Supplemental information: Structural Trends in the Dehydrogenation Selectivity of Palladium Alloys

Alloying is well-known to improve the dehydrogenation selectivity of pure metals, but there remains considerable debate about the structural and electronic features of alloy surfaces that give rise to this behavior. To provide molecular-level insights into these effects, a series of Pd intermetallic alloy catalysts with Zn, Ga, In, Fe and Mn promoter elements was synthesized, and the structures were determined using in situ X-ray absorption spectroscopy (XAS) and synchrotron X-ray diffraction (XRD). The alloys all showed propane dehydrogenation turnover rates 5–8 times higher than monometallic Pd and selectivity to propylene of over 90%. Moreover, among the synthesized alloys, Pd3M alloy structures were less olefin selective than PdM alloys which were, in turn, almost 100% selective to propylene. This selectivity improvement was interpreted by changes in the DFT-calculated binding energies and activation energies for C–C and C–H bond activation, which are ultimately influenced by perturbation of the most stable adsorption site and changes to the d-band density of states. Furthermore, transition state analysis showed that the C–C bond breaking reactions require 4-fold ensemble sites, which are suggested to be required for non-selective, alkane hydrogenolysis reactions. These sites, which are not present on alloys with PdM structures, could be formed in the Pd3M alloy through substitution of one M atom with Pd, and this effect is suggested to be partially responsible for their slightly lower selectivity.


Supplemental methods
Catalyst Synthesis Pd A Monometallic 2% Pd catalyst was prepared by strong electrostatic adsorption (SEA). 5 grams of davasil 646 silica was dispersed in 50 mL of deionized (DI) water and the pH was adjusted to 11 with 32% ammonium hydroxide. 2.8g of 10% palladium (II) tetraammine nitrate solution was diluted to a total volume of 25 mL and the pH was adjusted to 11 by the addition of 32% ammonium hydroxide. The silica suspension and the palladium solution were mixed for 15 minutes. The silica was then allowed to settle out of solution and removed by filtration. The Pd-SiO 2 was then washed 3 times with DI water and dried at room temperature for 3 hours and then overnight at 125°C. After drying the catalyst was calcined at 300°C for 3 hours. The Pd-SiO 2 was then reduced in 5% H 2 in steps from 100°C to 250°C at a 2.5°C/min ramp rate with 15 minute dwells every 25°C and then a fast ramp (10 C/min) to 550°C and a 30 minute dwell. The catalyst was then cooled to room temperature and passivated in air. A second monometallic Pd catalyst was synthesized by the above method except the calcination temperature was 200°C for 3 hours and the mass of 10% palladium (II) tetraammine nitrate was adjusted to give a weight loading of 1%.

Pd-In
A 2% Pd 3% In catalysts was synthesized according to the procedure for Pd-In 0.8 in reference 28 .
Briefly, In-SiO 2 was synthesized by incipient wetness impregnation of an In(NO 3 ) 2 and Citric acid solution pH adjusted to 11 with ammonium hydroxide. The In-SiO 2 was dried at 125°C overnight and then calcined in air at 200°C. Palladium was then added by incipient wetness impregnation of a pH 11 solution of palladium (II) tetraammine nitrate. The Pd-In-SiO2 catalyst was then dried overnight at 125°C and calcined at 200°C. The catalyst was then reduced using a slow ramp to 200°C and then a fast ramp to 600°C in 5% H 2 (balance N 2 ). The catalyst was cooled to room temperature in nitrogen and then passivated in air.

Pd-Fe
A 2% Pd 3%Fe catalyst was synthesized according to the procedure for "Pd 3 Fe small" in reference 29 . A 2:1 molar ratio of citric acid to iron (III) nitrate nonahydrate was pH adjusted to 11 using concentrated ammonium hydroxide. The solution was impregnated dropwise onto 5g of SiO 2 and dried at 125°C and then calcined at 400°C for 3 hours. Pd loading was accomplished using a pH 11 solution of palladium (II) tetraammine nitrate. The Pd-Fe-SiO 2 catalyst was then dried at 125°C and then calcined at 250°C for 3 hours. Reduction was performed in 3% H 2 (balance Ar) at 200°C for 30 minutes and then at 600°C for 30 minutes. The catalyst was then cooled to room temperature and passivated in air.

Pd-Ga
A 2.5% Pd 2.5% Ga catalysts was prepared by sequential incipient wetness impregnation of gallium and palladium on silica. 1.25 g of gallium (III) nitrate hydrate and 2 grams of citric acid were dissolved in DI water to a total volume of 5 mL. The pH was adjusted to 11 with 32% ammonia solution and subsequently impregnated into 5 g of davasil 646 silica. The Ga-SiO 2 was dried at 125°C overnight and then calcined at 400°C for 3 hours. Pd impregnation was done using 3.3 g of 10% palladium (II) tetraammine nitrate diluted to 5 mL total volume and pH adjusted to 10 with 32% ammonia solution. The Pd solution was then impregnated to the pore volume of the 5 g of Ga-SiO 2 catalyst and then dried at 125°C overnight. The Pd-Ga catalyst was then calcined at 250°C for 3 hours and then subsequently reduced in 5% H 2 with a slow ramp (2.5°C/min) through 200°C and then a fast ramp (10°C/min) to 600°C with a 30 minute dwell. The reduced catalyst was then cooled to room temperature in 5% H 2 and passivated in air.

