Prediction and characterization of chemically complex σ − phase intermetallics with graph neural network
Gespeichert in:
| Titel: | Prediction and characterization of chemically complex σ − phase intermetallics with graph neural network |
|---|---|
| Autoren: | Zhang, Wenhao, Forti, Mariano, Xie, Runan, Barreteau, Céline, Abe, Taichi, Joubert, Jean-Marc, Hammerschmidt, Thomas, Crivello, Jean-Claude |
| Weitere Verfasser: | Crivello, Jean-Claude |
| Verlagsinformationen: | 2025. |
| Publikationsjahr: | 2025 |
| Schlagwörter: | σ-phase, [CHIM.INOR] Chemical Sciences/Inorganic chemistry, [INFO.INFO-AI] Computer Science [cs]/Artificial Intelligence [cs.AI], [CHIM.MATE] Chemical Sciences/Material chemistry, Machine learning, Intermetallic, [CHIM.CHEM] Chemical Sciences/Cheminformatics, Thermodynamic modeling |
| Beschreibung: | Determining the thermodynamic stability of intermetallic phases in multicomponent alloys from Compound Energy Formalism (CEF), requires the formation enthalpy of up to millions of end-member configurations. Here, a graph neural network (GNN) model, trained from high-throughput DFT calculations, was used to construct such a database of intermetallic σ-phase for 19 metallic elements with a root mean squared error (RMSE) of 10 meV/atom. We show that a good prediction accuracy of GNN, even without prior knowledge of the relaxed structure, can be achieved by focusing on a single phase and using a regression model for structure prediction. The resulting 19 5 formation enthalpy data, corresponding to all end-member configurations up to quinaries, enables the characterization of the stability and chemical trends of σ-phase both at 0 K (configurations on the convex hull) and at high temperature (with disorder using CEF, also up to to quinary systems). We quantified the important role of entropy in stabilizing the σ-phase in multicomponent systems and investigated the relationship between chemical |
| Publikationsart: | Article |
| Dateibeschreibung: | application/pdf |
| Sprache: | English |
| DOI: | 10.1016/j.actamat.2025.121427 |
| Zugangs-URL: | https://hal.science/hal-05262526v1/document https://hal.science/hal-05262526v1 |
| Rights: | CC BY |
| Dokumentencode: | edsair.od.....10692..a049b146e4b14c91a9d11a9bdfd98f30 |
| Datenbank: | OpenAIRE |
Nájsť tento článok vo Web of Science | Abstract: | Determining the thermodynamic stability of intermetallic phases in multicomponent alloys from Compound Energy Formalism (CEF), requires the formation enthalpy of up to millions of end-member configurations. Here, a graph neural network (GNN) model, trained from high-throughput DFT calculations, was used to construct such a database of intermetallic σ-phase for 19 metallic elements with a root mean squared error (RMSE) of 10 meV/atom. We show that a good prediction accuracy of GNN, even without prior knowledge of the relaxed structure, can be achieved by focusing on a single phase and using a regression model for structure prediction. The resulting 19 5 formation enthalpy data, corresponding to all end-member configurations up to quinaries, enables the characterization of the stability and chemical trends of σ-phase both at 0 K (configurations on the convex hull) and at high temperature (with disorder using CEF, also up to to quinary systems). We quantified the important role of entropy in stabilizing the σ-phase in multicomponent systems and investigated the relationship between chemical |
|---|---|
| DOI: | 10.1016/j.actamat.2025.121427 |