Abstract Scope |
I build a deep learning model that predicts melting temperature from chemical composition in milliseconds. The model, along with its database that contains approximately 10K melting points, also serves as a handbook for experimental melting temperatures. The model is deployed and publicly available at my group’s webpage. We utilize this model to design refractory materials with extremely high melting temperatures. We also employ this model to study melting temperatures of 4,700 naturally formed minerals, which correlates reasonably well with two major events in Earth's history. This extremely rapid model complements the SLUSCHI method we previously built for accurate, robust, and automated density functional theory melting point calculations. We are integrating the deep learning model and the first principles method to build a framework for rapid and accurate melting temperature prediction. |