Analysis of degenerate cross-diffusion population models with volume filling (Q501533): Difference between revisions

In this interesting and well-written paper the authors analyze a class of multi-species population cross-diffusion systems with volume-filling effects, which appear for instance in spatial segregration of interacting species, chemotactic cell migration in tissues and ion transport through membranes. The system (of strongly coupled equations) for the unknown \(\mathbf{u}=(u_1,\ldots,u_n)\) reads as follows: \[ \begin{cases} D_tu_i=\sum_{j=1}^n\text{div}(A_{ij}(\mathbf{u})\nabla u_j),&\text{in } (0,+\infty)\times\Omega, i=1,\ldots,n,\\ \sum_{j=1}^nA_{ij}(\mathbf{u})\nabla u_j\cdot\nu=0,&\text{in } (0,+\infty)\times\partial\Omega, i=1,\ldots,n, \end{cases}\tag{S} \] \(\Omega\) being a bounded domain of \(\mathbb R^d\) (\(d\geq 1\)) and \(\nu\) denoting the outward normal vector to \(\partial\Omega\). Here, the entries of the diffusion matrix \(A(\mathbf{u})\) (which in general is not positive definite) are given by \[ A_{ij}(\mathbf{u})=\delta_{ij}p_i(\mathbf{u})q(u_{n+1})+u_ip_i(\mathbf{u})q'(u_{n+1}) +u_iq(u_{n+1})\frac{\partial p_i}{\partial u_j}(\mathbf{u}), \] for \(i,j=1,\ldots,n\), where \(u_{n+1}=1-\sum_{i=1}^nu_i\); \(q\in C^3([0,1])\) is positive in \((0,1)\), \(q'\geq\gamma q\) in \((0,1)\), for some \(\gamma>0\), \(q(0)=0\); \(p_i(x)=\exp\left (\frac{\partial\chi(x)}{\partial x_i}\right )\) for \(i=1,\dots,n,\) \(x\in D:=\{x\in\mathbb R^n: x_i> 0\) for \(i=1,\ldots,n, \sum_{j=1}^nx_j<1\}\) and some nonnegative convex function \(\chi\in C^3(\overline D)\). The choice of the matrix \(A\) implies that system (S) has a gradient-flow structure, with entropy density \(h\) defined by \(h(x)=\sum_{i=1}^n(x_i\log(x_i)-x_i+1)+\int_a^{1-\sum_{i=1}^nx_i}\log(q(s))ds+\chi(x)\) for any \(x\in D\), where \(a=1\) if \(q(1)\leq 1\) and \(a=q^{-1}\) otherwise. The paper contains three main results: (a) global existence of solutions to system (S), (b) their long-time behaviour and (c) uniqueness. As far as (a) is concerned, the authors prove that, for every \(T>0\) and every measurable function \(\mathbf{u}^0:\Omega\to D\) such that \(h(\mathbf{u}^0)\in L^1(\Omega)\), problem (S) admits at least one weak solution \(\mathbf{u}:(0,T)\times\Omega\to\overline D\) such that (i) \(\mathbf{u}(0,\cdot)=\mathbf{u}^0\) in the sense of \(H^1(\Omega)'\); (ii) \(u_i\in L^{\infty}((0,T);L^{\infty}(\Omega))\), \(D_tu_i\in L^2((0,T);H^1(\Omega)')\) and \((1+\sqrt{u_i}+u_ip_i(\mathbf{u}))\sqrt{q(u_{n+1})}\in L^2((0,T);H^1(\Omega))\) for each \(i=1,\ldots,n\). Moreover, \(u\) satisfies the entropy inequality \[ \begin{aligned}\int_{\Omega}h(\mathbf{u}(t,\cdot))dx +c_0\int_0^tdt\int_{\Omega}\bigg (q(u_{n+1})^2\sum_{i=1}^n|\nabla\sqrt{u_i}|^2+|\nabla \sqrt{q(u_{n+1})}|^2\bigg )dx \\ \leq \int_{\Omega}h(\mathbf{u}^0)dx\end{aligned} \tag{E} \] for some positive constant \(c_0\) (explicitly computable). Finally, if \(\log(q)\notin L^1((0,b))\) for all \(b\in (0,1)\), then \(u_{n+1}>0\) a.e. in \((0,T)\times\Omega\). The proof of this result relies on some generalized Aubin-Lions and Aubin-Lions-Dubinskiĭ lemmata, here proved by the authors, which allow them to apply a compactness argument to a sequence of solutions to a time discretization and regularization of the differential system in (S). This new system is solved by applying the Leray-Schauder fixed-point theorem, and the positive semi-definiteness of the matrix \(A(x)(\text{Hess}\,h)(x)^{-1}\) plays