Large eddy simulation (LES) of compressible periodic channel flow is performed using a fourth-order finite difference scheme for a Reynolds number based on bulk density, bulk velocity and channel half-width equal to 3000. Two configurations are studied: a subsonic case (M-0 = 0.5) that corresponds to the experiments of Niederschulte et al. ['Measurements of turbulent flow in a channel at low Reynolds numbers', Exp. Fluids, 9, 222-230 (1990)] and a supersonic case (M-0 = 1.5) that corresponds to the direct numerical simulation (DNS) results by Coleman et al. ['A numerical study of turbulent supersonic isothermal-wall channel flow', J. Fluid Mech., 305, 159-183 (1995); 'Compressible turbulent channel flows: DNS results and modeling', J. Fluid Mech., 305, 185-218 (1995)]. In order to determine the influence of the discretization, two cases are computed using two different meshes, a coarse one and a fine one. Two subgrid-scale models are tested: the first one is an extension to compressible flows of the Smagorinsky model, while the second one is a model based both on large and small scales of turbulence, a hybrid Bardina-selective mixed scale model. Various statistical comparisons are made with experimental and DNS data at similar Reynolds numbers, including higher-order statistics. Copyright (C) 2000 John Wiley & Sons, Ltd.

A comparative assessment of six subgrid-scale models is presented in the case of a compressible plane channel flow over isothermal walls, four of which are based on the recent mixed-scale model. A supersonic and a subsonic case in temporal development at Mach 1.5 and 0.5, respectively, are considered for a Reynolds number equal to 3000, with comparison to the direct numerical simulations (DNS) of Coleman et al. in the first case, and the incompressible DNS of Kim et al. in the second case. For each Mach number, two grids are considered, with about 20 and 5 times less grid points as in the reference DNS, and the results are globally satisfactory, although not excellent. In particular, the wall friction is globally underestimated by about 10%, both in the subsonic and the supersonic cases. Among the models tested are several original combinations, which vanish at the wall without requiring any damping function