Improved electron transport mechanics in the PENELOPE Monte-Carlo model A.F. Bielajew a, * , F. Salvat b a Department of Nuclear Engineering and Radiological Sciences, University of Michigan, 1906 Cooley Building, 2355 Bonisteel Boulevard, Ann Arbor, MI 48109-2104, Michigan, USA b Facultat de F ısica (ECM), Universitat de Barcelona, Societat Catalana de F ısica, Diagonal 647, 08028 Barcelona, Spain Received 17 March 2000 Abstract We describe a new model of electron transport mechanics, the method by which an electron is transported geo- metrically in an infinite medium as a function of pathlength, s, the accumulated elastic multiple-scattering angular deflection characterized by Hs, the polar scattering angle, and U, a random azimuthal angle. This model requires only one sample of the multiple-scattering angle yet it reproduces exactly the following spatial moments and space–angular correlations: hzi, hx sin H cos Ui, hy sin H sin Ui, hz cos Hi, hx 2 i, hy 2 i and hz 2 i. Moreover, the distributions associated with these moments exhibit a good improvement over the PENELOPE transport mechanics model when compared self- consistently with the results of analog simulations. When we split the transport step into two steps with equal path- length, we observe excellent agreement with the distributions, indicating that the algorithm nearly matches higher order moments when employed in this way. The equations described herein are relatively inexpensive to employ in an iterative Monte-Carlo code. We have employed the new model to demonstrate the usefulness of the new mechanics for several examples that span the dynamic range of application. Ó 2001 Elsevier Science B.V. All rights reserved. PACS: 02.50Ng; 13.60Fz; 25.30Bf; 34.80Bm Keywords: Monte-Carlo simulation; Condensed history; Elastic scattering; Multiple-scattering; Coulomb scattering 1. Introduction One of the most challenging problems in the Monte-Carlo simulation of high-energy electron (and positron) transport is the generation of spa- tial displacements of the particle. In each step of the simulation, the electron is moved a certain pathlength, s, through the medium. The angular deflection after this pathlength is determined by the polar multiple-scattering angle, Hs, and the azimuthal angle U, which is distributed uniformly on 0; 2p. For a given elastic cross-section, the theory of Goudsmit and Saunderson [1,2] provides the multiple-elastic scattering distribution from which Hscan be sampled. The diculty comes Nuclear Instruments and Methods in Physics Research B 173 (2001) 332–343 www.elsevier.nl/locate/nimb * Corresponding author. Tel.: +1-734-7646364; fax: +1-734- 7634540. E-mail address: [email protected] (A.F. Bielajew). 0168-583X/01/$ - see front matter Ó 2001 Elsevier Science B.V. All rights reserved. PII: S 0 1 6 8 - 5 8 3 X ( 0 0 ) 0 0 3 6 3 - 3
12
Embed
Improved electron transport mechanics in the PENELOPE Monte … · 2003-10-23 · Improved electron transport mechanics in the PENELOPE Monte-Carlo model A.F. Bielajew a,*, F. Salvat
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Improved electron transport mechanics in the PENELOPEMonte-Carlo model
A.F. Bielajew a,*, F. Salvat b
a Department of Nuclear Engineering and Radiological Sciences, University of Michigan, 1906 Cooley Building, 2355 Bonisteel Boulevard,
Ann Arbor, MI 48109-2104, Michigan, USAb Facultat de F�õsica (ECM), Universitat de Barcelona, Societat Catalana de F�õsica, Diagonal 647, 08028 Barcelona, Spain
Received 17 March 2000
Abstract
We describe a new model of electron transport mechanics, the method by which an electron is transported geo-
metrically in an in®nite medium as a function of pathlength, s, the accumulated elastic multiple-scattering angular
de¯ection characterized by H�s�, the polar scattering angle, and U, a random azimuthal angle. This model requires only
one sample of the multiple-scattering angle yet it reproduces exactly the following spatial moments and space±angular
correlations: hzi, hx sin H cos Ui, hy sin H sin Ui, hz cos Hi, hx2i, hy2i and hz2i. Moreover, the distributions associated with
these moments exhibit a good improvement over the PENELOPE transport mechanics model when compared self-
consistently with the results of analog simulations. When we split the transport step into two steps with equal path-
length, we observe excellent agreement with the distributions, indicating that the algorithm nearly matches higher order
moments when employed in this way. The equations described herein are relatively inexpensive to employ in an iterative
Monte-Carlo code. We have employed the new model to demonstrate the usefulness of the new mechanics for several
examples that span the dynamic range of application. Ó 2001 Elsevier Science B.V. All rights reserved.
