Light-induced Ammonia Generation over Defective Carbon Nitride Modified with Pyrite

Photocatalytic nitrogen fixation under ambient conditions is currently widely explored in an attempt to develop a sustainable alternative for the Haber-Bosch process. In this work we combine defect-rich C 3 N 4 , one of the most investigated photocatalysts reported in literature for ammonia generation, with earth-abundant and bioinspired FeS 2 to improve the activity for ammonia production. By this combination, an activity enhancement of approx. 400 % compared to unmodified C 3 N 4 was achieved. The optimal FeS 2 loading was established to be 1 wt.%, with ammonia yields of up to 800 µg L -1 after irradiation for 7 hours. By detailed material characterization of the electronic and material properties of the composites before and after the


Introduction
The Haber-Bosch process is a large-scale industrial process for the production of ammonia (NH3) from hydrogen (H2) and nitrogen (N2) gas at elevated pressures and temperatures.It is a wellestablished and optimized process, that is crucial for the production of fertilizers.Nevertheless, it still suffers from sustainability issues due to high energy requirements and the utilization of natural gas for the supply of H2 in large, centralized power plants. [1]The photocatalytic nitrogen reduction reaction (NRR), that directly converts N2 into NH3 under (sun)-light irradiation at ambient conditions, presents a feasible alternative. [2,3]However, the NRR is a thermodynamically and kinetically unfavorable process due to the stable and inert nature of the N2 molecule.Furthermore, the similar potentials for N2 reduction and H2 evolution impede a high selectivity in aqueous media so far.This imposes strident demands on an efficient and selective photocatalyst, that are not even close to be met even by the state-of-the art materials, thus highlighting the importance of continued research on the topic. [2,4,5]The low NH3 yields further pose additional difficulties when it comes to accurate quantification, necessitating careful experiment control to avoid impurities and erroneous results.9] Carbon nitrides like polyheptazine imide, C3N4, are low cost, non-toxic polymeric materials with good light absorption characteristics, due to a band gap of approx.2.7 eV, which have received a lot of attention as an n-type photocatalyst during the past decade. [10,11]C3N4 is commonly synthesized by thermal polymerization of organic precursor molecules, such as urea, melamine or cyanamide.The polymerization conditions, as well as the choice of the precursors have a strong effect on its structure and properties, by influencing the defect concentration, degree of polymerization, optical band gap, and surface area. [12]As-synthesized C3N4 suffers from high recombination rates.5][16][17][18][19][20][21] Some of the highest NH3 yields in photocatalytic NRR have been reported using defective C3N4. [22- 26] wing to the polymeric structure of C3N4, the number of possible defects is vast and versatile. [18,20,21,27,28]A frequently exploited strategy for activity enhancement in C3N4 is the introduction of nitrogen vacancies, which are of the same size as the nitrogen atoms in molecular N2, and can thus act as efficient adsorption and activation centers. [27]other important class of defects are cyano or cyanamide groups, that act as electron-withdrawing groups, assisting in charge separation and suppressing recombination. [14,24,29]The NRR is supposed to proceed via a Mars-van-Krevelen mechanism for both defect types, making a distinction between the influence of cyano groups and nitrogen vacancies difficult.Intercalation of potassium was reported to assist in the replenishment of nitrogen in the structure, more specifically of the cyano groups, resulting in an NH3 production rate of 3.42 mmol g -1 h -1 . [25]Treatment of bulk C3N4 with KOH, or direct incorporation of KOH into the synthesis recently emerged as a promising strategy for both the introduction of vacancies and of cyano groups.Zhou et al. treated bulk C3N4 with KOH in ethanol, followed by solvent evaporation and annealing.They observed an abundance of cyano groups that assisted in charge separation and N2 adsorption. [24]KOH has also directly been involved in the thermal polymerization of urea, likewise promoting the formation of cyano groups. [14]A similar result was obtained by Wang et al., using KOH in the thermal polymerization of dicyandiamide. [25]Etching with KOH was reported to mostly lead to the formation of vacancies.
Still high NH3 yields of 3.632 mmol g -1 h -1 were reported, together with a quantum efficiency of over 20 %. [22] Apart from defect engineering, research on semiconductor materials for artificial photosynthesis is still in its infancy compared to the process optimization in nature.The reduction of molecular N2 has been realized under ambient conditions at the active centers of selected enzymes, termed nitrogenases.Three different classes of nitrogenases are distinguished, based on the composition of the active centers: all contain sulfur and iron atoms, but differ in the nature of additional constituents, one containing vanadium, one molybdenum and one solely iron. [30][33][34][35][36][37][38][39] One of the simplest iron and sulfur containing compounds is FeS2, which is a non-toxic, stable, and earthabundant mineral.It shows very high optical absorption and high charge carrier mobility, but its use in photoelectrochemical applications suffers from charge trapping and recombination. [40][43] By itself it is reported to be inactive in photocatalytic NRR, due to unsuitable band positions, but it might still offer beneficial contribution in combination with other materials. [44,45][48][49][50][51] Apart from improving the light absorption properties in composites, FeS2 can act as an efficient electrocatalyst, which has been shown separately for the H2 evolution, as well as for the NRR. [52- 56] here have been also recent reports on the formation of heterojunctions of FeS2 with C3N4 and their photocatalytic application in organic dye and antibiotics removal. [57,58]The reported mechanism and band positions on which the proposed electron transfers are based, are notably different between the reports, requiring further investigation of the interaction between both materials.
In a bioinspired approach, we herein report the combination of the two earth-abundant semiconducting materials FeS2 and defective C3N4 for light-induced NH3 generation.Exploiting the good light absorption and N2 activation characteristics of the Fe-S system together with the reported high activity of defective C3N4, NH3 yields of up to 800 µg L -1 can be achieved in the course of the reaction (7 h), which equals to an activity enhancement by around 400 % compared to bulk C3N4.By detailed characterization of the electronic and material properties of the composites before and after reaction, we show that FeS2 decoration weakens the bonds of C3N4terminating imine groups in the vicinity of nitrogen defects by back donation.This facilitates the reduction of terminal cyano groups under light irradiation towards NH3, with H2 being the only byproduct.

