Optimization Computations in Transputer Systems

Transputer, Occam, optimization methods, criterion function, analogue diagram, gradient method

Mathematical analyses have proved that the use of the algorithms devised by us for numerical integration with a direct use of Taylor series brings an improvement in computation speed in comparison with the classic integration algorithms. Further substantial increase in computation speed can be achieved by using parallel architectures. As the problems connected with the subject are rather numerous, this paper aims at drawing attention to the developed methodology of parallel computations to solve systems of linear differential equations in a transputer network. The computation algorithm is written for a one-transputer system; however the computation can easily be generalized for a multiple-transputer system or for solving more complex systems of diferntial equations. For clearness' sake, we have used the 1st order Taylor series method (in this case the method is identical with the Euler integration method) - only the first terms of Taylor series are included in the computation. The well-known method of the analogue and hybrid computer techniques is used. It is based on procedures describing the activities of integrators and invertors. The interconnection of integrators and invertors is implemented via channels according to an analogue diagram.