Chemoselective lactonization of renewable succinic acid with heterogeneous nanoparticle catalysts

. The production of chemicals from renewable resources, resulting in the establishment of bio-refineries, represents a challenge of increasing importance. Here we show that succinic acid, a C4 compound increasingly being produced on a kiloton scale by the microbial fermentation of sugar, can be selectively converted into a variety of important chemicals. Optimal performance in terms of activity, selectivity and reusability is observed with Al 2 O 3 -supported Pd nanoparticles, which mediate the selective, hydrogenative lactonization of succinic acid to gamma-butyrolactone at > 90 % selectivity, even at high levels of conversion (< 70 %). Through a variety of kinetic, spectroscopic and microscopic studies, preliminary structure-activity relationships are presented, and the roles of the reaction conditions, the choice of metal and the nature of the support in terms of guiding the overall process selectivity, are also investigated. On a broader level, these studies demonstrate the suitability of succinic acid to act as a platform for renewable chemical production in future bio-refineries. a determined by SEM/EDX analysis, b calculated from porosimetry analysis by N 2 adsorption by the BET method, c calculated from TEM images. d n.d not determined, due to poor contrast in TEM between particle and support.


Introduction.
Over the last century, the world has become increasingly dependent on fossil feedstock for the production of fuels, base chemicals, and commodities, including fibers, pharmaceuticals, detergents, plastics, pesticides, fertilizers, lubricants, solvents and much more. 1,2 In fact, over 99 % of all plastics are produced from chemicals sourced from fossil fuels. 3 However, the decreasing availability and volatile pricing of crude oil, the geopolitical and economic issues relating to its unequal distribution, coupled to the harmful nature of the by-products generated through its consumption, have stimulated the chemical society to search for alternative sources for fuels and chemicals. [4][5][6] Although a variety of renewable vectors hold potential for energy production, the production of chemicals requires carbonbased resources. When considering alternative carbon-based raw materials for the chemical industry, options are mainly limited to plants or atmospheric carbon (CO2). Although routes exist for converting atmospheric CO2 to liquid products, several technical limitations currently impact the feasibility of these approaches. Thus, bio-renewables, such as lipids, starches and celluloses, are emerging as the most sustainable class of feedstock for chemicals. [7][8][9] In fact, the high functionality and chemical diversity of bio-renewables makes them an excellent raw material for the production of value-added chemicals and polymers, 10,11 Given their high oxygen content and functionality, the utilization of biomass as industrial feedstock also eliminates the requirement for some functional group insertions, thereby reducing the total number of processing steps required. Moreover, it permits the production of higher carbon chain length molecules (C4+) that are not typically present in other emerging feedstock, such as shale gas. As cellulose-based derivatives account for 60-90 % (by weight) of all bio-renewable material, they therefore represent the most viable and sustainable source of carbon chemical production. Of particular interest is cellulose derived from second-generation biomass and energy crops, 12 which offer exhibit several advantages including 1) an ability to be grown on marginal land; 2) higher levels of productivity in terms of sugar produced per acre and per unit time, and; 3) lower lifecycle CO2 emissions. For example, the lifecycle CO2 emission from energy cane is 3-4 times lower than those of corn and sugarcane, and 7 times lower than that of gasoline. Moreover, second-generation biomass does not compete with food sources, unlike other firstgeneration biomass, such as bio-ethanol. 13 Amongst possible strategies for converting renewables into chemicals, catalytic methodologies offer several advantages, particularly in the context of process intensification. Of these, one of the most promising involves the catalytic conversion of various 'platform' molecules, which are themselves obtained via controlled depolymerization or fermentation of cellulose. 14 Examples of such platform molecules include glucose, fructose, 5-hydroxymethylfurfural, 2,5-furandicarboxylic acid, 3hydroxypropionic acid, levulinic acid and glycerol, amongst others. 15,16 In fact, the catalytic conversion of several of these platforms has received an enormous amount of attention over the last decade.
