Amine-Functionalized Covalent Organic Framework for Efficient SO2 Capture with High Reversibility

Removing sulfur dioxide (SO2) from exhaust flue gases of fossil fuel power plants is an important issue given the toxicity of SO2 and subsequent environmental problems. To address this issue, we successfully developed a new series of imide-linked covalent organic frameworks (COFs) that have high mesoporosity with large surface areas to support gas flowing through channels; furthermore, we incorporated 4-[(dimethylamino)methyl]aniline (DMMA) as the modulator to the imide-linked COF. We observed that the functionalized COFs serving as SO2 adsorbents exhibit outstanding molar SO2 sorption capacity, i.e., PI-COF-m10 record 6.30 mmol SO2 g−1 (40 wt%). To our knowledge, it is firstly reported COF as SO2 sorbent to date. We also observed that the adsorbed SO2 is completely desorbed in a short time period with remarkable reversibility. These results suggest that channel-wall functional engineering could be a facile and powerful strategy for developing mesoporous COFs for high-performance reproducible gas storage and separation.

functionalization of COFs via the azide-alkyne click reaction 35,36 or the high-throughput ring-opening reaction 37 . Recently, Bein et al. have suggested a new functionalization method by introducing modulator agents into the one-pot synthesis 38 .
The storage capabilities of COFs for gases, such as hydrogen [39][40][41] , methane 39,42,43 , ammonia 44 , and carbon dioxide 36,37,39 have also been widely investigated; however, to our knowledge, SO 2 sorption COFs have yet to be reported and may be one of the most promising areas of SO 2 sorption research with the potential to dramatically improve performance.
Given the above, we designed a new SO 2 sorbent using imide-linked COFs (PI-COFs) that have high levels of physical and chemical stability 34 , even though most COFs and MOFs have very weak hydrolytic and thermal stability 45 . Also, quite a lot of SO 2 sorption MOFs have been reported so far, however, the materials have limitation in long-term stability and reversibility [46][47][48] . We successfully synthesized a new series of imide-linked COFs that incorporated 4-[(dimethylamino)methyl]aniline (DMMA) as the modulator with various ratios (i.e., X = 10, 20, 40, and 60) in Fig. 1, according to a scheme in Fig. 2. The dimethyl functional group of DMMA (i.e., pK b = 4.3) has strong basicity for forming a charge transfer complex with SO 2 (i.e., pK a = 1.76). Thus, chemical adsorption of SO 2 could be achieved through the surface-functionalized channel of imide-linked COF.
Furthermore, adsorbed SO 2 on the imide-linked COF are completely desorbed under desorption conditions, and thus the imide-linked COF can be reused. We were able to obtain functionalized imide-linked COFs via microwave heating in two hours, which was 60 times faster than the five-days reaction time required of the synthesized approach using the conventional solva-thermal method 49,50 .

