Design of an inherently-stable water oxidation catalyst

While molecular water-oxidation catalysts are remarkably rapid, oxidative and hydrolytic processes in water can convert their active transition metals to colloidal metal oxides or hydroxides that, while quite reactive, are insoluble or susceptible to precipitation. In response, we propose using oxidatively-inert ligands to harness the metal oxides themselves. This approach is demonstrated by covalently attaching entirely inorganic oxo-donor ligands (polyoxometalates) to 3-nm hematite cores, giving soluble anionic structures, highly resistant to aggregation, yet thermodynamically stable to oxidation and hydrolysis. Using orthoperiodate (at pH 8), and no added photosensitizers, the hematite-core complex catalyzes visible-light driven water oxidation for seven days (7600 turnovers) with no decrease in activity, far exceeding the documented lifetimes of molecular catalysts under turnover conditions in water. As such, a fundamental limitation of molecular complexes is entirely bypassed by using coordination chemistry to harness a transition-metal oxide as the reactive center of an inherently stable, homogeneous water-oxidation catalyst.

A current challenge in the development of molecular wateroxidation catalysts 1 is to overcome their inherent susceptibilities to oxidative or hydrolytic degradation under turnover conditions in water 2,3 . Advances in the past decade have led to remarkably fast rates of O 2 formation 4-6 , along with the parallel development of more stable catalysts 7 , better able to withstand strongly oxidizing conditions in water. Nevertheless, organic ligands are susceptible to oxidation 1,2,[8][9][10][11] , which, combined with hydrolysis of their active transition-metal centers, often results in the formation of colloidal metal oxides. Entirely inorganic polyoxometalate (POM)-based water-oxidation catalysts 12,13 , by contrast, are thermodynamically stable to oxidation [12][13][14][15] , can utilize earth-abundant metals 5,[14][15][16][17] , and feature remarkably rapid rates 5 . At the same time, molecular POM catalysts are stable to hydrolysis only under specific, albeit welldefined and by now, well-understood conditions, concerning variables such as pH and the nature and concentration of buffer [17][18][19] . Fundamentally, however, their catalytically active transition metals are in equilibrium with trace concentrations in the aqueous solvent 20 , which can lead to hydrolysis under nonoptimal conditions 18 . Hence, as advances in ligand design have led to impressively rapid rates and longer catalyst lifetimes under turnover conditions, catalyst stability nevertheless remains an ongoing topic of discussion and experiment 1,10,20 .
One reason for this heightened level of concern is that colloidal metal oxides can be extremely active [21][22][23] , such that considerable efforts are required to prove that the molecular catalyst, and not its products of decomposition in water, are the kinetically competent species in catalytic water oxidation. Given this situation, a compelling solution to catalyst stability might be to embrace the colloidal metal-oxide nanocrystals (NCs) themselves as thermodynamically stable water-oxidation catalysts. Apart from a few exceptions 23 , however, colloidal metal oxides are either insoluble at most pH values in water, or rapidly aggregate and precipitate from water under turnover conditions. And, despite impressive advances in solvothermal syntheses of metal-oxide NCs 24 , the requisite organic stabilizing ligands are not only susceptible to oxidation, but usually limit solubility in water and block access to the metal-oxide surface.
In this context, we recently discovered that heteropolytungstate cluster anions (POMs) could serve as covalently attached oxodonor ligands for 6-nm anatase-TiO 2 NCs 25 , giving water-soluble POM-complexed nanostructures. We now deploy Fe(III)-substituted [α-P V W VI 11 O 39 ] 7cluster anions as oxo-donor ligands for hematite (α-Fe 2 O 3 ) cores, giving water-soluble catalysts, 1, uniquely positioned between molecular iron-oxide clusters 26 and colloidal hematite 27,28 . The new catalyst features 3-nm hematite (α-Fe 2 O 3 ) cores comprised of ca. 300 Fe atoms, sufficiently large to possess the visible-light semiconductor properties of hematite. Unlike colloidal hematite, however, covalent coordination of each α-Fe 2 O 3 core by ca. 15 cluster anions renders the POMcomplexed structures stable to aggregation in water, giving optically transparent solutions at pH values of 2.5 to 8. Moreover, fitted with entirely inorganic tungsten(VI)-oxide ligands, and formed in water at 220°C, 1 is thermodynamically stable to both oxidative degradation and hydrolysis. As an inherently stable visible-light activated water-oxidation catalyst, 1 is capable of continuous operation for 7 days under turnover conditions (corresponding to 7600 turnovers) with no decrease in activity.

