Introduction

Supported metal nanoparticles (NPs) are pivotal as heterogeneous catalysts in numerous industrial and environmental processes1,2,3,4,5,6. Early research established structure-activity correlations7 and rationalized the impact of NPs size on catalytic activity and selectivity8,9. Metal NPs are also active entities with their own dynamic behavior in the presence of adsorbates and/or under reaction conditions10,11,12,13. Understanding the structural evolution of metal NPs in various reaction environments is crucial for precise control over their size and reactivity, both of which are essential for catalyst design, as well as for the regeneration of deactivated catalysts14,15.

Platinum NPs are among the most studied catalytic systems. The CO-induced reconstruction of Pt surfaces and of Pt NPs is a classic example of adsorbate-induced surface reconstruction11,16,17,18,19,20,21. Similarly, H2 can induce electronic and morphological restructuring of supported Pt clusters22,23,24,25,26. At room temperature and sub-ambient pressure, H2 splits on Pt surfaces, forming various Pt hydrides. In nanoscale Pt particles, hydride formation is accompanied by structural reconstruction, as demonstrated by Mager-Maury et al.27 using density functional theory (DFT) calculations on a model Pt13 cluster supported on γ-Al2O3. Their findings revealed that, as the H/Pt ratio increases, the Pt13 cluster transitions from a biplanar morphology, strongly interacting with the support, to a more symmetric cuboctahedral structure, which almost detaches from the support. Similar behavior was observed for Pt37 clusters on graphene23, suggesting a universal nature of this process across diverse NP sizes and supports of very different nature. The biplanar morphology is not predicted to be the most stable in the absence of the support, and even the cuboctahedral morphology is unstable in the absence of hydrogen28. Hence, the morphology of supported Pt NPs is strictly related to both the extent of interaction with the support and the H/Pt ratio29.

The available theoretical calculations for Pt clusters on γ-Al2O327 propose three primary descriptors for this restructuring phenomenon as a function of the hydrogen partial pressure: (i) changes in the relative amounts of linear and bridged Pt hydrides, (ii) elongation of the average Pt-Pt distance (i.e., the “breathing” of the Pt NPs), and (iii) increased Pt-support distance (i.e., the detachment of the Pt NPs). To date, only limited experimental evidence supports these theoretical predictions. As regards the amount of Pt hydrides, titration experiments30,31,32, as well as X-ray Absorption Spectroscopy (XAS) data24,30,33,34,35, found values of the H/Ptsurf ratio much greater than 1 for ultra-dispersed Pt NPs, in good agreement with models where the Pt NPs are completely hydrogenated (e.g., for supported Pt13 NPs, a H/Pt ratio as high as 2.5 is predicted by DFT calculations)27. Recently, we provided experimental evidence of the conversion of linear Pt hydrides into bridged species on 5 wt% Pt/Al2O3 using Fourier transform infrared (FT-IR) spectroscopy and inelastic neutron scattering (INS) spectroscopy36,37,38, in excellent agreement with the theoretical prediction27. With respect to structural descriptors, only the breathing of Pt NPs in the presence of H2 was confirmed experimentally. XAS showed a contracted fcc lattice39 and higher structural disorder in naked Pt NPs compared to bulk Pt, while hydrogen adsorption induces lattice relaxation and more structural order (e.g., restoring of the fcc symmetry)36,40,41,42,43. In contrast, apart from the mobility of Pt clusters detected by environmental STEM under H244, the direct observation of the H2-induced detachment of Pt NPs from the support remains elusive. This is due to the minute structural effects associated with NP detachment from the support. A decrease in the intensity of a weak contribution ascribed to Pt-support distance upon increasing the H2 pressure up to 20 bar was inferred from XAS data of a Pt/Al2O3 sample with an average particle size of 0.8 nm, and ascribed to a restructuring of the Pt NPs45. To date this has not been observed by synchrotron X-ray pair distribution function (PDF) analysis46,47,48, despite its ability to describe the structural changes in 1–3 nm sized Pt NPs on Al2O3 induced by adsorbates from the gas phase49. Finally, the reversibility of the H2-induced restructuring and its dependence on the reaction environment are also unexplored yet critical areas, with practical implications for catalyst stability and performance under hydrogenation conditions.

