scieee Science in your language
[en] (orig)

Validations of the radiation transport module NEUTRO: A deterministic solver for the neutron transport equation

Abstract

We present significant improvements and validations of a deterministic neutron transport code (NEUTRO) dedicated to solving the Boltzmann Transport Equation. The code is integrated as a module in the Alya software package developed by the Barcelona Supercomputing Center which uses the Discrete Ordinates Method on angular coordinates, multi-group for energy discretization and FEM on unstructured meshes to treat special complex domains. The anisotropy of the scattering medium is introduced into the scattering kernel using real base expressions for spherical harmonics. In order to build the total cross-section and the respective group matrix for the elastic cross-section, we use the NJOY code. We test the solver using different geometries, orders of integration for the angular discretization and number of energy groups. Finally, we compare our results against benchmarks obtained from an NEA database that reported measurements of leakage spectra of several materials.

Read accessible full text

Validations of the radiation transport module NEUTRO: A deterministic solver for the neutron transport equation

Author: Soba, Alejandro,Cazado, Mauricio E.,Houzeaux, Guillaume,Gutierrez Milla, Albert,Mantsinen, Mervi J.,Saez Pous, Xavier
Publisher: Elsevier
Year: 2021
DOI: 10.1016/j.fusengdes.2021.112497
Source: https://upcommons.upc.edu/bitstream/2117/342302/1/postprint-4.pdf
Valida ions agains SINBAD benchma k o he adia ion anspo module,
a de e minis ic sol e o he neu on anspo equa ion.
Alejand o Soba1,2, Mau icio E. Cazado2, Guillaume Houzeaux1, Albe Gu ie ez-Milla1,
Me i J. Man sinen1,3 and Xa ie Saez1
1 Ba celona Supe compu ing Cen e , Ba celona, Spain.
2 CNEA-CONICET, Buenos Ai es, A gen ina.
3 ICREA, Ba celona, Spain.
Keywo ds: Radia ion anspo , neu on-wall in e ac ion, usion eac o s shielding, SINBAD.
Abs ac
We p esen signi ican imp o emen s and alida ions o e a de e minis ic neu on anspo
code (NEUTRO) dedica ed o sol ing he Bol zmann T anspo Equa ion. The code is
in eg a ed as a module in he Alya sys em de eloped by he Ba celona Supe compu ing Cen e
which uses he Disc e e O dina es Me hod o e an angula po ion, mul i-g oup o ene gy
disc e iza ion and FEM o e uns uc u ed meshes o ea special complex domains. Ma e ial
aniso opy o he sca e ing medium is in oduced in o he sca e ing ke nel using eal base
exp essions o sphe ical ha monics. In o de o build he o al c oss-sec ion and he espec i e
g oup ma ix o he elas ic c oss-sec ion, we use he NJOY code. We es he sol e using
di e en geome ies, o de s o in eg a ion o he angula disc e iza ion and numbe o ene gy
g oups. Finally, we compa e some esul s agains benchma ks ob ained om an IAEA da abase,
basically measu ing leakage e ec s o se e al ma e ials.
1. In oduc ion
In his pape , we p esen some imp o emen s and alida ions o a de e minis ic adia ion
anspo code based on he Bol zmann T anspo Equa ion and in eg a ed in o he Alya sys em
a Ba celona Supe compu ing Cen e [1], which is in ended o mul i-physics applica ions in
he nuclea usion ield [2, 3]. Imp o ing he unde s anding o his ma e is essen ial o he
de elopmen o u u e usion eac o s, which equi es highly demanding simula ions o he
neu on anspo p ocesses. In pa icula , sol ing his kind o adia ion anspo equa ions in
o de o p edic he beha iou o neu on damage and adia ion-ma e in e ac ion demands high
memo y equi emen s and a high compu a ional complexi y [4].