Material Synthesis and Characterization
C3N4 was prepared via thermal polymerization of melamine. [59]1 g of melamine (Sigma Aldrich, 99%) was calcined at 550 °C for 4 h in a closed crucible in air, using a heating ramp of 5 K/min.
The synthesis was repeated several times and the obtained C3N4 powder was ground and thoroughly mixed, before it was used for further modifications and composite formation.
Vacancy-rich C3N4 (VN-C3N4) was obtained by dispersing 2 g of the as-synthesized C3N4 in 36 mL of 1 M KOH for 3 h under stirring.Subsequently, the material was collected via centrifugation and washed until neutral with ultrapure water. [22]the composite formation, respective amounts of commercial FeS2 (Sigma Aldrich, 99.8%, 325 mesh) and VN-C3N4 were ground for 10 min in a mortar, under addition of low amounts of ipropanol (p.a.).Subsequently, the mixture was subjected to heat treatment for 2 h at 200 °C in air, to improve interfacial contact.
The composites were analyzed before and after the photocatalytic experiments.Powder X-ray diffraction (XRD) was measured on a Malvern PANalytical Empyrean device with Cu Kα irradiation (λ1 = 1.5406Å; λ2 = 1.54443Å).Acceleration voltage and emission current were set to 40 kV and 40 mA, respectively.Peak assignment was performed with X'Pert Highscore plus.
Following reference cards were used for the reflection assignment in FeS2 and FeS, respectively: 00-042-1340, 00-023-1123.The diffraction pattern for C3N4 was calculated with Vesta, using crystallographic data from the group of Irvine. [60]Diffuse-reflectance UV/vis spectra were obtained using a Perkin Elmer Lambda 750 spectrometer with a Praying Mantis (Harrick) and spectralon as white standard.The Kubelka-Munk function was used for the calculation of pseudo-absorption, f(R). [61]() = (1−) 2
For diffuse reflectance infrared Fourier transformed spectroscopy (DRIFT) a Bruker Alpha II spectrometer and the software OPUS were used.Sample scans were taken from 400 to 4000 cm -1 , with a resolution of 4 cm -1 .
Fluorescence measurements were conducted on a FluoTime 300 spectrometer from PicoQuant, with the software EasyTau2.Emission spectra were recorded at different excitation wavelengths from a 300 W Xe lamp at room temperature in air.Time correlated single photon counting (TCSPC) spectra were measured using 355 nm laser excitation.The software EasyTau2 was employed for fitting of the decay curves, using a tailfit with three exponentials, according to:

𝑖=1
For steady state measurements, the sample was placed in a holder for solid powder samples.For quantum yield (QY) measurements a thin film on the inside of a cuvette was prepared.The cuvette was placed in an integrating sphere.Measurements were conducted for "out" geometrymeaning that the film was positioned outside of the direct excitation path.For the calculation of the quantum yield, the intensity of the fluorescence emission was integrated, and the area AS was divided by the total integral excitation intensity as measured in an empty reference cuvette (ABE) minus the excitation intensity that is not absorbed by the sample, ASE.

𝑄𝑌 = 𝐴 𝑆 𝐴 𝐵𝐸 − 𝐴 𝑆𝐸
X-ray photoelectron spectroscopy (XPS) was performed with a Physical Electronics PHI VersaProbe III Scanning XPS Microprobe device.Monochromatic Al Kα X-ray irradiation with a beam diameter of 100 µm was used, with the beam voltage being set to 15 kV and x-ray power to 25 W. The sample surface was pre-cleaned by argon cluster sputtering with a gas cluster ion-beam.
To avoid surface charging, samples were continuously flooded with slow-moving electrons and Ar + .For survey scans, pass energy and step size were set to 224 eV and 0.4 eV, respectively.Highresolution spectra were measured with a pass energy of 26 eV, a step size of 0.1 eV and a step time of 50 ms.For data analysis a CASA XPS 2.3.17software was used.The background was corrected using Shirley subtraction.Peak fitting was done with Gaussian-Lorentzian line shapes, with 30 % Lorentz ratio.For charge correction C 1s was set to 284.8 eV.
Transient absorption spectroscopy data was collected in diffuse reflectance geometry with an LP980 spectrometer (Edinburgh instruments).Pump laser pulse excitation was set at 355 nm (3 rd harmonic of an Nd:YAG laser produced by Ekspla, NT340), while for the probe pulse a 150 W xenon arc lamp was used.Prior to the measurements the powder sample was filled in a cuvette, stored under Argon and sealed directly before the measurement.TAS data was normalized and decays were fitted with exponential decay functions.
N2 physisorption measurements were conducted on a Quadrasorb Evo device from Anton Paar QuantaTec at 77 K to determine Brunauer-Emmet-Teller (BET) surface areas, using the software ASiQwin for data evaluation.Samples were degassed for 12 h at 120 °C prior to measurements.
Due to the small surface area, Kr at 77 K was used for FeS2, in an AS-iQ-MP-MP-AG instrument from Anton Paar QuantaTec.
CHNS elemental analysis was performed with an Unicube instrument from Elementar, using sulfanilamide as standard.Approximately 2 mg of the respective sample were weighed into a tin boat, sealed and combusted at temperatures up to 1143 °C in an oxygen/ argon atmosphere.
Thermogravimetric Analysis (TGA) with gas evolution detection via mass spectrometry (MS) was conducted with a Netzsch Jupiter STA 449C thermobalance together with a Netzsch Aeolos QMS 403C quadrupole MS, heating the sample at a rate of 5 K/min up to 900 °C in synthetic air.
Scanning electron microscopy (SEM) images were recorded on a Zeiss Leo 1530 device with an acceleration voltage of 3 kV after sputter-coating with platinum (Cressington Sputter Coater 208 HR).Energy dispersive X-ray diffraction spectroscopy (EDX) measurements were conducted on the same instrument, using and an acceleration voltage of 20 kV.An ultra-dry EDX detector by Thermo Fisher Scientific NS7 was employed.

