I m pl e m e n t a ti o n a n d
c o m p u t a ti o n al a s p e c t s of a 3 D
e l a s ti c w a v e m o d e lli n g b y
P U F E M
M a h m o o d , M S, L a g h r o u c h e , O, T r e v ely a n , J a n d Ka ci mi, AEl
h t t p :// dx. d oi.o r g / 1 0 . 1 0 1 6 /j. a p m . 2 0 1 7 . 0 5 . 0 1 3
T i t l e
I m pl e m e n t a ti o n a n d c o m p u t a ti o n al a s p e c t s of a 3 D e l a s ti c
w a v e m o d elli n g b y P U F E M
A u t h o r s
M a h m o o d , M S, L a g h r o u c h e , O, T r e v ely a n , J a n d Ka ci mi,
AEl
Typ e
Ar ticl e
U RL
T hi s v e r si o n is a v ail a bl e a t :
h t t p :// u sir. s alfo r d . a c . u k /i d/ e p ri n t/ 4 2 3 5 2 /
P u b l i s h e d D a t e
2 0 1 7
U S IR is a d i gi t al c oll e c ti o n of t h e r e s e a r c h o u t p u t of t h e U n iv e r si ty of S alfo r d .
W h e r e c o p y ri g h t p e r m i t s , f ull t e x t m a t e r i al h el d i n t h e r e p o si t o r y is m a d e
f r e ely a v ail a bl e o nli n e a n d c a n b e r e a d , d o w nl o a d e d a n d c o pi e d fo r n o
n-c o m m e r n-ci al p r iv a t e s t u d y o r r e s e a r n-c h p u r p o s e s . Pl e a s e n-c h e n-c k t h e m a n u s n-c ri p t
fo r a n y f u r t h e r c o p y ri g h t r e s t r i c ti o n s .
ContentslistsavailableatScienceDirect
Applied
Mathematical
Modelling
journalhomepage:www.elsevier.com/locate/apm
Implementation
and
computational
aspects
of
a
3D
elastic
wave
modelling
by
PUFEM
M.S.
Mahmood
c,a,
O.
Laghrouche
a,∗,
J.
Trevelyan
b,
A.
El
Kacimi
aaInstituteforInfrastructureandEnvironment,Heriot-WattUniversity,EdinburghEH144AS,UK bSchoolofEngineeringandComputingSciences,DurhamUniversity,DurhamDH13LE,UK cSchoolofComputing,ScienceandEngineering,SalfordUniversity,ManchesterM54WT,UK
a
r
t
i
c
l
e
i
n
f
o
Articlehistory:
Received30August2016 Revised7February2017 Accepted7May2017 Availableonline12May2017
Keywords:
PUFEM Planewavebasis Elasticwaves Pressurewaves Shearwaves 3Dscattering
a
b
s
t
r
a
c
t
Thispaperpresentsanenrichedfiniteelementmodel forthreedimensionalelasticwave problems, inthe frequency domain,capable ofcontainingmanywavelengths pernodal spacing.Thisis achievedbyapplyingtheplane wavebasisdecompositiontothe three-dimensional (3D)elasticwaveequationand expressingthedisplacementfieldas asum ofbothpressure(P)andshear(S)planewaves.Theimplementationofthismodelin3D presentsanumberofissuesincomparisontoits2Dcounterpart,especiallyregardinghow S-wavesareusedinthebasisateachnodeandhowtochoosethebalancebetweenPand S-wavesintheapproximationspace.Variousproposedtechniquesthatcouldbeusedfor theselectionofwavedirectionsin3Darealsosummarisedandused.Thedeveloped ele-mentsallowustorelaxthetraditionalrequirementwhichconsiststoconsidermanynodal pointsperwavelength,usedwithloworderpolynomialbasedfiniteelements,and there-foresolveelasticwaveproblemswithoutrefiningthemeshofthecomputationaldomain ateachfrequency.Theeffectivenessoftheproposedtechniqueisdeterminedby compar-ingsolutionsforselectedproblemswithavailableanalyticalmodelsortohighresolution numericalresultsusingconventionalfiniteelements,byconsideringtheeffectofthemesh sizeandthenumberofenriching3Dplanewaves.Bothbalancedandunbalancedchoices ofplanewavedirectionsinspaceonstructuredmeshgridsareinvestigatedforassessing theaccuracyandconditioningofthis3DPUFEMmodelforelasticwaves.
© 2017 Published by Elsevier Inc.
1. Introduction
Growingresearchactivitieshavebeentakingplaceinvariouswavenumericalmodellingfieldssuchasacoustics,surface waterwaves,radarwaves,seismology,geophysicsandbiomedicalultrasound.Therelatedproblemsinthefrequencydomain aremodelledusingmainlytheHelmholtzequation [1–3],Maxwellequation [4]andNavierequation [5–8]dependingonthe wavepropagationmediumandthetypeofapplication.
Anumberofdifferentnumericalmethodshavebeenusedtosolvesuch problemsandthemostcommonlyusedisthe finiteelement method(FEM)dueto itsflexibility inhandlingcomplexgeometriesanditsabilityto modeldifferenttypes ofmedia. However, theuse oflow orderpolynomial basedfiniteelementsforsolving wave problemsrequires finemesh
∗ Correspondingauthor.Tel:+44(0)1314513100.
E-mail addresses: [email protected] (M.S. Mahmood), [email protected](O. Laghrouche),[email protected] (J. Trevelyan),
[email protected](A.ElKacimi).
http://dx.doi.org/10.1016/j.apm.2017.05.013
gridswithmanynodalpointsperwavelength.Atmedium andhighfrequencies, thisleadstolarge memoryrequirements andhighCPUtime.Moreover,becauseoftheso-calledpollutionerror,evenfinermeshgridswouldberequiredtomaintain anacceptableaccuracy.
