High-Dimensional Fluctuations in Liquid Water: CombiningChemical Intuition with Unsupervised Learning

The microscopic description of the local structure of water remainsan open challenge. Here, we adopt an agnostic approach to understanding water'shydrogen bond network using data harvested from molecular dynamics simulationsof an empirical water model. A battery of state-of-the-art unsupervised data-sciencetechniques are used to characterize the free-energy landscape of water starting fromencoding the water environment using local atomic descriptors, throughdimensionality reduction andfinally the use of advanced clustering techniques.Analysis of the free energy under ambient conditions was found to be consistentwith a rough single basin and independent of the choice of the water model. Wefind that thefluctuations of the water network occur in a high-dimensional space,which we characterize using a combination of both atomic descriptors andchemical-intuition-based coordinates. We demonstrate that a combination of bothtypes of variables is needed in order to adequately capture the complexity of thefluctuations in the hydrogen bond network atdifferent length scales both at room temperature and also close to the critical point of water. Our results provide a general frameworkfor examiningfluctuations in water under different conditions.

Automated summary

This study used data science techniques to characterize the free-energy landscape of water, revealing that fluctuations of the water network occur in a high-dimensional space and a combination of different variables is needed to capture the complexity of the hydrogen bond network fluctuations.