Applied Catalysis A, General Hydrogenation of levulinic acid to γ -valerolactone over anatase-supported Ru catalysts: E ﬀ ect of catalyst synthesis protocols on activity

γ -Valerolactone (GVL) is a value-added renewable chemical with great potential and can be obtained from biomass by the hydrogenation of levulinic acid (LA) using metal-based catalysts, such as Ru/TiO 2 . We here report an in depth study of the e ﬀ ect of catalyst synthesis parameters on the performance of Ru/TiO 2 (anatase), varying the nature of the Ru-precursor and the conditions of the calcination and/or reduction step. Catalyst performance was evaluated under batch conditions at a hydrogen pressure of 45 bar and using either water (90 °C) or dioxane (150 °C) as solvent. The experiments showed that catalyst activity depends greatly on the Ru precursor used (RuCl 3 , RuNO(NO 3 ) 3 , Ru(NH 3 ) 6 Cl 3 ). Best results when considering the turn-over frequencies (TOF) of the catalysts were obtained using the RuNO(NO 3 ) 3 precursor, whereas RuCl 3 performed better when considering the initial rate based on Ru intake. An intermediate calcination step and the use of a hydrogen-rich sweep gas during the ﬁ nal reduction step were shown to have a negative impact on catalyst activity. Characterization of the fresh catalysts by BET and TEM provided valuable insight in the relation between the catalyst structure and its activity.