Results and Discussion
Synthesis of PI-COFs via microwave-assisted reaction. Imide-linked COFs were synthesized as a result of the co-condensation reactions of tris(4-aminophenyl)amine (TAPA) with 1.5 equiv of pyromellitic dianhydride (PMDA) 34 . These two building blocks were then suspended in a 1:1:0.1 mixed solution of N-methyl-2pyrrolidone (NMP), mesitylene, and isoquinoline under microwave-assisted conditions at 200 °C for 2 h. The advantage of this microwave-assisted reaction is that direct microwave heating is able to reduce chemical reaction times and is also known to reduce side reactions, increase yields, and improve reproducibility of synthesis condition [49][50][51] .
The condensation reaction of PMDA and TAPA yielded a crystalline brown solid which is insoluble in water and typical organic solvent, such as acetone, hexane, chloroform, tetrahydrofuran (THF), N, N-dimethylformamide, or m-cresol. Among mixed solvents, NMP and mesitylene could control the solubility of building blocks and isoquinoline, being the catalyst, could accelerate reaction time by enhancing the rearrangement of iso-imide to imide 52 . The reversibility of the imidization reactions involves an error-correction mechanism that enables the conversion of kinetic intermediates (amorphous) to thermodynamically stable forms (crystalline) 53 .
FT-IR spectra confirmed C=O groups of the imide rings that corresponded to an asymmetric stretching peak at 1,770 cm −1 and a symmetric stretching peak at 1,720 cm −1 (see Supplementary Fig. S1). In addition, we observed a stretching vibration of C-N-C groups in the imide at peak 1,375 cm −1 and aromatic C-N stretching vibration of the TAPA core at peak 1320 cm −1 . Furthermore, no bands appeared that corresponded to the starting monomers (i.e., amino around 3,340 cm −1 and anhydride at 1,765 cm −1 ) or amic acid intermediate (i.e., amide around 1,650 cm −1 ), demonstrating that the products are fully imidized via the microwave-assisted reaction (PI-COF-m), just as in the solva-thermal method (PI-COF-s). These PI-COFs exhibit high thermal stability regardless of synthetic methods, as determined by thermogravimetric analysis (TGA) (see Supplementary  Fig. S2).
Properties of PI-COFs via microwave-assisted reaction. In PXRD, the peaks at 3.1°, 5.3°, and 6.  It is well matched to results presented by Yan et al. 34 , as shown in Fig. 3a. The experimental patterns also agreed with simulated PXRD patterns which exhibited eclipsed stacking structure slipped by 1/4 of the unit cell (see Supplementary Fig. S3). This structure has advantage for gas absorption effectively 34 .
We also demonstrated the influence of concentration of building blocks as a driving force for the error correction on the crystalline structure. The best optimized condition was 0.25 M according to the highest crystallinity of PI-COF-m in PXRD (see Supplementary Fig. S4). The appropriate concentration was crucial here, because suitable solubilized building blocks could be polymerized in the crystalline PI-COF-m. Depending on the PXRD data, we concluded that the fully solubilized building blocks would be polymerized in amorphous PI-COF-m because of the insufficient error correction.
Furthermore, we investigated the surface area and pore width of PI-COF-s and PI-COF-m via the Brunauer-Emmett-Teller (BET) method by measuring nitrogen gas (N 2 ) sorption at 77 K, revealing reversible isotherms. The discrete step in the sorption isotherms at P/P 0 = 0.05 could indicate the very well-defined porosity of the frameworks. In addition, the absence of hysteresis during desorption is a common feature of materials containing hexagonally aligned one-dimensional mesopores, which agrees with the crystallinity of frameworks attributed to the PXRD results 54 .
The BET surface area has been recorded up to 1,003 m 2 g −1 with a pore width of 29 Å for PI-COF-m, which is comparable with the 1,016 m 2 g −1 and 29 Å for PI-COF-s 34 . Scanning electron microscopy (SEM) images of PI-COF-s and PI-COF-m could exhibit homogeneous morphologies, consisting of the aggregation of bid-shaped porous structure (see Supplementary Fig. S5). The above results suggest that our PI-COF-m exhibited hexagonal network, a large surface area, and good morphology, all comparable to the control, PI-COF-s. In addition, the short reaction time of 2 h for PI-COF-m was 60 times faster than the 5 days required by the conventional solva-thermal method.

Functionalization of PI-COFs with modulators.
To functionalize PI-COF-m, we introduced the three equiv of the modulator agents being substituted for fractions X of TAPA, as indicated in Fig. 2. In our study, ratio X of the modulator was systematically varied from 0% to 60%. The modulator, 4-[(dimethylamino)methyl]aniline (DMMA), which included tertiary amine, that constructs the charge transfer complex with SO 2 , being a relatively unstable ionic structure that could also reduce the energy requirements for SO 2 desorption, would be the suitable functional building blocks for the SO 2 sorbent. With the increase in the contents of DMMA in PI-COF-m, the number of functional groups for imidization per the monomers decreased below two, which can interfere with the formation of a network structure and form broken framework (see Supplementary Table S1). With more than 30% of DMMA contents, it is difficult to obtain a porous crystalline structure 55 .
The FT-IR spectra have confirmed chemical functionality attributed to the modulators and pronounced broad peaks at 1,603, 1,450, and 1,250 cm −1 , which correspond to N-CH 2 bending, N-CH 3 bending, and C-N stretching for dimethyl amine group of modulator, respectively (see Supplementary Fig. S6). Fig. 3a, the crystallinity of PI-COF-mX has been monitored by PXRD. We observed the successful formation of the hexagonal network of the PI-COFs for up to X = 20. As noted above, the number of functional groups for imidization per building block below two, i.e., 1.82 for PI-COF-m40 and 1.62 for PI-COF-m60, could not construct the network structures corresponding to the decreased crystallinity. PI-COF-m10 and PI-COF-m20 revealed broad and relatively decreased peaks at hexagonal Bragg reflection mentioned above, which could be an evidence that the amorphous regions were increased, because the modulator disturbed the construction of the regular stacking of the PI-COFs. Furthermore, we examined the porosity of PI-COF-mX by measuring nitrogen gas (N 2 ) sorption at 77 K, revealing reversible isotherms, as shown in Fig. 3b.

