Machine Learning–Accelerated Discovery of Earth-Abundant Bimetallic Electrocatalysts for the Hydrogen Evolution Reaction

Overview

This repository contains the computational data and results for the discovery of earth-abundant bimetallic HER electrocatalysts using machine learning and density functional theory.

Key Achievement: Identification of four synthesis-ready catalysts (Fe₂Sb₄, Cu₆Sb₂, Cu₆Sn₂, Ni₂Sb₂) with platinum-like performance at significantly lower cost.

Methodology

• Chemical Space: 14 elements (Ir, Pt, Ru, Ni, Co, Fe, Mn, Ta, Ti, Nb, Mo, Cu, Sn, Sb)
• ML Model: EquiformerV2 within AdsorbML framework
• DFT Validation: VASP with RPBE functional
• Screening Scale: ~560,000 initial adsorption sites → 361 final candidates

Data Files

Core Results

• 361_slab_DFT_optimized_energy.csv - Final 361 candidates after DFT optimization
• 500_slab_EqV2_SP_DFT.csv - 500 configurations with ML and DFT validation
• AllBulks.pkl - All 439 bulk structures
• ASCIrank.csv - Activity-Stability-Cost Index rankings
• EquiformerV2_H_ads_prediction.csv - ML adsorption energy predictions
• InDomainBulk.pkl - In-domain bulk structures (OC20)
• OutOfDomainBulk.pkl - Out-of-domain bulk structures (reoptimized)

Analysis Data

• performance_comparison.csv - Activity-Stability-Cost Index rankings

Structure Files

Compressed Archives (ZIP format)

• MLrelaxedTrajMinHadsData.zip - ML-relaxed trajectories and minimum H adsorption data
• VASP_InputandOutputFiles.zip - VASP input and output files for DFT calculations
• VolcanoPlot-VASP_configurations.zip - Structures used in volcano plot analysis
• Out_Of_Domain_Vasp.zip - VASP input and output files for fully optimized out-of-domain bulk systems

File Formats

• CSV files: Comma-separated with headers, energies in eV
• PKL files: Python pickle format containing bulk structure data
• ZIP files: Compressed archives containing VASP structure files and trajectories

Usage

Load data using standard tools:

import pandas as pd
import pickle

# Load final candidates
candidates = pd.read_csv('data/361_slab_DFT_optimized_energy.csv')

# Load bulk structures
with open('data/AllBulks.pkl', 'rb') as f:
    bulks = pickle.load(f)

Structure files in ZIP archives are in VASP POSCAR format, compatible with ASE, pymatgen, VESTA.

Notebooks

Analysis notebooks for data exploration and visualization:

• demo_InDomain_H_ads.ipynb - In-domain hydrogen adsorption analysis
• SurfaceEnergy_ASCat_demo.ipynb - Surface energy Calculations and Surface reconstruction detection
• Material_Screening.ipynb - Screens bimetallics and classifies them as in- or out-of-domain (OC20)

Citation

If you use this dataset, structures, or analysis from our manuscript, please cite:

@article{oguz2025machine,
  title={Machine Learning--Accelerated Discovery of Earth-Abundant Bimetallic Electrocatalysts for the Hydrogen Evolution Reaction},
  author={Oguz, Ismail Can and Khossossi, Nabil and Brunacci, Marco and Bucak, Haldun and Er, Süleyman},
  journal={ACS Catalysis},
  volume={15},
  pages={19461--19474},
  year={2025},
  publisher={ACS Publications},
  doi={10.1021/acscatal.5c04967},
  url={https://pubs.acs.org/doi/full/10.1021/acscatal.5c04967}
}

Published Article: Machine Learning–Accelerated Discovery of Earth-Abundant Bimetallic Electrocatalysts for the Hydrogen Evolution Reaction
ACS Catalysis 2025, 15, 19461–19474. DOI: 10.1021/acscatal.5c04967

Contact

Corresponding Author: Süleyman Er (s.er@differ.nl)
DIFFER – Dutch Institute for Fundamental Energy Research