Time Step Bias Improvement in Fixed-Node Diffusion Monte Carlo : a Size Consistent and System Independent Solution

Simulations of realistic materials with fixed-node diffusion Monte Carlo (DMC) are becoming appealing for its comparable accuracy with high level quantum chemistry methods and much less computational cost for large systems. In this method, ground state is accessed by applying a series of short time propagators which introduce a bias. The usual way to reduce this bias is by performing calculations at a few time steps and extrapolate the measured properties to zero time step. In practice, properties like binding energy can be made less sensitive to this bias if size consistency is ensured in the method and the expensive calculations for extrapolation can be avoided. The work by A. Zen et al. [1] introduced a scheme to improve the size consistency of the branching term in the DMC algorithm and our work further improves it by making it system independent. In addition, we also study the size consistency of DMC when pseudo-potentials and T-moves [2] are used.
[1] A. Zen et al. Physical Review B 93, 241118(R) (2016)
[2] M. Casula et al. The Journal of Chemical Physics 132, 154113 (2010)