Surface Electron-Hole Rich Species Active in the Electrocatalytic Water Oxidation

Iridium and ruthenium and their oxides/hydroxides are the best candidates for the oxygen evolution reaction under harsh acidic conditions owing to the low overpotentials observed for Ru- and Ir-based anodes and the high corrosion resistance of Ir-oxides. Herein, by means of cutting edge operando surface and bulk sensitive X-ray spectroscopy techniques, specifically designed electrode nanofabrication and ab initio DFT calculations, we were able to reveal the electronic structure of the active IrOx centers (i.e., oxidation state) during electrocatalytic oxidation of water in the surface and bulk of high-performance Ir-based catalysts. We found the oxygen evolution reaction is controlled by the formation of empty Ir 5d states in the surface ascribed to the formation of formally IrV species leading to the appearance of electron-deficient oxygen species bound to single iridium atoms (μ1-O and μ1-OH) that are responsible for water activation and oxidation. Oxygen bound to three iridium centers (μ3-O) remains the dominant species in the bulk but do not participate directly in the electrocatalytic reaction, suggesting bulk oxidation is limited. In addition a high coverage of a μ1-OO (peroxo) species during the OER is excluded. Moreover, we provide the first photoelectron spectroscopic evidence in bulk electrolyte that the higher surface-to-bulk ratio in thinner electrodes enhances the material usage involving the precipitation of a significant part of the electrode surface and near-surface active species.

Electrode preparation. Graphene was grown by chemical vapor deposition (CVD) on a 20 µm thick Cu foil (Alfa Aesar 99.8%) as catalysts and CH 4 (diluted in Ar and H 2 ) at 1000 °C using an Aixtron BM Pro (4 inch) reactor yielding a continuous polycrystalline film with grain size in the range of ~20 µm, which was confirmed by scanning electron microscopy. The graphene layer was fixed to 500 nm of Poly(methyl methacrylate) PMMA (4 wt.% in anisole, 950k molecular weight) deposited by spin coating. After that, the copper support was eliminated by floating on a 50 mM aqueous solution of (NH 4 ) 2 S 2 O 8 . The graphene/PMMA layer was floated in deionized water (DI-water) and transferred onto another graphene/copper foil and dried at 50 °C for 5 minutes. The resulting sample was floated again in the (NH 4 ) 2 S 2 O 8 solution to remove the copper foil before rinsing in DI-water. The PMMA/BLG layer was transferred to a Norcada ® Si 3 N 4 TEM grid with 500 nm diameter holes or onto a SiN x membrane 100 nm thick for the TFY measurements 1,2 . Finally, the PMMA was eliminated with acetone and the adherence between the BLG and the substrate is due to Van der Waals interaction ensuring stability for the electrodes. It produces a continuous film that can be used as an electrode in aqueous environments as electrocatalytic applications among others 3,4 .
Fig. S1 A and B show the SEM measurements of the free standing graphene on the Si 3 N 4 grid. The influence of the defects in the graphene layer (i.e. holes, tears etc) is suppressed by the addition of a second graphene layer. Raman spectroscopy was used to check the lattice vibrations of the crystallized BLG which is sensitive in order to determine the graphitic character of this sample. Fig. S1 C shows the Raman spectra of different reference samples such as HOPG, single layer graphene (G), graphene oxide (GO), reduced graphene oxide (RGO), and the transferred BLG used in this study. The origin of the graphitic Raman spectrum is well established and is accepted that it presents three first order bands between 1000 cm -1 and 3000 cm -1 : D band at ~1350 cm -1 , G band at ~1580 cm -1 with a shoulder at ~1620 cm -1 . 5,6 The D band is associated with defects that perturb the breathing modes of carbon rings. The G band is due to the in-plane phonons at the Brillouin zone centre. The 2D band is due to excitation of two phonons with opposite momentum in the highest optical branch near the K point and is sensitive to the number of graphene layers 7 . The Raman spectrum of HOPG does not show a D band, which attributed to the small number of points defects, and attenuated 2D due to the excitation of two phonons with opposite momentum in the underlying layers. On the other hand, the graphene reference spectrum does not show a D band and the ratio of the 2D and G peak intensities, i.e., I 2D /I G is approximately ~2 indicating a good graphitization and likely the absence of more underlying layers 21 . Graphene oxide (GO) shows a I 2D /I G ratio decreases and the D peak intensity increases, indicating an increase in the defect concentration. This trend in the peak intensity is not due to the presence of oxygen, as the similar behavior of the reduced graphene oxide (RGO) spectrum proves, indicating that the origin of this behavior is the existence of defects in the graphene lattice. The absence of the D band confirms the BLG is of high-quality with a low density of defects in the graphene lattice. In addition, the reduction in the I 2D /I G ratio to approximately ~1.7 suggests the presence of an underlying graphene layer due to the excitation of phonons with opposite momentum, which reduces the intensity of the 2D.
