Local Structure of Pd 1 Single Sites on the Surface of PdIn Intermetallic Nanoparticles: A Combined DFT and CO-DRIFTS Study

: Local structure of Pd 1 single sites on the surface of Pd 1 In 1 intermetallic nanoparticles supported on α -Al 2 O 3 was investigated by the combination of CO-DRIFTS spectroscopy and DFT. CO-DRIFTS spectra of PdIn/Al 2 O 3 catalyst exhibit only one asymmetric absorption band of linearly adsorbed CO comprising two peaks at 2065 and 2055 cm − 1 attributable to CO molecules coordinated to Pd 1 sites located at (110) and (111) facets of PdIn nanoparticles. The absence of bridged or hollow-bonded CO bands indicates that multipoint adsorption on PdIn nanoparticles is signiﬁcantly hindered or impossible. DFT results show that on (110) facet multipoint CO adsorption is hindered due to large distance between neighboring Pd atoms (3.35 Å). On (111) facet multipoint CO adsorption on surface palladium atoms is impossible, since adjacent Pd atoms are located below the surface plane.


Introduction
Single-atom alloys (SAA) is an important class of so-called single-site catalysts widely used in state-of-the-art catalysis [1][2][3][4]. As a rule, SAA catalysts contain supported nanoparticles of substitution alloy, on the surface of which the active metal atoms are isolated from each other by the atoms of inactive (or less active) component, thus forming a system of isolated single-atom sites. SAA catalysts have received increased attention in recent years due to their enhanced selectivity in a number of reactions and regeneration capability.
A significant drawback of these systems is insufficient stability of "single site" surface structure, especially under conditions of adsorbate induced segregation. Thus, the adsorption of molecules with a high adsorption energy (for example: CO, ethylene, acetylene) leads to the enrichment of the surface with the active component, which is frequently accompanied by the formation of multiatomic centers deteriorating surface SAA structure [5][6][7][8].
Unlike substitution alloys, intermetallic compounds (IMCs) can provide a much more stable structure of isolated surface active sites due to several reasons. Thus, the IMCs possess a well-organized crystal structure, which is more stable than solid solution alloys due to the high formation enthalpy. The enthalpic driving force stabilizes IMC surface structure by preventing or minimizing surface segregation even in the presence of adsorbates [9,10]. It is important that IMCs tend to form a system of single-atom active sites on their surface [1,11,12]. Thus, theoretical analysis of formic acid decomposition over SAA-like CuPt catalysts revealed intermetallic Cu 3 Pt as a promising material for this process [13], since its surface structure exhibits only Pt 1 sites isolated by Cu component. This conclusion is in agreement with the results reported elsewhere [14][15][16].
An excellent example of single-site intermetallic catalyst is PdIn compositions, which demonstrate a favorable performance for selective semihydrogenation of substituted alkynes, direct formic acid fuel cell processes, production of enhanced magnetic resonance signals using parahydrogen, etc. [17][18][19][20][21]. The specific catalytic properties of PdIn IMCs are attributed to the formation of single Pd 1 sites on their surface, isolated from each other by indium atoms [12,17,20,22,23], which prevents or hinders multipoint adsorption of reacting molecules. Thus, the authors of [17] using computational chemistry modeling with density functional theory (DFT), predicted high ethylene selectivity in acetylene hydrogenation for PdIn(110) surface due to destabilization of multisite adsorption on Pd 1 sites, which favors ethylene desorption rather than its overhydrogenation to ethane. However, the local structure of single-atom Pd 1 sites on atomic level is not clear yet.
It should be noted that the study of the structure of isolated atomic centers on the surface of bimetallic nanoparticles is a difficult task. The most surface-sensitive techniques (XPS, Auger spectroscopy, etc.) do not provide detailed information about the top atomic layer, but analyze surface and subsurface atomic layers (up to 5 nm in depth). Fortunately, the convenient method for studying surface structure of Pd single-site catalysts is the infrared spectroscopy of adsorbed CO. This is due to the fact that CO is adsorbed only by surface atoms, and it is possible to distinguish isolated Pd 1 and multiatomic Pd m (m > 1) centers by the presence of linear-, bridge-or threefold hollow-bonded CO bands. Note that bridged and threefold hollow-bonded CO adsorption is energetically more favorable than linear CO adsorption on Pd surface [24,25] and usually the signals of multiply coordinated CO tend to prevail in infrared spectra. Therefore, convincing evidence of the formation of Pd 1 single-site surface structure is the complete disappearance of bridged and triple bonded CO adsorption signals and the presence of linearly bonded CO only, suggesting that CO coordination to several Pd atoms is unfavorable [18][19][20][26][27][28][29][30][31][32].
There are several indications that the formation of PdIn intermetallic compound leads to a decrease in the relative intensity, or disappearance of multiply-bonded CO (bridged and threefold hollow bonded) and to a "red" shift of linearly adsorbed CO by 10-25 cm −1 toward lower frequency in comparison with monometallic palladium surface. Thus, the authors of [27] reported disappearance of bridged CO for freshly reduced PdIn catalyst in contrast to monometallic Pd. Similar observations were presented elsewhere [33,34]. Results of our previous CO-DRIFTS study of Pd 1 In 1 /Al 2 O 3 are in good agreement with these data and demonstrate the disappearance of multiply-bonded CO species on the surface of PdIn nanoparticles [18,30,35]. However, it is still unclear which characteristic features of PdIn surface structure impede the multipoint adsorption of CO.
It should be noted that the formation of isolated Pd 1 sites on SAA surface usually occurs in the catalyst with a low fraction of palladium and is explained by a dilution of Pd atoms and their surroundings by atoms of the second component, which is present in significant abundance. However, for Pd 1 In 1 intermetallics isolation of Pd atoms by their surrounding with In atoms seems unlikely, since the fraction of In atoms is not sufficient, and the disappearance of bridged CO should be explained by a specific local atomic structure around surface Pd atoms.
Therefore, this study was focused on investigation of the local atomic structure of Pd 1 isolated single sites on the surface of PdIn intermetallic nanoparticles by combining experimental (CO-DRIFTS) and theoretical (DFT calculations) approaches. Since CO-DRIFTS provides information only on the upper surface layer, the combination with DFT modeling can be an informative method for revealing the local atomic surroundings of Pd 1 surface sites.

