Loading field map
For over a century, chemical engineers have faced a persistent puzzle: how to make chemical reactions faster and more selective without being consumed themselves. Catalysis—the acceleration of reactions by substances that remain unchanged—has been the central answer, but the way engineers think about catalysts has undergone repeated transformations. Each major framework in catalysis emerged because the previous way of thinking hit a limit: it could not explain why certain catalysts worked, could not predict new ones, or could not connect atomic-scale events to industrial performance. Today, eight frameworks coexist, each with its own tools, assumptions, and domain of application, and their relationships—replacement, coexistence, absorption, and productive tension—define the subfield.
The earliest framework was not a theory but a practice. From the 1880s through the mid-twentieth century, catalyst development was driven by trial and error, guided by intuition and systematic variation of recipes. Industrial giants such as the Haber–Bosch process for ammonia synthesis and the Fischer–Tropsch process for hydrocarbon fuels were discovered this way. Paul Sabatier's work on hydrogenation, for which he won the 1912 Nobel Prize, exemplified the empirical approach: he identified that certain metals catalyzed hydrogenation reactions, but the explanation remained at the level of observation. The framework's strength was its direct industrial impact; its weakness was that it offered no mechanistic understanding. A catalyst that worked was simply a good catalyst, and there was no way to know why another formulation failed. This lack of explanatory power created the pressure for a more quantitative approach.
Irving Langmuir's 1916 paper on the adsorption of gases on plane surfaces introduced a radical shift: catalysis could be understood through the behavior of molecules on surfaces. Langmuir proposed that a solid catalyst surface consists of a finite number of equivalent sites, each of which can adsorb one molecule, and that adsorption and reaction rates follow from the fraction of occupied sites. This gave rise to the Langmuir isotherm and, later, Langmuir–Hinshelwood kinetics, which modeled surface reactions as sequences of elementary steps. For the first time, engineers could write rate equations based on assumed surface mechanisms and fit them to experimental data. The framework's core assumption—that all surface sites are identical and independent—was a deliberate simplification that made kinetic modeling tractable. It remains in use today for many industrial reactions, but its limitations became clear as evidence accumulated that real surfaces are far more complex.
While heterogeneous catalysis dominated industry, a parallel tradition emerged in which the catalyst and reactants were in the same phase. The discovery of Wilkinson's catalyst (tris(triphenylphosphine)rhodium chloride) in the 1960s demonstrated that soluble organometallic complexes could hydrogenate olefins with remarkable selectivity under mild conditions. This framework introduced a new unit of analysis: the molecular catalyst, whose activity could be tuned by modifying the ligands around a metal center. Homogeneous catalysis did not replace heterogeneous catalysis; rather, it coexisted with it, occupying a different niche. Homogeneous catalysts excel at reactions requiring high selectivity—asymmetric synthesis, for example—while heterogeneous catalysts remain preferred for large-scale, high-temperature processes. The organometallic insights later fed back into heterogeneous catalysis through concepts such as single-atom catalysts, where isolated metal atoms on a support mimic the well-defined active sites of homogeneous complexes.
Nature had solved the selectivity problem long before chemists did. Enzymes—protein catalysts evolved for specific biochemical transformations—operate with extraordinary rate enhancements and substrate specificity. The Michaelis–Menten model, developed earlier but widely adopted in the 1960s, provided a kinetic framework analogous to Langmuir–Hinshelwood kinetics but with a crucial difference: the active site is a three-dimensional pocket that binds the substrate through multiple weak interactions, not a flat surface site. Enzyme biocatalysis expanded from a niche biochemical curiosity to an industrial tool as engineers learned to immobilize enzymes for continuous operation and, later, to engineer them through directed evolution. The framework's distinctive contribution was to demonstrate that catalysis could be understood at the level of individual active-site geometry and binding energy, a lesson that later frameworks would absorb.
