Structurally Diverse Covalent Triazine-Based Framework Materials for Photocatalytic Hydrogen Evolution from Water

A structurally diverse family of 39 covalent triazine-based framework materials (CTFs) are synthesized by Suzuki–Miyaura polycondensation and tested as hydrogen evolution photocatalysts using a high-throughput workflow. The two best-performing CTFs are based on benzonitrile and dibenzo[b,d]thiophene sulfone linkers, respectively, with catalytic activities that are among the highest for this material class. The activities of the different CTFs are rationalized in terms of four variables: the predicted electron affinity, the predicted ionization potential, the optical gap, and the dispersibility of the CTFs particles in solution, as measured by optical transmittance. The electron affinity and dispersibility in solution are found to be the best predictors of photocatalytic hydrogen evolution activity.


■ INTRODUCTION
Hydrogen is essential in many industrial processes, such as the hydrogenation of unsaturated fats, 1 production of stainless steel (ss) 2 and ammonia, 3 and crude oil purification. 4 Hydrogen also has potential as a clean fuel for fuel cells that does not result in the emission of greenhouse gases at the point of use. 5 Currently, hydrogen is made mostly by steam reforming of methane, which releases carbon dioxide directly and indirectly as a result of the high temperatures and pressures required for this process. 6 Photocatalytic water splitting has been studied extensively as a green method for the production of hydrogen, releasing oxygen as the only side product. Light-driven water splitting was first demonstrated in 1972 for a photoelectrochemical cell with a TiO 2 anode and platinum cathode. 7 After this first report, many inorganic photocatalysts have been studied, also in suspension systems that do not require an applied bias. 8−10 Conjugated polymer photocatalysts were studied as early as 1985, 11−14 but it took the first report on carbon nitride in 2009 15 to trigger real interest in this field. Some limitations arise from the synthesis temperatures needed for carbon nitrides, which are typically above 300°C, and this can result in a lack of synthetic control. A large number of studies reported improved hydrogen production rates under sacrificial conditions by varying the synthesis conditions and starting materials, 16−21 by introducing doping, 22,23 by controlling end groups, 24−26 or by using heterojunctions 27 or donor/acceptortype systems. 28 Conjugated organic polymers are typically made under much milder conditions 29 than carbon nitrides and offer potential advantages in terms of synthetic control via the incorporation of different building blocks 30−32 and by monomer sequence control. 33 Conjugated polymers can be tailored in terms of their optical properties, as well as their driving force for proton reduction and water/sacrificial electron donor oxidation, respectively. 33,34 A number of conjugated organic materials have now been studied for sacrificial hydrogen and/or oxygen production from water using light, including oligomers, 35−37 linear polymers, 34,38−45 covalent organic frameworks, 46−54 and networks. 42,55−62 Covalent triazine-based frameworks (CTFs) are a subclass of conjugated microporous polymers that have been reported to be photocatalytically active for hydrogen and oxygen evolution from water. 30,63,64 The most common reactions used to prepare CTFs are the acid-catalyzed trimerization of aromatic nitriles, 57,65−68 sometimes combined with sulfur and phosphorous doping, 69,70 and high-temperature ionothermal synthesis. 71−73 A two-step route was also reported whereby acidcatalyzed trimerizations were used to make CTFs that were then further condensed using thermal methods, 74,75 or doped using ammonium halides. 76 We previously used acid-catalyzed trimerization to prepare a series of CTFs with an increasing linker length between the triazine vertices (phenylene to quarterphenylene; Figure 1, CTF-1−4). In that series, the highest hydrogen evolution activity was observed for the acidcatalyzed CTF-2, which is based on a biphenylene linker. 57 Alkyne-linked CTFs have also recently been made using Sonogashira−Hagihara polycondensation linking tris(4-ethynylphenyl)-1,3,5-triazine with a range of nitrogen-and sulfurcontaining compounds. 