Acceleration of Multidisciplinary Analysis Solvers
by Inexact Subsystem Simulations
Matthias Heinkenschloss
Mary Beth Hribar
Department of Computational and Applied Mathematics
Michael Kokkolaras
Department of Mechanical Engineering and Materials Science
Rice University
AIAA Paper 98-4712
7th AIAA/USAF/NASA/ISSMO Symposium on Multidisciplinary
Analysis and Optimization
September 2-4, 1998, St. Louis, MO
Abstract
This paper is concerned with the numerical solution of
systems of blocked nonlinear equations arising in the
solution of Multidisciplinary Analysis (MDA) problems.
We consider the case where individual discipline
solvers/simulators are given and are iterative methods.
Thus, an MDA solver consists of an outer iteration
for the solution of the system of blocked nonlinear
equations and of inner iterations in the discipline simulators.
We will show how the control of the truncation of the inner
iterations can be effectively used to accelerate the
overall iteration. The key is the interpretation of
the outer iteration with inexact inner iteration as an
iteration of a related system with same solution
as the MDA problem.
PDF file (175 kB).