![MathJax Logo](/templates/jsp/_style2/_tandf/pb2/images/math-jax.gif)
Abstract
In this manuscript, we consider a non-local porous medium equation with non-local diffusion effects given by a fractional heat operator
in two space dimensions for
Global in time existence of weak solutions is shown by employing a time semi-discretization of the equations, an energy inequality and the Div-Curl lemma.
1. Introduction
In this manuscript, we study existence of weak solutions to a porous medium equation with non-local diffusion effects:
(1)
(1)
Here
denotes the density function and
the pressure. We analyze the problem when
and
The model describes the time evolution of a density function u that evolves under the continuity equation
where the velocity is conservative,
and p is related to
by the inverse of the fractional heat operator
Problem (Equation1(1)
(1) ) is the parabolic-parabolic version of a parabolic-elliptic problem recently studied in [Citation1]. In [Citation1], the authors proved the existence of sign-changing weak solutions to
(2)
(2)
For and
EquationEq. (2)
(2)
(2) reads as
The presence of makes our system quite different from (Equation2
(2)
(2) ). For example, techniques such as maximum principle and Stroock-Varopoulos inequality do not work. We overcome these significant shortcomings with the introduction of ad-hoc regularization terms, together with suitable compact embeddings and moment estimates. See later for a more detailed explanation.
A linear parabolic-elliptic version of (Equation1(1)
(1) )
(3)
(3)
was studied by the first author and collaborators in a series of papers: existence of weak solutions for (Equation3
(3)
(3) ) is proven in [Citation2–4] and Hölder regularity in [Citation5]. The case s = 1 also appeared in [Citation6] as a model for superconductivity.
Systems (Equation3(3)
(3) ) and (Equation1
(1)
(1) ) are reminiscent to a well-studied macroscopic model proposed for phase segregation in particle systems with long range interaction:
(4)
(4)
Any system that exhibits coexistence of different densities (e.g., fluid and vapor or fluid and solid) has equilibrium configurations that segregate into different regions; the surface of these regions are minimizers of a free energy functional. The relaxation to equilibrium of the density function u(x, t) can be described in general by nonlinear integro-differential equations of type (Equation4(4)
(4) ). One example is the model proposed in [Citation7], in which the mobility is
and the kernel K is bounded, symmetric and compactly supported. Such model describes the hydrodynamic (or mean-field) limit of a microscopic model undergoing phase segregation with particles interacting under a short-range and long-range Kac potential. Several other variants of (Equation4
(4)
(4) ) are present in the literature [Citation7–11]. We also mention [Citation12] for the study of a deterministic particle method for heat and Fokker–Planck equations of porous media type where the non-locality appears in the coefficients. The long time behavior of weak solutions to (Equation1
(1)
(1) ) was studied in [Citation13]. There the authors show algebraic decay in time toward the stationary solutions u = 0 and
The condition that the pressure satisfies a parabolic equation introduces non-trivial complications in the analysis of (Equation1(1)
(1) ). The non-local structure prevents the equation from having a comparison principle. Moreover, maximum principle does not give useful insights, since at any point of maximum for u we only know that
We overcome the lack of comparison and maximum principles with the introduction of several regularizations. Stampacchia’s truncation arguments yield non-negativity of the solutions and the Div-Curl lemma will be used to identify the limit for
The main result of this manuscript is summarized in the following theorem:
Theorem 1.
