Introduction

Intermetallic nanoparticles (NPs) demonstrate outstanding performance across a range of applications, including optical devices1,2, electrolysis3,4,5,6, and critical processes in the chemical industry such as propane dehydrogenation (PDH) for the targeted production of propylene7,8,9,10. Incorporating a second metallic component not only reduces the consumption of noble metals in catalysts but also tailors the catalytic properties. The regular structure of intermetallic NPs contributes to uniformity of active sites enabling a deeper understanding of structure-activity relationships. Despite the thermodynamic stability of ordered intermetallics (IMAs) compared to their disordered counterparts, the disorder-to-order transition needs to surmount the kinetic energy barrier associated with atom ordering11,12,13.

The most commonly used method for preparing supported intermetallic nanoparticles is impregnation (IMP), which involves saturating the pore volume of the support with various metal precursor solutions. Though it is straightforward, it often leads to particles with a wide size distribution and inhomogeneous alloying14. These issues could be alleviated through electrostatic adsorption. However, it’s constrained by surface area and acid-base resistance of the support15. Alternatively, the atom migration (AM) strategy has proven effective in constructing ordered interfaces, leveraging the energetic driving force that prompts atoms to migrate from their original positions to the equilibrium positions within the crystal lattice16,17,18. A notable example of this strategy is atomic layer deposition (ALD), which, due to its self-limiting reactions, enables the deposition of uniform layers on high-surface-area or porous solids19,20,21. However, it is limited to the intermetallic alloys that the different ALD reactions share similar temperatures and surface chemistries22,23, which is constrained by the availability of precursors and the requirement for specialized and sophisticated reactors. An industrially feasible, controllable, and scalable method for preparation of supported intermetallic nanoparticles is therefore desired.

In this work, we demonstrate a robust approach of atomic gas-migration different from the general methods to synthesize homogeneous intermetallic alloys. The gas-migration and trapping processes were realized through a physical mixture of separately supported metal species which are simple and versatile, in one reactor. Specifically, immobile metal (such as Pt) clusters were first supported on inert silica. Gaseous mobile metal atoms (M, such as Zn, Ga, In) migrate in situ to Pt surface to form Pt1M1 intermetallic alloys. Under industrially relevant conditions for PDH (550–600 °C), the in-situ formed Pt1M1 intermetallic catalyst achieves an exceptionally high propane conversion rate and propylene selectivity (>95%) for over 1300 hours, outperforming the state-of-the-art Pt-based catalysts.

Results

Formation of Pt-M intermetallic alloys via atomic gas-migration approach

In a representative process (Fig. 1), the optimized quantities of Pt and the second metal were individually supported on silica. These metals were subsequently ground to a precise particle size range of 20–40 meshes or 230–400 meshes and mixed as stacked granules (see details in the Supplementary Information Fig. S1, S2). The physically intimate mixture was heated in a hydrogen flow, during which the loaded Pt is firstly reduced and nucleated at a low temperature of 170 ~ 350 oC (Fig. S3, S4). Then the other metal was gradually reduced and diffused as gaseous atoms along the hydrogen stream at the higher temperatures (> 350 oC), which could be captured by Pt seed to form a homogenous Pt-M intermetallic alloy.

Fig. 1: Schematic illustration of the atomic gas-migration approach.
figure 1

a A mixture of Pt/SiO2 and MOx/SiO2 was loaded in a quartz tube. b Evolution of Pt seeds to Pt1M1 ordered alloy in the process of M migration-trapping during H2 pretreatment at high temperature.

