes of a Rh tube
and crystal positions using Monte Carlo simulations. When compared to a flat crystal, it is
possible to have increases in the excitation intensity in a zone diameter excitation ¼ 20 mm
(for example) on the sample by the factors of 17.2, 11.1, and 2.6 for focal spot diameters of
0.1 mm, 1 mm, and 6 mm of a Rh tube, respectively. Experimental examinations with a
focal spot diameter of focus 1 mm have approximately confirmed this improvement by a
Copyright © 2002 Marcel Dekker, Inc.
factor of 8.1 for the (002) reflection. Focal spot diameters of focus 1 mm can be realized
with the end-window x-ray tubes only in conjunction with reduced powers. However, even
tube powers of 50 W are sufficient for the most applications.
C. Applications
Because only a limited number of elements can advantageously be excited with Bragg
systems, the range of applications of an EDPXRF instrument optimized exclusively for
Bragg polarization is limited. The determination of the thickness of Al layers on diverse
substrates (e.g., SiO2) is an interesting application of this kind. For a layer thickness of
d
90 nm, for example, reproducibilities [relative standard deviation (RSD)] <1%
(99.86% confidence level) were obtained with a measuring time of 50 sec per measuring
point using a spherically bent HOPG(002). Due to the monochromatic excitation of Al
and Si, there were no disturbing peaks in the measured spectrum.
The optimization criteria described in Sec. III, for the generation of highly intense
Barkla polarized radiation, is crucial for optimization of the geometry of an EDPXRF
system for Bragg as well as Barkla polarization. Typical applications for Bragg systems
within the given framework include, for example, the determination of the elements from
Na to Fe in ceramics, refractory materials, cements, ores, ferro-alloys, slag, or additives
in oil.
The detection limits for 12 selected elements in base oil displayed in Table 8 de-
monstrate that EDPXRF is also well suited to the simultaneous trace detection of lighter
elements (see also Fig. 7).
Protective foils, 4 mm prolene, were used in front of the detector and between crystal
and sample for the measurements listed in Table 8. These easily replaceable protective foils
are necessary in routine work. It is, however, possible to obtain much better detection
limits for Na (120 mg=g) without the protective foils. The relatively poor detection limits
for Na is caused in part by an overlap of the NaKa line with the escape line from non-
polarized PdLa radiation backscattered from the sample. Greatly improved detection
limits and sensitivities can be obtained for Fe, Cu, and Zn by using a Barkla scatterer
analogous to that in Sec. III. F. However, these improvements are not required for the
given application, additives in oil; an additional measurement unnecessarily increases the
measuring time. The detection limits for P, S, and Cl, in particular, open the range of
possible applications for EDPXRF (e.g., the determination of S in diesel fuels and ga-
soline). The comparison shown in Table 9 demonstrates the brightness of curved crystals
in combination with fine focus x-ray tubes in EDPXRF.
Table 8 Detection Limits and Sensitivities, S, for Various Elements in an Oil Matrix Using
Excitation with Polarized Radiation
DL S DL S DL S
Element (mg=g) [cps=(mg=g)] Element (mg=g) [cps=(mg=g)] Element (mg=g) [cps=(mg=g)]
The polarization geometry described here is also suited to the analysis of fusion
samples. Preparation of lithium tetraborate glasses from oxidic samples, for example, is
utilized to conduct precise determination of major and minor elements and also, in some
cases, trace elements. Excellent sensitivity and reproducibility are required for simulta-
neous analysis of the elements from Na to Fe with dilution ratios from 1 : 6 to 1 : 20.
RSD ¼ 0.12% for the reproducibility of major elements and root-mean-square deviations
RMSs for the calibration lines [see Eq. (28)], presented in Table 10 confirm the ad-
vantageous parameters of the analytical technique based on Bragg polarization:
Table 9 Comparison of Sensitivities, SN , Normalized to Tube Current (mA) and Detector Area
(mm2 ) for Various Elements in Base Oil
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PM
i ðCi
certified
Cicalibrated Þ2
RMS ¼ ð28Þ
Mf
where M is the number of standards used (M ¼ 35) and f is the number of regression
coefficients determined.
The HOPG crystals are characterized as very good Barkla targets in Sec. III and as ex-
cellent Bragg polarizers in Sec. IV. Polarization of the entire spectrum emitted from a tube
has been successfully performed with these crystals. Excitation by monochromatic Bragg
reflections dominates the energy range E < 10 keV; polychromatic Barkla polarized ra-
diation dominates in the energy range E > 10 keV. Optimization of this Barkla–Bragg
combination system is matrix dependent due to the count-rate limitation of the detection
system and possible pileup lines. Using modern Si(Li) detectors or Si-drift chambers, it is
possible to process 104–105 cps so that optimization is facilitated by taking advantage of
the ability to control the tube current. Additional targets are only required for special
applications, because it is possible to excite all of the elements between Na and U with this
combination target. Of course, it is possible to optimize the excitation conditions using
radiation filters between the target and sample, analogously to direct excitation with
nonpolarized primary tube radiation. Being able to use air-cooled low-power x-ray tubes
(50 W) as the source of primary radiation is an important advantage of Barkla–Bragg
combination targets. The spectrum shown in Figure 8 was measured with an integral input
count rate of 25,000 cps (dead time: 18%) using a Peltier-cooled Si-drift chamber with an
active area of 5 mm2 and an energy resolution of 159 eV for MnKa.
The basic principles when working with secondary targets are to generate nearly mono-
chromatic radiation and to use it for excitation. Because the characteristic radiation of the
target material is nonpolarized, the detection system is burdened by backscattered char-
acteristic radiation from the target material, particularly when analyzing light matrices.
The advantageous application of secondary targets in the Cartesian excitation
geometry required for polarization is described by Bisgård et al. (1981). The suppression of
Copyright © 2002 Marcel Dekker, Inc.
Figure 8 Spectrum (log scale) of a geological standard sample, GnA, excited with the polarized
primary spectrum of a 50-W low-power Pd x-ray tube, measured with a Peltier-cooled Si-drift
chamber. Some of the concentrations are as follows: Si, 334 mg=g; K, 21.8 mg=g; Ca, 4.4 mg=g; Ti,
120 mg/g; Mn, 1.3 mg=g; Fe, 41.4 mg=g; Cu, 18 mg=g; Zn, 78 mg=kg; Rb, 2020 mg=g; Zr, 70 mg=g; Nb,
94 mg=g; Mo, 100 mg=g; Sn, 1900 mg=g; Bi, 220 mg=g.
primary tube radiation scattered on the target plays an important role, especially when
using secondary targets with low atomic numbers, such as Al. This is because radiation
filters cannot be placed between the target and the sample in this case. The detection limits
listed in Table 4 show that, compared to polarization targets, better results can be ob-
tained with a given x-ray tube for a limited number of elements. For example, a detection
limit of 0.2 mg=g was achieved for Hg in an organic solution using a measuring time of
500 s with a Zr secondary target.
Secondary targets in the Cartesian excitation geometry are often utilized to de-
termine trace elements with atomic numbers Ztrace < Zmajor without exciting the major
element Zmajor. For example, a Co secondary target is used for the detection of Mn traces
in an iron matrix; an Fe filter is used to absorb the CoKb line that can excite Fe. A Si
secondary target is used for the trace detection of Na, Mg, and Al in a SiO2 matrix.
A detection limit of 11 mg=g for Al in glass sand prepared as a pressed powder pellet is
achievable.
Thus, secondary targets are an ideal complement to Barkla and Bragg targets in a
Cartesian excitation geometry.
VII. CONCLUSION
In the fundamental publications in the 1980s and 1990s about the application of polarized
radiation in EDXRF by Ryon and Zahrt for Barkla polarization and Aiginger and Wo-
brauschek for Bragg polarization, tube powers in kilowatt ranges are a condition for
presentation of the advantages of EDPXRF over direct excitation with nonpolarized x-ray
radiation. In recent years, the development of Cartesian excitation geometries has been
Copyright © 2002 Marcel Dekker, Inc.
propelled forward so that when coupled with powerful semiconductor spectrometers that
enable the processing of about 100,000 cps with resolution FWHM < 200 eV, it is now
advantageous to use 50-W tubes in PXRF.
In practice, PXRF has asserted itself in many industrial laboratories and research
institutes as a powerful method for elemental analysis. The following typical applications
of EDPXRF instruments* have been identified:
Determination of heavy metals in soils and sludges
Major, minor, and trace element determination in geological samples
Precise major and minor element determination in ceramics and refractory materials
Trace elements determination in biological and organic samples
Analysis of aerosols
Analysis of additives in oils
Determination of trace elements in polymers and pharmaceutical products
Rapid determination of heavy metals and halogens in wastes
Analysis of slag
In the future, the adjustment of tube and crystal parameters will be additionally
improved so that new applications with EDPXRF will be possible (e.g., major, minor and
trace element determination within sample areas 1 mm2). It will be a long time before the
developments in the analytical fields become completed, especially the use of curved
crystals in EDXRF.
REFERENCES
Anders Rindby
Chalmers University of Technology and University of Götëborg, Götebörg, Sweden
Koen H. A. Janssens
University of Antwerp, Antwerp, Belgium
Up to the end of the 1980s, elemental microanalysis based on the emission of characteristic
x-rays was largely restricted to charged-particle-beam techniques (electrons, protons, and
heavier nuclei). Although the use of primary x-rays for inducing the production of the
fluorescent radiation was widespread in bulk x-ray fluorescence (XRF) instrumentation,
the lack of brilliance of conventional x-ray tubes and the inherent difficulties of focusing
x-rays in the 1–30-keV energy range hampered the development of the microscopic variant
of XRF as a microanalytical tool.
Nevertheless, Glockner and Schreiber (1928) quite early demonstrated the possibility
of doing chemical analysis by fluorescence spectroscopy with small x-ray beams. In the
mid-1950s, Long and Cosslett (1957) reported detection limits at the picogram level from
an x-ray microbeam device. Zeits and Baez (1957) predicted a theoretical detection limit
from an optimal instrument to be below 1014 g. They concluded that compared to the
electron microprobe, the x-ray microprobe was superior in relative sensitivity, but not in
terms of absolute detectability. Localization was also poor for x-rays; however, the energy
impact is much smaller in comparison to charged-particle beams.
In the early 1980s, the availability of highly intense radiation beams from syn-
chrotron sources increased the interest in using microscopic x-ray beams for performing
m-XRF experiments (Chapter 8); also around this period, capillary optics were being used
as simple, compact, and relatively inexpensive means of ‘‘concentrating’’ a beam of x-rays
down to a smaller size. This innovation lead to the development of microbeam XRF in-
strumentation employing (conventional) x-ray tubes as the source of primary radiation.
During the last two decades, monocapillary and polycapillary focusing devices have been
incorporated into a variety of m-XRF spectrometers; an overview of their characteristics is
provided in Table 1.
More recently, the introduction of commercially available compact minifocus and
microfocus tubes has considerably increased the potential of microbeam XRF, making
possible the use of microbeams in small-sized and easily transportable XRF equipment.
Another significant development of recent years is the maturation of the techniques
Diffraction tube (a) n.s. n.s. 10, 30, 100 mm coll. 1,000 n.s. Cu
W tube þ Mo secondary target (b) n.s. n.s. 200 mm str. cap n.s. n.s. W
Philips interchangeable tubes (c) n.s. n.s. 30 mm coll. 2,500 n.s. Cu
Fine-focus diffraction tube (d) n.s. n.s. 10, 30, 100 mm coll. 500–1,000 n.s. Cu, Mo
n.s. (e) n.s. n.s. 22, 7.5 mm par. cap. n.s. n.s. Mo
Low-power tube (f) < 0.25 0.25 n.s. 10, 30, 100 mm coll. 50 > 0.8 Mo
Long, fine-focus diffr. tube (g,h) 0.4 12 40 200 mm str. cap. 1,900–3,000 0.4–0.6 Cr, W
Radiographic microfocus tube (i) 0.030 5 10 mm str. cap. 15–21 0.5–0.7 Mo, W
Diffraction tube ( j) 0.4 8 n.s. 18, 23, 27 mm str. cap. n.s. n.s. Cu, W, Mo
Low-power tube (k) 0.25 0.25 n.s. 70 mm coll. 50 0.8 Mo
Long, fine-focus 0.4 8 40 18 mm con. cap. 1,000 0.3 Cr
diffraction tube (l,m) 7.5 mm ell. cap.
Idem (n) 15, 65, 70 mm con. cap. 1,500 0.5 Mo
Idem (o) 7.5 mm ell. cap. 1,000 0.3 Cr, Cu
n.s. (p) n.s. 40 15 mm con. cap. n.s. n.s. Mo
Radiographic microfocus 0.015 0.1 5 4, 8, 29, 100 mm str. cap. 1.6 (Cu, W) 1.0 Cu, W, Mo
tube (q) 1.4 (Mo) 0.9
Rotating anode tube (r) 0.5 10 70 10, 15 mm ell. con. cap. 18,000 3.6 Cu, Mo
Rotating anode tube (s) 0.5 10 70 160 mm polycap. 18,000 3.6 Mo
Minifocus tube (t) 0.25 0.25 40 50, 100 mm polycap. 60 0.96 Mo
Radiographic microfocus tube (u) 0.03 0.006 22 21 mm polycap. 12 67 Mo
a
(a) Nichols and Ryan, 1986; (b) Rindby, 1986; (c) Boehme, 1987; (d) Nichols et al., 1987; (e) Yamamoto and Hosokawa, 1988; (f ) Wherry et al., 1988; (g) Engström et al.,
1989; (h) Rindby et al., 1989; (i) Carpenter, 1989; Carpenter et al., 1989; Carpenter and Taylor, 1991; (j) Furata et al., 1991; Furata et al., 1993; Hosokawa et al., 1997;
(k) Pella and Feng, 1992; (l) Larsson et al., 1990; (m) Shakir et al., 1990; (n) Rindby, 1993; (o) Attaelmanan et al., 1995; (p) Holynska et al., 1995; (q) Carpenter et al., 1995b;
(r) Janssens et al., 1996a; (s) Vekemans et al., 1998; (t) Bichlmeier et al., 2000; (u) Gao et al., 1997.
b
str.cap., con.cap., par.cap., ell.cap.: resp. straight, conical, paraboloidal, ellipsoidal capillary; coll.: collimator; polcap.: monolithic polycapillary lens; n.s.: not specified.
C. Chapter Overview
In this chapter, after providing some information on the basic x-ray–matter interactions
which are employed for microfocusing x-ray and comparing the way in which x-ray and
charged-particle microbeams interact with matter, the instrumentation required for per-
forming microbeam XRF is discussed, with some emphasis on recent developments in this
field (tubes, optics, detectors). Next, data acquisition strategies for point, line, and image
XRF analysis are described and the appropriate procedures for evaluating, segmenting,
and quantifying (large amounts of) spectral data are discussed. Finally, an overview of
applications of the microbeam XRF technique in various fields is presented: the latter
include the use of m-XRF in the industrial, environmental, cultural, and forensic sectors of
activity.
Table 2 Penetration Depth of Electrons, Protons, and X-ray Photons in Various Materials
For a multielectron atom, f 0 can be written as a sum of scattering factors for all the
electrons. For energies far below any absorption edge, the corresponding electrons will
only contribute in a minor way to the atomic scattering factor. However, for energies far
above the absorption edge, the corresponding electron will contribute with unity. Thus, for
energies far above the K edge, the scattering factor will be equal to the atomic number.
Close to any edge, the scattering factor will have a ‘‘resonance structure,’’ as shown in
Fig. 5.
2. X-ray Reflectivity
If a plane wave propagating in a medium characterized by the refractive index n1 ¼ 1 d1
is incident on a smooth-plane boundary of another medium with refractive index
n2 ¼ 1 d2 then if d1 6¼ d2 , a reflected and a transmitted wave is generated. The reflectivity
(R) as well as the transmittance (T ) can be calculated from the Fresnel formula that simply
assumes the electric and magnetic field components to be continuous at the boundary
interface (Azaroff et al., 1974). Using Snell’s law, the parallel and perpendicular compo-
nents of the reflectivity Rp and Rs can be expressed in terms of the incident angle yi and the
indices of refraction of the two media:
yi ðy2i 2dÞ1=2
rs ¼ rp ¼ ð6Þ
yi þ ðy2i 2dÞ1=2
Note that this expression becomes unity if y2 ¼ 2d, provided that d is real. This situation
corresponds to total reflection of the incident wave front and thus the critical angle
yc ¼ ð2dÞ1=2 / 1=E, where E is the x-ray energy. Because d, in general, is complex (taking
the absorption into account) the reflection coefficient will not be exactly unity. However,
for most of the optical materials considered here, the reflectivity will be close to unity
within a small angular range from zero degrees up to a few minutes of arc and then
suddenly drop to about d2 =4 (see Fig. 6).
Because d is complex ðd ¼ a þ ibÞ, the following substitution will be convenient:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
y2 2d ¼ y2 2a 2ib ¼ p þ iq ð7Þ
Thus, R can be written as
ðy pÞ2 þ q2
Rðy; EÞ ¼ ð8Þ
ðy þ pÞ2 þ q2
where
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1
p ¼ ð ðy2 2aÞ2 þ 4b2 þ y2 2aÞ;
2
2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð9Þ
1
q2 ¼ ð ðy2 2aÞ2 þ 4b2 y2 þ 2aÞ
2
3. Geometrical Aberrations
As the reflectivity for x-rays is extremely small except within a small range from zero up to
critical angle yc , most x-ray optics are based on the principle of grazing incidence, although
optical devices based on Bragg diffraction are also in use (Ice, 1997). The different types of
geometrical aberration that occur for such an optical system can be described in terms of
spherical aberration, astigmatism, curvature of the field, and coma. However, in contrast to
the imaging system of light optics, there is no axial symmetry. Thus, the normal optical
treatment cannot be applied and the classification of aberration will be somewhat different.
Spherical aberration arises when rays (originating from a point source), which will be
incident on the mirror surface at different angle, are not intercepting at the
same point. Thus, the image of a point object formed by such a nonideal
reflector will become a disk on confusion rather then a point.
Astigmatism will arise if the meridian and sagittal rays are not focussed at the same
distance from the mirror.
Coma is a kind of field aberration that is due to different magnifications being
produced by different regions of the mirror.
Curvature of the field means the focal plane is slightly curved due to differences in
focal lengths at different position in the object plane. This type of aberration
will determine the physical ‘‘depth of focus’’ for the system.
Copyright © 2002 Marcel Dekker, Inc.
Figure 6 Left: The reflectivity for glass as a function of the glancing angle at E ¼ 5.411 keV; the absorption coefficient was chosen to
be mglass (i.e., the normal value), 10mglass and 0.1mglass. Right: Calculated reflectivity of glass at two energies (CrKa and MoKa) while
using the normal value for absorption coefficient.
Table 1 summarizes the characteristic features of several tabletop instruments recently de-
scribed in the scientific literature. A m-XRF setup essentially consists of four major parts:
1. An x-ray source
2. A focusing or collimation device
3. A motorized sample stage þ microscope
4. One or more detectors
As we can see from Table 1, capillary optics is, by far, the most popular type of optics for the
tabletop instruments. Most of the tabletop setups are equipped with either microfocused
tubes in combination with collimators or straight capillaries or high-flux tubes with conical
capillaries. When simple pinhole or cross-slit systems are used to aperture the beam, the
problem is to get close enough to the actual source in order to pick up the highest flux density.
Straight capillaries have been used by many authors as a means of doing so and guiding it out
to the sample position without the 1=r2 geometrical loss that occurs when conventional
collimators are employed. As shown in Figure 7, these setups may utilize specially designed
microfocused tubes where the capillary optics have been integrated in the tube construction.
By using conical capillaries, one can operate further away from the source, as the
‘‘squeezing’’ of the beam will compensate for the reduced flux density. As a ‘‘nonimaging
device,’’ the conical capillaries can utilize virtual brilliance that is achieved in standard
diffraction tubes by the projection of long fine-focal spots. For any focusing device,
accurate alignment is important; thus, the focusing element has to be completely in-
tegrated and permanently fixed to the source or be mounted on a flexible stage so that
realignment can be done. A more recent development is the use of polycapillary x-ray
lenses in combination with compact low-power tubes for generating x-ray beams in the
size range above 15–30 mm.
Figure 7 Microbeam XRF setup with the capillary focusing unit built into a microfocus x-ray
tube. (From Carpenter et al., 1988.)
B. X-ray Optics
As can be seen from Table 1, in laboratory m-XRF setups, monocapillary or polycapillary
optics is employed almost exclusively for beam focusing or concentration. At synchrotron
Nominal Distance
Nominal Maximum Maximum Maximum maximum anode to
Manufacturer= focal size voltage current power brilliance Be window
model Anode (mm2) (kV) (mA) (W) (kW=mm2) (mm)
Figure 8 The principle of x-ray propagation inside straight, conical, and ellipsoidal capillaries and
in monolithic polycapillary x-ray lenses. In the ellipsoidal and polycapillary cases, a true focus is
formed by rays emitted from the focal point of the device. Any ray emitted outside the optical axis
will propagate inside the ellipsoidal by multiple reflections similar to the conical capillary.
Figure 9 The capillary gain factor (Ig) of a straight capillary is defined by the ratio of DOeff to DO.
Figure 10 The capillary gain factor versus x-ray energy for a straight capillary without any
absorption.
As long as the final angle of incidence is below the critical angle, the x-rays propagate
inside the capillary by successive reflections without losing almost any intensity. If the
intensity of the x-rays from such a capillary is compared with the intensity where no re-
flections occur, which can be determined by replacing the capillary with an aperture
positioned at the site corresponding to the far end of the capillary, the reflectivity of the
capillary wall material can be determined. The intensity gain factor, which can also be
defined as the ratio of the intensity with and without reflection, can then be written as
Zymax
1
Ig ¼ Rtot ðy0 ; EÞ2p sin y0 dy0 þ 1 ð13Þ
DO
ymin
3. Ray-Tracing Models
Although explicit expressions can be obtained for calculating the reflection of radiation
inside a capillary (Engström, 1991; Lechtenberg, 1994; Vincze, 1995c), they are only valid
for perfectly shaped conical capillaries and can only be applied for point sources. Any
model that would intend to describe ‘‘real’’ capillaries applied to ‘‘real’’ extended sources
have to rely on some kind of Monte Carlo model where all the source and capillary
parameters are used as input data.
There are many different factors that have an influence on the capillary performance
and it is hard to quantitatively assess the significance of each individual factor by inter-
pretation of simple test experiments. This problem is augmented by the fact that the
manufacture of capillary optics is still very much an art rather than a well-controlled
technique, making it difficult to produce series of capillaries where one parameter is sys-
tematically varied. The need to gain insight into the relative importance of the various
above-mentioned factors has prompted the development of a detailed ray-tracing code
(Vincze, 1995c) that is able to simulate the beam-forming properties of realistic capillary
devices assuming various experimental conditions.
A number of different factors such as capillary material, surface roughness, devia-
tions in shape that real devices assume in comparison to the ‘‘ideal’’ straight, conical,
or ellipsoidal shapes, need to be properly described in such a simulation calculation.
In addition to that, the size, divergence, and distance of the x-ray source and the energy
distribution of the photons that enter the capillary also strongly influence the performance
of the capillary beam and thus also have to be specified in detail. Such a full-scale ray
tracing simulation model has been described by Vincze et al. (1995c).
Because the capillaries in such a simulation are treated as truly three-dimensional
devices having numerically defined shapes, the only restriction on the modeled shape is the
assumed circular cross section in the plane perpendicular to the capillary axis. This allows
overall distortions of the capillary shape, such as bending of the capillary axis, to be
included in the model. Also, the composition of the capillary material as well as the
roughness of the reflecting surface can be freely chosen. The implemented surface
roughness model describes both the attenuation of the specular reflectivity and the effects
of diffuse scattering by the rough surface. An additional feature of the program is that the
possible transmission of photons through the capillary wall is also taken into account. In
the case of a perfectly conical capillary coupled to a point x-ray source, the simulation is in
perfect agreement with the analytical models, as is demonstrated in Figure 12. However,
the ray-tracing models, is also able to simulate all kinds of capillary–source combination
which cannot be simplified so that they may be described by simple analytical models.
As we can see from Figure 12, the transmission efficiency versus energy will appear as
a steplike function, where the ‘‘steps’’ represents the maximum number of reflections that
can occur inside the capillary before the critical angle is reached. This distinct feature
disappears for an extended source, as the initial inclination angle will vary over a wide
range corresponding to the source size. The graphs in Figure 12 also show that the effi-
ciency goes down for an extended source and the ‘‘energy of maximum transmission
efficiency’’ will be shifted to lower energies.
Copyright © 2002 Marcel Dekker, Inc.
Figure 12 Comparison of the analytical model and the ray-tracing model as applied to a perfect
conical-shaped capillary and a point source. The solid curve shows the transmission efficiency (out
going flux=ingoing flux) calculated for a capillary 5 cm away from a point source with the
dimensions l ¼ 7 cm, d0 ¼ 31 mm, and d1 ¼ 5 mm. The symbols correspond to ray-tracing calculations
for a point and mm extended x-ray source (100 100 mm2 ).
4. Practical Considerations
According to most models of capillary optics, almost any gain factor could be obtained if
there are no geometrical constraints on the capillary dimensions. However, in practice, the
capillaries have to be designed within a narrow range of geometrical limits. In most cases,
the source characteristics and the beam diameter is fixed or dictated by the analytical
Copyright © 2002 Marcel Dekker, Inc.
Figure 13 Intensity distribution from a conical capillary 5 cm away from the opening. left; ray
tracing calculation for a point source, center; ray tracing calculation for an extended source
(400 400 mm2) and right; recorded image from a synchrotron set-up. The central spot represents the
direct beam and the rings correspond to different reflection modes (first ring ¼ one reflection, second
ring ¼ two reflections etc.).
requirements and, thus, length, shape and capillary material can be optimized. In some
cases, the source–sample distance is determined by the mechanical arrangement so that
only the shape (i.e., the capillary cone angle) can be varied. Figure 14 shows an example
Figure14 Calculated effective solid angle as a function of entrance diameter for different capillary
lengths. The calculation was performed at an energy of 7 keV and a fixed output opening of 5 mm
assuming a point source and a source–capillary distance of 5 cm. The surface roughness was
characterized by assuming a slope error of 30 Å (see Vincze, 1995c). The result was also obtained for
an ideal surface.
of how the lengths and cone angle will affect the ultimate gain factor at a given set of
experimental parameters.
These calculations were performed with the introduction of a realistic surface
roughness model. The impact of surface roughness is indicated in Figure 15 by showing
the gain factor with perfect reflecting inner walls. Although the gain factor will be
dependent of the capillary length, the flux density will be about the same for all three
capillaries in Figure 15 and they will all have an optimum at an entrance opening of about
30 mm. If the distance between the source and capillary is substantially longer than the
capillary itself, then the optimum will be reached for a certain cone angle for any capillary
length, but the flux density obtained will depend on how long one can make the capillary.
If the source is extended, then the effective solid angle in Figure 14 will decrease, and
the maximum values will be reached at a slightly higher entrance diameter as the capillary
can pick up more of the ‘‘off-axis’’ rays emitted from the source (see Fig. 15).
5. Capillary Alignment
For (mono) capillary optics, it is necessary to have two linear and two rotational stages for
a complete alignment procedure (see Fig. 16). Most alignment procedures are monitored by
the output flux (normally recorded by some simple device such as an ion chamber, a diode,
etc.) and the operator has to find the position for the maximum throughout. This is not a
straightforward task, as there always will be ‘‘secondary maxima’’ when operating the four
different stages. The ‘‘secondary maximum’’ problem can be avoided by combining the
linear and rotational stages in such a way that rotational axes are intercepting the x-ray
Copyright © 2002 Marcel Dekker, Inc.
Figure 16 Photograph of the COX Analytical Instrument’s capillary holder and gimbal
mechanics. The capillary rest in a highly viscous liquid that has a high cross section for x-ray
absorption. The ‘‘semiliquid’’ efficiently damps out microvibration while preventing any x-rays from
passing the holder outside the capillary. The capillary and the liquid are encapsulated into a
stainless-steel tube which is put into the gimbal.
source. However, a simple systematic alignment procedure normally works well and
realignment is a matter of a few minutes if there is some measurable flux being transmitted
through the capillary. If the beam is completely lost, realignment will take a longer time.
In this case, prealignment of the optics holder with a wider capillary can be a solution.
In general, the alignment of polycapillary lenses is easier and less critical than that of
monocapillary tubes; this is due the larger acceptance angle of polycapillary lenses.
Normally, only a stage having two transational stages is sufficient; in some cases,
alignment of the lens can even be achieved by using plastic spacers of various thicknesses
to appropriately position the lens relative to the anode focal point of the tube (Worley
et al., 1999).
For the ‘‘direct inspection’’ camera setup, it is necessary to have at least one
(probably two) mechanical linear stage to adjust the camera focal plane so that it will
coincide with the sample surface plane. For the ‘‘indirect inspection’’ system, the camera
can be fixed to the table, but it will be necessary to recalibrate the exact distance (in motor
steps) between the central point in the camera focal plane and the location of the beam at
the measuring position. The camera (or microscope) can be equipped with different types
of magnification lens, and optical micrographs from the sample are usually recorded
through a framegrabber to a video screen.
Copyright © 2002 Marcel Dekker, Inc.
D. Detectors
Most microbeam XRF tabletop systems are equipped with a conventional solid-state x-ray
detector such as Si(Li) or HP (high purity) Ge devices. For monocapillary-based mi-
crobeam XRF, it is important to have a large detector solid angle, as the primary photon
flux normally is very slow, especially for applications in trace element mapping. Thus, the
detector area needs to be large and the detector crystal should be close to the entrance
window (Fukumoto et al., 1999). It is important that the end-cap design is such that the
detector entrance window can come close to the sample. Normally, the count-rate cap-
ability is not critical; however, for fast mapping of mineral samples (or any material with
relatively high concentrations of heavy elements such as metal alloy samples), the count
rate can induce variations in the electronic dead time, generating a distortion in the ele-
mental maps. Deviations from the linear relation between the count rate of characteristic
radiation and the image gray-scale level will occur in areas where local concentrations are
high. This problem can be avoided by using the lifetime settings of the MCA (multi-
channel analyzer), but only when ‘‘static’’ scanning is used (see Sec. IV.A).
With the introduction of the solid-state drift-chamber x-ray detectors (Bertucci,
1992; Lechter et al., 1996; Strüder et al., 1998), the linear range can be extended for count
rates beyond 20,000 counts without any serious deviation and still with reasonable energy
resolution and low dead time. Other types of detectors that can be integrated in the sys-
tems are x-ray p-i-n diodes used for monitoring the primary intensity and for recording the
transmitted intensity. In this way, microradiographic images of the sample can be re-
corded and the sample thickness or density can be determined with high accuracy at each
image pixel (Attaelmanan et al., 1994). A large-area p-i-n diode is also useful for alignment
of capillary optics when the operator has to find the transmission maximum. Standard
CCD detectors have also been used as beam monitors, especially for studying the profile of
the primary beam (Attaelmanan et al., 1993). This is important for precise and accurate
alignment of many types of x-ray optics. It is also necessary to have detailed information
of the beam profile in order to perform any kind of image deconvolution.
Figure 19 EDXRF spectra of NIST SRM 1833 irradiated with Mo-anode-derived x-ray
microbeams.
Antwerp (Janssens et al., 1996), employing a 10-mm tapered capillary (15-mm effective
beam diameter at the sample surface) are compared to values obtained for the NSLS beam
line X26A m-SRXRF spectrometer (8-mm beam diameter) (Jones and Gordon, 1989). For
the laboratory setup, net yield values which are lower by a factor of 30–50 than those
obtained for the synchrotron setup. The absolute detection limits derived from the same
spectra (Fig. 20b) indicate that a comparable absolute detectability to that obtained with
other laboratory-scale m-XRF setups (Carpenter et al., 1995b) is obtained, situated at the
0.1–1.0-pg level. For thick samples, rather than the absolute DL values, the peak-to-
background ratios (P=Bs) found in experimental spectra determine, to a large extent, the
lowest measureable concentration levels.
Figure 3 compares the EDXRF spectrum of NIST SRM 1577a bovine liver ob-
tained using the above-mentioned monocapillary setup to the spectral distribution ob-
tained by using a polychromatic synchrotron (micro)beam of comparable cross section
at NSLS (Jones and Gordon, 1989) on the same sample. In Figure 20c, the corre-
sponding relative DLs corresponding to a 1000-s counting time are plotted. Clearly, in
the spectrum of Figure 3a (polychromatic synchrotron excitation), a much better peak-
to-background ratio is obtained: overall, the laboratory m-XRF spectrum features P=B
ratios which are about 10–30 times worse (Fig. 3b). The abundant scatter background
in the region of 10–16 keV makes fairly long counting times (>1000 s) necessary in order
to obtain information on elements present at or below the 10-ppm level (e.g., Br:
18 ppm). As can be seen in Figure 20c, for the elements Fe to Rb, DL values are
situated in the 3–10 ppm range for the lab-scale spectrometer, whereas they are about an
order of magnitude better in the case of polychromatic synchrotron excitation. One can
conclude that by means of the laboratory m-XRF instrument involved, trace-level
Copyright © 2002 Marcel Dekker, Inc.
microanalysis at the 10-ppm=10-mm level is possible, provided that sufficiently long
counting times are employed.
2. Polycapillary Setups
In contrast to monocapillary concentrators, polycapillary lenses have the ability to actu-
ally collect divergent radiation in a fairly large solid angle and focus it onto a point which
is located several centimeters away from the other end of the lens (see Fig. 8).
a. Beam Size and Divergence
In Figure 21, the beam size by vertically scanning a 10-mm Cr wire through the beam
produced by two polycapillary lenses is plotted for different values of the lens-to-wire
distance d. Both lenses had an input and output widths of 5.5 mm and lengths in the
range of 50–64 mm. The distance of the lenses to the Mo anode of a minifocus tube was
4 cm so that they collect radiation in a solid angle of 0.015 sr. The lenses consist of
250,000 individual monocapillary channels, each with an internal diameter of 7 mm (35%
open area) (Vekemans et al., 1998). Lens A yields a focal spot with a diameter of 60 mm
at a distance d of 15.7 mm from the end of the lens; in the case of lens B, a focus of around
120 mm is reached at d 42 mm.
Figure 21 Beam FWHM obtained by perpendicular scanning of a 10-mm Cr wire through the
beams produced by two polycapillary lenses of different output lengths. Lens A has an output focal
length of 15 mm and focuses the radiation into a spot of 60 mm FWHM; lens B has an output
focal length of 42 mm and produces a focal spot of around 120 mm in diameter.
FL =dL2 lL2
Ig ¼
FA =dA2 lA2
where FL and FA are the measured beam fluxes at a distance lL and lA , respectively, and
dL and dA are the beam FWHM produced by both devices. Because sometimes slightly
different definitions for the gain factor are used in the literature, is it difficult to directly
compare the individually reported values; however, under optimal conditions, values of a
few hundred to several thousand have been measured (Gao et al., 1998). When due to
external constraints, the x-ray source or lens characteristics are less than optimal, gain
factors between 10 and 100 are obtained. An example of such situation was described by
Worley et al. (1999), where a polycapillary lens was designed to fit into an existing m-XRF
spectrometer. The fact that the lens was only 29 mm long caused the gain factors to vary
between 124 (at 4 keV) and 2.5 (at 16 keV). When using a rotating anode as the x-ray
source (where the distance between anode spot and lens is of the order of 90 mm due to the
size of the anode tower), gain factors of 7–20 were observed (Vekemans et al., 1998),
despite the relatively high brilliance of the source (see Table 1). In a study to find the
Copyright © 2002 Marcel Dekker, Inc.
Figure 22 Fluorescent intensity profiles derived from a lens by perpendicular scanning of (a) a 20-
mm Cu=Sn wire through the beam at 50 kV tube operating voltage and (b) a 250-mm Pd wire at
various tube operating voltages in the range 30–50 kV. The importance of the transmission halo
increases with tube voltage.
optimal match between various commercially available x-ray sources and polycapillary
lenses, Bichlmeier et al. (2001) obtained gain factors between 30 and 300 for different
lenses coupled to a Mo-anode minifocus tube.
d. Analytical Characteristics
In Figure 20a, the elemental yields [in counts=s=(mg=cm2)] obtained by irradiating
NIST SRM 1832 and 1833 (thin-glass standards) using Mo excitation at 40 kV ex-
citation and three polycapillary lenses are compared to values obtained for the NSLS
beam line X26A m-SRXRF spectrometer (8 mm beam diameter) and to those obtained
by means of the monocapillary concentrator. For the polycapillary lenses, net yield
Copyright © 2002 Marcel Dekker, Inc.
values which are a factor of 3–5 lower are obtained than for the synchrotron setup. In
the setup employed, the lenses yielded a beam size of 160 mm. Considering a ratio of
(160=8)2 ¼ 400 in spot sizes, this indicates that by means of polycapillary lenses, a
flux density is obtained which is 1000–2000 lower than at the synchrotron beam line.
The absolute detection limits derived from the same spectra (Fig. 20b) are situated in
the 1–10-pg range, whereas the corresponding values for the monocapillary are one
order of magnitude better. Similar values were found by Bichlmeier et al. (2001),
whereas Gao et al. (1998) and Ding et al. (1998) reported absolute DL values in the
0.1–10-pg level. The relative detection limits (Fig. 20c) derived by irradiation of NIST
SRM 1577a bovine liver sample by means of the three lenses are situated at the 2–3-
ppm level (i.e., intermediate between the monocapillary characteristics and those of the
NSLS X-26A beam line). In a glass matrix (NIST SRM610), Bichlmeier et al. (2001)
reported relative DL values in the 10–100-ppm range for the elements Ca to Rb. It can
be concluded from these measurements that the polycapillary lenses are able to pro-
duce highly intense beams with moderate flux density and relatively small size, per-
mitting the determination of minor and trace constituents with a lateral resolution in
the range 20–150 mm.
F. Commercial Equipment
At present, four commercial companies market microbeam XRF instruments. Kevex-
Spectrace Inc. (Sunnyvale, CA) has been offering the Omicron instrument since 1986
(Nichols and Ryan, 1986; Yamamoto and Hosokawa, 1988), based on a 50-W minifocus
tube (250 250-mm spot on a Rh anode) and 10-, 30-, and 100-mm apertures. The
apertures are about 6 cm away from the anode spot and 3 mm from the sample; fluor-
escent radiation is collected with a 50-mm2 Si(Li) detector. Such a device can also be
retrofitted with a polycapillary lens (Worley et al., 1999). Hosokawa et al. (1997) of
Horiba Ltd. (Kyoto, Japan) developed a scanning x-ray analytical microscope (the XGT
2000 instrument) based on capillaries with paraboloidal inner walls to focus radiation
produced in a compact Rh-anode x-ray tube with a 200 200-mm2 spot size. A high-
purity Si(Li) detector collected fluorescent radiation. The capillary and detector nose are
situated in a small vacuum chamber sealed with a thin resin foil so that the airpath that
primary and fluorescent x-rays need to pass on their way to and from the sample is less
than 1 mm. Beams of 10 and 100 mm can be used in this device. The XGT 2000 is dis-
tributed in the United States and Europe by Oxford Instruments. COX Analytical In-
struments is a Swedish company (Göteborg, Sweden) that is marketing the
monocapillary-based ITRAX instrument offering m-beam XRF capacity down 5 mm; the
instrument is equipped with a high-flux x-ray tube (Attaelmanan et al., 1995). COX
Analytical Instruments also provides modular units for converting existing XRF equip-
ment into microbeam spectrometers. EDAX (Tilburg, The Netherlands) has introduced
the EAGLE m-probe device; this m-XRF instrument device uses a microfocus x-ray tube
and a polycapillary focusing unit to generate a 30-mm x-ray beam. All commercial
equipment includes a motorized scanning stage and optical microscopy system for sample
viewing (Haschke et al., 1998). More specifically intended for thickness gauging and
(other) applications in the microelectronics industry, Veeco Inc. (Plainview, NY) offers
several microscopic and small-spot XRF systems, the smallest beam size available being
50 mm. Seiko Instruments (Torrance, CA) also offers ‘‘small-spot’’ XRF equipment for
use in the microelectronics industry.
Copyright © 2002 Marcel Dekker, Inc.
IV. COLLECTION AND PROCESSING OF l-XRF DATA
There are three different types of measuring mode for the microbeam XRF instrument: point
analysis, line scanning, and area scanning. The scanning procedure is promoted by a regular
movement of the sample in the beam with a simultaneous recording of the intensities of
different characteristic x-ray energies by the x-ray detector. The operator normally controls
the sample position by manually inducing precise movements of the different sample stage
motors while the camera is viewing the sample surface. When the desired sample position is
found, the operator can normally select a point, line, or an area of interest.
During line and area scanning, the important choices for the operator are the step
size and exposure time. These parameters are interdependent and the choice is related to
the type of analytical data required and the time available for the analysis. Normally, the
step size is set equal to the effective diameter of the beam and the exposure time is selected
from the expected concentration of the elements being analyzed. However, in some cases,
it might be wise to operate with smaller step sizes (increasing the number of pixels) and
decreasing exposure times. Attaelmanan et al. (1994) described a scanning system where
the step size is always set equal to the pixel size of the optical image from the monitoring
microscope. The fluorescent data can then be superimposed directly on the micrograph
image (Rindby et al., 1992). This allowed the operator to check for correlations between
the distribution of certain elements and well-defined structures within the specimen.
As the amount of information obtained by the scan is entirely determined by the
time of analysis, no information is lost by scanning in steps smaller than the beam size.
The small step size might also allow for an increased image resolution by applying de-
convolution techniques if precise knowledge of the beam profiles can be implemented into
the algorithms (Zahrt, 1989).
The simplest way of implementing coordinated sample movement and spectrum
collection is to employ a motorized stage and detector electronics, which are both con-
trolled via PC plug-in cards or the equivalent. From within a scanning program, syn-
chronized motor movement and data collection and storage commands can then be issued.
50 50 0.25 ¼ 625 s (i.e., in about 10 min). If after this period of time, (some of) the
obtained images are too noisy, one or more additional frame scans can be performed and
the collected x-ray distribution can be added to the one collected during the previous
frame(s) until images of sufficient contrast and clarity are obtained. Usually, the number
of required frames is different for each elemental signal; as an illustration, Figure 23 shows
dynamically collected x-ray images using two, four, and ten 0.25-s=pixel frames obtained
from a sectioned piece of corroded Roman glass (see the electron micrograph shown in
Fig. 24a). Also shown are the corresponding images obtained using the ‘‘static’’ mode of
collection (60 s dwell time per pixel) to highlight the fact the both modes of operation in
the end yield essentially the same information.
Note that the images of Figure 23 result from integration of the spectral intensity
within various windows positioned around characteristic peaks and that, therefore, no
spectral background correction was performed; in Figure 24b, the complete set of
Copyright © 2002 Marcel Dekker, Inc.
Figure 24 (a) Backscatter electron micrograph of a polished section of Roman glass showing a cut
through a hemispherical corrosion growth. The complex morphology of the corrosion layer is
visible. The embedding resin and the original glass are visible in respectively the left bottom and right
upper corners. Scale bar length is 100 mm. (b) Full set of background-corrected m-XRF images
(see also Fig. 25).
(background corrected) elemental maps is shown. These maps are obtained by recording
the intensity from characteristic lines in the spectra for each pixel and storing these values
as two-dimensional image files. Thus, these images files can later be subjected to any kind
of image processing (see the next subsection). The recording can be done directly by
reading the counts from operator predefined ROIs (regions of interest) in the spectra or by
storing a complete spectrum from every pixel and doing peak-area evaluation and ele-
mental map reconstruction in a second phase.
B. Spectrum Processing
The data-reduction procedures associated with m-XRF spectrometry share a number of
features with those for m-PIXE data. In contrast to their bulk equivalents, scanning
m-PIXE and m-XRF are able to rapidly produce massive amounts of raw data: The col-
lection of a 64 64-pixel x-ray image may involve, for example, the collection of 8 Mbyte
of raw data and the processing of 4096 separate x-ray spectra. Although, in principle,
the same spectrum evaluation procedures that are used for the bulk technique data can be
Copyright © 2002 Marcel Dekker, Inc.
Figure 24 Continued
employed, the large size of these microscopic data sets makes the adaptation of these
procedures mandatory so that they can operate in a reliable way with minimal user in-
tervention on large numbers of spectra.
