Introduction

Solar-driven conversion of carbon dioxide (CO2) into multi-carbon (C2+) solar fuels has garnered significant attention as a promising green solution to address global energy demands and mitigate the impacts of climate change1,2,3. Although recent advances have demonstrated the potential of photocatalytic systems for CO2-to-C2+ conversion4, achieving high selectivity remains a critical challenge, because the formation of C2+ products relies heavily on the stability of carbon-based intermediates at the catalytic active sites to enable efficient C–C coupling5,6,7.

Among various developed photocatalysts for CO2 reduction reaction (CO2RR), transition metal sulfides, such as CdS, Bi2S3, CulnSnS4, have attracted particular interest due to their tunable d-orbitals8,9,10. However, these materials often exhibit weak adsorption of C1 intermediates, e.g., *CO, leading to premature desorption11,12,13 and preferential formation of C1 products, such as CO, CH4, and HCOOH, rather than the desired C2+ compounds (Fig. 1a).

Fig. 1: Schematic diagram showing regulatory effect of coordination environment on *CO adsorption.
Fig. 1: Schematic diagram showing regulatory effect of coordination environment on *CO adsorption.
Full size image

a On the pristine ZnS, the weak adsorption of *CO intermediates reduce the surface covering and then yields C1 product. b The low-coordinated Mn single-atom sites enhance the adsorption of *CO intermediates via charge modulation, facilitating asymmetric C–C coupling.

To address this limitation, several effective strategies, including heterojunction construction, crystal facet engineering, doping, and defect engineering, have been explored to enhance C2+ selectivity by stabilizing carbon-based intermediates14,15,16,17. For instance, in MoSx/Fe2O3 system, d-p orbital hybridization at Mo-Fe sites reduced electrostatic repulsion between *CO and *COH intermediates, promoting ethylene (C2H4) formation18. Co-O-Fe triatomic sites in partially oxidized FeCoS2 increased local charge density, enriching C1 intermediates and directing the reaction pathway toward C2 products19. Incorporation of Co into NiS2 tailored the coordination number and oxide state of Ni sites, facilitating asymmetric C–C coupling between *CO species in atop and bridge adsorption configuration, achieving 75% selectivity for C2H420. Despite these advances, achieving high activity and selectivity for photocatalytic CO2-to-C2+ conversion remains an unmet goal.

Single-atom catalysts (SACs), with atomically dispersed metal centers and well-defined coordination environments, provide a powerful platform for tuning intermediate adsorption and enhancing CO2RR activity and selectivity18,21,22,23,24. Particularly, low-coordination SACs can break geometric symmetry and modulate the local electronic structure, strengthening the adsorption of carbon-based intermediates and facilitating C–C coupling25,26. Therefore, integrating low-coordination SACs into binary sulfides holds great potential for stabilizing *CO and enabling efficient photocatalytic CO2-to-C2+ conversion (Fig. 1b).

Herein, we develop a low-coordination Mn single-atom embedded zinc sulfide (Mn1–ZnSv) via a microwave irradiation-induced targeted defect engineering strategy for enhancing CO2-to-C2H4 photoreduction. Structural analyses reveal that Mn atoms are atomically dispersed and coordinated with a reduced number of sulfur atoms, forming Mn–Zn coupled sites with modulated electronic structures. Density functional theory (DFT) calculations and in-situ spectroscopic studies demonstrate that these low-coordination Mn sites effectively stabilize C2H4-related intermediates and promote asymmetric *CO–CHO coupling into *COCHO, a key intermediate for C–C bond formation. As a result, Mn1–ZnSv exhibits attractive photocatalytic performance for CO2 reduction, achieving a C2H4 production rate of 76.6 μmol g−1 h−1 with nearly 100% selectivity, significantly emphasizing the capability of regulating the coordination environment of sing-atom catalytic sites for photoconversion CO2 into C2+ products.

Results

Enhanced C–C coupling via low-coordination Mn sites

To assess *CO formation during the CO2RR, we calculated the free energy profiles for CO2-to-*CO conversion on ZnS, Mn1–ZnS, and Mn1–ZnSv (Fig. 2a, Supplementary Figs. 1–3, and Supplementary Data 1). Among these, Mn1–ZnSv exhibits the lowest energy barrier (0.56 eV), significantly lower than that of ZnS and Mn1–ZnS (Fig. 2b), indicating that low-coordination structure of Mn single-atom is favorable for CO2 activation and *CO generation.

Fig. 2: Theoretical guidance.
Fig. 2: Theoretical guidance.
Full size image

a Schematic of calculation models for pristine ZnS, Mn1–ZnS, and Mn1–ZnSv (the atoms marked with numbers represent metal atoms in different coordination states). b The free energy diagram for CO2 reduction reaction to display the conversion of CO2-to-*CO over ZnS, Mn1–ZnS and Mn1–ZnSv. c Illustration of the interaction between the 5σ and 2π* of *CO and metal surfaces. d Adsorption energies of *CO on Mn1–ZnS and Mn1–ZnSv. e, f Differential charge density and Bader charge of *CO adsorbed on Mn1–ZnSv (e) and Mn1–ZnS (f) (the isosurface is 0.005 e/Å3). Blue and purple represent charge gaining and losing, respectively. g, h The partial density of states and the d-band center of the Mn atom for Mn1–ZnS and Mn1–ZnSv, εd represents the d-band center. i The free minimum-energy diagram for *CO hydrogenation and subsequent C–C coupling on Mn1–ZnSv (111) facet. Source data are provided as a Source Data file.

