COAST : Catalysts for a sustainable future.

The project will combine digital, experimental, and methodological achievements:

• Machine learning (ML) and digital developments – ML approaches aimed at frugality are capable of making predictions and analyses that are accurate enough to be usable with a minimum of data. The development of frugal approaches is necessary to reduce the environmental impact of artificial intelligence. This goal can be achieved by developing models dedicated to specific tasks rather than large generic models requiring gigantic databases. On the one hand, we want to accelerate surface science analyses by integrating experimental data (tunneling microscopy, photoemission, spectroscopy, surface diffraction) to constrain the space of numerical parameters of the structural models needed for interpretations (tunneling microscopy image simulations, electronic structure calculations). On the other hand, we want to continue developing ML methods for predicting quantities (adsorption energies, barriers, etc.) with a minimum of ab initio data. In terms of numerical simulations, we will use global optimization methods, approaches based on density functional theory, and approaches based on machine learning potentials, which we will refine for our systems of interest. Finally, exploration of the chemical space will enable us to select one or more phases for catalytic testing under real conditions, with a view to designing effective catalysts.
• Database – An experimental database compiling the characterizations of sp-TM surfaces under ultra-high vacuum, as well as under various in-situ conditions. This part of the project will make extensive use of the Soleil and Tube Daum experimental platforms at the IJL, as well as GENCI computing resources. The catalytic performance of the synthesized materials will be evaluated. A database compiling the adsorption energies of molecules of interest on a wide variety of surfaces will be developed. At the experimental level, the database will be constructed using Pt-Sn and Pd-(Ga,Sn) systems, oxidized under controlled conditions (Pt-Sn-O and Pd-(Ga,Sn)-O). Our choice is based on their extensive tunability through chemical substitutions, giving them a wide variety of electronic properties. Modeling under reactive conditions will be considered, in particular the possible formation of oxide layers on the surface, which plays an important role in chemical reactivity.