Controlled synthesis of Bi- and tri-nuclear Cu-oxo nanoclusters on metal–organic frameworks and the structure–reactivity correlations

Precisely tuning the nuclearity of supported metal nanoclusters is pivotal for designing more superior catalytic systems, but it remains practically challenging. By utilising the chemical and molecular specificity of UiO-66-NH2 (a Zr-based metal–organic framework), we report the controlled synthesis of supported bi- and trinuclear Cu-oxo nanoclusters on the Zr6O4 nodal centres of UiO-66-NH2. We revealed the interplay between the surface structures of the active sites, adsorption configurations, catalytic reactivities and associated reaction energetics of structurally related Cu-based ‘single atoms’ and bi- and trinuclear species over our model photocatalytic formic acid reforming reaction. This work will offer practical insight that fills the critical knowledge gap in the design and engineering of new-generation atomic and nanocluster catalysts. The precise control of the structure and surface sensitivities is important as it can effectively lead to more reactive and selective catalytic systems. The supported bi- and trinuclear Cu-oxo nanoclusters exhibit notably different catalytic properties compared with the mononuclear ‘Cu1’ analogue, which provides critical insight for the engineering of more superior catalytic systems.

Formic acid (FA) reforming 5 mL 0.7×10 -3 g/L catalyst was added into 5 mL 2% (v/v) FA in a quartz reactor. The reactor was then flushed with N 2 for 15 min to ensure an inert atmosphere. The photocatalytic reaction was performed using a xenon lamp (GLORIA-X500X, 500 W, 200-2500 nm, Zolix Instruments Co., Ltd.). The gaseous products were analyzed by gas chromatography (GC-2060). TCD was used for the detection of hydrogen with helium as the internal standard. FID was used for the detection of carbon monoxide and carbon dioxide.

Methods
High-resolution synchrotron X-ray powder diffraction (SXRD) and Rietveld refinement High-resolution SXRD data were collected at Beamline BL02B2, SPring-8, Japan. The energy of the incident X-ray flux was set at 18 keV. The tuned energy for each beamline emits the optimum X-ray flux to achieve high contrast (signal-to-noise ratio) and high angular resolution. The wavelength and the 2θ-zero point were calibrated using a diffraction pattern obtained from a high-quality standard CeO 2 powder. High-resolution SXRD data were obtained from the zeolite samples (loaded in 0.5-mm borosilicate capillaries) using the MYTHEN detector. Each diffraction pattern was collected for an hour for good statistics. In total, there are 490 hkl reflections measured (within the region of refinement (2θ = 2-60°), of which at least 100 independent hkl reflections are observed. From a mathematical perspective, the number of variables should not exceed the number of observables. In the Rietveld refinements performed in this work, the number of structural parameters has not exceeded 40. Using the TOPAS 6.0 software, the lattice parameters were obtained using Pawley and Rietveld refinement analyses of the diffraction patterns were performed. The background curve was fitted by a Chebyshev polynomial with an average of 16 coefficients. The Thompson-Cox-Hastings (pseudo-Voigt) function was applied to describe the diffraction peaks [2] . The scale factor and lattice parameters were allowed to vary for all the histograms. The final refined structural parameters for each data histogram were carried out using the Rietveld method with the fractional coordinates (x, y, z) and isotropic displacement factors (B eq ) for all atoms. In addition, the R wp and goodness-of-fit values (gof = R wp /R exp ) are helpful to indicate the quality of fit, where R exp represents the quality of the data.
(i) Framework atoms The crystallographic locations of the framework atoms may change slightly -a small deviation from the model by Wragg et al. Therefore, before the refinement of the entire structure with the guest Cu site, the framework atoms and the positions of atoms were first refined. This was performed to avoid a miscalculation of the structure that reaches the global minimum of the refinement by changing the entire framework.
(ii) Fourier analysis The Fourier analysis was used to identify the positions with the highest remaining electron density in the framework, once the positions of the host atoms have been determined.
