c European Mathematical Society 2013

J. Eur. Math. Soc. 15, 201227

DOI 10.4171/JEMS/360

Peter Constantin Diego Cordoba Francisco Gancedo Robert M. Strain

On the global existence for the Muskat problem

Received October 9, 2010 and in revised form April 15, 2011
Abstract. The Muskat problem models the dynamics of the interface between two incompressible
immiscible fluids with different constant densities. In this work we prove three results. First we
prove an L2 (R) maximum principle, in the form of a new log conservation law (3) which is
satisfied by the equation (1) for the interface. Our second result is a proof of global existence for
unique strong solutions if the initial data is smaller than an explicitly computable constant, for
instance kf k1 1/5. Previous results of this sort used a small constant   1 which was not
explicit [7, 19, 9, 14]. Lastly, we prove a global existence result for Lipschitz continuous solutions
with initial data that satisfy kf0 kL < and kx f0 kL < 1. We take advantage of the fact that
the bound kx f0 kL < 1 is propagated by solutions, which grants strong compactness properties
in comparison to the log conservation law.
Keywords. Porous media, incompressible flows, fluid interface, global existence

Contents
1. Introduction . . . . . . . . . . . . . . . . . .
2. L2 maximum principle . . . . . . . . . . . .
3. A global existence result for data less than 1/5
4. Global existence for initial data smaller than 1
References . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

201
205
207
211
226

1. Introduction
The Muskat problem models the dynamics of an interface between two incompressible
immiscible fluids with different characteristics, in porous media. The phenomenon has
P. Constantin: Department of Mathematics, Princeton University, Fine Hall 206, Washington Road,
Princeton, NJ 08544-1000, USA; e-mail: const@math.princeton.edu
D. Cordoba: Instituto de Ciencias Matematicas, Consejo Superior de Investigaciones Cientficas,
Serrano 123, 28006 Madrid, Spain; e-mail: dcg@icmat.es
F. Gancedo: Department of Mathematics, University of Chicago, 5734 University Avenue,
Chicago, IL 60637, USA; e-mail: fgancedo@math.uchicago.edu
R. M. Strain: Department of Mathematics, University of Pennsylvania, David Rittenhouse Lab,
209 South 33rd Street, Philadelphia, PA 19104, USA; e-mail: strain@math.upenn.edu
Mathematics Subject Classification (2010): 35A01, 76S05, 76B03

202

Peter Constantin et al.

been described using the experimental Darcy law that is given in two dimensions by the
following momentum equation:

u = p g(0, ).

Here is viscosity, permeability of the isotropic medium, u velocity, p pressure, g

gravity and density. Saffman and Taylor [18] related this problem to the evolution of an
interface in a Hele-Shaw cell since both physical scenarios can be modeled analogously
(see also [7] and reference therein). Recently, the well-posedness has been shown without
surface tension in [8] (for previous work on this topic see [1], [21], [2] and [9]) using
arguments that rely upon the boundedness properties of the Hilbert transforms associated
to C 1, curves. Precise estimates are obtained with arguments involving conformal mappings, the Hopf maximum principle and Harnack inequalities. The initial data have to
satisfy the RayleighTaylor condition initially, otherwise the problem has been shown to
be ill-posed in the sense of Hadamard (see [19] and [9] for more details). With surface tension, the initial value problem becomes more regular, and instabilities do not appear [13].
The case of more than one free boundary has been treated in [11] and [14].
In this paper we consider an interface given by fluids of different constant densities i ,
with the same viscosity and without surface tension. The step function is represented
by
 1
, x 1 (t),
(x, t) =
2 , x 2 (t) = R2 \ 1 (t),
for i (t) two connected regions. As the density is transported by the flow
t + u = 0,
the free boundary evolves with the two-dimensional velocity u = (u1 , u2 ). The Biot
Savart law allows one to recover u from the vorticity given by = x1 u2 x2 u1 , via the
integral operator
u(x, t) = 11 (x, t).
Darcys law then provides the relation = x1 where / and g are taken equal to 1
for the sake of simplicity. Then the velocity field can be obtained in terms of the density
as follows:
Z
1
u(x, t) = PV
K(x y)(y, t) dy (0, (x, t)).
2
2
R
Here the kernel K is of CalderonZygmund type:


1
x1 x2 x 2 x 2
K(x) =
2 , 1 22

|x|
2|x|
(see [20]). As a consequence of L (R2 R+ ) it follows that the velocity belongs
to BMO. Moreover, as K is an even kernel, it has the property that the mean of K (in the
principal value sense) is zero on hemispheres [4], and this yields a bound on the velocity
u(x, t) in terms of C 1, norms (0 < < 1) of the free boundary [11].

On the global existence for the Muskat problem

203

In order to have a well-posed problem we need to consider initially an interface parameterized as the graph of a function with the denser fluid below: 2 > 1 as in [9]. The
interface is characterized as the graph of the function (x, f (x, t)). This characterization
is preserved by the system and f satisfies
Z
(x f (x, t) x f (x , t))
2 1
,
PV d 2
ft (x, t) =
2

+ (f (x, t) f (x , t))2
(1)
R
f (x, 0) = f0 (x), x R.
The above equation can be linearized around the flat solution, which yields the following
nonlocal partial differential equation:
2 1
3f (x, t),
2
f (, 0) = f0 (), R,

ft (x, t) =

(2)

where the operator 3 is the square root of the Laplacian. This linearization shows the
parabolic character of the problem in the stable case ( 2 > 1 ), as well as the illposedness in the unstable case ( 2 < 1 ). We use the term ill-posedness here to mean
that some solutions do leave the H s spaces right away even for arbitrarily small data.
The nonlinear equation (1) is ill-posed in the unstable situation and locally well-posed
in H k (k 3) for the stable case [9]. Furthermore the stable system satisfies a maximum
principle kf kL (t) kf kL (0) (see [10]); decay rates are obtained for the periodic
case:
kf kL (t) kf0 kL eCt ,
and also for the case on the real line (flat at infinity):
kf kL (t)

kf0 kL
.
1 + Ct

Numerical solutions performed in [12] further indicate a regularizing effect. The decay
of the slope and the curvature is observed to be stronger than the rate of decay of the
maximum of the difference between f and its mean value. Thus, the irregular regions in
the graph are observed to be rapidly smoothed and the flat regions are smoothly bent. It
is shown analytically in [10] that, if the initial data satisfy kx f0 kL < 1, then there is a
maximum principle so that this derivative remains in absolute value smaller than 1.
The three main results we present in this paper are the following:
1) In Section 2, we prove that a solution of (1) formally satisfies
 
 
Z
Z
Z
2 1 t
f (x, s) f (, s) 2
2
kf kL2 (t)+
ds
d
dx ln 1+
= kf0 k2L2 .
2
x
0
R
R
Furthermore, we have the inequality
 


Z
Z
f (x, s) f (, s) 2
dx
d ln 1 +
Ckf kL1 (s).
x
R
R

(3)

204

Peter Constantin et al.

