An in situ exploration of subsurface defect migration to a liquid water‐exposed rutile TiO2(110) surface by XPS

The ability of titanium dioixide to split water into OH− and H+ species is heavily dependent on the behaviour of defects in the crystal structure at or near the surface. We present an in situ study of defect migration in rutile TiO2(110) conducted using X‐ray photoelectron spectroscopy (XPS). First, surface and subsurface defects were created in the crystal by argon ion sputtering. Subsequent in situ exposure of the defective crystal to liquid water healed the surface defects, whereas the subsurface remained defective. The sample was then annealed while XPS was used to monitor the concentration of titanium defects. At low annealing temperatures, Ti3+ was observed to migrate from the subsurface to the surface. Further annealing gradually restored the surface and subsurface to the defect‐free Ti4+ form, during which the changes in abundance of Ti1+, Ti2+ and Ti3+ defects are discussed.

The ability of titanium dioixide to split water into OH − and H + species is heavily dependent on the behaviour of defects in the crystal structure at or near the surface.
We present an in situ study of defect migration in rutile TiO 2 (110) conducted using X-ray photoelectron spectroscopy (XPS). First, surface and subsurface defects were created in the crystal by argon ion sputtering. Subsequent in situ exposure of the defective crystal to liquid water healed the surface defects, whereas the subsurface remained defective. The sample was then annealed while XPS was used to monitor the concentration of titanium defects. At low annealing temperatures, Ti 3+ was observed to migrate from the subsurface to the surface. Further annealing gradually restored the surface and subsurface to the defect-free Ti 4+ form, during which the changes in abundance of Ti 1+ , Ti 2+ and Ti 3+ defects are discussed.

K E Y W O R D S
ambient pressure XPS, defects, titanium dioxide, water, X-ray photoelectron spectroscopy

| INTRODUCTION
The production of hydrogen through photocatalytic water splitting is an appealing next-generation energy technology with potential to provide a low-cost, sustainable and environmentally responsible fuel. 1,2 For an overview of photocatalysis, we would direct the reader to one of the many recent review articles on the topic. [3][4][5][6] This paper discusses titanium dioxide, an attractive photocatalyst due to its chemical stability, abundance and nontoxicity and has been widely considered as a candidate not only for water splitting 7 but also water purification 8,9 and numerous other photocatalytic applications. 10 Further, titanium dioxide is well suited to being sensitised with dye molecules to increase visible light absorption 11 or can be doped to modify photocatalytic properties, 12 for example, stabilising reactive facets on the surface of the catalyst. 13 The catalytic activity of titania depends heavily on the surface crystallinity. For anatase, the photocatalytic activity can be tuned by the exposure of (101) and (001) facets, which act as reduction and oxidation sites, respectively. 14 The rutile (110) surface, which although not as catalytically active as anatase, can be used to split water and produce stoichiometric quantities of hydrogen and oxygen. 15 As part of the effort to develop suitable photoelectrodes for water splitting, single-crystal surfaces have been extensively studied in an attempt to better understand the fundamental surface chemistry governing catalytic performance. [16][17][18][19] The structure of the rutile titanium dioxide (110) surface is shown in Figure 1. The bulk crystal consists of sixfold coordinated titanium atoms and threefold coordinated oxygen atoms, but at the (110) surface, there is additionally rows of twofold coordinated oxygen, located at the bridge sites running in the [001] direction and fivefold coordinated titanium. 