A
fluid‐particle
system
related
to
Vlasov‐Navier‐Stokes
equations
Franco Flandoli
(Universita
diPisa)
Abstract
Preliminaryresultsonthe convergencetotheVlasov‐Navier‐Stokesequationsofa
systemofparticles interactingwithafluid areannounced. Mainemphasisisgivento
thedifficulties that arise and hints forsolutionsaregiven.
1
Introduction
Theaim of this workis to
investigate
afluid‐particle
systemwhichseemsto converge, inthe hmit of
infinitely
manyparticles,
toaVlasov‐Navier‐Stokes(VNS)
system. We restrictthis
preliminary investigation mainly
to dimension d=2 and\mathrm{W}shall assumeto beon atorus
$\Gamma$^{2}
withperiodic boundary
conditions. The facts described in thisnotehaveonly
thecharacter ofa
preliminary investigation
andannouncement ofpartial
results.Let $\epsilon$\in
(0,1)
and N\in \mathrm{N} begiven,
where N is the number ofparticles;
consider thesystem:
\displaystyle \frac{\partial u^{N}}{\partial t}= $\Delta$ u^{N}-u^{N}
.\displaystyle \nabla u^{N}-\nabla$\pi$^{N}-\frac{c_{0}}{N}\sum_{i=1}^{N}(u_{ $\epsilon$}^{N}(t, X_{t}^{i})-V_{t}^{i})$\delta$_{X_{\mathrm{t}}^{i}}^{ $\epsilon$}
\displaystyle \frac{d}{dt}X_{t}^{i}=V_{t}^{i}
\displaystyle \frac{1}{N}dV_{t}^{i}=\frac{c_{0}}{N}(u_{ $\epsilon$}^{N}(t, X_{\mathrm{t}}^{i})-V_{t}^{i})dt+\frac{1}{N^{2}}\sum_{j=1}^{N}K(X_{t}^{i}-X_{t}^{j})dt+\frac{$\sigma$_{p}}{N}dW_{t}^{i}.
The first
equation
isthe usualNavier‐Stokessystemfor thevelocity
andpressure(u^{N}, $\pi$^{N})
ofa
fluid,
forcedby
thepresenceof Nparticles;
aprecise
description
of the interactionbetween
particles
and fluid is adifficulttopic
(just
as aninstance,
see[2], [5],
[8], [9], [16]),
outside the scopeof this
preliminary
note, hence weadopt
apartially phenomenological
description,
whereparticles
act as delta Diracforces,
withintensity
proportional
to thevelocity
difference between fluid andparticle.
For technicalreasons, but alsoasatrace offorce;
andanalogously
thevelocity
differenceiscomputed
between theparticle velocity
V_{\mathrm{t}}^{i}
andalocalaverage atparticle
centerX_{t}^{i}
of the fluidvelocity,
u_{ $\epsilon$}^{N}(t, X_{t}^{i})
.The
smoothings
usedinthe firstequation
above aregiven by
classical mollifiers of theform
$\theta$_{ $\epsilon$}^{0}(x)=$\epsilon$^{-d}$\theta$^{0}($\epsilon$^{-1}x)
,where$\theta$^{0}
isasmoothprObability density
withcompact supportwhich includesa
neighbor
of theorigin,
andaredefinedas$\delta$_{X_{t}^{i}}^{ $\epsilon$}(x)=($\theta$_{ $\epsilon$}^{0}*$\delta$_{X_{t}^{i}})(x)=$\theta$_{ $\epsilon$}^{0}(x-X_{t}^{i}) , u_{ $\epsilon$}^{N}=$\theta$_{ $\epsilon$}^{0}*u^{N}.
The lasttwo
equations
of thesystemabove describe the Newtoniandynamics
ofparti‐
cles andweassumethevelocity
V_{t}^{i}
satisfies astochasticdifferentialequation
drivenby
theBrownianmotion
W_{t}^{i}
in\mathbb{R}^{d}
; the BrownianmotionsW_{t}^{i},
i=1,
N areindependent
anddefined on a
probability
space( $\Omega$, \mathcal{F}, P)
.Weassumethe
particles
havemass\displaystyle \frac{1}{N}
; theforceacting
onparticle
ihas threecompo‐nents: the Stokes
drug
force duetothefluid,
aninteractionforcegiven by
the interactionkernel K andanoise
perturbation.
Remark 1 Recall that Stokes
drag force
isgiven
by
6 $\pi$ r $\mu$ \mathrm{v}
wherer isparticle radius,
\mathrm{v} isthe relative
velocity
of
particle
and $\mu$ isviscosity.
Hence theinterpretation
of
thescalings
in N chosen above is: the
particle
mass isof
order\displaystyle \frac{1}{N}
; theparticle
radius isof
order\displaystyle \frac{1}{N}
and
c_{0}\sim 6 $\pi \mu$
. Particles with a massdensity
similar to thefluid
should have massof
theorder
\displaystyle \frac{1}{N^{\mathrm{d}}}
, while here we assumeitof
order\displaystyle \frac{1}{N}
, muchbigger.
Thiscorresponds
to aregime
of
sparseheavy partides.
Example
2 Theinteractionkernelisusually
absentinclassicalformulations
of
VNSsys‐ tem. We include it here since it may beinteresting
in someapplications.
Forinstance,
thinktometastaticcancercells
flowing
inthe bloodstream, anexample
where the conditionof
sparseheavy
particles
may be realistic. These cells do notonly
interact with thefluid
but also between themselves and
possibly
with otherspecial
cells.Example
3Having
in mindapplications
tobiological fluids,
aninteresting
variationscould betointroduceadeath‐rate
of
theform
$\lambda$_{t}^{i}=g(u_{ $\epsilon$}^{N}(t,X_{t}^{i})-V_{t}^{i})
, motivatedby
thefact
thatstress mayinduce cell death. Such additionaltermleadtothe
term-g(u(t,x)-v)F(t, x, v)
inthe limit. PDE.
Aswesaid
above,
ouraimisproving
convergence toaVlasov‐Navier‐Stokessystem.We
would hke toprove
that,
as N\rightarrow\infty and $\epsilon$\rightarrow 0 withappropriate
link betweenthem,
thepair
u^{N}(t,x) , S_{t}^{N}=\displaystyle \frac{1}{N}\sum_{i=1}^{N}$\delta$_{X_{\mathrm{t}}^{i},V_{\mathrm{r}^{i}}}
(S_{\mathrm{t}}^{N}
(
dx,
dv)
isatime‐dependent
randomprobability
measure, calledempirical
measureofthe
particle
system)
converges tothe solution(u, F)
of\displaystyle \frac{\partial F}{\partial t}+v\cdot\nabla_{x}F+\mathrm{d}\mathrm{i}\mathrm{v}_{v}((u-v)F+(K*F)F)=\frac{$\sigma$_{p}^{2}}{2}$\Delta$_{v}F
(2)
Remark 4 The term-\displaystyle \int(u(t,x)-v)F(x,v)dv=\int vF(x,v)
dv‐u
(t,x)\displaystyle \int F(x,\prime v)dv
is the so called Brinkmans
force, usually
denotedinthephysical
literatureby
j-u $\rho$
We shall see that
proving
this hmit result is adifficultproblem.
