A simple step size selection algorithm for ODE codes (Q1899962)

An economical method for controlling the stepsize in a variable step numerical algorithm for approximating the solution of an ordinary differential equation (ODE) is to choose the step so that the change in the approximate solution on each interval of the calculation remains below a preassigned tolerance \(\tau\). The authors compare this method of control to methods based on choosing the stepsize to minimize a known local error estimate. For an algorithm of \(p\)th order control of stepsize by the change in solution this leads to a global error proportional to \(\tau^p\). This compares with controls which minimize the local error with tolerance \(\tau\) whose global error may be proportional to \(\tau\), \(\tau^{p/p+ 1}\), \(\tau^{p+ 1/p}\). The method of control by change in the solution serves to stabilize a computation. The authors have written an experimental program which utilizes control of stepsize by the change in the solution. The program was tested on four difficult test problems and the results compared with ones obtained with the code RFK45. The method is recommended for problems where control by the local error is not applicable or too expensive.