Adsorbate-induced structural evolution changes the mechanism of CO oxidation on a Rh/Fe3O4(001) model catalyst

The structure of a catalyst often changes in reactive environments, and following the structural evolution is crucial for the identification of the catalyst’s active phase and reaction mechanism. Here we present an atomic-scale study of CO oxidation on a model Rh/Fe3O4(001) “single-atom” catalyst, which has a very different evolution depending on which of the two reactants, O2 or CO, is adsorbed first. Using temperature-programmed desorption (TPD) combined with scanning tunneling microscopy (STM) and X-ray photoelectron spectroscopy (XPS), we show that O2 destabilizes Rh atoms, leading to the formation of RhxOy clusters; these catalyze CO oxidation via a Langmuir–Hinshelwood mechanism at temperatures as low as 200 K. If CO adsorbs first, the system is poisoned for direct interaction with O2, and CO oxidation is dominated by a Mars-van-Krevelen pathway at 480 K.


Introduction
Supported rhodium catalysts show high activity for a number of reactions, including CO oxidation. [1][2][3][4] It is generally accepted that Rh nanoparticles supported on rigid, irreducible substrates catalyze CO oxidation via the Langmuir-Hinshelwood (L-H) or Eley-Rideal (E-R) mechanisms, where the CO and O 2 molecules either coadsorb on the surface (L-H), or one of the molecules reacts directly from the gas phase with the other molecule adsorbed on the surface (E-R). [6][7][8][9][10][11][12] However, the reactivity of the system is often affected by the choice of support, and reducible supports enable an alternative pathway, the so-called Mars-van-Krevelen (MvK) mechanism. Here, oxygen is extracted from the metal oxide lattice to oxidise CO, and the lattice is repaired by O 2 from the gas phase. 14,[16][17][18][19] Metal-support interactions are not always beneficial, however, and encapsulation by the support can completely deactivate the catalyst. 13,14,20,21 Improving a catalyst's activity, selectivity, and stability requires the optimization of atomic-scale processes that are very difficult to ascertain on working catalysts. This motivates the use of welldefined model systems studied by the surface science approach, which can provide atomic-scale insights.
In this paper we present how CO oxidation proceeds on a Rh/Fe 3 O 4 (001) model "single-atom" catalyst. Using a combination of temperature-programmed desorption (TPD), scanning tunneling microscopy (STM), and X-ray photoelectron spectroscopy (XPS), we show the system is active for CO oxidation by both L-H and MvK mechanisms, depending on which of the reactants the surface is exposed to first. Oxygen adsorption leads to formation of superoxo (O 2 ) − species bound to Rh 1 adatoms, which rapidly agglomerate to small Rh x O y clusters. The clusters contain weakly bound oxygen, which reacts with CO below room temperature. In contrast, CO adsorption on the bare Rh 1 species poisons the system for O 2 adsorption, but CO oxidation nevertheless takes place via a MvK mechanism at ca. 480 K. Two different deactivation mechanisms are observed: the oxidised clusters become inactive after the first desorption of the weakly bound oxygen, while the Rh 1 adatoms incorporate into the support lattice as the CO desorbs.

Methods
The experiments were carried out using two independent ultra-high vacuum (UHV) setups. The STM data were acquired in a two-vessel UHV chamber consisting of a preparation chamber ( p < 10 −10 mbar) and an analysis chamber ( p < 7 × 10 −11 mbar) equipped with an Omicron μ-STM operated at room temperature in constant-current mode with electrochemically etched tungsten tips. The chamber further includes commercial XPS, Auger electron spectroscopy and low-energy electron diffraction instruments. The STM images were corrected for distortion and creep of the piezo scanner as described in ref. 22.
