Catalytic oxidation of ammonia to NO is the core reaction for producing nitric acid, one of the top-ten bulk chemicals by annual consumption (over 100 million tons per year) and widely applied in the manufacturing of fertilizers, nylon, dyes, and explosives, among others1,2,3. Since the early 20th century, the Ostwald process (4NH3 + 5O2 → 4NO + 6H2O), which utilizes Pt-Rh alloy (5-10% Rh) gauzes as the catalysts for ammonia oxidation at high temperatures (800–950 °C) and pressures (up to 12 atm), has been the industrial standard1,4,5,6,7,8,9. The NO yield can reach 90−98% depending on the operation parameters, rendering it one of the most efficient catalytic reactions1. However, Pt loss through volatilization (0.05–0.3 g per ton of nitric acid produced) at high-temperature and strongly oxidizing atmosphere results in significant catalytic performance decays and high operating costs1,6,7. Moreover, nitric acid production is the largest source of the greenhouse gas N2O in the chemical industry (125 million tons CO2-equiv per year)10. The above economic challenges and environmental concern stimulate research towards developing cost-efficient catalysts and new processes with favorable NO yield while reducing N2O emission.

Over the past decades, extensive efforts have been devoted to replacing noble metal catalysts with transition metal oxides2,3,11,12,13,14,15,16,17,18. A large number of metal oxide catalysts have claimed promising performance for ammonia oxidation, including spinels (Co3O4)17,19, perovskites (ABO3 where A = Ca, Sr, La, and B = Mn, Co, Fe)3,12,13,15 and K2NiF4 structured oxides (La2BO4 where B = Co, Cu, Ni)1. The industrial implementation is impeded, however, by the insufficient NO selectivity and rapid catalyst deactivation. For instance, Co3O4 exhibited good activity towards NH3 oxidation with a promising NO selectivity over 90%1,17,19. However, the rapid deactivation due to the reduction to less active CoO, despite in the presence of considerable excess O2, has impaired its commercial deployment. Recently, an alternative approach based on oxygen-conducting membrane reactor was proposed, where the O2 and NH3 were supplied at feed and permeate side of the membrane reactor separately3,15. Pérez Ramírez et al. demonstrated that by exploiting Ca- and Sr-substituted lanthanum ferrite membranes for the high-temperature NH3 oxidation, high NO selectivities (92–98%) can be attained with NH3 conversion ranges between 80 and 95%3. However, the stability of the oxygen ionic transport membranes are often impaired by cation segregation and/or physical degradation. Therefore, it is imperative to research into new intensified processes for NH3 oxidation.

One method towards process intensification is chemical looping, whereby oxidation reactions are decoupled into two-step, spatially separated cyclic redox processes20,21,22,23. Chemical looping is a new frontier for producing valuable chemicals in a clean and efficient manner24,25,26,27,28, and has been demonstrated for processing methane28,29,30,31,32,33,34,35, biofuels36,37, syngas38,39, coal and carbonaceous feedstocks22,40,41 with low cost, reduced emissions and higher energy efficiency. Herein, we proposed and validated the chemical looping ammonia oxidation (CLAO) process. As depicted in Fig. 1, during the reduction step, NH3 is oxidized by the active lattice oxygen derived from the metal oxide redox catalyst, producing concentrated NO. In the subsequent oxidation step, the reduced metal oxide redox catalyst is re-oxidized with air, completing the redox cycle. The CLAO process is potentially advantageous compared to the conventional co-feed approach where NH3 and air are introduced simultaneously. First, since CLAO process decouples the reaction into two, the operation parameters can be optimized independently in order to achieve improved performances. Second, catalyst deactivation could be avoided as the reduced catalyst is regenerated in each oxidation half cycle. Third, the utilization of lattice oxygen instead of nonselective gaseous oxygen may effectively suppress the formation of undesirable N2O, thus reducing the greenhouse gas footprint. That last point is that as large quantities of inert N2 (about 3/4 of the total flow in a current plant)3 are excluded from the product NO, allowing for a radically intensified process for nitric acid production.

Fig. 1: Schematic of the chemical looping ammonia oxidation (CLAO) process.
figure 1

The red and gray spheres represent metal and oxygen atoms, respectively.

The metal oxide redox catalyst functions as the reactive intermediate to transport oxygen ions between the cyclic steps. Hence, the techno-economic viability of the CLAO technology is critically dependent upon the redox properties of the metal oxide catalyst. By virtue of the excellent catalytic activity in various redox reactions, vanadium oxide catalysts have been intensively investigated in chemical and environmental industries, including oxidative dehydrogenation of alkanes, oxidation of methanol to formaldehyde, mass production of sulfuric acid, and selective catalytic reduction of NOx42,43,44,45,46,47,48. In recent studies, V2O5-based redox catalysts have emerged as attractive candidates for oxidative dehydrogenation of ethane via chemical looping, due to the easy accessibility of lattice oxygen and appropriate bonding strength49,50. Generally, three types of lattice oxygen can be distinguished within the unit cell of V2O5 (Fig. S1): (i) singly coordinated terminal oxygen OI, namely vanadyl (V=O); (ii) doubly coordinated oxygen OII bridging two vanadium atoms; and (iii) triply coordinated oxygen OIII bridging three vanadium atoms51,52. The exact role of different lattice oxygen is crucial in determining the product distribution. However, controversy remains on what type of oxygen is the active site for selective oxidation reactions.

In this study, V2O5 was employed as a prototype catalyst to investigate the feasibility of the proposed CLAO process given the aforementioned distinct redox properties. The cyclic NH3 conversion activity and NOx selectivity were evaluated in consecutive isothermal cycles under alternating NH3 and O2 atmospheres. Encouragingly, superior NH3 conversion (97.0%) and NO selectivity (99.8%) were achieved with largely suppressed N2O emission (0.1%) at a low temperature of 650 oC, demonstrating exceptional efficacy of the CLAO technology. In comparison, the maximum NH3 conversion and NO selectivity in the co-feed mode were only 87.1% and 32.4%, respectively, with a fairly high N2O selectivity of 16.8% at 600 °C. On the basis of detailed physicochemical characterization, we endeavored to gain a detailed understanding of the NH3 oxidation mechanism complemented by density functional theory (DFT) calculations. In particular, the catalytic roles of different lattice oxygen of V2O5 were investigated in detail. In situ Raman spectroscopy revealed that terminal oxygen (V=O) being the active center for catalytic NH3 oxidation. Furthermore, XPS and XANES indicated that the reaction involved the reversible reduction of V5+ to V4+, with V=O participated in the oxidation of NH3 via a modified, temporally separated Mars-van Krevelen (MvK) mechanism. The subsequent re-oxidation of vanadium cation to V5+ with molecular oxygen replenished V=O and completed the catalytic cycle. The facile reaction between V=O and NH3 (activation energy: 16.0 kJ/mol) was responsible for the extraordinary NH3 oxidation performance at a relatively low temperature, as substantiated by DFT calculations.


