Introduction

Photocatalysts are promising materials to harvest solar energy1 or purify polluted air and water2. Recent pandemics showcased the significance of the antibacterial and antiviral properties of photocatalytic surfaces3,4. One promising material is TiO2, a widely applied photocatalyst with strong oxidizing properties. TiO2 is low-cost, non-toxic5, chemically and biologically stable, and shows antiviral and antibacterial properties6,7,8. The interest in TiO2 as a photocatalyst increased in 1972 when Fujishima and Honda used a TiO2 semiconductor photoanode for water splitting under UV light9, and since then, this oxide was considered a component for solar cells1. Today, the photocatalytic properties of TiO2 were studied for a range of reactions, such as water splitting10 for hydrogen generation11, and were tested in field studies for the degradation of pollutants in air and water6,8, e.g., as an additive for concrete12. Rutile is the most stable polymorph of TiO2 and is a widely studied model system for science on metal oxide surfaces13. Therefore, research on rutile contributes to a deeper understanding of the nature of photocatalysts.

One well-studied heterogeneous catalytic model reaction is CO oxidation due to its simplicity and its character as a benchmark system14. CO oxidation has a single product, CO2. CO photooxidation on rutile and anatase TiO2 was studied using Infrared Reflection Absorption Spectroscopy as well as X-ray Photoelectron Spectroscopy (XPS)15,16,17. As expected from studies on powdered samples18, stoichiometric anatase (101) is the most photocatalytically active surface under UV-illumination, exhibiting faster reaction kinetics compared to reduced anatase (101) as well as reduced and stoichiometric rutile (110) in converting CO to CO2. CO oxidation on both TiO2 polymorphs was only observed in an O2 atmosphere under concurrent UV-illumination.

In these experiments, the UV-illumination initiates a photocatalytic reaction, as an electron-hole pair is generated in the conduction and valence band19. Gas-phase O2 dissociates by trapping the generated electron, resulting in adsorbed oxygen. The chemisorbed oxygen ion reacts with adsorbed CO to form CO215. The adsorption of oxygen is necessary for the reaction, as studies show, that lattice oxygen is not an oxygen source for this photoreaction20. The adsorption and photoactivation of oxygen is the crucial step, initiating the CO oxidation as an electron-mediated reaction, and thus competing with charge carrier recombination21. The efficiency of this process directly impacts the efficiency of the catalyst. A longer lifetime of charge carriers increases the probability of interacting with a gas-phase oxygen molecule. Different studies16,22 found a shorter bulk lifetime of charge carriers in rutile compared to anatase. The reason is that anatase has an indirect band gap that inhibits electron-hole pair recombination and therefore enables a higher percentage of generated charge carriers to initiate this reaction pathway. The direct band gap of rutile results in a shorter lifetime of charge carriers, thus lowering the catalytic efficiency16.

The studies on the lifetimes of charge carriers focus on the bulk properties of the materials leaving a knowledge gap of catalytic reactions that occur at the surface. The general observation of longer lifetimes of bulk electron-hole pairs in anatase is in alignment with the higher photocatalytic efficiency. Still, it ignores the influence of adsorbates on the catalytic surface under reaction conditions. In addition, the lifetimes of surface charge carriers may differ from bulk charge carriers and are influenced by band bending, defects, surface traps, and adsorbates23. It is therefore important to study the reaction dynamics in a catalytic environment to elucidate the reaction mechanism21.

The ultrafast real-time dynamics during the CO oxidation on anatase TiO2 was previously studied by Wagstaffe et al.24 in a pump-probe experiment at FLASH. On anatase (101) CO photooxidation to CO2 induced by a 770 nm laser was observed with a delayed onset between 1.2 and 2.8 (±0.2) ps after illumination. Based on Density Functional Theory (DFT) calculations an O2-TiO2 charge transfer (CT) complex was proposed, that enabled a direct charge transfer from the anatase-TiO2 valence band to the O2 molecular states in the bandgap. It was proposed, that the directly excited adsorbed oxygen dissociates and provides the oxygen adatoms for the CO oxidation. This indicated that charge transfer can occur on a faster timescale than previously reported25,26.

