Three-Phase Equilibria Using Equations of

State
It is demonstrated that a single equation of state may be used to
describe all three phases in liquid-liquid-vapor equilibrium situations.
Wilson’s version of the Redlich-Kwong equation is shown to predict accu- ROBERT A. HEIDEMANN
rately the water solubility in normal paraffins with interaction parameters
k12 = 0.50. A simple procedure for three-phase flash computations is pre- Deportment of Chemical Engineering
University of Calgary
sented. Results obtained using this procedure for the system methane,
Calgary, Alberta T2N 1 N4, Canada
n-butane, water exhibit many of the characteristics of the experimental
data of McKetta and Katz (1948). In particular, the water content of the
vapor phase and the liquid hydrocarbon phase are accurately predicted.

SCOPE
Equations of state, such as the Benedict-Webb-Rubin et al., 1966) have been reported to show liquid miscibil-
equation and the Redlich-Kwong equation, have been ity gaps.
increasingly used to describe both phases in vapor-liquid Although this paper is limited to water, normal paraffin
equilibrium computations. In this paper it is shown that examples and only Wilson’s version of the Redlich-Kwong
it is possible to perform three-phase liquid-liquid-vapor equation is considered, the general approach taken is
computations using the same equation of state to describe applicable to other equations of state and to other kinds
all three phases. of mixtures. The ability of an equation of state to predict
Three-phase equilibria occur in the production and liquid immiscibility and quantitatively correct solubility
processing of natural gas and petroleum wherever signifi- data provides yet another criterion for evaluation.
cant amounts of water are present. At cryogenic tempera- The computational problems in three-phase equilibria
tures and high pressures, systems such as methane, n- are not trivial. A new algorithm is proposed here for three-
heptane (Kohn, 1961) and nitrogen, propane (Schindler phase flash calculations.

CONC1US IONS AN D SIGNI FICANCE

It has been demonstrated that the liquid phase de- Using the new computational scheme, it was possible
scribed by Wilson’s version of the Redlich-Kwong equa- to show that Wilson’s version of the Redlich-Kwong equa-
tion can be unstable and can separate into two liquid tion predicts phase behavior in three-phase systems which
phases of different compositions at equilibrium, depend- exhibit many of the characteristics of experimental sys-
ing on the values of the binary interaction parameters. A tems. In a detailed study of the system methane, n-butane,
value of kI2 = 0.50 for normal paraffin, water pairs is water it was found that the water content of both the
found to predict the water solubility in the hydrocarbon liquid hydrocarbon and the vapor phase were predicted
liquid with reasonable accuracy. These results open the with satisfactory precision for design purposes. Liquid-
possibility of using a single equation of state to describe liquid equilibria computed at pressures above the three-
all three phases in liquid-liquid-vapor equilibrium com- phase critical pressure also agree with the available ex-
putations. perimental data.
A new computation scheme for multiphase flash prob- The general approach used in this paper of describing
lems was presented. This scheme, which is based on free all three phases by the same equation of state in three-
energy minimization, is reliable and easily programmed phase equilibrium situations can be applied to equations
and may be extended to equilibrium computations with other than Wilson’s version of the Redlich-Kwong equa-
more than three phases present. tion.

WILSON‘S REDLICH-KWONG EQUATION where

The version of the Redlich-Kwong equation which is b = t xi bi (2)
used in this paper was proposed by Wilson (1969). It in- and
cludes some modifications of equations proposed earlier by a = st xi xj a
5 (3)
Wilson ( 1964), ( 1966). The pure component coefficients bi are found in the usual
The equation is
way
RT RT a bi = 0.0865 RT,i/Pd (4)
p=-- (1)
V-b V(V+b) Coefficients uij are temperature dependent in a way which

AIChE Journal Wol. 20. N o . 5 ) September. 1974 Page 847

is intended, in part, to match pure component vapor pres-
sure data. The equation is 2.0t-

[ 1+ (1.45 + 1.6205) ( -
Tcj
T
- l)] kij

+ 2.467 bj (-) T T d ktj 0.12

-2.0-

[ 1+ (1.45 + 1.620J (-TdT - l)] kij (5)

Coefficient kj is the so-called ‘Tnteraction parameter” and 0 0.2 0.4 0.6 0.8 1.0
must be obtained by fitting binary data. For pure compo-
nents, MOLE FRACTION WATER IN n-BUTANE
kit = 1 (6) Fig. 1. Standard free energy of mixing for water, n-butane a t 289°K
for various interaction parameters.
Wilson ( 1969) has reported comparisons between ex-
perimental vapor-liquid equilibria data and equilibria com-
puted using the above equations to describe both the vapor
and the liquid phase. In many cases the match of experi- TABLE1. EFFECTOF THE INTERACTIONPARAMETER ON

mental data is very good. CALCULATED MUTUALSOLUBILITIES OF LIQUIDn-BUTANE

AND WATER AT 289°K
Equation ( 1 ) is taken to apply to both the liquid and
the vapor phase. At a given temperature and pressure,
Equation ( 1 ) is satisfied by as many as three real positive Interaction Mole fraction Mole fraction water
parameter n-butane in water in n-butane
values for the molar volume V. I t is conventional to take
the smallest and the largest ositive values to represent
the volume of potential l i q u i i and vapor phases, respec- 0.92 0.00768 0.4603
0.90 0.001738 0.2911
tively. Then the fugacity of component i in the phase hav- 0.80 1.42 x 10-6 0.04224
ing volume V is given by 0.70 1.175 x 10-9 0.007717
lnft = In [XiRT/(V - b ) ] + (2 - l)b& 0.60
0.50
1.108 x 10-11
1.256 x 10-15
1.514 x 10-3
3.040 x
+ [bt 22
jk
Xj xk Ujk/b2 - 2 t Xj ~ j / b ]* In [ (V + b ) /V]
f
0.40 - 6.846 x

