E.Canès,N.Crouseilles,H.Guillard,B.Nkonga,andE.Sonnendrüker,Editors
NUMERICAL APPROXIMATION OF KNUDSEN LAYER FOR THE
EULER-POISSON SYSTEM
Fréderique Charles
1
, Niolas Vauhelet
1
, Christophe Besse
2
,
3
, Thierry
Goudon
2
,
3
, Ingrid LaroixViolet
2
,
3
, Jean-Paul Dudon
4
and Laurent
Navoret
5
Abstrat. Inthiswork,weonsidertheomputationoftheboundaryonditionsforthelinearized
EulerPoissonderivedfromtheBGKkinetimodelinthesmallmeanfreepathregime. Boundary
layersaregeneratedfromthefatthattheinomingkinetiuxmightbefarfromthe
thermody-namialequilibrium. In[2℄, theauthors proposeamethodto omputenumeriallythe boundary
onditionsinthehydrodynamilimitrelyingonananalysisoftheboundarylayers. Inthispaper,
wewillextendthesetehniquesintheaseoftheoupledEuler-Poissonsystem.
Résumé. Dansetravail,nousnousintéressonsàl'évaluationnumériquedeonditionsauxlimites
pourlesystèmed'EulerPoissonlinéariséobtenuàpartirdumodèleinétiqueBGKdanslerégime
depetitlibre parours moyen. Des ouhes limites peuvent apparaître en raison du fait que les
uxinétiquesentrantspeuventdiérerdel'équilibrethermodynamique. Laréferene[2℄introduit
uneméthodedealulnumériquedesonditions auxlimitesdansuntel régimehydrodynamique
baséesurl'analysedesouheslimites. Ii,nousétendonsestehniquesauasdusystèmeouplé
d'EulerPoisson.
1. Introdution
Atakinetilevel,astatistialdesriptionofthedynamisofaplasmasubjettoaneletrield
E
(
t, x
)
an be obtained thanks to the Boltzmann equation governing the evolution of the partiles distribution
funtion
F
(
t, x, v
)
. Intheonedimensional framework,thisequationreads:∂
t
F
+
v∂
x
F
+
qE∂
v
F
=
1
τ
Q
(
F
)
,
(1)where
q
∈ {−
1
,
1
}
orrespondtothesignofthehargeofpartiles. Herethetimevariablet
≥
0
,thepositionvariable
x
belongs to(
−
ω, ω
)
⊂
R
and theveloityv
∈
R
. The parameterτ
is theKnudsen number;it isrelatedtothemeanfreepathandisassumedtobesmall. Inthiswork,wehoosefortheollisionoperator
Q
theBGKollisionoperator:Q
(
F
) =
M
(
ρ,u,θ
)
−
F,
wheretheMaxwellian
M
(
ρ,u,θ
)(
v
) =
ρ
√
2
πθ
exp
−
|
v
−
u
|
2
2
θ
,
(2)1
LaboratoireJaques-LouisLions-UMR7598CNRS&UniversitéPierreetMarieCurie,ParisVI
2
ProjetTeamSIMPAF,INRIALilleNordEurope
3
LaboPaulPainlevéUMR8524,CNRS&USTLille
4
ThalesAleniaSpae,Cannes
5
InstitutMathématiquedeToulouseCNRS&UniversitéPaulSabatier,Toulouse
EDPSienes,SMAI2011
andthemarosopiquantities
(
ρ, u, θ
)
aredened bydensity:
ρ
(
t, x
) =
Z
R
F
(
t, x, v
)
dv,
veloity:
ρu
=
Z
R
vF
(
t, x, v
)
dv,
temperature:
ρu
2
+
ρθ
=
Z
R
|
v
|
2
F
(
t, x, v
)
dv.
Inthispaperwerestrittoonsiderasinglespeieofhargedpartiles,thepartilesoftheoppositeharge
beingagivenbakgroundwithaxedandonstantdensity. Theeletrield
E
=
−
∂
x
V
isself-onsistentlydenedthroughthePoisson equationsatisedbythepotential
V
:−
∂
xx
V
=
q
(
ρ
−
1)
.
(3)Wereallthat
(1
, v,
|
v
|
2
)
areollisioninvariants:
Z
R
1
v
|
v
|
2
Q
(
F
)
dv
= 0
.
(4)Itis well-knownthat when
τ
→
0
,thedistribution funtionF
onvergesto aMaxwellian,whose maro-sopiparametersρ
,u
andθ
solvestheEulersystem. ThequestionraisedbythisworkisthedeterminationoftheboundaryonditionsonthisEuler systemorrespondingto theonesimposed onthekinetisystem.
