Ammonia synthesis by plasma catalysis in an atmospheric RF helium plasma

The in-plasma-catalytic synthesis of ammonia from nitrogen and hydrogen admixed to a helium RF plasma is studied with infrared absorption spectroscopy, optical emission spectroscopy, and through chemical kinetics modeling. Sandblasted glass is used as support for iron, platinum, and copper catalysts up to a surface temperature of 150  ∘C . It is shown that the optimum ammonia production occurs at very small N2/(N2+H2) ratios with an increase of concentration with plasma power. The global kinetic modelling agrees well with the data for a variation of the N2+H2 admixture and the absorbed plasma power. The introduction of the catalyst enhances ammonia production by up to a factor of 2. Based on the comparison with the modelling, this is linked to a change in the electron kinetics due to the presence of the catalyst. It is postulated that introducing the catalyst increases the reduced electric field because it reduces the secondary electron emission coefficient. As a result, the dissociation of N2 is stimulated, thereby enhancing the NH3 formation. These experiments show that the impact of the catalyst on the plasma performance in noble gas-diluted RF plasmas can be more important than the impact of the plasma on any catalytic surface process.


Introduction
Ammonia is a widely-used chemical in the industry and essential for the production of nitrogen based fertilisers.Its largescale production from methane and steam is based on the Haber Bosch (HB) process.This involves producing a significant amount of CO 2 that should be avoided for addressing the climate change challenge.Therefore, many alternative methods for ammonia synthesis are being explored.Ammonia synthesis in a plasma reactor is one of the candidates for smallscale, on-demand ammonia production, which is needed for a decentralised energy infrastructure.Non-thermal plasmas have a lower theoretical limit for energy consumption than Original Content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence.Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
HB and can easily be switched on-off [1].Currently, dielectric barrier discharges-in filamentary mode-are the most studied plasma sources for ammonia synthesis, since they show promising results and are operated at atmospheric pressure [2][3][4].In such systems, NH 3 is predominantly produced by surface reactions [5,6], which motivated the combination of plasmas with catalysts.This combination, however, is difficult to characterise, because changing the surface properties-e.g. by introducing a metallic catalysts-has a significant impact on the discharge properties [2,3,[7][8][9][10].This makes it difficult to separate plasma and plasma-catalytic effects in NH 3 formation.
A valuable contribution to the understanding of plasmabased ammonia synthesis comes from modelling NH 3 synthesis.Usually, ions and the electronic and vibrational excitation of species are included to explain the experimental results [11][12][13][14][15][16], which creates a complex model.For instance, Mehta et al [17] postulated-based on a microkinetic modelthat vibrational excitation of nitrogen stimulates dissociative adsorption to allow for a lower synthesis temperature in plasma synthesis in comparison to regular thermal catalysis.However, recent findings question the importance of this excitation pathway [18][19][20].
The contribution of catalytic surface processes in plasma synthesis of NH 3 has been analysed by several authors.For example, Bayer et al [19] demonstrated that surface reactions are faster and more selective towards ammonia.Several catalysts are explored in the literature: (i) copper is an inexpensive known catalyst for plasma-synthesis of NH 3 [3,[21][22][23] and can also catalyse nitrate reduction to ammonia [24,25] and the oxidation of ammonia to nitrates [26].Unfortunately, there are some stability issues with copper as a hydrogen plasma easily reduces the oxidised copper, which alters the properties of the coated surface.(ii) Iron is commonly used in thermal catalysis and has very different adsorption energies compared to copper.Its dissociative N 2 adsorption energy is similar to ruthenium, which is known to be one of the best catalytic materials for NH 3 synthesis [27].Also, iron catalyst seems to yield the highest product yield in low-pressure RF plasmas [28].(iii) Finally, platinum has a nitrogen binding energy in between copper and iron [29] and was used in a hybrid electrochemical plasma reactor [30].Moreover, the dissociative N 2 adsorption is suggested to be stimulated by vibrational excitation for Pt and Cu but not for Fe, as the effect of vibrational excitation relies on the N-binding energy [17].Therefore, differences in the yield between these materials can be used to study the importance of improved nitrogen dissociative adsorption on the surface.
The aim of this study is to characterise the effect of catalytic surfaces on ammonia plasma synthesis in a simplified reactive chemical environment.This is achieved by igniting an atmospheric RF discharge in helium and mixing a small amount of N 2 and H 2 to the gas flow.This capacitively coupled RF plasma is based on the widely studied COST-jet [31][32][33][34].Two glass plates cover the electrodes and suppress the transition to a filamentary mode, thereby increasing the power range at which the gas conversion can be studied.This reactor was previously used to study CO 2 dissociation and oxidation of n-butane [35][36][37].The ammonia concentration is quantified with Fourier transform infrared (FTIR) absorption spectroscopy, and the excitation of the plasma is characterised by studying the power coupling into the plasma and using optical emission spectroscopy.The experimental results using different catalysts are compared to a kinetic model.

