Solar-driven co-thermolysis of CO2 and H2O promoted by in situ oxygen removal across a non-stoichiometric ceria membrane

First-time experimental demonstration of simultaneous thermolysis of CO2 and H2O in a solar-driven membrane reactor.


Introduction
The utilization of the vast solar energy resource for electricity, heat, and fuels has become a key objective in research and development. 1 The conversion and storage of solar energy in fuels is especially appealing as a means to transition from fossil fuels to a "CO 2 economy". 2 With this, a solar refinery needs to be developed where solar energy is collected and used to convert CO 2 and H 2 O to fuels by some method. Existing research spans solar-driven electrochemical, photoelectrochemical, and photocatalytic paths for direct conversion, as well as indirect routes via the solar thermochemical production of syngas (H 2 and CO). 3 Solar thermochemical redox cycles utilize the entire spectrum of solar radiation concentrated to high-temperature process heat to drive the splitting of CO 2 and H 2 O and produce CO and H 2 at high rates, selectivity, mass conversions, and efficiencies. [4][5][6] However, the temperature swing required between the redox steps induces significant material stresses and energy irreversibility, which prompted the search for alternative isothermal processes. [7][8][9] One promising approach is the use of a dense, ceramic, mixed ionic-electronic conducting (MIEC) membrane for the continuous separation of oxygen and fuel (H 2 and/or CO) derived from the thermolysis of CO 2 and H 2 O at high temperatures, as pioneered for solar water splitting by Fletcher and coworkers. 10,11 We recently demonstrated the proof-of-concept utilization of a solar-driven membrane reactor for splitting of CO 2 . 12 Other investigations of thermochemical membrane reactors, both theoretical and experimental, have also only focused on either CO 2 -or H 2 O-splitting. 10,[13][14][15][16][17][18] This work goes further and demonstrates the feasibility of co-feeding both CO 2 and H 2 O and assesses the relative favorability between the two thermolysis reactions occurring simultaneously. The desired dissociations are chemical equilibrium reactions in the gas phase described by: The reactions are analogous; that is, both are endothermic and thermolytic, but their reaction energetics differ. This is described by the standard Gibbs free energy changes at equilibrium (ΔG = 0): where G i  is related to the equilibrium constant K i which in turn is a function of temperature only. Fig. 1a  O are fed to the inner side of a capped tubular non-stoichiometric ceria membrane. Ceria has become the benchmark material for oxygen-cycling applications due to its stability and fast kinetics. [22][23][24] In our previous work it was also found to be an effective material for oxygenconducting membranes. 12 The supply of concentrated solar process heat at high temperatures drives the thermolysis, producing CO, H 2 , and O 2 . The latter adsorbs on the inner membrane surface, dissociates, and is transported across the membrane in an ionic form along a chemical potential gradient. O 2− then associates into O 2 at the outer membrane surface and desorbs into an inert sweep gas contained in a shell tube. This in situ removal of one of the reaction products drives the reactions forward towards dissociation and avoids downstream recombination. The counter-flow configuration of the reactant and product gases favorably maximizes the gradient of p O 2 along the length of the membrane. By placing this reactor in a solar cavity receiver, the high-temperature heat for the reactions is provided by concentrated solar radiation incident on the shell tube.
Typical operating conditions require temperatures around 1773 K and partial pressures of O 2 down to 1 Pa. These high temperatures eliminate the need for catalysts but pose significant constraints on the construction materials which must withstand these conditions over extended periods of time. Materials must also resist thermal shock that may occur due to cooling overnight, unless the reactor is equipped with an alternative heat source such as a hightemperature thermal energy storage system. 25 A modular tubular membrane design could avoid costly maintenance by allowing for simple replacement of degraded membranes. Maintaining low partial pressures of oxygen is crucial, requiring additional energy for vacuum pumping or gas separation to regenerate the inert sweep gas (such as N 2 , though here we use Ar for gas analytic considerations). 26,27 Alternatively, some studies have reported solar-driven production of pure O 2 and inert gas with low partial pressures of oxygen using thermochemical oxygen pumps driven by low-grade process heat. [27][28][29] Experimental Materials CeriumĲIV) oxide (ceria, CeO 2 , powder, particle size <5 μm, 99.9% purity), polyĲoxy-1,4-phenylene sulfonyl-1,4-phenylene) (PES, (C 12 H 8 O 3 S) n , pellets), polyvinylpyrrolidone (PVP, (C 6 H 9 NO) n , powder, average M.W. 40 000), and 1-methyl-2pyrrolidinone (NMP, C 5 H 9 NO, liquid, ≥99.0% purity) were from Sigma Aldrich. Al 2 O 3 membranes (Alsint 99.7, 7 mm outer diameter (OD), 5 mm inner diameter (ID), 250 mm length) were from Intertechno-Firag AG. High-purity alumina adhesive (Aremco Ceramabond 569) and glass-filled

