Going Beyond the d-band Center to Design Intermetallic Catalysts for Nitrogen Reduction: A High-Throughput DFT and Machine Learning Study
Abstract
This study combines high-throughput density functional theory (DFT) calculations with machine learning (ML) to uncover the key descriptors governing the nitrogen reduction reaction in 47 intermetallic compounds (IMCs) composed of Co, Ni, Al, Zn, V, Fe, Cu, and Pt, and the adsorption energies of key intermediates (*N2, *N2H, and *NH3) were systematically evaluated across all accessible surface sites, yielding approximately 1,200 data points. Among the IMCs studied, Fe9Co7 and Fe3Co emerge as the most balanced catalysts, exhibiting favorable adsorption energies across all three intermediates. By incorporating intrinsic material properties along with local and global electronic descriptors including s-, p- and d-band centers and fillings, as well as Bader charges of atoms neighboring the adsorbate, predictive ML models were developed with mean absolute errors (MAEs) of 0.27 eV for *N2, 0.39 eV for *N2H, and 0.17 eV for *NH3 adsorption. Importantly, accurate adsorption-energy predictions were achieved using only 20 key features for *N2H and *NH3 and 38 features for *N2, enabling the use of simple and computationally efficient ML models. SHAP analysis indicates that p-band and s-band characteristics play a more prominent role in determining adsorption strength than the traditionally used d-band center, particularly for *N2. Beyond their established importance in systems containing p-block elements or nearly filled d-band metals, s- and p-orbitals are also found to contribute significantly to transition metal alloy activity such as Fe-Co. By challenging the d-band-centric paradigm and identifying s- and p-band descriptors as critical yet overlooked contributors, this work redefines the electronic descriptor space for intermetallic NRR catalysts and lays the groundwork for DFT-ML-guided discovery of non-noble, compositionally complex materials for sustainable ammonia synthesis.
Related
- AI-guided high-throughput discovery of iridium- and ruthenium-free palladium-oxide catalysts for durable acidic oxygen evolution
- Machine Learning for High-Entropy Catalysts: Methods and Applications
- Sustainability-Aware Discovery of Dual-Atom Hydrogen Evolution Catalysts via a Unified Machine Learning Framework for Activity and Carbon Cost
- Embracing Flow Unsteadiness: A High-Throughput Learning Platform Enables Vortex-Exploiting Bioinspired Propulsion
- Improved screening of doped ZrO2 catalysts for CO2 hydrogenation from graph-based machine learning
BibTeX
@misc{agharezaei2026going,
title = {{Going Beyond the d-band Center to Design Intermetallic Catalysts for Nitrogen Reduction: A High-Throughput DFT and Machine Learning Study}},
author = {Parastoo Agharezaei and Kulbir Kaur Ghuman},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2607.17463},
eprint = {2607.17463},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2607.17463},
}
RIS
TY - UNPB
TI - Going Beyond the d-band Center to Design Intermetallic Catalysts for Nitrogen Reduction: A High-Throughput DFT and Machine Learning Study
AU - Parastoo Agharezaei
AU - Kulbir Kaur Ghuman
PY - 2026
DA - 2026/09/23
PB - arXiv
DO - 10.48550/arxiv.2607.17463
UR - https://arxiv.org/abs/2607.17463
AB - This study combines high-throughput density functional theory (DFT) calculations with machine learning (ML) to uncover the key descriptors governing the nitrogen reduction reaction in 47 intermetallic compounds (IMCs) composed of Co, Ni, Al, Zn, V, Fe, Cu, and Pt, and the adsorption energies of key intermediates (*N2, *N2H, and *NH3) were systematically evaluated across all accessible surface sites, yielding approximately 1,200 data points. Among the IMCs studied, Fe9Co7 and Fe3Co emerge as the most balanced catalysts, exhibiting favorable adsorption energies across all three intermediates. By incorporating intrinsic material properties along with local and global electronic descriptors including s-, p- and d-band centers and fillings, as well as Bader charges of atoms neighboring the adsorbate, predictive ML models were developed with mean absolute errors (MAEs) of 0.27 eV for *N2, 0.39 eV for *N2H, and 0.17 eV for *NH3 adsorption. Importantly, accurate adsorption-energy predictions were achieved using only 20 key features for *N2H and *NH3 and 38 features for *N2, enabling the use of simple and computationally efficient ML models. SHAP analysis indicates that p-band and s-band characteristics play a more prominent role in determining adsorption strength than the traditionally used d-band center, particularly for *N2. Beyond their established importance in systems containing p-block elements or nearly filled d-band metals, s- and p-orbitals are also found to contribute significantly to transition metal alloy activity such as Fe-Co. By challenging the d-band-centric paradigm and identifying s- and p-band descriptors as critical yet overlooked contributors, this work redefines the electronic descriptor space for intermetallic NRR catalysts and lays the groundwork for DFT-ML-guided discovery of non-noble, compositionally complex materials for sustainable ammonia synthesis.
ER -