Atomic control of active-site ensembles in ordered alloys to enhance hydrogenation selectivity

Intermetallic compounds offer unique opportunities for atom-by-atom manipulation of catalytic ensembles through precise stoichiometric control. The (Pd, M, Zn) γ-brass phase enables the controlled synthesis of Pd–M–Pd catalytic sites (M = Zn, Pd, Cu, Ag and Au) isolated in an inert Zn matrix. These multi-atom heteronuclear active sites are catalytically distinct from Pd single atoms and fully coordinated Pd. Here we quantify the unexpectedly large effect that active-site composition (that is, identity of the M atom in Pd–M–Pd sites) has on ethylene selectivity during acetylene semihydrogenation. Subtle stoichiometric control demonstrates that Pd–Pd–Pd sites are active for ethylene hydrogenation, whereas Pd–Zn–Pd sites show no measurable ethylene-to-ethane conversion. Agreement between experimental and density-functional-theory-predicted activities and selectivities demonstrates precise control of Pd–M–Pd active-site composition. This work demonstrates that the diversity and well-defined structure of intermetallics can be used to design active sites assembled with atomic-level precision. Advances in the design of heterogeneous catalysts are limited by our ability to synthesize atomically precise active-site ensembles. Now, the controlled synthesis of Pd–M–Pd catalytic sites (M = Zn, Pd, Cu, Ag and Au) has been demonstrated. Stoichiometric control identifies that Pd–Pd–Pd sites are active for ethylene hydrogenation, whereas Pd–Zn–Pd sites are not.

A dvances in heterogeneous catalyst active-site design are limited by our ability to exert atomically precise synthetic control over active-site ensembles 1 . Tailoring the electronic and geometric structure of the active site has mainly been accomplished by carefully tuning bimetallic alloys, in which the addition of the second metal modifies the electronic and structural properties of the primary reactive metal. Development of model bimetallic catalysts has traditionally relied on vapour deposition of a second (minority) metal onto a single crystalline form of the host metal, followed by annealing treatment to form a bimetallic surface [2][3][4][5] . Vapour deposition methods lead to a distribution of ensemble sizes of ill-defined geometry and composition, and large uncertainty in surface composition due to redistribution of the deposited metal between the surface and subsurface post-annealing.
A more recent application of this approach, in which low loadings of a second metal are physically deposited on a well-defined surface of a second metal, has led to a class of 'single-atom alloy' (SAA) catalysts [6][7][8] . As the name implies, SAAs may possess well-defined active sites consisting of isolated metal atoms in an inert or less active host metal. Sykes and coworkers demonstrated the enhanced reactivity of Cu single crystals for acetylene hydrogenation after doping with single Pd atoms 7 . The authors proposed that Pd enabled the facile dissociation of hydrogen (compared with activated dissociation of H 2 on Cu). The atomic H bound to Pd subsequently spilled over to Cu, catalysing the hydrogenation of acetylene. Single-atom catalysts (SACs) are a related class of catalysts in which individual metal atoms are isolated on refractory or reducible oxides [9][10][11] . A major limitation of both SAAs and SACs is their inherent limitation to mononuclear assemblies (that is, single atoms) because site isolation is accomplished through low surface densities. Therefore, SAAs and SACs do not offer the ability to create well-defined homomultinuclear (M n , n > 1) or heteromultinuclear (M n M′, n > 1) assemblies.
Goodman and coworkers controlled the coverage of Pd atoms on Au single crystals to demonstrate precise spacing of monomer 'pairs' that facilitated synthesis of vinyl acetate monomer 6 . They deposited a low coverage of Pd atoms on Au(111) and Au(100) single crystals, which have different surface interatomic Au spacing, to control spacing between the deposited Pd atoms. The spacing of Pd atoms on Au(100) was sufficient to allow the Pd atoms to act as a monomer 'pair' capable of synthesizing vinyl acetate monomer, while Pd atoms on Au(111) were too distant to catalyse vinyl acetate monomer formation. This study controlled nuclearity and spacing of active metal atoms (Pd) within a relatively inert host (Au) to provide active and selective catalytic sites. These sites, however, were prepared through a statistical distribution of low-coverage Pd atoms rather than targeted synthesis of a specific active metal atom arrangement.
