Effect of the numerical dissipation and resolution on large-eddy simulation of turbulent square duct flow

Performance of large-eddy simulation (LES) in the presence of numerical dissipation due to the Rhie-Chow interpolation is assessed for the prediction of turbulent flow in a square duct. A wide range of resolutions and bulk Reynolds numbers Re-b = 2,500 and 5,693 are investigated. A second-order colocated finite-volume solver with the dynamic Smagorinsky (DS) subgrid-scale (SGS) model is employed. To distinguish between the role of the numerical and SGS dissipations, LESs without an SGS model are also performed. LESs without the Rhie-Chow interpolation did not experience numerical instabilities. Use of the Rhie-Chow interpolation, however, increased the error in the wall shear stress and enstrophy predictions, which were more pronounced at coarse resolutions. Significant mis-predictions in Reynolds stress and its anisotropy were also observed at coarse resolutions, which were improved, to some extent, by omitting the Rhie-Chow interpolation. Convergence behaviour of important flow statistics, towards the DNS, were also assessed and resolution requirements were explored.