Radiation Protection Dosimetry (2005), Vol. 116, No. 1–4, pp.

293–296
doi:10.1093/rpd/nci107

ON THE THEORY AND SIMULATION OF MULTIPLE COULOMB

SCATTERING OF HEAVY-CHARGED PARTICLES
S. I. Striganov

Accelerator Division, Fermi National Accelerator Laboratory, MS 220, Batavia, IL 60510-0500, USA

The Moliere theory of multiple Coulomb scattering is modified to take into account the difference between processes of
scattering off atomic nuclei and electrons. A simple analytical expression for angular distribution of charged particles passing
through a thick absorber is found. It does not assume any special form for a differential scattering cross section and has a

Downloaded from http://rpd.oxfordjournals.org/ at National Dong Hwa University Library on April 4, 2014
wider range of applicability than a gaussian approximation. A well-known method to simulate multiple Coulomb scatterings is
based on treating ‘soft’ and ‘hard’ collisions differently. An angular deflection in a large number of ‘soft’ collisions is sampled
using the proposed distribution function, a small number of ‘hard’ collision are simulated directly. A boundary between ‘hard’
and ‘soft’ collisions is defined, providing a precise sampling of a scattering angle (1% level) and a small number of ‘hard’
collisions. A corresponding simulating module takes into account projectile and nucleus charged distributions and exact
kinematics of a projectile–electron interaction.

INTRODUCTION In this case, the angular distribution can be rewritten

in a simple form
Multiple scattering of charged particles in the
Coulomb field of nuclei is of interest for numerous expðfÞ
applications related to particle transport in matter. F ðy, tÞ ¼ ½1 þ rL2 ðfÞ þ 3r2 L4 ðfÞ þ    ,
A comprehensive comparison of the Moliere theory(1) phy2 i
with experimental data on multiple scattering of ð4Þ
1 MeV to 200 GeV protons shows that this theory,
with the Fano correction(2), is accurate to better than where Lk are the Laguerre polynomials.
1% on average(3) except for thick absorbers(4) and for
hydrogen targets at high energies(3). x2 y2
L2 ðxÞ ¼ 1  2x þ , f¼ ;
2 hy2 i
NEW ASYMPTOTIC FOR THICK hy4 i 1
r¼ / :
SCATTERERS 2hy2 i2 t
An angular distribution of a charged particle after For a large value of thickness, the parameter r
passing through an absorber of a length t, can be becomes small and one or two terms in Equation 3
written as: are enough to provide a reasonable accuracy.
Z 1 Note that the Moliere theory is based on a single
1 scattering cross section, which has infinite moments
F ðy, tÞ ¼ J0 ð pyÞexp½tAðpÞp dp, ð1Þ
2p 0 (Equation 3). Therefore, this theory has a wrong
asymptotic at large thicknesses. This problem can
where be overcome if a finite nuclear size is taken into
Z account. The solution for an angular distribution
1
d was obtained(4) under the assumption that the charge
tAðpÞ ¼ t d ½1  J0 ð pyÞ , ð2Þ distribution in nuclei is a gaussian distribution.
0 d
It coincides with Moliere for a thickness of 0.1–1
where d/d is a single scattering cross section. If radiation lengths and reaches gaussian for 100–1000
thickness t is large enough, then only small p are radiation lengths. This modification of the Moliere
important in Equations 1 and 2, then theory can be used to find a range of the applicability
of the approximation (Equation 4). As shown in
p2 2 p4 4 Figure 1, the first two terms in Equation 4 provide
tAðpÞ ¼ hy i  hy i, 1% agreement with a precise calculation for four
4 64
Z 1 orders of magnitude at r  0.01. One term in
d Equation 4 (gaussian distribution) is enough to
hyk i ¼ t d yk : ð3Þ
0 d describe first two decades only. For r > 0.05, the
new approximation with first two terms provides
better than 10% accuracy. In more convenient units,


