Skip to content
New issue

Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? # to your account

Cholesky decompositon of spd BandedBlockBanded matrices #104

Open
TSGut opened this issue Apr 27, 2022 · 1 comment
Open

Cholesky decompositon of spd BandedBlockBanded matrices #104

TSGut opened this issue Apr 27, 2022 · 1 comment

Comments

@TSGut
Copy link
Contributor

TSGut commented Apr 27, 2022

How much work would it be to get cholesky working on infinite spd BandedBlockBanded matrices?

I'm wondering whether decomposition methods also allow for interesting weight OPs in higher dimension and want to test this with the Jacobi matrices on the Zernike disk. I can always experiment with finite blocks for now, just wanted to put the thought out there.

@dlfivefifty
Copy link
Member

dlfivefifty commented May 4, 2022

First you would need to finish block Cholesky which was started here:

JuliaArrays/BlockArrays.jl#126

As you can see it uses LAPACK's cholesky. This should work for BlockBandedMatrix. I believe then it would be pretty easy to get the infinite case working.

A more interesting question is how to take advantage of structure in BandedBlockBandedMatrix. Since Cholesky has a simple rank update formula (which I believe is stable, unlike Woodbury formula) it is possible to

  1. parallelise the Cholesky factorization.
  2. Reduce complexity to O(N^(3/2)) for N unknowns.

But this is a bigger undertaking.

# for free to join this conversation on GitHub. Already have an account? # to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants