Swelling and Morphology of the Skin Layer of Polyamide Composite ...

Environ. Sci. Technol. 2004, 38, 3168-3175Swelling and Morphology of theSkin Layer of Polyamide CompositeMembranes: An Atomic ForceMicroscopy StudyVIATCHESLAV FREGER*Laboratory for Desalination and Water Technology Research,Institutes for Applied Research, Ben-Gurion University of theNegev, P.O. Box 635, Beer-Sheva 84105, IsraelThe paper introduces a new methodology for studyingpolyamide composite membranes for reverse osmosis (RO)and nanofiltration (NF) in liquid environments. Themethodology is based on atomic force microscopy of theactive layer, which had been separated from the support andplaced on a solid substrate. The approach was employedto determine the thickness, interfacial morphology, anddimensional changes in solution (swelling) of polyamidefilms. The face (active) and back (facing the support) surfacesof the RO films appeared morphologically similar, inagreement with the recently proposed model of skinformation. Measured thickness and swelling data inconjunction with the intrinsic permeability of the membranessuggest that the selective barrier in RO membraneconstitutes only a fraction of the polyamide skin, whereasNF membranes behave as nearly uniform films. For NFmembranes, there was reasonable correlation between thechanges in the swelling and in the permeability of themembrane and the salinity and pH of the feed.IntroductionPolyamide composite membranes currently dominate thefield of reverse osmosis (RO) and nanofiltration (NF) (1, 2).Membrane desalination based on RO has today become thekey technology for the production of freshwater from salinesources. As opposed to RO membranes, which effectivelyreject all inorganic solutes, polyamide membranes for NFare capable of selective removal of certain solutes, whileallowing the passage of others (2, 3).RO and NF polyamide membranes are made up of anultrathin (

FIGURE 1. Dark field optical micrograph (135 × 102 µm) of stripsof polyamide (SWC1) on a silicon wafer. Bare silicon appears black.Local film thickness presumably increases in going from blue tored.surface by adhesion. For complete removal of the polysulfone,the substrate with the attached membrane was thoroughlyrinsed in clean DMF and dichloromethane. Once cleaned ofpolysulfone, the films prepared from RO membranes usuallyacquired a bright color due to interference of reflected light.The thinner NF films were colorless, yet easily discernibledue to a more diffuse light reflection as opposed to thespecular reflection of the bare substrate.The choice of DMF for dissolving off the polysulfone wasbased on three factors. First, DMF is known to dissolvepolysulfone, but the dissolution is relatively slow and doesnot lead to fast and poorly controlled nonhomogeneousdeformations. Second, the solubility parameter of DMF (24.8Mpa 1/2 ) is fairly close to the value suggested by Aharony (10)for fully aromatic polyamide networks (23 Mpa 1/2 ), whichmeans that DMF should be a good swelling agent for the skinlayer. Swelling and softening (without dissolution due to thehigh cross-linking) should help the polyamide to accommodateirregular stresses occurring during rapid deformationof polysulfone upon dissolution. Third, polar DMF wets boththe glass or native (oxidized) Si surface and the membrane.When placed on the membrane, a drop of DMF spreads anddrags the edges of the film in the direction that preventsfolding and wrinkling. This was particularly important duringthe very first moments of dissolution of polysulfone. It is nottotally impossible that some fraction of the polyamide couldbe dissolved in DMF. However, comparison of the thicknessof the free films measured by AFM with TEM images (seebelow) and our preliminary spectroscopic data obtained usingattenuated total reflection IR spectroscopy (11) indicate thatmost of the polyamide did not dissolve. Furthermore, evenif some dissolution had occurred, it would have affected onlythe loosest and very lightly cross-linked parts of polyamidethat do not contribute to separation and are actuallyundesirable for our analysis.Prior to AFM imaging, random and nearly parallel multiplescratches were made in the film by gently scratching it witha sharp metal needle to form narrow strips of polyamide.The optical