Local Order Parameters Classifying Water Networks of Ice and Cyclopropane Clathrate Hydrates

Abstract

Clathrate hydrates are crystalline solids with guest molecules in a hydrogen-bonding network of frozen water. Cyclopropane hydrate has a four-phase coexistence point of guest gas, ice, and two hydrates, suggesting ice–hydrate and hydrate–hydrate structural transitions, but little is known about the details. Microscopic approaches such as molecular simulations are expected as a means of elucidation, but these require reaction coordinates that are sensitive to structural changes (e.g., order parameters). Here, 1220 local order parameters (LOPs) and their combinations were used to classify cyclopropane hydrate and ice structures in the vicinity of the four-phase coexistence point. A total of 21 599 490 LOP combinations were automatically and systematically evaluated via supervised machine learning. The optimal (set of) LOP(s) distinguished between structure I and structure II cyclopropane hydrates and ice Ih with high accuracy and robustness.