Very Important Publication Selective Aerobic Oxidation of 5-(Hydroxymethyl)furfural over Heterogeneous Silver-Gold Nanoparticle Catalysts

Bimetallic silver-gold alloy nanoparticles on zirconia with varying Ag/Au ratios were designed by a rational approach and tested as catalysts for the selective oxidation of the promising biomass platform molecule 5-(hydroxymethyl)furfural (HMF). For this purpose, colloidal AgxAu10-x particles with molar compositions x=1/3/5/7/9 were prepared by laser ablation in liquids, a surfactant-free method for the preparation of highly pure nanoparticles, before adsorption on zirconia. In-depth characterization of the supported catalysts evidenced alloyed nanoparticles with distinct trends of the surface and bulk composition depending on the overall Ag/Au molar ratio as determined by X-ray photoelectron spectroscopy (XPS) and Xray absorption spectroscopy (XAS), respectively. To uncover the synergistic effect of the Ag/Au ratio, the catalysts were further studied in terms of the catalytic activity and selectivity in HMF oxidation. Either the aldehyde moiety or both functional groups of HMF were selectively oxidized depending on the Ag/Au composition resulting in 5-hydroxymethyl-2-furan-carboxylic acid (HFCA) or 2,5-furandicarboxylic acid (FDCA), respectively. Optimization of the reaction conditions allowed the quantitative production of HFCA over most catalysts, also after re-use. Only gold rich catalysts Ag1Au9/ZrO2 and particularly Ag3Au7/ZrO2 were highly active in FDCA synthesis. While Ag3Au7/ZrO2 deactivated upon re-use due to sintering, no structural changes were observed for the other catalysts and all catalysts were stable against metal leaching. The present work thus provides fundamental insights into the synergistic effect of Ag and Au in alloyed nanoparticles as active and stable catalysts for the oxidation of HMF.


Introduction
The production of chemicals from renewable raw materials is becoming increasingly important as the growing demand for chemicals meets dwindling fossil resources, which currently account for the largest share of raw materials in the chemical industry by far. [1] One promising approach for the future synthesis of chemicals is the use of platform molecules from inedible biomass like lignocellulose. [1b,2] Partial fragmentation of biomass to sugars followed by acid-catalyzed dehydration results, for example, in 5-(hydroxymethyl) furfural (HMF), one of the most promising biomass platform molecules. [2a,3] Among other reactions like hydrogenation, [4] dehydrogenation [5] or hydrodeoxygenation [6] the selective oxidation of HMF has gained increasing interest in recent years, [1c,7] particularly due to the structural similarity of 2,5furandicarboxylic acid (FDCA) to terephthalic acid. [1c,8] Because of its significance as a bio-derived monomer, FDCA was even listed as one of twelve important molecules that could be produced in an industrial scale by the U.S. Department of Energy. [9] Selectively oxidizing the aldehyde moiety of HMF yields 5hydroxymethyl-2-furancarboxylic acid (HFCA), which also has possible applications as a monomer [1c,8] as well as a building block in high-value fine chemicals, like drugs. [10] The synthesis of FDCA starting from HMF has been studied using stoichiometric oxidants like KMnO 4 . [11] As the production of stoichiometric waste should be circumvented, [12] catalytic routes have been studied using homogeneous, heterogeneous, electro-, and biocatalysis. [1c,7] Currently, the AMOCO process is used for the large-scale industrial synthesis of FDCA. [13] FDCA is produced in 61% yield over a homogeneous Co/Mn/Br catalyst at 125°C and 70 bar air in acetic acid as solvent. [13a] Advantages such as simple separation of the catalyst have now shifted the focus more on the more demanding heterogeneously catalyzed oxidation. High FDCA yields are mostly obtained over noble metal catalysts like Ru, Pd, Pt or Au (Figure 1), [1c,7,14] which often require the addition of a homogeneous base to enhance FDCA solubility and accelerate the reaction. [15] As it is mostly considered an intermediate in FDCA synthesis, a small number of studies focus on the targeted synthesis of HFCA. Besides a chemical synthesis using the Cannizzaro reaction, [16] which is limited to a maximum HFCA yield of 50%, biocata-lytic reactions were employed for HFCA synthesis. [17] Also, heterogeneous catalysts gave high yields in organic solvents, which however affects the sustainability of the reaction. [18] In contrast, supported Agbased catalysts allow the quantitative production of HFCA in water under mild reaction conditions ( Figure 1). [19] Bimetallic catalysts are of particular interest in HMF oxidation as they were reported to be more active and stable upon re-use. [1c,7,20] The previously mentioned high selectivity of Ag towards HFCA and Au towards FDCA formation suggests that tuning the oxidation catalyst by alloy formation of Ag with Au appears to be especially attractive. However, to the best of our knowledge, this has not yet been conducted in HMF oxidation.
For other reactions it has been shown that bimetallic AgÀ Au catalysts were more active in selective [21] as well as total oxidation [22] reactions, e. g. the superior activity of bimetallic AgÀ Au catalysts in glucose oxidation was attributed to the activation of oxygen on Ag. [21a] It is well known that oxygen activation is often challenging on Au-based oxidation catalysts, [23] which on the other hand is favored on Ag. Hence, synergistic effects can be expected. In addition, both metals are fully miscible in every molar ratio.
To study how Ag modifies the Au catalyst for different molar ratios, catalysts with alloyed and comparable particle sizes are required. Despite their miscibility, the synthesis of alloyed AgÀ Au nanoparticles is not straightforward. Preparation methods include sequential deposition-precipitation, [24] coimpregnation, [25] flame-spray pyrolysis, [26] or colloidal preparation methods with chemical reduction. [27] The most commonly used precursors HAuCl 4 and AgNO 3 have several disadvantages, such as reduction of Au before Ag leading to the formation of a composition gradient throughout the bimetallic nanoparticle with a Au-rich core and a Ag-rich shell. [28] Further, AgCl may precipitate parallel to the reduction and high chloride concentrations can lower the catalytic activity of Au-based catalysts. [29] Working below the solubility limit of AgCl is impractical for the preparation of supported catalysts as these high dilutions complicate subsequent deposition. [30] Finally, capping agents or ligands might also affect the catalytic activity by blocking the active sites. [31] This may alter oxygen activation, catalytic activity, and selectivity in addition to the nanoparticle composition.
