Crystallinity-dependent structural evolution of CoS₂ catalysts for enhanced oxygen evolution reaction

Abstract

Transition metal sulfides (TMSs) are promising catalysts for water oxidation due to their unique electronic properties, but their application is limited by structural instability during the oxygen evolution reaction (OER). The specific mechanisms by which crystallinity variations in TMSs influence their structural evolution and catalytic performance remain poorly understood. Here, we investigate the structural evolution of CoS₂ catalysts with varying crystallinities and their effect on the OER. In-situ characterization and density functional theory (DFT) calculations reveal that lower crystallinity enables rapid equilibrium among surface sulfur-oxygen exchange, metal site activation, and oxygen intermediate evolution, enhancing the transformation of bulk phase into small amounts of sulfur-stabilized oxyhydroxide and improving OER performance. Conversely, moderate crystallinity leads to self-corrosion and progressive structure collapse from the exterior inward, resulting in deactivation. This work highlights the potential of crystallinity control in optimizing the electronic states and catalytic behavior of water oxidation catalysts.

Introduction

Electrocatalytic overall water splitting technologies, powered by renewable electricity to produce green hydrogen, are essential for addressing environmental concerns and meeting the growing global demand for energy^1,2. A critical factor in advancing these technologies is the development of low-cost, highly efficient catalysts capable of overcoming the sluggish kinetics of the anodic oxygen evolution reaction (OER). Transition metal sulfides (TMSs), with diverse compositions and crystal structures, offer significant potential in the field of energy storage due to their unique d-electron configurations and well electrical conductivity^3,4. However, the high oxidative potential and harsh corrosion environment at the anode have thus far limited the use of TMSs in improving the slow kinetics of OER⁵. Therefore, monitoring the kinetic structural transformation during the OER process can not only deliver insights into catalytic mechanisms of sulfides, but also offer guidance on structural design for their efficient application in related catalytic system^6,7.

According to equilibrium potential vs pH diagrams, the oxidation of sulfur into soluble species (S²⁻ to SO₄²⁻ at 0.15 V vs RHE) takes place below the equilibrium potential for O₂ evolution (H₂O to O₂ at 1.23 V vs RHE)⁸. This indicates that TMSs are thermodynamically less stable and prone to phase transformations, driven by surface amorphization, bulk phase reconstruction, and elemental leaching⁹. With the advancement of in situ characterization techniques, it is now widely recognized that surface/bulk restructuring occurs in heterogeneous catalysts during the OER process^10,11. Recent studies on complete structural reconstruction into oxyhydroxide and partial surface oxidation, forming a sulfide-oxyhydroxide core-shell structure, have demonstrated that these in situ-generated amorphous catalysts are responsible for enhanced activity^12,13,14,15. For example, edge-sharing octahedral CoO₆ nanoclusters, converted from Co₉S₈ with di-μ-oxo Co–Co motifs on the surface, have been identified as the true catalytic centers. The oxygen atoms located at the top of CoO₆ octahedra can adsorb reaction intermediates (OH^*), while the Co–O–Co bridging bonds are prone to breaking, creating unsaturated coordination sites that enhance the OER performance¹³. In the reconstructed CoFe₂S₄, the presence of spin-polarized Fe₃S₄/Co(Fe)O_xH_y interface can act as the spin channel, optimizing the energetics of OER steps and facilitating the spin-sensitive electron transfer during OER¹⁶. Differences in the structures of these reconstructed catalysts are believed to affect the binding energetics of oxygen intermediate species, potentially explaining the observed variations in activity. However, the detailed mechanistic origins driving the varying degrees of sulfide reconstruction under OER conditions remain poorly understood.

Amorphous materials, formed through reconstruction during the electrochemical process or directly synthesized, are characterized by inherent disorder. These disorder leads to the creation of abundant “dangling bonds” and defects, resulting in a higher density of active sites compared to that of crystalline materials^17,18,19. To directly gain insights that the surface amorphous phases serve as the stable active sites for enhanced electrocatalytic activity, Zhang et al. employed time of flight secondary ion mass spectrometry (TOF-SIMS), a technique sensitive to elemental composition at the scale of several atomic layers, to confirm that the amorphous phase is composed of oxysulfide²⁰. Moreover, by comparing the sulfur and oxygen element signals after 1 h, 10 h, and 100 h of OER, they found that all samples exhibited a similar elemental distribution pattern. The only change was an increase in the thickness of the amorphous layer, indicating that the surface oxysulfide acts as a stable active site, supporting sustained OER activity. However, Stephens et al. argued that despite amorphous IrO_x exhibiting a higher geometric current density, the intrinsic reaction rates per active state on IrO_x and rutile IrO₂ are comparable at given potentials, after quantifying the densities of redox-active centers and analyzing the correlation between intrinsic reaction rates and active-state energetics²¹. Additionally, Chung et al. showed that no significant enhancement in OER activity was observed in amorphous nickelates regardless of the Fe doping method. According to the high OER activity in Fe-exchanged crystalline films, they suggested that the random atomic arrangement in amorphous films is insensitive to doping methods²².

Despite numerous studies examining how local structure differences between crystalline and amorphous materials affect the electronic structure of catalysts^23,24, it remains unclear how crystallinity-crystallinity discrepancy influences activity trends from one catalyst to another under the same phase, particularly in sulfides. The scarcity of research in this area can be attributed to the fact that, compared to oxides, sulfur’s larger atomic radius and lower electronegativity allow it to arrange in multiple ways and bond with metals in various configurations²⁵. Furthermore, sulfides generally exhibit lower thermodynamic stability than oxides, making them more susceptible to phase transitions within the same temperature range in response to external conditions²⁶. To gain a comprehensive understanding of the relationship between the structural evolution of sulfides and their OER performance, it is crucial to elucidate how crystallinity differences influence the electronic states of sulfides, while minimizing the introduction of additional interfering factors.