Let uspreliminarily
describeatechnical
difficulty
withthe Navier‐Stokespart.Concerning
thehterature,
thereareresultsonparticle
systemsrelatedtoVNSequations
but
only
underspecial conditions,
causedby
thefact that atruefluid‐particle
interaction isimposed,
see[1], [2], [6], [15];
and there are resultson convergence of PDEs toPDEs,
although
motivatedby particle
arguments,see[7], [10], [11].
1.1
Difficulty
with the Navier‐Stokesforcing
Letus restricthereto d=2. Theequation
\displaystyle \frac{\partial u^{N}}{\partial t}= $\Delta$ u^{N}-u^{N}
.\displaystyle \nabla u^{N}-\nabla$\pi$^{N}-\frac{c_{0}}{N}\sum_{i=1}^{N}(u_{ $\epsilon$}^{N}(t, X_{t}^{i})-V_{t}^{i})$\delta$_{X_{\mathrm{t}}^{i}}^{ $\epsilon$}
còntainsasubtle
difficulty.
Ifweput $\epsilon$=0,weforceNavier‐Stokesequations
withaninput
which isworsethan
H^{-1}
(recall
that intwodimensions the delta Dirac isonly
inH^{-1- $\gamma$}
forevery
$\gamma$>0
)
andwepretend
tospeak
ofu^{N}(t, X_{t}^{i})
(for
$\epsilon$=0)
whichrequires u^{N}
tobecontinuous.
For $\epsilon$>0 wedo not seethis
regularity
issue;
butweneed uniform estimates in( $\epsilon$, N)
to pass to the
limit,
andthus,
sooner orlater,
wemeet thedifficulty
just
described.Remark 5 The need
for
continuous‐in‐space
velocity field
in this areahas beenrecognized
also
dealing
with otherquestions,
seel141
who assumesu\in L^{2}(0, T;C(D))
.Letus
explain
thisdifficulty
also with thefollowing
argument.
Tosimplify,
assumewehave the heat
equation
inplace
of the Navier‐Stokesone andwehaveonly
onefixed
point
particle
atposition X_{0}
:\displaystyle \frac{\partial u}{\partial t}=\frac{1}{2} $\Delta$ u+(u_{ $\epsilon$}(t, X_{0})-V_{0})$\delta$_{X_{0}}^{ $\epsilon$}
The solution with
u_{0}^{N}=0
iswhere
p_{t}(x)
isthe heat kernel. Ifwetake the hmit as $\epsilon$\rightarrow 0wehaveu(t, x)=\displaystyle \int_{0}^{t}\frac{1}{(2 $\pi$(t-s))^{d/2}}e^{-\frac{|x-X_{\mathrm{f}1}|^{2}}{2(t- $\epsilon$)}}(u(s, X_{0})-V_{0})ds.
Already
ind=2, forx=X_{0}
, wesee that|u(t, X_{0})|=+\infty!
1.2
Why
theproblem
should be solvableNotice however that the
conjectured
limitequation,
the Valsov‐Navier‐Stokes system is better: nodelta Diracappearthere.What is
meaningless,
as remarked intheprevious section,
is the model with a finitenumber
N_{0}
ofpoint
particles
ofmass\displaystyle \frac{1}{N_{0}}
, ifwetake the limit $\epsilon$\rightarrow 0. But this is notwhatwewant to do: we want totake the limit of
infinitely
manyparticles,
with infinitesimalinteraction
strength.
Wemayhope that,
inthe limitas $\epsilon$\rightarrow 0,wemaycontrol thequantities
\displaystyle \frac{\mathrm{b}\mathrm{e}1}{N}
.cause
we also take N\rightarrow\infty and theintensity
offluid‐particle
interaction isrescaledby
However,
torealize thisprogram, itisessentialto provethatparticles
donot concentrate toomuch,
otherwisewe take the riskto haveagain,
in thelimit, cơńcentrated
masses ofparticles
withfinite interactionstrength. Therefore,
amainpurposeof the estimates belowis
proving
aform ofnonconcentration.2
Energy
balance
Lemma 6
Setting
\displaystyle \mathcal{E}_{t}=\frac{1}{2}\int|u^{N}(t, x)|^{2}dx+\frac{1}{2N}\sum_{i=1}^{N}|V_{t}^{i}|^{2}
if
u^{N}
is aregular
solution thenwehaved\displaystyle \mathcal{E}_{t}+(\int|\nabla u^{N}(t, x)|^{2}dx+\frac{1}{N}\sum_{i=1}^{N}(u_{ $\epsilon$}^{N}(t, X_{t}^{i})-V_{t}^{i})^{2})dt
= (\displaystyle \frac{1}{N^{2}}\sum_{i,j=1}^{N}V_{t}^{i}K(X_{t}^{i}-X_{t}^{j})+\frac{$\sigma$_{p}^{2}}{2}) dt+\frac{$\sigma$_{p}}{N}\sum_{i=1}^{N}V_{t}^{i}dW_{\mathrm{t}}^{i}.
The
proof
iselementary
by
Itôformula. Notice that theprevious
result alsogives
us acontrolon
because it is bounded
(up
toconstants)
by
\displaystyle \frac{1}{N}\sum_{i=1}^{N}|V_{t}^{i}|^{2}
plus
\displaystyle \frac{1}{N}\sum_{i=1}^{N}(u_{ $\epsilon$}^{N}(t, X_{t}^{i})-V_{t}^{i})^{2}
that areboth controlled
(the
secondoneintegrated
intime).
Using
theprevious
apriori
estimatesonecanprove, under theassumptions
u_{0}\in \mathrm{L}_{ $\sigma$}^{2}($\Gamma$^{2})
E[\displaystyle \frac{1}{N}\sum_{i=1}^{N}(|X_{0}^{i}|^{2}+|V_{0}^{i}|^{2})] \leq C
(
\mathrm{L}_{ $\sigma$}^{2}($\Gamma$^{2})
is the usual space ofdivergence
freeperiodic
zero mean vector fields on$\Gamma$^{2}
)
existence and
uniqueness
of solutions(for
finiteN)
such thatE[\displaystyle \sup_{t\in[0,T]}\int|u^{N}(t, x)|^{2}dx] \leq C
E[\displaystyle \int_{0}^{T}\int|\nabla u^{N}(t, x)|^{2}dxdt] \leq C
E[\displaystyle \sup_{t\in[0,T]}\frac{1}{N}\sum_{i=1}^{N}(|X_{t}^{i}|^{2}+|V_{t}^{i}|^{2})] \leq C
E[\displaystyle \int_{0}^{T}\frac{1}{N}\sum_{i=1}^{N}(u_{ $\epsilon$}^{N}(t, X_{\mathrm{t}}^{i})-V_{t}^{i})^{2}dt] \leq C
E[\displaystyle \int_{0}^{T}\frac{1}{N}\sum_{i=1}^{N}u_{ $\epsilon$}^{N}(t,X_{t}^{i})^{2}dt] \leq C.
From these
bounds,
withrelatively
classicalcompactnesstheorems,
onecanshow thatthe
family
of laws of(u^{N}, S^{N})
aretight
and thus there existsubsequences
whichconvergein
law;
changing
probability
spaceitispossible
to assume a.s. convergenceinappropriate
topologies.
