Differentiating the Reactivity of ZrO2-Bound Formates Formed on Cu/ZrO2 during CO2 Hydrogenation

The surface species formed during the hydrogenation of CO2 with H2 over a ZrO2supported Cu catalyst were investigated by operando diffuse reflectance FT-IR spectroscopy at 220 ◦C and 3 bar. The reactivity of two different formates located on zirconia could be unraveled. The data pointed to ZrO2 hydroxyl groups at 3755 cm−1 as the sites on which carbonates and then formates were hydrogenated to methoxy species. Formate hydrogenation appeared as the slowest step. The most reactive ZrO2-bound formates exhibited a rate constant of reaction about 65 times higher than that of the slower formate.


Introduction
The hydrogenation of CO 2 to produce methanol enables recycling CO 2 into a base chemical that can also be directly used as a fuel [1][2][3][4]. Cu/ZnO/Al 2 O 3 is the reference industrial catalyst for methanol synthesis and has been developed primarily for CO-containing feed [3]. A major difference between using CO (Equation (1)) and using CO 2 (Equation (2)) for manufacturing methanol is the formation of water, which can also be obtained through the reverse water gas-shift reaction (RWGS, Equation (3)): CO + 2H 2 →CH 3 OH (1) The increased presence of water in the reactor will impact the long-term stability of Cu/ZnO/Al 2 O 3 catalysts due to the poor hydrothermal stability of alumina. In fact, Prašnikar et al. showed that a correlation existed between alumina surface area and Cu particle sizes during accelerated aging tests [5]. Golunski and Burch [3]. stressed the need to stabilize alumina against densification or find a more durable support. ZrO 2 is a strong candidate to be used as a support for methanol synthesis catalysts because of its well-known hydrothermal stability. The melting point of zirconia is 2715 • C, in comparison to 2072 • C for alumina. ZrO 2 -based catalysts have been successfully used for many demanding reactions such as the dry [6] and steam [7] reforming of methane, steamreforming of methanol [8] (the reverse reaction of that of interest here) and aqueous-phase reactions in acidic media [9].
Unpromoted ZrO 2 present some activity by itself for CO 2 hydrogenation, its combination with ZnO yield even greater interesting activity, although less than those obtained when combined with Cu [10]. In fact, the combination of Cu and ZrO 2 is one of the most efficient to obtain high reaction rates to methanol with CO/CO2/H2 [3,11] or CO2/H2 mixture [12][13][14][15] Cu supported on amorphous zirconia was found to be more active than the catalysts made from monoclinic or tetragonal polymorphs, though ageing was not considered [16].
Numerous IR studies have been reported on the Cu-catalyzed hydrogenation of CO [17,18] and CO2 [19][20][21][22][23] to methanol, with formates often proposed as being a main reaction intermediate. Fischer and Bell investigated the hydrogenation of CO2 over Cu/ZrO2/SiO2 and proposed that Cu favored the activation of H2 that spilled over to ZrO2 to promote carbonate/hydrogenocarbonate hydrogenation into formates and then to methoxide species [21].
The role of formates in the hydrogenation of CO2 over ZrO2-supported Cu has actually been the subject of a controversy. Two distinct pathways have been proposed ( Figure  1). Kattel et al. [22] proposed a formate-free pathway, starting with the reverse WGS (RWGS) and the formation of a CO-OH hydroxycarbonyl intermediate. In contrast, Larmier et al. [24] proposed a formate-based route, in which a ZrO2-bound formate located at the interface with Cu nanoparticles is hydrogenated by H2 activated on Cu to methoxy, still on the support. We have previously investigated formate decomposition over alumina-supported cobalt used for CO hydrogenation [25]. The reactivity of formates was determined by inducing a chemical transient, i.e., by removing CO from the feed. Interestingly, two types of formate reactivity were observed, despite the fact that both formate species exhibited identical DRIFTS spectra. These facts were rationalized by proposing that the reactivity of these alumina-bound formates differed depending on their proximity to the aluminametal interface.
