Author's personal copy

Microporous and Mesoporous Materials 192 (2014) 43 51 Contents lists available at ScienceDirect Microporous and Mesoporous Materials journal homepage: www.elsevier.com/locate/micromeso Mixed-linker zeolitic imidazolate framework mixed-matrix membranes for aggressive CO 2 separation from natural gas Joshua A. Thompson, Justin T. Vaughn, Nicholas A. Brunelli, William J. Koros, Christopher W. Jones, Sankar Nair School of Chemical & Biomolecular Engineering, 311 Ferst Dr. NW, Atlanta, GA 30332-0100, United States article info abstract Article history: Available online 10 July 2013 Keywords: Zeolitic imidazolate framework Mixed-matrix membranes Gas separation Plasticization CO 2 capture Zeolitic imidazolate framework (ZIF) materials are a promising subclass of metal organic frameworks (MOF) for gas separations. However, due to the deleterious effects of gate-opening phenomena associated with organic linker rotation near the limiting pore apertures of ZIFs, there have been few demonstrations of improved gas separation properties over pure polymer membranes when utilizing ZIF materials in composite membranes for CO 2 -based gas separations. Here, we report a study of composite ZIF/polymer membranes, containing mixed-linker ZIF materials with ZIF-8 crystal topologies but composed of different organic linker compositions. Characterization of the mixed-linker ZIFs shows that the mixed linker approach offers control over the porosity and pore size distribution of the materials, as determined from nitrogen physisorption and Horváth Kawazoe analysis. Single gas permeation measurements on mixedmatrix membranes reveal that inclusion of mixed-linker ZIFs yields membranes with better ideal CO 2 / CH 4 selectivity than membranes containing ZIF-8. This improvement is shown to likely occur from enhancement in the diffusion selectivity of the membranes associated with controlling the pore size distribution of the ZIF filler. Mixed-gas permeation experiments show that membranes with mixed-linker ZIFs display an effective plasticization resistance that is not typical of the pure polymeric matrix. Overall, we demonstrate that mixed-linker ZIFs can improve the gas separation properties in composite membranes and may be applicable to aggressive CO 2 concentrations in natural gas feeds. Ó 2013 Elsevier Inc. All rights reserved. 1. Introduction Corresponding authors. E-mail addresses: cjones@chbe.gatech.edu (C.W. Jones), sankar.nair@chbe.gatech.edu (S. Nair). Membranes are an attractive alternative to established industrial technologies for gas and liquid separations, because of lower energy cost and smaller footprint requirements in process operations [1,2]. Mixed-matrix membranes (MMMs), typically composed of a polymer and a molecular sieve, are a promising membrane architecture because they combine the ease of processibility associated with the polymeric matrix and improved molecular sieving properties from the filler phase [3 5]. In addition, MMMs can be a promising platform for evaluating the gas separation properties of microporous materials prior to developing fabrication techniques for pure microporous membranes [6 9]. The use of metal-organic frameworks (MOF) in MMMs has become increasingly popular for several reasons, including the large diversity in separation properties of MOFs and the good adhesion compatibility between the polymer and MOF phases during membrane formation (in contrast to zeolites). There have been several successful demonstrations of improving the gas or liquid separation properties of pure polymeric materials by use of different MOF materials as fillers [10 17]. An important observation from previous studies is that zeolitic imidazolate frameworks (ZIFs) used in MMMs often do not improve the membrane permselectivity [18 20]. ZIFs are a subclass of MOF materials that are structural analogues to microporous zeolites and have better chemical and thermal stability in comparison to many other types of MOFs [21 23]. ZIFs have been shown to exhibit a gate-opening phenomenon, whereby the organic linkers that bridge between transition metal centers rotate on the bond axis in the presence of different gases [24 27]. This linker rotation increases the effective pore size in ZIFs and affects gas transport through the ZIF crystal, usually leading to gas separation properties that are lower than would be expected based on static crystal structures [28,29]. Recently, a strategy for changing the linker composition in ZIFs has shown the capability of removing or controlling gate-opening characteristics in a ZIF structure [30,31]. By diminishing the gate-opening behavior, it may be possible to tailor the kinetic separation properties of these molecular sieves. In the present work, ZIF materials are synthesized utilizing synthesis techniques that allow control of the linker composition. By using mixed-linker compositions, it is shown that gate-opening behavior is not observed in these ZIFs, and there are noticeable 1387-1811/$ - see front matter Ó 2013 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.micromeso.2013.06.036