Redox performance and stability

In order to validate the feasibility of the proposed CLAO process, consecutive redox cycles were carried out to investigate the product distribution and the overall chemical looping process efficiency. Figure 2a illustrates the ammonia conversion and product distribution over V2O5 as a function of temperature in a typical chemical looping mode. The corresponding products evolution upon NH3 injection are reproduced in Fig. S2a–h, which indicates that NO, N2O, and N2 were the only N-containing products. While no H2 was observed in the reaction temperature range, water being the only H-containing product that has been condensed out to avoid interference. As depicted in Fig. 2a, ammonia conversion increases from 78.5% to 97.0% at 300–650 °C and plateaus around 97.0% in the temperature range of 580 to 650 °C. Meanwhile, the selectivity to nitrogen-containing products depends critically upon temperature: N2 is the dominant product at low temperature (<500 °C), whereas NO takes over at high temperature (580–650 °C). The selectivity to NO starts at 400 °C and increases continuously with increasing temperature, reaching a maximum of 99.8% at 650 °C. Accordingly, the increase of NO selectivity is coupled with the progressive decrease in N2 and N2O selectivity. It is worth noting that N2O formation is greatly inhibited (0.1% selectivity at 650 °C). In comparison, V2O5 redox catalyst exhibits notably different results in a conventional co-feed mode. As shown in Fig. 2b, ammonia conversion increases from 26.1% to 87.8% at increasing temperature and reaches a plateau from 500 to 650 °C. The maximum ammonia conversion (87.8%) is considerably lower than that in the chemical looping mode (97.0%). In terms of product selectivity, NO peaks at 300 °C with an inferior selectivity of 34.2% and decreases significantly when rising the temperature to 400 °C. A further increase in temperature resulted in increased formation of NO. After passing through a maximum at 600 °C, NO formation slightly decreases with a further increase in temperature. The decreased NO formation with a concurrent increase in N2 formation at higher temperatures (600–650 °C) can be related to the selective catalytic reduction (SCR) of NO with NH3. A similar parallel SCR mechanism was also proposed in a recent study of Decarolis et al.53, which indicates the oxidation of NH3 to NOx by PdO and the subsequent catalytic reduction of NOx by NH3 to produce N2. In addition, a high N2O selectivity (13.6–26.9%) is observed over the reaction temperature range. Fig. S3a, b illustrates the NH3 oxidation performance with different O2 partial pressure. Furthermore, we also investigated the ammonia oxidation performance for typical single and binary redox metal oxide catalysts in chemical looping mode. The optimized NH3 conversion and selectivity towards NO, N2O, and N2 are compared in Table S1. The NH3 conversions are close to 100% irrespective of the catalyst used. Nonetheless, the NO selectivity is relatively lower than V2O5 in chemical looping mode and industrially relevant Pt-Rh in the conventional co-feed mode. Hence, superior NH3 conversion can be achieved over V2O5 via chemical looping with close to complete NO selectivity at 650 °C, which is sufficiently lower than the industrial ammonia oxidation reaction temperature of 810–950 °C over Pt-Rh gauzes. Moreover, the NH3 conversion and NO selectivity for V2O5 also compared very favorably with those reported in the membrane mode that carried out at significantly higher temperatures (Table S1), evidencing the great efficiency of the proposed CLAO process.

Fig. 2: Reactivity and selectivity comparison over V2O5.
figure 2

NH3 conversion and NO, N2O, N2 selectivity versus temperature in NH3 oxidation over V2O5 in a chemical looping NH3 oxidation process and b steady-state NH3 oxidation (co-feed mode). c Arrhenius plots of surface lattice oxygen removal of V2O5 in chemical looping NH3 oxidation process at 300–650 °C. d NH3 conversion, NO, N2O, N2 selectivity as well as total lattice oxygen consumption versus the number of pulses with NH3 over V2O5 at 650 °C. e NH3 conversion, NO, N2O, N2 selectivity versus the number of pulses over V2O5 at 650 °C during five consecutive chemical looping NH3 oxidation redox cycles.

Kinetically, chemical looping reactions are generally slower than conventional catalytic oxidation/reduction reactions, which are commonly encountered in chemical looping process. This is because the rates of redox reactions are limited by the removal and replenishment of lattice oxygen from/to the bulk of the redox catalyst. The CLAO process proposed here resulted in a normalized NO production rate of 56.4 ml gcatalyst−1 h−1 (Table S1). The NO production rate of conventional co-feeding mode reported here ranged between 105.2-127.6 ml gcatalyst−1 h−1 (Table S1). While the NO production rates are on the same order of magnitude, the superior NH3 conversion and NO selectivity of CLAO approach can potentially eliminate the costly separation units, ensure safer operation and inhibit nonselective oxidation reactions, which contribute to the economic attractiveness of the NH3 oxidation process. Furthermore, to ensure a relatively faster surface oxygen removal rate, two potential strategies have been proposed for the chemical looping process23. One is through doping or lattice site substitution, in an attempt to introduce crystallographic defects that facilitate oxygen ion diffusion. The second strategy is to promote the redox metal oxide with catalytically active species on the surface to decrease the activation energy for reactant activation and to enhance the lattice oxygen consumption from the bulk. Both strategies may open up extensive possibilities for the CLAO process.

