Surface Chemistry of Perovskite-Type Electrodes During High Temperature CO2 Electrolysis Investigated by Operando Photoelectron Spectroscopy.

Any substantial move of energy sources from fossil fuels to renewable resources requires large scale storage of excess energy, for example, via power to fuel processes. In this respect electrochemical reduction of CO2 may become very important, since it offers a method of sustainable CO production, which is a crucial prerequisite for synthesis of sustainable fuels. Carbon dioxide reduction in solid oxide electrolysis cells (SOECs) is particularly promising owing to the high operating temperature, which leads to both improved thermodynamics and fast kinetics. Additionally, compared to purely chemical CO formation on oxide catalysts, SOECs have the outstanding advantage that the catalytically active oxygen vacancies are continuously formed at the counter electrode, and move to the working electrode where they reactivate the oxide surface without the need of a preceding chemical (e.g., by H2) or thermal reduction step. In the present work, the surface chemistry of (La,Sr)FeO3-δ and (La,Sr)CrO3-δ based perovskite-type electrodes was studied during electrochemical CO2 reduction by means of near-ambient pressure X-ray photoelectron spectroscopy (NAP-XPS) at SOEC operating temperatures. These measurements revealed the formation of a carbonate intermediate, which develops on the oxide surface only upon cathodic polarization (i.e., under sufficiently reducing conditions). The amount of this adsorbate increases with increasing oxygen vacancy concentration of the electrode material, thus suggesting vacant oxygen lattice sites as the predominant adsorption sites for carbon dioxide. The correlation of carbonate coverage and cathodic polarization indicates that an electron transfer is required to form the carbonate and thus to activate CO2 on the oxide surface. The results also suggest that acceptor doped oxides with high electron concentration and high oxygen vacancy concentration may be particularly suited for CO2 reduction. In contrast to water splitting, the CO2 electrolysis reaction was not significantly affected by metallic particles, which were exsolved from the perovskite electrodes upon cathodic polarization. Carbon formation on the electrode surface was only observed under very strong cathodic conditions, and the carbon could be easily removed by retracting the applied voltage without damaging the electrode, which is particularly promising from an application point of view.

Introduction

In view of environmental problems such as global warming, it is one of the main scientific and technological challenges to improve possibilities for reducing emission of CO2 while concurrently meeting the continuously rising energy demand of the population.[1,2] As a consequence, sustainable energy sources such as solar, wind, or hydroelectric power experience a strong upsurge. However, most of the renewable energy sources suffer from nonhomogeneous availability in both temporal and geographical senses. For a global change toward a carbon-neutral energy supply, the storage and distribution of sustainably produced energy is thus inevitable.

Electrolysis is a highly promising technology for fulfilling this need for energy storage, since it allows the transformation of energy from a volatile source into a persistent form. Though electrolysis is currently often highlighted as a novel and innovative method for energy storage, its most popular example, the electrochemical splitting of water into hydrogen and oxygen, is already known since more than 200 years.[3] One of the main limiting factors for a widespread use of hydrogen as an energy carrier is its nontrivial storage[4] and the associated need to establish a new distribution infrastructure, which makes the change toward a hydrogen-powered society economically challenging. From a storage and distribution point of view, hydrocarbons offer a couple of advantages over hydrogen such as high volumetric energy density, safe and easy storability, and compatibility with existing fuel infrastructure.[4,5] One way out of this dilemma is hydrocarbon formation from H2O and CO2, which uses electrochemically produced hydrogen together with carbon dioxide via a methanation reaction (CO2 + 4H2 → CH4 + 2H2O). This is commonly referred to as a power-to-gas (P2G) process. Another technologically relevant process for the synthesis of hydrocarbons is the Fischer–Tropsch process (e.g., (2n + 1)H2 + nCO → CH2 + nH2O), which, in contrast to the direct hydration of CO2, is already industrially well-established. In current industrial large scale reactors, however, P2G methanation as well as Fischer–Tropsch and similar processes utilize locally available mixtures of CO2, CO, and H2, whereby the stoichiometric CO/H2 ratio for the respective product needs to be established individually via a preceding reverse water–gas shift step (rWGSR: CO2 + H2 → CO + H2O).[6]

In contrast to this chemically driven CO2 reduction, an electrochemically performed CO2 splitting offers a couple of advantages: (i) No additional reduction agent such as H2 is needed, which offers the possibility of higher efficiencies. (ii) Owing to the absence of both reducing agent (H2) and byproduct (H2O), no blocking of active sites by these species can occur, and the entire catalyst surface is available for CO2 reduction. (iii) Contamination of the produced CO by the second product (H2O in case of chemical reduction, O2 in the electrochemical case) can easily be prevented via separating the gas compartments of working and counter electrodes. Electrolysis of CO2 and H2O thus offers a very attractive opportunity for avoiding the additional rWGSR step, allowing control of the final H2/CO ratio (slightly above 2:1 for Fischer–Tropsch and methanol synthesis, 3:1 for synthetic methane) and, at the same time, utilizing excess renewable electricity. Consequently, electrolysis of CO2 and H2O, performed either separately or together as coelectrolysis, is highly attractive for running both power-to-gas and power-to-liquid processes using renewable resources with high efficiency. With sustainably produced CO and H2 (commonly referred to as syngas) these processes thus offer the possibility of storing excess renewable energy in a form that is compatible with the existing fuel infrastructure. A further, though somewhat exotic, application of CO2 electrolysis is the production of storable fuel as well as oxygen during space missions such as a long-term human exploration of planet Mars.[7,8]

