Introduction

Capacitive deionization (CDI) is an emerging desalination technology that adsorbs ions in water by a porous electrode based on an electric double layer capacitor (EDLC). Compared with commercialized desalination technologies such as reverse osmosis (RO) and thermal processes, CDI is known to have advantages in terms of energy consumption and maintenance for low salt water1,2.

During the operation of CDI, the ion removal performance is highly dependent on the properties of the electrodes such as the specific surface area, effective pore size, electric conductivity, water wettability, and ion accessibility1,3,4,5. In practice, CDI electrodes are generally produced in a metal plate form by coating various electrode materials on the surface of non-permeable metal plates. Among various electrode materials, carbon allotropes including activated carbons6,7,8,9, carbon fibers and aerogels10,11, carbon nanotubes12,13, and graphene14,15, have been most frequently used as CDI electrodes incorporating with metal (oxide) nanoparticles15,16,17,18 (Table S1). However, there are several challenges for the application of carbon materials as CDI electrodes that the ion transport strongly depends on the electric field-assisted diffusion, which induces a slow adsorption rate of ions onto the electrode19. The electrical double layer (EDL) overlapping or a co-ion repulsion effect also hinders the removal of ions20,21. In addition, irreversible adsorption of multivalent ions such as Ca2+, Mg2+, or SO42− also lead to a significant decrease in the ion removal capacity, especially during repeated uses22,23. These issues are mainly caused by the ion removal mechanism of the carbon nanomaterial-based electrodes: target ions adsorbed on the surface of the electrode by non-Faradaic reaction without any charge neutralization.

To resolve the above issues, both physical and chemical approaches have been explored. As an example, we reported a CNT-based hollow fiber electrode (CHF) to improve the physical characteristics of the electrode24. In this electrode, the salty water can flow through the electrode, thus, the ion transfer is mainly occurred much faster by convection, while that of a conventional plate-type electrode is slower by diffusion. As a result, the CHF electrode achieved faster ion removal kinetics (i.e., <10 min to reach maximum ion adsorption and desorption capacity) and higher ion removal capacity (58.2 mg g−1 at 500 mg L−1 NaCl solution). Despite the improvement in the study, the CHF did not solve the problem of non-selective and irreversible ion adsorption at the presence of multivalent ions such as Ca2+.

To overcome the irreversible adsorption of multivalent ions onto carbon-based CDI electrodes, various methods have been applied including the incorporation of nanoparticles and membrane-assisted CDI (MCDI). Metal nanoparticles such as MnO2, Fe3O4, Co3O4, Mn-polypyrrole composite4 and a metal-organic framework18 have been applied to form the composite structure with carbon materials and exhibited enhanced removal of ions3,4,15,17,18. However, the ion removal mechanism of these electrodes were highly dependent on a non-Faradaic reaction25, therefore, multivalent ions were still preferably adsorbed on the surface of a CDI electrode compared to monovalent ions such as Na+, as well as exhibited more irreversible desorption. For a stable and reversible CDI process under the presence of multivalent cations, the MCDI with ion exchange membranes26 or ion exchange resin coatings27 has been developed. In these electrodes, multivalent ions can be desorbed by applying a reverse potential to the electrodes, thus, the CDI performance can be sustained under the presence of multivalent ions such as Mg2+ or Ca2+. However, the performance of MCDI is significantly varied by the performance and configurations of ion exchange membranes. Moreover, MCDI cannot solve the EDL overlapping phenomenon, which reduces ion removal capacity of carbon-based, non-Faradaic ion removal electrodes.

To solve both issues in CDI electrodes, the incorporation of Faradaic reaction-based Na+-selective materials to the cathode materials can be the potential solution. When Na+-selective materials are incorporated on the cathode, the removal of multivalent ions will be effectively hindered based on the selectivity of the Na+-capturing materials. Among various candidate materials, NaTi2(PO4)3 (NTP) has been reported to possess the high ionic conductivity, reversible capacity, and chemical characteristics28,29,30,31. In addition, NTP affords not only reversible capture and release of Na+ ion into its crystal structure by Faradaic reaction, but also provides the selective removal of Na+ in the presence of multivalent cations such as Ca2+ and Mg2+32. According to the theoretical calculations, the maximum Na+ removal capacity of NTP is 4.78 mmol g−1, which is much higher than that of metal oxide-carbon composite-based CDI electrodes33.