Crystallinity and porosity of functionalized PI-COFs. As shown in
While the modulator-free synthesis produced PI-COF-m with a surface area of 1,003 m 2 g −1 , substitution of the modulators decreased the surface area. As shown in Fig. 3c, the surface areas were recorded up to 831 m 2 g −1 for PI-COF-m10, 548 m 2 g −1 for PI-COF-m20, 279 m 2 g −1 for PI-COF-m40, and 93 m 2 g −1 for PI-COF-m60. Because the network could snap due to the outer functional groups and the amorphous regions, which extended and disturbed the formation of well-defined pores, the modulators were increased. In general, PXRD results indicate that the broken network occurs above the 40% ratio. SEM images also indicate that the aggregation of bid-shape porous crystals collapsed as ratio X increased for the modulator agents (see Supplementary Fig. S5). SO 2 sorption and desorption on functionalized PI-COFs. PI-COF-mX was tested as a SO 2 sorbent to study the effect of functional groups and as a correlation of BET surface and crystallinity with the sorption capacity under anhydrous conditions using an apparatus similar to the one described in literature 27,[56][57][58] . In a typical experiment, the adsorbent (1 g) was loaded into a 25 mL sorption tube equipped with an electrical heater, temperature controller, and inlet and outlet valves. SO 2 (99.9%) was introduced into the sorption tube at 25 °C at a rate of 30 mL/min. The weight change during the SO 2 adsorption was monitored using a balance (accuracy: 0.001) and recorded on a computer until equilibrium was attained. The amount of SO 2 absorbed by the adsorbent was calculated by subtracting the mass of the initial adsorbent and the mass of SO 2 in an empty glass tube (0.084 g) from the total mass in the tube. In desorption process, the absorbed SO 2 was desorbed at 100 °C by flowing N 2 into the SO 2 loaded sample at a rate of 30 mL/min. The weight loss during SO 2 desorption was measured and noted until SO 2 removed completely.
As shown in Fig. 4a, molar SO 2 sorption capacities of PI-COF-mX were found to decrease as the ratio of modulators increased. PI-COF-m and PI-COF-m10 showed outstanding SO 2 sorption capacity, recording up to 6.50 mmol SO 2 g −1 (41 wt%) and 6.30 mmol SO 2 g −1 (40 wt%), respectively in Table 1 (see Supplementary Fig. S7). Within a short time period of 20 min, PI-COF-m and PI-COF-m10 could adsorb ca. 40 wt% of SO 2 gas and consistently maintain the sorption capacity just before desorption.
The imide backbone, which is fundamentally polar, provided a great advantage in efficiently absorbing the SO 2 gas 59 . The large surface area could afford the channels that can be gas was able to pass through. Furthermore, in PI-COF-m10, the strong basic dimethylamine (i.e., pK b = 4.3) could construct charge-complex structures with acidic SO 2 gas (i.e., pK a = 1.76) attributed to the Lewis base-acid chemical interaction. These structures were relatively unstable forms, i.e., Zwitterions, that could reduce the energy requirement for SO 2 desorption.
Furthermore, we also have conducted the SO 2 absorption experiment using water-saturated SO 2 gas. The concentration of water in SO 2 flow was 3.5 vol%. The SO 2 sorption capacity of PI-COF-m10 with the water-saturated SO 2 was 0.35 g SO 2 g −1 , that is lower than that of dry SO 2 (0.4 g SO 2 g −1 ) (see Supplementary Fig. S8a). However, the capacity was also maintained during 5 cycles. The FT-IR spectrum of the SO 2 absorbed PI-COF-m10 with the water-saturated gas also clearly showed the two peaks centered at 1,323 and 1,145 cm −1 , corresponding to asymmetric and symmetric SO 2 stretching peaks, however the intensities of those peaks were lower than those of the one treated with dry SO 2 (see Supplementary Fig. S8b). This result may indicate that the water block the SO 2 sorption sites in PI-COF-m10, thereby decreasing the SO 2 sorption capacity on the sorbent.