Among the available types of CO2 electrolyzers, the solid oxide electrolysis cell (SOEC) is the most promising one.[9] Owing to entropy reasons, its high operation temperature in the range of about 800 °C leads to a significant reduction of the cell voltage and thus to an increased efficiency.[5,10] From a kinetic point of view, carbon dioxide splitting in solution based electrolyzers is challenging owing to the very stable bonds in the CO2 molecule[11] and the limited CO2 solubility in aqueous electrolytes.[12] Moreover, CO2 electrolysis cannot be done in proton exchange membrane (PEM) cells, since especially the product CO poisons the fuel electrode catalyst by very strong adsorption.[13] Also, in this respect, the high operation temperature of SOECs and the use of gas phases as reactants are highly advantageous. Typical electrode (SOEC cathode) materials for CO2 reduction are Pt/YSZ,[7,14] Ni/YSZ,[10,15−18] and ceria based mixed conductors,[19−22] and perovskites such as (La,Sr)(Cr,Mn)O3δ[22−24] or donor doped SrTiO3.[25] Most of the existing studies on high temperature CO2 electrolysis were performed on porous electrodes in symmetrical cells or entire SOECs, aiming at an increase of the maximum current density.[7,10,14−25] In order to develop and improve efficient SOEC cathodes, a more detailed fundamental understanding of the kinetic processes and the mechanisms behind cathodic CO2 reduction or CO2 and H2O co-reduction is required. However, studies on model systems with the goal of obtaining a detailed mechanistic understanding of the involved elementary processes are rare.[26−28] For gaining such an in-depth understanding of the surface chemistry of operating electrodes, the combination of electrochemical techniques and operando X-ray photoelectron spectroscopy (XPS) is a very powerful tool and thus has attracted much attention in recent years.[26,27,29−36] However, for CO2 electrolysis again only very few studies exist. Those deal with model-type ceria based electrodes[26,27] while comparable mechanistic investigations on perovskite-type SOEC cathodes have, to the best of our knowledge, not been published so far.

Perovskite-type electrode materials could be a very attractive alternative to ceria based electrodes for CO2 reduction since they offer a much larger compositional diversity and thus many options for optimizing the electrode stability, ionic and electronic conductivity, electrochemical properties, and catalytic activity.[37−41] In particular, the latter can be further affected by introduction of reducible transition metals, which may exsolve under reducing conditions, forming metallic particles on the perovskite surface.[38,41−47] In a recent study we could show by simultaneously performing electrochemical polarization experiments and near-ambient pressure XPS (NAP-XPS) measurements that the exsolution of metallic iron particles from (La,Sr)FeO3δ electrodes strongly improves their water splitting kinetics.[36] It is an open question whether surface decoration of perovskite electrodes by exsolved metallic particles also improves CO2 electrolysis.

In the present study we investigated the surface chemistry of perovskite-type oxides during carbon dioxide electrolysis by means of operando NAP-XPS measurements on La0.6Sr0.4FeO3δ and (La,Sr)(Cr,Ni)O3δ thin film model electrodes with different cation composition. The CO2 reduction behavior of lanthanum ferrite based electrodes can be compared with their H2O reduction properties reported recently.[36,48] The chromite based oxides were chosen since they offer the possibility of a reduction stable backbone[49−52] and exsolve Ni without the risk of decomposition as in the case of La0.6Sr0.4FeO3δ.[53] NAP-XPS revealed the voltage dependent evolution of different carbon species on the surface of the electrodes. Under cathodic polarization a carbon species at relatively high XPS binding energies (ca. 290 eV) can be observed, and its evolution is discussed in light of a possible reaction mechanism: A bidendate (CO3)•3– adsorbate is suggested as the decisive intermediate of CO2 activation. The formation of this surface species requires oxygen vacancies as well as electrons in the acceptor doped oxide electrode. Hence, perovskite-type oxides with high vacancy and high electron concentration might be particularly attractive for CO2 electrolysis. As a second reduction process, the development of graphitic carbon was investigated, and it is shown that any carbon deposits can be removed by retracting the cathodic polarization, thus allowing for a complete recovery of the electrode performance. These insights into the relationships of surface chemistry and electrochemical performance not only provide a large step toward an in-depth understanding of CO2 electro-reduction on perovskite-type oxides, but also provide a valuable basis for the future optimization of porous SOEC cathodes for CO2 electrolysis, since the surface process on mixed conducting oxides is usually responsible for the largest contribution to the polarization resistance of “real” porous electrodes.

Experimental Methods

Preparation of Materials

Three different compositions of chromite based perovskites were examined, see Table . Those were prepared as follows via a modified Pecchini method:[54] Appropriate amounts of SrCO3 (99.995% trace metals basis, Aldrich) and La2O3 (99.99% trace metals basis, Aldrich) were dissolved in HNO3 (redistilled, 99.999% trace metals basis, Aldrich); Cr(NO3)3·9(H2O) (99.99% trace metals basis, Alfa Aesar) and, in case of Ni containing perovskites, Ni(NO3)·6(H2O) (99.995% trace metals basis, Aldrich) were dissolved in double distilled water. The cation solutions were merged, citric acid (99.9998% trace metals basis, Aldrich) was added as a complexing agent in a molar ratio of 1.2 with respect to the total amount of cations, and the obtained solution was heated in a quartz beaker. After evaporation of large parts of the solvent a darkish foam was formed, which spontaneously decomposed upon further heating. The black residue was calcined at 750 °C in a box furnace, milled in a stainless steel mortar, isostatically pressed (ca. 4 kbar), and finally sintered at 1400 °C for 2 h in air to obtain targets for pulsed laser deposition (PLD). La0.6Sr0.4FeO3δ (LSF) targets were prepared from commercial powder (purchased from Aldrich) by isostatic pressing (ca. 4 kbar) and sintering at 1250 °C in air.

Table 1

Composition of Investigated Perovskite-Type Working Electrodes and the Abbreviations Used

composition of perovskite electrode	abbreviation used throughout the text
La0.6Sr0.4FeO3δ	LSF
La0.8Sr0.2Cr0.9Ni0.1O3δ	LSCrNi8291
La0.7Sr0.2Cr0.9Ni0.1O3δ	LSCrNi7291
La0.8Sr0.2CrO3δ	LSCr

Cell Preparation

The electrochemical cell was prepared by depositing electrodes on (100)-oriented 9.5 mol % yttria-stabilized zirconia (YSZ) single crystals (Crystec, Germany), which served as the electrolyte. For preparation of counter electrodes (CE), 20 mol % gadolinia doped ceria (GDC; Treibacher) paste, LSF paste, and Pt paste (Gwent Electronics) were consecutively brushed onto the bottom side of YSZ single crystals. GDC and LSF pastes were prepared from the respective powders, ethylcellulose, and terpineol by mixing them in a mortar. GDC was used to avoid reaction between LSF and YSZ and was sintered at 1300 °C. Subsequently, LSF and Pt paste were deposited and annealed for 2 h at 1050 °C in air. In these porous electrodes, LSF is the electrochemically active material and Pt acts as an electronic current collector. The GDC is a buffer layer to avoid formation of insulating zirconate phases at the LSF/YSZ interface during sintering at 1050 °C.