To overcome such shortcomings, a rational catalyst design is needed. [32] Pulsed laser ablation in liquids (LAL) has shown to be a promising alternative for the preparation of metal, oxide, and alloy nanoparticle catalysts. [33] In this method, a pulsed laser beam is focused onto the target immersed in a liquid, leading to rapid heating combined with the formation of a plasma plume. The superheated and evaporating material subsequently forms a jet of superheated matter which gets injected into the overlying liquid phase on a timescale of 10 to 100 nanoseconds, inducing rapid cooling with up to 10 12 K s À 1 . [34] Due to the heat transfer into the liquid, a cavitation bubble emerges, containing the nucleating and growing nanoparticles. The bubble collapses about 200 μs after the laser hits the target thereby releasing the nanoparticles. [35] Because of the uniform and rapid evaporation and nucleation kinetics, LAL is an efficient method to generate surfactant-free AgÀ Au nanoparticles with a homogeneous as well as average elemental distribution and high purity, rendering them as attractive systems to study the selectivity of Ag and Au for HMF oxidation to FDCA and HFCA.
Within this study, (1) Ag x Au 10-x /ZrO 2 catalysts were prepared by supporting on ZrO 2 nanoparticles generated via LAL with systematically varying the molar composition (Ag/Au ratio x = 1/3/5/7/9) with comparable size and (2) applied in the selective oxidation of HMF to HFCA and FDCA. The overall strategy is shown in Figure 2a. After thorough catalyst characterization concerning structure and composition, the effect of alloy formation on the catalytic activity and selectivity in aldehyde and alcohol oxidation was explored. The catalytic reactions were carried out in an aqueous medium using air as the oxidant.

Rational Design and Characterization of the Catalysts
The general approach of this study is depicted in Figure 2a. Nanoparticles were synthesized by LAL with a monomodal size distribution with average hydrodynamic diameters of 6-7 nm, except for Ag 3 Au 7 (12 nm) based on analytical disc centrifugation (ADC). The synthesis is described in detail in the experimental part and a scheme ( Figure 2a) and photograph of the setup is given in the supporting information ( Figure S1).
Respective particle size distributions are shown in Figure S3. From UV-Vis spectra of the resulting colloidal suspensions, shown in Figure 2b, single absorbance peaks that are shifted linearly with the composition of the particles can be observed (Figure 2c), indicating successful alloy formation. [36] Although the optical properties of the nanoparticles also depend on their size and shape, [37] the respective shifts are in line with literature where comparable particle sizes and shapes were investigated. [38] Slight differences arise from the presence of stabilizing ligands during biological or chemical syntheses, [39] whereas laser-generated alloy nanoparticles give the lowest surface plasmon resonance (SPR) wavelengths due to the absence of such agents. [36a,40] Hence, LAL proves to be an ideal method for the synthesis of alloyed AgÀ Au nanoparticles (AgAu NPs) compared to wet-chemical preparation. [41] Successful electrostatic adsorption of the nanoparticles on ZrO 2 was achieved by colloidal deposition under careful pH adjustment as previously described [42] and was observed both by the significant color change of the initially white ZrO 2 powder ( Figure 2a) and the depletion of AgAu NPs in the colloidal suspensions (Figure 2b) during the deposition process. After removing the aqueous supernatant, Ag x Au 10À x /ZrO 2 catalysts were dried and characterized by DR-UV-Vis confirming the successful adsorption of AgAu NPs on the support ( Figure S4a). In accordance with UV-Vis measured of the initial colloids, a composition-dependent linear shift of the SPR peak was found ( Figure S4b). The presence of the nanoparticles on the ZrO 2 support was further evidenced by STEM analysis. Particle size distributions, exemplarily shown for Ag 3 Au 7 /ZrO 2 in Figure 2d (for all other catalysts cf. Figures S14-S15), and mean particle sizes were derived from STEM images of the supported catalysts (Table 1). From the comparison to the initial size distributions measured for the colloids by ADC, the nanoparticle size of colloidal AgAu NPs were maintained after the deposition process for all catalysts (see Figure S15). In accordance with ADC measurements conducted before colloidal deposition, AgAu NP sizes of 7-11 nm were also found by STEM on the ZrO 2 support after deposition. Consequently, particle sizes were maintained during the deposition reported previously in other studies for laser-generated catalysts. [33b,43] EDX line scans of representativeAg 1 Au 9 /ZrO 2 and Ag 3 Au 7 /ZrO 2 catalysts showed a homogeneous distribution of both metals along the particle diameters ( Figure 2e, Figures S16-S17). ICP-OES verified the metal loading to be close to 1.5 wt.% in all but one case (Table 1). Additionally, the Ag/Au molar ratios are well in line with the nominal compositions of the target alloys. It can be observed from N 2 physisorption, that the specific surface area was maintained during the colloidal AgAu NP deposition process (Table 1). Since the deviations are within the margin of error, we can conclude that the support was not changed during the adsorption process and no pores were blocked by the metal particles, further confirmed by similar pore sizes of the pure ZrO 2 and the supported catalysts.
