The Role of Cu1-O3 Species in Single-Atom Cu Catalyst for Directional Methanol Synthesis from CO2 Hydrogenation

Cu-based catalysts have attracted much interest in CO 2 hydrogenation to methanol because of their high activity. However, the effect of interface, coordination structure, particle size and other underlying factors existed in heterogeneous catalysts render to complex active sites on its surface, therefore it is dicult to study the real active sites for methanol synthesis. Here, we report a novel Cu-based catalyst with isolated Cu active sites (Cu 1 -O 3 units) for highly selective hydrogenating CO 2 to methanol at low temperature (100% selectivity for methanol at 180 o C). Experimental and theoretical results reveal that the single-atom Cu-Zr catalyst with Cu 1 -O 3 units is only contributed to synthesize methanol at 180 o C, but the Cu clusters or nanoparticles with Cu-Cu or Cu-O-Cu active sites will promote the process of reverse water gas shift (RWGS) side reaction to form undesirable byproducts CO. Furthermore, the Cu 1 -O 3 units with tetrahedral structure could gradually migrate to the catalyst surface for accelerating CO 2 hydrogenation reaction during catalytic process. The high activity isolated Cu-based catalyst with legible structure will be helpful to understand the real active sites of Cu-based catalysts for methanol synthesis from CO 2 hydrogenation, thereby guiding further design the Cu catalyst with high performance to meet the industrial demand, at the same time as extending the horizontal of single atom catalyst for application in the thermal catalytic process of CO 2 hydrogenation.


Introduction
The excessive use of fossil fuels in recent decades has led to a dramatic increase in the amount of CO 2 in the air, which has caused serious damage to the natural environment. Hence, the related C1 chemistry has become an important research area because one of the most challenging scienti c issues is to nd alternative energy sources for petroleum in the 21st century 1 . Especially, the conversion of excess CO 2 into high value-added chemicals and energy fuels in industry can not only effectively mitigate the greenhouse effect, but also realize the sustainable use of resources by closing the natural carbon ring 2-6 . Methanol (CH 3 OH), as a basic industrial raw material, can be used to synthesize a series of important industrial chemicals such as low-carbon Ole ns and gasoline by means of MTO and MTG [7][8][9][10][11][12] . It is obvious that the process of CH 3 OH catalysis from CO 2 is of great commercial value. Therefore, in the past few decades, a lot of research work has been studied to nd catalysts with good performance, including metal-metal oxides (Cu/ZnO/Al 2 O 3 13,14  Cu catalyst has attracted much attention because of its excellent catalytic activity and stability to CH 3 OH synthesis among the CO 2 hydrogenation catalysts 18,26 . However, Cu-based heterogeneous catalyst is di cult to have a clear understanding of the active sites due to its complex microstructure and various effects between the metal and the support. The diversity of valence states of Cu and the particularity of hydrogenation reactions make Cu mostly exist in mixed valence states during the reaction. It is confused that the previous literatures show that Cu 0 , Cu + or Cu δ+ may be active sites [27][28][29] . Moreover, the physical size of Cu nanoparticles also profoundly affects the catalytic performance 17,30−32 . Taking Cu/ZrO 2 catalyst as an example, Witoon etc. 33 found that the Cu/ZrO 2 catalyst with amorphous phase is favorable for methanol synthesis, but Samson etc. 16 claimed that the Cu/ZrO 2 catalyst with tetragonal phase has the best catalytic activity in methanol synthesis; Ma etc. 18 believed Cu-Zr interface played a crucial role for methanol synthesis from CO 3 *→HCOO*→CH 3 O* route via stabilizing the activated Cu + , but Lercher etc. 32 thought Cu-O-Zr could only activate CO 2 to synthesize CO rather than methanol.
Although substantial progress has been made in the active sites of Cu catalysts, there are still many controversies on the structure-performance correlation between catalyst and reaction.
