• Aucun résultat trouvé

An example of high order Residual Distribution scheme using non Lagrange elements

N/A
N/A
Protected

Academic year: 2021

Partager "An example of high order Residual Distribution scheme using non Lagrange elements"

Copied!
21
0
0

Texte intégral

(1)

HAL Id: inria-00403691

https://hal.inria.fr/inria-00403691

Submitted on 12 Jul 2009

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

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

An example of high order Residual Distribution scheme using non Lagrange elements

Remi Abgrall, Jirka Trefilick

To cite this version:

Remi Abgrall, Jirka Trefilick. An example of high order Residual Distribution scheme using non

Lagrange elements. Journal of Scientific Computing, Springer Verlag, 2009, 45 (1-3), pp.3-25. �inria-

00403691�

(2)

Lagrange elements

R. Abgrall

(1,2)

and J. Treik

(1,3)

(1) INRIABordeaux Sud Ouest,33 405 Talene, Frane

(2)Institut Polytenhique de Bordeaux, 33 405 Talene,Frane

(3) CTU Prag,

June 8,2009

Abstrat

Weareinterestedinthenumerialapproximationofnonlinearhyperboliproblems. Thepartiular

lassof shemesweare interestedinare thesoalled ResidualDistributionshemes. Intheirurrent

form, theyrelyonthe Lagrangeinterpolationof thepoint valuesofthe approximatedfuntions. This

interpretationofthedegreesoffreedomaspointvaluesplaysafundamentalroleinthederivationofthe

shemes.ThepurposeofthepresentpaperistoshowthatsomenonLagrangeelementsanalsodothe

job,andmaybebetter. ThisopensthedoortoisogeometrianalysisintheframeworkofRDSshemes.

We are interested in the numerial approximation of linear and non linear hyperboli problems. The

partiularlassofshemesweareinterestedinarethesoalledResidualDistributionshemes. Theyanbe

traedbaktotheearlyworkofP.L.Roe[1℄andNi[2℄,butalsotothestabilizedniteelementshemessuh

astheHughes'SUPGsheme[3,4,5℄. Theirmain harateristisarethefollowing: (i)theyhaveanatural

formulation on unstrutured meshes, (ii) their stenil is the most possible ompat one to reah a given

orderofauray,(iii)theirparallelizationisstraightforward. Thesethreepropertiesaresharedinommon

with the Disontinuous Galerkin sheme, but here, thanks to the onformalnature of the approximation,

thenumberofdegreesoffreedomisreduedbyalargefator,asthisanbeseenontable 1.

2D 3D

Order DG RDS DG RD

2

6n

s

n

s

24n

s

n

s

3

12n

s

4n

s

40n

s

8n

s

4

20n

s

9n

s

80n

s

27n

s

Table1:Numberofdegreesoffreedomforthirdandfourthorderapproximationintheaseofatriangular/tet

mesh. DGstandsforDisontinuousGalerkin,RD forResidualDistribution.

Inpreviouspapers,we,andothers[6,7,8,9,10,11, 12,13℄,haveshownhowto ombinemonotoniity

preservingpropertiesandveryhighauray (

≥ 2

)ongeneralonformalmeshes, ornon onformalmeshes

[14,15℄. Oneof thekeyingredientin theonstrutionisthat thedegreesoffreedomanbeinterpretedas

pointvalues. Thepurposeofthepresentpaperistoshowthat somenonLagrangeelementsanalsodothe

job. Thisopensthedoorto isogeometrianalysis [16℄intheframework ofRDSshemes.

The format of the paper is as follows. In a rst part, we reall what are these Residual Distribution

shemes,andshowtheonstrutionofhighordershemes. Amonotoniitypriniple,orvariationdiminishing

one, plays a key role. In the seond part, we provide examples for salar steady non linear hyperboli

equations. Thethird partdisusstheextensiontotheunsteadyaseforawavemodel. Conlusionfollows.