Thin film working electrodes (WE) of 200–250 nm thickness were grown at 650 °C on the polished top side of the YSZ electrolyte by pulsed laser deposition (PLD). To ensure a laterally homogeneous electron distribution and thus a well-defined electrochemical polarization of the thin film electrodes, a buried Pt current collector grid was fabricated prior to the PLD process by magnetron sputtering, photolithography, and Ar ion beam etching; more details on the working electrode preparation are given elsewhere.[36,48,55,56] Application of a GDC buffer layer between WE and YSZ was omitted since no zirconate formation is expected at temperatures of 800 °C and below. A sketch of the mounted sample’s cross section and an electron microscopy image of a thin film working electrode with buried current collecting grid are depicted in Figure a,b, respectively.

Figure 1

(a) Sketch (cross section) of an electrolysis cell mounted for NAP-XPS measurements. (b) SEM (top view) of a PLD-deposited LSF working electrode with embedded Pt thin film grid for current collection.

Operando NAP-XPS Experiments

NAP-XPS measurements were performed at beamline ISISS of synchrotron facility BESSY II in Berlin. The main part of the setup is a “high pressure cell” (up to 7 mbar), which is attached to a differentially pumped hemispherical analyzer (modified SPECS Phoibos 150); for details regarding the XPS equipment the reader is referred to ref (57). Photoelectron spectra were recorded with different photon energies (410 eV for C 1s, 490 eV for Ni 3s, 650 eV for O 1s, 845 eV for Fe 2p); pass energy was 20 eV. These correspond to kinetic photoelectron energies between 110 and 130 eV, which lead to an approximate information depth (inelastic mean free path) of ca. 0.5 nm. This is of the order of the rhombohedral lattice parameter of LSF,[53] thus allowing very surface sensitive measurements.

For mechanical fixation and electrical contact the electrochemical cell was mounted onto the Pt back plate of the sample holder by means of Pt/Ir clamps. The Pt/Ir clamps served both as mechanical fixations and electric contact of the working electrode. The Pt back plate contacted the counter electrode. A sketch of such a cell mounted for spectroscopy is shown in Figure a. Heating of the cell was done by illuminating the CE-sided Pt sheet with an IR laser. Temperature was controlled by adjusting the power of the infrared laser heating the back plate of the sample holder. Temperature measurement was done by a pyrometer as well as by the ionic conductivity of the YSZ electrolyte obtained from impedance measurements; details regarding the latter procedure are given in refs (58 and 59). Temperature values resulting from both methods were in reasonable agreement of ±15 °C. For electrochemical polarization experiments as well as impedance spectroscopy, a Novocontrol Alpha Analyzer equipped with a POT/GAL interface was used. For these measurements the WE was grounded, and the voltage was applied to the counter electrode. The overpotential η at the WE was obtained by subtracting the ohmic drop in the electrolyte (and the contact resistances and wires) from the applied voltage Uset byIn eq Rohm denotes the ohmic resistance (obtained from the high frequency intercept in impedance spectra, see Figures S1 and S2 in the Supporting Information), and IDC is the steady state DC (direct current) current flowing during polarization experiments. The potential drop at the counter electrode is small compared to the thin film working electrode by the following considerations: (i) The surface area of the porous CE is large compared to the surface area of the thin film electrode; thus, the polarization resistance of the porous CE can be regarded to be rather small compared to the WE. (ii) The CE acts as an oxygen evolution electrode during CO2 splitting at the WE (as can be seen from MS data below), and its polarization resistance for this reaction was already shown to be negligible.[48] Consequently, the CE also acts as a reversible reference electrode, thus allowing two-point measurements. For all experiments, the background pressure in the chamber was 0.25 mbar CO2. The gas composition in the chamber was measured by a quadrupole mass spectrometer (Pfeiffer Prisma). Please note that all the experiments were performed in a single chamber setup, i.e., without separation of anodic and cathodic gas compartments. Thus, both gaseous products of CO2 electrolysis (O2 and CO) can be found in the chamber (further details see Section ).

Results and Discussion

Electrochemical and MS Results

In a first step, we demonstrate that electrochemical CO2 reduction indeed takes place upon cathodic polarization. A plot of the electrochemical DC current IDC versus the overpotential η applied to the thin film working electrode is depicted in Figure . This diagram shows a comparison of all investigated perovskite-type materials at 720 °C in 0.25 mbar CO2. For a mixed conducting electrode with surface limited kinetics the overpotential η determines the deviation of the oxygen chemical potential in the electrode from its equilibrium value (ΔμO), i.e., the difference between the chemical potential of oxygen in the electrode (μO,el) from the value in the gas phase (μO,gas). This is the thermodynamic driving force of the reaction:[48]The DC current IDC is a direct measure of the reaction rate r since in the case of a purely ionic current through the electrolyte both are connected by Faraday’s lawwith z and F denoting the number of transferred electrons and Faraday’s constant, respectively. In our case the reaction rate is the rate of CO2 reduction at the SOEC cathode (i.e., working electrode)and owing to charge neutrality also the rate of oxygen evolution at the anode (i.e., counter electrode)The total CO2 electrolysis reaction, which is the sum of both partial reactions, is thusFrom Figure a, a higher activity of LSF for CO2 splitting can be concluded, since it shows similar electrolysis currents as the chromite based materials already at ca. 200 mV lower polarization; i.e., a lower thermodynamic driving force is needed for the same reaction rate.

Figure 2

Current–overpotential characteristics of the four different WE materials measured at 720 °C in 0.25 mbar CO2. (a) Linear plot. The filled symbols are regular data points. The open symbols indicate measurements influenced by deposition of graphitic carbon (owing to large cathodic overpotentials). The corresponding degradation of the current is indicated by the arrows. The lines indicate the curves obtained from fitting the data in the logarithmic diagram below. (b) Tafel plot, i.e., logarithmic I–V characteristics, of the data in part a. Data points influenced by carbon deposition related degradation are not shown. The lines were obtained by fitting the data points to the Tafel equation (see eq ). The open points at η = 0 V were not considered in these fits.

The shape of the measured curve indicates an exponential relation between IDC and η. A logarithmic plot of the current versus the overpotential, which is commonly referred to as the Tafel plot, thus yields a linear correlation (see Figure b). This can be analyzed in terms oforwith R and T denoting gas constant and absolute temperature, respectively. The fit parameter I0, which can be extracted from the axis intercept of the linear curve in Figure b, is the exchange current density under equilibrium conditions. It is connected to the equilibrium exchange rate of the CO2 reduction reaction (eq ) by Faraday’s law (eq ). Such an exponential relation between current and overpotential is frequently found in electrochemistry, and in aqueous electrochemistry this often indicates a rate-limiting charge transfer.[60] However, a simple interpretation of such a relation may fail for electrochemical reactions at gas/solid interfaces.[61] Not only does the overpotential η differ from the relevant change of the surface potential step, but also the variation of defect concentrations by applying a voltage (η = ΔμO/2F) may be highly important.[62] Hence, also the meaning of the parameter α in eq is not straightforward. Fit parameters are summarized in Table .

