The most active Cu facet for low-temperature water gas shift reaction

Identification of the active site is important in developing rational design strategies for solid catalysts but is seriously blocked by their structural complexity. Here, we use uniform Cu nanocrystals synthesized by a morphology-preserved reduction of corresponding uniform Cu2O nanocrystals in order to identify the most active Cu facet for low-temperature water gas shift (WGS) reaction. Cu cubes enclosed with {100} facets are very active in catalyzing the WGS reaction up to 548 K while Cu octahedra enclosed with {111} facets are inactive. The Cu–Cu suboxide (CuxO, x ≥ 10) interface of Cu(100) surface is the active site on which all elementary surface reactions within the catalytic cycle proceed smoothly. However, the formate intermediate was found stable at the Cu–CuxO interface of Cu(111) surface with consequent accumulation and poisoning of the surface at low temperatures. Thereafter, Cu cubes-supported ZnO catalysts are successfully developed with extremely high activity in low-temperature WGS reaction.

Review of "The Active Cu Structure for Low-Temperature Water Gas Shift Reaction" Summary: This paper describes the reactivity of Cu nanocrystals (NCs) with distinct exposed facets towards the water-gas-shift reaction. The NCs were synthesized as Cu 2 O, and were characterized and probed both before and after reduction. The authors provide evidence that the (100) facet of Cu is very active. This kind of work is important to merge the fields of single crystal surface science and catalysis engineering. However, the authors have not fully characterized their catalysts to the level that Nat. Commun. would require and have made claims that over-step what the data allows. The kinetic test analysis has flaws as does the computational part. For these reasons I do not recommend acceptance of the manuscript in the present form. My concerns are listed below in more detail: Major Concerns:

1)
It is well known that Cu 2 O is relatively unstable. The Auger peaks for Cu(I) and Cu(II) come at nearly identical kinetic energies, do the authors have any evidence that CuO isn't also present on the surface of these nanoparticles? 2) On p. 6 the authors compare the activity of the NCs to Cu/ZnO/Al 2 O 3 . But it is not an accurate comparison to calculate catalytic activity based on total Cu atoms for the Cu/ZnO/Al 2 O 3 and only surface Cu atoms for the NCs. The activity calculation needs to be the same for all catalysts being compared. The commercial catalyst is several orders of magnitude more active than the NCs. Reaction rates (per g catalyst or per SA) may be used for comparison, but not TOFs.

3)
On p. 7 the authors discuss that both the Cu and Cu 2 O NCs are equivalent in activity under reaction conditions. The authors then state this a evidence for Cu(0) as the active site, as opposed to Cu(I). However, this statement is inaccurate and should be removed, as the authors go on to show how a partially oxidized surface is much more active than the bare metal. The data seems to suggest that the catalyst is very dynamic and doesn't exist in purely one oxidation state, rather a mix of the oxidation states is likely created.

4)
Related to the previous point, the authors should characterize their postreaction NCs to quantify how the catalyst has changed, which would shed light on the active state of the catalyst.

5)
The poisoning by formate mechanism is surprising at first because formate should be much more weakly bound than the other listed carbon species. What seems to lead to the poisoning is that the formate is made at the active site and blocks it. Is it possible for the other carbon species to be removed in order to more definitively identify the formate species as the poisoning adsorbate? 6) On p. 9 the authors state that the CO oxidation takes place on the Cu -Cu suboxide interface. However the data only shows this oxygen species as being active and there is no evidence for the reaction taking place exclusively at the interface, as opposed to on top of the suboxide. It should be made clearer to the reader that the data presented does not identify if the reaction takes place at the interface or on top of the suboxide regions. 7) The interface as the locus of activity for the inverse structures of ZnO/Cu single crystal surfaces has been demonstrated by the work of Rodriguez et al. (not included in the refs. here; ref. 27 reaches the same conclusion for ceria). At the interface, the oxidation state of the active Cu species has been debated in the literature, and convergence to Cu(I) is shown. In the work by Flytzani-Stephanopoulos et al., the WGS reaction has been found to occur on cationic metal species, including Au(I), Pt(II), Cu(I), etc. In other words, the reduction to the metal is not complete. See for example, Rui Si et al., "Structure sensitivity of the low-temperature water-gas shift reaction on Cu-CeO 2 catalysts", Catal. Today, 2012, 180, 68. How to probe the interface with any spectroscopy in the preponderance of metallic Cu as in the authors' NCs is a formidable and perhaps untenable task.

8)
A cautionary statement in the kinetics work. Have the authors conducted their tests at steady-state and from high-T to low-T? if not, structural changes at the high temperatures, could have caused lower rates, i.e. an erroneous apparent activation energy, which can explain the values of ~ 30 kJ/mol reported for the Cu/ZnO catalysts. There is no reason for the activation energy to be different, if the same cuprous species is the active site in all such catalysts, i.e. the mechanism does not change. Cu should not be compared to other metals like gold. The latter has its own "signature" with Ea lower than that of Cu (or Pt) on any support. 9) The oxidation of flat copper surfaces has been extensively studied. Although the exact structure of Cu under reaction conditions is not known, the design of the DFT model for the Cu-Cu 2 O interface should resemble some of the motifs found on carefully studied single crystal surfaces (Besenbacher and Norskov, Prog. Surf. Sci. 1993, 44, 5). For example, O on Cu(111) readily reconstructs into ring structures over a wide range of O coverages and temperatures (Matsumoto et al. Surf. Sci., 2001, 471, 225;Soon et al. Phys. Rev. B, 2006, 73, 165424;Yang et al. J. Phys. Chem. C, 2010, 114, 17042), but a structure like the zig-zag pattern that the authors chose as their model has never been observed experimentally. This is unfortunate given the large library of data on partially oxidized Cu surfaces, and dramatically diminishes the impact of the calculations with the structure chosen. The authors should provide some justification to the reader for the design of their DFT model, and should also provide the reader with information of observed O/Cu structures from the literature, including some of the references listed above, and how their structures differ.
Minor Concerns:

10)
Recently there has been a study on Cu single crystals from the Salmeron group showing that the (111) facet has the best activity towards CO oxidation. (Eren et al. J. Phys. Chem. C 2015, 119, 14669) CO oxidation is a crucial step in the water-gas-shift reaction; could the authors comment on the inverse in the activity hierarchy for these two reactions? 11) I suggest a modification to the title from "the active Cu structure…" to "the most active Cu facet…" because all Cu structures investigated are indeed active, this work is aimed towards finding the most active facets (as apposed to general structure such as nanoparticle size or shape).