To investigate the behavioral of *CO intermediates on different coordination structures, the *CO adsorption energy and protonation energy barriers of Mn1–ZnS and Mn1–ZnSv were compared. The interaction between the Mn 3d orbital and the 2π* (LUMO) and 5σ (HOMO) orbitals of *CO underpins the modulation of adsorption behavior (Fig. 2c). The low-coordination Mn site in Mn1–ZnSv strengthens the Mn-CO π backbonding, resulting in a significantly enhanced *CO adsorption energy of −1.36 eV, much stronger than that of coordination-saturated Mn1–ZnS (Fig. 2d). Furthermore, the incorporation of low-coordination Mn single atoms ingeniously balances the competitive adsorption of *CO intermediates and *H species (Supplementary Fig. 4). The appropriate binding inhibits premature *CO desorption and enables its participation in subsequent protonation steps. It is worth noting that the energy barrier for *CO hydrogenation on Mn1–ZnSv is reduced to 0.75 eV, compared to 1.25 eV on Mn1–ZnS, confirming the regulatory role of the low-coordination structure in steering the *CO reaction pathway (Supplementary Fig. 5).

To probe the origin of the stronger *CO adsorption, we investigated the electronic structure differences between Mn1–ZnS and Mn1–ZnSv. Differential charge density analysis reveals a significant electron transfer (1.08 |e|) from the Mn single atom to the adsorbed *CO in Mn1–ZnSv (Fig. 2e), compared to just 0.32 |e| in Mn1–ZnS (Fig. 2f), highlighting the enhanced electronic interaction in the low-coordination structure. Furthermore, in Mn1–ZnSv (coordination number, CN = 2), the Mn 3d orbital center shifts upward relative to Mn1–ZnS (Fig. 2g), resulting in a higher d-band center (Fig. 2h). This upshift implies an increased population of anti-bonding states, favoring stronger adsorption of reaction intermediates and facilitating the key C–C coupling process.

To validate this mechanistic insight, we calculated the free energy barriers for C–C coupling through different C1 species (*CO, *CHO, *COH) on Mn1–ZnSv (Fig. 2i). The results suggested that direct dimerization of two *CO species into *COCO is energetically unfavorable (ΔG = 1.91 eV). Instead, *CO preferentially undergoes hydrogenation to *CHO, which subsequently couples with another *CO to form the *COCHO intermediate via an asymmetric C–C bond formation pathway (Supplementary Fig. 6). This pathway is thermodynamically more favorable and constitutes the lowest-energy pathway for CO2-to-C2H4 conversion (Supplementary Figs. 7 and 8 and Supplementary Table 1). Thus, low-coordination Mn single-atom sites in Mn1–ZnSv significantly enhance *CO adsorption and facilitates the asymmetric coupling of *CO and *CHO, ultimately leading to the selective generation of C2H4. The theoretical screening of coordination structures serves as a guide for establishing trends in *CO adsorption and the subsequent reaction pathways.

Catalyst synthesis and characterization

Guided by the DFT-screened Mn1–ZnSv structure, a Mn single-atom substituted ZnS catalyst with targeted sulfur vacancies (Sv) was successfully synthesized via a microwave irradiation-induced targeted defect strategy (Supplementary Fig. 9), followed by H2O2 etching under an Ar atmosphere. Inductively coupled plasma optical emission spectrometry (ICP-OES) confirmed a Mn content of 0.6 wt% (Supplementary Table 2). X-ray diffraction (XRD) patterns (Supplementary Fig. 10) confirmed the formation of cubic-phase ZnS (PDF #05-0566)27 with an average crystallite size of 41.3 ± 0.2 nm, as estimated using the Scherrer equation (Supplementary Fig. 11).

High-resolution transmission electron microscopy (HRTEM, Fig. 3a) revealed distinct lattice fringes with an interplanar spacing of 0.312 nm, corresponding to the (111) plane of ZnS. Scanning transmission electron microscopy–energy-dispersive X-ray spectroscopy (STEM-EDS) elemental mapping images (Fig. 3b) and the STEM-EDS line scanning profiles (Supplementary Fig. 12) corroborated the even distribution of Zn, Mn, and S throughout the sample, with no observable aggregation of Mn into large clusters or particles. Aberration-corrected annular dark-field STEM (ADF-STEM, Fig. 3c) further confirmed the atomic dispersion of Mn, with isolated single atoms highlighted by yellow circles and localized intensity dips in the corresponding line-scan profile across a representative region.

