A stable covalent organic framework for photocatalytic carbon dioxide reduction

A metal-decorated alkene-linked covalent organic framework is a robust, selective photocatalyst for CO2 reduction.


Introduction
There is now little doubt that the human production of CO 2 has contributed to climate change, which will affect life on earth. 1 One potential approach to reducing CO 2 emissions is its conversion into value-added products using solar energy. 2,3 A range of inorganic semiconductors has been investigated as photocatalysts for CO 2 reduction, such as TiO 2 , 4 ZrO 2 , 5 In 2 O 3 , 6 CdS, 7 or ZnGa 2 O 4 . 8 Unfortunately, many inorganic photocatalysts either have unsuitably aligned conduction/valence band positions or relatively large band gaps, hence limiting visible light absorption. By contrast, the band gap in organic semiconductors can be tuned readily through the incorporation of a diverse range of monomers. [9][10][11] In recent years, porous organic materials such as carbon nitrides, 12-14 conjugated microporous polymers (CMPs), 15,16 covalent triazine-based frameworks (CTFs) 17 and hyper-crosslinked polymers (HCPs) 18 have been studied for photocatalytic CO 2 reduction. Those organic materials are typically amorphous; by contrast, covalent organic frameworks (COFs) can combine porosity with crystallinity. [19][20][21][22][23] COFs have been investigated as photocatalysts for water splitting, 24,25 and for electrocatalytic CO 2 reduction. 26,27 These materials also have potential for direct photocatalytic CO 2 reduction: for example, an azine-based COF, N 3 -COF, was shown to exhibit gas phase photocatalytic CO 2 reduction. 28 Likewise, a 2D imine triazine-COF loaded with rhenium 29 and a b-ketoenamine-linked COF decorated with both nickel and a light-absorbing dye 30 were studied for the same reaction. All of these COFs have limited effective conjugation lengths in the 2D plane of the framework because they are based on imine, azine, or b-ketoenamine-linkers. This results in blue-shied absorption on-sets, which limit the ability of the materials to absorb visible light. 31,32 Here, we used Knoevenagel condensation (Fig. 1a) such that olens become the COF linkers. 31,33,34 Our aim was to increase the conjugation length in the framework and hence, perhaps, to improve the performance of these materials for CO 2 reduction. The cyanovinyl-groups as a result of the Knoevenagel condensation have been shown to be benecial for CO 2 uptake 35 which might increase the efficiency of CO 2 reduction. The COF was loaded with [Re(CO) 5 Cl] giving a heterogeneous analogue of the well-studied homogeneous catalyst [Re(bpy)(CO) 3 Cl] with enhanced stability. 36
Nitrogen sorption experiments were performed at 77 K and the Brunauer-Emmett-Teller surface area (SA BET ) for Bpy-sp 2 c-COF was calculated to be 432 m 2 g À1 . This SA BET is lower than that predicted for the atomistic model of a perfectly crystalline structure (2041 m 2 g À1 ), but this is commonly observed for sp 2 c-COFs which typically have surface areas ranging from 322 m 2 g À1 for sp 2 c-COF-2 (ref. 37) up to 692 m 2 g À1 for sp 2 c-COF. 31 The pore size distribution prole shows a narrow pore size distribution with a pore width of 2.4 nm (Fig. 2a, inset curve), further supporting an AA stacking sequence that is predicted to have a pore size of 2.4 nm. Fourier-transform infrared (FT-IR) spectroscopy shows a distinct peak at 2217 cm À1 relating to a -C^N vibration band, indicating the formation of Bpy-sp 2 c-COF (Fig. 2c).