The electrodes were deposited by sputtering using Cressington 208HR sputter coated machine loaded with an Ir target (99.99% from Elektronen-Optik-Service GmbH, Germany) in a 0.1 mbar Ar atmosphere at a current of 40 mA. The sputter time determined the thickness of the electrode: the nanoparticles were deposited for 5 s yielding the formation of a homogeneous well distributed Ir NPs in the whole surface with NPs size ranging from 2 to 5 nm in the form of metallic iridium (Ir 0 ). The 20 nm thin-film electrodes were fabricated by sputtering onto a SiN x membrane (100 nm thick and 5 mm x 5 mm, with a Si frame of 1 cm x 1 cm and 200 µm thick sourced from Norcada) in a 0.1 mbar Ar atmosphere at a current of 40 mA for 160 s (see Fig.  1A). There are stoichiometric iridium oxide NPs that remain unaltered under the electron beam, being more stable than the non-stoichiometric iridium oxide (Ir x O y ) nanoparticles. These particles are in the form of IrO 2 rutile with orientation (210) and 0.20 nm lattice constant and IrO 2 (201) and lattice plane distance of 0.26 nm both with tetragonal crystal structure, as shown in Fig. 1B. At higher magnification it was found that some of the iridium NPs are reduced to Ir 0 under vacuum conditions and in presence of the electron beam. It is a well know problem and is related to the existence of a non-stoichiometric iridium oxide (Ir x O y ) giving Ir 0 (111) cubic structure with a lattice plane distance of 0.23 nm as a product of the reduction by the beam. The thickness of the electrode was calibrated for the different sputter time, pressure and current conditions using cross section images acquired with an electron microscope. We found this procedure is more accurate than the readings provided by the micro-quartz balance in the sputtering chamber.
The electrodes were oxidized/activated for their use in OER by potential cycling by continuous cyclic-voltammogram (CV) between 0 V and 1. In situ electrochemical cell. The EC-cell used for hard X-ray measurements is shown in Fig. S2A, where the flow of liquid was assured with a peristaltic micro pump. This cell is based on a 100 nm thick SiN x membrane, transparent to the incoming X-ray photons, possible photonin/photon-out (PIPO) techniques possible in the hard X-ray regimen, thereby enabling the study of electrochemical reactions with aqueous electrolyte. The SiN x window is 100 nm thick, with an area of 5 mm x 5mm and silicon frame of 1 cm x 1 cm and 200 µm thick 9 .
For the in situ XPS characterization, the liquid flow cell 10,11 was operated inside the main chamber of the AP-XPS endstation of the ISISS of the synchrotron facility BESSY II (Berlin, Germany) under a main chamber of ~10 -7 mbar background pressure while aqueous electrolyte (0.1 M H 2 SO 4 ) is flowed (1 ml/min) on the back side of the free standing graphene membrane described above, which is used as the working electrode. The continuous flow of liquid was assured by a syringe pump, 260D Teledyne Isco (Lincoln, USA). The main body of the electrochemical cell is made of polyether ether ketone (PEEK), which is an exceptional electrical insulator and chemically inert. The sealing was assured with several Kalrez O-rings, which have good chemical stability. The cell contains two additional electrodes, the counter-(platinum) and reference-electrode (Ag/AgCl DRIREF-2SH, World Precision Instruments, USA). A crosssection of the cell is depicted in Fig. S2B. The Ir 4f spectra were collected with a kinetic energy of the photoelectrons of 600 eV at different potentials. As far as the minimum kinetics energy needed for the photoelectron to penetrate the graphene electrode and reach the detector is 500 eV (see Fig. S3). Thus the collected spectra at 600 eV excitation energy are surface informative (equivalently at around ~100 eV excitation energy in membrane free systems). At open circuit potential (OCP) the Ir 4f peak consists of two main peaks ascribed to the spin orbit split Ir 4f7/2 and Ir 4f5/2 states.