44 J.A. Thompson et al. / Microporous and Mesoporous Materials 192 (2014) 43 51 changes in the pore size distribution of the materials, as determined by Horváth Kawazoe analysis of nitrogen physisorption isotherms. Control over the pore size distribution and the linker composition of different ZIFs allows for modification of the effective permeability and diffusivity properties of mixed-matrix membranes prepared with these materials using the commercial poly(imide), Matrimid Ò 5128, based on single gas permeation experiments. Both the linker functionality and composition are shown to have significant effects on the permeability of CO 2 and CH 4 in the mixed-linker ZIF mixed-matrix membranes. Mixed gas permeation experiments demonstrate that although some enhancement in ideal permselectivity is exhibited in single gas permeation experiments, there is still little improvement in the permselectivity of mixed-matrix membranes compared to the pure Matrimid Ò polymer at high operating pressures. However, highpressure mixed gas permeation experiments show stabilization of CH 4 permeability when mixed-linker ZIF materials are used as the filler phase, indicating improvement in the plasticization resistance of the membranes compared to membranes made from the pure polymer. Finally, process calculations for a hollow fiber membrane module are performed to demonstrate that the major benefit of including these filler materials in mixed-matrix membranes is the significant reduction of membrane area for a given feed flow rate and desired product purity, due to large increases in the effective permeability. 2. Experimental methods 2.1. Materials Sodium formate (99%, NaCO 2 H), Zn(NO 3 ) 2 6H 2 O (99%), 2-methylimidazole (99%, 2-MeIM), 2-aminobenzimidazole (97%, 2-amBzIM), benzimidazole (BzIM), and chloroform (HPLC grade, CHCl 3 ) were obtained from Alfa Aesar. Dimethylformamide (99%, DMF) and methanol (99%, MeOH) were obtained from BDH. Matrimid Ò 5128 was obtained from Ciba. All materials were used without further purification. 2.2. Synthesis of ZIF-8 ZIF-8 was synthesized using a scaled-up procedure that is reported elsewhere [32]. In one solution, 80 mmol (6.57 g) 2-MeIM was dissolved in 200 ml MeOH. In a separate solution, 20 mmol (5.95 g) Zn(NO 3 ) 2 6H 2 O was dissolved in 200 ml MeOH. After stirring separately for 5 min, the Zn salt solution was poured into the 2-MeIM solution, and after 5 min of mixing, a milky white precipitate formed. The combined mixture was allowed to stir for 1 h at room temperature. The precipitate solution was then centrifuged at 10,000 rpm for 5 min. The supernatant was removed, and the precipitate was washed with MeOH. This procedure was repeated 3 times, and then the product was collected by vacuum filtration and dried at 85 C. The yield for the ZIF-8 product was approximately 25% based on Zn added to the solution. 2.3. Synthesis of ZIF-8-ambz-(x) ZIF-8-ambz-(15) and ZIF-8-ambz-(30) samples were synthesized using a procedure reported elsewhere [31]. First, 2 mmol (0.266 g) or 8 mmol (1.06 g) of 2-amBzIM, 18 mmol (1.48 g) or 12 mmol (0.985 g) 2-MeIM, and 5 mmol (0.34 g) NaCO 2 H were dissolved in 50 ml deionized (DI) H 2 O by heating the solution to 70 C for 2 h to form the imidazole solution for ZIF-8-ambz-(15) or ZIF-8- ambz-(30). A separate solution containing 5 mmol (1.49 g) Zn(NO 3 ) 2 6H 2 O was prepared with 50 ml DMF. After the imidazole solution cooled to room temperature, the Zn salt solution was added, and a precipitate formed almost immediately. The solution was allowed to react for 1 h at room temperature before centrifugation at 10,000 rpm for 10 min. The supernatant was removed, and the precipitate was washed with MeOH. This procedure was repeated 3 times, and then the product was recovered by vacuum filtration and dried at 85 C. The yield of ZIF-8-ambz-(x) samples was approximately 30 40% based on Zn added to the solution. 2.4. Synthesis of ZIF-7-8-(20) ZIF-7-8-(20) was synthesized by utilizing the non-solvent induced crystallization method, similar to the ZIF-8-ambz-(x) samples. A solution of 8.4 mmol (0.992 g) BzIM, 111.6 mmol (9.16 g) 2-MeIM, and 120 mmol NaCO 2 H was prepared in 300 ml MeOH. A separate solution of 30 mmol (8.92 g) Zn(NO 3 ) 2 6H 2 O was prepared in 300 ml DMF. Once both solutions were clear, the Zn salt solution was poured into the imidazole solution. The combined mixture was then placed in an oven in a sealed polyethylene bottle and allowed to stir for 24 h at 50 C. After cooling to room temperature, the milky solution was centrifuged at 10,000 rpm for 5 min. The supernatant was removed, and the precipitate was washed with MeOH. This washing procedure was repeated 3 times, and then the product was recovered by vacuum filtration and dried at 85 C. The yield of ZIF-7-8-(20) was approximately 10% based on Zn added to the solution. 2.5. Mixed-matrix membrane fabrication Mixed-matrix membranes were prepared by first dispersing ZIF samples in a solvent, and then after addition of polymer, films were prepared using the solution-casting technique [33]. In a typical film preparation procedure, 0.15 g ZIF sample was sonicated in 5 ml CHCl 3, using a sonication bath with a sonication intensity of 3.8 W cm 2. Once particles appeared well-dispersed after approximately 1 2 h of sonication, 0.2 g Matrimid Ò 5128 (chemical structure shown in Fig. 1) was added and allowed to dissolve while sonicating in the sonication bath. The addition of Matrimid Ò acts as a primer to cover the ZIF surface with polymer to help promote adhesion to the polymeric phase and to prevent particle aggregation during film preparation [34]. Once the polymer was fully dissolved, the primed ZIF dispersion was sonicated using an ultrasonication horn with a sonication intensity of 160 W cm 2. To prevent solvent evaporation, primed ZIF dispersions were sonicated with the horn in 30 s intervals, and this procedure was repeated twice. The primed dispersion was then sonicated in the bath again for 1 h. The primed ZIF dispersion was sonicated with the horn for two more 30 s intervals and then poured over the remaining balance of Matrimid Ò (0.65 g) to obtain the desired ratio of ZIF to polymer (0.15:0.85 w/w). The membrane dope solution was allowed to tumble overnight on a roller. A glove bag was prepared with N 2 /CHCl 3 atmosphere, and using a casting knife of 200 lm, a film was manually cast and allowed to vitrify and dry overnight at room temperature. All films were roughly 30 50 lm in thickness, measured using a micrometer, and films were annealed at 225 C for 24 h under vacuum [33,34]. Fig. 1. Chemical structure of commercial polyimide, Matrimid Ò 5128.

