Wider research context
A pivotal aspect of sustainable development is closing the carbon cycle by converting CO2 to more valuable chemical feedstocks. The thermocatalytic CO2 hydrogenation to methanol is a highly promising route to that aim. Binary oxides are an excellent alternative to metal-based catalysts for this reaction. However, unraveling the synergistic effect of combining two oxides and predicting their catalytic activity is challenging. Density Functional Theory (DFT) calculations are usually combined with microkinetic models to predict catalytic activities, but they are highly computationally demanding, hindering catalysts' theory-driven tailored design. In this context, data-driven approaches and artificial intelligence techniques are emerging tools in the computational design of materials, but several challenges remain in the generative design of heterogeneous catalysts.
Hypotheses and Objectives
The fundamental atomistic properties of binary metal oxides determining the CO2 hydrogenation activity are lacking, hindering their application as methanol synthesis catalysts. The project aims to investigate the parameter space of mixed oxides to provide structure-property-activity relationships in CO2 hydrogenation and drive this knowledge to the generative design of new catalytic materials. To aid the methodological developments, we will use complex binary oxides active in the CO2 hydrogenation to methanol: In2O3/ZrO2, ZnO/ZrO2, and ZnM2O4 spinels.
Approach and Methods
We will evaluate the physicochemical properties of the selected materials via advanced DFT-based methods to use them as descriptors in supervised machine learning (ML) algorithms to predict their experimental catalytic performance. After extensively testing the algorithms and via surrogate ML models to obtain the most computationally demanding features, we will perform the high throughput screening of binary oxides via the developed tools. Then, we will formalize the inverse generation of binary oxides active in the target reaction by using generative conditional learning combining variational autoencoders with diffusion-based methods.
Level of Originality and Innovation
The project envisages an original, artificial intelligence-driven, and top-bottom approach to the computational design of heterogeneous catalysts, contrasting with the state-of-the-art bottom-up strategy. It will pave the way to the inverse design of heterogeneous catalysts while providing unprecedented knowledge of oxides as thermocatalyzed CO2 hydrogenation catalysts.
Primary researchers involved
Prof. Aleix Comas-Vives (TU Wien) will lead the proposal and supervise two co-workers funded by the project. Two international experimental collaborators with whom the PI has active collaborations: Prof. Christophe Copéret and Prof. Christoph Müller (ETH Zürich) will provide reliable experimental data to benchmark the methodologies and test the potential of the theoretical predictions.
