All chemicals were used without further purification. Sodium chloride (NaCl), calcium chloride (CaCl₂), magnesium chloride (MgCl₂), and sodium ferrocyanide (Na₄Fe(CN)₆) were purchased from Sigma-Aldrich. All solutions were prepared with deionized water from a Milli-Q ultrapure water-purification system (resistivity > 18.2 MΩ·cm, Millipore). An anion exchange membrane (AEM) (ASE, ASTOM) and a cation exchange membrane (CEM) (CSE, ASTOM) served as separators. Porous carbon electrodes (Pureechem) were employed in the preliminary electrochemical water softening tests, while carbon felt (ACC-5092-10, Kynol Europa GmbH) was used as the electrode material in the scaled-up custom-made redox-ED cell. Anion exchange resin (AXR) with a total capacity (minimum) of 1.3 eq/L (Cl⁻ form) (A400, Purolite) and cation exchange resin (CXR) with a total capacity of 1.9 eq/L (Na⁺ form) (HCR-S/S, DOWEX) were utilized as IXRs in the subsequent redox-ED ion removal study.

Three major categories of electrochemical tests were performed in this study: (1) tests within the Preliminary Study of Water-Softening Redox-ED System (Section 3.1), (2) electrochemical water softening using a scaled-up redox-ED system and its optimization, and (3) tests within the Effects of Conductive Ion Exchange Resin (Section 3.5). For each experiment, suitable configurations of electrochemical cells were set up. Detailed cell designs and configurations for the tests in the Preliminary Study of Water-Softening Redox-ED System and the Effects of Conductive Ion Exchange Resin can be found in the Text S1 of Supplementary Materials.

For electrochemical water softening with a scaled-up redox-ED system and its optimization, a custom-made redox-ED cell was designed based on prior research [28], with modifications (see Fig. S1a and S1d in the Supplementary Materials). The thickness of the feed channel, referred to as the “middle channel” (MC), was adjustable by stacking MC units (0.2 mm each, 100 × 100 × 0.2 mm³ inside). The cell included two SCs (100 × 100 × 20 mm³ inside) functioning as the positive and negative electrode channels, respectively, and one MC (100 × 100 × 0.4 mm³ inside) for the feed solution channel. The MC was effectively isolated from the SCs by a CEM and an AEM to prevent stream mixing while allowing ion transfer. Three sheets of carbon felt electrodes (100 × 100 mm² each) and four titanium current collectors (100 × 100 mm² each) were alternately stacked and precisely aligned to ensure uniform voltage distribution across the carbon electrodes. The current collectors also served as spacers to enhance mass transfer of the redox couple.

In all electrochemical tests of this study, hard water containing 1.85 mM CaCl₂ and 0.65 mM MgCl₂ (i.e., 250 ppm as CaCO₃) served as the feed water. The ratio of Ca²⁺ to Mg²⁺ was established in accordance with IEC 60734:2012 [44]. Hard water feed solutions were continuously supplied to the MCs. Flow rates for MC and SC were regulated by peristaltic pumps (YZ1515x, Shenchen). Effluents were continually monitored and analyzed employing conductivity (3574-10C, HORIBA) and pH meters (9615S-10D, HORIBA), and the data underwent further processing as required. All electrochemical tests utilized a potentiostat (VSP, Biologic), recording currents and cell potentials (E_cells) throughout operations for subsequent analysis. When preparing a new redox couple solution, preliminary charging at E_cell = +1.2 V was carried out to equilibrate the species ratios in the SC. Prior to beginning ion removal steps, the effluent conductivity without applied potential was allowed to stabilize. Ion chromatography (IC) analyses were conducted employing a Thermo Scientific Dionex Aquion Ion Chromatography system with an AS-AP Autosampler (Thermo Fisher Scientific).

