+ All Categories
Home > Documents > Thermodynamics of Fluid-phase Equilibria for Standard Chemic

Thermodynamics of Fluid-phase Equilibria for Standard Chemic

Date post: 08-Apr-2016
Author: jenkeiz-khan
View: 59 times
Download: 0 times
Share this document with a friend
phase equilibrium
Embed Size (px)
of 23 /23
ANNIVERSARY ARTICLE Thermodynamics of Fluid-Phase Equilibria for Standard Chemical Engineering Operations John M. Prausnitz Dept of Chemical Engineering, University of California, Berkeley, CA 94720 and Chemical Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720 Frederico W. Tavares Dept of Chemical Engineering, University of California, Berkeley, CA 94720 and Escola de Quı ´mica, Universidade Federal do Rio de Janeiro, Caixa Postal 68542, CEP, 21949 –900, Rio de Janeiro, RJ, Brazil DOI 10.1002/aic.10069 Published online in Wiley InterScience (www.interscience.wiley.com). Thermodynamics provides one of the scientific cornerstones of chemical engineering. This review considers how thermodynamics is and has been used to provide phase equilibria as required for design of standard chemical engineering processes with emphasis on distillation and other conventional separation operations. While this review does not consider “modern” thermodynamics for high-tech applications, attention is given to 50 years of progress in developing excess-Gibbs-energy models and engineering- oriented equations of state; these developments indicate rising use of molecular physics and statistical mechanics whose application for chemical process design is made possible by increasingly powerful computers. As yet, results from molecular simulations have not had a major influence on thermodynamics for conventional chemical engineering; how- ever, it is likely that molecular simulation methods will become increasingly useful, especially when supported by quantum-mechanical calculations for describing intermo- lecular forces in complex systems. © 2004 American Institute of Chemical Engineers AIChE J, 50: 739 –761, 2004 Keywords: fluid-phase equilibria, chemical thermodynamics, equations of state Introduction Separation of fluid mixtures is one of the cornerstones of chemical engineering. For rational design of a typical separa- tion process (for example, distillation), we require thermody- namic properties of mixtures; in particular, for a system that has two or more phases at some temperature and pressure, we require the equilibrium concentrations of all components in all phases. Thermodynamics provides a tool for meeting that re- quirement. For many chemical products (especially commodity chemi- cals), the cost of separation makes a significant contribution to the total cost of production. Therefore, there is a strong eco- nomic incentive to perform separations with optimum effi- ciency. Thermodynamics can contribute toward that optimiza- tion. In this article, we indicate some of the highlights of progress in the thermodynamics of phase equilibria since the AIChE Journal began about 50 years ago. Although it is clear that, in addition to phase equilibria, thermodynamics also provides useful caloric information (en- thalpy balances) that affects separation operations, we do not consider that here. Furthermore, we recognize that in recent years, thermodynamic research has given rising attention to product (as opposed to process) design. There is good reason to Correspondence concerning this article should be addressed to J. M. Prausnitz at [email protected] © 2004 American Institute of Chemical Engineers THERMODYNAMICS AIChE Journal 739 April 2004 Vol. 50, No. 4

    Thermodynamics of Fluid-Phase Equilibria forStandard Chemical Engineering Operations

    John M. PrausnitzDept of Chemical Engineering, University of California, Berkeley, CA 94720

    andChemical Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720

    Frederico W. TavaresDept of Chemical Engineering, University of California, Berkeley, CA 94720

    andEscola de Qumica, Universidade Federal do Rio de Janeiro, Caixa Postal 68542, CEP, 21949900, Rio de Janeiro, RJ, Brazil

    DOI 10.1002/aic.10069Published online in Wiley InterScience (www.interscience.wiley.com).

    Thermodynamics provides one of the scientic cornerstones of chemical engineering.This review considers how thermodynamics is and has been used to provide phaseequilibria as required for design of standard chemical engineering processes withemphasis on distillation and other conventional separation operations. While this reviewdoes not consider modern thermodynamics for high-tech applications, attention isgiven to 50 years of progress in developing excess-Gibbs-energy models and engineering-oriented equations of state; these developments indicate rising use of molecular physicsand statistical mechanics whose application for chemical process design is made possibleby increasingly powerful computers. As yet, results from molecular simulations have nothad a major inuence on thermodynamics for conventional chemical engineering; how-ever, it is likely that molecular simulation methods will become increasingly useful,especially when supported by quantum-mechanical calculations for describing intermo-lecular forces in complex systems. 2004 American Institute of Chemical Engineers AIChE J,50: 739761, 2004Keywords: uid-phase equilibria, chemical thermodynamics, equations of state


    Separation of uid mixtures is one of the cornerstones ofchemical engineering. For rational design of a typical separa-tion process (for example, distillation), we require thermody-namic properties of mixtures; in particular, for a system thathas two or more phases at some temperature and pressure, werequire the equilibrium concentrations of all components in allphases. Thermodynamics provides a tool for meeting that re-quirement.

    For many chemical products (especially commodity chemi-cals), the cost of separation makes a signicant contribution tothe total cost of production. Therefore, there is a strong eco-nomic incentive to perform separations with optimum ef-ciency. Thermodynamics can contribute toward that optimiza-tion. In this article, we indicate some of the highlights ofprogress in the thermodynamics of phase equilibria since theAIChE Journal began about 50 years ago.

    Although it is clear that, in addition to phase equilibria,thermodynamics also provides useful caloric information (en-thalpy balances) that affects separation operations, we do notconsider that here. Furthermore, we recognize that in recentyears, thermodynamic research has given rising attention toproduct (as opposed to process) design. There is good reason to

    Correspondence concerning this article should be addressed to J. M. Prausnitz [email protected]

    2004 American Institute of Chemical Engineers


    AIChE Journal 739April 2004 Vol. 50, No. 4

  • believe that modern chemical engineering will be increas-ingly concerned not with large-scale production of classical(commodity) chemicals, but with new (specialty) materials forapplication in biotechnology, medicine, pharmaceuticals, elec-tronics, and optics, as well as in the food and personal-careindustries. Thermodynamics can contribute to the developmentof these new areas and, indeed, much progress toward thatend has been reported in the ever-growing literature. In ourshort review here, it is not possible to give adequate attentionto new thermodynamics. This review is essentially connedto progress in thermodynamics for what is often called tradi-tional chemical engineering.

    Furthermore, during the last 50 years, the literature on ap-plied thermodynamics has grown to vast proportions, evenwhen we limit our attention to phase equilibria. Therefore, thisreview cannot be comprehensive. Indeed, it is unavoidablyselective, reecting the authors (right or wrong) opinions andpreferences. Space and time are nite, and some subconsciousprejudice is always with us. The authors, therefore, apologizeto their colleagues whose work is not sufciently mentionedhere or worse, not mentioned at all. Because the editor hasnecessarily limited the total number of pages available to us,we must, with regret, omit much that may merit inclusion.

    Fifty years ago, most chemical engineering thermodynamicswas based on representation of experimental data in charts,tables, and correlating equations that had little, if any, theoret-ical basis. Fifty years ago, most chemical engineering thermo-dynamics was in what we may call an empirical stage. How-ever, the word empirical has several interpretations. Whenwe represent experimental data by a table or diagram (forexample, the steam tables or a Mollier diagram), we call suchrepresentation empirical. When we t experimental ideal-gasheat capacities to, say, a quadratic function of temperature,wechoose that algebraic function only because it is convenient todo so. However, when, for example, we t vaporpressure dataas a function of temperature, we inevitably do so by expressingthe logarithm of the vapor pressure as a function of the recip-rocal absolute temperature. This expression is also empiricalbut in this case our choice of dependent and independentvariables follows from a theoretical basis, viz. the ClausiusClapeyron equation. Similarly, when for a binary mixture, werepresent vapor-liquid equilibrium data with, say, the Margulesequation for activity coefcients, we also call that representa-tion empirical, although it has a theoretical foundation, viz. theGibbsDuhem equation. We should distinguish betweenblind empiricism, where we t experimental data to a totallyarbitrary mathematical function, and thermodynamically-grounded empiricism where data are expressed in terms of amathematical function suggested by classical thermodynamics.

    If now, in addition to thermodynamics, we introduce into ourmethod of representation some more-or-less crude picture ofmolecular properties, for example, the van der Waals equationof state, we are still in some sense empirical but now we are inanother realm of representation that we may call phenomeno-logical thermodynamics. The advantage of proceeding fromblind to thermodynamically-grounded to phenomenolog-ical is not only economy in the number of adjustable param-eters but also in rising ability to interpolate and (cautiously)extrapolate limited experimental data to new conditions, whereexperimental data are unavailable. Because phenomenologi-cal thermodynamics uses molecular concepts, an alternate

    designation is to say molecular thermodynamics. The moststriking engineering-oriented examples of molecular thermo-dynamics are provided by numerous useful correlations, basedon the theorem of corresponding states or on the concept ofgroup contributions.

    In contrast to what we (somewhat carelessly) call empiri-cal thermodynamics, we also have modern theoretical ther-modynamics that uses statistical mechanics and molecular sim-ulations. In this theoretical area, we relate macroscopicthermodynamic properties to microscopic characteristics in amore-or-less rigorous manner.

    Although thermodynamically-grounded thermodynamicswas well known fty years ago, and whereas some correlationsbased on corresponding states or group contributions have along history, the last 50 years have provided signicantprogress in phenomenological or molecular thermodynam-ics with numerous applications in chemical process design.During the last 25 years, there has also been much progress instatistical thermodynamics and molecular simulation. How-ever, regrettably, with few exceptions, that progress has not yetseen signicant application in engineering design of uid-phase chemical processes.

    For engineering application, applied thermodynamics is pri-marily a tool for stretching experimental data: given somedata for limited conditions, thermodynamics provides proce-dures for generating data at other conditions. However, ther-modynamics is not magic. Without some experimental infor-mation, it cannot do anything useful. Therefore, for progress inapplied thermodynamics, the role of experiment is essential:there is a pervasive need for ever more experimental results.Anyone who does thermodynamics is much indebted to thosewho work in laboratories to obtain thermodynamic properties.It is impossible here to mention even a small fraction of thevast body of new experimental results obtained over a period offty years. However, it is necessary here to thank the hundredsof experimentalists who have provided essential contributionsto progress in chemical thermodynamics. Particular recognitionmust go to the laboratories of Grant Wilson (Provo, Utah) andD. Richon (Fontainebleau, France) who have pioneered inobtaining experimental results for difcult industrial sys-tems.

    A useful compilation of experimental phase-equilibrium datais provided by the multivolume DECHEMA series (Behrensand Eckermann, 19801997).

