Low-nuclearity Pt-Mo ensembles with defined carbon hosts. The NC hosts with tunable N defects were prepared by the pyrolysis of self-made polyaniline derived from the established oxidative polymerization method49. The N contents of 2.06 and 5.93 wt.% were obtained by moderating the annealing temperatures at 1073–1473 K (Table S2). Another NC host with a higher N content of 9.4 wt.% was derived from our recently developed “ring opening-pyrolysis” approach53. These hosts were hereafter coded as NCx (x = 0.02, 0.07, and 0.13), wherein x denoted the molar N:C ratios that were determined by both elemental analysis and X-ray photoelectron spectroscopy (XPS, Table S2). N2 sorption isotherms of NCx presented the H4 shapes typical for the micropore-rich carbon-based materials with the specific surface areas (SBET) of 464–571 m2 g− 1 (Fig. S1). In addition, transmission electron microscopic (TEM) analysis revealed the worm-like morphologies of NCx, and the presence of more graphitic structures at the edges on NC0.02 (Fig. S2), hinting the higher graphitization probably due to the elevated pyrolysis temperature applied. The as-prepared NCx were adopted to accommodate the mono- and bimetallic Mo and/or Pt species following the strategies presented in Fig. 1. The mono-metallic catalysts were prepared by incipient wetness impregnation and subsequent calcination and reduction, while for the bimetallic analogues, sequential impregnation and calcination of the Mo and Pt precursors were applied before the final reduction. Details on the synthesis and all characterization techniques employed are provided in the Methods. Inductively coupled plasma-optical emission spectroscopic (ICP-OES) analyses confirmed close contents to the nominal set values for both metals (Table S2), and similar hysteresis loops to those of NCx were verified after deposition of the metals (Fig. S3). The powder X-ray diffraction (PXRD) patterns exhibited two broad peaks related to the amorphous carbon hosts and only a fainted peak at 2θ = 39.4o corresponding to the Pt(111) facet was detected for Pt-Mo/NC0.02 (Fig. S4), suggesting the high dispersion of these metals. Agreeing with these results, no clear nanoparticles could be observed by TEM on all the catalysts, except for Pt-Mo/NC0.02 (Fig. S5). High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) was adopted to visualize the metal dispersion (Fig. 2a-e and Fig. S6). Many white dots decorating on the grey backgrounds were observed for the mono-metallic catalysts. Statistic counting above 500 ensembles suggested the average sizes of 0.13 ± 0.11 and 0.22 ± 0.14 nm, respectively, for Pt/NC0.07 and Mo/NC0.07, hinting predominantly atomic-level dispersion of the metals. The role of the nitrogen defects in tuning the ensemble sizes was exemplified in the bimetallic series. Pt-Mo/NC0.07 showed broadened size distribution centered at 0.25 nm with the formation of more clusters, as compared with the mono-metallic counterparts. These ensemble sizes were further tunable within the subnano regime (between 0.17 ± 0.09 and 0.66 ± 0.76 nm, for Pt-Mo/NC0.13 and Pt-Mo/NC0.02, respectively) by simply varying the numbers of the N defects. The high density of N defects was likely responsible for maintaining the atomic dispersion for Pt-Mo/NC0.13. On the contrary, elemental color mapping of Pt-Mo/NC0.02 with reduced N defects revealed more homogeneous distribution of C, N, and Mo than Pt, unambiguously pointing to the clustering of the Pt atoms (Fig. S7). By adopting a spherical model with Pt radius of 0.139 nm, the average numbers of the atoms within the clusters were estimated to be 1 ~ 4 and 13 ~ 133, respectively, for Pt-Mo/NC0.07 and Pt-Mo/NC0.02. Considering the high-temperature reduction treatment (973 K), our approach provides a straightforward strategy to design highly sintering-resistant low-nuclearity bimetallic ensembles with tunable and narrow size distributions.
