Asymmetric Framework for Predicting Liquid-Liquid Equilibrium of Ionic Liquid-Mixed Solvent Systems: II.

Prediction of Ternary Systems

Luke D. Simoni, Alexandre Chapeaux, Joan F. Brennecke and Mark A. Stadtherr* Department of Chemical and Biomolecular Engineering University of Notre Dame, Notre Dame, IN 46556, USA

March, 2009

*Author to whom all correspondence should be addressed. Phone (574) 631-9318; Fax (574) 631-8366; Email: markst@nd.edu

Abstract
A new asymmetric framework for modeling liquid-liquid equilibrium (LLE) in electrolyte/mixed-solvent systems is demonstrated, with focus on systems involving a dilute aqueous solution of an ionic liquid (IL). The extent to which this approach is able to predict ternary LLE, using parameters obtained from binary and pure component data only, is evaluated. For this purpose, ternary IL/solvent/water systems are used as examples. Comparisons of predicted LLE are made to experimental data representing various types of ternary LLE behavior, as well as to predictions obtained from standard symmetric models. Results indicate that an asymmetric NRTL/eNRTL model provides better predictions of ternary LLE for systems containing water than standard symmetric models.

Keywords: Liquid-liquid equilibrium; Ionic liquids; Electrolyte models; Extraction; Aqueous systems

1. Introduction
In Part I of this two-part contribution, we introduced an asymmetric framework for modeling liquid-liquid equilibrium (LLE) in electrolyte/mixed-solvent systems. This approach allows for the use of different excess Gibbs energy models, representing different degrees of electrolyte dissociation, in different phases. As a first approximation, we focused on the case in which the electrolyte (IL) is either completely dissociated or completely paired (molecular), with the state of the IL depending on the dielectric constant of the mixed solvent and on the concentration of IL in the phase in question. For these assumptions, the applications of interest are systems involving an aqueous phase that is dilute in IL. In this second part of the contribution, we demonstrate the use of this asymmetric approach in modeling ternary LLE, with focus on systems of the form IL/solvent/water. Of particular interest is the extent to which this approach is able to predict ternary LLE, using parameters obtained from binary and pure component data only. To determine this, we will make comparisons of predicted LLE to experimental data representing various types of ternary LLE behavior, as well as to predictions obtained from standard symmetric models. Before presenting these examples, we first discuss the methods used for estimating model parameters and for performing the ternary phase equilibrium calculations.

2. Model Parameters
In this section we outline the general approach used to determine model parameters. Additional problem-specific details, including sources of data, will be provided in Section 4. The models for which parameters are needed are the conventional (symmetric) NRTL, electrolyte-NRTL (eNRTL) and UNIQUAC models and the new asymmetric NRTL/eNRTL model. 2.1 Fully Adjustable Parameters For each of the models considered, we use only two fully adjustable parameters per binary. For components i and j, these are the binary interaction parameters ij and ji , as defined in Appendix B (Part I) for NRTL and eNRTL and in Appendix A here for UNIQUAC. For the NRTL and eNRTL models, the nonrandomness parameter ij = ji is not treated as fully

adjustable, as discussed in Section 2.2. The binary interaction parameters ij and ji will be determined using experimental binary data only. If data at the exact system temperature are not available, then, unless otherwise noted, the parameters are determined using data at nearby temperatures, and the parameter values at the system temperature are obtained by assuming a linear temperature dependence. 2.1.1 Partially immiscible binaries For binaries exhibiting a miscibility gap, the model parameters are determined from LLE data (mutual solubility) using the equal activity conditions. For IL/water binaries, the asymmetric NRTL/eNRTL model parameters are determined using the method described in detail in Part I, and the symmetric NRTL, eNRTL, and UNIQUAC parameters are determined using the method given previously by Simoni et al. 1,2 For IL/solvent and solvent/water binaries, only the symmetric-model parameters are needed (since there are no phases with dissociated IL), and these are obtained using published values, if available, or by using the method of Simoni et al.1 2.1.2 Miscible binaries It is possible that the IL/solvent and/or solvent/water binaries may be completely miscible. For these cases, only the symmetric-model parameters are needed, and these are determined from either VLE or excess enthalpy (hE) data. If published parameters obtained from VLE data are available then these are used. Otherwise, except for the solvent/water binary in Section 4.1, we use isothermal VLE data and obtain the binary model parameters by globally minimizing the relative least squares objective function

i ( Piexp Picalc )

Pi exp . If no VLE

data is available, which is the case only for the IL/solvent binary in Section 4.2, then we obtain the binary model parameters from excess enthalpy data by globally minimizing the relative least square function

i ( hiE,exp hiE,calc )

hiE,exp . To perform the global optimization we follow Gau

et al.3 and use an interval-Newton method. The approach used here differs in that a somewhat more efficient interval-Newton algorithm4 is used. By using this approach, we obtain a rigorous guarantee that the relative least squares objective function used has been globally minimized.

2.2 Other Parameter Values In this section, we discuss the strategies used for setting other model parameters. For the NRTL and eNRTL models, used symmetrically or asymmetrically, we treat the nonrandomness parameter, ij = ji , as fixed at a value that depends on the type of binary data used. If previously published values have been used for the binary interaction parameters, then the accompanying value of ij is used. Otherwise, we use the conventional values 5,6 of ij = 0.2 when LLE data are used to obtain the binary interaction parameters, and ij = 0.3 when VLE data are used. For the one case of a miscible binary in which hE data are used, we use the arbitrary high value of ij = 0.8 . In our experience, when fitting to hE data, particularly if endothermic, a relatively large value of ij is needed to ensure that the resulting parameters lead to a prediction of complete miscibility (convex Gibbs energy curve). It is well known 6 that complete miscibility can be forced by using a sufficiently large value of ij . For the closest ionic approach parameter in the eNRTL model, we consider values of 5, 8.94, 14.9 and 25, the last three of which also have been used elsewhere. 1,2,7,8 The effect of increasing is to reduce the importance of the long-range electrostatic contribution in the eNRTL model. As observed previously, 1,2 when a relatively small value of is used, it may not be possible to find suitable binary parameter values that fit given binary LLE data. Thus, it may be necessary to adjust to a higher value until suitable binary parameters can be found. For symmetric use of the eNRTL model, we use = 14.9, as suitable binary parameters can be found using this value for all of the example systems considered here. Although a physical relationship exists9,10 between and = 1, the ionic center-to-center distance used in the asymmetric framework (see Part I, eq 10), this is only roughly approximated here. The default values used here in the asymmetric NRTL/eNRTL model for and are = 5 and = 51010 m. If suitable binary parameter values cannot be found using these and values, then we systematically increase and , according to the sequence given in Table 1, until suitable binary parameter values can be found for all binaries in the ternary system of interest. The specific values of and

used in each example are stated in Section 4.

