Applied Catalysis B: Environmental PdIn intermetallic nanoparticles for the Hydrogenation of CO 2 to Methanol

Direct hydrogenation of CO 2 to methanol could offer signiﬁcant environmental beneﬁts, if efﬁcient cata- lysts can be developed. Here, bimetallic Pd-In nanoparticles show good performance as catalysts for this reaction. Unsupported nanoparticles are synthesised by the thermal decomposition of Pd(acetate) 2 and In(acetate) 3 precursors in a high boiling point solvent (squalane), followed by reduction using dilute H 2 gas (210 ◦ C). Adjusting the ratio of the two metallic precursors allow access to 5–10 nm nanoparticles with different phase compositions, including metallic Pd(0), In 2 O 3 and intermetallic PdIn. Liquid phase methanol synthesis experiments (50 bar, 210 ◦ C, H 2 :CO 2 = 3:1) identify the intermetallic PdIn nanoparticles as the most efﬁcient. The catalysts exhibit around 70% higher methanol rates (normalised to the overall molar metal content) compared to the conventional heterogeneous Cu/ZnO/Al 2 O 3 catalyst (900 and 540 (cid:2) mol mmol PdInorCuZnAl − 1 h − 1 , respectively). In addition, the optimum Pd/In catalyst shows an improved methanol selectivity over the whole temperature range studied (190–270 ◦ C), reaching >80% selectivity at 270 ◦ C, compared to only 45% for the reference Cu/ZnO/Al 2 O 3 catalyst. Experiments showed an improvement in stability; the methanol production rate declined by 20% after 120 h run for the opti- mum PdIn-based compared with 30% for the Cu/ZnO/Al 2 O 3 catalyst (after 25 h). The optimum catalyst consists of ∼ 8 nm nanoparticles comprising a surface In-enriched PdIn intermetallic phase as charac- terised by XRD, HR-TEM, STEM-EDX and XPS. Post-catalysis analysis of the optimum catalyst shows that the same PdIn bimetallic phase is retained with only a slight increase in the nanoparticle size. © (http://creativecommons.org/licenses/by/4.0/).


Introduction
In the last decades, anthropogenic carbon dioxide (CO 2 ) emissions have become the focus of attention due to the implications for climate change. In this context, the hydrogenation of CO 2 to methanol, using H 2 produced from renewable energy or off peak loadings, is an attractive potential route to recycling CO 2 . Methanol is an interesting fuel as it can be used directly or blended with conventional fuels, or as a generic liquid-energy carrier. It is also important in the production of chemicals such as formaldehyde, ethylene, propylene, dimethyl ether, acetic acid, and may be applied in fuel cell applications [1][2][3]. Indeed, methanol is one of the new catalyst formulations have been developed and systems based on Pd/Ga 2 O 3 [16][17][18][19], Au/ZnO [20], Pd/ZnO [21], Cu/CeO 2 [22], or In 2 O 3 [11], all show promise. Among them, In 2 O 3 nanoparticles (NPs) supported onto ZrO 2 have recently shown high activity and selectivity, as well as good stability [11]. However, these catalysts operate best at above 270 • C, indeed high temperatures are required to obtain methanol yields comparable to those from the Cu/ZnO/Al 2 O 3 catalysts, but under these conditions the thermodynamic equilibrium disfavours the formation of methanol and thereby reduces the CO 2 conversion [23].
Fisher and Muhler and co-workers reported colloidal mixtures of Cu and ZnO NPs for the liquid-phase synthesis of methanol from syn-gas [24][25][26][27]. We were inspired by this approach and have reported a series of Cu/ZnO colloidal catalysts some of which show promising performance for the liquid phase hydrogenation of CO 2 to methanol [28][29][30]. The catalyst preparations allow production of small NPs that lead to an intimate intermixing of the Cu/ZnO components increasing the quantity of Cu/ZnO interfaces, where the active sites for the methanol synthesis are proposed to be located [31,32]. More generally, colloidal catalysts offer a broad range of possibilities to 'tune' or control properties such as the composition, particle size, solubility and functionality [33]. In addition, the use of a liquid reaction medium may facilitate heat removal and enable the reaction to be performed isothermally, minimising catalyst deactivation commonly attributed to a poor temperature control [5]. Recently, we extended the use of colloidal Cu/ZnO catalysts to bimetallic Pd-Ga NPs for the hydrogenation of CO 2 to methanol [34]. The resulting Pd 2 Ga-based colloids showed an improved activity and stability compared to the reference Cu/ZnO/Al 2 O 3 catalysts. Nevertheless, a significant rWGSR activity (CO production) was observed, at the expense of the methanol selectivity. Interestingly, the novel synthetic methodology used, consisting of the simple thermal decomposition and further reduction of organometallic precursors (Pd(II) acetate and Ga(III) stearate) is relevant to the production of other catalytic systems, as will be demonstrated here.
