Enhancing sodium removal in capacitive deionization via cathode asymmetry and faradaic reactions suppression

2.1 Experimental setup

The study utilized a commercially available activated carbon cloth (ACC-5092-15, Nippon Kynol, Japan), which has a reported surface area of 1478 m²∙g^-1 and a mass of 0.20 ± 0.01 g (Uwayid et al., 2021; Kim et al., 2017; Cohen et al., 2013). The ACC cloth was used as received, without undergoing chemical pretreatment. The design of the FT-CDI cell employed in this work was detailed in a prior publication (Algurainy and Call, 2020). The anode was placed as the upstream electrode in the cell (Anode: Cathode; A:C) because ORRs occurred at much greater extents than the upstream cathode. In the symmetric electrode configuration, both the anode and cathode consisted of equal mass (0.20 ± 0.01 g for each electrode). For asymmetric designs, the cathode mass was intentionally varied to be either half or double that of the anode, resulting in three electrode mass ratios: 1A:1C, 1A:2C, and 2A:1C. In this notation, #A:#C refers to the relative number of anode (A) and cathode (C) components used. These configurations were achieved by adjusting the number of activated carbon cloth (ACC) layers within each electrode. Specifically, each ACC layer contributed approximately 0.20 ± 0.01 g of carbon mass, so mass ratios were controlled by stacking either one or two layers per electrode as needed. Each ACC layer had a diameter of 3 cm and a thickness of 0.5 mm. Before use, ACC layers are thoroughly rinsed and cleaned with deionized water to remove dust and impurities, and then dried in the oven at 105°C for at least 3 h.

2.2 CDI cycling

A 10 mM sodium chloride (NaCl) solution (>99% purity, Sigma-Aldrich, USA), prepared with deionized water, was used as the feed solution for all desalination trials. The solution was continuously supplied to the FT-CDI cell in single-pass mode at a flow rate of 3 mL/min using a peristaltic pump (Masterflex, Cole-Parmer, USA). Solution properties were monitored in real time using flow-through conductivity (ET908, eDAQ, USA) and DO sensors (NEOFOX-KIT-Probe, OceanOptics, USA) positioned at the cell outlet.

Short-term CDI cycling was selected in this study to isolate the effects of electrode mass loading asymmetry and capture the transient electrochemical behavior of the cathode, including electrode potentials, transferred charge, and the influence of the potential of zero charge (E_pzc) and cathodic side reactions (i.e., ORRs). Focusing on short-term operation enabled a clearer evaluation of cathodic Na⁺ adsorption mechanisms without interference from long-term degradation phenomena, such as carbon oxidation at the anode (Algurainy and Call, 2022).

2.3 Analytical methodology

Sodium (Na⁺) and chloride (Cl⁻) concentrations were quantified using dedicated ion chromatography (IC) systems: a cation-specific ICS-2000 and an anion-specific ICS-5000+ (Thermo Fisher, USA). Calibration employed traceable Na⁺ and Cl⁻ standards (Environmental Express, USA). Instrument detection sensitivity was verified using independent Na⁺ and Cl⁻ standards (Inorganic Ventures, USA) to ensure measured ion concentration variations in samples exceeded the analytical margin of error. Established protocols defined minimum reporting limits of 180 mg∙L^-1 for cations and 280 mg∙L^-1 for anions.

The levels of H₂O₂ in effluent samples during adsorption/desorption cycles were measured via the I₃⁻ technique publication (Algurainy and Call, 2020), selected based on its low detection threshold (<1 µM), matching the anticipated H₂O₂ concentration range.

2.4 Desalination performance

To evaluate the performance on the CDI cells in this work, two metrics were determined: ion adsorption capacity (IAC) and cell charge efficiency (CE). IAC, defined as the adsorption quantity of a specific ion (e.g., Na⁺ or Cl⁻) per unit mass of the total electrodes, was determined as

where Q denotes volumetric flow rate (mL/min), C_o and C_t represent influent and effluent concentrations of ions (μmol/L), t is time, (min), and m is the mass of all ACC layers added to both electrodes (g).