Single-atom catalyst is an ideal model for active site study due to its possessed uniform metal sites which is embodied in the homogeneous catalyst [34][35][36] . However, the related work is rarely reported because it is di cult to construct the effective active sites in the thermal catalytic hydrogenation of CO 2 to CH 3 OH [37][38][39] . Here, an e cient Cu 1 /ZrO 2 single-atom catalyst is synthesized to realize the reaction of CO 2 hydrogenation to CH 3 OH at relatively low temperature (180 o C). Compared to the typical Cu/ZrO 2 catalysts, monoatomic dispersed Cu 1 /ZrO 2 displayed highest TOF MeOH and 100% CH 3 OH selectivity.
Based on the above catalytic model, the relationship between Cu active sites structure and the formation of methanol or CO is in depth exploration.

Results
The design of various type Cu/ZrO 2 catalysts with different coordination structure models. A series of Cu/ZrO 2 catalysts with different Cu loading (1-20 wt%) were synthesized by modi ed co-precipitation method. And the Cu-based amorphous/monoclinic ZrO 2 with different Cu amount (x wt%) were named as CAZ-x and CMZ-x, respectively. The actual Cu concentration in various Cu/ZrO 2 samples were measured by Inductively Coupled Plasma Optical Emission Spectrometry (ICP-OES) ( Table S1). As shown in Figure   S1a, all the Cu/a-ZrO 2 catalysts with different Cu loading amount exhibited amorphous state of ZrO 2 and no XRD peaks of CuO species, suggesting that the Cu species were highly dispersed on a-ZrO 2 33 . Similar speci c surface areas and pore structures also indicated a high degree of dispersion of Cu species (Table   S2). Notably, the amorphous ZrO 2 transformed to tetragonal phase without adding Cu 2+ precursor during the same co-precipitation procedure ( Figure S1b). The changed phases indicated that strong interaction effect was existed between Cu and ZrO 2 . The broad peak at 523cm − 1 also con rmed the amorphous zirconia were formed in different Cu loading Cu/a-ZrO 2 catalysts from the Raman spectroscopy ( Figure   S2) 40 .
To con rm the hyper ne structure of the Cu/a-ZrO 2 catalysts at the atomic scale, high angle dark eld scanning transmission electron microscope images (HAADF-STEM) and X-ray absorption ne structure (XAFS) spectroscopy were used. The HAADF-STEM images (Fig. 1a, Figure S3) of CAZ-1 showed that the amorphous state ZrO 2 occupied the sample with no Cu nanoparticles on it. The elements mapping ( Fig. 1d) and N 2 O titration test (Table S3) also con rmed the highly dispersed Cu sites were located in the CAZ-1 catalyst consistent with the XRD and TEM results. With Cu loading amounts increasing to 15 wt%, the Cu species still kept in a highly dispersed state even though a small amount of Cu particles were started to form on the ZrO 2 substrate ( Fig. 1b and Figure S4). Extended x-ray absorption ne structure spectroscopy (EXAFS) in R space and corresponding WT spectra results could supply more important information about the catalysts structure, as shown in Fig. 1 (g, h) and Figure S5 (a, b) Figure S6a) and at 291 cm − 1 in Raman spectroscopy ( Figure S2) were corresponded to the existence of CuO nanoparticles 42 . The highly agglomerated CuO particles (15-20 nm) were also observed in the CMZ-15 catalyst from the results of TEM (Fig. 1c, Figure S7), elements mapping ( Fig. 1f, Figure S7) and N 2 O titration (Table S3). Moreover, an additional smaller peak corresponding to CuO at 2.81 Å was observed in EXAFS and WT spectra (Fig. 1g, Figure S5b), ascribed to Cu-(O)-Cu scattering at second shell, con rming the formation of CuO particles 43  over all the catalysts was controlled less than 10% to study their intrinsic activity, which was far beneath its CO 2 equilibrium conversion at this condition (29.7% at 180 o C, Figure S8). As shown in Fig. 2(a, b) and Table S5, only CH 3 OH could be detected without any other byproducts over CAZ-1 catalyst and it preferentially produced CH 3 OH with a TOF Cu value up to 1.37h − 1 . Therefore, a "homogeneous" active site catalyst with excellent stability was synthesized for CH 3 OH production from CO 2 hydrogenation. However, CO 2 could not be activated in CAZ-1-r (refer to CAZ-1 pre-reduced by H 2 at 370°C to form Cu large particles), CMZ-15 and CS-15, where the Cu species were almost entirely composed of large metallic Cu particles, con rmed by XRD ( Figure S6a, b, Figure S9) and H 2 -TPR ( Figure S10). Therefore, the highly dispersed isolated Cu species might be the real active sites for the CH 3 OH synthesis from CO 2 at low temperature. As