#include "petscmat.h" PetscErrorCode MatSetNullSpace(Mat mat,MatNullSpace nullsp)Logically Collective on Mat and MatNullSpace
| mat | - the matrix | |
| nullsp | - the null space object |
For inconsistent singular systems (linear systems where the right hand side is not in the range of the operator) you also likely should call MatSetTransposeNullSpace(). This allows the linear system to be solved in a least squares sense.
You can remove the null space by calling this routine with an nullsp of NULL
The fundamental theorem of linear algebra (Gilbert Strang, Introduction to Applied Mathematics, page 72) states that the domain of a matrix A (from R^n to R^m (m rows, n columns) R^n = the direct sum of the null space of A, n(A), + the range of A^T, R(A^T). Similarly R^m = direct sum n(A^T) + R(A). Hence the linear system A x = b has a solution only if b in R(A) (or correspondingly b is orthogonal to n(A^T)) and if x is a solution then x + alpha n(A) is a solution for any alpha. The minimum norm solution is orthogonal to n(A). For problems without a solution the solution that minimizes the norm of the residual (the least squares solution) can be obtained by solving A x = \hat{b} where \hat{b} is b orthogonalized to the n(A^T).
Krylov solvers can produce the minimal norm solution to the least squares problem by utilizing MatNullSpaceRemove().
If the matrix is known to be symmetric because it is an SBAIJ matrix or one as called MatSetOption(mat,MAT_SYMMETRIC or MAT_SYMMETRIC_ETERNAL,PETSC_TRUE); this routine also automatically calls MatSetTransposeNullSpace().