Abstract:Machine learning potentials (MLPs) trained on data from quantum-mechanics based first-principles methods can approach the accuracy of the reference method at a fraction of the computational cost. To facilitate efficient MLP-based molecular dynamics (MD) and Monte Carlo (MC) simulations, an integration of the MLPs with sampling software is needed. Here we develop two interfaces that link the Atomic Energy Network (ænet) MLP package with the popular sampling packages TINKER and LAMMPS. The three packages, ænet, TINKER, and LAMMPS, are free and open-source software that enable, in combination, accurate simulations of large and complex systems with low computational cost that scales linearly with the number of atoms. Scaling tests show that the parallel efficiency of the ænet-TINKER interface is nearly optimal but is limited to shared-memory systems. The ænet-LAMMPS interface achieves excellent parallel efficiency on highly parallel distributed memory systems and benefits from the highly optimized neighbor list implemented in LAMMPS. We demonstrate the utility of the two MLP interfaces for two relevant example applications, the investigation of diffusion phenomena in liquid water and the equilibration of nanostructured amorphous battery materials.

What problem does this paper attempt to address?

### Problems the Paper Aims to Solve The paper aims to address the issues of efficiency and accuracy in modeling complex material systems under experimental conditions. Specifically, the authors have developed two interfaces that integrate machine learning potentials (MLPs) based on the atomic energy network (ænet) with popular molecular dynamics (MD) and Monte Carlo (MC) simulation software TINKER and LAMMPS. These interfaces enable efficient large-scale and complex system molecular dynamics simulations while maintaining high accuracy. ### Main Issues and Challenges 1. **Large-scale Simulation of Complex Materials**: - **Lithium Transport in Nanostructured Amorphous Silicon**: Nanostructured silicon is a promising anode material for high-energy-density lithium batteries. However, studying lithium transport in silicon electrodes is challenging due to its complex amorphous structure, scale-dependent mechanical properties, and low lithium conductivity. - **Diffusion Phenomena in Liquid Water**: Studying proton defect transport in liquid water requires long simulations because this process involves alternating between fast-moving and stationary states, requiring several nanoseconds to reliably estimate the diffusion coefficient. 2. **Computational Efficiency and Accuracy**: - **Limitations of Traditional Methods**: First-principles methods (such as DFT) are accurate but computationally expensive, while traditional molecular force field methods are computationally efficient but lack accuracy. - **Advantages of MLPs**: Machine learning potentials can achieve near-reference method accuracy at a lower computational cost by learning atomic interactions from quantum mechanical calculations. ### Solutions 1. **ænet-TINKER Interface**: - Suitable for shared memory systems with approximately optimal parallel efficiency. - Suitable for small-scale simulations on desktops or laptops. 2. **ænet-LAMMPS Interface**: - Suitable for highly parallel distributed memory systems with excellent parallel efficiency. - Utilizes the highly optimized neighbor list in LAMMPS, suitable for large-scale system simulations. ### Experimental Validation - **Application Examples**: - **Lithium Transport in Nanostructured Amorphous Silicon**: Demonstrated different crystalline and quasi-amorphous LiSi structures and variations in lithium content. - **Proton Defect Diffusion in Liquid Water**: Showed the diffusion process of proton defects through long simulations, including alternating fast-moving and stationary states. - **Performance Testing**: - Tested computation time for systems of different scales, showing that computational cost increases linearly with system size. - Parallel efficiency tests showed that ænet-LAMMPS maintains high parallel efficiency when using up to 96 CPU cores across multiple nodes. ### Conclusion By developing the ænet-TINKER and ænet-LAMMPS interfaces, the paper successfully addresses the problem of efficient and accurate simulation of complex material systems under experimental conditions. These interfaces not only generate long trajectories but also efficiently utilize computational resources on modern hardware, providing powerful tools for reliable prediction of molecular and material properties.

AENET-LAMMPS and AENET-TINKER: Interfaces for Accurate and Efficient Molecular Dynamics Simulations with Machine Learning Potentials

Amber Free Energy Tools: Interoperable Software for Free Energy Simulations Using Generalized Quantum Mechanical/molecular Mechanical and Machine Learning Potentials.

NNP/MM: Accelerating molecular dynamics simulations with machine learning potentials and molecular mechanic

NNP/MM: Accelerating Molecular Dynamics Simulations with Machine Learning Potentials and Molecular Mechanics

Accelerating Metadynamics-Based Free-Energy Calculations with Adaptive Machine Learning Potentials.

PiNNAcLe: Adaptive Learn-On-The-Fly Algorithm for Machine-Learning Potential

Machine Learning for Parameter Auto-tuning in Molecular Dynamics Simulations: Efficient Dynamics of Ions near Polarizable Nanoparticles

LAMMPS Implementation of Rapid Artificial Neural Network Derived Interatomic Potentials

Accurate and efficient molecular dynamics based on machine learning and non von Neumann architecture

Accurate Free Energy Calculation via Multiscale Simulations Driven by Hybrid Machine Learning and Molecular Mechanics Potentials

Learning Interatomic Potentials at Multiple Scales

ANI Neural Networks Meet Electrostatics: A ML/MM Implementation in Amber

DeePKS + ABACUS as a Bridge between Expensive Quantum Mechanical Models and Machine Learning Potentials

Accelerated Quantum Mechanics/Molecular Mechanics Simulations Via Neural Networks Incorporated with Mechanical Embedding Scheme.

Developing General Reactive Element-Based Machine Learning Potentials as the Main Computational Engine for Heterogeneous Catalysis

Improving Molecular Dynamics Simulations for Solid‐Liquid Interface with Machine Learning Interatomic Potentials

DeePKS+ABACUS as a Bridge between Expensive Quantum Mechanical Models and Machine Learning Potentials

Taming Multi-Domain, -Fidelity Data: Towards Foundation Models for Atomistic Scale Simulations

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

Refining Potential Energy Surface through Dynamical Properties via Differentiable Molecular Simulation

Development of Machine Learning Atomistic Potential for Molecular Simulation of Hematite–Water Interfaces