Catalysts are widely used to increase reaction rates. They function by stabilizing the transition state of the reaction at their active site, where the atomic arrangement ensures favourable interactions1. However, mechanistic understanding is often limited when catalysts possess multiple active sites—such as sites associated with either the step edges or the close-packed terraces of inorganic nanoparticles2,3,4—with distinct activities that cannot be measured simultaneously. An example is the oxidation of carbon monoxide over platinum surfaces, one of the oldest and best studied heterogeneous reactions. In 1824, this reaction was recognized to be crucial for the function of the Davy safety lamp, and today it is used to optimize combustion, hydrogen production and fuel-cell operation5,6. The carbon dioxide products are formed in a bimodal kinetic energy distribution7,8,9,10,11,12,13; however, despite extensive study5, it remains unclear whether this reflects the involvement of more than one reaction mechanism occurring at multiple active sites12,13. Here we show that the reaction rates at different active sites can be measured simultaneously, using molecular beams to controllably introduce reactants and slice ion imaging14,15 to map the velocity vectors of the product molecules, which reflect the symmetry and the orientation of the active site16. We use this velocity-resolved kinetics approach to map the oxidation rates of carbon monoxide at step edges and terrace sites on platinum surfaces, and find that the reaction proceeds through two distinct channels11,12,13: it is dominated at low temperatures by the more active step sites, and at high temperatures by the more abundant terrace sites. We expect our approach to be applicable to a wide range of heterogeneous reactions and to provide improved mechanistic understanding of the contribution of different active sites, which should be useful in the design of improved catalysts.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
A.M.W. acknowledges support from the Alexander von Humboldt Foundation. We acknowledge support from Deutsche Forschungsgemeinschaft (DFG) and the Ministerium für Wissenschaft und Kultur (MWK) Niedersachsen, and the Volkswagenstiftung under grant INST 186/952-1. C.T.C. acknowledges the Göttingen Academy of Sciences and the US National Science Foundation (grant CHE-1665077) for support.Reviewer information
Nature thanks E. Hasselbrink and the other anonymous reviewer(s) for their contribution to the peer review of this work.
Extended data figures and tables
a, b, Ion images of the product CO2. a, Raw data after background subtraction, with the intensity of each pixel proportional to the density of CO2. The pixel distance from the laser position (dashed red line) is proportional to the velocity, that is, velocity increases to the right. b, Data after multiplying each pixel by its corresponding velocity, with the resulting pixel intensity proportional to the flux and the hyperthermal channel then becoming more readily apparent. The red and green rectangles indicate typical velocity integration windows used to produce kinetic traces for the thermal and hyperthermal channels. The vertical span of the boxes subtends ±3° around the surface normal. This small angular range justifies using rectangular integration windows. For defining kinetic traces, we take the average intensity within the rectangles as the product flux normal to the surface. c, d, Raw kinetic trace data. c, Average density of the CO2 product integrated over the blue (thermal channel) and red (hyperthermal channel) rectangular areas of the ion image (inset). d, Same data converted to flux. This kinetic trace and the ion images above were measured at 350 °C, a time-averaged CO flux of 2.2 × 1012 s−1 cm−2 and a time-averaged O2 flux of 1.1 × 1013 s−1 cm−2.
a, Product speed distribution. The branching ratio between the thermal and hyperthermal channels is obtained from this distribution, calculated from the ion image in Extended Data Fig. 1. The thermal product channel is fit to a Maxwell–Boltzmann function (Ttrans = 483 K; dashed blue line) and the hyperthermal channel is fit to a flowing Maxwell–Boltzmann function42 (Ttrans = 894 K, α = 190 meV; dashed red line). The velocities used for extraction of the kinetic traces are indicated by the hatched areas. b, The angular distribution for the ion image shown in c. c, The flux-corrected ion image that shows an angular distribution subtending ±25° around the surface normal. The angular distribution was measured at a surface temperature of 400 °C. d, The kinetic trace. We obtain the product flux as a function of reaction time for two channels with different speed and angular distributions (details are given in the text). It is now clear that the hyperthermal channel is much weaker than the thermal channel.
a, Kinetic traces of the thermal and the hyperthermal channel fitted by a convolution of an exponential decay over the incoming beam. b, Kinetic model for the effective lifetime of the thermal channel of CO oxidation. c, Effective lifetimes plotted against time-averaged O2 flux. Solid circles are experimental points for . The y-intercept is non-zero owing to CO desorption. In these measurements, the time-averaged CO flux was 2.2 × 1012 s−1 cm−2.