To capture experimentally all these dynamic aspects of NPs restructuring, we combined high-energy x-ray diffraction (XRD) and PDF analysis, in a modulated excitation (ME) experimental approach, to study two Pt/Al2O3 catalysts characterized by different particle size distributions, previously investigated by operando XAS, operando IR spectroscopy and in situ INS36,37,38,50. Compared to extended X-ray absorption fine structure (EXAFS) spectroscopy, which could deliver comparable information in the range 2–5 Å, PDF provides enhanced sensitivity due to the possibility of comparing changes occurring simultaneously to short and long distances. The experiments were designed to recreate the conditions in which NPs are expected to either be completely hydrogenated and (partially) detached from the support or, conversely, to strongly interact with the support. A time resolution of 1 s/pattern, coupled with the ME approach, enabled tracking the fast dynamics of the Pt-hydride species without the need to slow down the process, as done in previous experiments36. While none of the previously mentioned techniques provided direct information on Pt-support interactions, the combination of a structure-sensitive technique such as PDF with experiment design gave conclusive evidence of the simultaneous H2-induced breathing and reversible (partial) detachment of Pt NPs in both gas and liquid environments. These findings bridge the gap between experiments and theoretical calculations on supported Pt NPs, demonstrating experimentally the ductility of supported NPs in the presence of adsorbates in terms of particle reconstruction and deformation, opening up fascinating new perspectives on their catalytic behavior in various reaction environments.

Results and discussion

Characterization of the pristine catalysts

The two Pt/Al2O3 catalysts show similar homogeneous distribution of Pt NPs but different average NP size (Fig. 1). The pre-reduced sample, PtAl(R), displays a log-normal distribution with a mode around 3.0 nm (Fig. 1a–c), while PtAl has smaller particles, with a bimodal distribution peaking at about 1.0 nm and 2.2 nm (Fig. 1b–d). Both distributions are narrow enough to enable reliable XRD and PDF analysis irrespective of the type of normalization (Figure S4), which ensures that outlier NPs with larger size do not dominate the distribution.

Fig. 1: HAADF STEM.
figure 1

Selected STEM images of pristine (a) PtAl(R) and (b) PtAl. Corresponding particle size distributions (c, d) using the Rice estimator for bin widths.

Full size image

The XRD patterns of the catalysts (Fig. 2a, c) are dominated by the peaks of the industrial Al2O3 support, which consists of a mixture of the δ- and θ-phases with a minor contribution from γ-Al2O3 (see Fig. S3 in Supporting Information for details). To isolate the contribution of Pt in the patterns of PtAl and PtAl(R), the data of the same Al2O3 support used to prepare the two catalysts was subtracted from those of the two catalysts. From a rigorous perspective, the difference pattern also contains information on the Pt-support interaction, whose magnitude is, however, far less relevant than the Pt-only contributions. As a result, the difference diffraction pattern of PtAl(R) (Fig. 2a) exhibits clear diffraction peaks of face-centered cubic (fcc) Pt. An average crystallite size of 2.8 nm was estimated based on the XRD peak widths (Table S2). In contrast, the difference pattern of PtAl (c) shows broader, weaker peaks (ca. 2.6 and 4.0 Å-1), which can be attributed to an oxidized phase46,51, in line with previous measurements on the same sample36 and with observations of well-dispersed metal hydroxide in catalysts prepared by deposition-precipitation reported in the literature52,53,54,55.

Fig. 2: Total scattering data of the pristine catalysts and the support.
figure 2

a XRD and (b) PDF patterns of PtAl(R) and of the Al2O3 support and their difference. c XRD and (d) PDF patterns of PtAl and of the Al2O3 support and their difference. Difference in (c) is magnified by a factor 3 for clarity. The main diffraction peaks and Pt-Pt correlations for fcc Pt are labeled in (a) and (b), respectively. Difference patterns are offset for clarity.