To elucidate the role of the oxygen activation timescale for the difference in catalytic activity of TiO2 rutile in comparison to anatase, we studied the dynamics of CO photooxidation on rutile (110) at the FEL FLASH at the Deutsches Elektronen-Synchrotron (DESY) in Hamburg. Free-electron Lasers with pulses in the femtosecond timescale allow the study of ultrafast surface dynamics and possible reaction intermediates that are observable on picosecond timescales27,28. Using superconducting RF accelerator technology, FLASH provides high-repetition rate photon pulses suitable to observe chemical dynamics with sub-picosecond temporal resolution29,30. The dynamics of the photoinduced CO oxidation to CO2 was monitored on rutile (110) with a temporal resolution of 250 fs. The formation of CO2 is observed within 200 to 800 fs after illumination with an optical laser with a wavelength of 770 nm. Time-Dependent Density Functional Tight-Binding (TD-DFTB) calculations showed the formation of an O2-TiO2 charge transfer complex as a possible pathway for ultrafast oxygen activation. Our results demonstrate that, although anatase is the more active photocatalyst compared to rutile, the dynamics of the CO oxidation on rutile is faster. The observation of different reaction dynamics on rutile and anatase is a further step to link the electronic structure of a material to its dynamics and the charge transfer to reactants.

Results

In this experiment, we studied the ultrafast dynamics of the CO oxidation to CO2 on rutile (110) in a controlled gas atmosphere of CO and O2 each with a partial pressure of 3 ⋅ 10−8 mbar at a sample temperature of 80 K. An optical pump laser (770 nm/1.6 eV) and the third harmonic of the FEL (hv = 643 eV) as a probe beam were spatially and temporally overlapped. The relative timing between the pump and probe pulses was controlled by a mechanical delay stage as part of the optical laser setup. The rutile TiO2(110) sample surface was prepared as described in the experimental section. With the FEL the O 1s, Ti 2p, and C 1s core levels were probed. Because of the increasing amount of water on the surface during the experiment, data were analyzed by integration over the first 15 min after flash-annealing the sample to observe the intrinsic behavior of the system, as displayed in Fig. 1.

Fig. 1
Fig. 1
Full size image

Left panel: Deconvoluted time-resolved O 1s core level spectra binned by 200 fs steps reveals ultrafast CO (red) oxidation to CO2 (blue) on rutile (110) at 80 K at FLASH. Right panel: CO2 concentration obtained from the integrated O 1s core level. The data were taken within the first 15 min after flash-annealing the surface.

As a prerequisite for the pump-probe experiments, the temporal resolution could be determined to be 250 fs. For this, the Full Width Half Maximum (FWHM) of the sidebands at zero delay in the Ti 2p (Fig. S1) and O 1s (Fig. S2) core level spectra have been evaluated. Sidebands appear when the optical laser and the FEL are temporally and spatially overlapped, and are represented as a replica of the original photoemission line shifted by the energy of the optical laser through absorption or stimulated emission31,32. The Ti 2p spectra (Fig. S1c) exhibit a small shoulder on the lower binding energy side of the Ti4+ 2p3/2 lattice peak, assigned to Ti3+ surface defects. The area of the Ti3+ 2p3/2 peak amounts to 5 ± 1% compared to the Ti4+ contribution. As one single oxygen vacancy contributes to two Ti3+, the estimated amount of oxygen vacancies at the surface is 2.5 ± 0.5%33. Defects can also occur in the form of Ti3+ interstitials.

In prior XPS experiments, the kinetics of CO oxidation to CO2 on rutile TiO2 (110) under UV light (365 nm) was studied with an X-ray laboratory source (hv = 1486.6 eV) at the DESY Nanolab34. In the O 1s core level region, adsorbed CO gives rise to a component at a binding energy of 536.1 eV and CO2 at 535.0 eV on rutile (110) with a difference of 1.1 eV between the two components (Fig. S3). The assignment of the binding energies of the O 1s core level of CO and CO2 in this study is in agreement with previous results17. The CO peak arises at 536.5 eV on rutile (110) at 80 K in 3 ⋅ 10−8 mbar CO atmosphere. At this temperature, it was reported that half a monolayer of CO adsorbs on rutile (110)35. Slight differences in the binding energy are expected due to coverage-dependent shifts36. Here we monitored the ultrafast dynamics of the conversion of CO to CO2 in the oxygen O 1s core level. The O 1s core level was chosen to monitor the time-resolved CO oxidation, as the favorable stoichiometry of the reaction product CO2, consisting of two oxygen atoms, results in higher absolute count rates. Additionally, the cross-section for photoionization at 643 eV is higher for the O 1s region than the C 1s region37. To obtain enough statistics to extract the time-resolved data shown in Fig. 1, the data were binned by steps of 200 fs. The detected photoelectrons of the O 1s and Ti 2p core level region have a kinetic energy of 100 eV and 175 eV, respectively, and the inelastic mean free path (λ) of TiO2 for electrons at those energies is calculated to be 5.6 Å and 6.8 Å38. The XPS probing depth, from which 95% of the measured photoelectron originates, corresponds to 3λ39. In TiO2, which has a layer spacing of 3.25 Å in the [110]-direction40, the photoelectrons measured for the Ti 2p region originate from the first 6.3 atomic layers and for O 1s from the first 5.2 atomic layers. This emphasizes the high surface sensitivity of this technique.

