New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add weighted GMRES(m) #325
Comments
Interesting. So is the approach to simply replace the Euclidean inner products with |
I sent a PR for the H-norm So far, I could not implement the H-norm Line 50 in af81d67
I think it would be judicious to create another issue (and related PR) later on for the |
|
According to Householder orthogonalization with a non-standard inner product (Meiyue Shao), it sounds like Householder reflections for H-inner products are of the form (see section 2):
Householder-based implementation sounds more complicated than what I was expecting. I think we are fine with the MGS-based H-GMRES. In the appropriate setting, |
Refer to Some observations on weighted GMRES (Güttel, Stefan and Pestana, Jen) Algorithms 1&2 p. 13. |
I propose to implement the weighted GMRES(m), as this variant can lead to mesh independent convergence results for FEM/BEM simulations (see e.g. this article).
The weighted version requires only a few changes to the code as compared to its classic Euclidean version.
It would take
H
(a supposedly Hermitian positive definite)LinearOperator
as an input.I think it could be interesting to add it to
pyamg
, as I expect theH^1(\Omega)
Sobolev norm GMRES to outperform the Euclidean one (I am not so familiar withpyamg
).Let me know if you have further comments. I have already some work in progress and will try to send a PR soon.
Also refer to: scipy/scipy#15693 (comment) for its scipy variant (also work in progress).
The text was updated successfully, but these errors were encountered: