Long-range electrostatics for machine learning interatomic potentials is easier than we thought.

Area of Science:

• Computational materials science
• Quantum chemistry
• Machine learning

Background:

• Machine learning interatomic potentials (MLIPs) are crucial for simulating materials.
• A key limitation of current MLIPs is the absence of long-range electrostatic interactions.
• This deficiency restricts their application in areas like interfaces, charge-transfer reactions, and polar materials.

Purpose of the Study:

• To present the Latent Ewald Summation framework for incorporating long-range electrostatics into MLIPs.
• To distill two core design principles for capturing electrostatic interactions and electrical response.
• To demonstrate the framework's flexibility and broad applicability.

Main Methods:

• Developing a Coulomb functional form incorporating environment-dependent charges.
• Avoiding explicit training on density functional theory (DFT) partial charges.
• Augmenting existing short-range MLIPs with the Latent Ewald Summation framework.

Main Results:

• The Latent Ewald Summation framework effectively captures long-range electrostatic interactions, charges, and electrical response.
• The framework can be integrated with various MLIPs and charge equilibration schemes.
• Inferred or fine-tuned dipoles and Born effective charges are achievable.

Conclusions:

• Incorporating long-range electrostatics into MLIPs is more straightforward than previously thought.
• The presented physics-guided design rules offer a broadly applicable solution.
• This advancement enables more reliable MLIP simulations for a wider range of materials and phenomena.