XP spectra of the O 1s core level were recorded within a delay range from −2 ps before to 12 ps after optical excitation. The results presented in Fig. 1 and discussed in detail in this work were only recorded within the pump-probe delay range from −0.6 to 2.0 ps in which the CO2 formation was observed. After the sample preparation, data were collected for 50 min. In the range of the first 800 fs after the pump laser initiates the reaction, a peak shoulder located at 535.5 eV appears, which is assigned to the formation of CO217. The maximum amount of CO2 is detected at 400 fs with 13 ± 4% normalized to the CO signal. The complete binding energy and delay range from −2 to 12 ps is shown in Fig. S4 as spectra and in Fig. S5 as a XP color map. No other CO2 signal was resolved in that delay range up to 12 ps within the sensitivity of our experiment.

The data averaged over the whole delay window in 5 min bins after the flash-annealing is shown in Fig. 2. During data acquisition at 80 K in a CO/O2 atmosphere with a partial pressure of 3 ⋅ 10−8 mbar for CO and O2 each, residual water in the UHV system of the experimental chamber accumulates on the surface (residual pressure: 4 ⋅ 10−10 mbar). The peak at 534 eV is assigned to water41 and it increases with time. Water partially dissociates to hydroxyls (OH) on the rutile surface and appears as a shoulder of the lattice O2− peak at 531.8 eV42. This shoulder is already visible within the first 5 min and before the H2O peak at 534 eV appears. The OH signal increases with the water coverage. Water impedes the adsorption of CO and promotes the stabilization of adsorbed CO243. As H2O binds more strongly to the rutile (110) surface than CO, the CO peak decreases over time as H2O blocks adsorption sites and inhibits readsorption of CO from the gas-phase as seen in the time-averaged data (Fig. 2). The data collected in the first 5 min in the absence of adsorbed water does not offer sufficient statistics for time-resolved binning. In the first 15 min, used for the time-resolved analysis in Fig. 1, a low amount of water and a negligible amount of CO2 is adsorbed on the surface. In the corresponding time-resolved data in Fig. S4 the water signal is not visible because the high statistical noise of the small signal. This is in contrast to the spectra from 15 to 50 min, which show a clear CO2 signal as a broad shoulder and a further increasing water signal, indicating that the surface is covered with a non-negligible amount of CO2 and H2O. After 40 min the CO2 signal does not increase anymore, indicating that no further CO is oxidized. Water accumulates further on the surface, blocking the adsorption sites for CO and O2. The CO2 formation was observed using the time-resolved spectra within the first 800 fs with a binning of 200 fs recorded in the first 0–15 min after flash-annealing. For comparison, the data for 0–15 min and 15–30 min after heating were binned in 500 fs time windows and CO2 formation at 535.5 eV was observed for both data sets in the spectra 250–750 fs after time zero as seen in Fig. S6.

Fig. 2: Deconvoluted O 1s core level average spectra during CO oxidation at FLASH binned in 5 min steps after heating and cooling the sample to 80 K.
Fig. 2: Deconvoluted O 1s core level average spectra during CO oxidation at FLASH binned in 5 min steps after heating and cooling the sample to 80 K.
Full size image

Adsorbed CO (red), produced CO2 (blue) and adsorbed water (light blue).

In the O 1s spectra, no peak can be assigned to adsorbed atomic or molecular oxygen under the experimental conditions of 3 ⋅ 10−8 mbar O2 and 3 ⋅ 10−8 mbar CO at 80 K. Physisorbed O2 was observed below 60 K on anatase (101) with a binding energy of 537.3 eV as a double peak in a triplet state44. On stoichiometric rutile TiO2(110), molecular oxygen only interacts weakly with the surface and physisorbs at low temperatures below 85 K45. When oxygen vacancies are present, O2 chemisorbs on the surface in a peroxo \({{{{\rm{O}}}}}_{2}^{2-}\) state in the oxygen vacancy itself or in the direct vicinity, on top of a five-fold coordinated Ti (Ti5c) atom. Even at low temperatures below 80 K, adsorbed O2 can heal an oxygen vacancy leaving oxygen adatoms on the TiO2 surface46. Due to defects (oxygen vacancies) of 2.5%, chemisorption of O2 is probable on the surface but due to the very low amount, it is below the detection limit in the O 1s spectra. Note, that the O2 signal overlaps with the strong O 1s signal from CO at 536.5 eV, making it impossible to detect it independently in our time-resolved measurements.

