Identification of K⁺-determined reaction pathway for facilitated kinetics of CO₂ electroreduction

Abstract

Cations such as K⁺ play a key part in the CO₂ electroreduction reaction, but their role in the reaction mechanism is still in debate. Here, we use a highly symmetric Ni-N₄ structure to selectively probe the mechanistic influence of K⁺ and identify its interaction with chemisorbed CO₂⁻. Our electrochemical kinetics study finds a shift in the rate-determining step in the presence of K⁺. Spectral evidence of chemisorbed CO₂⁻ from in-situ X-ray absorption spectroscopy and in-situ Raman spectroscopy pinpoints the origin of this rate-determining step shift. Grand canonical potential kinetics simulations - consistent with experimental results - further complement these findings. We thereby identify a long proposed non-covalent interaction between K⁺ and chemisorbed CO₂⁻. This interaction stabilizes chemisorbed CO₂⁻ and thus switches the rate-determining step from concerted proton electron transfer to independent proton transfer. Consequently, this rate-determining step shift lowers the reaction barrier by eliminating the contribution of the electron transfer step. This K⁺-determined reaction pathway enables a lower energy barrier for CO₂ electroreduction reaction than the competing hydrogen evolution reaction, leading to an exclusive selectivity for CO₂ electroreduction reaction.

Introduction

Recently, extensive discussions have emerged on how electrolyte cations modify the activity and selectivity of electrocatalytic reactions such as CO₂RR^1,2,3. For example, density functional theory (DFT) studies predicted that K⁺ could reduce the energy barrier of each step during CO₂RR through the interaction with adsorbed reagent species⁴. However, results based on ab initio molecular dynamics (AIMD) simulations instead proposed that K⁺ ions could stabilize the key intermediate to activate the reaction⁵. Another hypothesis is that K⁺ could modulate the electric field distribution that impedes the side reaction⁶. To date, these discussions on how the cations behave and whether the interaction between K⁺ and chemisorbed CO₂⁻ intermediate exists are still ongoing⁷. These debates were mainly generated for two reasons. Theoretically, conventional computation methods such as DFT and AIMD are not able to fully address questions in electrocatalysis due to the difficulty in mimicking experimental conditions, which may lead to false impression^7,8. Meanwhile, the lack of strong experimental evidence that clarifies the role of K⁺ in catalysis makes it difficult to propose an empirically guided model. Therefore, a higher-level approach such as grand canonical potential kinetics (GCP-K) combined with a corresponding experimentally derived model would be highly beneficial to address simulation needs^7,9. It is, however, enormously challenging to obtain direct evidence to identify how K⁺ shapes the reaction pathway and what, if any, the nature of the interaction between K⁺ and chemisorbed CO₂⁻ is⁷. The very recent experimental observations by Waegele et al. proved that K⁺ ions could gather on the surface of Au during the CO₂RR¹⁰, but no evidence was detected of an interaction between K⁺ and CO₂⁻. Even though the facilitating effect of K⁺ is apparent over the Au surface^5,6, related literature suggests the Au surface may not be an ideal choice to investigate the mechanistic influence of K⁺⁷. Because the rate-determining step (RDS) of the CO₂RR over the Au surface is typically electron transfer^11,12, the introduction of K⁺ is unlikely to shift this RDS and makes discerning mechanistic changes from Tafel slopes difficult. The weak adsorption of CO₂ and the irrepressible side reactions on the Au surface also hinder the identification of the cardinal value of the Tafel slope^7,13, limiting its use in kinetics experiments. Consequently, we propose that a new system is needed to investigate the role of K⁺ in catalysis during CO₂RR.

Regarding suitably performant but well-defined systems, Nickel single-atom catalysts have been reported as state-of-the-art in converting CO₂RR to CO, particularly the highly symmetric Ni-N₄ structure¹⁴, despite ground-state DFT calculations indicating that the Ni-N₄ site remains largely inert for both CO₂RR and hydrogen evolution reaction (HER)^11,15,16. This contradiction between theory and experiment likely arises from calculations taking concerted proton-electron transfer (CPET) as the RDS of the reaction without considering the catalytic role of K⁺, as a significant promoting effect of K⁺ during CO₂RR has been observed experimentally. Because the Ni-N₄ structure has a reaction pathway towards only a single product., it allows us to more easily isolate the contribution of K⁺ to the reaction and the correlating structural changes. The well-defined and unambiguous structure of the Ni-N₄ structure is also ideally suited to rationalize these findings with computational methods, which enables us to integrate experiment and theory into a clear picture of the K⁺-determined reaction pathway. Meanwhile, considering the easy transfer of electrons from nickel sites to CO₂ and the lower thermodynamic energy of chemisorbed CO₂⁻ compared to that of adsorbed COOH on Ni-N₄ sites¹¹, the potential stabilization effect from K⁺ to chemisorbed CO₂⁻ may instigate an RDS alteration, which is instrumental to give explicit experimental evidence for the interaction between K⁺ and chemisorbed CO₂⁻. We, therefore, choose a Ni-N₄ single-atom catalyst as a model system in our study, which possesses the properties laid out above.

Herein, we performed in situ XANES and in situ Raman spectroscopy to track the structure evolution and adsorption behaviors of the Ni-N₄ site during the reaction, and we complement this with electrochemical analysis to quantify the effect of K⁺ on the reaction kinetics. These findings elucidate that the origin of enhanced catalytic activity of K⁺ lies indeed in its ability to stabilize chemisorbed CO₂⁻ by a non-covalent interaction. The stabilized chemisorbed CO₂⁻ further serves as a reactant, leading to an RDS shift from CPET to proton transfer, lowing the reaction barrier and thus delivering excellent performance for CO₂RR. By incorporating these experimentally derived constraints into our initial GCP-K calculations model, we achieve a harmonious match between simulated CO₂RR performance and experimental results. This reconciliation bridges the longstanding gap between theoretical predictions and empirical observations and provides solid evidence for a K⁺ determined reaction mechanism shift on a Ni-N₄ catalyst.