    Thermodynamic Properties of Pure Fluids

    For common uids (for example, water, ammonia, lighthydrocarbons, carbon dioxide, sulfur dioxide, and some freons)we have detailed thermodynamic data conveniently compiledin tables and charts; the outstanding example of such compi-lations is the steam table with periodic improvements andextensions (Harvey et al., 1998; Harvey and Parry, 1999).

    Although thermodynamic data for less common uids areoften sketchy, a substantial variety of thermodynamic proper-ties for normal uids can be estimated from corresponding-states correlations, especially those based on Pitzers use of theacentric factor (19551958) for extending and much improvingclassical (van der Waals) corresponding states (Lee and Kesler,1975). Here normal applies to uids whose nonpolar orslightly polar (but not hydrogen-bonded) molecules are not

    740 AIChE JournalApril 2004 Vol. 50, No. 4

  • necessarily spherical but may be quasi-elliptical. Although theLee-Kesler tables are useful for numerous uids, with fewexceptions, correlations that are linear in the acentric factorcannot be used for strongly polar molecules or for oligomers,or other large molecules whose acentric factors exceed(roughly) 0.4.

    Two Procedures for Calculating Phase Equilibria

    For calculating uid-phase equilibria, it is common practiceto use either one of two methods. In Method I, we use fugacitycoefcients for all components in the vapor phase, and activitycoefcients for all components in the liquid phase. In MethodII, we use fugacity coefcients for all components in all uidphases. We nd fugacity coefcients from an equation of state(EOS), and activity coefcients from a model for the molarexcess Gibbs energy, as discussed in numerous textbooks (forexample, Smith et al. 2001; Kyle, 1999; Prausnitz et al., 1999;Sandler, 1999; Tester and Modell, 1997). Each method hassome advantages and some disadvantages.

    For Method I, for every component i, the essential equationof equilibrium is

    yiiVP xiifi0 (1)

    Here y and x are mole fractions in the vapor and the liquid,respectively; P is the total pressure; V is the vapor-phasefugacity coefcient, and is the liquid-phase activity coef-cient. Liquid-phase reference fugacity f0 is typically the pure-liquid vapor pressure at system temperature with (usuallysmall) corrections for pure-uid vapor-phase nonideality, andfor the effect of total pressure (Poynting factor).

    For liquidliquid equilibria, where the same standard state isused in all phases, we have

    xii xii (2)

    where and refer, respectively, to the two liquid phases.For Method II, for every component i, the essential equation

    of equilibrium is

    yiiV xiiL (3)

    for vapor-liquid equilibria or

    xiiL xii

    L (4)

    for liquidliquid equilibria. Here superscripts V and L refer,respectively, to the vapor phase and liquid phase; while and refer, respectively, to the two liquid phases.

    Before 1950, Method I was dominant because there wasreluctance to use an equation of state for condensed uids.Although many years before, van der Waals had clearly shownthat his (and similar) equations of state are applicable to bothgases and liquids, there was little condence in the ability ofsuch equations to represent the properties of liquids with suf-cient accuracy. Furthermore, before computers becamereadily available, equilibrium calculations were prohibitive ifboth phases were described by an equation of state. For liquid-

    phase mixtures, there was a preference to use well establishedactivity-coefcient models (typically the Margules and vanLaar equations). However, a pioneering application of MethodII was presented by Benedict et al. (1940, 1942) whose Bene-dict-Webb-Rubin (BWR) equation of state provided the basisfor an extensive correlation of high-pressure vaporliquid equi-libria of parafn mixtures. However, because computers werethen in their infancy, the cumbersome BWR equation was notused much. Furthermore, because the BWR equation of statefor a mixture (at that time) contained only pure-component (nobinary) constants, its accuracy was limited.

    Activity-Coefcient Models for Liquid Mixtures ofNonelectrolytes

    More than one-hundred years ago, Margules proposed tocorrelate isothermal binary vaporliquid equilibria (VLE) witha power series in a liquid-phase mole fraction to represent ln1, where 1 is the activity coefcient of component 1. Theactivity coefcient of component 2, 2, is then obtained fromthe GibbsDuhem equation without requiring additional pa-rameters. About 15 years later, van Laar derived equations forln 1 and ln 2 based on the original van der Waals equationof state. After introducing a key simplifying assumption forliquids at modest pressures (no volume change upon isothermalmixing), van Laar assumed that the isothermal entropy ofmixing at constant volume is equal to that for an ideal solution.Furthermore, upon assuming that the cross coefcient in thevan der Waals equation of state a12 is given by the geometricmean (a12 a11a22), van Laar obtained expressions for ln 1 and ln 2 that require only pure component parameters.However, regrettably, agreement with experiment was notgood.

    Figure 1. Predicted VLE for C6H6(1)/n-C7H16(2) at 70 Cwith the original regular solution theory.Here y is the vapor-phase mole fraction and x is the liquid-phase mole fraction.

    AIChE Journal 741April 2004 Vol. 50, No. 4

  • About 1930, Hildebrand and (independently) Scatchard, pre-sented a derivation similar to that of van Laar but, instead ofvan der Waals constant a, they used the concept of cohesiveenergy density, that is, the energy required to vaporize a liquidper unit liquid volume; the square root of this cohesive energydensity is the well-known solubility parameter . In the nalHildebrand expressions for ln 1 and ln 2, the square root ofthe cohesive energy density appears because of a geometric-mean assumption similar to that used by van Laar. Because ofthis geometric-mean assumption, the original regular-solutiontheory is predictive, requiring only pure-component experi-mental data (vapor pressures, enthalpies of vaporization, andliquid densities). For simple mixtures, Hildebrands regular-solution theory often gives a good approximation as illustratedin Figure 1.

    Results from regular-solution theory did not have apprecia-ble inuence in chemical engineering thermodynamics untilabout 25 years after that theory was published. To make thetheory more exible, the geometric-mean assumption is cor-rected by introducing a single binary coefcient l12 that verymuch improves agreement with experiment. Figure 2 illustrateshow binary parameter l12 can signicantly increase the accu-racy of the regular-solution theory. Numerous efforts have notsucceeded in correlating l12 in terms of pure-component prop-erties. Such a correlation is possible only for limited situations,as illustrated in Figure 3.

    When the modied regular-solution theory for liquid mix-tures was combined with the RedlichKwong equation of statefor vapor mixtures, it was possible to correlate a large body ofVLE data for mixed hydrocarbons, including those at highpressures found in the petroleum and natural-gas industries(Prausnitz et al., 1960). The resulting ChaoSeader correlation(Chao and Seader, 1961) was used extensively in industry untilit was replaced by other simpler methods, based on a cubicequation of state applied to all uid phases.

    Until about 1964, most chemical engineering applications ofactivity coefcients were based on either the Margules or thevan Laar equations, although in practice, the two binary coef-cients in the van Laar equation were not those based on thevan der Waals equation of state, but instead, those obtainedfrom reduction of binary VLE data.

    In an inuential article, Wohl (1946) showed how activity-coefcient equations can be systematically derived from aphenomenological model where the molar excess Gibbs energygE is expressed as a function of liquid-phase composition.Wohls systematic method had two primary effects: rst, itencouraged the development of new models (variations onMargules and van Laar), and second, very important for dis-tillation column design, it showed how binary VLE data can besystematically scaled up to predict VLE for ternary (andhigher) liquid mixtures with or without requiring any ternary(or higher) VLE data (Severns et al., 1955).

    For liquid mixtures containing strongly polar or hydrogen-bonding components, the van Laar equations for activity coef-cients are often not satisfactory, especially when applied tomulticomponent liquidliquid equilibria (LLE). For LLE, we

    Figure 2. VLE with the regular solution theory with bi-nary parameter 12.Effect of 12 on calculating relative volatility for the 2,2-dimethylbutane(1)/benzene(2) system.

    Figure 3. Binary parameter 12 for aromatic-saturatedhydrocarbon mixtures at 50 C.Binary systems shown are: 1. Benzene(2)-Pentane(1); 2. Ben-zene(2)-Neopentane(1); 3.Benzene(2)-Cyclopentane(1); 4.Benzene(2)-Hexane(1); 5. Benzene(2)-Methylpentane(1); 6.Benzene(2)-2, 2-Dimethylbutane(1); 7. Benzene(2)-2, 3-Dim-ethylbutane(1); 8. Benzene(2)-Cyclohexane(1); 9. Ben-zene(2)-Methylcyclopentane(1); 10. Benzene(2)-Heptane(1);11. Benzene(2)-3-Methylhexane(1); 12. Benzene(2)-2,4-Dimethylpentane(1);13. Benzene(2)-2, 2, 3-Trimethylbu-tane(1); 14. Benzene(2)-Methylcyclohexane(1); 15. Ben-zene(2)-Octane(1); 16. Benzene(2)-2, 2, 4-Trimethylpen-tane(1); 17. Toluene(2)-Hexane(1); 18. Toluene(2)-3-Methylpentane(1); 19. Toluene(2)-Cyclohexane(1); 20.Toluene(2)- Methylcyclopentane (1); 21. Toluene(2)-Hep-tane(1); 22. Toluene(2)-Methylcyclohexane(1); 23. Tolu-ene(2)-2, 2, 4-Trimethylpentane(1).

    742 AIChE JournalApril 2004 Vol. 50, No. 4

  • can use the Margules equations but for good agreement withexperiment for ternary (and higher) systems, it is often neces-sary to use many empirical coefcients.

    If ternary (and higher) coefcients are omitted, Wohlsmethod for multicomponent liquid mixtures assumes additiv-ity; for a ternary mixture, gE is essentially given by the sum g12E g13E g23E where the subscripts denote binary mixtures.However, whereas molecular considerations indicate that ad-ditivity is (approximately) correct for hE, the excess molarenthalpy (heat of mixing), there is no physical basis for theadditivity of gE. Wohls method, in effect, emphasizes thecontribution of hE, while neglecting the contribution of sE, themolar excess entropy (gE hE TsE).

    With a fundamentally different method that emphasizes sE(rather than hE), Wilson (Wilson, 1964) derived an equation forgE based on a generalization of Florys theoretical expressionfor the entropy of mixing noninteracting spheres and chains ofspheres (polymers). To take molecular interactions into ac-count, Wilson used the concept of local composition that, inturn, is based on Guggenheims quasi-chemical theory fornonrandom mixing, that is, the tendency of molecules in aliquid mixture to show preferences in choosing their immediateneighbors. For example, in a mixture of methanol and hexane,because of hydrogen bonding between two (or more) methanolmolecules, a methanol molecule prefers to be near anothermethanol molecule, rather than near a hexane molecule. How-ever, in a mixture of chloroform and acetone, because ofhydrogen bonding between the CH group of chloroform andthe C O group of acetone, a chloroform molecule prefers tobe near an acetone molecule.

