The influence of bulk stoichiometry on near-ambient pressure reactivity of bare and Pt-loaded rutile TiO2(110)

The interaction of catalyst particles with reducible support materials can drastically change their reactivity. On rutile TiO2, processes like particle encapsulation (caused by the “strong metal–support interaction”, SMSI) have long been known to depend on the initial reduction state of the oxide. Despite this knowledge, sample stoichiometry has rarely been controlled in a reproducible manner in the surface science literature. Here, we use scanning tunnelling microscopy (STM) to explore systematically how near-ambient pressures (0.1–1.0 mbar) of O2, H2, CO and CO2 affect blank and Pt-loaded rutile TiO2(110) surfaces of different bulk stoichiometry at 600 K. To this end, we present preparation recipes that result in a sample stoichiometry always converging back to the same value, which allows us to use the same samples with constant reduction state over hundreds of preparation cycles. Comparing a highly reduced and a near-stoichiometric TiO2 sample, we find that surface reactivity to all four gasses differs even without Pt loading. Most surprisingly, we find that the highly reduced TiO2(110) is oxidized by CO2, but this reaction is completely inhibited on the near-stoichiometric sample. Pt nanoparticles, in turn, become encapsulated after vacuum annealing on the reduced, but not on the near-stoichiometric sample. Encapsulation on the near-stoichiometric sample is achieved only after exposing it to 0.1 mbar H2 at 600 K. Interestingly, we also see a further modification of the already encapsulated particles on the reduced sample under the same conditions, such that they become embedded deeper in the TiO2(110) surface.


Supplementary Figures
) as a func on of the ac va on barrier for bulk diffusion, shown for relevant temperatures and mes.The inters al is assumed to follow a one-dimensional random walk, resul ng in a normal distribu on with σ equivalent to the root mean square distance from the original posi on.(b) Simulated concentra on profiles as shown in Figure 5 for a 2 mm thick ru le TiO2(110) crystal aGer a given number of cycles of spu ering and annealing in O2 for 20 minutes at 900 K.In contrast to Figure 5, here, the bulk diffusion barrier was set to 0.5 eV.The surface reac on barrier determining the rate of reoxida on was kept at 1 eV.Solid and dashed lines show equilibra on when star ng from a fully stoichiometric and from a homogeneously reduced ini al state, respec vely.(c) Same simula on as in (b), adding a 10 minute annealing step at 1100 K in each cycle.No further oxida on or reduc on is assumed during this higher-temperature step.