Results
Inorganic coordination complexes of hematite. Complex 1 is prepared by converting micron-sized particles of γ-FeO(OH) to α- remain bound via bridging oxo linkages to 3-nm hematite cores ( Fig. 1a and Supplementary Figures 1-7). The reaction gives an orange, optically transparent pH-7 solution of 1 (Fig. 1b, left inset). Dynamic light scattering (DLS) of the clear-orange solution reveals a number-weighted hydrodynamic radius of 1.9 nm (Fig. 1b). Cryogenic-TEM images 29 of the same (vitrified) solution (right inset to Fig. 1b) reveal numerous freely diffusing particles with an average size of 2.8-3.0 nm, too small for resolution of their surface structures in cryo-TEM images (additional images in Supplementary Figure 8). (This limitation is consistent with extensive studies of POMs on gold nanoparticles, in which even well-ordered POM monolayers are not discernible on gold cores smaller than ca. 5 nm 29 .) The cores of 1 are single NCs of hematite (α-Fe 2 O 3 ). Highresolution TEM images (Fig. 1c) reveal an inter-planar spacing of 2.69 Å, in agreement with the (104) crystal planes of hematite 30 . Indexing of the well-defined rings found in electron diffraction of dry samples precisely matches hematite ( Supplementary Figure 9a) 31 , and the related dark-field images reveal individual 2.8 ± 0.5 nm ( ± denotes s.d.) diameter NCs (Supplementary Figure 10). The hematite structure was further confirmed by powder X-ray diffraction (XRD) 30 , for which the Debye-Scherrer equation 32 gave a crystallite size of 3.5 ± 0.5 nm ( Supplementary  Figures 9b and 11).
Isolation and stability to aggregation. To isolate pure samples of 1, a clear-orange solution (inset to Fig. 1b) was made 2 M in NaCl. This gave a cloudy solution of salted-out 1, an orange-red solid isolated by centrifugation. Re-dissolution of 1 in pure water (10 mL) returned clear-orange solutions, even after multiple cycles of NaCl addition, centrifugation, and re-dissolution (Supplementary Figure 2). This solubility and remarkable stability to  aggregation differs dramatically from the properties of electrostatically stabilized colloids. Behaving more like a molecular macroanion, the hydrated Na + salt of 1 can be stored indefinitely, and then readily dissolved in water. These findings provided the first indication that, as shown in Fig. 1a, heteropolytungstate cluster anions are covalently bound to the hematite cores.
Cluster-anion ligands coordinated to hematite cores of 1. Energy-dispersive X-ray spectroscopic (EDX) analysis of 1 ( Differential pulse voltammetry (DPV) 25 performed after adding LiClO 4 (0.1 M) to aqueous solutions of 1 (Fig. 2c), provided further support for the presence of [α-PW 11 O 39 Fe] anions at the hematite surface. While conceptually analogous to electrochemical studies of ferrocene-covered SiO 2 nanoparticles 33 , to our knowledge, the observation of reversible redox chemistries of ligands bound to metal-oxide NCs in solution has few precedents in the literature 25 . It was made possible here by the abundance of cluster anions on the hematite surface, in combination with the water solubility imparted by their negative charges and numerous alkali-metal counter-cations. After adding LiClO 4 (0.1 M), differential pulse voltammetry (DPV) revealed reversible redox processes consistent with the presence of the cluster anions identified by ESI-MS (DPV data for the independently prepared molecular cluster anions are provided in Supplementary Figure 15b).
Surface coverage by the POM anions, based on a reasonable footprint of 1.9 nm 2 , places 15 POMs on the surface of an (idealized) spherical 3-nm-diameter hematite core (Supplementary Table 1). Such a structure would contain 165 W atoms and 279 Fe atoms, giving a relative atomic composition of 37% W and 63% Fe, nicely matching the %-atom values observed by EDX (Fig. 2a). The W to Fe ratio also gives ca. 18-20 Fe atoms per POM ligand.
