Distinguishing faceted oxide nanocrystals with 17O solid-state NMR spectroscopy

Facet engineering of oxide nanocrystals represents a powerful method for generating diverse properties for practical and innovative applications. Therefore, it is crucial to determine the nature of the exposed facets of oxides in order to develop the facet/morphology–property relationships and rationally design nanostructures with desired properties. Despite the extensive applications of electron microscopy for visualizing the facet structure of nanocrystals, the volumes sampled by such techniques are very small and may not be representative of the whole sample. Here, we develop a convenient 17O nuclear magnetic resonance (NMR) strategy to distinguish oxide nanocrystals exposing different facets. In combination with density functional theory calculations, we show that the oxygen ions on the exposed (001) and (101) facets of anatase titania nanocrystals have distinct 17O NMR shifts, which are sensitive to surface reconstruction and the nature of the steps on the surface. The results presented here open up methods for characterizing faceted nanocrystalline oxides and related materials.

F aceted oxide nanocrystals have attracted much research attention in a variety of fields, including catalysis 1-4 , photocatalysis [5][6][7][8] , solar hydrogen generation 9 , photoelectrochemical application 10 , gas sensoring 11 , and energy storage 12 , owing to their specific surface structures. Identification of the exposed facets is thus fundamental to the preparation and applications of oxide nanomaterials. Current characterization tools for studying the surface structure of nanocrystals are mostly based on electron microscopy [13][14][15][16][17][18] . At a resolution that the exposed facet can be determined, however, the field of view of microscopy techniques is often so small, or the particles may show considerable aggregation that it is possible that the region investigated is not representative of the whole sample 19 . Therefore, the development of complementary characterization methods that can give detailed structural information concerning the nature of the exposed facets of nanocrystals is urgently required.
Solid-state NMR spectroscopy is a powerful technique that has been widely used in studying the local environments of solids 20 . 17 O NMR spectra, e.g., can give detailed structural and dynamic information of important functional oxygen-containing materials [21][22][23][24][25][26] , benefiting from the large 17 O chemical shift range (>1000 ppm). However, few publications are available on the 17 O NMR studies of nanosized oxides, in spite of their widespread applications, largely owing to the high cost of 17 O and structure change during isotopic labeling. Recently, Wang et al. 27 developed a surface-selective labeling method for oxide nanomaterials at low temperatures and revealed that the 17 O species on the first few layers of ceria nanomaterials are associated with different 17 O chemical shifts. However, direct experimental evidence is still missing concerning the relationship between the 17 O chemical shifts and the nature of the exposed facets.
Here, we demonstrate a new approach based on NMR and surface-selective 17 O labeling to determine the structures of the exposed facets on the technologically important anatase titania nanocrystals 6,9,10,[28][29][30] . With the help of density functional theory (DFT) calculations, oxygen species on different facets can be distinguished by their NMR shifts. The nature of surface steps and reconstructions of these surfaces, particularly on reaction with water, are also revealed.

Results
Morphology of anatase TiO 2 nanosheets and nano-octahedra. Two types of anatase TiO 2 nanocrystals with different tailored facets were examined, i.e., anatase TiO 2 nanosheets with dominant exposed (001) facets (NS001-TiO 2 ), and nano-octahedra preferentially exposing (101) facets (NO101-TiO 2 ) 31 . Their crystal forms were confirmed with X-ray diffraction (XRD) (Supplementary Fig. 1). High-resolution transmission electron microscopy (HRTEM) results show that NS001-TiO 2 ( Supplementary  Fig. 2) are nanosheets with a thickness of 6-7 nm, while NS101-TiO 2 ( Supplementary Fig. 3) are nano-octahedra with an average size of 14 nm. According to the statistical analysis of the data ( Supplementary Figs. 2, 3), an average of 77% of the exposed surfaces of NS001-TiO 2 are (001) facets, while 96% of the exposed surfaces of NO101-TiO 2 are (101) (see Supplementary Table 1, Supplementary