Structural Insights into Poly(Heptazine Imides): A Light-Storing Carbon Nitride Material for Dark Photocatalysis

Solving the structure of carbon nitrides has been a long-standing challenge due to the low crystallinity and complex structures observed within this class of earth-abundant photocatalysts. Herein, we report on two-dimensional layered potassium poly(heptazine imide) (K-PHI) and its proton-exchanged counterpart (H-PHI), obtained by ionothermal synthesis using a molecular precursor route. We present a comprehensive analysis of the in-plane and three-dimensional structure of PHI. Transmission electron microscopy and solid-state NMR spectroscopy, supported by quantum-chemical calculations, suggest a planar, imide-bridged heptazine backbone with trigonal symmetry in both K-PHI and H-PHI, whereas pair distribution function analyses and X-ray powder diffraction using recursive-like simulations of planar defects point to a structure-directing function of the pore content. While the out-of-plane structure of K-PHI exhibits a unidirectional layer offset, mediated by hydrated potassium ions, H-PHI is characterized by a high degree of stacking faults due to the weaker structure directing influence of pore water. Structure–property relationships in PHI reveal that a loss of in-plane coherence, materializing in smaller lateral platelet dimensions and increased terminal cyanamide groups, correlates with improved photocatalytic performance. Size-optimized H-PHI is highly active toward photocatalytic hydrogen evolution, with a rate of 3363 μmol/gh H2 placing it on par with the most active carbon nitrides. K- and H-PHI adopt a uniquely long-lived photoreduced polaronic state in which light-induced electrons are stored for more than 6 h in the dark and released upon addition of a Pt cocatalyst. This work highlights the importance of structure–property relationships in carbon nitrides for the rational design of highly active hydrogen evolution photocatalysts.


■ INTRODUCTION
A new class of (semi)crystalline two-dimensional (2D) carbon nitrides has emerged over the past few years, which exhibits outstanding photocatalytic performance 1−3 along with intriguing photophysical properties, such as the potential for timedelayed fuel production ("dark photocatalysis"), 4 which also make them attractive candidate materials for solar batteries. 5 A hallmark of this carbon nitride family is its ability to store electrons in the form of a long-lived photoreduced state, 4,6,7 which allows for the separation of light absorption (the light reaction) and catalytic conversion (the dark reaction) akin to natural photosynthesis. This material has been postulated to exhibit a 2D poly(heptazine imide) (PHI)-based structure, but so far a comprehensive elucidation of its local and long-range structure is lacking. PHI has been synthesized by both, molecular precursor approaches 8,9 as well as top down strategies starting from the well-known melon polymer. 2,10 And yet, even though the synthesis procedures may vary widely, the use of salt melts, i.e., ionothermal synthesis, seems to be vital to all of these approaches. 2,8−10 Ionothermal synthesis routes often yield fundamentally different products as compared to high-temperature solidstate approaches. A prominent example is the synthesis of melon, which is often referred to as graphitic carbon nitride, g-C 3 N 4 . Although the solid-state synthesis of melon yields a onedimensional heptazine-based polymer, 2D networks are typically formed by a salt-melt approach using the same types of precursors. Under these conditions, the ionic flux and its components, typically alkali metal halides, serve as a hightemperature solvent and structure-directing agent at the same time. 11−13 The structure-directing influence of alkali ions has been observed by Savateev et al. and is also known to be the governing factor during the formation of secondary building units in zeolites. 9,14 Interestingly, the composition of the melt also seems to have a profound influence on the type of carbon nitride backbonetriazine or heptazine-basedthat is formed during the reaction. 8 Herein, we have used a molecular building block approach to synthesize highly crystalline potassium containing PHI terminated by cyanamide(NCN − )-functional groups, abbreviated as potassium PHI (K-PHI), and its metal-free counterpart H-PHI. A comprehensive investigation of their local and long-range structures has been carried out through a combination of electron diffraction (ED) and transmission electron microscopy (TEM) imaging, solid-state nuclear magnetic resonance (ssNMR) spectroscopy, supported by quantum-chemical calculations, X-ray powder diffraction (XRPD), and pair distribution function analysis (PDF). We show that both materials feature an extended 2D network constituted by tri-s-triazine (heptazine) units connected via imide bridges. These layers assemble in either ordered or disordered long-range stacking relationships driven by interactions with the pore content, resulting in hitherto unknown structural modifications in this PHI family.

