Controlling the Coke Formation in Dehydrogenation of Propane by Adding Nickel to Supported Gallium Oxide

Atomic layer deposition was applied on mesoporous silica to synthesize a highly dispersed gallium oxide catalyst. This system was used as starting material to investigate different loadings of nickel in the dehydrogenation of propane under industrially relevant, Oleflex‐like conditions. The formation of NiGa alloys was confirmed by X‐ray diffraction analysis and electron microscopy. Surprisingly, the nanoalloys enhanced the selectivity towards C3H6 while decreasing the tendency for coking. Herein, in situ thermogravimetry, and measured mass fractions of carbon revealed that the coking rate was reduced by over 50 % compared to the pristine gallium oxide. Generally, the increased selectivity can be explained by the partial hydrogenation and reduction of the gallium oxide surface. The optimum temperature for the removal of deposited carbon was evaluated by a temperature programmed oxidation. Finally, the best‐performing Ni−GaOx catalyst was employed in a cycled experiment with periodic reaction and regeneration tests. After regeneration, the selected Ni−GaOx catalyst provided a higher yield of propylene compared to the unmodified gallium oxide.


Introduction
Propylene is an essential intermediate in the chemical industry and is mainly used for the production of polypropylene, acrylonitrile, and propylene oxide.Most propylene is obtained as a byproduct from steam cracking and fluid catalytic cracking processes which in recent times are unable to satisfy the everincreasing demand alone. [1]Therefore, on-purpose technologies emerged targeting the production of propylene from abundant shale gas. [2]The non-oxidative propane dehydrogenation (PDH) is of particular interest as it generates hydrogen as a valuable byproduct. [3]The most prominent commercial techniques for the dehydrogenation of light alkanes are the Catofin (CB&I, ABB Lummus), Oleflex (UOP Honeywell), and STAR process (Thys-senKrupp Uhde). [4] The dehydrogenation of propane is a highly endothermic reaction, limited by its thermodynamic equilibrium [Equation (1)].Hence, elevated reaction temperatures are required (up to 650 °C) to enable industrially relevant propane conversions. [1,4]However, high temperatures diminish the chemo-selectivity to propylene and favor unintended cracking or deep dehydrogenation. [5,6]Both can eventually result in the formation of coke which accelerates the deactivation of the catalyst. [6,7]s the deposition of coke cannot be fully avoided, industrial catalysts are periodically regenerated by burning the coke in air. [3,8][11][12] Consequently, restrained coke formation and stability under regeneration conditions are important requirements for the catalyst.The Catofin process employs supported CrO x catalysts whereas the Oleflex and STAR processes apply different bimetallic PtÀ Sn systems (Oleflex: KÀ PtÀ Sn/Al 2 O 3 , STAR: PtÀ Sn/ZnAl 2 O 4 /CaO-Al 2 O 3 ). [4]The use of a CrO x catalyst requires frequent regeneration (every 15-25 min) and is environmentally questionable due to its toxicity. [13]herefore, Pt-based catalysts are preferable, providing a propylene selectivity of around 90 % and reaching operation periods of up to 10 days. [3,4]Particularly PtÀ Ga alloys have been reported with high propylene selectivity (around 95 %), propane conversion (up to 40 %), and inhibited coke formation. [14,15]evertheless, noble metals should be replaced by more abundant elements to reduce catalyst costs.In this regard, gallium oxide alone was found to be active in the PDH reaction. [16,17][20] Furthermore, two patents have been filed claiming the usage of gallium oxidebased catalysts, applying silica-and alumina-containing support materials. [21,22]nterestingly, the addition of preferably 50-200 ppm nickel was shown to have a beneficial effect on the overall propylene yield (36 % at 580 °C). [21]However, the actual role of nickel was not discussed and analysis of possibly formed NiGa alloys was not considered.Moreover, the catalysts were only tested for a few minutes in which the initial activity could be preserved. [21]n the end, this catalyst type has never found application on an industrial scale, despite the high dehydrogenation activity.
A similar principle was reported for nickel when being used for the non-oxidative dehydrogenation of alkanes. [23]For instance, Ni was alloyed with Cu, [42,43] Au, [44] Mo, [45] Sn [46][47][48] or Zn [49,50] leading also to beneficial geometric and electronic modifications.However, there are only a handful of studies specifically targeting the PDH reaction using alloyed Ni. [46,51,52] Only two previous studies are known to have tested the combination of Ni and Ga in the dehydrogenation of light alkanes. [51,52]Only one of them studied this system for PDH. [51]It was stated that Ga-rich surface compositions on 1 : 1 NiGa alloys provide the highest propylene selectivity (ca.90 %) and conversion of propane (20-10 %).Yet, the propane feed concentration was only 10 % which is far from industrially relevant conditions.
Furthermore, results from one of the previous studies [51] indicated that the use of SiO 2 as a support material is not beneficial, whereas using Al 2 O 3 led to the highest propylene activity.Although the intrinsic activity of Al 2 O 3 is well known in PDH, [53][54][55] the mentioned report [51] lacked performance data of the pure Al 2 O 3 used as support material.Hence, the individual contributions of NiGa alloys and Al 2 O 3 to the catalyst's performance were not transparent.
In this study, SiO 2 was chosen as the support material as it is inert in the PDH reaction. [46]That way, the catalytic performance derives exclusively from the deposited components.58] We report the synthesis of a gallium oxide catalyst, modified with NiGa alloys, combining atomic layer deposition (ALD) and incipient wetness impregnation.ALD is a well-established technique used for the deposition of uniform, nanoscale films in the semiconductor industry. [59]In the last decade, ALD also emerged as a precise tool for the synthesis of heterogeneous catalysts. [60][69] So far, only one publication is known which applied ALD for the synthesis of gallium oxide catalysts used for propane dehydrogenation. [70]owever, alumina was used as support material and the dehydrogenation reaction was CO 2 -assisted.
Here, ALD was applied to deposit a homogeneous interface of gallium oxide on silica.The high dispersion of GaO x ensured intimate contact with the subsequently impregnated nickel precursor.A schematic synthesis procedure is depicted in Figure 1 and an investigation of the ALD method is reported in detail elsewhere. [71]The modified gallium oxide catalysts were tested for the PDH reaction under industrially relevant, Oleflexlike conditions.Moreover, the resulting NiGa species and respective tendencies for coke formation were determined.

