Hydrogen is considered to be a clean and renewable energy source, which might address global concerns regarding future fossil fuel shortages. Photocatalytic water splitting into hydrogen and oxygen on the surface of a semiconductor is widely accepted to be an ideal means for hydrogen production owing to the inexhaustible nature of solar radiation and the abundance of water on earth [1-6]. Among various photocatalysts, titanium dioxide (TiO₂) has continued to attract considerable attention because of its superior chemical stability, low-toxicity, and cost effectiveness [7, 8]. Nevertheless, drawbacks such as its wide band gap (3.2 eV for anatase) and fast electron-hole recombination rate substantially limit the use of TiO₂ practical applications. Considerable efforts have been devoted to modifying TiO₂ to improve its light absorption and photocatalytic activity. The doping of heteroatoms (such as N [9], C [10], Fe [11], S [12]) into TiO₂ has successfully been used to enhance the light absorption of TiO₂; however, challenges remain as these heteroatoms may act as extra electron-hole recombination centers [13].

After the pioneering work of Chen et al. [14], self-doping TiO₂ (i.e., Ti³⁺ species) might be a promising alternative strategy to modifying this wide band gap semiconductor. Further studies have shown that Ti³⁺ species and their charge balancing counterpart-oxygen vacancies are responsible for band-gap narrowing and the separation of photogenerated electrons and holes [15-17]. Various preparation methods and synthetic techniques have been used to increase the oxygen vacancy concentration in TiO₂ including hydrogen plasmas [15], aluminum reduction [16], chemical oxidation [17], and electrochemical reduction [18]. The approach involving Ti³⁺ in self-doped TiO_2-x is distinct from these methods owing to the simpler and safer procedures (no requirement for the use of hydrogen) and better visible light photocatalytic activity. To date, methods to induce Ti³⁺ have generally been energy intensive and involved unstable Ti raw materials (such as TiO [19] and TiH₂ [17]). In this context, it is highly desirable to develop an alternative method for preparing Ti³⁺ self-doped TiO₂.

Here, we successfully prepared self-doped TiO₂ using metallic Ti and H₂O₂ as raw materials. Our method shows great advantages over previous methods in that the concentration of Ti³⁺ species and oxygen vacancies can be easily tuned by controlling the amounts of added H₂O₂ and the hydrothermal reaction time. The involvement of H₂O₂ during the synthesis not only allows control over the dopant concentration but is also critical for controlling the microstructures. Nanorod-type self-doped TiO₂ can be synthesized by the addition of appropriate amounts of H₂O₂. Possible mechanisms for the formation of such microstructures have been proposed. Photocatalytic hydrogen experiments have confirmed the high photocatalytic activity of self-doped TiO₂. The correlation between the catalytic activity and the concentration of Ti³⁺ species is discussed.

Ti³⁺ self-doped TiO₂ was prepared by annealing in a N₂ atmosphere after a hydrothermal reaction. In a typical preparation procedure, 0.3 g of Ti power (99.99%, Aladdin) was mixed with 20 mL of water, 8 g of sodium hydroxide and a certain amount of H₂O₂ solution (30%) (0, 20, 50, and 100 μL). The suspension was transferred into an autoclave for the hydrothermal reaction. Various hydrothermal reaction times at 180 ℃ were used (12, 18, 24, and 30 h). The resultant precipitates were then collected, washed with hydrochloric acid and water until the pH of the washings was neutral. The obtained precipitates were grinded for approximately 30 min with an agate pestle and mortar and then calcined at 500 ℃ for 3 h. The samples were named according to the calcination atmosphere, the amount of H₂O₂ and hydrothermal reaction time. The samples calcined in N₂ were designated as 500-0-12, 500-20-12, 500-50-12, 500-100-12, 500-100-18, 500-100-24, and 500-100-30. An uncalcined sample was designated similarly as 100-24 (with 100 μL H₂O₂ and a hydrothermal reaction time of 24 h).

The photocatalytic hydrogen evolution was evaluated in a top-irradiation-type reactor connected to a gas closed circulation and evacuation system (Perfect Light, Labsolar-IIIAG). In a typical experiment, 0.1 g of the sample powders were dispersed in 100 mL of aqueous solution containing 10% methanol by volume. Pt was loaded onto the samples at 1 wt% by a photo-deposition method: H₂PtCl₆ aqueous solution was reduced into Pt nanoparticles over 1 h under full spectra irradiation [20]. A 300 W xenon lamp (Perfect Light, PLX-SXE300) was used as a light source. An AM1.5 filter was applied to simulate natural sunlight. The gas component within the reactor was then analyzed using an online gas chromatograph (Techcomp, GC7900) with a TCD detector.

