Engineering nanoscale H supply chain to accelerate methanol synthesis on ZnZrO_x

Abstract

Metal promotion is the most widely adopted strategy for enhancing the hydrogenation functionality of an oxide catalyst. Typically, metal nanoparticles or dopants are located directly on the catalyst surface to create interfacial synergy with active sites on the oxide, but the enhancement effect may be compromised by insufficient hydrogen delivery to these sites. Here, we introduce a strategy to promote a ZnZrO_x methanol synthesis catalyst by incorporating hydrogen activation and delivery functions through optimized integration of ZnZrO_x and Pd supported on carbon nanotube (Pd/CNT). The CNT in the Pd/CNT + ZnZrO_x system delivers hydrogen activated on Pd to a broad area on the ZnZrO_x surface, with an enhancement factor of 10 compared to the conventional Pd-promoted ZnZrO_x catalyst, which only transfers hydrogen to Pd-adjacent sites. In CO₂ hydrogenation to methanol, Pd/CNT + ZnZrO_x exhibits drastically boosted activity—the highest among reported ZnZrO_x-based catalysts—and excellent stability over 600 h on stream test, showing potential for practical implementation.

Introduction

Aiming to mitigate anthropogenic CO₂ and utilize it as a valuable product, CO₂ hydrogenation to methanol has received extensive interest from academia and industry^1,2,3,4,5. Carbon Recycling International established the world’s first large-scale CO₂-to-methanol plant using Cu/ZnO/Al₂O₃, a classical catalyst originally developed and optimized for syngas-to-methanol conversion⁶. Numerous efforts are still being undertaken to elucidate the active site and optimize catalyst activity^7,8,9,10. Nevertheless, one of the critical limitations of Cu/ZnO/Al₂O₃ is deactivation due to the unsatisfactory stability of the catalyst under exposure to heat and moisture, hampering long-term process operation^11,12,13,14. Consequently, several Cu-free metal oxides have been reported as new classes of catalysts for CO₂ hydrogenation to methanol^{15,16,17,18,19}, among which, ZnO-ZrO₂ solid solution catalyst (ZnZrO_x) reported by Li and co-workers¹⁸ is undergoing pilot-scale testing²⁰. One distinctive feature of ZnZrO_x is its superior long-term stability; however, a downside is its lower activity compared to Cu/ZnO/Al₂O₃. Achieving both high activity and long-term stability is a challenging mission for developing next-generation CO₂ hydrogenation catalysts.

Since H₂ activation over metal oxide catalysts is often kinetically limited^16,21,22, the metal promotion has been widely explored for improving methanol formation activity. In particular, for ZnZrO_x, various metals such as Cu²³, Pt²³, Ga²⁴, and Pd^22,23,25 have been reported as effective metal promoters. It is a general belief that the metal promoter and oxide support should be placed close to each other to maximize interfacial synergy (including strong metal-support interaction), and thus, various synthetic techniques were developed to anchor metal promotors directly on the oxide catalyst surface^{26,27,28,29,30,31,32,33,34,35,36,37,38}. Previously, we studied Pd-promoted ZnZrO_x catalyst (Pd/ZnZrO_x) for methanol formation²², revealing that the major role of Pd is to activate and split H₂ into H atoms that are subsequently transferred to adjacent Zn sites for methanol synthesis. Thus, the rate of H supply to the active sites was identified as a key factor governing the rate of CO₂ reduction to methanol. However, at present, there are limited ways to enhance the interfacial area between the Pd promoter and ZnZrO_x catalyst for traditional metal/oxide nanocomposite catalysts.

Herein, we address this challenge by combining ZnZrO_x with Pd supported on carbon nanotubes (CNT): Pd/CNT + ZnZrO_x. CNT is a well-known material for hydrogen storage^39,40,41, and doping CNT with metal- or metal oxide- greatly enhances its H storage capacity, hinting at swift hydrogen transfer between CNT and the dopants^{40,41,42,43,44}. Moreover, Crossley and co-workers conducted an elegant study of the nature of active sites on Pd/TiO₂ and Cu/TiO₂ catalysts in furfural and anisole hydrogenations by separating metal from the support by a controlled distance using CNT, further substantiating that CNT may act as a bridge between metal and oxide⁴⁵. Torrente-Murciano and co-workers reported the superiority of graphitized CNT as a support for Ru catalysts in NH₃ cracking, due to the electronic modification of Ru particles by CNT⁴⁶. Based on these previous reports, and the unique property of CNT, we presumed that Pd/CNT can deliver hydrogen to ZnZrO_x via hydrogen activation on Pd and spillover via CNT. Our experiments showed that Pd/CNT + ZnZrO_x features drastically improved methanol formation activity from CO₂ compared to not only ZnZrO_x, but also Pd/ZnZrO_x with the same Pd content. The Pd/CNT + ZnZrO_x additionally exhibits excellent long-term stability, surpassing the performance of the state-of-the-art Cu/ZnO/Al₂O₃ industrial catalysts during the on-stream test from 200 h onwards.