We demonstrate that the intermetallic γ-brass phase of Pd-Zn can be used to directly synthesize both well-defined homomultinuclear and heteromultinuclear active sites. Intermetallic compounds contain at least two metals with a well-defined crystal structure with fixed atom positions and site occupancies, leading to long-range order; this offers distinct advantages over SACs and SAAs for controlling active-site ensembles beyond single, isolated atoms. The well-defined crystal structure and long-range atomic ordering of intermetallic compounds offer consistent site isolation throughout the catalyst, as well as control over the number of atoms per active site. Beyond this exquisite control over site nuclearity (n > 1), homo-and heteromultinuclear sites can also be prepared with high density on the surface; no comparable synthetic methods exist for SAAs or SACs.
Intermetallics demonstrate distinct catalytic properties from their corresponding monometallic catalyst for many reactions including semihydrogenation of alkynes 12 . Armbrüster and coworkers demonstrated Pd-Ga intermetallics have high acetylene semihydrogenation selectivity compared to monometallic Pd [13][14][15][16][17] . The selectivity is believed to result from the isolation of the active sites, which control the adsorption energy and configuration of acetylene 14 . Associated electronic effects due to the alloying event were found to play only a minor role in controlling selectivity 18 . In addition, the presence of the Ga prevented PdH formation 19 . Li et al. studied MgO-supported Ni-Ga intermetallics and found that selectivity towards phenylacetylene semihydrogenation was much higher than with pure Ni 20 . Intermetallic Pd-Zn catalysts expose a chequerboard arrangement of alternating Pd and Zn atoms, which exhibited a distinct selectivity towards acetylene semihydrogenation 21 . The high selectivity at high conversion (90% C 2 H 4 at nearly 100% C 2 H 2 conversion) was attributed to the presence of these isolated Pd atoms, which induced C 2 H 4 to bond in a weakly bound π-bonded configuration. In their comparison of supported Pd-In catalysts, Feng et al. found the selectivity increased from 21% to 92% as the Pd/In ratio shifted from 3:1 to 1:1, suggesting that isolation of Pd was critical to high selectivity 22 . These examples demonstrate that intermetallics offer more precise control over active-site local composition than vapour-deposited bimetallic systems. In the referenced studies, however, control over the geometry and composition of the active-site ensembles was limited to isolated single Pd or Ni atoms. Previous work has demonstrated active assemblies of n atoms (n = 3, trimers) in Pd 3 In(111) (ref. 22 ), Pt 3 Sn(111) and Pt 2 Sn(111) (ref. 23 ), but these 'trimer' active sites are interconnected rather than being isolated by In or Sn atoms. A combined surface-science/computational examination of PdGa(111) demonstrated the presence of isolated Pd trimers, but their reactivity and selectivity during semihydrogenation were not reported 24 . To date, there has been no demonstration of the ability to synthesize isolated active-site ensembles composed of more than one catalytic active metal, to the best of our knowledge.
Here we utilize the intermetallic (Pd, Zn) γ-brass phase to control the geometry of isolated, active catalytic ensembles dispersed within an inert matrix, as well as the composition of isolated multinuclear active sites. The γ-brass structure is a 52-atom unit cell with four symmetry-inequivalent positions: outer tetrahedral (OT), inner tetrahedral (IT), octahedral (OH) and cuboctahedral (CO; Fig. 1a) 25 . This precise and homogeneous metal ensemble distribution is maintained at the surface, exposing isolated OT-OH-OT catalytic ensembles. The occupation of the OT-OH-OT ensemble is determined by the composition of the γ-brass phase. The (Pd, Zn) binary γ-brass phase exposes single-atom Pd 1 monomers where Pd atoms are separated by zinc (Pd-Zn-Pd) and elongated Pd 3 trimer (Pd-Pd-Pd) ensembles (insets in Fig. 1b

Results and discussion
Ethylene hydrogenation activities on Pd-Zn binary γ-brass intermetallics. Unsupported Pd-Zn γ-brass catalysts were synthesized using a well-established bulk synthesis technique and characterized by inductively coupled plasma optical emission spectroscopy; scanning electron microscopy (SEM) and energy dispersive X-ray spectroscopy; and X-ray diffraction to ensure phase purity and bimetallic composition. Brunauer-Emmett-Teller surface area was determined to calculate and compare normalized reaction rates. Small changes in the Pd-Zn ratio result in enormous differences in ethylene hydrogenation activity, which we establish (below) is due to transition from inactive Pd 1 monomers to active Pd 3 trimers. As shown in Fig. 1b (Fig. 1c). However, a Pd-Zn γ-brass catalyst with 15.4 at.% Pd (22.8 wt%) shows no measurable activity for ethylene hydrogenation. Based on a conservative flame ionization detection limit, the hydrogenation rate over the 15.4 at.% Pd catalyst was less than 10 −13 mol m −2 s −1 , at least four orders of magnitude lower than the rate measured over the 17.3 at.% Pd catalyst at 307 K.
