The structural identification of VO_x/SiO₂ has been the subject of many spectroscopic studies including Raman, infrared, UV-vis-NIR, EXAFS, NMR, and EPR ^{[1, 3, 24, 25, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40]}. It is generally agreed that vanadium oxide is present on the dehydrated silica surface in two different forms, i.e., as monomeric VO₄ species at low loadings (typical < 2 VO_x/nm²) and as mixture of monomer and V₂O₅ crystals at high loadings (> 2 VO_x/nm²) ^[38]. It has been long accepted that a pyramidal model (O=V-(O-Si)₃) is the only structure for dehydrated VO_x/SiO₂ at low VO_x loading (< 2 V/nm²) ^{[31, 41]} and thus VO_x/SiO₂ has been considered as an example of a nearly single-site heterogeneous catalyst.

Recently, there has been renewed experimental ^{[39, 40, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51]} and theoretical ^{[42, 49, 50, 52, 53, 54, 55, 56]} interest in what precisely the geometrical molecular structure of surface vanadia on silica is under dehydrated condition. This is because the pyramidal model cannot satisfactorily explain all the experimental data collected over the years especially the uncertainties about specific spectral assignments in the vibrational spectra as illustrated in studies by Magg et al. ^[42], Gijzeman and co-workers ^{[43, 44, 45]}, and Moisii et al. ^{[46, 47]}. It has been realized that there is strong vibrational coupling between the vanadia species and the silica support ^[42], which makes the vibrational spectra of vanadia/silica more complex and difficult to interpret. Especially controversial is the assignment of the vibrational band at around 915 cm^-1 which was assigned to V-O-V ^{[40, 57]}, V-O-support mode ^{[6, 36, 42, 48, 49, 52, 53, 54, 55, 56]}, O-O mode ^{[43, 44, 58]}, or V=O mode ^[46]. Controversies also exist on the assignment of the Raman bands above 1000 cm^-1 where the vanadyl stretch and silica modes are close in vibrational energy and thus strongly coupled together ^{[42, 46]}. Based on the different assignments, three structures have been suggested for monomeric VO₄ species anchored at the silica surface under dehydrated co

nditions, namely, pyramidal, bi-grafted species (two V-O-Si, one V=O and one V-OH), and the so-called umbrella structure (one V-O-Si, one V=O and a perturbed oxygen molecular bonded to V). A recent study by Wachs and coworker ^[59] demonstrates definitively with Raman and UV-Vis spectroscopy that the surface vanadia umbrella-like structure is not present for both hydrated and dehydrated supported vanadia catalysts via comparing with the vibrational spectrum of K[VO(O₂)(heida)] where the umbrella structure is present. Up to date, both experimental and theoretical work have concluded that the traditionally accepted pyramidal model is the most likely structure for stable VO_x species under highly dehydrated condition. However, even the exclusive presence of monomeric vanadia species on silica surface was challenged recently by Schlogl and co-workers ^{[60, 61]} via comparing the theoretical and experimental NEXAFS spectra of a silica SBA-15 supported vanadia catalyst. They suggested that under in situ conditions different molecular vanadia species, in particular non- monomeric VO_x, might exist at the catalyst surface. This was resonated in a recent UV-Vis study ^[62] of vanadia supported on hexagonal mesoporous silica where differently coordinated oli gomeric VO_x species are proposed to exist on silica surface in addition to the presence of monomeric VO_x. Clearly, there is still controversy on the exact molecular structure of VO_x species present on a dehydrated silica surface despite of the extensive experimental and computation studies.

In this case study, we investigate the molecular structure of VO_x/SiO₂ catalytic system using MW Raman spectroscopy including RR and non-resonant Raman ^[27]. The advantages of MW Raman spectroscopy over conventional single wavelength Raman spectroscopy have been highlighted in recent studies of supported VO_x catalysts ^{[6, 7, 8, 21, 42, 46]}. Excitation of Raman spectra within the electronic absorption region of surface vanadia species will produce RR spectra from a subset of VO_x species and/or local V-O bonding with absorptions at/near the excitation wavelength. Through measurement of the Raman spectra at several wavelengths, more detailed structural information can be obtained not only on the various VO_x species in the distribution but also the local V-O bondings in a specific VO_x species.

The molecular structure of vanadia on silica over wide range of surface VO_x density (0.0002-8 V/nm²) has been investigated after dehydration at 773 K by in situ MW Raman spectroscopy (laser excitations at 244, 325, 442, 532, and 633 nm) and UV-Vis diffuse reflectance spectroscopy. Fig. 2(A) shows an example of the MW Raman spectra of a dehydrated 0.25V sample (referring to a VO_x/SiO₂ sample with surface VO_x density of 0.25 V/nm²). Generally, the spectra excited by visible lasers (442, 532 and 633 nm), the “visible Raman” spectra, are similar to each other while quite different from those excited by UV lasers (244 and 325 nm), the “UV Raman” spectra, in terms of spectral contour and bands position. This is due to the fact that the 0.25V sample barely absorbs in the visible region (see inset in Fig. 2(A) for UV-Vis DRS spectrum) while it has strong electronic absorption in the UV region (below 400 nm) which may offer resonance enhancement effect in the UV Raman spectra. The visible excited spectra show a sharp band at 1041 cm^-1 that is related to the surface VO_x species and usually assigned to vanadyl (V=O) stretch. In the range below 1000 cm^-1, Raman bands are observed at 976, ~800, 605, 484 cm^-1, all due to the SiO₂ support, have been assigned to Si-OH stretch, Si-O-Si stretch, D2 defect mode, and D1 defect mode, respectively ^{[36, 48, 63]}. For UV excited RR spectra, no apparent features due to SiO₂ can be observed, consistent with the stro ng absorption of the surface VO_x species in the UV region. The characteristic V=O stretching band is observed at 1032 cm^-1 in the UV Raman spectra for both 325 and 244 nm excitation. This band is about 9 cm^-1 lower than that observed in the visible Raman spectra and is probably responsible for the lower Raman shift tailing of the 1041 cm^-1 band in the visible Raman spectra. The observation of two different V=O stretching bands suggests the presence of two types of VO_x species on the silica surface. The 325-nm excited spectrum exhibits an additional sharp band at 461 cm^-1 and a weak one at 920 cm^-1. The band at 920 cm^-1 is assigned to interface V-O-Si mode. The band at 461 cm^-1 is usually ascribed to the bending mode of V-O because of its low frequency. However, its sharpness is in general conflict with spectral characteristics of bending modes. It is thus likely a stretching mode and has recently been assigned to primarily V-O-Si stretch ^[46]. The 244-nm excited spectrum shows a shoulder at 1060 cm^-1 on the intense band at 1032 cm^-1. The shoulder band at 1060 cm^-1, apparently resonance enhanced by 244-nm excitation, was previously assigned to silica network TO mode and Si(-O^-)₂ and Si-O^- functionalities ^{[32, 36]} but recent theoretical work suggests that it is due to the V-O-Si stretching mode ^{[42, 46, 55, 56]}. Weak bands are also observed at 920, 690 and 570 cm^-1 in 244-nm-excited spectrum, assignable to V-O-Si stretch, and V-O bending modes, respectively. The observation of these new bands below 1000 cm^-1 in the UV Raman spectra is strong indication of resonance enhancement of certain modes of surface VO_x species. Detection of different V=O stretching bands is evidence for the presence of at least two different VO_x species. The edge energy obtained from the UV-Vis-DRS spectrum is about 3.4 eV, very close to that of monomeric vanadia ^{[31, 64, 65]}. So we propose that the two VO_x species are monomeric in nature.

Fig. 2. (A) In situ multi-wavelength Raman spectra of dehydrated 0.25V/SiO2 sample. Inset gives the in situ UV-Vis DRS spectrum. (B) Distribution of surface monomeric species A and B and crystalline V2O5 as a function of surface VOx density. The structures of species A and B are illustrated. Adapted from Ref. [27].

The MW Raman spectra of VO_x/SiO₂ were also collected as a function of VO_x density up to 8 V/nm². V₂O₅ crystals are developed at surface density above 2 V/nm² in the visible Raman spectra but barely observable in the UV Raman spectra. Under non-resonant Raman condition (442, 532 and 633-nm excitations), the 1032 cm^-1 band is weak and buried in the stronger V=O stretching band at 1041 cm^-1. This implies that the monomeric VO_x species giving V=O stretch at 1041 cm^-1 (type A) dominates on the silica surface while the other monomeric species (type B) exhibiting V=O stretch at 1032 cm^-1 is the minor one. Assuming that the Raman cross section of the V=O in the two species are the same under 532 nm excitation (no resonance enhancement) and the Raman cross section of V₂O₅ is known 10 times larger than dispersed vanadia species ^[32], a quantitative determination of the distribution of different surface VO_x species can be obtained by peak fitting the 532-nm excited spectra (not shown here, see Ref. ^[27]). As shown in Fig. 2(B), the percentage of species B (with V=O stretch at 1032 cm^-1 detected by UV Raman) continues to drop while that of species A (with V=O stretch at 1041 cm^-1 detected by visible Raman) increases at low loadings and then decreases at higher loadings. This indicates that VO_x species may be anchored on the silica surface initially as species B which transforms partially into species A at higher VO_x loadings. This is con sistent with the detection of species B by UV excitations at VO_x density as low as 0.0002 V/nm^{2 [27]}. Both VO_x species transform to crystalline V₂O₅ at higher surface density as evidenced by the increasing percentage of V₂O₅.

