eslickj
commented
2 years ago Fairly sure I checked the constraint bounds before calling fg_read(). When using PETSc I don't have inequity constraints, and I guess the allocated memory is usually full of zeros. It looks like this mistake wouldn't affect solutions, just checking for inequalities.
Sometime the PETSc nl-solver wrapper wrongly classifies equities as inequalities and fails due to degrees of freedom. It doesn't happen very often and it seems kind of random, in a way that makes me think it may be a memory mismanagement issue. The solver seems to work fine otherwise, so whatever the issue is, it's sneaky.