Dominique
Dominique
@tcovert The two solvers are slightly different, even on unconstrained problems. Both are similar trust-region methods, but they use slightly different parameters, different trust-region management, etc. That's not to say...
Thank you for the report. If I understand well, the issue is not with multiple right-hand sides, but rather with successive solves, where the factorization is reused?!
Also, each call to `jac_coord` evaluates the constraints and allocates a vector. That would be easy to preallocate.
Could we just leave `Tf` as an (unspecified) parameter?
As you pointed out earlier, we need to use `conj` in appropriate places to ensure correctness, but I think I would leave `Tf` free.
I would expect the permutations to be identical though. SYMAMD is deterministic as far as I know and only depends on the sparsity pattern. @VikasChidananda Have you confirmed that the...
One possible difference is that we don't set default options explicitly: * https://github.com/DrTimothyAldenDavis/SuiteSparse/blob/dev/CCOLAMD/MATLAB/csymamdmex.c#L90 vs * https://github.com/JuliaSmoothOptimizers/AMD.jl/blob/main/src/COLAMD.jl#L169 (should be the same but we should double check). Another is these options: https://github.com/DrTimothyAldenDavis/SuiteSparse/blob/dev/CCOLAMD/MATLAB/csymamdmex.c#L99...
@VikasChidananda I wonder if #61 fixes your issue.
@geoffroyleconte Would anything break if we changed `Real` to `Number` here?
@abelsiqueira Does your CUTEst installer also install the single precision library? If not, we're going to need it.