Pd Thickness Optimization on Silicate Sheets for Improving Catalytic Activity

Maximizing surface‐to‐body ratio demands ever smaller metallic palladium (Pd) nanoparticles for catalytic applications. The quest for miniaturization is now reaching the single‐atom limit. However, if the supported Pd is below a critical size, the Pd hybridization with the supporting material can detrimentally reduce the labile electrons that facilitate the catalytic reactions. Thus, the smallest attainable size, i.e., single‐atom Pd, may not offer the best efficiency. Here, it is demonstrated that Pd with at least six atomic layers (or thickness of ≈1 nm) on the silicate sheets, synthesized via the partial exfoliation of a layered silicate, exhibits a metallic‐like electronic property, yielding an excellent catalytic activity (e.g., turnover frequency) for dehydrogenating formic acid higher than both isotropic Pd nanoparticles and single‐atom Pd.


Pd Thickness Optimization on Silicate Sheets for Improving Catalytic Activity
Esmail Doustkhah,* Nao Tsunoji, M. Hussein N. Assadi others. [2] Formic acid, nonetheless, has become a favorite for energy production [3] because of its facile processing and high hydrogen reservoir capacity. [3b] A prevalent class of catalysts for the dehydrogenation of formic acid is based on Pd nanomaterials [4] that efficiently dehydrogenate formic acid to hydrogen and carbon dioxide. Although tremendous efforts have been dedicated to improving Pd-based catalysts, [5] such as the catalytic activities of the Pd's crystal planes, [6] some critical mechanistic aspects of Pd's catalytic dehydrogenation of formic acid remain unexplored. [7] Decreasing the size of Pd particles, that is, moving toward a single-atom catalyst, has been hinted to boost the Pd's catalytic activity. [8] However, no optimal Pd particle size has been identified for optimal catalytic formic acid dehydrogenation. [9] Moreover, the support material's role, if any, in determining Pd's optimal shape and size is to be established. Here, we demonstrate that plate-like Pd nanoparticles supported on the silicate must have at least six atom-layer-thick (≥1 nm) for an optimal catalytic property toward formic acid dehydrogenation that is considerably higher than single-atom Pd catalysts.
To design plate-like Pd architecture, we use silicate sheets obtained via partial exfoliation of a commercially available layered silicate that possesses a richer surface silanol group (SiOH/SiO − ) compared with the other layered silicates. [10] The rationale behind our choice is that a richer hydroxyl-layered surface appears more efficient in immobilizing metals and grafting organic functional groups necessary to immobilize metals. [11] However, the exfoliation of silanol-rich layered silicates has been recognized as challenging due to several factors, including higher-layer charge density (stronger Coulombic interaction between layers). In this study, we utilize iminodiacetate (IDA) immobilized on the silicate surface as an exfoliating agent and Pd-adsorption assisting agent precursors, [12] which is subsequently eliminated when the plate-like Pd formation attaches to the silicate nanosheets' surface ( Figure 1).

Materials Synthesis and Characterization
Layered materials often show relatively slow diffusibility to reactants (ions and molecules) due to the absence of open pores in the interlayer space. [13] Therefore, the exfoliation of layered materials while immobilizing catalytically active sites is crucial Maximizing surface-to-body ratio demands ever smaller metallic palladium (Pd) nanoparticles for catalytic applications. The quest for miniaturization is now reaching the single-atom limit. However, if the supported Pd is below a critical size, the Pd hybridization with the supporting material can detrimentally reduce the labile electrons that facilitate the catalytic reactions. Thus, the smallest attainable size, i.e., single-atom Pd, may not offer the best efficiency. Here, it is demonstrated that Pd with at least six atomic layers (or thickness of ≈1 nm) on the silicate sheets, synthesized via the partial exfoliation of a layered silicate, exhibits a metallic-like electronic property, yielding an excellent catalytic activity (e.g., turnover frequency) for dehydrogenating formic acid higher than both isotropic Pd nanoparticles and single-atom Pd. to designing a highly active catalytic architecture. Here, we present a strategy for exfoliating layered silicate and immobilizing Pd on the surface by modifying the interlayer surface with IDA. In this stage, IDA serves a dual role as both an exfoliating agent and a collecting agent for the Pd precursor, which will be discussed in detail later. First, the interlayer space (1.58 nm) of the original layered silicate (Na-SiO 2 ) is expanded to 3.24 nm via a cation exchange of the interlayer Na ions with cetyltrimethylammonium (C 16 ) ions to form an intermediate named C 16 -SiO 2 to modify the interlayer surface with a bulky (3-aminopropyl) triethoxysilane. The obtained NH 2 -SiO 2 , having a basal spacing of 3.20 nm, was set to further react with sodium chloroacetate to convert the attached amino groups to IDA, causing the basal spacing of the structure to shrink to 1.58 nm to form IDA-SiO 2 , as monitored by powder X-ray diffraction (XRD) in Figure 2A. Solid-state 13 C cross-polarization magic angle spinning nuclear magnetic resonance (CP MAS NMR) revealed the presence of aminopropyl groups and propyliminodiacetate in NH 2 -SiO 2 and IDA-SiO 2 , respectively, while a minor amount of C 16 was detected in the two products ( Figure 2B). Notably, the Pd formation through reduction onto the layered silicate was accompanied by a significant loss in the carbon species' intensity.
