Influence of surface atomic structure demonstrated on oxygen incorporation mechanism at a model perovskite oxide

Perovskite oxide surfaces catalyze oxygen exchange reactions that are crucial for fuel cells, electrolyzers, and thermochemical fuel synthesis. Here, by bridging the gap between surface analysis with atomic resolution and oxygen exchange kinetics measurements, we demonstrate how the exact surface atomic structure can determine the reactivity for oxygen exchange reactions on a model perovskite oxide. Two precisely controlled surface reconstructions with (4 × 1) and (2 × 5) symmetry on 0.5 wt.% Nb-doped SrTiO3(110) were subjected to isotopically labeled oxygen exchange at 450 °C. The oxygen incorporation rate is three times higher on the (4 × 1) surface phase compared to the (2 × 5). Common models of surface reactivity based on the availability of oxygen vacancies or on the ease of electron transfer cannot account for this difference. We propose a structure-driven oxygen exchange mechanism, relying on the flexibility of the surface coordination polyhedra that transform upon dissociation of oxygen molecules.

. Secondary-electron emission spectra measured on a surface bi-crystal prepared as described in the main manuscript, showing both (4 µ 1) and (2 µ 5) reconstructions. Spectra acquired on the two regions are represented in black and red, respectively. The horizontal axis is corrected for the negative bias voltage applied to the sample during measurement (see Supplementary Note 4 for details). Cutoff energies of the secondary electron emission, i.e., work-functions -whose positions are highlighted by vertical dashed lines -were determined by fitting a sigmoidal step function to the main peak in each spectrum. Secondary electrons were collected in normal emission, from an approximately circular region of~0.5 mm diameter, with 3.5 eV pass energy.  Fig. 6 Surface structure-independent band alignment. Core level x-ray photoelectron spectra acquired on the SrTiO 3 (110) surface bi-crystal on which the oxygen exchange experiment -described in the main manuscript -was performed. Spectra acquired on the (4 µ 1)-and (2 µ 5)reconstructed areas are represented in black, and red, respectively. The spectra were acquired with electron lens settings such that electrons are collected at normal emission from an approximately circular area of~1.5 mm diameter, and at 10 eV pass energy. The two sets of spectra are vertically offset for clarity. See Supplementary Note 5 for details.

