Electric field modulated ion-sieving effects of graphene oxide membranes

Abstract

Precise and selective separation of ions using two-dimensional (2D) laminar membranes is a budding research field with potential applications in water treatment, desalination, sensing, biomimicry and energy storage. However, most experiments are designed around pressure or concentration–driven processes without leveraging on the charged nature of ionic species. By applying an electric field across a 2D graphene oxide-based (GO-based) membrane, we demonstrate how ionic selectivity can be enhanced by changing the polarity or increasing the strength of the field. As opposed to traditional permeation experiments, the effects of selectivity are not limited to the membrane itself and can be dynamically tuned during ion removal by changing the strength of the applied electric field. Two types of membranes were investigated in a series of experiments using binary solutions of monovalent and divalent salts. The first is a classical, unmodified GO membrane and the second is a GO membrane with d-spacing restricted by functionalization with ethylenediamine (EDA). Monovalent ion selectivity was detected for both membranes with the EDA–GO membrane achieving a monovalent ion selectivity between 1.5–5 as compared to 1.3–3.5 for the GO membrane. Using density functional theory (DFT) calculations, we determined that the presence of EDA enhanced the formation of K⁺ and Na⁺ complexes which would cause distortions in local interlayer distances.

---

1. Introduction

Graphene oxide (GO) membranes have, in recent years, emerged as highly versatile and functional alternatives to polymeric membranes in areas of desalination,^1–4 water treatment,^5,6 ion sieving,^7–10 molecular separation^11,12 and gas sensing.^13–15 The interest, in part, comes from an urgency to resolve pressing issues of water scarcity and environmental degradation with a focus on sustainability and energy efficiency. Reverse osmosis desalination, for example, has traditionally been serviced by polymeric membranes composed of cellulose acetate or polyamide. Although these membranes can be cheaply produced, they often require high pressures to efficiently reject salt ions and allow water to permeate. Furthermore, polymeric membranes are vulnerable to scaling and biofouling effects from a variety of contaminants such as pharmaceuticals, heavy metals and organic matter.^16,17 GO membranes, on the other hand, are more resilient and offer unique opportunities to decouple the relationship between water permeation and ion selectivity.

GO membranes can be described as well-defined laminar macrostructures consisting of two-dimensional (2D) planar nanochannels between GO sheets. By varying the physicochemical properties and organisation of GO sheets, GO membranes have been derived with a variety of structural and chemical effects. When dehydrated, GO membranes are completely impermeable to gas and liquids² but in aqueous environments, these membranes allow intercalation of up to three monolayers of water.^18,19 The aqueous swelling of GO membranes can be attributed to the presence of oxidised moieties (hydroxyl, epoxy, carboxylic groups etc.²⁰) which interact strongly with water molecules. sp³ regions further serve as spacers separating adjacent sheets, giving GO membranes its characteristic d-spacing. In a seminal work,² Nair et al. showed that ultrafast permeation of water was possible due to non-oxidised regions (pristine sp²-hybridized zones) acting as near-frictionless graphene capillaries similar to that observed in carbon nanotubes. Joshi et al.^10,12 further showed that these graphene capillaries were also responsible for incredibly fast ion transport and that restriction of d-spacing enabled ionic and molecular sieving. In their work, GO membranes possessed a strict permeation cut-off for ions with hydrated radii greater than 4.5 Å which meant that smaller ions such as Na⁺ or Cl⁻ could easily pass through. To limit the passage of smaller ions, the interlayer spacing can be reduced through a myriad of schemes including incorporating heterogenous interlayers of graphene flakes,¹⁰ physical confinement using epoxy,¹⁰ nitrogen doping of GO sheets,²¹ intercalation of cations^7,8 or by covalent cross-linking of small molecules.^4,5,22 The restriction of d-spacing below the hydrated size of an ion essentially increases the energy barrier for entry and only ions with weakly bonded hydration shells are able to slip through either by reorienting its hydration layers^23,24 or through dehydration.^25–27 The concept of restricting interlayer spacing is not limited to GO and recent studies have employed 2D nanomaterials such as MoS₂ (ref. 28) and MXene Ti₃C₂T_x.^29–31 For example, Wang et al.³⁰ fabricated Al³⁺ intercalated MXene membranes which were highly resistant to swelling and exhibited outstanding ion rejection properties. In addition, permeation experiments showed fast water fluxes up to 8.5 l m⁻² h⁻¹ and high stability even after 400 h of operation. Despite this, concentration-driven desalination experiments require a long period of time for equilibrium to be met and increasing the pressure of the feed side might cause irrevocable damage to the membranes.

An alternative to pressure or concentration-driven systems is to use electric field modulated techniques. By applying a voltage across or directly on the membrane, researchers have been able to modify properties such as transmembrane or surface potentials to induce disparate transport mechanisms for different ions. As such, highly selective ion removal systems can be potentially constructed to control the composition of an output stream. However, we are currently at a stage where high ion selectivity can only be reliably achieved on nanopore devices^29,32 applicable for molecular or ionic sensing. That said, there are attempts to achieve ion selectivity on macrostructures such as membranes. In one such example, Li et al.³³ applied a voltage directly to a GO-based membrane to modulate the diffusion rates of various common ions such as Li⁺, Na⁺, K⁺ and Cs⁺ paired with Cl⁻ and SO₄²⁻. By controlling the polarity and strength of the voltage applied, the electrical double layer structures within nanochannels can be altered to enhance the diffusion of specific ions up to 7 times within a 0.5 V range. More importantly, their work highlighted the merits of using voltage to discern between different ion types based on valency. In another study, Sun et al.³⁴ investigated the permeation of ions through a GO membrane when an electric field was imposed across it. He showed how selective permeation of salts could be controlled by simply changing the direction of the electric field.

