You are on page 1of 13

Diffusion-Controlled Formation of Porous Structures in Ternary Polymer Systems

CLAUDE COHEN, School of Chemical Engineering, Cornell University, Ithaca, New York 14853, G. B. TANNY, Department of Plastics Research, T h e Weizmann Institute of Science, Rehovot, Israel, and STEPHEN PRAGER, Department of Chemistry, University .f Minnesota, Minneapolis, Minnesota 55455
Synopsis
We propose a theory for the appearance of two-phase structures during the formation of polymer membranes from a casting solution immersed in a coagulant bath. Our model is based on diffusion induced phase separation at the spinodal in the ternary nonsolvent-solvent-polymer system. A simplified treatment of the interdiffusion process by the diffusion layer method permits the formulation of criteria for the formation of two-phase structures in the course of the solvent-coagulant exchange. Our criteria are expressed in terms of the composition dependence of the chemical potentials in the stable and metastable region of the ternary phase diagram. Comparison with experimental results shows qualitative similarities with theoretical predictions.

INTRODUCTION
When a solution of a polymer in a solvent is cast as a film on a supporting there are, depending surface and then immersed in a bath of a nonsolvent (Fig. l), on the recipe, three possible outcomes as the solvent diffuses into the bath and is replaced to a greater or lesser extent by nonsolvent: (i) a uniform film containing little void space and having a density close to that of bulk polymers; (ii) a porous material possessing a considerable void fraction; or (iii) formation of a porous structure as in (ii), but with a thin dense layer a t the polymer solutionhonsolvent interface. Case (iii) is of particular interest in the preparation of membranes for desalination and other separation processes; membranes with this type of structure are often referred to as asymmetric in the reverse osmosis literature. Although the theory of equilibrium phase separation in polymer solutions1-3 is well developed, only recently has attention been paid to the structures which may be generated in a process such as the casting of a perm-selective membrane.3-11 In qualitative terms, two distinct sets of factors have been recognized as playing an important role in the structures obtained by this process: (a) the equilibrium thermodynamic properties of the three component system nonsolvent-s~lvent-polymer,~-~ (b) the magnitude of the overall material transfer and between fl and bath during the coagulation pro~ess.~-ll im Anderson and Ullman have suggested that the formation of the thin dense layer at the interface in case (iii) could be understood in terms of the high concentration gradients produced a t the surface by a rapid drop in the diffusion coefficient at high-polymer concentrations. However, their calculations were made only for a binary solventpolymer system, where the solvent leaves by evaporation and no phase separation
Journal of Polymer Science: Polymer Physics Edition, Vol. 17,477-489 (1979) 01979 John Wiley & Sons, Inc. 0098-1273/79/0017-0477$01.OO

478

COHEN, TANNY, AND PRAGER


Non-solvent bath
JiL

JzI

2.0

surface

Diffusion layer

- - - -- --- --- - - - z = E ( t )
Casting solution Support

Fig. 1. Schematic representation of the cast polymer solution layer. At time t = 0 the whole layer is homogeneous and its composition is that of the initial casting solution. At time t a propagating diffusion layer bounded on one side by the nonsolvent bath and on the other by the casting solution has developed.

occurs. On the other hand, studies of phase separation kinetics are generally carried out for macroscopically homogeneous solutions with no overall concentration gradients. Clearly cases (ii) and (iii) involve the diffusion controlled evolution of a two-phase structure, with solvent and nonsolvent fluxes perpendicular to the film surface playing a dominant role. The intention of the present paper is to provide a first step toward a general theory which combines both the thermodynamic and the diffusion aspects of the problem, and which can give some insights into the structures likely to develop for a given system. Our approach resembles that taken by Kirkaldy and Brown12 and by Ruschak and Miller13in treating, respectively, the formation of metal joints and microemulsions, but differs from the work of these authors in allowing deviations from local equilibrium in regard to phase separations.