Effect of exchange duration, and derivation of k* from SIMS data
In order to evaluate the influence of the total exchange time on the amount of incorporated oxygen, and exclude possible saturation effects, a second surface bi-crystal was prepared with a procedure similar to the one described in the main manuscript. Subsequently, this surface bi-crystal was annealed in 16 O 2 (16 h, 450 • C, 0.1 mbar), and 18 O 2 (4 h, 450 • C, 0.1 mbar), following a procedure analogous to the one described in the main text. In this case, however, the preparation of the surface bi-crystal -first with a uniform (4 × 1) surface structure and then with the additional (2 × 5) reconstruction -was carried out by pulsed laser deposition (248 nm-KrF excimer laser; spot-size: 0.7 × 1.7 mm 2 ; fluence: 2.5 J cm −2 ; repetition frequency: 1 Hz; target-substrate distance 55 mm) 1 , from a single-crystalline rutile TiO 2 target (pre-ablated before each deposition at the growth parameters), in O 2 background (5 × 10 −6 mbar). A total of 27 laser pulses were deposited at 600 • C onto the uniform (4 × 1)-reconstructed SrTiO 3 (110) surface -while shading half of the sample with a dedicated mask -to obtain a well-ordered SrTiO 3 (110)-(2 × 5) surface structure. Comparison of Supplementary Fig. 2a,b and Fig. 2a,b (main manuscript) shows that this PLD-based preparation method yields analogous results for both the SrTiO 3 (110)-(4 × 1) and -(2 × 5) surface structures.
In particular, similar morphologies (main panels of Fig. 2a,b, and Supplementary Fig. 2a,b) and structures (insets of Fig. 2a,b and Supplementary Fig. 2a,b) are obtained by either employing epitaxial growth of Ti (Fig. 2a,b -main text), and pulsed laser deposition of TiO 2 ( Supplementary  Fig. 2a,b). Supplementary Fig. 2c,d shows the corresponding samples in Supplementary Fig. 2a,b after the annealing treatment described above (i.e., 20 h at 450 • C, 0.1 mbar O 2 ). Comparison of large-and small-scale STM images shows that morphology and atomic-scale structure are respectively retained even after such a harsh treatment. Minimal morphology changes, comparable to those described in the main text, occur only for the SrTiO 3 (110)-(2 × 5) structure, which appears somewhat rougher after annealing, with the appearance of a few, single-atomic-layer-high islands.
It should be stressed that, while STM imaging allows to confirm that the SrTiO 3 (110)-(4 × 1) and -(2 × 5) structures are stable at the atomic and mesoscopic scales, an analogous conclusion can be extended to the macroscopic scale by analysis of the LEED patterns of the samples (bottom-right insets of Supplementary Fig. 2). In particular, several LEED patterns on the whole surface of the sample (not shown here) have been acquired by macroscopically moving the sample in the electron beam. All LEED images acquired on the two parts of the surface bi-crystal showed analogous structures at each step of the preparation. Several SIMS depth profiles ( Supplementary Fig. 3) were measured on such a surface bi-crystal after the 18 O isotope exchange treatment described above, i.e., 16 h anneal in 16 O 2 , followed by 4 h anneal in 18 O 2 . The first measurement point is excluded. As in the main text, the surface exchange coefficient, k*, was determined from where M represents the area-specific total amount of incorporated tracer In Supplementary Eq. 3 c 18 O (z) is the volume concentration of exchanged tracer, and A the crosssectional area measured by SIMS, while z represents the depth coordinate. Since each SIMS measurement has been performed on uniformly-terminated regions of the SrTiO 3 bi-crystal, the volume integral in Supplementary Eq. 3 can be reduced to an integration along the depth coordinate, yielding .
(Supplementary Eq. 5) As for the profiles presented in the main text, direct fitting of the depth profiles measured after 4 h 18 O exchange did not yield reliable results due to their extremely shallow extension of only a few nanometers. In such a depth range, SIMS-related profile broadening effects cannot be neglected, as discussed in Supplementary Note 3 below.

Experimental broadening of SIMS profiles
Broadening of depth profiles occurs in SIMS due to atomic mixing as a consequence of the sputter ions used for depth profiling (Cs + ions, 2 kV, ∼ 100 nA in the present study), and of the impact and implantation of the primary ions (Bi ++ 3 clusters, 25 kV, ∼ 0.02 pA) 2,3 . As a result, the isotope composition attributed to deeper atomic layers is affected by the composition of the higher-lying ones. Such an effect is of minor importance when isotope profiles with diffusion lengths larger than 10 nm are investigated. However, the SIMS results reported in Fig. 3b (main manuscript) and Supplementary Fig. 3 show depth profiles rapidly decaying towards the natural abundance of 18 O within a few nanometers.
In order to address if these 18 O depth profiles are effectively representative of diffusion profiles, or if they are rather dominated by broadening effects, Supplementary Fig. 4 compares the shape of the decay of the 18 O concentration on (4 × 1)-and (2 × 5)-reconstructed areas. In the former case, given that (i) both reconstructed areas belong to the very same bulk crystal, and (ii) we measure no difference in band-bending/surface potential by XPS -meaning that the whole sample is characterized by a uniform 'bulk' diffusion constant -the difference in k* values between the two surface structures should lead to a different decay length, and, as a result, to different profile shapes. Conversely, in case readily-decaying profiles are dominated by SIMS broadening effects, the very same shape as a function of depth should be observed. As can be observed in Supplementary Fig. 4a, the rescaled 18 O concentration profiles measured on (4 × 1)-and (2 × 5)-reconstructed areas after short annealing times (1 h) are essentially superimposed, indicating that such profiles are dominated by SIMS broadening effects. Upon longer annealing periods (4 h, Supplementary Fig. 4b) the decay lengths of the 18 O concentration profiles on (4 × 1) and (2 × 5) are still similar, but no longer virtually identical. The visible difference arises in particular from some broadening of the profile in the (4 × 1)-reconstructed area. Therefore, we conclude that the decay of these latter depth profiles at least partly originates from true tracer diffusion dynamics.
However, a strong indication that SIMS-related broadening effects still significantly contribute to the data in Supplementary Fig. 4b comes from the comparison of the scaled profiles measured on (2 × 5)-reconstructed regions ( Supplementary Fig. 4c), whose shape appears essentially unaffected by the four-fold increase in exchange duration. This observation strengthens the conclusion that on SrTiO 3 (110)-(2 × 5) areas oxygen exchange involves exclusively very near-surface layers, on a length scale shorter than the depth resolution of SIMS, consistently with the findings of DFT calculations (see main manuscript, and Supplementary Notes 7 and 8). For this reason, and since comparable decay lengths are observed for all depth profiles in Supplementary Fig. 4, SIMS-related broadening of the 18 O profiles are important also for long annealing times ( Supplementary Fig. 4b). As a result, fitting such profiles -as it is customarily done in these cases -using Fick's diffusion law does not yield satisfactory results, and provides unreliable values for k* and D* coefficients. Therefore, only k* values have been extracted from the total 18 O amount incorporated, as described in Supplementary Note 2.

