HAL Id: hal-01398874
https://hal.archives-ouvertes.fr/hal-01398874
Preprint submitted on 17 Nov 2016
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
problem with discontinuous permeability
Rafael Granero-Belinchón, Steve Shkoller
To cite this version:
problem with dis ontinuous permeability
RafaelGranero-Belin hón
UnivLyon,UniversitéClaudeBernardLyon1 CNRSUMR5208,InstitutCamilleJordan
43blvd. du11novembre1918 F-69622Villeurbanne edex, Fran e
email: graneromath.univ-lyon1.fr Steve Shkoller Departmentof Mathemati s University of California Davis, CA 95616 USA email: shkollermath.u davis.edu November 17,2016
Werstprovelo al-in-timewell-posednessfortheMuskatproblem,modelinguidowin a two-dimensional inhomogeneous porous media. The permeability of the porous medium is des ribed by a step fun tion, with a jump dis ontinuity a ross the xed-in-time urve
(x
1
, −1 + f (x
1
))
,whiletheinterfa eseparating theuidfromtheva uumregionisgivenby thetime-dependent urve(x
1
, h(x
1
, t))
. Ourestimatesarebasedonanewmethodologythat reliesupona arefulstudyofthePDEsystem, ouplingDar y'slawandin ompressibilityof theuid,rather thanthe analysisofthe singularintegral ontourequationfor theinterfa e fun tionh
. We are able to develop an existen e theory for any initial interfa e given byh
0
∈ H
2
andanypermeability urve-of-dis ontinuitythatisgivenbyf ∈ H
2.5
. Inparti ular, our method allows for both urves to have (pointwise) unbounded urvature. In the ase that the permeability dis ontinuity is the set
f = 0
, we prove global existen e and de ay to equilibrium for small initial data. This de ay is obtained using a new energy-energy dissipation inequality that ouples tangential derivatives of thevelo ity in the bulk of the uidwiththe urvatureoftheinterfa e. Tothebestofourknowledge, thisistherstglobal existen eresultfor theMuskatproblem withdis ontinuouspermeability.Contents
1 Introdu tion 2
2 The Muskat problem in the ALE formulation 10
3 A priori estimates 14
6 Proof of Theorem 2: Global well-posedness when
f = 0
281 Introdu tion
1.1 The Muskat problem
TheMuskatproblem,introdu edin[38 ℄,modelsthedynami sofanevolvingmaterial inter-fa eseparating two uids owing through aporous medium, i.e. a medium onsistingof a solidmatrix withuid-lled pores. Porousmedia owismodelled byDar y'slaw
µ
β
u = −∇p − (0, gρ)
T
,
(1.1) where
µ
is the vis osity of the uid,ρ
denotes the density,u
is the in ompressible uid velo ity,andp
isthepressure fun tion;additionally,β > 0
denotes thepermeability ofthe solid matrix, andg
is the a eleration due to gravity, whi h we shall hen eforth set to1
. Dar y's law (1.1) is an empiri al relation between momentum and for e (see, for example, [3, 39℄), and repla es onservation of momentum, whi h is used to model theevolution of invis id uidows.Thepurposeofthispaperistostudytheevolutionofaninterfa emovingthroughporous mediawithadis ontinuouspermeability. Asthepermeabilitytakestwodierentvalues,this ase is known in the literature as theinhomogeneous Muskat problem. Spe i ally, we are interested in the well-posedness and de ay to equilibrium for the inhomogeneous Muskat problem.
Γ
bot
Ω
−
β
−
Ω
+
(t)
β
+
µ
−
,
ρ
−
Vacuum
Γ
perm
Γ
(t)
β
+
Figure 1: The solid urve (blue) is the interfa e
Γ(t)
and the dashed urve (red)denotes the interfa eΓ
perm
,a rosswhi hthepermeabilityisdis ontinuous.We let
S
1
denote the ir le, so that fun tions
h : S
1
→ R
are identied with
[−π, π]
-periodi fun tionsonR
. AsshowninFigure1,we onsidera porousmedium o upying an opentime-dependent subsetΩ(t) ⊂ S
1
× R
su hthat
where
Ω
+
(t) = {(x
1
, x
2
) ∈ S
1
× R, −1 + f (x
1
) < x
2
< h(x
1
, t)} ,
(1.2a)Ω
−
= {(x
1
, x
2
) ∈ S
1
× R, −2 < x
2
< −1 + f (x
1
)} ,
(1.2b)Γ
perm
= {(x
1
, −1 + f (x
1
)), x
1
∈ S
1
} ,
(1.2 )and where the fun tions
f
andh
satisfymin
x
1
∈S
1
f (x
1
) > −1
andh(x
1
, 0) > −1 + f (x
1
).
(1.3)The xed-in-time permeability interfa e
Γ
perm
denotes the urve, a ross whi h the perme-abilityfun tionβ(x)
isdis ontinuous; spe i ally,thepermeabilityfun tionβ(x)
isdened asβ(x) =
β
+
inΩ
+
(t)
β
−
inΩ
−
,
forgiven onstants
β
±
> 0
. Thedomainfor thisproblemisalsoanunknown;thus,we must tra k theevolution of the time-dependent interfa e or free-boundaryΓ(t)
,whi h isdened asthesetΓ(t) = {(x
1
, h(x
1
, t)), x
1
∈ S
1
} .
For simpli ity, we shall set the uid density
ρ
and vis osityµ
to1
. As the uid is in ompressible, itfollows that[[u · n
perm
]] = 0
onΓ
perm
× [0, T ] .
With the domains dened,the Muskat problem onsistsof thefollowing systemof ou-pledequations:
u
±
β
±
+ ∇p
±
= −e
2
,
inΩ
±
(t) × [0, T ] ,
(1.4a)∇ · u
±
= 0,
inΩ
±
(t) × [0, T ] ,
(1.4b)[[p]] = 0
onΓ
perm
× [0, T ],
(1.4 )[[∇p · n
perm
]] = −
hh 1
β
ii
u
+
· n
perm
onΓ
perm
× [0, T ],
(1.4d)p
+
= 0
onΓ(t) × [0, T ],
(1.4e)V(Γ(t)) = u
+
· n
onΓ(t) × [0, T ] ,
(1.4f)u
−
· e
2
= 0
onΓ
bot
× [0, T ],
(1.4g )where
V(Γ(t))
denotes the normal omponent of the velo ity of the time-dependent free-boundaryΓ(t)
,n
isthe (upward) unitnormal toΓ(t)
,n
perm
is the(upward-pointing) unit normaltoΓ
perm
,and[[f ]] = f
+
− f
−
denotes thejumpof adis ontinuous fun tion
f
a ross1.2 A brief history of the analysis of the Muskat problem
Dar y'slaw(1.1 ) isastandardmodelforowinaquifers,oilwells,orgeothermal reservoirs, and it is therefore of pra ti al importan e in geos ien e (see, for example, [9, 28℄ and the referen es therein). Furthermore, the Muskat problem is equivalent to the Hele-Shaw ell problemwithgravity (see[36 ℄)for owbetween twothinly-spa ed parallelplates.