X-ray absorption spectroscopy (XAS) in junction with XPS was further applied to unveil the electronic and geometric properties of the developed key materials. Comparison of the normalized X-ray absorption near edge structures (XANES) revealed quite different features between the mono-metallic catalysts and the respective metal foils (Fig. 2f,g). Pt/NC0.07 displayed a pre-edge at 11570.3 eV with an upshift of 2.1 eV as compared to Pt foil and an increased whiteline intensity (1.61 vs. 1.36, Fig. 2f). Mo/NC0.07 also exhibited a pre-edge at 20006.3 eV that was absent for Mo foil (Fig. 2g). These results confirmed the ionic feature of the metal species, i.e., Pt2+ and Mo6+, which were further corroborated by the XPS showing perfect doublets centered at 72.5 and 232.6 eV, respectively, for Pt 4f7/2 and Mo 3d5/2 (Fig. S8). Fittings of the Fourier transformed extended X-ray absorption fine structures (EXAFS) of the mono-metallic catalysts indicated the predominant Pt − N scattering path for Pt/NC0.07 and both Mo − O and Mo − N contributions for Mo/NC0.07 (Fig. 2h,i and Table S3). In comparison to the mono-metallic catalysts, the pre-edge of Pt-Mo/NC0.07 in Pt 4f XANES spectrum was downshifted by 0.8 eV with a slightly reduced whiteline intensity while the Mo 3d XANES spectrum remained essentially identical. Besides, the Pt 4f and Mo 3d XPS spectra also only slightly downshifted by 0.1 eV. These observations jointly suggested the ionic nature of both metals in the bimetallic catalyst, probably increased electron density for the Pt species. Indeed, the Pt 4f FT EXAFS spectrum of Pt-Mo/NC0.07 differed greatly from that of Pt/NC0.07 with the evidenced scattering paths of Pt − Mo and Pt − Pt bonds, which therefore explained the increased electron density of the Pt species and their agglomeration phenomenon.
Nitrogen defects played an important role in anchoring the metal sites, which was evidenced by the presence of Pt − N bonds detected in mono- and bimetallic catalysts as well as the increasing sizes of the bimetallic Pt − Mo ensembles at decreasing N dopant. To further discriminate the contribution of different N functionality, the N 1s XPS spectra of Pt-Mo/NCx was assessed (Fig. S9). Spectra deconvolution revealed the increasing shares of planar N functionalities (both pyridinic- and pyrrolic-N) at higher doping levels. This well corroborated with the inversed trend in the ensemble sizes of the supported catalysts, thus suggesting the critical roles of planar N defects in stabilizing the metal species. The possible configuration of the metal ensembles in Pt-Mo/NC0.07 was then explored by DFT simulations. Based on the above thorough characterizations, ten different optimized stable models were derived (Fig. S10) considering i) the maximal metal numbers of four (Pt2Mo1 or Pt3Mo1), and the presence of ii) Pt-Mo bonds and iii) Pt-N bonds related to sp2-hydridized N defects (both pyridinic- and pyrrolic-N). Comparison on the formation energy of the metal clusters and their binding energy on various defects revealed that the pyridinic-N-stabilized Pt3Mo1N3 with a stable tetrahedron configuration of the metal atoms possessed the highest formation energy of − 16.84 eV and a high binding energy of − 7.14 eV. Therefore, Pt3Mo1N3 ensemble was later adopted for further mechanistic investigations (vide infra).
Catalytic performance in formic acid decomposition. The performance of the developed low-nuclearity catalysts in the continuous gas-phase decomposition of formic acid was then evaluated in a home-made fixed-bed quartz reactor at ambient pressure (Scheme 1). The catalytic data were first collected in a temperature-ramping mode from 353–523 K at a total gas hourly space velocity (GHSV) of 15000 cm3 gcat.−1 h− 1. At first, the performance of the NC0.07-supported catalysts was compared, which clearly demonstrated the superior activity of Pt-Mo/NC0.07 over Pt/NC0.07 and Mo/NC0.07 (Fig. 3a). Namely, full conversion was reached at a low temperature of 398 K over Pt-Mo/NC0.07, whereas it was delayed to 423 K on Pt/NC0.07 and avove 523 K on Mo/NC0.07. Meanwhile, the CO2 selectivity remained nearly 100% over all the three catalysts at < 423 K, with trace CO (selectivity < 1%) formed on Pt-Mo/NC0.07 but only at elevated temperatures. These promising results strongly hinted the synergistic effect between the Pt and Mo species in the bimetallic catalyst. To verify this point, another contrast experiment was performed, wherein two physically mixed mono-metallic catalysts (Pt/NC0.07 + Mo/NC0.07) comprising of the same metal loadings as those of the bimetallic analogue were packed into the reactor and evaluated. This mixture exhibited similar activity to that of Pt/NC0.07 but with poorer CO2 selectivity at > 423 K. The contrast test thus highlighted the importance of intimate contact and/or certain geometric structures between the Pt and Mo species in boosting the reaction rate. Next, the impact of the bimetallic ensemble sizes on the catalytic performance was examined by using the Pt-Mo/NCx series (Fig. 3b). All these catalysts exhibited similar activity profiles with steep curves observed between 373–423 K, but the temperatures corresponding to full conversion were delayed for the other two catalysts as compared with that of Pt-Mo/NC0.07. Furthermore, the CO2 selectivity as a function of the temperature on Pt-Mo/NC0.13 was close to that on Pt-Mo/NC0.07, while CO was formed already at 398 K and increased more rapidly at higher temperatures on Pt-Mo/NC0.02. Our results suggested a critical size in the subnano regime for the maximized activity, which was likely linked to the unique geometry of Pt-Mo/NC0.07 as simulated by DFT.