Trade off relation between surface area and functional groups. In addition, PI-COF-m20 and
PI-COF-m40 have almost similar SO 2 capacity despite the BET surface area of PI-COF-m20 being twice as large. This could suggest that there is a trade-off relation between BET surface area and the quantity of functional groups. Even when SO 2 sorption capacity for PI-COF-m60 was recorded at its lowest capacity of 4.74 mmol SO 2 g −1 (30 wt%) among the PI-COF-mX series, it was still at a high capacity among the reported SO 2 sorption materials. As shown in Fig. 4b, the SO 2 molar sorption capacities per the unit surface area of PI-COF-mX series exponentially increased up to 0.05 mmol SO 2 m −2 for PI-COF-m60, attributing to the decreased BET surface  area of 93 m 2 g −1 . In contrast, the N 2 molar absorption capacities per unit surface area of PI-COF-mX was almost identical regardless of modulator content because there have not been any functional groups for chemical interaction with N 2 . This also indicates that the dimethyl amine group of the modulator is efficient for capturing SO 2 . Moreover, as shown in Fig. 5, PI-COF-m10 showed much slower desorption kinetics than that of PI-COF-m under 25 °C conditions, resulting from the chemical interaction with SO 2 . Under general desorption conditions at 100 °C, adsorbed SO 2 was completely desorbed in a short time period.
Further inspection for the the interaction of PI-COF-m and PI-COF-m10 with SO 2 , we investigated FT-IR spectroscopy in Fig. 6. The FT-IR spectrum of PI-COF-m confirmed that there were not any new peaks or shifts upon contact with SO 2 . This suggests that there was no chemical interaction between SO 2 and PI-COF-m, indicating that PI-COF-m physically absorbed SO 2 via porous channels. However, for PI-COF-m10, upon contact with SO 2 , we observed new peaks at 1,323 and 1,145 cm −1 . The appearance of these peaks, being asymmetric and symmetric stretching peaks, respectively, indicates a chemical interaction between SO 2 and the tertiary nitrogen on the modulators 60 . After the desorption process, these new peaks disappeared, and the peak at 1,320 cm −1 , attributed to the aromatic C-N stretching vibration of TAPA moiety, reappeared respectively.
Comparison of the effect of functional groups on SO 2 sorbent. To clarify the influence of the broken framework resulting from incorporated modulator, we synthesized the PI-COF series with an amine-free modulator, i.e., 4-(tert-butyl)aniline (see Supplementary Fig. S9). These amine-free COFs (AF-COF X) also exhibit lower crystallinity along with increases of modulator ratio X, which is a result of disturbing the construction of the regular PI-COFs (see Supplementary Fig. S10). This trend is consistent with the trend of PI-COF-mX. However, in SO 2 sorption and desorption tests, we observed that SO 2 is only physically absorbed on the surface of AF-COF 10 and AF-COF 20 by the FT-IR results (see Supplementary Fig. S11). Thus, we could conclude here that the broken structure from incorporating the amine-free modulator barely influenced on SO 2 chemical sorption. Fig. 7, we tested the stability and reproducibility of the SO 2 adsorbency for PI-COF-m and PI-COF-m10 through five sorption-desorption cycles. For both sorbents, the release of SO 2 was completed within 5 min at 100 °C under flowing N 2 . For PI-COF-m, the sorption capacities gradually decreased to 0.33 g SO 2 g −1 (80%) as the cycles repeated. We observed that PI-COF-m10 repeatedly recycled without any loss of SO 2 sorption capacity, indicating that the process of SO 2 sorption via functionalized PI-COF-m10 is completely reversible.

