Brazilian Journal

of Chemical ISSN 0104-6632
Engineering Printed in Brazil
www.abeq.org.br/bjche

Vol. 25, No. 03, pp. 571 - 583, July - September, 2008

PHASE STABILITY ANALYSIS OF
LIQUID-LIQUID EQUILIBRIUM WITH
STOCHASTIC METHODS
G. Nagatani1, J. Ferrari1, L. Cardozo Filho2, C. C. R. S. Rossi2, R. Guirardello3,
J. Vladimir Oliveira1 and M. L. Corazza1*
1
Department of Food Engineering, URI, Campus de Erechim, Phone: +(55) (54) 3520-9000,
Fax: +(55) (54) 3520-9090, Av. Sete de Setembro 1621, CEP: 99700-000, Erechim - RS, Brazil.
E-mail address: mlcorazza@uricer.edu.br
2
Department of Chemical Engineering, Maringá State University, UEM,
Av. Colombo 5790, CEP: 87020-900, Maringá - PR, Brazil.
3
DPQ-FEQ, College of Chemical Engineering, State University of Campinas,
UNICAMP CEP: 13081-970, Campinas - SP, Brazil.

(Received: February 19, 2007 ; Accepted: March 5, 2008)

Abstract - Minimization of Gibbs free energy using activity coefficient models and nonlinear equation
solution techniques is commonly applied to phase stability problems. However, when conventional
techniques, such as the Newton-Raphson method, are employed, serious convergence problems may arise.
Due to the existence of multiple solutions, several problems can be found in modeling liquid-liquid
equilibrium of multicomponent systems, which are highly dependent on the initial guess. In this work phase
stability analysis of liquid-liquid equilibrium is investigated using the NRTL model. For this purpose, two
distinct stochastic numerical algorithms are employed to minimize the tangent plane distance of Gibbs free
energy: a subdivision algorithm that can find all roots of nonlinear equations for liquid-liquid stability
analysis and the Simulated Annealing method. Results obtained in this work for the two stochastic algorithms
are compared with those of the Interval Newton method from the literature. Several different binary and
multicomponent systems from the literature were successfully investigated.
Keywords: Phase stability; Liquid-liquid equilibrium; Stochastic algorithms; Thermodynamic models.

INTRODUCTION Phase stability analysis of multicomponent
mixtures is usually investigated by the Gibbs tangent
The phase split phenomenon in liquid mixtures is plane distance criterion (Baker et al., 1982;
related to thermodynamic stability of the system and Michelsen, 1982). For a given temperature and
comprises important information for the projection, pressure, the necessary and sufficient condition for a
modeling and simulation of industrial processes that phase with composition z to be stable is that the
involve liquid phase separation. Phase stability Gibss free energy surface of the mixture is not
analysis allows determination of the exact number of intercepted by the tangent hyper plane associated
coexisting phases in stable equilibrium and also with this surface in x. To fulfill this condition, the
provides estimation of phase compositions, affording Gibbs tangent plane distance function, TPD(x), must
a suitable initialization for phase equilibrium be nonnegative for any acceptable x. Therefore, it is
calculations. possible to examine whether a phase is stable by

*To whom correspondence should be addressed

determined by the first-order derivatives with respect However. J. 1982): of an algebraic nonlinear set of equations that represents the stationary points of the Gibbs tangent n plane distance function and (ii) direct minimization of the TPD(x) function.. respectively. the application of where γ i and γ *i are the activity coefficient of conventional mathematical models presents component i of tried and tested phases. n . n − 1 (Smiley and Chun. L... 1968). such as homotopy continuation. J. The stochastic techniques. The second approach. the Gibbs tangent plane unstable and will be subdivided into new phases.: for the set of equations. In both cases. Zhu et al... wich are 1995. R. Nagatani. Some work is available in the current literature regarding application of deterministic techniques for global As discussed previously. the first strategy function represented in Eq.. the aim of this work is to provide a systematic comparison between the two mentioned {ln  x i γi ( x ) − ln  zi γ i ( z )} − (3) strategies for phase stability analysis: Simulated Annealing (Press et al.. multiple roots may be found. like Simulated Annealing and stationary points of TPD(x) can be obtained by Genetic Algorithm. 2007) – henceforth denominated SubDivNL – employed for When the activity coefficients are estimated using solution of the stationary points on TPD of liquid the NRTL model. Vladimir Oliveira and M. and a subdivision method i = 1.. If the tangent plane distance the approach used in this work with Interval Newton function at the global minimum point has a value analysis is also provided. The Gibbs free energy for liquid mixtures at low then the tangent plane lies above the Gibbs free or moderate pressures can be modeled in terms of energy surface. Corazza minimization of the TPD(x) function. Ferrari.. Guirardello. C.. just a few studies make use of global to the ( n − 1 ) independent mole fractions. especially when liquid Since a phase is stable if TPD ( x ) ≥ 0 . McDonald and Floudas. subject to the mass balance restriction: 2000... for phase equilibrium and phase solving the following set of algebraic nonlinear stability calculations (Zhu and Xu.. Cardozo Filho. difficulties. x n ) and z = (z1 . and hence the examined phase is activity coefficients. Corazza et al.. then the analyzed phase is thermodynamically stable because the TPD(x) function is also nonnegative for all x vectors PHASE STABILITY ANALYSIS contained in the permissible region. this set of equations may have mixtures modeled by the NRTL activity coefficient multiple roots in addition to the trivial solution x = z. taking into account the TPD ( x ) = ∑ x ln x γ ( x ) − ln z γ ( z ) i =1 i i i * i i (1) constraints imposed by the mass conservation principle. subject to mass model (Renon and Prausnitz. Since composition of tried and tested phases. respectively. C.. in contrast. R. z n ) represent the converges to a trial root or to a local minimum. Souza et al. this optimization optimization. In this context.. 2001). 1) or computation of the (Sun and Seider. L. 2006. If the TPD(x) function is negative at its global minimum point. i. n ∑x The majority of numerical methods applied to (2) subject to : 1 − i =0 phase equilibrium calculations has local convergence i =1 characteristics resulting in only local solutions and with 0 ≤ x i ≤ 1 and i = 1. the phases are under inspection.. two at specified temperature T and pressure P is given by distinct approaches can be employed: (i) resolution (Baker et al. Lima et al. S. 2001. distance for a mixture with n number of components For the computational evaluation of this test.e. are very sensitive to initial guesses. Comparison of balance constraints. Rossi. 1995.. greater than or equal to zero. 1992). wich is a global { } ln  x n γ n ( x )  − ln  z n γ n ( z )  = 0 optimization procedure.. such as dependence on an arbitrary initialization or the possibility that the solution and x = (x1 . (1) must be minimized requires a mathematical method able to find all roots with respect to all possible compositions. Thus.. 1999. stationary points of the TPD(x) function.. Rangaiah. Brazilian Journal of Chemical Engineering . requires a reliable global optimization min TPD ( x ) method. 2006).572 G. equations. problem can be solved by direct minimization of the branch-and-bound and Interval Newton analysis objective function (Eq..

