Light-induced agglomeration of single-atom platinum in photocatalysis

With recent advances in the field of single-atoms (SAs) used in photocatalysis, an unprecedented performance of atomically dispersed co-catalysts has been achieved. However, the stability and agglomeration of SA co-catalysts on the semiconductor surface may represent a critical issue in potential applications. Here we describe the photoinduced destabilization of Pt SAs on the benchmark photocatalyst, TiO 2 . In aqueous solutions within illumination timescales ranging from few minutes to several hours, light-induced agglomeration of Pt SAs to ensembles (dimers, multimers) and finally nanoparticles takes place. The kinetics critically depends on the presence of sacrificial hole scavengers and the used light intensity. DFT calculations attribute the light induced destabilization of the SA Pt species to binding of surface-coordinated Pt with solution-hydrogen (adsorbed H atoms), which consequently weakens the Pt SA bonding to the TiO 2 surface. Despite the gradual aggregation of Pt SAs into surface clusters and their overall reduction to metallic state, which involves >90 % of Pt SAs, the overall photocatalytic H 2 evolution remains virtually unaffected.


Introduction
In recent years, highly dispersed speciesmost commonly, in the form of supported single metal atomshave been introduced to the catalysis field with great success and high impact.
[29][30] Photocatalytic reactions are based on a reaction sequence of i) the generation of photoexcited charge carriers (electron-hole pairs) in a semiconductor, ii) charge carrier separation and migration to the semiconductor surface, and iii) the reaction of the excited electrons and holes at the semiconductor surface with reactants in the environment. [31,32]A particularly important example of such a photocatalytic reaction is solar hydrogen production that can be accomplished via transfer of conduction-band electrons from a suitable semiconductor to water that then is reduced to form H2. [33,34] While electron transfer and water reduction for many semiconductors is thermodynamically feasible, H2 generation directly on the semiconductor surface often occurs very sluggishly.To overcome this slow kinetics, the H2 evolution reaction is often co-catalyzed using noble metals such as Pt, Pd, or Rh that are deposited on the semiconductor in form of nanoparticles. [33,35]The often strong acceleration of H2 generation provided by such co-catalysts of a few nanometers in size is ascribed to a combined effect of i) a facilitated electron-hole separation due to the Schottky-junction at the metal/semiconductor interface and ii) a facilitated recombination of 2 H 0  H2.To deposit such nanoscale noble metal particles, various techniques have been developed, most commonly based on noble metal salt precursors that are impregnated on the semiconductor surface and then are reduced chemically, thermally or photo-induced to metallic particles. [36,37][40][41][42][43][44] To date, several studies have suggested the ideal size of the co-catalytic sites (e.g. 1 nm for Pt [45] , or in contrast 0.2 nm for PtO [46] ).In spite of such controversial reports, maximizing the surface to volume ratio of a co-catalytic site is most desirable not only from an efficiency-(per site or per atom) but also from an economic perspective.The maximum surface to volume ratio is achieved when the dimensions of a co-catalytic entity has reached the size of a single atom (SA).Nevertheless, as pointed out several times, for SAs, downscaling laws may no longer hold. [47]Unusual reaction pathways may become favored that may be disadvantageous or advantageous. [30,48]cordingly, a key issue when dealing with SAs is their stability on a chosen substrate and agglomeration is conceived to be of a paramount importance for the applicability and performance of SA catalysts. [49,50]"Stable" refers in a classic catalyst setting mainly to thermal stability, i.e. suppressing surface mobility of SAs at elevated temperatures.Many approaches to stabilizing SAs focus on a classic strong electrostatic adsorption [51] or on substrate lattice defects such as kinks, atomic steps, or often more ideally on ionic vacancies in combination with favorable surface chemistries. [1,49,50]Reliable SA trapping is regarded as a key to ensuring their stable location and catalytic reactivity, which may otherwise be disrupted by SA agglomeration at elevated temperatures. [52]In contrast to such classic thermally activated processes, in photocatalysis the activation takes place by photoinduced charge carriers but only very few SA-studies in photocatalysis report on the observation of photoinduced agglomeration of SAs to NPs. [43,53,54] Such conversion of SAs to NPs may affect the photocatalytic performance by altering the co-catalyst surface distribution, surface-to-volume ratio, and energetics of surface reactions.Correspondingly, a clear difference in the H2 production rates by SAs and NPs has been shown in [55] .Therefore we believe that disregarding the photoinduced mobility and agglomeration of SAs may potentially lead to erroneous evaluations of their "true" photocatalytic activity, e.g.due to the presence of "unstable" (excessive) SA sites that do not contribute significantly to the H2 evolution, in line with previous findings [56] .