Full size image

Figure 2b, d compare the PDF curves of both catalysts to that of the Al2O3 support. Figure S5 shows the same data in a wider R range, along with the fit used to estimate the NPs size. The difference PDF contains the interatomic distances within the Pt NPs, as well as the Pt-support contributions, which are however too small to be appreciated in static patterns. For PtAl(R) (Fig. 2b), the most intense peaks appear at ca. 2.76, 4.79, and 7.32 Å, corresponding to the 1st, 3rd and 7th shell Pt-Pt distances in the fcc Pt phase (i.e., those with the highest multiplicity). The intensity falloff as a function of R can be modeled using a spherical particle diameter of 3.3 nm (Table S2), in agreement with the estimate from XRD and consistent with the particle size estimated from STEM. The difference PDF of PtAl contains only a few clearly discernible peaks up to 6 Å (Fig. 2d), among which the peak at about 2.0 Å is typical of Pt-O distances in Pt oxide/hydroxide.

Isothermal reduction in gas and liquid phase

The reduction of the catalysts was performed in H2 under isothermal conditions in both gas phase (H2/Ar, 150 °C) and liquid phase (H2-saturated cyclohexane, 70 °C). No changes were observed for PtAl(R) (Figs. S6–S7 and Tables S1, S2), confirming that the Pt phase was already reduced and a possible oxide passivation layer was undetectable. In contrast, the reduction of the Pt-oxide phase was clearly observed for PtAl, in both gas (Fig. 3) and liquid phases (Fig. S8). Difference patterns (Fig. 3c, d) highlight the relatively small changes to the XRD patterns during reduction in H2/Ar at 150 °C.

Fig. 3: Time-resolved reduction of PtAl.
figure 3

Time-evolution of (a) XRD and (b) PDF patterns of PtAl during the isothermal reduction in H2/Ar flow at 150 °C. Time evolution (yellow to blue) of difference (c) XRD and (d) PDF patterns obtained by subtracting the last pattern before H2 was admitted. The entire process takes place in 40 s.

Full size image

Both XRD and PDF data show that the Pt-oxide phase is gradually reduced to metallic fcc Pt by H2. Difference XRD (Fig. 3c) shows broad negative peaks at the positions assigned to PtO2 (around 2.6 and 4.0 Å−1), which pinpoints its disappearance. Positive peaks appear at positions characteristic for metallic Pt (2.8, 3.2, 4.5, 5.3 and 7.0 Å−1). In the PDF (Fig. 3d), the sharp peaks corresponding to the 1st, 3rd, and 7th shell Pt-Pt distances emerged clearly during reduction. From the temporal evolution of their normalized intensity, we obtained kinetic information on the Pt oxide reduction (Fig. 4a): reduction by H2 in the gas phase was completed within 10 s, while it was almost ten times slower in the liquid phase. We attribute this difference to a combination of factors: the higher reduction temperature in the gas phase experiment (150 °C vs. 70 °C, the latter imposed by the boiling point of the solvent), the poor solubility and the lower diffusivity of H2 in cyclohexane, resulting in lower H2 concentration in the liquid-phase experiment.

Fig. 4: Comparing reduction of PtAl between gas and liquid environments.
figure 4

a Kinetics of PtAl reduction in gas and liquid phase, determined from normalized intensity of the nearest-neighbor Pt-Pt peak at about R = 2.77 Å in the PDF patterns. b Difference PDF patterns (dots) of PtAl at the end of the isothermal reduction in H2/Ar flow at 150 °C (gas) and in H2-saturated cyclohexane at 70 °C (liquid) and their fit (solid lines), compared to simulated PDF patterns of the Pt NPs in the Pt55H91/γ-Al2O3 model reported in (a). Only the contribution from the Pt phase has been simulated. Patterns are offset for clarity.