■ RESULTS AND DISCUSSION
The ionothermal bottom-up synthesis of PHI starts from the molecular precursors potassium melonate, dicyandiamide, and potassium thiocyanate (KSCN), which are heated under their autogenous pressure in an ampoule to yield highly crystalline potassium poly(heptazine imide) (K-PHI), terminated with NCN-functional groups. In contrast to other PHI-derived materials (mainly synthesized from LiCl/KCl eutectics), the herein discussed synthesis is based on KSCN as the salt component and hence flux. When K-PHI is treated with dilute acid, it is converted into the proton-bearing H-PHI, thus pointing to the Brønsted acid−base character of the polymer. 9 Note that the reaction is insensitive to which acid (e.g., HCl, H 3 PO 4 , H 2 SO 4 , or HClO 4 ) is used for the transformation (Figures S50−S52). In the following text, we elucidate the local structure of K-PHI and H-PHI by solid-state NMR spectroscopy and characterize the 2D projection of the structure by means of TEM, whereas XRPD, PDF analyses, and recursive-like simulations of planar defects lead to the classification of the material in the third dimension.
TEM Analysis. The TEM image in Figure 1a reveals the sheetlike morphology of the crystallites and high crystallinity of the carbon nitride backbone over a large area of up to several 100 nm (for overview TEM images see Supporting Information Figures S1 and S3). The lattice fringes correspond to an interval of 11 Å within the layer planes, illustrated by the high-resolution (HR) TEM image in Figure 1a. The good agreement between the measured (Figure 1b) and simulated (1c) selected area electron diffraction (SAED) patterns along the [001] zone axis based on the structural model shown in Figure 1d indicates trigonal layer symmetry of the PHI backbone in H-PHI. The fast Fourier transformation (FFT) obtained from large crystallites (Figure 1a, inset) also fits the depicted SAED pattern. The 100 lattice spacing of 11 Å corresponds to the pore-to-pore distance of 12.8 Å and is consistent with a hexagonal arrangement of trigonal pores and a 2D network topology, similar to that found in the triazine analogue poly(triazine imide) (PTI). 11,12 The TEM data suggest that ideal H-PHI layers crystallize with trigonal symmetry, forming a "graphitic" 2D framework of imidebridged heptazine building units akin to melamine-intercalated PHI, which crystallizes in the space group P31m. 15 Particularly in the H-PHI case, slight intensity deviations from the perfect hexagonal distribution of the electron diffraction spots ( Figure  1b) and the fact that the diffraction data of H-PHI are relatively tolerant against tilting of the specimen for up to ±10°i ndicate structural disorder along the [001] zone axis. Intensity deviations might also arise from losing pore content (i.e., water) in ultra high vacuum and local heating through the electron beam, which may induce on-spot incipient degradation and enable the layers to shift to the eclipsed case, as also seen in thermogravimetric XRPD experiments in Figure S27. The diffraction spots of K-PHI (Figure 1f, [001] zone axis) imply that the material exhibits monoclinic layer symmetry. The SAED pattern can be reproduced by ED simulation with   Figure 1g). The HR-TEM image (Figure 1e) also points toward the existence of large crystalline domains, with the FFT (inset) reproducing the pattern obtained from the simulation in Figure 1g. The difference in symmetry suggests an altered stacking behavior of the PHI backbones in K-PHI and H-PHI, likely driven by the pore constituents (water in H-PHI or potassium ions in K-PHI). Note that the water content nearly doubles from 10 wt % for K-PHI to 20 wt % for H-PHI, determined by combustion analysis and inductively coupled plasma (Tables S8 and S9). Solid-State NMR Spectroscopy. To probe the local structure of PHI, especially the intralayer connectivity and the distribution of cations within the pores, 13 C, 15  The direct excitation 13 C NMR experiment shows five distinct signals for K-PHI, which can be assigned according to the inset in Figure 2a. Like in melem, 24 the central carbon atom (C1) in the heptazine unit exhibits a chemical shift of 157 ppm. The signal at 164 ppm is attributed to a peripheral carbon atom next to a NH group (C2) as its signal is the most intense in the 13 C cross-polarization experiment confirming its proximity to a proton ( Figure S12). Additionally, the 