Error tolerance [*1e-3]

[Silverbrook]

Hello,

I'm not sure what this should be set at. The default value is 0.1 [*1e-3]. What justification is there for changing this from the default in a model that is not converging?

Thank you,

SB

[Pete Sinton]

non-convergence in a feflow model is often due to a problem with the way the model is set up.  If you are not sure what is causing the problem, a good way to identify problems is to run tests in which you set up the model with  a few inputs as possible, then add complexity until the problem shows up (or do the same in reverse: remove things until convergence improves).

After you get a model that behaves better, you can refine the solver settings.

Pete

[Silverbrook]

Thanks Pete. I found a problem with a boundary condition that was causing the issues.

Out of interest, and for future reference, what does the 0.1 [*1e-3] error tolerance represent, and under what circumstances would we consider altering it?

[Pete Sinton]

It depends on which tolerance type you choose.  L1 compares the tolerance to an average of all "changes" across the domain, L2 compares the tolerance to the deviation around the average change, and max compares the tolerance to the max change.  "Change" depends on what's being solved (head, solute mass, heat energy).  L1 is the least stringent and "max" is the most stringent.  Which you choose depends on the complexity of the model: more complex often require more stringent comparison.  More complex models also may require smaller tolerance value.  To help you decide, run tests on your model.  There is no better way to select the right combination of settings.

Pete 

[Silverbrook]

So would it be right to say that when applying the error norm to the max error that it is essentially working to try and solve by ensuring that any areas that are numerically problematic are focused on during the solving routine?

[Bastian Rau]

Some additional information: The absolute value of error depends on the maximum value of the primary variable in the model. For flow models this is usually the head. If your model has high elevations, you may need to set a smaller error criterion. For more details, please refer to the help-system: http://www.feflow.info/html/help/HTMLDocuments/reference/dialogs/problemsettings/numerical_parameters.htm
 

[Pete Sinton]

Silverbrook: yes, using max error as the measure will focus model solver on areas with largest error.