2004. Furthermore. Since the solutions are confined to the Petriciolet. As suggested by Faber et ( ) if F x ij ≤ 2−i + τij (4) al. the SA modeled with an equation of state was proposed.. (2005).. al. x k +1 = x k = xopt . subdivision level i). pp. 1999. optimization technique suitable for detecting a global minimum hidden among many local minima (Press Subdivision Algorithm et al. The selection criterion is based on otherwise.. Teh and Rangaiah. 2001). 2005. This method explores the analogy with the way that. a metal forms a According to Smiley and Chun (2001) the crystalline structure with minimum energy. x k +1 . July .September. when cooled slowly. subdivision algorithm consists in a given set of Metropolis et al. interval (hyper rectangle) R ∈ R d . Franceschi et and Chun. where d is the The Simulated Annealing algorithm is based on problem dimension. In (the index j identifies the subrectangles produced at a accordance with Faber et al. where the probability acceptance criterion used is Pr ( ϕ ( x k+1 ) ) ≥ xr . Phase Stability Analysis of Liquid-Liquid Equilibrium with Stochastic Methods 573 Simulated Annealing (1992). an inappropriate 1 d  ∂F ( x )  change in the annealing temperature can result in τij = ∑  max 2 k =1  y∈R ij ∂y k  bk − a k  2 (5) convergence to a local minimum. a new vector. and ak and bk are the lower and upper bonds of applied to several types of phase equilibrium unknowns. 2004. 2006). is randomly generated from the earlier state For each subdivision level.0 . where just as excessively fast cooling may produce a structure with a higher energetic content. No. equilibrium calculations of systems at high-pressures α = 0. the annealing schedule can be set by TAj+1 = αTAj for j = 1. a and flash equilibrium calculations by minimizing the conventional method (e.. the Newton-Raphson Gibbs free energy (Zhu and Xu. 2001. and the aim of a subdivision the Monte Carlo method and makes use of an algorithm is to determine all values objective function ϕ ( x ) to replace energy E. (1953) proposed that the thermal algebraic nonlinear equations F ( x ) and a primitive equilibrium of an annealing process could be simulated by the Boltzmann probability distribution. results obtained were quite satisfactory (Corazza et Brazilian Journal of Chemical Engineering Vol. ζ . taking 2000. 2006. 2006) and parameter estimation of retained subintervals. in which 0 ≤ α ≤ 1 .. equations: The progressive decrease in annealing temperature reduces the probability that a more energetic state be selected. The algorithm used was that presented in Press et al. However. Carvalho Jr. 1992). In this work. method) is employed to determine the roots.. (variables) into two equal parts so as to obtain 2d analogous to the annealing temperature. then the new vector is obtained are tested for the existence of roots in order that only those containing one or more solutions are unconditionally accepted. 03. as initial values the middle points of the last Corazza et al. the convergence of the adopted thermodynamics models (Henderson et al. a subdivision algorithm for phase values for phase stability calculations are TA0 = 1000.1].g. is employed subrectangles R ij from one rectangle in the R d space for evaluation of the Boltzmann factor. 2003. the acceptance probability is determined calculation of the Euclidean norm of the set of by the Metropolis criterion. Bonilla- subintervals. Preliminary studies indicated that the best A recent study.. it is also essential to start from a conveniently high temperature in order to allow a more detailed then the subrectangle R ij is retained or else. The * { d ( ) x = x∈R F x = 0 } through systematic subdivisions vector x is defined in the n-dimensional space of the and trials in R. 571 .. Pr is the Boltzmann probability Simulated Annealing (SA) is a stochastic global distribution and xr is a random number [0.583. Rangaiah. the subintervals x k . method becomes more efficient and reliable (Smiley Moura et al. 25. A parameter TA . Souza et al. it is exploration of the solution space of the problem.9 and ζ = 10000 . The first partition of R to congruent optimization variables and represents a particular subrectangles is by division of all coordinates configuration of the system.. 2008 ..e. i. (2005). 2001. If ϕ ( x k +1 ) ≤ ϕ ( x k ) . Zhu et al... maintained. discarded (y is the vector of the randomly sampled Over the past years the SA algorithm has been points). calculation problems involving phase stability tests After a finite number of subdivisions.

1500. C.9000) 1.5000) 0.1282 0.8709) 0.1151 –2 (0.1991×10–3.1795 0. 0. 0. L. RESULTS AND DISCUSSION solutions obtained in this work are in perfect agreement with the roots found by Gecegormez and To validate the algorithms used in this work Demirel (2005).55 (0. 0. Rossi. 0.3044.8000) 1.7096.0701×10–3.5442) –7.1696×10−2.70) (0. 0.9930) –9.1147 0.6614 (2. x2) (x) (s) (x1.6409×10–4 24 (0.2500.5248×10–17 1.3013 × 10 .9988) –0.9989) –0. which used an interval analysis (Simulated Annealing and SubDivNL subdivision algorithm.7500.11 (0.9987) –3 –0.7096.85) (0.08 (1.6000) 0.1147 (0.1806. 0.0739×10–3.50.50) (0. 0. 0.66GHz with 512MB RAM.9344×10–3 (2.5716×10−3. from the citric acid (1) + butan-2-ol (2) binary system at preliminary tests.5767×10–2 (1. the appropriate maximum number 25ºC and different overall compositions are of subdivisions for phase stability tests was presented. z2) Stationary points TPD CPU Global minimum TPD CPU Stationary points TPD CPU (x1.8308 (1.6614 (0.5705. model were taken from Gecegormez and Demirel (2005).8500) 1.2500.5716×10–3.10 (1. 0.60) (0.30.75) (0.5000.1147 (1.38 (3.6956) −4.20.9989) –0. 0. 0.2000. 0.11 (0. 0.8500) 0.4000. All the calculations were performed in a algorithm reported on Gecegormez and Demirel PC Pentium 4 2.90) (0.6957×10–3 (0.4558.1991×10−3.7500) 1. 0.0000 (1. R.9989) –0. One can see in Table 1 that. 0.1795 (0.75. L.2535 0.6555 × 10−3.10. 0.0000×10–9 0.0000 23 (0.9989) –0.8194) 2.5767×10–2 (0.0000×10–9 0. 0. to our Case 1: citric acid (1) + butan-2-ol (2) system knowledge.7500.10 –0.5442) −7.1806. SubDivNL Simulated Annealing Interval Newton* (z1.11 (1.2904×10–2 (0. 0.3695.8323 26 (0. The investigated systems results) and was shown to be much less time- include binary. x2) (x) (s) (7.0500. 0. 0.1570 0. 0.4295) –0. x2) (x) (s) (x1.2500) –8.4558.6410×10–4 0.1800 (1. 0. 0.9000) 0.9981) 3.6614 0. 0.2539.0000×10–8 (0. In this work.58 (0. (2001). for this system. 0. 0.6000) 1.0000 (1.2904) –0.6957×10–3 (7.. Nagatani.9973) −5. 0.0000 25 (0.25.9988) –0. 0.6305) −1.1800 0. 0. 0.6956) –4.8835×10–3.0000×10–9 0.4997) 1. 0. S.11 (0.9883) 0. 574 G.7671×10 0.1151 (1.40. 0.0000 (1.1291.7500) 0. 0.9500) 0.3013×10–3. It may be interesting to note that.09 (0. 0. 0.5003.0000×10–8 (0. 0.7000) 1.3695. 0.8194) 2.0000×10–9 0.9344 × 10–2 (0.9883) 0.0000 22 (0.0000×10–8 (0.1500. 0.4295) –0. 0.7096.7461) 2.3000.2000.1282 24 (0.6625 × 10–7 0.0000×10–9 0.9500) 1. Corazza al.3412×10−2.0701×10−3.4295) −0.1800 (0. 0.1991×10–3.9981) 3.1570 (1.0315×10–3.9989) –0. 0.5000.7500. It can also be observed that the method).05.0500.95) (0. Guirardello.0315×10−3.6555×10–3. which were presented by Lintomen et al. ternary and multicomponent consuming than the Interval Newton (IN/GB) mixtures.56 (0. The binary parameters of the NRTL determined to be six.5716×10–3.1291.9984) –0.3046. 0.8709) 0.53 (0.9666) 0. 0.1000.5705.25) (3.2539.1151 0.2500) 1.0000×10–9 (0.6305) –1.55 (0.9984) −0. 0.09 (1. 0.3191×10–4 (1.9989) (1.0000 * Results obtained from Gecegormez and Demirel (2005) Brazilian Journal of Chemical Engineering .4000. 0.7461) 2. J.5705.2904) –0.8000) 0.56 (0.0000 23 (0. 0.2904×10–2 (1. 0. J. 0. C.8323 0.3246×10–2 (0. 0.2535 (1.7671×10–2 22 (0.0739×10–3. (2005).0000 22 (0.3000.0701×10–3.7000) 0.3190×10–4 (0. Table 1: Results of phase stability analysis for the citric acid (1) + butan-2-ol (2) system at 25ºC.2904) –0.9987) –0. 0. 2007). 0. 0. 0.0739×10−3.8835×10−3.1000. Vladimir Oliveira and M. 0.59 (0.2500) 0. Cardozo Filho. 0. 0. an application of a subdivision algorithm to liquid-liquid equilibrium calculations is not In Table 1 the results of phase stability tests for available in the literature.5000) 1. 0.9973) –5.1570 (0.9984) –0. some results presented in the literature for SubDivNL algorithm had lower computational effort liquid-liquid stability calculations with the NRTL in terms of CPU time that the SA algorithm (our model were selected.48 (0.3013×10−3. 0.80) (0.1795 (1. 0.8308 (0. 0. 0. Ferrari. R. 0. 0. 0.1696×10–2.9988) –0. 0.9930) −9. 0.3412×10–2. 0. 0. 0.9987) –0. 0.2535 (0.15.9666) 0.3246×10–2 (1. 0.