On the basis of this rationale, NTP nanoparticles were synthesized and then incorporated into CHF structures (NTP-CHF) to provide both convective transport of ions into cationic adsorbents and Faradaic selective removal of Na+-. Moreover, the cation adsorption capacity and reversibility of NTP-CHF were evaluated at the presence of Ca2+ as a model multivalent cation to verify selective Na+ removal characteristics of NTP-CHF in the CDI process.

Results and discussion

Physicochemical properties of CHF and NTP-CHF

SEM images of CHF and NTP-CHF are shown in Fig. 1. Both electrodes have a uniform hollow-fiber structure with the outer and inner diameters around 1.4 mm and 0.8 mm, respectively. The SEM image of NTP-CHF showed that NTP nanoparticles were successfully combined and formed an effective electric network in the entangled MWCNT structure (Fig. 1b). The EDX mapping of the surface of NTP-CHF also exhibited that titanium and phosphorous, major organizing elements of NTP, were detected on the whole region of NTP-CHF (Supplementary Fig. 2). XRD analysis of CHF, pristine NTP nanoparticles, and NTP-CHF were carried out, and the results are presented in Fig. 2. It showed that the XRD pattern of NTP-CHF clearly matched with that of CHF (PDF no. 01-075-1621) and bare NTP nanoparticles (PDF no. 00-033-1296). In addition, there was no noticeable peak shift or new peak from NTP-CHF compared to CHF or the NTP nanoparticles, which indicated that the NTP and MWCNTs in the NTP-CHF structure did not experience any chemical structural changes. In summary, NTP nanoparticles were successfully anchored and incorporated in the structure of MWCNT networks in NTP-CHF.

Fig. 1: SEM images at various magnifications.
figure 1

a CHF and b NTP-CHF.

Full size image
Fig. 2
figure 2

XRD patterns of CHF, NTP nanoparticles, and NTP-CHF.

Full size image

The specific surface area and the pore characteristics summarized in Table 1 showed that NTP-CHF exhibited a decreased specific surface area compared to that of CHF due to the higher density of NTP nanoparticles than MWCNTs in NTP-CHF. Both electrodes have a mesoporous pore structure, which enables faster ion transport through pores.

Table 1 Physical properties of CHF and NTP-CHF.
Full size table

Electrochemical characteristics of CHF and NTP-CHF

Figure 3a displays the CV curves of CHF and NTP-CHF in various concentrations of NaCl. In this analysis, CHF presents the typical capacitive behaviour of a non-Faradaic reaction-based electric double layer capacitor (EDLC). However, NTP-CHF showed not only a non-Faradaic reaction but also a reduction peak at −0.9 V and an oxidation peak at −0.7 V in 1.0 M NaCl solution. The reduction and oxidation peaks indicate that NTP-CHF is successfully adsorbing and desorbing Na+ ions by a Faradaic reaction of NTP nanoparticles in addition to non-Faradaic reaction of MWCNTs28, while CHF adsorbs Na+ by only a non-Faradaic reaction. At 8.6 mM NaCl solution, the reduction peak was shifted to −1.2 V due to the decrease in ionic conductivity of the feed solution. The calculated specific capacity at the standard condition (i.e., 1 M NaCl) exhibited that NTP-CHF was significantly higher (16.4 C g−1) than that of CHF (6.3 C g−1) as shown in Table 2. At various NaCl concentrations, NTP-CHF always showed about two times higher specific capacity than CHF, and as discussed above, it is mainly originated from NTP nanoparticles in NTP-CHF. During the increase of scanning rate of cyclic voltammetry from 5 mV s−1 to 100 mV s−1 with NTP-CHF (Fig. 3b), overpotential for Na+ absorption-desorption reaction gradually magnified and specific capacity decreased from 16.4 C g−1 to 4.3 C g−1 (Supplementarty Table 2) due to the slower kinetics of Faradaic reaction of NTP nanoparticles than non-Faradaic reaction by CHF.