Here, we used crystalline CoS₂ as a research model to distinguish the differences in dynamic structural evolution behaviors of sulfides (Supplementary Fig. 1). Experimental characterizations and density functional theory (DFT) calculations confirm that changes in crystallinity not only alter the electric charge distribution between Co and S but also endow them with different surface and bulk sulfur–oxygen exchange capabilities. By monitoring dynamic surface reconstruction and analyzing sulfur oxidation corrosion behavior of catalysts, we identified that 1.55 V vs RHE as the threshold potential for initiating significant sulfur–oxygen exchange reactions on the catalyst surface. Time-dependent monitoring of bulk phase reconstruction via in situ X-ray absorption spectroscopy (XAS) revealed that rapid equilibrium between sulfur–oxygen exchange on the catalyst surface, the activation of metal sites, and the evolution of key oxygen intermediates is crucial for transforming the catalyst with the lowest crystallinity from sulfide to oxyhydroxide in the shortest time, thereby enhancing its catalytic performance. To further understand the influence of CoS₂ structural evolution on OER performance, we established a relationship between the weighted p-band center of sulfur and the catalytic activity of CoS₂ within a specific structural range. Our findings show that the closer the weighted p-band center of sulfur is to the Fermi level, the better the catalyst’s OER performance. Finally, according to the effects of catalyst reconstruction differences on the distribution of residual sulfur in active species, we used DFT calculations to elucidate the essential role of residual sulfur in regulating the structure of the active catalytic phase. Residual sulfur disrupts the geometry symmetry of the Co–O octahedral, reduces the degeneracy of the d-orbitals, induces electronic rearrangement in the bonding orbitals, and enhances the ability of oxygen intermediate evolution. Together, these insights reveal the mechanistic origins of the varying degrees of reconstruction among sulfides with different local structures and provide a rational design strategy for efficient water oxidation catalysts with in situ structural evolution capabilities.

Results

Quantifying the crystallinity of sulfides and analyzing its impact on structure

In this study, we acquired three catalysts with increasing crystallinity (Supplementary Fig. 2a, b). The characteristic peaks in the X-ray diffraction (XRD) pattern are ascribed to CoS₂ (PDF: 41-1471). The quantified relative crystallinity (RC) values were determined to be 69%, 76%, and 83%, respectively (Supplementary Note 1 and Supplementary Table 1). For simplicity, RC_69%, RC_76%, and RC_83% are used to denote three catalysts in this work unless otherwise mentioned. Obviously, as the crystallinity of the catalyst increases, the full width at half maximum (FWHM) of the corresponding characteristic peaks gradually decreases (Fig. 1a and Supplementary Fig. 2c). According to the Scherrer equation, this leads to an increase in grain size. Transmission electron microscopy (TEM) images reveal that the CoS₂ nanosheets consist of numerous crystalline grains (Supplementary Fig. 3). As the annealing temperature rises, the grain size grows, reflecting the enhanced crystallinity of the catalysts (Fig. 1b–d). This trend is further confirmed by the improved clarity of the diffraction spots in the corresponding FFT images. Moreover, the uniformity of the lattice fringes in the grains across all three catalysts indicates that structural order improves with increasing crystallinity. To ensure the reliability of these observations, grain sizes in various regions were characterized, all of which align with this trend (Supplementary Fig. 4). Importantly, during the control of catalyst crystallinity, the morphology, size, and element contents of the nanosheets in three CoS₂ catalysts show no significant changes. (Fig. 1e–g, Supplementary Figs. 5 and 6, and Supplementary Table 2).

Fig. 1: Quantification of crystallinity and structural characterization of three CoS₂ nanosheets.

a FWHM and CN of RC_69%, RC_76%, and RC_83%. b–d TEM images and corresponding FFT of RC_69% (b), RC_76% (c), and RC_83% (d). The colored lattice fringes and labels on the TEM images correspond to the diffraction spots in the associated FFT pattern. e–g Field-emission scanning electron microscope (FE-SEM) images of RC_69% (e), RC_76% (f), and RC_83% (g). h Ex situ k²-weighted FT-EXAFS spectra and fitting of FT-EXAFS spectra collected at the Co K-edge of RC_69%, RC_76%, and RC_83%. Experimental data (scatter plots) and fits (solid lines) are shown. i Peak position of cobalt L₃-edge from the EELS spectra and the rising edge position from the Co K-edge XANES spectra of RC_69%, RC_76%, and RC_83%. The Co K-edge positions are extracted from the half-height values of the normalized rising absorption edge. j Local atomic configurations and Co–S bond lengths at different positions of RC_69%, RC_76%, and RC_83%. k p-band center of sulfur for surface in RC_69%, RC_76%, and RC_83%. Green, gray, and yellow lines or points represent RC_69%, RC_76%, and RC_83% in this work. Source data are provided as a Source Data file.

To eliminate the interference of sulfur defects on the changes in catalyst crystallinity, Fourier-transform (FT) extended X-ray absorption fine structure (FT-EXAFS) spectra were collected to identify the local coordination environment of Co centers (Fig. 1h). The intense peaks present within the interval R + α = 1–3 Å, validate the coordination environment of the first scattering shell Co–S. All the samples exhibit another peak at R + α of 3.55 Å, corresponding to the second coordination shell of Co–Co, meaning that all of them possess long-range order⁹, which are consistent with the XRD results. Moreover, the fitting results on the FT-EXAFS spectra indicate that the coordination numbers (CN) for the Co–S bond of RC_69%, RC_76%, and RC_83% are 5.8, 6.0, and 6.0, respectively (Fig. 1a, Supplementary Fig. 7, and Supplementary Table 3). The subtle change in the Co–S CN, and the absence of significant signals in the electron paramagnetic resonance (EPR) spectra of all catalysts (Supplementary Fig. 8), suggest that sulfur defects are not the primary factor contributing to the changes in catalyst crystallinity. In addition, we also exclude the possibility that atomic arrangement disorder contributes to strain generation via PDF analysis (Supplementary Fig. 9).

X-ray photoelectron spectroscopy (XPS), electron energy loss spectroscopy (EELS), and X-ray absorption near-edge structure (XANES) spectra were employed to investigate the effects of local structure modification on the electronic states of cobalt from both surface microregion and bulk phase perspectives (Fig. 1i and Supplementary Figs. 10–12). The results exhibited a decreasing trend across the catalysts, following the order: RC_69% > RC_83% > RC_76%. Interestingly, contrary to theoretical expectations, the S valence followed the same trend as Co. This suggests that the bond length between local Co and S varies with crystallinity, leading to changes in the electronic distribution around Co and S in the system. To deepen understand the effects of local structure modifications on the electronic structure of active sites and subsequent sulfur–oxygen exchange behavior, we used ab initio molecular dynamics (AIMD) to model three CoS₂ catalysts with different crystallinities (Fig. 1j, Supplementary Data 1, and Supplementary Fig. 13). Based on these models, the p-band center of sulfur for surface and bulk in all catalysts were calculated (Supplementary Fig. 14). Generally, the closer the sulfur p-band center is to the Fermi level, the more likely sulfur–oxygen exchange will occur via binding the dissociative hydroxide ions in electrolyte^27,28. The results show that the ability of surface sulfur–oxygen exchange is RC_76% > RC_69% > RC_83% (Fig. 1k), while for the bulk it is RC_69% > RC_83% > RC_76% (Supplementary Fig. 15).