The Choice of Catalytic System
The reference Pd/Al 2 O 3 and bimetallic PdIn/Al 2 O 3 catalysts were prepared via incipient wetness impregnation of α-Al 2 O 3 by aqueous solution of Pd and In nitrates followed by hydrogen reduction in order to avoid the formation of monometallic Pd or In species. Experimental details can be found in Supplementary Materials. The α-Al 2 O 3 was used as a carrier for the catalyst preparation. This choice was dictated by several factors. First, the highly-crystalline structure of α-Al 2 O 3 allows reliable study of Pd 1 In 1 nanoparticles by XRD analysis, since α-Al 2 O 3 diffraction pattern exhibits narrow reflexes that do not overlap characteristic signals of metallic Pd, In, and PdIn. Second, since relatively large bimetallic particles are formed over α-Al 2 O 3 due to low BET surface area, the fraction of edged and corner atoms is low and contribution of CO adsorption on these atoms can be neglected. Third, high chemical inertness of this carrier allows one to minimize metal-support interaction effects.

Transmission Electron Microscopy
The microstructure of the samples was studied by transmission electron microscopy (TEM), using an HT7700 instrument (Hitachi, Japan). More details can be found in Supplementary Materials. Figure S1 presents TEM images of Pd/Al 2 O 3 and PdIn/Al 2 O 3 catalysts. The monometallic catalyst ( Figure S1a) contains almost spherical well-distributed Pd particles with average diameter of ca. 30 nm. Bimetallic catalyst contains the regular spherical nanosized PdIn particles ranged from 10 to 30 nm.