In m-PIXE setups, which frequently employ list mode data acquisition strategies,
x-ray line scans and x-ray maps for a particular element can be obtained by sorting the
recorded events using an appropriate energy window, centered around the peak energy of
the element in question. A similar procedure is also often employed during on- or off-line
processing of m-XRF data. This simple procedure is very useful for explorative
data analysis but implicitly assumes that within the energy window used, a single,
nonoverlapped peak is present with a high peak-to-background ratio so that the integrated
intensity within the window is a good estimate of the net intensity of the peak. Un-
fortunately, both for m-PIXE and m-XRF spectra, in general, these assumptions are not
valid: peak overlap frequently occurs in energy-dispersive x-ray spectra, whereas, espe-
cially for peaks corresponding to trace constituents, the background intensity below the
peak may be of the same order or larger than the net peak intensity. In these cases, the use
of too simple spectrum evaluation procedures may negate all the efforts that are made
Copyright © 2002 Marcel Dekker, Inc.
both during the data collection and during the further quantitative processing of the data
to increase the reliability of the final (trace) element concentration profiles or images.
As an example, in Figure 25, the result of processing in various ways the elemental
maps derived from the Roman glass sample of Figure 24 are shown. These images were
obtained respectively by plotting the integrated ROI intensities (Fig. 25a), by using a
simple ‘‘straight-line’’ background compensation interpolation approach between the ROI
boundaries (Fig. 25b), and by using net peak intensities obtained via least-squares fitting
of the individual spectra (Fig. 25c) (Vekemans et al., 1994; Van Espen et al., 1986;
Vekemans et al., 1995). See Chapter 4 for a detailed discussion on spectrum evaluation and
nonlinear least-squares fitting. During the latter processing, the evaluation of the spectra is
largely unsupervised: the fitting model (containing a list of photopeaks to be fitted and a
specification of the background modeling function to be used) is made a priori using a few
spectra from the set or by means of the sum spectrum. As a background compensation
algorithm, the use of a robust procedure is recommended when processing is unsupervised.
In this case, the background of each spectrum is described by linear combination of
mutually orthogonal polynomials, typically of the order of 10–15 (Vekemans et al., 1994,
1995). This algorithm permits the calculation of the appropriate background shape of the
spectra while this shape changes throughout the series of spectra (e.g., for spectra collected
from different phases in the scanned area) and also results in correct background values in
case spectra with much noise are processed. In Figure 24b, the full set of x-ray maps
obtained via method c is shown.
The most striking difference between the results of method a (no background
compensation) and methods b and c is that the high scatter background contribution in
the Ni and Br ROIs is eliminated. For images with much contrast between phases (see,
e.g., the Mn map), the three methods yield quasi-identical results. Method c resolves the
overlap between the KKb and CaKa signals, which results in a more detailed Ca map.
Similarly, instead of being a replica of the Mn distribution, the Fe-image appears quite
different in reality after proper processing. In the case of Ni, the use of the fitting pro-
cedure results in a less noisy image than when the straight-line background (method b) is
used for background correction. Finally, from the Br image, one can conclude that in the
corroded area, in general, more Br is present than in the original glass but that the noise
level does not permit to discern details.
It can therefore be concluded that for m-XRF data, in addition to resolving peak
overlap and correct background compensation in the neighborhood of intense matrix
lines, especially the ability of the (unsupervised) software to do reliable background
subtraction below the x-ray peaks of the heavier trace elements (atomic number >26) is
valuable. Using current-day workstations or personal computers, the computational effort
associated with the nonlinear least-squares evaluation does not represent a large fraction
of the total time required for acquisition and processing of the x-ray image data. As de-
scribed by Janssens et al. (1996b), by proper synchronization of different tasks, the eva-
luation of a spectrum corresponding to a particular pixel may be accomplished on-line
while the spectrum of the next pixel is being collected.
identified. However, for trace element distribution, the intensity is poor and only in-
dividual counts are recorded in each pixel, which makes it more difficult to identify any
structure in the sample. This is why it is so important to be able to correlate trace element
maps directly to the optical micrograph image. For trace element distributions in general,
the question is normally ‘‘where’’ rather than ‘‘how much.’’ Figure 26 shows a typical
example of the identification of a structure from the distribution of trace elements
(Stocklassa and Nilsson, 1993). For many of these maps, it is difficult to identify any
recognizable structure, but if you know what to look for (‘‘24’’), then most of the maps
show a significant increase in trace element concentration for that particular structure.
Although the contrast from trace mapping can be very poor, it can normally be
enhanced by a standard type of image processing technique, such as dynamics optimi-
zation, gray-level histogram equalization, and so forth. With good pixel statistics, de-
convolution or simple smoothing techniques can be used in order to achieve higher
lateral resolution or to utilize the screen dynamics of a color screen in a more efficient
way. For example, by utilizing the full dynamic of a color screen, three different ele-
mental maps can be superimposed at the same time using the RGB channels as separate
gray-scale images, as shown in Figure 27. In this way, correlations between different
elements can be studied in a convenient way; for noncorrelating elements, however, the
color mixing will complicate the interpretation.
et al., 1994) allows one to reduce this dataset (typically 1024 50 50 datapoints) to
dimensions nel ncol nrow , where nel is the number of elemental intensities derived from
each spectrum; nel typically has values in the range 5–20. In either case, the data cube can
be considered as a multivariate dataset, consisting of a number of objects (the pixels in the
image), each characterized by a number of properties (either nchan channel contents or a
vector of nel x-ray intensities).
Because during image scans, usually fairly short collection times are employed per
pixel (typically in the range 1–100 s), the photopeaks in individual spectra are often buried
in the background noise. This results in noisy images for the corresponding elements (see
Figs. 24–26). For performing quantitative analysis with a precision of the same order as
the accuracy offered by the calibration model employed, usually a better signal-to-noise
ratio is required.
The heterogeneous materials that are investigated in many cases are built up out of a
limited number of (quasi)homogeneous phases (Carpenter, 1997; Vekemans et al., 1997;
Rindby, 1993). Thus, m-XRF elemental maps usually can be segmented into a limited
number of regions, in which all pixels provided equal or similar compositional informa-
tion, or in the multivariate sense, they display a similar pattern of properties. After seg-
mentation of m-XRF images, it becomes possible to calculate the sum spectrum
corresponding to each region (by simply adding all spectra corresponding to each of the
pixels inside a region) having a much better signal-to-noise level than the spectra corre-
sponding to individual pixels. A similar approach is also employed in m-PIXE analysis
(Ryan et al., 1988; Svietlicki et al., 1993).
Segmentation of individual images can be done in various ways, by using, for
example, edge-enhancement filters or, on the basis of their corresponding, gray-level
histograms (Russ, 1995; Bonnet, 1995a). In the case of multivariate datasets (produced by
m-XRF or a similar technique), it is better to employ the information in all images
simultaneously during the segmentation process (Bonnet, 1995b; Geladi, 1995). This can be
done by employing appropriate mathematical procedures for (1) eliminating redundancies,
(2) distinguishing the significant information from random noise, and (3) splitting the
information in mutually orthogonal (i.e., noncorrelated) components. Techniques for
doing this have been reviewed by Bonnet (1995a, 1995b) and illustrated with datasets
Copyright © 2002 Marcel Dekker, Inc.
taken from electron probe microanalysis (Paque et al., 1990; Bright et al., 1988; Bright,
1995), auger electron spectrometry (Prutton et al., 1990), electron energy loss spectroscopy
(EELS) mapping (Jeanguillaume, 1985), and secondary ion microscopy (Van Espen et al.,
1992). For the specific case of m-XRF datasets, Cross et al. (1992) have described the use of
principal-component analysis (PCA) (see Geladi and Kowalski, 1986; Johnson and
Wickern, 1988; Press et al., 1992; Trebbia et al., 1995 for details) for colinearity removal
and dimensionality reduction.
By manually grouping pixels in the space of the resulting principal components
[a procedure called interactive correlation partitioning (ICP); see Paque et al., 1990],
semiautomated or supervised image segmentation was shown to be feasible for datasets in
which a limited number (three) of principal components were present. A procedure for
automated correlation partitioning (ACP) was proposed by Vekemans et al. (1997); this
method can also be applied to datasets in which more than three principal components are
present so that it is possible to automatically extract a small number of region-specific sum
spectra from a m-XRF image dataset.
Instead of attempting to reduce the dimensionality of the ncol nrow nel dataset by
looking for collinearity (or covariance) between the various elemental signals within the
images (as is done in PCA), in order to perform automated image segmentation, all
(ncol nrow ) pixels can be considered as objects characterized by a vector of nel properties
(a multivariate ‘‘fingerprint’’) and by using an algorithm such as K-means clustering
(KMC) (Bonnet, 1995a; Vekemans et al., 1997) to generate a limited number of pixel
groups having similar fingerprints.
A more sophisticated approach is to use a combination of PCA and K-means
pixel clustering: In a first step, the eigenvalues and principal-component images (or
eigenimages) of the original dataset are calculated using PCA; this serves to separate
the meaningful structure in the data from the noise. Second, a limited number of
these (noise-free) eigenimages are used as input to the KMC algorithm, allowing the
original elemental maps to be segmented. A concise overview of the mathematical
background of the PCA and KMC procedures can be found in Vekemans et al. (1997),
in addition to a discussion on the behavior of this procedure with datasets of variable
complexity.
The usefulness of this combined procedure can be illustrated by again considering
the x-ray images derived from the glass sample shown in Figure 24. The elemental maps
and electron micrograph show quite a complicated picture: In addition to the Mn-rich
areas where precipitation occured, several bands of different composition can be dis-
tinguished. The top layer corresponds to the original glass (high Ca signal) and the bottom
layer is the embedding resin. Many of the trace element maps (e.g., Ti, Cr, Ni, Zn, Br, Pb),
however, have a noisy appearance.
In Figure 28, the first nine principal-component images, resulting from PCA pro-
cessing of all the original maps of figure 24b are shown (step 1 in the procedure described
earlier). Images PC1–PC7 are contrast-rich, whereas in the remaining PC’s noise dom-
inates the pictures; PC1–PC7 explain more than 99% of the variance in the data. When
considering score and loading scatterplots [see Vekemans et al. (1997) for details], it ap-
pears very difficult, if not impossible, to find a pair of principal components where all pixel
clusters can be straightforwardly separated.
By using the seven meaningful PCA images as input for the KMC procedure, the
pixel grouping indicated in Figure 29 was obtained. The corresponding sum spectra are
shown in Figure 30. Cluster 7, which is associated with a high Compton scatter signal and
to none of the elemental signals, corresponds to the embedding resin; the corresponding
Copyright © 2002 Marcel Dekker, Inc.
Figure 28 Score images obtained by PCA of the x-ray maps of Figure 24b.
sum spectrum (Fig. 30) shows (almost) only background and a high Compton=Rayleigh
ratio. Sum spectrum 2 shows abundant Ca, Mn, and Fe signals, a clear Sr peak, and a
much lower Compton=Rayleigh ratio; this image segment corresponds to the original
glass. Cluster 3 shows a similar pattern but features clear Cu and Bi signals and probably
corresponds to a Bi-rich precipitate in the crack between the original glass and the cor-
roded layer. Clusters 6 and 9 contain only a few pixels of Bi ‘‘hot spots’’ but essentially
show the same pattern. Clusters 4 and 1 corresponds respectively to the MnO2 pre-
cipitation area and the phase it originated from (high Mn and Fe signals), whereas clusters
8 and 10 refer to respectively the top part of the leached layer, parallel with the surface,
and the precipitate crust formed on top of the surface (high Ca signal associated with a
relatively low Sr signal).
Copyright © 2002 Marcel Dekker, Inc.
Figure 29 Individual and compound segmentation masks obtained by KMC on the first seven
score images shown in Figure 28.
Comparison of the x-ray maps of Figure 24b and of the sum spectra in Figure 30
shows that although some of the x-ray maps show an excessively noisy distribution, most
of the sum spectra feature (sometimes weak but) predominatly noisy-free photopeaks,
allowing the corresponding major and trace constituents to be determined with an un-
certainty which mainly derives from the errors in the calibration model and not from
counting statistics.
D. Quantitative Analysis
1. General Considerations
X-ray fluorescence is well known for its ability to yield reliable quantitative data and,
as such, is used in many (industrial and research) laboratories for routine analysis of a
variety of material types. In many cases, in order to ensure high accuracy, wavelength
dispersive instrumentation and an empirical form of quantitative calibration is employed.
Relative accuracies better than a few percent can be reached for specific matrix types
(see Chapter 13). In laboratories which are more research oriented and=or where the type
of material being analyzed is more diverse, qualification models which (1) allow a rapid
Copyright © 2002 Marcel Dekker, Inc.
Figure 30 Image and cluster sum spectra corresponding to the clusters shown in Figure 29.
changeover from one matrix type to another, (2) do not require many calibration stan-
dards, and (3) readily allow for quantitative analysis of samples of different and inter-
mediate thickness are preferred and the so-called ‘‘fundamental parameter’’ (FP) method
is often employed (Shiraiwa and Fujino, 1966, 1968b).
As explained in detail in Chapter 13, these models essentially attempt to correct
mathematically for all phenomena (of first and higher order) that influence the detected
intensity of the characteristic radiation emitted by the sample. Because in these corrections
Copyright © 2002 Marcel Dekker, Inc.
a number of atomic constants of the elemental species involved are used (mass-absorption
coefficients, photoelectric cross section, fluorescent yields, transition probabilities, etc.),
the uncertainty on these fundamental parameters propagates into the uncertainty on the
elemental concentrations calculated with these methods. In addition, as the mathematical
expressions for the corrections are established for ideal(ized) geometrical conditions (e.g.,
parallel primary beam, perfectly flat sample surfaces, etc.), systematic deviations may be
introduced by using these models when the experimental reality does not conform to their
implicit assumptions. When models of sufficient sophistication are employed and
calibration is done with standards very similar to the unknown samples, accuracies of the
order of 2–5% can be obtained (He and Van Espen, 1991) for infinitely thick samples. For
partially transparent samples, some of these models include the estimation of the sample
mass (rd from the energy-dispersive spectrum; because this introduces additional errors,
the accuracy in these cases is of the order of 5–10% (Shiraiwa and Fujino, 1966, 1968b).
For quantification of m-XRF spectra, which normally are collected from samples
featuring some form of heterogeneity (either in sample shape, composition, or both)
(Janssens et al., 1996c), the use of FP-based models rather than empirical calibration
schemes is the most appropriate. Examples of heterogeneous sample types are, for ex-
ample, particles on a filter backing, fluid inclusions, in a geological material, multilayered
samples, and interfaces between two minerals inside a thin rock section.
2. Information Depth
As already mentioned (see Table 2), in most materials, x-rays have a much larger pene-
tration range compared to charged particles. The absorption corrections for the char-
acteristic x-rays are therefore much larger in the case of XRF than with EPXMA or
m-PIXE, at least when intermediate or thick samples are concerned. To correct for matrix
effects in EPXMA and m-PIXE, corrections have to be applied for the (continuous)
slowing down of the particles and for the attenuation of the x-rays on their way out of the
sample toward the detector. For protons, the slowing down does not strongly depend on
the matrix composition; also, the corrections due to this effect are partly compensated by
the self-absorption of the characteristic x-rays. In XRF, there is no such compensation,
as both the primary and fluorescent x-rays are attenuated. In EPXMA and PIXE, usually
the penetration range of the primary projectiles determines the depth of information; in
XRF, this is not the case: Because the characteristic radiation is usually more strongly
absorbed than the primary radiation, the information depth in most nonthin samples is
strongly element dependent.
One way of defining the information depth for element i (with characteristic)
radiation of energy Ei) is to equate it to the samples thickness d99%,i that would yield 99%
of the intensity that might be derived from an infinitely thick target with the same com-
position. In the case of monochromatic excitation, d99%,i satisfies the equation
dZ99%;i Z1
wrz lnð1 0:99Þ
e dz ¼ 0:99 ewrz dz , rd99%;i ¼ ð14Þ
wðE0 ; Ei Þ
0 0
where wðE0 ; Ei Þ ¼ mðE0 Þ cscðaÞ þ mðEi Þ cscðbÞ: mðE0 Þ and mðEi Þ represent the mass
absorption coefficients of the sample at the primary and fluorescent energy. a and b are the
incidence and takeoff angles, respectively (see Fig. 31). In the case of polychromatic
excitation, Eq. (14) needs to be integrated over the appropriate excitation energy range
and a value of d99%,i may be obtained numerically.
Copyright © 2002 Marcel Dekker, Inc.
Figure 31 Size of area at beam impaction point required to be homogeneous to permit application
of conventional calibration models to heterogeneous samples.
In Table 4, the information depth d99% is calculated for a few matrices, corre-
sponding to monochromatic MoKa excitation (E0 ¼ 17.4 keV); in addition to the energy
of the characteristic radiation, the major composition and density of the sample also
influences d99% is a significant way. This means that for a sample of given thickness, the
self-absorption can also vary considerably, the same sample being infinitely thick for
low-energy radiation while being (partially) transparent for the more energetic lines. To
illustrate this, in Table 5, the absorption factors Ai ¼ ½1 expðwi rd Þ=wi rd are listed for
a 100-mm-thick foil of the same materials as in Table 4. Ai is the ratio of the elemental
intensity derived from this foil to that when no self-absorption would take place in it.
Table 4 Information Depth (d99 %) for Different Elements in a Few Matrices (E0 ¼ 17.4 keV,
a ¼ b ¼ 45 ) (in mm)
the concentration to the x-ray intensity of element i derived from an infinitesimal sample
volume with thickness dz situated at a depth z would become (in first order)
Z z
dIi 0 0 0
/ I0 ðE0 ÞrðzÞci ðzÞti ðE0 Þ exp ½mL ðE0 ; z ÞcscðaÞ þ mL ðEi ; z Þ cscðbÞdz
dz
Z0 z
0 0
¼ I0 ðE0 ÞrðzÞci ðzÞti ðE0 Þ exp wL ðE0 ; Ei ; z Þdz ð15Þ
0
where mL ðE; zÞ ¼ mðE; zÞ rðzÞ is the (depth-dependent) linear absorption coefficient. In the
case that this sample consists of a number of discrete layers (index k from 1 to N ), each
homogeneous and which its own concentration ci,k, mass absorption coefficient mk ðEÞ,
density rk , and thickness dk , the contribution dIi;k to the intensity of element i from a slice
of thickness dz in the kth layer becomes
" #
dIi;k Xk
/ I0 ðE0 Þri ci;z ti ðE0 Þ exp ½mj ðE0 Þ cscðaÞ þ mj ðEi Þ cscðbÞrj zj
dz j¼1
" #
X k
¼ I0 ðE0 Þri ci;z ti ðE0 Þ exp wj ðE0 ; Ei Þrj zj ð16Þ
j¼1
with
8
< dj if j < k
zj ¼ P
k1
:z dn if j ¼ k
n1
For other heterogeneous samples with comparable (simple) heterogeneity, analogous ex-
pressions such as that of Eq. (16) may be established. As can be seen from this equation,
even for samples with limited heterogeneity, fairly complex expressions relate the observed
fluorescent intensity Ii to the characteristics of each layer. In the general case (sample
heterogeneous in three dimensions), the relation between Ii and the sample structure can
only established by numerical integration of the sample concentration and absorption
characteristics over the interaction volume of the primary and fluorescent x-rays.
What is important to note when considering Eqs. (15) and (16) (and their equivalents
for other cases) is that the magnitude of the exponential absorption factors depends very
Copyright © 2002 Marcel Dekker, Inc.
strongly on how the total path of, for example, the fluorescent x-rays [total length z csc(b)
in Eq. (16)] is distributed over the various phases. Small variations in this repartition may
cause large fluctuations in the integrated absorption coefficient along the total path. As an
example,P in Table 6, for various thickness combinations the integrated attenuation
exp½ j mj ðEi Þrj zj cscðbÞ is listed of several characteristic line energies Ei in a hypothe-
tical three-layered sample of 100 mm total thickness consisting of parallel layers of KAl-
Si3O8, NaAlSi3O8, and SiO2 bottom layer), assuming that this radiation originates from a
depth of 90 mm.
The data in Table 6 illustrate that in order to reliably correct for strong self-ab-
sorption (CaKa, CrKa, and FeKa), the matrix composition and detailed spatial distribu-
tion of the various phases that make up a heterogeneous sample must be known with high
accuracy. Whereas, in many cases, the matrix of these phases may be approximately
known (e.g., from measurements on more homogeneous parts of the material or by using
other methods), the details on the spatial distribution of the various phases inside natural
samples (e.g., a rock sample composed of different minerals) is hard to find. Table 6 also
shows that when the self-absorption is intermediate to weak (PbLa, SrKa, MoKa, in the
example), only an approximate knowledge of the repartition of phases is required.
In the former case (strong self-absorption), even when the sample geometry is known,
the position of the microbeam relative to this heterogeneous structure must also be known
in order to allow experimental spectra to be corrected for self-absorption. This is illustrated
for a very simple case in Figure 32. This figure shows experimental data collected
by scanning a NIST K961 glass microsphere of 32 mm diameter through a microbeam
of 8 8 mm2 in the x and y direction( cf. Fig. 31). When the sample movement is done in the
beam–sample–detector plane (x direction, see Figs. 32a and 32b), the characteristic ra-
diation of K, Ti, and so forth must pass through the body of the glass sphere on its way to
the detector and the intensity profiles are strongly and assymetrically deformed by self-
absorption in the glass. [see Voglis and Rindby (1995) for a more detailed discussion]. Note
that the maxima of the K, Ti, and so forth profiles do not coincide with each other or with
the geometrical center of the particle. Experimentally, it is therefore difficult to position the
beam exactly at the center of the particle because this position does not correspond with a
maximum in the detected count rate. The situation is simpler when the sample movement is
executed perpendicularly to the beam–detector plane ( y direction, see Figs. 32c and 32d);
in this case, the fluorescent radiation leaves the particle sideways toward, the detector and a
Figure 32 Experimental and simulated net intensity profiles resulting from scanning a 32-mm-
diameter K961 glass microsphere through a 8 8-mm2 white SR beam (a, b) in the beam–sample–
detector plane (horizontal sample movement). (c, d) perpendicular to this plane (vertical sample
movement), (e) m-XRF spectra collected during scan (a) at 10 mm to the left (lower curve) and right
(upper curve) of the geometrical center of the particle.
In the case of radiation for which the sample is infinitely thick, Lz;i ¼ d99%;i ; for more
penetrant radiation ðd99%;i > dsample Þ, the sample thickness determines the lateral size Lx of
the interaction volume in the x direction and therefore also the lateral resolution in this
direction. In the y direction and perpendicular to the beam–detector plane), only the beam
size dbeam,y defines the size of the interaction volume, at least when the x-ray detector
Copyright © 2002 Marcel Dekker, Inc.
subtends a small solid angle, as is usually the case at sychrotron m-XRF setups. In labo-
ratory instrumentation, sometimes large-area detectors are used (e.g., 80 mm2 at a fairly
close distance to the sample (e.g., 2 mm) so that the detector solid angle O becomes quite
large. In this case, the conditions for local homogeneity are somewhat stricter and the
following expressions must used for Lx and Ly (with tan2 g ¼ O=p):
Cert. Calc. Cert. Calc. Cert. Calc. Cert. Calc. Cert.a Calc. Cert.a Calc.
Element (ppm) (ppm) (ppm) (ppm) (%) (%) (%) (%) (ppm) (ppm) (ppm) (ppm)
K 1.47% 1.41% 0.97% 0.92% 0.39% 0.53 0.16% 2.24% 1.64 0.17%
Ca 2.09% 2.1% 0.012 < DL 7.95% 7.1 0.2% 1.20% 0.94 0.09%
Ti 2.94 2.74 0.46 0.52 1.53% 1.4 0.1% 0.15% 0.11 0.02%
Cr 19.7 18.4 17.5 17.3 280 417 148 4 (180)b 177 34
Mn 91 87 10.3 7.9 0.137% 0.18 0.03% 0.086% 0.10 0.02%
Fe 300 258 270 261 0.46 0.41 19.2 19.16 8.79% (ref ) 2.59% (ref )
Co 11.55 11.9 0.82 –
Ni 53.3 53.4 51.9 50.8 110 124 25 8
Cu 12 14.8 193 204
Zn 25 22.7 130 134 110 123 14 150 153 10
As 14 14.5 0.055 < DL
Br (10) 9.2 9
Rb 12 12.7 18.3 19.6 9 10 4 62 65 4
Sr 37 35.2 370 391 33 93 99 5
Y 26 22 6 102 105 4
Zr 155 177 29 540 557 15
Nb 5.38 5.30 15 13 5 59 52 3
Mo 10.3 10.4 3.24 3.35
Ba 121 153 43 600 592 16
La 13.0 12 7 52 57 3
Ce 32.5 34 13 119 125 5
Pr 4.6 54 14 13 2
Nd 21.3 29 11 58 72 3
Pb 45 48.9 0.34 <DL
a
The geological materials were irradiated as fused glass; the certified concentrations refer to the parent basaltic rocks.
b
Note: For details see Vincze et al., 1993, 1995a, 1995b.
Beam position
inelastic beam scatter. The effective mass thickness of the particle is approximated by
an iterative technique that employs experimentally detected elastic and inelastic scat-
tered intensities. The method is restricted to particles with homogeneous element dis-
tributions. In addition, the beam size had to be greater than the size of the particle and
the method worked best with light-element matrices. When used to analyze the NIST
glass standards K-411 and K-3267, over the particle size range, 50–150 mm, the results
for K-411 showed reasonable agreement with certified values. For K-3267, however, the
results for Si and Ca were nearly 30% different from the certified values. The poor
results for K-3267 were attributed to the more irregular and elongated shape of those
particles.
performed on the same particle after it was rotated over 90 relative to the beam,
equivalent maps are obtained which show, as expected, the ‘‘hot spots’’ at other locations
in the projection maps. In a simple case, two perpendicularly collected elemental ‘‘views’’
may be sufficient to ‘‘triangulate’’ the approximate position of the metal-rich areas inside
the particulate (see, e.g., the Cr and Fe maps in Figures 34a and 34b). However, when the
distribution of the element involved is more complicated, it is recommended to collect
more than two views.
If, instead of collecting XRF spectra while the samples is moved according to an XY
scanning pattern, the sample height Y is fixed and instead a XY scan is performed, during
which the horizontal position and orientation angle of the particle relative to the beam are
varied, instead of elemental maps, one obtains so-called ‘‘sinograms.’’ Figure 35 shows a
photograph and schematic drawing of the sample environment during XRF tomography
experiments. During tomographic data collection, the sample (mounted on a goniometer
head and on a rotation stage) is translated in a stepwise manner through the microbeam
(with a horizontal increment Dx of, e.g., 4 mm), after which the sample=goniometer head
assembly is rotated (with angular increment Dy of, e.g., 3 ) around the vertical axis and
the translation movement starts again until an angular range of 180 or 360 has been
covered.
In this manner, instead of performing a conventional XY scan, yielding multiple XY
elemental maps, an XY-scanning movement of the sample through the beam is executed,
which results in several elemental sinograms (see Fig. 36). Through the use of the ap-
propriate reconstruction algorithms, the latter can be back-projected into XZ maps of the
internal distribution of chemical elements in the horizontal plane through the sample at a
particular vertical position Y on the sample.
Copyright © 2002 Marcel Dekker, Inc.
Figure 35 Photograph and schematic drawing of the sample environment during XRF
tomography experiments: The sample is placed at the end of a thin glass tube mounted in a
manually adjustable goniometer head. This allows the center of the sample to be aligned with the
rotation axis of the motorized XYZY stage and with the capillary.
Figure 36 Conventional elemental XY maps, elemental sinograms (or XY maps), and elemental
tomographs (or XZ maps) of the fly-ash particle shown in Figure 34. Size in sinograms: 30 120
pixels, pixel size: 5 mm 3 ; size of resulting tomograms: 30 30 pixels size: 5 5 mm. T.B.
tomographic back-projection.
As an example, Figure 36 (middle panel) shows the sinograms obtained from the
particle shown in Figure 34 at the indicated height by employing an angular increment of
3 and a horizontal translation increment of 5 mm while the sample was kept at constant
height. In this way, the distribution of elements inside a plane situated at 35 mm above
the lower edge of the particle can be determined. From the elemental sinograms, by means
Copyright © 2002 Marcel Dekker, Inc.
of a mathematical technique called ‘‘weighted back-projection,’’ the equivalent elemental
tomograms (i.e., the distributions of the elements in the XZ plane through the particle at a
specific vertical coordinate) may be obtained (Fig. 36, right panel). For a more detailed
discussion on tomographic data collection strategies and back-projection theory and al-
gorithms, see, for example, Herman (1980) and Russ (1995).
As can be seen from the superposition of the S, V, and Cu tomograms, the metal-rich
areas of the particulate appear to be located on the outside of the particle. From the S
tomogram, it also appears that this particle is hollow and that the sulfur is distributed in a
shell around a central cavity. This type of structure has also been observed during other
experiments (Rindby, 1997a, 1997b). The ‘‘hot spots’’ apparently are small, quasi spherical
metal-rich particles which have become attached to this S-rich shell. However, not all ‘‘hot
spots’’ contain the heavy metals in the same relative amounts: for example, the three
enriched areas visible in the tomograms of Figure 36 all contain the elements Ca, V, Cr, Ni
and Mo, but only two show a detectable Cu signal. It is clear that by varying the height of
the sample relative to the beam, equivalent tomograms at various vertical positions inside
the particle may be obtained allowing a full three-dimensional visualization of the dis-
tribution of various chemical elements throughout the entire particle volume and this with
a spatial resolution equal to the primary beam size.
Whereas the example shown in Figure 36 highlights some of the analytical possi-
bilities the use of small, penetrant photon beams may have for three-dimensional in-
vestigation of certain sample types, it is appropriate also to point to some of the present
limitations of the technique in its current state. First, the technique is limited to fairly
small samples: Because it is the intensity of the fluorescent signals which is used to
construct the elemental tomograms, obviously only samples which are smaller or of
the same diameter as the sampling depth of the fluorescent radiation can be studied in
this way. In the case shown in Figure 36, this presents no problem, as the effective
density of the fly-ash sample (mainly consisting of a carbon skeleton and hollow areas) is
fairly low.
However, in the metal-rich areas, although they are fairly small in diameter
( 10–15 mm), the local density and atomic number is much higher and considerable self-
absorption occurs, especially for the fluorescent radiation of the lighter elements (e.g., S
and Ca). This can be seen in the S sinogram in Figure 36 in which the sinuous bands that
correspond to specific S-rich area in the XZ cross section under observation do not show
the same intensity under all angles, as parts of the particle that absorb the SKa radiation to
a different extent rotate into the path of the SKa x-rays on their way to the detector. The
self-absorption, which influences both the intensity of the primary beam on its way to a
particular volume element inside the particle and the intensity of the emerging fluorescent
radiation, must be corrected for in order to obtain quantitatively reliable tomographic
maps. In order to perform such a quantitative correction of the experimentally corrected
sinograms, the variation of the linear absorption coefficient mL(x, z, E) within the slice
under consideration (x, z) and at the primary and various fluorescent energies (E) involved
must be known, or, equivalently, the (heterogeneous) matrix composition within the slice
must be approximately known.
Studies with simulated data indicate that under such conditions, via the use of an
iterative correction procedure, self-absorption phenomena may be corrected and the initial
compositional distribution iteratively refined, provided that the self-absorption of the
fluorescent signals is not too severe (Nellessen, 1996; Vincze et al., 1999b, 1999c); con-
siderable efforts, however, still are necessary to solve this heterogeneous calibration pro-
blem in a general way (Simionovici et al., 2000). Because, at the moment, such correction
Copyright © 2002 Marcel Dekker, Inc.
algorithms are not (yet) available in validated form (Sakellariou et al., 1997a, 1997b), the
tomograms in Figure 36 can only be used in a semiquantitative manner (by implicitely
assuming either a nonexistent or a constant self-absorption inside the particle) (Vekemans
et al., 2000).
Note: The elemental XY projection images, sinograms, and XZ tomograms show in
Figure 36 were obtained at the m-SRXRF setup at HASYLAB Beamline L by means of an
ellipsoidal lead glass capillary, providing a beam of 4–5 mm diameter (Janssens et al.,
1998). Similar measurements, but with less good lateral resolution, could also be per-
formed by employing a monocapillary-based laboratory m-XRF spectrometer using at
15-mm x-ray beam from a Mo anode operated at 40 kV.
V. APPLICATIONS
A. Industrial Applications
X-ray fluorescence has been used in the industrial laboratory and on the factory floor to
analyze small specimens or small areas for many years (Adler et al., 1960; Miller, 1961;
Shiraiwa and Fujino, 1968a; Stebel and Silverman, 1984). By simply restricting the x-ray
beam with vendor-supplied apertures or other special attachments, spatially resolved
quantitative data could be obtained without using expensive electron beam equipment.
Special attachments for conventional XRF instrumentation reduced beam sizes to 100 mm
for application to a variety of problems in the industrial laboratory. It became obvious
that many materials problems such as plating and film thickness measurement could be
solved with the lower resolution of the ‘‘m-XRF’’ technique.
In the mid-1980s, two developments occurred which led to increased industrial ac-
ceptance of m-XRF. The first instrument specifically designed for energy-dispersive m-XRF
was developed and marketed in 1986 (Wherry and Cross, 1986). Perhaps more sig-
nificantly, it was capable of element imaging, allowing the user, instead of simple point
analysis, to superimpose elemental maps and optical images. The availability of specialized
instrumentation led to applications involving microelectronic devices, industrial waste
screening, forensics, wood, paper, and other problems in materials science. The other
significant innovation in the mid-1980s was the development of more intense laboratory
sources using glass-capillary optics.
2. Microelectronics Industry
In the microelectronics industry, thin-film technology is being used to solve a number of
connector and packaging problems that have developed with the downsizing of in-
tegrated circuits. The lead counts on VLSI (very large-scale integration) circuits have
increased in proportion to the increase in the number of components per chip. Close
tolerances and narrow statistical quality control parameters have placed great demands
on the characterization of these thin films, m-XRF is the method of choice for the
measurement of film thickness and composition in the small areas involved in electronic
devices.
Wherry and Cross (1986) described a m-XRF instrument that had several advantages
over conventional small-spot, proportional detector systems with respect to applications
in the microelectronics industry. The instrument used a solid-state Si(Li) detector and had
imaging capabilities for revealing hidden chemistry information and qualitative in-
formation on film thickness homogeneity. Anderson and co-workers (1991) presented
several examples of the utility of the new m-XRF analyzer to support the manufacture of
integrated circuits. Tape Automated Bonding (TAB) is a method for bonding
input=output leads to VLSI circuits. They accurately measured the Sn film thickness on
TAB tape by monitoring the SnKa line with a 100-s counting time. The m-XRF technique
was said to be superior to Auger and SEM–EDXRF analyses because of the larger
thickness range covered, lower detection limits for impurity elements, and simplified
measurements procedure (no sample preparation needed and ability to operate in air,
helium, and=or low vacuum). They also demonstrated the possibility of monitoring the
Cu=Sn intermetallic formation by measuring the CuK and SnL lines via the use of a thin-
window solid-state detector.
Copyright © 2002 Marcel Dekker, Inc.
3. Waste Characterization
The characterization of industrial wastes in an area of concern because of potential en-
vironmental pollution due to improper wastes disposal. The identity and quantity of
hazardous elements present must be determined before disposal. In addition, new disposal
methods must be developed to contain hazardous materials and prevent leaching into the
soil. m-XRF has played a role in both of these areas.
a. Electronic Components
Tissot and Boehme (1994) suggested the use of m-XRF as a replacement for the Toxic
Characterization Leaching Procedure (TCLP) in the classification of waste from electronic
components. TCLP is an expensive (US $2000=sample) wet-chemical process that is re-
quired by the United States Environmental Protection Agency as the basis for compliance
with the Resource Conservation and Recovery Act (RCRA). It involves leaching the waste
material with dilute acetic acid; if specific toxic element concentrations exceed as pre-
scribed amount in the leachate, then sample and leachate must be disposed of in a ha-
zardous landfill. In the m-XRF-based procedure, large areas of fractured waste are scanned
in air; samples required essentially no special preparation. No other analytical imaging
method could have been used to obtain the required data.
Electronic components from dismantled nuclear weapons at Sandia National La-
boratory were broken up using an industrial forge hammer to create samples having
particle sizes in the 2–10 mm range. These samples were used for the TCLP and for m-XRF
analysis. Sample chambers (4 in. 5 in., filled one layer deep with granulate) were scanned
with a 100 mm x-ray beam from a Mo-target x-ray tube (50 kV). In a sample in which
TCLP found Cd (10 1 ppm) and Pb (30 10 ppm) levels above the regulatory limit (1 and
5 ppm, respectively), these elements were also detected by means of m-XRF. In addition,
Cr was found to be present by m-XRF, but this element was not present at hazardous levels
in the leachate. Though present at relatively low concentration levels in the sample as a
whole, the presence of these elements could readily be determined by m-XRF because in
the crushed electronic components, they are concentrated in small areas that are still
significantly larger than the beam. Thus, one small particle containing a large con-
centration of Pb would yield a large XRF count rate and that could correspond to pm
quantities for the entire sample. Furthermore, element associations in the electronic
component waste, determined by comparing m-XRF element maps, were used to estimate
the relative toxicity of the hazardous elements. For example, Cr, associated in stable form
with Fe and Ni as stainless steel, was not expected to toxic and, therefore, was not
leachable in the TCLP, Pb, on the other hand, was associated with Sn as solder and was
expected to readily undergo acid attack.
4. Homogeneity Testing
Wide-area scans, using beam sizes ranging from a few hundred micrometers to 1 mm, are
used in the steel industry to determine gross element segregation. Welfringer and co-
workers described the m-XRF analysis of an area 150 150 mm2 on as cast 100C6 steel
using a spot size of 1 mm2 (Welfringer et al., 1992). The authors pointed out that the easily
detectable Cr and Mn tend to cosegregate with the minor elements P and S, which undergo
extensive segregation during melting. The detection and characterization of segregation
allows the metallurgist to predict the properties of the product and make adjustments to
the continuous casting machine. Quantitative maps of Cr and Mn were produced from
calibration curves. It was found that a milled surface finish was adequate for the m-XRF
technique and that the data were as good as or better than that obtained by metallographic
etching or the more expensive and complicated electron microprobe technique.
B. Environmental Applications
An overview of environmental application of laboratory and synchrotron-based
microbeam XRF can be found in Osan et al. (2000). In what follows, primarily the
applications that can be realized with laboratory m-XRF setups are briefly outlined.
Kuczumow et al. (1995) aimed to obtain density variational data within one in-
dividual growth ring (using 20 mm resolution) and to record long-range density changes
(80 mm resolution). The primary intensity obtained from a conventional x-ray tube (Cu
anode) with capillary focusing (7-, 18-, 23- and 79-mm openings) was sufficient for
transmission and coherent scattering measurements. Core samples of two similar tree
species (Pinus silvestris from Poland and Picea abies from northern Sweden) but growing
in different climatic conditions were analyzed. Ten-year scans showed that the transmis-
sion data and the temperature distribution were in good correlation; each increase in
ambient temperature caused a density decrease. From the long-range densitometric
measurements, year rings with extremely high mean density could be selected, as an evi-
dence of low-temperature years (Jones et al., 1995) (e.g., as a result of volcanic explosions,
causing a global temperature decrease due to dust emission into the atmosphere). By
means of a 18-mm capillary beam (Kuczumov, 1995), within individual year rings sig-
nificant differences between the Polish and Swedish wood were found. In the former wood
type, a typical yearly density variation of pine growing in the temperate zone was found
showing a thick layer of dense ‘‘late’’ wood. The Swedish wood lacked this layer; its
density rapidly decreased in spring and sharply increased in autumn. When the densito-
metric measurement was carried out in the scattering mode, Attaelmanan et al. (1994)
demonstrated that some nutritional elements such as Ca and Mn followed the density
variations, whereas others (K, Cl) were completely uncorrelated. Kuczumow et al. (1999a,
1999b) also studied differences in density and in composition of year rings in petrified
wood by means a combination of techniques.
Copyright © 2002 Marcel Dekker, Inc.
Laboratory x-ray tube systems have also been used for biomedical applications. This
type of investigation was reviewed by Fukumoto et al. (1999). Already in 1990, Fukumoto
et al. (1990) demonstrated the possibility of performing trace element mapping on tumor
tissues and living plant leaves by means of a 200-mm beam extracted from a conventional
x-ray tube. Trace element mapping on tree leaves was later demonstrated by Carpenter
(1995b), whereas Voglis et al. (1993) performed high-resolution mapping of Ca in bone.
4. Document Forgeries
The ability to analyze materials in air and in a nondestructive manner makes m-XRF an
excellent technique for the analysis of valuable documents (e.g., to determine their au-
thenticity). Typically, different inks, while having the same visible appearance, will have
different chemical compositions. Although m-PIXE and m-SRXRF has been used to ex-
amine documents such as the Gutenberg Bible and other early printed works (Cahil, 1981,
1984; Mommsen, 1996), the relative simplicity and nondestructive nature of m-XRF make
it an attractive analytical alternative.
Larsson (1991) and Stocklassa and Nilsson (1993) described analysis by m-XRF of an
500-year-old Swedish possession letter (dated April 1, 1499). The document, a sales
contract for an estate, showed signs of alteration. It is suspected that the alteration was
made in the 1530s when the Swedish king gave land back to certain nobles, who had been
Copyright © 2002 Marcel Dekker, Inc.
Table 9 Quantitative Results on Local Analysis of Different Locations on the Nefer-Hotep Statue Shown in Figure 37
Crown
Right Head
% (w=w) leg Front band Lower Upper Front Back Bottom A Bottom A Bottom B Bottom C
Fe 0.12 0.14 0.25 0.09 0.09 0.11 0.11 0.15 0.09 0.07 0.08
Cu 81.2 82 76.5 72.8 75.8 72.6 70.7 71.4 80 84.9 84.8
Ag 0.05 0.04 0.12 0.18 0.13 0.17 0.19 0.13 0.07 0.06 0.06
Sn 13.7 12.9 14.8 19.8 17.5 22.3 21.6 21 15.3 12.2 12.2
Pb 4.8 4.5 7.6 6.5 6.2 4.3 7 6.7 4.6 2.5 2.5
Figure 38 The Ca (bottom panel), Mn, Fe, and Zn (top panel) maps of an area of a Swedish
possession letter. In the Ca map, the family name ‘‘Gäsmested i Böre’’ is visible, corresponding to the
visible (but falsified) writing; in the Zn map of the same area, however, a completely different name
reading ‘‘Bøtinge i Asbo’’ is visible.
D. Forensic Applications
In view of the nondestructive nature of m-XRF and the minimal specimen preparation
required, it is an ideal tool for the analysis of material found at crime scenes and for the
investigation forgeries. The use of m-XRF permits to avoid the destructive of valuable
evidence, as is the case for many other methods commonly employed in forensic analytical
chemistry. In Stocklassa et al. (1992), the analysis of various forensic materials such as
single hair fibers, paint chips, glass particles, and ball-point pen ink using m-XRF is de-
scribed.
1. Hair Analysis
With respect to the analysis of trace elements in hair, m-XRF has the advantage of being
nondestructive, quick, and able to map element distributions both along the length of the
fiber and across its thickness.
Trace elements in hair originate from both internal (endogeneous) and external
(exogeneous) sources (Chatt and Katz, 1988). Exogeneous elements originate from en-
vironmental sources such as tap water or air pollution. The assumption is generally made
that exogeneous elements will be concentrated nearer the surface of hair strands and the
concentrations of those elements will increase with length, as long as the external de-
position continues; in contrast, endogeneous elements should be distributed relatively
uniformly throughout the thickness of the fiber, the relative amounts of each element
determined by diet and individual metabolism.
In a few studies, the claim is made that within a person’s hair, the compositional
fingerprint among individual hair is relatively constant and that this patterns is unique to a
single individual (Toribara et al., 1982; Seta et al., 1982, Valkovic et al., 1973); thus,
(m-XRF) analysis of a single hair strand could potentially provide the forensic scientist
with unique information to match hair fibers found at crime scenes. Engström et al. (1989)
showed that detection limits of 1 ppm were possible in the analysis of trace elements in a
single hair fiber using a m-XRF analyzer with a 200-mm beam. With the same instrument
and a small (19 mm) Cr-anode beam, Larsson et al. (1990) obtained self-absorption cor-
rected radial distributions of sulfur and calcium from single hair fibers without cross
sectioning the fiber. The Ca distribution was maximum at the surface of the fiber and
showed a minimum in the center of the fiber, consistent with other work indicating the
presence of significant amounts of exogenous Ca in hair from washing in tap water and
with PIXE observations by Cookson and Pilling (1975).
