Enhancing Electrochemical Water-Splitting Kinetics by Polarization-Driven Formation of Near-Surface Iron(0): An In Situ XPS Study on Perovskite-Type Electrodes

In the search for optimized cathode materials for high-temperature electrolysis, mixed conducting oxides are highly promising candidates. This study deals with fundamentally novel insights into the relation between surface chemistry and electrocatalytic activity of lanthanum ferrite based electrolysis cathodes. For this means, near-ambient-pressure X-ray photoelectron spectroscopy (NAP-XPS) and impedance spectroscopy experiments were performed simultaneously on electrochemically polarized La0.6Sr0.4FeO3−δ (LSF) thin film electrodes. Under cathodic polarization the formation of Fe0 on the LSF surface could be observed, which was accompanied by a strong improvement of the electrochemical water splitting activity of the electrodes. This correlation suggests a fundamentally different water splitting mechanism in presence of the metallic iron species and may open novel paths in the search for electrodes with increased water splitting activity.

Alpha-A high performance frequency analyzer equipped with a POT/GAL 30V 2A interface (both: Novocontrol, Germany). Impedance spectroscopy was performed in a frequency range between 10 mHz and 1 MHz and the ac root mean square voltage was limited to 5 mV to avoid XPS peak broadening.
For electrochemical polarization, dc bias voltages between +700 mV and -500 mV were applied to the working electrode. The individual set voltages were not applied in a linear sequence but rather randomly (with repeating some of the set voltages) to check for the reversibility and reproducibility of the current-voltage characteristics and XPS features. At each set voltage the collection of XPS data was started after a steady state dc current was reached and surface sensitive Fe2p spectra were recorded with incident photon energy of 845 eV. During simultaneously performed electrochemical and XPS experiments the partial pressures of H 2 and H 2 O were held constant at 0.25 mbar each, yielding a total pressure of 0.5 mbar in the chamber. Please note that in the given setup (both electrodes in the same atmosphere) a quantification of the produced amount of hydrogen is not feasible since hydrogen generated at one electrode is consumed at the other one. Thus the total composition of the atmosphere in the NAP-XPS chamber is not changed during electrochemical polarization of the sample. Results of electrochemical measurements 2

.1 DC measurements
In Fig. 2 of the main text the current-voltage characteristics of the LSF working electrode (i.e. a plot of the measured steady state dc current versus the applied overpotential) is shown. The overpotential η was calculated from the dc set voltages (U dc ) by subtracting the ohmic drop in the electrolyte and the contacting wires via η = U dc -R HF ×I dc .
Therein R HF and I dc denote the (ohmic) high frequency intercept measured by impedance spectroscopy (see below) and the dc current, respectively.
The current-overpotential curve in Fig. 2 is highly asymmetric with a strong non-linearity under cathodic polarization and a much weaker non-linearity in the anodic regime. Owing to this asymmetry, a fit to a function commonly applicable to electrode reactions in electrochemistry (e.g. based on exponential functions) was not successful over the entire voltage range with electrochemically meaningful parameters. To illustrate the unusual shape of the measured curve, we assumed charge transfer limited electrode kinetics with Butler- describing the current-overpotential curve. Therein I 0 is the exchange current, α the symmetry factor (with α = 0.5 indicating a symmetric activation barrier of the rate limiting charge transfer step), and z the number of transferred charges; e 0 , k B , and T denote elementary charge, Boltzmann's constant, and absolute temperature, respectively. A comparison of the measured data and simulation results is shown in Fig. s1. The current values at strong anodic and strong cathodic polarization can only be described by using significantly different i 0 and z values in the B-V simulation. The part of the curve between 0 and -100 mV, however, cannot be explained by these functions at all. Only when assuming rather unrealistic parameters (1-α = 0.85, z = 2) this part of the I-η relationship can be simulated (but without explaining the data measured at high polarization). As a consequence, the steep increase of the current at cathodic polarizations beyond -20 mV is a strong indication of a fundamental change in the reaction mechanism at the electrode surface.
It should further be emphasized that the dc curve shown in Fig. 2 -apart from a small degradation directly after the first heating of the sample -could be measured reproducibly.
(This is also true for the electrode's surface resistance measured by impedance spectroscopy. Here, also only during the first couple of measurements some minor degradation of the surface resistance could be observed.) Irreversible changes of the electrode during the electrochemical experiment can thus be excluded as origin of the asymmetrically shaped curves.

