High-capacity and selective ammonium removal from water using sodium cobalt hexacyanoferrate

A new NH4+ adsorbent with high capacity and selectivity, sodium cobalt(ii) hexacyanoferrate(ii) (NaCoHCF, NayCo(ii) [Fe2+(CN)6]x·zH2O), was prepared. The adsorption performance was investigated by varying the mixing ratio of [Fe(CN)6]4− to Co2+ during synthesis, Rmix. The ammonia capacity was found to be proportional to Rmix, indicating that the NH4+ capacity can be increased by increasing the Na+-ion content in NaCoHCF. To conduct a detailed study, we prepared homogeneous nanoparticles by flow synthesis using a micromixer with Rmix = 1.00. Even on the addition of a saline solution (NaCl) with an Na+-ion concentration of 9350 mg L−1, the capacity was maintained: qmax = 4.28 mol kg−1. Using Markham–Benton analysis, the selectivity factor, defined by the ratio of equilibrium constants for NH4+ to that for Na+, was calculated to be α = 96.2, and 4.36 mol kg−1 was found to be the maximum capacity. The high selectivity of NaCoHCF results in good NH4+-adsorption performance, even from seawater. In comparison with other adsorbents under the same conditions and even for a NH4Cl solution, NaCoHCF showed the highest capacity. Moreover, the coexisting Na+ caused no interference with the adsorption of ammonium by NaCoHCF, whereas the other adsorbents adsorbed ammonia only slightly from the saline solution. We also found that the pores for NH4+ adsorption changed their sizes and shapes after adsorption.


Introduction
Ammonium, NH 4 + , is a widely synthesized chemical, most of which is used as a fertilizer. This distributed ammonium is oxidized to nitrite, NO 2 À , and then further converted to nitrate, NO 3 À , in rivers, lakes, and oceans through nitrication processes. When humans and animals drink water containing ammonia (NO 3 À $ 45 mg L À1 or NO 3 -N $ 10 mg L À1 ), 1 they can become ill or even die. In aquatic systems, especially those near densely populated settlements or large-scale livestock facilities, the high concentrations of ammonia in water cause concern. Excess ammonia in aquatic systems can cause eutrophication, harmful algal blooms, and anoxic conditions in estuaries, rivers, and even the coastal environment. Consequently, biodiversity, sheries, and overall ecosystem health are adversely affected by this N nutrient imbalance. [2][3][4] Furthermore, N 2 O, 1 a byproduct of ammonium oxidation, is an important greenhouse gas and ozone-depleting agent. The Intergovernmental Panel on Climate Change (IPCC) has estimated that the N 2 O emissions from the open ocean represent 3.8 Tg N per year, 35% of the total natural emissions. 5 In fact, various nations have set standards for ammonia in effluent or environmental water. For example, the U.S. Environmental Protection Agency (EPA) lowered the limit in 2013 for aquatic life in ambient water bodies to 17 and 1.9 mg L À1 total ammonia nitrogen one-hour and 30 day averages, respectively. 6 China and India have also established standards for effluents of 15-50 and 50 mg L À1 ammonium nitrogen. 7,8 Reducing the ammonium concentration is also important for biogasication technology with anaerobic digestion because the digestion is inhibited by high concentrations of ammonium. Depending on the conditions, 1500-5000 mg L À1 of total ammonium nitrogen can cause the slow down or failure of digestion. 9,10 For the removal of NH 4 + from wastewater or digestion liquids, methods for the selective removal of NH 4 + are necessary because various ions coexist in these solutions. For the uptake of NH 4 + , removal with an adsorbent is an easily controllable and highly efficient method. Strong acid cation (SAC) exchange resins, analogs of Amberlite IR-120 (Alfa Aesar, UK), show the highest adsorption capacity, reaching 5.34 mol kg À1 for aqueous solution without coexisting cations and using glass-packed bed columns. 