Membrane fabrication
Capped tubular ceria membranes were produced as reported previously. 12 Briefly, membranes were fabricated using a phase-inversion/sintering method. 30,31 Two polymers, namely PES (5.7 wt%) and PVP (0.5 wt%), were dissolved in NMP (22.0 wt%). Ceria powder (71.8 wt%) was suspended in the polymer solution. The ceria slurry was coated onto membrane templates (highly flexible silicone tubing, 3 mm ID, 7 mm OD, RCT Reichelt Chemietechnik GMbH & Co.), which were placed in a water bath for phase inversion (unfiltered tap water coagulant under ambient conditions). The silicone templates were removed, and the dried membrane precursors were then sintered for 8 hours at 1873 K (oven model HTL 20/17, ThermConcept). The sintered membranes were 6-7 mm OD, 5-6 mm ID, and 150-250 mm in length. Typically, the membrane walls were about 0.5 mm thick. In-depth solid-state characterization of ceria membranes before and after use in the reactor was performed previously. 12

Experimental setup
The solar membrane reactor system is depicted schematically in Fig. 2. CO 2 and/or H 2 O was fed through a feeder tube into the inner side of the membrane while Ar sweep gas was fed into the reactor shell tube in a countercurrent flow. This assembly was placed in a thermally insulated solar cavity receiver with an aperture of 4 cm in diameter. A compound parabolic concentrator (CPC) was incorporated into the aperture to boost the solar flux concentration and generate a more uniform directional distribution of concentrated radiation entering the cavity. 32 Experimentation was performed using the high-flux solar simulator (HFSS) of ETH Zurich: an array of seven Xe arcs, close-coupled to truncated ellipsoidal reflectors, provided an external source of intense thermal radiation that closely approximated the heat transfer characteristics of highly concentrating solar energy facilities. The radiative flux distribution at the focal plane was measured optically using a calibrated CCD camera focused on a Lambertian (diffusely reflecting) target. The solar radiative power input to the cavity was calculated by integration of the radiative flux over the aperture area and verified with a water calorimeter. Temperatures were measured at the outer surface of the reactor shell at two heights along the tube (indicated in Fig. 2) using B-type thermocouples. Gas flow rates were regulated using electronic mass flow controllers (MFC, Bronkhorst F-201 C, accuracy 0.5% Rd + 0.1% FS), whereas steam flow was generated using a liquid flow controller (LFC, Bronkhorst Liqui-Flow L23-AAD-33-K-305, accuracy 1% FS) and steam generator (Bronkhorst CEM W-202A-333-K). The product gas composition in each stream was monitored on-line by gas chromatography (GC, Agilent 490 MicroGC).

Experimental runs
All volumetric flow rates are given under standard conditions (1 bar and 273 K). The HFSS heated the reactor to the desired temperatures in the range of 1723-1873 K with a radiative power input of 2.5-3.0 kW. For water-splitting experiments, steam was fed to the inner side of the membrane at a rate of 104 mL min −1 (5 g h −1 water to steam generator), carried in by 80-100 mL min −1 Ar. For co-feed experiments, the same flow of steam was carried by 75-200 mL min −1 CO 2 . At the outer side of the membrane, the reactor shell was purged with 200-1000 mL min −1 Ar. The compositions of both gas streams exiting the solar reactor were analyzed simultaneously using two gas chromatographs (GC). Steady state was defined as the condition at which the measured gas concentration was within 2% of the mean over the previous five consecutive measurements collected at a frequency of one every two minutes: where c i Ĳt j ) is the volumetric concentration of species i at time point j. Steady-state data were collected for at least 19 minutes and the arithmetic mean was used to summarize the results under each experimental condition.