Let . Moreover let
be functions such that
. There exist functions
such that for every T > 0
which satisfy the following weak formulation to (Equation1
(1)
(1) ):
(5)
(5)
(6)
(6)
as well as the mass conservation relation
The starting point about our analysis is the observation that
is a Lyapunov functional for (Equation1
(1)
(1) ) and satisfies the bound
Indeed, formal computations show that
Testing the equation for p against we obtain
which leads to
(7)
(7)
The major difficulty, in the approximation process, is the identification of the limit of The energy inequality (Equation7
(7)
(7) ) provides plenty of information for the pressure p, but only uniform integrability in
for u. At the moment it is unclear to the authors how to use the bounds for
to get useful bounds for
or u. To overcome the lack of compactness we employ the Div-Curl Lemma (see [Citation14]) to the vector fields
where
is a suitable approximate solution to (Equation1
(1)
(1) ). The argument yields
where U, V are the weak limits of
respectively. Strong convergence of
and standard result in compensated compactness theory [Citation15] yield strong convergence for
The application of the Div-Curl Lemma brings two restrictions on the system. The first one concerns the lower bound for s, the second one the dimension. It is unclear how to remove such restrictions, as they seem necessary to fulfill the integrability and compactness constraints on the quantities
The assumptions on s, β, and d are not satisfactory from the point of view of a general theory for weak solutions. As such, Theorem 1 is a first step to understand the complete behavior of (Equation1
(1)
(1) ). Most interesting, however, is the fact that the addition of a nonstationary term in the pressure equation radically changes the behavior of the system and calls for a different analytical setting than in [Citation3, Citation11]. We also point out that the successful use of the Div-Curl Lemma, a tool commonly employed in the study of fluid-dynamic systems, in the analysis of nonlocal diffusion equations is (to our best knowledge) a novelty and an unexpected connection between the two fields. Uniqueness of weak solutions is an important open question for our system. We expect it to hold for short time straightforwardly. For long time the only available result so far is the one in [Citation13], in which the authors show a weak-strong uniqueness result: if there exists a strong solution, then any weak solution with the same initial data coincides with it.
Existence of a solution for β = 1 appears to be out of reach with the present technique, as several other terms will lack compactness.
The article is organized as follows: in Section 2, we show two preliminary technical lemmas, and in Section 3 the proof of the main theorem.
2. Some technical results
Lemma 1
. Let be a continuous, nondecreasing function such that
. For
define the functional space
as
Then is compactly embedded in
for any
Proof.
Let be a uniformly bounded sequence in
We first notice that there exists a subsequence, still denoted with fn such that
Denote with BR the ball of center x = 0 and radius R. Since
is compactly embedded in
for any
there exists a subsequence of fn, still denoted with fn, such that
Thanks to a Cantor diagonal argument, the subsequence fn can be chosen to be independent of R. The uniform bound for fn in and Fatou’s Lemma imply that
Next we show that strongly in
for n big enough
by choosing R big enough. Interpolation between
and
implies that for any q with
the sequence fn strongly converges to f in
□
Lemma 2.
Define with
and for every
we set
. For s > 0 we have
Proof.
The result is a consequence of the scaling property of the fractional Laplacian:
3. Proof of the main theorem
Define the spaces
where
Thanks to Sobolev’s embedding and Lemma 1:
(8)
(8)
(9)
(9)
In particular, the embedding is compact.
For every measurable function we denote by
and
its positive and negative part, respectively.
For given constants functions
and
such that
a.e. in
consider the time-discrete problem
(10)
(10)
(11)
(11)
We divide the proof of Theorem 1 into several steps: we first show existence of solution to (Citation10, Citation11) by Leray-Schauder fixed point theorem. Then we perform the limits and
(in this order). The last limit is the most complicated because we need compactness for u without relying on the term
3.1. Existence for (10)–(11)
For given constants functions
and
such that
a.e. in
consider the linear problem in the variable w:
(12)
(12)
(13)
(13)
We first solve (Equation13(13)
(13) ). We have that
Lax-Milgram Lemma yields the existence of a unique solution
Standard elliptic regularity results imply that
and consequently
for every
We now solve (Equation12(12)
(12) ). Since
and
for every
the linear mapping
is continuous. The nonlinear operator
defined by
for every
is strictly monotone, coercive, hemicontinuous. Therefore, the standard theory of monotone operators [Citation16] yields the existence of a unique solution
to (Equation12
(12)
(12) ).
We can now define the mapping
where
is the unique solution to (Citation12, Citation13). Clearly
is a constant mapping. Moreover F is continuous and also compact due to the compact embedding
see Lemma 1.