(7)
The nec55sary criteria for equilibrium between two phases, butane and water at 289°K (60°F) and 6.8 atm (100 lb./
phase I aiid phase 11, are satisfied if, for each component of sq.in.) with various values of the interaction parameter kl2.
the phases, This figure has been constructed using Wilson’s version of
fir = ftII (8) the Redlich-Kwong equation with fugacities as given by
Equation ( 7 ) . The curves must indeed be tangent to the
Any computational procedure for phase equilibria using vertical at x1 = O + and x1 = I-, approaching from below.
equations of state rests, in the end, on finding phase com- Because the change in curvature takes place on such a
positions for which Equation ( 8 ) is satisfied. small scale near the axes it is impossible to show this be-
havior in the figure.
LIQUID INSTABILITY According to arguments first presented by Gibbs, if a
I t happens that the liquid phase described by Equations line can be drawn tangent to the free energy of mixing
( 1) through ( 7 ) ,when V is taken to be the smallest posi- curve at two points, any mixture of composition between
tive value which satisfies Equation ( l ) ,can be unstable. the two points of tangency must, at equilibrium, separate
That is, at equilibrium, the liquid must separate into two into two phases which have the end point compositions.
phases having different compositions. In Figure 1, it is possible to draw such a tangent line to
That this is true may be seen by looking at the Gibbs each of the curves except for k12 = 1.0. That is, liquid n-
free energy of mixing curve for the liquid phase. The text- butane, water mixtures at 289”K, 6.8 atm are predicted
book treatment of liquid immiscibility is usually presented to separate into two phases for interaction parameters less
with the use of such curves. [See for instance, Prausnitz than some value between 0.8 and 1.0. The composition of
( 1969) 1.
the two phases can, in theory, be determined from the
Taking the standard state to be the pure components at points of tangency of a common tangent line. For the water
unit fugacity and at the given temperature, the standard rich phase at least, this composition is apparently very close
free energy of mixing for a binary system is given by to being a pure phase.
Using a computational procedure which is described in
aGM/RT = xllnfl + x2lnf2 (9) detail in Appendix, the predicted mutual solubilities for
n-butane and water have been calculated for various values
Fugacities f l and f 2 must be in atmospheres in this expres- of k l z . These results appear in Table 1.
sion. At a given temperature and pressure, aGM/RT is a The predicted solubility of n-butane in water is con-
function of mole fraction x1 alone. The curve of aGM/RT sistently many orders of magnitude lower than the solubil-
versus x1 is tangent to the vertical at both XI = 0 and at ity of water in n-butane. It proves impossible to select a
x1 = 1. At x1 = 0 + , the derivative is - co and at x1 = 1- single value of the interaction parameter which predicts
the derivative is co.+ both mutual solubilities adequately. It was decided to con-
Figure 1 presents free energy of mixing curves for n- centrate on matching the data for the hydrocarbon-rich

Page 848 September, 1974 AlChE Journal (Vol. 20, No. 5)

TABLE
2. INTERACTION
PARAMETERS
FOR WATER-HYDROCARBON
SYSTEMS
AT 289°K

Interaction
parameter

Propane 0.46
n-bu tane 0.452
n-pentane 0.475
n-hexane 0.502
n-heptane 0.502
n-octane 0.502

TABLE
3. CALCULATED
SOLUBILITY
OF +BUTANE
IN HzO

k l z = 0.45 klz = 0.5

TOK mole fraction water in n-butane Experimental 0 02 40 60 80 I0
x,-MOLE FRACTION n-HEXANE

289 1.401 x 10-4 3.04 X 1.5 x 10-4 I

311 3.83 x 10-4 7.60 x 10-4 4.7 x 10-4
339 1.156 x 10-3 2.08 X 1.75 x 10-3 2c
367 3.140 x 10-3 5.160x 10-3 5.4 x 1 0 4
394 8.387 x 10-3 1.256 X 10-2 1.5 X
-
A G ~
RT

