• Aucun résultat trouvé

An asymptotic-preserving method for a relaxation of the Navier-Stokes-Korteweg equations

N/A
N/A
Protected

Academic year: 2021

Partager "An asymptotic-preserving method for a relaxation of the Navier-Stokes-Korteweg equations"

Copied!
18
0
0

Texte intégral

(1)

HAL Id: hal-02498895

https://hal.archives-ouvertes.fr/hal-02498895

Submitted on 4 Mar 2020

HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

An asymptotic-preserving method for a relaxation of the Navier-Stokes-Korteweg equations

Alina Chertock, Pierre Degond, Jochen Neusser

To cite this version:

Alina Chertock, Pierre Degond, Jochen Neusser. An asymptotic-preserving method for a relaxation

of the Navier-Stokes-Korteweg equations. Journal of Computational Physics, Elsevier, 2017, 335,

pp.387-403. �10.1016/j.jcp.2017.01.030�. �hal-02498895�

(2)

Contents lists available atScienceDirect

Journal of Computational Physics

www.elsevier.com/locate/jcp

An asymptotic-preserving method for a relaxation of the Navier–Stokes–Korteweg equations

Alina Chertocka,Pierre Degondb,∗, Jochen Neusserc

aDepartmentofMathematics,NorthCarolinaStateUniversity,Raleigh,NC27695,USA bDepartmentofMathematics,ImperialCollegeLondon,LondonSW72AZ,UK

cUniversitätStuttgart,InstitutfürAngewandteAnalysisundNumerischeSimulation,Pfaffenwaldring57,D-70569Stuttgart,Germany

a r t i c l e i n f o a b s t r a c t

Articlehistory:

Received15December2015

Receivedinrevisedform20December2016 Accepted17January2017

Availableonline24January2017 Keywords:

Asymptotic-preservingscheme Diffuse-interfacemodel

Compressibleflowwithphasetransition

The Navier–Stokes–Korteweg (NSK)equations are a classicaldiffuse-interface model for compressibletwo-phaseflows.AsdirectnumericalsimulationsbasedontheNSKsystem are quiteexpensive andinsomecaseseven impossible,weconsiderarelaxationofthe NSKsystem,forwhichrobustnumericalmethodscanbedesigned.However,timestepsfor explicitnumericalschemes dependonthe relaxationparameterandthereforenumerical simulations inthe relaxationlimit are veryinefficient.To overcome thisrestriction, we proposean implicit–explicitasymptotic-preservingfinite volumemethod. Weprovethat thenewschemeprovidesaconsistentdiscretizationoftheNSKsystemintherelaxation limitanddemonstratethatitiscapableofaccuratelyandefficientlycomputingnumerical solutionsofproblemswithrealisticdensityratiosandsmallinterfacialwidths.

©2017TheAuthor(s).PublishedbyElsevierInc.Thisisanopenaccessarticleunderthe CCBYlicense(http://creativecommons.org/licenses/by/4.0/).

1. Introduction

There are in general two approaches to describe the behavior ofmulti-phase fluids, the sharpinterface (SI) andthe diffuse interface (DI) approach. The first approach represents multiple phases with different sets of equations that are coupledby some interface conditions.Thesecond approach,which usedtodescribe, e.g. themergingprocess ofdroplets andbubbles,needsonlyonesetofequationstomodelthephasesanddoesnotrequirethelocationoftheinterface tobe trackedexplicitly.

Inthispaper,weconsidertwoDImodelsforahomogeneoustwo-phasecompressiblefluid:TheNavier–Stokes–Korteweg system(NSK)andarelaxationsystemfortheNSKsystem.TheNSKsystemgoesbacktotheworkofKorteweg[31]andwas formulatedinitspresentformin[18,2].TheNSKsystemusesaVan-der-Waalslikepressurefunctiontoidentifytwodistinct phasesandathird-ordertermtomodelphasetransitions.Manyauthorsachievedanalyticalresultsonthewell-posedness oftheNSKsystemanditsvariants,e.g.[4,7,24,32,37].Whilesomenumericalmethodshavebeensuccessfullydevelopedand usedfortheseandrelatedfamiliesofproblems,see,e.g., [6,8,16,21,25,39],there arestill manyopenproblems,forwhich accurateandefficientnumericalmethodsareyettobedesigned.Up toourknowledge,therobustcomputationofrealistic densityvaluesforliquidandvaporphaseshasnotbeensuggestedfortheNSKmodel.Additionally,numericalmethodsalso failincaseswhenverysmallinterface widthsclosetoasharpinterface aretobeconsidered.Inbothcases,theoccurring problems arerelated tosteep densitygradients. Anothersource of difficulty one comesacross while numericallysolving

* Correspondingauthor.

