Mining of Effective Local Order Parameters to Classify Ice Polymorphs.

Area of Science:

• Materials Science
• Computational Chemistry
• Chemical Physics

Background:

• Order parameters quantify structural ordering in materials, crucial for analyzing phase transitions.
• Local order parameters (LOPs) are valuable for understanding molecular structures during transitions, but identifying effective LOPs is challenging.
• Supervised machine learning offers a powerful approach to systematically screen and optimize LOPs.

Purpose of the Study:

• To investigate the efficacy of local order parameters (LOPs) in classifying solid and liquid water structures.
• To systematically compare a vast number of LOP combinations for accuracy in phase discrimination.
• To identify optimal LOP sets for water at specific coexistence and triple points.

Main Methods:

• Utilized supervised machine learning to analyze 179,738,433 combinations of 493 LOPs.
• Applied automated and systematic comparison of LOP classification accuracy.
• Focused on water structures at the ice Ih-Ic-liquid coexistence point and ice III-V-liquid and ice V-VI-liquid triple points.

Main Results:

• Identified optimal sets of two LOPs for classifying water structures at each investigated phase transition point.
• Demonstrated the effectiveness of machine learning in screening large LOP datasets.
• Achieved high classification accuracy for distinguishing between solid and liquid water phases.

Conclusions:

• LOPs, when optimized using machine learning, are highly effective for classifying water phases.
• Specific sets of LOPs can accurately distinguish between different solid and liquid water structures at coexistence and triple points.
• Suggests that sets of three LOPs can further enhance classification accuracy for complex phase transitions.