code_saturne User's Forum

In case of a polyhedral mesh generated by fluent mesher what are the recommendations in term of schemes and linear solver (highly turbulent flow, variable density and CHT) ?
The mesh non-orthogonality is less than 70°.

• Currently I use SOLU for convection terms and leastSquares for gradients calculation

• Looking at the solver output, the computational effort seems mainly affected for solving the Poisson equation (default setting). Could this be improved by changing the linear solver ?

Beside that, does code_saturne use a limiter for gradients calculation (I didn't find this topic on the theory guide)? Typically a slope limiter based on cell to cell or cell to face values (similarly as what OpenFOAM or Fluent do) ?

Best regards,

Hello,

For polyhedral meshes, our feedback is still limited, but your options seem reasonable.

Independently of the mesh type, small time steps usually lead to faster linear solver convergence, so for such cases, gradients and balances (whose cost is independent of the time step) can become dominant or at least 30% of the computational cost.

With larger time steps, more linear solver iterations are needed, so solvers (especially the Poisson solver) dominate.

We have seen that the multigrid-preconditioned solver's coarsenig and convergence can be very sensitive to mesh numbering and partitioning.

In the performance.log file, does the multigrid solver spend more time in the coarsest levels, or is time spent well distributed ?

As regards limiters, with the least squares gradient, v8.0 automatically used a limiter. In v9.0, the setting is decoupled from the gradient choice but limiters can be activated for each variable in.the GUI. This is different from slope tests (it is mainly for the diffusion part), and slope tests are also used by default (maybe not for SOLU, I need to check). In the development version, we can also clip reconstructed values to neighboring cell bounds as an alternative to gradient limiters (for better control/precision), and plan to decouple some diffusion and convection options, but this is still work in progress.

Best regards,

Yvan