Results and discussion
8]60,62] The set-up consists of a 7 nm thin anatase layer that was sputter-deposited on an 8 nm thick SiO2 window microfabricated in a Si wafer (Figure 1a, b; Figure S1).The thinness of the sputter-deposited layer not only enables the conversion of the layer to the anatase polymorph of TiO2, but more importantly provides a reliable substrate that allows high-quality imaging in an aberration-corrected high-angle annular dark-field imaging (HAADF) STEM and even moreit allows to monitor changes in the SA configuration at identical locations [68,69] after illumination intervals.These investigations are supplemented by investigations on single crystalline facetted anatase crystallites (nanosheets) (characterization is provided in the supporting information,

Figure S2
).To elucidate critical factors in SA stability and light induced agglomeration under photocatalytic H2 generation conditions, all samples (thin films and nanosheets) were decorated with Pt SAs using a "dark-deposition" approach previously studied in detail [55][56][57][58]60,62] and shortly described in the experimental section. This ws followed by hydrogen evolution experiments carried out under UV (λ = 365 nm) LED illumination in different aqueous solutions.min (e) and 24 h (f) of irradiation (600 mW cm -2 at λ = 365 nm) in 50 vol.%MeOH (the insets show corresponding SEM images).(g-i) HAADF-STEM images of TiO2 layers after 0 min (g), 1 h (h) and 24 h (i) of irradiation (65 mW cm -2 at λ = 365 nm) in pure water (the insets show Figure 1c shows the photocatalytic hydrogen evolution rates obtained from the titania anatase flakes (Pt SA-loaded in 2 mM H2PtCl6) in pure H2O and in presence of a sacrificial hole capture agent (MeOH).In both cases, a linear evolution of H2 over time is observed.The surface configuration of the Pt SAs during such an illumination process is illustrated in HAADF-STEM images of TiO2 membranes (Pt SA-loaded in 2 mM H2PtCl6) before, after 1 min, and after 24 h illumination (Figure 1d-f), respectively.Without illumination, HAADF-STEM shows uniformly dispersed Pt atoms (either as single entity or in small ensembles).A statistical evaluation of such images via a machine learning based segmentation (an example is provided in Figure S3) of an average area of 5660 nm² shows that SAs are present at an average density of 4.1 × 10 5 µm -2 .In addition, three other species are identified: i) dimers, ii) small clusters classified by an area <1 nm², and iii) SA clusters with an area >1 nm². I this context, less than 13% of the Pt is agglomerated to larger clusters with an average area of ~2.3 nm² corresponding to an equivalent diameter of a circle de of 1.7 nm.Evidently, already after 1 min illumination (Figure 1e), a significant rearrangement of the surface co-catalyst can be observed.Namely, only 11% of the Pt is either in the SA or dimer form and 10% in small clusters, whereas already 79% are contained in clusters with an average size of ~3.8 nm² (de=2.2nm).Finally, after 24 h illumination, the STEM visible agglomerate area increases to an average size of up to ~4-5.5 nm (de=2.6 nm).Notably, the fraction of SAs still takes nearly 7% of the total Pt amount.The complete statistical analysis is graphically shown in Figure 1j  Loading with a much lower precursor concentration (0.001 mM) leads to a SA density of only 0.2 × 10 5 µm -2 , which, however, are all in a SA state (see Figure S5).As observed for the high concentration sample, a large fraction of these SAs agglomerate within 1 min of illumination, whereas even after 24 h still 6% of the Pt atoms are in a SA state, which means that concentration independent nearly the same relative amount of Pt is in the SA state and agglomerated, respectively.
For the SA decorated surface, obviously the feature size is below the resolution of SEM.Also in this case clusters of a crystalline metallic Pt nanoparticle can be confirmed (Figure S9c).The distribution of these clusters suggests that diffusion of Pt SAs takes place over the entire surface (or facets) of the nanosheets.
