Identification of earth-abundant materials for selective dehydrogenation of light alkanes to olefins

Significance

Catalytic alkane dehydrogenation is of considerable importance in the synthesis of olefins industrially. However, discovery of highly active, selective, and stable heterogeneous catalysts to replace noble metal Pt remains challenging. By combining descriptor-based microkinetic modeling, high-throughput computations, machine-learning concepts, and experiments, we efficiently evaluated 1,998 bimetallic alloys and successfully identified Ni₃Mo as one of the most promising catalysts in selective ethane dehydrogenation. This work will open new possibilities of using earth-abundant materials as catalysts for essential heterogeneous catalytic reactions.

Abstract

Selective ethane dehydrogenation (EDH) is an attractive on-purpose strategy for industrial ethylene production. Design of an effective, stable, and earth-abundant catalyst to replace noble metal Pt is the main obstacle for its large-scale application. Herein, we report an experimentally validated theoretical framework to discover promising catalysts for EDH, which combines descriptor-based microkinetic modeling, high-throughput computations, machine-learning concepts, and experiments. Our approach efficiently evaluates 1,998 bimetallic alloys by using accurately calculated C and CH₃ adsorption energies and identifies a small number of new promising noble-metal–free catalysts for selective EDH. A Ni₃Mo alloy predicted to be promising is successfully synthesized, and experimentally proven to outperform Pt in selective ethylene production from EDH, representing an important contribution to the improvement of light alkane dehydrogenation to olefins. These results will provide essential additions in the discovery and application of earth-abundant materials in catalysis.

As an important feedstock in the petrochemical industry, ethylene has essential applications in the production of plastics, ethylene oxide, ethylene dichloride, and ethylbenzene (1). However, the current worldwide capacity cannot fulfill the increasing market demand of estimated 200 million tons per year (2). The conventional process, high-temperature alkane steam cracking (SC), still dominates the ethylene production despite of its low selectivity and high energy intensity. Nevertheless, due to the technological development in fracking, the large availability of cheap ethane from shale gas has provided an opportunity for on-purpose ethylene production such as ethane dehydrogenation (EDH) with the advantages of high selectivity and energy efficiency. (3,4) However, the key challenge of this process came from the common hurdle in heterogeneous catalysis, that is, design of active, selective, stable, and earth-abundant catalysts. Currently, Pt alloyed with elemental Sn is regarded as the most promising catalyst with all above merits except low cost (5).

Experimentally, systematic investigations have been reported to illustrate the role of Sn in promoting the performance of Pt and discover alternative catalysts. For example, an optimum Pt:Sn ratio of 3:1 with best ethylene selectivity was identified by Bell and coworkers (6). Other precious alloys like PdIn (7), PtIn (8, 9), PtIr (10), PtGa (11), and PtZn (12) were also reported to have improved EDH performance. Iron (13, 14), gallium (15), and molybdenum (16) supported on Zeolite Socony Mobil–5 (ZSM-5) were found to be good for EDH, where the substitution of conventional γ-Al₂O₃ support with zeolite was essential. In contrast, theoretical simulations of EDH reaction mechanisms are relatively scarce. Chen and Vlachos (17) identified the most stable adsorbates and favorable C-C scission pathways of ethane dehydrogenation and hydrogenolysis on Pt catalyst. Similar computations were performed on Ru(0001) surface by Ande et al. (18). Hook et al. (19) theoretically found that Sn weakens the adsorption strength of adsorbates while increasing C-C scission barriers. A kinetics study into the EDH reaction mechanisms at experimental relevant conditions was recently reported but failed to provide useful suggestions for further experimental EDH catalyst development (20). Computational and experimental investigations into propane dehydrogenation also provided useful information for our understandings of light alkane dehydrogenation mechanisms (21–25).

Despite the wide fundamental investigations into EDH, a large-scale industrial application competitive with SC for ethylene production is still challenging, partially due to the scarcity of Pt as a catalyst component. Computational studies of the mechanism on a case-by-case basis and experimental “trial-and-error” strategies for catalyst design could help identify and evaluate each catalyst but they lack the necessary efficiency. Therefore, more efficient catalyst screening methods are important and currently the microkinetic analysis of general catalytic trends across various catalysts has proven to be both accurate and effective (26, 27). This framework has been successfully applied to efficiently design metallic catalysts for essential reactions (28–31). In a recent review, the importance of expanding catalysis with earth-abundant elements was reported (32), again emphasizing the need for an effective and reliable methodology that can accelerate the discovery of promising earth-abundant EDH catalysts.