In thesequel,
to understand thedifficulties,
we assume forsimplicity
sucha.s. convergence. We do not want to
give
the detailshere,
which will be included in aforthcoming
technical work.Let
usonly
mention thatsubsequences
(u^{N_{k}}, S^{N_{k}})
, on thenew
probability
space, will have thepropertythat\bullet
u^{N_{k}}
convergesstrongly
inL^{2}(0, T;L^{2}($\Gamma$^{2}))
\bullet
u^{N_{k}}
convergesweakly
inL^{2}(0, T;W^{1,2}($\Gamma$^{2}))
and weakstar inL^{\infty}(0,T;L^{2}($\Gamma$^{2}))
\bullet
S^{N_{k}}
converges inthe weaktopology
ofmeasuresuniformly
in time.In the
sequel,
when weinformally
discussquestions
of convergence, wereplace
the3
A
difficulty
about
passage to
the
limit
in the Navier‐Stokes
system
Letusstressthat the existenceofaconvergent
subsequence
(u^{N_{k}}, S^{N_{k}})
(denoted
belowby
(u^{N},
S^{N}\backslash
inthetopologies
indicated at the endof theprevious section,
is trUe both ifwe
keep
$\epsilon$>0unchanged
with N, orifwehnkit to Nby choosing $\epsilon$= $\epsilon$ N\rightarrow 0
.However,
in the first case we canpass tothehmit,
in the secondonewe meet arelevant techmicaldifficulty,
that we nowexplain.
In weak formon
divergence
free smoothtest vectorfields$\phi$
, theNavier‐Stokessystemreads
\displaystyle \langle u^{N}(t) , $\phi$\rangle_{x}-\{u_{0}, $\phi$\rangle_{x}+\int_{0}^{t}\langle\nabla u^{N}, \nabla $\phi$\rangle_{x}ds
=\displaystyle \int_{0}^{t}\langle u^{N}\cdot\nabla $\phi$, u^{N}\rangle_{x}ds-\{\frac{1}{N}\sum_{i=1}^{N}(u_{ $\epsilon$}^{N}(t,X_{t}^{i})-V_{t}^{i})$\delta$_{X_{t}^{i}}^{ $\epsilon$} \mathrm{V}\mathrm{q}1), \mathrm{X}, $\phi$(\cdot)\}_{x}
where
\{f,
g\displaystyle \rangle_{x}=\int_{$\Gamma$^{d}}f(x)\cdot g(x)dx
for suitablevectorfieldsf,
g,and\cdotdenotes scalar
product
in
\mathbb{R}^{d}
. Thedifficulty
isonly
inthe convergenceof the lastterm, whenu^{N}
convergesonly
inthe usualtopologies
of weak solutions mentionedattheend of lastsection. What abouttheconvergenceof
u_{ $\epsilon$}^{N}(t, X_{t}^{i})=($\theta$_{ $\epsilon$}^{0}*u_{t}^{N})(X_{t}^{i})
?Itseemsnecessary to provesomeconvergenceof
u^{N}
intheuniform topology.
Butuniformestimatesarenot amongthea
priori
bounds.Although
notbeing
theonly
one,anaturalway to provebounds in theuniformtopology
for
u_{t}^{N}
isby
Sobolevembedding,
henceinvestigating
bounds on derivatives ofu_{t}^{N}
. Sincewe are on a torus and werestrict to d=2, we use
vorticity.
Thequestion
then is: canwe prove
enstrophy
type bounds? Consider thenthevorticity
equation,
which in thecased=2,for the
vorticity
function$\omega$^{N}=\nabla^{\perp}\cdot u^{N}
, is\displaystyle \frac{\partial$\omega$^{N}}{\partial t}= $\Delta \omega$^{N}-u^{N}
.\displaystyle \nabla$\omega$^{N}-\frac{\mathrm{c}_{0}}{N}\sum_{i=1}^{N}\nabla^{\perp}.
((u_{ $\epsilon$}^{N}(t,X_{t}^{i})-V_{t}^{i})$\delta$_{x_{t^{\dot{\mathrm{a}}}}^{N}}^{ $\epsilon$})
.(3)
A main
conceptual
remark is thatparticles
createvorticity.
Terms hke\partial_{i}$\delta$_{X_{{\$}}^{i}}^{$\epsilon$_{N}}
contributediverging
termsinNfor $\epsilon$= $\epsilon$ N\rightarrow 0and thusvorticity
doesnot seemtobe under control.4
Summary
of results
\bullet First we
develop
a twostepsapproach
which consists intwoseparate hmit theo‐rems.
1. The first oneis
only
the hmit as N\rightarrow\infty,given
aconstant value of$\epsilon$\in(0,1)
; itidentifies ahmit
\mathrm{P}\mathrm{D}\mathrm{E}_{ $\epsilon$}
;2. Thesecondoneisthe limit
PDE_{ $\epsilon$}\rightarrow PDE
as $\epsilon$\rightarrow 0.Linking artificialy
( $\epsilon$, N)
, therearesequences( $\epsilon$ k, N_{k})
where Particles($\epsilon$_{k},N_{k})\rightarrow PDE
as k\rightarrow\infty. Thisis notthe solutionwe were
looking
for,
butitisimportant
toknowthat atleast this
relatively simple
two‐stepapproach
works.\bullet
Second,
weconjecture
a local in time result of the form Particles ($\epsilon$_{N},N) \rightarrow PDE asN \rightarrow \infty. It is based on local‐in‐time uniform‐in‐x estimates on
u^{N}
,jointly
withestimatesonno concentrationof
particles
(these
twofactsproceed
together).
A fullproof
stillrequires
to solve\mathrm{t}\mathrm{e}\mathrm{c}\mathrm{b}\mathrm{u}\mathrm{c}\mathrm{s}^{\mathrm{R}}\mathrm{J}
problems,
so we limit ourselves to express areasonable
conjecture.
5
Two‐steps approach
5.1
Preliminaries
The
advantage
ofthetwo‐steps
orseparatelimitstrategy
isthat itworks with minimalingredients:
wedonotneedto proveno‐concentrationofparticles;
wedonotneed the noiseto
regularize;
wecansaysomething
alsointhecased=3(always
on atorus$\Gamma$^{3}
,tosimplify)
Weassume,for
simplicity
ofnotations,
K=0,
$\sigma$_{p}=0,
c_{0}=1
. However the result remainstrue when $\sigma$_{p}
\neq
0 and when Kis boundedLipschitz
continuous andpresumably
also insome caseswhen
K=K_{N}
isrescaled in aproper way.For sake of
clarity
(also
because here there is no average over therandomness),
werestatethe well
posedness
mentioned above forfinite N and theenergybounds.Lemma 7 Forevery
$\epsilon$\in(0,1)
and N\in \mathrm{N}, thesystem
\displaystyle \frac{\partial u^{N}}{\partial t}= $\Delta$ u^{N}-u^{N}\cdot\nabla u^{N}-\nabla$\pi$^{N}-\frac{1}{N}\sum_{i=1}^{N}(u_{ $\epsilon$}^{N}(t, X_{\mathrm{t}}^{i})-V_{t}^{i}).$\delta$_{X_{\mathrm{t}}^{i}}^{ $\epsilon$}
\displaystyle \frac{d}{dt}X_{t}^{i}=V_{t}^{i}, \frac{d}{dt}V_{\mathrm{t}}^{i}=u_{ $\epsilon$}^{N}(t, X_{t}^{i})-V_{t}^{i}
has a
unique
solution such that\displaystyle \sup_{t\in[0,T]}\int|u^{N}(t,x)|^{2}dx+\int_{0}^{T}\int|\nabla u^{N}(t,x)|^{2}dxdt\leq C.