One of the most challenging problems in theMonte-Carlo simulation of high-energy electron(and positron) transport is the generation of spa-
tial displacements of the particle. In each step ofthe simulation, the electron is moved a certainpathlength, s, through the medium. The angularde¯ection after this pathlength is determined bythe polar multiple-scattering angle, H�s�, and theazimuthal angle U, which is distributed uniformlyon �0; 2p�. For a given elastic cross-section, thetheory of Goudsmit and Saunderson [1,2] providesthe multiple-elastic scattering distribution fromwhich H�s� can be sampled. The di�culty comes
Nuclear Instruments and Methods in Physics Research B 173 (2001) 332±343
0168-583X/01/$ - see front matter Ó 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 1 6 8 - 5 8 3 X ( 0 0 ) 0 0 3 6 3 - 3
from the fact that the space displacement �x; y; z�at the end of the step, although strongly correlatedwith the angular de¯ection, is not known. Formalsolutions of the transport equation [3,4] provideclosed expressions only for the moments of thespace displacements and space±angle correlations.Since only a few of these moments can be evalu-ated and employed in a practical transport scheme,there is not enough information to characterize�x; y; z� unambiguously. The prescription that re-lates �x; y; z� to H�s�, U and s will be called the``electron transport mechanics'' [4].
The PENELOPE Monte-Carlo code system [5±7], a general purpose coupled e�c Monte-Carlocode, employs a ``random hinge'' electron trans-port mechanics' scheme that can be summarized asfollows:
x=s � r sin H�s� cos U;
y=s � r sin H�s� sin U;
z=s � �1ÿ r� � r cos H�s�;�1�
where r is a random number sampled uniformly on�0; 1�, and the pathlength, s, for which the multiple-scattering angle H�s� is calculated and interpretedas the total curved pathlength that the electrontravels through the medium.
Although this scheme is an ansatz, it produceshigh quality results as indicated through compli-ance [4] with Lewis' moments [3]. The Lewis mo-ments studied in the previous work were
hzi �Z s
0
ds0 eÿg1s0 ;
hzli � 1
3
Z s
0
ds0 eÿg1�sÿs0��1� 2eÿg2s0 �;hx sin H cos U� y sin H sin Ui
� 2
3
Z s
0
ds0 eÿg1�sÿs0��1ÿ eÿg2s0 �; �2�
hz2i � 1
3
Z s
0
ds0Z s0
0
ds00 eÿg1�s0ÿs00��1� 2eÿg2s00 �;
hx2 � y2i � 2
3
Z s
0
ds0Z s0
0
ds00 eÿg1�s0ÿs00��1ÿ 2eÿg2s00 �;
where the gs are moments of the single-scatteringcross-section r�l� with Legendre polynomials,
g` � 2pNAq
A
Z s
0
ds0Z 1
ÿ1
dlr�l��1ÿ P`�l��; �3�
in which NA is Avogadro's number, A is atomicweight (we assume single-element medium) and q isthe density of the medium. Here, the distance s isexpressed as a unit of length and the integrationvariable l is the cosine of the polar scattering angle.Since the scattering model we are considering isazimuthally symmetric, hx sin H cos Ui � hy sin Hsin Ui, hx2i � hy2i, and are combined in Eq. (2).