Not only is 1 remarkably stable to aggregation, it is soluble at pH values of 2.5-8. Over this wide pH range, its largely negative zeta potential values (ζ, −35 to −40 mV) remain pH-invariant (Fig. 2d). By contrast, colloidal α-Fe 2 O 3 precipitates at its isoelectric point (pH 5.5), at which ζ = 0 mV 34 . These findings represent three independent lines of evidence for strong coordination of the POM ligands to the α-Fe 2 O 3 cores. Given these inert linkages, metathesis of Na + counter-cations (of the POMs) by n-R 4 N + (R = hexyl or octyl) was used to render 1 organic-solvent soluble. This cation-exchange proceduretypical of POM salts-gave optically transparent MeCN or MeOH solutions with no loss of POM ligands (Supplementary Figure 16), further evidence for their covalent attachment to the hematite cores.
It is extraordinarily difficult to precisely determine the atomic connectivities of molecules bound to colloidal NCs. However, the presence of a μ 2 -oxo linkage between Fe(III) ions in the known oxo-  Figure 17). Notably, these W4f peaks are comparable with those of 2 ( Supplementary Figures 18 and 19), consistent with the presence of Keggin-anion derived ligands connected to the surface of 1 via μ 2 -oxo linkages.
This was investigated in more detail by FTIR spectroscopy (Fig. 3). The IR-allowed vibrational modes of the central PO 4 moieties within Keggin-derived structures are highly sensitive to changes in symmetry of the cluster anion. Removal of a single W (VI) ion from the plenary-Keggin anion, [α-PW 12 O 40 ] 3-, reduces the local symmetry of the central PO 4 moiety from T d to C 3v . As a result, rather than a single IR-active mode at 1080 cm -1 , two IRactive PO 4 modes are observed for mono-lacunary [α-PW 11 O 39 ] 7-. And, after complexing a transition-metal cation, the difference in energy between the two IR-active modes depends on cation size, and the degree to which it fits within the lacunary anion's pentacoordinate binding site 36 . As the ion is pulled farther out from the binding pocket, the local symmetry around the central PO 4 moiety deviates from Td, leading to a larger difference in energy between the two IR-active PO 4 bands. Hence, splitting of the PO 4 mode can be used to identify the ligand environments of substituted transition-metal cations.
The precise match between the PO 4 bands 36 of the POM ligands on 1 (1090 and 1051 cm −1 ; Fig. 3a) with the corresponding PO 4 bands of 2 (1090 and 1050 cm −1 ; Fig. 3b), is consistent with μ 2 -oxo linkages between the POM ligands and the hematite cores. Combined with the EDX data in Fig. 2a In the present work, the covalent attachment of POM ligands to the hematite cores of 1 gives substitutionally inert structures that occupy a unique position at the interface between molecular iron-oxide clusters and electrostatically stabilized colloidal ironoxide NCs. We now show that, when used as a soluble photocatalyst for visible-light driven water oxidation, 1 is inherently stable under turnover conditions in water.
Catalytic water oxidation. The activity of 1 as a visible-light water-oxidation catalyst was explored by irradiating pH-8 solutions of 1 with visible light (150 W Xe lamp, cutoff λ ≥ 420 nm, Supplementary Figures 21 and 22). Four oxidants were evaluated: Ce(IV), Ag + , S 2 O 8 2-, and periodiate (I VII O 4 -). Ce(IV) was not suitable as it is only stable at pH values below 1, at which hematite itself dissolves. Although Ag + is soluble at neutral pH values, these large cations form strong ion pairs with the negatively charged POM ligands, leading to precipitation at desired Ag Although written as IO 4 -, a recent report shows 43 that its dominant form in water is actually orthoperiodate, H 5 IO 6 , which behaves as a polyprotic acid, with pKa 1 , pKa 2 , and pKa 3 values of ca. 1, 7.5, and 11, respectively. For simplicity, entries in Table 1 refer to added periodate (IO 4 -).