Photocatalysis
Light-induced ammonia generation was performed in a flow setup using a doped Hg immersion lamp (Z4, 700 W Peschl Ultraviolet) placed in a water-cooled quartz-glass inlay and operated at 350 W. 200 mg of the photocatalyst were dispersed in approx.30 mL of water by ultrasonic treatment for 10 min.The dispersion was transferred to the glass reactor and diluted to 600 mL by the addition of water and methanol.The total amount of methanol was 20 vol.%.Nitrogen was bubbled through the stirred dispersion at a flow rate of 50 mL/min overnight to flush out residual air.For pre-purification of the inlet gas stream, it was first passed through a 0.1 M KMnO4 solution, followed by a 0.1 M KOH. [8]The dispersion was illuminated for 7 h, during which a constant temperature of 10 °C was ensured by cooling the reactor with the help of a cryostat (Lauda Proline RP845).Evolving gasses were passed through an acid trap containing 10 mL of 1 mM H2SO4, [63] dried and subsequently analysed by a quadrupole mass spectrometer (HPR-20 Q/C, Hiden Analytical).After the reaction, the dispersion was immediately filtered and tested for NH3 using the salicylate test method, a modification of the indophenol blue method. [64,65]r the salicylate test, a stock solution of sodium hypochlorite and a stock solution containing the catalyst, sodium nitroprusside (Carl Roth, >99%), and sodium salicylate (Carl Roth, >99%), were prepared.The solutions were prepared fresh weekly and stored at 4 °C in the dark.For the preparation of the salicylate/ catalyst solution, 2 g of sodium salicylate and 8 mg of sodium nitroprusside were dissolved in 15 mL of ultrapure deionized water, to which 5 mL of a 2 M sodium hydroxide solution was added.For the preparation of the hypochlorite solution, 200 μL of sodium hypochlorite solution (12 % Cl, Carl Roth) and 1 mL of 2 M NaOH were given to 18.8 mL of water.In a typical testing procedure, 500 μL of the hypochlorite solution were given to 2 mL of the reaction solution, which was filtered through a 0.2 µm syringe filter beforehand.Then, 500 μL of the sodium salicylate solution were added.The mixture was stored in the dark at room temperature overnight for color development, before being analyzed by UV/vis spectroscopy (Perkin Elmer Lambda 750 spectrometer), using a mixture of the two testing solutions and 20 % aqueous methanol as reference.For the calibration, ammonium chloride (Carl Roth, >99.7%) stock solutions were prepared in a concentration range from 0.1 μg/L to 10 mg/L of NH4 + .Since the experiment was performed in the presence of methanol as a scavenger, 20 % of methanol were present in the calibration as well.
To verify the results from the salicylate test, the filtered reaction solutions were additionally analyzed with ion chromatography (IC).A Dionex Aquion System from Thermo Fisher, equipped with a Dionex IonPac CS16 column with a CG16 guard column, CERS suppressor, and electric conductivity detector.30 mM methanesulfonic acid was used as eluent.
Ion chromatography was also used to analyze the reaction solution for nitrate by-products.The reaction solution was filtered with a 0.2 µm syringe filter.A Dionex Aquion system from Thermo Fisher, equipped with a Dionex IonPac AS9-HC column and IonPac AG9-HC guard column was used for the analysis.1 mM NaHCO3/ 8 mM Na2CO3 was used as eluent, a UV detector was employed for quantification at a wavelength of 207 nm.
For the quantification of hydrazine, the colorimetric method first reported by Watt and Chrisp was employed. [66]For the testing solution, 0.4 g of p-dimethylaminobenzaldehyde (Sigma Aldrich, 99%) were dissolved in 20 mL of ethanol (p.a.), to which 2 mL of concentrated HCl were added.
For the calibration curve, standard solutions of hydrazine sulfate (Sigma Aldrich, >99%) in water/ methanol mixtures were prepared.For the measurement, 1.5 mL of the filtered reaction solutions were mixed with 1.5 mL of the testing solution and stored in the dark for 20 min for color development, before analysis of the absorbance with UV/vis spectroscopy against a reference containing only the testing solution and a water/ methanol mixture.