Heresystem(1) (3)isompleted withthefollowingboundaryonditions:
γ
inc
F
(
t,
−
ω, v
) =
R
(
γ
out
F
(
t,
−
ω,
·
))(
v
) + Φ
data,L
(
t, v
)
,
forv >
0
,
(5)γ
inc
F
(
t, ω, v
) =
R
(
γ
out
F
(
t, ω,
·
))(
v
) + Φ
data,R
(
t, v
)
,
for
v <
0
,
(6)where
Φ
data,R
and
Φ
data,L
aregivenfuntions,and
R
isareetionoperator.ForinstaneassumingR
= 0
and
Φ
data
= 0
orrespondstoafullyabsorbingboundary. Weonsiderheretwotypesofreetionoperator:
thespeularreetionoperatorgivenby
R
(
F
)(
t, x, v
) =
αF
(
t, x,
−
v
)
,
(7)where the parameter
α
∈
[0
,
1]
represents thefration of reeted partiles, and thereetion operator oftheMaxwelldiuselawgivenby
R
(
F
)(
t,
−
ω, v
) =
α
M
w
(
v
)
Z
w
Z
v
′
<
0
|
v
′
|
F
(
t,
−
ω, v
′
)
dv
′
,
forv >
0
,
(8)where
M
w
(
v
) =
ρ
w
√
2
πθ
w
e
−|
v
|
2
/
(2
θ
w
)
,
Z
w
=
Z
v>
0
vM
w
(
v
)
dv,
with
θ
w
the temperature ofthe wall (seee.g. [15℄). Obviously, dealingwith therighthand boundary,+
ω
replaes
−
ω
andinomingpartileshavenegativeveloitiesv <
0
,wehangethesignofv
′
in(8)aordingly.
Theproblemisalsoompletedwithaninitialondition
F
(0
, x, v
) =
F
init
(
x, v
)
.
ForthePoissonequation, weimposeDirihletboundaryonditions
V
(
t,
−
ω
) =
V
L
,
V
(
t, ω
) =
V
R
.
Thisframeworkmightappearquiterudein omparisontotheomplexboundaryonditionsthatareused
alreadymakeseveraldiultiesappear;weshallgobaktoamoreambitiousmodelelsewhere.
Asthemean freepathgoesto
0
,thekinetimodelanbeapproahedbyasystemof onservationlawsforwhihthenumberofboundaryonditionsthatshouldbexeddependsonthesolutionitself(seee.g.[9℄).
Infattheboundaryonditionsforthehyperbolisystemdependonthenumberofinomingharateristis
attheboundary. Theaimofthisworkisto presentnumerialmethodstoomputetheboundaryuxesfor
thehydrodynamimodel,andto omparetheresultswithsimulationsofthekinetiequation(1).
Thepaperisorganizedasfollows. InthenextSetion,westudythesimplestasewherewelinearizethe
BoltzmannBGKequationaroundaMaxwelliansteady state. Werstintroduesomenotationsandreall
thetheoretialresultsneeded. Thenwepresentsomenumerialsimulations. InSetion3,weareonerned
with theouplingwith the Poisson equation. Weonsider boththe linearizedproblem and thenon-linear
ase.
2. Analysis for the linearized Euler system
Inthissetion,weareinterestedinthenumerialapproximationoftheboundarylayerforthelinearized
Eulersysteminthease
E
= 0
,i.e. thereisnoouplingwiththePoissonequation.2.1. Analysis of the boundary layer
Webriey reallhere thetheorialresultsdeveloppedin [9℄ (seealso [13℄). Let us assumethat
ρ
∗
>
0
,θ
∗
>
0
andu
∗
∈
R
are given. Weverifyeasily thatM
∗
:=
M
(
ρ
∗
,u
∗
,θ
∗
)
is astationary solutionof (1) withE
= 0
. Linearizing around the equilibriumM
∗
in the formF
=
M
∗
(1 +
f
)
, (1) leads to the linearizedequationfor
f
:∂
t
f
+
v∂
x
f
=
1
τ
L
∗
(
f
)
,
(9)where
L
∗
isthelinearizedBGKoperator. Morepreisely,forf
∈
L
2
(
M
∗
dv
)
wehaveL
∗
(
f
) = Π
f
−
f
,withΠ
theorthogonalprojetionofL
2
(
M
∗
dv
)
tothenite dimensional setspanned bytheollisionalinvariant{
1
, v,
|
v
|
2
}
(see[2℄):
Π
f
=
ρ
e
ρ
∗
+
v
−
u
∗
θ
∗
e
u
+
|
v
−
u
∗
|
2
θ
∗
−
1
e
θ
2
θ
∗
,
where
(
ρ,
e
u,
e
e
θ
)
aresuh that
e
ρ
e
ρu
∗
+
ρ
∗
u
e
e
ρ
(
u
2
∗
+
θ
∗
) + 2
ρ
∗
u
∗
u
e
+
ρ
∗
θ
e
=
Z
R
1
v
|
v
|
2
f M
∗
dv.
Weanverifythatthefollowingproperties 1
hold:
•
L
∗
isself-adjointfortheinner produtofL
2
(
M
∗
dv
)
,•
Ker(
L
∗
) =
Span{
1
, v,
|
v
|
2
}
,
•
Ran(
L
∗
) =
(Ker(
L
∗
))
⊥
fortheinnerprodutof
L
2
(
M
∗
dv
)
,•
wehavethedissipationpropertyZ
R
L
∗
f f M
∗
dv
≤
0
.
Equation(9)isompletedwiththeboundaryonditionsdeduedfrom(5)(6)
γ
inc
(
M
∗
f
)(
t,
−
ω, v
) =
R
(
γ
out
(
M
∗
f
)(
t,
−
ω,
·
))(
v
) +
φ
data,L
(
t, v
)
,
forv >
0
,
(10)γ
inc
(
M
∗
f
)(
t, ω, v
) =
R
(
γ
out
(
M
∗
f
)(
t, ω,
·
))(
v
) +
φ
data,R
(
t, v
)
,
forv <
0
,
(11)1
ThesepropertiesobviouslyholdforthelinearizedBGKoperatorbeauseitreduestoamereprojetion;butitisimportant
whereforinstanein thease
v >
0
,wehaveφ
data,L
= Φ
data,L
−
M
∗
+
R
(
M
∗
)
,
andwiththeinitialdata
f
(0
, x, v
) =
f
init
(
x, v
)
.
