A structural understanding of the active sites is essential for establishing the structure–property relationship in catalysis. Such insights are particularly critical for metal oxide catalysts, where multiple active sites can emerge. DSA, an archetype of MMO-based CER catalysts, typically consists of rutile RuO₂ and rutile TiO₂ with optional doping of other precious or transition metals [32]. Notably, DSA preserves its surface structure even after being operated on for over 10 years [33]. Given that rutile RuO₂ alone is prone to facile dissolution via ionization under acidic electrochemical operation conditions [33–35], the active sites of DSA with remarkable stability could originate from the interfaces where RuO₂ and TiO₂ are mixed.

Zeradjanin et al. imaged the surface of a RuO₂–IrO₂–TiO₂-based DSA electrode using scanning electron microscopy–energy dispersive X-ray spectroscopy (SEM–EDS), which showed random cracks and non-uniform distribution of Ru species (Fig. 2a) [36]. They prepared DSA electrodes with controlled crack sizes and correlated their sizes with their CER activities and conductivities [22]. As the channel width in the DSA cracks increased, CER activity increased proportionally. This correlation suggests that a high density of active RuO₂–TiO₂ interfacial species was distributed on the channels. Moreover, wider channels allowed for the facile detachment of Cl₂ bubbles, generating a large available surface area for the CER.

Scanning electrochemical microscopy (SECM) is an analytical method that can visualize the CER process in situ at the micron scale by reducing the generated Cl₂ molecules with a scanning tip [37]. Zeradjanin et al. identified from 3D SECM images that different types of active sites existed on the RuO₂–IrO₂–TiO₂ electrode (Fig. 2b). The tip currents near the sample, measured as a function of the applied potential over two different sample regions, clearly showed different catalytic activity trends (Fig. 2c) [36]. Chen et al. revealed using SECM that the area showing high CER activity became wider as the crack channel size in the RuO₂–SnO₂–TiO₂-based electrode increased [38].

In situ Raman spectroscopy can also serve as a versatile tool for analyzing the reaction intermediates adsorbed on catalyst surfaces, thus providing detailed information on the active sites. Zeradjanin et al. revealed the relationship between the Raman shift for Cl–O bonding and the CER overpotential for metal oxide catalysts (Fig. 2d) [39]. RuO₂ and IrO₂ occupy the near-top position in the volcano-shaped plot with intermediate Raman shifts among the examined compositions. This correlation suggests that the CER proceeds via Cl species adsorbed on the oxygen site rather than on the metal site, which has been verified by theoretical calculations (see below). Importantly, this result suggests that the vibration frequencies of the crystal lattice can serve as dynamic catalytic descriptors instead of thermodynamic descriptors, such as the adsorption energy of intermediates.

Recently, a growing body of theoretical studies, mainly based on density functional theory (DFT), has provided new insights into the fundamental understanding of CER electrocatalysis [40,41]. DFT calculations have helped to identify the active sites of metal oxide-based CER electrocatalysts and predict the reaction pathways in competitive electrocatalytic reactions between the CER and OER [9,40]. Over et al. proposed possible active sites on the stoichiometric surface of RuO₂(110) (Fig. 3a) [41–43]. Four active sites can exist on the 2×1 surface unit cell of RuO₂(110): coordinatively unsaturated Ru (Ru^c), bridge O (O^br), full 3-coordinated O (O^3f), and coordinatively unsaturated O (O^c). They later calculated the Gibbs free energy for the thermodynamic barrier in the overall CER process (ΔG_loss) and revealed that the O^c site had an optimum ΔG_loss value of +0.13 eV [43]. They also revealed that upon introducing a PtO₂ monolayer on the top of RuO₂(110), the reactivity of the O^c site was further improved by lowering the ΔG_loss (−0.08 eV).

