• Aucun résultat trouvé

hyperbolic conservation laws A new fifth order finite difference WENO scheme for solving Journal of Computational Physics

N/A
N/A
Protected

Academic year: 2022

Partager "hyperbolic conservation laws A new fifth order finite difference WENO scheme for solving Journal of Computational Physics"

Copied!
12
0
0

Texte intégral

(1)

Contents lists available atScienceDirect

Journal of Computational Physics

www.elsevier.com/locate/jcp

Short note

A new fifth order finite difference WENO scheme for solving hyperbolic conservation laws

Jun Zhu

a

, Jianxian Qiu

b,∗

aCollegeofScience,NanjingUniversityofAeronauticsandAstronautics,Nanjing,Jiangsu210016,PRChina

bSchoolofMathematicalSciencesandFujianProvincialKeyLaboratoryofMathematicalModelingandHigh-PerformanceScientific Computing,XiamenUniversity,Xiamen,Fujian361005,PRChina

a rt i c l e i n f o a b s t ra c t

Articlehistory:

Received16April2016

Receivedinrevisedform2May2016 Accepted3May2016

Availableonline7May2016

Keywords:

FifthorderWENOscheme Hyperbolicconservationlaws Finitedifferenceframework

Inthispaperanewsimplefifthorderweightedessentiallynon-oscillatory(WENO)scheme ispresentedinthefinitedifferenceframeworkforsolvingthehyperbolicconservationlaws.

ThenewWENOschemeisaconvexcombinationofafourthdegreepolynomialwithtwo linearpolynomialsinatraditionalWENOfashion.ThisnewfifthorderWENOschemeuses thesamefive-pointinformationastheclassicalfifthorderWENOscheme[14,20],couldget lessabsolutetruncationerrorsinL1 andL norms,andobtainthesameaccuracyorder in smooth region containing complicated numerical solution structures simultaneously escaping nonphysical oscillations adjacent strong shocks or contact discontinuities. The associatedlinearweightsareartificiallysettobeanyrandompositivenumberswiththe onlyrequirement that their sum equals one. New nonlinear weights are proposed for thepurpose ofsustainingthe optimalfifth orderaccuracy.The new WENOschemehas advantagesovertheclassicalWENOscheme[14,20] inits simplicityand easyextension tohigher dimensions. Some benchmark numerical tests are performed to illustrate the capabilityofthisnewfifthorderWENOscheme.

©2016ElsevierInc.Allrightsreserved.

1. Introduction

Recently, manyhighorderfinitedifferenceorfinitevolume numericalmethods havebeeninvestigatedtosolveforhy- perbolic conservation laws. The essentially non-oscillatory(ENO) and weighted ENO (WENO) schemes, which have been appliedquitesuccessfullytosolvetheproblemswithstrongshocks,contactdiscontinuitiesandsophisticatedsmoothstruc- tures,aretheprimary schemesusedasthecurrentstate oftheartintheliterature. Forthepurposeofachievinguniform highorderaccuracybothinspaceandtime,HartenandOsher[11]gaveaweakerversionofthetotalvariationdiminishing (TVD)criterion [8]andon whichthey established abasis forthe reconstructionofhighorder ENOtype schemes. Anda series ofENO schemes were developed by Hartenet al.[10] to solve one dimensional problems.The mostcrucial spirit oftheseENOtype schemesistoapply thesmoothestcandidatestencilamongallstencilsto approximatethe variablesat cellboundariesforthesakeofsustaininghighorderaccuracyinsmoothregionandnotintroducingspuriousoscillationsin nonsmoothregion[21,22].ThefirstWENOschemewasoriginally proposedbyLiu,OsherandChan[15],inwhichinstead ofusingthe optimalsmooth candidatestencil,alinearconvexcombinationofallstencilsincludingnonsmoothstencils is used.Afterthat,suchWENOschemewasimprovedbyJiangandShu[14],inwhichageneralframeworkforthedesigning

TheresearchispartlysupportedbyNSFCgrants11372005,11571290and91530107.

*

Correspondingauthor.