A similar experimental approach has been used here to investigate the hydrogenation of CO2 over a zirconia-supported Cu catalyst. The present work aims at highlighting We have previously investigated formate decomposition over alumina-supported cobalt used for CO hydrogenation [25]. The reactivity of formates was determined by inducing a chemical transient, i.e., by removing CO from the feed. Interestingly, two types of formate reactivity were observed, despite the fact that both formate species exhibited identical DRIFTS spectra. These facts were rationalized by proposing that the reactivity of these alumina-bound formates differed depending on their proximity to the aluminametal interface.
A similar experimental approach has been used here to investigate the hydrogenation of CO 2 over a zirconia-supported Cu catalyst. The present work aims at highlighting different types of formate species present at the surface of the catalyst and their relative reactivity.

Experimental Section
A 6 wt.% Cu/ZrO 2 catalyst was prepared by incipient wetness impregnation of zirconia (from MEL Chemicals, monoclinic, 131 m 2 g −1 ) using Cu(NO 3 ) 2 . 6 H 2 O. The impregnated sample was then dried at room temperature for 24 h before being oven-dried at 110 • C for 12 h. The sample was finally calcined at 450 • C for 12 h in synthetic flowing air. The sample was reduced in situ at 350 • C under flowing 80% H 2 /He for 1 h before the CO 2 hydrogenation experiments.
Copper dispersion was measured using N 2 O titration (frontal chromatography) using a 10 cm-pathlength FT-IR gas cell as detector. The sample was reduced in situ at 300 • C in 50% H 2 /He for 1 h, purge in He and brought to 75 • C to be exposed to 2000 ppm of N 2 O in a flow rate of 36 mL min −1 . It was assumed that each O deposited on the surface titrated two Cu surface atoms [26,27].
Powder X-ray diffraction patterns (XRD) were recorded to assess the crystallinity of the samples. Diffractograms were collected between 10 and 80 • (2Θ) with steps of 0.02 • and 1 s per step with a Bruker D8-Advance diffractometer using CuKα radiation at λ = 1.5418 Å. Nitrogen adsorption isotherms were measured at 77 K on an ASAP 2020 from Micromeritics. Samples were first outgassed under vacuum at 300 • C for 3 h.
High-purity gases He, CO, CO 2 and H 2 from Air Liquid were used for the operando catalytic tests. The experiments were carried out at 3 bars using a mixture of 20% CO 2 + 60% H 2 in He at a total flowrate of 75 mL min −1 , unless otherwise stated. Operando DRIFTS experiments were performed on a modified high-temperature DRIFT cell (from Spectra-Tech, Hong Kong) fitted with CaF 2 windows using a Collector II assembly. A description and properties of the cell can be found in earlier references [28][29][30]. The spectrophotometer used was a Nicolet 8700 (ThermoFischer Scientific, Waltham, MA, USA) fitted with a liquid-N 2 cooled MCT detector. The DRIFT spectra were recorded at a resolution of 4 cm −1 and 8 scans were averaged. The DRIFTS spectra are reported as log (1/R), where R is the sample reflectance. This pseudo-absorbance gives a better linear representation of the band intensity against surface coverage than that given by the Kubelka-Munk function for strongly absorbing media such as those based on metals supported on oxides [31]. The contribution of gas-phase CO was subtracted using a CO(g) spectrum collected under the same experimental conditions over KBr powder [32,33].
The reaction products were analyzed using a transmission FT-IR gas-cell (200 mL dead-volume) with a 2 m pathlength kept at 60 • C to prevent product condensation. The pressure in the system was controlled by a back-pressure regulator located after the DRIFTS cell and before the transmission FT-IR gas cell. The pressure in the line was precisely measured using an electronic gauge.

Results and Discussion
The surface area of the calcined 6 wt.% Cu/ZrO 2 catalyst was 123 m 2 g −1 . The sample exhibited a Cu surface area measured by N 2 O frontal chromatography of about 8.3 m 2 g −1 , corresponding to Cu spherical particles with an average diameter of 4.9 nm. In view of these two surface measurements, it can be concluded that the ZrO 2 surface represented the largest fraction (ca. 93%) of the surface area of the sample.
The diffraction patterns of the ZrO 2 support and Cu/ZrO 2 sample showed the presence of mostly monoclinic zirconia with traces of a tetragonal phase ( Figure 2A). Very weak and broad peaks associated with CuO were also observed in the case of the calcined Cu/ZrO 2 sample, which are consistent with the high dispersion of the Cu phase determined through N 2 O titration.
