Towards Atomically Precise Supported Catalysts from Monolayer-Protected Clusters: The Critical Role of the Support

Graphical Abstract

Small and shiny: X-ray absorption spectroscopy and density functional theory computations show that the acidity of the support defines the fate of gold–phospine clusters after deposition.

Abstract

Controlling the size and uniformity of metal clusters with atomic precision is essential for fine-tuning their catalytic properties, however for clusters deposited on supports, such control is challenging. Here, by combining X-ray absorption spectroscopy and density functional theory calculations, it is shown that supports play a crucial role in the evolution of monolayer-protected clusters into catalysts. Based on the acidic nature of the support, cluster-support interactions lead either to fragmentation of the cluster into isolated Au–ligand species or ligand-free metallic Au⁰ clusters. On Lewis acidic supports that bind metals strongly, the latter transformation occurs while preserving the original size of the metal cluster, as demonstrated for various Au_n sizes. These findings underline the role of the support in the design of supported catalysts and represent an important step toward the synthesis of atomically precise supported nanomaterials with tailored physico-chemical properties.

Introduction

Sub-nanometer metal clusters possess unique electronic, optical, magnetic and chemical properties.¹ These small metal clusters are particularly attractive for catalysis due to the accessibility of nearly all metal atoms to gas or liquid reactants rendering the highest possible efficiency in metal utilization.² In the sub-nanometer regime the properties of metal clusters are often non-scalable with their atomicity. For example, the addition/removal of a single metal atom to/from a cluster may lead to a drastic change in its properties, meaning that every metal atom in a cluster is important.³ Examples of dramatic effects of the nuclearity on catalytic properties include oscillatory behaviour of Au clusters with even or odd number of atoms for adsorption and activation of small molecules shown for gas-phase Au clusters;^3b-4 activity of Au clusters only with 5–10 atoms in the oxidation of thiophenol⁵ or with 3–6 atoms in coupling reactions;⁶ activation of methane over gold clusters;^{4d, 7} remarkable difference in hydrogenation activity for carbon-supported Pt₈, Pt₉ and Pt₁₀,⁸ etc.⁹ Furthermore, such cluster size effects in the sub-nm regime were observed even in reactions that are generally regarded as structure-insensitive.¹⁰ In order to understand such atomic level cluster size effects, an extremely challenging control over the nuclearity and uniformity of supported clusters is required.

For most applications, including catalysis, metal nanoparticles have to be supported on a high surface area solid carrier primarily to prevent them from agglomeration. Although a number of ways to synthesize well-defined metal clusters in the gas phase or solution exist, controlling the size and uniformity of clusters on supports is extremely challenging. For example, monodisperse organometallic precursors with the desired size are deposited onto porous supports with subsequent activation at high temperature to unblock the active metal surface from organic ligands.¹¹ However, the latter step often leads to cluster sintering with the loss of the original size and uniformity.¹²

Currently, oxide-supported low-nuclearity carbonyl clusters of group VII–IX metals are best understood systems in terms of immobilization, interaction with supports and de-ligation.^11a-11c Cluster-support interactions are known to play a major role in the formation of active metal nanoclusters.

Monolayer-protected Au, Pt and Ag clusters have attracted a lot of interest due to the possibility to precisely control their size and composition in a uniform manner.^{8, 11f, 11h, 13} Thiolate-protected Au clusters have been actively investigated in recent years as precursors for heterogeneous catalysts,^{11g, 11h, 13b, 13g, 13h, 14} however less knowledge is available for the phosphine-protected atomically precise clusters.¹⁵ For example, interaction between phosphine clusters and support materials is not fully understood and studied. Gaining a better understanding of the behaviour of metal-organic clusters on supports is of utmost importance for the design of well-defined supported nanomaterials.

Here, by employing phosphine-stabilized Au_n clusters (with n=1, 8, 9) and clusters with the average size n=101, and a combination of X-ray absorption spectroscopy methods and density functional theory (DFT) calculations, we have investigated the interaction between the clusters and four different supports: silica, graphitic carbon, titania and ceria. We demonstrate that these stable clusters readily react with these supports already at temperatures as low as 60 °C, undergoing fragmentation via two different mechanisms, governed solely by the surface chemistry of the support. We observe that the clusters break down by “oxidative fragmentation” into Au–PPh₃ species when placed on Brønsted acidic supports such as silica and oxidized graphitic carbon, while on Lewis acidic supports such as titania and ceria, the clusters break down by “ligand migration” into metallic Au_n clusters and PPh₃ species on the surface of the support.

