Introduction

The surface oxygen vacancy (Vo) of metal oxides usually shows significant influences on the molecular adsorption and reactivity on reduced oxide surfaces1,2,3. It can serve as the preferential adsorption site of many molecules, such as O2 4, 5, H2O6, alcohols7, and carboxylic acids7. Moreover, it directly promotes surface reactions such as the dissociation of alcohols and H2O8,9,10, the oxidation of CO over TiO2 1, the coupling and enolization of acetaldehyde on CeO2 11, and the aerobic coupling of amines over defective WO3 nanosheet12. Surprisingly, Vo also prohibits some photocatalytic reactions, e.g., photooxidation of trimethyl acetate13 and formaldehyde14 on TiO2(110) surfaces. It is generally thought that the excess electrons introduced by Vo play a crucial role in the modification of the surface chemical properties of metal oxides15, 16. These excess electrons usually induce the lattice distortion (polarization) in the metal oxides by Coulomb interaction. Such an electron together with its induced polarization in the oxide crystal forms a quasiparticle called a polaron. The polarons are divided into small and large polarons depending on the dimension of the lattice distortion region. For the bare reduced rutile TiO2(110) surface, both recent scanning tunneling microscopy (STM) and infrared spectroscopy (IR) studies confirmed that the excess electrons belong to small polarons17,18,19. That is, the lattice distortion region is of the order of the lattice constant (usually one unit cell), representing the strong interaction between electrons and phonons20. The resonant photoelectron diffraction (RPED) measurement revealed that these small polarons preferentially distribute in the subsurface21, 22.

On adsorbate-covered oxide surfaces, the polaron-involved molecular adsorption is the initial step of many chemical reactions and is important in understanding the surface chemical reactivity15, 16, 23. O2 was the most investigated probe molecule in polaron-involved molecular adsorption studies16 as the “electron scavenger” in photo-oxidation reactions. The O2 with different oxidation states plays different roles in CO oxidation and various photooxidation reactions involving organic molecules on TiO2 under anhydrous conditions24, 25, though the extent of electron transfer between O2 and TiO2 is still an open issue26. On rutile TiO2(110) surfaces, electron energy loss spectroscopy (EELS) studies26 showed that upon O2 adsorption the electronic Ti3+ defect state disappears and is replaced by a new energy loss at ~2.8 eV which is assigned to O2 . This further confirmed the O2 formation during O2 reduction on polycrystalline TiO2 detected by electron paramagnetic resonance (EPR) earlier27. On the other hand, Petrik et al. suggested that O2 chemisorbs on TiO2(110) either in O2 2− or in O2 state depending on the oxygen coverage28. Deskins et al. proposed that for the surface/adsorbate complex significant charge transfer from the reduced surface to the absorbate will occur if the electronegativity of the adsorbate is greater than the surface electronegativity29.

To better understand the microscopic mechanisms governing the interaction of polarons and adsorbates, the experiments using more probe molecules are urgently required. Besides O2 mentioned above, CO and NO are also important probe molecules. CO is the most frequently-used probe molecule whose adsorption behavior is often used to characterize the physical and chemical properties of oxide powder and single crystal surfaces23, 30. On TiO2(110) surfaces, STM study by Zhao et al.31 revealed that the preferentially adsorbed sites of CO are the next-nearest Ti5c sites of Vo rather than Vo sites themselves due to the delocalized distribution of polarons. In contrast to CO, NO molecule possesses an unpaired electron in its 2π* orbital, which leads to complicated coordination chemistry on oxide surfaces23, 32. Recent STM studies revealed that the surface hydroxyls result in an electrostatic stabilization of the adsorbates and significantly enhance the binding energy of NO33, which was also supported by the density functional theory (DFT) calculations33, 34. The contrast study on NO and CO adsorption will give more clear scenarios of polaron-involved molecular adsorption on single crystal oxide surfaces. We expect to throw new light on this important issue by probing the subtle interaction between CO/NO and polarons through detecting the vibration frequencies of adsorbed molecules by the highly sensitive ultra-high vacuum - Fourier transform infrared spectroscopy (UHV-FTIRS)35, 36.