To explain the experimental observation of ultrafast CO oxidation within 800 fs, we performed first principle TD-DFTB calculations for the adsorption of O2 and CO on the rutile TiO2(110) surface. The optimized geometry of CO and O2 adsorption can be found in Fig. S7 and the density of states (DOS) of these systems in S8. Two favorable configurations are found for the coadsorption of CO and O2. In both configurations, the CO molecule interacts via the carbon atom with Ti5c in agreement with previous reports45,47,48. The O2 molecules either adsorb perpendicular or parallel to the surface on top of the neighboring Ti5c site with a calculated adsorption energy of −0.667 eV for the perpendicular and −0.505 eV for the parallel O2 + CO configuration, respectively. The DOS calculations reveal that the presence of adsorbed O2 molecules is related to the appearance of electronic states in the band gap of the surface model, which are responsible for a new band at lower energies in the absorption spectrum. This indicates the formation of an O2-TiO2 CT complex that is activated by visible/infrared light via a direct electron transfer from the TiO2 valence band to adsorbed O2 molecules. Such a charge transfer complex was proposed previously by Wagstaffe et al.24 for the ultrafast CO oxidation on anatase initiated by a 770 nm laser and by Freitag et al.49 for the visible light activity of TiO2 with adsorbed nitrogen(II) oxide. In this study, we additionally calculated the absorption spectrum of TiO2 with and without adsorbed O2 using real-time TD-DFTB implementation. The absorption spectra are shown in Fig. 3. Stoichiometric TiO2 has no adsorption bands in the visible light region, but upon O2 adsorption new adsorption bands in the visible light region appear due to the formation of a CT complex. Via the CT excitation, adsorbed O2 is reduced to \({{{{\rm{O}}}}}_{2}^{-}\) which is the initial step of O2 dissociation.

Fig. 3
Fig. 3
Full size image

DFT calculation of an absorption spectrum of TiO2 before (black) and after (red) O2 adsorption.

Discussion

In this experiment, the oxidation of CO to CO2 is observed within the first 800 fs after initiation. Figure 4 illustrates the reaction mechanism. After 800 fs CO2 desorbs and new CO2 forms only in the next cycle (see Fig. 1). This gives evidence, that O2 dissociates on TiO2 by a light activated process as O2 only physisorbs on stoichiometric TiO2 at 80 K and does not dissociate. On the other hand, on defective rutile (110) oxygen adatoms were observed in small concentrations at 80 K after O2 adsorption and healing of an oxygen vacancy which leaves an isolated oxygen adatom46. However, this mechanism can be excluded for a reaction cycle, as the oxygen vacancy is healed after the O2 dissociative adsorption. Other CO2 signals were not detected in the 12 ps delay window, which would indicate another CO oxidation pathway with a different timescale (see Fig. S4). The CO2 signal, therefore, implies that oxygen was reduced, dissociated, and reacted with CO to CO2 within 800 fs.

Fig. 4
Fig. 4
Full size image

Schematics of the time-resolved reaction mechanism, see text for details.

For CO photooxidation on TiO2, the activation of oxygen is considered the rate-defining step. This includes charge transfer to oxygen and dissociation of the anion. To generate electron-hole pairs on TiO2, photons with either an energy larger than the bandgap or, in case of photons with a smaller energy, multi-photon absorption is required to excite electrons from the valence to the conduction band. Multi-photon absorption to match the bandgap of 3 eV for bulk rutile TiO250 was reported for laser pulses with pulse energies up to 3 μJ and wavelengths of 774, 800, and 813 nm from a chirped-pulse amplified Ti:Sapphire system51. The optical laser fluence used in this study fits within those parameters, having a pulse energy of 5–10 μJ and a wavelength of 770 nm. The excited electrons in the conduction band can either recombine, be trapped, or induce an oxidation/reduction pathway, in this case, the reduction of O2. Photogenerated electrons in the TiO2 band structure are generated in less than 100 fs and conduction band electrons are trapped at the surface within 200 fs52. The transfer of surface-trapped electrons to oxygen was only observed on a nanosecond timescale in less than 100 ns25 and is significantly faster than the transfer of conduction band electrons to oxygen in 10–100 μs53. In this activation pathway, the electron-driven oxygen activation competes not only with charge carrier recombination but also with the hole-driven desorption of oxygen54.