In the softening tests with the scaled-up redox-ED system, the cell operated in a single-pass mode for MC. In the SCs, a solution containing a total of 0.1 M ferri-/ferrocyanide redox couple (from Na₄Fe(CN)₆ salt) and 0.5 M NaCl circulated around the positive and negative electrodes in semi-batch mode at a flow rate of 30.0 mL/min. The optimization of the scaled-up redox-ED cell focused on four factors: (1) MC thickness (T), (2) MC flow rate (v_MC), and (3) cell potential (E_cell). Optimization of MC thickness (T) was conducted at a v_MC of 2.0 mL/min and E_cell = +1.2 V by varying T from 0.2 to 0.6 mm. Optimization of v_MC was performed with T = 0.4 mm and E_cell = +1.2 V by varying v_MC from 0.5 to 3.0 mL/min. Optimization of E_cell was performed with T = 0.4 mm and v_MC = 2.0 mL/min by varying E_cell from +0.8 to +1.6 V. Note that the E_cell optimization was performed with a freshly prepared SC solution without balancing. During all ion removal tests with the scaled-up redox-ED cell, potentials were applied for 2 h, except during long-term redox-ED operation. A 5.0 mL sample of MC effluent was collected 10 min before the end of the ion removal period. For prolonged ion removal in long-term redox-ED operation, the scaled-up redox-ED cell operated for 5 h of charging at T = 0.4 mm, E_cell = +1.2 V, and v_MC = 2.0 mL/min, with IC samples collected hourly since the first sampling at 40 min after the beginning of ion removal. The effluent samples were stored and later diluted for IC analysis. Detailed experimental methods for electrochemical ion removals with additional cells in the Preliminary Study of Water-Softening Redox-ED System and the Effects of Conductive Ion Exchange Resin can be found in Text S2 of Supplementary Materials.

In the softening tests with the scaled-up redox-ED system, the collected MC effluent samples were analyzed by IC for individual ion removals. Along with data extracted from the potentiostat, the C.E. was calculated using Eq. (1):

where z_i represents the charge (eq/mol) of the i-th cation, C_i,0 indicates the concentration (M) of the i-th cation in the effluent IC sample collected in a stabilized state without applied potential, C_i,n denotes the concentration (M) of the i-th cation in the effluent IC sample collected during the ion removal period, v_MC is the MC flow rate (L/min), Δt represents the time period (min) during which the effluent IC sample was collected, and I is the current (A) recorded during the effluent IC sampling period. Additionally, F is the Faraday constant (96,485 C/mol). In the C.E. calculation, not only were Ca²⁺ and Mg²⁺ from the hard water feed considered, but also Na⁺ that back-diffused from the SC. It is noteworthy that evaluating the softening performance based on the C.E.s obtained during the final minutes of ion removal was feasible, as the redox-ED cells demonstrated an almost constant reduction in effluent conductivity and ion removal at that stage.

T.H.R. was calculated using Eq. (2):

where only Ca²⁺ and Mg²⁺ removals were considered. Individual ion removals were calculated using Eq. (3):

Finally, the E.C. for hardness ion removal (kWh/kg hardness ion) was calculated using Eq. (4):

where E_cell is the applied cell voltage (V), MW_i is the molecular weight (g/mol) of the i-th ion, and only Ca²⁺ and Mg²⁺ were considered in this calculation. Detailed analytical methods for electrochemical ion removals with additional cells in the Preliminary Study of Water-Softening Redox-ED System and the Effects of Conductive Ion Exchange Resin can be found in the Text S3 of Supplementary Materials.

Before exploring the redox-ED system as a water-softening device, brief comparisons were conducted between preceding CDI-based systems and the RDF system: (1) MCDI vs. MC-MCDI, and (2) MC-MCDI vs. pre-redox-ED cells. CDI was excluded from this study. The redox-ED system was shown to have higher and prolonged removal of Ca²⁺ and Mg²⁺ compared to the preceding systems.

The first comparison (MCDI vs. MC-MCDI) highlighted the effects of an independent SC with a high-saline solution (Fig. S2a of the Supplementary Materials). In softening hard water of 250 ppm as CaCO₃, the ion removal (I.R.), C.E., and energy consumption for cation removal (E.C., kWh/kg cation) derived from MC-MCDI were 148.1%, 186.8%, and 69.8% of those from MCDI, respectively (Fig. S2b and S2c of the Supplementary Materials). The second comparison (MC-MCDI vs. pre-redox-ED) demonstrated the effects of the redox couple in the SC, which provides an additional driving force for ion removal. In the effluent conductivity profile, consistent with other studies [24,28], the pre-redox-ED system sustained high I.R. throughout 1 h of continuous charging (Fig. S3 of the Supplementary Materials).

As the initial step in the optimization of the scaled-up redox-ED system for softening processes, we evaluated the impact of T, a crucial systematic factor [16,45], on the softening performance (Fig. 1). We determined that an MC with T = 0.4 mm was optimal for achieving a high C.E. (75.9%) together with the highest T.H.R. of 96.1%.

Specifically, an increase in T resulted in a decrease in C.E., particularly when T exceeded 0.4 mm, while T.H.R. initially rose then decreased around T = 0.4 mm. A greater T allowed more time for the target ions to be separated across the CEM due to increased retention time. Yet, higher T (and the resulting decrease in conductivity inside the MC) also led to increased internal resistance [45], resulting in reduced C.E. at higher T levels. The retention time’s impact was more evident in ion removals and T.H.R., which initially improved as T increased from 0.2 mm to 0.4 mm but then declined beyond T = 0.4 mm. With the enlarged MC volume, the subdued mass transfer negatively influenced Ca²⁺ and Mg²⁺ removals, predominating the softening performance.