77 To date, studies on CTFs have shown rather limited structural diversity either because of the availability of monomers, 57,78 and/or the harsh synthesis conditions employed; for example, in ionothermal synthesis, partial carbonization might limit the materials' photocatalytic activity, 65,74 while acid-catalyzed cyclotrimerization is incompatible with functional groups that are acid sensitive. In this study, we present 39 new CTFs prepared by Pd(0)-catalyzed Suzuki−Miyaura polycondensation. We study these in detail for their photocatalytic hydrogen production activity ( Figure  2). As such, this single study covers a significantly more diverse set of CTF materials than the 18 structurally defined photocatalytically active CTFs published so far in the combined literature. 30 O) and sealed with a septum. The resulting suspension was ultrasonicated for 10 min until the photocatalyst was well-dispersed before degassing by N 2 bubbling for 30 min. The reaction mixture was side illuminated with a 300 W Newport Xe light source (model: 6258, ozone free) equipped with a λ > 420 nm filter for 5 h unless stated otherwise. The light source was cooled and IR light cut out by water circulating through a metal jacket with quartz windows. For the illumination under solar light, an Oriel LSH-7320 ABA LED solar simulator with an output of 1.0 sun was used (classification ICE 60904-9 2007: spectral match A, uniformity classification B, temporal stability A). Gas samples were taken with a gas-tight syringe and run on a Bruker 450-GC gas chromatograph equipped with a molecular sieve 13 × 60−80 mesh 1.5 m × 1/8 in. × 2 mm ss column at 50°C with an argon flow of 40.0 mL min −1 . Hydrogen was detected with a thermal conductivity detector referencing against standard gas with a known concentration of hydrogen. Hydrogen dissolved in the reaction mixture was not measured, and the pressure increase generated by the evolved hydrogen was neglected in the calculations. The rates were determined from a linear regression fit and the error is given as the standard deviation from the linear fit. No hydrogen evolution was observed for mixtures of water/triethylamine/H 2 PtCl 6 under λ > 420 nm illumination or under the solar simulator illumination in the absence of a photocatalyst. No hydrogen was observed for L-ascorbic acid solution under λ > 420, >350, and >295 nm illumination (300 W Xe light source) or solar-simulated light (ABA, 400−1100 nm) in the absence of a photocatalyst. Aqueous Na 2 S solution showed a limited hydrogen production rate of 0.15 μmol h −1 under λ > 295 nm illumination (300 W Xe light source). For longer-term stability experiments, the photocatalysis was carried out for 5 h and the polymer was recovered through centrifugation and after decantation of the liquids, water, methanol, and triethylamine (TEA) mixtures were added for the next run. After 100 h of combined photocatalysis, the photocatalyst was isolated by decantation of the aqueous phase, washed with water and THF three times each, and dried overnight at 120°C before analysis.
High-Throughput Hydrogen Evolution Experiments. The Agilent Technologies vials (crimp top, headspace, clear with graduation marks and write-on spot, flat bottom, 10 mL, 23 × 46 mm) were charged with 5.0 ± 0.1 mg of polymer powders and transferred to the Chemspeed Accelerator SWING for liquid transfer. Degassed jars with triethylamine, methanol, and a stock solution of H 2 PtCl 6 were loaded into the automated liquid handling platform.
The system was then closed and purged for 4 h with nitrogen. The automated liquid handling platform then dispenses the liquids as specified, which was unless stated otherwise degassed water (1.7 mL) or aqueous H 2 PtCl 6 solution (1.7 mL, 0.24 wt % in H 2 O) triethylamine (1.7 mL), methanol (1.7 mL). The vials were then capped using the capper/crimper tool under inert conditions. Once capped, the samples are taken out, shaken for 5 s, and transferred to an ultrasonic bath to disperse the photocatalysts. An Oriel Solar Simulator 94123A with an output of 1.0 sun was then used to illuminate the vials on a Stuart roller bar SRT9 for the time specified (classification IEC 60904-9 2007 spectral match A, uniformity classification A, temporal stability A, 1600 W xenon light source, 12 × 12 in. 2 output beam, air mass 1.5 G filter, 350−1000 nm). After photocatalysis, the gaseous products of the samples are measured on an Agilent Technologies 7890B gas chromatograph connected to an Agilent Technologies 7697A headspace sampler. No hydrogen evolution was observed for mixtures of water/triethylamine/methanol or water/triethylamine/methanol/H 2 PtCl 6 under solar irradiation (1.0 sun) in the absence of a photocatalyst. No hydrogen production was observed in the absence of a photocatalyst.