J.A. Thompson et al. / Microporous and Mesoporous Materials 192 (2014) 43 51 45 2.6. Characterization methods The ZIF materials were analyzed by powder X-ray diffraction (XRD) using an X Pert Pro PANalytical X-ray Diffractometer. Diffraction measurements were done from 3.5 50 2h using an X celerator detector. Nitrogen physisorption measurements were measured on a Micromeritics ASAP 2020 surface area analyzer at 196 C. All samples were degassed at 250 C for 18 h prior to physisorption measurements. The BET surface area and t-plot micropore volume methods were used to analyze the relative surface properties of each sample. Horváth Kawazoe pore size distributions were used to compare the relative changes in pore size distributions between the different ZIF samples. The methodology for this analysis has been described elsewhere [30,35]. To determine the linker composition in the ZIF framework, all samples were analyzed with solution 1 H nuclear magnetic resonance (NMR) spectroscopy on a Mercury Vx 400 MHz spectrometer after digesting samples using d 4 -acetic acid (CD 3 CO 2 D). Thermogravimetric analysis (TGA) was performed using a Netzsch STA-409- PG. Powder samples were heated from room temperature to 900 C with a heating ramp rate of 10 C min 1. Scanning electron microscopy (SEM) imaging of ZIF particles was done on a Zeiss Leo 1550 electron microscope. Samples were coated with gold by sputtering under vacuum, and images were taken with an accelerating voltage of 10 kv. Mixed-matrix membrane films were characterized using Fourier-transform infrared (FTIR) spectroscopy and SEM. FTIR measurements were done in transmission mode on a Bruker Vertex 80v FTIR analyzer from 4000 to 400 cm 1. SEM imaging was done on a Zeiss LEO 1550 SEM. Membrane microscopy samples were prepared by submerging films in liquid nitrogen and fracturing the film to examine the cross section of the mixed-matrix membrane. Samples were coated by sputtering gold under vacuum, and images were taken with an accelerating voltage of 10 kv. 2.7. Gas permeation measurements Permeation measurements were performed using a constant volume permeation cell described in earlier work [36]. A small area of the film was cut out from a larger membrane film, and using aluminum tape, a mask was prepared with approximately 1 cm in diameter of exposed membrane area. At least two areas of a film and two separate films were tested for each membrane reported for single gas measurements. The membrane area edges were sealed using Duralco 4525 high temperature epoxy obtained from Cortronics and allowed to set overnight before sealing the permeation cell. After insertion of the film into the permeation cell, the film was degassed at 35 C for at least 24 h before each permeation test. Leak tests were done before each permeation experiment, ranging from 10 10 to 10 9 bars 1. Subsequent permeation tests following the first test were performed after degassing both sides of the film under vacuum for 12 24 h and testing the leak rate again. Single gas permeation experiments were performed at 35 C with 3.45 bar of upstream pressure of either CO 2 or CH 4. Measurements started once upstream gas was introduced to the cell and the downstream was evacuated (<10 5 bar). Permeability of CO 2 and CH 4 were calculated by the following equation: P ¼ dp Vl dt RTA m Df where dp is the rate of pressure rise, V is the volume of the permeate dt side of the membrane, l is the thickness of the membrane, R is the ideal gas constant, T is the absolute temperature, A m is the measured area of the exposed membrane, and Df is the fugacity difference of the feed and permeate sides of the membrane. Additionally, ð1þ because the permeate side pressure is very low in comparison to the feed, the driving force of each gas for permeability calculations in both single and mixed gas measurement can be assumed as the fugacity of the feed. Time lag measurements were performed by analyzing the time for gas to permeate through the film. By this method, the approximate time to reach the permeation flux steady-state and the apparent diffusion coefficient of the film can be calculated by the following equation [12]: D app ¼ l2 6h where D app is the apparent diffusion coefficient of the membrane, l is the membrane thickness as measured by a micrometer, and h is the measured time lag of the membrane film. The approximate time to reach steady-state can be estimated by allowing each permeation measurement to take at least 4 6 h, but in most cases, 10 h was used for each measurement. Although fillers have been shown to affect the measured time lag in mixed-matrix membranes, this typically only occurs if the fillers strongly adsorb the gas that is being tested [37,38]. It has been shown previously that the primary amine on the 2-amBzIM substituting linker does not greatly influence the overall CO 2 affinity in these mixed-linker ZIF materials [31], making Eq. (2) a reasonable approximation for the apparent diffusion coefficient through a mixed-matrix membrane film. Mixed gas permeation experiments were performed on a constant volume apparatus with a feed mole fraction of 50:50 CO 2 :- CH 4. Prior to mixed gas testing, each membrane was tested for single gas permeation to confirm if there were noticeable differences in performance between mixed gas-tested films and single gas-tested films. Films for mixed gas measurements had approximately 2.5 times larger areas than films used for singlegas measurements, in order to obtain a better representative performance of each membrane sample. The data presented for the mixed gas experiments represent one membrane sample. The retenate flow rate was set to approximately 100 times the downstream gas flux in the permeate side of the apparatus (6 ml min 1 or more). Pressures tested for each film were approximately 6.9, 13.8, 27.6, and 41.4 bar of total feed pressure. To confirm that at each pressure point the membrane was at steady state, films were allowed to equilibrate over a 12 h period, which was well-beyond the time required to reach steady-state permeation as estimated by the pure gas time lag measurements. Additionally, permeability was measured 1 3 h after the steady-state time had been reached to confirm equilibrium. The composition of gas in the permeate side was determined by using a Bruker Daltronics Varian 450 gas chromatograph (GC) with a thermal conductivity detector and He reference gas. At each pressure point, at least 3 6 GC injections were done to obtain an average of the permeate composition, and at least two injections at different times were performed to check that each film was at steady state. The permeate composition was used to calculate the separation factor: SF ¼ ðy CO 2 =y CH4 Þ perm ðy CO2 =y CH4 Þ feed where y i is the mole fraction of component i in the feed or permeate side. Permeability values were calculated by the molar flux across the membrane (see Eq. (1)) and the gas composition on the permeate side obtained from GC measurements, and mixed gas permselectivity is the ratio of the CO 2 and CH 4 permeability values using the equation: a ¼ P CO 2 P CH4 ð2þ ð3þ ð4þ

46 J.A. Thompson et al. / Microporous and Mesoporous Materials 192 (2014) 43 51 where P i is the permeability of component i. It should be noted that as the feed pressure is reduced, the mixed gas permselectivity obtained from Eq. (4) will appear to be the same as the separation factor expressed in Eq. (3), due to minimal differences in driving force for a 50:50 CO 2 :CH 4 feed mixture. 2.8. Hollow fiber permeation calculations To compare the effects of adding ZIFs to mixed-matrix membranes, process calculations were done using the mixed gas permeation data obtained for each membrane sample. A single hollow fiber membrane module was modeled using a previously discussed methodology, assuming counter-current flow with a shell side feed [39]. The permeance of the hollow fiber module was predicted by assuming the selective skin layer of the membrane to be 500 nm in thickness, and permeance can then be predicted by the equation: Permeance ¼ Permeability l where l is the assumed selective skin layer. By this prediction, the required membrane area and resulting CH 4 recovery to obtain a 98 mol% CH 4 product stream was calculated as a function of feed pressure, assuming a 50:50 CO 2 :CH 4 feed stream of 1 m 3 h 1 volumetric flow rate. This approach is similar to procedures used to compare performance of different membrane materials in the literature [40]. In addition, the product stream was varied from 90 to 98 mol% CH 4 to understand the relative changes in required membrane area and CH 4 recovery for each membrane sample. ð5þ chosen here represent two hypotheses: (1) the organic functionality in small pore ZIFs affects the overall transport through the ZIF crystal based on differences in effective flexibility at ambient temperatures [29], and (2) increasing the amount of substitution of bulky organic linkers in the ZIF sample, while maintaining the same ZIF crystal structure, increases the overall gas transport permselectivity for CO 2 /CH 4. To confirm that the observed linker amounts are not an artifact of post-synthesis washing and purification of the mixed-linker ZIFs, TGA measurements (Fig. S1 and Table S1) reveal organic:inorganic ratios that are very close to those obtained from the NMR measurements. Fig. S2 in the Supporting information shows the powder XRD patterns of each sample compared with the simulated ZIF-8 structure, using Mercury crystal structure simulation software [41]. In all cases, even at the higher 2-amBzIM substitution, the cubic I- 43m ZIF-8 structure was maintained and was observed previously [30,31]. However, there appears to be some differences in peak positions and peak broadening related to framework substitution in the ZIF crystal structure. This may be due to increased crystalline strain or disorder associated with addition of bulky organic linkers in the ZIF framework. SEM images of the various ZIF particles are shown in Fig. S3 in the Supporting Information, and only the ZIF- 8 sample was observed to have very small nanoparticles, which could contribute to increased XRD peak broadening. Fig. 3a shows high-resolution, low-pressure N 2 physisorption isotherms of the ZIF samples used in mixed-matrix membrane 3. Results and discussion 3.1. ZIF synthesis and characterization ZIF samples with varied compositions were synthesized to understand the relative changes in effective permeability and diffusivity in mixed-matrix membranes made using hybrid ZIFs. Fig. 2 shows the 1 H NMR spectra (after dissolution) of each sample prepared, following degassing at 250 C to remove occluded solvent from the synthesis. The NMR peaks were assigned, and the calculated linker substitutions for the samples were determined to be: 100 mol% 2-MeIM in ZIF-8; 20 mol% BzIM and 80 mol% 2- MeIM in ZIF-7-8-(20); 15 mol% 2-amBzIM and 85 mol% 2-MeIM in ZIF-8-ambz-(15) sample; and 30 mol% 2-amBzIM and 70 mol% 2-MeIM in ZIF-8-ambz-(30) sample. The linker substitutions Fig. 2. Solution 1 H NMR of ZIF samples prepared for mixed-matrix membrane fabrication. Fig. 3. Nitrogen physisorption (a) and Horváth Kawazoe pore size distributions (b) of ZIF samples, showing decrease in surface area and micropore volume as ZIF composition changes and a shift to smaller PSD.

