Learning Electronic Polarization in Molecular Systems: Vibrational Spectroscopy of Ethanol–Water Mixtures

Mixtures of water and ethanol are an important solvent for biochemical reactions, as well as being a model system for hydrophobic hydration, since the ethanol molecule consists of polar and nonpolar regions. Experiments carried out over the last several decades have not yet reached agreement on what the addition of ethanol does to the structure of water, with some studies finding evidence for highly ordered cage-like structures and others finding no such evidence. We combine machine-learning force fields for the interatomic interactions in systems containing organic molecules with models that we develop for polarization, and show that we obtain good agreement with experimental infrared spectra, whose results are thought to imply the formation of cage structures. We show instead that in our simulations there is a decrease in the structure of water close to ethanol molecules and that the experimental observations can be explained by the change in structure further away from the ethanol.