Fig. 4 and additional discussion in Supplementary Note 1). X-ray photoelectron spectroscopy (XPS) spectra ( Supplementary Fig. 5) suggest that there is no evidence for the existence of F − or Cl − on the surface of either sample, while the concentrations of carbon (C) or nitrogen (N) impurities in both samples are also very small according to the elemental analysis (Supplementary Table 2 32 . Therefore, 17 O NMR spectrum of the anatase TiO 2 sample enriched with 17 O 2 at high temperature show a single sharp peak at 558 ppm ( Fig. 1), corresponding to O 3c species in the "bulk" part, consistent with previous reports 33 . It is clear that the 17 O NMR spectra of surface-labeled NS001-TiO 2 and NO101-TiO 2 differ significantly and are also distinct from the spectrum of the nonselectively labeled anatase TiO 2 (Fig. 1 Supplementary Fig. 11) can be found in Supplementary Fig. 12 and Supplementary Note 4 on the surface of titania nanostructure 27 . The broad signals at much lower frequencies (−150 to 300 ppm) can be attributed to hydroxyl groups in surface hydroxyls and/or water environments 27,34,35 . The peak centered at 150 ppm in the spectrum of NS001-TiO 2 can also be observed in 1 H→ 17 O cross-polarization (CP) MAS NMR spectra (Supplementary Fig. 13 and Supplementary Note 5), confirming that this signal arises from oxygen ions in close proximity to proton. Such signal is very weak in the spectrum of NO101-TiO 2 while an additional peak can be found centered at −75 ppm ( Fig. 1  Surface reconstruction of anatase TiO 2 (001). In order to help the spectral assignment, DFT calculations were performed on anatase titania structures with different exposed facets. Since water molecules prefer to dissociate on the high-energy (001) facets 36 , and surface reconstructions are likely to occur on (001) 39 , four possible surface models were constructed for NS001-TiO 2 , including the un-reconstructed clean TiO 2 (001) (CL), hydrated TiO 2 (001) at a water coverage of 1 / 2 ML (dissociative adsorption, DA), 1 × 4-reconstructed clean TiO 2 (001) (RC-CL), and hydrated 1 × 4-reconstructed TiO 2 (001) (RC-DA) (see Fig. 2 and Supplementary Figs. 15-18 for details). 1 / 2 ML means that every two surface Ti 5c take one water molecule, and it also corresponds to a fully hydrated surface state 36 . The calculated isotropic chemical shifts of each oxygen sites (δ iso ), quadrupolar coupling constant (C Q ), asymmetry parameter (η), and center of gravity of the NMR signals (δ CG ) are given in Supplementary Tables 5-8. In all the models investigated, the calculated chemical shifts (δ CG ) of oxygen ions in the "bulk" part (middle layers) of the anatase structures are close to 558 ppm, which is the observed chemical shift of O 3c in the nonselectively labeled anatase TiO 2 . The chemical shifts of the oxygen species in the first few layers, however, deviate noticeably from the "bulk" values and depend on the specific local structure.
The calculated results were used to simulate the 17 O NMR spectra at different external magnetic fields ( Fig. 2 and Supplementary Fig. 19) by considering the surface oxygen species only, whose isotropic chemical shifts have been marked in the structural models in Fig. 2. The simulated signals arising from the OH species generated in the DA and RC-DA structures give a fair match with the experimental data (450-0 ppm), further supporting that water dissociates on the (001) facets. Furthermore, they also allow us to assign a weak peak centered at approximately 400 ppm that overlaps with the sidebands from the surface oxygen sites to another OH environment. The calculation results also show that the majority species that give rise to the signals at 600-760 ppm in the experimental data are actually the O 2c environments, rather than the O 3c site, and that these species can only be ascribed to reconstructed surfaces (i.e., contributions from the RC-CL and/or RC-DA structures). Therefore, these results provide compelling evidence that structure reconstruction does indeed occurs on the (001) surface. On the basis of the 1 H NMR results (Supplementary Fig. 20 and Supplementary Table 9), the water coverage on this sample is 0.3 ML, indicating that a reconstructed surface is energetically favored at this state 36 , and both RC-CL and RC-DA surface conditions should exist, due to the insufficient water coverage. Therefore, it can be concluded that, at this specific water coverage (0.3 ML), surface reconstruction occurs on (001) surface of anatase titania, and water dissociates on this surface.