Full size image

The aberration-corrected high-angle annular dark-field scanning transmission electron microscopy (AC-HAADF-STEM) images (Fig. 2a–d and Fig. S5) and X-ray diffraction analysis (XRD) provided a clear identification of the Pt-M intermetallic alloy (Fig. 2e). The observed speckling effect and overlapping distributions (Fig. S6) indicated that the two metals were intimately mixed15,24. The structures of PtZn (tetragonal, P4/mmm, a = 4.0247 nm, c = 0.3491 nm, PDF#00-006-0604), PtGa (cubic, P213 (198), a = 0.491 nm, PDF#04-004-4951), PtIn2 (cubic, Fm\(\bar{3}\) m (225), a = 0.6353 nm, PDF#04-004-8870) were determined. For all the Pt-M alloys we prepared via gas migration, ordered intermetallic formed where Pt were geometrically isolated by inactive promoter atoms25,26,27,28. Compared to the face-centered cubic (fcc) Pt (PDF # 00-004-0802), a set of diffraction peaks of 31.1°, 41.8° and 44.8° appeared, which is ascribed to the (110), (111) and (200) planes of Pt1Zn1 IMA. In Pt1Ga1 alloys, the planes (110), (012), (211) and (321) were found at angles of 25.6°, 40.1°, 45.2° and 71.9°, respectively, which differ from the peaks at 39.8°, 46.2°, 67.5°associated with (111), (200) and (220) planes of Pt fcc structure. In Pt-In alloys, where the atomic ratio of Pt to In is 1:2, distinct crystallographic diffractions were observed at 24.2°, 40.1°, and 47.4° corresponding to the (111), (220), and (311) planes, respectively. Meanwhile, the fast Fourier transform (FFT) images from the AC-HAADF-STEM data, along with the selected area electron diffraction (SAED) patterns, revealed two distinct d-spacing distances of 0.351 nm and 0.287 nm. These distances are consistent with the [001] and [110] planes of PtZn, respectively (Fig. 2d and Fig. S7), indicating the formation of Pt1Zn1 IMAs, distinct from the Pt fcc structure (Fig. S8). At the surface of PtZn (110), periodic assemblies of [PtZn4] motifs were identified, wherein each Pt atom is surrounded by four adjacent Zn atoms. This configuration results in the formation of isolated Pt1 sites in Pt1Zn1 IMAs. The formation of PtZn IMAs was also demonstrated by the CO adsorption band that shifts from 2068 cm−1 to 2045 cm-1, the latter is attributed to the CO linear adsorption on single atom Pt1 site in ordered Pt1Zn1 intermetallic alloy25,29,30 (Fig. S9).

Fig. 2: Formation of Pt-M intermetallic alloy.
figure 2

a–c Low magnification and high magnification AC-HAADF-STEM images of Pt-Zn alloys prepared via the gas-migration strategy. The blue spheres in (b) represent Pt atoms and the pink spheres represent Zn atoms. d FFT images of PtZn1.4/SiO2 particle in (c). e XRD patterns of Pt-M alloys prepared via the gas-migration strategy. (f) Pt LIII-edge k2-weighted Fourier-transformed EXAFS spectra of Pt/SiO2 and PtZn1.4/SiO2 that are reduced and tested at 600 oC, as well as Pt-Pt and Pt-Zn scattering paths. g In situ CO-DRIFTS of PtZn1.4/SiO2 after H2 treatment at different temperatures from 100 to 500 oC. The loading of Pt is 1 wt.% for (a, f, g), 3 wt.% for (b–d) and 5 wt.% for (e), respectively.

Full size image

The EXAFS Fourier transform (FT) of PtZn1.4/SiO2, Pt/SiO2 (Fig. 2f, see Fig. S10 for more details for k-space and r-space data) reveals distinct peaks. The coordination configurations of PtZn1.4/SiO2 and Pt/SiO2 were analyzed using quantitative least-squares EXAFS curve-fitting. The fitting results indicate that the Pt-LIII edge spectrum of PtZn/SiO2 includes both Pt−Zn and Pt−Pt scattering paths (Table S1) in the first shell. The coordination numbers of Pt−Zn and Pt−Pt in the first coordination sphere are estimated to be 1.6 and 3.5 at distances of 2.67 and 2.76 Å, respectively. This first shell coordination aligns with our DFT model of intermetallic PtZn, which shows coordination numbers of 1.4 and 3.1 for the first shell Pt−Zn and Pt−Pt, respectively. Notably, despite the high Pt loading of 1 wt.%, extremely small PtZn NPs (ca. 0.9 nm) are uniformly distributed, lower than that of Pt/SiO2 (ca. 1.6 nm) (Fig. S8, Fig. S11 and Fig S12 for 3 wt.% PtZn NPs). This hints the gas-migration strategy benefits a reduced and controllable particle size with homogeneous alloying.

Inductively coupled plasma-optical emission spectroscopy (ICP-OES) analysis of the Zn:Pt ratio in the feed and sieved Pt/SiO2 (Fig. S13), combined with XPS and EDS analysis for PtZn/SiO2 NPs gave a clearer impression of the utilization of Zn feed throughout the catalyst bed (Table S2, S3). After reduction treatment, about 70% initial Zn content remains within the catalyst bed. When the zinc content in the feed is added to a sufficient level, specifically with a Zn/Pt feed ratio of 2:1 in our experiments, the atomic ratio of Pt to Zn in the resulting Pt1Zn1 alloy approximates 1:1, aligning with the typical atomic ratio observed in PtZn intermetallic alloys25. It is worth noting that the gas-migration strategy affords the smaller and more homogeneous PtZn nanoparticles compared to that via impregnation methods (ca. 2.2 nm, 1 wt.% of Pt for 1Pt1.7Zn/SiO2) as reported in our recent research25.