Formally,when
τ
→
0
thefuntionf
belongstoKer(
L
∗
)
andanbethereforedesribedbyaninnitesimalMaxwellian:
m
(e
ρ,
u,
e
e
θ
)(
t,x
)
(
v
) =
e
ρ
ρ
∗
+
v
−
u
∗
θ
∗
u
e
+
|
v
−
u
∗
|
2
θ
∗
−
1
e
θ
2
θ
∗
.
(12)
From(9)andthedenition ofKer
(
L
∗
)
wededuethemomentsystem:∂
t
Z
R
1
v
|
v
|
2
f M
∗
dv
+
∂
x
Z
R
v
1
v
|
v
|
2
f M
∗
dv
= 0
.
Substituting
f
bytheinnitesimalMaxwellianm
(e
ρ,
u,
e
θ
e
)
leadstothelinearizedEulersystem:∂
t
e
ρ
e
u
e
θ
+
u
∗
ρ
∗
0
θ
∗
ρ
∗
u
∗
1
0
2
θ
∗
u
∗
∂
x
e
ρ
e
u
e
θ
= 0
.
(13)Thesystem(13)writesinmatrixform
∂
t
U
e
+
A∂
x
U
e
= 0
,
(14)with
A
=
u
∗
ρ
∗
0
θ
∗
ρ
∗
u
∗
1
0
2
θ
∗
u
∗
.
Thissystemisobviouslyhyperboli: thespetrumofthematrix
A
is{
u
∗
−
√
3
θ
∗
, u
∗
, u
∗
+
√
3
θ
∗
}
. Therefore,thenumberofboundaryonditionsthat should bexed at theleft(resp. right)boundarydependsonthe
numberofpositive(resp. negative) eigenvalues. Moreover,weandenethefollowingquadratiform
Q
:
U
e
= (
ρ,
e
u,
e
θ
e
)
7→
Z
R
v
|
m
(e
ρ,
u,
e
θ
e
)
|
2
M
∗
dv.
(15)Following[2,9℄,weansplitthesetofinnitesimalMaxwelliansaordingtothesignofthequadratiform
Q
:Ker(
L
∗
) = Λ
+
⊕
Λ
−
⊕
Λ
0
,
where
Λ
±
orresponds to theeigenspaes assoiatedto the positive (resp. negative) eigenvaluesof
A
, andΛ
0
istheeigenspae assoiatedto 0whenit belongto thespetrumof
A
. WeshalldenoteI
±
= dim(Λ
±
)
thenumberofpositive(resp. negative)eigenvaluesof
A
.Then, onsidering theleft hand boundary
x
=
−
ω
, itis possibleto splitm
(
ρ,
e
u,
e
e
θ
)
intom
−
∈
Λ
−
(whih
orresponds to theoutgoingpart) and
m
+
∈
Λ
+
⊕
Λ
0
(whih orresponds to theinoming part at the
boundary
x
=
−
ω
):m
(e
ρ,
u,
e
θ
e
)
=
m
−
+
m
+
.
(16)Dealingwith therighthand boundary
x
= +
ω
, therole ofΛ
+
and
Λ
−
is inverted. Ateahboundarythe
outgoingpartisgivenbytheow,whereastheinomingparthastobeimposedasaboundaryonditionto
ompletetheEulersystem.
Theinominguxattheboundariesaredeterminedthankstoaboundarylayeranalysis. Letusonsider
thefollowinghalfspaeproblem
v∂
z
G
=
L
∗
G,
z >
0
, v
∈
R
where
Υ
data
hastobesuitablydened. Wereallthefollowingstatement(see [9℄):
Theorem1. Thereexistsalinearmapping(thatisusuallyalledthe generalizedChandrasekharfuntional)
C
∗
:
L
2
(
R
,
(1 +
|
v
|
)
M
∗
(
v
)
dv
)
→
Λ
+
⊕
Λ
0
Υ
data
7→
m
∞
,
where
m
∞
isthelimitasz
→ ∞
oftheuniquesolutionG
of (17). Moreover,G
∈
L
∞
(0
,
∞
;
L
2
(
R
, M
∗
(
v
)
dv
))
.Letusassumethat
f
expandsasfollowsf
=
m
(e
ρ,
u,
e
e
θ
)
+
G
L
t,
x
+
ω
τ
, v
+
G
R
t,
ω
−
x
τ
, v
+
o(τ
),
where
G
L
and
G
R
standforboundarylayersand aredened asfollows:
•
G
L
(
t, z, v
) =
G
(
t, z, v
)
with
G
thesolutionof (17)withinomingdataΥ
data
(
t, v
) =
γ
inc
f
(
t,
−
ω, v
)
−
m
(e
ρ,
u,
e
e
θ
)
(
t,
−
ω, v
)
,
(18)andimposing that
lim
z
→
+
∞
G
(
z, v
) = 0
,
(19)
•
G
R
(
t, z, v
) =
G
(
t, z,
−
v
)
with
G
thesolutionof (17)withinomingdataΥ
data
(
t, v
) =
γ
inc
f
(
t, ω,
−
v
)
−
m
(
ρ,
e
u,
e
e
θ
)
(
t, ω,
−
v
)
.
(20)andondition(19).