DFT calculations indicate Pd 1 sites are inactive for ethylene hydrogenation whereas Pd 3 sites demonstrate measurable activity. The {110} facet has two equally stable terminations (depicted here as (110) and (110) by virtue of the four-fold crystal symmetry of the γ-brass phase), one having Pd 3 sites and the other only Pd 1 sites. From a Wulff construction of a particle, ~45.5% of the exposed surface area contains trimer sites (in addition to monomer sites, which decrease in number per unit cell with Pd addition; Supplementary Tables 5 and 6). d, Relative energies of restructured Pd 9 Zn 43 (110) surfaces relative to the bulk-terminated, Pd 3 -trimer-containing surface. Pd 4 is constructed by swapping a subsurface Pd with a Zn in the surface. Pd 5 is constructed by swapping two Pd monomers on the surface with Zn atoms adjacent to the Pd 3 trimer. Pd 6 is constructed by swapping Pd from the subsurface and two Pd monomers on the surface with Zn atoms adjacent to the Pd 3 trimer. Inset images show the surface termination after each swap that was considered. An enlarged version of a ( Supplementary Fig. 2) is provided in Supplementary Section 2. The fit parameters for the refined X-ray diffraction data for Pd 9 Zn 43 (and the other γ-brass Pd-Zn intermetallics) are summarized in Supplementary Table 2. cell. All three γ-brass phase compositions show substantially weaker ethylene binding than Pd(111) due to the significant downshift in Pd-projected d-band centre (-1.36 eV for Pd(111) versus -1.96 to -2.02 eV for surface Pd atoms in 8-Pd and 9-Pd; Extended Data Fig.  1). Pd 1 monomers and Pd 3 trimer sites have very similar d-band centres, indicating that differences in their catalytic chemistry are dictated by ensemble rather than electronic effects. Zn surface atoms are inert to ethylene hydrogenation, as both ethylene and hydrogen binding are endothermic on Zn sites (Supplementary Fig. 13). The Pd 1 monomer site, exposed on the surface of all three γ-brass compositions, does not provide a sufficient Pd ensemble to bind ethylene, activate H 2 and form the ethylene hydrogenation transition state. This leads to a hydrogenation barrier significantly exceeding the ethylene desorption energy, a reliable descriptor for ethylene hydrogenation activity 26 . By contrast, the Pd 3 trimers in 9-Pd and 10-Pd provide a sufficient active-site ensemble and, therefore, the hydrogenation barrier is similar to the desorption energy and the catalyst is active for ethylene hydrogenation. Differences in relative values of the ethylene desorption energy and hydrogenation bar-rier among catalysts have been used previously as a descriptor for acetylene semihydrogenation selectivity 26 . Trends among similar surfaces should be considered given the strong exchange-correlation functional dependence of adsorption energies and the embedded assumption that pre-exponential factors for these two steps do not differ among catalysts.