There hasabeenagreatdealofmathemati alanalysisofboththeMuskatproblemand theHele-Shaw ell, andweshallonly reviewasmallfra tionoftheresultsthatare,insome sense,most loselyrelatedto our result.
For theMuskatproblemwitha ontinuouspermeability fun tion,existen eofsolutions in the Sobolev spa e
H
3
has been established by Córdoba & Gan edo [18 , 19 ℄, Córdoba, Córdoba&Gan edoin[17 ℄,andCórdoba,Granero-Belin hón&Orive[22 ℄,usingthesingular integral ontour equation for the height fun tion
h
. Cheng,Granero-Belin hón &Shkoller [12 ℄ introdu ed the dire t PDE approa h (modied for use, herein), and established anH
2
existen e theory (see also Cheng, Coutand & Shkoller [11℄ for a similar approa h to the horizontal Hele-Shaw ell problem). This was followed by an
H
2
existen e theory by Constantin, Gan edo,Shvydkoy&Vi ol[16 ℄usingthesingularintegralapproa h;theyalso obtained a nite-slopeglobal existen e result. Very re ently, lo alexisten e in
H
s
,
s > 3/2
hasbeen obtained by Matio [37℄. Inthepresen e ofsurfa e tension, lo al existen einH
6
wasalsoobtained byAmbrose [1,2℄.
In the aseof adis ontinuous permeability fun tion (with a jump a rossthe at urve
(x
1
, −1)
), the lo al-in-time existen e of solutions has been proved by Berselli, Córdoba & Granero-Belin hón [4℄. Inthe ase of two uids withdierent vis osities and densities and permeability fun tion with a jump given by an arbitrary smooth urve(f
1
(α), f
2
(α))
the lo al-in-time existen eofsolutions hasbeen establishedbyPernás-Castaño [40 ℄ .For the ase of a ontinuous permeability, there are a variety of results showing global existen e of strong solutions under ertain onditions on the initial data. In parti ular, Cheng,Granero-Belin hón &Shkoller[12 ℄ provedglobal existen eunder restri tionsonthe sizeof
kh
0
k
H
2
,whileCórdoba,Constantin,Gan edo&Strain[15℄andCórdoba,Constantin, Gan edo, Strain & Rodríguez-Piazza [14 ℄ proved global existen eunder restri tionson the sizeofk b
h
′
0
k
L
1
,whereh
ˆ
denotestheFouriertransform. Theglobalexisten eofweaksolution hasbeenprovedbyCórdoba,Constantin,Gan edo&Strain[15℄andGranero-Belin hón[35 ℄ for initial data satisfying restri tions onkh
0
k
˙
W
1,∞
andkh
0
k
W
1,∞
, respe tively. Note that the onditiononkh
0
k
L
∞
in [35℄isa onsequen e of having abounded porousmedia.Finitetimesingularitiesofturningtypeareknowntoo ur. Aturningwaveisasolution whi h startsasagraph, butthenturns-overand losesthegraphproperty. Theexisten eof su h waves in the Rayleigh-Taylor stable regime hasbeen established byCastro, Córdoba, Feerman,Gan edo&López-Fernández[8 ℄,Córdoba,Granero-Belin hón&Orive-Illera[22℄, Berselli,Córdoba &Granero-Belin hón [4℄ andGómez-Serrano &Granero-Belin hón [34 ℄.
interfa essu h that, afterturning, theinterfa e isno longeranalyti and, infa t,
lim sup
t→T
kz(t)k
C
4
= ∞,
for anite time
T > 0
.Gan edo & Strain [33℄ have shown that the nite-time splash and splat singularities (a self-interse tion of a lo ally smooth interfa e) annot o ur for the two-phase Muskat problem (see also Feerman, Iones u & Lie [32 ℄ and Coutand & Shkoller [27℄). However, in the ase of the one-phase Muskat problem, Castro, Córdoba, Feerman & Gan edo [7℄ provedthatthesplashsingularitiesmayo urwhileCórdoba&Pernás-Castaño [23℄showed thatsplat singularities annoto ur. See also Coutand&Shkoller[26℄ forsplashand splat singularities forthe 3-D Euler equationsand relatedmodels.
Let us also mention that several results for the multiphase Muskat problem have been obtained in the ompletely dierent framework of little Hölder spa es
h
k+α
by Es her & Matio [30 ℄,Es her, Matio &Matio [29℄and Es her, Matio &Walker[31 ℄.
Very re ently,a regularityresultinHölder spa es for theone-phase Hele-Shawproblem hasbeenobtainedbyChang-Lara&Guillén[10℄usingthehodographtransform. Also,Prüss &Simonett[41 ℄studiedthetwo-phaseMuskatprobleminamoregeometri frameworkusing theHanzawa transform. In parti ular,these authors showwell-posedness, hara terize and study thedynami stabilityoftheequilibria.
Finally,usinga onvexintegrationapproa h,Castro,Córdoba&Fara o[5℄veryre ently proved theexisten eofweaksolutionsfor the Muskatprobleminthe ase wherethedenser uidliesabovethelighteruid,so,itisintheRayleigh-Taylorunstableregime. Remarkably, these solutions develop a mixing zone (a strip ontaining parti les from both phases and, onsequently, withuidparti leshaving both densities), growing linearlyintime.
1.3 Methodology
Asnotedabove,mostpriorexisten etheoremshaverelieduponthesingularintegral ontour equation for the height fun tion
h
; in the ase of the innitely deep two-phase Muskat problemwith ontinuous permeability,theevolution equationforh
an be written ash
t
(x
1
) =
p.v.Z
R
h
′
(x
1
) − h
′
(x
1
− y)
y
1
1 +
h(x
1
)−h(x
1
−y)
y
2
dy;
(1.5)see, for example,[18℄ forthe derivation.