the Cu amount increased in the CAZ-x catalysts (Fig. 2c), the CO started to be produced from CAZ-4 catalyst, indicating additional active sites were formed in these high Cu loading amount catalysts ( Figure S11). It is inferred that the formed Cu clusters of small nanoparticles on CAZ-x catalyst (> 4 wt% Cu) played a crucial role in accelerating reverse water gas shift reaction (RWGS) since it was no activity over CAZ-1-r, CMZ-15 and CS-15 catalysts with large metallic Cu particles. It was found that the catalysts exhibited excellent intrinsic activity in the low loading range (< 2 wt% Cu). It showed an approximately linear growth relationship of CO 2 conversion between CAZ-1 and CAZ-2 with only CH 3 OH produced (the ratio of CO 2 Conv. CAZ-2/CAZ-1 and Cu amount CAZ-2/CAZ-1 is 0.96), indicating that all Cu active sites were uniformly dispersed and exposed on the catalysts surface. However, the increasing trend of conversion rate decreased as the Cu amount higher than 4 wt% (The ratio of CO 2 Conv. CAZ-4/CAZ-1 and Cu amount CAZ-4/CAZ-1 is 0.67), since the Cu clusters or small nanoparticles could not provide CO 2 activation ability as strong as isolated Cu active sites, thereby the CO selectivity was much lower than CH 3 OH selectivity. When the Cu loading amount was more than 8 wt%, the surface of the catalyst was lled up with Cu clusters and isolated Cu sites, at the same time as Cu large particles without CO 2 activation ability being gradually formed ( Figure S12), resulting in no further increase of the CO 2 conversion due to the saturation of surface active sites.
The conclusion was also evidenced by the catalytic performance and ICP-OES results ( surface were about 7 ~ 8 wt%. Meanwhile, it was proved that the active sites in the catalyst were in a very stable structure, even strong acid could not break its structure. The CAZ-1 catalyst was also conducted in CO 2 conversion reaction for continuous 100 h to evaluate its catalytic stability (Fig. 2d, Figure S13). After the induction period for a couple of hours, the catalyst with isolated Cu active sites gave both very stable CO 2 conversion and target products selectivity during this period, which revealed that the catalyst possessed an extremely steady structure in the CO 2 hydrogenation reaction, Furthermore, fresh and used CAZ-1 catalysts had similar weight loss and heat absorption/exothermic trend, which indicated that CAZ-1 catalyst had stable structure and no carbon deposition on its surface during low-temperature reaction ( Figure S14).
Surface electronic state and coordination structure. The catalysts electronic state and the coordination environment in the short range were investigated to study the relationships to their catalytic performance. The X-ray photoelectron spectroscopy (XPS) and X-ray near-edge structure spectroscopy (XANES) results were shown in Fig. 3. In comparison to the CAZ-15, the absence of Cu 2p satellite peaks between 940eV and 945eV implied that no Cu 2+ species exited in the fresh and used CAZ-1 catalysts 15,44 . Identical peaks situated at 932.9eV between Cu 0 (932.4eV) and Cu 2+ (934.6eV) were observed in CAZ-1 and CAZ-1-U (U: used), indicating the Cu species were at cationic Cu δ+ state 31,45 . The position of Cu 2p 3/2 was not shifted after the reaction, giving another strong evidence that the active Cu δ+ species was very stable in the CAZ-1 catalyst, at the same time as no metallic Cu nanoparticles forming to make sure supplying enough stable active sites in CO 2 hydrogenation process ( Fig. 2d and S13). Moreover, the peak of Cu L 3 M 45 M 45 Auger centred at 915.5eV without any shifting in CAZ-1 and CAZ-1-U also con rmed the stable Cu δ+ species in catalyst surface ( Figure S15) 31 . However, the satellite peaks of Cu 2p 3/2 that appeared in CAZ-15, suggesting that Cu 2+ species were formed in CuO clusters or nanoparticles (Fig. 3a) 46 . After the CO 2 hydrogenation reaction, the weakening of the satellite peak intensity and the shifting of Cu 2p3/2 from 933.6eV to low binding energy (932.5eV ~ Cu0) indicated that the Cu 2+ was partially in-situ reduced to Cu 0 at CAZ-15 during the reaction 45,47 . The XAES results also proved the valence state of Cu partially changed after reaction in CAZ-15 catalyst ( Figure S15). Therefore, the unstable surface structure led to its relatively poor CH 3 OH selectivity. The co-existed Cu 0 and Cu δ+ species in CAZ-15 would convert CO 2 to CO and CH 3 OH, respectively. The chemical states of Cu species were also investigated by XANES (Fig. 3b).