Introduction
Levulinic acid (LA) is a major product of the controlled dehydration of the C6-sugars (D-glucose, D-mannose and D-galactose) in lignocellulosic biomass. It is recognized as an important carbohydratederived renewable platform molecule and has attracted considerable interest from a number of chemical companies [1,2]. The family of LA derivatives is large and some have much potential for commercialization. For instance, LA can be converted to 2-methyltetrahydrofuran (MTHF) and various levulinate esters, which may be used as gasoline and biodiesel additives, respectively [3][4][5][6][7][8]. δ-Aminolevulinic acid is a well-known herbicide and β-acetylacrylic acid has been proposed as a (co-)monomer for novel acrylate polymers [1]. Arguably, most attention has been devoted to the conversion of LA to γ-valerolactone (GVL), as GVL is considered as an important platform chemical in its own right and can be used as food additive, solvent and as precursor for fuel additives and bulk polymers [9][10][11][12][13]. The conventional way to produce GVL involves the hydrogenation of LA or its esters [14][15][16], with molecular hydrogen or an alternative hydrogen donor (e.g. formic acid), preferably using heterogeneous catalysts (Scheme 1). The intermediate 4-hydroxypentanoic acid (4-HPA) may be found in relatively high amounts when using water as the solvent, its exact amount depending on the relative rate of hydrogenation versus the rate of the intramolecular esterification reaction.
Recently, Liguori and Barbaro [17] published a comprehensive review on the direct catalytic conversion of renewable sources to GVL with an emphasis on the heterogeneous catalysts that have been used for this reaction. Noble metal-based catalysts are most commonly and successfully employed, with Ru-based ones in particular showing high activity and selectivity to GVL [14,[18][19][20][21]. In addition to the nature of the active metal phase, the choice of support also has a large effect on catalyst performance, in particular on catalyst stability. Activated carbons are the most widely used support for Ru in LA hydrogenation [22][23][24], mainly due to their good performance and availability [24][25][26], and as such Ru/C catalysts can be regarded as one of the benchmark catalysts for this reaction. Under continuous flow conditions and using water as the solvent, however, slow though irreversible deactivation of the Ru/C catalyst was observed, presumably due to Ru sintering and a reduction in specific surface area as a result of the deposition of carbonaceous deposits [15,22,25,27].
Various metal oxides, including SiO 2 , Al 2 O 3 , Nb 2 O 5 , ZrO 2 and TiO 2 , have been tested as a support for LA hydrogenation with Ru as the active metal [16,19,24,25]. A major advantage of such metal oxides over carbon supports is their mechanical and thermal stability, which allows for repetitive regeneration and coke removal at elevated temperature. However, for some of these oxides, e.g. Al 2 O 3 and SiO 2 , the stability under hydrothermal, acidic conditions is known to be limited, which is of relevance for LA hydrogenations carried out in water or at high LA loadings [25,28]. Estimations of oxide support stability in pressurized water at 200°C were recently provided by Lange using Pourbaix diagrams [2]. SiO 2 has the tendency to be converted into silica gel at all pH values, whereas Al 2 O 3 is well-known to convert into boehmite (AlO(OH)) at pH values between 4.5 and 11.5. TiO 2 , calculated to be stable over the entire pH range, and ZrO 2 are attractive alternatives [28], even though hydrated Zr(OH) 4 was found to be the thermodynamically most stable phase in water. Indeed, a screening study on LA hydrogenation involving 50 catalysts in a flow reactor system revealed that Pt on TiO 2 (P25 from Degussa) and ZrO 2 performed best, with constant GVL yields for runtimes exceeding 100 h (> 95 mol% at 200°C and 40 bar H 2, 11 wt.% LA in GVL) [28]. LA hydrogenation studies in water as solvent comparing Ru/TiO 2 (P25) and Ru/ZrO 2 catalysts (both 1 wt.% Ru) in a batch set-up at 70°C revealed that quantitative LA conversion could be achieved within 4 h with the Ru/TiO 2 catalysts, while Ru/ZrO 2 was less active and 92% LA conversion was observed after 6 h [29]. This enhanced hydrogenation activity of Ru/TiO 2 was attributed to a better Ru dispersion on TiO 2 .
Recently, some of us reported on a catalyst screening study for LA hydrogenation in water (90°C, 45 bar H 2 ) using a wide range of supported Ru-catalysts (1 wt.% Ru on C, CNT, Al 2 O 3 , TiO 2 , ZrO 2 , Nb 2 O 5 and H-Beta-12.5) and again found that TiO 2 (anatase form, A100) performed better than for Ru/ZrO 2 in this solvent [30]. The results listed above thus show the potential of TiO 2 as support for this reaction and an overview of titania-supported Ru catalysts for LA hydrogenation is given in Table 1.
Some of these studies have shown that the phase composition of the TiO 2 support (anatase, rutile and combinations thereof) influences catalyst performance. For example, Al-Shaal et al. investigated the hydrogenation of LA using Ru supported on rutile and P25 TiO 2 (a 75:25 mixture of anatase-rutile) supports in ethanol and ethanol-water mixtures [25]. The rutile-supported catalyst gave no LA conversion in neither ethanol nor ethanol-water, whereas Ru/TiO 2 (P25) showed much better performance. Furthermore, a comparison of the catalytic transfer hydrogenation of levulinate esters using Ru(OH) x /TiO 2 [31], with anatase, rutile and anatase-rutile titanias, showed the anataserutile-based catalyst to perform best (86% for TiO 2 (A); > 99% for TiO 2 (R) and TiO 2 (A75:R25)). Recently, Ruppert et al. reported a detailed study on the influence of various TiO 2 supports (anatase, rutile and mixtures thereof) on the Ru-catalyzed LA hydrogenation [32]. A Scheme 1. Proposed reaction scheme for the catalytic hydrogenation of LA to GVL. Ru/TiO 2 (AR) catalyst (10-20% rutile) again proved to be more active than fully anatase-based ones. The lower activity of the latter was attributed to the presence of larger Ru particles and agglomerates on the surface and relatively small Ru nanoclusters predominantly found in micropores, thought to be inactive in the hydrogenation reaction. Activity of the anatase-based Ru catalyst could be improved by implementation of either a calcination step of the anatase support before impregnation to reduce the micropore volume or by avoiding a high temperature reduction step during catalyst synthesis by using a mild chemical reduction step [32]. Besides variations in support phase composition, many different synthetic procedures for titania-supported Ru catalyst preparation have been reported varying in the choice of, e.g., metal precursor, impregnation method and activation procedure (Table 1), which all may affect catalyst performance. Wet impregnation is most commonly used method for the synthesis of these Ru/TiO 2 catalysts [19,20,25,29], using Ru precursors such as RuCl 3 , Ru(NO)(NO 3 ) 3 , Ru(acac) 3 , Ru (NH 3 ) 6 Cl 3 or Ru 3 (CO) 12 [19,20,25,[29][30][31][32][33][34][35]. Limited information is available on the effects of these synthesis parameters on LA hydrogenation activity for Ru/TiO 2 catalysts. For a series of Ni/γ-Al 2 O 3 catalysts, the effect of catalyst preparation method on catalytic activity for LA hydrogenation to GVL has been investigated [36]. Four different Ni catalysts were prepared via wet impregnation, incipient wetness impregnation, precipitation and flame spray pyrolysis, leading to different textural and, as a result, catalytic properties with wet impregnation giving the best results. Another example concerns the hydrogenation of LA to pentanoic acid using Ru supported on H-ZSM-5 [37], for which the use of Ru(NH 3 ) 6 Cl 3 as precursor resulted in much better yields than when using RuCl 3 .
To the best of our knowledge, a systematic study on the influence of several important synthesis variables (Ru precursor, calcination and/or reduction step) in the catalyst preparation procedure for the hydrogenation of LA with supported Ru catalysts has not been reported to date. We here report such an approach using Ru supported on pure anatase TiO 2 . The support choice was fixed to reduce the complexity of possible variations during the catalyst synthesis procedure. In addition, we have recently shown good catalytic results for Ru on anatase TiO 2 [30]. Perusal of the literature on Ru/TiO 2 catalyzed LA hydrogenation ( Table 1) finally shows that catalyst performance not only depends on support choice and catalyst synthesis parameters, but that a strong dependence on the choice of solvent can also be expected. Therefore, catalyst activity and selectivity to GVL were determined in both water and dioxane, the latter being a typical example of an organic solvent. It is well known that higher activities are attainable for reactions run in water compared to dioxane at otherwise similar conditions. As such, an optimum catalyst preparation procedure for catalysis in water not necessarily be the best for dioxane. Catalyst characterization (N 2 physisorption, Transmission Electron Microscopy (TEM) and Scanning Electron Microscopy (SEM) were carried out to rationalize the results and provided insight into the relation between catalyst performance and catalyst structure.