micrograph of a typical scratched sample takenwith an optical microscope (Zeiss Axioscop 2 Math equippedwith a digital camera) is shown in Figure 1. The hard nativeSi or glass surface required only minimal care to preventscratching of the substrate. Perfect specular reflection of thebare substrate areas in the dark field optical images (uniformblack background in Figure 1) did indeed indicate the absenceof scratches. The use of strips of film gave the advantage ofa flat solid surface on both sides of the film, which madepossible reliable measurements of the strip thickness.AFM Characterization and Analysis. AFM images weretaken on a Dimension 3100 (Digital Instruments, Santa-Barbara, CA) and Autoprobe CPR (Park Scientific Instruments,Palo Alto, CA) instruments using the manufacturers’ fluidcells that allowed imaging of liquid-flooded samples. Thefollowing silicon nitride tips were used: DNP (DI-Veeco)with a nominal spring of constant 0.32 N/m and Parkmicrolevers (0.1 N/m), respectively. The contact mode wasemployed, which makes use of “hard core” repulsiveinteractions between the tip and the surface known to be ofvery short range (,1 nm). To ensure that AFM did not operatein the long-range repulsive electrostatic interaction associatedwith the double electric layer, we made use of the data ofBowen et al. (12, 13), who reported force-distance curvesfor the interaction of a SiO 2 sphere and polyamide membranesnormalized to the sphere radius. Although the shapeand material of the tip in our experiments differed from theSiO 2 sphere, we used these data and the nominal tip diameterof 20-60 nm to roughly estimate the maximal electrostaticforce; this was found to be

FIGURE 2. AFM image (5 × 5 µm) of a strip of the skin of a NF-270 membrane under dry conditions and the corresponding height histogram.FIGURE 3. AFM image (22 × 18 µm) of a strip of the skin of a SWC1 membrane under water and the corresponding height histogram.The inset shows the peak corresponding to the area covered by the film in more detail. Note the small third peak, at about three timesthe dominant height, corresponding to the folded area along the right edge of the film. The presence of the fold does not interfere withthe determination of the dominant height.images and the swelling. This was accomplished by usingheight histograms. The height distribution was bimodal, withtwo clear and well-determined maxima corresponding tothe bare substrate and film-occupied areas (Figures 2 and 3).The height difference between the maxima served as ameasure of the dominant film thickness. This parameter isrelatively insensitive to the presence of excessively swollenspots, folds, small parts of the film detached from thesubstrate or lifted by the moving tip at the film edge, dustparticles, and other artifacts. The degree of swelling on thewet basis (S) was found by relating the dominant thicknessesof the wet (d wet) and dry (d dry) samples measured for thesame region as follows:S ) (d wet - d dry )/d wet × 100% (1)It was also possible to use in eq 1 the average thicknessto calculate swelling. The average thickness was taken as thedifference between the averaged heights of the film-coveredand the bare substrate areas. This was obtained by averagingover the respective parts (peaks) of the histogram. Thischaracteristic could help detect the enhanced swelling ofthe more protruding parts of the film corresponding to themore elevated regions that show up as “tails” in the heighthistogram. When large artifacts were present, it was possibleto eliminate their influence by excluding highly elevated areas(i.e., those exceeding three to four times the dominant height).Results and DiscussionNF Membranes. Typical flattened images of the polyamidelayer of NF composites and the corresponding heighthistograms are shown in Figures 2 and 4. The data andcalculations of swelling in water are summarized in Table 1.As was expected, there were some deviations betweensamples, as is typical for synthetic membranes, but thedifferences were not dramatic. The swelling was fairly similarfor NF-200 and NF-270 membranes, the main differencebetween the two being the thickness. The dry thickness ofthe NF-270 membrane was determined to be about 14 nm,while the NF-200 membrane was about twice as thick (32nm), findings which are fully consistent with the fact theformer membrane is twice as permeable as the latter (Table1). Both membranes swelled in water to contain around 20%water on wet volume basis. The swelling of the film was nearlyunidimensional, as was concluded by examining the changesof the film dimensions in the normal and lateral directions,the latter being virtually undetectable. This mode of swelling,which apparently results from the anchoring of the mem-3170 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 11, 2004