(Time-Dependent) Density Functional Theory [(TD-)DFT] Calculations. The CTFs were modeled for the purpose of computationally screening their properties as triazine−linker−triazine (TLT) cluster models. The relevant three-dimensional structures of these TLT CTF models were generated using a multistep process starting from the SMILES codes for the linkers using the supramolecular toolkit (stk). 82,83 stk is a Python library for the assembly, structure generation, and property calculation of supramolecules (using functionality from RDKit 84 and Schrodinger PLC's Macromodel software) and allowed us to automate the assembly and conformer search process. The as-assembled TLT cluster models were energy minimized using the OPLS2005 85,86 force field. After the relaxation step, a high-temperature OPLS2005 molecular dynamics run with regular quenches (700 K, 50 ps of simulation time, and 1000 quenches) was employed to find low-energy conformers for each cluster model. The lowest-energy conformers were finally reoptimized using density functional theory (DFT), using the B3LYP 87,88 density functional and the DZP 89 basis set. The thermodynamic ability of the various CTFs, as modeled by TLT cluster models, to reduce protons and oxidize water and/or triethylamine was analyzed in the terms of the potentials associated with free electrons [electron affinity (EA)] and holes [ionization potential (IP)], as well as the exciton (IP* and EA*), relative to those of the different solution half reactions. These potentials are calculated using our standard ΔSCF approach 90,91 based on DFT/timedependent DFT (TD-DFT) calculations on polymer cluster models and the use of the COSMO 92 continuum solvation model (ε r 80.1, water) to approximate the dielectric environment of the polymers in the presence of water. As in our previous work 33,34,38,43,57,62,90,91 and as in the CTF model generation stage, the potential calculations use the B3LYP density functional and the DZP basis set, a combination that we previously found to result in the prediction of accurate potentials for polymeric solids when compared to experimental photoelectron spectroscopy data. 91 Potentials of solution halfreactions are taken from previous work. 33 The absorption spectra of the CTFs are approximated by the vertical excitation spectra of the TLT cluster models, as calculated by TD-B3LYP/DZP. Similarly, the optical gap, the energy of light above (and the wavelength below) which the CTF starts absorbing light, is approximated by the energy of the lowest vertical singlet excitation.

■ RESULTS AND DISCUSSION
A structurally diverse library of new CTFs was synthesized by using a trisphenyl-1,3,5-triazine building block as the node (M45) connected by various linkers (M5−44). The resulting CTFs are referred to as CTF-xx (whereby xx is the number of the dibromo monomers used in the synthesis, Mxx). The linkers cover a wide range of different aromatic units, which can be divided into the following subsets: (i) phenylene oligomers; (ii) five-membered aromatic heterocycles; (iii) substituted phenylenes; (iv) bipyridyls; (v) nitrogen heterocycles; (vi) planar heterofluorenes; and (vii) azoles and quinoxalines.
To probe the dispersibility of the polymer particles in the reaction mixture, the transmission of light through a suspension of the photocatalysts in the water/TEA/methanol mixture was determined. The transmission values were found to range from low values of 0.8%, indicating good dispersibility (i.e., almost opaque dispersions), to high values of up to 59.4%, indicating poor dispersibility in water/TEA (triethylamine)/ methanol mixtures.
The photocatalytic activity for proton reduction of the materials was tested using a high-throughput workflow: the polymers (5 ± 0.5 mg) were dispersed in mixtures of water, TEA, and methanol (1/1/1 ratio). Platinum (3 wt %) was in situ photodeposited onto the photocatalyst 57 acting as a cocatalyst, which has been shown previously to improve hydrogen evolution rates (HERs). 57,67 TEA acts as a hole scavenger, 35 and methanol is used to ensure miscibility between water and TEA. 39,93 Samples were illuminated with a solar simulator for 2 h before measuring the amount of hydrogen produced using an automated gas chromatograph.
