| A Newton-Krylov algorithm is presented for the compressible Navier-Stokes equations on hybrid unstructured grids. The Spalart-Allmaras turbulence model is used for turbulent flows. The spatial discretization is based on a finite-volume matrix dissipation scheme. A preconditioned matrix-free generalized minimal residual method is used to solve the linear system that arises in the Newton iterations. The incomplete lower-upper factorization based on an approximate Jacobian is used as the preconditioner after applying the reverse Cuthill-McKee reordering. Various aspects of the Newton-Krylov algorithm are studied to improve efficiency and reliability. The inexact Newton method is studied to avoid over-solving of the linear system to reduce computational cost. The ILU(1) approach is selected in three dimensions, based on a comparison among various preconditioners. Approximate viscous formulations involving only the nearest neighboring terms are studied to reduce the cost of preconditioning. The resulting preconditioners are found to be effective and provide Newton-type convergence. Scaling of the linear system is studied to improve convergence of the inexact matrix-free approach. Numerical studies are performed for two-dimensional cases as well as flows over the ONERA M6 wing and the DLR-F6 wing-body configuration. A ten-order-of-magnitude residual reduction can be obtained with a computing cost equivalent to 4,000 residual function evaluations for two-dimensional cases, while the same convergence can be obtained in 5,500 and 8,000 function evaluations for the wing and wing-body configuration, respectively, on grids with a half million nodes. |
