Two precursors [Ti6O6(OiPr)6(AB)6] (AB = 4-aminobenzoate; HOiPr = isopropoxide, denoted Ti6)44 and [Cu3(PyCA)3] (1H-PyCA = pyrazolate-4-carboxaldehyde, denoted Cu3)32 were prepared according to the previously reported methods. The synthesis of MCOF-Ti6Cu3 was carried out from the condensation of Ti6 and Cu3 in a mixture of mesitylene and N, N-Dimethylformamide (DMF) under the vacuum solvothermal conditions (Fig. 1a). Fourier-transform infrared (FT-IR) spectra confirmed the condensation reaction between the two starting materials by the obvious disappearance of the C = O (1685 cm−1) and N-H (3200–3500 cm−1) stretching vibration peaks found in Cu3 and Ti6, respectively (Fig. 1b). In addition, a new characteristic peak at 1625 cm−1 attributed to the C = N bond was observed, indicating the successful formation of an imine-linked network45. Powder X-ray diffraction (PXRD) pattern of MCOF-Ti6Cu3 displayed the strongest diffraction peak at 2θ 6.1° corresponding to (110) reflection plane, along with minor peaks at 12.4°, 18.8°, 20.4°, and 25.3° that are attributed to the (220), (330), (502), and (440) crystal planes, respectively (Fig. 1c). To elucidate the crystal structure of MCOF-Ti6Cu3, the staggered stacking model in the R3 and P63 space groups corresponding to ABC and AB stacking were built using Materials Studio28,31,46. The experimental PXRD pattern matches well with the simulated one for the ABC stacking model. Pawley refinement was applied to provide the refined unit cell parameters of a = b = 29.14 Å, c = 18.97 Å, with good residual factors of Rp = 2.36% and Rwp = 3.10%, suggesting the validity of the computational model. The structural information for the AB stacking model is in the Supplementary Information.
These structural analyses reveal that a rhombic channel along c axis with a theoretical pore size of 1.08 nm × 1.72 nm and interlayer distance of 5.6 Å (Fig. 1d, e). The porosity and surface area of MCOF-Ti6Cu3 were examined by N2 adsorption-desorption measurements at 77 K (Fig. 1f). The Brunauer-Emmett-Teller (BET) surface area was calculated to be 215.2 m2 g−1. Besides, the pore size distribution (∼1.01 and ∼1.68 nm) of MCOF-Ti6Cu3 were in good agreement with the results predicted from the theoretical structure model. The crystallinity and structure of MCOF-Ti6Cu3 remained by treating it with various solvents at room temperature, which was confirmed by PXRD analyses (Supplementary Fig. 9). The thermal stability of MCOF-Ti6Cu3 was confirmed by thermogravimetric analysis (TGA) (Supplementary Figs. 10, 11, and 12), which showed no noticeable weight loss up to 300 °C. The morphology of MCOF-Ti6Cu3 was characterized by scanning electron microscopy (SEM) and transmission electron microscopy (TEM). The SEM and TEM images of MCOF-Ti6Cu3 shown in Supplementary Figs. 13 and 14 display a sheet-like morphology, which is apparently different from the regular octahedral structure of Ti6 (Supplementary Fig. 15) and the needle-shaped particles of Cu3 (Supplementary Fig. 16). In addition, the high-resolution TEM (HRTEM) image distinctly reveals the high crystallinity of MCOF-Ti6Cu3 (Supplementary Fig. 17). The corresponding energy-dispersive X-ray spectroscopy (EDS) mapping spectra (Supplementary Fig. 18) indicate that Ti and Cu elements are uniformly distributed in MCOF-Ti6Cu3.
