Sustained Solar H2 Evolution from a Thiazolo[5,4-d]thiazole-Bridged Covalent Organic Framework and Nickel-Thiolate Cluster in Water.

Solar hydrogen (H2) evolution from water utilizing covalent organic frameworks (COFs) as heterogeneous photosensitizers has gathered significant momentum by virtue of the COFs' predictive structural design, long-range ordering, tunable porosity, and excellent light-harvesting ability. However, most photocatalytic systems involve rare and expensive platinum as the co-catalyst for water reduction, which appears to be the bottleneck in the development of economical and environmentally benign solar H2 production systems. Herein, we report a simple, efficient, and low-cost all-in-one photocatalytic H2 evolution system composed of a thiazolo[5,4-d]thiazole-linked COF (TpDTz) as the photoabsorber and an earth-abundant, noble-metal-free nickel-thiolate hexameric cluster co-catalyst assembled in situ in water, together with triethanolamine (TEoA) as the sacrificial electron donor. The high crystallinity, porosity, photochemical stability, and light absorption ability of the TpDTz COF enables excellent long-term H2 production over 70 h with a maximum rate of 941 μmol h-1 g-1, turnover number TONNi > 103, and total projected TONNi > 443 until complete catalyst depletion. The high H2 evolution rate and TON, coupled with long-term photocatalytic operation of this hybrid system in water, surpass those of many previously known organic dyes, carbon nitride, and COF-sensitized photocatalytic H2O reduction systems. Furthermore, we gather unique insights into the reaction mechanism, enabled by a specifically designed continuous-flow system for non-invasive, direct H2 production rate monitoring, providing higher accuracy in quantification compared to the existing batch measurement methods. Overall, the results presented here open the door toward the rational design of robust and efficient earth-abundant COF-molecular co-catalyst hybrid systems for sustainable solar H2 production in water.

Introduction

The conversion and storage of solar energy in the form of chemical bonds in “solar fuels” like H2 through light-driven water reduction has evolved into a key technology over the past decade due to the fast depletion of fossil energy sources and rapid global climate change.[1] To drive the proton reduction half reaction in an efficient way, the major challenge is to find a catalytic system that is robust and highly active, but at the same time low-cost and earth-abundant, in combination with a strongly absorbing, chemically stable photosensitizer (PS).[2] In this regard, covalent organic frameworks (COFs)[3] have recently emerged as an exciting class of photoactive materials for light-driven H2 production due to their tunable light-harvesting[4] and charge-transport properties.[5] In contrast to other porous materials, COFs are known for being mechanically robust and offering large accessible surface areas. By virtue of their modular geometric and electronic structures, COFs have attracted significant interest for a range of applications including adsorption, storage and separation,[6] chemical sensing,[7] electronics,[8] and catalysis.[9] In spite of their versatility, there are only few reports on COFs utilized as photoabsorbers for photocatalytic H2 evolution so far.[10] Although it is rare and expensive, all except one of these works has employed metallic platinum as the co-catalyst to reduce water efficiently, which appears to be the bottleneck in the development of scalable, economical solar H2 production. In addition, the use of nanoparticulate Pt co-catalysts precludes obtaining detailed insights into the nature of the catalytic sites and the intricacies of the photocatalytic cycle. Inspired by natural photosynthesis,[11] researchers worldwide are motivated by this shortcoming to search for single-site, earth-abundant, non-precious metal-based co-catalysts with well-defined catalytic centers. So far, only one molecular co-catalyst–COF system for photocatalytic H2 evolution has been demonstrated.[12] This system is based on an azine-linked COF (N2-COF)[10b] acting as the PS and a cobaloxime molecular proton reduction catalyst, which shows a H2 evolution rate of 782 μmol h–1 g–1 and a turnover number TONCo = 54.4. However, the limited photostability and especially the utilization of an organic solvent (acetonitrile/water mixture; 4:1) were major concerns.[12]

Notably, a majority of molecular catalysts decompose during prolonged catalysis, are inherently insoluble in water, and require the addition of organic solvents to accomplish water reduction.[13] With cobaloxime-based systems, for example, the catalyst often converts to an inactive form within a few hours (<6 h) of H2 evolution, possibly due to ligand decomposition or hydrogenation.[14]

To overcome these issues, the development of a scalable, earth-abundant, and low-cost co-catalyst system which is soluble in water and can be coupled efficiently to a molecularly defined heterogeneous photoabsorber is in high demand. In this regard, Ni-based synthetic photocatalytic H2 evolution catalysts[15] have attracted significant interest because of their robust and oxygen-tolerant nature and, importantly, their structural similarity to the active site in [Ni-Fe] hydrogenase.[16]

Likewise, small molecules and polymers conn class="Chemical">taining fused (bi)heterocyclic thiazolo[5,4-d]thiazole (TzTz) moieties have received much attention as semiconductors in organic electronics lately because of their n-type character featuring high oxidative stability and their rigid planar structure.[17a] The latter enables efficient intermolecular π–π overlap that affords high electron and hole mobility.[17] Such TzTz moieties further feature excellent photoabsorbing ability, which is likewise beneficial for photocatalysis.[2a] Nevertheless, TzTz-based COFs have not been explored so far. Notably, thus far, only a very limited number of COFs bearing photoactive functionalities such as triazine,[10a,10b] diacetylene,[10c] or sulfone moieties[10d] have been shown to produce H2 from water, with the noble metal Pt acting as co-catalyst.