Catalytic performance (PDH)
The objective of this study was to determine the influence of nickel on the catalytic behavior of gallium oxide in the dehydrogenation of propane.Mesoporous silica (SiO 2 ) was chosen as the support material on which gallium oxide was deposited.The advantage of SiO 2 is that it does not contribute to the catalyst's performance which makes the activities of the individual components more transparent. [72]ince nickel should be added subsequently, it is reasonable to disperse gallium oxide evenly on the support.This would ensure intimate contact between nickel and gallium oxide, inhibiting the segregation of nickel particles.
Previous studies have demonstrated that atomic layer deposition (ALD) can be used to distribute gallium oxide (GaO x ) homogenously on SiO 2 powder. [71]Therefore, ALD was applied to synthesize GaO x (ALD)/SiO 2 as reference material for this study.One ALD cycle was conducted yielding a Ga loading of 14 wt %.Furthermore, gallium oxide was applied in a comparable amount to SiO 2 using incipient wetness impregnation (IWI) as a reference.
The catalytic dehydrogenation performance of both starting materials, along with unmodified SiO 2 , is shown in Figure 2. The fresh catalysts were pre-treated in 10 % H 2 and afterward exposed to Oleflex-like conditions with high propane partial pressure and H 2 co-dosing (59 % C 3 H 8 and 29 % H 2 , 600 °C).After a rapid deactivation within one hour, the unmodified SiO 2 support material was found to be nearly inactive, with propane conversions below 3 %.Hence, the blind activity of SiO 2 and the quartz reactor itself are neglectable after the first hour.The initial higher activity of SiO 2 should also not affect the performance of the GaO x (ALD)/SiO 2 catalysts, as SiO 2 was homogeneously covered with GaO x after ALD. [71]In the initial phase, the GaO x (ALD)/SiO 2 catalyst exhibited a high propane conversion of around 30 %. Within 5 hours of time-on-stream (TOS), however, the conversion decreased to below 20 %.At the same time, the selectivity towards propylene gradually increased and reached a plateau at 81 %.Afterward, the selectivity towards propylene remained stable and the conversion of propane decreased only by 5 % over the last 7 hours.
The supported gallium oxide catalyst synthesized by IWI (Ga 2 O 3 (IWI)/SiO 2 ) showed a similar behavior.However, its propane conversion remained around 5 % below that of the GaO x (ALD)/SiO 2 .Moreover, the selectivity of propylene was similar for both catalysts which suggests that the same active phase was formed.Therefore, the superior activity of the ALD catalyst derived from its higher dispersion of gallium oxide.This is in line with several reported studies comparing ALD and IWIprepared catalysts. [36,65,69]onsequently, atomic layer deposition (ALD) was further used to prepare the starting material for the subsequent impregnation with nickel.In total, four different NiÀ GaO x /SiO 2 catalysts were synthesized containing 0.8 wt %, 1.6 wt %, 3.2 wt %, and 10 wt % Ni on GaO x (ALD)/SiO 2 .The resulting nominal atomic ratios are listed in Table 1, together with the specific surface areas (SSA) determined by N 2 physisorption.Each NiÀ GaO x /SiO 2 catalyst possessed an SSA of around 330 m 2 / g due to the support material SiO 2 , which provided 505 m 2 /g.The set of catalysts was tested in the dehydrogenation of propane (PDH) as previously demonstrated on the unmodified GaO x catalyst.The achieved yields of each catalyst are depicted in Figure 3 (b).Initially, all NiÀ GaO x /SiO 2 catalysts resulted in lower production of propylene compared to GaO x /SiO 2 .The reduction in yield can be directly correlated with an increase in nickel content.Generally, this trend can be attributed to the fact that the addition of nickel negatively impacted the overall conversion of propane.
However, it is noteworthy that the NiÀ GaO x /SiO 2 catalysts consistently exhibited a higher propylene selectivity than GaO x / SiO 2 .Moreover, low nickel loadings (0.8 and 1.6 wt %) led to only a minor decrease in conversion (À 1 % at 12 h TOS).Therefore, the propylene yield achieved with the Ni(1.6 wt %)À GaO x /SiO 2 catalysts reached that of GaO x /SiO 2 towards the middle of the experiment.
Still, the lower conversion (Figure 3 (a)) of the NiÀ GaO x /SiO 2 catalysts indicate that the formed NiGa species is less active than standalone GaO x or not active in the PDH reaction at all.The lower activity could partially derive from a reduction of available gallium oxide surface area.For instance, nickel might consume gallium atoms to generate alloys or block active sites
Unfortunately, the only published example which reports on the interplay of nickel and gallium for PDH does not discuss the activity of gallium oxide alone. [51]It was only stated that Garich alloys or Ga-terminated alloys are most active in PDH.Here, however, the propane conversion decreased whenever Ni was added to gallium oxide, regardless of the formed NiGa phase.
The actual benefit of adding nickel to gallium oxide becomes apparent when considering the product selectivity, as in Figure 3 (c).Each NiÀ GaO x /SiO 2 catalyst, possessing 3.2 wt % or less Ni, provided an over 5 % higher propylene selectivity than GaO x /SiO 2 during the entire experimental run.At 12 h TOS, the Ni(0.8 wt %)À GaO x /SiO 2 catalyst reached a propylene selectivity of 89 %, which was 7 % above that of the pure oxide.This level of selectivity actually approaches the values found for the PtZn [34][35][36] and PtGa [14,15] systems, with only the conversion rate being lower on NiÀ GaO x .
In contrast to the Ga-containing catalysts, the standalone Ni/SiO 2 showed a much lower tendency to produce propylene.The conversion of propane remained at 7 % and the propylene selectivity was 12 % at 12 h TOS.At the same time the selectivity towards methane (> 40 %) was significantly higher compared to the NiÀ GaO x /SiO 2 catalysts.This followed the expectations, as nickel is known for hydrogenolysis and cracking of propane which leads to methane. [27,28]hen Ni was added to GaO x , the methane selectivity dropped to below 3 % during the whole PDH tests.This is the first indicator, that the Ni deposited on GaO x did not form segregated particles.Therefore, the Ni atoms must be incorporated into the GaO x framework which inhibits its typical hydrogenolysis and cracking activity.
Only the Ni(10 wt %)À GaO x /SiO 2 provided a slightly higher tendency to produce methane than GaO x (ALD)/SiO 2 .At the same time, the propane selectivity never reached that of the other NiÀ GaO x /SiO 2 catalysts.This suggests that despite the high dispersion of GaO x , not all Ni was incorporated in the case of the highest Ni loading.Consequently, to increase the propylene selectivity of GaO x while maintaining the propane conversion, only small amounts of Ni, below or equal to a molar ratio of 1 : 9 (Ni : Ga) should be considered.
Regarding the gaseous side products like methane, ethane, and ethylene, the other NiÀ GaO x /SiO 2 catalysts showed the same selectivity distribution as the unmodified GaO x .Therefore, it appears that the observed selectivity advantage of nickel addition does not pertain to the inhibition of CÀ C cleavage (hydrogenolysis or cracking). [5,6]s mentioned before, the production of propylene can also be inhibited by deep dehydrogenation of adsorbed intermediates (excessive removal of hydrogen). [5,6,10]Therefore, the superior propylene selectivity observed for NiÀ GaO x /SiO 2 catalysts must derive from a decreased tendency for deep dehydrogenation.This reaction pathway leads to multiply dehydrogenated alkenes/alkynes acting as coke precursors and formation of high boiling hydrocarbons like aromatics. [6,10,73]nwanted, heavy compounds which were not detected by the GC are labeled as � C 5 H x in Figure 3 (c).In fact, the higher propylene selectivity of NiÀ GaO x /SiO 2 catalysts was accompanied by a decreased tendency for carbon loss (production of undetected � C 5 H x ).Herein, coke deposited directly on the catalyst might only be a fraction of the unwanted compounds.However, the quantity of formed coke should scale with the selectivity to the non-detected, heavier molecules.