Work function of SrTiO 3 (110)-(4 × 1) and -(2 × 5) surfaces
In order to assess any difference in work function between (4 × 1)-, and (2 × 5)-reconstructed SrTiO 3 (110) surfaces, x-ray-excited (Al Kα, hν = 1486.61 eV) secondary electron emission spectra were measured at a surface bi-crystal ( Supplementary Fig. 5). The bi-crystal was prepared as detailed in the main manuscript. It should be stressed that this approach allows to unequivocally measure work function differences solely arising from the two SrTiO 3 (110) reconstructions, independently of any extrinsic influence.
We measured the secondary electron cutoff energy, which can be straightforwardly related to the sample work function 4 . The electron emission spectra were acquired at normal emission, by negatively polarizing the sample with a 9 V (nominal) battery. Such a precaution prevents unwanted cutoffs due to the detection system, conferring the electrons a minimum kinetic energy higher than any analyzer work function (4-5 eV). To compensate for this bias voltage, and therefore allowing to determine absolute values of work functions rather than mere differences, O 1s core-level spectra were acquired on both regions, with the sample held at both ground and at the polarization potentials. This allows for a correction of the horizontal energy axis in Supplementary Fig. 5. In order to minimize electric-field-lines distortion, which can significantly affect work function measurements 4 , the x-ray source was significantly retracted from the surface (approximately 50 mm). All measurements were performed overnight, to limit the effect of any stray, time-varying magnetic fields. Subsequent spectra acquired on both regions show a deviation in the measured cutoff smaller than the standard error of the fitted position.
The secondary electron cutoff was evaluated by fitting a sigmoidal step function to the main, lowest-kinetic energy peak in each spectrum. The cutoff positions were evaluated as the energies corresponding to 5% of the step amplitude, resulting in work functions of (4.051 ± 0.010) eV and (4.470 ± 0.003) eV for SrTiO 3 (110)-(2 × 5) and -(4 × 1) surfaces, respectively. Uncertainties derive from the combined standard errors on the fitting parameters of the sigmoidal step, and the position of the O 1s core levels.
Additional features in the spectra of Supplementary Fig. 5 are possibly related to low-energy Auger emission, and were not present on secondary emission spectra measured on the sample plate. Their characterization is beyond the scope of the present manuscript.

XPS core level positions of (4 × 1)-and (2 × 5)-SrTiO 3 (110)
Supplementary Fig. 6 reports core-level x-ray photoelectron spectra (Al Kα, hν = 1486.61 eV) acquired on the SrTiO 3 (110) surface bi-crystal described in the main manuscript. Clearly the features corresponding to electron emission from O 1s, Ti 2p, and Sr 3d are all found at the same binding energy (within the experimental uncertainty of ±0.03 eV). This indicates that no difference exists between the surface potentials of (4 × 1)-(black), and (2 × 5)-reconstructed (red) SrTiO 3 (110) surfaces. Once more it should be stressed that the present measurement, performed on the very same bulk sample on which both surface structures are present at the same time, allows to unequivocally disentangle the structure-related surface potential from extrinsic effects deriving from, e.g., the bulk of the single crystalline sample, or the detailed configuration of the electric contact with the measuring set-up.