Combining these aforementioned leverages, in this work, we present a light-driven hybrid proton reduction system employing a newly designed TzTz-linked COF (TpDTz) as a photoabsorber and a molecular Ni-thiolate cluster (NiME)[18] assembled in situ from a Ni(II) salt and 2-mercaptoethanol (ME). The combination of the NiME cluster co-catalyst and TpDTz COF enables sustained H2 evolution with an excellent rate (941 μmol h–1 g–1) and a TONNi > 103 (70 h) in the presence of triethanolamine (TEoA) as the sacrificial electron donor (SED) in water under AM 1.5 light illumination. We thus report a single-site heterogeneous COF-based photocatalyst system that operates with a noble-metal-free co-catalyst in water as the solvent. We further carve out structure–property–activity relationships by comprehensively screening the parameter space of this heterogeneous–homogeneous hybrid photocatalytic system, including pH, SED, co-catalyst metal centers, different N/S-containing chelating ligands for co-catalysts, and a variety of PSs. Also, our study is built on a continuous-flow photocatalytic reactor system which enables a non-invasive and direct monitoring of the H2 evolution rate with high accuracy, in contrast to the routinely used standard photocatalytic batch reactors, and this allows gathering unique insights into the photocatalytic reaction modeling and kinetics. The results and understanding presented here thus contribute toward the rational development of robust and efficient single-site hybrid photocatalytic systems as a sustainable solution for solar H2 production in water.

Results and Discussion

COF Synthesis and Characterization

The precursor 4,4′-(thiazolo[5,4-d]thiazole-2,5-diyl)dianiline (n class="Chemical">DTz) was synthesized as described in the Supporting Information and characterized using single-crystal X-ray diffraction, nuclear magnetic resonance (NMR) spectroscopy, Fourier transform infrared (FTIR) spectroscopy, and mass spectrometry. TpDTz COF was synthesized by solvothermally reacting 1,3,5−triformylphloroglucinol; Tp (1.0 equiv) and DTz (1.5 equiv) in the presence of 6 M aqueous acetic acid using an o-dichlorobenzene and N,N-dimethylacetamide solvent combination in a high-precision glass vial, which was sealed and heated to 120 °C for 3 days (Figure and Supporting Information section S2). Following a similar protocol, TpDTP COF with a similar pore size was synthesized as a reference,[19] with the DTz linker replaced with the linear terphenyl linker.

Figure 1

Synthesis and structural characterization of TpDTz COF. (a) Schematic representation of TpDTz COF synthesis. (b) Space-filling model of TpDTz COF pores with π–π stacking of successive 2D layers (gray, C; blue, N; red, O; yellow, S; and white, H). (c) Indexed PXRD patterns of TpDTz COF with corresponding Pawley refinement (red) showing good fit to the experimental data (blue) with minimal differences (cyan); the inset shows close-up of the indexed experimental (blue) and simulated (black) PXRD patterns based on Pawley fits [final Rwp = 2.59% and Rp = 1.89%].

To verify crystallinity and phase purity, the as-synthesized TpDTz COF was analyzed via powder X-ray diffraction (PXRD). The PXRD pattern exhibits an intense first peak at 2.56° 2θ corresponding to the 100 reflection along with other diffraction peaks at 4.41, 5.23, 6.90, and 9.10° 2θ, attributed to the 110, 200, 210, and 220 reflections, respectively. In addition, at ∼26° 2θ a broad set of reflections is visible, with 00l being the most intense, which corresponds to the π–π stacking of the 2D layers (Figure c). The experimental PXRD pattern is in good agreement with the simulated AA eclipsed stacking model (Figure S6.2). The lattice parameters of TpDTz COF were extracted by Pawley refinement in the hexagonal space group P6/m (a = b = 39.27 Å, c = 3.46 Å, α = β = 90°, and γ = 120°) (Figure c). The relatively high level of order observed with PXRD may originate from effective π–π stacking interactions facilitated by the planarity of the DTz linker and, thus, the 2D layers. The measured pore aperture is ∼3.4 nm, and the π–π stacking distance between individual layers is ∼3.5 Å for TpDTz COF, as obtained from the structural model.