Thelasttwodecadessaw thedevelopmentofnumericaltechniquesaimingatmodellingwaveproblemswithimproved accuracyandreducedcomputationalcost,incomparisontotheconventionalFEM.Thesetechniquesinvolvethe incorpora-tionofa prioriknowledge ofthewavefield inthenumericalmodel.Indeed,thewave field isexpandedintosetsof ana-lyticalfunctions,forexample,intheformofplanewavespropagatingindifferentdirections.ForthecaseoftheHelmholtz equation, various methodshave beensuccessful inefficiently solving wave problems withcoarse mesh grids inboth 2D and3D withreducedcomputationaleffortandbetteraccuracy, incomparisontostandardpolynomial basedFEM.Among thesemethodsare the least-squares method [9],thepartition of unityfinite elementmethod (PUFEM) [10–16],the ultra weak variationalformulation (UWVF) [17,18],Plane wave discontinuousGalerkin (PWDG) [19], the generalizedfinite ele-mentmethod [20]andthediscontinuousenrichmentmethod(DEM) [21].Inthecaseofthe boundaryelementmethod,a partitionofunityboundaryelementmethod(PUBEM)hasbeendevelopedfortwoandthreedimensionalHelmholtz prob-lems [15,22]andanisogeometricwave-enrichedmethodwithin thecontextofboundaryelements(XIBEM)waspresented fortwoandthreedimensionalHelmholtzproblems [23,24].Othersolutionswerealsoconsideredsuchastheoscillatedfinite elementpolynomials [25]andinamorerecentworkthestablediscontinuousGalerkinmethod[26].Thevariationaltheory ofcomplexrays(VTCR) [27]andthephasereductionfiniteelementmethod(PR-FEM) [28]areotherdedicatedstrategiesfor solvingshortwaveproblemswithcoarsemeshgrids.
SomeofthesetechniqueshavebeenextendedtodealwithHelmholtzproblemsinnon-homogeneousmedia [1–3]and flowacoustics [29].Theywerealsoextendedtodealwithtwo-dimensionalelasticwaveproblems,forexample,PUBEM [30], UWVF [31],DEM [32]andPUFEM [5–7].Otherapplicationsincludefluid-solidinteraction [33],evanescentwaveproblemsat solid-fluidinterface [34],solid-solidinterface [35],acousticfieldsincavitieswithabsorbingmaterials [36]andsimulationof Biotswavesinporoelasticmaterials [37].Thesetechniqueshaverecentlybeenreviewedin [38]fortheHelmholtzequation andfurtherreferencesareindicated.
Inthecurrentwork,thePUFEMconceptisextendedtothree-dimensionaltime-harmonicelasticwaveproblems.Tothe authors’knowledge, only the works [39,40]have dealt withthree-dimensional plane wave enriched elementsfor elastic waveproblemswithinDEMandUWVFframeworks,respectively.InbothDEM [39]andUWVF [40]theapproximatingplane wavesare definedattheelementlevelandhenceinter-elementcontinuitymustbeenforcedintheformulation.However, inthepresentedPUFEMelasticmodel,theplanewavesdefinedatthenodesoftheelementsaremultipliedbytheLagrange polynomialshapefunctionsandhencecontinuityisautomaticallyensured.
Inthismodel,thecomponentsofthedisplacementfieldare firstexpressedassumsofbothpressure(P)andshear(S) wavecontributions,like inthetwo-dimensionalcase [5].The maindifferencewiththethree-dimensionalcaseconsistsin thedirectionofS-waveinduceddisplacement,whichisorthogonaltothedirectionofpropagation.Whileintwodimensions theorthogonaldirectioniswelldefinedviaasinglevector,inthree-dimensionstheorthogonaldirectioniscontainedwithin theorthogonalplanetothepropagationdirectionandthereforetwovectorsarerequiredforitsresolution.Furthermore,in twodimensions, thedirectionsofthe approximatingplanewavesare easily definedwhetherina uniformornonuniform distribution.This is not the casein three dimensionswhere the distributionof plane waves in space, either uniformor nonuniform,isnotstraightforward.
Indeed,intheimplementationofthethree-dimensionalversionofPUFEMtheuniformdistributionofdirectionsinspace isachallengingissue,unlikeinthetwo-dimensionalcase.Infact,theuniformdistributionofasetofpointsonaspherehas attractedtheattentionofresearchersfromvariousfieldsincludingmeteorologyandquantumtheory.Inthispaper,different approachesforuniformdistributionofdirectionsinspacearesummarised andthensome ofthemare usedtoobtain sets of linearlyindependent directions. Moreover, to keep the conditioning within acceptable limits, unbalanced numbers of directionsforthepressureandshearwavesareused.
Theoutlineofthepaperisasfollows.In Section2,themodelproblemandtheusednotationsarepresented. Section3is dedicated to thevariational formulation and Section 4presents the PUFEM approximated solution. In Section 5, various approaches to uniformly distribute sets of plane waves in space are presented. The PUFEM discrete model is given in
Sections6and 7 presentsavalidation ofthe modelagainst problemswithanalytical solutionssuch asprogressiveplane wavesinanelasticinfinitemediumandelasticwavescatteringbyasphericalrigidbody.Forthelatterproblem,a compari-sonbetweenPUFEMandthecommonlyusedloworderFEMisalsoattempted.Finallysomeconcludingremarksaredrawn in Section6.
2. Problem statement
Let
be the spatialdomain inR3 occupiedby an elastic medium. Letusdenote by (e
1, e 2, e 3) thecartesian vector
system,and x =x1e 1+x2e 2+x3e 3 a genericpoint inR3.Thepropagation ofelastic wavesina3D homogeneousisotropic
mediumisgovernedbythewaveequation [41]
ρ∂
tt U−∇
·σ
(
U)
=ρ
F, (1)classicalHooke’slaw
σ
(
U)
=λ
∇
·UI+μ
∇
U+(
∇
U)
, (2)where I isthe identitymatrixinR3×3 (3× 3matrix),
λ
andμ
aretheLamé parametersoftheelastic material,assumedconstant,and
∇
U =(
∇
U1,∇
U2,∇
U3)
.Wewilldenotehereby‘’thetransposeofagivenvectorortensor.Thedotproduct‘·’of
∇
withatensorfield A =(
A 1,A 2,A 3)
inR3×3 isdefinedby∇
·A=∇
·A 1∇
·A2∇
·A3. (3)
However,foravectorfieldinR3,‘·’istheusualdotproduct.
Thecaseofharmonicmotionhaspracticalinterest.Supposethat F =exp
(
iω
t)
f ,whereω
isthecircularfrequencyand i=√−1istheimaginaryunitnumber.If U =exp(
iω
t)
u isasolutionof (1)thenthestrongformoftheprobleminterms of u canberewrittenasfollows−
ρω
2u−∇
·σ
(
u)
=ρ
f, in, (4)
withthefollowingboundarycondition
σ
(
u)
n=i((
λ
+2μ
)
kP(
u·n)
n+μ
kS(
u·t)
t)
+g, on, (5)
where n and t denote,respectively,theoutwardunitnormalandtangentvectorstotheboundary
(
=
∂
)and g isthe sourceterm. Detailsonhowthisboundaryconditionisdeduced aregiveninreference [31].Notethat (5)isan absorbing boundaryconditionofthefirstorderforelasticwaveequationswhen g =0on
.Theaimofthisworkconsiststovalidate thepresentednumericalmodel andthereforethe sourceterm g isintroducedanalytically inexpression (5)toenablethe solutionoftheproblem.