homogeneous system.008 0. SubDivNL Simulated Annealing Interval Newton* (z1.80) (0. Results from the SA and ∂TPD SubDivNL algorithms are compared with solutions found is = 0. 0.04 0. 0. results obtained in this work are different from Sayegh and Ratcliff (1976).9150×10–3 (0.018 0. 0.2-dimetilbutane (2) system at 25ºC.9000) 1. 0.8500) 1.1294. x2) (x) (s) (0. In represented in Figures 1 and 2.9500) –3.2000.7775×10–5 0.9000) 9.1500. 0.8500) –1. x2) (x) (s) (x1.95) and x = (0. x2) (x) (s) (x1.1000.20. 0.7463×10 –5 (6.9349) 2.95) and z = (0.50 (0.11 (0. 0.14 (0. Nevertheless. 1991) for LLE computations is applied Gecegormez and Demirel (2005) are not true just to enable numerical comparison with the results of solutions for the phase stability test.016 0. 0.8500) 0.0500. 25.0000 × 10–9 0. z2) Stationary points TPD CPU Global minimum TPD CPU Stationary points TPD CPU (x1.08 (0.8500) –9. 0.0000 (6. 2008 .08 0.5086×10–2.9187×10 .8706). z = (0. Results of phase stability analysis obtained in this the Euclidean norms of the set of equations work for the n-pentanol (1) + 2.42 38 (0.9500) 1.2-dimetilbutane (2) system at 25 oC and z1 = 0.9308) –2 –4.20.0292.0000×10–8 (0. 0.5542 × 10–16 0.9000) 0. 0. 0. 0.006 0.2000.2-dimethylbutane (2) calculated for such roots are far from zero: for z = system at 25 ºC and several global compositions are (0. 0. the solutions presented in the literature are not stationary points on the TPD surface.1000.8000) 1. a parameter set from VLE (Demirel these figures. 0.8000) –1.05 (0.7765×10–5 26 (0. which just overall compositions.0025×10–5 (6..20 in mole fraction of n-pentanol are calculations is not recommended (Poling et al. system.9500) 0.583.02 0.0013×10–5 (0. the result is 0.8495 × 10–3 (6.61 (0.06 0. system Moreover. pp.004 0.8214×10–5 1.1000.0000×10–8 0.80).9168×10 .5086 × 10-2.1500.85) (0.1000. which refer to a VLE set of and x = (6.0000 * Results obtained from Gecegormez and Demirel (2005) 0.14 0.September.1500. The NRTL parameters. 0. so the system Gecegormez and Demirel (2005).05.1500. 0. Brazilian Journal of Chemical Engineering Vol.16 0. As clearly shown in this work. 0.7473×10–5 (0.90) (0. Table 2: Results of phase stability analysis for the n-pentanol (1) + 2. 0.010 TPD 0. the value presented in Table 2. 571 .1294.9308) –2 –4.05.8500) –9. the nontrivial solutions reported by and Gecegormez.9349).20. however. 0.05. Phase Stability Analysis of Liquid-Liquid Equilibrium with Stochastic Methods 575 Case 2: n-pentanol (1) + 2. parameters previously reported in the literature.9000) 9.2000.8511 × 10–17 1.8214×10–5 (6.10.05 and 0.0000 (0.1500.80) ∂x i i ≠ j reported by Gecegormez and Demirel (2005). 0.9308) 5. 0.2-dimethylbutane (2) has only one phase in both cases. those presented by Gecegormez and Demirel (2005). our calculations only indicated the existence It can be observed in Table 2 that for the two of a one-phase.8000) 0.05.53 25 (0.0229. July .9168×10–2. and for z = (0.0000 26 (0.9308) –2 5.9187×10 .002 0 -0.95) (0. 0. 0. 2000). 0.0500. 0. 0.1395 × 10–16 0.0000 × 10–8 0.12 0. corroborates phase equilibrium data reported by 0. 03.012 0.15. were For all overall compositions tested for this taken from Demirel and Gecegormez (1991). In fact. one should call attention to the fact that The TPD curves for mixtures containing respectively the use of a set of parameters coming from VLE to LLE 0.8706) –1.014 0. No.18 0.8500) –9.1500.10 0. 0.0500.020 0.002 0 0. 0.8204×10–5 (0. 0.20 x1 Figure 1: TPD curve for the n-pentanol (1) + 2. 0.

7500.4000) 0. 0.70) (0. 0.4091 × 10–2 1.20. 0.30 x1 Figure 2: TPD curve for the n-pentanol (1) + 2.8644.22 0.6230 × 10–17 4.09 (0. 0.0500.1356) –6. 0.08 (0.83 (0.6000) 1.5232) 2. 0.8000) 1. SubDivNL Simulated Annealing Interval Newton* (z1.2000. 0.0380×10–2 (0.0500.5432 × 10–16 4.9322×10–3 0. 0.1217) –9.4768.63 (0. 0.1597) 0.7671 × 10–3 1. 0.1217) –9. L. 0.2500) 1. for the water overall molar fractions of 0.3000. 0.1214) –9.7671×10–3 (0.14 0.60. 0.7044) 8. 0.6590.3761×10–3 (0.9322×10–3 23 (0.13 (0.1426.2025 × 10–2 1.2025×10–2 (0.05. 0.8439.11 (0.9000) 1.7044) 8. x2) (x) (s) (0. 0.25. 0.8403.1291) –7. 0.6000.5000) 0.1291) –7.8000) 0.8384×10–2 1. 0.6590.1000. respectively.6000) 0.9500) –4.5009×10–5 0.4000) 1.7671×10–3 (0. 0.008 0. 0.1198. 0. 0. x2) (x) (s) (x1.07 (0.8783.8783. 0.2151 × 10–2 1. 0. Rossi.1426.03 (0. 0.7895.8185) –8.8874) –8.002 0.75. 0. 0. 0.0000 (0.8874) –8.1000.2000.1815.3869×10–4 (0.4479×10–2 (0. Corazza 0.1269×10–4 24 (0.7355×10–2 (0. According to were taken from Gecegormez and Demirel (2005). Cardozo Filho. 0. 0. 0. L. 0.4999×10–5 26 (0. 0. 0.2-dimetilbutane (2) system at 25 oC and z1 = 0.8574) –3.6344) 3.8786.1484×10–2 (0.8786.7794×10–3 (0.7000) 1.010 0.26 0.8783.1126.80) (0.0000×10–9 0.10 0. R.2463.15.09 (0. Nagatani.50. this system has only one LLE for the water (1) + butyl glycol (2) system at stable phase at these compositions. 0.7537) –1. S.1815.7537) –1.1484 × 10–2 1.1562) –7.1217) –9.7500) 1. 0. 0.4768.7324. As depicted in Figures 3 and 4 the nontrivial The NRTL parameters for this system. 0. 0.20.8384×10–2 (0. 0. 0.4000. 0.0000 (0. z2) Stationary points TPD CPU Global minimum TPD CPU Stationary points TPD CPU (x1. are not true roots. C. x2) (x) (s) (x1. 0.6000.1271×10–4 0. Ferrari.1642) –2. Guirardello.1126.8874) –8.8438. Table 3: Results of phase stability analysis for the water (1) + butyl glycol (2) system at 5ºC.2025×10–2 (0.18 0.4091×10–2 (0.7355×10–2 (0. 0.0000×10–9 (0.0000 (0.3656.1356) –6.2463. 0.5000. 0.1562) –7. Vladimir Oliveira and M.6344) 3. 0.2105) –5.70 (0.8500) 1.5000.2105) –5.57 (0.4708 (0.8709.1291) –7.7794×10–3 (0.9500) 0.60) (0. 0. 0. 0. 0. 0.6938×10–2 22 (0.1198.1356) –6.3410) 6.8438.3000.1198.0000 22 (0.30.2500.9500) 1.2500. 0.4479×10–2 1.0000×10–9 (0.90) (0.8802) –6.8185) –8. 0.4000.85) (0.0000 24 (0.1642) –2.1000.0000×10–8 0. 0.2956.0000×10–9 0.1255 (0. 0. J.10.0000×10–8 0.5000) 1. 0.0000 23 (0. C.75) (0.1561) –7.4091×10–2 (0.3761×10–3 (0.9000) 0.0000×10–9 (0. J.50 (0.8802) –6.40.1500.1500.5232) 2.7895.2956.2766) 0.8358. 0.004 0.9000) –1.50) (0. 0.8802) –6. 0. 0.0000 (0. 0.09 (0. 0. 0.05 39 (0.95) (0.2500) 0. 0.45 (0.40) (0.02 0. 0.8500) 0.006 TPD 0. 0.0000 × 10–8 0. 0. Case 3: water (1) + butyl glycol (2) system 5ºC and several global compositions are presented in Table 3.1214) –9.1431.7000) 0. 0.002 0 -0.8644.8384×10–2 (0. R.8358.25) (0.4479×10–2 (0.8569) –3.91 (0.3410) 6.05 and 1991) and refer to a set of parameters from VLE. the results from the SA and SubDivNL algorithms Solutions of the phase stability test in relation to presented in this work. 0.8644.0000 (0.1126. which are roots reported by Gecegormez and Demirel (2005) presented in the literature (Demirel and Gecegormez.10.7500) 0.0000×10–8 (0.40 (0.6938×10–2 0.05 (0.7500.8709. 0.2151×10–2 (0. 576 G.3656.1484×10–2 (0. 0. 0.0380×10–2 (0.8709.06 0.0000 23 (0.04 30 (0.8574) –3.0000 * Results obtained from Gecegormez and Demirel (2005) Brazilian Journal of Chemical Engineering .3868×10–4 (0.8786.0500.0380 × 10–2 1. 0.0000×10–8 0.2151×10–2 (0.1214) –9.10 (0.

