Spectrochimica Acta Part B 71–72 (2012) 14–23

Contents lists available at SciVerse ScienceDirect

Spectrochimica Acta Part B

journal homepage: www.elsevier.com/locate/sab

High-resolution continuum source electrothermal atomic absorption spectrometry:

Linearization of the calibration curves within a broad concentration range
Dmitri Katskov a,⁎, Miranda Hlongwane a, Uwe Heitmann b, Stefan Florek c
a
Tshwane University of Technology, Chemistry Department, Pretoria 0001, South Africa
b
German Aerospace Center, Rose-Luxemburg Str. 2, 10178 Berlin, Germany
c
ISAS-Leibniz-Institut für Analytische Wissenschaften e.V., Albert-Einstein-Str. 9,12489 Berlin, Germany

a r t i c l e i n f o a b s t r a c t

Available online 2 April 2012 The calculation algorithm suggested provides linearization of the calibration curves in high-resolution
continuum source electrothermal atomic absorption spectrometry. The algorithm is based on the
Keywords: modification of the function wavelength-integrated absorbance vs. concentration of analyte vapor in the
High-resolution continuum source atomic absorption volume. According to the suggested approach, the absorption line is represented by a triangle for
absorption spectrometry low and trapezium for high analyte vapor concentration in the absorption volume. The respective semi-
Electrothermal atomization
empirical formulas include two linearization parameters, which depend on properties of the absorption line
Linearization of calibration curves in HR-CS ET
AAS
and characteristics of the atomizer and spectrometer. The parameters can be approximately evaluated from
the theory and determined in practice from the original broad-range calibration curve. The parameters were
found and the proposed calculation algorithm verified in the experiments on direct determination of Ag, Cd,
Cu, Fe, Mn and Pb in the solutions within a concentration ranges from 0.15 to 625 μg·L− 1 using tube,
platform tube and filter furnace atomizers. The use of various atomizers, lines, elements and atomization
temperatures made possible the simulation of various practical analytical conditions. It was found that the
algorithm and optimal linearization parameters made it possible to obtain for each line and atomizer linear
approximations of the calibration curves within 3–4 orders of magnitude with correlation coefficients close
to 0.999. The algorithm makes possible to employ a single line for the direct element determination over a
broad concentration range. The sources of errors and the possibility of a priori theoretical evaluation of the
linearization parameters are discussed.
© 2012 Elsevier B.V. All rights reserved.

1. Introduction employed for element determination. For the transient absorption

High-resolution continuum source (HR-CS) electrothermal atomic
absorption spectrometry (ET AAS) is recognized to be among the
most powerful analytical tools. Main attributes of the technique are
the possibility of fast sequential multi-element determination due to
the use of a single primary radiation source, perfect background
correction and low limits of detection, similar to or better than those
in traditional ET AAS with line radiation sources. However, although
having much better sensitivity in comparison with inductively
coupled plasma optical emission spectrometry (ICP OES) or flame
HR-CS AAS, it yields to these methods with respect of working range.
The reason for this limitation is a non-linear function of absorbance
vs. the amount of free analyte atoms in the absorption volume for
high concentrations of atomic vapor. This factor has little or no
importance for flame AAS since a non-linear calibration curve of
absorbance vs. concentration of the analyte in the solution can be
⁎ Corresponding author. Fax: +27 12 382 6286.
E-mail address: katskovda@tut.ac.za (D. Katskov).
signals in ET AAS, however, the limitation is critical. Since the
quantification of the analyte in ET AAS is performed using temporal
integration of the absorption pulse, non-linearity of the function
absorbance vs. content of analyte vapor in the atomizer causes deter-
mination error. The error depends on the temporal profile of the
absorption signal, which is affected by the design of the atomizer, the
temperature program and the thermal properties and mass of the
analyte and matrix. Thus, dilution of the samples is to be applied for
the quantification of high analyte content. However, HR-CS ET AAS,
compared to line source ET AAS, provides an advantage of sensitivity
reduction by using the measurement at the wings of specific absorption
line or various lines [1]. Both the dilution of the sample and selection
of proper wavelength for the measurements according to the concen-
tration of the analyte increase the total analysis time, especially when
the samples contain various analytes in a broad concentration range, the
element to be determined has a limited number of analytical lines or the
sample is to be analyzed as solid or slurry.
The objective of this work is to find out the pattern of formation of
absorption signals for high concentrations of analyte atoms in the
absorption volume and to employ the findings for HR-CS ET AAS data

