Breakdown of linear correlations in methane activation on doped nickel nanoparticles: physical origins and nonlinear relations revealed by symbolic regression

Abstract

Dry reforming of methane (DRM) both mitigates greenhouse gases (CH₄, CO₂) and produces syngas (H₂, CO) for fuels and chemicals. Designing efficient DRM catalysts requires quantitative links between reaction barriers and adsorption/reaction energies. Traditional Brønsted–Evans–Polanyi (BEP) scaling, although widely assumed, may fail to predict quantitative kinetics, motivating the discovery of nonlinear relations that better couple thermodynamics and kinetics. Here, we investigate CH₄ and CO₂ decomposition on 12 doped Ni nanoparticles (M–Ni) using density functional theory. We show that BEP breakdown in DRM arises not only from binding-site preferences but also from the intrinsic electronic properties of the dopant, as analyzed by diabatic state calculations; kinetic modeling further amplifies these deviations. To uncover governing connections, we employ symbolic regression to derive physically interpretable expressions linking dopant descriptors to kinetic and thermodynamic parameters. Compared with linear scaling, these relations capture nonlinear effects, reveal hidden correlations, and provide predictive formulas that accelerate high-throughput screening. Validation on 12 previously unexplored doped nanoparticles confirms transferability: the best candidate, Mo–Ni, shows overall reaction energy and rate-determining barrier predictions within 0.05 eV of DFT. This integrated mechanistic and data-driven framework offers design rules for highly active, coke-resistant DRM nanocatalysts beyond BEP limitations.