Solid-state 29 Si MAS NMR revealed that the integral Q 3 /Q 4 signal ratio, which reflects the amount of the surface Si-OH groups, for NH 2 -SiO 2 (≈15/75) was not only smaller than that for Na-SiO 2 and C 16-SiO 2 (≈40/60 for each) but also almost identical to that of IDA-SiO 2 ( Figure 2C). Moreover, NH 2 -SiO 2 and IDA-SiO 2 have T signals, assignable to organosylilic Si atoms   13 C CP MAS NMR spectra, and C) 29 Si MAS NMR spectra of materials. Asterisks in panels (B) and (C) indicate organic compounds (including C 16 and solvents) and spinning sidebands, respectively. www.advmatinterfaces.de covalently attached to the silicate layer via Si-O-Si bonds, and their integral ratios were almost identical (T/Q 3 /Q 4 ≈ 15/15/75). All the results described above indicate that aminopropyl groups are immobilized on the interlayer surface of the layered silicate, and subsequently, the amino functionality is quantitatively converted to IDA. In contrast, almost all the preintercalated C 16 ions are displaced.
As revealed by the solid-state 29 Si NMR, only 66%, calculated as (45−15)/45 × 100, of the interlayer Si-OH groups were modified with aminopropyl groups in NH 2 -SiO 2 ( Figure 2C), meaning that a significant amount SiO − /SiOH group remained unmodified. Upon the reaction of sodium chloroacetate with NH 2 -SiO 2 , besides the conversion of the immobilized amino group to IDA, a substantial amount of Na + ions is exchanged with the intercalated C 16 TMA + . This cation exchange recharge the layered silicate with Na + , rendering the composition comparable to the original Na-SiO 2 (Table S1, Supporting Information). Therefore, we can mention that Na + ions revert to the original position for IDA-SiO 2 due to the inhomogeneous modification with the organic functional group, as schematically shown in Figure 1.
To demonstrate that IDA-SiO 2 was exfoliated, we resorted to scanning electron microscopy (SEM) and atomic force microscopy (AFM) characterizations. Accordingly, Na-SiO 2 was composed of microplate-like crystallite aggregates with a lateral size of up to 2 µm and a thickness of 50-100 nm ( Figure 3A). In contrast with NH 2 -SiO 2 , which was composed of just disaggregated microplates ( Figure 3B), IDA-SiO 2 was composed of disaggregated microplates, and each plate formed thinly enough to be flexible-unlike Na-SiO 2 and NH 2 -SiO 2 ( Figure 3C). AFM confirmed the existence of thin sheets with a thickness of ≈5 nm corresponding to three single layers, [10] which were considerably thinner than plates (>40 nm) of Na-SiO 2 ( Figure 3A, inset). Thicker sheets with a thickness of up to 20 nm were also observed (more AFM images are shown in Figure S1, Supporting Information). These results indicate partial exfoliation of Na-SiO 2 due to the surface modification by IDA. A possible mechanism for the exfoliation involves a repulsion between negatively charged COO − groups on the layers. [14] Since the modification is inhomogeneous and some unmodified interlayer spaces still exist after treatment, the complete exfoliation into single layers is thought to be impossible ( Figure 1).
