HighHydrogenML · High-throughput Discovery of Catalysts for the Hydrogen Economy through Machine Learning

High-throughput Discovery of Catalysts for the Hydrogen Economy through Machine Learning

The 7th sustainable development goal of the United Nations Organization is to ensure access to affordable, reliable, sustainable, and modern energy for all, while the 13th goal is to take urgent action to combat climate change. Both objectives are closely linked to the use of clean and renewable energy sources for addressing the world’s rising energy needs while limiting environmental effects. However, further expansion of renewable energies (currently 28% of the world’s electricity) is limited because power provided by solar and wind energy (as opposed to coal, natural gas and nuclear) is intermittent and must be coupled with large energy storage capabilities. The most relevant large energy storage capabilities are pumping water uphill (pumped-storage hydroelectricity), batteries and hydrogen storage. While pumped-storage hydroelectricity and batteries have high efficiency, they do not scale to the energy storage needs of large grids powered by intermittent renewable energy due to land restrictions or cost, respectively. On the contrary, hydrogen energy storage offers a unique combination of scalability, long-term storage, and portability, leading to the so-called hydrogen economy. In fact, the difficulty with hydrogen economy is not with storage but with the production of hydrogen from water and the generation of energy by the oxidation of hydrogen into water. The former process is controlled by the Hydrogen Evolution Reaction (HER), the cathodic reaction during the electrochemical process of dissociation of the water molecule into hydrogen and oxygen. Among the various alternatives for transforming hydrogen into electrical energy, polymer electrolytic membrane fuel cells are among the most interesting and widely applied options in transportation. The main limitation for the industrial application of this technology is the reduced kinetics of the cathode Oxygen Reduction Reaction (ORR). As in the case of the HER, the sluggish kinetics of the ORR can only be improved by increasing the reaction temperature or by using Pt catalysts. Thus, the search for efficient, cheap, durable, and non-toxic catalysts that can replace Pt for the HER and ORR is of dramatic importance for the hydrogen economy but has been unsuccessful so far. The catalytic activity of a material is controlled by the electronic structure that can be modified by adding other elements to form an alloy or compound or by introducing defects in the crystal lattice. Furthermore, the electronic structure of the atoms on the surfaces is different from that of the atoms inside the material and it also varies between surfaces depending on the crystallographic orientation. Another mechanism to modify the catalytic activity of surfaces is based on elastic strain engineering (ESE), i.e., the application of large elastic deformations to modify the electronic structure. Nevertheless, a systematic application of ESE to search for catalysts for both HER and ORR has not been carried out. The main objective of the project High-throughput Discovery of Catalysts for the Hydrogen Economy through Machine Learning (HighHydrogenML) is to develop a high-throughput strategy based on first principles calculations and artificial intelligence tools to discover intermetallic compounds whose catalytic activity can be tuned to reach an optimum catalytic performance for the HER and ORR by means of elastic strain engineering. The successful completion of these objectives will provide unique information for experimental synthesis of intermetallic compounds with high catalytic activity for the HER and ORR and could, therefore, open a new avenue for a feasible and efficient hydrogen economy. Moreover, the strategies and tools developed in this project can be applied later to many other catalytic processes of large industrial and/or environmental interest (such as ammonia production, carbon sequestration, etc.). The specific objectives for achieving this goal are: i) Construction by means of density functional theory of a reference dataset of adsorption energies of intermetallic compounds for the adsorption of H, O, and OH including the effect of elastic strains. ii) Development of robust ML models for the prediction of adsorption energies of intermetallic compounds for the HER and ORR including the effect of elastic strains. iii) Finding of intermetallic compounds that can achieve superior catalytic performance for the HER and ORR through the application of elastic strains.

Data: CORDIS, © European Union