To explore how the introduction of Zn species inhibits the growth of Pt particles, in situ characterizations under chemical atmospheres were performed. The structural transformation was corroborated by in situ CO diffuse reflectance infrared Fourier-transform spectroscopy (DRIFTS) measurements by altering the reduction temperature from 100 to 500 oC. The peak at 2060 cm−1 at 100 oC could be attributed to the linearly bonded CO on the unsaturated Pt sites. When the temperature increased to 300 oC, the peak broadened and exhibited a slight redshift. This may be ascribed to that partial adsorption of Zn on the Pt surface, which weakens the strength of the Pt(5d)-CO(2π*) bonding interactions31. Quasi in situ X-ray photoelectron spectroscopy (XPS) (Fig.S14), in situ X-ray absorption near edge structure (XANES) (Fig. S15) and H2-TPR experiments (Fig. S4) showed that Pt was fully reduced while only a small amount of metallic Zn atom was released at 300 oC. When the temperature raised to 500 oC, the CO adsorption peak is shifted to a lower wavenumber at 2045 cm−1 in PtZn1.4/SiO2, which is usually attributed to the on-top CO adsorption on the single-atom Pt1 site30,32. In situ XPS showed that the binding energy of Pt 4f7/2 is reduced from 71.5 eV to 71.2 eV when temperature was elevated from 300 oC to 550 oC, which indicates the electronic contribution of Zn atoms caused by the formation of PtZn alloy (Fig. S14). After treatment at 600 oC, the slightly decreased white-line intensity of PtZn1.4/SiO2 compared to Pt foil and Pt/SiO2 also indicates electron transfer from Zn to the Pt 5d orbital, consistent with findings from CO-DRIFTS and XPS analyses. Based on the results from XANES and XPS, most of the ZnO was reduced at 550 ~ 600 oC while some oxidized Zn species remained. Due to the excess Zn in the feed, it is possible that some ZnO remained on the SiO2. In situ detection in a propane atmosphere showed minimal changes in the states of Pt and Zn on the catalyst, indicating that the prepared PtZn alloy can remain stable under reaction atmosphere. Based on discussion above, the size of Pt seeds was controlled before the Zn addition, then uniform PtZn alloys with close particle sizes can be obtained at the same Pt loadings (Fig. S16).

C–H activation studies and catalytic performance

To underscore the advantages of utilizing a Pt-Zn intermetallic alloy through an atomic gas-migration strategy, we conducted a thorough investigation of C–H activation and catalytic performance during the propane dehydrogenation (PDH) reaction. It is well established that metallic Pt serves as the primary active site due to its superior C–H activation capabilities compared to metallic Zn25,33,34,35. C3H8 temperature-programmed surface reaction (TPSR) shows that the initial C–H activation temperature for C3H8 on the synthesized PtZn1.4/SiO2 is 237 °C, slightly higher than that of 207 °C on Pt/SiO2 (Fig. S17). Consistent with previous reports on PtZn alloys, the introduction of Zn slightly increases the difficulty of C–H bond activation due to the electron contribution from Zn to Pt, but the formation of the alloy simultaneously weakens the side reactions of C–C cracking and carbon deposition25,34,35, thereby enhancing the selectivity of propylene. When the Zn content increased (Figs. S18, S19), we observed a gradual improvement in initial propane conversion, rising from 36% to 48%, with a corresponding increase in propylene selectivity from 66% to 99%. After ten hours of continuous reaction at 600 °C, the conversion of propane barely declined from 58.9% to 56.2%, with a deactivation rate (kd) of 0.009 h−1 (Fig. 3a, b). It is noted that conversions of propane on PtZn1.4/SiO2 are close to thermodynamic equilibrium values under temperatures of 550–600 °C, indicating its superior activity for PDH (Fig. 3c)25,33,34,35.

