Revisiting the Electrochemical Nitrogen Reduction on Molybdenum and Iron Carbides: Promising Catalysts or False Positives?

Abstract

The electrochemical dinitrogen reduction reaction (NRR) has recently gained much interest as it can potentially produce ammonia from renewable intermittent electricity and replace the Haber–Bosch process. Previous literature studies report Fe- and Mo-carbides as promising electrocatalysts for the NRR with activities higher than other metals. However, recent understanding of extraneous ammonia and nitrogen oxide contaminations have challenged previously published results. Here, we critically assess the NRR performance of several Fe- and Mo-carbides reported as promising by implementing a strict experimental protocol to minimize the effect of impurities. The successful synthesis of α-Mo₂C decorated carbon nanosheets, α-Mo₂C nanoparticles, θ-Fe₃C nanoparticles, and χ-Fe₅C₂ nanoparticles was confirmed by X-ray diffraction, scanning and transmission electron microscopy, and X-ray photoelectron and Mössbauer spectroscopy. After performing NRR chronoamperometric tests with the synthesized materials, the ammonia concentrations varied between 37 and 124 ppb and are in close proximity with the estimated ammonia background level. Notwithstanding the impracticality of these extremely low ammonia yields, the observed ammonia did not originate from the electrochemical nitrogen reduction but from unavoidable extraneous ammonia and NO_x impurities. These findings are in contradiction with earlier literature studies and show that these carbide materials are not active for the NRR under the employed conditions. This further emphasizes the importance of a strict protocol in order to distinguish between a promising NRR catalyst and a false positive.

1. Introduction

The activation of diatomic nitrogen has been one of the greatest challenges in nitrogen-related chemistry.^1,2 This is inherently related to the inert nature of the nitrogen molecule due to its high bond-dissociation energy, absence of a dipole moment, and low proton and electron affinity.³ Despite the inert nature, diazotrophic microorganisms successfully fixate nitrogen and play a key role in enriching the soil.⁴ However, due to the growing world population and the high demand for food, additional nitrogen containing nutrients in the form of artificial ammonia-based fertilizers must be provided to the soil. The majority of the ammonia produced worldwide is synthesized by the Haber–Bosch process (H-B), wherein energy intensive reaction conditions (T = 300–500 °C, P = 150–300 bar) are needed to activate dinitrogen.² The ammonia industry consumes approximately 1% of the global energy demand and emits roughly 0.75% of the anthropogenic CO₂ emissions (assuming 1.9 t_CO2 t_NH3^–1),^5,6 which is motivating the search for more energy efficient and sustainable alternatives.

The electrochemical nitrogen reduction reaction (NRR), wherein dinitrogen, water, and electrons from renewable sources react to form ammonia, has recently gained significant scientific interest and has been proposed as a potential replacement for the fossil fuel-based H-B.^7,8 NRR systems at high (>500 °C) and intermediate (100–500 °C) temperatures have proven to be successful in terms of faradaic efficiency (FE ≥ 75%) and NH₃ yield (≥4.5 nmol s^–1 cm^–2).⁹ Nevertheless, the present high temperature NRR systems tend to have a low energy efficiency compared to H-B.^10,11 Therefore, it would be beneficial to perform the NRR under ambient conditions. Significant FEs have been reported with iron electrocatalysts in ionic liquids and organic electrolytes by suppressing the parasitic hydrogen evolution reaction (HER).^12,13 However, the use of organic electrolytes is likely to be less economically feasible compared to aqueous electrolytes due to complex scalability, safety issues, high costs, and intense energy requirements.^8,14 Unfortunately, the kinetics under ambient conditions in aqueous electrolytes are sluggish and many attempts in examining transition metals, such as Au, Fe, Mo, Ru, Rh, and Re, as potential electrocatalysts for the NRR have resulted in low FEs (<1%) and ammonia yields (<0.1 nmol s^–1 cm^–2).^15,16

The active site of the nitrogenase enzyme, the biologic pathway for nitrogen fixation, is the FeMo-cofactor. The FeMo-cofactor contains a six iron atomic trigonal prism with a carbon-centered position. Each iron is bound to three sulfur atoms, with an additional iron and molybdenum in apical positions.^2,17 Attempts to mimic the FeMo-cofactor initiated investigation into Fe- and Mo-based heterogeneous NRR catalysts, such as carbides and sulfides. Both Mo₂C and MoS₂ show noble metal like properties, due to similar d-band configurations as Pt.¹⁸ Therefore, they can act as cheap and robust catalytic substitutes for many applications, including water electrolysis, water gas shift reaction, and ammonia decomposition.¹⁹ Despite the fact that these materials are good HER catalysts, several density functional theory studies have predicted favorable nitrogen binding energies.¹⁹⁻²¹ Experimental results are somewhat distinct; nitrogen reduction experiments with amorphous 2H-MoS₂ and metallic 1 T-MoS₂ did not produce quantities of ammonia exceeding the background level,^12,21 while other studies report reasonable FEs and ammonia yields using FeS₂, Mo₂C, and Fe₃C, thereby labeling these materials as promising NRR catalysts.²²⁻²⁴

The electrochemical NRR field is plagued by questionable results, mainly due to the large impact of extraneous ammonia sources on experiments performed on a small scale. Ammonia stemming for other sources can erroneously be assigned to ammonia synthesized by the NRR, which can lead to false positives. Ammonia impurities can be minimized by a proper experiment design and can be identified by applying the right control experiments, such as argon and open-circuit blank tests and ultimately ¹⁵N₂-labeled experiments. Recently, nitrogen oxides have been identified as another source of contamination, as these species are more easily reduced to ammonia than dinitrogen in the NRR.²⁵ The majority of the recently published studies have applied blank tests, but performing quantitative ¹⁵N₂-labeled experiments and monitoring nitrogen oxide species are done sporadically.²⁵ As a consequence, a handful of research groups have tried to reproduce electrocatalysts initially labeled as promising, such as Fe, Bi, Au, VN, CoMo, Mo₂N, and MoS₂,^{16,21,26−30} but discovered that the quantified ammonia must originate from sources other than the NRR. Here, we critically assess the electrocatalytic NRR activity of molybdenum and iron carbides, where more than 10 independent literature reports claim to observe superior or excellent catalytic performance.^{23,24,31−38} In the present work, α-Mo₂C nanodots from Cheng et al. (reported as the most promising carbide catalyst) are reproduced and compared with α-Mo₂C nanoparticles as a benchmark.²³ Additionally, nanostructured θ-Fe₃C and χ-Fe₅C₂ are synthesized and examined for their NRR activity. A key aspect of this work is the implementation of a strict protocol, which is designed to minimize the level of extraneous contamination,^15,25 allowing genuine quantification of ammonia produced by the NRR.