Since pyramidal structure has been considered the most stable form for dispersed VO_x on highly dehydrated SiO₂ ^{[42, 52, 53, 54, 55, 56]}, it is highly likely that the major VO_x species A observed in our non-resonant visible Raman spectra possesses such a structure with three V-O-Si bonds and one V=O bond (Fig. 2(B)). The exact molecular structure of species B observed by RR is not clear, but likely associated with surface hydroxyl groups. Since this species is anchored on silica surface prior to species A, the silica surface is more hydroxyl group-rich when species B is formed. This surface is analog to a partially hydrated one on which the V=O stretching frequency of hydrated VO_x is generally lower than that of a dehydrated species ^{[31, 32]}. Therefore, species B may have a partially hydroxylated pyramidal structure where a hydroxyl group is either hydrogen-bonded to the bridging oxygen atom in the V-O-Si bonds (designated as hydrogen-bonded pyramidal structure) or bonded directly to the V center (similar to bi-grafted structure) as illustrated in Fig. 2(B). Further theoretical modeling is warranted to clarify a clear picture of structure of the VO_x species B. It is noted that similar vanadia structures were later adapted by Chlosta et al. ^[51] for vanadia supported on silica supports with different surface areas. They also observed two Raman bands of V=O mode at 1027 and 1040 cm^-1, with the former band assigned to hydroxylated vanadia species while the later to dehydrated species, similar to our proposed models.

One of the characteristics of RR is the high possibility of observing overtones and combination bands which could provide insight for the symmetry and electronic properties of the fundamental vibrations and help for the vibrational assignment of fundamental modes. We observed overtones and combination bands for most of the VO_x/SiO₂ samples (loading > 0.001 V/nm²) only when excited by the two UV lasers due to RR effect. As an example, the full range spectra of 0.25V excited by 325 nm before and after ¹⁸O exchange are shown in Fig. 3. For unexchanged sample, fundamental bands are observed at 462 (V-¹⁶O-Si, denoted as ν_a), 920 (V-¹⁶O-Si, ν_b), and 1032 (V=¹⁶O, ν_c). Overtones are clearly observed at 2052, 3059, 4065, and 5064 cm^-1, corresponding to 2ν_c, 3ν_c, 4ν_c, and 5ν_c, respectively. To our knowledge, this is the first resonance Raman spectrum showing overtones of vanadyl stretch up to 4^th order. These observations are generally consistent with resonance enhancement by Albrecht A term ^{[66, 67, 68, 69]}, which indicates that ν_c is a totally symmetric mode with a large displacement of the vanadium oxo bond length in the excited state. Bands at 1362, 1493, and ~1831 cm^-1 are also observed beyond the fundamental vibrations and can be assigned to combination bands of (ν_a + ν_b), (ν_a + ν_c), and overtone of ν_b, respectively. Upon complete oxygen isotopic exchange, all the fundamental bands, overtones and combination bands are shifted to lower Raman shifts as shown Fig. 3. Since a control experiment on the pure SiO₂ support does not show obvious isotopic exchange under current experimental condition, the shifts upon oxygen isotopic exchange observed on 0.25 V confirms that the bands observed above fundamental region are indeed due to overtones and combination bands of surface vanadia species.

Fig. 3. Full range Raman spectra excited by 325 nm for 0.25V/SiO2 collected at room temperature before and after 18O isotopic exchange. Inset shows the plots of νV=O(n)/n versus n for the overtone progressions of V=O stretch. Adapted from Ref. [27].

The availability of overtones generally allows the calculation of the harmonic wavenumber and anharmonic constant of the fundamental V=O mode ^[8], permitting that the fundamental V=O mode is close to a diatomic stretch. This is not straightforward for VO_x/SiO₂ system since the V=O mode couples strongly with the silica support ^[42]. Fortunately, normal modes analysis of the vibrational property of vanadia supported on silica surfaces has been recently carried out using density functional theory (DFT) ^[56]. The two cluster models, VO_x/hexagonal prism (resembling a completely dehydrated silica surface without hydroxyl groups) and VO_x/Edingonite (resembling partially dehydrated silica surface with some hydroxyl groups) used in Dobler’s work ^[56] can be used to represent our dehydrated VO_x/SiO₂. The results on both models showed that the vanadyl mode in monomeric vanadia species had dominating contribution of ~90% from V=O stretch and thus was almost separated from the silica bulk mode. This provides strong support that the V=O mode in our VO_x/SiO₂ system can be treated as a pseudo-diatomic V=O stretch and the diatomic approximation analysis associated with overtones of V=O mode can be carried out.

The overtones of V=O (ν_c) up to 4^th order (Fig. 3) are used to calculate the harmonic wavenumber, ω_e (= ω_m, the vibrational frequency corrected for anharmonicity), and anharmonic constant ω_ex_e (= x_mm) ^{[8, 70]}. The observed wavenumbers, ν_m(n), for fundamental (n = 1), first to fourth overtone (n = 2-5) from a polyatomic anharmonic oscillator are given by the expression ν_m(n) = nω_m - n(n + 1)x_mm, where m is the normal mode identifier. A plot of ν_m(n)/n versus n should be a straight line of slope x_mm and an intercept, which gives ω_m - x_mm and thus ω_m. Plots of the overtone progressions of ν_V=O (ν_c) in 0.25 V before and after ¹⁸O exchange are shown in the inset in Fig. 3. The obtained harmonic frequency (ω_e) allows the determination of anharmonicity-corrected force constant (f_e) using the diatomic oscillator approximation from the equation ω_e = (2π)^-1(f_e⁄μ)^0.5, where μ is the reduced mass. The calculated f_e for V=O (1032 cm^-1) and V-O-Si (920 cm^-1) are 7.76 and 4.57 mdyn/Å, respectively, matching well with the general trend f(V=O) > f(V-O). The value of f_e for V=O is in agreement with the V=O force constant from VO_x/Al₂O₃ (7.37-7.58 mdyn/Å) ^[8] and vanadium bulk compounds (7.10-7.70 mdyn/Å) ^[71]. The value of f_e for V-O-Si is less than that for V-O-Al in VO_x/Al₂O₃ (~6.40 mdyn/Å) ^[8], consistent with the picture that vanadia species anchor weaker on SiO₂ than on Al₂O₃. The dissociation energy of the two modes at 1032 and 920 cm^-1 was also calculated in the range 628-608 and 517-574 kJ/mol, respectively ^[27].

Making use of the results from the analysis of the overtones and combination bands, namely, the symmetry, force constant, and dissociation energy associated with the fundamental modes at 920, 1032, and 1060 cm^-1, we were able to assign the three bands to in-phase V-O-Si, V=O stretch, and out-of-phase V-O-Si modes, respectively along with the aid of other experimental observations and the DFT calculations as illustrated in our recent work ^[27].

The redox property of the two monomeric VO_x species was tested by hydrogen reduction followed by in situ RR and non-resonant Raman spectroscopy. Fig. 4(A) and 4(B) respectively give the 244- (detecting species B, partially hydroxylated monovanadate) and 442-nm (detecting mainly species A, pyramidal monovanadate) excited Raman spectra for 0.25V sample collected at room temperature after different reduction temperatures. For the 244-nm excited RR spectra, the intensity ratio of the bands at 1032 to 1060 cm^-1 decreases as reduction temperature increases. This intensity ratio change is in the reverse direction to the case when anchoring more VO_x on silica surface ^[27], indicating the coverage decrease of species B due to hydrogen reduction. By comparing the intensity ratio of the two bands at 1032/1060 cm^-1 to that from the spectra of dehydrated VO_x/SiO₂ samples by 244-nm excitation, it is estimated that about 90% of species B has been reduced at temperature higher than 973 K. The observation of rather strong Raman features due to un-reduced VO_x even after 1073 K reduction is attributed not only to the resonance enhancement effect but also to the decreased self-absorption ^[72] because supported vanadia oxide shows less absorption in the UV region when reduced ^{[73, 74]}. While in the 442-nm excited spect ra, a continuous decrease in intensity of the band at 1041 cm^-1 is observed with increasing reduction temperature. This band disappears at reduction temperature of 973 K and above as a result of reduction of species A (and species B). But the disappearance of the V=O band does not necessarily indicate a complete reduction of VO_x species because the Raman intensity decrease can also be partially caused by the increased self-absorption effect ^[72] of the reduced sample in the visible Raman region ^[74]. Nevertheless, most of species A and B seems to have been reduced at high temperatures. Reoxidation in oxygen is able to restore the reduced vanadia oxide species back to oxidized ones since the Raman spectra of freshly calcined and reoxidized samples resemble with each other. Therefore, it appears that VO_x species A and B have no significant difference in reduction and oxidation behaviors and thus may both participate in redox reactions catalyzed by VO_x/SiO₂. Further in situ Raman study under redox reaction conditions such as alkane oxidative dehydrogenation may provide insights into the catalytic function of the two different VO_x species.