DESCRIPTION OF THE MODEL


Figure 1 represents the situation encountered in casting flat sheet (or large diameter tubular) membranes. Upon immersion in the nonsolvent bath, the surface layer of the cast film equilibrates with the bath in the sense that the chemical potential of nonsolvent (and solvent, if present in the elution bath) must be the same in the surface layer as in the bath. After a time t, nonsolvent 1and solvent 2 will have diffused in the directions shown in Figure 1,forming a diffusion layer within which the volume fractions 4 and 4 2 of nonsolvent and sol1 vent vary from their surface values &), 49) their values in the casting solution to 4Io, cbp)(in many instances cpp)and +{') are zero). We will treat the flow of solvent and nonsolvent in the membrane as one-dimensional diffusion processes, and will assume that: (I) the coagulation bath is well stirred, so that the exiting solvent does not remain in the vicinity of the interface, and the composition of coagulant a t the film boundary is a t all times that of the original bath. A t its surface the film is considered to be a t equilibrium with the coagulant mixture. Assumption (I) may in actual practice be modified by the presence of boundary layers, a point to which we shall return later. Eventually, the interdiffusion may generate a zone in which the local composition falls within the two-phase region of the equilibrium phase diagram, but the formation of a multiphase structure is relatively slow on the time scale established by the diffusional equilibration of solvent and nonsolvent in the membrane. We shall therefore assume

FORMATION OF POROUS STRUCTURES

479

(11) the thermodynamics appropriate to a one-phase system, even where the minimum free energy state calls for the appearance of a second phase. The experimentally observed fact that viscous solutions of polymers, metals, or glasses can exist as homogeneous phases for measurable periods of time in a thermodynamically metastable region lends support to our assumption. In general, the compositions reached at various positions in the film may be partly in the stable one-phase region (outside the binodal line of the three component phase diagram, Fig. a), partly between the binodal and spinodal, where the one-phase situation is metastable, and partly under the spinodal where it is not even locally stable. If the diffusion process occurring in the membrane generates at some point a composition on the boundary of the unstable region (the spinodal), we assume that (111) a three-dimensional structure, consisting of two interspersed equilibrium phases, will appear and spread to all compositions below the binodal. This, basically, is the model. As it stands it is capable of describing membrane formation only up to the point where precipitation occurs. The description of the three-dimensional structure which develops if there is phase separation is beyond the present model which could be extended somewhat beyond this stage by assuming that local phase equilibrium is maintained and that no transfer of polymer occurs between the two phases. However, such an extension is not part of this paper, which will concentrate primarily on distinguishing systems producing dense membranes (i) from those which lead to porous structures (ii) and (iii).*

Fig. 2. Phase diagram for a system with binary x parameters similar to those of the water (l), acetone (2), cellulose acetate ( 3 )system. The parameters used are listed in Table I. The stars are experimental coexisting compositions from ref. 4.

* Classification of reverse osmosis membranes into dense and porous is necessarily arbitrary; in our definition we will call the membrane porous if the phase separation takes place during the casting.

480

COHEN, TANNY, AND PRAGER

DIFFUSION PROBLEM
We now treat the diffusion process which determines the local compositions within the membrane. We will aim only at the zeroth-approximation level of the theory, simplifying as much as possible at every stage, in order to reduce the computational difficulties, and also to minimize the data required before predictions can be made. We shall (IV) neglect volume changes in the overall system of membrane plus bath. The region containing the polymer, the membrane itself, may of course change considerably in extent as nonsolvent enters and solvent leaves, but the volume change in the membrane is assumed to be exactly compensated by an equal and opposite change in the volume of the bath. We take the composition variables 1 to be the volume fractions 41,42, and 4 3 = 1 - 4 - 42 of nonsolvent, solvent, and polymer, and the diffusion flux of species i may then be defined as the volume of i passing through unit area per unit time. To avoid the complications arising from motion of the membrane-bath interface, we shall introduce (following Hartley and Crankl4) a position coordinate measured in terms of the volume m of polymer per unit area of membrane between the interface and the point of observation. The interface is always at m = 0, although its position z in a laboratory frame is a function of time