13 C− 15 N 2D HETCOR spectrum shows distinctly that C2 correlates with the NH group at −242 ppm (Figure 2g), identifying this NH as the connecting bridge between two heptazine units. The 13 C NMR signal at 168 ppm arises from a carbon atom next to a deprotonated imide bridge (C3). Indeed, this C3 is only connected to nitrogen atoms with a chemical shift at −173 ppm (Figure 2g), a typical value for nonprotonated nitrogen. In the case of an attached NCN moiety, the neighboring heptazine carbon atom (C4) is shifted downfield to 172 ppm whereas the carbon atom of the terminal NCN group (C5) is found at 120 ppm, which is likewise observed for cyanamide in molecular potassium melonate. 25 However, according to the integrated intensities in NMR spectra, these cyanamide groups are only present in about every second to fifth (batch variations) heptazine ring in K-PHI. Assuming only terminating cyanamide groups at the edges of the crystallites, especially for large crystals in the range of 100 nm, the expected NCN content would be lower than the observed amount. However, additional cyanamide could have been introduced as point defects within the PHI sheet. Such a point defect is equivalent to a missing heptazine unit, leveraging additional internal NCN termini, since the former NH bridges become terminating NCN groups inside the PHI layer. As discussed in more detail in the Supporting Information (SI) ( Figure S9), the concentration of these defects has been estimated to be approx. 5−12%. The number of NCN groups seems to be reduced upon the transformation to H-PHI, which can be attributed to partial hydrolysis of the functional group, resulting in an increased number of terminal NH 2 groups or urea terminations. 15 N CP MAS NMR of K-PHI confirms the presence of NH groups at −242 ppm (N2, inset Figure 2b) bridging the heptazine units, as well as a small amount of terminal NH 2 groups at −273 ppm (N1) (note that the intensity of the NH 2 CP signal is grossly overestimated). The 15 N direct excitation spectra ( Figure S11) show negligible (below detection limit) amounts of these NH 2 groups. The central nitrogen atom at −229 ppm (N3) has a very low intensity in the CP experiments, as the polarization transfer is very poor. This assignment is confirmed by the 13 C− 15 N 2D correlation experiment ( Figure 2g) in which the central nitrogen is only connected to the carbon at 157 ppm (C1). The peripheral nitrogen atoms at −160 to −200 ppm (N4) lie in the typical range of shifts known from heptazine-derived molecules. 24 Interestingly, the best agreement with our quantum-chemical models is found by assuming the "naked" anionic N-bridge to be screened by a potassium ion (Figures S23). In this case, calculations suggest that the signal of the anionic bridge (N5) lies at −173 ppm and is thus beneath the shifts of the heptazine core. Placing the potassium ion in the center of the pore ( Figure S24), a shift at >−140 ppm is calculated for the bridging nitrogen atom, which is not found in our measurements. Thus, our data suggest that K + , presumably with a partial hydration shell, is located off-center and moved toward the polymeric backbone.

Chemistry of Materials
Although the structure of the acid-treated polymer H-PHI is largely similar, some distinct changes in the NMR spectra accompany the transformation. The NMR data of H-PHI support the assumption that protonation occurs predominantly at formerly negatively charged bridging nitrogen atoms (H a , inset Figure 2d). Due to the conjugated charge in the network, however, protonation may also occur directly at the heptazine ring. Quantum-chemical calculations have identified H b and H c (inset Figure 2d) as potential protonation sites. The corresponding calculated chemical shift of these ring protonations are −247 and −261 ppm, respectively, which is in agreement with the upfield shift of the NH-group signal in Figure 2e. Molecular heptazine-based compounds such as cyameluric acid or protonated calcium melonate support this possibility, as they also bear the proton at the heptazine ring. 26,27 However, the overall low absolute amount of NH sites at the ring, along with the coexistence of different protonation sites and expected similar chemical shifts (see calculated pore models in Figures S18, S19, and S22) for this group compared with the bridging NH, does not allow for an unambiguous resolution of the protonation sites.
