Hypercube implementations of parallel shooting (Q1119354)

This paper is concerned with parallel shooting methods for the numerical solution of linear two-point boundary value problems with separated boundary conditions upon a hypercube. By a hypercube the authors mean a distributed memory computer with communication between processors by message passing via a communication network having the topology of an n- dimensional cube, where the vertices are considered as processors and the edges as communication links. In this context, the aim of the authors is to present two strategies to associate the tasks onto the processors of a hypercube. In the first technique referred to as the domain decomposition each processor has the responsibility for the computations associated to one subinterval between two consecutive shooting points. In the second technique, called the column decomposition, each processor is assigned the computations corresponding to certain columns of the fundamental matrices associated with each shooting interval. Finally under some assumptions the efficiency of the two strategies is compared in terms of several parameters. However, due to the large number of assumptions and the fact that some of them can be easily violated in practice, the conclusions are not clearly applicable.