Comparison of the electrocatalytic activity and stability of catalysts

OER performances of catalysts were evaluated by linear sweep voltammetry (LSV) with corresponding Tafel plots. RC_69% exhibits the lowest overpotential of 276 mV to reach a geometric area normalized current density of 10 mA cm⁻², followed by RC_83%, while RC_76% demonstrates the least favorable performance (Supplementary Fig. 16a). As for the kinetics criterion, RC_69% shows a small Tafel slope (79 mV dec⁻¹), smaller than RC_83% (85 mV dec⁻¹) and RC_76% (108 mV dec⁻¹). Furthermore, we used the continuous chronopotentiometry-cyclic voltammetry (CP-CV) model to evaluate the activation tendency and stability of catalysts. During 8 CP testing, RC_76% and RC_83% showed continuously increased voltage and a progressive decline in OER performance, with respective reductions of 6.1% and 3.6% from their initial values. In contrast, RC_69% exhibited superior activity and stability, with signs of secondary activation, and its voltage dropped by only 2.9% (Supplementary Fig. 17). Additionally, CV test results reveal a continuous decline in current density for both RC_76% and RC_83%. However, RC_69% exhibits an upward trend in current density after completing the final CP cycle, which aligns with the observed secondary activation phenomenon (Supplementary Fig. 18).

The characteristics of the separated positive and negative scanning curves at the OER region suggest a coupling between metal ion oxidation and sulfur oxidative corrosion. Evidently, RC_76% shows weaker overall performance in terms of metal site activation and sulfur–oxygen exchange ability. The negative shifts of cathodic peaks mean that the Co sites in catalysts are prone to being oxidized to high valence states, which are recognized as higher intrinsic activity sites²⁹. It can be observed that after completing 8 cycles of CP-CV testing, the order of the cathodic peak positions for three catalysts, from negative to positive, is RC_69% < RC_83% < RC_76% (Supplementary Fig. 18). Given the voltage and current variations observed during the OER tests for catalysts with different crystallinities, along with the leaching of active sites induced by sulfur–oxygen exchange, we conducted an analysis of metal ion concentrations in the electrolyte during OER process via inductively coupled plasma mass spectrometry (ICP-MS). The results revealed that RC_69% demonstrated structural stability during reconstruction, with metal ion leaching consistently measured below 1.5 ppm and therefore operated stably for 300 h at 10 mA cm⁻² (Supplementary Fig. 19). With the progress of reconstruction during the OER process, the three catalysts exhibited similar electrocatalytic active surface area (ECSA) after 8 cycles of CP-CV test (Supplementary Figs. 20 and 21), indicating that the intrinsic activity differences arise from the influence of local structural modifications on the electronic structure of active sites (Supplementary Fig. 16b). Moreover, the turnover frequencies (TOF) and Faraday efficiency (FE) of the catalysts were also provided (Supplementary Figs. 22 and 23). Considering CoS₂ with the lowest crystallinity exhibited the highest OER performance among the catalysts investigated in this work, we also synthesized amorphous Co_xS_y and characterized its structure and performance (Supplementary Figs. 24 and 25).

Impact of surface sulfur–oxygen exchange differences on catalytic performance

We further employed the electron localization function (ELF) and relevant quantitative calculations to conduct a comparative analysis of the chemical bonding types and strengths within the three catalysts (Fig. 2a and Supplementary Fig. 26)³⁰. The ELF visualizations of the surface regions reveal significant localization between sulfur atoms, as indicated by the yellow-to-red color gradients, signifying covalent interactions. Meanwhile, cobalt is shown in blue regions, suggesting a clear ionic interaction between cobalt and sulfur. Compared to RC_69% and RC_76%, RC_83% exhibits a more uniform bonding due to its higher structural order, resulting in stronger Co–S ionic and S–S covalent interactions. Interestingly, RC_69%, characterized by a higher degree of local structural disorder, shows enhanced interaction between some S–S and Co–S pairs. Therefore, the overall bonding strength among the samples increases in the order: RC_76% < RC_69% < RC_83%.

Fig. 2: In situ monitoring of the sulfur–oxygen exchange behavior on the surface of catalysts during OER.

a ELF analysis of the interactions between Co and S in RC_69%, RC_76%, and RC_83% from left to right. The color scale represents the dimensionless ELF index, with red regions indicating significant electron localization, while blue areas correspond to delocalized electrons or zones with extremely low electron density. b Potential dependence of Raman modes: peak positions and intensities. The spectral background is subtracted, and the intensity of the Co–S characteristic peak is normalized to that at open circuit voltage (OCV) for comparison. Peak positions were obtained by recording the maximum values of the characteristic peaks. c Potential dependence of IR intensities of the OOH_ad peak and the O–H stretching vibration peak. d In situ UV-vis absorption spectra for RC_69%. e In situ UV absorption vs time curve of RC_69%. f In situ UV current of RC_69%. g Relationship between charge and differential absorption of RC_69%. Source data are provided as a Source Data file.

In situ Raman spectra were performed to monitor the dynamic surface reconstruction and the oxidation corrosion of sulfur (Fig. 2b and Supplementary Figs. 27 and 28). For RC_69%, the peaks at 500 cm⁻¹ corresponding to E_g vibrational modes of Co–O bonds in CoOOH were observed at 1.05 V vs RHE³¹. With increasing applied potential, this peak shifted to higher frequencies, indicating gradual oxidation of Co³⁺ to Co⁴⁺. The peak stabilized at 575 cm⁻¹, corresponding to the A_1g mode of CoO₂. When the applied voltage drops back to 1.05 V vs RHE, the high-valence CoO₂ is reduced to CoOOH, as confirmed by the appearance of the characteristic peaks near 600 cm⁻¹. At voltages exceeding 1.55 V vs RHE, the disappearance of the Co–S characteristic peak around 400 cm⁻¹ suggests a complete reconstruction of the catalyst within the Raman-detectable surface region. In contrast, both RC_76% and RC_83% exhibit the formation of active CoO₂ only when the voltage exceeds 1.55 V vs RHE. For RC_76%, the disappearance of the Co–S peak coincides with the generation of high-valence CoO₂, indicating significant sulfur–oxygen exchange at the surface. This phenomenon severely compromises the stability of the sulfide, causing its structure to degrade from surface to bulk, which explains the severe metal ion leaching observed during OER tests. Although RC_83% also forms its active phase at higher voltage, the ELF analysis reveals stronger bonding within the material, which helps maintain structural integrity. Consequently, although the Co–S peak intensity decreases with increasing voltage, a residual Co–S peak remains detectable under Raman conditions.