    With two binary constants per binary mixture, Wilsonsequation for gE is often superior to the older two-constantequations, especially for VLE of mixtures where one (or more)components can hydrogen bond. Unfortunately, however, Wil-sons equation cannot be used for binary LLE without oneadditional binary parameter.

    Encouraged by Wilsons use of the local composition con-cept of 1964, two other models with the same concept wereproposed: Renons nonrandom two-liquid (NRTL) model of1968, and Abrams universal quasi-chemical (UNIQUAC)model of 1975. Although the theoretical basis of these local-composition models is not strong, subsequent to their publica-tion, they obtained some support from molecular simulationstudies (Hu et al., 1983; Nakanishi and Tanaka, 1983; Phillipsand Brennecke, 1993). Although NRTL uses three adjustableparameters per binary, one of these (nonrandomness parameter12) can often be set a priori; a typical value is 12 0.3. BothNRTL and UNIQUAC are readily generalized to multicompo-nent mixtures without additional parameters, and both may beused for VLE and LLE.

    Modications of the Wilson equation and the NRTL equa-tion have been used to describe phase equilibria in polymersolutions (Heil and Prausnitz, 1966; Chen 1993).

    The UNIQUAC equations use only two adjustable binaryparameters per binary and, because the congurational part ofthe excess entropy is based on Florys expression for mixturesof noninteracting short and long-chain molecules, UNIQUACis directly applicable to liquid mixtures that contain polymers.Furthermore, because UNIQUAC separates the congurationalentropy contribution to gE from the residual contribution that isprimarily because of attractive intermolecular forces whenever

    the components in the mixture are not identical in size andshape, extension to multicomponent mixtures (like the Wilsonmodel) does not rely on simple additivity of binary excessGibbs energies.

    NRTL and UNIQUAC have been extensively used for aboutthirty years, largely (but not totally) replacing the equations ofMargules, van Laar, and Wilson.

    Finally, for liquid mixtures containing strongly interactingmolecules (for example, alcohols), some models for gE arebased on chemical equilibria either with or without a contri-bution from physical interactions. Because molecules do notknow whether they are doing physics or chemistry, anydivision between physical and chemical interactions issomewhat arbitrary. Nevertheless, when guided by molecularphysics, such arbitrary division can be useful for correlation ofexperimental data.

    A chemical theory for activity coefcients was rst intro-duced nearly 100 years ago by Dolezalek who claimed that areal mixture is an ideal solution provided that we correctlyidentify the mixtures molecules. For example, VLE for mix-tures of ethanol and heptane can be represented by an ideal-solution calculation when we consider that some ethanol mol-ecules are dimers, or trimers, and so on, as determined bychemical equilibrium constants. Improvement is obtainedwhen, in addition, we allow the various true chemical speciesto interact with each other through physical forces as givenby an equation of the Margules, van Laar, NRTL or UNIQUACform. Because such chemical theories necessarily requirenumerous adjustable equilibrium constants, it is customary, inpractice, to make simplifying assumptions; a common one is toassume that the equilibrium constant for association (for ex-ample, alcohols) is independent of the degree of association.Figure 4 shows calculated and measured liquidliquid equilib-

    Figure 4. Experimental and calculated ternary LLE forethanol-ethylnitrile-n-hexane at 40 C.Concentrations are in mole fractions. Dashed lines are pre-dictive calculations with parameters obtained from binarysystems; points are experimental data, and full lines are ob-served phase-envelope and experimental tie lines (Nagata andKawamura, 1979).

    AIChE Journal 743April 2004 Vol. 50, No. 4

  • ria for the ethanolethylnitrilenhexane system at 40C. Cal-culations for the molar excess Gibbs energy use a chemical-plus-UNIQUAC equation g E. These calculations use twobinary physical (UNIQUAC) parameters and two chemicalparameters (Nagata and Kawamura, 1979). A more detailedchemical plus physical model is the ERAS model developedby Heintz and coworkers (Heintz et al., 1986; Letcher et al.,1995); chemical plus physical solution theory is frequentlyused for describing the properties of electrolyte solutions (forexample, Lu and Maurer, 1993).

    Although chemical theories are often successful for binarymixtures, generalization to ternary (and higher) mixtures withmore than one associated component is often not possiblewithout introduction of numerous additional adjustable param-eters or additional (doubtful) simplications. However, forsome cases, good results are achieved as indicated by Nagata etal. (2000).

    Effect of Temperature on VLE and LLE

    The effect of temperature presents a fundamental problem inthe application of activity-coefcient models because the ad-justable binary (or higher) parameters depend on temperature.Although thermodynamics provides exact equations that relatethat temperature dependence to either the excess enthalpy orexcess entropy of mixing, such equations are of little usebecause the required enthalpy or entropy data are only rarelyavailable. Fortunately, for VLE calculations, the effect of tem-perature on activity coefcients is often not large; the primaryeffect of temperature on VLE comes from the (large) knowneffect of temperature on pure-component vapor pressures. ForVLE, it is common practice either to neglect the effect oftemperature on ln I or, as predicted by regular-solutiontheory, to assume that at constant composition, ln i is pro-portional to 1/T. However, for LLE (where pure-componentvapor pressures play no role), the effect of temperature is likelyto be signicant. Regrettably, at present, we do not have anyconsistently reliable molecular thermodynamic methods forcalculating the effect of temperature on LLE.

    Solubilities of Gases and Solids

    At moderate pressures, the solubility of a gas j in a liquid iis given by Henrys constants H i j that depends on temperature.We have a reasonably large data base for these constants forcommon gases in a variety of common liquids. However, themajor part of that data base is for temperatures near 25 C; thefurther we go from 25C, the smaller the database. For somenonpolar systems, we can estimate Henrys constants withHildebrands (Hildebrand et al., 1970) or Shairs correlation(Prausnitz and Shair, 1961) based on solubility parameters. Foradvanced pressures, we can add a correction to Henrys lawusing partial molar volumes of the gaseous solutes; these areoften not known but for nonpolar systems, we can often esti-mate them with a correlation (for example, Lyckman et al.,1965; Brelvi and OConnell, 1972). Some correlations forHenrys constants are based on scaled-particle theory (Pierotti,1976, Geller et al., 1976) or on assumptions concerning theradial distribution function for a solute molecule completelysurrounded by solvent molecules (Hu et al., 1985).

    The effect of temperature on gas solubility follows, in part,

    from the effect of temperature on solvent density. For manycases, solvent density is the dominant inuence on Henrysconstant. An example is shown in Figure 5 that correlatessolubilities of ve gases in water over a large temperaturerange. In Figure 5, f1 is the fugacity of the solvent.

    Some attention has been given to correlating Henrys con-stant for a gas in a mixed solvent. For simple systems, for a gasj, good approximations can often be made with a volume-fraction average for ln Hj that requires knowing only Hj forevery solvent in the mixture. A better approximation is oftenobtained by assuming additivity of binary interactions. In thatcase, for a ternary mixture, we need not only Hj for gas j in bothpure solvents but, in addition, some information on interactionsin the binary (gasfree) solvent mixture (Campanella et al.,1987; Shulgin and Ruckenstein, 2002).

    In nonelectrolyte systems, the solubilities of solids are com-monly calculated by referring the solutes activity coefcient tothe solutes subcooled liquid. The ratio of the ctitious vaporpressure (or fugacity) of the subcooled liquid to that of thestable solid at the same temperature is found from knowingprimarily the solutes melting temperature and enthalpy offusion, and secondary, from the difference in heat capacities ofthe solid and subcooled liquid. Correlations of solid solubilitiesbased on such calculations have been presented by numerous

    Figure 5. Henrys constant for several gases (2) in water(1) from Japas and Levelt-Sengers (1989).The line is obtained from a linear equation where is solventdensity and parameters A and B are constants for each gas.The solvents fugacity is f1.

    744 AIChE JournalApril 2004 Vol. 50, No. 4

  • authors, notably by McLaughlin (McLaughlin and Zainal,1959, 1960; Choi and Mclaughlin, 1983); an example is shownin Figure 6.

    Thermodynamic Properties from GroupContributions

    Because desired thermodynamic data are frequently in shortsupply, many efforts have been made to estimate these prop-erties from known molecular structure. If we divide a moleculeinto its constituent groups, it then seems reasonable to assumethat each group contributes to a particular thermodynamicproperty, such as the molar volume, or the normal boilingpoint. For example, Seinfeld and coworkers (Asher et al., 2002)have used group contributions to estimate thermodynamicproperties of some oxygen containing liquids and solids thatare required for analysis of air-pollution data.

    Because corresponding-states-correlation methods are oftenuseful, it is particularly important to estimate the critical tem-

    perature and pressure of a pure uid; as a result, starting in themid-fties, the literature is rich in group-contribution methodsfor estimating critical properties; some of these are summarizedin Properties of Gases and Liquids by Poling et al. (2001).

    It is by no means simple to establish a reliable group-contribution method for pure uids, however, it is more dif-cult to establish such a method for uid mixtures, in particular,for activity coefcients of all components in that mixture.However, for applications in chemical process design, it isuseful to have such a correlation because, for many (indeed,most) mixtures, activity-coefcient data are at best sketchy andoften nonexistent.

    About 75 years ago, Langmuir briey discussed a possibleactivity-coefcient correlation based on group contributions.However, Langmuirs idea remained dormant for about 40years, primarily because the required data base was too small,and because the necessary calculations are too tedious withouta computer. Langmuirs idea was revived by Deal and Derr(1969) who presented an early version of their ASOG correla-tion . To use this correlation, we need temperature-dependentgroupgroup parameters; because Deal and Derr provided onlya few of these parameters, the usefulness of ASOG was se-verely limited. At present, its usefulness is somewhat largerthanks to a monograph (Kojima and Tochigi, 1979) and articlesby Tochigi et al. (1990, 1998).

    When the UNIQUAC model for activity coefcients is mod-ied toward a group contribution form, it leads to the UNIFACcorrelation (Fredenslund et al., 1975), UNIFAC is simpler touse than ASOG because (to a rough approximation) its param-eters are independent of temperature. UNIFAC was eagerlypicked up by numerous users because the authors of UNIFACsupplied necessary software; a monograph (Fredenslund et al.,

    Figure 6. Solubility of aromatic solids in benzene at dif-ferent temperatures.The ideal solubility is calculated with the equation:

    log x2 54.42.303R TmT 1where 54.4 Jmol-1 K-1 is an average value for the entropy offusion of solids considered. Tm is the melting temperature,and x2 is the mole fraction of the solute in the liquid solution.