Toribara (1987) constructed an XRF analyzer for the analysis of single hair fibers in
1–2-mm steps along the length of the hair. The authors pointed out that the data obtained
from multiple hair samples from a single individual did not appear to vary significantly
and that forensic applications were possible. Detection limits were found to be similar to
those for PIXE, or less than about 30 ppm. The longitudinal distribution of Hg in the hair
of an Iraqi woman who had consumed wheat produced by seed that had been treated
with methyl mercury insecticide was determined. From the fact that hair grows at an
average rate of about 1 cm per month, the period of ingestion of the Hg could be de-
termined. This is similar to PIXE data used to show slow poisoning of Napolean by As,
except that it was done faster and with higher spatial resolution (Forshufvud et al., 1961;
Copyright © 2002 Marcel Dekker, Inc.
Smith et al., 1962). The same instrument was also used to determine trace elements in
fibrous materials: chemical differences could be found in carpets made by different
manufacturers, pointing to the possibility to match=identify single textile fibers found on
victims and=or criminals.
Adler LA, Axelrod J, Branco JJR. Adv X-ray Anal 2:167, 1960.
Anderson TA, Evans KL, Carney F, Carney G, Berg H, Gregory R. Spectroscopy 28 (July=August
1991).
Asher RK, Ruiz ER. Plating Surf Finish 73:30, 1986.
Attaelmanan A. On resolution enhancement of capillary based scanning x-ray microfluorescence
spectroscopy. PhD dissertation, Chalmers University of Technology, Göteborg, Sweden, 1995.
Attaelmanan A, Rindby A, Voglis P, Shermeat A. Nucl Instrum Methods Phys Res B 82:481, 1993.
Attaelmanan A, Larsson S, Rindby A, Voglis P, Kuczumow A. Rev Sci Instrum 65:7, 1994.
Attaelmanan A, Voglis P, Rindby A, Larsson S, Engström P. Rev Sci Instrum 66:24, 1995.
Azaroff LV, Kaplow R, Kato N, Weiss RJ, Wilson ACJ, Young RA. X-ray Diffraction. New York:
McGraw-Hill, 1974.
Bertsch PM, Hunter DB, Sutton SR, Bajt S, Rivers ML. Environ Sci Technol 28:980, 1994.
Bertucci G. Nucl Instrum Methods Phys Res A322:583, 1992.
Bethe A. Ann Phys 5:325, 1930.
Bichlmeier S, Janssens K, Heckel J, Gibson D, Hoffmann P, Ortner H. X-ray Spectrom. 30:8, 2001.
Boehme DR. Adv X-ray Anal 30:39, 1987.
Bonnet N. Microchimi Acta, 120:195, 1995a.
Bonnet N. Ultramicroscopy 57:17, 1995b.
Bright D, Microbeam Anal 4:151, 1995.
Bright DS, Newbury DE, Marinenko R. In: Newbury DE, ed. Proceedings of the 17th Microbeam
Analysis Society Meeting. San Francisco: San Francisco Press, 1988, p 18.
Burton A. Biological monitoring of environmental contaminants. PhD dissertation, King’s College,
London, 1986.
Bush GT, Stebel MD. Connectors 85:129, 1985.
Cahil T, Kusko B, Schwab RN. Nucl Instrum Methods Phys Res B181:205, 1981.
Cahil T, Kusko B, Eldred RA, Schwab RN. Archaeometry 26:3, 1984.
Carpenter DA. X-ray Spectrom 18:253, 1989.
Carpenter D, Ed. X-ray Spectrom 26:300–387, 1997 (entire issue).
Carpenter DA. In: Janssens K, Adams F, Rindby A, eds. Microscopic X-ray Fluorescence Analysis.
Chichester: Wiley, 2000, p 347.
Carpenter DA, Taylor MA. Adv X-ray Anal 34:217, 1991.
Carpenter DA, Gorin A, Shor JT. Adv X-Ray Anal 38:557, 1995a.
Carpenter DA, Taylor MA, Lawson RL. J Trace Microprobe Technol 13:141, 1995b.
Carpenter DA, Taylor MA, Holcombe CE. Adv X-ray Anal 32:115, 1989.
Carpenter D, Haney L, Lawson R, Poirier P, Paper presented at the 36th Annual Denver X-ray
Conference, Denver, 1987.
Carpenter D, Lawson R, Taylor M, Poirier P, Morgan R, Haney L. In: Newbury DE, ed.
Proceedings of the 17th Microbeam Analysis Society Meeting. San Francisco: San Francisco
Press, 1988, p 391.
Chatt A, Katz SA. Hair Analysis: Applications in the Biomedical and Environmental Sciences.
New York: VCH, 1988.
Chesley FG. Rev Sci Instrum 18:442, 1947.
Chung PS, Pantell RH. Electron Lett 13:527, 1977.
Compton AH. Philos Mag Series 6 45:270, 1923.
Cookson JA, Pilling FD. Phys Med Biol 20:1015, 1975.
Cosslett VE, Nixon WC. Proc Roy Soc B140:422, 1952.
Criss JW, Birks L. Anal Chem 40:1080, 1968.
Cross BJ, Wherry DC. Thin Solid Films 166:263, 1988.
Cross B, Lamb RD, Ma S, Paque JM. Adv X-ray Anal 35:1255, 1992.
Czechanshi M, Measurement of precious metal coating thickness using x-ray fluorescence
instrumentation. Proceedings of the 9th International Precious Metals Conference, 1986, p 457.
Willy Maenhaut
Ghent University, Ghent, Belgium
Klas G. Malmqvist
Lund University and Lund Institute of Technology, Lund, Sweden
I. INTRODUCTION
In 1970, Johansson et al. (1970) demonstrated that the bombardment of a specimen with
protons of a few megaelectron volts (MeV) gives rise to the emission of characteristic x-rays
and that this can form the basis for a highly sensitive elemental analysis. This landmark
article formed the starting point of the x-ray emission analysis technique which became
known as particle-induced x-ray emission analysis (PIXE). Although PIXE is sometimes
considered a variant of x-ray fluorescence (XRF), such a classification is not correct in a
strict sense, because the technique does not rely on excitation of the sample by x-rays.
Instead, heavy charged particles (i.e., protons, a-particles, or heavy ions) are used in PIXE
to create inner-shell vacancies in the atoms of the specimen. As in XRF and electron-probe
microanalysis (EPMA), the characteristic x-rays produced by deexcitation of the vacancies
can be measured by either a wavelength-dispersive or an energy-dispersive detection sys-
tem. However, whereas the two detection systems are employed in both XRF and EPMA,
an energy-dispersive spectrometer with a Si(Li) detector is almost exclusively used in PIXE.
The incident charged-particle beams in PIXE are invariably generated by particle accel-
erators. For the great majority of PIXE work, protons of 1–4 MeV, which can be produced by
small accelerators (e.g., Van de Graaff accelerators or compact cyclotrons), are employed. Such
small accelerators are also used in other analytical techniques that utilize ions of a few mega-
electron volts per mass unit, such as Rutherford backscattering spectrometry (RBS), nuclear
reaction analysis (NRA), charged-particle activation analysis (CPAA), and accelerator mass
spectrometry (AMS). Because of their common use of ion beams, these techniques and PIXE
are often jointly referred to as ion beam analysis (IBA) techniques. Furthermore, because
the same incident particle type and energy may be used in several of the techniques, the
simultaneous analysis of a sample by PIXE and some other IBA techniques (particularly RBS
and NRA) is often feasible, so that the elemental coverage may be increased down
to the very light elements and=or information on the depth distribution may be obtained.
Compared to x-rays, protons or other heavy charged particles have the advantage
that they can be focused by electrostatic or (electro)magnetic lenses and may be trans-
ported over large distances without loss in beam intensity. As a result, incident fluence
1 dE
SðEÞ ¼ ð1Þ
r dx
where r is the density of the stopping material and x is the distance. As defined here, S(E )
is expressed in units of kiloelectron volt per gram per square centimeter.
Copyright © 2002 Marcel Dekker, Inc.
Numerous experimental measurements of stopping powers are available. They
formed the database to fit the parameters of semiempirical equations (Anderson and
Ziegler, 1977; Ziegler et al., 1985), which are now commonly used to obtain the stopping
powers for all elements of the periodic table. For the energy range of 1–4 MeV, which is
most important in PIXE, the uncertainty of the values calculated with the semiempirical
equations is estimated at less than 3%.
The stopping power for compounds or more complex matrices is obtained from
those of the constituent elements through the Bragg–Kleemann additivity rule:
X
n
Smatr ðEÞ ¼ wi Si ðEÞ ð2Þ
i¼1
where wi and Si ðEÞ are the mass fraction and stopping power of constituent element i,
respectively.
The total path length R of an ion may easily be obtained by integration of the
stopping powers:
Z0
dE
R¼ ð3Þ
SðEÞ
E0
where E0 is the incident ion energy. Although the total path length is larger than the
projected range, the difference between the two is smaller than 1% for incident protons of
a few megaelectron volts.
L1 ¼ sL1 o1
sX ð6Þ
I
0
L3 ¼ ½sL3 þ sL2 f23 þ sL1 ð f13 þ f12 f23 þ f13 Þo3
sX ð8Þ
I I I
0
where oi denotes the fluorescence yield, fij denotes the Coster–Kronig probability, and f13
is the small radiative intrashell vacancy transfer probability.
Calculation of the x-ray production cross section from the ionization cross sections
thus involves additional atomic parameters; that is, fluorescence yields and fractional
radiative widths in the case of the K shell and fluorescence yields, Coster–Kronig yields,
and fractional radiative widths for the L shell. As to the K fractional radiative widths,
either the experimental data of Salem et al. (1974) or the theoretical values of Scofield
(1974a) that are derived from Dirac–Hartree–Fock (DHF) calculations are considered
accurate within 1%. There is, however, a small failure of the theory in the atomic number
region 22 < Z < 30, where experimenters (Perujo et al., 1987) agree on a deviation of a
few percent. The K fluorescence yields are usually taken from Krause (1979). These data
Copyright © 2002 Marcel Dekker, Inc.
are quite accurate for atomic numbers above Z ¼ 20, but the situation for the lighter
elements is less clear (Maenhaut and Raemdonck, 1984; Paul, 1989). For these lighter
elements, one currently prefers the data from Hubbell (1989). For the case of the L x-rays,
Cohen and Clayton (1987) and Cohen (1990) suggest employing the fractional radiative
widths of Salem et al. (1974) and the Krause (1979) fluorescence and Coster–Kronig yields,
and Cohen and Harrigan (1986) used this approach to convert their table of ECPSSR L
ionization cross sections (Cohen and Harrigan, 1985) into a very useful table of produc-
tion cross sections for up to 16 individual L x-ray lines. Campbell (1988), however, ad-
vocates the use of the DHF radiative widths of Scofield (1974b) and of the DHS
fluorescence and Coster–Kronig yields of Chen et al. (1981) in combination with the
tabulated DHS L ionization cross sections of Chen and Crasemann (1985, 1989).
A very practical alternative to tabulated theoretical ionization and x-ray production
cross sections, particularly with computer calculations in mind, are parameterized or
analytical formulas which are obtained by fitting polynomial expressions to theoretical or
empirical cross-sectional data. The equations derived by Johansson and Johansson (1976)
have been widely used by PIXE analysts in the past, but it is now known that they pro-
gressively underpredict cross sections with increasing Z of the target element. More ac-
curate formulas, in which similar functions were often used as in the Johansson and
Johansson equations, have been presented by Paul (1984), Cohen and Clayton (1987),
Miyagawa et al. (1988), Johansson and Campbell (1988), and Sow et al. (1993).
For the proton energy range of 1–4 MeV typically used in PIXE, the ionization (and
x-ray production) cross sections increase with increasing proton energy and decrease with
Figure 1 K and L shell ionization cross sections in barns (1 barn ¼ 10 7 24 cm2) as a function of
proton energy and target atom. The values are theoretical ECPSSR predictions (Cohen and
Harrigan, 1985). (From Johansson and Campbell, 1988. Copyright John Wiley & Sons, Ltd.
Reproduced with permission.)
3. Elastic Encounters
Protons or other heavy charged particles that pass near atomic nuclei can be scattered
elastically (i.e., without causing nuclear or atomic excitation). The cross section for the
elastic scattering is strongly dependent on the scattering angle and on the atomic number
(Z ) of the scattering nuclide; it decreases with increasing angle and increases with increasing
Z. The energy of the scattered particles also increases with Z. By measuring the elastically
scattered particles, information can be obtained on the elemental composition of the sample
and on the distribution of the elements with depth. This has given rise to the widely used
technique of Rutherford backscattering spectrometry (Chu et al., 1978), which can often be
elegantly employed in combination with PIXE to provide complementary information. On
the other hand, the energetic scattered particles can also cause problems in PIXE. Indeed,
when they enter the Si(Li) detector, their large energy is deposited there, and this gives rise
to serious resolution deterioration and other electronic problems in many x-ray pulse
processing systems (Mingay, 1983). A solution to this problem is to interpose a light-ele-
ment absorber (preferably Be) between specimen and Si(Li) detector (e.g., Maenhaut and
Raemdonck, 1984), but this invariably results in loss of sensitivity for the light elements.
4. Nuclear Reactions
In addition to elastic encounters between incident particles and target nuclei, various in-
elastic interactions or nuclear reactions are possible. They include ( p, g), ( p, p0 g), and
( p, ag) reactions when protons are used as incident particles. The cross sections for those
reactions vary in a rather irregular way with target nuclide and with incident particle
energy. They generally increase with increasing energy but may exhibit intense resonance
peaks at particular energies. Also, because of the Coulomb barrier, the cross sections are
smaller for the heavier target elements than for lighter ones. By detecting the promptly
emitted g-rays or charged particles for these nuclear reactions, elemental analysis and
depth profiling of certain elements is possible. The analytical technique that employs these
possibilities is referred to as nuclear reaction analysis. A comprehensive treatment of this
technique can be found in the Handbook of Modern Ion Beam Materials Analysis (Tesmer
et al., 1995). The technique is also briefly presented in Sec. VII.B.
5. Other Interactions
As indicated earlier, the interactions of charged particles and matter occur mainly through
inelastic Coulombic encounters with bound electrons. This results in electron excitation
and ionization but also in secondary phenomena that contribute to the continuous photon
background in the PIXE spectrum. Another interaction which contributes to that back-
ground is projectile bremsstrahlung. All these interactions are discussed in some detail here.
Figure 2 Schematic representation of the relative contributions of QFEB, SEB, and AB to the
electron bremsstrahlung background. For an explanation of the acronyms and symbols, see the text.
(From Ishii and Morita, 1987; with permission from Elsevier Science.)
3. g-Ray Background
The prompt g-rays emitted as a result of nuclear reactions between the projectiles and the
target atoms are generally quite energetic and are therefore far outside the 0–30-keV
energy range typically observed in PIXE. However, through Compton interactions of
these g-rays with the Si(Li) detector, a wide spectrum of electron energies is generated,
and this, in turn, gives rise to a slowly varying continuum in the PIXE spectrum. This
Compton-scattering background is not predictable in the sense that the bremsstrahlung
background is predictable, as it does not depend on the matrix composition of the
target alone but rather on the presence of particular elements with large cross sections
for nuclear reactions yielding g-rays. Na and, particularly, F are examples of elements
that are often responsible for a noticeable or even significant g-ray background. A large
fraction of the g-ray background in the PIXE spectrum may originate from nuclear
reactions that do not take place in the specimen but instead in the beam collimators,
Faraday cup, or various other parts of the PIXE irradiation chamber [even the x-ray
absorber placed in front of the Si(Li) detector]. As discussed by Johansson and Campbell
(1988) and Johansson et al. (1995) and briefly indicated in Sec. III.B.1, the materials
for the collimators and various other parts of the chamber should therefore be selected
with care.
III. INSTRUMENTATION
A. Accelerators
The ion beams used for PIXE analysis are produced in an accelerator. The energy range
required, 1–4 MeV=u, means that relatively small accelerators are sufficient. Electrostatic
accelerators or cyclotrons are suitable sources (Scharf, 1989). Electrostatic accelerators are
most commonly used. They may be of a regular or modified Van de Graaff type, in which
a high-voltage terminal is charged by a belt or a metal chain. A second type of electrostatic
accelerator uses a high-frequency voltage multiplication stage to charge the terminal, the
ions being accelerated in the electrostatic field created by the terminal. Cyclotrons have
been employed at some laboratories. These are particularly suitable for PIXE at higher ion
energies (Durocher et al., 1988; McKee et al., 1989; Peisach and Pineda, 1990) and in
certain combinations of PIXE with complementary techniques, such as particle elastic
scattering analysis (PESA) or forward alpha scattering (FAST) (Cahill et al., 1984).
In most analytical situations in which PIXE is applied, the choice of accelerator is
governed by access to a particular machine rather than by free choice. When a dedicated
accelerator can be selected, however, usually a single-ended or tandem electrostatic ma-
chine of 2–3 MV is chosen. Such accelerators may also favorably be employed for a
combined use of ion implantation and ion-beam analysis, including PIXE.
Certain types of analysis, such as analyses by microbeam techniques, impose special
requirements on ion source brightness, energy stability, beam emittance, and minimum
scattering of the ion beam. Such requirements may play an important role in selecting an
accelerator.
It is often stated that the need for an accelerator in PIXE is a major drawback, for
reasons or both technical complexity and economy. Although such objections contain
some truth, it should be pointed out that the prices of small electrostatic accelerators are of
the same order as those for complex ‘‘traditional’’ analytical equipment, such as mass
spectrometers or electron microscopes, and that a comparable technical skill is required
of the operator. It is therefore unjustified to rule out the PIXE method for such reasons.
Copyright © 2002 Marcel Dekker, Inc.
B. Beam Handling and Target Chambers
1. Macro-PIXE inVacuum
Commercially available vacuum chambers used as scattering chambers in nuclear physics
experiments are generally not well suited for PIXE. Their shapes and dimensions hamper
obtaining large solid angles for detection and introducing multispecimen holders and
sample changers. As a consequence, few existing PIXE chambers are of commercial origin.
There are exceptions to this rule, however, and certain chambers have been built entirely
from commercially available components.
Figure 3 shows the design and principal components of a hypothetical ‘‘typical’’
setup for macro-PIXE analysis. In the following, the details of such a system are outlined.
The ion beam emerging from the accelerator first passes through an analyzing
magnet, which sorts out the ions of the correct mass and velocity, and is then normally
focused by electrostatic or magnetic quadrupole lenses onto the specimen.
A typical experimental facility for macro-PIXE analysis employs a system for pro-
ducing a homogeneous beam so that the specimen is evenly irradiated. This is required for
quantitative analysis of heterogeneous samples. There are various methods of achieving
this:
1. Focusing the beam onto a scattering foil, which distributes the ions homo-
geneously over the cross section selected by a diaphragm. The foil material
should be able to withstand the ion beam for long times. It is often made of thin
metal, such as gold, which is an excellent scatterer with low stopping power for
the ions.
2. Rastering the beam over the sample by electrostatic deflection, for example.
This does not give a genuinely homogeneous distribution and may, therefore,
produce high instantaneous count rates in heterogeneous samples. On the other
hand, this method has the advantage of relatively low loss of beam intensity,
whereas such a loss may be more than 90% in the case of method 1.
3. After the initial alignment of the beam, straightforward symmetric defocusing by
the quadrupole magnets.
After homogenization, the beam is defined by a pair of collimators (diaphragms), which
are normally circular with diameters between 1 and 10 mm and placed approximately
Figure 3 Typical experimental arrangement for routine PIXE, with detectors for x-rays and
g-rays. Normally, the collimators can be changed between 1 and 10 mm diameter, and the x-ray
absorber holder would contain 5–10 different filters.
these may penetrate into the Si(Li) detector and cause problems. Hence, it is advisable to
use an absorber that is sufficiently thick to stop all scattered particles, but this hampers or
precludes the detection of those light elements. According to Musket (1986) and Nejedly
et al. (1995), placing a magnetic deflection trap between the sample and the detector crystal
is a viable alternative to an absorber for removing the scattered ions. However, the
installation of such system also results in a lower solid angle of detection.
The Si(Li) detectors used for PIXE are the same as those in EDXRF (see Chapter 3).
They typically have a sensitive area from 10 to 80 mm2, but this area is often reduced by
inserting an x-ray collimator in front of the detector. The aim of such collimator is to
minimize the low-energy peak tailing that results from incomplete charge collection at the
edge of the detector crystal.
In TPIXE, the special geometry with a very small angle of incidence allows one to
position the x-ray detector very close to the specimen (van Kan and Vis, 1997), similarly as
in TXRF. Consequently, a very large solid angle can be obtained, thereby increasing the
sensitivity. This increased sensitivity in combination with the reduced bremsstrahlung
background is at the origin of the low limits of detection in TPIXE (van Kan and Vis, 1997).
The amplifiers=pulse processors in energy-dispersive x-ray spectroscopy require large
time constants for optimum energy resolution (see Chapter 3). Unfortunately, large time
constants also imply that pulse pileup already becomes a serious problem at relatively low
count rates. It is therefore common practice to incorporate an electronic pileup rejector
Copyright © 2002 Marcel Dekker, Inc.
and its associated dead-time correction circuitry in the pulse processing chain. Such so-
lution reduces the throughput of signals to the analyzer (i.e., it decreases the ratio of
output signal count rate to beam current) and this forms a drawback when analyzing
samples that deteriorate during bombardment. To minimize beam-induced damage during
PIXE analysis, an on-demand beam excitation system is recommended. Such a system
employs a set of deflection plates placed in the beam path upstream of the sample; when an
x-ray event is detected, the beam is deflected immediately and then held off until the event
has been processed (Malmqvist et al., 1982). It should be indicated here that pileup peaks
(or sum peaks) are not entirely eliminated by pileup rejectors or on-demand beam excita-
tion, because such systems have a finite pulse pair resolving time or beam-switching time.
In PIXE of thin specimens, the ions pass through and are dumped in a Faraday cup
(see Fig. 3) for charge integration. To minimize the g-ray background originating from
nuclear reactions in the Faraday cup, long cups in which the particles are collected far
from the detector are preferable. Furthermore, the escape of secondary electrons from the
cup as well as the entrance of secondary electrons from the specimen should be prevented
by placing a negatively biased ring in from of the cup. The charge integration itself is
accomplished by connecting the Faraday cup to a sensitive current integrator=digitizer.
When the samples are thick enough to stop the ions, the beam current must be measured
either on the whole irradiation chamber or through some indirect approach. One possible
method is to place a thin foil in the beam path upstream of the target and to measure the
intensity of the particles scattered from it with a surface barrier detector (Mitchell et al.,
1980). Another, related method uses a beam chopper (e.g., a thin metal strip, which
periodically passes through the beam) (Volvo et al., 1983).
The bombardment of an insulating thick (or semithick) specimen in vacuum gen-
erally gives rise to charge buildup, and the specimen may reach a positive potential of up
to several tens of kilovolts before breakdown and sparking. The high potential accelerates
electrons up to tens of kiloelectron volts, and as a consequence, a huge bremsstrahlung
background is produced in the PIXE spectrum. The peak-to-background ratios then de-
crease significantly (Ahlberg et al., 1975a). To avoid this, various methods can be used:
2. Nonvacuum Macro-PIXE
As has already been indicated, it is sometimes advantageous to use atmospheric pressure or
moderate vacuum instead of high vacuum during analysis (Williams, 1984). The advantages
Copyright © 2002 Marcel Dekker, Inc.
are that the heat conductivity is increased and the target temperature decreased, the charge
is conducted from the target, and the vacuum requirements are less, thus making it easier to
design a low-cost chamber. Because the accelerator requires a good high vacuum, the ion
beam must be extracted into the moderate vacuum or atmospheric pressure region through
a thin exit foil. The beam eventually deteriorates this foil, and, therefore, its material has to
be carefully selected. The best choice is the polyimide foil Kapton, which withstands high
intensities and a high radiation dose before mechanical breakdown. Supporting the foil by a
carbon grid and direct flow of liquid-nitrogen-cooled helium gas allows the use of high
beam intensities over extended time intervals (Hyvonen-Dabek et al., 1982). The chamber
gas is normally helium or nitrogen. In addition to its better cooling properties, helium
produces less bremsstrahlung than nitrogen. Some x-ray detector windows are not leak
proof to helium, however, and this has disastrous consequences for the cryostat vacuum
unless the detector is separated from the chamber gas by an additional window.
The choice of chamber atmosphere is determined by the objectives of the analysis
and by the samples to be analyzed. A rather low pressure of helium gas, for example,
suffices to improve the heat conductivity and to reduce thermal losses of elements or
compounds (Martinsson, 1987). On the other hand, an external beam in air is needed when
large objects are to be analyzed without sectioning of subsampling. This is very important
in archaeology and art science, for example. The disadvantages of bombardment in air are
that there is a danger of sample oxidation during irradiation and interfering argon x-ray
lines are present in the spectrum. The strong argon lines may also be used to monitor the
beam fluence, however, which is otherwise difficult to do in nonvacuum PIXE. In fact, one
generally relies on some indirect method of beam current measuring (Mitchell et al., 1980;
Volkov et al., 1983); although with special precautions, it is possible to measure the beam
current directly (Wookey and Rouse, 1987).
The advantages of using higher pressure during irradiation have been demonstrated
in various applications. For example, in studies on volatile organic compounds, using a
combination of PIXE and complementary techniques (PESA), it was shown that sig-
nificant losses of certain constituents occurred when bombarding in high vacuum, whereas
the losses were insignificant for bombardments of similar beam intensity in helium at a
pressure of 100 torr (Martinsson, 1987).
When using external beams, some safety precautions are necessary. Besides the
obvious hazard of direct exposure, there is the potential exposure to the scattered beam
and the activation of the air to form positron-emitting radionuclides. These hazards are
discussed in detail by Doyle et al. (1991) in a review of external beam work with various
analytical techniques.
3. Nuclear Microprobes
In nuclear microprobe or micro-PIXE analysis, the particle beam is collimated and=or
focused down to dimensions in the range of 1–50 mm. With the regular equipment used in
accelerator-based ion-beam analysis, beam sizes down to typically a few tenths of a milli-
meter diameter are easily obtained, but to produce a genuine microbeam, specially de-
signed equipment is required.
The simplest way of producing a microbeam is to employ a pinhole collimator
(Horowitz and Grodzins, 1975). For very small collimator sizes, however, the beam in-
tensity obtainable is much too low for practical use. In addition, a substantial fraction of
the ions are scattered at the edge of the collimator, and this gives rise to a halo (with an
intensity of several percent of the beam current) around the central beam. In most micro-
Copyright © 2002 Marcel Dekker, Inc.
PIXE systems developed since the early seventies, collimation is therefore combined with
an electrostatic or magnetic demagnification system, such as quadrupole doublets, triplets,
quadruplets, or superconducting solenoids (Legge, 1997). The initial collimation in an
object aperture normally provides a beam with dimensions of 5–100 mm, and this beam is
then demagnified by a factor of 5–100. New-generation microprobes with enhanced de-
magnification power (200–3006) are under development in a few laboratories (Breese
et al., 1999). The best systems currently available are able to produce a spatial resolution
of about 0.5 mm at the specimen while maintaining an ion current that is useful for PIXE
analysis ( > 100 pA). Figure 5 shows the layout of a nuclear microprobe. Although the
exciting particles form the only difference between micro-PIXE and energy-dispersive
EPMA, much better peak-to-background ratios and, consequently, lower detection
limits are obtainable by micro-PIXE. This is illustrated in Figure 6, where x-ray spectra
for a biological specimen obtained by both techniques are compared. The much better
peak-to-background ratio, particularly for the heavier elements (Z > 20), justifies the use
of the much more complex analytical equipment of the nuclear microprobe. In addition,
the high resolution of the nuclear microprobe makes it possible to combine PIXE with
several other analytical techniques for imaging and quantitative analysis.
As shown in Figure 5, the normal components of a nuclear microprobe comprise the
following:
1. A particle accelerator (normally electrostatic) with a very bright ion source
2. Precision collimators
3. Magnetic or electrostatic quadrupoles for focusing
4. A scanning system to raster the beam over the sample, as in a scanning electron
microscope
The detection system is, in principle, identical to that for macro-PIXE but usually includes
complementary surface barrier detectors for scattered particles and for particles emitted in
nuclear reactions to extend the analytical arsenal (Malmqvist, 1995a). Sometimes, a de-
tection system for secondary electrons (Kneis et al., 1982) and=or a detection system for
near-visible light, ionoluminescence (Yang et al., 1993) are also included, for imaging
of the specimen (see Sec. VII.C). Facilities for accurate positioning of the beam on the
specimen area of interest are also required. Hence, an optical viewing system with high
magnification and good resolution is needed, as well as a precision sample holder con-
trolled by stepping motors or piezoelectrical crystals with a position accuracy below 1 mm.
The data are acquired in dedicated computer systems that can produce both quantitative
results and qualitative elemental maps.
Various types of object collimators can be used. The collimator design should
minimize the contribution of scattered ions to the specimen. Such devices with precision-
polished cylinder surfaces (Fischer, 1988) to define the ion beam and local or remote
control of the collimator width are commercially available and are employed in most
nuclear microprobe setups. Other laboratories prefer to use fixed apertures of the same
design as for electron probes (Legge et al., 1982). The precision parts of the collimators are
normally protected against beam damage and excessive heating by adding a slightly larger
precollimator in which most of the energy is dissipated.
Each configuration of demagnification devices has its advantages and drawbacks.
The rather complex Russian quadruplet, which was used in the first micro-PIXE system at
Harwell (Cookson et al., 1972), produces a symmetric image but is difficult to align
mechanically and has an increased risk of parasitic aberrations. Because of its simpli-
city, some laboratories use the doublet configuration, although it requires rectangular
Copyright © 2002 Marcel Dekker, Inc.
Figure 5 Base components (not to scale) of a typical nuclear microprobe system: (a) electrostatic
particle accelerator; (b) primary object aperture; (c) secondary collimator; (d) focusing system;
(e) scanning system; (f ) video camera and microscope; (g) surface barrier detector for scattered
particles; (h) x-ray detector; (i) specimen; ( j) surface barrier detector for transmitted particles
(STIM); (k) front-end CAMAC with data bus; (l) main computer and display with elemental map.
collimation for obtaining a symmetric image. The most common system, at present, is the
commercially available triplet system from Oxford MicrobeamsTMLtd. (Oxford, UK).
When more than two quadrupoles are used, two magnets are connected in series to the
same current supply. Whatever configuration is used, the current supplies should be very
stable ( < 10 7 4), to reduce distortions due to chromatic aberrations.
The image size at the specimen is easily determined from first-order calculation. For
small apertures, however, the first-order calculation does not suffice and second- and third-
Copyright © 2002 Marcel Dekker, Inc.
Figure 6 X-ray spectra from a thin biological specimen (human brain) obtained with electron (a)
and nuclear microprobe (b) excitation. Note the very large difference in peak-to-background ratio,
particularly from about 3 keV (channel 100) up. (From Johansson and Campbell, 1988. Copyright
John Wiley & Sons, Ltd. Reproduced with permission.)
order calculations are required. The lens aberrations can be calculated if the details of the
configuration are known. For more information on this subject, the reader is referred to
the book by Grime and Watt (1984), which provides a comprehensive compilation of
various magnetic quadrupole systems. It should be noted here, that among the intrinsic
aberrations, the third-order, spherical (angular) aberrations are the dominating ones in
most nuclear microprobes. Hence, some systems make use of octupole magnets to reduce
the spherical aberrations (Jamieson and Legge, 1988). Other important factors are the
mechanical precision (to reduce parasitic aberrations) and, particularly, the rotational
defects in the magnetic fields (these can severely distort the image).
To allow full use of its powerful analytical capabilities, a nuclear microprobe setup
should include a scanning system for rastering the ion beam over the specimen surface.
Copyright © 2002 Marcel Dekker, Inc.
The scanning system could be installed before focusing ( predeflection), but such a system
deflects the beam out of the optical axis and could therefore, for large deflection angles,
significantly increase the aberrations. By proper selection of the position of deflection,
however, the beam can be made to pass through the optical center of the lens system so
that the aberrations are minimized. The scanning system can also be installed after the
demagnification lenses. Because of the short distance between lens and specimen, however,
scanning systems using electrostatic deflection plates require a high electric field (with
concomitant risk of electric discharges) to obtain a sufficiently large deflection amplitude.
An alternative is to employ magnetic scanning. By using scanning coils with ferrite cores,
a reasonably large amplitude and a scanning frequency of more that 5 kHz can be ob-
tained (Tapper et al., 1988). Such systems provide a good compromise between scanning
speed and amplitude. The scanning system is usually computer controlled and connected
to the data acquisition system (see Sec. III. C.2).
The detailed design of the irradiation chamber for the nuclear microprobe is beyond
the scope of this chapter. We will therefore limit ourselves to giving some recommenda-
tions for the essential components. The sample-positioning system may be a commercially
available x-y-z precision translator as designed for scanning electron microscopes. The
sample holder should preferably take many samples to avoid the need for frequent
opening of the vacuum chamber. The microscope used for viewing the specimen should
have a magnification of 200–4006 and preferably be equipped with a zoom lens. In ad-
dition to the x-ray detector, surface barrier detectors should be entered in the forward and
backward directions. Such detectors are needed for extending the elemental coverage and
for determination of the specimen thickness. The specimen thickness is of paramount
importance for PIXE quantification in thin and semithick samples (see Secs. IV.B. and
IV.C). The specimen thickness may also be provided by measuring the energy loss in each
specimen pixel, as is done in scanning transmission microscopy (STIM) (Bench et al.,
1992). Finally, it is recommended to install a secondary electron detector and to allow for
detection of visible light to facilitate imaging of the specimen surface.
A good high vacuum is crucial to reduce residual gas scattering that would de-
grade the beam quality. A high vacuum close to and in the specimen chamber is
maintained by direct pumping with oil diffusion, turbomolecular, or cryopumps. Al-
though they have many advantages, the two latter types may transmit vibrations and
magnetic disturbance to the demagnification lens system and to the specimen holder.
Hence, it is crucial to arrange those pumps to minimize such effects. The focusing of
small beams is adversely affected by these or other vibrations. It is therefore common to
place the whole microprobe system on a rigid optical bench on a fundament with good
damping. The accurate positioning of the optical elements on this bench is realized by
high-precision mechanical controls. Furthermore, to avoid the effects of the Earth’s
magnetic field and stray fields from surrounding equipment, beam tubes are sometimes
shielded by m-metal foils.
As is the case for macro-PIXE, micro-PIXE may also be carried out under non-
vacuum conditions (see Sec. III.B.2). Because of scattering of the beam by the gas, however,
nonvacuum micro-PIXE is only feasible for moderately small beam sizes (20–100 mm).
After collimation and=or focusing, the beam is passed to the nonvacuum region through a
pinhole or an exit foil. In the first approach, the high vacuum in the beam line is maintained
by means of differential pumping. If the spatial resolution requirements are not too high,
the nonvacuum micro-PIXE technique is rather straightforward and simple to use. It is very
useful when examining large samples or sensitive art objects, such as bronze figures and
ancient documents, and may, hence provide unique information (Swann, 1983).
Copyright © 2002 Marcel Dekker, Inc.
C. Data Acquisition
1. Macro-PIXE
The data acquisition systems used in PIXE show great similarity with the EDXRF data
acquisition systems discussed in Chapter 3. For x-ray detection, a high-resolution solid-
state detector is virtually always employed. This could be either a Si(Li) or high-purity Ge
(HPGe) detector, but the former is highly preferable in most PIXE work. Ge detectors give
rise to much more intense escape peaks than Si(Li), and this can render the spectra un-
acceptably complex. The higher detection efficiency of Ge for energies above 20 keV is not
very useful in 1–4-MeV PIXE because of the very rapid decrease in K ionization cross
sections with increasing Z (see Sec. II.A.2). On the contrary, the higher detection efficiency
may be a disadvantage, as the background contribution resulting from Compton-scattered
g-rays is larger than with Si. Most PIXE chambers contain only one Si(Li) detector for
x-ray detection. However, to improve the sensitivity for the heavier elements while re-
taining the capability of measuring the light elements during the same bombardment, the
use of two Si(Li) detectors has been advocated (Wätjen et al., 1990), and such a system is
now employed in an increasing number of PIXE facilities (Johansson et al., 1995). The
second Si(Li) detector, used for measuring the heavier elements, is provided with a thick
absorber to cut down the high count rate from the light elements and has a larger solid
angle of detection than the first Si(Li). As an alternative to such a second Si(Li), a Ge x-ray
detector is well worth considering. In addition to the Si(Li) detector(s) for x-rays, PIXE
chambers generally also contain detectors for the complementary IBA techniques (i.e.,
surface barrier detectors for scattered particles or for particles resulting from nuclear re-
actions and a Ge detector for measuring prompt g-rays). Hence, the acquisition system
must include several analog-to-digital converters. Furthermore, because the spectral in-
tensity in PIXE is proportional to the number of incident particles, the measurements are
generally carried out for a preset charge (or some parameter related to it in the case of
indirect beam current measurement) instead of for a preset live time. The charge (or re-
lated parameter) is usually measured by an external counter, and this unit forces the data
collection to stop when its preset is reached. For acquiring the spectra, either a personal-
computer (PC)-based multichannel analyzer (MCA) or a classical MCA may be used, but
the latter should be interfaced to a computer so that the spectra can be saved on disk and
evaluated by appropriate computer programs.
2. Micro-PIXE
The data acquisition in scanning nuclear microprobe analysis is more complex than in
macro-PIXE and is invariably controlled by computers. As in more advanced systems for
macro-PIXE, signals from several detectors must be handled, but, in addition, the posi-
tional information must be recorded. In Figure 7, a typical modern design of a data
acquisition system is outlined (Elfman et al., 1997). Two main principles are employed:
on-line display of elemental maps and event-by-event acquisition with off-line sorting of
data. Sometimes, a combination of both is used. Elemental maps on-line are obtained by
setting energy windows for the characteristic x-ray lines of interest (and for background
regions) and reading out the count rate within each window for each position of the beam.
The line intensity values (with or without background subtraction) for each beam position
(pixel) are stored in a computer, and maps are generated on-line with intensity modulation
by gray or color codes. This technique gives good feedback, so, that, during analysis, one
can concentrate on the more interesting regions of the specimen. This is quite important
Copyright © 2002 Marcel Dekker, Inc.
Figure 7 Schematic outline of a comprehensive data acquisition system for a nuclear microprobe including several detectors. A CAMAC system is
used as front end and several personal computers interact with the system control and the data acquisition. (Courtesy of M. Elfman.)
requirements on the spectrum model. As far as the modeling of the sum peaks is con-
cerned, this is generally done by representing them by a single pileup element, according to
an approach first proposed by Johansson (1982). Despite the many fine details, accurate
modeling of PIXE spectra is quite feasible, as was demonstrated in an intercomparison
exercise of five different PIXE spectrum analysis programs (Campbell et al., 1986). It is
also illustrated by the good agreement between modeled and experimental spectrum in
Figure 8.
where E0 ad Ef are the incident proton energy and the energy of the protons after passage
through the target (Ef ¼ 0 for an infinitely thick specimen), respectively, E is the proton
energy, Tp ðEÞ is the transmission of the x-rays from successive depths in the specimen, and
SðEÞ is the matrix stopping power. Tp ðEÞ is itself given by
0 1
ZE
Bmp sin y dE C
Tp ðEÞ ¼ exp@ A ð12Þ
sin f SðEÞ
E0
with mp the mass attenuation coefficient for line p in the sample matrix, and f the angle
between the specimen surface and the specimen–detector axis (i.e., the x-ray takeoff angle).
It should be noted here that the relation between the analyte element line intensity
Yp ðZÞ and the concentration CZ , as expressed by Eqs. (11) and (12), does not include
secondary or tertiary fluorescence enhancement effects. Detailed treatments of these effects
were given by Campbell et al. (1989, 1993). Although enhancement effects are less pro-
nounced in PIXE than in XRF and are, in fact, often negligible, secondary fluorescence
should be accounted for when the analyte elements are lighter than the matrix elements.
For example, in 3-MeV-thick target PIXE of stainless steel (with y and f both equal to
45 ), the CrKa intensity is raised by about 50% as a result of the secondary fluorescence
from the FeK x-ray lines (Campbell et al., 1989).
As for thin specimens, several quantitation approaches are possible for thick spe-
cimens. If one relies on the fundamental parameter approach and thus solves Eqs. (11) and
(12) for CZ , other parameters are needed in addition to those already required for the thin-
specimen case. These additional parameters are mp and SðE Þ, the mass attenuation co-
efficient for line p in the sample matrix and the matrix stopping power, respectively. The
values of those parameters for the sample matrix can be obtained from those for the
matrix constituents by employing Bragg’s additivity rule, as already indicated in Sec.
II.A.1 for the matrix stopping power. Such calculations require knowledge of the matrix
composition and databases for the mass-attenuation coefficients in the various elements
and for the elemental stopping powers. The stopping power database and its accuracy
were dealt with in Sec. II.A.1. The problem of selecting an accurate database for the mass-
attenuation coefficients is the same as in all other x-ray emission techniques and will not be
discussed here. The matrix elemental composition places the major burden on the calcu-
lations and usually contributes most to the uncertainty in the calculated mp and SðEÞ
Copyright © 2002 Marcel Dekker, Inc.
values for the sample matrix. For heavier-element matrices, in which all matrix elements
are detected in the PIXE spectrum, iterative procedures can be applied to obtain the
matrix composition, but for light-element matrices, one must resort to a priori information
(e.g., obtained by other techniques) or certain assumptions have to be made (e.g., that the
elements are present as oxides). In any case, this problem of the matrix composition
is common to all x-ray emission analysis techniques. When specimens of intermediate
thickness are analyzed, the transmitted proton energy Ef (or rather the energy loss
E0 Ef ) is also needed for evaluating the integral in Eq. (11). This implies knowledge (or
determination) of the specimen mass thickness, because the energy loss is related to the
latter through the matrix stopping power. Alternatively, the energy loss can be measured
experimentally. It is evident that any uncertainty in the specimen mass thickness (or in the
experimental energy loss) will also be transmitted to the value of CZ . The magnitude of
this uncertainty transmission increases with decreasing specimen thickness and, ultimately,
a given relative uncertainty in the specimen mass thickness produces an identical relative
uncertainty in the value of CZ , as is, in fact, also the situation for the infinitely thin
specimens just discussed. The uncertainty transmission from the matrix composition, from
the databases, and for intermediately thick specimens also from the specimen mass
thickness can be much reduced by the use of an internal standard element. Indeed, the
uncertainty transmitted in the integral of Eq. (11) is, to a large extent, in the same sense for
the analyte and the spike, so that a significant uncertainty reduction occurs when dividing
the two integrals.
An alternative to the pure fundamental parameter quantitation approach is to make
use of experimental thin-target sensitivities kp ðZÞ, as defined by Eq. (10), so that Eq. (11)
can be written as
ZEf
kp ðZÞNCZ pZ ðEÞTp ðEÞ
sX
Yp ðZÞ ¼ X dE ð13Þ
spZ ðE0 Þ= sin y SðEÞ
E0
By solving this equation for CZ , one basically uses a relative method (relative to thin-film
standards), but the correction for matrix effects is made by a fundamental parameter ap-
proach. As in the relative quantitation method for thin specimens, this mixed approach
requires no knowledge of the absolute detection efficiency or of the radiative transition
probabilities and fluorescence yields [the latter two parameters cancel out in the ratio of the
x-ray production cross sections in Eq. (13)], but ionization cross sections and Coster–
Kronig yields are still required. However, the division of the x-ray production cross sections
also has the effect that the impact of the Coster–Kronig yields is marginal and that for
ionization cross sections, essentially only their dependence on proton energy is needed,
which has a much smaller uncertainty than the absolute value of the ionization cross section.
In the analysis of infinitely thick specimens, one can also utilize experimental thick-
target calibration factors instead of relying on the fundamental parameter approach or on
experimental thin-target sensitivities. The thick-target calibration factors incorporate the
integral of Eq. (11) and are usually expressed in x-ray counts/mC and per mg=g. They are
commonly derived from PIXE measurements on samples with known trace element
composition (standards). In a strict sense, the thick-target factors are only valid for the
analysis of unknown samples with identical (matrix) composition to the standards, but in
practice, some variability in composition can be tolerated or corrected for. The necessary
correction factor is, in this case, the ratio of the integral of Eq. (11) for the standard to the
corresponding integral for the unknown.
