Computer Aided Property Estimation for Process and Product Design G.M. Kontogeorgis and R.

Gani (Editors) 2004 Elsevier B .V. All rights reserved.

113

Chapter 6: Association Models - The CPA Equation of State

Georgios M. Kontogeorgis 6.1 INTRODUCTION Associating systems are those, which contain compounds capable of hydrogen bonding e.g. alcohols, water, amines, acids, etc. Phase equilibria of complex associating systems is important in many practical cases. For example, in several hydrocabon and chemical processes as well as in order to meet environmental and low-emissions legislations, systems involving a highly polar compound, a non-polar one and a co-solvent, with both polar and non-polar functional groups, are typically present, e.g. systems involving water, alcohols or glycols and hydrocarbons. The role of the alcohol or glycol is that of a co-solvent: they improve significantly the mutual solubility of water and hydrocarbon in the hydrocarbon-rich and water-rich phases, respectively. Without such co-solvents, these solubilities are very low. Examples of cases where phase equilibria of associating systems is important: in the oil industry: - formation of gas hydrates - calculation of the amount of hydrate inhibitors - partitioning of alcohols and other chemicals between water and oil use of alcohols as additives in gasolines for 'Octane Number' improver - heterogeneous azeotropic separation in alcohol manufacturing glycol regeneration units - reid vapor pressure of gasoline in the presence of additives^ for environmental calculations: octanol-water and other partition coefficients - modeling of waste streams in biotechnology and biochemical engineering^'^ in the food product and process design"^ in the chemical industry^: - novel separation methods for pharmaceuticals (e.g. supercritical fluid extraction) - recovery of alcohols from aqueous solutions using high pressure propane (or other gases) [propane is often selected as solvent since no azeotropes are expected in mixtures of ethanol with propane and due to its high pressure extracting capability] in the polymer industry: polymer blends - phase equilibria in coatings of mixed water-organic solvents

114 Typical conventional models like cubic EoS and activity coefficient models (NRTL, UNIFAC - see chapter 4) are often insufficient for such systems, especially for multicomponent and multiphase equilibria, for predictions from binary information alone, for liquid-liquid equilibria, such as prediction of liquid-liquid equilibria for ternary water alcohol- hydrocarbons. Correlation and/or prediction of binary phase equilibria of some complex mixtures are often possible with simple models, such as cubic EoS (see Figure 1). However, prediction of multicomponent phase equilibria from binary data and/or the presence of multiphase equilibria, e.g., vapor-liquid-liquid equilibria are particularly difficult to model with simple cubic equations of state and activity coefficient models.
380 -1

370 -

"^AD -

'^'^n -

c )

0.2

0.4

0.6

0.8

mole fr. methanol

Figure 1. Correlation of vapor-liquid equilibria for water-methanol at 1.013 bar using the SRK EoS and the van der Waals one fluid mixing rules (ki2=-0.0752) Group-contribution UNIFAC-type models and simple cubic EoS with the classical (van der Waals one fluid) mixing rules predict or correlate VLE for some binary systems including associating compounds satisfactorily (Figure 1), less so however for LLE (Figure 2 for a glycol-heptane system), especially when water is present. The limitations of conventional models become further apparent when correlation of alcohol/alkane and water/hydrocarbon VLE and LLE (or similar mixtures) are considered^'^. Over the last years, a variety of models to account for the effects of hydrogen bonding in solutions have been proposed. These are called 'association models'. The association models are designed specifically to describe phase equilibria of mixtures containing highly polar and/or strongly associating compounds such as water, alcohols, acids, amines, phenols, and others, which have the capability of forming hydrogen bonds. The hydrogen bonding formation in the same molecule is called self-association (e.g, pure water or pure methanol), while forming of hydrogen bonding complexes between two different molecules is called cross-association (e.g. water-methanol). Several reviews on association models are

115 available^'^^. No attempt to review these theories will be made here. All theories lie in three main directions: i. ii. iii. chemical theories lattice-fluid theories perturbation theories
1.00E-01 -

1.00E-02 -

5 1 .OOE-03 o 1.00E-04 1 noF OR

o
X

o
X

O X

0 ^

310

320

330 T/K

340

350

360

Figure 2. LLE for ethylene glycol (1) - n-heptane (2) system with the Mathias Copeman (MC)-SRK with a kij = 0.094: x, x", experimental; o, X2, experimental; solid line, x", MC-SRK; dashed line, X2, MC-SRK. / -= Glycol-rich phase and / / = Hydrocarbon-rich phase [Derawi et al.^]. All approaches include two (or more) contributions: a so-called physical term (phys) accounting for the deviations from ideality due to physical forces and an association term (assoc) accounting for the effect of hydrogen bonding (and other "chemical" forces). In terms of the compressibility factor, the two contributions are often (but not always) separate:
2 _ ypf^ys , y assoc

(1)

The physical term can be either a non-cubic or a cubic equation of state. The approach used for describing the hydrogen bonding is different in the three categories of association models. Chemical theories are based on the formation of new species and the extent of association is determined by the number of oligomers formed, as a function of density, temperature and composition. Lattice (often called quasi-chemical) theories account for the number of bonds formed between segments of different molecules that occupy adjacent sites in the lattice. The number of bonds determines the extent of association. Finally, in the perturbation theories, the total energy of hydrogen bonding is calculated from statistical mechanics and the important parameter for hydrogen bonding is in this case the number of bonding sites per molecule.

116 All these theories have been developed especially in the period 1989-1999 and the engineer has now a number of alternatives to select in terms of 'closed-form' thermodynamic models (EoS), where the hydrogen bonding effect is accounted for as a separate term. An important conclusion, independently reached by a number of investigators^'^^'^^ is that all three theories, despite their different physical origin and formulation, yield essentially identical mathematical expressions for the hydrogen bonding contribution. This helps in e.g. the interpretation of the parameters of the popular perturbation theories (SAFT, CPA) with the help of the more familiar chemical theory, where the parameters have a well-established physical meaning (enthalpy and entropy of hydrogen bonding, equilibrium constant). In this chapter, we limit our discussion to the CPA (Chemical-Plus-Association) equation of state, which represents one of the widely applied association models. It combines a cubic equation of state (for the physical term) with the hydrogen bonding term employed in Statistical Associating Fluid Theory, SAFT^^'^"^.

6.2 THE CPA EQUATION OF STATE - MODEL DESCRIPTION The development of the CPA equation of state started as an industrial project, with the participation of industrial and academic partners from Europe. The purpose of 'the CPA project' was to develop an equation of state suitable for complex multicomponent multiphase equilibria for systems of the type: 'water - alcohols - hydrocarbons'. Later the equation of state was extended to cover other associating compounds e.g. glycols and sulfolane. The model can be in principle extended to other types of associating molecules. It was at an early stage realised that to better achieve the above goal without sacrificing the successful performance of cubic equations of state for hydrocarbon mixtures: the equation of state should be predictive for multicomponent systems. Prediction in this sense implies that only binary parameters should be employed, determined from binary data the cubic SRK EoS should be used as the base-model for describing the physical interactions the association term of SAFT can be adopted due to its applicability to different types of hydrogen bonding compounds.