All catalysts were examined via XAS to further unravel the local electronic structure (oxidation state), charge transfer phenomena, and alloy formation. The former two can be derived from the X-ray absorption near edge structure (XANES) while the extended Xray absorption fine structure (EXAFS) provides information about alloy formation, as it is sensitive to the different backscattering of Ag and Au neighbors. [44] The overall oxidation state of Au can be estimated from the XANES region by the area of white line features in the Au L 3 spectra, which is related to the density of d states of the absorbing Au atom. [45] This area can be correlated to the d-electronic structure by using the difference spectrum obtained from the Au L 3 -and the L 2 -edge XANES spectra, which gives a single distinct peak. [44a, [45][46] Figure 3a shows the normalized Au L 3 XANES spectra of the catalysts along with Au foil for comparison. From the XANES features, Au is in the metallic state in all catalysts. Table 1. Metal loading and specific surface area of the supported catalysts and the ZrO 2 support material.
Entry Catalyst/Support Metal loading/wt.% Specific surface area/m 2 g À 1 Mean particle size/nm [a] Ag:Au molar ratio [b]  [a] Mean particle size based on TEM. [b] Values from ICP-OES with theoretical values given in brackets for comparison. However, the spectra differ in the white line region (11920-11940 eV) and in the post-edge region (11960-11980 eV) as shown in the inset of Figure 3a.
The variation in the white line feature can be attributed to electron transfer between Au and Ag. Figure 3b shows the corresponding difference spectra obtained by subtracting the Au L 2 -edge spectrum of Au foil from the samples where a single peak with varying width and area is obtained ( Table 2, detailed procedure in Supporting Information and Figure S2). In general, the calculated areas for AgAu catalysts are larger as compared to pure Au foil, thereby indicating AgÀ Au interaction, i. e. due to alloy formation. The larger area of Ag 1 Au 9 compared to Au foil ( Table 2, entries 1-2) indicates an increase in the d-hole density. [45] Hence, even at a low Ag fraction an electron transfer between Au and Ag is evident confirming strong AgÀ Au interaction, e. g. by alloy formation. From Ag 1 Au 9 to Ag 5 Au 5 ( Table 2, entries 2-4), the areas decrease with increasing Ag content. This reflects a decreasing number of d-holes, but all areas remain larger as compared to pure Au foil, thereby indicating AgÀ Au interaction. Based on the variation in the d-hole density, the interaction between Au and Ag appears to vary with the molar ratios. From this trend, more Ag atoms appear present in the outer layer while higher Au atom concentration is located in the center for Agrich composition. On the other hand, a high Ag fraction in Ag 7 Au 3 and Ag 9 Au 1 seems to lead to diffusion of Ag atoms into the center as evidenced by the increase in the corresponding areas (Table 2, entries 5-6). The k 2 -weighted χ(k) EXAFS spectra for the Ag x Au 10-x catalysts are shown in Figure 3c in comparison to Au foil and corresponding Fourier transformed (FT) spectra are given in Figure 3d (fits given in Figures S12-S13). EXAFS oscillations in the k-region from 5-8 Å À 1 change with increasing Ag content, which correspond to the increase of Ag neighbors around the absorbing Au atom with different backscattering phase shifts at higher k. [47] In the FT spectra, the metal-metal scattering doublet peaks shift towards lower R with increasing Ag content, which also points towards replacement of Au atoms by Ag atoms i. e. an increase in AuÀ Ag coordination. The first shell contribution which appears as a double peak can be attributed to the interference between AuÀ Au and AuÀ Ag backscattering with different phase shift and amplitude. EXAFS analysis was performed to derive the degree of alloying by comparing the fitted coordination number (CN) with the composition (Table 3). For Ag 1 Au 9 , the CN AuÀ Au of 12.1 (AuÀ Au 2.85 Å) is close to its value in Au foil and no AuÀ Ag scattering was observed. On the other hand, a lower CN AuÀ Au of 10.5 was found for Ag 3 Au 7 . Here AuÀ Ag scattering occurred with a CN AuÀ Ag of 2.2 at 2.86 Å. For a homogeneously mixed alloy, the ratio of CN AuÀ Au to CN AuÀ Ag should be close to the molar ratio of Au:Ag. [48] Since XAS is a volume-sensitive method, spatially resolved information e. g. on the composition of the surface can only be derived indirectly from comparison of the ratio of AuÀ Au and AuÀ Ag CNs as well as their deviation from the expected values for homogeneous composition. The ratio of CNs is shown in the last column of Table 3. Evidently, the ratio of CNs is higher in gold while ICP verified that the overall composition agrees well with the nominal composition (Table 1). Hence, a small composition gradient with Ag slightly enriched in the surface-near volume is expected for some compositions. [49] For Ag 7 Au 3 , the ratio of CNs is closest to the molar ratio (Table 3, Table 2. Area under the marked region in Figure 3b.

Entry
Sample ΔA3/eV cm À 1  Table 3. EXAFS fitting results obtained from Au L 3 -edge absorption spectra of Ag x Au 10À x /ZrO 2 ; EXAFS spectra were fitted in the range of R = 1.0-3.4 Å and k = 2.4-12.7 Å À 1 using k w = 2. [a] based on CNs from EXAFS fitting. [b] theoretical values are given in brackets.
[c] fixed during the fitting.

FULL PAPER
asc.wiley-vch.de entry 5). Hence, a homogeneous alloy is more probable in this case. For Ag 9 Au 1 , no AuÀ Au scattering was observed, but a AuÀ Ag CN of 10.1 was found. Since the total CNs for Ag 7 Au 3 and Ag 9 Au 1 are less than its maximum value of 12, this indicates the presence of an increased Au atom fraction on the surface. In summary, apart from the successful alloy formation, slight compositional gradients of the alloy nanoparticles are indicated by the EXAFS and XANES analysis.