To elucidate the dynamics of surface lattice oxygen removal during the reduction step, the initial maximum lattice oxygen extraction rate for NH3 oxidation during the first pulse was determined from Fig. S2a–h, considering the following stoichiometry reaction pathways as no H2 is produced (all the H ends up into H2O, which was condensed out before entering the quadrupole spectrometer):

$${{{{{{\rm{NH}}}}}}}_{3}+2.5{{{{{{\rm{O}}}}}}}_{{{{{{\rm{lattice}}}}}}}\to {{{{{\rm{NO}}}}}}+1.5{{{{{{\rm{H}}}}}}}_{2}{{{{{\rm{O}}}}}}$$
$$2{{{{{{\rm{NH}}}}}}}_{3}+3{{{{{{\rm{O}}}}}}}_{{{{{{\rm{lattice}}}}}}}\to {{{{{{\rm{N}}}}}}}_{2}+3{{{{{{\rm{H}}}}}}}_{2}{{{{{\rm{O}}}}}}$$
$$2{{{{{{\rm{NH}}}}}}}_{3}+4{{{{{{\rm{O}}}}}}}_{{{{{{\rm{lattice}}}}}}}\to {{{{{{\rm{N}}}}}}}_{2}{{{{{\rm{O}}}}}}+3{{{{{{\rm{H}}}}}}}_{2}{{{{{\rm{O}}}}}}$$

And the results are summarized in Table S2. By fitting the maximum oxygen donation rate of V2O5 redox catalyst with the Arrhenius equation, the apparent activation energy for the surface lattice oxygen removal is estimated to be as low as 16.0 kJ mol−1 (Fig. 2c), which is in line with its superior low-temperature NH3 oxidation activity. To gain a quantitative picture of the effect of lattice oxygen extraction on the NH3 conversion and product distribution over V2O5 in the CLAO process, the NH3 conversion, NO, N2, N2O selectivity as well as the accumulated lattice oxygen withdrawn from the original sample (δ) is plotted as a function of the NH3 pulse numbers (Fig. 2d). Starting with a fully oxidized catalyst surface, NH3 conversion and NO formation are initially very high, with NH3 conversion amounts to 97% and NO selectivity approaching 100% in the first 10 pulses. The initial high NO selectivity can be associated with the presence of highly reactive and selective lattice oxygen species. Subsequently, the NO selectivity decreases with ongoing pulse number. The NH3 conversion, in contrast, is practically constant with increasing pulse numbers. The total lattice oxygen consumption increases linearly with increasing number of pulses. It should be noted that NH3 conversion above 97%, NO selectivity above 90% can be achieved, with insignificant N2O selectivity, even after 30 pulses of NH3. And this corresponds to the catalyst where ~2.4% of the total oxygen is removed, equivalent to overall composition of V2O5-0.12. In addition, we further explored the NH3 conversion and product distribution over V2O5 with 180 pulses of NH3. As shown in Fig. S4, the NH3 conversion remains close to 100%. Meanwhile, the N2O formation is largely inhibited. On the other hand, the NO selectivity decreases accompanied by the increase of N2 formation with increasing pulse number. The decreased NO selectivity can be tentatively attributed to the decreasing amount of active oxygen species available from the catalyst during the sequence. Figure 2e exemplifies NH3 conversion and product distribution over 5 cycles for <1300 min of continuous operation. As can be seen, NH3 conversion remains close to constant over the redox cycles. Despite the decrease in NO selectivity with increasing pulse number, a reoxidation of V2O5 catalyst fully restores the initial NO selectivity when pulsing NH3, indicating excellent recyclability through redox cycling. Based on the foregoing multipulse NH3 experiments, it can be concluded that the ammonia oxidation to nitric oxide over V2O5 in the proposed CLAO process proceeds via a modified, temporally separated MvK scheme. NH3 is first activated by the lattice oxygen of V2O5 redox catalyst in the absence of gas-phase oxygen, yielding NOx and a reduced VOx center. The reduction degree of V2O5 dictates the product distribution but does not alter the catalytic NH3 oxidation activity. The reduced vanadium oxide surface sites are subsequently replenished by the spatially separated reoxidation step, closing the redox cycle.

Redox catalyst characterizations

To rationally develop a process for ammonia oxidation to nitric oxide catalyzed by redox metal oxide, in-depth understanding of the redox chemistry and reaction pathways is of prime importance. In this section, thorough physicochemical characterization combined with density functional theory calculations were employed to reveal the catalyst structure–activity relationship. Powder X-ray diffraction (XRD) was conducted to investigate the possible crystalline phase transition and structural stability of as-prepared, reduced, and reoxidized catalyst. As illustrated in Fig. 3a, the XRD pattern of the as calcined sample corresponds well to single-phase orthorhombic V2O5 structure (JCPDS 01-089-0612). Upon NH3 pulse reduction, new reflections associated with V6O13 (17.76° and 30.15°, JCPDS 00-019-1399) and VO2 (26.92° and 27.88°, JCPDS 03- 065-2358), respectively, are identified apart from V2O5. This indicates that V2O5 was partially reduced due to the stoichiometric reaction of the active lattice oxygen species with NH3. In addition, peaks attributed to V6O13 and VO2 progressively increased in intensity with increasing NH3 pulse number as more lattice oxygen was extracted. After CLAO cycling, V6O13 and VO2 peaks vanished, implying that the reduced phase was completely reoxidized. Therefore, it was confirmed that V2O5 redox catalyst possesses excellent cycling stability over the redox cycles. For comparison, the XRD pattern of V2O5 in conventional co-feed mode is also shown, which indicates negligible structural change after the activity measurement (see Fig. S5 for more details).

Fig. 3: Structural analyses of V2O5 redox catalyst.
figure 3

a X-ray powder diffraction patterns of V2O5 as a function of NH3 pulses numbers. The following compounds were identified: V6O13 (black heart), VO2 (black diamond), and V2O5 (vertical cyan line). b V 2p XPS spectra of V2O5 sample: (i) as calcined, (ii) after 30 pulses of NH3 reduction, (iii) after 120 pulses of NH3 reduction, and (iv) after CLAO redox cycling at 650 °C. c Normalized V K-edge XANES spectra of V2O5 catalysts with different NH3 pulse numbers together with the standard compounds in different oxidation states. The insert shows the enlarged view of the pre-edge peak. d Plot of position of the absorption edge energies where the normalized absorption equals 0.5 (E1/2) vs the vanadium average oxidation state.