The CPA equation of state has the following form for mixtures:

ySRK _ _}_

/O"!

~V-h

RT{V + b)

1 ^ ax^^
2) dp,

r-

= pY^x^x^Y.X^'

117 The association term is the one employed in SAFT (with a simplified version of the radial distribution function though). For non-associating fluids, e.g, hydrocarbons, the association term disappears and the equation reduces to SRK. In the association term, the key parameter is X^^, the mole fraction of the molecule i not bonded at site A, i.e. the monomer mole fraction. X"^^ is related to the association strength between two association sites belonging in two different molecules (A ' 0 as:
A-I

XA-

l + pXIXyX^^A'^'^^J Bi (3)

r MBj ^ A
AA S ;

' =g(p) exp 'RT'

, M

-1

by^'j

The expression for z^^^^^^ presented in equation 2 is the one originally proposed for SAFT (and CPA) by Chapman et alP and Huang and Radosz^"^. However, in recent works, Yakoumis et alP and Michelsen and Hendriks^^ proposed a surprisingly simple general expression for the association term of CPA (which is identical to the one shown in equation 2):
assoc

-X^i)

(4)

where xi is the (analytical) mole fraction of component i. When comparing equations 2 and 4, it becomes apparent that using the simplified expression equation 4, there is no need for calculating the derivatives of the mole fraction of non-bonded molecules with respect to the density. This issue simplifies the calculations. For a pure associating fluid, equations 2 and 3 are simplified: 1 X^ 1 dx' 2^ dp (5)

x^ =

l + p^X^A^^ B V

X is now the fraction of a single associating molecule non-bonded at site A (=monomer fraction). The key quantity of the association term of CPA is, as in SAFT, the association strength between two sites A and B:

118
r^AB^

A^^=fif(p)| exp

RT

p6

(6)

where g(p) is the so-caUed 'radial distribution function'. Finally, the energy parameter of the EoS is given by a Soave-type temperature dependency, while b is, in agreement with most pubUshed EoS, temperature independent:

a = a(l + c,(l-V77))'

(7)

The parameter ci can be, for inert compounds e.g. hydrocarbons, obtained via the acentric factor as in classical cubic equations of state. For associating substances, it has to be estimated together with the other parameters from pure component (vapour pressure and Hquid density) data. 6.2.1 The radial distribution function: from the original to the simplified CPA SAFT and CPA in the pubHcations by Kontogeorgis et al}^ and all subsequent publications up to 1999 employ the hard-sphere radial distribution function (RDF):

SAFT uses also the hard-sphere radial distribution function. In the case of SAFT this is rigorously correct since the hard-sphere repulsive term is used, while for CPA only approximately correct, since CPA employs the van der Waals repulsive term (of SRK EoS). Kontogeorgis et al}^ suggested the following simplified hard-sphere radial distribution function first proposed by Elliott et ah ^^: g(p)-^ ^^^ . l-1.9y y= ^ W (9)

As shown in Figure 3, the two radial distribution functions yield very similar values, but the simplified hard-sphere radial distribution function offers some computational advantages. Thus, the simplified RDF is recommended. CPA with Eq. 9 is denoted as sCPA (simplified CPA).

6.3 THE CPA EQUATION OF STATE - PARAMETER ESTIMATION The first step in the development of the CPA EoS is to define (assign) the number and type of association sites for a specific associating substance. This is by no means a trivial issue and a carefully-designed parameter estimation procedure is required based on vapor pressures and liquid densities, taking also into account the physical significance of the parameters.

119 Hydrocarbons and other inert compounds, which do not self-associate, have zero association sites.

*2^M* hard sphere OQDQD simplified hard sphere

I 11 I I I I I I I u I M I I 11 I I I I 11 n I I 11 11 I 11 11 I 11 I I I I I I I I

0.0

0.1

0.2

0.3

0.4

0.5

y (=b/4V)
Figure 3. Comparison of the hard sphere RDF (employed in SAFT and CPA), Eq. 8, with the simplified radial distribution function (employed in sCPA), Eq. 9. [Kontogeorgis et al}^] From the investigations carried so far^^'^^'^^'^^, it has been verified that: - alcohols and phenols are best modelled as two-site molecules (with equal sites) (type - water is best modelled as a four site molecule (type 40^"^) The 2B model for alcohols and the 4C one for water are in agreement with the generally accepted physical picture for these two types of associating compounds, i.e. that alcohols form linear oligomers and water forms three-dimensional structures. The notation "2B" and "4C" stems from Huang and Radosz^"^. 6.3.1 Pure Fluids CPA has in total five pure compound parameters for associating substances, three parameters in the physical SRK term (as in all cubic EoS) and two parameters in the association term, the association energy and the association volume. One parameter, the co-volume b, is included in both the physical and the association term, as can be seen by equations 2 and 3.

120 As observed from equation 2, the exact functional form of the association term of CPA depends on the choice of association scheme. Acids are expected to be modeled as one-site molecules (1 in Huang-Radosz notation^"^), alcohols as two-site (2B) and water as 3- or 4- site molecules (3B, 40^"*). These schemes are in agreement with the fact that acids form dimers, alcohols linear oligomers and water three-dimensional structures. However, for other associating molecules (alkanolamines, glycols, etc.) the association schemes need to be investigated together with the parameter estimation. Other association schemes, which are not covered expcilitly in the Huang and Radosz study^"^, may be important for specific associating fluids e.g. inclusion of branches, closed ring structures which are known to be present for hydrogen fluoride, etc. Evidently, the need for determining the correct association scheme gives at the same time flexibility to the model and some problems, since this can be a rather time consuming process. The pure parameters are estimated from thermodynamic data for pure fluids. Vapor pressures and liquid densities are typically employed for this purpose. Multiple sets of parameters can be obtained which yield acceptable errors in both properties (within experimental uncertainty). Interestingly enough, the co-volume parameter seems to be approximately constant among the various sets, independent of the starting values of the regression^ ^. Moreover, b seems to be related to the van der Waals volume (Vw) and, moreover, in a universal way for many compounds. For example, as shown by Zeuthen"^^ for alcohols, glycols and acids the b/Vw ratio of CPA is approximately constant having a value between 1.35 and 1.47. Such considerations may provide a way of reducing the number of pure parameters to be adjusted. There seems that the b-value regressed from pure data for the CPA EoS 'forces' somehow the repulsive vdW term to assume values close to those of the Camahan-Starling expression, a^ shown by Kontogeorgis et al}^. The choice of physically correct set of pure parameters is crucial as emphasized many years ago by Anderko^^. Whether the regressed association parameters are physically correct may be determined upon comparing them with values for the heats and entropies of hydrogen bonding, which can in some cases be obtained from independent (spectroscopic) measurements. Table 1 gives a list of the sCPA parameters published so far and table 2 presents some values of CPA parameters compared to enthalpies of hydrogen bonding and association parameters from other models such as the SAFT, Anderko and APACT EoS. Additional sCPA parameters for other glycols are given by Derawi et al^. In several cases, especially for associating compounds for which it is unclear which association schemes can be used, consideration of some binary VLE/LLE data for the associating compound (under investigation) with (inert) hydrocarbons is a useful way for establishing the best set of pure parameters. This in no way implies, however, that mixture data enter directly into the regression for pure fluid parameters. Hydrocarbons and related compounds are treated as inert compounds, i.e. the association term is zero. In this case the pure parameters are three: the critical temperature, critical pressure and the acentric factor. Alternatively, the three pure fluid parameters (ao, b and Ci in equation 7) can be regressed from vapor pressure and liquid densities sacrificing, thus, the description of the critical point but improving the pure compound description. When this is done, slightly better VLE^^ and typically better LLE correlation is achieved^^.

