Polymers with Cavities Tuned for Fast Selective Transport of Small Molecules and Ions

See allHide authors and affiliations

Science  12 Oct 2007:
Vol. 318, Issue 5848, pp. 254-258
DOI: 10.1126/science.1146744


Within a polymer film, free-volume elements such as pores and channels typically have a wide range of sizes and topologies. This broad range of free-volume element sizes compromises a polymer's ability to perform molecular separations. We demonstrated free-volume structures in dense vitreous polymers that enable outstanding molecular and ionic transport and separation performance that surpasses the limits of conventional polymers. The unusual microstructure in these materials can be systematically tailored by thermally driven segment rearrangement. Free-volume topologies can be tailored by controlling the degree of rearrangement, flexibility of the original chain, and judicious inclusion of small templating molecules. This rational tailoring of free-volume element architecture provides a route for preparing high-performance polymers for molecular-scale separations.

Small-molecule and ion diffusion through cavities (i.e., free-volume elements) in soft organic materials is an inherently subnanoor nanoscopic phenomenon. It has important implications for membrane separation processes in chemicals production as well as energy conversion and storage applications [e.g., pharmaceutical separations (1), organic batteries (2), fuel cells (3), and gas separation (4)]. Transport of small gas molecules through polymers occurs by diffusion through transient free-volume elements or cavities formed by random, thermally stimulated motion of the flexible organic chains. Unlike pore sizes and shapes in rigid microporous inorganic materials such as zeolites (5) and carbon molecular sieve materials (6), cavity sizes and shapes are not uniform in amorphous polymers. The cavity radius (r) of the most selective polymers such as polyimides, polysulfones, and polycarbonates, as measured by positron annihilation lifetime spectroscopy (PALS), is 0.3 nm or less with a broad distribution of cavity sizes, and gas permeability is rather low (7).

Conversely, the most permeable polymer, poly(1-trimethylsilyl-1-propyne) (PTMSP), exhibits an approximately bimodal cavity size distribution centered at around r = 0.3 nm and r = 0.6 to 0.7 nm (8). The high concentration of large cavities and the high connectivity among cavities results in very high permeability for a polymer, but its ability to separate small molecules (kinetic diameter <0.45 nm) is too low to be useful, and the large cavities collapse over time due to physical aging (8). Thus, among known polymers, free-volume element size and distribution play a key role in determining permeability and separation characteristics. However, the broad size range of free-volume elements in such materials precludes the preparation of polymers having both high permeability and high selectivity.

We demonstrate that polymers with an intermediate cavity size, a narrow cavity size distribution, and a shape reminiscent of bottlenecks connecting adjacent chambers, such as those found elegantly in nature in the form of ion channels (9) and aquaporins (10), yield both high permeability and high selectivity. Central to our approach for preparing these intermediate-sized cavities is controlled free-volume element formation through spatial rearrangement of the rigid polymer chain segments in the glassy phase. It is known that a rearrangement, such as intramolecular cyclization, in glassy polymers could lead to changes in polymer structure for gas transport (11). For this purpose, aromatic polymers interconnected with heterocyclic rings (e.g., benzoxazole, benzithiazole, and benzimidazole) are of interest because phenylene-heterocyclic ring units in such materials have a flat, rigid-rod structure with high–torsional energy barriers to rotation between two rings (12). The stiff, rigid ring units in such flat topologies pack efficiently, leaving very small penetrant-accessible free-volume elements. This tight packing is also promoted by intersegmental interactions such as charge-transfer complexes between heteroatoms containing lone electron pairs (e.g., O, S and N) (13). The genesis of these materials was the demand for highly thermally and chemically stable polymers. However, their application as gas separation membranes was frustrated by their lack of solubility in common solvents, which effectively prevents them from being prepared as thin membranes by solvent casting, which is the most widely practiced method for membrane preparation.

We circumvented this fabrication challenge by using postfabrication polymer-modifying reactions (14, 15). Completely aromatic, insoluble, infusible polymers can be prepared from highly soluble precursors by irreversible molecular rearrangement at about 350° to 450°C for aromatic polyimides containing ortho-positioned functional groups (e.g., -OH and -SH) (Fig. 1). Two types of changes in chain structure occur during the rearrangement that alter chain packing: (i) random chain conformations resulting from the formation of meta- and para-linked chains (Fig. 1A); and (ii) relatively flexible, twisting pairs of short flat planes (α and β) that convert to single long flat planes (γ) (Fig. 1B) that are much more rigid than those of the parent moieties [e.g., the torsional angle (φ2) of benzoxazole-phenylene ring is close to 0° at the energy-minimized state because the coplanar conformation is favored due to resonance stabilization]. The use of stiff, rigid chain elements (e.g., benzoxazole-phenylene ring or benzithiazole-phenylene ring) prevents large intrachain, indiscriminant torsional rotation, increases the efficiency of cavity formation, and inhibits rapid collapse of the created cavities. These materials are thermally stable, and the structural rearrangements occurring during this process do not correspond to partial burning (or carbonization) of the underlying polymer structure, a process that has been used in other cases to enhance gas separation properties of polymers.