The angular distribution after a pathlength, s, isgiven by
f �l; s� �X1l�0
l�� 1
2
�eÿsg`P`�l�; �4�
and we also have
hli � eÿg1s; hl2i � 1� 2eÿg2s
3: �5�
Here, we have ignored the energy dependence ofthe single-scattering cross-section, which allowsfor greater analytic development. The aboveequations may be expressed in energy-dependentform, for example, employing the continuousslowing down approximation (CSDA), wherebythe integrals over pathlength, s, in Eq. (2) are re-placed by integrals over energy, and pathlength andenergy are related through a stopping-power rela-tionship. We leave this, or similar, adaptations tofuture work. Having ignored energy loss, the inte-grals in Eq. (2) may be performed with the result
hzis�1ÿeÿn
n;
hzlis� 1
3n1
�ÿeÿn�2
eÿnÿeÿcn
cÿ1
�;
hxsinHcosU�y sinHsinUis
� 2
3n1
��eÿcnÿceÿn
cÿ1
�;
hz2is2� 2
3cn2c�n�
�eÿn�ÿ�cÿ2��2eÿcnÿceÿn
cÿ1
�;
hx2�y2is2
� 4
3cn2cn
�ÿ�c�1�ÿeÿcnÿc2eÿn
cÿ1
�;
�6�where n � sg1 and c � g2=g1 which spans the rangefrom 0 (backward scattering) to 3 (high energy,
A.F. Bielajew, F. Salvat / Nucl. Instr. and Meth. in Phys. Res. B 173 (2001) 332±343 333
forward directed) and has the value unity for iso-tropic scattering.
Generally, much of the range of applicationinvolves small values of n which suggests an ex-pansion of the moments in a Taylor series in n.These results have been stated elsewhere [4], butwe include them here for completeness and for thenext order in n for later use. To O�n3�,
sÿhzis�n
2ÿn2
6�n2
24;
cov�z;l�s
�3ÿc3
n�c2�cÿ9
9n2ÿc3�c2�cÿ21
36n3;
hxsinHcosU�ysinHsinUis
�c3nÿc�c�1�
9n2�c
c2�c�1
36n3;
�7�
var�z�s2�2�3ÿc�
9n�c2�cÿ9
18n2ÿc3�c2�cÿ21
90n3;
hx2�y2is2
�2c9
nÿc�c�1�18
n2�cc2�c�1
90n3;
where cov�z; l� � hzli ÿ hzihli and var�z� � hz2iÿhzi2. We note that the O�n� terms of cov�z; l� andvar�z� vanish in the high-energy limit. This willhave interesting consequences as we shall seelater on.
With the above information, we can comparepredictions of spatial and angular moments ofPENELOPEs transport mechanics. As shownpreviously [4], to O�n2�,
hziPÿhzihzi � n2
12;
hzliPÿhzlihzli � 2c2ÿ4c�3
36n2;
hxsinHcosU� y sinHsinUiPhxsinHcosU� y sinHsinUi ÿ1
�8�
�ÿcÿ2
6n� c2ÿ c�1
36n2;
hz2iPÿhz2ihz2i � 3�2c�cÿ1�
36n2;
hx2� y2iPÿhx2� y2ihx2� y2i �ÿcÿ1
4n�3ÿ12c�13c2
240n2:
We note, in particular, that the lateral moments,hx sin H cos U� y sin H sin UiP and hx2 � y2iP haveO�n� discrepancies, while the longitudinal onesshown are O�n2�. This forms the main motivationfor attempting to improve the model. We will seethat our new form is able to reproduce the exactmoments shown above.
2. The improved model
It su�ces to say that we attempted manyschemes before settling on the following model:
x=s � fr sin H�s� cos U1 � r cos H�s� cos U2;
y=s � fr sin H�s� sin U1 � r cos H�s� sin U2;
z=s � k�1ÿ r� � c� �kr � d� cos H�s�;�9�
where r is a random number sampled uniformly on�0; 1�, and f, r, k, c and d are constants independentof r, H and the Uis. U1 and U2 are sampled uni-formly on �0; 2p� and are independent of eachother. On average, U1 is associated with the ran-dom azimuthal direction of scattering after path-length, s, while U2 provides some additionalstraggling about this direction.