Fig. 3: Structural characterization.
Fig. 3: Structural characterization.
Full size image

a, b TEM image (a), STEM image and the corresponding EDS elemental mappings (b) of Mn1–ZnS. Inset: An enlarged HRTEM of the dotted line area. c ADF-STEM image of Mn1–ZnSv with corresponding intensity profile of the representative region (red dotted line), and the isolated single atoms highlighted with yellow circles. d Normalized Mn K-edge XANES spectra of Mn1–ZnS, Mn1–ZnSv, and the related reference samples. e Fourier-transform (FT) spectra of EXAFS oscillations for Mn1–ZnS and Mn1–ZnSv. f, g EXAFS spectrum fitting result of Mn1–ZnS (f) and Mn1–ZnSv (g) in R space. (The inserted image shows the atomic structure model of Mn1–ZnS and Mn1–ZnSv, where the balls in blue, pink, and yellow represent Zn, Mn, and S atoms). h, i Wavelet transform (WT) contour plots of Mn foil (h) and Mn1–ZnSv (i), respectively. Source data are provided as a Source Data file.

To confirm the formation of Sv, electron paramagnetic resonance (EPR) spectroscopy was performed (Supplementary Fig. 13). A clear signal at g = 2.005 indicated the presence of Sv in Mn1–ZnSv28. Additionally, EDS point analysis (Supplementary Fig. 14) showed an increased metal-to-sulfur atomic ratio, further supporting the generation of Sv, which increased with prolonged H2O2 etching time. Quantitative analysis from ICP-OES and EPR29,30 (Supplementary Figs. 15 and 16 and Supplementary Table 2) revealed an isolated Sv concentration of 1.54% for Mn1–ZnSv, closely matching the theoretically identified optimal configuration by DFT (Supplementary Fig. 17 and Supplementary Data 1).

To probe the local coordination environment of Zn and Mn atoms, X-ray absorption spectroscopy was conducted. X-ray absorption near-edge structure (XANES) spectra (Supplementary Fig. 18a) showed that Zn maintained an oxidation state of approximately +2.0. The Mn K-edge positions in both Mn1–ZnS and Mn1–ZnSv were located between those of metallic Mn and MnO, indicating an average Mn valence between 0 and +2 (Fig. 3d). Notably, Mn in Mn1–ZnSv exhibited a lower oxidation state than in Mn1–ZnS, attributed to the low-coordination environment induced by Sv formation.

Fourier-transformed extended X-ray absorption fine structure (FT-EXAFS) analysis (Fig. 3e, Supplementary Fig. 18, and Supplementary Table 3) displayed distinct peaks corresponding to Zn–S and Zn–Zn coordination, while a prominent peak at 1.96 Å was assigned to Mn–S bonding. The absence of Mn–Mn and Mn–O coordination in both Mn1–ZnS and Mn1–ZnSv indicates that Mn atoms are atomically dispersed and coordinated with neighboring S atoms, consistent with ADF-STEM results31.

EXAFS fitting revealed that the first-shell Mn–S coordination number was 3.7 for Mn1–ZnS and 1.9 for Mn1–ZnSv (Fig. 3f, g and Supplementary Table 3). These values are in excellent agreement with DFT-optimized structures, confirming that Mn atoms in Mn1–ZnSv are coordinated with only two sulfur atoms (insets in Fig. 3f, g). No Mn–Mn path in second-shell coordination (Supplementary Fig. 19) and wavelet transform (WT) analysis further confirmed the atomic dispersion and low-coordination environment of Mn in Mn1–ZnSv (Fig. 3h, i).

To investigate the effect of low-coordination Mn single-atoms on light absorption efficiency, UV–Vis diffuse reflectance spectra and X-ray spectroscopy (XPS) valence band spectroscopy were conducted. The optical absorption band edge of Mn1–ZnSv is observed at 428 nm, corresponding to a smaller band gap of 2.88 eV (Supplementary Fig. 20). Ultraviolet photoelectron spectroscopy measurements revealed work functions (Φ) of 6.13 and 5.60 eV for Mn1–ZnS and Mn1–ZnSv, respectively (Supplementary Fig. 21). The valence band maxima for Mn1–ZnS and Mn1–ZnSv were located at 1.00 and 0.55 eV below the Fermi level (Supplementary Fig. 22), and the electronic band structure versus Normal Hydrogen Electrode could be elucidated (Supplementary Fig. 23), which is thermodynamically favorable for the CO2-to-C2H4 conversion (Supplementary Fig. 24). These results are well-consistent with those obtained from Mott–Schottky plots (Supplementary Fig. 25), clearly suggesting the ability for CO2-to-C2H4 conversion and O2 evolution. Upon irradiation, photogenerated electrons are transferred from S (electron donor) to Mn (electron trap), and then rapidly injected into the adsorbed CO2 molecules through the electron transfer channel (Supplementary Fig. 26). Moreover, improved charge-transfer dynamics induced by low-coordination Mn sites were recorded in Supplementary Figs. 27–29.

