Dismiss Notice
Join Physics Forums Today!
The friendliest, high quality science and math community on the planet! Everyone who loves science is here!

[PETSC] which method ?

  1. Jul 14, 2009 #1

    I have to solve a linear equation as a part of a more global simulation. The equation is :

    [tex]\mathbf{B} - \mu\Delta \mathbf{B} = \mathbf{S}[/tex]

    Where both [tex]\mathbf{B}[/tex] and [tex]\mathbf{S}[/tex] are 3 component vectors fields, depending on x and y (2D fields).

    My grid is regular, and has a typical size of [tex]n_x=n_y=512[/tex] (nx can be different from ny).

    Boundary conditions are periodic on x borders, but dirichlet or neuman on y borders (but are the same for both y borders). The choice of Dirichlet or Neumann depends on the Field component that is solved.

    There are then 3 equations (one for each field component) that have to be solved 4 times per time step of the simulation.

    Typical simulation is about 30000 time steps, thus the 3 equations must be solved 120000 times. I need an efficient and parallel method.

    I've begun to learn the PETSC library, which is a high level and parallel library for this kind of problem. But I'm not mathematician, and a bit lost with all the non stationnary iterative methods.

    I have made a sum up of my problem on the pdf here : http://nau.cetp.ipsl.fr/matrixinertia.pdf [Broken]

    Can someone first tell me if the matrices I've calculated and written in this document are ok ? (I'm not used to do this...)

    Then, I observe that, because of the boundary conditions, these two matrices are not symmetric... could someone tell me which method in Petsc should be the best for my problem ? (methods are presented here : http://www.mcs.anl.gov/petsc/petsc-as/documentation/linearsolvertable.html [Broken])

    Thanks a lot
    Last edited by a moderator: May 4, 2017
  2. jcsd
Share this great discussion with others via Reddit, Google+, Twitter, or Facebook

Can you offer guidance or do you also need help?
Draft saved Draft deleted