Chemical Engineering Science 55 (2000) 2121}2129

A powerful algorithm for liquid}liquid}liquid equilibria
predictions and calculations
Roumiana P. Stateva!, Georgi St. Cholakov", Anatolii A. Galushko!,
William A. Wakeham#,*
!Institute of Chemical Engineering, Bulgarian Academy of Sciences, Soxa 1113, Bulgaria
"Department of Petroleum and Solid Fuels, University of Chemical Technology and Metallurgy, Soxa 1156, Bulgaria
#Department of Chemical Engineering and Chemical Technology, Imperial College of Science, Technology and Medicine, Prince Consort Road, London SW7
Received 6 January 1999; received in revised form 27 July 1999; accepted 11 August 1999


The paper describes a new robust and e$cient algorithm for liquid}liquid}liquid equilibria (LLLE) predictions and calculation.
The architecture of the algorithm is a two-level one, and a judicious combination of a stability analysis, carried out in the "rst stage,
with phase identi"cation routines and liquid #ash calculations, carried out in the second stage, is realised. The numerical routines
require modest computational e!orts and demonstrate excellent convergence characteristics. The thermodynamic model applied to
a system does not in#uence the robustness of the new algorithm, which can treat any model and handle any number of components
and phases. The usefulness and e$ciency of the new algorithm is exempli"ed by using two model systems, a water#oil#non-ionic
surfactant system and the system ethylene glycol#lauryl alcohol#nitromethane. ( 2000 Elsevier Science Ltd. All rights reserved.

Keywords: Liquid}liquid}liquid-phase behaviour predictions and calculations

1. Introduction lations is presented; then, to demonstrate its capabilities,
two examples are given, each one of which illustrates
Systems exhibiting multiple liquid phases are of a con- some of the possible di$culties associated with the LLLE
siderable interest for a number of industrial processes, in problem.
particular those involving surfactants. These systems
are applied in the production of cosmetics, drugs and 1.1. Algorithm for multiphase-liquid equilibria predictions
herbicides, paints, lubricants, etc. They are especially and calculations
important in the petroleum engineering industry where
surfactants are used to recover residual oil, a process At present, there are only few studies concerning the
often referred to as microemulsion #ooding. numerical solution of the three liquid-phase equilibria
In this paper we address and concentrate on the design problem, applying excess Gibbs energy models. They are
of a robust and e$cient algorithm for liquid}liquid} discussed in some detail by McDonald and Floudas
liquid equilibrium (LLLE) predictions and calcula- (1995,1997) and Garcia-Sanchez, Schwartzentruber, Am-
tions in multicomponent systems, which is not a trivial mar and Renon (1996). It is important to point out that
task. the solution of this problem is particularly di$cult for
The paper is organised in the following way: "rstly, the strongly non-ideal multicomponent systems with com-
algorithm for LLLE equilibria predictions and calcu- plex interactions, for which the hypersurface of the Gibbs
free energy might have a very elaborate form. Therefore,
it is very easy to "nd either a solution, corresponding
with the local rather than the global minimum of the
* Corresponding author. Tel.: 0044-171-594-5005; fax: 0044-171-594-
Gibbs free energy (i.e. a `wronga solution), or a trivial
8802. solution during phase equilibrium calculations. Thus,
E-mail address: (W.A. Wakeham) special care should be taken in the design of an algorithm

0009-2509/00/$ - see front matter ( 2000 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 0 9 - 2 5 0 9 ( 9 9 ) 0 0 4 7 0 - 4

