Fine-tuning universal machine learning potentials for transition state search in surface catalysis

TL;DR

This paper introduces an active learning workflow to fine-tune universal machine learning potentials, enabling accurate and efficient transition state searches in surface catalysis, significantly reducing DFT calculations needed.

Contribution

The study develops a method to adapt universal ML potentials for specific catalytic transition states using active learning, improving accuracy and transferability.

Findings

Active learning reduces DFT calculations to about 8 per structure.

Modified Sella algorithm improves success rate in TS searches.

Fine-tuned uMLPs enable high-throughput catalyst screening.

Abstract

Determining transition states (TSs) of surface reactions is central to understanding and designing heterogeneous catalysts but remains computationally prohibitive with density functional theory (DFT). While machine learning potentials (MLPs) offer significant speedups, task-specific models have limited transferability across catalytic systems, and universal MLPs (uMLPs) lack the accuracy needed for reactive configurations. Here, we present a workflow based on active learning to iteratively fine-tune uMLPs for DFT-quality TS search. Using 250 TSs from the CO2 hydrogenation reaction network on metal and single-atom alloy surfaces, we first benchmark TS search algorithms, identifying the Sella algorithm as most robust, and propose a modification (Bond-Aware Sella) that substantially improves its success rate. We then explore sequential and batch active-learning strategies for fine-tuning and show that DFT-quality TS structures can be found using only 8 DFT single-point calculations on average per structure. This demonstrates the viability of fine-tuned uMLPs for high-throughput catalyst screening.