CE quantifies the ratio of cumulative salt removal to total charge transfer through the cell (Eq. 3). This parameter evaluates the proportionality between electrical energy consumption and ion removal efficacy, highlighting losses to competing mechanisms (e.g., Faradaic reactions).

where F equals the Faraday constant (96,485 C/mol), C_o and C_t denote influent and effluent NaCl concentrations, respectively, (mg∙L^-1), M is the molecular weight of sodium chloride (g/mol), and i(t) indicates instantaneous current (A) at time t (s).

3.1 Working potential window

To understand how electrode configurations influence electrochemical behavior in CDI, the effect of varying carbon mass loading on cathode and anode potential windows during operation was first examined. The working potential windows of the electrodes were determined and analyzed using three voltage values; (1) electrode potentials at the end of charging (cathode; E_ca and anode; E_an), (2) short-circuited (E_o) potentials at the end of discharging, and (3) potentials of zero charge (the cathode; E_pzc- and anode; E_pzc+) at the end of cycling. In the 1A:1C configuration, the 1.2 V applied voltage was equally distributed during charging (E_ca: −0.53 ± 0.03 V) and (E_an: 0.56 ± 0.02 V) (Fig. 1). During discharging, the short-circuit potential (E_o) of both electrodes was 0.2 ± 0.01 V. Under asymmetric cell configurations, increasing the number of ACC layers also proportionally increased the geometric thickness of the electrode, which could influence flow resistance and the distribution of the electric field within the cell (Algurainy and Call, 2022; Qu et al., 2015). Indeed, the applied cell potential showed differential allocation across electrodes, with the predominant voltage fraction applied to the lower-mass electrode. This behavior was observed in the 1A:2C cell configuration, where charging-phase potentials measured −0.53 ± 0.02 V (cathode) and 0.6 ± 0.01 V (anode). In that configuration, E_o​ decreased to 0.16 ± 0.00 V relative to symmetric cell baselines. Analogous voltage partitioning behavior was observed in the 2A:1C configuration, confirming mass-dependent potential distribution trends. The cathode potential was more negative (−0.68 ± 0.01 V), and the anode was less positive at only 0.49 ± 0.00 V. The E_o value in that cell increased to 0.29 ± 0.01 V, resulting in only a 0.2 V anode potential window.

Close

Fig. 1.

The potential window of the cathode and anode electrodes in CDI cells operated at 1.2V during charging and 0V during discharging. Three cathode-to-anode mass ratios were investigated: (a) 1anode:1cathode, (b) 1anode:2cathode, and (c) 2anode:1cathode, where the numbers represent the layer(s) of ACC added to the electrode. Potentials of zero charge of the cathode (E_pzc-) and anode (E_pzc+) were determined after the operation completed (10 charging/discharging cycles) for each cell configuration. Anode E_pzc was plotted as a domain within the electrode potential window because the measured value was a potential range rather than a distinct value. Highlighted ovals indicate the dynamics of anode potentials, when the anode electrode was operated at potentials less positive than E_pzc+, corresponding to a period dominated by co-ion release rather than counter-ion adsorption. The dynamic of the cathode potential is not shown because the electrode was almost always operated at potentials more negative than E_pzc- during the entire charging phase.

Export to PPT

Measurements of E_pzc of the cathode and anode were also determined in each configuration. E_pzc is a key electrochemical property of the electrode and is defined as the specific electrode potential at which the net surface charge on the electrode is zero. At this potential, the surface of the electrode has no excess positive or negative charge, leading to no electrostatic attraction or repulsion of ions. Operating the anode (cathode) at potentials that are more positive (negative) than the electrode E_pzc, would result in counter-ion attraction (i.e., Cl^- is attracted by the anode). Prior to cycling, the E_pzc value for the pristine electrode (E_pzc,o) was - 0.1 ± 0.01 V. After operating the cells, the E_pzc of the cathode remained stable and independent of cell configuration at 0 ± 0.05 V, but the anode E_pzc shifted over time towards more positive values. Notably, in contrast to electrode potential variations, the magnitude of the increase in the E_pzc of the anode remained nearly stable and did not change among cell configurations. In particular, the anode E_pzc in the symmetric, 1A:2C, and 2A:1C cells were 0.3 ± 0.12, 0.3 ± 0.11, and 0.3 ± 0.11 V, respectively, which were not significantly different (p = 0.06).