In this paper, we introduce a decision map based on a specific acquisition function, which efficiently evaluated 1,998 bimetallic alloys using accurately calculated C and CH₃ adsorption energies and identified a number of earth-abundant candidates with promising performances for EDH as potential alternatives to Pt. One of these predicted candidates (NiMo) was successfully synthesized and experimentally verified to have better activity, selectivity, stability, and much lower cost than Pt in EDH.

Results and Discussion

Reaction Networks for EDH.

In SI Appendix, Fig. S1, the reaction mechanism for the dehydrogenation of ethane is summarized, including molecular adsorption/desorption, C–H and C–C bond-breaking, as well as isomerization. For example, dehydrogenation of ethane (C₂H₆) will generate ethyl (CH₃CH₂) followed by two possible C-H scission routes to either the desired product ethylene (CH₂CH₂) or ethylidene CH₃CH. The isomerization of CH₃CH could also produce ethylene but usually with high energy barriers. Further dehydrogenation of these species will form the undesired side-product acetylene, while the C-C scission opens a channel for methane production and, depending on the carbon resistance, can result in coking and deactivation of the catalyst. Therefore, an ideal EDH catalyst should be able to selectively break the C–H bonds in ethane to produce ethylene while avoiding the deep dehydrogenation and C–C bond scission. In this work, a systematic density-functional theory (DFT) study of this complex reaction network on the close-packed surface of Cu, Ag, Au, Ni, Pt, Pd, Co, Rh, Ir, Ru, Re, Os, and Pt₃Sn catalysts is reported. The acquired energetic details and cross-correlations between data served as inputs into a microkinetic model enabling the simulation of production rates and selectivity.

Activity and Selectivity Volcanos.

A microkinetic simulation at 873 K, 0.2 bar ethane, C₂H₆/H₂ = 1/1 and 0.1% ethane conversion was performed. The computational much more facile C and CH₃ adsorption energies were chosen as descriptors to plot the activity volcano maps of different products, while the coverage maps of different species are shown in SI Appendix, Fig. S2. Fig. 1 shows the general activity trends of ethylene (A), acetylene (B), and methane (C) productions on different transition metals. Clearly, Pt is on the plateau of both the ethylene and the methane volcanos indicating its ability to produce ethylene at high rates and at the same time pinpoints its higher risk of coke formation due to favorable formation of CH_x species. This agrees well with available knowledge of Pt catalysts, i.e., high ethylene activity, but low coke resistance (6). Similarly, other reactive metals such as Ir, Ru, Rh, Re, and Os have similar issues as Pt with higher selectivity toward methane and the risk of coke formation. The coinage metals Cu, Ag, and Au all have high ethylene selectivity, but their practical application is curbed by the low activity as EDH catalysts. In addition to the pure metals, we also put the commercialized Pt₃Sn catalyst on the activity maps. It reveals that Pt₃Sn is near the plateau of the ethylene activity volcano in Fig. 1A but still far away from the top of the methane activity volcano in Fig. 1C, and this results in an improved ethylene selectivity and coke resistance. This simple mapping of activities clearly identifies the role of Sn in improving the coke resistance of elemental Pt and why this alloy is the material of choice for the selective production of ethylene from ethane in industry. This demonstrates that an analysis based on descriptors and microkinetic modeling can reasonably describe the general activity and selectivity trends of metallic catalysts, which provides a good foundation for further catalyst design.

Fig. 1.

Activity volcano plots and decision map for EDH. TOF maps for C₂H₄ (A), C₂H₂ (B), and CH₄ (C), and decision map for identifying promising EDH catalyst (D) as a function of C (E_C*) and CH₃ (E_CH3*) adsorption energies at 873 K, 0.2 bar ethane, C₂H₆/H₂ = 1/1, and 0.1% ethane conversion. The numbers 1 and 0 in decision map indicate promising and less promising catalyst candidates for EDH, respectively.

Decision Map and Acquisition Function.