If the reaction medium is changed to a plain water environment, SA agglomeration and the photocatalytic reaction become significantly affected.HAADF-STEM (Figure 1g,h,i) shows the state of the anatase thin films after the surface has been irradiated in pure H2O for 0 h, 1 h, and 24 h.In this case, still 23% of Pt is in a SA state and only small Pt clusters (not visible by SEM) are formed after 1 h, which is in stark contrast to the aqueous MeOH solution that shows a heavily agglomerated situation already after 1 min.In plain water, after 24 h illumination still 12% of the Pt is available in a SA state after 24 h (Figure 1k).Clearly, as compared to the MeOH solution, the agglomeration process is drastically slowed down in pure water.Pt δ+ with δ≈2 [55,70] while a Pt 4f7/2 position of 70.6 eV corresponds well to Pt 0 in the form of metallic nanoparticles. [55,70]The surface reaction of the [PtCl6] 2-precursor leads to a reduction of the Pt 4+ reduction to a Pt 2+ species while all coordinating Cl -species are lost (the latter is evident from the absence of Cl-species in the Cl 2p region XPS of a Pt SA decorated surface,

Figure S11
). Please note that the wide peak at 75.2 eV is a Ti 3s satellite peak.
In line with the microscope data, also XPS shows already after 1 min of illumination some conversion of the surface trapped Pt δ+ to metallic Pt 0 apparent in the spectra from the pronounced shoulder at lower binding energies, pending at ≈70-71 eV. Figure 2b gives an evaluation of the Pt δ+ and Pt 0 content over illumination time based on deconvolution of the Pt 4f signal into the two-valent and zero-valent contributions.For the data in Figure 2a, after 1 min, 51 % of the Pt on the surface is converted from SAs (Pt δ+ ) to Pt 0 , and after 10 min of illumination, >80 % of the Pt is in a metallic state.After 24 h, 7.2% of the Pt is still in Pt δ+ state, which perfectly fits to the SA fraction of 7% as determined from HAADF-STEM imagingthis allows to conclude that after 24 h illumination all Pt atoms, which are not SAs are in a metallic state.This is in-line with the observed transformation of SAs to a 2D loosely bound raft to a 3D nanocrystal as described above.A comparison of the XPS and the microscopy data after 1 min reveals that at this stage a large fraction of the agglomerated Pt atoms (the rafts) is still in the Pt δ+ state, implying the SA binding character for SA in the small agglomerates.(corresponding to Pt-O) and a maximum at 1.9 Å.The signal at 1.9 Å reflects a strong interaction between Pt and the TiO2 support via Pt-O bonds (as was reported for sub-nanometer PtOx clusters in [71] ).The samples illuminated for 1 min and 24 h show a weakened signal from Pt-O at 1.6 Å, and a characteristic peak at 2.65 Å (similarly to Pt foil reference), which is in the range of values reported for Pt-Pt bond length. [70]The k-space spectra are shown in Figure S12 in the ESI.These findings are well in line with the crystallization of Pt δ+ species to Pt 0 nanoparticles observed in STEM.
XAS measurements performed for nanosheets irradiated in pure H2O (Figure S13) show drastically slower rate of reduction of Pt SAs to metallic state.Here virtually no change can be seen after 1 min of irradiation, in contrast to observations in presence of MeOH.
Additional DRIFTS experiments were carried out with 1% CO diluted in Ar in a continuous flow for 1 h with a flow rate of 30 mL min -1 .The spectra of the non-illuminated Pt SA-anatase sheets (Figure 2e) show CO vibrational frequency at 2125 cm -1 which is characteristic of linearly bonded CO over Pt single sites. [70,72]In contrast, the samples illuminated for 1 min and 24 h, respectively, show a shift of the peak to 2082 cm -1 and 2090 cm -1 (Figure 2f, 2g), which is characteristic to CO adsorbed on metallic (crystalline) Pt. [70,72] Notably, slightly higher peak frequency after 24 h irradiation, as compared to 1 min, can originate from the overall higher coordination number of Pt, i.e. larger average Pt cluster size. [72]us XAS and DRIFTS measurements are in good agreement with TEM and XPS.In order to elucidate critical factors affecting agglomeration, we further evaluated the influence of the sacrificial agent concentration (Figure 3a,b) and the illumination intensity (Figure 3c,d).From Figure 3a,b it is evident that already the addition of sacrificial agent (MeOH) at 5 vol.% not only strongly affects the H2 evolution activity (Figure 1c) but also the agglomeration rate (Figure 3b).This suggests that the charge flux across the surface SAs significantly affects the agglomeration.This is also well in line with the effect of the light intensity (Figure 3c,d).Evidently, with a lower light intensity, a slower agglomeration is observed.Specifically, after 1 min of irradiation at different intensities we observed 13% (at 2.7 mW cm -2 ), 49% (at 65 mW cm -2 ), or 76% (at 600 mW cm -2 ) conversion of Pt δ+ to metallic Pt 0 (Figure 3d).In accordance with XPS, the corresponding SEM images (Figure S14) show relatively more particles on the surface for the nanosheets irradiated at higher intensities.
