Operando probing of the surface chemistry during the Haber–Bosch process – Nature

Figure 1a shows how surface-sensitive operando X-ray photoelectron spectroscopy (XPS) is measured together with reaction-product detection during the Haber–Bosch process in the POLARIS instrument¹³. XPS is a powerful technique for investigating the chemical state of catalytic surfaces through core-level shifts that traditionally required vacuum conditions, but operando studies can be conducted using a differential pumping scheme¹⁷. The Fe and Ru single-crystal surfaces are mounted in front of the electron spectrometer with a gap of 30 µm and gases are fed through the front cone of the electron lens, creating a localized virtual catalytic reactor of elevated pressure with a rapid gas flow¹³. The typical operational pressure for ammonia synthesis is 50–200 bar (ref. ¹⁸), at which the gas-phase equilibrium is strongly shifted towards the product, giving a high final conversion to ammonia. However, during the initial phase of the Haber–Bosch process, when not much ammonia has yet been produced, the reaction also proceeds with a high rate at our operational pressures of up to 1 bar (refs. ^19,20).

a, The sample faces a set of apertures that deliver the reaction gas while simultaneously gathering products and emitted electrons. The grazing incidence X-rays enter from the left, producing photoelectrons. The mix of gas and electrons is separated by an electrostatic lens and analysed in an electron analyser and a mass spectrometer. The inset shows XPS spectra of the chemical state of N at 200 mbar over the Fe(110) surface with a 1:3 N₂:H₂ gas ratio. b, Mass spectrometer readout of masses 15 and 16 corresponding to NH₃ production as the gas ratio changes from 150 mbar pure N₂ (blue region showing flow) to 300 mbar 1:1 N₂:H₂ (green region showing flow) over Ru at 673 K. Note that the flows of the gases are shown as the filled blocks plotted on the left axis. c, The enhanced mass spectrometer signals were time averaged during the interval of the 1:1 N₂:H₂ mixture to estimate the relative chemical reactivity. a.u., arbitrary units.

The incoming X-rays were set to an energy of 4,600 eV and the incidence at an angle below total reflection, allowing for high surface sensitivity despite high kinetic energy electron detection. The emitted photoelectrons will pass into the spectrometer through orifices in the front cone and be detected in a hemispherical analyser. The inset in Fig. 1a shows an example of an N1s spectrum of 1:3 N₂:H₂ gases at 1 bar at 673 K, indicating NH₃ (blue), NH₂ (purple), NH (red), surface N (green) and nitride surface (yellow) components. The measurements were conducted at a photon flux at which no detectable X-ray-beam-induced changes could be seen during the Haber–Bosch process (see Methods for further details).

To track the production of NH₃, masses 15 and 16 were monitored in the mass spectrometer (see Methods), as shown in Fig. 1b. The relative chemical reactivities shown in Fig. 1c were determined by measuring the mass spectrometer ammonia signal with respect to the signal of all constituents to compute the number of ammonia molecules formed per second per surface site, which is then further normalized to the highest activity shown by any surface at any temperature (see Methods for further details). The reaction rate increases with increasing temperature and is higher for the stepped Fe(210) than the flat Fe(110) surface, in agreement with previous high-pressure-reactor studies⁹. The highest rate is seen for the \({\rm{Ru}}(10\bar{1}3)\) surface, as expected based on polycrystalline studies showing that Ru has higher activity than Fe (ref. ²¹). The maximum rate for Ru is not at the highest temperature of 723 K, as for the Fe surfaces, but at 623 K, also in accordance with catalytic-reactor studies²².

On exposure to pure N₂ gas at 150 mbar, the two Fe surfaces have a delayed but eventually rapid increase in the N1s intensity, showing bulk nitride formation (Fig. 2a,b). On the basis of the binding-energy position of the N1s peaks in the spectra, this corresponds to the formation of γ′-nitride and ε-nitride plus some small amount of chemisorbed N atoms on the bare Fe surface (see Extended Data Table 1). The nitride formation is more rapid on the Fe(210) surface, specifically the γ′-nitride, whereas on the Fe(110) surface, there is an equal amount of the two nitrides and slower growth. The thicknesses of the nitride layers are greater than ten monolayers; exact quantification depends on the reaction time, as the surface continues to evolve even after hours of observation (see Methods for details on monolayer calculations). We attribute the faster growth on the Fe(210) facet to the higher probability of N₂ dissociation on the stepped surface²³. At temperatures below 523 K, no nitride formation is observed.