With the activity volcano plots in Fig. 1 A–C, EDH catalyst screening could simply be done by calculating C and CH₃ adsorption energies on metallic alloys. However, the evaluation of alloys by putting them on several different volcano maps will become complicated and tedious especially when the screening space includes hundreds or even thousands of candidates. To this end, we applied statistical analysis to aid our evaluation and screening of catalysts. Based on the available knowledge of EDH as well as the information in Fig. 1, neither Pt nor Pt₃Sn is perfect, i.e., Pt is limited by the selectivity as well as low coke resistance despite of the high activity, while Pt₃Sn is limited by the activity despite of the high selectivity. Therefore, an ideal catalyst for ethane dehydrogenation would possess comparable or even better activity than Pt as well as higher selectivity than Pt₃Sn. In reality, the volcano maps in Fig. 1 are plotted based on a coarse grid (SI Appendix, Fig. S3) and then interpolated during the evaluation of the kinetic model using CatMAP. In principle, each point in SI Appendix, Fig. S3 denotes a catalyst with a given C and CH₃ adsorption energy. To speed up materials search, we performed further analysis and attributed each point with a score based on a well-defined acquisition function inspired by machine-learning concepts as

$$\text{Acquisition Function:}\{\begin{array}{c}1:\hspace{0.17em}If\hspace{0.17em}TOF\ge TOF\left(P{t}_{3}Sn\right)\hspace{0.17em}and\hspace{0.17em}Selectivity\hspace{0.17em}Selectivity\left(Pt\right)\\ \\ 0:\hspace{0.17em}If\hspace{0.17em}TOF<\hspace{0.17em}TOF\left(P{t}_{3}Sn\right)\hspace{0.17em}and\hspace{0.17em}Selectivity<\hspace{0.17em}Selectivity\left(Pt\right)\end{array}$$

Based on the above acquisition function, if a catalyst at a given point in descriptor space (E_C*, E_CH3*) has comparable turnover frequency (TOF) to Pt₃Sn as well as comparable ethylene selectivity to Pt, it receives a score of 1, otherwise the score is 0. This process combines all the TOF maps in Fig. 1 A–C into a single volcano plot––a decision map as shown in Fig. 1D. The orange part of the map defines the region where promising candidates for catalyzing ethane selective dehydrogenation to ethylene can be found.

High-Throughput Computations and Screening.

With the decision map in Fig. 1D, metallic catalyst with calculated C and CH₃ adsorption energies can be rapidly evaluated for their activity and selectivity toward ethylene. At relatively low computational costs, we performed high-throughput DFT computations of C and CH₃ adsorption energies on closed-packed surface of 1,998 different ordered bimetallic alloys in the L1₀ and L1₂ structure with AB and AB3 composition, respectively. We combined 37 different elemental metals as shown in Fig. 2, which resulted in roughly 17,105 different types of active sites and 34,210 DFT simulations. The bulk structures of these alloys were acquired from the open-source Catalysis-hub database (33). To evaluate each alloy based on their C and CH₃ adsorption energies in a smarter way, the decision map was further transferred into a probability map with the help of machine-learning strategies. For example, the decision map in Fig. 1D was split into numerous points as shown in SI Appendix, Fig. S4A, where each point represents a metallic catalyst with a certain probability of being a promising candidate for EDH. The candidates that fall within the orange region are immediately selected as promising, whereas the candidates falling in the blue region are rejected. In this way, the decision map can be treated as a classification scheme in machine learning, categorizing the datapoints (E_C*, E_CH3*) into one of the two classes 1 or 0. We can now assign each point of the decision map a color based on its probability (0–1) as shown in SI Appendix, Fig. S4B. The K-Nearest Neighbors classifier as implemented in the Scikit-learn python library (34) was applied.

Fig. 2.

Theory-driven framework for EDH catalyst screening: The procedure of forming 1,998 bimetallic alloys by combining 37 metals in periodic table, the evaluation of each alloy with C/CH₃ binding energies using the decision map, as well as the price and stability checks of the alloys.