Bimetallic nanoparticles have recently shown great potential for catalysing reduction and oxidation reaction [35]. Supported intermetallic Pd-In compounds have shown good performances for methanol steam reforming (MSR) [36][37][38][39]; this process is simply the reverse of methanol synthesis from CO 2 (CH 3 OH + H 2 O CO 2 + 3H 2 ), driven under different operating conditions (typically 300-450 • C and atmospheric pressure) [38,40,41]. Remarkably, the Pd/In materials are highly CO 2 selective, showing low activity for methanol decomposition (CH 3 OH CO + 2H 2 ) and rWGSR reactions [39,41]. The catalysts are also known to be more stable and resistant to sintering than the conventional Cu-based systems [41,42]. Despite this excellent performance in the closely related MSR process, intermetallic Pd-In catalysts have not yet been explored for the hydrogenation of CO 2 to methanol. In this work, dispersions of unsupported bimetallic Pd-In NPs are prepared and investigated for the first time in the liquid phase hydrogenation of CO 2 to methanol. The Pd-In catalysts are characterised both before and after reaction, and the catalytic performance correlated with composition and nano-scale structure.

Catalyst preparation and catalytic experiments
Both the catalyst synthesis and the catalytic studies were conducted in a 300 mL continuous flow stirred tank reactor (CSTR, Parr), mechanically stirred at 1500 r.p.m., and with vertical baffles to ensure the homogenous mixing of the gas and liquid phase. The catalysts were prepared by modifying a recently reported method for the synthesis of Pd 2 Ga NPs [34]. In a typical preparation procedure, the required amount of Pd(II) acetate (Pd(OAc) 2 , Acros Organics 47.5% Pd metal basis) and In(III) acetate (In(OAc) 3 , Aldrich 99.99% metal basis) were added to the stainless steel vessel containing squalane (100 mL, 3.18, mmol Pd+In L −1 , specific quantities available in Table S1). Squalane (16 ppm water by Karl Fischer titration, Acros Organics 99%) is reported to be an excellent solvent for the liquid phase methanol synthesis [43]. The air was swept from the reactor containing squalane and the catalyst precursors by a flow of N 2 gas at 750 mL min −1 for 15 min. Then, the reactor was pressurised to 0.5 MPa with N 2 , and held at 210 • C (4.5 • C min −1 ) under a N 2 flow (400 mL min −1 ) for 1 h. Finally, the flow was switched to 200 mL min −1 of 5%H 2 /N 2 (vol%) for 2 h. The resulting catalysts are labelled as PdInx:y, where x:y is the Pd:In molar ratio used. Mixtures obtained before the 5%H 2 /N 2 treatment are labelled as PdInx:yN2.
To conduct the methanol synthesis experiments, the prepared Pd/In catalysts (still in the CSTR) were subjected to 5 , ground from pellet to fine powder, was tested as a reference material. The total amount of metal (Cu + Zn + Al + Mg) was kept the same as in the Pd/In catalysts (3.18 mmol metal .L −1 ). Before the runs were performed, the precursor was activated by using a 5%H 2 /N 2 stream at 0.45 MPa and 240 • C (2 • C min −1 ) for 3 h, according to a standard activation procedure for Cu/ZnO/Al 2 O 3 methanol synthesis catalysts in slurry reactors [44]. Finally, the catalytic experiment was performed under the same conditions described above.
The reaction products and unreacted material were analysed by an on-line gas chromatograph (Bruker 450-GC), equipped with a thermal conductivity detector (TCD), for the quantification of CO, CO 2 and Ar, and a flame ionization detector (FID), for the quantification of MeOH, and other oxygenates or hydrocarbons (if produced). Product condensation in the transfer lines from the reactor to the injection port of the GC was prevented by heating them at 180 • C. The catalytic results are given at the methanol peak rate (normalised per Pd + In or Cu + Zn + Al + Mg molar content), once the pseudo-steady state was reached, typically after ca. 7 h time-onstream. Selectivities remained nearly constant throughout each experiment. All the experiments were conducted under differential reaction conditions, with CO 2 conversions always below 3%. Under the reaction conditions employed, the maximum CO 2 conversions calculated at the thermodynamic limit (Aspen plus v 8.8 software), considering the methanol synthesis and the side rWGSR, were 36.9, 31.8, 26.4 and 24.3% at 190, 210, 240 and 270 • C, respectively. All catalytic experiments were, therefore, performed in the kineticallycontrolled regime in this study. Selectivities are given on a carbon basis, and the standard deviations determined by repeated runs was ±14% and ±2% for the methanol activity and selectivity, respectively.
Turnover frequencies (TOFs), as mol CH3OH m −2 s −1 , were estimated from the average size of the Pd(0), In 2 O 3 or PdIn NPs determined by TEM, and assuming a spherical shape of the NPs. For the reference Cu/ZnO/Al 2 O 3 catalyst, the methanol rate is generally correlated to the initial Cu(0) surface area [6,42], and therefore, the TOF was calculated from the average Cu(0) particle size after activation determined by XRD (6.3 nm, Fig. S1). In this case, Scherrer analysis of the XRD data was used owing to agglomeration of the heterogeneous Cu/ZnO/Al 2 O 3 catalyst hampering comprehensive TEM analysis.

