A three level finite element approximation of a pattern formation model in developmental biology (Q2510399)

The paper under review provides a thorough numerical analysis of a multilevel semidiscretization of a weak formulation of a reaction-diffusion equation. This approach, discretizing first in time and conceptually replacing the reaction-diffusion equation with a sequence of spatial PDE (in this case elliptic problems) is known as Rothe's method or more commonly as the transverse method of lines. Garvie and Trenchea, while not using this terminology, give a detailed historical introduction concentrating on pattern formation (morphogenesis). This paper gives rigorous stability results and error estimates for a fixed time-step three-level backward difference in time scheme. The method is called 2-SBDF or ``Extrapolated Gear'', with 1-SBDF as starting method. This paper also analyzes a fully discretized version where the spatial discretization is a standard Galerkin finite element method. Finally, the paper presents numerical experiments in an exemplary manner, with enough details -- including details of random number generation and seeds -- to allow reproduction of their experiments. I have not attempted such a reproduction but I believe the claim of Garvie and Trenchea that all necessary details have been reported.