Improving the Photocatalytic Reduction of CO2 to CO through Immobilisation of a Molecular Re Catalyst on TiO2

The photocatalytic activity of phosphonated Re complexes, [Re(2,2′-bipyridine-4,4′-bisphosphonic acid) (CO)3(L)] (ReP; L=3-picoline or bromide) immobilised on TiO2 nanoparticles is reported. The heterogenised Re catalyst on the semiconductor, ReP–TiO2 hybrid, displays an improvement in CO2 reduction photocatalysis. A high turnover number (TON) of 48 molCO molRe−1 is observed in DMF with the electron donor triethanolamine at λ>420 nm. ReP–TiO2 compares favourably to previously reported homogeneous systems and is the highest TON reported to date for a CO2-reducing Re photocatalyst under visible light irradiation. Photocatalytic CO2 reduction is even observed with ReP–TiO2 at wavelengths of λ>495 nm. Infrared and X-ray photoelectron spectroscopies confirm that an intact ReP catalyst is present on the TiO2 surface before and during catalysis. Transient absorption spectroscopy suggests that the high activity upon heterogenisation is due to an increase in the lifetime of the immobilised anionic Re intermediate (t50 %>1 s for ReP–TiO2 compared with t50 %=60 ms for ReP in solution) and immobilisation might also reduce the formation of inactive Re dimers. This study demonstrates that the activity of a homogeneous photocatalyst can be improved through immobilisation on a metal oxide surface by favourably modifying its photochemical kinetics.


Introduction
The efficient and selective reduction of CO 2 is the key challenge to access carbon capture and utilisation (CCU) technologies. The green conversion of CO 2 into the energy carrier CO is particularly attractive as it could be used, in combination with H 2 , as an important chemical feedstock to form syngas, and replace the steam reforming of fossil fuels in the petrochemical industry. Sustainable syngas provides a direct route to a greener chemical industry sector and the generation of transport fuel through Fischer-Tropsch chemistry. [1] Photoelectrocatalytic CO 2 reduction has been achieved on semiconductors, [2] such as p-type Si, CdTe, InP, GaAs, GaP, Cu 2 O [3] and n-type TiO 2 . [4] However, semiconductor materials often suffer from poor selectivity, producing a range of carbon-based products as well as H 2 . In contrast, enzymes display excellent selectivity and high turnover frequencies; a carbon monoxide dehydrogenase adsorbed on dye-sensitised TiO 2 reduced CO 2 selectively to CO with a turnover frequency (TOF) of 530 h À1 . [5,6] Major drawbacks of enzymes are their large size giving a correspondingly low 'per volume' activity, fragility and the restriction of their operation to specific temperatures, pH ranges and anaerobic atmospheres.
Synthetic molecular photo-and electrocatalysts can provide good stability combined with high product selectivity. [7] They may also be tuned by choosing from a range of metals and ligands. [8] For example, Fe porphyrin electrocatalysts have been reported to operate with high Faradaic efficiency and rate with overpotentials between 410 mV and 890 mV, depending on the molecular structure. Other examples are Re-, Ni-, Ru-and Pd-based catalysts, which operate under such overpotentials. [7,9] The photocatalytic reduction of CO 2 using synthetic molecular catalysts can be achieved through two main strategies: They can receive electrons from an excited photosensitiser with appropriate energy requirements. Rhenium catalysts for CO 2 reduction have been used in photocatalytic dyads of this type. [10] A second approach consists of using molecular photocatalysts that can perform both the functions of light harvesting and chemical catalysis. [11] The rhenium tricarbonyl bipyridine complex with a defined leaving group L, [Re(bpy)(CO) 3 (L)], is one of very few mononuclear compounds that falls into this second category and acts as both a light absorber and a catalyst for the photoreduction of CO 2 to CO. [11a-c] However, the two major drawbacks of such Re catalysts are their inefficient absorption of solar radiation and