E-mailaddresses:chertock@math.ncsu.edu(A. Chertock),p.degond@imperial.ac.ukm(P. Degond),jochen.neusser@mathematik.uni-stuttgart.de (J. Neusser).

http://dx.doi.org/10.1016/j.jcp.2017.01.030

0021-9991/©2017TheAuthor(s).PublishedbyElsevierInc.ThisisanopenaccessarticleundertheCCBYlicense (http://creativecommons.org/licenses/by/4.0/).

(3)

NSKsystemsisrelatedtotheVan-der-Waalslikeformofthepressureequation.Thelatterpreventsonefromusingupwind hyperbolic solvers, whichhave beensuccessfullyapplied, e.g.,to stabilizecomputationsforNavier–Stokes equationswith highReynoldsnumbers.

The issueofvery smallinterfacesisespeciallyimportantbecausetheNSKmodelcanonly providethecorrectamount ofcapillaryforcesiftheinterfaceisextremelysmall[17,25,27].One ideatoloosenthestrict couplingbetweeninterfacial width andcapillaryforces istointroduce an additionalCahn–HilliardorAllen–Cahn type equation foranewphase field variable.Thiswasdoneforexamplein[1,5,42].AnotheransatztoavoidsomeofthedifficultiesfortheNSKsystemssuggests tointroducearelaxationoftheNSKsystem,inwhichthethird-ordertermisreplacedbyafirst-ordertermandaPoisson equation, that defines a newphase field parameter, see, e.g., [37]. This modelis parametrized by a so-calledKorteweg parameter α. IftheKorteweg parameter tends tozero, therelaxationsystem formallyconverges to theNSKsystem. The most importantfeature of the relaxationsystem isthe fact, that the first-order partis purely hyperbolic for sufficiently small Korteweg parameter. It should also be observed that the addition ofthe Poisson equation to the systemdoesnot increase thecomputationalcostofnumericalsimulationsastest casesdemonstratethatthetime savings,thatcomefrom the fact that one does not have to solve a third order system, are greater than the loss of time that comes from the numerical solutionofthe Poissonequation. Thesepropertiescan be exploitedtoconstructrobust numericalschemes for therelaxationsystem.In[34],ithasbeenshownthattheoverallapproachisrobustforproblemswithlargedensityratios andsmallinterfacialwidths.However,thenumericalschemeproposedin[34]isanexplicitschemeandthusthetimesteps decreaseastheKortewegparameter αtendstozero.

It is the main purpose of thiscontribution to constructan asymptotic-preserving (AP) scheme [28] in the Korteweg limit, that is,a schemefortherelaxationsystemthat provides aconsistent approximation oftheoriginal NSKsystemas theKortewegparameter α tendstozero.TheAPapproachwasdevelopedintheframeworkoflineartransportindiffusive regimes[22,29,33]andhasbeenappliedtomanydifferentareas,e.g.fluidanddiffusionlimitsofkineticmodels,relaxation methodsforhyperbolicsystemsandlow-Mach numberlimitsforcompressibleflowproblems;see,e.g.,[11,12,14,15,20,23, 30,35,36].In[23],thetimeandspatial discretizationoftheisentropicEulerandNavier–StokesEquationsinthelowMach numberlimitwas investigated.Inspiredbythisresearch,weconstructhereaschemethatcapturestheKorteweglimitfor therelaxationsystemandprovetheAPpropertyoftheproposedscheme.TheAPpropertyofthenewschemeisachieved bysplittingtherelaxationsystemintoanon-stiffnonlinear,compressiblehyperbolicNavier–Stokeslikesystemandasystem that canbetreatedbyaPoissonsolver,andallowstheuseoftimeandspatialstepsthatareindependentoftheKorteweg parameter. Asthe resulttheproposednumericalscheme isveryefficientforsmallvaluesoftheparameter α,whichisa significant improvementcomparedto anexplicitschemefrom[34].Beyondthat,we expectourschemeto beasymptotic preservinginthesharpinterfacelimit.Wecannotgiveanalyticalprooftothat,butwesupportthisstatementbyanumerical example.

