Suppressing CO_x in oxidative dehydrogenation of propane with dual-atom catalysts

Abstract

Oxidative dehydrogenation of propane (ODHP) is a promising route for propylene production, but achieving high selectivity towards propylene while minimizing CO_x byproducts remains a significant challenge for conventional metal oxide catalysts. Here we propose a solution to this challenge by employing atomically dispersed dual-atom catalysts (M₁M'₁-TiO₂ DACs). Ni₁Fe₁-TiO₂ DACs exhibit an ultralow CO_x selectivity of 5.2% at a high propane conversion of 46.1% and 520 °C, with stable performance for over 1000 hours. Mechanistic investigations reveal that these catalysts operate via a cooperative Langmuir-Hinshelwood mechanism, distinct from the Mars-van Krevelen mechanism typical of metal oxides. This cooperative pathway facilitates efficient conversion of propane and oxygen into propylene at the dual-atom interface. The superior selectivity arises from facile olefin desorption from the dual-atom sites and suppressed formation of electrophilic oxygen species, which are preferentially adsorbed on Fe₁ sites rather than oxygen vacancies. This work highlights the potential of dual-atom catalysts for highly selective ODHP and provides insights into their unique catalytic mechanism.

Introduction

The rising demand for propylene, driven by its widespread use in the production of polymers and chemicals, has intensified research into efficient and sustainable production methods^1,2. Oxidative dehydrogenation of propane (ODHP) offers a promising alternative to conventional propane dehydrogenation (PDH)^3,4, as it is exothermic and less susceptible to coke formation. However, achieving high selectivity towards propylene remains a major challenge. Conventional metal oxide catalysts (MOCs), typically operating via a Mars-van Krevelen (MvK) mechanism^{5,6,7,8,9,10,11,12}, suffer from high CO_x selectivity (>25%) at high propane conversions (>25%) due to the formation of electrophilic oxygen species on oxygen vacancies^13,14. Despite extensive efforts, few MOC-based catalysts have achieved CO_x selectivity below 10% while maintaining high propane conversion.

To address this limitation, we explore the potential of cooperative dual-atom catalysis, where distinct active sites can synergistically activate propane and oxygen. This approach aims to minimize the formation of electrophilic oxygen species by promoting O₂ adsorption on specific metal sites rather than oxygen vacancies. While the concept of cooperative catalysis has been demonstrated in other systems^{15,16,17,18,19,20,21}, its application in heterogeneous ODHP remains largely unexplored.

Herein, we report a series of supported dual-atom catalysts (6 types, M₁M'₁-TiO₂ DACs) synthesized via a modified cation-exchange strategy^22,23,24. These catalysts, featuring atomically dispersed metal pairs on a TiO₂ support, demonstrate exceptional performance in ODHP. In particular, Ni₁Fe₁-TiO₂ DACs exhibit an ultralow CO_x selectivity of 5.2% at a high propane conversion of 46.1% and 520 °C, with remarkable stability for over 1000 h. This performance surpasses previously reported catalysts, which typically show limited stability beyond 300 h under continuous ODHP conditions.

Results

Catalytic performance

The catalytic performance of metal oxide catalysts (MOCs), M₁-TiO₂ single-atom catalysts (SACs), and M₁M'₁-TiO₂ dual-atom catalysts (DACs) was evaluated for oxidative dehydrogenation of propane (ODHP). Our optimization revealed that when the loading of M₁ and M'₁ in M₁M'₁-TiO₂ was fixed at ~3 wt %, DACs (e.g. Fe₁Ni₁-TiO₂) exhibited high performance. However, deviations from this optimal metal (e.g. Fe and Ni) loading led to decreased propene selectivity (Supplementary Fig. 1). As shown in Fig. 1a, Fe₁Ni₁-TiO₂ achieved an ultralow CO_x selectivity of 5.2% while maintaining a high propane conversion of 46.1% at 520 °C. This performance surpasses previously reported metal-oxide-based ODHP catalysts (Supplementary Table 1), where CO_x selectivity typically exceeds 10% at comparable propane conversions (>25%). In contrast, both SACs and MOCs generated higher levels of CO_x under identical conditions. Fe₁Ni₁-TiO₂ also exhibited excellent olefin selectivity (90.6%) across a wide temperature range (420–520 °C, Supplementary Fig. 2). This superior CO_x suppression was consistently observed for other M₁M'₁-TiO₂ DACs, demonstrating the inherent advantage of dual-atom structures over single-atom structures or MOCs in ODHP.

Fig. 1: Catalytic performance of different catalysts for ODHP.

a The CO_x selectivity of as-prepared Ni₁Fe₁-TiO₂ DACs with reported ODHP catalysts (Supplementary Table 1). b Comparison of the TOF of C₃H₈ versus time on as-prepared DACs and SACs (Reaction conditions: C₃H₈: O₂: N₂ = 1:1:1, Total flow = 36 mL/min, Catalyst weight = 0.1 g, Temperature = 520 °C). c Comparison of STY versus time on our Ni₁Fe₁-TiO₂ DACs with reported ODHP catalysts. The STY, defined as the olefin production rate per gram of catalyst per hour, is presented in Supplementary Table 2 along with all reaction conditions and catalytic results. d The durability of Ni₁Fe₁-TiO₂ powder and shaping samples. Reaction conditions: C₃H₈: O₂: N₂ = 1:1:1, Volume rates of C₃H₈ = 12 mL/min with weight hour space velocity (WHSV) = 8.8 h⁻¹, 520 °C.

