Facet-engineered TiO 2 nanomaterials reveal the role of water – oxide interactions in surface protonic conduction † Journal of Materials Chemistry A

Water adsorption and surface protonic conduction have been investigated at 25 – 400 (cid:1) C in wet (H 2 O and D 2 O) atmospheres on nanocrystalline TiO 2 hydrothermally grown to a predominance of di ﬀ erent crystal facets. In situ Fourier transform infrared spectrometry shows that a portion of the water molecules in the ﬁ rst chemisorbed layer adsorbs dissociatively on the {001} and {100} surfaces, but molecularly on the {101} surface. The next layer of physisorbed molecular water is solid (ice)-like on the {001} and {100} surfaces, a fact that is attributed to relatively strong bonds to the surface terminating hydroxyls. On the other hand, it is looser, liquid-like on the {101} surface due to the lack of rigid hydroxyl groups to bond to. As relative humidity (RH) increases with decreasing temperature, additional water layers physisorb and eventually become liquid-like also on the {001} and {100} surfaces. The di ﬀ erences can be attributed to the di ﬀ erent atomic arrangements and surface energies of the di ﬀ erent crystal facets. Surface protonic conduction on {001} and {100} dominated samples exhibits high pre-exponentials, indicating a high concentration of sites for dissociative chemisorption, and high activation energies re ﬂ ecting di ﬃ cult proton migration in the rigid water layers on those surfaces. The surface protonic conduction on the {101} dominated sample re ﬂ ects a lower coverage of dissociable water and lower activation enthalpies for protonic migration in the more loosely bonded water. This suggests that, more generally, surfaces with dissociative chemisorption may exhibit high surface protonic conductivity at the highest and lowest temperatures (corresponding to lowest and highest RH, respectively), while surfaces with molecular chemisorption may exhibit a relatively high conductivity (less deep minimum) at intermediate temperatures. H/D isotope e ﬀ ects show that the protonic conduction mechanism changes from Grotthuss to vehicle as the physisorbed water layers gain thickness and become liquid-like. The protonic conductivity decreased when the surface was terminated by more strongly bonded ﬂ uoride ions, believed to re ﬂ ect that they block the sites for water adsorption.


Introduction
Surface protonic conduction comprises Grotthuss migration of free protons (H + ) and vehicular migration mainly of hydronium ions (H 3 O + ) in chemisorbed and physisorbed water layers on oxides or other materials. 1 Mesoporous and other nanocrystalline TiO 2 materials exhibit high surface protonic conductivity in wet atmospheres, of importance for use as photocatalysts, sensors, and electrodes, and proposedly also as electrolytes in fuel cells. [2][3][4] As an electrolyte material, TiO 2 cannot compete with the traditional proton exchange membranes (PEM), such as Naon® that typically operates at temperatures between 80 and 100 C with high relative humidity (RH), 5 and proton conducting ceramics like Y-substituted BaZrO 3 (BZY) employed at higher working temperatures (>400 C). 6 However, over a broad range of temperatures between 400 C and room temperature, TiO 2 nanomaterials exhibit considerable protonic conductivities 7,8 that in the case of anatase nanolms have even been reported to exceed that of Naon® at low temperatures. 3 Previous studies on the surface protonic conduction of TiO 2based nanomaterials mainly focused on the effects of the specic surface area, pore networks, and grain size, as well as intrinsic or extrinsic dopants. 7,[9][10][11] For example, Maglia et al. found that the surface protonic conductivity of dense anatase samples with grain sizes between 24 and 56 nm increased strongly with decreasing grain size. 10 Stub et al. reported that donor and acceptor doped porous TiO 2 materials give rise to increased and decreased surface protonic conductivity, respectively, compared to undoped TiO 2 . Based on this, they suggested the possible effect of heterovalent doping on the acid-base properties of the TiO 2 surface. 11 The results of the past few years imply that the orientation of different crystal facets of TiO 2 signicantly affects water adsorption and dissociation, 12 however, to the best of our knowledge, their inuence on surface protonic conductivity has not been systematically investigated.