In this work, we investigate the selective removal of cations using GO-based membranes in the presence of an electric field. To direct our investigations at cations alone, we adopt a technique called capacitive deionization (CDI). In CDI, a voltage is imposed across two electrodes which induces the adsorption of anions at the anode and cations at the cathode. By superposing a GO-based membrane across the cathode, we can selectively control cation adsorption and hence, modify overall deionization. A schematic outlining this process is provided in Fig. 1(a). The use of membranes in CDI is not a novel implementation. In fact, ion-exchange membranes have routinely been used to overcome the effects of co-ion expulsion and improve charge efficiency in membrane capacitive deionization. However, this is the first time GO-based membranes were implemented in CDI to achieve ion selectivity. We showed that a more dynamic range of monovalent ion selectivities (1.5–5) could be obtained for a GO membrane with its nanochannels restricted by ethylenediamine (EDA) as compared to an unmodified version (1.3–3.5). Interestingly, there were stark differences in adsorption between a binary solution containing K⁺ and a binary solution containing Na⁺. Density functional theory (DFT) calculations performed on K⁺ and Na⁺ intercalated membranes showed different localised interlayer distortions which resulted in a reduced spacing for K⁺ and an increased spacing for Na⁺.

Fig. 1 (a) Schematic showing preferred adsorption of monovalent cations across GO/EDA–GO membrane. (b) Exploded view of CDI cell showing (1) acrylic housing, (2) carbon electrodes, (3) acrylic separator, (4) nylon mesh and (5) carbon paper or GO/EDA–GO membrane. Holes were punched into the device to allow feed solution to pass through.

2. Results

2.1 Characterization of membranes

Two types of vacuum filtered GO membranes were employed in our study, the first uses plain GO sheets and the second uses GO sheets cross-linked with ethylenediamine (EDA). The diamine molecule had previously been shown to be effective in restricting the interlayer expansion of GO sheets^35–37 in water and was hence employed in our work as a spacer molecule. Due to the atypical nature of the membranes, we refer our systems as GO/CDI and EDA–GO/CDI. The membranes were synthesized by vacuum filtration across two substrates, the first was a piece of carbon paper composed of interwoven carbon fibres and the second was a commercial PTFE membrane filter. Both substrates were hydrophilic materials which made fluid flow easier and enabled the formation of a 2D stacked membrane. As evidenced in the photograph (Fig. 2(a)), a GO-based membrane could be easily separated from its PTFE base without much damage. Fig. 2(b1) depicts a schematic of a GO/EDA–GO membrane with its underlying carbon paper scaffold and Fig. 2(c) and (d) show SEM images of a GO and EDA–GO membrane respectively. The texture of the membranes exhibited typical smooth morphologies as expected of most GO-based membranes except for the rod-like textures caused by the carbon fibres. No macroscopic defects were observed even though a minimal amount of GO/EDA–GO solution was used (∼10 mg). The thickness of the GO membrane was about 1 μm whereas the EDA–GO membrane was about 2 μm when measured along three different positions. The average mass loading of both types of membranes was 10 mg. XRD was used to study the d-spacings of the membranes in the dry state and immersed in various salt solutions. Under dry conditions, the GO membrane showed a d-spacing of 8.11 ± 0.1 Å whereas EDA–GO membrane showed an increased d-spacing of 9.28 ± 0.2 Å due to grafting of EDA. Both values were consistent with reported literature.^22,38 When immersed in pure water, EDA–GO membrane barely showed any swelling (9.15 ± 0.1 Å) as compared to its GO counterpart which had a huge increase to 11.6 ± 0.4 Å. The suppression of swelling was further investigated by immersing the membrane in various chloride salt solutions, each for at least an hour. A comparison of XRD results between GO and EDA–GO membranes (Fig. 2(e) and (f) respectively) clearly showed a leftwards shift in (001) peak for the GO membrane which was not observed in the EDA–GO membrane. Further tests were conducted using solutions comprising of equimolar monovalent and divalent salts (Fig. S1†) and GO membranes were observed to exhibit swelling effects between the range of monovalent and divalent salt swelling. This was expected given that XRD averages all reflections from an incident area. EDA–GO membrane once again, did not exhibit any noticeable swelling. Fig. S1(c)† is a summary of all d-spacings and an increase in d-spacing of up to 54% was observed for the GO membrane in contrast to only 4.5% increase for EDA–GO membrane.

Fig. 2 (a) Representative photograph of an EDA–GO membrane separated from its PTFE membrane filter. (b1) Schematic showing a GO/EDA–GO membrane with the underlying carbon paper scaffold. (b2) Schematic illustrating transmembrane cation flow modulated by d-spacing and moving in the same direction as the electric field. SEM images of (c) GO and (d) EDA–GO membranes. XRD results of (e) GO and (f) EDA–GO membranes in the dry state, immersed in ultrapure water and various 1 mmol salt solutions as indicated.