At present, we studied NO and CO adsorption on reduced TiO2(110) surfaces by using UHV-FTIRS and DFT calculations, and found out two typical scenarios of polaron-involved molecular adsorption on TiO2(110). For NO adsorption, the subsurface polaron transfers to a Ti:3d-NO:2p hybrid orbital causing the large redshift of vibration frequency (VF) of NO. While for CO adsorption, the polaron only transfers to the surface Ti5c cation underneath CO, inducing a weak shift of VF of CO.

Results and Discussion

Figure 1a shows the infrared reflection absorption spectroscopy (IRRAS) data of very low dosage of 0.01 L NO on the reduced TiO2(110) surface at 90 K. For the p-polarized incidence plane along both [001] and \([1\bar{1}0]\) directions, two negative IR absorption bands of adsorbed NO were observed at 1751 and 1626 cm−1, respectively, which are largely redshifted relative to that (1875 cm−1) on perfect TiO2(110) surfaces. Based on the sign judgment principle of IRRAS36, both bands correspond to the out-of-plane vibration mode of N-O stretching. To the best of our knowledge, these two IR absorption bands of adsorbed NO have never been reported on single crystal rutile TiO2(110) surfaces.

Figure 1
figure 1

(a) P-polarized IRRA spectra of 0.01 L NO adsorbed on reduced TiO2(110) surface at 90 K. The incidence plane is along \([1\bar{1}0]\) (black curve) and [001] (red curve) direction respectively. (b) P-polarized IRRA spectra of NO-CO co-adsorption on reduced TiO2(110) surface at 90 K: First exposure 1 L CO to the reduced TiO2(110) surface, then 0.01 L NO.

To determine the NO adsorption states on reduced TiO2(110) at 0.01 L NO dosage, co-adsorption experiments of NO with CO or O2 were performed. Firstly, the reduced TiO2(110) surface was first pre-adsorbed with adequate CO (1 L), and then exposed to 0.01 L NO. In this case, as shown in Fig. 1b, only one negative band at 1600 cm−1 was observed for NO besides the vibrational bands (2186 cm−1 and 2190 cm−1) of adsorbed CO. Here, the 1600 cm−1 band has the same origin as the 1626 cm−1 band shown in Fig. 1a, the redshift results from the interaction between different molecules37. Since CO can only adsorb on Ti5c sites rather than Vo sites27, the present 1600 (or 1626) cm−1 band is assigned to the stretching of NO adsorbed at empty Vo sites, while the absent 1751 cm−1 band is assigned to that of NO on Ti5c sites. Secondly, the reduced bare TiO2(110) surface was pre-adsorbed with adequate O2 at 90 K and subsequently exposed to 0.01 L NO. In this case, both 1626 and 1751 cm−1 bands no longer occur in the spectra (See Figure S1 in the SI), while new 1874 cm−1 and 1743 cm−1 bands appear. The latter two bands correspond to the symmetric and asymmetric stretching vibrations of (NO)2 dimer adsorbed on surface Ti5c sites38. Contrary to CO, saturatedly adsorbed O2 molecules occupy all the Vo sites and collect all the excess electrons of reduced TiO2(110) surface at low temperature15. The absence of 1626 cm−1 and 1751 cm−1 bands convincingly indicates that both bands should be related to the excess electrons.

DFT calculations were performed based on a 2 × 4 supercell to further clarify the adsorption configuration of NO and the role of polarons in molecular adsorption by means of comparing the calculated vibrational frequency (VF) of NO (see Table 1) with the experimental values. To verify the feasibility of the theoretical approach, we firstly calculated the distribution of polarons on the pristine TiO2(110) surface with one Vo. The lowest energy state among all possible polaron distributions corresponds to the case that two polarons are respectively localized at two nearest-neighboring Ti6c sites in the subsurface from Vo (See Figure S2 in the SI), which is in good agreement with previous reports39,40,41. Furthermore, the calculation shows that the parallel spin alignment of two polarons and the antiparallel one are almost degenerate in energy, suggesting the weak exchange interaction between two polarons. This is also consistent with the recent calculation results42.

Table 1 Calculated vibrational frequency of adsorbed NO on TiO2(110) and their dependence on the location of polarons.

