After heating the sample to 900°C in a vacuum, the bulk-truncated rutile TiO2 (110)-(1×1) surface undergoes reconstruction. The high angle annular dark field (HAADF) STEM profile images (as depicted in Fig. 1a) along the [001] zone axis reveal numerous periodically spaced protrusive bright-dot pairs on the (110) surface. These bright dots could be attributed to the Ti atomic columns, a conclusion supported by the Z-contrast of the HAADF STEM image, where Z represents the atomic number.34,35 The spacing between consecutive Ti column pairs was ~ 1.31 nm, confirming the anticipated (1×2) reconstruction. In bright field (BF) STEM image (Fig. 1b), the Ti columns appear as dark-dot pairs. Upon closer examination of this BF STEM image and specifically focusing on the magnified portion in Fig. 1b, we can identify three protruding O columns (slightly brighter dark-dots, highlighted with red arrows) positioned between the Ti-column pair and the subsurface layer. These observed features align well with the Ti2O3-(1×2) reconstruction model. However, an intriguing discovery is the varying heights of the Ti columns within each pair, with difference ranging from 0.02 to 0.07 nm. This suggests an inherent asymmetry in each pair of protruding Ti rows on the reconstructed TiO2 (110)-(1×2) surface, a finding that challenges the widely accepted Ti2O3-(1×2) reconstruction model.
We conducted experiments to delve deeper into the intrinsic properties of the (1×2) reconstructed rutile TiO2 (110) surfaces, focusing on how the surface structures evolve in various environments. Initially, the TiO2 (110) surface adopts a (1×1) structure when heated to 700°C under an oxygen pressure of 2.53×10− 3 Pa, as shown in Supplementary Fig. 1a. Upon evacuating the O2 gas, the (1×2) reconstruction begins to manifest progressively on the surface (illustrated in Supplementary Fig. 1b and 1c). At the beginning of the reconstruction, three TiOx double-rows can be observed with the dotted rectangles, presenting a 3d-3d pattern (with “d” denoting the periodicity of the 1×1 bulk terminated surface). As the reconstruction unfolds, the central double-row slowly bifurcates into two distinct double-rows, culminating in a 1d-1d-1d pattern. Supplementary Fig. 1d offers intensity profiles of the top layer, aiding in a deeper analysis of the surface structure. Several structural models have been proposed to elucidate the (1×2) reconstruction of the rutile (110) surface.
To support the experimental observations, systematic DFT calculations are performed. It is known that there exist multiple distributions of the small polarons/Ti3+ ions on the near-surface region of reduced TiO2.36 To locate the ground state and understand the corresponding electronic structures of the oxygen-deficient Ti2O3-(1×2) reconstructions, we have investigated different distributions of the small polarons/Ti3+ ions within the reconstructed titanium protrusion, the top two titanium layers, and some selected deeper titanium layers (Supplementary Fig. 2). It is found that the four small polarons/Ti3+ ions of the Ti2O3-(1×2) reconstructions have the tendency to stay at the near surface area and tend to separate from each other due to their repulsive electrostatic interactions. There exist symmetric and asymmetric Ti2O3-(1×2) reconstructions with the asymmetric reconstruction being more stable by around 0.3 eV, which is consistent with our in situ STEM observations that most of the Ti2O3-(1×2) reconstructions are asymmetric. We note that at high temperatures the vibrational entropy can affect the relative stability of different structural configurations of material surfaces.37 The free energy differences of the symmetric and asymmetric Ti2O3-(1×2) reconstructions at different temperatures, incorporating the vibrational entropy, are collected in Supplementary Table 1. The thermodynamic stability of the asymmetric reconstruction is even higher relative to its symmetric counterpart at high temperatures due to its disorder mediated higher vibrational entropy. The most stable configurations of the symmetric and asymmetric Ti2O3-(1×2) reconstructions are displayed in Fig. 2. For the symmetric Ti2O3-(1×2) reconstruction (Fig. 2a), the two columns of titanium protrusion have the same height relative to the substrate. While, for the asymmetric Ti2O3-(1×2) reconstruction (Fig. 2b), the lower titanium column has relaxed downward by 0.16 Å while the higher titanium column has relaxed upward by 0.42 Å relative to the symmetric Ti2O3-(1×2) reconstruction. These structural characteristics match well with our experimental STEM images.