Full size image

At the end of the reduction process, and in excess of H2, the Pt phases in PtAl were identical regardless of the reaction environment and temperature (dotted lines in Fig. 4b). Fits to the background-subtracted PDFs (solid lines in Fig. 4b) returned very similar lattice parameters and a NP size of 2.3 ± 0.1 nm after both liquid and gas-phase reductions (Table S2). This provides compelling evidence that the Pt oxide phase in PtAl is completely reduced in cyclohexane also at 70 °C, and that the reaction environment does not affect Pt particle size, which remains very close to the center of the distribution evaluated by STEM on the pristine (unreduced) PtAl catalyst (Fig. 1). The same applies also to PtAl(R) (Fig. S9), although the average NP size remains far larger than in the reduced PtAl (Tables S1, S2). The PDF patterns of the reduced catalysts (Fig. 4b), where the short-range order is largely dictated by bulk fcc Pt, are well reproduced by the highly symmetric PtxHy/Al2O3 cuboctahedral models from DFT calculations27, in which the Pt NPs are surrounded by hydrogen and detached from the alumina support (Pt55H91/Al2O3 in Fig. 4). The choice of hydrogenated Pt NPs models is justified by the experimental evidence from IR spectroscopy in both the gas and the liquid phase36,38 and INS in gas phase37 that, under very similar experimental conditions, various surface Pt hydride species exist on these catalysts.

In this regard, we estimated the average H/Pt ratio in the two catalysts as the ratio between the dispersion value measured by H2/O2 chemisorption (D) and the “geometrical” dispersion (Dgeom) determined from the particle size distribution measured by STEM (assuming spherical shape). For PtAl, Dgeom = 38%, which is 1.6 times smaller than the dispersion measured by H2/O2 chemisorption (D = 63%), and corresponds to an average H/Pt ratio of 1.6. Even though approximated, these calculations demonstrate that small Pt NPs in PtAl adsorb much more hydrogen than expected from their geometrical surface area, in agreement with the structural rearrangement predicted by DFT. For PtAl(R) the same calculation gives Dgeom = 26%, to be compared with D = 21% obtained from H2/O2 titration, resulting in a much smaller H/Pt ratio of 0.8.

Pt NPs breathing and detaching in the presence of H2

Hydrides are unstable in the absence of H2 and are decomposed once the surrounding is changed to Ar after reduction; as a consequence, the Pt NPs undergo surface relaxation accompanied by disordering. Figure 5a compares the simulated PDF for a supported Pt55 NP dehydrogenated and in two different hydrogenation states, which are the most stable structures determined by DFT for that H-coverage. Supported Pt34 and Pt13 NPs behave very similarly (Fig. S10). In the high H-coverage regime (Pt55H91/Al2O3, Fig. 5b), the simulated PDF peaks are sharp and largely determined by the Pt fcc structure. In the dehydrogenated state (Pt55/Al2O3, Fig. 5a,b), the Pt NP adopts an irregular morphology and interacts strongly with the alumina support. Correspondingly, the simulated PDF shows a more contracted first shell Pt-Pt distance compared to the two hydrogenated counterparts, and completely smeared-out higher-shell Pt-Pt peaks. In an intermediate hydrogenation state (Pt55H44/Al2O3) the simulated PDF is also intermediate. This effect is more pronounced upon decreasing the Pt particle size, in supported Pt34 and Pt13 NPs (Fig. S10), and is emphasized in the difference PDF.

Fig. 5: Comparing the experimental PDF to optimized structural models.
figure 5

a Simulated PDF patterns of Pt55H91/Al2O3, Pt55H44/Al2O3 and Pt55/Al2O3 models and the difference with respect to Pt55/Al2O3 (diff). b Pt55H91/Al2O3, Pt55H44/Al2O3 and Pt55/Al2O3 models. c Experimental PDF patterns of PtAl immediately after the isothermal reduction in H2/Ar at 150 °C and after purging in Ar for 10 min and their difference. The box indicates the region fitted as described in (d). d Fitting procedure applied to a PDF pattern, corrected by a linear background subtraction, of PtAl collected during one ME cycle at 150 °C. The experimental pattern is fit with four Gaussian components.