By the 1960s, evidence was mounting that the Langmuir assumption of uniform sites was wrong. Michel Boudart's landmark 1966 paper on the specific activity of platinum catalysts showed that the turnover frequency—the number of reactions per surface atom per second—depended on the size and structure of metal particles. Some reactions were "structure-sensitive," meaning their rates changed with particle size, while others were "structure-insensitive." This finding challenged the entire framework of uniform-site kinetics. Surface science techniques—low-energy electron diffraction, X-ray photoelectron spectroscopy, scanning tunneling microscopy—allowed researchers to probe catalyst surfaces at atomic resolution, revealing steps, kinks, and terraces with different reactivities. The framework's core insight was that catalytic activity is not a bulk property but a local one, determined by the atomic arrangement at the surface. This created a direct bridge to computational methods: if the structure of active sites could be observed, it could also be simulated.
Density functional theory (DFT) made it possible to calculate adsorption energies, activation barriers, and reaction pathways from first principles, without relying on experimental fitting. Computational catalysis emerged in the 1990s as a framework that could predict the behavior of catalytic surfaces at the atomic scale. Its dependence on structure-sensitive catalysis was explicit: DFT calculations require an atomic model of the active site, and the surface-science framework provided those models. The method's power is that it can screen hypothetical catalysts—alloy compositions, dopants, facet orientations—before any synthesis is attempted. Its limitation is that DFT calculations are computationally expensive and approximate; they work best for well-ordered surfaces and simple reactions, and they struggle with the complexity of real catalysts under operating conditions.
Computational catalysis produces energies, but engineers need rates. Microkinetic modeling, formalized by James Dumesic and colleagues in the early 1990s, bridges this gap. A microkinetic model constructs a detailed reaction mechanism from elementary steps—adsorption, surface reaction, desorption—each with a rate constant derived from DFT or experimental data. The model then solves the coupled rate equations to predict overall reaction rates, selectivity, and sensitivity to operating conditions. Unlike Langmuir–Hinshelwood kinetics, microkinetic models do not assume a rate-determining step or a single site type; they can incorporate multiple site types and coverage-dependent energetics. The framework's role is infrastructure: it translates atomic-scale insights into reactor-scale predictions. It coexists with computational catalysis, relying on DFT for input but adding the kinetic complexity that DFT alone cannot capture.
The most recent framework, rational catalyst design, is not a single method but an integrative workflow that draws on all prior frameworks. Its premise is that catalysts can be designed from the bottom up by identifying descriptors—measurable or calculable properties that correlate with activity and selectivity—and then searching for materials that optimize those descriptors. The concept was demonstrated in the late 1990s with imprinted nanostructured materials, where molecular templates created specific binding sites, mimicking enzyme active sites. Today, rational design combines computational screening (from computational catalysis), microkinetic validation, surface-science characterization, and, increasingly, machine learning to navigate the vast space of possible catalyst compositions. What distinguishes it from earlier frameworks is its explicit goal of prediction: rather than discovering catalysts and then explaining them, rational design aims to specify a target and then synthesize it.
Today, computational catalysis, microkinetic modeling, and rational catalyst design are the most active frameworks, but they do not operate in isolation. They agree on a core principle: catalytic performance can be understood and predicted from atomic-scale properties. The disagreement is about how much prediction is possible. One camp holds that first-principles DFT, combined with microkinetic modeling, can eventually replace most experimental trial and error. The other camp argues that the complexity of real catalysts—dynamic surfaces, solvent effects, promoter interactions—makes complete first-principles prediction impractical, and that empirical optimization, guided by machine learning, will remain essential. This tension is productive: it drives the development of better approximations in DFT, more comprehensive microkinetic models, and data-driven methods that learn from high-throughput experiments. The older frameworks—empirical catalysis, adsorption kinetics, homogeneous catalysis, enzyme catalysis, and surface science—remain active in their own domains, providing the experimental data and mechanistic insights that the newer frameworks depend on. The subfield's history is not a story of one framework replacing another, but of successive layers of understanding, each adding new tools while preserving the practical achievements of its predecessors.