12)
The authors may want to consider re-labeling of the "suboxide" species, because it suggests that the oxide is subsurface. However the authors find that there is O on the surface at these suboxide regions, so perhaps calling these sites simply "oxide" would be more fitting.

Reviewer #3 (Remarks to the Author):
This manuscript describes a promising approach to determine the active site structure of C u -based catalysts for the low temperature water-gas shift (WGS) reaction. The information is then used to prepare an improved catalyst sample. The WGS reaction is of great industrial importance and the currently used catalyst, C u/ZnO/Al2O2 is also C u-based; thus, this topic is of high relevance and could be appealing to a large audience.
Despite the potential impact of the work, there are several gaps in the argumentation. These gaps are described below and require major revisions before this manuscript can be reviewed again.
The interpretation of the TPRS data on page 7 leads the authors to claim that H2 desorption is the rate-limiting step of C u-catalyzed WGS. I don't see sufficient evidence for this conclusion. If the active site is the C u/C u-suboxide interface, C O2 desorption could simply be a result of C u-suboxide reduction by C O. I assume similar C O2 desorption peaks would be observed in the absence of H2O. Also, if H2 desorption was rate-limiting, why did the authors not simulate this step with DFT?
The adsorption/desorption of H2 in Table S5 is assumed to be quasi -equilibrated and no barrier is reported.
Ultimately, the authors attribute WGS reactivity to the presence of C u/C u-suboxide interfaces.
With this active site assignment, many of the performance data are questionable. The BET surface area would not be a good way to normalize the TOF. Instead, the authors should normalize by the number of interface sites. The observed differences may simply be a difference in how well the different C u facets oxidize to a suboxide, or how well the epitaxial relationship between the C u metal and the suboxide matches. It is likely that the thermodynamically most stable (1 11) facet is not easily oxidized. From the DFT results, the authors could easily extract the interfacial energy between C u(111) or C u(100) and C u2O, and determine if there is a significant stability difference.
Side note: a brief clarification why the BET surface area of larger o-C u NC s (937 nm) is larger than that of smaller c-C u NC s (877 nm) would be appropriate. It is likely due to the higher site density of the (111) planes.
The high surface roughness is a concern. A rough surface will necessarily h ave defect and step sites. Were there any efforts to quantify the amount of surface defects? The C O vibration experiments should exhibit signals of C O molecules strongly adsorbed on steps/defects. Did the authors try to compare the signal intensities to ge t a rough estimate of the ratio of step to terrace sites?
The DFT results suggest that there is no significant difference between C u(111) and C u(100) model interfaces with C u2O present. The only difference was speculated to be the formate coverage based on the formation and decomposition barriers. Such speculation is rather unreliable and often fails, even in simple cases. The coverage of formate depends on the coverage of H and C O2, which in turn depend on everything else. Also, if formate is a spectator species, it will be formed until it reached its saturation coverage. At this point the formation and decomposition barriers are no longer relevant to estimate the formate coverage at the interface under reaction conditions. Hence, I see value in the DFT work to assign the active site to the C u/C u2O interface, but I do not see agreement with the experimentally observed activity increase on the (100) facet.
Were the transition states found in the NEB method verified by vibrational analysis showing a single imaginary mode? I would be beneficial if the authors clearly stated that zero point energies and entropy corrections were neglected.
C ould the authors please clarify if the reaction energies and barriers are taken with respect to isolated reactants/products? The equation in the SI suggests this but an explicit statement is preferred. If energies are reported at infinite separation, could the authors also include the interaction energies for the co-adsorbed states that form the initial and final states of their NEB calculations?
The manuscript must be edited for grammar and spelling. Most importantly "morphology -reserved" should probably be "morphology-preserved".
Comment: This is a detailed study presenting the use of well defined, ligand free, metallic and oxidized copper nanocrystals to determine the most active phase for the water-gas-shift (WGS) reaction. The authors convincingly show that open surfaces of copper are more reactive than the thermodynamically preferred Cu(111) phase, but more importantly that the presence of Cu2O islands helps the metal nanocrystals to dissociate water, the rate limiting step on metal surfaces for the WGS reaction. The authors also show that by depositing a small coverage of ZnO on top of the Cu nanocrystals they obtain a catalyst that is even more active than the commercially use Cu/ZnO/Al2O3 catalyst, due to the formation of active bifunctional metal/oxide interphases. I find the paper worth of publishing, but there are a few comments that the authors should address before publication.
Author Reply: We appreciate the reviewer's positive and valuable comments very much. We have seriously considered the reviewer's comments and revised the manuscript accordingly. We hope that the revised manuscript will be suitable for the publication in Nature Communications.