Results

Preparation and characterization of the catalysts

The structures of ZnZrO_x, Pd/ZnZrO_x, and Pd/CNT + ZnZrO_x catalysts were characterized using various techniques. According to TEM and SEM analyses, ZnZrO_x is composed of ca. 10–15 nm crystallites, which are present as agglomerated ~µm-scale particles (Supplementary Fig. 1). Pd/CNT contains uniformly dispersed Pd nanoparticles (ca. 5 nm) inside and outside the carbon tubes. (Supplementary Fig. 2). XRD analysis proves that ZnZrO_x has a t-ZrO₂ phase-like crystalline solid solution structure¹⁸. The crystallinity of ZnZrO_x is well-preserved after Pd-impregnation as well as physical mixing with Pd/CNT (Fig. 1a). The TEM image of Pd/ZnZrO_x shows the presence of ca. 5 nm Pd nanoparticles on the surface of ZnZrO_x (Fig. 1b). In Pd/CNT + ZnZrO_x, the majority of Pd particles are present far away from the ZnZrO_x surface (Fig. 1c). As Pd/CNT and ZnZrO_x entangle with each other (Fig. 1c, d), EDX elemental mapping in the 2D-projection shows a uniform distribution of Pd over the catalyst domain (Fig. 1e). H₂- and CO-chemisorption analysis confirms that Pd/ZnZrO_x and Pd/CNT have a similar Pd dispersion of 18–20%, which is consistent with TEM analysis. Of note, Pd dispersion is not altered during physical mixing (Supplementary Table 1). Based on these results, we present an illustration of the three catalysts in Fig. 1f. Furthermore, Pd/GNP (GNP: graphene nanoplatelets), Pd/AC (AC: activated carbon), Pd/SiO₂ and Pd/TiO₂ were prepared and mixed with ZnZrO_x as control samples. The characterizations of these catalysts are included in Supplementary Figs. 3, 4 and Supplementary Table 1. The general observations about their structural features are very similar to those of Pd/CNT and Pd/CNT + ZnZrO_x.

Fig. 1: Structural characterization of catalysts.

a Powder XRD patterns of ZnZrO_x, Pd/ZnZrO_x, and Pd/CNT + ZnZrO_x. The peak shift to higher 2θ of ZnZrO_x compared to pure ZrO₂ indicates the substitution of Zr atoms to Zn atoms (i.e., solid solution formation). The XRD pattern of pure t-ZrO₂ is obtained from Materials Data on ZrO₂ by Materials Project, ID: mp-2574. b TEM image of Pd/ZnZrO_x. c TEM image of Pd/CNT + ZnZrO_x. d SEM image of Pd/CNT + ZnZrO_x. e EDX elemental maps of Pd/CNT + ZnZrO_x during TEM analysis. f Illustration of the structures for ZnZrO_x, Pd/ZnZrO_x, and Pd/CNT + ZnZrO_x catalysts (o.d. outer diameter).

Unique CO₂ hydrogenation performance of Pd/CNT + ZnZrO_x

The activities of ZnZrO_x and various Pd-promoted ZnZrO_x catalysts for CO₂ hydrogenation to methanol (MeOH) were investigated under the reaction kinetics-controlled region (conditions: 533 K, 5 MPa total pressure with CO₂/H₂/Ar = 19/76/5 (molar), conversion: 0.64–1.36%). The exclusion of external and internal mass transfer limitation is described in Section 3 in Supplementary Information.

CO₂ conversion and product selectivity of the catalysts are shown in Supplementary Table 2. For all catalysts, the major products are methanol and CO. Only Pd/SiO₂ + ZnZrO_x and Pd/TiO₂ + ZnZrO_x produced CH₄ (≤3.40%) and dimethyl ether (DME, ≤0.14%) as side products. Since standalone Pd catalysts have negligible methanol formation activity compared to ZnZrO_x (Supplementary Fig. 5), the methanol formation rate (r_MeOH) of ZnZrO_x-containing catalysts in Fig. 2a was represented based on the mass of ZnZrO_x. Pd/ZnZrO_x exhibits higher r_MeOH (3.96 mmol g_ZnZrOx⁻¹ h⁻¹) than ZnZrO_x (1.69 mmol g_ZnZrOx⁻¹ h⁻¹), illustrating the promotional effect from surface Pd nanoparticles. Unexpectedly, a higher r_MeOH is observed over Pd/CNT + ZnZrO_x (13.44 mmol g_ZnZrOx⁻¹ h⁻¹) compared to Pd/ZnZrO_x, which is an exceptional finding, since the metal promoter (Pd) induces a significantly more pronounced effect when away from the active sites on ZnZrO_x. Considering that Pd/ZnZrO_x and Pd/CNT + ZnZrO_x possess similar Pd dispersion and identical Pd loading with respect to ZnZrO_x (Supplementary Table 1), and that no promotion was observed when ZnZrO_x is mixed with CNT (CNT + ZnZrO_x), both Pd and CNT are concluded to be essential for the superior catalytic performance of Pd/CNT + ZnZrO_x. A strong promoting effect is also observed when Pd/ZnZrO_x is mixed with CNT (i.e., r_MeOH = 7.63 mmol g_ZnZrOx⁻¹ h⁻¹ for CNT + Pd/ZnZrO_x) as well as when ZnZrO_x is mixed with other carbon-supported Pd catalysts (i.e., Pd/GNP + ZnZrO_x, Pd/AC + ZnZrO_x). In contrast, when ZnZrO_x is mixed with Pd/oxide catalysts such as Pd/SiO₂ and Pd/TiO₂ (i.e., Pd/SiO₂ + ZnZrO_x, Pd/TiO₂ + ZnZrO_x), there is only slight promotion, demonstrating the unique role of carbon supports. Although TiO₂ is a well-known reducible oxide for fast hydrogen spillover, Pd/TiO₂ has an order of magnitude higher CO₂ hydrogenation activity toward CO compared to other Pd/support catalysts (Supplementary Fig. 5). Thus, Pd/TiO₂ converts CO₂ and H₂ by itself rather than acting as a ‘hydrogen dispenser’. These observations indicate promoting performance is related to hydrogen spillover efficiency, which strongly depends on the type of support as hydrogen moves as an H⁺/e⁻ pair^47,48,49.