The main idea behind his de elopmen has a double impac by in eg a ing wo complex
esea ch ields. On he one hand, i will be aligned wi h he Eu opean Fusion Roadmap [5] o
he ealiza ion o usion ene gy, being ocused on he i ium sel -su iciency om he Mission
4 and he design o he Tes Blanke Module (TBM). I will p o ide a cu ing-edge
compu a ional ool o add essing his complex mul i-physics p oblem bo h in exis ing and
u u e usion de ices which is in inc easing demand wi hin he usion communi y. On he o he
hand, i has he po en ial o become a d i e o u u e usion eac o design challenges gi en
i s High-Pe o mance Compu ing (HPC) capabili ies which make he analysis o di e en
design a ia ions easible.
This new modelling ool o usion is based on he pa allel compu a ional mechanics code Alya
[6, 7]. Alya is a amewo k c ea ed o sol e di e en physical phenomena in a coupled way in
la ge scale supe compu e s, accomplishing high pa allelism s anda ds and scalabili y. The code
0DQXVFULSW1HZ6XEPLVVLRQ
Final e sion a ailable ia: h ps://doi.o g/10.1016/j. usengdes.2021.112497
is o ganized in a modula way including se ices, ke nel and modules. The ke nel is he co e o
Alya and implemen s inpu /ou pu ope a ions, ou ines o he execu ion o he main loop and
all ope a ions ela ed o he mesh geome y. Se ices cons i u e he unc ionali ies o e ed o
he execu ion, o which he mos impo an is he se ice ha deals wi h he di ision o he wo k
and is esponsible o he communica ion be ween nodes. Alya’s modules, as in he case o he
NEUTRO module c ea ed o neu on anspo , implemen he se o Pa ial Di e en ial
Equa ions (PDEs) which de ine he physics o he p oblem. The modules a e coded and
compiled independen ly and can be easily coupled oge he . Using FEM in ol ing millions o
billions o elemen s equi es gene al spa se sys ems o equa ions sol ed using i e a i e
me hods. Mode n supe compu ing a chi ec u es consis o independen compu ing nodes,
ypically mul ico e and he e ogeneous a chi ec u es, in e connec ed by a sha ed communica ion
ne wo k. Alya uses he Message Passing In e ace (MPI) lib a y o implemen communica ions
be ween nodes and Open Mul i-P ocessing (OpenMP) o sha ed-memo y pa allelism. De ailed
desc ip ions o he s uc u e o he Alya code can be ound in e e ences [1, 6-9]. The isual
pos -p ocessing o he esul s was pe o med wi h Pa a iew so wa e [10].
The sol e uses he Disc e e O dina es Me hod [11-13] o e an angula po ion, a mul i-g oup
me hod o he ene gy disc e iza ion and he Fini e Elemen Me hod (FEM) [9, 14] o e
uns uc u ed meshes o ea special complex domains. We use he NJOY code [15] o build he
o al c oss-sec ion and he espec i e g oup ma ix o he elas ic c oss-sec ion. NJOY eads he
da a om an ENDF/B da abase [16] p o ided by Los Alamos Na ional Secu i y o ice. We es
he sol e in wo-dimensional geome ies (slabs o di e en dep hs), and h ee-dimensional
geome ies using laye s and sphe es. Quad a u es o se e al o de s in he equal-le el disc e e
o dina e se a e a emp ed [17, 18] and due o he sca e ing o high ene gy 14 MeV usion
neu ons being highly aniso opic, a new ype o quad a u e based in he FEM disc e iza ion
[19] is conside ed. In addi ion, we ha e es ed he sol e using di e en ene gy g oups
disc e iza ions, om 3 o 256, and he esul s ob ained we e alida ed agains o he codes and
he SINBAD (Shielding In eg al Benchma k A chi e Da abase) [20] benchma ks, ob ained
om an IAEA da abase, in which he leakage e ec s o se e al ma e ials we e measu ed.
Al hough he alida ions included in his wo k a e compa ed wi h he neu on anspo esul s,
he code is p epa ed o sol e p oblems o gamma adia ion anspo as well [21, 22].
Valida ions o he la e will be p esen ed in u u e wo ks.
