PCMulticolorGibbs#

A MulticolorMulticolorGibbs sampler wrapped as a PETSc PC.

Options database keys#

  • -pc_mcgibbs_omega - the SOR parameter (default is omega = 1)

Notes#

This implements a MulticolorGibbs sampler wrapped as a PETSc PC. In parallel this uses a multicolour Gauss-Seidel implementation to obtain a true parallel Gibbs sampler.

Implemented for PETSc’s MATAIJ and MATLRC formats. The latter is used for matrices of the form \(A + B \Sigma^{-1} B^T\) which come up in Bayesian linear inverse problems with Gaussian priors.

This is supposed to be used in conjunction with KSPRICHARDSON, either as a stand-alone sampler or as a random smoother in Multigrid Monte Carlo. As a stand-alone sampler, its usage is as follows: 

KSPSetType(ksp, KSPRICHARDSON);
KSPGetPC(KSP, &pc);
PCSetType(pc, "mcgibbs");
KSPSetUp(ksp);
...
KSPSolve(ksp, b, x); // This performs the sampling
This PC supports setting a callback which is called for each sample by calling 
PCSetSampleCallback(pc, SampleCallback, &ctx, NULL);
where ctx is a user defined context (can also be NULL) that is passed to the callback along with the sample.

Functions#

Warning

doxygenfunction: Cannot find function “PCMulticolorGibbsGetPetscRandom” in doxygen xml output for project “ParMGMC” from directory: build/xml

PetscErrorCode PCMulticolorGibbsSetOmega(PC pc, PetscReal omega)#

Sets the MulticolorGibbs-SOR parameter. Default is omega = 1.