-ksp_cg_type Hermitian | - (for complex matrices only) indicates the matrix is Hermitian, see KSPCGSetType() | |
-ksp_cg_type symmetric | - (for complex matrices only) indicates the matrix is symmetric | |
-ksp_cg_single_reduction | - performs both inner products needed in the algorithm with a single MPIU_Allreduce() call, see KSPCGUseSingleReduction() |
Notes: The PCG method requires both the matrix and preconditioner to be symmetric positive (or negative) (semi) definite.
Only left preconditioning is supported; there are several ways to motivate preconditioned CG, but they all produce the same algorithm.
One can interpret preconditioning A with B to mean any of the following:
(1) Solve a left-preconditioned system BAx = Bb, using inv(B) to define an inner product in the algorithm.
(2) Solve a right-preconditioned system ABy = b, x = By, using B to define an inner product in the algorithm.
(3) Solve a symmetrically-preconditioned system, E^TAEy = E^Tb, x = Ey, where B = EE^T.
(4) Solve Ax=b with CG, but use the inner product defined by B to define the method [2].
In all cases, the resulting algorithm only requires application of B to vectors.
For complex numbers there are two different CG methods, one for Hermitian symmetric matrices and one for non-Hermitian symmetric matrices. Use KSPCGSetType() to indicate which type you are using.
Developer Notes: KSPSolve_CG() should actually query the matrix to determine if it is Hermitian symmetric or not and NOT require the user to indicate it to the KSP object.