Table 2

Resulting Fit Parameters of Tafel Fits in Figure b (720 °C data) Using Equation

material	α	I0/mA cm-2
LSCrNi8291	0.590 ± 0.025	(10.1 ± 1.4) × 10–4
LSCrNi7291	0.613 ± 0.045	(8.7 ± 1.6) × 10–4
LSF	0.567 ± 0.066	(54 ± 18) × 10–4

As already mentioned above, these results reveal LSF as the more active material: Its exchange current density is almost 1 order of magnitude higher than those of the chromite based components. The difference in α of all materials, however, is statistically insignificant. This is a clear indication that the rate-determining step is of the same nature on all investigated perovskite-type WEs. The difference in I0 between LSF and LSCrNi may be caused by a different height of the activation barrier and/or by a different concentration of a species involved in the rate-determining reaction step. Correlation of these results with NAP-XPS data suggests that different defect concentrations (electrons and oxygen vacancies) are the more likely reason (see discussion below).

Moreover, the results in Figure strongly suggest that the exsolution of metallic particles does not play a significant role for the CO2 electrolysis performance, since the Ni free chromite electrode (red pentagons in Figure ) shows almost the same electrochemical behavior as the Ni containing ones (blue circles). Exsolution of metallic Ni from LSCrNi was verified by NAP-XPS (see Section below) as well as by SEM images (Figure S4 in the Supporting Information). The maximum catalytic effect of the Ni exsolutions (if there is any) is thus smaller than the experimental scatter between nominally identical samples. This is very little compared to that of H2O electrolysis, which is accelerated by more than an order of magnitude once metallic exsolutions are formed.[36,47] Further details on the role of exsolved metal particles for CO2 electrolysis will be discussed in the following sections.

For LSCrNi7291 and LSF, I–V curve measurements were also performed at 600 and 400 °C, respectively. The resulting Tafel plots are shown in Supporting Information Figure S3. The corresponding fit results are summarized in Table S1. Thus, a first estimate of activation energies of I0 can be made for LSCrNi7219 and LSF and leads to 1.7 and 1.3 eV, respectively. This difference in activation energy can originate from a different energy barrier height or a different formation enthalpy of a species in the rate-determining reaction step or from a combination of both.

Since the above-mentioned interpretations are only valid for a Faradaic relation between IDC and reaction rate, the gas composition in the reaction chamber was analyzed by mass spectrometry (MS). Owing to the large background of CO2 and its fragmentation into CO the accurate measurement of the reaction product CO was rather challenging.[63] However, since the experiments were performed in a single chamber setup, both products of the electrolysis reaction CO and O2 (which are connected by eq ) are observable. Thus, not only the CO but also the O2 signal can be used to check for a linear relationship with the DC current. Plots of the normalized O2 signal versus the DC current are depicted in Figure a,b for LSCrNi8291 and LSF, respectively. In either case a linear correlation between the amount of produced oxygen and the electrochemical current is found. This is a strong indication of a Faradaic electrolysis of CO2 despite rather strong cathodic polarization. Contributions of an electronic current in YSZ, which may become relevant for strongly reducing conditions,[64] can thus be neglected.

Figure 3

MS signal of oxygen normalized to the CO2 background plotted as a function of the measured DC current at 720 °C in 0.25 mbar CO2 for (a) LSCrNi8291 working electrode and (b) LSF working electrode. The lines indicate a linear fit in each case.

Furthermore, it should be noted that for very high polarization a time dependent decrease of the electrolysis current can be observed, which is indicated by the open symbols as well as by the arrows in Figure a. This degradation can be related to the deposition of graphitic carbon, which was observed by NAP-XPS measurements; see Section for spectroscopic details. However, both the degradation and the presence of surface carbon were completely reversible upon retracting the applied electrochemical potential. At the experimental temperature of 720 °C the reaction of carbon with carbon dioxide via Boudouard’s reaction (C + CO2 ⇄ 2CO) already yields significant amounts of CO[65] and is obviously sufficiently fast for regenerating all materials used in this study. This observation is particularly important from an application point of view, since it offers a rather simple method of resetting degradation caused by coking. Further details regarding coking and regeneration of the working electrodes and corresponding XPS data are discussed in Section .

NAP-XPS Results

At each applied electrochemical polarization (i.e., each point in Figure ) NAP-XPS measurements were performed. Here, the main focus was laid on the measurement of C 1s and O 1s spectra. Typical C 1s spectra of all four perovskite-type working electrodes under polarization are depicted in Figure . Depending on the electrochemical polarization, different features could be observed, corresponding to different carbon species. Qualitatively, however, the four different materials showed a very similar behavior.

Figure 4

Comparison of C 1s spectra measured on the four different perovskite-type working electrodes at 720 °C in 0.25 mbar CO2 under different cathodic polarizations. In this plot binding energies were corrected by the applied overpotential for the sake of easier comparability of different spectra assuming a shift of the electrode’s Fermi level upon polarization by −1 eV/V.[31,48] (This correction anticipates a result, which is discussed later in the text.)

