Browse Prior Art Database

Multilevel Iteration Algorithm

IP.com Disclosure Number: IPCOM000069536D
Original Publication Date: 1978-May-01
Included in the Prior Art Database: 2005-Feb-20

Publishing Venue

IBM

Related People

Authors:
Hsieh, HY Rabbat, NB [+details]

Abstract

In a conventional analysis, the Newton method for convergence is used to solve an entire system simultaneously. If any one variable does not converge, all the variables are updated until all variables have converged. In our macromodular approach, we solve one macromodel at a time by considering only its immediate loading macromodels and iterating in a multilevel Newton loop. The idea of a multilevel loop can be considered a new application of the Newton-SOR (Successive Over Relaxation) iterative scheme. In regular Newton-SOR, they are interested in solving a system of nonlinear equations by solving them one equation at a time; we are interested in analyzing a large nonlinear network by analyzing them one macromodel at a time. However, each macromodel may consist of a system of nonlinear equations.