Parallel ODE-solvers with stepsize control (Q919760): Difference between revisions

The present paper continues a recent investigation by the authors [J. Comput. Appl. Math. 25, 341-350 (1989; Zbl 0675.65134)] on parallel implementation of one-step methods \(y_{n+1}=F_{n+1}(y_ n)\), \(n=0,...,N-1\), for ordinary differential equations (ODEs), where the trajectory \((y_ 0,y_ 1,...,y_ N)\) is iteratively computed as the fixed point of the transformation \(\phi (u_ 0,u_ 1,...,u_ N)=(u_ 0,F_ 1(u_ 0),...,F_ N(u_{N-1}))\). Availing of p processors the block \((u^ k_ s,...,u^ k_{s+p})\) initialized as \((u^ 0_ 0:=y_ 0,...,u^ 0_ p)\) is handled concurrently each iteration and shifts forward at least one index each iteration. A moving mesh method based on some mechanism of self-adapting stepsize is developed. The strategy consists in performing the stepsize control by changing the mesh after each iteration. The practical implementation is explained in detail, the attainable speedup is studied and numerical examples illustrate the method.