Full size image

The experimental PDF patterns of PtAl in the fully hydrogenated (i.e., under H2/Ar at 150 °C) and dehydrogenated (i.e., under Ar at 150 °C) states (Fig. 5c) suggest that analogue changes indeed occur to the Pt NPs during dehydrogenation. While the experimental patterns in Fig. 5c are strongly affected by the scattering of Al2O3 (which constitutes 95 wt% of the sample), the contribution of the support is negligible in the simulated PDF patterns in Fig. 5a, because the models comprise only a thin slab of Al2O3. In the difference PDF (“diff” in Fig. 5c), where the scattering contribution of the support is cancelled, it is possible to appreciate a contraction of the first Pt-Pt distance of about −0.3 ± 0.05% and a higher dispersion of the 3rd (4.80 Å) and 5th (6.20 Å) shell Pt-Pt peaks, closely resembling the simulated difference pattern for a Pt55 NP (“diff” in Fig. 5a). The contraction of the first shell Pt-Pt distance upon removal of H2 can be interpreted as the breathing of the Pt NP induced by removal of hydride species, as well as the more irregular morphology responsible for the disappearance of the higher-shell Pt-Pt peaks.

To provide also kinetic information and to emphasize structural changes with increased sensitivity, we performed a series of experiments after the isothermal reduction adopting the ME approach56,57. The raw (i.e., not Al2O3 background subtracted) PDF data calculated from the XRD patterns collected during the ME experiment were analyzed to track the breathing of the Pt NPs in the presence/absence of H2, and their interaction with the support. In this latter case, difficulties arise because only very small changes can be expected to affect the Pt-support distances (less than 0.2 Å)27.

We focused on the region 2.5–3.75 Å range of the PDF (Fig. 5d), which consists mostly of two peaks: the peak at 2.76 Å, dominated by nearest-neighbour Pt-Pt correlations and including a minor contribution of the Al2O3 support, and the peak at 3.33 Å consisting mainly of Al-Al correlations (Fig. S11). Both peaks also contain information on Pt-support correlations: short and very heterogeneous Pt-support pair contributions, (Pt-supp)short, falling within the peak centered at 2.76 Å, predominate for Pt NPs in strong interaction with the support (Figure S12a); while longer Pt-support distances are found above 3.2 Å (i.e., within the peak centered at 3.33 Å) for hydrogenated NPs, (Pt-supp)long (Fig.  S12b). Attempts to fit the data in the 2.5–3.75 Å range with five components (Pt-Pt + 2 peaks for Al2O3 + (Pt-supp)short + (Pt-supp)long) (Fig. S13) were unstable, especially due to the strong parameter correlation between the Pt-Pt and (Pt-supp)short contributions. Therefore, we used four Gaussian contributions (Fig. 5d). The position of the first peak (blue) accounts for the first shell Pt-Pt distance; its full width at half maximum (FWHM) indicates the relative importance of the (Pt-supp)short contribution, i.e., the larger the FWHM, the larger the (Pt-supp)short contribution. The second and third Gaussian contributions (grey) account for the Al2O3 support and their parameters were fixed during the fit to the values obtained fitting the pattern of blank Al2O3. The amplitude of the fourth contribution (violet) estimates the relative weight of the (Pt-supp)long contribution, i.e., the different degree of interaction between the Pt NP and the Al2O3 support.

In the gas phase experiment at 150 °C, the first shell Pt-Pt distance of PtAl equals that of bulk Pt (2.774 Å) at the end of the reduction step (Fig. 6a). This distance contracts in the absence of hydrogen by ca. 0.7 ± 0.1% with respect to the hydrogenated situation and expands in its presence by the same amount, in a reproducible and reversible way over the ten modulation cycles. It is worth noticing that during the half cycles without H2 the Pt-Pt distance does not decrease to its H2-equilibrated value, indicating that the 10-min Ar flushing phase is insufficient to completely remove the Pt hydride species. This behavior is not due to structural limitations of the sample environment. For PtAl(R) in H2 (Fig. 6c), the Pt-Pt distance is slightly expanded with respect to the bulk. It behaves similarly to PtAl during dehydrogenation, but with a smaller perturbation of the Pt-Pt bond (0.2 ± 0.1%), as expected from its larger, less ductile NPs. The NP breathing in both catalysts is not very pronounced but can be followed very precisely by these PDF data.