The adsorption of the rst layer of water on oxides is generally considered to be the chemisorption of water molecules that bond to terminating cations, followed and stabilized by various degrees of dissociation into adsorbed hydroxyl groups and protons. The latter bonds to terminating surface oxide anions forming another surface hydroxyl group. As RH increases, a second layer of molecular water physisorbs fairly strongly by hydrogen bonds to the chemisorbed layer underneath. The possibility to make double hydrogen bonds to dissociated hydroxyl groups in a dissociated chemisorbed layer makes it rather crystalline and solid ("ice-like"). At high RH, a few layers more of water molecules physisorb by weaker hydrogen bonds that make them randomly oriented ("liquidlike").
Dissociated protons migrate by Grotthuss-type hopping in the chemisorbed and ice-like physisorbed layers, while at RH > 0.6 (60%) vehicular transport of H 3 O + ions in the liquid-like water takes over. 11,13 Generally, abundant dangling bonds, such as undercoordinated cations and oxide anions, serve as sites for water adsorption and dissociation, 12 16 Different surface energies, atomic arrangements, and steric hindrance lead to different degrees of molecular or dissociative adsorption characteristics, and we here investigate the effects of these variables on surface protonic conduction.
The extent and conguration of molecularly and dissociatively chemisorbed water on the different facets of anatase TiO 2 are debated. For example, Vittadini et al. 17 found that molecular adsorption on the {101} facet is energetically favored over dissociative adsorption by 0.44 eV at low coverage and by 0.28 eV at monolayer coverage. They report that on the {001} facet, low coverage is dominated by dissociative adsorption, which turns into mixed molecular/dissociative adsorption at a monolayer coverage. Based on synchrotron radiation core level photoelectron spectroscopy, Walle et al. 18 proposed mixed molecular and dissociative water adsorption on the {101} facet.
In this work, in situ Fourier transform infrared spectrometry (FTIR) has been utilized to study the adsorption, dissociation, and water layer formation as a function of RH and temperatures on anatase nanocrystallites with predominance of three lowindex facets {101}, {001}, and {100}. Surface protonic conductivity on such facet-engineered porous anatase samples was measured with impedance spectroscopy under controlled temperatures and xed partial pressure of water. Additionally, the effects of H/D isotope exchange and uoride ion termination were studied to improve our understanding of the processes underpinning the surface protonic conduction. We model the conductivity in terms of pre-exponentials and activation enthalpies of contributions from bulk, Grotthuss protonic conduction in the chemisorbed and ice-like physisorbed layers, vehicular conduction in the liquid-like layer, and correlate them with adsorption, dissociation, and layer rigidity on the differently terminated surfaces, an approach possibly applicable also to surface protonic conduction of other oxides.

Experimental
Synthesis of anatase TiO 2 nanoparticles with predominance of different facets TiO 2 nanoparticles with a predominance of the {101} facet (T101) were synthesized with a two-step hydrothermal procedure based on previous reports. 19 First, a precursor of potassium titanate powder with nanotubular morphology was prepared by adding 1 g of commercial TiO 2 powder (P25, Sigma-Aldrich) to 40 mL freshly made 17 M KOH (aq). The mixture was stirred for 20 min and introduced into a Teon vessel with an inner volume of 50 mL for the hydrothermal treatment at 110 C for 20 h. The resulting powder was rinsed with dilute HCl (aq) and deionized water until neutral (pH z 7). In the second step, the as-prepared potassium titanate powder was dispersed in 35 mL deionized water. The pH was adjusted to 3.5 with HCl (aq) during stirring and the mixture was transferred to a 50 mL Teon-lined autoclave for the hydrothermal reaction at 200 C for 12 h. The T101 powder was nally collected by centrifugation and washed with deionized water.