Fig. 1.

Two major factors contributing to structural change during thermal chain rearrangement of polyimides containing ortho-positioned functional groups (X is O or S). (A) Change of chain conformation–polymer chains consisting of meta- and/or para-linked chain conformations can be created via rearrangement. (B) Spatial relocation due to chain rearrangement in confinement, which may lead to the generation of free-volume elements [α plane, phthalic imide ring; β plane, XH-containing phenylene ring; γ plane, newly created phenylene-heterocyclic ring (if X is O, benzoxazole-phenylene ring; if X is S, benzithiazole-phenylene ring); φ1 and φ2, dihedral angle; θ, tilting angle after transformation].

If managed properly, these changes in chain conformation and topology create well-connected, narrow size distribution free-volume elements (i.e., cavities) appropriate for molecular separations. For example, PALS analysis of a precursor polymer [PIOFG-1, synthesized from 4,4'-(hexafluoroisopropylidene)-diphthalic anhydride (6FDA) and 2,2'-bis(3-amino-4-hydroxylphenyl) hexafluoropropane (bisAPAF) via thermal imidization up to 300°C] and its corresponding thermally rearranged samples (i.e., TR-1-350, TR-1-400, and TR-1-450, respectively) shows that the polymer undergoes microstructural change depending on the extent of rearrangement (16). Thermal degradation of the polymer chains is not observed within the heat-treatment temperature ranges, based on results from thermogravimetric analysis coupled with mass spectroscopy (fig. S1A) and elemental analysis (table S2). Spectroscopic analysis (i.e., Fourier transform–infrared) provides convincing evidence that the conversion from imide to benzoxazole is achieved (fig. S1B).

Figure 2 shows that the cavity radius of PIOFG-1 polymer (which is centered at about 0.28 nm and is very broad) increases to ∼0.4 nm, and the distribution of cavity sizes becomes narrow as the thermal rearrangement temperature increases to 450°C. PALS analysis reveals an increase in o-positronium (o-Ps) lifetime as rearrangement temperature increases from 300° to 450°C. In general, longer o-Ps lifetime indicates larger cavity sizes (17). The o-Ps intensity (%) increases by 700% as thermal rearrangement temperature is increased to 400°C, but decreases above this temperature. Notably, despite increasing o-Ps lifetime, the reduction of o-Ps intensity in the sample treated at 450°C indicates that an increase in mean cavity size is accompanied by a decrease in the number of cavities, suggesting coalescence of smaller cavities to form larger ones. Hourglass-shaped cavities, having narrow neck regions separating much larger adjacent chambers, are consistent with this scenario. To have excellent separation properties, the small neck regions must not be too large relative to the size of the molecules being separated, because large openings enable relatively nonselective flow mechanisms (e.g., Knudsen flow) (7). However, large cavities adjoining the necks will contribute to high rates of molecular transport. The large cavity size of the fully converted sample (TR-1-450) is smaller than that of PTMSP (0.675 nm) but substantially larger than that of common glassy polymers (e.g., 0.286 nm for polysulfone; 0.289 nm for polycarbonate) (17). Similar behavior is observed in other PIOFG samples prepared by a combination of other monomers based on the same methodology.

Fig. 2.

(A) Change of cavity radius (Å) distribution, measured by PALS, of 6FDA + bisAPAF polyimide (PIOFG-1) as a function of thermal treatment temperature. (B) SAXS profiles of PIOFG-1 for all four processing temperatures and fit of the Guinier knee of TR-1-450 polymer (black dotted line). (a) PIOFG-1; (b) TR-1-350; (c) TR-1-400; (d) TR-1-450 (FWHM, full width at half maximum from the o-Ps lifetime τ3 distribution).

Synchrotron small-angle x-ray scattering (SAXS) measurements also indicate structural changes over the q-range 0.1 to 0.5 Å–1 (q = 4πsinθ/λ, where λ is the x-ray wavelength and 2θ is the scattering angle). Specifically, a peak is apparent in the SAXS profiles for samples processed at 400° and 450°C, but not in those processed at lower temperatures (Fig. 2). Furthermore, the peak in the 450°C sample is more pronounced and centered at a lower q than that of the 400°C sample. If we attribute this peak to scattering from cavities, these data suggest that cavities increase in size (i.e., the peak shifts to small q) at higher treatment temperatures.