Itiswellknownfromtheelasticwavetheory [41]thatadisplacementfieldsolutionof (4)isasuperpositionoftwotypes ofpurelyoscillatorywaves:P-wave(primary,orcompressional)andS-wave(secondary,orshear).Thiscanbedemonstrated bywriting
∇
·σ
(
u)
=(
λ
+2μ
)
∇
(
∇
·u)
−μ
∇
×∇
×u, (6)ThePandSwavestravelwithspeeds
cP =
λ
+2μ
ρ
and cS =μ
ρ
. (7)ThecorrespondingcompressionalwavenumberkP andshearwavenumberkS aregivenby
kP =c
ω
P and kS =
ω
cS . (8)
Theselasttwo physicalquantitiesareusefulforthenumericalmodellingapproachthatwewilldiscussandinvestigatein thispaper.
3. Variational formulation
SincePUFEMisafiniteelementtypemethod,wewillneedtoestablishavariationalformulationforthetime-harmonic problem (4)and (5).WefirstintroducethefollowingSobolevspaces
V=
{
v∈H1()
3;v=0on}
. (9)Unless otherwise specified, standard notations of Sobolev spaces are adopted. Then the weak variational formulation of PUFEMcanbedescribedas:Find u ∈Vsuchthat
a
(
u,v¯)
=b(
v¯)
,∀
v¯∈V, (10)wherev ¯isthecomplexconjugateofatestfunction v in V and a, b arebilinearandlinearformsdefinedrespectivelyas:
a
(
u,v¯)
=−ω
2ρ
u·v¯d
+
σ
(
u)
:∇
v¯d−
σ
(
u)
n·v¯d,
b
(
v¯)
=ρ
f·v¯d
. (11)
Theproductoftwotensorfields A =
(
A 1,A 2,A 3)
and B =(
B 1,B 2,B 3)
inR3×3isgivenbyThecurloperator
∇
×isdefinedforavectorfield v =(
v
1,v
2,v
3)
by∇
×v=∂
2
v
3−∂
3v
2∂
3v
1−∂
1v
3∂
1v
2−∂
2v
1. (13)
Underthepreviousnotationsanddefinitions,thetensor
σ
canbewrittenasfollowsσ
(
u)
=λ
∇
·uI+μ
[∇
×u]×+2μ
∇
u, (14)wherethematrix[
∇
×u ]× isdefinedby[
∇
×u]×= 0∂
1u2−
∂
2u1∂
1u3−∂
3u1∂
2u1−∂
1u2 0∂
2u3−∂
3u2∂
3u1−∂
1u3∂
3u2−∂
2u3 0, (15)
with u =
(
u1,u2,u3)
.Letusnoticethattheproductσ
(
u)
:∇
v ¯ in (11)canbe written,using (14),undera forminvolvingtheoperators
∇
,∇
·and∇
×:σ
(
u)
:∇
v¯=λ
(
∇
·u)(
∇
·v¯)
+2μ
∇
u:∇
v¯−μ
(
∇
×u)
·(
∇
×v¯)
. (16)Thenthankstotheboundarycondition (5),werewrite a and b ofexpression (11)inthefollowingformulations:
a
(
u,¯v)
=−ω
2ρ
u·v¯d
+2
μ
∇
u:∇
v¯d−
μ
(
∇
×u)
·(
∇
×v¯)
d−
λ
(
∇
·u)(
∇
·v¯)
d−i
((
λ
+2μ
)
kP(
u·n)(
v¯·n)
+μ
kS(
u·t)(
v¯·t))
d,
b
(
v¯)
=ρ
f·v¯d
+
g·v¯d
. (17)
From the mathematical point of view, according to the boundary condition (5) on
, the problem in (10) and (17) is well-posed and an existence and uniqueness resultcan be established by virtue of Fredholm’s alternative theorem and continuation arguments [42,43], under weak assumptions, i.e., let us consider that
is a bounded Lipshitz domain,
g ∈H−12
()
×H−12()
,andthat f ∈ V ∗,where V ∗denotesthetopologicaldualspaceof V .4. PUFEM approximated solution
We now need to approximate the problem stated in expression (10) by the PUFEM, based on plane wave basis and polynomialshapefunctions.Forthispurpose,letusconsiderafiniteelementmeshcontainingnnodes,denotedz, z=1,n. Wedenoteby{Nz }thepartitionofunityby polynomialfiniteelementshapefunctions, andrespectivelybymP andmS the numbersofapproximatingPandSplanewaves.Followingthetwo-dimensionalPUFEMapproximation [5],thedisplacement field u isapproximatedbysuperimposingpressureandsheardisplacementsasfollows
uh =uP h +uS h . (18)
However,inthreedimensionstheshearwaveinduceddisplacementisresolvedintotwocomponentsasfollows
uS h =uh S,1+uSh ,2. (19)
Therefore,using the plane wave enrichment approach,the approximated displacement field u h is then expressed in the followingform
uh = z=1,n
l=1,m P
Nz AP z,l exp
(
ikP x·dl P)
dl P+
z=1,n
l=1,m S
Nz ASz,,1l exp
(
ikS x·dl S)
dlS,,1⊥+
z=1,n
l=1,m S
Nz ASz,,2l exp
(
ikS x·dl S)
dlS,,2⊥. (20)TheorthogonalunitS-vector d S ,⊥,totheunit P-vector d ,isresolvedintotwoorthogonalvectorsdenotedby d lS,,1⊥ and d lS,,2⊥ andaredefinedby
dlS,,1⊥=d×P and d l,2
S,⊥=d
l,1
S,⊥×d×P , (21)
suchthat
dl S 22=dl,1
S,⊥
22+dl,2
S,⊥
22, (22)andthen
dl,1
S,⊥=
d×P
dl S 2and dl,2
S,⊥=
dlS,,1⊥×d×P
dlS
2Table1
Examplesofsetsofdirections(points)relativelyuniform(equallyspaced)onunitspherefordifferent ap-proaches.