Then, in situ attenuated total reflection infrared (ATR-IR) spectra were further recorded within the OER potential range (Supplementary Fig. 29). The band around 1200 cm⁻¹ is ascribed to the O–O stretching mode of surface-adsorbed superoxide (OOH_ad)³². This means that the adsorbate evolution mechanism (AEM) dominated O₂ generation over the catalysts. We plotted the signal intensity of the OOH_ad peak against the electrode potential (Fig. 2c). It was found that as the applied potential changed, the signal intensity of RC_69% decreased the least, followed by RC_83%, while RC_76% exhibited a significant decrease when the voltage exceeded 1.55 V vs RHE. The broad band in the region of 3000-3800 cm^-1 was ascribed to the H₂O stretching mode and could be resolved into three distinct components, corresponding to three types of O–H stretching vibrations. The peak at approximately 3268 cm⁻¹, 3444 cm⁻¹, and 3584 cm⁻¹ are assigned to 4-coordinated hydrogen-bonded water (4-HB·H₂O), 2-coordinated hydrogen-bonded water (2-HB·H₂O), and K⁺ ion hydrated water (K·H₂O) respectively³³. Previous studies have shown that an increased frequency of ν_O–H corresponds to the decreased degree of hydrogen bonds, and the adsorption of 4-HB·H₂O on the surface favors the dissociation of H₂O and the occurrence of water oxidation³⁴. Clearly, when the applied voltage exceeds 1.55 V vs RHE, only the O–H stretching vibration peak center of RC_69% falls within the range of 4-HB·H₂O (Supplementary Figs. 30 and 31). Moreover, the vibrational frequency of the adsorbate varies as a function of electrode potential and has been attributed to the vibrational Stark effect³⁵. A steeper Stark slope indicates that the interfacial water molecules are closer to the active sites. The Stark slope for RC_76% is not perturbed so greatly compared with that of RC_69% and RC_83%, which means that for RC_76%, during its interaction with interfacial water molecules, there are fewer active sites available for the effective evolution of oxygen intermediates³⁶. Meanwhile, we utilized in situ UV-vis absorption spectroscopy to track the oxidation state changes of Co ions during the OER process and quantified the density of high-valence active species with a stepwise spectra electrochemical test conducted at 400 nm. The results showed that RC_69% exhibits a higher number of active sites across the entire OER reaction potential range (Fig. 2d–g and Supplementary Figs. 32 and 33). In addition, we analyzed the variation of the O–H stretching vibration peak intensity with electrode potential and found that the intensity of RC_76% exhibits a particularly sharp increase (Fig. 2c). Given its poor contribution to active oxygen intermediate evolution, we hypothesize that this is the primary reason for the intense sulfur–oxygen exchange on its surface, leading to structural collapse. The absence of a clear SO₄²⁻ absorption peak at approximately 1100 cm⁻¹ on the surface of RC_76% further supports this conclusion, as excessive sulfur–oxygen exchange likely resulted in severe surface corrosion, hindering the effective adsorption of ${{\rm{SO}}}_{4}^{2-}$³⁷.

Impact of bulk reconstruction differences on catalytic performance

Based on the above results, we identified that 1.55 V vs RHE is the threshold potential at which significant sulfur–oxygen exchange reactions are triggered on the catalyst surface. Accordingly, we employed in situ X-ray absorption fine structure (XAFS) to conduct time-dependent analysis on the bulk phase reconstruction and the valence states variations of the Co site for the catalysts under a constant potential of 1.55 V vs RHE (Supplementary Figs. 34–36). The Co K-edge XANES spectra reveal a prominent feature: upon applying voltage, the white line peak intensity of RC_69% significantly increases with test duration, followed by a similar but delayed response in RC_83% (Fig. 3a–c). This suggests that bulk reconstruction occurs quickly in RC_69% and then in RC_83%, resulting in a complete structure transition from sulfide to oxyhydroxide. However, RC_76% undergoes negligible bulk reconstruction under the same conditions, suggesting that 1.55 V vs RHE is insufficient to trigger bulk reconstruction in RC_76%. Furthermore, the Co valence state was unraveled by the investigation of Co K-rising edge position shifts at different potentials (Fig. 3d). The results show that RC_69% undergoes complete reconstruction into a stable CoOOH state within 32 min, while RC_83% requires 48 min. In contrast, RC_76% shows negligible changes in Co oxidation state under the same conditions, which aligns with the trend observed in the white line peak intensity. Pourbaix diagram obtained by DFT calculation also confirms that, under identical conditions, RC_69% exhibits a faster onset of CoOOH formation, which can be attributed to its inherent structural advantages (Supplementary Fig. 37). Thus, we propose that only when the surface sulfur–oxygen exchange is well-balanced by the activation of metal sites and the evolution of oxygen intermediates can the sulfides fully reconstruct into oxyhydroxides. Otherwise, sulfides are more likely to undergo surface-to-bulk decomposition and collapse.

Fig. 3: In situ monitoring of bulk structure evolution of catalysts during OER.

a–c Time-dependent contour map of XANES spectra collected at the cobalt K-edge of RC_69% (a), RC_76% (b), and RC_83% (c) during the OER process by potentiostatic electrolysis. d Co K-edge position from the XANES spectra as a function of the reaction time of RC_69%, RC_76%, and RC_83% at 1.55 V vs RHE. The Co K-edge positions are extracted from the half-height values of the normalized rising absorption edge. e Mass and current change of RC_69% measured by EQCM cyclic voltammetry in 1 M KOH at a scan rate of 5 mV s⁻¹. f, The overpotential at 10 mA cm⁻² of RC_69%, RC_76%, and RC_83% and DEMS signals of ³⁴O₂/³²O₂ collected from the reaction products for ¹⁸O-labeled catalysts in 1 M KOH. Here, catalysts refer to RC_69%, RC_76%, and RC_83% after 8 cycles of CV reconstruction and electrochemical reconstruction conditions consistent with the EQCM testing conditions. Source data are provided as a Source Data file.

The electrochemical quartz crystal microbalance (EQCM) was used to monitor the mass and current changes of three catalysts during the OER process (Supplementary Fig. 38). All three catalysts displayed an overall mass reduction trend within the testing conditions. However, the degree of mass loss differs among them. After 8 cycles of CV testing, the mass of RC_69% stabilizes, and its OER current remains almost unchanged, whereas the other two catalysts continue to experience varying degrees of mass and current decreases (Fig. 3e and Supplementary Figs. 39 and 40). Compared to RC_76% and RC_83%, which exhibit mass changes of 0.21 μg and 0.34 μg at the 8th CV cycle, respectively, RC_69% shows only a mass change of 0.05 μg, representing an order-of-magnitude improvement in mass retention on the electrode surface. Notably, the time required for RC_69% to stabilize under the wide voltage range in the CV test is almost the same as the time it takes to stabilize at a constant voltage of 1.55 V vs RHE, further demonstrating the critical role of crystallinity control in modifying the local structure of sulfides, endowing well catalytic activity and stability. To explain the differences in mass changes among the catalysts, differential electrochemical mass spectrometry (DEMS) was employed to assess the degree of lattice oxygen participation in the catalysts after 8 CV cycles (Supplementary Figs. 41 and 42). Following reconstruction, the extent of lattice oxygen involvement during the subsequent OER reaction process in the oxyhydroxide layer formed on the catalyst surface ranked in the order of RC_76% > RC_83% > RC_69% (Fig. 3f). Although previous studies have suggested that a moderate degree of lattice oxygen participation can enhance the OER performance of catalysts, excessive lattice oxygen involvement has also been reported to compromise the structural stability of catalysts. According to our experimental results, in sulfides that undergo significant sulfur–oxygen exchange and severe reconstruction, excessive lattice oxygen involvement tends to destabilize the oxyhydroxide structures formed on the surface, leading to a substantial loss of active sites.

