Publisher Summary In a direct numerical simulation (DNS), all scales of a turbulent flow are supposed to be resolved explicitly without any model assumptions. In the last decade, DNS has provided detailed databases for a small number of turbulent flows in simple geometries, which proved very useful for the validation and development of turbulence models. The simulation is part of a series aimed at studying flows around bluff bodies. The interequation coupling and nonlinearities are resolved iteratively in a predictor-corrector fashion within outer iterations. Linear equation systems are relaxed by conjugate-gradient-based solvers in inner iterations. Because outer iterations have to be performed to account for nonlinearities and coupling of equations, inner iterations do not have to converge to a very tight tolerance. The solution algorithm can be efficiently parallelized by domain decomposition techniques in both space and time. The communication, both local and global, can also be overlapped with computation (where hardware allows) by separating the loops for interfaces with neighbor subdomains and inner cell faces. A combination of time and space parallelization for a given grid size and number of processors usually leads to a higher efficiency than employing one strategy alone.
[1]
I. Demirdzic,et al.
Direct numerical simulation of flow around sphere
,
1997
.
[2]
Joel H. Ferziger,et al.
Computational methods for fluid dynamics
,
1996
.
[3]
William W. Willmarth,et al.
Some experimental results on sphere and disk drag
,
1971
.
[4]
E. Achenbach,et al.
Vortex shedding from spheres
,
1974,
Journal of Fluid Mechanics.
[5]
M. Perić,et al.
Space- and time-parallel Navier-Stokes solver for 3D block-adaptive cartesian grids
,
1996
.
[6]
I. Demirdzic,et al.
Numerical method for coupled fluid flow, heat transfer and stress analysis using unstructured moving meshes with cells of arbitrary topology
,
1995
.
[7]
John L. Lumley,et al.
Computational Modeling of Turbulent Flows
,
1978
.