Direct Imaging of Local pH Reveals Bubble-Induced Mixing in a CO2 Electrolyzer

Electrochemical CO2 reduction poses a promising pathway to produce hydrocarbon chemicals and fuels without relying on fossil fuels. Gas diffusion electrodes allow high selectivity for desired carbon products at high current density by ensuring a sufficient CO2 mass transfer rate to the catalyst layer. In addition to CO2 mass transfer, the product selectivity also strongly depends on the local pH at the catalyst surface. In this work, we directly visualize for the first time the two-dimensional (2D) pH profile in the catholyte channel of a gas-fed CO2 electrolyzer equipped with a bipolar membrane. The pH profile is imaged with operando fluorescence lifetime imaging microscopy (FLIM) using a pH-sensitive quinolinium-based dye. We demonstrate that bubble-induced mixing plays an important role in the Faradaic efficiency. Our concentration measurements show that the pH at the catalyst remains lower at −100 mA cm–2 than at −10 mA cm–2, implying that bubble-induced advection outweighs the additional OH– flux at these current densities. We also prove that the pH buffering effect of CO2 from the gas feed and dissolved CO2 in the catholyte prevents the gas diffusion electrode from becoming strongly alkaline. Our findings suggest that gas-fed CO2 electrolyzers with a bipolar membrane and a flowing catholyte are promising designs for scale-up and high-current-density operation because they are able to avoid extreme pH values in the catalyst layer.


General information
De-ionized water was used for all experiments. Detailed experimental results are available in the accompanying Excel file of the supporting information.

Methods
This section describes the engineering and methods in more detail.

CO 2 electrolysis setup
We used a 3-compartment flow cell made in our workshop to carry out the CO 2 electrolysis experiments ( Figure S1).
All tubing connections had an outer diameter of 1/8 inch. The cell components were pressed together with four M4 bolts tightened with a torque of 2 Nm. The anode flow plate was made of stainless steel and served as a current collector ( Figure S1)  The catholyte flow channel had a width of 2 mm and a depth of 4 mm ( Figure S2). The channel was cut from a PMMA sheet with a diamond milling bit to create a highly transparent surface. The GDE was manufactured by spray-coating an Ag catalyst layer on a commercial Sigracet 39BC (SGL Carbon, Germany). We described the deposition method in more detail in a previous publication. [1] The Nafion content in the catalyst layer (CL) was 20 wt%. The Ag loading was 0.94 mg Ag cm -2 . The GDE was aligned by two PET spacers with a thickness of 140 µm each. Assuming a GDE thickness of 325 µm, these spacers yielded a GDE compression of 14%. The gas flow plate was pressed against the GDE as a current collector. The gas flow channel had a width of 3 mm and a depth of 4 mm. The flow plate was sealed against the GDE spacer with butyl rubber O-rings.
The liquid outlets were located at the top to facilitate the removal of gas bubbles from the liquid channel ( Figure S3).
The gas outlet was located at the bottom of the cell to allow the drainage of any electrolyte that broke through the GDE into the gas channel during the start-up phase. The FLIM system recorded images of the catholyte channel in the x-y plane.  (1) Catholyte purge gas: N 2 purge (23 mLn min -1 ) or saturation with CO 2 (15 mLn min -1 ); (2) Liquid flow rate: 0.9 or 9.0 mL min -1 (≙ = 5 or = 50 in catholyte channel); (3) Current density: 10, 50, 100 mA cm -2 . Peristaltic pumps supplied the anolyte and catholyte to the CO 2 electrolyzer. Pulsation dampers were used to reduce the pressure fluctuations caused by the pumps. The anolyte and the catholyte channel were separated with a bipolar membrane (BPM). The backpressure of both electrolyte streams was set by check valves with a cracking pressure of 69 mbar. Both electrolytes were recirculated to their respective reservoirs, in which the gaseous products were removed with a purge gas. Magnetic stirrers were used to accelerate the mixing of the electrolyte. The bulk pH inside the reservoirs was measured with a pH meter. The CO 2 gas feed was humidified to 85% relative humidity (r.h.) at 20°C inside a bubble column and passed into the gas channel of the electrolysis cell. The gas backpressure was controlled with a manual needle valve. The composition of the cathode product gas was analyzed with gas chromatography (GC). The flow rate was measured with a bubble flow meter. The cathode potential was recorded with a Ag/AgCl micro-reference electrode for the CO 2 -saturated experiments.
The flow cell was installed in the CO 2 electrolysis setup ( Figure S4). All fluid lines were made of PE. They had an outer diameter of 1/8 inch and inner diameter of 1/16 inch.