121 Table 1. sCPA pure-compound parameters for water, methanol, ethylene glycol (MEG), diethylene glycol (DEG) and triethylene glycol (TEG) Compound Reference b ao (dm^ mol"^) (bar dm^ mof^) 0.014515 0.030978 0.0514 0.0921 0.1321 1.2277 4.0531 10.819 26.408 39.126
Cl

Water Methanol MEG DEG TEG

Kontogeorgis etal}' Kontogeorgis etal}^ Derawi et al^ Derawi et al^ Derawi et al^

0.67359 0.43102 0.6744 0.7991 1.1692

s (bardm^ mor^) 166.55 245.91 197.52 196.84 143.37

P
0.0692 0.0161 0.0141 0.0064 0.0188

Table 2. Experimental values of enthalpies of hydrogen bonding for some alcohols and water and the corresponding association parameter (association energy) of several association models. All values are in K.
Compound Expt CPA* APACT' APACT' Anderko^ Anderko^ SAFT' SAFT' SAFT'

water

1813

2003

2618

2418

3505

methanol ethanol Ipropanol 2propanol Ibutanol Ipentanol Ihexanol phenol

2630 25263007
ti

2957

2770 3022 3022 3022 3022 3022 3022

2770 3021 3021 3021

2360 2620 2712 2875

3127 3127 3127 3127 3127 3127 3127

1809 3 site 1368 4site 2714 2759 2619 2670 2605 2587 2556 1894

826

1635

2710 2803 2512 2459

-"n

-"n

2343

Expt: Experimental values. References for the experimental values in column 2 are from : water: Koh et al}^\ methanol: Nath and Bender^^; ethanol: Pimentel and McClellan^"^. Many authors accept that the 'experimental' values for the heats of hydrogen bonding of alcohols heavier than ethanol are close to that of ethanol, i.e. they do not depend considerably on chain length; phenol: Ksiazczak and Moorthi^^ APACT^: Vimalchand et al.^^; APACT^: Economou and Donohue^^ Anderko^: Anderko^^; Anderko^: Anderko^^ SAFT : original SAFT by Huang and Radosz^"^ and Economou and Tsonopoulos^^ SAFT^: simpHfied SAFT by Fu and Sandler^^ SAFT^: re-estimation of water parameters of original SAFT by Voutsas et al.^^ *CPA values for water: sCPA: 2003 K^^ CPA: 1794 K^^

122 An independent test of pure parameters can be made by investigating the performance of the EoS to properties other than vapor pressures and liquid densities, which are employed in the parameter estimation. As shown by Kontogeorgis et al}^, CPA yields very good prediction of second Virial coefficients for alcohols down to very low temperatures. At such extreme conditions where the second Virial coefficients achieve very low values, cubic EoS such as SRK fail completely. The prediction of second Virial coefficients through EoS is a very strict test. It is accepted that cubic EoS with parameters fitted to the critical point and the vapor pressures cannot simultaneously predict the second Virial coefficients. CPA seems to overcome this problem. However, the prediction of second Virial coefficients has not been systematically investigated with the CPA EoS, since such coefficients are not very important for practical applications. 6.3.2 Mixtures: Mixing and Combining Rules The conventional mixing rules are employed in the SRK-part for the energy and co-volume parameters. The geometric mean rule is used for the cross energy parameter and the interaction parameter kjj is, in the applications reported so far, the only adjustable binary parameter of CPA, which is estimated from experimental binary phase equilibrium data:

(10)

a = ^f3^(l-k^j)
No mixing rules are need in the association term, since, as can be seen from equations 2-4, the association theory of Wertheim provides a theoretical extension to mixtures. However, for extending the CPA EoS to cross-associating systems e.g. water-alcohols, alcohol-alcohol, etc., combining rules are needed for estimating the cross association parameters, the energy and volume ones (s,p). An alternative simpler way, which is often adopted, is the so-called Elliott rule. Suresh and Elliott^ suggested a combining rule for the cross-association strength, which is actually the property employed in mixtures:

AA^y=^/7A7^

(11)

Elliott mentioned that equation 11 could only be applied to mixtures of alcohols or other associating fluids where compounds have the same or similar associating behavior (scheme). However, as will be shown in the next section, this mixing rule has been also employed with success for water-alcohol systems. Still, it has been verified that the results become progressively worse as the size of alcohol increases and for systems containing water and heavy glycols. Thus, results are satisfactory for systems with water and methanol, ethanol and propanol, but already for water/butanol the results are not satisfactory. As shown by Voutsas

123 et alP, for water/n-butanol and similar water/heavy alcohols, correlation of liquid-liquid equilibria for this system with a single kjj using the Elliott rule is not successful. It is not possible to correlate at the same time the water-rich and the butanol-rich areas of the binodal curve. Similar results have been obtained for other water/heavy alcohol systems and for both vapor-liquid and liquid-liquid equilibria. We will show later that improved combining rules for the association parameters for such systems have been developped. It has been reported^^ that the use of Elliott rule yields increased computational speed.

6.4 THE CPA EQUATION OF STATE - CURRENT STATUS/RESULTS At its first stages of development, CPA has been successfully applied to: pure alcohols^ ^ pure water^^'^^ binary VLE and LLE for alcohols/hydrocarbons^^'^^ binary LLE for water/hydrocarbons^^'^^ VLE and LLE for cross-associating water/alcohols water/butanol) and water - alcohol - hydrocarbons^^'^^ Gas-liquid equilibria for gas/alcohols and water^^

(water/methanol

and

The CPA model is under continuous development. Recent results include binary and multicomponent systems (VLE, LLE, Henry's law constants) containing aromatic hydrocarbons and glycols, carboxylic acids, ketones, polymers and environmental calculations"^"^""^^. 6.4.1 Discussion of the Results Water / alkanes LLE CPA yields, for water/alkane systems, excellent correlation of the water solubility in the hydrocarbon-rich phase and very good results also for the very low hydrocarbon solubility in the water-rich phase, as shown in Figure 4. As shown in this figure and in the recent work of Voutsas et al ^\ CPA yields much better results than both the SRK and the SAFT equations of state, especially for the hydrocarbon solubility. Nevertheless, none of the models is capable of describing the minimum, which has been experimentally confirmed for the hydrocarbon solubility in the water-rich phase at ambient temperatures. Although, for this particular case, CPA seems to perform better than SAFT, for other systems such as methanol/pentane, both CPA and SAFT perform equally well. The simplicity of the physical term of CPA makes it an attractive alternative over the SAFT EoS. VLE for mixtures with associating substances Figures 5 and 6 show two typical results for water/methane and water/methanol. Yakoumis et al}^ have showed that CPA performs considerably better than SRK in describing VLE for