Pd-Zn
A 2% Pd 3% Zn catalyst was synthesized by sequential incipient wetness impregnation. 1.14 g of Zinc nitrate hexahydrate was dissolved in 2 mL of DI water and the pH was adjusted to 11 using 32% ammonia solution. Finally, the total volume of the Zn solution adjusted to 5 mL with the addition of DI water. The Zn solution was impregnated to the pore volume of 5 g of davasil 646 silica and dried overnight at 125 C. The Zn-SiO 2 catalyst was then calcined at 300°C for 3 hours.
0.281 g of palladium (II) tetraammine nitrate was dissolved in 3.5 mL of DI water and the pH was adjusted to 11 using 32% ammonia solution. The palladium solution was then impregnated to the pore volume of the calcined Zn-SiO 2 and dried at 125°C. The Pd-Zn-SiO 2 catalyst was calcined at 200°C for 3 hours and then reduced in 5% H 2 (balance N 2 ) with a slow ramp (2.5°C/min) through 250°C and a fast ramp (10°C/min) to 550°C and a 30 minute dwell at temperature. The reduced catalyst was then cooled to room temperature in hydrogen and passivated in air.

Pd-Mn
A 1% Pd 5% Mn catalyst was synthesized by sequential incipient wetness impregnation. 0.814 g of manganese (II) nitrate hydrate and 0.874 g of citric acid were dissolved in 5 mL of Millipore water to give a solution with a 2:1 molar ratio of citric acid to manganese nitrate. The pH of the solution was adjusted to 11 by the addition of 32% ammonium hydroxide. The solution was then added dropwise to 5 g of davasil 646 silica. The Mn-SiO 2 was then dried at 125°C overnight and calcined at 250°C for 3 hours. Pd loading was done by diluting 1.4 g of 10% palladium (II) tetraammine nitrate to 5 mL total volume and adjusting the pH to 11 with 32% ammonium hydroxide. The solution was then added dropwise to the Mn-SiO 2 and dried at 125°C overnight and calcined at 200°C for 3 hours. The Pd-Mn-SiO 2 catalyst was then reduced in 5% H 2 (balance N 2 ) with a slow ramp (2.5°C/min) through 250°C and a fast ramp (10°C/min) to 550°C with a 30minute dwell at temperature. The reduced catalyst was then cooled to room temperature in 5% H 2 and passivated in air.
Pd L 3 edge XANES Pd L 3 edge X-ray adsorption near edge structure (XANES) were measured at the 9BM line of the advanced photon source. Measurements were performed in fluorescence mode using a vortex 4 element detector. The samples were ground and pressed into a steel sample holder with the catalyst wafer at a 45-degree angle relative to the beam. The reactor used for treatment has been described elsewhere 1 , and is capable of heating and gas flow with kapton windows for transmission and fluorescence measurements. Samples were treated by heating to 500°C in 3.5% H 2 (balance He). After a 30-minute dwell at 500°C, the gas flow was switched to He at high temperature to desorb hydrogen and decompose any palladium hydride formed during the reduction. The samples were then cooled to room temperature and multiple scans were collected and averaged. L 3 edge XANES spectra were normalized using first order polynomial for the pre-edge region and a second order polynomial for the post-edge region. Due to the close proximity of the L 3 to the L 2 edge, the data collection range for post edge normalization is limited and the third order polynomial typically fit to the post edge region fits poorly. The absolute energy scale was calibrated using bulk PdO with an L 3 edge energy of 3174.4 eV. Figure S1 shows STEM images and EDX maps of Pd and Pd alloy catalysts. Pictured in figure S1a, the 1% Pd catalyst is monodisperse with small metal particles under 2 nm in diameter. In contrast, the 2% Pd catalyst, shown in figure S1b, contains both small (1-2 nm) particles and large (5+ nm) particles. The high temperature calcination treatment used in the 2Pd catalyst results in agglomeration of the palladium oxide resulting in larger metallic particles after reduction. EDX was used to observe the dispersion of the second metal, which is difficult to distinguish from the support in STEM images. Figure S1c shows a STEM image of 1Pd-5Mn with overlaid EDX maps for manganese and palladium. The manganese is well dispersed across the support, small clusters containing both Pd and Mn can be seen, consistent with the formation of a Pd-Mn bimetallic. Ga in the Pd-Ga catalyst (figure S1d) is also well dispersed on the support, but the bimetallic Pd-Ga clusters are better resolved owing to their slightly larger particle size. Due to the large number of overlapping peaks and imperfect background subtraction, it was not possible to determine a lattice parameter using the above pattern. However, simulation of the particle size broadening, shown in figure S5, gives an estimate of the particle size. The first major group of peaks, between 2.5-4 degrees, merges into a single asymmetric peak when the particle size is below 2 nm. In the 3 and 4 nm sized simulations, distinct shoulders start to emerge, and 4 major peaks can be resolved. In the 5 nm simulation, the most intense peak in the pattern around 3.1 degrees starts to split into two distinct peaks. In the Pd-Ga catalyst, the first cluster of peaks between 2.5-4 degrees most closely resembles the 3 nm simulation, which is larger the TEM measured particle size of 2.1. The difference between the XRD determined value and the TEM determined value can be attributed to the presence of microstrain broadening and the volume averaging nature of the XRD measurement. Electronic Characterization

