Many-body interactions and deep neural network potentials for water

Francesco Paesani,Yaoguang Zhai,Richa Rashmi,Etienne Palos

DOI: https://doi.org/10.26434/chemrxiv-2024-sm0gd

2024-03-21

Abstract:We present a detailed assessment of deep neural network potentials developed within the DeePMD framework and trained on the MB-pol data-driven many-body potential energy function. Specific focus is directed at the ability of DeePMD-based potentials to correctly reproduce the accuracy of MB-pol across various water systems. Analyses of bulk and interfacial properties as well as many-body interactions characteristic of water elucidate inherent limitations in the transferability and predictive accuracy of DeePMD-based potentials. These limitations can be traced back to an incomplete implementation of the “nearsightedness of electronic matter” principle, which may be common throughout machine learning potentials that do not include a proper representation of self-consistently determined long-range electric fields. These findings provide further support for the "short-blanket dilemma" faced by DeePMD-based potentials, highlighting the challenges in achieving a balance between computational efficiency and a rigorous, physics-based representation of the properties of water. Finally, we believe that our study contributes to the ongoing discourse on the de- velopment and application of machine learning models in simulating water systems, offering insights that could guide future improvements in the field.

Chemistry

What problem does this paper attempt to address?

This paper mainly explores the limitations of Deep Neural Network Potentials (DeePMD) in simulating the many-body interactions of water. The research focuses on the potential developed within the DeePMD framework and trained based on the MB-pol data-driven many-body potential function. Although DeePMD performs well in simulating various water systems, the analysis shows that it has limitations in reproducing the accuracy of MB-pol, especially in describing the many-body interactions of water. The paper suggests that these limitations may arise from the incomplete implementation of the "electronic matter's locality principle," which is common in machine learning potentials, especially when proper representations of self-consistent long-range electric fields are not considered. This results in the so-called "short-blanket dilemma" for DeePMD potentials, where they cannot accurately reproduce the properties ranging from clusters to liquid water and vapor-liquid coexistence. The study also compares the performances of different DeePMD potentials and finds that regardless of the training set, they all exhibit the "short-blanket dilemma" proposed by Zhai et al. These potentials are considered to be "mimics" rather than "real" representations of reference models, so caution is needed when predicting thermodynamic states outside the training set as their reliability cannot be predetermined. Despite the limited predictive ability of DeePMD potentials, they demonstrate excellent computational efficiency, enabling them to be used for simulating water in thermodynamic states well-represented by reference models. This may allow for more extensive simulations of larger water systems.

Many-body interactions and deep neural network potentials for water

Many-body interactions and deep neural network potentials for water

A "short blanket" dilemma for a state-of-the-art neural network potential for water: Reproducing experimental properties or the physics of the underlying many-body interactions?

DeePMD-kit: A deep learning package for many-body potential energy representation and molecular dynamics

Deep Potential Molecular Dynamics: a scalable model with the accuracy of quantum mechanics

NEP-MB-pol: A unified machine-learned framework for fast and accurate prediction of water's thermodynamic and transport properties

Comparison of permutationally invariant polynomials, neural networks, and Gaussian approximation potentials in representing water interactions through many-body expansions

Condensed Phase Water Molecular Multipole Moments from Deep Neural Network Models Trained on Ab Initio Simulation Data

A deep potential model with long-range electrostatic interactions

Transferability evaluation of the deep potential model for simulating water-graphene confined system

Adaptive coupling of a deep neural network potential to a classical force field

Enhanced Deep Potential Model for Fast and Accurate Molecular Dynamics; Application to the Hydrated Electron

DeePMD-kit v2: A software package for deep potential models

A Transferrable Range-Separated Force Field for Water: Combining the Power of Both Physically-Motivated Models and Machine Learning Techniques

Enhanced deep potential model for fast and accurate molecular dynamics: application to the hydrated electron

Learning DeePMD-Kit: A Guide to Building Deep Potential Models

Comparative Study of the Water Response to External Force at Nanoscale and Mesoscale

Toward High-level Machine Learning Potential for Water Based on Quantum Fragmentation and Neural Networks