Jutho

Actually, looking more in the computation of the convergence criterion, I don't understand why the authors want to use a relative precision where they divide by rho_j, being the eigenvalue....

Ok thanks for the comments; that is very helpful. When working with real arithmetic, a valid `which` argument should anyway be such that it doesn't distinguish between $$\lambda$$ and $$\lambda\^ast$$...

My apologies for the silence, I was unavailable all of last week. I will try to pick it up again this week and finally push this over the finish line.

My apologies for the silence here; I got a bit distracted by other code reviews. This week I am also not actively at work very much, but I will still...

Ok I finally got around spending some more time on this PR, which was really needed for me to make progress. I have changed quite a few things, but in...

@tipfom, the tests now run locally. I think the actual implementation is now complete, but the tests currently only cover the case with a single iteration of a small problem....

Thanks, I'll take a look asap, such that this can be merged and a new KrylovKit version can be tagged. Was there a specific reason to remove the `eager` option?

@tipfom , I finally made some time to look at this again, and would like to get this merged before tagging a new release of KrylovKit. However, currently, I get...

It seems to be the `eager` option; with `eager=false` everything works like it should. That's probably why you removed it earlier? However, it should be possible to make this work;...

Is there a way that you could provide an explicit example of what you are trying to accomplish. Does "tensor" in your use case refer to an actual multi-index (i.e....