Method Numberofdirections
Regularicosahedron[47] 12 42 162 642 Kuriharadistribution[44] 6 18 38 66 102 Cubeboundarymeshing[15] 8 26 58 98 154
CorrectedBeverlydistribution[48] 6 12 22 34 46 58 84 106 128 156
with
·2 denoting the L2-norm. Note that, in3D, there is an infinite numberof vectorsthat are orthogonal to a givenvector.Asaresult,findingtheperpendicularvector d ×P to d l
P canbechosenindifferentways,oneofthembeing
d×P =
0 −1 11 0 −1
−1 1 0
dl P , (24)
where d l P =
(
sinθ
l cosφ
l ,sinθ
l sinφ
l ,cosθ
l)
,withθ
l andφ
l beingthetwoanglesofthesphericalcoordinatesystemwith0≤
θ
l ≤π
and0≤φ
l ≤2π
.Inthecaseof d lP withthesamecomponentssuchas d lP =(
a,a,a)
thentheperpendicularvectorisgivenby d ×P =
(
−a/2,−a/2,a)
.Noticethat
∇
×u p h =0and∇
·u Sh ,1=∇
·u Sh ,2=0.Inexpression (20),hdescribesthecomputationalmeshsize.For sim-plicityofthenotations,thedependencyofthenumericalsolution u h onthenumbersofapproximatingPandSplanewaves,mP andmS ,isnotindicated.
The above approximation (20) can be derived fromthe Helmholtz decomposition theorem, as discussed in [30]. The unknownsarenolongerthenodalvaluesofthedisplacement u h ,butarenowtheamplitudesAzP ,l,ASz,,l 1andASz,,2l attachedto anodezandcorrespondingtoPandSplanewavestravellinginthedirections d l Pand d l S,respectively.
5. Choice of plane wave directions in space
Intwodimensions,thechoiceoftheplanewavesandtheiruniformspacingorclusteringarounddirectionsofpreference isatrivialproblem.Themovetothethreedimensionalcasepresentshoweversomedifficulties.Infact,inthreedimensions, it isneitherintuitive noran easy tasktodefine m uniformdivisionsofthe4
π
solid angle,mbeinganyintegernumber, a part froma few known solutions. In previous work [15], the authors used a uniformboundary meshing ofa cube. A reasonablywell spaced setofpointsis definedbythe vectorsjoiningthecentre ofthecube toeach node on thecube’s boundary.However,thisapproachisalsolimitedtoafewspecialcasesofmforwhichaboundary-meshedcubeisavailable. Specifically,thealgorithmislimitedtovaluesofm=p3−(
p−2)
3 forallp>1,thusgivingm=8,26,56,98,...etc.Theuniformdistributionofa setofpointson asphereattractedtheattentionofresearchersfromvariousfields.Early workinmeteorologicalmodelling,forexamplebyKurihara [44],resultedinalgorithmsthatproducedistributionsbasedon the mappingof a regular grid onan equilateraltriangleto each octant ofthe sphere.However, this islimited to values of m=4p2+4p−2,where p+1 isthe number ofgrid pointsfromthe north poleto the equator.The icosahedral grid
ofWilliamson [45] usestheverticesofthe regularicosahedron withthe possibilityofsubdividingeach triangularfacein a similar manner to the Kurihara grid [46] or in a symmetricalmanner such as performed in reference [47]. All these possibilitiesproduce distributions withlimitations onthe values ofm. Forexample,in reference [47], itis givenasm=
10p2+2 where p=2q withq ≥0 beingthe number ofequal intervalsinto which each side of theoriginal icosahedral
trianglesisdivided.
With the motivation to define sets of plane waves forming bases forthe DEM, Grosu andHarari [48] corrected the techniqueofBeverly [49]findingequallyspacedpointsonlatitudinallinesofaspherewhicharethemselvesequallyspaced alongthelongitudinallines.Thistechniqueprovidessetsofapproximatelyuniformdistributionsofpointsonaspherewith morepossibilities,incomparisontothepreviouslycitedapproaches(see Table1).
Inthe caseoftheUWVF [40],a librarywasusedinwhich theselection ofm directionsinspace, with4≤ m≤ 130, wasbasedonminimising thedistancesbetweenpointsona surfaceofasphere.The recursivezonalequal-area partition proposedbyLeopardi [50]offersthepossibilitytogenerateanynumbermofpointsonthespherethroughthepartitioning intoregionsofequalareaandsmallequivalentdiameter.Inthisapproach,itisrequiredtofindthecolatitudesofpolarcaps (zonesbetweenadjacentlatitudinallines), thendeterminethe collaranglewhichprovidesthenumberofregions ineach collar.
More recently,Peake etal. [51] presenteda methodof uniformlyspacinga set ofpointson a spherebased on static equilibriumof chargedparticles and hencenamed it theCoulomb force method.This work ismotivated by theneed to use uniformwave directionsin spacewithin PUBEM. It isshownthat the performance ofthe enriched elements, within PUBEM,isnotsensitivetosmallvariationsintheuniformityoftheplanewavedirectiondistribution.Theinterestedreaders inmethodsdealingwithuniformdistributionofpointsonaspherearereferredtothereferencesincludedin [51].
−1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1
Fig.1. Distributionexamplesof58pointsonthesurfaceofasphere:Cube-boundarymeshing(CM)[15](left),CorrectedBeverly(CB)[48](middle)and Leopardimethods[50](right).
consiststouse theCube-boundaryMeshing (CM) which waspreviouslyused inPUFEM for3D Helmholtz problems [15]. Thisallows anumberofdirectionsm=p3−
(
p−2)
3 forallintegerp>1,whichgivesm=8,26,56,98,...etc.Forevenpnumbers,twofurtherdirectionsareaddedpointingtothenorthandsouthpolesleadingtom=10,26,58,98,...etc. The secondapproachistheCorrectedBeverly(CB)method [48]whichconsiststofindequallyspacedpointsonlatitudinallines ofa spherewhich themselvesare equally spacedalong the longitudinallines. Theapproach sets thechord betweentwo adjacentpointsonthenthlatitudinallineequaltothechordbetweenthenthand
(
n+1)
thlatitudinallines.ThenumberNn ofpointsatthenthlatitudinallineisthengivenby
Nn =
π
/sin−1 sinsin(
(
φ
n0φ
/2)
0
)
, (25)
where
φ
0isaconstantanglebetweentwoconsecutivelatitudinallines.Fig.1 showsdistribution examples of 58 pointsobtainedby the cube-boundary meshing (left), the corrected Beverly method(middle)andtheLeopardimethod(right).Itisobviousthatthecube-boundarymeshingdistributionprovidespoints relativelymorespacedaroundtheequator incomparisontotheregions aroundthesouthandnorthpoles,whilethetwo othermethods providerelativelyuniformdistributions. Moreover,thecorrected Beverlymethodandthe Leopardimethod leadtoverysimilardistributionsform=6,12,22,34...etc.Ifthesouthcap,orthenorthcap,oftheLeopardimethodis rotatedbyacertainangle,bothmethodsleadtothesamedistribution.