h "ln z #ln u (z). Michelsen. These two steps may have to be a quantity kH corresponds. where equilibrium predictions and calculations in chemically k (yH)"k (yH)"2"k (yH). To each vector yH. and further extended to the case of phase directly. i (1) Abbreviations kH a number. component i i j number of ethoxy groups in the hydrophilic part U(y) function. U(y)" + [k (y)!k (y)]2. k (y)"ln u (y)#ln y !h . but uses a di!er. Locating all zeros of U(y) is not an easy task. however. N. Stateva et al. where the local tangent hyperplane at equilibrium is determined and then the governing at each yH is parallel to that at z. N (1b) posed by Stateva and Tsvetkov (1994) to the solution i i i of the liquid}liquid}vapour (LLV)-phase equilibrium and k (y) is equated to k (y). the procedure can be described as then pertaining to the `equation-solvinga methods as opposed to the methods which solve the problem by minimising U(yH)"0. / Chemical Engineering Science 55 (2000) 2121}2129 Nomenclature G total Gibbs free energy. possesses some interesting distinctions from these methods and kH"kH(yH)"ln u (yH)#ln yH!h i i i i those will be highlighted further on. 1982. J c mass fraction of the amphiphile in the mixture i number of carbon atoms in the hydrophobic of all three components part of the amphiphile u fugacity coe$cient. Our approach.2. 1994): positive. as N a search over the entire composition space is required. When all calculated kH are ent objective function (Stateva & Tsvetkov. Eq. (2) The "rst stage of the hierarchical procedure involves a rigorous phase stability analysis performed on the where kH geometrically represents the distance in the initial system with composition z.2. 1998).2. N. 2. 1982). the system is stable.2122 R. such that repeated several times. that when y"yH . corresponding to yH A water N total number of components in the system B oil P pressure. which is its minimum value. i"1. originally pro. z and the hyperplane at any yH. (1) of the amphiphile t phase split for the three-phase #ash k (y) function of the chemical potential di!erence. (1) i`1 i The search is further complicated by the existence of i/1 .2. i"1. equations are solved. In 1 2 N most general terms. From Eq. (1) it follows N`1 - problem. reacting systems by Stateva and Wakeham (1997). for any i"1. either positive or negative. (1a) The algorithm advocated in the present study imple. i i i i ments a two-stage hierarchical procedure.2. Eq. The zeros (yH) of U(y) correspond with points on the tion into two distinct steps * "rst the number of phases Gibbs energy surface. The stability check Gibbs energy hypersurface between the hyperplane at applies the idea of the tangent-plane criterion (Baker. Pa C amphiphile ¹ temperature. K LLE liquid}liquid equilibrium yH zero of the function '(y) for the initial system LLLE liquid}liquid}liquid equilibrium z mole fraction vector of the initial system Superscripts Greek letters * corresponding to a zero of the function '(y) a phase split for the two-phase #ash b mass fraction of the oil in the mixture of water Subscripts and oil i number of components for multiphase-liquid equilibria in order to ascertain that where the correct number of phases is found. (1c) the Gibbs energy of the system (Han & Rangaiah.P. The methods of the "rst group divide the problem solu. This distance can be Pierce & Luks.

