ON WEAK SOLUTIONS TO A DIFFUSE INTERFACE MODEL OF A BINARY MIXTURE OF COMPRESSIBLE FLUIDS

. We consider the Euler-Cahn-Hilliard system proposed by Lowengrub and Truskinovsky describing the motion of a binary mixture of com- pressible ﬂuids. We show that the associated initial-value problem possesses inﬁnitely many global-in-time weak solutions for any ﬁnite energy initial data. A modiﬁcation of the method of convex integration is used to prove the result.


1.
Introduction. The mathematical theory of fluid dynamics based on continuum models faces several fundamental difficulties including the problem of well-posedness of the Euler and Navier-Stokes systems on large time intervals and arbitrary initial data. The situation has become even more delicate in the light of the recent ground breaking discoveries of DeLellis and Székelyhidi [5], [6] based on the technique of the so-called convex integration. There is a body of evidence that the classical admissibility criteria of well-posedness of hyperbolic systems of conservation laws based on the Second law of thermodynamics fail to identify a unique weak solution. On the other hand, however, the weak solutions are the only available given the inevitable presence of discontinuities -shock waves -that may develop in a finite time regardless the smoothness of initial data.
Convex integration shows how vulnerable with respect to apparently non-physical perturbations are the weak solutions at the moment we are not able to control the mechanical energy dissipation. Rather surprisingly but by the same token, the same method can be used to "construct" global in time weak solutions to a vast class of sofar open problems, among which the compressible Euler system (see Chiodaroli [2], Chiodaroli, DeLellis and Kreml [3]). In the light of these arguments, uniqueness rather than existence of solutions to problems of gas dynamics became an issue.
Very roughly indeed, the method proposed by DeLellis and Székelyhidi [5] for hyperbolic problems of the form works as follows: • In the spirit of the seminal work of Tartar [12], the system (1) is understood as a linear problem ∂ t v + div x F = 0,

EDUARD FEIREISL
supplemented with a (nonlinear) constitutive relation • The constitutive relation (2) is replaced by an "implicit" one where G is a convex functional such that • The original problem is relaxed to finding "subsolutions" satisfying the inequality (4), more specifically, where E is a prescribed "energy". • The desired equality (3) is achieved by modulating suitable oscillations on a given subsolution. The existence of (infinitely) solutions is obtained via Baire's category argument. In addition, the solutions satisfy |v| 2 = E. In this paper, we extend the method to the case when the energy E = E[v] is allowed to depend on the field v. To be more specific, we focus on a physically motivated regularization of the Euler equations proposed in the seminal paper by Lowengrub and Truskinovsky [9]. The model describes the motion of a mixture of two immiscible compressible fluids in terms of the density = (t, x), the macroscopic velocity u = u(t, x), and the concentration difference c = c(t, x), where t ∈ (0, T ) is the time and x ∈ Ω ⊂ R 3 the reference Eulerian coordinate. The fluid is described by means of the standard Euler system coupled with the Cahn-Hilliard equation describing the evolution of c. The resulting system of equations reads: where for a given free energy function f 0 . Although the physical relevance of such a model might be dubious because of the absence of viscous stress in the momentum equation (7), the system (7 -9) can be seen either as a regularization of the Euler equations or as an inviscid limit of the associated Navier-Stokes-Cahn-Hilliard system. From the mathematical view point, the system is neither purely hyperbolic nor parabolic as the dissipation mechanism acts in a very subtle way through the coupling of the Euler and the Cahn-Hilliard systems. To the best of our knowledge, such a problem has never been studied in the framework of weak solution, in particular in the physically relevant 3-D setting.
As we show in this paper, the problem (6-8) is very close to the (inviscid) Euler system as the existence of weak solutions may be obtained by the method of convex integration. Thus, despite the presence of diffusion in (8), the system still admits large oscillations of the velocity field. Our method is based on a modification of DeLellis and Székelyhidi's argument, namely the energy E in (5) is replaced by a functional depending on the solution v itself provided the mapping v → E(v) is continuous or at least upper semi-continuous with respect to the topology of uniform convergence. The maximal regularity estimates for non-constant coefficients evolutionary equations play a crucial role in our arguments. A similar method has been applied to the Euler-Fourier system in [4]. As we shall see below, the present problem is much more delicate than [4] as we have to establish compactness of the gradient terms appearing on the right-hand side of (7).
The paper is organized as follows. In Section 2, we recall the standard definition of weak solution to the problem (6 -8) and state our main result. In Section 3, the system (6 -8) is rewritten in a form suitable for application of the machinery of convex integration. Global existence of (infinitely many) weak solutions is proved in Section 3. We finish by a brief discussion on possible extensions of the theory in Section 5.
2. Weak solutions, main result. We consider the system (6 -8) in the physically relevant 3-D setting. For the sake of simplicity, we impose the space-periodic boundary conditions. Accordingly, the physical domain Ω ⊂ R 3 is the flat torus, For technical reasons, we also impose certain restrictions on the function f 0 , the latter being typically of the form where α 1 , α 2 , β > 0 are positive constants. To avoid possible difficulties with singularities, we replace the function by a smooth (bistable) potential H. Specifically, we suppose • the functions , u, c belong to the class • the equation of continuity (6) is satisfied a.a. in (0, T ) × Ω; • the momentum equation (7) is replaced by a family of integral identities Remark 2. As a matter of fact, the attribute "weak" is appropriate only for the momentum equation (7), the remaining two equations being satisfied in the strong sense.

