Combining Artificial Intelligence, Quantum Mechanics and Statistical Mechanics for the Design of CO₂ Reduction Catalysts

Abstract

This Bachelor’s Thesis addresses, through an integrated scientific and business perspective, the development and potential commercialisation of a computational screening platform for heterogeneous catalysts aimed at CO₂ conversion and utilisation. From a scientific standpoint, the work develops a computational framework that combines three complementary methodologies: Density Functional Theory (DFT) as the reference ab initio approach, Grand Canonical Monte Carlo (GCMC) for exploring the compositional and configurational space of catalytic surfaces under reaction conditions, and Message Passing Atomic Cluster Expansion (MACE) potentials as an efficient surrogate for energy and force evaluations. This framework is applied to the thermodynamic characterisation of oxidised Cu(100) and Cu(111) surfaces, leading to the construction of surface phase diagrams that identify the most stable atomic configurations as a function of the oxygen chemical potential. From a business perspective, the thesis proposes the preliminary design of a university spin-off specialised in computational catalyst screening services for industries linked to catalysis, sustainable fuels, and the energy transition. The proposed value proposition consists of providing advanced simulation-based screening tools to research institutions, chemical companies, refineries, synthetic-fuel producers, and organisations involved in Carbon Capture, Utilisation and Storage (CCUS). Within this context, the Cu/Oₓ system investigated in the scientific part of the thesis serves as both a case study and a technological proof of concept. The economic analysis, based on realistic estimates of experimental characterisation costs, high-performance computing (HPC) resources, and cloud-computing services, indicates that the computational approach can reduce the cost per evaluated structure from approximately €2.300 using conventional experimental methodologies to around €28.31 through computational screening. This corresponds to a cost reduction of nearly 98.8%, while also shortening evaluation times from weeks to only a few hours per structure. Five-year financial projections suggest profitability from the third year of operation, an estimated Internal Rate of Return (IRR) between 22% and 28%, and an investment profile consistent with the growth expectations of a technology-based university spin-off. Overall, the results demonstrate both the scientific viability and the economic potential of integrating artificial intelligence, atomistic simulations, and high-performance computing into catalyst discovery workflows, highlighting their role in accelerating innovation for carbon-utilisation technologies and the broader energy transition.