Au/CeO₂ catalysts show high activity in the oxidation of CO, water–gas shift, selective oxidation and other reactions.¹⁷ Here, we report the catalytic activity of the well-defined Au_n clusters formed after ligand migration on the CeO₂ surface in the oxidation of carbon monoxide and show that the size of Au_n clusters significantly affects the catalytic activity. These insights open new strategies for designing atomically precise, monodisperse, supported heterogeneous catalysts that can take advantage of the size-dependent properties of the metal clusters for controlling catalytic reactions.

Results and Discussion

Cluster immobilization on supports at ambient temperature

We have performed experiments and analysis for Au_n clusters; n=1, 8, 9 and 101, however, here we focus on the results for n=9 and show our results for n=1, 8 and 101 in the Supporting Information. The clusters for n=1, 8, 9 and 101, correspond to AuL(NO₃), Au₈L₈(NO₃)₂, Au₉L₈(NO₃)₃ and Au₁₀₁L_x(NO₃)_y, respectively, where L=PPh₃. These clusters are deposited on four different supports, SiO₂, TiO₂, CeO₂ and oxidized graphitic carbon (for the latter support only data for the thermally treated sample was recorded). Note that the NO₃⁻ counter-anions are washed away during the sample treatment, before characterization. Table 1 shows surface acid characteristics. SiO₂ and graphitic carbon contain only Brønsted acid sites with silanols^11c on SiO₂ or carboxylic, lactonic and hydroxyl groups¹⁸ on carbon. TiO₂, in addition to the low density of Brønsted acid sites, also shows Lewis acidity, while CeO₂ has only Lewis acid sites.

Table 1. Support Lewis/Brønsted acid characteristics.

Support	Point of zero charge	Brønsted acid sites [nm^−2]^[a]	Lewis acid sites [nm^−2]^[a]
SiO2	4.1	0.2	–
Carbon	4.0	0.16^[18][b]	–
TiO2	3.8	0.003	0.74
CeO2	4.4	–	0.21

• [a] determined from Pyridine-IR (Figure S2). [b] Determined by acid–base titration.

Figure 1 a shows the structure of the unsupported [Au₉L₈]³⁺ cluster. The cluster has an icosahedral positively charged Au core of around 1 nm with NO₃⁻ as a counter-ion and a formal charge on each Au atom of +0.3. (electrospray ionization mass spectra and transmission electron microscopy image of the unsupported cluster are shown in Figure S1. Figure S5 shows the structure of the unsupported [Au₈L₈]²⁺ cluster and further structural characteristics of n=1, 8 and 9 clusters are listed in Table S1 in the Supporting Information). Phosphines are L-type ligands that bind to Au via a dative interaction, where PPh₃ acts as a Lewis base due to the lone pair of electrons on P, and Au acts as a Lewis acid.¹⁹

Figure 1

(a) Structure of [Au₉(PPh₃)₈]³⁺;¹⁶ (b) Au L_III XANES and (c) Fourier transformation of the EXAFS spectra of [Au₉L₈]³⁺ unsupported and when supported on oxides dried at 25 °C. The FT spectra are uncorrected for the phase shift.

Figure 1 b shows the X-ray absorption near edge structure (XANES) spectra for the unsupported cluster (red), supported clusters and a metallic Au foil (black). The peak at around 11 920 eV (“white line”) seen for the unsupported [Au₉L₈]³⁺, and different oxide-supported [Au₉L₈]³⁺ dried at 25 °C (denoted Au₉L₈/MO₂_25) confirms the positive charge on gold in these samples.

Figure 1 c shows the Fourier transform (FT) analysis of the Au L_III-edge extended X-ray absorption fine structure (EXAFS). For the unsupported [Au₉L₈]³⁺ (red), peaks due to Au−P and Au−Au bonds are observed at 2.28 and 2.7–3.1 Å, respectively, with both coordination numbers (CN) being one (Tables 2 and S1). The Au−Au CN is lower than expected for a cluster of 9 atoms. The underestimation of Au−Au CN is known for monolayer-protected clusters and arises from their disordered low-symmetry structures.^{13b, 14a} Our DFT calculated value for the average Au−P bond length is 2.37 Å and Au−Au lengths range between 2.86–3.02 Å, for the [Au₉L₈]³⁺ cluster in gas phase, in good agreement with the experimental values. The spectra of [Au₉L₈]³⁺ and Au₉L₈/MO₂_25 are similar with only small differences in the Au−Au bond region. This suggests that on all three supports considered here, the deposited [Au₉L₈]³⁺ clusters remain essentially unaltered and retain their core structure, the PPh₃ ligands and positive charge on gold when deposited on supports followed by drying at ambient temperature.