X-ray Diffraction
Powder XRD patterns were obtained on a Bruker D8 ADVANCE X-ray diffractometer (Cu Kα, Ni-filter, LYNX-EYE detector, reflection geometry). For details, please see Supplementary Materials.  [29,32,39]. It is important that no peaks characteristic of monometallicPd 0 were observed. It should also be mentioned that no metallic In 0 particles were detected by XRD due to the absence of characteristic XRD reflections at 32.9 • , 36.

CO-DRIFTS
Diffuse reflectance infrared Fourier transform spectroscopy was performed with a Tensor 27 Bruker spectrometer equipped with a high temperature cell (Harrick) and liquid-nitrogen-cooled MCT detector. Spectra were recorded at 50 • C under continuous 0.5% CO/He flow (30 cm 3 /min). For details, please refer to Supplementary Materials. Typical CO-DRIFT spectra of the reference Pd/Al 2 O 3 and the intermetallic PdIn/Al 2 O 3 are displayed in Figure 2. The DRIFTS spectrum of CO adsorbed on the reference monometallic Pd-catalyst exhibits two broad bands at 2150-2050 and 2200-1800 cm −1 (Figure 2a). The band at~2100-2000 cm −1 corresponds to CO molecules linearly adsorbed on Pd atoms, whereas the broad peak at~2000-1800 cm −1 is attributable to bridged and hollow-bonded CO on Pd surface [41]. Lower intensity of linearly adsorbed CO bands can be explained by the large size of Pd particles (~20-30 nm), and the prevalence of terrace atoms on their surface, which favors multiply adsorbed CO. In contrast to Pd/Al 2 O 3 , spectrum of CO adsorbed on PdIn/Al 2 O 3 does not exhibit bands of bridged or hollow-bonded CO within 2000-1800 cm −1 region. These observations are in a good agreement with the data reported by Furukawa [27], Hirano [33], and results of our previous studies [18,35], and indicate that multiple coordination of CO on the surface of PdIn intermetallic nanoparticles is unfavorable. As mentioned above, this observation can be explained by the disappearance of Pd n ensembles (n ≥ 2) on PdIn surface [27,33] capable of accommodating bridged CO species [35] similarly to the PdGa IMCs [22]. This idea is in line with the results of detailed study of bulk structure and physical properties of InPd IMC [21].
The band corresponding with linear CO on PdIn (2064 cm −1 ) is shifted toward a lower frequency relative to the band of CO adsorbed on monometallic Pd nanoparticles (~2084 cm −1 ), which is in a good agreement with the data reported by Wu [34] and Furukawa [27].
In order to gain insight into specific surface arrangements of surface structure of PdIn nanoparticles determining CO adsorption, it is informative to analyze possible reasons of the asymmetry of the band at 2064 cm −1 . Tentatively, its assignment can be made on the basis of the simplified PdIn intermetallic model proposed in [17]. In accordance to this model, PdIn crystallite surface is made up of two facets, (110) and (111) (see Figure 3a,b).
The contribution of the (110) facet to the total surface area is~83%, while only 17% of (111) facet. The model allows us to suggest that peak asymmetry of linear adsorbed CO in the FTIR spectra of PdIn/Al 2 O 3 catalyst is attributable to CO adsorption on (110) and (111) facets in accordance with the results of DFT calculations (see data below). Deconvolution of the linearly adsorbed CO peak revealed two bands with maxima at 2065 and 2055 cm −1 , corresponding to CO adsorption on (111) and (110) facets, respectively (Figure 2b). It should be noted that the ratio of integrated intensities of the absorption bands for (111) and (110) facets calculated from spectral data is approximately 48 and 52%. The discrepancy with the data reported in [17] may be related to differences in structure of bimetallic PdIn nanoparticles, or either different extinction coefficients for CO molecules adsorbed on various facets, different adsorption strengths (see below), or both.
To investigate the strength of CO adsorption on different PdIn facets the temperaturedependent experiment was performed (Figure 2b). It is shown that the increase in temperature leads to the change in peak shape and its shift towards lower wavenumbers. The peak maximum shifts from 2064 to 2057 cm −1 with temperature increase from 50 to 150 • C.
The results of spectra deconvolution show the pronounced decrease in the intensity of the peak corresponding to CO adsorption on (110) facet, while the intensity of the peak related to (111) facet changes to a much lesser extent ( Table 1). The steep increase in I CO(111) /I CO(110) intensity ratio indicated that CO molecules bound to (111) facet of PdIn much stronger than to (110). This observation is in good agreement with the results of DFT calculation (Table 2)     For spectra collected at 50 • C, the FWHM is 11.6 cm −1 and 25.5 cm −1 for peaks with maximum at~2065 cm −1 and 2055 cm −1 , respectively. This difference is significant, but not as large as in a previously reported study, where FWHM was~10, 25, and 40 cm −1 for CO adsorbed on metallic Pt [41]. The broadening of CO adsorption peak is presumably attributable to inhomogeneity of adsorption sites and can be related to different sizes of metallic particles [42].
In order to exclude the possible influence of gaseous or physisorbed CO, we performed the experiment by purging adsorbed CO with helium flow at 50 • C. The results of this experiment were found to be in a good agreement with the results obtained by variation of adsorption temperature and also revealed stronger adsorption of CO on (111) facet (see Figure S2).