Reversibility of functionalized PI-COF with 10% of modulators. As shown in
The FT-IR spectrum after five recycled desorption cycles is well matched with the first desorption spectrum, indicating that the adsorbed SO 2 was entirely desorbed even after several recycle processes. During the recycling test, dramatic change of sample temperature is also noted, which might influence the crystalline structure of PI-COFs. To demonstrate these crystallinity changes, the PXRD was measured after each of the desorption steps (see Supplementary Fig. S12). The relatively strong crystallinity of PI-COF-m steadily diminished with increasing fast recycle steps in PXRD results; this induced lower SO 2 sorption capacity than the initial state.
After one day of stabilization in a vacuum, PI-COF-m was recovered its initial crystallinity and SO 2 capacity, i.e., 0.38 g SO 2 g −1 (93%), by forming the thermodynamically stable crystal structure (see Supplementary  Fig. S13). In PI-COF-m10, which has relatively low crystallinity, there was no significant change in crystallinity, thus it was able to retain its initial SO 2 capacity over several recycling steps (see Supplementary Fig. S12b). This high stability of PI-COF-m10 could be attributed to its partially amorphous regions, which is the driving force in effectively maintaining and supporting the framework.
Given the above, a 10%-modulator-substituted PI-COF-m10 is the optimal point in the trade-off relation between surface area for physical absorption and amine functional groups for chemical absorption. In other words, a slightly disturbed crystal structure with 10% functionalization in PI-COF-m10 could accomplish both outstanding SO 2 capacity and high reproducibility.

Conclusion
In conclusion, we successfully synthesized new series of imide-linked COFs that incorporates DMMA as the modulator via a microwave-assisted reaction. The surface-functionalized channel of imide-linked COF for the interaction with SO 2 was achieved through the dimethyl amine functional groups of DMMA. In addition, well-defined large surface areas could afford the channels that can pass SO 2 gas through. These imide-based COFs utilizing SO 2 and tertiary amine-reversible interactions could be perfectly recovered in a recycling process.  Substituted as 10% of the functional group, the molar SO 2 sorption capacity was recorded up to 6.30 mmol SO 2 g −1 (40 wt%). As the ratio of the modulators increased, molar sorption capacities steadily decreased due to the remarkable decrease in surface area. However, capacities per unit surface area of the PI-COF-mX series were dramatically increased.
Furthermore, we found that functionalized PI-COF-m10 was completely reversible for SO 2 and highly stable on repeated sorption-desorption cycles. The slightly disturbed crystal structure with 10% tertiary amine functionalization in PI-COF-m10 could accomplish both outstanding SO 2 capacity and the high reproducibility. Our results suggest that channel-wall functional engineering could be a facile and powerful strategy for developing mesoporous COFs for high-performance gas storage and separation.

Methods
Materials. All chemicals were purchased from Sigma-Aldrich and used without further purification, except for tetrahydrofuran (THF) and dichloromethane (DCM), which were purified using a J.C. Metyer solvent dispensing system. DMMA was synthesized following the procedures in the Supplementary Information, while PI-COF-mX was synthesized with the modulator DMMA that we systematically increased from 0% to 60% (see Supplementary Information for details). Finally, SO 2 (99.9%) and N 2 (99.9%) were obtained from the Shin Yang Gas Chemical Co. Synthesis of PI-COF-m. PMDA (165 mg, 0.76 mmol) and TAPA (140 mg, 0.48 mmol) were evacuated for 2 h and then dissolved in a solution of mesitylene (3 mL)/NMP (3 mL)/isoquinoline (0.3 mL) in a glove box under N 2 atmosphere. The mixed solution was sealed under nitrogen in a 10 mL glass microwave tube, then heated by microwave irradiation at 200 °C with 300 W for 2 h using an Anton Paar microwave synthesizer (monowave 300) to afford a brown precipitate, which was isolated by filtration with purified THF (100 mL). The product was then immersed in THF (100 mL) for 8 h, during which the activation solvent was decanted and replaced four times. The solvent was removed in a vacuum at 100 °C to afford PI-COF-m as a brown powder (227 mg, 80%). Anal. Calcd for C 66 H 30  Characterization. 1 H NMR spectra were collected out using a Bruker DPX-300 (300 MHz) FT NMR system operating at 300 MHz. Fourier transform infrared spectra were recorded on a Cary 600 spectrometer equipped with a MCT-A (mercury cadmium telluride) detector with 5 mg samples. Elemental anlyses ware recorded on a Vario ELIII element analysis with 20 mg samples. PXRD data were carried out using a synchrotron radiation on the beam line 5A over the range of 2θ = 1.5-60.0° with a step size of 0.02° and 30 s per step at the Pohang Accelerator Laboratory (PAL), Pohang, Korea. BET data were recorded on a micrometrics ASAP 2010 equipment using N 2 gas at 77 K. TGA data were obtained under nitrogen atmosphere on a TGA Q50 anlayzer. FE-SEM was performed on a HITACHI S-4800 at 3 keV and 10 μA.