Modeling CoTAPc on CNTBuilding on our previous work, graphitized CNT (GCNT) was employed as the catalyst substrate to reduce the substrate defect level and suppress undesirable hydrogen evolution reaction (HER)34. As shown in Fig. 1a, the highly crystalline graphene layers, as revealed in the high-resolution transmission electron microscope (HRTEM) image of CoTAPc immobilized on GCNT (CoTAPc/GCNT), clearly indicate its higher graphitization degree and reduced level of defects. This observation is corroborated by the Raman spectra of the GCNT before and after loading with the CoTAPc. As shown in Fig. 1b, compared to the non-graphitized CNT (nGCNT) as purchased and purified, the much lower D band intensity of the GCNT manifests its significantly reduced defect level. More physical characterizations in Supplementary Fig. 2 further validate that carbon-based defects are the primary source of defects on CNTs. Negligible influence on the defect level of GCNT is observed upon CoTAPc immobilization. Notably, the high-angle annular dark-field (HAADF) scanning transmission electron microscope (STEM) from aberration-corrected (AC)-HRTEM images (Fig. 1c and Supplementary Fig. 3) reveal that CoTAPc tends to adsorb on the defect-rich domain (i.e., intersections) rather than the highly graphitized surface, suggesting that CoTAPc prefers to adsorb on the CNT via catalyst/defect interactions rather than the generally believed π-π interactions20,35,36,37. A similar observation was made when decreasing the loading of CoTAPc on GCNT (Supplementary Fig. 4), confirming that highly graphitized CNT surfaces exhibit reduced adsorption of CoTAPc molecules due to the reduced defect level. In contrast, CoTAPc immobilized on nGCNT (CoTAPc/nGCNT), as shown in the HAADF-STEM image in Supplementary Fig. 5, exhibits much more uniform dispersion of CoTAPc single sites on nGCNT. This is attributed to the high degree of defects present on nGCNT (Fig. 1b). Such dispersions resemble to those reported in other literature18,21,37, further confirming that CoTAPc tends to adsorb on defect-rich domain, akin to our previous observations with NiTAPc34. Previously, we also found that dipole interaction between the metal center in Ni-Pc and single vacancy is substantially stronger than conventional π-π interaction, providing a reasonable description of the Pc molecule adsorption on the CNT surface. Inspired by this, we established a catalyst model consisting of CoTAPc absorption on a single vacancy on graphene (1V-Gr) to investigate the Co-N4 plane configurations. Intriguingly, two distinct CoTAPc@1V-Gr configurations were identified. As shown in Fig. 1d, the quasi-flat Co-N4 configuration (CoTAPc@1V-Gr-flat, Supplementary Data 1) resembles those reported in the literature using perfect graphene as the substrate37,38, where the Co center is slightly out of the Pc plane, with a N−Co−N angle of ~164°. Besides, a distorted Co-N4 configuration (CoTAPc@1V-Gr-dis, Fig. 1e and Supplementary Data 2) was also observed, wherein the Co center exhibits significant out-of-plane distortion. Notably, the energy of the distorted configuration Co-N4 is 0.35 eV lower than that of the quasi-flat configuration (Fig. 1f), indicating its favored thermodynamic stability. We then compared the adsorption energy of these two CoTAPc configurations on 1V-Gr with that of CoTAPc on pristine graphene (CoTAPc@Gr, Supplementary Fig. 6). As shown in Supplementary Fig. 6b, the formation of CoTAPc@1V-Gr-flat is substantially favored (~0.58 eV) than CoTAPc@Gr. This result further validates the stronger interaction between CoTAPc and defects on CNT rather than π-π interactions. Similar observations were made for the pristine CoPc/CNTs (Supplementary Fig. 7), indicating that this CoPc/defect interaction is likely applicable to the adsorption of other planner molecules on CNTs.Fig. 1: Characterization and modeling of CoTAPc on GCNT.a HRTEM image of CoTAPc/GCNT, b Raman spectra of CoTAPc/GCNT, GCNT, and nGCNT, the intensity of G band is normalized for comparison; c HAADF-STEM image of CoTAPc/GCNT. d,e top-view and side-view of CoTAPc@1V-Gr-flat and CoTAPc@1V-Gr-dis, respectively; f differences in energies of CoTAPc@1V-Gr-flat (0 eV) and CoTAPc@1V-Gr-dis (−0.35 eV); g dipole moment of CoTAPc@1V-Gr-dis and CoPc@1V-Gr-dis; h,i top-view and side-view of adsorbed CO configurations on CoTAPc@1V-Gr-flat and