The paperisorganized asfollows.InSection 2,weintroduce theNSKandrelaxationsystemsanddescribe their main propertiestogether withthe basicthermodynamicalframework. Wecommentontheadvantagesoftherelaxationsystem andpointout whyitisnecessarytointroduceanewschemeinordertosolvetherelaxationsystemefficiently.Section 3 contains the basic outcome of this contribution. We propose the AP scheme forthe relaxation systemand perform an asymptotic analysisto show that the schemetransforms into a schemefor the NSKequations in theKorteweg limit.In Section4,wedemonstratethatthealgorithmprovidesamassiveimprovementcomparedtoastandardexplicitschemefor anumberofproblemsinoneandtwospacedimensions.

2. Navier–Stokes–Kortewegequations 2.1. TheNavier–Stokes–Kortewegsystem

Letanopen set⊆Rd,d∈ {1,2,3}uptothefinaltime T>0 begiven.TheisentropicNavier–Stokes–Korteweg(NSK) equationsinarbitraryspatialdimensionaregivenby

tρ+ ∇ ·v)=0,

tv)+ ∇ ·vv)+ ∇(p(ρ))= ∇ ·Tε[v] +γ ε2ρρ,(x,t)×(0,T), (2.1) where ρ=ρ(x,t)isthe densityofthefluid,v=(v1(x,t),...,vd(x,t))T∈Rd isitsmomentumand p(x,t)isthepressure.

Notethat εistheReynoldsnumberand γ ε2 isthecapillarynumber.Wereferto[26,41]foradetailedexplanationonthe physical meaningofthescaling ε0 and γ =O(1).The matrixTε[v]∈Rd×d in(2.1)stands fortheviscous partofthe stresstensorwhichisgivenfortheviscositycoefficients ν∈Rwith μ ≥0 and3ν+2μ >0 by

Tεi j:=εν∇ ·(vi j+2εμDi j, Di j:=1 2

vj,xi+vi,xj

, (i,j∈ {1,2}). (2.2)

Weaugment(2.1)withtheinitialdata

ρ(x,0)=ρ0, v(x,0)=v0, x, (2.3)

(4)

Fig. 2.1.Pressurepas a function ff densityρfor the valuesb1=b=3,T=0.85, andR=8.

andboundaryconditionsthatcorrespondtoaboundedbox:

v=0,ρ·n=0, x∂. (2.4)

Todescribeatwo-phasefluidwechoosetheVan-der-Waalstypepressure p(ρ)= R Tρ

bρ b1ρ2. (2.5)

Thereby,b,b1,R are positive constants and T isthe fixed temperature. If T is chosen smallenough, the pressure p is monotonedecreasinginsomenon-emptydensityinterval.Thisstructureallowsonetodefinephases.Ifthedensity ρ lies intheinterval(01],(12)),{[α2,b)}thecorrespondingfluidstateiscalledvapor(spinodal){liquid},seeFig. 2.1foran illustration.

Weobservethatthefirst-orderpartof(2.1)isnotpurelyhyperbolicforalldensityvalues.Indeed,consider,forinstance, theone-dimensional(1-D)case,d=1.ItiseasytocheckthattheeigenvaluesoftheJacobianofthefirstorderpartof(2.1) Df1(ρ,ρv)∈R2×2are

λ1(ρ,u)=v

p(ρ), λ2(ρ,u)=v+

p(ρ) (2.6)

withcorrespondingeigenvectors K1(ρ,v)=

1 v

p(ρ)

, K2(ρ,v)=

1 v+

p(ρ)

. (2.7)

Therefore,thefirstorderpartof(2.1)ishyperbolicifandonlyif ρ∈(0,b)\ [α12].

Thelackofhyperbolicityinthefirst-orderpartof(2.1)andthepresenceofthethird-orderderivativeinthemomentum balancemake thenumericalsolutionof(2.1)tobe achallenging task:Explicit schemessufferfromextremelysmalltime stepswhile implicitdiscretizations leadtobadly conditionedalgebraicproblems.The non-monotonicityof p preventsthe useofmostmodernshock-capturingschemes.

2.2. ArelaxationfortheNavier–Stokes–Kortewegsystem

ToovercomesomeoftheshortcomingsoftheclassicalNSKsystemweproposearelaxationfortheNSKsystem[37]

⎧⎪

⎪⎪

⎪⎨

⎪⎪

⎪⎪

tρα+ ∇ ·αvα)=0,

tαvα)+ ∇ ·αvαvα)+ ∇pα)= ∇ ·Tε[vα] + 1

α2ρα∇(cαρα), γ ε2cα+ 1

α2(ραcα)=0.