Plasma reactor
Figure 1 illustrates the experimental configuration.The plasma reactor is already presented [36], thus only a brief overview is given here.The reactor is enclosed in a low-pressure vessel, to minimise contamination from air.The front view of the reactor is sketched in figure 1(a).The plasma is ignited in a 1 mm gap, between two electrodes covered by glass plates.This creates a homogeneous plasma with a volume of 26 × 13 × 1 mm.The electrodes and glass plates are isolated from the metal housing using macro plates.One electrode is powered by an RF power generator (RFG100-13, Coaxial Power Systems) and the other one is grounded.An manual Impedance Matching Network (MMN 50-600 W, Coaxial Power Systems Ltd) is used to match the impedance of the reactor with the output of the generator.An in-house circuit is used to measure the voltage and current.Finally, the temperature of the electrodes is controlled with an oil circuit, which goes through the plasma chamber metal body (RE415S, LAUDA), see figure 1(b).
The glass plate of the bottom electrode is modified and coated with a catalyst.Details about the catalyst preparation procedure are described in [36,38].Nanoparticles of copper (25 nm, 774081-5G, Sigma-Aldrich), iron (35-45 nm, 746843-5G, Sigma-Aldrich), and platinum (200 nm, 771937-250MG, Sigma-Aldrich) are used.Despite that the copper and iron nanoparticles likely oxidise in air, the oxidation is presumed to be reversed by the reducing nature of the hydrogen admixture to the plasma.The coating is assumed to evenly cover the entire surface of the glass plate, following the laser scanning microscope images of Peters et al [38].However, on a smaller scale, the particles can be described as perfect spheres sticking on the surface.A loading of 3.0 mg cm −2 is used for copper and iron and 0.3 mg cm −2 for platinum.A lower loading for platinum is deemed sufficient since noble metals are usually very reactive.This gives an maximum specific surface area of 800, 510-650, and 4.2 cm 2 per cm 2 of glass plate for Cu, Fe, and Pt, respectively [39].
The gases are obtained from a helium gas bottle (99.999%He, Air Liquide), nitrogen gas bottle (99.999%N 2 , Air Liquide), and a hydrogen gas can (99.999%H 2 , Messer).A flow of 250 sccm of helium is mixed with about 2.5 sccm of nitrogen and hydrogen by mass flow controllers (MFC).The error bar of the gas mixture is estimated from the uncertainty of the MFC.The hydrogen fraction in the molecular gas mixture f H2 is defined by selecting specific hydrogen to nitrogen mixing ratio expressed as f H2 = ϕ H2 /(ϕ N2 + ϕ H2 ) for given flows ϕ H2 and ϕ N2 .The total molecular gas admixture c admix is defined as the total gas admixture ϕ N2 + ϕ H2 to the helium flow as c admix = (ϕ H2 + ϕ H2 )/ϕ He for given helium flow ϕ He .

Optical emission spectroscopy
Optical emission spectroscopy is a standard tool to investigate excitation pathways in the plasma.The electron properties and densities of helium metastables (He * ) are probed by observing the N 2 (C) and N + 2 (B) species, using the emission of the second positive system (SPS) and first negative system (FNS), respectively.
A spectrograph (SR-750-B1, ANDOR) is used together with an ICCD camera (ISTAR-SCMOS-18U-E4) to observe the light emitted by the plasma.The light is collected by placing an optical fibre (03345-REV.B, CeramOptec) in front of the plasma.The observed area of the reactor is indicated with a dashed circle in figure 1(a) with a radius of about 4-5 mm.Thereby 15%-20% of the plasma is probed.
Figure 2 gives a typical observed emission spectrum of the SPS, where each vibrational band is identified by its electronic and vibrational transition.The density of N + 2 (B) is related to the maximum of the FNS 0 → 0 transition.The density of N 2 (C) is estimated in arbitrary units from the summation of the emissions over the vibrational levels (up to v ′ = 4).The densities of the respective vibrational levels n v ′ are obtained from the N 2 (C,v ′ →B,v ′ ′ = v ′ +2) transitions as, where C is a constant, λ v ′ ,v ′ ′ is the transition wavelength, q v ′ ,v ′ ′ is the Franck-Condon factor, and S v ′ ,v ′ ′ is the signal from the respective vibrational band.The density of N 2 (C) is strongly related to the electron density and the fraction of electrons with an energy above 10 eV, following the model described by Dilecce et al [40].In this model, N 2 (C) is created by electron impact from the ground state of N 2 .Pooling reactions may be omitted under our conditions because only a small amount of N 2 is admitted to the plasma.The loss channel consists of radiative and collisional relaxation, where the latter depends on the gas mixture.
The density of N + 2 (B) is strongly related to the electron density and the fraction of electrons with an energy above 20 eV.Since the excitation energies of N + 2 (B) and of He * are very close, trends in the N + 2 (B) density should also reflect trends in the He * density.This is important, because, especially at low nitrogen admixtures, Penning ionisation of N 2 by He * is a dominant process [41][42][43].