CoTAPc@1V-Gr-dis, respectively; j comparison of CO and CO2 adsorption energies on two configurations, as well as corresponding C − O bond lengths of the adsorbed *CO. The calculated configurations and energies are provided in Source Data file.To explore the dynamics of the Co-N4 center on CoTAPc/GCNT under cathodic potentials, dipole moments of CoTAPc@1V-Gr-dis and CoPc@1V-Gr-dis were calculated to be 0.224 and 1.512 Debye towards the Pc plane, respectively (Fig. 1g). Molecules with larger dipole moment are in general have stronger interaction with electric field due to electrostatic interaction. With this result, we anticipate that the external electric field (i.e., cathodic potential) would pull the Co center towards the CNT substrate, leading to out-of-plane distortion, especially under more negative potentials. Therefore, we believe that potential-driven distortion occurs during electrocatalytic CO2R/COR, with the degree of distortion associated with the cathodic potential applied. This is confirmed by our in-situ XAFS measurements, vide infra. Furthermore, the much larger dipole moment of CoPc@1V-Gr-dis compared to that of CoTAPc@1V-Gr-dis suggests that CoPc@GCNT is likely prone to more severe out-of-plane distortion under cathodic overpotentials. This would compromise its stability, in good accordance with our experimental findings (Supplementary Fig. 8) and reports elsewhere18.We then investigated the corresponding *CO adsorption on the two configurations. As shown in Fig. 1h–i, *CO adsorbs on CoTAPc@1V-Gr-flat linearly (linear *CO, denoted as *COL, Supplementary Data 3) on top of the Co-site, such *CO binding configuration are often used in recent literature, however without considering the catalyst/substrate interactions20,21,22. In contrast, with the CoTAPc@1V-Gr-dis, we found *CO tends to bridge (bridge *CO, denoted as *COB, Supplementary Data 4) on the two adjacent pyrrole N atoms within the Co-N4 center. The *CO adsorption energies of these two binding modes are −0.52 eV and −1.80 eV, respectively. Additionally, the C − O bond length in *CO increases from 1.15 Å (*COL) to 1.21 Å (*COB) (Fig. 1j), indicating higher reactivity towards further hydrogenation steps for methanol production. In contrast, the adsorption energies of CO2 on these two catalyst configurations exhibit little difference and are much lower than that of CO (Fig. 1j), indicating that the low \({{\mbox{FE}}}_{{{{\rm{CH}}}}_{3}{{\rm{OH}}}}\) during CO2R is less likely resulted from competition of CO2 adsorption with CO. We also assessed the two *CO adsorption configurations on CoPc@1V-Gr (Supplementary Fig. 7d, e). Likewise, we observed four-folders of enhancement in *CO adsorption energy on CoPc@1V-Gr-dis compared to CoPc@1V-Gr-flat, together with prolonged C − O bond length. This result suggests that this behavior is likely universal among different Co-Pc/CNT systems. Furthermore, we calculated the vibration frequency of the C − O bond in these two configurations of CoTAPc. The calculated frequencies for the *COL and * COB are 2075 cm−1 and 1743 cm−1, respectively. Note that, despite the thermodynamically favored energetics of CoTAPc@1V-Gr-dis, the actual configuration of CoTAPc under cathodic potential is likely determined by the equilibrium between quasi-flat and distortion states. This equilibrium is influenced by the energy barrier associated with the distortion of the CoTAPc plane, which may be significant due to the large planar nature of the Pc molecule. Therefore, we believe that the above two *CO binding states may coexist during CO2R/COR, and the prevalence of linear or bridged *CO is dictated by the applied cathodic potential. Overall, our DFT results suggest that the strong interaction between the Co center and CNT defects will likely lead to the potential-driven distortion of the Co-N4 center. This distortion strengthens the adsorption of *CO and weakens the C − O bond, facilitating its hydrogenation and ultimately promoting the production of CH3OH.Electrochemical CO2R/COR on CoTAPc/GCNTThe CO2R/COR performances on CoTAPc/GCNT were assessed using a gas diffusion electrode-based flow cell (Supplementary Fig. 9). We first compared the COR performance of CoTAPc/GCNT and CoTAPc/nGCNT. As shown in Supplementary Fig. 10, we observed substantially higher selectivity and activity of CH3OH production with the former, along with suppressed HER, further