Results

It has been demonstrated that molecularly dispersed Nickel phthalocyanine (NiPc) catalysts (NiPc-MDE) exhibit nearly 100% selectivity for CO evolution during CO₂RR¹⁴. However, despite its specificity, its practical application is hindered by its high production cost and susceptibility to aggregation. Moreover, its low metal loading makes it unsuitable for spectral investigations, and its performance at industrial current densities is restricted by insufficient mass transfer efficiency^17,18,19. To address these limitations, a Ni-CN single-atom catalyst with uniform Ni-N₄ sites and high mass transfer efficiency (Ni-N₄-HM) was developed as depicted in Fig. 1a. Typically, Ni atoms were adsorbed into ZIF-8 at first, subsequently the dried powder was mixed with NaCl by grinding. The mixture was then annealed at 950 °C to achieve Ni-N₄-HM. During this process, molten NaCl introduces porosity to the substrate and facilitates atomic rearrangement^20,21, and high temperature favors the rearrangement of atoms around the Ni center to form the thermodynamically stable Ni-N₄ configuration²². For comparison, Ni-CN-ZIF is obtained by directly annealing ZIF-8 powder that absorbed Ni atoms in advance. For Ni-CN-BLE, ZIF-8 was directly annealed with NaCl powder after mixing, then the obtained CN substrate absorbed Ni atoms followed by annealing at 950° C, this method was adopted to separate the shearing effects of NaCl from the tendency of materials to restructure into their most stable configuration under ultra-high temperature conditions to form a catalyst with blended active sites.

Fig. 1: Schematic illustration for the preparation of Ni-N₄-HM and its CO₂RR performance.

a schematic presentation of the synthesis of Ni-N₄-HM. The insets represent partially enlarged illustrations. b LSVs of Ni-N₄-HM, Ni-CN-BLE, Ni-CN-ZIF, and NiPc-MDE conducted in CO₂-saturated 0.5 M KHCO₃ in H-type cell (without iR compensation). Inset shows the expanded partial region around onset potential. c CO Faradaic efficiencies at various potential versus RHE in H-type cell (without iR compensation). d CO Faradaic efficiencies at various current densities in a flow cell setup (1 M KOH).

A series of characterizations were performed to verify the successful synthesis of Ni-N₄-HM. Transmission electron microscopy (TEM) images (Supplementary Fig. 1) show that molten salt indeed restructures the ZIF-8 substrate²¹. Evidence from High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM), X-ray diffraction (XRD) patterns, and Extended X-ray absorption fine structure (EXAFS) spectroscopy demonstrates that the nickel atoms are dispersed atomically across all samples (Supplementary Figs. 1–5). In addition, N₂ adsorption-desorption isotherms highlight a significant increase in the BET surface area post-NaCl etching (Supplementary Fig. 6). The Raman spectrum and N 1 s XPS spectrum also prove the restructuring of the substrate (Supplementary Figs. 7, 8).

The CO₂ electroreduction performances were used to assess the active centers of the catalysts. During the CO₂RR process in an H-type cell, only CO and H₂ gas were detected by gas chromatography (GC) (Supplementary Fig. 9a, b), and no liquid product was formed as determined by ¹H NMR spectroscopy (Supplementary Fig. 10a). The increased BET surfaces enabled Ni-N₄-HM and Ni-CN-BLE to achieve a doubled current density compared to that of Ni-CN-ZIF (Fig. 1b). The onset potential of Ni-N₄-HM (− 0.524 V) aligns closely with the NiPc catalyst (− 0.557 V) (Fig. 1b). Moreover, when we normalized the current density of NiPc-MDE to the same Ni content (Supplementary Fig. 11), the on-set potential for NiPc-MDE is − 0.517 V, indicating that the active site of NiPc molecular catalyst possesses similar activity to that of Ni-N₄-HM. However, Ni-CN-BLE has a less negative onset potential (− 0.377 V), indicating a significant difference in the active sites of Ni-N₄-HM and Ni-CN-BLE. Meanwhile, a clear limiting diffusion current plateau emerges at − 1.0 V, and this plateau also can be found on NiPc-MDE, indicating the same exclusive selectivity for CO₂RR on Ni-N₄-HM^23,24. In Fig. 1c, Ni-N₄-HM maintains Faradaic efficiencies (FEs) over 99% within the potential range from − 0.6 V to − 1.1 V, and almost no hydrogen is produced at this potential window (Supplementary Figs. 12, 13). Such an exclusive selectivity is also in line with a NiPc molecular catalyst. On the contrary, the CO FE for Ni-CN-BLE rapidly decreased to below 95% upon reaching − 0.65 V. Subsequently, we tested Ni-N₄-HM in flow-cell setups to further identify the exclusive selectivity of Ni-N₄-HM (Supplementary Figs. 14–18). It achieves 1 A cm⁻² at − 0.8 V with a CO FE of over 99% (Fig. 1d). No liquid phase product is formed and hydrogen content is almost undetectable (Supplementary Figs. 7c, d, 8b). Although Ni-CN-BLE can achieve a high current density of 300 mA/cm², maintaining selectivity at industrial current density proved challenging, further confirming that Ni-N₄-HM and Ni-CN-BLE possess different active sites. Therefore, we conclude that Ni-N₄-HM has been successfully synthesized with a uniform Ni-N₄ active site. Its exclusive selectivity is also what enables it as a suitable catalyst to investigate the mechanistic influence of K⁺ on the CO₂RR.