0584-8547/$ – see front matter © 2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.sab.2012.03.007
 D. Katskov et al. / Spectrochimica Acta Part B 71–72 (2012) 14–23
quantification within a broad concentration range of the analyte. It is
expected that the search for an algorithm would allow using the most
sensitive analytical lines or those less prone to interferences, indepen-
dent of the concentration of the respective analyte in the sample.
2. The algorithm
2.1. Theory
The basic issues of the theory of continuum source atomic
absorption measurements relevant to this work were taken from
Refs. [1–5].
Pulse sample vaporization in the atomizer leads to appearance of
free atoms of the analyte in the absorption volume that causes
absorption of radiation from the primary light source. For low analyte
vapor concentrations, the absorbance at wavelength λ is described by
equation
At ðλÞ ¼ Log½I0 ðλÞ=It ðλÞ; ð1Þ
where It(λ) and I0(λ) are radiation fluxes from the primary source
passing through the atomizer with and without absorption, respectively.
The concentration of atoms in the absorption volume, nt (mol·m− 3), is
characterized by the wavelength-integrated absorbance within the line
profile,
St ¼ ∫ At ðλÞdλ ¼ γ·nt ; ð2Þ
λ Regarding possible extension of the working range, two calibration
where γ (m 4·mol − 1) is a coefficient, which characterizes the sensitivity
of the absorption line. For very high vapor concentrations, when full
absorption occurs in the line center and its further increase comes about
only on account of the line wings, the theory predicts,
0:5
St ∝ðγ·nt Þ : ð3Þ
There is no analytical expression proposed, which could describe
the function St vs. nt within the interim concentration range [1,5–7].
Atomic absorption is detected using a spectral instrument having a
transmittance profile F(λ). The wavelength distribution of the measured
absorbance, A′t(λ), is determined by convolution of the original line
profile At (λ) at given experimental conditions and the instrumental
transmittance profile [4],
∞ proposed theory the vaporization pattern of the analyte was considered
A′t ðλÞ ¼ ∫ F ðλ′ÞAt ðλ−λ′Þdλ′: ð4Þ
−∞
According to Eq. (4), the measured absorbance A′t(λ) depends on the
transmittance profile of the instrument: The highest output absorbance
can be obtained if the full width at half maximum (FWHM) of the
instrumental profile, δλF, is smaller than that of the original absorption
line, δλA, then
A′t ðλÞ≅At ðλÞ: ð5Þ
For a much broader instrumental profile,
∞ proposed in this work to introduce automatic correction in the running
A′t ðλÞ≅F ðλÞ· ∫ At ðλÞdλ′ ¼ F ðλÞ·St ; ð6Þ
−∞
where wavelength-integrated absorbance,
St ¼ ∫ At ðλÞdλ; ð7Þ
λ emission, which overlaps with the primary radiation passing through
15
is invariant respective to instrumental broadening of the original line
profile. If absorbance meets condition (2), full quantity of the analyte
in the sample (mol) can be characterized using temporal integration
∞ tr
N ¼ k· ∫ nt dt ¼ ðk=γÞ·Q ¼ ðk=γ Þ· ∫ St dt: ð8Þ
0 0
where k is the vapor transportation rate, (m3·s− 1), (considered invariant
towards mass of the analyte), tr is the registration time including the
vaporization and transport processes and Q is the time- and wavelength-
integrated absorbance.
For the instrument with broader line profile, according to Eq. (6),
absorbance at any wavelength within instrumental contour can be
employed for the quantification of the element, e.g. absorbance in the
line center, λ0
tr
N∝ ∫ A′t ðλ0 Þdt: ð9Þ
0
The realization of Eqs. (8) and (9) in practice, however, meets
limitations for high concentrations of the analyte atoms in the gas
phase: absorbance in the line center, A′t(λ0), above some limit reaches
saturation; further increase of wavelength integrated absorbance St
occurs only on account of broadening of the line contour. Respectively,
time- and wavelength-integrated absorbance Q in Eq. (8), in general,
includes under the integral functions (2) and (3), with partial
contribution depending on the absorbance magnitude and time.
options, wavelength-selected absorbance and wavelength-integrated
absorbance, were considered in Refs. [6,7]. In the first method (later
thoroughly investigated in Ref. [8]) the absorbance corresponding to
the CCD elements (pixels) at some distance from the center of the
absorption line was employed; calibration of the data related to specific
pixels was needed similar to that for different absorption lines. Prospects
of the second method were deemed more promising by the authors [6,7]
since the character of functions St vs. nt (Eqs. (2) and (3)) was known
and, hence, could be used for calibration of analytical Q(N) measure-
ments. A special calibration algorithm was proposed in which the value
Q was considered proportional to N or (N)0.5 for low and high sample
masses, respectively. The associated calibration curves (wavelength-
and time-integrated absorbance vs. mass of the analyte) were approx-
imated by a general equation representing the rotated half of a
hyperbola Ax2 + Bxy + Cy2 + Dx + F, where the coefficients A, B, C and
D were similar while F was specific for different elements. In the
not depending on the sample mass, which permitted constructing the
calibration curves within a broad concentration range. The condition,
however, contradicts to practice since transient absorption signals
normally can differ with magnitude and duration depending not only
on the quantity of the analyte but on the nature of the sample matrix,
presence of a chemical modifier, the temperature program or other
experimental conditions. Since it is hardly possible to provide a similar
vaporization pattern of the analyte in realistic samples and standards
over a broad concentration range, the method cannot be considered
feasible for practice.
Instead of calibration of direct measurements of time- and
wavelength-integrated absorbance, as suggested in Ref. [6,7], it is
absorbance, which after temporal integration should provide a linear
calibration curve independent of the vaporization pattern. To find out
the correction algorithm, the nature of absorption saturation is to be
understood and a respective approximation proposed for wavelength-
integrated absorbance including the range before and after saturation.
Atomic absorption in the atomizer is accompanied by resonance
16 D. Katskov et al. / Spectrochimica Acta Part B 71–72 (2012) 14–23
the atomizer. Contribution of emission causes distortion of the absor-
bance: instead of Eq. (1), the absorption line profile is to be described by
the equation
At ðλÞ ¼ Logf½I 0 ðλÞ þ Ire ðλÞ=It ðλÞ þ I re ðλÞg;
where Ire(λ) is the flux of the secondary emission entering the
spectrometer within the spatial angle determined by the primary light
beam through the atomizer and the concentration of the light absorbing
species. In the tube atomizer of length l and internal diameter d in
absence of quenching processes full absorption of the primary radiation
in the line center (It(λ0) = 0) should initiate an emission signal from the
center of the tube,


2 2
Ire ðλ0 Þ ¼ I0 ðλ0 Þ·d = 4·l ;
overlapping the primary radiation. Substitution of Eq. (11) in Eq. (10)
gives the absorbance saturation level As. Further increase of concentra-
tion of atomic vapor should cause spreading of saturation beyond the
line center.
For absorbance At(λ) b As, wavelength-integrated absorbance St is
proportional to the respective amount of analyte atoms, independent
of the line profile. This permits to render function At(λ) as a triangle
with base bt and base angle β (Fig. 1, a). Integration of Eq. (2)
provides the area of the triangle,
2
γ·nt ¼ At ðλ0 Þ·bt =2 ¼ ðbt Þ ·ðtg βÞ=4:
In Eq. (12) both parameters At(λ0) and bt increase together with nt
until At(λ0) reaches the saturation level As and bt/2 becomes equal to
As/(tg β). The angle β here is conditionally defined as for the triangle
with base equal to 2∙δλA and height of absorbance equal to unity, tg
β = 1/δλA. Further increase of concentration of atoms in the absorption
volume according to the suggested model causes transformation of the
triangle into a trapezium shaped area
St ¼ As ·bt −As ·δλA ;
After substituting parameter bt found from Eq. (12) in Eq. (13) respective
concentration of analyte atoms nt can be found,
2
γ·nt ¼ ðSt þ As ·δλA Þ =ð4·As ·δλA Þ:
Combination of Eqs. (13) and (14) provides the general expression:
h i
 2
γ·nt ¼ St ¼ St −0:5· St −ðSt þ C 1 Þ =ð4·C 1 Þ ·½1 þ SignðSt –C 1 Þ;
Fig. 1. The approximation of absorption line contour (a) before and (b) after instrumental broadening.
where
C 1 ¼ As ·δλA ð16Þ
ð10Þ and the function Sign (St −C1) is equal to 1 or −1 depending on the sign
of the argument. Thus, if the proper constant C1 is found, the linearized
function S⁎t vs. nt can be employed in temporal integration (8) instead of
St,
tr