To explore the separation properties of these polymers (hereafter referred to as TR polymers), we prepared dense membranes (thickness ∼20 to 30 μm) for pure and mixed-gas permeability experiments. Figure 3 shows the gas separation performance of several families of polymers considered for CO2/CH4 separation at 35°C. Such separations are vital in natural gas processing, landfill gas recovery, and enhanced oil recovery (18).

Fig. 3.

(A) Relation between CO2 permeability and CO2/CH4 selectivity of TR polymers (◼) [1, PIOFG-1; 2, TR-1-350; 3, TR-1-400; 4, TR-1-450; 5, HCl-doped TR-1-450; 6, dedoped TR-1-450; 7, HCl-redoped TR-1-450; 8, H3PO4-doped TR-1-450; 9 to 19, other TR polymers prepared at 450°C from homopolyimides and copolyimides containing thermally convertible segment units (16)]. These data were obtained from pure gas experiments at 35°C. Gas separation performance data of polyimides (PI) reported in the literature (◯) (32); other polymers (★) (8, 29, 32) [PET, poly(ethylene terephthalate); PSF, polysulfone; CA, cellulose acetate; PC, polycarbonate; PS, polystyrene; PPO, poly(phenylene oxide); PIM, polymer with intrinsic microporosity; PTMSP, poly(1-trimethylsilyl-1-propyne); CMS, carbon molecular sieve membranes (△) (6, 33)] are included for comparison. The upper bound is from (19), and the dotted line is provided to guide the eye. (B) Effect of CO2 partial pressure on mixed-gas CO2/CH4 selectivity in TR-1-450 at 35°C. Mixed-gas CO2/CH4 feed compositions (in mol% CO2:mol% CH4) were 10:90 (⚫), 50:50 (▢), and 80:20 (♢). Mixed-gas data for a fluorine-containing polyimide (32) and cellulose acetate (22) are included for comparison. The CO2 permeabilities of fluorine-containing polyimide, cellulose acetate, and TR-1-450 at 35°C and pressure = 10 atm are included for comparison. The dotted lines are provided to guide the eye.

TR polymers demonstrate excellent CO2/CH4 separation performance, surpassing the CO2/CH4 separation limitation (i.e., the “upper bound” line in Fig. 3A) (19) of typical polymer membranes (TR polymers also exceed the separation limit of other notable gas pairs such as O2/N2 and H2/N2). Counterintuitively, the CO2 permeability and CO2/CH4 selectivity are both high, in contrast to the behavior of conventional strongly size-sieving polymer membranes, where high CO2/CH4 selectivity invariably leads to low CO2 permeability (20). On the permeability-selectivity map, the separation performance of our polymer membranes is intermediate between the performance of common polymers and carbon molecular sieve membranes. As revealed by PALS and SAXS, the unusual microstructure of TR polymers (i.e., large cavities) provides an explanation for their high gas permeabilities, and the constriction formed by cavity coalescence is presumably responsible for their precise discrimination among gas molecules such as CO2 and CH4. In addition, gas separation results (Fig. 3A) reveal that the cavity size in TR polymers can be tuned by adding small acidic dopants (e.g., HCl and H3PO4) because TR polymers include basic nitrogen atoms (-C=N-) on the heterocyclic rings (e.g., benzoxazole ring). After doping, the CO2 permeability decreases but CO2/CH4 selectivity increases. However, after dedoping, the permeability and selectivity return to their original values, indicating that the cavity size and shape can be tailored.

In CO2/CH4 separation, CO2 typically acts as a plasticizer, swelling the polymer matrix, causing the permeation of CH4 to increase more than that of CO2, which decreases selectivity (21). Glassy polymers such as polyimides and cellulose acetate exhibit substantial decreases in CO2/CH4 selectivity in mixed-gas experiments, particularly at high CO2 fugacity. In contrast, TR polymer membranes do not exhibit substantially reduced selectivity, even at high CO2 concentration (∼80 mol%) and high CO2 fugacity (∼15 atm) (Fig. 3B). The small reduction in selectivity with increasing CO2 fugacity is caused by a stronger decrease in CO2 permeability as compared to CH4 permeability, which is typical for glassy polymer membranes and is due to the effect of competitive sorption in mixtures (22, 23). The CO2 and CH4 permeabilities slightly decrease with increasing CO2 fugacity in all CO2/CH4 mixtures (fig. S4), and there is no evidence of plasticization. That is, TR polymer membranes show excellent resistance to plasticization at CO2 partial pressures as high as 20 atm.