Elemental composition was investigated using XPS and low-resolution survey spectra (Fig. S2†) showed an obvious N1s peak for the EDA–GO membrane originating from the aliphatic diamine. High-resolution C1s spectra (Fig. S3†) was further analysed and deconvoluted into component peaks. C1s spectra of GO membrane was comprised of 4 distinct peaks centered at 284.5 eV (C=C/C–C), 286.7 eV (C–O–C), 287.0 eV (C=O) and 288.1 eV (O–C=O). The C1s spectra of EDA–GO membrane, on the other hand, showed five peaks at 284.5 eV (C=C/C–C), 285.1 eV (C–N), 286.2 eV (C–O–C), 286.5 eV (C=O) and 287.8 eV (O–C=O). Not only did the EDA–GO membrane exhibit a new C–N peak at 285.1 eV, it also showed decreased peak intensities and lower binding energies for oxygen moieties such as C–O–C. As such, these results indicated successful nucleophilic attacks by the diamine molecule on oxygen moieties. According to the literature, nucleophilic addition reactions most likely occurred between diamine and epoxy groups in a ring opening sequence.^22,35,36,38 The quantification of each C1s component peak is provided in Table. S3.† Overall, the atomic percentage of oxygen decreased from 26% to 21% while the nitrogen content increased from 0% to 7% which indicated a simultaneous amine functionalization and GO reduction. Amine functionalization had also altered the surface potential of GO membranes and the EDA–GO membrane showed a zeta potential of −18.6 mV at pH 7 which was slightly higher than that of the GO membrane (−20.5 mV). Although free EDA molecules are strongly positive due to protonation of the double amine groups, EDA molecules in our membranes were localised and bound between the GO layers due to formation of C–N bonds. Hence, there was not much change in the surface potential.

2.2 Dynamics of ion transport and selectivity

In the first study, we investigated the effects of passive cation movement induced by applying a positive potential at the anode. Under the influence of an electric field, anions were attracted and adsorbed on the anode whereas cations on the cathode. Four equimolar binary salt solutions were used as feed solutions under an anode voltage between 0.8 to 1.6 V using a step size of 0.2 V. Specifically, feed solutions are designated F1–F4 as shown in Table. S1† and ionic properties of the investigated cations are given in Table. S2.† The percentage decrease in concentration of each type of cation is provided in Figs. S5 and S6† and a separation factor, SF_M/D was used to quantify monovalent ion selectivity. SF_M/D of 1 indicated no selectivity while >1 indicated a monovalent ion adsorption selectivity. The results are summarized in Fig. 3 and we note that conventional CDI without any membranes did not exhibit any apparent monovalent ion selectivity except for two data points at the lower voltage range for F4. Close inspection of those points revealed a SF_M/D of 1.30 ± 0.16 at 0.8 V and 1.07 ± 0.13 at 1.0 V which decreased monotonically as the voltage was increased. Without considering other factors, we expected divalent cations such as Mg²⁺ and Ca²⁺ to experience larger electrostatic forces due to their higher valence. This would eventually lead to an equilibria where divalent cation adsorption was favoured over monovalent cation. However, we observed an apparent monovalent cation selectivity at lower voltages (<1.2 V) in F4. This was likely due to smaller electric field strengths which decreased the effects of valence. This reason, coupled with the fact that monovalent cations possess faster diffusion kinetics, caused the apparent monovalent cation selectivity. Similar trends were reported by Hassanvand et al.³⁹ who noted increased adsorptions of K⁺ and Na⁺ over Ca²⁺ at an even higher voltage of 1.5 V. In contrast, GO/CDI and EDA–GO/CDI systems showed more obvious signs of selectivity, especially when voltage was increased above 1.0 V in F2 and F4. This was likely due to the larger size of Mg²⁺ (R_hydrated: 4.28 Å) used in F2 and F4 as opposed to Ca²⁺ (R_hydrated: 4.12 Å) in F1 and F3. It stands to reason that for a nanochannel-modulated mass transport process, it would be easier to exclude Mg²⁺ than Ca²⁺. It could also be counterargued that the size of Ca²⁺ was only slightly smaller than Mg²⁺ and if effects such as partial dehydration or hydration shell reorientation were to be taken into account, we should see similar monovalent ion selectivity in F1 and F3. However, Mg²⁺ possesses a solvation energy (455.5 kcal mol⁻¹)⁴⁰ higher than Ca²⁺ (380.8 kcal mol⁻¹)⁴⁰ which meant Mg²⁺ was more likely to retain its hydrated configuration than undergo some form of distortion. Fig. 3(a) and (c) further show that for most voltages, SF_M/D was lesser than 1 or close to 1. In general, higher SF_M/D values obtained at higher anode voltages suggested an enhancement of nanochannel-modulated transport due to stronger electric field. This hypothesis was further supported by a weak linear trend observed across in feeds F2–F4. Moreover, the GO/CDI showed monovalent ion selectivity despite possessing an average nanochannel size (>8.5 Å after deducting thickness of single GO layer⁴¹) larger than the investigated cations. This could be explained from the perspective of Chen et al. who documented the effects of cation–π interactions in controlling the interlayer spacing of GO membranes.^7,8,42 He showed that monovalent cations such as K⁺ or Na⁺ possessed hydration energies comparable to interaction energies between the cation and sp² regions of GO sheets. As such, it was possible for a monovalent cation to be specifically adsorbed onto aromatic regions of GO sheets. This results in steric and charge distortion surrounding the vicinity of the adsorbed cation which effectively prevented other cations from traversing across the nanochannel. Additionally, the adsorption of K⁺ or Na⁺ can cause adjacent GO sheets to skew towards the cation and reduce local interlayer spacing. When considered within the context of our experiments, effects of cation–π interactions were much weaker due to lower concentrations of K⁺ or Na⁺ used (1 mM vs. 250–1500 mM). On the other hand, EDA–GO membranes showed higher and more consistent monovalent ion selectivity across all voltages and feeds due to strict restriction of d-spacing by EDA functionalization. An apparent lack of selectivity was observed in F3 which was attributed to a comparatively smaller difference in hydrated size of Na⁺ and Ca²⁺. As such, two factors which affected the selectivity of our membranes were disparity in size of monovalent and divalent cations and voltage applied.

