[christian]

I need help in understanding when a RANS computation can be considered to be converged. I monitor both the residuals as well as e.g. inlet pressure (when using a velocity inlet). Another option is to probe a location in terms of e.g. pressure.

1) If both residuals as well as inlet pressure have reached a level where no changes occur, can the solution be regarded as converged? I mean, is there no possibility of oscillating around a "false" solution?

2) When running a second order scheme (e.g. vanLeer) the residuals usually never drop three orders of magnitude as one was taught at university. What does this mean in combination with a monitored inlet pressure that has reachad some kind of constant oscillatory behaviour?

3) When running a second order scheme I start from the first order solution. This leads into my question concerning how the residual is computed. Depending on what solution one starts from, as well as how the residual is calculated, it could be hard to make the residuals drop three orders of magnitude. Comments?

Best regards,
Christian Svensson

[pbo]

Christian,

as regards the 'academic' rule of thumb for iteration convergence (ie dropping 3 orders of magnitude), this only applies to the *unscaled* residuals (ie something like sum_over_cells(RHS-LHS)). In OpenFOAM, the residuals you see are 'scaled' residuals (I am borrowing some terminology from Fluent here, sorry), which means you need not drop 3 orders of magnitude to claim convergence, just have your scaled residuals remaining smaller than a given tolerance (default values used in the tutorial are enough in most cases, however the smaller the better -- if you don't mind an increase in CPU time)

Cheers,

Patrick