Gas feed
The CO 2 gas feed was supplied from a gas cylinder ( Figure S4). The feed flow rate of 10 mLn min -1 (Normal conditions: 0°C, 1 atm) was controlled with a mass flow controller (F-201CV-020-RAD-3A-V, Bronkhorst BV, Netherlands). This flow rate corresponds to a maximum CO 2 conversion of 7% at 100 mA cm -2 (assuming = CO 100%). The gas was humidified to 85% relative humidity (r.h.) at 20°C inside a bubble column. [1] and passed through the gas channel. We used a manual needle valve (SS-SS2, Swagelok, Netherlands) to control the gas backpressure. This allowed us to set a flow-by flow regime at the GDE. [2,3] Electrolytes Both electrolytes were pumped with a peristaltic pump (Masterflex L/S, Cole Parmer). In-line pulsation dampers  Figure S4). The purge flow rate was set to 100 mL min -1 with a rotameter (Reference conditions: 20°C, 1 atm).

Process parameter 1: Catholyte Purge gas
The catholyte (0.4 M K 2 SO 4 , 0.1 mM fluorescent dye) was purged with N 2 (23 mLn min -1 ) or CO 2 (15 mLn min -1 ) to flush the product gases. The gas flow rate was controlled with a mass flow controller (calibrated for normal S5 conditions: 0°C, 1 atm). In the case of the N 2 purge, the catholyte had a low CO 2 saturation because the N 2 stripped dissolved CO 2 from the electrolyte. In the case of the CO 2 purge, the catholyte had a high CO 2 saturation because CO 2 bubbles were brought into close contact with the liquid.

Process parameter 2: Reynolds number / Catholyte flow rate
We set the hydraulic Reynolds number in the catholyte channel, , to 5 or 50 by adjusting the liquid flow rate of the peristaltic pump, , to 0.9 mL min -1 or 9.0 mL min -1 . We note that this Reynolds number differs from the gas L bubble Reynolds number from the main paper and the Reynolds number along a plate electrode, , used in B Section 3.3. The hydraulic Reynolds number in the catholyte channel, , is calculated with (S1). The hydraulic diameter of the channel is in mm. The superficial liquid velocity is in m s -1 . We approximate the kinematic H viscosity of the catholyte, , with the corresponding value of water at 20°C ( = 10 -6 m 2 s -1 ). [4] = H ⋅ (S1) The value of is a function of the channel width, , in mm and depth, , in mm (S2).

Process parameter 3: Current density
After setting the purge gas flow rate and the liquid flow rate, we used a potentiostat (XP20, Ivium technologies, Netherlands) to set three galvanostatic current density steps (10, 50, and 100 mA cm -2 ).

Steady state measurements
The system was given 20 min to equilibrate after setting each process parameter set. Then we recorded the cell potential with the potentiostat. For the experiments with CO 2 -saturated electrolyte, the cathode potential was recorded with a Ag/AgCl micro-reference electrode at the catholyte inlet ( Figure S4). We did not correct the potential for the ohmic resistance of the electrolyte. The bulk pH of the catholyte and anolyte were determined with a pH meter (913 pH Meter, Metrohm AG, Switzerland). The gas flow rate was measured three times with a bubble flow meter to calculate an average gas flow rate for the product mixture, . We used a syringe pump to take three mix gas samples from the product gas stream. These were analyzed with a gas chromatography (GC) system (CompactGC 4.0, Interscience BV, Nederland) to determine the volumetric concentration, , of the following gas species (CO 2 , CO, H 2 , N 2 , O 2 , CH 4 , and C 2 H 4 ). No CH 4 or C 2 H 4 was detected during our experiments.
The Faradaic efficiency of gas species (H 2 , CO) was calculated from the current density, , in mA cm -2 , the electrode area, = 1 cm 2 , Faraday's constant, = 96485 As mol -1 , the stoichiometric number of electrons exchanged, ( for H 2 and CO), and the molar flux of the species, , in mol s -1 (S4).
The molar flux of the species, , was estimated with the ideal gas law (S5), in which the gas pressure is = 1 bar, the gas temperature is = 293.15 K, and the ideal gas constant is = 8.3145 J K -1 mol -1 .