Fig. 3: PDH performance under near-industrial operating conditions.
figure 3

a, b PDH performance of Pt/SiO2, PtZn1.4/SiO2, and ZnO/SiO2. c The test and equilibrium conversion of propane over PtZn1.4/SiO2 at different temperatures. Reaction conditions: 550–600 °C, WHSV(C3H8) = 4 h−1, C3H8/H2/N2 = 8/8/34. d Long-term PDH performance of PtZn1.4/SiO2. Reaction conditions: 600 °C, WHSV(C3H8) = 24 h−1, C3H8/H2/N2 = 4/4/17. e Extended PDH performance of PtZn1.4/SiO2. Reaction conditions: i, 600 °C, WHSV(C3H8) = 8 h−1, C3H8/H2/N2 = 16/8/34; ii, 600 °C, WHSV(C3H8) = 4 h−1, C3H8/H2/N2 = 8/8/34; iii, 580 °C, WHSV(C3H8) = 4 h−1, C3H8/H2/N2 = 8/8/34; iv, 550 °C, WHSV(C3H8) = 4 h−1, C3H8/H2/N2 = 8/8/34; v, 550 °C, WHSV(C3H8) = 1 h−1, C3H8/H2/N2 = 2/2/46; vi, 550 °C, WHSV(C3H8) = 2 h−1, C3H8/H2/N2 = 4/4/42; vii, 550 °C, WHSV(C3H8) = 4 h−1, C3H8/H2/N2 = 8/8/34. Regeneration conditions: 30 min regeneration at 500 °C under air stream (10 mL·min−1). The mass ratio of loaded Pt/SiO2 and ZnO/SiO2 was ca. 1.5:1 for (a–e). For the regeneration step in (e), 50 mg of ZnO/SiO2 were added in regeneration cycle, constituting approximately one-third of the initial ZnO/SiO2 feed. The metal loadings of Pt and Zn were both 1 wt.%.

Full size image

Durability test and validation of the approach

In order to evaluate the catalytic performance under industrially relevant conditions, we conducted a comprehensive long-term test. During the 100-h test (WHSV(C3H8) = 10 h−1, C3H8:H2 = 2:1) (Fig. S20), the catalyst exhibited an extremely high propylene selectivity (>95%) without obvious decline in conversion. Then the conversion was controlled at a low level to assess the deactivation rate (Fig. 3d). The catalyst operated for 70 hours at a WHSV of propane  at 24 h−1, with a deactivation rate of 0.006 h−1. After undergoing four dehydrogenation-regeneration cycles (Fig. S21), the initial propane conversion gradually declined from 37.3% to 28.5%. However, by introducing approximately one-third of the initially loaded zinc oxide charge during the fifth regeneration step, we were able to successfully restore the propane conversion to 34.9%.

Considering the application of atomic diffusion strategy in the regeneration process, the optimal catalyst PtZn1.4/SiO2 was subjected to an extensive long-term stability evaluation with continuous dehydrogenation-regeneration cycles (Fig. 3e). Initially, it underwent 86 h of PDH testing (600 °C, WHSV(C3H8) = 8 h-1, C3H8/H2 = 2:1). Subsequent regeneration led to a higher conversion of 54% as the WHSV of propane (4 h−1) decreased. When running at 580 °C, the conversion decreased from 37.6% to 31.8% during 120 hour-test. Following regeneration and ZnO/SiO2 supplementation, the conversion was restored to 45.9%. The subsequent steps involved periodic propane dehydrogenation under varying WHSV of propane and partial pressures at 550 oC evinced the stable conversion and selectivity across a continuous 700-hour operation, outperforming the state-of-the-art Pt-based catalysts (Table S4). Moreover, the average particle sizes of the PtZn alloy tested under different atmospheric conditions at 600 °C for 10 and 100 h remained at 1.0 and 1.2 nm, respectively (Fig. S22), indicating the excellent high temperature stability of the as-prepared PtZn intermetallic catalyst.

Catalysts often undergo frequent regenerations to eliminate carbon deposits, a process that unfortunately also contributes to the loss of metal components36,37,38,39. In our system, while the catalysts demonstrate the ability for activity restoration during the short-term regeneration (WHSV(C3H8) = 4 h-1, with two regenerations within 10 h), it became increasingly challenging to fully recover the activity under more rigorous reaction conditions and frequent regenerations (WHSV(C3H8) = 20 h−1, four regenerations in 25 h) (Fig. S21 and S23). However, we observed a significant boost in initial activity, from 28.4% to 34.9%. We hypothesize that the unique regeneration properties of our catalyst can be attributed to the availability of an excess Zn source through atomic migration. This excess Zn source plays a crucial role in restoring the alloy’s state after regeneration, thereby addressing the issue of metal loss. After a long reaction process, the Zn/Pt ratio in the whole bed eventually aligns with the ratio present in the original Pt1Zn1 alloy (Fig. S24 and Table S3), further validating its suitability for industrial applications.