Next, we show that any fixed point is nonnegative and uniformly bounded in σ. We use a Stampacchia truncation argument. This method is generally used in nonlinear elliptic problems to show positivity, boundedness and higher regularity via the choice of particular test functions. In our case, by choosing and
as test functions, we get
from which it follows that
a.e. in
The nonnegativity of w and the
-regularity of p allow for the formulation
(14)
(14)
(15)
(15)
where we defined
We now search for uniform bounds with respect to σ: choosing in (Equation14
(14)
(14) ) leads to
On the other hand, multiplying (Equation15(15)
(15) ) by
and integrating in
yields
Given that
we deduce
(16)
(16)
The above estimate yields a bound for
in Y which is uniform in σ. Together with the embedding
we have that u belongs to X, with
bounded uniformly with respect to σ. Leray-Schauder fixed point theorem yields the existence of a fixed point
for
i.e., a solution
to
(17)
(17)
(18)
(18)
such that
a.e. in
and (Equation16
(16)
(16) ) holds for σ = 1:
(19)
(19)
3.2. The limit ![](//:0)
![](//:0)
The next step is to take the in (Equation17
(17)
(17) )–(Equation19
(19)
(19) ).
The uniform bound of in
(see (Equation19
(19)
(19) )) and Sobolev’s embedding insure that for every R > 0 there exists a subsequence
of
such that
The function u is the weak limit of in
By a Cantor diagonal argument we can find a subsequence (not relabeled) of
such that
as well as
a.e. in
As a consequence
(20)
(20)
Going back to the limit in (Equation18(18)
(18) ) and (Equation17
(17)
(17) ) we have that as
where we used (Equation20
(20)
(20) ) for the first limit, and (Equation20
(20)
(20) ) together with
in
to obtain the second limit (remember that
is relatively weakly compact in
). Summarizing, taking the limit
in (Citation17, Equation18
(18)
(18) ) and subsequently employing a standard density argument we get
(21)
(21)
(22)
(22)
Moreover a.e. in
and
(23)
(23)
Let
for every
By testing (Equation22
(22)
(22) ) against
and exploiting the fact that
one deduces the estimate
Taking the limit in the above inequality (by monotone convergence) yields
Let ηR as in the statement of Lemma 2. Multiplying (Equation22(22)
(22) ) by ηR, integrating in
and integrating by parts leads to
(24)
(24)
Since as
(see Lemma 2) and
the bound for the mass of p follows
At this point, we have proved the existence of sequences such that
and for
a.e. in
(25)
(25)
(26)
(26)
with the estimates
(27)
(27)
(28)
(28)
Choose T > 0 arbitrary. Define
Moreover define the backward finite difference w.r.t. time
as
We can rewrite (Equation25(25)
(25) )–(Equation28
(28)
(28) ) with the new notation. For all
and
we have
(29)
(29)
(30)
(30)
(31)
(31)
(32)
(32)
where the constant in (Equation32
(32)
(32) ) only depends on the entropy at initial time.
3.3. The limit ![](//:0)
![](//:0)
We first estimate the time derivative of the density function. Let R > 0 arbitrary, For any
using (Equation31
(31)
(31) ). This yields
(33)
(33)
In particular,
(34)
(34)
The compact Sobolev embedding valid for every
allows us to apply Aubin-Lions Lemma in the version of [Citation17] and obtain, for any R > 0, the existence of a subsequence
of
such that
The limit function u is unique and coincides with the weak-* limit of in
A Cantor diagonal argument allows us to find a subsequence of
(which we denote again with
) such that
and
(35)
(35)
Since a straightforward interpolation yields
(36)
(36)
Since thanks to (Equation35
(35)
(35) ) it follows
(37)
(37)
Hence as
Moreover directly from (Equation31(31)
(31) )
From (Equation33(33)
(33) ), (Equation37
(37)
(37) ) it follows
Since is uniformly bounded in
and
is uniformly bounded in
Gagliardo-Nirenberg and the entropy inequality (Equation31
(31)
(31) ) yield
(38)
(38)
where C only depends on the initial data. Hence there exists a subsequence of
(which we denote again with
) such that
In particular,
(39)
(39)
Also, by Sobolev’s embedding,
The strong convergence in
for every R > 0, the weak convergence of
in
and the assumption
imply
Let us look at the discrete time derivatives of the pressure function. Thanks to (Equation36(36)
(36) ) we have
while (Equation31
(31)
(31) ) implies
We deduce
(40)
(40)
It follows
(41)
(41)
Since is bounded in
and
is bounded in
we can invoke Aubin-Lions lemma to deduce, for every
the existence of a subsequence
of
such that
strongly in
for every
A Cantor’s diagonal argument yields the existence of a subsequence of
(which we call again
) such that
(42)
(42)
At this point we can take the limit in (Equation29
(29)
(29) ) and (Equation30
(30)
(30) ), which yields (after a suitable density argument)
(43)
(43)
(44)
(44)
where
is defined in (Equation36
(36)
(36) ).