The BET surface area (SA BET ) for Re-Bpy-sp 2 c-COF was calculated to be 323 m 2 g À1 (Fig. S38 †). Scanning transmission electron microscopy (STEM) and energy-dispersive X-ray spectroscopy (EDX) mapping images (Fig. 2b) show a uniform distribution of C, N, O, and Re in Re-Bpy-sp 2 c-COF, further suggesting that the Re moiety has been incorporated throughout the structure of COFs. Inductively coupled plasma- optical emission spectrometry (ICP-OES) measurements show that 18 wt% of Re has been incorporated into the material. This ratio corresponds to ligation of half of the bipyridine sites by the Re complex, which was also the ratio in the atomistic model built to represent Re-Bpy-sp 2 c-COF (Fig. 1c). The FT-IR spectrum for the Re-modied COF (Fig. 2c) shows new peaks at 1900 cm À1 , 1917 cm À1 , 2024 cm À1 , corresponding to the COstretching bands of the incorporated [Re(CO) 3 Cl] complex. 29,36 UV-visible diffuse reectance spectra (Fig. 2d) show a red-shi of the absorption edge from 589 nm to 694 nm for Re-Bpysp 2 c-COF compared to Bpy-sp 2 c-COF. Finally, X-ray photoelectron spectra for Re 4f, Cl 2p and N 2s regions look very similar for Re-Bpy-sp 2 c-COF compared to the molecular catalyst [Re(bpy)(CO) 3 Cl] indicating that complexation of Re is similar in both cases (Fig. S40 †).
We next studied the CO 2 uptake for this material up to 1200 mbar at both 273 and 298 K. Re-Bpy-sp 2 c-COF adsorbs 1.7 mmol g À1 CO 2 at 273 K and 1.1 mmol g À1 at 298 K (Fig. 2e). Re-Bpysp 2 c-COF has a high isosteric heat of adsorption (31 kJ mol À1 ), showing that the COF has good affinity toward CO 2 (Fig. S19 †).
Photocatalytic CO 2 reduction experiments were conducted in a quartz ask under 1 atmosphere CO 2 in a mixture of acetonitrile (MeCN) and triethanolamine (TEOA) in 30 : 1 ratio and under visible light illumination (l > 420 nm, 300 W Xe light source). TEOA acts as the sacricial electron donor and proton source, while MeCN is used to disperse the catalyst. Over a total of 17.5 hours irradiation under visible light ( Fig. S16 †), Re-Bpysp 2 c-COF produced CO with a rate of 1040 mmol g À1 h À1 and 81% selectivity over H 2 , which equals to a TON of 18.7 for CO, outperforming its homogeneous counterpart under the same conditions, which is deactivated aer 3 hours with a TON of 10.3 ( Fig. S16 †).
An apparent quantum yield (AQY) of 0.5% was measured at 420 nm for CO production. The small amount of H 2 possibly originates from competing proton reduction of water traces in the TEOA or oxidative dehydrogenation of TEOA. 13 In the absence of the Re-complex, Bpy-sp 2 c-COF only generated trace amounts of CO while H 2 was not detected. No other liquid phase products, i.e. HCOOH and methanol, were observed.
The combination of CO 2 conversion rate and CO/H 2 selectivity of Re-Bpy-sp 2 c-COF compares favorably with other reported COFs (see Table S3 †). For example, a rhenium modied 2D imine triazine-COF produced around 750 mmol g À1 h À1 CO with 98% selectivity, 29 and a b-ketoenamine-linked COF modi-ed with nickel, plus the use of an additional dye, gave a CO production rate of 811 mmol g À1 h À1 CO with 96% selectivity. 30 These comparisons should be made with caution, though, since photocatalytic rates also depend strongly on the precise experimental set-up that is used. 41 Control experiments were carried out to conrm that the source of the CO generated is indeed a photocatalytic process (Table S2 †). Under an argon atmosphere in absence of CO 2 , Re-Bpy-sp 2 c-COF generated 14.9 mmol g À1 h À1 CO and 285.3 mmol g À1 h À1 H 2 . The small amount of CO produced is possibly a result of decomposition of organic residues during photocatalysis 1 or decomposition of TEOA as an ineffective sidereaction. 13 No gas production was observed in the dark or in the absence of hole scavenger. Experiments with isotopically labelled 13 CO 2 resulted in the formation of 13 CO, strongly suggesting that CO 2 was the source of the produced CO (Fig. 3a).