Material Synthesis and Characterization
The X-ray diffraction (XRD) patterns of the KOH-treated C3N4 (VN-C3N4) show the typical two broad reflections at 13° 2θ and at 27.4° 2θ, corresponding to in-plane order and interplanar stacking, respectively, with a d-spacing of approx.0.326 nm (Figure S2). [60]The structure of C3N4 obtained via thermal polymerization is best described by a model of parallel melon chains connected via hydrogen bonds, as found by Lotsch et al. and confirmed in later studies.[69] This structure model will be assumed in the following.Powder XRD patterns of the composites show the reflections for FeS2 and for VN-C3N4 (KOH-etched, vacancy-rich).The intensity of the reflections of FeS2 increases with increasing FeS2 amounts in the composites (Figure 1).No additional phases could be observed.
Enlarged and normalized PXRD patterns of VN-C3N4 and the composites (Figure S2) show a shift of the (002) reflection at around 27 °2θ towards higher °2θ values compared to untreated C3N4.
The shift and a decrease in the intensity upon KOH treatment can be assigned to a decrease in interplanar stacking distance and a general loss of order, which has been ascribed to either introduced cyano groups or nitrogen vacancies. [13,70,71] further characterization of the morphology, SEM images and EDX maps were recorded.They show FeS2 particles in the µm range, distributed all over the VN-C3N4 matrix (Figure 2 and Figure S3).EDX analysis confirmed the particles to be FeS2 with a ratio close to the ideal value of Fe:S 1:2 (Table S1) and the surrounding matrix to consist of C3N4 with a C/N ratio of approx.0.52.This is lower than the ideal value of 0.75 and could be an indication of free amino groups in the sample, although it is also effected by the low sensitivity of EDX for light-weight atoms.Physisorption measurements were conducted to evaluate the surface area of the composites with the BET model (Table S2).Pristine C3N4 exhibits the highest surface area of 10.6 m 2 /g, which slightly decreases upon KOH treatment.This is in good agreement with the decreased interlayer stacking distance observed in the XRD patterns and otherwise retention of the morphology.The apparent further decrease of the specific surface area upon composite formation can be explained by a difference in material density and particle size between C3N4 and FeS2.
The optical properties of a photocatalyst are of utmost importance, since they vastly determine the efficiency in light harvesting.The color of the composites gradually gained a tinge of grey with increased FeS2 content, compared to the previously pale yellow coloring of both untreated C3N4 and VN-C3N4 (Figure S1).Diffuse reflectance UV/vis measurements were conducted to elucidate the effect of FeS2 addition on the absorption behavior (Figure 3).C3N4 is an indirect n-type semiconductor. [72,73]Therefore, an indirect Tauc plot was used for a more accurate determination of the band gap and compared to the values apparent in the Kubelka-Munk plots.KOH treatment results in a slight decrease of the band gap of C3N4 from 2.73 to 2.70 eV, and marginally improved absorption in the UV region.The band gap reduction -and thus red-shifted absorption -could be caused by the introduction of cyano groups, whose electron-withdrawing properties were reported to lower the conduction band edge and lead to a narrowing of the band gap. [20,74]The increased UV absorption might be caused by improved charge separation due to the decreased layer distance and the introduction of cyano groups, since transitions in the UV region are commonly ascribed to ππ* transitions in sp 2 hybridized centers of the aromatic system. [25,75]e composites exhibit essentially all the same band gap, which is an indication that the major contribution to the light absorption is given by C3N4.This is expected, since its concentration is much higher than that of FeS2.The band gap of the composites is slightly increased but very comparable to VN-C3N4 with a change from 2.70 eV to 2.78 eV (Figure 3).The UV absorption of the composites is increased compared to VN-C3N4.Both effects hint at a change in the electronic structure and availability of electrons in the π-system.The color change and increased absorption of visible light upon addition of FeS2 is reflected by diffuse absorption at higher wavelengths, visible in an offset of the baseline (Figure S5).The effect of band gap widening is especially pronounced for the composites with a FeS2 ratio of 2.5 to 10 wt.% which might indicate optimal charge separation at medium FeS2 loading.All band gaps derived from the Kubelka-Munk and Tauc plots are summarized in Table S3.DRIFT spectra were recorded to further elucidate possible structure changes upon composite formation (Figure 3).Comparable DRIFT spectra of C3N4 and VN-C3N4 were recorded, where the broad signal between 3000 and 3600 cm -1 can be assigned to O-H and N-H stretching vibrations, underlining the presence of free amino groups that in turn indicate only partial polymerization.The sharp peaks between 1700 and 1200 cm -1 belong to stretching modes of C=N and C-N in the heterocycles, as well as bridging units, and the band at approx.808 cm -1 can be assigned to the breathing mode of the s-triazine units. [24,76,77]Additionally, a band at 2150 cm -1 could be observed, that is ascribed to the presence of cyano groups, that appear to be present in C3N4 and VN-C3N4 as well as in the composites to varying extent. [20,24,77]KOH treatment increases the amount of cyano groups, as well as -OH and/ or -NHx groups.Additionally, the signals arising from stretching vibrations in the heptazine units are of slightly lower intensity for the defective C3N4, (relative to the signal at 1720 cm -1 ), as is the one for deformation vibrations at 808 cm -1 (Figure S6).This is an indication of structural damage inflicted on the heptazine framework, as would be expected by the formation of defects.
The spectra for the composites are fairly similar to that of VN-C3N4, indicating retention of the structure.An increased absorbance for the vibration of the heptazine units relative to that at 1720 cm -1 was observed for the composites with 10 and 15 % FeS2 in the normalized spectraan opposite effect to that caused by the KOH treatment.The signals for heterocycle-vibrations at higher wavenumbers (closer to 1700 cm -1 ) stem from C=N vibrations, while those at lower wavenumbers (closer to 1100 cm -1 ) arise from C-N vibrations. [78]Therefore, mainly C=N vibrations appear to be affected by the addition of FeS2.We additionally observe a slight shift of the vibrations for the heptazine unit towards lower wavenumbers, that hints at minor changes in the vibration energy of the entire heptazine framework (Figure S6).
To further investigate the structural evolutions upon KOH etching and composite formation, XPS measurements were conducted on bulk C3N4, VN-C3N4 and a composite containing 5 wt.% of FeS2.
Survey scans show the expected signals for carbon, nitrogen and low amounts of oxygen in several spots (Figure S7).The latter is mainly due to adventitious carbon at the surface and not OH-groups in the defective C3N4, since oxygen was also observed in several spots on bulk C3N4.C/N ratios for bulk C3N4, and VN-C3N4, are 0.62 and 0.71, respectively, after correction for adventitious carbon, with slight deviations depending on the measured spot (Table S4). The increased carbon ratio in VN-C3N4 might indicate the introduction of nitrogen vacancies.The C/N ratio in the composite is with 0.71 identical to that of VN-C3N4, underlining that the structure remains intact during composite formation.
Generally, KOH treatment on bulk C3N4 is expected to lead to partial hydrolysis of the structure.
Yu et al. proposed a deprotonation of an apex amine group, during thermal polymerization in the presence of KOH which led to a breaking of the topmost cycle of the heptazine unit and cyano group formation.The overall amount of amino groups is retained here. [77]The reaction conditions notably differ from those employed in this work, but similar structural changes, specifically the introduction of cyano groups in addition to possible vacancy formation, are expected.Nitrogen vacancies are generally believed to be introduced at C-N=C sites. [25]These kinds of possible defects are mainly considered in the following.
Two main peaks are observed in the C 1s spectra of C3N4, corresponding to adventitious carbon at 284.8 eV and N-C=N species in the aromatic system at 288.2 eV (Figure 4).The other peak at 281.6 eV can be assigned to carbon adjacent to amino groups. [24,76]Carbon atoms bound to cyano groups are expected to have a similarthough slightly higherbinding energy, that will overlap with this peak. [68,77]A minor shift from 286.1 eV to 286.2 eV is visible in the spectrum for VN-C3N4, hinting at an increase in the amount of cyano groups, as observed in the DRIFT spectra.This is further supported by changes in the ratios of the respective signal to that of C-C and N-C=N.
Another effect is the observation of a new π-π* satellite at 295.3 eV in case of VN-C3N4 and a growth in the intensity of the π-π* satellite at ~293 eV, indicating the presence of larger numbers of free electrons in the vacancy rich system, or more favorable excitation of these.This is in good agreement to the observations from UV spectroscopy and to literature. [25,80]Furthermore, a reduction of the asymmetry of the peak at 298.2 eV upon KOH-treatment indicates damage inflicted on the aromatic system, which is in good agreement with the assumed breaking of some heptazine units.The formation of a composite of VN-C3N4 and FeS2 results in a comparable C 1s XP spectrum only that the signals of carbon in both the -N-C=N and in the cyano group shift towards higher binding energy (Figure 4 and Table 1).This might indicate electron extraction from the entire heptazine framework, resulting in partial oxidation.Four peaks are identified in the N 1s spectra for bulk C3N4 at approx.398.6 eV, 399.2 eV, 401.4 eV and ~404 eV, in good agreement to the literature (Figure 4). [24,25,76,77]Correlation of the signals with corresponding nitrogen species is challenging, since peak assignment in the literature is ambiguous. [68]The most prominent peak at 398.6 eV in the spectra for bulk C3N4, and VN-C3N4 both can be assigned to -C-N=C species in the heptazine units (marked as N-(C)2 in the following), and a small peak at 404.9 eV corresponds to a π-π* satellite.The signals between 401 eV and 399 eV can be attributed to amino groups and to the nitrogen atom in the middle of a heptazine unit (N-(C)3), although the exact assignment of the signals to the two nitrogen species is discussed controversially. [25,68,76,79]Additionally, the signal for amino nitrogen should further be fitted by two, since the structure is not ideally graphitic, but closer to that of parallel melon chains, which results in the presence of both -NH and -NH2 groups, of which the primary amino group is expected at lower binding energy (Figure 4). [60,68]Cyano groups give rise to a signal at ~400.1 eV, which is indistinguishable from either the signal for the amino groups, or that of N-(C)3. [68]nerally, the nature of the N 1s spectra allows for the possibility of various fits that give good results mathematically but are meaningless in a chemical and physical sense, due to a large number of independent fitting parameters.This is demonstrated in Figure S8.If we assume that the structure of C3N4 obtained via thermal polymerization of melamine lies in between the model of parallel melon chains and of fully condensed sheets, the amount of NH should be smallest, compared to the other nitrogen species. [68]This expectation supports an assignment of the signal at 401.5 eV to NH instead of N-(C)3.Considering, that the signals for primary and secondary amines are likely to be adjacent, we follow assignment of the nitrogen species as: N-(C)2 at 398.7 eV, N-(C)3 at 399.5 eV, NH2 at 400.5 and NH at 401.5 eV, although we stress that a reverse assignment, as suggested in some XPS studies on pristine C3N4 also has its merits. [68,79]To allow for comparison of the spectra for VN-C3N4 and the composites with that for bulk C3N4, we therefore fitted the spectra using several constraints based on structural relationships (Figure S9 and Figure S10).
Further information about the fitting process is given in the SI.
The binding energies and atomic ratios for both the C 1s and the N 1s spectra with the most reasonable fitting result are summarized in Table 1.
In order to still derive meaningful insights into structural changes from the N 1s spectra, despite the variable fitting, we decided on normalization of the spectra (Figure S11).Two things become immediately obvious.One is a shift of the main peak in the N 1s spectra from 398.7 to 399.0 eV upon composite formation, indicating an increase in the binding energy of N-(C)2 in the heptazine units, as discussed above.Additionally, the 'shoulder' at ~401 eV is pronounced less sharply, indicating larger amounts of nitrogen species at medium binding energy.The same shift towards higher binding energy is observed in the C 1s spectra, together with an increased intensity of the π→π*-satellite.The lower amount of cyano groups in the untreated C3N4 is also shown by the lower intensity at 286 eV compared to both VN-C3N4 and the composite.These observations confirm the main conclusions drawn from the N 1s spectra above, without relying on arbitrary fitting results.Concluding these considerations, XPS analysis supports the introduction of both cyano groups and vacancies upon KOH treatment.Furthermore, it elucidates the changes in the electronic structure upon addition of FeS2 to the system, indicating partial electron extraction from the VN-C3N4 matrix.
Hydrolysis and introduction of OH-groups similar to what has been observed for hydrothermal treatment with NaOH, involving the breaking of NH-bridging bonds and introduction of OHgroups cannot be totally excluded, but is not likely to have a major influence, since oxygen contents are similar in both C3N4 and VN-C3N4. [69]A-MS measurements were conducted on the composite containing 5 wt.% of FeS2 and on both of the constituents, FeS2 and VN-C3N4, to confirm that no structural changes occur during synthesis while thermal treatment for 2 h at 200 °C (Figure S12).FeS2 was stable up until about 400 °C, above which a gradual extraction of sulfur in the form of SO2 was observed.Notably, the lack of an increase in the mass indicates an absence of significant oxidation during the initial heating phase.
For VN-C3N4, a major mass loss is observed starting at 570 °C, which is completed at around 720 °C.During the heating in synthetic air, NO, H2O and CO2 were found to be the main combustion products up until around 650 °C, after which an increase in the evolution of CO and NO2 was observed (Figure S13).The two steps of the combustion process are also apparent in the DSC curves (Figure S12).The TGA curves for the composite show many similarities compared to that of C3N4.However, both the mass loss curves and the ion currents for the evolving gasses are shifted by almost 100 °C to lower temperatures, signifying that the presence of FeS2 boosts the decomposition, likely acting as a catalyst and activating the heptazine units.Nevertheless, the observed thermal decomposition only occurs at temperatures far above the 200 °C, thus precluding decomposition during the composite formation.
The interaction with FeS2 mainly involves the nitrogen in VN-C3N4, as shown in the much decreased ion current for NO2 during the second step of the combustion process, indicating, that more nitrogen is extracted from the structure in the beginning.This is confirmed by the DSC curves, where the sharp peak at the end of the combustion process is much less pronounced (at 586 and 712 °C, respectively).While the ion currents for all gaseous combustion products in VN-C3N4 show one signal with a sharp peak current, those of the composites appear as double peaks.Perhaps they correspond to the combustion of areas in close proximity to FeS2 and to areas for which the influence of FeS2 is less.Since the ion currents for both NO and NO2 are significantly lowered compared to that in C3N4 (also in relation to the ion current for CO2, so this observation is not only due to the lower content of C3N4), new nitrogen containing reaction products might be formed, that were not detected.