When one NO molecule adsorbs at Vo sites and none of the surrounding Ti5c cations is occupied, named configuration NO-Vo, as shown in Fig. 2a and b, the binding energy (Eb) of NO molecules adsorbed at Vo is 1.09 eV. In this case, the calculated VF is 1636 cm−1 (2th entry), in good agreement with the experimental IR value (1626 cm−1) but markedly lower than that (1943 cm−1) of isolated NO adsorbed on defect-free surface (1st entry of Table 1) with Eb of 0.61 eV. The calculated density of states (DOS) is shown in Fig. 2c. Within the band gap, there are three states highlighted in different colors. The spin-up state lower in energy (peak I) consists of \({{\rm{Ti}}}_{{\rm{5c}}-{\rm{Vo}}}{:{\rm{d}}}_{{\rm{yz}}}/{{\rm{d}}}_{{{\rm{3z}}}^{{\rm{2}}}-{{\rm{r}}}^{{\rm{2}}}}\) and antibonding \({\text{NO}:{\rm{p}}}_{{\rm{y}}}^{\ast }\) orbitals, whereas the other one higher in energy (peak II) is composed of \({{\rm{Ti}}}_{\mathrm{5c}-\mathrm{Vo}}{:{\rm{d}}}_{{\rm{xz}}}/{{\rm{d}}}_{{\rm{xy}}}\) and antibonding \({\mathrm{NO}:{\rm{p}}}_{{\rm{x}}}^{\ast }\) orbitals. The spin-down state originates from the Ti6c-polaron state. Here, we found that the interaction between the Ti6c-polaron and the adsorbed NO is rather weak. Therefore, to get a clear view of the band structure, in the following discussions the Ti6c-polaron state is artificially located in the minority spin channel. Obviously, both hybridized \({\mathrm{NO}:{\rm{p}}}^{\ast }-{{\rm{Ti}}}_{{\rm{5c}}-{\rm{Vo}}}:{\rm{d}}\) states are occupied, in contrast to that of NO adsorbed on the defect-free TiO2(110) surface, where only one \({\text{NO}:{\rm{p}}}_{{\rm{y}}}^{\ast }\) molecule state is occupied within the band gap (See Figure S3 in the SI). The calculated band structure indicates that the NO adsorbed at Vo possesses an additional electron from the excess electrons of reduced TiO2(110) surface. Such result is further confirmed by the spatial charge distribution of the band gap states shown in Fig. 2a and b: there is only one Ti6c-polaron left but evident charge distribution mainly on NO is observed.

Figure 2
figure 2

Structure, charge distribution of band-gap states (purple and green isosurfaces in the left and middle panels) and the corresponding density of states of NO-adsorbed TiO2(110) surfaces (in the right panel) for different NO adsorption configurations: (a–c) NO adsorbed at Vo; (d–f) NO adsorbed on Ti5c of reduced surface with Vo (see the dashed red circles). Left panel: top view; middle panel: side view. For the DOS given in the right panel, the dashed vertical line denotes the Fermi level. The Ti:d-NO:p hybridized states near the Fermi level are shown by the purple peaks, and the spin-down gap state in green represents the polaron in the subsurface. The colorized states situated in the band gap are all rescaled by a factor of 2 for highlight.

When the subsurface Ti6c-polaron was removed from the supercell, the calculated VF of NO at Vo is 1640 cm−1 (3rd entry), very similar to 1636 cm−1. Such result indicates that the subsurface Ti6c-polaron has little influence on the VF of NO adsorbed at Vo. However, if both excess electrons are removed, the calculated VF of NO at Vo dramatically shifts to 1965 cm−1 (4th entry), similar to that on the defect-free surface (1943 cm−1).

