Homogenization of degenerate cross-diffusion systems (Q2284636)

The authors describe the asymptotic behavior of the solution of two parabolic systems. The first one is written as: \(\partial_{t}u_{i}^{\varepsilon}-\operatorname{div}(\sum_{j=1}^{n}P(\frac{x}{\varepsilon})a_{ij}(u^{\varepsilon})\nabla u_{j}^{\varepsilon})=f_{i}(u^{\varepsilon})\), \(i=1,\ldots ,n\). It is posed in \(\Omega \times (0,\infty )\), where \(\Omega \) is a bounded domain of \(\mathbb{R}^{d}\) (\(d\geq 1\)). The no-flux boundary condition \(\sum_{j=1}^{n}P(\frac{x}{\varepsilon})a_{ij}(u^{\varepsilon})\nabla u_{j}^{\varepsilon}\cdot \nu =0\) is imposed on \(\partial \Omega \times (0,\infty )\) and the solution starts from an initial condition \(u_{i}^{\varepsilon}(0)=u_{i}^{0}\), which is measurable from \(\Omega \) to \(\mathbb{R}^{n}\) and which takes its values in an open subset \(\mathcal{G}\) of \((0,1)^{n}\). In this problem, \(P\in L^{\infty}(Y;\mathbb{R}^{d\times d})\) is a diagonal matrix with periodic coefficients from the cell \(Y=(0,b_{1})\times \ldots (0,b_{d})\) to \(\mathbb{R}\), such that \(P_{i}(y)\geq d_{0}>0\) in \(Y\). The matrix \(A(u)=(a_{ij}(u))\in C^{0}(\mathcal{G};\mathbb{R}^{n\times n})\) satisfies \(\left| a_{ij}(u)\right| \leq C_{A}u_{j}^{s_{j}}\), \(z^{T}h^{\prime \prime}(u)A(u)z\geq \alpha \sum_{i=1}^{n}\left| u_{i}\right|^{2s_{i}}z_{i}^{2}\) and \((h^{\prime \prime}(u)A(u))_{ij}\leq Cu_{i}^{s_{i}}u_{j}^{s_{j}}\), for a convex function \(h\in C^{2}(\mathcal{G};\mathbb{R})\) such that \(h^{\prime}:\mathcal{G}\to\mathbb{R}^{n}\) is invertible. The source term \(f\in C^{0}(\overline{\mathcal{G}};\mathbb{R}^{n})\) satisfies \(f(u)\cdot h^{\prime}(u)\leq C_{f}(1+h(u))\). The authors do not impose the symmetry and the positive semi-definite property of the matrix \(A\). They first define the notion of weak solution to this problem as a function \(u^{\varepsilon}\in L^{\infty}(0,T;L^{\infty}(\Omega; \mathbb{R}^{n}))\) such that \((u_{i}^{\varepsilon})^{s_{i}+1}\in L^{2}(0,T;H^{1}(\Omega ))\) and \(\partial_{t}u_{i}^{\varepsilon}\in L^{2}(0,T;H^{1}(\Omega )^{\prime})\) and which satisfies a variational formulation of the above equation for every \(\varphi \in L^{2}(0,T;L^{\infty}(\Omega ;\mathbb{R}^{n}))\) and the initial condition in the \(L^{2}\) sense. The main result concerning this problem proves that up to some subsequence \(u^{\varepsilon}\) converges strongly in \(L^{2}(\Omega \times (0,T);\mathbb{R}^{n})\) to the solution \(u\in L^{\infty}(0,T;L^{\infty}(\Omega ;\mathbb{R}^{n}))\) such that \(u_{i}^{s_{i}+1}\in L^{2}(0,T;H^{1}(\Omega ))\) and \(\partial_{t}u_{i}\in L^{2}(0,T;H^{1}(\Omega )^{\prime})\) of the problem \(\partial_{t}u_{i}-\sum_{k,m=1}^{d}\sum_{l=1}^{n}\frac{\partial}{\partial x_{m}}(B_{mk}^{il}\frac{\partial u_{l}}{\partial x_{k}})=f_{i}(u)\), \(i=1,\ldots ,n\), with the boundary condition \(\sum_{k,m=1}^{d}\sum_{l=1}^{n}\nu_{m}B_{mk}^{il}\frac{\partial u_{l}}{\partial x_{k}}=0\). Here the homogenized matrix \(B_{mk}^{il}(u)\) is given through an explicit formula which involves the solution \(W_{j}^{kl}\) of a local problem. They specialize this result assuming that \(a_{ij}(u)=D_{i}(\delta_{ij}u_{n+1}+u_{i})\) where \(u_{n+1}=1-\sum_{i=1}^{n}u_{i}\) and \(D_{i}>0\). They here slightly modify the definition of weak solution. For the second problem, the authors consider a periodically perforated domain \(\Omega^{\varepsilon}=\Omega \setminus \bigcup_{\xi \in \Xi^{\varepsilon}}\varepsilon (Y_{0}+\xi )\). They write the parabolic problem \(\partial_{t}u_{i}^{\varepsilon}-\operatorname{div}(\sum_{j=1}^{n}a_{ij}(u^{\varepsilon})\nabla u_{j}^{\varepsilon})=f_{i}(u^{\varepsilon})\), \(i=1,\ldots ,n\), posed in \(\Omega ^{\varepsilon}\times (0,\infty )\). The boundary condition \(\sum_{j=1}^{n}a_{ij}(u^{\varepsilon})\nabla u_{j}^{\varepsilon}\cdot \nu =0\) is imposed on \(\partial \Omega^{\varepsilon}\times (0,\infty )\) and the solution starts from an initial condition \(u_{i}^{\varepsilon}(0)=u_{i}^{0}\). The authors adapt the notion of weak solution, still considering the special case of the the coefficients \(a_{ij}\) and the main result proves that \(u^{\varepsilon}\) converges strongly to the solution \(u\in L^{\infty}(0,T;L^{\infty}(\Omega ;\mathbb{R}^{n}))\) such that \(u_{n+1}^{1/2},u_{n+1}^{1/2}u_{i}\in L^{2}(0,T;H^{1}(\Omega ))\) and \(\partial_{t}u_{i}\in L^{2}(0,T;H^{1}(\Omega)^{\prime})\) of the problem \(\partial_{t}u-\operatorname{div}(D_{\mathrm{hom}}A(u)\nabla u)=f(u)\), with the boundary condition \(D_{\mathrm{hom}}A(u)\nabla u\cdot \nu =0\). The matrix \(D_{\mathrm{hom}}\) is given an explicit formula in terms of the solution \(w^{l}\) of a local problem. For the proof of the main results, the authors first establish uniform estimates on the solutions \(u^{\varepsilon}\). In the first case, they analyze the cell problem and they use the two-scale convergence to derive the limit problem. In the second case, the authors first establish an entropy inequality and they also use the two-scale convergence. In the final appendix, the authors give two examples of cross-diffusion systems with coefficients \(a_{ij}\) satisfying the special case.