Further confirmation of the enhanced activity of DACs was obtained through an assessment of propane reaction rates. With a TOF of 56.2 s⁻¹, the Ni₁Fe₁-TiO₂ catalyst exhibited significantly higher activity compared to the monometallic Ni₁-TiO₂ and Fe₁-TiO₂ catalysts. This translated to an ultrahigh olefin yield of 40.7% at 520 °C, surpassing Fe₁-TiO₂ and Ni₁-TiO₂ by twofold (Supplementary Fig. 3). This enhanced catalytic activity was consistently observed across various DAC and SAC (Supplementary Figs. 4–5), underscoring the benefits of the dual-atom architecture.

The long-term stability and productivity of Ni₁Fe₁-TiO₂ were assessed by evaluating the olefin space-time yield (STY) over 1000 h of continuous operation (Fig. 1c). While previous studies have reported high olefin STY values for catalysts composed of multi-elemental, non-metal, or alloy materials (e.g., V-Fe-O, BN, PtSn/Beta, Supplementary Table 2), MOC-based ODHP catalysts have generally struggled to achieve STY values exceeding 50 mmolg_cat⁻¹h⁻¹ after 300 h. Remarkably, Ni₁Fe₁-TiO₂ exhibited an ultrahigh olefin STY of 83.3 mmolg_cat⁻¹h⁻¹ after 1000 h at 520 °C, effectively overcoming the trade-off between STY and stability. This performance surpasses state-of-the-art catalysts in terms of olefin yield and long-term durability (Fig. 1c, d, Supplementary Table 2). Furthermore, we successfully scaled up production to kilogram-scale quantities while maintaining ease in shaping the catalyst structure, highlighting its potential for industrial applications (Supplementary Fig. 6). The shaped Ni₁Fe₁-TiO₂ catalyst demonstrated comparable performance as its powdered counterpart, and also operated stably with a high olefin selectivity of ~90% at 520 °C (Fig. 1d, right).

Structure and characterization of dual-atom catalysts

The catalysts were synthesized via a solid-state calcination, a robust and sophisticated method involving a mixture of TiO₂, metal oxide (M = Fe₂O₃, CoO, NiO, MnO, MgO,…) salts, and alkali metal carbonates heated to high temperatures (Supplementary Fig. 7). The supported and catalysis were thoroughly characterized. The TiO₂ nanosheet support exhibited a BET surface area of 78.2 m²/g, with an interlayer spacing calculated to be 0.95 Å (Supplementary Figs. 8–10). Both SACs and DACs, including Ni₁Fe₁-TiO₂, Ni₁-TiO_2, and Fe₁-TiO₂, exhibited a lamellar structure consistent with the BET surface area >100 m²/g (Supplementary Figs. 11–14, Supplementary Table 3). In Ni₁Fe₁-TiO₂, Ni and Fe atoms were homogeneously dispersed throughout the nanosheets without any discernible agglomeration.

X-ray absorption near-edge structure (XANES) analysis confirmed the dual-atom structure of Ni₁Fe₁-TiO₂. Compared to Ni₁-TiO₂ and Fe₁-TiO₂, Ni₁Fe₁-TiO₂ displayed a dual scattering pathway in the k²-weighted extended X-ray absorption fine structure (EXAFS) spectra (Fig. 2a, b, Supplementary Fig. 15, Supplementary Table 4), consistent with the two domains observed in the wavelet transform signature (Supplementary Fig. 16). This indicates that the Ti lattice of the TiO₂ nanosheet support effectively confines isolated Ni and Fe atoms in close proximity^25,26,27. Pre-edge features in the XANES spectra clearly distinguished the electronic structures of Ni₁Fe₁ dual-atom sites from those of the isolated Ni₁-TiO₂ and Fe₁-TiO₂ sites. Charge transfer within the Ni₁-Fe₁ pair was evident, with Ni in Ni₁Fe₁-TiO₂ displaying a higher valence shift compared to Ni in Ni₁-TiO₂, and Fe in Ni₁Fe₁-TiO₂ exhibiting a lower valence shift compared to Fe in Fe₁-TiO₂. (Supplementary Figs. 17–20).

Fig. 2: Structural characterization of Ni₁Fe₁-TiO₂ DACs.

a, b k²-weighted EXAFS spectra of (a) Fe K-edge and (b) Ni K-edge for Ni₁Fe₁-TiO₂ and reference samples. k denotes the wave vector of the photoelectron. c The AC-STEM image, the line intensity profile of line 1 (insert, bottom), and corresponding EELS (side) for Ni₁Fe₁-TiO₂. The top view of Ni₁Fe₁-TiO₂ DACs is shown in the top right corner. d Experimental ⁵⁷Fe Mössbauer spectra of Fe₁-TiO₂ and Fe₁Ni₁-TiO₂ measured at 293 K. The doublets D1 and D2 represent Fe³⁺ and Fe²⁺ states, respectively.

Aberration-corrected scanning transmission electron microscopy (AC-STEM) and atomic-resolution electron energy loss spectroscopy (EELS) provided direct visualization of the heteronuclear dual-atom structure in Ni₁Fe₁-TiO₂ (Fig. 2c). The distinct line intensity profiles of Ni and Fe signals provide direct evidence for observing Ni₁-Fe₁ pairs along the Ti column (inset in Fig. 2c, Supplementary Fig. 21). The measured distance between Ni₁ and Fe₁ (3.1 Å) aligned with the scattering distance observed in XANES analysis, confirming their substitution for Ti sites and the formation of Ni₁-Fe₁ dual-atom sites within the TiO₂ lattice (Supplementary Fig. 22). Notably, when mono-metal oxides (NiO or Fe₂O₃) were introduced during synthesis, Fe and Ni exhibited a preference for isolated FeO₅ (Fe₁-TiO₂) or NiO₅ (Ni₁-TiO₂) configurations. The single-atom structure of Fe₁-TiO₂ and Ni₁-TiO₂ was further confirmed by AC-STEM (Supplementary Fig. 23) and XANES measurements (Fig. 2a, b, Supplementary Fig. 24), which revealed well-defined monodispersed spots along the Ti lattice and distinct features corresponding to Ni(1)/Fe(1)-O(5) coordination in the isolated systems (Supplementary Table 5).

