A transition toward a circular carbon economy, in which carbon is continually recycled rather than emitted, represents a promising route to mitigate carbon emissions, particularly in energy and chemical sectors. Such a transformation relies on continued advances in catalyst developments, which underpins more than 85% of current chemical and fuel production processes. Single-atom catalysts (SACs), featuring isolated and atomically dispersed active metal centers, offer exceptional activity, selectivity, and atom efficiency, thereby reducing material costs and maximizing catalytic utilization. Metal–organic frameworks (MOFs) are ideal porous supports for SACs, owing to their well-defined crystalline architectures, tunable pore structures, and improved chemical and thermal stability. Consequently, MOF-supported SACs have emerged as a rapidly growing class of catalysts and a model platform for probing fundamental catalytic phenomena at the atomic scale.
Computational catalysis provides molecular insight into active site structures, reaction mechanisms, and material design principles, offering perspectives often inaccessible to experimental approaches. ... mehrCatalysis in MOFs, like in other heterogeneous systems, inherently spans multiple length and time scales, necessitating carefully benchmarked multiscale methodologies to achieve accurate thermodynamic and kinetic descriptions. A hierarchical, bottom-up strategy, linking quantum-mechanical, molecular-dynamics, and kinetic models, offers a robust framework to bridge these scales. This thesis addresses two central challenges in this direction: obtaining accurate reaction energetics and thermodynamic properties.
The first part investigates the activity of two Ni-based active-site motifs for ethylene dimerization via the Cossee–Arlman (linear insertion) mechanism. The more active configuration is identified as a NiIIO2-type site, which provides enhanced accessibility for ethylene coordination. The C–C coupling step is found to be rate-determining, yielding butene as the predominant product. The NU-1000 framework facilitates chain growth preferentially within its smaller trigonal pores as compared to the hexagonal mesopores. Truncated cluster models reproduce periodic energetics within 4 kJ mol−1, indicating these models can be used to provide computational efficiency without compromising accuracy. Free energy profile corrected to the TPSSh level predicts turnover frequencies (TOFs) in close agreement with experiment. Insights from the Ni-based system are extended to the Cr-NU-1000 catalyst, revealing a tetrahedral CrIII active site responsible for ethylene polymerization. The polymerization follows a Cossee–Arlman pathway with insertion barriers as low as 50 kJ mol−1 and chain-termination barriers exceeding these by more than 30 kJ mol−1, consistent with experimental observations.
In the second part, cost-efficient machine-learned force fields (MLFFs) are implemented to accelerate phase space sampling in two rigorous free energy estimation frameworks: 𝜆-thermodynamic integration (𝜆-TI) and the intermediate hard-sphere model (IHSM). This approach bridges the time-scale limitations of conventional DFT-MD while preserving DFT-level accuracy. Complementary statistical techniques, including free energy perturbation (FEP), machine learning thermodynamic perturbation theory (MLPT), Bennett’s acceptance ratio (BAR), and the phase-space overlap index, 𝐼𝑤, are used to assess MLFF-derived results and correct them to the desired DFT accuracy. The robustness of these MLFF-aided workflows is meticulously validated, and the workflows were applied to diverse adsorption phenomena, namely the chemisorption of oxygenated species, and the physisorption and dissociative adsorption of methane on Pt(111). The atomic oxygen adsorption exhibited nearly harmonic behavior and was reasonably described by the HA, whereas OH and OOH adsorption displayed strong anharmononicity, with the MLFF-TI method recovering up to 10% of the entropy loss estimated under HA. For CH4 adsorption, the influence of density functionals were evaluated. BEEF-vdW, which tends to underestimate adsorption strength, predicted higher anharmonic contributions relative to PBE-D3(BJ). At high temperatures or within the physisorption regime, where the adsorbate became more mobile, the impact of anharmonicity was found to be nearly independent of the employed density functionals. Using the MLFF-IHSM approach, the experimental adsorption entropy for CH4 physisorption was accurately reproduced at both theory levels, while the adsorption entropies from both levels converged at 700 K for dissociative adsorption process. This part highlights the limitations of the harmonic approximation and emphasize the necessity of employing more accurate methods that can reliably capture anharmonic effects to improve the estimation of adsorption free energies and their derived thermodynamics and kinetics. Finally, a proof-of-concept demonstration shows that DFT-accurate anharmonic free energies can be achieved using a pretrained foundation MLFF model with only 200 additional DFT single-point evaluations, paving the way for transferable and scalable applications to MOF systems using these general-purpose models.