Table 2. Structural characteristics of Au₉L₈ clusters on various supports dried at ambient and elevated temperature based on curve fitting analysis of Au L_III-edge EXAFS (see k³-weighted spectra in the Supporting Information).

Catalyst	Bond	CN^[a]	r [Å]^[b]	Δσ^2[c] 10^−13
Au9L8/SiO2_25	Au−Au	2.2 (6)^[d]	2.71 (6)	2.5 (3)
	Au−P	0.8 (5)	2.32 (1)	1.9 (3)
Au9L8/CeO2_25	Au−Au	3.0 (5)	2.69 (1)	1.6 (3)
	Au−P	0.8 (2)	2.29 (2)	1.0 (1)
Au9L8/TiO2_25	Au−Au	2.3 (5)	2.71 (5)	2.6 (3)
	Au−P	1.0 (3)	2.28 (3)	1.5 (2)
AuL/SiO2_120	Au−Au	1.2 (5)	3.01 (3)	4.1 (3)
	Au−P/Au−O	1.6 (3)	2.23 (3)	8.4 (4)
Au9/CeO2_120	Au−Au	4.5 (5)	2.83 (3)	4.1 (5)
	Au−O	0.5 (2)	2.23 (5)	1.2 (3)
Aun/TiO2_120	Au−Au	7.8 (2)	2.86 (5)	3.2 (2)
	Au−P/Au−O	–	–	–
AuL/Carbon_60	Au−Au	1.0 (1)	3.41 (3)	6.0 (7)
	Au−P/Au−O	1.8 (3)	2.19 (5)	6.0 (2)

• [a] CN coordination number. [b] Bond length. [c] Relative Debye–Waller factor: σ²=(σ_sample)². [d] Number (x) in brackets indicates a standard deviation of ±0.x to the value before it.

Oxidative fragmentation of gold-phosphine clusters on Brønsted acidic supports

We show that [Au₉L₈]³⁺ clusters supported on Brønsted acid supports, such as SiO₂ and carbon, undergo oxidative fragmentation and disintegrate into ensembles of Au-PPh₃ (AuL) fragments upon heating to temperatures of 120 °C on SiO₂ (AuL/SiO₂_120) and 60 °C on graphitic carbon (AuL/Carbon_60). This result is evidenced by several observations that we describe below.

Figure 2 a shows the UV/Vis spectra of the unsupported [Au₉L₈]³⁺ (in methanol), and Au₉L₈ supported on SiO₂ at 25 and 120 °C. The optical absorption features characteristic of the unsupported [Au₉L₈]³⁺ (in methanol), are also seen for Au₉L₈/SiO₂_25 but they disappear upon heating the system to 120 °C. This clearly indicates that [Au₉L₈]³⁺ clusters on SiO₂ undergo a structural change. Similar result for n=8 cluster is shown in Figure S6. (Due to the band gap peak of the semiconductor supports (CeO₂ and TiO₂) or a complete light absorption on a carbon-supported catalyst, the changes in the UV–visible spectra of the supported cluster upon thermal treatment were not as pronounced on other supports as on SiO₂).

Figure 2

(a) UV/Vis of the unsupported [Au₉L₈]³⁺ in methanol, DR UV/Vis and photographs (inset) of Au₉L₈/SiO₂_25 and AuL/SiO₂_120. (b) FT of Au L_III EXAFS (uncorrected for the phase shift) and (c) Au L_III XANES of supported [Au₉L₈]³⁺ (at total Au loading 0.2 wt.%, see Table 3) after thermal treatment and reference compounds. FT EXAFS intensity of Au foil was reduced 2.5 times to fit in the Figure. (d) HAADF-STEM of Au₉/CeO₂_120. Au nanoclusters are seen as white dots. (e) Experimental Au L_III XANES of Au₉/CeO₂_120 and calculated XANES of Au₁₃.