(3)

1.1 Introdution

Letusonsiderthefollowingsalarmodelequation,

div

f (u) = S(x) x ∈ Ω ⊂ R

d

u = g

weaklyontheinowboundary

Γ

(1)

where

Γ

= {x ∈ ∂Ω, ∇

u

f (u) · ~n(x) < 0}

,

~n(x)

istheoutwardunitnormalof

at

x

. In(1),

u

and

g

belong

to

R

, andtheux

f

has

d

omponents,namely

f = (f

1

, . . . , f

d

)

. Weassumethat

f

is

C

1 and

g

belongsto

L

(Γ)

. Thedisussionwillbedevelopedusingthatsalarmodel,with

d = 2

,howeverextensionstosystems

andthease

d = 3

areratherstraightforward.

Weonsider atriangulation of

denoted by

T

h. The triangles are

{T

j

}

j=1,...,ne. We denote by

h

=

j=1,...,ne

T

j. Thevertiesofthemesharedenotedby

{M

i

}

i=1,nv. Besidestheusualregularityassumptions weneed, wealsomakethestandardassumptionthat ifanelement

T

hasapartof anedgeon

Γ

h

:= ∂Ω

h,

thisfulledgeisinludedin

h.

Ineahelement

T

,weneedanapproximationofthesolution,say

u

h,andweassumethefollowingform

u

h|T

= X

σ∈T

u

σ

ψ

σ

|T

.

(2)

In(2),thesumisindexedbydegreesoffreedomthatareseenaspointsin

T

. AtypialexampleisaLagrange

interpolant. We will assumethat the funtion

u

h is ontinuous aross edges, i.e. the

ψ

σ are ontinuous

arosstheedgesof

T

h,sothat wewrite

u

h

= X

σ

u

σ

ψ

σ

.

Morepreisely, given

k ∈ N

, we assumethat foranyfuntion smooth enough

u ∈ C

k+1

(Ω)

, weandene

u

h

= π

h

(u)

of this type, suh that if

u

is a polynomial of degree

k

, we have

u = u

h. Then, standard

approximation results, se for example [17℄, show that in

L

p norms, we have

||u − π

h

(u)|| ≤ C(u)h

k+1.

ThesepropertiesaretrueforexampleusingLagrangepolynomials,Bezier,splinerepresentationsorNURBS

[18, 19℄. We assumethat degreesof freedom also liveon the boundary of

T

, this is true for any of these

examples. Notethat thisassumptionisonsistentwiththeontinuityassumption.

Thankstothis, wedene,ineahelement

T

,thetotalresidual

Φ

T as

Φ

T

= Z

∂T

f

h

(u

h

) · ~ndl − Z

T

S(x)dx

(3)

where

f

h is someapproximationof the ux

f

. Wepreise theassumptions on

f

h abit latterin the text.

One this hasbeendone, weonsider split-residuals,

Φ

Tσ, for

σ ∈ T

, so that they satisfythe onservation property:

X

σ∈T

Φ

Tσ

= Φ

T

.

(4)

Inorder to handle boundaryonditions, we needto onsider boundary residuals. Let

Γ

bean edge of

sometriangle

T

whihison

Γ

h,weonsidertheboundaryresidual

Φ

Γ

= Z

Γ

F(u

h

, u

, ~n) − g(x) · ~n

dl

(5)

where

(F(u

h

, u

, ~n)

isanumerialupwinduxthatdependsonthetraeof

u

hon

Γ

,theboundaryondition

u

= g

,withtheunderstandingthatthenumerialuxvanishesonthenonupwind partsoftheboundary.

Then,weonsidersplit-residuals

Φ

Γσ,for

σ ∈ Γ

,sothat theysatisfytheonservationproperty:

X

σ∈Γ

Φ

Γσ

= Φ

Γ

.

(6)

(4)

Onethishasbeendone,theshemewrites: nd

u

hsuhthatforanydegreeoffreedom

σ

,

If

σ 6∈ ∂Ω

h,

Σ(u

h

) := X

T∋σ

Φ

Tσ

= 0.