6. PUFEM discretization
Inthiswork,thedisplacementfieldisapproximatedviapropagatingplanewavesinspaceinwhichbothbalanced(mP = mS )andunbalanced(mP =mS )choicesofplanewavesareconsideredinexpression (20)oftheapproximateddisplacement field.Letusset
pl :=exp
(
ikPx·dl P)
dPl s1l :=exp(
ikSx·dl S)
dlS,,1⊥, s2l :=exp(
ikSx·dl S)
dlS,,2⊥. (26)Afiniteelementapproximationofthevariationalproblem (17)isderivedbyusingNz p ¯l ,Nz s ¯1
l andNz s ¯2l : −
ω
2ρ
uh ·
(
Nz pl)
d+
λ
(
∇
·uh)(
∇
·(
Nz pl))
d+2
μ
∇
uh :∇
(
Nz p¯l)
d−
μ
(
∇
×uh)
·(
∇
×(
Nz pl))
d−i
r
((
λ
+2μ
)
kP(
uh ·n)(
Nz pl ·n)
+μ
kS(
uh ·t)(
Nz pl ·t))
d=
ρ
f·
(
Nz pl)
d+
g·
(
Nz pl)
d, l=1,mP , (27)
−
ω
2ρ
uh ·
(
Nz s1
l
)
d+
λ
(
∇
·uh)(
∇
·(
Nz s1
l
))
d+2
μ
∇
uh :∇
(
Nz s¯1
l
)
d−
μ
(
∇
×uh)
·(
∇
×(
Nz s1
l
))
d−i
r
((
λ
+2μ
)
kP(
uh ·n)(
Nz s1l ·n)
+μ
kS(
uh ·t)(
Nz s1l ·t))
d=
ρ
f·
(
Nz s1
l
)
d+
g·
(
Nz s1
−
ω
2ρ
uh ·
(
Nz s2
l
)
d+
λ
(
∇
·uh)(
∇
·(
Nz s2
l
))
d+2
μ
∇
uh :∇
(
Nz s¯2
l
)
d−
μ
(
∇
×uh)
·(
∇
×(
Nz s2
l
))
d−i
r
((
λ
+2μ
)
kP(
uh ·n)(
Nz s2l ·n)
+μ
kS(
uh ·t)(
Nz s2l ·t))
d=
ρ
f·
(
Nz s2
l
)
d+
g·
(
Nz s2
l
)
d, l=1,mS . (29)
The mainfeatureofthe discretescheme (27),(28)and (29),thanksto thestress tensorform (14)andthenumerical ap-proximationofthedisplacementfield (20),isthattheresultingmatricescanbeeasilycomputedasinthecaseofthescalar Helmholtzequation [12],intermsoftheoscillatoryshapefunctions
NP z,l :=exp
(
ikP x·dl P)
Nz and NS z,l :=exp(
ikS x·dl S)
Nz . (30)Indeed,asNz p l =NzP ,l d l P ,Nz s 1
l =NzS ,ld l,1
S,⊥andNz s 2l =NzS ,ld l,2
S,⊥,themassmatrixcoefficientsaresuchthat
(
Nz pl)
·(
Nz pl)
d=d l P ·dl P
N P z ,l N¯P z,l d
,
(
Nz s1
l
)
·(
Nz pl)
d=dl S,,1⊥·dl P
N S z ,l N¯zP ,l d
,
(
Nz s2
l
)
·(
Nz pl)
d=dl S,,2⊥·dl P
N S z ,l N¯zP ,l d
,
(
Nz s1
l
)
·(
Nz s1l)
d=dSl ,,1⊥·dlS,,1⊥
N S z ,l N¯zS ,l d
,
(
Nz s2
l
)
·(
Nz s2l)
d=dSl ,,2⊥·dlS,,2⊥
N S z ,l N¯zS ,l d
,
(
Nz s1
l
)
·(
Nz s2l)
d=dSl ,,1⊥·dlS,,2⊥
N S z ,l N¯zS ,l d
,
(
Nz s2
l
)
·(
Nz s1l)
d=dSl ,,2⊥·dlS,,1⊥
N S
z ,l N¯zS,l d
. (31)
Theothermatricesinvolvedin (27),(28)and (29)canbewritten,inasimilarwayasforthemassmatrix,intermsofthe productofthenewoscillatoryshapefunctions.Thenumericalevaluationoftheaboveintegralsisperformedwithhigh-order Gauss–Legendrescheme.
Anodalpointz,withz=1,n,hasmP +2mS degreesoffreedom.Sothetotal numberofdegreesoffreedomisn
(
mP +2mS
)
, andtheunknown amplitudesAP z,l, ASz,,1l andAzS,,2l forz=1,n canbe storedinn blocksofvectorsin Rm P+2m S ofthe form(
A P z,A S z 1,A z S 2)
.TheglobalunknownamplitudevectorcanbestoredasfollowsA=
(
AP 1,AS 11
,AS 2
1
,...,AP n ,AS 1
n ,AS n 2
)
, (32)andtheglobalmatrixresultingfrom (27),(28)and (29)hasthefollowingform
W=
⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎝
WP1,,P1 WP,S 1
1,1 W
P,S 2
1,1 . . . W
P,P
1,n W1P,,S n 1 WP1,,S n 2
WS 1,P
1,1 W
S 1,S 1
1,1 W
S 1,S 2
1,1 . . . W
S 1,P
1,n W1S 1,,n S 1 WS 11,n ,S 2
WS 2,P
1,1 W
S 2,S 1
1,1 W
S 2,S 2
1,1 . . . W
S 2,P
1,n W1S 2,,n S 1 WS 12,n ,S 2
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
WP,P n,1 W
P,S 1
n,1 W
P,S 2
n,1 . . . W
P,P
n,n WnP,,S n 1 WPn,,S n 2
WS 1,P
n,1 W
S 1,S 1
n,1 W
S 1,S 2
n,1 . . . W
S 1,P
n,n WnS 1,,n S 1 WS n1,n ,S 2
WS 2,P
n,1 WS n2,1,S 1 WnS 2,,1S 2 . . . WnS 2,n ,P WnS 2,,n S 1 WS n2,n ,S 2
⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎠
, (33)wheretheblocks W Pz,,z P , W S z1,z ,S 1, W S z,2z ,S 2 aresquarematricesinRm P×m P,Rm S×m S andRm S×m S,respectively,withz,z=1,n.The
blocks W P,S1
z,z , W Pz,,z S2, W Sz,1z ,Pand W Sz2,z ,Pareeventually,ifmP =mS ,rectangularmatricesinRm P×m S orRm S×m P.Last,theblocks W S 1,S 2
0 1
2 3
4
0 1 2 3 4 0 1 2 3 4
0 1
2 3
4
0 1 2 3 4 0 1 2 3 4
0 1
2 3
4
0 1 2 3 4 0 1 2 3 4
Fig.2. Consideredmeshgrids:(left)h1,(middle)h1 2 and(right)h1 4 .