To further elucidate intermediate adsorption and reaction pathways in photocatalytic CO2 reduction, in situ diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) was employed (Fig. 4a, b and Supplementary Table 4). Under simulated solar irradiation, Mn1–ZnS exhibited characteristic signals at 1618, 1655, and 1695 cm−1 (*COOH species)26,32,33, 2078 cm−1 (*CO)34, 1122 and 1415 cm−1 (*CHO)33,35, and a key band at 1315 cm−1 corresponding to *COCHO (Fig. 4a)32, a critical intermediate in C–C coupling toward C2H436,37.

Fig. 4: In situ spectra analysis.
Fig. 4: In situ spectra analysis.
Full size image

a, b In situ DRIFTS for CO2 photoreduction reaction over Mn1–ZnS (a) and Mn1–ZnSv (b). c, d CO-adsorbed DRIFTS for Mn1–ZnS (c) and Mn1–ZnSv (d). e, f Reaction pathway schemes for ZnS and Mn1–ZnSv interface. Source data are provided as a Source Data file.

In contrast, Mn1–ZnSv exhibited a stronger *CO signal at 2078 cm−1 (Fig. 4b and Supplementary Fig. 30), indicating enhanced stabilization and coverage of *CO intermediate. The progressive intensification of the 1315 cm−1 band over time confirmed efficient *CO–*CHO coupling and *COCHO formation, which was further verified by the 13CO labeling experiment (Supplementary Fig. 31). Additional bands at 1442, 1462, and 1364 cm⁻1 were attributed to *C2H4 and *CH2 intermediates32,38, confirming the high activity and selectivity of Mn1–ZnSv toward C2H4 production.

To further probe CO adsorption behavior, CO-adsorbed DRIFTS was performed on both catalysts (Fig. 4c, d). Prominent peaks at ~2084 and 2036 cm−1 were assigned to linearly bonded CO (COL) on Mn single atoms. Mn1–ZnSv displayed significantly higher peak intensities than Mn1–ZnS, indicating a stronger CO binding capability, consistent with DFT predictions. Based on the in situ DRIFTS data and theoretical insights, a photocatalytic reaction pathway for CO2 conversion on Mn1–ZnSv was proposed (Fig. 4e, f), highlighting the critical role of low-coordination Mn single-atom sites in stabilizing reaction intermediates, enabling *CO–*CHO coupling, and ultimately promoting selective C2H4 production.

Evaluating catalyst performance for CO2 photoreduction

The CO2 photoreduction performance of Mn1–ZnSv catalysts was assessed under visible light irradiation (λ ≥ 380 nm) without the use of photosensitizers or sacrificial agents (Fig. 5, Supplementary Figs. 32–34, and Supplementary Table 5). As shown in Fig. 5a, pristine ZnS primarily produced CO at a rate of 64.6 μmol g−1 h−1, with a low C2H4 selectivity of only 5.6%. Compared to pristine ZnS, ZnS with sulfur vacancies (ZnSv) exhibits negligible enhancement in C2H4 yield and selectivity (7.3%) (Supplementary Figs. 35 and 36). Upon incorporation of saturated-coordination Mn single atoms (Mn1–ZnS), both activity and selectivity improved significantly, achieving a C2H4 formation rate of 47.5 μmol g−1 h−1 and a selectivity of 74.5%. Concurrently, CO evolution decreased to 28.4 μmol g−1 h−1 and the competing hydrogen evolution reaction was markedly suppressed (Supplementary Fig. 37), likely due to the *CO-stabilizing effect of Mn sites that promote its protonation and subsequent C–C coupling.

Fig. 5: Photocatalytic CO2 reduction performance.
Fig. 5: Photocatalytic CO2 reduction performance.
Full size image

a Product yields of products (HCOOH, CO, C2H4, and CH4) and the selectivity of C2H4 product over ZnS, Mn1–ZnS, and Mn1–ZnSv catalysts (T = 298 K; irradiation time = 4 h; solvent: H2O, 5 mL: amount of catalyst: 0.2 g). b The evolution of C2H4 product during photocatalytic CO2 reduction over time (n = 3 independent experiments, data are presented as mean values ± SD). c The products yield over ZnS, Mn1–ZnS and Mn1–ZnSv catalysts using CO2 and CO as reactant, respectively. d The AQE of C2H4 evolution catalyzed by Mn1–ZnSv catalyst and the correlation between the related absorption spectra and absorption wavelengths (n = 3 independent experiments, data are presented as mean values ± SD). e The control experiments of photocatalytic CO2 reduction performance over Mn1–ZnSv under altered conditions. f Cycling tests for CO2 photoreduction to C2H4 over Mn1–ZnSv photocatalyst under the same as reaction conditions. Source data are provided as a Source Data file.

Notably, the low-coordination Mn1–ZnSv catalyst delivered a significantly enhanced C2H4 formation rate of 76.6 μmol g−1 h−1, nearly 58.9 times higher than that of pristine ZnS, with a selectivity of 99.1% for C2H4 production (Fig. 5b and Supplementary Table 6). Meanwhile, CO formation further decreased to 4.2 μmol g−1 h−1 (Supplementary Fig. 37) and no liquid products were detected, underscoring the pivotal role of Mn coordination environment in modulating *CO adsorption and steering the reaction pathway toward selective C2+ hydrocarbon formation.