UNIQUAC requires for each pure component i, a size (relative volume) parameter ri and shape (relative area) parameter qi. For the IL species, these parameters were calculated using the Bondi method using IL segment values determined by Nebig et al. 11 For the solvent and co-solvent species, these parameters were taken from standard sources. 12-14 Traditionally, the reference species used in determining the relative area parameter qi has been the van der Waals CH2 group.14 However, as explained by Abreu et al., 15 this choice may make it impossible to find suitable binary parameter values for modeling binary LLE with UNIQUAC, especially when there are components of greatly different sizes and shapes. To alleviate this problem, Abreu et al.15 suggest using water as an alternative reference species. Following suit, we allow water as an alternative reference species, but use it only when use of CH2 does not permit finding suitable binary parameter values. calculations. A coordination number of Z = 10 is used for all UNIQUAC

3. Computation of Phase Equilibrium

For phase equilibrium at constant temperature T and pressure P, the total Gibbs energy G must be at a global minimum with respect to the number of phases present and their amounts and composition. To find this global minimum, we use a standard approach 16,17 in which an If necessary, the phase split equilibrium phase split calculation is used to determine a local minimum of G and then this point is tested for global optimality using phase stability analysis. calculation is then repeated, perhaps changing the number of phases assumed to be present, until a solution is found that meets the global optimality test. The key to this two-stage global optimization procedure for phase equilibrium is the correct solution of the phase stability problem, itself a global minimization problem. To ensure correct solution of the phase stability problem, we extend the approach of Tessier et al.18 This method is based on tangent plane analysis16,19,20 and uses a rigorous global minimization technique based on an interval-Newton approach, thus providing a mathematical and computational guarantee 21,22 of global optimality. Tessier et al.18 considered only the case of a symmetric model with no dissociation. For the computations reported here, we have extended this approach to the new case of an asymmetric model with dissociation, using the objective functions derived in Part I. The approach used here also differs from that of Tessier et al. 18 in that a somewhat more efficient interval-Newton algorithm4 is used.

4. Examples
In this section, we give four examples that demonstrate the use of the asymmetric framework for modeling LLE in electrolyte/mixed-solvent systems, in particular ternary IL (1)/alcohol (2)/water (3) systems. For each example, we discuss how the model parameters were obtained from binary data, and then we use these parameters to predict the ternary LLE. For the asymmetric framework we use NRTL for a molecular phase and eNRTL for a dissociated phase. Predictions from the asymmetric NRTL/eNRTL model are compared to experimental data, and to predictions based on the conventional (symmetric) NRTL, eNRTL, and UNIQUAC models. The example systems are listed in Table 2, which also indicates the type of ternary LLE observed experimentally for each system, and provides a key to the Figures in which the predictions (ternary diagrams) for each system can be found. All examples are at conditions of atmospheric pressure. The cut-off values (explained in Part I) used in the asymmetric modeling framework are c = 40 and xc = 0.10. The binary interaction parameters used in each example are compiled in Tables 3-5 for the symmetric NRTL, UNIQUAC and eNRTL models, respectively, and in Table 6 for the asymmetric NRTL/eNRTL model. Pure component properties used are tabulated in Table 7, which gives data for properties needed in the electrostatic term of the eNRTL model, namely molecular weight (Mi), density (di) and dielectric constant (i), and in Table 8, which gives data for the UNIQUAC size (ri) and shape (qi) parameters. 4.1 Example 1: [hmim][Tf2N]/Ethanol/Water at 295 K This example involves the ternary system of 1-hexyl-3-methylimidazolium

bis(trifluoromethylsulfonyl)imide ([hmim][Tf2N]) (component 1), ethanol (EtOH) (component 2), and water (component 3) at T = 295 K. This system exhibits Type 1 ternary LLE behavior experimentally.23 A Type 1 system exhibits a single phase envelope emanating from a single binary miscibility gap, which in this case corresponds to the [hmim][Tf 2N]/water binary. LLE data for [hmim][Tf2N]/water are given by Chapeaux et al.24 and these data were used to estimate the model parameters for this binary. The other two binaries, [hmim][Tf 2N]/EtOH and Unfortunately, VLE data for EtOH/water, are totally miscible at the system temperature.

IL/EtOH was not available at the temperature of the system. Therefore, VLE data measured by

Kato and Gmehling25 at 353.16 K were used to estimate the model parameters for this binary. For the EtOH/water binary, we used NRTL parameters from Kurihara et al., 26 including 23 = 0.1448. The UNIQUAC parameters for the EtOH/water binary were estimated using data at 295.75 K from Gmehling et al.27 by fitting directly to the equifugacity conditions for VLE. For the asymmetric model predictions, the closest ionic approach parameter values used in this example are = 5 and = 1109 m. In order to calculate the IL reference energy, the asymmetric model requires a value for the ILs static dielectric constant (see Part I). Dielectric constants have been measured28 for a series of 1-alkyl-3-methylimidazolium bis(trifluoromethylsulfonyl)imide ILs. Extrapolating these values with respect to alkyl chain length yields a dielectric constant estimate for [hmim][Tf 2N] of 1 = 11.4. For the symmetric use of UNIQUAC, the reference species used for the pure component shape parameters is the CH 2 group. For the symmetric NRTL, UNIQUAC, and eNRTL models, the predicted ternary behavior is shown in Figure 1. It can be seen that UNIQUAC provides a remarkably good prediction of the two-phase envelope and plait point, though the tie line slopes are somewhat steeper than those measured experimentally. NRTL provides a somewhat better slope for the tie lines than UNIQUAC, but with greater overestimation of the two-phase envelope. Overall, UNIQUAC is the best of the symmetric models in predicting the ternary LLE for this system. Figure 2 shows the prediction from the asymmetric NRTL/eNRTL model, with comparison to the UNIQUAC prediction, the best of the symmetric models for this system. A close qualitative inspection of Figure 2 indicates that the asymmetric NRTL/eNRTL prediction is somewhat better than the UNIQUAC prediction, in terms of both the two-phase envelope size and the slope of the tie lines. To obtain a more quantitative comparison, for each experimental tie line, we took the midpoint and used it as the feed composition to calculate phase equilibrium (endpoints of a calculated tie line). 1 N
TL

These calculated values were then compared to the

2 3

experimental endpoints of the tie line, and the average absolute deviation AAD =

x
k =1 j =1 i =1

( j) i , k ,exp

xi(,kj ),calc

(2)