0.583.September.4 0.0682. 0. 0.1609.1 0.1456.3 0.85). presented in the literature. experimental diagram for this system is presented in 0.05. 2008 .8 0.0855.4411) with TPD = -0.2 0.05 0 0.7 0. 0. pp. 0. = (0.25 0. For the global composition of z = (0.4743.2 0.4 0. some divergences between of two liquid phases. 0. 25. homogeneous region.05.10 0.05.9 1.6 0. 0. application of the SA and our results and those presented in the literature can SubDivNL algorithms did not result in the be verified. while this root Figure 5. 0.70) and b) by Lintomen et al.3 0. for the global the literature (Lintomen et al. 0..10.9 1.60 0. was not found by the SA and SubDivNL The LLE binary parameters of the NRTL model algorithms.6 0. Phase Stability Analysis of Liquid-Liquid Equilibrium with Stochastic Methods 577 0.0 x1 Figure 3: TPD curve for the water (1) + butyl glycol (2) system at 5oC and z1 = 0.5 0. No. a second root.40 TPD 0. Table 4 contains the z = (0.5 0.30. The experimental data were obtained from is not reported in the literature.05.8 0. originally reported overall compositions: a) z = (0. 03.10.7862) and b) z system at 25ºC and different overall compositions. Brazilian Journal of Chemical Engineering Vol.65) that correspond to mass results of phase stability analysis for this ternary fractions of a) z = (0. In 0.10.05 0 -0.20 0 0. found in the literature. indicating the formation As shown in Table 4. the ternary value found in this work was x = (8.7625). the composition with the lowest TPD order to elucidate these results.20 0 -0.4640×10-2. and as one can composition z = (0. July .0 x1 Figure 4: TPD curve for the water (1) + butyl glycol (2) system at 5oC and z1 = 0.7 0.15 TPD 0. 0. It can also be verified agreement with the results found in this work. the two feed points (a) and (b) lie in a and SA algorithms found an unstable composition. 0. 571 .05. 0. 0. which is in complete not reported in the literature. for this system were used as presented by Another discrepancy was verified for the two Gecegormez and Demirel (2005). Case 4: water (1) + citric acid (2)+ 2-butanol (3) from Table 4 that for the last third overall system composition. occurrence of separate liquid phases (see Table 4).30 0.1 0. respectively.85) the SubDivNL see. While Solutions found in this work are compared with the unstable roots from the phase stability test are roots presented by Gecegormez and Demirel (2005).20 0.25. (2001). 2001).1520.80 0. Also.