To further validate the role of low-coordination Mn SAs in promoting C–C coupling, a CO substitution experiment was carried out. This resulted in a C2H4 yield of 198.8 μmol g−1 (Fig. 5c), confirming that *CO is a crucial intermediate for C–C coupling and effectively activating and stabilizing by the low-coordination Mn sites. The apparent quantum efficiency (AQE) for CO2 photoreduction over Mn1–ZnSv reached 8.1% at 420 nm (Fig. 5d). Control experiments (Fig. 5e) demonstrated that C2H4 formation occurred only under light irradiation in the presence of both CO2 and H2O, excluding contributions from dark reactions or carbon-containing contaminants39. Isotope-labeling experiments using 13CO2 and D2O (Supplementary Fig. 38) further confirmed CO2 and H2O as the carbon and proton sources, respectively, for C2H4 formation. Considering that the activation of H2O into active hydrogen species may be an important factor that affecting the turnover-limiting step during the photocatalytic CO2RR to C2H4, the KIE of H/D value over the Mn1–ZnSv catalyst is close to 1 (Supplementary Fig. 39), further confirming the *CHO formation as the rate-determining step suggested by DFT results.

The photocatalytic stability of Mn1–ZnSv was evaluated over 50 consecutive cycles (200 h), during which no significant decline in activity or selectivity was observed (Fig. 5f). Post-reaction characterizations, including TEM, XRD, and XPS (Supplementary Fig. 40), confirmed the preservation of the structural and electronic integrity of the catalyst. No characterized signals of Mn–Mn bond over Mn1–ZnSv after long-term reaction further proved the great stability of the coordination environment of Mn atoms (Supplementary Fig. 41), ensuring effective photocatalytic performance. Furthermore, the well-balanced charge consumption between the oxidation and reduction half-reactions (RO/R ≈ 1) minimizes charge accumulation and effectively suppresses photocorrosion (Supplementary Table 5). Notably, under sacrificial-agent-free conditions, a comparative analysis of photocatalytic performance between our study and previous reported catalysts17,18,19,32,40,41,42,43,44,45,46,47,48,49,50 is displayed in Supplementary Table 6. Both the competitive C2H4 yield and selectivity of Mn1–ZnSv underscores its potential as a highly efficient and robust CO2-to-C2H4 conversion.

To evaluate the general applicability of low-coordination transition metal single atoms in ZnS matrices, analogous catalysts incorporating Fe, Co, Ni, and Cu were synthesized and tested (Supplementary Fig. 42). All M1–ZnSv (M = Fe, Co, Ni, Cu) variants exhibited substantially higher C2H4 yields than both pristine ZnS and their saturated-coordination M1–ZnS counterparts, indicating a certain enhancement in photocatalytic performance induced by low-coordination metal centers (Supplementary Fig. 43).

To elucidate the electronic origin of this enhancement, projected density of states analyses were conducted for the 5σ and 2π orbitals of adsorbed CO and the metal 3d orbitals (3dz², 3dxz, 3dyz) of the M1–ZnSv systems (Supplementary Fig. 44). Among all the systems, Mn1–ZnSv exhibits the highest d-band center at −1.85 eV, closer to the Fermi level. This upshift results in decreased population of antibonding d–5σ and bonding d–2π* states, thereby enhancing *CO adsorption.

Furthermore, integrated crystal orbital Hamilton population values for M–*CO bonds exhibited a clear linear decrease from Cu1–ZnSv to Mn1–ZnSv (Supplementary Fig. 45), further validating the stronger *CO binding affinity of Mn sites. In situ DRIFTS spectra (Supplementary Fig. 46) and free energy of reaction pathway (Supplementary Fig. 47) presented that Mn1–ZnSv exhibits significantly stronger *COL absorption peaks and comparatively favorable energetics for *CO protonation and coupling than that of Cu1–ZnSv, which is consistent with the results from the catalytic performance. These findings conclusively demonstrate that low-coordination transition metal single atoms, particularly Mn, enhance *CO stabilization and facilitate the critical C–C coupling step, thereby driving selective C2 product formation in CO2 photoreduction.

Discussion

In summary, we developed a Mn1–ZnSv photocatalyst with a low-coordination Mn SA active site for efficiently selective CO2 photoreduction to C2H4. The detailed studies revealed that low-coordination Mn single-atom active sites break the geometric symmetry of traditional SACs, resulting in anisotropic charge redistribution upon *CO adsorption. This not only enhances the adsorption of *CO intermediates, but also provides new sites for further C–C coupling. Both DFT calculations and experimental evidences confirmed the obtained Mn1–ZnSv enhanced *CO adsorption and promoted the coupling between *CO and *CHO intermediates, ultimately driving the formation of C2H4. As a result, the optimal Mn1–ZnSv photocatalyst achieved a C2H4 selectivity of 99.1% with a formation rate of 76.6 μmol g−1 h−1, and demonstrated long-term stability over 200 h. We anticipate that this strategy for tailoring low-coordination single-atom active sites will provide new insights into the rational design of highly efficient and selective photocatalysts for converting carbon dioxide into multi-carbon products.