    Figure 7. VLE calculations for the benzene(1)-sulfo-lane(2) system at several temperatures withUNIFAC (Wittig et al., 2003).In the vapor phase, the mole fraction y1 is essentially unity forpressures higher than 5 kPa.

    AIChE Journal 745April 2004 Vol. 50, No. 4

  • 1977), and a subsequent series of articles provided a largenumber of groupgroup interaction parameters (Gmehling etal., 1982; Macedo et al., 1983; Hansen et al., 1991; Freden-slund and Srensen, 1994; Wittig et al., 2003). UNIFAC issimple to use because it requires no experimental mixture data;as a result, UNIFAC became immensely popular, despite itslimitations, especially for dilute solutions. Numerous empiricalmodications, primarily by Gmehling and coworkers (includ-ing temperature dependence of some UNIFAC parameters)have improved the ability of UNIFAC to predict activity co-efcients in binary or multicomponent liquid mixtures of typ-ical subcritical liquids, including hydrocarbons, petrochemicalsand water. To illustrate, Figure 7 shows vapor-liquid equilibriafor the benzene(1)-sulfolane(2) system at several temperatureswith a recent set of UNIFAC parameters from Wittig et al.(2003). Figure 8 compares activity coefcients at innite dilu-tion, predicted by UNIFAC with experiment.

    Oishis extension of UNIFAC to polymer solutions (Oishiand Prausnitz, 1978) has been modied by others (for example,Holten-Anderson et al., 1987; Goydan et al., 1989). A fewefforts have been made to extend UNIFAC to solutions con-taining electrolytes; especially for dilute solutions, such exten-sions requires important corrections for the long-range forcesbetween charged particles. Furthermore, such extensions arenecessarily limited because the data base is essentially connedto aqueous systems.

    Efforts to include supercritical components (for example,hydrogen) have not had much success because UNIFAC isbased on a lattice model where each molecule is conned to theimmediate vicinity of a lattice position. A lattice model is notsuitable for a highly mobile gaseous solute. Furthermore, inUNIFAC, the activity coefcient refers to a standard statefugacity of pure liquid at system temperature. For a supercriti-cal component that fugacity is necessarily hypothetical.

    The primary application of UNIFAC is to estimate VLE formulticomponent mixtures of nonelectrolytes for screening andfor preliminary design of distillation or absorption operations.Although UNIFAC can often provide good results, like all

    group-contribution methods, UNIFAC is not always reliable,especially for liquid mixtures where the molecules of one (ormore) components have two or more close-by polar groups (forexample, ethylene glycol). As indicated later, some recentpromising developments with quantum mechanics are directedat reducing this limitation of UNIFAC. Although, UNIFACprovides an attractive method for estimating phase equilibria, itis important to keep in mind that, as yet, there is no substitutefor high-quality experimental data.

    Because ASOG and UNIFAC parameters are obtained frombinary VLE data, predictions from ASOG and UNIFAC areuseful only for VLE, not for LLE. In VLE calculations, theprimary quantities are pure-component vapor pressures; activ-ity coefcients play only a secondary role. However, in LLEcalculations, where activity coefcients are primary, a muchhigher degree of accuracy is required. UNIFAC correlations forLLE (discussed in Poling et al., 2001) are useful only forsemiquantitative predictions. Better results can be achievedwhen UNIFAC LLE parameters are regressed from (and thenapplied to) a limited class of mixtures. For example, Hooper etal., (1988) presented a set of UNIFAC LLE parameters foraqueous mixtures, containing hydrocarbons and their deriva-tives for temperatures between ambient and 200C.

    The DISQUAC group-contribution correlation is useful forestimating enthalpies of liquid mixtures (Kehiaian, 1983,1985).

    A semiempirical class of methods that tries to account for thecorrelation between close-by polar groups uses descriptors. Inthese methods, the structure of a molecule is represented by atwo-dimensional(2-D) graph, with vertices (atoms) and edges(bonds). The numerical values resulting from the operation ofa given descriptor on a graph are related to a physical property,for example, the activity coefcient at innite dilution (Faulonet al., 2003; He and Zhong, 2003). Because the method haslittle theoretical basis, the type and the number of descriptorsneeded is property-dependent, that is, the method is specic foreach thermodynamic property. Furthermore, for reliable pre-dictions, any correlation based on descriptors requires a largedatabase.

    The SPACE model with solvatochromic parameters for es-timating activity coefcients is an extension of regular-solutiontheory where the cohesive energy density is separated intodispersion forces, dipole forces, and hydrogen bonding. Thedipolarity and hydrogen-bond basicity, and acidity parameterswere correlated with the activity coefcient database by Hait etal. (1993). Castells et al. (1999) compare different methods(including SPACE) to calculate activity coefcients for dilutesystems. A method similar to SPACE is described by Abrahamand Platts (2001); their group contribution model is used tocalculate solubilities of several pharmaceutical liquids andsolids in water at 289 K. A recent method for calculatingsolubilities of pharmaceuticals is given by Abildskov andOConnell (2003).

    Equations of state (EOS)In the period 19501975, there were two major develop-

    ments that persuaded chemical engineers to make more use ofMethod II that is, to use an EOS for uid-phase (especiallyVLE) equilibria. First, in the mid-1950s, several authors sug-gested that successful extension of a pure-component equation

    Figure 8. Comparison of activity coefcients at innitedilution predicted by UNIFAC with experimentfor benzene, toluene, cyclohexene, hexane, 2,2, 4-trimethylpentane and undecane in sulfo-lane at several temperatures (Wittig et al.,2003).

    746 AIChE JournalApril 2004 Vol. 50, No. 4

  • of state to mixtures could be much improved by introducingone binary constant into the (somewhat arbitrary) mixing rules,that relate the constants for a mixture to its composition. Forexample, in the van der Waals EOS, parameter a for a mixtureis written in the form

    amixture i


    zizjaij (5)

    where i and j represent components and z is the mole fraction.When i j, van der Waals constant aij is that for the purecomponent. When i j, the common procedure is to calculateaij as the geometric mean corrected by (1 kij) where kij is abinary parameter

    aij aiiajj 1 kij (6)

    Parameter kij is obtained from some experimental data for thei j binary. It seems strange now, but it was not until about19551960 that the currently ubiquitous k12 became a commonfeature of articles in chemical engineering thermodynamics. Itwas during that period that lists of k12 appeared and that(mostly futile) attempts were made to correlate k12 with prop-erties of pure components 1 and 2.

    Second, about 1965, there was a growing recognition thatbecause an EOS of the van der Waals form can be used togenerate both vapor-phase fugacity coefcients or liquid-phase

    activity coefcients, it would be attractive to use one EOS forall uid phases. However, to apply that idea, the pure-compo-nent constants in the equation of state must be evaluated to twhat for VLE is the most important quantity, viz. the pure-component vapor pressure. If an EOS can correctly give thevapor pressure of every pure component in the mixture, VLEfor a mixture can be calculated, essentially, by interpolation asdictated by mixing rules. When these mixing rules are madesufciently exible through one (or sometimes two) adjustablebinary parameters, good results for VLE can often be achieved.To illustrate, Figure 9 shows experimental and calculated re-sults for methane-propane. Although k12 is very small com-pared to unity, it nevertheless has a signicant effect.

    Because methane and propane are simple and similar mol-ecules, a single binary parameter is sufcient to achieve goodresults. To represent phase equilibria for the much more com-plex system, water-hydrogen sulde, Evelein et al. (1976)introduced a second parameter c12 in the mixing rule for vander Waals size-parameter b. With two binary parameters, it ispossible to achieve good agreement with experiment as shownin Figure 10.

    As a result of these happy developments, the literature wassoon ooded with proposed EOS where the pure-componentconstants were t to pure-component vapor pressure data. Inthis ood, a favorite target was to modify the RedlichKwong(RK) EOS, published in 1949, where the authors had intro-duced a simple but remarkably effective modication of thedensity dependence in the van der Waals equation. BecauseRedlich and Kwong were concerned only with dense gases, notliquids, their particular temperature dependence was dictatedby second-virial coefcient (not vapor pressure) data. Afterabout 1972, numerous articles reported modications of theRK EOS where, for each pure uid, the characteristic attractiveconstant a is given as a function of temperature such that goodagreement is obtained with experimental vapor-pressure data.The best known modication of the RK EOS is that by Soave(1972) who was one of the rst to show that a simple EOS of

    Figure 9. Isothermal pressure-composition phase dia-gram for methane-propane.Calculation with the Redlich-Kwong-Soave equation of state(Prausnitz et al., 1999). Parameter k12 was adjusted to givegood agreement with experimental data from Reamer et al.(1950).

    Figure 10. Isothermal pressure-composition phase dia-gram for water-hydrogen sulde.Calculation with the Redlich-Kwong-Soave equation ofstate (Evelein et al., 1976). Parameters k12 and c12 wereadjusted to give good agreement with experimental datafrom Selleck et al. (1952).

    AIChE Journal 747April 2004 Vol. 50, No. 4

  • the van der Waals form is useful for calculating VLE of avariety of mixtures at both moderate and high pressures.

    In 1976, Peng and Robinson (PR) published their modica-tion of the van der Waals EOS (Peng and Robinson, 1976) that,unlike Soaves modication (SRK), introduces a new densitydependence in addition to a new temperature dependence intothe RK equation. Although Soaves equation and the PR equa-tion necessarily (by design) give good vapor pressures, the PREOS gives better liquid densities. The PR EOS and the SRKEOS are now the most common working horses for calcu-lating high-pressure VLE in the natural-gas, petroleum andpetrochemical industries. For application of the PR EOS tomixtures containing polar as well as nonpolar components, aparticularly useful correlation is that given by Vera and Stryjek(1986). For mixtures where one (or more) components are wellbelow their normal boiling points, a useful modication is thatby Mathias and Copeman (1983).

    The van der Waals EOS is a perturbation on a highlyoversimplied model for hard spheres; the perturbation is in-tended to account for attractive forces, although, in effect, italso corrects the oversimplied hard-sphere term. Equations ofthe van der Waals form can be improved with a more realistichard-sphere model or by changing the density dependence ofthe perturbation term or both. Many efforts along these lineshave been reported. However, except for the Soave and PRequations, they have not enjoyed much success in chemicalprocess-design calculations essentially for two reasons: rst,because for some improved EOS, additional constants mustbe determined from some experimental physical property thatmay not be readily available, and second, because the calcula-tions are often more tedious if the EOS is not (unlike the PRand Soave EOS) cubic in volume; in that event, iterations areneeded to nd the fugacity coefcient when temperature, pres-sure and composition are given. Although standard computerscan easily perform such iterations, when very many VLEcalculations are required for a particular design, impatientengineers prefer an EOS where calculations are relatively sim-ple and fast.

