Skip to content

Implement Exciting-Mixing, Scalar and Diagonal Broyden approximations #260

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

Closed
ChrisRackauckas opened this issue Jan 17, 2023 · 3 comments · Fixed by #303
Closed

Implement Exciting-Mixing, Scalar and Diagonal Broyden approximations #260

ChrisRackauckas opened this issue Jan 17, 2023 · 3 comments · Fixed by #303

Comments

@ChrisRackauckas
Copy link
Member Author

@avik-pal is this completed as well?

@avik-pal
Copy link
Member

No. Maybe let's transfer this to NonlinearSolve? We can handle all of them with branches in Broyden there

@ChrisRackauckas ChrisRackauckas transferred this issue from SciML/SimpleNonlinearSolve.jl Oct 25, 2023
@avik-pal
Copy link
Member

avik-pal commented Dec 6, 2023

LinearMixing is just J = -1 / alpha, and J is never updated, I don't think there is any point of having that, for small problems our Broyden operates in nanoseconds via StaticArrays and for larger problems this won't work anyways

Even ExcitingMixing is a very crappy update rule, unless we have some compelling usecase for that, we should not clutter the implementation.

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

Successfully merging a pull request may close this issue.

2 participants