Convergence rates for adaptive approximation of ordinary differential equations (Q1434048)

Consider the solution \(X:[0, T]\rightarrow \mathbb{R}^d\) of an ordinary differential equation \(\dot X(t)=a(t, X(t))\), \(0\leq t\leq T\), \(X(0)=X_o\), where \(a:[0, T]\times \mathbb{R}^d\rightarrow \mathbb{R}^d\). Denote by \(\overline {X}\) its approximation obtained by any numerical method, \(\overline X(0)=X(0)\). A general adaptive algorithm for \(\overline X\) is derived including a measure of the corresponding convergence rates and the analysis of asymptotic behaviour when the error tolerance parameters goes to zero. The algorithm uses the expansion of the global approximation error derived by \textit{K.-S. Moon, A. Szepessy, R. Tempone} and \textit{G. E. Zouraris} [Numer. Math., No. 1, 131--152 (2003; Zbl 1053.65059)]. The number of time steps is adapted by sucessfully dividing these steps to bound an approximation of the global error. Each refinement level decreases the maximal error indicator with at least a given factor until the process stops with the optimal number of steps up to a multiplicative constant factor. The true global error is then bounded asymptotically by the tolerance parameter as the tolerance tends to zero. The stopping, accuracy, and efficiency formulae are proved. The numerical results on the Lorenz problem and a problem with a singularity are included.