Fixed a bug in CFiniteElementStd::Assemble_RHS_M() #93
Conversation
|
@norihiro-w I am preparing the final version of ogs5, v5.8, which will include some material models developed by BGR and some bug fixing. If there is any fundamental changes, I would like to have your comment. For minor bug fixing and changes, I may merge it quickly. |
| @@ -795,13 +795,16 @@ double CRFProcessDeformation::Execute(int loop_process_number) | |||
| { | |||
| pcs_absolute_error[0] = NormU; | |||
| pcs_relative_error[0] = pcs_absolute_error[0] / Tolerance_global_Newton; | |||
| cpl_max_relative_error = pcs_relative_error[0]; | |||
| // To be improved by overloading CRFProcess::CalcIterationNODError | |||
| cpl_max_relative_error = pcs_absolute_error[0]; //pcs_relative_error[0]; | |||
There was a problem hiding this comment.
two comments
- why cannot you use
pcs_relative_error[0]? - it's quite confusing if
cpl_max_relative_erroris actually absolute error. isn't therecpl_max_abs_error?
There was a problem hiding this comment.
cpl_max_relative_error is only set here under if (ite_steps == 1). If cpl_max_relative_error is set to pcs_relative_error[0]( = (pcs_absolute_error[0] / Tolerance_global_Newton), the coupling convergence criterion of cpl_max_relative_error < 1 is never satisfied for some cases. Anyway, this implementation is not correct. Therefore, I made this change just to make sure that computation of the modelling of M related coupled processes can keep running, while its coupling convergence is controlled by other processes. As you can see in the comment above the changed line, the method in CRFProcess::CalcIterationNODError will be used to calculate the coupling error of M process.
There was a problem hiding this comment.
@norihiro-w Thanks for your comments. I am going to merge this PR.
|
@wenqing Thanks for preparing the new version! I have only one minor comment. If that is resolved, please go ahead. |
and some other minor changes.