Beyond averaging: δH, a chemical heterogeneity descriptor for single-phase formation prediction in multicomponent alloys

A new dimensionless descriptor for single-phase microstructure prediction in multicomponent alloys (MCA) and high-entropy alloys (HEA) is proposed: the normalised dispersion of binary mixing enthalpies, δ H . This descriptor quantifies the internal heterogeneity of chemical interactions among constituent element pairs —a contribution that classical descriptors do not capture, as they rely on weighted averages that mask the coexistence of strongly attractive and repulsive pairs. Combined with the atomic size mismatch parameter δ , the pair δ H – δ significantly outperforms the Ω – δ reference criterion established by Yang and Zhang (2012), with an improvement of +3.16 percentage points in cross-validated accuracy (83.16% vs 80.00%) and +0.063 in the Matthews Correlation Coefficient (0.663 vs 0.600), validated on an experimental dataset of 1675 alloys from three independent sources spanning binary to nonary compositions (McNemar test, p = 0.006). Although δ H and δ are moderately correlated at the global level (Spearman ρ = 0.27), partly because both tend to increase with compositional complexity, the significant improvement of δ H – δ over any univariate classifier confirms that δ H contributes predictive information that δ alone does not capture. The critical threshold δ H ≤ 1.1 has a direct physical interpretation: the solid solution is stable when the heterogeneity of binary chemical interactions does not exceed 1.1 times the mean thermal energy available at the solidification temperature.