11 Nevertheless, SAC resins have low selectivity to NH 4 + when other competing ions are present. He et al. studied alkaline-activated and lanthanumimpregnated zeolites 12 and found a maximum adsorption capacity, q max , of 1.54 mol kg À1 in a pure water solution. However, the removal efficiency, P R , decreased from 90% to 36% in the presence of Na + . Soetardji et al. reported that sodium-hydroxide-modied zeolite mordenite has q max ¼ 3.0 mol kg À1 in aqueous solution and that P R decreased from 81% to 66.9% when competing with other ions. 13 Guaya et al. studied a hydrated aluminum-oxide-modied zeolite, 14 which showed q max ¼ 2.14 mol kg À1 in aqueous solution and P R ¼ 12% with coexisting Na + ions. Thus, conventional adsorbents have lower selectivity. As mentioned above, for the removal of NH 4 + from wastewater or the digestion liquids, adsorbents with high-selectivity are crucial. In particular, for digestion liquids, a large capacity is also essential because the NH 4 + concentration is quite high. For these reasons, the development of new adsorbents with both large capacity and high selectivity is desirable. In our earlier study, we found that potassium copper hexacyanoferrate, KCuHCF, has a high ammonium adsorption capacity of 1.94 mol kg À1 , as well as high selectivity for dissolved ammonia. 15 6 ] vacancies. With respect to MHCFs, many researchers have studied their use in catalysis, [16][17][18] electrodes in secondary batteries, [19][20][21][22] electrochromism, [23][24][25][26][27] sensors, 28,29 gas storage, 30-32 photomagnets, [33][34][35] and adsorbents for radioactive Cs + ions. [36][37][38][39][40] The ionic radii of hydrated Cs and NH 4 are similar (3.29 and 3.31Å, respectively). Therefore, assuming a size-based adsorption model, MHCFs could also have substantial adsorption capability for NH 4 + . 41 The NH 4 + -adsorption mechanism is affected by the porous network in MHCFs. The crystal structure of MHCF is shown in Fig. 1, where two kinds of adsorption sites exist. One is an interstitial site, a cubic conned space surrounded by eight metal sites and twelve cyano-groups. Its pore size is less than 0.5 nm. The other is a vacancy site, represented as [Fe(CN) 6 ] vacancies. In the case of KCuHCF, both K + and NH 4 + are located in the interstitial sites, as shown by Rietveld analysis of the X-ray diffraction (XRD) patterns. 15 In addition, NH 4 + is adsorbed via the ion-exchange with the A + ions (K + in the case of KCuHCF). 15 Therefore, to enhance the adsorption capacity, [Fe(CN) 6 ] vacancies should be eliminated. This is because the number of A + ions increases as the number of vacancies decreases because of charge balance. The other reason is that the vacancy site does not play a role in NH 4 + adsorption.
Based on these considerations, in this paper, we investigated the use of another MHCF, sodium cobalt hexacyanoferrate (NaCoHCF), to enhance the adsorption capacity. The most important difference between KCuHCF and NaCoHCF is the difference between Cu and Co. In the case of KCuHCF, KCuHCF with fewer [Fe(CN) 6 ] 4À vacancies causes material instability in the aqueous solution. 42 However, with substitution with Co, the introduction of a small number of [Fe(CN) 6 ] 4À vacancies becomes possible. Additionally, the affinity of MHCF for the mono-cation is known to depend on the hydrated radius, implying that utilization of Na + instead of K + can increase the NH 4 + adsorption performance.
Our study has two parts. The rst is a compositiondependent study. Five kinds of NaCoHCF-nanoparticles (NaCoHCF-NPs), Na y Co[Fe(CN) 6 ] x $zH 2 O, were synthesized by changing the molar concentration ratio of the reagent solution (R mix ) using a batch method. The second part is a detailed study of R mix ¼ 1.00. Quantitative analysis into the adsorption capacity and selectivity was conducted. The changes to the crystal structure are also discussed. By comparison with earlier studies, we found that our NaCoHCF exhibits the a very high capacity when using the batch-adsorption method. Particularly for NH 4 + adsorption from saline solutions, the benets of NaCoHCF are enhanced by its high selectivity. We also demonstrate its potential for recyclability.