J.A. Thompson et al. / Microporous and Mesoporous Materials 192 (2014) 43 51 47 Table 1 Linker substitution amounts, framework density, BET surface area and t-plot micropore volume of ZIF samples used in mixed-matrix membrane films. Sample Linker substitution (mol%) Framework density (g cm 3 ) preparation. As different linkers substituted in the ZIF-8 crystal structure, there was a decrease in both the BET surface area and t-plot micropore volume (Table 1). Additionally, all mixed-linker ZIFs with bulky benzimidazole-type linkers showed no evidence of gate opening in the physisorption isotherms. The H K pore size distributions (PSD) of each sample were calculated and are shown in Fig. 3b. Interestingly, ZIF-7-8-(20) and ZIF-8-ambz-(15) had the same PSDs, making comparison of permeation data crucial to understanding the effects of organic functionality on bridging organic linkers in small pore ZIFs. As the substitution of 2-amBzIM was increased as in the case of ZIF-8-ambz-(30), the PSD shifted to smaller pore widths, possibly creating more difficult diffusion pathways for larger gas molecules, such as CH 4. The BET surface area and micropore volume also dropped sharply, thereby indicating a large reduction in the gate-opening effect of the ZIF-8 linker. 3.2. Mixed-matrix membrane performance BET surface area (m 2 g 1 ) ZIF-8 0 0.93 1860 0.67 ZIF-7-8-(20) 20 0.98 1100 0.39 ZIF-8-ambz-(15) 15 0.99 1150 0.40 ZIF-8-ambz-(30) 30 1.05 350 0.12 t-plot micropore volume (cm 3 g 1 ) Mixed-matrix membrane dense film cross sections shown in Fig. 4 demonstrate good adhesion to the Matrimid Ò polymeric matrix and adequate dispersion throughout the film. It is important to note that without the priming step described previously there is inadequate dispersion and adhesion of ZIF particles throughout the polymeric matrix; therefore, it is recommended that further MMM studies containing ZIF materials to use a polymeric priming step to obtain well-dispersed particles, which has been shown to be beneficial for other materials, such as carbon molecular sieves [33]. Unlike previous studies that utilized high-intensity sonication as a method for particle dispersion in MMM fabrication, the present ZIF particles did not exhibit any significant changes in particle size or morphology following membrane fabrication in this work [31,42]. We hypothesize that the solvent chosen for this study, CHCl 3, does not provide sufficient solubility for the ZIF framework to cause any changes in the particle size or morphology. In addition, the ZIF materials show good adhesion to the polymer matrix, but there is some delamination from the polymer that may be a result of fracturing the membrane in liquid nitrogen. It may be possible to use other characterization techniques that obtain a more representative visualization of the ZIF/polymer interfaces [43,44]. FTIR spectra of annealed films (Fig. S4 in the Supporting Information) show vibrations typical for Matrimid Ò, with both symmetric and asymmetric vibrations for the imide group in the polymer backbone at approximately 1700 cm 1. There was no apparent shift in the imide vibrations when ZIFs were added to the polymeric film. The presence of ZIF materials was evident in the FTIR spectra due to v(z N) at 450 cm 1 and to v(c H) at 3150 cm 1 corresponding to the imidazolate ring. The v( N H) band that appeared at 3400 cm 1 in the ZIF-8-ambz-(x) samples was due to the presence of the primary amine functional group in the mixed-linker ZIFs. Because there were no changes in the carbonyl regions of the spectra of the MMM samples containing ZIF-8- ambz-(x) materials, these spectra suggest no chemical reaction occurred between the primary amine moiety with the carbonyl group in the polymer backbone. Permeation data for pure Matrimid Ò and ZIF/Matrimid Ò membranes are shown in Fig. 5. As expected from previous studies [19], when ZIF-8 was used as the filler phase in a highly-selective, glassy polymeric matrix, there was no change (or only a slight decrease) in the ideal selectivity but there was a large increase in the effective permeability of the membrane. These permeation characteristics for ZIF-8 are consistent with that reported recently in [28]. To compare the effect of functional groups in small pore ZIFs, ZIF- 7-8-(20) and ZIF-8-ambz-(15) which have similar micropore volume and PSD were used with the same filler weight loading and membrane fabrication steps. As Fig. 5 shows, there was a slight increase in ideal permselectivity when ZIF-7-8-(20) was used as the filler phase and a decrease in the effective CO 2 permeability, compared with the ZIF-8/Matrimid Ò films. However, when ZIF-8- ambz-(15) was used as the filler phase, there was a further increase Fig. 4. SEM images of 15 wt% ZIF/Matrimid films fabricated, using priming and solution casting techniques: (a) ZIF-8; (b) ZIF-7-8-(20); (c) ZIF-8-ambz-(15); (d) ZIF-8-ambz- (30).