However, despite the surge of interest in renewable chemical production from these platform molecules, a number of other platforms with high potential have received relatively little attention. A key example is succinic acid (SA). Classically obtained by the hydrogenation of butane-derived maleic anhydride, recent breakthroughs in biotechnology mean renewable SA is now being produced via the fermentation of glucose on a kilotonne scale by several companies. [17][18][19] However, despite the advanced stage of these production processes, and its widely accepted high potential as a platform molecule, the catalytic conversion of succinic acid to value added products has received scant attention, despite the plethora of products that can in principle be obtained from it. In fact, an important number of commodity chemicals, 20,21 including 1,4-butanediol (BDO), tetrohydrofuran (THF) and gammabutyrolactone (GBL), can all be obtained from succinic acid under the guidance of a suitable catalytic material ( Figure 1). Amongst these processes, the conversion of succinic acid to GBL has been identified as being one of the most technologically attractive, as it opens up routes towards various commodity compounds and additional downstream monomers, such as pyrrolidone. Error! Bookmark not defined.,22 To the best of our knowledge, however, only a handful of reports concerning the catalytic conversion of succinic acid can be found in the open literature. [23][24][25][26][27][28] In each of these previous studies, the reductive valorisation of SA was investigated. Although classical hydrogenation catalysts, such as supported Ru and Pd nanoparticles, have been shown to be active heterogeneous catalysts for this reaction, very harsh conditions, such as high reaction temperatures (> 240 °C) and high pressures (< 80 bar) have previously been required (Table S1). [29][30][31][32][33] Even at these conditions, the productivities of the processes are also rather low. Moreover, and more crucially in the context of process intensification, it is clear that investigation of the overall reaction network, and particularly the selective production of one particular reaction product, has not yet been achieved. In fact, in previous reports a variety of reaction products are typically observed, particularly at elevated levels of productivity. [34][35][36] We note that although achieving high levels of selectivity during the upgrading of bio-renewables is especially challenging, given their high levels of functionality and reactivity, it is an essential task given the economic and energetic cost of downstream separation. As such, development of a catalytic system that exhibits both high selectivity and productivity at mild conditions has not yet been achieved.
In this manuscript, we demonstrate that various supported hydrogenation catalysts, such as Pd/Al2O3 and Ru/Al2O3, are able to convert SA selectivity to GBL at relatively mild conditions (140-170 °C, 15-30 bar H2). Although commercially available catalysts demonstrate high levels of performance, optimal performance in terms of activity and selectivity to GBL is achieved with 5 wt. % Pd/Al2O3, prepared by co-precipitation, which results in GBL being produced at greater than 90 % selectivity even at high levels of conversion (< 70 %). Interestingly, the choice of active metal dramatically impacts the overall selectivity of the reaction under otherwise identical conditions, with Ru resulting in comparable to selectivity to both GBL and propionic acid (PA), whereas utilization of Pd results in almost exclusive production of GBL. Kinetic, spectroscopic and microscopic studies indicate that overall performance depends on the reaction conditions, the composition of the catalyst (choice of metal, metal loading and nature of support), in addition to the method of catalyst preparation.

Results and discussion.
Optimization of reaction conditions. The catalytic hydrogenation of SA can result in a variety of important commodity chemicals, including THF, GBL and BDO, amongst others. Given how the reaction conditions can impact overall activity and selectivity, the choice of reaction conditions including time, temperature, pressure and catalyst loading require optimization for the highest levels of performance to be achieved. Given the reported ability of Ru nanoparticles to hydrogenate SA, optimization studies were performed with a commercially available Ru/Al2O3 catalyst (Sigma Aldrich, 5 wt. % Ru, reduced, henceforth denoted 5Ru/Al2O3(COM)). To identify the optimal temperature regime of the reaction, a number of kinetic experiments were performed at various temperatures and times, to be generated (Figure 2, Right). This revealed the activation energy of the system to be 78.9 kJ mol -1 .