points correspond with the global solution. Summary of the algorithm uration of a system. cannot determine unequivocally which is the stable phase con"g. and the distribution of the components among the easily di!erentiated analytically. among them there is a zero. Only the global optimisation implement. arranged in an ascending order depend- estimates allow performing the search with the BFGS ing on their density. `chainsa. The "rst group. tion of this set of initial estimates guarantees a highly ure are given by Stateva and Tsvetkov (1991) and will not diversi"ed search for the phase con"guration with the be discussed here.P. as well K-value approach. or the auxiliary initial estimates. identi"ed as unstable at the given The generalised scheme of the algorithm for LLLE temperature and pressure. the chance of using Thus. / Chemical Engineering Science 55 (2000) 2121}2129 2123 a number of trivial solutions. upon z only. . We practically do not exhibit a strong attraction to the use the term `almost alla zeros because there is no trivial solution. a non-linear minimisation technique for locating its The #ash calculations employed can either be based on stationary points. corresponding with the quence of two-phase LL and one LLL #ash calculations number of equilibrium phases present. 1991). which determines the best direction locally lead. If not. composition space. the `chainsa are continuously modi"ed during the divided into two groups * a group. is which can be accepted as a model of the light-liquid applied when a zero of the function is located. It has proved to be extremely reliable initial estimates guarantee a steady and non-oscillatory in locating almost all zeros of the function at a reasonable convergence of the two-phase #ash calculations which computational cost (Stateva and Tsvetkov. Stateva et al. 1. however. We apply the `negativea #ash routine as the implementation of any non-linear minimisation as discussed originally by Whitson and Michelsen (1989). 1987). This theoretically-based guarantee that our scheme will al. which involves a se. The speci"c form to determine the correct number of the phases at equili- of U(y) (its zeros are its minima) and the fact that it is brium. or starting initial estimates. if it is desired. Furthermore. a method has been created which leads. Thus. as will be demonstrated further on. unlike most of the `equation solvinga order and in parallel. stable adopt. Applying the (depending on their liquid-like or vapour-like density). The `startinga initial estimates are elements of duce an arrangement of the stationary points obtained. and the phase. estimates for the subsequent two-phase LL #ashes. The two-phase #ash calculations are performed in any a zero is missed. The stability analysis on its own. However. suggested originally by Stateva and Tsvetkov Once all zeros of U(y) are located they are automatically (1991). while the second * to resume should be used to the full. the search with the BFGS method once a minimum is To generate a reliable initialisation scheme we intro- located. Given z. in prac.1. If. to an `extensivea search in the multidimensional to infeasible solutions is eliminated. then the two-phase result is the "nal. We use the BFGS method with a the minimisation of the Gibbs energy. from very di!erent points in the LLL equilibrium all zeros located are liquid. zero with the vapour-like density and a group with all the which has already been obtained. technique. minimum Gibbs energy. solution. 1. our method is of the Gibbs energy is selected for further processing. as good initial estimates as possible. as will be demonstrated later. R. the stability analysis is performed only once and will lead to a further minimisation of the Gibbs energy. the LL #ash calculations it is very important to provide To do that we introduce two groups of initial esti. these composition space. Further. ¹. however. The `startinga initial rest of the zeros. which is a distinct di!erence from the one additional two-phase LL #ash calculation is re- approach that `equation solvinga methods generally quired. Its implementation. methods give such mathematical guarantee. simple scheme is totally automated and very easy to ways "nd them all. or can adopt the line-search (Fletcher. BFGS method needs a new estimate to restart the search. requires a set of multiple `gooda initial To assure a steady and non-oscillatory convergence of estimates. To determine whether the inclusion of a third phase more. Be- ing to the smallest increase in the value of U(y). with a vapour-like density. All details cause it is impossible to tell a priori which stationary of the organisation and implementation of the initialisa. each one of which originates from a corre. They lead to a variety methods which will eventually fail to determine correctly of two-phase LL solutions and that with the lowest value the phase identi"cation of the system. if that is desired. based on the notion of a heavy and a light liquid phase sponding pure component of the system. Then all zeros from the second group are combined To "nd such an estimate an uni-dimensional search is with the zero of the "rst group to produce a set of initial applied.1. guesses that may in the end lead to local optima or even tice. allows the application of phases. For the case of a possible method in parallel. the applica- tion strategy employed by the stability analysis proced. algorithm. This is the point mates. which contains the search in order to avoid the location of a minimum. To do that the algorithm predictions and calculations is summarised as follows: continues with the second stage. self-recovering. P. is where the information from the stability analysis routine used to begin the search. The second group.

