Imaging the Heterogeneity of the Oxygen Evolution Reaction on Gold Electrodes Operando: Activity is Highly Local

Understanding the mechanism of the oxygen evolution reaction (OER), the oxidative half of electrolytic water splitting, has proven challenging. Perhaps the largest hurdle has been gaining experimental insight into the active site of the electrocatalyst used to facilitate this chemistry. Decades of study have clarified that a range of transition-metal oxides have particularly high catalytic activity for the OER. Unfortunately, for virtually all of these materials, metal oxidation and the OER occur at similar potentials. As a result, catalyst surface topography and electronic structure are expected to continuously evolve under reactive conditions. Gaining experimental insight into the OER mechanism on such materials thus requires a tool that allows spatially resolved characterization of the OER activity. In this study, we overcome this formidable experimental challenge using second harmonic microscopy and electrochemical methods to characterize the spatial heterogeneity of OER activity on polycrystalline Au working electrodes. At moderately anodic potentials, we find that the OER activity of the electrode is dominated by <1% of the surface area and that there are two types of active sites. The first is observed at potentials positive of the OER onset and is stable under potential cycling (and thus presumably extends multiple layers into the bulk gold electrode). The second occurs at potentials negative of the OER onset and is removed by potential cycling (suggesting that it involves a structural motif only 1–2 Au layers deep). This type of active site is most easily understood as the catalytically active species (hydrous oxide) in the so-called incipient hydrous oxide/adatom mediator model of electrocatalysis. Combining the ability we demonstrate here to characterize the spatial heterogeneity of OER activity with a systematic program of electrode surface structural modification offers the possibility of creating a generation of OER electrocatalysts with unusually high activity.


S1 SEM Micrographs of the Working Electrodes
Figure S1: Scanning Electron Microscopy (SEM) image of the polycrystalline working electrode after annealing at 500 • C for 2 hours. The scale bar on the right hand bottom of each image is 50 µm. These images clearly show that the grain facets have length scales of 10s of µm. The image shown is of the electrode before any cleaning steps. Figure S2: SEM micrograph of the physical vapour deposited gold film working electrode. As described in the main text, we create these electrodes starting with an optically smooth borofloat glass substrate, then depositing 30 Å of Cr, and finally depositing 200 nm of Au. The length of the scale bar in the lower right hand corner of the image is 1 µm. Clearly the crystallites apparent in the image are structurally heterogeneous on length scales of 10-100 nanometers.

S2 Processing of the Second Harmonic Images
Measured second harmonic data in form of videos were first stabilized in ImageJ with the plugin "Image Stabiliser" using the Lucas-Kanade algorithm, the parameters that were used After image stabilisation we performed a flat-fielding procedure in MATLAB. The SH data in this paper is collected by a wide-field second harmonic (SH) microscope. As opposed to a rastering SH microscope, where every pixel is illuminated with the same laser fluence, here S3 we illuminate and collect the SH intensity from the whole imaged area at the same time.
The benefits of the wide-field approach are: i) we can effectively deliver more laser power to the sample during the time equal to one rastering cycle, which enables our approach to be orders of magnitude faster; ii) SH intensity from all pixels are comparable in real time in every image as there is no rastering delay between different pixels. The downside of the wide-field illumination is that the illumination has a Gaussian like intensity distribution in the x and y direction, i.e. in the plane of the surface, and therefore does not deliver the same laser fluence to every pixel of the field of view. To compensate for this we use a flat-fielding procedure to simulate a homogeneous illumination so we can directly compare the SH intensity pixel to pixel. For this reason, we acquire the SH beam profile immediately before every experiment.
The flat-fielding procedure is then a simple division of acquired SH images with the SH beam profile image.