The FTIR spectrum of the as-synthesized TpDTz COF shows bands at ∼1254 cm–1 (−C–N), ∼1571 cm–1 (C=C), and ∼1618 cm–1 (C=O) (Figure S9), which confirms the formation of the proposed β-keton class="Chemical">enamine-linked framework. The TzTz moiety was identified by appearance of C=N vibrations (∼1660 cm–1) and C–S stretching bands between 650 and 700 cm–1. The structural composition of TpDTz COF was further confirmed by 13C cross-polarization magic-angle spinning (CP-MAS) NMR spectroscopy (Figure a). The spectrum shows signals corresponding to the heterocyclic TzTz ring of the DTz building unit (δ ≈ 151 ppm), together with a characteristic signal of the carbonyl carbon (C=O) at ∼184 ppm, which further supports formation of the β-ketoenamine moiety. 15N NMR spectroscopy confirms the presence of two different kinds of nitrogen atoms with chemical shifts of −93 and −243 ppm, corresponding to the TzTz and enamine (=C–NH−) moieties, respectively (Figure a). All assignments are supported by quantum-chemical calculations of NMR chemical shifts (Tables S4 and S5) at the B97-2/pcS-2//PBE0-D3/def2-TZVP level using the FermiONs++ program package[20a,20b] based on a selected molecular model system (Figure S49). The corresponding structures were optimized at the PBE0-D3/def2-TZVP level using Turbomole (version 7.0.3).[20c,20d] Scanning electron microscopy (SEM) images of TpDTz COF reveal a flower-like morphology composed of flakes with 1–3 μm lateral dimensions (Figure S11). Transmission electron microscopy (TEM) images confirm the layered morphology of the crystalline network with clearly visible 2D honeycomb-type pores oriented perpendicular to the crystallographic c axis with a periodicity of ∼3.3 nm (Figure c). In order to evaluate the thermal stability of TpDTz COF, we further performed thermogravimetric analysis (TGA) in air. The TGA profile suggests that the COF pores are guest free and the material is thermally stable up to ∼400 °C (Figure S10).

Figure 2

Structural characterization of TpDTz COF. (a)13C and 15N CP-MAS solid-state NMR spectra of TpDTz COF. Calculated NMR chemical shifts for the TpDTz-NMR model (Figure S49) obtained at the B97-2/pcS-2//PBE0-D3/def2-TZVP level of theory (Tables S4 and S5) are shown as gray dashes. (b) Argon adsorption–desorption isotherm for TpDTz COF recorded at 87 K; inset shows calculated pore size distribution of TpDTz COF according to the QSDFT method. (c) TEM image of TpDTz COF showing the hexagonal pore structure with a periodicity of ∼3.3 nm (scale bar, 100 nm). (d) UV–vis diffuse reflectance (DR) spectrum for TpDTz COF measured in the solid state; insets show a plot of the Kubelka–Munk function to extract the direct optical band gap and a photograph of TpDTz COF powder. (e) Cyclic voltammogram (CV) of a TpDTz COF-modified FTO working electrode in 0.1 M NBu4PF6 as the supporting electrolyte in anhydrous acetonitrile at a scan rate of 100 mV/s.

The permanent porosity of TpDTz COF was assessed by Ar adsorption analysis measured at 87 K (Figure b and Figure S13). A Brunauer–Emmett–Teller (BET) surface area of 1356 m2 g–1 was obtained for TpDTz COF, which is comparable to some of the most porous β-ketoenamine-based porous COFs previously synthesized via solvothermal methods.[10c,10d,21] The experimental pore size of 3.4 nm obtained from the adsorption isotherm using the quenched solid state density functional theory (QSDFT) cylindrical-slit adsorption kernel for carbon (inset of Figure b) is in excellent agreement with the pore size obtained from the structure model (∼3.4 nm) and TEM (∼3.3 nm). Further, the measured water adsorption isotherm (total uptake 309 cm3 g–1, 25 wt% at STP) of TpDTz COF suggests its relatively hydrophilic nature, induced by the polar N/S containing TzTz group and should thus lead to higher dispersibility of the COF in water during photocatalysis,[10d,10e] as opposed to the non-TzTz TpDTP COF (total uptake 75 cm3 g–1, 6 wt% at STP) with similar pore sizes (Figure S15). This fact is also supported by the higher CO2 uptake for TpDTz COF compared to TpDTP COF (Figure S16).

Since chemical stability is a crucial criterion for any material to be considered for practical applications, we investigated the chemical stability of TpDTz COF under strongly acidic (12 M HCl) conditions and in boiling water up to 7 days. The retention of all characteristic peaks in the PXRD pattern suggests a high chemical stability under the tested conditions (Figure S7). It is important to note that TpDTz COF is stable only under mild basic conditions (1 M KOH) for up to 3 days, while at harsher basic conditions (12 M KOH for 7 days) the framework decomposes. The high chemical tolerance of TpDTz COF is ascribed to the combined effect of the stabilizing enol-to-keto tautomerism[21] and the planarity of the TzTz moiety,[17] which allows for strong π–π interactions between the layers.