z"(0. demonstrate a complex phase behaviour with peculiar 4. system "guration is a two-phase LL. 3. Run two-phase LL #ash calculations upon z.1997) use the UNIQUAC and UNIFAC (binary heavier of the two liquid phases of the above LL interaction parameters for the prediction of LL equilibria solution. If solution B represents the phase lations to identify the stable-phase con"guration of the con"guration with the minimum Gibbs energy then both system as either LL or LLL. Application of the algorithm.1 MPa. among which two deserve 1. B (Table 2). performed on the "rst stage of must be added or deleted. if unstable.3). the true equilibrium our algorithm. / Chemical Engineering Science 55 (2000) 2121}2129 2. represented by case (1). the system is a homogeneous mixture. If the initial solution is this particular example because it provides a ground for stable.2. etc.1.0. the For this system. However. in a four. nitromethane system at ¹"295 K and P"0. The other two zeros component system there might be several three-phase form the group of the heavy liquid zeros. 1995. a model of the light liquid phase. from Magnussen et al.5) additional brief comments. The zeros of the solutions found. END.1 MPa. We apply the original UNIFAC converged mole fractions of the solutions determined model (Fredenslund.0766 0. Furthermore.0. z"(0. for further analysis while solution A is discarded from in practice. method will "rstly perform a sequence of a stability Since phase ¸B of the solution with the lowest Gibbs 2 analysis and a LL #ash calculation on the initial system. sequentiala shown in Table 3.P.0. three sets of feed conditions are tested and analysed with Several characteristics of the above algorithm ensure our algorithm: its #exibility and robustness. of which only one corresponds to the two groups are paired and a set of initial estimates for the global minimum). Test examples step 3 Converged mole fractions of solution A The robustness and e$ciency of the algorithm will be Component Feed ¸A 1 ¸A 2 illustrated on two examples. LL . 3.0. McDonald & Floudas.9504. Two solu- by selecting on each step the solution with the minimum tions are obtained * solution A (Table 1) and solution Gibbs energy for further processing. Use as initial estimates the models for the system.1997). pour phase is not formed and there is a potential for three mates for the subsequent LLL calculations are found liquid phases at equilibrium. the equilibrium-phase con. z"(0.. 3 2 1995. The present algorithm overcomes this subsequent LL #ash calculations is obtained. The zero rich in lauryl alcohol is the components within the same phases (e. 1981) as the thermodynamic 5. a phase is added and a new #ash is run. Then. Feed conditions of case (1). Test this particular solution towards a LLL split by phase boundaries. Table 1 Ethylene glycol.4 0.1 0. To test that distribution among the phases for the latter case an assumption we perform a LL #ash calculation upon the additional three-phase LLL #ash is performed upon z. lauryl alcohol.2631 2 6 2 C H O 0. the present algorithm requires only Solution B is with a lower Gibbs energy and is chosen a single application of the stability analysis routine and. performing a liquid}liquid #ash on its lighter phase. 12 26 CH NO 0.5487 the system ethylene glycol#lauryl alcohol#nitrome. To obtain the component its phases would be stable to a phase split.2.5 0. Choose among the LL solutions the one with the systems. Continue with step 5. using as The second example presents the solution of the LLLE initial estimates the stationary points obtained in step problem for two water#oil#non-ionic surfactant 2. The commonly accepted 2. the a postulated solution is found to be unstable then a phase stability analysis routine.4. continue with step 3.0.g. energy among all initial phase splits is unstable (a is followed by a stability check on one of the phases of the converged LL solution. performance of other algorithms. END.4. Initial esti.0. END.3. We have chosen a"0. 1.3. Run a LLL #ash upon z. Stateva et al.1882 thane and is taken from the literature (Sun & Seider.5163 0. McDonald and Floudas by performing an additional LL #ash upon the (1995. Otherwise.2. further consideration. Ethylene glycol#lauryl alcohol#nitromethane ically acceptable bounds. any other `equation-solving. We have chosen these systems because they minimum G. The converged solution is comparison. rium predictions and calculations is to assume that if For the feed condition.1. In lighter phase of solution B. The "rst one discusses the phase equilibrium predictions and calculations for C H O 0. at ¹"295 K and P"0. Run a stability analysis upon z. a va- equilibrium-phase con"guration is LLL. Gmehling & Rasmussen.5) practice of equation solving methods for phase equilib. a maximum of three two-phase #ash calcu. two of solution may in fact be one with a di!erent distribution of which have negative kH. 1.2124 R.4071 0. GA "!12. If the phase split upon convergence is outside the phys. 1977) and in step 4.24534. "nds three distinct zeros of U(y). comparison of the performance of our algorithm with the Otherwise.