Mössbauer spectrometry was employed to probe the Fe environments in Ni₁Fe₁-TiO₂ and Fe₁-TiO₂. Compared to Fe₁-TiO₂, the proportion of Fe²⁺ species increased from 9.3% to 21.0% in Ni₁Fe₁-TiO₂ (Fig. 2d), indicating a decrease in the average oxidation state of Fe^28,29,30. This shift in oxidation states between individual Ni₁ and Fe₁ atoms can be attributed to charge compensation within the Fe-Ni pairs in Ni₁Fe₁-TiO₂. This charge redistribution facilitates pairing between divalent Ni²⁺ and trivalent Fe³⁺ to fulfill local coordination requirements and maintain overall charge neutrality, rather than dimerization involving solely +2/ + 3 charged species^31,32,33.

The synthetic procedure was extended to create a library of heteronuclear dual-atom catalysts in TiO₂. A series of M₁-TiO₂ catalysts (M = Fe, Co, Ni, Mn) (Supplementary Figs. 25–26, Supplementary Table 4) and M₁M'₁-TiO₂ DACs (M₁M'₁ = Ni₁Fe₁, Co₁Ni₁, Co₁Mn₁, Ni₁Mn₁, Fe₁Mn₁, Fe₁Co₁) were successfully synthesized using different precursors. Comprehensive structural characterizations, including XRD analysis, TEM imaging, AC-STEM observations, and XANES measurements (Supplementary Figs. 27–34, Supplementary Table 5), confirmed the dual-atom feature in these catalysts.

Catalytic mechanism

In the ODHP process, MOCs undergo either the commonly observed MvK mechanism (Fig. 3a, left) or an alternative Langmuir-Hinshewood (L-H) mechanism (Fig. 3a, right), depending on whether lattice oxygen exchange is involved. To verify the mechanism, we conducted a study using the ¹⁶O/¹⁸O isotope exchange method. Catalysts were initially pre-treated with ¹⁶O₂ to ensure lattice oxygen was predominantly ¹⁶O (Fig. 3a). Upon switching to a reaction gas mixture of ¹⁸O₂ and C₃H₈, Ni₁-TiO₂ and Fe₁-TiO₂ exhibited H₂¹⁶O production, indicative of lattice oxygen involvement (Fig. 3b). This H₂¹⁶O signal gradually decreased as lattice ¹⁶O was replaced by ¹⁸O from the gas phase, consistent with the MvK mechanism^14,34. In contrast, Ni₁Fe₁-TiO₂ showed negligible H₂¹⁶O production and a rapid H₂¹⁸O response (Fig. 3c), with no evidence of ¹⁶O₂ or ¹⁶O¹⁸O formation (Supplementary Fig. 35). This suggests that lattice oxygen exchange is minimal, and the ODHP reaction over Ni₁Fe₁-TiO₂ likely proceeds through the L-H mechanism, where C₃H₈ and O₂ co-adsorb on the catalyst surface.

Fig. 3: Catalysis mechanism studies.

a Schematic diagram of MvK and L-H mechanisms. b, c H₂¹⁶O and H₂¹⁸O response signals of reactions of ¹⁸O₂–C₃H₈ mixture on Ni₁-TiO₂, Fe₁-TiO_2, and Ni₁Fe₁-TiO₂ followed by He purge at 520 °C (10% C₃H₈/Ar, 10 mL/min and 10% O¹⁸₂/Ar, 10 mL/min). d, e EPR spectra of spent Ni₁-TiO₂, Fe₁-TiO_2, and Ni₁Fe₁-TiO₂ catalysts without and with pretreatment with 10% C₃H₈/Ar at 520 °C for 2 h.

To further probe the involvement of lattice oxygen, electron paramagnetic resonance (EPR) spectroscopy was employed to quantify oxygen vacancies in spent SACs and DACs before and after pre-treatment with 10% C₃H₈/Ar at 520 °C. The EPR intensity of oxygen vacancies increased approximately twofold for Ni₁-TiO₂ and Fe₁-TiO₂ after C₃H₈ pre-treatment (Fig. 3d, Supplementary Fig. 36), whereas only a 1.2-fold increase was observed for Ni₁Fe₁-TiO₂ (Fig. 3e). The generation of more oxygen vacancies in SACs compared to DACs further supports the notion that SACs experience an MvK pathway while DACs undergo an L − H one for oxygen activation.

Mechanistic insights from in-situ spectroscopy

To investigate the interplay between reactants (C₃H₈, O₂) and the active sites (Fe, Ni) in Ni₁Fe₁-TiO₂, we employed in-situ X-ray emission spectroscopy (XES). This technique is particularly sensitive to the chemical environment of transition metals and can effectively differentiate between ligands with similar atomic numbers, such as C, N, and O^35,36,37. We designed an in-situ XES experiment by gradually increasing the temperature to 520 °C within a helium stream and acquiring the ‘fresh’ spectrum after stabilization. Subsequently, different gases were sequentially introduced (pure C₃H₈ → C₃H₈ + O₂ → pure O₂), and spectra were collected following each stabilized condition.

For Ni₁Fe₁-TiO₂ catalysts, Fig. 4a and b present the Kβ_2,5 XES of Ni and Fe during distinct steps, respectively. The XES spectra revealed both Ni-C and Fe-C bonds in a pure C₃H₈ atmosphere with binding energies of 7102.1 eV and 8326.1 eV, respectively^38,39. Upon the addition of O₂, there was a gradual disappearance of Fe-C bonds compared to persistent but weakened Ni-C bonds observed. After removing C₃H₈, Ni-C bonds remained while Fe-C bonds almost completely disappeared indicating that Ni sites exhibit higher propensity for propane adsorption than Fe sites.