validating the significance of optimizing the CNT defect level for catalysis. We then conducted both CO2R and COR in three electrolytes with different bulk pH: 1 M KOH (pH 14), 1 M KHCO3 (pH 8.4), and 0.5 M K2SO4 (pH 2, adjusted by H2SO4). As shown in Fig. 2a–c, Supplementary Fig. 11 and Supplementary Table 1, 2, within a broad current density window, \({{\mbox{FE}}}_{{{{\rm{CH}}}}_{3}{{\rm{OH}}}}\) in COR substantially outperforms those in CO2R in all three electrolytes. Notably, the maximum \({{\mbox{FE}}}_{{{{\rm{CH}}}}_{3}{{\rm{OH}}}}\) in COR at 200 mA cm−2 in alkaline, neutral, and acidic electrolyte reached 47%, 53%, and 45%, respectively. In contrast, the maximum \({{\mbox{FE}}}_{{{{\rm{CH}}}}_{3}{{\rm{OH}}}}\) in CO2R attained at 800 mA cm−2 in alkaline, neutral and acidic electrolyte are only 12.8%, 10.5%, and 12.1%, respectively. However, the CH3OH production rates (i.e., partial current density) in the aforementioned scenarios are relatively similar. What merits more attention is that the cathodic potential applied (at the absolute potential scale, i.e., standard hydrogen electrode, SHE) for COR and CO2R to achieve these CH3OH production rates are also close, validating the critical role of electrode potential in CH3OH production for both COR and CO2R. In other words, this discrepancy in \({{\mbox{FE}}}_{{{{\rm{CH}}}}_{3}{{\rm{OH}}}}\) between COR and CO2R at identical overall current density is a result of different applied potential (Supplementary Fig. 12 and Supplementary Table 3). A larger cathodic potential affords a higher \({{{{\rm{FE}}}}}_{{{{{\rm{CH}}}}}_{3}{{{\rm{OH}}}}}\). To elucidate this correlation further, we plotted the CH3OH production rate for CO2R/COR at different current densities on the function of the applied cathodic potential on the SHE scale. As illustrated in Fig. 2d, the CH3OH production rates of CO2R and COR at similar potentials are, as anticipated, similar in both alkaline and neutral electrolytes. This further confirms the previously mentioned potential-determined CH3OH production in both COR and CO2R. Note, that similar trends are less apparent in the acidic electrolyte. We believe that COR would be further facilitated than CO2R due to the relatively lower local pH39.Fig. 2: Electrocatalytic CO2R/COR performance of CoTAPc/GCNT conducted at room temperature.a–c \({{\mbox{FE}}}_{{{\mbox{CH}}}_{3}{\mbox{OH}}}\) and cathodic potential of CO2R/COR at different current densities in electrolytes with various of pH: 1 M KOH (pH = 14), 1 M KHCO3 (pH = 8.4) and 0.5 M K2SO4 (pH = 2, adjusted by H2SO4), respectively; d correlations between CH3OH production rate and cathodic potential in different electrolytes; e applied cathodic potential and f resulted 13CH3OH/12CH3OH distributions for the 13CO2/12CO co-feeding electrolysis, in different electrolytes. Solid and dash columns indicate 12CH3OH and 13CH3OH, respectively; g production rates of CH3OH and H2 in CO2/CO co-feeding electrolysis in different electrolytes. Solid and dash columns denote CH3OH and H2, respectively. Note: All co-feeding experiments were executed at 500 mA cm−2. h Summary of \({{\mbox{FE}}}_{{{\mbox{CH}}}_{3}{\mbox{OH}}}\) and \({j}_{{{\mbox{CH}}}_{3}{\mbox{OH}}}\) of CO2R/COR for CoTAPc/GCNT. Catalyst loading amount is 1 mg cm−2. 100% iR compensation was applied for all potentials. Typical solution resistance for 1 M KOH, 1 M KHCO3 and 0.5 M K2SO4 were measured to be 1.1 ± 0.25 Ω, 1.5 ± 0.35 Ω and 1.4 ± 0.30 Ω, respectively. The error bars represent standard deviations from three independent measurements. Relevant source data are provided in the Source Data file.Modulating the density of surface-active sites on the electrode can significantly influence the operating potential in electrocatalytic CO2R/COR, and subsequently impact the CH3OH selectivity and productivity. To explore this effect, we reduced the CoTAPc loading on GCNT by adjusting the precursors weight ratio (CoTAPc: GCNT) from the previous 1:10 to 1:20 and 1:40. We confirmed that the decreased density of the surface-active sites aligns closely with the CoTAPc loading (Supplementary Fig. 13) by inductively coupled plasma optical emission spectrometry (ICP-OES, Supplementary Table 4). Subsequently, the CO2R/COR performance of CoTAPc/GCNT-1:20 and 1:40 was assessed under the same conditions (Supplementary Fig. 14). As anticipated, the overpotentials for both CO2R and