2. Experimental Section

2.1. Materials

Materials and chemicals were purchased from Sigma Aldrich, if not indicated otherwise. Ultrapure water (Millipore Milli-Q IQ 7000) was used for catalyst synthesis, electrolyte preparation, and cleaning procedures. Concentrated sulfuric acid (95–98 wt % H₂SO₄, trace metal purity) was used for glassware acid cleaning and diluted for other purposes. High purity N₂, Ar, and H₂ (99.999%, Linde) were used for electrochemical experiments and material synthesis.

2.2. Molybdenum and Iron Carbide Synthesis

2.2.1. Mo₂C Nanodot-Decorated Carbon Nanosheets

Mo₂C nanodots (Mo₂C NS) were synthesized by a molten-salt synthesis procedure as reported in detail elsewhere.²³ In short, a mixture of 1 mL of water and 4 mL of ethanol (96%, VWR) was mixed in a beaker and continuously heated and stirred on a hotplate. Once the mixture reached 70 °C, 0.4 g of bis(acetylacetonato)dioxomolybdenum(VI) and 0.14 g of sucrose (99.5%) were added. After the mixture turned green, an excess amount of sodium chloride (99.5%) was added until a green crystalline slurry was formed. The slurry was directly transferred to a ceramic boat and placed inside a tubular furnace (Blue, Lenton), where the specimen was heated to its carburization temperature under an Ar atmosphere (T_carb = 800 °C, heating rate = 5 °C min^–1), kept constant at this temperature for 2 h and the furnace cooled down to room temperature naturally. The resulting black catalyst/salt mixture was excessively rinsed with ultrapure water to remove the sodium chloride. The residue was filtrated (Durapore 100 nm, Merck) and dried in an oven at 80 °C overnight.

2.2.2. α-Mo₂C Nanoparticles

Gómez-Marín and Ticianelli reported a procedure for the synthesis of porous Mo₂C nanoparticles (Mo₂C NP) that was replicated here.³⁹ In a typical procedure, 0.15 g of Vulcan VC-72 (Cabot) was mixed with 0.51 g of MoO₃ (99.9%) in a beaker containing 30 mL of ethanol. The dispersion was heated to 60 °C overnight while continuously stirring to evaporate the ethanol completely. The powder was transferred to a ceramic boat for carburization inside a tubular furnace under 10 vol % H₂:Ar. The precursor was heated to 725 °C for 30 min with a slow heating rate (1 °C min^–1) and cooled down to room temperature.

2.2.3. Mesoporous Fe₃C

A combined hard-templating and carburization method developed by Kraupner and coworkers was used to create a mesoporous Fe₃C structure with a high surface area.⁴⁰ In brief, 0.5 g of FeCl₃ (99.9%) was dissolved in 1 g of 40 wt % SiO₂ in H₂O (Ludox AS40) in a borosilicate test tube. Additionally, 0.728 g of 4.5-dicyanoimidazole (99%) was added and stirred through the suspension and sonicated for 30 min to achieve a homogeneous yellow-colored thick slurry paste. The paste was transferred to a ceramic boat and carburized inside a tubular furnace at 700 °C (heating rate 2 °C min^–1) for 2 h under an Ar atmosphere and cooled down to room temperature.

2.2.4. χ-Fe₅C₂ Nanoparticles

The principle of thermo-decomposition of Fe(CO)₅ is a common used strategy to synthesize iron carbides and is discussed in detail elsewhere.⁴¹ A mixture of 0.2 g of polyvinylpiyrrolidone (PVP, 40000 g/mol) and 1 mL of Fe(CO)₅ (99.99%) was inserted in a homemade air-tight reactor consisting of Swagelock tubes and adapters (Figure S1). The reactor was purged with Ar at a flowrate of 20 mL min^–1 for approximately 10 min to remove residual oxygen, and immediately afterward, all Swagelock adapters were closed. The reactor was positioned inside a muffle furnace programmed with T_carb at 300 °C (heating rate 2.3 °C min^–1) for a duration of 24 h.

2.3. Characterization

2.3.1. X-ray Diffraction (XRD)

Samples were deposited on a Si510 zero background wafer and positioned inside a Bruker D8 Advance diffractometer in Bragg–Brentano geometry equipped with a graphite monochromator, a Vantec position sensitive detector, a variable divergence slit, and a 5 mm height scatter screen. Co Kα radiation (λ = 0.1789 nm) was used to avoid incident beam fluorescence effects on the Fe carbides. During each acquisition, steps with a size of 0.038° and 5 s per step were measured over a 10–110° 2θ range. Bruker DiffracSuite.EVA v6.0 was used to subtract the background, correct small displacements, and strip the Kα2 contribution from the patterns to enable crystallite size (D_XRD) estimation with the Scherrer equation (eq 1), where λ is the wavelength, and κ the shape factor taken as 1. Peak shapes were assumed Gaussian, and the full width at half-maximum, in this case β, was additionally corrected for instrumental line broadening effects.

1

2.3.2. Mössbauer Spectroscopy

Transmission ⁵⁷Fe Mössbauer spectra were collected at room temperature with a conventional constant-acceleration spectrometer with a ⁵⁷Co(Rh) source. Velocity calibration was carried out using an α-Fe foil. The Mössbauer spectra were fitted using the Mosswinn 4.0 program.⁴²

2.3.3. X-ray Photoelectron Spectroscopy (XPS)

A Thermo Scientific Kα spectrometer with a monochromatic Al Kα excitation source was used to acquire X-ray photoemission spectroscopy (XPS) spectra. The base pressure inside the analysis chamber was about 2 × 10⁹ mbar. HR-XPS spectra were recorded using a 400 μm spot size, 0.1 eV step size, and 50 eV pass energy (200 eV for survey). All spectra were charge-corrected to the C 1s adventitious carbon (284.8 eV). Subsurface layers were measured with a depth profile by argon ion etching (1000 eV) in between XPS measurements. The obtained XPS spectra were deconvoluted with CasaXPS v2.3 software.

2.3.4. Inductively Coupled Plasma Atomic Emission Spectroscopy (ICP-OES)

The materials were dispersed in 35 vol % HNO₃ overnight to dissolve the carbides. The samples were further diluted with 3 vol % HNO₃ with an amount depending on the expected metal content. All ICP-OES measurements were performed on a SPECTRO ARCOS measured against an external calibration, with a typical detection limit of 10 ppb.