Fig. 2: Analysis of reaction kinetics of ZnZrO_x and Pd-promoted ZnZrO_x catalysts in the CO₂ hydrogenation to methanol.

a Methanol formation rate (r_MeOH) over ZnZrO_x-based catalysts (Condition: 533 K, 5 MPa, CO₂/H₂/Ar = 19/76/5 (molar), gas hourly space velocities (GHSV) are shown in Supplementary Table 2). b Reaction orders with respect to H₂ and CO₂ for methanol formation. c Apparent activation energy for methanol formation. The values were determined by the Arrhenius plot shown in Supplementary Fig. 7. d Effect of catalyst mixing configurations on methanol formation in the combination of Pd/CNT and ZnZrO_x. Inset shows the illustration of each mixing configuration. The green dashed line in Fig. a, d indicates the activity of Pd/ZnZrO_x.

In reaction kinetic analyses (Fig. 2b and Supplementary Fig. 6), the reaction orders with respect to H₂ (n_H2) and CO₂ (n_CO2) for methanol synthesis over ZnZrO_x are ca. 1.0 and 0, respectively. The results imply that the surface coverage of H₂ is much lower than that of CO₂. The n_H2 = ca. 1.0 suggests that H₂ activation may be a rate-determining step. Pd/ZnZrO_x catalyst shows similar n_H2 and n_CO2 values to those of pristine ZnZrO_x. In contrast, there is a substantial decrease of n_H2 (0.84) alongside a simultaneous increase of n_CO2 (0.21) for Pd/CNT + ZnZrO_x. The result indicates the coverage of hydrogen on the ZnZrO_x surface increased considerably through combination with Pd/CNT. As shown in Fig. 2c and Supplementary Fig. 7, Pd/CNT + ZnZrO_x exhibits drastically decreased apparent activation energy (E_a = 0.50 eV) compared to ZnZrO_x (E_a = 1.18 eV), confirming the immense contribution of physical mixing with Pd/CNT on accelerating reaction kinetics.

For the ZnZrO_x and Pd/CNT combination, the effect of mixing configuration was studied to gain insight into the origin of promotion (Fig. 2d). Four different mixing configurations were prepared by varying the mixing method, namely separated bed, large pellet mix (size: 450–850 µm), small pellet mix (size: 180–450 µm), and mortar grind mix (size: 0.2–20 µm). The separated bed configuration gives a detrimental effect on methanol formation, implying the promoting effect in Pd/CNT + ZnZrO_x is not originated from the transport of intermediates in the gas-phase. On the other hand, a higher activity compared to pristine ZnZrO_x was detected for all other configurations. The promoting effect is strengthened as the contact between the two components (Pd/CNT and ZnZrO_x) increases, reaching the maximum in the mortar grind mix configuration. It is noteworthy that even when the two components are mixed in 180–850 μm-scale, the promoting effect (r_MeOH = 4.81–6.42 mmol g_ZnZrOx⁻¹ h⁻¹) is stronger than that of Pd/ZnZrO_x (Fig. 2a, r_MeOH = 3.96 mmol g_ZnZrOx⁻¹ h⁻¹). This result firmly demonstrates the effectiveness of the strategy for hydrogen delivery; hydrogen transfer efficiency is no longer limited by the number of Pd-ZnZrO_x interface sites, but rather by the closeness of ZnZrO_x particles and CNT.

We further tested the performance of Pd/CNT + ZnZrO_x under industry-relevant reaction conditions (5 MPa, GHSV = 24,000 cm³_STP g_cat.⁻¹ h⁻¹) to evaluate the practical potential of this catalyst. First, CO₂ conversion and selectivity of products at different temperatures (513–633 K) are summarized in Supplementary Table 3. The result indicates that methanol and CO are the major products under these reaction conditions. The selectivity of Pd/CNT + ZnZrO_x is lower than that of ZnZrO_x and Pd/ZnZrO_x due to the CO formation activity of Pd/CNT itself (Supplementary Fig. 5). Nevertheless, the MeOH selectivity gap between Pd/CNT + ZnZrO_x and ZnZrO_x is less than 10% at the same CO₂ conversion level (Supplementary Fig. 8).