NH 3 concentration quantification
The products of the plasma are measured ex-situ by FTIR (Vertext 70v, Bruker).A multipass cell (A136/2-LT, Bruker) is used to increase the sensitivity of the FTIR measurement by increasing the path length to 6.4 m.This path length is tested using a known CO 2 /He gas mixture.The cell is connected to the exhaust of the plasma, as illustrated in figure 1(b).The temperature of the cell is controlled at 180 • C with a heating jacket.An elevated temperature is chosen to lower the acquisition time by reducing the ammonia adsorption at the inside walls of the multipass cell, which prolongs the acquisition time.Simultaneously, this temperature is not high enough to induce thermal dissociation, since temperatures higher than 500 K are typically required [44].A typical NH 3 transmittance spectrum with the corresponding best fit, where T cell is 180 • C. The best fit for the molar fraction is 30.9 ppm and for the instrumental full-width-half-maximum is 0.266 cm −1 .The molar fraction of ammonia is obtained by fitting the vibrational bands between 800-1200 cm −1 using the HITRAN database [45,46] and a python script [47].A typical spectrum plus the corresponding best fit is plotted in figure 3, where the bottom figure gives the residual (= data-fit).The baseline is corrected for intensity fluctuations.The abundance of the isotopologue 14 NH 3 and 15 NH 3 is adjusted according to the natural abundances of 14 N 2 and 15 N 14 N. Furthermore, the line profile is approximated considering Doppler, pressure, and instrumental broadening.Since the gas mixture mostly exist of helium, the pressure broadening is calculated assuming helium as collision partner.The instrumental broadening of the FTIR spectrometer is approximated by a Gaussian line shape [48], where the line-width σ instr is a fit parameter.Table 1 gives the parameters used for the fitting of the transmittance spectra.Finally, the uncertainty of the best-fit concentration is quantified using the χ 2 red -method [48].The NH 3 concentrations are expressed as molar fractions, since such a fraction is independent of the gas temperature.Therefore, no correction factor must be used to extrapolate the values measured in the multipass cell back to the concentration exiting the plasma.The presented concentrations are obtained after an extended plasma operation time, which is in the order of hours.The long acquisition time guarantees that these concentrations result steady-state values.This is of importance for the kinetic model of the plasma chemistry.

Kinetic model
The measured quantities are compared with the output of a kinetic model of the chemistry.The model is based on the plug-flow scheme, where the time and spatial dimensions are related to each other.The chemistry set is based on models from literature [11][12][13][14][15]43] and it is explained in more detail in appendix.The gas kinetics of the plasma are combined with the surface kinetics based on Eley-Rideal (ER) and Langmuir-Hinshelwood (LH) reactions.
Including the surface kinetics in a plug flow model is not as straightforward as the gas kinetics.Unlike gaseous species, the adsorbed species do not flow through the reactor.Therefore, to better represent the reality of the surface kinetics, the model is solved until surface coverages reach a steady state.This is obtained by iteratively running the model while updating the initial surface coverages until the surface coverages are constant in time.The introduction of the metallic coating is considered by varying the sticking coefficients and diffusion energy barriers [14].The gas flow is not expected to change due to the surface modifications, since the coating thickness is expected to be much smaller than the gas gap distance [38].Hence, such an effect is not considered.
The complex chemistry of the helium-diluted N 2 +H 2 plasma is simplified by excluding electronic, vibrational, and rotational excitation of hydrogen and nitrogen.Approximations of the chemistry set are known to be possible without reducing the accuracy of the model [49].The low reduced electric field and the quenching of excited species at atmospheric pressure supports this simplification in our plasma system [19].Moreover, N 2 dissociation in the plasma volume is often the rate-limiting step for ammonia production.According to recent findings, the enhanced dissociative adsorption of N 2 by vibrational excitation does not contribute to the NH 3 formation [18][19][20].Rotational excitation merely leads to gas heating and does not directly enhance dissociation or dissociative adsorption.In contrast, a metastable state of helium is included.As previously mentioned, He(2 3 S) is able to ionise nitrogen molecules very efficiently, especially at low N 2 -admixtures as in this study.At our conditions, the N + 2 ion most likely relaxes back to N 2 , via N + 2 +N 2 +M→N + 4 +M with the subsequent dissociative recombination with electrons [50].Yet, the formation of atomic nitrogen via He * +N 2 →He+2 N is included.This is done to consider the impact of electronically excited nitrogen on the N 2dissociation process, which is discussed in more detail in appendix.
Finally, the electron kinetics are approximated by a fixed electron density and an electron energy distribution function (EEDF).The EEDF is calculated with LoKI-B [51,52].Following earlier results, the absorbed plasma power is set linearly proportional to the electron density [31,53], using n e (5 W) =1 • 10 11 cm -3 , and should not affect the electron energy distribution [54].

Results
The formation mechanisms of NH 3 are studied by varying the hydrogen fraction in the molecular gas mixture f H2 , the total molecular gas admixture c admix , and the power.The surfaces are either regular glass (blank), sandblasted glass, or sandblasted glass with a metallic catalyst with a loading of 0.3 mg cm −2 Pt or 3.0 mg cm −2 Fe or Cu.