Further control experiments revealed that in the absence of catalyst and H2, no conversion of SA was observed (Table S2), confirming the heterogeneous and hydrogenative nature of the reaction, respectively.
To gain better insight of how the choice of temperature and time influence the overall reaction performance, the product distribution obtained after 4 h of reaction between 170-230 °C was investigated in greater detail. Under the influence of 5Ru/Al2O3(COM), SA was primarily converted to GBL and propionic acid (PA). However, smaller amounts of butyric acid (BA) were also observed in solution, and traces (< 1 mol. % carbon) of CO2 were detected in the gas phase by GC-FID (methaniser) and GC-MS analysis. In total, these products and the remaining unconverted substrate accounted for > 95 % of the carbon balance in each of these reactions. Preliminary analysis indicates that GBL, the desired reaction product, is preferentially observed at lower temperatures. However, selectivity during heterogeneous catalysis is always a function of conversion. 37 Consequently, more  and comparable selectivity to both GBL and PA is always observed across the entire H2 pressure range.
Thus, it is clear that optimal performance in terms of activity, GBL selectivity and process favorability is achieved at 170 °C, H2 pressures of 30 bar, and catalyst loadings of 2 mol. % and lower. Catalyst design studies. Alongside the reaction conditions, the properties of a solid -such as the choice of metal, its level of loading on the support, and the method of preparation -can dramatically impact its overall performance as a heterogeneous catalyst. Accordingly, the design of more active and selective heterogeneous catalysts is an essential requirement during the development of new chemical processes. Given the high activity observed over Al2O3-supported Ru nanoparticles, a variety of analogous Al2O3-supported metal nanoparticles were screened for reactivity at the optimal conditions identified for catalyst screening (170 °C, 30 bar H2, 2 mol. % catalyst relative to SA). Initial catalysts were all prepared by co-precipitation, as this is known to permit the production of well-dispersed nanoparticles along with the generation of high surface area supports, both critical factors during heterogeneous catalysis. 38  selectivity to GBL could be maintained at higher levels of yield, extended time online analysis of 2Pd/Al2O3(COP) was performed. As can be seen ( Figure 5, Right), high selectivity (> 90 %) to GBL is maintained even at substantially higher levels of product yield (< 60 % product yield after 480 minutes of reaction). Consequently, in addition to being three times more active than the analogous 2Ru/Al2O3(COP) sample, 2Pd/Al2O3(COP) is also twice as selective, even at high levels of conversion.
Thus, 2Pd/Al2O3(COP) clearly represents the most suitable catalyst for SA valorization. To gain an understanding of structure-function relationships with 2Pd/Al2O3(COP), analogous samples with different metal loadings, and different methods of preparation, were investigated. Impregnation of 2 wt. % Pd onto commercially available γ-Al2O3 (henceforth 2Pd/Al2O3(IMP)) results in very poor levels of activity and selectivity being observed, even after reductive pre-treatment (200 C, 5 % H2/Ar, 3 h) ( Figure 6, Left). However, when 2 wt. % of Pd was impregnated onto a sample of Al2O3 that was itself prepared by co-precipitation (2Pd/Al2O3(IMP@COP)), similar levels of performance to 2Pd/Al2O3(COP) were observed, clearly demonstrating impregnation to be a valid method of preparation. Taking into account both impregnated results, it is clear that the properties of the support can dramatically alter performance of the catalyst. Characterization of both Al2O3 materials (commercial and co-precipitated) by XRD (SI Figure S1) and porosimetry (SI Table S3) revealed the very different nature of coprecipitated Al2O3 with respect to the commercial material. For instance, the XRD pattern of alumina prepared by co-precipitation method is consistent with a pseudo-boehmite phase of alumina, 39 which possesses a much larger surface area than that of commercially available gamma-alumina (245 vs 93 m 2 g -1 , respectively). Analysis of the textural properties of both materials following deposition of Pd (2Pd/Al2O3(IMP@COP) and 2Pd/Al2O3(IMP)) revealed that improved porosity for the boehmite phase was retained even after deposition of the metal (224 vs 85 m 2 g -1 ). However, this 2.5-fold decrease in surface area cannot fully account for the 6-fold decrease in activity of 2Pd/Al2O3(IMP), strongly indicating that other factors of the support (such as acidity) may influence performance.  To further probe how the acidity of the alumina supports may affect catalytic performance, pyridine DRIFT experiments were performed (SI Figure S3), by exposing both alumina samples to pyridine vapours. 40 These studies revealed commercial alumina to much more acidic than alumina prepared by co-precipitation. To examine whether increased levels of acidity could be responsible for poorer catalytic performance, a complimentary sample of 2Pd/H-Beta (SiO2/Al2O3 = 38) was also prepared by impregnation. Despite exhibiting a very high surface area (498 m 2 g -1 ), 41 much poorer levels of performance as also observed when this strongly acidic support was employed (Figure 6, Right).
