Prospects for Using Artificial Intelligence to Understand Intrinsic Kinetics of Heterogeneous Catalytic Reactions

TL;DR

This paper addresses the challenge of extracting intrinsic kinetic information from heterogeneous catalysis data, where many multiscale models can fit the same observables. It proposes self-driving models that couple atomistic, microkinetic, and reactor models with multimodal experimental data through agentic AI, generative models, and measurement-informed workflows; in practice, the workflow envisions solving on the order of $10^6 - 10^8$ forward problem instances to identify ensembles of models that agree with experiments. Key contributions include outlining an architecture for self-driving catalysis, surveying advances in ML reactive force fields, microkinetic methods, automatic mechanism generation, and operando/transient spectroscopy techniques, with exemplars like ensemble approaches using BEEF-vdW and adaptive experimental design. The approach promises to enhance speed, reproducibility, and uncertainty quantification, enabling knowledge engines that synthesize data into interpretable, transferable insights for catalysis and related domains.

Abstract

Artificial intelligence (AI) is influencing heterogeneous catalysis research by accelerating simulations and materials discovery. A key frontier is integrating AI with multiscale models and multimodal experiments to address the "many-to-one" challenge of linking intrinsic kinetics to observables. Advances in machine-learned force fields, microkinetics, and reactor modeling enable rapid exploration of chemical spaces, while operando and transient data provide unprecedented insight. Yet, inconsistent data quality and model complexity limit mechanistic discovery. Generative and agentic AI can automate model generation, quantify uncertainty, and couple theory with experiment, realizing "self-driving models" that produce interpretable, reproducible, and transferable understanding of catalytic systems.

Figures (5)

• Figure 1: Schematic illustration of how a self-driving model connects multiscale models with multimodal experiments by using AI approaches to automate common tasks in multiscale modeling.
• Figure 2: Illustration of results adapted from Kreitz et al. Kreitz2023, where an ensemble of exchange enhancement factors from the BEEF-vdW functional was used to propagate uncertainty through adsorption energies, reaction mechanism generation, and reactor modeling to generate an ensemble of light-off profiles for exhaust gas oxidation. The results were directly compared with experiments, allowing identification of the exchange-correlation functional that was most consistent with experimental observations.
• Figure 3: (a) Time-resolved DRIFTS spectra for CO2 production over Pd/$\gamma$-Al2O3 catalyst with time-resolved intensities of integrated CO* peaks for the (b) forward and (c) backwards transients, and (d) comparison of CO and CO2 absorbance during the forward (green) and backwards (magenta) transients. Direct comparison to simulated coverages and rates from microkinetic models are shown in panels (e-g). Reproduced with permission from Ref. OConnor2025.
• Figure 4: Illustration of model-driven adaptive design of experiments using a synthetic profile reactor setup. The architecture uses a feedback loop to identify models and parameters and suggest new experiments. Reproduced with permission from Ref. Kouyate2025.
• Figure 5: Workflow for CO/Pt(111) study done by an automated agentic system for setting up, running, converging, and evaluating DFT simulations of surfaces and adsorbates.