Sparse Iterative Solver cgna and cgnr

[Petros_Mamales]

Hi,

As emphasized in the mkl reference, the conjugate gradient sparse iterative solver is for positive symmetric matrices (call them A).
However there are two variations, that utilize A*A^t or A^t*A, that allow the usage of cg for arbitrary A.
Is there a way to fool the current cg to do it ? (the obvious way of forming the products before employing cg,
kills the sparsity, of course ).
Thank you in advance for your help,
Petros

[Alexander_K_Intel2]

Hi,

MKL CG doesn't support multiplication of A on A^t. Moreover, from my point of view, it's not a good idea to use CG for solving system like A^tAx=A^tb instead of Ax=b because number of CG iteration depend on condition number of stiffness matrix. In your case condition number of stiffness matrix is squaredcondition number of matrix A. If you matrix is not symmetric try to use fgmres solver from MKL

With best regards,

Alexander Kalinkin