Newton-type decomposition methods in large-scale dynamic process simulation

For the plantwide dynamic simulations in the chemical process industry, a parallel approach using a divide and conquer strategy is developed, and its realization on parallel computers with shared memory is discussed. The system of differential-algebraic equations (DAEs) is partitioned into blocks, and then it is extended appropriately such that block-structured Newton-type methods can be applied, which enable the application of relaxation techniques. Two types of block-structured methods are considered and compared with one another. Applied to industrial applications, the approach has gained considerable speedup factors for the dynamic process simulation of various large-scale distillation plants, covering systems with up to 60000 equations. (C) 2001 Elsevier Science Ltd. All rights reserved.