FIGURE 4. AFM images (8 × 8 µm) of a NF-200 film: (a) under dry conditions, (b) in deionized water, (c) in 15% NaCl solution at pH 3,and (d) 15% NaCl solution at pH 10. It is evident that in panel d the right edge of the strip had been partly detached by the tip. Fordetermination of the average thickness, the damaged area was excluded (the exclusion negligibly affected the determination of thedominant thickness).TABLE 1. Thickness, Swelling in Water, and Actual and Superficial Intrinsic Permeability of Four RO and NF Polyamide CompositeMembranesthickness (nm)dominantaverageswelling(%, wet basis)La pmembrane dry swollen dry swollen dominant average (L m -2 h -1 bar -1 )NF-270 13.9 18.0 14.1 19.0 22.8 25.813.7 16.0 14.5 17.7 14.4 18.113NF-200 30.2 38.1 32.2 44.0 20.7 26.8 7.1SWC1 74 84 108 115 11.9 6.1288 296 283 295 2.7 4.1ESPA1 back surface 146 157 162 165 7.0 1.8333 355 357 386 6.2 7.5ESPA1 face surface 240 252 283 296 4.8 4.4aDeduced from the manufacturers’ data (15). b Dominant thickness in swollen state used.10 19 L pδ b(m 2 s -1 Pa -1 )6.8 ( ‘1.20.39 2.1 ( 1.64.9 35 ( 14brane on the substrate by adhesion forces, closely correspondsto the actual conditions experienced by a workingmembrane, which is tightly held by the support.One of the goals of the work was to obtain an insight intothe nonuniformity across the skin. Our new model (4)proposed that the skin has a sandwiched structure in whichVOL. 38, NO. 11, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 3171

FIGURE 5. TEM images of uranyl-stained NF-200 (left) and ESPA1 (right) membranes. Uranyl stains the negatively charged (carboxyl-rich)polyamide, leaving the positively charged (carboxyl-free) polyamide unstained. The boundary between the dark and bright-colored polyamidescorresponds to the densest barrier. Adapted from ref 5.the central part (middle plane) is the densest, whereas towardthe outer surfaces (face and back) the polymer grows looser.This conjecture was in good agreement with TEM observationsfor RO polyamides, whereas for NF membranes someambiguity remained, suggesting that piperazine-based skinmight be more uniform than RO composites (5) (Figure 5).The present work offers another way to address this point,namely, to look at the difference between the swelling ratescalculated using the dominant and average thicknesses. Theswelling deduced from the average thickness was consistentlylarger than the swelling deduced from the dominant thickness(Table 1). This suggests that protrusions from the surface(outmost parts) are looser and that their swelling is strongerthan the film on average. Protruding areas with enhancedswelling, contributing mainly to the average thickness andmuch less so to the dominant thickness, can also be directlyobserved in Figure 4. Nevertheless, they occupy a ratherlimited area, while most of the skin looks rather smooth anduniform both before and after swelling. This suggests thatthe density nonuniformities across the skin, while definitelyexisting, are fairly moderate, in agreement with the TEMobservations.RO Membranes. In the determination of the swelling ofRO films by the present method, the accuracy of themeasurements became the critical issue. Although thethickness of each sample (d wet or d dry) could be determinedwith reasonable precision (a few percent), swelling measurementsrequired a higher precision to determine the smalldifference (d wet - d dry) between the thickness in the dry andswollen states (eq 1). This situation was further complicatedby the statistical noise coming from the rough morphologyof the samples, which made it difficult to accurately locatethe maxima in the histograms. This statistical noise is clearlyevident in the histograms in Figure 3. Smoothing of thehistograms or