The lack of ethylene hydrogenation activity on the 8-Pd catalyst is due to the complete isolation of all Pd atoms as Pd 1 monomers, surrounded by all Zn nearest neighbours. This catalyst exposes 'single-atom' Pd sites despite a relatively high 15.2 at.% Pd due to the intermetallic nature of the γ-brass Pd-Zn structure. The π-bound ethylene on Pd 1 monomers blocks H co-adsorption and formation of the hydrogenation transition state. The drastic increase in ethylene hydrogenation activity on 9-Pd occurs due to the formation of Pd 3 trimers, also isolated from other surface Pd atoms by Zn nearest neighbours. Further increasing the Pd content to 10-Pd doubles the ethylene hydrogenation rate without changing the apparent activation barrier (Fig. 1c), suggesting that the doubling of Pd 3 ensembles doubles the surface density of active sites. The apparent activation     [3][4][5], forming either one Pd 3 (9-Pd) or two Pd 3 (10-Pd) trimers per unit cell, which are exposed on the (110) facet (Fig. 2a) of the material. A first-principles-based cluster expansion method (CEM) 28 was employed to confirm the lowest-energy configuration among all the independent configurations (the total number of configurations is 2 26 or 6.7 × 10 7 ) of the 8-Pd and 10-Pd γ-brass phases using a 26-atom sublattice model of (Pd, Zn) OT 4 (Pd, Zn) IT 4 (Pd, Zn) OH 6 (Pd, Zn) CO 12 ( Fig. 2b; more details are in Supplementary Section 9). The CEM demonstrated that the lowest-energy configuration for 8-Pd places all eight Pd in OT sites, whereas for 10-Pd, eight Pd occupy OT sites and two Pd occupy OH sites. These results agree with our DFT calculations ( Supplementary Figs. 3 and 4) and X-ray diffraction Rietveld refinement results. In contrast to our previous research on the Ni-Zn γ-brass phase 29 , DFT calculations for the Pd-Zn γ-brass phase indicate that a trimer containing (110) surface termination is energetically stable (Supplementary Tables 5 and 6) and substantially (~50% area basis) exposed in the Wulff particle construction (Fig. 2c) for Pd-Zn γ-brass materials. DFT was also used to consider the possibility of surface reconfiguration in which Pd surface aggregation (Pd monomers exchanged with Zn to expand the Pd trimer ensemble) and subsurface-to-surface exchange (subsurface Pd atoms exchanged with Zn to expand the Pd trimer ensemble) were considered. All reconfigurations were found to have highly unfavourable formation energies because these restructuring events require Pd atoms to reside in Zn-occupied surface lattice positions-an inherently unstable local atomic composition ( Fig. 2d; Supplementary Fig. 11 shows that acetylene adsorption does not induce Pd aggregate formation). Thus, the combined experimental, CEM and DFT calculations demonstrated that Pd-Zn intermetallics with compositions ranging from 8-Pd to 10-Pd could be synthesized with high phase purity and well-defined lattice siting. DFT calculations further confirmed that Pd 3 trimers were stable and present on (110), a predominant exposed facet.

Influence of active-site nuclearity (Pd-M-Pd trimer) on catalytic semihydrogenation.
Acetylene hydrogenation serves as a probe reaction to demonstrate how the relative density of isolated Pd sites impacts both hydrogenation activity and selectivity. Our experimental studies indicate Pd 1 monomers hydrogenate C-C triple bonds at a rate two orders of magnitude slower than that of Pd 3 trimers. Computational studies demonstrate the central atom of the Pd-M-Pd active ensemble plays a controlling role in governing catalyst performance. DFT calculations predict that, just as for Pd 9 Zn 43 , replacement of a single Zn atom in Pd 8 Zn 44 by a coinage metal atom results in the formation of stable Pd-M-Pd ensembles (M = Cu, Ag, Au) both in the bulk and, more importantly, at the surface. Further, the catalytic performance (acetylene-ethylene competitive hydrogenation) of Pd-M-Pd ensembles is between Pd 1 and Pd 3 .
Experiments confirm these predictions. As shown in Fig. 3a, acetylene hydrogenation activity increased by over two orders of magnitude from 8-Pd to 9-Pd, indicating a distinct change in active-site morphology, which we attribute to the transition from Pd 1 to Pd 3 active sites. From 9-Pd to 10-Pd the areal rate doubles, due to the doubling of Pd 3 sites (as for ethylene hydrogenation), which overshadows any kinetic impact of the corresponding decrease in the number of Pd 1 sites. Pd foil exhibits an order of magnitude higher rate per area than 10-Pd and significantly lower ethylene selectivity. The 8-Pd, consisting of only Pd 1 monomers on the surface, demonstrates high selectivity towards ethylene due to its inactivity to hydrogenate ethylene, as previously discussed.