Given the remarkable low-temperature activity of Pt-Mo/NC0.07, the long-term stability performance at 373 K was further evaluated in a 50 h time-on-stream (Fig. 3c). The activity gradually increased after a few hours of stabilization and slightly fluctuated at ca. 90–96%, while CO2 was the only detectable product in our gas chromatography. Pt-Mo/NC0.07 after the stability test was thoroughly characterized by different techniques. PXRD analysis revealed the same amorphous nature of the spent catalyst and no diffractions of Pt- and/or Mo-related compounds were detected (Fig. S11), suggesting that these metal species remained highly dispersed. This was further corroborated by HAADF-STEM observations, showing almost homogeneous distribution of small ensembles (the white spots) scatting around the hosts (Fig. S12). Statistic counting revealed very close metal ensemble sizes as compared to those of the fresh catalyst (0.23 ± 0.21 vs. 0.22 ± 0.14 nm), thus indicating the high stability of the bimetallic ensembles. A further survey of the core-level XPS spectra of Pt 4f, Mo 3d, and N 1s also revealed negligible differences after catalysis (Fig. S13). All these characterizations thus demonstrated the superior robustness of the bimetallic Pt-Mo catalyst.
To better evaluate the potential of Pt-Mo/NC0.07, a detailed comparison of the key performance descriptors with those of previously reported catalysts for gas-phase formic acid decomposition was made taking into account of the stability and kinetic studies (Fig. 3d, Fig. S14, and Table S1). Pt-Mo/NC0.07 possessed a moderate apparent activation energy (Ea) of 54 kJ mol− 1, which was much lower than those of Pt/NC0.07 and Pt-Mo/NC0.02 (65 and 91 kJ mol− 1, respectively). Notably, the reaction rates of Pt-Mo/NC0.07 at 373 and 388 K reached 0.31 and 0.62 molHCOOH molPt−1 s− 1, respectively, significantly outperforming the-state-of-the-art precious metal-based catalysts by ca. one order of magnitude. This remarkable utilization efficiency of precious metals reflected the promising industrial prospect. Furthermore, owing to the maximized exposure of the Pt atoms, Pt-Mo/NC0.07 also stood out as the most active catalysts even by comparing the turnover frequency (TOF) based on the active surface metals. In addition, the full H2 selectivity and no activity deterioration in the long-term evaluation further highlighted the outstanding performance.
Kinetic insights from operando dual-beam Fourier transform infrared spectroscopy
To shed light on the high performance of Pt-Mo/NC0.07, the adsorption of formic acid molecules on the key catalysts was studied by employing a dual-beam Fourier transform infrared spectrometer (DB-FTIR). The advantages of DB-FTIR over conventional single-beam FTIR could be explained by the fact that the former can simultaneously collect the catalyst sample and the reference spectra. As such, it could often provide more precise structural fingerprints and high-quality spectra, as have been demonstrated in the previous works54–57. To our delight, this superior technique was able to tackle the adsorption of formic acid on our carbon-based materials that were otherwise extremely challenging for the conventional IR. The adsorption of pure formic acid on the selected catalysts at room temperature was first studied (Fig. S15). Pt/NC0.07 with predominant Pt single atoms exhibited two weak adsorption bands centered at ca. 1718 and 1595 cm− 1, which were assigned to the C − O vibration of the molecularly adsorbed HCOOH (HCOOHad) and the O − C−O vibration due to the adsorbed formate species (HCOOad)12,29. These bands were also detected on Mo/NC0.07 and Pt-Mo/NC0.07 but with much stronger intensities, suggesting favorable adsorption of HCOOH on the latter. By comparing the intensities of these typical bands (I1595:I1718), one can find a much higher ratio for Pt-Mo/NC0.07 over Mo/NC0.07. This might suggest the higher dissociation propensity of formic acid molecules on Pt-Mo/NC0.07. In contrast, Pt-Mo/NC0.02 did not exhibit obvious spectroscopic features at 1800 − 1400 cm− 1.