Besides the NO-Vo configuration, for the very low NO dosage at low enough temperature (e.g., 90 K in our experiments), another possible adsorption configuration is that one Ti5c site near Vo site is occupied while the Vo remains unoccupied, named NO-Ti5c & Vo. We found that the adsorption energy of NO on Ti5c sites is hardly related to the distance between NO and Vo. Therefore we select the third-neighboring surface Ti5c site relative to Vo to adsorb one NO, as shown in Fig. 2d. The structural optimization suggests a tilted orientation of the NO molecule, similar to that of NO adsorbed on the defect-free TiO2(110) surface43, 44. The DOS of such configuration with the lowest energy is shown in Fig. 3f, similar to that of NO-Vo configuration, where both hybridized \({\mathrm{NO}:{\rm{p}}}^{\ast }-{{\rm{Ti}}}_{\mathrm{5c}-\mathrm{Vo}}:{\rm{d}}\) states are occupied. The spatial charge distribution of the band gap states (Fig. 2d,e) shows that one Ti6c-polaron remains at the subsurface, and the other one transfers mainly to NO. The calculated VF is 1770 cm−1 (6th entry), in good agreement with our IR value in Fig. 1a (1750 cm−1). Therefore, we conclude that the IR-detected 1750 cm−1 band corresponds to the NO-Ti5c & Vo adsorption configuration with one excess electron (polaron) involved. The corresponding Eb is 0.84 eV. However, if both Ti6c-polarons remain at the subsurface (5th entry), that is, neither polaron involves in the adsorption, the calculated VF of NO is 1942 cm−1 similar to the case of NO adsorbed on defect-free surface (1st entry of Table 1). Then the corresponding Eb is only 0.44 eV.

CO adsorption was also studied to further understand the role of polaron in molecular adsorption. STM and IR results revealed that CO preferentially adsorb on the next-neighboring Ti5c site from Vo31, 45. For this configuration, the calculated electronic structure is shown in Fig. 3. It is found that CO adsorption induces one electron of subsurface Ti6c-polarons to transfer to the Ti5c cation under CO, forming a surface Ti5c-polaron. Unlike NO adsorption, no electron transfers to the orbital state of CO. As a result, although the polaron involves in CO adsorption and thus the surface Ti5c-polaron is no longer degenerate with the subsurface Ti6c-polaron, the VF of CO has got little effect from the surface Ti5c-polaron (exp.: from 2188 cm−1 to 2178 cm−1 (Supplemental Information, Fig. S4; ref. 45); DFT: from 2178 cm−1 to 2155 cm−1). Such little effect is also embodied in the little enhancement of the binding energy, which is less than 0.05 eV.

Figure 3
figure 3

Structure, charge distribution of band-gap states (purple isosurface in the left and middle panels) and the corresponding DOS of CO-adsorbed TiO2(110) (in the right panel). Here, the CO adsorbs on the next-nearest neighboring Ti5c of reduced surface from Vo (see the dashed red circles). Left panel: top view; middle panel: side view.

Deskins et al. proposed that the electronegativity difference between the adsorbates and surfaces governs the surface-to-adsorbate charge transfer29. If the electronegativity of the adsorbates is greater than that of the substrates, the surface-to-adsorbate charge transfer will occur29. For NO and CO on reduced TiO2(110) surfaces in our experiments, the electronegativity of CO (~6.1 eV) is higher than that of NO (~4.5 eV)29, however, the surface-to-adsorbate charge transfer occurs only for NO on TiO2(110) rather than for CO on TiO2(110).

To understand the different scenarios of polarons in NO and CO adsorption, herein, we analyze the energy alignment between the molecular frontier orbitals and surface electronic states. The calculated highest occupied molecular orbital (HOMO), the lowest unoccupied molecular orbital (LUMO) of gas-phased NO and CO as well as the band structures near the band gap of TiO2(110) relative to the vacuum level are shown in Fig. 4. Due to the strong spin-splitting of NO orbitals, only the lower energy levels in the majority-spin channel are shown in this level range. Clearly, the LUMO of NO is at −4.42 eV, higher than the occupied valence band maximum level (−6.93 eV) of defect-free TiO2 by 2.51 eV, but only higher than the polaron state (−5.61 eV) of the reduced TiO2 by 1.19 eV. When NO adsorbs on the Ti5c site of reduced TiO2(110) surface, the small energy level difference (1.19 eV) facilitates the hybridization between the NO LUMO and the polaron state, and the electron transfer occurs from the subsurface polaron to the Ti:3d-NO:2p hybrid orbital. Such involvement mode of polaron in NO adsorption causes the strong intramolecular modification and thus leads to the dramatic shift of NO vibrational frequency. In contrast, the LUMO of the free CO molecule is at −2.13 eV, much higher than that (−5.61 eV) of polaron states of TiO2(110) surface, so their hybridization hardly occurs, and the electron from TiO2 polarons cannot transfer to the molecular orbital of CO. That is, although the adsorption of CO induces the electron from the subsurface Ti6c polaron state to the surface Ti5c polaron state, such surrounding change only induces a weak shift of the CO vibrational frequency. Above analysis does not concern the barrier between two states for the charge transfer. According to our experiments, the barrier should be very low or be lowered by the coupling between adiabatic states proposed in recent DFT calculations for O2 adsorption on reduced TiO2(110) surfaces46.