Fig. 6: Gas-phase modulated excitation experiment.
figure 6

Fit of the experimental PDF data of (a, b) PtAl and (c, d) PtAl(R) during 10 modulation cycles of H2/Ar at 150 °C. a, c Variation of the Pt-Pt distance in the first five cycles; b, d variation of FWHM of the Pt-Pt peak and of the amplitude of the (Pt-supp)long peak, averaged over the ten cycles. Fit results are reported as dots, while lines are obtained upon averaging 5 neighboring points.

Full size image

Variations of the FWHM of the first Gaussian contribution and of the amplitude of the (Pt-supp)long contribution are smaller than those of the Pt-Pt distance. We report these two parameters averaged over ten cycles for PtAl in Fig. 6b and for PtAl(R) in Fig. 6d, thus providing an increase in signal-to-noise ratio (by a factor \(\sqrt{{N}_{{cycles}}} \sim 3.2\)) that facilitates resolving the small changes discussed. For PtAl, the FWHM of the first peak increases by ca. 7 ± 1% of its original value upon H2 removal, revealing that the (Pt-supp)short contribution becomes relevant in the inert (Ar) environment when hydrides are removed. Simultaneously, the amplitude of the (Pt-supp)long contribution decreases by 23 ± 1%. Therefore, the inert atmosphere promotes the decomposition of hydride species and the interaction of the NP with the support. While statistically relevant, the smaller variations of all three parameters for PtAl(R) are attributed to the larger particles in this sample, which are less prone to detachment from the support. Taken together, these two observations provide experimental evidence that, upon variation of hydrogen coverage, Pt NPs not only breathe but also move relative to the support. The distance from the support increases in the presence of H2 and hydride species, and decreases when hydrides are removed.

Similar behavior is observed in the liquid environment (Fig. 7), but here both breathing and movement relative to the support are less pronounced and kinetically slower than in the gas phase. The low H2 concentration in the liquid phase and competition between H2 and the solvent for adsorption sites at the Pt surface may both contribute to this. However, the behavior can be still captured by the ME experiment: the Pt-Pt distance of NPs in PtAl changes by 0.3 ± 0.1% and the amplitude of (Pt-supp)long by 3 ± 1%. Observation of similar phenomena in the presence of a solvent and in the gas phase indicates that the structural/morphological reconstruction is a general feature of Pt NPs.

Fig. 7: Liquid-phase modulated excitation experiment.
figure 7

Fit of the experimental PDF data of (a, b) PtAl and (c, d) PtAl(R) during 10 modulation cycles of H2-saturated cyclohexane/Ar-saturated cyclohexane at 70 °C. a, c Variation of the Pt-Pt distance in the first five cycles; b, d variation of FWHM of the Pt-Pt peak and of the amplitude of the (Pt-supp)long peak, averaged over the ten cycles. Fit results are reported as dots, while lines are obtained upon averaging 5 neighboring points.

Full size image

Our data demonstrate that supported Pt NPs subjected to a reversible hydrogenation change their structure and distance from the support. This is supported by conformational sampling via molecular dynamics27, which indicates that the most stable arrangement of Pt13 NPs under the given conditions of H2 pressure and temperature is a cuboctahedron detached from the support, while in absence of H2 the stable arrangement is biplanar, in strong interaction with the support. The structural rearrangement is ultimately driven by the energetic stability of the final structure under H2 atmosphere. This stems from the competition between several energetic factors: (i) Pt-Pt cohesion, (ii) Pt-H interactions, (iii) Pt-support interactions. While the latter is favoured at low H coverage, Pt-H interactions dominate at high coverage. As hydrogenated Pt NPs interact less strongly with the Al2O3 support, they might become more available for interaction with adsorbates, including substrates to be hydrogenated. This has implications for a better understanding and control of the sustained stability of hydrogenation catalysts, since the detachment of the Pt NPs from the Al2O3 support seems to be facilitated by hydrogen adsorption and could be a first step towards their mobility and leaching from the support, especially in liquid phase. The reproducibility of the modulation cycles in both gas and liquid phase, however, indicates that this must be a slow process under these experimental conditions and compared to the time scale of the ME experiments.