2000. 0. 0. 0. –6.7558×10−2.3523×10−2.1828×10−2.60) (0.9495) 0. 0.9495) (0.6500) 0.2177. 7. 0.2177. 0.0500. 0.1500.2501×10–2 (4.0000 (0.1910×10−2.1257.3989. 0.0500.1877×10−2. 1. 0.73 0.9495) (4. 0.1000.0500. 0.11 (0. 3.3676×10−3.6098×10−2.5985) –1.9789) (3.6709) −2 −3 –2 (2.7352) 6. 0.8972×10−2.3523×10−2.4964.5722) –5.2000.15.1080. 0. –9.59 0. 0. 0.70) (0. L.0321×10–2 (0. 3. 0.35) (0.5500. Guirardello.06 –0. 0. 0.15.7000) 0.2072×10−3. 0.7000) 0. 0.7500) 1.7500) 1. 5. 0.7323×10–2 (0.8972×10−2.5500. 0. 578 G.0000 (0.3189×10−2.61 (0.0000×10–9 0.6000.1878. –4.1506. 0. –0.91 –2.6000.6467) –4.20.0000 (0.1828×10−2.4000) 1. 2. 0.1506. Corazza Table 4: Results of phase stability analysis for the water (1) + citric acid (2) + 2-butanol (3) system at 45ºC.1500.1000.0500.1500. 0.2500.5649) –3.9554) 0.89 –4. 0.0000 2. 0.5000.7500) 0.2072×10−3.4411) –0.15.3218.0000 2.1000. 0.1126.05 (0.1000.7000) 1.7713×10–2 (0.3500) 1.4743.10. 2.50. 0. 1.9549×10–3 0. 0. 0.9554) (4.86 (0.1317×10–4 (0.0000×10–8 (0.1000. S.3189×10−2.1409 1.9195) 0. 2.1838. 9. 2.0000×10–9 0.66 0. 0.8000) (0.9133) * Results obtained from Gecegormez and Demirel (2005) Brazilian Journal of Chemical Engineering .2907×10−2.0052×10 –2 1.8812) –1. 0. 0. 0.0000×10–8 (0.60.8500) 1.0000×10–8 (0.1628×10−3.9798) 1.1748.7000) 1. 0. –5.9216) (7.3257×10−2. Nagatani. 0.70) (0.1000.75) (8. 7. 2. x2.2000.0000 (8.4285 × 10−2. 0.9195) 0.3000. L.3989.8376×10–3 (1.1534×10–2 (7.0909×10−3. –4. 0.5722) –5.8393×10−2.59 0.6141) –1. 0. 0. 0.5649) 0.7972) –2.6271×10–2 1. 0. (0.2530. –2 0.27 –9.1910×10 . Cardozo Filho. 0. 0.7717×10−3.1534×10 1. –2.3396×10−3.70) (9. R. 0. 0.3000. 0. 0. x2.3523×10−2.80 –8.10.55.0000×10–8 (0.5662×10–3 (1.2567. 0.8812) –1.8812) –1. 0. C. 3.9447 × 10–3 1. 0. (0.7500) 0.9094) 0. –0.6537×10−2. 0. 0.3639 (0. (0. 0.1272×10−3.15.0321×10–2 1.3500) 0.8393×10−2. 0.84 –6.2530.1000. x3) (x) (s) (x1. 0.05. 0.0000×10–8 1.8393×10−2.7352) 6.7713×10–2 1. 1. 0.5722) –5.0052×10–2 (0. –8.05. 0. (0.0000×10–9 0.6000) 0.75) (0.5000. 9.3981 0.3000.03 (0. 2. 0. 0. 0.15.27 (2. 0.1000.5662×10–3 (0.0500.3396×10−3.1272×10−3.0000×10–9 (0.1877×10−2.01 (0.1085.3676×10 .2500.0000 (4. 0.0909×10−3.8500) 1. 0.9754×10−2.9447×10–3 (0. 0.1738×10–2 (8.0500.0768×10−2. 0. 0.8336×10–3 0.1752. 2. 0.6000.9494×10−2.6000) 0.6000) 1. 0. 0.6500) 1.0500. 1.2000.10. 1.0500. 5. (0.1126. 0.1534×10–2 (0. 0.1828×10−2.0052×10–2 –9.05.7500) 0.85) (2.1878. 0. 0.3405×10–3 0. 0.2000. 3.1095.1192 1. 0.0000 (4. 0.3500) 0. –4.3698×10–4 (0.8376×10–3 (0. 0.75) (0.0000×10–8 (0.1500.67 0.2000. 3.30. 0. 0. 0. SubDivNL Simulated Annealing Interval Newton* (z1. J. 0.9763×10−2. z3) Stationary points TPD CPU Global minimum TPD CPU Stationary points TPD (x1.7500) 0. 9. 0.8830) 0.40) (0. 0.9536) (4.1006.1095.7717×10−3.9747) 1.8500) 0. Vladimir Oliveira and M.3230. 0.2501×10–2 (0. 0.1000.0000×10–9 (0.3230. 0.4738) –0.0500.0500. 0.10. 0. 0.0500. 0.1929×10–2 0.10.6500) 0.2374 (0.1000.3000.8624) –1.0000 2.4640×10−2. R. 0.0500.20.2000. 0.98 (0. 0. 1.2353.0000 0.75) (4.1609 1.9094) (0. 0.9798) (1.6996) 1.2000.1500.9747) 0.0500.3500) 0.72 0.4000) 0.4729×10–2 (0.1738×10–2 1.2000.5548) –1.0000 (0.6280×10−2. -4. 1. 0.6271×10–2 1.5985) –1.7972) –2.1568.1409 (0.7113×10−2.6353) –4.2657.2072×10−3.5875) –1. –5.6405×10−2. 0. 0. x3) (x) (6.0000 2.8877) 0.0500. 0. 0.1500. 5. 0.0500. –6.05.25.0909×10−3.1257.5875) –1.67 (0.9554) (9.9798) (4. 0. 2. 0.9534×10–3 0.1544. 3. 0.6280×10−2. 0. 1.2567.1929×10–2 (0.9094) (8.20. 0.3500) 0.2000. 0.4738) –0.0768×10−2.8004×10−3. 0. (1.05.10. 0.2907×10−2.7000) 0.7500) 1. 0.0000×10–8 0.4729×10–2 1.0000 (7.3698×10–4 (0.1095.1192 (0.0500.2500.7500) 1.7000) 0. –4.7558×10−2. 0.1000.5224×10−2. x3) (x) (s) (x1.1838.10.4700×10–3 (6.0000×10–8 0.80 (0.97 (0. 0. 0. (0.3104×10−3.1000.5345×10–2 (0. 0.4729×10–2 (0.0321×10–2 (0.1000. J.3104×10−3.0000 (4.1000. Ferrari. 0. 3.1192 (1.2773×10–4 (0. 0. 0.64 0.9789) –3.6467) –4. 0. 0.9754×10−2.1739 × 10–2 (0.6353) –4. 0. 0.4700×10–3 (0.1000.0000×10–8 1. 0. 0.3257×10−2.3396×10−3.3104×10−3. 0. 0. 0.7000) 1. –4. 0.65) 0. –2. 0. 5.0000 (0.3189×10−2. –8.64 0.2530.9494×10−2.9536) (8.70) (0.1500. 0.5500.91 (8.7000) 0.2501×10 1. 0. C. 0. 2.3048. z2.1000. 0.9789) (1.05.9536) 0.1000. 0. 0.0000 (0. 0.6000) 1.6537×10−2. 0.7323 × 10–2 (0.0000 (0.0500. 0. 0. 0. 0.30.3000.3500) 0.89 –5.1500. (0.8500) 0. 0. 0. 0. 0.30 0.1748.3405×10–3 (0.1748.59 0.0000×10–8 1.9447×10–3 (0.0593×10−2. 0. –2 0.8624) –1. Rossi.60) (0.61 (7. 0. 0.00×10–8 (0.20.4964. 0. 0. 0.2908. 0.7000) 1.2353. 0.1609 (0.8000) 0.1628×10−3.5345×10–2 (4.3218.0000×10–8 1. 0.0000 0. 0.1000.0500. 0.3257×10−2.1257.9763×10−2.4411) –0.5548) –1.0000 (0.8004×10−3. 0.3500) 1.10.9216) (0.1752. –4.8336×10–3 (0.05.1544.5500.9549×10–3 0.0500.1628×10−3.1877×10−2. x2.1272×10−3. 0. 0. 1.8972×10−2.7113×10−2. 0. 0.2000. 0.7717×10−3.1500.66 0. 0.80) (3. 0. 0. 1.8624) –1. 0.2772×10–4 0. 3.0000 (0.72 (0.4743.6996) 1.2908.2000.1500.97 (0.0000 2.7113×10−2. 0.5345×10–2 1.80 0.20.7349×10–2 (0.4640×10−2.1000.9216) 0.1000.35) 0. (0.10.1568. 0.7935×10−3. 3.6790) –3. 7.1080.20.6141) –1.85) (8. 0.7713×10–2 (4.8004×10−3.7349×10 (2.0000 (0.05. 0.

The maximum number of subdivisions was again (3) + water (4) system set to four.1860×10–2. taken from Tessier et al. z2.4000.02 0. 0.4500. the maximum number of (3) system subdivisions was set to four. 03.1085 6.0000×10–9 (0.9114. 0. (2000).6723×10–2. 0. (0.44 0. pp. 0. (2000).0000×10–8 1. 2. x2.05.0000 7. –8.3 0.7304) −5. 0. whose results are compared in this Case 6: n-propanol (1) + n-butanol (2) + benzene work. It can be verified that both algorithms employed in this work were able to Table 5 contains the results of phase stability find the same solutions as those presented in the analysis for this ternary system at 45ºC and different literature (Tessier et al.0500.3500) 0. 4. (0. 4.1320. 571 . Case 5: acetonitrile (1) + benzene (2) + n-heptane the computational effort. (0.06 0. x2.04 0.5500) 0.3584×10–2.50) (0. 0. (0.6610×10–2 7.1024) –2. solutions obtained with both system are presented in Table 6. 4.9049.5000) 1. 2. 0.16 (a) (b) 0. (z1.5 0.55) (0.5500) 1.2490×10–2 Brazilian Journal of Chemical Engineering Vol.8213) –8. 0. system the NRTL parameters were also taken from Tessier et al.12 0.1389×10–3 1. 0.4500.0000 (0.00 0. July .0816×10–2 1.6 0.0500.1 0.05.8013×10–2. x2.80 –0.92 –8.583.1320. 0.6723×10–2. 0.08 0.4 0. table.14 0.8013×10–2.9 w3 Figure 5: Experimental diagram of the water (1) + citric acid (2) + 2-butanol (3) ternary system at 25 oC. 2000).47 0.05.0500. x3) (x1.5000) 0. 0. 0. 0. 0. The binary mathematical methods employed in this work are in interaction parameters for the NRTL model were agreement with those reported by Tessier et al. Phase Stability Analysis of Liquid-Liquid Equilibrium with Stochastic Methods 579 0.7608) –1. 25.45.0500. 0.4775×10–2. 3.60. 0. 4.5255×10–2 1.2215.7304) –5.8 0. 0.4000. This additional root however (3) + ethanol (4) + water (5) system is not presented in the work of Gecegormez and Demirel (2005). x3) (0. application of the subdivision algorithm indicated the existence of a third root for all Case 7: n-propanol (1) + n-butanol (2) + benzene compositions analyzed. For this terms of CPU time than the subdivision method. 2.4978×10–2) (0.6000.8213) (0.1389×10–3 (0.3584×10–2.10 w2 0.7275×10–2. (2000).6000.1919.3500) 1. 4.0500.2 0.0816×10–2 (0. 0.9114. 4.18 0.35) 0.1919.4775×10–2. (0.7608) −1.8658. For this problem. No.6615×10–2 1.1320. –0..5255×10–2 (0.September. As shown in this for systems with larger number of components.45 0. z3) SubDivNL Simulated Annealing Interval Newton* Stationary points Global minimum Stationary points TPD (x) CPU (s) TPD (x) CPU (s) TPD (x) (x1.0000 6.7 0. (0. 4.81 –8. The CPU time values overall compositions. x3) (x1. 2008 .40. the Simulated Table 7 depicts the results of the thermodynamic Annealing algorithm had a better performance in stability test for this multicomponent system. 2.0367×10–2) (0. 0.9049.2215.6723×10–2. Solutions found in this work listed in this table confirm the advantageous are compared with the roots presented by performance of the Simulated Annealing algorithm Gecegormez and Demirel (2005). Although the Simulated Annealing algorithm had a better performance in terms of CPU Solutions of the phase stability test for this time than the SubDivNL. 0.1085 1. 0.0367×10–2) (0.8213) −8.0500. Aiming at reducing who used the Interval Newton method.4978×10–2) (0.0000×10–9 (0.0816×10–2 (0.7275×10–2. Table 5: Results of phase stability analysis for the acetonitrile (1) + benzene (2) + n-heptane (3) system at 45ºC.

