Andreas Noack

I'm not sure if it will fix the issue but I don't think the package is supposed to overload `promote_type`. Instead, I believe we should define `promote_rule`.

I think we need this. In particular now that we've also defined `isposdef` in https://github.com/JuliaStats/PDMats.jl/pull/123.

Possibly. After filing the issue, I thought a bit more about the problem and I'm not sure what the right solution is. My problem clear wants to set `σ` to...

It's traditional maximum likelihood optimization and most likely the model that triggered this is wrong. However, people will try to fit wrong models it's a matter of failing in a...

>I am very happy to fix this issue, I just need help with what you or other users would consider reasonable. I understand > Personally, since an invariant is violated...

Yes and no. I think much of it should be as simple as loosening the signatures but there might be places where the package makes use of the elements being...

Late here but this is an ARPACK limitation. See section 3.9 of http://li.mit.edu/Archive/Activities/Archive/CourseWork/Ju_Li/MITCourses/18.335/Doc/ARPACK/Lehoucq97.pdf > The reverse communication interface subroutine for the double precision complex eigenvalue problem is znaupd . This...

I'm in favor of this but I'm wondering if we should just use `Cholesky`. We only use this field for computing the uncertainty after the estimation, right, and the inverse...

But with this PR it would fail, no? I'm not sure using pinv is sound when computing a variance estimate so failing might be better.

My guess would be https://github.com/opencollab/arpack-ng/commit/d04bdf19b9189a971f8c3bf2561b8058994e9cb4 but I'm not really sure.