Alternatively, based on our DFT calculations, an O2-TiO2 CT complex is proposed. Adsorbed oxygen on stoichiometric rutile (110) introduces unoccupied states into its band gap with excitation energies relating to the visible light range as seen in Fig. 3. The direct excitation of the oxygen molecule may activate oxygen faster than by photogenerated conduction band electrons. As a result, the activated \({{{{\rm{O}}}}}_{2}^{-}\) dissociates and reacts with CO to CO2. The activation with a 770 nm laser only requires one photon absorption to excite the O2-TiO2 CT complex. The CT complex’s excitation probability is lower than band-to-band excitations since the excitation is limited by the number of acceptor states offered by the adsorbed oxygen molecules. But the CT complex excitation is more efficient since the charge is directly trapped in the reduced oxygen molecule and does not compete with recombination. A similar charge transfer complex was proposed for NO on TiO2 for the visible light for the degradation of NO49. The photonic efficiency under UV (bandgap excitation) and visible light (CT complex excitation) are in the same order of magnitude. However, the CT process is one order of magnitude less likely to occur. The CT pathway does not compete with charge carrier recombination and thus increases the photonic efficiency of the desired reaction.

The O2-TiO2 CT is also a possible mechanism for anatase TiO2(101). In a similar study at FLASH in which the ultrafast CO oxidation on anatase (101) at 60 K was investigated, the CO2 formation was assigned to the direct activation of oxygen24 and was observed from 1.2 ps to 2.8 ps after initiation. In contrast, this study on rutile (110) at 80 K reveals that CO2 is observed already at 0.2 ps after initiation. The temperature difference might influence the diffusion of the adsorbates on the surface, thus enabling a faster reaction between CO and O255. In our experiment, measurements at 60 K were not feasible due to rapid water adsorption on rutile (110), which blocked adsorption sites for the reactants. Therefore, we conducted measurements at 80 K instead.

In comparison, CO oxidation was observed in the first few picoseconds after initiation on catalytic metals such as Ru(0001)56,57, Pt(111)58 and Pd(111)59. Important to note is, that only atomic and no molecular oxygen was adsorbed on these metal surfaces. Furthermore, the oxygen was not dissociated during the reaction process. The role of the optical laser was to excite electrons in the substrate, resulting in an energy transfer to the adsorbate. This causes vibrational motions that induce the reaction between CO and O. Öström et al.57 studied the ultrafast CO oxidation on Ru(0001) with femtosecond X-ray laser pulses and reported the activation of O within 300 fs, of CO in 500 fs, and a formation time for CO2 of 800 fs. In our experiment we observe a faster CO2 formation onset time hinting at a different reaction mechanism. As mentioned above, atomic oxygen on rutile-TiO2(110) is only observed on reduced rutile at surface oxygen vacancies. Molecular oxygen chemisorbs on top of an oxygen vacancy, heals the vacancy by dissociation, and leaves an O adatom46.

We argue that atomic oxygen on oxygen vacancies is unlikely to be responsible for the fast CO oxidation. The low amount of 2.5 % oxygen vacancies in the Ti 2p spectra is in disagreement with the maximum of 13 ± 4% CO2. The amount of oxygen vacancies indicated by Ti3+ does not change during the CO oxidation (see Fig. S9). It also indicates that another CO oxidation mechanism without defect contribution is present, since the CO oxidation is also observed on the stoichiometric rutile (110) and anatase (101)-TiO2 surfaces17. Oxygen adatoms are observed to be unstable on anatase (101), so the mechanism including complexes of CO and oxygen adatoms is not a possible pathway on anatase60.

The rutile (110) and the anatase (101) surface results differ in three aspects: in the amount of observed CO2 compared to CO, the time between initiation and first CO2 signal detection, and the decay time of the CO2 peak. The CO2 signal on anatase (101) was detected for 1.6 ps with a maximal CO2 concentration of ~25 % relative to the CO signal. After 2.8 ps the formed CO2 was desorbed. For comparison, on rutile (110) no CO2 is observed after 1 ps and the maximum relative CO2 concentration was ~13% after 0.4 ps. The higher relative CO2 concentration aligns with the observation that anatase (101) shows higher photocatalytic activity than rutile (110) and it implies that the charge transfer process is more efficient on the anatase (101) surface.