( j) was determined. Here xi ,k ,calc is the calculated mole fraction of component i in phase j based on ( j) the midpoint of tie line k, and xi ,k ,exp is the corresponding experimental value. TL = 12 is the

number of tie lines used for the comparison, and N = (2)(3)(TL) is the number of terms in the summation. For the UNIQUAC prediction, AAD = 0.0311, and for the asymmetric NRTL/eNRTL prediction, AAD = 0.0202. Two experimental tie lines very close to the plait point were not used in this comparison, since UNIQUAC slightly overshoots the plait point, and asymmetric NRTL/eNRTL slightly undershoots it. For this example, the UNIQUAC model provides a remarkably good prediction of the ternary phase behavior. The new asymmetric framework, using NRTL and eNRTL, provides a remarkably good prediction as well, one that is even somewhat better than the UNIQUAC prediction. 4.2 Example 2: [HOhmim][Tf2N]/1-Butanol/Water at 295 K For this example, we consider the ternary system of 1-hydroxyhexyl-3-

methylimidazolium bis(trifluoromethylsulfonyl)imide ([HOhmim][Tf 2N]) (component 1), 1butanol (BuOH) (component 2), and water (component 3) at T = 295 K. This system exhibits a Type 2 ternary LLE system experimentally.29 A Type 2 system exhibits a phase envelope that spans the composition space and connects two binary miscibility gaps. In this example, the immiscible binaries are [HOhmim][Tf2N]/water and BuOH/water. LLE data for [HOhmim] [Tf2N]/water are given by Chapeaux et al. 30 and these data were used to estimate the interaction parameters for this binary. For the BuOH/water binary, LLE data from Srensen and Arlt 5 were used for the parameter estimation. The third binary, [HOhmim][Tf 2N]/BuOH is completely miscible at the system temperature. Since no VLE data is available for this binary, we used hE data,31 which is available for 40 C, to estimate the model parameters. The NRTL parameters found for [HOhmim][Tf2N]/BuOH (with 12 = 0.8) yield a miscible binary system. However, for UNIQUAC and eNRTL, the model parameters found for this binary yield only a partially miscible system. Thus, for symmetric model predictions we focus on NRTL only. For the asymmetric model predictions, the closest ionic approach parameter values used in this example are = 8.94 and = 5109 m. No experimental value of the dielectric constant for [HOhmim] [Tf2N] is available. We took this value to be 1 = 15. This is based on the estimated dielectric constant (11.4) of [hmim][Tf2N], from the previous example, and assuming that addition of a hydroxyl group may somewhat increase the dielectric constant. The symmetric NRTL prediction is shown in Figure 3. This predicts a three-phase region, with an additional two-phase envelope, terminating in a plait point, that emanates from

the IL/BuOH side of the three-phase region. This prediction is qualitatively incorrect, as neither of these features is observed experimentally. Figure 4 shows the asymmetric NRTL/eNRTL prediction for this system. This prediction of Type 2 ternary behavior is in qualitative agreement with the experimental measurements, and there is good quantitative accuracy as well. Note that for the IL/BuOH and BuOH/water binaries the model parameters are the same in the symmetric NRTL model as in the asymmetric NRTL/eNRTL model. The parameters in the IL/water binary are different in the asymmetric NRTL/eNRTL model in order to account for the dissociation of IL in the dilute aqueous phase. By using an asymmetric model that allows for dissociation of IL in a dilute aqueous phase, but not in other phases, we have been able to obtain a significantly improved prediction of the ternary behavior in this example system. 4.3 Example 3: [bmim][PF6]/Ethanol/Water at 290, 298 and 313 K The example focuses on the ternary system of 1-butyl-3-methylimidazolium hexafluorophosphate ([bmim][PF6]) (component 1), ethanol (component 2) and water (component 3) at T = 290 K, T = 298 K, and T = 313 K. Experimentally, this system exhibits Type 2a ternary LLE diagrams.32 In Type 2a ternary diagrams, there are two distinct phase envelopes corresponding to two binary miscibility gaps, and complete miscibility at intermediate compositions. As noted in a previous study,2 these types of ternary LLE systems are difficult to predict with gE models fit to binary data. It may be difficult to obtain even a proper qualitative prediction (i.e., Type 2a rather than Type 2). LLE data for [bmim][PF 6]/water are available from Najdanovic-Visak et al.33 and Anthony et al.,34 and these data were used to estimate the interaction parameters for this binary. For the [bmim][PF 6]/EtOH binary, experimental data from Najdanovic-Visak et al.32 were used to estimate the binary parameters. For neither of these binaries was it possible to obtain UNIQUAC parameters to represent the miscibility gaps. This is a well-known issue15 with UNIQUAC, which might be addressed by considering an even larger reference species for the pure component shape parameters (we considered only CH2 and water). Thus, symmetric predictions were made using only NRTL and eNRTL. For the miscible EtOH/water binary, we used NRTL parameters from Kurihara et al., 26 including 23 = 0.1448. For the asymmetric model predictions, the closest ionic approach parameter values used

in this example are = 8.94 and = 5109 m. The dielectric constant for [bmim][PF6] was measured by Weingrtner35 as 1 = 11.4. We have considered the application of the symmetric NRTL and eNRTL models to this system in a previous study.2 Figure 5 shows these symmetric-model predictions for the 298 K case, together with the experimental data. Note that experimental tie lines are not available, just cloud point measurements to obtain an approximate binodal curve. Also note that there are small discrepancies between the experimental and calculated binary miscibility gaps. This is due solely to an inconsistency between the experimental binary data used to perform the predictions and the experimental ternary data used for comparison. Both NRTL and eNRTL, applied symmetrically, predict Type 2 LLE behavior, not Type 2a, as observed experimentally. Figure 6 shows the asymmetric NRTL/eNRTL prediction at 298 K. Here we see a dramatic qualitative improvement in the prediction. We now see Type 2a behavior predicted on the aqueous side of the diagram, with this two-phase envelope terminating in a plait point. However, on the alcohol side of the diagram, the two-phase envelope is still of Type 2 character, as it does not terminate in a plait point. Instead this two-phase envelope will terminate when the mixed solvent dielectric constant in the IL-lean phase ( x1 < xc = 0.10 ) reaches the c = 40 cutoff for a molecular phase model. This type of discontinuity must be recognized as a possibility when an asymmetric model for LLE is used. It is interesting to note that if the cut-off on the mixed solvent dielectric constant is relaxed, so that the IL is considered dissociated in dilute solution ( x1 < xc ) with any mixture of water and ethanol, then the prediction of the asymmetric NRTL/eNRTL model becomes completely Type 2a (i.e., both two-phase envelopes terminate in a plait point). For the 290 K case, both the symmetric NRTL and eNRTL models again predict Type 2 LLE behavior,2 not Type 2a, as indicated by cloud point measurements. Figure 7 shows the cloud point data, along with the prediction from the asymmetric NRTL/eNRTL model at 290 K. The situation is very similar to that observed for the 298 K case. The asymmetric-model prediction captures the two-phase envelope on the aqueous side of the diagram, but not on the alcohol side. Finally for the 313 K case, which experimentally is Type 2a, the symmetric NRTL model again predicts2 Type 2 behavior, but the symmetric eNRTL model correctly predicts 2 Type 2a. Figure 8 shows that the asymmetric NRTL/eNRTL model also correctly predicts Type