m(z)=

I*
surface

43(Z)

dz

The two independent fluxes that define the local material transport are most Z conveniently taken as volume fluxes J 1 and J of nonsolvent and solvent with respect to the polymer, i.e., through surfaces of fixed m. Hence the motion of the polymer never has to be explicitly treated, since we are dividing the membrane into layers containing a fixed amount of polymer dm per unit area, and formulating our diffusion equations in terms of a material balance of small molecule species in each layer. The appropriate driving forces associated with the volume fluxes J 1and J2 are the gradients of chemical potentials dplldm and dp21dm;in writing the phenomenological relations, we will (V) neglect cross terms so as to obtain the simplified phenomenological relations

and

where D1 and 0 2 are the diffusion coefficients for nonsolvent and solvent in the polymer-fixed reference frame, and p1 and p 2 are the molar chemical potentials. Assumption (V) does not of course allow the solvent and nonsolvent to move independently of one another. In the first place, a gradient in the volume fraction of one component will produce gradients in the chemical potentials, and therefore fluxes, in both components. Second, since there is no overall change in volume, a flux in one component must generate compensating fluxes in the

FORMATION OF POROUS STRUCTURES

481

others. This displacement flow is concealed by our choice of the polymer as a reference frame, but it would be very much in evidence in a laboratory-fixed frame of reference (see Appendix). As indicated in eq. (l), diffusion coefficients are concentration dependent the and even a zeroth-order approximation cannot ignore changes which may amount to several orders of magnitude.10 However, we shall assume that (VI) a Waldens rule type of relation exists, so that the ratio (Y = [D1(41,4d/ D2(@1,&)] constant; since solvent and nonsolvent molecules are of comparable is size, we shall take (Y to be near unity. The diffusion problem at which we have arrived through these assumptions can be written as a pair of partial differential equations for &(m,t) and cbz(m,t), d(41/43)- bJ1 1 D141--. (2) at bm - R T d m b(42/43)- dJ2 - 12 D2422 bt dm - aRTbm with boundary and initial conditions

( 2) ( 2 1

cbz(0,t)= 4 P 41(0,t) = 4Y, (3) Ji(M,t) = J,(M,t) = 0 4l(m,O)= 4 1 , 42(m,O)= (0 where M is the total volume of polymer per unit area of membrane. We shall consider the cast film to be completely homogeneous just before immersion in the coagulation bath, and thus ignore the extra complications which may arise from surface phenomenal5 and predrying.16 In order to solve the diffusion problem expressed by eqs. (2) and (3), we would normally divide the membrane into N layers of thickness Am = M / N , and carry out a mass balance of components 1and 2 in each layer; the larger the number of layers, the more accurate would be our calculation of the concentration distribution. We shall however restrict ourselves to the lowest level of approximation, i.e., to N = 2: the membrane is divided into a diffusion layer extending from m = 0 to m = mD(t),and a layer of unchanged casting solution extending from m = mD(t)to M . In this description, there are two stages to the diffusion process: an initial period, during which mD(t)is less than M , which lasts until the composition next to the support begins to change; if phase separation does not occur beforehand, there is also a second stage during which m remains constant at M, with changes D in composition occurring throughout the membrane. We restrict ourselves here to membranes sufficiently thick so that phase separation, if it occurs, takes place during the first stage. We now assume (VII) that during this stage (mn < M ) the concentration distribution in the diffusion layer at a given time is the steady-state distribution for a membrane of thickness mD(t) with composition &), & ) at m = 0 and &, & at m = mD(t). This assumption implies that at any instant the fluxes J1and J 2 are independent of m throughout the diffusion layer. Their ratio can then be written from eq. (1)as

482

COHEN, TANNY, A N D PRAGER

The thickness of the diffusion layer increases so as to just accommodate the total net volume of small molecule species which has entered or left the membrane. So long as mD has not reached M , u in eq. (4) remains a constant independent of time and is determined by the compositions prevailing at m = 0 and m =
mD.

