Residuals and solution ratio

Fededea

Hi,

I am new to CFD FEM simulation. Should the residuals in a steady state simulation reach the same order as other CFD SWs (something like 1e-6) or can the default value of 1e-3 be fine? Also is the number of iterations, which defaults to 100, that numbers needs to be higher?

And what trend should the solution ratio have?

Both the residuals and solution ratio graphs have a decreasing trend with only a few spikes at times (but not that high).

I ask these questions because in my simulation I reach these values and the simulation stops after 20-30 iterations, and I have some doubt about the convergence (I checked the results with another SW and they are almost the same, the results change by something like 1-5%).

Thanks in advance

acupro

By default, AcuSolve looks at both Residual Ratio (a measure of how well the overall solution matches the equations) and Solution Ratio (a measure of how much the overall solution is changing from time step to time step).  By default the Solution Ratio values can be an order of magnitude larger than the Residual Ratio values. For Solution Ratio, with convergence tolerance at 0.001, eddy-viscosity would need to be below 0.1, pressure and velocity below 0.01.  You can review the CONVERGENCE_CHECK_PARAMETERS in the <problem>.run.ss.inc file for the default stopping criteria, based on the convergence tolerance.  'standard' means at or below the convergence tolerance value, looser_by_10 means one order of magnitude higher, etc.  You also need to determine what results are of interest.  If those have stabilized to what you would term 'steady' maybe that is enough.  These automatic settings are just tools for guidance.  'Convergence' is more of a convention/practice rather than a hard definition.

Generally 1e-3 for AcuSolve is similar to 1e-5/1e-6 for some other CFD solvers.  The normalization factors used for residual and solution ratios are updated each time step rather than being computed at the start and held constant.

Although the default convergence/stoppage may be reached in a just a few time steps (you mentioned 20 to 30 in your case) there are many linear solver iterations done during each of those time steps.  Those show in the GMRES / CGP iterations within the time step.