Propylene is one of the most diverse petrochemical building blocks used for the production of many chemicals (e.g., polypropylene, propylene oxide, and acrylonitrile). The co-production of propylene from steam and fluidized crackers is anticipated to be insufficient to satisfy the rapidly growing demand1. Consequently, there is a need for the development of economic on-purpose production techniques to produce additional propylene. The direct dehydrogenation of propane (DDP) is thermodynamically limited and is highly endothermic (Δ r  = 29.70 kcal/mol), requiring temperatures that may exceed 973 K for significant propylene yields2. In principle, the introduction of CO2 as a mild oxidant into the feed alters the dehydrogenation pathway by oxidizing the abstracted hydrogen from the alkane and consequently releasing the heat of reaction that reduces operating temperatures2,3. The presence of CO2 can also increase the equilibrium conversion of propane by consuming H2 through the reverse water gas shift reaction (RWGS), as seen in the thermodynamic calculations in Fig. 1a. Additionally, unlike regular oxidative dehydrogenation with molecular oxygen, CO2 as a mild oxidant suppresses over-oxidation and thus minimizes the production of carbon oxides. The reactions of propane and CO2 also have the potential to employ two underutilized4,5,6 reactants to supply propylene as well as to mitigate detrimental CO2 emissions7,8.

Fig. 1
figure 1

Thermodynamic equilibrium plots. Equilibrium calculations were performed through HSC Chemistry 8 software, which utilizes a Gibbs free energy minimization algorithm. a C3H8 equilibrium conversion for CO2-ODHP and direct dehydrogenation of propane; b product amounts for CO2+C3H8 system and c conversions of propane, ethane, and methane dry reforming; all vs. temperature at 1 atm

The reactions of CO2 with propane may occur through two distinct pathways, oxidative dehydrogenation (CO2 + C3H8 → C3H6 + CO + H2O) and dry reforming (3CO2 + C3H8 → 6CO + 4H2). The two reactions should occur simultaneously at temperatures around 823 K and above with considerable conversions (Fig. 1b), allowing the formation of both dehydrogenation products (propylene) and reforming products (synthesis gas). The oxidative dehydrogenation of propane by CO2 (CO2-ODHP) can reach an equilibrium conversion of 33% as opposed to 17% for DDP at 823 K. At that same reaction temperature, as seen in Fig. 1c, CO2 equilibrium conversion for the dry reforming of propane (DRP) can reach up to 98% at a temperature 150 K less than that of methane dry reforming (DRM). This in turn would reduce catalyst deactivation due to coking and phase transformations triggered by the relatively high temperatures commonly used in DRM9,10. Furthermore, in the CO2+C3H8 system unreacted CO2 can remove surface carbon via the Boudouard reaction (CO2 + Cs → 2CO) at temperatures as low as 773 K with moderate rates11,12. Thus, it is of great interest to identify catalysts that can either selectively break the C–H bond to produce propylene or the C–C bonds to generate synthesis gas (CO + H2).

Previous work in CO2-ODH primarily focuses on supported chromium catalysts13,14,15 as a result of their ability to exist in multiple oxidation states16, but implementation is limited due to short lifecycles and high toxicity of chromium17. Ni is mainly used for dry reforming, but catalyst deactivation due to severe coking is still a problem18,19,20. To alleviate coke formation, precious metal catalysts (e.g., Rh, Re, Ru) have also been investigated on high surface area Al2O321,22. However, large scale catalytic conversion of CO2 into valuable products would require the development of cost effective, selective, and coking-resistant catalytic systems. While there are studies that examine the CO2-ODHP or DRP separately, a thorough examination utilizing supported bimetallic catalysts at a temperature range that allows both pathways to occur is still lacking. Ceria (CeO2) is a good choice of oxide support because it has the ability to store/release oxygen and thus may induce direct C–O bond scission of CO2, while also providing available lattice oxygen for coke suppression9,23,24,25.

The present work will explore ceria supported bimetallic catalysts, non-precious metal Fe3Ni as well as precious metal- based Fe3Pt and Ni3Pt, that are active at 823 K. In summary, steady-state flow reactor studies indicate that Fe3Ni shows promising selectivity toward propylene via the CO2-ODHP pathway, whereas Ni3Pt is active for the DRP with high selectivity toward CO. Density functional theory calculations of the energetics for the C–H and C–C bond scissions over the two catalysts are in agreement with experimental results.