Relationship between sulfide structural evolution and OER performance

Through in situ analysis of the structural evolution of CoS₂ with three kinds of crystallinities during OER, we have discovered that modifications in the local structure of the catalysts result in distinct sulfur–oxygen exchange capacities at the surface (RC_76% > RC_69% > RC_83%) and varying capacities for bulk phase participation in reconstruction (RC_69% > RC_83% > RC_76%). Both trends in structural evolution align with changes in the p-band center of sulfur in the surface and bulk. Evidently, the bulk phase reconstruction ability of the catalysts positively correlates with their OER performance, but catalysis mainly occurs on the catalyst’s surface. Considering the intense reconstruction behavior of sulfides and their bulk phase structural participation during OER, along with the general proportionality of surface and bulk structures, we have weighted the p-band center of sulfur in catalysts. This enabled us to calculate the overall p-band center for each catalyst and establish its correlation with OER performance (Fig.4a and Supplementary Table 4). We found that within a certain structural proportion range, the closer the overall p-band center of CoS₂ is to the Fermi level, the better its OER performance, even when the surface structure accounts for up to 60%.

Fig. 4: Relationship between sulfide structural evolution and OER performance.

a Relationship between the weighted p-band center of sulfur and the catalytic activity of sulfides. b Integrated COHP values of Co–S bonds at different positions in RC_69%, RC_76%, and RC_83%. The positions of Co1, Co2, and the S atom are shown in Fig. 1f. c Atomic fraction of sulfur as a function of the reaction time of RC_69%, RC_76%, and RC_83% at 1.55 V vs RHE. Atomic fractions were collected from HADDF-STEM images. The detection limit of the instrument for elemental content is 0.5%. Source data are provided as a Source Data file.

The actual active site structure of sulfides during the OER process has long been debated, and it remains unclear whether sulfur residues remain after reconstruction of sulfides. Based on CoS₂ models with three crystallinities, we utilized crystal orbital Hamilton population (COHP) to compare the strength of Co–S bonds in different samples (Supplementary Fig. 43). By calculating Co–S bonds at the same position, we found that, whether stretched or shortened, the Co–S bond in RC_69% exhibits the strongest bonding capability (Fig. 4b). Meanwhile, we used high-angle annular dark field scanning TEM (HAADF-STEM) to characterize the sulfur content of different catalysts during the OER process (Supplementary Figs. 44 and 45). When a voltage of 1.55 V vs RHE was applied, after 30 min of reaction, only RC_69% retained 0.89% sulfur, while the sulfur content in the other two catalysts had already fallen below the detection limit of the instrument (Fig. 4c). Therefore, we believe that the modification of the local structure endows RC_69% with Co–S bonding capability, which not only ensures its activity but also enhances its structural stability.

The crucial role of residual sulfur

To provide further insights into the distribution of residual sulfur and its role in regulating the structure of the reconstructed catalyst, we utilized XAS to investigate the oxidation state and coordination environment of metal in RC_69% reacted at 1.55 V vs RHE for 1 h, 5 h, and 10 h (denoted as RC_69%—1 h, RC_69%—5 h, and RC_69%—10 h). It can be observed that the Co K-edge energy of the catalysts at different reaction times nearly coincides with that of amorphous CoOOH, prepared by electrochemical oxidation (Fig. 5a and Supplementary Fig. 46). This is more intuitively observed in the magnified view of region 2. The pre-edge peak at around 7710 eV corresponds to the Co 1s–3d transition, and the intensity of the pre-edge peak is closely correlated to the centrosymmetry of the octahedron in CoOOH³⁸. As shown in the magnified view of region 1, the pre-edge peak intensity of catalysts after electrochemical reconstruction are higher than that of amorphous CoOOH and crystalline CoOOH, showing a lower degree of centrosymmetry of the octahedron in RC_69%—1 h, RC_69%—5 h, and RC_69%—10 h. Furthermore, the rising edge and the white line are generally related to the electron jumping from 1s to 4p orbitals³⁸. A more delocalized 4p orbital is typically represented by a broader white line with lower intensity. Therefore, these results shown in the magnified view of region 3 suggest that the 4p orbitals become delocalized in RC_69%—1 h, RC_69%—5 h, and RC_69%—10 h. To summarize, RC_69% achieves complete structural transformation within 1 h from CoS₂ to CoOOH, or potentially as early as 32 min, supported by the nearly identical Co K-edge energy shown in the in situ XANES results. Importantly, the reconstructed structure remains stable, showing no changes with prolonged reaction times. This is further validated by the corresponding continuous Cauchy wavelet transform (CCWT) results (Fig. 5b). Subsequently, we conducted a comparative analysis using HAADF-STEM to investigate the sulfur content and distribution in samples at various reaction times (Fig. 5c and Supplementary Figs. 47–49). Based on the elemental mapping and their corresponding extraction of the elemental distribution curve, residual sulfur particles are predominantly located within the bulk of the catalysts. The sulfur content in RC_69%—1 h, RC_69%—5 h, and RC_69%—10 h was 0.85%, 0.9%, and 0.89% (Fig. 5d), respectively, which is nearly identical to the sulfur content in RC_69% after reaction for 30 min. Therefore, we propose that the residual sulfur remains stable in the bulk of RC_69% after reconstruction and substitutes some of the oxygen sites in CoOOH, leading to a decrease in the symmetry of the Co–O octahedron. Based on these findings, we performed a comprehensive analysis of the existence form of residual sulfur and explored the regulation of its content by increasing the reaction potential and extending the reaction time. The results demonstrate that the sulfur content, corresponding to a small amount of sulfur chemically bonded with Co and stable within the bulk phase, remains invariant under varying external conditions (Supplementary Figs. 50–53).

Fig. 5: Identification of the catalytically active phase and the distribution of residual sulfur.

a Ex situ XANES spectra collected at the cobalt K-edge of catalysts. b Corresponding Co K-edge EXAFS-CCWT images for RC_69%—1 h, RC_69%—5 h, and RC_69%—10 h. c HAADF images, elemental mapping, and the corresponding elemental distribution curve extracted from elemental mapping of RC_69%—1 h. d Atomic fraction of oxygen and sulfur of RC_69%—1 h, RC_69%—5 h, and RC_69%—10 h. Atomic fractions were collected from HADDF-STEM images. Source data are provided as a Source Data file.

