Dataset for: Understanding methane cleavage on medium-entropy oxides

# Dataset for “Understanding methane cleavage on medium-entropy oxides” This dataset accompanies the article **“Understanding methane cleavage on medium-entropy oxides”** by S. Yuan, H. Song, Z. Wu and D. Jiang (*Catalysis Science & Technology*, 2026). The dataset provides site-resolved, DFT-computed adsorption energies and climbing-image nudged elastic band (CI-NEB) activation barriers for the first C–H bond cleavage of methane (CH4 → CH3 + H) on the (100) surface of the medium-entropy oxide Mg0.25Zn0.25Ni0.25Cu0.25O. It also includes the feature tables used to learn the compact analytical activation-energy descriptor reported in the article, enabling direct reproduction of the machine-learning analysis. ## Contents This dataset contains four data files and one documentation file. ### `methane_activation_MEO_activation.csv` This file contains 77 elementary C–H cleavage reactions: 30 heterolytic, 21 homolytic-trans and 26 homolytic-cis pathways. It includes nine columns: the reaction identifier, mechanism, site indices and identities of the CH4-, CH3- and H-hosting sites, and the CI-NEB activation energy, Ea. ### `site_adsorption_energies.csv` This file contains the adsorption energies of H, CH3 and CH4 at each of the 16 lattice sites of the MEO (100) cell: eight metal-cation sites and eight lattice-oxygen sites, indexed from 1 to 16. These site indices are the same as those used in the reaction identifiers, so the per-site adsorption energies can be attached to any reaction by joining on the site index. ### `methane_activation_MEO_features_full.csv` This file contains the full candidate-feature pool used for descriptor discovery. It includes 77 rows and 42 features, together with `reaction_id`, `mechanism` and the target activation energy, `Ea`. Each reaction is augmented with electronic-structure descriptors of the participating sites in the initial CH4-adsorbed state and the final CH3 + H state. These descriptors include Bader charges, d-band occupations and centres, electronegativity, covalent radius, cross-site adsorption energies and surface-aggregate quantities. ### `methane_activation_MEO_features_15.csv` This file contains the curated 15-feature subset retained after removing constant, near-zero-variance and redundant columns. These 15 features were used as inputs to the SISSO activation-energy descriptor reported in Section 3.4 of the article. Each column maps directly onto a symbol in the published descriptor. ### `README.md` This file provides full documentation, including file and column definitions, units, numerical precision, the site-naming scheme and computational methods. ## Numerical precision All numerical values in the CSV files are rounded to two decimal places. Integer columns, including site indices and mechanism indicators, are stored as integers. Identifier and label columns are stored as text. ## Site-naming scheme Each reaction is identified by a three-number code, `i_j_k`, where `i` is the cation site on which CH4 is adsorbed, `j` is the site that hosts the CH3 fragment after cleavage, and `k` is the lattice-oxygen site that hosts the abstracted H atom. For heterolytic cleavage, the CH3 group remains on the original cation site (`j = i`). For the homolytic pathways, `j` and `k` are both lattice-oxygen sites. The labels `trans` and `cis` denote the relative geometry of the two oxygen sites around the cation. ## Sign convention All energies are reported in eV. Adsorption energies follow `E_ads(X) = E(X/slab) − E(X,gas) − E(slab)` where `X` denotes the adsorbate. Under this convention, a more negative adsorption energy indicates stronger binding. Activation energies are reported as positive values. ## Computational details All calculations were performed with VASP 6 using the PBE functional and PAW pseudopotentials, with the `Ni_pv` potential used for Ni. A plane-wave cutoff of 450 eV and a 3 × 3 × 1 Γ-centred k-point mesh were used. Geometry optimisations were converged to a force threshold of 0.05 eV Å−1 and an electronic energy threshold of 10−5 eV. A Hubbard correction was applied to the Ni 3d states using the Dudarev DFT+U approach, with an effective U − J value of 6.2 eV. Grimme DFT-D3 dispersion corrections and spin polarisation were included in all calculations. The surface was modelled as a five-layer 4 × 4 Mg0.25Zn0.25Ni0.25Cu0.25O (100) slab with at least 15 Å of vacuum. The top two atomic layers were relaxed, while the bottom three layers were fixed at their bulk positions. The cation sublattice was generated by randomly redistributing the equimolar Mg, Zn, Ni and Cu cations over 200 independent realisations. One of the lowest-energy configurations was retained as the representative model.Transition states were located using the climbing-image nudged elastic band method with five intermediate images. ## Reuse This dataset enables reproduction of the descriptor-based machine-learning analysis reported in the article and supports follow-up studies of methane activation on medium-entropy oxides and related multicomponent-oxide catalysts.