- When a symmetric matrix (Hessian) is involved, MA57 is used (ldl), which is faster than suitesparse.
- The symbolic factorization is actully the reordering that produces sparser factors. It's done by default with amd (metis is another option).
Symbolic factorization of a large sparse matrix
7 views (last 30 days)
I'm solving a series of problems Ax=b, where A changes, but still has the same sparsity pattern. To exploit that, in the initialization, I only perform symbolic factoriazation once.
Currently, I use pardiso:
I was wondering how come matlab doesn't offer something like this, considering that the mldivide (based on Tim Davis's SuiteSparse) is supposed to be the fastest option?