phase in this work. I C

The data of Table 1 plot as a straight line on semi-

logarithmic paper with k12 on the linear axis. This fact
permits interpolation to find the value of k l z which matches
the experimental data. Data for the solubility of water in
0 0 2 04 06 08 I0
hydrocarbons are available in the API Technical Data X, -MOLE FRACTION n-HEXANE
Book (1970). In Table 2 are tabulated values of the in-
teraction parameter for water and selected normal paraffins
which permit Wilson's version of the Redlich-Kwong equa-
tion to predict the experimental solubility of water in the
hydrocarbon.
\ (CI I
TEMPERATURE VARIATION OF WATER SOLUBILITY
Calculations were performed for the predicted mutual
solubilities of water and n-butane over a range of tempera-
tures with k12 = 0.45 and k l z = 0.50. The results of these
computations appear in Table 3.
It is clear that the calculated solubilities show the cor- 0
r I I
02
I
04
I I I
06
I I
08
I
I0
rect qualitative behavior, increasing exponentially with x I -MOLE FRACTION n-HEXANE
temperature. I t is also clear that quantitatively accurate Fig. 2. Standard tree energy of mixing for water, n-hexane at 378°K
solubilities of water in n-butane cannot be obtained with a and various pressures. (A) P = 0.68 atm, (6) P = 2.04 atm,
single constant value of k12. (C) P = 4.08 atm.
Based on the computed results of Tables 2 and 3 it was
decided to recommend that a constant value k12 = 0.50
be used for water and normal paraffins. There will un-
doubtedly be some loss in accuracy of calculated solubili- V at a given temperature and pressure for a binary mix-
ties at higher temperatures. Considering, however, that the ture, it is possible to prepare a free energy of mixing curve
predicted solubility of hydrocarbons in the water layer is for the vapor phase as well as for the liquid phase by using
low by many orders of magnitude it appears inconsistent to the largest positive value of V as the volume of the vapor.
add the complexity of making k,, a temperature function. By inspection of these two curves it is then possible to de-
A value of k I 2 = 0.50 permits reasonably accurate solubil- cide whether, for a given binary composition, the condition
ity predictions for water in n-butane over a temperature of minimum Gibbs free energy is met when the system is
range from 300" to 400°K. homogeneous in the liquid or in the vapor phase or
whether the system must separate into two phases.
LIQUID-LIQUID-VAPOR EQUILIBRIA The necessary condition of equilibrium between two
phases is met if the free energy of mixing curves have a
If three phases are present in a system at equilibrium common tangent line, just as in the liquid-liquid equilib-
it is necessary for each component which is present in all rium case.
three phases, that Figures 2 and 3 present free energy of mixing curves
f i r = f i n = fizz1 (10) and their evolution with pressure for n-hexane and water
This equation follows from the requirement that the Gibbs at 378°K (22OOF) and 450°K (350"F), respectively, with
free energy be a minimum. k12 = 0.5. Figure 2 is somewhat more simple than Figure
Free energy of mixing curves can be of value in deter- 3. At each of the pressures considered in Figure 2 and over
mining equilibrium conditions even in liquid-vapor or the whole composition range, the equation of state is satis-
liquid-liquid-vapor systems. On solving Equation ( 1) for fied by three volumes.