decreasing the number of subdivisions (tomake location of a maximum less ambiguous) reduced theprecision to the level of the swelling rate itself. This was alsomanifested in the absence of any systematic differencebetween the swelling based on average thickness and thatbased on the dominant thickness (Table 1). In addition, thedry thickness of the RO samples varied much more significantlythan that of the NF samples. For this reason, the resultsshould be viewed as semiquantitative at best. Nevertheless,the data presented in Table 1 indicate that the swelling ofthe analyzed RO composites (SWC1 and ESPA1) is unlikelyto significantly exceed 10% (most probably being of the orderof 5%), which is appreciably lower than that of NF membranes.The low swelling agrees well with the fact that fullyaromatic networks are more rigid (16) and presumably moreregularly packed (17) than the semi-aromatic network of NFcomposites.Arthur (18) measured the mass sorption of water vaporby a powder polyamide prepared from the same monomersas RO membranes, which he presumed would represent agenuine membrane. He reported that, upon equilibrationwith water vapor of 95-100% humidity, weight gain increasedfrom about 20% after a short exposure to 28.4% after 7 daysand that no free (freezable) water was detected by differentialscanning calorimetry (18). This gain is much larger than ourAFM estimates of the swelling rate in pure water (Table 1).Typically, during our AFM scans, samples were exposed towater for less than 1 h and over this time no noticeableincrease of swelling was observed in repeated scans. For thisreason this time was considered sufficient. As compared withArthur’s experiments, the shorter time of our experimentswas presumably set off by the small thickness of themembrane and direct contact with liquid, which is knownto largely enhance the rate of sorption as compared to vaporsorption. To explain the large discrepancy between ourfindings and those of Arthur, we could only presume that atleast part of the measured water uptake in his experimentscould have been due to capillary condensation in pores smallenough to suppress freezing. The effect of considerablesupercooling of water at distances of a few nanometers (atypical size of the voids in a RO skin) from a confining interfaceis well-known (19) and could make the water of swelling andcapillary condensation indistinguishable. In contrast, ourmethod detects only the dimensional changes caused byswelling (i.e., mixing of polymer and solvent at the molecularlevel rather than pore filling) and is thus less prone to artifactsof this kind. In addition, the present method is not limitedto pure water, as are vapor sorption experiments, andtherefore offers the possibility of observing the effect ofsalinity and pH on the swelling response of polyamide filmsin direct contact with a feed.As already mentioned, the film is supposedly densestapproximately in the middle of the cross-section and getslooser toward the two interfaces (face and back). The complexnature of the interfacial polymerization process leads to thedevelopment of a rough pattern in the polymer (5, 20). Suchpatterns have routinely been observed in AFM scans of theactive surface (face) of the polyamide composites (21-24)3172 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 11, 2004

FIGURE 6. AFM images of the face and back surfaces of ESPA1 and SWC1 films. All images are 5 µm × 5 µm: (a) active surface of originalESPA1, (b) face side of a film prepared from ESPA1, (c) back side of a film prepared from ESPA1, (d) active surface of original SWC1,and (e) back side of a film prepared from SWC1.and may be considered as an intrinsic property of this typeof membrane. A remarkable finding was that the morphologyof the back surface of the skin (originally facing the support)is strikingly similar to that of the active surface.To confirm that the morphology does not change duringisolation of the skin, we prepared a sample in which the skinof an ESPA1 membrane was prepared on the substrate withthe active surface facing down; thereafter, a gentle flow ofDMF over the specimen served to pull the free film over sothat the active surface finally faced up. Characterization ofthis sample gave the same results as samples prepared withtheir face down, and