gf_precond

Purpose

Constructor for getfem preconditioners (which can be used with gf_linsolve).

Synopsis

gf_precond('identity')
gf_precond('cidentity')
gf_precond('diagonal', vec D)
gf_precond('ildlt', spmat M)
gf_precond('ilu', spmat M)
gf_precond('ildltt', spmat M [, int fillin [, scalar threshold]])
gf_precond('ilut', spmat M [, int fillin [, scalar threshold]])gf_precond('superlu', spmat M)

Description

The preconditioners may store REAL or COMPLEX values. They accept getfem sparse matrices and Matlab sparse matrices.

gf_precond('identity') : create a REAL identity precondioner.

gf_precond('cidentity') : create a COMPLEX identity precondioner.

gf_precond('diagonal', @dcvec D) : create a diagonal precondioner.

gf_precond('ildlt', M) : create an ILDLT (Cholesky) preconditioner for the (symmetric) sparse matrix M. This preconditioner has the same sparsity pattern than M (no fill-in).

gf_precond('ilu', M) : create an ILU (Incomplete LU) preconditioner for the sparse matrix M. This preconditioner has the same sparsity pattern than M (no fill-in).

gf_precond('ildlt', M [, fillin [, threshold]]) : create an ILDLT (Cholesky with filling) preconditioner for the (symmetric) sparse matrix M. The preconditioner may add at most fillin additional non-zero entries on each line. The default value for fillin is 10, and the default threshold is 1e-7.

gf_precond('ilut', M [, fillin [, threshold]]) : create an ILUT (Incomplete LU with filling) preconditioner for the sparse matrix M. The preconditioner may add at most fillin additional non-zero entries on each line. The default value for fillin is 10, and the default threshold is 1e-7.

gf_precond('superlu', spmat M) : uses SuperLU to build an exact factorization of the sparse matrix M. This preconditioner is only available if the getfem-interface was built with SuperLU support. Note that LU factorization is likely to eat all your memory for 3D problems.

See Also

gf_linsolve