Copyright © 2002 Marcel Dekker, Inc.
As discussed by Johansson and Campbell (1988) and Johansson et al. (1995), still
other quantitation approaches are possible, for example, making use of thick single-
element standards.
Before closing this section on quantitation for semithick and infinitely thick speci-
mens, it should be warned that Eqs. (11)–(13) are, in a strict sense, only valid for perfectly
flat homogeneous samples and that, for specimens made up of particulate material, the
particle size should be as small as possible (ideally below 1 mm when quantifying light
elements, such as Na through Si). The surface roughness effects in PIXE were dealt with by
Cookson and Campbell (1983). With regard to particle size effects in PIXE, these are
especially of concern when analyzing deposits of atmospheric particulate material (aero-
sols) on filter or cascade impactor samples. Model calculations for proton-induced x-ray
generation in a monolayer of spherical particles having realistic matrix compositions were
presented by Jex et al. (1990). In aluminosilicate particles of 2 mm diameter, the attenua-
tion is over 20% for the elements Na to P.
It thus appears that xL is proportional (or inversely proportional) to the square root of the
experimental parameters FWHM, Ep ; N, and the specimen mass thickness. Hence, to op-
timize xL , a detector with very good resolution has to be used, and, rather obviously, the
solid angle of detection should be made as large as possible, but improvement in this
Copyright © 2002 Marcel Dekker, Inc.
parameter is limited by the area of the detector and by the fact that the detector can only
get to about 2 cm from the specimen. Much more flexibility is provided by the number of
incident protons N (preset charge), which can be increased either by a longer measurement
time or by an increase in beam current.
Both theoretical calculations and experimental measurements have been performed
to obtain xL values for thin specimens [e.g., Folkmann (1976), Johansson and Johansson
(1976), Ishii and Morita (1988)]. In most of this work, one adopted (or employed) a light-
element matrix (typically carbon or an organic polymer) and a specimen mass thickness of
1 or 0.1 mg=cm2. Johansson and Johansson (1976) produced a very useful contour plot of
xL values as a function of incident proton energy and atomic number of the analyte ele-
ment for the case of 0.1-mg=cm2-thick carbon matrix. Their plot, reproduced here in
Figure 9, was based on experimental measurements of the continuum background, and it
was further assumed that elements with atomic number up to about 50 are determined
through their Ka x-ray line and the heavier elements through their La line. As can be seen
in Figure 9, there is a valley of optimum detection limits for both the K and L cases, with
the best K xL values (less than 0.5 mg=g) obtained at lower proton energy than the best L xL
values (0.5–1 mg=g). Furthermore, within either the K or L case, the bombarding energy for
optimum detection limits depends on the atomic number of the analyte elements of
interest, with higher bombarding energies favoring the heavier elements. Selection of the
energy should thus be made with the objective of the analysis in mind, but, in practice,
some compromise is necessary. Johansson and Johansson (1976) concluded from their
contour plot that the optimum proton energy is about 2 MeV for the analysis of trace
elements in biological and environmental samples. Such bombarding energy also has the
advantage that the xL values show rather little variation (about one decade only) for the
analyte elements with Z between 15 and 90. More recently, Ishii and Morita (1988)
produced a contour plot similar to that of Johansson and Johansson (1976), but they
based it solely on theoretical calculations and adopted a pure oxygen matrix (which was
considered representative for biological samples). The conclusion of this study was that the
Figure 9 Contour plot of the limit of detection (xL) as a function of incident proton energy and
atomic number of the analyte element for the case of a 0.1-mg=cm2-thick carbon matrix.
Experimental conditions: detector FWHM 165 eV, solid angle of detection 38 msr, collected charge
10 mC. The background interval selected for calculating xL was equal to one FWHM. (From
Johansson and Johansson, 1976, with permission from Elsevier Science.)
VI. APPLICATIONS
The applicability of PIXE to various analytical problems has been amply demonstrated in
many publications. Furthermore, numerous studies have been carried out in which PIXE
provided part or all of the requested trace element concentration data. This section pre-
sents a brief selection of the applications of PIXE and is mainly based on publications
from the period 1992 through mid-1998. Many more examples can be found in the pro-
ceedings of the international conferences on PIXE (Johansson, 1977, 1981; Martin, 1984;
Van Rinsvelt, 1987; Vis, 1990; Uda, 1993; Moschini and Valković, 1996; Malmqvist, 1999)
and nuclear microprobe technology (Grime and Watt, 1988; Legge and Jamieson, 1991;
Lindh, 1993; Yang et al., 1995; Doyle et al., 1997), in the proceedings of the international
symposia on Bio-PIXE (Ishii et al., 1992; Zheng et al., 1996), in two textbooks on the
PIXE technique (Johansson and Campbell, 1988; Johansson et al., 1995), and in the
chapter on PIXE in the first edition of this Handbook of X-Ray Spectrometry (Maenhaut
and Malmqvist, 1992).
2. Examples of Macro-PIXE
The macro-PIXE application examples presented here were chosen somewhat arbitrarily.
However, their selection was guided by the aim of demonstrating the particular potential
of the PIXE technique. Furthermore, most examples were taken from the medical field.
a. Body Fluids
Blood, blood plasma, and serum have always been popular study objects for PIXE, despite
the fact that only a few real trace elements (Fe, Cu, Zn, Se, Br, Rb) can be measured in
these sample types. Compared to the early days of PIXE, there is now increasing com-
petition from other highly sensitive analytical techniques, such as inductively coupled
plasma–mass spectrometry (ICP–MS), which are more suitable for analyzing liquid
samples. Therefore, it becomes hard to justify the use of PIXE. The motivation often is
that one has a good-running and cost-effective bio-PIXE research program set up at a
small accelerator or baby cyclotron and that the blood, plasma, or serum samples form
part of a wider variety of sample types which are studied in the research project. When
working with blood or blood components from small experimental animals (e.g., mice or
rats), the small sample size provides additional justification for the use of PIXE.
Shenberg et al. (1995) used a combination of PIXE and instrumental neutron acti-
vation analysis (INAA) for measuring K, Fe, Zn, Se, Br, and Rb in blood and blood
fractions (plasma and red cells) of colorectal patients. It was concluded that the two
techniques complement each other. The mean values for K, Fe, Se, Br, Rb, and the Br=Rb
ratio were significantly lower for cancer cases than for healthy individuals, and it was
suggested that this may be applicable as additional information for differentiating ma-
lignant cases from normals. It was also found that the Br=Rb ratio was much lower in
Belgium than in Israel, which was explained by the influence of differences in dietary
habits and environmental factors on the Br level in the blood of both populations. Miura
et al. (1996) used PIXE for measuring 7 elements in serum of patients with acute myo-
cardial infarction (AMI), and they examined the correlations between the elements and the
vascular cell adhesion molecule-1 (VCAM-1), which is closely related to various types of
Copyright © 2002 Marcel Dekker, Inc.
inflammation. Soluble VCAM-1 was markedly enhanced in the sera of the patients with
respect to controls, and two of the elements measured (i.e., Ca and Zn) were significantly
correlated with it. The authors suggested that there was an interaction between the di-
valent cations Ca2 þ and Zn2 þ and VCAM-1, but concluded that further investigations
are needed to elucidate the actual mechanism of the alterations.
Blood plasma samples from experiments with animal models were analyzed by PIXE
by Sato et al. (1995) and Tamanoi et al. (1995). The study of Sato et al. (1995) dealt with
the effects of diethyleneaminepentaacetic acid (DTPA). This chelating agent is re-
commended for eliminating several kinds of radionuclides from persons contaminated
with them. It is generally administered as calcium-trisodium salt (Ca-DTPA) or zinc-tri-
sodium salt (Zn-DTPA), but both compounds have been reported to induce injuries, be it
that Zn-DTPA is less toxic than Ca-DTPA. Male Wistar rats were intraperitoneally (i.p.)
injected with the salts, and blood was collected from the tail vein just before and at 3, 6,
and 24 h after the injection. Na, K, Ca, Fe, Cu, and Zn were measured in the plasma. Ca-
DTPA significantly lowered the concentration of Zn, whereas no significant changes were
observed for the other metals. Zn-DTPA did not lower any metal concentrations, but
a significant increase of plasma Zn was observed. Tamanoi et al. (1995) used PIXE to
measure Cl, K, and Ca in blood plasma of mice in which EL-4 lymphatic tumor cells had
been transplanted. The motivation for employing PIXE was that blood samples were
taken at several intervals after the transplantation and that it was desirable to collect
samples as small as possible in order to induce no stress. A sample size of 5 mL plasma was
used for the PIXE analysis.
Certain body fluids, such as cerebrospinal fluid (CSF), exhibit very low concentra-
tions of trace elements, even for Fe, Cu, and Zn. PIXE of CSF requires special sample
preparation and analysis procedures. Kupila-Rantala et al. (1996) used ultrathin (20–
30 mg=cm2) Formvar films as sample support and a 625-mm-thick Kapton foil with a
pinhole as the x-ray absorber. The detection limits for Fe, Cu, Zn, and Br were 6, 4, 8, and
18 mg=L, respectively, for a CSF sample size of 50 mL.
3. Examples of Micro-PIXE
The fact that one can focus the ion beam to small groups of cells or even single cells as well
as to different cell strata in various tissue, in combination with the high analytical sensi-
tivity and accuracy of the PIXE technique, makes the nuclear microprobe a strong ana-
lytical tool in biomedicine. Its high lateral resolution represents an obvious advantage for
studies on the role of elements in cellular functions. By selecting some good examples, we
will try to illustrate the great importance of micro-PIXE in biomedicine.
a. Single Cells
The elemental composition of single cells can be studied on cells that are either present in
their natural surrounding tissue or are grown free in cell cultures. In most studies of modern
biology, human or animal isolated cells are used as experimental models. The advantage of
analyzing ‘‘normally’’ grown cells is that one can assume that they are not perturbed ele-
mentally. Cultured cells, due the difficulties of exactly simulating normal growing condi-
tions, may display some elemental deviations. However, the study of individual cells in a
living organism is impracticable because of the difficulty in identifying the revelant cells, in
manipulating their behavior in a controlled manner, and in separating effects due to in-
trinsic properties of the cells from effects due to the interaction among the many cell types
present in the tissue. Hence, the use of cell cultures has increased during the last decades, in
particular for studies with x-ray microanalysis, including micro-PIXE by using the nuclear
microprobe. The purpose of working with isolated cells as a experimental models is to
check, in vitro, consequently in reproducible conditions, a cellular function or the effect of
any molecule with a biological activity (Moretto and Llabador, 1997).
One limitation of PIXE, other x-ray techniques, or most other elemental analysis
techniques is that no information on the ionic form or speciation of the element is obtained.
For instance, analysis of calcium ions in cells is important for the understanding of cell
physiology. Also, although micro-PIXE is more sensitive than traditional x-ray micro-
analysis as used in an electron microprobe, both techniques provide information only on
total Ca and not on the Ca2 þ concentration, which is the relevant physiological parameter.
In the studies by Pålsgård et al. (1994) and Pålsgård and Grime (1996) on the role of Ca2 þ
as the second messenger, this problem has been overcome by using a Ca analog. In se-
cretory cells, such as the insulin-synthesizing b-cells, a rise in the Ca2 þ level triggers the
exocytosis of secretory vesicles and the release of insulin. To discriminate between
Copyright © 2002 Marcel Dekker, Inc.
endogenous Ca2 þ and Ca2 þ entering as a result of stimulation, Sr was used as a tracer.
Due to its chemical affinity, Sr2 þ is known to enter the Ca2 þ channels and to mimic the
role of Ca2 þ in the secretory process (Wroblevski et al., 1989). This elegant approach
makes use of the multielemental character and high sensitivity of PIXE. To assure that this
analog substitution works as assumed, Pålsgård et al. (1995) performed another study in
which they examined whether a different analog (Ba2 þ ) behaves the same.
When one is interested in measuring trace metals rather than electrolytes in cells,
micro-PIXE has particular relevance. Examples are research on heavy-metal intoxication,
but also studies on elemental changes induced by the physiological or pathological state of
the cells. In a study of Menkes disease, the Melbourne nuclear microprobe group in-
vestigated the intracellular Cu concentration in cultured Menkes fibroblasts by means of
individual cell microanalysis (Allan et al., 1994). There was a highly significant difference
of a factor of 6 between the Cu levels of Menkes and normal fibroblasts. By using the
individual cell analysis, it was also found that there was large biological variability be-
tween the individual cells, despite the fact that they were grown under uniform conditions.
Some transition metals delay important regulatory roles in gene expression. The
disturbance of their cellular levels could be involved in oncogene expression and tumo-
rigenesis. In a study by the Bordeaux nuclear microprobe group (Ortega et al., 1997),
micro-PIXE was used to measure Mn, Fe, Cu, and Zn in cultured human neuroblastoma
cells. The study indicated that the nuclear microprobe is a useful tool for investigating
transition metal concentrations in cultured cells and that further research is required to
investigate the detailed role of these trace elements.
Another interesting example of single-cell analysis is that of individual marine
plankton. The occurrence of annual marine phytoplankton blooms is becoming a global
problem. Within a European Union-funded project, it is investigated if the unbalanced
nutrient composition of the water promotes the dominance of harmful phytoplankton
species. A combination of new advanced methods is used to allow simultaneous de-
termination of toxin content, elemental composition, genetic expression, and bio-optical
properties of toxic phytoplankton at the species level in natural communities. One of the
tasks is the determination of the elemental composition of single phytoplankton cells. This
is carried out with the Lund nuclear microprobe, using both micro-PIXE and nuclear
reaction analysis (NRA) with deuterons, with a special focus on C, N, P and K. The
method used to isolate single living cells while reducing their salt environment is an im-
portant part of the analytical procedure. Figure 10 shows elemental maps obtained by
nuclear microprobe analysis of a phytoplankton. Also, the results from STIM measure-
ments and an optical micrograph are shown (J. Pallon, personal communication, 1999).
b. Soft Tissues
With the nuclear microprobe, one can raster the focused ion beam over selected areas on
various biological structures and form elemental or structural maps which are displayed
on-line. This facilitates the combination of overall analysis and detailed spot investiga-
tions. By overlaying a video image or an optical micrograph, it is possible to localize and
identify particularly interesting regions or cell strata. Because of the stringent requirements
on the sample preparation in order to preserve biological trace element information, the
tissue specimens are normally not stained but only freeze-dried. Hence, the lack of contrast
in optical mode of such specimens makes it necessary to combine all means of imaging and
analysis available (Malmqvist, 1995a).
The central nervous system has been extensively studied using the nuclear micro-
probe. Starting in Oxford and continuing at the National University of Singapore (NUS),
Copyright © 2002 Marcel Dekker, Inc.
Figure 10 Elemental maps and an off-axis STIM map from a single phytoplankton, Dinophysis
sp., obtained by nuclear microprobe analysis with 2.5-MeV protons. P to Ca are determined by
micro-PIXE, and C and N with backscattering. The scan size is 64 mm. (Courtesy of J. Pallon.)
Watt and co-workers have carried out several studies on brain tissue taken postmortem
from diseased patients diagnosed as suffering from Alzheimer disease. The presence of
elevated levels of certain elements, especially Al, in neurofibrillary tangles and neuritic
plaques from such patients has been the subject of great controversy (Jacobs et al., 1989;
Landsberger et al., 1992). In a recent study on neurofibrillary tangles, Makjanić et al.
(1997) used freeze-dried samples for analysis by both micro-PIXE and STIM. The results
indicated that the elevated Al concentration measured is an artifact introduced during
sample preparation. Using animal models, the Lund group performed several studies on
rat brains in order to understand the mechanisms behind cell death in special cell strata of
the brain as a consequence of stroke or epileptic seizures. The micro-PIXE results for
groups of cells in the substantia nigra indicated that the onset of the cell death was
characterized by a change in Ca concentration, and this before any histopathological ef-
fects could be observed (Themner et al., 1988; Inamura et al., 1989; Tapper et al., 1991).
Atherosclerosis is a major cause of death in the Western world. The understanding of
mechanisms behind the calcified depositions on the artery wall is of paramount im-
portance. Several studies using micro-PIXE and the nuclear microprobe have been carried
out on this subject (Rokita et al., 1991; Pinheiro et al., 1996). The detailed information
obtained, including data on several trace elements, provides clues on the atherogenesis.
Being a very well-stratified tissue, skin lends itself very easily to microprobe analysis.
Many studies have been carried out with the Lund nuclear microprobe by Forslind and co-
workers (Pallon et al., 1992, 1996, 1997). Both normal skin and pathological states (e.g.,
Copyright © 2002 Marcel Dekker, Inc.
atopic and psoratic skin) have been extensively investigated. The trace elemental profiles
across the various skin strata are also useful in the basic understanding of the skin barrier
function. This is a very essential mechanism of epidermal skin to protect the body while
still being permeable. Perturbations of the barrier have recently been investigated using a
nuclear microprobe by the Lawrence Livemore Laboratory group (Mauro et al., 1998).
Figure 11 shows the Ca distribution before and after acetone-induced barrier disruption.
Another type of tissue that has been studied frequently using the nuclear microprobe
is kidney cortex. In an attempt to understand the etiology of Menkes disease, Kirby et al.
(1998) used an animal model (mice) and studied the distributions of Cu, Fe, and Zn in
kidney cortex. The tissue specimens were prepared by cryosectioning and freeze-drying.
Localization was facilitated by enzyme histochemical staining of adjacent sections. The
micro-PIXE technique clearly showed how the three elements in question were distributed.
c. Hard Tissue
Due to the convenience in sampling (hair, nails), the elemental long-term stability (mi-
neralized tissues), ‘‘time-recorder capability’’ (hair, nails, shells), and stability under the
ion beam, hard tissues have been rather popular specimens for nuclear microprobe stu-
dies. Besides fish scales, various aragonite structures in fish and squid have been analyzed,
in particular for Ca and Sr (Tang et al., 1997). The elemental variations in the slowly
growing statoliths or otoliths (ear stones) reflect variations in living conditions. In most
countries with extended coastlines, fishing is of great economical importance, and hence
the migration patterns of fish are obviously of great economical as well as ecological
interest. Lipinski et al. (1997) studied squid statoliths using micro-PIXE. The results were
Figure 11 Calcium distribution in skin as measured by micro-PIXE with and without acetone
provocation. A 3-mm microbeam of 3-MeV protons was used for analysis. (From Mauro et al., 1988,
by permission of Blackwell Science, Inc.)
Figure 12 Elemental maps (Zn and Sr) from an otolith from brown trout. Analysis was done by
2.5-MeV protons and the scan size was 4 mm. Dark areas indicate high concentration and light areas
indicate low concentration. (Courtesy of K. Limburg.)
B. Atmospheric Aerosols
The multielement analysis of airborne particulate material (atmospheric aerosols) has been
a very popular and highly successful application of PIXE since the early days of the
technique. Actually, the pioneering paper of Johansson et al. (1970) already presented a
PIXE spectrum of such material. A more recent spectrum, obtained from the bombard-
ment of an urban aerosol sample (Campbell et al., 1986), is shown in Figure 13.
Considering that atmospheric aerosols are often collected as a thin-sample layer on
some thin filter or substrate film, that such samples can be analyzed nondestructively by
PIXE without sample preparation, that the sample matrix consists of light elements, and
that a 5–10 min bombardment suffices to detect up to 20 elements, including interesting
anthropogenic elements such as S, V, Ni, Cu, Zn, As, and Pb, the analysis of aerosol samples
forms almost an ideal application of PIXE. Compared to EDXRF, which shares several
favorable characteristics, PIXE offers sensitivities (expressed as characteristic x-ray count
rate per microgram of element actually exposed to the analysis) that are typically at least one
order of magnitude better (Maenhaut, 1989), so that it requires much less sample mass and
thus allows the use of compact samplers with high time and size resolution. Another ad-
vantage of PIXE over EDXRF is that it can be complemented with other ion-beam analysis
techniques (Cohen, 1993, 1998; Cahill, 1995; Swietlicki et al., 1996a), so that the light ele-
ments (H, C, N, and O) that make up most of the aerosol mass can be measured as well.
Motivation for the study of atmospheric aerosols and of their physical and chemical
characteristics is that they reduce visibility, affect human health, contribute to acidification
of terrestrial and aquatic ecosystems, and cause damage to structures and buildings (Elsom,
1992; Boubel et al., 1994; Cahill, 1995; Seinfeld and Pandis, 1997). During the past decade,
renewed interest in the health risks of aerosols has been generated by the finding of strong
correlations between increased mortality and the concentrations of (fine) airborne particles
in metropolitan areas in the United States (Pope et al., 1995). Furthermore, it has become
clear that aerosols play a much more important role in climate than previously thought
Copyright © 2002 Marcel Dekker, Inc.
Figure 13 PIXE spectrum of an urban aerosol sample. (From Campbell et al. 1986, with
permission from Elsevier Science.)
(Charlson and Heintzenberg, 1995). At the same time, one has come to realize that there
remain serious gaps in our knowledge of atmospheric aerosols. With regard to the observed
correlations between aerosols and increased mortality, it is uncertain whether these are due
to the particles as a whole or to certain specific components, such as fine sulfate, the car-
bonaceous aerosol (organic carbon and elemental carbon), or heavy metals. There is
consensus, though, that the fine particles are more harmful to human health than the coarse
ones and this has led to the replacement in 1997 in the United States of the PM10 (parti-
culate matter smaller than 10 mm) standard by the PM2.5 (particulate matter < 2.5 mm)
standard for aerosol measurements in atmospheric monitoring networks. As to the climatic
effect of aerosols, this stems from the fact that they physically affect the heat balance of the
Earth, both directly by reflecting and absorbing solar radiation and by absorbing and
emitting some terrestrial infrared radiation and indirectly by influencing the properties and
processes of clouds, and, possibly, by changing the heterogeneous chemistry of reactive
greenhouse gases (e.g., O3) (Charlson and Heintzenberg, 1995; Schwartz, 1996). Changes in
the heat balance due to anthropogenic or externally imposed changes are referred to as
forcings. There is a very large uncertainty associated with the aerosol forcing estimate, and
this is essentially due to our poor state of knowledge regarding the sources, spatial and
temporal variability, and chemical, physical, and optical properties of atmospheric aero-
sols. As for the impact on human health, fine (submicrometer-sized) particles are more
important than the coarse ones for both the direct and indirect effect on climate. However,
there is still much unknown about the formation, evolution, and composition of the fine
particles. Quite often, all measured aerosol constituents in fine-aerosol characterization
projects do not add up to the measured fine-particulate mass (PM), and a major fraction of
it remains unexplained. The recent findings about the effects of aerosols together with the
realization of the serious gaps in our knowledge have given a strong new impetus to aerosol
Copyright © 2002 Marcel Dekker, Inc.
research. The PIXE technique can continue to provide an invaluable contribution to this
research by giving data for major, minor, and trace elements in large numbers of fine and
highly size-resolved aerosol samples. The data for the major elements (S, Na, Cl, Al, Si, Fe)
are needed for assessing the climatic effect of aerosols, for estimating the contributions of
important aerosol types (e.g., sea salt, crustal material), and for arriving at chemical mass
closure, whereas the multielemental data set as a whole (which comprises data for various
anthropogenic tracers) can be used for resolving the contributions from different source
categories. On the other hand, one should realize that PIXE only provides part of the
desired information. With regard to the chemical composition, it is highly desirable to
perform also measurements for important ionic species (e.g., ammonium, nitrate) and for
organic carbon (OC) and elemental carbon (EC). Furthermore, in order to arrive at a
complete aerosol characterization, the chemical measurements have to be complemented
with physical and optical (radiative) aerosol measurements.
Recent insights in the importance and role of atmospheric aerosols were provided by
Andreae and Crutzen (1997). The chapter by Cahill (1995) in the textbook of Johansson
et al. (1995) gives a comprehensive overview of the applications of PIXE to aerosol
research up to the early 1990s and gives valuable information on various aspects of
aerosols. In a review paper by Maenhaut (1992), PIXE applications of the early 1990s to
aerosols and other environmental samples are discussed.
2. Examples
There is a wealth of examples on the application of PIXE to atmospheric aerosol samples.
It is comforting to note that publications in which PIXE was used are not only published
in key journals on PIXE and the IBA techniques, such as Nuclear Instruments and Methods
Copyright © 2002 Marcel Dekker, Inc.
Figure 14 Cross section of the SDI. Only 4 of 12 impaction stages are shown. (From Maenhaut
et al., 1996d.)
in Physics Research B and the International Journal of PIXE, but also in atmospheric and
environmental chemistry journals, such as Atmospheric Environment, Journal of Geophy-
sical Research (section Atmospheres), and others. The aerosol samples analyzed by PIXE
range from those collected indoors (e.g., in work environments) or near specific pollution
sources to samples dealing with urban, regional, and global air-pollution problems, and to
samples collected in areas as remote as the Arctic and Antarctic.
a. Work Environments and Near Specific Sources
An example of a study in a work environment is that of Formenti et al. (1998b). These
researchers collected aerosols inside an arc-welding shop with a single-stage streaker
sampler, using a time resolution of 1 h. By examining the temporal variability and the
correlations among the airborne particulate elements, they extracted an arc-welding source
profile (i.e., the composition of the emitted particles). It was also found that the welding
shop activities gave rise to concentrations of up to 100 mg=m3 for Fe and up to 7 mg=m3 for
Cr. Studies on heavy metals in the atmosphere near copper, zinc, and steel smelters were
conducted in Chile (Romo-Kröger et al., 1994), China (Zhu and Wang, 1998), and Italy
(Prati et al., 1998). The study near the copper smelter in Chile (Romo-Kröger et al., 1994)
was done over a period of 1.5 months, but because of a major labor strike, there was a
total shutdown of the smelter during 22 days. It was found that S, Cu, Zn, and, especially,
As were quite enriched in the air during normal working periods relative to the strike
period. The average fine ( < 2.5 mm EAD) As level at the sampling site (at 13 km from the
smelter) during normal operation was 240 ng=m3.
Also, the aerosol emissions from nonindustrial and larger-scale sources, such as the
1991 Kuwaiti oil fires and tropical forest and savanna biomass burning, have been in-
vestigated using PIXE [e.g., Cahill et al. (1992), Reid et al. (1994), Echalar et al. (1995),
Gaudichet et al. (1995), Maenhaut et al. (1996c)]. However, in most of these studies, the
PIXE measurements were complemented with other analyses [e.g., for black carbon (BC)]
in order to arrive at a more complete characterization of the emitted particles. Some of the
collections for the Kuwaiti oil fires (Cahill et al., 1992) were done with SFU samplers that
were installed on an aircraft. The samples were subjected to optical, gravimetric, scanning
electron microscopy (SEM), and PIXE analyses, yielding information on the morphology,
mass, and composition of the aerosols. It was found that the mass in the coarse size
fraction was dominated by soil-derived particles. In the fine fraction, organic matter and
Copyright © 2002 Marcel Dekker, Inc.
fine soils each accounted for about one-fourth of the mass, whereas salt and sulfates
contributed about 10% and 7%, respectively. The work on the aerosol emissions from
tropical forest and savanna biomass burning (Echalar et al., 1995; Gaudichet et al., 1995;
Maenhaut et al., 1996c) resulted in source profiles which can be used in chemical mass-
balance receptor (CMB) modeling work on ambient aerosol datasets in order to apportion
the contribution from biomass burning to the PM and various aerosol constituents
(Maenhaut et al., 1996a, 1996c).
b. Urban Areas
Numerous PIXE application papers have dealt with the study of atmospheric aerosols in
urban areas. Considering the renewed interest in the health risks of aerosols in such areas
(Pope et al., 1995), PIXE may continue to find an important application niche in this field.
Miranda (1996) presented an overview of the studies in large urban areas in which PIXE
was used and which were published up to 1995. Besides assessing the airborne levels of the
heavy metals and of important elements such as S, major objectives in most studies are to
identify the major sources (source types) of the heavy metals and elements and to ap-
portion them to the sources. For this source identification and apportionment, one mostly
relies on multivariate techniques, such as absolute principal-component analysis (APCA),
absolute principal factor analysis (APFA), and multiple regression analysis. Occasionally,
these approaches are complemented or compared with CMB analysis and wind direction
analysis.
Several urban aerosol studies with PIXE were carried out in Latin America, parti-
cularly in Mexico City, Santiago de Chile, and Sao Paulo city (Miranda et al., 1994, 1998;
Andrade et al., 1994; Cahill et al., 1996; Aldape et al., 1996). Andrade et al. (1994) applied
APCA to their coarse and fine aerosol data from Sao Paulo city and also to a dataset of
various meteorological variables, and they subsequently performed a correlation analysis
between the various matrices with ‘‘absolute principal-component scores’’ (APCS). This
indicated, for example, that the industrial components of both size fractions were posi-
tively correlated with the northeast wind direction. For Mexico City and Santiago de
Chile, the aerosol data were compared with those from Los Angeles, California (Cahill
et al., 1996). A rather surprising result was the rough equivalence among the three cities in
terms of several classes of fine particles. Coarser particles were, however, far more pre-
valent in Mexico City and Santiago than Los Angeles.
In Europe, extensive work with PIXE has been done on urban aerosols from
Hungary (Molnár et al., 1993; Ali et al., 1994; Borbély-Kiss et al., 1996; Salma et al.,
1998). In a study by Molnár et al. (1993), data for downtown Budapest were compared
with those for suburban Budapest and for rural air, and selected suburban and rural data
were subsequently used in a source–receptor model for estimating the dry deposition for a
number of elements. Urban aerosol studies in which PIXE was used were also done
in several other European countries (Climent-Font et al., 1994; Luis-Simon et al., 1995;
Braga Marcazzan, 1996; Swietlicki et al., 1996b; Harrison et al., 1997; Maenhaut and
Cafmeyer, 1998; Wrobel and Rokita, 1998). Swietlicki et al. (1996b) combined their PIXE
data for Lund, Sweden with results of gaseous species (SO2, NO2, O3) and examined the
combined dataset with APCA. About one-third of the SO2, most of the fine particulate S,
and about half of the fine Zn and Pb were attributed to a regional background source
originating from ferrous and nonferrous smelters. Also, Harrison et al. (1997) used PIXE
aerosol data together with results from other measurements and examined the combined
datasets with receptor modeling. In addition to two European cities (Birmingham, UK,
and Coimbra, Portugal), Lahore, Pakistan was included in this study. It was found that
Copyright © 2002 Marcel Dekker, Inc.
there were considerable similarities between the two European cities, but large contrasts
with Lahore, with its dryer climate and poorly controlled pollution sources. Other cities in
Asia where aerosol studies with PIXE have been performed are Singapore (Bao et al.,
1995) and Kyoto, Japan (Kasahara and Gotoh, 1995).
An example of a comprehensive study on urban aerosols from Australia is the work
by Chan et al. (1997). PM10 samplers were used during 1 year at five sites in Brisbane (and
a dichotomous sampler at one of the sites), and the samples were analyzed by a variety of
techniques, including PIXE, other IBA techniques, ion chromatography (IC), and atomic
spectrometry. The major components in the PM10 aerosols were crustal matter (25% by
mass), organics (17%), sea salt (12%), elemental carbon (10%), and ammonium sulfate
(7%). Furthermore, apart from significant local influences at some of the sites (such as a
cement factory), most anthropogenic emissions appeared to be rather evenly and widely
distributed in Brisbane. Also, urban aerosols from Africa have been analyzed by PIXE
[e.g., from Khartoum, Sudan (Eltayeb et al., 1992, 1993), from Lagos and Ile-Ife, Nigeria
(Ogunsola et al., 1993), and from Soweto, South Africa (Formenti et al., 1998a)].
e. Micro-PIXE
Whereas normal macro-PIXE is very extensively used for analyzing atmospheric aerosols,
the application of micro-PIXE and complementary nuclear microprobe techniques to this
sample type has been much more limited. The emphasis in contemporary aerosol research
is clearly placed on the submicrometer-sized particles, and to examine individual aerosol
particles of such sizes, a nuclear microbeam of 0.1 mm with an intensity of 100 pA would be
very welcome. However, progress in the development of such beam sizes has been slow.
Also, atmospheric aerosols consist of a heterogeneous population of particles of widely
variable size and composition. In order to obtain results that are representative for the
Copyright © 2002 Marcel Dekker, Inc.
aerosol as a whole, thousands of individual particles should preferably be analyzed.
However, due to the fact that the nuclear microprobe analysis is rather time-consuming, at
most a few hundred particles are analyzed in micro-PIXE analyses of aerosols. For the fast
multielemental analysis of large numbers of submicrometer-sized particles, EPMA is
certainly to be preferred over micro-PIXE, and this is despite the fact that its detection
limits (in micrograms per gram) are about two orders of magnitude worse and that ac-
curate quantification in it is much more problematic. Rather than trying to compete with
EPMA in the analysis of aerosols, micro-PIXE should be used as a complementary
technique, for example, for obtaining additional information and=or for special aerosol
samples or problems where accurate quantitative results and=or data for real trace ele-
ments are needed.
Orlić (1995) reviewed the nuclear microprobe work on atmospheric aerosols that was
carried out up to about 1994. Examples of micro-PIXE applications on ambient (outdoor)
aerosols since then are the studies by Orlić et al. (1995) on particles from Singapore, of
Maenhaut et al. (1997a) on aerosols from Israel, of Gerab et al. (1998b) on Amazon Basin
aerosols, and of Rajta et al. (1998) on aerosols from Debrecen, Hungary. In the study by
Maenhaut et al. (1997a), micro-PIXE was used in combination with micro-Rutherford
back-scattering (RBS) to determine the quantitative composition of individual particles
and to differentiate, for calcareous aerosols, between particles consisting of pure com-
pounds and of simple mixtures. For the latter type, the contribution from each compound
was quantitatively determined. In the study by Rajta et al. (1998), a quantitative PIXE
analysis was performed on 412 single particles (of 2–12 mm in diameter). The resulting
dataset was examined by hierarchical cluster analysis, and several types of soil particles
were identified. De Bock et al. (1996) employed micro-PIXE as a complement to EPMA in
their study on indoor aerosols from the Correr museum in Venice, Italy. They used the
Oxford SPM to produce elemental maps for giant particles ( > 8 mm). Because of the better
detection limits, micro-PIXE allowed the detection of elements such as Ti, Cr, Fe, and Zn.
1. Combustion Sources
The physicochemical characterization of particles from combustion units is of importance
from a purely environmental viewpoint, but also to understand and improve the com-
bustion technology. The environmental impact comes from particles that are emitted
through the stack into the atmosphere, but also from the particulate material that is
collected by electrostatic precipitators (ESPs) or other cleaning devices, or is recovered as
bottom ash. For ESP ash and bottom ash, the levels of heavy metals are of critical im-
portance for making decisions on their use as construction material, their distribution to
agricultural or other land, or their disposal as waste.
Copyright © 2002 Marcel Dekker, Inc.
Several studies involving micro-PIXE have been done on particles from coal com-
bustion (Jakšić et al., 1993; Caridi et al., 1993; Bogdanović et al., 1994, 1995; Cereda et al.,
1995a, 1995b, 1996; Rousseau et al., 1997; Hickmott et al., 1997). The PIXE groups from
Zagreb and Milan, in cooperation with the Oxford SPM group, have been especially active
in this area. The fly-ash and other ash particles usually had sizes of several micrometers,
which makes them more appropriate objects for nuclear microprobe investigations at the
individual particle level than, for example, submicrometer-sized atmospheric aerosols.
Bogdanović et al. (1994) used STIM in combination with micro-PIXE in order to obtain
information on the areal density and topology of individual fly-ash particles and to be able
to convert the elemental intensity maps into actual concentration distributions. Cereda
et al. (1995b) examined about 100 particles that were collected at the outlet of a pilot ESP
of a coal-fired power plant. As a result of the combustion of the different components of
the mineral matter in the parent coal, seven particle classes with characteristic matrix
compositions and different trace element concentrations were observed. The trace elements
were found to be nonuniformly distributed over the particles. Elements which are asso-
ciated with volatile compounds, such as S, Ni, Zn, and Ga, were clearly more concentrated
in the smaller particles ( < 2 mm) than in the larger ones.
Macro-PIXE in combination with other techniques has been used for the analysis of
cascade impactor samples that were collected at various locations in the flue gas stream of
coal burners (Maenhaut et al., 1993c; Lind et al., 1994, 1996) and recently also of biomass
combustion units. In their study on the volatilization of heavy metals (Cu, Zn, Cd, Pb)
during circulating fluidized bed combustion of forest residue, Lind et al. (1999) found that
none of these metals were enriched in the fine particles at the inlet of the ESP. In order to
obtain source profiles from the combustion of herbaceous and wood fuels, Turn et al.
(1997) conducted biomass burning experiments in a wind tunnel, collected the emitted
particles with a variety of samples, including DRUM impactors, and analyzed the samples
by PIXE, XRF, and some other techniques.
2. Aqueous Environment
Liquid samples are not very suitable for direct analysis by PIXE. They are often analyzed
after pipetting a fraction on a suitable backing film, letting it dry, and then bombarding the
residue. However, in order to obtain detection limits that can compete with those in ICP–MS
or AAS, one has to resort to preconcentration techniques. Savage et al. (1995) experimented
with dried algae for preconcentrating metallic trace elements from water for subsequent
PIXE analysis. For the preconcentration of precipitation samples, Tschierch et al. (1996)
built a spray-drying apparatus that was similar to that developed by Hansson et al. (1988).
Examples of applications of PIXE to rainwater are the work of Ghorai et al.
(1993) in Alabama (USA) and of Kasahara et al. (1996) in Kyoto, Japan. In the latter
study, rainwater was collected with 0.1 mm of rainfall from the start of each rain, and the
samples were filtered through a 0.2-mm pore-size Nuclepore filter, so that the soluble and
insoluble components could be determined. In accordance with earlier studies, it was
found that the concentrations of the elements in the rain decreased quickly until about
0.5 mm of rainfall. Si, Ti, and Fe were mainly present in the insoluble component of the
rain, whereas virtually all the S and Cl was soluble. River water from India and Portugal
was analyzed by PIXE (Kennedy et al., 1998; Araujo et al., 1998). Cecchi et al. (1996) gave
an overview of their comprehensive study on the Venice Lagoon. Somewhat special water
samples are ice cores drilled from glaciers or inland ice. Such samples are collected
to investigate variations in their composition and trace element content with depth
Copyright © 2002 Marcel Dekker, Inc.
(e.g., along the core) and, by doing so, to obtain information on past environmental or
climatic changes or to assess the impact of major volcanic eruptions. PIXE offers the
possibility of analyzing ice cores with millimeter resolution, so that the seasonal and an-
nual elemental variation can be studied in great detail (Hansson et al., 1993). Samples for
PIXE may be prepared by placing ice sections on a thin backing film and removing the ice
by sublimation. Hansson et al. (1993) used this approach for samples from a deep core,
which was drilled at Dye 3, South Greenland. The spatial (and thus temporal) variations in
the elements measured by PIXE were used to assist in the dating of some sections from the
core. Laj et al. (1996) coupled macro-PIXE with EPMA for characterizing atmospheric
aerosol particles and soluble species in ice cores from Summit, Central Greenland.
Several PIXE studies have been devoted to the study of sediments in rivers, estuaries,
or the sea (Randle et al., 1993; Limić and Valković, 1996; Valković and Bogdanović, 1996;
Martı́n et al., 1996, 1998a, 1998b; Benyaich et al., 1997; Dutta et al., 1998; Tang et al.,
1998), mostly to assess the anthropogenic input of heavy metals and other elements. Randle
et al. (1993) compared INAA and PIXE for the determination of heavy elements in es-
tuarine sediments and concluded that the two techniques should be regarded as com-
plementary rather than competitive. Valković and Bogdanović (1996) used a combination
of PIXE and XRF. A total of 19 elements were measured by PIXE and 12 by XRF. For
elements that were determined by both techniques, reasonable agreement was observed.
Also, Benyaich et al. (1997) made a comparison of PIXE and XRF, and they concluded
that the two techniques are complementary. Dutta et al. (1998) analyzed ferromanganese
oxide deposits from different locations of the Indian Ocean by PIXE. Based on the Mn=Fe
ratio, the deposits were classified into hydrogenous and hydrothermal types, and the dis-
tribution of various minor and trace elements in both types were studied. Several elements
exhibited higher concentrations in the hydrogenous deposits than in the hydrothermal
ones. This was attributed to suitable physicochemical oceanic conditions prevailing at the
depositional sites and the growth rate of the deposits. The concentrations of Co, Ni, and Cu
were dependent on ocean depth. It was also found that the hydrothermal processes seem to
be more controlling in the uptake of V and As than the hydrogenous processes.
The most spectacular use of PIXE in the past few years has undoubtedly been the
deployment of the portable a-proton x-ray spectrometer (APXS) on Mars (Rieder et al.,
1997a, 1997b). The APXS was a compact (600 g) device installed on board the Sojourner
rover of the Mars Pathfinder. Alpha-particles from a 50-mCi 244Cm source were used for
the excitation, and the x-rays were detected with a thermoelectrically cooled Si p-i-n de-
tector with a resolution (FWHM) of 250 eV at 6.4 keV. Besides PIXE, RBS and NRA
[using (a, p) reactions] were employed with the APXS. The PIXE results indicated that the
Martian rocks were high in Si and K, but low in Mg compared to Martian soils and
Martian meteorites. The analyzed rocks were similar in composition to terrestrial ande-
sites and close to the mean composition of the Earth’s crust. Addition of a mafic com-
ponent and reaction products of volcanic gases to the local rock material is necessary to
explain the soil composition.
E. Materials Analysis
Materials analysis is a vast field of application for ion-beam analysis (IBA) techniques.
A comprehensive treatment of the various IBA techniques (with the exception of PIXE) in
the field of materials analysis can be found in the Handbook of Modern Ion Beam Materials
Analysis (Tesmer et al., 1995). Rutherford backscattering spectrometry (RBS) in particular
is very commonly used for examining surface layers and determining elemental depth
distributions in various materials. PIXE generally serves only as a complementary tech-
nique. Because most matrices in this area are made up of elements with relatively high
atomic number, the characteristic x-rays of the matrix elements show up in the PIXE
spectra, thereby worsening the detectability of the trace elements. When the matrix is
rather light, however, as in the case of polymers, some semiconductor materials, and
cellulose fibers, PIXE can provide concentration data down to the microgram per gram
level. For microscale characterization of materials, the nuclear microprobe has emerged as
an important technique (Breese et al., 1996), in particular when micro-PIXE is combined
with complementary IBA techniques. It has therefore also been implemented in the in-
dustrial environment (e.g., in development and production plants in the microelectronics
industry). Although the micro-PIXE technique provides essentially only elemental maps
and=or concentration data, the combination with various imaging techniques makes the
nuclear microprobe an invaluable tool in materials analysis. It forms a fine complement to
the vast array of surface and near-surface characterization techniques, such as Auger
electron spectroscopy, photoelectron spectroscopy, scanning electron microscopy, sec-
ondary ion mass spectrometry (SIMS), and many other techniques (Brune et al., 1997).
1. Semiconductor Materials
The majority of analyses on this type of material are performed as routine PIXE and RBS
measurements in semiconductor plants, with the aim of monitoring wafer processing.
However, PIXE and nuclear microprobes are also frequently used in research and de-
velopment environments for the characterization of silicon-based and other semi-
conductor materials, such as GaAs. Examples of macro-PIXE studies on the latter
material are the work by Kuri et al. (1996) and Wendler et al. (1996). Kuri et al. (1996)
used PIXE and XRF to study the effect of MeV Au ion implantation on the composition
of GaAs upon vacuum annealing. Practically no As loss was observed from the implanted
Copyright © 2002 Marcel Dekker, Inc.
region, whereas such loss was significant for the unimplanted region. Optical micrographs
taken on both regions showed the formation of Ga droplets on the unimplanted region
above 600 C annealing. No such features were observed in the implanted region up to
850 C. This indicated that MeV Au ion implantation inhibits the As release from GaAs.
For a given annealing temperature and duration, the amount of As loss was dependent on
the dose of implanted Au. Wendler et al. (1996) measured displacements in 2-MeV Seþ
implanted GaAs by RBS and PIXE channeling experiments. It is also possible to combine
micro-PIXE and micro-RBS with channeling. Under channeling conditions (i.e., when the
projectiles are ‘‘guided’’ along crystal axes or crystal planes), the interaction probability is
decreased and this leads to a reduced yield of induced x-rays or scattered particles.
