Existence of weak solutions to a cross-diffusion Cahn-Hilliard type system (Q2020131)

The paper is concerned with the analysis of partial differential equations describing the evolution of multicomponent systems partly subject to phase-separation. The \(n+1\) chemical substances are assumed filling a bounded region \(\Omega\) of \(\mathbb{R}^d\) (\(d = 1,2,3\)). In the cylinder domain \((0,T) \times \Omega\), the volume fractions \(u_0, \, u_1, \ldots, u_n\) of the species are subject to \[ \partial_t u_i = \operatorname{div}v \Big(\sum_{j=0}^n M_{ij}(u_0, \ldots,u_n) \, \nabla \mu_j\Big), \;\; 0 \leq u_i \leq 1 \, \text{ for } \ i = 0,\ldots,n \,,\;\; \sum_{i=0}^n u_i = 1 \, . \tag{1} \] The mobility matrix \((M_{ij}(\mathbf{u}))_{0\leq i,j\leq n}\) obeys the particular Ansatz \[ M_{ij}(\mathbf{u}) := \begin{cases} - K_{ij} \, u_i \, u_j & \text{ for all } 0 \leq i \neq j \leq n \, ,\\ u_i \, \sum_{0 \leq \ell \neq i \leq n} K_{i\ell} \, u_j & \text{ for all } i = 0, \ldots, n, \ j = i \, , \end{cases}\tag{2} \] with \(K_{ij} = K_{ji} > 0\) for all \(0 \leq i \neq j \leq n\). Hence, \(M_{ij}(\mathbf{u})\) tends to zero for \(u_i \rightarrow 0+\) (degenerate mobility). The chemical potentials \(\mu_0, \ldots,\mu_{n}\) are defined by \[ \mu_i := \ln u_i \ \text{ for } \ i = 1,\ldots,n, \qquad \mu_0 := \ln u_0 - \epsilon \, \Delta u_0 + \beta \, (1-2u_0) \, ,\tag{3} \] and they correspond to the variational derivatives \(D_{u_i}E\) of the energy functional \[ E(\mathbf{u}) := \int_{\Omega} \sum_{i=0}^n (u_i \, \ln u_i - u_i + 1) + \frac{\epsilon}{2} \, |\nabla u_0|^2 + \beta \, u_0 \, (1-u_0) \, dx \, . \] The first contribution \(\sum_{i=0}^n (u_i \, \ln u_i - u_i + 1) \) describes the entropic mixing of the components. The term \(\frac{\epsilon}{2} \, |\nabla u_0|^2 + \beta \, u_0 \, (1-u_0) \) with \(\epsilon, \, \beta > 0\) is a Ginzburg-Landau phase-field regularisation for the interfacial energy between the zeroth species and the other species in the case that the surface tension coefficient is independent on composition. The main new contribution of the present paper is the global weak solution analysis for the fourth order PDE system (1), (2), (3) in connection with the initial and boundary conditions \begin{gather*} \mathbf{u}(0, \, x) =\mathbf{u}^0(x) \ \text{ in } \ \Omega \, , \\ \sum_{j = 0}^N M_{ij}(\mathbf{u}) \, \nabla_{\nu} \mu_j = 0 \text{ for } i = 0, \ldots,n, \qquad \nabla_{\nu} u_0 = 0 \ \text{ on } \ (0,T) \times \partial \Omega \, . \end{gather*} For comparison: In the literature, existence results for multicomponent Cahn-Hilliard systems were provided -- for instance by C.~Elliott and H.~Garcke -- in the case that all components separate from each other. Likewise, well-posedness studies for entropy-driven mixtures subject to cross-diffusion are widely available in the context of the \emph{boundedness-by-entropy method}, for instance in the work by A.~Jüngel. However, the quite natural scenario that one component separates from a mixture seems to have not yet been studied. In this sense, we can agree with the authors claiming that the paper extends both the theory of multicomponent Cahn-Hilliard systems and the boundedness-by-entropy method. As to this natural interpretation of the model -- the binary phase-separation between the zeroth species and, on the other hand, the mixture of the remaining \(n\) species -- let us remark that it should still await the results of the asymptotics \(\epsilon \rightarrow 0\), which