IDA-SiO 2 was reacted with a Pd precursor (K 2 PdCl 4 ) followed by reduction with NaBH 4 to yield Pd-SiO 2 . The XRD showed no significant change in the basal spacing during the reaction, while few organic groups were detected in the 13 C NMR spectrum (Figure 2A,B). In the 29 Si NMR spectrum, the T environments were hardly seen, while the Q 3 /Q 4 signal ratio was comparable to that of the original Na-SiO 2 . Moreover, elemental analyses showed that Pd-SiO 2 still preserved a significant amount of Na ions similar to IDA-SiO 2 while having 0.57 mass% of Pd (Table S1, Supporting Information). Additionally, SEM, transmission electron microscopy (TEM), scanning TEM (STEM), and high-resolution TEM (HRTEM) revealed the presence of Pd nanoparticles on the flexible silicate sheets for Pd-SiO 2 ( Figure 3E-G and Figure S2, Supporting Information). We can thus conclude that almost all the formed Pd nanoparticles are directly attached to the exfoliated sheets. We can consider the following possible mechanism for the Pd formation: the Pd precursors, that is the Pd ions, are collected around IDA functionalities immobilized on the silicate sheets via strong interactions between Pd ions and IDA. [12] Afterward, NaBH 4 treatment reduces the collected Pd precursors to metallic Pd and pulls out the organic units from the silicate surface, [15] resulting in the formation of Pd metals directly attached to the silicate surface.
To accurately investigate the Pd structure and morphology, we conducted high-resolution scanning probe microscopy (AFM mode). This characterization helped us measure nanoparticles' thickness and lateral size in a high resolution, which a usual AFM does not deliver. As revealed in Figure 4, most Pd nanoparticles (>75%) for Pd-SiO 2 had a lateral dimension of 5-60 nm and a thickness of 1-3 nm and were gently tapered plate-like particles. The average lateral size/thickness aspect The inset of panel A shows the thickness distribution of each microplate-like particle.

www.advmatinterfaces.de
ratio is calculated to be 14 from the inverse of the slope of a regression line of data points, indicating the plate-like structure of the Pd. The rough linear relationship between the lateral size and thickness of the Pd also suggests that Pd growth is controlled by the 2D surface of the silicate nanosheets as a scaffold. Therefore, further study is worth conducting to precisely control the Pd thickness by changing Pd deposition methods.

Pd's Electronic Structure via Density Functional Calculations and XPS
We now examine the atomic and electronic interactions at the interface of the silicate nanosheet and Pd with density functional calculations to understand how the Pd growth progress at this interface influences the catalytic activity. Since the exact structure of the Na-type layered silicate used in this study is still elusive; [10] here, we use the structure of a known sodium silicate (Materials Project, mp-1210861) for the calculation, that to a great extent, resembles the stoichiometry and the morphology of Na-SiO 2 . In Figure S3 (Supporting Information), we first establish Pd's stronger affinity toward the Na-SiO 2 surface over chelation with IDA, so we focus our discussion around Pd/Na-SiO 2 interface. Figure 5A,B shows the optimized interface structure and the layer-resolved Pd 4d and O 2p partial density of states (PDOS). The O 2p states are of those O ions at the outermost layer that directly interact with Pd, while Pd states are presented for all layers. As in Figure 5A-7, the O 2p states gravitate toward the bottom of the valence band (−9 to −3 eV). The Pd 4d states of the layer directly at the interface (Figure 5A-6) are greatly delocalized and stretch to ≈−7 eV below the Fermi level. Consequently, to a large extent, the 4d states of these Pd ions hybridize with the O 2p state. This hybridization, which pulls the 4d states to lower energies, is highlighted in yellow in Figure 5A-7 and A-6. Figure 5C shows the electronic localization function (η) at the sodium silicate and Pd interface. η indicates the likelihood of finding a spin-like electron in the vicinity of a reference electron at any given point of space and provides a qualitative mean to map electron pair probability. [16] η has a range between 0 and 1. The higher η is at a given point, the more likely it is to find an electron pair. When coupled with charge density analysis, η can measure the degree of covalency of a bond. [17] η in Figure 5C has significant values along the PdO bond at the interface. More quantitatively, Figure 5D, which shows the η line profile along the PdO bond (the dotted line in Figure 5C,E), indicates a second peak (marked with an arrow) at the middle of the bond (η = 0.560 at 1.74 Å). Combined with the clue of nonvanishing charge density along the same bond (marked with a cross in Figure 5E), we infer that the hybridization between O and Pd at the interface is significantly covalent. This covalent interaction is mainly responsible for delocalizing the Pd states to lower energies at the first Pd layer ( Figure 5A-6).