N ¼ ðk=γÞ·U ¼ ðk=γ Þ· ∫ St dt; ð17Þ
0
where U is the time integral of the wavelength-integrated absorption
ð11Þ after correction (15). According to Eqs. (10) and (11), for the graphite
tube atomizers normally employed in ET AAS (e.g., having length and
diameter close to 20 and 6 mm, respectively [9]) and an optical design
providing focusing of the primary light source in the center of the tube,
contribution of the secondary emission can reach to about 2.25% of the
primary radiation. The estimation gives a saturation level As equal to 1.6.
Increase of vapor concentration in the absorption volume should cause
formation of the spatial light source and, respectively, increase of
emission output and reduction of the saturation level.
The parameter δλA in Eq. (16) characterizes the dependence of
line broadening on experimental conditions. In ET AAS absorption
line width is affected by a number of factors, of which Doppler and
collisional broadening, caused by motion of analyte atoms and their
ð12Þ interaction with particles of ambient atmosphere, are dominating. Both
types of broadening are described by formulas including wavelength,
atom mass, temperature and collision cross-section (see, e.g. [1]).
Superposition of all contributing factors causes the absorption profile
described by a Voigt function. In the suggested model the range of
variations for the Voigt profiles, δλv, is considered similar to that for
δλA. The examples of FWHM of Voigt profiles calculated for several
elements investigated in this work in the temperature range charac-
teristic for ET atomization are presented in Table 1. According to the
calculations, the difference between FWHM for various lines can reach
ð13Þ a factor of 2.75, while the deviations related to temperature do not
exceed 18–19%.
The influence of the transmittance profile on the absorbance
measurements can be evaluated for each particular δλA/δλF combina-
tion using the simplifying suggestion that both, initial absorbance
ð14Þ contour and instrumental profile are described by Gaussian functions.
In this case FWHM of the absorption profile broadened by instrument
line (δλA′) can be determined from the equation [4],
ð15Þ 2 2
δλA’ ¼ δλv þ δλF ;
2
ð18Þ
 D. Katskov et al. / Spectrochimica Acta Part B 71–72 (2012) 14–23 17

Table 1
Comparison of the FWHM for the absorption lines approximated by Voigt profile, before and after instrumental broadening in the HR-CS spectrometer.

Element Wavelength/nm δλF/pm Temperature/K; δλv, δλA′/pm

2800 2400 2000

δλv δλA′ δλv δλA′ δλv δλA′

Ag 328.068 4.41 1.21 4.57 1.12 4.54 1.02 4.52

Cd 228.802 3.08 0.82 3.19 0.76 3.17 0.71 3.16
Cu 324.754 4.36 1.55 4.63 1.44 4.59 1.32 4.55
Fe 248.327 3.34 1.26 3.57 1.17 3.54 1.07 3.51
Mn 279.482 3.75 1.43 4.02 1.33 3.98 1.22 3.95
Pb 217.001 2.92 0.57 2.98 0.53 2.97 0.48 2.96
Pb 283.306 3.81 0.75 3.88 0.71 3.87 0.64 3.86

and the respective absorption maximum for the broadened line profile where W is the time integral of wavelength-integrated and twice
found, corrected absorbance, within a broad range of analyte masses.

A′t ðλ0 Þ ¼ At ðλ0 Þ·½δλv =δλA′ : ð19Þ 2.2. Calculation methodology

The data on Voigt profiles are compared in Table 1 with FWHM of To perform determinations in a broad concentration range using a
the instrumental contour of the HR-CS ET AAS instrument described in spectral instrument with CCD detector, instant values of wavelength-
[1]. Linear interpolation between the indicated δλF values 2.7 and integrated absorbance corresponding to sequential ti moments are to
6.7 pm for wavelengths 200 and 500 nm, respectively, is taken into be measured,
account. The data regarding to δλv and δλA′ show that the instrumental
pX
0 þθ
broadening should cause a factor of 3 to 6 reduction of absorbance Si ¼ Δp ⋅ Ap;i ; ð23Þ
compared to the detection using an ideal narrow-band instrument. p¼p0 −θ
After the broadening the difference in FWHM for various elements
turns out to be reduced by a factor of 1.5, and for the same lines with where Δp is the spectral interval related to a single CCD element
different temperatures by 1–1.5%. (pixel), p0 is the pixel corresponding to the line center, p0 ± θ is the
Taking into consideration Eqs. (15) and (19), the absorption profile interval of 2 ∙θ + 1 pixels, sufficient for registering the line contour.
after exposure to the instrumental broadening can also be simulated by Next, the temporal profile of the wavelength-integrated absorbance St
a triangle, A′t(λ), with reduced maximum and respectively enlarged should be registered in digital format as a set of Si data. Summation of
base (Fig. 1b), with an area equal to S⁎t from Eq. (15). In practice the Si data would provide a calibration curve
opportunities to measure high absorbance A′t(λ) are limited by stray
light, which is detected together with the primary radiation in the X
t r =i
selected spectral interval [1,5–7]. Similar to the secondary emission in Q ðN Þ ¼ Δi⋅ Si ; ð24Þ
Eq. (10), the presence of stray light causes distortion of the A′t(λ) profile i¼0

that eventually results in saturation of the absorption at A′s level. After

the saturation, further increase of the analyte vapor concentration (Δi is the temporal interval of data collection).
causes an increase of the wavelength-integrated absorbance only on If the linearization parameters C1 and C2 are known, first and
account of the line wings. This puts a limitation to the use of Eq. (15) for second correction to each Si value can be sequentially introduced
the linearization of St vs. nt function. according to Eqs. (15) and (20) and, respectively, summarized within
The approximation of the broadened contour A′t(λ) by triangle tr time that gives the intermediate,
and trapezium for low and high concentrations of atomic vapors,
respectively, (Fig. 1b) leads to the expression similar to Eq. (15), X
t r =i

