Open bvdmitri opened 11 months ago
What is the rationale behind this line? https://github.com/timholy/PositiveFactorizations.jl/blob/master/src/cholesky.jl#L96 It basically makes the factorization completely wrong in marginal cases, where the diagonal entry is slightly below the threshold.
@bvdmitri I have a problem where doing a modified Cholesky is definitely the way to go. Is this (currently) unresolved issue drastic enough mean that this package shouldn't be used, and I should try coding it up for myself rather than trusting the output here?
Consider this matrix:
It is not particularly good, but also not awfully bad, eg
Here is what I get with the package against the Julia base
Basically the factorizations are almost identical with the exception of the last element. It also gives the wrong inverse