In the Pd layers away from the interface, as shown in the partial DOS of Figure 5A-3 through A-7, the Pd 4d states become more and more localized toward higher energies closer to the Fermi level. For instance, in the fifth Pd later, the Pd 4d states only stretch to −5 eV below the Fermi level instead of −7 eV for the first Pd layer ( Figure 5A-6). Furthermore, in the outermost Pd layer, the Pd 4d states gravitate to energies closer to the Fermi level, as marked with an arrow in Figure 5A-7. This partial density of states distribution drastically resembles that of the pure Pd surface. [18] We conclude that as the Pd layer grows thicker at the interface, the Pd 4d states move from being covalent-like to being more metalliclike, therefore, more labile for participation in the catalytic activity.
To further show Pd 4d states liability changes with the layer thickness, we calculated the partial electronic population in the Pd 4d states confined to the energy range −3.5 eV < E < E Fermi . Pd's electrons within 3.5 eV below the Fermi level are known to make a more significant contribution to its catalytic properties. [19] This partial electronic population (q) was calculated by integrating the 4d PDOS over the range of −3.5 eV < E < E Fermi , and normalizing the integral with the integrated 4d PDOS over all energies below E Fermi times, all Pd electrons expressly www.advmatinterfaces.de considered in the pseudopotential. As shown in the A panels of Figure 5, q for the Pd layer hybridizing with O is 4.67 e per Pd. However, q rises to 6.14 eV per Pd at the outermost layer, indicating a 1.47 e per Pd gain. Figure S4 (Supporting Information), which reports a structure with only three Pd layers, shows a q value of 5.79 e per Pd at the Pd surface, which is still smaller than the surface q of 6 Pd layers. Additionally, Figure S5 (Supporting Information) shows that the surface q for the structure containing nine Pd layers is 6.15 e per Pd, a value quite similar to that of the structure with six Pd layers. Consequently, our simulation demonstrates that at six atomic layers, the Pd at the surface acts more like a metal by having a higher electronic population near the Fermi energy. Decreasing the Pd thickness markedly reduces the Pd 4d electronic population near the Fermi energy, while increasing the Pd thickness does not significantly increase q.
According to the d band model for catalysis, the higher, i.e., closer to the Fermi level, the center of the d band of transition metal is, the stronger the adsorption is between the surface and the catalytic species. [20] In essence, the d band locationor approximately its center's location-from the Fermi level determines the hybridization degree between the antibonding (σ * ) orbital of the adsorbate and the d band of the surface. The σ * -d hybridization is itself antibonding and offsets the stability of the bonding σ-d hybridization. [21] The balance between the bonding σ-d and antibonding σ * -d dictates the adsorption rate of the initial molecules and the desorption rate of the catalyzed species. In the case of Pd for formic acid dehydrogenation, closer d band centers to the Fermi surface results in higher catalytic activity. [5a,22] Our calculations indicate that for a thickness of at least six atomic layers, the 4d band is adequately and optimally close to the Fermi level. Spreading the Pd load thinner,

www.advmatinterfaces.de
although it increases the surface area, lowers the d band center. Thicker Pd particles reduce the overall Pd surface area for a given Pd load. Both scenarios decrease the catalytic activity. As a result, a thickness of six atomic layers, or about ≈1 nm, is more likely an optimal thickness balancing the surface area and the d band center for catalytic performance.