Step edges of anatase TiO 2 (101). For NO101-TiO 2 , three defect-free structure models, including clean anatase TiO 2 (101) (CL), hydrated anatase TiO 2 (101) under a water (molecular adsorption) coverage of 1 / 2 ML (MA), and hydrated anatase TiO 2 (101) with dissociatively adsorbed water under the coverage of 1 / 2 ML (DA, which is energetically less favorable 37, 38 ), were constructed first to calculate the NMR parameters (Supplementary Figs. 21-23 and Supplementary Tables 10-12). However, the simulated spectra do not match the experimental data (for surface O 2c sites in particular) (Supplementary Fig. 24). Surface defects, however, often occur on the (101) facets according to scanning tunneling microscopy investigations 40,41 as well as first-principles calculations 41 . Particularly, "step edges", associated with higher reactivity 41 , are considered as the most common defects on this surface. Gong et al. have proposed several types of step-edge defects 42 with monoatomic height along trapezoidal or triangular islands on (101) surface 40,43 . The so-called type-D steps occur along two nonparallel sides of the trapezoidal islands (or two sides of the triangular ones), and they are also the most prevalent ones among all the steps. Accordingly, in the current work, an anatase TiO 2 (134) vicinal surface with such type-D steps In both adsorption modes, water has higher adsorption energies than that found at flat (101) surface (Supplementary Table 15). Since the adsorbed water molecules in two orientations have similar adsorption energies, each orientation is weighted the same and only 14 different surface/subsurface oxygen species are considered in the spectral simulation. The calculated structures, NMR parameters, and simulated spectra, along with the experimental data, are shown in Fig. 3, Supplementary Fig. 27, and Supplementary Table 16. For clarity, the simulated spectrum of the 14 oxygen species in OA is also presented as colored and shaded peaks in Fig. 3b.
The simulated spectra agree remarkably well with the experimental data at different external magnetic fields (Fig. 3b and Supplementary Fig. 27b), except for the center of gravity of the NMR signal for the adsorbed water species (Fig. 3b, peak 1).
The experimental line width of this peak is smaller than the calculated one, which can be attributed to the motion of the adsorbed water molecules (see Supplementary Fig. 28 and Supplementary Note 6). Other signals from surface sites probably originate from the dissociation of H 2 17 O at oxygen vacancies generated in the vacuum-drying pretreatment at 100°C (see Supplementary Fig. 29 and Supplementary Note 7) and possible subsequent migration of oxygen ions within the structure of TiO 2 , since water molecules are not expected to dissociate on type-D step edges 37,44 . The major resonance at 730 ppm (peak 3) arises from O 2c species at the step edges (Fig. 3). In comparison, peak 2, corresponding to O 2c species at the middle of (101) plane, has much smaller intensity. Considering the fact that there is only a small fraction of oxygen ions at step edges (4 ± 1.5%) 42 , the much stronger intensity of peak 3 implies that O 2c at the step edge has higher activity in the initial labeling process than the species on (101) plane. The other relatively strong peak owing to O 2c ions occurs at 640 ppm (peak 5). Such oxygen species is at flat terraces below the adjacent step edge and is attached with the adsorbed water through hydrogen bond. The signals at 480-560 ppm can be assigned to surface and subsurface O 3c species. The much stronger intensity of the O 2c species compared to the O 3c ones confirms that the 17 O-enrichment method adopted in this work does achieve an effective surfaceselective labeling.

Discussion
17 O solid-state NMR spectroscopy, in combination with DFT calculations, can be used to distinguish two anatase TiO 2 nanocrystals with different exposed facets and explore the details of their unique surface local environments. The 17 O NMR spectra provide definitive evidence that surface reconstruction occurs when (001) faceted anatase TiO 2 nanosheets adsorb a small amount of water, while "step edges" are the main defects present on the anatase TiO 2 (101) surface. The results indicate that 17 O solid-state NMR spectroscopy is a sensitive method to probe the local environments of the exposed facets of oxide nanocrystals, the structures of these facets playing a vital role in determining their properties. Further studies based on this approach can be readily envisaged to study possible changes that may occur on the faceted oxide nanocrystals in catalytic processes and other related applications.

Methods
Sample preparation. The anatase TiO 2 nanosheets, mainly dominated by exposed (001) facets, i.e. NS001-TiO 2 , were prepared according to Han's work 45 . (101) facets dominated anatase nano-octahedra (NO101-TiO 2 ), and non-faceted anatase TiO 2 nanoparticles (NF2-TiO 2 ) were prepared hydrothermally according to Liu's work 31 . The obtained materials were washed thoroughly with NaOH aqueous solution and water to remove F − or Cl − on the surface, which were introduced in the preparation. Experiment details are given in the Supplementary Methods. Another non-faceted anatase TiO 2 sample with smaller surface area, NF1-TiO 2 , was purchased from Sigma-Aldrich Corporation, and used as received.
