How does the oxidation and reduction time affect the chemical looping epoxidation of ethylene?

The chemical looping epoxidation (CLE) of ethylene was performed over an Ag catalyst supported on strontium ferrite perovskite (SrFeO3-δ). CLE consists of a reduction step in ethylene where oxygen is transferred from the support to the Ag catalyst to form ethylene oxide (EO), and a separate regeneration step, in which the support is reoxidised in air. The effect of altering the reduction and oxidation times was investigated, analysing changes in the conversion of ethylene and selectivity to EO. Experiments were conducted at 270 °C in a packed bed of Ag(15 wt.%)/SrFeO3-δ with a gas hourly space velocity of 9600 h−1 and a total operation time > 40 h. When the time of the reduction step was increased from 1 to 3 min, selectivity to EO only decreased by 0.4%, demonstrating that CLE can run with prolonged reduction times while maintaining high selectivity. Increased duration of the reoxidation step resulted in both selectivity and conversion increasing, but when varying the oxidation time from 10 to 15 min, the overall improvement in the performance was minimal; thus, CLE can feasibly operate at shortened oxidation times. By increasing the pressure during the oxidation step to 2.5 bar, the duration of the reoxidation step was further shortened to 5 min without impacting the CLE performance. With 1.5 min reduction and 5 min reoxidation steps, a CLE installation producing EO in a pseudo-steady manner would require 4 packed bed reactors operating in parallel. The role of Ag2O in the CLE process was also investigated, demonstrating that the oxide was not selective towards EO.

runaway reaction [5]. Moreover, because the selectivity towards EO decreases in an approximately linear manner with increasing ethylene conversions [2], in the incumbent direct epoxidation of ethylene, conversions are limited to < 15% [6].
Recently, a new process for producing EO has been proposed, in which a solid metal oxide (e.g., SrFeO 3 ) plays a role of active catalyst support, capable of providing oxygen to the reaction [7]. The gaseous oxygen feed in traditional direct epoxidation is thus replaced by lattice oxygen (O lattice ), provided from the metal oxide, an oxygen carrier. Because of the finite oxygen supply from the solid, the process is carried out in two steps: (1) the silver-catalysed epoxidation of ethylene using O lattice , followed by (2) metal oxide regeneration, i.e. replenishment of O lattice by reoxidation in air. Repeating these two steps introduces cyclic operation (loops); thus, the new epoxidation process follows the chemical looping approach, resulting in a new process: chemical looping epoxidation (CLE). In CLE, ethylene and air never mix, which eliminates the risk of forming explosive gaseous mixtures. Similarly, the chance of the epoxidation reaction entering thermal runaway is considerably reduced. The chemical looping approach is also particularly appropriate in the context of smaller, distributed EO production sites, reducing emissions and costs associated with EO transportation, and allowing EO production in remote areas.
So far, the work on CLE has focussed on improving the performance parameters of CLE, namely the selectivity and yield of EO, by modifying the catalytic material, so the composite made of a metal oxide and catalytic metal, Ag. Modification of SrFeO 3 , used as the solid oxide support, with CeO 2 , introduced as dopant, was shown to improve selectivity to EO, most likely because CeO 2 acts as an oxygen gateway and enhances the ionic oxygen transport in the bulk of the metal oxide [8]. The oxygen release characteristics in CLE were also influenced by the type of perovskite phase used for the oxygen carrier, namely by the content of the cubic phase, SrFeO 3 , and the Ruddlesden-Popper phase, Sr 3 Fe 2 O 7 . Overall, previous investigations demonstrated that oxygen carriers could be engineered to increase the yield of EO [9]. Further improvements in CLE's selectivity to EO and conversion of ethylene were achieved by optimising the catalyst-oxide preparation procedure, with the best-performing composites having relatively large (~ 200 nm) Ag particles deposited on SrFeO 3 . The synthesis that resulted in the creation of large Ag particles also promoted strong interaction between the metallic Ag and the SrFeO 3 support [10].
Previous studies on CLE have not included variations in the time needed for each process step. Instead, the relative durations of the oxide regeneration (oxidation) and oxide reduction (epoxidation) were kept the same. However, the time of effective production of EO will influence not only the achievable selectivity and yields of EO, but also determine the economic viability of CLE. Given the looping nature of the CLE process, multiple CLE reactors will need to run in parallel to achieve a pseudo-steady output of EO. The ratio of time spent producing EO in the reduction step to the time for regenerating the oxygen carrier will determine the number of reactors required to achieve a pseudosteady operation. Minimising the number of required CLE reactors will reduce operating and capital expenditure, improving process viability.