This identity shows a major difference with the linear equation (2) where the evolution of
the L2 norm provides a gain of half derivative for 2 > 1 :
Z t
2
2
1
kf kL2 (t) + ( )
ds k31/2 f k2L2 (s) = kf0 k2L2 ,
(4)
0

or equivalently
kf k2L2 (t) +

2 1
2

Z
ds

Z
dx
R

d
R

f (x, s) f (, s)
x

2

= kf0 k2L2 .

Notice that this linear energy balance (4) directly implies compactness, whereas compactness does not follow from the nonlinear L2 energy (3).
2) Our second result proves global existence of unique C([0, T ]; H 3 (R)) solutions if
initially the norm (7) of f0 is controlled as kf0 k1 < c0 where
Z
kf0 k1 =
d | | |f0 ( )|.
R

Of course here f denotes the standard Fourier transform of f . There are several results
of global existence for small initial data (small compared to 1 or   1) in several norms
[7, 19, 9, 14] taking advantage of the parabolic character of the equation for small initial
data. For example in [19] and [9], in order to measure the analyticity of the solution,
global existence is shown when kf0 k1 for very small (compared to 1) and
Z
d | |2 eb(t)| | |f0 ( )| eb(t) (1 + |b(t)| 1 ),
(5)
R

where 0 < < 1 and b(t) = a ct/2. Here c depends on the RayleighTaylor condition
and a ct/2.
For the Hele-Shaw problem, [7] proves the global existence in time for small analytic
perturbations of the circle, and nonlinear asymptotic stability of the steady circular solution. Also recently [14] considers the Muskat problem in a periodic geometry, and proves
the well-posedness as well as the exponential stability of a certain flat equilibrium.
The key point for our result, in comparison to previous work [7, 19, 9, 14], is that the
constant c0 can be easily explicitly computed (see (8)). We have checked numerically that
c0 is not that small: it is greater than 1/5.
3) In Section 4 we prove global in time existence of Lipschitz continuous solutions in
the stable case. Being a solution of (1) is understood in its weak formulation:
T

dx (x, 0)f0 (x)

dx t (x, t)f (x, t) +

dt
R

dt

=
0

2 1
dx x (x, t)
PV
2
R

f (x, t) f (, t)
d arctan
x
R


(6)

On the global existence for the Muskat problem

205

for all Cc ([0, T )R). For initial data f0 satisfying kf0 kL < and kx f0 kL < 1
we prove that there exists a solution of (6) that remains in the spaces C([0, T ] R)
L ([0, T ]; W 1, (R)) for any T > 0. We point out that, because of the condition f
L (R), the nonlinear term in (6) has to be understood as a principal value for the integral
of two functions, one in H1 , the Hardy space, and the other in BMO [20].
Kim [16] studied viscosity solutions for the one-phase Hele-Shaw and Stefan problems. For the Muskat problem, previous results of global existence [7, 19, 9, 14]
need small initial data more regular than Lipschitz. We show here that we just need
kx f0 kL < 1, therefore


 f0 (x) f0 () 
 < 1.



x
Notice that if we consider the first order term in the Taylor series of ln(1 + y 2 ) (absolutely
convergent for |y| < 1), then the identity (3) becomes (4).

2. L2 maximum principle
In this section we provide a proof of the identity (3). As we are in the stable case, we take
without loss of generality ( 2 1 )/(2 ) = 1 to simplify the exposition. The contour
equation (1) can be written as follows:


Z
f (x, t) f (x , t)
ft (x, t) = PV x arctan
d.

R
We multiply by f , integrate over dx, and use integration by parts to observe


Z
Z
f (x, t) f (x , t)
1 d
d fx (x) arctan
kf k2L2 (t) = dx
2 dt

R
R


Z
Z
f (x, t) f (, t)
= dx
d fx (x) arctan
.
x
R
R
We use the splitting



Z Z 
1 d
f (x, t) f (, t)
f (x, t) f (, t)
2
kf kL2 (t) =
arctan
dx d
2 dt
x
x
R R



Z Z 
fx (x)(x ) (f (x, t) f (, t))
f (x, t) f (, t)

arctan
dx d
x
x
R R
= I1 + I2 .
We also use the function G defined by
Z
p
G(x) = x arctan x ln 1 + x 2 =
0

dy arctan y.

206

Peter Constantin et al.

With these, it is easy to observe that



Z Z
f (x, t) f (, t)
I2 =
(x )x G
dx d.
x
R R
The identity


f (x, t) f (, t)
=0
lim (x )G
|x|
x
allows us to integrate by parts to obtain


Z Z
f (x, t) f (, t)
I2 =
G
dx d
x
R R
s


Z Z
f (x, t) f (, t) 2
= I1
ln 1 +
dx d.
x
R R
This equality gives
1 d
kf k2L2 (t) =
2 dt

Z Z

ln 1 +
R R

f (x, t) f (, t)
x

2
dx d,

and integrating in time we get the desired identity.

The above equality indicates that for large initial data, the system is not parabolic at
the level of f in L2 . We prove below the inequality
Z Z
R



 

f (x, t) f (, t) 2
ln 1 +
dx d 4 2 kf kL1 (t),
x
R

which shows that there is no gain of derivatives for the stable case. If the initial data are
positive, then kf kL1 (t) kf0 kL1 follows from [10], so that the dissipation is bounded in
terms of the initial data with zero derivatives.
For the proof of the inequality, we denote by J the integral
Z Z


ln 1 +

J :=
R R

f (x) f (x )

2 
dx d.

We now use that the function ln(1 + y 2 ) is increasing to observe that



Z Z
2|f (x)|2
2|f (x )|2
J
ln 1 +
+
dx d.
2
2
R R
The inequality ln(1 + a 2 + b2 ) ln(1 + a 2 ) + ln(1 + b2 ) yields




Z Z
Z Z
2|f (x)|2
2|f (x )|2
J
ln 1 +
dx d +
ln 1 +
dx d,
2
2
R R
R R

On the global existence for the Muskat problem

and therefore

207



2|f (x)|2
dx d = K.
ln 1 +
2
R

Z Z
J 2
R

For K it is easy to get

Z

K=2

dx
{x : |f (x)|6=0}

so that an easy integration in provides

Z
K = 4 2
{x : |f (x)|6=0}



2|f (x)|2
d ln 1 +
,
2
R

dx |f (x)| = 4 2 kf kL1 .