DFT Calculations
For interpretation of the observed spectral data the DFT calculations were performed. Details can be found in Supplementary Materials. Table 2 shows calculated CO adsorption energies and wavenumbers for various adsorption sites on the surfaces of Pd and PdIn.
The calculated values for linear, bridge, and hollow-bonded CO for Pd(111) and Pd(100) facets were in agreement with DRIFTS results. Calculated CO adsorption energies are consistent with the fact that CO preferentially adsorbs on bridge and hollow sites.
With DFT calculations the adsorption behavior of CO at the two different facets of PdIn intermetallic compound were investigated and the results are presented in Figure 3a-e and Table 2. It is of interest to analyze the DRIFTS data on CO adsorption on the surface of intermetallic nanoparticles by DFT calculations in two aspects: (1) the absence of signals from multiply coordinated CO species, and (2) the position of absorption bands and the adsorption energy of linear forms of adsorbed CO.
DFT calculations indicate that CO does not bind to In atoms, but only to the sites consisting of Pd. This limits the possible adsorption modes of CO on PdIn surface significantly. DFT results also show that on the (110) surface CO is capable of binding both bridged and linearly to Pd (Figure 3 c,d respectively). However, the energy of CO adsorption in bridged position on the (110) facet is negligible (−0.09 eV), therefore this band is not detectable under the conditions of our experiments. Such a low energy of adsorption compared to the energy of adsorption of the bridged form of CO on monometallic palladium (−1.51 to 1.71 eV, see Table 2) stems from the different geometry of the bridge site on intermetallic PdIn (110) and monometallic Pd. For monometallic Pd the distance between neighboring Pd atoms is 2.8 Å, whereas on PdIn(110) the corresponding distance is 3.34 Å (Figure 3a and Figure S3). Therefore, to facilitate the CO adsorption at the bridge site a quite heavy reconstruction of the surface takes place, where the interatomic Pd-Pd distance at the bridge site is shortened to 2.96 Å.
On the (111) surface, CO can only bind linearly to Pd as shown in Figure 3e. Adsorption of bridge-bonded CO on PdIn(111) facet is impossible due to specific atomic geometry of Pd site on the surface of (111) facet. As can be seen from Figure 3e and Figure S3, the Pd atoms adjacent to the Pd atom are located significantly below the surface plane. Since the nearest Pd surface atom is at a distance of 4.73 Å, this makes bridging CO adsorption impossible. Thus, the calculated data is consisted with the absence of additional bands in the range of 2000-1800 cm −1 in the CO-DRIFTS spectra of PdIn/Al 2 O 3 sample and clearly show that only linear CO adsorption on isolated Pd 1 sites is possible on the surface of intermetallic PdIn nanoparticles.
DFT study of linear CO adsorption indicates that CO adsorbed on (111) and (110) facets vibrates with a frequency value of 2065 and 2058 cm −1 , which is in qualitative agreement with experimentally observed values: 2064 and 2055 cm −1 ( Table 2). It is remarkable that DFT calculations also show stronger binding of linear CO on the (111) facet with an adsorption energy of −0.77 eV compared to an adsorption energy of −0.39 eV on the (110) surface. Different strength of CO adsorption is in agreement with the experimental data on different stability of linear CO species revealed by variation of CO adsorption temperature (Figure 2b). Thus, the preferential decrease in the intensity of the peak at 2055 cm −1 with increasing adsorption temperature is in agreement with weaker CO binding on (110) facet.
Different stability of CO adsorbed on (111) and (110) facets can be explained as follows. The surface Pd-atoms in both surface terminations have a coordination of 8 nearest neighbors; however, in the surface layer of the (111) surface each Pd-atom is surrounded by 4 Pd and 4 In, whereas each surface Pd in the (110)-surface is surrounded by 2 Pd and 6 In. In order to address the effect of the local environment on the reactivity of surface Pd, we calculated the d-band center of the surface Pd-atom to which CO bind. According to the d-band model proposed in [43], the strength of the adsorption decreases as the d-states are shifted down in energy. We find a downshift in the d-states energy going from PdIn(111) (−1.6 eV) to PdIn(110) (−2.4 eV), which explains the weaker binding of CO to Pd on the surface of PdIn(110).

