On nonlinear cross-diffusion systems: an optimal transport approach (Q1650886): Difference between revisions

The authors consider a domain \(\Omega \subset \mathbb{R}^{d}\), with \(C^{1}\) boundary, \(T>0\), \(m>1\) and the coupled transport problem \(\partial _{t}\rho ^{1}-\nabla \cdot ((\nabla p+\nabla \Phi _{1})\rho ^{1}=0\), \(\partial _{t}\rho ^{2}-\nabla \cdot ((\nabla p+\nabla \Phi _{2})\rho ^{2}=0\), where the pressure \(p\) is given as \(p=\frac{m}{m-1}(\rho ^{1}+\rho ^{2})^{m-1}\). The solution starts from an initial data \(\left( \rho _{0}^{1},\rho _{0}^{2}\right) \in (L^{1}(\Omega ))^{2}\). Here \(\Phi _{1}\) and \(\Phi _{2}\) are given functions which satisfy \(-\partial _{x}\Phi _{1}\geq -\partial _{x}\Phi _{2}\) and \(x_{1}\geq x_{2}\) for every \(x_{i}\in \{\rho _{0}^{i}>0\}\), \(i=1,2\). The authors observe that this problem may be seen as the gradient flow in the Wasserstein space of the free energy \(\left( \rho ^{1},\rho ^{2}\right) \mapsto \int_{\Omega }\frac{1}{m-1}\left( \rho ^{1}+\rho ^{2}\right) ^{m}dx+\int_{\Omega }\Phi _{1}\rho ^{1}dx+\int_{\Omega }\Phi _{2}\rho ^{2}dx\). They also consider the problem obtained as \(m\rightarrow +\infty \) \(\partial _{t}\rho ^{1,\infty }-\nabla \cdot ((\nabla p^{\infty }+\nabla \Phi _{1})\rho ^{1,\infty }=0\), \(\partial _{t}\rho ^{2,\infty }-\nabla \cdot ((\nabla p^{\infty }+\nabla \Phi _{2})\rho ^{2,\infty }=0\), where the pressure \(p^{\infty }\) is supported in the region \(\{\rho ^{1,\infty }+\rho ^{2,\infty }=1\}\). The first main result of the paper proves a stay ordering property for the scheme defined through \(\rho _{k+1}^{\tau }=\left( \rho _{k+1}^{1,\tau },\rho _{k+1}^{2,\tau }\right) =\arg \min_{\rho \in \mathcal{P}^{M_{1}}(\Omega )\times \mathcal{P} ^{M_{2}}(\Omega )}\{\mathcal{F}_{m}(\rho )+\frac{1}{2\tau }\mathbf{W} _{2}^{2}(\rho ,\rho _{k}^{\tau })\}\), where \(\mathcal{F}_{m}(\rho )=\int_{\Omega }\frac{1}{m-1}\left( \rho ^{1}+\rho ^{2}\right) ^{m}dx\) if \( \left( \rho ^{1}+\rho ^{2}\right) ^{m}\in L^{1}(\Omega )\) and \(+\infty \) otherwise, \(\mathbf{W}_{2}^{2}\) is the Wasserstein distance on the product \( \mathcal{P}^{M_{1}}(\Omega )\times \mathcal{P}^{M_{2}}(\Omega )\) of spaces of finite nonnegative Radon measures on \(\Omega \) with masses \(M_{1}\) and \( M_{2}\) respectively: \(\mathbf{W}_{2}^{2}(\mu ,\nu )=W_{2}^{2}(\mu ^{1},\nu ^{1})+W_{2}^{2}(\mu ^{2},\nu ^{2})\) and \(\tau =T/N\), \(k=0,\ldots ,N\). The first main result indeed proves that \(\rho _{t}^{1,\tau }\) is supported on the right of the support of \(\rho _{t}^{2,\tau }\) for all \(t>0\). Moreover \( \rho ^{i,\tau }\) converges weakly in \(L^{2m-1}([0,T]\times \Omega )\) and pointwise a.e. to \(\rho ^{i,m}\) as \(t\downarrow 0\). The pair \((\rho ^{1,m},\rho ^{2,m})\) solves the first above-indicated problem in a weak sense and stays ordered for each times. The authors first prove regularity properties of the minimizers \(\left( \rho _{k+1}^{1,\tau },\rho _{k+1}^{2,\tau }\right) \). They then introduce the notion of weak solution for the above-indicated problems and the notions of interpolations between the densities for which they prove a priori estimates, which allow to take the limit when \(t\downarrow 0\). They then study segregation properties of the densities, with a focus in the 1D case. The second main result of the paper proves that \((\rho ^{1,m},\rho ^{2,m})\) converges weakly in \( L^{p}([0,T]\times \Omega )\) for any \(1<p<\infty \), up to a subsequence, to a weak solution \((\rho ^{1,\infty },\rho ^{2,\infty })\) of the second above-indicated problem. The third main theorem proves that \(\rho ^{i,\infty }\) remains a patch solution to the second above-indicated problem, for all \( t>0\), that is \(\rho ^{i,\infty }=\chi _{(a^{i}(t),b^{i}(t))}\), if \(\partial _{xx}^{2}\Phi _{i}\geq 0\) and \(\rho _{0}^{i}=\chi _{(a^{i}(0),b^{i}(0))}\), with a density pair which is a solution to a two-phase Hele-Shaw type flow with drifts.