Thermodynamic analysis
We calculated the thermodynamic equilibrium limits of thermolysis of CO 2 and H 2 O in a membrane reactor to compare to experimental results. Relatively fast rates are expected for each reaction step: gas-phase thermolysis, heterogeneous surface reaction, and oxygen bulk diffusion. When each of these serial processes is sufficiently fast, global kinetics are fast, and thermodynamics govern the net reaction. In this case, kinetics can be neglected. Previous observations with a solar cavity receiver containing a porous ceria structure directly exposed to highflux irradiation reported that the overall kinetics are not controlled by solid-state diffusion within the crystal lattice. 33 This is also expected for a ceria membrane because the measured values of ambipolar diffusion coefficients of oxygen in ceria (1.5 × 10 −5 -4 × 10 −4 cm 2 s −1 in the range of 1673-1823 K 22 ) translate to diffusion times in the order of seconds for the length scales across the 0.5 mm-thick membrane. Thus, as far as solid-state diffusion is concerned, the transport of oxygen vacancies through the membrane is almost instantaneous compared with the time scales of data collection.
Reaction rates have an exponential dependence on temperature, scaling with expĲ−E A /RT), as seen in the Arrhenius equation. The high temperature in the range of 1723-1873 K and consequent high activity of reactive oxygen vacancies at the surface of the membrane are expected to lead to fast sur-face exchange of oxygen from the gas into the solid phase. 19 While studies of non-isothermal processes show that the heating rate limits the oxygen release rate, heat transfer should not be limiting in this isothermal process because the heat of the reaction is much lower than the heat input. 23,33 Likewise, the high temperature, along with the catalytic effect of ceria, implies very fast thermolysis reactions in the gas phase. 34 Therefore, a purely thermodynamic model is expected to adequately predict reactor performance.
However, to reach the thermodynamic limit in the countercurrent-flow reactor, there must be a sufficient membrane area and sweep gas flow rate relative to the flow rate of the reactant. To account for the oxygen capacity in a given flow of sweep gas, a thermodynamic model described by Bulfin was applied, which is specific to countercurrent-flow reactors. 35 This approach guarantees compliance with the second law of thermodynamics and conservation of mass along the entire reactor by means of a dimensionless oxygen exchange coordinate, κ, defined as the number of moles of O 2 crossing the membrane up to a certain point along the length, x, per mole of oxidant fed: where ṅ oxidant is the molar flow rate of H 2 O and/or CO 2   n n Bulfin detailed the full methodology in an example tailored to thermolysis in a membrane reactor with a counter-current sweep flow. 35 The countercurrent-flow thermodynamic model was implemented and solved numerically with Matlab using thermodynamic data from NIST JANAF. 36 The model input parameters are T, p, p O 2 , the relative flow of sweep to oxidant (ṅ sweep /ṅ oxidant ), and the relative flow rates of CO 2  reaction temperature) and a simple heat transfer model were used to estimate the reaction temperature. Consequently, this analysis presents the thermodynamic limits as a shaded region, where the upper and lower bounds are the limits at the maximum measured shell temperature and the estimated reaction temperature, respectively. In the range of operation, the difference between these temperatures is approximately 50 K. The simple heat transfer model used to calculate this temperature difference is described in the ESI. † The outputs of the thermodynamic model are the partial pressures of the products and conversion of the reactants at equilibrium. These experiments used a roughly equimolar feed of CO 2 and H 2 O and produced fuel with a relatively higher fraction of CO than H 2 . This product proportion is consistent with the more favorable change in Gibbs free energy for thermolysis of CO 2 at high temperatures, as seen in Fig. 1a. The average molar ratio O 2 : fuel over 19 experiments was 0.53 ± 0.07, corroborating a closed mass balance. Furthermore, no other by-products were detected, indicating the absence of undesired side reactions. Specifically, Fig. 3a shows production rates over time at steady state at a nominal T of 1873 K and p O 2 of 0.4 Pa. Note that T refers to the maximum measured shell temperature and p O 2 refers to the partial pressure of O 2 at the Ar inlet, which is the minimum p O 2 in the system and determined by the impurity in the sweep gas. Gas evolution rates leveled off at a constant incident radiative flux of about 3500 suns (1 sun = 1 kW m −2 ) over the cavity's aperture, and thus the reactor demonstrated continuous, steady-state operation under isothermal conditions. Fig. 3b shows the steady-state average production rates as a function of T in the range of 1723-1873 K. Mass flow rates were kept constant at 5 g h −1 H 2 O with 100 mL min −1 CO 2 to the inner side and 500 mL min −1 Ar to the outer side of the membrane (L denotes standard liters). The measured p O 2 ranged from 0.4-1.7

