February 1948 I N D U S T R I A L A N D E N G I N E E R I N G CHEMISTRY
y,, 7 2 = activity coefficients
Vl, Vz = molal volumes of the liquid components
V , = molal volume in the gaseous state
B = second virial coefficient
Po, To = critical pressure and temperature
T = an arbitrarily fixed temperature
s = slope factor, Equations 11 and 12
(Y = relative volatility
LITERATURE CITED
(1) American Petroleum Institute Research Project 44, National
Bureau of Standards, “Selected Values of Properties of Hydro-
carbons,” Tables 2k (March 31,1944) and 5k (June 30, 1944).
(2) Beatty, H.A,,and Calingaert, G., IND.ENG.CHEH.,26,504,904
(1934).
(Thermodynamics of Nonelectrolyte Solutions)
ALGEBRAIC REPRESENTATION OF THERMODYNAMIC
PROPERTIES AND THE CLASSIFICATION OF SOLUTIONS
OTTO REDLICH
Shell Development Company, Emeryville, Calif.
T h e utilization of laboratory data for the design of distillation columns and other separation equip-
ment requires the efficient representation of extensive experimental data. A flexible, nonarbitrary,
and convenient method is developedfor systems of two or more components. This method furnishes
an immediate distinction between various types of solutions.
E XPERIMENTAL results can be improved or damaged on
their way from the laboratory to the practical application.
in plant design and operation. The treatment of experimental
data should eliminate inconsistencies without distorting the
results by imposing arbitrary conditions, it should be flexible
enough to cover all important cases, and it should be pimple in
operation, From this viewpoint the following method was
developed for the representation of thermodynamic properties
of nonelectrolyte solutions. The present discussion starts from
binary solutions and is later extended to systems of more com-
ponents.
SELECTION OF A USEFUL FUNCTION
The use of the activity coefficients y1 and y 2 is not advisable
since they are redundant. The use of two functions necessitates
the imposition of a condition-namely, Duhem’s equation. This
complicates any correct smoothing procedure. For this reaeon
Scatchard’s “excess free energy” is preferable. Dividing this
function by 2.303RT, one obtains
Q = x log YI + (1 - Z) log -i2
(z mole fraction of the first component) , which y a y be considered
a little more convenient in some numerical calculations.
* For various reasons still ’more suitable is the function
dQ/dx = log ( Y I ~ Y ~
The degree of this function is one unit lower than Q as well as the
functions
log 71 = Q + (1 - s)dQ/dx and log y~ = Q - xdQ/dx
This is a considerable advantage in practical calculations,
realized in a special case already by Benedict et al. (1). Another
advantage of the function log (y1/y2) is its simple relation to the
experimental data and to the technically important relative
volatility :
345
(3) Benedict, M.,Johnson, C. A,, Solomon, E., and Rubin, C. C.,
Trans. Am. Inst. Chem. Engrs., 41,371 (1945).
(4) Brunjes, A. S.,and Bogart, M. J. P., IND. ENG.CHEM.,35,255
(1943).
(5) Lewis, G.N., and Randall, M., “Thermodynamics,” New York,
McGraw-Hill Book Co., 1923.
(6) Othmer, D. F., and Benenati, R. F., IND.ENQ.CHEM.,37, 299
(1945).
(7) Scatchard, G.,and Raymond, C. L., J . Am. Chem. SOC.,60,1278.
(1938).
(8) Tongberg, C. O., and Johnson, F., IND.ENG.CHEM.,25, 733
(1933).
(9) York, R., and Holmes, R. C., Ibid., 34,345 (1942).
RECEJVED
November 26,1946.
AND A. T. KISTER
- z ) / x ( l - y)
for
a = y(1 (4)
log a = log ( Y l l r n ) -I- log (PYP,”) (5)
where g is the mole fraction in the vapor and p i and p i are the
vapor pressures of the pure components, the vapor being 118-
sumed to be perfect.
Furthermore, the function log (-y~/r~) provides an efficient
tool for eliminating inconsistencies in the experimental data.
Since &, according to Equation 1, assumes the value zero for
x = 0 and x = 1, we derive from Equation 2
Figure 1 shows a simple example for the application of this condi-
tion. The only curve for log (rl/r3which represents the data in
accordance with Equation 6 is the iero line-that is, the system
is perfect within the limits of experimental error. The deviations
(1)
for low concentrations of either component are safely recognized
t o be due to experimental errors. One could hardly arrive so
quickly and cogently at the same result by another method.
The relation
(2) (7)
following from Equations 5 and 6 is sometimes useful. If the
values of .(Y r:fer to a constant temperature, the integral is equal