In the analysis associated with the presentwork, it became apparent that the simple PE-NELOPE picture of a particle traveling a certainrandom distance, scattering and traveling the re-mainder of the pathlength, would have to beabandoned. Yet, the anticorrelation of the twolongitudinal parts of k (with and without cos H�s�)would have to be nearly preserved except for themodi®cation by c and d. The correlation betweenthe cos H�s�-dependent part of z and the lateralde¯ection is also important but the correlation isbroken to some degree by the di�erent modifyingfactors, k and f. Physically, the r-factor was mo-tivated to break the absolute correlation betweenthe azimuthal direction of scatter and the azi-muthal direction of transport. This e�ect was rec-ognized in the algorithm described by Kawrakow[8] although our approach is di�erent. Thecos H�s�-factor modifying the r-term was found tobe necessary to reduce some overprediction of thetail in lateral straggling distributions.
334 A.F. Bielajew, F. Salvat / Nucl. Instr. and Meth. in Phys. Res. B 173 (2001) 332±343
The requirement that the model Eq. (9) repro-duce the ®ve exact Lewis moments given in Eq. (6)allows us to solve for f, r, k, c and d with theresult
f � 2hx sin H cos U� y sin H sin Ui
1ÿ hl2i ; �10�
r �������������������������������������������������hx2 � y2i ÿ f 2
3�1ÿ hl2i�
hl2i
s; �11�
k �������������������������������������������������������������12
which is expressed above to O�n2�. The apparentsingularity of these expressions is an artifact ofleading order terms in n vanishing in the limitc! 3. All this means that the series expansionexpressed in Eq. (15) is of limited use. For nu-merical calculations we will have to resort to Eqs.(10)±(14) and Eq. (6) for most cases except veryclose to n � 0 where rational expressions of twoTaylor series in n were employed.
The c! 3 limit of Eqs. (10)±(14) is
f � 1� 1
6nÿ 1
6n2;
r � 1
9n2;
k � 1���3p 1
�� n
5� 19
1800n2
�; �16�
c � 4ÿ ���3p
6ÿ 20� 3
���3pÿ �
90n� 2
27
�ÿ 19
3600���3p�
n2;
d � 2ÿ ���3p
6ÿ 3
���3p ÿ 5
ÿ �90
nÿ 1
54
�� 19
3600���3p�
n2:
The ``exact'' forms of the ®ve factors, f, r, k, c andd are plotted in Figs. 1±5 over the ranges 06 n6 1and 06 c6 3 ± su�cient for any practical simula-tion. These should be contrasted to the standardPENELOPE model, f � k � 1, r � c � d � 0. Allthe surfaces are ¯at with the most structure beingexhibited near c � 3 which may have been ex-pected from the previous discussion.
3. Implementation of the new algorithm
To generate random electron trajectories, thetransport mechanics algorithm needs to be sup-plied with polar de¯ections, H�s�, sampled fromappropriate multiple-scattering distributions.Ultimately, the reliability of the simulation isgoverned by the physical quality of the single-scattering model adopted and the numerical ac-curacy of the pre-calculated multiple-scatteringangular distribution. In the limit of small path-lengths, Larsen has shown [9] that condensedsimulation should reproduce the exact solution ofthe transport equation, independently of the formof the underlying elastic cross-section. However, ithas been argued [10] that the multiple-scatteringalgorithm must also be ``robust''. That is, thesimulated spatial and angular distributions after agiven pathlength should be the same no matterhow that total pathlength is subdivided into sub-steps, each with its own de¯ection and displace-ment. A robust multiple-scattering algorithm has
A.F. Bielajew, F. Salvat / Nucl. Instr. and Meth. in Phys. Res. B 173 (2001) 332±343 335
been previously developed using the screenedRutherford cross-section [10]. However, we havedecided to employ a di�erential cross-section(DCS) with more physical content, one based on apartial-wave analysis.
3.1. Generation of multiple-scattering angles
DCSs for elastic scattering of electrons byneutral atoms have been calculated using thePWADIR code of Salvat and Mayol [11]. This
Fig. 1. f surface for 06 n6 1 and 06 c6 3.
Fig. 2. r surface for 06 n6 1 and 06 c6 3.