Supplementary Note 6 V O concentration estimates
As detailed below, according to defect chemical models in the literature 5-7 , at our experimental conditions the nominal equilibrium concentration of V O s in 0.5 wt.% Nb-doped SrTiO 3 is approximately 6 × 10 −7 cm −3 , or about 10 −29 with respect to oxygen sites. However, it is possible that this thermodynamic lower limit of oxygen vacancies is not reached within our experiments. Therefore, here we also determine an upper limit of oxygen vacancy concentration by using an estimate of the decay length of the tracer profiles.
Electron-conducting donor-doped SrTiO 3 may undergo a re-oxidation process when exposed to high temperatures 6 , accompanied by a complex time dependence of the corresponding defect concentration profiles due to the changing band bending during re-oxidation. SrO evolves at the surface, and at equilibrium Sr vacancies largely compensate the donor dopants. Since our samples were annealed at about 1000 • C in 3 × 10 −9 bar O 2 for 1 h, such a re-oxidation may have taken place also in our case, at least in the relevant region close to the surface. With the database of Ref. 6 we can estimate the resulting equilibrium defect concentrations for the given pressure and temperature: An oxygen vacancy concentration of about 5 × 10 14 cm −3 and an electron concentration of about 5 × 10 18 cm −3 result. Hence, a remaining effective donor concentration of 5 × 10 18 cm −3 is found, corresponding to those Nb donors still uncompensated by Sr vacancies.
This cation defect chemical state is 'frozen' while cooling the sample after annealing. Possibly some minor cation changes take place also during cooling, and the frozen-in temperature is thus somewhat lower than 1000 • C; this may further reduce the effective donor concentration very close to the surface, but the following conclusion is not affected by this uncertainty: Before the diffusion experiment, the SrTiO 3 sample was exposed to 0.1 mbar oxygen at 450 • C for about 4 h and the oxygen stoichiometry equilibrates according to these conditions. In contrast to annealing at 1000 • C, however, all cations -and thus also Sr vacancies -can be considered immobile at this temperature 6 . As a consequence, the equilibration takes place without changing the effective donor concentration of about 5 × 10 18 cm −3 . From the defect chemical data in the literature 6 we then get a nominal oxygen vacancy concentration of 6 × 10 −7 cm −3 , i.e., virtually all oxygen vacancies that were formed at 1000 • C should become refilled at 450 • C during oxygen exposure. This is also in accordance with our experimental observation that no oxygen vacancies are observed on the SrTiO 3 surfaces by STM.
However, as discussed in Supplementary Note 3 the tracer diffusion profile of the (4 × 1)-reconstructed SrTiO 3 (110) surface seems to broaden to a small extent after 4 h diffusion time, compared to the 1 h exchange period, i.e., at least some oxygen vacancies seem to exist. An upper limit of this concentration can be determined from an estimated decay length L D in the profile of about 2 nm. With L D ≈ √ D*t we find D* = 3 × 10 18 cm 2 s −1 . Using data 6 of the vacancy diffusion coefficient D V we get from D* = 0.69 D V C V an upper limit of the vacancy concentration C V of about 6 × 10 12 cm −3 , or a vacancy fraction with respect to oxygen sites of 10 −10 (possibly with a spatial variation within the diffusion zone).

DFT formation energy for V O s at (4 × 1) and (2 × 5) surfaces
Several vacancy sites were inspected in the surface (S), subsurface (S-1) and interface region (I) between the reconstructed overlayer and the bulk-like SrTiO 3 substrate, as indicated in Supplementary  Fig. 7. The results, collected in Supplementary Table 1, reveal that the formation of V O in the (2 × 5) surface layer is by 2.2 eV more favorable than at the very surface of the (4 × 1) reconstruction. We note that the DFT slab effectively models the (2 × 4) member of the (2 × n) family; to avoid any confusion with the experimental results we adopt the notation (2 × 5). As already reported in Ref. 8 we find that the energetically more advantageous V O site in the (4 × 1) is in the interface layer.

