Optimal adaptive solution of initial-value problems with unknown singularities (Q945916)

The authors consider initial value problems of a scalar nonlinear ordinary differential equation \(y'(t) = f(y(t))\). The function \(f\) is Lipschitz continuous and strictly positive, i.e., the solution is monotonically increasing. The first \(r\) derivatives of \(f\) are continuous except for a possible discontinuity at one unknown singular point. Thereby, the existence of finite limits is assumed. Although strong assumptions are imposed on the problems, a sophisticated analysis is required to achieve error bounds for corresponding numerical techniques. The authors investigate numerical methods based on Taylor expansion in a standard form as well as involving an integral. In \textit{L. Plaskota} and \textit{G.W. Wasilkowski} [Numer. Math. 102, No.~1, 123--144 (2005; Zbl 1083.65031)], adaptive quadrature is performed by detecting a singularity of the integrand via divided differences. The authors transfer the idea of this strategy by using a criterion based on the derivatives of \(f\). An adaptive grid is constructed, where an a priori equidistant grid is refined at the singularity only. While a singularity of \(f\) is given, the corresponding singularity of the solution is detected correctly in time domain. The authors prove that the errors for the considered class of numerical methods using the adaptive grid exhibit the same order as in the case of regular functions, i.e., the error depends on \(r\). Furthermore, lower bounds of the error demonstrate that each method of the class produces an error independent of \(r\) in case of more than one singularity. Although numerical simulations are not presented in the article, the authors mention that numerical experiments exist, which confirm the bound of the crucial theorem.