Fig. 4: The adsorption capabilities of DACs towards C₃H₈ and O₂.

a, b In-situ Kβ _2,5 and Kβ′′ mainline XES spectra recorded Ni₁Fe₁-TiO₂ under different steps at 520 °C. c, d In-situ XANES absorption edge spectra for the Ni K-edge and Fe K-edge of Ni₁Fe₁-TiO₂ under different steps at 520 °C. e Schematic illustration of C₃H₈ (Ni site) and O₂ (Fe site) reaction over Ni₁Fe₁-TiO₂. f Schematic illustration of C₃H₈ (Ni site) and O_lattic reaction over Ni₁-TiO₂. In-situ XES and XANES were measured by gradually increasing the temperature to 520 °C within a helium stream after stabilization; different gases were sequentially introduced (pure C₃H₈ → C₃H₈ + O₂ → pure O₂), and spectra were collected at each stabilized condition.

This preferential adsorption behavior was further corroborated by temperature-programmed desorption (TPD) experiments. C₃H₈ desorption capacity was lowest at the Fe₁ site in Fe₁-TiO₂ compared to Ni₁-TiO₂ and Ni₁Fe₁-TiO₂ (Supplementary Fig. 37), while a minor O₂ desorption peak at 432.0 °C was observed for Ni₁-TiO₂, which was absent in the other two samples (Supplementary Fig. 38). These results are consistent with the XES data, suggesting preferential adsorption of C₃H₈ on Ni and O₂ on Fe in Ni₁Fe₁-TiO₂.

In-situ XANES analysis provided insights into the valence state evolution of Fe and Ni in Ni₁Fe₁-TiO₂ under the same reaction conditions as the XES experiments. Changes in the Ni K-edge and Fe K-edge XANES spectra (Fig. 4c, d) revealed that: (1) C₃H₈ purging at 520 °C induced a shift to lower energies, indicating reduction of both Ni and Fe; (2) Introducing O₂ significantly increased the Fe valence state with minimal impact on the Ni valence state; (3) Removing C₃H₈ under O₂ flow led to a further increase in the Fe valence state. These findings, supported by DFT calculations (Supplementary Fig. 39), suggest that Fe sites are primarily responsible for oxygen activation. The limited change in the Ni valence state upon O₂ introduction is attributed to the strong adsorption of residual C₃H₈ on Ni sites. Further in-situ XANES studies confirmed the facile reduction of Ni by C₃H₈ (Supplementary Figs. 40–43) and the strong interaction between Fe₁ sites and O₂ (Supplementary Figs. 44–45). Collectively, the in-situ XES, C₃H₈/O₂-TPD, and in-situ XANES data provide compelling evidence for distinct roles of Ni and Fe in Ni₁Fe₁-TiO₂: Ni sites preferentially adsorb C₃H₈, while Fe sites favor O₂ adsorption and activation.

DFT calculations were performed to elucidate the ODHP reaction pathway over Ni₁Fe₁-TiO₂ (Supplementary Figs. 46–47). The results suggest a highly efficient L-H mechanism (Fig. 4e) where C₃H₈ dehydrogenation occurs at Ni₁ sites, while activated O₂ on Fe₁ sites promotes hydrogen coupling to form H₂O and C₃H₆. In contrast, DFT calculations for the single-atom catalysts (Ni₁-TiO₂ and Fe₁-TiO₂) suggest an MvK mechanism (Supplementary Figs. 48–51, Fig. 4f), highlighting the distinct mechanistic behavior of DACs and SACs. The key difference lies in the involvement of lattice oxygen exchange in the MvK mechanism, where oxygen vacancies play a crucial role in driving propane dehydrogenation.

Suppression of CO_x formation

One of the major challenges faced by ODPH catalysts is the generation of significant amounts of valueless CO_x (>30% in selectivity for MOCs). A consensus among literature reports suggests that the majority of CO_x species originate from propylene oxidation reaction with electrophilic oxygen species generate by lattice oxygen exchange through Mv-K pathways (Supplementary Fig. 52)^14,40,41. To assess the propensity for C₃H₆ over-oxidation, we analyzed C₃H₆-TPD profiles (Fig. 5a). Ni₁Fe₁-TiO₂ exhibited a lower C₃H₆ desorption temperature (138.7 °C) compared to Ni₁-TiO₂ (170.1 °C) and Fe₁-TiO₂ (162.2 °C), indicating weaker C₃H₆ binding and enhanced propylene release. This observation was corroborated by in-situ diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) analysis, which revealed significantly faster C₃H₆ desorption rates over Ni₁Fe₁-TiO₂ compared to the single-atom counterparts (Supplementary Figs. 53–55). DFT calculations further support this trend, showing lower C₃H₆ desorption energies from the Ni₁ site in Ni₁Fe₁-TiO₂ compared to Ni and Fe sites in Ni₁-TiO₂ and Fe₁-TiO₂ (Supplementary Fig. 56). The facile desorption of C₃H₆ from Ni₁Fe₁-TiO₂ effectively mitigates its over-oxidation.

Fig. 5: Mechanism of CO_x inhibition.

a C₃H₆-TPD analysis of of Ni₁-TiO₂, Fe₁-TiO₂ and Ni₁Fe₁-TiO₂. b In-situ FT-IR spectra of Ni₁Fe₁-TiO₂ with 10 vol% C₃H₈/Ar (10 mL/min) and 10 vol% O₂/Ar (10 mL/min), (350–500 °C, recorded at intervals of 50 °C, down to top), Fe₁-TiO₂ (500 °C) and Ni₁-TiO₂ (500 °C) at enlarge view 1850–1550 cm⁻¹. c, d EPR spectra of Ni₁-TiO₂, Fe₁-TiO_2, and Ni₁Fe₁-TiO₂ after pretreatment with 10% O₂/Ar. e Schematic diagram of CO_x formation.