2.3.5. Scanning Electron Microscopy (SEM)

Prior to analysis, the aluminum cylindrical sample holder was washed in isopropanol in an ultrasonic bath for approximately 2 min. An isopropanol based catalyst ink was drop-casted on the sample holder and positioned in a 25 mm working distance. The SEM measurements were executed on a Jeol JSM 6500F instrument at an acceleration voltage of 15 kV, coupled with an energy dispersed X-ray analysis detector (Ultradry, Thermo Scientific).

2.3.6. Transmission Electron Microscopy (TEM)

A dispersion of catalyst and isopropanol was drop-casted on a TEM grid with a holey carbon film on a copper 400 mesh (EM-resolutions). All materials were analyzed with a JEOL JEM1400plus TEM at a 120 kV acceleration voltage using a single-tilt specimen holder. The TEM was equipped with a TVIPS TemCam-F416R high-resolution camera based on a custom designed CMOS architecture. ImageJ was used to estimate the particle size distribution.

2.4. Electrochemical Measurements

A Biologic VSP-300 potentiostat in combination with EC-Lab software was used for all electrochemical measurements. The uncompensated resistance (R_u) of the system (the resistance between the reference electrode (RE) and working electrode (WE)) was measured before each cyclic voltammetry (CV) and chronoamperometry (CA) measurement. R_u was determined with potentiostatic electrochemical impedance spectroscopy at open-circuit potential, with a frequency range between 200 kHz and 0.1 Hz. The distance between the origin and the first line intersection on the Z_Real-axes within the Nyquist plot represents R_u and was extracted by manual data fitting. Subsequently, the EC-Lab build-in IR compensation allowed 85% R_u compensation without adding to much distortion to the CV and CA results. Only for the CA experiments, the other 15% R_u was compensated after the measurement by using eq 2.

2

A polyether ether ketone (PEEK) three-electrode cell design adapted from the Jaramillo group was used for all electrochemical experiments.⁴³ It consisted of two separate compartments that accommodate 5 mL of electrolyte and 3 mL of gas headspace. An additional plate was added to the overall cell design (Figure S2), which fixated the WE. A leak-free Ag/AgCl micro reference electrode (Innovative Instruments, LF-1-45) was used for potential control, wherein all potentials were recalculated versus the reversible hydrogen electrode scale following eq 3.

3

A Pt foil (50 × 50 × 0.025 mm, 99.99%, Mateck) functioned as the anode and was rinsed with water and flame annealed before each experiment. A fresh sheet of membrane (Celgard 3401) was used for every run, thereby preventing accumulation of NH₃. The WE was prepared by drop-casting 3 droplets of 10 μL of a freshly prepared catalyst ink (2 mg_cat ml^–1, 950 μL 2-propanol (98%, VWR) and 50 μL of Nafion 117-containing solution (5 wt %)) on a carbon paper disk (1 cm², Toray carbon paper, Aesar) with a loading of 0.06 mg cm^–2 and stored under vacuum once prepared. The WE was soaked in a fresh 1 M KOH (99.95%), 0.1 M KOH, 0.5 M Li₂SO₄, or 0.05 M H₂SO₄ solution before it was fixated in the cell by a glassy carbon plate (25 × 25 × 1 mm, HTW). The back of the glassy carbon was taped with a Cu strip (AT528, 10 mm width, RS Components) and connected to the potentiostat wires. The catholyte was saturated by purging N₂ or Ar for 30 min before each experiment. After cyclic voltammetry and chronoamperometric measurements, aliquots of both catholyte and anolyte were collected with a syringe and transferred to several test tubes for further quantification.

2.5. Minimizing Effects of Impurities

Feed gas contamination in the form of NH₃ and NO_x in both high purity Ar and N₂ have been reported previously.^15,25 In order to remove residual contaminants, a certified commercial gas filter (Entegris GPUS35FHX) was installed upstream of the electrochemical cell (see Figure S3). The cell components were always acid cleaned with 10 vol % H₂SO₄ for at least 1 h and rinsed with ultrapure water prior to each experiment. Syringes, needles, pipet tips, and sample tubes were also excessively washed with ultrapure water and dried under Ar flow before use. A microporous membrane (Celgard 3401) with a gas repellent coating was selected as a more suitable separator compared to the more commonly used Nafion membrane to avoid accumulation of ammonia contaminations as was reported previously.^15,44,45 A downstream acidified liquid trap is often used to measure volatile ammonia that could potentially be present in the effluent gas. As NH₃ dissolves very well in aqueous electrolytes (∼500 g/L), this suggests that low concentrations of NH₃ readily dissolves in the used electrolyte. This means that an acid trap is often redundant and can potentially be an extra source of contamination.¹⁵ Therefore, we did not incorporate a downstream acidified trap in the experimental design.

Precursors and catalysts containing nitrogen species are potential sources of impurities and should be avoided.^28,30 The selection criteria for our catalyst synthesis procedures was mainly motivated by minimizing the use of N-containing precursors. The Mo₂C nanoparticles and Mo₂C nanodots do not contain N-based materials for the preparation, while the use of N–C compounds was unavoidable for the synthesis of iron carbide nanomaterials. The latter motivated us to use a catalyst loading of 0.06 mg·cm^–2 to minimize the effects of the N–C precursor during the electrochemical experiments. We used a method adopted from Chen et al. to monitor impurities in our materials,²⁶ such as catalyst powders, membranes, carbon paper, and Pt foil. Strategies to effectively remove impurities will be discussed in a future study.⁴⁶

Li-salts are notorious for containing trace levels of NO_x^– species as was previously reported by Li et al.⁴⁷ Therefore, Li₂SO₄ is suspected of having these labile N-species and the suggested thermal annealing step was implemented to remove trace impurities. For the annealing step, the as received Li₂SO₄ (99.5%) was transferred to a tubular furnace and thermally annealed at 800 °C for 4 h in Ar with a heating rate of 10 °C·min^–1 before preparing a solution.

