Machine Learning–Assisted Prediction of Anharmonicity-Corrected Vibrational Spectra

We have recently developed a tool to calculate anharmonicity-corrected vibrational spectra [Withanage et al. "Incorporating Anharmonicity within Density Functional Theory Calculations." arXiv preprint arXiv:2509.02961 (2025)] based on the Vibrational Configuration Interaction (VCI) method. This framework enables the inclusion of anharmonic effects beyond the harmonic approximation through accurate evaluations of the potential energy surface (PES) on both Gauss-Hermite quadrature grid and highly efficient variational meshes. In the present work, we explore the applicability of the UMA [Wood et al., “UMA: A Family of Universal Models for Atoms,” preprint arXiv:2506.23971, (2025)] machine learning model for efficient PES evaluations, enabling the computation of anharmonic vibrational properties of molecules. We apply this approach to a diverse set of molecules to compute their vibrational spectra and further model proton tunneling to investigate its effects on the vibrational characteristics of the systems.