The greater reactivity of IO 4relative to S 2 O 8 2- (Table 1) is not unique to visible-light driven reactions of 1. Notably, IO 4is also more effective than S 2 O 8 2in trapping photoexcited electrons from visible-light irradiated WO 3 Table 2). At the same time, when using IO 4 -11 , decomposition or oxo-transfer reactions 46 , that could generate O 2 without removing four electrons from water, must be ruled out 47,48 . This required numerous control experiments (see Supplementary Figure 25 and Table 3), necessitated by the fact that the oxide ligands of periodate equilibrate rapidly with water, precluding the use of 18   Rate optimization and stability under turnover conditions. Unlike most polyoxometalate-based water-oxidation catalysts, 1 is stable in water over a wide range of pH values, from 2.5 to 8. And, in contrast to colloidal hematite that precipitates from solution at its isoelectric point (i.e., at pH 5.5), 1 is soluble over this entire pH range (see Fig. 2d). Hence, unlike most purely molecular or traditional colloidal catalysts, the activity of 1 can be investigated over a wide range of pH values. The results (Fig. 4a) reveal an approximate doubling of rate from pH values of 5 to 8. At pH 8, the rate of O 2 formation is 800 ± 50 µmol g −1 h −1 (at λ ≥ 420 nm using a 150 W Xe lamp), somewhat exceeding the fastest reported rates for visible-light driven water oxidation by colloidal α-Fe 2 O 3 (see Supplementary Table 4 for reported values and related light sources).

Fe-OH 2 H H
The change in pH from 5 to 8 spans the isoelectric point of hematite (near pH 5.5; Fig. 2d). As such, the increase in rate may correlate with deprotonation of water molecules bound to the hematite surface, to give more reactive (negatively charged) hydroxide ligands, in combination with the more favorable, pHdependent Gibbs free energy for water oxidation itself. For periodate to trap photoexcited electrons, it must diffuse to within a close proximity to the α-Fe 2 O 3 surface. The cluster-anion ligated surface of 1, including [H 3 I VII O 6 ] 2-(orthoperiodate-the dominant periodate species present at pH 8-drawn to scale), and Na + counter-cations (without their hydration shells), is illustrated in Fig. 4b The rate of visible-light driven water oxidation by 1 is enhanced by the relatively small size of the hematite cores, which is critically important 28 due to the short, 2-4 nm, hole-diffusion length 55 of α-Fe 2 O 3 . This leads to a quantum yield of 3.9%, closely matching values reported for optimized reactions of similarly sized colloidal hematite 28,56 .
A preliminary schematic of reasonable mechanistic steps is provided in Fig. 4c. Trapping of an excited electron by [H 3 I VII O 6 ] 2-(the dominant form of periodate at pH 8) 43 would result in oxidation of a surface Fe III -OH moiety to Fe IV =O (A and B, respectively, in Fig. 4c), a species recently  the solutions remained optically transparent. After each 8-h reaction, the catalyst was quantitatively isolated by salting out with NaCl (2 M) and centrifugation, followed by re-dissolution in fresh water (Supplementary Figure 28a). Notably, this had no effect on its activity. The same remarkable retention of activity was observed in a single 7-day (1-week) reaction (7600 turnovers; Fig. 4d). No decrease in activity was observed. The TOF remained constant at 44 ± 1 h -1 , the solution remained optically transparent (inset to Fig. 4d), and no aggregation was observed by DLS (Supplementary Figure 29).
The TON of 7600 is much larger than that reported for colloidal hematite. For example, under identical conditions, the initial rate of O 2 formation by 5-nm colloidal α-Fe 2 O 3 (no POM ligands) 56 was ca. 320 µmol g −1 h −1 , but decreased dramatically within a few hours due to aggregation processes typical of colloidal metal oxides in water (Supplementary Figure 28b). Moreover, the present TON is more than seven times that reported for the most stable, water-soluble Fe-based catalysts 11 which, although much faster than 1 (with respect to TOF), become inactive within a few hours under turnover conditions in water. And, while acknowledging that 1 contains ca. 300 Fe atoms, many more than typically found in traditional molecular catalysts, its relatively small (ca. 20 Å) α-Fe 2 O 3 center is   nevertheless large enough to retain the photochemical properties of bulk hematite. It is from this perspective that 1 can be viewed as a soluble complex of a reactive metal-oxide core.