After determining the appropriate MC thickness, we evaluated the effects of v_MC. Among the tested v_MC values, operating at v_MC = 2.0 mL/min was identified as the optimal point where C.E. and T.H.R. were balanced with acceptable E.C. (Fig. 2). In conductivity profiles (Fig. 2a), following an initial stabilization stage, effluent conductivities remained relatively consistent but exhibited minor variations in reduction magnitude. Once stabilized, higher v_MC resulted in smaller conductivity reduction, indicating lower ion removal due to decreased ion retention time within the MC. Further comparisons in C.E. and T.H.R. (Fig. 2b) and analyses of individual cation removals (Fig. 2c) revealed the complex ion removal behavior of the system. Notably, while C.E. increased progressively as v_MC rose, T.H.R. experienced a significant decline when v_MC exceeded 2.0 mL/min. This occurs because the term Z_i × (C_i,0 – C_i,n) × v_MC in the C.E. equation (Eq. (1)) largely controls the change in C.E. and influenced by the removals of Ca²⁺, Mg²⁺, and the back-diffused Na⁺, while T.H.R. is only affected by the removal of Ca²⁺ and Mg²⁺. Particularly, as v_MC increased from 2.0 mL/min to 3.0 mL/min, Na⁺ removal (C_Na+_,0 – C_Na+,n) remained essentially unchanged, despite a 1.5-fold increase in v_MC. Conversely, the removals of Ca²⁺ and Mg²⁺ noticeably decreased due to shorter retention times, negating the impact of the v_MC increase. Consequently, the influential term for Na⁺ overshadowed those of Ca²⁺ and Mg²⁺, resulting in an enhanced C.E. when v_MC increased from 2.0 mL/min to 3.0 mL/min. The back-diffusion of Na⁺ proved mildly beneficial in our system. In membrane-based systems, insufficient charge transfer (i.e., ion removal) across IEMs can elevate internal resistance, consequently increasing the E.C.. This problem can be more pronounced in softening systems, asdivalent ions usually exhibit lower diffusivities (or mobilities) within CEMs due to their strong interactions compared to monovalent ions [46–50]. Nonetheless, in our water-softening redox-ED system, the slow transfer of divalent ions was partially offset by the removal of back-diffused Na⁺ [51].

E.C. was compared as the last parameter (Fig. 2d). Interestingly, E.C. at v_MC = 1.0 mL/min was significantly higher than the others due to excessive ion removal within the MC at v_MC = 1.0 mL/min, which substantially increased the internal resistance across the MC. This increase in E.C. at lower feed flow rates can occur once the ion removal exceeds a certain threshold. For instance, although typical changes in feed channel flow rates did not impact the energy consumption of systems [28,45], one study recorded an increase in energy consumption (kJ/mol) when the flow rate was reduced from 20 mL/min to 10 mL/min, the lowest in that study [28]. Fortunately, in our study, the rise in C.E. and the decline in T.H.R. coincided at v_MC = 2.0 mL/min, where E.C. also exhibited gradual changes. This confirmed that v_MC = 2.0 mL/min is the optimal MC flow rate for our redox-ED configuration.

We optimized the E_cell for the redox-ED cell by varying E_cell around +1.2 V (Fig. 3). The redox-ED cell operated with a substantial amount of fresh SC solution to mitigate the effects of the ratio changes between redox couple species in the SC during operation. Through a series of experimental comparisons, we ascertained that E_cell = +1.2 V is optimal for the removal of Ca²⁺ and Mg²⁺ in our system. We also underscored the importance of maintaining a balanced state of redox couple species in each SC.

In the effluent conductivity profiles (Fig. 3a), operations with E_cell < +1.2 V failed to stabilize their conductivities, whereas other conditions succeeded. This instability resulted from inadequate participation of the redox couple in ion removal, yielding profiles akin to those found in CDI-based systems (i.e., EDL-based ion adsorption and electrode surface saturation). Although the redox-ED can enable additional ion removal mechanisms with the aid of the redox couple in its SC, ion removal primarily depends on EDL formation if the potentials are insufficient to facilitate redox reactions. Consequently, a gradual increase in effluent conductivity was observed, becoming more notable at lower E_cell. Conversely, starting from E_cell = +1.2 V, the conductivities remained stable throughout the operations, highlighting the increased involvement of the redox couple and confirming its role in supporting prolonged ion removal. Additionally, as ion removal activities (both Faradaic and non-Faradaic) improved with increasing E_cell, the lowest conductivity value recorded during an operation continued to decrease, albeit marginally.

