STEADY-STATE PROCESS SIMULATION ON MIMD MACHINES - SOLVING NONLINEAR EQUATIONS

The feasibility of using MIMD machines for steady state process simulation is shown, by a parallel implementation of an inexact Newton/Quasi-Newton algorithm. A restarted version of GMRES is used to obtain the Newton step with a suitable parallel preconditioner. The algorithm is tested on two problems, a simple model for a distillation column using Antoine coefficients for the physical properties and the discretization of a PDE. Two MIMD machines are used for the tests. A Meiko CS1 and a cluster of workstations using the TCGMSG message passing package. Problems of up to 16384 variables are tested on the CS1 and up to 129600 variables on the cluster. The numerical results show very good speedups obtained, even for the smaller problems.