To further investigate the oxidation state and the surface-near alloy composition, a complementary XPS study has been conducted with the as-prepared Ag x Au x-10 /ZrO 2 catalysts and the results are given in Figure 4. The molar fraction of Au in the outer surface layers of Ag x Au xÀ 10 /ZrO 2 catalysts gained from XPS is given in Figure 4a, where the as-prepared, gold-rich Ag 1 Au 9 /ZrO 2 and Ag 3 Au 7 /ZrO 2 catalysts in Figure 4a show an enrichment of the surface with Ag. This is in line with EXAFS analysis. On the other hand, the surface composition is closest to the nominal composition for Ag 5 Au 5 . At higher Ag content, more Au appears to be present on the surface of the catalysts before the reaction. Note upfront that after the catalytic reaction, the compositional gradient disappeared for low gold contents (shift toward nominal composition), while it was still present after reaction at gold molar fractions above 50%. From XPS peak deconvolution summarized in Figures S6-S9, the surface oxidation of the Au 4f and Ag 3d peaks has been extracted and summarized in Figure 4b. Therein, a correlation of the amount of oxidized atomic species of gold and silver is evident, respectively. In this context, XRD patterns of the bulk catalyst showed that the (111) reflections of Ag and Au at 38.2°decreased with higher Ag content ( Figure S10). Taking the large extent of oxidized Ag surface atoms found by XPS into account, the decrease of the (111) reflection with increasing Ag content appears to indicate the presence of small Au crystallites and amorphous oxidized Ag phases on the surface. On the other hand, for gold-rich compositions a higher crystallinity is evident in XRD, which is in line with the slight silver enrichment in the surface near volume observed in XPS for Ag 3 Au 7 and Ag 1 Au 9 (compare Figure 4a). Additionally, for gold-rich compositions, less oxidized gold atoms are present in the outer nanoparticle region while the percentage of oxidized silver atoms rises to a limited extent. In turn, a high silver content leads to a higher content of gold atoms present in oxidized form, such as silver renders gold to become less noble as previously discussed in the literature. [50] Consequently, an electronic interaction of Au and Ag atoms is evident and in line with the results from XANES. In summary, by combining different X-ray based, optical and element-specific techniques, a successful alloy nanoparticle preparation of the desired composition has been proven. Slight element-enrichment at the nanoparticle surface has been observed with about 10% more Ag present at the surface in case of high Au content or about 10% enrichment of Au at the surface, in case of high Ag content. Compared to wet-chemical nanoparticles, which always show strong enrichment of Ag in the surface-near volume due to preferential reduction of more noble Au-based precursors, the laser-generated catalysts show a gradual elemental composition of the surface-near volume. In addition, they hit the right size regime reported to be active for Au-based catalysts for HMF to FDCA oxidation. [51] Consequently, these lasergenerated catalysts are ideal candidates for studying the influence of the Ag content on the catalytic activity and selectivity, presented in the following section.  Figures S6-S9), shown for the asprepared catalyst before reaction as well as the same catalysts after reaction.

Selective Oxidation of HMF by the Supported Ag x Au 10À x Particles
After thorough characterization, the catalysts were tested in the selective oxidation of HMF. First, the oxidation reactions were performed at different temperatures to investigate its influence on the activity and selectivity. The reactions were performed in the presence of 4 equivalents of NaOH, 10 bar air pressure and a constant HMF-to-metal molar ratio (HMF:M; M = sum of Ag and Au content) of 100:1 ( Figure 5).
All catalysts were found to be active in the selective oxidation of HMF even at room temperature, where HFCA was produced as the main product. The highest HFCA yield of 95% under these conditions was achieved over Ag 1 Au 9 /ZrO 2 . Increasing the reaction temperature caused the HFCA yield to pass through a maximum for Ag 3 Au 7 , Ag 5 Au 5 , and Ag 7 Au 3 at 50°C, 75°C, and 75°C, respectively, where HFCA was quantitatively produced over Ag 5 Au 5 /ZrO 2 . Increasing the reaction temperature led to decreasing HFCA yields for all catalysts. Yet, FDCA production started at 100°C over Ag 1 Au 9 , Ag 3 Au 7 , and Ag 5 Au 5 , reaching up to 31%, 74%, and 29% yield at 125°C, respectively. Carbon balances of less than 100% in some reactions can be attributed to the formation of unquantified socalled humins by multiple side-reactions of HMF in alkaline solution. [52] The formation of HFCA as the primary oxidation product at room temperature confirms that the oxidation of the aldehyde moiety of HMF proceeds readily and that the oxidation of the hydroxymethyl group is the rate-limiting step in alkaline conditions. [53] The fact that the reaction temperature barely affected the activity and HFCA yield of the Ag 9 Au 1 catalyst up to 100°C can be attributed to the high Ag content ( Table 1, entry 6), in line with literature on pure silverbased catalysts. [19] The oxidation of the hydroxymethyl group to finally form FDCA proceeded both at higher temperatures and increasing Au content, probably because Ag is inactive in this reaction. [19a] Surprisingly, Ag 1 Au 9 with the highest Au content gave a lower FDCA yield of 31% at 125°C compared to Ag 3 Au 7 (74%), while Ag 5 Au 5 gave a similar yield of 29%. The optimum of FDCA production over Ag 3 Au 7 , may be attributed to two effects: particle size or alloy composition. In general, the particle size is an important factor in pure Au catalysis, although Au catalysts in the given size range ( � 7-11 nm, Table 1) were all found active in FDCA production. [51,54] Note, that for CO oxidation, a reduced effect of particle size was found for alloy formation with Ag. [55] Thus, the particle composition seems to have a stronger effect on FDCA production, with an optimal Ag/Au composition for both aldehyde and alcohol oxidation.Generally, the catalytic tests uncovered a pronounced effect of the catalyst composition on the product distribution at different temperatures.
