Data-oriented approach for pd-nanocatalyst design

Boucher, Alexandre 2024. Data-oriented approach for pd-nanocatalyst design. PhD Thesis, Cardiff University. Item availability restricted.

	PDF - Accepted Post-Print Version Restricted to Repository staff only until 20 February 2026 due to copyright restrictions. Download (17MB)
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (136kB)

Abstract

Palladium-based catalysts are employed in a large number of reactions, such as hydrogenation and multiple cross-coupling reactions. The development of highly active, selective, and robust catalysts is at the heart of the catalyst industry, and over the past decade, the increase in computational capacity proved a driving force for the accelerated discoveries of catalysts. The present thesis starts by introducing the reader to some of the challenges chemists face in the development of high-performance catalysts, namely the sintering problem and the metastable ensemble. The method employed in the present thesis to overcome these challenges, e.g. the data-oriented approach, is also introduced. The subsequent chapters introduce different models, derived from atomistic simulations, starting with the prediction of the surface energy. Surface energy is a key thermodynamic parameter of supported metal clusters, and the sintering driving force. The proposed model describes a new approach for the prediction of i) the surface energy of individual atoms on metal slabs and ii) the average surface energy of random metal gas-phase clusters. The resulting model is accurate and validated on nine different late transition metals. The kinetics of sintering were studied through the prediction of diffusion coefficients of single palladium. The atom was supported on different surfaces including Pd1/Pd(111), Pd1/Pd(001), and Pd1/SiO2(001). A probabilistic model associating the probability laws, the transition-state-theory, the Einstein’s equation of diffusion, and the density functional theory (DFT) calculations was developed. The diffusion of an adatom on both Pd surfaces compare well against the existing literature, and the results obtained on the silica surface demonstrate the model is applicable also on oxides supports surfaces. The diffusion coefficients, expressed as a function of temperature, provide useful information to experimentalists regarding the sintering rate and the strength of the interaction between the metal and its support. The thesis also investigates the problem arising from the nanoparticles metastable ensemble, namely, the exponential number of structures coexisting at a fixed number of atom in a metal particle. A bottom-up machine learning approach is introduced for the prediction of atomic energies and forces. Combining graph and classic feedforward neural networks, the energy free approach is first tested on Pd-pure gas-phase structures of size from 1 to 55 atoms, and AuPd-alloys gas-phase particles with a size in range 15 to 40 atoms succesfully. The proposed approach is then tested on Pdn/SiO2(001) clusters, with n = 1 – 8 atoms with good accuracy, providing a reliable DFT-derived forcefield that can be used during atomic simulations. Such forcefields proved a powerful tool for the exploration of the metastable ensemble, where classic computational methods , e.g. DFT, are overwhelmed by the computational cost associated to the number of structures that must be explored.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Schools > Chemistry
Date of First Compliant Deposit:	20 February 2025
Last Modified:	20 Feb 2025 14:59
URI:	https://orca.cardiff.ac.uk/id/eprint/176354

Actions (repository staff only)

Edit Item