2.6. Ammonia and Nitrite Quantification

Ammonia was quantified by the Berthelot reaction.⁴⁸ In a routine analysis, a volume of 1.33 mL of either 1 M KOH, 0.1 M KOH, 0.5 M Li₂SO₄, or 0.05 M H₂SO₄ was neutralized with dilute concentrations of H₂SO₄ or KOH. Then, phenol nitroprusside and alkaline hypochlorite (0.2 wt % sodium hypochlorite in an alkaline solution) were both added in an amount equal to 25 vol % of the neutralized solution. The mixture was stirred thoroughly on a vortex shaker. After 30 min of incubation time, the solution color and its intensity differed from light green to dark blue with increasing NH₃ content. The samples were transferred to PMMA cuvettes (10 × 10 × 30 mm) for further analysis with the UV–Vis spectrophotometer (Hach DR6000). For constructing a calibration line, a series of six different concentrations of NH₄Cl (99.99%) in 1 M KOH, 0.1 M KOH, 0.5 M Li₂SO₄, and 0.05 M H₂SO₄ were prepared with respective concentrations of 0.01, 0.05, 0.1, 0.5, 1, and 2 ppm. The fitted calibration lines shown in Figure S4 were reproducible and resulted in the following linear relationships: A_1MKOH = 0.5642C_NH3 – 0.0045 with R² = 0.9997, A_0.1MKOH = 0.7279C_NH3 – 0.001 with R² = 0.9999, A_0.5MLi2SO4 = 0.7992C_NH3 – 0.0033 with R² = 0.9997, A_0.05MH2SO4 = 0.6613C_NH3 – 0.00405 with R² = 0.9997.

The concentration of NO₂^– was quantified by the photometric Griess test. A commercially available Griess reagent mixture was used with a detection range between 0.007 and 3.28 ppm NO₂^– (Spectroquant, Merck). Typically, a sample of 2 mL of 0.1 M KOH was neutralized with 168 μL of 0.5 M H₂SO₄. Subsequently, 30 mg of the Griess reagents were added and mixed with the solution with an incubation time of 10 min. Five different concentrations of 0.02, 0.05, 0.1, 0.5, 1 ppm KNO₂ in 0.1 M KOH were prepared to construct a calibration line with a perfect linear fit: A = 0.8071C_NO2^– – 0.0001 and R² = 1 (Figure S5). The UV–Vis spectroscopic measurements to detect ammonia and NO₂^– were always performed versus a blank 0.1 M KOH electrolyte stock solution to exclude the influence of electrolyte background contaminations.

3. Results and Discussion

3.1. Material Characterization

The X-ray diffraction patterns of Mo₂C NS, Mo₂C NP, Fe₃C and Fe₅C₂ are shown in Figure 1. The Mo₂C samples (Figure 1a) show three sharp peaks at 40.2, 44.3, and 46.1° that are identical to the reference spectrum of α-Mo₂C (PDF 04-003-0962). Three other peak features at 30.3, 43.2, and 63.0° suggest the existence of MoO₂ (PDF 04-013-3645) in the Mo₂C NS sample. This is most likely related to an incomplete carbothermal reduction of the molybdenum oxide precursor, which was not observed for the Mo₂C NP. The ″hill-like″ peak between 20 and 25° is typical for amorphous carbon and reflects its dominant presence in the Mo₂C NS, Fe₃C, and Fe₅C₂ samples.⁴⁹ The peaks between 45 and 60° in Figure 1b correspond to orthorhombic iron carbide (θ-Fe₃C, PDF 00-035-0772). The formation of other Fe oxidation states, such as reduced Fe (53.3°), Fe₃O₄ (41.3, 35, and 74.2°), and Fe₂O₃ (38.6°) are inevitable by-products of the carburization process.^41,50 Also, small fractions of Fe₃O₄ (41.4° and 74°) were identified in Figure 1c after the thermal decomposition of Fe(CO)₅,⁴¹ while the multiplet between 49 and 55° is very typical for χ-Fe₅C₂ (PDF 01-080-4102). The average crystallite size was calculated with the Scherrer equation (eq 1) and summarized in Table S1.

Figure 1.

X-ray diffractions patterns of (a) Mo₂C NS (green) and Mo₂C NP (blue) with the α-Mo₂C (PDF 04-003-0962, black) and MoO₂ (PDF 04-013-3645, gray) reference patterns. The patterns in (b) and (c) represent Fe₃C (purple) and Fe₅C₂ (red) with the corresponding θ-Fe₃C (PDF 00-035-0772, black) and χ-Fe₅C₂ (PDF 01-080-4102, black) reference patterns.

The fitted Mössbauer spectrum of the Fe₃C sample (Figure 2a) shows a sextuplet with an isomer shift (IS) of 0.19 mm s^–1, and a hyperfine field of 20.8 T. θ-Fe₃C is identified as the major spectral contributor (67%).⁵¹ Additionally, a metallic Fe sextuplet (IS = −0.004 mm s^–1, 33 T) was clearly observed and is in agreement with the sharp peak at 53.3° in the Fe₃C diffractogram. A doublet indicates a quadrupole peak splitting, which means the absence of magnetic field spin coupling. This indicates the presence of (super)paramagnetic Fe³⁺ nanostructures. It is difficult to allocate the specific Fe³⁺ phase, as multiple subdoublets can be superimposed in one doublet.⁵² However, the low intensity XRD peaks of Fe₂O₃ suggests that the doublet contains mostly nanostructured Fe₃O₄. The presumably low quantities of Fe₂O₃ are covered by a sextet (IS = 0.31 mm s^–1, 49.3 T); therefore, it is unlikely that Fe₂O₃ has a spectral contribution in the doublet. Three sextuplets (IS = 0.27, 0.21, and 0.16 mm s^–1 with B_hyp = 21.7, 18.1, and 10.3 T) covered 78% of the spectral area in Figure 2b, which were attributed to the three iron lattice sites in the Fe₅C₂ crystal structure.^53,54 Fe₃O₄ has a small spectral contribution located in the outer spectrum with an octahedral (IS = 0.28 mm s^–1, 49.1 T) and a tetrahedral site (IS = 0.71 mm s^–1, 46 T).⁵⁵ Fe₂O₃ was not identified in the Fe₅C₂ diffractrogram, which again suggests that the doublet is nanostructured Fe₃O₄. In conclusion, the Mössbauer data confirms the synthesis of the intended Fe₃C and Fe₅C₂ compounds with limited amounts of iron and iron oxide species. The remainder of the Mössbauer data is summarized in Table S2.

Figure 2.

Room temperature transmission ⁵⁷Fe Mössbauer spectra in the (a) prepared Fe₃C sample with a large intensity sextet (purple) identified as θ-Fe₃C and (b) synthesized Fe₅C₂ powder with three intense sextets of χ-Fe₅C₂ (I maroon, II red, III salmon).