The weak pre-edge peak located at 8977-8978eV (1s→3d) and 8984eV(1s→4p) suggested that the Cu species were exited in the form of cationic state in CAZ-1 and CAZ-15, which was a little similar to the CuO with tetrahedral structure 41,48,49 . However, compared to the CuO, the obvious "step" of the pre-edge peak located at ~ 8984eV(1s→4p) disappeared since the tetrahedral structure was highly asymmetric according to its unsaturated coordination state. Meanwhile, it was inferred that the Cu species in the bulk structure of the CAZ-15 catalyst were also highly dispersed, but the different surface Cu state led to its different catalytic performance in CO 2 hydrogenation. The apparent pre-edge peak at 8984eV was detected in CMZ-15, which con rmed that CuO particle was embodied in CMZ-15 49 . The EXAFS tting data for Cu/ZrO 2 were shown in Fig. 3c, Figure S16 and Table S4, suggesting the local structure of CAZ-1 was composed of one isolated Cu atom coordinated with ~ three oxygen atoms (Cu 1 -O 3 unit) in a tetrahedral geometric con guration with defects. The stable isolated Cu 1 -O 3 units rather than Cu-(O)-Cu in CAZ-15 were the real active sites for CH 3 OH synthesis from CO 2 .
The active sites for Cu/ZrO 2 catalysts. It was supposed that isolated Cu 1 -O 3 sites might be favor of CO 2 directional converting to CH 3 OH. Because no Cu particles were detected in used CAZ-1( Fig. 4a and Figure   S17a), which con rmed that isolated Cu 1 -O 3 active sites in Cu/ZrO 2 were really stable. However, the local Cu particles with 0.21nm spacing for Cu (111) planes 15,48,50,51 detected in used CAZ-15 by HRTEM (Fig. 4b, Figure S17b) revealed that Cu species were partially aggregated and reduced during the catalytic process, consistent with the XPS results ( Fig. 3a and Figure S15). H 2 -TPR was also tested to understand the evolution of the Cu species under reduced atmosphere ( Figure S10). The Cu δ+ species in CAZ-1 were hard to be reduced because only one peak at 360 o C was appeared, which was much higher than actual reaction temperature, indicating the stability of Cu 1 -O 3 active sites. As the Cu content increased from 1 wt% to 15 wt%, the reduction peak shifted to lower temperature, demonstrating that the interaction between the Cu species and ZrO 2 carrier became weak gradually. The reason was the Cu species would be more easily aggregated thereby further reduced as its loading amount increasing. It was also supported by XRD results (Fig. 4c, Figure S12) that no Cu particle diffraction peaks were detected and the crystal structure still kept in amorphous state for the used CAZ-1 catalyst. In contrast, two obvious sharp peaks (43.3 o and 50.4 o ) 52,53 were appeared in used CMZ-15, suggesting the formation of Cu from CuO reduction ( Figure S6). Furthermore, in-situ XRD was employed to investigate the dynamic evolution of Cu δ+ species in Cu/a-ZrO 2 under H 2 atmosphere. As shown in Fig. 4d, no signal of Cu 0 was observed in  Figure S18), the peak at ~ 420 o C were assigned to the H 2 absorbed sites at Cu species and a-ZrO 2 substrate. Compared with CAZ-1, the peak at ~ 420 o C in CAZ-15 migrated to a higher temperature (415 o C → 430 o C), possibly caused by the more supplied adsorption sites on CAZ-15 surface, indicating that the structure of Cu species in CAZ-15 was more complex and non-unique. Based on above results, it is concluded that the dispersion of Cu species in CAZ-1 were still in single-atom state; Cu single-atom with partially reduced clusters or nanoparticles were coexisted in CAZ-15 during the reaction. As for CMZ-15, both XRD ( Figure S6) and H 2 -TPR ( Figure S10) results indicated that almost all Cu species were reduced to larger Cu particles during the reaction.
