[petsc-users] Cholesky matrix factorization

[Barry Smith]

On Sep 8, 2012, at 3:36 PM, Jed Brown <jedbrown at mcs.anl.gov> wrote:

> On Sat, Sep 8, 2012 at 3:19 PM, Pearl Flath <pflath at ices.utexas.edu> wrote:
> Hi all,
> 
> I have a matrix M, for which I set up a Cholesky decomposition M = L L^T  using MUMPS and PETSc. I then obtained the factored matrix with
> 
>   PCFactorGetMatrix(pc,&factor);  
> 
> However, I do not want to solve M*x = b, which seems to be what MatSolve does. How do I tell it to solve Lx = b instead? 
> 
> Looks like you want MatForwardSolve().

   We don't have an interface for just the forward solve for the MUMPS from PETSc. If MUMPS provides separate routines for forward and backward solve (check the mumps documentation)
let us know and we'll add the forward and backward solve for our MUMPS interface.

   Barry

> 
> Note that there is also PCApplySymmetric{Left,Right}.