Structural evolution by DFT and MD simulations

To further elucidate the intricate formation mechanism of the intermetallic alloy through atomic migration, we employed density functional theory (DFT) calculations and molecular dynamics (MD) simulations. The results indicate that the adsorption energy of Zn atoms is significantly lower when they occupy terrace sites on Pt nanoparticles, compared to step sites at the edges or top sites (Fig. 4a). Based on the optimal adsorption pattern obtained from DFT calculation, we then examined the diffusion behavior of Zn atoms on Pt particles. Interestingly, compared to a Pt particle that is already partially covered by Zn (Fig. 4b, c), the diffusion of Zn atoms on a pristine Pt particle exhibits a higher energy barrier (Fig. 4d), which means the Zn atoms will cover the Pt particle firstly and then initialize the alloying process. At high temperatures (600 °C), the adsorbed Zn atoms would diffuse into the interior of Pt particles to form the homogeneous PtZn intermetallic alloy (Fig. 4e), corresponding well to the formation process evidenced by in-situ characterizations above.

Fig. 4: DFT calculation.
figure 4

a Adsorption energy of Zn atoms on different sites of Pt particle. b Zn diffusion process on pristine Pt particle. c Zn diffusion process on Pt particle covered by Zn atoms. d Potential energy diagram of two diffusion mechanisms in (b, c). e Iterative MD simulations of formation process of PtZn alloy.

Full size image

Discussion

To validate the universality of this approach, we extended this method to other binary systems of Pt-In and Pt-Ga, where diffusion temperatures of In and Ga were optimized and set as 550 °C and 600 °C respectively, according to the H2-TPR results (Fig. S25)40,41. The atomically resolved Z-contrast images made a clear distinction of Pt-In and Pt-Ga intermetallic alloy (Fig. S5). As shown in Fig. S26, although their performance is slightly inferior to PtZn alloy, Pt-In and Pt-Ga catalysts exhibit up to 90% selectivity for stable propane dehydrogenation at high temperatures (600 °C). While highlighting the advantages, it is worth noting that one limitation of this atomic diffusion strategy lies in the restrictions of physical properties of the second metals, such as reducibility and melting/boiling points, which somewhat limits the range of available metals.

In summary, we demonstrate the in-situ atomic gas-migration approach to synthesize the ultra-small and homogeneous Pt-M (M = Zn, In, Ga) intermetallic alloys, which significantly improved the selectivity and stability for propane dehydrogenation. By introducing the adjunct metal M during the growth process of Pt particles, it enables higher dispersion of Pt under a broader metal loading range and addresses the issue of metal loss during prolonged reactions compared to conventional impregnation methods. In situ characterizations combining DFT calculations have revealed that M atoms adsorb on the particle surface under high temperature conditions and then diffuse inwards to form highly ordered intermetallic alloys. Further improvements are expected in controlling the diffusion process inside the catalyst and reactor. This study opens the possibility for the preparation and application of homogeneous intermetallic alloys for catalysis via the atomic gas-migration strategy.

Methods

Materials

Tetraammineplatinum nitrate ((NH3)4Pt(NO3)2, ≥ 50.0 wt. % Pt basis), zinc nitrate hexahydrate (Zn(NO3)2·6H2O, reagent grade, 98%), gallium(III) nitrate hydrate (Ga(NO3)3·xH2O, complexiometric EDTA, wt. % Ga< 26.3%), indium(III) nitrate hydrate (In(NO3)3·xH2O, titration with ZnSO4, wt.% In=28–37 %), ammonia solution (25% ~ 28% wt. %), silica (with a particle size of 230–400 mesh, the surface area of 480 m2/g) were purchased from Sigma-Aldrich. All reagents are used as received. The 5 vol.% CO in Ar, air (22 vol.% O2 and 78 vol.% N2), pure propane (99.9%), hydrogen (99.9%), nitrogen (99.9%) were supplied by Air Liquide.

Catalyst preparation

All the SiO2-supported monometallic catalysts were synthesized by the incipient wetness impregnation method. The metal precursor solution was prepared before determining the pore volume of the silica support. The metal loading x% is based on the weight ratio between the metal and the support.

Synthesis of Pt/SiO2. The Pt precursor supported on SiO2 was implemented by careful control over the impregnation procedures. According to Brunelle’s postulation, strong electrostatic adsorption (SEA) facilitates the deposition of the metal complex onto the surface of the certain supports42. The strongest SEA happens in the wet impregnation when the final pH meets the pH range in which electrostatic interactions are strongest. For the synthesis of the supported Pt on silica, the precursor solution was kept at basic condition (PH > 11) by dropwise adding ammonia, so hydroxyl groups of the silica were sufficiently deprotonated and such cation as (NH3)4Pt2+ could be strongly adsorbed. In this way, the metal precursor was anchored on the catalyst and the repulsion between them assists the dispersion. Tetraammineplatinum nitrate ((NH3)4Pt(NO3)2 was used as the precursors and tuned with different weight loadings (0.1 wt.%~5 wt.%). Additional ammonia was added dropwise to keep the pH of Pt precursor solution greater than 11. The catalysts were dried overnight at 80 °C. Before the reaction test, the catalysts were reduced by 50 mL·min−1 10 vol. % H2/ N2 at 550 oC for 1 h. For the test of particle size control, samples were calcined at 300 oC or 400 oC for 1 h with a rate of 2 oC·min-1 before the reduction treatment.