Characterisation techniques
Thermogravimetric analysis (TGA) was performed in a Mettler Toledo TGA/DSC 1, coupled with an integrated Hiden HPR-20 QIC EGA mass spectrometer (MS), under a flow of N 2 from 100 to 700 • C, using a heating rate of 10 • C min −1 . Additional experiments simulated the temperature profile during the synthesis of the Pd/In catalysts, by heating from room temperature to 210 • C (4.5 • C min −1 ), and then maintaining this temperature for 3 h under N 2 flow.
Powder X-ray diffraction (XRD) was carried out using an X'pert Pro MPD diffractometer (PANalytical B. V) operating at 40 kV and 40 mA, using nickel-filtered Cu K ␣ radiation ( = 0.1542 nm). The samples were covered by a polyimide film tape (Agar Scientific), so as to prevent contact with air during the measurements. The average crystallite size of Pd(0), PdIn, In 2 O 3 and Cu(0) particles were determined by applying the Scherrer equation (shape factor of 0.9) to the most intense and not overlapped reflections at 40. Transmission electron microscopy (TEM) samples were prepared by drop-casting a toluene solution onto a 300 mesh Au holey carbon grid with an ultrathin 3 nm carbon film (Agar Scientific). Annular dark field scanning TEM (ADF-STEM) and energy dispersive X-ray (EDX) mapping were performed with a JEOL JEM 2100F scanning transmission electron microscope operated at 200 kV, and equipped with an Oxford X-Max 80 SDD EDX detector. High-resolution TEM (HR-TEM) images were obtained with an aberration-corrected FEI Titan operated at 300 kV. For HR-TEM imaging, samples were loaded and sealed into a Gatan Environmental Holder, under a nitrogen atmosphere in a glove box, preventing any atmospheric exposure that might lead to inadvertent sample oxidation during transfer to the microscope. The average NP sizes were determined using HR-TEM and ADF-STEM by measuring approximately 200 NPs from the images of each sample. The Pd:In composition was determined by EDX performing 13-18 measurements over large, independent regions of the samples (Table S2). The method used for phase identification, as well as (hkl) reflections and the corresponding lattice spacings used are described in Table S2.
The surface of the samples was characterised using X-ray photoelectron spectroscopy (XPS). The spectra were recorded on a Thermo Scientific K-Alpha + X-ray photoelectron spectrometer system operating at 2 × 10 −9 mbar base pressure. This system incorporates a monochromated, microfocused Al K␣ X-ray source (h = 1486.6 eV) and a 180 • double focusing hemispherical analyser with a 2D detector. The X-ray source was operated with a 6 mA emission current and 12 kV anode bias and a flood gun was used to minimize sample charging. Samples were mounted using conductive carbon tape and transferred to the spectrometer using a special glove box module which ensured that samples were never exposed to air. Data were collected at 20 eV pass energy for core level (Pd 3d, In 3d, and In 4d), for In MNN Auger, and for valence band spectra using an X-ray spot size of 400 m 2 . All data were analysed using the Avantage software package. For the calculation of the inelastic electron mean free path (IMFP) following the Tanuma, Powell and Penn (TPP-2 M) method, the QUASES software package was used [45].
With the aim of preventing any air exposure, Pd/In pre-and post-catalysis samples analysed by XRD, XPS, and TEM were prepared using air-sensitive techniques. As such, the samples were withdrawn from the reactor to a Schlenk tube, under a N 2 atmosphere, via a cannula and a high flow rate of N 2 (400 mL min −1 ), washed several times with dry toluene (distilled from sodium and degassed by performing at least three free-pump-thaw cycles, O 2 < 5 ppm), and subsequently dried under vacuum (10 −2 Pa).