The in situ spectrum at 220 • C of the Cu/ZrO 2 sample reduced at 350 • C prior to exposure to the feed ( Figure 2B, Red) showed two main broad bands at about 3755 and 3665 cm −1 typical of zirconia surface hydroxyl groups [34]. A group of bands near 1505 cm −1 was also observed, indicated the residual presence of strongly bound carbonates that were not decomposed during the reduction step.
cm −1 typical of zirconia surface hydroxyl groups [34]. A group of bands near 1505 cm −1 was also observed, indicated the residual presence of strongly bound carbonates that were not decomposed during the reduction step. Methanol and CO were the only C-containing products observed at 220 °C and 3 bar when the 20% CO2 + 60% H2 feed was introduced. A steady state was achieved within one hour on stream in terms of methanol and CO concentrations measured at the exit of the DRIFTS cell and in terms of DRIFTS signals of the various adsorbates observed. Differential conditions were obtained with a CO2 conversion of about 0.8%. The selectivity to methanol was about 47%, and the productivity was 1.0 μmol(CH3OH) s −1 gCatalyst −1 . These values are similar to those reported by Fischer and Bell [21] at 250 °C and 6.5 bar for a Cu/ZrO2/SiO2, that were 1.3 μmol(CH3OH) s −1 gCatalyst −1 and a selectivity to methanol of 43%. The operando spectrum obtained at steady state under the CO2 and H2 feed at 220 °C is shown in Figure 3A (top). The set of bands at 2873, 1575 and 1380 cm −1 corresponded to formate species [35,36], mostly, if not wholly, adsorbed on the zirconia support ( Table 1). The band around 1468 cm −1 can be assigned to polydendate carbonates formed from CO2 adsorption on zirconia basic sites [21]. The bands at 2927, 2822, 1150 and 1049 cm −1 were assigned to methoxy groups [34,37]. The methoxy species with the band at 1150 cm −1 (corresponding to the stretching C-O vibration ν(O-C)) was proposed to be formed over zirconia hydroxyl groups located near 3755 cm −1 and be a monodendate species. In contrast, the methoxy with a band at 1049 cm −1 was proposed to derive from the adsorption on the site associated with hydroxyl groups at 3665 cm −1 and be a bridged methoxy [34,37].
Interestingly, no evidence of CO adsorbed on Cu was apparent. To assess whether or not adsorbed CO could ever be observed, an experiment was carried out replacing CO2 with CO. A large band centered around 2010 cm −1 assigned to CO adsorbed on metallic copper [38] was observed when CO was used ( Figure 3A (bottom)). This indicates that the coverage of the Cu nanoparticles is very different in the presence of CO2, as compared to the case of CO. More work would be needed to determine the nature of the main species covering the Cu surface in this case (e.g., H, O). It is interesting to note that, apart from the CO(ads) signal, the rest of the spectra was almost identical, pointing to similar Methanol and CO were the only C-containing products observed at 220 • C and 3 bar when the 20% CO 2 + 60% H 2 feed was introduced. A steady state was achieved within one hour on stream in terms of methanol and CO concentrations measured at the exit of the DRIFTS cell and in terms of DRIFTS signals of the various adsorbates observed. Differential conditions were obtained with a CO 2 conversion of about 0.8%. The selectivity to methanol was about 47%, and the productivity was 1.0 µmol(CH 3 OH) s −1 g Catalyst −1 . These values are similar to those reported by Fischer and Bell [21] at 250 • C and 6.5 bar for a Cu/ZrO 2 /SiO 2 , that were 1.3 µmol(CH 3 OH) s −1 g Catalyst −1 and a selectivity to methanol of 43%. The operando spectrum obtained at steady state under the CO 2 and H 2 feed at 220 • C is shown in Figure 3A (top). The set of bands at 2873, 1575 and 1380 cm −1 corresponded to formate species [35,36], mostly, if not wholly, adsorbed on the zirconia support ( Table 1). The band around 1468 cm −1 can be assigned to polydendate carbonates formed from CO 2 adsorption on zirconia basic sites [21]. The bands at 2927, 2822, 1150 and 1049 cm −1 were assigned to methoxy groups [34,37]. The methoxy species with the band at 1150 cm −1 (corresponding to the stretching C-O vibration ν(O-C)) was proposed to be formed over zirconia hydroxyl groups located near 3755 cm −1 and be a monodendate species. In contrast, the methoxy with a band at 1049 cm −1 was proposed to derive from the adsorption on the site associated with hydroxyl groups at 3665 cm −1 and be a bridged methoxy [34,37].