U ðNÞ ¼ Δi· Si ð25Þ
h i  i¼0
     2   
γ·nt ¼St ¼ St −0:5· St − St þC 2 =ð4·C 2 Þ · 1þSign St –C 2 ; ð20Þ
and the final calibration curve,

where X
t r =i

W ðNÞ ¼ Δi· Si : ð26Þ
C 2 ¼ A′s ·δλA′ ð21Þ i¼0

depends on the transmittance profile of the spectrometer and the The main problem regarding practical realization of the approach
level of stray light. is related to the selection of proper C1 and C2 parameters defined in
According to estimations [5,6], stray light as 3–5% of primary Eqs. (16) and (21). Since the values of the constituents of the
radiation is to be considered that causes saturation of absorbance at a respective formulas are not exactly quantified by the theory, it makes
level of A′s equal to1.3–1.5. Taking this and δλA′ values from Table 1, the sense to determine them experimentally for a particular instrument,
parameter C2 can be evaluated and employed in Eq. (20). Finally, atomizer, experimental conditions and analyte. The selection is to be
wavelength-integrated absorption S⁎⁎ t is to be used in the temporal performed by variation of the C1 and C2 parameters around the values
integration (8) instead of St, which should provide a linear calibration estimated from the theory. The appropriate C1 constant should
curve provide a calibration curve (25) having an extended linear part in
comparison with Eq. (24) and a similar slope. The correct parameter
tr C2 should provide an additional linear part to the function (25). It

N ¼ ðk=γ Þ·W ¼ ðk=γÞ· ∫ St dt; ð22Þ is expected that C1 and C2 parameters thus justified for specific
0 experimental conditions, besides practical application, could provide
18 D. Katskov et al. / Spectrochimica Acta Part B 71–72 (2012) 14–23
the basis for comparison with respective theoretical values and
generalization of the model and calculation algorithm.
3. Experimental
Experimental investigation of the linearization algorithm was based
on HR-CS AAS measurements in three different atomizers for the group
of metals listed in Table 1. The choice of elements and atomizers was
expected to provide broad spectrum of variables regarding physical and
thermal properties of the analytes, atomization kinetics and wavelength
range, which could affect the measurements and calculations.
3.1. Instrumentation
The prototype HR-CS AAS instrument [1] (built at ISAS — Leibnitz-
Institut für Analytische Wissenschaften, Berlin, Germany) used for the
investigations was based on a Model AAS 6 Vario system (Analytik Jena
AG, Jena, Germany), from which the entire optical compartment
including detector and associated controls had been removed and
replaced by a high-resolution double monochromator (DEMON), similar
to the system described in Ref. [10]. The instrument, which utilizes a
300 W high-intensity xenon short-arc lamp (XBO 301, GLE, Berlin,
Germany) as a continuum source, and a linear CCD array detector
(S 7031-0906, Hamamatsu, Japan) with 512 pixels, offers high resolv-
ing power and linear dispersion, resulting in a spectral bandwidth of
λ/140 000 per pixel. It allows the simultaneous detection of arbitrary
wavelength intervals up to 1 nm within the total spectral range from
189 nm to 900 nm. In addition, the instrument is equipped with the
original electrothermal atomizer and autosampler (both from Analytik
Jena). Three types of transverse heated atomizers were employed in the
experiments: graphite tube 20 and 6 mm in length and diameter,
respectively; platform tube of similar to the tube geometry and filter
furnace [11] with central channel (absorption volume) 11 and 2.5 mm in
length and diameter, respectively.
3.2. Measurements
Atomic absorption measurements were performed for Ag, Cd, Cu,
Fe, Mn and Pb using solutions with concentrations ranging from 0.15
to 625 μg·L − 1 each. The upper range was limited at 625 μg·L − 1 to
avoid excessive deposition of the elements on the walls of the furnace
housing and possible memory effects. Each set of samples was
prepared from 1000 mg·L − 1 stock solutions (Merck, South Africa) by
sequential dilution with water containing 0.2% of nitric acid. For each
measurement a volume of 20 μL of the analyte solution was injected into
the atomizer. Each set of samples was employed for the measurements
with all three atomizers, different lines of the same element (for Pb
and Mn) or temperature settings (for Ag). The measurements were
performed in the sequence from low to high concentration, which
reduced possible errors due to memory effects. The data are presented as
the calibration curves W=γ′·m, where m is mass concentration of the
analyte in the solution (μg·L− 1) proportional to N in Eqs. (8), (17) and
(22), and time- and wavelength-integrated absorbance W is expressed
in pm·s units.
After selection of the analytical line the solutions were injected in
the atomizers and the temperature program shown in Table 2,
executed. Wavelength-integrated absorbance Si (pm) was calculated
according to Eq. (23) by summation of the absorbance data for 19
CCD pixels multiplied by Δp, equal to 2 pm. Each measurement was
performed in triplicate and the respective data averaged. Upon
completion of the measurements in a particular atomizer for the full
concentration range, the Si data for each concentration were transferred
to an Excel worksheet thus forming a data matrix concentration
(columns) vs. time (rows). Altogether the data could be presented as a
set of St curves specific for each element and atomizer. Examples of St
curves are shown in Fig. 2, a–c for Ag, Pb and Mn, respectively.
Table 2
Furnace temperature program.
Stage Temperature/°C Temperature Hold time/s Internal
ramp/°C s− 1 gas flow
1 130 10 15 On
2 180 10 5 On
3 700 100 15 On
4 701 100 1 On
5 1600–2600a 3000 5–8 Off
6 1650–2650b 200 5 On
a
According to the element and atomizer, see Table 3.
b
Cleaning temperature, 50 °C above that used in the previous stage.
3.3. Calculation of the linearization parameters
The Si values from each column of the data matrix were summarized,
and the calibration curves representing time- and wavelength-
integrated absorbance, Q, vs. concentration of the analyte, m, in the
injected solution were plotted. The example of calibration curves for
Ag in three atomizers corresponding to the experimental data of Fig. 2a,
is presented in Fig. 3. The linear part of each graph for low analyte
concentrations (enlargement in the top corner of Fig. 3) was employed
for sensitivity determination: the linear regression trend line Q = γ′∙m
and the respective correlation coefficient R 2were found. The range of
Q(m) data was taken as broad as possible within the linear trend.
Including the data beyond the linearity range caused reduction of both
slope of the trend line and of correlation coefficient.
At the next stage of calculations each component of data matrix Si
was introduced in Eq. (15), where an arbitrary value of the parameter
C1 (Eq. (16)) was selected within a reasonable range of variations of
the constituents As (e.g. 1.0–1.6) and δλA (e.g.0.5–1.5). Thus, a
provisional S⁎i matrix was built, integrated values U(m) for each
column were found, function Log U vs. Log m was plotted and, finally,
the linear regression trend line,
Log U ¼ x·Log m þ Log γ′; ð27Þ
and the respective correlation coefficient were visualized in the
respective chart. The change of the linearization parameter C1 automat-
ically transformed the plot, slope of the trend line and correlation
coefficient, which made the optimization easy. In Fig. 4 the transforma-
tion of the initial calibration curve (dotted line) is shown for various C1
parameters for the example of Ag in the platform furnace. The variations
did not affect parts of the graphs for low concentration; the optimal
C1 parameter provided a coefficient x in Eq. (27) and a correlation
coefficient close to unity for an as broad as possible concentration range.
The correct data are indicated in Fig. 4 by squares; parameters x = 0.9995
and R2 =0.9999 of the linear regression (27) are close to unity. The range
of the linear part of the U(m) function increased compared to Q(m) by
about a factor of 10.
At the final stage of the calculations each component of the data
matrix S⁎i was introduced in Eq. (20), where a parameter C2, 3–5 times
larger than C1, has been selected; thus, data matrix S⁎⁎ i was built and
integrated values W(m) were found. The optimization of parameter
C2 was performed similar to that already described for C1. The
example of the linearization results is shown in Fig. 5 for Mn in the
platform furnace. The arrows in Fig. 5b indicate the limits for the
linear parts of Q(m) and U(m) graphs. A full list of correction
parameters for the elements investigated and the experimental
conditions is presented in Table 3.
4. Results and discussion
The use of different atomizers in the linearization experiments
was suggested in order to imitate diverse vaporization kinetics for the
 D. Katskov et al. / Spectrochimica Acta Part B 71–72 (2012) 14–23 19

