On the fly nodal searches in importance sampled fixed-node diffusion Monte Carlo using a parallel, fine-grained, genetic algorithm

A method for finding nodal surfaces on the fly in importance-sampled, fixed-node diffusion Monte Carlo calculations is described. The procedure relies on minimizing the difference between the nodal functions of the guiding wave function, psi T, and (H) over cap psi(T), where (H) over cap is the Hamiltonian. This is done by allowing the trial function to depend on a set of parameters whose values are then optimized using a parallel genetic algorithm (e.g., the Pikaia code developed in astrophysics). Application is made to the calculation of several excited states of a non-integrable two-dimensional quartic oscillator and to excited states of the He-C2H2 complex. (C) 2014 Elsevier B.V. All rights reserved.