This concludes our discussion of the L2 maximum principle (3) for (1).
3. A global existence result for data less than 1/5
In this section we prove global existence of C([0, T ]; H 3 (R)) small data solutions. A key
point is to consider the norm
Z
kf ks :=
d | |s |f( )|, s 1.
(7)
R

This norm allows us to use Fourier techniques for small initial data that give rise to a
global existence result for classical solutions. The key point is that (7) appears naturally
when taking the Fourier transform of the equation in our computations below (see (9) and
(10)). As a result, (7) provides sharper constants than the ones that one could obtain with
different norms.
Theorem 3.1. Suppose that initially f0 H 3 (R) and kf0 k1 < c0 , where c0 is a constant
such that
X
2
(2n + 1)2+ c02n 1
(8)
n1

for a fixed 0 < < 1/2. Then there is a unique solution f of (1) that satisfies f
C([0, T ]; H 3 (R)) for any T > 0.
Remark 3.2. We compute the limit case = 0, so that
X
2
(2n + 1)2 c02n 1
n1

for
s
q
1
14 52/3
3
7 p
+ 2 5(9 39 38) 0.2199617648835399.
0 c0
3
3
9 39 38
In particular,
2

X
(2n + 1)2.1 c02n < 1
n1

if say c0 1/5.

208

Peter Constantin et al.

The remainder of this section is devoted to the proof of Theorem 3.1. The contour
equation for the stable Muskat problem (1) can be written as
(9)

ft (x, t) = (3f + T (f )),

where we recall that = ( 2 1 )/2 > 0 and we have
1
T (f ) =

Z
R

f (x)f (x)
2

f (x)f (x)
2

x f (x) x f (x )

1+

d.

(10)

We define
1 f (x) :=

f (x) f (x )
.

We consider the evolution of the norm kf k1 given in (7):

d
kf k1 (t) =
dt

d | | sgn(f( ))ft ( )
Z
=
d | | sgn(f( ))(| |f( ) F(T )( )).
R

We will show that the first term dominates the second if initially
q
kf0 k1 <

(4 13)/6,

q
where

(4

13)/6 > 1/4.

The key point, again, is that the constant is given explicitly.

Notice that under the local existence theorem of [9], this bound will be propagated for
a short time. Then we may use the Taylor expansion

X
x2
=
(1)n+1 x 2n ,
1 + x2
n=1

to obtain
T (f ) =

1 X
(1)n
n1

x (1 f )(1 f )2n d.

(11)

Notice that
F(1 f ) = f( )m(, ),

F(x 1 f ) = i f( )m(, ),

m(, ) =

Therefore
F(x (1 f ) (1 f )2n ) = ((i fm) (fm) (fm))(, ),

1 ei
.

On the global existence for the Muskat problem

209

with 2n convolutions, one with i fm and 2n 1 with fm. Using (11) we obtain
F(T )( ) =

Z
Z
Z
i X
d
d1 d2n ( 1 )f( 1 )m( 1 , )
(1)n
n1
R
R
R

f(1 2 )m(1 2 , ) f(2n1 2n )m(2n1 2n , )f(2n )m(2n , )

Z

2n1
Y
XZ
f(i i+1 ) f(2n )Mn ,
d1 d2n ( 1 )f( 1 )
=
n1 R

i=1

where Mn = Mn (, 1 , . . . , 2n ) is given by
i
Mn := (1)n

Since m(, ) = i

R1
0

2n1

Y
m( 1 , )
m(i i+1 , ) m(2n , ) d.

Z
R

(12)

i=1

ds ei(s1) we obtain

Mn (, 1 , . . . , 2n ) = mn (, 1 , . . . , 2n ) (1 2 ) (2n1 2n )2n
with
i

i
=

mn =

Z
ds1

1 ei( 1 )

R
 2n1

X
exp i
(sj 1)(j j +1 ) + i(s2n 1)2n

Z
ds2n

j =1
1

Z
ds1

0
Z 1

ds2n
0
Z 1

d
PV exp(iA)
PV

d
exp(iB)

ds2n (sgn A sgn B),

ds1

where
A=

2n1
X

(sj 1)(j j +1 ) + (s2n 1)2n = 1 +

j =1

j =1

B = ( 1 ) +

2n1
X

(sj 1)(j j +1 ) + (s2n 1)2n

j =1

= +

2n
X

2n
X
j =1

sj j

2n1
X
j =1

sj j +1 .

sj j

2n1
X
j =1

sj j +1 ,

210

Peter Constantin et al.

It follows that
F(T )( ) =

XZ

d2n mn (, 1 , . . . , 2n ) ( 1 )f( 1 )

d1

n1 R

2n1
Y


(i i+1 )f(i i+1 ) 2n f(2n ),

i=1

with |mn (, 1 , . . . , 2n )| 2. We then have

Z
d | | |F(T )( )| 2

XZ

n1 R

d2n | | | 1 | |f( 1 )|

d1
R

|1 2 | |f(1 2 )| |2n1 2n | |f(2n1 2n )| |2n | |f(2n )|.

The inequality | | | 1 | + |1 2 | + + |2n1 2n | + |2n | yields
Z
d | | |F(T )( )| 2
R

Z
2n

d | | |f ( )|
d | | |f ( )|
,

Z

(2n + 1)
R

n1

and therefore
Z

Z
d | | |F(T )( )|

 X
d | |2 |f( )| 2
(2n + 1)kf k2n
1

n1


2kf k21 (3 kf k21 )
2

d | | |f ( )|
.
(1 kf k21 )2
R
q

2
2
2
2
Notice 2x (3 x )/(1 x ) < 1 if 0 x < (4 13)/6 0.256400964. If
q

kf0 k1 < (4 13)/6, then this inequality will continue to hold for some time so that
Z

d
kf k1 (t) 0,
dt
q

and we conclude that kf k1 (t) kf0 k1 if kf0 k1 < (4 13)/6.

Now we repeat the argument but with s > 1 in (7). Our goal is to obtain
d
kf k2+ (t) 0,
dt

0 < < 1/2.

(13)

Let us point out that

kf0 k2+ C(kf0 kL2 + kx3 f0 kL2 )
for 0 < < 1/2. Using the inequality
| |2+ (2n + 1)1+ (| 1 |2+ + |1 2 |2+ + + |2n1 2n |2+ + |2n |2+ ),

On the global existence for the Muskat problem

211

we proceed as before to get

Z
Z
X
(2n + 1)2+ kf k2n
| |2+ |F(T )( )| d
| |3+ |f( )| d 2
1 .
R

n1

In particular, taking kf k1 small enough we find

Z
Z
| |2+ |F(T )( )| d
| |3+ |f( )| d,
R

and bound (13) therefore holds.