Table2
Valuesofh/λS for variousmeshsizes andcircularfrequencies.
ω 1 10 15 20
h1 0.32 3.18 4.77 6.37 h1 2 0.16 1.59 2.39 3.18
h1
4 0.08 0.79 1.19 1.59
Galerkinweighting is used inthis work andhence theglobal matrixof the resultingsystem issymmetric andblock banded. It is stored using a steering vector to locate the elements andthe solution of the final system is obtained via a directsolver [52] based on LDLT decomposition where LT is the transposeofthe lower triangular matrixL andD is a diagonalmatrix.
7. Numerical analysis
Inthis section,the developed numericalmodel forselected elasticwave problems is investigatedwithrespect to the computationalmeshsizehandtheplanewaveenrichment,intermsofthenumbersmP andmS oftheapproximatingPand Splanewaves.Intheconvergenceanalysisconsideredbelow,specialcareistakensothattheincidentwavedirection d inc P
isnottooclosetotheplanewavedirectionsusedintheapproximatingmodel.Thecomputationaldomainisdefinedby
=
{
x=(
x1,x2,x3)
;1≤x1,x2,x3≤3}
. (34)Allparameters a,
λ
,μ
andρ
aretakenequalto1withtheirrespectivecorrespondingunits.Thefiniteelementsusedhere are 8-node hexahedral elements, for which thegeometry is interpolatedvia linearLagrange polynomials. Three uniform mesh grids are considered. They are denoted by h1, h12
and h1 4,
see Fig. 2. The coarser mesh h1 has 8 elements. It is
hierarchicallyrefinedto obtainh1
2 andh14 mesh gridswith64 and512elements, respectively.The EuclideannorminR
3
oftherealpartofthecomputeddisplacementfield u h isdenotedby|Re(u h )|andtheaccuracyofthenumericalsolutionis measuredbythefollowingL2 error
ε
2=||
uh −u
||
L 2()||
u||
L 2()×100% (35)
where u representstheanalyticalsolutionoftheconsideredproblem.
Toquantifythenumberofdegreesoffreedom(DOF) perPandS wavelengths,respectively,the parameters
τ
P andτ
S areintroducedinthefollowingforms [5]τ
P = kπ
P(
nmP)
1/3 and
τ
S =k
π
S(
n mS)
1/3, (36)
wheren,asmentionedbefore,isthenumberofnodalpointsinthemeshgrid.
Letusdenote by
α
the average rateofconvergence withrespect to hsuch that the L2 errorε
2=O(
hα)
,whereα
istobeevaluatednumericallyvia
α
=(
logε
2(1/4)−logε
2(1/2))
/(
logh1/4−logh1/2)
,whenconsideringmeshgridsh1/4andh1/2,forexample. Table2showsthenumberofshearwavelengths
λ
S perelementsizehoftheconsideredmeshgrids(Fig.2), forcircularfrequenciesω
=1,10,15and20.Thecomparisonofthemeshsizewiththeshearwavelengthrevealsthatfor lowcircularfrequencyω
,eachfiniteelement containsafractionofλ
S whereas athigherω
,each finiteelementbecomes multi-wavelengthsized.Forexample,forthemeshgridh12,
Table3
Progressiveplanewaveproblem:L2errorin%forω=10.
(mp,ms) (10,10) (26,26) (58,58) (98,98)
h1 83.6 14.4 0.964 0.0377 h1
2 3.93 0.50 0.0032 0.0019
h1
4 0.237 0.0123 0.00002 0.00002
α 4.2 5.1 7.8 5.4
Table4
Progressiveplanewaveproblem:L2errorin%forω=15.
(mp,ms) (10,10) (26,26) (58,58) (98,98)
h1 109.4 105.5 2.31 0.213 h1
2 44.4 1.13 0.097 0.019
h1
4 1.59 0.205 0.00066 0.00017
α 3.1 4.5 5.9 5.1
Table5
Progressiveplanewaveproblem:L2errorin%forω=20.
(mp,ms) (10,10) (26,26) (58,58) (98,98)
h1 113.6 103.4 76.3 40.2 h1
2 102.5 26.5 0.973 0.310
h1
4 3.91 0.711 0.0076 0.00089
α 2.4 3.6 6.7 7.7
theempiricalexpressiongivingngauss =10×
h/λ
S +2,toensureenoughintegrationpointsareused.Itispossibleto de-velopsemi-analyticalintegrationschemessuchasdone in [6,53]toavoidtheburdenofthehighorderscheme,butthisis nottheobjectiveofthecurrentwork.AllcomputationsarecarriedoutinFortranwithdouble-precisioncomplexnumbers onWorkstationIntelCorei73.5GHzwith32GBRAM.7.1. Progressiveplanewavetestproblem
Letusfirstconsiderasimplemathematicalmodelconsistingofaprogressiveplanewaveinathreedimensionalinfinite elasticmedium.Thisproblemhasananalyticalsolutionandtheexpressionofthedisplacementfieldisgivenby
u=exp
(
ikPx·dinc P)
dinc P +exp(
ikS x·dS)
d1S,⊥+exp(
ikS x·dS)
d2S,⊥, (37) where d inc P ischosensuchthat(
dinc)
=(
sinβ
1cosβ
2,sinβ
1sinβ
2cosβ
1)
,β
1=33.33o ,β
2=333.33o , (38)andthetwocomponents d 1
S,⊥ and d 2S,⊥ofvector d S orthogonaltothedirection d inc P aregivenby
d1
S,⊥=
(
dinc P)
× and d2S,⊥=d1S,⊥×dinc P , (39)where
(
d inc P)
×isdefinedinexpression (24).Thesourceterm g ofexpression (5)iscalculatedfromthetraction
σ
(u )n on,evaluatedwiththeanalyticalsolution u
giveninexpression (37).Notethatthisisonlypossiblewhenexactboundaryconditionsareprescribedon
.