Methanol yield and corresponding space-time yield of methanol (STY_MeOH) were monitored while varying reaction temperature (Fig. 3a). The STY_MeOH for Pd/CNT + ZnZrO_x is 0.780 g g_cat.⁻¹ h⁻¹ at 593 K, the highest value among all reported ZnZrO_x-based catalysts (Supplementary Table 4). Moreover, the long-term stability of each catalyst was monitored at 593 K while a commercial Cu/ZnO/Al₂O₃ catalyst was simultaneously introduced as a control sample. As shown in Supplementary Fig. 9, the optimal reaction temperature for Cu/ZnO/Al₂O₃ catalyst that shows the maximum STY_MeOH is 533 K (thermodynamic equilibrium MeOH yield = 24%), thus the long-term test was conducted at this temperature for Cu/ZnO/Al₂O₃. Not unexpectedly, Cu/ZnO/Al₂O₃ exhibits relatively high activity initially but shows rapid deactivation from the beginning (Fig. 3b); after 600 h of reaction, 46% of initial activity has been lost. In contrast, Pd/CNT + ZnZrO_x catalyst shows superior long-term performance; the methanol yield over Pd/CNT + ZnZrO_x reaches 13.5% (STY_MeOH = 0.90 g g_cat.⁻¹ h⁻¹) at 600 h time-on-stream, very close to the value at thermodynamic equilibrium (MeOH yield = 14.0% at 593 K). XANES and EXAFS analyses after the reaction reveal that there are no substantial changes in the chemical state of either Zn or Zr in Pd/CNT + ZnZrO_x, validating the superior stability of Pd/CNT + ZnZrO_x (Supplementary Figs. 10–11). Both catalysts were additionally tested at a higher GHSV (80,000 cm³_STP g_cat.⁻¹ h⁻¹) to shift conversion further away from thermodynamic equilibrium. Again, Cu/ZnO/Al₂O₃ exhibited continued deactivation while the Pd/CNT + ZnZrO_x system kept increasing its methanol productivity during a 104 h test. (Supplementary Fig. 12). Compared to ZnZrO_x and Pd/ZnZrO_x, distinct increases in conversion and methanol selectivity are observed for Pd/CNT + ZnZrO_x, particularly during the initial stage of reaction (Fig. 3c). Under the steady state (time-on-stream >400 h), Pd/CNT + ZnZrO_x shows 66% methanol selectivity at 20% CO₂ conversion, higher than the methanol selectivity of Cu/ZnO/Al₂O₃ at comparable CO₂ conversion (50–52% at 23–20% CO₂ conversion). As a result of the opposite activity variation trend, cumulative methanol production (Q_MeOH) of Pd/CNT + ZnZrO_x overwhelms that of commercial Cu/ZnO/Al₂O₃ after 360 h reaction (Fig. 3b, inset). At 600 h, Pd/CNT + ZnZrO_x offered a 1.44-fold higher methanol yield over the commercial Cu/ZnO/Al₂O₃ catalyst, and its STY_MeOH (g g_cat.⁻¹ h⁻¹) compare favorably with state-of-the-art catalysts, that were scrutinized in long-term stability tests in the literature (Supplementary Table 5), further validating the superior performance of the Pd/CNT + ZnZrO_x catalyst. In addition, Pd/CNT + ZnZrO_x contains 33 wt% CNT, which can act as a catalyst binder/diluent due to its high mechanical strength⁵⁰ and thermal conductivity⁵¹. Therefore, Pd/CNT + ZnZrO_x may not require further use of an extra binder/diluent, providing additional benefits for practical implementation.

Fig. 3: Catalytic performance of Pd/CNT + ZnZrO_x in industry-relevant conditions.

a Methanol yield and STY_MeOH of the ZnZrO_x-based catalysts as a function of reaction temperature. The dashed line indicates the methanol yield at thermodynamic equilibrium. b Long-term catalytic tests of ZnZrO_x-based catalysts (593 K), and commercial Cu/ZnO/Al₂O₃ (533 K). The inset shows cumulative methanol production (Q_MeOH, g g_cat.⁻¹) as a function of time-on-stream (Condition: 5 MPa, CO₂/H₂/Ar = 19/76/5, GHSV = 24,000 cm³_STP g_cat.⁻¹ h⁻¹). c CO₂ conversion and methanol selectivity changes during long-term reaction test over ZnZrO_x, Pd/ZnZrO_x, Pd/CNT + ZnZrO_x, and commercial Cu/ZnO/Al₂O₃ catalysts.

Characterization of the promotion effect in Pd/CNT + ZnZrO_x

The origin of promotion was elucidated by quantitative gas adsorption-desorption analyses. CO₂-TPD profiles reveal ZnZrO_x, Pd/ZnZrO_x, and Pd/CNT + ZnZrO_x have comparable adsorption strengths (Supplementary Fig. 13) and capacity of CO₂ (Fig. 4a), indicating CO₂ adsorption properties of the catalysts are similar. H₂–D₂ isotope exchange experiments were then conducted at 373 K (Fig. 4b) to determine their H₂ activation function. ZnZrO_x shows a low H–D exchange rate (8.93 mmol g_ZnZrOx⁻¹ min⁻¹), while Pd/ZnZrO_x and Pd/CNT + ZnZrO_x show accelerated rates at 42.9 and 49.8 mmol g_ZnZrOx⁻¹ min⁻¹, respectively. We draw two major conclusions from these isotope exchange experiments. First, Pd species easily activate H₂, regardless of whether Pd nanoparticles are directly supported on ZnZrO_x, or on CNT. Second, the H₂ activation function alone is not the critical factor in determining methanol formation activity, since the H–D exchange rates on Pd/ZnZrO_x and Pd/CNT + ZnZrO_x are comparable despite their pronounced difference in methanol production.