(7a)

If

σ ∈ ∂Ω

h

Σ(u

h

) := X

T∋σ

Φ

Tσ

+ X

Γ⊂∂Ω,Γ∋σ

Φ

Γσ

= 0.

(7b)

Weansummarize(7a)and(7b)by

Σ(u

h

) = X

E∋i

Φ

Tσ

= 0

(7)

where

E

standseitherforanytriangle

T

oredge

Γ

thatshares

σ

.

1.2 Design priniples

1.2.1 Consistenywith (1)

Whatarethedesignpriniplesonthesheme(7)with(4)sothatwehaveaonvergentsheme? Theanswer

tothisproblemhasbeenprovidedin[13℄,andwereproduetheresult.

Proposition1.1. Assumethatthemeshisregular,thattheuxapproximation

f

h

(u

h

)

isontinuousaross

edges anddenesaonvergentapproximation (in

L

1 ofthe

C

1 ux

f

. Assumethat the residualssatisfy the

onservation relations (4)and (6). Assumethat the sheme (7)denes aunique

u

hsuh that

1. there existaonstant

C(g)

independentof

h

suhthat

||u

h

||

L2

≤ C(g)

,

2. there exists

v ∈ L

2

(Ω)

suhthat asubsequeneof

u

h onvergesto

v

in

L

2,

then

v

isaweak solution of (1)

Theresultof[13℄was aboutarstorderintimeapproximationof

∂u

∂t +

div

f (u) = 0

withinitial ondition. The adaptationtothesteady ase(1)with boundaryonditionsand souretermis

straightforward,andusesexatlythesamearguments.

1.2.2 Auray

Again,wereallpreviousresults,see[13℄Thekeyremark istosee thatif onean solve (7)aurately,the

shemeisformally

r

orderaurateifthesplit-residualsatisfy

Φ

Tσ

= O(h

r+d

), Φ

Γσ

= O(h

k+d−1

).

Thereasonfollowsfrom asimpleerroranalysis. If

ϕ

isaompatlysupported testfuntion,letus denote

ϕ

h itsLagrange interpolationdened by

ϕ

h

(σ) = ϕ(σ)

. Sayingthat, weassumethat within eah triangle,

thesetofdegreesoffreedomisunisolvant. Theexamplesofsetion2willmakethatpointlearer. Thenwe

(5)

multiplytherelations(7)by

ϕ

hσ andadd,thenusingtheonservationrelationsweobtain

E(u

h

, ϕ

h

) = X

σ∈Ω

ϕ(σ) X

T∋σ

Φ

Tσ

+ X

Γ⊂∂Ω,Γ∋σ

Φ

Γσ

!

= Z

div

f

h

(u

h

) − S

h

(u

h

)

ϕ

h

(x) dx + X

T⊂Ω

1

#{σ ∈ T } X

σ,σ∈T

ϕ(σ) − ϕ(σ

)

Φ

Tσ

− Φ

T,cσ

+ Z

∂Ω

F(u

h

, u

, ~n) − f

h

(u

h

) · ~n

ϕ

h

(x)dl+ X

Γ⊂∂Ω

1

#{σ ∈ Γ}

X

σ,σ∈Γ

ϕ(σ) − ϕ(σ

)

Φ

Γσ

− Φ

Γ,cσ

= − Z

∇ϕ

h

(x) · f

h

(u

h

) + Z

∂Ω

ϕ

h

(x)f

h

(u

h

) · ~ndl + Z

ϕ

h

(x)S

h

(u

h

)dx +

Z

∂Ω

F(u

h

, u

, ~n) − f

h

(u

h

) · ~n

ϕ

h

(x)dl

+ X

T⊂Ω

1

#{σ ∈ T } X

σ,σ∈T

ϕ(σ) − ϕ(σ

)

Φ

Tσ

− Φ

T,cσ

+ X

Γ⊂∂Ω

1

#{σ ∈ Γ}

X

σ,σ∈Γ

ϕ(σ) − ϕ(σ

)

Φ

Γσ

− Φ

Γ,cσ

.