their low photostability and turnover numbers to generate CO (TON CO ). The highest TON CO reported to date for a Re complex acting as both light absorber and catalyst in solution in DMF/triethanolamine (TEOA) is 30, using l > 400 nm irradiation (L = Cl À ). [11a, 12] A Re derivative with a thiocyanate leaving group gave a TON CO of 26 (2 h, l > 400 nm) [8b] or a TON CO of 30 (25 h, l = 365 nm). [13] A dinuclear system gave a TON CO per Re centre of 30 after one hour of irradiation (l = 450 nm, LED). [14] Until recently, photocatalysis with Re complexes had only been performed in homogeneous solution. [Re(bpy)(CO) 3 L] complexes have been immobilised in various environments, but few of these systems have been reported as CO 2 reduction photocatalysts. [15] Emerging heterogeneous systems [16] are advantageous as they facilitate catalyst recycling and increase the flexibility of solvents that may be used. [12,17] However, heterogenised Re photocatalysts have to date shown only low efficiency, with a maximum reported TON CO of 7 in the absence of an additional dye. [12] Herein we report two novel phosphonated Re bipyridine photocatalysts (ReP) for selective CO 2 to CO reduction. Quantitative immobilisation of the ReP catalysts onto TiO 2 nanoparticles results in an improvement of the TON CO compared to previously reported homogeneous Re catalysts. To explain the nature of this enhanced photoactivity of ReP, the reaction mechanism and the kinetics of the catalytic intermediates were studied by time correlated-single photon counting and transient absorption spectroscopy. This spectroscopic study reveals an enhanced stability of the reaction intermediates when immobilising the ReP catalysts onto TiO 2 and provides evidence for a reduction in the concentration of inactive Re dimers formed during catalysis, resulting in a 26-fold increase in the CO 2 reduction yield for the heterogenised ReP on TiO 2 compared to homogeneous ReP.

Synthesis and characterisation of ReP
Rhenium catalysts with phosphonated bipyridine ligands for immobilisation on metal oxides, three CO ligands in facial configuration and a 3-picoline (ReP pic ) or bromide (ReP Br ) leaving group were prepared (Figure 1). Heating [ReBr(CO) 5 ] with tetraethyl 2,2'-bipyridine-4,4'-bisphosphonate in refluxing benzene gave Et ReP Br in 92 % yield. The bromide ligand in Et ReP Br was replaced upon heating with AgPF 6 and 3-picoline in THF to yield Et ReP Pic in 77 %. The ethyl ester groups were dealkylated with ISiMe 3 in CH 2 Cl 2 followed by methanol treatment to give ReP Pic in 39 % yield. ReP Br was synthesised by heating [ReBr(CO) 5 ] with 2,2'-bipyridine-4,4'-bisphosphonic acid in a refluxing toluene/methanol mixture in 44 % yield, or by reaction of Et ReP Br with BrSiMe 3 in 62 % yield.
The composition and purity of the ReP complexes were confirmed by 1 H, 31 C and 31 P NMR, UV/Vis and FT-IR spectroscopies, mass spectrometry and elemental analysis (see Experimental Section and Figures S1-S8 in the Supporting Information). Single crystals of Et ReP Br suitable for X-ray diffraction analysis were obtained from CH 2 Cl 2 /hexane and the crystal structure is shown in Figure 1 B. Et ReP Br displayed a Re1ÀN1 bond length of 2.163(7) , a Re1ÀBr1 bond length of 2.6291(10) and a torsion angle for NCCN in the 2,2'-bipyridine of 6.3(11)8. In comparison [ReCl(4,4'-dimethyl-2,2'-bipyridine)(CO) 3 ] showed a ReÀN bond length of 2.172 , a ReÀCl bond length of 2.489 and a torsion angle in the bipyridine of 3.018. [8a] ReP pic was detected by ESI-MS as a doubly deprotonated species in the negative ion mode and is likely to exist in a singly deprotonated form as a zwitterionic, overall charge-neutral complex in solution. The absence of signals for the PF 6 À counter ion in the 31 P NMR spectrum provided further evidence for a charge-neutral complex. Elemental analysis indicated that in ReP Br the phosphonic acid groups are fully protonated and therefore that ReP Br is also uncharged. ReP pic and ReP Br were both soluble in DMSO and H 2 O, but only ReP Br in DMF.