To experimentally confirm the metallic-like state of Pd nanoparticles in Pd-SiO 2 , we measured the sample's X-ray photoelectron spectroscopy (XPS). Metallic Pd shows a peak due to the 3d 3/2 electron at 335.2-335.7 eV in the Pd 3d-region XPS spectrum, [8a] which is observed for our reference sample, Pd-GO (isotropic Pd 0 crystals (18-34 nm size, average size of 21 nm) supported on graphene oxide [23] ) ( Figure 6A). Whereas Pd-SiO 2 shows two peaks; one is a prominent peak at 335.0 eV assigned to metallic Pd, and the other is a shoulder peak around 336.3, which can be assigned to Pd 2+ species. [8a] Here, we strongly believe that the shoulder at 336.3 eV originates from the Pd species that exist in the interface of the silicate layers (the first few layers) in the form of electron-deficient Pd δ+ species, which is previously observed for another silicate [24] and is consistent with our DFT results. The electron-deficient Pd δ+ species can occur due to the strong interaction of Pd with the surface O of the support. The Pd's d band has a larger energy difference than the Fermi level, leading to lower catalytic activity. We would thus expect a high catalytic activity for the Pd layers that are located above those six layers. Here, we can rule out the possibility of the surface oxidation of the Pd in Pd-SiO 2 since Pd is resistive to oxidation (even to harsh oxidations [25] and high temperatures [26] ). PdO bond formation is also observed in the case of other oxides, e.g., CeO 2 -supported Pd nanoparticles. [27] XPS study ( Figure 6B) also confirms the existence of a significant amount of Na in Pd-SiO 2 . The atomic% of Pd and Na determined by the spectrum were 0.1% and 2.3%, respectively, corresponding to the Pd/Na mass ratio of 0.20. This mass ratio is in good agreement with that obtained by the elemental analysis data (Table S1, Supporting Information). Although the type of support, as well as the size and the location of Pd nanoparticles, are critical in determining the binding energy of the Pd nanoparticles on SiO 2 , the coexisting another cation such as Na + in Pd-SiO 2 may not shift the binding energy of Pd. In this regard, a previous report indicates that Pd 0 nanoparticles (3.5 ± 0.6 nm) supported on the external surface of a Na-contained synthetic clay-sodium fluorohectorite ([Na] 0.5 [Mg 2.5 Li 0.5 Si 4 O 10 F 2 ], which may be supposed as a typical supporting, compared with Na-free support (alumina), shows a very slight shift while synthesizing similar Pd nanoparticles in the interlayers of the clay (but not on the external surface) significantly shifts the binding energy of Pd nanoparticles (Table S2, Supporting Information). [24]

Catalytic Activity Evaluation
There are two main mechanisms in the decomposition of formic acid, usually in competition. [28] One produces CO and H 2 O, which is an unfavorable reaction, and the other produces H 2 and CO 2 . Since selective H 2 generation (with a strict tolerance of at most 5 ppm of CO trace) is a crucial issue for fuel cell applications, attempts to produce a clean H 2 source are still under investigation. [29] For preventing CO generation, one approach is to use Pd as an alloy; [30] however, this method is more challenging and time-consuming. Another approach is to choose support for the synthesis of Pd that could lead to high efficiency.
Here, we studied the capability of Pd-SiO 2 in the catalytic decomposition of formic acid toward the selective generation of H 2 in the aqueous media in an effort to generate H 2 without any side product. Figure 7A shows the time course of H 2 production on Pd-SiO 2 at different temperatures of 20, 50, and 70 °C. The H 2 production rate increased with the increase in temperature, and the reaction was completed within 90 min at 70 °C. CO was not detected during H 2 production for all tests. We also monitored the reaction at 70 °C over 500 min by exchanging the reaction media's internal gas with Ar gas and restarting the reaction every 90 min, finding no significant decrease in the H 2 generation rate ( Figure 7B). Furthermore, a hot-filtration test was applied in the last cycle at 30 min. The reaction did not have any noticeable progress, implying that Pd species' leaching does not occur in the reaction media. These results demonstrate the catalytic activity and high durability of Pd-SiO 2 toward selective H 2 production from formic acid. The high durability of the catalyst is reasonable because the Pd nanoparticles are covalently attached to the silicate sheet component.