124 alcohols/alkanes with a single interaction parameter. For such highly non-ideal systems, conventional models with the classical mixing rules typically fail; occasionally SRK yields a false phase split for alcohol/alkanes. The interaction parameter ky of CPA has, moreover, much lower values than in SRK and is a weak function of temperature.
n-hexane In 10 N ooooo water in the DDoaD the water rich phase n-hexane rich phase sCPA (kij=0.0361) sCPA (kij=-0.08+0.0003*T)

10-

u
D

10-*^
O

^10-^-J

^o-'i
I I I I I I I I I I I I I I I I I I I I I I I I I M I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I

200

250

300

350

400

450

500

.T(K)
Figure 4. Correlation of the water/n-hexane liquid-liquid equilibria with the sCPA EoS using either a single binary parameter or a temperature dependent interaction parameter.[Kontogeorgis et al}^] The methanol/propane system This system is of special interest. It has been shown^^ that CPA correlates very well the phase diagram of methanol/propane and predicts the azeotrope at low methanol concentrations. For the existence of this azeotrope at propane splitters there are no experimental data but there is industrial evidence"^^ [Meijer H., 1999; Lauerman, 1999 - personal communication]. SAFT predicts also the azeotrope at approximately the same concentration"^"^: 0.6% methanol. Predicting this azeotrope is important for the design of propane/propene separation (C3 splitters). Comparison with other models Table 3 presents a comparison between various association models for alcohol/alkanes VLB. The models considered are: original SAFT, Simplified SAFT^^, Anderko's chemical EoS and CPA. All models perform satisfactorily. CPA yields very similar resuhs to the more compHcated models e.g. SAFT.

125
250ooooo Experimental data sCPA (l<,j=-0.0475) 200^ 0.20

S150H O _Q Q_100H

50

0.9975

'6.9980 ' ' 0.9985 ' ' 0.99 ^water

0.9995

1.0000

1.00

Figure 5. Correlation of the water/methane Figure 6. Correlation of the water/methanol vapor-liquid equilibra at 75C with the sCPA vapor-liquid equilibria at 25C with the sCPA EoS using a single binary parameter^^. EoS using a single binary parameter^^. Table 3. Comparison of the CPA EoS with other association models in the obtained average absolute percent error in VLE calculations (bubble-point pressures). SSAFT = Simplified SAFT, AEoS=Anderko equation of state [Yakoumis et al ^\
System Butane/octane Ethanol/butane Temperature 373.51 298.45 323.75 345.65 373.15 413.15 453.15 493.15 313.10 343.10 373.10 CPA 0.43 3.31 2.07 3.34 1.47 2.68 2.52 1.31 4.26 2.17 0.92 SAFT 2.29 1.31 2.84 2.09 3.81 3.67 1.83 5.25 2.64 1.66 SSAFT 1.75 0.35 1.20 2.50 4.47 4.76 2.73 4.92 2.33 1.45 AEoS 0.70 3.2 1.0

methanol/benzene

methanol/propane

It may be expected that a more advanced model such as CPA would be significantly more time-consuming for performing phase equilibrium calculations compared to the simpler

126 models (SRK, NRTL). As shown in table 4, this is not entirely correct. Only for binary systems, CPA is 6 times slower than SRK and NRTL, while for multicomponent systems, such as those of interest to practical applications, CPA is only 3.6 times slower. Such a difference may be allowable considering the improvement achieved in the calculations. Table 4. Speed of various models (in hundreedths of seconds on a 200 MHz Pentium PC for 100 identical flash calculations)^^

Model sCPA with Elliott sCPA without Elliott SRK NRTL

Testl 166 529 28 28

Test 2 3361 4938 830 -

Test 3 3218 5405 900 -

Test 1: equimolar mixture of water and methanol at 1 bar and 350 K Test 2: mixture of 29 hydrocarbons plus methanol at 10 bar and 350 K Test 3: mixture of 29 hydrocarbons plus methanol and water at 10 bar and 350 K LLE for alcohol/alkanes As shown in Figure 7 and in the publication by Voutsas et al?^, CPA provides much improved results for binary LLE for alcohol/hydrocarbons compared to conventional models (SRK, UNIFAC). Furthermore, a comparison with the LLE results reported for these systems in the literature with SAFT^ indicate that CPA and SAFT perform similarly. It is interesting to note that UNIFAC-LLE, although it contains special LLE-based parameters, does not yield very satisfactory results: this demonstrates the necessity of accounting explicitly for the hydrogen bonding effects dominant in these systems, especially when extended temperature ranges are involved (the UNIFAC-LLE parameter table is recommended at temperatures around 20-40 C). In some cases, e.g. for phenol/pentane, CPA yields excellent description of the binodal curve, while UNIFAC-LLE incorrectly predicts complete miscibility. Glycol/alkane liquid-liquid equilibria The extension of the CPA (and SAFT) equations of state to new associating compounds should be done with care: establishing the correct number of sites and the exact association scheme is not a trivial task, neither is selecting the best set of pure parameters among the various sets, which optimize equally well the pure compound vapor pressures and liquid densities. To illustrate the difficulties encountered, the following example can be mentioned: for compounds, which resemble to alcohols e.g. glycols, it might be expected that the 2B or 4C association schemes could be reasonable starting points. This has been indeed verified; the 4C scheme was proved to be the most successful choice, as shown recently by Derawi et al^. The choice of the correct association scheme and parameters is based on fitting vapor pressures and liquid density data. These choices may depend on the choice of data optimized, which, as shown in Figures 8 and 9, can in some cases vary substantially. Figures 8 and 9 show the "experimental data" generated from two DIPPR correlations as well as the actual

127 raw data. Moreover, the DIPPR correlations are often extended at much greater temperature ranges than the actual raw data^. The LLE correlation for two glycol/alkane systems shown in figures 10 and 11 is very good with a single interaction parameter. Such a behavior cannot be obtained with a conventional model such as SRK (figure 2).
40ooooo Experimental data sCPA (ko=0.009)

30^

20o

O
10-

OH

1 0

I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I M

I I I I I I I

0.0

0.2

0.4 ^methanol

0.6

0.8

1.0

Figure 7. LLE for methanol/hexane with the sCPA EoS state^^. Methanol is described with the 2B association scheme.