Catalytic evaluation with kinetics and deactivation patterns

Flow reactor studies measuring both CO2-ODHP and DRP activity simultaneously are summarized in Table 1 along with CO chemisorption values. All catalysts were synthesized via incipient wetness impregnation of metals onto commercially obtained CeO2 (35–45 m2/g, Sigma Aldrich). For additional details see Methods section or Supplementary Methods section. Results for conversions and product selectivity following time on stream for all catalysts are shown in Supplementary Fig. 1. The monometallic Ni1 catalyst exhibits 12%–87% C3H6 and reforming selectivity, respectively, with minimal cracking products (CH4 and C2 hydrocarbons), while the Fe3 monometallic catalyst is not active for either reaction. The bimetallic system, Fe3Ni, however, at steady-state demonstrates propylene production from the CO2-ODHP reaction, corresponding to 58.2% C3H6 selectivity. The differences among the propylene yields on a C3H8 basis provided in Supplementary Table 1 of Fe3Ni (1.6% C3H6 yield) and the respective monometallics (C3H6 yield of 0.4% over Ni and 0.2% over Fe) indicate that there is a synergistic effect from the formation of the bimetallic Fe3Ni catalyst.

Table 1 Catalyst flow reactor results for CO2 + C3H8 reaction

Exchanging Ni in the Fe3Ni catalyst with precious metal Pt (Fe3Pt) roughly reduces the activity by half, decreases the selectivity toward C3H6 to 32%, and is unstable compared to Fe3Ni (Supplementary Fig. 2). The other precious metal bimetallic catalyst, Ni3Pt, primarily performs the DRP reaction with 39% CO2 conversion, a robust selectivity toward CO of 88% at comparable reactant conversions (Supplementary Table 2) and is more stable compared to monometallic Ni3 (Supplementary Fig. 3). Thus, when Ni is coupled with non-precious Fe at a ratio of 1:3, higher dehydrogenation activity can be achieved and propylene is produced. In contrast, when Ni is alloyed with precious metal Pt, reforming activity is enhanced compared to monometallic Ni3. Further analysis, such as the comparison of CeO2 supported Ni3Pt with Ni3Fe and Fe3Ni catalysts along with CO selectivity following CO2 conversion plots can be found in Supplementary Notes 1 and 2, respectively.

Kinetic studies examining the influence of the reactant partial pressure and the reaction temperature on the activity of Fe3Ni and Ni3Pt were conducted to further evaluate the differences between the two types of catalysts. The apparent activation energies were derived by measuring production rates in the temperature range of 803–843 K. Over Fe3Ni, the activation barrier for propane CO2 oxidative dehydrogenation was found to be 115 kJ mol−1, while the activation barrier for reforming over Ni3Pt was 119 kJ mol−1. Arrhenius-type plots and additional values are available in Supplementary Fig. 4 and Supplementary Table 3, respectively. As seen in Fig. 2a, the reactant consumption rate of C3H8 for the Fe3Ni CO2-ODHP catalyst was initially unaffected by increasing the partial pressure of CO2 but upon reaching a C3H8:CO2 ratio of 1:1, the rate started to decline. The reforming catalyst, on the other hand, was positively influenced by the partial pressure of CO2 until the aforementioned ratio of 1:6. Increasing the C3H8 partial pressure produced similar trends and are shown in Supplementary Fig. 5. The declining rates signify that there are less catalytic sites available for one reactant when the other is in excess, indicative of competitive adsorption of adsorbates and/or surface intermediates. Particularly, the rates for both reactants decrease at high propane partial pressure, suggesting that as the reaction progresses intermediates from propane block surface sites and lead to a loss in activity.

Fig. 2
figure 2

Effect of CO2 partial pressure on the propane production rate. Plots for a Fe3Ni and b Ni3Pt. Total system pressure is 1 atm

To further evaluate how different reaction pathways may influence deactivation patterns, both thermogravimetric (TGA) and energy dispersive spectroscopy (EDS) experiments were conducted and results are provided in Supplementary Figs 6 and 7, respectively. The TGA results indicate that the Fe3Ni catalyst only loses less than half a percent of its original mass, therefore, it is unlikely that the main deactivation pathway is due to coking. The EDS of the spent Fe3Ni sample shows small regions of higher Ni content, and to a lesser extent regions with higher Fe. However, in-situ XRD measurements do not reveal obvious agglomeration formation during reaction, and the absence of metal diffraction peaks suggests that the metal particles are most likely less than 2 nm in size (Supplementary Fig. 8). The Ni3Pt catalyst loses about 8% of its original mass but does not illustrate signs of sintering. However, the coking over the Ni3Pt catalyst at comparable propane conversion to Fe3Ni is not significant.