E-mailaddresses:[email protected](J. Zhu),[email protected](J. Qiu).

http://dx.doi.org/10.1016/j.jcp.2016.05.010 0021-9991/©2016ElsevierInc.Allrightsreserved.

(2)

ofnewsmoothnessindicatorsandnonlinearweights wasspecified indetail.Forthesystemcase, theWENOschemesare basedonlocalcharacteristicdecompositionandnumericalfluxsplittingmethodtoavoidspuriousoscillationsnearbystrong shocksorcontactdiscontinuities. Andsome other famousWENOschemes areproposed forspecificalcomputation needs, suchasoptimizedWENOschemesforsolvinglinearwaveswithdiscontinuity[23],monotonicitypreservingWENOschemes forveryhighorderaccuracysimulation[2],hybridcompactWENOschemesforcomputingshockandturbulenceinteraction [17,18],androbustWENOschemesfordealingwithdistortedlocalmeshgeometryorthedegenerationofthemeshquality variesforcomplexcomputationaldomain[12,16,19]etc. ThevarioustypesofENOandWENOschemes[4,5,9,13–15,21,22, 25] are quitesuccessful innumericalsimulations forsteady andunsteady problemsharboring strongdiscontinuities and sophisticatedsmoothstructures.

Comparingwith the above mentioned literature, especially including classical WENO schemes proposed by Jiang and Shu[14,20], themajor advantages of thenewfinite difference fifth orderWENOscheme inthispaper are its simplicity, highorderaccuracyandeasyimplementationinthecomputation.ThisnewWENOschemehasaconvexcombinationofa fourthdegreepolynomialwhich shouldsubtractother twolinearpolynomialsbymultiplyingproper constantparameters, thencompoundthesethreeunequaldegreepolynomialswhichuseinformationondifferentspatialstencilsinaWENOtype methodologyincludingartificiallysettinglinearweightsfortheconservationwithoutlosingnumericalaccuracy,computing smoothnessindicatorsandproposingnewnonlinearweightformulawhichisdifferenttotheexpressionspecifiedin[3,6], forsolvingtheEulerequationsinone andtwodimensions. Theessentialmeritsofsuchmethodologyareits simplicityin spaceby thedefinition ofpositive linearweights, andonlyone five-pointstencil andtwotwo-point stencils are usedto reconstructthreedifferentdegreepolynomials. Aninnovationofmodifying thefourthdegreepolynomialis veryessential forsustainingthescheme’shighorderaccuracy,otherwise,sometraditionalrobustWENOmethodologieswilldegradetheir numericalaccuracytoalowerorderofaccuracy.Therefore,wetrytousefifthorderapproximationtothenumericalfluxfor simplicityandeasyimplementationforthesakeofobtaininghighorderaccuracyinsmoothregion,andswitchittoeither oftwosecondorderapproximationsonsmallerspatial stencilswhenthecomputingfield isadjacenttheshockorcontact discontinuities forthesake ofavoidingspuriousoscillations.Thereafter, thenewnonlinearweightformulasarepresented forthefinal highordernumerical fluxapproximation procedure.When alarge, centeredstencil permitsoptimalstability andaccuracytobereached,thismethodwillcertainlyreachthatstabilityandaccuracy.However,whenthesolutiononthat stencilisrough,itisbeneficialtoseekoutasmallerTVD-likestencil.Thetwosmallstencilsthatareusedensurethat(when thelargestencilmaybeignored)themethodiseffectivelyaTVDschemewithavanAlbada-likelimiter.Consequently,for non-smoothsolutionvector, thisschemeoperates likea TVDschemewithallthe attendant stabilityproperties ofaTVD scheme. In other words, the fact that a TVD-like property is built into the method makes it closer to a Monotonicity PreservingWENOthanapureWENO-JS(whichalwaysreliesonanensembleoflargestencils).Allinall,theinnovationsof thispaperlieinthreeaspects:thenewwayofreconstructingamodifiedfourthdegreepolynomialbysubtractingtwolinear polynomialswithproperparameters,anovelWENOtypecommunicationamongthreeunequal(modified)polynomialsfor highaccurateapproximationsinsmoothregionandnon-oscillationinnonsmoothregion,andthenewmannerofobtaining associatednonlinearweights.