The formation and depletion of nitride on the surface of each catalyst are shown as a function of time. At the top, the N₂ gas is introduced with a total pressure of 150 mbar and spectral collection begins. Then, after the nitride begins to stabilize, H₂ gas is introduced immediately in a 1:1 ratio with N₂ with a total pressure of 300 mbar, reducing the surface within the frame of the detector. Next to each time series are example spectra normalized to the background, with a grey arrow showing the frame it represents. a, The data for 673 K over Fe(110). b, The data for 673 K over Fe(210). c, The data for 623 K over \({\rm{Ru}}(10\bar{1}3)\). For Ru, the spectra shown are the summation of the entire time series. Note the difference in y-axis scale in the spectral figures.

The \({\rm{Ru}}(10\bar{1}3)\) reacts completely differently. Almost instantaneously after N₂ exposure, the N1s intensity saturates and remains constant, corresponding to a coverage of 5% of a monolayer, and there is no bulk nitride formation at 623 K (Fig. 2c). The coverage is comparable with previous work, which predicts 17% of a monolayer at 500 K and a pressure of 100 mbar (ref. ²³). The small amount of N₂ on the Ru surface indicates a much weaker N–metal interaction than on Fe, as expected from theoretical predictions¹⁶. The two components are at 397.4 eV and 397.9 eV, and we tentatively assign these to N adsorbed on terraces and steps, respectively (Extended Data Fig. 1). It is interesting that a weak, broad feature is seen at approximately 399–400 eV, with a binding energy consistent with adsorbed N₂ (ref. ²⁴); see Extended Data Fig. 1.

When the pure N₂ gas is replaced by 1:1 N₂:H₂ at 300 mbar, a marked change on the two Fe surfaces occurs within the first spectral sweep (90 s), shown at the bottom of Fig. 2a,b. The nitrides instantaneously disappear and only a small amount of adsorbed N atoms with a coverage of 2% of a monolayer on Fe(110) and 5% on Fe(210) remains. At the same time as the gas mixture is introduced, NH₃ is detected by the mass spectrometer. The rapid removal of the nitrides shows the strong reduction ability of the H₂. The slow growth of nitrides (10–15 min) compared with the fast reduction (<100 ms) shows the difference in rates of N₂ and H₂ dissociation. The adsorbed N atom coverage is also substantially lowered on the \({\rm{Ru}}(10\bar{1}3)\) surface following the introduction of the 1:1 N₂:H₂ mixture at 300 mbar and decreases from 5% to <0.05% of a monolayer as NH₃ is produced.

Next, we address the question of oxides potentially not being reduced on Fe under operando conditions owing to trace contaminations of water or CO₂ in the gas phase⁵. Iron is known to oxidize in trace amounts of water or CO₂ at room temperature, yet iron oxide is not readily reduced below 500 K and, as a result, even under pure hydrogen, iron will oxidize with high flows (see Methods for a detailed description). Figure 3 shows data collected at 500 mbar, 1:3 N₂:H₂ and various temperatures. The Fe 2p_2/3 peaks in Fig. 3a from metallic iron at 706.5 eV and 707.4 eV are split owing to exchange interactions with the ferromagnetic valence electrons, and there is a broad Fe oxide peak at 710.8–709.8 eV, indicated by the grey rectangle. The Fe(110) sample is fully reduced as the temperature reaches 523 K at 500 mbar and the Fe(210) surface requires a higher temperature of 573 K, as seen in Fig. 3b. Fe(210) needs a higher temperature because of the stronger binding of oxygen on a stepped surface. Ru is metallic at all conditions. All surfaces are in a metallic state during the Haber–Bosch process, as expected because of the high concentration of adsorbed hydrogen (Fig. 3c). Note that these measurements were gathered simultaneously with the data in Fig. 4.

Owing to trace contaminations in the gases, the surfaces can form oxides. a, Two cases in which a thick oxide forms at low temperatures and 500 mbar in a 1:3 N₂:H₂ gas mixture, but the oxide thins and disappears as the temperature increases. The grey rectangle shows the region in which iron oxide peaks are present. b, The ratio of oxide to metal as a function of pressure and temperature for the Fe catalysts. The Fe(110) is grey, whereas the Fe(210) is blue. The solid line shows the lower-pressure data at 200 mbar, whereas the dashed line is the higher-pressure data at 500 mbar; at no point was the Ru catalyst oxidized. c, Example spectra of the metal peaks during NH₃ formation at 623 K, showing a singular metallic peak for all catalysts. a.u., arbitrary units.