Methods

Chemical reagents

Zinc acetate (Zn(OAc)2), Manganese acetate (Mn(OAc)2), Copper acetate (Cu(OAc)2), Iron acetate (Fe(OAc)3), Nickel acetate (Ni(OAc)2), Cobalt acetate (Co(OAc)2), thiourea, ethylene glycol were purchased from Aladdin Reagent Co., Ltd., Shanghai. All chemicals are of analytical grade and used without further purification. The deionized water was supplied by a Millipore system (Outlet water resistivity > 18 MΩ cm).

Sample preparation

To synthesize ZnS catalyst, typically 6 mmol of Zn(OAc)2 and 12 mmol of thiourea were added to 100 mL of ethylene glycol. After continuous agitation for 30 min, the mixture was moved to a 250 ml three-necked flask with round bottom, and placed in a microwave reactor with a reflux unit. It was heated from room temperature to 150 °C with a heating rate of 25 °C min−1 and maintained for 10 min. After the reaction cooled to room temperature naturally, the resulting powder was washed with deionized water and ethanol for three times and dried under vacuum at 80 °C for 6 h. The obtained samples were named as ZnS.

To synthesize Mn1–ZnS catalyst, 6 mmol of Zn(OAc)2 and 12 mmol of thiourea were added to 100 mL of ethylene glycol. After continuous agitation for 30 min, the different mass ratios of Mn(OAc)2 (0.2%, 0.6%, 1%) was added to the suspension and continued stirring for 10 min. The mixture was moved to a 250 ml three-necked flask with round bottom, and placed in a microwave reactor with a reflux unit. It was heated from room temperature to 150 °C with a heating rate of 25 °C min−1 and maintained in Ar flow for 10 min. After the reaction cooled to room temperature naturally, the resulting powder was washed with deionized water and ethanol for three times and dried under vacuum at 80 °C for 6 h. The obtained samples were named as Mn1–ZnS-0.2, Mn1–ZnS-0.6 and Mn1–ZnS-1.0, respectively.

To synthesize Mn1–ZnSv, 100 mg Mn1–ZnS was immersed in H2O2 solution (0.05 mol L−1) under the protection of Ar in microwave reaction treated for 5, 10, and 15 s, respectively, all the obtained samples were carefully washed and dried using the method described above. The resulting samples were labeled as Mn1–ZnSv with different S-vacancy concentrations.

The preparation process for ZnSv is the same as that of Mn1–ZnSv, except that the precursor Mn salt was not added. The synthesis steps for Fe1–ZnSv, Co1–ZnSv, Ni1–ZnSv, and Cu1–ZnSv are the same as Mn1–ZnSv, except that the precursor salt was replaced with the corresponding acetate. Unless otherwise specified, the precursor transition metal salt loading for the ZnS sample was 0.6%.

Characterization

The phase structure of the photocatalysts was performed on a powder X-ray diffractometer (XRD, Shimadzu XRD 6000). The morphology was observed via field emission scanning electron microscopy (FEI Quanta 200 F), transmission electron microscopy (TEM, JEOL JEM 2100). STEM-ADF images and EDS mapping were obtained by Hitachi HF5000, working at an accelerating voltage of 200 kV. To interpret the light absorption characteristics of the catalysts, UV–Vis DRS were detected on an UV–vis spectrophotometer (Shimadzu UV-2600) over a range of 200–800 nm. To study on the chemical bonds or functional information, in situ DRIFTS were examined using the IR Affinity-1 FTIR spectrometer. The Zn, Cu, and Mn K-edge X-ray adsorption spectra were acquired from 4B9A beamline in Beijing Synchrotron Radiation Facility (BSRF). Isotopic labelling experiments were performed on an Agilent 5977B GC/MS system, using 13CO2 and D2O as reactants under the photocatalytic CO2 reduction conditions. The GC-MS spectra reflect the relationship between the m/z of all ion fragments and their relative abundance. Photoluminescence spectra were analyzed on FluoroMax+ spectrophotometer (Hitachi, Japan) with excitation wavelength at 310 nm, and the receiving fluorescence range was from 400 to 700 nm. X-ray photoelectron spectroscopy (XPS) and valence band-XPS tests were conducted on the PerkinElmer PHI-1600 ESCA spectrometer with a Mg Kα X-ray source. All the calibration for the binding energies were based on the C 1 s at 284.8 eV51. Carbon dioxide-temperature programmed desorption (CO2-TPD) was recorded on a BSD-C200 chemisorption system (BSD Instrument).

In-situ DRIFT spectra measurement

In-situ DRIFT spectra were obtained from a Thermo Scientific Nicolet iS50 FTIR spectrometer. The compressed catalysts were placed in an diffuse reflectance cell (Harrick) equipped with a mixed atmosphere of CO2 and H2O vapor for in situ experiments. In the pretreatment stage, the system was progressively heated from room temperature to 200 °C under continuous N2 purging to remove surface impurities. After cooling to room temperature, the data were collected at a flow rate of 50 mL min−1 in a mixed atmosphere. The infrared spectra obtained in the dark were used as background data. Subsequently, time-dependent IR spectra were collected under illumination for 30 min.