To synthesize anatase TiO 2 with a predominance of the {100} facet (T100), a similar procedure was used. In the rst hydrothermal process, instead of using P25, TiO 2 nanoparticles were prepared according to a previously reported method 20 and used as precursor to synthesize potassium titanate nanotubes. The as-synthesized nanotubes were washed with deionized water, dried at 80 C. In the second hydrothermal process, pH was kept neutral.
TiO 2 nanosheets with exposed {001} crystal surface (T001) were synthesized by a modied HF hydrothermal method according to the previous report. 21 In a typical procedure, 1.2 mL HF (aq) (0.6 mL 48% wt HF mixed with 0.6 mL deionized water) were added dropwise into the 10 mL of tetrabutyl titanate solution under stirring for 1 h. Then, the suspension was transferred into a 50 mL Teon-lined autoclave and kept at 200 C for 24 h. The precipitate was collected and washed with water and ethanol. Aer drying, the powder was dispersed in 0.1 M NaOH (aq), stirred for 3 h to remove the adsorbed uorine ions, and then washed with deionized water until neutral (pH z 7).

Sample characterization
Scanning electron microscopy (SEM, Hitachi SU8230, operated at 1 kV acceleration voltage) and X-ray diffraction (XRD, Bruker D8 Discover, Cu K a1 , Bragg-Brentano) were used to analyze the morphology and phase purity of the as-synthesized TiO 2 nanoparticles. Transmission electron microscopy (TEM) and highresolution TEM (HRTEM) were performed on an FEI Tecnai-G 2 F30 transmission electron microscope at 200 kV to determine the lattice parameters of the three samples. The surface adsorbates were detected using in situ FTIR (Bruker Vertex 70) equipped with a high-temperature and high-pressure transmission IR-cell (Harrick Scientic Products Inc, NY, US) and a mercury cadmium telluride (MCT) detector. Specically, the samples were outgassed under vacuum (p < 3 Â 10 À3 mbar). In situ measurements under owing O 2 (15 mL min À1 ) were performed during ramps from 30 to 400 C at 5 C min À1 . The applied water pressure was xed with 0.35 RH at 30 C. BET surface analyses were operated with nitrogen physisorption measurements in the range of pressure of 0.6587 kPa to 101.48 kPa at liquid nitrogen temperature.

Conductivity measurements
The as-prepared samples (T101, T001, and T100) were compacted into pellets with a diameter of 13 mm using a uniaxial cold press and then sintered at 400 C for 4 h in ambient air. The thickness was about 1.5 mm, and the relative density was about 50% calculated from the weight and geometry of the pellet. Ag ink was painted as electrodes on both sides of the pellet that was then dried at 120 C for 20 min. All the electrical measurements were carried out in a ProboStat™ sample holder cell (NORECS, Norway) with a standard 2-electrode 4-wire set up connected to a Novocontrol alpha-A impedance spectrometer via a ZG4 interface. A steel outer tube was used to electrically shield the cell. The impedance spectra were recorded from 10 MHz to 1 mHz at oscillation voltages of 0.1-0.3 V RMS in dry or wet (P H 2 O ¼ 0.026 atm or P D 2 O ¼ 0.023 atm) air as a function of temperature (25-400 C). The data were analyzed and modelled using the ZView equivalent circuit tting soware (ZView ver. 3.5a; Scribner Associates Inc.). The specic conductivity, s, of the porous material was calculated from the macroscopic dimensions, i.e., the area of the electrodes and thickness of the pellet.