(8)

where

ϕ

h

= π

h

(ϕ)

,

Φ

T,cσ

=

Z

T

ψ

σ

div

f (u

h

) − S(u

h

)

dx, Φ

Γ,cσ

= Z

Γ

ψ

σ

F(u

h

, u

, ~n) − f (u

h

) · ~n

dx

and

ψ

σ

∈ P

k

(T )

suhthat

ψ

σ

) = δ

σσ.

Followingagain[13℄, havethefollowingresult:

Proposition 1.2. If the solution

u

is smooth enough and the residual, applied to the

P

k interpolant of

u

satisfy

Φ

Tσ

(u

h

) = O(h

k+d

)

(9a)

and

Φ

Γσ

= O(h

k+d−1

),

(9b)

if moreoverthe approximation

f

h

(u

h

)

is

k + 1

-orderaurate,then the trunationerrorsatises

|E(u

h

, ϕ

h

)| ≤ C(ϕ, f, u) h

k+1

.

The onstant

C(ϕ, u)

depends onlyon

ϕ

and

u

.

Westartbyalemma

Lemma1.3. Forthe steady problem (1),ifthe solution

u

issmooth,wehave

Z

∂T

f

h

(u

h

) · ~ndl − Z

T

S(x)dx = O(h

k+d

)

and

Z

∂T

F(u

h

, u

, ~n) − f

h

(u

h

) · ~n

dl = O(h

k+d−1

)

provided that the approximation

f

h

(u

h

)

is

k + 1

th order aurate and the numerial ux

F

is Lipshitz

ontinuous.

(6)

Z

∂T

f

h

(u

h

) · ~ndl − Z

T

S(x)dx = Z

∂T

f

h

(u

h

) · ~n − f (u)

dl

= O(h

k+1

) × |∂T | = O(h

k+d

).

Ontheboundary,wehave

Z

∂T

F(u

h

, u

, ~n) − g(x) · ~n

dl = Z

∂T

F(u

h

, u

, ~n) − g(x) · ~n

dl + Z

∂T

F(u, u

, ~n) − g(x) · ~n

dl

= Z

∂T

F(u

h

, u

, ~n) − F (u, u

, ~n)

dl

and the result follows beause of the approximation inequality and sine the numerial ux is Lipshitz

ontinuous.

Proof of proposition 1.2. This inequalityisaonsequeneof (8)beausewehave

− Z

∇ϕ

h

(x) · f

h

(u

h

) + Z

∂Ω

ϕ

h

(x)f

h

(u

h

) · ~ndl + Z

ϕ

h

(x)S

h

(u

h

)dx =

− Z

∇ϕ

h

(x) · f (u) + Z

∂Ω

ϕ

h

(x)f (u) · ~ndl + Z

ϕ

h

(x)S

h

(u)dx

!

+ −

Z

∇ϕ

h

(x) ·

f (u) − f

h

(u

h

)

+ Z

∂Ω

ϕ

h

(x) f (u) − f

h

(u

h

)

· ~ndl + Z

ϕ

h

(x) S

h

(u) − S

h

(u

h

) dx

!

(10)

where

u

h

= π

h

(u)

. Fromstandardinterpolationresults[17℄, wehave

h

| ≤ C

and

|∇ϕ

h

| ≤ C

,

|f

h

(u

h

) − f (u)| ≤ C(u, f)h

k+1and

|S

h

(u

h

) − S(u)| ≤ C(u, S)h

k+1

.

sothat(10)isinnormsmallerthat

C(u, f, S)h

k+1

forasuitableonstant

C(u, f, S)

.

Fromlemma1.3,forany

T

and

Γ

,

T,cσ

| ≤ C(u, f, S )h

k+d and

Γ,cσ

| ≤ C(u, f, S )h

k+d−1 where

d

isthe

spaedimension.