Supplemental results
In addition to changing the catalyst structure, alloying can also electronically modify palladium, which can be studied by XANES. Figure S3a shows the Pd K edge XANES collected for 2Pd-3Zn, 2.5Pd-2.5Ga, 1Pd-5Mn and 1Pd after reduction in 3.5% H 2 (balance He) for 30 minutes. The alloy catalysts show small changes in the edge shape which are indicative of alloying. The first peak in the XANES is lower in intensity for the three alloy samples and the edge position for each is shifted   (magenta) at the K edge (a) and L 3 edge (b) after reduction at 550°C in 3.5% H2.

Electronic Structure Calculations
The surface Pd-atom projected Density of States (DoS) have been calculated for all the alloy surfaces to quantify the electronic modification of palladium in the alloy structures. Further, the 1 st moment (d-band center) and 2 nd moment (d-band width) are reported in Table S2. Apart from the Pd 2 Ga surface, all the Pd atoms on the surface of the alloys are crystallographically identical. Pd 2 Ga has two symmetrically distinct Pd atoms, hence an average of their moments has been reported. The results show that as the promoter content increases the Pd d- Adsorption energies of deep dehydrogenated intermediates   Effect of vdW functionals on Pd (111) and PdIn (110) To understand whether the functional influences the selectivity trends obtained, the binding energies of all the C 3 intermediates on Pd (111) and PdIn (110) have been recalculated with BEEF-vdW and optPBE functionals (Tables S6 and S7). The two surfaces have been chosen as representing the two extremes in terms of binding of adsorbates. The results show that the binding of intermediates are stronger with the vdW functionals, with optPBE having the strongest binding among the three functionals. Interestingly, on both the surfaces, the binding energies also have a linear correlation with the values obtained using PBE functionals (Figures S4 and S5). The slopes of the correlations are approximately 1, with the intercept being more negative for optPBE than for BEEF-vdW. This illustrates that the vdW functionals increase the strength of binding of intermediates by a constant value, given by the intercept. Therefore, the changes in reaction energetics and barriers for dehydrogenation on the surfaces would be very small. We expect these linear correlations to hold for other alloy surfaces as well. Furthermore, the calculated differences in binding energies of propylene, between Pd (111) and PdIn (110)

Surface Segregation Energies
To analyze the propensity of the alloy terrace surfaces to form clusters of either Pd or promoter on the top layer of the surface, we have performed a simple thermodynamic analysis by swapping the atoms on the surface with a counter-atom from 2 nd layer. This led to formation of surfaces with either a Pd-rich or a promoter rich top layer. The energies reported in Table S7 are the segregated energies of each alloy surface with respective to its clean surface counterpart. The analysis demonstrates that for the 1:1 and 2:1 alloys, the tendency to form segregated surfaces is very low, considering their large thermodynamic barriers (> 1 eV). For the 3:1 alloys, even though the energetics to form promoter-rich surfaces are still above 0.5 eV, there exist slight driving forces to form Pd-rich surfaces. Although we would not expect such small values to lead to substantial compositional changes in the surface layers, this effect could modestly contribute to the lower selectivity of 3:1 alloys towards propylene formation in comparison to 1:1 alloys.

Supplemental discussion
Because the structure of the alloy determines the local Pd coordination in an intermetallic compound, it also determines the electronic effect, and the two effects cannot be decoupled. In addition to the difference in ensemble size between the alloys, they also differ in the number of Pd-Pd bonds equal in length to the Pd-promoter bonds. As the local coordination changes, the d band of palladium is modified, which is quantified in the first and second moments of the d band.
In general, as the number of Pd-promoter bonds increases, the d band shifts away from the fermi level and increases in width (see table S2).
While L 3 edge XANES is commonly used to demonstrate d-band modification in platinum, the same information cannot be gained at the Pd L 3 edge due to the electron configuration of Pd.
Because Pd metal is d 10 , there are no unfilled d states, and the L 3 edge XANES cannot give information about the d-band. The lowest energy unfilled state accessible by the dipole selection rules of XANES is the 5s unfilled states. The change in whiteline shape observed for the alloys reflects the unfilled s states redistributing in energy due to overlap with neighboring promoter s orbitals, similar to how the 5d unfilled states in platinum are modified by promoters in platinum alloys 4 .