Synthesis of NaCoHCF-NPs
First, NaCoHCF-NPs with compositions of Na 4xÀ2 Co[Fe(CN) 6 ] x (water omitted) were prepared according to the following chemical reaction.  To study the composition dependence, the NaCoHCF-NPs were synthesized using a batch method by mixing two aqueous solutions of Na 4 [Fe(CN) 6 ]$10H 2 O (Wako Pure Chemical Ind., Ltd.) and CoCl 2 $6H 2 O (special grade from Wako Pure Chemical Ind., Ltd.) with different molar concentration ratios (R mix ¼ 0.50, 0.75, 1.00, 1.50, and 2.00). Here, R mix represents the mixing ratio of the concentration of [Fe(CN) 6 ] 4À to that of Co 2+ . The suspension was shaken using a multi shaker (SI-300C; AS One Corp.) for 3 min at 1700 rpm and room temperature. Aer shaking, the slurry solutions were centrifuged. The slurries were washed at least ve times with Milli-Q water. They were dried under vacuum at 60 C for 48 h.
For detailed studies conducted with a xed composition, we prepared NaCoHCF-NP samples using a ow synthesis method to guarantee the homogeneity of the particle size and chemical composition. 36 The NaCoHCF-NPs, denoted Flow-1.00, was synthesized by mixing 0.4 mol L À1 solutions of the Na 4 [Fe(CN) 6 ]$10H 2 O and Co(NO 3 ) 2 $6H 2 O (special grade from Wako Pure Chemical Ind., Ltd.) in a Y-type micro-mixer with a hole of F 250 mm, as shown schematically in Scheme 1. The mixed concentrations were the same as those for Batch-1.00. The ow rates of the two solutions were set to be equal. The total ow rate was 40 mL min À1 . The obtained slurries were washed using a hollow ber rinse system (DBW-24; OCT Science Co., Ltd.) to remove the NaNO 3 byproduct. Then, the NaCoHCF-NPs were dried in vacuum at 60 C for 72 h.

Characterization of NaCoHCF-NPs
The crystal structures of Flow-1.00 were studied before and aer NH 4 + adsorption using an X-ray diffractometer (D2 Phaser; Bruker Analytik GmbH, Germany) with Cu Ka (l ¼ 1.54Å) radiation in the 2q range of 5-60 at 30 kV and 10 mA. A Si (311) double-crystal monochromator was used to monochromatize the incident beam while reducing the high harmonics of the monochromatic beam. The XRD patterns were analyzed using the Pawley method to determine the space group and the lattice constants. For adsorption, a 500 mg L À1 NH 4 + aqueous solution was used. Other conditions are described in Section 2.3. The crystallite sizes were estimated using Scherrer analysis of the XRD patterns, assuming a Scherrer constant of 0.94. 43 Sample images were obtained using a eld-emission scanning electron microscope (FE-SEM, S-4800; Hitachi Hitec Corp.) with 5 kV accelerating voltage aer Pt-Pd coating using an ion sputter coater (E-1030; Hitachi Ltd., Japan). The chemical compositions and leaching concentration of CN À into treated water were determined using a Microwave Plasma-Atomic Emission Spectrometer (MP-AES, 4100; Agilent Technologies Inc., USA) with pre-decomposition using microwaves (MW, Multiwave 3000; PerkinElmer Inc., USA). The hydration numbers in each sample were ascertained through thermogravimetric analysis (Thermo Plus EVO2; Rigaku Corp.). The specic surface areas of several samples were estimated by tting the Brunauer, Emmett, and Teller (BET) equation to the N 2 adsorption isotherms obtained at 77 K. The typical pre-treatment condition was 100 C for 24 h.

NH 4 + adsorption tests
To evaluate the composition dependence, a batch-shaking method was used to evaluate the NH 4 + adsorption capacity of the NaCoHCF-NPs, as shown schematically in Fig. S1

Evaluation of recyclability
The potential for recyclability was also investigated. The experimental setup is shown in Fig. S1(b). † A membrane lter with NaCoHCF-NPs was prepared for the ow test. The Flow-1.00 powder was mixed with 2 mL Milli-Q (3.76 mg mL À1 ) using an ultrasonic cleaner (W-113MK-II; Honda). Then, 100 mL solutions were dropped on the membrane lter (F 25 mm, 0.45 mm pore size, JHWP01300; Merck), followed by drying at 60 C for 2 min. Thus prepared, the solutions were set on a circular plastic plate of F 25 mm with a hole of F 5 mm in the middle. They were pasted on the lm for effective adsorption-desorption. An FT-IR spectrometer (iD1 transmission iS5; Nicolet Biomedical Inc.) was used to conrm adsorption and desorption of NH 4 + . For the adsorption test, the NH 4 + solution of 500 mg L À1 was owed through the NaCoHCF-NP-dipped membrane for 30 min at the rate of 0.2 mL min À1 . For the desorption test, a NaCl solution of 5 mol L À1 was similarly owed for 2 h at a rate of 1 mL min À1 .