Fig. 4: Quantitative gas adsorption-desorption analysis of catalysts.

a CO₂ adsorption capacity of catalysts determined by CO₂-TPD. b H–D exchange rate of catalysts at 373 K. c H₂-TPD profile and quantification of reversible H₂ of catalysts. The quantification of reversible H₂ was conducted by integrating the peak area in the range of 323–533 K which is below the temperature of our reaction kinetic analysis. d Zn 2p_3/2 XPS of ZnZrO_x and Pd/CNT + ZnZrO_x catalysts before and after sequential gas treatments (gas treatment condition: H₂ at 673 K, CO₂ at 533 K, and CO₂ + H₂ (1:4) at 533 K).

Next, H₂-TPD profiles of various catalysts were examined to differentiate their ability in delivering activated hydrogen (H*) (Fig. 4c). ZnZrO_x shows an H₂ desorption peak above 423 K, while Pd-containing catalysts exhibit a peak starting from 343 K due to more facile desorption of H₂ from the Pd surface. The amounts of adsorbed H₂ on ZnZrO_x and Pd/ZnZrO_x were 2.3 and 4.1 µmol g_cat.⁻¹, respectively. In sharp contrast, Pd/CNT + ZnZrO_x shows an order of magnitude increase in H₂ adsorption (39 µmol g_cat.⁻¹). Considering Pd/CNT alone only possesses 7.1 µmol H₂ g_cat.⁻¹, and that Pd/CNT accounts for one-third of the mass in the mixture catalyst, over 90% of the adsorbed hydrogen in the Pd/CNT + ZnZrO_x mixture is stored at the interfaces between ZnZrO_x and CNT, which undergo reverse spillover during TPD analysis (Fig. 4c, right panels). In the case of Pd/SiO₂ and Pd/TiO₂, there are only marginal increments in the H₂ adsorption capacity after physical mixing with ZnZrO_x (Supplementary Fig. 14), confirming the unique property of CNT as a hydrogen delivery system. Band gap analysis suggests carbonaceous materials have much smaller band gap values (0.74–0.75 eV) compared to metal oxides (≥3.09 eV) including ZnZrO_x (Supplementary Fig. 15), suggesting higher electrical conductivity results in faster migration kinetics of hydrogen (H⁺/electron pair) on the carbon surface versus oxides such as ZnZrO_x.

X-ray photoelectron spectroscopy (XPS) after sequential gas treatments (without sample exposure to air) for ZnZrO_x and Pd/CNT + ZnZrO_x were carried out to provide more insight into the hydrogen spillover. The spectra of each element are demonstrated in Supplementary Figs. 16–20, while the XPS of Zn 2p_3/2 region is represented in Fig. 4d. Before gas treatment, ZnZrO_x and Pd/CNT + ZnZrO_x show identical symmetric Zn 2p_3/2 signals at 1022 eV which represent Zn²⁺ bound to lattice oxygen (Zn²⁺–O²⁻). After H₂ treatment at 673 K, ZnZrO_x shows no signal changes, whereas Pd/CNT + ZnZrO_x exhibits the formation of a shoulder peak at ~1 eV lower binding energy region. The transformation of Zn²⁺–O²⁻ to reduced Zn species (Zn^δ-)^52,53 is a result of electron transfer from spilt over hydrogen to the Zn site of the ZnZrO_x surface. Both H₂-reduced catalysts were further treated with CO₂ and CO₂ + H₂ after which a shoulder peak at ~1 eV higher binding energy region appears, indicative of CO₂ chemisorption on Zn–O pair (Zn²⁺–CO₃²⁻)⁵³. Notably, the existence of Zn^δ- species is still pronounced in Pd/CNT + ZnZrO_x after consecutive treatments with CO₂ and CO₂ + H₂. The portion of Zn^δ- is slightly decreased after CO₂ treatment (from 63 to 57%), but it recovers the original fraction when treated with CO₂ + H₂ (65%), providing evidence of effective hydrogen spillover in the Pd/CNT + ZnZrO_x system under CO₂ hydrogenation condition.