AlChE Journal (Vol. 20, No. 5) September, 1974 Page 849

 mixing curves have a common tangent line with points of
tangency at points a and b, respectively. If the mole frac-
tion of n-hexane exceeds that at point b the mixture re-
mains a homogenous vapor.
In Figure 2c the pressure is 4.08 atm (60 lb./sq.in.) and
is well above the vapor pressure of n-hexane. It can be seen
that a tangent line can be drawn between two points on the
liquid curve, representing almost pure water and almost
pure hexane, which line lies completely below both the
vapor and liquid curves. At some pressure, just below 4.08
atm, this line would be tangent to the vapor curve as well.
At 4.08 atm, two liquid phases must be at equilibrium. At
some specific pressure just below 4.08 atm it would be
x , - MOLE FRACTION n-HEXANE IN WATER
possible to have two liquid phases and a vapor phase at
equilibrium.
Figure 3, in which the temperature is 450"K, shows the
same qualitative behavior as pressure is increased. From
Figure 3a, at P = 3.40 atm (50 lb./sq.in.), we conclude
that any mixture of n-hexane and water is wholly in the
vapor phase at equilibrium. At 5.10 atm (75 lb./sq.in.)
there is one region of vapor-liquid equilibrium. At pressures
of 17.0 and 20.4 atm (250 and 300 lb./sq.in.) there are
two separate regions of vapor-liquid equilibrium. When
I
0
I I
02
I I
04
I I
06
1 I
08
1 J
I0
the conditions of Figure 3e are reached, with P = 27.2
q - M O L E FRACTION n-HEXANE IN WATER atm (400 lb./sq.in.), there can be no vapor phase and,
over a broad range of compositions, mixtures of n-hexane
and water must separate into two liquid phases.
The new feature of Figure 3 is that the free energy of
mixing curves for both the liquid and vapor phases are
shown to have discontinuities. These discontinuities occur
when, at some composition, the equation of state changes
from having three volume roots to having only one root. It
is possible that either the largest or smallest positive vol-
ume becomes discontinuous with composition.
0
I I
02
I I
04
I I
06
I I
08
I I
IC
In principle these features can be dealt with without
ambiguity and correct conclusions can be drawn regarding
conditions at equilibrium. It must be apparent, however,
that certain computational problems can occur. The disap-
pearance of a root to the equation of state corresponding
to a liquid volume could cause an iterative procedure to
fail. These difficulties, however, are not peculiar only to
liquid-liquid-vapor computations.
There are some additional problems possible which be-
2ot I 1 I I I I I I I I
come apparent on inspection of Figure 3e. In addition to
the line tangent to the liquid curve at two points in this
0 02 04 06 08 10 figure, it is possible to draw two lines each of which is
tangent to both the liquid curve and the vapor curve. On
each of these lines, the points of tangency obey the neces-
sary conditions for vapor-liquid equilibrium; that is, equal-
ity of fugacities. If a mixture containing 0.65 mole fraction
n-hexane in water is flashed it will be seen from Figure 3e
that it is possible to compute two different pairs of vapor-
liquid phase compositions at which Equation (8) is satis-
fied for both components. Neither composition pair will be
2 or
0
I
02
I I
04
I I
06
I I
08
I I I
I 0 correct since the correct equilibrium condition has two
liquid phases.
Fig. 3. Standard free energy of mixing for water, n- hexane a t 450°K
and various pressures. (A) P = 3.40 otm, (6) P = 5.10 atm, COMPUTATIONAL SCHEME
(C) P = 17.0 atm, (D) P = 20.4 atm, (E) P = 27.2 atm.
In performing the three-phase and liquid-liquid equilib-
rium computations in this work, we used a new calculation
In Figure 2a, where P = 0.68 atm (10 lb./sq.in.), the procedure which, it is believed, avoids some of the poten-
vapor free energy of mixing curve lies completely below tial difficulties outlined above.
the liquid curve. The homogeneous vapor therefore has Erbar (1973) has outlined a computational scheme for
the minimum possible Gibbs free energy, regardless of three-phase equilibrium problems. His formulation of the
composition. problem is similar to that of Deam and Maddox (1969).
At 2.04 atm (30 lb./sq.in.), the condition of Figure 2b, It involves evaluating phase equilibrium constants Ki and
the vapor pressure of water is exceeded. Over a composi- proceeds in an iterative manner until the sum of the mole
tion range, whose endpoints are indicated by points a and fractions in all phases is one. Erbar has reported results
b, the system must divide into a liquid phase, which is of computations for a hydrocarbon, water liquid-liquid-
almost pure water, and a vapor phase. The free energy of vapor system and for solid-liquid-vapor systems in which

Page 850 September, 1974 AlChE Journal (Vol. 20, No. 5)

 Water, Paraffin Mixture

COMPONENT
FEED
MOLES //’ I Three-phase flash computations were performed on a
system made up from 25 moles of propane, 25 moles of n-
butane, 30 moles of pentane, 10 moles of n-hexane, 20
moles of n-octane, and 40 moles of water. With the pres-
sure held at 23.8 aim (350 Ib./sq.in.), the number of moles
in each of the three phases and the phase compositions
were computed over a range of temperatures. The results
of these calculations are presented in Figure 4 along with
the condensation curves presented by Erbar (1973) for
this mixture.
Erbar’s condensation curves and the condensation curves
from this work are strikingly similar. The major difference
is that Erbar’s predicted bubble point and dew points are
3 to 5°K lower than those computed here. In Erbar’s calcu-
lations a separate description is used for each phase. In this
work, all three phases are described by Wilson’s version of
the Redlich-Kwong equation.
The detailed results of computations at 422°K (300°F)
are summarized in Table 4. These results also are very
similar to those of Erbar. The major difference is that in
this work the mole fraction of water in the hydrocarbon
liquid is 0.0229 and in Erbar’s procedure it is calculated
”
400 410 420 430 440 450 to be 0.0075. Procedure 9A1.5 in the API Technical Data
TEMPERATURE, O K Book predicts a water solubility of 0.036 mole fraction. If
the hydrocarbon-water interaction parameters were in-
Fig. 4. Condensation curves predicted for a hydrocarbon-water mix- creased, a higher water solubility would be predicted.
ture a t 23.8 atm. Values of kij = 0.50 were employed here.