Composition dependence
The dependence of the chemical composition on the mixing ratio, R mix , is presented in Table 1. For R mix < 1.00, x is an almost equal to R mix . The value of y, the number of Na + ions in NaCoHCF, also increased. In contrast, when R mix > 1.00, the chemical composition was almost unchanged, demonstrating that the composition can be controlled by changing the reaction R mix to x < 1.
The crystal structure of NaCoHCF depends on the chemical composition (see Fig. S2 †). When R mix $ 1.00, the crystal structure is rhombohedral (R 3), as reported. 44,45 On the other hand, the structure for R mix ¼ 0.50 is unclear. Earlier reports described the space group as monoclinic (P2/m or P2 1 /m) 46 or cubic (Pm 3m). 47,48 In our case, the XRD pattern is explained as a mixture of R 3c and P2 1 /m structures, as shown in Table 1. The NaCoHCF with R mix ¼ 0.67 is also explainable as a mixture. Such a mixture could be the result of the batch synthesis because homogeneous synthesis is difficult using the batch method, resulting in the uctuation of the chemical composition.
The BET surface areas of Batch-0.50, À0.67, and 1.00 were evaluated to be 28, 76, and 46 m 2 g À1 , respectively. The N 2 isotherms are shown in Fig. S3 in the ESI. † These values are much smaller than those of other Prussian blue analogs synthesized for gas adsorption. For example, that of Co [Fe(CN) 6 ] 0.60 was 848 m 2 g À1 . The smaller surface area originated from the inclusion of Na + cations. Because the interstitial sites of NaCoHCF are fully or partially occupied by Na + , it would be impossible for N 2 to penetrate into the NaCoHCF lattice. This presumption is also supported by the fact that the averaged pore diameter is comparable to the particle size, as shown later in the analysis of Flow-1.00. Fig. 2(a) shows that the NH 4 + adsorption capacity improved with increasing R mix . The amount of adsorbed ammonia of Batch-2.00 is about twice that of Batch-0.5. Fig. 2(b) shows that the amount of Na + from the adsorbent has an almost linear correlation with the NH 4 + adsorption amount, indicating the NH 4 + adsorption occurred through ion exchange with Na + .
These results demonstrate that the increase in the Na + composition in NaCoHCF enhances the NH 4 + adsorption capacity, and that NaCoHCF retains its structure even aer long-term shaking in water. However, for R mix > 1.00, the adsorption capacity increased by only 2.9-3.9% from R mix ¼ 1.00 because the upper limit of x is 1.0.
3.2 Detailed study with ow-synthesized NaCoHCF with R mix ¼ 1.00 Based on results of the composition dependence of the NH 4 + capacity, we chose to conduct a detailed study of NaCoHCF-NPs with R mix ¼ 1.00 because we obtained the desired chemical  composition by using this value of R mix and because it showed sufficiently high capacity. For our detailed study, we used the ow-synthesized sample, Flow-1.00, to avoid the uctuation of the chemical composition and particle size. Table 1 shows that the chemical composition of Flow-1.00 is almost identical to that of Batch-1.00. The adsorption kinetics was studied at an initial NH 4 + concentration of 500 mg L À1 , 30 C, and at 600 rpm for 8 h (see Fig. S4 †). The results showed that the NH 4 + adsorption was almost completed in 30 min. Such fast adsorption is similar to the case of KCuHCF. 15 Using BET analysis, we estimated the surface area to be 53 m 2 g À1 , which is also comparable to that of Batch-1.00, 46 m 2 g À1 . As described before, this value is not very high because the interstitial sites of NaCoHCF are lled with Na + , preventing the penetration of N 2 into the porous network in the crystal. The average pore size was estimated to 31 nm for Flow-1.00, consistent with the size of the crystallites, as shown later.
