Efficient Machine Learning Force Field for Large-Scale Molecular Simulations of Organic Systems

Aug 3, 2026

Upload by Ivy

Machine learning force fields address the costly trade-off between ab initio molecular dynamics accuracy and empirical force-field limitations, yet large-scale organic simulations remain hindered by long-range intermolecular complexity, diverse molecular conformations, and instability during long-time runs. A universal multiscale higher-order equivariant model with active learning is proposed to capture these interactions and conformations with superior predictive accuracy. A bond length stretching strategy further improves long-horizon simulation stability. Trained on 901 samples from a 120-atom dataset, the model extends high-precision performance to systems with hundreds of thousands of atoms while improving speed and memory efficiency.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages23

Reading Time58 min

AccessFree to download

Abstract

Molecular dynamics and machine learning force fields

Trade-off between AIMD and empirical force fields

Mapping molecular coordinates to quantum data

Review of MLFF representations and equivariance

Invariant vs equivariant graph neural networks

Higher-order equivariant models

Application gap in organic systems

Challenges in long-range interactions and long-time stability

arXiv :2312 .09490v1 [ cond-mat .soft] 15 Dec 2023 Efficient Machine Learning Force Field for Large-Scale Molecular Simulations of Organic Systems Junbao Hu1,2 , Liyang Zhou3*, Jian Jiang1,2* 1 Beijing National Laboratory for Molecular Sciences, State Key Laboratory of Polymer Physics and Chemistry, Institute of Chemistry, Chinese Academy of Sciences, Beijing, 100190, P. R. China. 2 University of Chinese Academy of Sciences, Beijing, 100049, P. R. China. 3 Juhua Group Co. , Ltd, Quzhou, 324004, P. R. China. *Corresponding author(s). E-mail(s): [fhzly@juhua.com](fhzly@juhua.com) ; [jiangj@iccas.ac.cn](jiangj@iccas.ac.cn) ; Abstract To address the computational challenges of ab initio molecular dynamics and the accuracy limitations of empirical force fields, the introduction of machine learning force fields has proven effective in various systems including metals and inorganic materials. However, in large-scale organic systems, the application of machine learning force fields is often hindered by impediments such as the complexity of long-range intermolecular interactions and molecular conformations, as well as the instability in long-time molecular simulations.Therefore, we propose a universal multiscale higher-order equivariant model combined with active learning techniques, efficiently capturing the complex long-range intermolecular interactions and molecular conformations. Compared to existing equivariant models, our model achieves the highest predictive accuracy, and magnitude-level improvements in computational speed and memory efficiency. In addition, a bond length stretching method is designed to improve the stability of long-time molecular simulations. Utilizing only 901 samples from a dataset with 120 atoms, our model successfully extends high precision to systems with hundreds of thousands of atoms. These achievements guarantee high predictive accuracy, fast simulation speed, minimal memory consumption, and robust simulation stability, satisfying the requirements for high-precision and long-time molecular simulations in large-scale organic systems. Main Molecular dynamics (MD) simulation has gained significant attention in recent years across various disciplines, spanning physics, chemistry, biology, and materials science. This cutting-edge technology, by simulating interactions between molecules or atoms, offers researchers a means to investigate the microstructure of substances and understand their macroscopic properties. This technique has proved invaluable for experimental design, development of new materials, and advances in biomedical research [1–4] . Traditional molecular simulations grapple with the dilemma of balancing the high computational cost of ab initio molecular dynamics (AIMD) against the low precision of empirical force fields. A resolution to this challenge is found in the application of machine learning, leveraging its powerful fitting capabilities. The fundamental idea of machine learning force field (MLFF) is to establish a mapping from molecular coordinates to the labels of high-precision quantum chemistry data, including potential energy and forces. This approach eliminates the need to solve the intricate Schr¨odinger equation, resulting in a significant acceleration and achieving a balance between prediction precision and simulation speed [5] . Since the advent of BPNN in 2007[6], numerous MLFF models have been proposed to improve prediction accuracy, and their performance has been systematically investigated in public datasets (MD17[7], MD22[8], OC22[9]) . From the perspective of tensor order (denoted by l), existing 3D molecular representation learning models can be categorized into two main classes: one is the invariant graph neural networks with only scalar features (i.e. , l = 0) , including SchNet[10], DeePMD[11], DTNN[12], PhysNet[13], ComENet[14], SphereNet[15]; the other is the equivariant graph neural networks with vector features (i.e. , l = 1) includin

FAQs

What problem does the document address in large-scale organic molecular simulations?

It targets the difficulty of applying machine learning force fields to large-scale organic systems due to complex long-range intermolecular interactions, complicated molecular conformations, and instability during long-time simulations.

What is the proposed model, and how does it improve prediction quality?

The document proposes a universal multiscale higher-order equivariant model combined with active learning. It captures long-range interactions and conformations efficiently and achieves the highest predictive accuracy compared with existing equivariant models.

How does the work improve stability for long-time simulations?

It introduces a bond length stretching method designed to enhance the stability of long-time molecular simulations.

Aug 3, 2026

Upload by Ivy