Time- and wavelength-integrated absorbance /pm·s

a. Ag 328.068 nm
10 8
Wavelength-integrated absorbance /pm

0.4
9 1
0.3
7
0.2
8 3
0.1
6 0
7 0 5
2 Ag 328.068 nm
6 5

5 3 4
2
4
3
3
1
2
2
1
1

0 0
0 0.5 1 1.5 2 2.5 3 0 100 200 300 400 500 600 700
Time /s

b. Pb 283.306 nm Fig. 3. Calibration curves for Ag corresponding to the sets of atomic absorption signals
7 in Fig. 2a for the (1) tube, (2) platform and (3) filter furnaces. The graphs in the upper
corner are related to the low concentrations of the analyte.
Wavelength-integrated absorbance /pm

1
6 3

the much faster signals in the tube atomizer (Fig. 3). For some metals,
5
2 shift of the signal maximums in the tube and platform furnaces
towards longer vaporization time is seen for high concentrations (e. g.
4
the shift relative to the dashed lines in Fig. 2a for Ag). No signs of
memory effects associated with high concentrations of the analytes
3
were observed, which indicates the possibility for further extension
of the calibration curves.
2
The suggested model helped to define the semi-empirical formulas,
calculation algorithm and parameters for the linearization of the
1
calibration curves within a broad concentration range. The data in
Table 3 show that optimal linearization parameters made it possible to
0
0 0.5 1 1.5 2 2.5 3 3.5 extend the linear part of the calibration curves for various elements and
Time /s atomizers by factors of 20 to 450 with correlation coefficients close to
0.999.
c. Mn 279.482 nm As expected, the linearization parameters differed for the
9 atomizers and elements investigated. However, the variations found
Wavelength-integrated absorbance /pm

8
could be associated either with objective factors (various FWHM of
1
absorption lines, difference in temperature of the atomic vapor, in
7 2 secondary resonance emission or dependence of stray light on the
6 radiation from the tube walls) or with experimental errors in the
measurement of low absorption signals (Q vs. m, as in the top corner
5 3
Log(Time- and wavelength-integrated absorbance

3 1
1

2 Ag 328.068 nm 2
0.5 3
4
1
5
0 0
0 1 2 3 4 5 6 -1 -0.5 0 0.5 1 1.5 2
/pm·s)

Time /s
-0.5
Fig. 2. Atomic absorption signals for (a) Ag, (b) Pb and (c) Mn in the (1) tube, (2)
platform and (3) filter furnaces, respectively, for a set of solutions from 0.15 to -1
625 μg L− 1.