Theorganizationofthepaperisasfollows:inSection2,weconstructthenewfinitedifferencefifthorderWENOscheme indetail.InSection 3,someclassical numericaltestsarepresentedtoverifythenumericalaccuracy andefficiencyofthe newWENOscheme.ConcludingremarksaregiveninSection4.

2. NewsimpleWENOscheme

Onedimensionalhyperbolicconservationlawsis

ut

+

fx

(

u

) =

0

,

u

(

x

,

0

) =

u0

(

x

),

(2.1)

andtheassociatedsemidiscretizationof(2.1)canbereformulatedas du

dt

=

L

(

u

),

(2.2)

where L(u) isthe highorderspatial discrete formulationof −fx(u). Forsimplicity,the uniformmeshis distributedinto some cells Ii= [xi1/2,xi+1/2],withtheuniformcellsizeisdenotedasxi+1/2xi1/2=handassociatedcellcentersare definedasxi=12(xi+1/2+xi1/2).ui(t)isdefinedasanodalpointvalueu(xi,t).Thereforetherighthandsideof(2.2)can bereformulatedas

L

(

ui

(

t

)) = −

1

h

( ˆ

fi+1/2

− ˆ

fi1/2

),

(2.3)

where ˆfi+1/2 isanumericalfluxwhichhasafifthorderapproximationofflux f(u)attheboundaryxi+1/2oftargetcellIi inthispaper.Ifthenumericalflux ˆfi+1/2istakenintoaccountthefifthorderapproximation,then 1h(ˆfi+1/2− ˆfi1/2)isthe fifthorderapproximationto fx(u)atx=xi.Foranordinaryflux f(u),itcanbesplitinto f(u)= f+(u)+f(u),satisfying

df+(u)

du0 and dfdu(u)0.A simplest Lax–Friedrichs splittingis applied as f±(u)= 12(f(u)±

α

u), inwhich

α

is set as

(3)

maxu|f(u)|overthewholerangeofu.ThedetailedflowchartofthenewsimplefifthorderWENOschemeisdescribedas follows,forsimplicity,weonlydothereconstructionof f+(u)atpointxi+1/2anddenoteitas ˆfi++1/2.

Step1.Choosethefollowingbigstencil:T1= {Ii2,Ii1,Ii,Ii+1,Ii+2}.Itiseasytoobtainafourthdegreereconstructed polynomial p1(x)whichisbasedonthenodalpointinformationofthefluxsplittingandsatisfying:

1 h

xj

+h/2 xjh/2

p1

(

x

)

dx

=

f+j

,

j

=

i

2

,

i

1

,

i

,

i

+

1

,

i

+

2

.

(2.4)

Itsexplicitexpressionissameasspecifiedin[1]

p1

(

x

) =

fi+

+ −

82fi+1

+

11fi+2

+

82fi++1

11fi++2

120

(

x

xi

h

) +

40fi+1

3fi+2

74fi+

+

40fi++1

3fi++2

56

((

x

xi

h

)

2

1 12

) +

2fi+1

fi+2

2fi++1

+

fi++2

12

((

x

xi

h

)

3

3 20

(

x

xi

h

)) +

4fi+1

+

fi+2

+

6fi+

4fi++1

+

fi++2

24

((

x

xi

h

)

4

3 14

(

x

xi

h

)

2

+

3

560

).

(2.5)

Choose another two smaller stencils: T2= {Ii1,Ii} and T3= {Ii,Ii+1}.It isalso very easy forus to get the two linear polynomialswhicharebasedonthenodalpointinformationofthefluxsplittingandsatisfying

1 h

xj

+h/2 xjh/2

p2

(

x

)

dx

=

f+j

,

j

=

i

1

,

i

,

(2.6)

and 1 h

xj

+h/2 xjh/2

p3

(

x

)

dx

=

f+j

,

j

=

i

,

i

+

1

.