Assembly and characterisation of ReP-TiO 2 hybrid
The quantitative immobilisation of ReP pic on dispersed TiO 2 nanoparticles (NPs; Evonik Aeroxide P25; 21 nm diameter) at sub-monolayer concentration was confirmed by UV/Vis spectrophotometry (l = 370 nm assigned to MLCT band in ReP pic ). No ReP pic remained in the supernatant after stirring ReP pic (0.1 mmol) in aqueous solution with TiO 2 (5 mg) for 1 h and centrifugation of the ReP pic -loaded TiO 2 NPs (see the Supporting Information, Figure S9). ReP was loaded in water due to the insolubility of ReP pic in most common organic solvents and to avoid adsorption of a carbon containing solvent on the TiO 2 surface (to facilitate XPS analysis; see below). The maximum loading capacity was determined by UV/Vis spectrophotometry as 0.4 mmol per 5 mg TiO 2 , which is comparable to the loading capacity of phosphonated cobaloxime and ruthenium tris(2,2'-bipyridine) complexes on TiO 2 . [18] Several techniques were subsequently used to characterise the ReP pic -TiO 2 NP hybrid ( Figure 2). An IR spectrum of ReP pic in MeOH solution displayed CO stretching vibrations at 2037 and 1932 cm À1 (Table 1), whereas an attenuated total reflectance (ATR) spectrum gave signals at 2025 and 1890 cm À1 . This difference in wavenumber is due to solid-state effects. Once loaded onto TiO 2 but before exposure to DMF/TEOA, the ATR spectrum gives wavenumbers of 2039 and 1931 cm À1 , which are very similar to the spectrum in MeOH solution. On the particle, the molecules of ReP pic are dilute and experience no solid-state effects as they cannot interact with one another.
Signals corresponding to phosphorus-oxygen bond stretches can be observed in the ATR spectra (Table 1). There is little difference between the P=O and PÀOEt stretches for Et ReP Br and Et ReP pic despite the difference in charge between the complexes. The P=O stretch in ReP pic shows a low wavenumber shift of 100 cm À1 compared with Et ReP Br and Et ReP pic . There is little difference between the P=O signal for unbound ReP pic and that when bound to TiO 2 .
X-ray photoelectron spectroscopy (XPS) measurements of ReP Br -TiO 2 confirmed the elemental composition upon immobilisation on TiO 2 . ReP Br was utilised for XPS studies due to the additional element, which provides a distinct handle for the bromide (as leaving group) compared to the picoline complex. XPS measurements showed a composition of 1.08 % Re, 0.49 % Br, 2.74 % P and 2.52 % N, in good agreement with the expected 1:1 ratio for P to N and 2:1 for N to Re (Figure 2, as well as Figure S11 and Table S1 in the Supporting Information). The low Br percentage suggests that the Br is rapidly substituted in the aqueous solution. [19] Photocatalytic activity The ReP-TiO 2 system displays CO 2 photoreduction activity under UV-filtered simulated solar light (100 mW cm À2 , l > 420 nm). Different amounts of ReP pic on TiO 2 in 5:1 DMF/TEOA (4.5 mL) were studied to identify optimal conditions for the heterogenised Re catalyst; different filters and metal oxide nanoparticles were also used. The turnover number (TON) is defined as molecules of CO produced per molecule of ReP catalyst after the system had stopped producing CO and is a measure of stability of the Re catalyst in the system. ReP pic loadings of 0.05, 0.1, 0.2, 0.3, 0.4, 0.5 mmol per 5 mg TiO 2 were tested and the highest TON (48) was achieved with a loading of 0.1 mmol ReP pic on 5 mg TiO 2 under l > 420 nm irradiation (see the Supporting Information, Table S2), which is hereafter denoted as the optimised system. ReP Pic -TiO 2 gives a Re-based turnover frequency (TOF CO ) of 8 h À1 during the first 3 h irradiation and a TON CO of 48 after 1 day visible-light irradiation ( Figure 3, Table 2). Control experiments, under comparable conditions but in the absence of either ReP pic , TEOA, TiO 2 , CO 2 or light, did not generate significant amounts of CO (Table S2).  (Table S3). The choice of metal oxide particle is therefore critical for the activity of immobilised ReP. Measurements with different neutral density filters and therefore reduced light intensity showed little effect on catalytic rate for ReP pic -TiO 2 (0.1 mmol on 5 mg; see the Supporting Information, Figure S12), demonstrating that light harvesting was not performance-limiting under optimised conditions. Employing a 300, 400 or 420 nm cut-off filter resulted in comparable results (Table 2). Unexpectedly, a high TON CO of 39 and 16 was observed with ReP-TiO 2 with a l > 455 filter and a l > 495 nm filter, respectively (Table 2). To our knowledge, catalytic CO production by Re complexes with wavelengths as long as 495 nm has not been previously reported.