Next, the influence of added NaOH on the product distributions was studied ( Figure 6). Temperatures that gave the maximum production of HFCA were chosen for all catalysts except Ag 3 Au 7 , which was highly active in FDCA production at 125°C. The respective temperatures are indicated for each composition in Figure 6. Lowering the amount of added base is relevant both ecologically and economically, as increasing the pH of a HMF solution prior to acidic separation of FDCA leads to the production of large quantities of salt. [12] As can be seen in Figure 6 (discussed from right to left), decreasing the amount of added NaOH by half had no effect on the production of HFCA over Ag 1 Au 9 , Ag 5 Au 5 , Ag 7 Au 3 , and Ag 9 Au 1 , all of which gave a constant HFCA yield. Further decreasing the added NaOH to one equivalent regarding HMF led to a slightly lower yield of HFCA; however, it had no effect for silver-rich alloys such as Ag 7 Au 3 and Ag 9 Au 1 , which both produce HFCA with a relatively constant yield of 80-85%. Hence, these catalysts are active also under conditions which are less alkaline and more environmentally favorable.
In the absence of NaOH, negligible or low HMF conversion and HFCA formation was observed. The lowest HMF conversion of 3% was observed for Ag 7 Au 3 , where no oxidation products were detected. On the other hand, for gold-rich catalysts � 20-25% of HMF was converted, while the yield of HFCA and FDCA was below 5% (selectivity below 15%). A maximum HFCA yield of 9% was gained from Ag 5 Au 5 and Ag 9 Au 1 catalysts at 13% and 15% HMF conversion, representing high selectivities of 68% and 62%, respectively.
As for Ag 3 Au 7 , for which a temperature of 125°C was chosen for FDCA production, reduction of the added NaOH to two equivalents even resulted in an increase of FDCA yield to 95%. At one equivalent of added NaOH, HFCA was produced as the main product in 74% yield (2% FDCA yield). Under base-free conditions, HMF conversion dropped to 24% while no oxidation products were formed.
These results show that for the present catalysts one equivalent of hydroxide is required to achieve high HMF conversion and HFCA yield. In addition, FDCA production was observed over Ag 3 Au 7 only when two or more equivalents of hydroxide were present. Hence, the reaction mechanism seems to involve a nucleophilic attack of a hydroxide ion on the aldehyde moiety of HMF with subsequent dehydrogenation of the formed geminal diol. [15,56] Another role of the homogeneous base may be to prevent catalyst deactivation by precipitation of poorly soluble FDCA onto the catalyst.
To compare the performances of all catalysts with literature (see Table S2), the productivity, i. e. moles of product formed normalized to the noble metal content and reaction time, was calculated. The highest productivity for HFCA of 19 mol HFCA h À 1 mol metal À 1 was obtained over Ag 5 Au 5 in the presence of two equivalents of NaOH. A similar FDCA productivity of 19 mol FDCA h À 1 mol metal À 1 was observed over Ag 3 Au 7 in the presence of two equivalents of NaOH.
In general, the variation in the amount of the added base revealed that the catalyst are active even in the presence of less base. Also, some activity was observed in the base-free oxidation of HMF over some catalysts; however, longer reaction times may aid to achieve higher yields of HFCA. Alternatively, solid bases could be used as support materials in future studies as reported recently. [57] In a next set of reactions, the influence of oxygen partial pressure was studied. An air pressure of 10 bar was chosen in the previous experiments. Oxygen participates indirectly in the reaction mechanism by removing hydrogen adsorbed on the catalyst surface. [15,56] The hydrogen is formed during the dehydrogenation of the intermediately formed geminal diol from nucleophilic addition of hydroxide ions to the aldehyde group. [15] Thus, despite gas phase oxygen not being directly incorporated into the oxidation product, its presence is crucial. Using air enhances the sustainability of the reaction, [12] but the low concentration of oxygen makes its use more challenging compared to pure oxygen. To further elucidate the role of oxygen, reactions were performed at ambient pressure under previously optimized conditions (Table 4).
In general, the lower air pressure resulted in lower product yields over all catalysts except of Ag 9 Au 1 , which gave a constant HFCA yield of 80% (Table 4, entry 5). The HFCA yield over Ag 1 Au 9 ( These results underline differences in activity and selectivity depending on the composition of AgÀ Au catalysts and demonstrate the crucial role of oxygen in the selective oxidation of HMF. Depending on the Ag/Au ratio, the effect of oxygen partial pressure differed. Previous studies on HFCA synthesis over pure Ag catalysts revealed a minor influence of air pressure, [19a] which also seems to be the case for bimetallic AgÀ Au catalysts rich in Ag like Ag 9 Au 1 . That might be attributed to favored adsorption of oxygen on Ag as compared to Au with the reaction proceeding on neighboring Au active sites. [21a] This is further confirmed by comparison of the Ag-rich Ag 9 Au 1 to Au-rich Ag 1 Au 9 , both of which were tested at room temperature. The more pronounced decrease in HFCA yield over silver-depleted Ag 1 Au 9 confirms that Au-based or in this case Au-rich catalysts are more influenced by air pressure due to limited oxygen adsorption. Hence, catalysts rich in gold generally require higher air pressure to achieve high product yields.
Neither Au nor Ag were found in the reaction solutions of all catalysts by ICP-OES, showing that the catalysts are stable against metal leaching in solution.
The productivity of Ag 3 Au 7 towards FDCA of 19 mol FDCA h À 1 mol metal À 1 is in the range or higher as compared to other alloys like AuPd [20a,58] or AuCu [59] in literature (Table S2), which underlines the potential of AgÀ Au alloys as highly active catalysts for HMF oxidation. However, note that comparability should be taken with caution due to different reaction conditions like the addition of a base.
