Open awvwgk opened 2 years ago
It's not practical to simply insert autodiff variables, aka var
into algorithms like lanczos. What's generally done is to implicitly calculate those derivatives using the derivatives of coefficient matrices. See https://jackd.github.io/posts/generalized-eig-jvp/. In that case, one can use AD for coefficient matrix derivatives.
Thanks for the reference, I'll have a look.
Is it possible to use this library for standard/general eigenvalue problems?
I have an application for quantum chemistry in mind which requires the iterative solution of an general eigenvalue problem.