its swelling was consistent with that ofother RO samples (Table 1). We could not detect anymorphological difference between the original ESPA1 membraneand the face-up film (Figure 6a,b), which both showedthe typical rough pattern resembling a large number of fusedparticles with sharp protrusions characteristic of the ESPAmembranes (5, 25, 26). The finite tip dimensions convert theprotrusions to characteristic pyramidal shapes observed inboth images, but their actual shape may be seen in Figure5. More important, however, is the fact that the back andface surfaces of the skin show similar morphologies (Figure6c). Admittedly, no sharp protrusions are present at the backsurface, yet the typical “fused particles” pattern is clearlyseen. The resemblance is even closer for the SWC1 membrane,for which the AFM images of the active surface of anoriginal membrane and the back surface of the film arevirtually indistinguishable (Figure 6d,e). This morphologicalsymmetry between the face and back surfaces of the skinpredicted by our new model of skin formation (4, 5) couldbe viewed as an additional indication of its correctness.However, it is still not clear what effect the support hason the formation of the above-described patterns on theback side of the film and why these patterns are observedas if no support is present. It may be assumed that the loosestparts of the polyamide connecting the skin to the supportcould have been destroyed during dissolution of polysulfonein organic solvents and washed away. As discussed above,if this were true, such loose parts would actually be undesiredin our analysis. Another explanation is that we could haveunintentionally introduced some bias to our results bychoosing strips free of major defects or peculiarities. Opticalmicrographs of the films (Figure 1) showed that most of theback surface of the films was densely covered with irregularrough bumps or bulges. They are easily detected, since theychange the thickness and hence of the color of the film, inthe places where they are present. Presumably, these bumpsfill large cavities on the support surface, like those seen inFigure 5, and it would be useless to scan them. To obtainmeaningful AFM scans, we usually chose more uniform areas,free of bumps, presumably where the support happened tobe smoother. In those places, the support could indeed havenegligible effect, as it would have to be covered with themonomer solution and formation of dense polymer wouldtake place at some distance from it, as classical works oninterfacial polymerization suggest (27, 28).Relationship between Swelling and Permeability of theMembrane. We would like to propose yet another way oflooking at the nonuniformity of the skin by analyzing thehydraulic permeability based on swelling and thickness data.This analysis relies on the observation that the dependenceof the intrinsic hydraulic permeability of many polymers ontheir swelling (in terms of equilibrium water volume fraction)follows a general trend: The intrinsic hydraulic permeabilityL pδ is the product of the thickness δ of a film and its hydraulicpermeability L p, the latter being defined from the relationshipJ w ) L p∆P, which connects the pure water flux through thefilm (membrane) J w to the applied trans-membrane pressure∆P (7). For a uniformly swollen homogeneous film L pδ isessentially a property of the material and is independent ofthe film thickness (7). Since polyamides interact with watermostly via hydrogen bonding (29, 30)sion solvation playinga relatively minor role due to the low content of the fixedcharges (about 2 orders of magnitude smaller than in ionicpolymers (4, 31, 32)sit would be better to compare thebehavior of polyamides with hydrogen-bonding hydrophilicpolymers in which the nature of polymer-water interactionis similar.VOL. 38, NO. 11, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 3173