2a. The symmetric eNRTL prediction significantly overestimates 2 the size of the two-phase envelope on the aqueous side, and the asymmetric-model prediction significantly overestimates the size of the two-phase envelope on the alcohol side. 4.4 Example 4: [bmim][Tf2N]/1-Butanol/Water at 288 K This example involves the ternary system of 1-butyl-3-methylimidazolium

bis(trifluoromethylsulfonyl)imide ([bmim][Tf2N]) (component 1), 1-butanol (component 2), and water (component 3) at T = 288 K. For this system, all three binary subsystems have a miscibility gap. There are multiple types of possible ternary behavior for the case of three binary miscibility gaps,5 of which Types 3a and 3b are of interest here. A Type 3b system has three distinct two-phase envelopes, each emanating from one of the binary miscibility gaps and terminating in a plait point. In a Type 3a system, two of these three envelopes merge into one, i.e., there is one two-phase envelope that connects two of the binary miscibility gaps and another two-phase envelope that emanates from the remaining binary miscibility gap and terminates in a plait point. Najdanovic-Visak et al. 36 have presented limited cloud point data for this system, interpolated to 288 K. This is consistent with either Type 3a or 3b behavior, though NajdanovicVisak et al.36 proposed the approximate Type 3b diagram shown for reference in Figure 9. However, recent experiments by Morton and Davis37 imply, at least qualitatively (we do not believe that their measurements are quantitatively correct, as discussed in Appendix B), that there is a much larger two-phase region extending roughly between the BuOH/water and [bmim] [Tf2N]/water miscibility gaps, even at the higher temperature of 298 K. This suggests that this system is actually Type 3a, with one two-phase region connecting the BuOH/water and [bmim] [Tf2N]/water miscibility gaps, and another emanating from the [bmim][Tf 2N]/butanol miscibility gap and going to a plait point. Davis and Morton 37 also state that they observed a three-phase equilibrium state for this system, but we believe that this is likely to be spurious, as discussed in Appendix B. For the [bmim][Tf2N]/water binary, model parameters were determined from the LLE data of Crosthwaite et al.,38 for the [bmim][Tf 2N]/BuOH binary from the data of NajdanovicVisak et al.,36 and for the BuOH/water binary from data given by Srensen and Arlt. 5 For the asymmetric model predictions, the closest ionic approach parameter values used in this example are = 5 and = 1109 m. The dielectric constant of [bmim][Tf 2N] was measured by Daguenet

10

et al.28 as 1 = 11.5. For the symmetric use of UNIQUAC the reference species for the pure component shape parameters is water. As can be seen from Figures 9 and 10, the symmetric-model predictions 2 using NRTL and UNIQUAC, respectively, yield qualitatively incorrect Type 3 ternary systems. Type 3 systems are characterized by the presence of three two-phase envelopes that emanate from their respective binary miscibility gaps and then converge in a 3-phase region. The size of three-phase region predicted by UNIQUAC, and hence the overall immiscibility of the system, is somewhat smaller than that predicted by NRTL. The symmetric eNRTL prediction 2 is shown in Figure 11 and has the Type 3a behavior that we believe to be qualitatively correct, as explained above. This is a significant improvement over the NRTL and UNIQUAC predictions, but still deviates considerably from the experimental cloud point observation at x2 0.23 (BuOH). Figure 12 compares the asymmetric NRTL/eNRTL model prediction to the symmetric eNRTL model prediction and the experimental cloud points. The asymmetric NRTL/eNRTL model also predicts a Type 3a system. Although there is not a dramatic change compared to the symmetric eNRTL prediction, the asymmetric NRTL/eNRTL prediction does better capture the cloud point observation at x2 0.23 (BuOH), improving the shape of the binodal curve for the two-phase region involving a dilute aqueous phase. The improved predictions provided by the asymmetric NRTL/eNRTL model and the symmetric eNRTL model on this problem, and on the previous one, indicate the importance of accounting for electrostatic forces in problems involving dilute aqueous solutions of ILs.

5. Concluding Remarks
In the examples above, we have used the asymmetric framework presented in Part I of this contribution to predict ternary LLE, using parameters obtained from binary and pure component data only. Comparisons of the predicted LLE were made to experimental data representing various types of ternary LLE behavior, as well as to predictions obtained from standard symmetric models. This is a stringent test of the suitability of various models for describing LLE in systems containing ILs. We have found that, for systems of the type considered here, namely systems involving an aqueous phase dilute in IL, an asymmetric NRTL/eNRTL model provides better predictions than standard symmetric models (NRTL,

11

UNIQUAC, eNRTL).

Examples 3 and 4, involving Type 2a and 3a ternary behavior,

respectively, suggest the importance of including an electrostatic contribution in the Gibbs energy for systems involving dilute aqueous solutions of ILs. Doing this asymmetrically, so that IL dissociation is allowed only in the aqueous phase, provides improvements over the symmetric electrolyte model (eNRTL) in all examples. In future work, we will consider partial dissociation and application to nonaqueous systems.

Acknowledgements
This work was supported in part by the Department of Energy under Grant DE-FG3608GO88020. Additional support was provided by a University of Notre Dame Arthur J. Schmitt Presidential Fellowship (LDS).