(2.8)

Here α>0 istheKortewegparameterandcα isanewunknown,thatisdefinedbytheadditionalPoissonequation,andthe stresstensorTε givenby(2.2).Therelaxationsystem(2.8)isaugmentedwiththeinitialconditions

ρα(x,0)=ρ0, vα(x,0)=v0, x, (2.9)

andtheboundaryconditions

vα=0,cα·n=0, x∂. (2.10)

(5)

Thesystem(2.8)hasastructuraladvantagethatbecomesevidentwhenwerewritethetime-dependentequationsas

⎧⎪

⎪⎩

tρα+ ∇ ·αvα)=0,

tαvα)+ ∇ ·αvαvα)+ ∇

p(ρα)+ 1 2α2ρ2

= ∇ ·Tε[vα] + 1

α2ραcα. (2.11)

Again,ifweconsiderthe1-Dcaseforthesakeofsimplicity,thentheeigenvaluesoftheJacobianofthefirst-orderpartof (2.11) Df1ααvα)∈R2×2 are

λ1ααvα)=vα

p(ρα)+ 1 α2ρα, λ2ααvα)=vα+

p(ρα)+ 1 α2ρα,

(2.12)

withcorrespondingeigenvectors K1ααvα)=

1 vα

p(ρα)+α12ρα

, K2ααvα)=

1 vα+

p(ρα)+α12ρα

.

(2.13)

Astraightforwardcomputation(see[38])showsthatweobtainapurelyhyperbolicsystemfor

1

α2>|min{p(s) : s(α12)|, (2.14)

where12)istheintervalofthedecreasingpressures,i.e.,p(s)<0,seeFig. 2.1.Thesystem(2.8)–(2.10)canbeseenas anapproximationoftheclassicalNSKsystemwithααvα,cα)(ρ,ρv,ρ)fortheKorteweglimitα0,where(ρ,ρv) is the solution of thecorresponding initial boundary value problem(2.1), (2.3), (2.4). We refer to [8,9,13,19,21] for first rigorousresultsontheKorteweg limit.Itispossibletoshowthat(2.8)formallyconvergesto(2.1).Wetaketheasymptotic expansion

ρα=ρ(0)+αρ(1)+α2ρ(2)+. . . ,

ραvα=ρ(0)v(0)+αρ(1)v(1)+α2ρ(2)v(2)+. . . , cα=c(0)+αc(1)+α2c(2)+. . . ,

(2.15)

forsmall α andlookatthebalanceswithin theequationsofsystem(2.8).Thereforewecompute theTaylorexpansion at

ρ(0)forthepressure

p(ρα)=p(0))+p(0))(ραρ(0))+p (0))(ραρ(0))2+. . .

=p(0))+αp(ρ(0))(ρ(1)+αρ(2)+. . . )+α2p (0))(ρ(1)+αρ(2)+. . . )2+. . . (2.16)

Ashortcomputationleadstothefollowingtermsforthedifferentpowersof α: O(α2):

ρ(0)=c(0) (2.17)

O(α1):

ρ(1)=c(1) (2.18)

O(1):

ρt(0)+ ∇ ·ρ(0)v(0) =0,

ρ(0)v(0)

t+ ∇ ·

ρ(0)v(0)v(0) + ∇

p(0)

= ∇ ·Tε[v(0)]

+ρ(0)c(2)ρ(2)+ρ(1)c(1)ρ(1)+ρ(2)c(0)ρ(0), γ ε2c(0)=

c(2)ρ(2)

.

(2.19)

(6)

Table 2.1

DiscreteL2()-distance.Thedistance decreasesasα does.The fifthlinecontainsthe experimentalorderof convergence(EOC)withrespecttoαandthethirdlinecontainstheCPUtime.

i 1 2 3 4 5

x=0.005,αi2= 1 10 100 1000 10000

CPU-time [s] 563 654 787 828 2178

Dhi= uhuαhL2 0.25 0.033 2.9e3 2.4e4 3.7e5

EOCi=lnln(Dihi+/1Dih+i1)) 0.879 1.056 1.082 0.812

Wesubstitute(2.17),(2.18)into(2.19)andobtain

ρt(0)+ ∇ ·(0)v(0))=0,

(0)v(0))t+ ∇ ·(0)v(0)v(0))+ ∇p(0)=εv(0)+γ ε2ρ(0)∇ ·ρ(0), (2.20)

whichistheclassicalNavier–Stokes–Kortewegsystem,see(2.1).