Fig. 3: Electrochemical properties of CHF and NTP-CHF.
figure 3

Cyclic voltammetry curves (a) in 1 M and 8.6 mM NaCl solutions at 5 mV s−1, (b) under various scanning rates, and c in the presence and absence of Ca2+, (d) Nyquist plots of CHF and NTP-CHF. Note that NTP-CHF was used as electrodes in (b) and (c), and all tests were performed in 1 M NaCl solution otherwise noted.

Full size image
Table 2 Electrochemical properties of CHF and NTP-CHF in NaCl solutions.
Full size table

The Na-selective adsorption of NTP-CHF in the mixture of Na+ and Ca2+ was evaluated by cyclic voltammetry (Fig. 3c). In 1 M CaCl2 solution, NTP-CHF did not show Faradaic reaction because NTP did not absorb Ca2+ from electrolyte. In case of 1 M NaCl or the mixture with CaCl2 (NaCl: CaCl2 = 100: 5), NTP-CHF successfully absorb and desorb Na+ ions from electrolyte with increased difference between anodic peak potential and cathodic peak potential from 172 mV to 335 mV. It is mainly caused by the selective adsorption of Na+ by NTP nanoparticles incorporated in NTP-CHF.

The electrochemical impedance spectroscopy (EIS) analysis showed that both hollow-fiber electrodes have one semi-circle at high frequency and linear region at low frequency in the Nyquist plot (Fig. 3d). Both electrodes showed similar semi-circle region with the electrode resistance of 10.7 Ω and 24.8 Ω for CHF and NTP-CHF, respectively. In the low frequency (or higher impedance) region, NTP-CHF showed lower slope than that of CHF, which implied that NTP-CHF had Faradaic capacitive regime, while CHF had a diffusion controlled adsorption regime34,35.

Capture and release characteristics of Na+ by CHF and NTP-CHF

Results from Fig. 3 and Table 2 strongly implied that NTP-CHF has a higher ion adsorption capacity than CHF. Thus, the Na+ removal capacity of both electrodes under 1.3 V were examined at 1.7, 4.3, 8.6, and 17.1 mM NaCl solutions. As shown in Fig. 4, NTP-CHF always exhibited higher Na+ adsorption capacity not only than CHF, but also than other carbon-based electrodes (Supplementary Table 1). It is due to the highly developed pores in a hollow-fiber structure as previously reported24, and the Faradaic ion adsorption characteristics of the NTP nanoparticles in NTP-CHF. More specifically, as shown in the CV analysis in Fig. 3, CHF adsorbs Na+ ions by the non-Faradaic reaction, while NTP-CHF could adsorb Na+ by both the non-Faradaic reaction of MWNCTs and the reversible Faradaic reaction by NTP nanoparticles.

Fig. 4: Na+ adsorption capacity of CHF and NTP-CHF at various NaCl concentrations.
figure 4

Error bars represent standard deviations.

Full size image

In general, the Faradaic reaction such as Na+ adsorption by NTP nanoparticles exhibits slower reaction kinetics than the non-Faradaic reaction. To evaluate the adsorption kinetics of NTP-CHF and CHF, Na+ adsorption capacities of NTP-CHF and CHF were tested under the 8.6 mM NaCl solution at the 1.3 V of applied potential. As shown Fig. 5a, both CHF and NTP-CHF reached their maximum adsorption capacity within 15 min, and NTP-CHF showed 1.5 times higher (2.2 mmol g−1) adsorption capacity than that of CHF. The Ragone curve of NTP-CHF located at the upper-right side of CHF (Fig. 5b), thus NTP-CHF had higher Na+ adsorption rate and capacity than those of CHF at the same condition36. It is due to the highly porous network of MWCNTs which provides sufficient electrons for the Faradaic Na+ adsorption reaction of incorporated NTP nanoparticles, and results in fast reaction kinetics comparable with the MWCNT-only electrode (CHF).

Fig. 5: Na+ adsorption/desorption characteristics of hollow-fiber electrodes under 8.6 mM NaCl solution.
figure 5

a Adsorption kinetics, (b) Ragone plot, (c) Galvanostatic curves for NTP-CHF at various current densities, and (d) Cyclic adsorption-desorption curve.