While Cu loading amount was less than 2 wt% in CAZ-x catalysts, the only product was CH 3 OH since the isolated Cu 1 -O 3 unit played the crucial role for CH 3 OH synthesis (Fig. 2c). As the Cu loading amount was increased to 4 wt%, a small amount of CO was produced, because a small number of reduced Cu clusters or nanoparticles were formed during the reaction ( Figure S11), which might be the active sites for RWGS reaction. By increasing the Cu content to 8 wt%, the proportion of CO in the products was also raised due to the more formed Cu clusters or small nanoparticles. In result, all the evidence suggested that singleatom Cu species with isolated Cu 1 -O 3 unit and Cu clusters or small nanoparticles was the active sites for generating CH 3 OH and CO, respectively. However, it was found that when the Cu loading amount was further increased to 12 wt%, 15 wt% and 20 wt%, the CO 2 conversion and products distribution did not change any more. Because large particles of Cu would be formed as the Cu content further increased, but large particles of Cu had no ability to activate CO 2 in low temperature, consistent with the reaction results under CMZ-15 and CS-15 catalysts (Fig. 2). Besides, more clearly Cu lattice fringe in HAADF-STEM ( Fig. 4b), more obvious Cu bulk structure in XRD (Fig. 4c), and the Cu reduction peak with lower temperature in H 2 -TPR ( Figure S10) together claimed that more Cu particles were generated by reducing Cu 2+ species over the CAZ-15 during the reaction.
Cu δ+ species migration during the reaction. Capturing the evolution of active metal species during the reaction is very essential for the in-depth understanding of the active site, especially in reactions involving hydrogen. A number of studies have shown that the gases in the pretreatment or reaction process greatly affect the structure of catalysts, including active species migration and surface reconstruction [54][55][56] . A similar effect also existed in our Cu/ZrO 2 catalyst system, where H 2 promoted the migration of Cu δ+ species to the surface of support. Time of Flight Secondary Ion Mass Spectrometry (TOF-SIMS) was an extremely sensitive method to analyze the dispersion of elements at the catalyst surface, where located at 1-3 atomic layers on the surface of catalyst can be detected (Fig. 5a, b). More bright red spots were detected in Fig. 5b, suggesting more Cu species appeared at the surface in used CAZ-1 catalyst. Semi-Quantitative analysis of the surface Cu (Fig. 5c) also displayed that the surface of used CAZ-1 catalyst contained more Cu species, which indicated that Cu species would migrate to the surface during the reaction. Same results were also supported by XPS analysis (Figure S19), the intensity of Cu 2p spectra for the used CAZ-1, CAZ-15 were much stronger than fresh catalysts respectively, which meant Cu species were getting enriched on used catalyst surface after the reaction. Meanwhile, Cu/Zr ratio calculated by XPS further con rmed the above conclusion. The surface Cu/Zr ratios of CAZ-1 and CAZ-15 samples were 0.0012 and 0.24, and the corresponding ratios for used Cu/ZrO 2 catalysts were 0.0018 and 0.38, respectively. Obviously, there were more Cu active sites on the catalysts surface appeared after reaction. Moreover, in-situ DRIFT was used to verify the Cu migration as well and supply another important evidence. The catalysts were adsorbed to saturation in CO atmosphere before the desorption experiment was carried out with Argon. The absorbed peak at 2102cm − 1 was assigned to linear adsorption of CO on the Cu species ( Figure S20a, b) 42,57 . For the fresh catalyst, the CO concentration decreased rapidly with the Ar purge time, and the surface residual concentration was close to zero at 20 min. In comparison, the desorption rate of CO on the used CAZ-1 was much slower than that on the fresh catalyst, at the same time as the obvious CO signal being still observed after 20 min gas purging. For more intuitive comparison, it was normalized the area of the CO absorption peak at 2102 cm − 1 and made a timedependent surface concentration attenuation spectrum with the Ar purging time ( Figure S20c). It was also concluded that the quantity of CO absorption on used CAZ-1 was much more than that on the fresh catalyst, which proved that the surface of the used catalyst contains more Cu sites.