Fluorescence lifetime imaging microscopy (FLIM)
The FLIM system measured the phase-shift fluorescence lifetime, , in the sample region of the flow cell ( Figure S5) using the frequency-domain technique. [5,6] The diode laser (405 nm) serves as a light source with a modulation frequency of 20 MHz. The laser light passes through a filter and enters the spinning disk confocal unit (Crest V2, CrestOptics, Italy). The microlense disk focuses the incident light onto the Nipkow pinhole disk (70 µm diameter pinholes), which restricts the light path to the imaged section in the focal plane while scanning the sample region. [7] The focused laser light passes through the objective (5x magnification) of the microscope (Zeiss Axiovert technique, a sine curve is fitted through the 6 different intensities recorded for each pixel, from which the fluorescence lifetime is determined. [5,6] This yields a total imaging time of 450 ms per FLIM image. We calibrated the FLIM system with an in-line titration setup ( Figure S6). The dye and K 2 SO 4 concentration were the same as for the CO 2 electrolysis experiments. We used a low phosphate buffer concentration of 0.0001 M to minimize the influence of phosphate ions on . [8] The electrolyte was continuously purged with N 2 to prevent ambient CO 2 from dissolving in the electrolyte and influencing the pH. We recirculated the electrolyte through the flow cell ( Figure S3  We used a quinolinium dye [8,9] to obtain the calibration curve for the pH and ( Figure S7). The curve has aplateau in the acidic region, which makes it impossible to distinguish the pH between 3.0 and 5.5. The dye is most sensitive to changes in for the pH region between 6.3 and 9. We kept the optical FLIM parameters of the calibration and the CO 2 electrolysis experiments constant to minimize systematic errors. The electrochemical stability of the fluorescent dye is discussed in Figure S8. Using a similar dye from the 1-methyl-7-amino-quinolinium fluorophore family (Compound 2b in Bleeker et al.). [8] , we carried out a series of cyclic voltammetry scans (Figure S8 a). The dye undergoes a reversible redox reaction 0.5 V and 1 V vs. SHE. The highly reproducible peak reduction current suggests that dye is not decomposing irreversibly into degradation products of the molecule. We assume the dye used in this study (Compound 2c in Bleeker et al.). [8] exhibits a similar electrochemical stability. In addition, we sampled the bulk fluorescence intensity from a random selection of FLIM images over the course of a CO 2 electrolysis run of this study (Figure S8 b). The dye does not show a decay in intensity over the course of electrolysis while a cathode potential of 0.9 V vs. SHE or less was applied. The variation in intensity is due to changes in the catholyte pH due to process conditions. . [8] (b) Bulk fluorescence intensity in counts per second (cps) measured over the course of the CO 2 -saturated electrolysis run of this study (Compound 2c in Bleeker et al.). [8] The geometry of the catholyte flow channel and the alignment with the microscope of the FLIM system are shown in Figure S9. The straight channel segment with a length of 7.5 mm was left without active area (Figure S9 a) to ensure a fully developed hydrodynamic flow profile. The microscope was focused on the outer channel wall ( -0 plane) as a reference point (Figure S9 b). Then the control software of the microscope was used to move the focal plane to the center of the channel depth in z-direction ( -plane) to carry out the imaging.