Then,Forany

T

,

| X

σ,σ∈T

ϕ(σ) − ϕ(σ

)

Φ

Tσ

− Φ

T,cσ

| ≤ X

σ,σ∈T

|ϕ(σ) − ϕ(σ

)|

Tσ

| + |Φ

T,cσ

|

|

≤ #

ofelements

× N × ||∇ϕ||

h × C(ϕ, f, S)h

k+d

where

N

isthenumberofdegreeoffreedomineahelement. Inaregularmeshforaboundeddomain,the

numberof elementssizes like

h

−d sothat in theend, wean nda onstant (againdenoted by

C

)whih

dependson

u

,

f

,

S

and

suhthat

| X

σ,σ∈T

ϕ(σ) − ϕ(σ

)

Φ

Tσ

− Φ

T,cσ

| ≤ C(u, f, S, Ω)h

k+1

.

Thelastestimation isto bedonefortheboundaryterms. Using theonsistenyofthenumerialux,

wersthave

Z

∂Ω

F(u

h

, u

, ~n) − f

h

(u

h

, ~n)

ϕ

h

(x)dl

≤ Z

∂Ω

F(u

h

, u

, ~n) − F(u

h

, u

h

, ~n)

ϕ

h

(x)dl

≤ L Z

∂Ω

|u

h

− u

| ≤ C(u, f, ∂Ω)h

k+1

(7)

Similarly,wehave,foranyboundaryedge,

Γ,cσ

| ≤ C(u, f )h

k+d. Iftheboundaryof

isregular,thenumber

ofboundaryfaesis oftheorderof

h

−(d−1).

Thus,weget,using againthesamearguments,

X

Γ⊂∂Ω

X

σ,σ∈Γ

ϕ(σ) − ϕ(σ

)

Φ

Γσ

− Φ

Γ,cσ

≤ C(u, f, ∂Ω)h

−d+1

h

k+d

= C(u, f, ∂Ω)h

k+1

.

Thisompletestheproof.

Letusonludethisparagraphbytwoimportantremarks.

Remark1.4. Weseethat theproof usestwokey elements:

The problem (1)issteady,

One isable toompute

u

h. This isdone in pratievia aniterative algorithm beause the system (7)

isingeneralnon linear. Inallthe numerial examples, wewillonsider asimpleJaobi-likeiteration,

u

k+1σ

= u

kσ

− ω

σk

Σ((u

h

)

k

)

(11)

where

ω

σk isarelaxationparameterthatanbethoughtastheratioofatimestep(onstraintbyaCFL

ondition)andanarea. The sequene

(u

h

)

k is initializedtosomevalue (say

u

h

= 0

)andmarhedup

toonvergene. The onvergeneissueofthe sequeneisasubtleone, asitwillbeseen.

Theaurayresultwillbetrue,inpratie,providedthatoneisabletoonstrutaonvergentsequene

((u

h

)

k

)

k∈N,that is,for any

ε > 0

,oneannd

N

εsuhthat

n ≥ N

ε

,

then

|Σ((u

h

)

k

)| ≤ ε.

The algorithm anbe stoppedprovidedthat

ε = O(h

k

)

.

1.2.3 Monotoniity preservation

Inthepreviousversions ofthe RDsheme,thedegreesoffreedom were Lagrangepoints,sothat

u

hσ isthe

valueof

u

h at

σ

. Inthatase,theiterativeshemeisdesignedin suh awaythat forany

k ∈ N

,

max

σ

|u

kσ

| ≤ max

σ

max(||g||

, max

σ

|u

0σ

|).

Indeed,theshemeisdesignedsothatforany

σ

,

σ

max

∈V(σ)

|u

kσ

| ≤ max

σ∈V(σ)

|u

k−1σ

|,

where

V (σ)

isthesetofneighborsof

σ

,

σ

inluded. Notethat inthisase,wearenotasking for

||(u

h

)

k

|| ≤ C

(12)

sine it is well known that the Lagrange interpolation, for degree larger than 2, suers from the Gibbs

phenomena.