Since changing the symmetry of electrodes resulted in noticeable changes in the electrode potentials but not the electrode E_pzc, new scenarios for the potential window of electrodes were created. For example, in the symmetric 1A:1C cell, the anode E_pzc after shifting was more positive than E_o. This shift affected the anode potential window in such a way that the initial adsorption window of ǀE_anode – E_oǀ altered, and a co-ion (Na⁺) expulsion window (i.e., ǀE_anode – E_pzcǀ) existed in this cell. Based on real-time measurements of electrode potentials, it required the anode potentials 2.66 minutes to be more positive than E_pzc, which implies that the anode during that time did not attract counter ions (Cl^-), but rather repulsed co-ions (Na⁺).

In the 1A:2C cell, although the shift in the anode’s E_pzc and the position of E_o resulted in a Na⁺ expulsion window (i.e., ǀE_anode – E_pzcǀ) that was slightly wider than that in the symmetric cell, the repulsive time of that window (i.e., the time corresponding to the period when the anode was polarized less positively than its E_pzc) was clearly shorter (1.97 minutes for the 1A:2C cell compared to 2.66 minutes). The shorter time in the asymmetric cell was likely due to the smaller anode requiring a more rapid potential change, as it served as the minority electrode and thus needed a greater driving force to overcome its limited charge transfer efficiency and compensate for its lower mass. The anode in the 1A:2C configuration showed a stronger polarization with more positive potentials by the end of charging compared to the symmetric 1A:1C cell. This elevated polarization generated an intensified electrostatic driving force, consequently yielding higher electrode charge density.

On the other hand, in the 2A:1C cell, the potential window of anode ǀE_anode – E_oǀ was minimized because the cathode required a greater electric driving force to overcome the electrode mass limitation. To meet this requirement, the anode potential was not only polarized at less positive potentials but also E_o was shifted more positive. This new scenario of E_o with respect to E_pzc reduced the (E_o – E_pzc) window from 0.1 V in the 1A:1C cell to 0.01 V, which was significantly different (p < 0.05). This decrease in the (E_o – E_pzc) domain reduces the time to release co-ions (Na⁺) from the anode by about three times. Besides, the cathode potential window in this configuration was 0.97V (33% larger than the symmetric cell), which translates into greater electrical driving force. These modified scenarios likely prompted adsorption of ions (Na⁺ removal).

3.2 Dissolved oxygen and hydrogen peroxide

Given the significant changes in electrode potential windows under different mass loadings, the potential influence of these variations on parasitic Faradaic processes was subsequently investigated, with particular focus on ORRs. To evaluate the presence and extent of ORRs, both DO and H₂O₂ were monitored in the effluent of each cell. While a decrease in DO suggests oxygen consumption, the detection of H₂O₂ confirms the occurrence of a two-electron ORR pathway commonly observed in CDI systems (Zhang et al., 2019). The concentrations of DO were measured over time at the effluent in each configuration. In the 1A:1C cell, the DO dropped from 8.09 ± 0.13 to 1.26 ± 0.49 mg∙L^-1 after 10 min of charging and then nearly plateau at the same value (Fig. 2a). In the asymmetric configurations, similar DO profiles were observed. In particular, the DO rapidly decreased during charging and nearly leveled at 1.52 ± 0.18 in the 1A:2C and 1.74 ± 0.26 mg∙L^-1 in the 2A:1C cells, which, based on statistical analysis, were not significantly different from the symmetric configuration (p = 0.07).

Close

Fig. 2.

Effluent DO (a) and hydrogen peroxide (b) measurements in each cell configuration during charging.

Export to PPT

The DO profile in FT-CDI cells can be influenced by electrochemical processes, such as ORRs at the cathode (He et al., 2016; Holubowitch et al., 2017; Zhang et al., 2018). Those electrochemical reactions can consume oxygen. For example, according to Eq. 4, oxygen is reduced at the cathode to generate hydrogen peroxide (H₂O₂), leading to a decrease in DO levels. To examine the presence of that reaction, H₂O₂ concentrations were measured over time in the effluent. In the 1A:1C cell, 72.5 ± 3.1 µM H₂O₂ was generated by the end of charging (Fig. 2b). H₂O₂ concentrations slightly increased to 87.7 ± 5.4 µM in the 1A:2C cell and 82.5 ± 2.6 µM in the 2A:1C cell, but, based on statistical analysis, these values were not significantly different from those of the symmetric configuration (p = 0.11).