To explore the mechanistic influence of K⁺ in CO₂RR, the performance of the Ni-N₄-HM was measured under different conditions with or without K⁺. Figure 2a shows I–V curves obtained in 0.5 M KHCO₃ solution with either Ar (blue) or CO₂ (red) introduced. The pink curve represents the H₂ partial current density derived from the red curve (with CO₂). The misalignment of the blue and pink curves indicates that the HER is inhibited by CO₂ in the presence of K⁺, suggesting that the reaction barrier of CO₂RR is lower than that of the HER in the presence of K⁺. For comparison, I–V curves taken in the absence and presence of K⁺ are compared in Fig. 2b. Here, the red and blue curves, which again represent the conditions with CO₂ and Ar, respectively, virtually coincide. This shows that in the absence of K⁺, CO₂ cannot hinder the HER. In contrast, the introduction of K⁺ leads to a misalignment of the purple and pink curves, indicating that the Ni-N₄-HM catalyst is inactive for HER but very active for CO₂RR in the presence of K⁺. Further enhanced activity for CO₂RR is also observed as the concentration of K⁺ increases (Supplementary Fig. 19). Drawing from these observations, it is evident that in the absence of K⁺, the catalyst is active for the HER and remains inactive for CO₂RR even when CO₂ is introduced, indicating the reaction barrier of HER is usually lower than that of CO₂RR. While in the presence of K⁺, the catalyst shows a pronounced activity for CO₂RR and largely sidelines the HER, suggesting that CO₂RR takes precedence over HER, and the reaction barrier of CO₂RR is now lower than that of HER. Subsequently, rotating disk electrode (RDE) measurements were also employed to further explore these kinetics. The results suggest that persistent sluggish kinetics impede the progress of CO₂RR (Supplementary Fig. 20). Subsequently, to determine how this drastic transition in reaction barrier is generated, we further investigated the RDS of this reaction. I–V curves with a variation of pH level are provided in Supplementary Fig. 21, where the slight difference between these two curves reveals that the reaction is independent of the H⁺ concentration in the presence of K⁺. Meanwhile, the kinetics isotope effect was studied by substituting H atoms with D atoms, showing that the RDS is related to the proton transfer (Fig. 2c). The value of the Tafel slope offers more direct insights (Fig. 2d). Without K⁺, the value of Tafel slope of J_CO is 122.6 mV dec⁻¹, indicating the RDS of this reaction is CPET. In contrast, with abundant K⁺, regardless of pH fluctuations, the value of the Tafel slope remains around 60 mV dec⁻¹, indicating that the RDS of this reaction is proton transfer. Therefore, we have identified an RDS shift of CO₂RR upon the introduction of K⁺ over the Ni-N₄ site. With K⁺, the RDS shifts towards proton transfer, which favors CO₂RR, whereas without K⁺, the RDS remains CPET, which favors HER.

Fig. 2: Kinetic studies for mechanistic role of K⁺ in CO₂RR.

a LSVs of Ni-N₄-HM in 0.5 M KHCO₃ solution. Red curve: With CO₂ introduced. Blue curve: With Ar introduced. Gray curve: Experimental partial current density for HER derived from the red curve (CO₂ introduced). b LSVs of Ni-N₄-HM in different solutions, but with the same pH value of 3. c Kinetic isotope effect (KIE) determined from comparing electrolytes prepared with D₂O vs H₂O. d Tafel slopes for different solutions. Note that the purple curve represents the Tafel slope of total current density. The variation of K⁺ concentration is modulated with the introduction of KCl. The voltage in this figure is not iR corrected.

To clarify how the RDS shift is modulated by K⁺, we performed in situ XANES measurements CO₂ saturated 1 M KHCO₃ to track the evolution of the Nickel active site during the reaction (Supplementary Figs. 22–24). We carried out XAS simulations both in the presence and absence of adsorbed carbon species (Supplementary Fig. 25). Figure 3a reveals the influence of COOH adsorption to the Ni site, and Fig. 3b depicts results when the Ni atom is moved out of its plane without introducing any carbon species. Specifically, a significant intensity increase occurs in the pre-edge region (peak A, corresponding to 1s – 3d transition) in both figures due to the D_4h symmetry distortion^25,26,27,28. Meanwhile, the intensity of peak B corresponds to the 1s – 4p_z transition and relates to the intensity of the covalent bond between metal and its coordinated atoms, with shakedown contributions²⁵. This transition is prominent in the square planar geometries and more subtle in the square-pyramidal geometries^25,29,30,31. Thus, upon COOH adsorption on the nickel site, the initial square planar geometry transitions to a square-pyramidal geometry, causing a marked decrease in peak B intensity (Fig. 3a), eventually leading to its disappearance. Conversely, in the scenario without any carbon species adsorbed, the square planar geometry remains with a slight elongation in bond length, ultimately leading to a diminished 1s – 4p_z transition and only a marginal decrease in peak B intensity (Fig. 3b). Moreover, we can find that peak C’s intensity rises and peak D’s decreases in both figures. This change can be tracked using the I_C/I_D ratio. As a COOH binds to the nickel site, with structure optimization in VASP, the nickel atom protrudes from the Ni-N₄ plane, leading to an increase of I_C/I_D (Fig. 3a)³². Meanwhile, as the Ni atom shifts out of its plane in the absence of carbon species, the value of I_C/I_D increases as well (Fig. 3b). Regarding the in situ XANES spectra (Fig. 3c), at − 0.4 V, both HER and CO₂RR remain inactive. Yet a subtle increase in I_C/I_D is observed, coupled with a decline in the rising edge (peak B) compared to the curves collected at OCV. As the applied potential shifts towards more negative values, approaching CO₂RR conditions, the value of I_C/I_D continues its upward trajectory and the intensity of peak B consistently decreases (Fig. 3d), indicating the adsorption of carbon species. However, at − 0.4 V, still at a potential before CO₂RR activation, the changes do not stem from adsorbed COOH. Thus, it is plausible to deduce the presence of chemisorbed CO₂⁻. This assumption is further supported by in situ Raman spectra (Fig. 3e). The peaks A and D are assigned to carbon substrate³³, remaining constant at different potentials. Peak B corresponds to the C-O bond, and peak C to the asymmetric O-C-O⁻ stretching^34,35,36. At − 0.4 V, the peaks B and C emerge. Shifting the applied potential to a more negative region results in a modest decline in peak B, while peak C sees an uptick. Similar tests conducted without the presence of K⁺ show no apparent changes (Supplementary Fig. 26). These phenomena indicate that at − 0.4 V, the chemisorbed CO₂⁻ is generated and adsorbed. Therefore, we propose that prior to the onset potential for CO₂RR, the physisorbed CO₂ transitions to chemisorbed CO₂⁻, and in the presence of K⁺, chemisorbed CO₂⁻ is stabilized.