After 24 h of irradiating ReP pic -TiO 2 (0.5 mmol on 5 mg), the nanoparticles were separated by centrifugation from the reaction solution, dried under vacuum and investigated with ATR-IR spectroscopy. Two metal carbonyl signals were clearly detected at n = 2017 and 1885 cm À1 , indicating that a significant quantity of the Re catalyst remains intact on the TiO 2 particles ( Figure 2 A and Figures S13 and S14 in the Supporting Information). The FT-IR spectrum suggests that the [Re(bpy)(CO) 3 (L)] (L = labile ligand) symmetry is retained and vibrations display a substantial low frequency shift with respect to fresh ReP pic -TiO 2 , indicating that the Re centre has changed from cationic to neutral. This is due to coordination of an anionic species, probably deprotonated TEOA. The post-catalysis vibrations are comparable in wavenumber to those reported for a solution spectrum of the [Re(bpy)(CO) 3 (OCH 2 CH 2 N(CH 2 CH 2 OH) 2 )] complex (2006, 1897 and 1881 cm À1 ), but show only one lowwavenumber band. [20a] The CO stretches after catalysis are broader than beforehand, suggesting that they are probably compound peaks from multiple species. Proposed deactivation products for homogeneous Re systems include dimerisation during electrocatalysis [22] and formation of Re formate complexes. [23] It is likely that dimer formation would be less favourable due to immobilisation of the catalyst on TiO 2 .
The catalytic ReP Br -TiO 2 system was characterised after 2 h of photocatalysis using XPS to investigate the stability of the molecular Re catalyst on the TiO 2 surface (Figure 2 B, as well as Figure S15 and Table S1 in the Supporting Information). The percentage composition was Re 0.82 %, Br 0.42 %, P 1.99 % and N 4.19 %. The values for Re, Br and P are in a comparable ratio to the pre-catalysis measurements (see above), whereas the nitrogen content has increased. This observation can be explained by coordination of DMF and/or TEOA to the Re, as in known photocatalytic intermediates for comparable Re complexes. [20] In addition, the three hydroxy groups in TEOA can also bind to TiO 2 and surface bound TEOA would result in a higher post-catalysis N content. The N 1 s signal at 400.3 eV, present before and after catalysis, has a very similar binding energy to signals previously assigned to the N in a Ru complex of 4,4'-dicarboxylic acid-2,2'-bipyridine adsorbed on TiO 2 (400 eV). [21] The additional signal after catalysis at 402.2 eV is assigned to contributions from DMF and/or TEOA. The XPS results are consistent with FT-IR measurements, which indicate that a significant quantity of the molecular catalyst is present on the surface during catalysis.
Spectroscopic characterisation CO 2 photoreduction is significantly more efficient for ReP catalysts anchored onto TiO 2 compared with analogous conc-   in solution. ReP-TiO 2 is also more efficient than an optimised homogeneous [ReCl(bpy)(CO) 3 ] system (0.871 mm in DMF/TEOA) under simulated solar light. [11a] To understand the difference in efficiency between homogeneous and TiO 2 immobilised catalysis, the ReP catalysts were investigated on TiO 2 and in solution by time correlated-single photon counting (TC-SPC) and transient absorption spectroscopy (TAS). The catalytic mechanism of ReP for the reduction of CO 2 was found to be independent of the nature of the labile ligand (Br or picoline).