Aer NH 4 + adsorption, the crystal structure was maintained, except for a slight trigonal distortion. Fig. 3 shows the XRD patterns obtained before and aer NH 4 + adsorption. Some splitting of the Bragg peaks is apparent. The slight structural transformation observed is the same as that in the case of ion exchange between Na + and K + . 45 Before adsorption, NaCoHCF had a rhombohedral (R 3c) structure. However, aer NH 4 + adsorption, it changed to a cubic lattice (Fm 3m; Z ¼ 4). Thus, the ion exchange reversibly changed the structure.
The structural change also inuenced the shape and size of the interstitial sites in which Na + or NH 4 + are located. The lattice parameters of the crystal are shown in Table 1. A schematic view of the relationship between the interstitial site and the lattice distortion is shown in Fig. 4. Note that the crystal symmetry is different before and aer the NH 4 + adsorption. Before adsorption, the lattice has the rhombohedral symmetry with slight distortion from the cubic lattice. That is, it was compressed along the (1,1,1) direction. In contrast, aer adsorption, the crystal maintained the cubic structure. This difference is consistent with previous reports of the ion exchange between Na + and K + but not NH 4 + . 45 Because the symmetry of the crystal is different before and aer NH 4 + adsorption, we investigated the distance between the nearest Co neighbors and the volume of the (pseudo)-cubic cage of the interstitial site to evaluate the changes to the interstitial sites. Before NH 4 + adsorption, the distance between the nearest Co neighbors was calculated as the length of the unit vector of the rhombohedral primitive cell, 7.24Å. On the other hand, aer NH 4 + adsorption, the separation was reduced to 7.18Å, resulting in reduction in the volume of the cage from 138 to 131 A 3 . It is interesting that the cage became smaller although the ionic radius of NH 4 + is larger than that of Na + . This could be due to the difference in the H 2 O accompanying the cations. Because the interaction with H 2 O is stronger with Na + than that with NH 4 + . Therefore, before NH 4 + adsorption, H 2 O is adsorbed along with Na + , even if cage expansion is energetically disadvantageous. This is consistent with the previous studies of the electrochemical injection of Na + into copper hexacyanoferrate thin lms. 49 No nanoparticle degradation occurred during adsorption. The crystallite sizes estimated by Scherrer analysis of the XRD patterns (Fig. 3) before and aer sorption were, respectively, 37.9 and 51.7 nm. This result is consistent with the SEM images in Fig. 5. The particle sizes estimated using SEM images were 33 AE 10 and 47 AE 13 nm. The data indicate no degradation, but there is a possibility of some particle growth. The reason for the growth remains unclear, but it could be due to the immobilization of the Co 2+ and [Fe(CN) 6 ] 4À ions eluted from the adsorbent onto the other part of the adsorbent. If so, the adsorbent would retain the eluted species. The surface morphology in SEM images shows no marked change aer adsorption. Furthermore, we also evaluated the release of CN À aer adsorption. The concentration of CN À in solution was only 0.33 mg L À1 , sufficiently smaller than effluent standard in Japan, 1 mg L À1 . 50 Fig. 6 shows the NH 4 + adsorption isotherms in aqueous NH 4 Cl solution and that in aqueous saline solution. In the saline solution, the concentration of the Na + solution was set to 9350 mg L À1 , the same as that of articial seawater. The curves t to the Langmuir, Freundlich, and Markham-Benton  equations are also shown in Fig. 6. The tting parameters for each equation are shown in Table 2. Concerning the Langmuir and Freundlich equations, individual tting parameters were obtained for the NH 4 Cl (aq) and saline solutions. The Langmuir equation is given by where C e , q e , q max , and K respectively represent the NH 4 + concentration in solution at equilibrium, loaded NH 4 + in the adsorbent, maximum adsorption capacity, and the equilibrium constant. The same data are also shown with other axes in Fig. S5, † and the adsorption behavior ts the Langmuir equation well. We also carried out tting to the Freundlich equation, For the Freundlich equation, only the region where the loaded NH 4 + concentration was less than 1.95 mol kg À1 , about a half the maximum capacity, was considered because the Freundlich equation is only suitable far from saturated loading. However, in this region, the Freundlich equation also well reproduced the experimental data.