By comparing the yields obtained for channeling and for random orientation, information
on the crystal structure and on interstitial atoms may be obtained (King et al., 1993;
Tesmer et al., 1995). To visualize the results from micro-PIXE and micro-RBS channeling
experiments, channeling contrast microscopy (CCM) can be used. The Microanalytical
Research Centre (MARC) in Melbourne is very active in studies on frontier materials,
mainly semiconductors (Jamieson, 1997). It applied micro-PIXE, micro-RBS, and CCM
in a study on diamond ion implantation and annealing. A diamond specimen was im-
planted with 4-MeV P þ ions, whereafter the implanted region was laser annealed and
regrown, and examined with the nuclear microprobe. Figure 17 shows the channeling
angular yield curves from the center of the laser-annealed and regrown region, together
with the CCM images. The reduction in the yield of PKa x-rays in the channeling
Figure 17 Yield of PIXE and RBS versus the tilt angle from the channeling direction in diamond.
CCM images are shown for both scattered particles and P x-rays. The CCM images cover an area of
1006100 mm2. (From Jamieson, 1997, with permission from Elsevier Science.)
3. Metals
In most of the nuclear microprobe studies on metallurgical samples, use was made of NRA
in order to determine the low-Z elements (Doyle et al., 1991; Breese et al., 1991). Because
metals and metal alloys consist generally of high-Z elements, they are far from ideal
specimens for PIXE. Nevertheless, by optimizing experimental conditions, it is possible to
study, for instance, corrosion processes and diffusion profiles in such materials. One ex-
ample of a micro-PIXE study is that done on proposed superconducting wires for magnets
in the International Thermonuclear Experimental Reactor (ITER). These wires contain Sn
and Nb in a matrix of ultrapure Cu, and they are plated with Cr. Micro-PIXE was used to
study the diffusion of Cr into the Cu matrix as a function of the temperature (Morse et al.,
1997). Already at rather low temperature, unexpectedly high Cr concentrations were
found in the Cu, and this was interpreted as being due to the formation of precipitates.
Other examples of metal analysis using micro-PIXE are the study of nodular cast in order
to draw conclusions regarding the nodularization process (Songlin et al., 1995) and the
investigation of lead in a chill-cast aluminum ingot of a bearing alloy (Breese et al., 1992).
Copyright © 2002 Marcel Dekker, Inc.
In the latter case, a combination of PIXE, backscattering spectroscopy, and STIM was
used in addition to traditional techniques, such as an electron microprobe. The lead
precipitates could be identified and localized.
There are also examples of the application of macro-PIXE to metals and alloys.
Feng et al. (1996) used it to determine the antinodularizing elements Al, Pb, and Bi in
nodular cast iron. Wätjen et al. (1996) used a combination of PIXE, RBS, and resonant
elastic backscattering to determine the depth profiles and doses of Y implants in Cr and
NiCr alloys, the mass contents of Cr and Ni, and the oxygen concentration in surface
layers of the alloys before and after implantation. Only through the simultaneous use of
PIXE was a severe error in the alloy preparation revealed. One of the nominal binary
alloys was found to contain about 5% Fe, which influences the corrosion behavior of this
material significantly. Nakae et al. (1994) used a combination of PIXE channeling and
RBS channeling for characterizing single crystals of type 304 stainless steel. He þ particles
were used as incident ions for both types of measurements. The study indicated that a
solution annealing process is absolutely necessary for producing a good single crystal. It
was further found that the P atoms were mostly on substitutional sites of the face-centered
cubic (fcc) structure and that the MeV He þ ion irradiation induced segregation of Si and S
atoms to the (110) surface. In a subsequent study (Kawatsura et al., 1996), the radiation-
induced segregation was further examined.
4. Cellulose Fibers
The quality and properties of paper used for printing and various other purposes are
highly variable, depending on the particular demand from the industry which uses the
produced paper. One field of great economical importance worldwide is the newsprint
industry, where extensive use is made of recycled fiber material and where very tight time
constraints exist in the printing phase. There is great demand for an enhanced under-
standing on the right combination of printing technique and paper quality. At Lund, a
project involving the nuclear microprobe, and partially funded directly from industry, was
initiated as part of a larger project on physical characterization methods for newsprint
Figure18 Copper and STIM map from investigation of newsprint paper by a nuclear microprobe.
The Cu distribution reflects the pigment distribution and the STIM reflects the mass areal density
of the cellulose fibers. Dark areas indicate high concentration and light areas indicate low
concentration. Scan size: 0.5 mm. (Courtesy of P. Kristiansson.)
On several occasions in this chapter, it has already been indicated that the MeV ion beams
employed for PIXE are also very useful for other ion-beam analysis (IBA) techniques. The
main interaction of MeV ions are close encounters, but also ‘‘shallow’’ encounters leading
to recombinations in the outer electron shells can be used by detecting the near-visible light.
In this section, the various IBA techniques are briefly presented. In particular, it is shown
through examples of applications that the other IBA techniques can provide information
that is complementary to that obtainable by PIXE (e.g., by extending the elemental
coverage down to hydrogen) and that such analyses may often be done simultaneously with
the PIXE analysis. This represents an important advantage of PIXE over other x-ray
spectrometric techniques. For comprehensive accounts on the various complementary IBA
techniques, we refer to a number of textbooks (Chu et al., 1978; Ziegler, 1975; Thomas and
Cachard, 1978; Bird and Williams, 1989; Tesmer et al., 1995; Breese et al., 1996) and the
proceedings of the biennial IBA conferences [see Gyulai et al. (1994), Culbertson (1996),
and da Silva et al. (1998) for the last three conferences in the series].
2. Ionoluminescence
When a significant amount of energy is transferred to a crystalline material, be it by elec-
trons, photons, or charged particles, near-visible light is emitted due to luminescence in the
material. The well-known techniques of photoluminescence (PL) and cathodoluminescence
(CL) are widely used in material science and geology in order to study the chemistry,
structure, or electrical properties of the material. The exact character of the impinging
radiation does not matter very much for the basic interaction mechanisms involved in lu-
minescence. As a consequence, much of the physical understanding from the PL and CL
techniques can be directly incorporated in the ionoluminescence technique (IL) that was
introduced and developed in a collaboration between the nuclear microprobe groups of
Lund and Melbourne (Yang et al., 1993). The intensity and wavelength of the ionolumi-
nescent light provide information concerning the nature of luminescence centers, such as
trace substituents and structural defects, present in the matrix. This makes IL a useful
complement to other IBA techniques, such as PIXE and, to a some extent, also RBS.
Luminescence can be divided into two subgroups (intrinsic and extrinsic) according
to its origin. The intrinsic luminescence, which usually contributes in emission through
crystal structural defects, is not related to impurities but to crystal-lattice properties. The
extrinsic luminescence depends on the impurities in a crystal through the processes acti-
vation, sensitization, and quenching. Luminescence phenomena are common in many
solids, yet the use of the light produced is restricted because of the complex physical
mechanisms behind the emission. For insulators, to which many minerals belong, crystal-
field theory can be used to explain luminescence (Henderson and Imbusch, 1989), whereas
for semiconductor crystals, band theory is often employed (Yacobi and Holt, 1990). As
long as no secondary effects such as lattice damage or crystal modification occur, the
general luminescence properties of a material are not dependent on the nature of the
excitation source. For excitation by energetic particles, the processes leading to lumines-
cence take place in three steps. First, the energy of the MeV particles is dissipated into the
excitation volume mainly through electronic stopping and partially through nuclear
stopping processes. Ionization takes place in the excitation volume. Recombination of the
electrons and the excited ions allows the crystal lattice to absorb the energy released and
the optical system becomes highly excited. The second stage involves the de-excitation of
the states of high excitation through radiationless transitions. The third stage, lumines-
cence emission, occurs when the atoms de-excite from a low-excitation state to the ground
level. The term activation is used when trace impurities cause a material to produce ex-
trinsic luminescence. Transition metal ions, with an electron configuration of 3d (1–9), can
interact strongly with the crystal field. This results in changes in the energy level structure
of the free ion. Usually, the luminescence activated by transition metal ions is char-
acterized by a broad peak width and a peak position strongly dependent on the host
matrix. The centroid and width of the peak are sensitive to the chemical surroundings in
the crystal and may be used to distinguish the host matrices involved. For example, with
Copyright © 2002 Marcel Dekker, Inc.
Mn2þ (3d5) ions located in a crystal with a strong crystal field, the luminescence produced
tends to be of longer wavelength. The luminescence of Mn2þ-activated calcite (Mn2þ ions
in octahedral coordination), for example, is orange, whereas the luminescence of Mn2þ -
activated willemite (Mn2þ ions in tetrahedral coordination) is greenish.
Trivalent rare-earth element ions and most divalent rare-earth element ions, except
for those of Ce, Gd and Lu, have the electron configuration 4f (k)[5s25p6]. These special
electron configurations give rise to characteristic narrow luminescence emission bands for
rare earth element (REE) ions with the configuration 4f (k1)f 1* through a f ! f tran-
sition. The narrow luminescence emission band of the REE ion (trivalent or divalent)
originates in the 4f subshell, which is partially shielded by [5s25p6] electrons. Therefore, for
REE ions with a 4f ðk1Þ f 1 configuration, the structure of the energy levels in the free REE
ions is basically the same in different host matrices. Figure 22 shows an IL spectrum from
a natural zircon grain, where three peaks from REE are superimposed on a broad intrinsic
luminescence band.
Experimentally, the IL light is detected by using mirrors or lenses and either directly
counting the photons in a PM tube or analyzing them in a spectrometer. The detailed
spectrum (see Fig. 22) provides information on the crystal structure and can be used to
determine the REE. A scanning grating spectrometer is a very slow system. Variation in
beam current, beam-induced lattice damage, and so forth can adversely affect the acquired
spectrum. The situation can be improved by implementing a photodiode-array detector
with, for instance, 1024 detector elements, each with a size of a few tens of micrometers.
Such a system allows the simultaneous recording of all wavelengths and can significantly
improve the analytical capacity, speed up the analysis, and reduce beam damage effects.
Furthermore, because ion-beam bombardment of a solid material modifies its properties
and changes its luminescence characteristics, the rapid spectrum acquisition makes it
possible to perform a detailed study of these transformations.
The most common experimental facility for IL is the nuclear microprobe, where the
imaging capability of IL is an invaluable asset, especially in studies on geological samples.
Figure 22 Ionoluminescence spectrum from a natural zircon with peaks, emanating from various
rare earth elements, superimposed on a broad yellow intrinsic luminescence band.
VIII. CONCLUSIONS
Particle-induced x-ray emission is undoubtedly an invaluable and very powerful x-ray
emission spectrometry technique. It has truly multielemental capability, covering a large
part of the periodic system (from Na to U), with detection limits that vary smoothly as a
function of atomic number. The use of x-ray absorbers can improve this Z dependence
further. The sensitivity of PIXE is high and detection limits are low, although they do
depend on the particular material being analyzed. The most favorable situation is the
determination of trace elements in a light-element matrix (Z < 11), where detection limits
are of the order of 0.1–1 mg=g. For samples with appreciable amounts of heavy and
medium–heavy elements, the detection limits are somewhat higher; for example, typically
a few micrograms per gram for elements between Z ¼ 20 and Z ¼ 50 in geological mate-
rials. In absolute terms, the detection limits are even more pronounced, down to 1012 g in
macro-PIXE and down to 1017 g in micro-PIXE. Other favorable features of PIXE are
the ability to analyze tiny samples (1 mg or less in macro-PIXE, and much less in micro-
PIXE), the speed of the analysis (1–10 min bombardment time per specimen), and the
possibility for automation. The degree of specimen preparation needed in PIXE varies
from zero at one extreme (e.g., aerosol samples and archaeological materials) through
modest (e.g., mineralogical samples and materials science) to significant at the other ex-
treme (e.g., some biomedical specimens). The specimens are normally in solid form, and
the irradiated target material is, in general, not affected by the particle beam at the current
densities needed for conventional macro-PIXE, so that the technique is nondestructive.
Even delicate materials such as paper or parchment are unaffected by so-called millibeams,
although in these cases, special precautions such as low beam intensity and=or external
beams are necessary. In micro-PIXE, the larger current densities incident on the specimen
can cause damage and alteration, but, in practice, analyses are often conducted at current
densities that do not cause such problems. However, as beam diameters become smaller
and current densities higher, damage will become more important.
Particle-induced x-ray emission is extremely versatile in terms of the size of sample
that it can accommodate. Depending on the size, different experimental arrangements may
be used: a conventional macrobeam in vacuum (with beam size of the order of several
millimeters), an external macrobeam or millibeam, or a genuine microbeam. The variety of
specimens analyzed to date is enormous, ranging from a large painting to a single blood
cell. They have in common, though, that they are normally in solid form. Micro-PIXE as
used in the nuclear microprobe makes it possible to perform analyses with excellent spa-
tial resolution (down to 1 mm or better), high sensitivity (detection limits of the order of
1 mg=g), and good accuracy. The importance of this can hardly be overestimated. For
example, microscopically small grains in minerals or single cells in biological tissue can be
analyzed. If the microbeam is used in the scanning mode, as is often done, elemental maps
can be produced, giving a much more detailed picture of trace element distributions than
measurements made at single points.
Copyright © 2002 Marcel Dekker, Inc.
Another favorable feature of PIXE is that it can be complemented with other IBA
techniques, such as RBS and other forms of elastic scattering spectrometry, NRA, and
PIGE. As a result, a simultaneous measurement of the light elements (H, Li, B, C, N, O,
and F), which cannot be measured by PIXE, is feasible. This is an important asset when
analyzing atmospheric aerosol samples, as discussed in detail by Cohen (1993, 1998),
Cahill (1995), and Swietlicki et al. (1996a), but also for various other sample types. Some
of the other IBA techniques, such as RBS and NRA, predate PIXE, and in certain fields of
application (e.g., materials analysis), RBS is the major IBA technique, and PIXE generally
serves only as a complementary technique. Some of the ancillary IBA techniques (e.g., off-
axis STIM and ionoluminescence) have been developed in the last decade as adjuncts to
micro-PIXE, demonstrating the continued development of the IBA field overall.
Particle-induced x-ray emission combines a particular and unique set of advanta-
geous features, but as any technique, it has also it limitations. Limitations which are also
shared by XRF are that it suffers from spectral interferences and matrix effects and that it
does not allow the direct measurement of ultratrace elements that are present at nano-
grams per gram levels. Unlike in some other techniques, however, the matrix effects are
well understood and can be corrected for. Consequently, PIXE can provide results with an
accuracy better than 5% relative. The most serious drawbacks of the technique are that it
requires a MeV particle accelerator and that commercial PIXE apparatus are not readily
available (the great majority of PIXE laboratories have built their own setup, and com-
mercial systems are usually custom-built). Most PIXE laboratories have been set up in
nuclear physics institutes, and although there have been interesting developments during
the past decade, with the installation of dedicated PIXE laboratories and nuclear micro-
probes, the growth in PIXE has remained slow. There are currently more than 100 PIXE
groups worldwide.
Part of the reason for the slow growth of PIXE is, of course, that the technique has
to compete with a wide plethora of other analytical techniques and that there have been
substantial advances in several of these techniques over the past 10–15 years, for example,
in ICP–MS. It is therefore worthwhile to compare the characteristics and capabilities
of PIXE with those of the other techniques and to assess for what sample types and=or
problems PIXE is most appropriate. Maenhaut (1990b) made such an evaluation for
macro-PIXE 10 years ago, which was retained in the Conclusions section in the chapter on
PIXE in the first edition of Handbook of X-Ray Spectrometry (Maenhaut and Malmqvist,
1992). Although many of the conclusions still stand, that evaluation is now somewhat
outdated. A more recent evaluation is that of Johansson and Campbell (1995) in the
chapter ‘‘Comparison with Other Methods: Future Prospects’’ in the textbook Particle-
Induced X-Ray Emission Spectrometry (PIXE). In the present Conclusions section, we have
drawn heavily on that excellent and insightful chapter of Johansson and Campbell (1995).
As indicated by them, any comparison with other techniques is fraught with risk. The
range of techniques available means that it is not an easy matter to select the most ap-
propriate for a particular task. It is often the case that proximity, familiarity, and ease of
access play just as large a role as the matching between task and technique. Furthermore,
because of the rapid progress in analytical techniques, any comparison becomes dated
after a few years. The recent progress in x-ray emission spectrometry techniques can be
judged from the various chapters in this handbook. For continued updates on this, the
biennial fundamental review on x-ray spectrometry in the June issue (of every even year) of
the journal Analytical Chemistry can be recommended [see Török et al. (1998) for the most
recent one in the series]. That same June issue of Analytical Chemistry also contains
fundamental reviews on many other analytical techniques. Furthermore, some other
Copyright © 2002 Marcel Dekker, Inc.
analytical chemistry journals, such as Journal of Analytical Atomic Spectrometry, also
publish technique-oriented reviews on a regular basis.
A natural starting point in a comparison of PIXE with other analytical techniques is
conventional tube-excited XRF. In actual numbers, the relative detection limits of PIXE
for thin samples and a light-element matrix vary within the range 0.1–1 mg=g for the ele-
ments of greatest interest (the transition metals and the heavy metals around Pb). With
careful optimization (a-particle excitation, tight geometry, and heavy shielding), detection
limits as low as 0.02 mg=g can even be reached in PIXE (Johansson, 1992). In EDXRF, the
corresponding detection limits are of the order of 1–10 mg=g. Even more pronounced is the
difference in detection limits in absolute terms. To obtain optimum relative detection limits
in XRF, the sample weight has to be at least 10–100 mg, which means that the absolute
detection limits are of the order of some hundred nanograms. In macro-PIXE, the amount
probed by the proton beam can be 0.1 mg or even lower, and the corresponding absolute
detection limits are of the order of a few picograms. However, as discussed by Johansson
and Campbell (1995), besides sensitivity and detection limits, several other factors have to
be considered. There are certainly many sample types and=or problems where EDXRF is
to be preferred over macro-PIXE as the technique of choice.
One substantial difference between PIXE and XRF is that with micro-PIXE, the
spatial distribution of trace elements can be studied down to the submicrometer level,
whereas conventional XRF does not offer this possibility. In the mid-1980s, the use of
glass capillaries for the concentration of the primary radiation from an x-ray tube was
introduced into XRF (see Chapter 11). This and subsequent developments have led to
micro-XRF and desktop x-ray microprobes, whereby a spatial resolution down to 10 mm
with reasonable incident x-ray intensity may be obtained. These are important technical
advances, conferring on tabletop XRF some, although not all, of the abilities of micro-
PIXE. Another significant step forward in XRF was the introduction of TXRF (see
Chapter 9). However, this technique is much more suitable for the analysis of dilute
aqueous solutions than of solid samples. It tends to compete more directly with other
techniques (e.g., optical atomic spectrometry) than with PIXE. There appears, for ex-
ample, to be little potential for high-throughput nondestructive analysis of aerosol par-
ticulate samples. A third far-reaching change in the XRF technique has occurred through
the introduction of synchrotron radiation (SR) as the primary source, which has led to
SRXRF (see Chapter 8). The main limitation of this technique is the same as for PIXE,
but to a much greater extent, namely the availability. In the case of PIXE, it is quite
feasible to set up an accelerator laboratory to be used solely for PIXE and related IBA
work, whereas it is clearly out of the question to use a SR source mainly for elemental
analysis. One is therefore limited to using existing facilities, whose number is small. The
potential of SRXRF undoubtedly lies in its microbeam capability. However, it seems
unlikely, for example, that the continued development and success of SRXRF will affect
the rapid growth of routine micro-PIXE analysis of mineral grains or the growing use of
micro-PIXE and associated IBA techniques in art and archaeometry. Perhaps the main
competition offered to micro-PIXE will be for in situ problems where the energy de-
position in the irradiated specimen is an important issue. Thus, microbeam SRXRF can
complement micro-PIXE by extending in situ microbeam analysis to specimens that
cannot withstand micro-PIXE.
Beyond doubt, the greatest competition for the nuclear microprobe and micro-PIXE
comes from electron probe microanalysis (EPMA) and other electron microscopic tech-
niques (see Chapter 13). Micro-PIXE has the advantage over EPMA that it offers relative
detection limits that are typically two orders of magnitude better, that it is easier to
Copyright © 2002 Marcel Dekker, Inc.
quantify, and that it can be complemented with other IBA techniques. EPMA and other
electron microscopic techniques, on the other hand, have as advantages that the in-
strumentation is more compact and is commercially available and that they offer a much
higher speed of analysis, better spatial resolution, and better imaging capabilities. It is
clear that EPMA will be preferred over the nuclear microprobe in many studies. However,
rather than being competitors of each other, EPMA and the nuclear microprobe can very
well complement each other, whereby micro-PIXE is used for problems where its excellent
detection limits and other specific advantages are required or at least highly desirable.
The majority of bulk element analyses are carried out by optical atomic spectrometry
or atomic mass spectrometry, in particular ICP–MS. In most of these techniques, the
sample material has to be introduced into the instrument as a liquid, and in ICP–MS, a
dilute solution is even preferred. Solid samples must therefore be dissolved, which means
an extra preparation stage with the risk of incomplete dissolution, losses, or contamina-
tion. The dissolution also implies that the analysis is destructive. Although some techni-
ques allow one to analyze powdered materials [e.g., electrothermal atomization atomic
absorption spectrometry (ETAAAS)] or can handle various types of solid samples (e.g.,
laser ablation mass spectrometry), and continued research is done on the introduction of
powders in other techniques, there are overall serious limitations with the analysis of solid
samples by optical atomic spectrometry or atomic mass spectrometry. Such analyses suffer
from severe matrix effects and difficulties in obtaining accurate quantitative results for a
wide range of elements, and they are invariably destructive. Also, certain techniques, such
as ETAAAS, can only measure one element at a time and require the addition of an
appropriate matrix modifier. It is clear that the optical atomic spectrometry and atomic
mass spectrometry techniques differ very much from PIXE. Also, the optical atomic
spectrometry techniques do not offer any equivalent of microbeam capability. Such cap-
ability does exist in some atomic mass spectrometry techniques [e.g., in secondary ion mass
spectrometry (SIMS)] and laser ablation inductively coupled plasma–mass spectrometry
[LA–ICP–MS], but their spatial resolution is clearly worse than in micro-PIXE and
accurate quantification is difficult.
One way to answer the question of which sample types or analytical problems are
better handled by PIXE and micro-PIXE than by other techniques is to survey the various
applications. Section VI provides such a survey, from which one can identify certain
applications where it appears difficult to find alternatives. To a large extent, these are
applications involving solid and particulate sample material and having minimum or
zero sample preparation.
Because of its inherent characteristics, macro-PIXE has been very much applied for
measuring trace elements in various types of biomedical samples. However, for several of
such samples and for ultratrace determinations, the optical atomic spectrometric and ICP–
MS techniques are now better suited. Also, in recent years, the emphasis has shifted from
‘‘total’’ element determination toward elemental speciation. PIXE and the other x-ray
techniques lend themselves much less to this speciation work than some of the other
techniques. In applying PIXE to biomedical problems, one should therefore look for cases
in which one can fully utilize the special advantages of PIXE, such as spatial resolution,
accurate quantitative analysis, and small samples. Particularly micro-PIXE, as used in the
nuclear microprobe and complemented with other IBA techniques such as STIM, is in-
valuable in biomedical trace element research.
One area of application where macro-PIXE has been and continues to be highly
successful is in the analysis of airborne particulate material (atmospheric aerosols). For the
analysis of very large numbers of small aerosol deposits, as collect by compact samplers
Copyright © 2002 Marcel Dekker, Inc.
which provide good time or size resolution, or both, there is virtually no competition to
PIXE. Conventional tube-excited EDXRF cannot be used because its absolute sensitivity
is too low. TXRF has the absolute sensitivity but is not physically appropriate and
SRXRF is much more expensive and not really applicable for routine analysis of large
numbers of samples. The optical atomic spectrometric techniques and ICP–MS require
dissolution, which is time-consuming and may be incomplete for certain important matrix
elements. Conventional ICP–MS also suffers from spectral interferences, causing the
problem that some important elements cannot be measured. PIXE provides at the same
time data for the major elements (S, Na, Cl, Al, Si, Fe), from which the concentrations of
important aerosol types (sulfate, sea salt, crustal material) can be estimated, and for
several anthropogenic and natural minor and trace elements (P, V, Mn, Ni, Cu, Zn, As,
Se, Br, Rb, Pb) that can be used in source type identification and apportionment. Fur-
thermore, by complementing PIXE with other IBA techniques, concentrations can be
obtained for the light elements (H, C, N, O) that make up most of the aerosol mass, and
the hydrogen concentration can then be used to estimate the concentration of the im-
portant organic aerosol type (Cahill, 1995). Considering the current interest in the effects
of fine particles on human health and in the role of aerosols in climate, it is expected that
the analysis of atmospheric aerosol samples will remain a successful application for PIXE
for many years to come. However, as indicated in Sec. VI.B, one should realize that PIXE
provides only part of the desired information. With regard to the chemical composition, it
is highly desirable to also perform measurements for important ionic species (e.g., am-
monium, nitrate) and for organic carbon (OC) and elemental carbon (EC). Also, in order
to arrive at a complete aerosol characterization, the chemical measurements have to be
complemented with physical and optical (radiative) aerosol measurements. PIXE re-
searchers should try to complement their work with these various other measurements or
otherwise cooperate with other groups who are involved in such research, and they should
try to become integrated in larger atmospheric aerosol and atmospheric chemistry research
projects. Fortunately, this is increasingly realized within the PIXE community.
Earth science is another field in which the use of PIXE or specifically micro-PIXE
has increased dramatically during the past decade. Here, however, there are some powerful
competing techniques (e.g., the x-ray spectrometric techniques of micro-XRF and
SRXRF). Furthermore, other physically based analytical techniques continue to develop,
such as SIMS and LA–ICP–MS. SIMS provides spatial resolution of a few micrometers
and relative detection limits that are frequently below 1 mg=g; it is clearly superior to
micro-PIXE for the rare earth elements. LA–ICP–MS routinely offers a spatial resolution
of 20–40 mm and detection limits of 0.5 mg=g. These two techniques also offer isotopic
discrimination, which is not an option with PIXE. However, in each of these two tech-
niques, the matrix effects are much more complex and quantification more difficult than in
PIXE. Micro-PIXE still maintains its advantage as a truly multielemental, in situ, non-
destructive technique with detection limits of a few micrograms per gram, 1 mm spatial
resolution, and straighforward matrix corrections based on simple, well-understood
physics. It can be used for surveys involving a large number of samples for ore pro-
specting, and for the detailed study of minerals. In the latter case, complementary use of
EPMA and micro-PIXE has turned out to be very fruitful. EPMA is a standard technique
in mineralogy for the determination of major and minor elements, and micro-PIXE allows
these studies to be extended in seamless fashion to trace elements. In studies of extra-
terrestrial materials, such as micrometeoritics and interstellar particles, the high sensitivity
and nondestructiveness of micro-PIXE are a prerequisite. Similarly, the study of ore body
emanations gives samples of extremely small mass and any other technique than PIXE
Copyright © 2002 Marcel Dekker, Inc.
seems to be excluded. Micro-PIXE is ideal for investigations of mineral and melt inclu-
sions in many materials (e.g., diamonds). Meanwhile, in the quite different context of
multielement bulk rock analysis with sub-microgram per gram detection limits, INAA,
ICP–AES, and ICP–MS remain the techniques of choice.
Because PIXE and related IBA techniques allow the analysis of delicate samples
without giving any visible damage and without introducing any radioactivity, they can
also very advantageously be used in studies in art and archaeology or for analyzing other
unique samples of which subsampling is impossible. XRF is often the natural choice for
the nondestructive analysis of for example, archaeological artifacts, but PIXE has other
advantages due to its somewhat better sensitivity and imaging capability. A situation in
which the greater sensitivity is desired is the determination of the trace element profile in
various items, which can help in the characterization and identification of material for
addressing questions such as provenance, manufacturing procedure, and trade routes.
Perhaps, the greatest asset of PIXE in these fields is its versatility. Bulk analysis can be
carried out with macro-PIXE, and microscopic details can be studied with a microbeam.
A very useful arrangement is an external beam with a cross section of 0.1–1 mm, called a
millibeam. This can be used for studies of details in paintings or the ink of single letters in
books and other documents. However, the developments in micro-XRF and microbeam
SRXRF will give rise to strong competition for PIXE in this type of application.
Finally, PIXE will certainly find further application in the field of materials research.
However, as indicated earlier, RBS is the major IBA technique in this field, and PIXE
generally serves only as a complementary technique. The complementarity of PIXE resides
in the fact that it is able to detect elements that are present at levels that are too low for
RBS or for which RBS does not offer sufficient Z discrimination.
REFERENCES
Ahlberg M, Akselsson KR, Brune D, Lorenzen J. Nucl Instrum Methods 123:385, 1975a.
Ahlberg M, Johansson G, Malmqvist K. Nucl Instrum Methods 131:377, 1975b.
Ahmed M, Faiz M, Al-Ohali MA, Fageeha O. Nucl Instrum Methods B132:507, 1997.
Aldape F, Flores J, Garcia R, Nelson JW. Nucl Instrum Methods B109=110:502, 1996.
Ali A, Borbély-Kiss I, Szabó G, Koltay E, Medve F, Bacsó J. Int J PIXE 4:9, 1994.
Allan GL, Camakaris J, Legge GJF. Biol Trace Element Res 40:103, 1994.
Amemiya S, Masuda T, Popa-Simil L, Mateescu L. Nucl Instrum Methods B118:403, 1996.
Amsel G, Heitz Ch, Menu M. Nucl Instrum Methods B14:30, 1986.
Andersen HH, Ziegler JF. The Stopping and Ranges of Ions in Matter, Vol. 3, Hydrogen: Stopping
Powers and Ranges in All Elements. New York: Pergamon Press, 1977.
Andrade F, Orsini C, Maenhaut W. Atmos Environ 28:2307, 1994.
Andreae MO, Crutzen PJ. Science 276:1052, 1997.
Annegarn HJ, Flanz M, Kenntner T, Kneen MA, Helas G, Piketh SJ. Nucl Instrum Methods
B109=110:548, 1996.
Aoki T, Katayama Y, Kagawa A, Koh S, Yoshida K. Nucl Instrum Methods B136–138:919, 1998.
Arai N, Sakamoto W. Int J PIXE 3:275, 1993.
Arai N, Sakamoto W, Maeda K. Int J PIXE 4:123, 1994.
Arai N, Takai N, Sakamoto W, Yoshida K, Maeda K. Int J PIXE 5:153, 1995.
Araujo F, Pinheiro T, Alves LC, Valerio P, Gaspar F, Alves J. Nucl Instrum Methods B136–
138:1005, 1998.
Artaxo P, Hansson H-C. Atmos Environ 29:393, 1995.
Artaxo P, Gerab F, Yamasoe MA, Martins JV. J Geophys Res 99:22857, 1994.
Artaxo P, Rabello MLC, Maenhaut W, Van Grieken R. Tellus 44B:318, 1992.
I. INTRODUCTION
The current trend in many technology-related fields such as electronics and materials
science is toward miniaturization and increased spatial resolution well below the micro-
meter scale. The corresponding requirements for microanalytical instruments needed to
meet the demands of this trend require analytical instruments that are capable of analyzing
micrometer and submicrometer regions of samples. A schematic diagram outlining the
generic features of a microanalytical instrument is shown in Figure 1. In such an instru-
ment, the primary radiation (ions, photons, or electrons) is focused to form a beam from
about 1 mm down to about 0.5 nm in size. The interaction of the primary beam with the
specimen results in the emission of secondary radiation that is then analyzed with a
spectrometer system to provide information on the structure and composition of the
sample. For the analysis of micrometer and submicrometer domains, the electron-probe
microanalyzer (EPMA), scanning electron microscope (SEM), and analytical electron
microscope (AEM) use electron beams as their primary radiation source and incorporate
electron-induced x-ray emission as one of the primary spectroscopies for obtaining ana-
lytical information.
In electron beam instruments, the electron beam is generated from a tungsten fila-
ment, a lanthanum hexaboride (LaB6) electron source, or a field-emission electron source.
The emitter serves as the cathode in the electron gun and is maintained at a negative
potential with respect to ground as shown in Figure 2 for a tungsten-filament source. In
systems employing a tungsten filament, the source of choice for quantitative electron probe
analysis due to its stability, the electron beam is produced by the thermal emission of
electrons from a ‘‘hairpin’’ filament consisting of a tungsten wire bent in a V shape. The tip
of the V is approximately 200 mm in diameter, as shown in Figure 3. The tungsten wire is
directly heated to a temperature of 2700–2900 K, resulting in an electron emission current
density of about 10 A=cm2.
Note: Certain commercial equipment, instruments, or materials are identified in this report to specify
adequately the experimental procedure. Such identification does not imply recommendation or
endorsement by the National Institute of Standards and Technology, nor does it imply that the
materials or equipment identified are necessarily the best available for the purpose.
Figure 2 Schematic of a W-filament electron gun. (From Goldstein et al., 1975a. Reproduced with
permission of Plenum Press.)
In systems that employ the LaB6 electron source, the cathode consists of a rod of
solid LaB6 that is milled on one end to a tip with a diameter of approximately 10 mm. The
LaB6 source is heated to a temperature of about 1950 K. At this temperature, the emission
current density for the LaB6 source is on the order of 100 A=cm2, a factor of 10 greater
than the current density for the tungsten filament. The higher current density of the LaB6
enables the use of smaller primary beams for a given current than the tungsten filament. In
the past, the LaB6 emission source has found only limited use in EPMA, primarily because
of instability in the emission current. Recent advances in electronics and gun design have
increased the stability of the electron emission from the LaB6 source to a level that allows
the analyst to perform relatively high-quality x-ray analysis. The LaB6 source, however,
is rapidly being replaced by field-emission sources.
In systems that employ a field-emission electron source, the emitter consists of a
single-crystal tungsten rod, the end of which has been formed into a sharp tip with a radius
less than 100 nm. In a cold field-emission source, a strong electric field is applied through a
series of extraction lenses to the tip, causing the emission of electrons without heating. In a
thermally assisted field-emission source, the tip is heated to 1800 K, increasing the beam
stability and eliminating the need to ‘‘flash’’ or clean the tip prior to each use. The main
advantage of the field-emission source is its inherent brightness on the order of 105 A=cm2,
a factor of 104 greater than the conventional tungsten wire that allows the formation of a
very small diameter probe on the order of 0.5 nm. This makes the field-emission source
particularly useful for x-ray analysis with low acceleration voltages and soft x-rays.
After emission, the electrons in an electron probe with a W or LaB6 filament are
focused through an initial crossover by the presence of the Wehnelt cap, which surrounds
the filament as shown in Figure 2. The cap is biased several hundred volts negative
compared to the filament, which creates an immersion field that focuses the electrons to a
crossover with a diameter d0 of approximately 50 mm for a conventional tungsten filament.
In a fielid-emission instrument, the crossover is done with the extraction lenses. After
passing through the initial crossover, the electrons are accelerated by an anode plate
Copyright © 2002 Marcel Dekker, Inc.
biased from 1 to 400 kV positive with respect to the cathode. Next, the initial electron
crossover or spot is demagnified by a series of apertures and electron optical lenses, in-
cluding both condenser and objective lenses. The final probe diameter used to interrogate
the sample is dependent on filament material, accelerating voltage, and emission current.
Figure 4 shows a plot of probe diameter versus probe current for different filaments at
30 kV acceleration potential. As the acceleration potential decreases, the probe diameter
increases as shown in Figure 5 (which shows a plot of probe diameter versus probe current
for 1kV acceleration potential). A schematic of an electron-probe microanalyzer is shown
in Figure 6. An excellent discussion of electron sources, optics, and electron probe for-
mation is given by Goldstein et al. (1992b).
Castaing developed the first successful EPMA and outlined the fundamental physical
concepts of quantitative analysis (Castaing, 1951). The electron microprobe that he de-
veloped made use of a focused beam of electrons to excite x-rays from a microscopic
domain on a sample surface.
In classical electron-probe microanalysis, the acceleration potential is on the order of
10–30 kV and the samples are polished flat, homogeneous, and opaque with respect to the
electron beam. The characteristic x-rays emitted as a result of the primary electron beam
interaction with the atoms in the specimen are analyzed by either wavelength-dispersive
spectrometry (WDS) or energy-dispersive spectrometry (EDS) to determine elemental
compositions. The volume of the specimen that is excited depends on the specimen
composition and the energy of the primary electron beam. Because the absorption path
lengths for x-rays are considerably greater than those for secondary electrons used for
electron imaging, the spatial resolution for classical x-ray microanalysis is 1–2 mm, com-
pared to 3–5 nm for electron imaging. This is shown schematically in Figure 7, in which the
x-ray emission volume is compared to the electron beam diameter of 1–5 nm. Specimens
that are inhomogeneous at dimensions below the x-ray resolution cannot be readily
analyzed by conventional microprobe analysis. These samples are best analyzed with low
Figure 4 Probe diameter versus probe current for different electron gun designs, plotted for a
30-keV accelerating potential. (From Goldstein et al., 1992a. Reproduced with permission of Plenum
Press.)
acceleration voltages and soft x-rays, or in the analytical electron microscope at high
voltages, 100 kV or more, as thin specimens.
where C refers to the elemental concentrations, expressed as weight fractions, and the
subscript j refers to all of the elements in the sample.
Equation (2) only applies to a system in which the sample and the standard are
identical and have been measured under identical experimental conditions. In practice,
as the similarities between the sample and the standard decrease, Eq. (2), even as an ap-
proximation, fails and a series of corrections must be applied to the k ratio to obtain an
Copyright © 2002 Marcel Dekker, Inc.
accurate quantitative analysis. The corrections that must be applied to the k ratio include
the following:
1. The atomic number correction for the difference between the electron scattering
and penetration in the sample and the standard
2. The absorption correction for the difference in the absorption of the x-rays as
they pass through the sample or standard before reaching the detector
3. The fluorescence correction for the fluorescence of x-rays induced by
the absorption of characteristic and continuum x-rays that are generated in
the sample by the primary electron beam and which propagate out from this
source
These corrections are applied to the various k ratios as part of theoretical or empirical
correction procedures in order to obtain quantitative results.
where the R and the S terms refer to the electron backscattering and the electron stopping
power, respectively, and Q is the ionization cross section. The limits on the integral are
from the incident electron energy, E0, to the critical excitation energy, Ec for the x-ray line
of interest.
Copyright © 2002 Marcel Dekker, Inc.
Electron stopping power. The electron stopping power S is defined in Eq. (6) as the
energy lost per unit electron path length in material of density r (Widdington, 1912):
1 dE
S¼ ð6Þ
r dX
One of the most commonly used terms for S is the approximation by Bethe (1930) and
Bethe and Ashkin (1953), which assumes a continuous function for the electron energy
loss:
Zi 1 1:166E
S ¼ 78;500 ln ð7Þ
Ai E Ji
where S is in keV cm2=g and E is in keV.
The value for the mean ionization potential, J, in Eq. (7) is not directly measured and
several different expressions have been used in the literature for the calculation of J.
Various literature values for J are listed in Table 1 and are plotted as a function of Z in
Figure 8. The various expressions for J all yield similar results for elements above Z ¼ 10
with relatively large discrepancies among expressions for Z < 10. The Berger–Seltzer ex-
pression for J is one of the most widely used in quantitative analysis procedures (Berger
and Seltzer 1964).
There are also several models in the literature that have been used to calculate Q,
all of which have the general form described by Bethe (1930):
ln U
Q¼C ð8Þ
UEc2
where U is the overvoltage defined as E0 =Ec .
Heinrich and Yakowitz (1970), however, have shown that the difference in the
various models resulted, in negligible changes in the final elemental concentrations for
elements with Z > 10.
Duncumb and Reed (1968) simplified the integration in Eq. (5), eliminating the need
for the numerical integration Eq. (5) and the evaluation of Q. In their procedure, they
assumed that the values for (1=S) dE were constant for the sample and standard over the
electron range used in electron-probe analysis and, therefore, could be removed from the
integral. As a result of this assumption, the integration of Q is unnecessary because it
appears in both the numerator and denominator and is the same for the unknown and the
standard. The stopping power factor can then be expressed as
Z 1:166E
S¼ ln ð9Þ
A J
where A is atomic weight, and J, from Duncumb et al. (1969), equals
J 75:5
¼ 14½1 expð0:1ZÞ þ Z=7:5 Zð100 þ ZÞ ð10Þ
Z Z
Equation (9) is one of the most commonly used formulations for classical ZAF micro-
analysis. Duncumb and Reed (1968) have shown from experimental work that the stop-
ping power for a multielement specimen can be expressed as a weighted sum of the
stopping power factor for each element:
X
j
Si ¼ Cj Sij ð11Þ
0
X1
ðln Y ÞF
LiðY Þ ¼ ln j ln Y j þ þD ð13Þ
F¼1
FðF !Þ
Similarly, Pouchou and Pichoir (1991), to avoid problems with the limiting behavior as E
approaches J in the Bethe expression, defined S as
2 3
X
1 4 Ci Zi 5 1
S¼ ð15Þ
J allðiÞ Ai f ðV Þ
where V ¼ E=J, J is same as Philibert and Tixier (1968), and Ji is from Zeller (Table 1).
Pk¼3
f ðV Þ ¼ k¼1 Dk Vpk with D1 ¼ 6:6 106 , D2 ¼ 1:12 105 ½ð1:35--0:45ÞJ , D3 ¼ 2:2
2
where the integration limit for Z is w0 ¼ Ec =E0 because electrons with energies less than Ec
cannot excite the x-rays of interest.
Similarly, It can be calculated from Eq. (21) with integration limits of E0 to Ec :
ZEc
Q
It ¼ dE ð21Þ
dE=drs
E0
Finally, substituting Eqs. (20) and (21) into Eq. (16) results in the following formulation
for R:
0 11
Z1 ZEc ZEc
dZ Q B Q C
R¼1 dE dw@ dEA ð22Þ
dw dE=drs dE=drs
w0 E E0
Several tabulations of R have been made for pure elements as a function of Z and U
(Duncumb and Reed, 1968); Green, 1963; Springer, 1966. Duncumb and Reed produced a
table of R values for various elements and several different overvoltage values. Their
values were determined indirectly from Bishop’s (1966) measurements of the energy dis-
tributions of backscattered electrons and were in agreement with the direct measurements
of R made by Derian and Castaing (1966). Figure 9 shows a plot of the Duncumb–Reed R
values versus Z at different overvoltages.
Duncumb derived an algebraic expression for R in terms of wq and Z from his
calculated values; Eq. (23) is still used in many analytical procedures (Heinrich, 1981c):
b. Absorption Correction, kA
The primary electron beam generates x-rays at varying depths within the sample. As a result,
the x-rays must pass through that portion of the specimen that lies between the x-ray gen-
eration point and the detector before they escape the sample and are measured. As shown
Figure 11, the distance A–B is referred to as the absorption path length because a percentage
of the generated x-rays undergo photoelectric absorption, interacting with specimen atoms
prior to escape. The effect of this absorption is an attenuation of the generated x-ray in-
tensity that is measured by the detector. Note that inelastic scattering of the x-rays is not a
significant process over the path lengths involved. The energy of the unabsorbed char-
acteristic x-ray is not modified during its passage through the specimen. The magnitude of
the attenuation is dependent on the composition of the specimen and a correction must be
considered when the sample and the standards used for the analysis are dissmilar.
Castaing (1951) described the characteristic x-ray intensity (without absorption)
generated in a layer of thickness dz, at a depth z below the specimen surface, in a sample of
density r as
dI ¼ fðrzÞ dðrzÞ ð28Þ
15 keV 30 keV
The introduction of x-ray absorption into Eq. (29) results in the following expression for
the x-ray intensity after absorption, I 0 :
Z1
0
I ¼ fðrzÞ exp½ðm=rðrzÞ cscðCÞÞ dðrzÞ ð30Þ
0
where ðrzÞ cscðCÞ is the absorption path length for the x-rays in the specimen, C is the
detector emergence angle, and m=r is the mass absorption coefficient of the specimen for
the characteristic line of the element of interest.
The absorption term for EPMA is referred to as fðwÞ or fp , where w ¼ ðm=rÞ cscðCÞ
and fðwÞ ¼ I 0 =I. From Eqs. (29) and (30), fðwÞ can be expressed in terms relating to the
specimen as
8 1 9
<Z = Z1
m
fðwÞ ¼ fðrzÞ exp ðrzÞ csc ðCÞ dðrzÞ I 1 ¼ fðrzÞ dðrzÞ ð31Þ
: r ;
0 0
2:39 105
s¼ ð36Þ
0 Ec
E1:5 1:5
Heinrich (1970) fit experimental values of fðwÞ from Green (1962) and proposed the
following formulation for s:
4:5 105
s¼ ð37Þ
E1:65
0 E1:65
c
The most common form of the absorption term found in the various analytical schemes is
Eq. (35) with h ¼ 1:2A=Z2 and Heinrich’s s. This form of Eq. (35) is often referred to as
the Philibert–Duncumb–Heinrich equation.