Photocatalytic NH3 Generation
Photocatalysis was performed in a semi-batch flow setup using 20 vol.% of methanol as a hole scavenger.The reaction was investigated for VN-C3N4 and composites therewith containing different amounts of FeS2.After the irradiation period of 7 h, the solution was directly filtered and analyzed for NH3 by the salicylate test.Additionally, the concentration of NH3 in an acid trap located behind the reactor was evaluated.The calibration curves can be found in the SI, Figure S14.The quantification of NH3 was performed after stable color development, which was only obtained after several hours in the dark (Figure S15).
The decoration of VN-C3N4 with FeS2 can significantly enhance the NH3 yield by a factor of 1.63 from 494 µg L -1 to 801 µg L -1 , which equals to 3.9 and 6.4 µmol h -1 for 200 mg photocatalyst, respectively.Compared to the untreated C3N4 the activity is enhanced by ~400 %, clearly showing that a combination of defect introduction and interaction with FeS2 is necessary for efficient ammonia generation.The ammonia yield is comparable for a FeS2 content between 1 and 5 wt.% (Figure 5 and Figure S17), although decreasing and dropping significantly for FeS2 ratios above 10 wt.%.Already low amounts of FeS2 are sufficient for the activation and further loadings decrease the activity due to shadowing effects and a reduced ratio of the active photocatalyst, because FeS2 itself is inactive in the N2 reduction, solely yielding H2 under illumination (Figure S17 and Figure S21).
Several control measurements were conducted, to elucidate the source of nitrogen and the selectivity of the reaction (Figure 5 and Figure S18).Firstly, a dispersion of a composite photocatalyst in water/ methanol was tested for NH3 after stirring in the dark.No NH3 was detected in this case.Secondly, a dispersion of VN-C3N4 was filtered and tested for NH3 to exclude amino groups in the photocatalyst interfering with the test.No NH3 was observed in both control measurements.We also tested a physical mixture of FeS2 (5 wt.%) and VN-C3N4 for photocatalytic NRR ("Mix 5 wt.%"), without performing the grinding and subsequent calcination steps that establish an interfacial contact between the two constituents.The activity was significantly lower than that of VN-C3N4 itself, due to the lower amount of active photocatalyst and lack of interaction between FeS2 and VN-C3N4 (Figure 5).This clearly indicates an enhancement effect in ammonia generation based on a direct contact between FeS2 and VN-C3N4.Additionally, the photocatalytic reaction for the composite containing 5 wt.% of FeS2 was repeated in an argon atmosphere ("5 wt.% -Ar").A similar activity compared to the reaction in N2 atmosphere was observed, strongly suggesting that the ammonia generation is not based on N2 feed gas reduction, but stems from the VN-C3N4 framework instead (Figure 5).
Literature for NRR over vacancy-rich and cyano-rich C3N4 likewise propose a N2 conversion pathway following a Mars-van Krevelen mechanism, therefore this observation is not surprising, [25] however significant NH3 yields for C3N4 in an argon atmosphere, like we show, were not reported.
Therefore, the performance of careful blind experiments in combination with detailed material characterization before and after photocatalysis is of high necessity.
The reaction solutions were additionally tested for nitrogen-containing by-products such as hydrazine and NO3 -.No hydrazine and only trace amounts of NO3 -around the lower detection limit were found for both VN-C3N4, and the composites (Figure S19, Figure S20), when the photocatalytic experiment was performed in N2, but not in an Ar atmosphere.The main side product was H2, with a production rate of around 200 µmol h -1 .The rate was similar for all composites, with a slightly increased H2 formation rate for higher amounts of FeS2 in the sample (Figure S21).FeS2 itself showed a remarkably high H2 production rate of 370 µmol h -1 .Thus, both the activity improvement for NH3 generation and the selectivity are highest for lower FeS2 loadings in the composites.
A gradual decrease in the H2 evolution rate over time is in good agreement with a possible Marsvan-Krevelen-type of structural changes during the illumination.When the photocatalytic experiment was performed in an argon atmosphere, similar H2 evolution rates were observed as compared to the results in a N2 atmosphere (Figure S21).
In order to elucidate the formation of methanol oxidation products, UV absorption spectra of the reaction solutions were recorded after the photocatalytic experiments.Critical evaluation of the conditions and evolving organic oxidation products is crucial, since the accuracy of the salicylate test can be influenced by a variety of parameters that are too often not considered.There is a correlation between the amount of NH3 produced and the amount of UV light absorbance by the filtered reaction solutions at 205 to 220 nm (Figure 6).This absorbance can mainly be attributed to formic acid that is formed alongside NH3 (Figure S22).This could have an effect on the salicylate test, resulting in a significant underestimation of the actual NH3 concentration. [81]For this reason a 12 % hypochlorite solution was used, which negates the effect of varying acid generation.Additionally, a reference sample of known ammonia concentration was always measured together with the reaction solution to avoid errors based on the testing solutions.All concentrations given in the paper are averaged between several quantification measurements with the salicylate test.Additionally, the spectra clearly show the presence of other ions that absorb UV-light, which is even more obvious in the derivation spectra (Figure S23).The overlap of multiple features renders an accurate identification nearly impossible, however.Nitrate and nitrite absorb UV light at 203 and at 210 nm, respectively. [82]No clear absorption feature could be identified for nitrite, while a very small contribution of nitrate might be discernible for VN-C3N4 and composites thereof, which is in good agreement to the results from ion chromatography.The absorption peak at 220 nm probably stems from π-π* transitions, possibly in carbonyl groups of amides or very short conjugated systems. [83]The broad absorbance from 230 nm towards higher wavelengths could in part be caused by very fine FeS2 particles, since a filtered FeS2 dispersion gives rise to a noticeable absorbance signal up to 450 nm (Figure S22).
To exclude significant errors in the determined ammonia concentration due to the large amounts of organic byproducts, the solutions were measured with ion chromatography and the values compared to those from the colorimetric determination (Table 3).The general trend is the same as observed with the salicylate tests, but the determined ammonia concentrations are slightly higher.This is to be expected, since the significant amounts of methanol and formed formic acid largely affect the colorimetric test, resulting in an underestimation of the actual ammonia concentration, and even with careful optimization of the testing procedure, some influence of the organic substances cannot be avoided.A graphic depiction of the concentrations determined with ion chromatography, as well as an exemplary chromatographic trace can be found in Figure S24.Since hardly any difference in the NH3 yield in argon and N2 atmosphere was observed after 7 h for the most active sample, we devised a longtime measurement.A dispersion of the composite containing 5 wt.-%FeS2 was first illuminated for 14 h in either N2 or argon atmosphere.Every two hours, a sample was taken and analyzed for NH3.After 14 h, the lamp was switched off and the solution was stirred in the dark for 6 h under continuous gas flow of the respective gas, before illumination was continued for another 6 h (Figure S25).During the first irradiation period, the generated amount of NH3 increased almost linearly with a rate of approx.82 µg h -1 in N2 and 90 µg h -1 in argon atmosphere, respectively.Once the lamp was switched off, the measured concentration slightly decreased in both atmospheres, likely due to NH3 carried out of the reactor by the gas flow.After continuing the light irradiation, the NH3 generation was increased again with about the same rate, as before the period in the dark.Afterwards, the determined concentrations seem to level out.This effect is likely caused by a combination of slow degradation of the structure and accumulation of oxidation products.A similar effect was observed for the H2 evolution rate, indicating that the observed effect is not solely caused by the accumulation of formic acid (Figure S25b).
In addition to the NH3 concentration reported in the reaction solution, NH3 was detected in the acid trap but not included in the concentrations given here.The amounts of NH3 were maximum around 6.5 µg and 3.5 µg total for the experiments in an argon and a nitrogen atmosphere, respectively (Table S8).Still, the use of an acid trap is sensible.The pH of a dispersion of 20 mg of VN-C3N4 in 20 mL of H2O was approx.9.5.Thus, both NH4 + and NH3 species can occur in significant amounts.For C3N4 the pH is with ~8 closer to neutral, perhaps due to higher amount of imine or amine groups in VN-C3N4.A slightly alkaline pH for C3N4 is expected, due to the large number of amino groups.