Structure characterization
To investigate the morphology and crystal structure of the synthesized TiO 2 samples, XRD, SEM, HRTEM, and BET characterizations were performed. Fig. 1 shows the XRD patterns, matching well with the anatase phase (tetragonal, JCPDS 21-1272). Obviously, the (004) diffraction line of the T001 sample weakened and its full width at half maximum (FWHM) broadened. According to the Scherrer equation, the broad peak represents the decreased crystallite size along the [001] axis vertical to (001) plane, corresponding to a smaller thickness in that direction. 22 In addition, the (200) peak intensied and its FWHM narrowed, revealing the increased side length of T001 nanosheets in the [100] direction. Therefore, the thickness and side length in the [001] and [100] direction, respectively, can be roughly estimated based these two peaks. The relative peak intensity of (004) and (200) was estimated in the insert. T001 has the lowest value, indicative of dominance of {001} facets. 23 For the T100 sample, changes of the (004) and (200)  The lattice fringe with a d-spacing of 0.35 nm corresponds to (101) planes of anatase. 26 Additional SEM images and indicative particle dimensions of each sample can be found in Fig. S1. † Specic surface areas as calculated from N 2 adsorption isotherms at 77 K (Table S1 †) are higher for the T001 sample than the other two, attributed to its thin sheet-like morphology.

H 2 O-TiO 2 interactions
The surface of TiO 2 has both Lewis basic (O 2c ) and acidic (Ti 5c ) sites. Water adsorbs molecularly, which may be followed by dissociation, 12 as illustrated in Fig. 3. Molecular adsorption involves a direct interaction of the oxygen atom of H 2 O (O w ) with the surface Ti 4+ acidic site (Ti 5c ), or/and the hydrogen atom of H 2 O (H w ) with the surface O 2À . In the later dissociation, a basic site (O 2c ) can extract a proton from the water molecule that was adsorbed on the adjacent Ti 5c , forming two surface hydroxyl groups (the terminal hydroxyl group, OH t , and the bridging hydroxyl group, OH b ). 27 As discussed above, the surface atom coordination and geometric arrangement, as well as the resulting steric hindrance considerably affect the behavior on water adsorption on different anatase facets. However, previous studies investigating water adsorption and desorption were mainly conducted at very low temperatures and ultra-high vacuum (UHV), 12 where p H 2 O is low, but RH is considerable. In practical applications at ambient temperatures, p H 2 O is higher, and RH is still high. At higher temperatures, RH is oen lower, resulting in a lower coverage of adsorbed water. A main effect of increased temperature, even at the same RH, is increased dissociation of adsorbed molecular water. The question is how the thermodynamics of adsorption and dissociation vary with the conguration of surface sites on different facets. Fig. 4 shows the FTIR spectra of the as-prepared TiO 2 samples outgassed at elevated temperatures. From the evolution of -OH absorbance with temperatures, the degree of dehydration and thermal stability of various -OH groups is indicated. We can distinguish three main IR absorption bands. A broad absorption band at around 3600-2800 cm À1 corresponds to the stretching vibration modes of -OH group, n(OH), and the H 2 O molecule perturbed by hydrogen bonding (Fig. 4a). The broadness is related to the length and strength of several possible O-H bonds in a dynamic intermolecular interaction. A series of sharp peaks at around 3600-3750 cm À1 correspond to the unperturbed terminal hydroxyls and water molecules. 28 Peaks below 1800 cm À1 can be assigned to the bond bending of -OH groups, d(OH). Clearly, the amount of adsorbed water at the same RH is more signicant on the {001} and {100} facets, probably due to their higher surface energy with a large amount of under-coordinated sites compared to that of the {101} facet.
The evolution of the FTIR spectra due to the changed surface hydration species by the progressive outgassing from 30 C to 400 C, was monitored. As shown in Fig. 4b-d, the integrated intensity in 3600-2800 cm À1 gradually decreases with increasing temperature, and is completely attened out at 400 C for all the three TiO 2 samples, reecting the condensation of -OH groups and desorption of water. 28 The change in peak intensities on the T101 and T001 surfaces as the temperature increased from 30 to 40 C indicates the massive loss of water on these two surfaces.
Previous studies suggested that the formation of the ice-like layer is almost completed at 0.3 RH, and at 0.3-0.6 RH, the second physisorbed water layer forms. 29 Here, we propose that the chemisorbed water layer dissociates and mainly consists of -OH t and -OH b groups on the {001}    higher RH compared to the ice-like layer on {001} and {100} facets.
