Investigation of Newton-Krylov algorithms for low Mach number compressible flow
Fully coupled Newton-Krylov algorithms are used to solve steady speed compressible flow past a backward facing step flow Mach and Reynolds numbers. Various preconditioned Krylov iterative methods are used to solve the linear systems that arise on each Newton step, specifically Lanczos-based and Arnoldi-based algorithms. Several preconditioning strategies are considered to improve the performance of these iterative techniques, including incomplete lower-upper factorization with various levels of fill-in [ILU(k)] and domain based additive and multiplicative Schwarz type preconditioning both with and without overlapping domains. The ILU(K) preconditioners were generally less reliable for lower values of the flow Mach number, and exhibited strong sensitivity to cell ordering. In addition, the parallel nature of the domain based preconditioners is exploited on both a shared memory computer and a distributed system of workstations. Important aspects of the numerical solutions are discussed.
- Research Organization:
- Idaho National Lab. (INL), Idaho Falls, ID (United States)
- Sponsoring Organization:
- USDOE, Washington, DC (United States)
- DOE Contract Number:
- AC07-94ID13223
- OSTI ID:
- 114646
- Report Number(s):
- INEL-94/00081; CONF-950853-7; ON: DE96001622
- Resource Relation:
- Conference: ASME/JSME annual summer meeting of the Fluids Engineering conference, Hilton Head, SC (United States), 13-17 Aug 1995; Other Information: PBD: [1995]
- Country of Publication:
- United States
- Language:
- English
Similar Records
Parallel implementation of domain-based preconditioning of Newton-Krylov methods for a natural convection model problem
Scalability of preconditioners as a strategy for parallel computation of compressible fluid flow