In order to track the surface chemical state evolution of vanadium during the CLAO process, X-ray photoelectron spectroscopy (XPS) measurements were employed. The high-resolution XPS spectra in the V 2p region of V2O5 redox catalyst after different treatments are presented in Figs. 3b, S6a–d, and S7. For the initial catalyst (i), a strong peak centered at 517.4 eV (red) along with a broad satellite peak at a higher binding energy (blue) was observed, which agrees well with the signature of V5+ in V2O554,55. After 30 pulses of NH3 reduction (ii), the V5+ peak decreased and an additional peak emerged at a lower binding energy of 516.3 eV. Deconvolution of the V 2p3/2 region suggests that V was predominantly present in two different oxidation states, namely, V5+ (79.7%) and V4+ (20.3%) (see Table S3 for further details). For V2O5 sample subjected to 180 pulses of NH3 reduction (iii), the V4+ component further increased (V4+, 38.4%) due to the enhanced reduction extent (Table S3). Note that no V3+ or V2+ was identified, evidencing that the vanadium existed only in the form of V5+ and V4+ after reduction with NH3. For V2O5 sample after the CLAO redox cycling (iv), the intensity of V5+ peak at 517.4 eV was recovered. Meanwhile, the contribution of V4+ completely disappeared, pointing toward a dynamic recovery of vanadium following reoxidation. For comparison, no apparent reduction was observed for V2O5 after NH3 oxidation reaction in the co-feed mode (Fig. S7). The above results highlight that vanadium could undergo reversible reduction and oxidation featuring a V5+/V4+ redox cycle in the CLAO process.

To obtain a deeper insight into the mechanistic details, XANES was used to determine the evolution of the vanadium oxidation state and coordination environment during the reaction. The position of the primary absorption and the intensity of the pre-edge feature serve as effective probes of the local electronic and geometric structure of vanadium56,57,58. Figure 3c depicts the normalized K-edge XANES spectra of the V2O5 samples as a function of NH3 pulses number alongside the standard references with nominal valences of +3 (VCl3), +4(C10H14O5V), and +5 (V2O5), respectively. With increasing NH3 pulses number, a monotonic shift of the absorption edge to successively lower energies and an associated decrease in pre-edge peak intensity were observed (see Fig. S8 for further details). The observed shift towards lower energies is consistent with increased reduction of vanadium sites. Notably, the absorption edges for all V2O5 samples after NH3 pulse reduction fell between V2O5 and C10H14O5V, revealing an oxidation state intermediate between +4 and +5, which is well in accordance with the XPS results mentioned above. In addition, linear correlation between the oxidation state of reference compounds and the edge position of E1/2 (where the normalized absorbance equals 0.5) was employed to evaluate the oxidation degree of vanadium after NH3 pulses reduction. The average vanadium oxidation state decreased to +4.6 after 180 pulses of NH3 reduction (Fig. 3d). The reduction degree of V2O5 determined by XRD, XPS, and XANES was compared in Table S3. The mean oxidation state of vanadium and the percentage of V4+ and V5+ determined by XRD and XANES (obtained by comparison of the main edge energies and XANES linear combination fitting in Fig. S9) are in qualitative agreement with each other. The mean vanadium oxidation state is slightly higher than the average value of +4.4 of a global composition of V2O5-0.6 presented in Fig. S4. The small difference is most likely induced by post-reaction air exposure due to the ex-situ XRD and XNAES measurement. Meanwhile, XPS results give a higher V4+ portion (a higher reduction extent) at different NH3 pulse numbers, which is due to the surface sensitivity of the XPS measurement. Overall, the XRD, XPS, and XANES techniques indicate a monotonic increase in vanadium reduction extent with increasing NH3 pulses number. EXAFS analysis of the data collected on the reduced samples are in line with the increased reduction of V2O5 (see Figs. S10 and 11 and Table S4 for more details). On the other hand, the decrease of the pre-edge peak intensity (inset in Fig. 3c) reflects an enhanced local symmetry around vanadium atoms59,60. Combined with the XRD results, the increase in local symmetry originates from the progressive evolution from strongly distorted square-pyramidal VO5 groups in V2O5 to less distorted octahedral VO6 groups in V6O13 and VO2 (Fig. S12). Moreover, it has been well documented that the pronounced pre-edge peak is assignable to the hybridization of V 3d with O 2p of the nearest oxygen, which is roughly proportional to the number of vanadyl (V=O) group present56,57,59. The characteristic pre-peak of the V K-edge XANES spectrum of V2O5 is concurrently related to the breaking of the central symmetry and the presence of the short V=O bond (1.585 Å). Density functional theory calculations also assign the pre-edge feature to the V 3d−O 2p hybridization of the vanadyl bond (V=O) and the local geometry of the distorted [VO5] square pyramid61. Thus, the decreased pre-edge intensity evidences that with increasing NH3 pulses number vanadyl bond diminished after reacting with NH3. Upon CLAO cycling, the absorption edge shifts back to higher energies while the pre-edge intensity increases, showing that the reduction and re-oxidation processes are entirely reversible.

In situ Raman spectroscopy was further employed to determine the nature of the active centers on V2O5 redox catalyst, as it constitutes a highly sensitive tool for monitoring the rearrangements of V–O at an atomic level62,63,64. Fig. S13a presents the Raman spectrum of the fresh V2O5 sample. A set of intense Raman bands at 100, 144, 198, 284, 303, 406, 481, 525, 701, and 994 cm−1 were identified, which correspond well to crystalline V2O5 nanoparticles62,63. The highest frequency mode at 994 cm−1 has been attributed to the shortest vanadyl (V=O) stretch within VO5 square pyramids. Whereas, the low-frequency modes at 100, 144 and 198 cm−1 have been assigned to external modes corresponding to the displacements of [VO5–VO5] units, which serve as a good baseline for reflecting the long-range order within V2O5. The next highest modes at 701 and 525 cm−1 are ascribed to the stretching vibration of V–OII–V bands, while the relatively weak mode at 481 cm−1 is associated with the bending of V–OIII–V bridging units. Three additional modes at 284, 303, and 406 cm−1 are assigned to bond rocking oscillations involving the terminal oxygen atoms. The in situ Raman spectra for V2O5 during NH3 reduction and the following O2 oxidation step are depicted Figs. 4 and S13b, c. As shown in Fig. 4a, both V = O and [VO5–VO5] bands lost intensity with increasing reduction time, implying the removal of terminal oxygen and the continual loss of long-range order within V2O5. After re-oxidation, the V = O and [VO5–VO5] bands were reinstated (Fig. 4b), indicating the reversible structural changes upon re-oxidation. Contrarily, no notable change in intensity was observed for V–OII–V and V–OIII–V peaks during the reduction and oxidation steps (Fig. 4a, b). By tracking the evolution of V=O vibration (994 cm−1) as a function of time, it was noticed that the V=O Raman band gradually diminished in intensity during the reduction step (Figs. 4c and S13b), which strongly indicate the destroying of V = O sites after reacting with NH3. Interestingly, during the following re-oxidation step (Figs. 4b and S13c), V=O Raman band increased slowly within the first 10 min. Subsequently, the V=O signal rapidly approached to a maximum within the next 5 min, indicating that the V=O sites were rapidly regenerated after re-oxidizing with O2. Hence, the above XANES and Raman spectroscopy results indicate that the reaction proceeds via a unique, temporally separated MvK mechanism associated with the breaking and regeneration of V=O bonds. While a classical MvK mechanism would involve concurrent removal and restoration of surface lattice oxygen species in separate steps, and often at spatially separated active centers65,66. Under the CLAO reaction scheme, the active V=O sites selectively activates and reacts with ammonia, while being removed from the surface. In a subsequent, temporally separated oxidation step, the active V=O sites were restored as confirmed by in-situ Raman spectra.