Mössbauer spectroscopy and XRD give information about the bulk phase of the material. XPS is surface sensitive and gives information regarding the surface phase and composition. The Mo₂C XPS survey scans (Figure S6a,b) contain peaks of Mo 3d, Mo 3p, C 1s and O 1s, from which the high-resolution scans of the Mo 3d photoelectrons (Figure 3a,b) were deconvoluted to identify different Mo oxidation states. The Mo 3d orbital has a spin–orbit Mo 3d_5/2-Mo 3d_3/2 doublet with a 3/2 peak intensity ratio that is separated by a binding energy of 3.15 eV. The full width at half-maximum (FWHM) was kept constant for each doublet during the deconvolution process. It is often ambiguous to assign an oxidation state to Mo₂C; therefore, it is often denoted in an aggregated term as Mo^0–3+.⁵⁶ The presence of Mo₂C surface bonded species was confirmed by the small Mo 3d_5/2 peaks at 229.3 and 228.2 eV for both Mo₂C NS and Mo₂C NP, respectively.^34,57 Other peaks at binding energies 232.5 and 235.5 eV for both Mo-carbide materials are identified as MoO₃ and must be solely present in the thin surface layers as MoO₃ was not identified in the diffractograms. These spontaneously formed metal oxide surface layers are inevitable due to exposure to ambient air. Post-mortem XPS analyses confirmed that the majority of the surface layer was Mo₂C.^23,58 This suggests that the Mo-oxide species are reduced during electrochemical reduction. Moreover, it is expected that the trans-passive Mo-oxide layers are not stable in alkaline conditions and form soluble MoO₄^2– even at moderate reduction potentials.^59,60

Figure 3.

X-ray photoelectron spectroscopy Mo 3d and Fe 2p spectra with deconvoluted peaks in (a) Mo₂C NS (green) and (b) Mo₂C NP (blue), (c) orthorhombic Fe₃C (purple), and (d) Fe₅C₂ (red).

The Fe 2p^3/2 peak was used to identify different Fe oxidation states. The broad peak between 714 and 709 eV contains a complex convolution of multiple subpentuplets of Fe₂O₃, Fe₃O₄, and FeOOH species, which all overlap in this region. We fitted one pentuplet as a general Fe-oxide term as indicated in Figure 3c,d by taking XPS reference data such as FWHM, relative peak area, and binding energies from Biesinger et al.⁶¹ For iron carbide, the majority of the surface is covered with a thin Fe-oxide layer. According to the Pourbaix diagram for Fe, this oxide-layer is reduced by applying mild reduction potentials.⁶² The presence of a single peak at 708.4 eV for Fe₃C and 708.6 eV for Fe₅C₂ is identified as the Fe carbide phase. Only Fe₅C₂ has an additional sharp metallic Fe peak at 707.2 eV. The low signal-to-noise ratio for the Fe₅C₂ Fe 2p spectra indicates a low Fe quantity (<1 at%), which is also reflected in a low intensity Fe oxide peak in the O 1s spectra (Figure S7e). From a depth profiling test (Figure S8), it becomes clear that the Fe 2p signal increases with a longer etching time, while the intensities of the O 1s and N 1s spectra decreases. This indicates that the top surface layer is covered with adventitious species due to atmospheric exposure.

The elemental Mo and Fe content in all metal carbides were analyzed by ICP-OES and are summarized in Table S3. The ICP-OES results revealed that the bulk concentrations of Mo and Fe are significantly higher with respect to the surface concentrations estimated by XPS. This suggests that the surface adsorption of advantageous species by air exposure is not only observed for Fe₅C₂ but also for the other metal carbides.

The Mo₂C NS are clearly visible in Figure 4a,e and confirm a successful synthesis. TEM imaging (Figure S9a) reveals that a relatively large proportion of the sample consists of undecorated carbon nanosheets. This explains why the majority of the surface composition, analyzed by XPS, is predominantly carbon (Table S4). The existence of Mo₂C nanodots (<20 nm), as proposed by Wang and coworkers, was not observed in our TEM analysis.²³ Despite the magnification limitations of the low-resolution TEM, distinguished nanoparticles up to 5–10 nm were detectable in other metal carbide samples, indicating that Mo₂C nanodots of the order 10–20 nm should be visible. In Figure 4e and Figure S9b, regions with a higher contrast indicate a layer of aggregated Mo₂C, with an average crystallite size (D_XRD) of 35 nm. The SEM–EDX results (Figures S10 and S11) support this observation and show that the carbon sheet is indeed covered with a nanocrystalline layer of Mo₂C.

Figure 4.

Scanning and transmission electron micrographs of (a, e) Mo₂C NS, (b, f) Mo₂C NP on a carbon support, (c, g) Fe₃C, and (d, h) Fe₅C₂.

The successful synthesis of Mo₂C NP on a carbon support was confirmed by TEM (Figure 4f and Figure S12). The particle size, D_TEM, was distributed between 10 and 50 nm, which is in agreement with the average crystallite size (D_XRD = 21 nm). The SEM images in Figure 4b and Figure S13 show a mesoporous morphology with a large surface area. The overall Mo₂C surface distribution is homogeneous as was confirmed by the SEM–EDX mapping (Figure S14). The carbon precursor in combination with the inert SiO₂ nanoparticles stimulates the spherical growth of nanosized Fe₃C particles and prevents it from forming larger aggregates. Most Fe₃C particles were between 40 and 60 nm. The Fe₃C sample contained nanosized hollow features as visible in Figure 4c,g and confirmed the successful removal of SiO₂ during the 1 M KOH treatment. The absence of the SiO₂ nanoparticles after the treatment was further supported by Fourier transform infrared spectroscopy (Figure S15).

Small and isolated Fe₅C₂ spherical nanoparticles with a narrow size distribution are observed in Figure 4d and Figure S16 (D_TEM is 5–35 nm and D_XRD is 11 nm). This highlights that PVP successfully stabilizes the nanoparticles from agglomeration during the carbothermal reduction of Fe(CO)₅. The material has a microporous structure (Figure S17b) with a high surface area because of the polymeric nature of the support. In contrast to the low Fe content measured in the first ∼10 nm-thick surface layer, well distributed and significant Fe concentrations were detected in the bulk surface layers (∼1 μm) by SEM–EDX mapping (Figure S18), which supports the XPS depth profiling and ICP-OES results (Figure S8).