    Because all analytic EOS based on the van der Waals modelare poor in the vapor-liquid (VL) critical region, some attemptshave been reported to obtain improvement by translation(Peneloux et al., 1982; Mathias et al., 1989), that is, by addinga correction (in the rst approximation, a constant) to thevolume in the EOS; this correction horizontally moves (trans-lates) the VL coexistence curve plotted on PV coordinates.The correction is designed to make the calculated criticalcoordinates (VC, TC, PC) agree with experiment. This procedurehas been used to calculate better saturated liquid volumes,however, because the correction toward that end introduceserrors elsewhere in the PV plane, this excessively empiricaltranslation procedure has only limited application (Valderrama,2003).

    Calculation of VLE in the critical region provides a severechallenge because any attempt to do so along rigorous lines(renormalization group theory) requires numerous approxima-tions and much computer time. Although some efforts havebeen reported toward better representation of VLE in the crit-ical region (for example, Anisimov and Sengers, 2000; Lue andPrausnitz, 1998; Jiang and Prausnitz, 1999; Kiselev et al.,2001; Kiselev et al., 2002), as yet, they are not sufcientlydeveloped for general engineering use.

    Toward further improvement of EOS for VLE, during the1980s signicant attention was given to establish better mix-ing rules. Because the EOS can be used to nd the molar excessGibbs energy, Vidal (1978, 1983) suggested that experimentaldata for liquid-mixture activity coefcients be used to deter-mine both mixing rules and binary constants that appear inthose rules. The resulting EOS can then be used to generateVLE at temperatures and pressures beyond those used to x theconstants that appear in the mixing rules.

    Unfortunately, the procedure suggested by Vidal (and oth-ers) leads to a thermodynamic inconsistency because the cor-responding mixing rules, in general, do not give the theoreti-cally correct quadratic composition dependence of the secondvirial coefcient. To avoid this problem, it was suggested thatthe mixing rules should be modied such that they simulta-neously reproduce the activity-coefcient data at high uiddensities, and at low uid densities, the correct compositiondependence of the second virial coefcient. The most success-ful of these suggestions is the Wong-Sandler (WS) mixingrules (Wong and Sandler, 1992). Subsequent experience withthese rules led to a monograph that gives details and pertinentcomputer programs (Orbey and Sandler, 1998). Figure 11shows a successful application of WS mixing rules.

    In another example, shown in Figure 12, the PR EOS withWS mixing rules was used to calculate LLE. Here, Escobedo-Alvarado and Sandler (1998) used experimental LLE data atlow pressure to determine binary parameters in the WS mixingrules. The PR EOS was then used to predict LLE at higherpressures up to 781 bar.

    The phenomenological basis of a traditional EOS of the vander Waals form is based on molecules that are spherical (orglobular). That basis is not appropriate for uids with chain-like molecules, especially polymers, not only because such

    Figure 11. Isothermal pressure-composition phase dia-gram for 2-propanol/water.Calculation with the PR EOS as modied by Stryjek andVera (1986), and Wong-Sandler mixing rules (Wong andSandler, 1992) with k120.326 (Orbey and Sandler, 1998).

    748 AIChE JournalApril 2004 Vol. 50, No. 4

  • molecules are not spherically symmetric in shape, but alsobecause such molecules (unlike small spheres) exercise rota-tions, and vibrations that, because they depend on density, mustbe included in a suitable EOS (Vera and Prausnitz, 1972).

    A phenomenological EOS for chain-like molecules

    (PHCT perturbed hard chain theory) was proposed in 1975(Beret and Prausnitz, 1975) based on Prigogines theory forliquid polymers (Prigogine, 1957); a particularly simple formof PHCT is the cubic EOS of Sako et al. (1989). In essence,PHCT is similar to the van der Waals model but, unlike thatmodel (in its original form), it allows for contributions fromso-called external degrees of freedom (density-dependent rota-

    Figure 13. Calculated and experimental hexane (weight-fraction) solubility in a polypropylene copoly-mer.Here, expt means experimental data. Calculations with theperturbed-hard-sphere EOS with Wong-Sandler mixingrules or with the SAFT EOS (Feng et al., 2001).

    Figure 14. Experimental and calculated (mole-fraction)solubility of water in the ethanerich phase forwater-ethane mixtures (Huang and Radosz,1991, 1993).

    Figure 12. Calculated and experimental LLE for the 2-butoxyethanol (C4E1)/water system, with the Peng-RobinsonEOS and Wong-Sandler-NRTL mixing rules where x is the mole fraction.Low-pressure LLE data were used to determine binary parameters in the EOS (Escobedo-Alvarado and Sandler, 1998).

    AIChE Journal 749April 2004 Vol. 50, No. 4

  • tions and vibrations) in addition to translations to the EOS. AnEOS similar to PHCT is the chain-of-rotators EOS (Chien etal., 1983; Kim et al., 1986). Variations of PHCT have been

    used extensively in representing phase equilibria for vapor-phase polymer-monomer-solvent mixtures at high pressure asused, for example, in the production of polyethylene (Feng etal., 2001), as shown in Figure 13.

    When attention is limited to the liquid phase, an EOS byFlory (1965, 1970), Eichinger and Flory (1968), and Patterson(1969) can be used to nd the so-called equation-of-statecontribution to the excess Gibbs energy of a polymer solution.These contributions that arise because of differences in freevolume of the polymer solutions components are neglected inthe classical FloryHuggins lattice theory for polymer solu-tions. These contribution are essential for explaining the oftenobserved high-temperature lower critical-solution temperatureof polymer solutions and polymer blends (Olabisi et al., 1979).An alternative method by Sanchez and Lacombe (1976, 1978)proposed a relatively simple EOS for polymers that is anextension of FloryHuggins lattice theory. The essential con-tribution of this equation is the inclusion of holes (empty sites)into the lattice. This inclusion provides the communal entropyat the ideal gas limit. Smirnova and Victorov (2000) give acomprehensive review of EOS based on lattice models.

    An elegant theory by Wertheim (19841986) has led toSAFT (Chapman et al., 1989 and 1990), a theoretically well-founded EOS for chain-like molecules where SAFT stands forStatistical Association Fluid Theory. Muller and Gubbins(2001) have given a user-friendly review of the derivation andapplications of SAFT. To illustrate, Figure 14 shows the ap-plication of SAFT to represent properties of a solution contain-ing a strongly associating component, and Figures 15 showscalculated and experimental phase equilibria for a solutioncontaining components with large size difference.

    Figure 15. Experimental and calculated solubility of bi-tumen in compressed carbon dioxide (Huangand Radosz, 1991).

    Figure 16. Comparison of experimental and calculated cloud points with the perturbedchain-SAFT EOS for mixturesof polyethylene and several solvents: ethylene, ethane, propylene, propane, butane, and 1-butene (Grosset al., 2003).

    750 AIChE JournalApril 2004 Vol. 50, No. 4

  • On the basis of some clever assumptions concerning thestructure of a chain-molecule uid, Chiew (1990, 1991) hasderived an EOS for polymer and polymer-like uids that, ineffect, is similar to SAFT. By considering the next-to-nearest-neighbor cavity correlation function, Hu et al. (1995) obtaineda more accurate EOS for hard-sphere chain uid mixtures in

    good agreement with computer simulation data. In otherSAFT-related studies, a number of successful efforts haveincorporated a more accurate description of the structure andfree energy of the monomer reference uid (Ghonasgi andChapman, 1994; McCabe et al., 2001, Paredes et al., 2001,Paricaud et al., 2002). For example, Gross and Sadowski(2001) applied second-order Barker-Henderson perturbationtheory to a hard-chain reference uid to improve the SAFTequation of state. To illustrate, Figures 16 and 17 show calcu-lated and experimental phase equilibria for some polymer andcopolymer solutions, respectively.

    In addition to poor agreement with experiment in the criticalregion, all currently available EOS for chain-like moleculessuffer from one limitation: for long chains, at dilute conditions,the calculated second virial coefcient has an incorrect depen-dence on chain length. As a result, SAFT (and similar) EOSgive unreliable results for the dilute polymer phase when usedto calculate uiduid equilibria when the polymer is concen-trated in one phase and dilute in the other.

    For strongly associating components it is tempting to super-impose a chemical theory onto a physical EOS. The gen-eral procedure for doing so was shown by Heidemann andPrausnitz (1976) who introduced a chemical association equi-

    Figure 17. Comparison of experimental and calculatedcloud points with the perturbedchain-SAFTEOS for mixtures of poly(ethylene-co-1-butene) and propane with varying repeat-unitcomposition (from 0 to 97 mass %) of1-butene in the copolymer (Gross et al.,2003).% PB represents the percentage of 1-butene in the polymer.

    Figure 18. Calculated and experimental VLE for sulfurdioxide (1)-propane (2) at 50 C.Full lines are for calculations with an optimized binaryinteraction parameter (k12), whereas dotted lines are fork120. Here x is mole fraction, and the points are experi-mental data.Calculations with the Association-plus-PR EOSof Anderko (1989).

    Figure 19. Calculated and experimental Pxy diagram forthe hydrogen uoride/refrigerant R12 sys-tem.Lines are for the Association-plus-PR EOS proposed byVisco and Kofke (1999). Solid lines use the van der Waalsmixing rules, whereas the dashed lines use the Wong-San-dler mixing rules (Baburao and Visco, 2002).

    AIChE Journal 751April 2004 Vol. 50, No. 4

  • librium constant into the van der Waals EOS. Following somereasonable simplications, an analytic EOS was derived withtwo physical interaction constants, van der Waals a and b, andone temperature-dependent equilibrium constant, K (T). On thebasis of similar ideas, several authors have represented theproperties of solutions containing one associating componentand one or more normal uids; an example is shown in Figure18 by Anderko (1989). Another example for particularlynasty mixtures containing hydrogen uoride is shown inFigure 19.

    Equations of state have been used extensively to designsupercritical-extraction processes (McHugh and Krukonis,1994). Because supercritical extraction uses high pressures andbecause of the large difference in size and shape between soluteand solvent, these systems present a variety of phase diagrams.For example, binary mixtures of ethane-linalool and ethane-limonene present double retrograde condensation, as discussedby Raeissi and Peters (2002, 2003). For supercritical-extractionconditions, calculation of phase equilibria with an EOS, and theequifugacity criteria may present multiple roots. The correctroot should be selected by a global phase-stability method. Ina series of articles, Brennecke and Stadtherr (Xu et al., 2000;Maier et al., 2000; Xu et al., 2002) used an EOS to obtainhigh-pressure solid-uid equilibria. To illustrate, Figures 20and 21 show the solubility of biphenyl in CO2 at differentpressures calculated with two methods: equifugacity and globaloptimization. Figure 21 shows that, at 333.15K the equifugac-ity calculation gives an unstable root at some pressures.