Condition (19)expressesthat the boundarylayeris expeted to vanish farfrom theboundary. Using the
deompositionon
Ker(
L
∗
)
in(16),itimpliesthattheunknownm
+
satisesC
∗
(
γ
inc
f
(
t, ω,
·
)
−
m
−
(
t, ω,
·
)) =
C
∗
(
m
+(
t, ω,
·
))
,
C
∗
(
γ
inc
f
(
t,
−
ω,
·
)
−
m
−
(
t,
−
ω,
·
)) =
C
∗
(
m
+(
t,
−
ω,
·
))
.
(21)
The questionwe addressis onernedwith the numerialapproximationof theoutgoingstate
m
+
, whih
arisesinthedenitionoftheboundaryuxes.
2.2. Numerial resolution of the linearized Euler system
2.2.1. Numerialmethod
Thematrix
A
of thesystem(13)anbediagonalizedunder theformA
=
u
∗
ρ
∗
0
θ
∗
ρ
∗
u
∗
1
0
2
θ
∗
u
∗
=
P
u
∗
0
0
0
u
∗
+
√
3
θ
∗
0
0
0
u
∗
−
√
3
θ
∗
P
−
1
,
with
P
=
1
2
√
θ
∗
2
√
θ
∗
ρ
∗
√
3
ρ
∗
−
√
3
ρ
∗
0
3
√
θ
∗
3
√
θ
∗
−
2
√
θ
∗
θ
∗
2
θ
∗
√
3
−
2
θ
∗
√
3
,
P
−
1
=
1
3
ρ
∗
√
3
θ
∗
2
√
3
θ
∗
0
−
ρ
∗
√
3
/
√
θ
∗
θ
∗
ρ
∗
√
3
θ
∗
ρ
∗
−
θ
∗
ρ
∗
√
3
θ
∗
−
ρ
∗
.
Weset
˜
ρ
diag
˜
u
diag
˜
θ
diag
=
P
−
1
˜
ρ
˜
u
˜
θ
thevariablesinthenewbasisdenedbythetransitionmatrix
P
,andweobtainthreeindependantequationson
ρ
˜
diag
,˜
u
diag
,θ
˜
diag
:
∂
t
ρ
˜
diag
+
u
∗
∂
x
ρ
˜
diag
= 0
,
(a)∂
t
u
˜
diag
+ (
u
∗
+
√
3
θ
∗
)
∂
x
u
˜
diag
= 0
,
(b)∂
t
θ
˜
diag
+ (
u
∗
−
√
3
θ
∗
)
∂
x
θ
˜
diag
= 0
.
()(22)
Equations(22)should beompleted withboundaryonditions,whihdependonthesignof
u
∗
,u
∗
+
√
3
θ
∗
and
u
∗
−
√
3
θ
∗
. More preisely, we denoteU
bd,l
= (
ρ
bd,l
, u
bd,l
, θ
bd,l
)
andU
bd,r
= (
ρ
bd,r
, u
bd,r
, θ
bd,r
)
themarosopiquantitieswhihhavetobedenedat theboundaries
x
=
−
ω
andx
=
ω
respetively,andwe setU
bd,l
=
U
1
bd,l
U
2
bd,l
U
3
bd,l
=
P
−
1
U
bd,l
,
U
bd,r
=
U
1
bd,r
U
2
bd,r
U
3
bd,r
=
P
−
1
U
bd,r
.
Boundaryonditionsofequations(22)are:
˜
ρ
diag
(
−
ω
) =
U
bd,l
1
ifu
∗
>
0
orρ
˜
diag
(
ω
) =
U
1
bd,r
ifu
∗
<
0
,
(a)˜
u
diag
(
−
ω
) =
U
bd,l
2
ifu
∗
+
p
3
θ
∗
>
0
oru
˜
diag
(
ω
) =
U
2
bd,r
ifu
∗
+
p
3
θ
∗
<
0
,
(b)˜
θ
diag
(
−
ω
) =
U
bd,l
3
ifu
∗
−
p
3
θ
∗
>
0
orθ
˜
diag
(
ω
) =
U
3
bd,r
ifu
∗
−
p
3
θ
∗
<
0
.
()(23)
We then introdue aregular subdivision
{
x
0
, . . . , x
I
+1
}
of the domain[
−
ω, ω
]
, withx
i
=
−
ω
+
i
∆
x
and∆
x
= 2
ω/
(
I
+ 1)
,andatimedisretisation∆
t
. Equations(22)aresolvednumeriallythanksto anupwindsheme. Weobtainforinstane,thanksto equation(22)-(a)andboundaryondition(23)-(a),thefollowing
approximations
ρ
˜
n
diag,i
ofρ
˜
diag
(
n
∆
t, x
i
)
:
˜
ρ
n
diag,
+1
0
=
U
bd,l
1
,n
+1
,
˜
ρ
n
diag,
+1
1
= ˜
ρ
n
diag,
1
−
u
∗
∆
t
∆
x
˜
ρ
n
diag,
1
− U
1
,n
bd,l
,
˜
ρ
n
diag,i
+1
= ˜
ρ
n
diag,i
−
u
∗
∆
t
∆
x
(˜
ρ
n
diag,i
−
ρ
˜
n
diag,i
−
1
)
fori
∈ {
2
, I
+ 1
}
,
if
u
∗
>
0
,
and
˜
ρ
n
+1
diag,i
= ˜
ρ
n
diag,i
−
u
∗
∆
t
∆
x
(˜
ρ
n
diag,i
+1
−
ρ
˜
n
diag,i
)
fori
∈ {
0
, I
−
1
}
,
˜
ρ
n
diag,I
+1
= ˜
ρ
n
diag,I
−
u
∗
∆
t
∆
x
U
bd,r
1
,n
−
ρ
˜
n
diag,I
,
˜
ρ
n
diag,I
+1
+1
=
U
bd,r
1
,n
+1
,
if
u
∗
<
0
,
where
U
1
,n
bd,l
=
U
bd,l
1
(
n
∆
t
)
andU
1
,n
bd,r
=
U
bd,r
1
(
n
∆
t
)
. Thesevaluesareomputedbysolving(21). Themethodtosolvetheseequationsisexplainedinthenextsubsetion. Wedenoteinthesequel
˜
U
i
n
=
˜
ρ
n
i
˜
u
n
i
˜
θ
n
i
=
P
˜
ρ
n
diag,i
˜
u
n
diag,i
˜
θ
n
diag,i
.