In the tting of the Langmuir or Freundlich equations, we used different parameters for NH 4 Cl (aq) and the salt solutions because the effect of the coexistent Na + ions can only be modeled by changing the tting parameters. For a more quantitative evaluation of the effect of the coexistent Na + ions, we also considered the Markham-Benton model for a solution with multi-alkali cations. As a type of extended Langmuir equation, the Markham-Benton equation 51 was used to examine adsorption isotherms for multiple components to estimate the ease of desorption of the adsorbents. The results also provide some understanding of the selectivity of the sorbents for some ions. The Markham-Benton equation is where C e , q e , and K respectively represent the NH 4 + ion concentration in equilibrium, the adsorption capacity, and the equilibrium constant. C 0 e and K 0 respectively denote the Na + ion concentration in equilibrium and the equilibrium constant.
Considering the Na + -ion exchange for NH 4 + ions, i.e., even in NH 4 Cl aqueous solution, an equal amount of Na + ions would be exchanged out, adversely affecting the adsorption capacity. When we consider both sources of Na + ions (those in NaCoHCF and that in the solution), the equation can be expressed by as   where C 0 0 , m, and V respectively denote the initial Na + ion concentration in solutions, adsorbent mass, and solution volume.
The tting parameters, q max , K, and K 0 , are shown in Table 2. Fig. 6 shows that the experimental data were well tted using the Markham-Benton model. Again, with the Markham-Benton model, we use the same parameter set for NH 4 Cl (aq) and the saline solutions. A selectivity factor, a, dened by the ratio of equilibrium constants for NH 4 + to that for Na + was calculated to be a ¼ 96.2, indicating the high selectivity of NH 4 + against Na + .
Such high selectivity is expected to lead to extremely high capacity, even in an aqueous saline solution. To clarify the high capacity of NaCoHCF among the various adsorbents, we used two approaches. First, we surveyed and compared results with those of earlier studies, and we also conducted experimental investigations to assess the adsorption capacity of adsorbents in identical conditions. For the literature survey, we picked reports of adsorption tests carried out using a batch style because column-style tests generally report higher capacities, rendering a comparison of results difficult between batch-style tests and column tests.
Information from earlier studies is presented in Table 3  The benets of NaCoHCF were amplied in the case of aqueous saline solution with 9350 mg L À1 -Na and 500 mg L À1 -NH 4 . Fig. 7 shows that NaCoHCF has an adsorption capacity that is almost identical to that of the case without Na + , whereas the other adsorbents showed little adsorption.
Although our main aim, the preparation of an NH 4 -adsorbent with high capacity and high selectivity, has been achieved, the recyclability of the adsorbent is also important for practical use. Therefore, nally, we demonstrate the potential for recyclability by attempting desorption tests. Fig. 8 shows that the adsorption-desorption-adsorption process was conrmed by measuring the infrared absorption corresponding to NH 4vibration mode at around 1415 cm À1 . 15 Using the continuous ow of NaCl solution for desorption, the peak height was found to decrease to 28% the original value before ow, indicating the potential for NaCoHCF recyclability. Next, further study on the sorption performances, including the column sorption, quantitatively reusability test would be conducted.
Finally, we mention concerning the cost of the materials. As mentioned above, NaCoHCF is synthesized only mixing two solutions immediately, resulting in the suppression of the manufacturing cost. For NH 4 removal from salt water, NaCoHCF would be cost effective in comparison with Amberlites and zeolites, despite the utilization of the cobalt, one of the rare metals. This is because the amount of the adsorbent can be drastically decreased with its high selectivity.

Conclusions
Na y Co[Fe(CN) 6 ] x $zH 2 O (NaCoHCF) was synthesized using a batch method and various chemical compositions. Synthesis was also carried out using a ow method and a xed composition. The adsorption capacity increased as the number of [Fe(CN) 6 ] vacancies decreased and the material was stable in water. Such stability is unlike that of copper hexacyanoferrate. We compared the NH 4 + adsorption performance with other high-capacity adsorbents under the same conditions: zeolites, Amberlite ion-exchange resin, and sepiolite. The results show that NaCoHCF exhibited the highest capacity in NH 4 Cl aqueous solution. Using aqueous saline solutions with a Na + ion concentration of 9350 mg L À1 , the benets of using NaCoHCF are enhanced drastically. The Markham-Benton model revealed the high selectivity for NH 4 + against coexisting Na + . In addition, ammonium desorption from the adsorbent (enabling recycling) was demonstrated using an NaCl aqueous solution.

Author contributions
The manuscript was written through contributions of all authors. All authors have given approval to the nal version of the manuscript.