About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Theory and Design of Metallic Glasses
|
| Presentation Title |
Identifying the Fundamental Structural Length Scale in Cu-Zr Metallic Glasses using Graph Neural Networks |
| Author(s) |
Emily Gurniak, Paulo S Branicio |
| On-Site Speaker (Planned) |
Emily Gurniak |
| Abstract Scope |
Crystalline materials are described by a primitive unit cell, yet an analogous structural unit for glasses has remained elusive. Here, we uncover a characteristic length scale for Cu–Zr metallic glasses by coupling molecular-dynamics simulations with graph-neural-network (GNN) analysis. Six amorphous states were generated by quenching a Cu64Zr36 liquid at rates from 109 to 1015 Ks−1. Atomic graphs built from 5,488-atom cubes (~4.4 nm per side) enabled the GNN to classify the quenched states with 99.8 % accuracy, far surpassing the 81 % achieved with 686-atom (~2.2 nm) samples reported previously. The sharp improvement pinpoints medium-range order spanning roughly 4 – 5 nm as the minimal representative volume required to uniquely encode glassy structures. Beyond resolving subtle structural variations, our graph-based workflow provides a quantitative route to defining “unit cells” for disordered materials, opening new avenues for the predictive design of metallic glasses. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, Modeling and Simulation, |