12

Literature Cited
(1) Simoni, L. D.; Lin, Y.; Brennecke, J. F.; Stadtherr, M. A. Reliable computation of binary parameters in activity coefficient models for liquid-liquid equibrium. Fluid Phase Equilib. 2007, 255, 138-146. (2) Simoni, L. D.; Lin, Y.; Brennecke, J. F.; Stadtherr, M. A. Modeling liquid-liquid equilibrium of ionic liquid systems with NRTL, electrolyte-NRTL, and UNIQUAC. Ind. Eng. Chem. Res. 2008, 47, 256-272. (3) Gau, C.-Y.; Brennecke, J. F.; Stadtherr, M. A. Reliable nonlinear parameter estimation in VLE modeling. Fluid Phase Equilib. 2000, 168, 1-18. (4) Lin, Y.; Stadtherr, M. A. LP Strategy for Interval-Newton Method in Deterministic Global Optimization. Ind. Eng. Chem. Res. 2004, 43, 3741-3749. (5) Srensen, J. M.; Arlt, W. Liquid-Liquid Equilibrium Data Collection; DECHEMA: Frankfurt/Main, Germany, 1979-1980. (6) Renon, H.; Prausnitz, J. M. Local compositions in thermodynamic excess functions for liquid mixtures. AIChE J. 1968, 14, 135-144. (7) Pitzer, K. S. Electrolytes. From dilute solutions to fused salts. J. Amer. Chem. Soc. 1980, 102, 2902-2906. (8) Chen, C.-C.; Britt, H. I.; Boston, J. F.; Evans, L. B. Local composition model for excess Gibbs energy of electrolyte solutions, part I: Single solvent, single completely dissociated electrolyte systems. AIChE J. 1982, 28, 588-596. (9) Pitzer, K. S. Electrolyte theory-improvements since Debye and Hckel. Accounts of Chemical Research 1977, 10, 371-377. (10) Pitzer, K. S.; Li, Y. -. Thermodynamics of aqueous sodium chloride to 823 K and 1 kilobar (100 MPa). Proc. Natl. Acad. Sci. USA 1983, 80, 7689-7693. (11) Nebig, S.; Blts, R.; Gmehling, J. Measurement of vapor-liquid equilbria (VLE) and excess enthalpies (HE) of binary systems with 1-alkyl-3-methylimidazolium

13

bis(trifluoromethylsulfonyl)imide and prediction of these properties and using modified UNIFAC (Dortmund). Fluid Phase Equilib. 2007, 258, 168-178. (12) Smith, J. M.; Van Ness, H. C.; Abbott, M. M. Introduction to Chemical Engineering Thermodynamics; McGraw-Hill: New York, NY, 2001. (13) Anderson, T. F.; Prausnitz, J. M. Application of the UNIQUAC equation to calculation of multicomponent phase equilibria. 1. Vapor-liquid equilibria. Ind. Eng. Chem. Process Des. Dev. 1978, 17, 552-561. (14) Abrams, D. S.; Prausnitz, J. M. Statistical thermodynamics of liquid mixtures: A new expression for the excess Gibbs energy of partly or completely miscible systems. AIChE J. 1975, 21, 116-128. (15) Abreu, C. R. A.; Castier, M.; Tavares, F. W. A phase stability analysis of the combinatorial term of the UNIQUAC model. Chem. Eng. Sci. 1999, 54, 893-896. (16) Michelsen, M. L. The isothermal flash problem. Part I: Stability. Fluid Phase Equilib. 1982, 9, 1-19. (17) Michelsen, M. L. The isothermal flash problem. Part II: Phase-split calculation. Fluid Phase Equilibria 1982, 9, 21-40. (18) Tessier, S. R.; Brennecke, J. F.; Stadtherr, M. A. Reliable phase stability analysis for excess Gibbs energy models. Chem. Eng. Sci. 2000, 55, 1785-1796. (19) Baker, L. E.; Pierce, A. C.; Luks, K. D. Gibbs energy analysis of phase equilibria. Soc. Petrol. Engrs. J. 1982, 22, 731-742. (20) Prausnitz, J. M.; Lichtenthaler, R. N.; Gomes de Azevedo, E. Molecular Thermodynamics of Fluid-Phase Equilibria; Prentice Hall: Upper-Saddle River, New Jersey, 1999. (21) Neumaier, A. Interval Methods for Systems of Equations; Cambridge University Press: Cambridge, UK, 1990.

14

(22) Kearfott, R. B. Rigorous Global Search: Continuous Problems; Kluwer Academic Publishers: Dordrecht, The Netherlands, 1996. (23) Chapeaux, A.; Simoni, L. D.; Ronan, T.; Stadtherr, M. A.; Brennecke, J. F. Extraction of alcohols from water with 1-hexyl-3-methylimidazolium bis(trifluoromethylsulfonyl)imide. Green Chem. 2008, 10, 1301-1306. (24) Chapeaux, A.; Simoni, L. D.; Stadtherr, M. A.; Brennecke, J. F. Liquid phase behavior of ionic liquids with water and 1-octanol and modeling of 1-octanol/water partition coefficients. J. Chem. Eng. Data 2007, 52, 2462-2467. (25) Kato, R.; Gmehling, J. Systems with ionic liquids: Measurement of VLE and gamma(infinity) data and prediction of their thermodynamic behavior using original UNIFAC, mod. UNIFAC(Do) and COSMO-RS(O1). J. Chem. Thermodyn. 2005, 37, 603-619. (26) Kurihara, K.; Nakamichi, M.; Kojima, K. Isobaric Vapor-Liquid Equilibria for Methanol + Ethanol + Water and the Three Constituent Binary Systems. J. Chem. Eng. Data 1993, 38, 446-449. (27) Gmehling, J.; Onken, U.; Arlt, W. Vapor-liquid Equilibrium Data Collection, Chemistry Data Series, Vol. I, Parts 1-8; DECHEMA: Frankfurt/Main, Germany, 1977-1990. (28) Daguenet, C.; Dyson, P. J.; Krossing, I.; Oleinikova, A.; Slattery, J.; Wakai, C.; Weingartner, H. Dielectric response of imidazolium-based room-temperature ionic liquids. J. Phys. Chem. B 2006, 110, 12682-12688. (29) Chapeaux, A.; Ronan, T.; Simoni, L. D.; Brennecke, J. F.; Stadtherr, M. A. Separation of water and alcohols using 1-(6-hydroxyhexyl)-3-methylimidazolium bis(trifluoromethylsulfonyl)imide and 1-hexyl-3-methylimidazolium tris(pentafluoroethyl)trifluorophosphate. 2009, Manuscript in preparation. (30) Chapeaux, A.; Andrews, N.; Brennecke, J. F. Liquid-Liquid behavior of ILs in water and 1-hexanol. J. Chem. Eng. Data 2009, Manuscript in preparation. (31) Chapeaux, A.; Brennecke, J. F. Excess enthalpies and enthalpies of vaporization for various ionic liquid-alcohol mixtures. 2009, Manuscript in preparation. 15