Comment 1:
The authors assign the IR features at 2107 and 2109 cm -1 in Figure 1.F to adsorption of CO on Cu(I) sites. This has been done in the literature on powder samples, but on well-defined single crystals it has been shown that these could be better assigned to under-coordinated Cu(0) sites, and that Cu(I) sites lead to CO adsorption with >2,110 cm -1 features (see J. Phys. Chem. C 2014, 118, 15902−15909 and references therein). Since Cu 2 O can be very easily reduced to Cu at room temperature, the top layer of a Cu 2 O NC can consist of metallic Cu sites after a short exposure to CO. In the same Figure 1.F, it is very unlikely that the features at 2193 and 2225 cm -1 are associated to CO 2 as assigned in the manuscript. Linear CO 2 interacts very weakly with any surface, and even when it does it would lead to IR features close to gas phase CO 2 (~2,350 cm -1 ) as it does when it is trapped in zeolites at room temperature.
Author Reply: We appreciate the reviewer's insightful comments very much and we have seriously considered these comments. We agree with that caution must be taken to assign the copper species probed with CO adsorption due to the likely surface reduction induced of copper oxides by CO. All previously observed surface reductions occur at certain temperatures, but during our investigations CO adsorption were carried out at 123 K, as described in the methods section of the supporting information, where the surface reduction of Cu 2 O NCs does not likely occur. Meanwhile, the vibrational feature of CO adsorbed at Cu(I) sites were reported to vary with the Cu(I) structure. Although mostly it locates above 2110 cm -1 at supported and well-dispersed Cu ( (2014)). It can be also seen from our results that CO adsorbed at Cu surface exhibits vibrational features below 2100 cm -1 . Therefore, we believe that it is reasonable to assign 2 the IR features at 2107 and 2109 cm -1 in Figure 1.F to CO adsorbed at Cu(I) sites of Cu 2 O NCs.
After CO adsorption on Cu 2 O octahedra at 123 K, vibrational bands at ∼2193 and 2225 cm -1 were also observed on o-Cu 2 O in addition to the vibrational features of adsorbed CO. We have studied this carefully in our previous publications (Ref. 12 and 19 in the revised manuscript), in which, by comparing with the IR spectra of CO 2 adsorption on Cu 2 O NCs, both features were assigned to CO 2 adsorbed at the coordination-unsaturated Cu(I) sites of o-Cu 2 O. The adsorbed CO 2 is formed by the reaction of CO with adsorbed oxygen species on o-Cu 2 O introduced during the controlled oxidation process of as-synthesized PVP-capped o-Cu 2 O NCs to remove the capping ligand. Again, due to the very low adsorption temperature, the formed CO 2 can stably adsorb at the coordination-unsaturated Cu(I) sites of o-Cu 2 O. In the present manuscript, we cite our previous results on Cu 2 O NCs to briefly explain the experimental observations. In reply to the reviewer, in the revised manuscript, we have added the CO adsorption temperature (123 K) in both the text and the Figure 1  ." (Please see Lines 6-12 on Page 6); we have also re-ordered the references accordingly.

Comment 2:
The authors referred to carbonite species (CO2--), while more likely this are carboxylate (CO2-) species. Due to the similarity in the XPS features, the assignments of carbonate, formate and carboxylate species in Figure S5 should be correlated to the corresponding IR data (similar to Figure 3.D) in order to obtain a more convincing set of assignments.
Author Reply: We appreciate the reviewer's insightful comments very much and we have seriously considered these comments. After carefully looking into the literatures and comparing the differences between the carbonite and carboxylate species, we agree with the reviewer that the carbonite species in our original manuscript should be re-assigned to the carboxylate species. Meanwhile, we have also followed the reviewer's suggestion to examine the in-situ DRIFTS spectra of as-synthesized Cu NCs with the spectra of corresponding Cu 2 O NCs as the reference spectra (shown below). The spectra are dominated by the significant attenuation of the Cu 2 O vibrational features (652, 794 and 1123 cm -1 ), but the formation of carboxylate (1277 cm -1 ) and carbonate (1235, 1455 and 1506 cm -1 ) species could be observed on all Cu NCs and formate (1588 cm -1 ) species could be observed exclusively on o-Cu NCs. These results are in consistence with the XPS results. 3 In reply to the reviewer, we have revised the "carbonite" to the "carboxylate" throughout all the revised manuscript; we have added the in-situ DRIFTS spectra of as-synthesized Cu NCs as Figure  S6 in the revised supporting information (Please see Page S18 in the supporting information) and described and discussed the relevant results as the following: "In the corresponding O 1s and C 1s XPS spectra ( Figure  We have also re-ordered the figures and the references accordingly.

Comment 3:
The authors use of "quasi in situ XPS" is not correct. In situ means "in place" of the reaction, like they do for their DRIFT measurements. They should just mention that they measure XPS on their samples after reactions without exposure to air.
Author Reply: We appreciate the reviewer's critical comments and kind suggestions very much and we have seriously considered them. During our experiments, two types of XPS measurements were carried: one was routine measurements; the other was measurements of the samples subjected to various treatments without exposure to air, which we defined as "quasi in situ XPS" in order to distinguish from the routine XPS measurements. However, we agree with the reviewer that XPS measurements of the samples subjected to various treatments without exposure to air are not appropriate to be defined as "quasi in situ XPS".
In reply to the reviewer, we have removed the term "quasi in situ XPS" throughout all the revised manuscript and mentioned it as "XPS without exposure to air" in the contexts needed.