The steady-state population of the N species on the surface is shown for each catalyst at 200 mbar and 500 mbar at 523 K and 673 K in a 1:3 N₂:H₂ gas mixture. Each set of spectra is normalized and corrected for the cross-section of the corresponding metal substrate. a–c, The data over Fe(110), Fe(210) and \({\rm{Ru}}(10\bar{1}3)\) at 523 K, respectively. d, The data over Fe(210) at 673 K and at 1 bar. e–g, The data over Fe(110), Fe(210) and \({\rm{Ru}}(10\bar{1}3)\) at 673 K, respectively. Note the change in scale owing to the Ru data in c and g; nitrogen coverage of N species on the Ru surface is incredibly low.

The adsorbed nitrogen species can be measured operando as NH₃ is produced. First, focusing on the two Fe single-crystal surfaces (Fig. 4a,b), we observe only adsorbed N atoms on the surface at a binding energy of 397.4 eV, consistent with previous surface-science vacuum experiments once the recoil effect of the emitted atoms is considered (see Extended Data Table 1). Adsorbed molecular N₂ could not be detected and would have been observed at 399.0, 401.2 or 405.9 eV (Extended Data Table 1), depending on the adsorption site and bonding type. The coverage of adsorbed N is 1.3% at 200 mbar and 0.6% at 500 mbar on the Fe(110) surface and increases on the Fe(210) surface to 5.0% and 1.5%, respectively. The higher coverage on the stepped surface is related to availability and stronger bonding of undercoordinated sites¹⁶. What is most surprising is that the coverage is not increasing at higher pressures; on the contrary, the coverage decreases slightly with increased pressure. Inspecting the N1s spectra in Fig. 4d, measured at 1 bar and 673 K, the peak is barely distinguishable from the noise, implying an even lower coverage. It would be tempting to expect an increase in N coverage with increasing pressure because the impinging rate of N₂ molecules increases, but obviously also does the rate of H adsorption. Although we cannot determine the H coverage with XPS, our data suggest that the hydrogenation ability of the surface increases with the total pressure; this would explain a more efficient further reaction of the adsorbed N atoms. Extrapolating to much higher pressures, we predict that the Fe surface is an almost pristine metal under realistic conditions. The fact that no amines (NH or NH₂) or NH₃ are observed at the reaction temperature of 673 K indicates that the rate-limiting step after N₂ dissociation is the hydrogenation of adsorbed N, and the rates of the other hydrogenation steps of NH and NH₂ as well as NH₃ desorption are much faster. At high temperatures, the Ru surface (Fig. 4g) has adsorbed N at 397.4 eV and the adsorbate coverage is almost negligible, with <0.1% of a monolayer of both NH and NH₂ species, independent of pressure within the noise limit. Here the surface is almost entirely clean of any species at conditions of high reaction rate.

At 523 K, for which the reaction proceeds very slowly, the population of the adsorbates changes. There is a slight increase of the adsorbed N on Fe(110) at 500 mbar to 2.3% of a monolayer (Fig. 4d). The Fe(210) surface shows large differences compared with the higher-temperature spectra (Fig. 4e). Further peaks at 398.0 eV, 398.9 eV and 400.2 eV formed, corresponding to NH, NH₂ and NH₃, as determined by previous XPS vacuum studies^9,25,26 and calculated relative peak positions (Extended Data Table 1). Note that the peak at 399 eV is not related to adsorbed N₂ because ex situ XPS studies observed the peak when the Fe catalyst was cooled down to room temperature in the reaction mixture and moved to a vacuum, in which all molecular N₂ would desorb. We observe a relatively high coverage of NH₂ (24.8%), adsorbed N (4.3%), NH (6.7%) and NH₃ (5.2%) at 200 mbar. There is a slight pressure dependence, for which—in particular—the NH₂ decreases to 9.3%. Clearly, there exist conditions in which the adsorbed N and NH_x species are strongly adsorbed on step sites owing to a substantially lower hydrogenation rate. Decreasing the temperature further to 423 K, adsorbed NH_x and NH₃ become visible on the Fe(110) surface. These trends are seen across 423 to 623 K (Extended Data Fig. 2).

On Ru at 523 K at 500 mbar (Fig. 4c), we still see very low coverages, although the coverage of adsorbed N at steps has increased to 0.5%, as well as adsorbed NH₂ to 0.1% and adsorbed NH₃ to 0.1% at around 400 eV. The NH signal increases with pressure, but the nitrogen coverage quantification of these results is nearly within the margin of error. If there is an increase in coverage with pressure for Ru, it may indicate that the H₂–metal interaction for Ru is weaker than for Fe, possibly leading to higher coverages at operational pressures. The adsorbed N species is much more reactive on Ru than Fe, supporting previous theoretical predictions¹⁶.