Fig. 4: In-situ Raman spectroscopy.
figure 4

In situ Raman spectra of V2O5 redox catalyst during the a reduction and b subsequent re-oxidation as a function of time. Reaction conditions: reduction, 5% NH3/Ar (2.2 ml/min); oxidation, 30% O2/Ar (2.2 ml/min); temperature, 650 oC. The corresponding V=O relatively intensity evolution during the c reduction and d re-oxidation step. V=O relatively intensity equals V=O reduced (oxidized)/V=O at full oxidation state.

The crystallinity and surface morphology of V2O5 redox catalyst were investigated by HRTEM and SEM techniques. The high-resolution TEM image and the corresponding Fast Fourier transform (FFT) pattern are demonstrated in Fig. 5a. The as-calcined sample exhibited well-resolved lattice fringes of 3.4 and 5.7 Å (Fig. 5a), which coincide with the (110) and (200) facets of rhombohedral V2O5, respectively. Figure 5b–d shows representative SEM images of the as-calcined V2O5 sample, V2O5 after 30 NH3 pluses reduction as well as V2O5 after the CLAO redox cycling. The as-calcined sample are platelet-like in shape with thicknesses ranging from 0.8–1.5 μm, lateral and longitudinal dimensions of 2–6 μm (Fig. 5b). In the case of V2O5 after 30 NH3 pluses reduction (Fig. 5c) and V2O5 after the CLAO redox cycling (Fig. 5d), severe sintering and substantial morphological changes were not observed, confirming the robustness and stability of V2O5 redox catalyst under the demanding reducing and oxidizing cycling.

Fig. 5: Morphological characteristics of V2O5.
figure 5

a HRTEM image of the as-calcined V2O5 sample and the corresponding Fast Fourier transforms (FFTs) image (inset). Representative SEM images of V2O5 sample: b as-calcined, c after 30 pulses of NH3 reduction at 650 °C and d after CLAO redox cycling at 650 °C.

DFT studies

To gain a more fundamental understanding of the CLAO reaction mechanism, DFT calculations were carried out to elucidate the reaction landscape associated with the dissociation of NH3 and the formation of NO, N2, N2O, and H2O. In particular, the catalytic roles of different oxygen sites are investigated in detail. The reaction is initiated by the adsorption of NH3 on V2O5 (001) surface (Fig. S14), where the single NH3 adsorption and two NH3 co-adsorption were both studied. As shown in Fig. S15, the reaction pathway for single NH3 adsorption is less favorable due to the very high reaction barrier (1.42 eV) of the rate-determining step (refer to Supplementary information for more details). Meanwhile, NO formation starting from the two NH3 co-adsorbed configurations (Eads = 3.95 eV) is also investigated, in which the surface of V2O5 (001) retains the original structure without serious reconstructions. As illustrated in Fig. 6a, the first hydrogen migrates to adjacent OII, forming an OIIH and NH2*, which is the rate-determining step endothermic by 0.21 eV with a medium barrier of 1.06 eV. The first hydrogen tends to continue migrating to the neighboring OI bonding with neighboring V, which is more energetically favorable with a much lower barrier of 0.47 eV in comparison with the direct dehydrogenation of the second hydrogen (1.80 eV, Fig. S16). This step liberates the OII site in the vicinity of NH2*, promoting the second hydrogen transfer with a relatively small barrier of 0.79 eV. Subsequently, the NHO* intermediate is formed via the combination of NH* and OI (activated by 0.15 eV), followed by H2OI formation via the second hydrogen transfer from OII site to the OIH site with an extreme small barrier of 0.09 eV. This step again liberates the OII site to accommodate the last hydrogen, resulting in NO* formation with a negligible barrier of 0.03 eV. The resulting NO* can be released from the surface with 0.16 eV energy increase. The above DFT calculations clearly indicate that the oxidation of NH3 to NO and H2O is favorably occurring on vanadyl oxygen OI site (V = O), while the successive stripping of hydrogen is catalyzed by both OI and OII sites. This result is in line with the in situ Raman spectroscopy, in which only V = O consumption was observed when reacting with NH3.

Fig. 6: DFT calculations of CLAO paths on V2O5.
figure 6

Computed energy potential profiles of each elementary step for the formation of a NO, b N2 and c N2O on V2O5 (001) surface. The kinetic barrier of each elementary step is given in eV.