Synthesis of ZnO/SiO2, Ga2O3/SiO2 and In2O3/SiO2. Zn(NO3)2·6H2O, Ga(NO3)3·xH2O and In(NO3)3·xH2O were used as the precursors respectively. The samples were dried overnight at 80 °C and followed by a calcination procedure in the air at 550 °C for 2 h. The loading of metals is based on the weight ratio between the metal and the support, which was in the range of 0.3 wt. % to 5 wt. % as needed.

Synthesis of PtM/SiO2. The method of preparing Pt-M alloy through gas-phase diffusion employs two experimental setups: the two-layer setup and the physical mixing setup. The physical mixing setup is selected as the more efficient approach (Detailed in Supplementary Fig. S1, S2). The Pt-Zn system was taken as an example, with Pt-Ga and Pt-In following a similar procedure. The Pt and Zn content was determined by inductively coupled plasma-optical emission spectroscopy (ICP-OES) and “x” in PtZnx means the atomic ratio of Zn to Pt.

Characterization

In-situ diffuse reflectance infrared Fourier-transform spectroscopy (DRIFTS) study of CO adsorption was conducted on Nicolet iS50 spectrometer, which was equipped with Harrick high-temperature reaction chamber and ZnSe windows. The mercury-cadmium-telluride (MCT) detector was cooled by liquid nitrogen prior to each experiment. The spectra were collected at 4 cm−1 resolution by accumulating 32 scans. For the prepared PtZnx/SiO2, about 40 mg of the powder samples were loaded into the cell and treated at 550 °C in H2 for 1 h following a 10 °C min−1 ramp rate. For the in situ monitoring of the formation of PtZn1.4/SiO2, the physical mixture of Pt/SiO2 and ZnO/SiO2 were treated at room temperature (RT), 300 °C and 500 °C for 30 min in 50 vol.% H2/N2. The Pt loading is 1wt.% and atomic ratio of Zn/Pt in the feed is 2. Then the samples were cooled to room temperature and purged with 40 mL·min-1 of N2 for half an hour. The background was collected in flowing N2 and subtracted automatically from the sample spectrum. Then diluted CO/Ar (5 vol.%) was introduced into the cell at 40 mL/min for 0.5 h. After purging with 40 mL·min-1 of N2 for 40 min, the CO adsorption spectra were recorded.

X-ray diffraction (XRD) measurements were conducted using a Bruker D8 diffractometer operating at 30 mA and 40 kV, utilizing Cu Kα radiation filtered through graphite. Data were collected through step size of 0.2° and a step time of 86.4 s over the range of 2θ = 5° to 80°. All the samples were reduced in diluted H2 (50 vol.%) at a rate of 20 mL·min-1 for 2 h before testing. The atomic ratio of M: Pt in the feed for Zn, Ga and In were 2, 4, 3 respectively. The Pt loading is 5wt.%. The reduction temperatures for Pt-Zn, Pt-In and Pt-Ga are 550 °C, 550 °C and 600 °C, respectively. All samples were ground into powder and then filled into round sample holders. Measurements were conducted under ambient conditions.

Quasi in situ X-ray photoelectron spectroscopy (XPS) was performed in ThermoFischer ESCALAB 250Xi photoelectron spectrometer using a mono Al Kα X-ray source (1486.71 eV, 5 mA). The angle between the sample surface and detector was 90°. The step size of 0.05 eV was employed with a dwell time of 100–400 ms. All the peaks were calibrated with C1s to 284.8 eV. Samples were reduced within an ultra-high vacuum (UHV) connected high-pressure gas cell and subsequently transferred to the analyzer chamber without exposure to ambient air. The calculated amounts of Pt/SiO2 and ZnO/SiO2 were mixed, ground, and pressed into a circular shape. The Pt loading is 1wt.% and atomic ratio of Zn/Pt in the feed is 2. The samples were subjected to heating in a flow of 10 vol.% hydrogen diluted in Ar at a rate of 20 mL·min-1, reaching temperatures of 300 °C, and 550 °C for 30 min respectively. Photoelectron spectra were acquired after each reduction at room temperature.