Nevertheless, the H2 evolution rate (Figure S15) remains hardly affected by the illumination for lower powers and only for intensities of >100 mW cm -2 show mild saturation effects.I.e., over the entire illumination intensity range the availability of charge carriers is determining the H2 evolution rate rather than the number and availability of co-catalytic sites.
It should be noted that the light induced agglomeration is in stark contrast to the thermal stability of the SA configuration at room temperature (the XPS spectra in Pt 4f region are virtually unchanged for samples stored for 6 months at room temperature, Figure S16a).This suggests sufficiently high barriers for Pt SAs diffusion due to strong Pt δ+ -TiO2 interaction.Also illumination in dry Ar atmosphere does not lead to agglomeration (Figure S16b), indicating that this process requires external charge transfer pathways (e.g an aqueous reaction medium) for photogenerated charge carriers, which otherwise recombine to preserve charge neutrality.
Clearly, illumination under reaction conditions has a crucial destabilization effect on the SA configuration on the surface, as it leads to agglomeration of SAs to dimers, multimers, rafts and NPs.Based on some preliminary experiments, this apparently applies to a wider context of SA-semiconductor systems, such as Pd SA-TiO2 and Pt SA-GaP (Figure S17-S18), as well as to other Pt SA deposition procedures, e.g.utilizing strong electrostatic adsorption of (NH3)4PtCl2 and/or thermal post-treatments in air or H2 (Figure S19-S21).The fact that the presence of a sacrificial hole capture agent (MeOH) leads to accelerated agglomeration indicates that an electron flux via surface anchored SAs and the follow-up reactions lead to the destabilizing effect.In other words, the reduced state of Pt δ+ , e.g.Pt (δ-1)+ or Pt-H, seems to become mobile on the anatase surface.The MeOH-induced enhancement of the electron flux to Pt SA surface sites can be ascribed to the "indirect" mechanism: a swift transfer of photogenerated holes to MeOH [73,74] leads to reduced e --h + recombination losses, thus allowing more electrons to reach the TiO2 surface.In order to elucidate these findings, DFT calculations were performed.When "pristine" Pt SAs are considered on a TiO2(100) surface, the binding is rather strong, with Pt binding energy (Eb,Pt) amounting to −5.29 eV, which is somewhat lower than Pt cohesive energy (absolute value 5.84 eV). [75]This suggests that Pt SAs could agglomerate if the Pt lattice could be restored somehow.This is supported by the fact that the formation of Pt dimer releases an integral binding energy of −5.73 eV, and differential binding energy of −6.17 eV, which overcomes the Pt cohesive energy.In the case of TiO2(001) surface we found SA binding energy of −6.57eV, while dimer formation releases differential binding energy of −8.42 eV, suggesting the preference of dimer configuration over isolated SAs.The situation is similar with trimers on both surfaces, whose formation releases integral energies of −6.39 eV ((100) surface) and −7.44 eV ((001) surface).It is important to say that in the case of the (001) surface all the initial configurations of Pt SAs converged to the same final configuration, suggesting that there is a clear site preference for Pt SAs on the (001) surface.
To investigate thermodynamic driving forces for Pt SA agglomeration, we also analyzed how H binding, corresponding to the intermediate formation of Pt-H during photocatalytic H2 evolution, affects the binding of SA Pt to the TiO2 surface.In the case of (100) surface, the results suggest that upon H binding to Pt, the preferred binding site shifts compared to a bare Pt SA (Figure 4a).This goes along with an elongated Pt-surface distance or a change of the Pt SA coordination, indicating the weakening of the Pt SAs bonding to the surface (Figure 4b,c).