X-Ray diffraction (XRD)
The crystal form of the TiO 2 support was determined using X-ray diffraction (D8 Advance, Bruker) with filtered Cu K α radiation and a wavelength of 1.5404 Å. The tube voltage was 40 kV, and the current was 30 mA. The 2θ scanning rate was 0.02°/min in a range from 5 to 80°.

Transmission electronic microscopy (TEM)
TEM measurements in bright field mode were conducted with a CM12 microscope (Philips), operating at 120 keV. Samples were made by ultra-sonication in ethanol and dropping the suspension onto carbon coated 400 mesh copper grids. Images were taken on a slow scanning CCD camera. The metal particle size distribution is calculated by measuring at least 100 particles with the software Nano Measurer 1.2.

Scanning electronic microscopy (SEM)
The average size of the TiO 2 particles was determined using Scanning Electron Microscope Jeol JSM-7000F (Field Emission Scanning Electron Microscope). For this purpose, the size of about 300 particles was measured.

Nitrogen physisorption experiments
Nitrogen physisorption experiments were carried out in a Micromeritics ASAP 2020 at −196.2°C. The samples were degassed in vacuum at 200°C for 10 h. The surface area was calculated using the standard BET method (SBET). The single point gas adsorption pore volume (V T ) was calculated from the amount of gas adsorbed at a relative pressure of 0.98 in the desorption branch. The pore size distributions (PSD) were obtained from the BJH method using the adsorption branch of the isotherms. The mean pore size was taken as the position of the PSD maximum. The t-plot method was employed to quantify the micropore volume (V M ).