(2.7)

Theirexplicitexpressionsare

p2

(

x

) =

fi+

+ (

fi+

fi+1

)(

x

xi

h

),

(2.8)

and

p3

(

x

) =

fi+

+ (

fi++1

fi+

)(

x

xi

h

).

(2.9)

Step 2. The main selection principle ofthe linear weights is solelybased on the consideration ofa balance between accuracy and ability to achieve essentially nonoscillatory shock transitions. In all of our numerical tests, following the practice in[7,26],wetake thepositivelinearweights as

γ

1=0.98 and

γ

2=

γ

3=0.01.Thelinearweightscan bechosen to beanysetofpositivenumberswiththecondition thatthesummationisoneandwouldnotpollutethenewscheme’s optimalaccuracy.

Step3.Computethesmoothnessindicatorsβn, n=1,2,3,whichmeasurehowsmooththefunctionspn(x), n=1,2,3, are inthetarget cell Ii.The smallerthesesmoothnessindicators,the smootherthefunctionsarein Ii.We usethesame recipeforthesmoothnessindicatorsasin[1,14,20]

β

n

=

r α=1

Ii

h2α1

(

d

αpn

(

x

)

dxα

)

2dx

,

n

=

1

,

2

,

3

.

(2.10)

Theassociatedexplicitexpressionsare

β

1

= (

82fi+1

+

11fi+2

+

82fi++1

11fi++2

120

+

2fi+1

fi+2

2fi++1

+

fi++2

120

)

2

+

(4)

13 3

(

40f

i+−1

3fi+2

74fi+

+

40fi++1

3fi++2

56

+

123 455

4fi+1

+

fi+2

+

6fi+

4fi++1

+

fi++2

24

)

2

+

781 20

(

2f

+

i1

fi+2

2fi++1

+

fi++2

12

)

2

+

1421461

2275

(

4fi+1

+

fi+2

+

6fi+

4fi++1

+

fi++2

24

)

2

,

(2.11)

β

2

= (

fi+1

fi+

)

2

,

(2.12)

and

β

3

= (

fi+

fi++1

)

2

.

(2.13)

Theexpansionsof(2.11)to(2.13)inTaylorseriesabout fi+areobtainedas

β

1

=

h2

((

fi+

)

)

2

+

13h4

((

fi+

)

)

2

/

12

+

h6

(

5467

((

fi+

)

)

2

14

(

fi+

)

(

fi+

)

(4)

336

(

fi+

)

(

fi+

)

(5)

)/

5040

+

O

(

h8

) =

h2

((

fi+

)

)

2

(

1

+

O

(

h4

)) =

O

(

h2

),

(2.14)

β

2

=

h2

((

fi+

)

)

2

h3

(

fi+

)

(

fi+

)

+

h4

( (

fi+

)

4

+ (

fi+

)

(

fi+

)

3

) +

O

(

h5

) =

h2

((

fi+

)

)

2

(

1

+

O

(

h

)) =

O

(

h2

),

(2.15)

and

β

3

=

h2

((

fi+

)

)

2

+

h3

(

fi+

)

(

fi+

)

+

h4

( (

fi+

)

4

+ (

fi+

)

(

fi+

)

3

) +

O

(

h5

) =

h2

((

fi+

)

)

2

(

1

+

O

(

h

)) =

O

(

h2

).

(2.16)

Itisassumedthat theindicatorsofsmoothnesscanbe rewrittenas:β1=D(1+O(h4))andβ2,3=D(1+O(h)),inwhich D=h2((fi+))2 isanon-zeroconstantindependentofnonconditionthat(fi+)=0.