I is impo an o men ion ha al hough he e a e many codes in he wo ld ha sol e he
adia ion anspo equa ion [23-26], none o hem is in eg a ed wi h compu a ional mechanics
amewo ks o mul i-physics coupled p oblems and hey ha e been de eloped as s and-alone
so wa e. In his sense, he in eg a ion o he NEUTRO module in o he Alya sys em has an
inno a i e aspec wi h espec o o he be e -known solu ions. The inal in en ion o his
de elopmen is o achie e a mul i-physics ea men o high- olume en i onmen s composed
o mul iple ma e ials, being able o sol e hei he mal, mechanical and neu onic beha iou a
he same ime, using he coupling capabili ies p o ided by he Alya sys em. In his espec , we
made a deep scalabili y s udy o he sol e , showing he beha iou o he code and he quali y
o he esul s as we inc ease he numbe o p ocesso s in ol ed in he p oblem sol ing. Some
o hese scalabili y measu es a e included in sec ion 3.
This pape is o ganized as ollows: in sec ion 2, we b ie ly desc ibe he sol ing amewo k o
he NEUTRO module included in Alya. In sec ion 3, we include some o he es s ca ied ou
wi h his module and we show some nume ical d awbacks ha had o be sol ed in o de o
achie e a good quali y in he inal esul s. In addi ion, we p esen some scalabili y esul s
including he beha iou o he code when a la ge numbe o p ocesso s a e in ol ed in he
calcula ions. In sec ion 4, we show he compa isons agains some selec ed expe imen s o he
SINBAD da abase. Finally, in sec ion 5, we p esen ou conclusions.
2. Nume ical adia ion anspo equa ion
The gene al s a iona y adia ion anspo equa ion sol ed a he momen in he NEUTRO
module included in Alya has he ollowing exp ession [7-10]:
Ω∙∇φ+Σ$(%,&,Ω)'=* * Σ-(%,&.→&,Ω.→Ω)'0Ω.0&.
12
3
4
(1)
Whe e
Ω: Solid angle o he possible di ec ion in whe e a pa icle a els.
E: Pa icle ene gy.
: Gene al spa ial a iable.
φ: Flux as a unc ion o , E and Ω.
Σ$( ,E,Ω): To al mac oscopic c oss-sec ion.
Σ-( ,E.→E,Ω): To al mac oscopic sca e ing c oss-sec ion.
The ene gy mul i-g oup app oxima ion [9, 27, 28] and he disc e iza ion o angula dependence
we e applied o equa ion (1). Fo he la e , he Me hod o Disc e e O dina es (equal-le el se ,
Sn) [11-14] o quad a u es se based on discon inuous ini e elemen s [19] may be chosen. The
mul i-g oup me hodology consis s o di iding he neu on ene gy ange in se e al g oups (G)
and e alua e all he ene gy-dependen pa ame e s in ela ion o hose ene gy g oups G.
Disc e iza ion o e he angula (di ec ional) a iable is usually pe o med by means o
mechanical quad a u e. The con inuous angula a iable Ω is eplaced by a se o disc e e
di ec ions, each o which has a ela i e weigh . The di ec ions may be ep esen ed as poin s on
he uni sphe e.
Using a mul i-g oup app oxima ion and an angula disc e iza ion, he in eg al-di e en ial
anspo equa ion is con e ed in o a sys em o G×M di e en ial equa ion as ollows:
78∙∇φ98 +Σ$;%,&9<φ98 = > > Σ-9?,9 8?,8(%)φ9?8?
@
8?AB
8.C8
D
9?AB
9.C9
(2)
Wi h F =1,…,I; K=1,…,L
Ma e ial aniso opy o he sca e ing medium is in oduced in o he sca e ing ke nel using eal-
based exp essions o sphe ical ha monics. Combining his app oxima ion wi h he ene gy
g oup di ision and disc e e o dina e me hods, he ollowing exp ession o he sca e ing e m
is ob ained:
    4
2!+1Y"#"(Ω#)
"
#"%&"
'
"%*
,
-.%/
0
3.%/ Y
5"#"(Ω#6) Σ#"3.,3 -6,-(:)φ36-6(:)
(3)
( he o de o he sphe ical ha monics was changed o ‘ml’ o a oid con usion). In his
exp ession, a maximum alue o L is chosen o cu he expansion. I L=0, he ke nel exp ession
in equa ion (2) is eco e ed. Any alue o L g ea e han ze o in oduces angula aniso opy.