To identify potential intermediates in the C₃H₆ over-oxidation pathway, we performed in-situ DRIFTS analysis under reaction conditions (Fig. 5b). Ni₁Fe₁-TiO₂ displayed a distinct C = C stretching vibration at 1650 cm⁻¹, characteristic of C₃H₆, whereas Ni₁-TiO₂ and Fe₁-TiO₂ exhibited a prominent peak at 1738 cm⁻¹, indicative of aldehyde groups^42,43. Furthermore, a distinct CO₂ gas-phase peak at 2349 cm⁻¹ was observed for Ni₁-TiO₂ and Fe₁-TiO₂, but was barely discernible for Ni₁Fe₁-TiO₂ (Supplementary Fig. 57). These findings suggest that acrolein, an aldehyde species, is a key intermediate in the deep oxidation of propylene to CO_x. The suppression of acrolein formation over Ni₁Fe₁-TiO₂ effectively limits CO_x generation.

Importantly, this suppression of over-oxidation was observed for all synthesized M₁M'₁-TiO₂ DACs (6 types, Supplementary Figs. 58–67), highlighting the general efficacy of this catalyst design. To understand the underlying cause, we examined the nature of oxygen species using electron paramagnetic resonance (EPR) spectroscopy. Fe₁-TiO₂ and Ni₁-TiO₂ exhibited a weak isotropic signal at g = 2.02 (Fig. 5c, d), indicative of electrophilic O₂⁻ species^44,45, which are known to promote CO_x formation. This signal was nearly absent in Ni₁Fe₁-TiO₂, suggesting a lower concentration of electrophilic oxygen species. This is attributed to the preferential adsorption of O₂ on Fe₁ sites in Ni₁Fe₁-TiO₂, rather than on oxygen vacancies as in MOCs, leading to a reduced contribution of electrophilic oxygen.

Figure 5e illustrates the proposed mechanisms of CO_x formation for NiO_x and Ni₁Fe₁ species. Trace amounts of electrophilic oxygen species and relatively low propylene desorption energy for the Ni₁Fe₁-TiO₂ catalyst effectively suppressed acrolein (a significant byproduct in the deep oxidation process) formation, thereby reducing CO_x generation. In contrast, the stronger propylene adsorption on Ni₁ or Fe₁-TiO₂ catalysts facilitated the formation of acrolein from propylene with electrophilic oxygen species, leading to further overoxidation to CO_x.

Discussion

Atomically dispersed dual-atom catalysts (DACs) offer a promising avenue for enhancing ODHP by mitigating CO_x formation. Here, we report a general synthetic strategy for M₁M'₁-TiO₂ DACs and demonstrate their exceptional performance in ODHP. Specifically, Ni₁Fe₁-TiO₂ exhibits an ultralow CO_x selectivity of 5.2% while maintaining a high propane conversion of 46.1% for 1000 h at 520 °C. Mechanistic investigations, combining experimental characterization and theoretical calculations, reveal a Langmuir-Hinshelwood mechanism facilitated by synergistic interactions between the Ni₁ and Fe₁ sites. Crucially, the Fe₁ sites promote O₂ adsorption, effectively suppressing the formation of electrophilic oxygen species that lead to CO_x generation. Furthermore, the Ni₁Fe₁-TiO₂ DAC exhibits a lower energy barrier for propylene desorption compared to single-atom Ni₁-TiO₂ and NiO catalysts, contributing to its enhanced catalytic efficiency. This work highlights the potential of precisely engineered DACs to optimize ODHP performance by manipulating the adsorption and activation of both propane and oxygen.

Methods

Materials

Titanium (IV) oxide (TiO₂, rutile, 99.5%, CAS: 1317-80-2) and potassium carbonate (K₂CO₃, 99.9%, CAS: 534–17-8) were purchased from Alfa Aesar. Manganese (II) oxide (MnO, 99.5%), nickel (II) oxide (NiO, 99%), iron(III) oxide (Fe₂O₃, 99.9%), and cobalt(II) oxide (CoO, 99.9%) were obtained from Innochem. Propane (C₃H₈, 99.9%), oxygen (O₂, 99.999%), and nitrogen (N₂, 99.999%) were supplied by Beijing Huatong Jingke Gas Chemical Co., LTD.

Catalyst synthesis

TiO₂ nanosheets

Briefly, K₂CO₃ (1.11 g), rutile TiO₂ (2.77 g), and Li₂CO₃ (0.2 g) were ground for 1 h and calcined at 900 °C for 1.5 h. The resulting product was ground again for 1 h and calcined at 1000 °C for 20 h to yield layered titanate (K_0.8Ti_1.73Li_0.27O₄). The titanate was then protonated by treatment with 1 mol L⁻¹ HCl (100 mL/g titanate) with daily replacement of the HCl solution for 3 days. The resulting catalysis was washed with distilled water until neutral and dried under vacuum.