The optimum thermodynamic adsorption energies of the reaction intermediates over a wide range of catalytic materials can be deduced using Sabatier’s volcano principle. Hansen et al. calculated the adsorption energetics of various sites on rutile (110) precious metal oxides (RuO₂, IrO₂, and PtO₂) for CER and OER (Fig. 3b) [44]. The highest CER activity of RuO₂(110) was obtained with the O₂^cc site (2O^c), where a single *Cl binds to two O^c sites. In contrast, the O^c and M^c sites were found to be the optimum active sites for the CER on IrO₂(110) and PtO₂(110), respectively.

The CER is typically performed in aqueous electrolytes in the oxidative potential range. Hence, the OER is unavoidably involved as a parasitic competitive reaction during the CER. Fig. 3c shows the most stable surface structure as a function of potential (U), which considers the intermediate of OER, *OH, and that of CER, *Cl [44]. In most metal oxide catalysts, the two intermediates, *Cl and *OH, commonly share the O^c active site, and the adsorption energy on the O^c site can serve as an activity descriptor for both the CER and OER [40–45]. Therefore, the CER and OER are correlated with each other by the scaling relationship, as encountered in other competitive electrochemical reactions such as the four- and two-electron (4e⁻/2e⁻) ORR [46,47]. Under these conditions, the optimized active site that binds Cl with the lowest energy can also bind OH with the lowest energy. Unlike 4e⁻/2e⁻ ORR, the CER/OER scaling relationship involves different adsorbates (*Cl vs. *OH). Hence, the scaling relation can be thermodynamically overcome by generating an active site with relatively larger adsorption energy toward *OH [41,42]. The doping of heteroatoms into precious metal oxide catalysts [8,20] or the introduction of a new class of catalysts have been explored [28,29] to achieve this goal, aspects of which will be described in detail in the next section.

The most prominent example of a doping strategy is the case of DSA [7,8]. Since Beer disclosed DSA in 1967 [48], this catalyst has been the mainstay of CER electrocatalysts for the past half-century [7]. Surprisingly, clear evidence underpinning the improvement in CER selectivity by doping RuO₂ with TiO₂ is still lacking. Exner et al. theoretically investigated the role of TiO₂ coating on RuO₂(110) in improving the selectivity of the CER [49]. Fig. 3d presents a volcano plot showing the change in ΔG_loss as a function of the Gibbs free energy of the Cl and OH species adsorbed on the O^c site (ΔG⁰[O^c]). The pure RuO₂(110) exhibited a lower ΔG⁰[O^c] value of 1.43 eV (symbol ×) than TiO₂ monolayer coated RuO₂(110) (2.88 eV, symbol ○), and the former showed lower ΔG_loss values both for the CER and OER than the latter. Interestingly, the ΔG_loss of the OER changes more sharply as a function of ΔG⁰[O^c] than that of the CER. The introduction of the TiO₂ monolayer resulted in a larger gap between the ΔG_loss values of the CER and OER. This result suggests that the doping method effectively improves the catalytic selectivity for the CER, which is consistent with experimental results [8,20,50]. From a mechanistic viewpoint, breaking the CER–OER scaling relationship via control of the adsorption energy of intermediates is also an effective method [40,45,51]. Once the adsorption energy of OH on the O^c site was weakened, the ΔG for the formation of *OH on the RuO₂(110) surface increased markedly, thus enhancing the selectivity for the CER (Fig. 3e) [45].

Exner constructed a CER activity map by combining the conventional volcano principle, overpotential-dependent microkinetics of the rate-determining step (RDS), and electrochemical step symmetry index (Fig. 3f) [52]. Notably, the CER via Cl adsorbate exhibited remarkably high selectivity over a wide pH range, whereas the CER via O^cCl adsorbate underwent a decline in selectivity with increasing pH. The currently prevalent CER catalysts (e.g., RuO₂ and IrO₂) prefer the formation of *O^cCl rather than *Cl. Instead, the introduction of metal oxides with weak adsorption energies with oxygen (e.g., PtO₂, TiO₂, and VO₂) can direct the formation of *Cl [52].