1.0E+01 1.0E+00 ^ 1.0E-01 -

,<^"'" / / ^^.-^^
wt^t^^^"^^^^^^

-2 -4 -6 - -8 10 .^ 12 14 16

o S 3^ ^
0)

^ :^ 1 .OE-02 ^ 1.0E-031.0E-04 1.0E-05 1 OE-06 250 350 450 T/K 550

"o E "^ "o 3 = o - 18 S

650

750

Figure 8. The difference between the DIPPR correlations for vapor pressure and saturated liquid density for ethylene glycol: solid lines, DIPPR 1989"^^ dashed lines, DIPPR 200145. experimental raw data [Derawi et al}\

128

i O F + 0 9

f^ f^

^^---'''C^---

CO

1.0E+00 1.0E-025 S 1.0E-041.0E-06/ / ^ ^ . ^ - ' ^ ^ ^ ' ' ^ ' '

-1.0 1

- 2.0
-3.0

i
c o

-4.0 "o

1.0E-081.0E-10 1 HF 19

/^
1 1 1 1

- 5.0
7 n

- 6.0 o
r / .U

200

300

400

500 T/K

600

700

800

Figure 9. The difference between the DIPPR correlations for vapor pressure and saturated Uquid density for tetraethylene glycol: solid lines, DIPPR 1989"^"^; dashed lines, DIPPR 2001"^^; X, experimental raw data [Derawi et al}\

1.00E-02

o j ^ 1.00E-03 o

1 .OOE-04 305

310

315

320 T/K

325

330

335

Figure 10. LLE for ethylene glycol (1) + n-hexane (2) system with sCPA with a ky = 0.059: X, xf , experimental; o, x\, experimental; solid line, xf , sCPA; dashed line, x\, sCPA. / = Glycol-rich phase and / / = Hydrocarbon-rich phase [Derawi et al}\

129

1.00E-01

I
o
(0

1.00E-02

1.00E-03

1.00E-04 300

310

320

330 T/K

340

350

360

Figure 11. LLE for triethylene glycol (1) + n-heptane (2) system with sCPA with a ky = 0.094: X, xf, experimental; o, Xj, experimental; solid line, x^', sCPA; dashed line, X 2 sCPA. / = Glycol-rich phase and / / = Hydrocarbon-rich phase. [Derawi et al}].

Multicomponent Equilibria As shown in Figures 12-14, prediction of ternary liquid-liquid equilibria for water/methanol/hydrocarbons is excellent using solely one interaction parameter per binary. The results for the multicomponent systems should be considered as straight predictions, since all three interactions parameters have been obtained from the corresponding binary systems. Even, when partition coefficients of the methanol are considered (a very strict test), the prediction is very good and consistently better than conventional thermodynamic models (SRK, NRTL) available in existing commercial simulators. Figure 14 provides results of the partitioning of methanol in a quaternary system, which contains the main types of compounds of a typical petroleum mixture (water, gases, oil and a hydrate inhibitor e.g, methanol). The accurate prediction of the partitioning of methanol between the two liquid phases for such a mixture is very important for the oil industry. A difference of the order of 10 in the calculated partitioning coefficients may result up to 50% increase in the amount of methanol injected for hydrate inhibition. To realise the economical importance of such a prediction, we mention that each oil platfomi spends on a yearly basis approximately 4 million British pounds for the addition of methanol in the oil [Moorwood, 1998 - personal communication]. Since, as shown in Figure 14, models such as SRK overestimate almost one order of magnitude the amount of alcohol in the organic phase, this would yield to an overestimation of the amount of added methanol of almost 50%. Using CPA more reasonable estimates of the methanol required will be made. It is interesting to note that successful "predictive" results with conventional models (SRK using an NRTLbased combining rule) have been also reported for the system

130 water/methanol/methane/heptane'^^. Future investigations may focus on further comparisons between the SRK (using advanced mixing rules) and CPA models for this type of systems.

0.10 H

' Experimental data at ZO'^C sCPA prediction g- 0.20 A

00000 Experinnental data at 20C sCPA prediction NRTL prediction SRK prediction

I 0.15
E D 0.04

0.10 i

E 0.00 0.2 0.3 0.4 0.5 0.6 0.7

Xmethanoi In oqueous

phase

0.2 Xmethanoi IH

0.4 0.6 Q.l OqUeOUS phOSe

Figure 12. Prediction with sCPA of the partitioning of methanol between the organic and the aqueous phase for the ternary system: water/methanol/propane^^.

Figure 13. Prediction with various models of the partitioning of methanol between the organic and the aqueous phase of the ternary system: water/methanol/n-hexane^^.

3 H s > * E x p e r l m e r r t a l d a t a a t 1 0 C i e i c c c i 0 . 1 5 -A o o o o o E x p e r i m e n t a l d a t a a t 2 0 C X > x X X E x p e r i m e n t a l d a t a a t 50**C < sCRA Rredfction SRK Prediction

0.05

f
^methanol ' '

0.3

0.4

0.5

aqueous

phase

Figure 14. Prediction with sCPA and SRK of the partitioning of methanol between the organic and the aqueous phase of the quaternary system: water/methanol/methane/nheptane^ ^.

6.5 USE OF THE CPA EOS

The CPA EoS is being used by several researchers:

-

At the Technical University of Berlin. CPA and other equations were tested for water / COz I aromatic hydrocarbons, cresol systems47.A software available on the internet also contains the CPA equation (httr,:iivt2~~8.vt2.tu-harbur~.de), well as other association as equations of state (SAFT, Anderlto). At Berkeley, Prausnitz and co-workers" developed a CPA-type EoS combining PengRobinson with the association term of Wertheim. The new feature of this work is the inclusion of an electrolyte term; thus the resulting EoS is suitable for associating mixtures in the presence of electrolytes. CPA is included in Infochem's commercial software (http:l/w\~v.infochemulc.com). From the company's web-site. we read: "The CPA model reproduces the partitioning of methanol between u'ater. and lqdrocarbon vapor and liquid phases more acczmtely than conventional models. Thus it will predict less conservative restdts for the anzount o f vzethanol required for uJixed inhibition, The diferences between the model predictions will be ~noslmarked for systems with low lvater content and/or signiJicant un~ountsof C6+". CPA is included in the software SPECS of the research group IVC-SEP (Institut for Kemiteknik. Technical Universitl- of Denmark, Denmark). The simplified version is incorporated (sCPA). Parameters for water: methanol and glycols are available and new parameters for other associating compounds are being inserted as they become available. I\4oreover, IVC-SEP has developed CPA as a CAPE-OPEN property package ( h t t p : / / c o l a n . a d d u c e . d e / p o r t a l / S o f t ~ l and thus the model can be ) used in commercial simulators. Recent applications of CPA include polymers38, octanol-water partition coefficient ca~culations~"~' (figure 19), mixtures with glycols - water - alkane^^"^ (figures 10.1 1,16,17), systems with chemicals such as sulfolane, gases (nitrogen, CO), water and ~~, hydrocarbons (not published in the open literature). systems with l ~ e t o n e s Henry's law constants for alkane-water systems40,and finally systems containing carboxylic (figure 20).