(24)2.2.2. Treatmentofthe boundary onditions
We explain in this setion the method presented in [2℄ to ompute, at eah time step, the
hydrody-namiboundaryonditions
U
n
bd,l
andU
n
bd,r
fromthe knowledgeof thehydrodynamiquantityU
˜
n
i
neartheboundaries
x
=
±
ω
andtheknowledgeofφ
data,L
,
φ
data,R
Wefousherethedisussionontheboundary
x
=
−
ω
(thetreatmentoftheboundaryx
=
ω
issimilar).Weintrodue,at eahtime
t
n
=
n
∆
t
, theinnitesimalMaxwellian
m
n
bd,l
(
v
) =
ρ
n
bd,l
ρ
∗
+
u
n
bd,l
v
−
u
∗
θ
∗
+
θ
n
bd,l
2
θ
∗
(
v
−
u
∗
)
2
θ
∗
−
1
andweassoiateto
m
n
bd,l
itsdeompositiononKer
(
L
∗
)
asin(16):m
n
bd,l
=
m
n
+
+
m
n
−
.
(25)Theoutgoingpart
m
n
−
∈
Λ
−
isgivenbytheprojetionofU
˜
n
0
onΛ
−
:
m
n
−
(
v
) =
X
k
∈
I
−
α
n
k
χ
k
(
v
)
,
α
n
k
=
R
R
v m
U
˜
n
0
χ
k
(
v
)
M
ρ
∗
,u
∗
,θ
∗
(
v
)
dv
R
R
v
|
χ
k
(
v
)
|
2
M
ρ
∗
,u
∗
,θ
∗
(
v
)
dv
,
where
m
˜
U
n
0
isdened in(12). Wedenote
(
χ
0
, χ
1
, χ
2)
anorthogonalbasisofKer
(
L
∗
)
whih isheredenedby(see[2℄)
χ
0(
v
) =
√
1
6
|
v
−
u
∗
|
2
θ
∗
−
3
,
χ
1(
v
) =
√
1
6
√
3
v
√
−
u
∗
θ
∗
+
|
v
−
u
∗
|
2
θ
∗
,
χ
2(
v
) =
√
1
6
√
3
v
√
−
u
∗
θ
∗
−
|
v
−
u
∗
|
2
θ
∗
,
and
I
+
=
k
∈ {
0
,
1
,
2
}
, χ
k
∈
Λ
+
,
I
−
=
k
∈ {
0
,
1
,
2
}
, χ
k
∈
Λ
−
,
I
0
=
k
∈ {
0
,
1
,
2
}
, χ
k
∈
Λ
0
.
Moreover,
Q
(
χ
0) =
ρ
∗
u
∗
,
Q
(
χ
1) =
ρ
∗
(
u
∗
+
p
3
θ
∗
)
,
Q
(
χ
2) =
ρ
∗
(
u
∗
−
p
3
θ
∗
)
,
where
Q
is the quadrati form dened in (15). Consequently, the sign of the eigenvaluesu
∗
−
√
3
θ
∗
,u
∗
,u
∗
+
√
3
θ
∗
allowsustoknowI
−
andthereforetoompute
m
n
−
.Then we need to ompute
m
n
+
. We rst notie that integrating the half spae problem (17) over theveloityvariablegives
d
dz
Z
R
v
1
v
|
v
|
2
G
L
(
z, v
)
M
∗
(
v
)
dv
= 0
,
andthen
Z
R
v
1
v
|
v
|
2
G
L
(0
, v
)
M
∗
(
v
)
dv
=
Z
R
v
1
v
|
v
|
2
G
L
(
∞
, v
)
M
∗
(
v
)
dv
= 0
.
(26)Moreover,we make anapproximationproposed by Maxwell in [19℄ (see also [1,18℄): weassume that the
outgoingdistributionattheboundary
γ
out
G
L
(0
,
·
)
oinides withthedistribution atinnity,that istosayγ
out
G
(0
, v
) = lim
z
→
+
∞
G
(
z, v
) = 0
,
v >
0
.
(27)
Then,using(26),(20)and(27),weobtain
Z
v>
0
v
1
v
|
v
|
2
γ
inc
(
f
)
−
m
n
bd,l
Z
v>
0
v
1
v
|
v
|
2
φ
data,L
(
t
n
, v
) +
R
(
M
∗
m
−
n
)(
v
)
−
m
n
−
(
v
)
M
∗
(
v
)
dv
=
Z
v>
0
v
1
v
|
v
|
2
m
n
+
(
v
)
M
∗
(
v
)
− R
(
M
∗
m
n
+
)(
v
)
dv.