Fig. 3: Simulated and experimental in situ XANES of Ni-N₄-HM.

a Comparison of simulated XANES spectra of Ni-N₄ and Ni-N₄-COOH, shaded regions represent different transitions. b Comparison of simulated XANES spectra: variations in Nickel atom protrusion from the plane. c Normalized in situ nickel K-edge XANES spectra of Ni-N₄-HM in CO₂ saturated 1 M KHCO₃ buffer at different applied potentials. d zoom-in difference spectra of (c). e in situ Raman spectra in CO₂ saturated 1 M KCl solution with a variation of applied potential.

So far, we suggest that K⁺ stabilizes chemisorbed CO₂⁻ during CO₂RR, influencing the reaction pathway. As illustrated in Fig. 4, the physisorbed ground-state CO₂ can transition into chemisorbed CO₂⁻ by fast electron transfer¹¹. Without the stabilization from the interaction with K⁺, the unstable chemisorbed CO₂⁻ reverts back to physisorbed CO₂, making CPET the RDS. Under these circumstances, the reaction barrier of CO₂RR is higher than that of HER, leading to an inactivity for CO₂RR. However, when K⁺ is present, the chemisorbed CO₂⁻ is stable long enough to serve as a subsequent reactant, leading to an altered RDS. The CPET is no longer the only viable reaction path, instead, separated proton transfer and electron transfer can occur, which translates to proton transfer being the actual RDS in this scenario. Consequently, the reaction barrier is significantly reduced and indeed lower than that of the HER, resulting in a strong preference towards CO₂RR.

Fig. 4: Proposed mechanism of CO₂RR in the presence or absence of K⁺.

The blue upward arrow signifies the energy barrier of this step. As we can see, physisorbed CO₂ can transition to chemisorbed CO₂⁻ with electron transfer, which is unstable. In the absence of K⁺, this chemisorbed CO₂⁻ remains unstable and reverts to a physisorbed CO₂, steering the reaction pathway towards CPET. In this pathway, the reaction barrier is higher than that of HER, leading to an inactivity for CO₂RR and a high FE for HER. In contrast, when K⁺ is present, the interaction with chemisorbed CO₂⁻ ensures its stability. Consequently, the RDS transitions to proton transfer, which boasts a substantially lower barrier than that of CPET, as well as that of HER. As a result, the Ni-N₄ site exhibits exceptional activity and selectivity for CO₂RR.

Complex electrocatalytic reactions in aqueous solution, such as the HER and CO₂RR, exhibit dependencies on multiple descriptors^8,37,38. Any electrocatalytic reaction is driven by an applied potential across the reaction cell, leading to a strong correlation between activity and applied potential. In addition, the ET processes in these reactions sometimes involve partial charge transfers rather than consuming a full electron (Supplementary Note 2). Therefore, the conventional approach of heterogeneous thermal catalysis, which relies primarily on temperature and pressure, may not suffice to describe intricate systems like CO₂RR³⁹. To address these, we employ an emerging method called Grand Canonical Potential Kinetics (GCP-K), an advancement over conventional Quantum Mechanics (QM, fixed numbers of electrons) that allows for variable electron numbers⁴⁰. Unlike the traditional Butler-Volmer description, where the electron transfer occurs between molecules and electrodes to achieve the final state (Fig. 5a), the GCP-K method maintains continuous changes in reaction barriers with the applied voltage (Fig. 5b, c). This innovative approach better captures the complexities of electrocatalytic systems, providing an accurate relationship between current density and potential. Consequently, it facilitates direct comparisons with experimental observations, offering a more comprehensive understanding of these intricate processes. (Supplementary Notes. 3, 4)³⁹.

Fig. 5: Schematic showing how voltage dependent electrochemical reactions are described by GCP-K.

a schematic of the traditional Butler-Volmer reaction kinetics, the blue curve represents the reactant, the red one is for the product at U₀, and the green one is for the product at U₁. As the potential changes from U₀ to U₁, the energy profile shifts from the red line to the green line. So, we can describe the traditional Butler-Volmer kinetics as a special case of the GCP-K scheme with the electron transfer instantaneously. b the linear relationship between the bond distance and charge in Grand canonical potential kinetics, the energy of TS at different applied potentials thus can be obtained. c Grand canonical potential kinetics methodology, illustrating the relationship between the TS geometry and its charge as applied potential changes. The red curve represents the reaction pathways for the conversion of COOH to CO on Ni-N₄ at − 0.43 V applied potential and the blue one is for − 0.75 V applied potential. Inserts are used to give intuitive images of the bond distance difference between the TS geometry at different applied potentials, and the TS at − 0.43 V can transform into the TS at − 0.75 V with a charge of 0.3 electrons.

To obtain a rational configuration for the K⁺-H₂O-CO₂ system on the Ni-N₄ site, the ab initio molecular dynamics (AIMD) simulations were carried out first (Supplementary Note. 5 and Supplementary Figs. 27, 28). As shown in Fig. 6a, the radial distribution function g(r) of the O-K⁺ pair was extracted from the equilibrated AIMD trajectory. Within a range of 2.75 Å to 3.0 Å, five O-K⁺ pairs were identified, with the O atom in the O-K⁺ pair at around 3.0 Å being associated with the CO₂ molecule. These results confirmed a proper description of the K⁺ coordination shell, as the estimated O-K⁺ distances d_O-K+ for H₂O molecules in the first solvation shell matched well with the previous reports⁵. In addition, to further probe the role of K⁺ on CO₂ stabilization, K⁺ was moved to a position away from the CO₂ molecule at the H₂O–M⁺–CO₂/Ni-N₄ interface after the equilibration of the H₂O–M⁺–CO₂–Ni-N₄ system. We found that the angle between O-C-O in CO₂ molecular rapidly increased from 140^◦ to nearly 180^◦ (Fig. 6b). This observation unequivocally underscored the pivotal role of K⁺ in facilitating CO₂ stabilization.