5.2 First limit:
N\rightarrow\infty,
$\epsilon$\in(0,1)
given
Wenowconsider thefollowing mollified
VNSsystem\displaystyle \frac{\partial u}{\partial t}= $\Delta$ u-u\cdot\nabla u-\nabla $\pi-\theta$_{ $\epsilon$}^{0}*\int(u_{ $\epsilon$}(t, \cdot)-v)F(\cdot, dv)
(4)
\displaystyle \frac{\partial F}{\partial t}+v\cdot\nabla_{x}F+\mathrm{d}\mathrm{i}\mathrm{v}_{v}((u_{ $\epsilon$}-v)F)=0
(5)
whereu_{ $\epsilon$}=$\theta$_{ $\epsilon$}^{0}*u.
Thanks to the
mollification,
we are allowed toinvestigate
thissystem and convergenceof theparticle
system when thedensity
ofparticles
is treatedjust
as a measure, not as adensity
function. Letusgive
theappropriate
definition. Denoteby
\mathrm{P}\mathrm{r}_{1}($\Gamma$^{d}\times \mathbb{R}^{d})
thesetof all Borel
probability
measures $\mu$on$\Gamma$^{d}\times \mathbb{R}^{d}
such that\displaystyle \int_{$\Gamma$^{d}}\prime\int_{\mathbb{R}^{d}}|v| $\mu$
(
dx,
dv)<\infty
we endow
\mathrm{P}\mathrm{r}_{1}($\Gamma$^{d}\times \mathbb{R}^{d})
with the weaktopology,
withconvergenceof firstmoment. The notation$\theta$_{ $\epsilon$}^{0}*\displaystyle \int(u_{ $\epsilon$}(t, \cdot)-v)
F.(dv),
when F \inC([0, T];\mathrm{P}\mathrm{r}_{1}($\Gamma$^{d}\times \mathbb{R}^{d}))
and\mathrm{u}_{ $\epsilon$}(t, \cdot)
is measurable andbounded,
standsfor($\theta$_{ $\epsilon$}^{0}*\displaystyle \int(u_{ $\epsilon$}(t, \cdot)-v)F(\cdot, dv))(x)=\int_{$\Gamma$^{\mathrm{d}}}\int_{\mathbb{R}^{d}}$\theta$_{ $\epsilon$}^{0}(x-x')(u_{ $\epsilon$}(t, x')-v')F(t, dx', dv')
.Beside the notation
\langle f,
g\}_{x}
already
introducedabove,
herewealsowrite( $\mu$,
f\}_{x,v}
for\displaystyle \{ $\mu$, f\}_{x,v}=\int_{$\Gamma$^{d}}^{\backslash }\int_{\mathbb{R}^{\mathrm{d}}}f(x, v) $\mu$
(dx
,dv)
.Definition8 Let u_{0} \in
\mathrm{L}_{ $\sigma$}^{2}($\Gamma$^{2})
andF_{0}
\in\mathrm{P}\mathrm{r}_{1}($\Gamma$^{d}\times \mathbb{R}^{d})
begiven.
Apair
(u, F)
is asolution
of
system
(4)‐(5)
with initial condition(u_{0}, F_{0})
if
u\in L^{\infty}(0, T;L^{2}($\Gamma$^{d}))\cap L^{2}(0, T;W^{1,2}($\Gamma$^{d}))
F\in C([0, T] ; Pr_{1}($\Gamma$^{d}\times \mathbb{R}^{d}))
\displaystyle \{u(t), $\phi$\}_{x,\backslash }-\{u_{0}, $\phi$)_{x}+\int_{0}^{t}\langle\nabla u, \nabla $\phi$\}_{x}ds
=\displaystyle \int_{0}^{t}\{u\cdot\nabla $\phi$, u\rangle_{x}ds-\int_{0}^{t}\langle$\theta$_{ $\epsilon$}^{0}*\int(u_{ $\epsilon$}(s, \cdot)-v)F(s, \cdot, dv) , $\phi$\rangle_{x}ds
\{F(t)
,$\varphi$\rangle_{x,v}=\{F_{0},
$\varphi$\displaystyle \rangle_{x,v}+\int_{0}^{t}\{F(s)
,v\displaystyle \cdot\nabla_{x} $\varphi$\rangle_{x,v}ds+\int_{0}^{t}\{F(s), (u_{ $\epsilon$}(s)-v)\cdot\nabla_{v} $\varphi$\}_{x,v}ds
Theorem 9 Let
$\epsilon$\in(0,1)
begiven.
Assumeu_{0}\in \mathrm{L}_{ $\sigma$}^{2}
(T2),
\displaystyle \frac{1}{N}\sum_{i=1}^{N}(|X_{0}^{i}|^{2}+|V_{0}^{i}|^{2})
\leq C,
and
S_{0}^{N}\rightarrow F_{0}
inthe weaksenseof probability
measures.1)
Let d=3. Given $\epsilon$\in(0
,1
)
, there exists asubsequence N_{k}\rightarrow\infty
such that thepair
(u^{N_{k}}, S^{N_{k}})
converges(in
the sense described at the endof
section2)
to a solution(u, F)
of
system
(4)‐(5)
with initial condition(u_{0}, F_{0})
.2)
Let d = 2.System
(4)‐(5),
with initial condition(u0, F_{0})
, has a
unique
solution(u, F)
and,
as N\rightarrow\infty, thepair
(u^{N}, S^{N})
converges to(u, F)
.Remark 10 In d=3
obviously
we donotknowuniqueness
duetotheNavier‐Stokespart;hence the convergence holds
only for
certainsubsequences
(for
everysubsequence
there isa
sub‐subsequence
whichconverges).
Remark 11 Existence
of
a solution(u, F)
of
the limitsystem
(with
given
$\epsilon$ \in(0,1))
eithercan be
proved
directly
oritfollows from
theconvergence resultitself,,
being
based on a compactnessargument. Uniqueness
of
(u, F) (
for
d=2)
has to beproved directly.
Letus
give
afew elementsoftheproof.
From the estimates\displaystyle \sup_{t\in[0,T]}\int|u^{N}(t,x)|^{2}dx+\int_{0}^{T}\int|\nabla u^{N}(t,x)|^{2}dxdt\leq.C
it isclassical
(cf. [17])
toapply
thecompactnessAubin‐Lions lemma. Duetothe estimate\displaystyle \sup_{t\in[0,T]}\frac{1}{N}\sum_{i=1}^{N}(|X_{t}^{i}|^{2}+|V_{t}^{i}|^{2}) \leq C
one can use a criterion based on Wasserstein distance to prove
compactness
ofS^{N_{k}^{(0)}}
in
C([0,T];\mathrm{P}\mathrm{r}_{1}($\Gamma$^{d}\times \mathbb{R}^{d}
From these fact one has the existence of asubsequence
(u^{N_{k}}, S^{N_{k}})
which converges as described at the end of section 2. Call(u, F)
the limitof such
subsequence.