Gas permeabilities of TR polymers are often two orders of magnitude higher than those of the original PIOFG polymers but are still lower than those of PTMSP, the most permeable polymer. However, selectivities for important gas separations (e.g., O2/N2 and CO2/CH4) are much higher than in PTMSP (24) but comparable to or slightly lower than in carbon molecular sieve membranes (25). For TR polymers, the order of permeability is CO2 > H2 > He > O2 > N2 > CH4, similar to that observed in ultrahigh free-volume polymers like PTMSP (24).

The outstanding performance results from largely unique cavity formation caused by random chain conformations during thermal molecular rearrangement. A few comparable studies are found in the literature. Barsema et al. (26) studied commercial polyimide membranes treated at different temperatures (300° to 525°C). They observed that the gas permeability of polyimide membranes treated below the thermal decomposition temperature (<450°C) did not change noticeably, but was slightly reduced due to polymer densification. At the decomposition temperature, the treated membranes exhibited a small increase in gas permeability.

A possible reason why these polymers have unique cavity sizes and shapes might be related to the role played by CO2 molecules escaping from the original polymer matrix. Therefore, we designed a carboxylic acid group–containing polyimide film and performed the same thermal treatment. Here, the decarboxylation also occurs in a similar temperature range (T = 400° to 500°C). However, no outstanding change in gas permeability was apparent relative to that of the parent polyimide. From our model study, the evolution of CO2 is not a decisive factor in the formation of gas-accessible free volume or cavities (fig. S5).

The polymers under investigation exhibit pseudo-microporous characteristics that can be probed by nitrogen adsorption/desorption, a technique usually applied to inorganic microporous materials rather than polymers. Conventional dense polymers are “nonporous” in that free-volume elements do not span the sample, so Brunauer-Emmett-Teller (BET) analysis is rarely used to characterize them (27). Nitrogen adsorption was used to study two PIOFG polymers [PIOFG-1 and PIOFG-2 (synthesized from 6FDA and 2,5-diamino-1,4-benzenedithiol (DABT)] and their thermally rearranged analogs at 450°C (TR-1-450 and TR-2-450) (Fig. 4). The PIOFG polymers exhibit an adsorption/desorption isotherm previously observed in glassy polymers (27). The nitrogen adsorption/desorption isotherms of TR-1-450 and TR-2-450 are of the irreversible Type I form with hysteresis. The BET surface areas are markedly large for polymers, 510 m2 g–1 (TR-1-450) and 410 m2g–1 (TR-2-450), which indicates the presence of substantial amounts of free volume. The hysteresis loops for the TR polymers do not correspond to any of the IUPAC isotherms (28), but they are also observed in polymers of intrinsic microporosity (29). Materials showing similar isotherms are typically understood to possess “throat and cavity” type microporosity characteristic of activated carbons (30). These results further support the hypothesis from PALS of hourglass-shaped cavities.

Fig. 4.

(A) Nitrogen adsorption/desorption isotherms at –195°C for (a) PIOFG-1, (b) PIOFG-2, (c) TR-1-450, and (d) TR-2-450. p/p0 is the ratio of gas pressure (p) to saturation pressure (p0), with p0 = 746 torr. (B) Apparent cavity size distributions of (a) TR-1-450, measured by BET, and (b) PTMSP (included for comparison).

There are two advantages to the materials described in this work. First, the original polyimides are soluble in common solvents; that is, they can be prepared in the form of hollow fibers and then continuously exposed to heat treatment because these TR polymers produce tough, ductile, robust films rather than brittle, fragile specimens such as zeolite or carbon membranes (table S3). This feature markedly enhances their potential utility and ultimate reduction to practice. Second, it is much easier and simpler to coat these polymers without any defects or cracks onto microporous ceramic support membranes than to coat zeolite, silica, and carbon membranes onto such supports. Recently, we observed that the separation performance of thin-layer coated composite membranes is comparable to that of thick, dense membranes and these membranes do not show permeability decay with time due to physical aging (fig. S5), as do high–free-volume glassy polymer membranes. These polymers with tailored porosity can also be applied as fuel cell membranes. When doped with acid molecules, these polymers exhibit high proton conductivity. For example, the proton conductivity of a H3PO4-doped TR-1-450 membrane reaches 0.15 S cm–1 at 130°C and low relative humidity (<30%). This value is higher than that of polybenzimidazole (PBI) (<0.1 S cm–1 at 200°C) (31), the most attractive proton-conducting polymer for high-temperature fuel cells. Currently we believe these polymers sequester water, bound acid, and free acid molecules in the cavity structure, and this phenomenon is responsible for the high conductivity. Most of all, the greatest benefit of these polymers is the ability to tune the cavity size and distribution for specific gas applications by using various templating molecules and heat treatments, with one starting material.

Supporting Online Material

Materials and Methods

Figs. S1 to S6

Tables S1 to S3


References and Notes

View Abstract

Navigate This Article