Fig. 3 Separation factors plotted against an increasing anode voltage for feed solution (a) F1, (b) F2, (c) F3 and (d) F4. Dotted line represents a separation factor of 1 which meant no selectivity between monovalent and divalent ions. Error bars denote standard deviations.

Dynamics of cation adsorption were also investigated as voltage was increased and adsorption curves of feed F2 were taken as an example to explain our findings. Fig. S4† shows the representative adsorption curves obtained at 0.8 V and 1.2 V. All systems show a steady decrease in cation concentrations when voltage was applied as expected of typical CDI systems and full-cycle conductivity curves register complete recovery of effluent conductivity upon desorption. Strict exclusion of Mg²⁺ was not observed due to assistance from the applied electric field. Without membranes, CDI showed a preferential adsorption of Mg²⁺ over K⁺ regardless of the voltage (Figs. S4(a) and (d)†). As voltage increased, the adsorption of both K⁺ and Mg²⁺ also increased with Mg²⁺ adsorption exceeding that of K⁺. In the case of GO/CDI, there was initially no discernible selectivity at 0.8 V and monovalent ion selectivity only manifested when voltage was ≥1.0 V. At 0.8 V, similar amounts of K⁺ (5.13%) and Mg²⁺ (5.01%) were removed which resulted in a SF_M/D of only 1.03. This was enhanced at 1.2 V where 12.4% of K⁺ was removed as compared to only 6.9% of Mg²⁺. Even though GO membranes showed some degree of monovalent ion selectivity, the transport of larger cations was only suppressed at higher voltages when size exclusion effects begin to become more apparent. Comparatively, EDA–GO/CDI operated more reliably and adsorption of K⁺ was favoured even at 0.8 V. About 11% of K⁺ and 7.3% of Mg²⁺ were removed at 0.8 V which was followed by 14.1% of K⁺ and 6.8% of Mg²⁺ at 1.2 V (Fig. S5(b)†). The change in effluent cation concentration was given as a percentage and a summary of the results was recorded in Fig. S5.† Generally speaking, total cation adsorption was comparable across all systems which indicated negligible adsorption by the GO laminates.

In our second study, a negative voltage was sourced to the cathode (membrane side) to induce active cation transport and field strengths were kept the same as before. In a parallel fashion, Fig. 4 shows the SF_M/D values of all feed solutions under different cathode voltages. A brief survey of the results showed that GO/CDI and EDA–GO/CDI systems exhibited monovalent ion selectivities approximately double that of the previous study in F1, F2 and F4 whereas F3 showed minor improvements. Among the membrane systems, EDA–GO/CDI showed the greatest improvement in F1, achieving SF_M/D up to 3.32 as opposed to only 1.47 in the first study. Conventional CDI showed no remarkable changes to performance and slight deviations could be attributed to the inclusion of the carbon paper at the cathode. A stronger linear trend was observed for the membrane systems which highlighted a difference in passive and active ion transport mechanisms. The first study employed a positive anode voltage which actively induced the transport of anions towards the anode. Cations, in that case, were passively transferred to the cathode mainly due to charge neutrality conditions. Overall ion transport was anion controlled and migration of cations was weaker. Some cations may even be adsorbed onto the GO membrane surface and not permeate through at all. As such, the effects of having a cation-based membrane diminished. In contrast, when a negative cathode voltage was applied, cations were actively attracted towards the cathode and ion transport became cation dominated instead. To further validate our claims, comparisons were made between the amounts of monovalent and divalent cations adsorbed in Fig. S5 and S6.† Under positive anode voltages, concurrent increases in both monovalent and divalent cation removal were observed which were direct responses to the adsorption of Cl⁻ ions. This was not the case when negative cathode voltages were used. The presence of a negative cathode potential enhances the transport of cations towards the already negatively charged GO membrane and forces cations to permeate through. Adsorption of monovalent cations increased linearly with increasing negative voltages yet divalent cation adsorption was muted. Thus, it could be argued that selectivity of monovalent cations was higher in the second study due greater suppression of divalent cation transport.

Fig. 4 Separation factors plotted against an increasingly negative cathode voltage for feed solution (a) F1, (b) F2, (c) F3 and (d) F4. Dotted line represents a separation factor of 1 which meant no selectivity between monovalent and divalent ions. Error bars denote standard deviations.

Another consequence of a cation dominated adsorption process was that lower amounts of monovalent and divalent cations were adsorbed due to greater influence from the membrane. Even though monovalent ion selectivity had improved, the amount of monovalent cation adsorbed actually decreased. The dynamics of cation adsorption were revisited for feed F2 and the results at 0.8 and 1.2 V are presented in Fig. S7.† Conductivity curves were noticeably smaller than the first study but adsorption and desorption were still performed reversibly. Adsorption curves of all systems almost parallel the results in Fig. S4† except for GO/CDI which showed a small but distinct K⁺ selectivity.

