Nuclear Quantum Effects on the Dielectric Properties of Water from Machine-Learning-Accelerated Path-Integral Molecular Dynamics

Water’s dielectric response plays a fundamental role in numerous physical, chemical, and biological processes, yet accurately predicting its dielectric constant from first principles remains a longstanding challenge. A major source of difficulty arises from nuclear quantum effects (NQEs), including zero-point motion and proton delocalization. In this work, we employ machine-learning-accelerated path-integral molecular dynamics based on the hybrid SCAN0 functional to systematically study the impact of NQEs on water’s dielectric properties. We demonstrate that a methodology using the second-order rotationally invariant expansion of the dipolar pair-correlation function achieves substantially improved computational efficiency relative to the conventional dipole-fluctuation approach. To analyze the isotope effect in dielectric permittivity between H₂O and D₂O, we decompose the total dielectric response into contributions from the electronic structure and from the molecular organization of the hydrogen-bond network. Our results show that NQEs enhance the molecular polarizability while the collective dipolar correlations undergo a more delicate competing effect under NQEs.