Here, the influence of altering the times of each CLE step over 15 wt.% Ag catalyst on SrFeO 3 solid oxide support (Ag(15wt.%)/SrFeO 3 ) was investigated. First, oxidation and reduction step times were changed independently, and their effects on the epoxidation performance, selectivity to EO, and ethylene conversion were observed. Next, the times for both CLE steps were varied simultaneously, aiming to achieve overall shorter CLE cycle times.

Material preparation
Strontium ferrite (SrFeO 3 ) was prepared by a solid-state synthesis method, adapted from Gabra et al. [9]. Stoichiometric amounts of SrCO 3 (0.72 mol, Sigma Aldrich, ≥ 98%) and Fe 2 O 3 (0.36 mol, Honeywell, ≥ 99%) were combined, adding ethanol (50 mL, 99.8%, Fisher Scientific) as a binder to improve mixing during wet ball milling. The mixing was carried out in a Fritsch, Pulverisette 6 ball mill for 30 min at 600 rpm, with the mixture of reagents contained in a 250 mL stainless steel grinding bowl, alongside 13 stainless steel balls (20 mm dia.). After ball-milling, the mixture of SrCO 3 and Fe 2 O 3 was dried in static air at 50 °C for 24 h and sieved to 180-355 µm. The obtained particles were calcined four times for 3 h in static air at 1000 °C, with a ramp rate of 5 °C min −1 . The material was again sieved to 180-355 µm. At room temperature and in air, the perovskite prepared with this method has a stoichiometry of SrFeO 2.82 [11] but will be denoted as SrFeO 3 .
Impregnation of Ag onto the surface of SrFeO 3 was done by incipient wetness impregnation. First, 1.3895 g of AgNO 3 (Alfa Aesar, ≥ 99.9%) was dissolved in 1.2 mL deionised water. The volume of the prepared solution matched the pore volume of 5 g of SrFeO 3 particles, as determined empirically beforehand. For the wetness impregnation, the AgNO 3 1 3 solution was added dropwise to 5.00 g of SrFeO 3 particles, which were frequently agitated with a spatula. The sample was then dried in static air at 120 °C for 12 h, followed by double calcination at 650 °C for 5 h with a ramp rate of 5 °C min −1 , with cooling to room temperature between calcinations. The impregnation resulted in 15 wt% Ag loading on SrFeO 3 .
To confirm that the oxygen involved in the epoxidation reaction during the CLE experiments originated as O lattice of SrFeO 3 , rather than O lattice from the reduction of Ag 2 O to Ag 0 , a batch of Ag 2 O on Al 2 O 3 was also prepared. Here, Ag 2 O was loaded onto Al 2 O 3 (355-425 µm) particles through a deposition precipitation method. A batch of 25 g of Al 2 O 3 particles was placed in 25 mL of deionised water. A required amount of AgNO 3 (6.9475 g) for 15 wt% silver loading was dissolved in 6 mL of deionised water, and the AgNO 3 solution was added to the Al 2 O 3 particles in water. Concentrated NaOH (31 mol% NaOH prepared by dissolving 10 g of NaOH (Fischer Scientific) in 10 mL of deionised water) was then added dropwise under continuous stirring until the pH increased to 10, resulting in Ag 2 O precipitating out of the solution. The resulting Ag 2 O/Al 2 O 3 particles were washed and filtered 7 times with deionised water, then dried in static air at 120 °C for 12 h. The material was calcined at 195 °C for 5 h. The temperature of 195 °C was selected to ensure that the Ag 2 O did not decompose to Ag 0 .

Experiments in a packed bed
Chemical looping epoxidation (CLE) experiments were carried out in a packed bed housed in a vertically held quartz tube of 8 mm i.d., 560 mm in length, equipped with a grade 1-porosity sintered Al 2 O 3 disc, located 280 mm from the bottom. The experimental setup is presented in Fig. 1. The bed material was introduced from the top of the tube in three layers: (1) a bottom layer of 0.5 g α-Al 2 O 3 (300-425 µm); (2) a middle active layer of 1.5 g Ag(15 wt.%)/SrFeO 3 or 2 g Ag 2 O/Al 2 O 3 catalyst; (3) a top layer of 4 g α-Al 2 O 3 , with the last layer used to preheat and uniformly distribute the feed gas, which was introduced from the top, ensuring that the feed gas was at the reactor operating temperature when reaching the catalyst. The quartz tube was placed in an electrical tube furnace and positioned such that the packed bed was located within the isothermal region of the furnace, which was determined by measuring the vertical temperature profile of the furnace. The temperature was measured in the middle of the packed bed with an N-type thermocouple in an Inconel sheath. The temperature measurement was used to control the heating of the furnace.