The situation is the same for any Pt adsorption site investigatedonce H is attached to Pt it changes the PtH binding-preference to a different adsorption site.Not only that the preferred binding site of PtH is different from that of Pt δ+ (the SA form), but the binding at the preferential site is also much weaker (reaching up to −1.59 eV).Thus, under H2 evolution, Hads formation at Pt SA causes the weakening of Pt binding to the TiO2 surface and makes it more mobile over the TiO2 surface.This is a straightforward conclusion as it was shown that the surface diffusion barriers of simple adsorbates scale with their binding energies. [76]Once the H adatom is formed on Pt SA, PtH diffuses over the surface and finds the most energetically favored site.We note that the situation is the same when one extra electron is added to the system (negatively charged slabs, see Figure S22).For negatively charged slabs we compared relative stability of Pt SAs at different adsorption sites and found that it perfectly correlates with the PtH binding energies at the same sites.This suggests that electron injection shifts the preferred binding sites compared to the "pristine" Pt SA.These are expected to be Pt agglomerates, as the formation of (PtH)2 dimer over TiO2 surface releases −3.27 eV, which exceeds PtH monomer binding on the surface by the factor of 2, suggesting thermodynamic preference of Pt agglomerates over Pt SAs.In the case of trimers, the formation releases −2.60 eV, which also surpasess PtH monomer binding energy significantly.The binding is favored due to the Pt-Pt interactions, which is reflected in the Pt-Pt distance in such dimer amounting to 2.47 Å (Figure 4d,e), being less than the corresponding interatomic distance in bulk Pt.In the case of (001) surface we found analogous situation.PtH binding is much weaker than Pt SA binding (−2.81 eV).PtH dimer and trimer formation releases −3.64 eV and −2.71 eV, respectively.
To address the activity trends, we investigated how strongly H atoms are bonded to Pt sites on the considered TiO2 surfaces.In the case of (001) surface, we find that SA, dimer, and trimer, all have quite similar |ΔGH|, around 0.5 eV, indicating that their activities towards H2 formation should be similar, while the formation of agglomerates takes place under reaction conditions, due to the weakening of Pt binding to the surface and thermodynamic preference for agglomerates over isolated PtH.On the (100) surface, where we managed to stabilize Pt SAs at different adsorption sites, we found that in the case of Pt SAs there is a direct correlation between the Pt binding energy to the TiO2 surface and the H binding energy to Pt SAsa stronger Pt binding to the TiO2 results in a better optimized H binding in terms of H2 evolution (Figure 4f).In contrast, the H binding at Pt atoms in Pt dimer is very strong, −3.50 eV (integral binding energy), as also in trimer (−3.01 eV), suggesting that such structures do not have a high photocatalytic activity for H2 evolution.Simply, H is bound too strongly, and the H2 formation is not energetically favored.In contrast, the preferentially bound Pt SAs provide the bestoptimized H binding in terms of the thermodynamic H2 evolution barrier, considering different adsorption sites of Pt SAs at the anatase TiO2(100) surface (Figure 4g).Thus, if the H2 evolution on the "truly" active sites is faster than the surface diffusion of the intermediately formed PtH, such an active site remains in the SA state, while other Pt atoms will agglomerate.Based on the results of DFT calculations, the observed linearity of H2 evolution plots (Figure 1c) at different MeOH concentrations (despite the apparent formation of less active agglomeration rates) can be now ascribed to the predominant H2 production by light-stable SA sites, whereas the contribution from mobile Pt species forming clusters and nanoparticles has an apparently small impact on the overall performance (in line with [56] ).Notably, stable H2 evolution rates even at the initial 0-1 h of irradiation (which corresponds to incomplete agglomeration in pure water) may indicate that the formation of inactive Pt species (e.g.Pt dimers) occurs at a time scale smaller than the resolution of the measured H2 evolution plots.This is supported by observations of abundant nanoparticles formed already during the initial 1  1e).Although the stable Pt δ+ sites constitute less than 10 % of the total surface Pt (according to XPS analysis after 24 h of irradiation in Figure 2b), we suggest that these sites can yield the majority of H2 owing to the superior surface coverage and favorable energetics of H-bonding.

Conclusion
The present study provides an important insight on the matter of stability of Pt SAs on TiO2 surfaces over the course of photocatalytic H2 evolution.Initiating photocatalytic reactions triggers surface migration and agglomeration of otherwise stable Pt SAs, which in our case affects >90% of the loaded Pt SAs on both thin film TiO2 layers and hydrothermal TiO2 nanosheets.Notably stable H2 evolution rates measured for different light intensities over 24 h lead us to identify charge carriers rather than the availability of co-catalytic sites as ratedetermining quantity and accordingly, the observed agglomeration behavior is independent of the initial SA loading and does not affect the overall reaction rate.Agglomeration can be subdivided into three stages i) initial fast agglomeration of SAs and dimers to loosely bound 2D rafts, ii) further agglomeration to 3D nanocrystals, and iii) slow ripening of the 3D nanocrystals.Further, we observe at least 6% of the initially loaded SAs in a SA state and could identify Pt δ+ states being present in the 2D rafts.The distinct correlation of the agglomeration rate with the presence of hole scavenger or irradiation intensity, points out that the agglomeration process is driven by photogenerated electrons, and their transfer to the surface Pt-atoms.DFT calculations point out the significant weakening of Pt SA bonding to TiO2 surface upon adsorption of H atoms which leads to surface migration of PtH units and finally to the agglomeration of the Pt co-catalyst species.