Conclusions
In summary, supported intermetallic PdIn nanoparticles were synthesized and their surface structure was studied by combination of experimental and theoretical methods: CO-DRIFT spectroscopy and DFT calculations. Results of CO-DRIFTS revealed only linear CO adsorption on PdIn nanoparticles as indicated by two CO stretching vibration bands at 2065 and 2055 cm −1 , corresponding to CO adsorption on (111) and (110) facets. The assignment of the bands was confirmed by DFT calculation. Both experimental and theoretical methods have shown that multipoint adsorption of CO on the surface of PdIn nanoparticles is unfavorable. On the PdIn (110) facet the energy of CO adsorption in bridged position was found to be negligible (−0.09 eV) because of too large a distance between neighboring Pd (3.35 Å). Adsorption of bridge-bonded CO on PdIn(111) facet is impossible since palladium atoms adjacent to the surface palladium atom are located significantly below the surface plane, while the nearest Pd surface atom is at a distance of 4.73 Å, which excludes bridging adsorption.  Figure S1: Representative TEM images for Pd/Al 2 O 3 (left) and PdIn/Al 2 O 3 (right) catalysts. Figure S2: CO-DRIFTS data for PdIn/Al 2 O 3 catalyst: (a) CO desorption from PdIn/Al 2 O 3 in He flow at 50 • C; (b) peak deconvolution after 0, 14 and 24 min of the experiment. Before the desorption experiment PdIn sample was reduced in situ at 500 • C in 5%H2/Ar flow. After cooling to 50 • C He flow, flow of 0.5 vol.% CO/He was introduced to the cell and the spectra were collected. Then the adsorbed CO was purged by helium flow (30 L/min) at 50 • C for 30min with subsequent recording of spectra. Figure S3: Scheme of the neighboring Pd positions on the Pd (110) and Pd (111) surface planes. Pd atoms are blue and In atoms are brown. Table S1: Parameters used for DFT calculations.

Conflicts of Interest:
The authors declare no conflict of interest.