336 A.F. Bielajew, F. Salvat / Nucl. Instr. and Meth. in Phys. Res. B 173 (2001) 332±343
code computes relativistic phase shifts from thenumerical solution of the radial Dirac equation,for a given interaction ®eld, and determines the
corresponding DCS. We have adopted the pa-rameterization of the Dirac±Hartree±Fock±Slater(DHFS) screened potential given by Salvat et al.
Fig. 3. k surface for 06 n6 1 and 06 c6 3.
Fig. 4. c surface for 06 n6 1 and 06 c6 3.
A.F. Bielajew, F. Salvat / Nucl. Instr. and Meth. in Phys. Res. B 173 (2001) 332±343 337
[12], which leads to essentially the same DCSsas the numerical DHFS ®eld. The e�ect of ex-change between the projectile and the electronsin the target atoms has been accounted for bymeans of the approximate local ®eld correctionof Furness and McCarthy [13]. With thisscheme, elastic DCSs can be calculated for in-cident electrons with energies up to a few MeV.For higher energies, the numerical calculationbecomes prohibitively lengthy and one must relyon approximate factorization methods. It shouldalso be noted that the physical model (static®eld approximation) loses validity for projectileswith energies of the order of 1 keV and less,since slow projectiles may cause appreciablepolarization of the target atom. Fortunately,these energies are below the range of interest ofmost transport calculations. Our computer pro-gram generates a table of DCS values for a gridof about 600 scattering angles suitably distrib-uted (logarithmically for small de¯ections anduniformly for large de¯ections). The DCS atother angles is obtained by linear interpolation.Computed DCSs for the elements C (Z � 6) andPB (Z � 82) and electrons and positrons withdi�erent energies are displayed in Figs. 6 and 7,respectively.
The moments g`, Eq. (3), of the single-scatter-ing distribution, disregarding energy loss, can beexpressed as
g` � NAqA
r0� � f`�; �17�
where
r0 � 2pZ 1
ÿ1
dlr�l� �18�
Fig. 6. Partial wave DCS for elastic scattering of electrons with
the indicated kinetic energies by carbon atoms.
Fig. 5. d surface for 06 n6 1 and 06 c6 3.
338 A.F. Bielajew, F. Salvat / Nucl. Instr. and Meth. in Phys. Res. B 173 (2001) 332±343
is the total atomic cross-section and
f` � 2pZ 1
ÿ1
dlr�l�P`�l�: �19�
The quantities f` have been calculated numericallyfrom the partial-wave DCS by using the followingalgorithm. First, the integration interval (ÿ1; 1) issplit into a number of subintervals in such a waythat the DCS varies by less than a factor of 10within each subinterval. Then, a 500-point Gaussquadrature formula is used to evaluate the integralwithin each subinterval. Since the Legendre poly-nomials are generated by using the upward re-cursion relation, the algorithm can be coded tocompute simultaneously all moments up to a givenorder, 600 in the present work. The calculation ofthese moments is very fast, a few seconds on a 366MHz IBM compatible PC.
To check the accuracy of the calculated mo-ments, we compared the original DCS with theresult of adding up its Legendre expansion,
r�l� �X1`�1
2`� 1
4pf`P`�l� ÿ `�l�: �20�
For low-energy electrons, when the DCS is rela-tively wide and its Legendre series converges rap-idly, this comparison is satisfactory. Di�erences
between the original DCS and the ``reconstructed''one are less than 0.001%. In principle, the accuracyof the calculated moments should be independentof the energy, since the DCS varies slowly in eachsubinterval. The Goudsmit±Saunderson distribu-tion is obtained by summing up its Legendre ex-pansion (after removing the no-scattering part)and is expected to be as accurate as the recon-struction of the DCS, provided that the Legendreseries actually does converge. At high energies,convergence with the 600 calculated moments isobtained only if the pathlength is large enough.