Given the structural consistency and similar sulfur content of RC_69%—1 h, RC_69%—5 h, and RC_69%—10 h, RC_69%—1 h was selected as a representative catalyst for further study. To elucidate the role of residual sulfur in RC_69%—1 h on its OER performance and structural modulation following a reduction in Co–O octahedral symmetry, we evaluated the OER performance of each catalyst using a CP-CV system (Supplementary Figs. 54–56). A plot of the steady-state voltage during CP testing vs the number of cycles revealed that RC_69%—1 h and RC_69% exhibited comparable OER performance, both outperforming the amorphous and crystalline CoOOH prepared directly (Supplementary Fig. 57).

Based on the identification of the active phase, we conducted PDOS to further investigate the impact of changes in the coordination environment of the catalyst’s active center on its electronic rearrangement behavior (Fig. 6a and Supplementary Note 2). The Co sites in bulk CoOOH exhibit symmetric and well-ordered bond lengths and angles, resulting in the classic e_g and t_2g splitting of the five d-orbitals under the octahedral field³⁹. Since catalytic reactions mainly occur on the surface of catalysts, when crystalline CoOOH undergoes catalysis, the surface unsaturated Co sites often serve as catalytic centers (Supplementary Data 1). Taking the surface pyramidal Co site as an example, the absence of one coordinated oxygen atom in the z-direction leads to the decreased d-orbital degeneracy⁴⁰, with the orbital and electron configurations shown in Fig. 6b, c. Moreover, according to the Co–O bond length of RC_69%—1 h and amorphous/crystalline CoOOH (Supplementary Fig. 58 and Supplementary Table 5), we stretched the bond lengths of amorphous CoOOH in the computational model and found that the bond length changes were more significant in the fully coordinated xy plane, causing a corresponding decrease in the energy of the d_xy orbital. Then we replaced one oxygen atom in the z-axis with a sulfur atom to build the model of RC_69% —1 h based on the above structure. Due to the significantly larger atomic radius of sulfur compared to that of oxygen, the degree of geometric symmetry breaking in the system further increases, leading to a reduction in the degeneracy of the d-orbitals and an increase in electron spin. Except for d_xy orbitals, which do not bond with oxygen intermediates, all other bonding orbitals have one electron. The unique electronic distribution in the bonding orbitals of RC_69%—1 h enables ideal adsorption behavior of oxygen intermediates from the x, y, and z directions when interacting with the active sites. These not only avoid intermediate poisoning due to overly strong adsorption but also prevent catalyst deactivation caused by weak adsorption, ultimately enhancing the oxygen intermediate evolution ability and reducing the rate-determining step energy barrier, thus improving the OER activity of the catalyst (Supplementary Fig. 59).

Fig. 6: Structural regulation of the catalytically active phase by residual sulfur.

a PDOS of Co in crystalline CoOOH, amorphous CoOOH, and RC_69%—1 h, respectively. b Structural models of bulk CoOOH, crystalline CoOOH, amorphous CoOOH, and RC_69%—1 h (from left to right). c d-Electron configurations of Co at Co–O octahedral of bulk CoOOH, crystalline CoOOH, amorphous CoOOH, and RC_69%—1 h. d Polarization curves of the RC_69%-based AEM at different reaction temperatures. e The CP curves of RC_69%-based AEM electrolyzer operated at 1 A cm⁻². Source data are provided as a Source Data file.

Finally, to further emphasize the critical role of crystallinity control in regulating the structural evolution of sulfides and consequently influencing the stability of residual sulfur, as well as to validate the practical application potential of the catalyst in industrial electrocatalytic water splitting for hydrogen production, we evaluated the stability of RC_69% under a three-electrode system at a current density of 100 mA cm⁻². Structural changes before and after the test were systematically analyzed (Supplementary Figs. 60 and 61). The results revealed that after the stability test, the material retained its identity as residual sulfur-stabilized amorphous CoOOH, with no significant alteration in morphology. Based on these findings, we utilized RC_69% as the anode catalyst and Pt/C as the cathode catalyst to assess the catalytic performance of this material within an anion exchange membrane (AEM) electrolyzer (Fig. 6d). Furthermore, its stability was evaluated under high current density conditions (Fig. 6e). The test results indicate that the catalyst can sustain stable operation for 500 h at a current density of 1 A cm⁻², with no noticeable performance degradation observed.

Discussion

In summary, we prepared three CoS₂ catalysts with different crystallinities and provided a platform for investigating the mechanistic origin of the varying degrees of sulfide reconstruction. Experimental and theoretical calculations confirmed that CoS₂ with different local structures possesses distinct Co–S bonding strength, which leads to significant differences in the structural evolution of the surface and bulk of the catalysts during the OER process. Among them, RC_69% exhibited bulk reconstruction and active phase transition capabilities, as it could quickly activate metal sites and evolve key oxygen intermediates, thereby displaying favorable OER performance. Conversely, the excessive sulfur–oxygen exchange on the surface of RC_76% could not be efficiently utilized by the active phase transition. This not only inhibited the transition of the bulk structure to the active phase but also caused a progressive structural collapse from the exterior inward, leading to a decline in stability and activity. Considering the sulfur–oxygen exchange capacity on the catalyst surface (RC_76% > RC_69% > RC_83%), bulk reconstruction ability (RC_69% > RC_83% > RC_76%), and overall catalytic performance (RC_69% > RC_83% > RC_76%), we established a relationship between the weighted p-band center of sulfur and the catalytic activity of CoS₂ within a specific structural range, revealing the closer the weighted p-band center of sulfur is to the Fermi level, the better the catalyst’s OER performance. Finally, by combining HADDF-STEM and XAS to characterize the metal coordination environment and the distribution of residual sulfur in the reconstructed samples, we conducted PDOS to further investigate the impact of changes in the coordination environment of the catalyst’s active center on its electronic rearrangement behavior. The larger atomic radius of sulfur disrupts the Co–O octahedron symmetry and reduces the d-orbitals degeneracy, leading to an increase in electron spin. These orbital distributions and electron occupancies of the catalyst ultimately enhance the oxygen intermediate evolution ability and reduce the rate-determining step energy barrier, thus improving the OER activity of the catalyst. Beyond sulfides, our work highlights the role of crystallinity-crystallinity discrepancy influence on the electronic states of water oxidation catalysts and the subsequent catalytic structural evolution behavior.