Step 4. Calculate the non-linear weights based on the linearweights and the smoothnessindicators. Forinstance, as shownin[3,6],weusenew

τ

whichissimplydefinedastheabsolutedeferencebetweenβ1,β2 andβ3,andisdifferentto theformulaspecifiedin[3,6].SincethetwodifferenceexpansionsinTaylorseriesabout fi+are

β

1

β

2

= (

fi+

)

(

fi+

)

h3

+ (

5

((

fi+

)

)

2

6

(

fi+

)

(

fi+

)

3

)

h4

+

O

(

h5

)

=

O

(

h3

), (

fi+

)

=

0

, (

fi+

)

=

0

,

O

(

h4

), (

fi+

)

=

0

, (

fi+

)

=

0

,

(2.17)

and

β

1

β

3

= − (

fi+

)

(

fi+

)

h3

+ (

5

((

fi+

)

)

2

6

(

fi+

)

(

fi+

)

3

)

h4

+

O

(

h5

)

=

O

(

h3

), (

fi+

)

=

0

, (

fi+

)

=

0

,

O

(

h4

), (

fi+

)

=

0

, (

fi+

)

=

0

.

(2.18)

Itiseasytoverify

τ = (

1

β

2

| + |β

1

β

3

|

2

)

2

=

O

(

h6

), (

fi+

)

=

0

, (

fi+

)

=

0

,

O

(

h8

), (

fi+

)

=

0

, (

fi+

)

=

0

.

(2.19)

Thenwedefine

ω

n

= ω ¯

n

3

=1

ω ¯

, ω ¯

n

= γ

n

(

1

+ τ ε + β

n

),

n

=

1

,

2

,

3

.

(2.20)

Here

ε

isasmallpositivenumbertoavoidthedenominatortobecomezero.Bytheusageof(2.14)to(2.16) and(2.19) in thesmoothregion,itsatisfies

τ

ε + β

n

=

O

(

h4

),

n

=

1

,

2

,

3

,

(2.21)

(5)

oncondition that

ε

βn.Therefore,thenonlinearweights

ω

n, n=1,2,3,satisfytheorderaccuracycondition

ω

n=

γ

n+ O(h4)[3,6],providingtheformalfifthorderaccuracytotheWENOschemeelaboratedin[14,20].Wetake

ε

=106 inour computation.

Remark.Forcase(fi+)=0,(fi+)=0,itisalsoeasytoverifythat

ω

n=

γ

n+O(h4). Step5.Thenewfinalreconstructionsofthenumericalfluxatx=xi+1/2isgivenby

ˆ

fi++1/2

= ω

1

(

1

γ

1

p1

(

xi+1 2

)γ

2

γ

1

p2

(

xi+1 2

)γ

3

γ

1

p3

(

xi+1

2

)) + ω

2p2

(

xi+1

2

) + ω

3p3

(

xi+1

2

).

(2.22)

The first termof the right hand side of (2.22) looks very complicatedat the first glance. If (2.22) is simply definedas

ω

1p1(xi+1

2)+

ω

2p2(xi+1

2)+

ω

3p3(xi+1

2)as usual, the schemewould degrade its optimalfifth order accuracy because of p2(xi+1

2)andp3(xi+1

2)areactiveandtheconvexcombinationtogetherwithp1(xi+1

2)wouldnotofferhighorderapproxi- mation atpoint xi+1/2 tonumericalfluxinsmoothregion.Thusweshoulddosomethingtowipeofftheircontributionin smoothregion.Bydoingso,ifabigspatialstencilpermitsoptimalstabilityandaccuracytobereached,(2.22)couldobtain thatstability andaccuracyobviously.However,whenthesolutiononthebigspatialstencilisrough,itisbeneficialtoseek outsmallertwo-pointTVD-likestencils.Suchtwosmallspatialstencilsthatareusedensurethat(whenthelargefive-point spatial stencil may be inactive andignored) the equation (2.22) can gradually transit to become a TVD scheme with a vanAlbada-likelimiter.So,forthepurposeofsolvingnon-smoothnumericalsolutions,(2.22)performs likeaTVDscheme withalltheattendant stabilitypropertiesofaTVDscheme.Inthisway,itisnotverycrucialtodeliberatelychooselinear weights forthepurposeofsustaininghighorderaccuracyinsmoothregionandcouldkeeptheshocktransitionsharplyin nonsmoothregion.