If kf kC 2, remains bounded (0 < < 1), then from previous work [9], we can deduce
global existence in C([0, T ]; H 3 (R)) for any T > 0. Note that the Holder seminorm
|g(x + y) g(x)|
|y|
x, y6=0

|g|C = sup
is bounded by



Z
Z
 C

|g(x + y) g(x)|
ix iy

C
| | |g(
)| d,
=
sup
g(

)e
(e

1)
d
 |y|


|y|
R
R
x, y6=0
x, y6=0
sup

and therefore
kf kC 2,



Z
Z
2+

C kf kL +
d | | |f ( )| +
d | | |f ( )| .
R

We conclude that the solution can be continued for all time if kf0 k1 is initially smaller
than a computable constant c0 , and kf0 k2+ is bounded. The constant c0 is defined by the
condition
X
2
(2n + 1)2+ c02n 1,
n1

which has been numerically verified to be no smaller than say 1/5.

4. Global existence for initial data smaller than 1

We prove now the existence of a weak solution of the system (1) which can be written as
follows:


Z

f (x) f (x )
ft = x PV arctan
d,
(14)

R
where = ( 2 1 )/2. We first extend the sense of the contour equation with a weak
formulation: for any Cc ([0, T ) R), a weak solution f should satisfy (6). We show
here that this is the case if kx f0 kL < 1. Then it follows that kf kL (t) kf0 kL
and kx f kL (t) kx f0 kL < 1 as in [10]. Then the solution is in fact Lipschitz
continuous by Morreys inequality. The main result we prove below is the following:

212

Peter Constantin et al.

Theorem 4.1. Suppose that kf0 kL < and kx f0 kL < 1. Then there exists a global
in time weak solution of (6) that satisfies
f C([0, T ] R) L ([0, T ]; W 1, (R)).
In particular f is Lipschitz continuous.
The rest of this section is devoted to the proof of Theorem 4.1. The first step is to prove
global in time existence of classical solutions to the regularized model (15) below. This
is done in Section 4.2. Prior to that, in Section 4.1 we prove some necessary a priori
bounds. Then, in Section 4.3 we explain how to approximate the initial data. Section 4.4
shows how to prove the existence of solutions of (6), subject to the strong convergence
established in Section 4.5.
From now on, in the next two subsections we write f = f for the solution to (15)
for the sake of simplicity of the notation. The regularized model is given by
Z

(15)
ft (x, t) = C31 f + fxx + x PV d arctan(1 f (x)).

R
where C > 0 is a universal constant fixed below, the operator 31 f is given by the
formula
Z
f (x) f (x )
31 f (x) = c1 ()
d,
(16)
||2
R
with 0 < cm c1 () cM for 0 1/4, and we define
1 f (x) :=

f (x) f (x )
,
()

with () = () = /|| and > 0 is small enough. Initially we consider the data
f0 W 1, (R) with kx f0 kL (R) < 1. We will explain how to approximate this initial
data later on in Section 4.3.
4.1. A priori bounds
In this section we show two a priori bounds for the regularized system (15). We prove the
following result:
Proposition 4.2. Let f (x, t) be a regular solution of the system (15). Then
kx f kL (t) kx f0 kL < 1,

kf kL (t) kf0 kL ,
for any t > 0.

In order to prove the first estimate, we check the evolution of

M(t) = max f (x, t) = f (xt , t).
x

On the global existence for the Muskat problem

213

Then M is differentiable for almost every t and

M 0 (t) = ft (xt , t) = C31 f (xt ) + fxx (xt ) +

x PV

arctan(1 f (xt )) d

(see [5, 10] for more details regarding the differentiability of M(t)). Using formula (16)
it is easy to check that the second and the third term above have the correct sign. We have
to deal with the third one. Now
Z
Z

I (x) = x PV arctan(1 f (x)) d = x PV arctan(1x f (x)) d,

(17)
R

and thus
1
(x)
d
x f (x) PV

2
R 1 + (1x f (x))
f (x)f ()
Z
|x|2
(1 ) PV
d.

2
R 1 + (1x f (x))

I (x) =

(18)

Therefore I (xt ) 0 (since x f (xt ) = 0). Then M 0 (t) 0 for a.e. t (0, T ] and
M(t) M(0). Analogously, we check the evolution of
m(t) = min f (x, t) = f (xt , t)
x

and find m(t) m(0).

From (15) and (17) we have
fxt = C31 fx + fxxx +

Ix .

Using (18) we rewrite I (x) = J 1 (x) + J 2 (x) where

Z
1
fx (x)(x ) (f (x) f ())
1
J (x) = PV
d,

2
2
|x

|
1
+
(1
R
x f (x))
Z
1
f (x) f (x )
J 2 (x) = PV
d,
2
f (x))2
||
1
+
(1
R

to find
Jx1 (x)

1
1
d

2
(x

)
1
+
(1
R
x f (x))
Z
()
fx (x) f (x)f
1
x
(2 ) PV
d

2
|x |
1 + (1x f (x))2
R
Z
21x f (x)
fx (x)(x ) (f (x) f ())
PV
|x |2
(1 + (1x f (x))2 )2
R
fx (x)(x ) (1 )(f (x) f ())

d,
|x |2

= fxx (x) PV

214

Peter Constantin et al.

and we split further Jx1 (x) = K 1 (x) + K 2 (x) + K 3 (x) + K 4 (x) where
Z
1
1
1
d,
K (x) = fxx (x) PV
()
1 + (1 f (x))2
R
Z
fx (x) f (x)f (x)
1
2
K (x) = PV
d,
2
f (x))2
||
1
+
(1
R

Z
fx (x) f (x)f (x)
2
d,
K 3 (x) = PV
2
f (x))2
||
1
+
(1
R

Z
fx (x) f (x)f (x)
21 f (x)
K 4 (x) = PV
d
2
||
(1 + (1 f (x))2 )2
R
(fx (x)|| (1 )1 f (x)).
For J 2 it is easy to check that
Z
fx (x) fx (x )
d
2
Jx (x) = PV
2
f (x))2
||
1
+
(1
R

Z
fx (x) fx (x ) 2(1 f (x))2 d
PV
.
||2
(1 + (1 f (x))2 )2
R
Next, as we did before, we consider M(t) = maxx fx = fx (xt , t). Then M(t) is differentiable (as we explained previously). It follows that
M 0 (t) = fxt (xt , t) = C31 fx (xt ) + fxxx (xt ) +

Ix (xt ).

Now we claim that if M(t) < 1 then M 0 (t) 0 for a.e. t. We can conclude analogously
for m(t) = minx fx > 1, m0 (t) 0 for a.e. t.
We check that if M(t) < 1, then
C31 fx (xt ) + fxxx (xt ) +

Ix (xt ) 0.

We can use in some cases the following formula for the operator 31 fx :
31 fx (x) = c2 ()

Z
R

fx (x) f (x)f (x)

d,
||2

where 0 < cm c2 () cM for 0 1/4.