Comment 4:
In the figures describing the DFT calculations is almost impossible to see the configuration of the intermediates. Is the COOH species they mention the text a carboxyl species with the carbon on the metal or what/where? The authors should ad bigger figures with the critical intermediates (such as the carboxyl COOH species) in order to clarify what and where are these adsorbates.
Author Reply: We appreciate the reviewer's critical comments and kind suggestions very much and we have seriously considered them. We accept the reviewer's suggestions and have split the original Figure 4 into Figure 5 and Figure 6 (shown below) in the revised manuscript (Please see Figure 5 on Page 33 and Figure 6 on Page 34) and included an additional Figure S18 with the structures of all important adsorbates and intermediates (shown below) in the revised supporting information manuscript (Please see Figure S18 on Page S31) in order to provide high-quality figures. We have also made revisions in the texts and re-ordered the figures accordingly.  Author Reply: We appreciate the reviewer's careful reading and kind suggestions very much and have replaced "reserved" with "preserved" throughout the revised manuscript and supporting information.
Minor comments: Page 4: the paragraph where the sentence ".., and reasonably are smaller than the corresponding .." is used is not clear and should be re-written.
Author Reply: We appreciate the reviewer's careful readings very much and have rewritten the sentence as the following in the revised manuscript: "However, the acquired Cu NCs exhibit rough surfaces likely resulting from the lattice mismatch between Cu 2 O and Cu. The sizes of acquired Cu NCs are smaller than those of corresponding Cu 2 O NCs." (Please see Lines 21-23 on Page 4).
Minor comments: Page 5: same with "..was while no feature was observed on.." Author Reply: We appreciate the reviewer's careful readings very much and have rewritten the sentence as the following in the revised manuscript: "Agreeing with previous reports, 12,19-21 a vibrational band was observed at ∼2108 cm -1 on o-Cu 2 O and d-Cu 2 O and assigned to CO adsorbed at the Cu(I) sites exposed on the {111} and {110} facets while no feature was observed on the Cu 2 O cubes enclosed with the O-terminated Cu 2 O (100) facets;" (Please see Lines 7-10 on Page 12).
Summary: This paper describes the reactivity of Cu nanocrystals (NCs) with distinct exposed facets towards the water-gas-shift reaction. The NCs were synthesized as Cu2O, and were characterized and probed both before and after reduction. The authors provide evidence that the (100) facet of Cu is very active. This kind of work is important to merge the fields of single crystal surface science and catalysis engineering. However, the authors have not fully characterized their catalysts to the level that Nat. Commun. would require and have made claims that over-step what the data allows. The kinetic test analysis has flaws as does the computational part. For these reasons I do not recommend acceptance of the manuscript in the present form. My concerns are listed below in more detail.
Author Reply: We appreciate the reviewer's positive and valuable comments very much. We have seriously considered the reviewer's comments and revised the manuscript accordingly. We hope that the revised manuscript will be suitable for the publication in Nature Communications.
Comment 1) It is well known that Cu 2 O is relatively unstable. The Auger peaks for Cu(I) and Cu(II) come at nearly identical kinetic energies, do the authors have any evidence that CuO isn't also present on the surface of these nanoparticles?  Figure S2 on Page S13, Figure S5 on Page S17, Figure S15 on Page S27, and Figure S16 on Page S28). We have also made revisions in the texts and re-ordered the figures accordingly.  Comment 2) On p. 6 the authors compare the activity of the NCs to Cu/ZnO/Al2O3. But it is not an accurate comparison to calculate catalytic activity based on total Cu atoms for the Cu/ZnO/Al2O3 and only surface Cu atoms for the NCs. The activity calculation needs to be the same for all catalysts being compared. The commercial catalyst is several orders of magnitude more active than the NCs. Reaction rates (per g catalyst or per SA) may be used for comparison, but not TOFs.  Ed. 2007Ed. , 46, 1329Ed. -1332.) and adopted their methods. But we definitely agree with the reviewer that such comparisons are not appropriate and accept the reviewer's suggestion to revise. In reply to the reviewer, we have used the surface Cu atom-specific reaction rates, instead of TOF, to describe the catalytic activity of Cu NCs in the revised manuscript (Please see Figure 2 on Page 30) and the mass-specific reaction rates of to describe the catalytic activity of Cu/ZnO/Al 2 O 3 in the revised supporting information (Please see Figure S8 on Page S20). We have also removed the data of Cu/ZnO/Al 2 O 3 from the revised Figure 2A and compared the intrinsic catalytic activity of Cu NCs and Cu/ZnO/Al 2 O 3 with the calculated apparent activation energies ( Figure S2C). We have revised the relevant texts and discussion in the revised manuscript as the following: "Various types of Cu NCs exhibit morphology-dependent catalytic activity in the WGS reaction. Comment 3) On p. 7 the authors discuss that both the Cu and Cu 2 O NCs are equivalent in activity under reaction conditions. The authors then state this a evidence for Cu(0) as the active site, as opposed to Cu(I). However, this statement is inaccurate and should be removed, as the authors go on to show how a partially oxidized surface is much more active than the bare metal. The data seems to suggest that the catalyst is very dynamic and doesn't exist in purely one oxidation state, rather a mix of the oxidation states is likely created.
Author Reply: We appreciate the reviewer's careful readings and insightful comments very much and we have seriously considered them. In the original manuscript we wanted to express that Cu, instead of Cu 2 O, as the active copper phase in the WGS reaction. Such an argument is reasonable since Cu(I) is present on the metallic Cu surface. In reply to the reviewer, we have re-written the sentence as the following in the revised manuscript: "These results demonstrate Cu, instead of Cu 2 O, as the active copper phase in the WGS reaction and further support the Cu{100} facet as the most active facet." (Please see the last sentence on Page 8).

Comment 4)
Related to the previous point, the authors should characterize their postreaction NCs to quantify how the catalyst has changed, which would shed light on the active state of the catalyst.
Author Reply: We appreciate the reviewer's insightful comments very much and we have seriously considered them. In the supporting information we have already provided XRD patterns ( Figure S12) and SEM images ( Figure S13) for supporting our argument that Cu 2 O NCs undergo the in-situ morphology-preserved reduction into the corresponding Cu NCs during the WGS reaction up to 548 K. We have measured the Cu LMM AES spectra of Cu 2 O NCs and Cu NCs mixed with Al 2 O 3 after the catalytic activity evaluations in the WGS reaction up to 548 K (shown below). The results show that the Cu LMM AES spectra of Cu 2 O NCs and Cu NCs mixed with Al 2 O 3 after the catalytic activity evaluations in the WGS reaction up to 548 K are very similar. However, at that time, we did not have access to the XPS facility capable of treating the catalysts and measuring XPS without exposure to air, thus the spectra were measured with the catalysts taken out from the reactor and exposed to air, resulting in that the Cu(I)/Cu(0) ratios on the used Cu NCs surfaces are higher than those measured without exposure to air ( Figure S15) due to the occurrence of the surface oxidation. But all these structural characterization results adequately prove that the Cu 2 O NCs undergo the in-situ morphology-preserved reduction into the 11 corresponding Cu NCs in the view of both bulk structure and surface structure.
In reply to the reviewer, we have added the Cu LMM AES spectra of Cu 2 O NCs and Cu NCs mixed with Al 2 O 3 after the catalytic activity evaluations in the WGS reaction up to 548 K as Figure S14 on Page S26 in the revised supporting information to demonstrate their surface compositions and explained the results in the figure caption. We have rewritten the relevant texts in the revised manuscript as the following: "We found that, when directly used as the catalysts for the WGS reaction, various types of Cu 2 O NCs exhibited the steady-state catalytic activity and calculated apparent activation energy almost identical to those of the corresponding Cu NCs ( Figure S11). The structural characterization results (Figures S12-S14) demonstrate that the Cu 2 O NCs undergo the in-situ morphology-preserved reduction into the corresponding Cu NCs with the presence of both Cu(0) and Cu(I) on their surfaces during the WGS reaction up to 548 K." (Please see Lines 16-22 on Page 8).