Catalyst preparation
A total of sixteen Ru/TiO 2 catalysts were synthesized via a standardized wet impregnation procedure. The anatase titania support was dried (120°C for 4 h in air) before use. The ruthenium loading was kept constant at 1 wt.% Ru for all catalysts. A number of parameters were varied: Ru precursor (RuCl 3 ·xH 2 O; RuNO(NO 3 ) 3 or Ru(NH 3 ) 6 Cl 3 ), calcination step (with or without) and the amount of hydrogen in the gas phase during the reduction step (5, 10, 50 or 100 vol.% H 2 ). Throughout this paper, the catalysts are abbreviated based on the precursor used for its preparation and the calcination/reduction procedure. For instance, Ru/TiO 2 (Cl,N,10) is a catalyst made using RuCl 3 as the precursor, without calcination and reduced in an atmosphere with 10 vol.% H 2 in nitrogen.
A typical catalyst synthesis entailed the dissolution of a required amount of the precursor in 25 mL of water while stirring (1100 rpm) for 30 min at 30°C to obtain a homogeneous solution. The titania support (1 g) was added gradually to the precursor solution under stirring. Subsequently, the temperature was increased to 85°C and kept until complete evaporation of the water (after about 19 h). After the impregnation, the catalyst was calcined and/or reduced. Catalysts prepared without a calcination step were directly reduced after the impregnation. When the catalysts were calcined, this was done after impregnation, followed by the reduction step. Calcination was performed using a Micrometrics AutoChem II 2920 system at 450°C (heating rate 2°C/min) in a nitrogen atmosphere (N 2 flow of 100 mL/ min) for 4 h. The reduction step was carried out in the same system at 400°C for 4 h. Either a N 2 /H 2 mixture (with variable composition) or pure H 2 was used, the total gas flow rate was maintained at a flow rate of 100 mL/min.

Catalytic hydrogenation of LA
LA hydrogenation reactions were performed according to previously published standard conditions for water [38], and dioxane [39]. The hydrogenation reactions using water as the solvent were performed in a stainless steel batch autoclave (100 mL, Parr Instrument Company). The heating mantel of the autoclave was equipped with electric heating rods and a cooling coil (using water) for proper temperature control. The reactor content was agitated using an overhead stirrer (Heidolph, RZR 2102 control). The temperature (90°C) and hydrogen pressure (45 bar) were measured online with a Eurotherm 2208e. For liquid sampling during the reactions, the reactor was equipped with a dip-tube. LA (2.9 g) was dissolved in water (40 mL, 0.6 M) and catalyst (0.06 g, LA/ Ru molar ratio of 4350) were introduced in the autoclave. The stirrer was started (2000 rpm) and the system was flushed with nitrogen for 5 min. The mixture was heated to the desired temperature and subsequently hydrogen was admitted to the reactor to 45 bar. This moment is set as t = 0 min. During the reaction, hydrogen was regularly admitted to the reactor to keep the pressure constant at 45 bar. The information has been added in the experimental section in the manuscript. 16 catalysts were prepared and tested for the hydrogenation reaction in water and dioxane. The hydrogenation result for one of the catalysts (Ru/ TiO2(NO,N,100, particle size 4.7 nm) in water was discarded from the dataset as hardly any reactivity was observed, likely due to an unknown experimental error.
The reactions in dioxane were performed in a 50 mL Parr batch autoclave at a temperature of 150°C for 4 h using a hydrogen pressure of 45 bar and a stirring speed of 1250 rpm. The tests were typically performed using 27.8 g of dioxane, 7.37 wt.% of levulinic acid (2.2 g, 0.63 M) with 0.06 g of catalyst (LA/Ru molar ratio of 3200). Then the autoclave was purged three times with argon after which the reaction mixture was heated to reaction temperature and charged with H 2 . This was taken as the starting point of the reaction; during the reaction samples were taken regularly, filtered and 1 wt.% of anisole was added as internal standard to the samples. At the end of the reaction, the autoclave was cooled rapidly to room temperature in an ice bath, after which the remaining H 2 was released.
An overview of reaction conditions is given in Table 2.  (1) where N a is the number of hydrogen atoms of the NMR resonances used for the calculation (δ 2.1 ppm for LA (3 hydrogen atoms), δ 1.03 ppm for 4-HPA (3 hydrogen atoms), δ 1.3 ppm for GVL (3 hydrogen atoms)) and N S is 8, being the number of hydrogen atoms of dioxane at δ 3.6 ppm. The concentrations of LA, GVL and 4-HPA in the samples were calculated using Eq. (2).
Where V t is the volume of the mixture in NMR tube and D f the dilution factor, which was calculated as follows (Eq. (3)

Analyses of the reaction products from LA hydrogenation in dioxane
Reaction products were analyzed using a Shimadzu GC-2010A gas chromatograph equipped with a CPWAX 57-CB column (25 m × 0.2 mm × 0.2 μm) and FID detector, using authentic samples for calibration.