(29)
Relation (29) allows us to ompute
m
n
+
. We notie that we havethree equations form
n
+
to be satisedwhereasdim
(Λ
+
⊕
Λ
0
)
∈ {
0
,
1
,
2
,
3
}
. Inthease
u
∗
= 0
i.e. when thesignatureofthequadratiformQ
is(1
,
1)
,weusethefatthat wehavetheadditionalonservationlaw(see[2,14℄)d
dz
Z
R
v
2
χ
0(
v
)
G
(
z, v
)
M
(
ρ
∗
,
0
,θ
∗
)(
v
)
dv
= 0
.
Amethodtosolvethisoverdeterminatedproblemistopikthenumberofequationsneededin(29)(see[14℄).
Here wewill follow insteadthe method in [2℄ where thedetermination of
m
n
+
relies onthe resolution ofaleastsquaresproblemunderonstraints. Wereferto[2℄forfurther details.
2.3. Numerial omparisons
Theequation (9) is solvedthanksto asplitting sheme : wesolvesuessivelythe freetransport
equa-tion
∂
t
f
+
v∂
x
f
= 0
(with upwind sheme) and the ollision part∂
t
f
=
1
τ
L
∗
(
f
)
. We notie that thislast equation anbe solvedexatlysine byonservation properties, asolution
f
of thisequation satisesd
dt
R
R
(1
, v, v
2
)
f
(
v
)
dv
= 0
,andtherefore
Π
f
isnotmodiedduringthisstepandanbeonsideredasagivensoureterm. Wereferto[2℄forfurtherdetails. Inthesimulationpresentedbelow,wetake
∆
v
= 5
·
10
−
4
and
τ
= 10
−
3
,andforbothhydrodynamiandkinetisimulation
ω
= 0
.
5
and∆
x
= 1
/
500
(inonvenientunits).Thenaltimeis
t
f
= 0
.
1
andthetimestep∆
t
isdeterminedbytheCFLondition. Wethenomparethenumerialsolutionofthelinearized Boltzmannequation(9)withthenumerialsolutionof linearizedEuler
system(14).
OnFigure1,weonsidertheequilibriumstategivenby
(
ρ
∗
, u
∗
, θ
∗
) = (1
,
−
0
.
1
,
1)
,
with speularreetion at boundaries(operator
R
is given by (7)), areetionoeient ofα
= 0
.
1
andnosoureterm (
φ
data,L
=
φ
data,R
= 0
). Inpartiular, thesignature of
Q
is(1
,
2)
. Inthis ase,resultsaresatisfatory : the density, marosopi veloity and temperature given by theresolution of the linearized
Eulersystemareverylosefrom thosegivenbythelinearizedBoltzmann equation.
Intheseond exampleonFigure2weonsider theequilibrium stategivenby
(
ρ
∗
, u
∗
, θ
∗
) = (1
,
0
,
1)
,
withdiuse reetionatboundaries(operator
R
isgivenby(7)),areetionoeientofα
= 0
.
8
andno soureterm(φ
data,L
=
φ
data,R
= 0
). Thisis adegenerate ase: the signatureof
Q
ishere(1
,
1)
. Thus asnotiedin theprevioussetion,wehavefouronservationrelationsandtwoomponentstobendinorder
todene
m
n
+
.Finallyin thelastexampleongure3wepresenttheaseofnoreetionat boundarieswithnosoure
term(
φ
data,L
=
φ
data,R
= 0
). Wehoosethe signature
(3
,
0)
forQ
. Againin this asetheapproximation−0.5
−0.4
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
0.5
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
Density
hydrodynamic
kinetic
−0.5
−0.4
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
0.5
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
Macroscopic velocity
hydrodynamic
kinetic
−0.5
−0.4
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
0.5
0.4
0.5
0.6
0.7
0.8
0.9
1
Temperature
hydrodynamic
kinetic
Figure 1. Comparison betweenthe kinetisimulation and the hydrodynami simulation
at
t
f
= 0
.
1
s,for(
ρ
∗
, u
∗
, θ
∗
) = (1
,
−
0
.
1
,
1)
,speularreetionatboundariesandα
= 0
.
1
.3. Coupling with the Poisson equation
3.1. Analysis of the linearized VlasovPoisson system
Inthis setion, weonsider the linearizationof the oupledBoltzmannPoisson system(1)(3). Let us
assumethat
ρ
∗
>
0
andθ
∗
>
0
aregiven. WesetM
∗
(
x, v
) =
ρ
∗
e
−
v
2
2
θ
∗
√
2
πθ
∗
e
−
qV
∗
(
x
)
/θ
∗
R
ω
−
ω
e
−
qV
∗
(
y
)
/θ
∗
dy
=
M
(
ρ
∗
,
0
,θ
∗
)(
v
)
e
−
qV
∗
(
x
)
/θ
∗
R
ω
−
ω
e
−
qV
∗
(
y
)
/θ
∗
dy
.
(30)Weverifyeasilythat forsuhaMaxwellian,wehave
v∂
x
M
∗
−
q∂
x
V
∗
∂
v
M
∗
= 0
.
(31)Wehoosefor
V
∗
thesolutionoftheproblem
−
∂
xx
V
∗
=
q
ρ
∗
e
−
qV
∗
(
x
)
/θ
∗
R
ω
−
ω
e
−
qV
∗
(
y
)
/θ
∗
dy
−
1
!
,
V
∗
(
−
ω
) =
V
L
,
V
∗
(
ω
) =
V
R
.