We claim that

C31 fx (xt ) + (K 2 (xt ) + Jx2 (xt )) 0.

We will show that

C 1

3 fx (xt ) + K 2 (xt ) 0,
2

(19)

On the global existence for the Muskat problem

215

using (19) and that

C 1

3 fx (xt ) + J 2 (xt ) 0,
2

by (16). In fact

C 1
3 fx (xt ) + K 2 (xt )
2

Z
fx (xt ) f (xt )f (xt )
= PV
||2
R

Cc2 ()
Cc2 ()

2
2 (1 f (xt )) +
2
1 + (1 f (xt ))2

d.

The mean value theorem gives

|f (x) f (x )|/|| kfx kL .
Thus if we take C 2/(cm ) we obtain the first inequality. Also

C 1

3 fx (xt ) + J 2 (xt )
2

Z
Cc ()
Cc ()
fx (xt ) fx (xt ) 21 (1 f (xt ))2 + 21
= PV
||2
1 + (1 f (xt ))2
R
Z
fx (xt ) fx (xt ) 2(1 f (xt ))2 d
PV
0.
||2
(1 + (1 f (xt ))2 )2
R

Thus the term above has the desired sign.

We find fxxx (xt ) 0 and K 1 (xt ) = 0. We still have to deal with K 3 and K 4 .
Considering K 3 (xt ) + K 4 (xt ), we realize that if
P () = 2 + 2(1 f (xt ))2 + 2(1 f (xt ))(fx (xt )|| (1 )1 f (xt )) 0,
then we are done. We rewrite
P () = 2 + 2(1 f (xt ))2 + 2(1 f (xt ))fx (xt )|| ,
and therefore we need
|(1 f (xt ))fx (xt )|| | 1.
This fact holds if

|f (xt ) f (xt )|
< 1.
||12
6=0

|f |C 12 = sup

Now we will check that if kf kL kf0 kL and kfx kL < 1 then |f |C 12 < 1 for
small enough uniformly. We replace 2 by without loss of generality. If kf0 kL = 0 or
kfx kL = 0 then there is nothing to prove. Otherwise
|f (xt ) f (xt )|
kfx kL
||1

216

Peter Constantin et al.

for 0 < || , and

kf0 kL
|f (xt ) f (xt )|
2 1
1
||

for || . We take 1 = 2kf0 kL /kfx kL and therefore

1/(1)

|f |C 1 max{kfx kL , kfx kL

(2kf0 kL )/(1) }.

Now it is clear that given kf0 kL , if kfx kL < 1 there exists 0 > 0 such that |f |C 1
1 for any 0 0 .
4.2. Global existence for the regularized model
In this section we use the a priori bounds to show global existence. Local existence can be
easily proved using the local existence proof for the non-regularized Muskat problem (1),
as in [9]. We use energy estimates and the Gronwall inequality. We have the following
result.
Proposition 4.3. Let f (x, t) be a regular solution of the system (15). Then
Z t

kf k2L2 (t) + kx3 f k2L2 (t) (kf0 k2L2 + kx3 f0 k2L2 ) exp
C()G(s) ds

(20)

for
G(s) = kf k4L + kf k2L + kfx k4L (s) + kfx k2L (s) + 1
and any t > 0.
Estimate (20) allows us to find f C([0, T ]; H 3 (R)) for any T > 0 by the a priori
bounds.
Furthermore, as we did for (1), it follows that

d
kf k2L2 (t) =
dt
2

Z Z
R R



 
f (x, t) f (, t) 2
1
ln
1
+
dx d
|x |
(x )

2Ck3(1)/2 f k2L2 (t) 2kfx k2L2 (t).

Therefore kf kL2 (t) kf0 kL2 .
Remark 4.4. The above maximum principle for the regularized system shows that Theorem 4.1 can also be proved with
kf0 kL2 < instead of kf0 kL < .
We picked the version above because it is more suitable. We see that if the solution satisfies initially an L2 bound then f L ([0, T ]; L2 (R)).

On the global existence for the Muskat problem

217

Next, we consider the evolution of

Z
R

x3 f x3 ft dx Ck3(1)/2 x3 f kL2 (t) kx4 fx k2L2 + L1 + L2 ,

where

L1 =

x3 f (x)x3

Z
PV
R

L2 =


fx (x) fx (x )
d dx,
()

 Z

fx (x) fx (x ) (1 f (x))2 d
3
3
x f (x)x PV
dx.
()
1 + (1 f (x))2
R
R

The term fx (x) cancels out in L1 due to the PV. An integration by parts further shows
that
Z

L1 = C() x3 f (x)31 x3 f (x) dx 0.

R
For L2 one finds
 Z

fx (x) fx (x ) (1 f (x))2 d
x4 f (x)x2 PV
dx,
()
1 + (1 f (x))2
R
R

L2 =

which splits as L2 = M1 + M2 + M3 with

M1 =

x4 f (x)

x3 f (x) x3 f (x ) (1 f (x))2 d
dx,
()
1 + (1 f (x))2

x2 f (x) x2 f (x ) fx (x) fx (x )
()
()
R
R
2(1 f (x)) d

dx,
(1 + (1 f (x))2 )2

Z
Z 
fx (x) fx (x ) 3 (2 6(1 f (x))2 ) d

4
M3 =
f (x)
dx.
R x
()
(1 + (1 f (x))2 )3
R
3

M2 =

x4 f (x)

We will now estimate each of these terms from above.

For M1 we proceed as follows:
|M1 | =

Z

Z
d

||>1

Z
dx +

Z
d

||<1

dx

C()(kf kL + 1)kx3 f kL2 kx4 f kL2 .

The identity
x2 f (x) x2 f (x ) =

Z
0

x3 f (x + (s 1)) ds

218

Peter Constantin et al.

yields
|M2 |

Z
ds

||<1

d
||12

dx |x4 f (x)| |x3 f (x + (s 1))|

(|fx (x)| + |fx (x )|)

6
+

Z
ds

||>1

d
||23

dx |x4 f (x)| |x3 f (x + (s 1))|

R

(|fx (x)| + |fx (x )|)(|f (x)| + |f (x )|),

and therefore
|M2 | C()(1 + kf kL )kx3 f kL2 kx4 f kL2 kfx kL .
In M3 we use the splitting M3 = N1 + N2 where
Z
Z
Z
Z

d
dx,
N2 =
d
dx,
N1 =
||>1
R
||<1
R
and then
16
|N1 |
kfx k2L

Z
||>1

d
||33

dx |x4 f (x)|(|fx (x)| + |fx (x )|)

