Development of composite control-variate stratified sampling approach for efficient stochastic calculation of molecular integrals

TL;DR

This paper introduces the CCSS method, a Monte Carlo-based approach for directly calculating molecular integrals without AO transformation, improving efficiency for large systems and broad applications.

Contribution

The paper presents the novel CCSS method that combines stratified sampling, control variates, and direct integration for efficient MO integral calculation without AO transformation.

Findings

CCSS reduces computational cost for large molecular systems.

It achieves variance reduction through control variates.

Applied successfully to excitonic properties in quantum dots.

Abstract

In this work, the composite control-variate stratified sampling (CCSS) method is presented for calculation of MO integrals without transformation of AO integrals. The central idea of this approach is to obtain the 2-electron MO integrals by direct integration of 2-electron coordinates. This method does not require or use pre-computed AO integrals and the value of the MOs at any point in space is obtained directly from the linear combination of AOs. The integration over the electronic coordinates was performed using stratified sampling Monte Carlo method. This approach was implemented by dividing the integration region into a set of non-overlapping segments and performing Monte Carlo calculations on each segment. The Monte Carlo sampling points for each segment were optimized to minimize the total variance of the sample mean. Additional variance reduction of the overall calculations was achieved by introducing control-variate in the stratified sampling scheme. The composite aspect of the CCSS allows for simultaneous computation of multiple MO integrals during the stratified sampling evaluation. The main advantage of the CCSS method is that unlike rejection sampling Monte Carlo methods such as Metropolis algorithm, the stratified sampling uses all instances of the calculated functions for the evaluation of the sample mean. The CCSS method is designed to be used for large systems where AO-to-MO transformation is computationally prohibitive. Because it is based on numerical integration, the CCSS method can be applied to a wide variety of integration kernels and does not require \textit{a priori} knowledge of analytical integrals. In this work, the developed CCSS method was applied for calculation of excitonic properties in CdSe quantum dots using electron-hole explicitly correlated Hartree-Fock (eh-XCHF) and geminal-screened electron-hole interaction kernel (GSIK) methods.