Comment 5)
The poisoning by formate mechanism is surprising at first because formate should be much more weakly bound than the other listed carbon species. What seems to lead to the poisoning is that the formate is made at the active site and blocks it. Is it possible for the other carbon species to be removed in order to more definitively identify the formate species as the poisoning adsorbate?
Author Reply: We appreciate the reviewer's insightful comments very much and we have seriously considered them. As described and discussed in the Section Reaction mechanism, our comprehensive characterization results shown in Figure 3 and Figure S15 clearly demonstrate the accumulation of formate and hydroxyl intermediates during the WGS reaction up to 548 K on the o-Cu surface but not on the c-Cu and d-Cu surfaces. In connection to the observations that c-Cu and d-Cu are catalytic active while c-Cu is poisoned by a stable hydrogen-containing intermediate on the surface, it is reasonable to conclude that the accumulating formate and hydroxyl intermediates should lead to the self-poisoning and subsequent inactivity of o-Cu NCs. However, it is noteworthy that the accumulation of formate and hydroxyl intermediates on the o-Cu surface can be alleviated at higher reaction temperatures, as the o-Cu NCs were observed to 12 become active in catalyzing the WGS reaction at 623 K and above (shown below). But the catalysts seriously restructure at such high reaction temperatures, thus we did not show these results in the original manuscript. However, we have repeatedly emphasized in the manuscript that our conclusions are applicable for the low-temperature WGS reactions up to 548 K.
In reply to the reviewer, we have added the behaviours of c-Cu and o-Cu in the WGS reaction up to 673 K as Figure  Comment 6) On p. 9 the authors state that the CO oxidation takes place on the Cu -Cu suboxide interface. However the data only shows this oxygen species as being active and there is no 13 evidence for the reaction taking place exclusively at the interface, as opposed to on top of the suboxide. It should be made clearer to the reader that the data presented does not identify if the reaction takes place at the interface or on top of the suboxide regions.
Author Reply: We appreciate the reviewer's insightful comments very much and we have seriously considered them. In the DRIFTS of CO adsorption on Cu NCs at 123 K ( Figure 1F), only vibrational features of CO adsorbed at Cu surfaces were observed. This suggests that CO species in CO oxidation at the Cu-Cu x O interfaces should be CO adsorbed at the Cu surfaces. In reply to the reviewer, we have clarified this issue as the following in the revised manuscript: "Since CO adsorbs on the metallic Cu surface instead of the Cu x O surface ( Figure 1F) (111) surface to accumulate and poison the surface at low temperatures. Our results also demonstrate the Copper facet effect is also applicable for the ZnO/Cu catalysts with ZnO/c-Cu more active than ZnO/o-Cu. However, presently we are not sure the copper species at the Cu-ZnO interface. We are going to work on it although it is very challenging and time-consuming, as the reviewer commented. But our advantage is that we can compare the results of ZnO/c-Cu and ZnO/o-Cu with the same composition but different Cu structures to gain the insights. We hope that we will succeed and present another story in the future.

Comment 8)
A cautionary statement in the kinetics work. Have the authors conducted their tests 14 at steady-state and from high-T to low-T? if not, structural changes at the high temperatures, could have caused lower rates, i.e. an erroneous apparent activation energy, which can explain the values of ~ 30 kJ/mol reported for the Cu/ZnO catalysts. There is no reason for the activation energy to be different, if the same cuprous species is the active site in all such catalysts, i.e. the mechanism does not change. Cu should not be compared to other metals like gold. The latter has its own "signature" with Ea lower than that of Cu (or Pt) on any support.
Author Reply: We appreciate the reviewer's insightful comments very much and we have seriously considered them. As we described, the reported reaction data were all acquired at the steady state. At each temperature, three data were measured at different reaction times and reproducible. As shown in Figure S7, each catalyst was measured for two cycles and the reaction data were also reproducible. We have also done a Weisz-Prater analysis and a Mears analysis proposed by Oyama to confirm the absence of mass and heat transfer limitations under our reaction conditions. And the Arrhenius plots were made using reaction data with CO conversions below 20%. Thus we believe that the reported kinetics in our manuscript is reliable. However, we have accepted the reviewer's suggestion to evaluate the catalytic performance of c-Cu and ZnO/c-Cu from low temperature to high temperature and then from high temperature to low temperature. The results (shown below) also confirm the reliable kinetic measurements. Thus the lower apparent activation energy of ZnO/c-Cu catalysts than of Cu/Zn/Al 2 O 3 and ZnO/o-Cu should be due to the Cu facet effect, although the underlying mechanism needs further studies and is not the topic of the present manuscript. It is likely that the Cu(I) species on Cu(111) could exhibit different reactivity from that on Cu (100); it is also likely that the ZnO on Cu (111)  In reply to the reviewer, we have added the results of activity evaluation of c-Cu and ZnO/c-Cu from low temperature to high temperature and then from high temperature to low temperature respectively as Figure S7B on Page S19 and Figure S26 on Page S43 in the revised supporting information. We have also described the Weisz-Prater analysis and a Mears analysis in the revised supporting information (Please see Page S6-S8). We have also rewritten the relevant texts as the following in the revised manuscript: "The c-Cu and d-Cu NCs are also catalytically stable at 548 K ( Figure 2B and Figure S7), and the SEM images ( Figure S9) demonstrate that the Cu NCs preserve their morphologies after the activity evaluation up to 548 K. Results of a Weisz-Prater analysis and a Mears analysis 26 in Supplemental Information show the absence of mass and heat transfer limitations under our reaction conditions." (Please see Lines 13-17 on Page 7) and "Meanwhile, the ZnO/c-Cu catalyst is stable ( Figure S26)." (Please see the last sentence on Page 15). We have also removed the comparison between our catalysts with other metals. We have also re-ordered the references and figures accordingly.

Comment 9)
The oxidation of flat copper surfaces has been extensively studied. Although the exact structure of Cu under reaction conditions is not known, the design of the DFT model for the Cu-Cu2O interface should resemble some of the motifs found on carefully studied single crystal surfaces (Besenbacher and Norskov, Prog. Surf. Sci. 1993, 44, 5). For example, O on Cu(111) readily reconstructs into ring structures over a wide range of O coverages and temperatures (Matsumoto et al. Surf. Sci., 2001, 471, 225;Soon et al. Phys. Rev. B, 2006, 73, 165424;Yang et al. J. Phys. Chem. C, 2010, 114, 17042), but a structure like the zig-zag pattern that the authors chose as their model has never been observed experimentally. This is unfortunate given the large library of data on partially oxidized Cu surfaces, and dramatically diminishes the impact of the calculations with the structure chosen. The authors should provide some justification to the reader for the design of their DFT model, and should also provide the reader with information of observed O/Cu structures from the literature, including some of the references listed above, and how their structures differ. Minor Concerns: 16 Author Reply: We appreciate the reviewer's insightful comments very much and we have seriously considered them. Actually during our DFT calculations, we built the oxide structure based on ring-like structures, which resemble very much the structures proposed in literatures. However, to simulate the reaction at the interface between Cu and Cu 2 O domain but remain computational affordable for exploration of full reaction network, we use a zig-zag chain representing the edge of Cu2O ring-like structures (Shown below). For simplicity, the lattice constant of Cu 2 O ring-like structure was assumed same with Cu metal underneath.
In reply to the reviewer, we have clarified this issue by showing the whole structures in Figure 5A on Page 33 and rewriting the relevant texts with additional references suggested by the reviewer as the following in the revised manuscript:

"The mechanisms of morphology-dependent catalytic performance of Cu NCs in the WGS reaction were further explored by DFT calculations employing Cu(100) and Cu(111) surfaces with Cu 2 O islands to respectively model the c-Cu and o-Cu surfaces with Cu x O (Figure 5A). According to previous results, 30-33 ring-like structures were adopted for Cu 2 O islands on both Cu(100) and Cu(111) surfaces. This gives a zig-zag chain structure of Cu 2 O at the Cu-Cu 2 O interface that was used during simulations of reactions at the Cu-Cu 2 O interface."
(Please see the Last paragraph on Page 12). We have also re-ordered the references accordingly. We have also described the calculation methods in very detail in the revised supporting information (Please see Page S8-S10).

Comment 10)
Recently there has been a study on Cu single crystals from the Salmeron group showing that the (111) facet has the best activity towards CO oxidation. (Eren et al. J. Phys. Chem. C 2015, 119, 14669) CO oxidation is a crucial step in the water-gas-shift reaction; could the authors comment on the inverse in the activity hierarchy for these two reactions?
Author Reply: We appreciate the reviewer's insightful comments very much. We have read the paper (denoted as JPC paper thereafter) mentioned by the reviewer very carefully. In the JPC paper the reaction of CO with chemisorbed oxygen on three low-index faces of copper was studied and it was found that on Cu(111) the rate was one order of magnitude faster than that on Cu(100) and two orders of magnitude faster than that on Cu(110). The story is very different from ours. Firstly, only CO+O(a) surface reaction was studied in the paper while we studied the catalytic WGS reaction with a complex surface reaction network. Such a surface reaction rate can not simply represent the catalytic activity. Actually it was reported that among Cu, Cu 2 O and CuO, Cu exhibits excellent initial activity in catalyzing CO oxidation, but then facilely deactivates (ChemCatChem, 2011, 3, 24); Secondly, chemisorbed oxygen species on Cu surfaces were studied in the paper while the suboxide species was the active species in our case. Our DFT calculation results ( (0.56 eV), similar to the CO+O(a) surface reaction reported in the JPC paper. However, in the WGS reaction at the Cu-Cu 2 O interface, the H 2 production is the rate limiting step.
Comment 11) I suggest a modification to the title from "the active Cu structure…" to "the most active Cu facet…" because all Cu structures investigated are indeed active, this work is aimed towards finding the most active facets (as opposed to general structure such as nanoparticle size or shape).
Author Reply: We appreciate the reviewer's nice suggestion and take the pleasure to accept it. The tile of our revised manuscript is "The Most Active Cu Facet for Low-Temperature Water Gas Shift Reaction".

Comment 12)
The authors may want to consider re-labeling of the "suboxide" species, because it suggests that the oxide is subsurface. However the authors find that there is O on the surface at these suboxide regions, so perhaps calling these sites simply "oxide" would be more fitting.
Author Reply: We appreciate the reviewer's kind suggestion very much and we have seriously considered them. The species labelled as copper suboxide in our manuscript exhibits a Cu(I) feature and an O feature with an O 1s binding energy at 530.3 eV. The O 1s feature of the suboxide species is very different from that in copper oxides (Cu 2 O and CuO). Such a species was previously observed and labelled as copper suboxide in Ref. 14 cited in the manuscript and the paper of Schedel-Niedrig et al. Partial methanol oxidation over copper: active sites observed by means of in situ X-ray absorption spectroscopy. Phys. Chem. Chem. Phys. 2, 3473-3481 (2000), and we then followed the references labelling. The suboxide does not mean subsurface oxide, but means its Cu:O stoichiometric is much larger than copper oxides. Thus we consider that it is appropriate to use copper suboxide instead of copper oxides. In reply to the reviewer's comments, we have clarified this issue by defining the suboxide and cited Phys. Chem. Chem. Phys. 2, 3473-3481 (2000) as Ref. 13 in the revised manuscript as the following: "Cu suboxide (Cu x O, x≥10)" (Please see Line 8 on Page 2 and Line 20 on Page 5). We have also re-ordered the references accordingly. 18

Authors' Reply to Reviewer 3's Comments and Revisions
Comment: This manuscript describes a promising approach to determine the active site structure of Cu-based catalysts for the low temperature water-gas shift (WGS) reaction. The information is then used to prepare an improved catalyst sample. The WGS reaction is of great industrial importance and the currently used catalyst, Cu/ZnO/Al2O2 is also Cu-based; thus, this topic is of high relevance and could be appealing to a large audience.
Despite the potential impact of the work, there are several gaps in the argumentation. These gaps are described below and require major revisions before this manuscript can be reviewed again.
Author Reply: We appreciate the reviewer's positive and valuable comments very much. We have seriously considered the reviewer's comments and revised the manuscript accordingly. We hope that the revised manuscript will be suitable for the publication in Nature Communications.

Comment:
The interpretation of the TPRS data on page 7 leads the authors to claim that H 2 desorption is the rate-limiting step of Cu-catalyzed WGS. I don't see sufficient evidence for this conclusion. If the active site is the Cu/Cu-suboxide interface, CO 2 desorption could simply be a result of Cu-suboxide reduction by CO. I assume similar CO 2 desorption peaks would be observed in the absence of H 2 O. Also, if H 2 desorption was rate-limiting, why did the authors not simulate this step with DFT? The adsorption/desorption of H 2 in Table S5 is assumed to be quasi-equilibrated and no barrier is reported.
Author Reply: We appreciate the reviewer's insightful comments very much and we have seriously considered them. In our manuscript, we discussed that, on the basis of TPR results, the H 2 production, not the H 2 desorption commented by the reviewer, is the rate-limiting step of Cu-catalyzed WGS reaction. The barrier of H Cu +H Cu →H 2 was calculated to be 0.90 and 0.85 eV at the Cu 2 O-Cu interface respectively on Cu(111) and Cu(100). It can be found from the data listed in Tables S3-S5,  As demonstrated in the XPS results, Cu suboxide and hydroxyl groups exist at the suboxide-Cu interface on the Cu surface subjected to the WGS reaction, thus the CO 2 production can be due to both the reduction of suboxide by CO and the CO oxidation by hydroxyl groups, as demonstrated by the calculated three reaction mechanisms. We carried the experiments of the CO reduction of Cu NCs activated by water decomposition at 250 °C, and the acquired CO 2 production is similar to those in the CO+H 2 O TPRS on Cu NCs but with much reduced intensities (shown below). However, we do not consider that the CO reduction results of Cu NCs activated by water decomposition at 250 °C are important and thus decide not to include them in the manuscript.