Plasma performance
The plasma properties are studied by examining the power coupling and the emission of the N 2 (C) and N + 2 (B) states, which are strongly related to the electron properties of the plasma.Figure 4(a) shows the power versus voltage curves for the different f H2 , as well as for a pure helium discharge.The near-linear relation between power and voltage is unaffected by changing f H2 , and is very similar to the pure helium discharge.This indicates that there is no change in the discharge mode, which is consistent with literature [34,[55][56][57].
A varying shift in the required voltage for plasma ignition is observed with varying f H2 .The required root mean square voltage to attain a plasma power of 4 W decreases when increasing f H2 .This voltage shift agrees with the Paschen criterion for plasma ignition, where the required ignition voltage increases with the electron-neutral cross section [58].Since H 2 is more difficult to ignite than He, a voltage shift is seen between the pure helium and the pure hydrogen discharge (f H2 = 1).When changing the gas mixture by adding nitrogen (from f H2 = 1 to 0), this shift increases due to the impact of N 2 to the electron kinetics.Summarizing, one can state that lower voltages are required to obtain the same plasma power when increasing f H2 , but the discharge mode itself remains unaffected.
The emission of the plasma is strongly related to the EEDF.In figure 5, the trends of N 2 (C)/N 2 and N + 2 (B)/N 2 density ratios are plotted with increasing power at different gas mixtures and surfaces.Also, a linear relation is fitted to the data (dashed-dotted and dotted lines for the blank and Fe-coated measurements, respectively).The linear relation between these density ratios and power agrees with the assumption that the electron density is linearly proportional to the plasma power.Thereby, a simple global model of the plasma kinetics when modelling the power scan is justified.
The gas mixture has a notable effect on the densities of both excited species.The N 2 (C) emission depends on collisional relaxation since N 2 (C) is quenched more efficiently by H 2 compared to N 2 [59].This explains the decrease in the slope from f H2 = 0% to 50% and 75%.On the other hand, the increase from 75% to 95% is attributed to a more energetic EEDF with increasing f H2 , see figure A1.This indicates that the electron dynamics are strongly affected when changing f H2 .
Furthermore, Penning ionisation by He * is a dominant excitation pathway.The trends of N + 2 (B) follow the known trend of He * with nitrogen admixture, where the He * density is inversely proportional to the nitrogen admixture [42,50].Finally, the marginal difference in the power-voltage curves-in figure 4(b)-and in the emission data between using the blank and the catalyst surfaces indicates that the plasma is affected when introducing a catalytic material, but also that these changes are rather small.In summary, one can state that f H2 has a significant effect on the plasma properties, whereas changing the surface morphology only slightly affects the electron kinetics.

Plasma synthesis of ammonia
The NH 3 fraction is plotted in figure 6 when varying f H2 for c admix = 1% and 4 W (a), c admix for f H2 = 95% and 4 W (b), and the plasma power for f H2 = 95% and c admix = 1% (c) for different surfaces.The lines in figure 6 follow the NH 3 fraction for a constant hydrogen conversion χ H ≡ 3[NH3]out 2[H2] in or energy yield (EY), as indicated in the figure.The EY describes the energy required to create 1 g of NH 3 .The dashed line indicates the lower limit for the blank experiment, the dashed-dotted line the upper bound for the blank experiment, and the dotted line the upper bound when including a catalytic material in the reactor for both χ H and EY.
Furthermore, a repeatability study shows that the experiments are reproducible within ±2.5 ppm.This is estimated from the statistical variation at (f H2 , c admix , power) of (95%, 1%, 4.0 W).The test of reproducibility indicates a larger scatter of the data compared to the estimated uncertainty of the measurement techniques themselves.This is attributed to the large sensitivity of ammonia production on f H2 and maybe also to the impact of minor impurities in the reactor.
First, we regard the data of the experiments using regular glass plates.Increasing f H2 leads to an increase of both the NH 3 fraction and χ H . Increasing c admix and power leads to an increase of the NH 3 fraction as well, but also to a lower χ H and energy yield, respectively.These trends indicate that NH 3 formation is favoured at high H 2 concentrations but is less efficient at higher total gas admixtures and powers.
Secondly, we regard the data of the experiments using sandblasted glass and observe only a marginal increase in the NH 3 fraction.Only at higher powers, the difference exceeds the estimated uncertainties, e.g. up to 15% at 7 W.This could be caused by a larger surface area of the sandblasted glass plates.Also, this indicates that surface processes are contributing to ammonia production.
Finally, we regard the data of the experiments when introducing a catalyst and observe an overall increase in the NH 3 fraction.According to figure 6(c), this becomes more significant at higher powers, which is in line with the blank versus sandblasted data points.On top of this, especially when introducing the iron catalyst, a smaller f H2 leads to a larger difference in ammonia production when comparing the blank and the catalytic surface.For instance, there is an 55% increase at f H2 = 95% but more than a 100% increase at f H2 = 50%.Consequently, the slope of χ H with f H2 is smaller when introducing a metallic catalyst, e.g. the ammonia production for Fe catalysts nearly follows the χ H = 0.43% line.
The results show that the state of the surface has a significant impact on NH 3 formation.Therefore, the temperature of the reactor varied to test for any contribution of thermally activated reaction steps within the accessible temperature range of the setup.In figure 7, the change in the ammonia concentration when increasing the reactor temperature from 20 • C up to 160 • C is plotted for the blank and Fecoated surfaces and for f H2 = 50% (production of 5 ppm for the blank, 12 ppm for the Fe catalyst at room temperature) and 95% (production of 18 ppm for the blank, 27 ppm for the Fe catalyst at room temperature).Only a very small increase in ammonia production by 10% for the Fe catalyst at high temperatures is observed, although the uncertainty margin of all data is of the same order.
To conclude, the comparison between the plasma and plasma-catalytic data shows that the NH 3 production is enhanced by the presence of the metallic catalyst.This is more evident at lower f H2 , i.e. higher N 2 admixtures to the molecular gas stream.