3.2. Electrochemical Characterization

The current-potential (I–V) relationship of each material was investigated by executing multiple CV cycles at a scan rate of 20 mV·s^–1 in a N₂ presaturated electrolyte to estimate the onset potential and an expected potential window for the NRR. A possible pH dependency on the Mo₂C activity of the NRR was investigated by executing CV in 0.05 M H₂SO₄ (pH = 1), 0.5 M Li₂SO₄ (pH = 8.3), 0.1 M KOH (pH 13), and 1 M KOH (pH 14). The stability of Fe-carbides in acidic-to-neutral conditions is low as the material tends to dissolve.⁶³ Therefore, we decided to only use 0.1 M and 1 M KOH for the evaluation of the Fe-carbides. Mo₂C is generally stable in both acidic and alkaline environments, allowing CV measurements in all electrolytes.^58,64 The uncompensated resistance (R_u), measured with open-circuit electrochemical impedance spectroscopy was consistent for each material tested and ranged between 25 and 30 Ω for 0.1 M KOH, 3 and 4 Ω for 1 M KOH, 12 and 13 Ω for 0.5 M Li₂SO₄, and 24 Ω for 0.05 M H₂SO₄ (Figure S19). These quantities for R_u are below the acceptable range of the R_u compensator used for all electrochemical measurements.

Figure 5 shows that all metal carbide I–V relationships in acidic, neutral, and alkaline conditions display an increase in current density at increasingly more negative potentials, characteristic for an HER I–V profile. Other distinctive reduction peaks that might be identified as the NRR were not observed in the voltammetry measurements. In addition to this, there was no indication of a metal oxide reduction peak within the examined potential window, which suggests that the metal oxide surface layer is removed immediately. Mo₂C NP reaches higher current densities compared to Mo₂C NS at all pH values, which could be explained by a larger electrochemical surface area due to the mesoporous structure of Mo₂C NP. Another explanation might be the higher Mo₂C loading content in Mo₂C NP, since the ICP-OES analysis resulted in a higher concentration of elemental Mo in Mo₂C NP.

Figure 5.

Cyclic voltammograms (6th cycle) at 20 mV·s^–1 for Mo₂C NS (green), Mo₂C NP (blue), Fe₃C (purple), Fe₅C₂ (red), and carbon paper (black) in (a) 0.1 M KOH, (b) 1 M KOH, (c) 0.5 M Li₂SO₄, and 0.05 M H₂SO₄. The gray dotted line at −0.345 V vs RHE represents the theoretically estimated onset potential for the NRR.

The onset potential is used to indicate the minimum activation potential of a redox reaction in cyclic voltammograms.⁶⁵ A theoretical approximation of the NRR onset potential in alkaline media can be calculated using the equilibrium potential and the activation overpotential. First-principles density functional theory calculations suggest that the minimum overpotential for the NRR is approximately −0.4 V, due to scaling relationships between intermediates.^2,66 The NRR equilibrium potential was calculated by equilibrium thermodynamics and is 0.054 V vs RHE, which results in a minimum required onset potential (E_onset, NRR) of −0.35 V vs RHE. Here, we estimated the experimental E_onset for different pH values by plotting the first derivative of the voltammogram (dj/dE) versus the applied potential (Figure S20). The lift-off point where the slope of the dj/dE curve starts to increase is set as the E_onset.

For 0.1 M and 1 M KOH, we do not see clear evidence for an alkaline pH effect for Mo₂C materials as both Mo₂C NP and NS have a similar I–V curve at both KOH concentrations. The onset potential for Mo₂C NP of −0.11 V vs RHE is similar for both electrolytes and is in agreement with earlier observations.⁶⁴ Moreover, the onset potentials for Fe₃C and Fe₅C₂ are also similar and varied between −0.22 and −0.23 V vs RHE for both 0.1 M and 1 M KOH. The I–V relationship in Figure 5c for Mo₂C in 0.5 M Li₂SO₄ is remarkably different showing a more negative onset potential of −0.28 V vs RHE and −0.32 V vs RHE for Mo₂C NP and NS, respectively. As a consequence, the activation overpotential at −10 mA·cm^–2 for Mo₂C NP is −0.25 V lower than in alkaline conditions, which can be related to the low availability of either protons or hydroxide ions. At acidic conditions (pH = 1), Mo₂C NP displays a similar I–V relationship with respect to alkaline conditions, which highlights the unique properties of Mo₂C showing similar catalytic activity in both acidic and alkaline conditions.⁶⁴ However, the onset potential is slightly more negative (−0.17 V vs RHE), indicating that the catalyst is more active in alkaline conditions.

E_onset for Mo₂C is above the theoretically estimated threshold in both acidic and alkaline pH, where the current density E_onset,NRR is roughly −4 mA·cm^–2 for Mo₂C NP. From this analysis, it is unlikely that the NRR is a dominant contributor to the I–V profile of Mo₂C because the HER kinetics are more facile in these conditions. Interestingly, E_onset for Mo₂C in 0.5 M Li₂SO₄ is below E_onset,NRR and suggests that operating at near-neutral conditions might be ideal for the NRR. It is important to note that Cheng and coworkers reported high NH₃ yields with Mo₂C NS using the same electrolyte.²³ For Fe₃C and Fe₅C₂, the majority of the I–V profile exceeds E_onset,NRR, suggesting that both iron carbides might be promising catalysts for the NRR.

The NRR activity of the metal carbides was qualitatively screened by measuring the ammonia concentration after executing 40 scans of cyclic voltammetry with a scan rate of 20 mV s^–1. The results are summarized in Figure S21 and show that NH₃ concentrations for Mo₂C in acidic-neutral pH is close to the detection limit <30 ppb, while levels up to 100 ppb were observed in alkaline conditions. Operating at alkaline conditions is therefore more beneficial for studying the NRR, and subsequently, the main electrochemical experiments herein were performed in alkaline conditions.

To ultimately verify Mo- and Fe-carbides as conceivable NRR catalysts, a series of 2 h chronoamperometry (CA) measurements were performed at five different potentials in a N₂-saturated 0.1 M KOH electrolyte. The current densities for the metal carbides are stable in alkaline conditions as illustrated in Figure S22. As a comparison, two additional CA measurements in Ar-saturated electrolytes were performed with Mo₂C, which gave slightly higher current densities for two potentials. The difference in current density is however rather small and can be caused by slight variations in the experiments. This observation was also made elsewhere,²⁸ questioning the reliability of N₂ vs Ar voltammetry and CA experiments as an initial indicator for successful dinitrogen reduction.