With the help of our developed framework, one can easily acquire the probability of a given intermetallic or alloy being a promising catalyst candidate for EDH by just entering the C and CH₃ adsorption energies into the code. This approach efficiently identifies 166 out of 1,998 candidates as promising. To further reduce the candidate space, the stability and price of the elements are taken into consideration as shown in Fig. 2. Firstly, we inquired the inorganic crystal structure database (ICSD) (35), which contains the crystal information of all available experimentally synthesized materials. We hypothesized that any candidate found in our screening with a corresponding crystal structure in the ICSD will have high stability in nature. Secondly, a price check then ruled out all the precious elements, which eventually reduced the initial search space from 1,998 to a limited set of promising earth-abundant alloys with specific crystal information in the ICSD. The theoretical framework developed in this work suggests that these candidates should display high selectivity as well as high activity. However, further state-of-the-art experimental verification is needed.

Reaction Mechanism Simulations on Predicted Candidates.

Based on our established framework, each bimetallic catalyst can be evaluated based on simple calculations of the C and CH₃ adsorption energies. To check the reliability of our strategy, we performed detailed calculations of energies for all reaction intermediates and transitions states between them in the EDH on three alloys falling into different regions on the decision map in Fig. 1D. Ni₃Mo falls inside the orange region, Ni₃Cr near the boundary, and NiAl₃ is outside the orange region of high activity and selectivity. SI Appendix, Fig. S5 summarizes the calculated potential free-energy diagrams for the most important elementary steps in ethane dehydrogenation to ethylene and acetylene on the (111) surface of Pt, Ni₃Mo, Ni₃Cr, and NiAl₃ catalysts at 873 K. It clearly shows that the Ni₃Mo catalyst has lower energy barriers for each important elementary step compared with Pt, thus indicating its potential better performance as an EDH catalyst. Ni₃Cr located on the boundary of the active region in Fig. 1D has similar reaction energetics as Pt and therefore should have comparable performance. Finally, the NiAl₃ catalyst located outside the active region shows much higher reaction energy barriers for each elementary step indicating a worsening in EDH performance relative to Pt. Based on the data, we further established a microkinetic model and calculated the TOF of different products as well as surface coverages of adsorbates on Ni₃Cr(111) and Ni₃Mo(111) surfaces, summarized in SI Appendix, Tables S1 and S2. Our theoretical modeling identifies Ni₃Mo to have both high activity and selectivity for C₂H₄, whereas Ni₃Cr shows very limited activity and selectivity. To provide a rationale for future refining and design of promising catalyst active site for EDH, we further analyzed the differences of Ni₃Mo and Ni₃Cr. It is known that CH₃ binding energy is linearly correlated with C binding energies on transition metals (the black dashed line in SI Appendix, Fig. S6), that is, the scaling relation (36). As shown in SI Appendix, Fig. S6, promising EDH catalysts in the orange region need to have weak C binding but strong CH₃ binding, which clearly needs to break the scaling relation between C and CH₃. We identify that Ni₃Cr follows the C/CH₃ scaling relation very well because both C (SI Appendix, Fig. S6B) and CH₃ (SI Appendix, Fig. S6B) form most stable adsorptions on threefold hollow sites. However, on Ni₃Mo(111) surface, C forms most stable adsorption on the threefold hollow site but CH₃ on Mo top site, which provide an ideal scenario for breaking away from the traditional scaling relation. In brief, an ideal EDH catalyst needs to have weak C binding strength and simultaneous strong CH₃ binding strength, which is clearly an outlier on the C/CH₃ scaling line.

Experimental Synthesis and Test of Ni₃Mo Catalyst.

To further evaluate the reliability of our theoretical predictions, a Ni-Mo catalyst on MgO support (Ni₃Mo/MgO) was prepared and tested experimentally for ethane selective dehydrogenation. Fig. 3 A and B summarize the conversion of ethane to ethylene and the selectivity to ethylene on Ni₃Mo/MgO as a function of weight hourly space velocity (WHSV). The applied reaction conditions were as follows: 873 K, atmospheric pressure, an ethane inlet concentration of 16.7%, a total gas flow rate of 24 mL/min balanced by Ar, and WHSV was tuned by varying the amount of the catalyst. As shown in Fig. 3 A and B, the initial conversion of ethane decreases with increasing ethane WHSV in the presence of Ni₃Mo/MgO, whereas the initial selectivity toward ethylene shows an increasing trend.

Fig. 3.

