Abstract
The Hume–Rothery rules have guided alloy design for nearly a century by evaluating atomic radius and electronegativity compatibility at ambient pressure. Here we extend these classical descriptors from static ambient values to pressure-dependent response functions spanning 0–300 GPa, testing whether the dynamic behaviour of atomic properties under compression contains predictive information for binary alloy mixing absent from single-point evaluation.
Using 780 binary pairs from 40 elements with mixing enthalpies from the DFT calculations and pressure-dependent atomic data from the SHARC database, we constructed 83 features: 13 ambient-only (classical Hume–Rothery baseline) and 70 pressure-dependent (gradients, variabilities, and trajectory stabilities). Statistical hypothesis testing demonstrates that pressure-dependent features improve prediction of the mixing enthalpy sign by Δ = +11.5% over ambient-only descriptors (p = 0.0001, Cohen's d = 2.69).
Pipeline-aware SHAP analysis attributes 84.6 ± 0.1% of predictive contribution to pressure-dependent features across three independent runs. The dominant descriptors—electronegativity variability and gradient under compression—directly extend the classical electronegativity-mismatch criterion: favourable mixing correlates not merely with small ambient Δχ, but with electronegativity trajectories that preserve electronic compatibility across the pressure domain. Similarly, atomic radius gradients extend the 15% size-mismatch rule by capturing differential compressibility. External validation on 29 lanthanum-containing pairs not included in training achieved 80% accuracy.
These results establish that the logic of Hume–Rothery—evaluating atomic compatibility to predict mixing—gains substantial predictive power when extended from single ambient values to pressure-response trajectories.
Supplementary materials
Title
Pressure-Dependent Atomic Properties Extend Classical Hume–Rothery Descriptors for Binary Alloy Mixing Prediction
Description
the atomic data files are attached here. the link to the codes is provided in the methods description in the manuscript
Actions
Supplementary weblinks
Title
XCE workflow
Description
all the code used in the paper and data used are uploaded in this github page
Actions
View 


![Author ORCID: We display the ORCID iD icon alongside authors names on our website to acknowledge that the ORCiD has been authenticated when entered by the user. To view the users ORCiD record click the icon. [opens in a new tab]](https://www.cambridge.org/engage/assets/public/coe/logo/orcid.png)