| About this Abstract |
| Meeting |
2022 TMS Annual Meeting & Exhibition
|
| Symposium
|
Phase Stability, Phase Transformations, and Reactive Phase Formation in Electronic Materials XXI
|
| Presentation Title |
The Challenge of Machine Learning the Stability of Materials |
| Author(s) |
Chris Bartel |
| On-Site Speaker (Planned) |
Chris Bartel |
| Abstract Scope |
The realization of high-throughput materials discovery depends critically on the ability to predict whether a given material will be synthesizable. This problem is generally addressed in silico by calculating the thermodynamic stability of candidate materials using density functional theory (DFT). There have been many recent claims that this step can be greatly accelerated using machine learning instead of DFT to predict the formation energy of inorganic solids. However, thermodynamic stability is ultimately dictated by relative formation energies between chemically similar competing phases, and not by the absolute formation energy of any single compound. It is therefore essential to assess how well these machine learning predictions of intrinsic thermodynamic quantities generalize to the problem of relative materials stability. In this work, we systematically probe this question by testing a diverse set of materials representations and learning algorithms on a variety of problems that simulate real applications of materials discovery. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Computational Materials Science & Engineering, Machine Learning, Phase Transformations |