48 J.A. Thompson et al. / Microporous and Mesoporous Materials 192 (2014) 43 51 Table 2 Diffusion coefficients of Matrimid Ò and ZIF/Matrimid Ò films, determined by the time lag method. Estimated error of each measurement is shown in parentheses. Sample CO 2 diffusivity (10 9 cm 2 s 1 ) CH 4 diffusivity (10 9 cm 2 s 1 ) Diffusion selectivity Matrimid Ò 8 (2) 0.8 (0.2) 11 (1) 15 wt% ZIF-8/ 25 (2) 2.3 (0.2) 10 (1) Matrimid Ò 15 wt% ZIF-7-8-(20)/ 18 (2) 1.5 (0.1) 11 (1) Matrimid Ò 15 wt% ZIF-8-ambz- 10.6 (0.7) 0.86 (0.04) 12 (1) (15)/Matrimid Ò 15 wt% ZIF-8-ambz- (30)/Matrimid Ò 8.8 (0.5) 0.67 (0.05) 13 (1) Fig. 5. Single gas permeation results of ZIF/Matrimid membranes at 3.45 bar and 35 C. There is an increase in ideal selectivity when there is a substitution in the ZIF- 8 framework for a bulkier organic linker. in CO 2 /CH 4 ideal permselectivity compared to the pure polymer, the ZIF-8/Matrimid Ò membrane, and the ZIF-7-8-(20)/Matrimid Ò membrane samples. Additionally, there was a significant reduction in CO 2 permeability compared to the ZIF-8/Matrimid Ò membrane. Because the ZIF-8 unit cell has eight pore windows in its unit cell and the amount of substitution for 2-amBzIM in the sample was roughly more than 1/7 of total linkers in the framework, there may be substantial intermolecular interaction between the 2-amBzIM linkers, possibly via hydrogen bonding, that could change the effective flexibility of the ZIF pore window in comparison to ZIF-8 or ZIF-7-8-(20), thus altering the diffusion pathway through the ZIF crystal [28,45]. Other mixed-matrix membrane studies using both amine-functionalized and unfunctionalized MOF fillers have shown substantial differences in CO 2 and CH 4 permeation characteristics between membrane samples, likely due to structural changes in the MOF filler and not due to the pendant amine enhancing separation performance based on affinity for CO 2 due to the similar CO 2 adsorption properties of the functionalized and unfunctionalized MOF samples [14,38]. As the effective substitution of 2-amBzIM was further increased to 30 mol%, there was a concurrent increase in ideal permselectivity with a slight decrease in the effective CO 2 permeability compared to ZIF-8-ambz-(15)/Matrimid Ò. It is likely that at the studied linker loadings there was not significant hindrance of CO 2 diffusion as the substitution of 2-amBzIM increases. However, when the mixed-linker ZIF/Matrimid Ò membranes are compared, there are apparent differences in the CO 2 permeability when BzIM or 2-amBzIM was used in the mixed-linker ZIF structure. Regardless, the permeation results suggest that the mixed-linker ZIF approach does provide enhancement in CO 2 /CH 4 selectivity for a ZIF-8-type structure, and this approach may be useful for tuning not just the CO 2 /CH 4 gas separation properties, but other molecular separations in a variety of ZIF materials [17,28,46]. Effective diffusion coefficients for each film were calculated using the time lag method and are shown in Table 2. When the diffusion coefficients are compared with the permeability values obtained for each membrane, it is apparent that the ideal permselectivity increase observed for the mixed-linker ZIFs was related to an increase in apparent diffusion selectivity, and the decrease in CO 2 permeability was linked to a decrease in the apparent diffusivity coefficient. This implies that substitution of 2-MeIM for a bulkier organic linker, such as BzIM or 2-amBzIM, enhanced the effective transport selectivity for CO 2 /CH 4. This substitution may also enhance other gas pair separations for these mixed-linker ZIFs. For instance, ZIF-8 has been shown to have low C 2 H 4 /C 2 H 6 kinetic selectivity [47,48], but careful substitution with a bulky linker may improve the diffusion selectivity without severe hindrance of diffusion for C 2 H 4. It should be noted that at the highest linker substitution in the ZIF-8 framework, ZIF-8-ambz-(30), the diffusion coefficient of CO 2 was close to the Matrimid Ò diffusion coefficient. If there is any further increase in substitution, there may be a substantial drop in diffusion for CO 2, resulting in a plugged-sieve type behavior for mixed-matrix membranes [33]. This behavior implies that there is an effective trade-off in the amount of linker substitution that can be used for improving the molecular separation properties in small pore ZIF materials. Mixed gas permeation isotherms can give insight into the behavior of membranes under realistic feed conditions for natural gas purification [49]. Single-gas permeation data can be misleading if used to predict the permselectivity under mixed-gas conditions, due to the effects of competitive sorption and diffusion, as well as CO 2 plasticization at high operating pressures [50,51]. Separation factors of membranes determined from mixed gas permeation experiments are shown in Fig. S5 in the Supporting Information. As previous studies have shown [40], the ideal permselectivity of pure Matrimid Ò is often lower than the mixed gas separation factor and permselectivity, which is evident when comparing Fig. 5 and Fig. S5 for pure Matrimid Ò membranes in the low feed pressure region. The composite membranes showed interesting behavior as the partial pressure of CO 2 is increased in the feed. The separation factor is close to or even less than pure Matrimid Ò, below the expected plasticization pressure even though some composite membranes showed ideal permselectivity enhancement in pure gas measurements. However, above the plasticization pressure, the ZIF-8-ambz-(30) composite membrane showed higher separation factor than pure Matrimid Ò and the other composite membranes. Additionally, the composite membranes showed a nearly constant separation factor as the partial pressure of CO 2 was increased