(3) to log ( p l / p z ) . If (Y is derived from equilibrium measurements
over a moderate temperature interval, the integral can be easily
estimated since usually varies only little with the tempera-
ture. Equation 7 is based only on the assumptions that the
vapor is perfect and that the dependence of the activity coeffi-
cients on the temperature may be neglected.
 INDUSTRIAL AND ENGINEERING CHEMISTRY
Figure 1. 2,2,4-Trimethylpentan~Methylcyclohex-
ane at 741 Mm.
Measurements of Harrison and Berg (3). Zero line represent.
perfect solution.
SERIES DEVELOPMENT
The development of functions like Q, log y, etc., into power
aeries with respect to x has been suggested already by hlargules.
Por nonelectrolytes it is always permissible and at least in nu-
merous cases useful. The usefulness, however, depends greatly
on the form of the series. Kohl ( 7 ) , in reviewing the various
methods of representing activity coefficients, pointed out that the
series should be developed in such a way that the higher. terms are
oorrections of the terms of lower order. In this way he avoids
the inconvenience of having the result appear as a small difference
of large terms. Also important is the fact that the coefficients
of a properly chosen series furnish a natural classification of
various systems, as will be s h o m later.
Since Q = 0 for x = 0 and x = 1, each term must contain the
factor x(1 - 2). It is desirable to develop the series with respect
to a variable which is somehow symmetric with respect to the two
components. The simplest variable of this kind is 22 - 1, which
merely changes its sign on exchange of the components. Thus
the most useful development appears to be
Q -
= ~ ( ls ) [ B + C(22 - 1) + D ( 2 2 - 1)' + 1 (8)
The coefficients are determined either by plotting
voi. 40, No. 2
= 0, is represented by a straight line in a diagram of log(yl/r2)
against x. This straight line must, according to Equat,ions 6 or
10, pass through zero at x = 0.5.
TYPE3. The type B # 0; C # 0 ; D = . . . = 0 corresponds
to what is frequently called t'he equation of ?vlargules. If a large
number of experimental points of extremely high accuracy is to be
represented, the method of least squares with properly chosen
weights will be adopted for the determination of the constants B
and C. In general, however, the following method furnishes
in a few minutes the best results.
The method consists of plotting the experimental data for
log(yl/yz) against 2 , drawing a preliminary curve, and reading the
values of log(yl/yz) for a number of characteristic points listed in
TableI. For type 3-that is, D = &only points 1, 3, 5, 7, and
9 need be considered.
In points 3 and 7 the value of log(yl/yJ is independent of C.
In addition, the values of the function in these two point,s are
equal if three terms are sufficient in Equation 10. This condition
may be assumed to be satisfied in all cases except for extreme
deviations from the perfect solution-that is, especially for in-
completely miscible systems. The equality of the values of
log(yl/rt) for p0int.s 3 and 7 constitutes, therefore, in many cases,
a check of the experimental data. The value of B - D / 3 , or, if
D = 0, of B, is calculated from the best value of log(yl/y2) for
point,s 3 and 7.
The value of C is derived from point 5 . If D = 0, the quitnti-
- -
ties ( B C) and ( - B C) must represent reasonable values for
points 1 and 9. 9new curve is drawn through the points 1, 3, 5,
7, and 9 calculated according to Table I with the assumed values
of B and C.
The deviations of the experimental points from this curve are
to be judged with regard to Equation 6-that is, only such varia-
tions of the representative curve are permissible which do not
change the total area under the curve. With a little practico one
sees immediately whether a slight adjustment of B and C will im-
prove the agreement.
TYPE4. The type B # 0, D # 0, and c = 0 actual1,y exists.
Several systems of methanol and hydrocarbons can bo approxi-
mately represented by a corresponding function. The diagram of
1og(y1/y2) indicates this type immediately by the S-shape of the
representative curve. The absolute values of the function at, z =
0 and x = 1, are equal, and the curve passes through zero a,f,x =
0.5. The values of B and D are easily derived from points 2, 4,6,
and 8 (Table I).
TYPE5 , If none of the constants B, C, and D equals zero, C is
calculated from point 5 (Table I), B from points 2 and 8, and D
from point,s 4 and 6 in which the value of t'he 11 term reaches a
maximum.