2.3 DFT calculations of cation interactions in EDA–GO membrane

The enhancement of selectivity in EDA–GO membrane was considered from the perspective of cation interactions within the vicinity of the EDA molecule using DFT. A model of EDA–GO was first constructed with an EDA molecule occupying a perpendicular position across two GO sheets possessing epoxide groups and allowed to relax to an energetically stable state as shown in Fig. 5(a). Only the epoxide group was considered since it was theoretically more stable⁴³ and was the dominant basal oxygen group based on XPS results. After EDA was successfully bound to GO, the interlayer distance was calculated to be 6.02 Å which corresponded to a d-spacing of about 9.32 Å (protrusions from C–O were ignored). This was consistent with the experimental result (9.28 Å) obtained from XRD. Next, we considered cation interactions with EDA. Our calculations revealed that it was possible for K⁺ to react with an epoxide group near EDA (Fig. 5(b)) and for Na⁺ to form a complex with nitrogen from EDA and oxygen from an adjacent epoxide group (Fig. 5(c)). Simulations for Ca²⁺ and Mg²⁺ also showed stable complex formations but their individual hydration energies were far too high for such a configuration to be feasible. In an aqueous environment, divalent cations prefer coordinate with water molecules and screening from their hydration shells prevent them from interacting with EDA or epoxide groups. Fig. 5(d) depicts a comparison between binding and hydration energies for all cations. Only monovalent cations possessed binding energies higher than that of their respective hydration energies which meant that it was easy for K⁺ and Na⁺ to reorganize their hydration shells and accommodate a new, more optimal position. Fang's group^7,42 had performed similar calculations and showed that among the two, the hydrated K⁺ was more easily distorted and could be adsorbed onto aromatic regions of GO. In our case, the unique ensemble of EDA and GO resulted in binding energies more than 5 times than what was calculated by Fang et al.

Fig. 5 DFT calculations of the (a) EDA–GO structure and the adsorption of (b) K⁺, (c) Na⁺. Yellow and cyan areas correspond to charge gain and loss respectively. (d) Comparison between binding energies calculated in this work and hydration energies obtained from references.

Our simulations showed that K⁺ preferentially reacts with the epoxide group along with a hydrogen from EDA to form a stable protonated complex. Although no bond was directly formed with EDA, the distortion caused by the K⁺ complex resulted in a decreased interlayer distance of 5.82 Å. Conversely, a different type of complex was formed between Na⁺, EDA and an epoxide group which extended the interlayer spacing to 6.04 Å. This increase in spacing was attributed to rigid bonds formed between Na⁺ and EDA and between Na⁺ and the epoxide oxygen. As such, the exclusion of divalent cation was supplemented by the formation of the K⁺ complex and suppressed by the Na⁺ complex. This could explain why feed solutions employing K⁺ performed slightly better than those containing Na⁺.

3. Discussion

Two types of membranes were investigated in our CDI system, an unmodified GO membrane and a d-spacing restricted EDA–GO membrane. Equimolar binary solutions of monovalent and divalent salts were used in our experiments and monovalent ion selectivity was recorded for both GO/CDI and EDA–GO/CDI which confirmed the effects of nanochannel-modulated ion transport. The nature of monovalent ion selectivity was found to be largely dependent on size exclusion effects and was enhanced when a negative voltage was applied at the cathode to actively induce cation transport. A further increase in negative voltage prompted higher removal of monovalent cations while divalent cation removal remained more or less the same. This phenomenon was in direct contrast to using a positive voltage at the anode side where cation transport was observed to compensate for Cl⁻ adsorption. DFT calculations further revealed the unique role of the EDA spacer molecule in GO. K⁺ could react with an epoxide group near EDA to form a complex which induces distortion of the GO bilayer. Consequently, the interlayer distance within the vicinity would decrease. Na⁺ could also form a complex with EDA and a nearby epoxide group. However, the bonds of this complex would cause the interlayer distance to increase beyond the original EDA–GO structure and reduce divalent cation exclusion. Thus, selectivity was higher for binary solutions containing K⁺ than Na⁺.

Although our findings have shown that it was possible for selectivity to be tuned using an applied electric field, selectivity remained largely dependent on the properties of the membrane. The flexibility to dynamically alter selectivity could prove useful for precise ion sieving applications. Additionally, the desorption phase of CDI releases these ions back into a secondary stream which meant that a user could selectively remove and concentrate ions in a composition of the user's choosing. Thus, specialised compositions of fluids could be produced efficient and rapidly which could be useful in pharmaceuticals and healthcare.

4. Methods

4.1. Materials

Chemicals and reagents used in this study were of analytical grade and used as received unless otherwise stated. Pristine GO solution was purchased from Graphene Supermarket and diluted for further use. Hydrophilic carbon paper was purchased from Shanghai Hesen Electronics Co. Ltd and ultrapure water (18.2 MΩ) was provided by an ultrapure water system (arium pro UV, Sartorius).

4.2. Synthesis and fabrication of GO and EDA–GO membranes

EDA–GO was prepared according to a previous method.²² Prior to synthesis, commercial GO solution was diluted to 0.5 mg ml⁻¹ and dispersed using ultrasonication. Next, EDA was added dropwise to a stirring GO solution at a mass ratio of 15 : 1 and the resulting mixture was allowed to stir for an additional 1 h before ultrasonication. Carbon paper circles of 41 mm diameter were cut and pressed against a 47 mm hydrophilic polytetrafluoroethylene (PTFE) membrane filter (Omnipore, 0.2 μm pore size) which was placed on top of a fritted glass before vacuum filtration. 20 ml of GO or EDA–GO solution was allowed to filter for at least 6 h before the membranes were collected and placed in an oven at 80 °C for 1 h. Heating removes intercalated water molecules and induces cross-linking between EDA and GO. The membranes were removed carefully from the PTFE substrate and the EDA–GO membrane was immersed in anhydrous methanol for 12 h to remove excess EDA before it was dried and stored along with the GO membrane.