Unfortunately, complimentary TEM images of Pd supported on commercial gamma-alumina showed that the NPs in this material were highly agglomerated, unlike those found for the co-precipitated catalyst or the IMP@COP catalyst. As such, the poorer levels of performance of 2Pd/Al2O3(IMP) cannot be unambiguously attributed to its porosity or acidity. However, it remains clear that the combination of higher surface area and lower acidity makes Al2O3(COP) a more suitable support than γ-Al2O3 for the preparation of catalytic materials for the system of study, either by hindering interaction of the catalyst with the acidic substrate, and/or by resulting in the deposition of lower activity Pd nanoparticles. Focusing on catalysts involving co-precipitated Al2O3, further analysis of 2Pd/Al2O3(COP) and 2Pd/Al2O3(IMP@COP) by TEM revealed that the co-precipitated sample possessed both a smaller mean particle size and a narrower particle size distribution than the impregnated analogue (2.85 ± 0.76 nm vs. 4.51 ± 2.8 nm, respectively) ( Figure 7). Given their otherwise very similar properties (Table 1), including specific surface area and dominance of metallic Pd (SI Figure S4), the higher levels of activity exhibited by 2Pd/Al2O3(COP) can tentatively be assigned to its more uniform distribution of smaller Pd nanoparticles. Further indication that catalytic activity has a particle size dependency was provided by investigation of the catalytic performance of 5Pd/Al2O3(COP). This material exhibits a comparable particle size distribution (4.83 ± 1.3 nm) to 2Pd/Al2O3(IMP@COP), and almost identical catalytic activity when tested at the same substrate-to-metal molar ratio. From an intensification perspective, the activity of a material per mass charge is one of the most critical factors, given that it choice, design and size of the eventual catalytic reactor. Therefore, although the intrinsic activity i.e. turnover frequency, of 5Pd/Al2O3(COP) is 27 % lower than that of 2Pd/Al2O3(COP) (2.9 h -1 versus 3.8 h -1 ), the 2.5-fold increase in metal loading does substantially boost the productivity of the catalyst, measured on an activity per gram basis, by a factor of two (0.12 h -1 versus 0.06 h -1 ). Additionally, 5Pd/Al2O3(COP) shown to be an excellent catalyst to perform the reaction at extended times and achieve high GBL yield maintaining the high selectivity towards GBL (> 90 %) even after 24 h of conversion (SI Figure S5). Accordingly, for future intensification studies, the higher loaded catalyst (5Pd/Al2O3(COP)) likely presents the optimal compromise between intrinsic activity and catalyst productivity. To further probe this, an extended time online reaction for 5Pd/Al2O3(COP) was also performed at higher temperature ( Figure 8, Left). As can be seen, in only 4 h, a GBL yield of 64 % was obtained during this reaction, representing a productivity of 0.33 g (GBL) g -1 (catalyst) h -1 . Further improvement in terms of GBL productivity was also obtained at higher initial SA concentration, i.e. at higher substrate metal ratio. As shown in Figure 8, Right, reactions performed with SA solutions of 0.6 M (3 times more concentrated), but with the same mass of catalyst (resulting in a Pd content of 0.67 mol. %, as opposed to 2 mol. %) resulted in an increased GBL productivity of 0.39 g (GBL) g -1 (catalyst) h -1 being obtained (SI Table   S1, Entry 12). In this case, high selectivity towards GBL could be maintained over extended periods of time and at elevated conversion, as demonstrated by running a reaction for 16 h at the higher concentration of SA (SI Figure S6).  