(32)
Lemma 1. Assume
V
L
and
V
R
are given in
R
andρ
∗
>
0
,θ
∗
>
0
. Then there existsa unique solution−0.5
−0.4
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
0.5
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
1.6
Density
hydrodynamic
kinetic
−0.5
−0.4
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
0.5
−0.25
−0.2
−0.15
−0.1
−0.05
0
0.05
0.1
0.15
0.2
0.25
Macroscopic velocity
hydrodynamic
kinetic
−0.5
−0.4
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
0.5
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Temperature
hydrodynamic
kinetic
Figure 2. Comparison betweenthe kinetisimulation and the hydrodynami simulation
at
t
f
= 0
.
1
s,for(
ρ
∗
, u
∗
, θ
∗
) = (1
,
0
,
1)
,diusereetionatboundariesandα
= 0
.
8
.Proof. Letus dene
V
b
∈
C
2
([
−
ω, ω
])
an extensionoftheboundaryondition:
V
b
(
−
ω
) =
V
L
and
V
b
(
ω
) =
V
R
. This results relies on the remark that a weak solution of (32)is a ritial point in the ane spae
V
b
+
H
0
1
(
−
ω, ω
)
ofthefuntionalJ
(
V
) =
1
2
Z
ω
−
ω
|
∂
x
V
|
2
dx
−
ρ
∗
θ
∗
ln
Z
ω
−
ω
e
−
qV /θ
∗
dx
−
Z
ω
−
ω
qV dx.
Therefore,itremainstoprovethatthisfuntionaladmitsauniqueminimizer. Infat,
J
islearlyontinuous andonvexonH
1
(
−
ω, ω
)
andwehavetheinequality
ln
Z
ω
−
ω
e
−
qV /θ
∗
dx
≤
k
V
θ
k
∞
∗
+ ln(2
ω
)
≤
C
k
V
k
H
1
+ ln(2
ω
)
.
Hene,applyingthePoinaréinequality,weget
J
(
V
)
≥ k
V
k
2
H
1
−
C
0
ρ
∗
k
V
k
H
1
−
C
1
.
Then
J
is oerive and bounded from below onH
1
(
−
ω, ω
)
. It admits aunique minimizer
V
∗
. Applyingstandardtehniques,weanshowthat
V
∗
∈
C
2
(
−
ω, ω
)
.
Thus,thepair
(
M
∗
, V
∗
)
isastationarysolutionoftheBoltzmann-Poissonproblem(1)(3). Welinearize−0.5
−0.4
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
0.5
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
1.2
Density
hydrodynamic
kinetic
−0.5
0
−0.4
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
0.5
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Macroscopic velocity
hydrodynamic
kinetic
−0.5
−0.4
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
0.5
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
Temperature
hydrodynamic
kinetic
Figure 3. Comparison betweenthe kinetisimulation and the hydrodynami simulation
at
t
f
= 0
.
1
s,for(
ρ
∗
, u
∗
, θ
∗
) = (1
,
1
,
0
.
25)
,withnoreetionandnosouretermatbound-aries.
problem:
∂
t
f
+
v∂
x
f
−
q∂
x
V
∗
∂
v
f
+
q
v
θ
∗
∂
x
V
e
=
1
τ
L
∗
f.
(33)The linearized ollisionoperator
L
∗
has been dened in the previoussetion. Equation(33)is oupled tothePoisson equation
−
∂
xx
V
e
=
q
Z
R
f
M
∗
dv,
V
e
(
−
ω
) =
V
e
(
ω
) = 0
.
(34)Thus, as
τ
goes to0
thefuntionf
looks likean innitesimal Maxwellianm
(e
ρ,
e
u,
e
θ
)
. The properties ofL
∗
desribedaboveyield
Z
R
L
∗
f
1
v
|
v
|
2
M
(
ρ
∗
,
0
,θ
∗
)
dv
= 0
.
Thereforetakingthemomentsof (33),weareledtothefollowingsystem:
∂
t
Z
R
1
v
|
v
|
2
f M
(
ρ
∗
,
0
,θ
∗
)
dv
+
∂
x
Z
R
v
1
v
|
v
|
2
f M
(
ρ
∗
,
0
,θ
∗
)
dv
−
q∂
x
V
∗
Z
R
v
1
|
v
|
2
∂
v
f M
(
ρ
∗
,
0
,θ
∗
)
dv
+
q
∂
x
V
e
θ
∗
Z
R
v
1
v
|
v
|
2
Z
R
1
v
|
v
|
2
∂
v
f M
(
ρ
∗
,
0
,θ
∗
)
dv
=
Z
R
v
θ
∗
1
v
|
v
|
2
f M
(
ρ
∗
,
0
,θ
∗
)
dv
−
Z
R
0
1
2
v
f M
(
ρ
∗
,
0
,θ
∗
)
dv
Substituting
f
bym
(
ρ,
e
e
u,
θ
e
)
,weusetheidentitiesZ
R
m
(e
ρ,
e
u,
θ
e
)
M
(
ρ
∗
,
0
,θ
∗
)
dv
=
ρ,
e
Z
R
vm
(e
ρ,
u,
e
θ
e
)
M
(
ρ
∗
,
0
,θ
∗
)
dv
=
ρ
∗
u
e
;
Z
R
v
2
m
(e
ρ,
u,
e
e
θ
)
M
(
ρ
∗
,
0
,θ
∗
)
dv
=
ρ
∗
θ
e
+
e
ρθ
∗
,
Z
R
v
3
m
(e
ρ,
u,
e
e
θ
)
M
(
ρ
∗
,
0
,θ
∗
)
dv
= 3
ρ
∗
θ
∗
u.