4.3. Fabrication of electrodes

Activated carbon (AC) electrodes were fabricated using a typical slurry-based method. Highly porous AC powder (Kuraray Co. Ltd) was dry mixed with polyvinylidene fluoride (PVDF, Mw ∼180 000) and carbon black (Super P) in a mass ratio of 8 : 1 : 1 and ground using an agate mortar and pestle. Approximately 3 ml of N-methyl-2-pyrrolidone (NMP, 99.5%) was added to form a slurry which was then spread across graphite sheets (Latech Scientific Supply Pte. Ltd, Singapore) serving as current collectors. The electrodes were left to dry in an oven at 90 °C for 12 h before they were weighed. The slurry occupied a circular disk 35 mm in diameter and had a mass of around 75 mg. The average coating thickness was around 25 μm.

4.4. Characterization of GO and EDA–GO membranes

A field emission scanning electron microscope (FE-SEM, JEOL JSM-7600F) was used to study the morphology of the as-synthesized membranes and estimate the thickness of the membrane. Powder X-ray diffraction (XRD) experiments were performed using a Bruker D8 Advanced X-ray diffractometer equipped with a Ni filtered Cu Kα radiation (λ = 1.5406 Å, 40 kV, and 40 mA) source to determine the d-spacing. X-ray photoelectron spectroscopy (XPS) studies were performed using a Thermo Scientific ESCALAB 250Xi system equipped with an Al Kα beam source. Post-analysis of XPS spectra was carried out using CasaXPS software (version 2.3.18) after a Shirley background subtraction and peak fitting using Lorentzian functions. Zeta potential measurements were obtained using a Zetasizer Nano ZS90 (Thermo Fisher Scientific) at pH 7. An inductively coupled plasma optical emission (ICP-OES) spectrometer (ICPE-9820, Shimadzu) operating in radial view mode was used to determine the composition and amount of cations removed during electrosorption.

4.5. Electrosorption experiments

Electrosorption was performed using a standard CDI cell as described in our previous works.^44,45 The CDI cell consisted of a pair of symmetrical AC electrodes, a GO or EDA–GO membrane, an acrylic separator layer with an attached nylon mesh and two acrylic housing panels secured by stainless steel screws. The positive electrode was denoted anode whereas the negative electrode was cathode. To account for ion adsorption due to carbon paper substrate, a plain piece of carbon paper was cut and placed in front of the cathode for the membraneless CDI. A schematic depicting the assembly of the CDI system is shown in Fig. 1(b). A batch-mode operation was adopted and salt solutions of varying compositions were cycled through the CDI cell. Each cycle lasted 20 min with equal half-cycle times for adsorption and desorption. Other parameters such as temperature (298 K) and flow rate (25 ml min⁻¹) were kept constant.

Two types of constant voltage modes were used in our experiments. In the first mode, a positive voltage was applied to the anode during adsorption whereas a negative voltage was applied to the cathode during adsorption. Depending on the polarity of the ion, the influence of the electric field could be considered active or passive. For instance, the electric field in the first mode actively induced the movement of anions towards the anode whereas cations were passively transferred to the cathode. Ion desorption for both modes was achieved by applying a short-circuit to the electrodes. All voltages were supplied using a programmable sourcemeter (SMU 2450, Keithley) and controlled using a script. Effluent solution conductivity was measured using a conductivity meter (DDSJ-308F, Leici) and cation removal was determined by taking aliquots of the effluent solution for ICP measurements. 1 mM KCl, NaCl, CaCl₂ and MgCl₂ solutions were used in our experiments in single or binary compositions and complete dissociation of salt in stoichiometric amounts was assumed. To avoid any spurious readings, all experimental results were recorded after the 3^rd cycle. The percentage decrease in molar cation concentration was calculated as:

where C_i and C_j are the effluent cation concentrations (mM) at the beginning and the end of the adsorption cycle respectively. Monovalent to divalent cation selectivity was quantified by a separation factor,^8,46 SF_M/D given as:

4.6. DFT calculations of cation interaction with EDA

A model of a single GO sheet was first constructed by binding 4 oxygen atoms onto a 4 × 4 supercell of graphene and two such monolayers were used to construct a GO bilayer structure. All computational simulations were performed using the Vienna Ab Initio Simulation Package (VASP 5.4) based on DFT.⁴⁷ The projector augmented wave (PAW)⁴⁸ approach in conjunction with a generalized gradient approximation (GGA)⁴⁹ in the form of Perdew, Burke, and Ernzerhof (PBE)⁵⁰ was adopted to describe the exchange-correlation potential. DFT-D3 (ref. 51) correction proposed by Grimme was employed to describe van der Walls (VdW) interactions. Besides that, a 12 Å vacuum distance was adopted to avoid interactions between atoms in neighbouring cells. The energy cut-off was set to 500 eV. The convergence criteria for energy and force were 10⁻⁴ eV and 0.05 eV Å⁻¹ respectively. A k-grid of 3 × 3 × 1 was utilized for geometry relaxation and total energy calculation.

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

This research is supported by CGH-SUTD Health Tech Innovation Fund (HTIF) Joint Grant (grant number: CGH-SUTD-HTIF-2018-003). This work is also supported by the 111 Project (D20015).