Discussion
A kinetic model is used to clarify the underlying chemistry.The set of rate equations for volume and surface processes are listed in appendix.According to the model, no changes in the NH 3 fraction are expected in the effluent of the plasma, e.g.due to loss reactions.Therefore, the presented data for the NH 3 fractions should correspond to densities at the end of the plug flow plasma channel at the residence time t = t res .The EEDF is calculated for varying E/N and f H2 for c admix = 1% and 4 W. The f H2 dependency on the distribution is considered following the observations of figure 5.

Kinetic model
The model is fit to the experimental data by varying E/N and by enhancing the helium excitation by electron impact, which indirectly promotes nitrogen dissociation.In a first attempt, E/N is set to 6.4 Td, which serves as a first guess, see left column of figure 8 and is denoted as Original.Then, the e+He→e+He * reaction is enhanced to better fit the power scan, see the middle column and is denoted as Enhancement He * .Here, the electronic excitation to the helium metastables is enhanced with a factor of 15, allowing us to set the reduced electric field to 5.6 Td.Lastly, for the third attempt, only the reduced electric field for data points with the Fe-coated surface is increased to 5.75 Td, with respect to the second attempt.This is denoted with Enhancement He * , different E/N.
For the first attempt, in the Original-column of figure 8, E/N is set to 6.4 Td to reach best agreement to the data obtained at the reference condition, i.e. (f H2 , c admix , power) = (95%, 1%, 4 W).The trends in the data are somewhat consistent with the model.Already, the model shows that a hydrogen-rich environment is optimal for ammonia synthesis.This is also observed using atmospheric dielectric barrier discharges plasmas [8,18,60].The optimal f H2 (≈ 63%) is close to 3/4 that corresponds to N 2 :H 2 = 1:3 being the stoichoimetric ratio of NH 3 .Thus, this NH 3 vs f H2 likely originates from the dominance of stepwise hydrogenation of nitrogen atoms to NH 3 , e.g. by the 3-body gas phase reactions.
However, the f H2 scan shows a stronger trend in the experiment than in the model.Such a deviation is not surprising given the sensitivity of the NH 3 production on the electron kinetics.The current model assumes that the electron density and reduced field E/N do not change with the gas mixture, i.e. assuming a 'pure' helium discharge.This seems to be an oversimplification because changing f H2 likely affects the ignition process, as discussed above in section 3.1.A dependence of the electron density and reduced field E/N on f H2 is more appropriate.Nevertheless, for a fixed gas mixture, the scaling of excited nitrogen with power is a regular linear dependence, which hints at a linear relation between the power and electron density.As a consequence, this parameter scan is more easily described and will be mainly considered in the further adjustment of the model parameters.
Furthermore, the electron-induced back-reactions at higher powers are overestimated, and the influence of the introduction of the Fe-coated surface is not properly reproduced.The former is seen in the negative relation between the NH 3 concentration and the power for > 4 W. There, the increasing electron density-with the power-promotes electroninduced back-reactions, e.g.e+NH 3 →e+NH 2 +H.Next, the power scan will be improved by promoting the indirect dissociation of nitrogen.In a second attempt, Enhancement He * , the nitrogen dissociation is indirectly promoted by increasing the rate coefficient for the electronic excitation of helium by a factor of 15.This is a proxy for the various electronic excitation pathways involving excited N 2 species, which are not included in the model.These pathways should indirectly enhance nitrogen dissociation.When directly increasing the nitrogen dissociation, the back-reactions of the NH x -radicals are promoted as well.Simultaneously, E/N is lowered to 5.6 Td to fit the model to the reference condition.Now, the experimentally observed trends of NH 3 with c admix and the power are well described by the model for the blank case, see the middle column of figure 8.
In this fitting attempt, however, the ammonia production is indifferent to the surface coating.A minor increase when introducing the Fe-coating can be seen at higher powers and lower f H2 and c admix .In this model, the increase in the Habstraction rate-by incident hydrogen atoms-increases the back-reaction to H 2 .This compensates the increasing production over the surface, e.g. by higher rates for the LH reactions.Therefore, the change in the surface reaction rate presented in the literature cannot explain the experiments.
For the third attempt, Enhancement He * , different E/N, only the reduced electric field in the model for the Fe-coated surface data points is increased to 5.75 Td in comparison to the second attempt, see right column of figure 8. Now, a decent agreement can be found between the blank and Fe-coated results.It should be mentioned that a good agreement could not be found when arbitrarily altering the sticking coefficients and diffusion energy barriers for ER and LH reactions, respectively (not shown).Also, the slightly lower E/N assumed for the glass surfaces (E/N = 5.6 Td) compared to the metal surfaces (E/N = 5.75 Td) is consistent with the negligible changes in plasma emission when scanning the plasma operation parameters.