FIGURE 7. Comparison of Meares’ data on the intrinsic permeabilityof various nonionic hydrophilic polymers (7) with the permeabilityof NF and RO membranes. The error bars show standard deviations.Figure 7 presents the results compiled by Meares (7) onthe permeability of homogeneous films of various hydrophilicnonionic polymers as a function of equilibrium watervolume fraction. Despite the scattering, the results do indeedfollow a trend. The existence of a “universal” trend canbe rationalized by assuming that it represents a purelygeometric random restriction imposed by the matrix on theindividual or collective motion of diluent molecules. Thisassumption presumably holds true as long as (a) the motionof the solvent molecules forming random “percolationchannels” inside the effectively immobile matrix remainsessentially similar to the one in a bulk solvent, being restrictedsolely by the channel walls, and (b) the effective pore diameterof the above-mentioned channels at a given equilibrium watervolume fraction is weakly dependent on the chemicalstructure of the polymer. Obviously, these conjectures arehighly questionable, but the data presented in Figure 7 suggestthey are approximately correct.Let us now examine where the permeability of thepolyamides of the NF and RO membranes are located in theplot in Figure 7. We can use the permeabilities reported bythe manufacturer for dilute feeds (14), which also agree wellwith our own measurements of water flux in laboratory cells,and combine them with our thickness and swelling measurements.The values and the standard deviations of theparameter L pδ for the membranes are shown in Table 1.Note that since the NF membranes showed similar properties,we combined them into a single data set. We can see fromFigure 7 that NF membranes fit the general trend well, whichsuggests that their behavior is not far from that of ahomogeneous film. This observation once again suggeststhat the nonuniformity of NF membranes is not particularlystrong, even though it might indeed be present. Theassumption of the skin as a uniform layer, though notaccurate, is likely to be sufficiently adequate for practicalconsiderations.FIGURE 8. Effect of the salinity and pH on the flux of NF-200membrane. Trans-membrane pressure 10 bar. Adapted fromref 33.The data for the SWC1 RO membrane are visibly higherthan the general trend, and for the ESPA1 membrane a largedeparture is obvious. We believe that this departure is anindication of the significant nonuniformity of the RO membranes:the actual thickness of the dense barrier (whichshould be multiplied by the permeability coefficient L p togive the intrinsic permeability) is much smaller than thesuperficial overall thickness of the skin. A “uniform skin”would therefore be a very poor model for RO membranes.It should be stressed, however, that the present data are stillapproximate and that more experimental work is requiredto verify these conclusions.Dependence of the Permeability of Polyamide on pHand Salinity. In a recent work on the separation of chargedorganics from highly saline brines with an NF-200 membrane,we encountered a problem of substantial decrease of theflux with increasing feed salinity (33). The salt rejection waslow under these conditions, and corrections due to theeffective reduction of the trans-membrane pressure by thedifference in osmotic pressure were insufficient to explainthis effect. Moreover, salinity also increased the sensitivityof flux to pH. The flux of NF membranes for dilute feedsolutions is known to depend only slightly on pH (see Figure8). In contrast, a change of pH of brines containing 9% saltfrom acidic to basic caused a 2-fold increase in the flux (Figure8). It could be argued that concentration polarization couldalso contribute to the flux reduction, but it should not changethe general trend.It was postulated (33) that the reduction in flux in brinesresulted from a change in swelling of the skin in concentratedsolutions. The present method allowed direct testing of thispremise. To enhance the effect, we chose to use 15% saltsolutionssrather than the 9% brine used in permeationTABLE 2. Swelling and Flux of NF-200 Membrane in Different Feed Solutionsthickness (nm)swelling (%, wet basis)solution dominant average dominant averagemeasured flux(L m -2 h -1 )(∆P ) 10 bar)dry film 30.2 32.215% NaCl, pH 3 35.8 39.4 15.6 18.3 10 (9% NaCl, pH 3)15% NaCl, pH 10 a 37.1 42.3 18.6 23.9 23 (9% NaCl, pH 9)pure (DI) water, pH 7 38.1 44.0 b 20.7 26.8 69 (water, pH 7)aThe area at the right edge of the strip lifted by the tip was excluded (Figure 4d). b Areas higher than 100 nm were excluded.3174 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 11, 2004