20,21 It is now well understood that defects in the crystal structure play a critical role in the photocatalytic performance of titanium dioxide. For example, the defect-free rutile TiO 2 surface does not dissociate/split water. 22,23 Water splitting on the rutile surface is instead largely dictated by oxygen vacancies in the rows of bridging oxygen atoms and is often dependent on the nature and density of these surface defect sites. [24][25][26][27] The dissociation of a water molecule into H + and OH − occurs near an oxygen vacancy site where the OH group fills an oxygen vacancy (V bridge ) and the H binds to a nearby bridging oxygen site (O bridge ) forming an additional OH species on the surface. This mechanism can be outlined as follows: Defects in TiO 2 (110) can be easily prepared by Ar + ion sputtering, which damages the crystallinity and preferentially removes oxygen atoms from both the crystal surface and subsurface layers. 28 The resulting reduced titania mainly consists of two types of defect: oxygen vacancies (missing atoms from the bridging oxygen rows) and titanium interstitials (titanium atoms not assigned to a lattice point that can move through the crystal). 20 After sputtering, it is well known that the defect-free rutile structure can be restored by annealing in vacuum. This process is entirely reliant on the mobility of the defects, which can be explained through two mechanisms. First, during 'oxygen vacancy diffusion', the undamaged crystal bulk supplies O 2− anions to the reduced surface/subsubsurface. Second, 'Ti 3+ interstitial diffusion' describes the bulk receiving excess Ti 3+ species from the surface/subsurface. In both cases, the migration of defects to and from the bulk allows the defect-free rutile structure to be restored. 28 Given the importance of defect chemistry on the catalytic activity of titania, it is not surprising that the interaction between water and defective titanium dioxide surfaces has been extensively studied using an array of techniques. 26,[29][30][31][32][33][34] In particular, near-ambient pressure X-ray photoelectron spectroscopy (NAP-XPS) is particularly well suited having been used to study the role of oxygen vacancies as water nucleation sites on rutile 24 and anatase. 35 Current understanding is that oxygen vacancies on the surface of anatase migrate to the subsurface 35 where they are more stable and less reactive. 36 Water dissociation occurring on the surface is enhanced at sites above a subsurface defect, which causes the binding to be more favourable. 37 This differs from rutile where oxygen defects are more common on the surface (compared with the subsurface), where they are stable and tend to remain. 35,38 Water dissociation on rutile occurs directly at these surface oxygen vacancies. 39 This difference, which has also been studied computationally, 40 is thought to explain the relatively increased catalytic reactivity of the anatase polymorph compared with rutile as the defects, which can trap photoexcited charge carriers, are not so easily quenched by adsorbates, including water, if they remain subsurface.
In this study, highly defective rutile TiO 2 (110) was prepared by Ar + sputtering; the surface of which was healed in situ by exposure to liquid water in a NAP-XPS instrument. The resulting combination of a defect-free surface and highly defective subsurface is used as a platform for an XPS study of annealing-induced defect migration.