The CLE experiments were carried out at 270 °C and with a gas hourly space velocity (GHSV) of 9600 h −1 and flowrates of 150-200 mL min −1 . The gas flow rates were controlled with calibrated rotameters and measured with an electronic flowmeter (ADM, Agilent). A single CLE experiment consisted of 5-33 cycles, each cycle comprising four steps: (i) purge with N 2 ; (ii) reduction with 5.4 vol.% C 2 H 4 in N 2 ; (iii) purge with N 2 ; (iv) oxidation with blended air (21 vol% O 2 in N 2 ). All gases were supplied by BOC, > 99.998%. The switching between gases was performed with solenoid valves controlled with a time-based programme. The duration of the reduction step was varied from 5 to 90 s, whilst the oxidation step was varied from 15 to 900 s for different CLE experiments. The same batch of the material  was used when varying reduction and oxidation times; thus, the operating CLE time exceeded 40 h; nevertheless, the performance of the catalyst remained stable.
Experiments operated at elevated pressures during the oxidation step were performed in a second reactor, made of a stainless steel tube, 15 mm i.d. The packed bed was created the same way as in the quartz reactor. A K-type thermocouple was placed in the middle of the packed bed and used to control electrical heating. The operational pressure was regulated manually with a pressure gauge and backpressure regulator located at the outlet of the reactor. The reduction step was carried out for 1.5 min at 1 bar (absolute), while pressure in the oxidation step was varied from 1 to 2.5 bar, keeping the time at 5 min. Mechanical properties of catalyst particles were investigated before and after CLE experiments at 2.5 bar with a texture analyser (TA.XT Plus). Results given in Fig. S10 show no significant change in strength after 40 h of the overall experimental time for CLE cycles.
Gas outflow from CLE experiments was passed to a Fourier Transform Infrared (FTIR) analyser (MKS Instruments, Multigas 2030), equipped with a liquid N 2 cooled HgCdTe detector. Before each CLE experiment, the detector was allowed to cool for 2 h; during this time, the packed bed was heated to 270 °C and kept at this temperature under a flow of air. A single FTIR scan was performed for 0.97 or 1.87 s, measuring the signal in the range of 100-5000 cm −1 at a resolution of 0.5 cm −1 . The collected FTIR spectra were analysed for C 2 H 4 , EO, CO, CO 2 and H 2 O using MKS MG2000 software.
For the analysis of the obtained results, the start time (t start ) of each CLE cycle was taken to be the time when CO 2 was detected above 20 ppm for the first time in the cycle, and the end time (t end ) was taken to be when the ethylene concentration reached the maximum value in the cycle. The carbon balance in each reduction step was calculated assuming that the total molar flow rate was constant, which is justified for dilute gases. The carbon balance was calculated as: where y i is the measured mole fraction of species i, t is time, and y C 2 H 4 feed is the mole fraction of C 2 H 4 in the feed. The average conversion achieved during a reduction step was estimated as: The mean selectivity towards EO was calculated as: The cumulative oxygen released during each CLE cycle was calculated as: where F is the total molar flow rate, and the 3y CO 2 term accounts for the formation of CO 2 as well as H 2 O associated with the complete combustion of ethylene 0.5C 2 H 4 + 1.5O 2 → CO 2 + H 2 O . The instantaneous oxygen release during a CLE cycle was determined as: When calculating the instantaneous oxygen release, the collected FTIR results were first smoothed with a Savitzky-Golay filter (polynomial order 2, frame length 15) and deconvoluted using a first-order CSTR-model: where y dec i is the deconvoluted molar fraction of species, i, y smoothed i is the smoothed molar fraction of species i, and t c is the characteristic time assessed with a step-change experiment performed with an inert bed of Al 2 O 3 at 270 °C. Results Total carbon in all products Total carbon detected = ∫ t end t start y EO + 0.5y CO 2 + 0.5y CO dt ∫ t end t start y C 2 H 4 + y EO + 0.5y CO 2 + 0.5y CO dt dt of the deconvolution and the ethylene profile recorded during the step-change experiment are presented in Fig. S1 in the Supplementary Information (SI). To prevent the results from overshooting or reaching negative values, the effect of deconvolution was managed by decreasing the experimentally determined t c to 24.5 s. Deconvolution reduced the time needed for a concentration of a gas component to increase from 10 to 90% of its nominal value from 48 to 35 s, as presented in Fig. S1 in the SI.