The TPD and XPS data were acquired in a second chamber equipped with a liquid-He cryostat, a home-built molecular beam source, a quadrupole mass spectrometer (Hiden HAL 3F PIC), a monochromatized Al/Ag twin anode X-ray source (Specs XR50 M, FOCUS 500), a hemispherical analyzer (Specs Phoibos 150), a low-energy electron diffraction setup (Specs ErLEED), an ion source (Specs IQE 12/38) and a UV source (Specs UVS 10/35). This chamber was specifically designed for surface chemistry studies of single-crystal oxide samples; full details are given in ref. 23. The samples were mounted on a Ta sample plate using Ta clips, with a thin Au foil placed between the sample plate and the sample to ensure good thermal contact. The sample plate was attached to the liquid-He cryostat via Ta rods, and the temperature was measured by a K-type thermocouple spot-welded on the sample plate. A molecular beam is formed by expansion of 0.53 mbar 18 O 2 or 0.27 mbar 13 CO through two differentially pumped stages. This results in a well-defined beam spot on the sample surface with a diameter of 3.35 ± 0.17 mm and a top-hat intensity profile. 23 The TPD spectra were acquired with a 1 K s −1 heating ramp.
All the experiments were conducted on natural Fe 3 O 4 (001) single crystals (SurfaceNet GmbH or Surface Preparation Laboratory) prepared by sputtering (1 keV Ar + or Ne + , 10 min) and annealing in UHV (930 K, 10 min). Every other annealing step was done in a partial pressure of O 2 ( p O 2 = 5 × 10 −7 mbar, 20 min) to avoid reduction of the surface. Rhodium was deposited using Focus e-beam evaporators, and the deposition rate was calibrated by temperature-stabilized quartz crystal microbalances (QCM). One monolayer (ML) is defined as one atom per (√2 × √2)R45°unit cell of the Fe 3 O 4 (001), which corresponds to 1.42 × 10 14 atoms per cm 2 .

Results
Rh 1 species on Fe 3 O 4 (001) Fig. 1a and b shows STM images of the Fe 3 O 4 (001) surface before and after deposition of Rh. The undulating rows running in the [110] direction correspond to 5-fold coordinated Fe atoms in surface octahedral sites (Fe oct , see model in the inset of Fig. 1a). In each pair of neighbouring terraces, the surface Fe oct row direction is rotated by 90°, as expected for the spinel structure (not shown here). Surface oxygen atoms are not visible in the STM images because they possess no density of states near the Fermi level, but their positions are precisely known from diffraction studies and density functional theory calculations. 24,25 This Fe 3 O 4 (001) surface reconstruction is based on subsurface cation vacancies (therefore it's referred to as the SCV reconstruction), and it has direction. The small bright species on the Fe oct rows are surface hydroxyls. 5 The inset shows a model of the Fe 3 O 4 (001) surface with the octahedral Fe oct , tetrahedral Fe tet and O atoms being represented by dark blue, light blue and red spheres, respectively. (b) After deposition of 0.2 ML Rh at room temperature, the majority species are Rh 1 adatoms (red circles), but we also observe slightly elongated species which we identify as Rh 2 dimers (cyan circles). (c and d) The Rh 2 dimers exhibit characteristic hopping motion in subsequent STM images (broken lines for guidance).
(e and f ) Scanning with a +3 V bias results in dissociation of the Rh 2 dimers (cyan arrows in panel e) into pairs of single Rh 1 adatoms ( paired cyan arrows in panel f ). The +3 V scan was done between acquisition of images shown in (e) and (f ). been shown previously to support dense arrays of single metal atoms, preventing the adatom mobility and cluster formation even at elevated temperatures. 15,24,[26][27][28] This is due to the existence of highly stable adsorption sites with a nearest-neighbour distance of 0.84 nm, in which the adatoms are 2-fold coordinated to the surface oxygen atoms, essentially occupying a bulk continuation tetrahedral site (schematic representation shown in the uppermost inset of Fig. 2a). After deposition of Rh at room temperature, isolated Rh 1 adatoms appear as bright protrusions (red circles in Fig. 1b) sitting between the Fe oct rows. In addition to the single Rh 1 adatoms, we also observe larger elongated species (cyan circle in Fig. 1b), which in subsequent STM scans exhibit a distinctive hopping motion over the adsorption site ( Fig. 1c and d). We identify these species as metastable Rh 2 dimers that can be split into two Rh 1 adatoms by bias pulsing or high-bias scanning. Fig. 1e and f shows this process on two STM images acquired on the same spot: in Fig. 1e two Rh 2 dimers are observed (cyan arrows), Fig. 1f shows pairs of isolated