Heinrich et al. (1972) empirically derived a simplified absorption term that was based
on experimental data and the Philibert equation. In the formulation of this term, it was
noted that the compositional dependence in Philibert’s h term was small compared to the
scatter in the available experimental data. The resulting formulation for fðwÞ, given in
Eq. (38), is referred to by Heinrich as the quadratic model. In this equation, the
compositional dependence in the h term has been eliminated, making the model in-
dependent of the atomic weight and atomic number of the target:
1 2
¼ 1 þ 1:2 106 gw ð38Þ
fðwÞ
where g is the quantity E1:65
0 E1:65
c .
In Eq. (35), h is dependent on target composition and must be averaged for the
various elements in multielement targets. A compositionally weighted average for h,
Eq. (39), is generally employed in analytical procedures:
X
ht ¼ Cj hj ð39Þ
j
In addition, the mass absorption coefficient, m=r, for the characteristic line of element i
in a multielement target is the weighted sum over all elements in the target including i:
sam X j
m m
¼ Cj ð40Þ
r i j
r i
where ðm=rÞ Ij is the mass absorption coefficient for the line of element i in element j and Cj
is the concentration of element j.
The calculation of x-ray absorption from Eq. (35) is most accurate when the value of
fðwÞ is greater than 0.7. Toward this end, it is important in setting up an experiment to
minimize the degree of x-ray absorption by the following:
Copyright © 2002 Marcel Dekker, Inc.
1. Minimizing the absorption path length by selecting a low overvoltage E0 =Ec
2. Selecting high-energy x-ray lines that have low-absorption cross sections. It is
important to note that selecting a high-energy x-ray line requires a high
acceleration potential and will, therefore, increase the absorption for elements in
the sample that have relatively low-energy analytical lines
3. Measuring the intensities at the highest possible x-ray emergence angle (which
typically is fixed for any given instrument)
A study was done by Small et al. (1991) comparing experimental measurements of fðwÞ
with four different absorption models. The experimental data were collected for Si at both
20 and 30 kV and for Fe and S in FeS2 at 20 kV acceleration potential. The models studied
included the following:
1. Heinrich’s quadratic model, Eq. (38) (Heinrich, 1972)
2. Heinrich’s duplex model, Eq. (41) (Heinrich, personal communication, 1987)
3. Philibert’s model, Eq. (35) (Philibert, 1963)
4. Pouchou and Pichoir’s model, Eq. (42) (Pouchou and Pichoir 1991a)
Duplex model:
ð1 þ 1:65 106 gwÞ2
fðwÞ ¼ ð41Þ
1 þ að1:65 106 gwÞ
p ffiffiffi
ffi
where a ¼ 0:18 þ 2=g þ ð8 109 ÞEc þ 0:005 Z with Ec in electron volts.
Puchou and Pichior model:
F1 ðwÞ þ F2 ðwÞ
fðwÞ ¼ ð42Þ
F
where the formulation for F, F1, and F2 are given in Pouchou and Pichoir (1991a).
A comparison of the different models and the experimental data is shown in Figure 13.
The results of the study indicated that no one model is best for all targets and conditions.
The analyst interested in the highest accuracy should select the absorption correction that
works the best for the particular type of sample and experimental conditions.
c. Fluorescence Correction, kF
The characteristic fluorescence correction is necessary when the analysis involves the
following conditions:
1. The characteristic x-ray peak from element j at energy Ej is greater than the
critical excitation potential Ec;i for element i.
2. The energy difference ðEj Ec;i Þ < 5 keV.
When these conditions exist, the characteristic line of element j will excite the characteristic
line of element i in the specimen as shown in Figure 14. This fluorescence results in an
increased intensity for element i that must be taken into account in obtaining a quanti-
tative analysis. In this case, the excitation is caused by x-rays and not electrons; hence, the
generation of the secondary x-rays will originate from a much larger volume of the sample
that if it were generated directly by electrons. In addition, the measured intensity of the
fluorescence radiation is proportional to the x-ray emergence angle.
The basic formulation of the fluorescence correction is given by
! !1
X If; j X If; j
Fi ¼ 1 þ 1þ ð43Þ
j
Ii j
Ii
std sam
The factor If =Ii is the ratio of the x-ray-excited intensity for the characteristic line of
element i to the electron-excited intensity. The summation over j is required because the
total correction must be summed over all the elements in the specimen.
The most commonly used fluorescence correction factor was originally developed by
Reed (1965) and is given in Eq. (44) [from Reed (1997)]:
!
If mij ri 1 Ai ðUj ln Uj Uj þ 1Þ lnð1 þ mÞ lnð1 þ nÞ
¼ 0:5Cj oj þ ð44Þ
Ii mj ri Aj ðUi ln Ui Ui þ 1Þ m n
which is the most common form of the correction found in analytical procedures. Gold-
stein et al. (1981e) demonstrated the magnitude of the fluorescence correction with the
analysis of a 10 wt% Fe–90 wt% Ni alloy. The results of the analysis are summarized in
Table 3 and show that the intensity of the fluoresced iron, If =IFe ranges from 16.8% to
34.6% of the observed iron x-ray intensity. The magnitude of the correction is lower at the
smaller accelerating potential and detector takeoff angle.
Copyright © 2002 Marcel Dekker, Inc.
Table 3 Fluorescence of FeKa by Ni in 10 wt% Fe–90 wt%
Ni alloy
Several researchers, including Love (1983), Love and Scott (1988), Bastin et al.
(1986), Pouchou and Pichoir (1988), and Sewell et al. (1985a) have made experimental
measurements of x-ray depth distributions, fðrzÞ curves, in a large number of targets,
including pure elements, alloys, and oxides. In addition, other researchers [e.g., Heinrich
et al. (1988), Gauvin et al. (1995), Murata et al. (1983), and Karduck and Rehbach (1988)]
have made Monte Carlo calculations of x-ray depth distributions. The increased knowl-
edge regarding the behavior of these curves for different materials has made it possible to
empirically develop corrections based on the integration of fðrzÞ curves. These procedures
are particularly attractive for the analysis of low-energy x-ray lines from the elements
boron through fluorine, where fðwÞ is much less than 0.7 and the accuracy of the ZAF
method is low. In these procedures, fðrzÞ refers to the depth distribution of x-rays in a
bulk sample normalized to the x-ray intensity produced in an infinitely thin, unsupported
film of the same composition. In principle, the fðrzÞ curve as described by Castaing (1951)
contains the information necessary to correct experimental data for both the atomic-
number correction, Z (integrated area under curve), and absorption correction, A (shape
of the curve). The Gaussian model proposed by Packwood and Brown (1981) combines
the Z and A corrections in their model for fðrzÞ producing a single emitted intensity. It
should be noted, as pointed out by Armstrong (1988a), that the use of fðrzÞ expressions
Copyright © 2002 Marcel Dekker, Inc.
for the correction of x-ray absorption requires only that a given expression produce the
correct shape for the fðrzÞ curve. The use of the fðrzÞ expression for the atomic-number
correction is risky because this requires ‘‘that the thickness of the tracer films and the
normalizing thin films in the fðrzÞ experiments for different matricies be known to a high
degree of accuracy’’ (Armstrong 1988a).
In practice, several of the fðrzÞ methods use of the shape of the fðrzÞ curve for the
absorption correction and combine that with separate Z and F corrections. Independent of
the procedure used, the analysis of specimens using the fðrzÞ correction method is de-
pendent on the derivation of an accurate expression that describes the experimental fðrzÞ
curves and the degree to which that expression can be universally applied to systems with
unknown fðrzÞ curves.
The formulas for the different terms can be found in Packwood (1991).
Figure 16 Modified Gaussian function for fitting f(rz) curves. (From Packwood and Brown,
1981. Copyright John Wiley & Sons Limited. Reproduced with permission.)
Bastin. Bastin and Heijligers (1991) modified the Gaussian model of Packwood and
Brown combining the a and g terms with a b term based on the PAP F term or area
parameter. In Bastin’s PROZA program (Bastin et al., 1984), the b term is adjusted, for
given a and g, so that the integral under the fðrzÞ curve is equal to the PAP area para-
meter F, and the peak of the fðrzÞ curve is optimized with respect to position and height.
The a term is a function of the beam energy E0 , the critical excitation energy Ec , the
overvoltage for the x-ray line of interest U0 , the atomic number Z, atomic weight A, and
ionization potential J of the matrix. The g term is determined from one of two equations,
depending on whether U0 is greater than 6. Both equations are functions of U0 and Z. The
b term is derived from a; g and F and is divided into nine different segments based on the
value of Rðb=2aÞ. The equation f0 is taken from the PAP model. The various equations
for the parameterization of a; b, and g are given in Bastin and Heijligers (1991).
In addition to PROZA, Bastin et al. (1998) have developed a new parameterization
based on the double partial Gaussian model of Merlet (1994), called PROZA96. The
formulation of the two Gaussian functions are as follows:
For 0 rz rzm
Z and A are the atom-concentration weighted atomic number and atomic weight,
respectively. The surface ionization function, fð0Þ, can be determined from the expression
of either Reuter (1972) or Love and Scott (1978).
Love and Scott. Love et al. (1984) proposed a fðrzÞ method in which they in-
troduced a separate atomic-number correction. With the separate treatment of the atomic-
number correction, the fðrzÞ curve is used only for target absorption. As mentioned
earlier, target absorption requires only that the shape of the curve be correct and not the
absolute height. The authors proposed a quadrilateral profile, Figure 19, to fit the shape of
the curve. The quadrilateral is defined by the y intercept fð0Þ, the position and amplitude
of the peak, rzm ; fðmÞ, and the x-ray range, rzr . From the quadrilateral model, the
analytical expression for fðwÞ can be written as
where
To constrain the quadrilateral model when fðwÞ values exceed 0.5, the authors expressed
both rzm and rzr in terms of the mean depth of x-ray generation, rz (Scott and Love,
1991). In conjunction with the quadrilateral fit to the fðrzÞ curve for target absorption,
the atomic-number correction combines the stopping power factor from Eq. (14) with the
backscatter factor R, Eq. (58), from Love et al. (1978):
1 R 0:6
¼ IðU0 Þ þ ZGðU0 Þ ð58Þ
Z
where U0 is the overvoltage, Z is the backscatter coefficient,
1 0:25
IðU0 Þ ¼ 0:3 þ expð1:5 1:5U0 Þ
U0
and
GðU0 Þ ¼ ð0:368 0:075 ln U0 Þ expð1 2:3U04 Þ
A full description of the quadrilateral model can be found in Scott and Love (1991).
Figure 22 Results from the analysis of 826 binary compounds with several different quantitative
analysis procedures. (Adapted from Pouchou and Pichoir, 1991b.)
Figure 24 Result for 242 analyses involving samples where x-ray absorption is the predominate
correction. (Adapted from Pouchou and Pichoir, 1991b.)
concentration, expressed in Eq. (2), can be expanded to include real samples by the in-
troduction of an efficiency factor a for element a in a sample:
aa Ca
ka ¼
Sai Ci
where the summation is over all elements in the sample.
Ziebold and Ogilvie, using Eq. (59), expressed the relationship between C and k for a
binary system containing elements i and j as
Ci
¼ a þ ð1 aÞCi ð60Þ
ki
or
ki 1 1
¼ þ 1 ki ð61Þ
Ci a a
where a for the binary is referred to as the a factor and equals aj =ai .
Element No. of measurements Mean k 0=k Root mean square deviation (%)
Figure 26 k=C versus k for Ag–Au alloys at various electron beam energies. (From Goldstein
et al., 1981f. Reproduced with permission of Plenum Press.)
approach has been developed. In this procedure, the analyst calculates the necessary a
factors by assuming an appropriate composition and then running a ZAF, fðrzÞ, or
Monte Carlo procedure to back-calculate the corresponding k ratios. The a factors can
then be determined from Eq. (60). It is important that the analyst realize that the total
uncertainty in this combined approach includes the uncertainties of both the empirical and
the ZAF methods. Albee and Ray (1970) have used this procedure to determine the a
factors for 36 elements relative to simple oxides. Laguitton et al. (1975) and Bence and
Holzwarth (1977) proposed additional corrections along with more elaborate polynomial
representations of the a factors. Armstrong (1988b) calculated a large number of oxide a
factors using a combination of ZAF, fðrzÞ, and Monte Carlo methods combined with a
second-order polynomial for the a factors, Eq. (66):
2
Ci Ci
ai; j ¼ c þ d
i
þe ð66Þ
Ci þ Cj Ci þ Cj
where Ci and Cj are the oxide weight fractions of elements i and j in the binary and c; d,
and e are the coefficients of the polynomial fit. The results indicated that the C=k values
Copyright © 2002 Marcel Dekker, Inc.
from the second-order polynomial a factor were essentially identical to the ZAF, fðrzÞ, or
Monte Carlo method they were based on and the results can be expected to have the same
accuracy and precision as the conventional procedures.
The empirical method of analysis is most accurate when the difference between the
sample and standards is mainly the result of x-ray absorption. The procedure is less ac-
curate when there are large differences in atomic number and is least accurate when there
is a significant contribution from fluorescence. Given these limitations, the empirical
procedure works best for the analysis of materials such as geologic specimens and oxide
systems that have relatively low mean atomic numbers and minimal secondary fluores-
cence.
where the terms in brackets represent the excitation function: r is the density, N0 is
Avogadro’s number, A is the atomic weight, o is the fluorescence yield, Q is the ionization
cross section, dE=ds is the stopping power, E is the electron energy, E0 is the incident beam
energy, and Ec is the critical excitation energy. The other terms correct for the loss of x-ray
production due to electron backscattering (R), the self-absorption of x-rays propagating
through the solid ½ fðwÞ, and the efficiency of the detector, e. It is useful to consider the
confidence with which each of these terms can be calculated.
a. Excitation
Three terms are critical in the excitation function: the ionization cross section, the fluor-
escence yield, and the stopping power.
1. Ionization cross section: Several parameterizations of the K shell ionization cross
section are plotted in Figure 28. The variation among these choices exceeds 25%.
Although we cannot say that any of these is correct, it is certain that they cannot
all be correct. Moreover, because of the continuous energy loss in a solid target,
the cross section must be integrated from E0 to Ec , through the peak in Q and
the rapid decrease to U ¼ 1. This region of the cross section is poorly
characterized, so that it is difficult to choose among the cross-section
formulations based on experimental measurements. The situation for L and M
shell cross sections is even more unsatisfactory.
2. Fluorescence yield: Various experimental determinations of the K shell
fluorescence yield are plotted in Figure 29. Again, a variation of more than
25% exists for many elements. The situation for L and M shell transitions is
substantially more incomplete.
3. Stopping power: The classic Bethe formulation of the stopping power becomes
inaccurate at low beam energies (< 5 keV), and eventually with decreasing
energy, it becomes physically unrealistic with a sign change. The accuracy of the
stopping power matters for calculating Eq. (68) because the cross section must
be integrated to Ec , which for low-energy x-rays (e.g., C, N, O, F) involve
electron energies in this regime. As discussed earlier in this chapter, several
authors have suggested modifications to the Bethe formulation to correct for the
Copyright © 2002 Marcel Dekker, Inc.
Figure 28 Ionization cross section as a function of overvoltage U, as formulated by different
authors.
b. Backscatter Loss
The backscatter-loss correction factor R was initially formulated based on experimental
measurements of the total backscatter coefficient and the differential backscatter coeffi-
cient energy. Although careful, extensive measurements of total backscatter were available
in the literature, the database of the differential backscatter coefficient as a function of
energy, a much more difficult experimental measurement, was limited to a few elements
and was available at only one emergence angle. The development of advanced Monte
Carlo simulations has permitted the rigorous calculation of R over all scattering angles
and energy losses so that this factor is probably known to an accuracy within a few percent
across the periodic table and the energy range of interest.
c. X-ray Self-absorption
The self-absorption of x-rays in the hypothetical standard is calculated with the for-
mulation of the absorption factor used in matrix corrections. Fortunately, the absorption
correction is generally small for the x-rays of a pure element, so that at least for higher-
energy characteristic x-rays (e.g., greater than 3 keV), there is little uncertainty in this
factor. However, the self-absorption increases both as photon energy decreases and in-
Copyright © 2002 Marcel Dekker, Inc.
Figure 29 K Shell fluorescence yield as a function of atomic number.
cident electron energy increases, so that the error in calculating the intensity emitted from
an element emitting low-energy photons (e.g., carbon) could be significant.
d. Detector Efficiency
The last term in Eq. (69) is one of the most difficult with which to deal:
m m m
e ¼ exp rwin twin þ rAu tAu þ r tSiDL
r win r Au r SiDL Si
m m m
þ rcon tcon þ rice tice 1 exp rSi tSi ð69Þ
r con r ice r Si
In the traditional k-value approach, the detector efficiency cancels quantitatively in the
intensity ratio because the same x-ray peak is measured for the unknown and the standard
under identical (or at least accurately reproducible) spectrometer conditions. When
standardless analysis is performed, this cancellation cannot occur because x-ray peaks of
different energies are effectively being compared and, therefore, accurate knowledge of the
detector efficiency becomes critical. Detector efficiency is mainly controlled by absorption
losses in the window(s) and detector structure. The expression for detector efficiency, Eq.
(69), consists of a multiplicative series of absorption terms for each component: detector
window (win), gold surface electrode (Au), semiconductor ‘‘dead layer’’ (DL, actually a
partially active layer below the electrode and the source of incomplete charge phenomena),
and a transmission term for the detector thickness. Additionally, for most practical
measurement situations, there may be absorption contributions from contamination on
Copyright © 2002 Marcel Dekker, Inc.
the detector crystal, usually arising from ice buildup due to pinhole leaks in the window or
support (ice) and from contamination on the detector window usually deposited as hy-
drocarbons from the microscope environment (con).
An example of the detector efficiency as a function of photon energy for several
window materials is shown in Figure 30. The choice of the window material has a strong
effect on the detector efficiency for photon energies below 3 keV, and accurate knowledge
of the window and detector parameters is vital for accurate interelement efficiency cor-
rection across the working range of the detector, typically 100 eV to 12 keV. The change in
the detector efficiency with the accumulation of ice is illustrated in Figure 31. The buildup
of ice and other contaminants and the resulting loss in efficiency is referred to as ‘‘detector
aging.’’ Detector aging can result in a significant loss of the low-energy photons (< 3 keV)
relative to the higher-energy photons (3–12 keV) (Fiori and Newbury, 1978).
Figure 30 Detector efficiency for various windows (Al window coating ¼ 0.02 mm; Au electro-
de ¼ 0.01 mm; Si dead layer ¼ 0.03 mm): (a) beam energies 0–5 keV; (b) beam energies 0–1 keV.
elements measured at 20 keV is shown in Figure 32. Missing elements can be calculated by
simply fitting the available peaks and interpolating (e.g., in Figure 32, the intensity for
gallium could be estimated by fitting the smoothly varying data and interpolating). From
the smooth change in peak height with atomic number seen in Figure 32, such an inter-
polation should be possible with reasonable accuracy. The situation is not as satisfactory in
the L and M families, as illustrated in Figures 33 and 34, respectively, because the fluor-
escence yield is a much more complicated function of atomic number.
If the analysis must be perfomed at a beam energy other than that of the spectral
database, then Eq. (68) must be used to shift the intensities appropriately. Similarly,
if a different EDS detector is used, the detector efficiency must be corrected using Eq. (69).
The ‘‘fitted standards’’ standardless procedure is expected to be more accurate than the
first-principles standardless procedure because it is tied to actual experimental measure-
ments which directly incorporate the effects of the cross section and the fluorescence yield,
at least over the range of the elements actually measured.
composition was available from independent chemical analysis and for which micro-
homogeneity could be established. Compositions were carefully chosen to avoid serious
spectral overlaps (e.g., PbS, MoS2). Light elements such as B, C, N, O, and F were also
eliminated from consideration because of large errors due to uncertainties in mass ab-
sorption coefficients. In oxidized systems, the oxygen was calculated by means of assumed
stoichiometry, but the resulting oxygen values were not included in the error histograms
because of their dependence on the cation determinations.
Figure 35 shows an error histogram for the ‘‘first-principles’’ standardless analysis
procedure embedded in the National Institute of Standards and Technology–National
Institutes of Health Desktop Spectrum Analyzer (DTSA) x-ray spectrometry software
engine (Fiori et al., 1992). The error distribution shows symmetry around 0% error, but in
comparing this distribution with that for the conventional standards=ZAF procedure
shown in Figure 21, the striking fact is that the error bins are 10 times wider for the ‘‘first-
principles’’ standardless analysis procedure. Thus, the 95% error range is approximately
±50% relative rather than ±5% relative.
The error distribution for a commercial standardless procedure based on the ‘‘fitted
standards’’ approach is shown in Figure 36. This distribution is narrower than the first-
principles standardless approach, but the error bins are still five times wider than those of
the conventional standards=ZAF procedure, so that the 95% error range is ±25%
compared to ±5%. It must be emphasized that this distribution represents a test of only
one of the many implementations of standardless analysis in commercial software systems
and more extensive testing is needed.
Copyright © 2002 Marcel Dekker, Inc.
Figure 32 K-Family peaks from transition elements; E0 ¼ 20 keV.
Figure 35 Error distribution for the ‘‘first-principles’’ standardless analysis procedure embedded
in the NIST–NIH Desktop Spectrum Analyzer x-ray spectrometry software system.
values, which may imply far more apparent accuracy than is justified and which may lead
to a loss of confidence in quantitative electron-probe microanalysis when an independent
test is conducted.
where k depends on matrix parameters [atomic number (Z), atomic weight (A), and
density (r)], E0 is the incident beam energy, Ec is the critical excitation energy (‘‘critical
ionization potential’’), and the exponent n is in the range 1.5–1.7. The ‘‘conventional’’
energy range for quantitative electron beam x-ray microanalysis can be thought of as
beginning at 10 keV and extending to the upper limit of the accelerating potential, typically
30–50 keV, depending on the instrument. The lower limit of 10 keV for the conventional
operating range is selected because this is the lowest incident beam energy for which there
is at least one satisfactory analytical x-ray peak excited from the K, L, or M shells for every
element in the periodic table that is accessible to x-ray spectrometry (Goldstein et al.,
1992c). Only H, He, and Li are excluded due to a lack of x-ray emission or, in the case of
Li, because the photon energy is so low (54 eV) that it is completely reabsorbed in the
Copyright © 2002 Marcel Dekker, Inc.
target or the components of the x-ray spectrometer. Electron-excited x-ray microanalysis
thus extends from Be (EK ¼ 0.116 keV) to the transuranic elements. This elemental range
is based on establishing the minimum acceptable overvoltage as U ¼ E0 =Ec
1:25. At
E0 ¼ 10 keV, this overvoltage criterion involves the use of K lines for 4 (Be) Z 27 (Co)
L lines for 28 (Ni) Z 67 (Ho), and M lines for 68 (Er) Z to the transuranics, as il-
lustrated by the periodic table in Figure 37, shaded according to the appropriate shell for
analysis with E0 ¼ 10 keV and U
1.25. This broad elemental coverage of electron beam
x-ray microanalysis is one of its most important features.
No strict definition exists as to what energy value constitutes the beginning of the
low-beam-energy microanalysis regime, but from the previous discussion, the low-energy
regime can be considered those beam energies below 10 keV, and especially below 5 keV.
Figure 38 shows the excitation range for several characteristic x-rays in a silicon matrix as
a function of incident beam energy from 0 to 10 keV, as calculated with the Kanaya–
Okayama formulation of the excitation range (Kanaya and Okayama, 1972):
0:0276A 1:67
RðmmÞ ¼ E0 E1:67
c ð71Þ
Z0:89 r
Near E0 ¼ 10 keV, the excitation range is substantially greater than 1 mm for the various
elemental lines plotted, whereas at 5 keV and below, the range for these x-rays is below
500 nm. The excitation range as a function of overvoltage is shown in Figure 39. The range
is less than 150 nm for AlK in Si for an overvoltage of 2 and diminishes sharply as the
overvoltage is reduced, so that for an overvoltage of 1.25, the excitation range is
approximately 20–30 nm.
Figure 37 Selection of x-ray peaks available for analysis with E0 ¼ 10 keV for conventional EDS.
Note that elements in black squares either do not emit x-rays or else are too low in energy for prac-
tical detection.
3:29a
CMDL ¼ ð72Þ
½ntPðP=BÞ1=2
where ‘a’ is the ziebold–Ogilvie factor in the expression relating concentration to measured
intensity ratio [unknown sample to pure-element standard (Ziebold and Ogilvie, 1964); ‘a’
can be taken as approximately unity for general estimation purposes], n is the number of
measurements, t is the integration time per measurement, P is the peak counting rate, and B
is the background counting rate. As the overvoltage is lowered, both the peak counting rate
and the peak-to-background decrease. Figure 40a shows the experimentally measured
behavior of P=B and CMDL as a function of overvoltage for SiKa from pure silicon.
"
Figure 40 (a) Experimental measurement of the P=B for copper as a function of overvoltage and
calculation of the concentration limit of detection for SiKa in a matrix of similar atomic number
(e.g., Al). Detector: Si EDS (150 keV); 100 s integration. The vertical arrow designates the ‘‘low-
keV SEM’’ regime. (b) Calculated limit of detection (mass fraction), derived from experimental mea-
surements of P=B and peak counting rate, as a function of overvoltage, U. Si EDS (150 eV FWHM)
operated at 30–40% dead time. Three different accumulation times (nt) of 100 s, 200 s, and 1000 s are
plotted. The vertical arrow designates the ‘‘low-keV SEM’’ regime.
C K 0.284 0.00198
Na K 1.080 0.0192
Mg K 1.303 0.0265
Si K 1.848 0.0603
Zn L 1.022 0.00736
Ga L 1.117 0.00875
Ge L 1.217 0.0103
Sm M 1.080 0.00133
Eu M 1.130 0.00137
Figure 46 Selection of x-ray peaks available for analysis with E0 ¼ 2.5 keV for conventional EDS.
Note that elements in black squares are not sufficiently excited to be practically measured.
thickness is a small fraction of the excitation range, so the effect on quantitation is neg-
ligible. However, at low beam energy, the total sampling depth becomes so shallow that a
surface layer constitutes a significant fraction of the analyzed depth. The specimen should
thus be considered as a case of ‘‘special geometry,’’ that of an overlayer of C (or oxide,
etc.) on a substrate that is actually the specimen composition of interest.
Rigorous testing for the accuracy of low-beam-energy microanalysis is similarly
affected by the issue of proper test specimens. The situation is difficult enough at high
beam energy, where the lateral homogeneity is the key issue. At low beam energy, the
measurement becomes profoundly sensitive to the surface region of both the specimen and
the standards, especially if these are different. Thus, any testing of quantitation procedures
can only sensibly proceed when a suitable series of test specimens and standards has been
produced that do not suffer from limitations imposed by the quality of the surface.
Figure 49 Microcalorimeter EDS spectrum of carbon; logarithmic scale. Note the carbon coinci-
dence peak, and the discontinuities in the background due to absorption in the spectrometer materi-
als (Al window thermal protection coating and AG absorber).
energy, for example, in the analysis of particles during defect review in simiconductor
processing. As an example, Figure 50 shows the spectrum obtained at 5 keV from WSi2
with a 10-eV microcalorimeter EDS. The separation of the SiKa line from the WM family
is virtually complete, thus eliminating the severe interference encountered with semi-
conductor EDS. The high P=B ratios evident in the spectra in Figures 47–50 will result in
limits of detection similar to WDS. Finally, the resolution of the microcalorimeter EDS
will make it possible to actually detect low-energy L and M lines that have low fluores-
cence yield. As an example, Figure 51a demonstrates the detection of several of the low-
yield peaks of the titanium L family and the barium M family in BaTiO3, which can be
compared to Figure 51b, obtained with semiconductor EDS. With conventional EDS, the
TiL peaks are unresolved from the OK peak, and the BaM peaks, although visible, are
actually convolved with the BaM edge discontinuities in the background.
"
Figure 51 (a) Microcalorimeter EDS spectrum of BaTiO3, low-energy region. Note detection of
BaM-family members; E0 ¼ 10 keV. (b) Si EDS spectrum of BaTiO3, low-energy region. Note detec-
tion of BaM-family members; E0 ¼ 10 keV.
ð74Þ
where fA0and fAare the x-ray depth distribution functions for the layer of interest in
the specimen and the bulk standard, respectively, w0A and wA are the respective ab-
sorption factors, and rt1 and rt2 are the depths from the surface of the top and
bottom, respectively, of the layer of interest (i.e., rt1 ¼ 0 for the topmost layer, rt2 ¼ 1
for a bulk substrate). Only in a case in which the layer of interest is very similar in
average atomic number both to its underlying (and overlying) layer(s) and to the bulk
standard will f0A and fA be similar. In most cases, f0A needs to be modified from ex-
pressions developed for bulk specimens to account for the variable composition within
the excited volume of the electron beam. Similarly, w0A and wA are seldom identical, and
wA for all underlying layers needs to be modified to account for the different degrees of
absorption in the various overlying layers. If the order, thicknesses, and compositions of
the various layers in the specimen are known, it is simple algebra to determine the
absorption factor [cf. Pichou and Pichoir (1991a), Scott et al. (1995)]. However, even if
the sequence of thickness and compositions were known, conventional parameteriza-
tions of f(rz) for bulk specimens cannot be used for layered specimens without sig-
nificant modification.
The fundamental difference between f0A ðrzÞ and fA ðrzÞ lies in the number and en-
ergy distribution of the electrons that scatter through the layer rz DðrzÞ; these, in turn,
depend on the average atomic number, distance, and thickness of the layer(s) that lie
above and below rz DðrzÞ. For example, at the top of a reasonably thick surface layer
(with respect to the electron beam depth of penetration) on a substrate of considerably
different Z, both the upward traveling and downward traveling electron components of
the f(rz) are similar for both fA0 ðrzÞ and fA ðrzÞ. At the bottom of a relatively thin surface
layer over a bulk substrate, the downward-traveling electron component of fA0 ðrzÞ is
similar to a bulk specimen of the surface film composition, whereas the upward-traveling
component is similar to a bulk specimen of the substrate composition. The situation gets
increasingly complex as the layers get thinner and=or more numerous.
A number of investigators have tried to come up with simplified expressions to ac-
count for the differences in electron scattering between layered samples and thick speci-
mens, with varying degrees of success. Hutchins (1966) tried to modify the original
Copyright © 2002 Marcel Dekker, Inc.
Castaing expression for bulk with expressions involving the initial slope of fðrzÞ curves
and the backscatter coefficient, whereas Colby (1968) modified Castaing’s expression with
functions relating to the differences in the backscatter coefficients, backsatter correction
factors, and stopping powers between film and substrate compositions [cf. Scott et al.
(1995)]. Warner and Coleman (1973) modified Hutchin’s methods to apply it to biological
materials, and Duzevic and Bonefacic (1978) modified his procedure to measure alloy film
compositions on elemental metal substrates. Oda and Nakajima (1973) modified Colby’s
procedure and showed good results in analyzing Ag–Cu alloys on Fe–Ni substrates. None
of these early attempts, however, resulted in widespread application.
As more realistic models for fðrzÞ for bulk specimens were developed, further at-
tempts were made to apply them to thin-film and layered-specimen analysis. Reuter (1972)
modified the fðrzÞ parameterization of Philibert (1963) to better fit experimentally de-
termined fðrzÞ curves and applied it to the determination of the thickness of elemental
films on single-element and multielement substrates. His fðrzÞ parameterization was a
function of the electron backscatter coefficient, Z. In order to account for the effects of
variable average atomic number in the different layers, he introduced an effective back-
scatter coefficient, Zeff , which varied between the limiting values of the backscatter coef-
ficient of a bulk specimen of the composition of the film, ZF , and the backscatter coefficient
of the bulk substrate, Zsub :
nrz nrz
Zeff ¼ Zsub þ ZF 1 ð75Þ
n0 n0
Where nrz =n0 is the ratio of the number of electrons transmitted through layer rz com-
pared to the number of electrons passing through an infinitely thin film at the surface of a
sample and is expressed by
nrz 4 104 Z 0:5 rz
¼ expðsrzÞ or 1 ð76Þ
n0 E1:7
0
whichever is larger (the Lenard coefficient s ¼ 4:5 105 =E1:65 0 ; Z is the average atomic
number of the layer, E0 is the accelerating potential in kiloelectron volts). As noted by the
author, this approximation accounts for differences between the number of backscattered
electrons from film and substrate, but not for differences in their energy distributions.
Nevertheless, the author showed good agreement between calculated and experimental
results for elemental films on substrates, both having a wide range in Z. A number of
investigators implemented various versions of this correction [e.g., Armstrong and Buseck
(1975) in their original fðrzÞ expressions for thin films and particles].
Love and Scott (1978) modified the ‘‘square’’ fðrzÞ model of Bishop (1974) for
application to thin films on substrates:
Ithin 1 expðwrtÞ
kA ¼ ¼ ð77Þ
Istd 1 expð2wrzm Þ
where Ithin and Istd are emitted intensities from a thin film and bulk standard of the same
material, t is the film thickness, and zm is mean depth of x-ray penetration which Scott et al.
(1995) proposed should be expressed by the equation of Sewell et al. (1985b):
rsm
rzm ¼ ð78Þ
½2:4 þ 0:07Z þ ð1:04 þ 0:48ZÞ= ln U0
where, in turn, rsm is the mean electron range given by
Copyright © 2002 Marcel Dekker, Inc.
7:87E 6J 0:5 E1:5 þ 7:35E 7E02
rsm ¼ P 0 ð79Þ
Ci Zi =Ai
Bushby and Scott (1993) reported results of analyses of oxide films on metallic substrates
using these corrections. The authors noted that this procedure is limited to analysis of
films on substrates of similar atomic number, as the procedure does not correct for
differences in electron scattering. This correction, however, is particularly easy to calculate
and is thus useful for ‘‘back of the envelop’’ calculations during experiment design.
The various forms of the ‘‘modified, surface-centered Gaussian’’ fðrzÞ expression
[see Eq. (46)], as originally developed by Packwood and Brown (1981), have been widely
adapted for thin-film and multilayer-specimen analysis. As described earlier, these
equations have the general form
[e.g., Armstrong (1991)]. For a multilayer specimen, a series of f(rz) functions are cal-
culated, one for each element of interest in each layer. These are constructed by keeping
track of the arrays of summed electron path lengths through each rz layer corresponding
to the different array components of energy. For details of the procedures, consult Joy
(1995) or Armstrong (1991).
One of the first groups to utilize Monte Carlo calculations for thin-film analysis was
Bolon and Lifshin (1973). Utilizing the multiple scattering Monte Carlo method of
Curgenven and Duncumb (1971), they produced calculated calibration curves of relative
intensity versus thickness for thin elemental films on elemental substrates (e.g., Au on Si)
and showed good agreement with experimental data. Kyser and Murata (1974, 1976) used
the single-scattering Monte Carlo model of Murata et al. (1971) to construct similar ca-
libration curves for both elemental and compound (alloy) films on elemental and com-
pound targets. Using the examples of MnxBiy and CoxPty films on SiO2, they were able to
show self-consistent agreement in both concentration and film thickness between mea-
sured and Monte Carlo data.
Numerous groups more recently have used Monte Carlo calculations to predict
emitted x-ray intensities from a wide variety of layered samples, including thin surface
Copyright © 2002 Marcel Dekker, Inc.
coatings, thin and thick multiple-layer structures, surface layers containing implanted
species, surface and depth diffusion profiles, and surface-layer analysis during ion sput-
tering [e.g., Amman and Karduck (1990), Amman et al. (1992), Karduck and Rehback
(1991), Armstrong (1991, 1993), Armigliato et al. (1996), Valamontes and Nassiopolos
(1996), Karduck (1997), Chan and Brown (1997a, 1997b)]. These studies have emphasized
the advantage of analyzing the same specimens at different analytical conditions (such as
electron beam accelerating potential and incidence angle) in order to elucidate their
structure. If there are no elements in common in the different analyzed layers, then each of
these measurements is an independent estimate of layer thickness and composition. If
there are elements that are present in more than one layer, then multiple measurements are
essential to deconvolve the composition and structure. Joy (1995) and co-workers and
Gauvin and co-workers (e.g., Hovington et al., 1997a,b; Drouin et al., 1997) have de-
veloped public-domain Monte Carlo programs that are easily adapted to the analysis of
layered specimens.
Table 7 Experimental Data of Reuter (1972) Processed Through GMRFilm and CASINO
Substrate: B B Be Fe Au Pt
Mn–Bi films
C–Co rt C–Co rt rt rt
Close examinations of Tables 7–9 reveals differences in the levels of agreement of the
various correction schemes, both with respect to the independent measurement data and
with respect to ‘‘internal consistency’’ (e.g., giving the same answer for data collected at
different accelerating potentials) that might be used to choose one procedure over another
as providing superior accuracy. However, due to the absence of independently analyzed
and certified reference standards as well as to the absence of comparable raw data col-
lected by more than one laboratory on these samples, no ‘‘best’’ correction scheme can be
unambiguously determined. The internal agreement of the various procedures, however,
is generally reasonably good—typically at a level of about 5% relative. This level is ap-
proximately twice what might be expected for these types of materials if they were
Copyright © 2002 Marcel Dekker, Inc.
Table 9 Experimental Data of Pouchou (1993) Processed Through GMRFilm
Layers on Si
Layer 1 Layer 2
Figure 52 Measured kaSi factors for thin-sample analysis from Wood et al. (From Wood et al.,
1984. Reproduced with permission of Blackwell Science Ltd.)
K Lines
Lorimer et al. (1976, 1977) McGill and Hubbard (1981) Wood et al. (1984)
Element (100 keV) (100 keV) (120 keV)
Na 2.46
Mg 1.23 ± 0.08 1.16 0.96 ± 0.03
Al 0.92 ± 0.08 0.8 0.86 ± 0.04
Si 0.76 ± 0.08 0.71 0.76 ± 0.004
P — — 0.77 ± 0.005
S — — 0.83 ± 0.03
K 0.79 0.77 0.86 ± 0.014
Ca 0.81 ± 0.05 0.75 0.88 ± 0.005
Ti 0.86 ± 0.05 — 0.86 ± 0.03
Cr 0.91 ± 0.05 — 0.90 ± 0.006
Mn 0.95 ± 0.05 — 1.04 ± 0.025
Co 1.05 — 0.98 ± 0.06
Ni 1.14 ± 0.05 — 1.07 ± 0.06
Cu 1.23 ± 0.05 — 1.17 ± 0.03
Zn 1.24 — 1.19 ± 0.04
Nb — — 2.14 ± 0.06
Mo 3.38 3.80 ± 0.09
Ag 6.65 9.52 ± 0.03
L Lines
(1991) and Goldstein et al. (1986). Figures 53 and 54 show a comparison of the calculated and
measured k factors for K and L line radiation from various elements (Wood et al., 1984).
Sheridan (1989) has determined a large set of experimental kaSi factors for a wide
range of Ka, La, Ma lines. The measurements were performed on a series of sub-
micrometer particles ground from NBS multielement research glasses. After applying an
absorption correction, the experimental results were compared to theoretical kaSi factors
determined with several different ionization cross sections.
As the thickness of the sample increases, the electron transparency of the thin sample
or small particle decreases, eventually reaching a thickness for which the sample no longer
Copyright © 2002 Marcel Dekker, Inc.
Figure 53 Calculated and measured kaFe factors for K line radiation from several elemental thin
films. Solid lines represent range of theoretical values. (From Wood et al., 1984. Reproduced with
permission of Blackwell Science Ltd.)
Figure 54 Calculated and measured kaFe factors for L-line radiation from several elemental thin
films. Solid lines represent range of theoretical values. (From Wood et al., 1984. Reproduced with
permission of Blackwell Science Ltd.)
where ra is the absorption jump ratio for element a and Aa and Ab are the atomic numbers
for a and b, respectively.
The absorption factor was derived by Goldstein et al. (1986) from the work of Tixier
and Philibert (1969) and Konig (1976). Williams and Goldstein (1991) calculated the
thicknesses of materials above which the absorption of x-rays was outside the 0.97–1.03
limits. The results of these calculations are given in Table 11.
As pointed out by Williams and Goldstein (1991), the effect of fluorescence must be
investigated in cases where the concentration of the fluoresced element is at a minor to
trace level, < 10 wt%, in a sample composed mainly of the exciting element. In a study of
Fe–10 wt% Cr (Nockolds et al., 1980), the fluorescence correction was only about 5%.
The equations for the absorption and fluorescence involve the measurement of the
film or particle thickness and the calculation of the mass absorption coefficients for ele-
ments a and b in the specimen. As the result, it is necessary to measure the thickness at
each analysis location and to calculate the concentrations in an iterative loop similar to
that used for the conventional ZAF schemes.
AlKa NiAl 9
PKa Fe3P 22
MgKa, OKa MgO 25
Source: Data from Williams and Goldstein, 1991.
of particles by WDS difficult at best (Armstrong and Buseck, 1978). In addition, the use of
multiple spectrometers, each with a different view of the particle and, therefore, a different
particle geometry, as shown in Figure 55, limited the quantitative capabilities of the
technique. The evolution of the various quantitative procedures resulted in the separation
of particles into roughly three size categories that are based on the procedures and in-
struments used for analysis (see Fig. 56). The particles less than 0.5 mm in diameter are best
analyzed in the analytical electron microscope using the procedures described in the
previous subsection. The particles with dimensions below about 0.1 mm require very little,
if any, particle-geometry corrections for elements with atomic numbers greater than about
11 and can be analyzed as infinitely thin films using kAB factors. As the particle size in-
creases, approaching 0.5 mm and larger, the effects of particle geometry must be included in
the analysis procedure (Goldstein, 1979).
Once the particle dimensions are greater than approximately 0.5 mm, the particles are
too large to be analyzed in the AEM and are best analyzed in the electron-probe or
scanning electron microscope. From 0.5 to about 20 mm in diameter, particle geometry
may have a pronounced effect on the generation and emission of x-rays and can sig-
nificantly affect quantitative analytical results. Many different analytical schemes have
been developed for the analysis of the particles in this size range. Finally, when the dia-
meter of the particle is greater than about 20 mm, the particle is large enough with respect
to the electron beam than it is essentially a bulk sample and classical analytical corrections
apply. The remainder of this subsection is concerned with the quantitative analysis of
particles in the 0.5–10-mm size range.
Figure 57 Monte Carlo simulation of electron trajectories in a large and small particle. (From
Small, 1981.)
correcting for the mass effect in the analysis of particles less than about 3 mm in size will be
an underestimation of the composition for all elements analyzed in the sample.
The second effect that must be corrected for is the absorption effect. In the analysis
of most particles, the x-ray emergence angle and, therefore, the absorption path length
cannot be predicted accurately as it can for polished specimens. The magnitude of this
effect is largest when there is high absorption, as is typically the case for soft x-rays from
elements like Al or Si that have energies less than 2 keV. The difference between the ab-
sorption path length in a particle and a bulk flat sample can result in widely different
values of emitted x-ray intensities. For the particle shown in Figure 59, the path length A–
B in the particle is less than the path length A–C in the polished sample. The path length
A*–D in the particle, however, is greater than the path length A–C in the polished sample.
The result of the varying absorption path lengths is shown in Figure 60, which is the SiKa
x-ray intensity from particles normalized to the intensity from a bulk material of the same
composition plotted versus particle diameter. For these spherical particles, the absorption
path lengths in the particles are less than the bulk material, resulting in a higher emitted
x-ray intensity form the particles compared to bulk. This effect is detectable primarily for
the lower-energy x-ray lines, such as that of aluminum at 1.49 keV, which are highly ab-
sorbed. This increased intensity for the low-energy x-rays results in an overestimation of
the concentration for these elements and a corresponding underestimation of elements
with high-energy x-rays for which absorption is not significant.
The third effect is caused by the fluorescence of x-rays by either the continuum or other
characteristic x-rays. Because x-ray absorption coefficients in solids are relatively small
compared to electron attenuation, the secondary x-ray fluorescence occurs over a much
Copyright © 2002 Marcel Dekker, Inc.