Catalyst characterisation
Powder X-ray diffraction (XRD) measurements were taken of SrFeO 3 , Ag/SrFeO 3 , and Ag 2 O/Al 2 O 3 with a Bruker D8 Advance diffractometer using Cu-Kα radiation (40 kV and 40 mA). Diffraction patterns were obtained for 2θ ranging from 5° to 80°, with step size 0.05° and step time 1.0-1.5 s. Rietveld refinement was performed with Profex [12] to confirm the phase composition. The used reference patterns were ICSD-91062 (SrFeO 3  Approximately 50 mg of Ag/SrFeO 3 was placed in a 70 µL alumina crucible and introduced into the TGA. The sample was heated in air to 270 °C and kept there for 5 min. An isothermal experiment was carried out with a gas switching programme, mimicking CLE cycles. Each cycle comprised four steps where a reactive gas in the analyser was varied as follow: (i) N 2 purge for 2 min; (ii) 5.16 vol% C 2 H 4 in N 2 for 20 min; (iii) N 2 purge for 2 min; (iv) air for 60 min. Throughout the experiment, the TGA chamber was purged with 100 mL min −1 (NTP) of N 2 , while the 'reactive' gas was supplied at 50 mL min −1 through a capillary tube located above the crucible.
Scanning electron microscopy (SEM) images were taken using a Tescan Mira3 FEG-SEM microscope. The accelerating voltage was 5 kV, and the working distance was between 2 and 8 mm. Particles of investigated samples were sputtercoated with a thin layer of carbon (~ 15 nm) to prevent charging of the sample. Backscattered-electron (BSE) imaging was used in parallel with secondary electron (SE) imaging. An SEM image of Ag/SrFeO 3 and the results of Ag particle size distribution are presented in the SI file, Fig. S4. Particle size distribution for impregnated catalysts was estimated by manual measurements of the diameter of Ag particles (2 measurements per particle, total number of Ag particles = ~ 3500) from BSE images using ImageJ software [13]. The mean size of the Ag particles was 168.6 nm.

Results
A sample of Ag(15 wt.%)/SrFeO 3 was analysed in a cyclic experiment in the TGA, and the recorded relative mass is shown in Fig. 2. Noticeably, oxygen release during the reduction in ethylene was faster than oxygen uptake during the reoxidation in air. The greatest change in mass was observed in the first redox cycle, with the material not returning to the starting mass in subsequent cycles. The mass loss during the first cycle corresponds to the mass loss associated with removing surface impurities accumulated during storage (carbonates and hydroxides [10]) and the mass loss associated with oxygen release. In Fig. 2, the reference value for calculating the relative mass was taken at the end of cycle 1. Over cycles 2-4, the mass loss during reduction was 0.45% ± 0.01%, and was matched by the mass gain in reoxidation. Such a stable activity of the Ag(15 wt.%)/SrFeO 3 sample over the subsequent redox steps and looping cycles indicates the ability to consistently provide O lattice in CLE. Because a complete removal of surface carbonates requires treatments at higher temperatures [9], the small net-changes in the mass between cycles 2-4 was attributed to the further decomposition of surface impurities.