Above 0.3 RH, the second physisorbed water layer develops by weaker hydrogen bonds, and the outermost layer changes gradually from ice-like to liquid-like. 29 During the degassing process, a massive water loss at 40 C is observed for the {001} (Fig. 4b), reecting the weak hydrogen bonds in the outmost water layer. Therefore, it appears that the second physisorbed water layer that is thermally unstable forms on the {001} surface at 30 C with 0.35 RH, while its formation is less favorable on the {101} and {100} under these conditions.
As for the T101 sample, most water molecules desorb at temperatures up to 150 C, whereas they are still visible on the {001} and {100} facets at around 200-250 C (Fig. 4b-d), indicating the unstable nature of the less crystallized ice-like layer on {101} surface due to the lack of hydroxyls underneath, as well as the relatively weak interaction of water hydrogen and surface oxygen. Noticeably, a lower-lying absorbance maximum at around 3200 cm À1 is more visible for the T100 than the other two samples (Fig. 4c), which can be ascribed to the more crystalline (ice-like) water. In comparison, the one at around 3400 cm À1 is from the more amorphous and liquid-like water. 32 Hence, it is reasonable to infer that a crystallized ice-like water layer is established on the surface of these three samples during the in situ FTIR test, but the ice-like layer is more crystallized on the {100} surface than the other two.
A detailed analysis at the 3600-3750 cm À1 region where peaks are persistent even at 400 C is displayed in Fig. 5. For the T001 sample, two peaks at around 3651 cm À1 and 3677 cm À1 can be ascribed to OH b and OH t , respectively. When increasing the temperature to 400 C, they positively shi to 3660 cm À1 and 3683 cm À1 , respectively, due to the disappearance of the additional water molecules bonding to these -OH species. 33,34 A peak at around 3695 cm À1 is only seen in the low-temperature range (below 100 C), which can be assigned to the free H 2 O in the second physisorbed layer where the outermost layer with less Hbonds is involved. A similar change takes place on the T100 sample, in which the two peaks at 3663 cm À1 (OH b ) and 3697 cm À1 (OH t ) shi to 3670 cm À1 and 3715 cm À1 , respectively, as temperatures increase to 400 C. Noticeably, compared to T001, the OH t signal is more pronounced than OH b on T100 surfaces, which is probably due to its higher density of Ti 5c (100%) than O 2c (50%).
With regard to the T101 sample (Fig. 5), only one distinctive peak appears, which positively shis from 3691 cm À1 to 3695 cm À1 with the temperature increasing from 30 to 150 C, then negatively shis to 3670 cm À1 at 400 C. Based on the above discussion, this single peak belongs to H 2 O molecularly adsorbed on the {101} surface, and the peak-shiing probably originates from the desorption of water from Ti 5c -H   Based on the above discussion, water layers adsorbed on different facets of TiO 2 are displayed in Fig. 6. As the RH increases, different water layers gradually form on the surface of these three TiO 2 samples. Here, we use the {001} facet as an example to show all these water layers. The rst chemisorbed water layer forms at temperatures above 275 C with mainly hydroxyls (OH b and OH t ) on {001} and {100} facets and mainly molecularly adsorbed water on {101} facets. These results are in line with several theoretical ndings in the literature. [35][36][37] As RH increases, a hierarchical ice-like water layer starts forming on the top of the chemisorbed layer on {001} and {100} facets. On the {101} surface, the incoming water molecules rst bond to O 2c , 30 and then to the surface adsorbed H 2 O with hydrogen bonds are not as strong as those on the {001} and {100} facets where water molecules form hydrogen bonds with the surface -OH groups. The ice-like water layer is more rigid on the {100} facet than that on the {001} because of its special surface atomic arrangement. Previous theoretical calculations indicated that the coinciding orientations of water adsorbed on the surface active sites of {100} facets (with respect to the TiO 2 surface plane) yield hydration layers with an enhanced density, and lead to strongly ordered water layers. 16 As the RH further increases, a third water layer physisorbs rst on {001} surfaces, and propagates further into liquid-like layers at RH above 0.6.
