(573g) Machine Learning Approach for Gas Hydrate Polymorph Classification to Study Their Nucleation Via Molecular Simulation

Natural gas hydrates are the class of crystalline compounds which are formed from natural gases such as methane, ethane, carbon di oxide etc. entrapped in a three dimensional cages formed by water molecules via hydrogen bonds. There exist a three well defined unit crystals (sI, sII, sH) known to be the polymorphs of gas hydrates. These compounds form in deep sea beds and permafrost regions under the conditions of low temperatures and high pressures. Natural gas hydrates have tremendous scientific and technological importance. It is of vital technological importance to understand their kinetics and mechanisms of nucleation during crystallization processes. In this regard, molecular simulations play an important and complementary role to experiments in obtaining a detailed molecular picture of the nucleation process. Analysis of nucleation trajectories requires the use of crystalline order parameters that can classify molecules belonging to different phases. Designing such order parameters is in general not a trivial task and requires a lot of time and effort. In this work, we concentrate on obtaining order parameters that can identify natural gas hydrates and classify them into different polymorphs using machine learning techniques. We start with a pool of generic order parameters¹ and systematically obtain the minimal set that can identify and classify water molecules belonging to different polymorphs of gas hydrates using principles of machine learning. The XGBoost² machine learning model is trained to classify water molecules belonging to liquid, ice-1h, ice-1c, sI, sII and sH hydrates. The trained model is then used to analyze the nucleation of methane hydrates near a planar gas-liquid water interface. The nucleation rate is computed using the mean first passage time (MFPT) formalism³. We compare and contrast the performance of our machine learning model with previously developed order parameters in literature⁴. We conclude that a suitable minimal set of generic order parameters can classify different crystal polymorphs with high accuracy and thereby obviating the need to develop specialized tailor-made order parameters for any material under investigation. . Furthermore, the developed machine leaning model can help us to compute the distributions of various ice and hydrates polymorphs with the nucleus. Results show that the critical nucleus formed has a disordered mixture of hydrate phases, with no dominant crystalline structure, indicating an amorphous nature.

References

1. Doi, H., Takahashi, K. Z. & Aoyagi, T. Searching for local order parameters to classify water structures at triple points. J. Comput. Chem. 42, 1720–1727 (2021).
2. Chen, T. & Guestrin, C. XGBoost: A scalable tree boosting system. Proc. ACM SIGKDD Int. Conf. Knowl. Discov. Data Min. 13-17-Augu, 785–794 (2016).
3. Chkonia, G., Wölk, J., Strey, R., Wedekind, J. & Reguera, D. Evaluating nucleation rates in direct simulations. J. Chem. Phys. 130, 64505 (2009).
4. Nguyen, A. H. & Molinero, V. Identification of Clathrate Hydrates, Hexagonal Ice, Cubic Ice, and Liquid Water in Simulations: The CHILL+ Algorithm. J. Phys. Chem. B 119, 9369–9376 (2015).