Computational studies on hydrogen spillover in Pd/CNT + ZnZrO_x

Density functional theory calculations were applied to provide an atomistic description of hydrogen spillover in the Pd/CNT + ZnZrO_x nanocomposite. After identifying the most stable model for ZnZrO_x (Supplementary Figs. 21–24), we studied the adsorption and movement of H on the bare ZnZrO_x surface (Fig. 5 and Supplementary Fig. 25). H₂ dissociation on pure ZrO₂ is kinetically limited⁵⁴, which can be improved by doping the oxide with Zn atoms^18,55. We calculated Zn–O pairs in the ZnZrO_x surface to have the lowest dissociative adsorption energies of H₂ among all surface sites (Supplementary Fig. 25b). Since our ZnZrO_x supercell contains two Zn sites (Fig. 5a), we are able to perform simulations for the sequential dissociation of two H₂ molecules. The dissociation of the first H₂ molecule is a barrierless step (Fig. 5b) with reaction energy of ΔG = − 0.31 eV (Fig. 5c). Yet, the dissociation of the second H₂ molecule is impaired by a higher activation barrier (ΔG^a = 0.71 eV) and reaction energy of ΔG = 0.17 eV (Fig. 5c), indicating limited H coverage on unmodified ZnZrO_x under reaction conditions.

Fig. 5: Simulations of H atoms on the bare ZnZrO_x surface.

a Top view of four employed ZnZrO_x supercells illustrating 4 H atoms adsorbed per unit cell on Zn–O pairs, formed after successive dissociation of two H₂ molecules. b Potential energy profile along the reaction coordinate for barrierless dissociation of the first H₂ molecule (TS transition state, FS final state). c Potential energy profile of successive H₂ dissociation at ZnZrO_x for 2 H₂ molecules. d Energy profile for H movement from Zn to Zr, O to O, and Zr to Zr on the ZnZrO_x surface. The inset images show the transition state configurations. Turquoise, purple, red, and yellow colors represent Zr, Zn, O, and H, respectively.

Moreover, H atoms are calculated to have negligible mobility on the ZnZrO_x surface with barriers up to ΔG^a = 1.18 eV for diffusion between O sites and ΔG^a = 1.51 eV for migration of H atoms from Zn to Zr atoms (Fig. 5d). Although a much lower barrier of ΔG^a = 0.34 eV was calculated for H diffusion between Zr sites, these sites were calculated to be up to 2.22 eV less stable for H than O sites on ZnZrO_x surface based on the adsorption energy of one H atom. These suggest that directly supported Pd nanoparticles cannot deliver reactive H atoms to the active sites on ZnZrO_x since long-distance H spillover is hindered by high diffusion barriers of H atoms on the oxide surface.

In contrast to the bare ZnZrO_x, our calculations show that H can be fluently supplied to the active sites on the ZnZrO_x surface via spillover from Pd/CNT (Fig. 6). We created models of physisorbed CNT on ZnZrO_x using an adapted lattice matching algorithm, which ensured the minimal strain of −2.33% along the CNT length (Fig. 6a). According to our simulations, a single H atom bind stronger to ZnZrO_x than to CNT, which creates a significant thermodynamic driving force for the spillover and makes this process essentially barrierless (Fig. 6c). However, the spillover barrier increases when more H atoms are adsorbed on CNT. For example, the Gibbs barrier for the spillover of an H atom from CNT to ZnZrO_x increases to 0.43 eV when another H atom is adsorbed in the most stable position on the vicinal site on the inner side of the nanotube (Fig. 6d). In turn, H spillover from CNT to ZnZrO_x becomes endothermic by about ΔG ~ 2.5 eV when every C atom of CNT is hydrogenated. In conclusion, low H coverage on CNT can be regarded as the critical structural factor enabling hydrogen spillover from C to the oxide. Low H coverage on CNTs is also suggested by our XPS studies showing mostly sp² hybridization of C atoms in the CNT under reaction conditions (Supplementary Fig. 19).

Fig. 6: Simulations of hydrogen spillover in Pd/CNT + ZnZrO_x.

a Atomic models of CNT(5,5) on ZnZrO_x and Pd on carbon (a graphene sheet) for H spillover. b Energy profile for movement of one H from hydrogen-saturated Pd to the graphene sheet with a pre-adsorbed H atom (IS initial state, TS transition state, and FS final state). c Energy profile for spillover of the single H atom from CNT(5,5) to ZnZrO_x. d Energy profile for spillover of one H atom from CNT(5,5) with another H atom adsorbed on it to ZnZrO_x.

To evaluate the rate of H spillover from Pd to sp² carbon, we simulated the movement of H from a 1.56 nm large Pd₁₂₇ fcc nanocrystallite to a graphene sheet representing a broad nanotube. We chose the shape of the Pd₁₂₇ particle in line with TEM data and identified the most stable structure of the Pd₁₂₇/graphene interface by screening all high-symmetry interfaces (Supplementary Fig. 26). Our calculations show that the barriers for H spillover from Pd to graphene are essentially governed by the endothermicity of this process (Fig. 6b), which depends on the H coverage. Namely, the spillover of the first H atom from Pd₁₂₇ to graphene is endothermic by ΔG = 1.57 eV, which decreases to ΔG = 0.37 eV when another H atom is present on the graphene sheet. Since the Gibbs adsorption energy per H atom of one (two) H on the freestanding CNT(5,5) is 1.04 eV (0.70 eV/H atom), which is 0.84 eV (0.37 eV/H atom) stronger than on graphene, the curvature of the CNT should further favor the transfer of H from Pd to the carbon domain. Thus, the barriers for H spillover from Pd to CNT are estimated to be much lower than the activation energy of H₂ dissociation on ZnZrO_x (ΔG^a = 0.71 eV). In turn, CNTs are well known to be excellent transport channels for H atoms⁴⁵, whose movement over graphene was shown to be enabled by quantum tunneling⁵⁶. Thus, our simulations suggest a much faster H supply to the active sites on Pd/CNT + ZnZrO_x nanocomposite compared to the pristine ZnZrO_x.