Optoelectronic Properties and Photocatalysis

The UV–vis diffuse reflectance (DR) spectrum of n class="Chemical">TpDTz COF reveals efficient light absorption extending into the orange parts of the visible spectrum with an absorption edge at ∼598 nm (Figure d). Kubelka–Munk analysis yields a direct optical band gap of ∼2.07 eV. In contrast, TpDTP COF shows a blue-shifted absorption band edge at ∼531 nm, corresponding to a larger optical band gap of ∼2.28 eV (Figure S18), due to the absence of light harvesting TzTz units. The measured photoluminescence (PL) spectra (Figure S19) reflect this trend; TpDTz COF has a significantly red-shifted emission (λmax = 690 nm) compared to TpDTP COF (λmax = 630 nm). The fluorescence decays can be fitted with triexponential functions, and the amplitude-weighted average lifetimes for TpDTz and TpDTP COFs are 94 and 115 ps, respectively (Figure S20). Emission intensities were too low to measure accurate absolute emission quantum yields. The short excited-state lifetimes together with low emission quantum yields suggest more pronounced non-radiative rates in the COF systems, relative to the radiative rates, similar to our previously reported N3-COF system.[10b,10f]

Cyclic voltammograms (CV) of TpDTz COF films were measured to estimate the band positions and the thermodynamic driving force for H2 evolution. The voltammogram of a TpDTz COF-modified FTO working electrode[22] shows an irreversible reduction wave with an onset potential of Ered,onset ≈ −1.24 V vs saturated calomel electrode (SCE) (Figure e and Figure S22). From the optical bandgap (Eg = 2.07 eV) determined from the UV–vis DR spectrum, the valence band (VB) and conduction band (CB) edges of TpDTz COF can be estimated to be ECB = −3.46 eV and EVB = −5.53 eV vs the vacuum level, following the equations ECB = −(Ered,onset vs SCE + 4.7) eV and EVB = ECB – Eg,opt.[23] Quantum-chemical calculations of vertical ionization potentials and electron affinities on a TpDTz pore model (Figure S53), cut from a supercell built using the 2D periodic optimized unit cell of the TpDTz COF (Figure S52), support these findings (Table S8). Based on a comparison of these values with the oxidation potential of TEoA (0.57 V vs SCE)[23d] and the reduction onset potential of the NiME molecular catalyst system (−0.75 V vs SCE) (Figure S21), it is likely that TpDTz COF can transfer electrons to the NiME co-catalyst system, forming a reduced Ni(I) center and thereby enabling H2 evolution in successive steps.[18,23a] Also, TEoA can efficiently quench the photoexcited holes in the COF thereby replenishing its photoactivity.

Owing to the planar and conjugated structure, the electron-deficient nature of the heterocyclic backbone, and the optimal band gap and hence light absorption ability, the TzTz-linked TpDTz COF was investigated as the heterogeneous photoabsorber for photocatalytic H2 evolution in combination with the Ni-thiolate hexameric cluster (NiME) co-catalyst in water. The NiME cluster co-catalyst has a cyclic hexameric structure composed of six Ni(II) ions forming a planar ring, and the Ni centers are bridged by 12 ME units, which has been confirmed by DFT calculation and single-crystal X-ray diffraction analysis by others (Figure S35).[18,24b] This NiME cluster co-catalyst has been shown to produce H2 actively when sensitized by an organic xanthene dye (erythrosin B).[18] However, the generation of unstable PS radical species upon photochemical quenching of the excited-state dye (PS*) leads to a fast decomposition and hence poor photochemical stability of organic dye PSs, which hinders the long-term performance of the photocatalytic system.[15a,18] The strategy of combining a photochemically stable COF photoabsorber with the Ni-thiolate cluster co-catalyst in water could thus be a viable path to impart better long-term stability for H2 production.

The beauty of the aforesaid NiME complex lies in its simple, quick in situ synthesis in water upon addition of a Ni(II) salt and ME at room temperature. This in situ assembly strategy is different from those of most other Ni(II) and Co(II) co-catalyst complexes, featuring arduous ex situ synthesis and purification of a water-soluble analogue, thus adding to the cost-effectiveness of the NiME cluster co-catalyst approach.[12,15]

In addition, this cluster has been shown to be a potent H2 evolution co-catalyst producing H2 immediately after light illumination in the presence of a PS and SED, and hence does not require any photodeposition, nor does it show an activation time, contrary to Pt-based photocatalytic systems.[10]