The FT EXAFS spectra in Figure 2 b show that after thermal treatment the peaks due to Au−Au bonds in the original [Au₉L₈]³⁺ cluster (red curve) are absent in AuL/SiO₂_120 (orange curve) and AuL/Carbon_60 (pink curve), and only a small Au−Au bond peak is observed at 3.4 Å, suggesting that Au−Au bonds have broken. Furthermore, we also show that the FT EXAFS of both supported samples are very similar to that of the isolated AuPPh₃NO₃ complex (light green curve) in Figure 2 b. This strongly suggests the fragmentation of Au₉L₈ clusters into AuL species.

We provide additional support in Figure 2 c, where we show that the shape of the white line in the XANES spectra of AuL/SiO₂_120 and AuL/Carbon_60 is similar to that of AuPPh₃NO₃. The formal oxidation state of Au in AuLNO₃ is +1. The similarity between the XANES features of the samples and [AuL]¹⁺ suggests the same charge of +1 on Au. Hence, the thermal evolution of [Au₉L₈]³⁺ on SiO₂ and carbon is accompanied by an increase in the formal oxidation state of Au from +0.3 to +1.

Unsupported [Au₉L₈]³⁺ clusters are stable up to 250 °C under both oxidizing and reducing conditions (gravimetric analysis shown in Figure S3), therefore the observed fragmentation of [Au₉L₈]³⁺ clusters at a considerably lower temperature (<120 °C) is the result of their chemical interaction with the Brønsted acid surface groups on SiO₂ and oxidized graphitic carbon. Hence, by analogy with molecular acids, the hydroxyl groups on SiO₂ or carboxylic groups on carbon oxidize gold in [Au₉L₈]³⁺ clusters to yield surface-bound L-Au-O-Si/C complexes. On acidic carbon, clusters fully fragment already at 60 °C (cf. 120 °C for SiO₂) likely due to the higher Brønsted acid strength of the carboxylic groups on carbon relative to that of hydroxyls on SiO₂.

The formation of support-bound L-Au-O-Si/C complexes is evidenced by the increase in the Au-P/O coordination number relative to the unsupported clusters: in [Au₉L₈]³⁺ and [AuL]¹⁺ the Au-P CN is found to be 1, while in AuL/SiO₂_120 and AuL/Carbon_60, it increases to 1.6 and 1.8, respectively. This is also supported by the shortened average Au−P/O bond length due to the contribution from the shorter Au−O bond (Tables 2 and S1). The presence of the Au−Au bond peak at 3.4 Å, that is, at non-bonding distances, for AuL/SiO₂_120 and AuL/Carbon_60 shows that the Au-phosphine complexes exist on the supports as ensembles, similar to isolated Au^I-phosphine complexes that tend to form dimers/polymers due to aurophilic Au−Au interactions.²⁰

This oxidative fragmentation of phosphine-protected Au clusters aided by the Brønsted acidic groups of the supports is similar to the behaviour of carbonyl clusters of more oxophilic metals of Groups VII–IX (e.g. Ir, Rh, Os, Re).^11c Unlike phosphine stabilized Au clusters, the more reactive metal carbonyl clusters were shown to react with metal oxide surfaces already upon chemisorption, and fragment at elevated temperatures with the evolution of H₂ and CO, confirming oxidative addition of surface hydroxyls to metal.^{11c, 21} It is hence intriguing that the rather inert phosphine-protected Au clusters react with the Brønsted acidic groups of supports at elevated temperature in a similar manner as metal carbonyls.

Ligand migration from gold-phosphine clusters to Lewis acidic supports

In contrast to the behaviour of the [Au₉L₈]³⁺ clusters on Brønsted acid supports, on Lewis acid supports, such as CeO₂ and TiO₂, upon heating to a temperature of 120 °C, the ligands migrate from the cluster to the Lewis acid sites on the support leaving behind ligand-free metallic Au_n clusters on the surface. On CeO₂, the Au₉ cluster seems to maintain its size (Au₉/CeO₂_120), while on TiO₂, the cluster is found to agglomerate to form somewhat larger Au_n clusters (Au_n/TiO₂_120).

FT EXAFS of both samples indicate complete migration of phosphine ligands from gold evidenced by the absence of the Au-P bond peak (Figure 2 b and Table 2). The metallic state of Au in Au₉/CeO₂_120 and Au_n/TiO₂_120 is evidenced by the absence of the white line in the XANES spectra in Figure 2 c. The spectra seem similar now to the spectrum of the metallic Au foil. The XANES and FT EXAFS spectra for n=1 and 8 clusters on CeO₂ are shown in Figure S7. The structural characteristics of n=1, 8 and 101 clusters on CeO₂ are listed in Table S2. These results suggest that migration of the phosphine ligands to CeO₂, while preserving the original cluster size, occurs irrespective of the size of the cluster.