Fig. 4. RR spectra (244 nm) (A) and non-resonance Raman spectra (442 nm) (B) for 0.25 V sample collected at room temperature after hydrogen reduction at different temperatures (673-1073 K). Adapted from Ref. [27].

Although CeO₂ has been extensively used as catalyst and catalyst support for oxidation reactions, it has been relatively less researched as a support for vanadia than other oxide supports. Recent studies have shown that ceria supported vanadia are very active and selective for the oxidative dehydrogenation (ODH) of ethane, propane and methanol ^{[75, 76, 77, 78, 79, 80, 81, 82, 83]}. Consequently a few research has devoted to the structure of surface vanadia and the interaction between vanadia and ceria [3,75-78, 84-90]. Although some information was obtained on the formation of CeVO₄ from reaction between surface vanadia and ceria support ^{[75, 76, 77, 78, 84, 85, 88]}, less was provided on the surface vanadia species ^{[76, 77, 87, 89, 90]}. This is due to the fact that, in Raman spectroscopy which is considered the most powerful technique for characterizing surface metal oxide species ^{[2, 3]}, the signal from surface vanadia species is rather weak compared to the strong Raman scatterers, CeO₂ and CeVO₄ in the vanadia-ceria system. Only recently was the atomic structure of surface vanadia species deposited on CeO₂(111) thin film unambiguously resolved by combining high- resolution scanning tunneling microscopy (STM) and infrared reflection absorption spectroscopy (IRAS) in UHV system ^{[81, 91]}. A direct relationship was established between the nuclearity of vanadia species (monomeric vs. polymeric) by STM and the vanadyl frequency by IRAS. Still, the exact molecular structure of surface dispersed vanadia on polycrystalline ceria has not been completely identified yet.

The addition of vanadia onto ceria was proposed to titrate the most labile surface oxygen of ceria, thus taming the combustion property of ceria ^{[76, 77]}. The strong interaction between vanadia and ceria leads to the formation of CeVO₄, which is facilitated by increasing VO_x loading, thermal treatment and reducing environment ^{[76, 77, 78]}. The presence of CeVO₄ may explains the fact that V remains in the highest oxidation state (+5) upon vanadia anchorage while some of the surface Ce cations are reduced to Ce³⁺. Recent DFT calculations also suggest that ceria surface stabilizes the vanadia species in a + 5 oxidation state, which wet and reduce the ceria surface ^{[79, 91, 92, 93]}. Despite of these investigations, the nature of the interaction between surface vanadia and ceria is still intriguing and worth of further investigation.

In this case study, we take advantage of the MW Raman spectroscopy to study not only the detailed structure of vanadia on ceria nanoparticles but also how the two interact with each other ^[30]. Furthermore, the shape effect of ceria support on the speciation of vanadia and the formation of CeVO₄ were also studied through the MW Raman approach ^{[28, 29, 94]}. It is expected that structural information obtained from this study would provide fundamental basis for understanding the vanadia-ceria catalysis.

The MW Raman spectra for dehydrated 1V (VO_x/CeO₂ sample with VO_x density of 1 V/nm² on Aldrich ceria nanoparticles) are given in Fig. 5(A) to show the laser excitation effect. The full range spectra from visible excitations (442, 532, and 633 nm) are dominated by the ceria F_2g mode at 464 cm^{-1 [95, 96, 97, 98]}. A comparatively weak band at 1015 cm^-1 is observed for surface vanadia species, due to V=O stretch ^{[2, 3, 77, 86]}. The 325-nm excited Raman spectrum is also dominated by features due to ceria but with the defect-related band at 590 cm^-1 and the second longitudinal optical mode (2LO) at 1179 cm^-1 greatly enhanced ^{[95, 96, 97, 98]}, due to RR effect as observed recently ^{[95, 96]}. The 244-nm excited spectrum show rather low signal/noise ratio with the F_2g mode observable, this is likely due to the strong self -absorption of ceria in the deep UV region ^{[72, 96]}. Nevertheless, the full range spectra for all laser excitations of 1V are dominated by features from the ceria support as crystalline ceria is an excellent Raman scatterer and thus overwhelms signal from surface vanadia species. To get detailed information on the surface vanadia species, the spectral region was chosen to avoid the strong features of ceria and extended spectral acquisition time was used. As shown in the inset in Fig. 5(A), features due to surface vanadia are now prominently observed in the visible Raman spectra while the UV Raman spectra show little sign of surface vanadia species. The sharp band at 1015 cm^-1 along with a shoulder at 1033 cm^-1 are due to vanadyl stretch of surface dispersed vanadia, while the broad bands at ca. 857 and 708 cm^-1 are due to V-O-Ce bridging modes ^{[77, 86, 92]}. The signal from surface vanadia becomes more prominent when the laser excitations move to longer wavelength, due to the decreased self- absorption from the ceria support at longer wavelength ^{[96, 97]}. Therefore, in this case of VO_x/CeO₂, visible laser excitations, namely, non-resonance Raman, are better suited for further investigation of the structure of surface vanadia. Meanwhile, the RR with UV excitation at 325 nm can be utilized for studying the defect sites information on VO_x/CeO₂ system since the RR is very sensitive for monitoring the defect sites (mode at 590 cm^-1) in ceria ^{[95, 96, 97, 98, 99]}

Fig. 5. (A) Multi-wavelength Raman spectra of dehydrated 1V/CeO2 sample collected at room temperature. Inset shows the narrow range spectra with extended acquisition time. (B) 633 nm-excited Raman spectra of dehydrated VOx/CeO2 as a function of VOx density. Spectra for ceria and CeVO4 are also included for comparison. * denotes plasma lines from the laser. Adapted from Ref. [30].

Fig. 5(B) shows the 633-nm excited Raman spectra of dehydrated VO_x/CeO₂ samples with VO_x surface density ranging from 0.05 to 10 V/nm². In the V=O stretching region (950-1100 cm^-1), two weak bands are readily observed at 1008 and 1015 cm^-1 on the 0.05V sample. They grow in intensity and move to higher wavenumbers when more VO_x is loaded on ceria surface. On 5V and 10V samples, three distinct V=O modes are developed at 1022, 1030, and 1044 cm^-1. In addition, V₂O₅ also forms on 10V sample with its characteristic V=O stretch observed at 994 cm^-1. The observation of multiple V=O modes indicates the co-existence of different vanadia species on ceria surface. In the spectral region below 950 cm^-1, the V-O-Ce mode shifts from 830 to 857 cm^-1 with VO_x loading and is prominent when the surface VO_x density is low. Above 5 V this mode sharpens and is visible at 862 cm^-1. A separate broad feature at ca. 930 cm^-1 appears at VO_x loading above 1 V/nm², attributable to V-O-V mode in polyvanadate ^{[77, 86, 87, 90]}. On 10V sample, sharp peaks due to CeVO₄ develop at 787, 800, and 862 cm^-1 on the broad background of V-O-Ce bridging modes. It appears that the monolayer of surface dispersed vanadia species on ceria is valued somewhere between 5 and 10 V/nm².

The atomic structure of the surface vanadia species can be pictured by borrowing from a recent model study of vanadia on CeO₂(111) thin film ^[91]. By combining STM and surface infrared spectroscopy, a direct structure-spectroscopy relationship was established for VO_x/CeO₂(111) system: monomeric VO_x with V=O mode at the lowest wavenumber while polymeric VO_x having V=O stretch at higher wavenumbers. Considering that the V=O stretches observed in our study fall within the same range (1006-1045 cm^-1) as in the IR study of VO_x/CeO₂(111) system ^[91], we can analogously assign the band at 1008 cm^-1 to monomeric VO_x, ~1015 cm^-1 to dimeric VO_x, 1022-1030 cm^-1 to trimeric VO_x, and 1044 cm^-1 to polymeric VO_x. The increase in V=O frequency is due to the increasing dipole-dipole coupling between V=O oscillators in vanadia species with higher polymerization degrees. It clearly shows that surface vanadia on ceria evolves from monomeric to dimeric, trimeric, polymeric VO_x, and eventually to crystalline V₂O₅ and CeVO₄. This is the first time a detailed structural evolution of VO_x species has been obtained on a powder ceria support.