Oxidation states by in situ XANES

In situ X-ray absorption near edge spectroscopy (XANES) measurements were conducted in order to identify the local environment of the metals under reaction conditions, as shown in Fig. 3. Additional details are available in Supplementary Note 3. The XANES data identified that under reaction conditions the Ni3Pt catalyst consisted of metallic Pt (Supplementary Fig. 9) and that both the Fe3Ni and Ni3Pt catalysts consisted of metallic Ni (Fig. 3a). On the other hand, the Fe in the Fe3Ni catalyst was in the oxidized form. The extended X-ray absorption fine structure (EXAFS) fitting of Fe3Ni suggested the presence of an inserted oxygen through Fe–O–Fe as well as Fe–O bonds (Supplementary Table 4). Theofanidis et al. and Kim et al. studied DRM over higher loading Ni–Fe catalysts (8 wt.% Ni-5 wt.% Fe, and 8.8 wt.% Ni-2.1 wt.% Fe, respectively) supported on magnesium aluminate and they also observed oxidized Fe under in situ conditions but in an oxidation state of 2+26,27. For the Ni3Pt catalyst, the EXAFS fitting indicates that the coordination number of the Pt–Pt and Pt–Ni bonds is 3.4 and 6.4, respectively, confirming the formation of the Pt–Ni bimetallic bond.

Fig. 3
figure 3

In-situ XANES spectra. a Ni and b Fe K edges of all the bimetallic catalysts with respective references. The insets show more detailed comparison of Fe3Ni with model compounds

Reaction pathways and DFT calculations

Density functional theory (DFT) calculations were performed on bulk-terminated-Fe3Ni(111) and Pt-terminated-Ni3Pt(111) surfaces (Supplementary Fig. 10) to further gain insight into the potential reaction pathways for the oxidative C–H and C–C bond cleavage of propane to form *CH3CHCH2+H2O(g) and *CH3CH2+*CO+H2O(g), respectively. In these calculations, the surfaces are first modified by *O atoms assuming that *CO2 dissociates to form *CO + *O. The DFT optimized geometries in Supplementary Fig. 11 show that the intermediates *CH3CH2CH2O, *CH3CH2CHO, and *H2O interact with the surfaces via the oxygen atoms while other intermediates *CH3CH2CH2, *CH3CHCH2, *CH3CH2, and *CO interact with the surfaces via the carbon atoms. It is noted that, even though the binding configurations of intermediates are similar on both surfaces, all the intermediates bind more strongly on bulk-terminated-Fe3Ni(111) than on Pt-terminated-Ni3Pt(111) (Supplementary Tables 5 and 6). The DFT calculated binding energies were then used to calculate the change in energy for the oxidative C–H and C–C bond scission of propane. On bulk-terminated-Fe3Ni(111), Fig. 4a shows that the pathway for the oxidative C–H bond cleavage lies lower in energy than that for the C–C bond scission. In contrast, as shown in Fig. 4b on Pt-terminated-Ni3Pt(111), the pathway for the C–C bond cleavage lies lower in energy than that for the C–H bond.

Fig. 4
figure 4

DFT calculated energy profiles for the oxidative C–H and C–C bond scission pathways. a Bulk Fe3Ni(111) surface, b Pt-terminated Ni3Pt(111) surface, and c FeO/Ni(111) interface as well as the optimized geometries of d CH3CH2CH2O and e CH3CH2CH2 on FeO/Ni(111)

Overall the DFT results reveal that the C–C bond cleavage pathway is preferred on Ni3Pt(111), while bulk-terminated-Fe3Ni(111) favors the C–H bond cleavage pathway. Kinetically, this is also the case based on the comparison of activation energies (Supplementary Table 7). According to the DFT calculations, on Pt-terminated-Ni3Pt(111), the *O insertion reaction (*CH3CH2CH2 + *O → *CH3CH2CH2O + *) along the C–C bond cleavage pathway (∆E = −0.75 eV and Ea = 1.07 eV) is thermodynamically and kinetically more favorable than the oxidative dehydrogenation reaction (*CH3CH2CH2 + *O → *CH3CHCH2 + *OH) along the C–H bond cleavage pathway (∆E = −0.51 eV and Ea = 1.33 eV). In contrast, on bulk-terminated-Fe3Ni(111), the oxidative dehydrogenation reaction (∆E = 0.29 eV and Ea = 1.02 eV) is more favorable than the *O insertion reaction (∆E = 0.43 eV and Ea = 3.30 eV). These DFT predictions are in agreement with experimental observations, suggesting that the bulk-terminated-Fe3Ni(111) surface promotes the oxidative C–H bond cleavage of propane to form *CH3CHCH2 while the Pt-terminated-Ni3Pt(111) surface promotes the C-C bond cleavage of propane to form *CO.