Post-photocatalytic characterization
Both VN-C3N4 and the composites with FeS2 were thoroughly characterized after the photocatalytic experiments, in order to evaluate the stability and to gain further insights into the ammonia generation pathway.Post-photocatalytic XRD patterns still show the same reflections for phasepure C3N4 and FeS2 (Figure S26).The intensity for the FeS2 reflections is significantly decreased after the NRR.This effect could be a result of a loss of interfacial contact between FeS2 and C3N4 as verified by dispersing a composite in water/ methanol mixtures and subsequently regaining the material via centrifugation (Figure S27).FeS2 is known for its flotation tendency in mineral separation, due to its relative hydrophobicity. [84]Additionally, we tested the stability of FeS2 during both the formation of the composite (with annealing at 200°C) and storage of the sample in air.No changes in the crystal structure of FeS2 were observed (Figure S27).
UV/vis spectra show a significantly increased absorbance of UV light and a decrease of the band gap (Figure S28).Both VN-C3N4 and the composites showed a pronounced darkening after the NRR that decreased again after storage in air (Figure S29).The effect of increased UV light absorption is less obvious for the composite containing 15 wt.% compared to the other composites, for which the activity was also lowest.All other composites that exhibited a similar activity, show a similar increase in UV absorption, which likely correlates to higher degrees of structural change.The increased UV absorption is less pronounced for VN-C3N4 in agreement with the lower activity.The band gap decreases very slightly by 0.03 eV for VN-C3N4.For the composite containing 5 wt.%, the decrease is most drastic and the band gap is experiencing a change by 0.07 eV from 2.78 to 2.71 eV.The red shift of the absorption might be caused by defect formation and distortion of the structure. [85]IFT spectra for the composites after the NRR experiments show a marked decrease in the vibrations for the heptazine units (Figure S30 and Figure S31), especially in relation to the vibration at 1720 cm -1 , that falls into the range for C=N vibrations.This vibration is further shifted back to slightly higher wavenumbers.A decrease in the relative intensity of the 808 cm -1 vibration agrees with this observation, indicating structural damage to the heptazine framework.Both effects were also observed to a lesser extent upon introduction of vacancies and cyano groups upon KOH treatment and thus confirm further breakdown of the heptazine units during photocatalysis.
Additionally, a closer look at the vibration at 2147 cm -1 reveals a decrease in the intensity, supporting the assumption that cyano groups are consumed during the photocatalytic experiment (Figure S32).The effect is even more increased for long-term experiments, highlighting the further degradation of cyano groups with prolonged illumination times (Figure S32).
XP spectra of the composite containing 5 wt.% of FeS2 after the photocatalytic reaction show a slight shift for the carbon species adjacent to cyano-/ amino-nitrogen and the N-C=N peak towards lower binding energies (Figure S33).The intensity for both peaks in the C 1s spectra corresponding to the VN-C3N4 structure decreases markedly in relation to that of C-C, further indicating structural changes, that do not only extend to nitrogen, but are further inflicted on carbon in the structure.
The ratio might also partly be influenced by adsorbed organic residues from the sacrificial agent.
Additionally, the intensity for the satellite peaks is decreased, supporting damage to the aromatic system.The calculated C/N ratio from the survey scan was 0.88, without correction for adventitious carbon, because the amount of C-C or C=C bonds possibly present in the structure after partial extraction of nitrogen is unknown.Thus the C/N ratio is increased compared to 0.80 for the same composite before the photocatalytic experiment.This is a strong indication of nitrogen extraction from the VN-C3N4 matrix.
Sample composition was additionally evaluated by elemental analysis before and after photocatalytic experiments (Figure S34).The amount of sulfur and thus probably also FeS2 was noticeably decreased after the photocatalytic experiment for all composites.This is likely an effect of the imperfect interfacial contact and washing of the sample, as has already been observed in the XRD patterns.
For a more detailed comparison, the C/N ratio was calculated for the composite samples before and after the photocatalytic reaction.For all composites, the C/N ratio was about 0.555, which significantly differs from the compositional value of 0.75 and indicates incomplete polymerization and the existence of many free amino groups.The value is in good agreement to the EDX measurements, though.No significant differences of the C/N ratio were observed for different FeS2 loadings.For all composites, the C/N ratio was visibly increased after the photocatalytic reaction, supporting the extraction of nitrogen from the structure (Figure S34).Compared to the C/N ratios obtained from XPS measurements, the nitrogen content in the bulk is significantly lower, due to adsorbed carbon impurities and possibly nitrogen deficiency at the surface.