Reaction pathways
The inspection of the molar fractions and surface coverages as predicted by the model, Enhancement He * , different E/N, for the blank and the iron-coated surfaces is shown in figure 9.The fractions of the shown gas species increase when introducing the Fe-coated surface, except for the H-fraction.Increasing the electron temperature increases the nitrogen dissociation.Consequently, this stimulates the stepwise hydrogenation by The prediction of the molar fractions and surface coverages from the kinetic model for the blank experiment (solid lines) and Fe-coated surface (dashed lines) using input parameter from literature for Enhanced He * , different E/N of figure 8. the 3-body reactions that form NH, NH 2 , and NH 3 .On the contrary, the larger hydrogen abstraction rate for the metal stimulates the H-loss process, thereby lowering the H fraction.This is a back-reaction and lowers the NH 3 concentration, as previously mentioned when discussing the Enhancement He *results of figure 8.
The surface fractions of figure 9 show that the N-coverage increases while those of H, NH, and NH 2 decrease when introducing the metallic coating.These trends of N s and H s are due to the aforementioned changing E/N and H-abstraction rate, respectively.The decrease in NH s and NH 2,s is attributed to an increased rate for LH reactions on the metal.This increases the throughput to NH 3 whilst lowering the reservoir of NH s and NH 2,s .
Furthermore, the model tells us that most of the ammonia is made in the gas phase, i.e. through stepwise hydrogenation to NH 3 .Therefore, an increased N 2 dissociation rate is responsible for the higher NH 3 fractions.Only a small portion of the ammonia is produced on the surface.This likely explains the insensitivity of the NH 3 molar fraction on the reaction constants of the surface processes, as stated in the previous subsection.For the blank surface, only ER reactions contribute to the NH 3 synthesis at the surface.For instance, at the reference condition, the H 2 +NH s contributes to 8% of the total NH 3 production.When Fe-coating is introduced, the overall production at the surface increases.This is attributed to LH reactions since the diffusion energy barriers on a metal are lower than on a glass surface.For the reference condition, the overall surface contribution increases to 14%, where 4% to ER reactions and 10% is attributed to LH reactions.
The dominance of the gas-phase reactions over the surface reactions is in contrast to the results of Hong et al [14] and 't Veer et al [15].They observe that the surface reactions dominate the NH 3 formation step.The difference between their and our results likely occurs due to a difference in the surface-tovolume ratio.In packed-bed reactors, the typical length scale is of the order of tens of microns in between the beads, whereas in our reactor it is 1 mm in between the electrodes.In this way, the three body sections of radicals with the diluted helium can remain more important than wall collisions.
According to the model, the rate-limiting step consists of electron-induced nitrogen dissociation.This explains the insensitivity of ammonia production on the surface temperature, as presented in figure 7.Although thermal catalysis for ammonia production usually requires temperatures above 250 • C [4], differences have been seen between 20 • C and 150 • C in the literature [8,9].The minor increase in the concentration when using the iron catalyst could originate from the onset of the catalytic formation of ammonia.Nevertheless, it is more likely that this change is induced by an increase in the reduced electric field due to the lower gas density at higher gas temperatures, because the process is highly sensitive on the electron kinetics.Therefore, we conclude that thermally activated surface processes are not the predominant rate-limiting steps in our experiments.