Ckfx k2L kfx kL2 kx4 f kL2 Ckfx k2L (kf kL2 + kx3 f kL2 )kx4 f kL2 .
To finish, the equality
Z
fx (x) fx (x ) =
0

x2 f (x + (s 1)) ds

allows us to obtain (since 1/2 + 1/4 + 1/4 = 1)

Z 1 Z 1 Z
Z
16
d
|N2 |
kfx kL
dr
ds
dx
13

0
0
||<1 ||
R
|x4 f (x)| |x2 f (x + (r 1))| |x2 f (x + (s 1))|
Ckfx kL kx4 f kL2 kx2 f k2L4 .
The estimate
Z
Z
kx2 f k4L4 = (x2 f )3 x2 f dx = 3 (x2 f )2 x3 f x f dx 3kfx kL kx2 f k2L4 kx3 f kL2
R

yields
|N2 | Ckfx k2L kx4 f kL2 kx3 f kL2 .
Using Youngs inequality we obtain
d 3 2
k f k 2 C()(kf k4L + kf k2L + kfx k4L + kfx k2L + 1)(kf k2L2 + kx3 f k2L2 ),
dt x L
and therefore the Gronwall inequality yields (20).

On the global existence for the Muskat problem

219

4.3. Approximation of the initial data

The approximation of the initial data described below is needed in order to construct
a weak solution. First consider a common approximation to the identity Cc (R)
satisfying
Z
dx (x) = 1,

0,

(x) = (x).

Now the standard mollifier (x) = (x/)/ continues to satisfy the normalization condition above.
For any f0 W 1, (R) with kx f0 kL < 1, we define the initial data for the regularized system as follows:
( f0 )(x)
.
f0 (x) =
1 + x 2
Notice that f0 H s (R) for any s > 0, and
kf0 kL kf0 kL .
More importantly, kx f0 kL < 1 when kx f0 kL < 1 if is sufficiently small (here
will generally depend upon the size of kf0 kL ).
In particular
x f0 (x) =

( x f0 )(x)
( f0 )(x)
2x
,
2
1 + x
(1 + x 2 )2

and clearly


 ( x f0 )(x) 


 1 + x 2  k( x f0 )kL (R) kx f0 kL (R) .
On the other hand, by splitting into |x| 2/3 and |x| > 2/3 we have
2x(1 + x 2 )2 2 max{ 1/3 , }.
On the unbounded region we have

x(1 + x 2 )2 = (1/ x + x 3/2 )2 1.

Thus, the desired bound follows if is small enough. Therefore global existence of the
regularized system (15) holds for f0 if is small enough.
Now consider the solution to the regularized system (15) with initial data given by
the f0 just described above. For > 0 sufficiently small, we decompose
Z
Z
( f0 )(x)
(x, 0)
dx = I1 + I2 ,
(x, 0)f0 (x) dx =
2
1
+
x
R
R
where
I1

Z
=
R


(x, 0)( f0 )(x)


1
1 dx,
1 + x 2

I2

Z
(x, 0)( f0 )(x) dx.

=
R

220

Peter Constantin et al.

We apply the dominated convergence theorem to find that I1 0 as 0. For I2 we

write
Z

I2 =
((, 0))f0 (x) dx.
R

The

L1

approximation of the identity property shows that

Z
I2
(x, 0)f0 (x) dx.
R

Thus, it remains to check the convergence of the rest of the terms in (6).
4.4. Weak solution
In this section we prove that solutions of the regularized system converge to a weak solution satisfying the bounds
kf kL (t) kf0 kL ,

kx f kL (t) kx f0 kL < 1.

(21)

Given a collection of regularized solutions {f } to (15), we have the uniform (in > 0)
bounds
kf kL (t) kf0 kL , kx f kL (R) (t) 1, > 0.
(22)
This implies that there is a subsequence (denoted again by f ) such that
Z TZ
Z TZ
f (x, t)g(x, t) dx dt
f (x, t)g(x, t) dx dt,
0
T

Z
0

x f (x, t)g(x, t) dx dt
R

R
T

Z
x f (x, t)g(x, t) dx dt,

for f L ([0, T ]; W 1, (R)) and any g L1 ([0, T ] R) by the BanachAlaoglu

theorem. This yields weak* convergence in L ([0, T ]; W 1, (R)).
We denote BN = [N, N]. Then we claim that there is a subsequence (denoted again
by f ) such that
kf f kL ([0,T ]BN ) 0 as 0.
We will prove this in Section 4.5. Then, up to a subsequence, we infer uniform convergence of f to f on compact sets. Since f C([0, T ]R) we find that f is continuous.
The only thing remaining to check is that as 0 we have
T

Z
dt



Z
f (x) f (x )

PV d arctan

()
R


Z
Z T Z

f (x) f (x )
dx x (x, t) PV d arctan
dt

0
R
R

dx x (x, t)
R

where () = /|| . The other terms will converge in the usual obvious way (since
they are linear).

On the global existence for the Muskat problem

221

Choose M > 0 so that supp() BM . For any small > 0 and any large R  1
with R > M + 1, we split the integral as
Z
Z
Z
Z
d =
d +
d +
d.
R

BR B

BRc

We first prove that the first and last integrals separately are arbitrarily small independent
of for R > 0 sufficiently large and for > 0 sufficiently small. One finds that






arctan f (x) f (x )  .


()
2
Here we do not need any regularity for f , and we conclude that
Z T Z


Z


f (x) f (x ) 
 dt
dx x (x, t) PV
d arctan

 kx kL1 ([0,T ]R) .

()
0
R
B
Therefore this term can clearly be made arbitrarily small, depending upon the smallness
of .
We now estimate the term integrated over BRc . We note that
Z 1
Z 1
d
1
arctan y =
(arctan(sy)) ds = y
ds,
ds
1
+
s2y2
0
0
and therefore

Z
arctan y = y 1
0


s2y2
ds
.
1 + s2y2

This is morally the first order Taylor expansion for arctan with remainder in integral form.
From this expression we have


f (x) f (x )
d arctan
()
BRc
 Z

Z
f (x) f (x ) 3 1
= HR (f ) PV
d
()
BRc
0 1 + s2

Z
PV

s 2 ds
f (x)f (x)
2
()

Here HR is a (Hilbert-type) transform which has the form

Z
f (x )

HR (f ) := PV
d
.
()
BRc
The principal value is evaluated at infinity (if necessary). For the second term on the left
hand side notice that the integral is over BRc and the principal value is not necessary. In
particular, we have
Z
Z 1 
Z


Ckf0 k3
d

 Ckf k3

.
d
ds
L
 c

33
R
BR
0
R

222

Peter Constantin et al.

This term is therefore arbitrarily small if R is chosen sufficiently large. We are going to
show the same for
Z
M

IR :=
M

dx x (x, t)HR (f ).

