A priori estimates for fractional nonlinear degenerate diffusion equations on bounded domains (Q495900)

In this paper, the authors study the properties of non-negative solutions (in a suitable sense) to the nonlinear equation \[ \partial_t u + \mathcal{L}(u^m)=0 \text{ in } \Omega \times \mathbb{R}^+ \] for \(\Omega \subset \mathbb{R}^N\) a bounded, regular domain. Here, \(m>1\) and \(\mathcal{L}\) is the spectral fractional Laplacian, that is, \[ \mathcal{L} u = \sum_{k=1}^\infty \lambda_k^s \hat{u}_k\phi_k(x) \] (with \(0<s<1\)), where \[ \hat{u}_k = \int_\Omega u(x) \phi_k(x)\;dx \] and \(\{(\lambda_k,\phi_k)\}_{k=1}^\infty\) are the eigenvalues and the \(L^2(\Omega)\)-normalized eigenvectors of the Dirichlet Laplacian on \(\Omega\). For such \(u\), they show various pointwise and integral upper and lower bounds, for example, \[ u(x,t) \leq K_2 \frac{\Phi_1(x)^{1/m}}{t^{1/(m-1)}}\quad \text{for all } x\in \Omega, \quad t>0, \] and \[ u(x,t) \geq L_1 \frac{\Phi_1(x)}{t^{1/(m-1)}}\quad \text{for all } x\in \Omega, \quad t\geq t^*, \] where \(t^*\) depends on the initial data \(u_0(x)\) and \(\Phi_1\) is the first eigenvector of \(\mathcal{L}\). These inequalities, combined with Harnack inequalities known for \(\Phi_1\), imply the Harnack inequality for \(u\) (for all sufficiently large times), while the authors go on to discuss Harnack inequalities of a backward type. The main technique is the use of good estimates for the Green's functions of \(\mathcal{L}\), which the authors recall in detail in the appendix.