We next discuss the underlying formation mechanism of the asymmetric Ti2O3-(1×2) reconstruction. Interestingly, we found the four small polarons/Ti3+ ions are exclusively located at the first titanium layer for the most stable asymmetric Ti2O3-(1×2) reconstruction (Fig. 2b). While, there are three small polarons/Ti3+ ions located at the first titanium layer and one located at the second titanium layer for the most stable symmetric Ti2O3-(1×2) reconstruction (Fig. 2a). The distinct distributions of the small polarons/Ti3+ ions for the symmetric and asymmetric Ti2O3-(1×2) reconstructions can be explained in terms of the surface relaxation effects and the effects of the repulsive electrostatic interactions induced by the small polarons/Ti3+ ions. To maintain the symmetry of the Ti2O3-(1×2) reconstruction, the distribution of the small polarons/Ti3+ ions at the first titanium layer must possess a high degree of local symmetry (Supplementary Fig. 2 (g-j)). As shown in Supplementary Fig. 2i and 2j, the four small polarons/Ti3+ ions located at the first titanium layer of the symmetric Ti2O3-(1×2) reconstruction have two types of d-like orbital characters (\({d}_{{x}^{2}{-y}^{2}}\) and \({d}_{{z}^{2}}\)) with the \({d}_{{x}^{2}{-y}^{2}}\) orbital character centered on the six coordinated titanium sites and the \({d}_{{z}^{2}}\) orbital character centered on the five coordinated titanium sites. To reduce the repulsive electrostatic interaction between the first-nearest neighbored small polarons/Ti3+ ions, the small polarons/Ti3+ ions with \({d}_{{z}^{2}}\) orbital character will diffuse from the first titanium layer to the second titanium layer (Fig. 2b and Supplementary Fig. 2g) without breaking the symmetry of the distribution of the small polarons/Ti3+ ions at the first titanium layer. This process will release the surface energy by around 0.26 eV with the structural symmetry of the Ti2O3-(1×2) reconstruction maintained. Another alternative way to reduce repulsive interaction between the first-nearest neighbored small polarons/Ti3+ ions is the diffusion of the small polarons/Ti3+ ions from the first-nearest neighbored titanium site to the second-nearest neighbored titanium site within the first titanium layer by breaking the local symmetry of the distribution of the small polarons/Ti3+ ions to form the asymmetric Ti2O3-(1×2) reconstruction (Fig. 2a and Supplementary Fig. 2a). It is found the latter one is energetically more favorable by around 0.3 eV (the energy released is as large as 0.57 eV). This can be rationalized by the fact that the small polarons/Ti3+ ions can effectively relax at the first titanium layer compared to the second titanium layer. The average distance between the Ti3+ ion and the oxygen is 2.08 Å when Ti3+ ion is located at the first titanium layer while it is 2.06 Å when Ti3+ ion is at the second titanium layer, which supports the surface relaxation effect as well.
Furthermore, we found the surface electronic states are sensitive to the symmetry of the Ti2O3-(1×2) reconstructions due to the surface relaxation effects and the effects of the repulsive electrostatic interactions induced by the small polarons/Ti3+ ions. As shown in Fig. 3, the occupied 3d orbitals of the small polarons/Ti3+ ions are located between the occupied 2p orbitals of O and unoccupied 3d orbitals of Ti near the Fermi level. The occupied 3d orbitals of the small polarons/Ti3+ ions, for both symmetric and asymmetric Ti2O3-(1×2) reconstructions, have more than one peak due to different coordination environment of the located titanium sites. For symmetric Ti2O3-(1×2) reconstruction (Fig. 3a), majority of the occupied 3d orbitals are located at high energy levels. Conversely, for the asymmetric Ti2O3-(1×2) reconstruction (Fig. 3b) minority of the occupied 3d orbitals are located at high energy levels. This means that the center of the occupied 3d orbitals for asymmetric Ti2O3-(1×2) reconstruction is at lower energy than that of the symmetric Ti2O3-(1×2) reconstruction, which can also be used to explain the higher stability of the asymmetric Ti2O3-(1×2) reconstruction. Both the band gap and the work function of the asymmetric Ti2O3-(1×2) reconstruction are found to be smaller than that of the symmetric Ti2O3-(1×2) reconstruction (Supplementary Fig. 3), which indicate the asymmetric Ti2O3-(1×2) reconstruction possesses different chemical and physical properties compared to the symmetric Ti2O3-(1×2) reconstruction.
To further understand how the asymmetric surface reconstruction of rutile TiO2 (110)-(1×2) responds to gas environments, we examined its dynamic evolution under oxygen gas environment. The stages of this evolution for the (1×2) reconstructed rutile TiO2 (110) surface are depicted in Figs. 4a-f. Initially, the (1×2) reconstruction was in situ fabricated on the (110) surface at 700 ℃ under a vacuum environment of 6.00×10− 4 Pa. Upon introducing oxygen gas with a pressure of 6.00×10− 2 Pa, this reconstruction becomes unstable. In situ HRTEM images depict the dynamic change of a two-row structure (Figs. 4a-f). During this process, the distance between the two TiOx rows expands, leading to the emergence of a new adjacent TiOx row (Figs. 4e, f). This evolving pattern is mirrored in the intensity profiles of the reconstructed layer, as shown in Fig. 4g. Notably, the pronounced contrast at the reconstruction sites transitions to a more uniform contrast, aiding in the analysis of the intermediate states during the reconstruction. These findings show that the TiO2 (110)-(1×2) is more stable under low oxygen chemical potential conditions, while the unreconstructed surface is more dominant at high oxygen chemical potentials. This offers a method to modulate these structures by adjusting the oxygen chemical potential.
In summary, we elucidated the asymmetric surface reconstruction of rutile TiO2 (110)-(1×2) by leveraging both scanning transmission electron microscopy and density functional theory calculations. This specific reconstruction arises primarily from surface relaxation effects coupled with the repulsive electrostatic interactions induced by the small polarons/Ti3+ ions. The strong coupling of geometric and electronic structures induces different chemical and physical properties in the asymmetric Ti2O3-(1×2) reconstruction as compared to its symmetric counterpart. Additionally, the pronounced oxygen sensitivity of the asymmetric Ti2O3-(1×2) reconstruction emphasizes its inherent dynamism. These findings not only deepen our understanding of the structural and electronic properties of reconstructed TiO2 surfaces but also pave the way for fine-tuning surface reconstructions, which have important implications in the optimization of the performance of titanium dioxide-based devices.