6.6 CHALLENGES

6.6.1 Systems with Aromatic Hydrocarbons

Olefinic and aromatic hydrocarbons in mixhires with water and other associating compounds have been considered as well. In these cases, the cross-association b e t ~ e e n aromatic (and olefinic. to a lesser degree) hydrocarbons and water or alcohols has been taken into account. Cross-association is important in these systems. since many experimental studies indicate the existence of weak complexes formed between the benzene ring (due to TC-electrons) and proton donor molecule^^^^^^. The increased solubility of water and aromatic compounds as compared to uater and n-alltanes is a result of solvation between water and aromatic

132 compounds. Most results related to these systems have not as yet appeared in the open literature but further reveal the capabilities of the CPA equation of state. A sample result is given in Figure 15, obtained from a publication in a Greek journal.

ooooo benzene in the water rich phase nnnna water in the benzene rich phase sCPA (single interaction parameter)

10

q o
O 10-H

o
10

1 0

* I I I I I I I I I I I I 11 I I I I I I I I I I I I I I I I I I I I H I I I I I I I I I I H I I I I I I 11 I I I I I I I I M I I H I

200

250

300

350

400

450

500

550

^^^^
Figure 15. Correlation of the water-benzene liquid-liquid equilibria with the sCFA using a single binary parameter"^^. The solvation between water and benzene has been accounted for. 6.6.2 Cross-Associating Systems As mentioned previously, water with heavy alcohols and glycols cannot be modelled satisfactorily using a single interaction parameter and the Elliott rule. Improved combining rules are needed for such cross-associating systems. Voutsas et al^^ suggested the following set of combining rules (denoted as CRl in figures 16 and 17) which result to much improved representation of these systems:

(12)

pAey^^pA,.S,.p^B,
A theoretical justification for this new set of combining rules can be provided. The rules of equations 12 are based on the well-loiown theoretically accepted rules for the cross enthalpy and entropy of hydrogen bonding:

133 AH/ + AH J (13) AS, +AS/

AH a-

In order to show the equivalence between the combining rules in equations 12 and 13, we consider the similarities (analogy) between the chemical and the perturbation theory. The association strength (of perturbation theory) is equivalent to the equilibrium constant of chemical theory^^, thus: AocKRT
( exp

^.

f-AH^

(AS

RT ocexp I n RT =>oc-A/-/ (14)

expl

RT

p/)

(AS\

(AS

RT oc exp R [

^ P oc exp R [

In equation 14, we have ignored the density dependence of the association strength (expressed via the radial distribution function) and of the equilibrium constant. Notice also that in the equation for the association volume parameter, the b/RT part is also ignored. If this is included, then the Elliott equation (Eq. 11) is obtained (using equations 12 and 13). Based on equations 13 and 14, the 'new combining rules', equations 12, are derived. In some few cases, e.g., for the system water/tert-butyl alcohol/i-butane, satisfactory ternary LLE results are obtained using the Elliott rule for the cross association strength of water/alcohol. Despite the fact that the correlation of water/alcohol is only fair (4% deviation), the performance of the multicomponent calculations is not greatly affected. Still, in most other cases where the Elliott rule fails for the cross-associating systems, improved combining rules need to be used. Equations 12 may be successful for some systems e.g. water with heavy alcohols or glycols. Recent results for water/heavy glycols demonstrate that equations 12 provide a better representation of cross-association effects than the Elliott rule (equation 11). Some results are shown in figures 16 and 17 (CR-1 combining rule is equation 12). We note that investigations on cross-associating systems with association EoS such as SAFT are extremely scarce. Essentially, to our knowledge, only few systematic works have been presented'^^'^^'^^ Experimental investigations and prediction methods as well as ab initio (molecular orbital) calculations for the cross-enthalpy and entropy of hydrogen bonding may be very helpful in the future in defining appropriate combining rules for the association parameters^ ^"^^'^\ Of

134 particular challenge, in certain future developments, may be the account of hydrogen-bond co-operativity effects in alcohol-containing systems^"^ as well as special associating schemes e.g. the intramolecular association in systems containing oxy-alcohols, which include both an ether and a hydroxyl group^^.

Figure 16. sCPA-correlation and prediction of the MEG-water VLE at T = 363.15 K using the CR-1 combining rule with a kjj = -0.012 (solid line) and % = 0 (dashed line).[Derawi et al?^. Reprinted by permission]
9 n ^j ^ ^ z.u t 1.5^ <>
_^ TS
* ^ \ ^

* ^

1 10 '
Q.

^ ^ ^

**^ ^ \ r N ^

0.5-

0.0 H 0.0

I
i

^^""^C^
*
.0
'

0.2

0.4

0.6

' 0.8

3^!irt.

>

1.0

DEG mole fraction

Figure 17. sCPA-correlation and prediction of the DEG-water VLE at T = 393.15 K using the CR-1 combining rule with a ky = -0.115 (solid line) and ky = 0 (dashed line).[Derawi et al.^"^. Reprinted by permission]

135 6.6.3 Critical Area of Mixtures CPA parameters for associating fluids are obtained from vapor pressure and liquid density data. The critical point constraints are not used, contrary to the classical cubic equations of state. For gas/alcohol or gas/water systems, the critical temperature and pressure are used for the gas. It may be expected that problems are encountered in the description of the critical area of associating mixtures, especially those containing gases. This is indeed, to some extent, the case and typically SAFT and CPA overestimate the critical point, less though for CPA than for SAFT^\ In some cases, e.g. for methanol/methane (figure 18) and water/methane (figure 5) systems, satisfactory results are obtained. Describing satisfactorily the critical area for associating systems with equations of state such as SAFT and CPA remains an open area for investigation.
500' ooooo Experimental data sCPA (k]j=0.0134) 400-

S300H O

CL200H

100-

0 11111111111111111111111111 n 111111111 n 111111111111 n 1111 rri 11111111111 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00 ^methanol

Figure 18. Correlation of the methanol/methane vapor-liquid equilibria at 25C with the sCPA EoS using a single binary parameter^^. 6.6.4 New associating compounds We have already discussed the extension of sCPA to various associating compounds such as water, alcohols and glycols. But what about 'special' chemicals such as sulfolane, which has the following structure (CH2)2-S02-(CH2)2? A simple inspection of the structure of this compound may not be adequate for establishing whether it is or not a self-associating substance let alone to give some hints regarding the correct association scheme. Experimental activity coefficient data of sulfolane with n-alkanes have been reported^^ and the values at infinite dilution of the order of 40-50 indicate a behavior similar to methanol/alkanes, at least

136 from this point of view. A similar conclusion is reached by the experimental LLE data for sulfolane/n-alkanes. Although such experimental data will not be used directly in the pure parameter estimation, they may help us identifying the correct number and nature of the association scheme of this or similar compounds. Another family of compounds, which present special interest, is the ketones. It is not clear whether ketones should be treated as associating compounds or not. Jog et al.^^ state that the highly non-ideal behavior of e.g. ketone-alkane systems is due to the polarity of ketones. They have developed a version of SAFT, named polar-SAFT, and model ketones-alkanes without association (with much better results than SAFT). On the other hand, Yakoumis et al?^ treated ketones as two-site associating molecules with CPA and presented very satisfactory results for the prediction of the liquid-liquid partition coefficient of methanol in the methanol/acetone/n-hexane system (using solely binary parameters). For environmentally related chemicals, Polyzou et al?^ (figure 19) have demonstrated that the sCPA EoS can be applied to the prediction of octanol/water partition coefficients.

experimental

log(Kow)

Figure 19. Predicted against experimental octanol-water partition coefficients with CPA"39

Finally, the sCPA EoS has been recently extended to systems with carboxylic acids and a typical result is provided in figure 20. Excellent correlation is achieved at both temperatures, including the azeotrope, with a single interaction parameter. The lA (Huang and Radosz notation^"^) association scheme has been used for acetic acid, in accordance to the dominant physical picture (formation of dimers).