Anotherwayofthinkingispreiselyto trytoenforetheonstraint(12)globally. Assumethat wehave

ashemethatwrites:

Φ

Eσ

= X

σ∈T

c

Tσσ

(u

σ

− u

σ

)

(13a)

where

E

iseitheratriangle(aseofaninternaldegreeoffreedom)oraboundaryedge

Γ

(aseofaboundary

degreeof freedom),with

forany

σ, σ

, c

Tσσ

≥ 0.

(13b)

(8)

|u

k+1σ

| ≤ max

σ∈Vσ

|u

kσ

|

(14)

providedthat

ω

σ

≤ X

E∋σ

X

σ∈T

c

σσ

!

−1

where

E

iseither atriangleoraboundaryedge.

Ifthebasisfuntions

ψ

σ arepositiveweseethat

|(u

h

)

n+1

≤ max

σ

max(||g||

, max

σ

|u

0σ

|)

(15)

Anexampleofsuhasplit-residualisgivenbythefollowingLax-Friedrihlikeresidual: werstapprox-

imate

f (u)

by

f

h

(u

h

) := f (u

h

).

Φ

Tσ

= Φ

T

N

T

+ α

T

(u

σ

− u

T

)

(16a)

with

u

T

= P

σ∈T

u

σ

N

T

, α

T

≥ max

σ∈T

Z

T

????

(16b)

and

N

T beingthenumberofdegreesoffreedomin

T

. Thisfamilyofsplitresidualsdenesashemethatis

onlyrstorderaurate.

1.3 Constrution of high order shemes

How anwe onstruta shemethat is both monotoniity preserving andhigh order aurate. Using the

remarkontainedin Lemma1.3,onepossibilityistolook forrealnumbers

β

Eσ

(u

h

)

(

E

triangleorboundary

edge)suh that

Φ

Eσ

= β

σE

(u

h

T

,

(17)

thatareuniformlybounded. Thisensurethat

Φ

Tσ

= O(h

k+d

)

and

Φ

Γσ

= O(h

k+d−1

)

.

The question is to dene the

β

s suh that the sheme is both high order aurate and monotoniity preserving.

Arststepisthefollowing: usingamonotoniitypreservingsheme(thinkoftheLaxFriedrihssheme)

whihresidualsaredenoted by

Φ

L,Tσ whihsatises(13),weformallywrite

Φ

H,Tσ

= Φ

H,Tσ

Φ

L,Tσ

Φ

L,Tσ

= X

σ∈T

Φ

H,Tσ

Φ

L,Tσ

!

c

Lσσ

(uσ − u

σ

)

= X

σ∈T

c

Hσσ

(uσ − u

σ

)

with

c

Hσσ

=

ΦH,Tσ

ΦL,Tσ

c

Lσσ. Hene,sine

c

Lσσ

≥ 0

,wehave

c

Hσσ

≥ 0

providedthat ΦH,Tσ

ΦL,Tσ

≥ 0

. Setting

x

σ

= Φ

L,Tσ

Φ

T and

β

σ

= Φ

H,Tσ

Φ

T

,

(18)

(9)

X

σ∈T

x

σ

= X

σ∈T

β

σ

= 1

andforany

σ ∈ T, x

σ

β

σ

≥ 0.

(19)

The problem is to nd a mapping

(x

σ

)

σ∈T

7→ (β

σ

)

σ∈T that satises the onditions (19). This mapping

annotbelinearaordingtoGodunov'stheorem.

Anextensivedisussionoftheserelationsisdonein[13℄,inpartiularweprovideageometrialinterpre-

tationoftheserelations. Amongthemanymappingsthat satisfy(19),wehavehosen

β

σ

= x

+σ

P

σ∈T

x

+σ

(20)

whihisalwayswelldenedbeause

P

σ∈T

x

+σ

≥ 1

.