Interestingly, several types of dispersed VO_x species can co-exist on CeO₂ surface, indicating the presence of a variety of surface sites on ceria that are able to anchor and stabilize VO_x species. This was rarely observed for VO_x supported on other oxide materials where only one type of 2-dimensional surface VO_x species was usually detected ^{[2, 3, 5]}. For better control of surface vanadia species, a more homogeneous ceria surface, such as ceria nanoshapes with defined surface planes, may be helpful. Recent advancement of nanomaterial synthesis makes such materials available ^{[100, 101]}. Taking advantage of the nanoshaped ceria, rods (mainly (110) facet), cubes ((100)), and octahedral ((111)), we investigated how the surface structure of CeO₂ can affect the structure of surface vanadia (VO_x) species by in situ MW Raman spectroscopy ^{[28, 29]}. The speciation of surface VO_x was found si milar on the nanoshaped and polycrystalline ceria, that is, monomeric, dimeric, trimeric, and polymeric VO_x as well as crystalline V₂O₅ and CeVO₄, are observed as a function of surface VO_x density. However, the presence of a single monomeric VO_x species was only observed on the ceria nanoshapes but not on the polycrystalline ceria in which monomeric and dimeric VO_x always coexist even when the surface VO_x density is quite low, suggesting a rather uniform surface structure on the nanoshaped ceria.

Two intriguing characteristics of ceria as catalyst and catalyst support are the defect sites and the labile surface oxygen ^[102], so we investigated how these properties are altered when VO_x species are anchored on the ceria surface. UV RR is employed to study how the VO_x species interact with the defect sites in CeO₂ since this technique has shown to be very sensitive to the changes of defects in ceria ^{[95, 96, 97, 98]}. Fig. 6 gives the 325-nm excited Raman spectra of dehydrated VO_x/CeO₂ samples. Due to the strong self-absorption of ceria, Raman features due to surface vanadia species are observed only when it reaches certain surface coverage (≥ 1 V/nm²). Furthermore, the bridging modes (V-O-Ce) are more prominent than the V=O stretch which shows up at 1030 cm^-1 due to trimeric VO_x. These differ drastically from the 633-nm excited spectra (Fig. 5(B)) where up to three V=O modes are observed, again suggesting that visible or near-IR excitation is preferred for studying the structure of VO_x supported on ceria. The selective observation of just one V=O mode at 1030 cm^-1 by 325-nm excitation is likely a RR effect, indicating that VO_x trimer may exhibit electronic absorption band near 325 nm. Except for the features due to surface VO_x, the 2LO (1179 cm^-1) and defect-associated (D band, 590 cm^-1) bands are greatly resonance enhanced in intensity relative to the F_2g mode of ceria at 464 c m^-1. It is clear that all these bands due to ceria are attenuated when VO_x species are anchored on the surface, especially obvious is the decease of the D band intensity relative to the F_2g band. By plotting the ratio of the D band to F_2g band (I_D/I_F2g), the relative amount of defect sites in ceria can be described. As shown in the inset of Fig. 6, The I_D/I_F2g ratio, i.e., the amount of defect sites, decreases continuously with the increasing loading of VO_x, especially at lower VO_x loadings, implying that the surface VO_x interacts closely with ceria defect sites so that the defect sites are either annihilated or altered in Raman cross section. This observation is also clearly show in the case of ceria nanoshapes-supported VO_x systems ^[28].

Fig. 6. 325 nm excited Raman spectra of dehydrated VOx/CeO2 as a function of VOx density. Spectra for ceria and CeVO4 are also included for comparison. Inset shows the plot of band intensity ratio between 590 and 464 cm-1 as a function of surface VOx density. Adapted from Ref. [30].

To understand how the surface reactive oxygen of ceria interacts with VO_x species, we used visible Raman (532-nm excitation) spectroscopy coupled with O₂ adsorption to study how the induced defect sites in the VO_x/CeO₂ samples change after hydrogen reduction. Reduction removes the most reactive lattice oxygen and creates oxygen vacancies which can adsorb O₂ that is detectable by Raman spectroscopy at low temperatures ^{[95, 103]}. Fig. 7 shows the Raman spectra from O₂ adsorption at 100 K on 773 K-reduced ceria, 0.1V, 0.5V, 1V and 2V samples. On reduced ceria, three Raman peaks due to adsorbed oxygen species are detected at 1137, 890, and 834 cm^-1, attributed to superoxide (O₂^-), peroxide (O₂^2-) on clustered vacancies and peroxide on isolated vacancies ^{[2, 95, 103, 104]}, respectively. These three peaks are observed to decrease in intensity when VO_x species are loaded on the ceria surface. Particularly the two peaks at 1137/1130 and 890 cm^-1 decline faster than the one at 834/843 cm^-1, suggesting that surface VO_x species interact preferably with the lattice oxygen sites that would otherwise produce these one-electron vacancy and clustered two-electron vacancy sites upon reduction. It appears that the reactive surface oxygen sites of ceria are gradually consumed via anchoring VO_x species because the three Raman bands of adsorbed O₂ species are barely detectable on the reduced 2V sample. In other words, surface VO_x species passivate the reactivity of ceria surface oxygen sites via bridge bonding between them. This passivation is further supported by the H₂-temperature programmed reduction results ^[30] on the VO_x/CeO₂ samples which showed that the reducibility of ceria surface oxygen is suppressed by anchoring VO_x species.

Fig. 7. Raman spectra of O2 adsorption at 100 K on 773 K-reduced ceria, 0.1V, 0.5V, 1V, and 2V samples. Adapted from Ref. [30].

Therefore, it appears that the two intriguing properties of ceria, defect sites and labile surface oxygen, both play an essential role in the anchorage of VO_x species on ceria. This well explains the decreased total oxidation capability and increased selectivity of VO_x/CeO₂ in oxidative dehydrogenation reactions when compared with bare ceria ^{[75, 76, 77, 78, 80]}.

As show above, when the loading is high, vanadia can react with ceria to form CeVO₄. The V-O-Ce bond in CeVO₄ was considered as the active site for alkaFig.ne ODH reactions ^{[76, 77, 78, 105]} and therefore the formation mechanism of CeVO₄ has attracted considerable research interest [75-78,84,85, 88,105]. Its formation was found facilitated not only by high vanadia loading, but also by some other factors such as high temperature and reducing environment ^{[77, 84]}. However, the role of the surface structure of ceria support in the formation of CeVO₄ has not been investigated. The availability of ceria nanoshapes with different surface facets provides us an opportunity to carry out such study ^[28].

The formation of CeVO₄ from the thermal reaction of surface vanadia species with ceria was compared on the three ceria nanoshapes (rods, cubes, and octahedra) with a similar surface VO_x density (2V/nm²) to give similar structure (mixture of dimeric and trimeric vanadia). Taken 2V/CeO₂-rod as an example, the visible Raman (for monitoring VO_x change and CeVO₄ formation) and UV Raman (for monitoring defect sites change) spectra are shown in Fig. 8 as a function of calcination temperature in an oxidative environment. As shown in Fig. 8(A), The Raman bands due to dimeric and trimeric VO_x species start to decrease in intensity after calcination at 923 K, meanwhile the sharp band at 864 cm^-1 of CeVO₄ takes shape. Faster decrease of the V=O mode of trimeric VO_x than dimeric VO_x is observed as the calcination temperature elevates, indicating easier reaction of more polymeric VO_x species with ceria to form CeVO₄. At calcination temperature of 1073 K and above, CeVO₄ starts to evolve into a different compound, possibly CeVO₃, as evidenced by the appearance of two new sharp peaks at 921 and 953 cm^-1at the expense of the decrease of the peak at 864 cm^-1. The UV Raman spectra (Fig. 8(B)) following the same calcination process give quite different information. The appearance of Raman bands characteristic of CeVO₄ is observed at about 100 K higher temperature than the visible Raman, due to the strong absorption of ceria support. Big change is seen for the Raman band (D band) due to defects in c eria: it continues to drop in intensity with calcination temperature. The decrease can be ascribed to both thermal annihilation and chemical consumption (by reacting with VO_x) of defect sites in ceria. But the trend of the intensity change, shown in Fig. 8(C) as a function of temperature, of the D band relative to F_2g mode (I_D/I_F2g) in the 2V/CeO₂-rod sample is more in line with that of I_V=O/I_F2g (from visible Raman in Fig. 7(A)), rather than with that of I_D/I_F2g in pure ceria rods. This indicates that the formation of CeVO₄ plays a larger role in consuming the defect sites of CeO₂ than the pure thermal annihilation when the temperature elevates.

Fig. 8. Visible (A) and UV (B) Raman spectra of 2V/CeO2-rod calcined at different temperatures. (C) Plots of the ratio of ID/IF2g vs. surface VOx density for both CeO2-rod and 2V/CeO2-rod, and the ratio of IV=O/IF2g on 2V/CeO2-rod from the visible Raman spectra. Adapted from Ref. [28].