For Au₉/CeO₂, the Au−Au coordination number increased to 4.5 and Au−Au bonds elongated to 2.83 Å upon heating to 120 °C, thus indicating the rearrangement of the Au core with equalization of the Au−Au bond length (Figure 2 b, Table 2). Figure 2 d shows the HAADF-STEM of Au₉/CeO₂_120 with Au clusters of 1 nm in agreement with the size of the unsupported [Au₉L₈]³⁺ (Figure S1). For an icosahedral or cuboctahedral cluster of 13 atoms, the expected CN lie in the range of 5.5–6.5.²² Hence, the short Au−Au bond (2.83 Å),²³ the absence of the higher shell Au−Au bonds with distances above 4 Å, and the low Au−Au CN (4.5) for Au₉/CeO₂_120 show that the resulting phosphine-free Au clusters on ceria are very small and the majority of clusters maintain the original size of 9 atoms with negligible sintering, if any. Furthermore, a comparison of the experimental XANES of Au₉/CeO₂_120 and simulated XANES of a model Au₁₃ cluster (see Figure 2 e) shows that XANES features are less pronounced for Au₉/CeO₂_120, which supports our conclusion that the average cluster size in this sample is not higher than 13 atoms.²⁴

On TiO₂, phosphine-free Au⁰ clusters sinter to form larger particles as evidenced by the increased Au−Au CN (7.8), larger Au−Au bond lengths of 2.86 Å and the appearance of the higher shell Au−Au bond peaks at 4–7 Å (Figure 2 b, Table 2). The cluster size is preserved on CeO₂ likely due to its higher number of surface defects (oxygen vacancies) and stronger metal adhesion compared to TiO₂.²⁵ We note that electronic properties of CeO₂ in Au₉/CeO₂_120 in the vicinity of gold clusters are probably altered due to interaction with phosphine ligands (vide infra).

What causes the loss of phosphine ligands from gold clusters? P1s X-ray photoelectron spectroscopy (XPS) shows that phosphine species are present in Au₉/CeO₂_120 (see Figure S4) therefore they must reside on the support. Phosphine species function as Lewis bases and form dative bonds to Lewis acid sites (under-coordinated Ti⁴⁺ and Ce⁴⁺) on TiO₂ and CeO₂.²⁶ Therefore, the stronger interaction of the phosphines with the Lewis acid sites compared to the interaction with Au is the reason for the observed ligand migration from the clusters to the support. TiO₂ and CeO₂ surfaces are both hydroxylated,²⁷ and TiO₂ used in this work shows weakly Brønsted acidic properties (point of zero charge of ca. 3.8, Table 1), however no Au cluster fragmentation occurs on these supports. This suggests that the presence of Lewis acid sites determines the specific pathway for the evolution of these clusters when both Brønsted and Lewis acid groups co-exist on the support surface. This reasoning is also supported by our DFT calculations described below.

Oxidative fragmentation vs. migration of ligands of [Au₉L₈]³⁺ clusters on supports: DFT analysis

To investigate the thermodynamic driving forces behind the interactions of the [Au₉L₈]³⁺ cluster with Brønsted acid and Lewis acid surfaces, we resort to density functional theory (DFT) calculations. We study the interaction of the cluster with hydroxylated amorphous silica (HO-SiO₂) and hydroxylated ceria (111) (HO-CeO₂) surfaces; on both supports, we consider the two mechanisms: (1) oxidative fragmentation into AuL species and (2) ligand migration forming Au₉ and L species on the supports.

In the first case, the [Au₉L₈]³⁺ cluster breaks down into eight AuL species each removing a hydrogen from one of the surface hydroxyl groups thereby forming L-Au-O-MO₂; M=Si/Ce. The remaining Au in the [Au₉L₈]³⁺ cluster also removes a H from a hydroxyl group and forms Au-O-MO₂. Equation 1 shows the considered reaction:

(1)

Note that this equation is formally charge-imbalanced. However, we have implicitly taken into account the enthalpy contribution to reduce the [Au₉L₈]³⁺ in the left-hand side of Eq. (1) by performing calculations in a Born–Haber cycle (to be shown later). The lowest energy configurations obtained by DFT for the species L-Au-O-MO₂ and Au-O-MO₂ for M=Si are shown in Figures 3 a,b, respectively, while those for M=Ce are shown in Figures 3 e,f, respectively.