Experimental and theoretical approaches have provided useful guidelines based on Sabatier’s volcano principle for the development of new CER electrocatalysts with enhanced performance. However, the overall electrochemical reaction kinetics are determined by the microkinetics of the elementary reaction steps, which are dependent on the rates of charge transfer [53,54]. The CER involves 2e⁻ charge transfer, similar to the HER, and proceeds via elementary steps of adsorption and desorption [53,55,56]. Accordingly, the microkinetics of the HER have provided an important basis for studying the CER mechanism, and vice versa [53,54,57]. The established CER mechanism comprises one adsorption step and three desorption steps (Fig. 4a). The Volmer step, also known as the Volmer discharge, is the adsorption of Cl⁻ from the electrolyte on the adsorption site (S), forming *Cl (i.e., S–Cl_ads) with one-electron transfer (Equation 5). The three possible desorption steps determine the overall CER mechanism, which can vary depending on the catalyst materials and electrolyte conditions. The most typical desorption step for precious metal oxides is the Heyrovský step in which *Cl is combined with Cl⁻ from the electrolyte to generate a Cl₂ molecule (Equation 6). The Heyrovský step resembles Eley–Rideal-type desorption in a heterogeneous gas-phase reaction [58]. The Krishtalik step consists of two sequential reactions: a discharge of *Cl species to form *Cl⁺ followed by Heyrovský step-like desorption (Equation 7) [55]. The Krishtalik step is only applied for the CER of metal oxides because an electron-rich oxide lattice can stabilize the *Cl⁺ species. The Tafel step, or Tafel recombination, is a non-electrochemical desorption in which two adjacent *Cl atoms combine to generate a Cl₂ molecule without an additional electron (Equation 8). The Tafel step is similar to the desorption step of the Langmuir–Hinshelwood mechanism in heterogeneous gas-phase catalysis [58].

The RDS of metal oxide catalysts can be determined by electrochemical reaction order (ℛ) analyses [53,59], which relates to the number of reactants involved in the RDS. Hence, the reaction order analysis method can serve as a powerful tool for investigating the reaction kinetics of 2e⁻-transfer reactions, such as CER, as this reaction involves only one reaction intermediate. Ferro and de Battisti prepared RuO₂ on the stable and non-reactive boron-doped diamond (BDD) support (RuO₂/BDD) and identified the RDS of CER through kinetic analyses (Fig. 4b) [60]. ℛ is defined by the change of current density (j) as a function of the bulk concentration of reactant ([Cl⁻] for the CER) under constant potential (E) and temperature (T) (Equation 9) [59,61]. The ℛ value is also related to the coverage of the CER intermediate (θ_Cl).

When the CER activity of RuO₂/BDD was plotted against [Cl⁻], its ℛ value in the potential range of 1.5–1.56 V (vs. reversible hydrogen electrode, RHE) was 0.74 ± 0.02. This result suggests that the RDS of the CER over RuO₂/BDD is the Heyrovský step involving one reactant (Cl⁻ in the electrolyte), which is consistent with previous results [53,59]. This RDS analysis was carried out by assuming quasi-equilibrium conditions, in which θ_Cl was kept constant (Equation 10). Under such conditions, the current density can be expressed as a function of the reaction rate using the Butler-Volmer equation (Equation 11) and the Ferro-de Battisti test (Equation 12) [60].