For better measurement accuracy and to gain insights into the photocatalytic mechanism, we developed a continuous-flow system to monitor the H2 evolution performance of the hybrid photocatalytic system (Figure and Figure S23). In this measurement system, the molar flow entering the system, the pressure (0.5 bar), and the temperature (25 °C) of the reactor are continuously controlled, while bypassing some of the out-flow to an open sampling loop gas chromatograph (GC) autosampler. In this way, the rate of H2 production (RH) can be monitored directly using only two experimental inputs: Fin from the mass flow controller, and xH from an online GC (BID) detection system (eq ), where Fin is the carrier gas (in this case He) flow in to the system and xH is the molar fraction of H2 at the outlet.Typically, the run-to-run error with this method is below 3%, compared to at least 15–20% error with a standard batch system. Also, this continuous-flow system is independent of experimental conditions and does not require human intervention when sampling or local derivative approximations as with regular batch system measurements (Figure S34). In addition to the better accuracy of this method in monitoring kinetic trends in the photocatalytic H2 evolution process, the method also keeps the media unperturbed—since the presence of the GC sampling line does not affect the hydrogen balance—thus completely eliminating typical sampling losses and mathematical and experimental uncertainties associated with batch photocatalytic reactor systems.

Figure 3

Continuous-flow photocatalytic reactor design. Schematic diagram of the designed continuous-flow photocatalytic reactor system (red streamlines are the continuous-flow pathway of gas). In contrast, the batch configuration involves mass flow controllers as dead-ends after back purging the initial headspace and replaces the autosampler by a septa-port or a manual sampling valve.

In a typical photocatalytic experin class="Chemical">ment using our hybrid system, 5 mg of TpDTz COF was dispersed in 10 mL of H2O containing TEoA (10 vol%) as the SED, and the pH was adjusted to 8.5 by adding HCl. Ni(OAc)2·4H2O (10 wt%, 0.5 mg) and ME (10 equiv, 1.4 μL) were then charged to instantaneously form the brown-colored NiME cluster co-catalyst. When irradiated with 100 mW/cm2 AM 1.5 radiation, the resulting mixture produces H2 actively over a period of at least 70 h—with ∼40% of the highest production rate still preserved after this time—in a single run without adding additional TEoA or co-catalyst (Figure a). A maximum H2 evolution rate of 941 μmol g–1 h–1 with a TONNi > 103 (70 h) and a TOF = 2.3 h–1 when the system is fully active were obtained. A mathematically projected (Supporting Information section S8) TONNi > 443 (890 μmol of total H2 evolution) can be obtained for the photocatalytic H2 evolution performance corresponding to a complete depletion of the co-catalyst. The relation between co-catalyst, SED, and observed activity loss of the system in time was confirmed by in situ addition of loss-equivalent amounts of ME or TEoA independently after 72 h of illumination, which did not change the deactivation trends observed (Supporting Information, section S8). It must be noted that the TONNi mentioned above is only a lower limit calculated based on the total amount of Ni(II) salt used for the photocatalysis experiment. Under identical conditions the erythrosin B (EB) dye-sensitized system[18] produces H2 with a maximum rate of 49 297 μmol g–1 h–1 (attained in 1 h); however, the rate rapidly drops off, and the whole system becomes completely inactive within 7 h. A TONNi > 36.5 (73 μmol of total H2 evolution) was obtained after 7 h (TOFNi = 31.9 h–1), which is 12 times lower than the value projected for the TpDTz COF-sensitized system (Figure a). These results demonstrate the added value of using a heterogeneous PS to stabilize charge transfer in photocatalytic hybrid systems.

Figure 4

Photocatalytic H2 evolution. (a) Comparison of photocatalytic H2 evolution rates in water (H2O) and deuterium oxide (D2O), using TpDTz COF over 72 h and EB dye under AM 1.5 light irradiation [COF photosensitizer: 5 mg of TpDTz COF in 10 mL of H2O/D2O with 10 vol% TEoA, 0.5 mg of Ni(OAc)2, and 1.4 μL of ME at a final pH of 8.5; dye photosensitizer: 1.33 mg of EB in 10 mL of H2O with 10 vol% TEoA, 0.5 mg of Ni(OAc)2, and 1.4 μL of ME at a final pH of 8.5]. (b) Light on–off cycles for photocatalytic H2 evolution experiments with TpDTz COF in water over 26 h. (c) Photocatalytic H2 evolution with TpDTz COF in water using different co-catalysts. (d) Photocatalytic H2 evolution with TpDTz COF in water using different metal-ME co-catalysts. (e) Photocatalytic H2 evolution from water using different photosensitizers. (f) Overlay of the UV–vis DR spectra of TpDTz COF with apparent quantum efficiency (AQEs) for the photocatalytic H2 evolution reaction with TpDTz COF at four different incident light wavelengths.