Application of the algorithm. step 3 Converged mole fractions of solution B step 3 Converged mole fractions of solution 1 Component Feed ¸B ¸B Component Feed Liquid Liquid 1 2 1 2 C H O 0.1 MPa. report a local LL solu. a"0. the LLL #ash con- implementing a local optimisation method.174742 0.174742 3 2 3 2 a"0.6122 0.096779 0. Feed conditions of case (2). As a result a stability analysis routine. represented by case (2). G"!12. which.2360340 0.074977 The correct phase identi"cation of the system with feed 12 26 CH NO 0.0748627 0.035292 0. 3 2 a"0. Feed conditions of case (2).6665 0.540228 12 26 12 26 CH NO 0. lauryl alcohol.252575 2 1 2 2 6 2 C H O 0. Since the feed. nitromethane system at ¹"295 K and P"0. Applying these initial estimates.1784 a"0.69072.6891033 0.25863. The result shows that tangent plane distance function. Then the available initial the di$culties and the inherent dangers associated with estimates converge only to solution 1. they converge in less than three iterations.1409 0. nitromethane system at ¹"295 K and P"0. the zero rich in ethylene glycol is very mates is given in Table 7. the three equilibrium phases a LLL #ash is performed strates how the location of the feed in#uences the success upon z. they point out that Following the steps of the algorithm.4 0.0763 0. Application of the algorithm.1 MPa. GB "!12.2469 0. The stable-phase con"guration of the system is identi"ed as a LLLE. Let us for the sake of argument accept that the stability The second feed condition. G"!12. lauryl alcohol.5708 12 26 CH NO 0. nitromethane system at ¹"295 K and P"0. di$cult to locate. 1).5 0. Table 6 Ethylene glycol. Application of the algorithm. lauryl alcohol.285030 2 6 2 2 6 2 C H O 0.8606 0.5 0.1409 C H O 0.2572 0. lauryl alcohol.690811 2 6 2 C H O 0. P"0. feed condition. step 4 step 4 Converged mole fractions of the LL #ash upon phase ¸B of 2 solution B Component Liquid ¸F ¸F 2 1 2 Component ¸B ¸H ¸H C H O 0.1 MPa. / Chemical Engineering Science 55 (2000) 2121}2129 2125 Table 2 Table 4 Ethylene glycol. Application of the algorithm. Feed conditions of case (1).0748627 0.106458 0.2469 CH NO 0.1 MPa.8693. lauryl alcohol.234212 (1) is realised at a very early stage of the calculations and 3 2 is not computationally expensive * only three two- phase LL #ash calculations are required and. analysis on the "rst stage of our algorithm fails to deter- according to McDonald and Floudas (1995) exempli"es mine all zeros of the function.035722 0. P"0.857820 0. To obtain the distribution of the components among This example is interesting also because it demon.2508 CH NO 0. C H O 0.2360340 0. R. Feed conditions of case (1). nitromethane system at ¹"295 K and Ethylene glycol. might fail to verges in three iterations.2 0.0977 0. Application of the algorithm. a two-phase LL con"guration corresponds with a local rather than Component ¸F ¸H ¸H 1 1 2 a global minimum of the Gibbs energy. the stable phase con"guration of the system is LLL tion as the stable-phase con"guration for this particular (Table 5).867929 0.1 0.285030 0. nitromethane system at ¹"295 K and Ethylene glycol.0417 0. The set of the initial esti- 1 3 triangle (Fig. tested for stability by performing a LL #ash calcula- tion method to determine the stationary points of the tion upon its lighter phase.540228 0. Further.9235.6122 C H O 0. obviously. step 4 inside the physically accepted boundaries). the correct determination of the global solution by any has a very large domain of attraction (Table 4).577606 12 26 C H O 0.5877. represented by case (2) is located in by only one more LL #ash run on the heavier phase of a close proximity to the L L boundary of the LLL the above solution (Table 6).6891033 0. The component distribution for "nd it.P. the global solution is given in Table 8.5238. because of the excellent initial estimates applied.800515. Stateva et al.1 MPa. who apply a homotopy-continua.3 0. . local optimisation method. To secure a steady convergence of the #ash of the stability analysis and phase identi"cation rou. LL Table 5 Table 3 Ethylene glycol. routine an excellent set of initial estimates are obtained tines. solution 1 is Sun and Seider (1995). Feed conditions of case (2).169819 2 6 2 3 2 C H O 0.