3.3. NRR Measurements

CA measurements were used to further assess the activity toward the NRR. After each CA, aliquots of both catholyte and anolyte were taken from the cell for further quantification of NH₃ and NO₂^–. The amount of quantified NH₃ after each experiment varied between 37 and 123 ppb with no particular trend linking ammonia concentration and applied potential over time. A 2 h open circuit potential (OCV) test with a N₂-saturated electrolyte was used to obtain insights on the amount of impurities coming from either the feed gas stream or surface adsorbed species inside the cell. The OCV results for Mo₂C reveal a similar NH₃ concentration as obtained with the chronoamperometry experiments. The impact of feed gas impurities can be excluded due to the installed certified gas filter (<100 ppt) in front of the cell. It is more likely that adsorbed NH₃ in the cell components is released during the OCV experiments and inevitably during the NRR measurements. Long term CA experiments with an Ar-saturated electrolyte are useful to study the possible release of N-impurities from the catalyst and other cell components exposed to the electrolyte under electrochemical conditions. For Mo₂C, three CA experiments with Ar-saturated electrolyte at −0.20, −0.31, and −0.44 V vs RHE resulted in a somewhat lower NH₃ content (80, 53, and 60 ppb) compared to experiments with N₂-saturated electrolytes. Again, it is deemed unlikely that purified Ar (and N₂) introduces feed gas impurities. Therefore, this observation suggests that the Ar gas flowing through the electrolyte stripped a small part of the dissolved NH₃ from the electrolyte. Nevertheless, both the N₂ OCV and Ar CA experiments indicate that the majority of the quantified NH₃ is not from the NRR but originates from contaminations. Additional control experiments with ¹⁵N₂-labeled gas were not performed since the observed NH₃ concentrations were below or approximating the background level.

Small quantities of NO₂^– were detected after all CA experiments, suggesting that a part of the quantified NH₃ potentially stems from NO_x reduction. Jiao and coworkers observed that the electrochemical reduction of NO_x forms multiple N-products, such as ammonia, hydroxylamine, N₂, and N₂O depending on the transition metal.⁶⁷ Pt is more selective toward NH₃, which was also supported by Koper and coworkers who made a similar observation for NO₃^– electroreduction on Pt.⁶⁸ Mo₂C has similar noble metal-like properties as platinum;⁶⁹ therefore, it is reasonable to assume that NO_x species are reduced to ammonia at the investigated potentials. Fe₃C is also an efficient nitrate reduction electrocatalyst, where a previous study reported faradaic efficiencies (FE) higher than 90% to NH₃ at moderate reduction potentials.⁷⁰

Despite the thorough cleaning efforts for every part of the cell, a well-established background level of both NH₃ and NO₂^– was always observed after each experiment. We decided to analyze the removal efficiencies of our cleaning methods (elaborated in the caption of Figure S23) and found that NH₃ was sufficiently removed by simply rinsing with water. Surprisingly, significant quantities of released NO₂^– were detected that originated from the Celgard 3401 membrane, carbon paper, and Pt foil. This is a valuable observation, as two previous studies advised substituting the Nafion membrane with a microporous Celgard membrane to reduce NH₃ contaminations.^15,45 Our results indicate that NO₂^– is not only a surface adsorbed species but is also present in the inner membrane and carbon paper structure and is problematic to remove. Generally, the amount of released NO₂^– depends mainly on the exposed surface area, meaning that it can be lowered significantly by optimizing the cell design. Investigating the origin of the observed NO_x impurities is out of the scope of the present work and will be addressed in an upcoming study.⁴⁶

3.4. Literature Comparison

Previous studies using Mo- and Fe-carbides as NRR electrocatalysts are shown in a comparative overview (Figure 7a), including our own observations. It becomes clear that both our Fe₃C and Fe₅C₂ quantified NH₃ yields are within the NH₃ background level. The Mo₂C catalysts exceed this threshold slightly, but with a significant NO_x background, it becomes unlikely that any nitrogen reduction to NH₃ occurred. The study of Cheng and coworkers outperformed our Mo₂C NS, observing a 240 times higher NH₃ yield.²³ This motivated us to execute a direct comparison by increasing the catalyst loading to 3 mg·cm^–2 and using 0.5 M Li₂SO₄. The chronoamperometry measurements were comparable, but our NH₃ concentrations were below 100 ppb and close to the earlier defined background level as displayed in Figure 6. This is additional proof that Mo₂C cannot be perceived as a promising NRR catalyst.

Figure 7.

Literature overview of recently published NRR studies using Mo- and Fe-carbides as electrocatalysts differentiated by either high N-source in the support or used during synthesis (square) and free of N-source (triangular). The symbols are filled in case of quantitative ¹⁵N₂-labeled experiments and half-filled if analyzed qualitatively. Our own results are included as Mo₂C NS* (green), Mo₂C NP* (blue), Fe₃C* (purple), and Fe₅C₂* (red). (a) Faradaic efficiency vs NH₃ yield with thick line (gray) indicates the estimated NH₃ background level and the dotted line presents a hypothetical background level including the measured NO₂^– from Figure S24. (b) Faradaic efficiency vs j_NH3. More details regarding the literature studies included in the figure can be found in Table S4.

Figure 6.

NH₃ concentration measured from the electrolyte after two hour CA experiments in 0.1 M KOH. (a) Mo₂C NS (green, spherical) and Mo₂C NP in N₂ (blue, rectangular) and Ar (open rectangular). (b) Fe₃C (purple, spherical) and Fe₅C₂ (red, rectangular). The data points with the error bars were done in duplicates.

The majority of the earlier published literature observed orders of magnitude higher yields and FEs compared to this work. Firstly, all the literature studies shown in the overview did not quantify or consider NO_x as an influential factor on their measured NH₃ content. Secondly, the impurities in the feed gas stream were not removed by the installment of a certified gas filter. This is especially important when performing ¹⁵N₂-isotope labeled experiments as traces of ¹⁵N-labeled impurities (¹⁵NH₃ and ¹⁵NO_x) have been identified in several ¹⁵N₂-gas bottles.^15,25,71,72 Before using Li-based electrolytes for NRR experiments, Li-salts must be thermally annealed at 800 °C under inert conditions to remove trace levels (>1 ppm) of NO_x^– impurities.⁴⁷ We followed this procedure, while others, including Cheng et al., did not consider this extraneous source of impurities, and this might be one of the main factors contributing to their high NH₃ yields.^23,31,36 As final point, the Nafion membrane commonly applied in these studies is known for the uptake and release of NH₃ during electrochemical experiments.¹⁵ Substituting the Nafion membrane with another membrane is not straightforward as we detected a significant amount of NO₂^– in the microporous membranes (Celgard 3401), but selecting a suitable treatment method is advised.⁷³