Re-Bpy-sp 2 c-COF appears to be stable under the photocatalysis conditions, evident from the post-illumination FT-IR spectra (Fig. S7 †) of the sample aer 17.5 hours of continuous visible light illumination (l > 420 nm, 300 W Xe light source). The material also retains most of its crystallinity and PXRD patterns show that some order is retained aer photolysis for 17.5 hours (Fig. S11 †). This shows that the material has very good stability compared to other previous reports. 29,30 When the run was extended to a total of 50 hours a further loss of crystallinity is observed ( Fig. S43 and S44 †) along a loss of activity, highlighting that stability is still one of the important challenges in the eld. Nevertheless, it seems that in making a heterogeneous analogue of [Re(bpy)(CO) 3 Cl] an increase in stability is observed (Fig. S43 †), possibly by preventing the formation of the dimer of the Re-complex 36 which can occur with Re(bpy)CO 3 Cl in solution, or by the shielding of the Re-centre within the COF structure from photodecomposition side reactions. 42 Photoelectrochemical experiments were conducted using FTO glass as a photocathode in 0.1 M Na 2 SO 4 solution (Fig. 3b). All samples were tested at a constant voltage of 0.5 V vs. reversible hydrogen electrode (RHE). The photocurrent of Re-Bpy-sp 2 c-COF photocathode was about 2 mA cm À1 , which was more than four times higher than a Re-free Bpy-sp 2 c-COF photoanode. Additionally, Nyquist plots (Fig. 3c) showed the arc radii for Bpy-sp 2 c-COF and Re-Bpy-sp 2 c-COF under irradiation were smaller than those in dark, verifying that charge carriers were generated in Bpy-sp 2 c-COF and Re-Bpy-sp 2 c-COF under irradiation. The Nyquist plots of Re-Bpy-sp 2 c-COF under irradiation have smaller semicircles than those of Bpy-sp 2 c-COF. Both measurements taken together show that the Re bearing material acts as a better photo-electro catalyst indicating that the material is better at separating and transferring charges, which is also in line with computational predictions (vide infra).
We then went on to use emission spectroscopy to study the mechanism of the photocatalysis for the Re loaded Bpy-sp 2 c-COF. The photocatalyst Bpy-sp 2 c-COF in acetonitrile suspension shows the presence of two emissive states, with l max at 475 and 640 nm. The excitation spectrum shows the 640 nm emission arises from a broad range of absorption bands in the UV/vis spectrum (from 300 to 500 nm), in contrast the sharp emission band centred at 475 nm is a result of excitation into a single band at 390 nm (Fig. S20 †). Time-correlated single-photon counting (TCSPC) measurements show the lifetime of the 640 nm emissive state of Bpy-sp 2 c-COF is insensitive to the TEOA scavenger (Table S1 and Fig. S21 †) and the emission yield is also unchanged (Fig. S20 and S21 †). In contrast the 475 nm emission lifetime (2.48 ns to 0.72 ns) and yield is very sensitive to the presence of the TEOA electron donor (Fig. S20 and S21 †), indicating that reductive quenching of this excited state can occur. Previous studies on closely related sp 2 c pyrene COFs have also reported the presence of two emissive states for COF samples. 37 Therein, emission at 640 nm was attributed to the presence of a delocalised excited state across both pyrene and the sp 2 -carbon backbone on the basis of the signicantly red-shiing of the emission when compared to that typically measured for excimer state of pyrene systems alone (ca. 480 nm). Interestingly following exfoliation of the COF, a second emission at ca. 468 nm was observed, proposed to be due to exfoliated COF where the removal of the p-p stacking force allows twisting of the structure and a loss of conjugation across the backbone. Here, the samples are sonicated prior to use and a similar assignment is also proposed.
Addition of the Re catalytic site to the COF framework leads to a marked change in the measured photophysical behaviour. With Re-Bpy-sp 2 c-COF a single emissive state (l max ¼ 475 nm), proposed to be due exfoliated COF material remains. Such an assignment is in-line with the noted insensitivity of emission at this wavelength to the presence of the Re centre as the loss of conjugation of the exfoliated structure may be expected to prevent efficient electron or energy transfer from the COF framework to the Re centre. Signicantly the delocalised COF excited 640 nm emissive state of Bpy-sp 2 c-COF is completely absent in the Re-Bpy-sp 2 c-COF (Fig. S22 †). The quenching of the emission by the Re centre indicates possible electron transfer from the COF backbone to the catalytically active Re complex. The assignment of the sensitisation of the catalytic centre following photon absorption by the COF framework is supported by the DFT calculations below and the good agreement between the wavelength dependent CO measurement (Fig. S35 †) and the excitation spectrum of the 640 nm emission of the Bpy-sp 2 c-COF sample (Fig. S20 †).