Based on the above facts of Cu migration, it was inferred that the dynamic change of Cu 1 -O 3 active sites during the reaction process. EXAFS data have shown that the Cu species in Cu/a-ZrO 2 were mainly singleatom dispersed in the fresh catalyst, and the Cu species would migrate to the surface at a certain rate under the induction of reaction gas. That was the reason of the reaction had a certain induction period at initial hours (Fig. 2d, S13). It was concluded that the unique Cu 1 -O 3 structure in CAZ-1 was vital for CH 3 OH synthesis at the low temperature. The Cu sites in CAZ-1 remained Cu single-atom with isolated Cu 1 -O 3 structure after migrating to the surface due to the small amount of Cu. However, more Cu sites in CAZ-15 were accumulated and in-situ reduced during migration in hydrogen atmosphere process (Fig. 4cd). Therefore, the migration behavior of Cu species during the reaction could be divided into three types according to the catalytic activity results: 1. When Cu content was less than 2 wt%, the migrated Cu species were not accumulated during the reaction process and the Cu species still distributed in monodispersed state, hence only CH 3 OH was produced and CO 2 conversion was enhanced with the increasing Cu loading amount in linear; 2. When the Cu loading amount was between 4-8 wt%, the Cu species were migrated to the surface for partly agglomeration, and the activity showed that the CO 2 conversion and CO selectivity was increased while the CH 3 OH selectivity was decreased due to the formed Cu clusters or small nanoparticles producing CO via RWGS; 3. When the Cu loading amount was higher than 8 wt%, the partial Cu species would be kept aggregating to form large Cu particles without CO 2 activation ability, thus the conversion of CO 2 and the products distribution were not changed any more. The schematic diagram was shown in Fig. 5d.
The reaction mechanism of CO 2 hydrogenation on Cu/ZrO 2 catalysts. To clarify the structure in uence for absorbed species on the surface, in-situ diffuse re ectance Fourier transform infrared spectroscopy analysis was carried out at reaction situation. All the tests were performed at 180 o C and the assignments of all band vibration peaks were listed in Table S6. As shown in Fig. 6a for CAZ-1, the active species were mainly rst exited in HCO 3 * ; the peaks located at 1695cm − 1 , 1431cm − 1 were assigned to bicarbonate species, i-HCO 3 * and 1631cm − 1 , 1226cm − 1 were v as (HCO 3 ), v s (HCO 3 ) of b-HCO 3 *58 , respectively. As the reaction proceeded, the HCO 3 * were slowly transformed into formate species, according to the peaks at Therefore, the process of CO 2 hydrogenation to CH 3 OH in CAZ-1 was followed the formate path.