137

0.35 0.3 0.25 -

(5

0.2 MB) O (BfpQQQ,

Z 0.15^
0.1 -

0.05-

0J

()

0.2

'

0.4
xi,yi

0.6

0.8

Figure 20. Pressure-composition diagram for Acetic acid(l) + n-octane(2) system with sCPA at T= 323.2 K (lower curve) and T=343.2 K (upper curve) with a kj = 0.06 : o, experimental data; solid line, CPA employing the lA site model. From Zeuthen'^\ 6.7 CONCLUDING REMARKS CPA is a powerful tool for predicting multicomponent multiphase equilibria for 'water alcohol - hydrocarbons' and for other mixtures containing hydrogen bonding substances. The improved performance of CPA (and SAFT) should be attrributed to the Wertheim association term. Considering the similarity of the various association theories^'^\ equally good performance could be expected for the chemical and lattice theories. This is the case provided that the crucial 'details' discussed for the pertubartion theory {e.g. selection of association scheme, cross association) are carefully taken into account. The time requirement of CPA EoS does not seem to be a restrictive factor for its commercial application. Every extension of CPA (and SAFT) to new associating compounds/mixtures should be done with great care, taking into account the factors discussed in detail in this chapter. The most important of them are summarized here: careful estimation of pure parameters and selection of the appropriate association scheme the pure parameter values for the association energy and volume should be compared, when possible, with experimental (or ab initio) values for the heats and entropies of hydrogen bonding phase equilibria data for a specific associating compound with an inert non-associating alkane is a useful way in selecting the appropriate set of pure parameters as well as the most suitable association scheme the nature of cross-association should be established. It is important to choose appropriate mixing and combining rules for the association parameters as well as to investigate "hidden" (not apparent) cross-associations such as when aromatic (and to a lesser degree olefmic hydrocarbons) are present in mixtures with alcohols and/or water

138 for satisfactory multicomponent predictions, crucial binaries should be determined and described as accurately as possible tricks for faster/easier calculations are important e.g. the Elliott rule for the cross association strength, the use of homomorph method^^ for reducing the number of pure association parameters, the use of estimation methods for obtaining pure vapour pressures and liquid densities when experimental data are not available, etc.

Future applications of the CPA equation of state depend on specific interest from academia and industry and can involve both further theoretical and applied studies-extensions to new types of compounds. Current activities include systems containing carboxylic acids, ternary LLE for water-based systems with nitrogen-containing compounds e.g. pyridine, acrylonitrile and ethyleneamine, mixtures with aqueous alkanolamines and gases such as CO2, gas-containing systems which also include aromatic hydrocarbons and associating compounds e.g. water and alcohols or glycols, solid-liquid equilibria for glycol-waterhydrocarbons, as well as extension of CPA to mixtures with strong electrolytes.

LIST OF SYMBOLS energy parameter of SRK and CPA EoS, defined in equation 7 parameter in the energy term, equation 7 association site A in molecule i association site B in molecule j co-volume parameter of the SRK and CPA EoS parameter in the energy term (Soave expression), equation 7 density (in figure 3) radial distribution function, defined in equations 8 or 9 binary interaction parameter, defined in equation 10 equilibrium constant octanol-water partition coefficient mole fraction of molecule i not bonded at site A pressure ideal gas constant temperature mole fraction in the liquid phase mole fraction in the vapor phase reduced co-volume, defined in Eqs. 8 or 9 volume compressibility factor

ao
Ai

Bj b
Cl

K P R T
X

y y

z
Greek Symbols

association volume parameter association strength

139 AH AS
AiBj

enthalpy of hydrogen bonding entropy of hydrogen bonding association energy parameter molar density

Superscripts A B assoc phys sat Subscripts compound i compound] cross-parameter (interaction) between compounds i and j reduced property van der Waals association site A association site B association term of the EoS physical term of the EoS saturated

ij

r
w

LIST OF ABBREVIATIONS AEoS CPA CR-1 DEG EoS LLE MC MEG NRTL RDF SAFT SSAFT sCPA SRK TEG vdW vdWlf VLE Anderko Equation of State Cubic Plus Association (EoS) combining rule - 1 (defined in Eq. 12) diethylene glycol Equation of State Liquid-Liquid Equilibria Mathias-Copeman (mono)ethylene glycol Non Random Two Liquid (model) Radial Distribution Function Statistical Association Fluid Theory/EoS Simplified SAFT simplified CPA Soave RedUch Kwong EoS triethylene glycol van der Waals (EoS) vdW one fluid mixing rules Vapor-Liquid Equilibria