Figure 59 X-ray absorption in a particle compared to the absorption in bulk material. (From
Small, 1981.)
larger volume than the primary electron excitation. In bulk samples and standards, the
x-rays, for the most part, remain in the specimens. In the case of particles however, the
particle volume may be only a small fraction of the x-ray excitation volume. As a result, the
exciting x-rays will fluoresce relatively few x-rays before leaving the particle. In those
samples where the fluorescence is important, the effect of comparing a particle to a bulk
standard may be significant. This effect is shown in Figure 61, which is a plot of the range for
NiKa x-rays causing fluorescence of FeKa x-rays in a Ni–Fe alloy. The net effect of not
correcting for the secondary excitation of x-rays in particle analysis is as follows:
Figure 60 Normalized SiKa x-ray intensity plotted as a function particle diameter. (From Small,
1981.)
1. Normalization
One of the simplest methods for the quantitative analysis of particles is to normalize to
100% the concentrations, determined with bulk standards, from a conventional procedure
such as ZAF Eq. (109) (Wright et al. 1963):
Ca
Can ¼ P ð109Þ
i Ci
procedure does not accurately compensate for the absorption and fluorescence effects, the
most accurate results will be obtained on particle systems that meet the following con-
ditions: (1) systems for which all the analytical lines for the elements are above 4 keV
where the absorption effects are minimal; (2) if any of the analytical lines are below about
4 keV in energy, then the lines for all the elements should be as close together in energy as
possible so that the matrix absorption is approximately the same in all cases; (3) systems
for which there is no significant fluorescence.
Table 12 lists the results from the analysis of lead silicate glass particles. The first set of
results are taken from the analysis of the PbMa line at 2.3 keV, which is close in energy to the
SiKa line at 1.74 keV. Because the absorption and mass corrections are similar for these two
lines, the lead and silicon concentrations are in good agreement with the true values. The
second set of results were determined by analyzing the PbLa line at 10.6 keV. In this case, the
two analytical lines have very different energies and the particle absorption effect is not similar
in magnitude. As expected, the errors associated with this analysis are considerably higher than
those associated with the PbMa analysis. The lack of an effective absorption correction by
simple normalization can also be seen in Figure 63, which shows the relative error distribu-
tions, by element, for the analysis of K-411 glass microspheres (Roberson et al., 1995) with a
bulk K-411 glass standard. The error distributions show that the concentration of Mg with the
lowest energy x-ray, 1.25 keV, is overestimated, Si at 1.74 keV is centered at about zero error,
and the elements with harder x-ray lines (Ca, 3.7 keV and Fe, 6.4 keV) are underestimated.
2. Particle Standards
The analyst can use a conventional analysis scheme and substitute particle standards for
the normal polished standards (White et al., 1966). In this procedure, the assumption is
Copyright © 2002 Marcel Dekker, Inc.
Table 12 Analysis of Lead Silicate Glass K-229 by Normalization of ZAF Results
that the particle effects, particularly the absorption effect, will be approximately the same
for the sample and standard. This assumption is reasonably valid, providing the sample
and standard are close in composition and shape, and the particle diameter is above about
2 mm. Below 2 mm, as shown in Figure 60, any difference in size and shape between un-
known and standards will be critical because a small change in effective diameter will result
in a large change in x-ray intensity. Figure 64 shows the results of using particle standards
for the analysis of a series of K-411 glass microspheres. Figure 64a is the error distribution
for the normalized analysis of a population of spheres ranging in size from about 1 to
10 mm analyzed with a piece of bulk K-411 glass as the standard. The error distribution in
Figure 64a shows three distinct modes that correspond to the overestimation of under-
estimation of different elemental concentrations because of the lack of an effective ab-
sorption correction (see Sec. IV.C.1). In contrast, Figure 64b is the error distribution for
the normalized analysis of the same particle spectra used in Figure 64a, except a 5-mm
sphere was used as the standard rather than a piece of the bulk glass. The distribution in
Figure 64b has a single maximum, is centered on zero, and is narrower, ± 0.1, than the
distribution in Figure 64a.
for particles, where gðrx ; ry ; rz Þ is the distance from the point of x-ray generation to the
particle surface in the direction of the detector and fa ðrx ; ry ; rz Þ is the generated x-ray
intensity in the particle. Calculation of fa ðrx ; ry ; rz Þ requires the determination of the
number of x-rays, DI of element a, product per electron path length Ds as shown in
Eq. (112), which is identical to Eq. (99) except Ds is substituted for D(pt):
Ca oa pia N
DI ¼ ð112Þ
Aa Qa Ds
Introducing the integration limits for the particle volume and electron energy into
Eq. (112) yields Eq. (113) for the calculation of fðrx ; ry ; rz Þ:
ZEc Z2p Z2p
Ca N nðE; rx ; ry ; rz ; y; gÞ Qa ðEÞ
fðrx ; ry ; rz Þ ¼ dg dy dE ð113Þ
Aa oa pia drs=dV dE=drs
E0 y¼0 g¼0
3
Figure 64 (a) Error distribution for the normalized analysis of a population of spheres ranging in
size from about 1–10 mm analyzed with a piece of bulk K-411 glass as the standard. (b) Error distri-
bution for normalized analysis of the same particle spectra used in (a), except a 5 mm sphere was used
as the standard rather than a piece of the bulk glass.
The FðZÞ terms describe the continuum intensity as a function of the concentration-
weighted average atomic number for the hypothetical bulk material and standards. For
the simplest case, Kramers’ relationship (Kramers, 1923), the first part of Eq. (117) would
be Zbulk =Zstd , where the current estimation of concentration from the iteration loop in the
ZAF procedure is used for the calculation of Zbulk Bstd is the measured continuum in-
tensity for a given standard at the energy of the analyzed x-ray. For multielement stan-
dards, Bstd must be multiplied by the weight fraction of the element of interest. Once the
values for Bbulk are determined, they can be used in Eq. (116) to obtain the first estimates
for I . A set of k ratios can then be calculated and used as input to the ZAF routine. The
set of concentrations from each iteration is used to calculate new values of Bbulk and the
sequence is repeated until successive iterations agree within a predetermined limit.
Various mineral particles have been analyzed with the peak-to-background method,
FRAME P, (Small et al., 1979a). The results of these analyses are reported in Table 13
along with the result from the conventional ZAF routine, FRAME C (Myklebust et al.,
1979). In all cases, the analyses with the peak-to-background routine are within 10% and
usually better than 5% relative error of the stoichiometric values. In contrast, the errors
Copyright © 2002 Marcel Dekker, Inc.
Table 13 Analysis of Mineral Particles by the Peak-to-Background Method
Talc (Mg3[Si4O10](OH)2)
with the conventional ZAF routine range from 7.9% for S in ZnS to 47% for Mg and Si in
talc. In addition, standard deviations for individual measurements are less for the peak-to-
background routine than they are for the conventional ZAF routine.
In the analysis of particles, the assumption that the generation volumes for char-
acteristic and continuum x-rays are identical is only valid for particles larger than about
2 mm in diameter. Below this size, the anisotropic generation of the continuum results in a
significantly different excitation volume for the continuum compared to the isotropically
generated characteristic x-rays (Albee and Ray, 1970). This effect is shown in Figure 66,
in which isotropic and anisotropic cross sections have been used to calculate peak-to-
background ratios from K-309 glass particles normalized to bulk glass. The composition
of K-309 is 7.9% Al, 18.7% Si, 10.7% Ca, 13.4% Ba, and 10.5% Fe. These plots show that
the introduction of an anisotropic cross section for the continuum results in significantly
higher peak-to-background ratios for the smaller particles. As a result, it is necessary to
introduce a correction for anisotropic generation of the continuum for quantitative ana-
lysis with peak-to-background procedures.
In the peak-to-background method developed by Statham and Pawley (1978), the
peak-to-background ratio from a given element is compared to the ratio from a second
element:
Ca ðP=BÞa
¼ fab ð118Þ
Cb ðP=BÞb
In this case, fab is a correction factor that should have a minimal dependence on particle
size and can be calculated or determined empirically from standards. Table 14 lists the fab
Copyright © 2002 Marcel Dekker, Inc.
Figure 66 Peak-to-background ratios from glass spheres normalized to bulk ratios showing the
effects of isotropic and anisotropic cross sections for continuum generation. (From Small, 1981.)
values for five particles of K-961 glass, the composition of which is listed in Table 15.
These results show that the fab values do not exhibit any noticeable trend with particle size
and the standard deviation of fab is less than 8% relative for elements with a concentration
greater than 1 wt%. For the elements Mn and P, with concentrations less than 1%, the
standard deviation in fab is small considering the relatively poor counting statistics.
Copyright © 2002 Marcel Dekker, Inc.
Table 14 fab Values for K-961 Microspheresa
Diameter (mm) Mg Al Si P K Ca Ti Mn Fe
Element Wt%
Na 2.97
Mg 3.02
Al 5.82
Si 29.9
P 0.22
K 2.49
Ca 3.57
Ti 1.20
Mn 0.32
Fe 3.50
Figure 67 Secondary electron image of a fracture surface of Standard Reference Material 482.
(From Small, 1979b.)
Figure 68 Line profile of SKa x-ray intensity overlain on a target current image of a specimen
from a failed pressure vessel.
of the specimen. By storing WDS or EDS spectra at each position, the analyst can obtain a
full quantitative analysis at each pixel. Once the results from the analyses are stored in the
array, the variations in elemental compositions can be coupled to the gray-scale or color
output of a computer monitor. The completed array of pixels provides the analyst with a
spatially resolved map of the elemental composition across the specimen. In addition,
because the array is stored in the computer, the analytical information can be
postprocessed in several ways. For example, a row of pixels can be selected providing the
analysts with elemental x-ray line profiles across the sample in any direction or pattern
desired. This information can be displayed as a simple line plot or can be used to construct
contour maps of a given x-ray distribution.
The uncertainty associated with a quantitative analysis obtained for compositional
mapping is essentially identical to the uncertainty associated with a conventional analysis
obtained at the same electron dose. For maps where high accuracy is required on minor
constituents, the dwell time per pixel and=or beam current must be sufficiently large to
provide the appropriate x-ray intensity in the peak of interest. Compositional maps under
these circumstances may require several hours to accumulate and care must be taken to
ensure instrument stability over the required time interval.
As previously mentioned, both EDS and WDS analysis systems can be used for
compositional mapping. WDS detection of x-rays, however, is superior to EDS detection
for compositional mapping, particularly for samples requiring the analysis of con-
stituents at the minor and trace levels (Marinenko et al., 1987). One of the most im-
portant advantages of WDS detection is that the pulse processing time for WDS is at
least an order of magnitude shorter than EDS. WDS detectors have a limiting count rate
Copyright © 2002 Marcel Dekker, Inc.
of about 105 counts per second (cps) compared with a limiting count rate of 104 for EDS
detectors operating at maximum resolution. In addition, the WDS systems use diffrac-
tion crystals so that only a narrow energy band of x-rays are detected at any given time.
The energy is adjustable so that most of these counts are the characteristic counts from
the element of interest. This count rate compares to a limiting rate of about 102–103 cps
in the peak of interest on an EDS detector because the limiting count rate of 104 is
distributed over the entire energy range of the spectrum.
In addition to its higher limiting count rate, the WDS detector has a factor of 10
higher peak to background ratio than the EDS detector. Taking into account the higher
limiting count rate and peak-to-background ratio of the WDS detector, the WDS detector
has a lower detection limit of about 100 ppm compared to a lower limit of 1% or greater
for mapping on the EDS system.
One limitation of the WDS system compared to EDS systems is in the mapping of
samples which contain a large number of elements of interest. In WDS systems, the
number of elements mapped during a given digital scan is limited to the number of WDS
spectrometers. Typically, in EDS systems, all elements, with characteristic x-ray energies
above about 1 keV can be detected simultaneously.
Two important aspects that must be considered in the interpretation of composi-
tional maps are the counting statistics and resulting uncertainties associated with the
various point analyses at each pixel location. Under normal mapping conditions, it is
impractical to obtain the level of accuracy associated with conventional probe analysis
where 100 s or more are used for data accumulation.
Marinenko et al. (1987) have shown that for a 1286128 map at a magnification
below 5006, each pixel represents an area of about 2 mm in diameter on the specimen
surface. This means that the spot size of the beam can be on the order of 1 mm in diameter
and carry a current of about 1 mA. At this current density, the limiting count rate of 105
cps can be obtained on pure-element samples. Assuming a dwell time per pixel of 0.1 s,
a 1286128 map would require 1600 s to accumulate and have 104 counts per pixel. From
Poisson statistics, the 1s counting uncertainty per pixel is 1%. In the case where mulite-
lement standards are required, the dwell time per pixel must be increased to obtain the
same counting statistics, as each element is present at less than 100%. In general, dwell
times of about 0.4 s are sufficient for multielement standards and require less than 2 h per
1286128 map. As in conventional analyses, the standard maps can be archived and used
indefinitely, provided the instrument conditions are constant.
In the analysis of samples where one or more of the constituents are at the minor
or trace level (1% or less by weight), digital maps may require 10 h or more to ac-
cumulate. A 10-h scan on a 1286128 map translates to 2 s dwell time per pixel. If the
element of interest is at the 1% level by weight, then the x-ray intensity in the peak
would be 2000 counts, assuming a limiting count rate of 105 cps as stated above. The
associated counting uncertainty is 2.2% for 1s. The actual uncertainty will be larger
than this due to fluctuations in the intensity of the x-ray background, which must be
taken into account for quantitative analysis. In general, it should be possible to obtain
2s confidence intervals of 10% or better for constituents at the 1% weight con-
centration in an unknown sample.
The power of compositional mapping can be seen in Figure 70, which is a compo-
sitional map of the same Zn grain-boundary migration shown in Figure 69. The brighter
areas on the map correspond to a Zn concentration of 10% by weight. The faint band
extending to the left in the bottom of Figure 70 corresponds to a Zn concentration as low
as 0.2% by weight.
Copyright © 2002 Marcel Dekker, Inc.
Figure 70 Compositional map of the same Zn grain-boundary migration shown in Figure 69.
(From Piccone et al., 1982.)
One of the major difficulties in obtaining quantitative x-ray maps with crystal
spectrometers is the defocusing of the diffraction crystal as the electron beam is moved off
the crystal axis. The magnitude of the defocusing effect is proportional to the distance the
beam is from a point centered on the optical focus of the electron probe. The defocusing is
most severe at the low magnifications of 200–8006 routinely used in digital x-ray maps.
Corrections for the defocusing fall into four categories, (Marinenko et al., 1987).
The first method to avoid spectrometer defocusing is to employ, as mentioned above,
the digital movement in the microscope stage rather than the electron beam. In this si-
tuation, the electron beam position remains static at the optical focus of the diffraction
crystal and the sample is moved in a raster pattern under the beam. For WDS mapping, it
is critical that the sample remains at the focus of the optical microscope during the x–y
movement of the stage. This procedure has been used successfully by Mayr and Angeli
(1985). The success of this method, since it involves the mechanical movement of the stage,
is dependent on the reproducibility of the stage motion. If the reproducibility of the stage
movement is poor, then the accuracy of the quantitative maps will be poor except for
extremely low magnifications. In the more modern instruments, which have optical en-
coding and a guaranteed stage positioning of 0.1 mm, the results from moving the stage will
be comparable to the other corrections methods.
The second method used to correct for spectrometer defocusing is to move or rock the
diffraction crystal in synchronization with the raster of the electron beam, (Heinrich, 1981e).
By slightly rocking the crystal, the entire sample area under study will be maintained at the
focus of the x-ray crystal. The problems associated with crystal rocking are similar to those
involving stage motion. Because beam rocking requires mechanical movement of the crystal,
the results are dependent on the reproducibility of the crystal rocking mechanism. This
procedure has been described for multiple spectrometer by Swyt and Fiori (1986).
The third correction method is to collect a series of standard maps in conjunction
with the collection of a map from an unknown. If the standard and the unknown are
Copyright © 2002 Marcel Dekker, Inc.
collected under the same experimental conditions, then the standard maps can be used to
construct k ratios at each pixel location, thus normalizing out the effects of spectrometer
defocusing. Although this method does not require any mechanical movement, it does
require that the instrument remain stable over the period of time necessary to collect all
the maps from the standards and unknown. This is particularly important because a
pixel-by-pixel comparison of the unknown to the standard is made which requires a
constant crystal orientation with respect to the beam raster. Unless great care is taken
during an analysis, the use of archival standard maps is not recommended because of the
possibility of introducing large errors as a result of spectrometer drift. This procedure
was used by Marinenko et al. (1987) for the analysis of the Au–Ag standard reference
material 481 with the results shown in Figures 71 and 72. Figure 71 presents the un-
corrected maps from the alloys and the pure-silver standard, which show the char-
acteristic banding from the defocusing of the spectrometers. Figure 72 contains the pure-
silver map and the subsequent alloy maps that have been corrected for defocusing by
normalization to the standard map. The normalized maps show very little gradation or
structure in the image gray levels indicating the artifacts from defocusing have been
removed. Table 17 contains a comparison of the average composition obtained from the
digital map and the results from a conventional point analysis. In all cases, ‘‘the average
composition from each map compares quite well with the conventional point analysis as
well as with the NIST certified values’’ (Marinenko et al., 1987).
The fourth method used is to model the spectrometer defocusing. Marinenko et al.
(1988) observed that the defocusing artifacts (i.e., banding) were equivalent to an intensity
profile from a wavelength scan across the elemental peak of interest. This is shown in
Figure 73, which compares a wavelength line scan across the CrKa1,2 peaks to a line profile
taken from a Cr map perpendicular to the defocusing bands. The similarity of these two
traces implies that the intensity can be measured in the center band of a map and, from a
Figure 71 Elemental maps from Au–Ag alloys showing the effects of spectrometer defocusing
(upper left) pure Ag, (upper right) Au20–Ag80, (lower left) Au60–Ag40, and (lower right) Au80–
Ag20. The analytical line is AgLa. (From Marinenko et al., 1987.)
model of the line profile, the intensity at any other pixel location can be calculated. This
procedure eliminates the need for measuring a standard map and greatly reduces the time
required for obtaining a quantitative x-ray map.
Figure 74 is a schematic showing the relationship between the x-rays emitted from a
defocused point on the sample and the corresponding angular deviation from the line of
focus for a vertical crystal spectrometer. The electron probe is designed such that the
maximum x-ray intensity obtained on a crystal corresponds to the point A in Figure 74,
which is centered on the electron optic axis with the aid of an optical microscope. The
distance between point A and the crystal, S0, is
S0 ¼ 2R sin y ð119Þ
Alloy Element line Certified value Point beam analysisa Digital mapping
where R is the spectrometer radius and y is the Bragg angle. At a defocused point, point B,
which is a distance DS from the optic axis, the angular deviation from the exact Bragg
angle is Dy. Given that Dy is small, less than about 0.01 radians, which is the case for most
mapping applications, it can be approximated as
DS 0
Dy ¼ sin Dy ¼ ð120Þ
S0
where DS 0 ¼ DSðsin CÞ for a vertical spectrometer and C is the x-ray takeoff angle for the
electron probe.
Copyright © 2002 Marcel Dekker, Inc.
Figure 74 Relationship between the x-rays emitted from a defocused point on the sample and the
corresponding angular deviation from the line of focus for a vertical crystal spectrometer. (From
Marinenko et al., 1988.)
In calculating the background value for a given pixel in the compositional map, it is
necessary to determine DS, which is the distance of the pixel from the maximum focus line
of the spectrometer. The orientation and equation of the line of maximum focus for a
given spectrometer can be defined by the coordinates of two pixel points [i.e., (x1 ; y1 ) and
ðx2 ; y2 Þ] which lie on the line. The orientation of the line can be determined from any raw
intensity map by increasing the image theshold as shown in Figure 75. The line has the
general form Ax þ By þ C ¼ 0 and it follows that DS for any pixel point, represented by
coordinates ðx3 ; y3 Þ, in the x-ray map can calculated from
Dx3 þ Y3 E
DS ¼ ð121Þ
ðD2 þ 1Þ1=2
where D ¼ ðy2 y1 Þ=ðx2 x1 Þ and E ¼ y1 Dx1 .
DS is positive for points located above the line of focus and negative for points below
the line of focus. DS as defined in Eq. (120) is defined in units of pixel elements which must
be converted to centimeters in order to obtain the linear distances for the calculation of Ds.
Copyright © 2002 Marcel Dekker, Inc.
Figure 75 Line of maximum x-ray intensity on a crystal spectrometer determined by increasing
the image threshold. The threshold is low in the top image and high in the bottom image. (From
Marinenko et al., 1988.)
For this purpose, the scaling factor F is used, which is defined in terms of the magnification
M, the linear dimension of the CRT display L, and the number of pixel points in the
matrix N:
L
F¼ ð122Þ
MN
Dy can then be expressed as
0 F
Dy ¼ Ds ð123Þ
S0
The accuracy of this procedure is demonstrated in Figure 76. Figure 76A is an experi-
mentally measured intensity profile taken perpendicular to the focal axis and Figure 76B is
the modeled profile along the same axis. The intensity profiles shown together in
Figure 76C are identical within the limitations of the experiment.
The use of standard maps or modeling to characterize spectrometer defocusing is
magnification limited. At magnifications of about 1506 or less the deviation in intensity
from the edge to the center of a map is about a factor of 2 and cannot be accurately
3
Figure 76 Comparison of experimental and modeled x-ray intensity profiles for TiKa x-ray:
(A) Experimental x-ray profile for TiKa overlain on the corresponding Ti map; (B) modeled intensity
profile; (C) profiles from (A) and (B) together. (From Marinenko et al., 1988.)
Figure 78 Compositional maps for an aligned Cu–Ti eutectic: (A) Cu map; (B) Ti map. (From
Bright et al., 1988.)
B. Composition^Composition Histograms
The final aspect of electron-probe analysis that will be considered is composition–com-
position histograms. Compositional maps provide the analyst with a visual method of
interpreting the results from the elemental analysis of some 4000 individual points on a
sample (Prutton et al., 1987, Browning, 1987, Bright et al., 1988). By associating color and
intensity, the analyst is able to spatially relate the various elements and their compositions
within the analyzed region of the sample. One problem in the interpretation of compo-
sitional maps is that it is often difficult to visualize the compositional ranges and resulting
interelement correlations from a color composite image, particularly in the case of minor
or trace constituents. An alternative method of displaying the analytical information is in
the form of a composition–composition histogram (CCH) (Bright et al., 1988). The CCH
provides the analyst with an image that can be used to interpret the numerical relation-
ships among the various components in the sample.
Figure 77 is a schematic diagram describing the CCH. The concentration of element
a at each pixel in the compositional map of a, is associated with the corresponding pixel
Copyright © 2002 Marcel Dekker, Inc.
Figure 80 CCH and corresponding map areas for the high-Cu region of the eutectic: (A) outline
of the high-Cu region of the CCH; (B) pixels in the compositional map that correspond to the out-
lined area of the CCH. (From Bright et al., 1988.)
and concentration of element b in the map of b. This is represented by the top two blocks
in Figure 77. The associated concentrations for a and b at each pixel in the compositional
image are then plotted in the CCH, bottom block, as a single point in a scatter diagram.
Associations between elements are visible in the CCH as features, such as lines or areas,
which have a detectable density of points above background. In addition, the overlapping
of multiple-image pixels at one CCH point can be coupled to the intensity of the recording
CRT and displayed in an appropriate gray-level or color scale.
Copyright © 2002 Marcel Dekker, Inc.
An example of the type of information conveyed by the CCH is shown in Figure
78A, 78B, and 79, which are the elemental maps and the CCH for an aligned Cu–Ti eu-
tectic alloy. The CCH (Fig. 79) shows two lobes that have a high density of pixels and
correspond to the two different phases of the alloy (Bright et al., 1998). In addition to the
two lobes in Figure 79, there is a distribution of points between the lobes which has a
lower pixel density compared to the density in the lobes.
Figure 81 CCH and corresponding map areas for the region of the eutectic containing both Cu
and Ti: (A) outline of the connecting pixels between the high-Cu and high-Ti lobes; (B) pixels in
the compositional map that correspond to the outlined area of the CCH. (From Bright et al., 1988.)
REFERENCES
Martina Schmeling
Loyola University Chicago, Chicago, Illinois
I. INTRODUCTION
The preparation of samples presents, for any analytical method, one of the most important
steps for a reliable analysis. It is critical with regard to errors, but better precision as well
as lower detection limits can also be achieved by suitable sample pretreatment. The X-ray
fluorescence (XRF) techniques [energy-dispersive XRF (EDXRF) and wavelength dis-
persive XRF (WDXRF)] allow the analysis of almost all types of sample with usually less
effort of sample preparation. However, for some materials, the sample preparation step is
even more time-consuming than the analysis itself; therefore, suitable procedures should
be developed and introduced, leading to reliable results.
Usually, the specimens are presented to the XRF spectrometer as solids, powders,
briquettes, fusion products, liquids, or films, and for most of them, the pretreatment can
be kept to a minimum. Especially for metallic specimen, liquids and atmospheric particles
need only small effort. If required, possible sample preparation steps may include simple
procedures like cutting, grinding, milling, and mixing, or a combination of them. The
preparation step may also involve a conversion from the present state of the sample into
another one in order to receive a homogenous sample. It is important that the treatment
must be practicable for both samples and standards.
Metals and metallic alloys often need only to be polished to obtain a suitable surface
for analysis. Liquid samples, containing high enough concentrations for a direct analysis,
can be transferred into special cups or pipetted onto a surface target and dried to obtain a
thin film. Atmospheric particles, collected onto filters, need practically no sample pre-
treatment. However, if sample preparation is required, as for geological and/or in-
homogeneous samples, the procedure should be fast and reproducible, require small effort,
and avoid contaminations. A consultation of the literature is very important before the
application of special sample preparation procedures because usually the problem oc-
curred already earlier and some proposals or possible solutions can be found. Extensive
sample preparation should always be avoided because with each additional step, the risk
of losses and contaminations increases.
X-ray fluorescence is a traditional analytical method for the analysis of solid samples.
They may be presented to the spectrometer as such, as powders, as briquettes, or as
fusion products. In principle, it is possible to determine the bulk composition of solid
samples directly if the element distribution inside the solid specimen is homogeneous in
all three dimensions and the specimen shows a satisfying flat surface without scratches
and striations. Often such samples need only to be polished at the surface with suitable
abrasive materials and to be cut into the appropriate size for the sample holder of the
spectrometer. If the sample is too small for the sample holder, the positioning might be
critical, and it is useful to prepare a wax mold that fits into the holder. Such a mold
permits a series of samples and standards and can be replaced reproducibly. This tech-
nique especially allows the analysis of small manufactured parts (Bertin, 1970). Often
solid samples to be analyzed show surface roughness and=or damages, which cause
problems due to absorption effects. Such specimens must be finished at the surface before
analysis in order to reduce these damages. Several different techniques are available for
finishing a sample surface. The most common and effective one is mechanical polishing
with special abrasives like Al2O3 or SiC, or with coarse and fine abrasive paper. A final
finish of 100 mm surface roughness for short x-ray wavelengths and down to 10 mm for
long wavelengths may be satisfactory for a reliable analysis. Other polishing techniques
Copyright © 2002 Marcel Dekker, Inc.
are electropolishing or etching. However, these are not as universal as mechanical pol-
ishing because the sample might be destroyed or the surface composition changed.
A. Metallic Specimens
X-ray fluorescence is one of the most common techniques for the analysis of samples in
the metal production and, therefore, the preparation of such a specimen should be fast,
simple, and reproducible. Usually, the metallic specimens are prepared as solid disks by
conventional methods like cutting and milling. Hard alloys can be ground and pelletized
prior to analysis. However, for most metallic samples, it is required to polish the surface
in order to avoid striations, which give rise to the so-called shielding effects (Jenkins and
Hurley, 1965; Müller, 1972). For a reliable analysis, the surface roughness should not
exceed the path length at which 10% absorption on the radiation will occur (Jenkins
et al. 1981). In Table 1, some 10% absorption path lengths (presented as x90% in the
table) for common analytical situations of metallic specimens are indicated. For mea-
surements at short wavelength, a final finish of 100 mm might be satisfactory, which can
easily be reached with abrasive paper or diamond paste. If light elements are determined,
a 10–50-mm surface roughness should be claimed and fine abrasives are required to
obtain an almost mirrorlike surface. Striations, even after polishing, cause a reduction of
the fluorescence intensities, which is more serious for long wavelengths than for short
ones (Bertin, 1978; Tertian and Claisse, 1982).
Another aspect which should be taken into account is the orientation of the speci-
men inside the spectrometer. By simple rotating of the sample with 45 or 90 in the
spectrometer and repetitive measurements, the grooves decrease the fluorescence intensity
more when oriented in the perpendicular direction to the incident beam than in the parallel
direction. In Figure 1, a plot of the fluorescent intensity versus the groove size for Al, Fe,
Cu, Pb, and Mo is presented (Müller, 1972). Also, here it is obvious that the reduction of
the fluorescence intensity is much more severe for elements with low Z numbers. However,
today, most spectrometers are provided with a spinning mechanism, which averages this
effect. For a spectrometer without a spinning mechanism, the sample should be placed in
such a way that the grooves appear parallel to the incident radiation in order to minimize
the effect. Several mathematical models have been developed to determine the relationship
between surface roughness and emergent analyte line intensity (Berry et al., 1969; Müller,
1972).
However, simple mechanical polishing is not always possible or even advisable.
Particularly, when such elements as Si, Al, or Fe are to be determined, it is sometimes
difficult to find a suitable abrasive which does not contain one of these elements and does
not introduce contaminations into the sample. SiC and Al2O3, for instance, are very
Table 1 Mass Absorption Coefficients and 90% Transmission Path Lengths for Some Elements
common and effective abrasives, but both contain elements of high analytical interest. In
such cases, electrolytic polishing or etching might be more useful. When the surface of the
specimen is too rough (e.g., in the case of turnings) and suitable polishing and milling is
not practicable, the samples can be prepared properly by pressing them into pellets with a
hydraulic press at several hundreds of mega Pascals. With such a procedure, most of the
turning are compacted into satisfactory smooth surfaces.
If the sample is composed of soft, malleable multiphase alloys, smearing effects of
softer components like Pb or Al cause problems. The soft elements will be enriched onto
the surface and the fluorescent intensities of the softer phases increase, whereas those of
the harder ones decrease (Tertian and Claisse, 1982). In such cases, the sample should be
finished by electrolytical polishing or etching, instead of mechanical polishing.
In the case of metallic specimens with irregular shape, it is possible to embed the
piece in a special wax resin (e.g., acrylic resin and methyl acrylic resin). The resulting block
can be polished and ground to appropriate surface smoothness with abrasives like SiC and
diamond paste before analysis (Wybenga, 1979).
If the surface of the sample is not representative for the bulk composition, due to
heterogeneity, corrosion, smearing, and so forth, it is necessary to treat the sample in
different ways. Bronze alloys can be prepared, for example, by etching with a solution of
alcoholic ferric chloride (120 mL C2H5OH, 30 mL HCl, and 10 g FeCl3), and brass alloys
by a solution of aqueous ammonium persulfate [100 mL H2O and 10 g (NH4)2S2O8]. For
archeological objects, however, it is not recommended to treat them by etching because
many inclusions could be dissolved and lost.
Another possibility is the complete decomposition or dissolution of the sample.
However, such a radical method of sample preparation should only be carried out when
other, less aggressive preparation techniques are not suitable for obtaining satisfactory
results. It should always be taken into account that a digestion or dissolution step con-
tributes an additional source of contaminations or losses and is quite time-consuming
(Tschöpel et al., 1980). The procedure and the necessary reagents must be carefully se-
lected in order to get the highest benefit. An effective and common method is the digestion
Copyright © 2002 Marcel Dekker, Inc.
of the sample with aqua regia (HCl:HNO3, 3:1). Most of the alloys and metals, even Pd
and Pt, can be attacked by this reagent. The strength of this combination is based on the
formation of chloro complexes during the reaction and to the catalytic effect of Cl2 and
NOCl. The effectiveness can be improved additionally when the solution is allowed to
stand for 10–20 min before heating to 60 C. So-called inverted aqua regia (HNO3:HCl,
3:1) is an effective combination to oxidize sulfur- and pyrite-containing samples.
Another agressive decomposition reagens is the combination of HNO3=HF. Here,
the complexing effect of the fluoride ion is utilized and, with this combination Si, Nb, Ta,
Zr, Hf, W, Ti, Nb–Sn alloys, FeNb, Al–Cr alloys, FeSi, FeZr, Cu–Si alloys, FeP, FeW, as
well as Al–Si and Ca–Si alloys can be dissolved. Table 2 shows some possible combina-
tions and their applications (Bock, 1979).
For the determination of traces in Nb, Mo, W, Ti, and Ta, Eddy and Balaes (1988)
dissolved the metallic samples in HF=HNO3, and extracted the Ta by reversed-phase
extraction chromatography using tributyl phosphate. They pipetted the sample solution
either directly onto a filter paper or coprecipitated the trace elements with In. The de-
tection limits were between 0.05 and 0.7 mg, depending on the element to be determined.
Semiconductors as well as PbTe and GeTe can be destroyed by a combination of con-
centrated HNO3 and 10% sodium oxalate solution (1:2) (Fano and Zanotti, 1974).
However, some elements may be lost by application of a digestion step. Especially
elements like Se, Hg, and Sn form volatile components with various kinds of acid and their
combinations. Samples treated with hot mixtures of HNO3=HF=HClO4 for example, lose Se
and Cr completely, and Hg, As, Ge, Te, Re, Os, and Ru show particular losses (Tölg, 1962).
HCl alone is often not aggressive enough to attack metals and metallic alloys but
might show satisfactory results in some cases. For the determination of Cu in W alloys, the
samples are dissolved with 50% HCl, and the solution decanted from the undissolved W
powder. The undissolved W is washed with water, and both, the received solution and the
wash, are combined, diluted, and adsorbed on a piece of filter paper. Subsequently, the
dried filter is presented to the instrument (Ning and Zhao, 1995).
Sometimes, less common dissolution reagents are useful in XRF. Trace levels of Zr,
Mo, Hf, and W in Ta metal can be determined after removal of the matrix elements with
diantipyrylmethane and collection of the impurity elements onto ion-exchange paper
(Knote and Krivan, 1982). For the determination of Co in Fe, Kato (1990) dissolved the
sample with acid peroxide, mixed it with citrate, and percipitated Co from the solution of
pH 7.0 with 2-nitroso-1-naphthol.
Non-attacked
Reagents Reaction Attacked elements elements
HNO3 Oxidation Cu, Ag, Bi, Tl, Te, Au, Pt metals, Al, B, Cr, In,
Pb, Hg, Cd, V, U, alloys Nb, Ta, Th, Ti, Zr, Hf
HNO3=HCl (1:3) Oxidation with Au, Pt, Pd, Mo, Rh, In, Ru, Ti
complexation stainless steel, most alloys
HNO3=HF Oxidation with Nb, Ta, Mo, W, Ti, Sn, Sb,
complexation Si, Zr, Hf, Fe-alloy steel,
Fe alloys
HCl Solution Au, Pt metals
B. Powdered Specimens
A common technique for geological, industrial, and biological materials is the preparation
of powders and pellets. Powdered specimens are prepared when the original sample is too
heterogeneous for a direct analysis or too brittle to form self-supporting disks or when a
suitable surface finishing is not possible.
Powders can be derived from different materials: solids like minerals, rocks, ores,
slags, and so forth, metal fillings, metal oxides, precipitates, and residues from solutions,
ground fusion products, ion-exchange resins, and ashed or freeze-dried biological mate-
rials. Some of the materials are already available as powders, but most of them must be
pulverized by crushing, grinding, and milling. The powders are presented to the spectro-
meter directly as loose powders, packed in cells or spread out on film materials like
Mylar1 or other foils and as pellets or briquettes. Sometimes, fusions (see Sec. III) of the
powdered material may be a more suitable alternative, especially when particle size effects
are severe and need to be overcome. However, the preparation of a powdered sample
requires much less time than the preparation by using a fusion technique.
The analysis of a powdered specimen shows several advantages over bulk materials:
Powdered samples allow the addition of a standard and low- and high-absorption dilution.
The heterogeneity of the material will be reduced to a certain level, and standards can be
prepared easily in the same way. However, if the sample consists of elements with different
grinding properties, which is often the case for mineralogical and geological samples,
additives must be added to overcome such a problem. In general, 2–10% additive to aid in
the grinding, blending, and briquetting process is sufficient for most of the materials.
To fulfill the condition of a homogeneous sample surface with infinite thickness, on
which most of the correction algorithms are based (see Chapter 5), the sample must be
converted into a very fine powder, otherwise particle size effects can occur and interfere
with the quantitative analysis. For very fine particles, the fluorescent intensity is constant,
but it decreases with increasing particle size. This effect is more significant for long
wavelengths than for short or intermediate ones. Furthermore, the degree of packing of
Copyright © 2002 Marcel Dekker, Inc.
the specimen influences the fluorescence intensity additionally. However, this can be re-
duced by the preparation of pellets and by briquetting the sample; a higher pressure and a
longer pressing time for the pellets give an additional improvement.
Segregation and, hence, heterogeneity effects can arise when the particles are not of
the same size, shape, or gravity. This problem occurs particularly when the sample is
analyzed in the form of a loose powder. Claisse (Claisse, 1957; Claisse and Samson, 1972)
proposed a qualitative interpretation for the origin of such heterogeneity effects in some
publications. The effects can be a major source of errors because large variations in
fluorescence intensity may occur and the influence becomes more and more detrimental
with increasing wavelength; sometimes, errors of about 50% are possible (Berry et al.,
1969; Rhodes and Hunter, 1972; Madlem, 1976; Holynska and Markowicz, 1982). To
avoid severe errors, it is always recommended to tap the powdered sample for better
mixing of the powder prior to analysis.
Different mathematical methods for the correction of particle size effects have been
developed. However, with only few exceptions (see Chapter 6), they are not very useful, as
most of them require an a prior knowledge of the particle size distribution in the sample,
which is usually unknown.
1. Grinding
For the preparation of fine powders, the original sample should usually be split and milled
into smaller pieces, which are then suitable for further grinding. Splitting in smaller
subsamples can be carried out by various equipments, like jaw mills or crushers. The main
objective at this stage is to obtain a representative subsample from the original material
and to avoid sampling errors due to wrong handling or treatment. For further grinding of
the sample to appropriate small sizes, various methods and devices are available. The most
common equipments are disk mills or shatterboxes, in which samples of smaller sizes than
6 mm can be ground further.
However, both steps present a source of contaminations, as most of the grinding
tools contain elements, which might be of interest in the sample. These elements can be
introduced easily as blank values during the grinding process and, therefore, the material
of the grinding tools should be selected carefully for each particular application. Agate,
zircon, steel, alundum, and WC are the most common materials for such vessels. All of
them can introduce various blank values. Agate, for instance, introduces traces of SiO2,
Mg, and Ca to the specimen. This might be of less interest for geological materials, which
often contain these elements as main compounds, but for biological material, this re-
presents a major source of errors (Van Grieken et al., 1980). However, agate and zircon
equipments insert still lower blank values to the sample in comparison to other materials.
Unfortunately, agate is very expensive and brittle, and for this reason, it is mainly used for
the preparation of small sample amounts by manual grinding in a mortar. With zircon
devices, only blanks of Zr and Hf occur, which are negligible for most of the applications.
Despite its good mechanical properties, the use of WC vessels is only partially re-
commended. During the sample preparation, substantial amounts of W, Nb, Ta, and Co
can be introduced to the sample (Hergt and Sims, 1994). Especially the W contamination
can give serious problems because the L and M lines of this element interfere with the
elements of Z numbers between 28 and 35. However, despite this drawback, WC is the
most favorable material for the preparation of hard alloys due to its material properties.
Steel is also a frequently used material, and the preparation of rock samples with bakelite
phenolic resin in a steel mill showed good results (Longerich, 1995). However, a con-
tamination with Fe, Cr, Ni, and Mn is regular and can cause severe errors.
Copyright © 2002 Marcel Dekker, Inc.
When grease and oil-forming materials are analyzed, the selection of a suitable vessel
material should be done carefully, as well. The preparation of these samples should be
carried out in vessels with smooth, nonporous surfaces, like agate. Otherwise, parts of the
sample can cling to the surface of the device and losses will occur. Sieber (1993) obtained
homogenized samples of greases by mixing them with a chemplex binder in a high-speed
rotar. The mixture was heated up during the procedure and was analyzed directly after
cooling down as a cake. Oil samples can be ground with MgO and baked afterward at a
temperature of 270 C. The received solid cake is suitable for direct analysis (Liu et al., 1992).
Reagents or additives are used during the preparation procedure for different rea-
sons. Fine abrasives like Al2O3 or SiC are often added as grinding aids to accelerate the
process (Adler and Axelrod, 1995). The addition of wet grinding aids like ethyl or iso-
propyl alcohol leads to a better homogeneity and higher efficiency. Calcination in an
airstream at 800 C and subsequent grinding to a fine power yields good results for ma-
terials like Pt or Pd (Gokhale and Wuensche, 1990; Norrish and Thompson, 1990). Ma-
terials that are difficult to treat, like rocks or ores, can be prepared much faster and more
efficiently after addition of sodium stearate.
Normally, grinding is very fast and takes only a few minutes, but it should always be
taken into account that the determination of light elements requires a fine-particle size.
The effect of grinding time or particle size on the fluorescent intensities of the elements Si,
S, K, Ca and Fe in coal samples prepared by the powder technique (5-g sample was
ground in WC rotary swing mill together with a 1-g boric acid binder and 100 mg sodium
stearate) is shown in Figure 2a (Wheeler, 1983). From this figure, it is clear that a grinding
time of around 6 min will reduce the particle size in a coal sample to the order of 50 mm,
but no further increase in intensity is obtained for longer grinding times.
2. Loose Powders
After grinding, the specimen can be analyzed directly as loose powder or further pressed into
a pellet. For the irradiation as a powdered specimen, the ground sample is transferred into a
special cup and covered by Mylar foil. The cup should be filled in such a way that infinite
thickness is assured, which means 75% in most of the cases. The covering film should be
made of a material with homogenous density in order to avoid scattering effects. If a sample
will be irradiated several times, as in the case of standards, such films tend to become brittle
and may break inside the XRF unit. To avoid this, the film should be changed regularly.
Another possibility for the analysis of a powdered specimen is the preparation of a
thin film by dusting the powder on adhesive tape (e.g., Scotch1) or onto foils. This can be
carried out manually or, with higher benefit, by using the so-called ‘‘puff technique’’
(Bertin, 1978). A special device was developed for such purposes: The device is filled with
the powdered sample and subsequently the powder is dispersed into fine dust by blasting
small air volumes inside the fairly evacuated chamber of the equipment. This dust is then
deposited onto a filter or frit by further evacuation of the chamber.
A further approach is the performance of a slurry analysis. By mixing the fine-
powdered sample with a solution of 2–5% (w=v) nitrocellulose in 1 mg=L acetate, a
suspension is formed which can be transferred on a microscope slide or other supports
(Bertin, 1978). A simpler approach is slamming of a slurry solution with water and fine
powder and sucking this suspension through a filter (Nuclepore1, Mylar, etc.) with fine
pore sizes, to obtain a thin and very homogenous layer (Wybenga, 1979; Araujo et al.
1990). A major drawback of this method is the loss of water-soluble elements. To
avoid these problem, nonpolar solvents, like hexane, are sometimes used. An alternative
Copyright © 2002 Marcel Dekker, Inc.
Figure 2 (a) Grinding time versus intensity. (b) Pelletizing pressure versus intensity. (From Ref.
30. Reprinted by permission of Plenum Press.)
In 1956, Claisse proposed preparing fused samples for analyses by XRF in order to over-
come the problems arising with powders and pellets (Claisse, 1956, 1957). This technique
has become one of the most successful methods for the preparation of specimens in x-ray
fluorescence and can be applied to almost all types of samples (Feret, 1990, 1993). With a
fusion, all particle and mineralogical effects, which cause serious problems for analysis of
powdered and pelletized specimen, become negligible and the analysis shows a better ac-
curacy (Banajee and Olson, 1978; Frechette et al., 1979; Tertian and Claisse, 1982; Metz and
Davey, 1992). The basic procedure consists of heating a powdered sample together with the
flux to such high temperatures that the flux melts, dissolves the sample, and reacts with it by
the formation of a one-phase disk. Agitation during the fusion process is essential for fu-
sions with low fluidity, in order to avoid the formation of bubbles and to receive a homo-
genous sample (LeHouillier and Turmel, 1974). The reaction time varies between 5 and
30 min depending on the sample composition. For a 30-mm disk 6–7 g material (flux and
sample) is required. For most applications, the flux=sample ratio varies between 2:1 and 10:1
(Bower and Valentine, 1986; Eastell and Willis, 1990; Kvaratskheli and Kornienko, 1992;
Ossaka et al., 1994; Spangenberg et al., 1994). In some special applications this ratio may be
different (Haukka and Thomas, 1977; Fink et al., 1993; Sato and Kohori, 1995).