Without additional polarization, only one sharp C 1s peak at 293–294 eV could be detected. Since this peak can be suppressed by applying a DC bias to the nozzle of the photoelectron collecting aperture (see Figure S5) and owing to its characteristic binding energy,[66] it can be identified as a CO2 gas phase signal. Upon application of a cathodic overpotential to the WE, this gas phase signal is slightly shifted, and an additional broader peak appears at ca. 290 eV (indicated by CO in Figure ). This additional species is most probably decisive for CO2 reduction; owing to its position in the spectrum it may be a carbonate or carboxylate[67] or a related oxygenated carbon species. Further investigation and discussion of its nature and kinetic implications is the subject of the following sections. The shift of the gas phase peak can be explained by a change of the vacuum level gradient within the gas phase upon electrochemical polarization of the working electrode.[29,68,69] At very high cathodic overpotential, a further carbon species appears on the electrode surface as can be observed by the evolution of another C 1s peak at ca. 284 eV (indicated by C in Figure ). Owing to its position in the spectrum and its asymmetric shape, this species can be clearly identified as graphitic carbon.[66,70]

Interestingly, the appearance of the carbon species at ca. 290 eV also finds its counterpart in O 1s spectra, which is illustrated for a LSCrNi7291 electrode in Figure . It shows a comparison of O 1s (Figure a) and C 1s (Figure b) spectra for the polarized (top) and nonpolarized (bottom) situation. The appearance of the carbon species peak at ca. 290 eV is clearly accompanied by the evolution of a shoulder at ca. 532 eV in the O 1s spectrum, and this further supports an interpretation of this species in terms of a carbon–oxygen compound. Moreover, in Figure the fit results of individual components are indicated as colored peak areas below the curve. For fitting of O 1s spectra, up to four components were used: the CO2 gas phase at ca. 537 eV and two components to fit the signal from the perovskite at ca. 530.5 and 529 eV, which are interpreted in literature studies as surface and bulk oxygen species, respectively.[32,71−73] Moreover, one additional species accounts for the shoulder appearing at ca. 532 eV in the case of polarized electrodes. Further details on the fitting procedure of O 1s and C 1s spectra are given in the Supporting Information.

Figure 5

Comparison of O 1s spectra (a) as well as C 1s spectra (b) of cathodically polarized (○) and nonpolarized (□) LSCrNi7291 working electrode at 720 °C in 0.25 mbar CO2. In the case of both types of spectra, an additional feature is visible upon polarization (top spectrum in each case). The components used for fitting are indicated as the filled peak areas, and the envelope of the fit is the solid blue line. (BE was not corrected to visualize peak shifts.)

Please note that owing to the absence of a Fermi edge for most of the samples no binding energy correction for all of the spectra in this study was performed. The meaning of this approach should be briefly explained: Since the WE electrode was grounded, electrostatic charging can safely be excluded as a source of the XPS peak shift. The observed shifts are thus either due to chemical shift (for example different species in C 1s and O 1s spectra) or due to work function changes caused by changes of electron concentration (i.e., shift of the Fermi level), which is a result of the electrochemical polarization.[31,48] For the sake of better comparability the spectra in Figure are corrected for the latter effect. However, the fitting of data was performed on uncorrected data, since the Fermi level shift upon polarization itself is an interesting result, which is analyzed and discussed in detail below. To estimate the accuracy and the scatter of our nominally uncorrected binding energies, which arise, e.g., from monochromator mechanics at the beamline,[74] the carbon peak in C 1s spectra was used. In our case a mean value of 284.99 eV with a standard deviation of 0.18 eV results, which is in good agreement with literature values for disordered graphitic carbon.[70]

Identification of the Surface Species Appearing under Cathodic Polarization

To answer the question about the nature of the oxy-carbon adsorbate evolving under cathodic polarization, it is helpful to first discuss where, i.e., on which substrate, it is adsorbed. In principle, two different cases are possible:[11] adsorption on the perovskite-type oxide or on the exsolved metal particles. The latter case can be ruled out by two arguments: First, the appearance of the unidentified species does not correlate with the formation of a metallic species since, on LSF at 600 °C, η = −699 mV, already substantial amounts of metallic iron can be detected, whereas no C 1s signal at 290 eV can be observed in the C 1s spectrum. On the other hand, at 720 °C a small though observable signal at ca. 290 eV is visible in C 1s spectra already without polarization and thus without presence of metallic iron, cf. parts a and b of Figure . Second, the carbon species at ca. 290 eV also appears on LSCr electrodes, which do not contain any cations that may form metallic exsolutions, see Figure (red pentagons). Consequently, the species can be assigned to the oxide.

Figure 6

Comparison of Fe 2p (a) and C 1s (b) spectra measured on LSF electrodes at 600 °C in 0.25 mbar CO2 at a cathodic polarization of −299 mV. In Fe 2p clearly the signature of both oxidic and metallic iron can be observed; in C 1s only the sharp peak of the CO2 gas phase is present (BE in C 1s spectra were corrected by η for the sake of easier comparability). (c) Ni 3p spectra measured on LSCrNi8291 at 720 °C with and without cathodic polarization showing the exsolution of Ni at the electrode surface (BE corrected by η). The gray shaded areas indicate typical 3p binding energies for metallic nickel and Ni2+ in oxides.[75,76]

This irrelevance of the exsolved metal particles for CO2 electrolysis is also obvious from the electrochemical data in Figure . The presence of metallic nickel on the Ni doped chromite electrodes under cathodic conditions was confirmed by evolution of a lower energy species in Ni 3p spectra, see Figure c. (Please note that a measurement of Ni 2p was not possible owing to an overlap with La 3d lines.) This seems to contradict the assumption of an enhanced catalytic activity of exsolved Ni for CO2 electrolysis as published in recent studies.[33,77] One reason may be the different oxide/metal combinations investigated in the present work and the above cited literature studies. At least on chromite electrodes in a H2/H2O atmosphere, exsolution of Ni particles caused a much smaller enhancement effect on the electrochemical performance than Ru exsolution.[78] However, for CO2 splitting, to the best of our knowledge, such an effect was not reported so far. Thus, the most obvious difference between the present work and the literature studies on CO2 electrolysis employing Ni exsolution electrodes appears to be the comparably ill-defined surface areas of the porous electrodes, which were used in refs (33) and (77). The dependence of the inner surface of porous electrodes on the sintering behavior of the starting powder, which may be affected by composition changes such as addition of exsolvable transition metals, makes normalization of measured currents to the correct electrode surface area rather challenging and prone to systematic errors. Indeed, the claimed effects on the electrode performance in refs (33) and (77) are only in the range of 10%, which may be explainable by differences in inner surface of porous electrodes. Moreover, XPS measurements in ref (33) were conducted ex situ, thus not allowing the detection of a metastable adsorbate, which only forms under operation conditions. This result again highlights the importance of using geometrically well-defined model electrodes in combination with operando spectroscopic characterization for a detailed investigation of the relevant electrode surface chemistry.