Full size image

The electrochemical characteristics of hollow-fiber electrodes were also evaluated by galvanostatic charge-discharge test (Fig. 5c). In this experiment, NTP-CHF shows the pair of Faradaic charge and discharge reaction by NTP nanoparticles, which is not observed in CHF. Thus, Faradaic adsorption of Na+ by NTP nanoparticles is responsible for the increased adsorption capacity of NTP-CHF.

The cyclic adsorption-desorption curves in Fig. 5d exhibited that NTP-CHF could be operated without a significant loss of the Na+ adsorption capacity during 50 times of adsorption-desorption cycles at 1.3 V in 8.6 mM NaCl solution. It has been reported that bare NTP nanoparticles have highly irreversible capacity in the initial charging and discharging process due to their low electric conductivity28. However, NTP nanoparticles in NTP-CHF were incorporated in the porous structure of highly conductive MWNCTs, thus, MWCNT structures provided the better ion transport as well as the faster electron transfer. After 50 cycles of ion adsorption-desorption, Na+ adsorption capacity of CHF decreased to 52% of initial value (0.76 mmol g−1), while NTP-CHF kept 68% of its initial performance (1.51 mmol g−1). These results can be found in many other studies and explained that the reduction of adsorption capacity after the repeated cyclic operation mainly came from the changes in electrochemical surface properties of MWCNTs in CHF and NTP-CHF, where NTP nanoparticles in NTP-CHF kept original properties more than CHF28,37,38. In addition to the electrochemical surface modification, irreversible adsorption of Ca2+ also plays an important role for the decrease in Na+ removal efficiencies in long-term operation of CDI in the next section.

Impact of Ca2+ ions to Na+ removal efficiencies

For the carbon-based CDI electrode, multivalent cations such as Ca2+ bound irreversibly and led to low regeneration efficiency during the repeated use of CDI electrodes22. To evaluate the impact of Ca2+ ions during the removal of Na+, the molar ratio of Na+:Ca2+ was adjusted from 100:0 to 100:5, and the ion removal capacities were measured.

As expected, the impact of Ca2+ ions became more pronounced with an increase of Ca2+ concentration as shown in Fig. 6. The Na+ removal capacity of CHF dramatically decreased from 1.48 mmol g−1 to 0.29 mmol g−1 when 0.43 mM of Ca2+ was added into the feed solution (Na+:Ca2+ = 100:5). However, at the same condition, the Na+ removal capacity of NTP-CHF decreased from 2.20 mmol g−1 to 1.19 mmol g−1. The decrease in Na+ removal capacities of CHF and NTP-CHF was due to the occupation of ion capturing sites by more strongly bound Ca2+ ions as well as by the larger hydration radius of Ca2+ than Na+, which hindered the access of Na+ ions to the capturing sites22,39.

Fig. 6: Changes in ion adsorption capacity at various Na+:Ca2+ ratio.
figure 6

Note that the Na+ concentration and the applied electric potential were fixed at 8.6 mM and 1.3 V, respectively. Error bars represent standard deviations.

Full size image

Importantly, it should be noted that the impact of Ca2+ ions was only limited to the capturing sites on MWCNTs. As shown in Fig. 6, the decreased amount of Na+ removal capacities of CHF were 0.34 mmol g−1 and 1.19 mmol g−1 at 100:1 and 100:5 of Na+:Ca2+ ratios, respectively, and the values are almost identical with those of NTP-CHF (0.22 mmol g−1 and 1.01 mmol g−1) even at various ratios of Na+:Ca2+. It strongly implied that the impact of Ca2+ ions was limited to the ion capturing sites on MWCNTs in CHF and NTP-CHF. This will be discussed further in the following section.