19
In reply to the reviewer's comments, we have included the calculated barriers of H Cu +H Cu →H 2 at the Cu 2 O-Cu interface respectively on Cu (111) and Cu (100) in Tables S3-S5 in the revised supporting information and clarified the issue as the following in the revised manuscript: " Figure  3A presents TPRS spectra of WGS reaction over the Cu NCs. The CO 2 production was observed to occur prior to the H 2 production, suggesting the H 2 production as the rate-limiting step in the Cu-catalyzed WGS reaction." (Please see Lines 4-6 on Page 9) and "In all three pathways the elementary surface reaction with the largest activation energy is the reaction of O Cu2O H → O Cu2O + H Cu involved for the H 2 production. This agrees with the experimental observations that the H 2 production is the rate-limiting step in the Cu-catalyzed WGS reaction ( Figure 3A)." (Please see the texts between Line 21 on Page 13 and Line 2 on Page 14).
Comment: Ultimately, the authors attribute WGS reactivity to the presence of Cu/Cu-suboxide interfaces. With this active site assignment, many of the performance data are questionable. The BET surface area would not be a good way to normalize the TOF. Instead, the authors should normalize by the number of interface sites. The observed differences may simply be a difference in how well the different Cu facets oxidize to a suboxide, or how well the epitaxial relationship between the Cu metal and the suboxide matches. It is likely that the thermodynamically most stable (111) facet is not easily oxidized. From the DFT results, the authors could easily extract the interfacial energy between Cu(111) or Cu(100) and Cu 2 O, and determine if there is a significant stability difference.
Author Reply: We appreciate the reviewer's insightful comments very much and we have seriously considered them. We definitely agree with the reviewer that the catalytic activity should be expressed by the reaction rate normalized to the number of active Cu suboxide-Cu interface site. However, it is a very challenging task and even could be an impossible task to identify the number of active interface site on powder samples. Thus it is usually approximated by the reaction rate normalized to the catalyst mass and more accurately the reaction rate normalized to the number of surface sites. In our case we used the reaction rate normalized to the number of surface Cu sites. The difference in the activity can be demonstrated more evidently using apparent activation energy than using the normalized reaction rate. Our results of apparent activation energy ( Figure 2C) (111), respectively (shown below). Thermodynamically more stable Cu(111) interacts more weakly with oxide film, in line with the larger resistance toward oxidation as pointed by the reviewer. However, since there results are not related with our experimental results much, we considered not to include them in the manuscript.
Comment: Side note: a brief clarification why the BET surface area of larger o-Cu NCs (937 nm) is larger than that of smaller c-Cu NCs (877 nm) would be appropriate. It is likely due to the higher site density of the (111) planes.
Author Reply: We appreciate the reviewer's insightful comments very much and we have seriously considered them. We considered two likely factors giving larger specific BET surface areas of the o-Cu 2 O and o-Cu NCs with a larger average edge length respectively than the c-Cu 2 O and c-Cu NCs. One is the morphology effect: the surface area-to-volume ratios of an octahedron and a cube can be geometrically calculated respectively as 3√6/a and 6/a (a being the edge length); the other is the facet effect: the surface atom density is higher on the {111} facets exposed on the octahedral NCs than on the {100} facets exposed on the cubic NCs for both Cu 2 O and Cu with cubic phases.

21
In reply to the reviewer, we have clarified this issue as the following in the revised manuscript: "Although with a larger average edge length, the o-Cu 2 O and o-Cu NCs exhibit larger specific BET surface areas respectively than the c-Cu 2 O and c-Cu NCs due to the morphology and facet effects. The surface area-to-volume ratios of an octahedron and a cube can be geometrically calculated respectively as 3√ /a and 6/a (a being the edge length); meanwhile, the surface atom density is higher on the {111} facets exposed on the octahedral NCs than on the {100} facets exposed on the cubic NCs for both Cu 2 O and Cu with cubic phases." (Please see Lines 4-10 on Page 5).

Comment:
The high surface roughness is a concern. A rough surface will necessarily have defect and step sites. Were there any efforts to quantify the amount of surface defects? The CO vibration experiments should exhibit signals of CO molecules strongly adsorbed on steps/defects. Did the authors try to compare the signal intensities to get a rough estimate of the ratio of step to terrace sites?  (2000)). In Figure 1F, the absence of vibrational features above 2100 cm -1 suggests the low density of stepped sites and defects on the surfaces of all acquired Cu NCs. In reply to the reviewer, we have clarified this issue by discussing as the following with citing J. Phys. Chem. B 104, 6001-6011 (2000) as Ref. 25 in the revised manuscript: "The absence of vibrational features above 2100 cm -1 that correspond to CO adsorbed at stepped sites and defects on Cu surfaces 26 also suggests the low density of stepped sites and defects on the surfaces of all acquired Cu NCs." (Please see Lines 18-21 on page 6). We have also re-ordered the references accordingly.