Similar visible and UV Raman spectra were also collected on 2V/CeO₂-cube and 2V/CeO₂-octahedra samples ^[28]. It turned out that the initial CeVO₄ formation temperature is 150 K higher on ceria cubes than on ceria rods while there is no CeVO₄ formation on ceria octahedra up to calcination temperature of 1173 K. The barely unchanged feature of V=O bands on 2V/CeO₂-octahedra at all calcination temperatures also indicates that the ceria octahedra is not reactive to the surface VO_x species and thus no CeVO₄ is produced upon heating. Since the formation of CeVO₄ (Ce³⁺) is a reductive reaction process of surface/subsurface CeO₂ (Ce⁴⁺) with the dispersed VO_x species, the surface oxygen vacancy formation energy, an indicator for the reducibity of a surface, of the three ceria nanoshapes play an important role in the formation of CeVO4. It has been calculated that the surface oxygen vacancy formation energy follows the sequence {111} > {100} > {110} for ceria ^{[106, 107]}, so that the surface of CeO₂ rods ({110}) is the easiest to be reduced while the CeO₂ octahedra ({111}) surface is the most difficult one. Furthermore, it was shown that the presence of defect sites in ceria also promote its reducibility ^[108]. Our previous Raman study ^[95]indicated that the ceria rods have the most abundance of defect sites, followed by cubes, and the least on octahedra. Combining the two factors, we can understand that both the lowest surface oxygen vacancy formation energy and largest amount of defect sites on ceria rods promote the formation of CeVO₄ at the lowest calcination temperature on VO_x/CeO₂-rod in comparison to VO_x supported on CeO₂ cubes and octahedra.

鉴定SiO₂表面VO_x物种的结构一直是包括拉曼, IR, UV-Vis-NIR, EXAFS, NMR和EPR等很多光谱技术研究的课题^{[1, 3, 24, 25, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40]}.  通常认为VO_x在脱水SiO₂表面上以两种不同形态存在.  一种是在低担载量时(通常< 2VO_x/nm²)的单体VO₄物种, 还有一种是在高担载量时(> 2VO_x/nm²)单体与V₂O₅晶体的混合物^[38].  长久以来, 金字塔模型(O=V-(O-Si)₃)被当作是脱水VO_x/SiO₂在低担载量时(< 2VO_x/nm²)的唯一结构^{[31, 41]}, 因此VO_x/SiO₂通常被认为是具有单位点的多相催化剂的典型.  

最近, 新的一轮实验^{[39, 40, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51]}和理论^{[42, 49, 50, 52, 53, 54, 55, 56]}试图揭示在脱水条件下SiO₂表面VO_x确切的几何分子结构.  这源自金字塔模型不能很好地解释这些年所有的实验数据, 尤其是在Magg等^[42], Gijzeman及其合作者^{[43, 44, 45]}, Moisii等^{[46, 47]}研究的振动谱中关于特征谱峰归属的不确定性.  人们发现, 在VO_x物种和SiO₂载体之间存在很强的振动耦合^[42], 使得VO_x/SiO₂的振动谱变得复杂和难以解释, 如915 cm^-1处振动谱带, 曾被归属为V-O-V^{[40, 57]}, V-O-载体^{[6, 36, 42, 48, 49, 52, 53, 54, 55, 56]}, O-O^{[43, 44, 58]}或者V=O模式^[46].  同样地, 1000 cm^-1以上拉曼谱带的归属也存在争议.  在这个区间内, 氧钒伸缩振动和SiO₂的振动模式因在振动能量上相近而强烈地耦合在一起^{[42, 46]}.  基于不同的归属, 人们对脱水条件下固定在SiO₂表面的单体VO₄物种提出了三种结构:  即金字塔型、双嫁接物种(两个V-O-Si, 一个V=O和一个V-OH)以及所谓的伞结构(一个V-O-Si, 一个V=O以及一个扰动氧分子与钒键合).  Wachs及其合作者最近的研究表明, 通过与具有伞结构的K[VO(O₂)(heida)]的振动谱相比较, 水合和脱水的负载VO_x催化剂表面绝对不存在伞状结构^[59].  最新的实验和理论研究认为, 被人们所广泛接受的金字塔型是稳定的VO_x物种在高度脱水条件下最可能的结构.  然而, 甚至对于SiO₂表面唯一存在单体VO_x物种的观点, Schlogl及其合作者^{[60, 61]}在比较SBA-15负载VO_x催化剂的理论和实验NEXAFS光谱后也提出了质疑.  他们认为, 在原位条件下, 不同的VO_x物种尤其是非单体VO_x也可能存在于催化剂表面.  这个观点也在最近负载在六边形介孔SiO₂上VO_x的UV-Vis研究^[62]中得到佐证, 除了存在单体VO_x外, 不同配位的低聚VO_x物种也被认为存在于SiO₂表面.  很显然, 尽管有大量的实验和理论计算研究, 关于脱水SiO₂表面的VO_x物种的确切分子结构依然存在争议.  

我们采用多波长拉曼光谱包括共振和非共振拉曼研究了VO_x/SiO₂催化体系的分子结构^[27].  与传统单波长拉曼光谱相比, 多波长拉曼光谱的优点已经在最近关于负载VO_x催化剂的研究中突显出来^{[6, 7, 8, 21, 42, 46]}.  在表面VO_x物种的电子吸收区间内激发会导致一部分VO_x物种和/或局部的V-O键在接近激发波长处有吸收, 从而产生共振拉曼光谱.  通过测量在几个波长下激发的拉曼光谱, 我们不仅能得到关于不同VO_x物种分布的信息, 还能获得关于特定VO_x物种中局部V-O键等更多详 细的结构信息.  

我们采用原位多波长拉曼光谱(在244, 325, 442, 532和633 nm激发)和UV-Vis漫反射光谱研究了SiO₂上具有不同表面VO_x密度(0.0002-8 V/nm²)的VO_x在773 K脱水后的结构.  图2列举了脱水后样品0.25V的多波长拉曼光谱(VO_x/SiO₂表面上VO_x密度为0.25  V/nm²).  通常, 由不同可见光激发(442, 532和633 nm)得到的可见拉曼光谱基本一致, 但与由紫外光(244和325 nm)激发的紫外拉曼光谱在谱带位置和相对强度上有很大的不同.  这是因为样品0.25V在可见区几乎没有吸收(见图2(A)插图), 但它在低于400 nm的紫外区有很强的电子吸收, 这在紫外拉曼谱中可能引起共振增强效应.  在可见激发谱中1041 cm^-1处有一尖锐的谱带, 与表面VO_x物种有关, 通常被认为是V=O键的伸缩振动.  在低于1000 cm^-1出现的976, ~800, 605, 484 cm^-1谱带都来自SiO₂载体, 分别归属为Si-OH键, Si-O-Si键的伸缩振动, D₂缺陷模式和D₁缺陷模式^{[36, 48, 63]}.  紫外激发的共振拉曼光谱中没有明显的SiO₂信号, 这与表面VO_x物种在紫外区有很强的吸收相一致.  在325和244 nm激发的紫外拉曼光谱中, 1032 cm^-1处都出现了特征V=O伸缩谱带.  这个吸收带比可见拉曼光谱相应的峰低9 cm^-1, 它可能是造成可见拉曼谱中1041 cm^-1谱带在较低的拉曼位移有拖尾的原因.  出现两种不同的V=O伸缩振动带表明, 在SiO₂表面存在两种类型的VO_x物种.  在325 nm激发的光谱中, 461 cm^-1处出现了一个额外尖锐的峰, 同时在920 cm^-1处有一弱峰.  后者归属为界面V-O-Si模式, 前者因频率较低而通常被认为是V-O键的弯曲模式.  然而, 该峰的尖锐程度与弯曲模式谱的特征相矛盾, 因此很可能是伸缩振动模式, 最近有文献认为它主要是V-O-Si的伸缩振动^[46].  由244 nm激发的拉曼谱中, 在1032 cm^-1谱带高波数处有一肩峰, 位于1060 cm^-1.  它显然是由于244 nm激发产生共振增强引起的, 之前被归属为SiO₂网状TO模式和Si(-O^-)₂以及Si-O^{- [32, 36], 但是最近的理论研究表明, 它是V-O-Si伸缩模式[42, 46, 55, 56].  244 nm激发的谱中同样在920, 690和570 cm-1处出现弱的吸收峰, 分别归属为V-O-Si的伸缩振动和V-O弯曲模式.  这些在紫外拉曼谱中低于1000 cm-1处新出现的谱带强烈表明表面VO_x物种某些振动模式得到了共振增强.  检测到两个不同V=O伸缩峰证明至少有两种不同的VO_x物种存在.  由UV-vis-DRS谱得到的吸收边能是3.4 eV, 与单体VO_x的接近[31, 64, 65].  因此我们认为, 这两种VO_x物种本质上都是单体结构.}