Thanks to the lower weak semicontinuity of the Lp norm we deduce from (Equation31(31)
(31) ) the following entropy inequality:
(45)
(45)
Furthermore, thanks to the a.e. convergence of (42) we can apply Fatou’s Lemma in (Equation32
(32)
(32) ) and get
(46)
(46)
3.4. The limit ![](//:0)
![](//:0)
From the entropy inequality (Equation45(45)
(45) ) and the mass conservation (Equation46
(46)
(46) ) we deduce the following
uniform bounds:
(47)
(47)
(48)
(48)
Moreover, from (Equation43(43)
(43) ), (Equation44
(44)
(44) ), (Equation47
(47)
(47) ), (Equation48
(47)
(47) ) we deduce
uniform bounds for the time derivatives of u, p:
(49)
(49)
Estimates (Equation47(47)
(47) )–(Equation49
(49)
(49) ) and the compact Sobolev embeddings
valid for every bounded open
and
allow us to apply Aubin-Lions Lemma and deduce, for every
the existence of subsequences
of
such that
for every
Once again, a Cantor diagonal argument allows us to find subsequences (not relabeled) of
such that
for every
Bounds (Equation47
(47)
(47) ), (Equation48
(48)
(48) ) also imply (up to subsequences) the following weak convergence relations
Thanks to the convergence relations stated above, taking the limit in (Equation43
(43)
(43) ), (Equation44
(44)
(44) ) is at this point straightforward and leads to
(50)
(50)
(51)
(51)
where
is defined in (Equation36
(36)
(36) ).
The same convergence relations yield
(52)
(52)
We also point out that (Equation46(46)
(46) ) holds true also after taking the limit
3.5. The limit ![](//:0)
![](//:0)
In the rest of the article, we denote ρ1 with ρ.
As a preliminary step, we are going to prove a uniform bound for By interpolation we obtain
with
The assumption
allows for the choice
such that
and, therefore,
for some
Since
is bounded in
by Sobolev’s embedding it is also bounded in
Together with the uniform bound in
we conclude
(53)
(53)
Now we wish to prove a uniform bound for in
Let us choose
in (Equation50
(52)
(52) ), (Equation51
), respectively, and sum the resulting equations. We obtain
(54)
(54)
Let us bound the terms on the right-hand side of (Equation54(54)
(54) ) by using bounds (Equation46
(46)
(46) ), (Equation52
). Applying Hölder and Gagliardo-Nirenberg inequalities yields
for some
Let us then consider
thanks to (Equation53
(53)
(53) ). Next we notice that
Finally, Gagliardo-Nirenberg inequality allows us to write
for some
From (Equation54
(54)
(54) ) we conclude
which implies, via Young’s inequality,
(55)
(55)
Next we find a suitable bound for Since
and
are bounded in
and
respectively, then
is bounded in
On the other hand,
and
are also bounded in
and
respectively, so
is also bounded in
A straightforward interpolations leads to
(56)
(56)
Now we prove the strong convergence of From (Equation52
), (Equation55
(55)
(55) ) it follows that
(57)
(57)
From (Equation52) and (Equation57
(57)
(57) ) we deduce via Aubin-Lions Lemma and a Cantor diagonal argument that, up to subsequences,
Bound (Equation55(55)
(55) ) implies that, up to subsequences,
(58)
(58)
for some function
We are now going to show that
a.e. in
Let us now consider the vector fields
Let
It is easy to see that
Let
bounded open smooth domain.