Step 6.Thesemidiscretescheme(2.2)isdiscretizedintimebyathirdorderTVDRunge–Kuttamethod[21]

⎧ ⎪

⎪ ⎩

u(1)

=

un

+

t L

(

un

),

u(2)

=

34un

+

14u(1)

+

14 t L

(

u(1)

),

un+1

=

13un

+

23u(2)

+

23 t L

(

u(2)

).

(2.23)

Remark. Forthe systemsofconservationlaws,such asthecompressibleEulerequations,all ofthereconstruction proce- duresare implementedinthelocalcharacteristicdirectionsforthepurposeofavoidingspurious oscillations.Forthetwo dimensionalproblems,allofthesereconstructionproceduresarecarriedoutina dimension-by-dimensionfashion.

3. Numericaltests

In this section we presentthe results ofnumerical tests ofthe fifth ordernew WENO schemetermed as WENO-ZQ specified intheprevious section incomparisontotheclassical WENOschemetermedasWENO-JS narratedin[14,20] in one andtwodimensions. TheCFLnumberissetas0.6forWENO-JSandWENO-ZQschemesinourcomputations.Forthe purposeoftestingwhethertherandomchoiceofthelinearweightswouldpollutetheoptimalorderaccuracyofWENO-ZQ scheme ornot, we set differenttype of linearweights in thenumerical accuracycases as:(1)

γ

1=0.98,

γ

2=0.01 and

γ

3=0.01;(2)

γ

1=1.0/3.0,

γ

2=1.0/3.0 and

γ

3=1.0/3.0;(3)

γ

1=0.01,

γ

2=0.495 and

γ

3=0.495.Andset

γ

1=0.98,

γ

2=0.01 and

γ

3=0.01 inthelatterexamples.

Example3.1.WesolvethefollowingnonlinearscalarBurgersequation:

μ

t

+ ( μ

2

2

)

x

=

0

,

0

<

x

<

2

,

(3.1)

with the initial condition

μ

(x,0)=0.5+sin(

π

x)and periodic boundary condition. When t=0.5/

π

the solution isstill smooth,andtheerrorsandnumericalordersofaccuracybytheWENO-ZQschemeareshowninTable 3.1.Forcomparison, errors andnumericalordersofaccuracy bythe classicalWENO-JS schemeare showninthesame table.We canseethat bothWENO-ZQandWENO-JSschemesachievetheirdesignedorderofaccuracy,andWENO-ZQschemewithdifferenttype of linearweights produces lesstruncationerrors andachieves optimalorders ofaccuracy. Fig. 3.1 showsthat WENO-ZQ schemeneedslessCPUtimethanWENO-JS doestoobtainthesamequantitiesof L1 andL errors,soWENO-ZQscheme ismoreefficientthanWENO-JSschemeinthistestcase.

Example3.2.Wesolvethefollowingtwo-dimensionalnonlinearscalarBurgersequation:

μ

t

+ ( μ

2

2

)

x

+ ( μ

2

2

)

y

=

0

,

0

<

x

,

y

<

4

,

(3.2)

withtheinitialcondition

μ

(x,y,0)=0.5+sin(

π

(x+y)/2)andperiodicboundaryconditions.Whent=0.5/

π

thesolution isstillsmooth,andtheerrorsandnumericalordersofaccuracybytheWENO-ZQschemeincomparisontothatofWENO-JS

(6)

Table 3.1

μt+(μ22)x=0.Initialdataμ(x,0)=0.5+sinx).WENO-ZQschemeandWENO-JSscheme.T=0.5.L1andLerrors.