Supplementary Note 8 O 2 adsorption and dissociation: first principles molecular dynamics
The first principles molecular dynamics (FPMD) runs delivered several final structures for the process describing the adsorption and dissociation of O 2 on both types of surface reconstructions. First, in our MD runs we did not start from a 'true' gas phase (i.e., O 2 many ångstroms above the surface), as this would have required a much larger vacuum region and much longer MD runs. Considering the large size of our supercells, this would have been computationally too costly. To model the adsorption and dissociation process we have performed several MD runs starting from different initial configurations with the O 2 molecule located at different positions/orientations at about 2 Å above the surface. Depending on the initial configuration, we had cases in which the O 2 landed on the surface and dissociated, and cases in which the O 2 desorbed back in the vacuum region. The structures involving both adsorption and dissociation turned out to be by far the energetically most stable configurations. In the main text, we show the most representative ones. In Supplementary  Fig. 8 we have collected a few other models representing (a) dissociation of O 2 on the large hollow area of the (4 × 1) surface (E ads = −2.94 eV), (b) O 2 dissociation on the defect-free (2 × 5) surface along the Sr-chain (E ads = −0.87 eV), and (c) the unfavorable (positive formation energy) situation of O 2 adsorption in the middle of the (2 × 5) surface with one V O . Supplementary Table 2 summarizes all the DFT-calculated O 2 adsorption energies for the configurations reported in the main text and in this Supplementary Information.

Supplementary Note 9 O 2 dissociation paths and barriers calculated by nudged elastic band (NEB)
Supplementary Figure 9 shows the NEB paths discussed in the main text, and reports the initial, transition and final states, the corresponding activation energy (i.e., the one at the transition state), and the overall energy gain in the final state. unsaturated polyhedra with respect to the octahedral environment is also explained by the larger degree of asymmetry of the unsaturated clusters in terms of the A-O-A bond-angles, which are significantly altered in the unsaturated clusters as compared to the 90°configuration in the AO 6 phase. In AO 5 the out-of-plane A-O-A angle is reduced to 85°, in AO 4 the A-O-A angles are increased to 109°, and in AO 3 the three A-O-A angles measure 120°. The average Ti-O bond length, on the other hand, decreases monotonically from 1.89 Å in TiO 6 to 1.78 Å in the TiO 3 cluster (qualitatively similar trends are obtained for VO x and CrO x clusters).

Surface phase diagram with defects or adsorbates
The final defect structures obtained by combining chemical intuition and extensive molecular dynamic runs should represent stable phases. However, a phase diagram of the most stable phases would help the interpretation of the results and its connection with the observations. Following the standard recipe and prescriptions of ab-initio thermodynamics 11-13 (described in detail also in Ref. 14), which we have previously used for SrTiO 3 , 14 we have constructed the surface phase diagram as a function of the O and Ti chemical potentials for the two types of SrTiO 3 reconstructions under scrutiny, specifically the (4 × 1) and (2 × 5), considering the clean surfaces and the defected ones (with one O 2 , and with one O 2 +V O ). The results shown in Supplementary Fig. 11 clearly indicate that the five phases show areas of stability for specific ranges of the Ti and O chemical potentials. The only phase that does not appear in the phase diagram is (2 × 5) + (O 2 +V O ), i.e., the less stable one according to our results reported in Fig. 4 of the main text. Nonetheless, by inspecting the relative stability of the (2 × 5) + (O 2 ) and (2 × 5) + (O 2 +V O ) phases at Δμ Ti = −4 eV and −1 > Δμ O > 0 -corresponding to the region of stability of (2 × 5) + (O 2 ) -we find that these two phases are very close in energy, about 30 meV Å −2 , as shown in Supplementary Fig. 12.