(708e) Powering a Green Future: Accelerating Hydrogen Technologies with Machine Learning and Computational Modeling

Hydrogen has gained significant attention as a promising energy carrier in the drive toward net-zero emissions by 2050. Its high energy density by mass, abundant availability, and clean combustion make it well-suited for supporting the integration and storage of renewable energy. Furthermore, with versatile production pathways and a wide range of industrial applications, hydrogen is particularly valuable for decarbonizing hard-to-abate sectors such as heavy transport, metallurgy, and cement production, positioning it as a key pillar of the global energy transition. However, advancing hydrogen as a sustainable and affordable energy source requires two critical developments: 1) highly efficient catalysts for water-splitting for green hydrogen production through the hydrogen evolution reaction (HER), and 2) materials with superior proton conductivity for proton exchange membrane (PEM) systems in electrolyzers and fuel cells.

Electrochemical and photochemical water-splitting technologies represent key strategies for sustainable hydrogen production. Among candidate catalysts, metal sulfides offer great potential due to their favorable electronic properties, including narrow band gaps, high charge transfer efficiency, and abundant active sites. However, the variety of metal sulfide compositions, surface facets, and active sites presents a significant challenge for identifying optimal materials through experimental or DFT-based screening alone. To accelerate the discovery of optimal materials, we have developed a machine learning-aided screening framework to evaluate 881 metal sulfide lattices for the hydrogen evolution reaction (HER). DFT calculations were used to compute ~2000 hydrogen adsorption energies across 3680 unique structures, serving as training data for supervised ML models. Among the different ML models used, our optimized random forest regression model successfully predicted adsorption energies across 10,000+ structures, identifying 10 high-performance and 37 promising catalysts with favorable stability, band gap, and HER activity.

In parallel, we addressed the design of proton exchange membranes (PEMs)—critical for hydrogen electrolyzers and fuel cells—by evaluating proton permeation barriers in non-metallic 2D materials. Through a combination of ab-initio molecular dynamics (AIMD) simulations and ML, we built a dataset of ~500 materials, correlating structural and electronic descriptors (e.g., pore size, diameter, electron affinity) with proton transport efficiency. Additional AIMD studies evaluated H₂/H⁺ selectivity, highlighting 18 promising candidates, including both established materials (e.g., graphene, h-BN) and underexplored yet experimentally synthesized ones such as germanene, cubic silicon, TeC, and GeSe.

Results from this work showcase how the integration of first-principles computational methods (DFT and AIMD) and ML accelerates novel materials discovery and interprets physically meaningful rules to bridge theory and experiments, facilitating efficient materials exploration and catalyst design.

This work is financially supported by the Research & Innovation Center for Graphene and 2D Materials (project 2DMat4H2-D0001, RIC2D centre). Computational resources from the RICH Center at Khalifa University (project RC2-2019-007) are also gratefully acknowledged.