Fig. 6: AIMD and GCP-K calculation of the process of CO₂RR to CO with K⁺.

a The radial distribution function g(r) and the corresponding integrated coordination number of the O-K⁺ pair in an equilibrated AIMD trajectory. b Evolution of the Ni-C distance, O-C-O angle, and K-O (CO₂) distance before and after the K⁺ ion left the active site. The purple dashed line represents the leaving of K⁺. In the left region, where K⁺ remains near the surface and interacts with chemisorbed CO₂⁻, the Ni-C distance maintains around 2 Å, and the angle of O-C-O keeps steady below 140°, which means the interaction exists all the time and chemisorbed CO₂⁻ never leaves. The left insert is a partial view of chemisorbed CO₂⁻ stabilized by K⁺. In the right region, as K⁺ moves away from the surface, the O-C-O angle becomes a right angle, and the interaction between K⁺ and CO₂⁻ disappears, then the CO₂ without stabilization from K⁺ leaves the active center. The right insert shows the partial view of spontaneous desorption of CO₂ as K⁺ departs. c Calculated partial current density for CO formation on Ni-N₂C₂ (green curve), Ni-N₃C₁ (blue curve), and Ni-N₄ (red curve) along with experimental data of this work (dotted curve) for comparison. The purple dotted line indicates the onset potential at the current density of 1 mA cm⁻², with the equivalent nickel site numbers. d Partial current density for CO formation on Ni-N₂C₂ (green curve), Ni-N₃C₁ (blue curve), and Ni-N₄ (red curve). The dotted blue curve following the blue curve is also the calculated prediction, indicating the RDS changes to the CO leaving the active site. e Faraday efficiency calculated for Ni-N₂C₂ (green curve), Ni-N₃C₁ (blue curve), and Ni-N₄ (red curve).

Using the GCP-K method alongside AIMD-derived configuration (Supplementary Figs. 29–36), we compute the CO partial I–V curves starting from chemisorbed CO₂⁻ as the primary reactant (Fig. 6c), yielding a strong alignment with the empirical data. The predicted onset potential (at 1 mA cm^-2) for CO₂RR on the Ni-N₄ site is − 0.59 V versus RHE, diverging from the empirical value (− 0.52 V) by 0.07 V. Notably, the onset potential for Ni-N₄ site is more negative than that of others (− 0.50 V for Ni-N₃C₁, − 0.36 V for Ni-N₂C₂). Similarly, the simulated onset potential for HER on the Ni-N₄ site is about − 1.2 V, aligning closely with the observed value (Supplementary Fig. 37). Simulated onset potential for HER on the Ni-N₃C₁ site is about − 0.5 V and − 0.4 V on Ni-N₂C₂ site (Supplementary Figs. 38, 39), remarkably close to their respective CO₂RR onset potentials. These results reflect that the Ni-N₄ site possesses a more negative onset potential of total current density compared to the Ni-N₃C₁ and Ni-N₂C₂ sites, which is in line with observations in Fig. 1b that Ni-N₄-HM and NiPc molecular catalysts possess a more negative on-set potential, while Ni-CN-BLE with blend active sites possesses a less negative on-set potential. I–V curves in Fig. 6d present the capability of Ni-N_xC_4-x sites to achieve industrial-level current density. Remarkably, the Ni-N₄ site exhibits an exceptional current density of 1.2 A/cm², while the maximal CO partial current density of Ni-N₂C₂ and Ni-N₃C₁ sites lag significantly. This difference aligns with the empirical data in Fig. 1c that Ni-N₄-HM with Ni-N₄ site achieves an exclusive FE at a current density of 1 A/cm², while Ni-CN-BLE with blend sites can only reach a maximum CO partial current density of 250 mA/cm². This suggests that only the Ni-N₄ site effectively mitigates the interference from HER due to its HER onset region being more negative than its working condition, while the early-emerged HER on Ni-N₂C₂ and Ni-N₃C₁ sites curtail their CO partial current density. Furthermore, the inverse trend of current density in the blue curve is partially contributed by the high energy barrier of CO desorption (Supplementary Fig. 40). In this potential range, the RDS transitions to CO desorption. This observation underscores the limitation of predicting complex systems with a single descriptor. Considering the contest with HER, the FE of each site is graphically presented (Fig. 6e). The Ni-N₄ site maintains a FE of over 99.5% from the onset potential to − 0.95 V, followed by a modest decline to an FE of 95% at − 1 V, matching well with the empirical data in Fig. 1d. Ni-N₃C₁ site can reach a FE exceeding 90% within a narrow potential window of just 0.1 V, but the drop emerged at − 0.7 V, while Ni-N₂C₂ site is largely inert for CO₂RR. Moreover, the Tafel slope value is also presented (Supplementary Fig. 41), in which the Ni-N₄ site has a Tafel slope of 51.8 mV dec⁻¹. This strongly supports the argument that in the presence of K⁺, the RDS of CO₂RR is proton transfer. In conclusion, by comparing the simulations with empirical data, we validate our proposed mechanism that K⁺ stabilizes chemisorbed CO₂⁻, allowing it to bifurcate the CPET process. This results in a shift of the RDS to proton transfer and an alleviated CO₂RR reaction barrier, which is ultimately the cause for exclusive selectivity of the Ni-N₄ site.

