[Paper Review] Open Catalyst Experiments 2024 (OCx24): Bridging Experiments and Computational Models

Jehad Abed, Jiheon Kim+18 more|arXiv (Cornell University)|Nov 18, 2024

Catalytic Processes in Materials Science17 citations

TL;DR

OCx24 creates a large, diverse experimental dataset of intermetallic nanoparticles and couples it with extensive AI-accelerated DFT adsorption-energy screening to bridge experiments and computational catalysts models for HER and CO2RR.

ABSTRACT

The search for low-cost, durable, and effective catalysts is essential for green hydrogen production and carbon dioxide upcycling to help in the mitigation of climate change. Discovery of new catalysts is currently limited by the gap between what AI-accelerated computational models predict and what experimental studies produce. To make progress, large and diverse experimental datasets are needed that are reproducible and tested at industrially-relevant conditions. We address these needs by utilizing a comprehensive high-throughput characterization and experimental pipeline to create the Open Catalyst Experiments 2024 (OCX24) dataset. The dataset contains 572 samples synthesized using both wet and dry methods with X-ray fluorescence and X-ray diffraction characterization. We prepared 441 gas diffusion electrodes, including replicates, and evaluated them using zero-gap electrolysis for carbon dioxide reduction (CO$_2$RR) and hydrogen evolution reactions (HER) at current densities up to $300$ mA/cm$^2$. To find correlations with experimental outcomes and to perform computational screens, DFT-verified adsorption energies for six adsorbates were calculated on $\sim$20,000 inorganic materials requiring 685 million AI-accelerated relaxations. Remarkably from this large set of materials, a data driven Sabatier volcano independently identified Pt as being a top candidate for HER without having any experimental measurements on Pt or Pt-alloy samples. We anticipate the availability of experimental data generated specifically for AI training, such as OCX24, will significantly improve the utility of computational models in selecting materials for experimental screening.

Motivation & Objective

• Motivate catalyst discovery for green hydrogen production and CO2 utilization by addressing the gap between AI-predicted and experimentally observed performance.
• Provide a reproducible, industrially relevant experimental pipeline with diverse material samples for AI training.
• Combine high-throughput experimental data with large-scale computational screening to enable predictive models of experimental outcomes.
• Develop preprocessing and matching methods to link experimental results with computational descriptors.

Proposed method

• Synthesize 572 intermetallic nanoparticle samples using two techniques (chemical reduction and spark ablation) to maximize compositional and structural diversity.
• Characterize samples with X-ray fluorescence (XRF) and X-ray diffraction (XRD) to filter for single-phase, target-structure materials.
• Test electrochemical performance for CO2 reduction (CO2RR) and hydrogen evolution (HER) at industrially relevant current densities up to 300 mA/cm2 using a modified MEA setup.
• Compute adsorption energies for six intermediates (OH, CO, CHO, C, COCOH, H) across ~19,406 materials via the AdsorbML pipeline combining AI and DFT relaxations (685 million relaxations; ~20 million DFT points).
• Build predictive models linking computational descriptors to experimental outcomes using linear and random forest regressions, with both surface-level and bulk descriptors; evaluate with LOOC and LOCO cross-validation.

Figure 1 : A summary of the computational and experimental screening efforts, and the resulting outcomes. Computationally, six adsorbates were used as descriptors and their adsorption energies were calculated across a wide swath of materials. Experimentally, two synthesis techniques were used to pre

Experimental results

Research questions

• RQ1 Can AI-accelerated computed descriptors predict experimental HER and CO2RR performance across a diverse set of intermetallics?
• RQ2 How well do matched (XRD/XRF-confirmed) versus unmatched samples generalize to predict experimental outcomes?
• RQ3 What is the incremental predictive value of including adsorption-energy descriptors vs. bulk Matminer features for material-level performance?
• RQ4 Can a Sabatier-type volcano emerge from data without any Pt-containing training samples, and what does this imply for discovering low-cost catalysts?
• RQ5 How does dataset size affect the predictivity of models linking computation to experiment?

Key findings

• A data-driven Sabatier volcano appears for HER with Pt at the apex, despite Pt not appearing in the training data.
• Linear models using adsorption energies as features achieve R2≈0.59 (LOO) and ~0.61 (including more energies) for HER predictions, outperforming Matminer-only models in some cases.
• For CO2RR, LOCO indicates weaker generalization (H2 and liquids show modest correlations; CO shows near-zero correlation).
• Joint training on matched and unmatched samples improves predictive performance over using only matched samples.
• Predicted HER catalysis candidates include low-cost, non-Pt compositions; a substantial set of materials (including Se-containing and Mo-based alloys) emerge as potential catalysts.
• Model performance improves with larger training datasets, with projections suggesting substantial gains at 10^4–10^5 samples.

Figure 2 : This figure illustrates the experimental pipeline, detailing the process from synthesis to characterization and testing. (right) Cu baselines for the two synthesis techniques and a reference technique using spray coated nanoparticles. Similar trends across current densities are seen for a