Overall, DO and H₂O₂ results clearly showed that the extent of ORRs remained stable in the asymmetric cells despite the increased cathode surface area in the 1A:2C cell or the stronger electrical driving force at the cathode in the 2A:1C cell. One possible explanation for these results is that DO was already depleted in solution in the symmetric 1A:1C cell. Hence, the additional cathode surface area or driving force scenarios in the asymmetric cells did not promote ORRs (Fig. 3).

Close

Fig. 3.

Conceptual diagram showing the measured (solid lines) and hypothesized (dashed lines) DO gradients in the symmetric flow-through CDI (FT-CDI) configuration, along with the ORR schemes (left part). The scenarios created by the asymmetric FT-CDI configurations, used to explain the suppression of ORRs, are shown on the right. ‘2x’ indicates a cathode mass twice that of the anode, while ‘0.5x’ indicates a cathode mass half that of the anode. Concepts adapted from Algurainy and Call (2020) and Holubowitch et al., (2017).

Export to PPT

3.3 Sodium removal and charge efficiency

Having demonstrated that variations in electrode potentials affect the extent of Faradaic reactions at the cathode, the influence of these potential shifts on capacitive ion adsorption was subsequently evaluated by assessing ion removal performance and charge efficiency across the tested configurations. In particular, the removal of Na⁺ was determined during the charging cycle in each cell configuration. 59.5 ± 14.5 µmol of Na⁺ per gram of carbon was removed in the symmetric cell at the end of charging. Interestingly, Na⁺ removals increased by 98.7 ± 3.4% in the 1A:2C and 83.2 ± 5.1% in the 2A:1C cells relative to the symmetric cell. In particular, 117.8 ± 4.1 and 108.3 ± 2.4 µmol.g^-1 were removed in the 1A:2C and 2A:1C cells, respectively, at the end of charging (Fig. 4).

Close

Fig. 4.

The adsorption capacity of Na⁺ as a function of cell configuration. Error bars represent the standard deviation derived from triplicate experimental replicates.

Export to PPT

Several studies have claimed that a strong competition between cathodic reactions (Na⁺ adsorption vs. ORRs) occurs at the cathode over the electrons generated and transferred from the anode. This competition results in a poor Na⁺ removal and therefore decreases the overall cell charge efficiency in FT-CDI (Algurainy and Call, 2020; Shapira et al., 2016; Suss et al., 2015). To investigate this hypothesis and determine if the improved Na⁺ adsorption capacity in the asymmetric cells was associated with a more efficient usage of charges, the cell charge efficiency was measured. With the addition of an ACC layer to the symmetric cell, more charges are transferred from the anode to the cathode in the asymmetric cells. In particular, the total number of charges increased from 12.4 ± 0.4 C in the 1A:1C configuration to 16.2 ± 1.3 C (˜ 30.6% increase) and 17.5 ± 1.5 C (˜ 41.1% increase) in the 1A:2C and 2A:1C cells, respectively. After measuring increases in the total charges in the asymmetric cells, the usage of those charges (cell charge efficiency) was then examined (i.e., how well electrical current passing through the cell contributed to the removal of ions versus enhancing other processes, such as faradaic reactions). There were considerable improvements in cell charge efficiency in the case of asymmetric cells. The charge efficiency increased from 38.1 ± 1.6% in the 1A:1C configuration to 59.2 ± 0.8% (˜ 59.4% increase) in the 1A:2C and 52.7 ± 2.4% (˜ 40.8% increase) 2A:1C cells (Fig. 5). Overall, these increases in cell charge efficiencies in the asymmetric cells indicated a more sufficient usage of the total charge toward salt ions removals (Table 1).

Close

Fig. 5.