M₁-TiO₂ nanosheets

M₁-TiO₂ nanosheets (M = Mn, Fe, Co, Ni) were prepared by adapting a solid reaction method¹. Stoichiometric quantities of the respective metal oxide (MnO, Fe₂O₃, CoO, or NiO), K₂CO₃, and anatase TiO₂ (see below for molar ratios) were ground (grinding should be performed at ≤40% relative humidity to prevent K₂CO₃ deliquescence.) and calcined at 1000 °C for 1 h, followed by regrinding and calcination at 1000 °C for 20 h. The molar ratios of the reactants were determined according to the following equations:

$${{{\rm{For}}}}\,{{{{\rm{M}}}}}^{2+}({{{\rm{MnO}}}},{{{\rm{CoO}}}},{{{\rm{NiO}}}})\!\!:\,{{{{\rm{K}}}}}_{0.8}{{{{\rm{Ti}}}}}_{(5.2-{{{\rm{x}}}})/3}{{{{\rm{Li}}}}}_{(0.8-2{{{\rm{x}}}})/3}{{{{\rm{M}}}}}_{{{{\rm{x}}}}}{{{{\rm{O}}}}}_{4}$$

(1)

$${{{\rm{For}}}}\,{{{{\rm{M}}}}}^{3+}({{{{\rm{F}}}}}{{{{\rm{e}}}}_{2}}{{{{\rm{O}}}}}_{3})\!\!:{{{{\rm{K}}}}}_{0.8}{{{{\rm{Ti}}}}}_{(5.2-2{{{\rm{x}}}})/3}{{{{\rm{Li}}}}}_{(0.8-{{{\rm{x}}}})/3}{{{{\rm{M}}}}}_{{{{\rm{x}}}}}{{{{\rm{O}}}}}_{4}$$

(2)

With Ni₁-TiO₂ as an example, the doping ratio of Ni is x = 0.2. According to formula 1, we can calculate the molar ratios of different components (K_0.8Ti_1.67Li_0.13Ni_0.2O₄). K₂CO₃ (0.04 mol (K⁺ = 0.08 mol), 5.53 g), rutile TiO₂ (0.167 mol, 13.3 g), NiO (0.02 mol, 1.5 g) and Li₂CO₃ (0.0065 mol (Li⁺ = 0.013 mol), 0.48 g) were ground for 1 h and calcined at 900 °C for 1.5 h. The product after calcination was ground again for 1 h and calcined at 1000 °C for 20 h to yield layered titanate. The titanate was then protonated by treatment with 1 mol L⁻¹ HCl (100 mL/g titanate) with daily replacement of the HCl solution for 3 days. Acid washing with hydrochloric acid exchanges interlayer K⁺ and Li⁺ for protons and removes surface undoped metals, as reported in previous literature. The resulting catalysis was washed with distilled water until neutral and dried under vacuum.

M₁M'₁-TiO₂ nanosheets

M₁M'₁-TiO₂ nanosheets (M₁M'₁ = Ni₁Fe₁, Ni₁Co₁, Ni₁Mn₁, Fe₁Co₁, Fe₁Mn₁, Co₁Mn₁) were synthesized following the procedure for M₁-TiO₂ nanosheets, using the general formula K_0.8Ti_{(5.2−3(x+y))/3}Li_0.8/3M_xN_yO₄ to determine the molar ratios of the metal oxides. With Ni₁Fe₁-TiO₂ as an example, the doping ratio is x = 0.1 for Ni and y = 0.1 for Fe. According to the formula, we can calculate the molar ratios of different components (K_0.8Ti_1.53Li_0.26Ni_0.1Fe_0.1O₄). K₂CO₃ (0.04 mol (K⁺ = 0.08 mol), 5.53 g), rutile TiO₂ (0.153 mol, 12.22 g), NiO (0.01 mol, 0.75 g), Fe₂O₃ (0.005 mol (Fe³⁺ = 0.01 mol), 0.8 g) and Li₂CO₃ (0.013 mol (Li⁺ = 0.026 mol), 0.96 g) were ground for 1 h and calcined at 900 °C for 1.5 h. The product after calcination was ground again for 1 h and calcined at 1000 °C for 20 h to yield layered titanate. The subsequent processing method is consistent with the preparation method of the M1-TiO₂ catalyst.

Catalyst shaping process

A catalyst precursor was prepared by ball-milling 250 g of catalyst powder with 50 g of hydroxypropyl cellulose (HPC) binder in a jar mill for 10 h at 300 rpm. Water was added to the powder mixture in a kneader to achieve a water-to-powder mass ratio of 0.1. The water was introduced gradually over 10 min to ensure uniform distribution. Kneading was then continued to achieve a homogeneous paste. The resulting paste was extruded through a twin-screw extruder at a screw speed of 7−10 rpm to produce catalyst strips. The extruded strips were dried in an oven and subsequently calcined in a muffle furnace to obtain the final clover-shaped catalyst.

Material characterizations

The physicochemical properties of the catalysts were characterized using a variety of techniques.

X-ray diffraction (XRD)

Crystalline phases were identified by powder X-ray diffraction using a Rigaku D/Max-2500 diffractometer equipped with a Shimadzu detector and Cu Kα radiation (λ = 1.5418 Å). The operating conditions were 100 mA and 40 kV, with a scan range of 10–90° 2θ and a scan rate of 5.0°/min.

Transmission electron microscopy (TEM)

The morphology and microstructure of the catalysts were investigated using a JEOL 2100 F transmission electron microscope operating at an accelerating voltage of 200 kV. High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images were acquired on a JEM-ARM300F microscope.

X-ray photoelectron spectroscopy (XPS)

Surface elemental composition and chemical states were analyzed by X-ray photoelectron spectroscopy (XPS) using an ESCALAB250XI spectrometer (Thermo Fisher Scientific).

X-ray absorption near-edge structure (XANES)

XANES measurements at the K-edge were performed at beamline 1W1B at the Beijing Synchrotron Radiation Facility (BSRF) in total electron yield (TEY) mode. The electron energy was 2.5 GeV with a beam current of 250 mA. Data processing, including energy calibration, background subtraction, and normalization, was carried out using the Athena software package. Extended X-ray absorption fine structure (EXAFS) analysis was performed using the Artemis module of IFEFFIT. The χ (k) data were Fourier transformed to R-space using a Hanning window (d_k = 1.0 Å⁻¹) over a k-range of 2–12 Å⁻¹. Least-squares curve fitting was performed in R-space to obtain quantitative structural information.