The irradiation of [Re(bpy)(CO) 3 (L)] complexes (ReP) with visible light causes an MLCT transition generating a triplet state 3 (ReP*) that emits at 600 nm with a characteristic broad signal (see the Supporting Information, Figure S16). ReP pic displays a blue-shifted emission signal with respect to ReP Br . This feature of ReP pic is characteristic of a cation, despite the zwitterionic feature of ReP pic ( Figure S16). [24] Both in solution and when immobilised onto TiO 2 , the luminescence of ReP decays within t 50 % = 20 ns. In good agreement with previous reports, [13,25] TEOA was observed to quench approximately 90 % of the emission within t 50 % = 10 ns due to the reductive electron transfer to ReP (see the Supporting Information, Figure S17). The luminescence intensity of ReP immobilised onto TiO 2 is very similar to that of an analogous Al 2 O 3 film functionalised with ReP (see the Supporting Information, Figure S18). Since Al 2 O 3 is a semiconductor with a conduction band that does not allow for electron injection (À4.2 V vs. NHE), [26] these results indicate that electron injection from the photocatalyst to TiO 2 does not take place.
Photoexcitation of homogeneous ReP pic and ReP pic immobilised on TiO 2 (l ex = 415 nm) in the presence of TEOA results in two characteristic transient absorption features; a peak centred at l max = 500 nm and a broad transient signal in the near-IR region of the spectrum (800-900 nm). The latter band shows a smaller signal amplitude relative to the 500 nm peak, with this difference being more pronounced for ReP pic -TiO 2 than for the homogeneous system (see the Supporting Information, Figure S19). These signals are not detected in the absence of an electron donor. As reported previously for analogous [Re(bpy)(CO) 3 (L)]-type catalysts, we have assigned the l max = 500 nm transient absorption peak to the reduced catalytic species ReP À , an important reaction intermediate in the CO 2 photoreduction formed upon the reductive quenching of ReP* by TEOA along with the loss of the labile ligand (L). [25] This reaction intermediate was detected independently of the nature of the labile ligand (picoline or Br). It has previously been suggested that the catalytic mechanism of Re-based complexes involves the replacement of these ligands by DMF or TEOA. [20] However, the transient absorption in the near-IR has been previously assigned to the formation of catalytically inactive Re-dimers. [27] This assignment implies that the lower signal amplitude for ReP pic -TiO 2 compared to homogeneous samples results from the formation of dimeric species, this being less favourable when the molecules are immobilised onto the surface of a metal oxide.
The kinetics of ReP pic in solution and anchored on TiO 2 were also monitored by transient absorption spectroscopy in the millisecond to second timescales ( Figure 4). The measurements were performed under either an N 2 or a CO 2 atmosphere, using TEOA (1 m in DMF) as sacrificial electron donor. In all cases, the decays were probed at 500 nm, corresponding to l max of the reduced catalytic intermediate ReP À . Under N 2 , the lifetime of ReP À is more than one order of magnitude longerlived when the catalyst is anchored onto TiO 2 than in solution (t 50 % > 1 s for ReP pic -TiO 2 and t 50 % = 60 ms for ReP pic in homogeneous solution). The addition of CO 2 in ReP pic -TiO 2 samples shortens the lifetime of the transient absorption decay assigned to ReP À to t 50 % = 400 ms. The decay of ReP À in solution, in the presence of CO 2 , has a strong biphasic behaviour, with a fast component in the 1-10 ms timescale and a slow phase in the 100 ms-1 s timescale, indicative of a multiple step process. The effect of the lifetime of the reaction intermediates on the CO 2 reduction photocatalysis is further discussed below.
The increase in lifetime of the reduced reaction intermediate observed when the catalyst is immobilised onto TiO 2 suggests that the scaffolding provided by the metal oxide enhances the stability of this intermediate. Thus, the longer-lived catalytically active ReP À species formed in ReP pic -TiO 2 samples have a greater probability of encountering and consequently reacting with CO 2 as well as of undergoing the second reduction necessary for the release of CO. This is consistent with our catalytic measurements showing a better performance of the ReP-TiO 2 system. Our results are also in good agreement with a recent publication showing improved CO 2 reduction yields in a TiO 2 system loaded with a phosphonated Re catalyst and a molecular dye. [17] However, in this previous study the Re catalyst acted solely as an electrocatalyst with the dye acting as photosensitiser, whereas in our work ReP was employed as a photocatalyst without requiring an additional dye.