In direct excitation 13 C NMR of protonated samples (H-PHI), the previously visible signal C3 next to a deprotonated bridge now disappears, hinting that this nitrogen has been protonated, which would result in a chemical shift similar to that of C2. Indeed, the relative intensity of C2 increases for H-PHI in the directly detected 13 C spectrum. The resonance of C5 at 120 ppm slightly shifts upfield to 117 ppm as compared to that of K-PHI, which is also mapped by our calculations ( Figure S18). The carbon atom next to the protonated NCN group is now located beneath the signal of C2. The 15 N CP MAS experiment yields a picture for H-PHI very similar to that for K-PHI. However, the NH group at −242 ppm is broadened toward higher fields in H-PHI and the central nitrogen in the heptazine unit is slightly upfield-shifted to −233 ppm as well. Additionally, a shoulder of the carbon signal at 157 ppm toward higher fields is also observed. These small changes could be related to interactions with protons of water in close proximity to nitrogen atoms on the s-heptazine ring and the possibility of NH located at the ring.
The 1 H NMR measurements of K-PHI and H-PHI ( Figure  2c,f) help to give a clearer picture on the role of water and protons within those networks. Both spectra exhibit an intense signal at about 4.5 ppm, which is assigned to relatively free water inside the pores. Several additional overlapping peaks can be observed between 5 and 12.3 ppm for H-PHI. From the 1 H− 14 N 2D correlation spectra, the proton of the bridging NH can be assigned to a chemical shift of about 10.4 ppm for K-PHI ( Figure S14b) and 12.3 ppm for H-PHI (Figure 2h), confirming that the majority of the bridging nitrogen atoms are protonated. Note that this low-field shifted resonance varies with the amount of water present in the sample. For a "dry" H-PHI sample (vacuum-dried at 120°C for 2 days), this peak shifts upfield to about 7 ppm, as shown by its 1 H− 14 N 2D HETCOR spectrum ( Figure S14a). This is in agreement with the calculations, hinting at water as a factor that significantly influences the proton NMR shifts. A calculated model pore with one water molecule interacting with an imide bridge leads to a shift of the NH proton from 8 ppm (without water) to roughly 11 ppm (with water), which is close to the experimental results (cf. Figures S19 and S20). In H-PHI, this proton signal is downfield-shifted to 12.3 ppm compared to that of the potassium-containing anionic network provides both a higher resolution for 1 H and information on proximity between these protons. On the diagonal we observe the self-correlation peak of free water at 4.5 ppm and also the bridging NH self-correlation, although this second one is less intense due to a large distance between the NH units. In addition, we observe a quite intense correlation between one 1 H at 7.2 and one at 9.0 ppm. From the DQ−SQ experiment acquired with different recoupling times ( Figure S15), these two protons are close to each other (their distance is similar to the 1 H− 1 H distance in water). This would hint at a water molecule with each of its two protons creating hydrogen bonds to nitrogen atoms on the heptazine ring. Similar protons are also observed in the case of K-PHI ( Figure S16), i.e., also in close contact to the network. In general, ssNMR spectroscopy, in combination with our calculated pore models, corroborates the existence of a heptazine-based interconnected 2D network for K-PHI and H-PHI partially functionalized by NCN groups. In both materials, the network strongly interacts with water and hydrated potassium, respectively. This interplay represents a key factor for explaining the three-dimensional (3D) structure, as presented in the following sections. X-Ray Powder Diffraction and PDF Analysis. XRPD patterns show significant peak broadening for H-PHI, whereas K-PHI exhibits higher crystallinity and was therefore used as the starting point for modeling (Figure 3a). In line with the TEM data, the Bragg peaks could be indexed with a triclinic unit cell, which was confirmed by Pawley refinement 28 to be pseudohexagonal with a = b = 12.78 Å, c = 4.31 Å, α = β = 109.63°and γ = 120°, R wp = 3.87%. The structure model for the Rietveld refinement 29 (Figure 3b) was based upon the previously solved structure of PTI, reimagined for PHI sheets. 11,12 The cell metric indicates a single layer per cell, which is stacked in a slip-stacked fashion, given the non-90°α/ β angles. A planar heptazine dimer was defined as a rigid body and oriented in the ab-plane, forming large triangular pores with two heptazine units on each side and an interlayer distance of approximately 3.2 Å, in agreement with the prominent 001 stacking reflection at ca. 28°2θ. Reasonable relative peak intensities could only be obtained by including K and O atoms (i.e., water molecules) representing the pore content, indicating that the pores are indeed filled. Further H-PHI appears to be structurally related through welldefined 110, 010/100, 001, and 321/231 reflections and could be indexed by Pawley refinement with a similar cell ( Figure  S29). However, a suitable Rietveld refinement for H-PHI was not possible due to significant peak broadening. Since NMR and TEM data suggest an intact layer constitution of H-PHI, this broadening is attributed to planar defects occurring as stacking fault disorder that developed during K + removal, which is discussed in the next section.
