Accelerating convergence in iterative solution for large-scale complete active space self-consistent-field calculations

Abstract

An algorithm that accelerates the convergence of the iterative optimization of the complete active space self-consistent field (CASSCF) wavefunction so as to find a optimum solution in fewer macroiterations is described. The algorithm is oriented to large-scale CASSCF problems that are to be solved with a combination of density matrix renormalization group (DMRG) method for the configuration interaction (CI) process. The algorithm is based on the alternating (or two-step) CASSCF optimization in which the CI and molecular orbital (MO) parameters are optimized separately. Convergence ratio is improved by finding further optimized MOs from a linear extrapolation of the MO sets of the iteration history. The acceleration results in fewer diagonalizations in a total CASSCF calculation to save a considerable computational cost. The convergence performance is examined in a couple of realistic applications on SiC3 and poly(phenyl)carbenes. For poly(phenyl)carbenes, the large-size CASSCF calculations with CAS(30e,30o) that entails full π valence space as well as sp2 orbital space of carbenes are performed by using the practical implementation of DMRG-CASSCF in conjunction with the acceleration technique.

Additional details