www.advmatinterfaces.de
Since a highly recyclable catalyst should also be resistant during the early time intervals of the reaction, [31] we examined the recyclability of the Pd-SiO 2 in terms of turnover frequency (TOF) in the first 10 min of the reaction progress for six consecutive cycles ( Figure 7C, note that, here, in the calculated TOF (h −1 , bulk), through the following formula [H 2 evolution rate (mole h −1 )]/[the amount of all the loaded Pd (mole)], all Pd species (surface and subsurface Pd) are included. The rate per gram of used Pd was also calculated for each to have a further comparison (see Figure S6, Supporting Information). This examination indicated that Pd-SiO 2 has insignificant deactivation in the catalytic activity over each cycle where TOF exhibits a trivial decrease. We further measured the XRD pattern for the recovered sample and observed Pd's morphology by STEM images. Eventually, we confirm an insignificant structural and morphological change for the Pd particles ( Figures S2 and S7, Supporting  Information).
Importantly, Pd-SiO 2 exhibits state-of-the-art activity for the catalytic reaction ( Figure 7D). We prepared three reference catalysts, Pd-GO, [23] Pd-PMO (isotropic 1.8 nm-sized Pd supported within a periodic mesoporous organosilica) [4a] and Pd-HNC (mainly compose of single-atom Pd stabilized on hollow N-doped carbon cage), [32] and their catalytic activities were compared with that of Pd-SiO 2 under the identical conditions. The TOF of Pd-SiO 2 (131 h −1 ) was considerably higher than that of Pd-GO (0.9 h −1 ) and Pd-PMO (15 h −1 ). Additionally, to our surprise, the TOF on Pd-SiO 2 is much higher than that of Pd-HNC (38 h −1 ). Since calculating the surface density of Pd particles in Pd-SiO 2 is elusive, we compared the bulk TOF values of the samples, by counting the actual Pd-loading amount as the total number of Pd active sites. Although this assumption shows a TOF value lower than the real TOF value for Pd-SiO 2 , the results reveal that the single atom-based Pd catalyst (Pd-HNC) still shows a lower TOF (lower catalytic activity) than Pd-SiO 2 . Therefore, despite the high quantity of the active sites in Pd-HNC, the catalytic activity was lower than Pd-SiO 2 . In the case of the calculating the TOF for Pd-PMO, same strategy was taken. Pd-PMO with slightly lower average particle size (1.8 nm) than that of Pd-SiO 2 (2.4-nm thickness in average), despite its higher surface density, has significantly lower TOF than that of Pd-SiO 2 . Furthermore, the catalytic activity of a Pd-loaded layered silicate (same support as Pd-SiO 2 has), which has a significantly larger Pd particle size (≈36 nm, prepared with another method) than that of Pd-SiO 2 , had significantly lower catalytic activity (for characterization and catalytic TOF results, see Figure S8, Supporting Information). This observation confirms that when Pd nanoparticles are too thick, they suffer from lower active catalytic site densities and, thus, show lower catalytic activity. On the one hand, too thin Pd's thicknesses (<1 nm), due to strong hybridization with oxygen and consequently, a lowered energy level of the Pd 4d band with respect to the Fermi level, a low catalytic activity can be expected. Therefore, to achieve the best catalytic activity, the thickness of Pd nanoparticles on supports should fall within a goldilocks range (the ones that are ≥1 nm and close to 1-nm particle size Pd) to ensure its high electron lability.

Conclusion
We have synthesized several-atomic-layer thick Pd nanoplates supported on silicate sheets via the partial exfoliation of a layered silicate modified with iminodiacetate group and the subsequent removal of the organic functional groups. In addition to the high durability and reusability of the supported Pd on layered silicate, the catalytic activity toward H 2 production from formic acid in water was considerably more elevated than the existing supported Pd catalysts such as those composed of larger-size and single-atom Pd or those of that were synthesized on different supports. Our density functional calculations demonstrated the necessity of at least six atomic layers of the Pd attached to the silicate sheets to obtain the metalliclike electronic properties and then optimize catalytic activity. These results imply that despite the tremendous efforts that have been directed toward downsizing metal nanoparticles (toward single atom), controlling their 2D structures offer a more facile and robust solution for creating metal catalysts with better performances.

