Parallel domain decomposition and the solution of nonlinear systems of equations

Conference · Sun Dec 31 23:00:00 EST 1989

OSTI ID:6467142

Gropp, W D ^[1]; Keyes, D E ^[2]

---

1. Argonne National Lab., IL (USA)
2. Yale Univ., New Haven, CT (USA). Dept. of Mechanical Engineering

Many linear systems arise as subproblems in the solution of nonlinear equations, either as part of a simple fixed-point of a Newton's method iteration. This paper considers the use of domain decomposition techniques for the solution of these linear problems in the context of solving a multicomponent system of nonlinear equations on two types of parallel processors. One of the computations is drawn from fluid dynamics and includes locally refined grids. Such problems require great computational resources, and domain, decomposition seems to offer a way to efficiently solve these problems on computers with significant parallelism. The domain decomposition approach used is as in Gropp and Keyes, modified to achieve better parallelism and to reduce the computational work. 13 refs., 2 figs., 5 tabs.

• Research Organization: Argonne National Lab., IL (USA)
• Sponsoring Organization: DOE/ER; NSF
• DOE Contract Number: W-31109-ENG-38
• OSTI ID: 6467142
• Report Number(s): CONF-9005292-1; ON: DE91004442; CNN: ESC-8957475
• Country of Publication: United States
• Language: English

Related Subjects

640410 -- Fluid Physics-- General Fluid Dynamics
75 CONDENSED MATTER PHYSICS
SUPERCONDUCTIVITY AND SUPERFLUIDITY
99 GENERAL AND MISCELLANEOUS
990200* -- Mathematics & Computers
DIFFERENTIAL EQUATIONS
EQUATIONS
FLUID MECHANICS
ITERATIVE METHODS
MECHANICS
NONLINEAR PROBLEMS
PARALLEL PROCESSING
PARTIAL DIFFERENTIAL EQUATIONS
PROGRAMMING