3.200) 4.2279×10–2. 5. 0.0000 (0. 0.0015 × 10–0 5. x3. 2.1805. z2.0000 (0.1387.6805.98 × 10–3. 0. R. 0.74×10–3.15.1006.0000×10–9 1.35) (0.0000 0.4000) 0.1863. x2. 0.1 1. (0. 4.6000.0000 0.400.7528.8945×10–2.29×10–2) 4.20×10–2. 9. 0.1006. 0. 0.4500.977) 4. 0.05. 1. 0.0660×10–2 (3. (0. x2.600.67 × 10–2.55 0. 0. 0. (0. x2. 0.45.7327) –2.9260×10 . 8.1085) –1.200) 1.6500. 0.953) (0.20 (0.0000×10–8 (0.0477×10–2 1.206.39 –4.71×10–3.7718×10–2 1. 9. 0.133.3500) (0. 0. 0. 0.1279×10–3 303.8579×10–3 (0.37×10–3.4500.7718×10–2 (0.3110 (2.1536.8526. 0.150.8389) (0.6916×10–2 1.61×10 . 0.6518) –1.3398 2. (0.3398 (1.0000×10–9 1.250. 0.74×10–3.952) 6.1548. 7.67×10 . (0.60.6000.48×10–3.1029) –2.957) 6. x3) (0.5000.1000.0477×10–2 (0.41×10–2. 0. 7.1279×10–3 (2.250) 0.916.2000.07×10–2. 2. 0.1246. 0.0477×10–2 (0.7736.31 (0. 0. (0.70.952) (0. 0.1000.8945×10–2.3500) 0.48×10–3.7508) (0.6914) –5. 0.5500. 0. 0. 0.600.1538.55 0. 0. 0.140. 0.8 –0.6034×10–2 1. 0.4032×10–2 (0.052. 0.32×10–2.1387. R.977) (0. x3) (x1.0000 (0.0000 0.700.20.25) 5. 0.4029×10–3 * Results obtained from Gecegormez and Demirel (2005) Table 6: Results of phase stability analysis for the n-propanol (1) + n-butanol (2) + benzene (3) + water(4) system.8631×10–2 (0.957) 0. 3.613) (0.2500) 0.4500. 0.4601×10–2 9.69×10–3.53×10 .55 0. 0.3626×10–2 (3.2500) 0.150.1443) –1.1700.55. 0. 0. x4) (x1. 0. 2. x2.98×10 . 1.2500) 1. 8.148.0000 × 10–8 (0.86×10–2. 1.0000 –3. C.15. 0.1465. 0.75×10–3.25) 0. 0.970) 4.2086. 0. 4.25) (0.3398 4.6429×10–3 (0.25. 9.4000) 1.5500. 0. (0.200) 0. (0.1538. –0.0000×10–9 1.0075×10–2 (0.40. 0.37 × 10–3.3650×10–2 0.5500. 0. 7. 0.735) (0.69×10–3. 0.20) 0.250. 2. 580 G.148.15. 0. 0. 0.29×10–2) (0.4032×10–2 (0.41×10–2.250.47×10–2. 0.3500) 0.0000 8. z3. 0. Rossi.6500. 0.1000. 0.5606×10–2.7685) –8.77 –0.2094×10–2 (0. z3) SubDivNL Simulated Annealing Interval Newton* Stationary points Global minimum Stationary points TPD (x) CPU (s) TPD (x) CPU (s) TPD (x) (x1.9260×10–2.32×10–2. 0.74×10–3.71×10–3.20×10–2.58 0. 0. –2 –3 3.3500) 0.7508) –4.250.2000.970) 0. 1. 0. 0.250.4032×10–2 1. 0.45.2154. x5) (x1.1085) –1.1442) –1.100) 0. 0. –3.0660×10–2 278.02×10–2. 4.4573×10–2.81×10–2. x3) (x1.4601×10–2 1.3626×10–2 2. 6.1334 (0.5857) −8.25) (0.9712×10–3 (0.6914) −5. –2 –3 –3. SubDivNL Simulated Annealing Interval Newton* (z1. 9. 9.1118.5000.25. (8. (0. Corazza Continuation Table 5 (z1.9260×10–2. x2.5500.1000) 0.6730) –2. x4.148.560) 6.8455×10–2) (0. –2 –2 –3. (0.25. (4.0500. 0. 0. –3.114.1118. (0. 0. 0.1536.2500) 0.8133×10–3 (0.89×10 .1858×10–2. 0. 0. 0.1548.2086. 2. 0.8341.1309.957) (0. 3. 0. 0.1858×10–2.977) 0.56 0.1000. 1. 0.7976. 0.1500.0000×10–8 (0.5000.55 0.5257×10–2. 0.140. 0.0000 0.15. S.3500) 1.250) (3.27 –3. 0. 9. 0. 0.10.22 (0.6914) –5. (2000) Brazilian Journal of Chemical Engineering .40) (0.400.8526.970) (0.25. 0.1500. 0.89×10 .5257×10–2.052.1989.8666 × 10–2 2.0000 (0.7311.1309.47×10–2.7685) −8.195.0000×10–9 1. 9.133. 0.6916×10–2 (0.250) 0. Vladimir Oliveira and M. 0.20.5857) –8. 0.8455×10–2) (0.7718×10–2 (0.86×10–4.10.0000 (0.35) 0.6518) −1.75 0.3794×10–2 (0. 3. 0. 0.1720.206.87×10–2) (0.5 –0.22 (0. (0. J.250.0000×10–9 (0. C. –7.1000.81×10–2.48×10–3. 0.25.148. Guirardello. 0.5606×10–2. 0.560) (0. (0.1863. 0.0000×10–8 1. 0. 0. 0.1014) –2.0790×10–2 6. 7.8631×10–2 1.69×10–3.20×10–2.53×10–2.700.86×10–4. 0.71×10–3. –7. 0.200) 0. 0.0000×10–9 1.0000 (2.20×10–4. L. 0.2000.67×10–2.195.5000. 0. 0.250.16 –3. J. 0.6 –3. 1. 0.1538. (8.854. 1. 0.0000 (0.350.73×10–3. 0. –0.6034×10–2 (0. 0. 0. 0.0500. z2.1309. 0. Ferrari.6730) –2.5320×10–2 (0.2500) 1.953) 0.25 (0.8171.0790×10–2 (3.7327 −2. 5. 0.40) (0.3651×10–2 292.6680×10–2 (3.052.1811×10–2 (0.114. Cardozo Filho. 0. 9. 0. 0. 0. 6.250.19 0. 0. Nagatani. x4) (1. (1.6429×10–3 (0. 0.41×10–2.1460. 0.8171.1500.7508) −4. (0.200) (3. 0. 0. 0.0000×10–8 (0. –3.47 (4. 6.0000×10–9 1.250.59 0.1465.1180×10–2.1500.250. 0.53 0. 0.55. 0.15.700.8389) −0. 6.35) (0. z4) Stationary points Global minimum Stationary points TPD (x) CPU (s) TPD (x) CPU (s) TPD (x) (x1.5606×10–2.1500. 0.25) 0.10. 0.3500) 0. 0.600.250.052.50.73×10 . 0.87×10–2) 0.6051) –2.1039×10–2) (0. 9.7000.1989.148.98×10–3.02×10–2.1118. 3.1000. 0.250.1006. x2.250) * Results obtained from Tessier et al.1334 1. 0.20×10–4.1805.6731) −2.75×10–3.3110 2.1334 (0.48 0.953) 7.1465. 0.–2 (0.1811×10–2 1.3500) 1.3110 4.2500) 0.65. 0. 0.1720. 8.86×10–2.69 0.81×10–2.1000) (3. 0.350. 0.20×10–2. 1. 5. 0.9224×10–2 8. (0. 0. 0. 0.86×10–4.250. 0.8114.52 0.916. 0.3794×10–2 (0.6916×10–2 (0.8665×10–2 (3.3500) 1.150.735) 0. 0. 9.9467×10–2. –0. (0.2500) 1. x3. 6. 2. 5.0000×10–9 (0.052.4500. 0.7685) –8.32×10–2. L.1495) –1. –2 –2 –3.7528. (0.6051) –2.052. 0.1500.37×10–3.7000. 9. 0.150.2023. –0. 7.2085. 1.613) 7.2680×10–2 269.1039×10–2) (0. 1.1548.1805.1460. 0.1474) –1.2500) 1.61×10 .250) 5.20×10–4. 0.9224×10–2 1.50.1536. 0.2000.6051) −2. 0. (0.25) (0. 0.0000×10–9 1. 2. x3.2154.3794 × 10–2 1. 0.8389) –0. 0. –3.854. 2.07×10–2.35.2023.3630×10–2 6.5320×10–2 271.2 –7. 6.8670×10–2 7. 0.0000 (0.148. 0. 0.1863. 4.