For N2 formation, two possible reaction pathways were considered according to whether atomic N* intermediates are produced. The energetically more favorable pathway is depicted in Fig. 6b. Initially, two NH3 dehydrogenate into two NH2* species with the activation barriers of 1.06 (rate-determining step) and 0.89 eV, respectively. Instead of continuing dehydrogenation, the two NH2* tend to combine into the NH2NH2* intermediate due to the relatively low reaction barrier (0.90 eV). Upon NH2NH2* formation, the final 4 steps of hydrogen abstraction proceed, each with a relatively low activation barrier, resulting in N2 formation. For each dehydrogenation step, all possible hydrogen accepter sites were examined. Our calculations reveal that both OI and OII could be hydrogen accepters during the dehydrogenation steps due to the close energy barriers. For comparison, the atomic N* was also considered as the intermediates. As shown in Fig. S17, the entire reaction process is exothermic (−5.10 eV), however, all dehydrogenation and N–N combination steps are highly endothermic with the highest activation barrier being 1.49 eV, making this pathway less feasible. It should be noted that since both NO and N2 formation share the same rate-determining step (TS1 in Fig. 6a, b), their relative rates are largely determined by the second highest barriers as highlighted in orange, i.e. the second hydrogen transfer (0.79 eV) (NH2* → NH*) for NO formation and the two NH2* combination (2NH2* → NH2NH2*) (0.90 eV) for N2 formation. To take the temperature effects into account, the Gibbs free energy corrections over temperature for these two key steps were calculated as shown in Fig. S18. Results show that the free energy barriers of NH2* → NH* are always much lower than those of 2NH2* → NH2NH2*, which is coherent with experimental observations whereby NO prevails at high temperature. It should be noted that at this stage, the existing DFT calculations are not sufficient to justify the product selectivity considering the complexity of the reaction networks. This problem may be addressed in future studies by using more advanced kinetic analysis methods like microkinetic modeling. Additionally, the energy potential profiles as well as the possible intermediates for N2O formation is illustrated in Fig. 6c. Starting from the NO* + NH3* configuration, the first N–H bond cleavage is also the rate-determining step with a relatively large activation barrier of 1.31 eV higher than that of NO and N2 formation (1.06 eV). Thus, N2O is largely suppressed during the reaction, which is in well accordance with the experiments. Notably, the oxygen atoms in the resulting oxidation products also arise from OI. Based on combined experimental and theoretical studies, it is therefore concluded that OI sites are the major oxidants leading to the formation of oxidation products, meanwhile both OI and OII sites participate in the hydrogen transfer processes. The inhibition of N2O formation makes V2O5 a very selective catalyst for CLAO processes.

Reaction mechanism

Combining catalytic measurements, spectroscopic studies as well as theoretical calculations presented above allows for a molecular-level understanding of the potential reaction pathways for the proposed CLAO process. As illustrated in Fig. 7, this process follows a unique, temporally separated MvK redox mechanism whereby lattice oxygen species participated in the reaction. In the first reduction step, NH3 was activated by the lattice oxygen originated from V2O5 catalyst, leading to NO formation and a reduced VOx center (VO2 and V6O13). In the subsequent re-oxidation step, the reduced vanadium oxide site was replenished by gas-phase O2, closing the catalytic cycle. XPS and XANES studies demonstrated that the CLAO process involved the reversible reduction of vanadium cations with a V5+/V4+ redox couple easily accessible. Combined operando Raman and XNAES spectroscopy provided concrete evidence that among the three lattice oxygen sites on V2O5: the vanadyl (V = O), the bridge oxygen OII, and the triply coordinated oxygen OIII, the V = O site was the catalytic active site for the selective oxidation of NH3 to NO. The initial high NO formation can be attributed to the presence of highly reactive and selective V = O species, which decreased with increasing NH3 pulse numbers due to their removal for NH3 oxidation to NO. The V = O species were regenerated following oxidation, restoring the high NO selectivity. This temporally separated MvK scheme avoids direct competition of reduction and reoxidation on the same V = O active sites that would otherwise be inevitable in a co-feed scheme, leading to remarkable enhancement in catalyst performance. Furthermore, in the proposed CLAO process, the activation of oxygen is spatially and temporally separated from the catalytic NH3 oxidation to effectively suppress the formation of nonselective surface molecular oxygen species and thus the undesired nitrogen byproduct15,16. Additionally, the chemical looping strategy allows for the precise control of the residence time of reactants, the reduction extent of the redox metal oxide and ratios of the reaction intermediates (O/NHx) on the catalyst surface. Therefore, a high NH3 conversion can be achieved, with can potentially suppress the N2 formation via the catalytic reduction of NO with NH3. Consequently, a high selectivity for NO can be anticipated. Kinetic studies indicated that the apparent activation energy for the lattice oxygen removal is as low as 16.0 kJ mol−1, which is consistent with the superior low-temperature selective NH3 oxidation performance. In agreement with the experimental observations, DFT calculations indicate that only OI and OII are terminated and can participate in the reaction. The CLAO process is initiated by the co-adsorption of 2NH3 with Eads = 3.95 eV, followed by the first dehydrogenation step, activated by 1.06 eV and endothermic by 0.21 eV, which is catalyzed by adjacent two-coordinate oxygen site (OII) and represents the rate-determining step. The hydrogen continues to transfer from the OII to the neighboring OI to form OIH* (activated by 0.47 eV). The second dehydrogenation step occurs afterwards with a relatively small barrier of 0.79 eV, which is also catalyzed by the nearby OII site, forming NH* and OIIH. The resulting NH* can easily bonds with adjacent OI site to form the NHOI* intermediate due to the very small barrier of 0.15 eV. The two hydroxyl groups (OIIH and OIH) can also easily interact to form H2OI molecule due to the extremely low barrier of 0.09 eV. The liberated OII site in the vicinity of NHOI* induces the final dehydrogenation step, leading to the facile NO* formation with a negligible barrier of 0.03 eV. Subsequently, NO* can be easily released from the catalyst surface with a very low activation barrier (0.16 eV). The remaining NH3* on the surface can be co-adsorbed with another NH3 molecule to generate NO mediated by OI and OII species. Upon reoxidation, the oxygen-deprived V2O5 is regenerated, completing the CLAO redox cycle.

Fig. 7: Proposed reaction mechanism.
figure 7

Schematic of the proposed reaction mechanism for CLAO process over V2O5 redox catalyst. The elementary reaction barriers are given in eV.


In this contribution, catalytic ammonia oxidation to nitric oxide via a chemical looping approach was proposed and demonstrated in absence of noble metal catalyst. Among many metal oxide redox catalysts investigated, V2O5 exhibited superior NH3 conversion at a substantially lower temperature with an outstanding NO selectivity (99.8%), outperforming the expensive, commercially applied Pt−Rh gauzes at a temperature up to 300 °C lower. In addition, the selectivity to undesirable N2O was largely suppressed, which was distinctly different from NH3 oxidation in the conventional co-feed mode. Repeatable NH3 conversion and product selectivity were achieved during consecutive redox cycles over 1300 min. The superior performance was rationalized down to the atomic level by both operando and ex-situ characterization techniques complemented by DFT calculations. Results show that the proposed CLAO process operates via a modified, temporally separated MvK redox mechanism featuring a reversible V5+/V4+ redox cycle. The vanadyl (V=O) centers act as active sites for NH3 oxidation, leading to the formation of the oxidation products. Meanwhile both V=O and doubly coordinated oxygen (OII) participate in the hydrogen transfer process. The superior activity and selectivity can be attributed to the low activation energies for the successive hydrogen abstraction and facile NO formation facilitated by V=O. The fundamental understanding and the proof-of-concept for CLAO can open exciting perspectives for nitric acid manufacturing in a cleaner, more sustainable and cost-efficient manner.