Coke formation
When heavy compounds are formed by undesired side reactions during PDH, a fraction of it is deposited as coke on the catalyst surface. [6,10,73]Therefore, the higher propane selectivity of NiÀ GaO x /SiO 2 catalysts should be accompanied by a lower tendency to form coke. To confirm this, the coking rate on selected catalysts during propane dehydrogenation was investigated in situ using a thermogravimetric balance.
The respective sample was placed in a crucible on which the same propane feed was applied as in the PDH experiments.During the reaction, the change in mass of the crucible was monitored with a balance and the chamber geometry allowed some of the propane to pass alongside the crucible.The resulting mass changes over time on stream are shown in Figure 4 (b).In the case of Ni(10 %)À GaO x /SiO 2 , the mass increased significantly by over 10 % within the first two hours.Subsequently, the rate of coke deposition reduced and the mass increased by an additional 15 % over the last 10 hours.This suggests again, that not all nickel was sufficiently alloyed with gallium at this loading, leading to a cracking behavior characteristic of nickel.Moreover, the decreased rate in the final phase might be due to the blockage of surface sites by carbon and the deactivation of the gallium oxide.This is also reflected in the significant drop in propylene yield as described above.
GaO x /SiO 2 led to a constant mass gain during the thermogravimetric PDH process.After 12 h TOS, the mass increased by over 4.3 wt % which is significantly lower compared to the sample with Ni(10 %)À GaO x /SiO 2 .However, Ni(1.6 %)À GaO x /SiO 2 resulted in an overall lower deposition rate than the GaO x /SiO 2 , leading to a 2.7 wt % mass gain.This is the first indicator, that the lower propylene selectivity of GaO x /SiO 2 derived from an increased tendency to form coke.
Although the geometry of the balance chamber differed from that of the fixed bed reactor, the mass gains are in line with the carbon contents determined by combustion analysis.Herein, the amount of deposited carbon was measured by exsitu oxidation of the catalyst bed retrieved from the reactor.Calculated mass fractions of carbon formed after 12 h PDH are displayed in Figure 4 (a).
In fact, the catalyst with 10 wt % Ni led to the deposition of 24 wt % carbon, which is in the same range as the mass change observed using thermogravimetry.This outcome closely matches the coke content received when using pure nickel on SiO 2 with a carbon fraction of 30 wt %.When only gallium oxide is applied for PDH, the amount of carbon drops significantly to 4 wt %.Once more, this value aligns with the increase in mass detected by thermogravimetric analysis.
The lowest mass fraction of carbon was obtained when combining gallium oxide with the NiGa nanoalloys.The NiÀ GaO x catalysts in the range of 0.8 wt % to 3.2 wt % Ni all led to carbon fractions below 2 wt %.Especially the sample with 3.2 wt % Ni had a significantly decreased carbon content of only 1.1 wt %.Unfortunately, recent literature reports on the NiÀ Ga system do not target the actual amounts of deposited carbon or its removal in detail. [51,52]dditionally, the produced amounts of propylene were calculated considering the respective yields and inlet flows of propane (see also Figure S10).The deposited carbon mass was put in relation to the cumulative quantity of propylene to determine a coke production rate (Figure 4 (a) below).
The unmodified GaO x /SiO 2 provided a rate of 8.3 kg of coke per ton of propylene over 12 hours.In other words, the amount of coke produced equaled almost 1 % of the total propylene production in terms of mass.The NiÀ GaO x /SiO 2 catalysts with less than 10 wt % of nickel, however, resulted in only half the coke production rate (< 4 kg/t).
The best value was achieved by applying 0.8 wt % Ni, which also showed the highest selectivity towards propylene.This means, that by tuning the gallium oxide with small amounts of NiGa alloys, the propylene could be synthesized with half the coking rate.Therefore, the addition of NiGa is a trade-off between a slightly reduced single-pass conversion of propane and an increased selectivity to the desired product.
Especially the mitigation of coking is a crucial target for dehydrogenation processes. [3,8]The greater the loss of propane through conversion into coke, the higher the emission of CO 2 during the oxidative regeneration.In this case, especially the gallium oxide with 1.6 wt % Ni would be the most efficient candidate as it delivers a comparable propylene yield as the pure oxide while producing 50 % less coke.
[11][12] Figure 4 (c) shows the performance of the Ni(1.6 %)À GaO x /SiO 2 catalyst with and without H 2 co-feed, at the same conditions as before.Upon removal of H 2 from the stream, the selectivity towards propylene dropped to the value achieved with the GaO x /SiO 2 sample (ca.81 %).Furthermore, the fraction of heavier species (� C 5 H x ) increased to a similar level as for the unmodified oxide.This suggests that hydrogen directly influences the NiÀ GaO x system and the reaction path- way leading to propylene.This aspect will be the matter of further discussion in the last section.