Discussion

In this study, we provide experimental evidence of the simultaneous H2-induced breathing and movement of well-dispersed Pt nanoparticles (NPs) relative to their alumina support. Both phenomena, previously predicted only by theoretical models, were captured directly with unprecedented accuracy using high-energy total scattering (XRD and PDF), combined with a concentration modulation experimental approach and an experimental design aimed at reproducing conditions where Pt NPs are either completely hydrogenated or bare. Our findings reveal that both phenomena occur in both gas- and liquid-phase environments with a magnitude highly correlated with the size of the NPs. The process is fully reversible over the time scale of our experiment, demonstrating the stability of the catalyst towards further mobility or even leaching.

A better understanding of the dynamic behavior of metal NPs in the presence of adsorbates may open exciting opportunities for their rational control under reaction conditions. In this respect, our work not only validates the possibilities suggested by theoretical predictions, but also sets a new benchmark for experimental studies in the field of heterogeneous catalysis with supported metal NPs. It introduces a methodological approach to accurately  describe modifications at the metal-support interface in gas and liquid environments, where catalysis often occurs.

Methods

Catalyst synthesis and preliminary characterization

The two 5 wt% Pt/Al2O3 catalysts investigated in this work were prepared by Chimet S.p.A. (Viciomaggio, Italy), following a deposition-precipitation method58 using an industrial high-surface-area transitional alumina as a support (SSA = 116 m2 g−1; pore volume = 0.41 cm3 g−1; mixed phase). These two catalysts differ in whether or not a reduction step was performed after Pt deposition: one catalyst (the pristine sample, called PtAl) was not pre-reduced and the Pt phase is fully oxidized; the other (PtAl(R)) was reduced using sodium formate as reducing agent, hence, the Pt phase is metallic, but passivated by a thin oxide layer. The latter is a typical industrial catalyst for hydrogenation reactions. In both cases, after the synthesis the catalysts were thoroughly washed with water to ensure that reagents were removed (e.g., ICP measurements showed Na content ~200 ppm, comparable with values found on bare alumina) and then dried at 120 °C overnight. The nominal Pt dispersion determined by H2/O2 titration59 on samples reduced at 120 °C was D = 63%, and D = 21% for PtAl and PtAl(R), respectively. The lower dispersion of PtAl(R) matches the behavior of Pd/Al2O3 catalysts synthesized by the same method52, showing that reduction with sodium formate produces larger particles and stronger interaction with the support.

Transmission Electron Microscopy (STEM) imaging was performed using a probe-corrected JEOL JEM ARM-200 F (NeoARM) microscope equipped with a cold FEG gun operated at 200 keV. Scanning transmission electron microscopy (STEM) high-angle annular dark-field (HAADF) images were collected at 68 mrad < α < 280 mrad. Approximately 800 particles were counted in the HAADF-STEM images of each sample in order to estimate the particle size distribution using the approach by Alxneit60.

In-situ XRD measurements

In situ X-ray diffraction measurements were performed at beamline ID15A61 at the ESRF synchrotron (Grenoble, France) using an X-ray beam energy of 98 keV (λ = 0.1265 Å) and a photon-counting Pilatus3X 2 M CdTe detector (Dectris, Switzerland) positioned 330 mm from the sample. Beam size was 100 × 100 μm2 (vertical×horizontal) and the average flux on sample was 1012 photons s−1. Instrumental peak broadening was evaluated by measuring a standard reference material (NIST Cr2O3 674b, Fig. S1 and Table S1). A stainless-steel, low-dead volume cell designed to feed the gas/liquid bottom-up through a 9 mm3 catalyst bed provided leak-tight operation, minimal dead volume upon switching flows, low scattering background, and a wide exit angle for measurements enabling PDF analysis (see Supporting Information Fig. S2).