where K_V, η, F, R, k_H, and β are the equilibrium coefficient of the Volmer step, overpotential (E − E⁰), the Faraday constant, gas constant (8.314 J·K⁻¹·mol⁻¹), rate constant of the Heyrovský step, and the symmetry factor, respectively. Note that β is a charge transfer coefficient for a one-electron transfer step, i.e., an elementary reaction step [62]. For simplicity, a β of 0.5 is generally applied for the medium overpotential range, assuming a symmetric energy barrier [62]. However, the charge transfer of RDS in the CER is likely to be asymmetric, which depicts the large deviation from 0.5, due to its highly complex electrical double layer with evolving Cl₂ [62,63]. For example, β = 0.7 was obtained for the CER with the Volmer RDS on Pt–N₄ sites [29]. Equation 12 provides a simple test plot based on the experimental values, indicating that exp{(1 − β)·F·η/R·T}·j⁻¹ against exp{−F·η/R·T} should depict a straight slope for the Heyrovský RDS. Recently, Kuo et al. reported the microkinetics of the CER over a RuO₂ catalyst in the low current density range (0.001–1 mA cm⁻²), for which they assumed steady-state conditions instead of a quasi-equilibrium state to reflect very low θ_Cl in this region [64]. Under such conditions, RDS has a ℛ value approaching 2 instead of 1, suggesting that the Volmer step is RDS.

The above analyses indicate that the RDS of the CER should be determined on the basis of a rigorous microkinetics model that can reflect intermediate coverage rather than the slope (b) values of the Tafel equation (Equation 13), which considers simple quasi-equilibrium conditions [53,58–60,65].

Recently, Lim et al. reported platinum-based atomically dispersed catalysts (Pt ADCs) as a new class of CER catalysts. Pt ADCs exhibited different kinetic behaviors from conventional metal oxide catalysts (Fig. 4c,d) [29]. They evaluated the [Cl⁻]-dependent CER activity of Pt ADCs (Pt₁/CNT(P)-700, Fig. 4d) and a Pt-nanoparticle (NP)-based catalyst (PtNP/CNT, Fig. 4c). PtNP/CNT showed an ℛ value of around 1 in the low potential region (1.44–1.48 V vs. RHE), suggesting the Heyrovský step as the RDS. We note that under the CER operating potential, Pt NPs convert to PtO_x and behave like RuO₂ and IrO₂ catalysts. Interestingly, the ℛ of Pt₁/CNT(P)-700 approached 2 in a similar potential region with the Volmer RDS (Fig. 4d), suggesting unique reaction kinetics of Pt ADCs that are distinguishable from those of MMO catalysts. These results were further confirmed by in situ X-ray absorption near-edge spectroscopy (XANES) analyses (see Section 3.3).

The most widely adopted method for improving the CER selectivity of MMO catalysts is doping MMOs with foreign metals. Karlsson and Cornell extensively reviewed the CER selectivity of foreign metal-doped MMO catalysts reported until 2016 [8]. The addition or substitution of Ir [66], Co [67,68], SnO₂ [69], Ni [70], Mg [71], and Pt [72] can improve the CER selectivity of MMOs for the OER. In contrast, Fe [73], Zn [74], Zr [75], and Hf [68] were found to decrease the CER selectivity of MMOs. Recent research has focused on developing selective CER catalysts under low [Cl⁻] and neutral pH conditions, where the CER is thermodynamically unfavorable compared to the OER.

Kishor et al. synthesized Cu-doped RuO₂ catalysts by co-electrodeposition of Cu and Ru precursors on a TiO₂ electrode and investigated their CER selectivity under neutral pH conditions (0.5 M NaCl, Fig. 5a) [76]. Among the prepared catalysts, Ru_99.8Cu_0.02O_x exhibited the highest activity (η = 340 mV at 51 mA cm⁻²), with a high selectivity of 95% (Fig. 5a). The activity and selectivity declined for the catalysts with higher Cu ratios. DFT calculations suggested that the active site of the Ru_99.8Cu_0.02O_x catalyst is the O^br site in Cu-doped RuO₂. This site is energetically unfavorable for the formation of *OH, indicating its high selectivity for the CER.