by 7 bar above the plasticization pressure (from 13 bar to 21 bar of CO 2 partial pressure) while Matrimid showed a continuous decrease in separation factor. This indicates that the presence of mixed-linker ZIFs in the polymeric matrix suppressed the plasticization effect that is typically expected in glassy polymeric membranes [52]. The mixed gas CO 2 and CH 4 permeabilities (Fig. 6) reveal that at the highest pressure tested the CH 4 permeability was stabilized in the MMMs while the CO 2 permeability continued to decrease, similar to the behavior in pure Matrimid Ò. The near-constant CO 2 permeability at high CO 2 partial pressure is a result of saturated sorption sites in the surrounding polymeric matrix as the feed pressure increased [52 54]. The stabilized CH 4 permeability for the composite membranes above the plasticization pressure suggests that the observed improvement in separation factor found in membranes containing mixed-linker ZIFs was caused by

J.A. Thompson et al. / Microporous and Mesoporous Materials 192 (2014) 43 51 49 Fig. 7. Mixed gas permselectivity of membrane samples with increasing feed and CO 2 partial pressure. Error bars represent variance in GC injection measurements. Shaded area represents typical plasticization pressure for Matrimid Ò. Fig. 6. Permeability of CO 2 (a) and CH 4 (b) calculated from mixed-gas permeation, normalized by fugacity driving force. Error bars represent variance in GC injection measurements. Dashed lines represent permeability values from pure gas measurements. Shaded area represents typical plasticization pressure for Matrimid Ò. suppression of plasticization when the ZIF was used as the filler phase in Matrimid Ò. As the surrounding polymeric matrix likely continues to swell due to an increase in the partial pressure of CO 2, resulting in increased CH 4 permeability around the ZIF (as seen in pure Matrimid Ò ), the ZIF crystals may exhibit some hindrance to CH 4 transport with increasing pressure, giving the apparent enhancement in plasticization resistance. Chmelik et al. have explored the transport diffusion properties of ZIF-8 in pure gases and mixtures of CO 2 and CH 4 using IR microscopy [55]. The authors found that the diffusion coefficient of CH 4 increased considerably with loading (analogous to increasing feed pressure) in the absence of CO 2. When mixtures of CO 2 and CH 4 in different ratios were present in the ZIF-8 crystal, the CH 4 diffusivity did not exhibit an increase with adsorbate loading and instead diffusion was apparently hindered due to the presence of CO 2. Molecular simulations [55] revealed that the CO 2 adsorption sites exist near the pore window of ZIF-8, and this adsorption behavior slows the transport of CH 4 in gas mixtures as the adsorbate loading increases. Hindering diffusion effects have also been shown with other microporous materials, such as zeolite DDR, using molecular simulations [56]. It is possible that the introduction of bulky organic linkers while maintaining a ZIF-8-like structure results in further hindrance of CH 4 diffusion as the loading of CO 2 inside the ZIF crystal increases with feed pressure. This increased hindrance would result in apparent CH 4 permeability and permselectivity stabilization (Fig. 7) with increasing partial pressure of CO 2 if the mixed-linker ZIFs have higher permselectivity than the parent ZIF-8 material. The mixed-gas permselectivity shown in Fig. 7 reveal that the mixed-linker ZIFs have mixed gas permselectivity comparable to Matrimid Ò at lower feed pressures, but as the CO 2 partial pressure increases, the mixed gas permselectivity inside the ZIF crystal shows improvement and even suppresses transport of CH 4 through the ZIF crystal, resulting in an effective stabilization of CH 4 permeability in the composite membranes (Fig. 6a). Increasing either the weight loading of ZIF in the polymer or increasing the molar ratio of the substituting bulky organic linkers in the ZIF framework may further improve the permselectivity and separation performance of the composite membrane. However, as diffusivity measurements showed, there may exist an upper limit of the linker substitution to maintain highly permeable ZIF materials. More importantly, these results are the first to elucidate the permeation behavior of dense-film mixed-matrix membranes above the CO 2 plasticization pressure of the surrounding polymeric matrix and at CO 2 partial pressures that are relevant for industrial gas separations [49]. Although there is increased polymeric chain mobility associated with CO 2 -induced swelling and plasticization [52], there is no degradation of the polymer/zif interface that results in large changes in the CH 4 permeability or separation factor for CO 2 /CH 4 at high feed and CO 2 partial pressures. 3.3. Hollow fiber membrane simulations Hollow fiber membrane simulations are useful for comparing different membranes without determining proper hollow fiber spinning and fabrication techniques, and these simulations can help screen different filler materials without the need of testing each filler sample [39,57]. Furthermore, it has already been shown that the membrane properties of mixed-matrix dense film membranes containing ZIF-8 are transferrable to hollow fiber membranes with suitable spinning and fabrication [58]. Table S2 in the Supporting Information shows the permeance values assumed for each membrane studied, calculated by assuming a 500 nm selective skin layer and using the mixed gas permeability values of CO 2 and CH 4 calculated from mixed gas testing. Because Matrimid Ò shows a decrease in selectivity as the CO 2 feed pressure increases in mixed gas streams, it is important to consider anticipated separation performance under conditions more closely related to industrially relevant separations [49]. The dimensions of