We also observed that an unbalanced state of the redox couple (e.g., negligible Fe(CN)₆³⁻ in the CEM-side SC) can hinder the Faradaic ion removal mechanism of the redox-ED, particularly for divalent ions, even at a higher E_cell (≥ +1.2 V). In the individual ion removals (Fig. 3c), the removal rates of all three cations increased up to E_cell = +1.2 V but plateaued thereafter. As explained previously, redox couples participate in the ion removal process beyond a certain E_cell through continuous charge transfers at each electrode, providing additional driving forces for ion removal. However, in operations with fresh SC, there may not be a sufficient amount of Fe(CN)₆³⁻, the reactant at the negatively charged electrode during the ion removal period, near the electrode surface. The lack of a charge mediator in the SC not only increases internal resistance around the electrode but also fails to provide a meaningful Faradaic ion removal driving force. The absence of the Faradaic mechanism in the redox-ED with fresh SC (Fig. 3b and 3c) was more pronounced when compared to ion removals from the redox-ED with balanced SC (Fig. 2b and 2c). Under identical operational conditions (v_MC = 2.0 mL/min and E_cell = +1.2 V), Ca²⁺ and Mg²⁺ removals from the balanced system were 142.6% and 144.5% of those from the fresh system, respectively. Note that the transfer of divalent ions through IEMs is relatively slower than that of monovalent ions (refer to Section 3.3. Effects of MC Flow Rate). In our redox-ED system, the Faradaic reactions could accelerate the removal of Ca²⁺ and Mg²⁺ with the balanced SC. Removal of Na⁺ was already high even without sufficient redox reactions, consistent with findings from other redox-ED studies [28]. These observations suggest that a redox-ED with an unbalanced redox couple species would face increased internal resistance and subsequent performance degradation when focusing on water softening.

Additionally, previous studies have reported that mass transfer limitations in the SC contribute to increased system resistance [24]. In our case, each electrode in the operation with balanced SC had sufficient reactant amounts, while most of the redox couple in the fresh SC existed primarily as Fe(CN)₆⁴⁻. This scenario resembles the mass transfer limitation state, which partly explains the limited ion removal with fresh SC. Furthermore, it has been reported that increasing E_cell can prevent the selective removal of Ca²⁺ over Na⁺ across the CEM [7]. In the presence of both Na⁺ and Ca²⁺, the CEM preferentially adsorbed Ca²⁺, leading to Ca²⁺ depletion near the membrane. As E_cell increased further, this depletion became more severe, potentially causing back-diffusion of Ca²⁺ from the brine channel to the desalinated channel. Given the relatively stronger interaction of divalent ions with the CEM and that membrane-based electrochemical ion removals are significantly influenced by the CEM, the reported phenomena may also account for the inhibition of T.H.R. increase in our redox-ED operation with fresh SC solution at E_cell ≥ +1.2 V.

We closely examined the effects of E_cell in terms of changes in internal resistance. Generally, greater ion removal leads to increased current and significant reductions in effluent conductivity [52]. However, during this phase, the conductivity profiles displayed only minor increases as E_cell was raised in our experiment. Interestingly, even with an increase in E_cell, the currents measured at stable states remained largely unchanged, particularly beyond E_cell = +1.2 V (data not shown). We attribute this phenomenon to the use of feed water with an initially low ion concentration (1.85 mM CaCl₂ and 0.65 mM MgCl₂), combined with the substantial increase in internal resistance during ion removal. Numerous CDI-based studies utilize feed waters with higher concentrations than our study [53]. For instance, MCDI is typically recommended for treating waters with low to medium ion concentrations; however, the optimal range is suggested to be between 3,000~4,000 mg/L [54]. The rise in internal resistance during the treatment of diluted feeds is a critical concern in CDI-based systems [24,55]. When operated at E_cell = +0.8 V, the redox-ED cell already demonstrated substantial reductions in effluent conductivity (Fig. 3a), suggesting that further ion removal from hard water at this stage would likely escalate internal resistance within the MC during redox-ED operations, thereby limiting further enhancements in softening performance.

These complex electrochemical behaviors also influenced E.C., which was lowest at E_cell = +1.2 V (Fig. 3d). E.C. continuously decreased up to E_cell = +1.2 V, after which it began to rise. Reduction in E.C. with increases in E_cell up to +1.2 V resulted primarily from enhanced removal of Ca²⁺ and Mg²⁺ through non-Faradaic processes. Beyond E_cell = +1.2 V, the increase in Ca²⁺ and Mg²⁺ removals plateaued, immediately resulting in rising E.C.