Expressing the differentials of chemical potentials as functions of volume fractions and rearranging, we have

-- - ~ & ( ~ P ~ W I ) T ~, IP( ~, P+~ ~~ ~ I ) T , P , + Z d42 / d4l 4l(aPl/d4Z)T,P,*] - u42(dP2/d42)T.P.*1

(5)

Equation ( 5 ) is a first-order differential equation for the relationship between 41and 4 2 in the diffusion layer, defining a family of curves in the &, 4 2 plane; a particular member of the family is determined by specifying the surface and bulk compositions. In order to apply eq. (5), we must know the composition . dependence of the chemical potentials F~ and ~ 2 We shall assume (VIII) that the system follows Flory-Huggins

where ui is the molar volume of species i, and xi, is the usual binary interaction parameter between species i and j . In Figure 2, we show a nonsolvent-solvent-polymer phase diagram obtained for xi, and ui values similar to those found in the literature for practical systems (see Applications). We have plotted in Figure 3 various paths for different u's and different initial compositions to give a general idea of their behavior. If the composition path remains entirely in the stable-metastable region no phase separation will occur, and the result will be a dense nonporous film. If the composition path touches the spinodal, phase separation will ultimately produce a porous structure. To generate the actual concentration profiles &(m) and 42(m)in the diffusion layer, we may rewrite eq. (1) as

and perform the integration numerically along various composition paths. For the case m = mD(t), eq. (7) is a relationship between J l ( t ) and the thickness of the diffusion layer which can be used to eliminate J1:

From eq. (8)it is clear that to calculate the concentration profile it is no longer possible to avoid the explicit use of the diffusion coefficient, as we were able to do in the determination of the &(41) curves.

FORMATION OF POROUS STRUCTURES

483

Fig. 3. Phase diagram and composition paths modelling the water (I), acetone (2), cellulose acetate ( 3 )system. T h e various composition paths correspond to the following values of c: (a) -0.97, (b) -0.96, (c) -0.9, (d) -1.1, (e) -1.5, (f) -0.9, (g) -0.2, and (h) +2.0.

To obtain mD(t)we use the fact that the net volume flux of small molecules into the membrane is (J1 J2),and that change in the total volume of molecules 1 and 2 in the diffusion layer is given by

I t follows that

and, substituting from eqs. ( 1 )and (4), we obtain

where

is independent of time. Integration of eq. (11)with respect to t yields

J l ( t )= [ A / 2 ( l l / a ~ ~ ) t ] ' / ~
so that

That the flux and the diffusion layer thickness are inversely and directly

484

COHEN, TANNY, AND PRAGER

proportional to t 1/2, respectively, is generally true for any system in which diffusion coefficients depend on composition alone. I t can be shown that even if our assumption of the Waldens rule relationship [assumption (VI)] is abandoned the and the neglected cross terms restored to eq. (l), composition will still be a function of mlt1I2so long as the membrane can be treated as being infinitely thick. The data of Frommer and Lancet7 for the penetration of the precipitation boundary for cellulose acetate solutions in contact with water do indeed exhibit t ll2 behavior.

APPLICATIONS We have obtained results for three sets of parameters (Table I), chosen to represent real ternary systems: water-acetone-cellulose acetate (Fig. 3), water-acetic acid-cellulose acetate (Fig. 4), and ethanol-toluene-polystyrene (Fig. 5). The first two are often used as a basis for Loeb-Sourirajan recipes for producing porous asymmetric membranes,l7 but unfortunately have large deviations from Flory-Huggins behavior: presumably as a result of hydrogen bonding; the last is a nonaqueous system with much more limited hydrogen bonding possibilities.
TABLE I
Values of Parameters Utilized in the Systems 1nvestigat.ed System Water-acetone-cellulose acetate Water-acetic acid-cellulose acetate Ethanol-toluene-polystyrene
x12 x13 X28(0,/02)
010:: :203

-0.3 0.12 1.45

1 1.1 2.10

0.05 -0.3 0.25

1:4:500 1:3.34:500 1:1.8:1000

Fig. 4. Phase diagram and composition paths modelling the water (1) acetic acid (2) cellulose acetate (3) system. The parameters used are listed in Table I. The various u shown are (a) -0.9, (b) -0.865, (c) -0.855, (d) -0.825, (e) -0.8, (f) -0.85. The star indicates the only available precipitating composition (ref. 22).