Formation of the Pd/In catalysts
In the first synthetic step, a mixture of the Pd(OAc) 2 and In(OAc) 3 (Pd:In molar ratio of 1:1) in squalane was treated at 210 • C and 0.5 MPa N 2 , for 1 h. The TGA-MS data indicate that, after 1 h at this temperature, the thermal decomposition of Pd(OAc) 2 is complete, whereas In(OAc) 3 is only partially decomposed (Fig. S2). The degradation of the precursors mostly leads to the formation of acetic acid (boiling point 180 • C at 0.5 MPa), acetone (boiling point 113 • C at 0.5 MPa) and CO 2 ( Fig. S3-S4), which are expected to be released from the reaction medium under the N 2 flow (400 mL min −1 ). XRD of the resulting black suspension, sample PdIn1:1N 2 , revealed only the presence of Pd(0) crystallites, with a cubic F structure and an average diameter of 3.8 nm, whereas no diffraction peaks for Incontaning phases are observed ( Fig. 1, Table 1). Interestingly, the PdIn1:1N 2 peaks are shifted (∼0.9 • ) to lower 2 angles compared to the ICDD 087-0639 reference pattern, which can be attributed to an expansion in the Pd-Pd interatomic distance for Pd(0) crystallites of such a small size [46]. TEM analysis also confirmed the presence of Pd(0) NPs with an average size of 5.4 ± 0.5 (1.2) nm ( Fig.  S5), although in this case, a few In 2 O 3 NPs ( < 5 nm) were also identified (Fig. S6). The low concentration of In 2 O 3 NPs along with their expected small crystallite size is likely responsible for the absence of any observable In 2 O 3 diffraction peaks in the corresponding XRD. The formation of separate metallic Pd(0) and In 2 O 3 phases from the thermal decomposition of Pd(OAc) 2 and In(OAc) 3 , under a N 2 atmosphere, is consistent with previous publications [47,48].
Next, the mixture was subjected to a dilute H 2 flow, at 210 • C for 2 h, and the XRD pattern of the resulting suspension (sample PdIn1:1) showed only diffraction peaks attributed to the forma- In2O3 a Pd:In molar ratio determined from EDX analysis. The standard error of the mean was defined as standard deviation/sqrt (n • measures), and does not account for systematic errors in the k-factors used in standardless quantification which can be ∼10%.
b The standard error of the mean was defined as standard deviation/sqrt (n • measures). c Size values for the primary crystalline phase. tion of a PdIn crystalline phase, with a cubic CsCl structure (Fig. 1). The increased sharpness of the reflexions indicates an increase in the average crystallite size, which was determined to be from 3.8 to 8.2 nm. Assuming a 33% radius expansion upon In incorporation (based on the metals relative densities), it is clear that some ripening must also take place. HR-TEM (Fig. 2) identified spherical NPs with an average size of 10.1 ± 0.8 (2.8) nm (Table 1). Lattice spacing analysis of the NPs (both in Fig. 2), and the Pd:In composition determined by EDX (Table 1) were also consistent with the formation of the PdIn intermetallic alloy. Interestingly, a con-trol experiment using the same synthesis protocol but including only In(OAc) 3 precursor (sample PdIn0:1), did not show any In(0)containing phases; rather, the In(III) species were only reduced in presence of Pd(0) under the relevant conditions. Analogous observations have been reported for Pd/Ga systems [49], where it is proposed that Pd(0), either through hydrogen dissociation or by forming active hydrides [50,51], mediates the reduction of Ga(III) species which subsequently diffuse into the Pd(0) core to form Pd-Ga alloy.

Influence of the Pd/In composition
A series of samples were prepared varying the Pd:In composition. As determined by EDX, the average Pd:In molar ratios for the samples, after the two preparation steps, were consistent with the expected values assuming the conversion of the Pd and In precursors into the In-and Pd-containing phases. During the first step of the synthesis under N 2 , the In(OAc) 3 precursor decomposes at a slow rate (see TGA, Fig. S2) and is only completely decomposed after the second preparation step (H 2 /N 2 treatment). In agreement with this hypothesis, the PdIn1:1N 2 sample, submitted only to the first step, contained a higher than expected Pd:In ratio (1.35) presumably due to the removal of some unreacted In(OAc) 3 during work-up. As observed by XRD (Fig. 1), the use of the Pd(OAc) 2 alone (PdIn1:0) produced a metallic Pd(0) phase with an average crystallite size of 5.1 nm ( Table 1). The pure In(OAc) 3 yielded a product (PdIn0:1) consisting of In 2 O 3 crystallites of 5.6 nm. The presence of small crystalline In 2 O 3 NPs in this PdIn0:1 sample was also identified by HR-TEM (Fig. S7). XRD patterns for the samples containing both Pd and In mainly showed diffraction peaks attributed to the formation of the PdIn intermetallic phase, revealing larger crystallites with increasing In content, in accordance with TEM. As expected, the sample with excess In (PdIn1:2), shows also small diffraction peaks corresponding to In 2 O 3 , in addition to the PdIn intermetallic phase. Strikingly, the sample with an excess of Pd (PdIn2:1) exhibited only the PdIn phase, as also observed by HR-TEM (Fig. S8). Interestingly, in this case, many of observed NPs consisted of several PdIn crystalline domains (Fig. S8), which explains the larger discrepancy between the particle sizes ascertained by XRD and HR-TEM, for this PdIn2:1 sample, 4.2 and 7.8 ± 0.6 (2.4) nm, respectively. Previous work identified the formation of the CsCl-structured PdIn intermetallic, in preference to other possible phases across a range of stoichiometries [52]; in that case, excess Pd was proposed to form a core within the PdIn shell structures under Pd rich conditions. However, in the current system there is no evidence of a significant Pd(0) component in XRD, HR-TEM or XPS characterisation. Rather, XPS analysis for the PdIn 2:1 sample is more consistent with the formation of an In-deficient form of the PdIn intermetallic (vide infra).