2
To process the FLIM images, we determined the wall coordinates of the channel with intensity images (Figure S9 c). The sides of the fluorescence lifetime images were cropped with these coordinates. The PET gaskets forming the wall of the channel show a lower fluorescence lifetime than the bulk of the electrolyte, which most likely originates from the fluorescence of the gasket material. [10] We assume that the gaskets exhibit a constant fluorescence lifetime, , of about 11.5 ns. According to the calibration curve ( Figure S7), this phenomenon leads to a systematic error, which lets the edges of the image appear to be close to pH 6. Therefore, the local pH near the walls in our images is overestimated at low pH (< 6) and underestimated at high pH (> 6). ions. We calculated 1D pH profiles by averaging the pH value over the y-axis of the 2D segment. For these 1D profiles, pH avg (x) is the y-averaged value over the channel width. The corresponding standard error is  pH (x). The minimum pH of the profile is indicated by pH min ; the maximum is indicated by pH max .  Figure S11). Typically, gas bubbles form at the surface of the GDE or the BPM (Figure S11 a). They S10 grow until they reach a certain diameter and are released into the flow. This leaves the electrode surface uncovered for a while until the growth cycle starts again. Gas bubbles reduce the intensity of the fluorescence signal even if they are outside of the focal plane. The fluorescent lifetime, , of the focal plane can still be measured as long as there is a sufficient signal-to-noise ratio.
This means that is still available in proximity to growing, stagnant bubbles only casting partial shadows (Figure S11 a). However, the intensity close to the GDE often decreases because of intense shadows. In addition to bubbles forming at the surface, these shadows are caused by a lower fluorescence emission of the dye at high pH. [8] As a result, intense shadows cause a noisy signal close to the GDE (Figure S11 a). Another phenomenon of the gas bubble evolution are small, moving bubbles (Figure S11 b), which follow the flow of the catholyte. These bubbles typically just reduce the intensity slightly, so that experiences only little noise. Occasionally smaller bubbles coalesce and form large, moving bubbles (Figure S11 c) taking up most of the channel cross-section.
These bubbles cast intense shadows leading to a poor signal-to-noise ratio.

Supplementary results and discussion
This section contains an explanation of the calculations used for the results and discussion section of the main article. It further includes the pH profiles for all parameter sets.

Comparison of bubble evolution for N 2 -purged catholyte at Re = 5
Gas evolution at the surface of an electrode can influence the potential in multiple ways. First, gas bubbles forming at the surface of an electrode can increase the activation overpotential by partially blocking the electrocatalytic surface area. If a constant current is supplied to the electrode, the local current density through the remaining active area rises, which results in a higher overpotential. Second, bubbles introduce additional ohmic resistance in the electrolyte by reducing the cross-section available for ion conduction. Third, bubbles can reduce the concentration overpotential when being released from a surface because their movement induces convective mass transfer from the bulk of the electrolyte. [11] Figure S12: Comparison of bubble evolution for = 5 and N 2 -purged catholyte: Cell potential, , as a function of electrolysis cell time at the specified current density, . (a) 10 mA cm -2 . The arrows indicate drops in potential due to bubble release.
100 mA cm -2 . The average ± its sample standard deviation was evaluated from 1 min after electrolysis = cell start to the end of the current density step. The relative standard deviation (Rel. Std.) is the ratio of the sample standard deviation of to the average of .

cell cell
Bubble evolution leads periodic increases and drops in potential at micro [12] and macro electrodes. [13] We use the fluctuations in to compare the gas evolution as a function of (Figure S12). At 10 mA cm -2 , very little gas cell = evolution takes place. The drops in (indicated by arrows) show that during an electrolysis time of 22 min only cell 10 bubbles were released from the anode and cathode (Figure S12 a). Further, varies less than 0.2% around cell its average value of 2.7 V (Figure S12 a). We therefore conclude that bubble evolution and plays an insignificant role at this current density. In contrast, the amplitude and frequency of potential fluctuations are much stronger at higher (Figure S12 b and c).