COR increased for the CoTAPc/GCNT with reduced catalyst loadings. Regarding the CoTAPc/GCNT-1:40 sample, the electrocatalysis even failed at high current densities due to the large overpotential applied. Nevertheless, similar potential dependent trends on CH3OH production were observed with CoTAPc/GCNT-1:20 (Supplementary Fig. 15). Taken together, we can conclude that the intrinsic activity of CoTAPc in COR/CO2R for CH3OH production is comparable under the same applied potential. We believe that sufficiently high overpotentials are likely required for further reducing CO to form CH3OH, owning the strong binding of CO on the distorted CoTAPc (Fig. 1j). However, lower overpotential is often applied for CO2R to achieve the same current density compared to COR, due to relatively easy CO2 to CO conversion step. Consequently, the lower overpotential of CO2R will lead to lower selectivity towards CH3OH compared to that of COR. As suggested by the DFT calculations above, we attribute the origin of the potential-dependent effect to the potential-induced structural evolution of the Co-N4 center, specifically the distortion of the Co-N4 plane.We conducted electrolysis by co-feeding isotopically labeled CO2/CO to further validate our conclusion (Supplementary Fig. 16). As shown in Fig. 2e, f we present the cathodic potential and the corresponding CH3OH production rates obtained from the isotopically labeled 12CO/13CO2 co-feeding electrolysis at 500 mA cm−2. As anticipated, the overpotential decreased noticeably with an increase in CO2 partial pressure (\({p}_{{{\mbox{CO}}}_{2}}\)), leading to a decrease in CH3OH production. At a low current density, such as 200 mA cm−2, the corresponding small overpotential is likely not sufficient for activating *CO on the active sites resulting in diminished CH3OH production at low pCO (Supplementary Fig. 17). These observations are consistent with the discussions above. Notably, we also found that the majority of the CH3OH was originated from CO rather than CO2 in the co-feeding electrolysis, even at a low CO to CO2 ratio of 1:3 (Fig. 2f). This observation agrees well with the above DFT calculations, where we found that CO exhibited substantially stronger adsorption energy compared to CO2 on CoTAPc (Fig. 1j). Furthermore, as depicted in Fig. 2g, the production rates of CH3OH and H2 surprisingly track each other in both COR to CO2R, indicating that these two pathways are associated, at least to some extent. Future work in decoupling these two pathways is crucial for achieving improved selectivity towards CH3OH production. The Ar/CO co-feeding electrolysis was also carried out. As shown in Supplementary Fig. 18, \({{\mbox{FE}}}_{{{{\rm{CH}}}}_{3}{{\rm{OH}}}}\) at 100 mA cm−2 remains relatively stable when decreasing the pCO to 50%, indicating the strong binding of the *CO on the active-sites. However, further reduction in CO partial pressure or increase in current density leads to a noticeable decrease in \({{{{\rm{FE}}}}}_{{{{{\rm{CH}}}}}_{3}{{{\rm{OH}}}}}\), likely due to the insufficient mass diffusion of CO. Finally, we summarized the \({{\mbox{FE}}}_{{{\mbox{CH}}}_{3}{\mbox{OH}}}\) and partial current densities \(({j}_{{{{{\rm{CH}}}}}_{3}{{{\rm{OH}}}}})\) for CH3OH production on CoTAPc/GCNT in Fig. 2h. We have achieved high \({j}_{{{{{\rm{CH}}}}}_{3}{{{\rm{OH}}}}}\) in both CO2R and COR, along with high \({{{{\rm{FE}}}}}_{{{{{\rm{CH}}}}}_{3}{{{\rm{OH}}}}}\), validating our design strategies elucidated above. Overall, the results obtained from these co-feeding measurements further elucidated the effect of cathodic potential on CH3OH production in both CO2R/COR, which likely resulted from the potential-induced distortion of the Co-N4 plane.Exploring the Coordination Environment of CoTAPc during CO2R/CORWe utilized X-ray absorption spectroscopy (XAS) technique to investigate the coordination environment of CoTAPc/GCNT, particularly under catalytic conditions40. Figure 3a shows the ex-situ X-ray absorption near-edge structure (XANES) spectra of CoTAPc and CoTAPc/GCNT in CO atmosphere and vacuum, respectively. Note, the spectral features labeled as a and b can be attributed to 1 s→4px, y transition, and dipole-allowed 1 s→4pz shakeup transition, respectively41,42,43,44,45. In the case of CoTAPc, the feature a shifts to higher energy, and the intensity of feature b increases when switching the testing