Rh 1 adatoms (smaller paired arrows) appearing instead of the Rh 2 after the surface was scanned with a +3 V bias. The observed hopping of the Rh 2 is restricted to the given adsorption site, thus it might be also considered a restructuring of the dimer and does not lead to destabilization of the system by diffusion or agglomeration. Apart from the dimer hopping, no mobility of Rh species is observed in subsequently acquired STM images at room temperature under UHV conditions. The Fe 3 O 4 (001) surface supports the Rh 1 adatoms even at significantly higher coverages; At elevated temperatures, the Rh 1 adatoms are unstable against incorporation into the Fe 3 O 4 (001) support. Such a phenomenon has been observed for several different metal adatoms on Fe 3 O 4 (001) before and the foreign adatom can either substitute a 5-fold coordinated surface Fe oct or incorporate in the vacant octahedral site of the SCV reconstruction. [13][14][15]29 Schematic representations of the three possible sites of Rh 1 near the surface are shown in insets in Fig. 2a. The incorporation of Rh takes place at temperatures above ≈450 K and can be followed both in STM and XPS, as shown in Fig. 2a and b. STM images taken after annealing to 500 K show predominantly elongated (orange circle) and circular ( purple circle) features on the Fe oct rows, which is consistent with Rh 1 incorporated into the subsurface 6-fold site or surface 5-fold site, respectively (STM simulations of different metal adatoms in the same sites are provided in ref. 14, 15 and Fig. S2 †). Additionally, we observe an increased number of bigger brighter features, presumably Rh clusters, and a very small number of remaining 2-fold Rh 1 adatoms (red circle). Fig. 2b shows the Rh 3d region of XPS spectra taken after depositing 0.6 ML Rh and heating to different temperatures. After deposition at 300 K, the Rh 3d peak maximum is located at ≈307.7 eV, but its broad shape suggests convolution of several components. Apart from the main signal corresponding to 2-fold Rh 1 , the additional XPS components might come from the Rh 2 dimers or Rh 1 adatoms residing in the vicinity of surface defects or step edges. On different samples, small variations of the peak maximum between 307.7 and 308.1 eV were observed, most likely depending on the relative Only a very small number of Rh 1 remain in the 2-fold coordinated site between the Fe oct rows (red colour). We identify the new species as Rh 1 incorporated in the octahedral cation sites near the surface, consistent with previously reported studies of various metal adatoms on Fe 3 O 4 (001). 14,15 Schematic representations of these sites are shown in the insets and labelled with the color of the corresponding circle. (b) Rh 3d region of the XPS spectra (monochromatized Al Kα, grazing emission) taken after heating to different temperatures (measured after cooling back to 300 K). Upon heating to 400 K the Rh 3d peak maximum shifts to 308.0 eV, as the low binding-energy shoulder at 307.2 eV decreases. Heating above 450 K leads to an increase of the 309.5 eV component, which becomes dominant above 600 K. We attribute this component to Rh incorporated into the octahedral sites of the Fe 3 O 4 (001) surface. (c and d) The Fe 2p and O 1s peaks corresponding to the first and last spectra shown in (b). Notably, in the Fe 2p region the shoulder at ≈708.5 eV slightly increases upon Rh deposition, indicating an increased amount of Fe 2+ near the surface. This shoulder decreases again upon heating as the Rh cations diffuse into deeper layers. number of these species. After heating to 400 K we observe a decrease on the low-binding-energy side of the peak (around ≈307.2 eV) and the peak maximum shifts to 308.0 eV. Heating above 450 K leads to an increase of a higher-binding-energy component at 309.5 eV and decrease of the signal at 308.0 eV. After heating to 650 K, the 308.0 eV peak is no longer observed, and the 309.5 eV component is dominant. Prolonged heating above this temperature leads to a decrease and eventual disappearance of the Rh signal in XPS, consistent with diffusion of Rh into the Fe 3 O 4 bulk (Fig. S3 †). Thus, we attribute the 309.5 eV component to Rh incorporated into the Fe 3 O 4 (001) substrate.
The Fe 2p and O 1s regions only show small differences with Rh deposition and heating ( Fig. 2c and d). Upon Rh deposition, an Fe 2p shoulder at ≈708.5 eV slightly increases in comparison to the clean surface, indicating an increased amount of Fe 2+ near the surface. 30 This shoulder decreases again upon heating, presumably because the Rh incorporates and thus the number of near-surface cations decreases.