As discussed in the introduction, the formation of protonic charge carriers originates from the surface water-oxide interactions, while the conduction of these charge carriers takes place in all water layers. Therefore, the proton conduction mechanism and conductivity on different crystal facets are expected to be different.

Conductivity measurements
Impedance spectra were measured for all samples as a function of temperature in various atmospheres. Temperatures were kept below 400 C to avoid sintering and change of crystal facets. A representative Nyquist plot of the impedance at 50 C of the {101} facet dominated TiO 2 sample, and the corresponding simplied equivalent circuit used to model all spectra, are displayed in Fig. S3. † Based on the associated capacitances, the impedance at high frequencies is ascribed to the transport in the bulk and over the surface of the grains. Although the high frequency impedance arc is asymmetrical, our data don't allow a consistent further deconvolution of this impedance, and we here assign only one resistance and conductivity to it. The response at low frequencies is assigned to the electrode impedance. The inset in Fig. S3 † shows the impedance in nominally dry air that is orders of magnitude higher at this temperature.
Based on impedance spectra and sample geometries, the conductivities of the different samples under both dry and wet conditions were extracted and form basis for Fig. 7. The curves were tted with TableCurve 2D soware. The pre-exponentials (s 0 ) and activation energies (E a ) of total conductivities corresponding to the bulk transport, the chemisorbed water layer, and the ice-like physisorbed layer were obtained separately, as shown in Fig. 7a, and the numerical results are provided in Table S2. † For the T101 sample, we tted the s 0 and E a for four different contributions: bulk transport, and transport in the Ti 5c -H 2 O layer, O 2c -H 2 O layer, and ice-like physisorbed layer.
For nominally dry conditions, the data, shown only at the highest temperatures in Fig. 7c-e (region I), exhibit linear Arrhenius behaviors with activation enthalpies (E a ) calculated from log(sT) vs. 1/T in the range of 0.8-1.2 eV. This can be attributed to the ionic conductivity of nanometric anatase, 38,39 and the variation between samples may be understood by their   Below approximately 300 C, the surface protonic contribution becomes signicant under wet conditions, giving rise to activation enthalpies in region II (chemisorbed water layer) in the range 0.3-0.4 for T001, 0.4-0.5 eV for T100 and 0.4-0.6 eV for T101. The difference between samples is clear from the direct comparison of the conductivities in wet atmosphere (p H 2 O ¼ 0.026 atm) in Fig. 7b. The conductivity in range II is attributed to mainly chemisorbed water, with proton migration by the Grotthuss mechanism. 3 For T001 and T100 we expect that the chemisorbed layer is at full coverage and fully dissociated, so that the activation energies reect the enthalpy of proton charge mobility (diffusion). They are in overall agreement with the 58 kJ mol À1 (0.60 eV) found for bulk proton migration in polycrystalline rutile and within the range of proton diffusivities in the different directions of single crystal rutile, 40 knowing that surface transport has somewhat lower activation enthalpies than in bulk. The activation energy of T101 should, on the one hand, contain some dissociation enthalpy, but may on the other hand reect easier jumps between the loosely bonded water molecules on the surface of T101 than the more rigid and dissociated water on T001 and T100.
At temperatures below approximately 200 C (region III, 1 st physisorbed layer), the conductivity increased with decreasing temperature for T001 and T100, as expected, attributed to the protolysis of the dissociated protons from the chemisorbed layer into the increasing coverage of physisorbed water, where the mobility is higher. 41,42 The conductivities down to around 50 C follow remarkably straight lines in the Arrhenius plot corresponding to apparent enthalpies of around À0.8 eV. For T101, we get instead a shallow conductivity minimum in region III, suggested to originate from the continued adsorption of H 2 O on O 2c sites, and the sharper increase due to the more continuous molecular physisorbed layer appears only around 100 C. Whether this is ice-like on T101 like on T001 and T100 is uncertain, but it still comes with the high mobility.