Also, the TpDTz COF-NiME photocatalytic system produces H2 at a 17% higher maximum rate and has a TON nearly 8-fold as high as our previously reported N2-COF-cobaloxime-based system (782 μmol h–1 g–1, TONCo = 54.4), while operating in water.[12] The H2 evolution rate and the sustained activity of this simple TpDTz COF-NiME system are competitive with and even superior to those of many COF-based photocatalytic systems (Table S3) and other benchmark photocatalytic systems involving metallic Pt or molecular Ni co-catalysts. Examples include g-C3N4/Pt (840 μmol g–1 h–1),[25b] TP-BDDA/Pt (324 μmol g–1 h–1),[10c] N2-COF/Pt (480 μmol g–1 h–1),[10b] crystalline poly(triazine imide)/Pt (864 μmol g–1 h–1),[25a] sg-CN-Ni (103 μmol g–1 h–1),[25c] Ni12P5/g-C3N4(536 μmol g–1 h–1),[25d]NCNCN-NiP (763 μmol g–1 h–1),[25e] and carbon quantum dots (CQDs)-Ni (398 μmol g–1 h–1).[25f]

Control experiments were performed by sequentially removing one of the components, i.e., TpDTz COF, TEoA, Ni(OAc)2·4H2O, and ME, at a time from our photocatalytic system to identify their importance and role for the H2 evolution. Indeed, no H2 evolution was observed for a period of 12 h unless all individual components act in concert, signifying that each is essential for the photocatalytic system to work and efficiently produce H2 (Figure S27).

Furthermore, a 1:10 metal-to-ligand molar ratio and 10 wt% of catalyst with respect to the PS were observed to elicit the best photocatalytic performance (Figures S28 and S29). To confirm water as the source of H2, the photocatalytic reaction was performed in D2O under identical conditions (Figure a). A production rate for D2 in D2O similar to that for H2 in H2O was observed over 72 h, taking batch-to-batch variations into account. This result suggests that water is the hydrogen source responsible for the production of H2, assuming that no significant proton/deuterium exchange processes in the individual components are at play. This finding was further confirmed by an almost complete disappearance of the m/z = 2 signal for H2 in a mass spectrometric measurement of the headspace gas of the photocatalytic reaction performed in D2O (Figure S26). Note that D2 is evolved with a time lag compared to H2, which is likely due to the kinetic isotope effect (KIE) of deuterium as described below. Further, H2 evolution experiments performed under multiple light on–off cycles over a period of 26 h (Figure b) suggests a purely light driven H2 evolution process in water. Once the catalytic system is fully active, H2 evolution activity is seen to be restored even after a prolonged light off period.

SED and the reaction pH are known to have a profound influence on the activity of many H2 production systems.[10b,10c,12] In our case, we observed a similar effect; the rate of H2 generated from the photochemical reaction is the highest (941 μmol g–1 h–1) at pH 8.5 using TEoA as SED. However, at acidic conditions (pH 6.5) there was negligible H2 evolution (16 μmol g–1 h–1). This could be attributed to the protonation of TEoA or due to inhibition of proton loss from one-electron oxidized TEoA+.[13c] Notable H2 evolution is observed over 24 h under alkaline conditions (pH 11), albeit at lower rates (308 μmol g–1 h–1) as compared to pH 8.5 (Figure S30). This is possibly due to the reduced driving force for protonation of the Ni hydride intermediate co-catalyst species at higher pH to subsequently generate H2. Triethylamine (TEA) and Na2S were also explored as potential SEDs. Interestingly, they produce H2 but with significantly lower rates of 84 μmol g–1 h–1 and 7 μmol g–1 h–1, respectively (Figure S31). Higher TEoA concentrations were found to decrease H2 evolution rates; a TEoA concentration of 10 vol% in water was observed to result in the maximum H2 production rate (Figure S32).

H2 evolution rates of the photocatalytic systems containing TpDTz COF PS and different Ni(II)co-catalysts were measured (Figure c). Different sulfur-containing compounds, such as thiourea (TU) and 2-mercaptophenol (MP), were explored as potential ligands for in situ formation of Ni(II)co-catalyst complexes. However, neither NiTU nor NiMP produced any H2 with TpDTz COF, possibly due to unfavorable complexation of the ligands with Ni(II) in water: TU and MP are known to be poorer complexation agents as compared to ME.[26] Also, a reported ex situ synthesized Ni(abt)2 complex[15c] was studied as a potential H2 evolution co-catalyst under our experimental conditions, but no H2 evolution was seen, most likely due to its poor solubility in water. It is also interesting to note that TpDTz COF produces H2 with a significantly smaller rate of 23 μmol g–1 h–1 with metallic Pt co-catalyst and TEoA at pH 8.5 over a period of 24 h as compared to that with NiME. The significant difference between H2 evolution of the molecular co-catalyst and photodeposited Pt nanoparticles is difficult to explain by a single effect.[12,25f] However, it may be argued that the higher activity of the NiME co-catalyzed system in contrast to the surface bound Pt nanoparticles (Figure S48) is due to a more effective blocking of charge carrier recombination since the co-catalyst is physically separated from the framework (physisorbed), which may support better charge separation. We further screened the H2 evolution activity of other transition metal-ME complexes, such as CoME and CuME, with TpDTz COF as the PS following a similar method as that of NiME. Although all systems produced H2, they do so with a much lower rate following the order NiME (941 μmol g–1 h–1) > CoME (85 μmol g–1 h–1) > CuME (52 μmol g–1 h–1). This could be due to the poor solubility of CoME and CuME clusters in water compared to the NiME, which is in accordance with the reported dye sensitized molecular system[18] (Figure d).