References

1. H.-H. Huang, R. K. Joshi, K. K. H. De Silva, R. Badam and M. Yoshimura, Fabrication of reduced graphene oxide membranes for water desalination, Journal of membrane science, 2019, 572, 12–19.
2. R. Nair, H. Wu, P. Jayaram, I. Grigorieva and A. Geim, Unimpeded permeation of water through helium-leak–tight graphene-based membranes, Science, 2012, 335, 442–444.
3. N. Padmavathy, S. S. Behera, S. Pathan, L. Das Ghosh and S. Bose, Interlocked graphene oxide provides narrow channels for effective water desalination through forward osmosis, ACS Appl. Mater. Interfaces, 2019, 11, 7566–7575.
4. M. Hu and B. Mi, Enabling graphene oxide nanosheets as water separation membranes, Environ. Sci. Technol., 2013, 47, 3715–3723.
5. V. Kandjou, A. M. Perez-Mas, B. Acevedo, M. Hernaez, A. G. Mayes and S. Melendi-Espina, Enhanced covalent p-phenylenediamine crosslinked graphene oxide membranes: Towards superior contaminant removal from wastewaters and improved membrane reusability, J. Hazard Mater., 2019, 380, 120840.
6. Y. Zhang, S. Zhang, J. Gao and T.-S. Chung, Layer-by-layer construction of graphene oxide (GO) framework composite membranes for highly efficient heavy metal removal, Journal of membrane science, 2016, 515, 230–237.
7. L. Chen, et al., Ion sieving in graphene oxide membranes via cationic control of interlayer spacing, Nature, 2017, 550, 380.
8. S. Liang, S. Wang, L. Chen and H. Fang, Controlling interlayer spacings of graphene oxide membranes with cationic for precise sieving of mono-/multi-valent ions, Sep. Purif. Technol., 2020, 241, 116738.
9. P. Yu, et al., Electrochemically-derived graphene oxide membranes with high stability and superior ionic sieving, Chem. Commun., 2019, 55, 4075–4078.
10. J. Abraham, et al., Tunable sieving of ions using graphene oxide membranes, Nat. Nanotechnol., 2017, 12, 546.
11. H. Dou, et al., Bioinspired Graphene Oxide Membranes with Dual Transport Mechanisms for Precise Molecular Separation, Adv. Funct. Mater., 2019, 29, 1905229.
12. R. Joshi, et al., Precise and ultrafast molecular sieving through graphene oxide membranes, science, 2014, 343, 752–754.
13. Q. Liu, K. M. Gupta, Q. Xu, G. Liu and W. Jin, Gas permeation through double-layer graphene oxide membranes: The role of interlayer distance and pore offset, Sep. Purif. Technol., 2019, 209, 419–425.
14. Y. Shin, et al., Graphene oxide membranes with high permeability and selectivity for dehumidification of air, Carbon, 2016, 106, 164–170.
15. H.-Y. Lee, Y.-C. Heish and C.-T. Lee, High sensitivity detection of nitrogen oxide gas at room temperature using zinc oxide-reduced graphene oxide sensing membrane, J. Alloys Compd., 2019, 773, 950–954.
16. B. Van der Bruggen, M. Mänttäri and M. Nyström, Drawbacks of applying nanofiltration and how to avoid them: a review, Separ. Purif. Technol., 2008, 63, 251–263.
17. M. D. Firouzjaei, et al., Recent advances in functionalized polymer membranes for biofouling control and mitigation in forward osmosis, J. Membr. Sci., 2020, 596, 117604.
18. A. V. Talyzin, T. Hausmaninger, S. You and T. Szabó, The structure of graphene oxide membranes in liquid water, ethanol and water–ethanol mixtures, Nanoscale, 2014, 6, 272–281.
19. N. Wei, X. Peng and Z. Xu, Understanding water permeation in graphene oxide membranes, ACS Appl. Mater. Interfaces, 2014, 6, 5877–5883.
20. B. Gupta, N. Kumar, K. Panda, V. Kanan, S. Joshi and I. Visoly-Fisher, Role of oxygen functional groups in reduced graphene oxide for lubrication, Sci. Rep., 2017, 7, 1–14.
21. J.-h Song, H.-W. Yu, M.-H. Ham and I. S. Kim, Tunable ion sieving of graphene membranes through the control of nitrogen-bonding configuration, Nano Lett., 2018, 18, 5506–5513.
22. W.-S. Hung, et al., Cross-linking with diamine monomers to prepare composite graphene oxide-framework membranes with varying d-spacing, Chem. Mater., 2014, 26, 2983–2990.
23. M. Zwolak, J. Wilson and M. Di Ventra, Dehydration and ionic conductance quantization in nanopores, J. Phys.: Condens. Matter, 2010, 22, 454126.
24. R. Tivony, S. Safran, P. Pincus, G. Silbert and J. Klein, Charging dynamics of an individual nanopore, Nat. Commun., 2018, 9, 4203.
25. J. Chmiola, G. Yushin, Y. Gogotsi, C. Portet, P. Simon and P.-L. Taberna, Anomalous increase in carbon capacitance at pore sizes less than 1 nanometer, Science, 2006, 313, 1760–1763.
26. L. A. Richards, A. I. Schäfer, B. S. Richards and B. Corry, The importance of dehydration in determining ion transport in narrow pores, Small, 2012, 8, 1701–1709.