Catalyst synthesis

V2O5 was purchased from Energy chemical (+99.5%). The as-received powder was calcined under stagnant air at 650 °C for 4 h with a ramp rate of 2 °C min−1. Prior to use, the calcined sample was ground into a fine powder using a mortar and pestle. Additionally, commercially available CeO2 (99.5%, Aladdin Industrial Corp.), SnO2 (99.5%, Aladdin Industrial Corp.), Fe3O4 (99%, Aladdin Industrial Corp.), NiFe2O4 (99.5%, Aladdin Industrial Corp.), CoFe2O4 (99.5%, Aladdin Industrial Corp.), and ZnFe2O4 (99.5%, Aladdin Industrial Corp.) were also investigated for comparison.

Catalytic activity measurements

The ammonia oxidation process was conducted at atmospheric pressure in a fixed-bed quartz U-tube reactor (i.d. = 9.75 mm), which was enclosed by an electric furnace (AutoChem II 2920 Micromeritics). Both chemical looping and co-feed modes were carried out for reactivity test in the temperature range of 300–650 °C. In both modes, 200 mg of the catalyst was packed between two layers of quartz wool within the isothermal zone of the reactor. The temperature was controlled using a K-type thermocouple in direct contact with the catalyst bed. The inlet gas flow rates were regulated by calibrated mass flow controllers.

  1. (i)

    In typical chemical looping mode, pulses of 5% NH3 (765 μL, balance Ar) were injected into the reactor using Ar (30 mL min−1) as the carrier gas. Subsequently, a regeneration step was conducted using 30% O2 (30 mL min−1, balance Ar). To study the evolution of catalytic performances, 30 consecutive ammonia pulses were injected to reduce the redox catalysts, followed by an oxidation step with simulated air (30% O2/Ar) at a total flow rate of 30 mL min−1 for 30 min. In between the reduction and oxidation step, a purge step with pure Ar (30 mL min−1) was introduced for 15 min to assure well-defined conditions.

  2. (ii)

    In typical co-feed mode, steady-state ammonia oxidation was carried out using feeds containing 5% NH3/Ar (30 mL min−1) and O2 (12.86 mL min−1) with the effective O2 percentage of 30%.

Outlet gas concentrations were monitored online after condensation of moisture by a quadrupole spectrometer (Pfeiffer Omnistar QMS 200), using Ar as internal standard. And the corresponding molar flow rates (ni) were calculated. The spectra signals were calibrated before each experiment with calibration gas of known compositions (5.0% NH3/Ar, 6006 ppm NO/Ar, 3002 ppm N2O/Ar, 5007 ppm N2/Ar, 4995 ppm H2/Ar). The atomic mass units (amu) related to NH3 (17), NO (30), N2O (44, 30, 28), N2 (28), H2 (2), and Ar (40) were analyzed. The fraction originated from N2O was deducted from the signal at 30 and 28 amu before NO and N2 quantification. The accuracy of the measurement was assured by performing nitrogen balances and the deviations were closed within 10% for all reported experiments.

$${{{{{\rm{Nitrogen}}}}}}\; {{{{{\rm{balance}}}}}}:{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{NH}}}}}}}_{{3,}_{{{{{{\rm{in}}}}}}}}}{{{{{\rm{dt}}}}}}= {\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{NH}}}}}}}_{{3,}_{{{{{{\rm{out}}}}}}}}}{{{{{\rm{dt}}}}}}+{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{NO}}}}}},}_{{{{{{\rm{out}}}}}}}}{{{{{\rm{dt}}}}}}\\ +2{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{N}}}}}}}_{2}{{{{{\rm{O}}}}}}{,}_{{{{{{\rm{out}}}}}}}}{{{{{\rm{dt}}}}}}+2{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{N}}}}}}}_{{2,}_{{{{{{\rm{out}}}}}}}}}{{{{{\rm{dt}}}}}}\,$$

The NH3 conversion and NOx selectivity were calculated using the following formulas:

$${X}_{{{{{{{\rm{NH}}}}}}}_{3}}=\frac{{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{NH}}}}}}}_{{3,}_{{{{{{\rm{in}}}}}}}}}{{{{{\rm{dt}}}}}}-{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{NH}}}}}}}_{{3,}_{{{{{{\rm{out}}}}}}}}}{{{{{\rm{dt}}}}}}}{{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{NH}}}}}}}_{{3,}_{{{{{{\rm{in}}}}}}}}}{{{{{\rm{dt}}}}}}}$$
$${S}_{{{{{{\rm{NO}}}}}}}=\frac{{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{NO}}}}}},}_{{{{{{\rm{out}}}}}}}}{{{{{\rm{dt}}}}}}}{{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{NO}}}}}},}_{{{{{{\rm{out}}}}}}}}{{{{{\rm{dt}}}}}}+2{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{N}}}}}}}_{2}{{{{{\rm{O}}}}}}{,}_{{{{{{\rm{out}}}}}}}}{{{{{\rm{dt}}}}}}+2{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{N}}}}}}}_{{2,}_{{{{{{\rm{out}}}}}}}}}{{{{{\rm{dt}}}}}}\,}$$
$${S}_{{{{{{{\rm{N}}}}}}}_{2}{{{{{\rm{O}}}}}}}=\frac{2{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{N}}}}}}}_{2}{{{{{\rm{O}}}}}}{,}_{{{{{{\rm{out}}}}}}}}{{{{{\rm{dt}}}}}}}{{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{NO}}}}}}}_{,{{{{{\rm{out}}}}}}}}{{{{{\rm{dt}}}}}}+2{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{N}}}}}}}_{2}{{{{{\rm{O}}}}}}{,}_{{{{{{\rm{out}}}}}}}}{{{{{\rm{dt}}}}}}+2{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{N}}}}}}}_{{2,}_{{{{{{\rm{out}}}}}}}}}{{{{{\rm{dt}}}}}}\,}$$
$${S}_{{{{{{{\rm{N}}}}}}}_{2}}=\frac{2{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{N}}}}}}}_{{2,}_{{{{{{\rm{out}}}}}}}}}{{{{{\rm{dt}}}}}}}{{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{NO}}}}}},}_{{{{{{\rm{out}}}}}}}}{{{{{\rm{dt}}}}}}+2{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{N}}}}}}}_{2}{{{{{\rm{O}}}}}}{,}_{{{{{{\rm{out}}}}}}}}{{{{{\rm{dt}}}}}}+2{\int }_{0}^{{{{{{\rm{t}}}}}}}{{{{{{\rm{n}}}}}}}_{{{{{{{\rm{N}}}}}}}_{{2,}_{{{{{{\rm{out}}}}}}}}}{{{{{\rm{dt}}}}}}\,}$$