我们也测量了具有不同VO_x密度VO_x/SiO₂的多波长拉曼光谱, 密度最高达8 V/nm².  当表面密度大于2V/nm²时, V₂O₅晶体对应的峰出现在可见拉曼谱中, 但是很少在紫外拉曼光谱中被观察到.  在非共振拉曼条件下(采用422, 532和633 nm激发光), 1032 cm^-1处的峰很弱, 并被埋在1041 cm^-1处很强的V=O伸缩振动谱带中.  这表明了SiO₂表面主要被1041 cm^-1对应的单体VO_x物种A占据了, 而另一种对应于1032 cm^-1的单体物种B只是少量存在.  假设在532 nm激发(非共振增强)下两种物种V=O的拉曼横切面相同, 并且已知V₂O₅的拉曼横切面比分散VO_x物种的大10倍^[32], 我们就可以通过拟合在532 nm激发的光谱来定量不同表面VO_x物种的分布^[27].  正如图2(B)所示, 物种B(紫外拉曼中1032 cm^-1处的V=O伸缩振动)的比例持续下降, 而物种A(可见拉曼中1041 cm^-1处的V=O伸缩振动)在低担载量时不断增加, 然后在较高担载量时降低.  这表明VO_x起初作为物种B固定在SiO₂表面, 在较高担载量时部分转变成物种A.  这与在紫外激发下在VO_x密度低到0.0002 V/nm²的样品中检测到物种B相一致^[27].  V₂O₅比例不断增加也证实在较高表面密度时两种VO_x物种向V₂O₅转变.  金字塔型被认为是在高度脱水SiO₂上分散的VO_x物种最稳定的形式^{[42, 52, 53, 54, 55, 56]}, 因此我们认为，在非共振可见拉曼谱中观测到的主要VO_x物种A具有这种带三个V-O-Si键和一个V=O键的结构(见图2(B)).  共振拉曼谱观测到的物种B的确切结构还不清楚, 但可能与表面羟基有关.  因为物种B先于物种A固定在SiO₂表面, 因此当物种B形成时SiO₂表面是富含羟基的.  这种表面类似于部分水合的表面, 在该表面水合VO_x的V=O伸缩频率通常比脱水物种的要低^{[31, 32]}.   因此, 物种B可能具有部分羟基化的金字塔结构, 羟基可能以氢键的形式与V-O-Si键中的桥氧原子相接(氢键金字塔结构), 或者如图2(B)中所示的直接与V中心键合(类似于双嫁接结构).  需要进一步的理论模型来对VO_x物种B给出较为清晰的结构信息.  值得注意的是, 类似的VO_x结构后来被Chlosta等^[51]用于负载在不同比表面SiO₂上的VO_x的研究中.  他们在1027和1040 cm^-1处也观察到两个V=O拉曼谱峰, 前者被认为是由羟基化的VO_x物种引起, 而后者是脱水物种, 与我们提出的模型相似.  

共振拉曼的特性之一是有较高的可能性观测到和频及倍频谱带, 它们能提供基频振动的几何和电子性质信息, 并帮助归属基频的振动模式.  我们在两个紫外光激发下因为共振拉曼效应而观察到大部分VO_x/SiO₂样品(担载量小于0.001 V/nm²)的和频及倍频谱带.  图3列出了0.25 V样品在325 nm波长激发下¹⁸O交换前后的全谱.  交换前的样品在462 (V-¹⁶O-Si, 表示为ν_a), 920 (V-¹⁶O-Si, 表示为ν_b),和1032 cm^-1 (V=¹⁶O, 表示为ν_c)处出现基频谱带.  2052, 3059, 4065, 5064 cm^-1处出现很明显的倍频, 分别对应于2ν_c, 3ν_c, 4ν_c, 和5ν_c.  据我们所知, 这是首次在共振拉曼光谱上出现高达第四级的钒氧伸缩倍频.  这与由Albrecht A项引起的拉曼增强相一致^{[66, 67, 68, 69]}, 表明ν_c是完全对称的模式, 在激发态钒氧键键长有较大的位移.  除了基频的振动峰外, 在1362, 1493, 和1831 cm^-1处也观察到拉曼谱带, 分别为(ν_a + ν_b), (ν_a + ν_c)的和频和ν_b的倍频.  氧同位素完全交换后, 所有的基频谱带, 和频和倍频带都移向较低的拉曼位移, 列于图3中.  在当前实验条件下, 相同的实验在SiO₂载体上并没有显示明显的同位素交换, 因此, 在样品0.25V上发现的氧同位素交换后的位移证实了在基频谱带区域外观察到的峰确实是由表面VO_x物种的和频和倍频引起的.  

如果V=O基频振动模式接近于双原子伸缩振动, 那么倍频就可以被用来计算V=O基频模式的谐波数和非谐波数常数^[8].  因为V=O与SiO₂载体强烈耦合, 这种计算有可能不能直接用于VO_x/SiO₂体系^[42].  所幸的是, 负载在SiO₂表面VO_x的振动光谱的简正模分析法最近已经通过密度泛函理论(DFT)实现^[56].  可以用Dobler等^[56]研究中所用的两个簇模型来代表我们脱水的VO_x/SiO₂样品, 即VO_x/六方柱(类似不带羟基完全脱水的SiO₂表面)以及VO_x/斜方晶(类似于带有一些羟基的部分脱水SiO₂表面).  两种模型的结果表明, 在单体VO_x物种中钒氧模式有90%主要来自V=O伸缩, 基本与SiO₂体相振动模式无关.  这表明VO_x/SiO₂体系中的V=O模式可以当做伪双原子V=O伸缩, 并且可以对V=O模式的倍频进行双原子近似 分析.  

V=O(ν_c)振动的各个倍频(图3)可被用于计算谐波数ω_e (= ω_m, 非谐性振动频率校正)和非谐性常数ω_ex_e (= x_mm)^{[8, 70]}.  对由多原子非谐性振荡产生的基频(n = 1)以及一到四级倍频(n = 2-5), 观察到的波数ν_m(n)可以用ν_m(n) = nω_m - n(n + 1)x_mm方程来表达, m是简正模式标识符.  ν_m(n)/n对n作图应该得到一条斜率为x_mm, 截距为ω_m - x_mm的直线.  图3插图显示了样品0.25V在¹⁸O交换前后ν_V=O (ν_c)的倍频变化图.  采用双原子振荡近似, 由方程ω_e = (2π)^-1(f_e/μ)^0.5获得的谐频(ω_e)来确定非谐性校正力常数(f_e), μ是约化质量.  计算得到V=O (1032 cm^-1)和V-O-Si(920 cm^-1)的f_e分别为7.76和4.57 mdyn/Å, 与通常f(V=O) > f(V-O)的趋势十分符合.  V=O的f_e值与由VO_x/Al₂O₃ (7.37-7.58 mdyn/Å)^[8]和VO_x体相化合物(7.10-7.70 mdyn/Å)^[71]得到的V=O力常数相一致.  V-O-Si的f_e值小于在VO_x/Al₂O₃ (~6.40 mdyn/Å)中V-O-Al的值^[8], 这与VO_x物种与SiO₂的作用比在Al₂O₃上要弱相一致.  在1032和920 cm^-1处两种模式的离解能分别在628-608和517-574 kJ/mol之间^[27].  

利用分析和频和倍频得到的结果, 可以计算得到 与920, 1032和1060 cm^-1处基频模式相关的对称性, 力常数, 以及离解能, 并在我们最近的其他实验和DFT计算帮助下^[27], 可以把三个谱带分别归属为同相V-O-Si, V=O伸缩和异相V-O-Si模式.  

我们采用氢气还原来测试两种单体VO_x物种的氧化还原性质, 并用原位共振和非共振拉曼光谱来监测还原过程.  图4(A)和4(B)分别列出了样品0.25V在不同温度还原后由244(检测物种B, 部分羟基化的单钒酸盐)和422 nm (主要检测物种A, 金字塔型单钒酸盐)激发的室温下的拉曼光谱.  在244 nm激发的共振拉曼谱中, 1032与1060 cm^-1峰强度的比例随还原温度的增加而减少.  该比值的变化与更多VO_x负载在SiO₂表面时相反^[27], 这表明因为氢气还原引起物种B的覆盖度减少.  通过与由244 nm激发脱水VO_x/SiO₂样品的1032/1060 cm^-1峰强度比相比较, 我们可以估计, 大概90%的物种B在973 K以上时就已经还原了.  在1073 K处理后还没有还原的VO_x物种仍然具有可检测的拉曼谱峰, 这不仅是因为共振增强效应, 还由于自吸收的减少^[72], 因为还原后负载VO_x在UV区有较少的吸收^{[73, 74]}.  然而在422 nm激发的光谱中, 在1041 cm^-1处峰的强度随还原温度升高而不断降低.  该峰在还原温度达到973 K时由于物种A(和B)的还原而消失.  但是V=O峰的消失并不必然表示VO_x物种完全被还原, 因为拉曼强度的减弱也可能部分是由还原样品在可见拉曼区增强的自吸收效应^[72]引起的^[74].  虽然如此, 大部分物种A和B在高温下都被还原了.  重新用氧气氧化可以把已经还原的钒氧化物恢复到氧化状态, 这是因为刚焙烧样品和重新氧化样品具有相似的拉曼光谱.  由此可见, VO_x物种A和B在氧化还原性上没有很大差别,&# 8197;可能都会参与到由VO_x/SiO₂催化的氧化还原反应中.  进一步在氧化还原催化反应条件下进行原位拉曼光谱研究, 如烷烃的氧化脱氢, 可以用来洞察两种不同VO_x物种的催化功能.  

