Abstract:In molecular simulations, neural network force fields aim at achieving \emph{ab initio} accuracy with reduced computational cost. This work introduces enhancements to the Deep Potential network architecture, integrating a message-passing framework and a new lightweight implementation with various improvements. Our model achieves accuracy on par with leading machine learning force fields and offers significant speed advantages, making it well-suited for large-scale, accuracy-sensitive systems. We also introduce a new iterative model for Wannier center prediction, allowing us to keep track of electron positions in simulations of general insulating systems. We apply our model to study the solvated electron in bulk water, an ostensibly simple system that is actually quite challenging to represent with neural networks. Our trained model is not only accurate, but can also transfer to larger systems. Our simulation confirms the cavity model, where the electron's localized state is observed to be stable. Through an extensive run, we accurately determine various structural and dynamical properties of the solvated electron.

What problem does this paper attempt to address?

This paper focuses on improving the accuracy and speed of neural network potential fields (such as Deep Potential, DP) in molecular dynamics simulations. While existing methods can run with quantum mechanical precision, they are computationally expensive, and traditional empirical potential fields are fast but lack accuracy. The paper proposes enhancements to the DP model, including the integration of a message passing framework (DP-MP) and a lightweight implementation, to improve its expressive power and efficiency. The new model provides significant speed advantages while maintaining comparable accuracy to leading machine learning potentials, making it particularly suitable for large-scale, precision-sensitive systems. In addition, the paper introduces a new iterative model called Deep Wannier Iterative Refinement (DWIR) for predicting the Wannier center positions in insulating systems, addressing the issue faced by traditional Deep Wannier models in tracking non-unique atomic electron correlations. This approach allows for tracking the electron positions in atomic simulations, even when the electrons are not associated with specific atoms. The paper demonstrates the capabilities of the enhanced model by studying solvated electrons (e-(aq)). Solvated electrons play important roles in radiation chemistry and biology, and despite their seemingly simple nature, their structures are complex and challenging for neural network models. The authors successfully simulated solvated electrons using the DP-MP model and demonstrated the transferability of the model to larger systems. The simulation results confirmed the cavity model of electrons, showing that electrons form stable local states and revealing the mechanism of electron diffusion. Overall, the objective of the paper is to develop a faster and sufficiently accurate molecular dynamics simulation model by improving the Deep Potential model, introducing a message passing mechanism, and proposing a new method for predicting Wannier centers. These advancements enhance the efficiency and accuracy of the simulations.

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

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

A deep potential model with long-range electrostatic interactions

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

Molecular Dynamics Simulation of Zinc Ion in Water with an Ab Initio Based Neural Network Potential

Ab Initio Validation on the Connection Between Atomistic and Hydrodynamic Description to Unravel the Ion Dynamics of Warm Dense Matter

Accurate interatomic force field for molecular dynamics simulation by hybridizing classical and machine learning potentials

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

Large Scale and Quantum Accurate Molecular Dynamics Simulation:Liquid Iron under Extreme Condition

High-Dimensional Neural Network Potential for Liquid Electrolyte Simulations.

Bridging deep learning force fields and electronic structures with a physics-informed approach

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

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

Many-body interactions and deep neural network potentials for water

Neural Network Potential with Multi-Resolution Approach Enables Accurate Prediction of Reaction Free Energies in Solution

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

Calculation of solvation force in molecular dynamics simulation by deep-learning method

Simulating the ghost: quantum dynamics of the solvated electron

Thermodynamics of Water and Ice from a Fast and Scalable First-Principles Neuroevolution Potential

Deep neural network for Wannier function centers