[Trilinos-Users] What are the best parameter choices for ML?

JR Cary cary at txcorp.com
Thu Sep 1 19:24:40 EDT 2016


We have trilinos as the solver in our code, and we present to users the
array of choices of parameters for ML:

    maximum levels = 30
    smoother type = ["Gauss Seidel" "symmetric variable block Gauss Seidel"
        "Jacobi" "Chebyshev" "Aztec"]
    smoother sweeps = 3
    when to smooth = ["before" "after" "both"]
    coarse type = ["Jacobi" "KLU"]
    damping factor = 1.333
    threshold = 0.0
    increase or decrease = ["increasing" "decreasing"]

so we get asked, "What should I choose?"

Are there any guidelines for this?  Or is it more a matter of trial and
error?

For the present case we are doing Poisson with possibly highly varying
coefficients.

FYI, I did go to
https://trilinos.org/docs/dev/packages/ml/doc/html/index.html#ml_doc.
Thought you might want to know that the first link for documentation is
broken.

I did look at mlguide.pdf, which I think is the suggestion, but it seems
to be more
of "how to select foo" rather than "why select foo".

Thanks for any help!

John Cary
 


More information about the Trilinos-Users mailing list