The observed negative activation energy for the temperature dependency of proton conductivity in the ice-like physisorbed water in range III have several contributions: the enthalpy of adsorption of the rst physisorbed layer should correspond to something a bit more negative than that of condensation of water (À0.45 eV at RT). But the enthalpy of mobility goes in the opposite direction. To get to À0.8 eV, we must therefore expect that the increasing coverage of the rst physisorbed layer increases the degree of protolysis 31 and decreases the activation enthalpy of mobility. What appears as a negative temperature dependence with an enthalpy of À0.8 eV is actually a continuously decreasing (yet positive) activation enthalpy of conduction with increasing coverage, as pointed out by others before. 43 One may say that the enthalpy of conduction is positive, but the differential enthalpy is negative.
The contribution from entropy with respect to the formation and conduction of protons should also not be neglected in rationalizing the apparent negative enthalpy. As the rst physisorbed water layer gains in coverage, protons can move into it at lower density (concentration), which increases both the vibrational and congurational entropy and drives the water protolysis forward, and, in turn, dissociation to increase the total number of mobile protons.
At temperatures below 50 C (in ambient humidity, with p H 2 O ¼ 0.026 atm), region IV, the conductivities of T001 and T100 increase even more sharply with decreasing temperature, reecting increasingly high vehicular mobility of protons in the thicker and liquid-like physisorbed layers. The mobility is approaching that of bulk water, independent of the characteristics of the oxide surface. Thus, the different conductivities observed on these three facets are now mostly determined by the concentration of protonic species.
Sample T101, which displayed a lower activation energy of proton conductivity in the chemisorbed layer and hence had a higher conductivity than the others at the minima around 150 C, does not exhibit the behavior attributed to the rst physisorbed layer that we saw in T001 and T100. Instead, it appears that it goes more directly from conduction in chemisorbed water to liquid-like physisorbed water layers, in agreement with the conclusions from the FTIR studies. It should be noted that the apparent activation energy of T101 is less negative than that of T001 and T100 at temperatures from 150 to 50 C, during which the rst physisorbed water layer forms. For {001} and {100} facets, water molecules form hydrogen bonds with surface -OH groups, while they are bonded to the molecular water on {101} facets in the chemisorbed water layer, which decreases the magnitude of the enthalpy of the water adsorption/condensation process (makes it less exothermic). The increasing coverage of the rst physisorbed layer with decreasing temperature hence lags behind for T101 as compared to T001 and T100, the apparent enthalpy gets less negative, and the conductivity increases less sharply.
T101 ends up at 1-2 orders of magnitude lower conductivity than T001 and T100 at room temperature with a RH of 80%. Assuming that this reects the concentration of protons, this must, in turn, stem from a lower degree of dissociation and/or protolysis. A lower degree of dissociation is in agreement with what FTIR analyses concluded about molecular water adsorption on the {101} facet. Another possibility lies in lower adsorption energies of water on the {101} facet resulting in lower total thickness and less liquid-like character of water here.
Interestingly, the pre-exponential factors acquired from the tting of conductivities in the chemisorbed water are higher for T001 and T100 than for T101, see Table S2. † This reects only differences in entropy, attempt frequencies, and density of sites. We suggest that it reects that the {101} facet has much fewer active sites for adsorption and dissociation of water.
All in all, we nd that surfaces with dissociative chemisorption exhibit high surface protonic conductivity at the highest and lowest temperatures, due to a high concentration of sites for adsorption and dissociation and unrestricted mobilities for two very different reasons. At intermediate temperatures, however, they exhibit deep minima from the high activation energies for mobility in the chemisorbed and the rigid rst physisorbed layers. Surfaces with less dissociation and hence molecular chemisorption, on the other hand, appear to offer fewer protons (fewer sites for adsorption and dissociation) but cross the temperature range with a much shallower minimum at intermediate temperatures because the activation energies are smaller.