In situ CO-adsorbed DRIFT spectroscopy measurement

In situ CO-adsorbed DRIFT spectroscopy measurement was carried out on a Nicolet 6700 instrument equipped with a mercury cadmium telluride detector. 10 mg catalyst was placed in an infrared diffuse reflection high-temperature reaction cell with a ZnSe window (Pike Technologies), pretreated in 10% H2/N2 flow at 300 °C for 30 min, then cooled to room temperature. This was followed by the introduction of CO flow for 30 min and subsequent evacuation. The CO-adsorbed FT-IR spectra were recorded at an apart of 2 min.

Photoelectrochemical measurements

Photoelectrochemical (PEC) measurements were carried out in the three-electrode system by utilizing an electrochemical workstation (CHI660E). The prepared coated catalysts serve as the working electrode, with an effective surface area of 1 cm2. Pt net and Ag/AgCl electrode were employed as the counter electrode and reference electrode, respectively. The Na2SO4 solution (0.1 mol L−1) was acted as electrolyte (pH = 7.02 ± 0.05). In order to prepare the working electrode, 25 mg catalyst and 10 μL Nafion were dispersed in 384 μL deionized water and 96 μL ethanol, then followed by ultrasonication for 25 min. The evenly mixed ink (5 μL) was dropped onto the surface of the glass-carbon electrode, and dried naturally at room temperature. A 50-s light‑on/light‑off cycle was applied to record the transient photocurrent response. The Mott–Schottky measurements were executed with the different frequencies (100, 500, and 1000 Hz), and electrochemical impedance spectra was performed, traversing a frequency range from 0.1 to 100000 Hz. Resistance (R) was measured to be 0.60 ± 0.02 Ω at the open circuit potential.

Tests on photocatalytic CO2 reduction with H2O

The photocatalytic CO2 reduction tests were performed in a sealed reaction system by employing the Labsolar 6 A reactor (Beijing Perfectlight Technology Co., Ltd.). 0.2 g of photocatalysts was dispersed in 5 mL of deionized water within a glass dish (diameter is ~4.0 cm). The mixed sample was then placed inside a reactor (volume of ~370 mL) of the photoreaction system, at a distance of 15.0 cm from the light source. The reaction system was subjected to three cycles of vacuuming and refilling with high-purity CO2, achieving a final CO2 pressure of 80.0 kPa in the reactor. And a cooling circulating water device was utilized to maintain the reaction system a constant temperature at 25 ± 1.5 °C. To simulate the natural photosynthesis, a 300 W xenon lamp (~100 mW cm−2) equipped with a 380 nm external filter served as light source. The light intensity of catalyst surface was measured by a ThorLabs PM100D Power with a photodiode sensor, with the average value determined from multiple regions of the watch glass52. After starting the light source and the gas chromatograph automatic sampling device, the sample was analyzed every 30 min, and the reaction test was carried out for 4 h. The amount of C2H4, CH4, CO, H2, and O2 products evolved was analyzed online by a GC-9560 gas chromatograph (HuaAiSePu Company) using the external standard calibration method. For the liquid product, the amount of HCOOH was determined by a proton nuclear magnetic resonance (1H NMR) analysis, taking DMSO as an internal standard. No other products were detected above the detection limit of instrument.

The product selectivity for CO2 photoreduction to C2H4 was calculated according to the following equations:

$${{{\rm{S}}}}_{{{{{\rm{C}}}}_{2}{{\rm{H}}}}_{4}}\left(\%\right)=\frac{12n\left({C}_{2}{H}_{4}\right)}{12n\left({C}_{2}{H}_{4}\right)+2n\left({CO}\right)+8n\left({{CH}}_{4}\right)+\,2n\left({CHOOH}\right)}\times 100\%$$
(1)

where n(product) is the molar amount of product generated.

To quantify the efficiency of photocatalytic light-energy conversion, the apparent quantum efficiency (AQE) was measured. The incident photon numbers were determined using monochromatic light sources at specified wavelengths (λ = 400, 420, 450, 500, 520, and 600 nm). AQE was defined as the ratio of consumed photons to incident photons, and according to the following equations:

$${{{\rm{AQE}}}}_{{{{\rm{C}}}}_{2}{{{\rm{H}}}}_{4}}= \frac{12\times {{\rm{Number}}}\; {{\rm{of}}}\; {{\rm{reacted}}}\; {{\rm{electrons}}}}{{{\rm{Number}}}\; {{\rm{of}}}\; {{\rm{incident}}}\; {{\rm{photons}}}}\times 100\,\%\\= \,\frac{{12\times n}_{{{C}_{2}H}_{4}}\times {N}_{A}}{S\times I\times t\times \frac{\lambda }{h\times c}}\times 100\,\%\,$$
(2)

where S is the irradiated area (cm2), I is the irradiation light intensity (W cm−2), t is the irradiation time (s), λ is the equivalent wavelength (m), h is Planck’s constant (6.626 × 10−34 J s−1), c is the speed of light (2.998 × 108 m s−1) and NA is Avogadro’s number (6.022 × 1023 J mol−1). nC2H4 is the amount of C2H4 produced in 4 h.