Beamlines. In situ synchrotron radiation based experiments were performed at the ISISS beamline of BESSY II in Berlin (Germany). In this facility, the photons are sourced from a bending magnet (D41) and a plane grating monochromator (PGM) yielding an energy range from 80 eV to 2000 eV (soft X-ray range), a flux of 6x10 10 photons/s with 0.1 A ring current using a 111 µm slit and a 80 µm x 200 µm beamspot size. The operando measurements were accomplished in the ambient pressure X-ray photoelectron spectrometer (AP-XPS) end-station of the ISISS beamline, which is equiped with a SPECS PHOIBOS 150 NAP hemispherical analyzer.
Hard X-ray absorption measurements at the Ir L 3 -edge were performed at the beamline BL17C1 of the National Synchrotron Radiation Research Center (NSRRC) in Hsinchu (Taiwan). The photons are sourced from a 25 poles wiggler (W20) with 20 cm period length and a focus spot size of 2 mm x 6 mm. The excitation energy ranges from 4.8 keV up to 14.2 keV. The signal was collected in transmission mode for the powder samples and in total fluorescence yield (TFY) mode for the foil and in situ EC-cell experiments, using an ionization chamber detector.
Electrolyte preparation. The electrolyte was prepared by diluting 9.8 g of H 2 SO 4 (purity 98%, Alfa Aesar, Massachusetts, USA) in 1 L of Milli-Q water (18.2 MΩ) at room temperature (RT), 25 °C. The electrolyte was continuously saturated by bubbling pure N 2 gas, which minimizes the presence of other dissolved gases in the electrolyte. The electrolyte is acidic with pH~1.
Potentiostat. Potentiometric control was assured with a Biologic SP-300 (Seyssinet-Pariset, France) allowing different potentiometric and amperometric controls. The experiments were performed in the presence of aqueous electrolyte where the measurements were acquired using the electrochemically activated IrO x electrode in presence of 100 mM H 2 SO 4 aqueous electrolyte (pH~1) saturated in N 2 (gas) at the same time that the electrolyte was continuously refreshed by a peristaltic pump (Pt counter electrode and Ag/AgCl reference), which minimizes the possibility to form trapped gas bubbles that can potentially influence the electrochemical cell performance.

Reference samples.
Commercially available iridium powders were purchased from Sigma-Aldrich (99.9% trace metals basis) and AlfaAesar (Premion, 99.99% trace metals basis). In addition, Ir foil was purchased from Mateck (99.99% metals basis). The AlfaAesar powder was used as received and the SigmaAldrich powder was washed in Milli-Q water and calcined at 800 °C in O 2 for 50 hours. The AlfaAesar powder is X-ray amorphous (IrO x ) wile the calcined sample of SigmaAldrich exhibits the rutile-type structure (IrO 2 ). More details of the sample preparation/characterization can be found elsewhere 10 . The Ir foil was subjected to several cycles of sputtering in Ar + atmosphere and annealing in H 2 atmosphere and shows a purely metallic phase.
