Optimal Mixture Designs for Alloy Compositions
Abstract The quality of the experimental design critically determines the accuracy of surrogate models in computationally-driven materials discovery. We introduce scaled support points as a design strategy for Gaussian process surrogate modeling on alloy composition simplices and benchmark them against random sampling, Quasi-Monte Carlo samples, and K-medoids clustering across two refractory high-entropy alloy systems: a quaternary and a septenary alloy space. We found that scaled support points achieve lower or comparable median normalized mean absolute error relative to competing methods across the budgets tested, with smaller variance and near-elimination of catastrophic surrogate failures.
Authors
- Surya R. Kalidindi (ORCID: https://orcid.org/0000-0001-6909-7507)
- Pranoy Ray
- Wei-Yang Yu
- V. Roshan Joseph
Institutions
- Georgia Institute of Technology (US)
Publication Details
- Journal
- JOM
- Published
- 2026-11-13
- DOI
- https://doi.org/10.1007/s11837-026-08881-7
- Primary Topic
- Machine Learning in Materials Science
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- National Science Foundation