Bound (Equation56(56)
(56) ) means that
is bounded in
for i = 1, 2, while
is bounded in
thanks to (Equation55
(55)
(55) ). In particular
is bounded in
for i = 0, 1, 2.
On the other hand, (Equation52) and (Equation55
(55)
(55) ) imply that
is bounded in
while
is bounded in
for i = 1, 2 thanks to (Equation53
(53)
(53) ) and the trivial relation
which holds thanks to the hypothesis
It follows that
is bounded in
for i = 0, 1, 2.
Next we notice that
thanks to (Equation52
). On the other hand,
since
is a gradient field.
Therefore, we are able to apply [Citation14, Thr. 1.1] and deduce that
where U, V are the weak limits of
respectively. This implies, being
bounded in
(59)
(59)
where u,
are the weak limits of
respectively. However, we know that
strongly in
while
is bounded in
and
thanks to (Equation52
), (Equation53
(53)
(53) ). Therefore, Gagliardo-Nirenberg inequality allows us to deduce
strongly in
It follows
(60)
(60)
From the relations above and (Equation59(59)
(59) ) we deduce
(61)
(61)
Again, the local-in-space strong convergence of and the known uniform bounds for
in
and
imply via Gagliardo-Nirenberg inequality that
strongly in
This fact, together with the weak convergence
in
and relation
implies that
(62)
(62)
Summing (Equation61(61)
(61) ), (Equation62
(62)
(62) ), employing (Equation51
(51)
(51) ) and the uniform bound for
in
leads to
(63)
(63)
where v is the weak limit of
and
is the space of Radon measures, i.e., the dual of
We are going to show that (Equation63(63)
(63) ) implies the a.e. convergence of
in
Define the truncation operator Tk as
for every
Let
in
arbitrary. Relation (63) implies
and so
(64)
(64)
On the other hand [Citation15, Thr. 10.19] implies
(65)
(65)
The weak lower semicontinuity of the L1 norm yields
The uniform bound for in
implies
which implies
weakly in
as
and so
(66)
(66)
From (Equation64(64)
(64) )–(Equation66
(66)
(66) ) we deduce
which easily implies
(67)
(67)
However, elementary computations yield
which implies that the sequence
is nondecreasing and nonnegative. Moreover
thanks to (Equation67
(67)
(67) ). Therefore, ak = 0 for every
that is
In particular
(68)
(68)
where we defined
It is easy to prove the elementary relation
which, together with (Equation68
(68)
(68) ), leads to
(69)
(69)
Fix arbitrary. Let us consider
From (Equation69(69)
(69) ) and the uniform bound for
in
we deduce
Since the left-hand side of the above inequality does not depend on k, we conclude
(70)
(70)
By choosing such that
on
for
arbitrary (where BR is the ball of
with center 0 and radius R) we conclude from (Equation70
(70)
(70) ) that
is a Cauchy sequence in
(and, therefore, strongly convergent in such space) for every
In particular, for every R > 0 there exists a subsequence
of
that is a.e. convergent in QR. A Cantor diagonal argument yields the existence of a subsequence (not relabeled) of
that is a.e. convergent in QR for every
and, therefore,
a.e. in
The a.e. convergence of and the boundedness of
in
imply
(71)
(71)
Finally, since is bounded in
(see (Equation55
(55)
(55) )), while
are bounded in
(from (Equation52
)), we deduce
As a consequence
and so
(72)
(72)
Putting the previous limit relations together allow us to take the limit inside (Equation50
(52)
(52) ), (Equation51
) and obtain a solution to (Citation5, Citation6) (after a suitable density argument). Finally, we show the mass conservation property. Define the cutoff
Let arbitrary. Choosing
inside (Equation5
(5)
(5) ) yields
Since and
it follows
(73)
(73)
with
Choosing
in
taking the limit
inside (Equation73
(73)
(73) ) and applying the monotone convergence theorem yields
(since T > 0 is arbitrary). At this point we can apply the dominated convergence theorem to take the limit
inside (Equation73
(73)
(73) ) with
arbitrary and deduce
implying that the mass
is constant in time. This concludes the proof of Theorem 1.