尽管CeO₂常被用作氧化反应的催化剂和催化剂载体, 但相比其他氧化物载体, 它较少作为VO_x的载体.  最近的研究表明, CeO₂负载的VO_x对乙烷, 丙烷和甲醇的氧化脱氢有很好的反应活性和选择性^{[75, 76, 77, 78, 79, 80, 81, 82, 83]}.  因此, 一些研究者对VO_x表面结构以及它与CeO₂相互作用进行了研究^{[3, 75, 76, 77, 78, 84, 85, 86, 87, 88, 89, 90]}.  虽然得到了一些关于表面VO_x与CeO₂载体反应形成CeVO₄的信息^{[75, 76, 77, 78, 84, 85, 88]}, 但很少有关于表面VO_x物种结构的信息^{[76, 77, 87, 89, 90]}.  这可能是由于在表征表面金属氧化物最强有力的工具拉曼光谱中^{[2, 3]}, 表面VO_x物种的信号与VO_x-CeO₂体系中CeO₂和CeVO₄的强拉曼散射比相当弱.  最近, 通过结合高分辨扫描隧道显微镜(STM)和超高真空系统的红外反射吸收光谱(IRAS), 沉积在CeO₂(111)薄膜上的VO_x物种的原子结构得到了确认^{[81, 91]}.  ST M检测的VO_x物种的核性(单一的或多聚的)与IRAS检测的氧钒频率之间建立了直接的联系.  不过, 在多晶CeO₂上表面分散VO_x的确切分子结构还是没有完全确认.  

在CeO₂上负载VO_x的方法可以除去CeO₂上大部分不稳定的表面氧, 从而改进CeO₂的燃烧性能^{[76, 77]}.  VO_x和CeO₂之间强的相互作用导致形成CeVO₄, 增加VO_x的量, 热处理以及还原环境都能促进这一过程^{[76, 77, 78]}.  CeVO₄的存在可以解释为什么VO_x中钒以最高价态(+5)存在, 而一些表面铈离子被还原为Ce³⁺.  最近的DFT计算同样表明, CeO₂表面把VO_x物种稳定在+5价氧化态, 而VO_x物种覆盖并还原了CeO₂的表面^{[79, 91, 92, 93]}.  然而表面VO_x和CeO₂相互作用的本质依然耐人寻味, 值得进一步的研究.  

在这个研究中, 我们利用多波长拉曼光谱不仅研究了在CeO₂纳米粒子上VO_x的详细结构, 而且也研究了两者如何相互作用^[30].  此外, 我们也通过多波长拉曼方法研究了CeO₂载体的形状对VO_x物种结构以及CeVO₄形成的影响^{[28, 29, 94]}.  可以预见, 由该研究得到的结构信息将会为理解VO_x-CeO₂催化体系提供基本依据.  

图5(A)为脱水样品1V(在Aldrich氧化铈纳米粒子上VO_x密度为1V/nm²的VO_x/CeO₂样品)的多波长拉曼光谱来展示激光激发效应.  由可见光激发(442, 532和633 nm)的全谱中主要谱峰是464 cm^-1处CeO₂的F_2g模式^{[95, 96, 97, 98]}.  表面VO_x物种在1015 cm^-1处有一个相对较弱的峰,& #8197;归属为V=O伸缩振动^{[2, 3, 77, 86]}.  在325 nm激发的拉曼光谱中也主要显示为CeO₂的特征峰, 不过在590 cm^-1处与缺陷相关的谱峰和在1179 cm^-1处的第二纵向光学模式(2LO)因为共振拉曼效应大大地增强^{[95, 96]}.  244 nm激发的光谱显示了相对较低的信躁比, 仅能观察到F_2g模式, 这可能是因为CeO₂在深紫外区很强的自吸收引起的^{[72, 96]}.  虽然如此, 所有激光激发的样品1V全谱主要显示CeO₂载体的特征, 因为晶体CeO₂是很好的拉曼散射物质, 从而压制了来自表面VO_x物种的信号.  为了得到更多关于表面VO_x物种的详细信息, 可以通过选择光谱的测量范围来避开CeO₂强的特征峰, 并延长谱图的采样时间.  正如图5(A)插图所示, 在可见拉曼谱中可以明显地观测到表面VO_x物种的特征谱峰, 但是在紫外拉曼谱中仍然给出较弱的表面VO_x物种的信号.  伴随着1033 cm^-1肩峰的1015 cm^-1处的尖锐谱带是由表面分散VO_x的钒氧伸缩引起的, 而在857和708 cm^-1处的宽峰归属为V-O-Ce桥式模式^{[77, 86, 92]}.  当激发光向更长波长移动时, 表面VO_x的信号变得越来越突出, 这是因为来自CeO₂载体的自吸收在更长波长逐渐减弱^{[96, 97]}.  因此, 在VO_x/CeO₂的例子中, 可见光激发, 即非共振拉曼, 更适合表面VO_x结构的进一步研究.  此外, 由325 nm紫外激发的共振拉曼可以用来研究VO_x/CeO₂体系缺陷位的信息, 因为共振拉曼对于监测CeO₂中的缺陷位(在590 cm^-1)非常灵敏^{[95, 96, 97, 98, 99]}.  

图5(B)为VO_x表面密度从0.05到10 V/nm²脱水VO_x/CeO₂样品在633 nm激发的拉曼光谱.  在V=O伸缩区(950-1100 cm^-1), 很容易观测到样品0.05V在1008和1015 cm^-1处有两个弱峰.  当更多的VO_x负载到CeO₂表面时, 这两个峰的强度增加并向更高波数移动 .  样品5V和10V中, 在1022, 1030和1044 cm^-1处出现三个不同的V=O振动模式.  另外, 在样品10V中形成了V₂O₅, 它的特征V=O伸缩出现在944 cm^-1.  观察到多个V=O模式表明在CeO₂表面存在不同VO_x物种.  在低于950 cm^-1的谱区内, 随着VO_x的增加, V-O-Ce由830移到857 cm^-1.  当VO_x密度高于5V时, 该区域的谱峰变尖锐并出现在862 cm^-1处.  当VO_x量超过1V/nm², 在930 cm^-1 处出现一单独的宽特征峰, 归属为多钒酸盐中的V-O-V振动模式^{[77, 86, 87, 90]}.  对于样品10V, 在桥式V-O-Ce的宽背景中, 787, 800和862 cm^-1处出现由CeO₄引起的非常尖锐的峰.  可见在CeO₂表面上, 分散VO_x物种的单层密度应该在5到10 V/nm²之间.  

表面VO_x物种的原子结构鉴定可以借鉴最近在CeO₂(111)薄膜上VO_x的模型研究^[91].  结合STM和表面IR谱, Baron等^[91]确立了VO_x/CeO₂(111)体系中VO_x物种的结构和光谱间的直接联系: 在低波数时是具有V=O模式的单VO_x, 在较高波数时是具有V=O伸缩的多聚VO_x.  考虑到我们观察到的V=O伸缩振动峰落在VO_x/CeO₂(111)体系的IR谱中相同的区间(1006-1045 cm^-1)^[91], 可以类似地将1008 cm^-1归属为单体VO_x, ~1015 cm^-1为二聚VO_x, 1022-1030 cm^-1为三聚VO_x, 1044 cm^-1为多聚VO_x.  在较高聚合度的情况下, V=O频率的增加是由于VO_x物种上V=O之间的偶极耦合增强导致的.  很明显, CeO₂上的表面VO_x由单体VO_x到二聚, 三聚, 多聚, 最后到晶体V₂O₅和CeVO₄逐渐演变.  这是首次获得在粉末CeO₂载体上VO_x物种详细的结构演变过程.  

