[Paper Review] Isotope effects on molecular structures and electronic properties of liquid water via deep potential molecular dynamics based on SCAN functional

Jianhang Xu, Chunyi Zhang+4 more|arXiv (Cornell University)|Dec 9, 2020

Quantum, superfluid, helium dynamics2 references4 citations

TL;DR

This study combines deep potential molecular dynamics (DPMD) with the SCAN exchange-correlation functional and Feynman path-integral (PI) methods to model nuclear quantum effects (NQEs) in liquid water. It demonstrates that PI-DPMD based on SCAN accurately reproduces isotope effects in radial and angular distribution functions, hydrogen-bond structure, and electronic properties—showing lighter H2O exhibits stronger NQEs than D2O, with dipole moments increasing under quantum effects.

ABSTRACT

Feynman path-integral deep potential molecular dynamics (PI-DPMD) calculations have been employed to study both light (H$_2$O) and heavy water (D$_2$O) within the isothermal-isobaric ensemble. In particular, the deep neural network is trained based on ab initio data obtained from the strongly constrained and appropriately normed (SCAN) exchange-correlation functional. Because of the lighter mass of hydrogen than deuteron, the properties of light water is more influenced by nuclear quantum effect than those of heavy water. Clear isotope effects are observed and analyzed in terms of hydrogen-bond structure and electronic properties of water that are closely associated with experimental observables. The molecular structures of both liquid H$_2$O and D$_2$O agree well with the data extracted from scattering experiments. The delicate isotope effects on radial distribution functions and angular distribution functions are well reproduced as well. Our approach demonstrates that deep neural network combined with SCAN functional based ab initio molecular dynamics provides an accurate theoretical tool for modeling water and its isotope effects.

Motivation & Objective

• To investigate nuclear quantum effects (NQEs) on molecular and electronic structures of liquid water using advanced ab initio methods.
• To address limitations of standard DFT functionals in describing hydrogen bonding and electron delocalization in water.
• To evaluate isotope effects between H2O and D2O in radial and angular distribution functions, hydrogen-bond networks, and electronic properties.
• To validate the accuracy of deep potential molecular dynamics (DPMD) trained on SCAN functional data against experimental scattering and spectroscopic data.
• To demonstrate the feasibility and accuracy of combining machine learning (DPMD), high-level DFT (SCAN), and quantum nuclear treatment (PI) for simulating liquid water.

Proposed method

• Employed deep neural networks to construct a machine-learned potential energy surface (PES) trained on ab initio data from the SCAN meta-GGA functional.
• Applied path-integral molecular dynamics (PI-MD) to include nuclear quantum effects (NQEs) in both H2O and D2O simulations.
• Performed simulations in the isothermal–isobaric (NpT) ensemble at ambient conditions using the deep potential model trained on SCAN data.
• Calculated radial distribution functions (RDFs), angular distribution functions (ADFs), and dipole moment distributions to probe structural and electronic properties.
• Used maximally localized Wannier functions (MLWFs) to compute electronic properties such as dipole moments and charge distributions.
• Compared classical DPMD and PI-DPMD results to assess the impact of NQEs on O–H bond elongation, H-bond breaking, and electron delocalization.

Experimental results

Research questions

• RQ1 How do nuclear quantum effects (NQEs) influence the hydrogen-bond network structure in liquid H2O and D2O?
• RQ2 To what extent does the SCAN functional improve the description of water’s radial and angular distribution functions compared to standard GGA functionals?
• RQ3 How do isotope effects manifest in the O–O–O angular distribution and radial distribution functions in liquid water?
• RQ4 What is the impact of NQEs on the electronic structure of water, particularly in terms of O–H bond elongation and dipole moment distribution?
• RQ5 Can deep potential molecular dynamics based on SCAN functional accurately reproduce experimental scattering and spectroscopic data for liquid water?

Key findings

• PI-DPMD simulations based on the SCAN functional reproduce experimental radial distribution functions (RDFs) and angular distribution functions (ADFs) for both H2O and D2O with high accuracy.
• The average dipole moment in H2O increases to 3.17 ± 0.05 D under path-integral treatment, reflecting enhanced polarization due to NQEs, compared to 3.01 ± 0.03 D in classical DPMD.
• Nuclear quantum effects lead to increased O–H bond elongation and greater asymmetry in the bonding pair distribution, with a shift in the bonding peak from 0.492 Å to a broader, asymmetric profile.
• The lone pair electron distribution shifts slightly outward, from 0.328 Å in DPMD to 0.325 Å in H2O-PI-DPMD, indicating enhanced H-bond facilitation under NQEs.
• A fraction of water molecules exhibit dipole moments >3.5 D in PI-DPMD, indicating significant proton delocalization and enhanced H-bond network fluctuations under NQEs.
• The predicted isotope effects—softer structure and enhanced H-bond breaking in H2O compared to D2O—are consistent with experimental observations of isotope-dependent RDFs and ADFs.