XPS characterisation
The nature of the Pd and In species after submitting the precursors to the thermal treatment, sample PdIn1:1N 2 , was studied by XPS analysis. The Pd 3d core level is dominated by the characteristic asymmetric line shape and binding energy position (BE = 335.3 eV) of Pd(0) (Fig. 3(a)), in agreement with reports in the literature [53][54][55]. In addition, the valence band ( Fig. 3(b) and (c)) shows a high density of states at the Fermi energy E F , typical for Pd(0), which stems from Pd 4d states [51,55]. The In core levels ( Fig. 3(d) and (e)) are dominated by higher BE components at 444.3 eV (In 3d 5/2 ) and 18.3 eV (In 4d 5/2 ), which indicate the presence of In(III) in In 2 O 3 [53,55,56]. Second, lower BE components are assigned to In alloy environments (443.3 eV in In 3d 5/2 and 16.4 eV in In 4d 5/2 ). The In M 4,5 N 4,5 N 4,5 Auger lines further confirm the observation of a mixed In(0)/In(III) oxidation state (Fig. 3(f)). The overall line shapes and peak positions of Pd and In 3d for the sample PdIn1:1N 2 point towards a mixed system dominated by metallic Pd(0) and In 2 O 3 , which is in accordance with previous XRD and HR-TEM characterisation. However, in this case, a small contribution from PdIn intermetallic alloy phase was observed. Peak fit analysis of the In and Pd 3d 5/2 core levels indicates a clear surface enrichment of In, with an overall Pd:In molar ratio of 27:73 compared to 57:43 found from EDX analysis, mainly suggesting that the In 2 O 3 species are on the surface of the Pd(0) NPs (Fig. 4, Tables 1 and S3).
Having established the spectra of the Pd(0) and In 2 O 3 phases, the full range of samples reduced in H 2 (PdIn1:0, PdIn2:1, PdIn1:1, PdIn1:2, PdIn0:1) was also investigated using XPS (Fig. 3). A clear transition from Pd(0) to the Pd-In intermetallic phase can be observed in the Pd 3d core level. For the PdIn1:0 sample (not containing In) the core line shows the characteristic asymmetric line shape of Pd(0), which is directly related to a large local density of states at the Fermi energy E F caused by Pd 4d states. The peak maximum is at a BE of 335.3 eV. Upon addition of In, the asymmetry is lost, as the Pd 4d band shifts away from E F , and a continuous shift to higher BE is observed. In addition, a narrowing of the Pd 3d peak is observed upon alloying consistent with previous reports [55].  [57,58] Auger parameters ␣ after Gaarenstrom and Winograd [59] from the In 3d 5/2 core level and the M 5 N 4,5 N 4,5 Auger level were determined to be 846.0 ± 0.2 eV for the intermetallic phase and 844.0 ± 0.2 eV for the oxide phase.
Valence band (VB) spectra ( Fig. 3(b) and (c)) clearly reflect the change in surface chemistry of the catalyst samples following the transition from metallic Pd in PdIn1:0 to In 2 O 3 in PdIn0:1. The pure Pd sample has a large density of states at the Fermi energy E F , which is lost with increasing In content. Already in the PdIn2:1 sample, there is a clear shift of the VB maximum away from E F . The overall shape of the valence band of the PdIn1:1 sample is often described as Cu-like as the states above 1.5 eV appear very similar to those of Cu [54,56,60]. Three distinct regions can be identified within the valence band of the Pd-In intermetallic alloy and by comparison with theoretical calculations for the density of states (DOS) [60,61] their orbital character can be identified. Above 5 eV, low intensity states stemming from In 5 s and Pd 4d are observed. The large contribution to the DOS between 5 and 2 eV is dominated by Pd 4d with some In 5p (and a very small contribution from Pd s states). Finally, the intensity from 2 eV up to the Fermi energy E F is again caused by a mixture of Pd 4d and In 5p states. The non-zero intensity at E F leads to a metal-like behaviour of the In-Pd alloy. This change from Pd-like to Cu-like valence states has been reported previously upon transition from Pd-rich to In-rich intermetallic phases [54][55][56]. Finally, an excess of In in the PdIn1:2 sample leads to a strong contribution from In 2 O 3 states to the VB above 4 eV, which can be compared to the pure In sample (PdIn0:1) [62,63].