Corresponding author: strigano@fnal.gov range of the validity of this approximation can be

ª The Author 2005. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oxfordjournals.org
 S. I. STRIGANOV
The recoil imparted to atomic electron by incident
heavy particle cannot exceed a certain limit, so a
simple approximation of the inelastic cross section
is given by
din w2c qin ðwÞ
t ¼ , w  wmax , ð7Þ
d Zpw4
wmax is so defined that the mean-squared angle
resulting from Equation 7 is adjusted to the mean-
squared angle calculated from the precise cross
section.
Using Equations 1, 2, 5 and 7, the angular distri-

Downloaded from http://rpd.oxfordjournals.org/ at National Dong Hwa University Library on April 4, 2014
bution can be written as
Z    2 
1 1 uy u
Fðy,tÞ ¼ J0 exp  þUðu,BÞ udu,
2p 0 yM 4
2
 2 2 Z 1
u u 2u 1  J0 ðxÞ
U ¼ log þ dx , ð8Þ
Figure 1. Ratio of asymptotic (Equation 4) and precise 4B 4 ðZþ1ÞB wu x3
angular distribution(4). The dashed line represents the first w
term and the solid line is the first two terms respectively, in w ¼ max , y2M ¼ Bw2c :
yM
Equation 4.
B is defined from
estimated from w2c 1 w2
B  lnB ¼ ln 2
þ 1  2C þ ln a2 : ð9Þ
26X0 b 2 wa Z þ 1 wi
r
,
ðZ þ 6:4Þt At small thicknesses (w 1), the distribution
where X0 is a radiation length and Z is a charge of (Equation 8) is described by the Moliere function
the absorber nuclei. with parameters defined from Equations 6 and 9.
At large thicknesses (w 1), Equation 8 also
coincides with Moliere, but the parameters are
SCATTERING OFF ATOMIC ELECTRONS calculated using
AND MOLIERE THEORY
w2cM 1 w2
A charged particle traversing a medium is deflected B  ln B ¼ ln þ 1  2C þ ln max ,
primarily by elastic collisions in the Coulomb field of wa2 Z w2i
nuclei. Inelastic collisions with atomic electrons Z
should also be taken into account. To estimate the w2cM ¼ w2c : ð10Þ
Zþ1
contribution of inelastic collisions, Bethe(1) proposed
to replace the squared nuclear charge, Z2, with the Figure 2 shows a ratio of angular distributions cal-
sum of the squares of the nuclear and electronic culated using Equations 8–10 and the Moliere distri-
bution obtained using the Bethe Z(Z þ 1) approach.
charges, Z(Z þ 1). This procedure would be accurate
It is seen that a standard Z(Z þ 1) approximation
if the single scattering cross sections were the same
for nucleus and electron targets. The actual cross agrees well with more precise consideration (Equa-
sections are different at small and large angles. Let tion 8) for small angles, but it overestimates the
angular distribution by 50% at large angles for
us consider a modification to the Moliere theory,
hydrogen. Calculations for other targets show that
which takes into account these differences.
Elastic scattering cross section reads(1) this large-angle ratio is simply Z/(Z þ 1). The asymp-
totics (Equations 9 and 10) have been obtained by
del w2c qel ðwÞ Fano(2). He believed that the Equation 9 is valid for
t ¼ , ð5Þ incident electrons and Equation 10 can be applied
d pw4
for heavy particles. Our consideration shows that the
4pNtz2 ZðZ þ 1Þe4 above limits have different ranges of applicability. If
w2c ¼ , ð6Þ w 1, solution (Equation 9) should be used even
Ap2 b2
for heavy particles. As shown in Table 1, this con-
where ze, p, b are the charge, momentum and velo- clusion is supported by experiments.
city of the incident particle, qel(w) is a screening Low-w measurements(3) agree well with the
function, N is Avogadro’s number and A is an formula (Equation 10), but at large w formula
atomic weight of target material. (Equation 9) is much closer to data(5) than the