Tables 3–5 show theL2 errorin % aswell asthe average rateof convergence
α
for the threeconsidered mesh gridsh1,h1
2 andh14 whenthenumbers (mP ,mS )oftheapproximating planewavesare increasedforthe valuesofthecircular
frequency
ω
=10,15and30,respectively.Notethatnoneofthesetsofapproximatingplanewavescontainstheconsidered directionoftheimposedprogressiveplanewaveofexpression (38).Fromthenumericalresultsitisobviousthath-refinementimprovestheaccuracyoftheschemeforallcasesof frequen-ciesandplane waveenrichments. Increasingthenumbers ofapproximatingplanewaves,forafixed h,doesalsoimprove theaccuracyoftheplanewaveenrichmentmodel.Infact,thisisexpectedasboththemeshrefinementandtheincreasein thenumberofapproximatingplanewavesleadtohighernumbers
τ
P andτ
S ofDOFperPandSwavelengths,respectively. Nevertheless,tworemarksare worthmentioning.Thefirst oneconcernsthecasesofcoarsemeshgrids,incomparisonto the wavelength,withlownumbers ofapproximating plane waves.Insuch cases,theL2 errorishighandthereason liesin thefact that the discretizationlevel represented bythe numbers ofDOF per P- andS-wavelength,
τ
P andτ
S , isvery low.Forexample,forthemeshgridh12 at
ω
=15,theL2 errorisvery high,ε
2=44.4%,for(
mP ,mS)
=(
10,10)
forwhich(
τ
P ,τ
S)
=(
3.1,1.8)
.Forahigherenrichment,(
mP ,mS)
=(
26,26)
theL2 errorisa lotlowerincomparisontothepreviousto
(
τ
P ,τ
S)
=(
5.6,3.2)
,for(
mP ,mS)
=(
58,58)
,theL2errorisevenlower,ε
2=0.097%,andthenfor(
mP ,mS)
=(
98,98)
the L2 errorreducestoε
2=0.019%.Thesecond remarkisrelatedtotheaveragerateofconvergence
α
.Fromthe three Tables2–5,forfixed numbers(mP , mS )ofapproximatingplanewaves,theresultsshowanexponentialdecreaseoftheL2errorwithrespecttoh-refinement.Ingeneral,theaveragerate
α
increaseswiththeincreaseofthenumbersofapproximatingplanewaves,suchasshownforthe caseofω
=20.However,continuingtoincrease thenumberofapproximatingplane waves,fora fixedcircularfrequency, doesnotnecessarilyleadtoanincreaseintherateofconvergence.Thisisthecasefor(mP,mS)=(98,98),atω
=10and15wheretheratehasdecreasedfrom7.8to5.4andfrom5.9to5.1,respectively.Thisismostlikelyduetotheillconditioning issue,whichisaninherentfeatureoftheplanewavebasisfiniteelementapproaches.Thisaspectisinvestigatedinthenext testexampledealingwithelasticwavescatteringbyarigidsphere.
7.2.Wavescatteringtestproblem
Theproblemofharmonicwavesscatteredbya sphericalrigidbody containedinathreedimensionalhomogenousand isotropicinfinite elasticspace isconsidered. An upward P-wavewith amplitude
0,for whichthe thepotential isgiven
by
=
0exp
(
−ikP x3)
exp(
iω
t)
, (40)where
0= k ip,impingesonasphericalbodyofradiusa.Theincidentdisplacementfieldisobtainedfrom
uin=
∇
. (41)
Thetotaldisplacementfield u isasuperpositionoftheincident u in andscattered u sc fields, u =u in +u sc .Infact,aP-wave impingingonarigidbodywouldleadtobothscatteredPandSwaves,inaninfiniteelasticmedium.Thetotaldisplacement canbewrittenintermsofitscomponents
u=ur er +uθeθ+uφeφ, (42)
withrespecttothesphericalunitvectors e r , e θ and e φ with
ur =1r
ν=0,∞
0
ε
1p(0)(
ν
,r)
+Aνε
1p(1)(
ν
,r)
+Bνε
1s (1)(
ν
,r)
Pν
(
cosθ
)
, (43)uθ=1
r
ν=0,∞
0
ε
2p(0)(
ν
,r)
+Aνε
2p(1)(
ν
,r)
+Bνε
2s (1)(
ν
,r)
dPν(
cosθ
)
d
θ
, (44)uφ=0, (45)
wherePν areLegendrepolynomialsfor
ν
=0,∞.TheconstantsAν andBν arechosensuchthat u|
r=a =0representingthe totalreflectionconditionontheboundaryofthesphericalrigidbody.Thisisobtainedvia
0
ε
1p(0)(
ν
,a)
+Aνε
1p(1)(
ν
,a)
+Bνε
1s (1)(
ν
,a)
=0,0
ε
2p(0)(
ν
,a)
+Aνε
p(1)
2
(
ν
,a)
+Bνε
s ( 1)2
(
ν
,a)
=0, (46)forall
ν
inN.The associatedstress fieldσ
(u )is thenobtainedfromthedisplacement field (42)intermsofits spherical componentsσ
rr = 2rμ
2
ν=0,∞
0
ε
3p(0)(
ν
,r)
+Aνε
3p(1)(
ν
,r)
+Bνε
3s (1)(
ν
,r)
Pν
(
cosθ
)
, (47)σ
rθ = 2μ
r2
ν=0,∞
0
ε
4p(0)(
ν
,r)
+Aνε
p(1)
4
(
ν
,r)
+Bνε
s (1)
4
(
ν
,r)
dPν(
cosθ
)
d
θ
, (48)σ
θθ = rλ
2
ν=0,∞
0
ε
5p(0)(
ν
,r)
+Aνε
p(1)
51
(
ν
,r)
Pν
(
cosθ
)
+2μ
r2
ν=0,∞
0
ε
6p(0)(
ν
,r)
+Aνε
6p(1)(
ν
,r)
+Bνε
6s (1)(
ν
,r)
Pν
(
cosθ
)
+2μ
r2
ν=0,∞
0
ε
7p(0)(
ν
,r)
+Aνε
p(1)
7
(
ν
,r)
+Bνε
s (1)
7
(
ν
,r)
dPν(
cosθ
)
and
σ
φφ = rλ
2
ν=0,∞
0
ε
5p(0)(
ν
,r)
+Aνε
p(1)
5
(
ν
,r)
Pν
(
cosθ
)
+2μ
r2
ν=0,∞
0
ε
8p(0)(
ν
,r)
+Aνε
8p(1)(
ν
,r)
+Bνε
8s (1)(
ν
,r)
Pν
(
cosθ
)
+2μ
r2
ν=0,∞
0
ε
9p(0)(
ν
,r)
+Aνε
p(1)
9
(
ν
,r)
+Bνε
s (1)
9
(
ν
,r)
dPν(
cosθ
)
d
θ
. (50)The symbols
ε
ip(0),ε
p(1) i andε
s (1)
i ,fori=1,9, arecoefficientsobtainedformthe variousderivatives.The radialspherical