The ontour equation (1.5 ) depends ru ially on the geometry of the domain and the permeability fun tion. In parti ular, when the porous medium has nite depth (equal to
in[4 ℄ that(1.5 ) takesthe form:
h
t
(x
1
) =
β
+
(−[[ρ]])
4π
p.v.Z
R
(h
′
(x
1
) − h
′
(y)) sinh(x
1
− y)
cosh(x
1
− y) − cos(h(x
1
) − h(y))
dy
+
β
+
(−[[ρ]])
4π
p.v.Z
R
(h
′
(x
1
) − h
′
(y)) sinh(x
1
− y)
cosh(x
1
− y) + cos(h(x
1
) + h(y))
dy
+
1
4π
p.v.Z
R
̟
2
(y)(sinh(x
1
− y) + h
′
(x
1
) sin(h(x
1
) + 1))
cosh(x
1
− y) − cos(h(x
1
) + 1)
dy
+
1
4π
p.v.Z
R
̟
2
(y)(− sinh(x
1
− y) + h
′
(x
1
) sin(h(x
1
) − 1))
cosh(x
1
− y) + cos(h(x
1
) − 1)
dβ,
(1.6) where̟
2
(x
1
) =
β
+
− β
−
β
+
+ β
−
β
+
(−[[ρ]])
2π
p.v.Z
R
h
′
(y)
sin(1 + h(y))dy
cosh(x
1
− y) − cos(1 + h(y))
−
β
+
− β
−
β
+
+ β
−
β
+
(−[[ρ]])
2π
p.v.Z
R
h
′
(y)
sin(−1 + h(y))dy
cosh(x
1
− y) + cos(−1 + h(y))
+
β
+
−β
−
β
+
+β
−
2
√
2π
β
+
(−[[ρ]])
2π
G
β
∗
p.v.Z
R
h
′
(y) sin(1 + h(y))dy
cosh(x
1
− y) − cos(1 + h(y))
−
β
+
−β
−
β
+
+β
−
2
√
2π
β
+
(−[[ρ]])
2π
G
β
∗
p.v.Z
R
h
′
(y) sin(−1 + h(y))dy
cosh(x
1
− y) + cos(−1 + h(y))
,
(1.7) withG
β
(x
1
) = F
−1
F
cosh(x
sin(2)
1
)+cos(2)
(ζ)
1 +
β+−β−
β++β−
√
2π
F
sin(2)
cosh(x
1
)+cos(2)
(ζ)
,
a S hwartz fun tion, and where
F
denotes the Fourier transform. Let us emphasize that, duetothenon-lo al hara terof̟
2
givenby(1.7 ),the ontourequation(1.6)issigni antly more hallenging to analyse than (1.5 ). Note also, from the denition ofG
β
(x
1
)
that the highlynon-lo al onvolution termsin(1.7) arenot expli itlydened.Be ause of the ompli ations inherent in the singular integral approa h of (1.7 ), we shall instead analyze the system (1.4) dire tly. As (1.4) is set on the time-dependent a priori unknown domain
Ω(t)
, in order to build an existen e theory, we rst pull-ba k this systemof equations onto a xed-in-time spatialdomain. We usea arefully hosen hange-of-variables thattransformsthefree-boundary problem(1.4 ) into asystemof equationsset on asmooth andxed domain,but having time-dependent oe ients.Topull-ba k our problem, weemploya family ofdieomorphisms
ψ
±
whi h areellipti extensions of the interfa e parametrizations, and thus have optimal
H
s
Sobolev regularity. The time-dependent oe ients (in the pulled-ba k des ription) arise from dierentiation and inversion of these maps
ψ
±
; by studying the transformed Dar y's Law, we obtain a new higher-order energy integral that provides the regularity of the moving interfa e
Γ(t)
. Additionally,we obtainanL
2
gainfor solutionsto the heatequation, ex eptthatwe gaina
1/2
-derivative inspa erather than a full derivative. The regularity of the interfa eΓ(t)
as well as the improvedL
2
-in-time paraboli regularity gain are found from the non-linear stru ture of the pulled-ba k representation ofthe Muskatproblem. Inparti ular, wedonorelyontheexpli itstru ture of thesingular integral ontourequation, and assu h, we arefreeto study general domain geometriesand permeability fun tions.
1.4 The main results
Aswe willshow, theRayleigh-Taylor(RT)stability ondition, given by
−
∂p
∂n
> 0
onΓ(t)
,is a su ient ondition for well-posedness ofthe Muskat problem (1.4) inSobolev spa es. In parti ular, withp
0
:= p(·, 0)
andΓ := Γ(0) ,
and letting
N := n(·, 0)
denote theoutward unit normal toΓ
,we prove that for anyinitial interfa eΓ
ofarbitrarysize and of lassH
2
, hosen su hthatthe RTstability ondition
−
∂p
∂N
0
> 0
onΓ
(1.8)issatised,thereexistsauniquesolution
(u
±
(x, t), p
±
(x, t), h(x
1
, t))
totheone-phaseMuskat problemwithdis ontinuous permeabilityfun tion.Morepre isely,we prove the following Theorem 1 (Lo al well-posedness in
H
2
). Suppose the initial interfa e
Γ
isgiven as the graph(x
1
, h
0
(x
1
))
whereh
0
∈ H
2
(S
1
)
and
R
S
1
h
0
(x
1
)dx
1
= 0
,andthattheRT ondition (1.8) is satised. LetΓ
perm
be given as the graph(x
1
, −1 + f (x
1
))
for a fun tionf ∈ H
2.5
(S
1
)
. Assumealso that (1.3 ) holds. Then,there exists a time
T (h
0
, f ) > 0
anda unique solutionh ∈ C([0, T (h
0
, f )]; H
2
(S
1
)) ∩ L
2
(0, T (h
0
, f ); H
2.5
(S
1
)) ,
u
±
∈ C([0, T (h
0
, f )]; H
1.5
(Ω
±
(t))) ∩ L
2
(0, T (h
0
, f ); H
2
(Ω
±
(t))) ,
p
±
∈ C([0, T (h
0
, f )]; H
2.5
(Ω
±
(t))) ∩ L
2
(0, T (h
0
, f ); H
3
(Ω
±
(t))) ,
tothe system (1.4 ),satisfying
kh(t)k
2
L
2
(S
1
)
+ 2
Z
t
0
u
+
(s)
β
+
2
L
2
(Ω
+
(s))
ds + 2
Z
t
0
u
−
(s)
β
−
2
L
2
(Ω
−
)
ds = kh
0
k
2
L
2
(S
1
)
,
andkhk
C([0,T (h
0
,f )],H
2
(S
1
))
+ kh
t
k
L
2
(0,T (h
0
,f );H
1.5
(S
1
))
+ khk
L
2
(0,T (h
0
,f );H
2.5
(S
1
))
+ kpk
C([0,T (h
0
,f )],H
2.5
(Ω
+
(t)∪Ω
−
))
+ kpk
L
2
(0,T (h
0
,f );H
3
(Ω
+
(t)∪Ω
−
))
+ kuk
C([0,T (h
0
,f )],H
1.5
(Ω
+
(t)∪Ω
−
))
+ kuk
L
2
(0,T (h
0
,f );H
2
(Ω
+
(t)∪Ω
−
))
≤ C(h
0
, f )
