Accelerating molecular vibrational spectra simulations with a physically informed deep learning model

Abstract: In recent years, ML surrogate models have emerged as an indispensable tool to accelerate simulations of physical and chemical processes. However, there is still a lack of ML models that can accurately predict molecular vibrational spectra. Here, we present a highly efficient high-dimensional neural network potentials (HD-NNP) architecture to accurately calculate infrared (IR) and Raman spectra based on dipole moments and polarizabilities obtained on-the-fly via ML-molecular dynamics (MD) simulations. The methodology is applied to pyrazine, a prototypical polyatomic chromophore. The HD-NNP predicted energies are well within the chemical accuracy (1 kcal/mol), and the errors for HD-NNP predicted forces are only one-half of those obtained from a popular high-performance ML model. Compared to the ab initio reference, the HD-NNP predicted frequencies of IR and Raman spectra differ only by less than 8.3 cm^-1, and the intensities of IR spectra and the depolarizaiton ratios of Raman spectra are well reproduced. The HD-NNP architecture developed in this work highlights importance of constructing highly accurate NNPs for predicting molecular vibrational spectra.