or, preferably, from a diagram of I DISCUSSION O F TYPES

A principal advantage of a classification like that discussed in

the preceding section lies in its close connection with the nature of
+ D ( l - 2.r)ll - 8 r ( l - z)] + , . . (10) the solutions and of the components.
It is well known from the work of Hildebrand, Scatchard,
TYPES OF SOLUTIONS
Guggenheim, and others that type 2 very well approximates
systems the components of which are not associated, interact
TYPE1. The simplest case is the perfect solution for which only moderately, and have approximately equal molal volumes.
log(r,/rz) = 0. For systems which do not satisfy the last condition, Scatchard
TYPE 2. The next type, characterized by B # 0; C = D = . . , ( 5 ) suggested the relation (Equation 11).

I. CHARACTERISTIC
TABLE POINTS
h-0. 1 2 3 4 5

f,
NO.
B -;
0
+D 0.7071B
8
0.1464
- C/4
0.2113
0,5773 ( B
7
-D/3) 0.4082 ( B
0,2969
6
- 21)/3) + C/4 0.5
c/2
../
...
fog ( y I / y I ) - B -1C - D -0.7071B
0.8536
- C/4 0.7887
-0.5773 (B- D / a )
0.7041
-
-0.4082 ( B 2 0 / 3 ) + C/4 ...
February 1948 I N D U S T R I A L A N D E N G I N E E R I N G CHEMISTRY 341

2.303RTQ = AViVzX(1 - z)/[xVI + (1 - x)VZ] (11) The development of Q into a power series can be appropriateIy
generalized without difficulty. It is useful to rewrite Equation 8
where Vi and V2 are the molal volumes of the components. If in the more symmetric form
this equation is developed into a series similar to Equation 10,
the comparison of the coefficients furnishes Qlz = +
ZIXZ[BIZ Clz(~1 - 2,) + DIZ(ZI- X Z ) ~+ . . .I (17)
The series for a ternary system is then conveniently represented
B = 2AViV2/2.303RT(Vl f Vz)
C - B(Vi - Vz)/(Vi Vz) + by
+ +
D = B(Vi - VdZ/(V1 Vz)' + (12)
+ 212223 [C
Q Q23 Q3i

+ D i ( ~ z- + D2(23 - + . . . I
~ 3 )
Qiz

21) (181
The D term contributes less than 0.011B to log(yl/yz) if the The number of coefficients required for a series running to a
larger molal volume exceeds the smaller by less than 50%. certain degree has been discussed by Benedict et al. ( I ) . The
Scatchard's equation therefore can be replaced in this case for present representation offers again the advantage of a natural
data of moderate accuracy by a formula of type 3 with the classification by means of terms of decreasing importance.
value of C given in Equation 12, This formula is usually more Equations 16, 17,and 18 furnish for a ternary system
convenient than the original equation.
Figure 2 illustrates ?n application of this formula. Scatchard's log (7z/7i) = Blz(21 - 22) + Cm[3(21 - 2z)* - 11/2
equation has been found to approximate well all hydrocarbon + DlZ(Z1 - XZ)[(Xl - - 4x12221 + . . .
22)2
eystems except those containing benzene. + - B81 f Cz3(222 - + c31(221 -
23[B23 23) 23)
It will be shown in another paper of this series that association + Dz8(32s2 - 4x228 + - D31(32l2 - 42123 +
23') 23')
of one of the components produces a contribution to log(-yi/rz) + - + D1(-2321 f 2x122 + - zZZ)
c(Zl 22) 22x3
which has approximately the character of the ll term in Equation
10. Types 4 and 5 therefore will be discussed in connection with
+ +23zl + 22122 - - f . . I
02( 22x3 Xiz) s (19)
association. A few results, however, should be anticipated here. The actual calculation is simple, since the higher terms are always
Interassociation between the two components tends to diminish small and often negligible. Cyclic permutation of the subscripts
the influence of the term characteristic of association. This ex- furnishes log(7 3 / 7 2 ) and log(y~/y,~).
plains why systems of highly associating substances often belong For a system of four components Q is represented by the sum of
to type 3. the six binary functions of Equation 17, the four terms containing
The association term is maintained if both components associate 21x223, etc., and a term containing X I X Z X ~ X P . The further gen-
but not with each other, Apparently the system benaene- eralization is obvious.
cyclohexane belongs to this type. According to Hildebrand (4) EXAMPLE: ~HEPTANE-METHANOL-TOLUENE. The measure-
both substances are slightly associated. The measurements of ments of Benedict et al. ( I ) of vapor-liquid equilibria in the
Scatchard, Wood, and Mochel(6) can be represented by Equation system n-heptane-methanol-toluene furnish an opportunity to
10 with a finite value of D. show that even accurate measurements in a ternary system can be
It should be pointed out, however, thab the existqnce of this sufficiently well represented by the functions Q for the binary
small D term could be ascertained only by measurements of high systems alone, provided that adequate expressions for these
accuracy. Within the limits of error of the usual technique the systems have been found.
system would belong to type 3 rather than type 5. Thus the Systems with an associating component like methanol are
classification presented here furnishes approximations of various properly represented by means of the association function
degrees. This flexibility may be considered to be an advantage of
a power series. A ( K , x ) = 22 log (1 + K) -
MULTICOMPONENT SYSTEMS
210g (1 22 - + 11 + 4 K ~ ( l- ~)]')2)/'2(1- 2) (20)