a crucial role (here, \(\text{Hess}\,h)\) denotes the Hessian matrix of \(h\)). The solutions of this new system satisfy some a priori integral estimates, which are uniform in the discretization and regularization parameters \(\tau\) and \(\varepsilon\). These estimates and the above mentioned lemmata allow the authors to let, first \(\varepsilon\) and then \(\tau\) tend to \(0\) (along a sequence) to prove the existence of a solution to (S). It is worth mentioning that the above machinery works also if \(q(0)>0\) and, actually, things are indeed easier in that case. Uniqueness of the weak solution to (S), satisfying the initial condition \(\mathbf{u}(0,\cdot)=\mathbf{u}^0\), is proved assuming that \(p_i\equiv 1\) for each \(i=1,\ldots,n\). This choice of the \(p_i\)'s allows the author to write a single uncoupled equation for the function \(u_{n+1}\). Applying the \(H^{-1}\) method to this equation, the uniqueness of \(u_{n+1}\) is obtained. Based on this result, the uniqueness of \(\mathbf{u}\) is proved employing the E-monotonicity argument: if \(\mathbf{u}\) and \(\mathbf{v}\) are two different weak solutions to (S) with the properties (i) and (ii), then the (distance) function \[ t\mapsto d_{\varepsilon}(\mathbf{u}(t,\cdot),\mathbf{v}(t,\cdot))=\sum_{i=1}^n\int_{\Omega}\bigg [\xi_{\varepsilon}(u_i(t,\cdot))+\xi_{\varepsilon}(v_i(t,\cdot))-2\xi_{\varepsilon}\bigg (\frac{u_i(t,\cdot)+v_i(t,\cdot)}{2}\bigg )\bigg ]dx \] where \(\xi_{\varepsilon}(s)=(s+\varepsilon)(\log(s+\varepsilon)-1)+1\) for \(s>0\), vanishes as \(\varepsilon\to 0\), which implies that \(\xi_{\varepsilon}(u_i)+\xi_{\varepsilon}(v_i)-2\xi_{\varepsilon}\left (\frac{u_i+v_i}{2}\right )\) vanishes a.e. in \((0,T)\times\Omega\) as \(\varepsilon\to 0\). Taylor's formula shows that \(\xi_{\varepsilon}(u_i)+\xi_{\varepsilon}(v_i)-2\xi_{\varepsilon}\left (\frac{u_i+v_i}{2}\right )\geq \frac{1}{8}(u_i-v_i)^2\) and allows to conclude that \(u_i=v_i\) for each \(i\). Concerning (c), the authors prove the convergence as \(t\to +\infty\) of weak solutions to (S) to the constant steady-state solution to the same problem, when \(\Omega\) is convex. More precisely, let \(\mathbf{u}^0\in L^1(\Omega,D)\) and denote by \(\mathbf{u}^{\infty}\) the constant function whose components are the averages over \(\Omega\) of the corresponding components of \(\mathbf{u}^0\). Further, assume that \(q'\) is strictly positive, the function \(q/q'\) is concave on \((0,1)\) and let \(\mathbf{u}\) be a weak solutions to problem (S), with satisfies the conditions (i) and (ii). Then, \(u_{n+1}(t,\cdot)\) converges to \(1-\sum_{i=1}^nu_i^{\infty}\) in \(L^2(\Omega)\) with exponential rate as \(t\to +\infty\), and an estimate of that rate is provided. If, in addition, \(q\) nowhere vanishes on \([0,1]\), then, for each \(i=1,\ldots,n\), \(u_i(t,\cdot)\) converges to \(u^{\infty}_i\) in \(L^2(\Omega)\) as \(t\to +\infty\) again with an exponential rate (and an estimate of that rate is also provided). The proof is obtained using the entropy estimate (E) the convex Sobolev and the logarithmic Sobolev inequalities. Here, the concavity of the function \(q/q'\) is essential to make the above machinery work. Finally, some extensions of the obtained results to more general systems are discussed. See also [\textit{N. Zamponi} and \textit{A. Jüngel}, Ann. Inst. Henri Poincaré, Anal. Non Linéaire 34, No. 3, 789--792 (2017; Zbl 1417.35062)] which contains a corrected proof of Lemma 7.