The transient absorption signal of ReP pic -TiO 2 decays approximately one order of magnitude more rapidly when adding CO 2 , compared with the analogous system under N 2 . This faster decay can be most obviously assigned to the disappearance of ReP À species upon reaction with CO 2 in the 100 ms to 1 s timescale. Further evidence of the timescale of the interaction between ReP À and CO 2 is provided by comparing the kinetics of ReP pic in solution and when immobilised on TiO 2 . From the biphasic transient absorption decay under CO 2 of ReP pic in solution, the kinetics of the slow phase are very similar to those of ReP pic -TiO 2 (Figure 4 B, inset), suggesting that they involve the same process. Theoretical calculations supported the hypothesis that rhenium photocatalysts covalently attached to TiO 2 show an optimum orientation for maximum reduction capacity. [15l] Thus, the lack of steric hindrance when immobilising ReP onto TiO 2 suggests that the interaction of ReP À with CO 2 is expected to be comparable to a homogeneous system. We hypothesise that the fast phase of ReP pic in solution is due to the formation of a reaction intermediate involved in the deactivation pathway of CO 2 reduction, likely to correspond to the formation of the Re dimer. [22] The detailed nature of these catalytic and deactivation pathways are currently under investigation. We also note that the reduction of CO 2 to CO requires two electron transfers. It has been previously hypothesised that this second reduction can take place through the interaction of two singly reduced ReP À catalysts, or via the reaction of the carbon radical formed upon the deprotonation of [TEOA]C + species with ReP À . [13, 20b, 25] The kinetics of this second reduction, represented by dashed grey lines in Scheme 1, have not been observed in our transient absorption spectroscopy measurements. However, the higher CO 2 reduction yields observed when ReP is immobilised onto TiO 2 compared to homogeneous solution suggests that, in our system, the second electron transfer may occur through the sacrificial electron donor, as the immobilisation of ReP would hamper electron transfer between two Re molecules. The results of our kinetic studies are summarised in Scheme 1.

Conclusions
We have synthesised two novel phosphonated rhenium bipyridine complexes, which bind quantitatively to TiO 2 nanoparticles. The ReP-TiO 2 hybrids can photoreduce CO 2 to CO with high wavelength light (l > 495 nm), show good stability and show higher TONs than previously reported Re catalysts in the absence of dye as far as we are aware. The molecular structure is retained when the complex is immobilised onto the surface of the metal oxide before and during catalysis, although there are changes in the axial ligand, L. The unexpected enhancement in photocatalysis with ReP-TiO 2 is explained by the significantly longer lifetime of the reduced catalytic Re species on TiO 2 , an important intermediate in the photoreduction mechanism, formed upon reductive quenching by the sacrificial electron donor. The near-IR region of the transient absorption spectra provide evidence that a reduced amount of inactive Re dimers is being formed in the catalytic cycle, which may also contribute to the higher activity of the ReP-TiO 2 assembly. We have therefore shown that modification of the photochemical kinetics through photocatalyst immobilisation on metal oxide surfaces is a viable route to enhanced catalytic activity.