The random sampling of the scattering angle,from both the single-scattering DCS (analog sim-ulation) and from the GS distribution (class Icondensed simulation), is performed as follows.We start from a table of values of the corre-sponding probability distribution (not necessarilynormalized) at the points of the angular gridmentioned above. To generate random de¯ectionangles, we apply the inverse transform method tothe (continuous) distribution obtained by linearinterpolation within this table (i.e., the distributionof sampled values is done exactly from a piecewiselinear distribution). With the aid of a binary-search method, this sampling algorithm is very fast(about 300,000 random values generated per sec-ond on a 366 MHz personal computer).
3.2. Implementation of the transport mechanics
In the numerical implementation of the algo-rithm, use is made of the fact that the azimuthaldirection of either the scattering angle or the spa-tial displacement is arbitrary for unpolarizedscattering. We adopted the following approach:1. Start with an electron with initial position~x0 and
direction ~X0 and sample the multiple-scatteringangles H�s� (from the Goudsmit±Saundersondistribution) and U1 (uniformly in �0; 2p�).
2. Do a partial transport of the particle to thepoint at which the additional lateral stragglingtakes place. i.e.,
x=s � fr sin H�s� cos U1;
y=s � fr sin H�s� sin U1;
z=s � k�1ÿ r� � c� �kr � d� cos H�s�;�21�
Fig. 7. Partial wave DCS for elastic scattering of electrons with
the indicated kinetic energies by lead atoms.
A.F. Bielajew, F. Salvat / Nucl. Instr. and Meth. in Phys. Res. B 173 (2001) 332±343 339
relative to an initial direction along the z-axisbut account, via rotation and translation, forthe particle's actual direction and position.
3. Apply the additional lateral straggling
Dx=s � r cos H�s� cos U2;
Dy=s � r cos H�s� sin U2
�22�
in a plane perpendicular to the particle's initialdirection.
4. Rotate the particle's direction accounting forthe scattering angles, H�s� and U1.
4. Simulation results
To demonstrate the quality of the new trans-port mechanics algorithm, we performed simula-tions of spatial and angular distributions ofelectrons with various energies after traveling dif-ferent pathlengths in a number of elements. Thesimulations probed the practical limits of the cparameter and cross-section shape. However, herewe will only present the graphical results for 100keV electrons in Pb with pathlengths of 75 meanfree paths (MFPs). This corresponds to c � 2:185and n � 0:507. Results from the new algorithm arecompared here with those from equivalent analog(collision by collision) simulations using the samesingle-scattering DCS, which provide essentiallyexact results. The comparison also includes resultsfrom PENELOPEs transport mechanics, whichwere obtained for the same cases as the new me-chanics.
In Figs. 8±10, we show the distributions for z,r � ��������������
x2 � y2p
and x. The results were obtainedusing 6� 107 histories. The one-step distributionsdi�er manifestly from the analog distributions,re¯ecting the somewhat arti®cial nature of thetransport mechanics. The shapes of the distribu-tions obtained from the new mechanics are gen-erally closer to the analog distributions than theresults from PENELOPEs mechanics; in particularthe p�r� distributions are peaked at ®nite values ofr, in accordance with the analog results, whereasPENELOPE sets the most probable lateral dis-placement at r � 0.
When the electron pathlength is split into twoequal steps, the shapes of the simulated distribu-tions improve substantially for both schemes. Figs.11±13 display results of the two-step simulationsfor the same case studied above. Two-step distri-butions are twice as expensive to simulate thanthose with one step, but the extra cost is largely
Fig. 8. The distribution of z for 100 keV electrons in lead taking
a single-step of pathlength 75 elastic scattering MFPs.
Fig. 9. The distribution of r ���������������x2 � y2
pfor the simulation
described in the caption of Fig. 8.
340 A.F. Bielajew, F. Salvat / Nucl. Instr. and Meth. in Phys. Res. B 173 (2001) 332±343
compensated by the gain in accuracy. It is satis-fying that the crude details of the one-step distri-butions are almost completely washed out, forboth the new transport mechanics and PENELO-PE (which both use the the same physical infor-mation). The only visible artifact is the little bump
in the lateral distributions, which disappears whenthe pathlength is split in four or more steps. Theseresults do not imply that calculations with PEN-ELOPEs and the new mechanics are equally ac-curate. Inspection of the spatial moments andspace±angular correlations obtained by the twomethods shows that the new mechanics is more
Fig. 10. The distribution of x for the simulation described in the
caption of Fig. 8.