Diffusion simula ons
As discussed in the main manuscript, diffusion simula ons of Tiint were based on a one-dimensional random walk.This is a reasonable approxima on when the diffusing par cles are dilute enough that interac on between them is negligible.We further model the occupa on of each layer as a floa ngpoint concentra on, rather than an integer number of diffusing par cles.For each single Tiint, the probability of finding it at a distance z from its original posi on is given by a normal distribu on, with ∆ √ aGer n steps, where n depends on the elapsed me, the diffusion barrier and temperature, and a preexponen al factor (see main manuscript).We can then directly evaluate the concentra on profile of a sample aGer a given annealing step by convolu on of the ini al concentra on per layer with a normal distribu on, se[ng the standard devia on σ to reflect annealing me and temperature.This effec vely smears out each "par cle" in the original concentra on profile to reflect its likely posi on aGer annealing.Crucially, the result is exactly the same no ma er if the concentra on profile samples each atomic layer individually, or only every k th layer, as long as σ is chosen according to the actual layer thickness.This treatment is therefore extremely computa onally efficient, as only one calcula on is required for each annealing step.
Edges of the sample must be accounted for specifically.The simplest boundary condi on to implement is that when a par cle at the surface (the 0 th layer) would diffuse out of the surface (to the −1 st layer), it is instead considered to s ll be in the 0 th layer.This is easily achieved by performing the convolu on with the normal distribu on, then "folding back" the nega ve space, such that all concentra on in the −1 st layer is added to the 0 th layer, all concentra on in the −2 nd layer is added to the 1 st layer, and so on.It is easy to see that this is s ll an exact solu on, as any diffusion event from the −1 st layer is treated the same way as a diffusion event from the 0 th layer, with diffusion in one direc on having no effect, and diffusion in the other direc on leading away from the surface.Applying the same approach to the other edge of the crystal, we essen ally obtain periodic boundary condi ons, where the concentra on is flipped in every other period.Again, this is s ll an exact solu on within the bounds of the random walk approxima on, no ma er the point sampling density.
Modelling oxida on at the surface is more difficult.To avoid having to model a varying thickness of the sample, we approximate oxida on by some probability that every me a par cle would diffuse out of the surface, it disappears instead of staying in the surface layer.It is trivial to choose this probability to correspond to some surface reac on barrier by se[ng it to a Boltzmann factor , with ε being the difference between the bulk and surface barriers.
Ideally, we would implement this oxida on process in our model by applying that probability every me a par cle passes through the surface, i.e. from the 0 th to the −1 st layer or vice versa.However, the approach of simply convolving a normal distribu on then breaks down, as e.g.most atoms at the 0 th layer that would remain at the 0 th layer in the random walk approxima on have actually passed through the origin at least once, and likely many mes (assuming large n).This can be solved either by calcula ng the contribu ons of different paths to each point of the normal distribu on and applying the loss probability accordingly, or by choosing small me steps, such that few atoms diffuse out of the surface in each step.Both approaches are computa onally expensive.We have chosen the second, applying the loss factor twice in each me step to the concentra on in the out-of-surface space to account for the symmetrical nature of the problem (i.e., for each par cle found in the −1 st layer, one par cle in the 0 th layer is considered to have come from the −1 st layer for an arbitrary ini al distribu on).This qualita vely captures the oxida on behaviour, especially in the limits of no oxida on (where the model is exact) or full oxida on (where every atom diffusing out of the surface is lost).However, for arbitrary surface reac on barriers, we accept some error in capturing the exact value of the barrier, because we do not correctly capture par cles passing through the surface mul ple mes.The concentra on models shown in Figure 5 and Figure S14 are s ll qualita vely correct within the limits of the approxima on, but we only give concentra ons as "arbitrary units" to reflect this error.Similarly, we model the reducing effect of spu ering simply by se[ng the concentra on at the surface layer to an arbitrary (high) value, since we have no good es mate of how much excess Ti is introduced in each spu ering step.

Figure S2 .
Figure S2.Addi onal STM images aGer NAP O2 exposure of LR-TiO2, from the same experiment as the image shown in Figure 2(a).(a) LR-TiO2 aGer exposure to 0.1 mbar O2 at 600 K for 15 minutes, and (b,c) aGer post-annealing at 800 K for 10 minutes in UHV.Images were acquired at RT in UHV, with scanning parameters Usample and Itunnel (a) 1.7 V, 0.1 nA, (b) 1.7 V, 0.2 nA and (c) 1.9 V, 0.1 nA.

Figure
Figure S3.LEED (70 eV incident electron energy) of HR-TiO2 (a) directly aGer exposure to 0.1 mbar O2 at 600 K for 15 minutes [also shown as the inset in Figure 2(b)], and (b) aGer post-annealing at 1100 K for 10 minutes in UHV, corresponding to the STM image shown in Figure 2(b).

Figure S4 .
Figure S4.Addi onal STM images from the NAP CO2 experiment on HR-TiO2 shown in Figure 2(d).(a) HR-TiO2 in 1 mbar CO2, image acquired in gas atmosphere 10 minutes aGer reaching 600 K. (b) AGer exposure to 1 mbar CO2 at 600 K for 30 minutes, image acquired at RT in UHV, same measurement as the image shown in Figure 2(d).(c) AGer post-annealing at 800 K for 10 minutes in UHV.Scanning parameters Usample and Itunnel were (a) 1.8 V, 0.2 nA, (b) 2.1 V, 0.2 nA and (c) 1.7 V, 3.3 nA.

Figure S5 .
Figure S5.Addi onal STM images corresponding to the NAP H2 experiment shown in Figure 2(f), showing the HR-TiO2 sample post-annealed in UHV aGer exposure to 1 mbar H2 at 600 K for 105 minutes.The sample was annealed at (a) 773 K for 10 minutes, (b) 973 K for 10 minutes, and (c) 1100 K for 20 minutes.Images were acquired at RT in UHV, with scanning parameters Usample and Itunnel (a,b) 1.2 V, 0.1 nA and (c) 1.2 V, 0.2 nA.