environment from CO to vacuum, indicating the transformation towards planar D4h structure46. In contrast, CoPc exhibits characteristic moieties of a planar D4h structure in both CO atmosphere and vacuum (Supplementary Fig. 19a). This result suggests that CoTAPc tends to bind CO stronger compared to CoPc under ambient conditions, aligning with our DFT calculation (Fig. 1j and Supplementary Fig. 7). Simulation of XANES based on the corresponding DFT models were carried out to further verify our hypothesis. As shown in Fig. 3b, the simulated XANES of CoTAPc-flat exhibits diminished peak b but similar structure a at the white line region. However, *CO-CoTAPc-flat shows a very different white line structure compared with the former but similar to what we observed of CoTAPc in CO. Therefore, it is reasonable to believe that the change in white line feature a resulted from the axial CO ligand. Moreover, the simulated XANES also aligns with the proposed configurations of the immobilized CoTAPc shown in Fig. 1. As shown in Fig. 3c, both CoTAPc-flat and CoTAPc-dis exhibit an increase in the intensity of the pre-edge c at approximately 7710 eV, although the changes in CoTAPc-dis is substantially larger. Since this pre-edge is associated with the quadrupole-allowed 1 s → 3d − 4p transition, a characteristic descriptor of square-planar D4h symmetry46,47,48, we can conclude that D4h symmetry of the CoTAPc-dis is diminished more significantly compared to that of CoTAPc-flat, pinpointing the strong interactions between the CoTAPc plane and the defects on CNT substrate.Fig. 3: XAFS simulations and characterizations of the CoTAPc and CoPc composites.a Ex-situ XANES of CoTAPc and CoTAPc/GCNT in CO atmosphere and vacuum, b simulated XANES of CoTAPc, CoTAPc@1V-Gr-flat (denoted as CoTAPc-flat) and *CO-CoTAPc@1V-Gr-flat (denoted as *CO-CoTAPc-flat), c simulated XANES of CoTAPc, CoTAPc-flat, and CoTAPc@1V-Gr-dis (denoted as CoTAPc-dis); d in-situ XANES of CoTAPc/GCNT in CO2R under different potentials, e in-situ EXAFS of CoTAPc/GCNT at −0.36 V in R-space, f in-situ XANES of CoPc/GCNT in CO2R under different potentials; g–i in-situ XANES of CoTAPc/GCNT in COR under different potentials in 1 M KOH, 1 M KHCO3 and 0.5 M K2SO4 (pH=2), respectively. For CO2R, 1 M KOH is used as electrolyte. Catalyst loading amount is ~2.3 mg cm−2. 100% iR compensation was applied for all potentials. Typical solution resistance for 1 M KOH, 1 M KHCO3 and 0.5 M K2SO4 were measured to be 2.5 Ω, 3.5 Ω and 3.3 Ω, respectively. The larger solution resistance results from different cell setup and catalyst loading. Source data for XANES and EXAFS are provided in the Source Data file.We believe the CO adsorption on solid CoTAPc is a slow process under ambient conditions, as evident in the XANES and EXAFS spectra of CoTAPc after exposure to a CO atmosphere (Supplementary Fig. 20), which closely resembles that in vacuum. Upon immobilization of CoTAPc on GCNT, the intensity of shoulder peak b and the magnitude of peak a both decrease, indicating a further diminishment of the D4h planar configuration of the Pc ring. This observation is consistent with the simulated CoTAPc-flat. Besides, the XANES of CoTAPc/GCNT in a vacuum resembles that in CO and does not fully recover the flat configuration as the pristine CoPc, further confirming the interactions between CoTAPc and GCNT. The extended X-ray absorption fine structure (EXAFS) in R-space, shown in Supplementary Fig. 19b, reveals that both CoTAPc and CoTAPc/GCNT exhibit reduced coordination numbers of Co−N compared with their CoPc counterparts, suggesting the possible existence of adsorbed CO as an axial ligand. The EXAFS fitting results (Supplementary Fig. 19c) further verify our hypothesis. Furthermore, samples with lower CoTAPc loadings on CNT (CoTAPc/GCNT-1:20 and 1:40) exhibit similar XANES spectra (Supplementary Fig. 21), further validating the similar CoTAPc adsorption on defect-rich domains on CNT.To monitor the evolution of the CoN4 center in CoTAPc/GCNT during CO2R/COR, we employed the identical flow cell used for assessing CO2R/COR performance with minor modifications for conducting the in-situ electrochemical XAS measurements (Supplementary Fig. 22). As a result, reliable data were obtained at practical relevant current densities, such as up to 260 mA cm−2. Additionally, given the significantly low catalyst loadings and the