We then evaluated the H2 evolution ability of the n class="Chemical">NiME cluster co-catalyst with a variety of photoabsorbing materials; TpDTP COF,[19] N3-COF,[10b] an amorphous porous polymer containing TzTz groups (TzTz-POP-3),[27] and the diamine linker DTz were tested under identical conditions. Even though N3-COF is considered one of the most active COFs for photocatalytic H2 generation (reported rate of 1700 μmol g–1 h–1 when co-catalyzed by Pt),[10b] with NiME co-catalyst it produces H2 only at a very low rate of 40 μmol g–1 h–1 (Figure e). Under similar conditions TpDTP COF produces H2 at a rate of 160 μmol g–1 h–1, which is nearly 6 times less compared to that of the TpDTz COF sensitized system. The marked difference in photocatalytic activity between TpDTz COF and TpDTP COF may in part be rationalized by the reaction conditions which were not optimized specifically for TpDTP COF, but also by their different photon absorption characteristics. A redshift of ∼67 nm is observed for TpDTz COF with respect to TpDTP COF, which indicates that TpDTz COF absorbs photons more effectively in the visible range. This said, increased reactivity is only expected if the conduction band is not significantly lowered to maintain the thermodynamic driving force for the HER. In addition to that, the higher crystallinity and the higher BET surface area of 1356 m2 g–1 for TpDTz COF versus 736 m2 g–1 for TpDTP COF, along with a better dispersibility of the more hydrophilic TpDTz COF in aqueous solution, are likely determining factors for the enhanced photocatalytic activity. Notably, the amorphous polymer TzTz-POP-3 and the diamine linker DTz are completely inactive at producing H2 with the NiME co-catalyst (Figure e). Overall, the significantly lower reactivity of other PSs in producing H2 with the NiME co-catalyst is rationalized by a combined effect of unfavorable charge-transfer processes, reduced light harvesting, low crystallinity and surface area, and poor dispersibility in water.

The apparent quantum efficiency (AQE) was calculated using four different bandpass filters with central wavelengths (±20 nm) at 400, 500, 550, and 600 nm to quantify the spectral contribution toward n class="Chemical">H2 evolution activity of the TpDTz COF photoabsorber. Figure f shows that TpDTz COF has a maximum AQE of 0.2% at 400 nm. Under AM 1.5 illumination, the AQE was estimated to be 0.044%, which is higher than that of our previously reported N2-COF-cobaloxime H2 evolution system (0.027%).[12]

We further verified the photochemical stability of TpDTz COF after a 72 h long photocatalysis experiment. The isolated TpDTz COF sample was fully characterized using PXRD, ssNMR, SEM, and TEM, and it was found that the framework structure, crystallinity, and morphology of TpDTz COF are retained (Supporting Information, section S9), thus supporting the high chemical stability of this COF (vide supra). A small additional signal at 56.6 ppm in the 13C ssNMR spectrum possibly corresponds to trapped ME molecules inside the TpDTz COF pore. This, together with the observation of traces of Ni in the post-photocatalytic TpDTz COF using SEM-EDAX (Figure S46) may hint to chemisorption of small amounts of co-catalyst to the COF walls. However, 15N ssNMR of the TpDTz COF sample does not show any noteworthy difference in the chemical shifts of the N signals before and after photocatalysis (Figure S41), suggesting that there is no substantial direct interaction between the residual Ni and the nitrogen centers of the COF. Also, the as-recovered TpDTz COF sample does not produce any H2 under identical photocatalytic conditions except for the absence of Ni(OAc)2·4H2O and 2-ME ligand. This suggests that the interaction between TpDTz COF and NiME is mostly physical, and no lasting chemical interaction exists between the two components. Our finding thus suggests an outer-sphere electron transfer to be at play, which nevertheless is efficient enough to allow facile charge transfer from the photoabsorber to the NiME co-catalyst (Figure a).

Figure 5

Reaction limitations insights. (a) General schematic of the proposed pathway for H2 evolution (color code: gray, C; red, O; yellow, S; blue, N; and light pink/white, H). (b) Proposed key steps of the photocatalytic H2 evolution reaction with TpDTz COF and NiME cluster co-catalyst. [Ni-L] denotes a ligand-coordinated co-catalyst state which is attained fast compared to the [R] state, [R] denotes the catalyst resting state, which is catalytically active nickel cluster species, [D] denotes the deactivated species, and [I] denotes an intermediate reduced catalyst species able to run the HER step.