The EDH performances of Pt/MgO and Ni₃Mo/MgO catalysts: The ethane (C₂H₆) conversion (A) and ethylene (C₂H₄) selectivity (B) vs. WHSV in the presence of Ni₃Mo/MgO. Reaction conditions: atmospheric pressure, 873 K, balance Ar for the total flow rate of 24 mL·min⁻¹; The ethane (C₂H₆) conversion (C) and ethylene (C₂H₄) selectivity (D) as a function of time during EDH in the presence of Pt/MgO and Ni₃Mo/MgO catalysts. Reaction conditions: atmospheric pressure, 873 K, C₂H₆ = 4 mL·min⁻¹, WHSV = 3.21 h⁻¹, Ar = 20 mL·min⁻¹, and 100 mg of catalyst.

For comparison, we studied a Pt/MgO (a content of 0.82 and 0.80 wt % of fresh and used Pt sample was identified by inductively coupled plasma atomic emission spectroscopy, ICP-AES, respectively) catalyst for the ethane to ethylene process. As shown in Fig. 3C, both catalysts show a slight loss in conversion during the first few hours of time on stream but then reach a nearly constant conversion. Interestingly, the Ni₃Mo/MgO catalyst has a constant ethane conversion of 1.2%, which is a factor of 3 higher than that for Pt/MgO (0.4%) under the identical reaction conditions. As shown in Fig. 3D, the initial ethylene selectivity for Ni₃Mo/MgO is 66.4% and then increases to 81.2% after 12 h, whereas that of Pt/MgO is 75.2% initially and then slightly rises to 79.3%. On both catalysts, methane was observed as the major side product from EDH. Moreover, under the same reaction conditions, the ethane conversion was low on Pt/MgO while higher conversion on Ni₃Mo/MgO was obtained after 12 h. Most importantly, no obvious deactivation was observed for Ni₃Mo/MgO during the EDH reaction, demonstrating its high stability. These results clearly demonstrate the better catalytic performance of Ni₃Mo/MgO catalyst than Pt/MgO in ethane dehydrogenation, which agrees very well with our theoretical predications.

To gain more insights into the structure of the catalyst, the as-prepared and used Ni₃Mo/MgO and Pt/MgO (SI Appendix, Fig. S7) catalysts were carefully characterized. ICP-AES showed that the contents of Ni and Mo for fresh Ni₃Mo/MgO were 4.11 and 1.85 wt %, respectively (corresponding to a Ni/Mo mole ratio of 3/1), whereas those for used were 3.95 and 2.04 wt %, demonstrating the slight change of these two catalysts. The transmission electron microscopy (TEM), high-resolution TEM, and TEM energy-dispersive X-ray spectroscopy (EDS) maps for as-prepared Ni₃Mo/MgO are shown in Fig. 4. The TEM characterization in SI Appendix, Fig. S8 showed that the NiMo nanoparticles (NPs) dispersed well on the MgO surface with a size distribution of 20–30 nm, and the HAADF (high-angle annular dark-field)-STEM (scanning transmissionelectronmicroscopy)-EDS mapping for the Ni₃Mo/MgO catalyst in Fig. 4 A–C corroborates the formation of a Ni-Mo alloy. The HAADF-STEM cross-line profile of the sample is shown in Fig. 4 D and E (corresponding to the red line). The overlap of the Ni and Mo profiles proved the coexistences of both metals in the nanoparticle and was suggestive of alloy formation. Moreover, no X-ray powder diffraction (XRD) peaks for isolated Ni and Mo nanoparticles were detected after EDH reaction indicating the high stability of the NiMo alloy during operation. The electronic structures of the supported NPs were probed with X-ray photoelectron spectroscopy (XPS) as shown in Fig. 4 G and H. In the fresh Ni₃Mo/MgO catalyst, the peaks at 855.8 eV are attributed to Ni 2p and the peaks at 232.6 and 335.7 eV correspond to Mo 3d, which confirms the presence of metallic Ni²⁺ and Mo⁶⁺ in the alloy structure. The binding energies (B.E.) of Ni 2p in the used Ni₃Mo/MgO shift to lower values compared to the Ni 2p B.E. in fresh Ni₃Mo/MgO, while the Mo 3d in used Ni₃Mo/MgO moves to higher values compared with the Mo 3d B.E. in fresh Ni₃Mo/MgO. Such changes indicate the alloying of Ni-Mo NPs during the reaction and the electronic variations of Ni with Mo. To quantify coke formation, thermogravimetric analysis was conducted, where Ni₃Mo/MgO showed a coke amount of < 8.0 wt % after 12 h (SI Appendix, Fig. S9). Furthermore, the properties of the deposited carbon on the catalyst were investigated by Raman spectroscopy. As shown in SI Appendix, Fig. S10, there are two peaks, i.e., the D band represents the amorphous carbon and the G band belongs to ordered graphitic carbon. XRD pattern of the Ni₃Mo/MgO catalysts in Fig. 4I also exhibits crystalline features corresponding to Ni-Mo alloy. Moreover, both the XPS and XRD patterns of the used Ni₃Mo/MgO shown in SI Appendix, Figs. S11–S12 are similar to that of the fresh catalyst, further corroborating the stability of the bimetallic catalyst during EDH.