Comment:
The DFT results suggest that there is no significant difference between Cu(111) and Cu(100) model interfaces with Cu 2 O present. The only difference was speculated to be the formate coverage based on the formation and decomposition barriers. Such speculation is rather unreliable and often fails, even in simple cases. The coverage of formate depends on the coverage of H and CO 2 , which in turn depend on everything else. Also, if formate is a spectator species, it will be formed until it reached its saturation coverage. At this point the formation and decomposition barriers are no longer relevant to estimate the formate coverage at the interface under reaction conditions. Hence, I see value in the DFT work to assign the active site to the Cu/Cu 2 O interface, but I do not see agreement with the experimentally observed activity increase on the (100) facet. In reply to the reviewer, we have made the revisions in the revised manuscript as the following: 1) We have moved the Cu LMM AES and O 1s XPS results of water activation on Cu NCs from the supporting information to the revised manuscript as Figure 4 on Page 32 in order to demonstrate the facile water activation on c-Cu than on d-Cu and o-Cu and to provide experimental results in accordance with the DFT calculation results shown in Figure 5. We have re-ordered all figures in the revised manuscript. 2) We have described how the DFT calculation results agree with the experimental results in the contexts of the revised manuscript as the following: "Thus the water activation is much more facile at the Cu 2 O-Cu interface than at the Cu surface, and the water dissociation at the Cu-Cu 2 O interface into OH Cu and O Cu2O H should proceed more facilely on Cu(100) than on Cu(111). These agree with the above experimental observations (Figure 4)." (Please see Lines 11-14 on page 13); "All three pathways can proceed at the Cu-Cu 2 O interface of both Cu(100) and Cu(111) surfaces with similar largest activation energies of around 1 eV. In all three pathways the elementary surface reaction with the largest activation energy is the reaction of O Cu2O H → O Cu2O + H Cu involved for the H 2 production. This agrees with the experimental observations that the H 2 production is the rate-limiting step in the Cu-catalyzed WGS reaction ( Figure 3A)." (Please see the texts between Line 20 on Page 13 and Line 2 on Page 14); "Thus, it can be expected that the formate species will accumulate at the Cu-Cu 2 O interface of Cu(111) surface due to the large barrier for the HCOO Cu2O decomposition during the low-temperature WGS reaction but not at the Cu-Cu 2 O interface of Cu(100) surface, which will eventually block the active Cu-Cu 2 O interface of Cu(111) surface. This is consistent with the experimental observations that the c-Cu catalyst is active in catalyzing the low-temperature WGS reaction up to 548 K while the o-Cu catalyst is not and is with accumulated formate species on the surface (Figures 2 and 3)." (Please see Lines 11-18 on page 14).

Comment:
Were the transition states found in the NEB method verified by vibrational analysis showing a single imaginary mode? I would be beneficial if the authors clearly stated that zero point energies and entropy corrections were neglected.
Author Reply: We appreciate the reviewer's kind suggestions very much and accept it. We have clarified the issue as the following in the revised supporting information: "The transition states were verified by vibrational analysis showing a single imaginary mode. Zero point energies and entropy corrections were neglected in DFT calculations." (Please see Lines 10-12 on Page S10).

Comment:
Could the authors please clarify if the reaction energies and barriers are taken with respect to isolated reactants/products? The equation in the SI suggests this but an explicit statement is preferred. If energies are reported at infinite separation, could the authors also include the interaction energies for the co-adsorbed states that form the initial and final states of their NEB calculations?
Author Reply: We appreciate the reviewer's kind suggestions very much and have accept it. We have clarified the issue as the following in the revised supporting information: "The activation energies of elementary reactions (Ea) and reaction energies of elementary reactions (E r ) were taken with respect to isolated reactants/products." (Please see Lines 12-14 on Page S10). In order to address the reviewer's comments on the interaction energy, four typical elementary surface reactions were selected to calculate the interaction energy between isolated species and co-adsorbed states. The results are given in the Table below. It can be found that most of them have modest interaction energies (less than 0.09 eV), and therefore can be neglected. The small interaction might come from the relative large supercell (2×4) used in present work. Large interaction energies were observed for water dissociation at Cu 2 O-Cu interfaces, being about 0.47 eV for Cu 2 O-Cu(111) and 0.41 eV for Cu 2 O-Cu(100). This comes from the cost of hydrogen bond between dissociated fragments at the co-adsorbed states. In reply to the reviewer, we have clarified this issue in the revised supporting information (Please  see the texts and Table between Line 14 on Page S10 and Line 13 on Page S11).

Comment:
The manuscript must be edited for grammar and spelling. Most importantly "morphology-reserved" should probably be "morphology-preserved".
Author Reply: We appreciate the reviewer's careful reading and kind suggestions very much and have replaced "reserved" with "preserved" throughout the revised manuscript and supporting information. We have also done our best to polish the English of the revised manuscript and supporting information.
catalysts. However, a very preliminary observation indicates that surface oxidation extent of Cu in Cu-ZnO-based catalysts in the WGS reaction should depend on the Cu particle size.
In the revised manuscript, we have rewritten the relevant texts as the following: "It can be seen that both Cu(0) and Cu(I) exist on the surface of Cu NCs in ZnO/Cu catalysts. The ZnO/Cu NCs catalysts are much more active than the corresponding Cu NCs catalysts in catalyzing the WGS reaction ( Figures 7E and S25E), and the calculated apparent activation energy of ZnO/c-Cu (32.4±0.8 kJmol -1 ) ( Figure 7F) and ZnO/o-Cu (55.9±3.9 kJmol -1 ) ( Figure S25F) catalysts are much smaller than the corresponding Cu NCs catalysts. These results suggest that the copper-ZnO interface in ZnO/Cu NCs catalysts exhibits much higher intrinsic activity than the Cu-CuxO interface in Cu NCs catalysts. However, the detailed mechanism needs further study. Meanwhile, the ZnO/c-Cu catalyst is stable ( Figure S26)." (please see Line 19 on Page 15 to Line 5 on Page 16) and "The ZnO/o-Cu catalyst exhibits a similar apparent activation energy to the commercial Cu/ZnO/Al2O3 catalyst but much larger than that of ZnO/c-Cu catalyst. Therefore, the active structure of ZnO/c-Cu catalyst should be intrinsically more active than those of ZnO/o-Cu and commercial Cu/ZnO/Al2O3 catalysts. This demonstrates that the Cu structure in the Cu-ZnO based catalysts plays a key role in determining the catalytic activity in the WGS reaction and the ZnO/c-Cu catalyst with the Cu{100} structures is a highly efficient catalyst. Meanwhile, it could also be inferred that the copper structure of commercial Cu/ZnO/Al2O3 catalyst should be dominated by the Cu{111} structures and its catalytic activity can be improved by engineering the Cu structure from the dominant Cu{111} structure into the Cu{100} structure." (please see Lines 13-23 on Page 16). We believe such description and discussion are appropriate.