Impact of Ca2+ during the repeated capture-release cycles

To evaluate the cyclic impact of Ca2+ on the removal of Na+ in the CHF and NTP-CHF electrodes, the capture (15 min at 1.3 V) and release (15 min at 0 V) were repeated three times in the presence of 0.43 mM CaCl2. As shown in Fig. 7, the total ion removal capacity of NTP-CHF was decreased continuously, retained 52% of the initial Na+ removal capacity after three iterations of capture and release cycles, while that of CHF was only 10%. As discussed above, CHF quickly lost its Na+ removal capacity by the irreversible attachment of Ca2+ ions on MWCNTs during the repeated uses, while NTP-CHF could retain the Na+ removal capacity from NTP nanoparticles under the existence of multivalent cations. In Fig. 7, the decreased amount of ion removal capacities during the capture-release cycles in the presence of Ca2+ was identical in CHF and NTP-CHF electrodes as observed in Fig. 6. The decrements of the Na+ removal capacity in each cycle were 1.19, 1.30, and 1.37 mmol g−1 for CHF, and 1.00, 1.08, and 1.12 mmol g−1 for NTP-CHF, respectively. This implies that Ca2+ ions were mainly adsorbed onto MWCNT structures, and did not inhibit the electrochemical Na+ ion removal by NTP nanoparticles in NTP-CHF. The selectivity of NTP nanoparticles are firstly arose from the difference in the behaviour of alkali metal and alkali earth metal in NTP. First, in the Na super ionic conductor (NASICON) type crystal structure (NaM2(PO4)3), Ca2+ exists at the MO6 octahedral site, not at the interstitial site for Na+ 40. As a result, Ca2+ does not impede the Na+ capture-release reaction. Secondly, the ionic conductivity of multivalent ions in the NASICON crystal under room temperature is too low to transfer multivalent ions from the surface to the inner side of the crystal41. As result, it is clear that Ca2+ significantly less affects the performance of NTP-CHF as a negative electrode of CDI than CHF due to the selective Na+ removal reaction of NTP nanoparticles. Based on the overall results, NTP-CHF can be applied to an ion removal system where the selective removal of Na+ from water is required such as in drinking water production from deep seawater.

Fig. 7
figure 7

Changes in cation adsorption capacities during repeated adsorption and desorption cycles in the presence of 0.43 mM of Ca2+ in 8.6 mM of Na+ (Na+:Ca2+ = 100:5).

Full size image

Concluding remarks

In this work, NTP-CHF, a NaTi2(PO4)3-incorporated MWCNT hollow-fiber electrode, was fabricated and evaluated for Na+-selective removal electrode in the CDI process. The prepared NTP-CHF has a mesoporous and water-permeable structure, which enhances the ion migration into the electrode. The incorporation of NaTi2(PO4)3 provides synergistic effects where the MWCNTs serves as an highly conductive network for the NTP nanoparticles and NTP nanoparticles provide selective Na+ removal with preventing the irreversible adsorption of multivalent cations. The novel selective removal of NTP-CHF is due to the unique crystal structure and electrochemical characteristics of NTP, and thus, shows both higher Na+ removal capacity and regeneration efficiency than CHF at the presence of Ca2+. Overall, this study proved that the incorporation of Faradaic reaction-based electrode materials such as NTP could significantly enhance ion adsorption capacities in the CDI process treating low salt solution, and moreover, the ion removal capacity was not influenced by the presence of multivalent ions such as Ca2+ unlike conventional carbon-based CDI electrodes.

Methods

Preparation of MWCNTs and NTP nanoparticles

Functionalized multi-walled carbon nanotubes (MWCNTs, diameter: 50~80 nm, length: 10~20 μm) were purchased from Cheap Tubes (USA), and further purified using 10 M HCl solution for 18 h. NTP nanoparticles were synthesized by the solvothermal method28. Briefly, equal molar amounts of titanium(IV) butoxide and sodium acetate were dissolved into anhydrous ethanol. The solution was then slowly mixed with a phosphorous acid solution (85 wt%) to a volumetric ratio of 1:3. After complete mixing, the solution was transferred into an autoclave and reacted at 160 °C for 3 h. After the reaction, white precipitates were separated by centrifugation and washed with deionized water. Finally, NTP nanoparticles were obtained by drying the white precipitate with a freeze dryer (Labconco, FreeZone 6 Plus) and annealing at 600 °C in an Ar environment.