Ammonia synthesis in He diluted RF plasmas
Based on the comparison between data and model, one may conclude that the impact of the catalyst on ammonia production is due to an increase in the electron temperature.This is motivated as follows: it is known that the secondary electron (SE) emission coefficient depends sensitively on the nature of the solid surface and the energy of the incident ions [61,62].For example, if we compare oxides and metals (i.e.thermal catalysts) one could state that oxides exhibit usually higher SE yields for kinetic emission, because the energy of the incident ion is more efficiently transferred to the electrons in a dielectric compared to a metal [63].As a result, more electrons are able to overcome the work function barrier, and they are ejected as SEs despite the larger work function of an oxide compared to a metal.On top of this, the sandblasted (+ nanoparticle catalytic coating) surface should show a lower SE emission coefficient than the blank surface, because emitted electrons are re-trapped by surface irregularities [64].
Therefore, the SE emission coefficient is expected to be higher for the blank experiments than for the metal catalyst surfaces.This reduces the net loss of electrons from the plasma to the surface, thereby improving the confinement of electrons in the plasma.According to the energy balance of standard global models, the improvement of the plasma confinement should decrease the electron temperature [49,65].This is exactly, what is needed in the model to explain the data.
Furthermore, the aforementioned postulated change in SE emission is consistent with the N + 2 (B) and N 2 (C) emission data.When comparing the data for the catalyst versus the dielectric, the emission lines for N + 2 (B) are slightly more intense for the cases for the blank surface.In contrast, the N 2 (C) lines are a bit less intense when using this surface.SEs injected into the plasma-after acceleration in the plasma sheath-contribute to the high-energy tail of the EEDF [66].They should enhance the excitation of species with larger excitation thresholds such as N + 2 (B) (threshold 20 eV).This is consistent with the experiment, where the N + 2 (B) emission is more intense for glass surfaces having a higher SE.At the same time, the lower energy part of the EEDF can be described by a smaller electron temperature leading to a bit smaller N 2 (C) intensity (threshold 10 eV).The iron coating is more effective in enhancing the NH 3 synthesis than copper, especially when lowering f H2 and when considering the 'active' surface area.This is in contrast with other studies [22,23], where copper usually outperforms iron.The enhancement of N 2 dissociative adsorption by electronic and vibrational excitation should occur for copper but not for iron because it depends on the nitrogen binding energy [17,29].Thus, the excitation of nitrogen molecules for dissociative adsorption is not as important as previously thought, which agrees with the recent findings [18][19][20].
The results of this study show that the comparison of different plasma experiments and catalysts in literature must be very carefully performed, to separate the effects of the plasma on the catalyst and of the catalyst on the plasma.This interplay depends heavily on the nature of the plasma type and its kinetics.For the noble gas diluted RF plasma, the impact of the catalyst on the electron kinetics in the plasma dominates.However, our plasma is not as energetic as other plasma sources.In our reactor, the ammonia production seems heavily related to the amount of energetic electrons, thus increasing E/N by increasing f H2 or introducing a metallic catalyst has a noticeable effect.On the contrary, in a more energetic plasma, increasing the reduced electric field is less likely to increase the ammonia production [14].Increasing E/N by increasing the hydrogen content is therefore less beneficial.Hence, some studies find that a nitrogen-rich environment is more beneficial [67,68].

Conclusion
An atmospheric helium RF discharge with admixtures of nitrogen and hydrogen and a catalyst at the surface is used to study the plasma catalysis of ammonia.The experiments are reproduced by a kinetic model.By comparison of measured ammonia production rates with a kinetic model, the contributions of surface effects and volume kinetics are separated.The model shows a large sensitivity to the EEDF.Therefore, a change of the reduced electric field by 0.15 Td can explain the difference in the data comparing the catalysts and blank experiment.It is postulated that a difference in the SE emission coefficient induces this difference.In the future, the characterisation of the interplay between plasma and catalytic surfaces must performed with much higher accuracy to eventually pinpoint the impact of the catalyst on the plasma directly.Also, in-situ surface coverage measurements will help in understanding the surface kinetics, e.g. by validating the current models.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).
Thanks to Christian Oberste-Beulmann and Jonas Hiepel for preparing the catalyst and Laura Chauvet for helpful discussions.The DFG (German Science Foundation) supported this project within the framework of the collaborative research centre SFB 1316 at the Ruhr-University Bochum.There is not conflict of interest to report.

Appendix. NH 3 synthesis model
The chemistry is simulated using a 0D kinetic model.The concentration of gaseous species is described using the molar fraction x M defined as the ratio between the partial density n M and the total density n tot , as x M ≡ n M /n tot .The concentration of the adsorbed species is described as the surface coverage θ M .This is defined as the ratio between the surface site density S M and the total surface site density S tot ; θ M ≡ S M /S tot , with S tot = 1 • 10 15 cm −2 [13].
The volume and surface species are related to each other with a factor ζ = ntotV StotA , with volume V and surface area A. The surface area is approximated as the geometric area of the glass plates in contact with the plasma, therefore the V/A factor reduces to d/2, where d is the distance between the electrodes.For instance, the adsorption rate of atomic hydrogen is obtained by the surface coverage of empty sites θ f with the adsorption rate k ads , which follows from Chantry's formula [13].Then, the reaction rate for H s (R Hs ) is related to the corresponding rate of H (R H ), see relation (A.1), R Hs = −ζR H = +ζk ads θ f x H . (A.1)

A.1. Electron kinetics
The electron kinetics are approximated with a timeindependent electron density and energy distribution function.At atmospheric pressure, the EEDF heavily relies on the gas composition.Even for a helium-dominated gas mixture, a change in the H 2 and N 2 admixture has a noteworthy effect on the distribution, see figure A1.Therefore, the LoKI-B script [51,52] is used to calculate the EEDF with the reduced electric field E/N and f H2 as input parameters.The electron density is set to 8 • 10 10 cm −3 , i.e. n e at 4 W, but has no significant impact on the distribution.The total admixture is not varied for the EEDF-calculations, since the NH 3 vs c admix shows a less intricate trend.The resulting distributions are converted to reaction rates for the electron-impact reactions.The model obtains these values using a look-up table, which interpolates the rates between E/N and f H2 values at which the EEDFs are calculated.