single-atom nature of the catalysts (Supplementary Table 4), we believe that the XAFS results precisely represent the authentic state of the Co-N4 active center during bulk CO2R/COR electrolysis. Likewise, we also focus on analyzing the characteristic moieties of the planar D4h symmetry: a, b, and c mentioned above. As shown in Fig. 3d, within the in-situ XANES of CoTAPc/GCNT, the intensity of peak a in the white-line region gradually decreases as the overpotential increases. Meanwhile, the intensity of shoulder peak b decreases, and the intensity of pre-edge peak c increases. These changes indicate that the CoN4 structure has deviated from its initial D4h symmetry, likely due to axial interaction with other atoms. Since we have excluded the possibility of gas molecule adsorption, the out-of-plane distorted configuration towards substrate defects under catalytic conditions can be inferred, which aligns well with the simulated XANES. Furthermore, the reduced coordination number of Co−N and the increased Co−N length derived from the EXAFS in R-space (Fig. 3e and Supplementary Fig. 23) provide additional evidence for the above-mentioned structural changes. However, due to the relatively small changes in the bond lengths of Co−N and Co−C from flat to distorted configurations (Supplementary Table 5), the changes in the interatomic distance in EXAFS are not as significant as the symmetry changes observed in XANES. Moreover, we observed similar XANES trends for CO2R on CoTAPc/GCNT in acidic electrolyte (Supplementary Fig. 24), indicating the same potential-driven structure evolution during catalysis. Reducing the CoTAPc loading on GCNT accentuates the spectroscopic changes mentioned above (see Supplementary Fig. 25), owing to more pronounced interactions between fewer CoTAPc molecules and the substrate defects. In-situ EXAFS spectra of post-electrolyzed catalyst, as shown in Supplementary Fig. 26, reveal that three characteristic moieties a, b, and c present partial recovery back to those of initial CoTAPc/GCNT OCV. This result suggests that the recovery of the Co center back to the initial planar state is a relatively slow process after the bulk electrolysis in absence of external electric field.We observed similar trends in Co-N4 structural evolution in the CoPc/GCNT system. As shown in Fig. 3f, while the deviation from square-planar symmetry is small at low overpotential (−0.36 V), noticeable changes occur in moieties a, b, and c when a more cathodic potential is applied (−0.5 V), and these changes align well with those observed in CoTAPc/GCNT. Furthermore, in-situ XANES spectra of COR in 1 M KOH, 1 M KHCO3, and 0.5 M K2SO4 (pH 2) are shown in Fig. 3g–i, respectively. Under all conditions, the spectroscopic tendencies are similar to those observed in CO2R, with the edge position shifting to lower energy, suggesting a more reduced state of Co-center under more cathodic potentials. Taken together, all these in-situ XANES spectra correspond well with our DFT simulations and strongly support our hypothesis that the out-of-plane distortion of CoN4 would occur under cathodic potential.Revealing the *CO binding configurations on CoTAPc/GCNT during CO2R/CORHaving demonstrated the potential-driven distortion of the Co center, we systematically conducted in-situ Fourier-Transform Infrared (FT-IR) studies using attenuated total reflectance surface-enhanced infrared absorption spectroscopy (ATR-SEIRAS) to track important reaction intermediates in COR and CO2R. In a typical measurement, ATR-SEIRAS over the full wavenumber range exhibits no noticeable signals for intermediates other than *CO, such as C − H, *CHO, *CH2O, etc. (Supplementary Fig. 27), suggesting that the hydrogenation of *CO is likely the rate-determining step for CH3OH formation21. As depicted in Fig. 4a, b, two *CO bands in the IR spectra of CoTAPc/GCNT during COR and CO2R in 0.2 M KHCO3 appear at ~2110 − 2090 cm−1 and ~1950 − 1900 cm−1, labeled as I and II, respectively. It is evident that the band II exhibits stronger intensity with a noticeable tail towards the lower frequency region, suggesting the possible existence of a third species. As a comparison, ATR-SEIRAS of the pristine GCNT under COR conditions does not show any peaks (Supplementary Fig. 28) related to the catalysis, confirming that the above bands are associated with CoTAPc.Fig. 4: In-situ FT-IR analysis of CoTAPc/GCNT