where \({{{{{{\rm{n}}}}}}}_{{{{{{{\rm{NH}}}}}}}_{{3,}_{{{{{{\rm{in}}}}}}}}}\) is the molar flow rate of NH3 introduced into the reactor, \({{{{{{\rm{n}}}}}}}_{{{{{{{\rm{NH}}}}}}}_{{3,}_{{{{{{\rm{out}}}}}}}}}\), \({{{{{{\rm{n}}}}}}}_{{{{{{{\rm{NO}}}}}},}_{{{{{{\rm{out}}}}}}}}\), \({{{{{{\rm{n}}}}}}}_{{{{{{{\rm{N}}}}}}}_{2}{{{{{\rm{O}}}}}}{,}_{{{{{{\rm{out}}}}}}}}\), and \({{{{{{\rm{n}}}}}}}_{{{{{{{\rm{N}}}}}}}_{{2,}_{{{{{{\rm{out}}}}}}}}}\) is the molar flow rate of NH3, NO, N2O, and N2 from the gas outlet.

Catalyst characterization

Powder X-ray diffraction (XRD) measurements were carried out to analyze the crystalline phases of the catalysts during the redox reaction. XRD patterns were recorded on a PANalytical diffractometer with Cu Kα radiation (λ = 1.5418 Å) at 40 kV and 40 mA. Recording was performed in the 2θ range of 10−80° with a step size of 0.02°. HighScore plus software was employed for phase identification with reference to the International Center for Diffraction Data (ICDD) database. The peak position gave information about the lattice spacing based on the Bragg law of diffraction.

X-ray photoelectron spectroscopy measurements were adopted for near-surface chemical states analysis using Thermo Fisher Scientific ESCALAB 250 equipped with a monochromatic Al Kα (hυ = 1486.6 eV) in ultrahigh vacuum conditions (base pressure P 3 × 10−8 Pa). Pressed samples were attached to conductive carbon adhesive tapes and the XPS spectra were recorded at a constant pass energy 20 eV and a resolution energy of 0.1 eV. Peak fitting was performed with XPSPEAK program Shirley background subtraction together with a combination of Gaussian–Lorentzian functions. The C 1 s peak at 284.6 eV was used for calibration of all binding energies.

X-ray absorption near-edge spectroscopy (XANES) measurements around the V K-edge (5465 eV) were performed at beam line 4W1B of Beijing Synchrotron Radiation Facility (BSRF). The storage ring was operated at 2.2 GeV with a typical electron current of 50 mA. The redox catalysts were homogeneously smeared on Scotch adhesive tape and folded to reach the optimum absorption thickness. Each sample was packed in a polyethylene bag under an argon atmosphere. Spectra were collected at ambient temperature in transmission mode with a double crystal Si(111) monochromator and vanadium foil as a reference material for edge-shift calibration. Reagent grade C10H14O5V and VCl3 were used as V4+ and V3+ standards, respectively. Data were normalized in a standard curve-fitting procedure using the Athena software program.

In situ Raman spectroscopy studies were performed using a high-resolution LabRam HR800 confocal Raman microprobe (Horiba-Jobin Yvon, France) equipped with three laser excitations (532, 442, and 325 nm). The experiments were performed in an in situ high-temperature Linkam CCR1000 cell attached to a gas manifold. The spectra were collected using the 532-nm laser with five accumulations at 10 s per scan and a 200-μm laser hole size. Prior to the measurements, the sample was heated to 650 °C (10 °C min−1 ramp) under 2.2 ml/min Ar. The redox behavior and molecular structures of the catalyst was fingerprinted via the following two-step procedure. (i) Reduction under 5% NH3/Ar (2.2 ml/min). Raman spectra were collected during the reduction process until a constant intensity (or complete disappearance of the peak) was achieved. (ii) Oxidation under 30% O2/Ar (2.2 ml/min). Raman spectra were taken during the oxidation process until a constant intensity was achieved.

The surface morphology of the fresh and cycled catalyst was observed by scanning electron microscopy (SEM) on a JSM-7800F instrument using an accelerating voltage of 1 kV. Samples were dusted on an adhesive conductive carbon belt attached to a copper disk. High-resolution transmission electron microscopy (HRTEM) characterization was implemented on a JEOL JEM-2200F microscope operated at 200 kV with a linear resolution of 0.10 nm. Prior to measurement, catalyst sample was ultrasonicated in ethanol and deposited on copper grids coated with a holey carbon film, followed by drying at room temperature.

Computational details

DFT calculations were performed using the Vienna ab initio Simulations Package (VASP)67 with the frozen-core all-electron projector augmented wave (PAW) model68 and Perdew-Burke-Ernzerhof (PBE) functional69. A kinetic energy cutoff of 400 eV was used for the plane-wave expansion of the electronic wave functions, and the convergence criteria of force and energy were set to 0.01 eV/Å and 10−5 eV respectively. A Gaussian smearing of 0.1 eV was applied for optimizations and a k-point grid with a 9 × 7 × 3 Gamma centered mesh was chosen for single V2O5 bulk structure to sample the first Brillouin zone, and the same k-mesh spacing was kept for supercells. The GGA + U method was chosen to describe partially filled 3d orbitals in V with U = 4 eV and J = 0.68 eV for Hund’s exchange interaction70. The SSW HOWTOs program was applied to search for reasonable testing structures which were used for later transition state (TS) optimization with the climbing image nudged elastic band (CI-NEB) method71. For a reaction that has multiple final states, the most stable one was selected. All the computed energies are the total electronic energies unless otherwise stated. The Gibbs free energy corrections were calculated using VASPKIT, which includes the zero-point energy (ZPE), enthalpy, and entropy contributions. For surface adsorbates, free energies were approximated using the harmonic approximation that treats all degrees of freedom as vibrational modes72. The V2O5 (001) surface (Fig. S14a) was chosen for all the studied reactions as it is the most thermodynamically stable plane exposed in crystalline V2O573.