One of the major advantages of the fusion technique is the possibility of adding a
heavy absorber for the reduction of matrix effects, especially for the determination of light
elements. Other advantages are the relatively simple preparation of synthetical standards
and the wider calibration range (Feret, 1990, 1993). The main disadvantage of this tech-
nique presents the dilution effect of the sample, which renders the determination of traces,
especially of light elements, more difficult, as their fluorescence intensity is decreased.
However, in the last years, several attempts were carried out to overcome this dis-
advantage [e.g., by application of the low dilution fusion technique (LDF) (Haukka and
Thomas, 1977; Thomas and Haukka, 1978; Lee and McConchie, 1982; Eastell and Willis,
1990; Maruta et al., 1992; Yamamoto et al., 1995) or a semilow dilution technique
(Malmqvist, 1998)]. Further disadvantages include the higher costs in comparison to the
powder methods and the more complicated preparation procedure of the fusion. Some
experience is required and most of the analysts have their special ‘‘tricks’’ to obtain the
best results. Today, several automated flux preparers are commercially available with
more or less sophisticated technology (Stephenson, 1969; Schroeder et al., 1980; Kvar-
atskheli and Kornienko, 1992).
For some elements like S, however, the briquetting technique is preferred and offers
better accuracy, because high losses due to volatilization during the fusion procedure occur.
Flux
Flux base composition Properties Applications
B. Fusion Procedures
For the fusion procedure, the sample should be ground with at least 150-mesh grain sizes
to yield a fast and homogenous fusion. For samples that are difficult to fuse, like
Copyright © 2002 Marcel Dekker, Inc.
refractories, grain sizes down to 200 mesh are recommended in order to obtain a homo-
genous fusion and accelerate the entire fusion process (Norrish and Hutton, 1969; Hutton
and Elliot, 1980). The sample is weighed out and mixed fairly. The flux, if consisting of
different compounds, is also mixed well and transferred into the crucible. The sample is
placed on top of the flux and both are mixed. A selection of different materials for
the crucible is available (Giles et al., 1995); common ones are Pt, Pt–Au alloys (95%:5%),
Pt–Rh–Au alloys (85%:15%:5%), zirconium grain-stabilized platinum (ZGS), and Ir or C
vessels. The latter ones act also as oxidants, but C vessels are easily attacked by sulfides
and other reducing compounds. Preferable to Pt is the Pt–Au alloy, as it is harder and
more difficult to degrade than the pure metal. This combination is nonwetting and rela-
tively long lasting. Recently, Ir crucibles became available, which show good properties,
such as corrosion resistance to all metals, insolubility in acids, even aqua regia, and
maintenance of their properties under normal atmosphere up to 1600 C.
The fusion procedure can be carried out in high-frequency induction furnaces, gas
burners, or muffle ovens. The first two show the advantages of higher temperatures and
faster heating. The sample is previously heated until the glass-forming agent is melted;
with a further increase of the temperature to 1050–1100 C for 10–15 min, the flux reacts
with the sample, forming a glass bead. Agitation during the fusion process is re-
commended for fusions with low fluidity in order to avoid bubble inclusions (Luke, 1963;
Harvey et al., 1973; LeHouillier and Turmel, 1974; Pella et al., 1982; King and Vivit, 1988;
Norrish and Thompson 1990). The melted fusion is casted into a preheated mold, for the
formation of a disk to be analyzed, and allowed to cool down. For easier removal of the
fusion from the crucible and mold, special ‘‘antiwetting’’ agents can be added. Iodides and
bromides of light alkali and alkaline earth elements are favorable for this purpose. With
application of these agents, the crucible shows a much longer lifetime and its cleaning is
much easier. Only a few droplets of these agents are required to be added on top of the
sample=flux mixture before heating.
Sometimes, the fusion is annealed at 200 C for some minutes in order to avoid
thermal shocks (Bower and Valentine, 1986), but, usually, the mold is preheated to the
same temperature and the melt can be transferred without suffering. For a reliable ana-
lysis, it may be required to clean the disk with alcohol and polish it with abrasives. The
resulting disks are stable over a long time but should be stored in a desiccator under
vacuum or in other moisture-free environments. Borates are hygroscopic and extended
exposure to ambient air may destroy the disk’s surface.
Various additives may be used for different reasons, according to the sample be-
havior during the fusion procedure. The addition of glass-forming agents like SiO2 or
Al2O3 is required to obtain a more stable disk when a higher amount of alkaline earth
elements is present in the sample. Transparency-increasing agents (e.g., LiF, NaF, or KF)
are added for higher fluidity of the fusion and to assure better homogeneity.
For the fusion of organic material, preignition for several hours at 450 C is necessary
to volatize all CO2. Otherwise, a weight loss will occur during the fusion, which cannot be
compensated for by mathematical calculation.
Evaporation of moisture or crystal water can also introduce some weight loss. If this
loss is similar for both the sample and flux, it can be calculated using the method of
Hutton and Elliot (1980) or compensated for by the addition of SiO2, if less than 20%
(Harvey et al., 1973). Any additional mathematical correction procedure is then super-
fluous. The losses of volatiles and=or flux usually increase with increasing temperature.
King and Vivit (1988) developed the following simple method for the prevention of losses
in geological samples: The powdered sample was dried for 2 h at 105 C and 2 g was
Copyright © 2002 Marcel Dekker, Inc.
weighed exactly and calcinated in a muffle furnace at 925 C until a constant weight was
achieved.
Cleaning of the mold can be carried out by reheating with a nonwetting agent or
inserting the crucible into a warm HNO3 bath (Tertian and Claisse, 1982).
Sometimes, the fused disks are ground and pelletized. For the determination of Nb
and Ta in geological samples, the fused disks were pelletized with cellulose powder for a
reliable analysis (Basu Chaudhury et al., 1987). Also, for the analysis of submarine
polymetallic nodules, the samples were fused by the application of the low-dilution fusion
technique and later pelletized without binder (Cai et al., 1992).
The high-dilution technique of Tertian and Claisse (1982) was investigated for the
determination of SiO2 in powdered plant materials. In this case, a 100-mg sample was
dissolved in 9.9 g lithium metaborate flux (Garivait et al., 1997) and satisfactory results
could be achieved.
Applications were also carried out in the field of ceramic materials. One study dealt
with the determination of Si as SiO2 in Si3N4. Here, the ground material was first pyrolized
at 1500 C before fusing with LiBO2 or a mixture of LiBO2 and B2O3 (Kaiser et al.,
1995).
When only small sample amounts are available, as in the case of air-dust samples,
microbeads might be formed with LiF and Li2B4O7 (Moore, 1993).
Liquid samples represent an almost ideal specimen for analysis by x-ray fluorescence. They
are homogenous, particle effects are eliminated, and the obtained analytical results are
representative for the whole sample. Furthermore, normally the matrix consists of ele-
ments with low atomic numbers, where absorption effects are small and enhancement of
the analyte lines is negligible. High absolute sensitivity is observed because the penetration
depth of the primary beam is high and low absorption of the emitted secondary radiation
exists. For the analysis of liquid-specimen special cells, made of stainless steel, poly-
ethylene, or polytetrafluorethylene (PTFE or Teflon), are in use, into which the liquid
sample is poured. These cells are usually covered by Mylar foils with thicknesses of 3–
6 mm; a lower thickness ensures less background scattering but is mechanically less stable
than the thicker ones. Sometimes, also microporous films are used, which allow the
equalizing of pressure differences between the liquid cell and the environment by being
permeable for air only. For the quantification of liquid samples, internal standardization
can be applied, but also calibration standards in each concentration range and blank
samples for the background evaluation are easily prepared. Solids, which are problematic
to analyze, such as powders, pellets or fusions, can be converted into liquids by dissolution
and digestion or, if trace analysis is required, the traces might be extracted with a suitable
solvent from the bulk material.
The major drawback of the analysis of liquid samples is the highly scattered back-
ground. This makes it difficult, or sometimes impossible, to determine light elements at
low concentrations and it increases the detection limits. Another disadvantage is the de-
terioration of standards with time. In contrast to standards prepared by fusion or pelle-
tizing, liquid standards change their concentration with time due to adsorption and=or
precipitation effects. Therefore, they should be prepared fresh or renewed each few days in
order to avoid systematic errors in quantitative XRF analysis. Also, the samples and the
standards may be altered by irradiation. During irradiation, the sample can be heated up
Copyright © 2002 Marcel Dekker, Inc.
and evaporation might occur, which influences the concentrations of the elements in the
sample. Bubbles can be formed by heating of the sample and influencing the analyte-line
intensity. Sometimes, precipitation occurs during the irradiation process, and the pre-
cipitate can be deposited at the bottom and change the properties of the sample.
Typical detection limits for the energy-dispersive XRF analysis of liquid samples
with 30 min counting time are in the ppm range (Van Grieken et al., 1976), which is sa-
tisfactory for several applications. Chlorine and iodine were determined by Morse (1992
and 1994) directly in photographic developer and fixer, respectively, with cadmium
standard solutions in polypropylene cups. Relatively high concentrations in the micro-
gram per gram level of Fe, Ni, and V in crude oils can be determined easily by using XRF
(Shay and Woodward, 1991). In the last few years, the field of direct applications ex-
panded, particularly in the oil sector (Kendall et al., 1995; Kira et al., 1995; Lyamina et al.,
1995; Untenecker et al., 1995). It appears that XRS is an excellent technique for the
analysis of various kinds of oil from highly raffinated greases (Untenecker et al., 1995) to
petroleum products and used oils (Kendall et al., 1995; Kira et al., 1995; Lyamina et al.,
1995).
However, in most cases, the elemental concentrations in liquid samples (e.g., in
environmental waters) are too low for a direct analysis of traces and the samples must be
preconcentrated prior to analysis. In general, all preconcentration techniques, which are
applied to other analytical methods, are also practicable for XRF analysis. Pre-
concentration has the additional advantage that possible matrix effects can be reduced or
even avoided (Van Grieken, 1982), and the accuracy of the results is improved. Un-
fortunately, a preconcentration step is often time-consuming, involves the risk of con-
tamination and losses, and may not always be specific for each elemental species.
Ideal for the presentation to XRF are thin-film targets, in which the x-ray intensity is
proportional to the mass of the element on the film. To obtain a thin-film specimen from
suspended matter in liquid samples, the liquid sample is filtered through a membrane
filter—preferable Nuclepore or Millipore—and dried. Vanderstappen and Van Grieken
(1976) found that filtration through a Nuclepore filter leads to interference-free detection
limits of 5–10 ng=cm2 for Mn, Ni, Cu, Zn, As, and Br, 10–20 ng=cm2 for Ti, V, and Cr, and
about 50 ng=cm2 for K and Ca by using EDXRS and counting times of 2000 s. Recently,
Civici (1994) used the filtration method for the separation of suspended material from
seawater. For the internal standardization, the standard element is added before filtration
and deposited together with the suspension onto the filter surface in order to get a
homogenous distribution.
X-ray fluorescence also offers an ideal technique for on-line process control in the
industry using a continuous flow of the sample through a sample chamber. Ceasy (1994)
applied this for the on-line analysis of molten metal samples with a portable EDXRF
instrument, equipped with two radioisotope sources, namely 55Fe and 244Cm, and a flow
cell. Another on-line application was carried out by Davidson (1994): he determined As,
Cu, and S in solutions from a lab-scale Cu electrolyte purification cell.
In fact, the XRF analysis of liquids increases from year to year. Numerous appli-
cations are cited in the review articles by Ellis et al. (1996, 1997) and Bacon et al. (1995).
A. Physical Preconcentration
Physical preconcentration methods are practicable for the analysis of rainwater, sewage,
and wastewater, and several different approaches have been carried out to remove the
liquid matrix in various ways. A simple and fast process of preconcentration is the
Copyright © 2002 Marcel Dekker, Inc.
evaporation of the liquid, leaving the elements to be determined as a residue on a surface,
which can be analyzed directly. This technique shows several advantages because of its
easy handling and direct approach. The probability of contaminations is low because no
complicated and extra sample preparation steps are required. The costs are reasonable:
only suitable carrier materials are necessary and all nonvolatile compounds remain
quantitatively on the surface. Even large amounts of water can be evaporated or freeze-
dried, and the residue is later mixed with binding materials, pelletized, and analyzed.
Freeze-drying of 250 mL wastewater on 100 mg graphite followed by grinding and pelle-
tizing of the residue leads to detection limits of 5 mg=L for many elements with accuracies
around 10% with EDXRF (Smits and Van Grieken, 1997).
Several attempts have been made to deposit analyte solutions on different surfaces.
In general, all materials with a smooth surface texture composed of low-atomic-number
elements are suitable. Especially, Nuclepore filters and Mylar foils show low-background
and blank values for XRF analysis. However, there are also some drawbacks to this
technique. Volatile species and elements get lost during the evaporation step. In the case of
high salinity or hardness, a formation of finite crystals will occur, which introduces particle
size effects. Special background correction methods need to be applied for different types
of water sample. Despite these problems, evaporation is still a popular technique for the
preconcentration of various liquids. Often, the samples are pretreated or predissolved in
solvents in order to increase their capability of spotting.
For the determination of U and Ce in nuclear fuel, the sample was first dissolved in
nitric acid solution and, subsequently, microdroplets of this solution were spotted onto a
filter. The XRF analysis of the dried filter showed an accuracy of better than 2% (Hanif
et al., 1995) for the elements of interest. Contents of less than 1 mg=g of Ca, K, S, P, Cl,
Mg, and Na could be determined in liquids, by spotting such a nitric acid solution on a
highly pure polycarbonate membrane (Gödiker et al., 1990).
Mylar foil is one of the most popular substrates for the deposition of samples
(Baryshev et al., 1995). Several interesting studies were carried out by use of this medium.
Oil sample, for example, can be deposited after homogenization with an organic solvent,
leaving a thin film glued onto the Mylar (Civici, 1995). Other substrate materials are
polypropylene and polycarbonate films. Cations (Ca, K, Mg, Na) and anions (Cl and
SO24 ) of drinking water were preconcentrated onto polypropylene foil. The addition of a
polysaccharide (GELERITE) reduced the hydrophobic effect of the foil and, thus, de-
tection limits from 0.12 mg=L for Ca to 1.78 mg=L for SO2 4 were claimed (Warner et al.,
1995). The advantage of this technique is the possibility of a simultaneous determination
of cations and anions. In another application in this direction, the sample is transferred as
a droplet onto a polycarbonate film, which is cemented on a 35-mm plastic photographic
slide mount. After evaporation, the resulting salt deposit is more or less circular. Detection
limits in the microgram per liter region could be reached (Meltzer and King, 1991).
Particularly, thin-film supports (like the type A2) have recently been developed by
Process Analytics (Orem, UT). The film can accumulate a small amount of solution in a
small spot because the film is treated to have a hydrophilic center in a hydrophobic field or
the film has a dimple in its center. These films are clean and resistant to high temperatures.
Wastewater samples with low salinity can be preconcentrated by a special vapor
filtration procedure (Luke, 1963; Rickey et al., 1977) in which the sample is deposited into
a container, with a cellophane foil bottom. The foil is permeable for water vapor, but not
for liquid water or dissolved material. Subsequent evacuation of the container leads to
deposition of the dissolved solids onto the membrane or foil, which is then presented for
analysis. In combination with particle-induced x-ray emission (PIXE), detection limits in
Copyright © 2002 Marcel Dekker, Inc.
the range of 0.1–3 mg=L were claimed. However, this method shows several disadvantages:
It is time-consuming and samples with high salinity cannot be applied successfully. Crystal
formation can occur, even with very low salinity, which has to be corrected by special
matrix correction procedures (Van Grieken, 1982).
Sometimes, metal surfaces can also be used for the deposition of extracted solutions.
A thin-film technique was applied for the determination of SO2 4 and Cl in small volumes
of saline samples, deposited onto suitable machined Cu disks (Dhir et al., 1995).
Other procedures are a nebulization technique (Giauque et al., 1977) and a multi-
droplet spotting technique (Camp et al., 1974), which result in a fairly homogeneous
distribution of the sample onto the surface.
The evaporation of larger samples can be carried out by impregnation of the filter
material (Whatman 41) with a sample solution. A wax ring 29 mm in diameter keeps the
spotted sample solution inside the area to be irradiated when the water is evaporated by
passing an airstream from underneath. The reported detection limits were below 50 mg=L
for 2000 s analyzing time, with optimum secondary target excitation (Giauque et al., 1977).
Water samples can be properly analyzed by TXRF without any preconcentration
step. Some microliters of the sample are pipetted directly onto a highly polished quartz
glass carrier and evaporated. The detection limits achieved with this techniques are one to
sometimes two orders of magnitude better than for conventional XRF. However, if
samples with higher salinity are analyzed, a separation step is also required.
Probably the best solution, when high precision and low detection limits are ne-
cessary, is a sample preparation involving a chemical preconcentration or separation step.
B. Chemical Preconcentration
Chemical preconcentration may be grouped into three main classes:
1. (Co)precipitation
2. Ion exchange
3. Chelation and sorption immobilization
Numerous applications to different water samples have been established in these
three groups (Camp et al., 1974; Rickey et al., 1977; Bruninx and Van Ebergen, 1979;
Andrew et al., 1982; Chakravorty and Van Grieken, 1982; Van Grieken, 1982; Becker
et al., 1985; Hirayama and Leyden, 1986; Shan et al., 1988) and most of them include a
final step in which the preconcentrated sample is deposited onto a surface (filter or foil) by
sorption or filtration. This surface serves as sample carrier and is subsequently exposed to
the x-radiation.
1. (Co)precipitation Methods
Precipitation methods are quite popular for the trace–matrix separation and they have
been used in analytical chemistry for a long time. They can be carried out very quickly and
with little effort. Selective precipitation of one or two species is usually the main task and
much of the time is dedicated to finding suitable reagents for this purpose. However, in the
case of XRF as a multielement technique, it is much more important to obtain a pre-
cipitate containing all elements of interest. Many attempts were carried out in this di-
rection and numerous reagents with different properties are practicable for this. A simple
and traditional method is the application of inorganic reagents like Fe or Al hydroxide,
which coprecipitate many elements at selected pH values. Sulfides are less common,
Copyright © 2002 Marcel Dekker, Inc.
because of the higher solubility of their precipitates. However, most of the inorganic re-
agents show an incomplete precipitation and a strong dependence on accurate pH values.
More common and effective are organic components, which form complexes of very
high stability. Often a spike element is added, acting as a carrier and an internal standard
at the same time. The most suitable organic reagents are the carbamates (Van Grieken,
1982). They form very stable and strong metal chelates. Luke (1968), for example, in-
vestigated, using his ‘‘coprex’’ technique, the coprecipitation of trace elements by sodium
diethyldithiocarbamate (NaDDTC) in the presence of a suitable metal ion spike. In a
further work, Kessler and Vincent (1972) improved the detection limits, obtained by Luke
(1968), by filtration of the received NaDDTC=hydroxide precipitate and analyzing of the
residue with a highly collimated XRF setup.
Toxic elements were successfully determined in plastic compounds for toys by ap-
plying NaDDTC. The samples were first mineralized with molten NaOH. Then Cd, Ba,
Sb(III), Cr(III), Hg, Pb, and As(III) were coprecipitated at pH 8.5 with NaDDTC and
sodium rhodizonate and Fe3 þ ions as the carrier. The results achieved by XRF were
comparable with the AAS results (Gimeno-Adelantado, 1993). When a high sample
throughput is required and all samples are of the same kind, automation might be pos-
sible. For such purposes, Tanoue et al. (1979) developed an automated device for the
preconcentration of trace metals in wastewater samples with DDTC and subsequent XRS
measurement.
Ammonium pyrrolidine dithiocarbamate (APDC) acts more efficiently, especially for
concentrations below 10 mg=L at a pH of 4. APDC is the most successsful nonspecific
reagent for the preconcentration of traces at this pH in comparison to other reagents.
Quantitative recoveries are obtained for a dozen elements in water, independent of the
alkaline earth element content. Various elements can be also determined in seawater, using
APDC as reagent. A small amount of Mo as carrier ion is added to the filtered seawater
sample, then the pH is adjusted at 4, and a portion of the 1% APDC solution is added.
After 15–20 min the precipitate is filtered off, dried, and analyzed. Ulrich and Hopke
(1977) achieved detection limits between 4 ng=cm3 for Br and 80 ng=cm3 for Ca by ap-
plication of this method. In another approach, traces of soluble metals were separated
from estuarine waters. In the first step, the water was filtered and acidified to pH 3. Fresh
APDC solution (2%) was added and then the complexes filtered. The obtained detection
limits were between 2 and 5 m=L (Mazo-Gray et al., 1997).
Quantitative recoveries are also received by application of dibenzyl dithiocarbamate
(DBDTC) for Mn, Fe, Co, Ni, Cu, Zn, Se, Sb, Hg, Tl, Ag, Cd, and Pb with detection
limits around 1 mg=mL for 100-mL samples (Watanabe et al., 1986). The very low solu-
bility of this reagent also eliminates the need for a metal carrier (Lindner et al., 1978; Ellis
et al., 1982). A detection limit of 10 mg=L could be reached for Mo in water samples by
preconcentration with DBDTC at pH 3 (Saitoh et al., 1984).
1-(2-Pyridylazo)-2-naphthol (PAN) represents another attractive reagent for several
water samples. This organic compound is soluble in hot water and also in ethanol, but not
in cold water. This behavior can be utilized for the separation of special elements. In one
study, 2 L of water were heated to 70 C and a 20-mg portion of PAN, dissolved in ethanol,
was added. After cooling down, the precipitate was filtered off and the obtained filter
target was transferred to the XRF unit. With this procedure, about 15 cations could
successfully be preconcentrated and detection limits about 0.5 mg=L achieved (Vander-
stappen and Van Grieken, 1978).
Other organic reagents, which are less known than the carbamates, are rhodizonate
and tannin. Bauer-Wolf et al. (1993) applied these reagents for the coprecipitation of rare
Copyright © 2002 Marcel Dekker, Inc.
earth elements (REE) in geological samples. The samples were previously decomposed and
the alkaline earth elements removed by an ion-exchange procedure. Subsequently, the
REE were precipitated at pH 13 with rhodizinate and tannin, the residue collected onto a
filter, and analyzed. Tannin, in combination with methyl violet, was used for the de-
termination of Hf, Nb, Ta, and Zr in geological samples after digestion with HCl. The
detection limits of the filtered samples were in the range of 0.1–0.4 mg (Zhang and Ma,
1994). Oxine or 8-hydroxychinoline is a very common reagent for the complexation of
elements in water, but its chelates are much less stable than the carbamate chelates. De-
spite this, it can be successfully applied for the preconcentration of Cu, Fe, Mn, Ni, and
Zn in sea salt. Therefore, the salt is first dissolved in HNO3 and filtered through a
membrane filter. After addition of Al as the carrier element, and oxine and thionalide
solutions, the sample is filtered again and subsequently analyzed. Iwatsuki et al. (1996)
achieved detection limits from 0.01 mg=g for Mn to 0.05 mg=g for Fe with this procedure.
Less efficient than the organic chelates, and therefore less popular, for the applica-
tion with XRF are inorganic reagents (e.g., Fe hydroxide and Al hydroxide). Aluminium
hydroxide might be selected for the coprecipitation of trace metals in environmental
waters. The optimum procedure involves the addition of a 1-mg Al(OH)3 carrier to a 200-
mL water sample at pH 7.3. Quantitative recoveries are obtained for Ti(IV), Cr(II), As(V),
Pb(II), and Th(IV). The detection limits lay in the range between 0.2 and 0.8 mg=L
(Eltayab and Van Grieken, 1992). In one study, a combined technique of organic and
inorganic precipitants was used for the determination of As(III) and As(V) as well as the
total arsenic content in plant material (peach tree). The samples were first treated by acid
digestion and then the As was separated and preconcentrated by DBDTC. Finally, the
arsenic components were precipitated Fe(OH)3. Detection limits lower than 0.1 mg=g and a
precision better than 5% were achieved (Pascual Brugues and Cortazar Castilla, 1995).
Sometimes, other inorganic compounds are also applied for the multielement (co)preci-
pitation and subsequent analysis by XRF; many references of earlier studies can be found
in Van Grieken (1982).
If only a few or even a single element should be separated, reagents, which form
selective complexes of low solubility with the elements of interest, are required. The main
problem, which must be solved, is the prevention of coprecipitation of other, interfering
elements; accurate chemical conditions play an important role for such selective proce-
dures. Tanaka et al. (1987) determined arsenic and antimony quantitatively after their
reduction to As(III) and Sb(III). First, the elements were treated with KBr and HCl at
80 C for 1 h. Both elements were then coprecipitated at pH 9 with a solution of ZrOCl2.
Detection limits of 0.3 mg=L for As and 6.1 mg=L for Sb were achieved. Sometimes, even a
speciation is possible by careful adjustment of the reaction conditions. Trace amounts of
V(V) and V(IV) can be quantitatively recovered from water at pH 1.8 and 4.0, respectively,
by reaction with NaDDTC. However, the concentration of Fe, Co, Ni, Zn, and Pb should
not exceed 100 mg=L in order to avoid interferences and receive a quantitative recovery
(Hirayama and Leyden, 1986). Microgram quantities of Cr(II) and Cr(VI) were success-
fully separated and determined by Peräniemi and Ahlgren (1995). In the first step, Cr(VI)
is isolated from the original sample by collection onto zirconium-activated charcoal at pH
3.9 and the charcoal is filtered. The filtrate is spiked in a second step with FeCl3 and the
pH is increased to 9. Subsequently, the formed precipitate is mixed with activated charcoal
and filtered. Detection limits better than 0.05 mg=L are reached for both species. PdCl2
serves as the precipitation agent for the determination of iodine in urine and water samples
after digestion with chromic acid. The iodate was converted into iodite by Na2SO3 and
finally precipitated as Pbl2 (Mwaura et al., 1994).
Copyright © 2002 Marcel Dekker, Inc.
However, it should always be kept in mind that inferences can occur by the presence of
organic material or high concentrations of alkaline elements. Most of the organic material
should be destroyed by acid treatment, but, often, stable components (e.g., humic sub-
stances) are difficult to digest and still present in the solution. Sometimes, the organic
content is so high that it is recommended to decompose the sample prior to further treat-
ment. For the separation of high alkaline earth element contents, an ion-exchange or
masking step might be useful before further precipitation. Another problem arises from
elements which are present in the sample in different chemical states. Some of the element
species may be incompletely enriched or even escape from the enrichment procedure. Cr3þ,
for example, is collected quantitatively by Fe hydroxide (Osaki et al., 1983; Mullins, 1984;
Ahern et al., 1985; Leyden et al., 1985), whereas Cr6þ escapes as CrO2
4 . The Cr
3þ
species can
be then quantitatively collected by DBDTC or APDC in the presence of Co as the carrier
(Ahern et al., 1985; Leyden et al., 1985). This fact, however, can also be utilized for the
species analysis of Cr. In general, some drawbacks are present for each preconcentration
procedure and therefore a selection of the optimum method requires a careful consideration.
2. Ion-Exchange Methods
The most popular techniques for the enrichment of traces from liquid samples are the
different ion-exchange methods. A very large number of studies dealt with various types of
ion exchangers, applied as resins packed in columns, in liquid form or impregnated onto
filters. Ion-exchange resins are available in the form of macroporous beads, membranes,
impregnates onto filters, or foamed plastics. The analyte, from which the traces should be
separated, can be taken up by different processes. One easy and fast solution is the batch
procedure, where the analyte solution is mixed with the ion exchanger in one container.
The whole solution is filtrated or decanted after some reaction time, and the remaining
residue is washed several times to eliminate possible impurities. The dried powder might be
briquetted for analysis or spread out onto a support material. Another common appli-
cation of ion-exchanger materials is the filtration of a liquid sample through a filter,
previously impregnated with an ion exchanger or the column technique. However, con-
ventional ion exchangers are of less interest for the enrichment of elements from liquid
samples, as they show only limited capacity and are relatively unspecific by the means of
elements. In the presence of high alkali or alkaline earth element content, for example, the
resins can easily be overloaded and most of the required elements pass either through the
column or are only partially enriched.
Despite of these drawbacks, the acid or basic resins are still quite popular for the
enrichment of traces from samples with low alkali and alkaline earth element content.
The most well-known chelating resins are Dowex A-1 and Chelex 100, containing
iminodiacetate functional groups for the complexation. Both have a wide range for the
collection of transition elements, and Co, Cu, Mn, Ni, Pb, Rb, Sr, and Zn are adsorbed
quantitatively from water samples. Chelex 100 also shows a rather high tolerance of alkali
and alkaline earth element content in natural water samples (Florkowski et al., 1976;
Clanet and Deloncle, 1980).
An anion-exchange separation of gold with Dowex 2-X10, after acid digestion of
gold containing ores, enabled the quantitative removal of the interfering elements Hg and
Zn for this application. The analysis of the dried and homogenized resin by XRF, using
the AuLa line for quantification, gave good results (Cojocaru and Spiridon, 1993).
Also, other Dowex resins (e.g. Dowex 168 with 100–200-mesh grain size) are ap-
plied for the separation of rare earth elements from alkaline earth metals in geological
Copyright © 2002 Marcel Dekker, Inc.
samples. The removal of interfering elements, after digestion of the original sample by HF,
HNO3, and HCl, is achieved by mixing the solvent with the anion exchanger. The rare
earth elements are retained on the ion-exchange column while the alkaline earth elements
pass through. The column is washed several times with HNO3 and propanol-2 to remove
residuals and, finally, the rare earth elements are eluted by HCl (Bauer-Wolf et al., 1993).
Rock samples with low concentrations of REE are enriched successfully by using
fibres impregnated with m-acetylchlorophophonazo. The rock sample is first acid digested
and then a masking agent containing EDTA, CDTA and EGTA, and ascorbic acid is
added. Portions of this solutions are allowed to pass though the impregnated fiber in the
form of an ion exchange column. Subsequently, the loaded fiber is digested with HNO3
and HClO4 and, finally, the precipitate is obtained with MgCl2 in aqueous ammonia.
Gong et al., (1995) reached detection limits of 0.05–0.21 mg=g with this procedure.
Speciation and preconcentration can be performed by using a mixture of Ag1-X8 (in
Cl 7 form) and AG5OW-X8 (in Naþ form) and Chelex 100 beads, in order to separate
Cr(III) and Cr (VI) (Prange and Knöchel, 1985).
For a quantitative separation of transition metals from most water samples the ion-
exchange material should show less affinity to alkaline and alkaline earth metals, but
should be selective for the trace metals of interest. Several attempts have been carried out
in this direction (Leyden and Luttrell, 1975; Burba and Lieser, 1977; Ducos-Frede et al.,
1995). Polyamine–polyurea resin columns, for example, were prepared from tetra-
ethyepentamine and toluene diisocyanate for the preconcentration of Ni, Cu, and Zn from
seawater at neutral pH by Leyden and Luttrell (1975). Burba and Lieser (1977) used
Cellulose hyphan effectively as the complexing agent for the enrichment of Cu and U from
seawater, freshwater, and mineral water samples. The reactive agent was received by
immobilization of 1-(2-hydroxyphenylazo)-2-naphtol on cellulose powder and by diazo-
tization of o-aminophenolcellulose and subsequent coupling with b-naphthol.
Uranium traces might be also trapped in another way—by treatment of a Dowex
168 resin with aqueous KMnO4 solutions. In this case, amorphous manganese oxides
were generated in the reticulated structure of the resin. After elution, the sample is pre-
sented on a thin Mylar foil to the XRF unit. Furthermore, the resin has interesting
properties for the separation of inorganic micropollutants, which are frequently present in
water, without enrichment of macrocomponents (Ducos-Frede et al., 1995).
However, all preconcentration procedures based on ion-exchange columns suffer
from several drawbacks. The resins or eluates obtained by the enrichment step need to be
treated further before they can be presented in a suitable form to the spectrometer. Resins
are usually ground, mixed with binder, and pelletized prior to analysis. Eluates must be
fixed by filtration through a membrane or even a coprecipitation step is required. All of
these procedures prolong the time of the sample preparation and include a risk of con-
tamination and losses.
Of much higher interest are filters, which are impregnated with an ion-exchange resin
or specific agent for the enrichment of elements. The loaded filters can be then presented
directly to the spectrometer, and any further sample pretreatment is superfluous. There-
fore, the risk of contaminations and losses can be kept very low and time-consuming
preparation steps are avoided.
AsO33 ; AsO4 ; CrO4 ; MoO4 ; SeO3 ; SeO4 ; VO2 ; and WO4 from different water
3 2 2 2 2 3 2
V. BIOLOGICAL SAMPLES
The amount of publications dealing with the analysis of atmospheric particles by x-ray
fluorescence increased in the last 5 years and showed the suitability of this technique for
such samples. XRS is still the most popular technique for this kind of specimen, especially
for routine analysis, in the field of x-ray analysis. Several studies were also carried out with
TXRF or PIXE, but easy handling of the sample and less effort make the conventional
XRF more attractive, despite its higher detection limits.
Heavy metals are nearly always present in the particulate phase of air and can be
easily separated by filtration or impaction. High enrichment factors are reached by fil-
tration of large air volumes, and homogeneously loaded filters are an ideal target for XRS.
Nevertheless, there are several requirements, which should be considered carefully before
reliable results can be obtained.
The main problems which occur are bound to particle size and x-ray absorption
effects due to the filter media (‘‘filter penetration effect’’). If the particle size distribution of
a particular element is known, the particle size effect can rather easily be corrected.
However, in most cases, the particle size distribution is unknown or difficult to measure. If
this is the case, the particle size effect should be kept low by selection of suitable filter
materials, which collect particles on the surface like Nuclepore or Millipore filters. The
absorption of x-rays by the filter medium depends on the depth distribution of the par-
ticles, which can be calculated approximately by measuring the fluorescence intensities
from the back and front sides of the loaded filter. The mass absorption coefficient of the
particles can be evaluated by transmission measurements and, later, the particle at-
tenuation can be calculated (Adams and Van Grieken, 1975).
If thin membranes, like Nuclepore or Millipore, are selected for the collection, the
absorption effects in the filters are small and often negligible above 5 keV, because,
Copyright © 2002 Marcel Dekker, Inc.
normally, all particles should stay at the filter surface. However, for light elements such as
Si, P, S, and as well as for elements having very small particle sizes, like Pb and S, the
absorption effect needs to be corrected by using a suitable correction procedure.
When choosing thick filters, for the collection of large particle amounts, an easy
and proper way to minimize serious absorption effects is the so-called sandwich geo-
metry (Van Grieken and Adams, 1975). This technique is based on simple inward
folding of the loaded filter to obtain a kind of sandwich. The heterogeneous dis-
tribution of the particles is minimized, despite the higher absorption effect for the filter
in this geometry, as it is assumed that all particles are on the surface and therefore in
the center of the sandwich, which also averages the unknown deposition depth of the
particles.
Both methods (the back–front measurement and sandwich geometry) show accurate
results when the absorption cross section of the filter is not excessively high. The particle
size distribution on the filter surface influences the accuracy of the correction more than
the depth distribution inside the filter (Davis et al., 1977).
For the establishment of filter reference material for aerosol particles, different types
of filter were tested with a high-volume aerosol sampler. As conditions for the reference
material were claimed, low particle size effects, the filter thickness should not exceed
0.5 mg=cm2 and the lateral uniformity of the filter must be within certain limits. Fiber
filters were generally excluded from this study, as they collect particles in their bulk. Only
two types of filter (polycarbonate and cellulose nitrate membranes) fulfilled these condi-
tions and showed an almost uniform deposition onto the surface, which was determined
by analysis of different spots from different places at the loaded surface (Wätjen et al.,
1993).
With filtration of the atmospheric particles, usually all particle sizes present in the
aerosol down to the filter cutoff recorded and particle size effects can easily occur. With
stacked-filter units, operating with different filters of different pore sizes, and dichot-
omous samplers, some kind of size fractionation can be established. With such col-
lection methods, the particle size effects can almost be overcome, as the particles are
divided into two or more fractions. For particles smaller than 2 mm, the particle size
effects are negligible in general, and, therefore, the determination of elements with a low
Z number is also possible. If impactors with several stages are applied for the aerosol
collection, the size distributions of the particles become more uniform and fulfill the
requirements.
Another possibility for overcoming effects bound to different particle sizes and to
obtain reliable results is the analysis of filters as pressed pellets. For this, the loaded
filters are pelletized and sometimes also previously mixed with binder and analyzed
(Vigayan et al., 1997). Coal fly ash, for example, was ground, mixed with poly(vinyl
alcohol) binder, and pelletized (Bettelini and Taina, 1990). In one study, it was shown
that pressing the sample into a pellet after grinding to a powder yields better reprodu-
cibility than applying the sample as a thin film (Matherny and Balgavá, 1993). The
preparation of fusions is another way to eliminate particle size effects (Pella et al., 1978;
Balcerzak, 1993). Good results were obtained by fusion of 70 mg of small dust samples
with lithium fluoride and lithium tetraborate (20:80). To yield a homogenous microbead,
the sample was treated at 1050–1100 C for 7 min in a muffle oven. Subsequently, the
fusion was poured into a specially designed mold and finally analyzed in a normal
sample holder, modified with a spacer in order to fix the sample (Moore, 1993). How-
ever, it should always be considered that each additional sample preparation step might
introduce blank values to the material or dilute the sample. Especially by working in the
Copyright © 2002 Marcel Dekker, Inc.
trace element region, which is usually the case for atmospheric particle samples, this can
cause serious problems.
One point, which should not be underestimated, is the partial or complete loss of
volatile components of the sample. Especially in the wavelength-dispersive mode, voltaile
compounds such as Cl, Br, and, in particular, S may be lost during evacuation of the
sample chamber. This effect is even more severe when working with PIXE. Hereby, the
sample is irradiated with a highly focused charged particle beam under high vacuum and
losses of about 50% of sulfur may occur. In addition to the losses of sulfur due to vo-
latilization, chemical reactions can take place and lead to significant losses of S when
2-MeV protons or 18-MeV a-particles are used (Hansen et al., 1980).
For the collection of aerosols onto filters, a variety of filter materials is available.
Each filter material provides special properties with respect to collection efficiency, me-
chanical stability, hygroscopy, and so forth and every selection is usually a compromise
among the filter properties, the collection purpose, the available costs, and the compat-
ibility with the analytical technique to be applied.
Before choosing a filter material for the collection of atmospheric particles, the blank
values of the material itself must be determined. The filter should consist of materials with
low blank values for the required elements. Furthermore, the spectral background should
be reasonably low in order to allow a reliable trace element determination. Cellulose
membrane filters, for example, especially those made of cellulose nitrate and acetate, are
unsuitable for the determination of P, S, K, and Ca, as they show high blank values of
these elements (Krivan et al., 1990). Very practicable are Nuclepore membranes because
their blank values are very low and the particles are collected only on the surface. Un-
fortunately, the price for these filters is quite high. Whatman 41 filters have a high col-
lection efficiency and are mechanically stable, clean, and reasonable in price. However, the
high hygroscopicity requires a controlled humidity for weighing and handling of the filter.
They also show a much higher background signal in comparison to the Nuclepore
membranes.
The interpretation of the filter efficiency is often complicated, as the test data are
usually based on unloaded or ‘‘clean’’ filters. However, during a sampling event, the filters
change their properties depending on the particle load. With increasing amounts of col-
lected particles, the efficiency increases as well, despite the higher flow resistance of the
loaded filter. In general, the real efficiency of the filter is higher than in the published data.
Furthermore, the collection efficiency increases with higher surface velocity and larger
particle size. For the Whatman 41 filter, for example, the collection efficiency is quite low
for particle sizes of 0.26 mm and surface velocities of 1.5 cm=s. With surface velocities of
100 cm=s, the collection efficiency increases to 95% for both small and large particles. A
10-mm pore-size Teflon filter shows collection efficiencies of 60–90% for low surface ve-
locities and particle sizes between 0.003 and 0.1 mm. For 0.5- and 1-mm pore-size Teflon
filters, the efficiency is always > 99.99%, independent of surface velocity and particle size
(Lippmann, 1978).
Membranes with different pore sizes can be used in stacked-filter units for size-
fractionated collection or with dichotomous samplers.
Quartz-fiber filters present another alternative. They show high collection efficiencies
and are very clean and less hygroscopic than cellulose filters. Particles within the range of
100–0.1 mm are retained on this filter. Furthermore, they are heat resistant. Haupt et al.
(1995) tested different techniques for the preparation of standards for the direct analysis of
quartz-fiber filters. One preparation technique was based on the generation of aerosols and
their deposition onto the filter surface. The other one required spotting of a multielement
Copyright © 2002 Marcel Dekker, Inc.
standard solution onto the filter and subsequent air-drying. The aerosol generation pro-
cedure showed much better accuracy and is, therefore, preferable for the preparation of
standards. Glass-fiber filters are not recommended for trace analysis of inorganic com-
ponents in aerosols. They are much less expensive than the quartz filters, but very high and
variable blank values of diverse elements are their major drawback. However, they are
quite useful for the collection and determination of organic compounds. Sometimes, these
filters are also applied for the analysis of one single component in the aerosol sample.
Lead, for example, can be determined in total suspended air particulates, after collection
onto glass-fiber filters without problems. LaFisca et al. (1991) cut disks of 32-mm diameter
from the filters and irradiated them successfully. Koutrakis et al. (1992) collected house-
hold dust onto Teflon filter with a Harvard impactor. The fine-particle mass was grav-
imetrically determined and, subsequently, the concentrations of elements associated with
the fine mass determined by XRS.
In some cases, filters may also be used for the collection of reactive gases. For this
purpose, the filter is impregnated with special chemicals which retain selected components
of the gases. Hydrogen sulfide, for example, can be trapped by coating the filter with ferric
ion solution. The excess Fe is removed from the filter surface by washing with appropriate
solvents, whereas the Fe sulfide precipitation stays on the filter. Sulfur can be determined
directly or indirectly by measuring the Fe x-radiation. Suitable standards are available for
such measurements (Leyden et al., 1984).
Because XRF is a well-established method, which shows good reliability and easy
handling, the technique is widespread for the determination of aerosols. The use of
radioisotope sources for the excitation makes it independent of the laboratory placement
and suitable for field application. The samples can be analyzed in situ and contaminations
and losses, which might occur during transport and handling, are avoided or kept at a
minimum (Gilfrich and Birks, 1978).
In addition to conventional XRF, TXRF is more frequently applied for the analysis
of aerosols. To utilize the low detection limits and to reduce the matrix effects originating
from the collection material, filter samples need to be digested. With this procedure, ele-
ment concentrations between 0.2 ng=m3 for Cu and 1 ng=m3 for Mn were determined
(Einax et al., 1994; Schmeling and Klockow, 1997). By collection with suitable impactor
systems (e.g., Battelle impactor), the aerosols are separated directly onto the polished
sample carriers required for the instrument, and detection limits of 2 ng for Cr and 15 ng
for Ca were obtained (Injuk and Van Grieken, 1995; Klockenkämper et al., 1995).
In addition to the specimen preparation procedure, the position of the sample inside the
spectrometer plays an important role for a reliable and precise analysis. It should be
ensured that the sample and the calibration standards are in the same form and pos-
ition for analysis, in order to avoid deviations and ensure reliable quantification.
Usually, the sample size is determined by the size of the sample holder, in which the
specimen is presented to the spectrometer. In some cases (e.g., for customized instru-
ments), the sample chamber is adapted for the nondestructive analysis of special objects
like archeological and art specimens. Usually, the sample holders of commercial in-
struments are cylindrical with a diameter of 5.1 cm and allow placing samples with a
maximum thickness of 4 cm. Smaller samples can be fixed by a special mask, which is
inserted to the conventional holder. The bottom of the sample holder is normally
Copyright © 2002 Marcel Dekker, Inc.
Table 4 Degradation Resistance Properties of Selected Thin-Film Materials
covered with a thin film to prevent losses of the sample and contamination of the
spectrometer.
As ideal targets for the XRS analysis, samples prepared as thin films are used. For
infinitely thin samples all interelement and mass absorption effects are negligible. How-
ever, in practice, infinitely thin means that for most x-rays, the sample thickness should be
between 10 and 200 mm, which is difficult to obtain.
Figure 3 Analyte line transmittance for various thin-film substances and thicknesses. (From
Ref. 182. Reprinted by permission of American Laboratory.)
REFERENCES