The tolerance criteria used to check for convergence

Forum for OpenSees users to post questions, comments, etc. on the use of the OpenSees interpreter, OpenSees.exe

Moderators: silvia, selimgunay, Moderators

Post Reply
Gholamreza
Posts: 84
Joined: Tue Nov 07, 2017 7:47 am
Location: University of Central Florida

The tolerance criteria used to check for convergence

Post by Gholamreza » Mon Feb 08, 2021 3:20 pm

Hello Engineers,

Can anyone tell me what is the best value for the tolerance in:

test NormDispIncr $tol $iter <$pFlag> <$nType>

I assume the smaller value, the more accurate response I get but I might face convergence issues.
I modeled 9 story benchmark building, yielding material and I am using 1.0e-4 as the tolerance. Is it accurate enough? how can I know that? or in general how I find the best value for the tolerance.

selimgunay
Posts: 879
Joined: Mon Sep 09, 2013 8:50 pm
Location: University of California, Berkeley

Re: The tolerance criteria used to check for convergence

Post by selimgunay » Tue Feb 09, 2021 2:56 pm

1e-4 seems to be a bit large. The best way is to use a smaller tolerance (e.g., 1e-8) and if non-convergence happens, try to overcome that using adaptive strategies (including increase of the tolerance only at the problematic time step). You can also consider using energyincr to start with as it considers both force and displacement

If your analysis was successfully completed with 1e-4 tolerance, one rough way of checking if that is accurate enough, is to run it with smaller tolerances and compare the results

Gholamreza
Posts: 84
Joined: Tue Nov 07, 2017 7:47 am
Location: University of Central Florida

Re: The tolerance criteria used to check for convergence

Post by Gholamreza » Tue Feb 09, 2021 3:20 pm

Thanks selimgunay, I am using adaptive strategies you shared before but it seems it has to reduce the tolerance in almost every steps to converge....is this shows my model has some issues? I developed a new yielding material in C++ and I am using that in 9-story building.....the new material works just fine in SDOF and I verified that with MATLAB. In my MDOF structure, eigen analysis is just fine and I get 5 first frequencies of the structure as I expected but in order to do the time-hisotry analysis, if I use a very small tolerance and the adaptive strategies, it takes so long.....is this reasonable?

selimgunay
Posts: 879
Joined: Mon Sep 09, 2013 8:50 pm
Location: University of California, Berkeley

Re: The tolerance criteria used to check for convergence

Post by selimgunay » Tue Feb 09, 2021 8:40 pm

If the convergence issues start at low levels of inelasticity, that may indicate that there might be some issues with your material. But, overall if it takes long with many iterations, but if it gets completed with meaningful results, it should be okay

Post Reply