Alongside high levels of catalytic activity and selectivity, heterogeneous catalysts must also possess high levels of stability, and an ability to operate for more than one reaction period. 42 Unfortunately, several factors make the design of reusable catalysts for biomass conversion a major challenge, including the harsh reaction conditions typically required (presence of solvent at high temperature and pressure), and the chelating nature of the oxygenated substrates, such as sugars and acids, which are known to cause leaching and agglomeration of various metallic nanoparticles. 43,44 Thus, to gain a preliminary understanding of the reusability properties of 2Pd/Al2O3(COP), recyclability experiments were performed. To do so, the catalyst was filtered out of the reactor following one kinetic experiment, and subsequently reused in a second and third catalytic experiment. To achieve the most accurate level of insight, and to probe intrinsic reusability of the catalyst, no intermediate treatments were performed on the material i.e. the catalyst was simply filtered from solution, dried at room temperature, and then placed back in the reactor. Furthermore, comparison of the initial activity of the material was done, to As can be seen (Figure 9), excellent reusability is observed for 2Pd/Al2O3(COP), with no loss of activity being observed over three kinetic cycles. The high levels of stability exhibited by the catalyst is a promising discovery for future intensification studies. 42 Error! Bookmark not defined. Accordingly, to gain a greater understanding of the reaction system, in addition to gaining some insight into the more selective mechanism of Pd during SA valorization, product stability studies were performed. These were performed by substituting the SA substrate for other feasible products of the reaction network, including PA, BA, GBL and BDO, and testing their reactivity as substrates under similar reaction conditions. For these studies, 5Ru/Al2O3(COM) was employed, as this catalyst exhibited the highest degree of consecutive and/or parallel reaction pathways. As can be seen ( with only GBL demonstrating more than trace levels of conversion to any other product. The low levels of reactivity observed in each of these cases indicates that there are no major consecutive reactions present, and that the main products observed during SA conversion with 5Ru/Al2O3(COM) (GBL and PA) occur from two separate pathways, allowing disclosure of the full reaction network (Figure 10). This is further supported by the observation that the selectivity obtained to GBL during SA hydrogenation with 5Ru/Al2O3(COM) is almost invariant at all levels of product yield (Figure 3, right).