Effect of catalyst structure on catalytic results in CO₂-to-methanol

A number of studies using metal/support catalysts (e.g., Cu or Pd on ZnO, ZrO₂, or ZnZrO_x) refer to hydrogen spillover as an important process for methanol synthesis from CO/CO₂^{23,57,58,59,60,61,62,63}_, but it is difficult to disentangle catalytic activity generated by sites adjacent to metals from that generated by sites farther away⁶⁴. In Fig. 7, we present the effect of the catalyst structure on CO₂ hydrogenation to methanol over Pd-promoted ZnZrO_x catalysts. ZnZrO_x is an excellent material to chemisorb CO₂, but methanol formation is limited due to low H₂ activation and H* delivery capability (Fig. 7a). In turn, Pd particles supported on ZnZrO_x significantly enhance H₂ activation, but the delivery of H* toward the active sites is hampered due to slow H-migration kinetics within ZnZrO_x surface (Fig. 7b). It is estimated the contact area between Pd particles and ZnZrO_x is about 0.5 m² g_cat.⁻¹, only 1% of the entire catalyst surface area (estimation procedure available in Section 4 in Supplementary Information). Our experimental and computational studies support Prins’s rigorous critique that hydrogen may not travel farther than the immediate interface between metal and basic oxide support⁴⁷, and Li et al.’s hypothesis that the rate-determining step of CO₂ hydrogenation to methanol over various catalysts (i.e., ZnZrO_x, Cu/ZnO/Al₂O₃, and Pd/ZnO) is the migration of H* on the catalyst surface⁶⁵. In the case of Pd/CNT + ZnZrO_x, CNT acts as a hydrogen dispenser, effectively delivering activated H* from Pd particles to the ZnZrO_x surface (Fig. 7c). Since the promotion range is determined by the contact area between CNT and ZnZrO_x, a ten-times larger promotion area is generated (as determined by H₂-TPD result) under mortar grind mode compared to traditional metal doping on the catalyst surface, leading to a drastically enhanced methanol formation rate. In the literature, non-contact integration of multiple catalytically active components, especially physical mixing, has been adopted in catalyst development for bifunctional- or cascade catalytic reactions for which each catalyst is responsible for one step of the entire transformation^{66,67,68,69,70,71,72,73,74}. Our work, in contrast, reports a bicomponent catalyst in which one component provides active sites for the formation of a target product while the other serves as an activation and delivery system for a kinetically hindered intermediate, which to our knowledge, has not been demonstrated for CO₂ hydrogenation.

Fig. 7: Schematic illustration of the effect of catalyst structure on CO₂ hydrogenation to methanol.

a ZnZrO_x catalyst without any metal promotion. b Pd-supported ZnZrO_x catalyst which represents the traditional metal/oxide catalyst harnessing metal promotion. c Pd/CNT + ZnZrO_x catalyst which represents a conceptually new catalyst system harnessing metal promotion with a hydrogen dispenser.

Furthermore, our preliminary tests of physical mixing of commercial Cu/ZnO/Al₂O₃ with CNT results in 1.4 to 2.0-times higher methanol formation activity versus pristine Cu/ZnO/Al₂O₃ in the pressure range of 0.1–1.0 MPa (Supplementary Fig. 27), suggesting this promotion strategy is not limited to ZnZrO_x. For the successful implementation of the CO₂-to-MeOH process on large scales, recycling of the product stream (including unconverted CO₂, H₂, and produced CO) is necessary to intensify process economics^75,76. Future research should be directed to studying the effect of CO on nanoscale H supply in newly developed systems.

Discussion

Insufficient H₂ activation has long been known to limit the activity of oxide catalysts, such as ZnZrO_x, for CO₂ reduction to methanol. In this work, we report CNT as a hydrogen dispenser for enhancing hydrogen delivery to active sites on ZnZrO_x. The combination of Pd/CNT and ZnZrO_x leads to an exceptional catalytic system for methanol formation, far better than Pd/ZnZrO_x catalyst, as CNT effectively delivers Pd-activated hydrogen to a broad range of ZnZrO_x surface sites. The Pd/CNT + ZnZrO_x shows the highest STY_MeOH among reported non-Cu/ZnO-based catalysts to date and is durable over 600 h continuous operation, demonstrating its potential for practical implementation; concurrently, the physical mixing strategy offers a robust and transferable method for material integration to design better catalytic systems. This study provides an example in oxide catalyst design for which the transition metal promoter in non-contact mode induces a greater effect than the transition metal promoter directly attached to the surface of the catalytically active phase, expanding the concept of metal promotion beyond the traditional focus of metal-support interfaces.