One major argument found in the literature for the higher photocatalytic activity of anatase is the charge carrier lifetime difference in rutile and anatase, which were studied for single-crystals and powders16,22,23,61,62. However, those measurements always detected the concentration of bulk charge carriers and not e−/h+ at surface sites for charge transfer to adsorbates. In stoichiometric rutile, charge carrier recombination is more likely and charge carriers are shorter-lived than in stoichiometric anatase. Maity et al.22 found using transient absorption spectroscopy, that bulk charge carriers decaying in ~0.5 ps for stoichiometric and reduced rutile (110) single crystals, whereas in stoichiometric and reduced anatase the lifetime was 32 ps and 24 ps, respectively. The timescale for rutile in that study is similar to the timescale of the CO2 formation in our study. The longer lifetime of charge carriers in anatase than rutile is in agreement with measurements from Xu et al.16 who observed a direct band gap with faster charge carrier recombination for rutile and an indirect bandgap for anatase with inhibited charge carrier recombination which links the bulk properties of rutile and anatase to its photocatalytic performance. In an indirect bandgap semiconductor, recombination requires a phonon to conserve momentum due to the mismatch between the valence band maximum and conduction band minimum63 and results in a prolonged lifetime of charge carriers. But the lifetime of surface charge carriers depends highly on surface adsorbates and not only on bulk properties19. Adsorbates can not only act as traps for photogenerated charge carriers but also induce band bending, promoting the migration of charge carriers to the bulk. The separation of charges directly influences the charge carrier lifetimes23. In this experiment, the surface dynamics are measured on a picosecond timescale, in contrast to the bulk lifetimes of charge carriers of several nanoseconds. The decay of the CO2 signal within 1 ps can also indicate that the direct excitation of O2 via the CT complex and not charge carriers from the bulk are responsible for the ultrafast oxidation of CO observed in this study. The reaction time is therefore dependent on the lifetime of the excited oxygen. We cannot exclude a further oxygen activation mechanism mediated by trapped charge carriers from the TiO2 conduction band on a nanosecond timescale, which is not detected in our study.

It is important to note that, although the surface is not deliberately exposed to water in this experiment, a growing water peak is observed after 5 min at 80 K in the CO/O2 atmosphere. Since the time-resolved data requires 15 min of each run for sufficient statistics water is present in the time-resolved spectra but in too low amount for a quantitative analysis. On rutile (110) water adsorbs on top of Ti5c64 and partially dissociates into hydroxyl (OH) groups. Water influences CO oxidation and can promote or inhibit the reaction. In one study65, the CO oxidation rate under UV light increased until a coverage of up to 1/2 ML of water was reached and decreased for higher coverages. It was proposed that on rutile (110) under UV light in the presence of water H2O2 as well as surface peroxo-species such as Ti–O–O–H and Ti–O–O–Ti are formed66. The CO oxidation in the presence of water appears to correlate with the amount of peroxide species formed. Several studies65,66,67,68,69 agree, that water blocks the adsorption sites for CO decreasing CO adsorption and therefore decreasing the CO2 formation rate. We also observe a decrease in CO adsorption with increasing water adsorption (see Fig. 2). Water could also influence the activation of oxygen and the charge transfer from TiO2 to O2 as the initial step of the CO oxidation. Wagstaffe et al.41 reported the ultrafast hole transfer from anatase-TiO2(101) to water within 285 fs as well as the formation of a hydrogen bond between water and the O2c site.

Tilocca et al.70 investigated the adsorption of O2 on the hydroxylated rutile (110) surface with molecular dynamics simulations. Physisorbed O2 can interact with OH-groups without going through a chemisorbed state. The adsorption structures included hydrogen bonds between chemisorbed O2 and OH, structures resulting from proton transfer as the formation of hydroperoxyls (HO2), and hydrogen peroxide (H2O2), and less stable structures resulting from dissociative O2 adsorption as OHt and Oa. Hydroperoxyls HO2 and bridging hydroxyls OHt were observed experimentally by Scanning Tunneling Microscopy71,72, Kelvin Probe Force Microscopy, and Atomic Force Microscopy73. Molecularly chemisorbed O2 next to OHbr was not observed. Calculations found an increased O2 adsorption mediated by adsorbed OH groups due to a charge transfer from OH to TiO274. Upon adsorption, the charge is transferred to O2 thus stabilizing the adsorption. Local Contact Potential Difference measurements suggest experimental evidence for the charge transfer from Ti5c atoms to oxygen Oa73. More recent calculations confirmed that O2 adsorption is favored on rutile (110) in the presence of OH groups75 and that the energy barrier for the O=O scission, necessary for the CO oxidation, is lowered by proton transfer, which is induced by adsorbed water76. The interaction of OH/H2O with O2 might thus facilitate the interfacial charge transfer, leading to a enhanced O2 and CO interaction, and therefore CO2 formation. We cannot exclude, that the faster CO2 formation on rutile TiO2(110) compared to anatase TiO2(101) could be due to the presence of water.