To gain further insights into the local structure, PDFs ( Figures S32 and S33) were obtained from synchrotron X-ray total scattering data for H-PHI and K-PHI, in Figure 3c. A direct comparison shows highly similar local structures in both compounds. The heptazine units are present and conformationally rigid. Broad, interlayer peaks repeating with a wavelength of ∼3.19 Å for K-PHI and ∼3.17 Å for H-PHI indicate that the sheets are stacked with high fidelity, on the order of at least several hundred Angstroms, shown in Figures  S32 and S33. The lack of longer wavelength modulations expected from periodic porous channels indicates that the pores are populated by uncorrelated electron density. Small differences in medium-range, low-amplitude features indicate possible differences in local ordering of pore content and neighboring layers, whereas more drastic long-range deviation in the behavior of the interlayer peaks provides additional evidence of differences in the stacking behavior. Model refinements to the K-PHI and H-PHI PDFs (Figures S34− S36) were consistent with the results from Rietveld refinements, as shown in Figures S30 and S31. In terms of shortrange ordering, the most notable difference between H-PHI and K-PHI is the increase in peak intensity around 2.7−3.3 Å, which correlates with an increase in K + ( Figure S37) concentration. This feature is connected to potassium−oxygen distances found in water-solvated potassium ions observed via MD simulations ( Figure S25), providing direct evidence of water−K + coordination within the pores. Notably, in all Rietveld and PDF refinements, K + ions consistently moved off center, in-plane, toward one corner of the triangular pore. Complexation of K + ions with azines has previously been shown favorable (planar orientation/similar bond distances). 30 It is reasonable that K + could occupy both fully watercoordinated and partial nitrogen-coordinated positions.  Both TEM and XRPD data are consistent with the comparably high crystallinity of the carbon nitride backbone. The overall high in-plane order likely results from the use of molecular precursors that have a higher mobility in the salt melt, which serves as a source for error correction during bond formation. The fact that the reflections in the XRPD pattern of H-PHI are particularly broadened will be discussed in the next section.

Chemistry of Materials
Stacking Fault Simulations. Information on the 3D structure of the bulk PHI is difficult to extract from electron diffraction patterns due to the local nature of the method and the pronounced 2D morphology of the crystallites. The structural coherence of the PHI materials in the third dimension is governed by the stacking order of the layers and in the case of H-PHI is vastly affected by planar defects occurring as stacking faults. Thus, systematic simulations 31 of distinct faulting scenarios have been performed and compared qualitatively to the measured XRPD pattern, which yields information on all three dimensions of the bulk material. H-PHI in Figure 4a shows significant peak broadening and triangular, Warren-type line shapes, 32 which are indicative of planar defects in the crystal structure. We therefore conclude that H-PHI has a higher degree of out-of-plane disorder compared with K-PHI.
In Figure 4b, possible interactions of pore water with the poly(heptazine imide) backbone of H-PHI are depicted. These interactions direct the stacking of the layers and induce a layer offset; i.e., a layer is shifted toward the pore of the preceding layer. Due to the intrinsic trigonal symmetry of the heptazine subunit, this type of stacking exhibits three equivalent directions (Figure 4c). The eclipsed case was also considered for the simulations, i.e., layers stacked directly atop each other. Different faulting scenarios were developed using these four stacking relationships and used for qualitative simulations on the diffractional effect of structural disorder (detailed information is given in chapter 12 of the SI). The faulting scenario ("Faulting scenario IV", Figures S47−S49) that was evaluated as the most suitable based on the match between simulated and measured patterns describes completely random transitions between all three pore directed stackings. This leads to a loss in the orientational coherence perpendicular to the layer planes and therefore to vast broadening of most reflections in the XRPD pattern in Figure 4d. Small sections of homogenously stacked layers are, however, still apparent, as indicated by the presence of the remaining 010 reflection.
