Improving the accuracy of fixed-node diffusion Monte Carlo for non-covalent interactions

Diffusion quantum Monte Carlo (DMC) and coupled cluster theory [CCSD(T)] are widely used benchmark methods for noncovalent interactions (NCIs). However, recent studies have reported notable discrepancies for several hydrogen-bonded and dispersion-dominated systems, raising questions about the accuracy of the approximations underlying each approach. In DMC, the dominant residual error is expected to stem from the fixed-node (FN) approximation, where the nodal surface is typically taken from a single Slater determinant (SD) derived from density functional theory or Hartree–Fock calculations. In this work, we assess the impact of nodal-surface optimization on DMC predictions for 12 compounds spanning diverse NCIs. In particular, we compare binding energies obtained with a recently proposed antisymmetrized geminal power Ansatz parameterized with natural orbitals (AGPn) and with the conventional SD ansatz. AGPn is a generalization of the SD Ansatz and, when variationally optimized at the DMC level, yields lower total energies. We find that for hydrogen-bonded systems, AGPn improves the agreement with CCSD(T), whereas its effect on dispersion-dominated systems is negligible within the statistical uncertainty. The obtained mean absolute deviations between CCSD(T) and FN-SD-DMC binding energies are 0.43(2) and 0.21(2) kcal/mol for the hydrogen-bonded and dispersion-dominated systems, respectively, and those between CCSD(T) and FN-AGPn-DMC binding energies are 0.18(2) and 0.28(3) kcal/mol for the hydrogen-bonded and dispersion-dominated systems, respectively. These results suggest that the reported discrepancies between DMC and CCSD(T) for hydrogen bonds originate primarily from the mean-field nodal surface of the SD ansatz, while the origin of the discrepancies for dispersion-dominated systems remains an open question.