Remark 1. Itis easyto see thatif
h
0
(x
1
) = f (x
1
) = 0
,thesolution isgiven byu
±
(x, t) = 0, h(x
1
, t) = 0, p
±
(x, t) = −x
2
,
(1.9)and theRT ondition is satised. There exist innitelymany initial data
h
0
satisfyingthe RT ondition; for example, small perturbations of (1.9) satisfy the RT ondition (1.8 ) via impli it fun tiontheorem arguments.Theorem 2 (Global well-posedness and de ayto equilibrium in
H
2
). Suppose the initial interfa e
Γ
isgiven as the graph(x
1
, h
0
(x
1
))
whereh
0
∈ H
2
(S
1
)
and
R
S
1
h
0
(x
1
)dx
1
= 0
. LetΓ
perm
be given asthe graph(x
1
, −1)
. Then, there existsa onstantC
su h that if|h
0
|
2
< C ,
the RT ondition (1.8) is satised and there exists a unique solution
h ∈ C([0, ∞); H
2
(S
1
)) ∩ L
2
(0, ∞; H
2.5
(S
1
)) ,
u
±
∈ C([0, ∞); H
1.5
(Ω
±
(t))) ∩ L
2
(0, ∞; H
2
(Ω
±
(t))) ,
p
±
∈ C([0, ∞); H
2.5
(Ω
±
(t))) ∩ L
2
(0, ∞; H
3
(Ω
±
(t))) ,
tothe system (1.4 ),satisfying
kh(t)k
H
2
(S
1
)
≤ kh
0
k
H
2
(S
1
)
e
−γt/2
for a onstant
γ(h
0
, β
±
)
, whi h dependsonh
0
andβ
±
.Remark 2. Notethatthe questionofwhetherthefreeboundary
Γ(t)
anrea hthe urveΓ
perm
innitetime, ina asituation thatresembles thesplash/splat singularity,remains an open problem. In fa t, su h behavior an be seen as a singular phenomena (for instan e, some ofthe (non-singular) termsin(1.6 ) and(1.7 ) be ome singular integraloperators). As Theorem2impliesthatΓ(t)
annotrea hthe urveΓ
perm
innitetimeifh
0
issmallenough, this resultrules out the possibilityof interfa e ollisioninnite timefor smallinitial data. Remark 3. Wenotethatthedryzone (theregionwithoutuid)liesabovethe urveΓ(t)
, and so, as long as (1.3) holds, the dry zone lies aboveΓ
perm
. The question of whether a dry zone an forminsideΩ
−
remains an open problem. In other words, assumethat there exists a solution
h(x
1
, t)
up to timeT
and assume also thatΓ(t)
interse tsΓ
perm
at the point(x
0
, t
′
) ∈ S
1
× (0, T )
,i.e.
h(x
0
, t
′
) = −1 + f (x
0
).
Then, it isnot lear ifthe urve
Γ(t)
may ross the urveΓ
perm
,i.e.h(x
1
, t) < −1 + f (x
1
), ∀ (x
1
, t) ∈ (x
0
− ǫ, x
0
+ ǫ) × (t
′
, t
′
+ δ), ,
for ertain
ǫ, δ > 0
. Notealso that,ifthis happens,thenthe region{(x
1
, x
2
), x
1
∈ (x
0
− ǫ, x
0
+ ǫ), h(x
1
, t) < x
2
< −1 + f (x
1
)} ⊂ Ω
−
Remark 4. The exponential de ay of the solution
h(t)
is a onsequen e of an energy-energydissipation inequality establishingarelationshipbetweentheinterfa eregularityand theregularityof the semi-ALE velo ity(see Se tions5and 6):kh
′′
(t)k
L
2
(S
1
)
≤ Ckw
′′
k
L
2
(S
1
×(−2,−1)∪S
1
×(−1,0))
.
Remark 5. Note that the linearized evolution equation for a small perturbation of the at interfa e an be written as
h
t
= −Λ
D
+
h,
where
Λ
D
+
is theDiri hlet-to-Neumann map asso iated withtheellipti system(5.1 ):Λ
D
+
h(x
1
) = δψ,
2
(x
1
, 0).
An integration byparts shows that
Z
S
Z
0
−1
δψ
+
∆δψ
+
dx
2
dx
1
=
Z
S
δψ
+
(x
1
, 0)δψ
+
,
2
(x
1
, 0)dx
1
−
Z
S
δψ
+
(x
1
, −1)δψ
+
,
2
(x
1
, −1)dx
1
−
Z
S
Z
0
−1
|∇δψ
+
|
2
dx
2
dx
1
,
sothat,Z
Γ
Λ
D
+
hhdx
1
=
Z
D
+
|∇δψ
+
|
2
dx.
We also have the following Poin aré-Wirtinger inequality
Z
D
+
|δψ
+
(x)|
2
dx =
Z
D
+
Z
1
0
δψ
+
,
2
(x
1
, sx
2
+ (1 − s)(−1))(x
2
+ 1)ds
2
dx
≤
Z
S
Z
0
−1
Z
1
0
δψ
+
,
2
(x
1
, sx
2
+ (1 − s)(−1))
2
(x
2
+ 1)
2
dsdx
2
dx
1
=
Z
0
−1
(x
2
+ 1)
Z
S
Z
x
2
−1
δψ
+
,
2
(y)
2
dy
2
dy
1
dx
2
≤
Z
0
−1
(x
2
+ 1)
Z
D
+
δψ
+
,
2
(y)
2
dydx
2
≤
1
2
Z
D
+
δψ
+
,
2
(y)
2
dy.
Thus,using the tra etheorem, we on lude that
Z
Γ
Λ
D
+
hhdx
1
≥ 0.5kδψ
+
k
2
1,+
≥ ν|h|
2
0.5
≥ ν|h|
2
0
,
1.5 Notation used throughout the paper For amatrix
A
,we writeA
i
j
forthe omponentofA
lo atedinrowi
and olumnj
. Weuse theEinstein summation onvention, whereinrepeated indi esaresummed from1
to2
. We denotethej
th anoni albasisve tor inR
2
by
e
j
. Fors ≥ 0
,we setkuk
s,+
:= ku
+
k
H
s
(D
+
)
, kuk
s,−
:= ku
−
k
H
s
(D
−
)
, kuk
s,±
:= ku
+
k
s,+
+ ku
−
k
s,−
and
|h|
s
:= khk
H
s
(Γ)
.
For fun tions
h
dened onΓ
perm
,weshallalso denotetheH
s
norm by
|h|
s
:= khk
H
s
(Γ
perm
)
, wheneverthe ontext is lear.Wewrite
f
′
=
∂ f
∂ x
1
, f,
k
=
∂f
∂x
k
,
andf
t
=
∂f
∂t
.
For a dieomorphismψ
,weletA = (∇ψ)
−1
,and denecurl
ψ
v = A
j
1
v
2
,
j
−A
j
2
v
1
,
j
,
(1.10)div
ψ
v = A
i
j
v
j
,
i
.
(1.11)2 The Muskat problem in the ALE formulation 2.1 Constru ting the family of dieomorphisms
ψ
(·, t)
2.1.1 The idea for the onstru tionOuranalysisoftheMuskatproblem(1.4 )isfoundedonatime-dependent hange-of-variables whi h onverts thefreeboundaryproblemto one set onsmooth referen edomains
D
±
D
+
= S
1
× (−1, 0) , D
−
= S
1
× (−2, −1) ,
(2.1)The boundaries ofthedomains
D
±
aredened asΓ
bot
= {(x
1
, −2), x
1
∈ S
1
} , Γ
perm
= {(x
1
, −1), x
1
∈ S
1
} ,
andΓ = {(x
1
, 0), x
1
∈ S
1
} .