The incident photon-to-current efficiency is defined as the ratio of the incident monochromatic photons converted to collected electrons and can be calculated by using Eq. (3):

$${{\rm{IPCE}}}=\frac{\frac{{hc}}{\lambda }\times j}{{Ie}}$$
(3)

where j is the photocurrent density (A cm−2), λ is the equivalent wavelength (m), and I is the irradiation light intensity (W cm−2).

Density functional theory (DFT) calculations

DFT calculations were carried out in the framework of the Vienna ab initio Simulation Package (VASP). The exchange and correlation energies were established by the Perdew–Burke–Ernzerhof functional with spin-polarized generalized gradient approximation53. The core electrons were described by the projected augmented wave pseudopotentials. For geometry optimization, a cutoff energy of 450 eV was used to expand the wave functions, and the Brillouin zone was sampled with 2 × 2 × 1 k-points. All the structures and energy were allowed to relax below 0.05 eV Å−1. The DFT-D3 method of Grimme et al. was employed to involve the van der Waals interaction54. The calculations of charge density difference were employed to analyze the movement and distribution of the charge. For the elementary reaction barriers, the transition states were determined by the climbing image nudged elastic band method and were confirmed by further frequency calculations showing one and only one imaginary frequency.

According to the XRD results, a 5 × 5 slab model composed of three layers along the (111) direction was constructed with the space group F-43m model of ZnS carrier. A 20.0 Å vacuum region between the slabs was constructed to avoid the interlayer interaction. The doped ZnS with other transition metal elements was simulated by replacing one of the Zn atoms. The S-vacancy was simulated by removing S atoms off, and the theoretical value of Sv is 1.33%. Considering the difference in the performance of as-prepared catalysts, ICP quantified vacancy concentrations, specific configuration of sulfur vacancies was adopted to model the surface. The atomic coordinates of the DFT models are provided as a Supplementary Data 1 file. The formation energy of sulfur vacancy \(\left(\Delta {{{{\rm{E}}}}_{{{\rm{V}}}}}_{{{\rm{S}}}}\right)\) was calculated as follows:

$$\Delta {{{{\rm{E}}}}_{{{\rm{V}}}}}_{{{\rm{S}}}}={{{\rm{E}}}}_{{{\rm{slab}}}-{{{\rm{V}}}}_{{{\rm{S}}}}}-{{{\rm{E}}}}_{{{\rm{slab}}}}-{{\rm{E}}}_{{\rm{S}}}$$
(4)

in which \({{\mbox{E}}}_{{\mbox{slab}}-{{\mbox{V}}}_{{\mbox{S}}}}\) and \({{\mbox{E}}}_{{\mbox{slab}}}\) represent the energy of the slab after and before Vs. The adsorption energy Eads was calculated by a standard formula:

$${{{\rm{E}}}_{{\rm{ads}}}}={{{\rm{E}}}_{{{\rm{catalyst}}}{\mbox{-}}}} * {-} {{{\rm{E}}}_{{\rm{catalyst}}}}{-}{{\rm{E}}} * $$
(5)

in which Ecatalyst - * and Ecatalyst is the total energy of the slab with and without intermediates.

The Gibbs free energy change (ΔG) for the reaction was calculated by

$$\Delta {{\rm{G}}}=\Delta {{\rm{E}}}+\Delta {{{\rm{E}}}}_{{{\rm{ZPE}}}}-{{\rm{T}}}\triangle {{\rm{S}}}$$
(6)

in which \(\triangle E\) is reaction energy obtained from DFT calculations, \(\triangle {{\mbox{E}}}_{{\mbox{ZPE}}}\) represent the change of the zero-point energy, T∆S is the entropic contribution (T was set to be 300 K), which were obtained from the vibrational frequency calculations through the VASPKIT code. The free energy of the proton–electron pair is equal to half the free energy of the hydrogen molecule according to the computational hydrogen electrode method55.

Ab initio molecular dynamics simulations are carried out by CP2K package and the QUICKSTEP module with fully explicit solvent water molecules56. Godecker-Teter-Hutter (GTH) pseudopotentials were adopted to model the core electrons, and double-ζ valence single polarization (DZVP)-molecularly optimized (MOLOPT)-short-ranged-GTH basis set was employed to obtain optimized structure57,58. Valence electrons were expanded in an orthonormal plane-wave basis using a cutoff energy of 400 Ry, along with Grimme D3 dispersion correction59. The self-consistent field convergence criterion is set to 10−5 Ry for the total energies, and the simulation temperature (300 K) was controlled through a Nosé-Hoover thermostats with time step of 1.0 fs. DFT offers valuable insights for screening coordination structures and exploring reaction pathways, yet it remains inherent limitations in simulating practical catalytic conditions (e.g., actual illumination, solvation effects). Thus, the computationally identified coordination structures should serve as qualitative guides for catalyst design and mechanistic investigation.