An Amplitude-Based Importance Sampling Algorithm for Wavefunction Optimization in Diffusion Monte Carlo

Efficient and robust wavefunction optimization algorithms are key to the success of diffusion Monte Carlo methods, as they play a crucial role in improving the trial wavefunction. An accurate trial wavefunction is essential as it determines both the variance of the local energy and the nodal (or phase) structure. A small variance is vital for reducing statistical error, while the proper nodal or phase structure minimizes systematic errors. To our knowledge, all current methods are based on an importance sampling probability density defined by the square of the trial wavefunction's amplitude; accordingly, configurations are biased away from the node. Although this procedure is very effective for evaluating the energy, it might be inefficient for optimizing the node. In this talk, we discuss an amplitude-based importance sampling algorithm that, when combined with the self-healing algorithm, efficiently optimizes a multideterminant wavefunction for second-row atoms. We compare our results with those obtained via stochastic reconfigurations and standard SHDMC.