Large eddy simulations of incompressible turbulent flows using parallel computing techniques

This paper presents a detailed procedure to solve incompressible high Reynolds number turbulent flows using large eddy simulations (LES) on distributed memory machines. The filtered Navier-Stokes equations are discretized using a partial-staggered variable arrangement and solved using a finite difference grid. A second-order central difference scheme and sixth-order compact scheme are employed for the spatial derivatives. A third-order low storage Runge-Kutta method is used for the temporal derivatives. Validation of the numerical scheme is performed first by simulating a driven cavity flow and flow over a backward-facing step. The dynamic Smagorinsky subgrid turbulence model is then validated for flow in a channel. Simulations are validated with relevant data available in literature. Since LES is computationally expensive, the solver is parallelized. using message passing interface. An efficient parallel linear equation solver is utilized for solving the elliptical pressure Poisson equation. The parallel program is tested for solutions of flow in a complex flow configuration and preliminary results are compared with experimental data. Performance of the program for the same geometry is tested on a parallel cluster up to 256 processors. The novel approach in this work is the use of a partial-staggered variable arrangement for LES of turbulent flows, obviating the need for any form of artificial dissipation that might mask the subgrid effect on the solution. Copyright (C) 2007 John Wiley & Sons, Ltd.