TABLE4. THREE-PHASE
FLASHRESULTS Methane, n-Butane, Water
Phase equilibrium computations were performed on the
T = 422”K, (300°F) P = 23.8 atm, (350 lb./sq.in. ) system methane, n-butane, water which has been studied
experimentally by McKetta and Katz ( 1948). The interac-
Moles tion parameter for the methane, n-butane pair was fixed
Hydro- Moles
carbon water at k12 = 1.01. This value was selected to match the binary
Moles rich rich Moles vapor liquid equilibrium data at 311°K (100°F) and 20.4
Component feed liquid liquid vapor KHC atm (300 Ib./sq.in.) . The n-butane, water interaction pa-
rameter was set at 0.50 as in the preceding calculations.
Propane 25 7.93 - 17.07 2.33 Although the methane, water interaction parameter could
n-butane 25 11.55 - 13.45 1.26 have been manipulated to match some of the three-phase
n-pentane 30 17.74 - 12.26 0.7483 data of McKetta and Katz it was found unnecessary to
n-hexane 10 7.03 - 2.97 0.4567 do so since a value of 0.50 provided a very good match of
n-octane 20 17.27 - 2.73 0.1714 the water mole fraction in both the hydrocarbon rich
Water -40 -1.44 -
28.91 - 9.65 -7.26 phase and the vapor phase at 311°K.
Total 150 62.96 28.91 58.13 The methane, n-butane vapor liquid equilibrium results
are presented in Figure 5. Calculated results are presented
Predicted water solubility in the hydrocarbon liquid is 0.0229 mole for the binary two-phase data and for the ternary three-
fraction.
phase data. In both cases there are two degrees of freedom
and the temperature and pressure fix the composition of
various solid phases may form. all three phases. In the case of the ternary data the com-
Lu et al. (1974) have developed a modified regula-falsi positions of the vapor and hydrocarbon liquid phases are
method for calculating three-phase vapor-liquid-liquid reported on a water free basis.
equilibria. Lu’s modification of the Redlich-Kwong equa- One of the more interesting features of the McKetta and
tion was used with different parameters for each compo- Katz data is that the presence of water has a large effect
nent in each of the three phases at equilibrium. on the hydrocarbon vapor liquid equilibrium in some
Dluzniewski et al. (1973) have outlined an application ranges of temperature and pressure. At 344°K (160°F)
of free energy minimization to solve a liquid-liquid-vapor and 378°K (220°F) it was found experimentally that the
equilibrium problem. water saturated vapor was relatively more rich in methane
The procedure used here is also based on free energy than the dry vapor.
minimization. In detail it is much different from the pro- In the calculated equilibria, the presence of water proves
cedures of Dluzniewski et al., Erbar, or Lu et al. It is pre-
to be important only at temperatures higher than 378°K.
sented in the Appendix.
The computed water saturated vapor is relatively less rich
in methane than the computed dry vapor.
EXAMPLES The data of McKetta and Katz include the three-phase
Two numerical examples of three-phase equilibrium critical region. They found that the three-phase critical
results will be presented. The first deals with the same pressure at a given temperature was consistently slightly
hypothetical mixture used as an example by Erbar (1973). higher than the methane, n-butane critical pressure at the
The second deals with the system methane, n-butane, water same temperature.
and those results are compared with the experimental data Although the calculation scheme employed is not suit-
of McKetta and Katz (1948). able for locating critical points exactly, it was possible to

AlChE Journal (Vol. 20, No. 5) September, 1974 Page 851

 - CALCULATED
A.m, DATA O F
MCKETTA 0 KATZ
0
5)
0
_I

z
Pn
3
[L
a
00
.. 378'K
>
I
W
I
k-

. . ...
n
W

k
3
z
Q
c
0
.. 344-K
**
/

s
LL

W
0
J
I

0 001

311-K

I I I I I I I I I
0 20 40 60 80 100 120 140 160
PRESSURE, A T M

Fig. 7. Water mole fraction in the hydrocarbon liquid; three-phase

equilibria in the system methane, n-butane, water.

MOLE FRACTION METHANE IN n-BUTANE,WATER FREE BASIS

Fig. 5. Vapor-liquid equilibrium in the system methane, n-butane complete computations close to the two-phase and three-
water. phase critical points. The qualitative behavior of the three-
phase critical pressure given by Wilson's version of the
0 I(

xoo Redlich-Kwong equation is consistent with the experimen-

tal data in being slightly higher than the computed binary
critical point at a given temperature.
The agreement with experimental critical data is vari-
able. At low temperatures, calculated critical pressures
are high, at 378°K the agreement is very .good, and at
41i'K\- -CALCULATED higher temperatures the calculated critical pressure is too
low.
n The water content of the t w o hydrocarbon phases in the
8
3 three-phase region is presented in Figures 6 and 7. The
W
\ \344-K calculated water mole fraction in the vapor phase agrees
+ extraordinarily well with the experimental data at 311"
z
z and 344°K but is low at the higher temperatures. This
5 00
agreement can, no doubt, be attributed to the success
of Wilson's version of the Redlich-Kwong equation at
W
matching the vapor pressure of pure substances and at de-
0
J
scribing the vapor phase. Since the water phase is almost
n
W \ 378OK pure water, the vapor pressure of water plays a large role
k in determining the water content of the vapor phase.
The computed water mole fraction in the hydrocarbon
liquid, as can be seen in Figure 7, is in agreement with
the experimental data at 311°K and is somewhat low at
higher temperatures. The agreement is comparable to that
obtained for pure n-butane. The match between the ex-
perimental data and computed results could be improved
40 100
0 20 60 80 120 140 160
by making the water, hydrocarbon interaction parameter
PRESSURE, AT M
an increasing function of temperature. Nonetheless, the
Fig. 6. Water mole fraction in the vapor phase; three-phase equilib- calculated mole fractions are no more than 30% low over
ria in the system methane, n-butane, water. the whole of the experimental range and at the lower pres-

Page 852 September, 1974 AlChE Journal (Vol. 20, NO. 5)