From peak fits to the Pd and In 3d 5/2 core lines, overall atomic ratios of Pd:In could be determined (Table S1). In all samples, a surface enrichment of In is observed compared to the ratios determined by EDX analysis (Table 1). In order to study the In speciation of this enrichment, relative ratios of the In(0) and In(III) components for the In 3d 5/2 and 4d 5/2 core lines were determined. Due to the difference in the kinetic energy E K of the photoelectrons, with E K in the order of 400 eV, a variation in the In(0):In(III) ratio was found. The surface sensitivity of the two signals is different, with In 3d being more surface sensitive than In 4d. Using the Tanuma, Powell and Penn (TPP-2 M) method, the inelastic electron mean free path (IMFP) was calculated, which is a good measure for the depth sensitivity of certain core levels [45]. As a close approximation the IMFP values were correlated for In 2 O 3 at average binding energies of the In 3d 5/2 and 4d 5/2 core levels of 445 eV and 18 eV, respectively. Calculations based on these values result in a 30% higher IMFP for In 4d 5/2 , with the exact values being 18.54 Å for In 3d 5/2 and 24.11 Å for In 4d 5/2 . As the intensity of the In 2 O 3 component is comparatively larger in the In 3d than in the 4d core level, it can be concluded that the surface of the PdIn intermetallic NPs is enriched with In 2 O 3 . Analogous Ga 2 O 3 enrichment was found for Pd 2 Ga NPs prepared by a similar method [34]. The PdIn1:1 sample is the only mixed Pd/In sample dominated by the intermetallic alloy contributions for both core levels.

Influence of the Pd/In composition
The different Pd/In-based catalysts were tested in the liquid phase hydrogenation of CO 2 to methanol at 210 • C and 5.0 MPa. Typically, after 7 h of stabilisation, the catalyst showed a very low loss in activity, and nearly constant selectivity with TOS. The only co-product detected was CO, derived from the competitive rWGSR. As summarised in Fig. 5, the PdIn1:0 catalyst, consisting of Pd(0) NPs, is barely active for the methanol synthesis, whereas the catalyst based on In 2 O 3 NPs (PdIn0:1) shows a modest activity. Nevertheless, the combination of Pd and In leads to a dramatic enhancement in the methanol rate, reaching a maximum for the Pd:In molar ratio of 1:1 (PdIn1:1). Significantly, the optimum methanol rate is about 67% higher than that for the benchmark Cu/ZnO/Al 2 O 3 (900 and 540 mol mmol PdInorCuZnAl −1 h −1 , respectively), when normalising for the total molar metal content (Pd + In or Cu + Zn + Al + Mg). Since this most active catalyst consists of PdIn NPs (see section 3.1), it is clear that this bimetallic phase is particularly suited for the hydrogenation of CO 2 to methanol. In order to normalise the methanol synthesis activity in terms of surface area, the TOF, as mol MeOH m −2 s −1 , were estimated (Fig. 5) considering the NP size of the main phase ( Table 1). The TOFs of the Pd/In samples clearly differed according to the Pd:In composition, leading to a maximum for the catalysts PdIn1:1 and PdIn1:2. The similar TOF (within the error) suggest that once there is enough In to form the PdIn alloy, any further addition of In, which leads to the formation of In 2 O 3 NPs, has no substantial impact on the methanol synthesis activity. In fact, In 2 O 3 NPs (sample PdIn0:1) exhibited a negligible TOF under the reaction conditions studied. The optimum TOFs are quite similar to that obtained for the reference Cu/ZnO/Al 2 O 3 catalyst, normalised to the Cu(0) surface area [6,42].
Currently, there are no reports investigating the nature of the methanol synthesis active sites upon formation of the PdIn intermetallic phase. Nevertheless, simultaneous changes in the electronic and structural properties of the materials can be assumed. The higher activity of the Pd-In intermetallic phases respect to the corresponding isolated pure metallic or oxide phases (Pd(0) or In 2 O 3 ) has also been reported for the methanol steam reforming process [38,41]. Some studies, using both unsupported PdIn intermetallic catalysts, as well as PdIn supported on In 2 O 3 , have attributed the enhancement in methanol steam reforming activity to a bimetal-oxide (PdIn-In 2 O 3 ) synergy [37,56,64]. Interestingly, a synergistic mechanism was also reported for bimetallic PdGa NPs supported on Ga 2 O 3 in the hydrogenation of CO 2 to methanol, where the role of the bimetallic PdGa NPs is to provide atomic hydrogen via spillover to the reaction intermediates adsorbed on Ga 2 O 3 [17]. It is also worth noting that oxygen vacancies in In 2 O 3 , whose concentration can be modulated by co-feeding CO or using supports like zirconia, have been recently reported to be active sites for the methanol synthesis from CO 2 [11]. As observed by XPS analysis (section 3.3), the surface of all the Pd/In-containing catalysts was enriched in In 2 O 3 species. Although the current study does not rigorously exclude a possible promotion of the methanol synthesis activity by the In 2 O 3 species, there was no relationship between the surface concentration of In 2 O 3 species and the intrinsic activity. Catalysts with PdIn1:1, PdIn1:2 and PdIn0:1 exhibited an excellent methanol selectivity, with values always above 90% (Fig. S9). No CO was detected for the catalyst based on In 2 O 3 NPs (PdIn0:1), either, in agreement with a recent publication [11]. However, the very low activity of this PdIn0:1 catalyst means that any CO production may be close to the GC detection limit. Interestingly, the sample PdIn2:1, with an In-deficient PdIn intermetallic alloy, revealed a dramatic drop in methanol selectivity (51% MeOH) in favour of CO selectivity (49%). Some authors have attributed the CO production to the rWGSR activity of unalloyed Pd(0) in catalysts for the methanol synthesis or methanol steam reforming [38,65]. However, in this case, the catalyst based on Pd(0) NPs (PdIn1:0) was not active for the rWGSR (no CO production) under the reaction conditions employed here, which is in agreement with some other studies [34,66]. The origin of the higher CO production observed for PdIn2:1 is still not clear, however, it can be tentatively attributed to the different electronic or structural properties of the In-deficient PdIn intermetallic phase, as revealed the characterisation results (section 3.2). Some authors have proposed that changes in selectivity upon PdIn intermetallic are due to the diluting effect of the In which ensures that no adjacent Pd atoms are present on the surface, leading to the formation of new active sites [52]; further studies are expected to shed light on this aspect.