27. G. Feng, R. Qiao, J. Huang, B. G. Sumpter and V. Meunier, Ion distribution in electrified micropores and its role in the anomalous enhancement of capacitance, ACS Nano, 2010, 4, 2382–2390.
28. W. Hirunpinyopas, E. Prestat, S. D. Worrall, S. J. Haigh, R. A. Dryfe and M. A. Bissett, Desalination and nanofiltration through functionalized laminar MoS2 membranes, ACS Nano, 2017, 11, 11082–11090.
29. C. E. Ren, et al., Voltage-Gated Ions Sieving through 2D MXene Ti3C2T x Membranes, ACS Appl. Nano Mater., 2018, 1, 3644–3652.
30. L. Ding, et al., Effective ion sieving with Ti 3 C 2 T x MXene membranes for production of drinking water from seawater, Nat. Sustain., 2020, 3, 296–302.
31. C. E. Ren, K. B. Hatzell, M. Alhabeb, Z. Ling, K. A. Mahmoud and Y. Gogotsi, Charge-and size-selective ion sieving through Ti3C2T x MXene membranes, J. Phys. Chem. Lett., 2015, 6, 4026–4031.
32. L. Cantley, et al., Voltage gated inter-cation selective ion channels from graphene nanopores, Nanoscale, 2019, 11, 9856–9861.
33. C. Cheng, G. Jiang, G. P. Simon, J. Z. Liu and D. Li, Low-voltage electrostatic modulation of ion diffusion through layered graphene-based nanoporous membranes, Nat. Nanotechnol., 2018, 13, 685–690.
34. P. Sun, F. Zheng, K. Wang, M. Zhong, D. Wu and H. Zhu, Electro- and Magneto-Modulated Ion Transport through Graphene Oxide Membranes, Sci. Rep., 2014, 4, 6798.
35. H.-L. Ma, Y. Zhang, Q.-H. Hu, D. Yan, Z.-Z. Yu and M. Zhai, Chemical reduction and removal of Cr (VI) from acidic aqueous solution by ethylenediamine-reduced graphene oxide, J. Mater. Chem., 2012, 22, 5914–5916.
36. N. H. Kim, T. Kuila and J. H. Lee, Simultaneous reduction, functionalization and stitching of graphene oxide with ethylenediamine for composites application, J. Mater. Chem. A, 2013, 1, 1349–1358.
37. Y. Liu, L. Pan, X. Xu, T. Lu, Z. Sun and D. H. C. Chua, Enhanced desalination efficiency in modified membrane capacitive deionization by introducing ion-exchange polymers in carbon nanotubes electrodes, Electrochim. Acta, 2014, 130, 619–624.
38. Y. Qian, X. Zhang, C. Liu, C. Zhou and A. Huang, Tuning interlayer spacing of graphene oxide membranes with enhanced desalination performance, Desalination, 2019, 460, 56–63.
39. A. Hassanvand, G. Q. Chen, P. A. Webley and S. E. Kentish, A comparison of multicomponent electrosorption in capacitive deionization and membrane capacitive deionization, Water Res., 2018, 131, 100–109.
40. C. S. Babu and C. Lim, Theory of ionic hydration: Insights from molecular dynamics simulations and experiment, J. Phys. Chem. B, 1999, 103, 7958–7968.
41. S. Zheng, Q. Tu, J. J. Urban, S. Li and B. Mi, Swelling of Graphene Oxide Membranes in Aqueous Solution: Characterization of Interlayer Spacing and Insight into Water Transport Mechanisms, ACS Nano, 2017, 11, 6440–6450.
42. Y. Yang, L. Mu, L. Chen, G. Shi and H. Fang, Precise control of the interlayer spacing between graphene sheets by hydrated cations, Phys. Chem. Chem. Phys., 2019, 21, 7623–7629.
43. H. S. Moon, J. H. Lee, S. Kwon, I. T. Kim and S. G. Lee, Mechanisms of Na adsorption on graphene and graphene oxide: density functional theory approach, Carbon Lett., 2015, 16, 116–120.
44. Z. Y. Leong, G. Lu and H. Y. Yang, Three-dimensional graphene oxide and polyvinyl alcohol composites as structured activated carbons for capacitive desalination, Desalination, 2019, 451, 172–181.
45. Z. Y. Leong and H. Y. Yang, Capacitive Deionization of Divalent Cations for Water Softening Using Functionalized Carbon Electrodes, ACS Omega, 2020, 5, 2097–2106.
46. K. Singh, Z. Qian, P. Biesheuvel, H. Zuilhof, S. Porada and L. C. de Smet, Nickel hexacyanoferrate electrodes for high mono/divalent ion-selectivity in capacitive deionization, Desalination, 2020, 481, 114346.
47. G. Kresse and J. Furthmüller, Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set, Phys. Rev. B: Condens. Matter Mater. Phys., 1996, 54, 11169.
48. P. E. Blöchl, Projector augmented-wave method, Phys. Rev. B: Condens. Matter Mater. Phys., 1994, 50, 17953.
49. D. Singh and J. Ashkenazi, Magnetism with generalized-gradient-approximation density functionals, Phys. Rev. B, 1992, 46, 11570.
50. M. Ernzerhof and G. E. Scuseria, Assessment of the Perdew–Burke–Ernzerhof exchange-correlation functional, J. Chem. Phys., 1999, 110, 5029–5036.
51. S. Grimme, J. Antony, S. Ehrlich and H. Krieg, A consistent and accurate ab initio parametrization of density functional dispersion correction (DFT-D) for the 94 elements H-Pu, J. Chem. Phys., 2010, 132, 154104.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: 10.1039/d0ta09109j

---