Interestingly, no evidence of CO adsorbed on Cu was apparent. To assess whether or not adsorbed CO could ever be observed, an experiment was carried out replacing CO 2 with CO. A large band centered around 2010 cm −1 assigned to CO adsorbed on metallic copper [38] was observed when CO was used ( Figure 3A (bottom)). This indicates that the coverage of the Cu nanoparticles is very different in the presence of CO 2 , as compared to the case of CO. More work would be needed to determine the nature of the main species covering the Cu surface in this case (e.g., H, O). It is interesting to note that, apart from the CO(ads) signal, the rest of the spectra was almost identical, pointing to similar coverages of methoxy, formates and carbonates over the ZrO 2 support whether CO or CO 2 were used. coverages of methoxy, formates and carbonates over the ZrO2 support whether CO or CO2 were used.  CO2 was removed from the feed, and the spectrum obtained after 60 min in H2 at 220 °C is shown in Figure 3B and compared to the steady-state spectrum. The carbonate and formate signals decreased significantly, while that of the methoxy at 1150 cm −1 increased. In contrast, the methoxy at 1049 cm −1 remained essentially unchanged. These observations show first that all the surface species were bound rather strongly to the surface and were poorly reactive under H2 at 220 °C, all likely bound to ZrO2 sites. The data also suggest that a fraction of carbonates and formates had been converted into the methoxy at 1150 cm −1 . These results would suggest that zirconia hydroxyl groups at 3755 cm −1 , associated with the 1150 cm −1 methoxy, are the location at which carbonate and formate species are hydrogenated to methoxy. Bensitel et al. [34] had actually reported that only the 3755 cm −1 hydroxyl species reacted with CO2 to form hydrogenocarbonate species, stressing the unique reactivity of this site.
The decay of the signal in the formate region was investigated by integrating over the range 1415.5-1313.1 cm −1 , using a single point baseline located at 1900 cm −1 (see inset in Figure 3B). The signal decay in the first minute was quite complex, consisting of a rapid evolution of overlapping increasing and decreasing contributions, and will be dealt with in a subsequent contribution.
The signal decay in the time range 1-60 min is presented in Figure 4. The total signal could be decomposed into a slow linear contribution and a faster decay ( Figure 4A). A semi-logarithmic plot of the faster signal decay showed a linear behavior, indicating a uniform reactivity ( Figure 4B). The difference DRIFTS spectrum obtained by subtracting  CO 2 was removed from the feed, and the spectrum obtained after 60 min in H 2 at 220 • C is shown in Figure 3B and compared to the steady-state spectrum. The carbonate and formate signals decreased significantly, while that of the methoxy at 1150 cm −1 increased. In contrast, the methoxy at 1049 cm −1 remained essentially unchanged. These observations show first that all the surface species were bound rather strongly to the surface and were poorly reactive under H 2 at 220 • C, all likely bound to ZrO 2 sites. The data also suggest that a fraction of carbonates and formates had been converted into the methoxy at 1150 cm −1 . These results would suggest that zirconia hydroxyl groups at 3755 cm −1 , associated with the 1150 cm −1 methoxy, are the location at which carbonate and formate species are hydrogenated to methoxy. Bensitel et al. [34] had actually reported that only the 3755 cm −1 hydroxyl species reacted with CO 2 to form hydrogenocarbonate species, stressing the unique reactivity of this site.
The decay of the signal in the formate region was investigated by integrating over the range 1415.5-1313.1 cm −1 , using a single point baseline located at 1900 cm −1 (see inset in Figure 3B). The signal decay in the first minute was quite complex, consisting of a rapid evolution of overlapping increasing and decreasing contributions, and will be dealt with in a subsequent contribution.