Methods

Chemicals

Co(NO₃)₂·6H₂O (99.0 %), CoSO₄ (99.0 %), urea (99.0%), KOH (99.0%), sublimation sulfur (99.0%), C₂H₅NS (99.0%), and ethanol (99.0%) were purchased from Aladdin. The demonized (DI) water was obtained from a Millipore Autopure system (18.2 MΩ, Millipore Ltd., USA). The commercial carbon fiber paper (CFP) was purchased from Fuel Cell Store. Argon gas was of 5 N quality (99.999%, Air gas). All the materials were used without any pre-treatment or purification.

Catalyst preparation

Co(NO₃)₂·6H₂O (2.008 g) and urea (0.48 g) were added in 80 mL DI water and stirred for 20 min to form a homogenous solution. The above solution was transferred into a 100 mL Teflon-lined autoclave, and a piece of hydrophilic CFP (3 × 4 cm²) was placed vertically in the autoclave. The autoclave was sealed and heated at 120 °C for 16 h and then cooled to room temperature naturally. After being washed with deionized water and ethanol three times, respectively, the cobaltous dihydroxycarbonate/CFP was dried in a vacuum oven at 60 °C overnight. Then, the cobaltous dihydroxycarbonate/CFP was used to synthesize CoS₂ catalysts with different crystallinities by adjusting the evaporation rate of sulfur powder based on annealing temperature control. 500 mg sulfur powder and one piece of cobaltous dihydroxycarbonate/CFP were placed in two crucibles, which are 20 cm apart in the tube. Then the tube was heated at 350 °C, 400 °C, and 450 °C with a rate of 10 °C min⁻¹ for 2 h under the flowing Ar atmosphere. Finally, the system was cooled to 50 °C with a rate of 100 °C h⁻¹ under a flowing Ar atmosphere, and the RC_69%, RC_76%, and RC_83% nanosheets could be obtained, respectively. Co_xS_y—300 °C and Co_xS_y—500 °C were prepared via the same procedure, only annealing at 300 °C and 500 °C. C₂H₅NS (0.3 g) was added to 60 mL of ethanol and stirred for 10 min to form a homogenous solution and transferred into a 100 mL Teflon-lined autoclave. Then a piece of cobaltous dihydroxycarbonate/CFP was placed vertically in the autoclave and heated at 120 °C for 5 h, and then cooled to room temperature naturally. After being washed with deionized water and ethanol three times, respectively, the amorphous Co_xS_y was dried in a vacuum oven at 60 °C overnight. Amorphous CoOOH was fabricated onto the CFP substrates in 0.2 M CoSO₄ aqueous electrolyte at a constant anodic deposition potential of 1.60 V in a three-electrode electrochemical cell. Co(NO₃)₂·6H₂O (0.5821 g) and urea (0.6006 g) were added to a 50 mL H₂O₂ solution with a mass fraction of 15% and stirred for 15 min to form a homogenous solution. The above solution was transferred into a 100 mL Teflon-lined autoclave and heated at 150 °C for 8 h and then cooled to room temperature naturally. After the reaction, the final products were centrifuged at 10,000 rpm (11,800×g) for 3 min and washed with water and ethanol to obtain crystalline CoOOH.

Structural characterization

XRD measurements were conducted on a Rigaku MiniFlex 600 diffractometer with Cu Kα radiation (λ = 0.1542 nm) from 15° to 75° under a constant voltage of 40 kV and a current of 15 mA. Catalyst morphologies were analyzed by ThermoFisher Apreo S field-emission scanning electron microscopy (FESEM) at an acceleration voltage of 30 kV. TEM and high-resolution transmission electron microscopy (HR-TEM) images were obtained on a Tecnai G2 F30 field emission transmission electron microscope, and the corresponding data processing software is Gatan GMS3. HADDF-STEM images were recorded on a probe aberration-corrected STEM (Cubed Titan G2 60-300, FEI, USA) operated at 300 kV. XPS analysis was made with a Kratos Axis Supra device and a VG ESCALAB 220I-XL device. All XPS spectra were corrected using the C1s line at 284.8 eV. ICP-MS analyses were performed on an iCAPQc spectrometer. Ex situ Co K-edge XAFS data were collected in transmission mode using a Si (111) double-crystal monochromator at the 1W1B station of the Beijing Synchrotron Radiation Facility (BSRF).

Electrochemical analyses

The continuous CP-CV (with 90% iR-correction) tests were performed in O₂-saturated 1.0 M KOH (pH = 13.97 ± 0.04, measured by Hanna precision pH meter) on a Bio-Logic (SP-150) using a standard three-electrode system. CV tests with iR-correction (85%) for averaging the positive-going and negative-going scans to obtain LSV plots⁴¹. The electrochemical impedance spectroscopy (EIS) measurements were conducted in a frequency range from 10⁵ to 0.01 Hz with an amplitude of 5 mV. The measured resistance was reported as 2.0 ± 0.1 Ω. The long-time stability test was performed in O₂-saturated 1.0 M KOH on CHI-760E. The 1 M KOH electrolyte was prepared by accurately weighing 1 mole of KOH and dissolving it in 1 L of DI H₂O. The solution was stirred until complete dissolution was achieved and then allowed to cool to room temperature. It was subsequently stored in a plastic bottle under ambient conditions for use in the experiments. Electrochemical measurements were performed using a Hg/HgO as reference electrode, a Pt plate as counter electrode, and the catalysts grown on the CFP served as working electrode (2 × 0.5 × 1 cm²) in 1 M KOH. The loading amount of the catalyst can be determined by comparing the mass of the CFP before and after loading, which is approximately 0.2 mg cm⁻². The reference electrode was calibrated within a three-electrode system, where platinum plates were employed as both the working electrode and the counter electrode, in an H₂-saturated 0.1 M HClO₄ electrolyte solution. All potentials were recorded vs a reversible hydrogen electrode (RHE), according to the Nernst equation (E_RHE = E_Hg/HgO + 0.098 V + 0.059 × pH). It needs to be noted that the voltages applied in all in situ characterizations without iR-correction. The Tafel slope was calculated by the equation below: η = a + b × logj. Where η stands for the overpotential, b stands for the Tafel slope, and j stands for the current density. The electrochemically active surface area (ECSA) of the catalyst was calculated by the equation: ECSA = C_dl/C_s, where C_s is the specific capacitance of a planar surface with an atomically smooth under identical electrolyte conditions. We use the general value of 40 μF cm⁻². TOF was calculated using the formula: TOF = (j × S)/(4 × n × F), where j represents the current density of the catalyst at 1.55 V (mA cm⁻²), S denotes the geometric surface area of the working electrode (cm²), F is the Faraday constant (96485.3 C mol⁻¹), and n corresponds to the mole number of active metal atoms on the working electrode. We assumed that all Co atoms participated in the catalytic reaction process in this work. The FE of the catalyst during the OER process was calculated using the following formula: FE = I_r/I_d × N, where I_d represents the disk current, I_r represents the ring current, and N denotes the current collection efficiency of the RRDE. Please refer to the supplementary information for details.