To obtain deeper insights into the photocatalytic mechanism, an overall coarse-grain mathematical model (Supporting Information eq S6.2) of the photocatalytic reaction was developed (see Supporting Information section S8 for details) by taking advantage of the quantification of hydrogen evolution rates in our flow detection platform. Our model was based on three primary experimentally observed trends: the activation time required by the photocatalytic system to reach maximum rates in the first run (Figure a), the absence of this activation time during light on–off cycles or long dark periods before illumination (Figure b), and the KIE in D2O, namely, smaller deuterium evolution rates and delayed response, together with the observation of a similar initial activation time as with H2O (Figure a). A possible reaction model is outlined in Figure b. The absence of an activation time during light on–off cycles (Figure b) suggests a light enhanced formation of a catalyst resting state [R] of the NiME complex upon illumination, as seen by the initial activation time required for the system to reach maximum efficiency. The reaction network can then be reduced to the following core steps using a microkinetics analysis. For the heterogeneous (COF) fast cycle: COF photoexcitation (hν), exciton recombination in the COF (krec), reductive quenching of the COF (kq), and electron transfer from COF•– to [R] to form the active intermediate species [I] (ka). Quantum-chemical calculations on the TpDTz pore model system (Figure S55) identify the lowest photoexcitation energy to be 2.30 eV (Table S9), the difference density of this excited state (Figure S57) visualizing the exciton. The spin density of the radical anion as a result of the reductive quenching of this state is shown in Figure S58. For the homogeneous (catalyst) cycle: formation of a rapidly coordinated complex [Ni-L] (Keq), slow assembly of the catalytically active species (kT, kT–1), an apparent first-order activation step from [R] to [I] (ka), an irreversible deactivation step [D] (kd), and the closing of the catalytic cycle via a dark step that produces H2 (kHER). If in such a system kT and kT–1 are significantly slower compared to the rest of the steps, the on–off behavior can be explained, because in the absence of light the dark equilibrium is slow and the amount of [R] and [I] will not change significantly over time. Furthermore, the absence of HER in the dark during the light on–off cycles observed in Figure b suggests that a kHER-limited homogeneous cycle is unlikely. Then, once nickel enters the cycle as [R] more rapidly due to a light-shifted equilibrium, [R] will build up until the rate of [D] leaving the system irreversibly is equal to the rate of formation of [R], [I] being stationary. This will lead to an expression for activation of the photocatalytic system such that the activation curve is an exponential asymptote with time constant ta, which is dominated by electron-transfer rate ka, a linear deactivation with an apparent time constant td, an HER apparent kinetic constant RH, and the apparent transient time t0 only which corresponds to the initial time delay necessary for the intermediate [I] to be pseudo-stationary (Supporting Information eq S6.2). Our model further provides an accurate fit to the data obtained in both H2O and D2O, in line with an expected trend of a KIE. In this case, RH, td, and t0 changed as expected, but the time constant for the activation step (ta) being independent of the isotopic mass is almost unchanged (Table S2), further corroborating our proposed rate-limiting steps (RLS). Our reaction model not only explains these qualitative trends but also provides an accurate fit with standard errors below 5% for different data sets. It is important to note that acquiring detailed insights into the H2/D2 evolution reaction mechanism for the TpDTz COF-NiME photocatalytic system became possible only with the use of a flow reactor system.

Our reaction modeling results suggest that as long as a slow catalyst activation time is observed, the RLS of the system is seemingly the electron transfer from the COF to the NiME complex. While this outcome is fully consistent with the assumed outer-sphere electron-transfer process, it reinforces the idea of studying the kinetics of such processes in more detail as this will be crucial to improve the HER rate by rational design of the COF–co-catalyst interface.

Conclusions

We report the first COF photosensitizer and noble-metal-free molecular co-catalyst photocatalytic system for sustained solar H2 production from water. This single–site system comprises the newly designed N/S-containing TpDTz COF PS that absorbs strongly in the visible region of the solar spectrum and is robust for long-term hydrogen evolution. In combination with an earth-abundant Ni-thiolate cluster co-catalyst which self-assembles in water, we obtain solar H2 evolution rates as high as 941 μmol h–1 g–1 and a TONNi > 103 (70 h) with persistent H2 evolution for more than 70 h in a single run, which surpasses many benchmark photocatalytic H2 evolution systems based on COFs and carbon nitrides. To map out the parameter space of this hybrid photocatalytic system, we comprehensively screened the influence of various reaction components, including pH, SED, co-catalyst metal centers, different N/S-containing chelating ligands, and a variety of PSs on the photocatalytic activity. In addition, we have introduced a newly designed continuous-flow system enabling the non-invasive, direct detection of the H2 production rate. This platform not only provides higher accuracy in quantification; it also paves the way for unprecedented insights into the reaction mechanism which are difficult to obtain with the existing batch measurement methods. Microkinetic modeling of the reaction system suggests that an outer-sphere electron transfer from the photoabsorber to the catalyst is the rate-limiting step, thus spotlighting the importance of the rational design of the COF–co-catalyst interface.