Quantum Monte Carlo Methods Describe Noncovalent Interactions with Subchemical Accuracy.

An accurate description of noncovalent interaction energies is one of the most challenging tasks in computational chemistry. To date, nonempirical CCSD(T)/CBS has been used as a benchmark reference. However, its practical use is limited due to the rapid growth of its computational cost with the system complexity. Here, we show that the fixed-node diffusion Monte Carlo (FN-DMC) method with a more favorable scaling is capable of reaching the CCSD(T)/CBS within subchemical accuracy (<0.1 kcal/mol) on a testing set of six small noncovalent complexes including the water dimer. In benzene/water, benzene/methane, and the T-shape benzene dimer, FN-DMC provides interaction energies that agree within 0.25 kcal/mol with the best available CCSD(T)/CBS estimates. The demonstrated predictive power of FN-DMC therefore provides new opportunities for studies of the vast and important class of medium/large noncovalent complexes.