Machine Learning-Guided Design of L1₂-Type Pt-Based High-Entropy Intermetallic Compound for Electrocatalytic Hydrogen Evolution.

Abstract

Rational design of high-entropy intermetallic compounds (HEICs) remains challenging due to complex structure-property relationships and the lack of predictive tools. Here, a data-driven framework is presented to evaluate the hydrogen evolution reaction (HER) activity of L1₂-type quinary Pt₃M(4) HEICs, where M comprises any four elements from six 3d transition metals (Cr, Mn, Fe, Co, Ni, Zn). Guided by the Pm-3m space group, 15 distinct compositions with numerous microstates are designed. A deep neural network, trained on 453 computed datasets, predicts hydrogen adsorption energy (∆E_H*) across 20 000 microstructures per composition, enabling statistical mapping of site-specific performance. To capture the effect of local atomic environments, a novel statistical evaluation approach is introduced that quantifies the number of microstates falling within the optimal ∆E_H* range, advancing beyond conventional mean-based evaluations. Among all candidates, Pt₃(CrMnFeCo) emerges as the most promising HER catalyst, validated experimentally over a wide pH range. Further in-depth data mining reveals that surface Co, Cr, and Fe optimize Pt-Pt-M sites, while subsurface Ni and Co modulate Pt-Pt-Pt interactions. This study establishes a new paradigm for HEIC catalyst design and deepens the mechanistic understanding of activity origin in complex multimetal systems.