| METHOD
Measurements were taken using a SPECS DeviSim NAP-XPS instru- highly defective sample was subsequently exposed to liquid water with the aim of healing the defects at the very surface of the crystal (the in situ data collected during this process are included in the supporting information). The resulting spectrum (water-exposed NE) in Figure 2 shows that the additional Ti oxidation states decrease in relative intensity. This preparation method repeatedly produced very similar defect concentrations; measurements showing this repeatability are included in the supporting information.
The surface sensitivity of XPS can be varied by changing the angle of the sample with respect to the spectrometer. 45 We predict the escape depth of Ti 2p photoelectrons at NE (0 ) and GE (70 ) to be 7 and 2 nm, respectively, calculated by considering the inelastic mean free path (IMFP) needed to attenuate the substrate signal by 95% (method described elsewhere 46 ). The NE and GE traces in Figure 2 therefore indicate that the water-exposed surface is less defective than the bulk as, when compared with the NE data, the more surface-sensitive GE spectrum (water-exposed GE) shows fewer defect states relative to the Ti 4+ . This conclusion is consistent with an independent study where defects at the surface were healed more readily than those subsurface. 47 The defect states are not expected to disappear completely in the GE spectrum presented in Figure 2 as the calculated information depth is still 2 nm and it is assumed that water exposure only heals defects directly accessible on the surface.
Annealing this 'water-healed' sample in UHV produces the data shown in Figure 3, which provides information about how the defect states change as a function of time and temperature. During the first 3 min of annealing, the defect states increase in intensity. After which we see the defect concentration slowly diminish until the spectrum is dominated by Ti 4+ , which has occurred by the time the temperature approaches 700 K.
For better understanding, five individual spectra have been extracted from Figure 3 at points labelled (a)-(e). These are plotted, alongside the corresponding O 1s spectra, in Figure 4 where curve fitting allows the spectral contribution of the different Ti oxidation states to be determined. The peak located at 458.5 eV corresponds to the Ti 2p 3/2 of the Ti 4+ oxidation state. 48 The three peaks located at 457.2, 455.9 and 454.6 eV are attributed to the Ti 3+ , Ti 2+ and Ti 1+ oxidation states, respectively. The corresponding Ti 2p 1/2 components were determined in the fitting procedure; a fixed spin-orbit splitting F I G U R E 2 Ti 2p XPS of 'clean' (after UHV cleaning/annealing), 'sputtered' (after 2-keV Ar + sputtering) and 'water-exposed' (after liquid water exposure) surfaces where peak fitting tracks the behaviour of the Ti oxidation states. All three surfaces were measured in normal emission (NE). Additionally, the 'water-exposed' surface was measured at the more surface-sensitive angle of 70 grazing emission (GE) of 5.70 eV led to Ti 2p 1/2 components with relative areas of 0.45 ± 0.03 compared with the corresponding Ti 2p 3/2 components. The Ti 3+ , Ti 2+ and Ti 1+ peaks are likely to be broadened with a rather complex line shape due to multiplet splitting. 49 We have opted to keep the peak fitting as simple as possible and not modelled the multiplet splitting, which in reality will lead to an asymmetric broadening of the Ti 3+ , Ti 2+ and Ti 1+ on the higher BE side and the intensity of the multiplet peak is likely very small in comparison with the main peak so likely only makes a small difference to the quantitative analysis we present.
During the first 3 min, the features attributed to Ti 3+ are observed to grow during annealing as shown in Figure 4a The O 1s spectra in Figure 4 are fitted with three peaks. The peak at 530.05 eV is attributed to lattice oxygen, whereas the features located at 1.3 and 2.6 eV higher binding energy correspond to adsorbed hydroxide and water, respectively. The relative positions of these features are consistent with a previous study of a waterexposed rutile TiO 2 (110) surface. 24 The intensity of the adsorbed hydroxide and water features remains comparable during the first two spectra (a,b). In spectrum c, at t ≈ 5 min and T ≈ 490 K, we start to see a loss of absorbed water. By the fourth spectra d at t ≈ 17 min and T ≈ 620 K, the water is no longer observed and the OH peak has started to diminish. Finally, by t ≈ 38 min and T ≈ 700 K (spectrum e), we see no hydroxide or absorbed water features. This is consistent with previous measurements of water-exposed TiO 2 surfaces showing that hydroxide and adsorbed water do not desorb below 550 K. 47 It is noteworthy that we do not see a reduction of the hydroxide/water features at lower temperatures perhaps implying that the water is not playing a dramatic role healing any Ti 3+ defects that may have migrated to the surface, potentially as physisorbed water is not kinetically active enough when compared with the gas and liquid phases, which are known to heal defects. In region A, that is, during the first 3 min, Figure 5 shows a decrease in Ti 4+ concentration with a corresponding increase in Ti 3+ , whereas the contribution of lower oxidation states remains constant.
This implies that at low annealing temperatures, Ti 3+ defects are likely diffusing from the subsurface to the surface until the concentration Normal emission Ti 2p XPS measured in situ during annealing of 'water-exposed' rutile TiO 2 (110). Horizontal lines labelled (a)-(e) were extracted for detailed analysis in Figure 4 F I G U R E 4 Left: Ti 2p spectra (extracted from Figure 3) where peak fitting tracks the behaviour of the Ti oxidation states during annealing. Right: associated O 1s spectra highlighting the relative change of lattice oxygen, hydroxyl groups and adsorbed water. All the spectra were measured at normal emission gradient between the surface and subsurface is minimised as the Ti 4+ reaches a minima at 3 min.
In region B, between 3 and 6 min, we observe that the Ti 1+  Figure 4) as a function of time and temperature. Shaded regions around the data points indicate an estimated error bar calculated from the standard deviation of the fit residual. Regions A to C represent stages of the annealing process as discussed in the text. Arrow indicated label (a)-(e) corresponds to the spectra presented in Figure 4 49. Gupta RP, Sen SK. Calculation of multiplet structure of core pvacancy levels.