TY - JOUR
T1 - Parallel computation of forced convection using domain decomposition
AU - Wang, Mingyu
AU - Georgiadis, John G.
N1 - Funding Information:
nology at Duke University for financial support. John G. Georgiadis acknowledges tha financial support of the National Science Foundation under grant CTS-8909119. Both authors are also grateful to Dr. Henry S. Greenside for access to the computing resources of the Department of Computer Science at Duke University.
Copyright:
Copyright 2016 Elsevier B.V., All rights reserved.
PY - 1991/9
Y1 - 1991/9
N2 - Boundary-fitted coordinate transformation broadens the applicability of finite difference methods. However, for a large class of geometries, coordinate transformation introduces singularities and increases grid skewness, which results in large numerical error and slow convergence rate. In this paper, we present the results of combining a finite difference scheme with domain decomposition to obtain a parallel scheme. This scheme is used to simulate steady-state forced convection in irregular axisymmetric and two-dimensional domains. The irregular domain is first dissected into subdomains that have smooth curves as their boundaries. Curvilinear coordinate systems are then generated for each subdomain. Each subdomain is mapped onto a processor in the BBN Butterfly computer. After the tasks of inner domain computation are completed in parallel, the inner boundary values are updated (also in parallel).
AB - Boundary-fitted coordinate transformation broadens the applicability of finite difference methods. However, for a large class of geometries, coordinate transformation introduces singularities and increases grid skewness, which results in large numerical error and slow convergence rate. In this paper, we present the results of combining a finite difference scheme with domain decomposition to obtain a parallel scheme. This scheme is used to simulate steady-state forced convection in irregular axisymmetric and two-dimensional domains. The irregular domain is first dissected into subdomains that have smooth curves as their boundaries. Curvilinear coordinate systems are then generated for each subdomain. Each subdomain is mapped onto a processor in the BBN Butterfly computer. After the tasks of inner domain computation are completed in parallel, the inner boundary values are updated (also in parallel).
UR - http://www.scopus.com/inward/record.url?scp=0025790770&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=0025790770&partnerID=8YFLogxK
U2 - 10.1080/10407799108944993
DO - 10.1080/10407799108944993
M3 - Article
AN - SCOPUS:0025790770
SN - 1040-7790
VL - 20
SP - 41
EP - 59
JO - Numerical Heat Transfer, Part B: Fundamentals
JF - Numerical Heat Transfer, Part B: Fundamentals
IS - 1
ER -