 5. HIGHPRESSURE
TABLE SOLWBIL~TY
OF WATER SUMMARY OF EXAMPLES
IN METHANE,BUTANE
The three-phase hydrocarbon water equilibrium exam-
ples discussed here demonstrate that it is feasible to de-
Mole fraction water
scribe all three phases by a single equation of state. Al-
Mole
fraction though the Wilson version of the Redlich-Kwong equation
CH4, was used here, the same approach is likely to yield similar
water P = 136.1 a h P = 204.1 a h results with other equations of state.
T”K free Calc. Exptal. Calc. Exptal. Many of the important qualitative characteristics of these
three-phase systems are correctly predicted in. this ap-
311 0.732 0.000849 0.000804 0.000745 0.00065 proach. The quantitative agreement between the calcu-
344 0.630 0.00303 0.00328 O.QO251 0.00274 lated and experimental values of the water content in the
378 0.4925 0.00832 0.01056 0.00661 0.00795 two hydrocarbon phases is sufficiently good to consider the
411 0.219 0.01640 0.0253 0.01330 0.0193 use of the proposed procedures for design purposes.

NOTATION
sures the agreement is considerably better. a, b = parameters in the Redlich-Kwong equation
The instability of the liquid phase continues at pressures aij = coefficient defined by Equation (5)
above the three-phase critical pressure for the mixture. It bi = coefficient defined by. Equation
- (4)
has been possible to calculate the water mole fraction in f = fugacity
the hydrocarbon liquid at 136.1 atm (2000 lb./sq.in.) and G’ = defined by Equation (A.18)
204.1 atm (3000 lb./sq.in.) and compare these results kii = interaction parameter for the i, i pair
with experimental data. The composition of the hydrocar- n, p , q, r = mole numbers
bon liquid has been taken to be the composition at the P =pressure
three-phase critical at the given temperature. The calcu- P, = critical pressure
lated and experimental results are presented together in R = universal gas constant
Table 5. T = temperature
Quantitative agreement of the calculated results with T , = critical temperature
the experimental results is better at lower temperatures. V = molar volume
As before, the agreement at higher temperatures could x = mole fraction
be improved by making the interaction parameters an Z = compressibility factor
increasing temperature function. Perhaps more interesting
is that the qualitative effects of pressure on the water solu- Greek Letters
bility in the hydrocarbon phase are duplicated by Wilson’s aGM = standard Gibbs free energy of mixing
version of the Redlich-Kwong equation. I n both the ex- w = acentric factor
perimental and computed results, the equilibrium water Subscripts
content decreases as pressure is decreased, following the
trend which exists at pressures below the three phase criti-
i, i, k = substance index
cal point in the vapor phase. Up to the three-phase critical
I, I I , I l l = phase index
pressure the water content of the hydrocarbon liquid is
seen to increase as pressure is increased.
LITERATURE CITED
The hydrocarbon content of the water rich liquid is,
of course, not accurately predicted at any temperature API Technical Data Book, American Petroleum Inst., Div. of
or pressure. It was possible to select k12 to match equilib- Refining, Washington, D. C. ( 1970).
rium data in one of the liquid phases but not in both and, Deam, J. R., and R. N. Maddox, “How to Figure Three-Phase
as stated before, it was decided to concentrate on the Flash,” Hydrocarbon Proc., 78(7 ) , 163 ( 1969).
hydrocarbon liquid. If accurate predictions of hydrocarbon Dluzniewski, J. H., S. B. Adler, H. Ozkardesh, and H. E.
Barner, “Aid to Correlation of Complex Equilibria,” paper
mole fractions in the water liquid are required, some other presented at the 75th National AIChE Meeting, Detroit
procedure must be used. (1973).
However, in some respects the behavior of the computed Erbar, J. H., “Three Phase Equilibrium Calculations,” Proc.
results does correspond qualitatively with the experimen- 52nd Ann. National Gas Processors Ass3oc. Meeting, 62
tal data. The methane solubility is computed to increase ( 1973).
sharply with pressure while the n-butane solubility de- Kohn, J. P., “Heterogeneous Phase and Volumetric Behavior
creases. At 378”K, the computed methane mole fraction of the Methane n-Heptane System at Low Temperatures,”
varies from 0.396 X 10-6 to 1 X lop5 as pressure varies AlChE J., 7,514 ( 1961).
from 20.4 to 103.4 atm. The corresponding experimental Lu, B. C.-Y., P. Yu, and A. H. Sugie, “Prediction of Vapor-
Liquid-Liquid Equilibria By Means of a Alodified Regula-
values are 0.5 x and 1.30 x respectively. On Falsi blethod,” Chem. Eng. Sci., 29, 321 ( 1974).
the other hand, the computed n-butane solubility decreases hlcKetta, J. J., Jr., and I). L. Katz, “Methane-n-Butane-Water
slightly from 0.576 x 10-lo to 0.422 x over the System in Two-and Three-Phase Regions,” Ind. Eng. Chem.,
same pressure range. At 378”K, experimental data are 40,853 ( 1948).
0.286 x 10-3 and 0.98 x for the mole fraction of Prausnitz, J. M., “;2lolecular Thermodynamics of Fluid-Phase
n-butane in the water phase at 20.4 atm and 103.4 atm, Equilibria,” Prentice-Hall, Englewood Cliffs, N. J. ( 1969).
respectively. Schindler, D. L., G. \V. Swift, and F. Kurata, “blore Low
In the experimental data, the total hydrocarbon solubil- Temperature V-L Design Data,” Hydrocarbon Proc., 45 ( l l ) ,
ity in the water phase is shown to pass through a minimum 205 (1966).
\I’ilson, G. l l . , “Vapor-Liqaid Equilibria, Correlation by Xleans
as temperature increases at constant pressure. This impor- of a hlodified Redlich-Kwong Equation of State,” Adu.
tant qualitative result is not in any way apparent in the Cryogenic Eng., 9, 1G8 ( 19G4).
computed equilibrium solubilities. At every pressure, the ., “A hlodified Rcdlicli-Kwong Equation of State, Ap-
computed mole fraction of the total hydrocarbons in the plication to General Physical Data Calculations,” paper pre-
water phase increases exponentially with temperature. sented at the 65th National AIChE hltg, Cieveland ( 1969).