To further explore the photophysics of the system we have also carried out transient absorption (TA) spectroscopic studies on both Re-Bpy-sp 2 c-COF and Bpy-sp 2 c-COF (Fig. 4). Following excitation at 400 nm, 800 mW (5 kHz) of Bpy-sp 2 c-COF we observe complex TA spectra with broad negative bands between 450 to ca. 700 nm that formed within 0.5 ps. There is minimal absorption by the ground state of Bpy-sp 2 c-COF (Fig. 2d) at wavelengths longer than 600 nm. Therefore, the negative signal is proposed to be the overlap of stimulated emission from both the conjugated Bpy-sp 2 c-COF structure (l max ¼ 640 nm) and the exfoliated Bpy-sp 2 c-COF (l max ¼ 475 nm), overlapped with the ground state bleach, giving rise to the complex shape. The complex nature of the bleach makes determining accurate kinetics challenging, therefore we use t 50% (the time taken for the bleach to decrease by 50%) as a rough measure of the lifetime of the photogenerated excited state. For Bpy-sp 2 c-COF at 550 nm, t 50% ¼ ca. 5 ps (Fig. S46 †). Within 0.5 ps a photoinduced absorption (PIA) is present at 770 nm, which decays within 10 ps to form a new PIA centred at 700 nm that decays over the course of the experiment to leave only a small PIA by 3 ns.
Re-Bpy-sp 2 c-COF shows a simpler TA spectrum following 400 nm excitation (Fig. 4). A negative band is formed, centred at 540 nm, again assigned to a combination of ground state bleaching and stimulated emission from the exfoliated COF framework (l max ¼ 475 nm; Fig. S23 †). It is notable that this negative feature is substantially narrower than that observed for Bpy-sp 2 c-COF and for Re-Bpy-sp 2 c-COF at 550 nm, t 50% ¼ ca. 200 ps, signicantly longer than observed in the absence of the Re. A PIA centred at ca. 770 nm is again observed to be formed within 0.5 ps, with a blue shi of this initially formed PIA observed within the rst 5 ps, forming a band centred at $720 nm. This state continues to decay over the course of the time-delays probed, as the negative band assigned to ground state bleaching and stimulated emission, recovers. Although no direct spectral ngerprint is observed for the formation of the reduced Re centre by TA spectroscopy in the UV/vis spectral region, it is clear from the simplication of the TA spectra, combined with the greatly increased lifetime of the ground state bleach, that the presence of the Re centre within the COF leads to the formation of a long-lived charge separated (non-emissive) state.
Density functional theory (DFT) and time-dependent (TD) DFT calculations-performed on representative molecular models Bpy-sp 2 c(L) and Re-Bpy-sp 2 c(L) of Bpy-sp 2 c-COF and Re-Bpy-sp 2 c-COF, respectively-show that the electron affinity (EA) and the ionization potential (IP) of both COFs straddle the reduction potential of CO 2 to CO, as well as the proton reduction potential, and the oxidation potential of TEOA ( Fig. S47 and  S48 †). This provides a thermodynamic explanation for the ability of Re-Bpy-sp 2 c-COF to drive CO 2 reduction to CO, in the presence of the sacricial agent TEOA. Relative energy levels of the dye and the molecular COF models conrm that it is thermodynamically allowed for excited electrons on the dye to be transferred to Re-Bpy-sp 2 c-COF (Fig. S48 †), in line with the dyesensitization effects observed experimentally.
TD-CAM-B3LYP calculations predict that the lowest-energy, excited electronic state (S1) for both Bpy-sp 2 c(L) and Re-Bpysp 2 c(L) corresponds to the LUMO ) HOMO transition, with a strong oscillator strength (Table S4 †). Electron distributions of the excited-state frontier orbitals show that for both Bpysp 2 c(L) and Re-Bpy-sp 2 c(L) the HOMO orbital is predominantly located on the pyrene unit of the COF, with the LUMO orbital mainly located on the bipyridine unit (with or without ligated Re complex; Fig. S49 and S50 †). Analyses of excited-state, interfragment charge transfer between the building units of the COFs indicate that appreciable amounts of electrons are transferred from the pyrene fragment to the bipyridine fragment (Table S5 †), with a sizable electron-hole distance as measured by the charge centroids of the orbitals involved (Dr in Table S4 †). Our computational results clearly support that there is electron transfer from the COF backbone to the catalytically active Re complex upon electronic excitation.