Fig. 11. The distribution of z for 100 keV electrons in lead
taking two equal steps comprising a total pathlength of 75
elastic scattering MFPs. 6� 107 histories were employed in this
simulation. This case corresponds to c � 2:185 and n � 0:507.
Fig. 12. The distribution of r ���������������x2 � y2
pfor the simulation
described in the caption of Fig. 11.
Fig. 13. The distribution of x for the simulation described in the
caption of Fig. 11.
A.F. Bielajew, F. Salvat / Nucl. Instr. and Meth. in Phys. Res. B 173 (2001) 332±343 341
accurate and, therefore, it allows simulating agiven pathlength in fewer steps.
This is illustrated in Table 1, where we givespatial moments and space±angular correlationsfor 100 keV electrons in lead; the pathlength is halfa transport mean free path (i.e., n � 0:507); eachsimulation involved the generation of 107 histories.We see that PENELOPEs mechanics gives resultsthat deviate in the expected way, according to Eq.(8). In particular, it gives values of the longitudinalmoments, hzi and hz2i, that are systematically toolarge, even when the pathlength is divided into sixsteps. On the other hand, the new mechanics givesthe correct moments in a single step. The cost ofimplementing the new algorithm is indicated in thelast line of the table, where we give the number ofseconds to simulate 106 histories on a 366 MHzPII. We see that the computation cost of the newalgorithm is about three times that of the randomhinge in this particular example.
5. Concluding remarks
The new electron transport mechanics algo-rithm provides a more accurate description ofspatial displacements than previous approaches.Although its accuracy can be matched by PENE-LOPEs mechanics, the latter will usually requiresplitting the pathlength into a larger number ofsteps. For pure elastic scattering, it has beenshown that two-step simulations already yieldfairly accurate space±angular distributions, which
have their ®rst moments correct (apart from sta-tistical ¯uctuations).
In principle, the new algorithm could be im-proved, e.g. by straggling some of the parametersin Eq. (9), to yield qualitatively improved one-stepdistributions. However, this is far from trivial. Inpractice, it may be equally expedient (and proba-bly faster) to appropriately increase the number ofsteps. It should be noted that in real simulationsthe number of steps per electron trajectory will beof the order of 10 or larger. Under these circum-stances, the e�ect of any possible improvement ofthe transport mechanics on the ®nal results will behardly seen.
Although the present paper has been limited topure elastic scattering for simplicity, the newtransport algorithm can be readily combined withthe continuous-slowing-down approximation toinclude energy losses. It is also particularly ame-nable for use in mixed, class II simulations, whereit can allow increasing the cuto� angle (i.e. re-ducing the number of elastic hard events to besimulated) considerably. Work along these lines isin progress.
Acknowledgements
We gratefully acknowledge the numerous dis-cussions with Prof. Ed Larsen (Department ofNuclear Engineering and Radiological Sciences,University of Michigan), Dr. Jos�e Mar�õaFern�andez-Varea (Universitat de Barcelona),
Table 1
Spatial moments and space±angular correlations of the multiple-scattering distributions of 100 keV electrons after traveling 75 MFPs
in lead (n � 0:507). A � hx sin H cos U� y sin H sin Uia
a Numbers in parentheses are statistical uncertainties (3r) in units of the last signi®cant ®gure of each value.
342 A.F. Bielajew, F. Salvat / Nucl. Instr. and Meth. in Phys. Res. B 173 (2001) 332±343
Dr. Josep Sempau (Universitat Polit�ecnica deCatalunya, Barcelona) and Mr. Ernest Benedito(Universitat de Barcelona) who kindly provided afast calculational method for computing Legen-dre moments. Additionally, Mr. David Lorch(University of Michigan) is thanked for assis-tance with the ®gures. One of us (AFB) grate-fully acknowledges partial ®nancial support fromLLNL and ADAC Laboratories (Milpitas, Cali-fornia) as well as a PVI grant from the Univer-sity of Barcelona.