50 J.A. Thompson et al. / Microporous and Mesoporous Materials 192 (2014) 43 51 Table 3 Hollow fiber membrane module dimensions and parameters. Hollow fiber module properties Feed flow rate (m 3 (STP)h 1 ) 1.00 Temperature ( C) 35 Fiber outer diameter (lm) 300 Fiber inner diameter (lm) 150 Active fiber length (m) 0.8 Permeate pressure (kpa) 103 Feed location Shell side CO 2 :CH 4 feed composition 50:50 mol% the hollow fiber module and the conditions assumed for the fiber module are shown in Table 3 and represent typical commercial hollow fiber module properties [49]. The first case considered for hollow fiber membrane process simulations was done under the requirement that 98 mol% CH 4 and 2 mol% CO 2 are the product composition based on the feed flow rate of 1 m 3 h 1. The flow rate chosen for these case studies represents a base case, and the membrane area can be scaled linearly for larger feed flow rate [40]. The required membrane area to reach this composition, and the total CH 4 recovery, are shown in Fig. S6. Because the separation factor and permselectivity do not change greatly when ZIF materials are added to the polymeric matrix, the major difference in performance between pure Matrimid Ò and the composite hollow fiber membranes is the area needed to reach the product composition. Therefore, if the cost of ZIF materials is the same or lower than the polymeric matrix, the inclusion of ZIFs in the membrane module provides a significant cost benefit by reducing the overall area required for CO 2 /CH 4 separations. Fig. S6a also shows a significant reduction in membrane area with increasing feed pressure due to higher effective flux through the hollow fiber module [40]. The approximate membrane area reduction by adding ZIFs to the polymeric matrix is at least 25% and varies based on the feed pressure and the mixed-linker ZIF used in the composite membrane. Because membrane area typically scales linearly with feed flow rate, a 25% reduction of membrane area can be very cost-beneficial as the desired feed flow rate increases, which ultimately requires more membrane area. The second case considered for process simulations is varying the required product composition with a constant feed pressure of 42 bar. The membrane area and CH 4 recovery are shown in Fig. 8. Again, because the separation performance of all the membranes studied here does not change substantially with inclusion of ZIF in the polymer, the CH 4 recovery does not vary greatly between the membranes. The benefit of ZIF materials is manifested in the reduction of required membrane area to reach the target CH 4 product composition. To avoid a great loss in CH 4 recovery, it may be beneficial to use these membranes for a rough cut of the CO 2 / CH 4 feed stream that can then be further purified by other separation techniques. For instance, if a product composition of 8 mol% CO 2 is desired for transport in pipelines to a gas processing facility [45], these simulations predict CH 4 recovery of approximately 93% can be achieved with one module. Additionally, it may be possible to increase the overall CH 4 recovery with these mixed-matrix membranes by utilizing higher weight loadings of ZIF in the polymer or using a smaller selective skin layer [59,60]. As these simulations and mixed gas experiments have shown, inclusion of mixed-linker ZIFs can reduce the overall required membrane area to perform natural gas purification compared to the case of the comparable pure polymer membrane. It may be possible to use mixed-linker syntheses on smaller pore ZIF materials (e.g., ZIF- 11) to further enhance the observed separation performance in ZIF-containing mixed-matrix membranes without significantly altering the diffusion of CO 2 inside the ZIF crystal. 4. Conclusions This work has shown that tuning of the organic linker composition in mixed-linker ZIF materials allows control of permeation of gases in mixed-matrix membranes. This ultimately results in improved separation properties, as shown by single and mixed gas CO 2 /CH 4 permeation data for mixed-matrix membranes containing mixed-linker ZIFs. The improvements occur either due to enhanced CO 2 permeability, CO 2 /CH 4 permselectivity, or CO 2 plasticization resistance. High-pressure mixed gas permeation experiments show that ZIF materials effectively stabilize the permselectivity with increasing pressure by suppressing the transport of CH 4 in the composite membrane, thereby achieving a permselectivity of 40 at 20 bar of total CO 2 partial pressure. Hollow fiber membrane process simulations revealed that the largest benefit of including ZIF materials in mixed-matrix membranes is reduction of membrane area for a target gas separation and feed conditions. These simulations show that mixed-linker ZIFs have promise as fillers in MMMs, and that hollow fiber membranes containing these materials should be prepared as the next step in investigating the effective separation properties of mixed-linker ZIF materials and bringing them closer to potential applications in gas separations. Acknowledgment This work was supported by King Abdullah University of Science and Technology under Award No. KUS-I1-011-21. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.micromeso.2013. 06.036. References Fig. 8. Required membrane area and CH 4 recovery as a function of target CH 4 composition in the product stream. Closed symbols: membrane area; open symbols: CH 4 recovery. [1] W.J. Koros, R.P. Lively, AIChE J. 58 (2012) 2624 2633. [2] S. Sridhar, B. Smitha, T.M. Aminabhavi, Sep. Purif. Rev. 36 (2007) 113 174. [3] B. Zornoza, C. Téllez, J. Coronas, J. Gascon, F. Kapteijn, Microporous Mesoporous Mater. 166 (2013) 67 78. [4] H.B.T. Jeazet, C. Staudt, C. Janiak, Dalton Trans. 41 (2012) 14003 14027. [5] Y. Zhang, J. Sunarso, S. Liu, R. Wang, Int. J. Greenh. Gas Con. 12 (2013) 84 107. [6] K.-S. Jang, H.-J. Kim, J.R. Johnson, W. Kim, W.J. Koros, C.W. Jones, S. Nair, Chem. Mater. 23 (2011) 3025 3028.