Fabrication of TiO2 supports: 7 nm thick TiO2 films were sputtered on the 8 nm SiO2 membranes by magnetron sputtering, followed by annealing in air at 450 ○ C for 1 h.
First, 1.2 mL HF was added dropwise to 10 mL of Ti(OBu)4 and vigorously stirred for 40 min at room temperature.The solution was then sealed in a Teflon-lined autoclave and heated to 200 ○ C in an oven.After 24 h of heat treatment, the autoclave was cooled naturally, and the precipitated reaction products were collected by centrifuging and washed with ethanol and deionized water (this sequence was repeated for 3 times).The obtained powders were dried in air at 75 ○ C for 24 h.
8]60,62] TiO2 films on supporting membranes were immersed in 100 ml of an aqueous solution of H2PtCl6•6H2O for 1 h.The membranes were washed with deionized water and dried in N2 gas flow.
Similarly, 20 mg of TiO2 nanosheets were dispersed in an aqueous solution of H2PtCl6•6H2O (concentration range of 0.0005-2 mM), followed by stirring in dark at room temperature for 24 h.The suspensions were then centrifuged, and the collected powders were washed with deionized water (this sequence was repeated for 3 times).The obtained powders were dried in air at 70 ○ C for 24 h.The software employs a random forest classifier-based algorithm based on "Ilastik" [77] to separate regions with and without SAs present.This advanced procedure allows segmentation not only based on image grey-values (intensity), but also on edge properties, texture and localization in the 2D image.In a second step, the generated segments are classified, corresponding to their calculated area, as SA (< 0.11 nm²), dimer (0.11-0.16 nm²), cluster < 1 nm 2 (0.16-1.0 nm²) and cluster > 1 nm 2 structures (>1.0 nm²).Finally, the estimated area occupied by each type of Pt structures is divided by their total area to obtain the corresponding area fraction values.

Morphology and composition characterization:
The surface morphology of Pt SA-loaded TiO2 supports before and after irradiation was characterized by field-emission scanning electron microscopy (SEM, Hitachi S-4800), transmission electron microscopy (TEM, Philips CM30).
The chemical composition was studied by X-ray photoelectron spectroscopy (XPS, PHI 5600).
The specific surface area of TiO2 nanosheets was measured by Brunauer-Emmett-Teller (BET) method.Pt content in Pt SA-loaded TiO2 nanosheets was determined by electrothermal atomization atomic absorption spectroscopy (ETA-AAS, Analytik Jena AG contrAA® 600) using a high-resolution double monochromator and Xe light source.
X-ray absorption spectra were collected at P65 (Petra III) beamline at the Deutsches Elektronen-Synchrotron (DESY) Hamburg, Germany.All the spectra were measured at Pt L3edge (11.564 eV) using a Si(111) double crystal monochromator to scan the incident energy.
The energy was calibrated with a reference platinum metal foil.The measurements were conducted in transmission mode (reference samples were diluted with cellulose and pressed in a pellet) and in fluorescence mode (as a powder in a quartz capillary without any dilution).
Fluorescence X-rays were collected using a 4 pixel silicon drift detector (SDD).DFT calculations: Density Functional Theory (DFT) calculations were done using QUANTUM ESPRESSO software. [78]Calculations were done at the GGA-PBE level. [79]The kinetic energy cut-off for the plane-wave basis set was 50 Ry, while the charge density cut-off was set to 800 Ry. Atomic positions were relaxed until the residual forces acting on atoms were below 0.01 eV Å −1 .Anatase (100) slab was created using a 2×2 supercell containing 120 atoms, following the addition of 20 Å of vacuum to generate the desired surface plane.The (001) surface was modelled using 3×3 surface cell with 108 atoms in the pristine slab.The first irreducible Brillouin zone was sampled with a 2×2×1 set of k-points generated using the general Monkhorst-Pack scheme. [80]In the first step, Pt adsorption was investigated on different adsorption sites on the TiO2 surface.Pt binding (Eb,Pt) energy was calculated as: b,Pt =  Pt@TiO 2 − ( TiO 2 +  Pt ) where  Pt@TiO 2 ,  TiO 2 , and  Pt , are the total energy of the TiO2 slab with Pt adatom, the total energy of the TiO2 slab, and the total energy of an isolated Pt atom.
