Predicting mechanical properties of CO2hydrates: machine learning insights from molecular dynamics simulations

Abstract

Understanding the mechanical properties of CO₂hydrate is crucial for its diverse sustainable applications such as CO₂geostorage and natural gas hydrate mining. In this work, classic molecular dynamics (MD) simulations are employed to explore the mechanical characteristics of CO₂hydrate with varying occupancy rates and occupancy distributions of guest molecules. It is revealed that the mechanical properties, including maximum stress, critical strain, and Young's modulus, are not only affected by the cage occupancy rate in both large 5¹²6²and small 5¹²cages, but also by the distribution of guest molecules within the cages. Specifically, the presence of vacancies in the 5¹²6²large cages significantly impacts the overall mechanical stability compared to 5¹²small cages. Furthermore, four distinct machine learning (ML) models trained using MD results are developed to predict the mechanical properties of CO₂hydrate with different cage occupancy rates and cage occupancy distributions. Through analyzing ML results, as-developed ML models highlight the importance of the distribution of guest molecules within the cages, as crucial contributor to the overall mechanical stability of CO₂hydrate. This study contributes new knowledge to the field by providing insights into the mechanical properties of CO₂hydrates and their dependence on cage occupancy rates and cage occupancy distributions. The findings have implications for the sustainable applications of CO₂hydrate, and as-developed ML models offer a practical framework for predicting the mechanical properties of CO₂hydrate in different scenarios.

Keywords: CO2 hydrate; guest molecular vacancy; machine learning; mechanical properties; molecular dynamics.