Convergence analysis of spatially adaptive Rothe methods (Q486668): Difference between revisions

Among the different strategies to solve numerically evolution equations of the parabolic type, the one analyzed in this paper corresponds to the so-called horizontal method of lines. In this approach, also known as Rothe's method, the equation is discretized first in time and then in space. More specifically, to cope appropriately with stiffness, an \(s\)-stage implicit time-solver is first applied, so that a system of linear elliptic equations per time step has then to be treated. This can be formulated as an abstract Cauchy problem defined by an ordinary differential equation in a suitable function space, which can subsequently be analyzed and eventually solved by some discretization scheme involving adaptive strategies. The purpose of the paper is to provide a convergence analysis of the resulting treatment. The first step in this analysis is to consider a uniform discretization in time with an \(s\)-stage (implicit) integrator and then formulate the resulting equations at each time step as an abstract problem involving two types of operators: the inverse of a (linear) elliptic differential operator and certain (nonlinear) evaluation operators. Then this problem is tackled by applying inexact variants of these operators (mimicking adaptivity) and studying the convergence of the resulting approximate solution. Several theorems provide conditions for adjusting the tolerances of the inexact schemes so that the approximations still converge to the exact solution and the resulting output inherits the approximation order of the exact scheme (which is typically the order of the time integrator used in the first instance). It turns out that large classes of time integrators can be fit into this procedure, and thus rigorous convergence results can be established for them. With respect to concrete space discretization methods, the authors are particularly interested in spatially adaptive schemes based on wavelets, due to their asymptotical optimality. They then determine an upper bound for the degrees of freedom for the overall scheme that are needed to construct an approximate solution in an adaptive way for a prescribed tolerance. The paper, although technically quite involved, is very readable and contains important results on the convergence of schemes in this setting.