The mos gene al bounda y condi ion o he anspo equa ion is he Albedo condi ion [1,9],
which consis s o :
φ<=( ,E,Ω!")= φ!" + αφ( , E, Ω%)
(4)
whe e φ!" is a ixed lux, α akes he alue 0 o he acuum bounda y condi ion and α=1 o
he gene al e lexi e condi ion. Gene ally, o model a e lec o su ounding he co e he alue
o α is 1. O e all, he Albedo coe icien α may a y wi h he e lec o p ope ies and should
also cap u e angula and ene ge ic edis ibu ion o he e lec ed neu ons due o hei di usion
h ough he e lec o .
The e a e wo main s eps when implemen ing mi o bounda y condi ions. Fi s ly, i is
necessa y o calcula e he no mal ec o o he su ace on o which he ays a e e lec ed.
Secondly, he calcula ion o he e lec ion angle is equi ed. Since we a e wo king wi h disc e e
o dina es we need o p ojec he ajec o y on o one o he disc e ized di ec ions ha we ha e.
In each bounda y whe e he e lec ing condi ion is u ned on, he algo i hm de eloped
de e mines which di ec ion is he closes o he cu en di ec ion o e lec ion o use i in he
building o he bounda y con ibu ion. In he disc e e o dina es me hod, gi en a di ec ion used
in he angula disc e iza ion, i is necessa y o e alua e he co esponding specula e lec i e
di ec ion a e he neu on a i ed a he e lec i e medium. To do so, a special unc ion was
p og ammed in o de o de e mine a he beginning o he simula ion he co esponding
di ec ion o each no mal p esen in he limi s o he domain o ou p oblem. Using his
p ocedu e, bo h angula di ec ion and e lec i e di ec ion a e e alua ed o e each bounda y
whe e a e lec i e condi ion is equi ed.
In o de o build he o al, cap u e and elas ic sca e ing c oss-sec ions and he espec i e g oup
ma ix o elas ic and ission c oss-sec ion, he NJOY code [15] was used. NJOY eads and
p ocesses da a om ENDF/B da abase [16]. F om SINBAD benchma k da abase, some
exe cises des ined o analyse he leakage e ec o se e al ma e ials we e selec ed. Ini ially, he
inpu iles wi h he o al and sca e ing ma ix o i on, nickel, aluminium, ungs en, manganese
and silicon wi h NE-ene gy g oups we e c ea ed. We es ed wi h NE = 238, 118 and 59 g oups
o compa e wi h he benchma k cases. Fo mos o he shielding expe imen s aken om he
SINBAD da abase, we used 59 ene gy g oups. In addi ion, o es ing coa se de ini ions, 29-
ene gy g oups and 3-ene gy g oups we e used.
3. Nume ical d awbacks and pe o mance analysis
3.1 Angula disc e iza ion: Once he de elopmen o he mul i-g oup equa ion was comple ed
and be o e going o he alida ion s age, we ca ied ou a se ies o nume ical expe imen s o
e i y he unc ioning o his new e sion o he code wi h espec o wha was p e iously
p og ammed, as well as i s in e ac ion wi h he Alya sys em. In he i s place, he solu ion o a
sphe e wi h an in e nal hole wi h an inne lux in he adial di ec ion and acuum in he ex e nal
bounda y was deeply analysed. We used a e ahed al mesh o linea elemen s o disc e ize he
eal space. As he in en ion was o analyse he solu ion wi h he di e en quad a u e o mulas
implemen ed in he code, a h ee-ene gy g oups disc e iza ion was used. Fo compa ison, we
analysed he solu ion wi h di e en app oxima ions o he equal le el me hods and he
implemen ed angula disc e iza ion based on he ini e elemen me hod wi h 512 poin s (called
in e nally in he code FE64). This me hod consis ed o he disc e iza ion o he medium plane
o he sphe e using a iangula mesh and he p ojec ion o e he sphe e su ace o each iangle
wi h a well-de ined weigh [19]. In Figu e 1, we p esen a gene al iew o he code capaci ies.