Regeneration of NiÀ GaO x /SiO 2
The Ni(1.6 %)À GaO x /SiO 2 catalyst resulted in the highest propylene yield among the NiÀ GaO x samples.Therefore, this system was further investigated in multiple consecutive PDH cycles.
Generally, the coke being deposited on the surface of the catalyst induces deactivation during the dehydrogenation process.In order to regenerate the catalyst without removing it from the reactor, the coke was burned off.Temperatureprogrammed oxidation (TPO) experiments were conducted to identify conditions at which the coke is removed.The results of TPO are shown in Figure 5 (a) and Figure S11.The spent catalyst was assembled in a thermogravimetric scale and heated to 1100 °C in syn-air (20 % O 2 in N 2 , 10 K/min).The blue dashes along the mass curve indicate the maximum slopes deriving from the minima of the first derivatives.
At 148 °C, physisorbed water is evaporated leading to the first drop in mass.The second rapid decrease at 540 °C accounts for the removal of carbon content as CO 2 .This was also observed in the online mass spectrometer.The two elevations around 540 °C point to an increase in mass due to the oxidation of metal content.Therefore, re-oxidation of the NiÀ Ga species might not be fully avoided when burning off the deposited carbon.Despite the overlay of the two mass-changing effects, the order of magnitude of the mass loss is in line with the measured carbon content (below 2 wt %).Identified conditions were applied for an oxidative treatment during the PDH process to execute three consecutive reaction cycles, which is shown in Figure 5 (b).In each cycle, the Ni(1.6 %)À GaO x /SiO 2 catalyst was exposed to the reaction conditions for 12 h and afterward purged with 20 % O 2 at 550 °C for one hour.The next cycle started with activation in diluted hydrogen at 600 °C for one hour, followed by applying the standard PDH conditions.As a reference, the GaO x /SiO 2 sample was regenerated under the same conditions and tested in two cycles.
In each of the three testing segments, the yield of propylene progressively decreased over time as observed for the first 12 hours.The initial yield level in the first cycle using Ni(1.6 %)À GaO x /SiO 2 was restored by 85 % in the second segment.Nevertheless, along all PDH cycles, the NiÀ GaO x catalyst always provided a selectivity towards propylene of over 86 %.This rather points towards a maintained nature of the active phases or its successful regeneration during the oxidative treatment.
Moreover, the yield provided by the Ni(1.6 %)À GaO x /SiO 2 sample exceeded the yield of the unmodified gallium oxide, in the second cycle.Its conversion was similar to that of the oxide while its propylene selectivity was up to 5 % higher.The conversion and selectivity of both catalysts at the end of the second cycle is displayed in Figure 5 (c).
Generally, the decreased yield originated from a slightly lowered conversion, which was the case for both, Ni(1.6 %)À GaO x /SiO 2 and GaO x /SiO 2 .Therefore, the slight decrease in yield between the first two cycles might be rationalized by the reduction of available oxide surface area.In fact, STEM EDX mappings demonstrated that the gallium oxide agglomerated after three cycles (Figure S12).Generally, areas, where gallium was detected, appeared more concentrated compared to the images after only one cycle.The structural integrity will be further evaluated in the next section.
Nevertheless, in the third cycle, the initial propylene yield of the second cycle was recovered by 97 % which indicates full restoration of the catalyst state.As this was not the case from the first to the second cycle, the most active sites might be altered rapidly at the beginning of the first segment.Moreover, this indicates that the absolute reduction of the surface area of gallium oxide and its mobility is limited.At this point, future studies might elucidate the deactivation behavior of NiÀ GaO x catalysts in more detail.Generally, the demonstration of the recyclability of the NiÀ GaO x catalyst is significant to promote the usage of nickel-based systems for PDH. [74]The maintained selectivity towards propylene and recovered conversion of propane make the NiÀ GaO x system attractive for future studies.