Exposing Rh 1 /Fe 3 O 4 (001) to O 2
When the Rh 1 /Fe 3 O 4 (001) system is exposed to O 2 at room temperature, the Rh 1 adatoms become mobile and begin to agglomerate. Fig. 3a-c shows three frames of an STM movie acquired while keeping a constant background pressure of 2 × 10 −10 mbar O 2 in the analysis chamber. Single adatoms (highlighted by red circles) form well-resolved oxidised Rh 2 O x dimers (yellow rectangles) and subsequently bigger Rh x O y clusters ( purple rectangle). Over the course of the STM movie, no O 2 adsorption was observed on the Rh 2 dimers (cyan circle).
After dosing a higher amount of O 2 in the preparation chamber (50 Langmuir, where 1 L = 1.33 × 10 −6 mbar s), the majority species observed in STM are small Rh x O y clusters, although some adatoms remain (Fig. 3d). Scanning of the oxidised Rh x O y species is difficult because the tip is very unstable, with frequent tip changes suggesting the presence of weakly bound species that interact with the STM tip. A comparison of the cluster density following saturation exposure with the initial adatom coverage suggests that the majority contain at most 2-3 Rh atoms. The fact that similar Rh agglomeration was observed after O 2 exposure in the preparation chamber suggests that the mobility observed in STM movies is not solely tip-induced. After heating the surface with oxidised Rh x O y clusters to 400 K, no structural differences are observed, but the imaging becomes much more stable, suggesting desorption or trapping of the weakly bound species. Fig. 4a shows O 2 TPD measurements from the clean Fe 3 O 4 (001) surface (grey curve) and from the surface with 0.6  Compared to the pristine Fe 3 O 4 (001), we identify two new components in the O 1s region at 533.7 and 536.9 eV. The spectra taken after heating above room temperature look identical to the orange curve taken at 300 K. The whole O 1s region is shown in the inset in (b). (c) Adsorption of O 2 at 60 K causes a ≈0.9 eV shift of the Rh 3d peak, and a further shift is observed upon heating to 300 K. Heating above ≈500 K leads to growth of the 309.5 eV component corresponding to Rh incorporated into the substrate. Up to 300 K, the spectra were taken at the temperatures given; for higher temperatures, the spectra were taken after cooling the sample back to 300 K.

Nanoscale Paper
This journal is © The Royal Society of Chemistry 2020 ML Rh (orange and brown curves). On the pristine Fe 3 O 4 (001) surface, O 2 physisorbs below 70 K (desorption peak labelled α in Fig. 4a, full spectrum shown in Fig. S4 †), but in our work this phase could not be saturated due to the O 2 dosing temperature (60 K) being within this desorption peak. Thus, desorption already takes place before the start of the heating ramp and the magnitude and shape of the α peak varies in repeated measurements due to slightly varying time between dosing and heating. Above 100 K, the O 2 TPD acquired on the pristine surface exhibits two distinct peaks, labelled β and γ in Fig. 4a and Fig. S4, † which presumably correspond to desorption from defects, domain boundaries, and/or step edges. The presence of Rh 1 adatoms on the surface prior to O 2 dosing leads to an increase of these two desorption peaks and induces a sharp new desorption peak, δ, at ≈330 K. When the TPD measurement was repeated on the same sample (following termination of the previous heating ramp at 570 K), the δ peak was no longer present (brown curve in Fig. 4a), and a similar result was observed in a separate experiment when the first heating ramp ended already at 380 K, before any Rh incorporation took place (evidenced by XPS spectra). XPS characterization of the O 2 adsorption on Rh/Fe 3 O 4 (001) is shown in Fig. 4b and c. Spectra taken directly after Rh deposition at 200 K and dosing 1.5 L O 2 at 60 K reveal two new components in the O 1s region at 533.7 and 536.9 eV (red curve in Fig. 4b). The 536.9 eV signal quickly decreases over time and completely disappears within several minutes, which is consistent with the O 2 physisorbed on Fe 3 O 4 (001) and readily desorbing at 60 K. The 533.7 eV component remains unchanged upon heating to 105 K, but disappears upon heating above 250 K (blue and orange curves in Fig. 4b). After heating to 300 K (orange curve), the spectrum in the O 1s region closely resembles the spectrum acquired on a pristine Fe 3 O 4 (001) surface (grey curve), save for a very small and broad shoulder between 530-535 eV, which can be attributed to a small amount of water adsorbed on surface defects. 31 Upon heating to higher temperatures up to 730 K, no significant change is observed in the O 1s region.