The signal decay in the time range 1-60 min is presented in Figure 4. The total signal could be decomposed into a slow linear contribution and a faster decay ( Figure 4A). A semilogarithmic plot of the faster signal decay showed a linear behavior, indicating a uniform reactivity ( Figure 4B). The difference DRIFTS spectrum obtained by subtracting the spectra collected after having removed CO 2 for 2 min and 20 min, when most of the fast signal Catalysts 2022, 12, 793 6 of 10 species had gone, is shown in Figure 5. It shows that the lost signal corresponding to this fast-removed species was essentially a formate similar to that prevailing at steady state. The larger band width observed was possibly due to a larger distribution of heterogeneous sites. The slow remaining species left after more than 30 min under H 2 was already discussed above ( Figure 3B) and also corresponded to ZrO 2 -bound formates.
Catalysts 2022, 12, x FOR PEER REVIEW 6 of 10 the spectra collected after having removed CO2 for 2 min and 20 min, when most of the fast signal species had gone, is shown in Figure 5. It shows that the lost signal corresponding to this fast-removed species was essentially a formate similar to that prevailing at steady state. The larger band width observed was possibly due to a larger distribution of heterogeneous sites. The slow remaining species left after more than 30 min under H2 was already discussed above ( Figure 3B) and also corresponded to ZrO2-bound formates.  The rate constant of reaction/decomposition of the two normalized signals of the socalled "slow" and "fast" formates on ZrO2 were approximately equal to 2 × 10 −3 min −1 and 1.3 × 10 −1 min −1 , respectively. The fast formate thus exhibited a ca. 65-fold higher reactivity the spectra collected after having removed CO2 for 2 min and 20 min, when most of the fast signal species had gone, is shown in Figure 5. It shows that the lost signal corresponding to this fast-removed species was essentially a formate similar to that prevailing at steady state. The larger band width observed was possibly due to a larger distribution of heterogeneous sites. The slow remaining species left after more than 30 min under H2 was already discussed above ( Figure 3B) and also corresponded to ZrO2-bound formates.  The rate constant of reaction/decomposition of the two normalized signals of the socalled "slow" and "fast" formates on ZrO2 were approximately equal to 2 × 10 −3 min −1 and 1.3 × 10 −1 min −1 , respectively. The fast formate thus exhibited a ca. 65-fold higher reactivity The rate constant of reaction/decomposition of the two normalized signals of the so-called "slow" and "fast" formates on ZrO 2 were approximately equal to 2 × 10 −3 min −1 and 1.3 × 10 −1 min −1 , respectively. The fast formate thus exhibited a ca. 65-fold higher reactivity than the slow formate. It must be stressed here that the products (adsorbed or gas-phase) to which these formates decomposed into could not be directly measured.
The evolution of the methoxy DRIFTS band area at 1150 cm −1 was plotted as a function of those of the total formate band ( Figure 6A) and fast formate-only band ( Figure 6B) following the removal of CO 2 . These plots show that a strong correlation existed between these quantities, primarily the fast formate band (since the slow formate hardly changed over the duration of the experiment) and the methoxy at 1150 −1 . This quantitative correlation supports the afore-mentioned model that the ZrO 2 hydroxyl groups at 3755 cm −1 , on which the 1150 cm −1 methoxy is formed, were the sites where carbonates and then formates were hydrogenated to methoxy species. It is interesting to note that the 3755 cm −1 hydroxyl group was still totally missing after 60 min exposure to H 2 at 220 • C ( Figure 2B, Black), indicating that those were still involved in bonding some adsorbates (formates or methoxy). than the slow formate. It must be stressed here that the products (adsorbed or gas-phase) to which these formates decomposed into could not be directly measured.