In summary, our studies identified the RDS shift upon the introduction of K⁺ into solution by kinetics studies. We further tracked the structure evolution and adsorbed species of the Ni site by in situ XANES and Raman spectroscopy to clarify exactly how the RDS was modulated by K⁺, which suggested that the chemisorbed CO₂⁻ can be stabilized below the on-set potential. These integrated findings revealed that K⁺ modulates the reaction pathway by non-covalent interaction with chemisorbed CO₂⁻, shifting the RDS from CPET to independent proton transfer, leading to a reduced reaction energy that makes CO₂RR preferentially proceed over HER in the presence of K⁺. Using this K⁺-stabilized model and taking proton transfer as the RDS, GCP-K simulations achieved well-matched results with the experimental data, compounding the rationality of this proposed mechanism. Consequently, we obtained a revised CO₂RR mechanism in an aqueous solution, which further harmonizes the gap between theoretical predictions and practical catalysis on Ni single-atom catalysts.

Methods

Computational methods

The spin-polarized calculations of the reactants, intermediates, products, and reaction process involved in CO₂RR on the Ni single-atom catalysts were carried out with the density functional theory (DFT) within the PBE⁴¹ exchange-correlation functional with the generalized gradient approximation (GGA)⁴², as implemented in the VASP⁴³. The ion-electron interaction was described with the projector-augmented plane-wave (PAW) method⁴⁴. To simulate the Ni single-atom catalysts, a 5 × 5 graphene supercell was chosen and two adjacent C atoms were removed to construct a divacancy to load the single Ni atom, which is coordinated to 2–4 N/C atoms through strong covalent bonds, denoted as Ni-N₂C₂, Ni-N₃C₁, and Ni-N₄. To avoid the interlayer interaction, the vacuum layer is set to be 15 Å between two graphene layers. To simulate the role of K⁺ and explicit H₂O in the process of CO₂RR, one K⁺ with 7 H₂O molecules around was placed nearby the active site of catalysts. For geometry optimization, the cut-off energy was set to be 500 eV and the Brillouin zone was sampled with 5 × 5 × 1 k-points. The systems were relaxed until the energy and force reached the convergence threshold of 10⁻⁶eV and 0.015 eV/Å, respectively. We describe the van der Waals (vdW) interactions by utilizing the DFT-D3 method⁴⁵. Atomic coordinates of structures can be found in Supplementary Data 1.

In the ab initio molecular dynamics (AIMD) simulation model, 68 water molecules, leading to an average density of ∼ 1 g cm^–3 were placed to construct the solvent water environment. Under the canonical ensemble (NVT), the AIMD simulations last 10 ps with a timestep of 1 fs, in which the temperature is controlled at 300 K using Nosé–Hoover thermostats^46,47.

For the elementary reaction barriers, free energy for each point along the NEB⁴⁸ reaction path was calculated by VASP, thus, the configuration of the transition state could change adiabatically with the applied potential. To obtain the transition states and reaction pathway, the transition state tools for VASP (VTST) package⁴⁹ with climbing image nudged elastic band (CI-NEB)⁵⁰ was used.

After obtaining the optimized geometry, we performed the single-point jDFTx⁵¹ calculations to obtain the combined DFT and solvation free energy. The jDFTx computations were elected due to the accuracy of including constant potentials for electrochemical reaction and in describing continuum solvation. However, because the obtained configurations are similar in both jDFTx and VASP, so a single-point computation is sufficient for the stable species. In addition, to describe the solvation implicitly, the CANDLE solvation model was applied in jDFTx. μ_e;SHE = 4.66 eV was applied for all structures. With a k-point mesh of 5 × 5 × 1, we applied a plane wave basis set with an energy cutoff of 20 Hartree. The systems were relaxed until the free energy convergence threshold of 10⁻⁸ Hartree.

Chemicals

All the chemicals were analytical grade and used without further purification. Analytical grade Sodium chloride (NaCl, 99.5%), Zinc nitrate hexahydrate (Zn(NO₃)₂·6H₂O, 99.0%), Nickel (II) nitrate hexahydrate (Ni(NO₃)₂·6H₂O, 98.0%), Ferric nitrate hexahydrate (Fe(NO₃)₃·9H₂O, 98.5%), 2-methyl imidazole (98.0%), Potassium hydroxide (KOH, 85.0%) and Potassium bicarbonate (KHCO₃, 99.5%) were obtained from Shanghai Chemical Reagents, China. Iridium (III) chloride hydrate (IrCl₃·xH₂O, 99.9%) was purchased from Alfa Aesar. Nafion solution (5%) was purchased from Aldrich. Deionized (DI) water from a Milli-Q System (Millipore, Billerica, MA, USA) was used in all experiments.

Synthesis of ZIF-8

Typically, 1.116 g Zn(NO₃)₂·6H₂O was dissolved in 30 ml methanol and 1.232 g 2-methyl imidazole was dissolved in 30 ml methanol, followed by adding Zn(NO₃)₂ solution into the other one under ultrasound for 100 min at room temperature. Then the ZIF-8 was set aside for growing overnight. The as-obtained precipitates were centrifuged and washed three times with methanol, then dried at 65° C in a vacuum for overnight.

Synthesis of Ni-N₄-HM

In a normal procedure, 200 mg Nickel (II) nitrate hexahydrate was dissolved in 250 mL ethanol, and the powder of ZIF-8 (100 mg) and powder of NaCl (3 g) were mixed by grinding in mortar, then 1.5 ml Nickel (II) nitrate hexahydrate solution was added to be adsorbed into the ZIF-8. Next, the dried powder was transferred into a tube furnace and heated to 950 °C (heating rate 10 °C/min) for 2 h in a stream of Ar (20 mL/min). The annealed powder was washed with a mixture of ethanol and water three times and then dried in a vacuum to obtain the Ni-N₄-HM. The Ni content was measured to be 0.52 wt% based on ICP-AES analysis.