Impedance spectroscopy
Typical impedance spectra under equilibrium and upon polarization are depicted in Fig. s2.
Qualitatively, each of the spectra consists of the same features: (i) a Z re -axis intercept, (ii) a relatively well separated (semicircle like) shoulder in the high frequency regime, (iii) an additional shoulder in the medium frequency range and (iv) a slightly depressed low frequency arc. The high frequency axis (R HF ) intercept can mainly be assigned to the resistance of ion conduction in YSZ [7] and a ca. 5 Ω resistance of the measurement wires and contacts. The electrode polarization (dc measurements) was corrected for the ohmic losses caused by this high frequency offset resistance (R HF ) -see above. The very small high frequency semicircle like shoulder is most probably caused by an ion transfer resistance at the electrode|electrolyte interface and an according interfacial capacitance [7] and partly also the counter electrode may contribute to this impedance feature. In agreement with a similar study on SrTi 0.7 Fe 0.3 O 3-δ electrodes [8] the medium frequency shoulder and the low frequency arc are attributed to in-plane electron transport within the MIEC thin film and the electrochemical reaction at the MIEC surface, respectively. The capacitance involved is the chemical capacitance of the mixed conducting electrode material LSF. [9] In all cases considered here the low frequency arc and thus the surface reaction resistance dominates the polarization resistance of the LSF electrode. Therefore we can conclude that the polarization of the working electrode is virtually homogeneous for the geometry used in this study. As shown in Fig. s2 the low frequency arc is decreased by any dc bias. This bias dependence of the differential impedance fits well to non-linear current-voltage characteristics and is a common behavior of electrode reactions. Uncommon, however, is the very strong bias dependence already found for rather small cathodic bias values, see -47 mV in Fig. s2. For evaluation of the resistance of the electrochemical reaction at the electrode surface, the low frequency part of the spectra (i.e. only the low frequency arc) was fitted to the equivalent circuit shown in the inset of Fig. s2a. In this circuit the oxygen exchange reaction at the LSF surface and the chemical capacitance of LSF are represented by the resistor R surface and the constant phase element [10] CPE, respectively. The serial resistor R offset accounts for all resistive contributions with higher relaxation frequencies and is thus larger than R HF . Even though this equivalent circuit strongly simplifies the impedance response of the investigated electrodes, a reasonable estimate of the dominating surface-related resistance R surface is possible for homogeneous polarization. [8] 3 XPS results (i) Between about -20 mV and the highest anodic overpotential two pronounced peaks at ~710 eV and ~724 eV binding energy (BE) were observed, which can be attributed to Fe2p 3/2 and Fe2p 1/2 , respectively. Both binding energy values are in reasonable agreement with literature values of Fe +II and/or Fe +III . [11] A Fe2p 3/2 satellite at ~716 eV is weakly visible in the spectrum measured without electrochemical polarization and corresponds well to a Fe +II species. This suggests a significant amount of Fe +II at the surface, while bulk-sensitive studies on La 0.6 Sr 0.4 FeO 3-δ [12] indicate mostly Fe +III to be present. A high reducibility of oxide surfaces was already observed for ceria-based anodes [13] and highlights the importance of in-situ investigations. [14] (The binding energy of an additional feature at ~730 eV principally accords to the satellite of Fe2p 1/2 . However, owing to its high intensity and relatively sharp peak shape additional contributions -e.g. of Auger lines -cannot be excluded.)