The adsorption of O 2 can also be followed in the Rh 3d region. Fig. 4c shows spectra taken after Rh deposition (0.6 ML), dosing O 2 at 60 K, and heating to different temperatures. Directly after Rh deposition at room temperature the shape of the Rh 3d 5/2 peak has a maximum at ≈307.7 eV, but as mentioned previously, its shape suggests convolution of several components. A small shoulder at ≈309.5 eV corresponds to Rh incorporated in the octahedral sites of the support (see Fig. 2); this component can be minimized by depositing Rh at lower temperature (Fig. S5 †). After dosing O 2 at 60 K, the maximum of the Rh 3d 5/2 peak shifts by ≈0.9 eV to 308.6 eV. Heating to 300 K induces a further shift to ≈309.0 eV, and this position remains the same with heating to 400 K. At this temperature, a small shoulder at 308.2 eV appears. Heating above 460 K causes growth of the 309.5 eV component corresponding to incorporated Rh and slight increase of the signal at 308.2 eV. Above 700 K only the signal from the incorporated Rh remains. The Fe 2p region does not show any significant changes with O 2 adsorption and heating (spectra shown in Fig. S6 †). The question of how the changes in XPS spectra correlate to the O 2 -induced Rh 1 agglomeration observed by STM is covered in the discussion section.

Exposing Rh 1 /Fe 3 O 4 (001) to CO
The same experimental approach as utilized above for O 2 was used to study CO adsorption on the Rh 1 adatoms. Fig. 5 shows five frames of an STM movie recorded while keeping a background pressure of 5 × 10 −9 mbar CO in the analysis chamber. In panel (a), the vast majority of the Rh 1 species have an apparent height of ≈180 pm (red circle). Over the course of the movie, the apparent height of the individual Rh 1 species abruptly decreases to 95-125 pm (panels b-d) and eventually almost all of the Rh 1 appear darker (panel (e)). We attribute this change to the formation of monocarbonyl RhCO, and the reduction of the apparent height to the modification of empty Rh states near the Fermi level through the interaction with CO. With higher doses of CO, we also observe double-lobed species oriented perpendicular to the surface Fe rows, which we assign as Rh(CO) 2 dicarbonyls (orange arrow in Fig. 5e and Fig. S7 †). Noncontact atomic force microscopy (ncAFM) measurements of the Rh carbonyls acquired with a CO-terminated tip (Fig. S7 †) appear strikingly  Fig. S5. † similar to those observed in the previously published study of Ir 1 /Fe 3 O 4 (001), 14 and one would indeed expect similar behavior for these two metals. As in the Ir case, the concentration of the dicarbonyl species never exceeded 25% of the total, which we attribute to the low pressures used in our experiments. Neither the monocarbonyls nor dicarbonyls appear to interact with molecular O 2 , as no changes were observed upon subsequent O 2 exposure in room-temperature STM measurements.
