SimScale CAE Forum

Convergence Criteria confusion

This is regarding a doubt in my “Sandwich simulation” project ->“quarter” ->Run14. In the numerics section I have opted for a value of 5e-4 as convergence criteria tolerance. However when I went through the solver logs, the absolute and relative residuals both had gone below 1e-4.


This picture is to give an idea of what the columns are assigned for. We are looking at the absolute and global residual. As per the tolerance should be less than5e-4


This picture shows the residuals for the newton iterations done for the step involving in increment instant of 5.85937E-04 s. Especially for the 23rd and 24th iterations which have values of the order e-5.

In spite the above residuals being less than the tolerance, I got an error for this run ( Run 14) for newton convergence. Could you explain whether my understanding for tolerance is wrong, or is there some other reason why convergence is not being met ( a criteria other than the tolerance input for numerics) ?

Hi there, can you also provide a link to the project please?

Best regards,
Fillia

1 Like

Hey there,
My bad, I forgot to add the link.
https://www.simscale.com/workbench/?pid=5398130486588740399

Thank you,
Marwan

Hey Marwan!

The two convergence residuals are column 3 and 4, corresponding to Relative and Absolute values, respectively.

As you point out, the residual has met the convergence criteria, especially the Absolute Residual, which is the specified one. You can tell that from the table because there is not an ‘X’ next to the residual value.

But following this indication, you can see that there is another criteria not being met. You can find the ‘X’ on column 8, which corresponds to the geometric residual for the physical contact solution, with a value of 2.5e-4, which is higher than the 1e-5 specified in the Physical Contacts panel.

You can solve this by allowing more Newton iterations (when using Newton solution of physical contacts I usually set the max iterations to 50, under Numerics), and allowing more room for the solver to work. As a less ideal option, you can lower the tolerance to 1e-4 for example.

2 Likes

Thank you very much. I understood the reason why the solution did not converge.
shall try increasing the max number of newton iterations.

1 Like