Conclusion

In conclusion, we investigated the dynamics of the CO oxidation on rutile TiO2(110) by optical pump, FEL probe X-ray photoemission spectroscopy. In an O2/CO atmosphere at 80 K, CO adsorbs on the rutile (110) surface and is oxidized to CO2 within the first 800 (±200) fs after excitation by the 770 nm laser. We propose, that O2 adsorbs molecularly on the surface and is activated via an O2-TiO2-CT complex. Residual water in UHV blocks CO adsorption sites and reduces the CO2 oxidation but might in low coverages facilitate charge transfer. With time-resolved XPS, several oxygen-containing components in the O 1s core level were monitored simultaneously, allowing studying reaction dynamics of co-adsorbed reactants or several products non-destructively in real-time.

While on anatase TiO2(101) the CO oxidation is observed within 1.2 and 2.8 ps after initiation, the CO2 signal on rutile is visible between in the first 0.8 ps. This indicates a shorter activation time of the oxygen species on rutile (110) likely related to a faster charge transfer. Although anatase is the more active photocatalyst compared to rutile, the dynamics of the CO oxidation on rutile is observed to be faster. The observation of different reaction dynamics on rutile and anatase is a further step to link the electronic structure of a material to its dynamics and the charge transfer to reactants. Our study sets the experimental basis for future time resolved full band structure theoretical studies which will further elucidate the mechanistic origin of the different reaction dynamics. Such a deeper understanding will help in tailoring photocatalytic systems which is crucial for developing more efficient materials for green energy production, as water splitting or photovoltaics.

Methods

Experimental

The time-resolved photoemission data were taken at the plane grating monochromator beamline PG277,78 of the free-electron laser FLASH29,30 located at DESY in Hamburg, Germany. The fundamental wavelength of FLASH was 5.79 nm (214 eV) with a pulse energy of 25–40 μJ. To probe the core level of oxygen O 1s the monochromator was tuned to the third harmonic of 1.93 nm (643 eV). The FEL pulses were delivered with a macrobunch repetition rate of 10 Hz with each macrobunch consisting of 400 bunches with a 1 MHz repetition rate. The temporal FWHM of each FEL pulse was <100 fs, though stretched in the monochromator to 150 fs. The optical pump laser with a wavelength of 770 nm (1.6 eV) matches the pulse pattern of the FEL. The maximum single pulse energy of the optical laser was 5-10 μJ with a spot size of ~300 μm under normal incidence. The fluence could be calculated to be 7–14 mJ ⋅ cm−2 under the measurement geometry of 55∘ sample tilt with respect to the incoming laser beam. The laser beam is coupled in collinearly with the FEL. To prevent CO desorption the laser fluence was attenuated to 2.2 mJ cm−2. The temporal FWHM of the optical laser pulse was ~120 fs. A mechanical delay stage set the temporal delay of the optical laser with respect to the FEL beam. The experimental setup used at the beamline was the wide-angle electron spectrometer (WESPE)79 chamber. WESPE consists of a sample preparation chamber with an ion gun, heating station, and low energy electron diffraction (LEED). The main experimental chamber is equipped with a Themis 1000 high-resolution time of flight spectrometer with a three-dimensional delay line detector (3D-DLD4040-4Q, Surface Concept), beamline connection, and leak valves for dosing gases. The spectra were recorded with a pass energy of 20 eV. The used gases were Ar (purity 99.999%) for sample preparation and CO (purity 99.97%) and O2 (purity 99.999%). The cryostat in the manipulator, which holds the sample under investigation, allowed cooling with liquid He. The rutile TiO2(110) single crystal (7 mm × 7 mm × 1 mm) was cleaned under ultra-high vacuum (UHV) conditions with a base pressure of 3 ⋅ 10−10 mbar by repeated cycles of 1 keV Ar+ ion sputtering and flash-annealing to 650 ∘C (for less than 1 min) and cooled in 1 ⋅ 10−6 mbar O2 until a (1 × 1) LEED pattern was obtained (Fig. S11). X-ray photoelectron spectra confirmed the absence of carbon contaminations (Fig. S10). Although the sample was annealed and cooled in oxygen, the Ti 2p core level spectra show 4–6% of Ti3+ as a small shoulder next to the lattice peak of Ti4+, as seen in Fig. S1c. Ti3+ indicates defects in the form of oxygen vacancies80. During the CO oxidation, the sample was cooled by liquid He to 80 K and was held in an atmosphere of CO and O2 with partial pressures of both 3 ⋅ 10−8 mbar. To avoid potential laser-induced damage, the incident pulses were scanned across the sample surface. During data acquisition at 80 K in a gas atmosphere of CO and O2 each with a partial pressure of 3 ⋅ 10−8 mbar, residual water from the UHV environment adsorbed on the cold sample surface resulting in a growing peak at 534.8 eV, as seen in Fig. 2. To limit the influence of water on the reaction dynamics, only spectra recorded until 15 min after brief flash-annealing of the surface to 600 K are used to study the dynamics of the CO oxidation. The sample was cleaned by sputtering and annealing in O2 after two measurement cycles with flash-annealing to obtain a stoichiometric surface. The binning of the extracted spectra was were 200 fs in temporal domain and 200 meV in electron energy. In total, the data of the first 10–15 min of 22 runs was used, depending on the amount of adsorbed water. Due to shifts of the FEL photon energy, the Ti 2p spectra of each run were calibrated by aligning the Ti4+ 2p3/2 to 459.0 eV and the O 1s spectra by calibration the lattice O2− to 530.4 eV13. For each run time zero, the temporal overlap of FEL and optical laser was determined by fitting the sidebands of the O 1s lattice peak. The spectra were fitted in CasaXPS with Gaussian/Lorentzian curves on a Shirley or linear background. The Shirley background emulates the inelastic electron scattering of the intensive O 1s lattice peak. Regions with lower counts were fitted with a linear background as the modeling of the inelastic scattering did not improve the fit (see Supplementary Data).

