Quantum Monte Carlo for Noncovalent Interactions: An Efficient Protocol Attaining Benchmark Accuracy

Abstract: Reliable theoretical predictions of noncovalent interaction energies, which are important e.g. in drug-design and hydrogen-storage applications, belong to longstanding challenges of contemporary quantum chemistry. In this respect, the fixed-node diffusion Monte Carlo (FN-DMC) is a promising alternative to the commonly used "gold standard" coupled-cluster CCSD(T)/CBS method for its benchmark accuracy and favourable scaling, in contrast to other correlated wave function approaches. This work is focused on the analysis of protocols and possible tradeoffs for FN-DMC estimations of noncovalent interaction energies and proposes an efficient yet accurate computational protocol using simplified explicit correlation terms with a favorable O(N^3) scaling. It achieves an excellent agreement (mean unsigned error ~0.2 kcal/mol) with respect to the CCSD(T)/CBS data on a number of complexes, including benzene/hydrogen,T-shape benzene dimer, stacked adenine-thymine and a set of small noncovalent complexes A24. The high accuracy and reduced computational costs predestinate the reported protocol for practical interaction energy calculations of large noncovalent complexes, where the CCSD(T)/CBS is prohibitively expensive.