Representative concentration profiles of EO and CO 2 recorded during CLE experiments in the packed bed of Ag(15 wt.%)/SrFeO 3 are shown in Fig. 3. The results were collected over 8 CLE cycles with oxidation and reduction steps lasting 15 and 1.5 min, respectively. In all performed CLE experiments, the highest concentrations of both CO 2 and EO were produced during the first cycle. Afterwards, the bed activity decayed, stabilising around the third cycle and producing similar results in the last 5 cycles. Thus, further on, reported values for (1) oxygen release from SrFeO 3 , (2) CLE selectivity to EO, and (3) CLE conversion of ethylene, represent the average results from the last 5 CLE cycles. No coking was observed during any CLE experiment over Ag/SrFeO 3 , as indicated by the lack of CO 2 during re-oxidation steps (CO 2 < 100 ppm) and by the closed carbon balance of 100% ± 7% for all experiments with a reduction time greater than 15 s. For experiments with reduction times of 5 and 15 s, the carbon balance did not close due to the measured concentrations being strongly affected by the dispersive effects within the reactor at these short reduction times rather than due to coking. Vol:.(1234567890)

Research
Discover Chemical Engineering (2022) 2:4 | https://doi.org/10.1007/s43938-022-00011-4 For reference to the experiments with Ag/SrFeO 3 , CLE cycles were also performed over 2 g of the Ag 2 O/Al 2 O 3 material. Figure 4A shows EO and CO 2 concentration profiles, and Fig. 4B shows ethylene conversion and selectivity to EO. The Ag 2 O(15.9 wt.%)/Al 2 O 3 was prepared to have an amount of Ag comparable to that in Ag(15 wt.%)/SrFeO 3 . During the first reduction step in ethylene, no EO was detected, but up to 7000 ppm CO 2 was produced, indicating the oxygen from Ag 2 O was used to combust ethylene. The CO 2 production dropped to less than 400 ppm in subsequent cycles, and EO was still not detected. Overall, 6% conversion of ethylene was achieved in the first cycle, dropping to 0.5% in the subsequent 8 cycles. Experiments with Ag 2 O indicate that if Ag deposited on SrFeO 3 were in an oxide form, for example, if SrFeO 3 were continuously donating oxygen to a thermodynamically unstable Ag 2 O, then  For the packed bed CLE experiments with 1.5 g of Ag(15 wt.%)/SrFeO 3 , the influence of oxidation time on the analysed process parameters, namely the conversion of ethylene, selectivity to EO, and oxygen release during a CLE cycle, are shown in Fig. 5. Here, each point represents the average value from the last 5 CLE cycles. The obtained catalytic performance achieved in experiments with a 15 min oxidation step is the best published to date for CLE, in terms of both the conversion of ethylene (19%) and selectivity to EO (67%), even though a smaller amount of catalyst was used in comparison to the previous studies (1.5 g vs 2 g, keeping the same GHSV of 9600 h −1 ) [7][8][9][10]. Interestingly, when the oxidation time increased from 10 to 15 min, conversion and selectivity improved slightly, but the amount of oxygen released showed a negligible change. The change in oxygen release is explained by the stoichiometry of the oxidation reactions, as 3 atoms of oxygen are required to facilitate the non-selective combustion rather than a single oxygen atom required for epoxidation. Thus, for a given conversion, a higher selectivity results in a decrease in oxygen release. Here, both selectivity and conversion increased in a manner that kept the total oxygen release constant. When decreasing the oxidation time from 10 to 1.5 min, both the conversion of ethylene and selectivity to EO dropped, the former by over a half, from 17.7 to 7.4%, while the latter by about a quarter, from 63.1 to 45.3%. Thus, whilst both parameters are influenced by the oxidation step, conversion appears to be more affected. Strongly associated with ethylene conversion is the amount of released oxygen, which also dropped when the oxidation time was reduced.
Also shown in Fig. 5 are the amounts of oxygen that could be released from (i) adsorbed oxygen on the silver surface, based on the assumption that oxygen atoms chemisorb onto the surface silver atoms at a 1:1 ratio (see Eq. S5 in the SI) [14]; (ii) oxygen that dissolved into the bulk silver during the oxidation step, with oxygen solubility taken to be 10 -6 mol O /mol Ag (see Eq. S4 in the SI) [14]; and (iii) the reduction of Ag 2 O to Ag 0 (see Eq. S6 in the SI). The amount of oxygen available from absorbed or dissolved oxygen is an order of magnitude smaller than the amount of oxygen released in a CLE cycle. Oxygen release from Ag 2 O reducing to Ag 0 can account for the oxygen released in CLE; however, the results presented in Fig. 4 show that Ag 0 does not reoxidise in CLE cycles. Taking 8 cycles CLE with the oxidation time of 15 min, the total amount of oxygen in the gas products adds up to ~ 1600 mol , exceeding the capacity of Ag 2 O, thus, confirming that SrFeO 3 was the source of the oxygen in CLE.
The measured CO 2 and EO concentration profiles associated with the results in Fig. 5 are presented in Fig. S5 in the SI. For each set of CLE cycles with decreasing oxidation step times of 15, 10, 5, and 1.5 min and a constant reduction step time of 1.5 min, the production of EO and CO 2 still stabilises in the third-fourth cycle. The experiment was repeated in the reverse order, shown in Fig. S6 in the SI. The production of EO and CO 2 was markedly reduced when the oxidation time ascended from 1.5 min to 15 min, showing that the operation history influences the catalyst performance.