The combined approach of PDF and XRPD along with stacking fault simulations reveals the overall 3D structures of H-PHI and K-PHI. It has been shown that the stacking of the pores is governed by rather weak pore−water interactions in the case of H-PHI, resulting in highly defective stacking. Nevertheless, pore water clearly mediates interlayer interaction. When the water molecules are removed by heating H-PHI up to 400°C under an inert, dynamic atmosphere in a TG-experiment, an ex situ XRPD analysis of the residue ( Figure S27) reveals a significant decrease in the crystallite size perpendicular to the layer planes. The complete removal of pore water also shifts the layer arrangement toward an eclipsed stacking order. The presence of potassium within the pores leads to a more ordered stacking, which is indicated by much sharper reflections in the corresponding XRPD pattern (see Figure 3a). Therefore, the range of the interlayer interaction is much larger in K-PHI than in H-PHI. This can be attributed both to the large ionic radius of the potassium cation and its more isolated positive charge. As potassium is located offcenter in the pores, it interacts with the PHI framework. In addition, a partial hydration sphere is formed around the cation by the pore water molecules ( Figure S25), which may indirectly mediate interactions to further distant layers.
Photocatalysis. Knowledge of the structure−property relationships in this class of carbon nitrides is key for the rational design of photocatalysts with improved properties. We therefore investigated the photocatalytic activity of H-PHI and K-PHI, obtained from ampoule synthesis, for the hydrogen evolution reaction (HER) with a focus on carving out those

Chemistry of Materials
Article structural features that affect the activity the most. We find that H-PHI, like K-PHI, is active toward HER in the presence of a Pt cocatalyst and methanol as the sacrificial electron donor but shows only moderate activity (H-PHI: 88 μmol/gh, K-PHI: 193 μmol/gh). This can be directly traced back to the large crystallites that tend to form aggregates that disperse only weakly in aqueous solution, thus having less accessible surface area and a reduced number of edge sites available. The latter seems to be vital for the catalytic activity 33 for cocatalystsupported systems through the presence of cyanamide terminations at the edges of the platelets. These terminations can bind more efficiently to the Pt cocatalyst and provide a higher degree of ionicity and thus colloidal stability. 2 To test this hypothesis, we synthesized variants of H-PHI and K-PHI with smaller crystallite sizes (∼20 nm) and an increased amount of external cyanamide groups accordingly, by using a mixture of melon and KSCN as precursors at ambient pressure under an argon flow, followed by treatment in diluted HCl. For the as-obtained, optimized K-PHI, we recorded a 3fold increased activity of 600 μmol/gh compared to that of the highly crystalline sample, whereas we registered a 38-time increase in activity for H-PHI (3364 μmol/gh, Figure 5b). This is well beyond the activity of the prototypical carbon nitride melon with an activity of just 25 μmol/gh measured under the same conditions. Note, however, that a quantitative comparison of the intrinsic photocatalytic activities of K-PHI and H-PHI is difficult since the observed hydrogen evolution rates depend on a complex interplay between various parameters, which, in addition, may scale differently with the particle size. These include exfoliation and agglomeration of the crystallites, the nature and amount of defects, or the amount of cocatalyst incorporated in the sample.
For a more detailed structural characterization of these optimized compounds, see Figures S7 and S8. It is apparent that the optimized materials exhibit lower long-range in-plane order, as indicated by broad, diffuse reflections. In particular, both 110 and the 010 reflections are vastly broadened, which is attributed to a loss of coherence within the layer plane, i.e., smaller platelets ( Figure S5). By using the Scherrer equation from the XRPD data in Figure S8, the crystallite sizes in the lateral dimension have been estimated to be about 20 nm for the optimized samples and up to 100 nm (Table S1) for the most crystalline samples. The extension in c-direction remains fairly similar for all samples at approx. 20 nm. Small crystallites have a higher amount of accessible cyanamide groups at their surface and are suspended more easily in water, enabling them to participate more efficiently in the photocatalytic reaction.