We can discriminate the various proposed hypotheses and put forward ideas consistent with the data on the chemical state of the catalysts and reaction mechanism in terms of rate-limiting steps. We have shown that nitride formation is far slower than nitride reduction and that the surface states are all metallic with low coverages of atomic nitrogen. There is no evidence for interstitial nitrogen, oxides or high coverage of any species of nitrogen, especially over the most active catalysts. It is interesting to compare the hydrogenation reactions of CO and N₂, which are isoelectronic molecules. In the case of the Fischer–Tropsch reaction on Fe(110), a thick carbide is formed¹⁵, whereas in the Haber–Bosch process, on the same surface, only a pristine metallic phase is generated. Clearly, the difference in the bond breaking of the CO molecule with respect to N₂ and the strength of the adsorbed C and N play an essential role.

The different reaction steps in NH₃ synthesis have been proposed as the following¹⁰:

in which * means surface species and θ* indicate empty sites available for bonding.

The simplest case is the \({\rm{Ru}}(10\bar{1}3)\) surface, for which we can directly explain that steps 3–6 are extremely rapid with no build-up of intermediates, pointing to 1 and 2 as the rate-limiting steps. We observe that the population of adsorbed N₂ is extremely low at high temperatures. The adsorbed molecular state is indeed observed at the low reaction temperature of 523 K, at which its dissociation limits the reaction. We conclude that the rate-limiting step of NH₃ production is the dissociation of the adsorbed N₂, fully in line with theoretical estimations¹². Even at low temperatures, the surface is mostly adsorbate free, with little adsorbed NH_x seen, because of the strong bonding to step sites in comparison with terrace atoms²³. Although we have not observed definitive pressure dependence in the population of adsorbed N, it is plausible that the step sites will become more populated but are expected to remain well below a monolayer.

On Fe it is well established that the rate-limited steps is the molecular dissociation^7,8,9, supported by the correlation between the NH₃ production rate and the N₂ dissociative sticking coefficient for the different single-crystal surface facets^9,27. However, the results here show that, at all temperatures, a factor of around 100 times higher population of adsorbates is observed in comparison with the stepped Ru surface and we can no longer postulate that the reaction proceeds with a high rate after the molecular dissociative steps. Furthermore, there are no signs of molecularly adsorbed N₂ even at the lowest temperatures, indicative of a much higher rate of step 1b. Above 573 K, we observe adsorbed N that is more populated on the stepped crystal, indicating that the hydrogenation step 3 also partly controls the rate¹².

The coverage of N species on the Fe surfaces decreases with increasing total pressure at a constant N₂:H₂ ratio, implying that the N₂ dissociation step is slower than the hydrogenation step¹⁰. Most likely, the coverage of adsorbed H increases with pressure, resulting in faster hydrogenation. Because the coverage of H₂ at the reaction temperatures is expected to be low, we can assume that there is no inhibition of N₂ dissociation caused by the adsorbed hydrogen²⁷.

The population of intermediates shows that, as the reaction temperature lowers, the rate-limiting step switches to become hydrogenation of N, NH and NH₂ as well as NH₃ desorption (steps 3–6), demonstrating differences in the bonding at different high and low coordinated Fe sites. This agrees with earlier observations of the activation energy for hydrogenation being much higher than for N₂ dissociation¹⁰ and the difference in the barriers of these two steps thus becoming prominent at low temperatures: although the N₂ dissociation rate at high temperatures is low owing to a low sticking coefficient that limits N₂ adsorption¹⁰, we see a large population of amines NH_x and NH₃ on Fe at low temperatures. This trend, not seen with Ru, points to the hydrogenation steps affecting the overall rate on Fe. At higher pressures at which more N₂ is converted and the NH₃ content is higher, the back reaction may become important. Indeed, for Ru, it has been theoretically predicted that the coverage of nitrogen species may become substantially higher²⁸.

In closing, we note that, although concerns over the environmental impact of ammonia synthesis have spurned interest in low-pressure alternatives and these might indeed be feasible²⁹, the Haber–Bosch process looks set to remain the primary method of ammonia production for many years to come. A better understanding of the mechanism at play might help to further improve the efficiency and, thereby, lower the environmental impact of this important industrial process. We anticipate that our approach to operando studies will contribute to this endeavour, by making it possible to explore the surface chemistry associated with ammonia formation in the presence of promotors and by making it possible, once measurements at higher pressures and with a higher NH₃ content are feasible, to explore the impact of the ammonia decomposition back reaction.