Rate constants for desorption of CO from Pt(111) terraces (black and red triangles) are taken from ref. 43. The plus signs are rate constants for desorption of CO from steps; see ref. 18. The filled red circles show rate constants for desorption of CO from steps as measured previously27. The zero-coverage rate constants from Extended Data Fig. 3 are shown as filled blue circles.
The transition state resembles reactants; it is a so-called ‘early-barrier’ reaction. Such reactions channel the energy released in the reaction primarily into product vibration. Experimentally we observe that approximately 20% of the barrier-height energy (0.38 eV) appears as product transitional energy in the hyperthermal channel. We take this as evidence that the hyperthermal reaction takes place on platinum terraces.
Extended Data Fig. 6 Examples of the kinetic model fit to experimentally derived kinetic traces for Pt(111).
Nine out of 18 (every second) kinetic traces at 340 °C are shown. Similar fit quality was obtained for 18 plots for seven temperatures between 290 and 350 °C. Information on the total oxygen coverage [Oa], the fractional coverage on steps and terraces (θOT and θOS, respectively) and the time-averaged O2 flux is shown above each plot. The time-averaged CO flux was 2.2 × 1012 s−1 cm−2.
a, A comparison of the values of total adsorbed oxygen [Oa] obtained from titrations on Pt(111) with [Oa] values obtained from the numerical solution. The black squares with estimated uncertainty (1 s.d.) are the total amount of oxygen on the surface, [Oa], obtained from titration measurements. The red dots show [Oa] = [OTerr] + [OStep] obtained from the numerical solution at 340 °C. The time-averaged CO flux was 2.2 × 1012 s−1 cm−2. b, The fractional step coverage (θOS) on Pt(111) from the numerical solution compared to the partition function simulation. The black squares are the result of the numerical fit shown in Extended Data Fig. 6. All data are at 340 °C. c, A comparison of values of [Oa] obtained from titrations on Pt(332) with [Oa] values obtained from the numerical kinetic model. Note that the titration results are found to be independent of surface temperature under our conditions. d, The fractional step coverage (θOS) plotted against [Oa] for Pt(332). The red lines are results from the partition function calculated for different binding energy differences. The black triangles are the results from the numerical analysis of the kinetic model. Two extreme cases—no oxygen-atom binding preference for steps (dashed) and large oxygen-atom binding preference for steps (solid)—for the partition function are shown as black lines.
Extended Data Fig. 8 Examples of the kinetic model fit to experimentally derived kinetic traces for Pt(332).
All three reactive contributions are shown. Here TS = 320 °C and the oxygen coverage varies from 0.044 to 0.168 monolayers (denoted as [Oa]). The fractional coverage on terraces (θOT) and steps (θOS) is also indicated. The time-averaged O2 flux is stated above each kinetic trace, the time-averaged CO flux was 2.2 × 1012 s−1 cm−2.
a, Simulation of previous experiments. We show the kinetic trace integrated over velocity (hollow circles). The solid line shows an exponential fit to the simulated data. b, Previously reported activation energies7,41 (hollow circles) for CO oxidation that were based on product velocity unresolved measurements are compared to the results of this work when integrated over product velocity. See text for details.
Extended Data Fig. 10 Model predictions of CO oxidation on a Pt(111) crystal with a step density of 0.25%.
a, Total CO–CO2 conversion efficiency as a function of temperature and oxygen coverage. The yellow boxes indicate (at low temperature) where past studies have been carried out and (at high temperatures) where industrial catalysts are used. b, The relative importance of the hyperthermal (terrace) reaction as a function of temperature and oxygen coverage. c–e, Contour plots showing the total fit residual as a function of two activation energies. y against x: TT versus TS (c), TS versus SS (d), TT versus SS (e).