July . 0.0 3.267) * Results obtained from Tessier et al. –0.152. 3.46×10–2. 0.18×10–4.784. –4. 0. SubDivNL Simulated Annealing Interval Newton* z4.60×10–2.162.105) (0.500. –8.918) 9.267) 0. 1.31×10–2. 0. x2.00×10–8 (0. 0.15×10–2. z3. –0. 0.7 1.2085×10–3 0.26×10–2. 2. x5) (x1.052. 2.500. 0.90×10–2.2383×10–5 (0.14×10–2.13×10–2. 0.160) 4.180) 0. 0.518) (0.2373×10–5 0.052.95×10–4.946) (2. 5. 1.639.0396×10–4 (0.200) 0. 0.180) 3. 0. –1. 0. 5.136.52×10 .49×10 .520. z2. 3. –2 –4 –0. 0. 0. 4. 6.98 (6. 1. 03. 2.128.24×10–2) (2.052.1086 (2.1284 1.83×10–4.811. –8. 5. 0.822.148.46×10–2. x3. Simulated Annealing f(x) nonlinear equation (-) was shown to be reliable and robust for systems with F(x) system of nonlinear (-) Brazilian Journal of Chemical Engineering Vol.3 1. z5) Stationary points Global minimum Stationary points TPD (x) CPU (s) TPD (x) CPU (s) TPD (x) (x1.42×10–3.86×10–3.63×10–3.98×10–2. 8.560.9019×10–3 (0.September. No.6572×10–2 4. 0. 0. 2. 3. –8.45×10–4. 4.60×10–2. 0. 1.32×10–2) 2. 8. 0. In a general sense.92×10–2. 5.43×10–2.108. 0.946) 0. are two stochastic global algorithms was examined for easy to implement and require relatively low binary. 0.385) 0.942) 0.73×10–3.918) (0.080.79×10–2. 5. (0.25×10–4. 4.24×10–2) 0.73×10–3.60×10–2. 5.148.4793×10–3 1269.1075×10–4 0.100.2085×10–3 (0. 6.148.180) (2.89×10–2. CONCLUSIONS larger numbers of components.26×10–2.00×10–9 1074. x4.55×10–2. 0.784.68×10–2.200) 5. 0. 4. (z1.32 × 10–2.96×10–2.71×10–3. 5.23 (7. 3. 6.59×10–3.81×10–4. 0. 0.397. x3.04×10–2) 3.160) (2.148.89×10–2. 4. 0. 2.63×10–2. 0.00×10–8 (0.77×10–2. ternary and multicomponent liquid mixtures computational effort.163.43×10–2. (0. 4.684.052.83×10–2. 0.04×10–2) (0.012×10–4 (0.942) 4.30×10–2.148. 0. x2.180) (2.108.09×10–4. 6. 2.052.52×10 . 8.100. 1. 8. using the NRTL activity coefficient model. 0.152. 5. –4.0000×10–8 (0. 3.0000 0.60×10–2.540.6580×10–2 1.93×10–2.42×10–3.9581×10–3 0. 1. (2000). 0. –4. x4. –4.81×10–4.540.052.00×10–8 (0.89×10–2.63×10–2. –0.100. x4.120. 1. –0. –0. 0. 0.080. 0. x5) (2. 0. 0. 0.938) (2. –2 –4 –0. 0. –0. 0. 0.148. 5. 5.583.99×10–2.052. 4. 0. 2. 0.83×10–2. 3.73×10–3. 4. 2.918) 4. 0.14 (6.92×10–2.6575×10–2 (2. 0. 0.052.2107×10–3 1314. 0.271.148.148.518) 0.85×10–2.119. 3.45×10–2. –4.148. 8.60 × 10–2) (2.1043 1. 1. 0. (7. 0.55×10–2.64×10–2. 8.136. 5.55×10–2.052. 2008 . 3.271.2 1.794.45×10–4.180) 0. 0. (0.100.64 × 10–2.684. 915. 25. x2.96×10–2.946) 4.938) 0.052. 8.148.0000 0.180) 4. –1.95×10–2. –4.148. 0. (6. 0.560. 5. 0. 0.180) 4.180) (0. 1.1284 7. 0.1068 8.13×10–2.933) (7.080. 2.99×10–2. x5) (x1.052.59×10–3.4793×10–3 0.33×10–2) (0.2107×10–3 0.7480×10–2 2. 0.148.26×10–2. 0. Results of application of the subdivision algorithm to liquid- liquid equilibrium calculations show that it was able NOMENCLATURE to find all roots of phase stability tests for the systems investigated with relatively short CPU times d problem dimension (-) for binary and ternary systems.560.30 × 10–2.156.163. 0.811. –1.942) (2. 0.100.200) (2.639.99×10–2. 0.052.77×10–2. 4.080. 0.49×10 . (0.105) 9.120.500.14×10–2.500.520.540. 0.52×10–2) 3. 0.68×10–2. 0. 0.90×10–2. 0. 0. 0. 8.1068 1.385) (0.148. 0.500.93×10–3.68×10–2. 4.55×10–2. 0.100.7479×10–2 (2.30 (0.81×10–2.116.148.1043 4.080. –4.156. –0.120. pp. 571 . 8.60×10–2) 0.162. x3.0397×10–4 (0.1076×10–4 (0.98×10–2. 0.1284 (2.32×10–2. 6. 5.18×10–4.128.31×10–2.62×10–2.822. –1. 0. 0.052. 5. 1.0107×10–4 0.9581×10–3 1296.52×10–2) (0. 0. 0.080.0000 0.0000 0. –1.933) 3.68×10–2.15×10–2.85×10–2. 0.45×10–4. 2.520. 3.81×10–4.119. 1.180) 0. 2.938) 5.1043 (2.93×10–3.7402×10–2 4.116.31×10–2. –1. 0.9019×10–3 (7. 1.95×10–2.93×10–2. 0. 0.934) 0. 0.13×10–2.148.42×10–3.160) 0. 5.052.052. the two algorithms are relatively simple. (6. 5. provide In this work thermodynamic phase stability with reliable solutions for phase stability analysis.25×10–4.397.25 (0.43×10–2. 0. 0. 0.794. 0.71×10–3. 0.6 1. 1.11×10–3.0000 0. 0. 0.500. Phase Stability Analysis of Liquid-Liquid Equilibrium with Stochastic Methods 581 Table 7: Results of phase stability analysis for the n-propanol (1) + n-butanol (2) + benzene (3) + ethanol (4) + water (5) system.