Influence of the operation conditions and stability
The performance of the optimum PdIn1:1 catalyst for the CO 2 hydrogenation to methanol reaction was also compared to the benchmark Cu/ZnO/Al 2 O 3 catalyst, across the temperature range 190-270 • C (Fig. 6a). The methanol rate for the PdIn1:1 catalyst reaches a maximum at 240 • C. The subsequent drop in rate at higher reaction temperatures can be mostly explained by the ripening of the PdIn NPs. In fact, XRD analysis of the catalyst NPs after the reaction at 270 • C indicated an average PdIn crystallite diameter of 16.6 nm (Fig. S10), around 105% higher than the original catalyst. Apart from this run at 270 • C, the PdIn1:1 catalyst showed methanol rates 50-100% higher than those for the reference Cu/ZnO/Al 2 O 3 material at a given reaction temperature. Remarkably, the increase in rate is 110% at 190 • C, which highlights the potential of the PdIn intermetallic phase to carry out the methanol synthesis process from CO 2 at lower temperatures. Both PdIn1:1 and Cu/ZnO/Al 2 O 3 catalysts experienced a drop in methanol selectivity with increasing reaction temperature, as expected due to the more favourable rWGSR (vide infra). Only the runs at 270 • C produced any alkanes, in the form of a small amount of CH 4 and short chain hydrocarbons (C 2 C 4 ), with a total selectivity lower than 5%. Interestingly, the methanol selectivity for the PdIn1:1 catalyst is significantly higher than that for the reference Cu/ZnO/Al 2 O 3 catalyst across the entire temperature range studied. In fact, the methanol selectivity for the Pd/In catalysts only dropped from 96 (190 • C) to 83% (270 • C), whereas the methanol selectivity for the Cu/ZnO/Al 2 O 3 system decreased from 94 (190 • C) to 44% (270 • C).
From Arrhenius plots (Fig. S11), the apparent activation energies for the methanol synthesis and rWGSR were calculated from the MeOH and CO production rates. Apparent activation energies of 41 and 117 kJ mol −1 were obtained for methanol and CO formation on Cu/ZnO/Al 2 O 3 , respectively, agreeing well with reported values [19,67]. In comparison, the PdIn1:1 catalyst exhibited a similar apparent activation energy for the methanol synthesis (35 kJ mol −1 ), but a substantially lower value for the rWGSR (76 kJ mol −1 ). The lower activation energy for the rWGSR would indicate a lower methanol selectivity for the PdIn1:1 catalyst compared to Cu/ZnO/Al 2 O 3 unless the occurrence of the corresponding active sites is much lower. Interestingly, the lower activation energy for the CO formation indicates that the methanol selectivity is relatively less sensitive to changes in temperature than for the conventional catalyst. In contrast to the proposed similarities in the reaction pathway to form methanol for both catalytic systems, the differing apparent activation energies for the rWGSR suggests mechanistic differences. Analogous differences in the apparent activation energies for the rWGSR were found for bimetallic Pd-Ga catalysts compared to the conventional Cu/ZnO/Al 2 O 3 when used in the hydrogenation of CO 2 to methanol [19].
