HEGEN: A Data-Free Physics-Informed Framework for Inverse Design of High-Entropy Materials

Read full text Summarise with VIRAT
Authors: Ziqi Yin, Jack Yang, Sean Li
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Manipulating the lattice thermal conductivity (κ) of high-entropy perovskites (HEPs) through phonon engineering is a persistent challenge that is critical for developing materials for applications in thermoelectrics and thermal barrier coatings. While the configurational entropy of HEPs offers conceptual framework for stabilizing the chemically disordered phases to enhance phonon scattering thus suppressing the thermal transport, the vast combinatorial space of multicomponent compositions limits our capacities in exploring the best material candidates through exploitative search. To bridge this gap, we propose HEGEN (High-Entropy GENerator), a reward-guided generative model for inverse design composition of high entropy materials. HEGEN employs a Gumbel-Softmax relaxation to sample the discrete distributions that represent the chemical elements. It is trained using a physics-informed reward function derived from the Klemens theory of phonon scatterings in alloys by optimizing the competitions among mass and ionic radius mismatch (disorder), bond strength, and structural feasibility constraints. By avoiding the explicit values of thermal conductivity in the training data, it enables HEGEN to operate in a data-free regime. Our model successfully identifies promising both low and high κ L compositions when validated against explicit computation of κ via molecular dynamic simulation that is combined with the Green-Kubo formalism, demonstrating that ionic radius disorder is a critical contributing factor for suppressing phonon heat transport in high-entropy systems. This work thus establishes a physics-informed workflow for the inverse design of complex multicomponent materials.

Fields of Study & Subjects

Machine Learning in Materials Science High Entropy Alloys Studies Thermal properties of materials

Publication details

Journal / source
ChemRxiv
Publication date
26 March 2026
Language
EN
Document type
Preprint