In order to conduct a more in-depth study on the hetero-metallic cluster catalyst of MCOF-Ti6Cu3, we also analyzed two monometallic cluster framework materials named MOF90128 and FDM-71-ABC32 as control samples, which were synthesized from Ti6 with p-benzaldehyde and Cu3 with p-phenylenediamine, respectively (Fig. 2a, see Supplementary Information for details). For photocatalytic reduction of CO2, the separation and migration behavior of photogenerated electron-hole pairs is one of the critical factors. The photoelectro-chemical properties of the samples were performed with the combination of transient photocurrent measurements, electrochemical impedance spectroscopy (EIS) and photoluminescence (PL) spectra. Figure 2b shows that all samples exhibit obvious photocurrent signals and have an excellent reproducibility of the response intensity in the process of on-off cycles. The transient photocurrent intensity of MCOF-Ti6Cu3 is ∼1.5 times that of FDM-71-ABC and much higher than that of MOF901. Moreover, the photocurrent intensity of Cu3 is about three times as high as Ti6, indicating that the Cu3 cluster may have better photosensitivity (Supplementary Fig. 19). The EIS spectra reveal that compared with MOF901 and FDM-71-ABC, MCOF-Ti6Cu3 exhibits a smaller semicircle radius in the high-frequency region in Nyquist plots, suggesting a higher separation and transfer efficiency of charge carriers in MCOF-Ti6Cu3 (Supplementary Fig. 20). As presented in Fig. 2c, the spectral peak position of FDM-71-ABC which originates from Cu3 cluster (Supplementary Fig. 21) is about 400 nm32, while that of MOF901 which originates from Ti6 cluster (Supplementary Fig. 22) is about 500 nm28. In contrast, MCOF-Ti6Cu3 exhibits two shifted peaks with the weakest PL intensity, indicating the existence of electron transfer and the lowest recombination efficiency of photoinduced electron-hole pairs. Meanwhile, time-resolved fluorescence decay spectra were measured to determine the specific charge carrier dynamics (Fig. 2d). The results showed that the average lifetime of the photogenerated charge carriers in MCOF-Ti6Cu3 was 1.90 ± 0.26 μs, which is comparable to that of FDM-71-ABC (1.67 ± 0.09 μs) and longer than that of MOF901 (1.28 ± 0.15 μs). The difference of fluorescence lifetime between MCOF-Ti6Cu3 and FDM-71-ABC can also be directly evident from the fluorescence decay data (Fig. 2d), which shows that the fluorescence decays with a lower rate in MCOF-Ti6Cu3 compared with those in the other two samples, illustrating that the photogenerated charge carriers survived longer on the surface of MCOF-Ti6Cu3. These results further demonstrate that the spatially separated hetero-metallic cluster structures can effectively capture photogenerated electrons and holes respectively, reduce the change of recombination of charge carriers and facilitate their separation and migration in the framework in favor of the higher photocatalytic activity.
The solid-state UV-vis diffuse reflectance spectra (UV-vis DRS) of the samples were recorded. As shown in Fig. 2e, FDM-71-ABC and MCOF-Ti6Cu3 exhibited considerable absorption in the UV and visible regions, whereas the absorption edge of MOF901 was only located near 650 nm. Evaluated by the Tauc plots, the corresponding band gap energies (Eg) of MOF901, FDM-71-ABC, and MOF-Ti6Cu3 were calculated to be 2.5, 2.71, and 2.25 eV, respectively (Supplementary Figs. 23, 24 and 25). Furthermore, Mott-Schottky measurements were performed to elucidate the semiconductor character of the samples (Supplementary Figs. 26, 27 and 28). The flat band potentials of MOF901, FDM-71-ABC, and MOF-Ti6Cu3 were determined to be −1.06, −1.19, and −0.96 V vs. Ag/AgCl (i.e., −0.86, −0.99, and −0.76 V vs. NHE), respectively, which were equal to their conduction band (CB) potentials47. Based on the equation Eg = EVB − ECB, their valence band (VB) positions were accordingly calculated to be 1.39, 1.51, and 1.95 V vs. NHE, respectively (Fig. 3a). Considering the CB/VB and LUMO/HOMO, the energy of CB/VB is equivalent to the electrochemical energy potentials of LUMO/HOMO. Obviously, they are thermodynamically suitable for photocatalytic reduction of CO2 together with oxidation of H2O to various fuels.