Structural investigation
Powder X-ray diffraction (XRD) and scanning transition electron microscopy (STEM) were performed to investigate how Ni is incorporated into GaO x .The diffractograms of the spent samples (after PDH) are displayed in Figure 6 (a) and the range is adjusted to the main reflections of the relevant metals and alloys.As a result, the addition of Ni to GaO x /SiO 2 led to the appearance of distinct maxima matching different NiGa alloys.None of the samples displayed clear reflections as it would occur for crystalline Ni metal or NiO. [75,76]Furthermore, diffractograms of Ni/SiO 2 contained a different pattern than the NiÀ GaO x /SiO 2 samples (Figure S3).
Only the Ni(10 wt %)À GaO x catalyst showed indications of NiO at 43.2°(2θ) or Ni at 44.5°and 52°(Figure S2).This rationalizes its higher selectivity to methane as the un-alloyed and segregated nickel still catalyzed hydrogenolysis and/or cracking reactions.Despite the presence of a molar excess of Ga in all catalyst compositions, higher Ni loadings led to Ga-poor alloys (Figure 6 (a)).Moreover, each NiÀ GaO x /SiO 2 material displayed reflections for gallium oxide after PDH (Figure S2).Hence, the reduction of GaO x might primarily occur in the vicinity of the deposited nickel, thereby restricting its mobility to a certain extent.
In fact, the X-ray photoelectron spectroscopy (XPS) of the 1.6 wt % sample revealed, that nickel is metallic after PDH (Figure S3).Herein, its signal is similar to the one acquired for pure nickel supported on SiO 2 , with a maximum of 853.3 eV.Additionally, NiO can be ruled out as it typically appears at lower binding energies. [77]he XRD of the NiÀ GaO x /SiO 2 catalysts with fractions of 0.8 wt % and 1.6 wt % Ni showed twin maxima that are typical for Ni 2 Ga 3 alloys.Moreover, the samples with increased Ni loading resulted in Ni-richer alloys.Herein, the main reflection of the 3.2 wt % Ni sample can be assigned to Ni 1 Ga 1 and for the 10 wt % Ni sample, a mixture of Ni 1 Ga 1 and Ni 3 Ga 1 alloys was observed.These findings indicate that GaO x is partially reduced to Ga metal which alloys with the impregnated Ni under the highly reductive reaction conditions. [17]This also underlines the hypothesis, that the observed decrease in activity of the NiÀ GaO x /SiO 2 catalysts might stem from the consumption of available GaO x surface sites through alloying with Ni.
Yet, the decreased activity might also derive from the interplay between the NiGa alloys and GaO x phase, as the selectivity and coke formation are also significantly altered.To rationalize the performance differences between the NiÀ GaO x / SiO 2 and GaO x /SiO 2 catalysts, the influence of H 2 must be considered.Especially the synergetic relationship between the NiGa alloys and GaO x via a potential H 2 spillover is conceivable.Saerens et al., [10] showed the positive impact of H 2 co-dosing during propane dehydrogenation on Pt(111) surface.The theoretical simulations, supported by experimental data, clearly showed that by increasing the hydrogen surface coverage, the formation of deeply dehydrogenated species, such as ethylidene and methylidyne, can be reduced.
As evident from past studies on CO 2 hydrogenation, [78,79] NiGa alloys are known to activate hydrogen and transfer it to Ga 2 O 3 for the hydrogenation of intermediates to methanol.We suggest that a similar H 2 spillover mechanism takes place on the NiÀ GaO x /SiO 2 surface during dehydrogenation reaction.The formed NiGa alloys would activate H 2 from the co-feed or generated during the dehydrogenation, and transfer it to adjacent GaO x sites.
This would not only rationalize the increased propylene selectivity but also the decreased conversion.The hydride donation from HÀ NiGa to GaO x partially reduces Ga(III) to Ga(� II) which diminishes the number of adsorption sites for propane.Simultaneously, the higher abundance of hydro- genated sites decreases the dehydrogenation potential of already adsorbed propane.Consequently, coking is mitigated because the hydrogen enrichment would inhibit the deep dehydrogenation pathway to propylidene, ethylidene or methylidyne. [6,10,46,73]This hypothesis is in agreement with the results shown in Figure 4 (c).Herein, removing the H 2 cofeed decreased the propylene selectivity and increased the carbon loss.Additionally, adding H-donating NiGa alloys to gallium oxide significantly reduced the coke formation when H 2 was codosed.
In a recent study by Zhang et al., [80] Pt was shown to promote the H 2 dissociation and increase the surface coverage of hydrogen species on GaO x .In the case of PdÀ GaO x , Collins et al. demonstrated that the reduction effect was indeed accomplished by hydrogen spillover from the noble metal to gallium oxide. [81]enerally, the hydrogen spill-over and partial reduction of gallium oxide were validated using XPS analysis. [80,81]ere, XPS analysis was applied to the selected NiÀ GaO x /SiO 2 catalyst to determine the reduction degree of GaO x after PDH.The spectra of the Ga 2p 3/2 region of GaO x /SiO 2 and Ni(1.6 wt %)À GaO x /SiO 2 are displayed in Figure 7.
The spectrum of the unmodified GaO x catalyst consisted exclusively of the Ga(III) signal originating from Ga 2 O 3 .The NiÀ GaO x samples could be deconvoluted into three peaks at 1118.9, 1117.6, and 1116.8 eV, which can be assigned to Ga(III), Ga(II or I), and Ga(0). [18,80,82,83]Herein, the partially reduced GaO x , in the oxidation state one or two, is summarized as Ga δ + .The binding energy deviation from metallic and oxidic gallium mainly derives from the difference in charge of Ga. [80,82,83]round 2.3 wt % of Ga should be captured as metallic gallium in the alloys, considering that mainly Ni 2 Ga 3 was formed and 1.6 wt % Ni was deposited on 14 wt % Ga.This would correspond to a molar fraction of all gallium of 16.6 mol% Ga(0).
In fact, when the signal areas of the three gallium species resulting from XPS are juxtaposed, an allocation of 13 % is ascertained for metallic Ga(0).This value is in good agreement with the estimated mole fraction which validates these XPS results.Moreover, the Ga(III) makes up for a mole fraction of 70 mol% and the partially reduced Ga δ + accounts for 17 mol%.Since the pure oxide did not show partially reduced gallium, the reduction must be accomplished by the NiGa.This essentially supports the hypothesis of a hydrogen spill-over from NiGa to GaO x during PDH.
STEM was conducted to reveal the structure of the supported NiGa alloys after PDH.Results for the Ni-(1.6 wt %)À GaO x /SiO 2 catalyst are shown in Figure 6 (b), while images of the other samples are listed in the SI (Figure S4-6).The HAADF image revealed the presence of individual nanoparticles on the SiO 2 surface with an average diameter of 19 nm (� 6).
Moreover, energy dispersive X-ray diffraction (EDX) scans were conducted to determine the location of nickel and gallium along the sample.The mappings disclosed that the nanoparticles consist of nickel and gallium, while the nickel counts are always situated in close proximity to gallium.Applying EDX also compliment the Ni : Ga compositions found by XRD.For example, one area selective mapping quantified a signal count ratio of 38 to 62 (Ni : Ga), which is in line with the X-ray diffractions indicating Ni 2 Ga 3 alloys.Herein, a nanoparticle was scanned exclusively situated at the edge or a shallow part of the support material.That way, signal noise originating from the support or other particles underneath is minimized, and only the selected nanoparticle accounts for the EDX detections.
An EDX line scan along two particles of the Ni(1.6 wt %)À GaO x specimen underlines these findings (Figure 6 (b), right).The average counts of nickel were approximately two-thirds that of gallium along both particles.Consequently, nickel and gallium are always situated next to each other in a constant composition.Additionally, the darkfield image showed sharp edges of the nanoparticles which can also be observed as a steep rise of counts in the line scan.Hence, core-shell particles are ruled out.
The observed alloy formation between nickel and gallium rationalizes again the respective selectivity towards propylene.Once NiGa nanoalloys are formed, Ga atoms impart the geometric modification of Ni, which inhibits its tendency for CÀ C cleavage. [27,28]t is noticeable that nickel and gallium no longer form uniform nanoparticles after the regeneration of Ni(1.6 wt %)À GaO x /SiO 2 (Figure S12).Therefore, the regeneration process distorted the NiGa alloys by the harsh oxidation and subsequent reduction.A temporary formation of oxides agrees with the slight increase in mass observed during the TPO (Figure 4 (a)).
Furthermore, XRD analysis revealed that gallium oxide crystals are present after three PDH cycles, whereas it was still XRD-amorphous after only 12 h TOS (Figure S12).Additionally, the reflection associated with Ni 2 Ga 3 alloys nearly vanished while crystalline Ni 3 Ga 1 was formed.As the catalyst maintained its propylene selectivity, the molar ratio of the nanoalloys is in fact not decisive.This suggests that as long as all nickel is alloyed with gallium, the hydrogen spillover and partial reduction of the oxide takes place.These observations complement the STEM results (Figure S12), which suggested that the integrity of the nanoalloys was disturbed by the regeneration procedure.Ultimately, it is evident that gallium oxide is partially reduced and becomes mobile under reaction conditions.This Area selective EDX mappings were also performed to determine whether the nickel is spread atomically on the gallium oxide.Respective mappings along the 0.8 wt % NiÀ GaO x catalyst are shown in Figure 6 (c).This sample was chosen as it led to the highest propylene selectivity.However, the resulting spectrum, received from an area where no nanoparticle is located, shows no indication of nickel.Moreover, the scan of a nanoalloy on the same sample provided clear signals at energies assigned to diffracted radiation deriving from the Ni(Kα) shell. [84]Hence, the EDX results show no indication of smaller nickel clusters or even single atoms.
An exemplary image of the 10 wt % Ni sample is displayed in Figure 6 (d).It was observed, that not only the Ni-fraction of the nanoalloys increased with higher Ni-loading, but also the nanoparticle count.The STEM images suggested that the 1.6 wt % Ni catalyst held 350 nanoparticles per μm 2 , whereas the 10 wt % sample reached 3300/μm 2 (Table S3).Therefore, a higher density of nanoalloys translates to a lower propane conversion, which again supports the idea, that gallium oxide is the active species and not NiGa.
Since the present alloy compositions were found, final DFT calculations could be conducted to complement the findings for the coking behavior.In several DFT studies, the correlation between the adsorption energy of a single carbon atom on a metallic surface and the proneness for the formation of coke was observed.In the case of Ni, this was reported for the watergas-shift reaction, [85] the dry reforming of methane, [86,87] and for other transition metals and alloys in PDH. [88]These studies demonstrated that the C-atom adsorption energy is a descriptor for the formation of coke.Here, DFT calculations were carried out to validate the coking ability of different NiGa surfaces.A detailed description of the theoretical process is given in the Supporting Information.
The data points in Figure 8 represent the statistical average according to the Boltzmann distribution of different facets in Ni x Ga y alloys.The adsorption energies on various facets and surface terminations, used to determine this average, were calculated and listed in Table S2.Upon initial observation, it is evident that incorporating gallium into nickel leads to lower absolute values of adsorption energies of carbon.Herein, the Ni and Ni 1 Ga 3 surfaces provided the highest Boltzmann averages of around À 7 eV and À 6.3 eV.To demonstrate the influence of Ga on the carbon deposition on NiGa alloys, we also present the values for pure metallic Ga.Thereby, the insertion of more gallium leads to a gradual decrease in energy, ultimately reaching À 5.5 eV, as observed on the Ni 2 Ga 3 facets.
At the same time, the catalysts holding pure nickel or not fully alloyed Ni led to the highest amount of coke.Samples containing alloys like Ni 1 Ga 1 or Ni 2 Ga 3 , however, had a significantly diminished tendency for coking.Therefore, the NiGa nanoalloys might not only prevent the formation of coke on gallium oxide but also on their own surfaces.8]