In a final step after optimizing reaction temperature, added NaOH and air pressure, the temporal progression of the reactions was studied (Figure 7). For Ag 1 Au 9 and Ag 9 Au 1 , which were tested at room temperature, a classical batch process conversion profile starting from no HMF conversion at t = 0 was observed. HMF was very rapidly and selectively converted to HFCA over Ag 1 Au 9 , which gave HFCA in 64% yield (76% selectivity) after 30 min. Quantitative HMF conversion was observed after 1 h giving 83% HFCA yield at 83 mol HFCA h À 1 mol metal À 1 productiv-ity. Prolonged reaction time led to an increase in HFCA yield up to 95% after 5 h, despite the quantitative conversion of HMF after just 1 h, which might be attributed to adsorbed intermediates or the product itself on the catalyst surface. On the other Ag 9 Au 1 [b,c] 100 80 0 100 83 0 Individual reaction conditions: [a] two equivalents of NaOH. [b] one equivalent of NaOH.
[e] 75°C. hand, HMF conversion and HFCA yield steadily increased with time over Ag 9 Au 1 , which can be attributed to the milder reaction conditions. However, longer reaction times are required to achieve high HMF conversion and HFCA yield under ambient conditions. In case of the intermediate compositions Ag 3 Au 7 , Ag 5 Au 5 , and Ag 7 Au 3 , conversions of HMF in Figure 7 were found to be high already at t = 0. This observation is linked to the definition of the starting time t = 0, which was set to reaching the desired reaction temperature for the first time. For example, reaching 125°C for Ag 3 Au 7 took about 15 min. Consequently, for Ag 3 Au 7 , high product yields of 85% HFCA and 10% FDCA were already achieved at the nominal start of the reaction. This shows Ag 3 Au 7 is extremely active in aldehyde oxidation, which is already quantitatively converted during heating up. With longer reaction times, HFCA yield decreased linearly with a linear increase in FDCA yield, pointing out alcohol oxidation as the rate-limiting step. [53] Therefore, a reaction time of 5 h is required to produce FDCA in 95% yield. The fact that no 5-formyl-2-furancarboxylic acid (FFCA) is found, which is the intermediate of HFCA oxidation to FDCA, [15] underlines the high activity in aldehyde oxidation. Note that for all catalysts not tested under ambient conditions, the aldehyde oxidation activity is even higher, since high yields are obtained despite harsher reaction conditions in terms of HMF stability.
The influence of the HMF:M ratio was studied by decreasing the amount of catalyst added to the reaction solution, (Figure S19) resulting in increased productivity rates up to 254 mol HFCA h À 1 mol metal À 1 for Ag 5 Au 5 at a HMF:M ratio of 300. Decreasing HMF conversion or product yields upon the addition of less catalyst showed that the reactions performed so far were not limited by solid-liquid mass transfer. If reactions are scaled up at a later stage, however, mass transfer limitations have to be considered. To study possible mass transfer effects, the oxidation of HMF over Ag 9 Au 1 /ZrO 2 was scaled up to 30 mL by a factor of three. As in the previous reactions, the catalyst was used as a fine powder to minimize possible internal mass transport limitations. With increasing reaction volume, good mixing becomes increasingly important to maintain a slurry, which minimizes solid-liquid transfer limitations. To study the effect of mixing at the increased volume of 30 mL, the stirring speed was varied by almost an order of magnitude, (150-1400 rpm, Figure S20) and the HFCA yield was increased accordingly (5-30%), revealing that the reaction was limited to some extent by solid-liquid mass transfer in the bigger reaction volume. Next, the reaction was performed in a special reactor equipped with a mechanical gas entrainment impeller for better mixing of the liquid phase and enhancement of the gas-liquid interface, thus better gas-liquid mass trans-fer. It was found that a high HFCA yield of 82% could be reached within 2 h (41 mol HFCA h À 1 mol metal À 1 ) with optimal mixing of the different phases of the reaction. The profiles are similar but with a higher productivity, evidencing that with careful considerations, a future scale-up is indeed possible.

Catalyst Stability and Reuse
Catalysts were separated from the reaction mixture and dried overnight before testing in a subsequent reaction under identical conditions for reusability studies (Figure 8 a-e). A slight loss of catalyst mass during this recycling process was taken into account by adjusting the amount of reactant and total volume to the recovered catalyst mass. Ag 1 Au 9 , Ag 5 Au 5, and Ag 7 Au 3 proved very stable upon reuse, and the product yield and productivity rate remained constant for five reuse cycles. Particle sizes did not severely change for these catalysts (Figure 8f) and as HFCA synthesis is not highly affected by particle size, this explains their constant activity upon reuse. Interestingly, the surface composition of Ag 5 Au 5 and Ag 7 Au 3 changed slightly (Figure 4a) but maintained their activity. For the silverrich Ag 9 Au 1 , a slight decrease in activity was observed after the third catalytic cycle, but the particle size was unaffected by recycling (Figure 8f, Figure S15). Thus, the decreasing activity of Ag 9 Au 1 might be related to the deposition of impurities from HMF degradation, [53] because the carbon balance was less than 100% and no drastic changes in the surface composition were found. Since monometallic Ag-based catalysts deactivate more rapidly under comparable reaction conditions, [19] alloy formation with less than 90% of Ag greatly enhances the catalyst stability.
Ag 3 Au 7 , which was tested at an elevated reaction temperature of 125°C for FDCA synthesis, decreased in activity after the first catalytic reaction, giving HFCA as the main product with a yield of 38% in the second reaction. For this catalyst, the particle size increased from 6.9 to 10.2 nm after the first catalytic reaction (Figure 8f, Figure S15) probably due to the higher reaction temperature applied for this catalyst. On the other hand, EDX scans and the surface composition derived by XPS did not show severe changes ( Figures S6-S9, Figure S17). Although the particle size effect may be limited for AgÀ Au alloys, this catalyst was the only one active in FDCA production, thus probably having active Au sites. Together with the high Au content and smaller particles favoring FDCA synthesis, [51] sintering seems to be the primary reason for the deactivation of this catalyst.
In general, the reuse stability of catalysts active in HFCA synthesis was greatly enhanced by alloy formation of Ag and Au and the catalysts were also structurally stable.