S3 Full Cyclic Voltammogram and Integrated SH vs. Potential Curve
Prior work on the oxygen reduction reaction (ORR) on gold has shown that this reaction leads to a cathodic current starting at 0.9 V vs RHE that increases with decreasing potentials in alkaline media in the absence of mass transfer limitations. This reaction is irreversible at 0.9 V and below and thus is expected to add a negative 'tilt' to all CVs on Au in which O 2 is present in the electrolyte. 1 We see such a 'tilt' at potentials 0.7 V vs. RHE and below suggesting that our electrochemical cell contains dissolved O 2 (see the lower panel of Figure S3 for data).
Between 0.7 V vs. RHE and the start of oxidation at ≈ 1.3 our CV appears to show small current features. Similar features have been observed previously -they are generally more pronounced in alkaline electrolyte than acid 2,3 -and attributed to surface defects, or nonequilibrated surface domains. 4,5 The spatially integrated second harmonic intensity is plotted as a function of potential S4 (along with the CV) in Figure S3. Clearly with increasing potential the SH intensity increases up to 0.85 V vs. RHE, is relatively stable between 0.85 and 1.1 and drops abruptly at potentials above 1.3: the onset of oxidation. The signal recovers at 1.15 V vs. RHE (with the dissolution of the oxide) and is afterward reversible. This general behaviour of the integrated SH signal is in agreement with previous reports: 6,7 at comparable fundamental wavelengths, these and other authors also found an approximately linear increase of SH intensity with increasing potential until the oxidation of the surface where a sharp drop in intensity was reported.

S4 SH Micrographs during Potential Cycling Cathodic of Au Oxidation
The SH images at potentials below oxidation show features with a length scale of tens of µms (see Figure S4 for data). The length scales of these features are similar to those observed in SEM micrographs (see data in Figure S1). Interestingly, as shown in Figure S4, the dependence of integrated second harmonic intensity on potential varies widely from feature to feature.
Prior work suggests a variety of possible reasons why SH intensity vs. potential profiles on Au might show such strong spatial variability. Polycrystalline gold typically exhibits facets of (111) or (110) like surfaces. 8 These crystal faces have strongly distinctive work functions, solvation structures, ion sorption behaviour and oxidation/reduction characteristics. These differences are reflected in such electrochemical observables as the point of zero charge for gold single crystal electrodes: much prior work suggests they vary with surface structure up to 400 mV. 9,10 Adding to the complication we expect that, in addition to having both (111) and (110) like crystal faces, our sample may contain domains of either type that are rotated with respect to each other. A strong azimuthal dependence of SH intensity vs. potential profiles has been demonstrated for gold single crystal electrodes suggesting that the histogram of domain orientation with respect to the laboratory reference frame will strongly influence the measured second harmonic intensity. 11

S5 Appearance, Growth and Detaching of Oxygen Bubbles and Calculating their Volume
As described in the main text, increasing the potential positive of 2 V vs. RHE leads to the formation of oxygen bubbles on the electrode surface. We monitor this formation by observing the bubbles and their shadows. In most cases the bubble itself is hardly visible because it extends well above the focal plane (which we place on the Au/electrolyte interface) and is S6 Figure S4: Surface heterogeneity as apparent in the second harmonic generation micrographs together with SHG intensity vs. potential curves in different positions of the surface. The electrolyte was 0.5 M Na 2 HPO 4 , pH = 9, the scanning speed 60 mV/s. thus out of focus. In contrast, the shadow is always in focus, because it lies in the plane of the electrode surface. An example where bubble and shadow can be seen together is Figure 3 of the main text. The offset between an oxygen bubble and its shadow was taken into account by determining the centre of a bubbles shadow and then correcting in MATLAB assuming a spherical bubble and accounting for the 34 • angle of incidence.
The shadows are slightly elliptical due to illuminating at 34 • , therefore the shadow diameter perpendicular to the direction of illumination was used for calculating the bubble volume.
The amount of charge that is passed in order to form a bubble was calculated as follows.
Given the measured diameter of a bubble we calculate its (assumed spherical) volume. The volume is then used to calculate the amount of oxygen molecules inside the bubble assuming pure oxygen and a pressure of 1 atm. It is worth noting that 1 atm is a lower bound of bubble pressure: bubbles with a radius of ≈ 10 µms (the average size of the bubbles, in the moment they detach from the surface, evaluated for Figure 2 of the main text) will be slightly higher due to Laplace pressure. However, this additional pressure is only 0.14 atm (with larger bubbles Figure S5: Illustration of an oxygen bubble on the gold surface. Panel A) shows a bubble and its shadow in side view along with the collimated illumination at 34 • . In panel B), the surface, bubble and shadow are seen from above. It is apparent, that the real centre of the oxygen bubble is shifted from the centre of the shadow by a size specific offset. Figure S6: The life cycle of an oxygen bubble from appearance to detachment. The same FOV is shown as in Figure 2 of the main text. The last frame before the oxygen bubble detached from the surface was used to determine its diameter, which was taken perpendicular to the direction of the illumination.