After extensive experimentation and comparison, we established the optimal operational conditions as T = 0.4 mm, v_MC = 2.0 mL/min, and E_cell = +1.2 V. We then operated the redox-ED cell under these conditions for 5 hours, taking samples every hour with the first sample at 40 min. During long-term operation, ion removal remained consistently stable (Fig. 4). Following the initial stabilization, effluent conductivity was consistently low. T.H.R. measurements, conducted five times at regular intervals, gradually stabilized, culminating in a T.H.R. of approximately 90%. This demonstrated the redox-ED system’s capabilities for water softening and stability, while also providing insights into the operational factors impacting softening performance.

In a further study, we assessed the possibility of integrating conductive IXRs within the MC of the redox-ED system for water softening. Prior research suggested that conductive IXRs in the dilute channel could enhance the transport of ions [56] and/or mitigate increases in ohmic loss in ion-removal cells, thereby ensuring consistent desalination performance [28,57]. A combination of AXR and CXR was incorporated in the MC of a small-sized redox-ED cell (Fig. S1a, S1b, and S1c of the Supplementary Materials).

Regardless of the MC flow rate (v’_MC), the redox-ED cell with conductive IXRs within the MC (termed “resin-redox-ED”) demonstrated relatively larger conductivity variations during charging and discharging than the identical cell structure without IXRs (termed “control”) (Fig. 5). The resin-redox-ED cell’s C.E.s were between 130.6 and 131.5% of those from the control. The effects of conductive IXRs were more pronounced in the individual ion removals, particularly for Ca²⁺ and Mg²⁺. At v_MC’ = 0.5 mL/min, Na⁺, Ca²⁺, and Mg²⁺ removals in the resin-redox-ED cell were 183.9%, 219.5%, and 279.5% of those from the control, respectively. Furthermore, at v_MC’ = 1.0 mL/min, the resin-redox-ED cell achieved Na⁺, Ca²⁺, and Mg²⁺ removal rates of 112.5%, 318.5%, and 259.7% of those from the control, respectively.

The enhanced removal of Ca²⁺ and Mg²⁺ was attributed not only to the buffered ohmic loss across the cell but also to the greater affinities of IXRs for divalent cations [7,49,58]. Specifically, a portion of the divalent ions was first attracted to the IXRs, and subsequently transferred toward and through the CEM. Interestingly, this increase in divalent ion removal is notable because filling the MC with IXRs physically reduces the MC’s volume and the retention time of the hard water feed, which should theoretically impede the removal of Ca²⁺ and Mg²⁺ according to prior studies. Indeed, during the v_MC optimization of the scaled-up redox-ED cell, the removal of Ca²⁺ and Mg²⁺ was significantly declined when v_MC exceeded 2.0 mL/min (Fig. 5). We therefore conclude that the attraction of divalent ions by the conductive IXRs played a pivotal role in ion transfers within the MC, counteracting the adverse effects of reduced volume. These findings clearly demonstrate that employing conductive IXRs in the redox-ED system can substantially enhance its softening performance by facilitating charge transfer within the MC and selectively trapping Ca²⁺ and Mg²⁺. We demonstrated that IXRs are a viable option for water-softening redox-ED cells to improve both the quality (i.e., C.E. and T.H.R.) and quantity (i.e., feed channel flow rate) of treated water.

Further optimization of the redox-ED cell is possible, notably by adjusting the number of electrodes in the stack. Typically, increasing the number of electrodes in a stack enhances ion removal performance and proves more efficient than augmenting the number of unit cells [28]. However, issues such as electrode imperfections, stack misalignment, uneven voltage distribution, or biased water flow must be considered, as these could degrade performance despite an increased electrode count [45]. Thus, in this study, we maintained a constant number of stacked electrodes in a single redox-ED cell to minimize potential issues arising from altering the electrode quantity. Another consideration involves the IEMs used in the redox-ED cell. Although the redox-ED system employs ion removal mechanism of CDI-based approaches, the effectiveness of such membrane-based electrochemical ion removal heavily relies on the IEMs comprising the cells [7]. In membrane-based systems, target ions initially traverse the IEMs via electro-diffusion, being influenced by the interactions between the ions and the IEMs [59]. However, as our focus is on practically optimizing the water-softening redox-ED device, an in-depth study on IEMs (e.g., alterations to polymer functional groups) would exceed the scope needed to demonstrate the system’s feasibility.