    Most supercritical-extraction processes use carbon dioxideas the solvent because of its attractive environmental proper-ties. Because many compounds have very low solubilities incarbon dioxide, even at high pressures, a supercritical-extrac-tion process may require an undesirable high solvent ow rate.To increase solubilities dramatically, it has been suggested toadd a surfactant that can form micelles of the extracted com-pounds in the solvent. However, ordinary surfactants (intended

    for oil-water systems) are not effective in dense carbon diox-ide. In a signicant contribution that shows the importance ofchemistry in applied thermodynamics, Beckman (Ghenciu etal., 1998; Sarbu et al., 2000) synthesized entirely new surfac-tants suitable for micellization in dense carbon dioxide.

    Equations of state can be used to describe adsorption of puregases and their mixtures on solid surfaces. An early discussionwas given by Van Ness (1969); a more recent one is by Myers(2002). Here, the parameters of the EOS depend not only on theadsorbate but also on the adsorbent, that is, the solid surfaceprovides an energetic eld that affects the forces between theadsorbed molecules.

    Equations of state are useful for design of crystallizationprocesses where it is important that the precipitated solids havea narrow size distribution (Chang and Randolph, 1989, 1990),and for calculating hydrate formation in moist natural gases asdiscussed in the monograph by Sloan (1990).

    Equations of state are also useful for describing particleprecipitation from liquid solution. For that case, it is convenientto write the EOS in terms of the McMillanMayer framework(1945) where the potentially precipitating particles are dis-solved in a continuous liquid medium. The pressure is replacedby the osmotic pressure; the density now is not the density ofthe system but that of the particles in the liquid medium.Although this type of EOS has been used for many years incolloid science, it has received only little attention from chem-ical engineers in the conventional chemical industries. Ogston(Laurent and Ogston, 1963; Edmond and Ogston, 1968) hasused an EOS in the McMillan-Mayer framework to describedilute solutions of two (or more) polymers, while Wu et al.(1998, 2000) have shown how an EOS of this type can be usedto describe precipitation of asphaltenes from heavy petroleum.

    Regardless of what EOS is used, perhaps the most importantengineering application of an EOS lies in the estimation of

    Figure 21. PR EOS calculation of the solubility of biphe-nyl in CO2 at 60 C.At this temperature, there are multiple equifugacity roots forpressures below 160 bar. At low pressure, the lowest solu-bility root corresponds to stable equilibrium, but at higherpressure, it is the highest solubility root that is stable. Athree-phase line, indicating solid-liquid-vapor coexistence,occurs at about 45.19 bar (Xu et al., 2000). Here y is molefraction.

    Figure 20. PR EOS calculation of the solubility of biphe-nyl in CO2 at 35 C.At this temperature, there is only one equifugacity root ateach pressure, and that root corresponds to stable solid-uidequilibrium (Xu et al., 2000). Here y is the mole fraction.

    752 AIChE JournalApril 2004 Vol. 50, No. 4

  • multicomponent VLE with only parameters obtained from sin-gle-component and binary experimental data (or from correla-tions based on such data). In an EOS of the van der Waalsform, we consider only two-body interactions. Therefore, oncewe have properly extended that EOS to a binary mixture, nofurther assumptions are required to achieve extension to ternary(and higher) mixtures. In a mixture, two-body interactions arereected by mixing rules that are quadratic in composition. Ifmixing rules use higher-order terms, extension from binary toternary (and higher) mixtures presents signicant theoretical problems as noted by Michelsen and Kistenmacher (1990) and

    by Mollerup and Michelsen (1992).Extensive experience in applying an EOS to calculate uid-

    phase equilibria has shown that for typical uid mixtures, therole of details in the EOS itself or in its mixing rules is lessimportant than that of the choice of constants obtained fromsome experimental source.

    Electrolyte Solutions

    Because electrostatic forces between ions are long-range, thephysical chemistry of electrolyte solutions is qualitatively dif-ferent from (and more difcult than) that for solutions ofnonelectrolytes. Because electric neutrality must be main-tained, the concentrations of cations and anions are not inde-pendent. As a result, conventional experimental thermody-namic data for electrolyte solutions do not give the activity ofthe cation and that of the anion but instead, a mean ionicactivity coefcient. Furthermore, because salts are not volatileat ordinary temperatures, mean ionic activity coefcients refernot to the pure electrolyte but to an ideal dilute solution of theelectrolyte in the solvent.

    For very dilute solutions of strong electrolytes (completedissociation into ions), we have the DebyeHuckel (DH) the-ory of 1923; this theory gives the mean ionic activity coef-cient arising from electrostatic ionion forces in a medium ofknown dielectric constant. In its rigorous, highly dilute limit,DH theory neglects the nite sizes of the ions and van der

    Figure 22. Calculated and experimental solubilities ofsalts in the ternary system NaCl/KCl/H2O atseveral temperatures.Intersections of isothermal curves represent calculated ter-nary invariant points where three phases are in equilibrium:pure NaCl solid, pure KCl solid, and aqueous solutioncontaining both salts (from Prausnitz et al., 1999). Here, m mobility.

    Figure 23. Calculated and experimental water activitiesat different electrolyte concentrations forvarious aqueous sodium carboxylates: meth-anoate (C1), ethanoate (C2), propanoate (C3),butanoate (C4), pentanoate (C5), hexanoate(C6), heptanoate (C7),octanoate (C8), non-anoate (C9), and decanoate (C10).Calculations with a gE model that combines NRTL for ionicsystems with NRTL for oligomers are able to describe theabrupt change in the water activity at the critical micelleconcentration (Chen et al., 2001).

    Figure 24. Solubility of Zn(NO3)2 in water as a function oftemperature.In addition to the anhydrous salt, ve hydrates are formed inthe solution containing zinc nitrate: nonahydrate, hexahy-drate, tetrahydrate, dihydrate, and monohydrate. Calcula-tions using a gE model that combines UNIQUAC with theDebyeHuckel theory are able to describe the ve eutecticpoints and one peritectic point (P) of the aqueous Zn(NO3)2system (from Iliuta et al., 2002).

    AIChE Journal 753April 2004 Vol. 50, No. 4

  • Waals attractive forces between ions. To describe the thermo-dynamic properties of concentrated electrolyte solutions, nu-merous phenomenological extensions of DH theory have beenpresented. Perhaps the most successful is the one by Pitzer(1973, 1995) which, in effect, expresses the excess Gibbsenergy (relative to an ideal dilute solution) as a sum of twoparts: the rst part is based on a slightly modied DH theory,

    and the second part is, essentially, an osmotic virial series inelectrolyte concentration. Regrettably, this power series re-quires several system-specic coefcients that depend on tem-perature. However, because we have a large body of experi-mental results for aqueous salt solutions over a reasonabletemperature range, we now have a fair inventory of Pitzerparameters. To extend Pitzers model to multisalt solutions, itis necessary to make some simplifying assumptions or else tointroduce one or more ternary parameters. Pitzers model hasbeen applied toward optimizing process design for salt recov-ery from Trona mines as discussed by Weare (Harvey et al.,1984), and for designing a recovery process for radioactivesalts from aqueous solutions (Felmy and Weare, 1986).

    To illustrate Pitzers theory for an aqueous system contain-ing two salts (potassium chloride and sodium chloride), Figure22 shows experimental and calculated salt solubilities in theregion 0 to 200 C.

    An alternate model for activity coefcients in aqueous elec-trolyte solutions was developed by Chen et al. (1982, 1986,2001) who used the NRTL equation to account for ionion andionsolvent interactions beyond those given by DH theory. Anadvantage of Chens model is that, in at least some cases, itrequires fewer binary parameters than Pitzers model. Figure23 shows an application of Chens model to aqueous organicelectrolytes. A similar theory, based on the UNIQUAC equa-tion, was presented by Iliuta et al. (2002) who gave particularattention to solubilities of heavy-cation salts. Figure 24 showscalculated and observed solubilities of various hydrates ofZn(NO3)2.

    When augmented by chemical equilibria, a much simpliedversion of Pitzers model has been used by Edwards (Edwardset al., 1978) to correlate multicomponent vapor-liquid equilib-ria for aqueous solutions of volatile electrolytes (NH3, H2S,CO2, SO2) that are frequently encountered in chemical pro-

    Figure 25. Simultaneous solubilities of NH3 and SO2 inwater at 100 C, calculated by Edwards et al.(1978); experimental results are from Rumpfet al. (1993).

    Figure 26. Calculated and experimental mean ionic ac-tivity coefcients of KCl in mixtures of meth-anol and water at 25 C: pure water (), 90 wt.% (salt-free) water (), 80 wt. % water (), 60wt. % water (), 40 wt. % water (E), 20 wt. %water (F), 10 wt. % water (), 0 wt. % water() (Papaiconomou et al., 2002).

    Figure 27. Calculated and experimental partition coef-cients for three dilute proteins in an aqueoustwo-phase system containing PEG 3350, dex-tran T-70, and 50 mM KCl (overall) at pH 7.5and 25 C.The partition coefcient is dened by the ratio of the proteinmolalities. Experimental and calculated results are for albu-min (circles), chymotrypsin (squares), and lysozyme (trian-gles) (Haynes et al., 1991).

    754 AIChE JournalApril 2004 Vol. 50, No. 4

  • cesses. To illustrate, Figure 25 shows the total pressure as afunction of SO2 concentration for aqueous mixtures of SO2 andNH3 at 100C.

    The early theory by Edwards has been much improved byapplying the full Pitzer theory. Maurer and coworkers havepresented an extensive correlation of VLE for aqueous solu-tions of weak electrolyte gases with or without selected addedsalts (Rumpf et al., 1993; Bieling et al. 1995, Kamps et al.,2002, Kamps et al., 2003).

    Integral-equation theory can be used to establish a theoreti-

    cal basis for describing electrolyte solutions as discussed byPapaiconomou et al. (2002). To account for electrostatic andfree-volume interactions of ions in solution, including the con-centrated region, these authors used the integral theory ofsolution coupled with Blums mean spherical approximation;the effects of van der Waals attractive forces are provided by anequation similar to NRTL. This theory is readily applicable toa salt in a solvent mixture; to illustrate, Figure 26 showscalculated and observed mean-ion activity coefcients for KClin methanolwater mixtures at 25C.

    Figure 28. Comparison of Monte Carlo simulations with experimental data for adsorption of binary mixtures ofpropane (1)- H2 S(2) on H-mordenite at 8.13 kPa and 30C.Parameters were obtained from each pure-component adsorption isotherm. (a) gives the phase diagram, and (b) gives the total amount of gasadsorbed (Cabral et al., 2003).

