Abstract:Machine learning potentials (MLPs) are poised to combine the accuracy of ab initio predictions with the computational efficiency of classical molecular dynamics (MD) simulation. While great progress has been made over the last two decades in developing MLPs, there is still much to be done to evaluate their model transferability and facilitate their development. In this work, we construct two deep potential (DP) models for liquid water near graphene surfaces, Model S and Model F, with the latter having more training data. A concurrent learning algorithm (DP-GEN) is adopted to explore the configurational space beyond the scope of conventional ab initio MD simulation. By examining the performance of Model S, we find that an accurate prediction of atomic force does not imply an accurate prediction of system energy. The deviation from the relative atomic force alone is insufficient to assess the accuracy of the DP models. Based on the performance of Model F, we propose that the relative magnitude of the model deviation and the corresponding root-mean-square error of the original test dataset, including energy and atomic force, can serve as an indicator for evaluating the accuracy of the model prediction for a given structure, which is particularly applicable for large systems where density functional theory calculations are infeasible. In addition to the prediction accuracy of the model described above, we also briefly discuss simulation stability and its relationship to the former. Both are important aspects in assessing the transferability of the MLP model.

What problem does this paper attempt to address?

This paper explores the transferability assessment problem of the Deep Potential (DP) model in simulating water-graphene confinement systems. Two DP models, Model S and Model F, were constructed to investigate their performance in predicting the behavior of liquid water near the graphene surface. The researchers adopted the concurrent learning algorithm DP-GEN to explore the configuration space beyond the scope of traditional ab initio MD simulations. The paper points out that although the DP model can combine accurate quantum mechanical predictions with the efficiency of classical molecular dynamics simulations, the evaluation and development of its model transferability still need to be strengthened. Through the analysis of Model S, they found that accurate prediction of atomic forces does not necessarily mean accurate prediction of system energy. Deviations based solely on relative atomic forces are not sufficient to evaluate the accuracy of the DP model. With Model F, the researchers propose a metric for evaluating the accuracy of model prediction structures, which is the relative magnitude of model deviation to the root mean square error of the original test dataset (including energy and atomic forces). This is particularly applicable to large-scale systems that cannot afford expensive density functional theory calculations. In addition, the paper discusses the relationship between simulation stability and prediction accuracy, both of which are important aspects in evaluating the transferability of the MLP model. The researchers validated these findings through simulations in different systems and temperatures, and explored issues related to simulation stability. Overall, the paper attempts to address the problem of how to effectively evaluate and improve the prediction accuracy and transferability of deep potential models in simulating complex systems such as water-graphene interfaces.

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

Transferable Performance of Machine Learning Potentials Across Graphene-Water Systems of Different Sizes: Insights from Numerical Metrics and Physical Characteristics.

Interfacial Transport Across Graphene-Water from Machine Learning Molecular Dynamics

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

A deep potential model with long-range electrostatic interactions

Many-body interactions and deep neural network potentials for water

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

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?

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

Transferable Water Potentials Using Equivariant Neural Networks

Deep Potential Molecular Dynamics: A Scalable Model with the Accuracy of Quantum Mechanics

Data-Efficient Construction of High-Fidelity Graph Deep Learning Interatomic Potentials

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

Reliable machine learning potentials based on artificial neural network for graphene

Perspective: Atomistic Simulations of Water and Aqueous Systems with Machine Learning Potentials

Discrepancies and Error Evaluation Metrics for Machine Learning Interatomic Potentials

Extending the atomic decomposition and many-body representation, a chemistry-motivated monomer-centered approach for machine learning potentials

Evaluating the Transferability of Machine-Learned Force Fields for Material Property Modeling