Control experiments become even more essential when catalysts have a high N-content, such as metal nitrides, N-doped supports, or leftover NO_x/NH₃ traces from the catalyst synthesis. Evidence was found that for catalysts with a high N content, such as VN and Nb₄N₅, the decomposition of the N-atomic lattice in acidic media released significant amounts of NH₄⁺ during the initial stages of the electrochemical experiment.²⁸ Similar observations were also reported for Mo₂N.³⁰ Additionally, several commercially available metal oxide powders, such as Fe₂O₃ and Bi₂O₃, released a large amount of NO_x impurities. This eventually led to the retraction of a study, as it was proven that the origin of observed NH₃ was from NO_x reduction and not the NRR.^26,74 Therefore, we decided to analyze the N-content of all four materials by XPS and UV–Vis spectroscopy (method described in the SI). The N 1s spectra of Mo₂C could not be identified because of overlapping peak features with the Mo 3p orbital. Nevertheless, the absence of N KLL Auger peaks in both Mo₂C NS and NP XPS surveys (Figure S6a,b) indicate that the N-content might be negligible. Figure S7c,f shows two distinct N 1s peaks for Fe₃C (398.3 and 399.9 eV) and Fe₅C₂ (398.9 and 400.1 eV), suggesting pyridinic N–C and pyrrolic N–C bonds from the precursor (4.5-dicyanoimidazole and PVP).⁷⁵ Nevertheless, the samples were exposed to air before XPS analysis, indicating that the peaks could be also from adventitious N species, such as −NH₂, which have similar binding energies.⁷⁵⁻⁷⁷ It is therefore challenging to assign these peaks to a specific N-functional group. From the spectrophotometric analysis (Figure S25), directly performed after the material synthesis, it becomes clear that NH₃ impurities from an unidentifiable source were present in all catalysts (Mo₂C NS = 8.9 μmol_NH3 g_cat^–1 , Mo₂C NP = 16.5 μmol_NH3 g_cat^–1 , Fe₃C = 21.9 μmol_NH3 g_cat^–1, Fe₅C₂ = 4.5 μmol_NH3 g_cat^–1). This effect was suppressed by using a low catalyst loading (0.06 mg) for each experiment. In the case of the most contaminated sample, the expected release of NH₃ from 0.06 mg Fe₃C is limited to a negligible 1.3 nmol. Nevertheless, the NRR measurements performed with 3 mg Mo₂C NS did not result in an increase in the NH₃ concentration. It remains unlikely that impurities in the catalyst resulted in exceptionally high NH₃ yield reported by Cheng et al.²³ This suggests that other factors lead to their positive result.

The NH₃ partial current density, j_NH3, is a useful performance indicator, wherein cases with j_NH3 smaller than 100 μA cm^–1 are too low to be promising. From a back-of-the-envelope calculation, we estimated that the NH₃ concentration at j_NH3=100 μA cm^–1 is in the 1 ppm order of magnitude range assuming typical parameters, such as A_WE = 1 cm², V_catholyte = 20 mL, and t_CP = 1 h. These levels of NH₃ can easily be reached when the earlier mentioned sources of contamination are not identified or even considered. This has implications on the reliability and usefulness of reporting the FE, wherein the focus should be initially on j_NH3 or the NH₃ yield rate. From Figure 7b, it becomes clear that most literature studies did not exceed 100 μA cm^–1, while a FE > 20% was reported (see Table S4). Therefore, we suggest that future publications explicitly report the NH₃ partial current density as the main catalyst performance indicator.

Interestingly, the role of metal carbides is also under debate for hydrazine oxidation.⁷⁸ Fe–N–C catalysts are common used catalysts for this reaction and contain iron carbides because of the high temperature pyrolysis required for the synthesis. Early studies claimed that Fe₃C plays an active role in the reaction,^79,80 while a recent study revealed the true role of Fe₃C by executing a rigorous comparison study between Fe–N–C materials with different amounts of Fe₃C.⁷⁸ This approach led to the conclusion that Fe₃C is mostly inactive for hydrazine oxidation, and should be removed by nonoxidizing acid solutions. This is yet another example of how a rigorous and well-designed experimental procedure can aid in clarifying the activity of electrocatalysts for reactions in the nitrogen cycle.

4. Conclusions

Nanostructured molybdenum carbide and iron carbide were reported earlier as promising electrochemical nitrogen reduction catalysts. In this study, the NRR activity of both molybdenum and iron carbide materials were reassessed with the implementation of a strict experimental protocol that allowed us to reduce the effects of extraneous impurities to a bare minimum and identify false positives. The successful synthesis of nanostructured Mo₂C, Fe₃C, and Fe₅C₂ was confirmed by X-ray diffraction, scanning and transmission electron microscopy, and X-ray photoelectron and Mössbauer spectroscopy. The current–potential relationship of the metal carbides is characteristic for the HER, where the current increases with increasing negative overpotential. Moreover, specific reduction peaks that could be related to the NRR were not identified. NH₃ quantification was done after 40 scans of cyclic voltammetry, where we indeed measured NH₃ (50–100 ppb) for both Mo₂C and Fe₅C₂ in alkaline conditions. To further assess the NRR catalytic activity of molybdenum and iron carbides, we performed a series of 2 h chronoamperometry measurements at different potentials in N₂-saturated 0.1 M KOH. For Mo₂C NP and NS, the NH₃ concentration was between 41 and 124 ppb, exceeding the NH₃ background level (84–88 ppb) for potentials at −0.2, −0.31, and −0.46 V vs RHE. We noticed that the yield earlier reported by Cheng et al. was considerable higher than measured with our Mo₂C NS.²³ A direct comparison by performing chronoamperometry experiments with an increased loading (3 mg·cm^–2) and 0.5 Li₂SO₄ did not result in elevated NH₃ concentrations. This is additional proof that Mo₂C cannot be conceived as a promising NRR catalyst. The NO_x content after the NRR, Ar, and OCV blank tests revealed NO₂^– concentrations in the same order of magnitude (55–122 ppb). This implies that NH₃ arises from NO₂^– reduction and not from the NRR. These NO₂^– impurities originated from the Celgard membrane, since we found that the membrane, even after rinsing excessively with water, released a considerable amount of NO₂^– impurities (109 ± 31 ppb). This emphasizes the importance of NO_x monitoring, which is often overlooked in the literature and might result in a false positive. The quantified NH₃ from the iron carbide catalysts did not exceed the NH₃ background level, indicating that these materials are not active for the NRR. With our experimental approach, we succeeded in establishing a minimized and reproducible background level that allowed us to critically assess promising NRR catalysts. We believe that our methods and detailed analysis will equip researchers entering the field with clear guidelines to perform NRR experiments in a more reliable manner.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acscatal.2c04491.

• Additional information on the experimental setup, UV–vis calibration curves, XRD analysis, Mössbauer data, XPS, SEM–EDX, TEM, ICP-OES, FT-IR, electrochemical data, impurity assessment, and a literature summary (PDF)

The authors declare no competing financial interest.