Data-Driven Design of Photocatalysts for CO₂ Conversion

Research Overview

The increasing concentration of atmospheric CO₂ creates an urgent need for technologies that can not only mitigate emissions but also convert captured carbon into useful fuels and chemical feedstocks.

Among the available approaches, photocatalytic CO₂ reduction is particularly attractive because it offers a route to drive chemical conversion using light.

Transition-metal complexes are promising molecular photocatalysts because their electronic structure, redox properties, excited-state lifetimes, and ligand environments can be systematically tuned.

Their rational design nevertheless remains challenging because catalytic performance is governed by a complex interplay between photoexcitation, electron transfer, intermediate stability, reaction barriers, and excited-state dynamics.

From Re(I) to Earth-Abundant Mn(I) Photocatalysts

My research will focus on the computational design of earth-abundant molecular photocatalysts for CO₂ reduction , using Re(I) and Mn(I) tricarbonyl complexes as a model platform.

Re(I) complexes are particularly attractive because they can function simultaneously as photosensitisers and catalysts. However, the scarcity and cost of rhenium limit their large-scale application.

Structurally analogous Mn(I) complexes offer a more sustainable alternative, but their photochemistry and catalytic behaviour remain considerably less well understood.

A central objective will therefore be to determine how the electronic structure, reaction energetics, and excited-state dynamics of Re(I) photocatalysts can be systematically related to those of their Mn(I) analogues.

Static Description of the Catalytic Mechanism

The first stage will establish a comprehensive static description of the catalytic mechanism .

Ground- and excited-state electronic-structure calculations will determine the energetics of CO₂ binding, reduction, intermediate formation, bond cleavage, product formation, and catalyst regeneration.

Reaction-pathway calculations will provide complete energy profiles for representative Re(I) and Mn(I) complexes, allowing the effects of bidentate and ancillary ligands on individual catalytic steps to be quantified.

Direct comparison of corresponding Re and Mn systems will then establish quantitative relationships between their electronic structures, redox behaviour, activation barriers, and catalytic intermediates.

Excited-State Dynamics of Photocatalytic Activation

A purely static description cannot fully capture the role of photoexcitation in initiating catalysis.

Surface-hopping and related non-adiabatic dynamics approaches will therefore be used to investigate the evolution of initially photoexcited states, metal-to-ligand charge-transfer character, relaxation pathways, electron-transfer processes, and the formation of catalytically active species.

These simulations will examine competition between productive relaxation into long-lived charge-transfer states and access to deactivating or dissociative metal-centred states.

Connecting these ultrafast dynamical processes with the subsequent ground-state catalytic cycle will provide a unified description of the mechanism from photon absorption to CO₂ conversion.

Excited-State Machine-Learning Potentials

Direct quantum-chemical non-adiabatic dynamics remains one of the principal computational bottlenecks in transition-metal photochemistry.

These systems involve internal conversion, intersystem crossing, charge-transfer relaxation, and access to dissociative metal-centred states across dense manifolds strongly influenced by spin–orbit coupling.

To overcome this limitation, the project will establish a transferable framework for multistate machine-learning potentials for excited-state dynamics .

The initial benchmark will use fac-[Re(bpy)(CO)₃Cl] and related Re(I) CO₂-reduction complexes.

The models will describe the competition between relaxation into long-lived metal-to-ligand charge-transfer states that may support productive electron-transfer chemistry and access to ligand-field or metal-centred states associated with deactivation and CO dissociation.

Development Workflow

Methodological Targets

Machine-Learning-Assisted Catalyst Discovery

The computational cost of performing static and dynamical calculations across large families of transition-metal complexes presents an important challenge.

Initial machine-learning models will therefore be trained on high-quality quantum-chemical data to predict properties relevant to photocatalysis, including:

• Excited-state energies and characters
• Ground- and excited-state redox potentials
• CO₂-binding energetics
• Catalytic-intermediate stability
• Reaction and activation barriers
• Ligand-dependent catalytic trends
• Excited-state relaxation pathways
• Photochemical branching ratios

Active-learning strategies will identify regions of molecular and configurational space where additional electronic-structure calculations provide the greatest improvement to the models.

Integrating Static and Dynamic Molecular Information

Dynamic information obtained from non-adiabatic simulations will subsequently be incorporated into the discovery datasets.

This will allow the models to account not only for static reaction energetics but also for the excited-state processes that determine whether a photoexcited complex successfully enters the catalytic cycle.

Molecular candidates will therefore be evaluated according to both conventional catalytic descriptors and dynamical properties such as:

• Excited-state lifetimes
• Charge-transfer relaxation rates
• Intersystem-crossing probabilities
• Population of productive photoreactive states
• Access to deactivating metal-centred states
• Competing ligand-dissociation pathways

Chemical-Space Exploration and Inverse Design

The resulting models will be combined with chemical-space exploration and inverse-design approaches to identify ligand environments and molecular architectures that optimise multiple properties simultaneously.

Candidate photocatalysts will not be selected on the basis of a single favourable quantity. Instead, design objectives will include light absorption, excited-state lifetime, electron transfer, CO₂ activation, catalytic barriers, molecular stability, and elemental abundance.

Principal Research Objectives

Computational Framework

• Ground- and excited-state density-functional theory
• Multireference electronic-structure methods
• Reaction-pathway and transition-state calculations
• Spin–orbit and non-adiabatic coupling calculations
• Surface-hopping molecular dynamics
• Multistate machine-learning potentials
• Uncertainty-aware active learning
• Automated high-throughput workflows
• Chemical-space exploration
• Multi-objective optimisation and inverse design

Long-Term Research Vision

The long-term aim is to establish an iterative discovery cycle in which quantum chemistry, excited-state dynamics, machine learning, and inverse molecular design work together to accelerate the transition from scarce Re-based photocatalysts towards efficient and sustainable Mn-based alternatives.