Methods

Catalysts preparation

ZnO-ZrO₂ (Zn/Zr = 1/5) solid solution catalyst (ZnZrO_x) was prepared via a coprecipitation method with reference to the earlier literature²². An 1 wt% Pd/ZnZrO_x catalyst was prepared by wet impregnation using Pd(NO₃)₂·xH₂O as Pd precursor. To make the Pd/CNT + ZnZrO_x catalyst, 2 wt% Pd was first supported on CNT by wet impregnation followed by reduction at 673 K, and then Pd/CNT was physically mixed with ZnZrO_x with Pd:ZnZrO_x = 1:100 mass ratio. Physical mixing was carried out by mortar grinding, unless otherwise stated. For comparison, various other physically mixed catalysts (Pd/support + ZnZrO_x) were also prepared in a similar manner while varying the support material (support = type of support used; AC: activated carbon, GNP: graphene nanoplatelet, SiO₂: silica, TiO₂: titania P25). CNT + ZnZrO_x and CNT + Pd/ZnZrO_x were also prepared by physical mixing of CNT and ZnZrO_x or Pd/ZnZrO_x. Prior to the characterizations and catalytic reactions, the catalysts were mortar ground. As shown in Section 5 in Supplementary Information, the mortar grinding itself does not significantly affect the surface area and catalytic activity of ZnZrO_x.

Catalyst characterization

The structures of catalysts were characterized using transmission electron microscopy (TEM) coupled with energy dispersive X-ray (EDX) elemental analysis, scanning electron microscopy (SEM), powder X-ray diffraction (XRD), N₂ physisorption, H₂ and CO chemisorption, CO₂- and H₂-temperature-programmed desorption (TPD), H₂–D₂ isotope exchange, UV-Vis-NIR spectroscopy, and X-ray photoelectron spectroscopy (XPS).

Catalytic test

CO₂ hydrogenation was carried out using an automatic multi-channel high-pressure flow reactor designed by PID Eng & Tech. Prior to the reaction, the catalysts were pressed, cracked, and sieved to make pellets. The sieved catalyst pellets (450–850 µm) were loaded into the reactor, and pretreated with H₂ (40 cm³_STP min⁻¹) at 673 K for 2 h (For Cu/ZnO/Al₂O₃ catalysts, pretreatments were conducted at 573 K). After the pretreatment, the reactor was cooled down to a target temperature (493–633 K), and was pressurized with a reaction gas mixture (CO₂/H₂/Ar, typically 19/76/5). The outlet gas line was kept at 453 K to prevent condensation of products, and products were analyzed by online gas chromatography (Agilent 8890) equipped with a flame ionization detector (FID) and a thermal conductivity detector (TCD). Agilent HP-PoraplotQ and Restek ShinCarbon were used as columns. Reported data are given as values of time-on-stream at 3 h unless otherwise stated.

Computational studies

The computational approach was based on spin-unpolarized density functional theory within the Perdew-Burke-Ernzerhof⁷⁷ exchange-correlation approximation as implemented in the Vienna Ab initio simulation package (VASP)^78,79. The zero-damping D3 correction was added to improve the description of dispersive interactions⁸⁰. To treat strong correlation effects for d electrons, the simplified implementation of Hubbard corrections⁸¹ was adopted with U − J = 4 eV on Zr atoms. Atomic charges were assigned using Bader charge analysis^82,83. The Gibbs adsorption energies of H were calculated as ${G}_{{ad}}=G[{{{{{\rm{H}}}}}}/{{{{{\rm{cat}}}}}}]-E[{{{{{\rm{cat}}}}}}]-0.5\,G[{{{\mbox{H}}}}_{2}]$, where, $G[{{{{{\rm{H}}}}}}/{{{{{\rm{cat}}}}}}]$ is the Gibbs total energy of the system with the adsorbate, while $E[{{{{{\rm{cat}}}}}}]$ and $G[{{{{{{\rm{H}}}}}}}_{2}]$ are the total and Gibbs total energy of the isolated catalyst model and H₂ molecule in the gas phase. All energetic properties except for the relative energies in Figs. 5b and 6b–d, include Gibbs corrections calculated at 533 K and 4 MPa of H₂ pressure under either the ideal gas or harmonic approximations^84,85. In Fig. 6, the H movement coordinate is the quadratic distance of each structure with respect to the corresponding initial state. The k-point sampling was performed using Monkhorst-Pack meshes yielding converged adsorption energies of H, namely, the k-mesh parameters were 8 × 1 × 1 for all systems containing CNT(5,5), 2 × 2 × 1 for the 10×10 supercell of graphene, while the gamma point was sufficient for calculations involving only the ZnZrO_x supercell. All models had a vacuum region of about 11 Å between adjacent slabs and all lateral distances between periodic images were at least 6 Å to keep interactions between periodic images minimal. All systems were optimized with the convergence criteria of 0.03 eV/Å and 10⁻⁵ eV for the force on each atom and total electronic energy. To optimize transition states, the climbing image nudged elastic band^86,87 and dimer⁸⁸ methods were employed and the optimized first-order saddle points were confirmed to have one imaginary vibrational frequency leading to the reported local minima.

Additional details of the methods for catalyst preparation, catalyst characterization, catalytic test, and computational studies are provided in the Supplementary Information.