We write IR = JR + KR where
Z

JR := lim

n M
Z M

KR := lim

n M

dx x (x, t)

d
n
n

Z
dx x (x, t)

d
R

f (x )
,
()

f (x )
.
()

We shall show how to control JR ; the same follows for KR . We write

Z M
Z n
f ()
dx x (x, t)
d
JR = lim
.
n M
(x )
x+R
An integration by parts yields


Z M
Z n
f (x + R)|R|
f ()
dx (x, t)
.
JR = lim
+ (1 )
d
n M
R
|x |2
x+R
Then for  (0, 1/2) we have
|JR | 2kkL1 kf0 kL /R 1/2 .
Since the same estimate holds for |KR |, one finds that IR is arbitrarily small if R is
arbitrarily large.
It remains to prove the convergence of


Z TZ
f (x) f (x )
d x (x, t) arctan
.
()
0
BR B
Recall that we have uniform convergence on compact sets. Let us consider
G =

f (x) f (x )
,
()

where x BM and BR B . Since arctan is a continuous function, we have

arctan(G ) arctan(G0 ) uniformly. Thus the integral of arctan(G ) over a bounded
region also converges. Hence for any R > M + 1 and any small > 0 as 0 we have


Z T Z
Z

f (x) f (x )
dt
dx x (x, t)
d arctan
BR B
()
0
R


Z T Z
Z

f (x) f (x )

dt
dx x (x, t)
d arctan
.
BR B

0
R
We conclude by first choosing R sufficiently large and > 0 sufficiently small and then
sending 0. Note that R and will generally depend upon the size of kf0 k , but this
does not affect our argument.

On the global existence for the Muskat problem

223

4.5. Strong convergence in L ([0, T ]; L (BR ))

In order to prove the strong convergence in L ([0, T ]; L (BR )), the idea is to use the
non-standard weak space W2, (BR ) which will be defined below. Crucially, we will
have the uniform bounds:
sup kf (t)kW 1, (BR ) Ckf0 kW 1, (BR ) ,

t[0,T ]

f


Ckf0 kL (R) ,
sup
t (t) 2,
t[0,T ]
W
(BR )

(23)

where C does not depend on R or . From this we will conclude that for any finite R > 0
there exists a subsequence such that f f strongly in L ([0, T ]; L (BR )).
We define the space W2, (BR ) as follows. For v L (BR ) we consider the norm

Z


.

(x)v(x)
dx
kvk2, =
sup


W02,1 (BR ): kk2,1 1

BR

Here W02,1 (BR ) is the usual set of functions in W 2,1 (BR ) which vanish on the boundary
of BR together with their first two weak derivatives. Now the Banach space W2, (BR )
is defined to be the completion of L (BR ) with respect to the norm k k2, . In general
this is all we need for our convergence study. This will be explained after the proof of
Lemma 4.5 below. The full space W2, may be large, but because we are going to deal
with f and df /dt both in L ([0, T ]; L (BR )), it is not difficult to find the norms of
both functions in W2, (BR ).
These spaces are suitable because
W 1, (BR ) L (BR ) W2, (BR ).
Now the embedding L (BR ) W2, (BR ) is continuous, and the embedding
W 1, (BR ) L (BR ) is compact by the Arzel`aAscoli theorem.
We now proceed to discuss the convergence argument. Arguments related to Lemma
4.5 below are described for instance in [6]. However in [6] reflexive Banach spaces are
used. None of the spaces used here are reflexive.
Lemma 4.5. Consider a sequence {um } in C([0, T ] BR ) that is uniformly bounded
in the space L ([0, T ]; W 1, (BR )). Assume further that the weak derivative dum /dt
is in L ([0, T ]; L (BR )) (not necessarily uniformly) and is uniformly bounded in
L ([0, T ]; W2, (BR )). Finally suppose that x um C([0, T ]BR ). Then there exists
a subsequence of um that converges strongly in L ([0, T ]; L (BR )).
Proof. Notice that it is enough to prove that the convergence is strong in the space
L ([0, T ]; W2, (BR )) because of the following interpolation theorem: for any small
> 0 there exists C > 0 such that
kukL kuk1, + C kuk2, .

224

Peter Constantin et al.

This holds for all u W 1, (BR ). See, for example, [6, Lemma 8.3]. Here we can replace
reflexivity with the BanachAlaoglu theorem in W 1, (BR ).
Let t, s [0, T ] be arbitrary. We have
Z t
um
um (t) um (s) =
d
( ).

s
This holds rigorously in the sense that
Z
Z
Z t
Z
um (t) dx
um (s) dx =
d
for any

BR
2,1
W (B

BR
R ).

BR

um
( ) dx

(24)

Clearly,

kum (t) um (s)kW 2, (B

R)

um

sup
|t s|,
( )
2,
[0,T ]
W
(BR )

and therefore
kum (t) um (s)kW 2, (B

where

R)

L|t s|,

(25)

um

L = sup sup
.
( )
2,
mN [0,T ]
W
(BR )

Now we consider {tk }kN = [0, T ] Q. We have um (tk ) W 1, (BR ) for any m and k.
By the standard diagonalization argument, we can get a subsequence (still denoted by m)
such that
um (tk ) u(tk )
in L (BR ) for any k as in the Arzel`aAscoli theorem.
Consider  > 0. Since [0, T ] is compact, there exists J N such that

J 
[


[0, T ]
tkj
, tk +
.
6L j
6L
j =1
Then there exists Nj such that for all m1 , m2 Nj ,
kum1 (tkj ) um2 (tkj )kW 2, (B

R)

< /3.

Taking N = maxj =1,...,J Nj , it is easy to check that for all m1 , m2 N,

sup kum1 (t) um2 (t)kW 2, (B

t[0,T ]

R)

< .

We find that the sequence is uniformly Cauchy in L ([0, T ]; W2, (BR )), and it converges strongly to an element of L ([0, T ]; W2, (BR )).
t
u
Now we apply Lemma 4.5 to prove the strong convergence which was claimed in
Section 4.4. It remains to prove that for any solution f to (15) we have f /t
L ([0, T ]; L (BR )) (but not uniformly) and that the second inequality in (23) holds
for all sufficiently small > 0 and for any R > 0.