Weillustratethisresultbythefollowingnumericalexperimentthatistakenfrom[34].

Example2.1(NumericalverificationoftheKorteweglimit).

Weconsiderthe1-Dsystem(2.8) with ε=0.01 oninterval =(−1,2)subjectto theboundaryconditions(2.10) and thefollowinginitialdata

ρ0(x)=

0.3, x(0.3,0.6)(0.85,1.05)

1.8, otherwise ,

v0(x)=0.

(2.21)

Fromthephysicalpointofview,theseinitialconditionsdescribetwovaporbubblessurroundedbyliquidfluid.Anumerical solutionof thisinitial-boundaryvalue problem, denoted by uα

h =hαhαvα h,cα

h)T andanumerical solutionofthe corre- spondingclassicalNSKsystem,denoted byuh=hhvh)T.Both solutionswerecomputedin[34]usingan explicitlocal discontinuousGalerkin(LDG)method,see,e.g.,[3,10,16].

Thenumericalresults,presentedinTable 2.1,indicatethattherelaxedmodelisanO(α2)-approximationoftheoriginal system.As onecanalsoseefromthisTable,fordecreasingvaluesof α theCPU-timeisincreasingduetothedependence on α oftheeigenvalues(2.12).Themaximumwavespeedforsystem(2.8)is λmax= |vαmax|+

pmaxα )(where ρmaxα and vα

maxarethemaximumvaluesofthedensityandvelocity,respectively).Foranexplicitschemeoneneeds

t=cD Gmin x

λmax

,x2

ε

=cD Gmin

x

|vαmax| +

p(ρmaxα ),x2

ε

, (2.22)

forsome 0cD G<1 to satisfythe CFLcondition forstability.Forsmall α oneneeds t=O(αx). Thisrestrictionisa hugedrawbackfornumericalsimulationsandweareinterestedinfindingawaytocircumventthisrestriction.

3. Anasymptotic-preservingschemeintheKorteweglimit

Havinginmindtheshortcomingsoftherelaxationsystem(2.8),weproposeanewAPnumericalscheme.Notethatfrom hereonwesuppresstheindex α ofalloftheprimalvariablesinordertoshortenthenotation.

3.1. Ahyperbolicsplitting

Thenumericalsolutionoftherelaxationsystem(2.8) requiresa resolutionoftwo scales:the(fast)wavescale,coming fromthe Korteweg part andthe (slow)convection scale. Inorder to obtain an accurate andefficient numericalscheme, whichisabletohandlebothscales,weimplementasplittingapproach.Tothisendwefirstintroduce aparametera and rewritethesystem(2.8)inthefollowingform(byaddingandsubtractingaρρ):

⎧⎪

⎪⎪

⎪⎨

⎪⎪

⎪⎪

tρ+ ∇ ·v)=0,

tvα)+ ∇ ·vαv)+ ∇ ˜p(ρ)= ∇ ·Tε[v] + 1

α2ρ(cρ)+aρρ, γ ε2c+ 1

α2(ρc)=0.

(3.1)

Herethepressureisdefinedas

˜

p(ρ)=p(ρ)+a

2ρ2 (3.2)

Références

Documents relatifs

As long as rigid solids are concerned, a local result was proved in [12], while the existence of global weak solutions is proved in [5] and [7] (with variable density) and [19]

To recover compactness in space, we will use the regularity of the limit system and extend the method used in [12] to the case of general initial data as was done in [16]..

In this paper, we consider global weak solutions to com- pressible Navier-Stokes-Korteweg equations with density dependent vis- cosities, in a periodic domain Ω = T 3 , with a

First, we establish a result in the same spirit as Theorem 1.1, but in a more general functional framework, then we prove that the solutions constructed therein are also Gevrey

In the present article, following what we did in the whole space for the non-local system (see [7]) and using lagrangian methods from [9], we will prove that under smallness

Danchin: Well-posedness in critical spaces for barotropic viscous fluids with truly nonconstant density, Communications in Partial Differential Equations, 32 , 1373–1397 (2007).

We analyse a bi-dimensional fluid-structure interaction system composed by a viscous incompressible fluid and a beam located at the boundary of the fluid domain.. Our main result is

Finally, in our previous work [3] we proved the existence and uniqueness of strong solutions in the case of an undamped beam equation but for small initial deformations.. The outline