Fig. 4.

Characterizations of the Ni₃Mo/MgO after EDH at 873 K for 12 h: HAADF image (A); TEM-EDS mapping images of Ni (B) and Mo (C); TEM-EDX line concentration profiles of Ni (D) and Mo (E); Elemental composition analysis (F); XPS of (G) Ni 2p and (H) Mo 3d spectra of the fresh and used Ni₃Mo/MgO; (I) XRD patterns of the fresh and used Ni₃Mo/MgO catalysts.

Clearly, the discovered Ni₃Mo/MgO catalyst outperforms Pt/MgO in activity, selectivity, and stability for EDH at the same reaction condition, which strongly proves the reliability of our theory-based framework in correctly and efficiently identifying promising earth-abundant bimetallic catalysts for this reaction. Furthermore, the NiMo catalyst is economically more viable for practical industrial application compared with Pt considering the huge difference in cost, i.e., Pt (27 $/g) VS Ni (0.025 $/g)/Mo (0.016 $/g). We believe that there are still ample spaces to further optimize this NiMo/MgO catalyst experimentally for its application in different alkane dehydrogenations. We are also confident that the other theoretically predicted promising candidates will trigger further state-of-the-art experimental efforts to engineer them, representing essential contributions to the development of industrial alkane transformations with on-purpose techniques.

Conclusion

We have developed and experimentally validated a theoretical framework for the efficient identification of promising bimetallic catalysts for EDH. A descriptor-based microkinetic model was established based on systematic DFT simulations of ethane dehydrogenation mechanism on a group of transition metals, where the activity and selectivity of different metallic catalysts were plotted as a function of C and CH₃ adsorption energies. A decision map was designed using a well-defined acquisition function, denoting a straightforward and convenient tool for efficient catalyst discovery. Combined with high-throughput computations of C/CH₃ adsorption energies on 1,998 different bimetallic catalysts (34,210 DFT calculations), this framework efficiently identified 166 promising candidates with good EDH performance, where only few of them were found to be stable and comprised earth-abundant materials. The proof-of-concept simulations of detailed EDH reaction mechanisms on three different catalysts reveal that the theoretical predictions using only C and CH₃ adsorption energies can properly identify interesting catalysts. We successfully synthesized one of the most promising candidates experimentally, Ni₃Mo, and found that it outperforms Pt as a promising EDH catalyst. We envision that the framework introduced herein will dramatically accelerate the discovery of earth-abundant materials as catalysts for selective alkane dehydrogenation to olefins, thus representing an important contribution to the material science community and heterogeneous catalysis in general.

Materials and Methods

Computational Details.