Physical measurements
1 H and 31 P NMR spectra were recorded either on a Bruker AVIII HD Smart Probe (400 MHz), a Bruker AVIII HD (400 MHz) or a Bruker ECS400 (400 MHz) spectrometer. 1 H NMR spectra were referenced to the solvent residual peaks as an internal reference, [29] and 31 P NMR spectra were referenced to an external standard (85 % H 3 PO 4 in D 2 O). UV/Vis spectra were recorded on a Varian Cary 50 UV/Vis spectrophotometer using quartz glass cuvettes. Electrospray ionisation mass spectra (ESI-MS) were recorded on a Quattro LC spectrometer, Bruker Esquire 6000 or a Bruker microTOF instrument (for high-resolution (HR) ESI-MS measurements) and the theoretical and experimental isotope distributions were compared. m/z values are quoted for 64 Zn, 185 Re and 79 Br. Elemental analysis was carried out by the microanalysis service of the Department of Chemistry, University of Cambridge. IR spectra were recorded on a Mattson RS FTIR instrument, averaging 64 scans at resolution 2 cm À1 or on a Thermo Scientific Nicolet iS50 FT-IR spectrometer. ATR spectra of the particles were recorded with an average of 256 scans. IR spectra of the KBr discs of the particles were taken with a PerkinElmer RX I instrument. ATR-IR spectra of particles before and after catalysis were run with 0.5 mmol ReP pic on 5 mg TiO 2 to obtain the strongest possible signals. X-ray photoelectron spectroscopy (XPS) measurements were carried out using an XR6 monochromated Al Ka X-ray source (hn = 1486.6 eV) with a 900 mm spot size in an ultrahigh vacuum chamber (ESCALAB 250Xi). The pass energy was 20 eV. ReP Br was chosen over ReP pic (see below) for XPS due to the additional measurable element (Br). The sample after catalysis was washed with DMF, and dried under vacuum, followed by exposure to the XPS operating vacuum. [Re(3-picoline)(tetraethyl 2,2'-bipyridine-4,4'-bisphosphonate) (CO) 3 ][PF 6 ] ( Et ReP pic ): A 250 mL two-neck round bottom flask was fitted with a reflux condenser and a gas valve and flame dried. AgPF 6 (430 mg, 1.70 mmol) was added under Ar followed by THF (60 mL) and 3-picoline (3 mL). Et ReP Br (600 mg, 771 mmol) was added and the solution refluxed until the reaction was completed after approximately 20 h. The reaction was monitored by IR spectroscopy (n CO ). The solution was filtered and the filtrate evaporated to dryness. The oil residue was dried under vacuum overnight and then redissolved in CH 2 Cl 2 . The product was precipitated with hexane, collected by filtration and washed exhaustively with hexane. The product was purified by column chromatography (SiO 2 with EtOAc and CH 2 Cl 2 /CH 3 OH (95:5) and Sephadex LH20 with THF and CH 3   [ReBr(2,2'-bipyridine-4,4'-bisphosphonic acid)(CO) 3 ] (ReP Br ):

Synthesis and Characterisation
Method 1: A Teflon capped high-pressure vessel was charged with [ReBr(CO) 5 ] (316 mg, 778 mmol), 2,2'-bipyridine-4,4'-bisphosphonic acid (246 mg, 778 mmol), toluene (20 mL) and methanol (3 mL), and the reaction mixture was refluxed overnight. The tube was allowed to cool and the solids collected by filtration. These were redissolved in methanol and diethyl ether added to precipitate a yellow solid. The suspension was filtered to yield an orange solution and the solvent was removed under reduced pressure to yield an orange solid. Single crystal X-ray diffraction studies Diffraction data were collected at 110 K on an Oxford Diffraction SuperNova diffractometer with MoK a radiation (l = 0.71073 ) using a EOS CCD camera. The crystal of Et ReP Br was cooled with an Oxford Instruments Cryojet. Diffractometer control, data collection, initial unit cell determination, frame integration and unit-cell refinement was carried out with "CrysAlisPro". [30] Face-indexed absorption corrections were applied using spherical harmonics, implemented in SCALE3 ABSPACK scaling algorithm. [31] OLEX2 [32] was used for overall structure solution, refinement and preparation of computer graphics and publication data. Within OLEX2, the algorithm used for structure solution was "charge-flipping" using the superflip program. [33] Refinement by full-matrix least-squares used the SHELXL-97 [34] algorithm within OLEX2. [32] All non-hydrogen atoms were refined anisotropically. Hydrogen atoms were placed using a "riding model" and included in the refinement at calculated positions. Crystal data and structure refinement details (see the Supporting Information, Table S4), selected bond distances and angles (Table S5) and details about the X-ray crystal structure refinement are given in the Supporting Information.

