petsc-3.13.6 2020-09-29
Report Typos and Errors

MatGetSchurComplement

Obtain the Schur complement from eliminating part of the matrix in another part.

Synopsis

#include "petscksp.h" 
PetscErrorCode  MatGetSchurComplement(Mat A,IS isrow0,IS iscol0,IS isrow1,IS iscol1,MatReuse mreuse,Mat *S,MatSchurComplementAinvType ainvtype,MatReuse preuse,Mat *Sp)
Collective on A

Input Parameters

A - matrix in which the complement is to be taken
isrow0 - rows to eliminate
iscol0 - columns to eliminate, (isrow0,iscol0) should be square and nonsingular
isrow1 - rows in which the Schur complement is formed
iscol1 - columns in which the Schur complement is formed
mreuse - MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX, use MAT_IGNORE_MATRIX to put nothing in S
ainvtype - the type of approximation used for the inverse of the (0,0) block used in forming Sp: MAT_SCHUR_COMPLEMENT_AINV_DIAG, MAT_SCHUR_COMPLEMENT_AINV_BLOCK_DIAG, or MAT_SCHUR_COMPLEMENT_AINV_LUMP
preuse - MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX, use MAT_IGNORE_MATRIX to put nothing in Sp

Output Parameters

S - exact Schur complement, often of type MATSCHURCOMPLEMENT which is difficult to use for preconditioning
Sp - approximate Schur complement from which a preconditioner can be built

Note

Since the real Schur complement is usually dense, providing a good approximation to newpmat usually requires Section 1.5 Writing Application Codes with PETSc-specific information. The default for assembled matrices is to use the inverse of the diagonal of the (0,0) block A00 in place of A00^{-1}. This rarely produce a scalable algorithm. Optionally, A00 can be lumped before forming inv(diag(A00)).

Sometimes users would like to provide problem-specific data in the Schur complement, usually only for special row and column index sets. In that case, the user should call PetscObjectComposeFunction() on the *S matrix and pass mreuse of MAT_REUSE_MATRIX to set "MatGetSchurComplement_C" to their function. If their function needs to fall back to the default implementation, it should call MatGetSchurComplement_Basic().

MatCreateSchurComplement() takes as arguments the four submatrices and returns the virtual Schur complement (what this returns in S).

MatSchurComplementGetPmat() takes the virtual Schur complement and returns an explicit approximate Schur complement (what this returns in Sp).

In other words calling MatCreateSchurComplement() followed by MatSchurComplementGetPmat() produces the same output as this function but with slightly different inputs. The actually submatrices of the original block matrix instead of index sets to the submatrices.

Developer Notes

The API that includes MatGetSchurComplement(), MatCreateSchurComplement(), MatSchurComplementGetPmat() should be refactored to remove redundancy and be clearer and simplier.

See Also

MatCreateSubMatrix(), PCFIELDSPLIT, MatCreateSchurComplement(), MatSchurComplementAinvType

Level

advanced

Location

src/ksp/ksp/utils/schurm/schurm.c
Index of all KSP routines
Table of Contents for all manual pages
Index of all manual pages