Catalyst GFlowNet for electrocatalyst design: A hydrogen evolution reaction case study

TL;DR

Hydrogen energy storage relies on efficient, inexpensive electrocatalysts; this work presents Catalyst GFlowNet, a generative framework that builds catalyst surfaces via Crystal-GFN, relaxes them with M3GNet, and evaluates adsorbate energy with a FAENet-based proxy (DepFAENet). Samples are drawn proportionally to a reward defined as $R(x)=\exp(-b\,η^2)$ with $η=E_H(x)+E_{corr}$, $b=100$, and $E_{corr}=-0.24$, enabling diverse high-performing candidates to be generated. In a proof-of-concept HER case study with a restricted search space, the method rediscoveres Pt (and Rh) as top catalysts and demonstrates stable structure generation across samples. The work lays a foundation for de novo catalyst discovery beyond HER, including extension to the oxygen evolution reaction and larger, more realistic search spaces, potentially accelerating the search for affordable catalysts.

Abstract

Efficient and inexpensive energy storage is essential for accelerating the adoption of renewable energy and ensuring a stable supply, despite fluctuations in sources such as wind and solar. Electrocatalysts play a key role in hydrogen energy storage (HES), allowing the energy to be stored as hydrogen. However, the development of affordable and high-performance catalysts for this process remains a significant challenge. We introduce Catalyst GFlowNet, a generative model that leverages machine learning-based predictors of formation and adsorption energy to design crystal surfaces that act as efficient catalysts. We demonstrate the performance of the model through a proof-of-concept application to the hydrogen evolution reaction, a key reaction in HES, for which we successfully identified platinum as the most efficient known catalyst. In future work, we aim to extend this approach to the oxygen evolution reaction, where current optimal catalysts are expensive metal oxides, and open the search space to discover new materials. This generative modeling framework offers a promising pathway for accelerating the search for novel and efficient catalysts.

Figures (4)

• Figure 1: Overview of the Catalyst GFlowNet framework. The leftmost section, Catalyst GFlowNet, samples the catalyst surface. The middle section determines atom positions, relaxes the structure, and converts to a graph. The rightmost section obtains the adsorption energy from a predictive model and the reward function to train the GFlowNet.
• Figure 2: Proportion of sampled structures (right) for the hydrogen evolution reaction case study. Low overpotential (left) is a predictor for efficient catalysts and aligns with the higher sampling rates. The values are detailed in Table \ref{['tab:HER_results']} in the appendix. Pd(229) does not have experimental or DFT overpotentials, corresponding values for Pd(225) are displayed with shaded colors for reference.
• Figure 3: For the HER case study, the platinum samples generated by the GFlownet relax to the minimum energy every time. The green line represents the total energy of the system.
• Figure 4: A linear fit that maps log of exchange current density to experimental overpotential using data from trasatti1972work and danilovic2012enhancing. These values are used column 3 (from the left) of Table \ref{['tab:HER_results']}.