Direct Optimization of Quasi-2D Atomic Structures with Diffusion Monte Carlo

Quasi-2D materials display great optical sensitivity to lattice strain, simultaneously with large lattice flexibility. Predictive characterization of the optical properties of these materials therefore requires precise determination of lattice parameters. We present a novel parallel line search method to simultaneously determine multiple structural parameters of 2D materials with sparse sampling of the diffusion Monte Carlo (DMC) potential energy surface (PES). The method relies on density functional calculations as a surrogate to guide the search along the most rapidly converging directions in parameter space. Resampling techniques are used on the surrogate model to predict and further minimize the computational cost of the DMC PES search. We present examples of the method as applied to 2D GeSe and flake-like 2D molecules.