FORMATION OF POROUS STRUCTURES

485

Fig. 5. Phase diagram and composition paths modelling the ethanol (1) toluene (2) polystyrene (3) system. The parameters used are listed in Table I. The various c shown are (a) -0.95, (b) -0.75, (c) -0.67, (d) -0.5.

The Flory-Huggins parameters x12 for the water-solvent-cellulose acetate systems were obtained from heat of mixing data7 fitted to a Van Laar expression, xi3 from equilibrium sorption data,4and the x23 were average values taken from the literature.18Jg The molar volumes for the small molecule species are of course readily available, and the phase diagram is as usual fairly insensitive to the choice of u3 so long as u3/u1 is sufficiently large. The resulting binodal curve differs from that observed experimentally, and the spinodal must therefore show a similar deviation. It is probably best to look on Figures 4 and 5 as illustrative of the various type of behavior that might be exhibited in a Loeb-Sourirajan process, rather than as true representations of the particular systems which served as bases for our choices of the x and u parameters. For the ethanol-toluene-polystyrene system one would expect closer adherence to Flory-Huggins thermodynamics. The values of xlz and x13 in this case were estimated from Hildebrand solubility parameters,'O and x23 was taken from the literature.21 Since the film surface is supposed to equilibrate with the coagulant adjacent to it [Assumption (I)], only surface compositions on the binodal or outside it are eligible for consideration; both mixtures containing only precipitant and casting solvent will generate surface compositions on the binodal; surface compositions outside the binodal can be obtained by adding appropriate solutes (electrolytes, for example)22 the bath to reduce the activities of precipitant and solvent there. to Curve b of Figure 3, which is the only one capable of connecting the given casting composition P to a composition B in the stable region by a path that contacts the spinodal, divides the stable region into two parts: an area whose points can be joined to P by constant-a lines lying entirely outside the spinodal of Figure 3, and a smaller area BAC' consisting of points for which this is not possible. By Assumption (11),a casting solution P whose surface is maintained a t a composition B' will undergo no phase separation at all, producing instead a dense