Fe2p and valence band spectra
(ii) In addition to the peaks described above, at cathodic overpotentials more negative than -20 mV two further peaks at about 707 eV and 720 eV were observed. These peaks can be related to Fe2p 3/2 and Fe2p 1/2 of metallic iron, respectively. [11] It should be mentioned that by variation of the measurement position beam damage of the LSF was excluded to be the origin of the described chemical changes during XPS experiments.
XPS valence band spectra of LSF were recorded with a photon energy of 140 eV (the photon energy was select in that way, that the kinetic energy of the emitted photoelectrons is the same as in case of Fe2p spectra and therefore the information depth is identical) and at the same electrochemical polarization as the Fe2p spectra -see Fig. s3b. Interestingly, for cathodic polarization leading to metallic iron peaks in the Fe2p spectra also a Fermi edge appeared in the very low binding energy range (see arrows in Fig. s3b). Correlation of these two changes in the XPS spectra can be regarded as a strong evidence for the near-surface formation of a metallic Fe phase. It should be emphasized that the evolution of a metallic Fe species at the electrode surface did not correspond to an irreversible decomposition of the LSF electrode.
Rather, a (surprisingly) reversible behavior was found in electrochemical and XPS experiments (cf. Figs. 2 and 3a of the main text). The Fermi edges appearing under cathodic polarizations are indicated by arrows.

Sr3d, La3d and O1s spectra
Besides Fe2p spectra also La3d, Sr3d and O1s spectra were recorded at each electrochemical polarization state -they are depicted in Fig. s4. However, only minor changes occurred in these XPS spectra, which were by far not as pronounced as in case of Fe2p and valence band edge spectra. Apart from some minor shifts in binding energy (which may be due to insufficient BE correction owing to a missing Fermi edge) no significant changes in the La3d, Sr3d and O1s spectra were found. This observation further supports the assumption that the metallic iron formed under cathodic polarization is not part of the perovskite phase. Otherwise much more pronounced changes in the La3d, Sr3d and O1s spectra are to be expected.
Near-surface Sr3d spectra measured on (La,Sr)CoO 3-δ (LSC) electrodes in oxygen atmospheres under electrochemical polarization were reported in Ref. [15]. In the cited study two Sr species -one at lower BE and one at higher BE -were observed and interpreted as lattice and surface species, respectively. Similarly, in our reducing atmosphere the shape of the spectra may be explained by contribution of two Sr species (cf. Fig. s4c). However, a quantitative analysis of the spectra (except Fe) is beyond the scope of the present paper.

Fitting of Fe2p XPS data
Fe2p spectra were deconvoluted (software: CasaXPS) using a simplified peak model with the main objective to quantify the relative proportions of metallic and oxidic iron. For this means, only the Fe2p 3/2 peak was considered in the fit. A line-shape related to the asymmetric Lorentian was used for the oxidic and metallic contribution -commonly referred to as LF line-shape [16] . This line-shape has an exponential decay at the high-binding energy tail of the synthetic component, so the synthetic peak intensity (nearly) meets the background at the end of the fitting region. Peak parameters LF(1.3,2,10,0) and LF(0.8,2,15,0) were used for the Fe +II/+III and Fe 0 peaks, respectively. The authors are aware that this method of peak fitting is a strong simplification of the actually very complex Fe peak structure. [17,18] However, this model is -due to the low number of fitting parameters and constraints -very reliable to separate the peak contributions of metallic and oxidic iron. A fit of the Fe2p 3/2 spectra measured at -333 mV and +285 mV is shown in Fig. s5, illustrating the pronounced differences between surfaces of anodically and cathodically polarized LSF. In this fit we considered a metallic (Fe 0 ) and one oxidized iron species (Fe +II and/or Fe +III ). The resulting binding energies of the Fe2p 3/2 peak for the oxidic and the metallic Fe species are 710.3 ± 0.2 eV and 706.8 ± 0.1 eV, respectively. These results are in good agreement with literature data. [11] The quantification of the relative amounts of +II and +III species, however, was not unambiguously possible. This is at least partly owing to the energy calibration of XP Kationen einer polarisierten Elektrode liefert. [11][12][13] In situ NAP-XPS Studien wurden bereits auf Perowskit-Elektroden in oxidierender Atmosphäre [14][15][16] und auf Ceroxid-basierten Elektroden unter reduzierenden Bedingungen durchgeführt. [17][18][19] [5,6] .