DFT calculations provide elementary reaction energetics that agree with the experimentally observed impact of Pd n nuclearity  Fig. 3d). The acetylene hydrogenation barrier is similarly tunable as shown in Fig. 3c. DFT results, therefore, predict that the addition of coinage metals in Pd-M-Pd sites will lead to acetylene hydrogenation activity and selectivity values intermediate to 8-Pd and 9-Pd. DFT calculations of reaction energies clearly demonstrate that the nuclearity of the Pd-M-Pd active site can finely tune activity and selectivity. We demonstrate that these predictions can be confirmed experimentally utilizing a synthetic approach with precise control of active-site composition.
Solid-state diffusion in a flame-sealed and evacuated quartz ampoule at 800 °C was used to prepare ternary γ-brass intermetallics. X-ray diffraction confirms that phase pure (Pd, M, Zn) (M = Cu, Ag, Au) γ-brass materials can be synthesized ( Supplementary Fig. 1). Using similar Rietveld refinement and DFT calculations, as described earlier (Supplementary Sections 2-5, Tables 3 and 5 and Figs. 6-10) for (Pd, Zn), it is further confirmed that Pd-M-Pd active sites form in (Pd, M, Zn) γ-brass materials because (1) M preferentially replaces Zn from OH sites and (2) exposure of Pd-M-Pd trimers at the surface is energetically favourable in all these materials.
We used 13 C 2 -ethylene and 12 C 2 -acetylene to independently track conversion and allow for rigorous quantification of semihydrogenation selectivity during the hydrogenation of acetylene in excess ethylene (Supplementary Section 10). Pd 8 Fig. 4a) similar to ethylene selectivity for non-competitive acetylene hydrogenation (Fig. 3a). More importantly, the net semihydrogenation selectivity (equal to [  ; Fig. 4a) on Pd 3 trimers on Pd 9 Zn 43 is far below its intrinsic value because of the simultaneous hydrogenation of 13 C 2 -ethylene. Net ethylene selectivity for Pd 8 Zn 44 is within ~15% of its intrinsic value (and far superior to Pd 9 Zn 43 ) even at ~100% acetylene conversion (Fig. 4b). The single Pd atoms cannot supply hydrogen to C-C double bonds because π-bound ethylene blocks H 2 dissociative adsorption, H co-adsorption and C=C hydrogenation, whereas acetylene di-σ-binding across a Pd monomer 'pair' allows H co-adsorption and C≡C hydrogenation. Pd 8   for activity and E act C 2 H 4 hydro − E C 2 H 4 des for selectivity), where E act C 2 H 2 hydro is the activation energy of the first hydrogenation step, E C 2 H 4 des is the desorption energy of ethylene, R is the ideal gas constant and T is absolute temperature, with experimental performance (Fig. 5). The first acetylene C-H formation barrier is a reasonable descriptor for acetylene hydrogenation activity given the strong adsorption of acetylene. The difference between the ethylene hydrogenation barrier and desorption energy is used as a selectivity descriptor, as applied previously 26,30 . Both activity and selectivity show strong agreement between experimental kinetics and DFT, thus providing evidence of excellent control of these multinuclear, heteroatomic active ensembles through subtle stoichiometric tuning and introduction of an additional metal atom.
In conclusion, we demonstrate that the (Pd, M, Zn) γ-brass crystal structure can be manipulated to reliably access well-defined, isolated, catalytically relevant three-atom active sites of the form Pd-M-Pd that are distinct among themselves and from fully coordinated Pd sites. These model surfaces demonstrate the impact of active atom nuclearity on catalytic performance in Pd-catalysed hydrogenation chemistry. These precise Pd-M-Pd ensembles are experimentally easy to access and convenient for DFT modelling due to their single-crystal-like nature to enable probing the impact of active-site nuclearity on catalytic activity and selectivity. From this work, the use of intermetallic compounds can be extended to create active sites that are potentially tuned to the molecular footprint of any hydrocarbon reactant or product of interest. Such materials will be key in the pursuit of atomically precise active-site generation for selective conversion with high atom efficiency.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41557-021-00855-3.