In 13 CO TPD experiments on 0.5 ML Rh/Fe 3 O 4 (001), we observe a 13 CO desorption peak with a maximum at ≈530 K, which has an additional shoulder at ≈450 K (Fig. 6a). Comparison to the spectrum acquired on the pristine Fe 3 O 4 (001) reveals that all the desorption signal above 250 K is related to Rh, and that the presence of Rh leads to disappearance of a defect-related 13 CO desorption peak at ≈190 K. On Rh/Fe 3 O 4 (001), a small 13 CO 2 signal is observed at ≈480 K, which we attribute to CO oxidation via a Mars-van-Krevelen mechanism. This phenomenon has been observed for Ir single atoms and Pt clusters on Fe 3 O 4 (001). 14,18 No CO desorption is observed above 580 K and in XPS spectra the C 1s region is featureless after heating to these temperatures (Fig. 6c). The CO desorption features below 200 K in the TPD spectra correspond to CO adsorption on the clean Fe 3 O 4 (001) surface. 32 Focusing on the Rh 3d and C 1s regions of the XPS spectra, CO adsorption at 60 K induced a shift of the Rh 3d peak maximum to 308.5 eV accompanied by the emergence of C 1s peak components at 290.6 eV and 287.3 eV. The 290.6 eV peak corresponds to adsorption of CO on the bare Fe 3 O 4 (001) surface. 32 Heating to 300 K leads to desorption of CO from Fe 3 O 4 (001), so that only CO bound to the Rh species remains (287.3 eV component in the C 1s region). The Rh 3d peak maximum shifts to lower binding energy (308.2 eV), and this position stays constant upon heating to higher temperatures. After the CO desorption from the Rh species above ≈500 K, the majority of the Rh incorporates into the surface, clearly evidenced by the increase in intensity of the component at 309.5 eV. Repeated 13 CO TPD measurements feature a small desorption peak at ≈410 K (Fig. S8 †), which we attribute to CO desorption from Rh 1 substituting an Fe atom in the 5-fold surface site. The Fe 2p region of the XPS spectra taken after CO adsorption and heating to different temperatures shows the same shape as the spectra taken without the CO (see Fig. 2c and Fig. S6 †).

Sequential dosing experiments
Sequential 13 CO and 18 O 2 dosing experiments were performed to find out whether the Rh adatoms or Rh x O y clusters can coadsorb CO and O 2 . Fig. 7a-d shows the CO (red) and CO 2 (blue and orange) signals desorbing from a 0.5 ML Rh/Fe 3 O 4 (001) sample exposed to CO and O 2 at different temperatures and in a different order. The blue area corresponds to CO 2 formed via a MvK mechanism ( 13 C 16 O 2 , mass 45), while the orange area indicates CO 2 formed by a L-H mechanism ( 13 C 16 O 18 O, mass 47). The corresponding 18 O 2 signal (mass 36) acquired simultaneously is shown in Fig. S9. † Fig. 7a provides a reference in which the sample was saturated by 13 CO at 300 K. Dosing O 2 on the sample predosed with 13 CO at 130 K ( panel b) leads to a lower CO desorption signal, and the overall amount of CO 2 produced is increased. The CO 2 signal produced via a MvK mechanism is comparable to the previous case, but additionally we observe a low, broad 13 C 16 O 18 O signal between 200-500 K corresponding to CO 2 formed by a L-H mechanism. Overall, roughly 35% of the CO 2 produced is formed by the L-H channel.
The situation changes dramatically when the same gases are dosed with the order reversed ( panel c). Dosing the 18 O 2 before the 13 CO at 130 K leads to a significantly enhanced mass 47 signal, which now exhibits a clear maximum at ≈340 K. The sum of masses 29, 45 and 47 is however signifi- shifts the Rh 3d photoemission peak maximum to 308.5 eV, but heating to 300 K shifts the maximum to 308.2 eV. With annealing to temperatures above 500 K, a new component develops at 309.5 eV, due to Rh incorporated into the Fe 3 O 4 (001) support. (c) In the C 1s region of the XPS spectra, adsorption of CO at 60 K results in a dominant feature at 290.6 eV, corresponding to CO adsorbed at the Fe 3 O 4 (001) surface, 32 and a smaller peak at 287.3 eV, corresponding to CO adsorbed on the Rh species. The 287.3 eV component is still observed after heating to 450 K. Above this temperature, no C 1s signal is observed in XPS.
cantly reduced compared to the previous CO-first experiments, which suggests that, overall, less CO can adsorb on the system once it is saturated with O 2 at 130 K, but a significantly higher proportion of adsorbed CO is converted to CO 2 . The overall amount of CO 2 produced is comparable to the case shown in Fig. 7b, but now the majority (≈60%) forms via the L-H channel. If the system is predosed with O 2 at 300 K prior to CO adsorption, the observed CO 2 TPD signal shown in Fig. 7d is significantly enhanced between 400 and 500 K in both reaction channels. In this experiment the amount of formed CO 2 is the highest (≈35% higher than in the previous two cases), and approximately 50% is formed by the L-H channel. These experiments, with O 2 predosing at 130 and 300 K, likely differ in that the oxygen-induced Rh 1 diffusion should be suppressed at low temperature, and additional oxygen is bound at sites linked to Rh species (see Fig. 4a). No CO 2 formation by any mechanism is observed in the absence of Rh species (see Fig. S10 †).