Taking
the limitinthe first fourtermsof the weak formulation ofNavier‐Stokesequa‐tions
(see
Definition8)
is classical(cf. [17]).
Concerning
the last term, wehaveto provethat
k\displaystyle \rightarrow\infty \mathrm{h}\mathrm{m}\{\frac{1}{N_{k}}\sum_{i=1}^{N_{k}}(u_{ $\epsilon$}^{N_{k}}(t,X_{t}^{i,N_{k}})-V_{t}^{i,N_{k}})$\delta$_{X_{t}^{i,N_{h}}}^{ $\epsilon$}, $\phi$\}_{x}
Thetermon theleft‐hand sideis
equal
to\displaystyle \int_{$\Gamma$^{d}}\frac{1}{N_{k}}\sum_{i=1}^{N_{k}}(u_{ $\epsilon$}^{N_{k}}(t, X_{t}^{i,N_{k}})-V_{t}^{i,N_{k}})$\theta$_{ $\epsilon$}^{0}(x-X_{t}^{i,N_{k}}) $\phi$(x)dx
=\displaystyle \int_{$\Gamma$^{d}}\int_{$\Gamma$^{d}}\int_{\mathbb{R}^{d}}(u_{ $\epsilon$}^{N_{k}}(t, x')-v')$\theta$_{ $\epsilon$}^{0}(x-x') $\phi$(x)S^{N_{k}}
(dx
,
dv')dx
=\displaystyle \int_{$\Gamma$^{d}}\int_{\mathbb{R}^{d}}(($\theta$_{ $\epsilon$}^{0}*u^{N_{k}}(t))(x')-v')($\theta$_{ $\epsilon$}^{0-}* $\phi$)(x')S^{N_{k}}
(dx
,
dv')
where
$\theta$_{ $\epsilon$}^{0-}(x) =$\theta$_{ $\epsilon$}^{0}(-x)
. The term$\theta$_{ $\epsilon$}^{0}*u^{N_{k}}(t)
converges‐uniformly
to$\theta$_{ $\epsilon$}^{0}*u(t)
fora.e. t(passing
to asubsequence).
With httleadditionalcarewe cantake the limitas k\rightarrow\inftyand get
\displaystyle \int_{$\Gamma$^{d}}\int_{\mathbb{R}^{d}}(($\theta$_{ $\epsilon$}^{0}*u(t))(x')-v')($\theta$_{ $\epsilon$}^{0-}* $\phi$)(x')F(t, dx', dv')
=\displaystyle \langle\int_{$\Gamma$^{d}}\int_{\mathbb{R}^{d}}$\theta$_{ $\epsilon$}^{0}(\cdot-x')(u_{ $\epsilon$}(t, x')-v')F(t, dx', dv') , $\phi$\rangle_{x}
Toprove that F satisfies the weak
identity
inDefinition 8 we have first to derive anidentity
forS^{N_{k}}
.By
chain ruleapplied
to$\varphi$(X_{t}^{i,N_{k}}, V_{t}^{i,N_{k}})
weget\langle S_{t}^{N_{k}},
$\varphi$\rangle_{x,v}=\langle S_{0}^{N_{h}},
$\varphi$\displaystyle \rangle_{x,v}+\int_{0}^{t}\langle S_{8}^{N_{k}},
v\displaystyle \cdot\nabla_{x} $\varphi$\rangle_{x,v}ds+\int_{0}^{t}\langle S_{8}^{N_{k}},
(u_{ $\epsilon$}(s)-v)\cdot\nabla_{v} $\varphi$\rangle_{x,v}ds
(in
thedeterministic caseit is a well knownfact that theempirical
measure isalready
asolution of the limit
PDE;
inthe stochasticcase,$\sigma$_{p}\neq 0
,onehastoapply
Itô formula andan additional
martingale
term appears,whichhowever,
converges tozero).
Thenonecanpass to the limit.
Finaly,
for d=2wehaveto proveuniqueness
for the limitsystem(4)-(5)
. Inprinciple,
oneof the difficultiesisthatwedeal with solutions F whichare
only
measures.However,
wemayuse awellknown method
(see
for instance[3])
basedonWasserstein distanced_{1}
(
$\mu$,iỷ)
between $\mu$,
\mathrm{v}\in \mathrm{P}\mathrm{r}_{1}($\Gamma$^{d}\times \mathbb{R}^{d})
. Assume that(u, F)
,
(u', F')
aretwosolutions. One hasd_{1}(F(t), F'(t))\leq E[|X_{t}-X_{t}'|+|V_{t}-V_{t}
where
(X_{t}, V_{t})
satisfies\displaystyle \frac{d}{dt}X_{t}=V_{t}
whereuisthefirstcomponentof the solution
(u, F)
, andsimilarly
for(Xt, V_{t}'
) (with
respectto
(
u',
F Theinitial conditions for thesetwoproblems
arethesame,(X_{0},
V_{0}
with lawF_{0}
. One caneasily
provethatE[|X_{t}-X_{t}^{l}|+|V_{t}-V_{t} \displaystyle \leq C\int_{0}^{t}E[|X_{s}-X_{s}|+|V_{s}-\acute{V}_{s}'|]ds
+C\displaystyle \int_{0}^{t}|u_{ $\epsilon$}(s, X_{s})-u_{ $\epsilon$}'(s, X_{s})|ds.
Then one has to repeat classical energy type
computations
of the 2‐dimensionaìtheory
of Navier‐Stokesequations
(cf. [17])
to control u-u' inthe normsL^{\infty}(0, T;L^{2}($\Gamma$^{d}))\cap
L^{2}(0, T;W^{1,2}($\Gamma$^{d}))
, boundsto be usedjointly
with theprevious
one. Theonly
nonclas‐sicaltermis
|\displaystyle \langle$\theta$_{ $\epsilon$}^{0}*\int(u_{ $\epsilon$}(t, \cdot)-v)F(\cdot, dv)
,u(t)\displaystyle \rangle_{x}-\langle$\theta$_{ $\epsilon$}^{0}*\int(u_{ $\epsilon$}'(t, \cdot)-v)F'(\cdot, dv)
,u'(t)\rangle_{x}|
whichis controlled
by
theprevious
normsofu-u'plus
theterm|\displaystyle \langle$\theta$_{ $\epsilon$}^{0}*\int(u_{ $\epsilon$}(t, \cdot)-v)(F(\cdot, dv)-F'(\cdot, dv)) , u(t)\rangle_{x}|
=|\displaystyle \int_{$\Gamma$^{d}}\int_{\mathbb{R}^{d}}($\theta$_{ $\epsilon$}^{0-}*u(t))(x)(u_{ $\epsilon$}(t, x)-v)(F (dx, dv)-F' (dx, dv))|
\leq C\cdot d_{1}(F(t), F'(t))
(the
last boundrequires
somework,
omittedhere).These
recursive estimates aJlow onetoapply
Gronwall lemma andprove that(u, F)=(u', F')
.5.3 Second limit: $\epsilon$\rightarrow 0
Untilnowwehaveproventhat the
fluid,
coupled
with theparticles,
converges, as N\rightarrow\inftyto system
(4)-(5)
, where the molhfication with $\epsilon$ \in(0,1)
survives. Called(u^{ $\epsilon$}, F^{ $\epsilon$})
thesolution in d=2
(or
asolutionind=3)
of(4)-(5)
, it remains toinvestigate
thehmit as$\epsilon$\rightarrow 0.