To account for the potential FeO–Ni interfacial active sites based on the in situ experimental observation of oxidized Fe in the Fe3Ni catalyst, further DFT calculations were carried out to investigate the pathways for the oxidative C–H and C–C bond cleavage of propane on the FeO/Ni(111) interface. For the FeO x clusters supported on Ni(111), both Fe6O9 and Fe3O3 clusters on three layer 7 × 7 Ni(111) and 5 × 5 Ni(111) surfaces (Supplementary Fig. 12) were considered. The oxygenated species (*O, *CO, *CH3CH2CH2O, *CH3CH2CHO, and *CH3CH2CO) prefer to adsorb at the interfacial sites while *C x H y species (*CH3CH2CH2, *CH3CHCH2, and *CH3CH2) most favorably adsorb on Ni(111) sites (Supplementary Table 8 and Supplementary Fig. 13). The energy diagram in Fig. 4c, calculated based on the DFT obtained binding energies of the potential intermediates, show that the first steps in oxidative C–C and C–H bond cleavage pathways are competitive. The subsequent step to form *CH3CHCH2 is downhill in energy along the oxidative C–H bond cleavage pathway; in contrast, the subsequent steps are uphill in energy along the oxidative C–C bond cleavage pathway. Again, such thermodynamic predictions are fully supported by the calculated Ea, showing that the oxidative dehydrogenation reaction (∆E = −0.40 eV and Ea = 0.29 eV) is highly favorable over the *O insertion reaction (∆E = 0.01 eV and Ea = 2.13 eV) on the Fe3O3/Ni(111) surface. This indicates that the oxidative dehydrogenation pathway should be more favorable than reforming, consistent with experimental observation.

Finally, on the three surfaces studied the desorption of *CO is expected to be a facile process due to the contribution of entropy at 823 K. *C2H5 is one of the reaction intermediates that undergoes O-insertion, C–H and C–C bond scission reactions to eventually produce CO and H2. The *O species on Pt-terminated-Ni3Pt(111) react with *C x H y to form the *C x H y O intermediate, which promotes the C–C bond scission. In contrast, the more stable *O on bulk-terminated-Fe3Ni(111) and the FeO/Ni(111) interface are expected to remain on the surface, which facilitates the selective C–H bond scission of propane to produce propylene.


Overall, the oxidative dehydrogenation of propane with CO2 has the potential to combine two underutilized4,5,6 reactants to produce propylene or syngas. Two types of bimetallic catalysts have been identified for the CO2 + C3H8 system. The DFT calculation results indicate that the bulk Fe3Ni(111) surface and the FeO/Ni(111) interface should favor C–H bond scission for the CO2-ODHP pathway, whereas the Pt-terminated Ni3Pt(111) surface should favor the C–C bond cleavage for the DRP pathway. Flow reactor results are consistent with the DFT calculations as it was observed that the Fe3Ni catalyst is selective for propylene production, while the Ni3Pt catalyst shows good activity and CO selectivity. The oxidation states of the different metals provided by in situ XANES measurements reveal that Fe3Ni consists of oxidized Fe and metallic Ni. Future efforts should be geared toward enhancing propylene yield through the discovery of more stable and selective catalytic materials.


Density functional theory calculations

Spin polarized28,29 density functional theory (DFT) calculations were performed as an attempt to elucidate the possible pathways of C–C and C–H bond cleavage of propane over Fe3Ni(111), Ni3Pt(111) surfaces, and FeO/Ni(111) interface using the Vienna Ab Initio Simulation Package (VASP) code30,31. Projector augmented wave potentials were used to describe the core electrons with the generalized gradient approximation (GGA)32,33 using PW91 functionals34. The Kohn−Sham one-electron wave functions were expanded by using a plane wave basis set with a kinetic energy cutoff of 400 eV. The Brillouin zone was sampled using a 3 × 3 × 1 k-point grid in the Monkhorst−Pack scheme35. Ionic positions were optimized until Hellman–Feynman force on each ion was smaller than 0.02 eV/Å. The transition state of a chemical reaction was located using the climbing image nudged elastic band (CI-NEB) method implemented in VASP36. The activation energy (Ea) of a chemical reaction is defined as the energy difference between the initial and transition states while the reaction energy (ΔE) is defined as the energy difference between the initial and final states.

Catalyst preparation and flow reactor studies

The catalysts evaluated in this study were synthesized through incipient wetness impregnation of metals onto commercially obtained CeO2 (35–45 m2/g, Sigma-Aldrich). Flow reactor experiments were performed under atmospheric pressure utilizing a 1/4” quartz U-shaped reactor. All catalysts were reduced at 723 K for 1 h under a 1:1 H2/Ar flow (40 mL/min total). Subsequently, the temperature was increased and held at 823 K in the presence of 1:1:2 CO2, C3H8, and Ar for 12 h. Apparent activation barrier and reaction order experiments were conducted at slightly different reaction conditions to ensure operation in a true intrinsic kinetic regime and minimize transport effects. XANES measurements were conducted using a custom in situ micro-channel cell holding ~200 mg of catalyst (60–80 mesh) and a 4-channel vortex fluorescence detector.

Data availability

The data that support the findings of this study are available from the corresponding author upon request.