(iii) Inclusion of guest molecules Based on the Fourier analysis, a Monte Carlo-based simulated annealing technique in which the guest Cu complex as the rigid body was used to locate their positions in the UiO-66-NH 2 . The guest Cu complex rigid body Z-matrices were refined while keeping the fractional coordinates of the framework atoms fixed. It was first applied to be simulated annealed using the Rietveld method (while the fractional coordinates of the framework were fixed). Thus, the simulated annealing technique ensures the correct number of Cu complexes. After identifying the number and location of the Cu sites, the site occupancy factors (SOFs) were refined. Also, the bond and dihedral angles of Cu-OH in the rigid-body were refined with restraining to ± 20% from the optimized tetrahedral structures using Chem3D software. Then, the relevant parameters were relaxed to be refined by simulated annealing repeatedly for an hour to ensure the global minimum has been reached. The global minimum is indicated by the lowest R wp and gof values. Several criteria were met to ensure the high quality and reliability of the refinement, namely, (i) the global minimum has been reached, (ii) the derived crystal structure fits chemical sense, (iii) reasonable systematic error values for all the refined parameters, and (iv) sensible SOF and B eq values. The B eq were constrained in the following way: (i) all the framework Zr-sites share the same value as 0.6 Å 2 , and the values for the framework O-sites as 2 Å 2 , and (ii) the B eq of the extra-framework atoms were all arbitrary fixed at 8-12 Å 2 (as B eq is broadly accepted to be about proportional to measurement temperature as that of framework sites) [3][4][5][6][7] . The position errors of the Cu complex were estimated from the percentage errors of the translation and rotation axes of the rigid bodies.
High-throughput SXRD High-throughput SXRD measurements for in situ desorption study were collected at beamline BL02B2 at SPring-8, Japan. The energy of the incident X-ray flux was set at 18 keV 2 . The wavelength (λ = 0.689556(2) Å) and the 2θ zero-point (ZP = -0.000015(2)°) were calibrated using a diffraction pattern obtained from a highquality CeO 2 powder (NIST SRM674b). High-throughput SXRD data were obtained from the zeolite samples (loaded in 0.5-mm borosilicate capillaries)) using the MYTHEN detector. The patterns were collected in the 2θ range 2-78° with 0.006° data binning. Each SXRD pattern was collected for 5 min for each MYTHEN-2θstep, i.e., 10 min in total for MYTHEN data summation. This produced patterns with a good signal-to-noise ratio (S/N). Therefore, the quality of the Rietveld refinement should be best judged by the difference between the fitted and observed data.
In-situ PXRD measurements Pre-treated samples were finely sieved and loaded in 0.5-mm-borosilicate capillaries to reduce the X-ray absorption problem. High-energy X-ray using Mo anode was used to optimize the spatial and angular resolution of Bragg's reflections. Dynamic measurements were conducted at elevated temperatures at 5 °C min -1 . Each powder XRD pattern required 15 min of scanning time for a suitable and reliable signal-to-noise ratio.
Extended X-ray absorption fine structure spectroscopy The extended X-ray absorption fine structure (EXAFS) spectroscopy data were collected at BL07A at Taiwan Light Source using transmission mode, with an average scanning time of 20 minutes. Artemis and Athena software were used for data treatment and analysis. [8] The detailed fitting parameters are summarized in the caption of the EXAFS fittings. The Hamma software was used for wavelet transform [9] .
Fourier-transform Infrared (FTIR) spectroscopy FTIR spectroscopy experiments were performed in Thermo Scientific Nicolet IS50 with attenuated total reflection (ATR) mode. The samples (10 mg, 1CuO and 2CuO) were treated by adsorbing 30 μL FA. The samples without FA adsorption were used as backgrounds. The spectra of adsorbed FA on 1CuO and 2CuO were collected by subtracting the corresponding backgrounds.
Calculation Setup for 1Cu 2+ -2Cu 2+ -, 3Cu 2+ , and 4Cu 2+ -UiO-66-NH 2 For all the calculations within this work, we have applied the DFT calculations within the CASTEP code. [10] The GGA and PBE exchange-correlation functionals are selected for all the calculations. [11,12] The cutoff energy of plane-wave basis sets based on the ultrasoft pseudopotential has been set to 440 eV with the selection of the algorithm Broyden-Fletcher-Goldfarb-Shannon (BFGS) for all the geometry optimizations. [13] The Monkhost-Pack reciprocal space integration was performed using coarse k-points with a mesh of 2×2×1 [14] , which was guided by the initial convergence test. With these settings, the overall total energy for each step is converged to less than 5.0 ×10 -5 eV per atom. The Hellmann-Feynman forces on the atom were converged to less than 0.001 eV/Å. Calculation Setup for 1CuO-, 2CuO, and 3CuO-UiO-66-NH 2 We have applied the DFT calculations based on the CASTEP packages to investigate the electronic