294
 THEORY OF MULTIPLE COULOMB SCATTERING
It can be shown that the ‘continuous’ distribution
is given by
expðfÞ
F ðy, tÞ ¼ ½1 þ rs L2 ðfÞ, ð11Þ
phy2s i
where
Z yb
d y2
hyks i ¼ t d yk ; f¼ ;
0 d hyks i
hy4s i 1
rs ¼ / :
2hy2s i2 t

Downloaded from http://rpd.oxfordjournals.org/ at National Dong Hwa University Library on April 4, 2014
A range of the applicability of Equation 11 is
defined as
rs ðyb , tÞ ¼ d 1: ð12Þ
For any step of length t, the boundary angle yb can
Figure 2. Ratio of the modified Moliere theory and Bethe be calculated from Equation 12. For elastic scatter-
approach for 10 GeV c1 muon on a hydrogen absorber. ing, using Equations 5 and 6, Equation 12 can be
Solid line is Equation 8, dotted line is Equation 9 and rewritten as
dashed line is Equation 10. Symbols are Monte Carlo

simulations. 1 8dw2 y2b
x  ln x ¼ b ¼ ln 2cM þ 1 , x ¼ :
2 wa 8dw2cM

Table 1. Ratio of measured and calculated widths of angular

ð13Þ
distributions. A solution of Equation 13 with better than 10%
accuracy is given by
Z w Exp./Bethe Exp./Fano Exp./Fano

r 
(Equation 9) (Equation 10) b ln b 1
x¼ 1þ 1þ 1  : ð14Þ
2 b1 b
1 9.7 0.99 0.01 0.96 0.01 0.88 0.01 The derivation between the exact and approximate
4 10.3 1.02 0.04 1.00 0.04 0.96 0.04
solutions can be considered as a redefinition of d.
6 9.8 1.03 0.04 1.01 0.04 0.98 0.04
4 0.12 1.00 0.06 0.98 0.06 1.03 0.06 For very small thicknesses (0.001 of a radiation
6 0.10 0.97 0.04 0.95 0.04 0.98 0.04 length) Equation 13 cannot be resolved for an arbit-
rary d. In this case, a maximal possible value of d can
be used. To perform a simulation with an arbitrary
original Fano result (Equation 10). Note, that nuclear form factor, the boundary angle should be
widths of angular distributions were measured at limited at a large step
small angles (y/yM < 2). Therefore, the Bethe ap- 0:3
proach looks perfect in Table 1. But as was shown y2b  0:1y2nuc ¼ , ð15Þ
p2 r2nuc
above, this approximation overestimates the large
angle scattering. where r2nuc is an average nuclear radius squared. At
large thicknesses, a number of ‘hard’ interactions
MONTE CARLO APPROACH becomes not small. In this case, the angular distribu-
An efficient method to simulate multiple scattering is tion is well described by the approximation (Equa-
based on a separate treatment of ‘soft’ and ‘hard’ tion 4). Scattering angles can be sampled directly
interactions. Angular deflection in a large number of from Equation 4. A comparison of angular distribu-
‘soft’ collisions is sampled from a ‘continuous’ dis- tions simulated by the above algorithm and calcu-
tribution, ‘hard’ scatterings are simulated expli- lated using analytical solutions(1,4) shows that results
citly(6). There is an obvious correlation between agree within 1% for d ¼ 0.03. As shown in Figure 3,
precision and efficiency of the algorithm and the a number of ‘hard’ elastic scatterings is small for this
value of a boundary angle yb between ‘soft’ and value of d.
‘hard’ collisions. For small yb, a number of discrete The boundary angle yin for inelastic collisions can
interactions is large and precision is high, for larger be obtained also from Equation 13. In this case
yb, the efficiency increases but the accuracy