    AIChE Journal 755April 2004 Vol. 50, No. 4

  • Integral-equation theory can also be used to establish anEOS for describing electrolyte solutions as discussed by Jinand Donohue (1988, 1988), Wu and Prausnitz, (1998), andMyers et al. (2002).

    Depending on pH, proteins carry an electric charge.Therefore, a description of the thermodynamic properties ofprotein solutions must include electrostatic effects in addi-tion to van der Waals forces, as discussed, for example, inAlbertssons (1986) book on separation of protein mixturesby extraction in aqueous two-phase systems. Such systemsare formed upon dissolving two water-soluble polymers, forexample, dextran and poly(ethylene glycol) (PEG). A mea-sure of how these two aqueous phases differ is provided bythe length of a tie line on a plot where the percent PEG inone phase is plotted against the percent dextran in the other.Figure 27 shows calculated and observed distribution coef-cients for three dilute proteins in a two-phase aqueoussystem as a function of the difference between the twophases (Haynes et al., 1991). Because this system alsocontains a small amount of KCl that partitions unequallybetween the two aqueous phases, and because the charges onthe three proteins are not identical, the distribution coef-cients differ widely, facilitating protein separation.

    Molecular Simulations

    An alternative to algebraic expressions for activity coef-cients or equations of state is provided by molecular simula-tions as discussed in several textbooks, notably that by Sadus(1999) and that by Frenkel and Smit (2002).

    Molecular simulations are attractive because they requireas input only quantitative data for molecular structure, andfor the potential of moleculemolecule interaction. The dis-advantage of molecular simulations is that results are re-stricted to a particular case; these results are not easilygeneralized.

    The last 15 years have produced a large number of articlesshowing how molecular simulation can be used to calculatephase equilibria for a large variety of systems based on theGibbsensemble method of Panagiotopoulos (1987). Although

    numerous authors have contributed to this important develop-ment, particularly noteworthy are the articles by Cummingsand coworkers (for example, McCabe et al., 2001; Rivera et al.,2003) and those by de Pablo et al. (for example, Nath et al.,1998; Yan and de Pablo, 2001; de Pablo and Escobedo, 2002;Jendrejack et al., 2002). To illustrate applicability to processdesign, Monte Carlo simulations can be used to describe theadsorption of pure gases and their mixtures on solid surfaces asdiscussed, for example, by Smit and Krishna (2003), and bySteele (2002). To illustrate, Figure 28 shows calculated andexperimental gas-solid adsorption equilibria for mixtures ofpropane (1)- H2S (2) on H-mordenite (Cabral et al., 2003).

    Because the results from molecular simulations are sensitiveto the potential function that describes intermolecular forces, itis necessary to obtain that potential function from the reductionof some experimental data. For some cases, the extent ofrequired experimental data can be reduced by quantum me-chanics.

    Application of Quantum Mechanics

    One of the most promising recent developments in chemicalengineering thermodynamics is provided by applying quantummechanics for calculating thermodynamic properties, in partic-ular, activity coefcients of components in liquid mixtures.Quoting from Sandlers review (Sandler, 2003):

    In the most direct and computational intensive form, com-putational quantum mechanicsis used to obtain information onthe multidimensional potential energy surface between mole-cules, which is then used in computer simulation to predictthermodynamic properties and phase equilibria. At present, thismethod is limited to the study of small molecules because ofthe computational resources available. The second method ismuch less computationally intensive and provides a way to

    Figure 29. Calculated and experimental VLE phase dia-grams for acetronitrile(1)-methanol(2) at60.31 C.Phase diagrams show (a) vapor vs. liquid mole fraction and(b) pressure vs. vapor and liquid mole fraction. Filled circlesare experimental data (DECHEMA), and open squares anddiamonds are predictions from Gibbs-ensemble MonteCarlo NPT and NVT simulations, respectively; the solid lineis the best t of the experimental data (Sum et al., 2002).

    Figure 30. Calculated and experimental VLE for ben-zene(1)-N-methyl formamide(2) at 45 and 55C; calculation with the COSMO-SAC model(Lin and Sandler, 2002).Circles and triangles are experimental.

    756 AIChE JournalApril 2004 Vol. 50, No. 4

  • improve group-contribution methods by introducing correc-tions based on the charge and dipole moment of each functionalgroup that is unique to the molecule in which it appears. Thethird method is based on the polarizable continuum model, inwhich the free energy of transferring a molecule from an idealgas to a liquid solution is computed, leading directly to valuesof activity coefcients and phase equilibrium calculations.

    For typical polar molecules, such as acetonitrile or methyluoride, it is now possible to establish a reliable two-bodypotential that depends on all distances between the atoms ofone molecule and those of the other. In some cases, the poten-tial can be simplied by considering only the distance betweenthe center of mass of one molecule and that of the other inaddition to angles of orientation. For polar molecules havingless than (about) 100 electrons, knowing the geometric andelectronic structures of the molecules is sufcient to establishthe two-body potential; for more difcult cases (for example,methanol), a well-measured thermodynamic property (typicallythe second virial coefcient) is used to augment results ob-tained from quantum mechanics.

    For a binary mixture containing components 1 and 2, weneed three two-body potentials: one each for 11, 22 and 12interactions. Some, or perhaps all of these potentials may beobtained from quantum mechanics. These potentials are thenused in a Monte Carlo-simulation program to generate vapor-liquid or other phase equilibria. Although this promising typeof calculation is likely to see increasing popularity, at present,for industrial application, it suffers from two disadvantages:typical simulation calculations are limited by the additivityassumption (the total potential energy of a system is given bythe sum of all two-body interactions), and by insufcientlypowerful computers. Although corrections for nonadditivityare not simple, they are often signicant, especially for hydro-gen-bonding systems. A highly computer-intensive method forcalculating multibody potentials is provided by Car and Par-rinello (1985), but as yet this method is not sufciently sensi-tive for application to mixtures of ordinary liquids (Trout,2001).

    Figure 29 shows a successful application of quantum me-chanics-plus-Monte Carlo simulation for vapor-liquid equilib-ria for methanol-acetonitrile at 333.46 K (Sum et al., 2002). Itis remarkable that, although no mixture data were used togenerate Figure 29, the calculations give the correct pressureand composition of the azeotrope.

    More than a century ago, Mossotti derived an equation forthe change in energy experienced by a dipolar molecule whenit is transferred from an ideal gas into a continuous liquidmedium characterized by its dielectric constant (Israelachvili,1992). Similarly, more than eighty years ago, Born indicatedhow the free energy of a charged molecule changes when itgoes from one dielectric medium to another (Israelachvili,1992). In the same spirit, but with more powerful physics,Klamt and coworkers (Klamt, 1995; Klamt and F. Eckert,2000) have developed a method for calculating the activitycoefcient of a solute dissolved in a continuous polarizablemedium. This method does not use functional groups but usessurface charges for atoms that depend not only on the particularatom, but also on the identity of other atoms in the samemolecule. Thus, Klamts method, in effect, overcomes one ofthe serious limitations of UNIFAC. Klamts method is attrac-tive for engineering because computational requirements are

    relatively low. However, at present this method is limited toactivity coefcients of solutes in dense liquids, that is, liquidswell below their critical temperatures; it is not (yet) applicableto gaseous mixtures or to low-density liquid mixtures encoun-tered in the vapor-liquid critical region. Figure 30 shows asuccessful example of Klamts COSMOSAC model for vapor-liquid equilibria in the benzeneNmethyl formamide system(Lin and Sandler, 2002)

    Figure 31. Comparison of calculated and experimentaldistribution coefcients for a large number ofsolutes distributed between water (W) andoctanol (O) at high dilution, near room-tem-perature.The partition coefcient is dened by the ratio of solutemolar concentrations (mol/L). Crosses are for monofunc-tional molecules, and circles are for multifunctional mole-cules (Sandler, 2003).

    AIChE Journal 757April 2004 Vol. 50, No. 4

  • Quantum Mechanics for Group-ContributionParameters

    The popularity of UNIFAC (and other group-contributionmethods) has encouraged numerous authors toward seekingimprovements that overcome some of UNIFACs well-knownlimitations. Perhaps the most important limitation of UNIFACis its neglect of neighbor effects; in UNIFAC, the interactionbetween a functional group X and a functional group Y isassumed to be independent of the identities of whatever func-tional groups are bonded to X or Y. For example, in UNIFAC,a chloride group in say, CH3 CH2Cl CH3, is equivalent tothat in say, CH3 CH2Cl CH2OH. With quantum mechan-ics, it is now possible to correct UNIFAC groupgroup inter-action parameters for the proximity effect because of neigh-boring bonded groups. For molecules that contain only onepolar functional group, proximity corrections are not large.However, for molecules that contain two or more polar func-tional groups, proximity corrections are often signicant, es-pecially if two polar functional groups are in close proximity asfound, for example, in biomolecules and pharmaceuticals. Toillustrate, Figure 31 presents calculated and experimental dis-tribution coefcients for a large number of dilute solutes dis-tributed between water (W) and octanol (O) near room-tem-perature (Lin and Sandler, 2000; Sandler, 2003). Part (a) ofFigure 31 shows UNIFAC calculations without proximity cor-rections, whereas Part (b) shows calculations with quantum-mechanical proximity corrections. For monofunctional mole-cules there is little difference; however, for multifunctionalmolecules, proximity corrections produce a large improvementin agreement with experiment.

    There is good reason to believe that, as computer speed rises,we will see increasing use of molecular simulations, and in-creasing use of quantum mechanics for the calculation ofthermodynamic properties. It is likely that extensive use ofrigorous ab initio calculations is still in the indenite future.However, it is now clear that the time is ripe for with molecularsimulations and quantum mechanics to extend and improvecurrent methods for calculating phase equilibria.


    This brief and unavoidably incomplete survey is limited to50 years of progress in applications of thermodynamics formore-or-less classical operations in conventional chemical en-gineering. That progress follows primarily from two mutually-supporting fortunate developments since 1950: rst, increasingavailability of ever more powerful computers, and second,increasing willingness of chemical engineers to base correla-tions and design procedures on insight from physical chemis-try, molecular physics and statistical mechanics. As chemicalengineering expands into a variety of new (high-tech) areas, itis clear that these two developments will provide the essentialbackground for future application of thermodynamics in chem-ical engineering science and practice.

    AcknowledgmentsThe authors are grateful to John OConnell, Juan Vera, Ying Hu, and

    Chau-Chyun Chen for helpful comments. We dedicate this work to Prof.Affonso Carlos Seabra da Silva Telles, an outstanding academic leader inBrazil, who pioneered graduate programs and contribution to thermodynamicsand other topics in Brazilian chemical engineering. For nancial support, the

    authors are grateful to the National Science Foundation, to the Ofce for BasicSciences of the U.S. Dept. of Energy, to the Donors of the Petroleum ResearchFund administered by the American Chemical Society, and to the BrazilianMinister of Education, CAPES/Brazil, for grants BEX 0621/021.