Synthesis of Ni-CN-ZIF

Ni-CN-ZIF was synthesized by using an ionic exchange method¹⁵. The ZIF-8 powder was uniformly dispersed in 10 ml of n-hexane by ultrasonic treatment for 5 min at room temperature. Once a homogeneous solution was obtained, 50 µL of Ni(NO₃)₂ aqueous solution(100 mg/mL) was slowly added into the homogeneous solution, followed by ultrasonication for 2 min at room temperature. To make the Ni species adsorbed, the obtained solution was stirred for 3 h. The mixture was then centrifuged, and the powder was dried in a vacuum. Next, the dried sample was transferred to a tube furnace and heated at 1000° C (with a heating rate of 10° C/min) for 2 h at an Ar atmosphere (20 mL/min) to yield Ni-CN-ZIF. The Ni content was measured to be 1.13 wt% based on ICP-AES analysis.

Synthesis of Ni-CN-BLE

The powder of ZIF-8 (100 mg) was mixed with NaCl (3 g) by grinding, followed by annealing at 950° C for 2 h in tube finance, the product was then washed with the mixture of ethanol and water and dried in vacuum to obtain the CN-substrate. Next, the powder of CN-substrate (100 mg) was dispersed in 50 mL Ni(NO₃)₂ aqueous solution (800 mg/L) and stirred for 3 h. Then the substrate was centrifuged and dried. Finally, the sample was transferred into a tube finance and annealed at 950° C (heating rate 10° C/min) for 2 h in a stream of Ar (20 mL/min). The cooled powder was Ni-CN-BLE. The Ni content was measured to be 0.67 wt% based on ICP-AES analysis.

Synthesis of NiPc-MDE

In a normal procedure^14,52, the Carbon nanotubes (CNTs) were cleaned by calcination at 500° C in air for 2 h followed by washing with a 5 wt% HCl aqueous solution. The purified CNTs were filtered and washed with DI water. Then, 30 mg of the purified CNTs was dispersed into 30 ml of DMF under sonication. The DMF solution with NiPc dissolved was subsequently added to the CNT suspension in sonication for 30 min to make sure it was a well-dispersed solution which was further stirred at room temperature overnight. Subsequently, the product was centrifuged, and washed with DMF and ethanol, followed by lyophilization. The Ni content was measured to be 0.23 wt% based on ICP-AES analysis.

Material characterizations

Powder X-ray diffraction patterns of samples were collected using a Rigaku Miniflex-600, operating at 40 kV voltage and 15 mA current with CuKα radiation (λ = 0.15406 nm). Aberration-corrected high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images and energy-dispersive X-ray spectroscopy (EDS) mapping of samples were provided by a JEOL JEM-2010 LaB6 high-resolution transmission electron microscope operated at 200 kV. Elemental content analysis of Ni of all samples was identified by inductively coupled plasma atomic emission spectrometry (ICP-AES) on an Optima 7300 DV. Micromeritics Tristar II3020 M was used to determine the Brunauer-Emmett-Teller (BET) specific surface area, samples were degassed at 300° C for 3 h to prepare for adsorption-desorption isotherm measurements. X-ray photoelectron spectroscopy (XPS) spectra were obtained using a PHI 5000 Versa microprobe with Al Kα radiation, referencing the C1s peak at 284.8 eV. During chronoamperometry, effluent gas from the cell was analyzed using a PANNA A91 gas chromatograph. The gas chromatograph, equipped with molecular sieve 5 A, Porapak Q 80/100 mesh, SE-30, and HP-Al₂O₃/S capillary columns, used ultra-high purity helium as the carrier gas to determine gas product concentrations.

In situ XANES measurements

The Ni K-edge (8333 eV) XANES spectra were measured at the BL14W1 beamline of the Shanghai Synchrotron Radiation Facility (SSRF). The storage ring of SSRF was operated at 3.5 GeV with a maximum electron current of 250 mA. The hard X-ray beam was monochromized using a Si(111) double-crystal monochromator and further suppressed for harmonics by 30% detuning. In situ, XANES measurements were performed in the fluorescence mode using a custom-designed reaction cell. The Ni K-edge (E₀) position was calibrated using a Ni foil reference, and all XANES data were collected within a single beam time to ensure consistency. The collected XANES spectra were processed using the ATHENA module of the IFEFFIT software package⁵³, which includes background subtraction, normalization, and energy calibration. Further analysis, including linear combination fitting and principal component analysis, was used to elucidate the evolution of Ni species during the reaction process.

XANES simulations

Reasonable parameters are selected for the convolution of the calculated spectra employing an energy-dependent arctangent shape of the Lorentzian profile (details can be found in the manual for the FDMNES program⁵⁴). The calculated convolution spectrum after the common shift is compared with the p-state projection of the Ni absorber (pDOS). The maximum white line peak of the XANES spectrum is aligned with the maximum p-state projection of the Ni absorber to evaluate the rationality of parameters.

In situ Raman measurements

The Raman spectroscopy measurements were performed on a Horiba LabRAM HR Evolution Raman microscope with a 785 laser. We measure in static mode with an Olympus LUMFLN60XW water-dipping objective, 2.25 mW laser power, and an acquisition time of 10 s, we measured two different Raman windows separately (1200–1450 cm^–1, 1450–1800 cm^–1). All measurements were carried out in the solution of 1 M KCl with CO₂ gas purged continuously.

Preparation of the cathode electrodes

For the H-type cell, the area of the working electrode was fixed at 1 cm⁻². The catalyst ink consisted of 5 mg catalyst, 40 $\mu L$ Nafion solution, and 1 mL 1:1 (vol/vol) ethanol/water mixed solvent was sonicated for 60 min at room temperature to form the homogeneous ink. The well-dispersed ink was dip-coated onto the electrode to reach the loading of 0.4 mg/mL.

For the Flow cell, the carbon paper (Sigracet 35 BC) with a microporous layer (MPL) was purchased from the Fuel Cell Store. The well-dispersed ink was airbrushed onto the gas diffusion layer using an Anest Iwata RG-3L airbrush (Japan) pumped by an air compressor at a certain pressure of 60 p.s.i to reach a loading of 0.4 mg/cm² for the Flow cell test. The gas diffusion electrodes (GDEs) were dried in a vacuum chamber before use.