The hmit cannot be taken at thelevelofmeasure solutions F, orat least this looks
verydifficult. Ońecan
give
ameaning
tothe weak formulation of the VNS(1)-(2)
when F isonly measure‐valued,
butattheprice
ofimposing
apriori
thatuiscontinuous bounded. This direction could beinvestigated
butrequires
afully
original
approach
toVNS system which isbeyond
the scope of this note. And inaddition,
as remarkedbelow,
oneshouldexpect
only
local‐in‐time solutions.Therefore let us consider the modified VNS system and the true one when F is a
fmction.
First,
onehasto prove that themodified VNSsystemhas asolution(u^{ $\epsilon$}, F^{ $\epsilon$})
inbe
umique
also inthe weaker class ofmeasure valued solutions. Then one should proveconvergenceof
(u^{ $\epsilon$}, F^{ $\epsilon$})
to asolution(u, F)
ofsystem(1)-(2)
. We have checked that boththese steps are
plausible following
theapproach
of[18];
see also[4];
however there areseveral details and results willappearina
forthcoming
work.When this is
done,
it ispossible
to extract suitable sequences( $\epsilon$ N)
such that(u^{N_{k}}, S^{N_{k}})
converges to(u, F)
as k\rightarrow\infty.Here, by
(u^{N_{k}}, S^{N_{k}})
, wemean thoseobtainedby
thefluid‐particle
systemwith $\epsilon$= $\epsilon$ k.6
Joint
limit
6.1
Introduction
As described inSections 1.1 and
3,
uniform estimatesonthevelocity
arerequired
to passto the hmit
simultaneously
in N\rightarrow\infty and $\epsilon$\rightarrow 0. A naturalapproach
to proveuniformbounds on
u^{N}
is to getW^{ $\epsilon$,2}
(for
$\epsilon$ > 0)
bounds onthevorticity $\omega$^{N}
=\nabla^{\perp}\cdot \mathrm{u}^{N}
, whichsatisfies
equation
(3):
Bounds on theenstrophy
arenotsufficient,
sincethey
are bounds ontheW^{1,2}
‐normofu^{N}
which donotimply
uniform boundsonu^{N}
. Hence wework withsemigroups
and look for boundsinmoreregular topologies.
Notice however thatenstrophy
boundsmeet thesamedifficultieswehave with
semigroups.
Denoting by e^{t $\Delta$}
thesemigroup
associatedtotheLaplacian
operatorinƯorC^{ $\alpha$}spacesonthe torus, wehave
\displaystyle \frac{\partial$\omega$^{N}}{\partial t}= $\Delta \omega$^{N}-u^{N}
.\displaystyle \nabla$\omega$^{N}-\frac{c_{0}}{N}\sum_{i=1}^{N}\nabla^{\perp}.
((u_{ $\epsilon$}^{N}N(t, X_{t}^{i})-V_{t}^{i})$\delta$_{X_{t}^{i}}^{$\epsilon$_{N}})
.$\omega$^{N}(t)=e^{t $\Delta$}$\omega$^{N}(0)-\displaystyle \int_{0}^{t}e^{(t-s) $\Delta$}u^{N}(s) \nabla$\omega$^{N}(s)ds
-\displaystyle \int_{0}^{t}e^{(t- $\epsilon$) $\Delta$}\nabla^{\perp}. \frac{c_{0}}{N}\sum_{i=1}^{N}(u_{$\epsilon$_{N}}^{N}(s, X_{s}^{i})-\dot{ $\psi$}_{s})$\delta$_{X_{s}^{i}}^{ $\epsilon$}ds.
Wewant to estimate
$\omega$^{N}(t)
inW^{2 $\alpha$,2}
(Td),
hencewe usetheinequality
\Vert(I- $\Delta$)^{a}$\omega$^{N}(t)\Vert_{L^{2}($\Gamma$^{d})} \leq \Vert(I- $\Delta$)^{ $\alpha$}$\omega$^{N}(0)\Vert_{L^{2}($\Gamma$^{d})}
+\displaystyle \int_{0}^{t}\Vert(I- $\Delta$)^{ $\alpha$}e^{(t-s) $\Delta$}u^{N}(s)
.\nabla$\omega$^{N}(s)\Vert_{L^{2}($\Gamma$^{d})}ds
+\displaystyle \int_{0}^{t}\Vert(I- $\Delta$)^{\frac{1}{2}+ $\alpha$}e^{(t-s) $\Delta$}\nabla^{\perp}(I- $\Delta$)^{-\frac{1}{2}}\cdot\frac{\mathrm{c}_{0}}{N}\sum_{i=1}^{N}(u_{$\epsilon$_{N}}^{N}(s, X_{s}^{i})-V_{8}^{i})$\delta$_{X_{8}^{i}}^{ $\epsilon$}\Vert_{L^{2}('$\Gamma$^{d})^{ds}}
.(6)
Let us
only
concentrate on the term(6)
withoutV_{8}^{i}
, which is the sóurce of the maindifficulty.
For every T> 0, denoteby
\Vert\cdot\Vert_{T,\infty}
the supremumnorm over[0, T]
\times$\Gamma$^{d}
. Wehave
|\displaystyle \frac{1}{N}\sum_{i=1}^{N}u_{$\epsilon$_{N}}^{N}(s, X_{s}^{i})$\delta$_{X_{s}^{i}}^{ $\epsilon$}(x)|\leq\Vert u_{$\epsilon$_{N}}^{N}\Vert_{T,\infty}F_{S}^{0,N}(x)
where
F_{t}^{0,N}=$\theta$_{$\epsilon$_{N}}^{0}*(\displaystyle \frac{1}{N}\sum_{i=1}^{N}$\delta$_{X_{t}^{i}})
and therefore theterm(ô)
withoutV_{ $\epsilon$}^{i}
isbounded aboveby
(
\nabla^{\perp}(I- $\Delta$)^{-\frac{1}{2}}
is aboundedoperatorinL^{2}(.$\Gamma$^{d})
)
\displaystyle \Vert\nabla^{\perp}(I- $\Delta$)^{-\frac{1}{2}}\Vert_{L^{2}($\Gamma$^{d})\rightarrow L^{2}($\Gamma$^{d})}\Vert u_{$\epsilon$_{N}}^{N}\Vert_{T,\infty}\int_{0}^{t})
\leq C\Vert u_{$\epsilon$_{N}}^{N}\Vert_{T,\infty}T^{\frac{1}{2}- $\alpha$}
\displaystyle \sup\Vert F^{0,N}\Vert_{L^{2}($\Gamma$^{d})}
(7)
t\in[0,T] because
\displaystyle \Vert(I- $\Delta$)^{\frac{1}{2}+ $\alpha$}e^{(t-s) $\Delta$}\Vert_{L^{2}('$\Gamma$^{d})\rightarrow L^{2}($\Gamma$^{d})} \leq\frac{C}{(t-s)^{\frac{1}{2}+ $\alpha$}}
by
well knownanalytic semigroup
estimates.We need an estimate on
\Vert F^{0,N}\Vert_{L^{2}( $\Gamma$)^{d}}
: this is the property of no concentration ofparticles,
asannouncedinSection 1.2.6.2 No concentration of
particles
Set
F_{t}^{N}=$\theta$_{$\epsilon$_{N}}*(\displaystyle \frac{1}{N}\sum_{i=1}^{N}$\delta$_{X_{\mathrm{t}}^{l},V_{\mathrm{r}^{i)}}}
, wherenow$\theta$_{ $\epsilon$}=$\theta$_{ $\epsilon$}(x, v)
are sm\cdot
tablemolhfiers inboth
variables,
related to$\theta$_{ $\epsilon$}^{0}
. Hereweneed$\sigma$_{p}\neq 0
. One hasdF_{t}^{N}= (\displaystyle \frac{$\sigma$_{p}^{2}}{2}$\Delta$_{v}F_{t}^{N}-\nabla_{x} . $\theta$_{$\epsilon$_{N}}*(vS_{\mathrm{t}}^{N}))dt
-(\nabla_{v}\cdot$\theta$_{$\epsilon$_{N}}*((u_{$\epsilon$_{N}}^{N}(t, x)-v)S_{t}^{N}))dt+dM_{t}^{N}
where
M_{t}^{N}(x, v)=\displaystyle \frac{1}{N}\sum_{i=1}^{N}\int_{0}^{t}\nabla_{v}$\theta$_{$\epsilon$_{N}}(x-X_{s}^{i}, v-V_{s}^{i})$\sigma$_{p}dB_{s}^{i}.