AlChE Journal (Vol. 20, No. 5) September, 1974 Page 853

APPENDIX A. THREE-PHASE FLASH motivate the use of an integration procedure which is simple
CALCULATlON PROCEDURE but which permits use of a relatively large increment At.
In this work the integration of Equations (A.lO) and ( A . l l )
The following development is applicable to flash type cal- is approximated by
culations only. While extensions of the approach taken here to
dew point or bubble point calculations may be possible, there pi’ + = pi’ ( f i z z I / f i z ) A t (A.12)
are some conceptual difficulties. and
A given mixture consisting of n1, n2, .. ., n~ moles of the qij+ = qij ( f i z z z / f i z I ) A t (A.13)
respective substances is postulated to separate into three phases
.
having pi, , . ., p~ moles in phase I , 91, . ., q~ moles in phase where pi’+’ and p$ are trial values of the mole number pi at
11, and 71, .. ., TN moles in phase 111. Conservation of mass successive steps of the integration or in successive iterations.
requires that These formulas are virtually Euler’s forward integration of the
+ +
pi 9i Ti = ni (-4.1) differential equations.
It is possible to assure that negative mole numbers can never
and all the mole numbers must, of course, be nonnegative. be generated by using Equations ( A . 1 2 ) and (A.13) only if
Given these constraints, the system must split into three phases fir11 < fiI and fizz1 < fizr; that is, only if both pi and 9i are de-
at the given temperature and pressure in such a way that the creasing. Otherwise, if fir < fizz and fiI < f i I r r then equations
Gibbs free energy is minimized.
There are two independent mole numbers for each substance 4ij + 1 = 9ij (fiI/firz ) “ *. (A.14)
in the problem as stated and it is desirable to leave some flex-
ibility tor a given substance, say substance i, in selecting pi, and
9i or pi, Ti or qi, ri as the independent pair. Whichever pair is rij+l = rij (f.11/ f lZII
. )At (A.15)
chosen as independent, it is quite a simple matter to form the
gradient of the Gibbs free energy. Taking pi and qi as inde- are used. If fir1 < fir and fizz < fizIz, then equations
pendent pi’ + = Pi’ ( fizz/fiz )At (A.16)
and
rij+ 1 = r$ (fizI/fizzI)At (A.17)
are used. With these equations the mole numbers being iterated
(A.3 1 upon are certain to be decreasing. The remaining mole number
is calculated from Equation ( A . 1 ) and cannot become negative.
Taking pi and ri as independent It remains to find the increment At in this procedure. The
criterion that must be met is that a t each stage of the computa-
tion the total Gibbs free energy of the system decreases. It is
(‘4.4 1 possible to show that, in the vicinity of the minimum point,
there is a maximum At for which the free energy decreases and
that this A t is also the maximum increment for Euler’s forward
integration to be stable. Furthermore, this maximum A t is a
constant and convergence is assured if any value less than this
Taking qi and ri as independent constant value is used in the iterations.
The iterative procedure which was used involves determining
at each step whether the free energy of the system has de-
creased. I t is equivalent that G’ decrease where
G’ Li (pi Infir + 9i lnfizz + ri WirII) (A.18)