Grid points WENO-ZQ (1) scheme WENO-JS scheme

L1error Order Lerror Order L1error Order Lerror Order

10 1.64E2 5.32E2 1.91E2 7.48E2

20 1.44E3 3.51 9.73E3 2.45 2.06E3 3.21 1.21E2 2.63

40 7.29E5 4.31 6.93E4 3.81 1.24E4 4.05 1.03E3 3.54

80 2.38E6 4.93 3.06E5 4.50 4.41E6 4.81 4.72E5 4.46

160 7.07E8 5.07 9.31E7 5.04 1.64E7 4.75 1.38E6 5.09

320 2.09E9 5.07 2.78E8 5.06 4.76E9 5.11 7.28E8 4.25

Grid points WENO-ZQ (2) scheme WENO-ZQ (3) scheme

L1error Order Lerror Order L1error Order Lerror Order

10 3.48E2 1.10E1 3.76E2 1.16E1

20 4.55E3 2.93 1.55E2 2.83 5.61E3 2.74 1.95E2 2.58

40 1.41E4 5.01 8.15E4 4.25 1.68E4 5.06 1.13E3 4.11

80 2.57E6 5.77 3.05E5 4.74 2.68E6 5.97 3.04E5 5.21

160 7.10E8 5.18 9.31E7 5.03 7.12E8 5.23 9.31E7 5.03

320 2.09E9 5.08 2.78E8 5.06 2.09E9 5.09 2.78E8 5.06

Fig. 3.1.μt+(μ22)x=0.Initialdataμ(x,0)=0.5+sinx).Computingtimeanderror.NumbersignsandasolidredlinedenotetheresultsofWENO-ZQ schemewithdifferentlinearweights(1),(2)and(3);squaresandasolidlinedenotetheresultsofWENO-JSscheme.(Forinterpretationofthereferences tocolorinthisfigurelegend,thereaderisreferredtothewebversionofthisarticle.)

schemeareshowninTable 3.2.WecanseethattheWENO-ZQschemewithdifferenttypeoflinearweightsachievesclose to its designed order of accuracy and generates less absolute truncation errors. And the Fig. 3.2 shows that WENO-ZQ schemeneedslessCPUtimethanWENO-JSschemedoestoobtainthesamequantitiesofL1 andLerrors.

Example3.3.

t

ρ ρμ

E

⎠ +

x

ρμ ρμ

2

+

p

μ (

E

+

p

)

⎠ =

0

.

(3.3)

In which

ρ

is density,

μ

is the velocity in x-direction, E is total energy and p is pressure. The initial conditions are:

ρ

(x,0)=1+0.2sin(x),

μ

(x,0)=1,p(x,0)=1,

γ

=1.4.Thecomputingdomainisx∈ [0,2

π

].Periodicboundarycondition isappliedinthistest.Theexactsolutionis

ρ

(x,t)=1+0.2sin(xt).Thefinaltimeist=2.Theerrorsandnumericalorders ofaccuracyofthedensitybytheWENO-ZQschemeandWENO-JSschemeareshowninTable 3.3andthenumericalerror againstCPUtime graphsareinFig. 3.3.Wecanobservethat thetheoreticalorderis actuallyachievedandthe WENO-ZQ schemecangetbetterresultsandismoreefficientthanWENO-JSschemeinthistestcase.

(7)

Table 3.2

μt+(μ22)x+(μ22)y=0.Initialdataμ(x,y,0)=0.5+sin(π(x+y)/2).WENO-ZQschemeandWENO-JSscheme.T=0.5.L1andLerrors.

