[Paper Review] Atomistic evolution of active sites in multi-component heterogeneous catalysts

Cameron J. Owen, Lorenzo Russotto+5 more|arXiv (Cornell University)|Jul 18, 2024

Catalysis and Oxidation Reactions6 citations

TL;DR

This study develops and validates a single, equivariant machine-learned force field (MLFF) for Pd–Au catalysts, enabling atomistic, time-resolved simulations of bulk, surfaces, and nanoparticles under thermal and hydrogen environments to reveal active-site evolution and deactivation mechanisms.

ABSTRACT

Multi-component metal nanoparticles (NPs) are of paramount importance in the chemical industry, as most processes therein employ heterogeneous catalysts. While these multi-component systems have been shown to result in higher product yields, improved selectivities, and greater stability through catalytic cycling, the structural dynamics of these materials in response to various stimuli (e.g. temperature, adsorbates, etc.) are not understood with atomistic resolution. Here, we present a highly accurate equivariant machine-learned force field (MLFF), constructed from ab initio training data collected using Bayesian active learning, that is able to reliably simulate PdAu surfaces and NPs in response to thermal treatment as well as exposure to reactive H$_2$ atmospheres. We thus provide a single model that is able to reliably describe the full space of geometric and chemical complexity for such a heterogeneous catalytic system across single crystals, gas-phase interactions, and NPs reacting with H$_2$, including catalyst degradation and explicit reactivity. Ultimately, we provide direct atomistic evidence that verifies existing experimental hypotheses for bimetallic catalyst deactivation under reaction conditions, namely that Pd preferentially segregates into the Au bulk through aggressive catalytic cycling and that this degradation is site-selective, as well as the reactivity for hydrogen exchange as a function of Pd ensemble size. We demonstrate that understanding of the atomistic evolution of these active sites is of the utmost importance, as it allows for design and control of material structure and corresponding performance, which can be vetted in silico.

Motivation & Objective

• Motivate the need for atomistic understanding of dynamic multi-component catalysts under realistic operating conditions.
• Develop a unified, accurate ML force field capable of describing Pd–Au bulk, surface, and nanoparticle behavior under heat and H2 exposure.
• Demonstrate the MLFF's accuracy against DFT across compositions, structures, and reactive environments.
• Use ML-MD to predict active-site evolution, Pd segregation, and nanoparticle morphology under pretreatment and reaction conditions.
• Provide a pathway for in silico catalyst design through atomistic control of active-site ensembles.

Proposed method

• Combine FLARE active learning to generate diverse ab initio training data for Pd–Au systems.
• Train an equivariant Allegro MLFF with angular cutoff and pairwise interactions optimized for Pd–Au-H chemistry.
• Validate MLFF against DFT for bulk, surface, and nanoparticle properties across compositions.
• Perform ML-MD simulations to study annealing, Pd segregation, surface ensemble distributions, and reactivity under H and H2.
• Compute reaction barriers and pre-exponential factors for H2 dissociation and H recombination on various PdAu ensembles to compare with experiments.

Experimental results

Research questions

• RQ1 What is the atomistic evolution of active-site ensembles in Pd–Au catalysts under thermal treatment and H2 exposure?
• RQ2 Can a single MLFF accurately describe bulk, surface, and nanoparticle Pd–Au systems across compositions and reactive environments?
• RQ3 How do Pd distribution, surface coverage, and nanoparticle morphology evolve during annealing and how do these affect catalytic reactivity?
• RQ4 What are the activation energies and pre-exponential factors for H2 dissociation and recombination on dilute Pd ensembles in Au, and how do they compare with experiments?

Key findings

• An equivariant MLFF (Allegro) trained via FLARE accurately reproduces DFT energies, formation energies, and surface energies across Pd–Au compositions, surfaces, and hydride phases (errors < 10 meV/atom).
• ML-MD with the MLFF captures Pd segregation into subsurface layers and predominantly monomeric Pd surface ensembles after annealing, across NP sizes (1–2 nm) and compositions (4–30% Pd).
• Simulated NP evolution shows pentatwinned symmetry emergence and alloying behavior consistent with experimental observations, including Pd distribution and surface ensemble statistics.
• Adsorption energies and H2 dissociation/recombination barriers on Pd–Au surfaces predicted by MLFF align with DFT results (activation energies within ~2 meV/atom for many states; E_act values tabulated for multiple Pd–Au configurations).
• For HD exchange, simulated activation energies and pre-exponential factors span multiple Pd–Au configurations, enabling direct comparison to experimental values (see Table 2).
• The dominant catalytic active sites under operating conditions are dynamic and can be identified in silico, enabling design/control of catalyst structure and performance.