In situ X-ray absorption spectroscopy (XAS) at Pt LIII-edge and Zn K-edge was measured at Beamline 14 W at the Shanghai Synchrotron Radiation Facility (SSRF). The sample is ground to a powder and pressed onto a stainless-steel holder in a custom-made in-situ reaction cell, with gas passing through the sample. The cell thickness was adjusted to optimize absorption levels. Standard samples were measured in transmission mode, while catalyst samples were measured in fluorescence mode. For in situ monitoring experiments of PtZn1.4/SiO2, the Pt loading is 1wt.% and atomic ratio of Zn/Pt in the feed is 2. The calculated amounts of Pt/SiO2 and ZnO/SiO2 were mixed, ground, and pressed into a circular cell. Samples were first tested at room temperature with a closed atmosphere. Then open 10 vol.% H2/He at 30 mL·min-1 and raise the temperature to 300 °C and then to 600 °C at a rate of 10 °C/min, respectively. The temperatures were maintained for 60 min before testing. Then the atmosphere was switched to 10 vol.% C3H8 in N2 at 600 °C, and the sample are tested after 60 min treatment. The samples were tested at corresponding maintained temperatures. and the data analysis was conducted using the ATHENA, ARTEMIS, and HEPHAESTUS software packages43. Photon energy calibration was achieved by referencing the first inflection points of the Pt LIII edge and Zn K-edge, using platinum and zinc foils.

H2-TPR was tested on chemisorption apparatus (Micromeritics AutoChem II 2920) with a TCD (thermal conductivity detector) detector. Typically, 100 mg catalysts were treated at 400 °C for 1 h in Ar and then cooled to 80 °C. The analysis was conducted in a mixture of 10 vol.% H2 in Ar (30 mL·min−1), ramping temperature from 100 °C to 800 °C at 10 °C/min. For C3H8-TPSR tests, mass spectrometer (Hiden QIC-20 mass spectrometer) was connected instantaneously. A 100 mg sample was heated in 10 vol% H2 in Ar at 600 °C for 1 hour, followed by a cooling process to 50 °C using an Ar flow (30 mL·min−1). Subsequently, the atmosphere was switched to a 10 vol.% C3H8 in Ar stream (20 mL·min−1) with a heating rate of 10 °C·min−1 until reaching 650 °C. The metal loading is 1wt.%. The resulting products included C3H8, C3H6, CH4, and H2, with m/e ratios of 29, 41, 16, and 2, respectively.

The bulk compositions of catalysts were determined by inductively coupled plasma-optical emission spectroscopy (ICP-OES), which was tested on a SPECTRO ARCOS FHM22 spectrometer. A certain amount of the sample was placed in aqua regia and digested by microwave at 250 °C for 1 h. The supernatant was then diluted and filtered for testing. Before testing the samples, calibration curves for each element to be measured were obtained using standard solutions (TraceCERT®, in 10% nitric acid). Then obtained calibration equations were used to convert the response signals into concentrations of the elements in the samples.

Field emission transmission electron microscopy (FE-TEM) was obtained using a JEOL JEM-F200 FE-TEM along with energy dispersive spectroscopy (EDS) images. The aberration-corrected scanning transmission electron microscopy (AC-STEM) images were acquired using the JEOL JEM-ARM200F at Tianjin University, and capable of achieving sub-angstrom resolution. The acceleration voltage was 200 kV. The sample powder was dispersed in ethanol and placed in ice water and sonicated for 30 minutes. The sample was then dripped onto copper grids covered with carbon films or micro-mesh grid. For the particle sizes distribution statistics, at least 100 particles are included. The fitting curve based on Gaussian Distribution was used to give mean size and standard deviation.

Catalytic performance test

The catalysts were loaded in a quartz tubular reactor with an inner diameter of 9 mm. The catalysts with particle sizes of 230–400 meshes or 20–40 meshes were packed inside the quartz fixed-bed reactor with ca. 20 mg quartz wool. In a typical run, the catalyst was preheated in the reductive atmosphere illustrated in Fig. S2. Then the temperature was elevated to the reaction temperature such as 600 °C and retained at 600 °C for 30 min. The reaction atmosphere, temperature, and the weight hourly velocity (WHSV) of propane were altered to test the catalyst’s performance. For instance, a mixture of C3H8/H2/N2 = 8:8:34 was fed at a rate of 50 mL·min−1 over 250 mg catalysts. The reaction was conducted at atmospheric pressure and the weight hourly velocity (WHSV) of propane was 4 h−1. The equilibrium conversions were calculated by HSC chemistry44.