Grid points WENO-ZQ (1) scheme WENO-JS scheme

L1error Order Lerror Order L1error Order Lerror Order

10×10 1.84E2 5.38E2 2.06E2 7.49E2

20×20 1.73E3 3.41 9.49E3 2.50 2.16E3 3.25 1.21E2 2.63

40×40 7.46E5 4.54 6.93E4 3.78 1.27E4 4.09 1.03E3 3.54

80×80 2.41E6 4.95 3.06E5 4.50 4.47E6 4.83 4.72E5 4.46

160×160 7.11E8 5.08 9.31E7 5.04 1.65E7 4.76 1.38E6 5.09

320×320 2.09E9 5.08 2.78E8 5.06 4.77E9 5.11 7.28E8 4.25

Grid points WENO-ZQ (2) scheme WENO-ZQ (3) scheme

L1error Order Lerror Order L1error Order Lerror Order

10×10 3.81E2 1.10E1 4.11E2 1.17E1

20×20 4.82E3 2.98 1.56E2 2.82 5.85E3 2.81 1.92E2 2.60

40×40 1.42E4 5.08 7.94E4 4.30 1.70E4 5.10 1.10E3 4.12

80×80 2.61E6 5.77 3.05E5 4.70 2.71E6 5.97 3.04E5 5.18

160×160 7.14E8 5.19 9.30E7 5.03 7.17E8 5.24 9.30E7 5.03

320×320 2.09E9 5.09 2.78E8 5.06 2.09E9 5.09 2.78E8 5.06

Fig. 3.2.μt+(μ22)x+(μ22)y=0.Initialdataμ(x,y,0)=0.5+sin(π(x+y)/2).Computingtimeanderror.Numbersignsandasolidredlinedenotethe resultsofWENO-ZQschemewithdifferentlinearweights(1),(2)and(3);squaresandasolidlinedenotetheresultsofWENO-JSscheme.(Forinterpretation ofthereferencestocolorinthisfigurelegend,thereaderisreferredtothewebversionofthisarticle.)

Table 3.3

1D-Eulerequations:initialdataρ(x,0)=1+0.2sin(x),μ(x,0)=1 andp(x,0)=1.WENO-ZQschemeandWENO-JSscheme.T=2.L1andLerrors.

Grid points WENO-ZQ (1) scheme WENO-JS scheme

L1error Order Lerror Order L1error Order Lerror Order

10 1.36E3 3.51E3 4.49E3 6.79E3

20 3.04E5 5.49 5.97E5 5.87 2.15E4 4.38 3.75E4 4.18

40 9.62E7 4.98 1.60E6 5.21 6.74E6 5.00 1.28E5 4.87

80 3.01E8 5.00 4.75E8 5.07 2.07E7 5.02 3.99E7 5.01

160 9.39E10 5.00 1.47E9 5.01 6.38E9 5.02 1.15E8 5.11

320 2.93E11 5.00 4.60E11 5.00 1.92E10 5.05 3.16E10 5.19

Grid points WENO-ZQ (2) scheme WENO-ZQ (3) scheme

L1error Order Lerror Order L1error Order Lerror Order

10 8.59E3 2.27E2 1.08E2 2.70E2

20 1.30E4 6.05 3.98E4 5.84 1.88E4 5.85 5.31E4 5.67

40 9.98E7 7.02 5.07E6 6.29 1.15E6 7.34 6.80E6 6.28

80 3.01E8 5.05 6.48E8 6.29 3.01E8 5.27 7.35E8 6.53

160 9.39E10 5.00 1.54E9 5.39 9.39E10 5.00 1.58E9 5.53

320 2.93E11 5.00 4.60E11 5.07 2.93E11 5.00 4.61E11 5.10

Références

Documents relatifs

8a shows the density from the forward facing step simulation at a time of 3 units when the WENO-AO(7,5,3) scheme was used. 8b shows the same for the WENO-ZQ scheme. The vortex

In Table 17, Table 18 and Table 19, number of iterations required for convergence, the final time and total CPU time when the scheme converges are reported for the FE Jacobi scheme

Parallel adaptive mesh refinement method based on WENO finite difference scheme for the simulation of.. multi-dimensional

We de- scribe the Lax-Wendroff time discretization of this alternative formulation of high order conservative finite difference schemes in detail, using fifth order WENO

[18] R. Zhao, A multigrid block lower-upper sym- metric Gauss-Seidel algorithm for steady Euler equation on unstructured grids, Numer.. Jiang and C.-W. Shu, Efficient implementation

The main objective of this paper is to design a finite difference WENO scheme for the gas dynamic equations with gravitational source terms, which maintains the well-balanced

Keywords: positivity preserving; high order accuracy; compressible Euler equations; gas dynamics; finite difference scheme; essentially non-oscillatory scheme; weighted

squares: the results of WENO scheme. From left to right: the third order scheme; the fifth order scheme.. From top to bottom: density; density zoomed in; the points where the