Waveform relaxation of partial differential equations (Q1625758)

Coupled reaction-diffusion equations arise from several areas of science and engineering in the form \[\begin{cases} \mathbb{u}_t - \Delta \mathbb{u} = \mathbb{f}(\mathbb{u}) & \text{in } \Omega \times (0,T], \\ \mathbb{u}(x,t) = 0 & \text{in } \partial \Omega \times [0,T], \\ \mathbb{u}(x,0) = h(x) & \text{in } \overline{\Omega}. \\ \end{cases}\] Through discretization in space, the system of partial differential equations transforms into a system of ordinary differential equations leading to numerical methods for approximate solutions. However, the numerial methods have slow convergence for granular meshes. Following the advances in [\textit{B. Zubik-Kowal} and \textit{S. Vandewalle}, SIAM J. Sci. Comput. 21, No. 1, 207--226 (1999; Zbl 0945.65107)], the authors studied schemes without ordinary differential equations in [\textit{Y. Jiang} and \textit{Z. Miao}, J. Comput. Math. 36, No. 4, 542--562 (2018; Zbl 1424.65204)]. Here the reaction term splits into \(\mathbb{f}(\mathbb{u}) = F(\mathbb{u}, \mathbb{u})\) with evaluation at \(F(\mathbb{u}^{(k+1)}, \mathbb{u}^{(k)})\). The split allows the reaction term to depend linearly on the iterate \(\mathbb{u}^{(k+1)}\) and nonlinearly on the iterate \(\mathbb{u}^{(k)}\). For example, a possible splitting is \[\mathbb{F}(\mathbb{u}^{(k+1)}, \mathbb{u}^{(k)}) \approx \mathbb{F}(\mathbb{u}^{(k)}) + \operatorname{diag}\left(D F \right) \left(\mathbb{u}^{(k+1)} - \mathbb{u}^{(k)}\right).\] The scheme updates \(\mathbb{u}^{(k)}\) to \(\mathbb{u}^{(k+1)}\) from the solution to a semi-linear partial differential equation. Note that numerical methods implementing the scheme can separately update the entries of \(\mathbb{u}^{(k+1)}\) in parallel. The authors prove convergence of the scheme for suitable splitting of the reaction term through the energy estimate \[||\mathbb{u}^{(k)} - \mathbb{u} ||_{L^{\infty}\left(0,T; L^2(\Omega) \right)} \leq C ||\mathbb{u}^{(0)} - \mathbb{u} ||_{L^{\infty}\left(0,T; L^2(\Omega) \right)} \] where \(C:=C\left(k,t, ||DF||_{L^2(\Omega)} \right)\).