Conclusions
This study brings attention to the synergetic benefits between gallium oxide and nickel for propane dehydrogenation (PDH).Silica was decorated with gallium oxide using atomic layer deposition (ALD), resulting in a superior propane conversion and gallium dispersion compared to the synthesis through impregnation.Therefore, the ALD catalyst was used as the starting material on which different loadings of nickel were deposited and tested under Oleflex-like conditions.
XRD measurements indicated the formation of Ni 3 Ga 1 , Ni 1 Ga 1 , and Ni 2 Ga 3 alloys while lower nickel loadings yielded gallium-richer species.The presence of alloy nanoparticles with an average diameter of 12-19 nm was confirmed by scanning transition electron microscopy (STEM).Furthermore, EDX mappings complimented the atomic ratios found by XRD.
Surprisingly, the formed NiGa alloys did not directly affect the conversion of propane.However, the selectivity to propylene was significantly enhanced.Conducting in situ thermogravimetric analysis under PDH conditions revealed that the addition of Ni to GaO x reduced the coking rate substantially.Compared to the unmodified gallium oxide, the formed NiGa alloys decreased the coke deposition by over 50 %.
Similar to previous reports, the decreased tendency for coking and higher propylene selectivity could be rationalized by a partial reduction of Ga(III) sites.In fact, XPS studies suggested the formation of a partially reduced gallium species only when NiGa alloys were present.As NiGa alloys are recognized for their capability to activate hydrogen, the reduction can be explained by a hydrogen spillover.The hydrogen enrichment of the gallium oxide prevents the deep dehydrogenation of adsorbed intermediates which inhibits coking.Additionally, DFT calculations indicated that NiGa alloys exhibit a low tendency for coke deposition compared to Ni.
Finally, the Ni(1.6 wt %)À GaO x /SiO 2 system was regenerated in a periodic PDH process by an oxidative treatment.The coke was successfully burned off between the PDH cycles, yet XRD and STEM indicated that the structural integrity of the nanoalloys and gallium oxide was disturbed.Nevertheless, the propylene selectivity was maintained in all three PDH periods.
Generally, the NiÀ GaO x /SiO 2 samples come close to the selectivities reported for PtGa catalysts.In future studies, tuning of the metal loadings might improve the NiÀ GaO x system even further.Moreover, the contribution of NiGa could be fully unraveled by applying colloidal synthesis to prepare defined NiGa nanoparticles.Thereby, their catalytic behavior can be detangled from gallium oxide and hydrogen activation studies might help to understand the hydrogen spillover.