Figure 3

Lowest energy geometries obtained from DFT for different species on hydroxylated amorphous silica and hydroxylated ceria (111) surfaces. (a)–(d) AuL and Au species formed by oxidative fragmentation, and Au₉ and L formed by ligand migration, respectively on the SiO₂ surface. (e)–(h) AuL and Au species formed by oxidative fragmentation, and Au₉ and L formed by ligand migration, respectively on the CeO₂ surface. Au, P, C, O, Si and Ce are shown by yellow, blue, dark grey, pale green, pink and pale blue colours, respectively. H atoms in the phenyl ring and on the surface hydroxyl groups are shown by light grey and light pink colours, respectively.

In the second case, the [Au₉L₈]³⁺ cluster breaks down into metallic Au₉ and L species. The Au₉ cluster adsorbs on the support by interacting with surface O atoms, while the ligand L forms a dative bond with the cation in the oxide. The reaction (formally charge-imbalanced, see above) can be written as in Equation 2:

(2)

The lowest energy configurations obtained for Au₉ and L species on HO-SiO₂ are shown in Figures 3 c,d, respectively, while those on HO-CeO₂ are shown in Figures 3 g,h, respectively.

We show the energetics for the two mechanisms on both supports in Figure 4. The top panel (a) describes oxidative fragmentation of the [Au₉L₈]³⁺ cluster into AuL species by removal of hydrogen from the surface. The reaction enthalpy for the complete reaction is given by ΔH_tot and is calculated to be −9.20 eV and −10.26 eV, for HO-SiO₂ and HO-CeO₂, respectively. The energy contributions to the reaction enthalpy can be split into three parts in this case: ΔH_tot=ΔH_rd+9ΔH_rh+ΔH_int, where ΔH_rd is the energy to reduce and disintegrate the cluster into AuL species in gas phase, ΔH_rh is the energy to remove a neutral H atom from a hydroxyl group on the surface and ΔH_int is the energy due to the interaction between the disintegrated fragments and the surface after removal of H. ΔH_rh for both surfaces were calculated for the geometries of O-MO₂ that correspond to the lowest energy geometry obtained for L-Au-O-MO₂.

Figure 4

The Born–Haber cycle for (a) oxidative fragmentation and (b) ligand migration for the two supports, HO-SiO₂ and HO-CeO₂. All numbers in pink correspond to values for the hydroxylated amorphous SiO₂ surface, while the numbers in blue correspond to energy values for the hydroxylated CeO₂ (111) surface. ΔH_tot is the total reaction enthalpy for the processes, ΔH_rd is the energy to reduce and disintegrate the cluster into AuL species in gas phase, ΔH_rp is the energy to reduce and peel the ligands from the cluster forming Au₉ and L species in gas phase, ΔH_rh is the energy to remove a H atom from the surface and ΔH_int is the energy due to the interaction between the end species and the surface. Note that the Born–Haber cycle implicitly takes into account the charge imbalance in the total reaction enthalpy ΔH_tot.

The second panel (b) in Figure 4 describes ligand migration from the [Au₉L₈]³⁺ cluster to the surface, leaving behind metallic Au₉. The reaction enthalpy, ΔH_tot, for the reaction on the two surfaces is calculated to be −5.44 and −11.62 eV for the HO-SiO₂ and HO-CeO₂ surfaces, respectively. In this case, the energy contributions to the reaction enthalpy is split into two parts: ΔH_tot=ΔH_rp+ΔH_int, where ΔH_rp is the energy to reduce and peel the ligands from the Au₉ cluster in gas phase and ΔH_int is the interaction energy between the disintegrated fragments, Au₉ and eight L, with the surface.

By comparing the total reaction enthalpies for the two mechanisms on silica, −9.20 eV for oxidative fragmentation vs. −5.44 eV for ligand migration, we find that DFT supports the experimental observation: there is a driving force for the surface-cluster interaction to form AuL species. Note that we assume here that the entropy contributions are similar for both mechanisms. By comparing the energetics for the disintegration of the cluster in gas phase, we see that [Au₉L₈]³⁺ energetically prefers to reduce and peel the ligands forming Au₉ and L (−7.54 eV) than to reduce and disintegrate forming AuL species (−0.63 eV). However, the formation of Au₉ and L on SiO₂ is not favoured due to the endothermic interaction (+2.10 eV) between the species and the silica surface. The Au₉ cluster binds weakly to the surface with a binding strength of −0.12 eV with respect to Au₉ in gas phase and the bare support, however the binding of PPh₃ on the surface is unfavourable with an endothermic binding energy of +0.28 eV, calculated with respect to PPh₃ in gas phase and the bare support. This is expected on a surface such as SiO₂. The Si atoms are already four-coordinated and do not interact with the ligand species. The formation of the AuL and Au species, however, is highly favoured on this surface due to the high interaction energy between AuL and O-SiO₂. Thus, the silica surface induces the breaking of Au−Au bonds and the formation of O-Au-L bonds.

