Machine Learning for Transition Metal Catalysts

Machine learning (ML) is becoming a powerful tool in chemistry, though its use is often challenged by the lack of big experimental data. This problem can be tackled by using computational data, following the quantum-based ML approach. We started this new research line in 2018, focusing on the prediction of oxidative addition barriers in the chemical space surrounding Vaska's complex. More recently, we published a large dataset with transition metal Quantum Mechanics properties for 86k complexes and a method for the automated optimization of transition states. Our long-term goal is to maximize the accuracy, transferability, and explainability of ML models in the metal-organic chemical space.


Single-Site Catalysis

Homogeneous catalysts can be designed to be highly active and selective but suffer from robustness and recyclability. Single-site catalysis has the potential to overcome these limitations by offering the advantages of heterogeneous and homogeneous catalysts. We are applying DFT methods for (1) spectroscopy simulations enabling the characterization of the active sites, (2) reaction mechanism determination, including confinement effects, and (3) designing multifunctional catalysts based on metal-organic frameworks (MOFs) activating inert substrates like methane in highly selective processes. 


Catalytic CO2 Reduction

The conversion of CO2 to fuels by means of renewable energies involves catalysts that require further optimization for their large-scale implementation. We are using DFT methods and microkinetic models to gain quantitative insight into the mechanism of these reactions and the steps governing catalytic activity and selectivity. Further, we aim at designing catalysts able to perform multiple steps in an energy-efficient manner. We are interested in both homogeneous and single-site heterogeneous catalysts working under thermal and electrochemical conditions.