Unfortunately, as we see in the next setion, the resulting sheme (i.e. (7) with (17) and (20) using

the Lax Friedrihs sheme) is over ompressive. The same problem would our with other rst order

spli=residuals,forexamplethoseonstrutedformstandardrstorderux,see[10℄forsomeexamples. The

fundamental reasonisthatthe limitationisdoneaordingto monotoniitypreservingonstraintsonly, in

ompleteignoraneofwhat isthephysisoftheproblem, i.e. howup-windinghasto betriggered intothe

sheme. Hene,weneedtoaddsomedissipationmehanismwithoutdestroyingtheformalaurayinorder

toorretthatdrawbak. . Onewayofdoingthat istoaddto (17)adissipative term,namely

d

T

h

, u

h

) = |T | X

xquad

ω

quad

"

u

f (u

h

) · ∇ϕ

σ

(x

quad

)

u

f (u

h

) · ∇u

h

− S

(x

quad

)

#

(21)

suhthatthequadratiform

(v

h

, u

h

) 7→ X

σ

v

hσ

X

E∋σ

Φ

Eσ

+ X

T

θ

T

h

T

d

T

h

, u

h

)

is dissipative. Again,

E

stands for any element or edge that share

σ

. In (21),

h

T is a the radius of the

irumsribedirle/sphere,and

θ

T isaparameterthatisoftheorderof

0

indisontinuitiesand

1

elsewhere.

In(21),

x

quad anbeinterpretedasquadraturepointsand

ω

quadasweights. Saying,weinterpret(21)asa

disreteversionof

Z

T

u

f (u)ϕ

σ

·

u

f (u)∇u

h

− S(x)

dx.

However,in [20℄,wehaveshownthat,atleastforlinearux

f (u) = ~λu

,isthataneessaryonditionis

thatthequadratiform

q

K

(v

h

) := X

xquad

ω

quad

~λ · ∇v

h

(x

quad

)

2

is positive denite whenever the polynomial

λ · ∇v

h is not identially zero. In the ase of polynomial

interpolation,weneedonlyonequadraturepoint(and

ω

quad

= 1

),forquadratipolynomials,weneedthree nonalignedpoints(in pratiethevertiesoftheelement,andwetake

ω

quad

==

13,andso on. Detailsan

befoundin thisreferene,wewillusethistehniquein thepresentpaper.

Therearemanypossiblehoiesfortheparameter

θ

T. Forexample,

θ

T isagoodhoie,eveninthease

ofdisontinuoussolutionswherewehaveexperimentallynotiedthatno(visible)spuriousosillationour.

However,thebesthoiewehaveexperimentedis

θ

T

= max

σ∈T

max

T∋σ

max

σ∈T

|u

σ

− u

T

|

|u

σ

| + |u

T

| + ε

!

(22)

with

ε ≈ 10

−10. Here,

u

T

= ( P

σ∈T

u

σ

)/N

.

Références

Documents relatifs

Here, we show that following acute myocardial infarction in mice, CD8 + T lympho- cytes are recruited and activated in the ischemic heart tissue and release Granzyme B, leading

An AMSDL program is composed of three main parts (described in the following paragraphs): the first one is for the declaration of the monitored resources, the second

Références Bibliographiques Annexes.. 18 Figure 4: Diagramme de Ficus carica syconium expliquant la terminologie des fruits ..… 20 Figure 5: Différences entres les fleurs

The implementation of the numerical scheme resulted in a parallel numerical code for the discretization of the steady Euler, Navier-Stokes and RANS equations with the

Multidimensional Upwind Residual Distribution Schemes for the Euler and navier-Stokes Equations on unstructured Grids.. PhD thesis, Universit´ e Libre de Bruxelles,

Comparison of the present scheme (B2/B1) and its second order Eulerian version for the Colella case.. Abgrall, On a class of high order schemes for hyperbolic problems, in

Instead of actually considering a numerical flux along the faces of the elements, as happens in the DG or Finite Volume schemes, the approach adopted in this work consists in

In order to obtain a fully well-balanced Lagrange-Projection scheme, we exploit the idea of using a high-order well- balanced reconstruction operator for the Lagrangian