-1.5

same elements, and to provide various sensitivity and curvature of

the calibration curves for the same absorption lines. The graphs in
Figs. 2 and 3, and the sensitivity data in Table 3 confirm the approach:
the atomization peaks are substantially different; each set of curves in
Fig. 2 is characterized by a specific behavior; the highest sensitivity is
characteristic for slow signals in the filter furnace and the lowest for
-2
Fig. 4. Linearization of the calibration curve for Ag in the platform furnace in Log–Log
coordinates: optimization of parameter C1 (Eq. (15)). 1–5: the curves, the respective
linear regressions and correlation coefficients for various C1 parameters.
20
a strongly affected by experimental conditions within a range typical for
Time- and wavelength-integrated absorbance /pm·s
1000
Mn 279.482 nm
900
800
700
600
500
400
300
200
100
0
0 100 200
b instrument transmittance profile.
Log(Time- and wavelength-integrated absorbance
2 correlation between the first linearization parameter (Table 3) and
LogW= 1.0098·Logm-1.12; R² = 0.9995
1.5
1
0.5
/pm·s)
0 The range of application of the linearization algorithm depends on
-1 -0.5 0 0.5
-0.5
-1
-1.5
-2
Fig. 5. Calibration graphs for Mn in a platform furnace in (a) linear and (b) Log–Log
coordinates: original curve (1) and those after first (2) and second (3) step of
linearization according to Eqs. (15) and (20), respectively. The arrows indicate the
upper linearity range for the curves 1 and 2.
of Fig. 3). The errors in sensitivity evaluation for low signals could
result in a change of the respective correction parameters. It is shown
in Table 4 at the example of Mn at 279.842 nm in the platform furnace
that ±5% sensitivity variation could be associated with 20–30% and
40–50% changes of the linearization parameters C1 and C2, respec-
tively. On the contrary, change of both parameters within the range of
deviations measured for various atomizers has relatively low effect on
the respective regression graphs. This issue is illustrated by the data
from Table 5, where the sensitivities obtained in the measurements
for each atomizer are compared with those calculated using the
linearization parameters averaged for all three (two for Cd and two
for Ag at 1800 °C) atomizers. It is seen that for 85% of the data the
general slope of the calibration curves in the Log–Log coordinates
deviates from unity not more than by 5% with a correlation coefficient
close to 0.999. The higher deviations are accompanied by a sensitivity
change and reduction of the correlation coefficients for the respective
linear regressions. Both types of deviations look random and, most
probably, can be attributed to the experimental errors in the sensitivity
measurements for low analyte concentrations. Thus, it can be concluded
that, in spite of substantial changes in the vaporization kinetics and
sensitivity (e.g. shown in Figs. 2 and 3), the combination of the
linearization parameters C1 and C2 for each individual line is not
D. Katskov et al. / Spectrochimica Acta Part B 71–72 (2012) 14–23
ET AAS measurements. This gives a hope to use once determined
parameters in practical applications.
The measurement accuracy in this work and simplifications in the
theory could not yet guarantee the precision necessary for the
attribution of the linearization parameters to particular physical or
W= 0.0817·m; R² = 0.9995
3 technical factors. In the theory, the attribution of FWHM of the
absorption line δλA, to the triangle with the height At(λ0) equal to
unity was arbitrary. In the experiments, similar program settings for
different atomizers could not provide exactly equal temperature of
the atomic vapor.
2 In spite of the limitations, some tendencies regarding conformity
of the measurements to the theory could be observed. Under the
admission that different atomizers provide more or less similar level
1 of secondary emission and stray light, the respective linearization
parameters for specific lines were to be close. Only slight variation of C1
parameter for a specific element could be caused by the dependence of
300 400 500 600 700 the Voigt line profile on temperature (Table 1), while for various
elements the difference in C1 values could reach a factor of 2–3. The C2
to C1 ratio should vary by a factor of 3–5 according to the line width and
Similar to the predicted tendency for each line, a certain degree of
3 FWHM of the Voigt profiles (Table 1) for different elements can be
observed in Fig. 6. The difference between C1 parameters for the same
Mn 279.482 nm
2 line in three atomizers is less than 20–30%, while it can reach a factor
of 2–3 for different lines. The average of the C2 to C1 ratio for the
1 atomizers found for different lines ranges between 2 and 5.7, which is
also in conformity with the theory.
1 1.5 2 2.5 3 the spectral resolution of the continuum source instrument. Accord-
ing to Eq. (2), in a hypothetical narrow band instrument (δλF b bδλA),
the model attributes low vapor concentration nt to the measured
absorbance as A′t(λ0) ∝ (nt) 0.5. Although highest sensitivity could be
obtained in such instrument, the calibration curve Q(N) cannot
contain any linear part. Contrary, the instrumental broadening of the
absorption line eventually leads to the reduction of the measured
value A′t(λ0) that is accompanied by an increasing weight of function
(6) or, in other words, by appearance of a linear part of the calibration
curve Q(N).
On the other hand, significant reduction of the measured absor-
bance also puts restrictions to the application of the algorithm for
instruments with broad instrumental profile. This case is highlighted
for a low-resolution continuum source (LR-CS) instrument (FWHM of
instrumental profile about 200 pm) employed in [12] for simultaneous
multi-element determinations. Although the wavelength-integrated
absorbance St in a LR-CS instrument theoretically was not affected by
instrumental broadening, absorbance magnitude A′t(λ0) in the line
center turned out to be about 100 times lower than that in the HR-CS
spectrometer used in this work. Correspondingly poor limits of detection
for lower and distortion of the line profile for higher vapor concentra-
tions made the linear part of St vs. nt function in Eq. (15) narrow (or
practically non-existing). In fact, only signals corresponding to high nt
and, therefore, to the case St > C1 in Eq. (15) could be measured, which
led to the particular case of Eq. (14) or close to the theoretical prediction
(3). It made possible simple squaring of current A′t(λ0) values in the
algorithm applied in simultaneous multi-element determination [12].
Discussing the suggested calculation algorithm, it should be noted
that secondary resonance emission had not been earlier listed among
the factors affecting saturation of the absorption lines. However, the
estimation shows that contribution of this factor should be substantial
and different for various atomizers and resonance lines; the emitting
volume can change its configuration along with the concentration of the
atomic vapor and the geometry of the primary light beam. Excluding of
emission from the list of factors, which control absorbance saturation,
makes vulnerable the conclusions [7] regarding direct dependence of
 D. Katskov et al. / Spectrochimica Acta Part B 71–72 (2012) 14–23 21

Table 3
Experimental conditions, linearization parameters and characteristics of the calibration curves for the element determination with the HR-CS ET AAS instrument.