In Fig. 6, the results from CLE experiments with a reduction time of 1.5 min and oxidation time between 1.5 and 15 min are further represented as the total oxygen released (Fig. 6A), the instantaneous conversion of ethylene and selectivity to EO (Fig. 6B), oxygen released in combustion (Fig. 6C) and epoxidation (Fig. 6D). The slow initial release of oxygen to both combustion and epoxidation, and thus, production of EO and CO 2 , is likely a result of the deconvolution model not adequately accounting for all dispersive and transport effects in the reactor system (see Fig. S1 in the SI for the effect of the deconvolution model). As illustrated in Fig. 6A, C and D, the point of maximum oxygen release to both products, EO and CO 2 , occurred later in the reduction step as the oxidation time increased. Considering the total oxygen release (Fig. 6A), the trends in experiments with 10 and 15 min oxidation times were very similar, changing only when the oxidation time was decreased to 5 and 1.5 min. Because the total release of oxygen is strongly affected by the oxygen released to the combustion products (see Eq. 4), similar behaviour is observed in Fig. 6C when analysing oxygen that participates in combustion. Clearly, the duration of the oxidation time affects the performance in the reduction step, but as seen in Fig. 6B and D, the effect is more pronounced for the epoxide product than for CO 2 . When the oxidation time lasted 10 or 15 min, the amount of produced EO increased, and the peak EO concentration appeared later in the cycle, translating, however, to a relatively less pronounced improvement of the instantaneous selectivity to EO. Finally, the instantaneous conversion of C 2 H 4 (Fig. 6B) decreased linearly throughout the reduction step for all oxidation times, with the overall decrease being more pronounced in experiments with shorter oxidation steps.
The pressure at which the oxidation step was performed was increased from 1 to 2.5 bar (absolute) for an oxidation time of 5 min; the resulting conversion to ethylene and selectivity towards EO are shown in Fig. 7. The oxidation time of 5 min was selected, as from Fig. 5, the conversion and selectivity achieved when operating CLE at 5 min oxidation time were notably lower than in experiments with 10 and 15 min oxidation times. Both conversion and selectivity increased approximately linearly with increased pressure, from 14 to 17% and 49 to 60%, respectively. Notably, the selectivity and conversion achieved at 5 min oxidation at 0.525 barO 2 gave a comparable performance to the 10 and 15 min oxidation at 0.21 barO 2 . Clearly, increasing the pressure in the oxidation step, and thus oxygen partial pressure, increased the rate of oxygen uptake and resulted in performances comparable to 1 bar CLE operated with longer oxidation times. Figure 8 shows the influence of reduction time on the conversion of ethylene, selectivity to EO, and total oxygen released. Here, the duration of the reduction step was varied, but the oxidation time was fixed at 15 min. Hence, Fig. 8 presents results where the ratio of reduction to oxidation times decreased from 20 (0.75 min reduction) to 5 (3 min reduction). Despite the change in the reduction time, the performance for EO production and ethylene conversion always stabilised around the third cycle. Transient results from the reduction steps are shown in Fig. S8, presenting that the instantaneous selectivity to EO and conversion of C 2 H 4 overlapped in all experiments. Because with time, the instantaneous selectivity slightly increases while conversion decays (see Fig. S8B), the averaged values in Fig. 8 are higher for selectivity and lower for conversion the longer the time of the reduction step. The most significant change was observed  Conversion of ethylene and selectivity towards EO achieved at various oxidation pressures during the last 3 cycles of CLE experiments consisting of 5 cycles in total. All experiments were conducted with reduction and oxidation times of 1.5 and 5 min, respectively. The oxidation step was carried out at 1, 1.5, 2, and 2.5 bar (abs). Conditions: 270 °C, 1 atm, 2 g of catalyst, feed of 200 mL min −1 (as measured at 293 K, 1 atm) in respect to the conversion of ethylene, which decreased from 24% in the 0.75 min reduction step to 13% in the 3 min reduction step. The associated oxygen release increased over the same range from 86 to 420 µmol O, while the selectivity to EO remained roughly constant.