shall yet have to be motivated in the present context. An interesting natural additional question would also be considering a coefficient of surface-tension depending on the composition of the mixture in the energy functional. The weak solution concept for (1) essentially uses the energy dissipation principle (gradient flow property) \[ E(\mathbf{u})(t) + \int_0^t \int_{\Omega} \sum_{i,j=0}^n M_{ij}(\mathbf{u}) \, \mu_i \, \mu_j \, dxds = E(\mathbf{u}^0) \, \ \text{ for all } \ 0 \leq t \leq T. \] If the initial data have bounded energy, then \(\sup_{t \leq T} E(\mathbf{ u})(t) < + \infty\), implying a bound for the component \(u_0\) in \(L^{\infty}(0,T; \, H^1(\Omega))\). A refined analysis of the total dissipation \(\int_0^T\int_{\Omega} M(\mathbf{u}) \, {\boldsymbol \mu} \, {\boldsymbol \mu} \, dxdt\) provides bounds for \(u_i\) in \(L^2(0,T; \, H^1(\Omega))\), for \(u_0\) in \(L^2(0,T; \, H^2(\Omega))\) and, moreover, a bound in \(L^2( (0,T) \times \Omega))^d\) for the auxiliary quantity \( u_0 \, (1-u_0) \, \nabla(-\epsilon \, \triangle u_0 + \beta \, (1-2u_0) ) =: J\). An interesting technical point is that these natural estimates do not allow making sense of third-order space derivatives of \(u_0\), while these occur in the diffusion fluxes. However, using the special structure (2) of the matrix \(M\), the third order contribution to diffusion for \(i=1,\ldots,n\) can be written as \[ K_{i0} \, u_0 \, u_i \, \nabla(-\epsilon \, \triangle u_0 + \beta \, (1-2u_0) ) = K_{i0} \, \frac{u_i}{1-u_0} \, J \, . \] Since \(J\) is controlled in \(L^2\), the fluxes are therefore controlled in \(L^2\) as well. Overall, a weak solution thus consists of \(u_0, \, u_1, \ldots,u_n: \, (0,T) \times \Omega \rightarrow [0, \, 1]\) and of an auxiliary vector field \(J \in (L^2((0,T) \times \Omega))^d\) such that \begin{gather*} u_1, \ldots, u_n \in L^2(0,T; \, H^1(\Omega)), \quad u_0 \in L^{\infty}(0,T; \, H^1(\Omega)) \cap L^2(0,T; \, H^2(\Omega)) \, , \\ \partial_t u_i \in L^2(0,T; \, (H^1(\Omega))^{\prime}) \ \text{ for } \ i = 0,\ldots,n \, , \end{gather*} subject to \(u_0 = 1- \sum_{i=1}^n u_i\), and for which the integral relations \[ \int_{0}^T \langle u_i^{\prime}, \, \phi_i\rangle \,dt = - \int_{0}^T \int_{\Omega} \left[\sum_{1\leq j \neq i \leq n} K_{ij} (u_j\nabla u_i-u_i\nabla u_j) + K_{i0} (u_0\nabla u_i-u_i\nabla u_0 -\kappa_i \, J)\right] \cdot \nabla \phi_i \, dxdt\, , \] \[ \int_{0}^T \int_{\Omega} J \cdot \eta \, dxdt = -\int_{0}^T \int_{ \Omega} (-\epsilon \Delta u_0 + \beta \, (1-2u_0)) \, \operatorname{div}v(u_0 \, (1-u_0) \, \eta) \, dxdt \, , \] hold for all \(\phi \in L^2(0,T; \, (H^1(\Omega))^d)\) and \(\eta \in L^2(0,T; \, (H^1(\Omega))^d) \cap (L^{\infty}( (0,T) \times \Omega))^d\). Here \(\kappa_i :=\chi_{\{u_0 < 1\}} \, u_i / (1-u_0)\). for \(u_0 \leq 1\) and \(\kappa_i = 0\) otherwise For arbitrary initial data \(\mathbf{u}^0\) in \((H^2(\Omega))^d\) subject to \(0 \leq u^0_i \leq 1\), \(\sum_{i=0}^n u^0_i = 1\), the paper shows the existence of solutions by means of the method of approximation via regularised time-discrete systems. In the course of the proof, a careful analysis must be provided for the auxiliary semilinear elliptic system \[ \epsilon \, \triangle u_0 + \ln u_i - \ln u_0 = f_i \text{ in } \Omega, \quad \nabla_{\nu} u_0 = 0 \text{ on } \partial \Omega \, . \] By given \(f_1,\ldots,f_n\), this problem is shown to possess a unique strictly positive solution, a result which is worth reading for itself.