Element, Atomizer Temperature Calibration curves W = γ′∙m

line/nm setting/°C
Linearization Sensitivity, Linearity Correlation
parameters in (15), γ′/pm∙s∙L∙μg− 1 rangea/μg·L− 1 coefficient R2
(20)/pm

C1 C2

Pb Platform tube 1800 – – 0.014 30 0.994

217.001 0.39 – 0.014 80 0.999
0.39 2.3 0.014 625 0.999
Filter furnace 2000 – – 0.018 30 0.980
0.46 – 0.018 80 0.998
0.46 1.9 0.018 625 0.999
Tube 1800 – – 0.011 20 0.993
0.5 – 0.011 50 0.996
0.5 1.7 0.011 625 0.999
Cdb Platform tube 1600 – – 0.081 1.2 0.992
228.802 0.21 – 0.081 10 0.999
0.21 3.6 0.082 40 0.999
Filter furnace 1800 – – 0.149 1.4 0.994
0.29 – 0.147 20 0.999
0.29 3.2 0.142 625 0.998
Mn Platform tube 2400 – – 0.074 3 0.997
279.482 0.46 – 0.075 20 0.999
0.46 2.3 0.074 625 0.999
Filter furnace 2600 – 3.2 0.072 10 0.997
0.34 0.072 40 0.999
0.34 0.072 625 0.999
Tube 2400 – – 0.047 5 0.999
0.54 – 0.047 20 0.999
0.54 1.9 0.047 625 0.999
Mn Platform tube 2400 – – 0.055 5 0.995
279.827 0.5 – 0.055 40 0.998
0.5 2.7 0.054 625 0.999
Filter furnace 2600 – – 0.060 10 0.997
0.38 – 0.060 80 0.998
0.38 3.0 0.059 625 0.998
Tube 2400 – – 0.038 5 0.995
0.54 – 0.038 10 0.997
0.54 2.3 0.037 625 0.999
Pb Platform 1800 – – 0.011 15 0.998
283.306 0.56 – 0.011 80 0.999
0.56 5.0 0.011 625 0.999
Filter furnace 2000 – – 0.018 15 0.996
0.56 – 0.018 80 0.998
0.56 6.0 0.018 625 0.999
Tube 1800 – – 0.008 15 0.998
0.62 – 0.008 80 0.999
0.62 6.0 0.008 625 0.998
Fe Platform tube 2300 – – 0.025 20 0.995
248.327 0.29 – 0.025 150 0.998
0.29 2.7 0.025 625 0.999
Filter furnace 2500 – – 0.021 40 0.994
0.22 – 0.021 150 0.999
0.22 3.0 0.021 625 0.999
Tube 2300 – – 0.017 5 0.904
0.31 – 0.017 80 0.987
0.31 2.7 0.017 625 0.999
Cu Platform tube 2000 – – 0.034 40 0.998
324.754 0.46 – 0.034 150 0.999
0.46 4.0 0.034 625 0.999
Filter furnace 2200 – – 0.038 40 0.999
0.64 – 0. 038 312 0.999
0.64 4.0 0.038 625 0.999
Tube 2000 – – 0.032 10 0.997
0.62 – 0.032 40 0.999
0.62 3.6 0.032 625 0.998
Ag Platform tube 2400 – – 0.049 5 0.999
328. 068 0.45 – 0.051 20 0.999
0.45 3.4 0.052 625 0.999
1800 – – 0.056 5 0.999
0.49 – 0.057 20 0.999
0.49 4.0 0.057 625 0.999
Filter furnace 2400 – – 0.051 10 0.999
0.42 – 0.051 19.5 0.999
0.42 3.2 0.051 312.5 0.991

(continued on next page)

22 D. Katskov et al. / Spectrochimica Acta Part B 71–72 (2012) 14–23

Table 3 (continued)
Element, Atomizer Temperature Calibration curves W = γ′∙m
line/nm setting/°C
Linearization Sensitivity, Linearity Correlation
parameters in (15), γ′/pm∙s∙L∙μg− 1 rangea/μg·L− 1 coefficient R2
(20)/pm

C1 C2

Tube 2400 – – 0.038 5 0.988

0.48 – 0.039 20 0.996
0.48 4.8 0.038 625 0.997
1800 – – 0.046 5 0.967
0.45 – 0.044 20 0.990
0.45 4.4 0.044 625 0.998
a
Upper limit of the linearity range or highest concentration measured.
b
The measurements for tube atomizer are excluded from the table because of Cd vapor losses during the pyrolysis stage.

Table 4
Variation of linearization parameters C1 and C2 in Eqs. (15) and (20) that should accompany ± 5% sensitivity changes for Mn using 279.48 nm absorption line in the platform tube.

Atomizer Experimental data for low concentrations and Obtained by linearization in a broad range Correction parameters,
suggested ±5% deviation (in brackets) Eqs. (13) and (18)/pm

Sensitivity/pm∙s∙L∙μg− 1 R2 Sensitivity/pm∙s∙L∙μg− 1 R2 C1 C2

Platform tube (0.0711) 0.0709 0.9994 0.54 1.8

0.0745 0.9967 0.0743 0.9994 0.46 2.3
(0.0784) 0.0785 0.9994 0.40 2.8
Filter furnace (0.0687) 0.0689 1.0 0.38 2.7
0.0723 0.9971 0.0722 1.0 0.34 3.2
(0.0759) 0.0761 0.9999 0.30 3.9
Tube (0.0451) 0.0449 0.9996 0.66 1.4
0.0475 0.9999 0.0474 0.9994 0.54 1.9
(0.0499) 0.0498 0.9996 0.46 2.4

Table 5
Calibration equations obtained using linearization parameters from Table 3, averaged for each line.