functionsaredefinedas,see [54]:
ε
p(i )1
(
ν
,r)
=A{
nξ
p(i )
ν
(
kP r)
−kP rξ
νp(i )(
kP r)
}
,ε
p(i )2
(
ν
,r)
=A{
ξ
p(i ) ν
(
kP r)
}
,ε
p(i )3
(
ν
,r)
=An2−n−1
2
(
kS r)
2
ξ
p(i )ν
(
kP r)
+2kP rξ
νp+(i 1)(
kP r)
,ε
p(i )4
(
ν
,r)
=A{
(
n−1)
2ξ
p(i )
ν
(
kP r)
−kP rξ
νp+(i 1)(
kP r)
}
,ε
p(i )5
(
ν
,r)
=A{
−k 2P
ξ
νp(i )(
kP r)
}
,ε
p(i )6
(
ν
,r)
=A{
−n 2ξ
p(i )ν
(
kP r)
−kP rξ
νp+(i 1)(
kP r)
}
,ε
p(i )7
(
ν
,r)
=A{
−cot(
θ
)
ξ
p(i ) ν
(
kP r)
}
,ε
p(i )8
(
ν
,r)
=A{
n 2ξ
p(i )ν
(
kPr)
−kPrξ
νp+(i 1)(
kPr)
}
,ε
p(i )9
(
ν
,r)
=A{
cot(
θ
)
ξ
p(i )
ν
(
kP r)
}
, (51)wherei=0ori=1suchthat
ξ
νp(0)(
kP r)
=Jν(
kP r)
isthesphericalBesselfunctionoffirstkindandordernandP0n
(
cosθ
)
isLegendre’spolynomialofordernanddegreem=0.A=
(
−i)
ν(
2n+1)
andξ
νp(1)(
kP r)
=Hν(
kP r)
,A=1ε
s (1)1
(
ν
,r)
=−n(
n+1)
Hν(
kS r)
,ε
s (1)2
(
ν
,r)
=−n(
n+1)
Hν(
kS r)
+kS rHν+1(
kS r)
,ε
s (1)3
(
ν
,r)
=−n(
n+1)
{
(
n−1)
Hν(
kS r)
−kS rHν+1(
kS r)
}
,ε
s (1)4
(
ν
,r)
=−n2−n−1
2
(
kS r)
2
Hν
(
kS r)
−kP rHν+1(
kS r)
,ε
s (1)5
(
ν
,r)
=−n(
n+1)
Hν(
kS r)
,ε
s (1)6
(
ν
,r)
=−n(
n+1)
{
(
n+1)
Hν(
kS r)
−kS rHν+1(
kS r)
}
,ε
s(1)7
(
ν
,r)
=−cot(
θ
)
{
(
n+1)
Hν(
kS r)
−kS rHν+1(
kS r)
}
,ε
s (1)8
(
ν
,r)
=−n(
n+1)
Hν(
kSr)
,ε
s (1)9
(
ν
,r)
=cot(
θ
)
{
(
n+1)
Hν(
kS r)
−kS rHν+1(
kS r)
}
. (52)Thestressfieldisthenusedintheevaluationofthesourceterm g ofexpression (5)tobeusedintheboundarycondition oftheelasticwavescatteringproblem.Theseriesusedinexpressions (43)and (44)aretruncatedwhenasufficientnumber
Nt ofterms isincluded.Thisnumber istakensuch that Nt ≥kSR whereR isthe radiusofthe mostdistant pointof the
considered computational domain [55].Forillustrationpurpose, Fig.3showscontour plotsofthe componentsof|Re(u h)|
in the computational domain forthe case of
ω
=40 when a P-wave travelling upward along the x3-axis impingeson asphericalrigidobjectofunitradius.ThecentreoftherigidscatterercoincideswiththecentreoftheCartesiansystemand hencetheconsideredcomputationaldomainisinthevicinityofthescatteringobject.Thescatteredfieldcontainsbothbody waves,PandS,whichinterferearoundthehardscatterer.Thefirstandsecondcomponentsareoppositeeachotherandthe thirdcomponenthashighervaluesduetotheincidentwavepropagatingalongx3-axis.Toobtainthecontourplotsof Fig.3,
thecomputednodalvaluesfromtheproblemsolutionareusedinconjunctionwiththemodelofexpression (20).
Thesamenumericalanalysisisfollowedforthiselasticwavescatteringproblemconsideringthesameparametersalready usedforthecaseoftheprogressiveplanewavetestproblembutwithmoreoptionsregardingtheusedsetsof approximat-ing plane waves.Indeed, inthiscase, plane wave distributions obtainedformthe Cube-boundaryMeshing (CM) andthe Corrected Beverly(CB)methodareused.Thetwo approaches,CMandCB,offermoreflexibilityinthechoiceofuniformly distributedplanewavesinspaceandevenwhentheyallowthesamenumberofplanewavestheCM andCBdistributions aredifferent. Tables6–8summarisetheerroranalysisresultsforthemeshgridsh1,h1
Fig.3. Elasticwavescatteringbyarigidsphere,contourplots(a),(b)and(c)offirst,secondandthirdcomponentsofRe(uh)respectively,ω=40.
approximatingplanewavesareincreased,for
ω
=10,15and20,respectively.Theyalsogivetheaveragerateofconvergence computednumericallyfromtheobtainedL2errors.From Tables6–8,itcanbeseenthat,ingeneral,theobtainedresultsleadtosimilarconclusionsalreadydrawnfromthe caseofthe progressiveplanewave test problem.Basically,the L2 errorimproves asthe numbers(mP ,mS ) ofthe
approx-imatingPandS plane wavesincrease forall casesofmesh grids andfrequencies presentedhere. However, we noticein certaincasesafterreachingalowL2 errorafurtherincreaseinthenumberofenrichingplane wavesdoesnotnecessarily
improvetheaccuracysuch asreflectedinthecaseofh1 4
and
ω
=10whenincreasing(mP ,mS )form(58,58)to(106,106). Inallthesecases,theL2errorremainedunchangedatε
2=0.00002%.Forfrequenciesω
=15andω
=20,thisisagainseenforthemeshgridh1
4 when(mP ,mS )increasedfrom(98,98)to(106,106)andforbothenrichmentstheL2errorstagnated
at
ε
2=0.00004%and0.00006%,respectively.Moreover,inthecaseoffrequencyω
=20andthemeshgridsh1 andh1 2,forhighnumbers ofenrichingplanewavesthequality oftheresultsdeterioratesinsteadofimproving,oratleaststagnating. Thisisseenwhen(mP ,mS )increasedfrom(98,98)to (106,106), forwhichtheL2 errorincreasedfrom4.00% to7.19%in
thecaseofmeshgridh1andincreasedslightlyfrom0.0022%to0.0024%forthemeshgridh1
2.Thisisknowntobecaused
bytheillconditioningaspect,whichisinvestigatedhere.