Theroretical

The DFT periodic calculations on the neutral TiO2 rutile (110) surface were performed with the Vienna ab initio simulation package (VASP) code81,82,83,84 The TiO2 rutile (110) surface was modeled by a slab model of 40 Ti atoms and 80 O atoms using the following lattice parameters: a = 5.9612 Å, b =  13.0834 Å, and c = 30.0000 Å. For the optimization of the structures with the Perdew-Burke-Ernzerhof (PBE) functional85, the projecto augmented-wave method86,87 was used and the Brillouin zone was sampled with a (2 × 2 × 1) Monkhorst-Pack k-points grid with an energy cutoff of 400 eV. In the systems with O2 molecule presence, spin polarization calculations are performed to include the triplet nature of the O2 molecule in the ground state. Van der Waals interactions were included by using the DFT-D3 dispersion corrections with Becke-Johnson damping88,89. All minima were confirmed by frequency calculations. The adsorption energies Eads were calculated by: \({E}_{{{{\rm{ads}}}}}=1/n\,({E}_{{{{\rm{complex}}}}}-{E}_{{{{{\rm{TiO}}}}}_{{{{\rm{2}}}}}}-n{E}_{{{{\rm{mol}}}}})\) where: Ecomplex, \({E}_{{{{{\rm{TiO}}}}}_{{{{\rm{2}}}}}}\), Emol, n are the total energy of the molecule-TiO2 complex formed by molecule-TiO2 rutile (110) surface, the TiO2 rutile (110) surface, the molecule and number of molecules, respectively. Within this definition of adsorption energy, a negative value indicates an exothermic process. For the DOS analysis a (4 × 4 × 1) k-points grid, PBE functional85 at DFT level of theory with the semi-empirical nonlocal external potentials was used90,91,92. The absorption spectra of O2 were calculated using DFTB+ code93. The repulsive potential for the Ti-O pair was improved for a better description of the physisorption of O2 over TiO2 rutile (110) surface. The set of DFTB parameters tiorg-0-194 were modified and used. To determine the absorption spectra at the real-time TD-DFTB level, a cluster model was used to simulate the TiO2 rutile (110) surface. The cluster is formed by Ti21O68H52 formula, in which the peripherical O atoms were saturated with H atoms to keep the cluster neutral. To obtain the absorption spectra, an initial perturbation to the initial ground-state matrix is introduced. This perturbation has the shape of a Dirac delta pulse, and the density matrix evolves in time. Its evolution can be resolved by time integration of the Liouville-von Newmann equation of motion. For this, an initial electric field of 0.001 V/Å was used.