Experimental Section
Materials Characterization: XRD patterns of powder samples were collected using a powder X-ray diffractometer (Smart Lab, RIGAKU) with Cu Kα radiation at 40 kV and 30 mA. 29 Si MAS and 13 C CP MAS NMR spectra were recorded at 119.17 and 150.87 MHz, respectively, on a Varian 600PS solid NMR spectrometer using a 6-mm diameter zirconia rotor. The products' composition (Na, Si, and Pd) was determined by the inductively coupled plasma optical emission spectroscopy (ICP-OES) of the dissolved samples using an Agilent 5800 manufactured by Agilent Technologies. The morphology of powder samples was observed with a HITACHI SU-8000 SEM. TEM images were taken with a JEOL JEM-2100F microscope (operated at 300 kV). The AFM images for measuring the Pd thicknesses were recorded on a high-resolution scanning probe microscope Shimadzu SPM-8000FM using dynamic and phase mode (tapping mode), with a cantilever namely Olympus AC160TC. XPS spectra were recorded on a PHI Quantera SXM instrument (ULVAC-PHI). The energies were calibrated as C 1s peak as 285.0 eV.
Synthesis of IDA-SiO 2 : NH 2 -SiO 2 (0.2 g) was dispersed in sodium chloroacetate (20 mmol, Sigma-Aldrich, 98%) in 20 mL H 2 O. Then, ethanolic (20 mL) solution of triethylamine (30 mmol) was added to the dispersion and stirred under reflux conditions for 48 h. The solid product obtained by centrifugation was washed with ethanol (30 mL) three times and dried at 70 °C for 4 h.
Synthesis of Pd-SiO 2 : An aqueous solution of K 2 PdCl 4 (4.6 µmol, Strem Chemicals, Inc., 99%) in H 2 O (10 mL) was added dropwise to IDA-SiO 2 (0.1 g) dispersed in ethanol solution (40 mL). The mixture was then sonicated for 3 min and stirred at room temperature for 15 min, and dried at 70 °C. Then, the obtained powder was redispersed in H 2 O:EtOH (1:1, 40 mL), and then, a methanolic solution (5 mL) of NaBH 4 (5 mg) was added dropwise to the mixture while it was stirring. After 30 min, the reaction was stopped, and the final solid was centrifuged, washed with distilled water and ethanol three times, and dried at 40 °C in a vacuum. Pd deposited on the outer surface of the layered silicate (Pd@Na-SiO 2 ) was prepared by the similar method used for synthesizing Pd-SiO 2 except that Na-SiO 2 was directly modified.
Density Functional Calculations: Density functional calculations were performed with VASP code, [34] using the projector-augmented potentials. [35] Van der Waals interactions were included based on the DFT-D3 method developed by Grimme, Ehrlich, and Krieg. [36] The energy cutoff was set at 400 eV, while the k-points were sampled by 3 × 3 × 1 mesh produced by the Monkhorst-Pack scheme. All other precision keys for electronic minimization were set to accurate. The convergence of these settings has been examined for sodium metal oxides elsewhere. [37] The sodium silicate/Pd interface was simulated by cleaving the sodium silicate (Materials Project [38] mp-1210861) surface in the (001) direction. The thickness of the sodium silicate slab was set at two conventional unit cells containing 16 Na atoms, 48 silicon atoms, and 96 O atoms. The O atoms at the loose surface were capped with H. A Pd (001) slab of the 3 × 3 × 6 dimensions containing 108 Pd atoms was placed on the other end of the sodium silicate, making six atomic Pd layers. Configurations with three and nine atomic Pd layers were also calculated for comparison and were presented in the Supporting Information. The lattice parameters mismatch between Pd and sodium silicate slab was at a reasonable 2.85%. A vacuum slab of an ample length of 30 Å was added to the top of the Pd surface to build the supercell. All Pd and H atoms and three atomic layers from sodium silicate closest to the interface were allowed to relax to forces smaller than 0.01 eV Å −1 . The relaxed structure was supplied in the Supporting Information in the POSCAR format.
Catalytic Tests: The sample powder was added into an aqueous solution of formic acid (6.5 mmol, 5 mL; supplied by FUJIFILM Wako Pure Chemical Corporation) in a Pyrex glass tube (34 mL), sonicated, and deaerated via bubbling with Ar gas. Then, the tube was sealed with a rubber septum and stirred at different temperatures (20, 50, and 70 °C). At a given time, the headspace gas in the glass tube was withdrawn with a gastight syringe and quantified on a Shimadzu GC-2010 plus gas chromatograph equipped with a barrier-discharge ionization detector.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.