486

COHEN, TANNY, AND PRAGER

nonporous film. The nonexistence of a constant-a connection between P and composition B in BAC is an indication that the quasi-steady-state approximation is failing, and that we should divide the diffusion layer into two or more zones, each with its own a; the result is a path consisting of constant-a segments, the determination of whose actual course requires knowledge of D1(41,&)and 02(&,$2). Indeed the exact path predicted from eqs. (2) without recourse to a quasi-steady-state approximation will show a continuous variation of a along its length, and will not enter the unstable region at all, since the chemical potential gradients and fluxes are all zero at spinodal compositions, where all the d p i l d 4 j vanish. Surface compositions which can not be connected to P by a constant-a line will therefore be connected instead by a variable-a line touching the spinodal, or a t least passing close to it-failure to do so would mean that a constant-a path exists after all. We accordingly offer the interpretation that (IX) two-phase structures will only be formed from a given casting solution P if the film surface is maintained at a composition which can notbejoined to P by a constant-a path lying wholly outside the spinodal. The introduction of finite transfer coefficients a t the film-bath interface through inadequate stirring of the bath will produce a transient period during which the effective surface concentration (and therefore a), will vary with time. The transient solvent-nonsolvent ratio at the film surface will then be higher than the equilibrium value but it will not in general affect the ultimate formation of two-phase structures predicted on the basis of Assumption (IX). The Flory-Huggins simulations of the three systems considered indicate that only water-acetone-cellulose acetate (Fig. 3) is capable of forming porous structures from a 10%polymer solution. Neither water-acetic acid-cellulose acetate (Fig. 4) nor ethanol-toluene-polystyrene (Fig. 5) modeled systems show regions outside the binodal which cannot be joined to the casting composition. The water-acetic acid-cellulose acetate system comes close to forming porous structures, but the area BAC in Figure 4, although it does satisfy the requirement that it cannot be joined to P, lies entirely in the metastable region. From Figure 3 we conclude further that bath compositions from pure water to 30 vol % water in acetone (corresponding to the surface composition A ) may produce two-phase structures for casting composition P. However, the model predicts that any attempt to form porous films by air drying a solution of cellulose acetate in a water-acetone mixture will fail because all stable casting compositions can be joined to the pure polymer vertex. These predictions are in reasonable accord with experiments. Water-acetone-cellulose acetate and water-acetic acid-cellulose acetate can form either dense or porous membrane^.^^^ Preliminary experimentsz3on the membranes formed from ethanol-toluene-polystyrene system indicate a very low porosity consistent with our results. It should be pointed out that for cellulose acetate membranes, the composition paths are in fact very sensitive to the choice of u and will lead to different structures as can be seen from Figure 3. This may be the cause of the experimental difficulty in exactly reproducing membranes under apparently identical c ~ n d i t i o n s . ~ , ~ J ~ The distinction between membranes that are merely porous [type (ii)] and those consisting of a dense layer on a porous support [type (iii)] has not really been addressed in this paper, but our model suggests that if the film surface is maintained at a binodal composition, formation of two phase structures will still

FORMATION OF POROUS STRUCTURES

487

leave a dense surface, since the volume fraction of the polymer-poor phase separating outright a t the surface layer will be zero. If the surface is instead maintained a t a composition in the interior of the stable region, a dense film of finite thickness will be formed. Such a situation is shown in Figure 6, where a profile of & ( m / m ~ as calculated from eq. (8), is shown for case b of Figure 3, ), with diffusion coefficients decreasing exponentially with the polymer fraction

D 1= D 2 = D o exp[A(l- 43)]
For a surface composition B and a casting solution P of 10%cellulose acetate in acetone, taking A = s7 (correspondingto a 1000-fold increase in D as $3 goes from 0 to l),we find that at the instant of phase separation there is a thin one-phase layer, containing about 1%of the total polymer in the diffusion zone, between the two-phase region and the film surface. It is this dense layer that will ultimately provide the separative element in the finished membrane.

CONCLUSION
We have developed a model for the diffusion controlled formation of porous membranes obtained by immersing a film of polymer solution in a coagulating bath. We have treated the flow of solvent and nonsolvent in the film as a one dimensional diffusion process under the following assumptions: (I) The polymer film surface is in equilibrium with a stirred bath. (11) Phase separation begins at compositions lying at the limit of metastability (the spinodal curve). (111) If and when phase separation occurs, a three-dimensional structure consisting of two interspersed equilibrium phases will appear and spread to all compositions below the binodal. (IV) Volume changes in the ouerall system of film plus bath are negligible.

t
SKIN

*
4

2
in m/mD

Distance

Fig. 6. Polymer volume fraction profile in the diffusion layer from the surface to the bulk casting composition for curve b of Figure 3 .