Figure S8 .
Figure S8.STM images showing the evolu on of Pt nanopar cle on LR-TiO2 in H2. (a,d) As-sintered nanopar cles in UHV, annealed 30 minutes at 1000 K. (b,e) The same par cles in 0.1 mbar H2 at (e) 620 K, ≈15 minutes aGer the temperature reached 600 K and (b) 621 K, aGer ≈21 minutes.(c,f) Images acquired aGer cooling to room temperature and returning the sample to UHV aGer a total of 90 minutes at T ≥ 600 K in H2.Some internal structure is resolved on some of the par cles, but we did not find any well-defined superstructure.Scanning parameters Usample and Itunnel were (a) 1.4 V, 2.0 nA, (b) 1.4 V, 1.5 nA, (c) 1.2 V, 0.3 nA, (d) 1.4 V, 0.2 nA, (e) 1.4 V, 1.4 nA, (f) 1.2 V, 0.8 nA.

Figure S9 .
Figure S9.STM images of the Pt/LR-TiO2 and Pt/HR-TiO2 samples used in the C 18 O TPD experiments shown in Figure 4(a).(a) Pt par cles on HR-TiO2 aGer annealing in UHV at 1000 K for 15 minutes.(b) Pt par cles on LR-TiO2 aGer annealing in UHV at 1000 K for 15 minutes, and (c) aGer annealing in UHV at 1100 K for 75 minutes.Slight linear distor ons in (b) and (c) are due to thermal driG of the STM scanner.Scanning parameters Usample and Itunnel were (a) 1.5 V, 1.0 nA and (b, c) 2.0 V, 0.3 nA.

Figure
Figure S10.XPS (Al Kα, normal emission, 50 eV pass energy) of the Pt/LR-TiO2 and Pt/HR-TiO2 samples used in the C 18 O TPD experiments shown in Figure4(a).Dashed lines show spectra acquired directly aGer deposi ng Pt.Solid lines show spectra acquired aGer annealing at 1000 K for 15 minutes.Note that the signal in the Ti 3s region is convoluted with duplicates of the Pt 4f peaks due to X-ray satellites from the non-monochroma c Al Kα source (α3: ΔE = 9.8 eV, α4: ΔE = 11.8 eV rela ve to α1,2, with rela ve intensi es of 6.4% and 3.2%, respec vely).1

Figure
Figure S12.XPS (monochromated Al Kα, normal emission, 30 eV pass energy) of Pt/LR-TiO2 and Pt/HR-TiO2 corresponding to the LEIS data shown in Figure 4(b).Dashed lines show spectra acquired directly aGer deposi ng Pt.Solid lines show spectra acquired aGer annealing at 1000 K for 30 minutes.The dot-dashed, dark blue line was taken aGer exposing the LR-TiO2 to 0.1 mbar H2 and hea ng to 600 K for 30 minutes.Note the more pronounced difference between as-deposited and sintered Pt than seen in Figure S10, possibly indica ng lower Pt loading in the TPD experiment.

Figure
Figure S13.NAP-XPS (monochromated Al Kα, normal emission, 30 eV pass energy) of the (a) O 1s, (b) Ti 2p and (c) Pt 4f / Ti 3s regions of Pt/LR-TiO2 exposed to 0.1 mbar H2 at 600 K.The spectra before (black) and aGer (orange) H2 exposure were acquired in UHV at room temperature, while the blue curve was acquired in 0.1 mbar H2 at 600 K.The inset in panel (b) is a magnified view of the Ti 2p3/2 peak, indica ng a slight increase in the Ti 3+ component aGer H2 exposure.The black and orange curves in panel (c) are the same as shown in Figure S12 and correspond to the LEIS data shown in Figure 4(b).

Figure
Figure S14.(a) Standard devia on σ for the posi on of a Tiint inters al diffusing perpendicular to the (110) surface in the bulk of ru le TiO2(110) as a func on of the ac va on barrier for bulk diffusion, shown for relevant temperatures and mes.The inters al is assumed to follow a one-dimensional random walk, resul ng in a normal distribu on with σ equivalent to the root mean square distance from the original posi on.(b) Simulated concentra on profiles as shown in Figure5for a 2 mm thick ru le TiO2(110) crystal aGer a given number of cycles of spu ering and annealing in O2 for 20 minutes at 900 K.In contrast to Figure5, here, the bulk diffusion barrier was set to 0.5 eV.The surface reac on barrier determining the rate of reoxida on was kept at 1 eV.Solid and dashed lines show equilibra on when star ng from a fully stoichiometric and from a homogeneously reduced ini al state, respec vely.(c) Same simula on as in (b), adding a 10 minute annealing step at 1100 K in each cycle.No further oxida on or reduc on is assumed during this higher-temperature step.