Catalyst synthesis
The used chemicals and resulting metal loadings are listed in the supporting information.Atomic Layer Deposition (ALD) of gallium oxide was carried out in a self-designed setup of which a detailed description is given elsewhere. [89]Trimethylgallium (TMG) and HPLC-grade water were used as a precursor-reactant combination and the overall ALD process is described in detail elsewhere. [71]r the catalyst synthesis, mesoporous silica (SiO 2 ) powder (500 m 2 / g) was filled into a tubular fixed bed reactor made of quartz glass (20 mL).The ALD process was conducted under a constant gas flow of 100 mL/min at atmospheric pressure.The powder substrate was maintained at 150 °C while the TMG and water dosing units were kept at room temperature.Both reactants were sequentially fed into the reactor using argon as the carrier and purge gas.The applied ALD sequence (cycle) was TMG/Ar-purge/H 2 O/Ar-purge.The point of precursor or reactant saturation was determined by online mass spectrometry.
Both reactants were dosed into the reactor until the mass traces for m/z = 69 (TMG) or m/z = 18 (H 2 O) broke through and reached a plateau.For each sample, one ALD cycle was conducted providing 14 wt % Ga.The resulting material is denoted as GaO x (ALD)/SiO 2 .
The preparation of the NiÀ GaO x /SiO 2 catalysts is schematically depicted in Figure 1.First gallium oxide was distributed on mesoporous silica powder by ALD.Subsequently, in order to enable scalability of the nickel content, incipient wetness impregnation (IWI) was selected as a convenient deposition technique.Nickel nitrate hydrate was dissolved in HPLC-grade water which equaled the maximum amount of water absorbed by the GaO x (ALD)/SiO 2 powder.The solution was distributed on the support and dried in air at 80 °C for 12 h.Afterward, the precursor material was reduced at 600 °C (5 K/min rate), in 5 % H 2 (in N 2 , 200 mL/min), for 3 h, yielding NiÀ GaO x (ALD)/SiO 2 .
As a reference, gallium oxide was supported on the silica powder by IWI.Gallium nitrate hydrate was dissolved in HPLC-grade water which equaled the maximum water absorption of the SiO 2 powder.The solution was deposited onto the SiO 2 support and dried in air at 80 °C for 12 h.Afterwards, the precursor was calcined at 500 °C (5 K/min), in 20 % O 2 (in N 2 , 200 mL/min), for 3 h, yielding Ga 2 O 3 (IWI)/SiO 2 with 12 wt % Ga.