structures and energetic trend of formic acid decomposition. [10] To accurately describe the exchange-correlation energy, we have selected the generalized gradient approximation (GGA) and Perdew-Burke-Ernzerhof (PBE) for this work. [11,12,15] The cutoff energy of the plane-wave basis has been set to 440 eV with the ultrafine quality with the utilization of the ultrasoft pseudopotentials for all the geometry optimizations. The Broyden-Fletcher-Goldfarb-Shannon (BFGS) algorithm has been applied in this work with the coarse quality of k-points for all the energy minimizations in this work. [13] To accomplish the geometry optimizations, the convergence test requires the total energy difference should be less than 5×10 -5 eV per inter-ionic displacement as 0.005 Å per atom, respectively. To calculate the number of moles of ATA organic linker:

Results and Discussion
To calculate the number of moles of ZrO 2 :
The crystal structure of UiO-66-NH 2 is preserved. No new crystalline peak is found, indicative of the absence of new phases, e.g., Cu or CuO.  Mass of water and solvent to be determined using 1Cu 2+ -UiO-66-NH 2     Investigation on whether the Cu sites are homogeneously dispersed inside MOFs. Typically, these structural parameters can be obtained from the quantitative analysis of the diffraction measurements. It should also be noted that we have particularly chosen to perform the SXRD measurements on BL02B2 at SPring-8 because of the highly optimised optical and instrumental configurations for the analysis of homo/heterogeneity. [16] The (111) reflection (2theta = 3.27°) as an example, peak patterns of the UiO-66-NH 2 , 1CuO, 2CuO, and 3CuO samples are compared in Figure S24. Theoretically, if the extra-framework species were disordered inside MOFs (such as more metals closer to the surface of the MOF), incoherent diffraction would be generated. As a consequence, the peak position will be partially moved, leading to symmetric peaks. Also, the Bragg peaks of 1CuO, 2CuO, and 3CuO are all symmetrical (see Table S6; all at 0.0001), indicating the samples are very homogeneous. This suggests that the metalation processes (in 1CuO, 2CuO, and 3CuO) are also homogeneous, which agrees with our recent study on the single-atom metalation on UiO-66-NH 2 . [1] By further quantitative analysis of the SXRD patterns (Table S6), we have obtained the volume-weighted mean column heights, LVol which are calculated using FWHMs and integral breadths (IB) assuming intermediate crystallite size broadening modelled by Voigt function. Similarly, the micro-strain, e 0 (dislocations, vacancies, and other defects) was calculated from FWHMs. We found the symmetry of UiO-66-NH 2 (peak asymmetry parameter of 0.0001) remains after the fabrication of 1CuO, 2CuO, and 3CuO, which confirms the homogeneous and ordered dispersion of Cu sites inside MOFs.   Cu-modified UiO-66-NH 2 samples with and without the Cu-oxo species to improve the reliability and confidence of the fittings. As presented in Figures S23-25 and Table S7, the impact on the removal of the Cuoxo species from the refinement is substantial where the mismatch in fitting is particularly obviously between 2θ = 5-10°, which primarily have been ascribed to the occupation of the extra-framework species. [17] Furthermore, we have also systematically presented a comparison between the Bragg peak intensities of 1CuO, 2CuO, and 3CuO in Table S8, as it is difficult to virtually see the differences in a standard graphical presentation. Clearly, obvious differences in the peak intensities can be seen in reflections such as (222),    In the following, we compared the Rietveld refinement structures with the XAS experimental results and investigate the scattering contributions coming from different neighbouring atoms. The single scattering paths from the Cu absorber of all the samples include the Cu-O path in the first shell and the Cu-Zr path in the second shell. However, 2CuO-UiO-66-NH 2 and 3CuO-UiO-66-NH 2 contain Cu-Cu paths in the second shell. Because Cu has a much higher electron density than C, O, and N, Cu has a much greater influence on the EXAFS profiles. Cu 1 st and 2 nd shell bonding information for the simulated EXAFS profiles (k-space ( Figure S27), R-space ( Figure S28) and WT ( Figure S29) (1) 0.005(1) S 0 2 was fixed at 0.867 and enot was returned to -4.86 ± 1.12 eV. Data range 3 ≤ k ≤ 10 Å -1 , 1.0 ≤ R ≤ 3.0 Å. 4 variable parameters used; 8.6875 independent data points; R factor = 0.80%.  (1) 0.006(2) Scale factor was fixed at 0.848 and enot = -2.54 ± 1.97eV. Data range 2 ≤ k ≤ 11 Å -1 , 1.0 ≤ R ≤ 2.1 Å. R factor = 1.28%. All Cu-O scattering paths were generated using monoclinic CuO (space group C2/c).  (1) 0.003(2) Scale factor was fixed at 0.848 and enot = -0.63 ± 2.20 eV. Data range 2 ≤ k ≤ 11 Å -1 , 1.1 ≤ R ≤ 2.1 Å. R factor = 1.22%. Cu-O 12 (fixed) 2.53(1) 0.008(5) Scale factor = 0.848 ± 0.049 and enot = 3.48 ± 0.61 eV. Data range 3 ≤ k ≤ 14 Å -1 , 1 ≤ R ≤ 3 Å. R factor = 0.55%. The calculated scale factor was used for the subsequent fitting analyses.