Lim et al. prepared RuO₂ NPs on an Nb-doped TiO₂ support (RuO₂/Nb:TiO₂) using a hydrothermal method, which demonstrated improved CER activity and selectivity via control of the d-band structure of the Ru–Ti interface (Fig. 5b) [77]. In the RuO₂/Nb:TiO₂ catalyst, the Nb dopant and RuO₂ NPs of 20–30 nm were uniformly distributed. The authors suggested that Nb doping allowed the thermal diffusion of Ti atoms into the rutile RuO₂ lattice to form the active Ru_xTi_1−xO₂ with a thin TiO₂ layer, and enhanced the electronic conductivity of TiO₂ support. RuO₂/Nb:TiO₂ exhibited a high CER selectivity of over 90% in a 0.6 M NaCl neutral electrolyte and a high activity with an η of 22 mV at 10 mA cm⁻². RuO₂/Nb:TiO₂ also preserved its initial CER activity after 50 potential cycles. The authors explained that the high CER selectivity of Ru_xTi_1−xO₂ could be achieved by decreasing the free adsorption energy of the top O atom via replacing the topmost Ru with Ti. Moreover, the thin TiO₂ layer enveloped and protected the active Ru–Ti surface from catalytic degradation. This work also demonstrated that control of the support composition can enhance the durability of the active catalysts.

Although DSA and its derivatives exhibit high CER performance, much effort has been devoted to reduce the cost of CER catalysts by developing non-precious-metal-based CER catalysts, such as SnO₂–Pb₃O₄ [24] and VO_x–SnO_x–SbO_x [25]. Moreno-Hernandez et al. prepared transition metal antimonates (MSb₂O₆, M = Ni, Co, and Mn) with unique crystalline phases via a plasma co-sputtering method [26]. Based on the theoretically predicted Pourbaix diagrams, they confirmed that all MSbO_x phases were stable under highly concentrated NaCl and acidic pH conditions. In particular, CoSbO_x exhibited very high stability for the CER, preserving a high current density of the CER (100 mA cm⁻²) for 250 h, which is much superior to the RuTiO_x catalyst that showed declining activity during interval tests (Fig. 6a). The catalysts showed initial CER activity in the order RuTiO_x > CoSbO_x > MnSbO_x > NiSbO_x (Fig. 6b). Noticeably, CoSbO_x exhibited a very high CER selectivity (97.4 ± 3.0%), which was slightly higher than RuTiO_x (94.8 ± 0.9%). This work is an excellent example demonstrating that even inexpensive non-precious metal-based catalysts can show high CER activity and selectivity compared to RuO₂–TiO₂-based catalysts.

Ha et al. exploited Co₃O₄ NPs as non-precious metal CER catalysts, which showed high CER activity under neutral pH conditions [27]. In the low current density region, Co₃O₄ exhibited similar CER activity to that of RuO₂–TiO₂-based DSA (Fig. 6c) and showed a CER selectivity of over 85% (Fig. 6d). However, Co₃O₄ NPs, in the presence of phosphate (Pi) buffer used to preserve neutral pH, can form Co–Pi species on their surfaces, which are highly active for the OER. Hence, under prolonged operation, the selectivity of Co₃O₄ NPs rapidly declined [78]. In the in situ XANES study, Co₃O₄ NPs preserved the oxidation state in the resting state, even under oxidizing CER conditions. This anomalous oxidation state of the Co₃O₄ NPs suggests that their CER microkinetics are governed by the Krishtalik RDS involving the *Cl⁺ intermediate. In situ Raman spectra confirmed that H₂O was not involved in the reaction under CER conditions. By combining spectroscopic and electrochemical evidence, they suggested Co metallic sites as active sites instead of O^c or O^br sites, which have been suggested as active sites for MMOs.

ADCs, also termed single-atom catalysts, have recently emerged as a new frontier in heterogeneous catalysis [79–84]. ADCs can take advantage of both homogeneous (~100% utilization of metal sites and high activity and selectivity) and heterogeneous catalysts (recyclability and easy separation). For precious metals, the realization of ADCs can substantially reduce their usage and cost compared to NP-based catalysts while achieving equivalent or superior activity. Importantly, the isolated nature of atomically dispersed sites can promote unique catalytic selectivity by altering the adsorption mode of the reactants compared to NP-based catalysts.