The influence of the total cycle time is shown in Fig. 9. Here, the ratio of oxidation to reduction ( s red. ∕s ox. ) time was kept as 3 throughout, with reduction/oxidation steps of 5/15, 15/45, 45/135, and 90/270, all in seconds. Ethylene conversion varied slightly with the overall cycle times, while the selectivity to EO increased from 44 to 49% between the 15/45 and 45/135 experiments and remained constant between the 45/135 and 90/270 experiments. The amount of oxygen released doubled as cycle times doubled from 45/135 to 90/270 s red. ∕s ox. . Overall, the perovskite cycled stably, despite CLE operating at altered cycle times and with the oxidation to reduction ratio of 3, which is significantly lower than in the published studies, where the ratio was always kept at 10 [7-10].

Discussion
The nature of the oxygen species involved in the direct epoxidation of ethylene has been extensively studied but is still not completely understood. Many of the conclusions drawn from the direct epoxidation process have yet to be established for CLE, particularly since the oxygen is essentially supplied to the catalyst as lattice oxygen from the solid support  rather than gaseous oxygen. However, some previous computational studies on direct epoxidation have indicated that Ag nanoparticles may not be purely metallic under reaction conditions and that Ag 2 O may be an active phase participating in EO formation [15,16]. Although Ag 2 O is not expected to be thermodynamically stable at CLE conditions [17,18], its decomposition to Ag 0 is governed by slow kinetics in air at the CLE reaction temperature, 270 °C [19]. The possibility of oxygen being sourced from silver oxide during CLE was explored with Ag 2 O/Al 2 O 3, demonstrating that Ag 2 O is unselective for CLE, and its activity diminishes with cycles (see Fig. 4A). The conversion of ethylene to CO 2 , and therefore the release of oxygen from Ag 2 O, during the first cycle, resulted from Ag 2 O being reduced by ethylene to Ag 0 . While surface Ag 2 O(001) sites could be catalytically active in the direct epoxidation with O 2(g) [15], particles of Ag 2 O are in contrast unselective for ethylene epoxidation in CLE. The deactivation of Ag 2 O(15.9wt.%)/Al 2 O 3 after the first cycle demonstrates that Ag 2 O cannot be efficiently used for CLE.
For successful CLE experiments with Ag/SrFeO 3 , the oxidation step time must be sufficient to replenish the oxygen released over the reduction step. From the TGA experiments (Fig. 2), we see that reduction of Ag/SrFeO 3 is faster than reoxidation, and to regenerate the material at ambient pressure, the oxidation step needs to be ~ 3 times longer than the reduction step. Indeed, if the ratio of the oxidation to reduction time is kept at 3, the performance in CLE in the packed bed experiments is similar for short and long cycles (see Fig. 9).
Typically, the higher the ratio of the oxidation to reduction time, the better the CLE performance, namely the conversion of ethylene and selectivity to EO both increase (see Fig. 5). Selectivity is the critical performance parameter for selective oxidation reactions [2] because low EO yields can be improved by recycling the ethylene stream when the single-pass conversion is low. However, prolonging the oxidation step to increase the selectivity to EO also lowers the effective EO production time. Producing EO continuously in a CLE arrangement requires a number of reactors in parallel, with at least one operating in the reduction stage while the other reactors undergo regeneration. The number of reactors required for pseudo-steady EO production must be equal to or greater than t ox ∕t red . To illustrate the trade-off between improved selectivity and effective EO production time, consider the situation with a fixed reduction time of 1.5 min and altered oxidation times presented in Fig. 5. A 50% increase in the oxidation time from 10 to 15 min resulted in a 7% increase in selectivity. Whilst improved selectivity is desirable, the 50% increase in oxidation time decreases the effective EO production time by 30%, requiring at least 4 additional reactors to operate with continuous EO production. CLE processes should thus be designed to maximise both the effective EO production time and selectivity to EO. Increasing the pressure during the oxidation step (Fig. 7) was shown to be a promising method for increasing the effective production of EO whilst maintaining high EO selectivity. For the presented arrangement, with a reduction step of 1.5 min and a 5 min reoxidation step at 2.5 bar, 4 reactors operating in parallel would be required for continuous production of EO in a CLE arrangement. Scaling the number of CLE reactors would multiply the amount of catalyst required and the capital costs for building more reactors. Silver catalysts supported on Al 2 O 3 are currently used industrially for ethylene epoxidation at similar loadings to the one used here (i.e., ~ 15-20 wt%) [1,2]. Also, the manufacturing of the oxygen carrier, SrFeO 3 , is relatively straightforward, consisting of a two-step procedure of ball-milling SrCO 3 and Fe 2 O 3 precursors, followed by calcination [11]. Therefore, the expected cost of Ag/SrFeO 3 should be low and comparable with the currently used catalysts. Hence, the capital cost of multiple reactors and the operating costs, which increase when operating at elevated pressures, would be the deciding factors in an optimisation exercise. A techno-economic analysis will be essential to determine the most profitable CLE configuration.