From this, it can be ruled out selectivity to GBL is not lost with 5Ru/Al2O3(COM) due to its ability to convert GBL to consecutive products. Rather, the presence of Ru opens up additional reaction pathways that are not observed over 2Pd/Al2O3(COM).  To gain further insight as to the more selective nature of Pd/Al2O3 versus Ru/Al2O3, preliminary spectroscopic experiments focused upon the activation of carboxylic acids by these materials were performed by Diffuse Reflectance Fourier Transform Infra-red (DRIFT) spectroscopy. Unfortunately, the poor volatility of SA prohibits its study by this technique. To overcome this, and to focus exclusively upon the binding of the acidic functionality, DRIFTS measurements were thus performed on pivalic acid, a tertiary carboxylic acid that cannot undergo lactonization. After dosing with pivalic acid, all three samples exhibit vibrational patterns consistent with physisorbed pivalic acid (SI Figure   S7). 45 However, following evacuation at 100 °C and removal of physisorbed pivalic acid, obvious differences between both active catalysts (5Ru/Al2O3(COM), 5Pd/Al2O3(COM)) and the inactive support material alone are observed (Figure 11, SI Figure S8), and stretches related to the deprotonated pivalate are also observed. 45 Since no conversion occurs in the absence of hydrogen (Table S2) and the probe molecule pivalic acid cannot undergo lactonization, it can be discounted that these changes are due to the formation of a reaction product.  Figure 11 presents the DRIFT spectra of 5Ru/Al2O3(COM), 5Pd/Al2O3(COM) and Al2O3, following dosing with pivalic acid and subsequent heat treatment at 100 °C (Left) and 300 °C (Right). When compared to the inactive support material itself, two distinct vibrations in both metal-doped catalysts can be distinguished at approximately 1650 cm -1 and ± 1415 cm -1 . These can readily be attributed to the assymetric (vas) and symmetric (vsym) stretches of the metal-coordinated carboxylate, respectively. 46 Interestingly, the precise wavenumbers of these stretches differs according to the choice of metal, with the difference () in energy between vas and vsym ( = vasvsym) increasing from 214 cm -1 for 5Ru/Al2O3(COM) to 264 cm -1 for 5Pd/Al2O3(COM). The value of relates to the type of binding mode in metal-acetate species, with increasing values of  indicating greater inequivalence between the C-O bonds, and a shift towards more unidentate coordination. 45 As such, these studies indicate that carboxylic acids, such as pivalic acid and SA, bind differently to 5Pd/Al2O3(COM) and 5Ru/Al2O3(COM).
As can be seen, this difference in coordination is still evident even at higher desorption temperatures ( Figure 11, Right and Figure S8).
Unfortunately, the poor volatility of SA (Vide Supra) prohibits its detailed vibrational study by these methods. Hence, to further explore how SA coordinates to both 5Pd/Al2O3(COM) and 5Ru/Al2O3(COM), computer simulations focused on the adsorption of SA onto metal surfaces were performed (Figure 12).
The most stable surfaces for Ru and Pd, i.e. (0001) and (111) respectively, were employed. These calculations reveal that SA adsorbs with a similar strength on both metal surfaces, with values of 1.1 and 0.9 eV determined for Ru and Pd, respectively. Likewise, free SA binds to both surfaces through the carboxylic groups, with negligible levels of electron transfer occurring (< 0.1 eV) in both cases.
However, although calculations indicate SA binds through the same functional group on both surfaces ( Figure 12), the precise geometry of this coordination and the type of binding differs on both metals.
As can be seen (Figure 12, top right), whereas the coordination of SA over Ru appears to be bidentate Ru, in excellent agreement to the DRIFTS studies of pivalic acid. In further agreement to the vibrational spectra, the  value observed in the predicted spectra is also greater for Pd than for Ru, at 251 cm -1 and 208 cm -1 respectively. Although detailed reaction coordinate analysis is required before definitive mechanistic hypotheses can be made, these initial vibrational and computational studies clearly indicate that the choice of metal impacts the coordination and hence, the geometry, of SA at the active site at reaction conditions, and likely accounts for the improved selectivity of Pd versus Ru during SA hydrogenation. Such computational studies, focused upon generating such a detailed molecular level mechanism, represent the focus of our on-going work.

Conclusions.
In this manuscript, we show that succinic acid, a C4 compound increasingly being produced on a several kiloton scale by the microbial fermentation of sugar, can be selectively converted into a variety of important chemicals. Optimal performance in terms of activity, selectivity and reusability is

Succinic acid
Deprotonated succinic acid observed for 2 wt. % Pd supported on Al2O3, which converts succinic acid selectivity to gammabutyrolactone at levels above 95 %. To the best of our knowledge, this represents the first report where succinic acid can be selectively converted to gamma-butyrolactone. Furthermore, the reaction conditions optimized in this study represent the mildest set of conditions for succinic acid valorization reported to date, being > 100 °C lower in temperature, and > 120 bar lower in pressure than previously reported. Complimentary spectroscopic and microscopic studies reveal that optimal activity depends on the choice of support and the size of the supported Pd nanoparticles.
