UCLA

Transition metals play a key role as catalysts in applications ranging from chemical and energy production to environmental remediation. Ample evidence indicates that catalyst structures do not remain unchanged during use. Instead, substantial restructuring occurs due to the rearrangement of the metal atoms to make new structures. This restructuring has major consequences on catalytic properties, sometimes beneficial and sometimes not. Understanding how surface-active sites evolve under reaction conditions is of central importance to designing improved catalysts but presents experimental and theoretical challenges. In this dissertation, we use Density Functional Theory (DFT) with Machine Learning based interatomic potentials, atomistic thermodynamics, and global optimization in combination with in-situ experimental observations to uncover the mechanism of such restructuring processes. We primarily focus our attention to CO-induced Pt reconstruction, but also discuss the on-going work on Cu restructuring. We developed an adsorbed CO bond length-based correction to solve the GGA-level DFT errors in describing the Pt/CO system. This corrected both the site prediction and adsorption energy relative to experiments and helped reproduce experimental in-situ STM imaging results at high coverage. Using a NN-based potential trained on a large number of reference DFT data, we perform large scale global optimizations to understand CO organization on stepped Pt surfaces. (111) steps lead to formation of quasi-hexagonal structure of CO on terrace with a 2:1 ratio of top:bridge sites occupied, while (100) steps lead to a majority of multiply bonded CO on the terrace. By improving this NN-based potential further, we explored the mechanism of step reconstructing. High CO coverage at a Pt step edge triggers the formation of atomic protrusions which then detach from the step edge to create subnano-islands on the terraces. The under-coordinated sites on these islands are stabilized by the strongly bound CO adsorbate. Using the computational study, we discover that small (<12 atoms) islands are metastable, while islands with 12 atoms and more are thermodynamically stable, in agreement with the experimental observation. Finally using a similar approach, we have developed an accurate NN-based potential for Cu/CO system which we are using to understand its reconstruction in operando conditions.