A.2. Volume reactions
Table A1 gives the used gaseous reactions with the corresponding reaction rate.This chemistry set is simplified by excluding reactions with ions and excited species, e.g.electronically excited atoms.The 3-body rate coefficients are used for helium as the third collision partner.For R24, we used the rate coefficients for argon instead of helium, since the rate with helium is not found in literature [69].The reaction N+H 2 → NH+H is omitted as it is only significant when considering the presence of electronically excited molecular hydrogen, which is omitted in this study.Also, reactions R9 and R10 actually create He * 2 and He + 2 , but we assume that these species quickly relax to He.
The electron-impact reaction rate coefficients for R1, R1, R3, R7, and R8 are calculated from a cross-section list using the previously described EEDF.The dissociation of H 2 is described as e+H 2 →e+H 2 (b 3 Σ + u )→e+2 H, since this b-state is a repulsive state [70].The cross-section is obtained from the LXCat/Morgan database [71].Following recent findings [72,73], we updated the calculations for the cracking reactions of NH 3 by electron impact.Before, these rates were a function of the electron temperature [13].Now, we derive it from a cross section set and the calculated EEDF, using the LXCat/Hayashi database [74].
The dissociation of N 2 is considered with two processes: direct electron impact-calculated from the cross sections found in the LxCat/Morgan database-and via an intermediary state He * .For low admixtures of N 2 in a helium plasma, the Penning ionisation of N 2 by He * is a prominent reaction.Despite that most of the resulting N * 2 will react back to N 2 (via N + 4 ) [50], reactions R11 and R12 are included.In this way, the impact of electronically and vibrationally excited nitrogen to the N 2 -dissociation process is simultaneously considered, because He * could act as a proxy for such species.This concept keeps the model as simple as possible but making is versatile enough to explain the experiments.Also, this ambiguity allows for tweaking the involved rate coefficients.

A.3. Surface reactions
The surface kinetics are included as ER and LH reactions of tables A2 and A3, as documented in the literature [6,[12][13][14]23].The sticking coefficients are mostly obtained from [14], who estimated different sticking coefficients and activation energy barriers when comparing iron (metal) and alumina (dielectric).The latter values are used when considering surface reactions at the glass plates, except for S5 and S6 where values are found for SiO 2 [76].Despite that only one of the two glass plates is coated with a catalyst, the values for iron are used for the entire surface, since this oversimplification should help to highlight potential changes as result of the surface in the model.The transport of species to the surface is based on a diffusion model, where the coefficients are calculated considering that helium is the collision partner and using the relation for binary gas systems given by Fuller et al [77].The dissociative adsorption of N 2 and H 2 is considered similar to Hong et al [14].Here, the sticking coefficient γ N2 is calculated from the vibrational level coefficients, which are set to be in equilibrium with the gas temperature.Table A1.Volume processes, where the gas temperature Tg is given in K, the electron temperature Te is given in eV, and the reference temperature T 0 = 300 K. Reaction number in brackets refers to the nomenclature in the cited references.• exp(1700/Tg) (R49) [11]

Figure 1 .
Figure 1.The cross-section of the front view of the reactor (a) and a general overview of the overall setup (b).

Figure 2 .
Figure 2. A typical optical emission spectrum observed at 4 W, f H2 = 0%, and t = 1%.The emission bands with their respective vibrational transitions are indicated, where the v ′ → v ′ ′ notation is used.

Figure 3 .
Figure 3.A typical NH 3 transmittance spectrum with the corresponding best fit, where T cell is 180 • C. The best fit for the molar fraction is 30.9 ppm and for the instrumental full-width-half-maximum is 0.266 cm −1 .

Figure 4 .
Figure 4.The absorbed plasma power as a function of the applied root-mean-squared voltage at different hydrogen fractions in the gas mixture (a) and with different surfaces for f H2 = 95% (b) for c admix = 1%.

Figure 5 .
Figure 5.The concentrations of N 2 (C) (a) and N + 2 (b) normalised to the N 2 admixture in arbitrary units with increasing power for different f H2 at c admix = 1% and for the blank and Fe-coated surface.

Figure 6 .
Figure 6.The molar fraction of NH 3 with increasing hydrogen fraction to the gas mixture for c admix = 1% and 4 W (a), total gas admixture for f H2 = 95% and 4 W (b), and the plasma power for f H2 = 95% and c admix = 1% (c).Four different surfaces are being employed and all data is obtained at 20 • C.

Figure 7 .
Figure 7.The change in the NH 3 fraction when increasing the reactor temperature at different applied gas mixtures and for different surfaces for c admix = 1% and 4 W.

Figure 8 .
Figure 8.A comparison of the kinetic model using surface parameters for an oxide (solid blue lines) or a metal (dashed red lines) and the data of figure 6 for the blank surface (blue symbols) and the iron coated surface (red symbols).The three columns denote: Original-E/N = 6.4 Td; Enhancement He * -E/N = 5.6 Td and the He * production is enhanced by a factor 15; enhanced He * , different E/N-E/N = 5.6 and 5.75 Td for the blank and iron coated surface, respectively, and the He * production is enhanced by a factor 15.

Figure 9 .
Figure 9.The prediction of the molar fractions and surface coverages from the kinetic model for the blank experiment (solid lines) and Fe-coated surface (dashed lines) using input parameter from literature for Enhanced He * , different E/N of figure 8.

Table 1 .
The parameters of the fitting routine.