The series used for binary systems can be extended without

difficulty to multicomponent systems. The definition of the 0.1 0 0
function Q by Equation 1is now replaced by
t o
Q = X Z log
~ ~k (13)
0.05
The relation of this function with the free energy furnishes for the
activity coefficient
I

t
2 0
The differentiation is to be performed at constant pressure and
temperature. Also, all mole fractions are to be kept constant
--
h
m
0
except the one indicated in the differential quotient.
It is convenient for some calculations to represent Q as a sum
of terms Q h which are homogeneous of the degree h in the mole
- 0.05
fractions xk. Equation 14 can then be written as
log y r = ZBQh/aXr - f(h - 1)Qh ' (15)
-0.1
The functions 1 I 0
log (rr/y*) = a Q / a ~ r- aQ/axa (16)
0 x 0.5
Figure 2. Toluene-2,2,4-Trimethylpentane at
offer the same advantages as the function defined in Equation 2. 760 Mm.
Their degree is one unit lower than that of 6? or logy?,and they are Menmaremento of Driakamar, Brown, and White (3). Curve
closely connected with the relative volatility. repremntm Santahard'a equation.
348 INDUSTRIAL A N D ENGINEERING CHEMISTRY Vol. 40, No. 2

coefficients-namely, eight from the binary jystems and three

TABLE
11. RELATIVEVOLATILITY OF ~-HEPTANE-~~ETHAN O L from the ternary system. The practical importance of
derived
A T 760 AIM.
the restriction of the number of empirical coefficients need not be
RIethanol/Heptane strrssed
1, oc. x2 Obsvd. Benediet This paper
60.60 0.138 16.1 16.45 16.5
59.47 0.178 12.6 12.65 12.55 CONCLUSIBSS
58.93 0,390 4.43 4.45 4.36
58.82 0.668 1.46 1.49 1.47 The use of the funct'ion log(rl/rz) for the representation of
58.81 0.810 0.696 0.691 0.702
59.01 0,885 0,423 0,407 0.419 free energy data of binary solutions is advisable. A suita-
59.90 0.946 0,242 0,243 0.246 bly chosen power series represents this function in a flexible and
unbiased ~ i a y . The coefficients can be derived quickly from
rxperimental data.
111. RELATIVE
TABLE VOLATILITY
OF >\IETHBSOL-TOLCENF: The nature of various classes of solutions is expressed in the
AT 760 MM. magnitude of these coefficients. The behavior of log(y , / y 2 )
RIethanol/Toluene
t, oc. 22 Obsvd. Benedict This paper
therefore can be predicted to a certain degree. Scatchard's
cquation furnishes a good approximation for systems of hydro-
70.25 0.130 19.0 15.07 17,s
66.44 0.266 9.90 9.75 9.83 carbons except those containing benzene.
65.58 0,407 5,94 5.72 5.93 The series used for the free energy can be advantageously used
64.47 0.593 3.10 3 07 3.13
64.10 0.692 2.16 2.17 2.20 also for other propertie., such as heat content and entropy.
63.79 0.779 1.54 1.65 1.58
63.67 0,843 1.23 1.19 1.21
63.58 0,852 1.01 0.99 1.01 ACKNOWLCDGIIENT