experimentssand a higher pH value (10) on the basic sideof the pH range. Figure 4 shows the images of the sameNF-200 film under four different conditions, and Table 2summarizes the calculations of swelling.It may be seen that, irrespective of the way the swellingis calculated (dominant or average), the increase of swellingin the order acidic brine-basic brine-pure water is readilyevident. Enhanced swelling of the more elevated looser partsmakes this increase clearly visible in the images shown inFigure 4. We see that changes in swelling and flux with salinityand pH are in close correspondence (Table 2). Obviously,the permeability changes more markedly, as expected fromthe steep rise of permeability with water content in Figure7 (the dependence is close to a power law with an exponentequal to about 5). These data seem to support the connectionbetween the swelling and the flux of NF membranes. Finally,it is worth mentioning that we were unable to detect anysignificant change of swelling in similar experiments withRO membranes, which seems be another indication of theexceptionally high rigidity of the fully aromatic RO polyamide.AcknowledgmentsThe kind help of Mr. Juergen Jopp and Ms. Roxana Golan ofthe Minerva Center at BGU in performing AFM measurementsand interpreting the results is gratefully acknowledged.The author thanks Prof. Ora Kedem, Dr. Sofia Belfer, and Dr.Charles Linder for stimulating discussions and valuablecomments.Literature Cited(1) Cadotte, J. E.; King, R. S.; Majerle, R. J.; Petersen, R. J. J. Macromol.Sci. Chem. 1981, A15, 727.(2) Petersen, R. J. J. Membr. Sci. 1993, 83, 81.(3) Linder, C.; Kedem, O. In Nanofiltration Principles and Applications;Schaefer, A. I., Fane, A. G., Waite, T. D., Eds.; Elsevier:Amsterdam, 2002.(4) Freger, V.; Srebnik, S. J. Appl. Polym. Sci. 2003, 88, 1162.(5) Freger, V. Langmuir 2003, 19, 4791(6) Crank, J.; Park, G. S. Diffusion in Polymers; Academic Press:London, 1975.(7) Meares, P. Philos. Trans. R. Soc. London, B 1977, 278, 113.(8) Muhr, A. H.; Blanshard, J. M. V. Polymer 1982, 23, 1012.(9) Mulder, M. Basic Principles of Membrane Technology; KluwerAcademic Publishers: Dordrecht, The Netherlands, 1991.(10) Aharony, S. M. J. Appl. Polym. Sci. 1992, 45, 813.(11) Wasserman, A.; Oren, Y.; Freger, V. Manuscript in preparation.(12) Bowen, W. R.; Doneva, T. A. J. Colloid Interface Sci. 2000, 229,544.(13) Bowen, W. R.; Doneva, T. A.; Stoton, J. A. G. Colloids Surf. A2002, 201, 73.(14) Nanotec Electronica. Homepage; www.nanotec.es. AccessedDecember 11, 2003).(15) The Dow Chemical Company. www.dow.com/liquidseps; Hydranautics,A Nitto Denko Company. www.membranes.com.(16) Aharoni, S. M.; Edwards, S. F. Adv. Polym. Sci. 1994, 118, 1.(17) Hirose, M.; Minamizaki, Y.; Kamiyama, Y. J. Membr. Sci. 1997,123, 151.(18) Arthur, S. D. J. Membr. Sci. 1989, 46, 243.(19) Clifford, J. In Water in Disperse Systems; Franks, F., Ed.; Plenum:New York and London, 1975; pp 75-132.(20) Sundet S. A. J. Membr. Sci. 1993, 76, 175.(21) Zhu, X.; Elimelech, M. Environ. Sci. Technol. 1997, 31, 3654.(22) Kwak, S. Y.; Ihm, D. W. J. Membr. Sci. 1999, 158, 143.(23) Bowen, W. R.; Doneva, T. A.; Stoton, J. A. G. Colloids Surf. A2002, 201, 73.(24) Vrijenhoek, E. M.; Hong, S.; Elimelech, M. J. Membr. Sci. 2001,188, 115.(25) Kwak, S.-Y.; Jung, S. G.; Yoon, Y. S. Environ. Sci. Technol. 2001,35, 4334.(26) Hirose, M.; Ito, H.; Masatoshi, S.; Tanaka, K. U.S. Patent5 614 099, 1997.(27) Morgan, P. W. Condensation Polymers: By Interfacial andSolution Methods; Interscience: New York, 1965(28) Nikonov, V. Z.; Savinov, V. M. In Interfacial Synthesis, Vol. 2;Millich, F.; Carraher, C. R., Jr., Eds.; Marcel Dekker: New York,1977.(29) Starkweather, H. W. In Water in Polymers; Rowland, S. P., Ed.;ACS Symposium Series 127; American Chemical Society: Washington,DC, 1980.(30) Loo, L. S.; Cohen, R. E.; Gleason K. K. Macromolecules 1998, 31,8907.(31) Childress, A. E.; Elimelech, M. Environ. Sci. Technol. 2000, 34,3710.(32) Schaep, J.; Vandecasteele, C. J. Membr. Sci. 2001, 188, 129.(33) Freger, V.; Arnot, T. C.; Howell, J. A. J. Membr. Sci. 2000, 178,185.Received for review July 24, 2003. Revised manuscript receivedDecember 15, 2003. Accepted March 8, 2004.ES034815UVOL. 38, NO. 11, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 3175