Acknowledgment
MPG would like to thank NCTS Mathematics Division Taipei for their kind hospitality.
Additional information
Funding
References
- Biler, P., Imbert, C., Karch, G. (2015). The nonlocal porous medium equation: Barenblatt profiles and other weak solutions. Arch. Rational Mech. Anal. 215(2):497–529. DOI: 10.1007/s00205-014-0786-1.
- Caffarelli, L., Vazquez, J.L. (2016). Regularity of solutions of the fractional porous medium flow with exponent 1/2. St. Petersburg Math. J. 27(3):437–460. DOI: 10.1090/spmj/1397.
- Caffarelli, L., Vazquez, J.L. (2011). Nonlinear porous medium flow with fractional potential pressure. Arch. Rational Mech. Anal. 202(2):537–565. DOI: 10.1007/s00205-011-0420-4.
- Serfaty, S., Vazquez, J.L. (2014). A mean field equation as limit of nonlinear diffusions with fractional Laplacian operators. Calc. Var. 49(3–4):1091–1120. DOI: 10.1007/s00526-013-0613-9.
- Caffarelli, L., Soria, F., Vazquez, J.L. (2013). Regularity of solutions of the fractional porous medium flow. J. Eur. Math. Soc. 15(5):1701–1746. DOI: 10.4171/JEMS/401.
- Ambrosio, L., Serfaty, S. (2008). A gradient flow approach to an evolution problem arising in superconductivity. Commun. Pure Appl. Math. 61(11):1495–1539. DOI: 10.1002/cpa.20223.
- Giacomin, G., Lebowitz, J. (1997). Phase segregation dynamics in particle systems with long range interactions. I. Macroscopic limits. J. Stat. Phys. 87(1–2):37–61. DOI: 10.1007/BF02181479.
- Giacomin, G., Lebowitz, J. (1998). Phase segregation dynamics in particle systems with long range interactions. II. Interface motion. SIAM J. Appl. Math. 58(6):1707–1729. DOI: 10.1137/S0036139996313046.
- Giacomin, G., Lebowitz, J., Marra, R. (2000). Macroscopic evolution of particle systems with short- and long-range interactions. Nonlinearity. 13(6):2143–2162. DOI: 10.1088/0951-7715/13/6/314.
- Rezakhanlou, F. (1991). Hydrodynamic limit for attractive particle systems on Zd. Commun. Math. Phys. 140(3):417–448. DOI: 10.1007/BF02099130.
- Stan, D., del Teso, F., Vazquez, J. L. (2019). Existence of weak solutions for a general porous medium equation with nonlocal pressure. Arch. Rational Mech. Anal. 233(1):451–496. DOI: 10.1007/s00205-019-01361-0.
- Lions, P. L., Mas-Gallic, S. (2001). Une méthode particulaire déterministe pour des équations diffusives non linéaires. C. R. Acad. Sci. Paris Sér. I Math. 332(4):369–376. DOI: 10.1016/S0764-4442(00)01795-X.
- Daus, E., Gualdani, M., Zamponi, N. (2020). Long time behavior and weak-strong uniqueness for a nonlocal porous media equation. J. Differ. Equations. 268(4):1820–1839. DOI: 10.1016/j.jde.2019.09.029.
- Gasser, I., Marcati, P. (2008). On a generalization of the Div-Curl Lemma. Osaka J. Math. 45:211–214.
- Feireisl, E., Novotný, A. (2009). Singular Limits in Thermodynamics of Viscous Fluids. Basel: Birkhäuser.
- Zeidler, E. (1990). Nonlinear Functional Analysis and Its Applications. II /B. New York, NY: Springer-Verlag.
- Chen, X., Jüngel, A., Liu, J.-G. (2014). A note on Aubin-Lions-Dubinskií lemmas. Acta Appl. Math. 133(1):33–43. DOI: 10.1007/s10440-013-9858-8.