[Paper Review] Machine Learning, Density Functional Theory, and Experiments to Understand the Photocatalytic Reduction of CO$_2$ by CuPt/TiO$_2$

Vaidish Sumaria, Takat B. Rawal+7 more|arXiv (Cornell University)|Feb 14, 2024

Machine Learning in Materials ScienceMaterials Science3 citations

TL;DR

This study integrates machine learning, density functional theory (DFT), and experiments to uncover the mechanism of CO₂ photocatalytic reduction on CuPt/TiO₂. A bias-corrected machine learning interatomic potential enables efficient basin-hopping Monte Carlo simulations, revealing that the CuPt/TiO₂ interface stabilizes key intermediates (*CO₂, *CO, *CH, *CH₂), enhancing methane selectivity—validated by synthesis and gas chromatography showing ~92% CH₄ selectivity.

ABSTRACT

The photoconversion of CO$_2$ to hydrocarbons is a sustainable route to its transformation into value-added compounds and, thereby, crucial to mitigating the energy and climate crises. CuPt nanoparticles on TiO$_2$ surfaces have been reported to show promising photoconversion efficiency. For further progress, a mechanistic understanding of the catalytic properties of these CuPt/TiO$_2$ systems is vital. Here, we employ $ extit{ab-initio}$ calculations, machine learning, and photocatalysis experiments to explore their configurational space and examine their reactivity and find that the interface plays a key role in stabilizing *CO$_2$, *CO, and other CH-containing intermediates, facilitating higher activity and selectivity for methane. A bias-corrected machine-learning interatomic potential trained on density functional theory data enables efficient exploration of the potential energy surfaces of numerous CO$_2$@CuPt/TiO$_2$ configurations via basin-hopping Monte Carlo simulations, greatly accelerating the study of these photocatalyst systems. Our simulations show that CO$_2$ preferentially adsorbs at the interface, with C atom bonded to a Pt site and one O atom occupying an O-vacancy site. The interface also promotes the formation of *CH and *CH$_2$ intermediates. For confirmation, we synthesize CuPt/TiO$_2$ samples with a variety of compositions and analyze their morphologies and compositions using scanning electron microscopy and energy-dispersive X-ray spectroscopy, and measure their photocatalytic activity. Our computational and experimental findings qualitatively agree and highlight the importance of interface design for selective conversion of CO$_2$ to hydrocarbons.

Motivation & Objective

• To understand the mechanistic basis of high photocatalytic activity and methane selectivity in CuPt/TiO₂ systems.
• To overcome the computational bottleneck of exploring large configurational spaces in bimetallic/CO₂/TiO₂ systems.
• To develop a bias-corrected machine learning interatomic potential that improves accuracy for surface and interface atoms critical to catalysis.
• To validate computational predictions through synthesis and experimental measurement of photocatalytic CO₂ conversion.
• To establish design principles for metal/oxide interfaces that enhance CO₂ activation and selectivity toward hydrocarbons.

Proposed method

• A bias-corrected machine learning interatomic potential (MLIP) is trained on DFT data to enable accurate, efficient exploration of potential energy surfaces.
• Basin-hopping Monte Carlo (BHMC) simulations are used to sample diverse CO₂@CuPt/TiO₂ configurations, accelerating exploration of complex configurational spaces.
• A modified loss function with weighted force errors and Gaussian density function (GDF)-based sampling bias correction improves model training for underrepresented atomic environments.
• The MLIP is trained using symmetry functions (G) to encode local atomic environments within a cutoff radius, ensuring invariance to rotation and translation.
• Experimental synthesis of CuPt/TiO₂ via photo-deposition on P25 TiO₂ is performed, followed by gas chromatography analysis of photocatalytic products.
• The model’s predictions are validated by comparing simulated intermediate stabilities and adsorption sites with experimental activity and selectivity.

Experimental results

Research questions

• RQ1 How does the CuPt/TiO₂ interface influence the adsorption and stabilization of CO₂ and key reaction intermediates?
• RQ2 What role do oxygen vacancies and metal nanoparticle composition play in enhancing CO₂ activation and selectivity for methane?
• RQ3 Can a bias-corrected machine learning interatomic potential accurately predict the potential energy surface of complex CO₂@CuPt/TiO₂ systems at reduced computational cost?
• RQ4 To what extent do computational predictions of intermediate stability correlate with experimentally observed photocatalytic activity and selectivity?
• RQ5 How can machine learning models be optimized to reduce error in predicting forces for low-abundance, catalytically critical atomic environments?

Key findings

• CO₂ preferentially adsorbs at the CuPt/TiO₂ interface with the C atom bonded to a Pt site and one O atom occupying an O-vacancy site.
• The interface stabilizes *CO₂, *CO, *CH, and *CH₂ intermediates, which collectively promote higher activity and selectivity toward methane.
• The machine learning interatomic potential achieves high accuracy (sub-1 meV/atom energy, <0.1 eV/Å force error) while enabling orders-of-magnitude faster sampling than DFT.
• Experimental synthesis of CuPt/TiO₂ yielded a methane selectivity of approximately 92%, consistent with computational predictions of high CH₄ activity.
• The bias-corrected loss function significantly improves model performance on low-frequency atomic environments, especially surface and interface atoms critical to catalysis.
• The combined computational and experimental results demonstrate that interface engineering is pivotal for selective CO₂-to-hydrocarbon conversion.