. Greek Letters Faber. D.. Iglesias- “parent” rectangles at “k” Silva. A. F. and (camphor “j” at the subdivision level +CO2+propane). Muniz. components of the system Lima. and glycidyl τ ij subdivision parameter of the (-) methacrylate at high pressure.. Chem.... R. Rubira.. C. Eng. 4764 y vector of random point in Rij (-) (2006). R. Fluid Phase Equilibria. 237.. J. and Gecegormez. E. A subdivision algorithm for phase z vector of global mole (-) equilibrium calculations at high pressures. dimension Eng. P.. Ind. C.. rectangle “j” at the Data. Res. M. 51. 22.. 273 (2005). 65. Modeling and analysis of 0 pure component (-) the isothermal flash problem and its calculation n number of components in (-) with the Simulated Annealing algorithm. L. Bk upper coordinate of the (-) Bonilla-Petriciolet. A. C. Cardozo-Filho. Cardozo Filho. Nagatani. A.. Tavares. Kunita. Chem. at “k” dimension E. and Luks. L.P. Corazza equations ACKNOWLEDGEMENTS R multidimensional rectangle (-) ak lower coordinate of the (-) The authors thank CENPES/PETROBRAS. Vladimir Oliveira and M.. H.. Oliveira.. H. Vazquez-Roman. 51. point of “parent” rectangle Franceschi.582 G. fraction and Dariva. Jr. Chem. 40. Biscaia. and Meireles. Chem. F. R. J. K.. Phase stability αm.. subdivision level “i” Gecegormez. C. L. R. Guirardello. 113 (2004). L. composition J.. Amaral Souto. Filho. Oliveira Jr. Ferrari. x vector of liquid phase mole (-) Corazza. S. N. K. A robust strategy for SLV equilibrium iRan sample number in Rij (-) calculations at high pressures. H. 6028 (2001). 45. 111 (1991). and Hall. A. M... N. J. and Marques. 997 “i”’ (2006). Subscripts Henderson. A. C. “child” rectangles at “k” FAPERGS and CNPq for the scholarships and dimension financial support of this work. C. Res. E. Corazza. Fluid Phase Equilibria. E. Jockenhövel.. J. E. A. and Tsatsaronis.. Soc. Performance of dimension stochastic global optimization in the calculation NRTL nonrandom two liquid (-) of phase stability analyses for nonreactive and i subdivision level (-) reactive mixtures.n matrix of zero and one (-) analysis using Interval Newton method with elements NRTL model. 48 (2005). Cardozo Filho. Eng. Phase behavior of binary and ternary systems component “i” involving carbon dioxide.. M. W. L.. Rossi. νk coordinate of the middle (-) Computers & Chem. R.. T. L. Eng. M. L. and Dariva. 686 (2006). V. Phase equilibrium for xij middle point of the rectangle (-) (camphor+CO2). (camphor+propane). 731 (1982).. Simultaneous iCov maximum coverage number (-) representation of excess enthalpy and vapor- TPD tangent plane distance (-) liquid equilibrium data by NRTL and UNIQUAC models. G. Y. Dynamic optimization with Simulated Annealing. A.. C. Corazza... F. Ind. Data. γi activity coefficient of (-) C. R.. Brazilian Journal of Chemical Engineering . In press (2007).. and vapor phase mole Carvalho Jr. propane. Pierce. T temperature (-) Corazza. Eng. Braz.. M.. R. Chem.. H. M. Y.. Fluid Phase iMax maximum subdivision (-) Equilibria. L. L. bk upper coordinate of the (-) “child” rectangles at “k” dimension REFERENCES Ak lower coordinate of the (-) “parent” rectangles at “k” Baker... C. G. J. J. R. J. R. J. Eng. Corazza. C. Oliveira. and T0 initial annealing temperature (-) Dariva. V. and Demirel. Pet. 221. number Demirel. the mixture and nth Eng. 29. fraction L.

C. 41. Application to NRTL and UNIQUAC equations... Eng. and Rangaiah. 1. Numerical Recipes in Reliable phase stability analysis for excess Gibbs FORTRAN . G. T.. M. Excess Gibbs Annealing algorithm. Fluid Phase Equilibria. vulgare) extract + CO2. and Wen. energies of binary systems of isopentanol and n.. Renon.. J. M. Cardozo-Filho. Fluid Phase Equilibria..M. 2008 .. Computers & Chem. and Rosenbluth. T.... and Seider. H. Fluid Phase Equilibria. Eng. R. L. Chem. B. W. Eng. 154.E. 1 (1982). A. and Sun. 49. 135 (1968). and Xu. energy models. Press. J. 9.. L.The Art of Scientific Computing. M. C. Eng.. Data. Fluid Phase Equilibria. S. Global Application of interval analysis for Gibbs and optimization for the phase stability problem. 55.. global optimization of continuous functions with The Properties of Gases and Liquids. 137. Teh. 55 (1999). G. Veterlling. New York (2000).. M. E. and Tessier.. computing machines. 1087 Guirardello. M. Equation of state calculations by fast Souza.15 K. A. Meirelles. Eng. J. A. J. equilibrium of the water+citric acid+short chain Computational and Applied Math. and Floudas. Phase Stability Analysis of Liquid-Liquid Equilibrium with Stochastic Methods 583 Utilização de Algoritmos Heurísticos de pentanol with hexane isomers at 25 oC: Otimização para Cálculo de Equilíbrio de Fases. N. W. Souza. C.September. Cardozo-Filho. 1785 2nd ed. Part I. 103. and O’Connell. P. Lintomen. Chem. J. Chem. Brazilian Journal of Chemical Engineering Vol. Phys. Phase (1953). F. W. M. Flannery. and Ratcliff. A. Data. 571 . A. Y.. 14. Chem.. S. 1657 (2005).. 203 (1995). J. 50.. M. Eng. M. C. Prausnitz. 117 (2006). 23. Eng. 46. Local compositions in Zhu. Homotopy- Meireles. Chem.. Science.. J. and Wolff.. N. A. 71 Química – COBEQ. An algorithm for A. Zhu. A. Pinto. Corazza. R. Metropolis. Y. (1992). Cardozo-Filho. B. application to phase equilibrium calculations.. and Chun. F. L. P. McDonald. J. W. 5th ed. Evaluation of genetic algorithms global phase stability for liquid-liquid equilibrium and Simulated Annealing for phase equilibrium through the Simulated Annealing algorithm: and stability problems. Chem. Data. No. Phase equilibrium continuation method for stability analysis in the measurements for the system fennel (Foeniculum global minimization of the Gibbs free energy.. Stability. Science. 83 (2001).. 352 (2004). P. 03. Tabu search for Poling. A.. J. 1665 (2003). S. W.. (Eugenia caryophyllus) oil + CO2. July .. Liquid-liquid finding all solutions of a nonlinear system. 55. Y. AIChE J. Brennecke. McGraw Hill. S. 25. J. L. Santos (2006). equilibrium measurements for the system clove Michelsen. and Stadtherr. A. S. 293 alcohol+tricaprylin system at 298. J.P.. A. Moura.. A reliable prediction of the Rangaiah. 187-188. P. T. Rosenbluth. high pressures using the enhanced Simulated Sayegh.. L. A. Z. 3451 (2000). M. G. Braz. Teukolsky. G.. (1976).. L. R.. R. and Praunitz.. Batista. H. A. Global stability thermodynamic excess functions for liquid analysis and phase equilibrium calculations at mixtures. Cambridge University Press. 27. M. W. and Meireles. Data.583. 546 (2001). 21. L. J.. Helmholtz free energy global minimization in AIChE J.. A. Corazza. The isothermal flash problem. and Maciel. New York (2000). Eng.. (2001).. Chem. H. D.... T. Xu. Measurement and prediction by analytical group In: XVI Congresso Brasileiro de Engenharia solution model. 21. pp. 1798 (1995). Smiley. phase stability analysis. Chem. Z.