Research Projects

The urgent global demand for advanced materials in sustainable energy, storage systems, and next-generation electronics requires a transformative approach to materials discovery. High-entropy materials (HEMs), composed of four or more principal elements in near-equimolar ratios, offer a promising platform due to their entropy-stabilized single-phase structures and superior thermal, mechanical, and functional properties. Yet, fewer than a thousand synthesizable HEMs have been experimentally realized, only a tiny fraction of the vast thermodynamic space. This project proposes an integrated high-throughput framework combining thermodynamic theory with machine learning (ML) to accelerate the discovery of novel HEMs for energy conversion, storage, and electronic applications. At its core is the Mixed Enthalpy–Entropy Descriptor (MEED), a physics-grounded metric recently developed by the PI, which captures the competition between formation enthalpy and configurational entropy to predict synthesizability. While MEED has shown exceptional predictive power in identifying viable high-entropy carbides and chalcogenides, scaling it to broader classes of materials and integrating it with learning models requires a robust, automated infrastructure. The central hypothesis of this work is that embedding MEED within a physically interpretable ML architecture will enable scalable and generalizable prediction of synthesizable and functionally relevant HEMs across a wide compositional spectrum. This hypothesis will be tested via a five-phase strategy: 1. High-throughput MEED-based screening will be carried out across quaternary to hexanary compositions in oxides, nitrides, halides, carbides, and layered 2D materials. This will result in a large dataset (~20,000 entries) annotated with thermodynamic descriptors. 2. Thermodynamic refinement will apply convex hull analysis, valence balancing, and oxidation-state filters to curate a synthesizable materials dataset. 3. Machine learning models, based on Random Forest algorithms, will be trained to classify synthesizable vs. non-synthesizable compositions and predict properties such as bandgap, ionic conductivity, among others. 4. Active learning loops will identify high-uncertainty regions and suggest new candidates for DFT evaluation, continuously improving model accuracy. 5. Deployment of an automated framework including Application Programming Interface (APIs) and user-friendly interfaces for broader access to the framework. If successful, this project will result in a powerful, open-access computational tool that transforms the discovery of high-entropy materials from an ad hoc, trial-based endeavor into a predictive, scalable process. The outcomes will directly support India’s clean energy, electronic materials, and digital innovation missions by enabling accelerated identification of strategic material platforms.