during CO2R/COR.ATR-SEIRA spectra of a COR and b CO2R in 0.2 M KHCO3 electrolyte, c comparison of frequencies and normalized intensities of the corresponding *CO bands in KHCO3; ATR-SEIRA spectra of d COR and e CO2R in 0.2 M KDCO3 electrolyte, f comparison of frequencies and normalized intensities of corresponding *CO bands in KDCO3. ATR-SEIRA spectra of g COR and h CO2R in 0.2 M phosphate buffer saline (PBS) electrolyte (pH = 8), i comparison of frequencies and normalized intensities of corresponding *CO bands in PBS. All potentials are versus SHE. 100% iR compensation was applied for all potentials. Typical solution resistance for 0.2 M KHCO3, 0.2 M KDCO3 and 0.2 M PBS were measured to be 25.8 Ω, 42.9 Ω and 29.7 Ω, respectively. 100% iR compensation was applied for all potentials. Source data for ATR-SEIRAS are provided in the Source Data file.Notably, band I in both COR and CO2R disappear around −0.95 ~ −1.0 V vs. SHE (Fig. 4a, b), indicating that a more cathodic potential drives the configuration evolution of the *CO. Considering that −1.0 V vs. SHE is too positive for *CO to be protonated to form *CHO, the disappearance of band I is likely not a result of *CO hydrogenation49. Additionally, we found that the disappearance of band I also occurs at a similar potential, ~ −1.0 V vs. SHE, in KDCO3/D2O (Fig. 4d, e). As we observed substantially hindered CH3OH production in COR in both KDCO3 and KOD (Supplementary Fig. 29), we can reasonably exclude the possibility of *CO hydrogenation driving the disappearance of band I. According to the above XAFS result (Fig. 3d), a noticeable out-of-plane distortion has already occurred at potentials lower than −1.1 V vs. SHE. Taken together, we attribute the disappearance of band I to the evolution of CoN4 coordination occurring around −1 V vs. SHE, leading to the altering of the *CO adsorption configuration. Taking into account that this structural transformation occurs at a relatively low overpotential region, and the wavenumber of this *CO band is close to that of the *COL we previously calculated via DFT (2075 cm−1), we attribute this *CO band to the *COL on the quasi-flat CoN4 center. As we discussed earlier, the *CO adsorption on CoN4 is likely determined by the equilibrium between the quasi-flat (*COL) and distorted (*COB) configurations of the CoN4 plane. Therefore, depending on the applied cathodic potential, one could anticipate the bridge *CO lying between the two predicted wavenumbers (2075 ~ 1743 cm−1), aligning with the wavenumber of band II. Hence, we attribute band II to the bridged *CO on the CoN4 center (Fig. 1i).Regarding the tail of the band II (at a very low wavenumber of 1880 ~ 1800 cm−1), we believe it is likely associated with relatively inert *CO species, which are difficult to both desorb and undergo subsequent hydrogenation. To validate this hypothesis, we conducted similar in-situ IR measurements on FePc/GCNT (Supplementary Fig. 30). As a result, *CO species was observed at a similar wavenumber, validating the above hypothesis since Fe-N4 is known to bind CO strongly50,51. Note, its intensity gradually increases with an increase in overpotential, likely due to the more severe distortion occurring to the catalyst center. Besides, the ATR-SEIRAS, collected during the second cycle after recovering the catalyst from the first run via anodic scan, reveals a slightly more pronounced low-frequency band in both KHCO3 and KDCO3 compared to the first scan at the same potential (Supplementary Fig. 31, denoted as band II’ here). We attribute this to the ineffective recovery of the Pc plane distortion during the cycling (Supplementary Fig. 26), causing distorted CoTAPc molecules to accumulate during the second cathodic scan, thus leading to more *CO adsorption. Besides, compared with transition metal surfaces, where *COL and *COB are typically located at ~2100 cm−1 and ~1900-1800 cm−1, respectively52,53,54,55,56,57, we tentatively suggest that the band around 1950 ~ 1900 cm−1 is attributed to the bridge *CO instead of atop *CO on the Co-Pc center.With an increase in overpotential, the frequencies of all *CO bands exhibit a red shift, owning to the Stark tuning effect58,59. Notably, as depicted in Fig. 4c, the frequencies of band I and II in both CO2R and COR are very close, indicating that the *CO binding strength is predominately influenced by the applied potential. Separately, the