Discussion
The experimental data acquired on the Rh/Fe 3 O 4 (001) model catalyst clearly show that Rh 1 adatoms sinter rapidly when exposed to O 2 at room temperature, forming small Rh x O y clusters that are active for CO oxidation. Given the extremely low O 2 pressure used in the STM movies, it seems likely that just one O 2 molecule is sufficient to create a mobile oxidised Rh species, probably Rh 1 O 2 . This suggests that oxygen adsorption significantly weakens the interaction of the Rh adatom with the support. DFT calculations for oxygen adsorption on the Rh 1 /Al 2 O 3 system predict significant electron transfer from Rh to O 2 , and the formation of superoxo (O 2 ) − species. 3 In the O 1s XPS data taken after O 2 adsorption below 100 K (Fig. 4b) we do observe a peak at 533.7 eV, which is close to the position expected for superoxo species. 33 However, this peak disappears after heating above 250 K, suggesting a further reaction with Rh takes place. We also observe a strong positive binding energy shift in the Rh 3d XPS peak maximum (Fig. 4c) after O 2 dosing at 60 K (308.6 eV), which shifts further to ≈309.0 eV upon heating above 250 K. Even though the position observed at 60 K is already close to that expected for RhO 2 (308.6 eV) 34 or Rh 2 O 3 (308.3 eV), 34,35 and thus could be already interpreted as a signal corresponding to the oxidised Rh clusters, the further shift above 250 K accompanied with the disappearance of the superoxo component in the O 1s region suggests that the agglomeration takes place between 100 K and 250 K, and the O 1s signal of the resulting Rh x O y nanoparticles is indistinguishable from that of the O 2− in the Fe 3 O 4 support.
Our room-temperature STM movies of O 2 exposure (Fig. 3a-c) suggest that the mobile Rh 1 O 2 species interact strongly with bare Rh adatoms to form a well-defined double protrusion ( presumably with Rh 2 O 2 stoichiometry). The addition of yet another mobile Rh 1 O 2 seems common, and would in principle  13 CO at 130 K leads to significant CO 2 signal produced by the L-H mechanism, the MvK channel is suppressed. (d) Dosing 18 O 2 at 300 K, then 13 CO at 130 K leads to similar results as (c), but with more CO 2 produced above 400 K in both reaction channels. yield clusters of nominal Rh 3 O 4 stoichiometry. These clusters appear larger, and have a scratchy appearance due to the interaction with the STM tip. The cluster density observed following 50 L O 2 exposure suggests that the clusters rarely grow larger than 2-3 Rh atoms, but it cannot be concluded how many O 2 molecules can ultimately be adsorbed on each cluster, nor whether the formation of these clusters results in the abstraction of one or more O atoms. Nevertheless, the O 2 TPD measurements clearly show that some of the O atoms are weakly bound, and desorb as molecular O 2 just above room temperature in the absence of CO (Fig. 4a). The Rh 3d spectrum is barely changed by this desorption; only a very small shoulder appears at ≈308.2 eV (green curve in Fig. 4c), which suggests the clusters remain oxidised. This shoulder gets significantly more pronounced after heating to 460 and 575 K ( purple and cyan curves in Fig. 4c), and at the same time the component corresponding to incorporated Rh appears (309.5 eV). In the TPD spectra there is no O 2 desorption signal observed above 380 K (up to 660 K, where the acquisition was stopped), therefore it seems likely that the oxygen from Rh x O y clusters gets accommodated in the Fe 3 O 4 lattice as the Rh incorporates, and the 308.2 eV component in this case indicates an intermediate step between Rh x O y on the surface and Rh incorporated in the lattice. Repeating the oxygen exposure (after its desorption at ≈350 K, but before the Rh incorporation at higher temperatures) does not replenish the weakly bound O 2 , suggesting that the clusters restructure to a stable configuration, and that the weakly bound oxygen was a consequence of the agglomeration process.