Calculations. Density functional theory calculations were performed at the PBE+U level with U=0 using the Quantum ESPRESSO package 12 using pseudopotentials from the PSLibrary 13 with a kinetic energy (charge density) cutoff of 60 Ry (600 Ry). While PBE has been shown to accurately recover the electronic structure of IrO 2 , some electron/hole localization beyond this U=0 level may be expected 14 . In fact, using a Hubbard U < 2 eV is known to maintain good agreement between the computed density of states and the measured valence band of iridium oxide 14 and can also influence surface adsorption energies on iridates 15 . Thus, we also explored the effect of applying a Hubbard U of 1-2 eV to the Ir d orbitals using the simplified rotationalinvariant scheme implemented in Quantum ESPRESSO 16 . For bulk rutile-type IrO 2 calculations a (12 × 12 × 12) k-point mesh was used with Marzari-Vanderbilt cold smearing using a 0.01 Ry smearing parameter 17 . The structure for bulk Ir 2 O 5 was found using USPEX 18 with Quantum ESPRESSO. A starting population of 28 individuals was employed with a fixed composition of Ir 2 O 5 . The search was allowed to run until the lowest energy structure remained unchanged for 5 generations. The resulting low energy C2/m structure from this search was further optimized with Quantum ESPRESSO using a (6 × 6 × 6) k-point mesh and is shown in Fig. S4A. The structure for bulk IrO 3 was taken from the Materials Project (ID:mp-1097041). 19 The orthorhombic Cmcm structure was further optimized Quantum ESPRESSO using a (6 × 6 × 6) kpoint mesh and is shown in Fig. S4B. Rutile-type (110) surfaces were modeled with symmetric 5-layer slabs separated by 20 Å of vacuum using a (6 × 12 × 1) k-point mesh and 0.01 Ry cold smearing. Rutile-type (111) surfaces were modeled with 11-layer slabs separated by 20 Å of vacuum using a (4 × 4 × 1) k-point mesh and 0.01 Ry cold smearing. Surface calculations included spin-polarization. A 7-layer rutile-type IrO 2 (001) surface was also included to access higher Ir surface oxidation states and was computed using a (6 × 6 × 1) k-point mesh and 0.01 Ry cold smearing with 20 Å of vacuum separating periodic images. Transition states were found using the climbing image nudged elastic band method. The images were relaxed until the force dropped below 0.05 eV/Å and the change in total energy dropped below 0.02 eV between iterations. These simulations were performed using (2 × 1) and (2 × 2) supercells for the (110) surfaces, for the (111) surface, the corresponding heat of reaction was computed using a (2 × 2) supercell.
XAS spectra were computed with a resolvent-based Bethe-Salpeter Equation (BSE) approach 20 using the wavefunctions from Quantum ESPRESSO with the core-level BSE solver in the OCEAN package 21 . For these calculations normconserving pseudopotentials were used with a kinetic energy cutoff of 120 Ry. Empty bands were included to up to 200 eV above the Fermi energy (E F ). Other parameters matched the total energy calculations. All XAS spectra were broadened with a 5.2 eV wide Lorentzian to capture the lifetime broadening at the Ir L 3 edge 22 . Spectra were aligned using ΔSCF calculations 23 . The excited state of each absorbing atom was computed separately using a 2p core-hole on the absorbing atom, and the excited electron was included in the simulation. For bulk IrO 2 ΔSCF calculations were performed using a (4 × 4 × 4) supercell, while for the (110) and (111) surfaces (3 × 6) and (3 × 3) supercells were used, respectively. The spectra were aligned to experiment using bulk rutile-type IrO 2 . The effect of a Hubbard U was tested on bulk IrO 2 , the (111) surface with µ 1 -OH and the (110) surface with µ 1 -O. No experimentally observable changes were seen for U≤2 eV, see Figure SIU.
XPS spectra were computed using a Hopfield perturbation model 24 . Following reference 25 , the initial density of states (DOS) of the photoexcited atom were taken from the project Ir 5d DOS of the ground state, and the final density of states were taken from the project Ir 5d DOS of the atom with an Ir 4f corehole using the same supercells as the ΔSCF calculations used for XAS alignment. Constant Gaussian broadening (0.3 eV) was included along with energy dependent Lorentzian broadening. Lorentzian broadening used an empirical Seah-Dench like model 26 , where the energy dependent line width above E F is given by: Γ (E) = Γ 0 + Γ max [1/2+1/π arctan(e-1/e 2 )] with e = (E -E F )/(E c -E F ). The parameters Γ max , and E c were taken as 22 and 8 eV, respectively, while Γ 0 was taken as the tabulated natural line width, 0.3 eV. 27 The N 6 and N 7 edges were broadened separately and the spin-orbit splitting was taken from the free atom value. The Ir 4f shifts were computed by way of ΔSCF calculations using the total energy differences from the ground state and excited calculations. The spectra were aligned to experiment using bulk rutile-type IrO 2 .        Correlation between Ir formal oxidation state and Ir 4f shift

Fig. S8
Correlation between Ir formal oxidation state and satellite position.

Fig. S9
Ball and stick models of the initial and final states in O-O bond formation on the surfaces used in Figure 7 of the main text. Iridium are shown as grey, oxygen red, and hydrogen white.  Ir L 3 -edge spectra computed with a Hubbard U of 1 to 2 eV on Ir. The (110) surface is covered in µ 2 -O and µ 1 -O while the (111) surface has µ 2 -O and µ 1 -OH (Fig. S6E). These were chosen as the most relevant surfaces since they were assigned from experiment under OER.