(32) Najdanovic-Visak, V.; Esperana, J. M. S. S.; Rebelo, L. P. N.; da Ponte, M. N.; Guedes, H. J. R.; Seddon, K. R.; de Sousa, H. C.; Szydlowski, J. Pressure, Isotope, and Water Co-solvent Effects in Liquid-Liquid Equilibria of (Ionic Liquid + Alcohol) Systems. J. Phys. Chem. B 2003, 107, 12797-12807. (33) Najdanovic-Visak, V.; Esperana, J. M. S. S.; Rebelo, L. P. N.; da Ponte, M. N.; Guedes, H. J. R.; Seddon, K. R.; Szydlowski, J. Phase behaviour of room temperature ionic liquid solutions: an unusually large co-solvent effect in (water + ethanol). Phys. Chem. Chem. Phys. 2002, 4, 1701-1703. (34) Anthony, J. L.; Maginn, E. J.; Brennecke, J. F. Solution thermodynamics of imidazolium-based ionic liquids and water. J. Phys. Chem. B 2001, 105, 10942-10949. (35) Weingrtner, H. The static dielectric constant of ionic liquids. Z. Phys. Chem. 2006, 220, 1395-1405. (36) Najdanovic-Visak, V.; Rebelo, L. P. N.; da Ponte, M. N. Liquid-liquid behaviour of ionic liquid-1-butanol-water and high pressure CO2-induced phase changes. Green Chem. 2005, 7, 443-450. (37) Davis, S. E.; Morton III, S. A. Investigation of ionic liquids for the separation of butanol and water. Sep. Sci. Technol. 2008, 43, 2460-2472. (38) Crosthwaite, J. M.; Aki, S. N. V. K.; Maginn, E. J.; Brennecke, J. F. Liquid Phase Behavior of Imidazolium-Based Ionic Liquids with Alcohols. J. Phys. Chem. B 2004, 108, 51135119. (39) Smyth, C. P.; Stoops, W. N. The Dielectric Polarization of Liquids. VI. Ethyl Iodide, Ethanol, Normal-Butanol and Normal-Octanol. J. Am. Chem. Soc. 1929, 51, 3312-3329. (40) Fernndez, D. P.; Goodwin, A. R. H.; Lemmon, E. W.; Levelt Sengers, J. M. H.; Williams, R. C. A formulation for the static permittivity of water and steam at temperatures from 238 K to 873 K at pressures up to 1200 MPa, including derivatives and Debye-Hckel coefficients. J. Phys. Chem. Ref. Data 1997, 26, 1125-1166.

16

Appendix A
For the UNIQUAC model,14 g E gcomb g res , = + RT RT RT n n g comb = xi ln i + 5 qi xi ln i , RT xi i i =1 i =1
n n g res = qi xi ln j ji , j =1 RT i =1

(A1) (A2) (A3)

i =

ri xi

rj x j
j =1 n

(A4)

i =
and

qi xi

qj xj
j =1

(A5)

uij ij = exp (A6) . RT Here, ri and qi represent a relative volume (size) and surface area (shape), respectively, for component i. It follows that i and i are volume and surface area fractions, respectively, for component i. The binary interaction parameters ij = uij are estimated from experimental binary data.

17

Appendix B
Davis and Morton37 have recently reported experimental measurements (two-phase tie lines) of LLE in the ternary systems [bmim][Tf2N]/1-butanol/water (Example 4) and [hmim] [Tf2N]/1-butanol/water at 298 K. Based on our own measurements 23 of LLE in the latter system, we do not believe that the results of Davis and Morton 37 are quantitatively correct for either system. Figure B1 shows the results of our measurements23 of tie lines in the two-phase region for [hmim][Tf2N]/1-butanol/water at 295 K. Each tie line was based on three replicates of the feed composition. The estimated uncertainty in the equilibrium compositions was 3 mol%. The weight fraction of each component was measured independently, and these summed to one within 5 mol%. Figure B1 also shows the measurements of Davis and Morton 37 for this same system at 298 K. There is clearly a large quantitative discrepancy between the two sets of data. Davis and Morton37 did not use replicate feed compositions and did not estimate the uncertainty in their equilibrium composition measurements. They measured the weight fractions of only two of the components, [hmim][Tf 2N] and 1-butanol, and calculated the water composition to make the weight fractions sum to one. Because they did not measure the water composition independently, there was no check on likely errors in the measurements of the other two components. Potential sources for these errors will be discussed elsewhere. Because their measurements for the [hmim][Tf2N]/1-butanol/water system are quantitatively incorrect, we do not believe that their measurements for the [bmim][Tf 2N]/1-butanol/water system are quantitatively correct either. For both of the ternary systems [bmim][Tf2N]/1-butanol/water and [hmim][Tf2N]/1butanol/water at 298 K, Davis and Morton 37 have reported observing a three-phase equilibrium state while exploring the two-phase region. They do not give the exact feed composition at which this was seen, nor the compositions of the three phases. In our study 23 of the [hmim] [Tf2N]/1-butanol/water system at 295 K, we did not observe any three-phase equilibrium states. The three-phase observation of Davis and Morton37 is likely due to the occurrence of a phase density inversion that occurs as the fraction of 1-butanol increases. At low 1-butanol fraction, the aqueous phase is the less dense of the two equilibrium phases. As the fraction of 1-butanol is increased in the two-phase region, eventually the aqueous phase becomes the denser of the two equilibrium phases. Thus, at some point in the two-phase region, the density of the two phases is 18

the same. When at or near this point, the density driving force for phase separation may not overcome surface tension forces, and a situation may arise in which part of the denser phase is trapped above the less dense phase, thus creating the appearance of a three-phase state. We believe that, in both of the systems that they studied, Davis and Morton 37 likely observed such a spurious three-phase state.

19

Table 1: Priority list for pairs of closest approach parameters and to use in the asymmetric NRTL/eNRTL model. See Section 2.2 for discussion.

5 5 8.94 8.94 14.9 14.9 25 25

(m) 51010 1109 1109 5109 5109 1108 1108 5108

20

Table 2: Summary of ternary system examples.

Example No. 1 2 3 4

Figure No. 1, 2 3, 4 5, 6, 7, 8 9, 10, 11, 12

Ternary Type 1 2 2a 3a/3b

System (1)/(2)/(3) [hmim][Tf2N]/Ethanol/Water [HOhmim][Tf2N]/1-Butanol/Water [bmim][PF6]/Ethanol/Water [bmim][Tf2N]/1-Butanol/Water

Temperature (K) 295 295 290, 298, 313 288

21

Table 3: NRTL binary interaction parameters (J/mol) estimated from binary data.