J.A. Thompson et al. / Microporous and Mesoporous Materials 192 (2014) 43 51 51 [7] A.J. Brown, J.R. Johnson, M.E. Lydon, W.J. Koros, C.W. Jones, S. Nair, Angew. Chem. Int. Ed. 51 (2012) 10615 10618. [8] S.R. Venna, M.A. Carreon, J. Am. Chem. Soc. 132 (2010) 76 78. [9] H. Bux, F. Liang, Y. Li, J. Cravillon, M. Wiebcke, J. Caro, J. Am. Chem. Soc. 131 (2009) 16000 16001. [10] B. Zornoza, A. Martinez-Joaristi, P. Serra-Crespo, C. Téllez, J. Coronas, J. Gascon, F. Kapteijn, Chem. Commun. 47 (2011) 9522 9524. [11] T.-H. Bae, J.S. Lee, W. Qiu, W.J. Koros, C.W. Jones, S. Nair, Angew. Chem. Int. Ed. 122 (2010) 9863 9866. [12] R. Adams, C. Carson, J. Ward, R. Tannenbaum, W.J. Koros, Microporous Mesoporous Mater. 131 (2010) 13 20. [13] S. Basu, A. Cano-Odena, I.F.J. Vankelecom, Sep. Purif. Technol. 81 (2011) 31 40. [14] O.G. Nik, X.Y. Chen, S. Kaliaguine, J. Membr. Sci. 413 414 (2012) 48 61. [15] X.-L. Liu, Y.-S. Li, G.-Q. Zhu, Y.-J. Ban, L.-Y. Xu, W.-S. Yang, Angew. Chem. Int. Ed. 50 (2011) 10636 10639. [16] T. Li, Y. Pan, K.-V. Peinemann, Z. Lai, J. Membr. Sci. 425 426 (2013) 235 242. [17] C. Zhang, Y. Dai, J.R. Johnson, O. Karvan, W.J. Koros, J. Membr. Sci. 389 (2012) 34 42. [18] M.J.C. Ordoñez, K.J. Balkus, J.P. Ferraris, I.H. Musselman, J. Membr. Sci. 361 (2010) 28 37. [19] Q. Song, S.K. Nataraj, M.V. Roussenova, J.C. Tan, D.J. Hughes, W. Li, P. Bourgoin, M.A. Alam, A.K. Cheetham, S.A. Al-Muhtaseb, E. Sivaniah, Energy Environ. Sci. 5 (2012) 8359 8369. [20] B. Zornoza, B. Seoane, J.M. Zamaro, C. Téllez, J. Coronas, ChemPhysChem 12 (2011) 2781 2785. [21] X.-C. Huang, Y.-Y. Lin, J.-P. Zhang, X.-M. Chen, Angew. Chem. Int. Ed. 45 (2006) 1557 1559. [22] K.S. Park, Z. Ni, A.P. Côté, J.Y. Choi, R. Huang, F.J. Uribe-Romo, H.K. Chae, M. O Keeffe, O.M. Yaghi, Proc. Natl. Acad. Sci. 103 (2006) 10186 10191. [23] A. Phan, C.J. Doonan, F.J. Uribe-Romo, C.B. Knobler, M. O Keeffe, O.M. Yaghi, Acc. Chem. Res. 43 (2010) 58 67. [24] D. Fairen-Jimenez, S.A. Moggach, M.T. Wharmby, P.A. Wright, S. Parsons, T. Düren, J. Am. Chem. Soc. 133 (2011) 8900 8902. [25] S. Aguado, G. Bergeret, M.P. Titus, V. Moizan, C. Nieto-Draghi, N. Bats, D. Farrusseng, New J. Chem. 35 (2011) 546 550. [26] L. Zhang, Z. Hu, J. Jiang, J. Am. Chem. Soc. 135 (2013) 3722 3728. [27] J. van den Bergh, C. Gücüyener, E.A. Pidko, E.J.M. Hensen, J. Gascon, F. Kapteijn, Chem. Eur. J. 17 (2011) 8832 8840. [28] C. Zhang, R.P. Lively, K. Zhang, J.R. Johnson, O. Karvan, W.J. Koros, J. Phys. Chem. Lett. 3 (2012) 2130 2134. [29] E. Haldoupis, T. Watanabe, S. Nair, D.S. Sholl, ChemPhysChem 13 (2012) 3449 3452. [30] J.A. Thompson, C.R. Blad, N.A. Brunelli, M.E. Lydon, R.P. Lively, C.W. Jones, S. Nair, Chem. Mater. 24 (2012) 1930 1936. [31] J.A. Thompson, N.A. Brunelli, R.P. Lively, J.R. Johnson, C.W. Jones, S. Nair, J. Phys. Chem. C 117 (2013) 8198 8207. [32] J.A. Thompson, K.W. Chapman, W.J. Koros, C.W. Jones, S. Nair, Microporous Mesoporous Mater. 158 (2012) 292 299. [33] T.-S. Chung, L.Y. Jiang, Y. Li, S. Kulprathipanja, Prog. Polym. Sci. 32 (2007) 483 507. [34] D.Q. Vu, W.J. Koros, S.J. Miller, J. Membr. Sci. 211 (2003) 311 334. [35] G. Horváth, K. Kawazoe, J. Chem. Eng. Jpn. 16 (1983) 470 475. [36] K. O Brien, W.J. Koros, T. Barbari, E. Sanders, J. Membr. Sci. 29 (1986) 229 238. [37] D.R. Paul, D.R. Kemp, J. Polym. Sci. Polym. Symp. 41 (1973) 79 93. [38] X.Y. Chen, H. Vinh-Thang, D. Rodrigue, S. Kaliaguine, Ind. Eng. Chem. Res. 51 (2012) 6895 6906. [39] D.T. Coker, B.D. Freeman, G.K. Fleming, AIChE J. 44 (1998) 1289 1302. [40] A. Bos, I.G.M. Pünt, M. Wessling, H. Strathmann, Sep. Purif. Technol. 14 (1998) 27 39. [41] C.F. Macrae, I.J. Bruno, J.A. Chisholm, P.R. Edgington, P. McCabe, E. Pidcock, L. Rodriguez-Monge, R. Taylor, J. van de Streek, P.A. Wood, J. Appl. Cryst. 41 (2008) 466 470. [42] L. Ge, W. Zhou, V. Rudolph, Z. Zhou, J. Mater. Chem. A 1 (2013) 6350 6358. [43] M.E. Lydon, K.A. Unocic, T.-H. Bae, C.W. Jones, S. Nair, J. Phys. Chem. C 116 (2012) 9636 9645. [44] T. Rodenas, M. van Dalen, E. García-Pérez, P. Serra-Crespo, B. Zornoza, F. Kapteijn, J. Gascon, Adv. Funct. Mater. (2012), http://dx.doi.org/10.1002/ adfm.201203462. [45] E. Stavitski, E.A. Pidko, S. Couck, T. Remy, E.J.M. Hensen, B.M. Weckhuysen, J. Denayer, J. Gascon, F. Kapteijn, Langmuir 27 (2011) 3970 3976. [46] G. Yilmaz, S. Keskin, Ind. Eng. Chem. Res. 51 (2012) 14218 14228. [47] H. Bux, C. Chmelik, J.M. van Baten, R. Krishna, J. Caro, Adv. Mater. 22 (2010) 4741 4743. [48] H. Bux, C. Chmelik, R. Krishna, J. Caro, J. Membr. Sci. 369 (2010) 284 289. [49] R.W. Baker, K. Lokhandwala, Ind. Eng. Chem. Res. 47 (2008) 2109 2121. [50] M.D. Donohue, B.S. Minhas, S.Y. Lee, J. Membr. Sci. 42 (1989) 197 214. [51] C. Staudt-Bickel, W.J. Koros, J. Membr. Sci. 155 (1999) 145 154. [52] A.F. Ismail, W. Lorna, Sep. Purif. Technol. 27 (2002) 173 194. [53] W.J. Koros, R.T. Chern, V. Stannett, H.B. Hopfenberg, J. Polym. Sci. Polym. Phys. Ed. 19 (1981) 1513 1530. [54] H.D. Kamaruddin, W.J. Koros, J. Membr. Sci. 135 (1997) 147 159. [55] C. Chmelik, J. van Baten, R. Krishna, J. Membr. Sci. 397 398 (2012) 87 91. [56] S.E. Jee, D.S. Sholl, J. Am. Chem. Soc. 131 (2009) 7896 7904. [57] R.T. Chern, W.J. Koros, P.S. Fedkiw, Ind. Eng. Chem. Process Des. Dev. 24 (1985) 1015 1022. [58] Y. Dai, J.R. Johnson, O. Karvan, D.S. Sholl, W.J. Koros, J. Membr. Sci. 401 402 (2012) 76 82. [59] R.W. Baker, Membrane Technology and Applications, McGraw-Hill, New York, 2000. [60] R.P. Lively, M.E. Dose, L. Xu, J.T. Vaughn, J.R. Johnson, J.A. Thompson, K. Zhang, M.E. Lydon, J.-S. Lee, L. Liu, Z. Hu, O. Karvan, M.J. Realff, W.J. Koros, J. Membr. Sci. 423 424 (2012) 302 313.