Further strong evidence that the unidentified species is indeed adsorbed on the oxide is its binding energy shift upon electrochemical polarization. In Figure a the difference between the C 1s binding energies of the adsorbate and graphitic carbon is plotted versus the applied overpotential. Graphitic carbon is assumed to form an additional phase on top of the electrode, and thus, its Fermi level cannot be affected by defect chemical changes in the electrode. A linear fit including data points from all materials yields a slope of almost exactly −1 eV/V. (Please note that the error of the slope can almost exclusively be explained by the standard deviation of our non-normalized binding energy values, which amounts to 0.18 eV, cf. Section ). This clearly shows that the species is associated with the perovskite lattice, since it almost exactly follows the shift of the Fermi level, which is a consequence of the electrochemical polarization in our predominantly vacancy-compensated mixed conducting electrodes.[31,48] The slope of almost exactly −1 eV/V in Figure a is thus a strong argument for ruling out the possibility of the 290 eV peak in C 1s spectra to originate from a gaseous species. Even though gas phase signals also show polarization induced shifts,[68,69] the changes in binding energy are significantly smaller as can be seen in Figure S6 in the Supporting Information. Moreover, the rather broad peak shape of the CO species provides further evidence of its adsorbate character, since gas phase peaks are usually sharper, such as the CO2 peak in Figures –6.

Figure 7

(a) Difference between the binding energy of the unidentified C 1s species at ca. 290 eV and graphitic carbon plotted versus the applied overpotential. The solid line is a linear fit considering all data points. (b) Plot of the XPS peak area of the shoulder in O1s spectra at 532−533 eV versus the peak area of the species in C 1s at ca. 290 eV. The solid gray line is a linear fit considering all data points. The dashed and the dash-dotted line indicate the corresponding peak area ration from CO2 and CO gas phase measurements, respectively. (c) Difference in binding energy of the unidentified species in O 1s and C 1s spectra plotted versus the applied overpotential. The dashed and the dash-dotted line indicate the corresponding values from CO2 and CO gas phase measurements, respectively. The black symbols indicate the values extracted from the CO2 gas phase peak of “regular” NAP-XPS measurements (i.e., performed on perovskite electrodes). The dotted line indicates the corresponding BE difference between O 1s and C 1s reported for SrCO3.[80,81]

Additional information for identifying the nature of the adsorbate appearing upon polarization comes from its oxygen to carbon ratio. In Figure b the peak area of the O 1s shoulder at ca. 532 eV BE is plotted versus the peak area of the C 1s peak at ca. 290 eV BE (cf. also Figure ). With the assumption of a linear relationship, the obtained fit curve (solid gray line in Figure b) is significantly steeper than the expected slopes extrapolated from CO2 and CO gas phase measurements shown by the dashed and dash–dotted lines, respectively. After correcting the C 1s/O 1s area ratio of the fit curve by the X-ray intensities at different photon energies as well as by the different probabilities of photoelectron emission of the two different elements,[79] an oxygen to carbon ratio of 2.84 ± 0.15 can be obtained. A very similar value (ca. 2.8) was obtained for intensity calibration by the gas phase peaks of CO2, which have a molar ratio of 1:2. This result of a value rather close to 3 strongly suggests that both the 290 eV peak in C 1s spectra and the O 1s shoulder at ca. 532 eV can be attributed to a carbonate species. This interpretation of a carbonate adsorbate is further supported by the difference in binding energy of the corresponding O 1s and C 1s signals, which is shown in Figure c. Comparison with gas phase CO and CO2 as well as bulk SrCO3[80,81] clearly identifies its carbonate nature.

Polarization Dependence of the Surface Carbonate

In Figure the evolution of the peak area of the carbonate XPS signal with overpotential η is depicted. Under moderate cathodic polarization the amount of carbonate quickly increases. The observations that the formation of the carbonate only occurs under cathodic polarization and that the surface coverage first increases upon increasing the polarization indicate an electron transfer to be necessary to form the surface carbonate. In other words the presence of electrons in the semiconductor-type oxide is crucial for the formation of the surface carbonate. This increase in electron concentration upon electrochemical polarization is also reflected by the observed −1 eV/V slope in Figure a, which is caused by an increase of the Fermi level of the perovskite under cathodic polarization.[31,48] It should also be emphasized that the formation and/or increase of the amount of surface carbonate cannot be explained by a change of the vacancy concentration, since under the applied conditions the acceptor dopant is already to a large degree vacancy compensated (especially in case of LSF).[53] Consequently, a further reduction cannot significantly increase the oxygen vacancy concentration since vacancies are already majority charge carriers.

Figure 8

Peak area of the carbonate C 1s peak plotted as a function of the applied overpotential for four different electrode materials measured at 720 °C in 0.25 mbar CO2.

Upon further increasing the cathodic polarization the carbonate coverage reaches a plateau; see LSF (green triangles) and LSCrNi7291 (orange squares) in Figure . A maximum amount is found for LSF and LSCrNi7291, while in the case of LSCrNi8291 (blue circles) only the onset of the plateau is visible since carbon deposition started at very high cathodic voltages thus impeding further measurements of the carbonate species under higher polarization values. Interestingly, this plateau in carbonate surface coverage differs by almost 1 order of magnitude between the different perovskites, see Figure . This can be interpreted in terms of different number of available adsorption sites. From a defect chemical point of view the most obvious difference between the materials is their degree of acceptor doping (with LSF > LSCrNi7291 > LSCrNi8291 ≈ LSCr). Consequently, they exhibit significantly different concentrations of oxygen vacancies under reducing conditions. Assuming oxygen vacancies as the adsorption sites for CO2 may thus partly explain the trend of different saturation limits in Figure . Moreover, the existence of a saturation limit also suggests a rate-limiting elementary step subsequent to carbonate formation.

For LSF and LSCrNi7291 the carbonate coverage is observed to decrease again under very high cathodic bias; see green triangles and orange squares in Figure , respectively. This behavior suggests the rate-limiting step subsequent to carbonate formation also depends on the electron concentration of the mixed conducting electrode material. A second electron transfer leading to electrochemical carbonate reduction would be a potential explanation for this polarization induced carbonate decomposition.

Discussion of the Mechanism of CO2 Electrolysis on Perovskite-Type Electrodes

Before a mechanism is suggested, the most important observations should be briefly summarized:

The formation of the carbonate correlates with an increasing electron concentration in the perovskite electrodes. Thus, an electron transfer is concluded to be necessary to form the carbonate. Since the carbonate only exists under sufficiently cathodic polarization but immediately vanishes upon retracting the bias, it is regarded as an intermediate of carbon dioxide reduction.

The maximum coverage of the surface carbonate shows the same trend as the oxygen vacancy concentration in the investigated materials, which suggests oxygen vacancies as the adsorption sites of CO2 forming the surface carbonate. Moreover, the saturation behavior suggests that an elementary step of the carbonate reduction to CO rather than the carbonate formation is rate-limiting.

