Abstract:Despite the widespread applications of machine learning force field (MLFF) on solids and small molecules, there is a notable gap in applying MLFF to complex liquid electrolytes. In this work, we introduce BAMBOO (ByteDance AI Molecular Simulation Booster), a novel framework for molecular dynamics (MD) simulations, with a demonstration of its capabilities in the context of liquid electrolytes for lithium batteries. We design a physics-inspired graph equivariant transformer architecture as the backbone of BAMBOO to learn from quantum mechanical simulations. Additionally, we pioneer an ensemble knowledge distillation approach and apply it on MLFFs to improve the stability of MD simulations. Finally, we propose the density alignment algorithm to align BAMBOO with experimental measurements. BAMBOO demonstrates state-of-the-art accuracy in predicting key electrolyte properties such as density, viscosity, and ionic conductivity across various solvents and salt combinations. Our current model, trained on more than 15 chemical species, achieves the average density error of 0.01 g/cm$^3$ on various compositions compared with experimental data. Moreover, our model demonstrates transferability to molecules not included in the quantum mechanical dataset. We envision this work as paving the way to a "universal MLFF" capable of simulating properties of common organic liquids.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the challenges of applying machine - learning force fields (MLFF) in complex liquid electrolytes. Although the application of MLFF in solids and small molecules has been very extensive, there is still a significant gap in its application in liquid electrolytes. Due to the complex local structures in liquid electrolytes, such as the co - existence of solvent - separated ion pairs (SSIP), contact ion pairs (CIP) and aggregates (AGG), it is difficult for existing MLFFs to accurately predict various properties, such as density, viscosity and ionic conductivity. In addition, molecular dynamics (MD) simulations based on MLFF usually encounter qualitative and quantitative instability problems, which limit their practical applications. Therefore, the paper aims to develop a general and stable MLFF framework that can efficiently and accurately simulate various properties of liquid electrolytes and has good generalization ability to adapt to systems with different solvents and salt concentrations. Specifically, the paper proposes BAMBOO (ByteDance AIMolecular Simulation Booster), which is a new MLFF framework for molecular dynamics simulations of liquid electrolytes. The main contributions of BAMBOO include: 1. **Novel MLFF model**: It combines the graph - equivariant transformer (GET) architecture and a partitioning method based on physics - based semi - local, electrostatic and dispersion interactions, and can learn from quantum - mechanical simulations and model various types of molecules and salts. 2. **Integrated knowledge distillation algorithm**: By integrating multiple independently - trained GNN models to improve the stability of MD simulations and reduce prediction deviations caused by the inherent randomness of machine learning. 3. **Density alignment algorithm**: Align the MLFF with experimental data to ensure the consistency between simulation results and experimental measurements while maintaining the generalization ability of the model. These methods together improve the accuracy and stability of BAMBOO in predicting properties of liquid electrolytes such as density, viscosity and ionic conductivity, making it an important tool for studying the design of new electrolytes.

BAMBOO: a predictive and transferable machine learning force field framework for liquid electrolyte development

Machine learning force fields for molecular liquids: Ethylene Carbonate/Ethyl Methyl Carbonate binary solvent

High-Dimensional Neural Network Potential for Liquid Electrolyte Simulations.

Scalable molecular simulation of electrolyte solutions with quantum chemical accuracy

Electron Density Learning of Z-Bonds in Ionic Liquids and Its Application

Liquid to Crystal Si Growth Simulation Using Machine Learning Force Field

Computational Prediction of Solvation Structure in Calcium Battery Electrolytes

Size Dependent Lithium-Ion Conductivity of Solid Electrolytes in Machine Learning Molcular Dynamics Simulations

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

Electrolytomics: A unified big data approach for electrolyte design and discovery

Machine Learning-Driven Investigation of the Structure and Dynamics of the BMIM-BF4 Room Temperature Ionic Liquid

Transferability and Accuracy of Ionic Liquid Simulations with Equivariant Machine Learning Interatomic Potentials

Machine learning 3D-resolved prediction of electrolyte infiltration in battery porous electrodes

Uni-ELF: A Multi-Level Representation Learning Framework for Electrolyte Formulation Design

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

Differentiable modeling and optimization of non-aqueous Li-based battery electrolyte solutions using geometric deep learning

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

Chemistry-Informed Machine Learning for Polymer Electrolyte Discovery

Molecular simulations of electrolyte structure and dynamics in lithium-sulfur battery solvents

Machine-learned molecular mechanics force fields from large-scale quantum chemical data