(2.2) We let
N = e
2
denotethe unitnormal ve toronΓ
(outwards),Γ
perm
andΓ
bot
.Asour analysis ru iallyrelies onobtaining aparaboli regularity gain,we need a refer-en e domain
D
+
with
C
∞
boundary. In parti ular, the initial domainΩ
+
(0)
annot serve asareferen e domain.
Weadapttheideas from[12 ℄to onstru tthetime-dependent familyofdieomorphisms with optimal Sobolev regularity,
ψ(x, t)
,that we shalluse to pull-ba k (1.4) onto thexed domainD
±
. Before detailing this onstru tion, let us sket h the pro edure. First, we onstru ta dieomorphismwith optimalSobolev regularityat
t = 0
:ψ
+
(0) : D
+
→ Ω
+
(0), ψ
−
: D
−
→ Ω
−
.
•
For0 < δ ≪ 1
a su iently small parameter (to be xed later), we dene auxiliary domains,D
±,δ
(0)
. Theseauxiliarydomainsare onstru tedviamolli ationof
h(x
1
, 0)
andf (x
1
)
and,thus,they areinnitelysmooth. We dene thegraph dieomorphismφ
±
1
: D
±
→ D
±,δ
.
Thesedieomorphisms areof lass
C
∞
be auseofthesmoothnessof thedomainsD
±
andD
±,δ
.
•
Weneed anotherdieomorphismsfromthe auxiliarydomainD
±,δ
to
Ω
±
(0)
. Weneed these dieomorphisms to gain1/2 derivatives with respe t totheregularity of
Ω
±
(0)
. Inorderthatthisoptimalregularityisobtained,we makeuseofthedenitionofD
±,δ
and the propertiesof ourmolliers. We dene
φ
±
2
: D
±,δ
→ Ω
±
(0)
asthesolution toLapla e problems withappropriate boundary onditions. Usingthe boundary dataand the inverse fun tion theorem, these mappings
φ
±
2
areH
2.5
−
lass dieomorphisms.•
Finally,wedeneψ
+
(0) = φ
+
2
◦ φ
+
1
, ψ
−
= φ
−
2
◦ φ
−
1
.
As ompositionof dieomorphisms,ψ
±
(0)
is adieomorphism.On ethe initialdieomorphismwithoptimalSobolevregularityis onstru ted,we solve Poissonproblems(tobedetailedbelow)for
ψ
±
(x, t)
. Anappli ation oftheinverse fun tion theorem together with standard ellipti estimates will show that these mappingsψ
±
(x, t)
area familyof dieomorphisms withthedesiredsmoothness.
2.1.2 Constru ting the initial regularizing dieomorphism
ψ(·, 0)
Given a fun tionh ∈ C(0, T ; H
2
)
withinitial data
h(x
1
, 0) = h
0
(x
1
)
,we x0 < δ ≪ 1
and dene our auxiliary domainsand boundariesD
+,δ
(0) = {(x
1
, x
2
), x
1
∈ S
1
, −1 + J
δ
f (x
1
) < x
2
< J
δ
h
0
(x
1
)},
D
−,δ
= {(x
1
, x
2
), x
1
∈ S
1
, −2 < x
2
< −1 + J
δ
f (x
1
)},
Γ
δ
(0) = {(x
1
, J
δ
h
0
(x
1
)), x
1
∈ S
1
}, Γ
δ
perm
= {(x
1
, −1 + J
δ
f (x
1
)), x
1
∈ S
1
}.
Aswesaid previously,we dene thegraph dieomorphism
φ
+
1
(x
1
, x
2
) = (x
1
, (x
2
+ 1)J
δ
h
0
(x
1
) − (−1 + J
δ
f (x
1
))x
2
) ,
φ
−
1
(x
1
, x
2
) = (x
1
, x
2
+ J
δ
f (x
1
)(x
2
+ 2)) ,
where
J
δ
denotes the onvolution withastandard Friedri h's mollier. This fun tionisa
C
∞
dieomorphism.Next, we have to dene the regularizing dieomorphisms
φ
±
2
: D
±,δ
(0) → Ω
±
(0).
We dene these mappingsasthe solution to thefollowing ellipti problems:
∆φ
+
2
= 0
inD
+,δ
(0) ,
(2.3a)φ
+
2
= (x
1
, x
2
) + [h
0
(x
1
) − J
δ
h
0
(x
1
)]e
2
onΓ
δ
(0) ,
(2.3b)φ
+
2
= (x
1
, x
2
) + [f (x
1
) − J
δ
f (x
1
)]e
2
onΓ
δ
perm
,
(2.3 )∆φ
−
2
= 0
inD
−,δ
,
(2.4a)φ
−
2
= (x
1
, x
2
) + [f (x
1
) − J
δ
f (x
1
)]e
2
onΓ
δ
perm
(2.4b)φ
−
2
= (x
1
, x
2
)
onΓ
bot
.
(2.4 )Using standard ellipti regularitytheory,wehave that
kφ
2
− ek
H
2.5
(D
±
,δ
)
≤ C(|h
0
− J
δ
h
0
|
2
+ |f − J
δ
f |
2
),
where
e = (x
1
, x
2
)
denotes the identity mapping. Using the Sobolev embedding theorem, and takingδ > 0
su iently small,we have thatkφ
2
− ek
C
1
(D
±
,δ
)
≪ 1,
so,due to the inverse fun tiontheorem, we obtain that
φ
±
2
is anH
2.5
- lass dieomorphism. Asin[12 ℄, we deneψ
+
(0) = φ
+
2
◦ φ
+
1
: D
+
→ Ω
+
(0), ψ
−
= φ
−
2
◦ φ
−
1
: D
−
→ Ω
−
.
(2.5)Then, this mapping isalso an
H
2.5
- lassdieomorphism.
2.1.3 Constru tingthe time-dependent family of regularizing dieomorphisms
ψ(·, t)
Wedenethetime-dependentfamilyofdieomorphisms
ψ(t) = ψ(·, t)
assolutionstoPoisson equationswithfor ingdependingonψ(0)
. Themainpointofthis onstru tionisthatdueto the ontinuityintime oftheinterfa eh
andstandardellipti estimates,thetime-dependent familyofdieomorphismsψ(t) = ψ(·, t)
isgoingtoremain losetotheinitialdieomorphismψ(0)
.In parti ular,we onsider thefollowing ellipti system:
∆ψ
+
(t) = ∆ψ
+
(0)
inD
+
× [0, T ] ,
(2.6a)ψ
+
(t) = (x
1
, x
2
) + h(x
1
, t)e
2
onΓ × [0, T ] ,
(2.6b)and
ψ
−
(t) = ψ
−
. Be auseofthefor ingterm present in(2.6 a), wehavethatψ
+
(t) − ψ
+
(0)
solves∆(ψ
+
(t) − ψ
+
(0)) = 0
inD
+
× [0, T ] ,
ψ
+
(t) − ψ
+
(0) = (h(x
1
, t) − h(x
1
, 0))e
2
onΓ × [0, T ] ,
ψ
+
(t) − ψ
+
(0) = 0
onΓ
perm
× [0, T ] .
