Machine Learning for Vibrational Spectroscopic Maps

• DOI: 10.1021/acs.jctc.9b00698
• Publication Date: 2019-10-15T22:46:08Z
• Authors: Alexei A. Kananenka; Kun Yao; Steven A. Corcelli; J. L. Skinner
• Keywords: Complex system

Maps that relate spectroscopic properties of a vibrational mode and collective solvent coordinates have proven useful in theoretical spectroscopy condensed-phase systems. It has been realized the predictive power such an approach is limited there no clear systematic way to improve its accuracy. Here, we propose adaptation Δ-machine-learning methodology goes beyond maps. The machine-learning part our combines Gaussian process regression used generate data set with artificial neural network predict interest. A specific application OH-stretch frequencies transition dipoles water presented. Our method approximates these about two times more accurately than spectroscopic-maps-only-based approach. results suggest new may become study