Example No. 1 2 3 4

Figure No. 1 3 5 9

Temp. (K) 295 295 298 288 g12 2222.5 1948.7 4309.1 3411.0 g21 4969.0 1914.6 16476 13019 g13 702.08 -5045.1 685.01 428.14 g31 21820 21276 16672 19105 g23 2568.6 2566.9 2556.2 2555. 4 g32 6949.2 12505 6974.5 12217

22

Table 4: UNIQUAC binary interaction parameters (J/mol) estimated from binary data.

Example No. 1 2 4

Figure No. 1, 2 NA 10

Temp. (K) 295 295 288 u12 2646.6 1313.9 67.986 u21 743.47 113.41 1431.9 u13 4812.5 3006.3 5811.7 u31 151.97 200.48 1368.1 u23 245.79 782.66 1959.4 u32 1870.2 1471.3 2333.4

23

Table 5: eNRTL binary interaction parameters (J/mol) estimated from binary data.

Example No. 1 2 3 4

Figure No. 1 NA 5 11, 12

Temp. (K) 295 295 298 288 g12 8259.0 2169.0 7219.1 1152.2 g21 14230 2005.3 16431 800.05 g13 2707.1 2684.6 3992.3 3636.3 g31 21287 21072 17535 20083 g23 2568.6 2566.9 2556.2 2555.4 g32 6949.2 12505 6974.5 12217

24

Table 6: NRTL/eNRTL binary interaction parameters (J/mol) estimated from binary data.

Example No. 1 2 3 3 3 4

Figure No. 2 4 6 7 8 12

Temp. (K) 295 295 298 290 313 288 g12 2222.5 1948.7 4309.1 3879.9 5060.2 3411.0 g21 4969.0 1914.6 16476 16887 15747 13019 g13 167.62 -4440.3 223.68 110.02 1263.3 133.55 g31 11805 16197 10530 10188 12015 8862.8 g23 2568.6 2566.9 2556.2 2588.7 2493.4 2555.4 g32 6949.2 12505 6974.5 6908.2 7102.4 12217

25

Table 7: Solvent properties for eNRTL model.

Temp. Solvent ethanol ethanol ethanol ethanol 1-butanol water water water water water (K) 290 295 298 313 288 288 298 295 290 313 M 46.07 46.07 46.07 46.07 74.12 18.02 18.02 18.02 18.02 18.02 d(g/cm3) 0.78 0.78 0.78 0.76 0.81 1.00 1.00 1.00 1.00 0.99 24.4 23.7 23.2 21.0 17.8 82.1 78.4 79.6 81.4 73.4 Reference Smyth and Stoops39 Smyth and Stoops39 Smyth and Stoops39 Smyth and Stoops39 Smyth and Stoops39 Fernndez et al.40 Fernndez et al.40 Fernndez et al.40 Fernndez et al.40 Fernndez et al.40

23

Table 8: Pure component size and shape parameters for UNIQUAC model.

Example No. 1 2 4

Figure No 1 3 8

Temp. (K) 295 295 288

UNIQUAC q reference CH2 CH2 Water q1 11.6 2 14.3 6 7.29 r1 12.5 1 13.9 2 11.2 0 q2 1.97 3.67 2.62 r2 2.11 3.92 3.92 q3 1.40 1.40 1.00 r3 0.92 0.92 0.92

24

List of Figures
Figure 1: Ternary diagram (mol fraction) for [hmim][Tf 2N]/ethanol/water at 295 K (Example 1). Type 1 experimental data from Chapeaux et al.23 All these symmetric-model predictions correctly predict a Type 1 system. Figure 2: Ternary diagram (mol fraction) for [hmim][Tf2N]/ethanol/water at 295 K (Example 1). Type 1 experimental data from Chapeaux et al. 23 and Type 1 predictions from the asymmetric NRTL/eNRTL model and the symmetric UNIQUAC model (best of symmetric models). Figure 3: Ternary diagram (mol fraction) for [HOhmim][Tf2N]/1-butanol/water at 295 K

(Example 2). Type 2 experimental data from Chapeaux et al. 29 Symmetric NRTL model incorrectly predicts a 3-phase region. Figure 4: Ternary diagram (mol fraction) for [HOhmim][Tf2N]/1-butanol/water at 295 K

(Example 2). Type 2 experimental data from Chapeaux et al. 29 Asymmetric NRTL/eNRTL model correctly predicts Type 2 behavior. Figure 5: Ternary diagram (mol fraction) for [bmim][PF6]/ethanol/water at 298 K (Example 3). Type 2a experimental cloud point data from Najdanovic-Visak et al. 32 Symmetric NRTL and eNRTL incorrectly predict Type 2 behavior. Figure 6: Ternary diagram (mol fraction) for [bmim][PF6]/ethanol/water at 298 K (Example 3). Type 2a experimental cloud point data from Najdanovic-Visak et al. 32 NRTL/eNRTL model predicts Type 2a behavior on aqueous side of diagram. Figure 7: Ternary diagram (mol fraction) for [bmim][PF6]/ethanol/water at 290 K (Example 3). Type 2a experimental cloud point data from Najdanovic-Visak et al. 32 NRTL/eNRTL model predicts Type 2a behavior on aqueous side of diagram. Asymmetric Asymmetric

25

Figure 8: Ternary diagram (mol fraction) for [bmim][PF6]/ethanol/water at 313 K (Example 3). Type 2a experimental cloud point data from Najdanovic-Visak et al. 32 NRTL/eNRTL model predicts Type 2a behavior. Figure 9: Ternary diagram (mol fraction) for [bmim][Tf2N]/1-butanol/water at 288 K (Example 4). Experimental cloud point data from Najdanovic-Visak et al. 36 is consistent with either Type 3a or Type 3b behavior, though Najdanovic-Visak et al. 36 proposed the Type 3b phase envelopes shown. However, recent experiments by Morton and Davis37 imply, at least qualitatively, that this system is actually Type 3a (i.e., a two-phase region extends from the IL/water binary miscibility gap to the 1-butanol/water miscibility gap). Symmetric NRTL model incorrectly predicts Type 3 behavior. Figure 10: (Example 4). Ternary diagram (mol fraction) for [bmim][Tf 2N]/1-butanol/water at 288 K Experimental cloud point data from Najdanovic-Visak et al. 36 Symmetric Asymmetric

UNIQUAC model incorrectly predicts Type 3 behavior. Figure 11: Ternary diagram (mol fraction) for [bmim][Tf 2N]/1-butanol/water at 288 K

(Example 4). Experimental cloud point data from Najdanovic-Visak et al. 36 Symmetric eNRTL model predicts Type 3a behavior. See text for discussion. Figure 12: Ternary diagram (mol fraction) for [bmim][Tf 2N]/1-butanol/water at 288 K

(Example 4). Experimental cloud point data from Najdanovic-Visak et al.36 The asymmetric NRTL/eNRTL model predicts Type 3a behavior. See text for discussion. Figure B1: Ternary diagram (mol fraction) for [hmim][Tf 2N]/1-butanol/water, comparing

experimental data sets of Chapeaux et al.23 at 295 K and Davis and Morton37 at 298 K.