H binding energy at Pt atoms and the binding energy of the PtH fragment to the TiO2 surface were calculated in an analogous way following Eq.( 2).In the case of Pt dimer (n = 2) and trimer (n = 3), differential and integral binding energies were obtained using the following equations: b,Pt  = ( Pt 2 @TiO 2 − ( TiO 2 +  Pt )) / b,Pt  =  Pt 2 @TiO 2 − ( (−1)Pt@TiO 2 +  Pt ) Analogous calculations were done for the (PtH)2 dimer and (PtH)3 trimer.Further, to convert hydrogen binding energies on Pt to Gibbs free energies of Hads formation (ΔGH), in terms of H2 evolution, the procedure of Nørskov et al. was used. [81]The obtained structures were visualized using VESTA software. [82]atistical analysis: Experimental data processing was performed using the Origin software.No Pre-processing of the XPS spectra included shifting to a standard Ti 2p peak position in TiO2 at 458.5 eV and smoothening.Fitting of the XPS spectra was performed using a Shirley background with symmetric Pt δ+ , asymmetric Pt 0 , and Ti 3s satellite peaks until reaching convergence (Chi-Squared value below 5).XAS data reduction was carried out using the Demeter software packages.Data reduction (alignment and normalization) was accomplished using the Athena code (version 0.9.26). [83]

Figure 1 .
Figure 1.Characterization of Pt SA-loaded TiO2 layers on TEM membranes.(a) Top view of a TEM membrane; (b) Schematic cross-section of a TEM membrane; (c) Photocatalytic H2 evolution by the Pt SA-loaded TiO2 nanosheets (2 mM H2PtCl6) in aqueous solutions at different MeOH concentrations; (d-f) HAADF-STEM images of TiO2 layers after 0 min (d), 1 images).(j-k) Area fraction of Pt single-atoms and clusters, and (l) average cluster area, as calculated from the HAADF-STEM images of TiO2 layers irradiated for 0-24 h in 50 vol.%MeOH (j) or pure water (k).
also including 10 min and 4 h illumination.The results are in-line with a first fast agglomeration of SAs and their agglomerates to larger clusters within 1 min, afterwards the fraction of SAs stabilizes and even regenerates after 10 min.In contrast, the already nucleated agglomerates dissolve or further coarsen, as evident by the size evolution of the clusters (Figure1l) and discussed in more detail below.The results of HAADF imaging are complemented by chemical information from STEM EDX mapping (see FigureS4), which confirms a uniform Pt distribution after SA decoration, as well as formation of Pt clusters upon light exposure.
However, the first agglomerated clusters become SEM visible after 1 min of illuminationthus SEM allows to follow SA Pt agglomeration on larger length scalestherefore SEM images are shown in the insets of STEM images and are given again in FigureS6.It is noteworthy that after 24 h of illumination, the HAADF-STEM (Figure1f) shows some agglomerates that can be identified as metallic crystallite nanoparticles, evident from the d-spacing of the crystallized particlesan example is shown in Figure1fwhere d = 0.23 nm which corresponds well to the (111) plane of metallic Pt.This illustrates that SAs not only agglomerate to 2D rafts on the surface but also to 3D crystallites.The site-specific nature of the HAADF-STEM analysis allows tracing the evolution of individual agglomerates and clusters.FigureS7depicts the evolution of agglomerates as function of illumination time.From 1 min to 4 h a constant change of cluster morphology can be observed, moreover the appearance of the clusters is cloudy and single spots can still be observed, this impression of a 2D object was further supported by a "defocus"-series, where the agglomerates are nearly simultaneously in focus with the substrate atoms and the surrounding SAs.Between 4 h and 24 h coarsening of the agglomerates is observed, in addition the cloudy character gets partially lost and as discussed above lattice planes can be detected implying the transformation to a 3D object.Two other examples of identical location electron microscopy are displayed in FigureS8, here on a larger field of view, the dissolution of a small agglomerate can be observed in a) and even cluster migration accompanied by coarsening in b).All observations point towards a high mobility of Pt atoms under photocatalytic conditions, i.e. not only Pt in the form of SAs but also Pt species that are already in an agglomerated state.In order to confirm the general validity of the observations on the TiO2 thin film membranes, we additionally carried out experiments on facetted TiO2 nanosheet-powder.For this we decorated hydrothermally fabricated preferentially facetted (001) anatase nanosheet powder with Pt SAs.After synthesis (as described in the experimental section), the nanosheets were uniformly loaded with Pt SAs (see HAADF-STEM, FigureS9a,d and SEM, FigureS10a).By XPS analysis no Cl signal was detected, indicating that the precursor complex fully reacted with the surface Ti-O species.Similarly to the thin film anatase layers on the TEM-grid, already after 1 min of light exposure in 50 vol.%MeOH, aggregation of Pt SAs was observed on the dominant (001) facets of the nanosheets (FigureS9b,e, FigureS10b) and extended irradiation leads to an overall increase in size and number of Pt clusters (FigureS9c,f, FigureS10c-e).
