A Framework for Catalytic Selectivity via Reactive Intermediate Networks

Thesis Abstract

In heterogeneous catalysis, unpredictable product selectivity often arises from competing reactive intermediate pathways that complicate mechanistic control and scale-up, leading to inefficiencies and undesired byproducts. This study develops an integrative framework—Reactive Intermediate Network Theory (RINT)—to elucidate and predict catalytic selectivity by mapping, quantifying, and constraining the network of transient species within catalytic cycles. The aim is to (i) formalize a node-and-edge model of reactive intermediates across representative catalytic systems, (ii) identify principal network topology features that govern selectivity, and (iii) implement a predictive toolkit that combines kinetic modeling with probabilistic inference to guide catalyst design. Specific objectives include (a) cataloguing common reactive intermediates (e.g., carbocations, metal-alkyls, radicals) and their interconversions in hydrocarbon hydrogenation and oxidation catalysts; (b) deriving a mechanistic hierarchy of rate-determining and branch-point steps through sensitivity analysis and global optimization; (c) developing a Bayesian network-augmented microkinetic model that integrates density functional theory (DFT) energetics with experimental rates to infer path probabilities; (d) validating the framework against three case studies—selective hydrogenation of dienes over Pd/C, oxidative coupling on Cu-based catalysts, and hydroformylation-selective carbonylation over Rh-based systems; and (e) delivering a decision-support dashboard that predicts product distributions under varying reaction conditions. Methodologically, the research adopts a mixed-methods design anchored in systems chemistry and catalysis theory. The population comprises published kinetic datasets and high-fidelity experimental measurements from three catalytic platforms, with a sample of 60 publicly available rate datasets and 20 in-house experiments conducted under carefully controlled conditions. Data collection employs high-pressure microreactor experiments, operando infrared and Raman spectroscopy for intermediate identification, and in situ X-ray absorption spectroscopy to corroborate oxidation and coordination states. Instrumentation includes GC-MS for product quantification, LC-MS for intermediate profiling, and DRIFTS for surface species. For theoretical insight, DFT calculations (PBE0-D3) are used to estimate activation barriers for key steps, while microkinetic modeling employs differential-algebraic equation solvers to simulate steady-state coverages and branching ratios. The analysis framework uses Bayesian inference to update intermediate probabilities with experimental data, global sensitivity analysis to rank pathway impact, and Markov state modeling to capture temporal evolution of the intermediate network. Validation proceeds through cross-system comparison, posterior predictive checks, and out-of-sample testing on the three case studies. Expected findings include (i) identification of universal network motifs—such as high-probability bottlenecks at branched steps and low-probability dead-end routes—that robustly govern selectivity across catalysts; (ii) quantitative links between network topology metrics (e.g., node centrality, branching entropy, and path diversity) and observed selectivity; (iii) a calibrated predictive model capable of estimating product distributions within ±5–10 mol% under defined temperature and pressure windows; and (iv) validated guidelines for catalyst engineering, such as tuning active-site geometry to suppress deleterious intermediates or redesigning support interactions to alter network connectivity. The study contributes to knowledge by providing a formalized, transferable theory that connects reactive intermediate dynamics with macroscopic selectivity, bridging microkinetics, network theory, and operando spectroscopy. It demonstrates a generalizable methodology for incorporating intermediate-level information into predictive catalysts design, enabling more efficient exploration of catalyst libraries and reaction conditions. The main conclusion anticipates that catalytic selectivity can be anticipated and steered through deliberate modulation of intermediate networks rather than solely by optimizing global reaction energetics. Consequently, recommendations include adopting RINT-informed catalyst screening pipelines, integrating Bayesian network models into standard microkinetic analyses, and expanding operando measurement campaigns to capture transient species critical to network topology.