Two materials, CTF-15 (2,5-linked benzonitrile) and CTF-34 (3,7-linked dibenzo[b,d]thiophene sulfone), were identified by this screen as the best-performing catalysts. The results of the screening measurements were compared to the traditional kinetic experiments for each photocatalyst over 5 h using a 300 W Xe light source equipped with a λ > 420 nm filter (Figures 4, S-73−S-94, and Table S-7). Again, the two most active photocatalysts were found to be CTF-15 (2946 μmol g −1 h −1 ) and CTF-34 (2373 μmol g −1 h −1 ), which had significantly higher values than that of the previously reported CTF-3 Suzuki (158 μmol g −1 h −1 ) when tested under the same condition. Overall, there is a good correlation between the high-throughput screening measurements and the kinetic  measurements, especially when one considers the marked differences in the spectral output of the light sources for the two measurements, the different head spaces of the reaction vessels (much smaller in the high-throughput screen), and the different agitation methods (rolling/rocking in the screen versus stirring in the kinetic measurements). 94 Following our previous work on linear conjugated polymer photocatalysts, 34 we attempted to analyze the performance of the CTFs in terms of four properties: their electron affinity (EA), ionization potential (IP), and optical gap, as well as the optical transmittance of the reaction mixture after the CTF is dispersed in it. Figure 2a−d shows the dependency of the measured hydrogen evolution rate on those four properties and dashed lines envelope the data points for all CTFs. The measured hydrogen evolution rate shows the clearest correlation with the predicted EA values (Figure 2a). EA, often approximated by (the negative of) the lowest unoccupied molecular orbital energy, governs the driving force for water reduction. None of the CTFs with EA values above −1.4 V relative to the standard hydrogen electrode have high hydrogen evolution rates (here defined to have HER >50% of the CTF with the highest activity in this study), while CTFs with even more positive EA values effectively produce no hydrogen. The observed HER is highest at around −1.7 V, but as the photocatalytic performance is not only controlled by the EA of the materials several CTFs with EA values in this range evolve only a small amount of hydrogen or show no activity ( Figure  2a). Where EA governs the driving force for proton reduction, IP, often approximated by (the negative of) the highest occupied molecular orbital energy, determines the driving force for water oxidation or, as in this study, the oxidation of the hole scavenger (TEA). The measured hydrogen evolution rates lie in a broad envelope of increasing hydrogen evolution rates with increasing predicted IP. A maximum is observed at approximately +1.6 V, and then the measured hydrogen evolution rates decrease again for higher IP values (Figure 2b). In contrast to the case of EA, CTFs with high hydrogen evolution rates, as defined above, can be found across most of the IP range, and there are no IP regions in which exclusively CTFs occur that produce little or no hydrogen.

Chemistry of Materials
The apparent higher dependency of the measured hydrogen evolution rates of the materials on their predicted EA values (Figure 2a) when compared to the correlation of photocatalytic activity and predicted IP values of these materials (Figure 2b) can be explained when the effect of triethylamine on the pH of the photocatalysis mixtures is considered. The high pH of these mixtures of approximately 11.5 effects the potentials; hence, the proton reduction potential will lie at −0.7 V, which means that CTFs with EA values more positive than ∼−1.0 V therefore are expected to have little or no driving force for proton reduction (Figure 2a). The overall oxidation of triethylamine to give diethylamine and acetaldehyde has a predicted potential to be at −0.7 V. Therefore, it is expected that the CTFs studied here all have significant driving force for scavenger oxidation. Even the one-hole oxidation step of triethylamine (+0.7 V) to the triethylamine radical [N(CH 2 CH 3 )CHCH 3 ], which can be an effective thermodynamic obstacle for the overall oxidation reaction, is predicted to favor most CTFs considered.
Plots of measured hydrogen evolution rates versus predicted excited-state ionization potential (IP*) and electron affinity (EA*), where the excited state (exciton) provides the electron or hole to drive the solution chemistry, look very similar to their EA and IP equivalents (Figure S-126). As a result, the correlation between the hydrogen evolution rate and the predicted IP* values is also stronger than that between the measured and predicted EA* values. Exciton dissociation, or not, thus appears to make little difference for these materials in terms of the predicted potentials.
In the case of the (computed) optical gap (Figure 2c), we find that the CTFs with the highest activity have large optical gaps. Because, as discussed above, the predicted optical gap values and experimental absorption onsets are strongly correlated, the plot of the measured HER as a function of absorption onset looks similar ( Figure S-117), and we made the same observation. The fact that the most active materials have large optical gaps is counterintuitive and would have been surprising had we not observed the same behavior for linear conjugated co-polymers. 34 As for those linear polymers, plotting EA/IP as a functional of optical gap in this work ( Figure S-125) shows that EA and optical gap are strongly correlated; that is, the CTFs with the largest optical gaps also have the most negative EA values. Hence, for both families of materials, the linear polymers and the CTFs, the hydrogen evolution rates, at least when using TEA, are limited more by the thermodynamic driving force for proton reduction than by light absorption.