Following consistent observations of light-induced Pt SA agglomeration on the thin sputtered TiO2 nanolayers and the faceted anatase nanosheets, further surface analysis was performed.From the XPS data for anatase flakes (Figure 2a, and Figure S10), the Pt 4f peak shifts over the illumination time from a position of Pt 4f7/2 at 72.4 eV to 70.6 eV.A peak position for Pt 4f7/2 at ~72.4 eV corresponds well to literature data for surface coordinated single atom

Figure 3 .
Figure 3. XPS characterization of the Pt SA-loaded TiO2 nanosheets irradiated at different conditions.(a) XPS spectra in the Pt 4f region after irradiation in pure H2O for different duration; (b) Pt δ+ and Pt 0 content evaluated from XPS spectra after irradiation in aqueous solutions with different MeOH concentrations; (c) XPS spectra in the Pt 4f region after 1 min irradiation in 50 vol.%MeOH solutions at different light intensities; (d) Pt δ+ and Pt 0 content evaluated from XPS spectra after irradiation in 50 vol.%MeOH solutions at different light intensities.

Figure 4 .
Figure 4. (a) Surface model of TiO2(100) with indicated diffusion path of Pt SA upon formation of Hads and its release, (b) optimized structure of Pt SA on TiO2(100), (c) optimized structure of PtH on TiO2(100), (d) optimized structure of Pt dimer and (e) PtH dimer on TiO2(100) surface, (f) correlation between Pt SA binding energy to the TiO2 surface (Eb,Pt), and the H binding energy on the correspondins Pt SAs (Eb,H), (g) reaction profiles for H2 evolution on Pt SA, Pt dimer, and Pt trimer on TiO2(100) surface.
HAADF-STEM imaging: Identical location high-angle annular dark-field transmission electron microscopy (HAADF-STEM) images were acquired with a Thermo Fisher Scientific Spectra 200 C-FEG.The microscope is probe-corrected and equipped with a cold field emission gun (X-CFEG).Images were taken at 200 kV with dwell times between 5 and 40 µs, a collection angle of 56-200 mrad and a beam current below 20 pA.Machine learning-based segmentation of HAADF-STEM images: Machine learning-based segmentation was performed with the commercial software Arivis Vision 4D (version 3.2).
Photocatalytic H2 evolution: Photocatalytic H2 evolution by Pt SA-loaded nanosheets was studied at different light intensities in absence or presence of a hole scavenger (MeOH). 2 mg of powders were ultrasonically dispersed in 10 ml aqueous MeOH solutions, purged with Ar gas for 20 min, and irradiated under vigorous stirring in sealed quartz cells using LED light source (λmax = 365 nm, Imax = 600 mW cm -2 ).The concentration of H2 gas produced in the cells statistical analysis was performed for the SEM, XPS, XANES, EXAFS, CO-DRIFT data.The reproducibility of morphology and composition of the as-formed Pt SA-loaded samples was confirmed by SEM and XPS in at least 3 experiments.Photocatalytic H2 evolution curves were measured in single experiment with error bars corresponding to a typical range of experimental error.Statistical analysis of the area fraction and the cluster area of the surface Pt after different illumination periods was performed by calculating mean average of the values obtained by machine learning-based segmentation of HAADF-STEM images taken in 4 different locations of the sample with the total area of 5660 nm², and the standard deviation is represented by the error bars.XPS spectra were pre-processed, analyzed and fitted using a PHI Multipak software.