Temperature-programmed desorption (TPD)

TPD of O₂, C₃H₈, and C₃H₆ was performed using an AutoChemII 2920 chemisorption analyzer. Before each experiment, the sample was pretreated by heating to 300 °C at a rate of 10 °C/min under a He flow (30 mL/min) and held for 1 h to remove any adsorbed species. After cooling to 50 °C, the sample was exposed to a 10% O₂/He mixture (30 mL/min) for 2 h for O₂-TPD, or to pure C₃H₈ or C₃H₆ (30 mL/min) for 2 h for hydrocarbon TPD. Weakly adsorbed species were removed by purging with He (30 mL/min) for 1 h. Finally, the temperature was ramped to 800 °C at a rate of 10 °C/min under He flow, and the desorbed species were monitored using a thermal conductivity detector (TCD).

In-situ diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS)

In-situ DRIFTS measurements were conducted using a Bruker Tensor II spectrometer equipped with a mercury cadmium telluride (MCT) detector and CaF₂ windows. Spectra were acquired in the range of 900–3800 cm⁻¹ with a resolution of 4 cm⁻¹ and an accumulation of 32 scans. A 100 mg sample of catalyst was loaded into the in-situ reaction cell and pretreated in pure Ar at 250 °C for 1 h (flow rate: 10 mL/min) to remove adsorbed species. After cooling to 50 °C, a background spectrum was collected under Ar flow. A reaction gas mixture of 10 vol% C₃H₈/Ar (10 mL/min) and 10 vol% O₂/Ar (10 mL/min) was then introduced into the cell, and spectra were recorded. The cell was subsequently heated to the desired temperature over 30 min, and further spectra were acquired.

C₃H₆-desorption FT-IR experiments were performed using the same spectrometer. The catalyst was pretreated in pure Ar at 250 °C for 1 h (flow rate: 10 mL/min) and then cooled to room temperature. A background spectrum was collected. C₃H₆/Ar (10 vol% C₃H₆) was introduced into the cell at a flow rate of 10 mL/min for 1 h to ensure saturated adsorption of C₃H₆ on the catalyst surface. The gas flow was then switched to pure Ar (5 mL/min) to initiate the desorption process. IR spectra were recorded at 2 min intervals. Desorbed species were analyzed using a downstream quadrupole mass spectrometer, and the characteristic peaks for water (m/z = 18) and argon (m/z = 20) were integrated to quantify desorption products.

In-situ XANES measurements

In-situ X-ray absorption near-edge structure (XANES) measurements at the Ni and Fe K-edges were performed in transmission mode at beamline 1W1B of the Beijing Synchrotron Radiation Facility (BSRF). The catalyst sample was pressed into a self-supporting wafer and loaded into a reaction microdevice equipped with Be windows. The sample was heated to 200 °C under a high-purity He flow and held for 30 min. A reaction gas mixture of C₃H₈/He 10 vol% (10 mL/min) and 10 vol% O₂/Ar (10 mL/min) was then introduced into the cell, and XANES spectra were collected. The cell was subsequently heated to the desired temperature over 30 min, and further XANES spectra were acquired. All XANES data were processed using the Athena software package.

In-situ XANES measurements at different atmospheres

The catalyst sample was heated to 520 °C under a high-purity He flow and held for 30 min before acquiring XANES spectra. A gas mixture of C₃H₈ (10 mL/min) was then introduced into the cell, and spectra were collected after 1 h. O₂ (10 mL/min) was then added to the gas flow, and XANES spectra were acquired at 1 h intervals. Finally, the C₃H₈ flow was stopped, and after 30 min, the final XANES spectra were collected.

In-situ X-ray emission spectroscopy (XES) measurements at different atmospheres

In-situ XES measurements were performed using a von Hamos-geometry wavelength-dispersive X-ray spectrometer at the BL14W1-XANES beamline of the Shanghai Synchrotron Radiation Facility (SSRF). Ni Kβ_2,5 and Fe Kβ_2,5 emission lines were collected using Si (551) and Si (531) cylindrically bent crystal analyzers, respectively, with a radius of curvature of 25 cm. A Pilatus 100 K detector equipped with CMOS sensors was used to collect the dispersed fluorescence. Kohzu positioning stages were used for the precise alignment of the spectrometer components. The electron storage ring operated at 3.5 GeV with a beam current of 240 mA in top-up injection mode. The photon flux of the BL14W1-XANES beamline was 5 × 10¹² phs s⁻¹ at 10 keV, with a Si (111) double-crystal monochromator and a beam size of 100 × 200 μm² (vertical × horizontal).

The catalyst sample was heated to 520 °C under a high-purity He flow and held for 30 min before acquiring XES spectra. A gas mixture of C₃H₈ (10 mL/min) and N₂ (10 mL/min) was then introduced into the cell, and spectra were collected after 1 h. O₂ (10 mL/min) was then added to the gas flow, and XES spectra were acquired at 1 h intervals. Finally, the C₃H₈ flow was stopped, and after 30 min, the final XES spectra were collected.

¹⁸O₂ isotopic tracer studies

Isotopic tracer studies were conducted using ¹⁸O₂ in a high-temperature reaction chamber. A 100 mg sample of catalyst was placed in the cell and pretreated in 10% ¹⁶O₂/90% He at 773 K for 1 h. After purging with pure He for 10 min, the gas flow was switched to a mixture of 10% ¹⁸O₂/He and 10% C₃H₈/He at a flow rate of 5 mL/min. The isotopic composition of the reactor effluent was monitored using an online mass spectrometer (Shimadzu QP2010Plus) at 6 s intervals.

