superLU occasionally iterates on a matrix that has obviously become singular. fault_check can be put into the system to supplement this check, but there may be options inside superLU to fix it.
An approach to this may be to look for #INDs and other nasty results during the convergence check and variable update, failing out before superLU "converges" to invalid answers.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
There do not appear to be any superLU flags to make it more "robust". Deferring this to 3.0 so as more examples pop up, we can handle them.
Diff:
An approach to this may be to look for #INDs and other nasty results during the convergence check and variable update, failing out before superLU "converges" to invalid answers.