Mechanistic Outlook
In summary, differences in initial activity of the prepared catalysts were observed and can be mainly attributed to the particle composition due to similar particle sizes. The catalytic activities at room temperature and at 125°C, i. e. HFCA and FDCA synthesis conditions, respectively, are summarized in Figure 9. An optimal composition for HMF oxidation activity can be concluded with Ag 3 Au 7 being active both in aldehyde and alcohol oxidation, giving FDCA as the main product. The alcohol oxidation activity is lower for Ag 1 Au 9 and Ag 5 Au 5 and no activity in FDCA synthesis was observed at higher Ag content (Figure 9). Since the activation energy for aldehyde oxidation is lower, [60] this step proceeds rapidly over all catalysts. The rate-limiting alcohol oxidation, on the other hand, has a higher energy barrier and only occurs on Au catalysts, as Ag is inactive in FDCA synthesis. [19a] Besides activation barriers, oxygen activation is a crucial factor, which is limited on Au [61] or limited to active Au sites possibly forming a perimeter with the support similar to literature on selective alcohol oxidation with Au/TiO 2 . [62] Since hydroxide ions form a geminal diol in solution by nucleophilic attack on the aldehyde moiety of HMF, which is then dehydrogenated, oxygen is not incorporated in the oxidized molecule and thus has an indirect role in the mechanism by removal of electrons and adsorbed Hspecies. [60,63] During oxygen activation, hydrogen peroxide is formed, whose decomposition has a high energy barrier on Au but proceeds readily on Ag or AgÀ Au alloys. [64] Consequently, two different kinds of active centers for HMF oxidation are present in AgÀ Au catalysts (Figure 9c). The catalysts differed in bulk and surface composition (Table 1, Figure 4a). This alters the oxygen activation and respective selectivity to either only oxidize the aldehyde moiety or both functional groups of HMF with Ag active in oxygen activation and Au active in HMF conversion. [21a] While catalysts with higher amounts of either one of both metals favor alcohol oxidation (Figure 9a-c) an ideal catalyst for FDCA synthesis should have an optimal Ag:Au-ratio, which may result in a favorable amount of Ag atoms next to Au atoms. Based on indepth characterization and extensive catalytic testing, Ag 3 Au 7 was identified to have the most favorable Ag:Au ratio of 38:62 with active centers for alcohol oxidation i. e. FDCA synthesis.
All catalysts other than Ag 3 Au 7 gave HFCA as the main product with enhanced stability. While pure Au and Ag catalysts deactivate by particle growth, [19a] sintering was significantly reduced for the bimetallic catalysts except for Ag 3 Au 7 .

Conclusion
The influence of alloy formation of Ag and Au on the catalytic activity in HMF oxidation was investigated. For this, we took a rational approach starting from the synthesis of bimetallic nanoparticles with varying Ag/Au ratio followed by extensive catalytic studies. LAL allowed the production of highly pure bimetallic nanoparticles with comparable moderate particle sizes without any surfactant or organic ligand. Thus, catalytic oxidation activities can be linked directly to the catalyst composition. Particle sizes were maintained upon adsorption of colloids on ZrO 2 , as shown by TEM. The supported catalysts still contained alloyed nanoparticles with slight Ag-enrichment in the surface-near volume for high Au content and about 10% enrichment of Au in the surface-near volume at high Ag content as shown by a combination of XAS and XPS. Catalytic tests in HMF oxidation revealed differences in selectivity towards HFCA or FDCA depending on the composition. Optimization of the reaction conditions for each catalyst allowed the production of both HFCA and FDCA in yields above 90% and high productivity rates. At high Ag content, catalysts were active in HFCA synthesis, whereas the high Au content in Ag 3 Au 7 and Ag 1 Au 9 featured the FDCA synthesis, with an optimal activity for Ag 3 Au 7 . Thus, a synergistic effect between Ag and Au was shown in varying selectivity and enhanced catalyst stability against sintering for most catalysts. The observed optimal compositions for either HFCA (Ag 9 Au 1 ) or FDCA (Ag 3 Au 7 ) formation might be linked to an ideal ratio of surface Ag and Au atoms, and respective oxygen activation to either promote selective aldehyde oxidation or both aldehyde and alcohol oxidation. For the future, it might be rewarding to substitute the homogeneous base by a solid base as catalyst support. All catalysts were stable against metal leaching under the reaction conditions applied in this study, and recycling experiments revealed great catalyst stability of Ag 1 Au 9 , Ag 5 Au 5 and Ag 7 Au 3 , which also preserved their structure. Hence, the highly active and stable bimetallic AgÀ Au catalysts may be suitable for future tests on the scale-up or the continuous oxidation of HMF.

Catalyst preparation
In a first step, the alloy nanoparticles (AgAu NPs) were synthesized by laser ablation of an alloy target in flowing micromolar aqueous saline, followed by electrostatic adsorption Figure 9. Schematic representation of the influence of the catalyst composition and structure on the activity in (a) aldehyde and (b) alcohol oxidation of HMF over Ag x Au 10À x / ZrO 2 catalysts as well as (c) active centers. The respective product yields are taken from the temperature screening at 25°C (active conditions for aldehyde oxidation) and 125°C (active conditions for alcohol oxidation). Lines are added to guide the eye. Reaction conditions are shown in Figure 5.
of the AgAu NPs in a dispersion of the ZrO 2 support material by adjusting the pH to 3.8 between the isoelectric points (IEP) of both reactants (Figure 2a). [42,65] In brief, the experimental laser ablation setup to synthesize the AgAu NPs consists of a flow chamber containing the alloy target, a piston pump (Ismatec ISM321 C) and a magnetically stirred beaker that was filled with an aqueous (Milli-Q water, resistivity 18.2 MΩ cm) 200 μM NaCl solution.