\displaystyle \frac{1}{2}d\int_{$\Gamma$^{\mathrm{d}}}\int_{\mathbb{R}^{d}}(F_{t}^{N})^{2}dxdv+\frac{$\sigma$_{p}^{2}}{2}\int_{$\Gamma$^{\mathrm{d}}}\int_{\mathbb{R}^{d}}|\nabla_{v}F_{t}^{N}|^{2}
dxdvdt=\displaystyle \int_{$\Gamma$^{d}}\int_{\mathbb{R}^{d}}($\theta$_{$\epsilon$_{N}}*(vS_{t}^{N}))\nabla_{x}F_{t}^{N}
dxdvdt+\displaystyle \int_{$\Gamma$^{d}}\int_{\mathbb{R}^{d}}($\theta$_{$\epsilon$_{N}}*((u_{$\epsilon$_{N}}^{N}(t, x)-v)S_{t}^{N}))\nabla_{v}F_{t}^{N}
dxdvdtplus
termsrelatedtothemartingale
partthat wedonotdiscussexplicitly
here. Letus seehowto treatthe mostdifficult term: since
|($\theta$_{$\epsilon$_{N}}*(u_{$\epsilon$_{N}}^{N}(t, x)S_{t}^{N}))(x, v)|
=|\displaystyle \int_{$\Gamma$^{\mathrm{d}}}\int_{\mathbb{R}^{d}}$\theta$_{$\epsilon$_{N}}(x-x', v-v')u_{$\epsilon$_{N}}^{N}(t, x')S_{t}^{N}
(dx
,
dv')|
\displaystyle \leq\int_{\mathrm{N}^{d}}\'{I}_{\mathbb{R}^{d}}$\theta$_{$\epsilon$_{N}}(x-x', v-v')|u_{$\epsilon$_{N}}^{N}(t, x')|S_{t}^{N}(dx', dv')
\leq\Vert u_{$\epsilon$_{N}}^{N}\Vert_{T,\infty}($\theta$_{$\epsilon$_{N}}*S_{t}^{N})(x, v)
wehave
|\displaystyle \int_{$\Gamma$^{d}}\int_{\mathbb{R}^{\mathrm{d}}}($\theta$_{$\epsilon$_{N}}*(u_{$\epsilon$_{N}}^{N}(t, x)S_{t}^{N}))\nabla_{v}F_{t}^{N}dxdv|
\displaystyle \leq\Vert u_{$\epsilon$_{N}}^{N}\Vert_{T,\infty}\int_{$\Gamma$^{d}}\int_{\mathbb{R}^{d}}F_{t}^{N}|\nabla_{v}F_{t}^{N}|dxdv
\displaystyle \leq $\epsilon$\int_{\mathrm{T}^{d}}\int_{\mathbb{R}^{d}}|\nabla_{v}F_{t}^{N}|^{2}dxdv+\frac{\Vert u_{ $\epsilon$}^{N}N\Vert_{T,\infty}^{2}}{ $\epsilon$}\int_{$\Gamma$^{d}}\int_{\mathbb{R}^{d}}(F_{t}^{N})^{2}dxdv.
Summarizing,
\displaystyle \frac{1}{2}d\int_{$\Gamma$^{d}}\int_{\mathbb{R}^{\mathrm{d}}}(F_{t}^{N})^{2}dxdv+\frac{$\sigma$_{p}^{2}}{4}\int_{$\Gamma$^{\mathrm{d}}}\int_{\mathbb{R}^{d}}|\nabla_{v}F_{t}^{N}|^{2}
dxdvdt=\displaystyle \int_{$\Gamma$^{d}}\int_{\mathbb{R}^{\mathrm{d}}}($\theta$_{$\epsilon$_{N}}*(vS_{t}^{N}))\nabla_{x}F_{\mathrm{t}}^{N}
dxdvdt+\displaystyle \frac{\Vert u_{$\epsilon$_{N}}^{N}\Vert_{T,\infty}^{2}}{ $\epsilon$}\int_{$\Gamma$^{d}}\int_{\mathrm{N}^{d}}(F_{t}^{N})^{2}dxdv
plus
termsrelatedto themartingale.
Heuristically,
itseems that for small T,using
(7),
theprevious
estimates close andgive
aboundon\Vert u^{N}\Vert_{T,\infty}
andHowever,
therearestill several nontrivial technicalproblems
tobeovercome. In theprevi‐
oussectionwe needed anestimateon
\Vert F_{t}^{0,N}\Vert_{L^{2}($\Gamma$^{d})}
.Here,
inthissection,
wehaveshowna controlon
F_{t}^{N}
,notF_{t}^{0,N}
. Onecanprovethe estimate\displaystyle \int F_{t}^{0,N}(x)^{2}dx\leq C\int\int|v|^{3}F_{t}^{N}(x, v)dxdv+C\int\int F_{t}^{N}(x, v)^{4}dxdv
(this
is avariant of Lemma 1 of[4],
which avoids\Vert F_{t}^{N}\Vert_{\infty}
, sinceit looks too difficulttoestimate
\Vert F_{t}^{N}\Vert_{\infty}
).
But then thetwoquantities
ontheright‐hand‐side
of thisinequality
have tobe controlled. Wepresume that all thesesteps can be done but dueto thecom‐
plexity
of theseestimates,
instead offormulating
aresult,
weprefer
to limit ourselvestostatea
conjecture.