All computations were carried out with periodic plane-wave-based DFT method as implemented in QUANTUM ESPRESSO code (37) with pseudopotentials GBRV version 1.5 (38). The energy cutoffs for plane wave and electron density were set to be 500 and 5,000 eV, respectively. BEEF-vdW (Bayesian Error Estimation Functional - van der Waals) functional was used to describe the exchange-correlation contribution to the electronic energy. The close-packed surfaces of alloys were simulated using four-layer (2 × 2) supercells with relaxed topmost two layers and constrained bottom two layers. The (4 × 4 × 1) Monkhorst–Pack k-point grids (39) were applied for sampling. Structure optimizations were done when forces became smaller than 0.05 eV/Å and the energy difference was lower than 10⁻⁵ eV. A vacuum layer of 12 Å was set between periodically repeated slabs. Transition-state geometries were identified with the climbing-image nudged elastic band method (40, 41). Spin polarization was included for Fe, Co, Ni, and Mn systems to correctly describe magnetic properties. The formation energies (∆E) of all the species are calculated with references to gaseous CH₄ and H₂ as ∆E_CxHy = E(C_xH_y) – E(slab) – x(E_CH4 – 2E_H2) – y E_H2/2, where E(C_xH_y), E(slab), E_CH4, and E_H2 denote the electronic energies of the surface slab with adsorbates, clean surface slab, gas phase CH₄ and H₂ molecules, respectively. The Gibbs free energy (∆G) including enthalpy and entropic contributions at different temperatures is calculated within the harmonic approximation for surface species and the ideal gas approximation for gas-phase species. Our DFT-calculated enthalpy change of C₂H₆(g) → C₂H₄(g) + H₂(g) in gas phase within ideal gas approximation at 873 K is 1.44 eV, in good agreement with the experimental result of 1.48 eV from CRC Handbook and National Institute of Standards and Technology (NIST) (42). This indicates that our methods could reasonably describe the thermodynamics of this reaction. The related energies in this work are summarized in Dataset S1, while the structures and coordinates of all the adsorbates and TS on different catalyst surfaces are available in the open electronic structure database for surface reactions https://www.catalysis-hub.org. The scripts and data for plotting volcanos in Fig. 1 are included in Dataset S2.

Experimental details.

Catalyst preparation.

The Ni₃Mo/MgO catalyst was prepared by incipient wetness coimpregnation method. Typically, (NH₄)₆Mo₇O₂₄·4H₂O (177 mg) and Ni(NO₃)₂·6H₂O (874.8 mg) were used as precursors and dissolved in water and MgO (5,000 mg) was used as support. After impregnation, the catalysts were placed in the atmosphere statically overnight and then dried in static air at 353 K for 12 h. Afterward, the catalyst precursor was calcined at 1073 K at a rate of 10 K min⁻¹ and retained at 1073 K for 4 h. The Pt/MgO was prepared using the similar method replacing Mo and Ni aqueous by H₂PtCl₆·6H₂O (100 mg) solution.

Catalyst characterization.

TEM was performed on an FEI Talos instrument operated at 200 kV high tension. Energy-dispersive X-ray spectroscopy (EDX) mapping was used for elemental characterization. ICP-AES was used to determine the content of Ni, Mo, and Pt in all samples.

Catalytic performance measurements.

Catalytic tests were performed in a fix-bed reactor with 8-mm inner diameter and 45-cm length under atmospheric pressure. Typically, 500 mg of calcined sample with the particle size of 20–40 mesh was packed between quartz wool plugs. The inlet gas flow rates were tested by mass-flow controllers. The sample was first heated to 873 K at a rate of 10 K min⁻¹ and retained at 873 K for 2 h in flowing 10 vol % H₂/Ar. Next, a mixture gas of C₂H₆ and Ar with a volume rate of 4:20 was fed at a rate of 24 mL min⁻¹. The product gas was analyzed by an online gas chromatography (GC) equipped with a flame ionization detector (HP-AL/S) and a thermal conductivity detector (Al₂O₃ Plot column).

The conversion of ethane and the selectivity to product were determined from the following equations:

$$\text{Conversion \%}=100\text{×}\left({\left[{\text{F}}_{\text{C}2\text{H}6}\right]}_{\text{inlet}}-{\left[{\text{F}}_{\text{C}2\text{H}6}\right]}_{\text{outlet}}\right)\text{/}{\left[{\text{F}}_{\text{C}2\text{H}6}\right]}_{\text{inlet}},$$

$$\text{Selectivity\%}=100\text{×}{\left[{\text{F}}_{\text{C}2\text{H}4}\right]}_{\text{outlet}}\text{/}\left({\left[{\text{F}}_{\text{C}2\text{H}6}\right]}_{\text{inlet}}-{\left[{\text{F}}_{\text{C}2\text{H}6}\right]}_{\text{outlet}}\right).$$

Data and Code Availability

All study data are included in the article and/or Supporting Information. The code used to perform the microkinetics modeling this work is available at GitHub, https://github.com/SUNCAT-Center/catmap.