Due toellipti estimates,we havethebound
kψ(t) − ψ(0)k
2.25,±
≤ C|h(t) − h
0
|
1.75
.
(2.8)By takingsu iently small time
t
and re alling thath ∈ C(0, T ; H
2
)
,we have thatkψ(t)k
2.25,±
≤ C|h(t) − h
0
|
1.75
+ C(|h
0
|
1.75
+ |f |
1.75
+ 1) ≤ 2C(|h
0
|
1.75
+ |f |
1.75
+ 1).
WritingJ(t) =
det(∇ψ(t)) = ψ
1
,
1
ψ
2
,
2
−ψ
2
,
1
ψ
1
,
2
,
wehave the bound
kJ(t) − J(0)k
1.25,±
≤ C|h(t) − h
0
|
1.75
.
(2.9)Consequently,using
h ∈ C(0, T ; H
2
)
,for su iently smalltime
t
,we have thatmin
x∈D
±
J(0)
2
< J(t) < 2 max
x∈D
±
J(0),
and,thanksto (2.8) ,wehave that
kψ(t) − ψ(0)k
C
1
≤ C|h(t) − h
0
|
1.75
≪ 1.
Due to the inverse fun tion theorem and using thefa t that
ψ(0)
is a dieomorphism, we seethatψ
±
(t) : D
±
→ Ω
±
(t)
isa dieomorphism. Fromthe ellipti estimate
kψ(t)k
2.5,±
≤ C(|h(t)|
2
+ |f |
2
+ 1),
wehave that
ψ(t)
is anH
2.5
- lass dieomorphism.
2.1.4 The matrix
A(·, t)
We writeA = (∇ψ)
−1
. Thus,A
i
r
ψ
r
,
j
= δ
j
i
and we obtaintheuseful identities
(A
t
)
i
k
= −A
i
r
(ψ
t
)
r
,
j
A
j
k
,
A
′′
= −2A
′
∇ψ
′
A − A∇ψ
′′
A.
(2.10)We will also make useof thePiola'sidentity:
(JA
k
2.2 The Muskat problem in the referen e domains
D
±
With
ψ(t) = ψ(·, t)
dened in Se tion 2.1, we dene our new variables in the referen e domainsD
±
:v = u ◦ ψ, q = p ◦ ψ
. Welet˜
τ = ψ
′
, ˜
n = (ψ
′
)
⊥
, g = |ψ
′
|
2
denote the (non-normalized) tangent and normal ve tors and the indu ed metri , respe -tively,on
Γ(t)
. Wealsodene theunittangentve torτ = ˜
τ /√g
andtheunitnormalve torn = ˜
n/√g
. In the same way, we deneτ
˜
perm
, ˜
n
perm
, g
perm
, τ
perm
, n
perm
as the analogous quantities onΓ
perm
. Re allthatJA
k
i
N
k
= ˜
n
i
onΓ, JA
k
i
N
k
= ˜
n
i
perm
onΓ
perm
.
Hen e,theALErepresentationoftheone-phaseinhomogeneousMuskatproblemisgiven by
(v
±
)
i
β
±
+ (A
±
)
k
i
(q
±
+ ψ
±
· e
2
),
k
= 0
inD
±
× [0, T ] ,
(2.11a)(A
±
)
k
i
(v
±
)
i
,
k
= 0
inD
±
× [0, T ] ,
(2.11b)h
t
(t) = (v
+
)
i
J
+
(A
+
)
j
i
N
j
onΓ × [0, T ] ,
(2.11 )q
+
= 0
onΓ × [0, T ] ,
(2.11d)[[q]] = 0
onΓ
perm
× [0, T ] ,
(2.11e)[[q,
k
A
k
i
JA
j
i
N
j
]] = −
hh 1
β
ii
v
i
JA
j
i
N
j
onΓ
perm
× [0, T ]
(2.11f)v
−
2
= 0
onΓ
bot
× [0, T ] .
(2.11g ) 3 A priori estimatesIn this se tion we establish the a priori estimates for theone-phase Muskat problem with dis ontinuouspermeability (1.4).
Wedene the higher-order energy fun tion
E(t) = max
0≤s≤t
|h(s)|
2
2
+
Z
t
0
kv(s)k
2
2,±
+ |h(s)|
2
2.5
ds.
Remark 6. Anotherpossible denitionfor ahigher-order energy fun tionis (see[12℄)
E(t) = max
0≤s≤t
|h(s)|
2
2
+
Z
t
0
kv(s)k
2
2,±
.
Infa t, aswill be shown,
As in[12 ℄, our goalisto obtain thepolynomial inequality
E(t) ≤ M
0
+ Q(E(t))t
α
,
for ertain
α > 0
,ageneri polynomialQ
,anda onstantM
0
dependingonh
0
andf
. WhenE(t)
is ontinuous, theprevious inequality impliestheexisten eofT
∗
(h
0
, f )
su hthat
E(t) ≤ 2M
0
.
(3.1)We assume that we have a smooth solution dened for
t ∈ [0, T ]
. We take0 < T ≤ 1
small enough su h that the following onditions hold: for a xed onstant0 < ǫ ≪ 1
(possiblydependingonh
0
andf
) andfort ∈ [0, T ]
,kψ(t) − ψ(0)k
L
∞
+ kA(t) − A(0)k
L
∞
+ kJ(t) − J(0)k
L
∞
≤ ǫ ;
(3.2a)kh(t) − h
0
k
L
∞
+ k∇q(t) − ∇q(0)k
L
∞
≤ ǫ ;
(3.2b)E(t) ≤ 3M
0
;
(3.2 )min
0≤t≤T
x
min
1
∈S
1
q,
2
(t) ≥ min
x
1
∈S
1
q,
2
(0)/4.
(3.2d)Wewillshowthat onditionsevenstri terthan(3.2 ,d)a tuallyholds. Letusemphasize that, due tothe RT ondition, we have that
min
x
1
∈S
1
q,
2
(0) > 0.
Again, we let
C = C(h
0
, f, δ)
denote a onstant thatmay hange fromline to line. We letP(x)
denoteapolynomialwith oe ientsthatmaydependonh
0
(·) := h(·, 0), f, δ
. This polynomialmay hange fromline to line.3.1 Estimates for some lower-order norms
In thefollowing, we olle t some estimates of lower-order norms. Theproofs aresimilar to those in[12 ℄,so,we omit them.