To acquire in-depth kinetic insights, operando DB-FTIR was performed to study the adsorption and activation of HCOOH on Pt-Mo/NC0.07 as well as the respective mono-metallic catalysts (Fig. 4). Besides the two most intense bands at ca. 1718 and 1595 cm− 1, another two bands at 1340 and 1190 cm− 1 corresponding to the symmetric O − C−O vibration in the HCOOad species and the C − O vibration in molecular HCOOH12,29 were detected at room temperature on all the three catalysts (Fig. 4a). These bands gradually attenuated to different degrees at increasing temperatures. As exemplified in the zoomed spectra (Fig. 4b), the bands at ca. 1718 cm− 1 gradually disappeared on all the catalysts, whereas the band at 1595 cm− 1 showed divergent evolution on the different catalysts. These bands greatly attenuated at high temperatures for Mo/NC0.07 and Pt-Mo/NC0.07 but remained essentially stable on Pt/NC0.07. To better compare these features, the normalized band intensities over different catalysts were presented (Fig. 4c,d). The band related to HCOOHad at 1718 cm− 1 disappeared at a comparably low temperature of 405 K on Pt/NC0.07 and Pt-Mo/NC0.07 and at about 473 K on Mo/NC0.07. On the contrary, the desorption of HCOOad at 1595 cm− 1 remarkably decreased with temperature on Pt-Mo/NC0.07 and Mo/NC0.07, while on Pt/NC0.07 it only mildly decreased by about one quarter at 473 K. Furthermore, an additional band at 2190 cm− 1, probably associated with CO vibration at electron-deficient Mo species58, gradually built up at higher temperatures on the bimetallic catalysts but was absent on the other two mono-metallic catalysts. This hinted the accelerated dehydration of HCOOH on the bimetallic catalyst, thus agreeing well with their catalytic testing results. In general, the above DB-FTIR study evidenced the different kinetic fingerprints of formic acid molecules with the distinct mono- and bimetallic sites as illustrated in Fig. S16. On the representative single-atom Pt sites, strong adsorption of formic acid was observed, but the deprotonation was much difficult. The mono-metallic Mo catalyst also showed a higher propensity toward HCOOH adsorption and its dissociation but displayed the poorest decomposition activity. This suggested that other fundamental steps, such as the abstraction of another H atom, might be more energy-demanding than the dissociation of HCOOH. Another explanation could be the too strong adsorption of HCOOH on the Mo6+ sites as indicated by the much higher desorption temperatures in the operando DB-FTIR study. In contrast, these two fundamental steps of formic acid adsorption and dissociation were both more favorable on the low-nuclearity Pt-Mo ensembles, thus accounting for the highest activity.
Mechanistic insights from density functional theory calculations
Platinum has been recognized as one of the best components among the platinum-group metals for gas-phase decomposition of formic acid. The seminal works by Bulushev et al. pointed at the geometric and electronic effects of N-doping of the carbon hosts on the Pt nanoclusters (1-2.3 nm) 31,32, leading to significantly improved decomposition activity. In particular, they found that the single Pt atoms anchored by a pair of pyridinic N exhibited superior catalytic performance than the analogues supported on the N-free hosts. In our study, the impact of molybdenum modification on the catalytic performance of different Pt ensembles in the sub-nano region, induced by defect-driving nanostructuring, was disclosed. The promotional effect was most prominent on the low-nuclearity Pt than the single atoms and bigger nanoclusters. In addition, our contrast experiment demonstrated the importance of intimate contact between Pt and Mo for efficiently activating formic acid molecules. These results thus highlighted the size effect and the synergistic catalysis of the Pt-Mo ensembles.