Furthermore, it was proved that the CH 3 OH was not produced from the in-situ CO hydrogenation via the RWGS route at initial according to the reaction results of showing no catalytic activity over CAZ-1 in CO hydrogenation process under same reaction condition (Table S5). For CAZ-15 catalyst, the CH 3 OH signal located at 1007cm − 1 was detected during the in-situ DRIFT test 61 , which indicated the process of CH 3 OH formation was greatly promoted on CAZ-15 due to possessing more de nitely quantity of active sites for CH 3 OH synthesis ( Figure S22). In fact, the CO 2 adsorption capacity of the catalysts was not signi cantly changed by only enhancing the loading amount of Cu species, because CAZ-1 and CAZ-15 had similar adsorption mode and adsorption amount for CO 2 (3.0mmol/g cat in CAZ-1, 2.9mmol/g cat in CAZ-15, Figure   S23). It was concluded that the increase of methanol production rate was not affected by the adsorption capacity in these two catalysts, but mainly caused by the different active sites numbers. As increasing Cu loading amount to 15 wt%, the absorption of intermediate species on the surface of CAZ-15 was much more complex than that on CAZ-1. In addition to the adsorbed bicarbonate (1621cm − 1 , 1225cm − 1 ), carbonate (1247cm − 1 , 1324cm − 1 , 1455cm − 1 ) formate (1360cm − 1 , 1384cm − 1 , 2864cm − 1 , 2974cm − 1 ) and methoxy species (1070cm − 1 , 1146cm − 1 , 2836cm − 1 , 2921cm − 1 ), the carboxylate signals were also captured at 1287cm − 1 and 1756cm − 1 63 . The CH 3 OH was formed by accompanied with the reverse water gas shift reaction since more complex Cu active sites (single-atom, cluster and particle) were provided under reaction, leading to the decrease of CH 3 OH selectivity. As for CMZ-15 with Cu large nanoparticles ( Figure S24), all the absorbed species were CO 3 * or HCO 3 * but no further hydrogenation intermediates were observed, demonstrating that the Cu large nanoparticles could not realize the further hydrogenation of carbonate at low temperature, which was corresponded to the results of activity test.
Based on the above information, intensity-time shift spectra supplied the dynamic behavior of intermediates converting on CAZ-1 and CAZ-15 in the reaction (Fig. 6b). The peaks at 2877cm − 1 , The excellent selectivity to CH 3 OH of CAZ-1 could be attributed to its unique Cu 1 -O 3 catalytic center as con rmed by DFT calculations. Figure 7 and Figure  possible channels for CO 2 hydrogenation, i.e., to HCOO * or to COOH * , our calculations showed that the formation of COOH * group had a high barrier of 0.78 eV ( Figure S27), but the formation of HCOO * was barrierless (0.01 eV). Thermodynamically, the COOH * group was also 0.63 eV less stable than the HCOO * group. The formed HCOO * adopted a bidentate con guration with two O ends linking with the Zr and Cu It should be mentioned that the identi ed HCOO * and CH 3 O * intermediates were consistent with the in-situ spectroscopy from experiment. The rate determining step in the pathway was the CH 3 O * hydrogenation to CH 3 OH * that has a barrier of 1.06 eV. The data also agreed reasonably with the barrier (~ 1.15 eV) deduced from the experiment TOF (1.3 s − 1 ) for CAZ-1 based on microkinetics.
Based on the above research results, we established the reaction model diagram for CO 2 hydrogenation over the CAZ-x series catalysts. As shown in Fig. 8, when the Cu species were distributed in a single atom level with uniform Cu 1 -O 3 catalytic centers on the surface of ZrO 2 , CO 2 would convert to methanol with 100% selectivity. When the Cu species were existed in the form of clusters or small nanoparticles, CO 2 could only produce CO. As the Cu species were in larger particles, there was almost no catalytic effect for activating CO 2 . Therefore, it was believed that the Cu 1 -O 3 sites in Cu single atoms and Cu clusters/small nanoparticles were the catalytic active sites for methanol and CO synthesis from CO 2 hydrogenation, respectively, while the larger Cu particles were not the active sites for CO 2 hydrogenation at this condition.

Conclusions
This work reveals the in uence of different Cu structure models on CO 2 hydrogenation process. The unsaturated coordination Cu 1 -O 3 species is favorable for generating HCOO * , which is a vital intermediate during the CH 3 OH synthesis. It is also proved that HCOO * pathway is the only viable route for CO 2 hydrogenation. Moreover, the hydrogen is more easily dissociated and further hydrogenated in the Cu atom with its adjacent O atoms. Therefore, the isolated Cu 1 -O 3 active sites contained in Cu single-atom catalyst realizes the directional selective of CO 2 to CH 3 OH (100%) at 180 o C, at the same time as migrating dynamically to the catalyst surface with stable structure during the catalysis process. The Cu clusters and/or small nanoparticles is the active sites for CO formation via the RWGS route. By contrast, the Cu large particles cannot activate CO 2 in the low temperature at all. The Cu single-atom catalyst with clear geometry and uniform active site structure provides guidance for further understanding of the Cu active center in CO 2 hydrogenation to CH 3 OH reaction, at the same time as extending the horizontal of single atom catalyst for application in the thermal catalytic process of CO 2 hydrogenation.