The final property that we considered was the light transmittance of dispersions of the CTF in the photocatalysis mixture, i.e., TEA/MeOH/water (Figure 2d). Values range from 100% (indicating quick settling or "creaming" of the CTF) to 0% (indicating that the CTF is well-dispersed and scattered and/or absorbed light effectively). Light transmittance is a descriptor of the dispersibility of the CTF in the photocatalysis mixture, which is affected by interaction with water and scavenger, density, and size of the CTF particles. Hydrogen evolution rates in Figure 2d are generally higher for CTFs when lowering the transmittance, and all CTFs that show high rates have a transmittance of less than 50%, therefore showing a positive correlation between the dispersion of the CTF in the reaction mixture and the performance as a photocatalyst for proton reduction.
For completeness, we also considered a number of other properties beyond those considered in our previous work but found no correlation with the hydrogen evolution rate, other than that for the fitted fluorescence lifetimes measured via TCSPC. It appears that materials with shorter lifetimes have a higher photocatalytic activity ( Figure S-119). However, as in the case for the optical gap, this correlation appears counterintuitive. As the TCSPC measurements are performed  in the absence of scavenger, these short lifetimes are not due to quenching but rather a sign of inherently short-lived excited states, which should be a negative effect rather than a positive one. We suspect that this is a noncausal correlation with the hydrogen evolution rate, as for the optical gap, that arises because the fluorescence lifetimes are correlated with another material property. The above analysis demonstrates that CTFs with the highest photocataytic activities occur for specific combinations of (material) properties. The word combinations in the above sentence is important, as there are clearly also CTFs with (near-)"ideal" EA, IP, optical gap, and/or transmittance values that yet only evolve a very small amount of hydrogen or are not active at all (see also . In line with our previous observations for linear conjugated polymers, each of these four properties appear to be a necessary but not sufficient condition for a CTF to be an active photocatalyst.

Chemistry of Materials
The high-throughput system was further used to study the best-performing material, CTF-15 (2,5-linked benzonitrile), with various different scavenger systems in combination with platinum as the co-catalyst (Figure S-97 and Table S-9). The screening shows that the highest performance was obtained in this case for water/TEA/MeOH mixtures (1909 μmol g −1 h −1 ), followed by 5 vol % TEA in water (1830 μmol g −1 h −1 ). Moderate hydrogen production was observed when L-ascorbic acid and sodium sulfide were used as scavengers (116 and 139 μmol g −1 h −1 , respectively) and low performance was observed for disodium ethylenediaminetetraacetate, L-cysteine, and thiophenol (101, 82, and 77 μmol g −1 h −1 ). FeCl 2 , methanol, phenol, and hydroquinone did not act as a scavenger and no hydrogen production was observed under these conditions. Furthermore, CTF-15 and CTF-34 were compared to CTF-16 and CTF-3 Suzuki in a HT screening using water/TEA/ MeOH mixtures as well as sodium sulfide and L-ascorbic acid as scavengers (see Figure S-103). In all cases, CTF-15 and CTF-34 were the best-performing materials compared to CTF-16 followed by CTF-3 Suzuki, which indicates strongly that superior photocatalytsts were obtained, rather than materials that were simply better optimized for the oxidation of TEA.

■ CONCLUSIONS
A library of 39 new CTFs was made via Suzuki−Miyaura polycondensation, where the modularity of this approach allowed us to obtain CTFs with a much richer structural diversity than has been studied previously. All materials were characterized and studied for their photocatalytic performance using a high-throughput workflow, as well as conventional kinetic measurements. The performance of the best-performing CTFs are among the highest for this material class. We showed that the photocatalytic performance of these CTFs can be analyzed by their predicted electron affinity, predicted ionization potential, optical gap, and dispersibility in the reaction mixture. Electron affinity and dispersibility were found to be the dominant variables under the conditions used.
In this study, we measured all materials using the rapid screening method (up to 50 samples per day) and traditional kinetic methods (1−2 samples per day). This was done to establish the reliability of the screening method. The generally good correlation (Figure 4) between the screening results and the kinetic measurements suggests that this could be a powerful method in the future to discover new photocatalysts of any materials class, as well as providing valuable insights into the structure−property relationships across a diverse range of chemical functionality.

* S Supporting Information
The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.chemmater.9b02825.

Chemistry of Materials
Article Synthetic procedures, FT-IR spectra, TGA traces, UV−