The sample (approximately 30 mg) was placed in the cell between two quartz wool plugs used to both diffuse the gas/liquid feed and suppress sample movement. Throughout the following protocol, we collected scattering data continuously with a time resolution of 1 s/pattern. The sample was kept 1 min at ambient temperature under Ar (gas phase experiment) or Ar-saturated cyclohexane (liquid phase experiment). Then the temperature was increased at a rate of 10 °C/min to 150 °C (gas phase) or 70 °C (liquid phase), under Ar (flow of 100 ml/min, gas phase) or Ar-saturated cyclohexane (1 ml/min, liquid phase). At this point, the gas flowing through the cell was changed to 5 vol% H2/Ar (gas phase) or to H2-saturated cyclohexane (liquid phase) to perform an isothermal reduction for 30 min. Lastly, the sample was washed in Ar (gas phase) or Ar-saturated cyclohexane (liquid phase) at the same temperature for 30 min. Modulated excitation experiments were started at this point by exposing the sample to repeated cycles of 5 vol% H2/Ar (10 min) and Ar (10 min) or H2-saturated cyclohexane (5 min) and Ar-saturated cyclohexane (5 min). Reference data such as blank Al2O3 in different phases were collected in the same conditions.

The use of a stainless steel valve and a ceramic valve (VICI AG, Switzerland) enabled repeated, no-dead time switching between the two different gas/liquid flows. Gas feeds were delivered through mass flow meters (Bronkhorst, Netherlands) and stainless steel tubing (1/16”) while liquid feeds were circulated by a peristaltic pump (Spetec, Germany) installed upstream of the cell and Teflon as well as stainless steel tubing. Glass bottles used as reservoirs for cyclohexane solvent (HPLC purity, Sigma Aldrich) were fitted with a frit to allow for saturation with pure gases (Ar or H2, both 99.999 vol%). Cyclohexane was selected as it can be considered non-interacting with the catalyst and to remain consistent with the previous IR study of Pt hydrides on these catalysts38. A quadrupole mass spectrometer (Hiden Analytical, UK) connected to the cell outlet monitored the gas feed and its evolution in the gas phase experiments.

Data reduction and PDF calculation

Two-dimensional X-ray scattering data were azimuthally integrated using FabIO62 and pyFAI63. Data were corrected for the flat-field response and spatial distortion of the detector61, X-ray beam polarization, and variations in the incident photon flux. Azimuthally integrated data were converted to PDF using PDFgetx364, using a range of momentum transfer 0.8 ≤ Q ≤ 22 Å−1 and the scattering patterns of the empty cell and/or blank Al2O3 support collected under identical conditions as background. This enabled isolating Pt-Pt correlations related to the NPs from interatomic correlations involving the support.

Data fitting and simulation

Structural models were fitted to XRD and PDF data using Topas v765. Crystallite size was estimated either from fits to PDF data using a spherical-particle dampening model or from fitting high-resolution XRD data, collected at the ID22 beamline66. With the exception of the pristine PtAl sample, the PDF data were fitted with Gaussian peaks67 corresponding to Pt-Pt distances in the Pt fcc structure, support-support and Pt-support distances, using lmfit68. The reported parameter uncertainties are based on the covariance matrices of the corresponding least-squares fits (see Supporting Information). Additionally, they were compared to a set of calculated PDFs corresponding to a series of PtxHy/γ-Al2O3 models, consisting of PtxHy nanoparticles of various sizes (x = 13, 34, 55) and hydrogen coverages supported on a slab of dehydroxylated γ-Al2O3(100) surface. Notably, the local structure of γ-Al2O3 is indistinguishable from that of the other alumina phases (Fig. S3c). The most stable structures of the Pt13Hy/γ-Al2O3 models were obtained previously by using DFT calculations based on velocity scale molecular dynamics followed by a quenching procedure27, while the larger models (x = 34 and x = 55) were obtained by static optimization because of computational cost. The Pt34H54/γ-Al2O3, Pt55H44/γ-Al2O3 and Pt55H91/γ-Al2O3 structures were adopted previously to explain the behavior of supported Pt NPs in the same PtAl sample in different hydrogenation conditions37, while the naked Pt34/γ-Al2O3 and Pt55/γ-Al2O3 systems were constructed as detailed in Supporting information. The selected sizes of the Pt clusters, about 1 nm (Pt13) to 1.5 nm (Pt55), represent the best compromise between the range of STEM sizes and the limitations imposed by computational costs. Total and partial PDFs of the PtxHy/γ-Al2O3 models were calculated over the same range of Q used for the experimental PDF data using DebyeCalculator69.