About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Computational Thermodynamics and Kinetics
|
| Presentation Title |
Encoding Locality in an Alloy Cluster Expansion for More Efficient Atomistic Simulations |
| Author(s) |
Jacob Jeffries, Enrique Martinez |
| On-Site Speaker (Planned) |
Jacob Jeffries |
| Abstract Scope |
First-principles simulations of alloys are notoriously expensive, often becoming intractable for systems exceeding a thousand atoms. Cluster expansion techniques have proven effective in circumventing these computational costs for concentrated alloys. However, current formulations of the cluster expansion do not explicitly encode the locality of interatomic interactions. As a result, existing tools often recompute redundant interaction energies during Monte Carlo simulations. In this work, we reformulate the cluster expansion to explicitly encode locality by incorporating the system’s topology into the feature representation prior to fitting. This encoding allows us to write the energy difference in terms of local interaction energies without approximation, avoiding the computation of redundant interaction energies. We demonstrate the method on several alloy systems of interest, observing significant reductions in simulation time. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Computational Materials Science & Engineering, Modeling and Simulation, Machine Learning |