Predicting Miscibility in Binary Compounds: A Machine Learning and Genetic Algorithm Study

TL;DR

This study combines machine learning and genetic algorithms to predict miscibility in binary compounds, successfully identifying stable phases and providing insights to guide experimental synthesis in materials science.

Contribution

It introduces a data-driven approach using atomic-level data and genetic algorithms to predict miscibility and discover new stable phases in binary compounds.

Findings

Random forest model accurately predicts miscibility.

Identified three novel stable phases in the Co-Eu system.

Integrated large experimental datasets for comprehensive analysis.

Abstract

The combination of data science and materials informatics has significantly propelled the advancement of multi-component compound synthesis research. This study employs atomic-level data to predict miscibility in binary compounds using machine learning, demonstrating the feasibility of such predictions. We have integrated experimental data from the Materials Project (MP) database and the Inorganic Crystal Structure Database (ICSD), covering 2,346 binary systems. We applied a random forest classification model to train the constructed dataset and analyze the key factors affecting the miscibility of binary systems and their significance while predicting binary systems with high synthetic potential. By employing advanced genetic algorithms on the Co-Eu system, we discovered three novel thermodynamically stable phases, CoEu8, Co3Eu2, and CoEu. This research offers valuable theoretical insights to guide experimental synthesis endeavors in binary and complex material systems.