[34a] The respective alloy target foil (dimensions: 50 mm*20 mm*0.5 mm) was fixed inside the ablation chamber. The liquid layer in front of the target was set by the chamber window and fixed to 4 mm. [34a] The piston pump was set to a flow rate of 50 mL min À 1 and the liquid was circulated through PTFE tubes. For the ablation, a nanosecond laser from Edgewave (model: IS160-1-T) operating with a pulse duration of 8 ns, a repetition rate of 5 kHz, a pulse energy of 8 mJ, and a wavelength of 532 nm was focused onto the target. The second harmonic laser wavelength was chosen to additionally reduce particle size by laser fragmentation of already produced colloids when passing through the ablation zone during cycling. [34a] During ablation, the laser pulses were directed in a rectangular pattern using a scanner (Raylase SS-IIE-12) at a scan rate of 2000 mm s -1 to minimize cavitation bubble shielding. The presence of the 200 μM NaCl solution allowed an in situ size quenching and to prevent agglomeration as well as particle growth. [66] The ablation was conducted between 30 min (Ag 1 Au 9 ) up to 120 min (Ag 9 Au 1 ) depending on the target composition, as the ablation rate scales with the target density. [67] The mass concentration of colloids was determined gravimetrically by differential weighting of the target with a microbalance (ABT 120-5DM from KERN) before and after LAL, as well as UV-Vis extinction spectroscopy of the colloids (Figure 2b).
For supporting the nanoparticles, 5 g of ZrO 2 was suspended in 200 ml of deionized water (Milli-Q water, resistivity 18.2 MΩ cm) in individual, continuously stirred batches. 0.1 g of AgAu NPs were added as colloids in respective volumes based on the measured concentrations after ablation to reach a mass loading of 2 wt.%. To improve the electrostatic adsorption of the AgAu NP to the ZrO 2 microparticles, the pH was adjusted to pH 3.8. [42] After that, the suspension was allowed to sediment. The deposition efficiency was derived from the supernatant phase via UV-Vis extinction spectroscopy. The sedimented catalyst was separated by decantation and freeze-drying (Christ Alpha 1-4 LSCplus).

Catalyst Characterization
The catalysts were thoroughly characterized using multiple methods. Analytical disc centrifugation (ADC) was performed to determine AgAu NP sizes of the freshly prepared particles. UV-Vis and diffuse reflectance (DR) UV-Vis spectra were recorded to study alloy formation of the freshly prepared and the supported AgAu NPs, respectively. The particle size of supported AgAu NPs was determined by transmission electron microscopy (TEM). Inductively coupled plasma optical emission spectroscopy (ICP-OES) was used to verify metal loadings and molar compositions of the supported catalysts and N 2 physisorption using the Brunauer-Emmet-Teller method was employed to determine the specific surface area of the solid catalysts. Powder X-ray diffraction (XRD) patterns were measured to study crystalline phases of supported catalysts. The surface composition and oxidation states of both metals were investigated using X-ray photoelectron spectroscopy (XPS). The degree of alloy formation and the local structure in the bulk AgAu NPs was investigated by X-ray absorption spectroscopy (XAS), which was recorded at the CAT-ACT beamline of the KIT synchrotron [68] and Athena and Artemis software from the IFEFFIT/Demeter package (version 0.9.25) [69] were used for data analysis.
For further details and experimental procedures, consult the Supporting Information of this article.

Catalytic Tests
Selective oxidation of HMF was carried out in Teflon ® inlets (52 mL), which were placed in home-built magnetically stirred stainless steel autoclaves equipped with pressure inlet and outlet, manometer, thermocouple, and a relief valve. The reactors were charged with appropriate amounts of 0.2 M HMF solution, 2.5 M NaOH and distilled water to give a total volume of 10 mL in each reaction. The catalyst powders were added at a constant HMF-to-metal molar ratio (HMF:M; M = sum of metal content Ag and Au) based on the metal loading of each catalyst, and the reactors were sealed and purged three times with synthetic air before the desired pressure was adjusted. The reaction temperature was set in a controller and the autoclaves were heated with heating sleeves based on the thermocouple inside the reaction solution. The temperature was kept constant with a maximum variation of 1°C and the starting point of each reaction was set after the solution first reached the set reaction temperature. Additional reactions were performed in a batch reactor with the same specifications that was equipped with a mechanical gas entrainment impeller. After the reactions, the reactors were quenched in an ice bath, depressurized, and the catalysts were separated by decantation.
Samples for analysis by high-performance liquid chromatography (HPLC) were taken before and after the catalytic tests, filtered with 0.45 μm Teflon filters and diluted according to the calibration range. HPLC was measured on a Hitachi Primaide at 50°C and 25°C (Bio-Rad Aminex HPX-87H column, solvent 5 mM H 2 SO 4 ). The compounds were detected by a refractive index (Hitachi Chromaster 5450) and a diode array (Hitachi 1430) detector. Reference solutions of HMF, HFCA, FDCA, DFF and FFCA were used for calibration as external standards and concentrations of reaction solutions were derived from the peak areas. HMF conversion (eq 1), product yield (eq 2) and selectivity (eq 3), as well as the carbon balance (eq 4), were calculated based on the concentrations determined by HPLC.  (4) Error bars were derived from Student's t-distribution with 95% confidence interval (two tailed t-test), a degree of freedom of 4 (5 recycling cycles) and the standard deviation of the respective recycling experiment. The standard deviation was calculated from samples shown in Figure 8c) as the sample showed good cycle stability. Consequently, as slight degradation might add to the error evaluation, error bars can be considered a worst-case estimation.
FDCA could be extracted by acidification of the reaction solutions with high FDCA yield to a pH value of 1 with HCl. For HFCA, the acidified solution was extracted with ethyl acetate and the solvent was removed under reduced pressure. The purity of both products was confirmed by 1 H and 13 C NMR spectroscopy, which was measured at room temperature in DMSO-d 6 on a Bruker Avance 250 and a Bruker Avance 400 spectrometer ( Figure S21).