Adaptive Runge-Kutta methods for nonlinear two-point boundary value problems with mild boundary layers (Q1067370)

This paper extends the methods of the author [ibid. 9, 257-265 (1983; Zbl 0521.65069)] to include non-uniform meshes. The meshes are adaptively generated by asymptotically equidistributing an estimate of the global truncation error, obtained from comparing pairs of implicit Runge-Kutta formulas. An implementation of these ideas for first-order systems has been tested on the standard set of stiff linear problems due to \textit{C. E. Pearson} [J. Math. Phys. 47, 134-154 and 351-358 (1968; Zbl 0167.158 and Zbl 0165.505)] (limited to moderate size \(\epsilon)\). Comparisons with \textit{M. Lentini} and the reviewer's NAG version (DO2GAF) of the PASVA3 code [Lect. Notes Comput. Sci. 76, 67-88 (1979; Zbl 0434.65067)], shows that the methods are competitive with it (including CPU times in the comparison would have been helpful). The author recommends these techniques especially for large systems and low required accuracy, because of their more sparing use of storage, as compared to PASVA3. Similar ideas are applied to a second order linear equation.