Figu e 1 a) shows a domain di ision o be dis ibu ed among 16 nodes, used o sol e he eal
space using FEM. Figu e 1 b) p esen s he neu onic cu en in he z di ec ion showing he
symme y o he solu ion ob ained. Figu e 1 c) is a con ou zone o one ene gy g oup as an
example solu ion o he domain. All hese solu ions we e ob ained using he mo e accu a e
angula disc e iza ion (FE64). Las ly, in Figu e 1 d), we plo he o al lux in all he nodes on
he su ace o he sphe e o he di e en angula disc e iza ions implemen ed in he code. I
can be app ecia ed how he solu ions ob ained wi h he equal le el me hod Sn4 (24 poin s) and
Sn8 (80 poin s) p esen ed a g ea dispe sion and nega i e alues o he solu ion. When we
passed o Sn10 (120 poin s) and Sn12 (168 poin s) posi i e alues and less dispe sion in he
solu ion we e ob ained. Finally, o he quad a u e based on FE he esul s achie ed show he
leas dispe sion, ob aining he uni o mi y ha he symme y o he solu ion p oposes. Le us
highligh ha he Sn12 me hod has 168 in eg a ion poin s in all he sphe e while he FE64 has
512 in eg a ion poin s, inc easing he compu a ion ime o he p oblem p opo ionally. Bu he
mo e impo an esul is ha using he highes quad a u e solu ion, we a oid he spu ious
nega i e alues in he lux solu ion, a ecu ing p oblem in he use o disc e e me hods o sol e
he anspo equa ion.
Figu e 1: a) Pa i ioned mesh o pa allelize he solu ion o he eal space. b) Neu onic cu en in z
di ec ion. c) Flux solu ion o one selec ed ene gy g oup d) Dispe sion o he su ace solu ion o di e en
modes o angula disc e iza ion.
3.2 Compa ison be ween 2D and 3D solu ions: An impo an and e ec i e way o accele a e a
p oblem om he nume ical poin o iew is o educe he dimensions o he domain o be

analysed when some ypes o symme y a e p esen . This educ ion, in he case o anspo
p oblems, in ol es educing he physical domain bu also he angula disc e iza ion by omi ing
ou oc an s o angula di ec ions. On he one hand, his sys em inc eases he esolu ion speed,
bu a he same ime enla ges he e o o he solu ion. In his sub-sec ion, we compa e a solu ion
ob ained on a h ee-dimensional laye and a slab o equal wid h, bu wo-dimensional. Bo h
cons i u ed by he same ma e ial, and wi h he same inne lux om he adial di ec ion (x in
he slab2d and z in he 3D laye ). A acuum bounda y condi ion is conside ed in he ou pu
di ec ion and e lec i e bounda ies in he es o he limi s. Figu e 3 shows he o al lux
ob ained on he ex e nal ace as a unc ion o one o he signi ican coo dina es. The solu ion is
di ided by he maximum ob ained in he h ee-dimensional case, eaching a di e ence o 14%
in he esul be ween bo h simula ions.
Figu e 2: Solu ion o a wo-dimensional slab compa ed wi h a h ee-dimensional laye o he
same wid h
3.3 Pe o mance Analysis: Be o e discussing he pe o mance analysis, a ew ema ks should
be made abou he pa alleliza ion in Alya. Fi s o all, he implemen ed pa alleliza ion s a egy
is he main-seconda y echnique. This means ha when he pa alleliza ion akes place, he main
p ocess is in cha ge o dis ibu ing he da a and coo dina ing he o he p ocesses (seconda y).