The CO 2 reduction mechanism of the COF compared to the homogenous catalysts is therefore different. Here we propose pyrene excitation to a bipyridine based LUMO. In contrast, in solution excitation upon irradiation forms a metal to bipyridine excited state ( 3 MLCT) which is then quenched by an electron donor. 43 Crystallinity 25 and accessible surface area 44 have been shown to be important factors for the photocatalytic activity of organic photocatalysts. To probe whether these factors also inuence the performance of the COF in photocatalytic CO 2 reduction, we synthesized an amorphous analogue by using 1,4-dioxane instead of a 1,2-dichlorobenzene/1-butanol mixture under otherwise exactly the same experimental conditions, Re-Bpysp 2 c-P (PXRD, see Fig. S12 †), which shows low CO 2 uptakes and BET surface area ( Fig. S19 and S39 †). Despite having comparable FT-IR, UV-visible and PL spectra (Fig. S8, S15 and S24 †), the amorphous polymer showed signicantly lower activity for CO 2 reduction (Table S2 †) aer being loaded with Re with a TON of 2.3 aer 12 hours compared to 12.9 for Re-Bpy-sp 2 c-COF. This highlights that morphological properties, such as crystallinity and porosity, are important in these materials.
We showed previously that COFs that have accessible pores can potentially act as a host for dyes giving rise to increased photocatalytic activity for hydrogen production. 25 Here, we used (Ir[dF(CF 3 )ppy] 2 (dtbpy))PF 6 (ppy ¼ 2-phenylpyridine, tbpy ¼ 4,4 0 -di-tert-butyl-2,2 0 -dipyridyl) in conjunction with Re-Bpy-sp 2 c-COF to further enhance the photocatalytic performances. Different amounts of the dye were used, and the CO production rates were enhanced by 32% and 84% compared to the unsensitized COF when using 0.3 mmol and 1.0 mmol of the dye, respectively, with 1 mg COF over 5 hours (Fig. 3e). The H 2 production rates were unaffected. The highest CO production rates were 1400 mmol h À1 g À1 , with a selectivity of 86% for CO, over 5 hours from Re-Bpy-sp 2 c-COF loaded with 1.0 mmol dye. It appears to be an electron transfer mechanism between the dye and the COF via oxidative quenching as suggested by emission quenching experiments (Fig. S32 †).
Finally, we explored Re-Bpy-sp 2 c-COF loaded with additional in situ photodeposited colloidal Pt as a photocatalyst for syngas production, hence, simultaneous evolution of CO and H 2 . Syngas is used in chemical industry on large scale for processes, such as Fischer-Tropsch, and control of the ratio is important. The production of syngas with tunable ratio of CO and H 2 has been reported for electrocatalysts 45,46 and inorganic photocatalysts. 47,48 By adding different amounts of Pt, Re-Bpy-sp 2 c-COF could produce high rates of CO-rich or H 2 -rich mixtures ranging from approximately 4 : 1 to 1 : 10 for CO : H 2 (Fig. 3f).

Conclusions
In conclusion, we have synthesized a new porous, crystalline bipyridine-containing sp 2 c-COF, which was post-synthetically modied with a rhenium complex to enhance the photocatalytic CO 2 reduction performance. Re-Bpy-sp 2 c-COF achieved a CO production rate of 1040 mmol g À1 h À1 with 81% selectivity over H 2 over 17.5 h illumination. This performance was enhanced over 5 hours by up to 84% by dye-sensitization, giving a CO production rate of 1400 mmol h À1 g À1 and a CO/H 2 selectivity of 86%. Based on a range of experimental and computational techniques it appears that Re-Bpy-sp 2 c-COF operates by a markedly different mechanism compared to the homogeneous catalyst [Re(bpy)(CO) 3 Cl] which the Re-Bpy-sp 2 c-COF also outperforms in terms of stability. Crystallinity and porosity seem to be important in these materials since an amorphous, lowporosity analogue showed almost no photocatalytic activity.