Data-driven design of mechanically hard soft(-magnetic) high-entropy alloys (HEAs)

This repository contains data, scripts, and examples for data-driven discovery and optimization of high-entropy alloys (HEAs) that aim to combine:

• mechanical hardness proxies (e.g., bulk modulus, elastic descriptors), and
• soft-magnetic proxies (e.g., high saturation magnetization / magnetic moment and Curie temperature).

It consolidates two tightly connected components:

1. High-throughput EMTO–CPA database + ML surrogate modeling (Advanced Science, 2025).
2. Multi-objective Bayesian optimization (MOBO) loop for accelerated composition search (arXiv:2509.05702).

---

Related publications

• Advanced Science (2025): Data-Driven Design of Mechanically Hard Soft Magnetic High-Entropy Alloys
  DOI: 10.1002/advs.202500867
  Full text (open access / PMC): https://pmc.ncbi.nlm.nih.gov/articles/PMC12097130/

• arXiv (2025): Accelerated Design of Mechanically Hard Magnetically Soft High-entropy Alloys via Multi-objective Bayesian Optimization
  arXiv:2509.05702
  https://arxiv.org/abs/2509.05702

If you use this repository, please cite the papers above (see Citation).

---

Project highlights

1) High-throughput database (HTP-DFT / EMTO–CPA)

• Large-scale DFT dataset (EMTO–CPA) spanning equimolar quaternary and quinary HEAs.
• For each composition, multiple magnetic states and crystal prototypes are evaluated:
  • magnetic: FM vs PM
  • structure: BCC vs FCC
• Target properties include:
  • B0: bulk modulus (EOS-fit)
  • mag / M_tot: total magnetic moment (proxy for saturation magnetization)
  • TC: Curie temperature (mean-field estimate in the database workflow)

2) Ensemble ML surrogate modeling

• Train predictive models to map (composition + structure/volume descriptors) → (phase/structure + target properties).
• Use ensemble learning to improve robustness and generalization in a large and diverse chemical space.

3) MOBO extension (accelerated search)

The MOBO workflow (arXiv:2509.05702) demonstrates:

• a 10-element pool (Sc, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn),
• exploration of equimolar and non-equimolar quinary compositions,
• multi-objective optimization over:
  • Pugh’s ratio (B/G) and Cauchy pressure (C12 − C44) as mechanical descriptors,
  • Total magnetization,
  • Curie temperature (TC),
• with an ensemble surrogate (bootstrapping + stacking) and Monte-Carlo-based acquisition evaluation.

---