Charge Carrier Dynamics
Due to the large variety of defects that can potentially be present in VN-C3N4, the electronic structure is rather complex and should be studied in detail, in order to get explanations for the different results in photocatalytic experiments.To investigate the charge carrier dynamics, transient absorption spectroscopy (TAS) in diffuse reflectance geometry was employed.In Figure 7 the ns-TAS measurements of a composite of VN-C3N4 and 5 wt.% of FeS2 in an argon atmosphere are presented.[88][89] The same signal is generally apparent in the measurement of the composite, but an increase of the relative absorption intensity in the range between 650 nm and 750 nm is noticeable, in comparison to the main signal between 750 and 900 nm.This might indicate a change in the electronic structure upon composite formation with FeS2 and an increased relative amount of photo-generated electrons in a second excited state.For a better comparison the main signal of the photo-generated electrons in both materials was analyzed regarding the lifetime of these electrons.The lifetimes monitored at 800 nm for both VN-C3N4 and the composite are comparable and in the order of ns (Figure S35).The signal is best fitted with two different lifetimes, that are 3.4 ns and 59.0 ns for VN-C3N4 and 4.3 ns and 31.7 ns for the composite, respectively.The shorter lifetime of the electrons is comparable for both materials and could be explained with the fast recombination of charge carriers in the materials.Electrons with fast recombination rates can generally not participate in photocatalytic reactions.The other lifetime for the composite indicates that the introduction of FeS2 to the VN-C3N4 matrix results in the prevention of electron accumulation in deep traps. [89,90]e additionally used photoluminescence (PL) spectroscopy to gain insights into the radiative energy relaxation levels of the excited photocatalyst.The most dominant PL emission in semiconductors is the band gap emission, where the excited states (charge carriers) recombine and relax to the ground state by emitting radiation.Thus, low PL emission is often an indicator for efficient charge carrier separation and low recombination rates.C3N4 as a polymeric material with a structure in between that of molecules and solid crystals, generally exhibits a prominent blue fluorescence, that has been shown to be significantly influenced by the introduction of defects. [91,92]on irradiation at wavelengths below the band gap energy, a broad fluorescence signal between 410 and 640 nm is observed for VN-C3N4 and the composites, with a maximum emission at 470 nm and a slightly less intense emission at 445 nm, as shown in Fehler!Verweisquelle konnte nicht gefunden werden.8.Generally, the fluorescence emission of the composites is decreased compared to VN-C3N4, indicating better charge separation and thereby reduced recombination.eS2 itself does not show any fluorescence emission.The intensity for the emission at 445 nm in relation to that at 470 nm is clearly increased for the composites (Figure 8).The emission at 445 nm can be attributed to direct band gap emission, which is in good agreement to the band gap energy determined via absorption spectroscopy. [85][94]ission at 470 nm corresponds to transitions from intra band gap states, likely involving lone electron pairs at N-(C)3 sites. [95]An additional tail towards higher wavelengths includes contributions from defect emissions, such as transitions from lone pair states of nitrogen atoms in the s-triazine unit, from amino groups, NH-bridging units, or possibly graphitized areas at even higher wavelengths above 500 nm. [91,92]The significantly decreased intensity of lone pair and The excitation maximum for VN-C3N4 is located at approx.320 nm which corresponds to band gap transitions (Figure S36 and Figure S37). [95]The ratio of emission intensity at 320 nm excitation to emission intensity at 380 nm excitation in the normalized excitation spectra is significantly decreased for the composites compared to VN-C3N4, along with a slight red-shift of the entire emission signal.The group of Gan et al. ascribes excitation at around 380 nm to transitions from the valence band to lone pairs of -N-(C)3 species, that appears to be more favored for the composites. [95]The absolute emission intensity is decreased at higher wavelengths, since the light absorption is diminished (Figure S37).
The lifetime of an emission is regarded as a more reliable parameter for judging charge recombination rates.The lifetime was measured for both detection at 445 and 470 nm (Figure S38).The decay can be fitted with three exponential functions, as is usually done in literature, as well. [14,91]The fluorescence lifetimes are increased again, for the composites, compared to VN-C3N4 (Table S10).
The shortest lifetime (τ3 for emission at 445 nm/τ3' for emission at 470 nm) for both the composites and VN-C3N4, respectively, can be attributed to recombination in the aromatic system. [70]An increase in the second lifetime, τ2 , for the composites compared to defective C3N4 implies improved intraplanar and intrachain charge separation (Figure S39). [70]This might be due to polarization of the heptazine framework and electron transfer.The greatest difference is observed in the time constants for τ1, which can be attributed to charge separation between layers (along the π-stacking direction).Hence, charge separation in the π system is significantly improved again by FeS2, affecting both the direct band gap emission at 445 nm and the emission of lone pairs of ternary nitrogen species at 470 nm.The lifetimes are fairly similar for all composites.Fluorescence decay is longer at 470 nm compared to at 445 nm for all samples, supporting the assumption of the main emission at 470 nm mainly being caused by recombination at defect sides (involving free electron pairs of nitrogen species).
The quantum yield (QY) was determined at an excitation energy of 355 nm, to ensure that the increased fluorescent lifetimes truly correlate with improved charge carrier separationTable 4 Quantum Yield at 355 nm for the composites containing different amounts of FeS2..The QY is decreased in the composites with increasing FeS2, implying that recombination is indeed reduced in the composites.

