Open albertomercurio opened 1 year ago
Hi @albertomercurio,
I'm not sure to understand your issue, do you want an eigenvalue solver in Krylov.jl?
This package is dedicated to Krylov processes and Krylov methods for linear problems. it's not a catch-all package like IterativeSolvers.jl and KrylovKit.jl.
I'm open to add singular value / eigenvalue solvers but I'm not an expert in that field.
Those could easily go into an extension package though.
I'm not sure to understand your issue, do you want an eigenvalue solver in Krylov.jl?
This package is dedicated to Krylov processes and Krylov methods for linear problems. it's not a catch-all package like IterativeSolvers.jl and KrylovKit.jl.
Ok, clear.
Anyway, the eigenvalues solvers use the Krylov subspaces, and I thought to this package since it supports these algorithms also on GPUs.
Hi,
I think it could be great to introduce the support for eigenvalue solvers. I noticed that the implicit restarted method has already been passed. The modern faster method should be the Krylov-Schur method, which is already used in the KrylovKit.jl and ArnoldiMethods.jl packages.
A good reference should be this one.
Looking at the KrylovKit code, and using the ExponentialUtilities.jl arnoldi struct, I wrote the following code, which is way faster that implicit restarted method. And it support already the GPU arrays.
I opened an issue (https://github.com/JuliaLang/julia/issues/47291) to Julia for support the LAPACK
hseqr!
solver natively, but for the moment I wrote it by my own.