Calculation of unbuffered pH limit
We calculate the unbuffered pH limit, pH unbuffered , by making the following assumptions:  OH  is perfectly mixed across the channel width (x-direction)


No neutralization with H + occurs  No homogenous buffering reactions with CO 2 take place Figure S13: Mass balance to calculate the unbuffered pH limit, pH unbuffered , over a channel segment with the height . The molar 1 flux of OH  produced in the reaction is .
To set up the mass balance, we further assume that the catholyte flows through the channel segment at a volumetric flow rate, , (Figure S13). At the inlet ( ), the initial concentration of is determined by pH feed using (S6).
We consider the pH increase over the height of a channel segment with the height (Figure S13). For example, 1 releases a uniform flux of OH  , , in mol cm -2 s -1 , which depends on according to (S7).
Over the considered electrode height of and depth of (z-direction), the cumulative flux The released OH  mixes with the catholyte stream and increases the concentration of OH  according to (S9).
The value of pH unbuffered is then calculated with (S10).

CO 2 bubble evolution at bipolar membrane
The electrolysis process splits water inside the bipolar membrane (BPM) and transports H + ions to the interface with the catholyte. If the catholyte feed is already saturated with CO 2, the reduction of local pH caused by the H + ions can lead to an oversaturation with dissolved CO 2 due to the carbonate equilibrium (H + + HCO 3 - H 2 CO 3  CO 2(aq.) + H 2 O). An example of CO 2 bubbles forming at the surface of the BPM is given in Figure S14.

Catholyte Reynolds number interferes with bubble-induced mixing
We calculate the limiting partial current density for CO with (S11). This quantity corresponds to the convective The value of is determined by the correlation (S13) of the Sherwood number, , with the Reynolds number conv ℎ over the height of the electrode, , and the Schmidt number, . [15] The height of the electrode, , is 2.5 cm. The diffusion coefficient for CO 2 in water at 25°C is . [14] CO 2 = 1.9 ⋅ 10 -5 cm 2 s -1 conv = We calculate the Reynolds number over the height of the electrode, , with (S14). We note that this Reynolds number is different than the hydraulic Reynolds number, , defined by (S1). The superficial liquid velocity, , is given by (S3). We approximate the kinematic viscosity of the catholyte, , with the corresponding value of water at 20°C ( = 10 -6 m 2 s -1 ). [4] The value of is given by (S15).
= ⋅ (S14) S14 = CO 2 = 532 (S15) The resulting values for and are listed for hydraulic Reynolds numbers of and in Table S1.
conv CO,lim = 5 = 50 The effect of increasing the Reynolds number on the is shown in Figure S15.

Calculation of CO 2 buffering capacity
We estimate the nominal pH buffer capacity of the dissolved CO 2 , , in mol s -1 with (S16). This value CO 2 ,buff corresponds to the molar flux of OH  that the dissolved CO 2 could absorb through the formation of CO 3 2-. For simplification, we neglect equilibrium reactions and assume that every dissolved molecule of CO 2 absorbs 2 molecules of OH  .

Complete set of pH profiles
The following section shows the 2D and 1D pH profiles for each parameter set ( Figure S16 -Figure S23). Figure S16: 1D pH profiles for Re = 5 and N 2 -purged catholyte. The pH profile, pH avg , was averaged over the height of the channel segment shown in the upper panel. The shaded red area indicates the standard deviation of the pH value. The minimum value of pH avg is pH min . The maximum value of pH avg is pH max . The pH value of the catholyte feed, pH feed , was measured with a pH meter. S16 Figure S17: 2D pH profiles for Re = 5 and N 2 -purged catholyte. The pH value of the catholyte feed, pH feed , was measured with a pH meter. S17 Figure S18: 1D pH profiles for Re = 5 and CO 2 -saturated catholyte. The pH profile, pH avg , was averaged over the height of the channel segment shown in the upper panel. The shaded red area indicates the standard deviation of the pH value. The minimum value of pH avg is pH min . The maximum value of pH avg is pH max . The pH value of the catholyte feed, pH feed , was measured with a pH meter. S18 Figure S19: 2D pH profiles for Re = 5 and CO 2 -saturated catholyte. The pH value of the catholyte feed, pH feed , was measured with a pH meter.  S22 Figure S23: 2D pH profiles for Re = 50 and CO 2 -saturated catholyte. The pH value of the catholyte feed, pH feed , was measured with a pH meter.