Element, Atomizera Data from Table 3 W = γ′*∙m LogW=x∙Log (γ′-γ′*)/γ′/%

line/nm m+Logγ′

γ′/pm∙s∙L∙μg−1 Averaged/pm γ′*/pm∙s∙L∙μg−1 R2 x R2

C1 C2

Pb 217.001 1 0.014 0.45 1.9 0.015 0.9987 1.001 0.9992 +1

2 0.018 0.018 0.9996 0.993 0.9987 0
3 0.011 0.011 0.9990 1.005 0.9991 0
Cd 1 0.082 0.26 3.4 0.063 0.9983 0.956 0.9971 − 23
228.802 2 0.142 0.156 0.9991 1.010 0.9993 + 10
Mn 1 0.074 0.45 2.5 0.074 0.9995 1.009 0.9995 0
279.482 2 0.72 0.070 0.9998 0.992 0.9993 −3
3 0.047 0.049 0.9995 0.992 0.9994 +4
Mn 1 0.055 0.47 2.6 0.051 0.9998 1.001 0.9993 −7
279.827 2 0.059 0.049 0.9994 1.007 0.9993 − 16
3 0.038 0.042 0.9996 1.011 0.9987 + 10
Pb 1 0.011 0.58 5.6 0.010 0.9989 0.988 0.9998 −9
283.306 2 0.018 0.018 0.9940 1.001 0.9998 0
3 0.008 0.008 0.9992 1.019 0.9995 0
Fe 1 0.025 0.27 2.8 0.026 0.9983 1.000 0.9976 +4
243.327 2 0.021 0.019 0.9995 0.961 0.9956 −9
3 0.017 0.019 0.9981 1.060 0.9990 − 11
Cu 1 0.034 0.57 3.9 0.037 0.9997 1.040 0.9986 +9
324.754 2 0.038 0.036 0.9997 1.000 0.9997 −5
3 0.032 0.033 0.9985 0.999 0.9991 +3
Agb 1 0.051 0.45 3.5 0.052 0.9969 0.997 0.9987 +2
328.068 2 0.051 0.052 0.9930 0.985 0.9991 +2
3 0.038 0.048 0.9996 1.034 0.9989 + 26
Agc 1 0.057 0.47 4.6 0.057 0.9996 1.033 0.9988 0
328.068 3 0.045 0.045 0.9977 0.985 0.9980 −2
a
1—platform tube; 2—filter furnace; 3—tube;
b
2400 °C.
c
1800 °C.
 D. Katskov et al. / Spectrochimica Acta Part B 71–72 (2012) 14–23
1.8
1.6
1.4
1.2
δAv and C1 /pm
1
0.8
0.6
123
0.4
23
0.2
0 References
Ag 328.068 Cd 228.802 Cu 324.754 Fe 248.327 Mn 279.482 Pb 217.001 Pb 283.306
Fig. 6. Calculated FWHM of Voigt contours of the lines at 2800 (*) and 2000 K (**) and
measured optimal first linearization parameters (C1) from Table 3 for the tube (1),
platform (2) and filter furnace (3).
the structure of the hyperbolic calibration curve from the ratio of Doppler
and collisional broadening. It can be also suggested that more precise
approximation than that in Fig. 1 would provide better correlation of
correction factors and physical parameters of the absorption line. For
example, the hyperbolic equation employed in [6,7] for the approxima-
tion of the calibration curve could be used instead for modification of the
running transient absorption, which would be followed by temporal
integration.
5. Conclusion
It has been verified using the prototype HR-CS ET AAS instrument
and solutions of several elements in various experimental conditions
that the approximation suggested and the respective calculation
algorithm make possible a linearization of the calibration curves
within 4 or more orders of magnitude of the analyte concentration
range. Obviously, the suggested methodology should be confirmed using
commercial instrumentation with standard atomizers and various real
samples. In the determination of the linearization parameters for typical
experimental conditions and further usage of the algorithm in analytical
23
practice, standard Excel software can be employed. It is expected that
further investigation of the issue should make possible prediction of the
linearization parameters on the base of physical properties of the element
to be determined and technical characteristics of the spectrometer.
Acknowledgment
The authors express their gratitude to ISAS — Leibnitz-Institut für
Analytische Wissenschaften (Berlin, Germany) for the donation of the
prototype HR-CS AAS spectrometer to Tshwane University of Technology
(TUT) and to National Research Foundation of South Africa for the
financial support.
[1] B. Welz, H. Becker-Ross, S. Florek, U. Heitmann, High-resolution Continuum
Source AAS: the Better Way to Do Atomic Absorption Spectrometry, Wiley-VCH,
Weinheim, 2005.
[2] A.C.G. Mitchell, M.W. Zymansky, Resonance Radiation and Excited Atoms,
Cambridge University Press, Cambridge, 1961.
[3] B.V. Lvov, Atomic Absorption Spectrochemical Analysis, Adam Hilger (Elsevier),
London (New York), 1970.
[4] K.I. Tarasov, The Spectroscope, Adam Hilger, Bristole, 1974.
[5] J.M. Harnly, The future of atomic absorption spectrometry: a continuum source
with a charge coupled array detector, J. Anal. At. Spectrom. 14 (1999) 137–146.
[6] J. Harnly, C.M.M. Smith, B. Radziuk, Extended calibration ranges for continuum
source atomic absorption spectrometry with array detection, Spectrochim. Acta
Part B 51 (1996) 155–1079.
[7] D.N. Wichems, R.E. Fields, J. Harnly, Characterization of hyperbolic calibration
curves for continuum source atomic absorption spectrometry with array
detection, J. Anal. At. Spectrom. 13 (1998) 1277–1284.
[8] U. Heitmann, B. Welz, D.L.G. Borges, F.G. Lepri, Feasibility of peak volume, side
pixel and multiple peak registration in high-resolution continuum source atomic
absorption spectrometry, Spectrochim. Acta Part B 62 (2007) 1222–1230.
[9] B. Welz, M. Sperling, Atomic Absorption Spectrometry, 3rd ed. WILEY-VCH,
Weinheim, 1999.
[10] U. Heitmann, M. Schütz, R. Becker, S. Florek, Measurements on the Zeeman-
splitting of analytical lines by means of a continuum source graphite furnace
atomic absorption spectrometer with a linear charge coupled device array,
Spectrochim. Acta Part B 51 (1996) 1095–1105.
[11] U. Heitmann, H. Becker-Ross, D. Katskov, Feasibility of filter atomization in
high-resolution continuum source atomic absorption spectrometry, Spectrochim.
Acta Part B 61 (2006) 351–360.
[12] D.A. Katskov, G.E. Khanye, Simultaneous multi-element electrothermal atomic
absorption determination using a low resolution CCD spectrometer and
continuum light source: the concept and methodology, S. Afr. J. Chem. 63
(2010) 45–57.