Discussion of the interaction of FeS2 and VN-C3N4 resulting in enhancement of ammonia generation
Based on the observations for structural changes in VN-C3N4 induced by the presence of FeS2, we propose an activation of the structure of VN-C3N4 by FeS2.Since KOH treatment is supposed to introduce defect sites that have two nitrogen species with free electron pairs in the vicinity (iminetype nitrogen in =N-CN units and amino groups), [77] an interaction similar to ligand to metal coordination to Fe 2+ appears to be feasible, that results in an activation of the C3N4 structure via πback-donation and possibly light induced metal to ligand charge transfer (Figure 9).This model is based on following observations: 1) Increased UV absorbance observed for the composites (more π-π*-transitions) (Figure 3).
2) Band gap widening upon FeS2 addition (Figure 3): possibly due to deformation/ polarization of the structure upon coordination of defects in VN-C3N4 to iron centers due to partial electron transfer (σ-Donor) to FeS2.
3) The shift of both nitrogen and carbon binding energies towards higher values suggests partial oxidation of the structure, thus supporting the σ-donor effect (Figure 4).
4) The increased satellite peaks in the XP spectra suggest higher number of electrons in the πsystem, which might be due to π-back-bonding, (Figure 4).Possibly, coordination of the cyano groups and/ or amino groups to Fe 2+ centers induces a marked electron withdrawing effect and partial charge transfer to iron, reducing the electron delocalization.
5) Fluorescence quenching and decreased QY indicate improved charge transfer (Figure 8 and Table 3).
6) Decreased ratio of emission from lone electron pair states in fluorescence measurements (Figure 8) indicates electronic interaction of defects in VN-C3N4 with FeS2.
7) Increased fluorescence lifetime, τ1, for the composite (Table S10): electrons in the π-system might be influenced and stabilized by the interaction with iron.
8) An additional signal for photo-generated electrons in the TAS measurements upon composite formation with FeS2 might arise due to complex interactions of defect states with iron centers, such as light-induced M→L-charge transfer (Figure 7).11) π-back-donation from Fe-centers to imine bonds are known, as is back-donation and charge transfer with cyano group. [96,97]) The oxidation products in the photocatalytic reaction are the same for both, composites and VN-C3N4, (Figure 6), suggesting oxidation on VN-C3N4 13) The HER activity is decreased for the composites (Figure S21), even though FeS2 is present, indicating that electrons in FeS2 are used for other redox reactions, such as π-back-bonding and/or re-oxidation of Fe 3+ to Fe 2+ .
14) Slightly enhanced ammonia generation was also observed for untreated C3N4 decorated with FeS2 (Figure S40) and was expected due to minor defects in C3N4 (cf.DRIFT results).
This further underlines the boosting effect in ammonia generation, when FeS2 and highly defective, KOH-treated C3N4 are present.
15) Interaction of the composite with N2 is indicated in the TGA-MS measurements, due to coordination to Fe 2+ .Such an effect could be advantageous for re-incorporation of nitrogen into the structure.

Conclusion
We have shown that a combination of defect introduction into C3N4 and subsequent composite formation with FeS2 can significantly improve the amount of ammonia generated in photocatalytic NRR experiments, resulting in four times higher ammonia yields under illumination compared to unmodified C3N4.The optimal FeS2 loading was established to be 1 to 5 wt.%.The system only employs inexpensive, earth-abundant and non-toxic materials.However, knowledge about the exact structure of C3N4 and the presence and character of defects proved to be crucial, as they significantly influence the interactions between the two constituents.Charge transfer between FeS2 and VN-C3N4 was established to proceed at the defect sites, resulting in an electronic activation of the structure.Therefore, NH3 generation was found to occur not photocatalytically from N2, but rather via a novel activation route of VN-C3N4, involving reduction of the =N-CN group adjacent to nitrogen vacancies.A replenishment of nitrogen in the structure could, however, not yet be verified.FeS2 acts similar to a co-catalyst, enhancing the ammonia generation, although the process is different to that of (metal) electrocatalysts typically employed for photocatalytic HER or OER.
Here, π-back-donation from Fe-centers to imine nitrogen and amino groups of the defect-rich C3N4 reduces the activation barrier for the reduction of terminal cyano groups upon illumination.
Although there are numerous reports on C3N4 for the photocatalytic NRR, the complexity of the system, with a broad variety of defects that can theoretically be present, necessitates a strict control of the synthesis parameters and thorough characterization experiments.We deduce that photocatalytic ammonia generation with defective C3N4 is most likely always a product of selfdegradation, which can be enhanced by back-donation with suitable coordinating materials.

ASSOCIATED CONTENT
Supporting Information.Additional XRD patterns, XPS data, SEM images, TGA analysis, DRIFT data, more details on salicylate test, hydrogen evolution rate transients, post-catalytic analysis, elemental analysis, TAS transients, PL excitation spectra, PL emission spectra,

Figure 1 .
Figure 1.Powder XRD patterns for composites of FeS2 and VN-C3N4 and a physical mixture of

Figure 2 .
Figure 2. SEM images of a composite of C3N4 and pyrite (5 wt.%), clearly showing the distribution

Figure 5 .
Figure 5. Ammonia concentration after 7 h obtained for composites with a different FeS2 ratio.

Figure 6 .
Figure 6.UV spectroscopy for the filtered reaction solutions after the NRR experiments.

Figure 8
Figure 8 Fluorescence emission spectra for the composites for λexc.= 355 nm (a), normalized defect emission in relation to direct band gap emission in the composites, coupled to fluorescence quenching might indicate the inhibition of radiative recombination at defect sites, likely due to interaction of VN-C3N4 with FeS2 involving free electron pairs.Since N-(C)3-sites are also effected, the interaction with iron influences the entire electronic structure of VN-C3N4, as also observed in DRIFT and XPS measurements.Charge transfer might assist in exciton separation and retard recombination.

Figure 9 .
Figure 9. Proposed coordination of nitrogen lone pairs to Fe 2+ at defect sites and π-back-donation

9 )
TGA curves show a shift in the decomposition of VN-C3N4 promoted by FeS2, indicating structure activation by FeS2 (Figure S12).10) Shift of the vibration energy to lower values for the heptazine units additionally indicate structure activation: signals for C=N vibrations are apparently more influenced by the FeS2 addition than C-N vibrations, with the relative intensity decreasing (Figure S6).

Table 1
Binding energy and atomic ratios for different species in the N 1s and C 1s spectra.

Table 3
Ammonia concentrations determined with the salicylate test and with ion chromatography.

Table 4
Quantum Yield at 355 nm for the composites containing different amounts of FeS2.