Definitions
The conversion of LA and the yield and selectivity for the products (4-HPA and GVL) were calculated according to Eq. (4)-(6).
Where X LA is the conversion of LA (mol.%); C LA,0 the initial concentration of LA (mol/L); C LA the concentration of LA at a certain time t (mol/L); Y GVL the yield of GVL (mol.%); Y 4-HPA the yield of 4-HPA (mol. %) and S i the selectivity to GVL or 4-HPA (mol.%). The initial reaction rate (R 0 , mol/mol Ru s) was determined from the experimentally obtained concentration-time profiles using a procedure given by Fogler [40]. For this purpose, the concentration time profile was modeled using a higher order polynome. The initial rate was determined by differentiation of the polynome and setting the value for the time at zero.
Where R 0 is the initial reaction rate (mol/mol Ru s) and D Ru is the dispersion of ruthenium, which was calculated as follows (Eq. (8)): where V Ru is the volume occupied by a bulk Ru atom (0.01365 nm 3 ), a Ru is an area per Ru atom (0.0635 nm 2 ) and d Ru is the average diameter of a Ru particle (nm, taken from the TEM data).

Catalyst characterization
Sixteen Ru/TiO 2 catalysts were synthesized using RuCl 3 ·xH 2 O, RuNO(NO 3 ) 3 , Ru(NH 3 ) 6 Cl 3 as Ru precursors, in/excluding a calcination step, and with different hydrogen partial pressures during the reduction step (5, 10, 50 or 100 vol.% H 2 ). Relevant properties of the various Ru/ TiO 2 catalysts were determined (BET surface area, micropore volume and the average Ru nanoparticle size) and the results are provided in Table 3. N 2 -physisorption analysis of fresh catalysts show that the surface areas are between 13 and 9 m 2 /g, which are lower than for the bare anatase TiO 2 support (17 m 2 /g). The average Ru particle size was determined with TEM (Table 3) and two representative TEM pictures are shown in Fig. 1. The average Ru nanoparticle size varied considerably over the series, ranging from 1.1 to 14.2 nm. In some cases, especially when the Ru nanoparticle sizes are larger than 6 nm, considerable particle agglomeration was observed.
The observation on the basis of TEM that well-dispersed Ru on TiO 2 (A100) with small average Ru nanoparticle sizes are attainable for some of the catalysts is surprising considering a recent study on Ru/TiO 2 (A100) [32]. Here, the wet impregnation of ruthenium (Ru(acac) 3 on large surface-area TiO 2 (A100, 83-336 m 2 /g), followed by calcination at 200°C in air and a reduction at similar conditions with hydrogen resulted in an inhomogeneous metal distribution with very large Ru nanoparticles (as seen by TEM) and small ones in the micropores. A Ru/ TiO 2 (A100) catalyst for which the support was calcined at high temperature before impregnation to reduce microporosity showed no particle agglomerates and a rather homogeneous distribution was obtained with an average Ru particle size of 4.5 nm. Comparison between our results and these given above is difficult as the anatase support used in our study has a far lower BET surface area (17 m 2 /g versus 83-336 m 2 / g in [32]). However, it shows that the catalyst preparation procedure is of high importance and under optimum conditions, highly dispersed Ru/TiO 2 (A100) can be obtained.

Hydrogenation experiments using the anatase-supported Ru catalysts
Initial hydrogenation experiments were performed with Ru/ TiO 2 (Cl,N,10) in both water and dioxane. Standard reaction conditions were selected based our previous results [19,38,39], and are given in Table 2. The lower reaction temperature selected for the reaction in water (90°C) compared to dioxane (150°C) reflects the surprisingly high activity of Ru-based hydrogenation catalysts under aqueous conditions, for which the underlying reasons have been recently discussed [15,17,19,38,41,42]. The hydrogen pressure was set at 45 bar for all experiments. Concentration-time profiles for the main components (LA, 4-HPA and GVL) under these standard reaction conditions are shown in Fig. 2.
In the case of water as the reaction medium, substantial amounts of 4-HPA are formed, in line with our own observations and literature data  [23,38,41]. Other intermediates (e.g. α-angelicalactone) or consecutive hydrogenation products of GVL (e.g., 1,4-pentanediol or 2-methyltetrahydrofuran) were not detected. Full conversion of LA was achieved within 6 h with 89% selectivity to GVL, the remainder being 4-HPA (Fig. 2). The intramolecular esterification of 4-HPA is known to be an equilibrium reaction with an equilibrium constant of about 16 [38]. The ratio of GVL/4-HPA obtained here after 6 h of reaction is 5, indicating that the two components are not yet at equilibrium. The reaction in dioxane at 150°C gave almost quantitative LA conversion after 4 h (Fig. 2), with full selectivity to GVL. No 4-HPA was detected in dioxane, as the intramolecular esterification reaction to GVL is much faster in organic solvents than in water due to equilibrium considerations (Scheme 1) [38].

