Exciting DeePMD: Learning excited state energies, forces, and non-adiabatic couplings

TL;DR

This paper extends the DeePMD neural network to accurately predict excited state energies, forces, and non-adiabatic couplings, enabling improved non-adiabatic dynamics simulations in quantum chemistry.

Contribution

It introduces a novel approach to learning non-adiabatic coupling vectors directly from local chemical environment descriptors, overcoming previous approximations.

Findings

Accurately predicts excited state properties for CH2NH2+

Demonstrates improved non-adiabatic dynamics simulation capabilities

Validates the method's efficiency and accuracy

Abstract

We extend the DeePMD neural network architecture to predict electronic structure properties necessary to perform non-adiabatic dynamics simulations. While learning the excited state energies and forces follows a straightforward extension of the DeePMD approach for ground-state energies and forces, how to learn the map between the non-adiabatic coupling vectors (NACV) and the local chemical environment descriptors of DeePMD is less trivial. Most implementations of machine-learning-based non-adiabatic dynamics inherently approximate the NACVs, with an underlying assumption that the energy-difference-scaled NACVs are conservative fields. We overcome this approximation, implementing the method recently introduced by Richardson [J. Chem. Phys. 158 011102 (2023)], which learns the symmetric dyad of the energy-difference-scaled NACV. The efficiency and accuracy of our neural network architecture is demonstrated through the example of the methaniminium cation CH$_2$NH$_2^+$.