Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.

Detailed Log information - No convergence because of spatial or time discretization ?

Please login with a confirmed email address before reporting spam

Hello,

see attached the detailed log of my time dependent simulation with linear material properties. The model fails to converge and I don't know why. By looking at the log, can one see where the problem might be?

From my point of view, "SolEst" and "LinRes" look okay. I think the problem might be with"LinErr". But what does this mean practical at all?

Kind regards, Jan



4 Replies Last Post 5 lug 2018, 07:25 GMT-4

Please login with a confirmed email address before reporting spam

Posted: 6 years ago 3 lug 2018, 05:16 GMT-4

I have to add, that the model basically works. I only changed one little thing from the working model: a rotation was added.

I have to add, that the model basically works. I only changed one little thing from the working model: a rotation was added.

Henrik Sönnerlind COMSOL Employee

Please login with a confirmed email address before reporting spam

Posted: 6 years ago 4 lug 2018, 02:12 GMT-4
Updated: 6 years ago 4 lug 2018, 05:28 GMT-4

Hi Jan,

This indicates that the system matrix is ill-conditioned. There can be many underlying reasons, ranging from the actual physical properties of the problem you are solving to erroneous boundary conditions or material properties.

If the solver for the 'innermost' problem (the linear solver) in your iterations reports this kind of problem, the 'outer' nonlinear solver will not accept the solution as converged, even though its own criteria are fulfilled.

You can force the nonlinear solver to accept the solutions nevertheless. Change the settings in the 'Error' section in the 'Direct' node in the solver sequence. This is usually safe in a nonlinear problem, since a 'bad' solution will give extra iterations. Your graph indicates that there are no large residuals, so suppressing the check on the solution from the linear solver seems OK.

Regards,
Henrik

-------------------
Henrik Sönnerlind
COMSOL
Hi Jan, This indicates that the system matrix is ill-conditioned. There can be many underlying reasons, ranging from the actual physical properties of the problem you are solving to erroneous boundary conditions or material properties. If the solver for the 'innermost' problem (the linear solver) in your iterations reports this kind of problem, the 'outer' nonlinear solver will not accept the solution as converged, even though its own criteria are fulfilled. You can force the nonlinear solver to accept the solutions nevertheless. Change the settings in the 'Error' section in the 'Direct' node in the solver sequence. This is usually safe in a nonlinear problem, since a 'bad' solution will give extra iterations. Your graph indicates that there are no large residuals, so suppressing the check on the solution from the linear solver seems OK. Regards, Henrik

Please login with a confirmed email address before reporting spam

Posted: 6 years ago 4 lug 2018, 05:24 GMT-4
Updated: 6 years ago 4 lug 2018, 05:30 GMT-4

Hi Henrik,

thank you very much for your reply. I am testing the model without the Error check right now.

Regards,
Jan

Hi Henrik, thank you very much for your reply. I am testing the model without the Error check right now. Regards, Jan

Please login with a confirmed email address before reporting spam

Posted: 6 years ago 5 lug 2018, 07:25 GMT-4

Hi again,

I tried Henriks tip with the Error check and the model does solve for a few time steps now (see the attached plot of the Convergence). Unfortunately, after some succesfull time steps, the nonlinear solver fails to converge.

I checked if the dependent variables show questionable behaviour. They look very smooth and not suspicious. From my point of view, the results look physically correct, but the solver says otherwise...

Kind regards, Jan

Hi again, I tried Henriks tip with the Error check and the model does solve for a few time steps now (see the attached plot of the Convergence). Unfortunately, after some succesfull time steps, the nonlinear solver fails to converge. I checked if the dependent variables show questionable behaviour. They look very smooth and not suspicious. From my point of view, the results look physically correct, but the solver says otherwise... Kind regards, Jan

Note that while COMSOL employees may participate in the discussion forum, COMSOL® software users who are on-subscription should submit their questions via the Support Center for a more comprehensive response from the Technical Support team.