Discussion
Oxidatively inert heteropolytungstate cluster anions serve as covalently coordinated oxo-donor ligands for 3-nm hematite (α-Fe 2 O 3 ) cores, giving anionic complexes soluble in water over a wide range of pH values (from 2.5 to 8), and that catalyze visiblelight driven water oxidation with no need for added photosensitizers. Like molecular macroanions, 1 is highly resistant to aggregation processes that typically lead to the precipitation of electrostatically stabilized hematite and other colloidal metal oxides. And, formed by reaction with entirely inorganic tungstenoxide-based ligands at 220°C in water, 1 is inherently (thermodynamically) stable to the oxidative and hydrolytic processes that can limit the active lifetimes of molecular water-oxidation catalysts. As such, 1 can continuously catalyze water oxidation for 7 days (1 week) with no detectable decrease in activity. Moreover, the method used to prepare 1 is not limited to iron oxide, but can be modified according to the pH-controlled aqueous speciation chemistries of numerous other transition-metal ions. As such, the covalent coordination of oxidatively inert polyoxometalate ligands to metal-oxide nanocrystal cores represents a conceptually new and general approach to the design of inherently stable water-oxidation catalysts. Visible-light driven water oxidation. Photochemical O 2 production was carried out in a gas-tight quartz cuvette connected to an upper glass bulb with a headspace volume of 16 mL. Carefully weighed samples of 1 were dissolved in 2.7 mL of water along with sodium periodate (NaIO 4 ; 12.8 mg in 0.1 mL of water) to give a final NaIO 4 concentration of 20 mM. The addition of NaIO 4 led to a drop in pH from 6.5 to 5, and either 0.2 N KOH or 0.2 N HCl solutions were used to adjust the pH to desired values (see Fig. 4a). The headspace within the cuvette was evacuated and refilled with pure Ar (g) four times to remove most of the air, after which the solution was purged with Ar (g) for 30 min to remove dissolved oxygen. The solution was then irradiated with a 150 W Xe lamp (USHIO Inc. Japan), after inserting a 420-nm long-pass filter between the light source and the cuvette, which was held at a distance of 12 cm from the light source. The illumination area was 2 cm 2 and the average light intensity was 565 mW cm −2 at λ = 450 nm measured using a 1928-C Optical Power Energy Meter (Newport Corp.) equipped with a model 919P-250-35 Thermopile sensor (UV-calibrated silicon detector). The headspace gas (0.5-mL volume portions) was injected every 2 hours into a gas chromatograph (Focus GC, Thermo Scientific) operating at isothermal conditions (40°C) using a ShinCarbon ST micropacked column (0.53-mm diameter, 2-m length), and equipped with a thermal conductivity detector (TCD) and Ar as carrier gas. Moles of O 2 per gram reported for 1 were calculated based on the mass of its α-Fe 2 O 3 cores, which comprised ca. 50% of the total mass of 1.
Water oxidation for 7 days under turnover conditions. A solution of 1 (5.4 μM) and NaIO 4 (20 mM) in 3 mL of water at pH 8 (adjusted using 0.2 N KOH) was degassed as described above and irradiated with visible light (λ > 420 nm) for 1 day, during which, amounts of O 2 in the headspace were quantified at regular intervals (see Fig. 4d). After 1 day, 6.4 mg of solid NaIO 4 dissolved in 0.1 mL of water was added to the solution (an additional 10 mM concentration of IO 4 -) and the reaction continued for a second day. After day 2, 1 was quantitatively separated from accumulated iodate (IO 3 -; ca. 5.5 mM) by making the solution 0.5 M in NaCl, and isolating the salted-out catalyst by centrifugation. (A control experiment later carried out after recharging the supernatant solution with periodate, followed by irradiation, showed no activity (i.e., no O 2 in 8 h). This demonstrated that possibly unidentified soluble components were not responsible for the catalysis.) The pellet of 1 isolated after day 2 was then dissolved in 3 mL of water containing freshly added periodate (20 mM), adjusted to pH 8, degassed, and irradiated for day 3.
(Similar separations of 1 from accumulated IO 3were repeated after days 4 and 6.) After days 3 and 5, additional 10 mM concentrations of IO 4were added. Turnover numbers were calculated based on the moles of O 2 produced per mole of 1.

Data availability
All data generated or analyzed during this study are included in this published article (and its supplementary information files).