1. / Chemical Engineering Science 55 (2000) 2121}2129 Fig. Application of the algorithm. Feed conditions of case (2).252738 system H O#n-C H #C E at atmospheric pressure.074977 0. lauryl alcohol. t "0. Stateva et al.1 MPa. in equilibrium with an excess water phase. in equi- 12 26 10 22 CH NO 0. the t "0.577605 12 26 CH NO 0.096779 0. A surfactant-rich aqueous phase. step 5 Actual global LLL solution.234212 0. t "0. is denoted as 21 . Table 7 Ethylene glycol.035292 0. mole fractions Component L L L 1 2 3 Fig.1 MPa.035232 0.4842.074737 0. is denoted as 2.096766 0. nitromethane system at ¹"295 k and P"0.2126 R.1 MPa.0466. 1 2 3 . LLLE is denoted as 3. nitromethane system at ¹"295 k and P"0. step 4 The set of the initial estimates (mole fractions) Component ¸H ¸H ¸F 1 2 2 C H O 0.691110 0.690811 0. Application of the algorithm.4692.234153 0. H O : n- 2 6 2 2 10 22 4 1 2 C H O 0. two and three-liquid phases in the ternary C H O 0. The position of the three studied feed conditions are shown as well.578284 C H ratio is 50 : 50 wt%. lauryl alcohol.868002 0. Feed conditions of case (2).169819 3 2 Table 8 Ethylene glycol.168978 librium with an excess oil phase.867929 0.252575 2 6 2 C H O 0. Regions of one.P. a surfactant-rich oil 3 2 phase. Liquid}liquid}liquid equilibria of the ethylene glycol#lauryl alcohol#nitromethane system at ¹"295 K and P"0. 2.

/ Chemical Engineering Science 55 (2000) 2121}2129 2127 Fig.P. In#uence of the temperature on the LLL equilibrium of the ternary system H O#C H (CH ) H#C E at atmospheric pressure. R. Stateva et al. 2 6 5 28 4 2 . 3.

. P. (1990). (1982). Langmuir. M. Strey. without any numerical di$culties the appearance and/or Kahlweit. b"B/(A#B) is the mass fraction of the oil in the mix. search..2128 R. 499}511. These systems exhibit three liquid phases. B. High-pressure vapour}liquid purposes of validation the new algorithm we are advocat. Stateva. S. L. Vapour}Liquid phenyloctane#C E system. body will be obtained. New York: Wiley. & Fredenslund. R. growth. The corresponding phase frac. M. We used normal alkanes and phenylalkanes R.. & Haase.0838. R.3897. 207}225. K. t "0. & Firman... M. t"0. * water#n-C #C E system. G. 785}790.Ch. & Rasmussen. Haase & phiphiles. L. D. G.. for case (1) they are: for predicting and calculating LLLE in multicompo- t "0. 1829}1836. General disappearance of a three-phase region. 22. change accordingly. A. the algorithm reports correctly and 2 lar solvents. A. Conclusion sented by case (1) and case (3) are the same and that is why they are not given. level one. F. robust. the algorithm does not su!er from some of the water#oil#non-ionic surfactant systems well-known problems like: convergence to a trivial solu- In order to demonstrate further the usefulness. . Schwartzentruber. Aa. Strey. Larsen. D. M. Cholakov. ature. Gmehling. D. 2274}2286. energy). c"C/(A#B#C). & Luks. Properties of eventual disappearance of the LLLE for the system the three-phase bodies of H O}oil}nonionic amphiphile mixtures. where ations. Schomacker. 4.4592. characteristics. Strey. a failure to converge to a physically feasible tant systems. E. 121. In our study we used the MHV2 model (Dahl & Michelsen. and A. as (1996). Kahlweit. 4 2 Kahlweit. (1987). 26. nal. Garcia-Sanchez. t "0. A method for multiphase equilib- the mixture of all three components. 36. convergence to an incorrect solution (corresponding ness and e$ciency of our algorithm we consider several with a local rather than a global minimum of the Gibbs numerical tests involving water#oil#non-ionic surfac. L. and electrolytes. (i denotes the number of carbon atoms in the hydropho- bic alkyl chain and j the number of hydrophilic ethoxy groups) * to model non-ionic surfactants. nonpolar solvents. 1989). The architecture of the algorithm is a two- 1 2 3 * t "0. for case (3) nent systems. 2 shows the simulated `"sha pro"le of the three. The #ash routines require modest com- which exist only within a well-de"ned temperature and putational e!orts and demonstrate excellent convergence pressure interval. Langmuir. (1988b). Fig. and the numerical routines of the second 1 level exploit to the full the results of the stability analysis check.2.. Gibbs energy analysis of phase equilibria. We exemplify the results on two model systems Fletcher.4882. / Chemical Engineering Science 55 (2000) 2121}2129 The component distribution for feed conditions..I. and an excess Gibbs energy model (modi"ed UNIFAC.. If a vertical section through the temper.. & Schomacker.1511. The present paper advocates a novel robust algorithm tions. Stateva et al. Aa. patterns of the phase behaviour of H O.5. Amsterdam: Elsevier. which is given by rium calculations. am- 2 mentally for these systems (Kahlweit. observed experi. because it has 731}748.. J. Phase behaviour predictions and calculations for more.. Lar. R. Rasmussen & Fredenslund. Firman. Jour- ing. however. 1988b. Fluid Phase Equilibria. 2. Ammar. P. 4 2 Equilibria Using UNIFAC. 5. It resembles the shape of a `"sha and can be used to represent graphically the in#uence of the surfactant concentration on the phase behaviour of Acknowledgements the system. and Kahlweit. Pierce. G. phase body for the water#n-C #C E system. P. 305}315. The algorithm can employ successfully any thermo- ture of water (A) and oil (B).. P. Schomacker riua and heat of mixing.. P. solution at all. Kahlweit. H. & Rangaiah. Practical Methods for Optimization (2nd Edition). carried out on the "rst level. Fig. & Michelsen.. General patterns of the phase behaviour of H O. (1989). M.. Jen. St. Langmuir. a pro"le of the three-phase dynamic model. 2 water#phenyloctane#C E with increasing temper. sen. Rasmussen. amphiphiles. A modi"ed Firman. 3 presents the origin (appearance).2.E. R. Strey. (1987). 897}911. & Renon. Computers in Chemical Engineering.. Haase.. J. (1998). M. Industrial and Engineering Chemistry Re- & Haase. Strey. (1977). Firman. qualitatively the correct behaviour and serves the present Dahl.. Further- 1. and electrolytes.4280. equilibrium with UNIFAC-based equation of state. R. and water# 10 4 1 Fredenslund. Convergence of the LL and LLL #ash ature * composition phase prism for a water#oil# calculations was usually achieved in less than four iter- surfactant system is plotted at b"0.. 1987). M. Baker. Galushko (C H )(CH ) H to model the oil components and n- 6 5 2k gratefully acknowledge "nancial support from the British alkyl polyglycol ethers H(CH ) (OCH CH ) OH-C E 2i 2 2+ i j Council. (1988a). t"0. 1990). repre.. Modeling of multiphase liquid equilibria for multicompo- 10 4 1 nent mixtures. J. P. tion. a function of the mass fraction of the amphiphile (C) in Han. Jen UNIFAC group contribution model for prediction of phase equilib- & Schomacker. Strey. Haase. D. References which combines the Soave}Redlich}Kwong CEOS. 22(7}8). C. Haase. 1. nonpo- In all cases studied. A.. 4.P. 2. Society of Petroleum Engineering Journal. 1988a.