Similarly, when comparing the total reaction enthalpies for the two mechanisms on ceria: −10.26 eV for oxidative fragmentation and −11.62 eV for ligand migration, we see that again, DFT supports the experimental observation that the [Au₉L₈]³⁺ cluster interacts with the surface to form Au₉ and L species. The energy to reduce and peel the ligands is highly exothermic (−7.54 eV) and the fragments bind favourably to the surface (−4.07 eV). DFT suggests that the Au₉ cluster would have tendency to further decompose to 9 Au atoms bound on the surface (the calculated enthalpy is −4.06 eV with respect to the Au₉ cluster deposited on ceria). This is not observed in the experiments however, indicating that a kinetic barrier prevents this process. On the other hand, the oxidative fragmentation to AuL complexes bound to the surface requires the removal of H from the surface hydroxyl groups, which costs considerable energy. The binding strengths for the adsorption of Au₉ and the L on the HO-CeO₂ surface are calculated to be −1.76 and −0.29 eV, respectively, calculated with respect to the corresponding species in gas phase. The ceria surface favours the breaking of Au−P bonds and formation of P−Ce bonds.

Interestingly, the fragmentation of the cluster on the support is possible only in the presence of the ligands bound to gold. The energetics for oxidative fragmentation of an Au₉ cluster into Au atoms on the two surfaces is described in Figure S8. The considered reaction is Au₉+9 HO-MO₂→9 Au-O-MO₂+9/2 H₂. The reaction enthalpy, ΔH_tot, for the reaction on the HO-SiO₂ and HO-CeO₂ surfaces, is calculated to be +14.60 and +20.60 eV, respectively. The process is highly unfavourable. The fragmentation of the [Au₉L₈]³⁺ cluster is favoured on HO-SiO₂, in the presence of the ligands, only due to the high binding strength of the O-Au-L bond, which is calculated to be −4.57 eV, with respect to the [AuL]⁰ in gas phase and O-SiO₂ (after H is removed).

In Figure S9, we show energetics for the interaction between the n=1 cluster, [AuL]¹⁺, and two surfaces. On HO-SiO₂, the reaction enthalpies for oxidative binding and ligand migration are −6.35 and −4.00 eV, respectively, suggesting that the cluster prefers to bind to the surface forming L-Au-O-SiO₂ species. While on HO-CeO₂, the reaction enthalpies for oxidative binding and ligand migration are −6.57 and −6.88 eV, respectively supporting the finding that the ligand migrates from the Au atom to form Au adatoms and L species on the surface.

As an alternative approach to the one used in Eqs. (1) and (2), and Figure 4, one can explicitly introduce the charge balance by considering the enthalpy of the reactions 3 and 4:

(3)

(4)

Here, we define the chemical potential of an electron (μ_e) as the energy to take an electron from the valence band of the oxide support to the lowest unoccupied molecular orbital (LUMO) of the [Au₉L₈]³⁺ cluster in gas phase. The μ_e values were calculated to be −0.65 and −1.99 eV, for the supports HO-SiO₂ and HO-CeO₂, respectively. These energies were calculated after referencing the Kohn–Sham energies of the valence band of the surface and LUMO of the cluster, with respect to the same vacuum energy. Using this approach, we calculate the total reaction enthalpies for Eqs. (3) and (4) for the two surfaces and find the following. On HO-SiO₂, the enthalpies for the two processes are −7.24 and −3.48 eV, respectively, suggesting once again that oxidative fragmentation is the preferred disintegration mechanism on HO-SiO₂. On the other hand, on HO-CeO₂, the enthalpies for the two processes are −4.30 and −5.66 eV, respectively, suggesting that ligand migration is the preferred mechanism on HO-CeO₂.