S8
having still lower pressures). From the amount of oxygen molecules we calculated the number of electrons consumed per bubble and then convert this number into a current. The conversion from current into a current density requires calculation of a size of the active area and is discussed further below.

S6 Is Bubble Growth Dominated by Diffusion, or the Activity of the Active Site?
As discussed in the main text, it is not immediately apparent that the formation of bubbles exclusively in the active area is actually a consequence of its superior activity compared to the surroundings. One might alternatively imagine that bubble nucleation occurs from a supersaturated solution at nucleation sites that are particularly active because of roughness, or a specific composition. In this scenario bubble growth would be dictated by mass transport, i.e.
convection or diffusion, of dissolved O 2 to the forming bubble. In such a scenario, the growth rate of the bubble is expected to be a function of bubble surface area. Evaluating the growth rate of a bubble over the course of 26 s during the linear sweep experiment of Figure 2 of the main text clarifies that this is not the case: within experimental error the growth rate is relatively insensitive to bubble size (see Figure S7 and note that the possible slight increase in growth rate for large bubbles may be the result of the increase in bias during the course of bubble growth. As it is a linear sweep experiment with a sweep rate of 1 mV/s, the potential at the end of the bubble growth will be 26 mV higher than at the start). We take this argument to suggest that the growth of bubbles, and the activity of the active area, are a consequence of the high intrinsic OER activity of the active area. Figure S7: Growth rate of a bubble versus its surface area during bubble growth. The bubble was part of the linear sweep depicted in Figure 2 of the main text, nucleated at around 2.08 V vs. RHE and grew for about 26 s before detaching. The experiment was conducted in 0.5 M Na 2 HPO 4 , the sweep rate was 1 mV/s.

S7 Estimation of the Current Associated with Formation of further Oxide and the Dissolution of Gold
It is in principle possible that the current features we observe may result from Au oxidation, the OER, double layer charging or some combination of all three. 12 Because current due to double layer charging should be small relative to Au oxidation, we address only the first two possible contributions. The amount of oxide that is formed during the linear sweep we estimate to be 330 V. This is a slight underestimate of the amount of oxide formed in our experiment due to the lower potential limit. Linear sweep voltammetry experiments described in the main text were performed at 1 mV/s. The rate of gold dissolution at this scan rate is estimated to be 0.5 ng cm 2 13 (the lowest scan rate measured by the authors was 2 mV/s and extrapolation to 1 mV/s suggests 0.5) which translates to a gold dissolution current of 1.67 microamps (assuming oxidation of S10 Au 0 to Au 3+ ) for our sample.
The combined currents of oxide formation (assuming one monolayer) over the course of the experiment and gold dissolution are compared to the measured current in Figure S8. It is apparent that the great majority of the current measured above 2 V vs. RHE comes from the OER. Figure S8: Comparison of the measured current (red curve) with the combined currents of oxide formation and gold dissolution (green curve) during a linear sweep experiment. The experiment was conducted in 0.5 M Na 2 HPO 4 , the sweep rate was 1 mV/s.