Lemma3 (Estimatesforsomelower-ordernormsof
h
,[12 ℄,Se tion8.4.1). Givena smooth solutionto the Muskat problem (2.11a-g),Z
t
0
|h
t
|
2
1
ds ≤ C E(t).
(3.3a)|h(t) − h
0
|
1
≤ C
p
E(t)t
1/2
.
(3.3b)Lemma 4 (Estimates for some lower-order norms of the ALE mapping
ψ
, [12 ℄, Se tion 8.4.2). Givena smooth solutionto the Muskatproblem(2.11a-g),kψ(t)k
2.5,±
≤ C(1 + |h(t)|
2
), kψ(t)k
3,±
≤ C(1 + |h(t)|
2.5
)
(3.4a)kψ(t) − ψ(0)k
2.25,±
+ kA(t) − A(0)k
1.25,±
+ kJ(t) − J(0)k
1.25,±
≤
4
√
tC
p
E(t).
(3.4b) Noti e that(3.4b) implies astri ter versionof (3.2 a). Asa onsequen e we obtain that|h(t)|
1.75
, kψ(t)k
2.25,±
, kJ(t)k
1.25,±
andkA(t)k
1.25,±
are bounded byC(h
0
, f )
uniformly for allt ∈ [0, T ]
. Furthermore,wealso havethat0 <
1
2
x∈D
min
+
∪D
−
J(0) ≤ J(t) ≤ 1.5
max
x∈D
+
∪D
−
J(0).
3.2 Basi
L
2
energy law
Lemma5(Estimatesforsomelower-ordernormsof
v
). ForasmoothsolutiontotheMuskat problem (2.11a-g),|h(t)|
2
0
+ 2
Z
t
0
s
J
β
v
2
0,±
ds = |h
0
|
2
0
.
(3.6)Proof. We test the equation (2.11 a) against
Jv
and integrate. Using Piola's identity, inte-grating by partsand using thedivergen e free ondition (2.11b),we obtain thats
J
β
v
2
0,±
+
Z
Γ
v
i
JA
k
i
(q + ψ · e
2
)N
k
dx
1
−
Z
Γ
perm
[[v
i
JA
k
i
(q + ψ · e
2
)N
k
]]dx
1
−
Z
Γ
bot
v
i
JA
k
i
(q + ψ · e
2
)N
k
dx
1
= 0.
Then, using the jumpand boundary onditions on
Γ
perm
andΓ
bot
,s
J
β
v
2
0,±
+
1
2
d
dt
|h|
2
0
= 0.
3.3 Estimates forh
∈ L
2
(0, T ; H
2.5
(Γ))
andh
t
∈ L
2
(0, T ; H
1.5
(Γ))
From (2.11 a)(v
i
+ β
+
δ
i
2
)τ
i
= 0
onΓ,
andv
′
i
τ
i
= −β
+
JA
2
i
A
2
i
q,
2
h
′′
g
3/2
onΓ .
(3.7)Lemma6(Paraboli smoothing,[12℄,Se tion8.4.6). GivenasmoothsolutiontotheMuskat problem (2.11a-g),
h ∈ C([0, T ], H
2
(S
1
)).
In parti ular,Z
t
0
|h
t
(s)|
2
1.5
ds +
Z
t
0
|h(s)|
2
2.5
ds ≤ C
max
0≤s≤t
|h(s)|
2
2
+
Z
t
0
kv(s)k
2
2,±
ds
.
(3.8)3.4 Pressure estimates Using (2.11 a)and(2.11 b),
q
±
solves−(β
±
J
±
(A
±
)
j
i
(A
±
)
k
i
q
±
,
k
),
j
= 0
inD
±
,
(3.9)q
+
= 0
onΓ ,
(3.10)[[q]] = 0
onΓ
perm
,
(3.11)[[βq,
k
A
k
i
JA
r
i
N
r
]] = −[[β]]δ
2
i
JA
r
i
N
r
onΓ
perm
,
(3.12)β
−
q
−
,
k
(A
−
)
k
i
J
−
(A
−
)
r
i
N
r
= −β
−
onΓ
bot
.
(3.13)Wehave that
A(0)A(0)
T
issymmetri and positivedenite:
[A(0)A(0)
T
]
i
j
ξ
i
ξ
j
≥ L|ξ|
2
;
onsequently, dueto (3.4 b),
kA
0
A
T
0
− A(t)A
T
(t)k
L
∞
≤ C
√
t
p
E(t) ,
and we seethatfor
t
su iently small,L
2
|ξ|
2
≤ [A(·, t)A
T
(·, t)]
i
j
ξ
i
ξ
j
≤ 2L|ξ|
2
.
Thus,A(t)A
T
(t)
form a uniformly ellipti operator for
t
on[0, T ]
, and ellipti estimates (following the same approa h asin[12 ℄ and[13℄) lead tokqk
2.5,±
≤ C
p
E(t) , kv(t)k
1.5,±
≤ C
p
E(t).
(3.14)Furthermore,using thesame argument asin[12 , Se tion 8.4.5℄, we obtainthat
kq(t) − q(0)k
2.25,±
≤ t
1/8
P(E(t)),
(3.15)and,
kq,
2
(t) − q,
2
(0)k
L
∞
(Γ)
≤ C|q,
2
(t) − q,
2
(0)|
0.75
≤ t
1/8
P(E(t)).
Asa onsequen e of thelatterinequality, theRayleigh-Taylor sign ondition holds in
[0, T ]
for small enoughT
. Furthermore, usingtheSobolev embeddingtheorem,k∇q(t) − ∇q(0)k
L
∞
≤ t
1/8
P(E(t)),
and astronger versionofthe bootstrapassumption (3.2 d)also holds.
3.5 The energy estimates
In this se tion we will perform the basi energy estimates. Integrals of lower-order terms will bedenoted by
R(t)
,meaning thatZ
t
0
R(s)ds ≤ M
0
We take two horizontal derivatives of (2.11a), test against
Jv
′′
, and integrate by parts to ndthatZ
t
0
Z
D
+
∪D
−
J
β
|v
′′
|
2
dxds+
Z
t
0
Z
D
+
∪D
−
J
h
A
k
i
(q + ψ
2
)
′′
,
k
+(A
k
i
)
′′
(q + ψ
2
),
k
i
v
′′
i
dxds+
Z
t
0
R(s)ds = 0.
Dueto the divergen e free ondition(2.11 b)we obtain that
A
k
i
(v
i
)
′′
,
k
= −(A
′′
)
k
i
v
i
,
k
+R(t).