Results and discussion
Pa due to variation in the small amount of air leakage into the reactor during different experimental runs. The specific fuel and O 2 production rates at steady state increased with temperature, in accordance with the thermodynamic dependence on expĲ−ΔG°/RT). Finally, Fig. 3c shows the steady-state average production rates as a function of p O 2 in the range of 0.2-0.9 Pa at 1873 K. Although the range of p O 2 tested was small, the gas production rates indeed decreased at higher p O 2 , as expected from the equilibrium relationship for the thermolysis of CO 2 and H 2 O described in eqn (3) and (4). However, to achieve this range of p O 2 , the flow rate of Ar was adjusted, thus changing the relative flow of sweep gas, ṅ sweep /ṅ oxidant , which is also a key thermodynamic parameter in this reactor system. Fig. 4 shows the experimental steady-state conversion of CO 2 to CO and H 2 O to H 2 as a function of (a) T in the range of 1723-1873 K at p O 2 = 1 Pa, ṅ sweep /ṅ oxidant = 2.4, and (b) ṅ sweep /ṅ oxidant ranging from 1-5 at T = 1873 K, p O 2 = 0.5 Pa. The thermodynamic limits are also indicated, denoted by a band whose upper and lower bounds are calculated as described above at T shell and the estimated reaction temperature, T shell − 50 K, respectively. The band is wider for CO 2 than that for H 2 O because its reaction favorability changes more steeply with T. Like the gas production rates, the conversion of reactants shown in Fig. 4a increased with T, as expected. The conversion also increased at higher relative sweep rates shown in Fig. 4b, because ṅ sweep /ṅ oxidant determines the total amount of O 2 that can be removed across the membrane at equilibrium. In fact, the thermodynamic analysis reveals that the trend in production rates observed in Fig. 3c is more attributable to varying ṅ sweep /ṅ oxidant than p O 2 . The p O 2 in the sweep gas must be lower than p O 2 in the oxidant stream at all points along the membrane to drive transfer of O 2 . At sufficiently low p O 2 , however, this parameter does not have a strong influence on the conversion of reactants, shown in Fig. S3 in the ESI. † In the range of p O 2 observed in the experiments (0.2-0.9 Pa), the theoretical conversion of each reactant at constant ṅ sweep /ṅ oxidant is almost flat, while the experimental data points exhibit a trend due to varying relative sweep rates. In contrast, the same experimental data plotted against ṅ sweep /ṅ oxidant shown in Fig. 4b match the shape of the equilibrium limit curves.

View Article Online
In general, the results indicate that the reactor performance indeed approaches the thermodynamic limit for a countercurrent flow reactor. Importantly, the experimental conversion does not exceed the theoretical limit. Furthermore, the experimental points lie closer to the lower bound of the equilibrium region, suggesting that the simple heat transfer model is necessary and effective to estimate the reaction temperature. While the experimental conversion of H 2 O closely follows the lower bound of the predicted thermodynamic limit, the experimental conversion of CO 2 falls short. The discrepancy is less than a factor of two and may be a result of the water-gas shift (WGS) reaction occurring at lower T downstream of the reactor, before the GC analysis. The WGS consumes some CO to produce additional H 2 and is thus the difference of the CO 2 and H 2 O dissociation reactions (eqn (2) minus eqn (1)): CO + H 2 O ↔ H 2 + CO 2 (7) In this case, the GC measurements may not be representative of the composition in the reactor; the actual conversion of CO 2 may have been higher, and the conversion of H 2 O lower, potentially equalizing the difference seen in Fig. 4 between the experimental data and the limit for each gas. Fig. 5 illustrates the effect of the molar feed ratio, CO 2 : H 2 O, on the: (a) steady-state average production rates of fuel and O 2 , (b) conversions of CO 2 and H 2 O, and (c) molar ratio of the two fuels produced, CO : H 2 . Fig. 5b and c show the comparison of the results measured experimentally with those predicted from thermodynamics. The relative flow rates of CO 2 and H 2 O were varied while maintaining the steam feed rate at 5 g h −1 H 2 O, temperature at 1873 K, and p O 2 at 0.5 Pa. Fig. 5a shows that the overall fuel production (sum of H 2 and CO) increased with CO 2 : H 2 O, as did the production rate of CO, which occurred because the total feed rate and CO 2 feed rate both increased. The production rate of H 2 , on the other hand, decreased with CO 2 : H 2 O. It must be emphasized that ṅ sweep /ṅ oxidant did not remain constant over experimentation, but rather decreased with CO 2 : H 2 O because the flow rate of sweep gas remained constant. A constant flow rate of sweep gas is less effective at maintaining low p O 2 as the amount of O 2 to be removed increases. In the case of CO, the effect of increasing CO 2 feed rate compensated for the  decreasing sweep ratio, so that the net production rate increased. However, the H 2 production decreased. The confounding factors of changing both the total feed rate and the relative sweep rate are accounted for by plotting conversion instead of the production rate in Fig. 5b. As seen in Fig. 4, the absolute values of H 2 O conversion in Fig. 5b again match the model results better than those for CO 2 . The experimental conversions of both CO 2 and H 2 O decreased slightly with increasing CO 2 : H 2 O, in agreement with the trend predicted at equilibrium. There are two contributions to the negative trend in conversion. First, as already mentioned, ṅ sweep / ṅ oxidant decreased with CO 2 : H 2 O, which decreased the conversion. However, the conversion of each reactant is predicted to decrease slightly with CO 2 : H 2 O even with a constant ṅ sweep / ṅ oxidant . The second reason for the trend is related to the difference in favorability of thermolysis of CO 2 and H 2 O. As the feed ratio increases, a higher proportion of the feed is CO 2 , which has a higher conversion than H 2 O under these conditions. In fact, the overall conversion of reactants to products actually increased slightly with increasing CO 2 : H 2 O. However, the O 2 capacity of the sweep gas was unchanged and therefore the conversion of each individual reactant must decrease to balance the production of O 2 with its removal. This result indicates that a feed of CO 2 requires a higher relative sweep rate than an equal feed of H 2 O, because its higher potential conversion leads to a larger amount of O 2 that must be removed.
Consistent with Fig. 5a, Fig. 5c shows that the product ratio CO : H 2 increased with increasing feed ratio CO 2 : H 2 O, as expected intuitively. Interestingly, CO : H 2 is always greater than the corresponding CO 2 : H 2 O, which further confirms that dissociation of CO 2 is more favorable than that of H 2 O under equivalent conditions. The observed experimental trend qualitatively matches equilibrium thermodynamics, though with a smaller slope, because the experimental conversion of CO 2 is lower than predicted. In consideration of downstream processing, the Fischer-Tropsch synthesis favors a syngas feed with 1 : 2 moles CO : H 2 . 37,38 According to Fig. 5c, the product ratio can be adjusted via the feed ratio, and a 1 : 2 product ratio would require a feed ratio smaller than the minimum tested here.
The maximum conversions observed experimentally were 1.0% CO 2 and 0.4% H 2 O at 1873 K, ṅ sweep /ṅ oxidant = 5, and 0.2 Pa O 2 (0.7% overall conversion of reactants). In general, the absolute values of both theoretical and experimental conversion were lower in the co-feed case tested here than in the pure-CO 2 feed case tested previously, 12 because the relative sweep rates were lower in this set of experiments. The base case ṅ sweep /ṅ oxidant was 8 in pure-CO 2 experiments and 2.4 in these co-feed experiments. The relative sweep rate is a significant parameter for sweep gas operation, and the application of the countercurrent flow model was essential to accurately predict the behavior of the reactor. Furthermore, although there was no effort to optimize the efficiency of the reactor in these proof-of-concept experiments, ṅ sweep /ṅ oxidant also impacts efficiency because it determines how much sweep gas must be heated and circulated per unit fuel produced.