The seconda ies a e he ones ca ying ou he calcula ions. All p ocesses communica e among
hemsel es h ough messages using he MPI s anda d, which p o ides synch oniza ion and
communica ion unc ionali ies be ween a se o p ocesses wi h dis ibu ed memo y (meaning
ha no all p ocesses can access he same memo y space and he e o e need o exchange da a
among hem). The Alya sys em is w i en in FORTRAN language: MPI is implemen ed as a
lib a y wi hin he code om which we can call he unc ions we may need once we ha e
included i , as we would do wi h any o he lib a y.
Ano he impo an ea u e o he Alya implemen a ion is he possibili y o mul iplying he ini ial
mesh. Using his op ion each ini e elemen is di ided in o up o 8 ini e elemen s o he same
cha ac e is ics: his implies an inc ease in he numbe o elemen s o he inpu mesh. In o de
o es he pe o mance, a h ee-dimensional sphe e o Fe56 ma e ial wi h an in e io sou ce in
he adial di ec ion using 59-ene gy g oups and Sn12 o he angula quad a u e (see igu e 1)
we e used. Mul iplying by 2 and by 3, meshes o 77984 and 623872 ini e elemen s a e de ined,
espec i ely. In Figu e 3 he speed up o each case is p esen ed. In he case o 77984 elemen s,
a 95% e iciency wi h 256 p ocesso s is achie ed, while in he case o 623872 elemen s, an
82% e iciency wi h 1024 p ocesso s and 70% wi h 2048 p ocesso s a e ob ained. Bo h cases
show sa u a ion due o he inc ease in communica ion. No e ha his e iciency is measu ed
wi h espec o he un on 16 ( o he small case) and 128 ( o he big case) MPI p ocesses.
Despi e he sa u a ion men ioned, he esul s show good pe o mance wi h mo e han wo
housand p ocesso s.
Fo all he calculus we used a gene alized minimal esidual algo i hm (GMRES) [29] o sol e
he nonsymme ic linea sys em p oduced by his ype o p oblem. We used a diagonal
p econdi ione , which esul ed in he mos sui able op ion o each con e gence in less ime.
Figu e 3: Speedup o wo cases, using wo meshes, 77984 elemen s wi h an a e age olume o
6.73x10-6 and 623872 elemen s wi h 8.4x10-7 a e age olume.
4. SINBAD es ing
In o de o alida e he de eloped NEUTRO module, we selec ed a numbe o benchma k cases
o es ing pu poses om SINBAD, an OECD/NEA Da a Bank (O ganiza ion o Economic
Coope a ion and De elopmen , Nuclea Ene gy Agency Da a Bank) and ORNL/RSICC (Oak
Ridge Na ional Labo a o y, Radia ion Sa e y In o ma ion Compu a ional Cen e) join p ojec
[20, 30]. The main objec i e o hese es s was o p oduce a da abase con aining an
in e na ionally es ablished se o adia ion shielding and dosime y da a ela i e o expe imen s
ele an o eac o shielding, usion blanke neu onics and accele a o shielding, o e ing a
s anda dized ool o alida ion o adia ion anspo compu e codes and nuclea da a lib a ies.
Ou pa icula analysis was based on he se o expe imen s ha ook place in Osaka Uni e si y
a he OKTAVIAN acili y. De ailed desc ip ions o he expe imen s, he de ec o sys em and
he sou ce spec um used can be ound in e e ences [31-36]. In summa y, his g oup o es s
used hollow sphe es wi h an inne neu on sou ce wi h a p onounced peak in he ange o 14
MeV. The sphe es we e made o di e en ma e ials (Fe, Ni, Al, W, Mn, Si, among o he s) and
wi h di e en diame e s in a ange o 14 cm o 100 cm. The main pu pose o hese benchma ks
was o s udy he neu on and gamma- ay leakage spec a om he ou e su ace o he sphe e.
The e a e wo kinds o measu emen s ca ied ou depending on he posi ion o he de ec o . Fo
some ma e ials (Mn, Si, Al, W), he neu on de ec o was loca ed wi h a 55° angle wi h espec
o he cen al axis o he sphe e [35, 36]. In o he cases, such as o Fe and Ni, he de ec o did
no co e he en i e su ace o he sphe e bu he solid angle o 17.28° om he sphe e cen e
due o he p esence o he collima o s [32-34].
