Permalink
Commits on Jun 10, 2009
Commits on Dec 9, 2008
Commits on Aug 15, 2008
  1. Add plain C support, testing conditioning of perturbed matrices

    Makefile: Allow a plain C version of PETSc to be used
    stokes.C: Testing conditioning of perturbed matrices
    
    Add option to filter matrix for testing effect of imposing
    Dirichlet conditions by zeroing diagonal.  The option are only effective if
    -pc_type 1 (Q1 finite element) preconditioning is chosen.  There are two new
    options:
    
      -zeroN Int      number of rows and columns to eliminate
      -zeroV Scalar   value to place on the diagonal
    
    We can test the conditioning of the velocity matrix with
    
      -vel_ksp_type preonly -vel_pc_type ksp -vel_ksp_ksp_rtol 1e-10 \
      -vel_ksp_ksp_gmres_restart 500 -vel_ksp_pc_type none \
      -vel_ksp_ksp_monitor_singular_value
    
    and the conditioning of the Schur complement with
    
      -schur_pc_type none -svel_ksp_type preonly -svel_pc_type ksp \
      -svel_ksp_rtol 1e-10 -svel_ksp_pc_type ilu -schur_ksp_gmres_restart 500 \
      -schur_ksp_monitor_singular_value
    
    Note that the divergence matrices are still applied to spectral order and
    matrix-free.  Hence the procedure here is only a first-order approximation
    to the effect on conditioning since we are not removing the appropriate rows
    and columns from these matrix-free operators.
    committed Aug 15, 2008
Commits on Jul 17, 2008
Commits on May 30, 2008
Commits on May 16, 2008
  1. Added an upper triangular preconditioner and flexible matrix format.

    Select the upper triangular preconditioner with -pc_saddle_type 1.
    committed May 16, 2008
Commits on Apr 30, 2008
Commits on Apr 28, 2008
  1. Added README.

    committed Apr 28, 2008
Commits on Apr 22, 2008
Commits on Apr 12, 2008
  1. Minor cleanup.

    committed Apr 12, 2008
Commits on Apr 9, 2008
  1. Implemented approximating the spectral matrix using a coloring. Unfor…

    …tunately,
    
    this is even worse quality than the finite element approximation.
    committed Apr 9, 2008
  2. Added formation of the element mass matrix. This still doesn't seem t…

    …o help.
    
    The finite difference preconditioner is still significantly stronger.
    committed Apr 9, 2008
  3. Finite element preconditioner works, but is actually not as strong as…

    … the finite difference one.
    
    After much bug hunting, I became convinced that the inner part is actually
    correct, but the Q1 finite element preconditioner is just not as good as the
    finite difference preconditioner, at least applied naively.  I believe the root
    of the problem is that we need to apply the mass matrix in order to get the
    scaling correct.
    committed Apr 9, 2008
Commits on Apr 8, 2008
Commits on Apr 7, 2008
Commits on Apr 5, 2008
Commits on Apr 4, 2008
Commits on Apr 3, 2008
  1. Quieter build.

    committed Apr 1, 2008
Commits on Apr 1, 2008
Commits on Mar 30, 2008
  1. Much cleaner design and theory.

    committed Mar 30, 2008
  2. Started decomposition solver.

    committed Mar 30, 2008
Commits on Mar 28, 2008
  1. Merge branch 'master' into stokes

    Jed Brown committed Mar 28, 2008
  2. Reduce order of polynomial for pressure.

    Jed Brown committed Mar 28, 2008
  3. Added polynomial interpolation.

    Jed Brown committed Mar 28, 2008
Commits on Mar 27, 2008
  1. Put constant pressure into the null space.

    Jed Brown committed Mar 27, 2008