Crystal structure was examined by X-ray powder diffraction (XRD) techniques (DX-2700B X-ray Diffractometer, Haoyuan). Raman spectra were collected on an inVia Raman Microscope (Renishaw). The morphologies of the samples were inspected with a field-emission scanning electron microscope (Hitachi S4800). The optical absorption spectra of samples were obtained at room temperature with a UV-Vis spectrophotometer (JASCO-750) and the JASCO software suite. BaSO₄ was used as a reference non-absorbing material. The X-ray photoelectron spectra (XPS) measurements were measured on a Thermo ESCALAB 250XI equipped with monochromatized Al K_α radiation using C 1s (284.8 eV) as the reference. Electron paramagnetic resonance (EPR) spectra were collected with the use of a JES FA200 spectrometer at room temperature.

XRD patterns of as-prepared samples are shown in Fig. 1(a). Analysis of these patterns suggested that 100 μL of H₂O₂ and a hydrothermal reaction time longer than 24 h were needed to fully convert all the metallic Ti. The subsequent calcination in a N₂ atmosphere at 500 ℃ was necessary to crystallize the hydrothermal precursors into an anatase phase (JCPDS Card no. 21-1272). Raman spectra were recorded to further examine the structure of the as-prepared sample. As shown in Fig. 1(b), five characteristic Raman peaks at 143.2 (E_g), 196 (E_g), 396 (B_1g), 515 (A_1g or B_1g) and 637(E_g) cm^-1 emerged for all samples after calcination, assignable to the Raman modes of the anatase phase [21, 22]. Thus, we confirmed the successful preparation of anatase TiO₂ from metallic Ti.

X-ray photoelectron spectroscopy (XPS) was further used to investigate the surface chemistry of the as-prepared samples. As shown in Fig. 2(a), binding energies for the Ti 2p state in the 500-100-12 samples were consistent with typical peaks of TiO₂ with Ti 2p_3/2 (458.43 eV) and 2p_1/2 (464.12 eV) peaks [9, 16]. The O 1s state in Fig. 2(b) featured two overlapping peaks: one locating at 529.69 eV was typical for lattice oxygen in TiO₂; the other one located at 531.29 eV was assigned to surface OH groups [15]. Compared with the 100-12 samples (Fig. 2(c) and (d)), there was a small decrease in the binding energy of Ti and O after calcining the sample in N₂ at 500 ℃, indicating the importance of the thermal treatment in weakening the bond strength between Ti and O. Interestingly, signals from elemental Ti were not detected in either of these two samples, likely because of the detection limitations of XPS, which can generally only penetrate to a depth of 1-3 nm from the surface. High-resolution TEM (HRTEM) images of the 500-100-12 samples (Fig. S1) showed that the distance in the crystal lattices of the outer layer was 0.352 nm, which is consistent with the (101) lattice planes of anatase. The binding energies of Ti in 500-100-24 and 500-100-30 samples were also ascribed to Ti⁴⁺ (Fig. S2). For all samples, no Ti³⁺ peaks were detected by XPS.

Electron paramagnetic resonance (EPR) is a useful technique to detect paramagnetic species owing to its high sensitivity toward unpaired electrons. Notably, the calcined samples exhibited strong responses at g-values of ~2.001 and ~1.997 in their EPR spectra (Fig. 3(a)), which could be attributed to surface oxygen vacancies [23, 24]. The amount oxygen vacancies appeared to increase as the amount of added H₂O₂ was increased and also correlated with the hydrothermal reaction time. However, the uncalcined sample (100-12) did not show oxygen vacancy signals. Notably, few nanorods were formed by the hot alkali method in the 100-12 samples and the nanorods length was short. Conversely, a number of long nanorods were formed in the 500-100-12 samples after calcination, as shown in Fig. 3(b) and (c) (larger version in Fig. S3). In addition, the nanorod structure was hardly found in the 500-20-12 (Fig. 3(e)), 500-0-12 (Fig. 3(f)), 500-100-24 (Fig. S4a), and 500-100-30 (Fig. S4b) samples except for the 500-50-12 (Fig. 3(d)) samples.

