Mohamed Tarek
Mohamed Tarek
Did you actually test this works? It doesn't on my machine.
Well that's odd I tested it on Mac M1 and WSL. Perhaps I need to test it on a more standard setup.
I tried it on my M1 and it still fails. Let me try it again on Linux and Windows and get back to you.
I can review this some time next week if that's not too late.
Seems like a duplicate of #223.
I am traveling this week, but I will give it a look next week. Thanks for the issue.
Ok this is interesting. I managed to reproduce this with a specific seed. I will look into it.
It is definitely possible that it is a bug in my code, but given the number of passed test cases, I am very curious where I might have gone wrong....
@mfherbst I think full orthogonalization of the basis with QR is an almost sure-fire way to make LOBPCG as stable as the QR alg (there is a loop-hole when using...
I think the following approach has potential. Let `U` be the basis `[X P R]`. 1. Remove components of `U` along constraints matrix `C` 2. Find the Gram matrix `gramB...