Figure 4
figure 4

Calculated HOMO and LUMO of gas-phased CO and NO molecules, and band structures of the stoichiometric TiO2(110) surface and the reduced surface (TiO2-Vo) with respect to the vacuum energy level. Note that all the results are spin-degenerate expect for NO.

Conclusion

In summary, the polaron-involved molecular adsorption on reduced TiO2(110) surfaces was studied by IR spectroscopic and DFT approaches using NO and CO as probe molecules. Two typical scenarios were proposed: for NO adsorption, the subsurface polaron electron transfers directly to the empty molecular orbital due to the hybridization between the NO LUMO and the polaron state, the strong intramolecular modification causes the large shifts of the molecular vibrational frequency; for CO adsorption, the subsurface polaron only transfers to the surface Ti5c cation underneath CO, the surrounding change results in a weak shift of the molecular vibrational frequency. These scenarios are determined by the energy-level matching between the polaron state and the LUMO of adsorbed molecules. Although such scenarios of polaron-involved molecular adsorption were obtained on rutile TiO2(110) surfaces, it is prospected that they should be also applicable to other reducible oxides.

Methods

Experimental details

The experiments were carried out in a state-of-the-art UHV-FTIRS system combining a vacuum FTIR spectrometer (Bruker, VERTEX 80 v) and a multichamber UHV system (PREVAC) with the base pressure better than 6 × 10−11 mbar, as described in detail previously36. The clean TiO2(110) surface was prepared by a sequence of Ar+ sputtering and annealing at 800 K in UHV conditions until a high-quality (1 × 1) low energy electron diffraction (LEED) pattern was obtained. During the preparation, Auger electron spectroscopy (AES) was used to monitor the clearance of the sample surface. After several times of such treatment, about 8% surface bridge oxygens were removed on the sample. The IR measurements were performed using infrared reflection absorption spectroscopy (IRRAS) mode with a fixed incidence angle of 80° respectively along \([1\bar{1}0]\) and [001] crystallographic directions. The IRRAS data were recorded at 90 K with 2048 scans at 4 cm−1 resolution. In experiments, high purity NO (99.9%) and CO (99.99%) were dosed via backfilling.

Computational details

First-principles calculations were performed using the Vienna ab-initio simulation package (VASP)47 with a cut-off energy of 500 eV for the basis set. Γ-point is used for Brillouin zone sampling. The projector-augmented wave method (PAW)48 with the PBE type49 exchange–correlation potentials was adopted. To correctly model the localization of excess electrons in rutile TiO2 17, the DFT + U method with UTi:3d = 4.2 eV50 is employed. Furthermore, a van der Waals dispersion correction based on the D2 method is included in the calculation34, 51. The vibrational frequencies are derived from Hessian matrix calculated by finite-displacement method. To model the TiO2(110) surface, the experimentally determined lattice parameters of bulk rutile TiO2, a = 4.594 Å and c/a = 2.95952, were used to build a slab including four tri-layers and a vacuum layer with a thickness of 15 Å. A supercell with a p(4 × 2) geometry along \([001]\) and \([1\bar{1}0]\) directions, respectively, was employed to perform the calculations. The atomic positions of top three tri-layers were optimized until the forces are less than 0.02 eV/Å, while the bottom tri-layer were fixed at bulk positions and were terminated by pseudo-hydrogen atoms53, 54. To manipulate the location of polaron, a preliminary lattice-distortion approach40 was employed, where the bonds between the targeted Ti cation and its neighboring O anion were initially stretched by ~0.1 Å before the structural optimization. During the calculation, removing the subsurface Ti6c-polaron from the supercell was realized by altering the number of valence electrons.