(3.16)Thus,integratingbyparts andusing (3.16 )andtheidentities
JA
k
i
N
k
= ˜
n
i
, JA
k
i
N
k
= ˜
n
i
perm
and
JA
k
i
N
k
= N
i
onΓ
,Γ
perm
andΓ
bot
,respe tively,we nd thatI
1
=
Z
t
0
Z
D
+
∪D
−
JA
k
i
(q + ψ
2
)
′′
,
k
v
′′
i
dxds
=
Z
t
0
Z
Γ
˜
n
i
(q + ψ
2
)
′′
v
′′
i
dx
1
ds −
Z
t
0
Z
Γ
perm
[[v
i
′′
(q + ψ
2
)
′′
n
˜
i
perm
]]dx
1
ds
−
Z
t
0
Z
Γ
bot
N
i
(q + ψ
2
)
′′
v
i
′′
dx
1
ds −
Z
t
0
Z
D
+
∪D
−
JA
k
i
(q + ψ
2
)
′′
(v
i
),
′′
k
dxds
=
Z
t
0
Z
D
+
∪D
−
J(A
′′
)
k
i
(q + ψ
2
)
′′
(v
i
),
k
dxds +
Z
t
0
Z
Γ
˜
n
i
h
′′
v
′′
i
dx
1
ds
−
Z
t
0
Z
Γ
perm
(q
+
+ f )
′′
[[v
′′
i
n
˜
i
perm
]]dx
1
ds +
Z
t
0
R(s)ds.
The2-Dintegralisnowalower-orderterm that an beestimatedwitha
L
2
− L
4
− L
4
− L
∞
Hölder argument togetherwiththeSobolevembeddingtheorem. Thus,we areleftwiththe integrals on the boundaries
Γ
andΓ
perm
. Due to the in ompressibility ondition, we have that[[v
i
n
˜
i
perm
]] = 0
,sothatI
1
=
Z
t
0
Z
Γ
h
′′
h
′′
t
dx
1
ds −
Z
t
0
Z
Γ
(
√
gn
i
)
′′
h
′′
v
i
dx
1
ds
+
Z
t
0
Z
Γ
perm
(q
+
+ f )
′′
[[v
1
]]f
′′′
dx
1
ds +
Z
t
0
R(s)ds
=
1
2
|h
′′
|
2
0
−
1
2
|h
′′
0
|
2
0
−
1
2
Z
t
0
Z
Γ
(h
′′
)
2
v
1
′
dx
1
ds +
Z
t
0
|q
′′
[[v
1
]]|
0.5
|f |
2.5
ds
−
1
2
Z
Γ
perm
(f
′′
)
2
[[v
′
1
]]dx
1
ds +
Z
t
0
R(s)ds
≥
1
2
|h
′′
|
2
0
−
1
2
|h
′′
0
|
2
0
−
√
tP(E(t)),
where wehave usedHölder inequality, the tra etheorem, (3.14 )and theinequality
The remaining high order term an be handled asfollows: using (2.10 )and integrating by parts,
I
2
=
Z
t
0
Z
D
+
∪D
−
J(A
k
i
)
′′
(q + ψ
2
),
k
v
i
′′
dxds
= −
Z
t
0
Z
D
+
∪D
−
JA
k
r
ψ
r
,
11j
A
i
j
(q + ψ
2
),
k
v
′′
i
dxds +
Z
t
0
R(s)ds
=
Z
t
0
Z
D
+
∪D
−
ψ
r
,
11
JA
j
i
(A
k
r
(q + ψ
2
),
k
v
′′
i
),
j
dxds −
Z
t
0
Z
Γ
ψ
r
,
11
JA
j
i
(A
k
r
(q + ψ
2
),
k
v
′′
i
)N
j
dx
1
ds
+
Z
t
0
Z
Γ
perm
[[ψ
r
,
11
JA
j
i
(A
k
r
(q + ψ
2
),
k
v
i
′′
)]]N
j
dx
1
ds
−
Z
t
0
Z
Γ
bot
ψ
r
,
11
JA
j
i
(A
k
r
(q + ψ
2
),
k
v
′′
i
)N
j
dx
1
ds +
Z
t
0
R(s)ds
=
Z
t
0
Z
D
+
∪D
−
ψ
r
,
11
JA
j
i
(A
k
r
(q + ψ
2
),
k
v
′′
i
),
j
dxds −
Z
t
0
Z
Γ
h
′′
n
˜
i
A
k
2
(q + ψ
2
),
k
v
i
′′
dx
1
ds
+
Z
t
0
Z
Γ
perm
f
′′
[[˜
n
i
perm
A
k
2
(q + ψ
2
),
k
v
i
′′
]]dx
1
ds +
Z
t
0
R(s)ds.
The integralinthe bulk ofthe uid an beestimated using(3.16) sothat
Z
t
0
Z
D
+
∪D
−
ψ
r
,
11
JA
j
i
(A
k
r
(q + ψ
2
),
k
v
i
′′
),
j
dxds ≥ −
√
tP(E(t)).
Theintegralover
Γ
perm
anbeestimatedusingtheH
0.5
−H
−0.5
dualityand(3.17 )asfollows:
Thus,
Z
t
0
s
J
β
v
′′
2
0,±
ds +
1
2
s
−q,
2
(0)
J(0)
h
′′
(t)
2
0
≤
1
2
s
−q,
2
(0)
J(0)
h
′′
0
2
0
+
√
tP(E(t)).
(3.18)3.6 Ellipti estimates via the Hodge de omposition Inthis se tion, we usethefollowing
Lemma 7 ([13 ℄). Let
Ω
be a domain with boundary∂Ω
of Sobolev lassH
k+0.5
,
k ≥ 2
. Letψ
0
be a given smooth mapping and denecurl
ψ
0
v
anddiv
ψ
0
v
as in (1.10 ) and (1.11 ), respe tively. Then forv ∈ H
k
(Ω)
,kvk
H
k
(Ω)
≤ C
h
kvk
L
2
(Ω)
+ kcurl
ψ
0
vk
H
k−1
(Ω)
+ kdiv
ψ
0
vk
H
k−1
(Ω)
+ kv
′
·
nk
H
k−1.5
(∂Ω)
i
,
where n= (ψ
′
0
)
⊥
/|ψ
0
′
|
.Sin e inea hphase,
curl u = 0
anddiv u = 0
,it follows (see [12 ℄,Se tion 8.4.8)thatZ
t
0
k
urlψ
0
vk
2
1,±
dy ≤
√
tP(E(t)),
(3.19)Z
t
0
k
divψ
0
vk
2
1,±
dy ≤
√
tP(E(t)).
(3.20)First, we want to use Lemma 7 to obtain an estimate for
kv
′
k
1,±
. The only term that is deli ate isthe boundaryterm|v
′′
·
n
|
−0.5
. For that term wehave thefollowingLemma 8 (Estimates for thenormal tra e of
v
). Given a smooth solution to the Muskat problem (2.11a-g),Z
t
0
|v
′′
·
n|
2
H
−0.5
(∂D
−
∪∂D
+
)
ds ≤ C
Z
t
0
kv
′′
k
2
0,±
ds +
√
tP(E(t)) ,
(3.21) where n= (ψ
′
0
)
⊥
/|ψ
0
′
|
.Proof. Inordertoestimate