Characterization methods
Synthesized materials were characterized using the following analysis methods: Inductively coupled plasma atomic emission spectroscopy (ICP-OES), combustion analysis (CHN), powder X-ray diffraction (XRD), nitrogen physisorption measurements (applying the B.E.T. method), X-ray photoelectron spectroscopy (XPS), scanning transmission electron microscopy coupled with energydispersive X-ray mapping (STEM-EDX), mass spectrometer-coupled thermogravimetric analysis (TG-MS) and temperature programmed oxidation (TPO).Detailed descriptions of the analysis procedures are given in the supporting information.

DFT Calculations
Density functional theory (DFT) calculations were used to determine the adsorption energy of carbon on the surface of different NiGa alloys.The adsorption energies were calculated based on the relaxation of a single carbon atom on different facets of NiGa alloys.Average values were determined according to the Boltzmann distribution of the formation energies. [90]A detailed description of the procedure and resulting data is listed in the Supporting Information (Table S2).

Propane dehydrogenation (PDH)
The pre-reduced catalysts were compared for their activity in the dehydrogenation of propane.Catalytic experiments were conducted in a continuous flow set-up equipped with a quartz tube as a fixed bed reactor, designed by Integrated Lab Solutions.The applied reactor had an inner diameter of 10 mm while the volume of the catalysts resulted in a bed height of 5 mm.In all runs, the amount of each catalyst was fixed to 500 mg.Prior to a catalytic test, the samples were activated in situ under a continuous flow of 10 % H 2 (50 mL/min, in He) at 600 °C (10 K/min rate) for 1 h.Subsequently, the reactor was purged with 50 mL/min He for 5 min and the gas flow was switched to 17 mL/min, containing 59 % C 3 H 8 and 29 % H 2 in He (1 bar).The resulting gas-hourly-space velocity (GHSV) was 2040 mL g À 1 h À 1 and the temperature at the catalyst bed was maintained at 600 °C.During the regeneration experiment, the catalyst was first exposed to 20 % O 2 (in He, 50 mL/min) at 550 °C and then re-activated as described above.The effluent gas stream was monitored by an online gas chromatograph (Agilent 7890A) equipped with a flame ionization and thermal conductivity detector.Propane conversion (X), product selectivity (S), and propylene yield (Y) were calculated according to Equations ( 2)-(4).
Where _ F represents the flow rates before entering the reactor (in) or in the effluent gas (out) respectively.The product selectivities (3) are calculated based on the respective product concentration and the amount of propane converted.n i represents the number of carbon atoms of the respective compound.Compounds which were not detected by the GC are labeled as � C 5 H x .These compounds are mostly heavy hydrocarbons and coke, which were deposited on the catalyst bed or condensed at colder regions after the reactor exhaust.
The in situ coke formation studies were conducted in a Rubotherm magnetic suspension balance (DynTHERM HP-ST, 2010-01001-D). 100 mg of a catalyst was filled in a quartz glass crucible attached to a quartz glass holder.The samples were heated (10 K/min) to 600 °C in 50 ml/min N 2 and afterwards activated in 10 % H 2 in N 2 .Afterwards, the gas feed was switched to the testing composition (C 3 H 8 /H 2 /N 2 = 1/0.5/0.33) at a pressure of 1.1 bar.The mass-change was tracked for 12 hours and the first measuring point, after switching to propane-rich conditions, was taken as a reference point.The influence of buoyancy was determined through measurement with an empty crucible and subtracted from each mass curve.

Figure 1 .
Figure 1.Schematic synthesis procedure of the NiGa-modified gallium oxide catalysts, supported on silica.The denotation ALD indicates atomic layer deposition and IWI stands for incipient wetness impregnation.

Figure 2 .
Figure 2. Conversion of propane and selectivity towards propylene against time-on-stream over two different gallium oxide catalysts.The selectivity of SiO 2 was below 40 % and is shown in the SI (Figure S9) to maintain the simplicity of the graph.The thermodynamic equilibrium conversion is indicated by a dashed grey line at 37 %.Conditions: T = 600 °C, 59 % C 3 H 8 , total flow: 17 mL/min (C 3 H 8 : H 2 : He 10 : 5 : 2 mL/min).

Figure 4 .
Figure 4. (a) Mass fractions of deposited carbon, determined by elemental analysis, and mass of carbon deposited per total amount of propylene synthesized (in kg/t) of different Ni-modified GaO x /SiO 2 catalysts (both after 12 h PDH).(b) in situ mass-change of the crucible filled with catalyst during PDH in a magnetic suspension balance.(c) Product distribution of one NiGa catalyst with and without H 2 co-feed.Conditions: T = 600 °C, 59 % C 3 H 8 , total flow: 17 mL/min (C 3 H 8 : H 2 : He 10 : 5 : 2 mL/min).

Figure 6 .
Figure 6.(a) X-ray diffractograms of the Ni-modified gallium oxide catalyst after 12 h TOS.The main reflections of crystalline NiGa alloys, Ni metal, and NiO are indicated according to literature.(b) Scanning electron microscopy (STEM) images (HAADF), EDX mappings and EDX line scan of GaO x /SiO 2 modified with 1.6 wt % Ni.Atomic counts of Ni and Ga were collected and determined by the EDX detector.Ga is indicated in purple and Ni in turquoise color.(c) EDX mappings of 0.8 wt % Ni on gallium oxide.(d) STEM images of 10 wt % Ni on gallium oxide.All measurements were conducted on spent catalysts, after 12 h PDH.

Figure 8 .
Figure 8. Adsorption energies of carbon on specific NiGa alloy compositions (calculated by DFT).Main points indicate the Boltzmann average of the respective composition and the lines indicate the energy bandwidth for all facets of the composition.Ni and Ga indicate a metallic, non-oxidic, surface.

Table 1 .
Compositions and specific surface areas of the NiÀ GaO x catalysts.