Effect of the Au cluster nuclearity in CO oxidation catalysis

Au_n/CeO₂_120 systems with n=1, 8, 9, 101, obtained after the migration of the phosphine ligands, were tested for their catalytic activity toward the oxidation of carbon monoxide. CO oxidation over Au catalysts depends on many parameters, such as Au particle size, oxidation state, nature of the support and presence of water.²⁸ The activity of our Au_n/CeO₂_120 catalysts (see Table 3) increased in the order: Au₁<Au₈<Au₉<Au₁₀₁, in line with previous findings.^{28a, 29} The activity of the phosphine Au cluster-derived Au_n/CeO₂ is lower compared to values reported for Au/CeO₂ catalysts prepared using more traditional methods, which could be related to the small size of the Au clusters in Au_n/CeO₂, and the presence of PPh₃ ligands bound to the surface around the Au clusters. The PPh₃ may hamper oxygen supply from the support and/or efficient contact between Au and the support, thus leading to a lower catalytic activity than that observed for the conventionally prepared Au/CeO₂ catalysts. As described in the previous section, ligand migration leads to the formation of the metallic Au₈ and Au₉ clusters on CeO₂ upon heating to 120 °C. Interestingly, we find that the catalytic activity of Au₉/CeO₂ is 2.2 times higher that of Au₈/CeO₂ (Table 3). This difference is substantial considering that these clusters differ by only one atom.

Table 3. Catalytic activity of Au_n/CeO₂ in the oxidation of CO at 120 °C.

Catalyst	Au loading [wt.%]	TOFAu surf [s^−1]^[a]	TOFAu total [s^−1]^[b]
Au1/CeO2	0.14	0.009	0.009
Au8/CeO2	0.24	0.025	0.025
Au9/CeO2	0.15	0.055	0.055
Au101/CeO2	0.16	0.066	0.051

• [a] TOF normalized per surface Au sites. For Au₁, Au₈ and Au₉ clusters, all atoms were assumed to be accessible to reactants; while for Au₁₀₁ a dispersion of 78 % was used for TOF calculation. [b] TOF normalized per total Au.

Au clusters with even or odd number of atoms are predicted to display the so called “even–odd oscillations” in their properties as a result of having closed- or open-shell electronic structure, respectively.^{3a, 30} The electronic structure of a cluster has a direct impact on the binding and activation of small molecules, such as O₂ and CO.^{3a, 30} Hence, we link the observed difference in catalytic activity between Au₈/CeO₂ and Au₉/CeO₂ to the different electronic structure of Au₈ and Au₉ clusters.

Conclusions

We demonstrate that supports are highly reactive towards otherwise stable monolayer-protected atomically precise clusters under very mild conditions. The properties of the support determine the type of cluster-support interaction, which ultimately defines the type of the evolving metal species. The phosphine-stabilized Au clusters deposited on different supports evolve according to two distinct pathways: a) oxidative fragmentation into ensembles of surface-bound monoatomic Au–phosphine complexes upon interaction with Brønsted acid sites on a support; b) formation of ligand-free Au⁰ clusters due to the interaction of the ligands with Lewis acid sites on the support. On supports that bind metals strongly, such as CeO₂, the formation of phosphine-free Au⁰ clusters occurs without a change in the original cluster size, thus enabling control over the size of supported metal clusters by tuning the size of the metal-organic precursor. Preliminary tests for the reactivity of these atomically precise clusters show that a difference in size of even one Au atom has a large impact on their catalytic activity. Our findings highlight the importance of cluster–support interactions for the design of well-defined materials and catalysts using monolayer-protected clusters. Study of these atomically precise supported clusters combined with theoretical work is expected to further advance our understanding of catalytic processes at the atomic level.

Acknowledgements

The experimental work was supported by Utrecht University and the Netherlands Organization for Scientific Research (Dubble beamline projects n. 195.068.1047 and 195.068.1127). The theoretical work was supported by the Academy of Finland [grants 294217 (H.H.), 319208 (H.H.), 277222 (K.H.), and H.H.′s Academy Professorship]. The computations were done at the CSC computer centre in Espoo, Finland, and at Jyväskylä’s node in the Finnish national FGCI infrastructure. Authors thank Frank de Groot for discussions, Mark Isaacs and HarwellXPS facility for XPS acquisition, Petra Keijzer for TEM acquisition, Dennie Wezendonk for gravimetric analysis, and Sami Malola and Ville Korpelin for technical help in setting up the DFT models, and Marko Melander for helpful discussions. The DUBBLE staff Dipanjan Baneerje, Dirk Detollenaare and Florian Ledrappier are acknowledged for their support during set up of the beamline.

Conflict of interest

The authors declare no conflict of interest.