S8 Determination of the Size of the Active Area
Nucleation initially leads to bubbles that are only a few nanometers large, 14 well below the resolution limit of the microscope. While such bubbles cannot be resolved, we can identify their presence by intensity fluctuations of a few pixels against an otherwise static background.
As might be expected, these fluctuations are spread over a circular area with a radius of 1-2 µms, but at this stage no clearly spherical shape can be identified and the centre of the bubble can not be determined reliably. The bubbles grow further and at around 3 µm in radius, the bubbles appear spherical. The centre of each bubble's nucleation sphere was then marked S11 with a red dot in Figure 2 of the main text and a yellow cross here in Figure S9. In the early stages of bubble growth (before it can be discerned as spherical), the bubbles geometry is ill defined and it is reasonable to assume an area for the bubble nucleation. For this purpose, we assume a circular nucleation area around the centre (blue circle), whose radius is defined by the standard deviation of a bubbles radius. This standard deviation was determined by fitting the the size of bubbles nine times by hand and then calculating the standard deviation from these nine values (the standard deviation obtained in this fashion is 1.8 µms). All nucleation sites lie in close vicinity, most nucleation areas overlap with neighbouring ones. This area was termed "active area", because, with few exceptions, every bubble originated from there. An estimate for the geometrical area of this active area is then obtained by choosing an envelope that contains all nucleation areas for which it appears sensible to be grouped together. This envelope is shown in black in Figure S9. For the present case the enveloped area was 69.6 µm 2 . The size of the active area determined in this manner is presumably not a function of the size of the actual active sites of the OER, rather this active area is determined by the effective spatial resolution of our images.
It is obvious that the calculated active area will linearly affect the current densities that are plotted in Figure 2 of the main text and that it depends to some degree on our choice of envelope (black line). However, this does not effect the argumentation in the main text for two reasons: i) the assumption of a spherical nucleation area with a radius of 1.8 µms is an overestimate that we are using because our microscope does not resolve the bubble nucleation on the submicron scale; ii) choice of a different envelope with the same nucleation areas (blue circles) would result in an active area that varies by ± 10 %, which would in turn alter the current density by a similar percentage. However, this can not explain the two orders of magnitude difference between the current densities of active areas and the average electrode. S12 Figure S9: The size of the active areas was determined by assuming a radius of bubble nucleation of 1.8 µms around every nucleation center (yellow crosses), overlapping these areas (blue circles) and taking the envelope (black). The radius of nucleation was determined from the standard deviation of how precisely we can determine a bubbles size.

S9 Can the Outstanding OER Activity of Active Areas be Explained by SHM Data Alone?
In an effort to explain the outstanding activity of active areas, cyclic voltammetry was performed from 0 -1.9 V vs. RHE at 60 mV/s in 0.5 M Na 2 HPO 4 , pH = 9, while performing SHM.
In so-called spatial correlation maps, we plot the correlation coefficient of every single pixel with the average FOV in a specified potential range. If a pixel has a value close to 1, then it performs like the average FOV and has a yellow hue. The farther the value is from one, the more it deviates from the average behaviour and is coloured blue. These maps identify areas that perform like the average FOV and areas that deviate from the average behaviour and should therefore highlight areas with distinctive surface chemistry. This analysis was done using the MATLAB function corrcoef().

S13
The correlation coefficient of A and B is defined: Where every element of vector A is the averaged SH intensity of all N pixels in the field of view at a given potential. Vector B represents a single pixel with SH intensities as entries for every potential value. µ is the mean and σ is the standard deviation. The correlation coefficient can be rewritten in terms of the covariance of A and B: Where cov(A,B) is defined as: From the definitions of covariance and standard deviation (not written out here) it is clear that the value of the correlation coefficient that we use in our analysis is normalized, therefore the differing SH amplitudes of the single pixels are accounted for.
The spatial correlation map for the potential range 1.5 -1.9 V RHE shows spatial features on a tens of µm scale that resemble the grains in the SEM images. This potential range was chosen, because the oxidation of the surface leading up to the onset of the OER is believed to be more influential on OER activity, than for example spatial differences in ion sorption before surface oxidation. The map demonstrates a distinctive oxidation behaviour of large portions of the surface, where most of it is either yellow, or either blue.
The nucleation sites for oxygen bubbles in a subsequent experiment (linear sweep from 1.8 -2.2 V at 1 mV/s) are overlaid onto the map in red circles. It is striking that most of the nucleation sites lie either directly in a transition region (from yellow to blue), or close by. However, some lie in a clearly blue, or yellow region far from a transition region. More S14 importantly, transitions from yellow to blue (or vice versa) exist over the entirety of the FOV, but oxygen bubble nucleation only occurs in a very confined area. So far, there is no straightforward explanation for the elevated OER activity of some active areas from the optical SHM data alone, but this is work in progress. The SH intensity of the area in which the bubble(s) was nucleated was integrated and displayed vs. time. Figure S11: Observation of the second type of active site under potential cycling conditions, corresponding to experiment one of Figure 4 in the main text. The scanning speed was 60 mV/s, the electrolyte 0.5 M Na 2 HPO 4 , pH = 9.
Figure S12: Observation of second type of active site under constant potential conditions, as mentioned prior to Figure 4 in the main text. The electrolyte was 0.5 M Na 2 HPO 4 , pH = 9. The scanning speed prior to and after the potentiostatic period was 60 mV/s.