In o de o pe o m he es , wo ypes o models we e used: a wo-dimensional model, using
slabs wi h a wid h equal o he sphe ical shell hickness; a h ee-dimensional laye wi h a wid h
equal o he sphe ical shell hickness and a h ee-dimensional sphe e wi h an in e nal hole. Table
I p esen s a summa y o he es s pe o med and he ma e ials used.
Table I: Desc ip ion o he Benchma k used in his wo k [31,36]
Model
Ma e ial
Ex .
Radi
us
(cm)
In e nal hole
adius
(cm)
E ec i e
hickness (cm)
De ec o posi ion
Sphe e o 40
cm
Silicon 20 10 10
55º de ec o
Sphe e o 60
cm
Silicon 30 10 20
55º de ec o
Sphe e o 60
cm
Mangane
se
30 10 20
55º de ec o
Sphe e o 40
cm
Aluminium
20
10 10
55º de ec o
Sphe e o 40
cm
Tungs en 20 10 10
55º de ec o
Sphe e o 32
cm Nickel 16 2 14
17.28º solid angle
om he
sphe e
cen e
Sphe e o
100 cm I on 50 10 40
17.28º solid angle
in he
sphe e
cen e
In Figu e 4, compa isons be ween expe imen al esul s and se e al benchma ks un wi h Alya
using wo-dimensional slabs a e p esen ed. In each case, he wid h o he slabs ep esen s he
e ec i e hickness o he sphe e in each expe imen . Al hough o he cases o aluminium o
10 cm hickness (Figu e 4 a), manganese o 20 cm hickness (Figu e 4 b), i on o 40 cm
hickness (Figu e 4 c) and ungs en o 10 cm hickness (Figu e 4 d) he esul s show some
dispe sion in he ange o ene gy be ween 0.01-1 MeV, he solu ion ob ained using he
NEUTRO module shows a good ep esen a ion o he expe imen al da a endency.
Figu e 4: Expe imen al measu ed and Alya esul s o Leakage s. Ene gy o SINBAD es using
wo-dimensional slabs. Be ween pa en heses he wid h o he slabs conside ed, co esponding
o he hickness o he sphe e on which he expe imen s we e ca ied ou . 59 ene gy g oups
we e used o ene gy disc e iza ion.
Figu e 5 p esen s he compa ison be ween expe imen al da a and calcula ed alues o h ee-
dimensional domains, bo h laye s and sphe es. In each case, we used meshes o linea
hexahed a o he laye s and linea e ahed a o he sphe es. The Sn8 quad a u e o he laye
and FE64 o he sphe es we e used. 59 ene gy g oups we e used o he ene gy disc e iza ion.
The esul s ob ained o a silicon laye o 20 cm hickness (Figu e 5 a), a manganese laye o
20 cm hickness (Figu e 5 b), a ungs en sphe ical shell o 20 cm hickness (Figu e 5 c) and a
nickel sphe ical shell o 14 cm hickness (Figu e 5 d) show some dispe sion in di e en anges
o ene gy, depending on he expe imen s.
In gene al, he esul s o h ee-dimensional geome ies a e hose ha p esen g ea e dispe sion
wi h espec o he expe imen al measu emen s. And among hese, hose co esponding o
sphe ical geome ies a e he a hes om he measu ed alues. The e a e se e al easons ha
con e ge o hese esul s. The i s is he e o ha an uns uc u ed mesh on a ci cula geome y
in oduces, which in luences he en i e calcula ion om he in e nal bounda y condi ion, whe e
he incoming low is ixed, o he ou e edge o he domain. Secondly, he ex ension o he
domain in ol ed, which, al hough i is ea ed by a la ge numbe o elemen s, is s ill mo e
di icul o ep esen by means o a mesh o e eal space. This di icul y is pa ly a esul o
using wo-dimensional domains, which allow dense meshing wi hou aising he compu a ional
cos . Fu he mo e, as in he laye s, he slabs do no ha e a meshing e o and i is possible o
co ec ly ep esen he en i e analysed su ace.