Probing nuclear quantum effects in HCl clusters with high accuracy machine learning potentials

Abstract

Nuclear quantum effects (NQEs) in hydrogen chloride (HCl) clusters up to the trimer are investigated using a combination of robust path integral simulations and highly accurate machine-learning potential energy surfaces (PESs). Fundamental-invariant neural network PESs, which inherently enforce permutational symmetry, are constructed from over 110 000 CCSD(T)-F12a/AVTZ level data points. Intra- and inter-molecular interactions are described using the many-body expansion approach, achieving remarkable accuracy for the 1-body, 2-body, and 3-body interactions, with root-mean-square errors of 0.003, 0.021, and 0.094 meV, respectively. Our PESs reveal new configurations of the HCl trimer, enhancing our understanding of its structural landscape and interconversion pathways. Employing our neural network potentials, path integral molecular dynamics simulations reveal that NQEs weaken the binding of HCl clusters at temperatures of 100 K and below, significantly destabilizing the trimer (by 50 meV) at 30 K. Ground state tunneling splitting of (H35Cl)2 is computed with the recently developed Eckart spring path-integral molecular dynamics method, agreeing with the experimental value within 10% difference. Our study provides quantitative insights into the influences of NQEs in these hydrogen-bonded clusters. In addition, the highly accurate many-body potentials developed herein lay the groundwork for future studies of HCl in the condensed phase.