TY - JOUR

T1 - Spectral element methods for large scale parallel Navier-Stokes calculations

AU - Fischer, Paul F.

AU - Rønquist, Einar M.

N1 - Funding Information:
The work of P. Fischer was supported by NSF under Grant ASC-9107674, in part by ONR contract N00014-90-J-4124, and by NSF Cooperative agreement CCR-8809615 during a post-doctoral fellowship at Caltech. The work of E. R~nquist was supported by Nektonics, Inc. and by NASA's material processing and micro-gravity programs under contracts NASI-19102 and NAS3-26132. The authors would like to thank Professor Anthony T. Patera for providing access to the 32 node iPSC/860 at MIT which is supported by DARPA contract N00014-91-J-1889, and to the NSF Center for Research on Parallel Computation and the Concurrent Supercomputing Consortium for providing time on the Intel Touchstone Delta System at Caltech.

PY - 1994

Y1 - 1994

N2 - We analyze the computational complexity of a recently developed two-level iteration scheme for spectral element solution of the time-dependent incompressible Navier-Stokes equations in complex domains. We present several algorithmic advances which significantly enhance the scalability of this approach, including: implementation of an advanced combine operation for degrees-of-freedom on subdomain edges, parallel solution of the (fine-grained) coarse-grid problem, and implementation of local low-order finite element preconditioners for the find-grid problem. Timings on the 512 node Intel Delta machine show that the combined improvements lead to a fourfold reduction in Navier-Stokes solution time for the particular case of a three-dimensional boundary layer calculation consisting of one million degrees-of-freedom.

AB - We analyze the computational complexity of a recently developed two-level iteration scheme for spectral element solution of the time-dependent incompressible Navier-Stokes equations in complex domains. We present several algorithmic advances which significantly enhance the scalability of this approach, including: implementation of an advanced combine operation for degrees-of-freedom on subdomain edges, parallel solution of the (fine-grained) coarse-grid problem, and implementation of local low-order finite element preconditioners for the find-grid problem. Timings on the 512 node Intel Delta machine show that the combined improvements lead to a fourfold reduction in Navier-Stokes solution time for the particular case of a three-dimensional boundary layer calculation consisting of one million degrees-of-freedom.

UR - http://www.scopus.com/inward/record.url?scp=0028450352&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0028450352&partnerID=8YFLogxK

U2 - 10.1016/S0045-7825(94)80009-X

DO - 10.1016/S0045-7825(94)80009-X

M3 - Article

AN - SCOPUS:0028450352

VL - 116

SP - 69

EP - 76

JO - Computer Methods in Applied Mechanics and Engineering

JF - Computer Methods in Applied Mechanics and Engineering

SN - 0374-2830

IS - 1-4

ER -