Fabrication of CHF and NTP-CHF

The wet-spinning technology was used for the fabrication of both a MWCNT hollow-fiber electrode (CHF) and NTP-incorporated CHF (NTP-CHF)24,42. Briefly, 4 g of MWCNTs (for CHF) or 3 g of MWCNTs with 1 g of NTP nanoparticles (for NTP-CHF) were dispersed in 34 g of N-methyl-2-pyrrolidone (NMP) by an ultrasonic homogenizer, and then 2 g of polyvinyl butyral (PVB) was added into the suspension as a dispersion stabilizer and binder. After five days of mixing, the homogeneous mixture was extruded through a hollow-fiber spinning apparatus with deionized water as a bore fluid. After the wet-spinning, hollow-fiber electrodes were dried at room temperature overnight, and then calcined at 350 °C for one hour in an Ar environment to remove PVB from the hollow-fiber electrode. It is worthy to note that >1 g of NTP nanoparticles in the spinning mixture induced a significant decrease in the mechanical strength during the preparation of NTP-CHF.

Physicochemical characterization of CHF and NTP-CHF

The microstructure and morphology of CHF and NTP-CHF were observed using a scanning electron microscope (SEM; Magellan400, FEI) equipped with an energy dispersive X-ray spectrometer (EDS). An X-ray diffractometer (XRD; Rigaku SmartLab) was used to verify the presence of NTP nanoparticles in the NTP-CHF structure. The specific surface area and pore size distribution of CHF and NTP-CHF were evaluated by a Brunauer-Emmett-Teller analysis (BET; Tristar II 3020, Micrometrics). The electrochemical characteristics of the fabricated CHF and NTP-CHF electrodes were examined by a 3-electrode cell connected to an electrochemical working station (ZIVE SP1, WonA tech, Rep. of Korea) at the room temperature. CHF or NTP-CHF was used as working electrodes with a graphite rod as a counter electrode, and a saturated calomel electrode (SCE) as a reference electrode. A cyclic voltammetry (CV) analysis was carried out to determine electrochemical characteristics of CHF and NTP-CHF in 1.0 M (58,400 mg L−1), 8.6 mM (500 mg L−1), and 1.7 mM (100 mg L−1) NaCl solutions as electrolytes with a 5 mV s−1 scan rate. Here, 1.0 M NaCl solution was selected as a standard condition to evaluate the specific capacitance of the electrode, and 8.6 mM (500 mg L−1) and 1.7 mM (100 mg L−1) NaCl solutions were tested as feed solutions for the CDI process to meet the drinking water standard. The selective adsorption of Na+ by NTP-CHF was measured in 1.0 M NaCl, 1.0 M CaCl2, and 1.0 M NaCl with 50 mM CaCl2 solutions with a 5 mV s−1 scan rate. Unless otherwise specified, all electrode potentials are subsequently referred to as SCE.

Ion capture and release test

The ion adsorption performance of CHF and NTP-CHF was evaluated in a batch-mode CDI cell at constant applied voltage of 1.3 V. In this cell, total 140 mg of CHF was used as an anode (positively polarized electrode), where 20 mg of CHF or NTP-CHF was used as the cathode (negatively polarized electrode) after the consideration of capacitance with 5 mm of the mean distance between anode and cathode (Supplementary Fig. 1). To evaluate the selective adsorption of Na+ ions by NTP nanoparticles, Ca2+ were chosen as representative cations due to their strong binding energy and abundance in natural waters22,23,27. The molar ratio of Na+:Ca2+ was varied from 100:0 to 100:5 by dissolving various amount of CaCl2 at the fixed concentration of Na+ (500 mg L−1 or 8.6 mM of NaCl).

The concentration of Na+ was measured by ion chromatography with cation column (Dionex Aquion IC, Thermo Fischer, USA), and Ca2+ concentration was measured by a colorimetric method (HA-71A, Hach). The adsorption capacities of Na+ and Ca2+ were calculated by following equation:

$${{{\mathrm{Cation}}}}\,{{{\mathrm{adsorption}}}}\,{{{\mathrm{capacity}}}}\left( {{{{\mathrm{mmol}}}}\,g^{{{{\mathrm{ - 1}}}}}} \right) = \frac{{\left( {C_0 - C_f} \right) \times V}}{{m_{cathode}}}$$
(1)

where C0 and Cf are the initial and final concentrations (mM) of Na+ or Ca2+, respectively, V is the volume of the solution in CDI cell (L), and mcathode is the mass of cathode electrodes (g).