An optimal Fe–C coordination ensemble for hydrocarbon chain growth: a full Fischer–Tropsch synthesis mechanism from machine learning

Fischer–Tropsch synthesis (FTS, CO + H(2) → long-chain hydrocarbons) because of its great significance in industry has attracted huge attention since its discovery. For Fe-based catalysts, after decades of efforts, even the product distribution remains poorly understood due to the lack of information on the active site and the chain growth mechanism. Herein powered by a newly developed machine-learning-based transition state (ML-TS) exploration method to treat properly reaction-induced surface reconstruction, we are able to resolve where and how long-chain hydrocarbons grow on complex in situ-formed Fe-carbide (FeC(x)) surfaces from thousands of pathway candidates. Microkinetics simulations based on first-principles kinetics data further determine the rate-determining and the selectivity-controlling steps, and reveal the fine details of the product distribution in obeying and deviating from the Anderson–Schulz–Flory law. By showing that all FeC(x) phases can grow coherently upon each other, we demonstrate that the FTS active site, namely the A-P5 site present on reconstructed Fe(3)C(031), Fe(5)C(2)(510), Fe(5)C(2)(021), and Fe(7)C(3)(071) terrace surfaces, is not necessarily connected to any particular FeC(x) phase, rationalizing long-standing structure–activity puzzles. The optimal Fe–C coordination ensemble of the A-P5 site exhibits both Fe-carbide (Fe(4)C square) and metal Fe (Fe(3) trimer) features.