All gas flows were carefully monitored by mass flow controllers. Exhaust gases were analyzed by an online GC (Agilent 7890B gas chromatography). FID (flame ionization detector) with Chromosorb 102 column and TCD (thermal conductivity detector) with Al2O3 Plot column was set. The conversion of propane and selectivity of propylene were calculated as follows:

$${{\rm{Con}}}\left(\%\right)=\frac{{n}_{{{\rm{propane}}} \, {{\rm{in}}}}-{n}_{{{\rm{propane}}}\, {{\rm{out}}}}}{{n}_{{{\rm{propane}}}\, {{\rm{in}}}}} * 100\%$$
(1)
$${{\rm{Sel}}}\left(\%\right)=\frac{{n}_{{\rm{propylene}}}}{{n}_{{{\rm{propane}}}\, {{\rm{in}}}}-{n}_{{{\rm{propane}}}\, {{\rm{out}}}}} * 100\%$$
(2)

Where npropane and npropylene are the number of moles of propane and propylene.

C3H6 formation rate were measured for ranges of C3H8 pressures (0–14 kPa), H2 pressures (0–14 kPa), and temperatures (550–600 oC). C3H6 forward rates \(({r}_{{\mbox{f}}})\) were calculated by correcting measured rates (rm) for the approach to equilibrium \((\eta )\) using:

$${r}_{{\mbox{m}}}={r}_{{\mbox{f}}}(1-\eta )$$
(3)

The stability test (Fig. 3d) was conducted far away from the equilibrium at 600 °C for 70 h, where WHSV(C3H8) = 24 h−1, C3H8/H2/N2 = 4/4/17 and the \(\eta\) was in the range of 0.03 ~ 0.07.

C3H6 formation rates were calculated by correcting forward rates \(({r}_{{\mbox{f}},{\mbox{t}}})\) for each C3H8 pressure, H2 pressure, and temperature for a duration of 60–120 min. These C3H6 forward rates followed a first-order trend at 550–600 oC given by:

$$({{r}}_{{\mbox{f}},{\mbox{t}}})={{r}}_{0}{{\mbox{e}}}^{-{{k}}_{{\mbox{d}}}{t}}$$
(4)

where \(r\left(f,t\right)\) is the C3H6 formation forward rate at time t, \({r}_{0}\) is the C3H6 turnover rate at zero time at each condition, and \({k}_{{\mbox{d}}}\) is the rate decrease constant. Values of \({k}_{{\mbox{d}}}\) at each condition were determined by regressing the time-dependent C3H6 forward rate at varying C3H8 and H2 pressures to the functional form of Eq. (3). The extrapolated C3H6 formation rates at zero time when C3H8 and products could be detected at about 10 min after C3H8 was first injected into the reactor and the stable values at 550–600 oC are reported. After the correction for equilibrium and measured rates, the C3H6 formation rates were similar at the same conditions, reflecting the accuracy for the corrections.

Then we get the expression for \({k}_{{\mbox{d}}}\) for a period time of t1 to t2:

$${k}_{{\rm{d}}}=\frac{{{\rm{ln}}}\left(\right.{r}_{{\rm{f}}}({t}_{1})-{{\rm{ln}}}({r}_{{\rm{f}}}({t}_{1}))}{{t}_{2}-{t}_{1}}$$
(5)

Computational details

The neural network potential driven molecular dynamics (NN-MD) simulation was conducted by LASP software45. NN-MD was performed at a constant temperature (T  =  873 K) within the NVT ensemble. All the spin-polarized DFT calculation was performed in the plane-wave pseudopotential method based on the GGA-PBE exchange-correlation functional46, which is implemented by Vienna ab initio simulation package (VASP, 5.4.4 version47). The valence wave functions were expanded by plane wave with a cutoff energy of 400 eV. We established the model of Pt particle with 19 Pt atoms, which size is about 1 nm and accordance with the real size of the particle in the catalyst. The vacuum layer is set as 15 Å. Then we first conducted Zn adsorption calculation. The Brillouin zone was sampled using 3 x 3 x 3 k points mesh for geometry optimization. All the structures were optimized until the force on each atom was less than 0.02 eV Å−1. After obtaining the optimal Zn adsorption sites, we conduct MD simulation to let the absorbed Zn atoms diffuse freely in the model for 10 ns. Then, we repeated the adsorption and MD calculation until the atomic ratio of Pt/Zn reached 1:1. Moreover, all clusters migration barriers calculations were performed by using the climbing-image nudged elastic band method (NEB)48. The activation barrier Ea was calculated based on following equation:

$$\Delta {E}_{{\mbox{a}}}\,=\, {E}_{{\mbox{TS}}}\!-\!{E}_{{\mbox{IS}}}$$
(6)

Based on the PtZn nanoparticle model constructed from MD simulation, we calculated the average coordination number of both Pt-Pt and Pt-Zn, the bonding length of Pt-Pt and Pt-Zn in PtZn ordered alloy were taken as the cutoff radii for coordination number calculation.