The quantity G’ is simple to compute from quantities which

For different substances different choices may be made about must be computed a t each iteration in any case. If G’ has in
the independent mole numbers. fact decreased, A t is increased,
Supposing that pi and qi are independent, the path of steep- Atj+l = 1.1 Atj (A.19)
est descent in the free energy, from any initial condition, is
given by If at some stage G’ is found to increase, then At is decreased
dpi
-
dt
+
= - hfir bfiIzI (A.8 1 Atj+l = 0.7 Atj (A.20)
It has been found necessary to prevent At from ever again
increasing after having once been decreased so that oscillations
can be avoided. The factors 1.1 and 0.7 in Equations ( A . 1 9 )
and ( A . 2 0 ) are arbitrary but appear to be satisfactory.
where t is a dummy variable. I t appears that Equations ( A . 8 ) It has also been found desirable to start iterations with a
and ( A . 9 ) , (or similar equations) can be integrated to reach small At, say A t 0.02, so that in the initial part, the process
the point of minimum free energy, at which point dpi/dt = 0 closely approximates integration and large changes in the mole
and dqi/dt = 0 and equality of fugacities is assured. There is, numbers can be avoided.
however, the possibility that one of the phases must disappear The speed of convergence of this procedure depends on the
at equilibrium so that there is no equilibrium point which cor- proximity of the equilibrium condition to any critical points
responds to nonnegative mole numbers. and, to a large extent, on the composition of the mixture being
On the other hand, the following equations also move toward flashed. Convergence is slow, in general, when the quantity of
the minimum Gibbs free energy but the solution is limited to one of the phases becomes small at equilibrium.
nonnegative pi and qi. It has been observed also that near the dew point or bubble
point conditions it is possible to converge to an incorrect solu-
(A.lO) tion in which there is only one hydrocarbon phase, depending
on the initial guess of the phase compositions.
Since for each component of the mixture there are three dif-
ferent possible sets of equations to be used in the computations,
(A.ll) there are 3N permutations in the overall set of equations. The
convergence properties are variable depending on exactly which
In integrating these (or similar) equations, accuracy of the set of equations is used.
integration is not important. I t is only important to approach Convergence depends to some extent on the scheme used to
the stationary state as rapidly as possible. These considerations obtain the increment At. Equations ( A . 1 9 ) and (A.20) were

Page 854 September, 1974 AlChE Journal (Vol. 20, No. 5)

used in this study, but it is obvious that other procedures could tended to moR than three phases which makes it a candidate
be employed. for use in computations where solid phases need be considered,
The convergence criterion used was that as in cryogenic applications.
Since the procedure minimizes free energy, it is possible to
avoid spurious solutions to the equilibrium problem such as
those pointed out in the discussion of Figure 5e. In the liquid-
With this criterion, convergence occurred, at best, with 50 liquid-vapor computations the starting point was that almost
iterations. At worst more than 1000 iterations may be required. all the water was in a water rich liquid with a few percent of
Computations were performed on a CDC 6400. On this ma- the total water being present in the two hydrocarbon rich
chine the computing time was roughly 0.003 sediteration per phases. The hydrocarbons were distributed so that the vapor
component. As indicated, computing time increases linearly with phase was rich in the light components and the liquid phase
the number of components. was rich in the heavy components. This assured convergence
The advantages of this procedure lie in its ease of formulation to the correct equilibrium conditions.
and programming, its tie to thermodynamic principles, and its Manuscript received January 31, 1974; revision received May 16
reliability. The formulation presented here can easily be ex- and May 20,1974.

Determination of the Kinetics of Secondary

Nucleation in Batch Crystallizers
The kinetics of secondary nucleation have been determined from mea-
surement of the supersaturation as a function of time following the addi- S. G. KANE
tion of seed nuclei to a supercooled solution in a well-stirred batch crys- T. W. EVANS
tallizer. Population balance mathematics have been used to show that the P. L. T. BRIAN
secondary nucleation kinetics may be inferred from the supersaturation-
and
time curve. The method has been applied to the determination of the A. F. SAROFIM
kinetics of the secondary nucleation of ice and found to give results in
excellent agreement with those obtained from tedious particle counts. In Department of Chemical Engineering
addition, it has been shown that the moment of the particle size distribu- Massachusetts Institute of Technology
tion that best correlates the nucleation rate data can be inferred from the Cambridge, Massachusetts 02 139
initial transient of the supersaturation-time history.

SCOPE
The use of batch crystallizers for the determination of crystals occurs with an insignificant variation in the
the kinetics of secondary nucleation is complicated by the supersaturation. Measurement of the number of crystals
variation during an experiment in both the number of during his period of relatively constant supersaturation
crystals and the supersaturation. Motivation for the use provides one method of obtaining nucleation kinetics.
of batch crystallizers is provided, however, by the rela- Particle counts, however, may be tedious or difficult, par-
tive ease with which a large range of operational variables ticularly for systems such as ice where crystals are rela-
may be studied in a short time. tively unstable. One major objective in this paper is
This paper develops techniques for obtaining kinetics therefore the development of methods for inferring nu-
of secondary nucleation for one mode of operation of a cleation kinetics from the supersaturation measured at
batch crystallizer. A seed crystal is introduced in a super- times greater than the induction time.
cooled solution and measurements are made of the varia- It should be noted that the inference of nucleation
tion with time of either the number of crystals or the kinetics is complicated by the fact that the rate of second-
supersaturation. Up to a critical time, designated here as ary nucleation may be proportional to the nth moment of
the induction time, the multiplication in the number of the particle size distribution

N = u
Correspondence concerning this paper should be addressed to A. F. Complete characterization of the nucleation kinetics
Sarofim. S. G. Kane is with Amoco Chemicals Corporation, Naperville,
Illinois. T. W. Evans is with T h e Upjohn Company, Kalamazoo, Michi- would therefore require the determination of two param-
gan. P. L. T. Brian is with Air Products Corporation, Allentown, Penn-
sylvania. eters n and an. It will be shown, however, that a single

AlChE Journal (Vol. 20, No. 5) September, 1974 Page 855