The simultaneous completion of CO2 reduction and H2O oxidation is a great challenge. Therefore, the performance of photocatalytic reduction of CO2 with H2O over the samples was systematically investigated under simulated sunlight irradiation. As shown in Fig. 3b, the main product of the CO2 photocatalytic reduction with MCOF-Ti6Cu3 was HCOOH, of which the yield reached 9.28 μmol (77.3 μmol g−1 h−1) after 12 h illumination, while no gaseous products were detected, resulting in a high selectivity. In sharp contrast, MOF901 or FDM-71-ABC showed inactive for CO2 reduction, indicating that neither Ti nor Cu cluster could complete the CO2 reduction coupled with H2O oxidation. In addition, control experiments with Ti6 and Cu3 as the photocatalysts for CO2 reduction have also been evaluated under identical conditions. It cloud be found that no HCOOH was observed, indicating that neither the precursor nor homo-metallic cluster catalyst can complete the overall reaction alone. The PXRD pattern and FT-IR suggested structural integrity and no noticeable loss of crystallinity for MCOF-Ti6Cu3 after reactions, demonstrating its excellent stability (Supplementary Figs. 29 and 30).
The structural units of MCOF-Ti6Cu3 are clusters, in which interlayer spacing is larger than that of the traditional two-dimensional COFs. Therefore, stripping MCOF-Ti6Cu3 into single layers is feasible employing high-frequency ultrasound48. As shown in Supplementary Fig. 31, after dispersing 10 mg of MCOF-Ti6Cu3 into 30 mL of deionized water, the nanosheets with the thickness of 4.83 nm were obtained by high-frequency ultrasound for 30 min. The resulting nanosheet structure exhibited a dramatically enhanced yield of HCOOH to 20.39 μmol in 12 h (169.8 μmol g−1 h−1) (Fig. 3c), which was superior to most of the previously reported noble-metal-free photocatalysts in overall reactions (Supplementary Table 5). Exfoliation has little effects on transient photocurrent responses and electrochemical impedance, but the gas adsorption and specific surface area were greatly improved (Supplementary Figs. 32, 33 and 34). The exfoliation exposes more adsorption sites and active centers, which may be the reason for the excellent photocatalytic performance. The physically mixed control experiments with Ti6 + Cu3 or MOF901 + FDM-71-ABC composites were performed. No products were detected, which demonstrates the necessity of the assembly of MCOF-Ti6Cu3 in the framework. In addition, in order to evaluate the catalytic stability and reusability of MCOF-Ti6Cu3, recycling experiments and long-term tests were carried out. As shown in Fig. 3d, MCOF-Ti6Cu3 retained the original photocatalytic efficiency for the HCOOH production during four consecutive cycles. MCOF-Ti6Cu3 nanosheets possessed excellent durability during the photocatalytic process, which was further confirmed by the time-dependent conversion yields of CO2 into HCOOH, which gradually increased with time (Fig. 3e, Supplementary Fig. 35). The yield of products no longer increased after the removal of catalyst from the reaction, further ruling out the possibility of leaching of Ti and Cu (Supplementary Fig. 36).
Additional control experiments demonstrated that negligible reduction products could be detected without the photocatalyst, CO2, or light irradiation. These results suggested that those factors were all indispensable for photocatalytic CO2 reduction, and the generated HCOOH originated from CO2 rather than the photolysis from other carbon-containing species in this photocatalytic system. In order to further verify the origin of HCOOH, the 13CO2 isotope trace experiment was performed. The 13C NMR spectrum gave the corresponding isotope H13COOH signal with a peak at 168.4 ppm, and there was no such signal under the same condition as for 12CO2 (Fig. 3f), unambiguously verifying that the carbon source of HCOOH indeed originated from the CO2 reduction. Alternatively, when H2O was replaced with H218O, 18O2 (m / z = 36), 18O16O (m / z = 34) and 16O2 (m / z = 32) were detected after the reaction, confirming that the generated 18O2 or 16O2 was from H218O or H216O (Supplementary Fig. 37)45,49.