Methods
Synthesis of CAZ-1. CAZ-1 was synthesized by modi ed co-precipitation with Na 2 CO 3 as precipitant.
Firstly, weighing Cu(NO 3 ) 2 ·3H 2 O (Sigma-Aldrich, 98%-103%) and Zr(NO 3 ) 4 ·5H 2 O (Macklin, AR) precursors according to the metal loading and dissolved them together into 100ml deionized water to make a 0.03M solution, which was recorded as solution A. Then it was weighed an appropriate amount of Na 2 CO 3 and dissolved it in 100ml deionized water to make a 0.06M solution, which was marked as solution B. After the dissolution completed, a peristaltic pump was used to slowly drip the two solutions A and B into another beaker containing 100ml of deionized water at a rate of 0.3ml/min. During the dropping process, the stirring maintained at 350 rpm/min and heated at 80°C until the dropping was completed. Then stopped stirring and aged at 80°C for 2h. After aging, cooled it to room temperature and washed with deionized water to pH=7, followed by drying in an oven at 80°C overnight. Finally, it was ground into powder and calcined in a mu e furnace at 2°C/min to 350°C for 5h.
Catalyst characterization. The Cu actual loading in different catalysts was determined by inductively coupled plasma optical emission spectroscopy (ICP-OES) which performed on an iCAP 7000 SERIES stabilized, the reacted gas ow was changed to 10ml/min and Shimadzu chromatography equipped with a ame ionization detector (FID) and a thermal conductivity detector (TCD) was used to analyze products online every 30 minutes. The catalytic performance of catalysts was evaluated by CO 2 conversion, product selectivity, space-time-yield (STY) and turnover frequency (TOF). The computational formula as follow: C CO2 -Conversion of CO 2 , %; X CO2, in -mole fraction of CO 2 in pristine mixed gas; X CO2, out -mole fraction of CO 2 in exit gas. Then, the NN potential is generated using the method as introduced in our previous work. 73,74 To pursue a high accuracy for PES, we have adopted a large set of power-type structure descriptors, which contains 324 descriptors for every element, including 148 2-body, 170 3-body, 6 4-body descriptors, and compatibly, the network utilized is also large involving two-hidden layers (324-50-50-1 net), equivalent to 75,000 network parameters in total. The min-max scaling is utilized to normalization the training data sets.
Hyperbolic tangent activation functions are used for the hidden layers, while a linear transformation is applied to the output layer of all networks. The limited-memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) method is used to minimize the loss function to match DFT energy, force and stress. The nal energy and force criterions of the root mean square errors are around 6.0 meV/atom and 0.151 eV/Å respectively.
To obtain a reasonable model of Cu SAC on amorphous ZrO 2 , SSW-NN method is applied to exhaustively search the phase space of Cu 1 O 1 /ZrO 2 . The model builds as follows: (i) From the most stable monoclinic ZrO 2 bulk phase, the most stable (-111) surface with (4 × 4) supercell and three ZrO 2 -layers thickness is built which contains 144 atoms (48 ZrO 2 formula unit).
(iii) One CuO formula unit (2 atoms) is randomly added around the amorphous ZrO 2 .
(iv) Starting from this initial model, more than 10000 minima are visited by SSW-NN simulation. From that, the most stable structure of Cu 1 O 1 on amorphous ZrO 2 is obtained with a special CuO 3 tetrahedron con guration with the Cu-O bond distances of 1.814, 1.920 and 1.921 Å ( Figure S28) which is well consistent with the EXAFS results. This model is veri ed by plane wave DFT calculations and then adopted for the CO 2 hydrogenation reaction.     In-situ DRIFT spectroscopy for CAZ-1. Analysis condition: the sample was rst pretreated under Ar atmosphere at 230oC for 60min, then insert CO2+H2 (1:3) to 3MPa and record data at 180oC for 100min.  The reaction diagram for different Cu-Zr catalysts.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SupplementaryInformation.pdf