有趣的是, 几种类型的VO_x物种能在CeO₂表面同时存在, 这表明在CeO₂上存在不同的表面位点, 能固定并稳定VO_x物种.  这很少在其他氧化物担载的VO_x上观察到, 通常只能检测到一种类型的二维表面VO_x物种^{[2, 3, 5]}.  一个比较均匀的CeO₂表面, 例如具有可控表面结构的纳米形状的CeO₂可能有助于更好的控制表面VO_x物种. 197; 最近纳米材料合成的发展使获得这种材料成为可能^{[100, 101]}.  利用具有纳米形状的CeO₂, 棒状(主要是{110}), 立方体({100})和八面体形({111}), 我们采用原位多波长拉曼光谱研究了CeO₂的表面结构如何影响表面VO_x物种的结构^{[28, 29]}.  发现在纳米形状和多晶的CeO₂上形成相似的表面VO_x物种, 即单体、二聚、三聚和多聚VO_x以及晶体V₂O₅和CeVO₄随表面VO_x密度变化都能被观测到.  然而, 仅在纳米形状的CeO₂上才能观察到只存在单体VO_x物种的情况, 在多晶CeO₂上总是单体和二聚共存, 甚至当表面VO_x物种的密度相当低时也是两种物种共存, 这表明纳米形状CeO₂具有相对均一的表面结构.  

CeO₂作为催化剂和催化剂载体的两个有趣的特征是缺陷位和不稳定的表面氧^[102], 因此我们研究了当VO_x被固定在CeO₂表面时会如何改变这些性质.  紫外共振拉曼用于研究VO_x物种如何与CeO₂的缺陷位相作用, 因为这种技术对CeO₂缺陷位的改变非常灵敏^{[95, 96, 97, 98]}.  图6给出了脱水VO_x/CeO₂样品在325 nm激发下的拉曼光谱.  由于CeO₂较强的自吸收, 表面VO_x物种的拉曼特征只有在它的表面覆盖度达到一定程度时(≥ 1V/nm²)才能观察到.  此外, 桥式V-O-Ce比在1030 cm^-1处由三聚VO_x引起的V=O伸缩更显著.  这些与由633 nm激发的谱(图5(B))上有很大的差异, 这再一次表明, 可见或近红外更适合研究CeO₂上VO_x的结构.  在325 nm激发的谱中于1030 cm^-1处只有一种V=O模式可以被观测到, 这可能是由共振拉曼效应引起的, 表明VO_x三聚体可能在325 nm处有电子吸收谱带.  除了表面VO_x的特征谱峰外, 2LO(1179 cm^-1)和缺陷相关谱带(D带, 590 cm^-1)的强度相对于在464 cm^-1处氧化铈F_2g模式都有极大地共振增强.  当VO_x担载在CeO₂表面时,所有这些CeO₂的特征谱带都减弱了.  相对于F_2g带, D带的强度减弱尤为明显.  通过对D带和F_2g带强度比例作图(I_D/I_F2g), 可以用来说明CeO₂缺陷位的相对量.  正如图6插图所示, I_D/I_F2g的比例, 即缺陷位的量, 随着VO_x负载量的增加而不断降低, 尤其是在较低的负载量时, 这意味着表面VO_x与CeO₂缺陷位紧密地相互作用, 导致缺陷位被覆灭或者其拉曼横截面被改变了.  该现象同样也在纳米形状CeO₂负载的VO_x体系中被很清楚地观察到^[28].  

为了理解CeO₂表面活性氧如何与VO_x物种作用, 我们采用可见拉曼光谱(532 nm激发)与氧气吸附相结合研究在VO_x/CeO₂样品上的缺陷位在氢气还原后会如何改变.  还原能除去大部分活性晶格氧并产生能吸附氧气的氧缺陷, 这可以由低温拉曼光谱检测到^{[95, 103]}.  图7为在100 K氧气吸附在773 K还原的CeO₂样品, 0.1V, 0.5V, 1V和2V上的拉曼光谱.  CeO₂样品在1137, 890和834 cm^-1处出现了三个吸附氧物种引起的拉曼峰, 分别归属为超氧(O^2-), 在团簇空位上的过氧(O₂^2-)和在孤立空位上的过氧化物^{[2, 95, 103, 104]}.  当VO_x负载在CeO₂表面时, 这三个峰的强度都逐渐减弱.  尤其是1137/1130和890 cm^-1处的峰比834和843 cm^-1处峰衰减得更快, 这表明表面VO_x更倾向与某些活泼的晶格氧位作用, 否则这些晶格氧位在还原后会产生单电子空位和团簇的双电子空位.  CeO₂的活性表面氧似乎逐渐通过固定VO_x物种而消耗, 因为在还原的2V样品上几乎检测不到吸附氧物种的拉曼谱峰.  换句话说, 表面VO_x物种通过桥键钝化了CeO₂表面氧位的活性.  这种钝化进一步被VO_x/CeO₂样品的H₂-TPR结 果证实^[30], 结果显示, 固定的VO_x物种抑制了CeO₂表面氧的还原性.  

因此, CeO₂的两个有趣的性质, 缺陷位和不稳定表面氧都对VO_x物种在CeO₂上的固定起了重要作用.  这能很好地解释VO_x/CeO₂在氧化脱氢反应中总体氧化能力比CeO₂低而选择性增加的原因^{[75, 76, 77, 78, 80]}.  

正如上述所示, 当负载量较高, VO_x能与CeO₂反应生成CeVO₄.  在CeVO₄中V-O-Ce键被认为是烷烃氧化脱氢反应中的活性位^{[76, 77, 78, 105]}, 因此, CeVO₄的形成机理得到广泛的研究^{[75, 76, 77, 78, 84, 85, 88, 105]}.  研究发现, 不仅高VO_x负载量, 而且高温以及还原环境等其他因素都会促进CeVO₄的形成^{[77, 84]}.  然而, 目前还没有关于CeO₂载体表面结构对CeVO₄形成机理的研究.  具有不同表面结构的纳米形状CeO₂给我们在这方面的研究提供了机会^[28].  

在具有相似表面VO_x密度(2V/nm², VO_x具有相似的二聚和三聚结构)的条件下, 我们比较了由表面VO_x物种与纳米形状CeO₂(棒状, 立方体形和八面体形)经热反应形成的CeVO₄的过程.  以棒状2V/CeO₂-rod为例, 图8给出了在氧化环境下, 可见拉曼(监测VO_x的变化和CeVO₄的形成)和紫外拉曼光谱(监测缺陷位的改变)与不同焙烧温度的关系图.  正如图8(A)中所示, 二聚和三聚VO_x物种拉曼峰的强度在923 K焙烧后开始降低, 同时在代表CeVO₄的864 cm^-1处形成一尖峰.  当升高焙烧温度, 三聚VO_x的V=O模式比二聚VO_x下降更快, 表明多聚VO_x物种更容易与CeO₂反应生成CeVO₄.  当焙烧温度在1073 K及以上, 在864 cm^-1处的峰逐渐减少, 并在921和953 cm^-1处出现两个新的峰, 证实CeVO₄开始逐渐转变成另一种化合物, 可能是CeVO₃.  相同的焙烧过程, 紫外拉曼光谱(图8(B))给出了完全不同的信息.  由于CeO₂强的吸收, 出现CeVO₄拉曼特征峰所需样品焙烧温度比在可见拉曼中高100 K.  CeO₂缺陷对应的拉曼谱带(D带)的强度随焙烧温度不断降低, 这可以归于热湮灭和CeO₂缺陷位的化学消耗(通过与VO_x反应).  图8(C)显示了不同温度下, 2V/CeO₂-rod样品中D带与F_2g(I_D/I_F2g)信号强度比的变化趋势更接近于I_V=O/I_F2g (由图7(A)中的可见拉曼得到), 而不是纯CeO₂棒中的I_D/I_F2g.  这表明当温度升高时, CeVO₄的形成在消耗CeO₂缺陷位上比单纯的热湮灭起到更大的作用.  

我们也对立方形2V/CeO₂-cube和八面体形2V/CeO₂-octahedra样品做了类似的多波长拉曼光谱研究^[28].  研究表明, 在CeO₂立方体上开始形成CeVO₄的温度比在CeO₂棒上要高150 K, 而在CeO₂八面体上直到温度高达1173 K也没有CeVO₄形成.  2V/CeO₂-octahedra上V=O的拉曼谱峰在所有焙烧温度都几乎保持不变, 这也表明了CeO₂八面体不与表面VO_x物种反应,  因此加热条件下没有CeVO₄生成.  CeVO₄(Ce³⁺)是由表面/次表面CeO₂(Ce⁴⁺)与分散的VO_x物种发生还原反应形成的, 因此三种纳米形状CeO₂表面氧空位的生成能, 即表面还原性标示, 在CeVO₄的形成中起重要作用.  计算得到CeO₂表面氧空位生成能的顺序为{111} > {100} > {110}^{[106, 107]}, 因此CeO₂棒({110})最容易被还原, 而CeO₂({111})八面体表面最难还原.  此外, CeO₂缺陷位的存在也促进了它的还原^[108].  我们之前的拉曼研究表明, CeO₂棒具有最丰富的缺陷位, 其次是立方体, 最后是八面体.  结合两种因素我们认为, 与负载在CeO₂立方体和八面体上的VO_x样品相比, 在VO_x/CeO₂-rod样品中CeO₂纳米棒上最低表面氧空位生成能和最大量缺陷位都是促进CeVO₄在最低焙烧温度下生成的原因.