To investigate the mechanism of the photocatalytic overall reaction XPS and in-situ XPS measurements combined with DFT calculations (method in Supplementary Information) were carried out. The calculation model was constructed based on the finite cluster structure of MCOF-Ti6Cu3 (Supplementary Fig. 38). XPS showed that the observed Ti 2p binding energy at 459.0 eV in Ti6 was consistent with Ti4+, while the Cu 2p binding energy at 932.8 eV and 934.6 eV in Cu3 were assigned to Cu1+ and Cu2+, respectively (Fig. 4a, b)18,30. Previous studies found that the production of Cu2+ was due to the surface oxidation, and the Cu valence could be switched between Cu1+ and Cu2+ without altering the backbone of Cu3(pyrazolate)3 triangular SBUs32,50. For MCOF-Ti6Cu3, the negative shift of the Ti 2p binding energy by 0.4 eV and the positive shift of the Cu 2p binding energy by 0.7 eV indicated a successful covalent-bond connection between the two precursors with the formation of an internal electric field, and the possible migration direction of electrons was from Cu to Ti. This interaction from cluster-to-cluster was conducive to the transfer of electrons. The XPS spectra of C, N and O elements are shown in Supplementary Figs. 39, 40 and 41.
The electron transfer process in the photocatalytic reaction could be distinctly revealed by the valence change using in-situ XPS measurements. As shown in Fig. 4c, d, under irradiation (black and red line), the Ti 2p binding energy increased from 458.46 eV to 458.61 eV. In comparison, the Cu 2p binding energy decreased from 932.94 eV to 932.78 eV, indicating that some electrons returned to Cu cluster with photoexcitation. After CO2 was introduced into the system (red and blue line), the binding energy of Cu 2p increased to 932.94 eV while that of Ti 2p was unchanged, indicating that the transfer direction of the photogenerated electrons was from the Cu cluster to CO2. Consequently, it could be inferred that the active center of the photocatalytic reduction of CO2 to HCOOH over MCOF-Ti6Cu3 was the Cu cluster. Finally, as shown in the Supplementary Figs. 42, 43 and 44, the in-situ XPS and in-situ DRIFTS spectra before and after the reaction were consistent, which proves the good stability of the catalyst. As shown in electrostatic potential (ESP) analyses (Supplementary Fig. 45), the positive charge was concentrated on the Cu-N cluster and benzene moieties, while the negative charge was concentrated on the Ti-O cluster.
The DFT and time-dependent DFT (TDDFT) calculation results could also prove that the photocatalytic CO2 reduction reaction on the Cu cluster in MCOF-Ti6Cu351,52. The TDDFT calculation results (Supplementary Fig. 46) show that the HOMO and LUMO of MCOF-Ti6Cu3 are close to each other and can be excitated by the excitation energy of infrared light (1704.16 nm). However, visible and ultraviolet lights correspond to multiple excitation modes. The visible lights with wavelengths of 487.42 nm and 453.25 nm corresponded to the excitation modes of HOMO → LUMO + 27, + 29, + 31, and HOMO → LUMO + 31, + 32, while the wavelengths of 352.79 nm and 351.89 nm in ultraviolet region correspond to the excitation modes of HOMO − 3 → LUMO + 1 and HOMO 1 → LUMO + 1, respectively. The orbital composition analysis displayes that the occupied orbitals of HOMO, − 1, and − 3 are located on the Cu cluster fragments, and the unoccupied orbitals of LUMO, + 27, + 29, + 31, and + 32 are also located on the Cu cluster fragments, other unoccupied orbitals are located on the Ti cluster fragments. It demonstrates that the electron cloud does not transfer in the process of infrared excitation, and indeed transfers from the Cu cluster to the Ti cluster under ultraviolet excitation. The electron cloud is distributed on the Cu cluster under visible excitation, which corresponds to the result that the Cu cluster receives electrons under optical excitation from the in-situ XPS measurement.