1 8dw2cM Zy2in
decreases. The optimal value of a boundary angle is bin ¼ ln þ 1 , x ¼ : ð16Þ
2 Zw2i 8din w2cM
important but still an open question.
295
 S. I. STRIGANOV
Figure 3. Average number of ‘hard’ elastic collisions.
The procedure of determination of the inelastic
boundary angle was checked by comparison of
Monte Carlo simulation based on simple cross
sections (Equations 5–7) and analytical solution
(Equations 8 and 9). The results agree within
1% if din < 0.1. Equations 13 and 16 were obtained
for small angles for which only the Rutherford part
of the cross section is important. Therefore, the
inelastic boundary angle should be limited by
b2
min b2 p2 y2in
¼ ¼ gin 1, ð17Þ
emax 2me emax
where
min and emax are the minimum and max-
imum energies transferred to the electron in a single
collision.
In inelastic scattering with electrons, the projectile
undergoes angular deflection and loses energy. For
discrete inelastic collisions, the correlation between
the energy loss and scattering angle is determined by
kinematics. ‘Continuous’ energy losses are described
well by the Vavilov distribution with redefined para-
meters(7). It is known(7) that a log–normal distribu-
tion fits the Vavilov function well for
x w2
k¼ ¼ cM2  0:3: ð18Þ

min Zyin
With the boundary inelastic angle determined from
Equations 16 to 18, the average number of inelastic
discrete collisions is small, in most cases it is about k.
A comparison between simulations and the modified
Moliere theory for hydrogen is shown in Figure 2.
The theory based on the approximation given in
Equation 7 reproduces main features, but at large
angles and small thicknesses Monte Carlo, which
uses a more precise cross section, should be applied.
Note that in such an approach the energy loss and
angular distributions are correlated, at least for
low-Z targets.
A comparison of the Monte Carlo simulation and
analytical calculation for high-Z target is presented
296
Downloaded from http://rpd.oxfordjournals.org/ at National Dong Hwa University Library on April 4, 2014
Figure 4. Angular distributions of 50 GeV c1 muon
passing through a uranium absorber. Symbols, Monte
Carlo simulation; solid line, theory(4).
in Figure 4. The theory(4) is based on a gaussian
model of a nuclear charge density, a more precise
Fermi model can be used in Monte Carlo. If the
gaussian form factor is used in simulation and
numerical integration, results agree within 1%. If
the Fermi charge density is used in Monte Carlo
the difference between theory and simulation
becomes noticeable at large angles.
ACKNOWLEDGEMENTS
I am grateful to Nikolai Mokhov and Dick Prael for
many useful discussions, suggestions and support.
This work was supported by the Universities
Research Association, Inc., under contract DE-
AC02-76CH03000 with US Department of Energy.
REFERENCES
1. Bethe, H. A. Moliere’s theory of multiple coulomb
scattering. Phys. Rev. 89, 1256–1266 (1953).
2. Fano, U. Inelastic collisions and the Moliere theory of
multiple scatterings. Phys. Rev. 93, 117–120 (1954).
3. Gottschalk, B., Koehler, A. M., Schneider, J. M.,
Sisterson, J. M. and Wagner, M. S. Multiple Coulomb
scattering of 160 MeV protons. Nucl. Instrum. Methods
Phys. Res. B 74, 467–490 (1993).
4. Baishev, I. S., Mokhov, N. V. and Striganov, S. I. On
the effect of the finite size of the nucleus in the theory of
multiple Coulomb scattering. Sov. J. Nucl. Phys. 42,
745–749 (1985).
5. Shen, G. and 17 others. Measurement of multiple
scattering at 50 and 200 GeV/c. Phys. Rev. D 20,
1584–1588 (1979).
6. Andreo, P. and Brahme, A. Restricted energy-loss strag-
gling and multiple scattering of electrons in mixed Monte
Carlo procedures. Radiat. Res. 100, 16–29 (1984).
7. Chibani, O. New algorithms for the Vavilov distribution
calculation and the corresponding energy loss sampling.
IEEE Trans. Nucl. Sci. 45, 2288–2292 (1999).