Porosimetry analysis. Porosimetry measurements were performed on a Quantachrome Autosorb, and samples were degassed prior to measure (120 °C, 6 h). N2 adsorption isotherms were obtained at 77 K and surface areas were calculated using the Brunauer-Emmett-Teller (BET) method based on adsorption data in the partial pressure (P/P0) range 0.05 -0.35.

Microscopy analysis.
A transmission electron microscope (TEM) JEM-2100 LaB6 operating at 300 kV was employed to determine how different synthesis methods affected the average particle size and particle distribution on the surface of the catalysts. The catalyst powder was dispersed in ethanol using ultra-sonication and 40 µL of the suspension was dropped on to a holey carbon film supported by a 300 mesh copper TEM grid before the solvent was evaporated prior the analysis. Additionally, a scanning electron microscope (SEM) Hitachi TM3030Plus equipped with a Quantax70 energy-dispersive X-ray spectroscope (EDX) was used at 15 kV and at EDX observation conditions to analyze the elemental composition and uniform distribution throughout all the sample.
X-Ray photoelectron spectroscopy (XPS). XPS spectra were recorded on a Kratos Axis Ultra DLD spectrometer using a monochromatic Al Kα X-ray source. X-ray source (75 -150 W) and analyzer pass energies of 160 eV (for survey scans) or 40 eV (for detailed scans). Subsequently, and in a manner analogous to the pyridine experiments, the system was gradually heated up to evaluate the probe molecule desorption.

Kinetic evaluation and analytical methods.
Liquid-phase hydrogenation of succinic acid (SA). Catalytic evaluation of SA hydrogenation to GBL Reusability studies. Catalyst reusability was carried out after simple filtration of the catalyst. Between measurements the catalyst was dried overnight at room temperature. Reaction conditions and its analytics were performed following the same procedure described above.

Computational experiments.
Periodic plane-wave DFT calculations were performed using the Vienna ab-initio simulation package (VASP), 47-50 the Perdew-Burke-Ernzerhof functional 51 and a kinetic energy of 550 eV to expand the plane-waves of the Kohn-Sham valence states. The inner electrons were represented by the projector-augmented wave (PAW) pseudopotentials considering also non-spherical contributions from the gradient corrections. 52 All the calculations include the long-range dispersion correction approach by Grimme. 53 The optimisation thresholds were 10 −5 eV and 0.01 eV/Å for electronic and ionic relaxation, respectively. The Brillouin zone was sampled by Γ-centred k-point mesh generated through a Monkhorst-Pack grid of 5 × 5 × 1 k-points, which ensures the electronic and ionic convergence. In order to improve the convergence of the Brillouin-zone integrations, the partial occupancies were determined using the first order Methfessel-Paxton method corrections smearing with a set width for all calculations of 0.2 eV. The metal surfaces are simulated by a p(4x4) slab model containing five atomic layers where the two uppermost layers were relaxed without symmetry restrictions and the bottom ones were frozen at the bulk lattice parameter. We added a vacuum width of 15 Å between periodic slabs, big enough to avoid the interaction between periodic images. Isolated molecules were placed in the center of a 15 × 16 × 17 Å 3 cell and optimized with the same criteria.

ASSOCIATED CONTENT
Supporting Information, including additional spectroscopic and kinetic information, is provided. The files are available free of charge.

Author Contributions
The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript.

Funding Sources
Engineering and Physical Sciences Research Council through the Catalysis Doctoral Training center (EP/L016443/1). The Royal Society, for provision of University Research Fellowships (UF140207 and UF160021).