    Literature CitedAbildskov, J., and J. P. OConnell, Predicting the Solubilities of Complex

    Chemicals I. Solutes in Different Solvents, Ind. Eng. Chem. Res., 42,5622 (2003).

    Abraham, M. H., and J. A. Platts, Hydrogen Bond Structural GroupConstants, J. Org. Chem., 66, 3484 (2001).

    Abrams, D. S., and J. M. Prausnitz, Statistical Thermodynamics of LiquidMixtures: a New Expression for the Excess Gibbs Energy of Partly orCompletely Miscible Systems. AIChE J., 21, 116 (1975).

    Albertsson, P.-., Partition of Cell Particles and Macromolecules, 3rd ed.,Prentice Hall PTR, N.J. (1986).

    Anderko, A., Calculation of Vapor-Liquid Equilibria at Elevated Pres-sures by Means of an Equation of State Incorporating Association,Chem. Eng. Sci., 44, 713 (1989).

    Anisimov, M. A., and J. V. Sengers, Critical Region, Equation of Statefor Fluids and Fluid Mixtures. Experimental Thermodynamics, Vol. 5.J. V. Sengers, R. F. Kayser, C. J. Peters, and H. J. White Jr., eds.,Elsevier, Amsterdam (2000).

    Asher, W. E., J. F. Pankow, G. B. Erdakos and J. H. Seinfeld, Estimatingthe Vapor Pressures of Multi-Functional Oxygen-Containing OrganicCompounds Using Group Contribution Methods, Atmospheric Envi-ronment, 36, 1483 (2002).

    Baburao, B., and D. P. Visco, VLE/VLLE/LLE Predictions For HydrogenFluoride Mixtures Using An Improved Thermodynamic Equation OfState, Ind. Eng. Chem. Res., 41, 4863 (2002).

    Behrens, D. and R. Eckermann, Chemistry Data Series, DECHEMA, Frankfurta.M., Vol. I, (subdivided into nineteen separate volumes) VLE Data Col-lection by J. Gmehling, U. Onken, W. Arlt, P. Grenzheuser, U. Weidlich,and B. Kolbe (19801996);Vol II, Critical Data by K. H. Simmrock(1986); Vol III, (subdivided into four volumes) Heats of Mixing DataCollection by C. Christensen, J. Gmehling, P. Rasmussen, and U. Weidlich(19841991); Vol V, (subdivided into four volumes) LLE-Data Collectionby J. M. Sorensen and W. Arl (19791987); Vol VI, (subdivided into fourvolumes) VLE for Mixtures of Low-Boiling Substances by H. Knapp, R.Doring, L. Oellrich, U. Plocker, J. M. Prausnitz, R. Langhorst and S. Zeck(19821987); Vol VIII, Solid-Liquid Equilibrium Data Collection by H.Knapp, R. Langhorst and M. Teller, 1987; Vol IX,(subdivided into fourvolumes) Activity Coefcients of Innite Dilution by D. Tiegs, J. Gmehling,A. Medina, M. Soares, J. Bastos, P. Alessi, and, I. Kikic, 19861994; VolXII, (subdivided into nine volumes) Electrolyte Data Collection by J.Barthel, R. Neueder, R. Meier et al., 19921997.

    Benedict, M., G. B. Webb, and L. C. Rubin, An Empirical Equation forThermodynamic Properties of Light Hydrocarbons and Their Mixtures. I.Methane, Ethane, Propane and n-Butane, J. Chem. Phys., 8, 334 (1940).

    Benedict, M., G. B. Webb, and L. C. Rubin, An Empirical Equation forThermodynamic Properties of Light Hydrocarbons and Their Mixtures.II. Mixtures of Methane, Ethane, Propane, and n-Butane, J. Chem.Phys., 10, 747 (1942).

    Beret, S., and J. M. Prausnitz, Perturbed Hard-Chain Theory - Equation ofState for Fluids Containing Small or Large Molecules, AIChE J., 21,1123 (1975).

    Bieling, V., F. Kurz, B. Rumpf, and G. Maurer, Simultaneous Solubilityof Ammonia and Carbon-Dioxide in Aqueous-Solutions of Sodium-Sulfate in the Temperature Range 313393 K and Pressures up to 3MPa, Ind. Eng. Chem. Res., 34, 1449 (1995).

    Brelvi, S. W., and J. P OConnell, Corresponding States Correlations forLiquid Compressibility and Partial Molar Volumes of Gases at InniteDilution in Liquids AIChE J., 18, 1239 (1972).

    Cabral, V. F., F. W. Tavares, and M. Castier, Monte Carlo Simulation ofAdsorption Using 2-D Models of Heterogeneous Solids, AIChE J., 49,753 (2003).

    Campanella, E. A., P. M. Mathias, and J. P. OConnell, Application of aFluctuation-Theory Model to Equilibrium Properties of Liquids Con-taining Supercritical Substances, AIChE J., 33, 2057 (1987).

    Car, R., and M. Parrinello. Unied Approach for Molecular Dynamics andDensity Functional Theory, Phys. Rev. Lett., 55, 2471 (1985).

    Castells, C. B., P. W. Carr, D. I. Eikens, D. Bush, and C. A. Eckert,Comparative Study of Semitheoretical Models for Predicting Innite

    758 AIChE JournalApril 2004 Vol. 50, No. 4

  • Dilution Activity Coefcients of Alkanes in Organic Solvents, Ind.Eng. Chem. Res., 38, 4104 (1999).

    Chao, K. C., and G. D. Seader, A General Correlation of Vapor-LiquidEquilibria in Hydrocarbon Mixtures, AIChE J., 7, 598 (1961).

    Chang, C. J., and A. D. Randolph, Precipitation of Microsize OrganicParticles from Supercritical Fluids, AIChE J., 35, 1876 (1989).

    Chang C. J. and A. D. Randolph, Solvent Expansion and Solute SolubilityPredictions in Gas-Expanded Liquids, AIChE J., 36, 939 (1990).

    Chapman, W. G., K. E. Gubbins, G. Jackson, and M. Radosz, SAFTEquationof- State Solution Model for Associating Fluids, Fluid PhaseEquilib., 52, 31 (1989).

    Chapman, W. G., K. E. Gubbins, G. Jackson, and M. Radosz, NewReference Equation of State for Associating Liquids, Ind Eng. Chem.Res., 29, 1709 (1990).

    Chen, C.-C. A Segment-Based Local Composition Model for the GibbsEnergy of Polymer-Solutions, Fluid Phase Equilib., 83, 301 (1993).

    Chen, C.-C., H. I. Britt, J. F. Boston, and L. B. Evans, Local CompositionModel for Excess Gibbs Energy of Electrolyte Systems. 1. SingleSolvent, Single Completely Dissociated Electrolyte Systems, AIChE J.,28, 588 (1982).

    Chen, C.-C., and L. B. Evans, A Local Composition Model for the ExcessGibbs Energy of Aqueous-Electrolyte Systems, AIChE J., 32, 444 (1986).

    Chen C.-C., C. P. Bokis, and P. Mathias, Segment-Based Excess Gibbs EnergyModel for Aqueous Organic Electrolytes, AIChE J., 47, 2593 (2001).

    Chien C. H., R. A. Greenkorn, and K. C. Chao, Chain-of-RotatorsEquation of State, AIChE J., 29, 560 (1983).

    Chiew, Y. C., Percus-Yevick Integral-Equation Theory for Athermal Hard-Sphere Chains. 1. Equations of State, Molecular Physics, 70, 129 (1990).

    Chiew, Y. C., Percus-Yevick Integral-Equation Theory for AthermalHard-Sphere Chains. 2. Average Intermolecular Correlation-Functions,Molecular Physics, 73, 359 (1991).

    Choi, P. B., and E. McLaughlin, Effect of a Phase-Transition on theSolubility of a Solid, AIChE J., 29, 150 (1983).

    Curl, R. F., andK. S. Pitzer, Volumetric and Thermodynamic Properties of Fluids- Enthalpy, Free Energy, and Entropy, Ind. Eng. Chem., 50, 265 (1958).

    Deer, E. L., and C. H. Deal, Jr., Analytical Solutions of Groups. Corre-lation of Activity Coefcients Through Structural Group Parameters,Proc. Int. Symp. Distill., 3, Inst. Chem. Eng., London, 40 (1969).

    De Pablo, J. J., and F. A. Escobedo, Molecular Simulations in ChemicalEngineering: Present and Future, AIChE J., 48, 2716 (2002).

    Edmond, E., and A. G. Ogston, An Approach to Study of Phase Separa-tion in Ternary Aqueous Systems, Biochemical J., 109, 569 (1968).

    Edwards, T. J., G. Maurer, J. Newman, and J. M. Prausnitz, Vapor-LiquidEquilibria in Multicomponent Aqueous Solutions of Volatile WeakElectrolytes. AIChE J., 24, 966 (1978).

    Eichinger, B. E., and P. J. Flory, Thermodynamics of Polymer Solutions.2.Polyisobutylene and Benzene, Trans. Faraday Soc., 64, 2053 (1968).

    Escobedo-Alvarado, G. N. and S. I. Sandler Study of EOS-G(ex) MixingRules for LiquidLiquid Equilibria, AIChE J., 44, 1178 (1998).

    Evelein, K. A., R. G. Moore, and R. A. Heidemann, Correlation of PhaseBehavior in Systems Hydrogen-Sulde-Water and Carbon-Dioxide-Wa-ter, Ind. Eng. Chem. Proc. Des. Dev., 15, 423 (1976).

    Faulon, J.-L., D. P. Visco, Jr. and R. S. Pophale, The Signature MolecularDescriptor. 1. Using Extended Valence Sequences in QSAR and QSPRStudies, J. Chem. Inf. Comput. Sci., 43, 707 (2003).

    Felmy, A. R. and J. H. Weare, The Prediction of Borate Mineral Equi-libria in Natural-Waters - Application to Searles Lake, California,Geochim. Cosmochim. Acta, 50, 2771 (1986).

    Feng, W., H. Wen, Z. Xu, and W. Wang, Perturbed Hard-Sphere-ChainTheory Modeling of Vapor-Liquid Equilibria of High Concentration Poly-mer and Coploymer Systems, Fluid Phase Equilib., 183-184, 99 (2001).

    Flory, P. J., Statistical Thermodynamics of Liquid Mixtures, J. AmericanChem. Soc., 87, 1833 (1965).