Effect of variation of the Ru-precursor on catalytic activity
Variation of the Ru precursor at otherwise fixed synthesis conditions (i.e., no calcination and direct reduction with 10 vol.% H 2 ) gave the catalysts Ru/TiO 2 (Cl,N,10), Ru/TiO 2 (NO,N,10) and Ru/ TiO 2 (NH 3 ,N,10) for which LA hydrogenation activity is shown in Fig. 3 and a compilation of the data is given in Table 4.
Remarkable differences in catalyst activity were observed for the three precursors, with the trends in activity being the same in both solvents.The catalyst prepared with RuCl 3 was the most active in both cases (Table 4), intermediate activity was found for RuNO(NO 3 ) 3 and lowest LA conversions was seen after 4 h for Ru(NH 3 ) 6 Cl 3 . In all cases, GVL was the major product, with, as expected, minor amounts of the intermediate 4-HPA for the reactions performed in water.
However, the general trend regarding LA conversion after 4 h is not exactly the one as described above for catalysts prepared by a direct reduction with 10 vol.% H 2 . In general, the catalysts prepared with RuCl 3 and RuNO(NO 3 ) 3 gave the highest LA conversion after 4 h ( Table 4). The highest activity for a catalyst prepared using the RuCl 3 precursor in water was observed for catalyst 1 (86% LA conversion), compared to 77% LA conversion when using the RuNO(NO 3 ) 3 precursor (catalyst 10). These highest conversions were obtained for the uncalcined catalysts which were reduced at relatively low hydrogen partial pressures. For catalysts made with the Ru(NH 3 ) 6 Cl 3 precursor, LA conversions were in general by far lower than for the other two precursors and best performance was found for catalyst 12 (24% LA conversion). The choice of Ru precursors will influence the average Ru particle size, which is expected to play a large role in determining catalyst activity. In this study, the highest activity is observed for the Ru catalyst with an intermediate Ru particle size (Ru/TiO 2 (Cl,N,10), 2.2 nm). In addition, the Ru oxidation state might play a role as well, with Ru 0 nanoparticles assumed to be the catalytically active species [43], even though RuO 2 species may also play a role [41,44,45].

