Self-optimizing diffusion quantum Monte Carlo calculation: the singlet-triplet splitting in CH2

Abstract In this paper we propose a novel diffusion quantum Monte Carlo (DMC) algorithm, which is a self-optimizing and self-improving procedure. In contrast to the conventional processes, the trial function is optimized synchronistically in the diffusion procedure, but not before beginning the DMC computation. In order to optimize the trial function, an improved steepest descent technique is used, with the step size automatically adjusted to obtain a procedure which converges superlinearly. We also use a novel trial function, which has both the correct electron-electron and electron-nucleus cusp conditions. The method and the trial function are employed to calculate the singlet-triplet splitting in CH 2 .