Abstract:Accurate modeling of long-range forces is critical in atomistic simulations, as they play a central role in determining the properties of materials and chemical systems. However, standard machine learning interatomic potentials (MLIPs) often rely on short-range approximations, limiting their applicability to systems with significant electrostatics and dispersion forces. We recently introduced the Latent Ewald Summation (LES) method, which captures long-range electrostatics without explicitly learning atomic charges or charge equilibration. Extending LES, we incorporate the ability to learn physical partial charges, encode charge states, and the option to impose charge neutrality constraints. We benchmark LES on diverse and challenging systems, including charged molecules, ionic liquid, electrolyte solution, polar dipeptides, surface adsorption, electrolyte/solid interfaces, and solid-solid interfaces. Our results show that LES can effectively infer physical partial charges, dipole and quadrupole moments, as well as achieve better accuracy compared to methods that explicitly learn charges. LES thus provides an efficient, interpretable, and generalizable MLIP framework for simulating complex systems with intricate charge transfer and long-range

What problem does this paper attempt to address?

This paper attempts to solve the problem of how to accurately model long - range interactions when simulating materials and chemical systems at the atomic scale. Specifically, standard machine - learning inter - atomic potentials (MLIPs) usually rely on short - range approximations, which limit their application in systems involving significant electrostatic and dispersion forces. This paper proposes and extends a method called implicit Ewald summation (LES) to capture long - range electrostatic interactions without the need to explicitly learn atomic charges or charge balance. ### Core of the problem 1. **Importance of long - range interactions**: Long - range interactions play a crucial role in determining the properties of materials and chemical systems, especially in scenarios involving electrochemical interfaces, charged - molecule dimers, ions and polar materials, and those involving charge transfer. 2. **Limitations of existing methods**: - Most existing MLIP frameworks rely on short - range approximations, assuming that the energy contribution of each atom is mainly determined by its local atomic environment. - Although this assumption is computationally efficient, it has obvious limitations for systems with significant long - range interactions. 3. **Improvement directions**: Introduce new methods that can effectively capture long - range interactions, especially those that can infer physical partial charges, dipole moments and quadrupole moments through the learning of energy and force without explicitly learning charges. ### Advantages of the LES method - **Efficiency and interpretability**: The LES method provides an efficient and interpretable MLIP framework by decomposing the total potential energy into short - range and long - range components and using hidden variables ("implicit charges") to predict the long - range potential energy. - **Wide applicability**: LES can be combined with any short - range MLIP architecture and is suitable for simulating fine charge transfer and long - range forces in complex systems. - **No need for explicit charge learning**: Compared with methods that require explicit learning of partial charges, LES can effectively infer physical partial charges and dipole moments when only learning from reference energies and forces. ### Main contributions of the paper 1. **Theoretical basis**: The theoretical basis of the LES method is introduced in detail, including range separation, long - range energy calculation, and the process of learning charges from energy and force. 2. **New extensions**: The ability to encode different global charge states is introduced, and how to impose charge - neutrality constraints during the training process is demonstrated. 3. **Application examples**: The effectiveness and superior performance of the LES method are verified through multiple test systems (such as random - charge gases, electrolyte solutions, charged - molecule dimers, polar dipeptides, etc.). In summary, this paper aims to solve the limitations of existing MLIP methods in dealing with long - range interactions and proposes a more efficient, general and accurate solution.

Learning charges and long-range interactions from energies and forces

Latent Ewald summation for machine learning of long-range interactions

Incorporating long-range electrostatics in neural network potentials via variational charge equilibration from shortsighted ingredients

On the role of long-range electrostatics in machine-learned interatomic potentials for complex battery materials

Machine Learning Interatomic Potentials and Long-Range Physics

Self-consistent Coulomb interactions for machine learning interatomic potentials

Electrostatic interactions in atomistic and machine-learned potentials for polar materials

Charge equilibration model with shielded long-range Coulomb for reactive molecular dynamics simulations

End-To-End Learning of Classical Interatomic Potentials for Benchmarking Anion Polarization Effects in Lithium Polymer Electrolytes

Density-Based Long-Range Electrostatic Descriptors for Machine Learning Force Fields

Modeling electronic response properties with an explicit-electron machine learning potential

Synergistic integration of physical embedding and machine learning enabling precise and reliable force field

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

Learning the Quantum Centroid Force Correction in Molecular Systems: A Localized Approach

A deep potential model with long-range electrostatic interactions

Kernel charge equilibration: efficient and accurate prediction of molecular dipole moments with a machine-learning enhanced electron density model

Machine Learning Potential for Electrochemical Interfaces with Hybrid Representation of Dielectric Response

High-Dimensional Neural Network Potential for Liquid Electrolyte Simulations.

Assessment of Embedding Schemes in a Hybrid Machine Learning/Classical Potentials (ML/MM) Approach

Investigating Ionic Diffusivity in Amorphous Solid Electrolytes using Machine Learned Interatomic Potentials