Catalytic performance evaluation

The catalytic performance for oxidative dehydrogenation of propane (ODHP) was evaluated in a fixed-bed reactor (Torch Instrument Co., LTD) using a quartz tube reactor (outer diameter: 8 mm, inner diameter: 6 mm, length: 480 mm) at a pressure of 0.1 MPa. Mass flow controllers (BROOKS Instruments Co., Ltd.) were used to regulate the flow of feed gases. The catalyst was initially received in tablet form. The catalyst was sieved to 60 mesh and 150 mg was loaded into the reactor, positioning it within the isothermal zone of the furnace The reaction temperature was controlled using two thermocouples, one placed in the furnace and the other in the quartz tube near the catalyst bed, with a heating rate of 10 °C/min. The feed gas consisted of a C₃H₈/O₂/N₂ mixture with a volume ratio of 1:1:1, and the weight hourly space velocity (WHSV) was maintained at 8.8 h⁻¹. The total gas flow used in the experiments was C₃H₈:O₂:N₂ = 12:12:12 (mL/min). The catalyst was tested without any pretreatment. The composition of the reactor effluent was analyzed online using a gas chromatograph (PANNA A91) equipped with a flame ionization detector (FID) and a thermal conductivity detector (TCD). A capillary column (Agilent, HP-AL/S, 25  m * 0.32 mm * 8 μm) with FID for hydrocarbons and TCD for inorganic gases (H₂, N₂, O₂, CH₄, CO, CO₂) was used for chromatographic analysis. The TCD detector utilized a multi-column system consisting of 1-meter and 2 meter HaySep columns followed by a 3 meter 5 A molecular sieve column for the separation of H₂, CO₂, O₂, N₂, CH₄, and CO. Propane conversion and product selectivity were calculated based on the corrected peak area normalization method.

Data Analysis

Propane and oxygen conversion were calculated using the following Eqs. (3–4):

$${{{{\rm{C}}}}}_{3}{{{{\rm{H}}}}}_{8}\,{{{\rm{conversion}}}}\!\!:{X}_{{{{{\rm{C}}}}}_{3}{{{{\rm{H}}}}}_{8}}\left(\%\right)=\frac{{\left[{F}_{{{{{\rm{C}}}}}_{3}{{{{\rm{H}}}}}_{8}}\right]}_{{{{\rm{in}}}}}-{\left[{F}_{{{{{\rm{C}}}}}_{3}{{{{\rm{H}}}}}_{8}}\right]}_{{{{\rm{out}}}}}}{{\left[{F}_{{{{{\rm{C}}}}}_{3}{{{{\rm{H}}}}}_{8}}\right]}_{{{{\rm{in}}}}}}\times 100$$

(3)

$${{{{\rm{O}}}}}_{2}\,{{{\rm{conversion}}}}\!\!:{X}_{{{{{\rm{O}}}}}_{2}}\left(\%\right)=\frac{{\left[{F}_{{{{{\rm{O}}}}}_{2}}\right]}_{{{{\rm{in}}}}}-{\left[{F}_{{{{{\rm{O}}}}}_{2}}\right]}_{{{{\rm{out}}}}}}{{\left[{F}_{{{{{\rm{O}}}}}_{2}}\right]}_{{{{\rm{in}}}}}}\times 100$$

(4)

where, ${F}_{{{{{\rm{C}}}}}_{3}{{{{\rm{H}}}}}_{8}}^{{in}}$ and ${F}_{{{{{\rm{C}}}}}_{3}{{{{\rm{H}}}}}_{8}}^{{out}}$ indicate the inlet and outlet flow rates of x (mL·min⁻¹), respectively.

Selectivity of hydrocarbon products is calculated as follows:

$${{{\rm{Sel}}}}\,\left(\%\right)=\frac{{n}_{{{{\rm{i}}}}}\times {\left[{F}_{i}\right]}_{{{{\rm{out}}}}}}{\left(\sum {n}_{{{{\rm{i}}}}}\times {\left[{F}_{{{{\rm{i}}}}}\right]}_{{{{\rm{out}}}}}\right)}\times 100$$

(5)

Where i represents the hydrocarbon products in the effluent gas, ni is the number of carbon atoms of the component i, and Fi is the corresponding flow rate.

The carbon balance of products was determined from below:

$${{{\rm{C\; balance}}}}=\frac{\left(\sum {{{{\rm{n}}}}}_{{{{\rm{i}}}}}\times {\left[{F}_{i}\right]}_{{{{\rm{out}}}}}+3\times {\left[{F}_{{{{{\rm{C}}}}}_{3}{{{{\rm{H}}}}}_{8}}\right]}_{{{{\rm{out}}}}}\right)}{3\times {\left[{F}_{{{{{\rm{C}}}}}_{3}{{{{\rm{H}}}}}_{8}}\right]}_{{{{\rm{in}}}}}}\times 100$$

(6)

Computational details

All theoretical calculations were performed using the Vienna Ab-Initio Simulation Package (VASP, version 5.4.4) based on the plane-wave pseudopotential density functional theory (DFT) methods. The projector-augmented wave (PAW) method was used to describe the electron-core interaction. The kinetic cut-off energy of the plane-wave basis set was 400 eV, and single gamma-point sampling was used for Brillouin zone integration. The total energy and forces convergence thresholds for the geometry optimization were set to 10–5 eV and 0.02 eV Å⁻¹, respectively. The vacuum spacing of 18 Å was added to Ni₁Fe₁-TiO₂ to avoid interactions between the layer plane along the z-direction in our manuscript. The TiO₂ surface was modeled for all DFT calculations, we finally used a hexagonal unit cell with a = 15.012 Å, b = 15.012 Å, and c = 18.342 Å.