The carbonate coverage again decreases under very high electrochemical polarization, which is an indication that also the rate-limiting step of carbonate reduction to CO can be accelerated by cathodic polarization.

To consider both oxygen vacancies as well as electronic charge carriers in the formation of the carbonate, adsorption of CO2 to an oxygen vacancy at the surface (denoted Vac), coupled with an electron transfer, is suggested as the first step of carbonate formation:The corresponding electron can either originate from a near surface polaron, e.g., a reduced Fe2+ or Cr2+ ion, or may already reside in the vacancy. In the first case the electron may be delocalized on the B–O sublattice of the perovskite owing to possible hybridization of the metal-d and oxygen-p states.[31,82] The latter case corresponds to an electron being trapped in the oxygen vacancy thus forming a singly charged vacancy (Vac + e– → Vac–). This defect species is reported in oxygen deficient oxides under strongly reducing conditions and may even act as a color center.[64,83−85] Owing to the electronic structure of CO2, the resulting species needs to be a radical (indicated by • in eq ; please do not confuse this with the relative charge in Kröger–Vink notation for point defects). In a following step, the carbonate can be formed by bonding to another surface oxide ion forming a bidentate carbonate:The interpretation of the carbonate as a bidentate is thus a direct consequence of the suggested CO2 adsorption to a surface vacancy. A sketch of the reactions in eqs and 9 is depicted in Figure . The weakening of the bonds associated with the radical nature of the carbonate may be responsible for the activation of the rather stable CO2 molecule. The fact that no intermediate before (CO3)•3– formation can be detected by NAP-XPS indicates a low surface coverage of the respective species. It even appears to be the general case for all investigated perovskites that only one reaction intermediate exhibits a measurable surface coverage.

Figure 9

Sketch of the proposed mechanism of the formation of the carbonate adsorbate detected by NAP-XPS. For the sake of simplicity only a cut through a (100) lattice plane occupied by B and O, with the surface parallel to (110), is depicted, thus not showing the full perovskite structure. (a) CO2 adsorbing into an oxygen vacancy, followed by an electron transfer from a polaron (or the conduction band). (b) CO2 adsorption and first electron transfer occurring in one step by adsorption into a singly charged oxygen vacancy (i.e., a vacancy with a trapped electron). In either case a or b, a bidentate carbonate radical results, which is regarded as a reaction intermediate of CO2 reduction to CO.

It should further be emphasized that formation of the suggested bidentate is not equally realistic on every perovskite surface, since the distance of two oxygen lattice sites needs to be in a certain range to allow for bidentate adsorption. The C–O bond length in a regular CO32– ion is 1.28 Å, and the O–O distance is ca. 2.22 Å.[86,87] Owing to the absence of a double bond in the proposed (CO3)•3– adsorbate, the bond lengths and thus the O–O distance can be expected to be slightly longer. With a comparison of these numbers with the oxygen lattice site distances in our pseudocubic perovskites, an adsorption on a AO-terminated (100) facet with an oxygen–oxygen distance between 3.8 and 3.9 Å is hardly possible. On a BO-terminated (100) facet, however, the distance between two neighboring oxygen sites is smaller by a factor of 1/√2 and thus in the range of 2.7 Å. Consequently, the adsorption of the suggested carbonate radical would be feasible on such a surface at least from a sterics point of view. AO-terminated (110) or oxygen-terminated (111) facets are other examples of perovskite surfaces capable of carbonate bidentate formation.

To the best of the authors’ knowledge, such a carbonate species has not been reported so far for perovskite-type electrodes. However, similar results were described for Gd and Sm doped ceria thin film electrodes.[26,27] In both studies an adsorbed species on ceria was identified as a carbonate. Also there, the carbonate intensity increased with cathodic overpotential, and its coverage was saturated under strong cathodic polarization. Moreover, in ref (27), the carbonate species is also interpreted as a surface carbonate with an excess electron, which is further reduced to CO in a following rate-determining step. In contrast to our interpretation, the resulting carbonate is suggested to be adsorbed on the ceria surface as a monodentate. While not being stable as a bulk species, such a carbon dioxide or carbonate adsorbate with excess negative charge has already been discussed in surface science, and it was suggested to play a role in CO2 activation.[11] The authors of ref (27) identified Ce3+ on the ceria surface as the crucial electron delivering point defect. These Ce3+ point defects are in fact electrons, which are localized on a Ce4+ ion, i.e., polarons.[88] Surprisingly, our perovskite-type electrodes show a very similar electrochemical as well as spectroscopic behavior, but electronic surface states comparable to the Ce3+ are not necessarily evident, and other sources of electrons (e.g., singly charged vacancies) may also be relevant.

For ceria surfaces, also formation of a tridentate carbonate was reported in literature reports based on DFT calculations.[89] For perovskite surfaces, however, no comparable study exists dealing with the conformation of surface carbonates. For an unambiguous proof of the model suggested in this study, further studies are urgently needed to address the question of the mechanism of CO2 splitting on perovskite-type electrodes in more detail. For example, vibrational spectroscopy may be an additional experimental method for future investigations of the carbonate species, which could shed further light on the reaction mechanism.[11,90] However, on the basis of the already available data we suggest that CO2 reduction with activation via a (CO3)•3– species requires an oxide electrode with many oxygen vacancies but also with sufficiently high electron concentration. High oxygen vacancy concentrations are generally achieved by acceptor doping. Then, however, electron concentration is often strongly suppressed by high hole concentrations. Our model with the requirement of both electrons and oxygen vacancies thus provides the first information regarding how to further proceed in the search for improved perovskite-type CO2 reduction electrodes.

Regarding the following steps of the CO2 electrolysis a second electron transfer as well as release of CO is necessary.Under very strong cathodic polarization and thus very high electron concentration in the mixed conducting perovskites, the rate-limiting step of eq is accelerated electrochemically, thus leading to the observed decrease of carbonate coverage, which is observed for LSF and LSCrNi7291 in Figure . This assumption of electron concentration directly affecting kinetics of the rate-determining step is supported by the exponential current voltage kinetics (see Figure ) together with the observed −1 eV/V binding energy shift of the carbonate intermediate (see Figure a). The fact that the carbonate completely follows the Fermi level shifts upon polarization clearly rules out the possibility of a Butler–Volmer-type mechanism, where the applied polarization changes the activation barrier of the rate-limiting charge transfer step thus yielding the typical exponential U–I curves.[60] On solid state electrodes such a type of mechanism can only occur if there exists an electrostatic surface potential step such as a surface dipole.[61] If such a surface dipole is affected by the applied electrochemical polarization the binding energy of the respective species shifts by value different from −1 eV/V.[30,35] The observed exponential U–I curve can thus not be explained by the applied polarization modifying an activation energy barrier. Rather, an increase in electron concentration accelerating the rate-determining step is a possible scenario. Since the concentration of electrons in the studied perovskites depends exponentially on the applied cathodic overpotential, a rate-determining elementary step depending on the concentration of electrons in the electrode material is a very likely explanation for the observed behavior.