Effect of reduction procedure on catalyst activity
For a number of catalysts, prepared with RuCl 3 as the precursor, the partial pressure of hydrogen was varied in the reduction step (10, 50 and 100 vol.% H 2 in N 2 /H 2 mixture, 1 bar). The results obtained with the catalysts (Ru/TiO 2 (Cl,N,10), Ru/TiO 2 (Cl,N,50) and Ru/ TiO 2 (Cl,N,100) in Table 4) under standard conditions in both solvents are presented in Fig. 4.
The partial pressure of hydrogen during the catalyst reduction step has a profound influence on activity. Again the order of activity is the same for both solvents, with the best results obtained with the catalyst reduced at 10 vol.% of H 2 . The TEM results show the same particle sizes of around 2.2 nm for the reduction with 10 and 50 vol.% H 2 , while 100 vol.% H 2 leads to agglomeration and an average particle size of 8.4 nm. The catalytic results obtained with Ru/TiO 2 (Cl,N,10) and Ru/ TiO 2 (Cl,N,50) differ considerably, despite the same TEM Ru particle size and indicate that factors other than mere particle size affect catalytic activity (vide infra).  (Table 2). The effect of a calcination step for Ru-catalyzed reactions has been reported in a number of studies. For instance, Luo et al. reported a negative effect on activity for the hydrogenation of LA in dioxane using a Ru/H-ZSM-5 catalyst [37]. The formation of highly mobile RuO x species during the calcination step and subsequent sintering, was mentioned as the most likely reason. Ruppert et al. showed that avoiding a thermal treatment step in the catalyst preparation protocol was beneficial for catalytic activity for LA hydrogenations using Ru/ TiO 2 (A100) in water [32]. This finding was attributed to a lowering of the average Ru particle size for the mild chemical treatment compared to calcination, leading to higher catalyst activities. The negative effect of a high temperature calcination step on the average Ru particle diameter (and catalytic activity) was also observed for the hydrogenation of o-xylene over Ru/TiO 2 , Ru/Al 2 O 3 and Ru/SiO 2 catalysts [46], again rationalized by agglomeration of Ru particles as a consequence of the high migration rates of RuO 2 species at elevated temperature [47,48].
We tested the effect of an intermediate calcination step during catalyst preparation on the catalytic activity in water for the catalysts prepared with the RuCl 3 precursors and the results are given in Fig. 5.
An intermediate calcination step negatively affects catalyst performance, as clearly demonstrated by the lower activity seen for all calcined catalysts when using RuCl 3 , in line with the literature data reported above. Calcination indeed led to an increase in average Ru nanoparticle size when using RuCl 3 as precursor for catalyst synthesis, see Fig. 6 for details. However, for catalyst prepared with the Ru(NH 3 ) 6 Cl 3 precursor, the calcined sample (catalyst 15) has a lower average Ru nanoparticle size than the uncalcined one (catalyst 14). Whether this is specific for catalysts prepared with the Ru(NH 3 ) 6 Cl 3 precursor cannot be concluded unequivocally as the dataset is too limited, with only two samples for Ru (NH 3 ) 6 Cl 3 compared to seven for RuCl 3 .
The effect of the partial pressure of hydrogen during the reduction step on catalyst activity of the calcined catalysts reveals the presence of an optimum value for intermediate hydrogen levels (Fig. 5). This might be the result of a balance in the average Ru particle diameter and extent of Ru reduction, which are expected to oppositely affect catalyst activity. TEM analysis (Fig. 6) shows that higher amounts of H 2 in the reduction step for the calcined samples led to sintering, which is expected to have a negative effect on catalytic activity.
3.6. Discussion 3.6.1. Effect of ruthenium particle size on catalytic activity for LA hydrogenation in water The effect of the Ru nanoparticle size on catalytic activity for the hydrogenation of LA in water was investigated by comparing the initial rate of all reactions (mol/mol Ru .s) and TOF (s −1 ) versus the average Ru nanoparticle diameter as determined by TEM, see Fig. 7. The TOF values are a better activity performance indicator as these are normalized on the surface Ru atoms (based on TEM particle size data).
For catalysts, prepared with the RuCl 3 precursor, a volcano type plot is found for the initial rate as function of Ru particle size, with the highest activity for Ru nanoparticles with an average diameter of about 2 nm (Fig. 7, left). A similar trend can be seen for the catalysts prepared with the other precursors.
A number of papers have reported the effects of Ru nanoparticle size on catalytic activity for hydrogenation reactions with Ru on TiO 2 catalysts. Ruppert et al. [32], reported that the optimum Ru particle size for LA hydrogenation (in terms of LA conversion after a fixed batch time) for Ru/TiO 2 (A100) is in the range of 3-4 nm. The authors proposed that the activity for smaller Ru particles (< 1 nm) is lower as these are present in the micropores and thus less available for the hydrogenation reaction amongst others due to diffusion limitations [32]. Primo et al. examined the influence of metal nanoparticle size of Ru/ TiO 2 (0.64 wt.% Ru) catalysts for the hydrogenation of lactic acid to 1,2-propanediol [49]. When increasing the Ru particle diameter from 2 to 10 nm by subjecting the catalysts to a calcination step at 400°C, the product yield decreased from 99 to 38%.
TOF values, expressed as mol LA converted per mol of surface Ru per s, as a function of the Ru nanoparticle size (Fig. 7) show volcano type plots with a maximum TOF within the range. This implies that LA hydrogenation is structure sensitive for these catalysts. Surface sites of different configuration (e.g. ensembles of surface atoms), of which the concentration is a function of particle size and shape, then contribute differently to the observed, averaged reaction rate. Detailed overviews on structure sensitive (hydrogenation) reactions are given by Murzin and van Santen et al. [50,51]. Recent research by Cao et al. on the hydrogenation of LA using both Ru/Al 2 O 3 and Ru/C catalysts also revealed the presence of typical volcano-type activity plots when the TOF was plotted as a function of the average Ru particle size [52]. A maximum TOF was found for catalyst with an average Ru particle diameter of about 1.5 nm for both supports.
Our dataset is too limited to draw definite conclusions regarding the optimal Ru nanoparticle size for the hydrogenation of LA, but seems to be between 1 and 3 nm for catalysts prepared with the two chloride precursors, RuCl 3 , Ru(NH 3 ) 6 Cl 3 and between 1 and 4 nm for RuNO (NO 3 ) 3 , in line with the data from Cao et al. for Ru/Al 2 O 3 and Ru/C catalysts.
Of interest is the observation that the highest TOF value in the range is observed for the catalyst prepared with the RuNO(NO 3 ) 3 precursor. One of the possible reasons for a better performance of the catalysts prepared with this precursor is the absence of chloride residues, which may be present when using the other two chlorine containing precursors. Analyses of the reduced catalysts by XPS shows the presence of only minor amounts of precursor-related elements. The content of Cl in Ru/TiO 2 (Cl,N,10) catalyst is below 0.6 at.%, whereas it is below the detection limit for Ru/TiO 2 (NO,N,10) and Ru/TiO 2 (NH 3 ,N,10). As such, there is not a clear relation between the activity of the catalysts and the chlorine content of the samples. The nitrogen content in the three samples was also below the detection limit, indicating that differences in residual nitrogen content do not play a role.
Catalyst activity thus depends on Ru nanoparticle size; the observation that the activity versus Ru-nanoparticle size plots are also a function of the precursor used during synthesis, also suggests that other factors play a role as well, however. These can include the ratio of various charged and non-charged Ru species, or the presence of minor amounts of residues from the catalyst precursors.  5. Effect of the composition of the reduction gas and the use of an intermediate calcination step on the initial rate for the catalytic hydrogenation of LA in water for various Ru/TiO 2 catalysts (prepared using RuCl 3 as the precursor) under standard reactions conditions ( Table 2). Fig. 6. Effect of the hydrogen amount in the reduction gas on the Ru particle size diameter for Ru/TiO 2 catalysts prepared with RuCl 3 precursor with and without a calcination step.