When the O 2 -sintered sample is exposed to CO at low temperature and TPD is performed, CO 2 is found to evolve from the sample in a broad signal between 200-500 K (Fig. 7d). Isotopic labelling of the reactants shows that the majority of the CO 2 formed below ≈400 K is produced by the L-H channel, with the molecular O 2 being the oxidising agent. The observation of a similar result when the initial O 2 exposure was performed at 130 K instead of 300 K hints that O 2 -induced sintering happens already at low temperatures. Interestingly, the CO 2 signal at higher temperature is reduced in intensity when the O 2 exposure is performed at 130 K. The comparison of these two datasets suggests that the oxygen bound above 300 K (δ peak in Fig. 4a) is facilitating the CO 2 formation by the L-H channel, and the oxygen adsorbed below 300 K likely blocks a small number of sites for CO adsorption. An alternative explanation is that at 300 K the 18 O 2 spills over to the Fe 3 O 4 surface, which leads to the increased CO 2 production above 400 K. In this case the process would be MvK, but due to the previous spillover from Rh it would be measured in both 13  desorption is not expected to facilitate further CO 2 production via a L-H mechanism, because the remaining Rh x O y clusters seem to be inert for further interaction with O 2 .
When the Rh 1 /Fe 3 O 4 (001) sample is exposed to CO first, the Rh adatoms do not become mobile at room temperature. We have recently shown that CO adsorption allows Ir 1 adatoms to achieve highly-stable square-planar environments, 14 and our STM/ncAFM images for Rh 1 (Fig. S7 †) appear very similar. In ref. 14, the enhanced stability was rationalized by analogy to Ir(I) complexes, and it seems reasonable to expect similar behavior for Rh(I), which can also adopt the preferred square planar d 8 configuration. The XPS spectra acquired on the COsaturated Rh/Fe 3 O 4 (001) system show almost no change in the Rh 3d region when heated to temperatures between 300 K and 450 K, suggesting most of the Rh carbonyls still exist at 450 K, and are still present before the CO 2 production takes place via a MvK mechanism at ≈480 K. Exposing the CO-saturated sample to O 2 does not lead to visible changes in STM or XPS, nor does it lead to significant low-temperature CO 2 production. We conclude that CO poisons the Rh adatoms for O 2 adsorption, at least under our low-pressure conditions. We thus do not form OCOO species, which have been proposed as possible intermediates in prior studies of CO oxidation by Rhbased single-atom catalysts. 36,37 This could be because the Rh atom is inaccessible to the O 2 molecule once the CO is present, or because adsorbing both molecules on the same Rh adatom is energetically unfavourable. The situation could be different at higher pressures where the kinetic limitations are more easily overcome. A more likely scenario, however, is that stable Rh(CO) 2 dicarbonyls would be formed and the catalyst would remain poisoned for CO oxidation via a L-H mechanism. Our TPD experiments show that a small amount of CO is oxidised to CO 2 at high temperature via a MvK mechanism, but this is followed by immediate incorporation of Rh into the support, which ultimately deactivates the catalyst.

Conclusions
This study presents an atomic-scale view on a range of phenomena relevant for heterogeneous catalysis by single atoms and sub-nano clusters. We show that O 2 adsorption initially leads to formation of superoxo species bound to Rh 1 adatoms, which already at room temperature rapidly agglomerate to small Rh x O y clusters. CO adsorption, on the other hand, completes the preferred square-planar coordination environment of the Rh 1 adatoms, stabilizing the Rh 1 in place while poisoning them for O 2 adsorption. Both the oxidised Rh x O y clusters and Rh 1 CO carbonyls show activity for CO oxidation, but via very different pathways: oxidised clusters catalyse the reaction via a Langmuir-Hinshelwood mechanism at temperatures as low as 200 K, while the presence of Rh carbonyls at 300 K allows CO 2 formation via a Mars-van-Krevelen mechanism at ≈480 K. These results highlight the importance of a careful characterization of the catalyst during the exposure to the reactants and show that the gas composition can play a crucial role in the structural evolution of the catalyst.

Conflicts of interest
There are no conflicts to declare.
This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.