Charge efficiency of the cathode towards Na⁺ adsorption vs. ORRs (primary y-axis) and the overall charge efficiency of the cell (secondary y-axis) for each cell configuration. For the cathode charge efficiency, the total Na⁺ removals were translated into the total charge consumed for capacitive process at the cathode (one mole of e^- consumed per one mole of Na⁺ removed) while both DO and H₂O₂ measurements at the effluent were used to estimate the total faradaic charges consumed to reduce O₂ (two moles of electrons consumed per one mole of H₂O₂ generated and four moles of e^- consumed per one mole of the remaining DO, which was calculated as, DO_remaining = DO_influent - DO_effluent - DO_H2O2).

Export to PPT

Table 1. Summary of various electrochemical parameters and desalination performance metrics for each cell configuration.

Cell orientation	Anode potential, E+(V)	Cathode potential, E- (V)	Anode potential of zero charge, Epzc+ (V)	Cathode potential of zero charge Epzc- (V)	Co-ion release time period (min)	Charges transferred during co-ion release (% from the total charge)	Effluent dissolved oxygen (mg/L)	Effluent H2O2 (μM)	Na+ adsorption capacity (nmol/g)	Cell charge efficiency (%)
+: anode; -: cathode	0.56 ± 0.02	-0.53 ± 0.03	0.3 ± 0.12	0 ± 0.05	2.66	7.1 coulombs (40.1%)	1.26 ± 0.49	72.5 ± 3.1	59.5 ± 14.5	38.1 ±1.6
	0.6 ± 0.01	-0.53 ± 0.02	0.3 ± 0.11	0 ± 0.05	1.97	4.9 coulombs (24.1%)	1.52 ± 0.18	87.7 ± 5.4	117.8 ± 4.1	59.2 ±0.8
	0.49 ± 0.00	-0.68 ± 0.01	0.3 ± 0.11	0 ± 0.05	0.87	2.3 coulombs (12.2%)	1.74 ± 0.26	82.5 ± 2.6	108.3 ± 2.4	52.7 ±2.4

The improved charge efficiencies in the asymmetric cells can be attributed to two main reasons. First, the portion of electrical charge consumed for co-ions repulsion was reduced. In particular, it was estimated from the current profile of the symmetric cell that about 40.1% of the total charge (7.1 coulombs) transferred from anode to cathode when the anode was polarized at less positive potentials than the E_pzc of the anode (the first 2.66 min in the charging phase). In the asymmetric cells, on the other hand, only 24.1% (4.9 coulombs) and 12.2% (2.3 coulombs) of the total charge were consumed to release co-ions in the anode of 1A:2C and 2A:1C cells, respectively (the time when the anode was operated at potentials less positive than its E_pzc). To further support this discussion regarding the effect of anode E_pzc on charge efficiency, the effluent solution conductivity profiles over the time corresponding to the potential window of co-ions release (E_o – E_pzc) were compared. If electrodes were polarized inside that window, this should be seen as a rise in conductivity readings during early times in the charging phase. Indeed, the magnitudes of the inversion peaks, shown in Fig. 6, agreed well with the unwanted charges (those dedicated for co-ions release) quantified above (1A:1C > 1A:2C > 2A:1C). The large reductions of this kind of charge in the asymmetric cells could be attributed to the response of electrodes potentials to varying the mass ratio of electrodes in a two-electrodes cell. In the 1A:2C cell, the minor anode required a faster potential dynamic (i.e., anode was the minority electrode and demanded a higher electrical driving force to counter its ineffective charge transfer and compensate for the mass limitation). Although the potential window of co-ions release (E_o – E_pzc) in that cell was larger, compared to in the symmetric cell, the dynamics of E₊ was much faster and therefore spent about 26% less time (1.97 minutes) to pass E_pzc. In the 2A:1C, the scenario of E_o with respect to E_pzc reduced the (E_o – E_pzc) window from 0.1 V in the 1A:1C cell to only 0.01 V, which translates into less time and therefore charge to pass that co-ions repulsion window.

Close

Fig. 6.

Effluent solution conductivity profile during the charge cycle as a function of cell configuration. The inversion peaks occurring at the beginning of charging are highlighted in an inset figure for better illustration.