In 2020, Lim et al. reported Pt-based ADC as a highly selective CER catalyst [28] inspired by the high chloride affinity of Pt porphyrin. The Pt ADCs were prepared by physical mixing of a Pt(II) mesotetraphenylporphine (PtTPP) precursor and a carbon nanotube (CNT) support, followed by annealing at 700°C. The resulting Pt₁/CNT catalyst consisted of uniformly dispersed, single atomic sites, as clearly visualized by its high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) image (Fig. 7a). Extended X-ray adsorption fine structure analysis of Pt₁/CNT suggested that the Pt center was coordinated to four nitrogen atoms (Pt–N₄ sites), preserving the core structure of the PtTPP precursor after annealing without aggregation. Importantly, Pt₁/CNT exhibited superior catalytic activity to RuO₂–IrO₂–TiO₂ DSA and a Pt NP-based catalyst (PtNP/CNT). Pt₁/CNT reached a current density of 10 mA cm⁻² at an η of 50 mV, which is much lower than those of DSA (η = 105 mV) and PtNP/CNT (η = 120 mV) (Fig. 7b). In the absence of NaCl, Pt₁/CNT delivered virtually no current, indicating that Pt₁/CNT was inactive for OER. Comparing the CER selectivity of Pt₁/CNT and DSA, both catalysts showed similarly high selectivity (> 95%) in an electrolyte of 0.1 M HClO₄ and 1.0 M NaCl (Fig. 7c). Interestingly, upon decreasing the concentration of NaCl to 0.1 M, DSA underwent a significant decline in selectivity to around 60%, whereas Pt₁/CNT preserved high selectivity. A similar phenomenon was observed for neutral electrolytes. A high selectivity shown by Pt₁/CNT suggests that the direct formation of the *Cl intermediate on the Pt–N₄ sites can avoid the scaling relation between the CER and OER, which has plagued MMO catalysts.

The origin of the excellent CER activity of Pt₁/CNT and the detailed active site structure were investigated by DFT calculations. Three plausible models representing atomically dispersed Pt–N₄ sites (PtN₄C₁₂, PtN₄C₁₀, and PtN₂₊₂C₄₊₄) were constructed (Fig. 7d). Calculation of the adsorption free energies (ΔG’s) of these structures for *Cl (Fig. 7e) indicates that the PtN₄C₁₂ site bearing a pyrrolic N moiety is the most plausible active site. This site exhibited the optimum adsorption energy of Cl⁻ but highly unstable adsorption properties with oxygen, corroborating the high CER selectivity of the Pt₁/CNT catalyst.

They later demonstrated that catalytically active Pt–N₄ sites could also be constructed from a combination of a metal chloride salt (H₂PtCl₆·6H₂O) and an ionic liquid in a bottom-up manner [29]. The resulting Pt₁/CNT catalysts also showed similar CER activity and selectivity to the catalyst prepared by the top-down method using the PtTTP precursor, demonstrating the general efficacy of Pt₁/CNT catalysts for the CER.

The viability of ADCs as selective CER catalysts was further explored theoretically by DFT calculations. Liu et al. calculated the ΔG’s of *Cl adsorbates on ADCs bearing the MN₄C₁₀ structure, which were formed by the coordination between transition metals (Ni, Cu, Pd, Pt, Ag, and Au) and pyridinic N sites (Fig. 8a) [85]. Among the examined structures, NiN₄C₁₀ exhibited the lowest ΔG for *Cl formation and was expected to show the highest CER activity. In contrast, NiN₄C₁₀ showed a very large overpotential for the formation of *OOH, which is considered the RDS for the OER, suggesting its high selectivity for the CER. The same group also proposed new carbon-based CER catalysts comprising Ni(S)₂–Li species as active sites using DFT calculations (Fig. 8b) [86]. Compared to analogous structures containing Co or Fe centers, Ni(S)₂–Li exhibited the smallest ΔG for *Cl formation, which originated from a lower p–d hybridization between Cl 2p and Ni 3d than other Fe 3d or Co 3d centers. The resulting weak Ni–Cl interaction of Ni(S)₂–Li species could benefit the desorption of Cl₂ gas, suggesting its potential utility as a highly active CER catalyst.