To further clarify the reasons for the excellent photocatalytic activity and HCOOH conversion selectivity, in-situ DRIFTS spectroscopy was performed to probe the intermediates53,54,55,56,57,58. As shown in Fig. 4e, new peaks obviously appeared and their intensities gradually increased with the illumination time from 0 to 60 min. The absorption peaks located at 1596, 1578, and 1275 cm−1 belong to bidentate carbonate (b-CO32−), and the peaks at 1541, 1506, 1473, 1351, and 1330 cm−1 are attributed to monodentate carbonate (m-CO32−). Additionally, the peaks at 1389 and 1403 cm−1 are assigned to polydentate carbonate (p-CO32−), and the peaks at 1453, 1434, and 1411 cm−1 are ascribed to bicarbonate (HCO3−). The peak intensities of these carbonates increase significantly, which suggests that MCOF-Ti6Cu3 can chemically adsorb and interact with CO2 and H2O molecules. The peak at 1641 cm−1 and the emerging peaks at 1742, 1761, and 1775 cm−1 maybe result from a formate, and the peak at 1628 cm−1 could be attributed to carboxylate. Meanwhile, the intensities of the peaks at 1560, 1526, and 1235 cm−1 corresponding to *COOH increased with the prolonged illumination time, which might be possibly caused by the favorable proton capture capability of CO2•− radicals. Most importantly, the intensity of the peak at 1709 cm−1 gradually increased with light progresses53,57. This peak is ascribed to *HCOOH which are important intermediates for the formation of HCOOH.
According to the results of in-situ DRIFTS analysis and DFT calculation, the possible mechanism of CO2 reduction and H2O oxidation over the hetero-metallic cluster catalyst MCOF-Ti6Cu3 can be reasonably proposed (Fig. 5a). The specific calculation models are in Supplementary Figs. 47, 48 and 49. We conjectured that the Cu cluster unit could be the active site for CO2 reduction, while the adsorption and activation of H2O were at the Ti cluster (Fig. 5b). To be specific, under irradiation, the photogenerated electrons and holes separated and transferred to the Cu cluster and Ti cluster, respectively. Then reduction and oxidation reactions took place simultaneously on these two clusters (Fig. 5c). On the Cu cluster, the conversion of CO2 to HCOOH requires two consecutive hydrogenation steps. The first step is the conversion of CO2 into *COOH, and then the formed *COOH undergoes a second reaction to generate HCOOH. Correspondingly, the oxidation of H2O at the Ti cluster involves twice the conversion processes of H2O into OH•. Finally, when HCOOH is desorbed from the system, the two OH• radicals combine with each other to generate O2 and H2O, thus maintaining the charge balance of the catalytic system. As shown in Fig. 5d, the formation of *COOH and OH• from CO2 and H2O is the rate determining step in the process of the photocatalytic overall reaction. In addition, two intermediates, *COOH and HCOO*, are generated in the first hydrogenation reaction. The *COOH is preferentially generated since the energy of the formation of *COOH is lower than that of HCOO*. The energy of *COOH dehydration into CO is much higher than that of *COOH hydrogenation into HCOOH, accounting for the preferential generation of HCOOH among the reduction products, which is consistent with the experimental results.
In summary, in order to realize an effective coupled photocatalytic oxidative and reductive overall reaction, we have elaborately designed and constructed a hetero-metallic cluster catalyst via linkage of dynamic covalent bonds, namely MCOF-Ti6Cu3, in which functional clusters with reductive and oxidative activities can cooperate for photocatalytic CO2 reduction and water oxidation, respectively. In addition, hetero-metallic cluster catalyst obtained by directed assembly inherits the oxidative and reductive properties of the clusters and achieves spatial separation. Also, the dynamic covalent bonds have more efficient electron transfer, enabling MCOF-Ti6Cu3 to exhibit excellent photocatalytic performance with a high HCOOH yield of 169.8 μmol g−1 h−1 in crystalline materials with only H2O. More importantly, the in-situ tests and DFT calculations prove that the reduction reaction occurs on the Cu cluster and the oxidation reaction on the Ti cluster, which is consistent with our pre-designed oxidative and reductive centers on hetero-metallic cluster catalyst, as well as distinctly reveals the charge transfer path and overall photocatalytic reaction mechanism on the atomic level. This research provides a strategy for the construction of hetero-metallic cluster catalyst from the chemical cooperation of MOF ∪ COF, and offers a significant guideline for the design of the photocatalytic overall reaction systems.