On the global existence for the Muskat problem

225

Recall that f C([0, T ]; H 3 (R)). Then in (15) the first two linear terms are
bounded easily. The last term can be written as
Z
fx (x) fx (x ) (1 f (x))2
1
e
d
N L = C()3 f
()
1 + (1 f (x))2
R
e a constant, and therefore
for C()
|N L(x, t)| C()kf kH 3 (t),
by Sobolev embedding.
The norm of f /t W2, (BR ) is given by
Z



f


f



dx
=
sup
(x, t)(x).

t (t) 2,
t
W
(BR )
W 2,1 (B ): kk
1 R
R

W 2,1

Since vanishes on the boundary of BR , we can think of (x) as being zero outside of
the ball of radius R. Then we are allowed to integrate over the whole space R, which
is important because we want to estimate the non-local operator 31 in this norm via
integration by parts. Then we have
Z
Z
Z
I=
31 f (x)(x) dx =
31 f (x)(x) dx =
f (x)31 (x) dx,
BR

and therefore
|I | kf kL (t)k31 kL1 .
We compute
1

Z
(x) = c
R

(x) (x )
d =
||2

d = J1 (x) + J2 (x),

d +
||<1

||>1

thus
Z

Z
|J1 (x)| dx
R

||>1

d
||2

It is easy to rewrite J2 as follows:

Z
J2 (x) = c
||<1

Z
dx (|(x)| + |(x )|) CkkL1 (BR ) .
R

(x) (x ) x (x)
d,
||2

and therefore the identities

1

x (x + (s 1)) ds,

(x) (x ) =
0

(x) (x ) x (x) =

Z
(s 1) ds

dr xx (x + r(s 1))
0

226

Peter Constantin et al.

allow us to find
Z
Z
|J1 (x)| dx

Z
ds

||<1 0

Z
dx |xx (x + r(s 1))| 2kxx kL1 (BR ) .

dr
0

Now it follows from the a priori bounds that

k31 f kW 2, (B

R)

+ kfxx
kW 2, (B

R)

Ckf kL (R) Ckf0 kL (R) .

Here the constant is independent of and R > 0.

For the last term in (15), integrating by parts in the definition of the norm, we are led
to estimate
Z
Z
dx x (x) PV d arctan(1 f (x)).
R

Using exactly the arguments from Section 4.4 with say R = = 1 we have

Z
Z


 dx x (x) PV d arctan(1 f (x)) Ckk 1,1 kf0 kL (R) .
W

We thus deduce (23), completing the proof.

Acknowledgments. PC was partially supported by NSF grant DMS-0804380. DC and FG were
partially supported by MCINN grant MTM2008-03754 (Spain) and ERC grant StG-203138CDSIF.
FG was partially supported by NSF grant DMS-0901810. RMS was partially supported by NSF
grant DMS-0901463.

References
[1] Ambrose, D.: Well-posedness of two-phase Hele-Shaw flow without surface tension. Euro. J.
Appl. Math. 15, 597607 (2004) Zbl 1076.76027 MR 2128613
[2] Bailly, J. H.: Local existence of classical solutions to first-order parabolic equations describing
free boundaries. Nonlinear Anal. 32, 583599 (1998) Zbl 0934.35211 MR 1612026
[3] Bear, J.: Dynamics of Fluids in Porous Media. American Elsevier, New York (1972)
Zbl 1191.76001
[4] Bertozzi, A. L., Constantin, P.: Global regularity for vortex patches. Comm. Math. Phys. 152,
1928 (1993) Zbl 0771.76014 MR 1207667
[5] Constantin, A., Escher, J.: Wave breaking for nonlinear nonlocal shallow water equations.
Acta Math. 181, 229243 (1998) Zbl 0923.76025 MR 0972259
[6] Constantin, P., Foias, C.: NavierStokes Equation. Chicago Lectures in Math., Univ. of
Chicago Press, Chicago, IL (1988) Zbl 0687.35071
[7] Constantin, P., Pugh, M.: Global solutions for small data to the Hele-Shaw problem. Nonlinearity 6, 393415 (1993) Zbl 0808.35104 MR 1223740
[8] Cordoba, A., Cordoba, D., Gancedo, F.: Interface evolution: the Hele-Shaw and Muskat problems. Ann. of Math. 173, 477542 (2011) Zbl 1229.35204 MR 2753607
[9] Cordoba, D., Gancedo, F.: Contour dynamics of incompressible 3-D fluids in a porous
medium with different densities. Comm. Math. Phys. 273, 445471 (2007) Zbl 1120.76064
MR 2318314

On the global existence for the Muskat problem

227

[10] Cordoba, D., Gancedo, F.: A maximum principle for the Muskat problem for fluids with different densities. Comm. Math. Phys. 286, 681696 (2009) Zbl 1173.35637 MR 2472040
[11] Cordoba, D., Gancedo, F.: Absence of squirt singularities for the multi-phase Muskat problem.
Comm. Math. Phys. 299, 561575 (2010) Zbl 1198.35176 MR 2679821
[12] Cordoba, D., Gancedo, F., Orive, R.: A note on the interface dynamics for convection in porous
media. Phys. D, 237, 14881497 (2008) Zbl 1143.76574 MR 2454601
[13] Escher, J., Simonett, G.: Classical solutions for Hele-Shaw models with surface tension. Adv.
Differential Equations 2, 619642 (1997) Zbl 1023.35527 MR 1441859
[14] Escher, J., Matioc, B.-V.: On the parabolicity of the Muskat problem: Well-posedness,
fingering, and stability results. Z. Anal. Anwend. 30, 193218 (2011) Zbl 1223.35199
MR 2793001
[15] Escher, J., Matioc, A.-V., Matioc, B.-V.: A generalized RayleighTaylor condition for the
Muskat problem. Nonlinearity 25, 7392 (2012) Zbl 1243.35179 MR 2864377
[16] Kim, I.: Uniqueness and existence results on the Hele-Shaw and the Stefan problems. Arch.
Ration. Mech. Anal. 168, 299328 (2003) Zbl 1044.76019 MR 1994745
[17] Muskat, M., Wickoff, R. D.: The Flow of Homogeneous Fluids through Porous Media.
McGraw-Hill, London (1937) JFM 63.1368.03
[18] Saffman, P. G., Taylor, G.: The penetration of a fluid into a porous medium or Hele-Shaw
cell containing a more viscous liquid. Proc. R. Soc. London Ser. A 245, 312329 (1958)
Zbl 0086.41603 MR 0097227
[19] Siegel, M., Caflisch, R., Howison, S.: Global existence, singular solutions, and ill-posedness
for the Muskat problem. Comm. Pure Appl. Math. 57, 13741411 (2004) Zbl 1062.35089
MR 2070208
[20] Stein, E.: Harmonic Analysis: Real-Variable Methods, Orthogonality, and Oscillatory Integrals. Princeton Univ. Press, Princeton, NJ (1993) Zbl 0821.42001 MR 1232192
[21] Yi, F.: Local classical solution of Muskat free boundary problem, J. Partial Differential Equations 9, 8496 (1996) Zbl 0847.35152 MR 1384001
[22] Yi, F.: Global classical solution of Muskat free boundary problem, J. Math. Anal. Appl. 288,
442461 (2003) Zbl 1038.35083 MR 2019452