488

COHEN, TANNY, AND PRAGER

(V) Cross terms in the phenomenological relations for fluxes relative to the polymer are unimportant when these are written in terms of chemical potential gradients. (VI) The ratio of the diffusion coefficients of solvent and nonsolvent can be taken as a constant of order unity. (VII) The fluxes in the diffusion layer will vary with time, but are at any instant constant throughout that layer (the quasi-steady-state assumption). (VIII) Flory-Huggins thermodynamics applies. (IX) If no quasi-steady-state can be found for a given pair of casting and surface composition, the actual composition profile in the diffusion layer will pass sufficiently close to the spinodal to initiate phase separation. To these formal assumptions we have added a tacit postulate which should be treated with caution: throughout the paper it is taken for granted that phase separation during the exchange of solvent and nonsolvent is equivalent to formation of pores in the finished membrane. There are of course many ways in which a two-phase structure can revert to a denser film in the final stages of the coagulation or during the subsequent processing (for example, the polymer-rich phase may coalesce to a structure anywhere between that of the original phaseseparated wet membrane and that of a dense film). Phase separation should be understood as a necessary but not sufficient criterion for the formation of porous membranes.
This work was initiated a t the Weizmann Institute of Science in Israel. Two of us (C.C. and S.P.) would like to thank the staff of the Polymer Department of the Institute for their hospitality and stimulating discussions.

APPENDIX
The volume current densities of component i relative (a) to an external laboratory-fixed coordinate frame, (b) to the local center of volume, and (c) to the center of volume of component k are given byZ4( a ) j i = &u,, ( b ) j i = &(ui -UO), and (c) J ; = @;(u; Uk),whereu, is the localvelocityofcomponent i and u, is defined by ug = B+;u;. One can then write the following relationship for a three-component systemZ4(with k = 3):

which clearly indicates that cross terms involving components 1 and 2 are adequately included in

FORMATION OF POROUS STRUCTURES

489

our equations (1)and that the coefficients of these cross terms in eq. (17) obey Onsager reciprocal relations as they should.

References
1. P. J. Flory, Principles of Polymer Chemistry, Cornell University, Ithaca, N.Y., 1953, Chap. 12. 2. H. Tompa, Polymer Solutions, Butterworths, London, 1956, Chap. 7. 3. R. E. Kesting and A. Mennefee, Kolloid 2. 2. Polym., 230,341 (1968). 4. H. Strathmann, P. Scheible and R. W. Baker, J . Appl. Polym. Sci.,15,811 (1971). 5. R. Bloch and M. A. Frommer, Desalination, 7,259 (1969). 6. M. A. Frommer, I. Feiner, 0. Kedem, and R. Bloch, Desalination, 7,393 (1969). 7. M. A. Frommer and D. Lancet, in Reverse Osmosis Membrane Research, H. K. Lonsdale and H. E. Podall, Eds., Plenum, New York, 1972. 8. D. R. Paul, J . Appl. Polym. Sci., 12,383 (1968). 9. A. Rende, J . Appl. Polym. Sci., 16,585 (1972). 10. J. E. Anderson and R. Ullman, J . Appl. Phys. 44,4303 (1973). 11. H. Strathmann and K. Kock, Desalination, 21,241 (1977). 12. J. S. Kirkaldy and L. C . Brown, Can. Metall. Q., 2,89,1963. 13. K. J. Ruschack and C. A. Miller, Ind. Eng. Chem. Fundam., 11,534,1972. 14. G. S. Hartley and J. Crank, Trans. Faraday SOC., 45,801 (1949). 15. G. B. Tanny, J . Appl. Polym. Sci., 18,2149 (1974). 16. G. J. Gittens, P. A. Hitchcock, D. C. Sammon, and G. E. Wakley, Desalination, 8, 369, 1970. 17. S. Loeb and S. Sourirajan, Ado. Chem. Ser., 38,117,1962. 18. W. R. Moore and R. Shuttleworth, J . Appl. Polym. Sci., Al, 733 (1963). 19. T . Kawai, J . Polym. Sci., 32,425 (1958). 20. J. Brandrup and E. H. Immergut, Polymer Handbook, Wiley, New York, 1966. 21. S. H. Maron and N. Nakajima, J . Polym. Sci., 42,327 (1960). 22. M. A. Frommer, R. Matz, and U. Rosenthal, Ind. Eng. Chem. Prod. Res. Deu., 10, 193 (1971). 23. G. B. Tanny, unpublished results. 24. D. D. Fitts, Non-Equilibrium Thermodynamics, McGraw-Hill, New York, 1962,Chap. 1.

Received August 22,1978

You might also like