Results in Fig. 5 show that prolonged oxidation time is essential for improving selectivity, conversion and the total oxygen release in the reduction step, but if extended beyond 10 min, further improvement is minimal. Interestingly, when considering the oxygen release profiles in Fig. 6, only oxygen release to the selective epoxidation reaction increased whilst oxygen release to the combustion reaction remained constant. The negligible change in the oxygen release between experiments with 10 or 15 min oxidation steps indicates that 10 min was sufficient to regenerate SrFeO 3 , replenishing the oxygen released over a 1.5 min reduction in ethylene. The observed improvements in selectivity and conversion between 10 and 15 min, in Fig. 5, indicate that the starting chemical potential of the perovskite is the key parameter for very good performance, alongside possible other phenomena occurring during the oxidation step, such as the removal of carbonaceous species, which can block selective Ag sites, oxygen regrouping in SrFeO 3 [20], or changes to the surface electronic state.
Increasing the reduction time to 3 min ( Fig. 8 and Fig. S8 in the SI) resulted in an approximately constant average selectivity, although at decreasing activity (lower conversion). Taking selectivity as the critical parameter means that longer reduction times are favourable for CLE operation as they increase the effective EO production time. The maximum possible reduction time achievable with 2 g of Ag(15 wt.%)/SrFeO 3 was determined by performing a prolonged reduction experiment, using 5.4 vol% C 2 H 4 in N 2 mixture. As presented in Fig. 10A, the measured CO 2 and EO concentrations reached peak values about 1 min after introducing the ethylene mixture; afterwards, concentrations of both products decayed. The EO production ceased after ~ 4.2 min, whereas CO 2 production tailed off to ~ 40 min, indicating long-drawn extraction of oxygen from SrFeO 3 leads to complete combustion of ethylene. Thus, the reduction step in the proposed experimental arrangement for CLE should be kept below 4.2 min. During the prolonged reduction in ethylene, the total amount of oxygen released was 0.0018 mol O , corresponding to SrFeO 2.82 reducing to SrFeO 2.65 . Interestingly, SrFeO 3 reacts for much longer when exposed to EO; a CO 2 profile from a prolonged reduction of 2 g Ag(15 wt.%)/SrFeO 3 in 1000 ppm EO mixture is presented in Fig. 10B. Here, CO 2 was measured for over 160 min, giving the total amount of oxygen release of 0.029 mol O , corresponding to SrFeO 2.82 reducing to SrFeO 2.55 . Taking that Ag(15 wt.%)/SrFeO 3 reduces more eagerly in EO than in C 2 H 4 , the CO 2 in Fig. 10A might have originated to some degree from SrFeO 3 combusting the EO product.

Conclusions
This work provides insight into the effect of the duration of reduction and oxidation steps on the selectivity to EO and conversion of ethylene in chemical looping epoxidation of ethylene. Both parameters were shown to improve with longer oxidation times; however, the changes were significant when the oxidation time increased from 1.5 to 10 min. If the oxidation step exceeded 10 min, further improvements in the CLE performance were very small. We conclude that shortening the oxidation step from 15 to 10 min will increase the effective production time of EO without severely impacting selectivity or conversion. The effective time of EO production can be further improved by increasing the pressure of the oxidation step; a 5 min oxidation step at 2.5 bar resulted in the selectivity to EO and conversion of C 2 H 4 comparable to a 15 min oxidation step at 1 bar. Increasing the reduction time from 1.5 to 3 min with a 15 min oxidation time resulted in decreased conversion of C 2 H 4 , but stable selectivity to EO. Longer reduction times increase the effective production time of EO; however, a practical time limit should be determined. For example, when working with a 2 g-batch of Ag(15 wt.%)/ SrFeO 3 , the epoxide product was measured for ~ 4.2 min. Finally, Ag 2 O was found unselective in CLE, with Ag 2 O/Al 2 O 3 combusting ethylene to CO 2 in the first cycle, with no capability to regenerate the oxide in the oxidation step. Metallic silver was thus confirmed as the superior catalyst for CLE.
Data availability A set of all collected experimental and supporting data for this work can be found at https:// www. repos itory. cam. ac. uk.

Code availability
No software was developed in this research.

Competing interests The authors declare no competing interests.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.