Petar Mlinarić

> > Thanks for working on this @TiKeil. > > Could also a COO format (e.g., a triple of row indices, column indices, and `Operators`) be an input to `__init__`?...

> > I meant doing it similarly to [`coo_matrix`](https://docs.scipy.org/doc/scipy/reference/generated/scipy.sparse.coo_matrix.html), which accepts different types as the first argument (in particular, a NumPy array and a COO format). > > Ah ok,...

So, I wanted to check if new SciPy sparse arrays allow object dtype, but then I realized I was mistaken, and actually some sparse matrix formats already do allow object...

I agree, it seems that there is a bug in `coo_matrix` (this https://github.com/scipy/scipy/pull/15828 changed the warning to an error). Using a special sparse format seems necessary.

> I would like to know how a `LoewnerReductor` relates to the `TFBHIReductor` we have. `TFBHIReductor` constructs the version of Loewner matrices for bitangential Hermite interpolation (used in TF-IRKA), so...

> @sdrave @drittelhacker @pmli Does this make sense to you? Should I move forward and implement this? I don't think it would become a problem but there are no proven...

> I am not sure, if the logger should produce a warning, if `scipy.linalg.cholesky` breaks down - maybe it is too bewildering? Adding more logs might be nice, I guess....

> However, if the input data is zero, this will still lead to an 'orthonormal' basis of zero vectors. I'd say the only practical approach would be to fail in...

> > > However, if the input data is zero, this will still lead to an 'orthonormal' basis of zero vectors. I'd say the only practical approach would be to...

It seems that the tests are failing because of an infinite loop when a zero vector array is given. I guess shifted Cholesky can cheaply check if the input is...