The ndings and conclusions above were supported by H/D isotope effects. The classical model predicts that the proton hopping rate in oxides is proportional to the O-H/O-D vibration frequency, but the model has uncertainties related to sticking probability and other mechanistic aspects. Instead, the isotope effect stems mostly from a higher activation energy of jumps  44 The signicant isotope effect ratio of $2 around 150 C (Fig. 8) shows that Grotthuss free proton jumps is the dominating mechanism. It decreases with increasing temperatures as the effect of the different activation energies diminish 44 and native bulk conductivity becomes signicant. The isotope effect also gradually diminishes at temperatures below 150 C (higher RH) and almost disappears as RH surpassed 60%, indicating a change to vehicular transport in liquid-like water layers.
The T001 sample was synthesized using HF as the capping agent. It is well-known that uoride ions can bond strongly to Ti 5c sites and stabilize the surface, 45 therefore, uoride ions would inevitably adsorb on the sample surface during synthesis. When the surface Ti atoms are terminated by uoride ions, the strong bonding between the Ti-5c and water oxygen (O w -Ti 5c ) is no longer possible. Moreover, the hydrogen bond between the water hydrogen and surface O-2c (H w -O 2c ) also becomes unfavorable due to the repulsive interactions between surface oxygen and uoride ions. In this case, water layers cannot be formed on the surface of TiO 2 at low temperatures, and as a result, the conductivity remained <10 À7 S cm À1 throughout the temperature range of interest ( Fig. 9). However, when the surface absorbed uoride ions were removed by washing with NaOH (aq), 46,47 the observed conductivity was greatly improved.

Conclusions
We systematically varied the dominance of different crystal facets and thereby surface chemistry of nanocrystalline TiO 2 to shed light on their role in water-oxide interactions and proton conduction. In situ FTIR suggests that the coordination of surface atoms on {001} and {100} dominated facets makes water chemisorb strongly and dissociatively, while chemisorption on {101} is weaker and more molecular. In ambient humidity (p H 2 O ¼ 0.026 atm) and below 300 C, surface protonic conduction in the chemisorbed layer becomes dominating, with activation enthalpies in the range of 0.4-0.6 eV for {001} and {100} and 0.3-0.4 eV for {101}. Then, the conductivity of all samples crosses a minimum, which is shallower for the {101} surface because of the low activation energy for mobility in the so molecular water. Below approximately 150 C, the coverage of the rst ice-like physisorbed layer becomes signicant, and its higher proton mobility increases the surface protonic conductivity with decreasing temperature. This sets in at higher temperatures and is steeper for {001} and {100} because physisorption is stronger (more exothermic) on the rigid dissociated than the so molecular chemisorbed water underneath. Further down at 50 C (RH z 30% for ambient humidity) this layer is considered complete and as we approach room temperature and RH surpasses 60%, liquid-like water layers physisorb, and the proton mobility and conductivity increase further. This is more pronounced on {001} and {100} due to more dissociation and thicker water layers, accommodating higher total concentrations of dissociated protons of higher mobility. This is supported by H/D isotope effect measurements showing a transition from Grotthuss H + hopping in the chemisorbed and rst physisorbed layers to vehicular H 3 O + transport in the liquid-like outer layers. The surface chemistry was further  studied by terminating {001} facets with uoride ions, which effectively suppressed surface protonic conduction by blocking H 2 O-Ti 5c interaction and protonation of surface O 2c . This study contributes mechanistic insights into surface protonic conduction governed by adsorbed water layers on specic surfaces of polycrystalline oxides.

Author contributions
XK synthesized the samples, designed and performed the measurements, analyzed the data and wrote the original dra. TA and CN did the TEM measurements and in situ FTIR tests, respectively. XS contributed interpretation of data and revision of the manuscript. AC and TN conceived and supervised the study. All authors reviewed the manuscript.

Conflicts of interest
There are no conicts to declare.