Since our operando DB-FTIR study evidenced the different interactions of formic acid on the distinct mono- and bimetallic sites, to gain further insights into the Pt-Mo synergy in Pt-Mo/NC0.07, the reaction mechanisms of HCOOH decomposition on different model catalysts were studied by DFT simulations. To examine the nuclearity effects of the Pt catalysts, different model systems were considered: Pt1N4 (single Pt atom stabilized by four pyridinic N in a planar mode), Pt3Mo1N3, and Pt(111). We first compared the adsorption and dissociation energies of HCOOH on these model systems, both displaying the same trends with the order of Pt3Mo1N3 > Pt(111) > Pt1N4 (Fig. S17). This suggested that the activation of HCOOH on Pt3Mo1N3 was thermodynamically favorable over the other two systems. Furthermore, the positive dissociation energy on Pt1N4 hinted the difficulty in deprotonation of HCOOH molecules. These results were therefore fully consistent with the DB-FTIR observations. The reaction paths of HCOOH decomposition on these catalyst models were further simulated (Fig. 5 and Figs. S18,19). The reaction coordinate on Pt(111) showed a small energy barrier of 0.30 eV for the adsorption of HCOOH, which underwent an exothermic H dissociation step to generate surface *HCOO species adsorbed in a bridge configuration. The dissociation of the second H in *HCOO for CO2 release was most energy-demanding with a barrier of 1.55 eV. Lastly, the combination of two adsorbed H species to release molecular H2 needed to overcome another barrier of 1.12 eV. On the contrary, the deprotonation of HCOOH at Pt1N4 should overcome a very high energy barrier of 2.60 eV, suggesting that the activation of HCOOH at Pt1N4 was also kinetically unfavorable. To probe the Pt-Mo synergy at Pt3Mo1N3, two different reaction paths were considered. When HCOOH was first adsorbed at the Pt site (Fig. S19), the H atom in the hydroxyl groups was subtracted and bonded on the Mo sites, accompanied with the adsorption of *HCOO through two O atoms at the top Pt sites. These steps were strongly exothermic (− 2.73 eV) without any barriers. The dissociation of the second H atom in *HCOO was most energy-demanding (2.70 eV), which needed to fist break one Pt − O bond, and then transfer the second H to the Mo sites with the release of CO2. In addition, the combination of *H should surpass another barrier of 1.17 eV. The significantly higher energy barrier than that of Pt(111) was contradictory to the catalytic activity of the low-nuclearity Pt-Mo/NC0.07 and thus suggested the possibility of other paths. Therefore, HCOOH firstly adsorbed at the Mo sites was considered (Fig. 5). The dissociation of the H atom in the hydroxyl groups was spontaneous and thermodynamically more favorable than on Pt(111) (reaction heat 0.76 vs. 0.40 eV). In this case, the H atom was bonded with two Pt atoms while *HCOO was adsorbed at the Mo and the top Pt sites through bonding with two O atoms. This unique configuration was quite beneficial for the activation of the second H in *HCOO, wherein the H atom was transferred to the top Pt sites with a lower barrier as compared with that on Pt(111) (1.18 vs. 1.55 eV). Based on the previous works on Pt single crystals59 and detailed kinetic and modeling study on Pd/C catalysts60, the dissociation of formic acid molecules into formate species was proposed to be the rate-determining step. Our DB-FTIR experiments and DFT modeling studies demonstrated the cooperative catalysis between Pt and Mo in Pt3Mo1N3 for lowering the overall energy barriers in HCOOH decomposition as compared with Pt1N4 and Pt(111), which might explain the superior catalytic performance.
In summary, we have successfully designed a new class of bimetallic platinum and molybdenum ensembles supported on nitrogen-doped carbon (NC) via a straightforward impregnation-reduction approach. Our defect-driven nanostructuring strategy coupling manipulation of the N defects and sequential metal depositions can systematically alter the geometric distribution of Pt species from single atoms to sub-nanoclusters. A high number of planar N defects was favorable for stabilizing the atomic dispersion of the Pt species, while accelerated agglomeration occurred at reduced defect numbers. The developed bimetallic catalysts together with the respective monometallic analogues were evaluated the continuous gas-phase decomposition of formic acid to generate CO-free hydrogen. The low-nuclearity Pt-Mo ensembles (Pt-Mo/NC0.07) displayed unprecedented high activity with a reaction rate of 0.62 molHCOOH molPt−1 s− 1, significantly outperformed i) the bimetallic analogues featuring single Pt atoms or bigger Pt sub-nanoclusters, ii) the respective mono-metallic catalysts, and iii) the state-of-the-art catalytic systems reported to date. More importantly, Pt-Mo/NC0.07 exhibited stable performance in a 50 hour time-on-stream without any apparent activity or selectivity deterioration, thus demonstrating the excellent structural robustness. The operando dual-beam Fourier transformed infrared spectroscopy coupling with density functional theory modeling jointly demonstrated that the superior catalytic performance of the low-nuclearity Pt-Mo ensembles rooted at the unique Pt3Mo1N3 configuration, over which the adsorption and dissociation of HCOOH were thermodynamically favorable. Furthermore, the synergistic catalysis between the Pt and Mo sites provided an alternative path for lowering the energy barriers for the consecutive HCOOH dissociations, which were otherwise most energy-demanding on the nanoparticle or single-atom Pt catalysts.