A joint data and model driven method for study diatomic vibrational spectra including dissociation behavior.

ABSTRACT

The details of quantum multi-body interactions are so rich and subtle which make it difficult to accurately model for some situations such as the behavior of diatomic long-range vibrations. In recent years, data-driven machine learning has made remarkable achievements in capturing complex relationships that are subtle. Combining the characteristics of these two fields, we propose a joint machine learning method to obtain reliable diatomic vibrational spectra including dissociation energy by using accessible heterogeneous micro/macro information such as low lying vibrational energy levels and heat capacity. Applications of this method to CO and Br2 in the ground state yield their state of the art of vibrational spectra including dissociation limit. The strategy introduced here is an exploration of combining the model-driven and data-driven method to cover subtle physical details that are difficult to study in a single way.