Assembly of colloidal ReP-TiO 2 hybrid system
ReP was pre-loaded onto the TiO 2 nanoparticles by stirring ReP with dispersed TiO 2 in H 2 O for one hour protected from light. The particles were separated by centrifugation and removal of the supernatant, followed by drying under high vacuum at room temperature overnight. In a typical procedure, a solution of ReP pic (1.36 mg) in H 2 O (10 mL) was added dropwise to a sonicated suspension of TiO 2 (100 mg) in H 2 O (20 mL) with stirring. The mixture was stirred for 1 h, centrifuged, the supernatant removed and the ReP pic -modified TiO 2 powder dried by high vacuum at room temperature overnight.

Photocatalytic studies
A reaction mixture of DMF/TEOA 5:1 (4.5 mL) and the catalyst were added to the light-protected photoreactor, which was subsequently sealed with a rubber septum and air was replaced by bubbling through a needle with CO 2 containing 2 % CH 4 (internal GC standard) before irradiation. Photocatalytic experiments were performed with a Solar Light Simulator (Newport Oriel, 100 mW cm À2 ) equipped with an air mass 1.5 global filter (AM 1.5G). The irradiation wavelengths were controlled using UV and blue-light cut offfilters (UQG optics) and IR irradiation was filtered by a water filter (path length 10 cm). The temperature of the photoreactor was kept constant at 25 8C in all experiments with a water circulator connected to a water-jacketed reservoir. The irradiated cross section of the solution in the vials was approximately 6.3 cm 2 . The photoreactors consisted of glass tubes with volumes between 14.2 and 14.5 mL with corresponding headspace volumes of 9.7 and 10 mL, respectively. Periodic headspace gas analysis (30 mL) allowed quantification of CO at regular time intervals with an Agilent 7800 Series gas chromatograph (GC) equipped with a HP-PLOT/Q column (0.53 mm diameter) attached to a HP-5 column (0.32 mm diameter). The GC oven temperature was kept constant at 45 8C, He was used as carrier gas at an approximate flow rate of 2 mL min À1 and a thermal conductivity detector (TCD) was used. TON CO assumes quantitative binding of ReP pic to the particle and is given as the average over a minimum of three runs. The errors are given as AE 1 standard deviation (s).

Spectroscopic measurements
Homogeneous samples for spectroscopic measurements were prepared by dissolving the photocatalyst (0.1 mm) in a water/DMF mixture (1:2). When required, TEOA was added to the solution to give a concentration of 1 m. Anatase TiO 2 films were employed for spectroscopic measurements, because their smaller particle size and decreased scattering effect offer an enhanced optical transparency compared to that of P25 analogues. Anatase TiO 2 films (8 mm thick, area 1 cm 1.5 cm) were prepared from a colloidal paste as reported previously and Al 2 O 3 was prepared as reported previously, [35] and deposited onto microscopic slides by Doctor blading. The films were functionalised by dipping the TiO 2 slides into a 0.3 mm aqueous solution of ReP during 48 h in the dark, to avoid degradation of the catalyst. The samples were then dried under vacuum for approximately 1 h to remove any traces of water. The samples were kept in the dark before and in between the measurements. The millisecond to second transient absorption measurements were performed using a home built system as reported previously. [36] The samples were excited with a 415 nm laser pulse (~300 mJ cm À2 , 0.5 Hz rep. rate, 6 ns pulse width) and, unless otherwise stated, transient absorption decays were monitored at 500 nm. The decays observed were the average between 500 and 1000 laser pulses. N 2 was used to purge the samples (> 15 min) before each measurement, unless otherwise stated. Time correlated-single photon counting measurements were performed using a Horiba Jobin Yvon TBX Fluorocube system. As excitation source, a pulsed laser with 404 nm nominal wavelength at a repetition rate of 100 kHz was used. The photoluminescence of a 0.1 mm ReP solution in N 2 -purged water, and a ReP-TiO 2 film in air was monitored at 600 nm for a fixed period of time (300 s). The instrument response was measured at the full width half maximum, showing typically a 200-250 ps value.