While the HER activity observed for H-PHI puts it on par with the best carbon nitride photocatalysts, 34−39 it bears another intriguing property, which has previously been documented for K-PHI; 4,6 under illumination and in the presence of an electron donor, H-PHI also forms a photoreduced state that is exceptionally long-living, clearly visible by its blue color. Figure 5a depicts the suspension during photoreduction, which is yellow initially and then turns dark blue upon irradiation in the presence of 4-methyl benzylalcohol as the electron donor. Under an argon atmosphere, the photoexcited photocatalyst is reduced by the electron donor and remains in the photoreduced state for over 200 h. The diminishing of the blue color is due to quenching of the anion radical by oxygen, which slowly diffuses into the solution over time (Figure 5a, right). The photoreduced radical state was also monitored by electron magnetic resonance spectroscopy (EPR), which shows a photoinduced paramagnetic signal for both K-PHI and H-PHI with g-factors of ≈2.003 ( Figure S10), indicative of a heptazine-centered π radical. 4,36,40,41 To test whether photoreduced H-PHI is able to store the trapped electrons and release them upon addition of a hydrogen evolution catalyst after a time delay, 15 mg of the sample was irradiated under a Xenon lamp (100 mW/cm 2 , AM 1.5 G) in phosphate buffer with 4-methylbenzyl alcohol as an electron donor for 2 h. Then, a colloidal platinum suspension was added in the dark (Figure 5c) either directly or 2, 4, or 6 h after illumination. Indeed, hydrogen was evolved for several hours (27 μmol/g, direct addition), which shows the ability of the system to separate the light (photoexcitation) from the dark reaction (catalysis) akin to the biological model of photosynthesis. Even after waiting for 6 h, 37% of the initial amount of H 2 could still be registered.

■ CONCLUSIONS
In conclusion, we have synthesized a 2D poly(heptazine imide) (PHI) polymer, which is comparable with PTI · LiCl 11,12 in terms of crystallinity but by far exceeds its photocatalytic performance. 13,42 Characterization of K-PHI and H-PHI provides insights into the structure−property− activity relationship of this new class of high-performance photocatalysts. By combining a suite of analytical techniques, including TEM, NMR spectroscopy, XRPD, PDF analysis, and quantum-chemical calculations, we were able to confirm the inplane PHI-based 2D structure and derive a detailed picture of the out-of-plane structure. Our analysis not only suggests that ionothermal synthesis conditions likely support error correction, thus leading to extended crystallites of over 100 nm in lateral dimension, but also that metal ions can act as out-ofplane structure-directing agents, leading to a pronounced longrange order in the c-direction. Furthermore, water inside the pores has been found to have a major influence on the 3D stacking structure of PHI networks. Our work thus adds another structurally characterized member to the large yet illdefined class of carbon nitrides, which is notoriously difficult to characterize. As such, this study provides synthetic and Figure 5. (a) H-PHI before and after 1 h of illumination and after storing the material in the stable, photoreduced "blue" state up to 208 h in the dark. A shade toward green is likely due to small amounts of oxygen leaking into the vessel. (b) Comparison of photocatalytic rates for hydrogen evolution for K-PHI and H-PHI, respectively. (c) Dark photocatalysis in phosphate buffer with various delay times for addition of platinum cocatalyst, i.e., 0, 2, 4, or 6 h. analytical tools that are crucial for the rational design of better photocatalysts and helps to carve out the structural features that are essential for observing an unusually long-lived charge separated state, enabling time-delayed photocatalysis in the dark.
Crystalline Ampoule Synthesis of Protonated Poly-(Heptazine Imide), H-PHI. Finely ground cryst. K-PHI (100 mg) was treated with 100 mL of 2 M HCl and stirred for 15 min. The suspension was filtered and the residue washed four times with DI water to yield H-PHI.
Further details on the synthesis of optimized samples for photocatalysis, melon, and potassium melonate can be found in Supporting Information, chapter 1.