The evolution of the methoxy DRIFTS band area at 1150 cm −1 was plotted as a function of those of the total formate band ( Figure 6A) and fast formate-only band ( Figure 6B) following the removal of CO2. These plots show that a strong correlation existed between these quantities, primarily the fast formate band (since the slow formate hardly changed over the duration of the experiment) and the methoxy at 1150 −1 . This quantitative correlation supports the afore-mentioned model that the ZrO2 hydroxyl groups at 3755 cm −1 , on which the 1150 cm −1 methoxy is formed, were the sites where carbonates and then formates were hydrogenated to methoxy species. It is interesting to note that the 3755 cm −1 hydroxyl group was still totally missing after 60 min exposure to H2 at 220 °C ( Figure 2B, Black), indicating that those were still involved in bonding some adsorbates (formates or methoxy). The findings here are consistent with a reaction model in which carbonates formed from CO2 adsorption on zirconia are readily reduced to formate species. The hydrogenation of formates to methoxy is then kinetically limiting, explaining the large signal of formate species. These steps are all favored by H2 activation over Cu followed by spillover of H onto the support, as proposed earlier by Fisher and Bell [21]. The accumulation of methoxy species may also occur, if methanol readsorption is significant and if the reductive elimination or hydrolysis of methoxy groups is slow. Fisher and Bell have actually observed of a similar sample that methoxy hydrolysis (by water formed in the reaction) was significantly faster than reductive elimination [21]. This may explain the accumulation of methoxy species in conditions under which water production is limited or totally absent, as in the present case under the H2/He stream.
Some of the main findings obtained here are summarized in Figure 7. Formate species are observed under reactions conditions that are mostly ZrO2-bound. The reactivity of ZrO2-bound formates is two-fold, with a fast formate exhibiting a rate constant of decomposition 65 times higher than that of a slower formate species. The fast formate species appears to be hydrogenated into methoxy groups that are associated with ZrO2 hydroxyl groups at 3755 cm −1 . It is not yet clear if the hydrogenation ability if solely related to the nature of the zirconia sites or if the distance to the copper-zirconia interface matters, for instance, as a result of H-spillover. It is possible that the slow formates were located on domains or ZrO2 crystallites on which no Cu nanoparticles were present and thus almost no spillover H would be available. On the contrary, the fast formates could be located on The findings here are consistent with a reaction model in which carbonates formed from CO 2 adsorption on zirconia are readily reduced to formate species. The hydrogenation of formates to methoxy is then kinetically limiting, explaining the large signal of formate species. These steps are all favored by H 2 activation over Cu followed by spillover of H onto the support, as proposed earlier by Fisher and Bell [21]. The accumulation of methoxy species may also occur, if methanol readsorption is significant and if the reductive elimination or hydrolysis of methoxy groups is slow. Fisher and Bell have actually observed of a similar sample that methoxy hydrolysis (by water formed in the reaction) was significantly faster than reductive elimination [21]. This may explain the accumulation of methoxy species in conditions under which water production is limited or totally absent, as in the present case under the H 2 /He stream.
Some of the main findings obtained here are summarized in Figure 7. Formate species are observed under reactions conditions that are mostly ZrO 2 -bound. The reactivity of ZrO 2bound formates is two-fold, with a fast formate exhibiting a rate constant of decomposition 65 times higher than that of a slower formate species. The fast formate species appears to be hydrogenated into methoxy groups that are associated with ZrO 2 hydroxyl groups at 3755 cm −1 . It is not yet clear if the hydrogenation ability if solely related to the nature of the zirconia sites or if the distance to the copper-zirconia interface matters, for instance, as a result of H-spillover. It is possible that the slow formates were located on domains or ZrO 2 crystallites on which no Cu nanoparticles were present and thus almost no spillover H would be available. On the contrary, the fast formates could be located on ZrO 2 crystals on which Cu nanoparticles would be present and spillover H would be readily available, enabling a uniform reactivity of formates on such domains.
ZrO2 crystals on which Cu nanoparticles would be present and spillover H would be readily available, enabling a uniform reactivity of formates on such domains.

Conclusions
This contribution reveals that on a zirconia-supported copper catalyst used for CO2 hydrogenation, two main types of zirconia-bound formates are present. A significant difference in reactivity is observed (ca. 65-fold), partly related to the nature of the hydroxyl group on which these formates are adsorbed. More work is under way to quantitatively relate the rate of formation of methanol to the reaction rate of the formate species present over zirconia.

Conclusions
This contribution reveals that on a zirconia-supported copper catalyst used for CO 2 hydrogenation, two main types of zirconia-bound formates are present. A significant difference in reactivity is observed (ca. 65-fold), partly related to the nature of the hydroxyl group on which these formates are adsorbed. More work is under way to quantitatively relate the rate of formation of methanol to the reaction rate of the formate species present over zirconia.