Main result.
Our main result concerning solvability of the initial-value problem for the system (6-8) reads as follows.
The rest of the paper is devoted to the proof of Theorem 2.1. The physical relevance of the result will be discussed in Section 5.
3. Apparatus of convex integration, oscillatory lemma. We start by rewriting the problem to fit the framework of convex integration. First we decompose where ∇ x Φ may be seen as the gradient part in the Helmholtz projection. Similarly, we write (15) Accordingly, the system (6-8) reads 3.1. Density ansatz. Next, we fix the density in such a way that , where Φ 0 is the potential determined by (15). Next, we identify the potential Φ as the unique solution of the elliptic problem Having fixed , Φ we observe that (16) holds and that obviously satisfies the relevant initial condition in (14).

Remark 3.
A striking feature of the method, already exploited by Chiodaroli [2], is that the density can be chosen in an almost arbitrary way in contrast with c that will be computed on the basis of and u.

Cahn-Hilliard equation.
With , Φ fixed, we claim that the equation (18), supplemented with the initial condition c(0, ·) = c 0 , admits a unique solution c = c[v] in the class specified in (11) for any given Indeed uniqueness for a given v can be seen by taking the difference c 1 − c 2 of two possible solutions and integrating the difference of the corresponding equations multiplied on c 1 − c 2 : where, by interpolation,

Moreover, by virtue of the hypothesis (10), the function H is globally Lipschitz and uniqueness follows (19) by Gronwall's argument.
Remark 4. This is the only point in the proof where we need the hypothesis (10).
It is worth noting that (19) is in fact independent of the field v. In particular, using the same argument we deduce that sup t∈[0,T ] c(t, ·) L 2 (Ω) + c L 2 (0,T ;W 2,2 (Ω)) ≤ C, where the bound is independent of v.
The proof of existence of c for any given v is a routine matter and relies on the available a priori estimates. These will be discussed in Section 4.1 below.

Velocity equation.
In the light of the previous discussion, the problem (16-18) reduces to finding a bounded measurable vector field where c = c[v] is the unique solution of the Cahn-Hilliard equation discussed in the preceding section and Λ = Λ(t) is a suitable spatially homogeneous function to be chosen below.

3.4.
Subsolutions. Let R 3×3 0,sym denote the space of all traceless symmetric matrices in R 3 , with λ max [A] denoting the maximal eigenvalue of a symmetric matrix A.
Following the strategy of DeLellis and Székelyhidi [5] , we introduce the set of subsolutions: Note that 1 2 for any w, V, whereas the equality holds only if In contrast with [5], the function E is allowed to depend on w.
The last ingredient we need to run the convex integration machinery is the following oscillatory lemma, see [8, Lemma 3.1]: sym,0 ), e, r ∈ C(U ), r > 0, e ≤ e in U are given such that Then there exist sequences and This lemma is a "singular" variant of similar results proved by Chiodaroli [2], and DeLellis, Székelyhidi [5]. We point out that the functions g, W are continuous but not necessarily bounded on the open set U , and, similarly, r need not be bounded below away from zero. Note, however, that such generality is not really needed in the present paper.

Existence of infinitely many solutions.
Having collected all the necessary material, we are ready to prove Theorem 2.1. To begin, fix the function Λ(t) in E (cf. (23)) large enough for the space X 0 to be non-empty. As a matter of fact, we can do it is such a way that the cardinality of X 0 is infinite.

4.1.
Boundedness of the set of subsolutions. The first necessary step is to show that the set X 0 is bounded in L ∞ ((0, T ) × Ω; R 3 ). Note that this is not completely obvious as we only know from (24) that Thus, in order to conclude that (26) yields a uniform bound on w we have to derive bounds for c, ∇ x c in terms of w.
Arguing by contradiction, we suppose that v ∈ X 0 is a point of continuity of I such that I[v] < 0.
Since I is continuous at v, there exists a sequence v m ∈ X 0 (with the associated In agreement with the definition of X 0 we have Fixing m we apply Lemma 3.1 choosing Note that, by virtue of the uniform bound (28), the energies E[v m ] are bounded uniformly in X 0 . We consider the sequence v m,n = v m + w m,n , where {w m,n } ∞ n=1 is the sequence contructed in Lemma 3.1. Clearly, v m,n and the associated fluxes V m,n = V m + V m,n satisfy ∂ t v m,n + div x V m,n = 0, div x v m,n = 0, v m,n (0, ·) = v 0 .
Moreover, in accordance with the conclusion of Lemma 3.1, Consequently, by virtue of (29), we may assume that for all n ≥ n(m) large enough, we have v m,n ∈ X 0 , and, extracting a suitable diagonal subsequence, we may suppose We have proved Theorem 2.1.
5. Discussion. Apparently, at least some if not all solutions the existence of which is claimed in Theorem 2.1 are not physically admissible. Similarly to their counterparts constructed by DeLellis and Székelyhidi [5], Scheffer [10], Shnirelman [11] for the incompressible Euler system, they violate the First law of thermodynamics in the sense that the total energy of the system, On the other, however, the method of the present paper can be used, exactly as in [4], to produce the following result: With the tools developed in this paper, the proof of Theorem 5.1 can be carried over in the same way as [4,Theorem 4.2].
It would be interesting to identify the class of the initial data u 0 for which Theorem 5.1 applies, in particular, whether or not such data may be attained by a regular solution in a finite time. The recent examples concerning solutions of the Riemann problem for the compressible Euler system indicate that it might be possible, see Chiodaroli, DeLellis, and Kreml [3].