Pt ADCs comprising Pt–N₄ active sites are expected to have fundamentally different CER kinetics and mechanisms than MMO catalysts, as the former involves the *Cl intermediate, whereas the *OCl intermediate is prevalent in the latter. Lim et al. combined electrochemical reaction order analysis and in situ XANES analyses to investigate the reaction kinetics and mechanism of Pt₁/CNT catalysts in the CER [29]. The PtN₄C₁₂ structure, which was identified as the most plausible active site structure for Pt₁/CNT [28], exhibits a unique behavior in its Pourbaix diagram (Fig. 9a) [28]; PtN₄C₁₂ preadsorbed with *Cl species is more stable than bare PtN₄C₁₂, even below the standard reduction potential of CER (1.36 V vs. SHE). Based on previous CER kinetics studies [53–61,64], the CER kinetics of Pt₁/CNT catalysts were modeled by considering quasi-equilibrium and steady-state approximations, which are schematically presented in Fig. 9b.

In situ electrochemical XANES spectra can provide valuable information on the redox behavior of the active sites during electrocatalytic reactions. Specifically, the whiteline (WL) intensity of the in situ XANES spectra reveals changes in the relative oxidation state of the active site. The average oxidation number of Pt can be calculated by integrating the area of the WL peak and interpolating the resulting value into those of Pt references with well-known oxidation numbers (e.g., K₂PtCl₄ for Pt²⁺ and H₂PtCl₆ for Pt⁴⁺). Importantly, for Pt₁/CNT catalysts, the coverage of the reaction intermediate (θ_Cl) on the Pt–N₄ active sites can be directly linked to the oxidation state of Pt, as it was demonstrated that no other intermediates are adsorbed on the Pt–N₄ sites in the CER-operating potential range [28]. In situ XANES data and potential-dependent oxidation states of the Pt₁/CNT catalysts with ascending and descending potentials are presented in Fig. 9c and 9d, respectively. Upon immersing the catalyst in the electrolyte, the average oxidation number of Pt₁/CNT was 2.2. With the potential being increased to 1.1 and 1.25 V (vs. RHE), the average oxidation number increased to 2.4, which originates from the preadsorption of Cl⁻ below E_CER (Fig. 9a). When the potential was increased to 1.4 and 1.55 V (vs. RHE), the average oxidation number gradually increased to 2.8 and 3.9, respectively. Hence, at a low potential of 1.4 V (vs. RHE), θ_Cl is low, and the reaction proceeds under the steady-state condition, which changes to almost saturated θ_Cl under the quasi-equilibrium state at a higher potential (1.55 V vs. RHE).

By combining of the analyses of reaction order, in situ electrochemical XANES spectra, and theoretical prediction, a schematic free-energy diagram (Fig. 9e, top) and a schematic model showing θ_Cl and RDS during the CER are proposed (Fig. 9e, bottom). Below E_CER, the preadsorbed Cl⁻ on Pt–N₄ sites makes the Heyrovský step and the Volmer step the first and second steps of the CER, respectively. At the low η region (η = 65–90 mV), the Pt₁/CNT exhibited the R value of ~2 and the Tafel slope of ~40 mV dec⁻¹ with low θ_Cl, in which the Volmer step is expected as the RDS. However, as the potential was increased, the R value decreased to ~1 and θ_Cl increased, with the Volmer step gradually becoming faster than the Heyrovský step. Consequently, in the high η region, the CER mechanism switches to the Heyrovský step as the RDS. The switch of the RDS by electrochemical and XANES analyses is also consistent with thermodynamic prediction based on free-energy diagrams [28]. This work provides fundamental insights into the CER catalyzed by atomically dispersed Pt catalysts, which exhibit different reaction kinetics and an unusual potential-dependent switch in the RDS.