Conclusion
We have demonstrated steady-state splitting of a mixed feed containing CO 2 and H 2 O into separate streams of syngas fuel and O 2 using an isothermal tubular ceria membrane reactor driven by simulated concentrated solar radiation. The experimental results generally agreed with trends predicted by thermodynamics. The conversion of CO 2 to CO was favored over H 2 O to H 2 , consistent with the energetics of the respective thermolysis reactions. The co-thermolysis of a mixture of CO 2 and H 2 O is more complex than feeding either CO 2 or H 2 O separately to the reactor, both experimentally and in the theoretical analysis. In the co-feed case, the mixture of CO 2 , H 2 O, CO, and H 2 could undergo additional reactions, especially the reverse water-gas shift (RWGS). As seen in eqn (7), the WGS reaction is not independent of the two thermolysis reactions, and thus does not change the equilibrium conversion of each reactant from the pure-feed values. Other possible reactions were found to be negligible from the lack of byproducts predicted by thermodynamics and measured experimentally under these conditions. The absence of by-products and a 2 : 1 ratio of fuel : oxygen together confirmed 100% selectivity for the desired splitting reactions.
The solar thermochemical membrane reactor unifies both CO 2 and H 2 O splitting in a single modular and scalable device and offers a technically viable pathway to single-step syngas production. However, determining an appropriate relative sweep rate is challenging in co-feed operation because the different favorability for thermolysis of CO 2 and H 2 O implies different optimums for each species. In addition, these energetic differences mean that H 2 O needs to be fed in large excess to achieve a desirable syngas composition. Therefore, it may still be attractive to produce CO and H 2 separately and mix them into syngas as needed. Furthermore, the singlestep approach incorporated in the membrane reactor must compete with multistep cycles currently available. Thus, further R&D and alternative membrane configurations are needed to boost mass conversions and consequently reach favorable solar-to-fuel energy efficiencies, a challenge because T and p O 2 determine the thermodynamic limits.

Conflicts of interest
There are no conflicts to declare.