Characterization. The powder XRD analysis was carried out on a Philips X'Pro X-ray diffractometer using Cu Kα irradiation (λ = 1.54184 Å) operated at 40 kV and 40 mA at 25°C. High-resolution TEM images were obtained on an FEI Titan 80/300 S/TEM with an acceleration voltage of 200 kV. Electron paramagnetic resonance (EPR) spectra were recorded on the samples with the same mass (50 mg) by a Bruker EMX-10/12 spectrometer at room temperature. The Brunauer-Emmett-Teller specific surface areas of the samples were measured by nitrogen adsorption at 77 K using a Micromeritics tristar ASAP 2020 instrument. The contents of C and N impurities of the samples were analyzed using a Heraeus CHN-0-Rapid analyzer. XPS spectra of both faceted samples were obtained on an Ulvac-PHI PHI 5000 VersaProbe instrument. 17 O enrichment. Faceted NS001-TiO 2 , NO101-TiO 2 , and non-faceted NF2-TiO 2 nanocrystalline samples were surface-selectively 17 O-labeled through a vacuum line   47 . The 17 O chemical shifts were calculated by using the linear response method. We used the project-augmented wave method 48 to describe the core-valence electron interactions in structure optimization, chemical shift, and electric field gradients (EFGs) calculations at a kinetic energy cutoff of 500 eV with Ti (3s, 3p, 3d, 4s), O (2s, 2p), and H (1s) electrons being treated as valence electrons. All of the atoms were allowed to relax during structure optimization with a force stopping-criterion of 0.02 eV/Å on each relaxed ion. During electronic minimization, we used an extremely high stopping criterion of 10 −8 eV for all the calculations 27 . With a 3 × 3 × 3 k-point mesh, we obtained optimized lattice parameters of a = 3.80 Å and c = 9.51 Å for bulk anatase TiO 2 , which is very close to the experimental values (a = 3.78 Å and c = 9.50 Å) 49 . It should be noted that the on-site Coulomb interaction of localized d electrons was also considered by using the DFT+U approach with an optimum Hubbard U value of 4.0 eV 50 , and lattice parameters of a = 3.86 Å and c = 9.53 Å was obtained. This indicates pristine DFT method can give reliable structural information. Since correct structural information is crucial to chemical shift calculations, we then used the pristine DFT method to do all the calculations.
The isotropic chemical shift (δ iso ) can be computed as δ iso = δ cal + δ ref 27 15-18, 21-23, 25, 26), which is close to the reported value of 52 ppm for CeO 2 27 . To calculate the quadrupole coupling constant (C Q ) and asymmetry parameter (η), we used the following equations: where h is the Planck constant, e is the absolute value of the electron charge, and V ii (ii = XX, YY, or ZZ) are the eigenvalues of the EFG tensor with |V ZZ | > |V YY | > | V XX |. We used the experimental quadrupole moment (Q) of −0.02558 barns 51 for 17 O. The adsorption energy of H 2 O (E ads ) was calculated as follows: where E H2O , E sub , and E H2O=sub are the DFT total energies of the gas phase H 2 O, the TiO 2 substrate, and the adsorption complex, respectively. For the simulated spectra from the models of (001) facet ( Fig. 2 and Supplementary Fig. 19b), only surface oxygen sites were considered, whose isotropic chemical shifts have been marked in the structural models in Fig. 2. The O 2c and O 3c sites have been given the same weight of peak area in the calculated spectra. Twice of the weight has been given to the hydroxyl groups centered around 420 ppm, and four times of the weight has been given to the hydroxyl groups centered around 150 ppm, for the sake of presentation. For simulating the NMR spectra of the defect-free (101) facet (as shown in Supplementary Fig. 24), a similar approach was used. Surface sites, i.e., sites 1-3 in Supplementary Fig. 21 and sites 1-7 in Supplementary Figs. 22 and 23, respectively, were considered. Twice the weight of the peak areas have been given to the signals of hydroxyl groups and adsorbed water, in comparison to those of the surface O 2c and O 3c sites. For simulating the spectra of (101) facet with type-D steps (Fig. 3b and Supplementary  Fig. 27b), NMR parameters of surface and subsurface oxygen sites (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14) in Supplementary Tables 13 and 14 were adopted, only with their percentage adjustable to achieve the best fitting. Furthermore, in Supplementary Fig. 28, C Q s of the adsorbed water in both adsorption orientations were also allowed to change in the simulation, in order to examine the influence of the motion of the adsorbed water on its NMR signal.