e
Wededuethat
∂
t
e
ρ
e
u
e
θ
+
0
ρ
∗
0
θ
∗
ρ
∗
0
1
0
2
θ
∗
0
∂
x
e
ρ
e
u
e
θ
=
q∂
x
V
∗
ρ
∗
e
u
θ
∗
e
θ
θ
∗
0
−
q∂
x
V
e
0
1
0
(35)holds. From(34),thissystemisoupledwiththelinearizedPoissonequation
−
∂
xx
V
e
=
q
ρ
e
e
−
qV
∗
(
x
)
/θ
∗
R
ω
−
ω
e
−
qV
∗
(
y
)
/θ
∗
dy
,
V
e
(
−
ω
) =
V
e
(
ω
) = 0
.
(36)Likeforthefreeproblem,weassumethat
f
admitsanexpansionoftheformf
=
m
(e
ρ,
e
u,
θ
e
)
+
G
L
(
t,
x
+
ω
τ
, v
) +
G
R
(
t,
ω
−
x
τ
, v
) +
o(
τ
)
.
Thentheboundarylayers
G
L
and
G
R
satisesthesamehalfspaeproblem(17). Thereforethedetermination
oftheboundaryonditionthatwehavetoimposetosolve(35)followsthetreatmentpresentedintheprevious
setion. Wenotiemoreoverthatweareneessaryinthedegenerateasewherethequadratiformassoiated
tothematrixofthehyperboliequationhasthesignature
(1
,
1)
,i.e.u
∗
= 0
. Infat,ifwehoosenon-zerou
∗
,therelation(31)isnotsatisedandthereforeM
∗
isnotastationarysolutionoftheBoltzmannequation.WepresentsomeomparisonsbetweenthesimulationofthelinearizedBoltzmann-Poisson equationand
thelinearizedEuler-Poisson system. Inthenumerialproedures weuse,system(32)issolvedthankstoa
xedpointiterativeproedure. Sine
ρ
∗
andθ
∗
arexed,thepotentialV
∗
is omputedonlyoneforallatthebeginningofthesimulation. ThelinearizedBoltzmannequation(33)issolvedwiththesameideathan
intheprevioussetion: wesplit bysolvingrstthetransportequationononetimestep
∂
t
f
+
v∂
x
f
−
q∂
x
V
∗
∂
v
f
+
q
v
θ
∗
∂
x
V
e
= 0
,
withanupwindsheme,thenwesolveononetime step
∂
t
f
=
1
τ
L
∗
(
f
)
.
The linearized Euler system(35)is disretized by asplitting method: rstwesolve (35)with zeroat the
righthandsideasdesribedinsubsetion2.2.1and thenwesolveimpliitely
∂
t
e
ρ
e
u
e
θ
=
q∂
x
V
∗
ρ
∗
u
e
θ
∗
e
θ
θ
∗
0
−
q∂
x
V
e
0
1
0
.
ThePoissonequationisdisretizedthankstoanitedierenemethod. Figures4and5presenttheresults
foralinearizationaroundtheequilibriumstategivenby
(
ρ
∗
, u
∗
, θ
∗
) = (1
,
0
,
0
.
5)
, withasouretermatthe−0.5
−0.4
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
0.5
0.9
1
1.1
1.2
1.3
1.4
1.5
Density
hydrodynamic
kinetic
−0.5
−0.4
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
0.5
−0.05
0
0.05
0.1
0.15
0.2
Macroscopic velocity
hydrodynamic
kinetic
−0.5
−0.4
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
0.5
0.48
0.5
0.52
0.54
0.56
0.58
0.6
0.62
0.64
0.66
0.68
Temperature
hydrodynamic
kinetic
−0.5
−0.4
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
0.5
0.86
0.88
0.9
0.92
0.94
0.96
0.98
1
Potential
Figure 4. Comparison betweenthe kinetisimulation and the hydrodynami simulation
fortheoupledproblem with
(
ρ
∗
, u
∗
, θ
∗
) = (1
,
0
,
0
.
5)
,noreetionat boundaries(R
= 0
),φ
data,L
=
m
(2
/
1
.
2
,
0
,
1
.
2
/
2)
,φ
data,R
=
m
(
ρ
∗
,
0
,θ
∗
)
,q
=
−
1
,V
L
=
V
R
= 1
at
t
f
= 0
.
1
.3.2. The non-linearEulerPoisson system
Finally,weomebaktothenon-linearproblem(1)(3)presentedin theintrodution. Asthemeanfree
pathgoesto
0
,wegettheEuler-Poissonsystem:
∂
t
ρ
+
∂
x
(
ρu
) = 0
,
∂
t
(
ρu
) +
∂
x
(
ρu
2
+
ρθ
) =
−
qρ∂
x
V,
∂
t
ρθ
+
ρu
2
2
+
∂
x
3
2
ρuθ
+
1
2
ρu
3
=
−
qρu∂
x
V,
(37)
oupledto
−
∂
xx
V
=
q
(
ρ
−
1)
.
(38)Wepresentarstattempttotreatnumeriallytheinitial-boundary-valueproblem,inorporatingthe
ompu-tationofrelevantboundaryonditionsforthehydrodynamis,designedfollowingtheapproahusedforthe
neutralproblemandthelinearized equation. Clearlynonlinearitiesandouplinggiverisetonewdiulties