21. 51}71. & Floudas. 53. Application to 1798}1814. Part I: Stability. 1. C.. C. tional tool for the phase and chemical equilibrium problem. D. A. P. & Tsvetkov. vapor}liquid}liquid systems. lations for chemically reacting systems. 5474}5482. St. (1997). 1}23. Stateva. Phase Equilibrium. Global optimisa.Ch. 4. A. P. A. A diverse approach for the tion for the phase stability problem. Homotopy } Contiuation for stability Stateva. R. 213}249. Sun. Journal. Technology Today. 41. Fluid to thermodynamic stability analysis as a "rst step when solving Phase Eqilibrium.. Fluid Phase Equilibrium. / Chemical Engineering Science 55 (2000) 2121}2129 2129 McDonald. C. & Seider.. 103. M. the isothermal multiphase #ash problem. G. C. M. G.. (1995). (1994). R. The negative #ash. Fluid 233}245. L. C. Com. & Tsvetkov.P. (1989). H.E. L. W. Stateva et al. The isothermal #ash problem. R. A. M. 9. R. 722}734. (1982). M. GLOPEQ: a new computa eering. . 72. A. St. 36. solution of the isothermal multiphase #ash problem. (1997). A rigorous approach analysis in the global minimisation of the Gibbs Free Energy.I. & Michelsen. & Floudas. Canadian Journal of Chemical Engin McDonald.. & Wakeham. (1991). W. Industrial Engineering Chem- Michelsen. (1995). Phase equilibrium calcu- puters in Chemical Engineering. Whitson.. Stateva. ical Research. P..