26

Ethanol 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 IL 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

0.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Water Experimental Tie Lines NRTL Tie Lines UNIQUAC Tie Lines eNRTL Tie Lies

Figure 1: Ternary diagram (mol fraction) for [hmim][Tf 2N]/ethanol/water at 295 K (Example 1). Type 1 experimental data from Chapeaux et al.23 All these symmetric-model predictions correctly predict a Type 1 system.

27

Ethanol 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 IL 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

0.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Water Experimental Tie Lines UNIQUAC Tie Lines NRTL/eNRTL Tie Lines

Figure 2: Ternary diagram (mol fraction) for [hmim][Tf2N]/ethanol/water at 295 K (Example 1). Type 1 experimental data from Chapeaux et al. 23 and Type 1 predictions from the asymmetric NRTL/eNRTL model and the symmetric UNIQUAC model (best of symmetric models).

28

1-Butanol 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 IL 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1
3-phases

0.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Water Experimental Tie Lines NRTL Tie Lines

Figure 3:

Ternary diagram (mol fraction) for [HOhmim][Tf2N]/1-butanol/water at 295 K

(Example 2). Type 2 experimental data from Chapeaux et al. 29 Symmetric NRTL model incorrectly predicts a 3-phase region.

29

1-Butanol 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 IL 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

0.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Water Experimental Tie Lines NRTL/eNRTL Tie Lines

Figure 4:

Ternary diagram (mol fraction) for [HOhmim][Tf2N]/1-butanol/water at 295 K

(Example 2). Type 2 experimental data from Chapeaux et al. 29 Asymmetric NRTL/eNRTL model correctly predicts Type 2 behavior.

30

IL 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Water 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

0.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Ethanol Experimental Binodal, 298 K NRTL Binodal, 298 K NRTL Tie Lines, 298 K eNRTL Binodal, 298 K eNRTL Tie Lines, 298 K

Figure 5: Ternary diagram (mol fraction) for [bmim][PF6]/ethanol/water at 298 K (Example 3). Type 2a experimental cloud point data from Najdanovic-Visak et al. 32 Symmetric NRTL and eNRTL incorrectly predict Type 2 behavior.

31

IL 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

0.0 Water 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Ethanol Experimental Binodal, 298 K NRTL/eNRTL Tie Lines, 298 K

Figure 6: Ternary diagram (mol fraction) for [bmim][PF6]/ethanol/water at 298 K (Example 3). Type 2a experimental cloud point data from Najdanovic-Visak et al. 32 NRTL/eNRTL model predicts Type 2a behavior on aqueous side of diagram. Asymmetric

32

IL 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Water 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

0.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Ethanol Experimental Binodal, 290 K AF: NRTL/eNRTL Tie Lines, 290 K

Figure 7: Ternary diagram (mol fraction) for [bmim][PF6]/ethanol/water at 290 K (Example 3). Type 2a experimental cloud point data from Najdanovic-Visak et al. 32 NRTL/eNRTL model predicts Type 2a behavior on aqueous side of diagram. Asymmetric

33

IL 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Water 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

0.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Ethanol Experimental Binodal, 313 K NRTL/eNRTL Tie Lines, 313 K

Figure 8: Ternary diagram (mol fraction) for [bmim][PF6]/ethanol/water at 313 K (Example 3). Type 2a experimental cloud point data from Najdanovic-Visak et al. 32 NRTL/eNRTL model predicts Type 2a behavior. Asymmetric

34

1-Butanol 0.0 0.1 0.2 0.3 0.4

3-phases

1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

0.5 0.6 0.7 0.8

0.9 1.0 Water

0.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 IL Experimental Binodal A Proposed Binodal NRTL Binodal NRTL Tie Lines

Figure 9: Ternary diagram (mol fraction) for [bmim][Tf2N]/1-butanol/water at 288 K (Example 4). Experimental cloud point data from Najdanovic-Visak et al. 36 is consistent with either Type 3a or Type 3b behavior, though Najdanovic-Visak et al. 36 proposed the Type 3b phase envelopes shown. However, recent experiments by Morton and Davis37 imply, at least qualitatively, that this system is actually Type 3a (i.e., a two-phase region extends from the IL/water binary miscibility gap to the 1-butanol/water miscibility gap). Symmetric NRTL model incorrectly predicts Type 3 behavior.

35

1-Butanol 0.0 0.1 0.2 0.3 0.4

3-phases

1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

0.5 0.6 0.7 0.8

0.9 1.0 Water

0.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 IL Experimental Binodal UNIQUAC Binodal UNIQUAC Tie Lines

Figure 10: (Example 4).

Ternary diagram (mol fraction) for [bmim][Tf 2N]/1-butanol/water at 288 K Experimental cloud point data from Najdanovic-Visak et al. 36 Symmetric

UNIQUAC model incorrectly predicts Type 3 behavior.

36

1-Butanol 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Water 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

0.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 IL Experimental Binodal eNRTL Binodal eNRTL Tie Lines

Figure 11:

Ternary diagram (mol fraction) for [bmim][Tf 2N]/1-butanol/water at 288 K

(Example 4). Experimental cloud point data from Najdanovic-Visak et al. 36 Symmetric eNRTL model predicts Type 3a behavior. See text for discussion.

37

1-Butanol 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Water 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

0.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 IL Experimental Binodal eNRTL Binodal eNRTL Tie Lines NRTL/eNRTL Binodal NRTL/eNRTL Tie Lines

Figure 12: Ternary diagram (mol fraction) for [bmim][Tf2N]/1-butanol/water at 288 K (Example 4). Experimental cloud point data from Najdanovic-Visak et al.36 The asymmetric NRTL/eNRTL model predicts Type 3a behavior. See text for discussion.

38

1-Butanol 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Water 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

0.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 IL Davis and Morton (2008) at 298 K Chapeaux et al. (2008) at 295 K

Figure B1:

Ternary diagram (mol fraction) for [hmim][Tf 2N]/1-butanol/water, comparing

experimental data sets of Chapeaux et al.23 at 295 K and Davis and Morton37 at 298 K.

39