In the case of Cu/ZnO/Al 2 O 3 catalysts used in the hydrogenation of CO 2 to methanol, decreased space velocity is known to reduce the methanol rate and selectivity [14,19,20]. This phenomenon is typically attributed to a product inhibition effect (mainly water) [14,19,20]. To explore this aspect, the optimum PdIn1:1 and the reference Cu/ZnO/Al 2 O 3 catalysts were studied at 210 • C and 5.0 MPa varying the inlet flow (Fig. 6b). Upon decreasing the flow rate, and thus, increasing the concentration of the reaction products in the medium, the methanol rate drastically decreases for both catalytic systems. Nevertheless, the methanol rate of the PdIn1:1 catalyst only appears to be affected when using space velocities lower than 12 (Fig. S12). The Cu/ZnO/Al 2 O 3 catalyst deactivated by ∼30% over this period, a well-known effect which is typically attributed to the sintering of the Cu and ZnO phases [11,68]. In contrast, the methanol rate of the PdIn1:1 sample dropped by only ∼15%, indicating a substantial improvement in stability. The metal ratio of the PdIn1:1 post-catalysis material determined by EDX analysis was 0.90 ± 0.01(0.05), compared to 0.99 ± 0.02(0.06) for the fresh catalyst; thus, within error, there is no change in average Pd:In composition during the catalytic run, although low levels of leaching cannot be completely excluded. XRD characterisation of the PdIn1:1 post-catalysis material revealed only the presence of the PdIn intermetallic crystallite phase (Fig. S13), in accordance with the diffraction patterns obtained from the HR-TEM images (Fig. S14). The results indicate that the crystalline structure of the sample is preserved during the reaction. Nonetheless, both techniques showed a slight increase in the crystallite/NP size from 8.2 to 11.6 nm by XRD, and from 10.1 ± 0.8 to 11.1 ± 0.8 nm by TEM. Therefore, the loss in the methanol synthesis rate can be mostly attributed to the reduction in the effective surface area related to ripening of the NPs. Finally, with the aim of studying the long term stability of the optimum PdIn1:1 catalyst, the reaction was studied for 120 h on stream. The methanol rate decreased by ∼20% for the first 30 h and then remained at a constant value (Fig. S15). These results confirmed an improvement in stability compared to Cu/ZnO/Al 2 O 3 catalyst even after 120 h. At the end of the extended reaction, the PdIn remained the only phase identified by XRD, although the crystallites were ripened to an average of 14.4 nm (Fig. S13); most of this coarsening occurred during the first 25 h time on stream. In order to reduce the ripening of the nanoparticles in the reaction liquid medium, ligands may be introduced to provide steric stabilisation in the future.

Conclusions
Dispersions of unsupported Pd-In bimetallic NPs have been prepared, simply and effectively, using a two-step procedure. In the first step, Pd(OAc) 2 and In(OAc) 3 (1:1 molar ratio) were thermally treated at 210 • C under N 2 flow, leading to the formation of Pd(0) NPs of ca. 4 nm. In the second step, the resulting mixture was reduced under a diluted H 2 flow at 210 • C, yielding Pd-In intermetallic alloy NPs of ca. 8 nm, attributed to the Pd-mediated reduction of the In(III) species and their subsequent diffusion into the core of the Pd(0) NPs. XRD and HR-TEM consistently identified the PdIn intermetallic phase. Exploiting the surface sensitivity of XPS analysis, In 2 O 3 species were identified on the surface of the resulting intermetallic NPs. Various control NP catalysts were prepared using different Pd/In ratios of the precursors, including the pure Pd(0) and In 2 O 3 systems.
Catalytic experiments using dispersions of the unsupported Pd/In-based NPs with different compositions were tested in the liquid phase hydrogenation of CO 2 to methanol (H 2 :CO 2 of 3:1 and 50 bar). Pure Pd(0) or In 2 O 3 NPs were hardly active, whereas the catalyst containing PdIn intermetallic NPs showed a dramatic increase in the methanol synthesis activity. Interestingly, among Pd/In catalysts, the use of excess Pd to form the PdIn phase (Pd:In molar ratio of 2:1) reduced methanol synthesis activity and selectivity (50%), whereas the catalyst consisting of PdIn intermetallic NPs (Pd:In = 1:1) showed the maximum methanol rate, which is ∼70% higher than that for the conventional Cu/ZnO/Al 2 O 3 catalyst, based on overall metal molar content (Pd + In or Cu + Zn + Al), and >90% methanol selectivity. This optimum catalyst was studied versus the reference Cu/ZnO/Al 2 O 3 catalyst under different reaction conditions. Runs at 190-270 • C showed a substantial improvement in the methanol synthesis activity and selectivity (low CO production) compared to the conventional catalyst. In fact, the methanol selectivity was higher than 80% at the highest studied temperature (270 • C), whereas the conventional catalyst gave values as low as 45%. Further, the optimum Pd/In catalyst exhibited somewhat higher stability than the Cu/ZnO/Al 2 O 3 catalyst for 120 h time-on-stream. The PdIn intermetallic phase was retained in the post-catalysis sample, although with a slight increase in the NP size. Finally, a space velocity study indicated that the reaction is product inhibited for both catalytic systems (Pd/In and Cu/ZnO/Al 2 O 3 ).
Considering the continued interest in using CO 2 as a feedstock, and the strong performance (activity, selectivity and stability) of the unsupported PdIn intermetallic NPs, there is potential to develop further catalysts. Further studies of catalytic active sites and pos-sible mechanisms are expected to guide improvements to the catalytic systems for methanol synthesis.