For completing the reaction cycle the surface of the perovskite has to be recovered by again forming a surface oxygen vacancy. In contrast to purely chemically driven CO2 conversion, where often hydrogen is employed for this purpose, electrochemically driven CO2 splitting regenerates the surface by vacancies formed at the counter electrode viaThese vacancies are then transported to the working electrode surface due to the electrochemical potential difference applied to the electrolysis cell. At the working electrode they can again start a CO2 splitting reaction. This method of electrochemically regenerating active surface sites can be much more efficient than a purely chemical reduction, since no additional reduction agent such as H2 is needed; thus, no competition of different gas species for adsorption sites takes place, and the byproducts (H2O in case of chemical reduction, O2 in the present case) can easily be prevented from contaminating the produced CO by separating the gas compartments of working and counter electrodes. These advantages further highlight the attractiveness of electrochemical CO2 splitting and thus emphasize the need for an in-depth understanding of this reaction, especially on oxide electrodes.

Coking and Regeneration Effects

The deposition of carbon was only observed under very strong electrochemical polarization. Figure a depicts C 1s spectra measured on a LSCrNi8291 working electrode at 720 °C in 0.25 mbar CO2 under different cathodic overpotentials. Only under 1450 mV does an additional asymmetric peak at 284–285 eV appear, which is characteristic for graphitic carbon.[70] When keeping the polarization, this peak grows further, as shown in Figure b. The fact that carbon deposition also occurs at these relatively high temperatures and with a pure CO2 gas feed can be explained by the extremely reducing conditions established by the strong cathodic polarization. Under these conditions the local CO partial pressure at the electrode surface is very high thus allowing carbon formation by CO disproportionation (2CO(ad) → C(ad) + CO2) or by a direct electrochemical reduction of carbon monoxide (CO(ad) + 2e– → C(ad) + O2–). However, upon retraction of the applied potential, the graphite peak immediately disappears, indicating a complete removal of the carbon deposit from the electrode surface (see also Figure b, top spectrum). A very similar behavior was also found on LSF electrodes, as shown in Figure S7 in the Supporting Information. Compared to zirconia or ceria based state of the art SOFC/SOEC fuel electrodes, where relatively harsh oxidizing conditions (which may also oxidize the Ni particles and lead to electrode damage) are needed to fully remove carbon deposits,[91,92] the regeneration on our perovskites is thus easy to achieve.

Figure 10

(a) C 1s spectra measured on LSCrNi8291 at 720 °C in 0.25 mbar CO2 at different cathodic polarizations. Coking occurs at the highest overpotential (η = −1450 mV) observable by evolution of an asymmetric peak at 284–285 eV. (b) With prolonged time the amount of deposited graphite increases, but it is immediately removed upon retracting the polarization. (BE in parts a and b not corrected.) (c) MS signals of O2 and CO (normalized to CO2 background) with visible steps corresponding to the indicated overpotential.

The quick removal of carbon can be explained by reaction of the carbon deposit with CO2 from the gas phase via the reversed Boudouard reaction (C + CO2 → 2CO). At the experimental temperature of 720 °C the “CO side” of the reaction is thermodynamically favored.[65] Differences in the regeneration behavior may originate from different types of carbon (such as graphitic carbon, disordered carbon, carbon nanotubes, and others) growing on different electrode materials. How far the metallic exsolutions play a role in the regeneration behavior cannot be concluded unambiguously from the data available so far; this question needs to be addressed in future work.

The electrochemical effects caused by carbon deposition can also be observed in current voltage curves (see open symbols in Figure a) and in the amount of produced O2 (and CO) as shown in Figure c: Under 1450 mV overpotential the production rate of O2 significantly decreased with time. MS data also nicely reflect the regeneration effect upon retracting the applied bias: When reapplying a cathodic voltage (1271 mV at ca. 9200 s in Figure c) the O2 production rate was even slightly higher than that before any coking had occurred (which might be due to a slightly increased roughness of the working electrodes). This again demonstrates the complete reversibility of the detrimental effects of carbon deposition, which is advantageous from an application point of view.

Conclusion

High temperature electrolysis of carbon dioxide was performed on different acceptor doped perovskite-type electrodes. In contrast to recent publications dealing with CO2 electrolysis on porous perovskite electrodes, the present work employs geometrically well-defined thin film electrodes. These model-type electrodes not only allow reliable comparison of electrode performance but also facilitate NAP-XPS measurements to study the surface chemistry of the electrodes under operation conditions. From the combined interpretation of the electrochemical results and the XPS data, the following conclusions are drawn:

All investigated electrode materials exhibit exponential current–voltage characteristics with LSF electrodes showing higher performance than chromite based materials. However, the virtually identical slopes in Tafel plots suggest that the reaction mechanism of CO2 reduction is the same on all investigated materials.

Exsolution of metallic particles from the perovskite lattice does not affect the kinetics of CO2 splitting, in contrast to high temperature H2O electrolysis. This contradicts conclusions made from electrochemical CO2 splitting measurements with porous perovskite electrodes and illustrates the importance of operando spectroscopic characterization of carbon dioxide electrolysis on model-type electrodes.

CO2 electrolysis on all investigated materials proceeds via a carbonate intermediate populating on the oxide surface under sufficiently reducing (cathodic) conditions. The carbonate adsorbate is interpreted as a bidentate radical (CO3)•3–, which forms via adsorption of CO2 on oxygen vacancies in combination with an electron transfer step. The electron originates from either a polaron or a singly charged oxygen vacancy.

In particular, the chromite based electrode materials exhibit a rather high coking resilience with carbon deposition only occurring under strongly cathodic polarization. Moreover, the detrimental effects of carbon are completely reversible upon retracting the applied bias without damaging the electrode, which is very promising from an application point of view.