63.62 0.927 0.81 0.80 0.80
63.94 0,969 0.61 0.62 0.81
The authors cxpress their
~ ~ . -- gratitude to It. W. Millar and
TABLE11'. G. Muller for helpful discus-
RELATIVE
\70L.ITILITICY OF n - I l c P . r . ~ ~ ~ - - 1 ~ ~ T H , 4 ~ O L - - T n L U ~760
.4T SE lIA>I, sions on t,he subject of this and
Heptanc/Toluene IIcthanol/Toluene
t, ae. El I2 Ohsvd. Bencdict This paper Obsvd. Benedict Thiq paper
the preceding paper. L. Korba
90.77 0.0684 0,8486 2.75 2.78 2.72 1.15 1.19 1.24
assisted the authors in the
69.96 o.1.51.0 0.7733 2.11 2.26 2.32 1.58 1.64 1.80 calculations.
61.59 0.0787 0,7494 2.58 2.63 2.48 1.91 1.77 1.88
59.97 0.3412 0.5409 1.65 1.8 1.83 3.8 3.9 4.20
m61.35 0.1717 0.6038 2.07 2.17 2.08 3.22 3.25 3.26 LITERATURE CITED
82.01 0.0911 0,6114 2.43 2.42 2.25 3.13 3.06 3.08
64.34 0.0962 0,3711 2.24 2.13 2.04 7.14 6.8 6.92
62.96 0.2533 0,3172 1,85 1.92 1.82 8.3 8.5 8.6 (1) Benedict, M., Johnson, C.
A., Solomon, E., a n d
Rubin, L. C., Trans. Am.
Inst. Chem. Engrs., 4 1 ,
371 (1945).
I n tkis equation the mole fract,ion z refers t o the associating ( 2 ) Drickarner, H. G., Brown, G. G., and IVhite, R. R., Ibid.,
component, and is a 'Onstant characteristic Of this component' 41, 555 (1945).
(3) Rarrison, J. M , , andBerg, L,, IND, Est, CHEM.,38, 117 (1946).
The three binary systems have been found to be represented bv (4) Hildehraild, J. H., J . them. phys., 7, 234 (1939).
means of the coefficients (1 = heptane, . 2 = methanol, 3 = (5) S c a t c h a r d , G., Chem. Revs., 8, 321 (1931).
toluene) (6) S c a t c h a r d , G . , JVood. S. E., and Moohel, J. M.,J . P h ~ s Chem.,
.
43, 119 (1930).
B23 = 0.95; Bai = 0.117; Biz = 1.178 ( 7 ) Wohl, K., T,.ans. Am,. h t . C'hem. Eiigis., 42, 215 (1946).

Kz = 3.8; c
31 = 0.018; D12 = 0.155 (21) RECEIVED November 2 5 , 1946

ail other coefficients being zero. Accordingly,

the followingequations were used for the ternary
system:

log (?1/?'3) =
+
0 2 2 8 ~ ~O.117(x3 - xi) + 0 O l 8 [ z , ( ~-~2s3)
-k Za(Z8 - 2Z1)] + 0.155~2(3~:- + 4ZlXz 2:)
(23)
The corrections for the imperfection of the vapor
calculated by Benedlct et at. ueie used.
The results are compared n ith the experinlental
and calculated values of BenediLt el al in Tables
11, 111, and IV. These tables contain the tem-
perature, the mole fractions in the Irqurd, and the
relative volatilities observed and calculated by
Bcnerhct et al. and by means of Equations 22 and
23.
The volatilities calculated by mean3 01 these
equations agree with the observed values ap-
proximately a5 well as the values calculated by
Benedict et al.; but Equations 22 and 23 con-
tain only six coefficients, all derived from the
binary systems. Bonedict et d.introduced eleven Furfural Solvent Extraction Unit for Lube Oil