Conjecture
12 Assumed=2,
u_{0}\in W^{2 $\alpha$,2}($\Gamma$^{d})
,\displaystyle \int_{$\Gamma$^{d}}\int_{\mathbb{R}^{d}}(F_{0}^{N})^{2}dxdv\leq C
. Let(u^{N}, X_{t}^{i}, V_{$\iota$^{i}})
be the solution
of
thefluid‐particle
interacting
system, with$\epsilon$= $\epsilon$ N\rightarrow 0
as N\rightarrow\infty. SetF_{t}^{N}=$\theta$_{$\epsilon$_{N}}*(\displaystyle \frac{1}{N}\sum_{i=1}^{N}$\delta$_{X_{t}^{i},V_{$\iota$^{i}}})
.Then,
for
smallT,
(u^{N}, F^{N})
converges totheunique
solution(u, F)
of
VNSsystem.The convergence should holds in several
topologies, including
the strongtopology
ofL^{2}(0, T;L^{2}($\Gamma$^{d}))
foru^{N}
, and ofL^{2}(0, T;L^{2}($\Gamma$^{d}\times \mathbb{R}^{2}))
forF^{N}.
6.3
Open questions
A first main hmitation of the results described here is the
phenomenological
description
of the
fluid‐body
interaction. We havealready
remarked in the Introduction about the difficulties metby
morereaJistic models.Thé
two‐stepapproach
iscomplete
and extendible to stochasticdynamics.
Ón
the contrary, themoreinteresting joint
limitapproach,
eveniftrue, containstwo restrictions: the short time and the presence of noise in theparticles
‐viscosity
in the PDE. Theshort timeis
due, conceptually,
tothevorticity
produced by
the immersedparticles,
which increases both with fluidvelocity
andparticle density,
thereforeintroducing
aquadratic
terminthe
equations.
Blow‐up
duetoquadratic
terms isprevented by
suitable conservation laws andwehavean energyinequality
butwemissaconservation law forvorticity,
duetothe
vorticity
production by particles.
Following
[18],
section4.1,
itcould be that uniform‐in‐z estimatesonu^{N}
canbereplaced
by
estimates inL^{4}
,whichareglobal;
correspondingly,
anL^{4}
‐controlonF^{N}
isneeded. Hereandfor otherpurposes, weseethe
importance
ofamajor
problem: F^{N}
doesnotsatisfy
acontinuity equation,
butanidentity
with weakergeometric properties.
Asa
remark,
as soon aswerestricttolocalintimeresults,
itseemspossible
toextend the result of thejoint
limit to the 3\mathrm{D} case,by working
in spaces ofsufficiently
regular
solutionsu;see arelated
problem
in[12].
Thisstressesoncemorethefact thatpresumably
wehave nottaken in full
advantage
theproperties
of 2\mathrm{D} fluids.The
viscosity
has been used in ourapproach
to obtain anL^{2}
‐estimate onF_{t}^{N}
, con‐ceptually
fundamentalas a meanto prove noconcentration ofparticles.
However,
for thelimit
equation
for F, Ư‐estimates areeasily
obtained in terms of the Ư‐norm ofinitialconditions,
without need ofanyLaplacian
(see
forinstance[13]).
This could beasignature
of the fact that noise is notneededto prove
L^{2}
‐estimateonF_{t}^{N}.
Finally,
theapriori
estimate on\displaystyle \int_{0}^{T}\frac{1}{N}\sum_{i=1}^{N}(u_{ $\epsilon$}^{N}(t, X_{t}^{i})-V_{t}^{i})^{2}dt=\int_{0}^{T}\int(u_{ $\epsilon$}^{N}(t,x)-v)^{2}S_{t}^{N}
(dx
,dv)dt
obtained
by
theenergybound lookspromising
tocontrol thedifficultquadratic
terms,butinall
computations they
seemto becoupled
withotherterms notunder control.Acknowledgement
13 The author thanks Christian Oliverafor
severalpreliminary
dis‐cussionsand the
organizers
of
workshops
atKomabaUniversity
an\grave{d}
WasedaUniversity
inTokyo
and at RIMSKyoto
in November 2016for
theirhospitality
and theopportunity
todiscuss the
topics
of
this note.References
[1]
G.Allaire, Homogenization
of the Navier‐Stokesequations
in open setsperforated
with
tiny
holes,
Arch. Ration. Mech. Anal. 113(1991),
209‐259.[2]
E.Bernard,
L.Desvilettes,
F.Golse,
V.Ricci,
A derivation of the Vlasov‐Navier‐ Stokes model for aerosol flows from kinetictheory,
arXivpreprint
2016,
n. 1608‐00422.[3]
$\Gamma$.Bolley,
A.Guillin,
F.MaJrieu,
Trendtoequihbrium
andparticle
approximation
fora
weakly
selfconsistent Vlasov‐Fokker‐PlanckEquation,
ESAIM44(2010)
867‐884.[4]
L.Boudin,
L.Desvillettes,
C.Grandmont,
A.Moussa,
Global Existence of Solutions for theCoupled
Vlasov and Navier‐StokesEquations,
Diff.
Int.Eq.
22(2009),
n.11‐12,
1247‐1271.
[5]
B.Desjardins,
M.J.Esteban,
Existence of weak solutions for the motion ofrigid
bodiesinaviscous
fluid,
Arch. Ration. Mech. Anal. 146(1999),
59‐71.[6]
L.Desvillettes,
$\Gamma$.Golse,
V.Ricci,
The mean‐field limitfor solidparticles
inaNavier‐Stokes
flow,
J. Stat.Phys.
131(2008),
941‐967.[7]
L.Desvillettes,
J.Mathiaud,
Some aspects of theasymptotics
leading
from gas‐particles equations
towardsmultiphase
flowsequations,
J. Stat.Phys.
141(2010),
[8]
E.Feireisl,
Y.Namlyeyeva,
Š.
Nečasová,
Homogenization
of theevolutionary
Navier‐ Stokessystem, Manuscr. Math. 149(2016),
n.1,
251‐274.[9]
D.Gérard‐Varet,
M.Hillairet,
Regularity
issues in theproblem
of fluid structureinteraction,
Arch. Ration. Mech. Anal. 195(2010),
375‐407.[10]
T.Goudon,
P.‐E.Jabin,
A.Vasseur, Hydrodynamic
hmit for theVlasov‐Navier‐Stokesequations.
I.Light particles regime,
Indiana Univ. Math. J. 53(2004),
1495‐1515.[11]
T.Goudon,
P.‐E.Jabin,
A.Vasseur, Hydrodynamic
limitfor the Vlasov‐Navier‐Stokesequations.II.
Fineparticles regime,
Indiana Univ. Math. J. 53(2004),
1517−1ó36.[12]
T.Goudon,
L.He,
A.Moussa,
P.Zhang,
The Navier‐Stokes‐Vlasov‐Fokker‐Plancksystemnear
equillibrium,
SIAM J. Math. Anal. 42(2010),
no.5,
2177‐2202.[13]
K.Hamdache,
Global existence andlarge
time behaviour ofsolutions for the Vlasov‐ Stokesequations,
Japan
J. Indvst.Appl.
Math. 15(1998),
no.1,
51‐74.[14]
P.‐E.Jabin,
Various levels of models foraerosols,
Math. Mod. Meth.Appl.
Sci. 12(2002),
n.7,
903‐919.[15]
P.‐E.Jabin,
$\Gamma$.Otto,
Identification of the diluteregime
inparticle sedimentation,
Comm. Math.
Phys.
250(2004),
415‐432.[16]
O.Glass,
A.Munmer,
$\Gamma$.Sueur,
Point vortexdynamics
as zero‐radius limit of themotionofa