AEMWE tests

The AEM was assembled with RC_69% as anode, with a loading of 2 mg cm⁻². Commercial Pt/C (70 wt% Pt) was utilized as the cathode catalyst (0.5 mgPt cm⁻²). Ni mesh was used as a gas diffusion layer. Versogen-80 was used as the AEM (5 cm²). The anode and cathode catalyst inks were separately prepared by dispersing the catalyst in a mixture of Nafion (5%), isopropanol, and distilled water. All inks were sprayed onto the AEM and immersed in 1 M KOH for 12 h before the assembly test. The catalyst-loaded AEM and gas diffusion layers fabricate the membrane electrode assembly (MEA), which was sandwiched by two Ni bipolar plates to assemble the AEMWE device. Both Ni bipolar plates featured parallel flow fields of 5 cm², with a reaction area of 5 cm². The AEMWE was operated at 70 °C using 1 M KOH as the electrolyte, delivered to the anode via a peristaltic pump. All measurements in the AEMWE were recorded by testing a power supply (NEWARE, CT-4004-5V100A) without iR-correction.

In situ Raman and ATR-IR measurements

Raman spectroscopy measurements were carried out on a LabRAM HR Evolution spectrophotometer with a 532 nm wavelength of the excitation light source. In situ Raman measurements were performed using a customized cell (031-2H) with a saturated Ag/AgCl reference electrode and a Pt ring counter electrode by changing the constant voltage in 1 M KOH. In situ ATR-FTIR was conducted on a BRUKER-Fourier Transform Infrared Spectrometer-TENSOR27 using a customized cell. The remaining test conditions are consistent with those used in in situ Raman. In situ electrochemical characterization was carried out at a specific potential for 10 min to obtain the surface chemical composition and structural information of materials.

In situ EQCM measurements

For EQCM measurements, it was assembled in a temperature-controlled eCell micro-chamber (Hettich, Germany) connected to an EQCM 15 M (Gamry, USA). An AT-cut quartz crystal with a fundamental frequency of 9.12 MHz ± 50 kHz coated was used. The Au polycrystals coated with thin films of catalysts by spin coating deposition were used as the working electrode. The working electrode was then mounted into a three-compartment electrochemical cell. An Ag/AgCl electrode and a Pt wire were used as the reference electrode and counter electrode, respectively. The frequency change was measured by a Q-sensor analyzer (QE 401) equipped with the QEC 401 electrochemistry module. The frequency transition signal of the potential is obtained through steady-state polarization and cyclic voltammetry. EQCM results were collected using Gamry Resonator software. For the treatment of data, the Sauerbrey equation was used to determine the deposited mass based on the parallel frequency (fp). Sauer-brey equation: ∆f = −Cfm; ∆f: Frequency variation. Cf: Correction factor (theoretical correction factor for 5 MHz chips: 56.6 Hz·cm²)/μG·m: quality changes occurring on the surface of the chip.

In situ DEMS measurements

The DEMS results were obtained by a QAS 100 device (Linglu Instruments, Shanghai). Typically, the working electrode was prepared by dropping ink on the surface of a PTFE film covered with a layer of Au. The content of the dropping ink of the catalyst was 0.051 mg cm⁻². A Pt wire was used as the counter electrode, and an Ag/AgCl electrode was used as the reference electrode. Then the catalysts were labeled with 1 M KOH diluted with H₂¹⁸O by performing CV cycles at the scan rate of 5 mV s⁻¹. Thus, the produced gas can be probed in real time. The mass signals of oxygen with different molecular weights were observed. Therefore, the participation of lattice oxygen from the catalyst in OER would be shown as ³⁴O₂.

In situ XAFS measurements

In situ XANES data were collected at the XAFCA beamline of Singapore Synchrotron Light Source (SSLS) using fluorescence mode⁴². Hg/HgO was used as the reference electrode and a carbon rod as the counter electrode. The storage ring of SSLS was running at 0.7 GeV with an electron current of 200 mA. Energy calibrations were carried out using a standard Co foil. Athena was used for energy calibration, background removal, and Fourier Transform (FT). Artemis was used for fitting⁴³.

Theoretical calculation

AIMD simulations were carried out with the CP2K simulation package⁴⁴. At the start of MD simulations, the CoS₂ module was assigned an initial temperature of 100 K, then the module was heated up to 5000 K by velocity scanning over 3 ps. The spin-polarized PBE exchange-correlation functional was adopted with double-ζ Gaussian basis sets⁴⁵, which were used for the valence electrons (3s² 3p⁴ for S and 3s² 3p⁶ 3d⁷ 4s² for Co). An auxiliary plane-wave basis set with a cutoff set to 500 Ry was used for computing the electrostatic terms. We select CoS₂ models with different crystallinity at 100 K, 1500 K, and 3000 K for subsequent calculation, as reported in the previous work method^46,47,48. All the static spin polarized calculations were performed in the framework of the density functional theory with the projector augmented plane-wave method, as implemented in the Vienna ab initio simulation package⁴⁹. The nuclei-electron and the electron exchange correlation interactions were described by the projector augmented wave (PAW) potentials with the Perdew–Burke–Ernzerh of (PBE) functional, respectively^50,51. The cutoff energy for plane-wave basis functions was set to 500 eV with the energy change convergence criterion of 1 × 10⁻⁵ eV. For CoS₂, we selected a bulk model with space group Pa3 and constructed a 5-layer Co slab with (200) facet, consistent with experimental characterization. The optimized unit cell lattice parameters are a = 11.54 Å, b = 11.54 Å, c = 27.74 Å, with angles α = 90°, β = 90°, and γ = 90°. For CoOOH, we adopted a bulk model with space group R-3m and created a 4-layer Co slab with the (012) facet, aligning with experimental characterization. The optimized lattice parameters are a = 8.55 Å, b = 14.05 Å, c = 24.24 Å, with angles α = 90°, β = 90°, and γ = 90°. To simulate amorphous CoOOH, we expanded the lattice by 10% to elongate bond lengths, resulting in optimized parameters of a = 9.40 Å, b = 15.45 Å, c = 26.67 Å, with angles α = 90°, β = 90°, and γ = 90°. Atomic positions were allowed to relax until the sum of the absolute forces reached down to 0.02 eV Å⁻¹. A vacuum layer of 15 Å is added perpendicular to the sheet to avoid artificial interaction between periodic images. The Monkhorst–Pack k-point samplings were set as 3 × 3 × 1 for the geometry optimization, and electronic structure. The Hubbard-U correction method was applied to improve the description of localized Co d-electrons with U = 3.40⁵². The COHP was calculated using Lobster^53,54. The long-range van der Waals interaction was also included in surface calculations⁵⁵.