formation and disappearance of band I for both CO2R and COR track each other, however, band II in COR typically exhibits earlier onset potential compared to those in CO2R, suggesting that the formation of *COB during CO2R is not favored due to the relatively small cathodic potentials applied. Similar trends were observed for CO2R/COR in KDCO3 (Fig. 4f), while CO2R shows a slightly more negative potential to reach the optimal intensity of *COB, which might correspond to the slower CO2R rates. In summary, the evolution of *COL and *COB can be divided into three stages based on overpotential, with the predominant *COL to *COB transformation occurs at −0.7 ~ −1.05 V vs. SHE.Previous literature has indicated that bicarbonate anion could influence the *CO signal in IR measurements (the equilibrium of HCO3− ↔ CO2 + OH− causing low local CO2 concentration)51. To avoid this effect, we also conducted similar in-situ ATR-SEIRAS measurements in 0.2 M phosphate buffer saline (PBS) at ~ pH 8 to exclude the influence of HCO3−. As shown in Fig. 4g, h, similar, if not identical, *CO bands were observed, confirming that the related IR peaks correspond to the *CO intermediates during CO2R/COR. Moreover, the potential dependent trends on the corresponding band frequencies and normalized intensities (Fig. 4i) also resemble those in CO2R/COR in KHCO3 and KDCO3. Finally, we also conducted similar in-situ ATR-SEIRAS measurements in 0.2 M KOH, where we found that band II appears to be broader (Supplementary Fig. 32), especially at negative potentials, and band I appears at more positive potential. This indicates that local pH would also influence the *CO binding configurations. Nevertheless, the potential dependent trends of both *CO bands remain the same. As anticipated, similar trends were observed in ATR-SEIRAS measurements during COR on CoPc/GCNT in different electrolytes (Supplementary Fig. 33).We further investigated the reversibility of these two *CO intermediates. As shown in Fig. 5a, b, when sweeping the electrode potential back to 0 V after the cathodic scan, the *COL band reappears, and the *COB band exhibits an obvious blue shift. This observation indicates that these two types of *CO are reversible depending on the electrode potential. Compared with the initial IR spectra recorded at 0 V, the existence of both *CO bands suggests that the distorted Pc plane is not fully recovered after the cathodic scan. Moreover, the N2 purging experiments further substantiated that the *COL is weakly bonded and can be easily removed (Fig. 5c, d), while bridge *COB binds stronger and remains stable under negative potentials, being stripped off only during anodic scans (Supplementary Fig. 34). As anticipated, similar observations were made in CO2R/COR in electrolytes of KDCO3 and PBS (Supplementary Fig. 35, 36), further validating our conclusions above. Overall, the results obtained from these in-situ ATR-SEIRAS experiments further validate our hypothesis regarding the potential-dependent structure evolution of the CoN4 plane (from quasi-flat to distorted configurations). This evolution, in turn, leads to distinct binding models of *CO on the active sites, as predicted by our DFT simulations. When comparing the onset potential of these different *CO species to the onset potentials for CH3OH production, we believe *COB (band II) is responsible for the CH3OH production in both CO2R and COR. As illustrated in Fig. 5e, the potential-driven distortion of the CoN4 center in CoTAPc/GCNT induces a shift in the binding mode of *CO. This change in binding mode simultaneously extends the C − O bond length, facilitating the subsequent CO hydrogenation and ultimately promoting CH3OH production.Fig. 5: *CO stability characterizations.Comparison of ATR-SEIRA spectra of a COR and b CO2R in 0.2 M KHCO3 after cycle back to 0 V from −1.15 V. ATR-SEIRA spectra of N2 purging experiments at fixed potential of c COR and d CO2R in 0.2 M KHCO3. Both CO or CO2 stream and saturated electrolyte were purged and saturated by N2. e Schematic illustration of two CoTAPc@1V-Gr models, their corresponding *CO binding configurations, and the predominant products. All potentials are versus SHE. 100% iR compensation was applied for all potentials. Typical solution resistance for 0.2 M KHCO3 was measured to be 25.8 Ω. Source data for ATR-SEIRAS are provided in the Source Data file.