Solvent effect
The hydrogenation reactions as reported in this study were carried out both in water and dioxane (see Table 2 for experimental details). Comparison of the LA conversion after 4 h reaction time for both water and dioxane in the form of a parity plot is provided in Fig. 8. A reasonable fit was obtained, showing that the performance in water and dioxane follow a similar trend. As such, the effect of Ru particle size distribution on catalytic activity as shown for water, seems to be valid for dioxane as well. In addition, it suggests that the molecular mechanism for the hydrogenation reactions in both solvents is similar.

Conclusions
We have shown that the performance of Ru/TiO 2 catalysts for the hydrogenation of LA in water and dioxane strongly depends on the preparation protocol (nature of the Ru-precursor, calcination and/or reduction step and the amount of hydrogen present in the sweep gas during reduction). Best results were obtained when using the RuCl 3 precursor (when considering initial rate on Ru intake) and the RuNO (NO 3 ) 3 precursor (when considering TOF) without an intermediate calcination step and 10% hydrogen in the reduction gas, whereas calcination and the use of a hydrogen rich sweep gas gave rise to lower catalytic activity. Catalyst characterization studies showed that the different preparation protocols lead to catalysts with differences in average Ru particle sizes. It was shown that catalyst activity (TOF and initial rates) in both water and dioxane is strongly correlated with the average Ru particle size, and an optimum activity was found for average Ru particle sizes of about 1-5 nm. This implies that LA hydrogenation in water is structure sensitive. The fact that catalysts with similar particle sizes though prepared with different precursors show different TOF values suggest that additional factors than particle size-dependent structure sensitivity must play a role. A possible factor is the difference in amounts of charged and neutral ruthenium species, which are expected to have different activity profiles. However, the influence of minor amounts of precursor residues such as Cl, as found for Ru/TiO 2 (Cl,N,10), cannot be ruled out. Fig. 7. Initial rate and TOF for LA hydrogenation in water versus the average Ru-nanoparticle for the Ru/TiO 2 catalysts prepared in this study according to TEM. Reaction conditions: see Table 2.