Export to PPT

The second possible explanation for the improved cell charge efficiencies in the asymmetric cells is that since the adsorption capacity of Cl^- approximately remained independent of cell configuration (143.5 ± 7.9 µmol of Cl⁻ per gram of carbon in the symmetric cell; 126.5 ± 3.6 µmol/g in the 1A:2C; 142.5 ± 16.9 µmol/g in the 2A:1C cell), it is expected that the improved charge efficiency in the asymmetric cell was primarily associated with an improvement in Na⁺ adsorption. To investigate this hypothesis, the cathode efficiency (i.e., the distribution of the charges towards Na⁺ adsorption vs. ORRs) was evaluated. This is the first attempt in the CDI field to determine the charge efficiency of the system via the efficiency of cathode to remove Na⁺ using parameters other than those used to obtain the conventional charge efficiency metric (total salt removed using measured conductivity divided by the charge passing through the cell as current).

Since the removal of Na⁺ in CDI during the charging phase occurs in the cathode, the efficiency of cathode to (1) adsorb Na⁺ (i.e., how much of the total charge transferred to the cathode was balanced by Na⁺ adsorption), and (2) facilitate ORRs were determine and compared for the first time in the CDI literature (Fig. 5). To determine this, it was assumed, under the operational conditions in this study, that the main reactions occurring in the cathode were Na⁺ adsorption and ORRs. Co-ions release was not considered to be a main reaction in the cathode because of the dynamic states of the cathode potentials in all the cells, as well as the stable cathode E_pzc values. According to real-time measurements of electrode potentials and E_pzc, the period of times when the cathode was operated at potentials less negative than the electrode E_pzc (i.e., the time when the charges were balanced by co-ions repulsion but not counter-ions adsorption) remained in milliseconds time-frame, and that the amount of charge passing as current during that time was less than 1% of the total charging generated. To estimate the total charge transferred to the cathode in each cell, the IC results was used to account for the capacitive charge (one mole of Na⁺ removed translated into one mole of charge consumed) and both DO and H₂O₂ measurements to quantify the faradaic charge. For the faradaic charge, the approach used by Holubowitch et al. (2019) to estimate the efficiency of FT-CDI to deoxygenate salt solutions was followed (Holubowitch et al., 2019). Briefly, the authors first estimated the portion of charges consumed to produce H₂O₂ by converting measured effluent H₂O₂ concentrations (Fig. 2b) into an equivalent quantity of charge according to Eq. 4. Next, from the remaining DO, which was calculated as, DO_remaining = DO_influent - DO_effluent - DO_H2O2, another equivalent amount of charge based on the four electron ORRs pathway (Eq. 5 or 6) was quantified. The summation of these two charges quantifies was donated as the faradaic charge, which was then added to the capacitive charge to result in the total charge transferred to the cathode. To determine the efficiency of the cathode to adsorb Na⁺ and reduce O₂, the capacitive charge and the faradaic charge were divided by the total charge in the cathode, respectively. The distribution of charge that was utilized for ORRs in the cathode dramatically decreased from 70.6 ± 4.1% in the 1A:1C to 43.5 ± 1.2% and 42.7 ± 2.4% in the 1A:2C and 2A:1C cell, respectively (Fig. 5). In other words, the efficiency of the cathode to electrosorb Na⁺ increased from 29.4 ± 0.9% in the symmetric cell to 56.5 ± 2.3% and 57.4 ± 1.7% in the 1A:2C and 2A:1C asymmetric configurations. It is noteworthy that the observed differences in cathode charge efficiency across the tested configurations were substantially greater than the potential measurement error associated with the DO sensor used in this study (±2-4% of the actual value).

Moreover, these results clearly showed an excellent agreement between the cell charge efficiency and the efficiency of the cathode to utilize charges for Na⁺ adsorption. In particular, when testing FT-CDI in short-term runs with monovalent salts, the results indicate that someone can predict a cell charge efficiency through the efficiency of the cathode to electrosorb Na⁺. In other words, the system charge efficiency is indeed controlled by the efficiency of the cathode to remove Na⁺. In addition, these results strongly suggest that minimizing the efficiency of ORRs in CDI is a key role in improving Na⁺ removal and therefore system charge efficiency. Further research is needed to validate these findings under more complex operational conditions (e.g., long-term cycling) as carbon oxidation typically accelerates over time and competes with Cl^- removal at the anode, which could have a major impact on salt adsorption capacity and charge efficiency.