The presence of oxygen vacancies may be ascribed to either the reduction of TiO₂ by the remaining metallic Ti or the loss of oxygen by the removal of surface OH groups. According to XRD results of the uncalcined sample (Fig. S5), the amount of Ti decreased as the hydrothermal reaction progressed and Ti was fully converted with a hydrothermal reaction period of 24 h, implying that the contribution from metallic Ti to the formation of oxygen vacancies should decreased over a long hydrothermal time. However, the opposite trend was found. More oxygen vacancies were found for the sampled subjected to long hydrothermal time. In this regard, the oxygen vacancies might be attributed to the removal of surface OH groups. This loss of oxygen from Ti-O might be associated with a change of the morphology. We propose the following mechanism. On the basis of the morphology of 500-0-12, 500-20-12, 500-50-12, and 500-100-12 samples, it is likely that H₂O₂ promoted the formation of hydroxyl groups bonded with Ti, such that Ti-O-Ti bonds were generated along one direction through selective dehydration of Ti-OH groups in a certain region under the calcination conditions. In other regions, the hydroxyl groups that were not dehydrated could be removed to form oxygen vacancies. On the basis of our EPR results for the 500-100-12, 500-100-24, and 500-100-30 samples, we deduced that an appropriate vacancy defect content enabled dehydration of Ti-OH along one direction owing to a change of surface tension and Coulomb forces acting on the surface, which induced the formation of nanorods from disordered structures [25]. In the calcination process, the oxygen vacancy formation process through the removal of hydroxyl groups occurs prior to the generation of Ti-O-Ti bonds via a dehydration process of Ti-OH groups. The reason is probably due to the presence of oxygen vacancies which may direct the formation of Ti-O-Ti bonds.

Fig. 4 shows the ultraviolet-visible (UV-Vis) absorbance spectra of freshly prepared samples. According to the results of 500-100-24 and 500-100-30 samples, the absorbance increased with higher Ti³⁺ content, which might also correlate with better photocatalytic activity. Hydrogen production from water splitting was performed to investigate the photocatalytic activity of the as-prepared samples, as shown in Fig. 5(a). Compared with the 500-100-12 samples, the 100-12 sample did not show any discernable photocatalytic activity, likely because of its poor crystallinity. On the basis of the results for 500-0-12, 500-20-12, 500-50-12, and 500-100-12 samples, the amount of H₂O₂ added during their synthesis was an important factor determining the photocatalytic activity thorough its control over the amounts of oxygen vacancies in the samples. However, there seemed to be certain oxygen vacancy content for optimal photocatalytic activity because further increases of oxygen vacancies at longer hydrothermal times (sample 500-100-30) decreased the activity. According to theoretical calculations [16, 26-30], oxygen vacancies tend to elevate the valence band maximum (VBM) 0.4 V higher than that of normal titania by forming a tail in the density of states, which is generally favorable for catalytic activity. Fig. 5(b) shows valence band (VB) XPS of 500-100-24 and 500-100-30 samples. The VB XPS of 500-100-24 shows that the main absorption onset was located at 2.16 eV below the Fermi energy, whereas the maximum energy associated with the band tail blue-shifted further toward the vacuum level at approximately 1.7 eV. The values of 500-100-30 were 2.62 and 2.02 eV, respectively. For the 500-100-24 sample, the tail was approximately 0.46 eV above the VBM which was consistent with the theoretical value that should result in better photocatalysis. Furthermore, oxygen vacancies also introduce localized states below the conduction band (CB) minimum [30, 31], and the width of the vacancy band increased together with the concentration of vacancies. Large amounts of oxygen vacancies in the 500-100-30 samples likely shifted the vacancy band to special location below the CB, resulting in low mobility of the electrons at these levels and their becoming electron/hole recombination centers. These findings resemble results for reduced rutile titania [27, 29]. Therefore, an optimal oxygen vacancy level exists for these samples.

In summary, Ti³⁺ self-doped anatase titania was prepared by removing hydroxyl groups through high temperature calcination. When the amount of H₂O₂ was 100 μL and the hydrothermal reaction time reached 12 h, appropriate oxygen vacancy defects formed after calcining in a N₂ atmosphere. These defects enabled dehydration of Ti-OH along one direction through a change of the surface tension and Coulomb forces to form nanorod structures from the disordered structure. Based on our EPR and morphology studies, we deduced that hydroxyl groups were removed prior to the dehydration process. Comparing the 500-100-24 with 500-100-30 samples, we found that more oxygen vacancies can improve visible-light absorption but do not necessarily contribute to photocatalytic activity. The VB of the samples determined by XPS and previous calculations suggested that these findings can be ascribed to the following: (1) The tail of the VBM was approximately 0.46 eV above the VBM of 500-100-24, which was consistent with the theoretical value believed to contribute to enhanced photocatalysis; (2) more oxygen vacancies in the 500-100-30 samples likely caused the vacancy band width reaching to special location below CB, where it acted as an electron/hole recombination center.