For the rotating disk electrode, the well-dispersed ink was dipped onto the glassy carbon RDE to obtain a loading of 0.1 mg/cm²

Preparation of the anode IrO₂-Ti mesh

The IrO₂-Ti mesh was prepared by a modified thermal decomposition and dip coating method. Briefly, acetone and de-ionized water were used to degrease the titanium mesh, which was then etched in a 6 M HCl solution for 20 min before coating. The dip coating solution consisted of 30 mg of IrCl₃·xH₂O, and 1 ml concentrated HCl was dissolved in 9 ml isopropanol. Then the fresh titanium mesh was dipped into the IrCl₃ solution followed by calcination in a furnace at 500 °C for 10 min. The process of coating was repeated until a certain loading was achieved (2 mg/cm²).

Preparation of FeNi-LDHs/Ni mesh

The Ni mesh washed with 5 M HCl aqueous solution was used to electrodeposit. The FeNi-LDHs was electrodeposited on the as-prepared Ni mesh in a 25 mL electrolyte bath containing Ni(NO₃)₂·6H₂O (3 mM) and Fe(NO₃)₃·9H₂O (3 mM) at − 1.00 V (vs. Ag/AgCl) at 25 °C for 120 s. After the reaction, the mesh was washed thoroughly with DI water. The FeNi-LDHs/Ni mesh anode was used in the Flow cell system.

Electrochemical measurements

The electrochemical measurements were conducted with a PGSTAT302N workstation with a 10 A Booster M204 and CHI760E. All the electrolytes are stored in a PFA bottle. For H-type cell tests, the two gas-tight cells were separated by Nafion212 membrane, a Pt foil was used as a counter electrode to generate OER, an Ag/AgCl electrode was used as a reference electrode, which is calibrated in a 0.1 M Na₂SO₄ Solution. 0.5 M KHCO₃ aqueous solutions pre-saturated with CO₂ were used as electrolytes for each chamber. For the electrolyte with a pH value of 4 (4.0 ± 0.1), 2 M KCl and 0.00005 M H₂SO₄ dissolved in the DI water with CO₂ gas saturated to make the electrolyte. The area of the working electrode was fixed as 1 $\times$ 1 cm², and CO₂ gas was purged in during the reaction. All potentials were measured against an Ag/AgCl reference electrode (saturated KCl, obtained from Tjaida and stored in a saturated KCl solution before use) and converted to the reversible hydrogen electrode (RHE) scale by

$$E\,\left({vs}. \, {RHE}\right)=E\,\left({vs}. \, {Ag}/{AgCl}\right)+0.22\,V+0.0592\times {pH}$$

For the gas-tight Flow cell system, a CO₂ gas compartment and two liquid compartments with channels of dimensions 2 cm × 0.5 cm × 0.3 cm were separated by carbon paper and an anion-exchange membrane respectively. The gaseous CO₂ could pass behind the GDL into the cathode electrolyte and then go out with gas product. The catholyte and anolyte were separated by an anion exchange membrane (Sustainion 37–50, Dioxide Materials). The FeNi-LDHs/Ni mesh was used as an anode, and an Ag/AgCl electrode was used as a reference electrode. The area of the working electrode was fixed as 2 $\times$ 0.5 cm², 1 M KOH aqueous solution was used as an electrolyte to be circulated, and the gas flow rate increased along with the current density increase. All potentials were measured as mentioned above. Resistance in the flow cell is 1.48 Ω for Ni-N₄-HM, 1.53 Ω for Ni-CN-BLE, and 2.16 Ω for Ni-CN-ZIF. An IR correction was compensated by:

$$E\,\left({vs}. \, {Ag}/{AgCl}\right)=E-0.8\,{I}_{{total}}R$$

A factor of 0.8 is applied for the ohmic potential drop.

The RDE experiment was done using a glassy carbon RDE with a diameter of 5 mm. The electrode was polished by alumina polishing powder (0.05 μm) before use. A Pt foil and an Ag/AgCl electrode were used as counter electrode and reference electrode respectively. 1 M KHCO₃ pre-saturated with CO₂ gas was used as an electrolyte. The linear sweeping voltammetry curves were collected at a scan rate of 5 mV/S with different rotating speed (400 to 2500 rpm). The electron transfer number (n) and kinetic current density (J_k) can be obtained according to the K-L equation:

$$\frac{1}{J}=\frac{1}{{J}_{K}}+\frac{1}{{J}_{L}}=\frac{1}{{J}_{K}}+\frac{1}{B{\omega }^{\frac{1}{2}}}$$

$$B=0.62{nFC}{D}^{\frac{2}{3}}{V}^{-\frac{1}{6}}$$

Where $J$ is the total current density from tests, ${J}_{K}$ is the kinetic current density, ${J}_{L}$ is the limiting diffusion current density, $\omega$ is the angular velocity of the disk, $n$ is the number of electrons, $F$ is the Faraday constant, $C$ is the bulk concentration of CO₂, $D$ is the diffusion coefficient of CO₂ in 1 M KHCO₃, and $V$ is the kinematic viscosity of the electrolyte.

CO₂RR product analysis

The Faradaic efficiency (FE) of gas products was calculated as

$${{FE}}_{{gas}}={x}_{i}\times v\times \frac{{P}_{0}}{{RT}}\times \frac{{z}_{i}F}{{I}_{{total}}}\times 100\%$$

where ${x}_{i}$ is the volume fraction of gas product $i,v$ the gas flow rate at the cathode outlet measured by a soap film flow meter. ${z}_{i}$ the number of electrons involved in the reaction to produce one molecule of product $i$, $F$ the Faraday constant, ${P}_{0}$ the absolute pressure (101.325 kPa), $R$ the gas constant, $T$ the temperature, ${I}_{{total}}$ the total current.

The Tafel slope was calculated based on the Tafel equation (ƞ = blog(j_CO/j_o)), where ƞ is the overpotential, b is the Tafel slope, j_CO is the current density for CO formation, and j_o is the exchange current density.