topic Re: Initial Vector in Krylov-Schur diagonalization in mkl sparse eigensolver in Intel® oneAPI Math Kernel Library

Initial Vector in Krylov-Schur diagonalization in mkl sparse eigensolver

VinceNeede — Sun, 14 May 2023 08:44:39 GMT

The Krylov-Schur method requires a starting vector for generating the Krylov Subspaces, It is usually selected at random since some vectors can happen to be in an invariant subspace and terminate the algorithm before, but since the probability to obtain them in a random choice is zero, there is no problem.

The point is that for the Density Matrix Renormalization Group (DMRG) that I'm implementing, has a first guess for the vector in order to fasten the convergence, but from the documentation of the mkl_sparse_d_ev routine is not clear how to set such vector, is there a way to do so?

Re: Initial Vector in Krylov-Schur diagonalization in mkl sparse eigensolver

andrew_4619 — Mon, 15 May 2023 13:42:51 GMT

You might be better asking this in the MKL forum rather than the Fortran forum

Re: Initial Vector in Krylov-Schur diagonalization in mkl sparse eigensolver

Barbara_P_Intel — Mon, 15 May 2023 15:07:56 GMT

I'll move this thread to the Intel MKL Forum.