140 REFERENCES 1. I. Hatzioannidis, E,C. Voutsas, E, Lois and D.P. Tassios, J. Chem. Eng. Data, 43 (1998)386. 2. U. von Stockar and L.A.M. van der Widen, J. Biotechnology, 59 (1997) 25. 3. J.M. Prausnitz, Fluid Phase Equilibria, 53 (1989) 439. 4. S. Bruin, Fluid Phase Equilibria, 158-160 (1999) 657. 5. S.Zeck, Fluid Phase Equilibria, 70 (1991) 125. 6. M.-L. Yu and Y.-P. Chen, Fluid Phase Equilibria, 94 (1994) 149. 7. S.J. Suresh and J.R. Elliott, Ind. Eng. Chem. Res., 31 (1992) 2783. 8. S.O. Derawi, M.L. Michelsen, G.M. Kontogeorgis and E.H. Stenby, Fluid Phase Equilibria, 209 (2003) 163. 9. I.G. Economou and M.D.Donohue, AIChE J., 37(12) (1991) 1875. 10. M.D. Donohue and I.E. Economou, Fluid Phase Equilibria, 116(1-2) (1996) 518. 11. E.M. Hendriks, J.M. Walsh and A.R.D. van Bergen, J, Stat. Phys., 87(5/6) (1997) 1287. 12. M.L. Michelsen and E.M. Hendriks, Fluid Phase Equilibria, 180 (2001) 165. 13. W.G. Chapman, K.E.Gubbins, G. Jackson and M. Radosz, Ind. Eng, Chem. Res., 29 (1990) 1709. 14. S.H. Huang and M. Radosz, Ind. Eng. Chem. Res., 29 (1990) 2284. 15.1.V. Yakoumis, G.M. Kontogeorgis, E.C.Voutsas, E.M. Hendriks and D.P. Tassios, Ind. Eng. Chem.Res., 37 (1998) 4175. 16. G.M. Kontogeorgis, I.V. Yakoumis, H. Meijer, E.M. Hendriks and T. Moorwood, Fluid Phase Equilibria, 158-160 (1999) 201. 17. J.R. Elliott, S.J. Suresh and M.D. Donohue, Ind. Eng. Chem. Res., 29 (1990) 1476. 18. G.M. Kontogeorgis, E.C. Voutsas, I.V. Yakoumis and D.P. Tassios, Ind. Eng. Chem. Res., 35(1996)4310. 19. I.V. Yakoumis, G.M. Kontogeorgis, E.C. Voutsas and D.P. Tassios, Fluid Phase Equilibria, 130(1997)31. 20. A. Anderko, Fluid Phase Equilibria, 45 (1989) 39. 21. C.A. Koh, H. Tanaka, J.M. Walsh, K.E. Gubbins and J.A. Zollweg, Fluid Phase Equilibria, 83(1983)51. 22. A. Nath and E. Bender, Fluid Phase Equilibria, 7 (1981) 275. 23. G. C. Pimentel and A.L. Mclellan, The Hydrogen Bond, Freeman: San Francisco, 1960. 24. A. Ksiazczak and A. Moorthi, Fluid Phase Equilibria, 23 (1985) 153. 25. P. Vimalchand, G.D. Ikonomou and M.D. Donohue, Fluid Phase Equilibria, 43 (1988) 121. 26. I.G. Economou and M.D. Donohue, Ind. Eng. Chem. Res., 31(1992) 2388. 27. A, Anderko, Fluid Phase Equilibria, 50(1989) 21. 28. A. Anderko, Fluid Phase Equilibria, 74(1992) 89. 29. I.G. Economou and C. Tsonopoulos, Chem. Eng. Sci., 52(4)(1997) 511. 30. Y.-H Fu and S.I. Sandler, Ind. Eng. Chem. Res., 34(1995) 1897. 31. E.C. Voutsas, G.C. Boulougouris, I.G. Economou and D.P. Tassios, Ind. Eng. Chem. Res., 39 (2000) 797.

141 32. E.G. Voutsas, I.V. Yakoumis, G.M. Kontogeorgis and D.P. Tassios, Fluid Phase Equilibria, 132(1997)61. 33. E.G. Voutsas, I.V. Yakoumis, and D.P. Tassios, Fluid Phase Equilibria, 158-160 (1999) 151. 34. S.O. Derawi, G.M. Kontogeorgis, M.L. Michelsen and E.H. Stenby, Ind. Eng. Ghem. Res., 42(7) (2003) 1470. 35. I.V. Yakoumis, E.N. Polyzou, H. Meijer, T. Moorwood, E.M. Hendriks and G.M. Kontogeorgis, Industrial applications of the sGPA equation of state. 2^^ Greek Ghemical Engineering Gonference, University of Thessaloniki, 16-29/5/1999, Book of Abstracts, 337-340 (in Greek) 36. I.V. Yakoumis, E.G. Voutsas and D.P. Tassios, An equation of state for systems exhibiting hydrogen bonding. 1^^ Greek Ghemical Engineering Gonference, University of Patras, 2937. -31/5/1997, Book of Abstracts, 319-324 (in Greek). 38. G.M. Kontogeorgis and I.V. Yakoumis, Novel model for phase equilibrium in the chemical and petroleum industry. Monthly Technical Review, March 1999, 39-43 (in Greek). 39. G.M. Kontogeorgis, I.V. Yakoumis and P.M. Vlamos, Gomputational & Theoretical Polymer. Science, 10(6) (2000) 501. 40. E.N. Polyzou, A.E. Louloudi, G.M. Kontogeorgis and I.V. Yakoumis, Prediction of octanol-water partition coefficients. 2"^ Greek Ghemical Engineering Gonference, University of Thessaloniki, 16-29/5/1999, Book of Abstracts, 101-104 (in Greek). 41. G.G. Boulougouris, E.G. Voutsas, I.G. Economou and D.P. Tassios, J. Phys. Ghem. B., 105(32) (2001) 7792. 42. J. Zeuthen, Extension of the GPA EoS to systems containing organic acids, M.Sc. Thesis, Institut for Kemiteknik, Technical University of Denmark, 2003. 43. G. M. Kontogeorgis, S. Derawi, J. Zeuthen, E.H. Stenby, N. v. Solms, Th. Lindvig, LA. Kouskoumvekaki and M. L. Michelsen, 2003. Applications of association models to problems of the oil, chemical and polymer industries, Invited (plenary) lecture to the 20th European Symposium on Applied Thermodynamics, 9-12 Ocrober 2003, Germany (article available in the proceedings of the conference). 44. P.K. Jog, A. Garcia-Guellar and W.G. Ghapman, Fluid Phase Equilibria, 158-160 (1999)321. 45. T.E. Daubert, R.P. Danner, Physical and Thermophysical Properties of Pure Gompounds: Data Gompilation, Hemisphere, New York, 1989. 46. T.E. Daubert, R.P. Danner, Physical and Thermophysical Properties of Pure Gompounds: Data Gompilation, Hemisphere, New York, 2001. 47. K.S. Pedersen, M.L. Michelsen and A.O. Fredheim, Fluid Phase Equilibria, 126 (1996) 13. 48. O. Pfohl, A. Pagel and G. Brunner, Fluid Phase Equilibria, 157 (1999) 53. 49. J. Wu and J.M. Prausnitz, Ind. Eng. Ghem. Res., 37 (1998) 1634. 50. J. Suresh and E.J. Beckman, Fluid Phase Equilibria, 99 (1994) 219. 51. S. Michel, H.H. Hooper and J.M. Prausnitz, Fluid Phase Equilibria, 45 (1989) 173. 52. J.P. Wolbach and S.I. Sandler, Ind. Eng. Ghem. Res., 37 (1998) 2917.

142 53.1.G. Economou, G.D. Ikonomou, P. Vimalchand and M.D. Donohue, AIChE J., 36(12) (1990) 1851. 54. J.P. Wolbach and S.I. Sandler, Ind. Eng. Chem. Res., 36(1997) 4041. 55. R.B. Gupta and R.L. Brinkley, AIChE J., 44(1) (1998) 207. 56. R.L. Brinkley and R.B. Gupta, Ind. Eng. Chem. Res., 37(1998) 4823. 57. W.C. Moollan, U.M. Domanska and T.M. Letcher, Fluid Phase Equilibria, 128 (1997) 137. 58. P.K. Jog, S.G. Sauer, J. Blaesing and W.G. Chapman, Ind. Eng. Chem. Res., 40(21) (2001)4641.