Fluid Phase Equilibria, 16 (1984) S-16 51

Elsevier Science Publishers B.V., Amsterdam - Printed in The Netherlands

CALCULATION OF CRITICAL POINTS AND PHASE BOUNDARIES

IN THE CRITICAL REGION

MICHAEL L. MICHELSEN

Instituttet for Kemiteknik, Danmarks Tekniske Hgjskole, Bygning 229, DK-2800 Lyngby
(Denmark)
(Received April 4, 1983; accepted in final form September 23, 1983)

ABSTRACT

Michelsen, M.L., 1984, Calculation of critical points and phase boundaries in the critical
region. Fluid Phase Equilibria, 16: 57-76.

A computationally efficient, generaI method is presented for calculating the critical

temperature and pressure of a multicomponent mixture of specified composition. The
procedure suggested also provides coefficients or, a,, and II relating changes in temperature,
pressure, phase fraction /3 and composition in the immediate vicinity of the critical point.
These variations are shown to be of the general form
T-T,=a,(l-2@)6
P-Pc=u,(l-2p)6
In K, = u,r;I/*&
where S is the distance parameter of the phase-boundary expansion. Finally, an explicit
first-order approximation is developed, based on derivatives of the Gibbs-energy surface,
giving the phase fractions and equilibrium phase composition for a mixture under near-criti-
cal conditions.

INTRODUCTION

A generally applicable procedure for the calculation of critical points for

multicomponent mixtures described by a single equation of state has recently
been developed by Heidemann and Khalil (1980). This procedure, which
requires no user-provided initial estimates of the critical temperature and
pressure, uses nested iterations, where the temperature corresponding to the
limit of intrinsic stability at a specified molar volume is determined in an
inner loop, while the molar volume is varied in the outer loop to satisfy the
condition that a cubic form evaluated at the stability limit must equal zero.
The method to be described here is based on the criticality criteria given
by Michelsen (1982b). The critical temperature and pressure are determined

0378-3812/84/$03.00 6 1984 EIsevier Science Publishers B.V.

58

by direct Newton-Raphson iteration, and a definite deficiency of the

method is that, using T and P as the independent variables, rather close
initial estimates are required. However, the method has the advantage of
being very cost-efficient with respect to evaluations of thermodynamic
properties and the associated algebraic operations. In addition, the partial
derivatives required by the algorithm are those normally applied in phase
equilibrium calculations.
A more important aspect of the method is that the quantities calculated
for critical-point determination can be related to relative changes in tempera-
ture, pressure and the composition and distribution of the equilibrium
phases in the critical region.

CRITICALITY CONDITIONS

Let the N-component mixture for which the critical temperature T, and
critical pressure PCare desired be of composition z. The criticality conditions
used here are based on tangent-plane stability criterion of Gibbs as modified
by Michelsen (1982a). The necessary and sufficient criterion for stability at a
specified temperature and pressure (T, P) is that

F(Y) = 1 + C y [In y + In &(Y) -In zi - In (p,(z) - l] 2 0 (1)

for any trial phase mole numbers Y. We shall use the term “locally stable”
for a mixture provided that there exists a closed region 1lY- zll< c, within
which F is nonnegative everywhere. At a critical point a specific behaviour of
F, to be described below, is required.
The first and second partial derivatives of F with respect to the y are
aF/ar; = gi = In & + In +i (Y) - In Z, - In +;(z) (2)
and
a+/ayaq = (11 y)sij +(a In $,/anj)y (3)
It is convenient to substitute as new independent variables a set of deviation
variables X given by
xi= (~.-zi)/ui (4)
with u. = z?/*
I . In terms of these variables, the above partial derivatives
becomk
aFlax, = uigi
and
a2F/axiaxj = u,uj(a2F/aqaq) (5)
 59

In particular,

(a*F/ax,ax,),=, = Bij = sij + uiOj(a In +i/anj)z (6)

The transformation leads to a scaling of the matrix of second derivatives
such that this matrix becomes the identity matrix for an ideal mixture.
We introduce a distance parameters and select X = su, where u is a vector
of unit length, u’u = 1. Considering Fin dependence on s along this line, the
following Taylor-series expansion is obtained:

F(X =SIU) = F,(s) = E (sm/m!)(d”F,/dP’)s=O

m=O
= as + bS2 + cs2 + G!s4+ o( S5) (7)
since F,(O) = 0. In addition, (I = C,n,gi(s = 0) = 0, and for sufficiently small
values of s the series is dominated by the s2 term. A necessary condition for
stability is therefore that

b = (1/2)~Bi,u,uj = (l/2)uTBu >, 0 (8)

i,j
regardless of the choice of u. The smallest value of the quadratic form (eqn.
(8)) is attained by choosing u as the eigenvector of B corresponding to the
smallest eigenvalue h min, or

Bu = X,i,u, uTu = 1 (9)

yielding b = A,,,/2. If this value of b is positive, the system is locally stable.
If b is negative, the system is intrinsically unstable, and if b equals zero, the
system is at the limit of intrinsic stability. The first condition that must hold
at a critical point is that b equals zero, i.e., X,, = 0. The second condition
that must hold is that the value of c corresponding to the eigenvector u is
also zero. A nonzero value of c with b = 0 implies that the mixture is
intrinsically unstable, since F, becomes negative for sufficiently small posi-
tive (if c is negative) or negative (if c is positive) values of s. Hence a critical
point is a point where, selecting u as the eigenvector of B corresponding to
Xmin, both b and c in eqn. (7) equal zero. Local stability at the critical point
requires at least that d > 0.
The numerical evaluation of b, c and d is performed conveniently as
follows. At the trial conditions (T, P), B is calculated and ( Ami,,, u) are
determined by inverse iteration (Wilkinson, 1965, Chap. 9). This yields the
coefficient b. For c, Michelsen (1982b) suggested the following numerical
determination. The derivative

d F,/ds = 26s + 3~s’ + 4ds3 + 0( s4) 00)

60

equals

c (aF/aq)dY,/ds = c uiuig, (11)

1 I
Since b is known, a single calculation of F and the gj (eqn. (2)) at a small
value of s, s = c, allows c (and d) to be determined. In particular, note that

OF, -c(dF,/ds), = 2br* + cc3 + O(c’) (12)

from which
c=(1/~~)[4F,(r)-r(dFr/ds),-2br*] +O(c*) (13)
Similarly,

d=(1/C4)[3F,(c)-<(fF,/ds),-be*] +0(r) (14)

Alternatively, the method of central differences, that is, evaluation of
dF,/ds ats=r ands= -c, yields

~=(1/6c*)[(dF,/ds),+(dF,/ds)_,] +O(c2) (15)

and

d= (1/8c3)[(dF,/ds), -(dF,/ds)_, -4661 +O(r*) (16)

that is, both c and d are obtained with an error proportional to c*.

ITERATIVE DETERMINATION OF CRITICAL POINT

The objective is to locate the temperature and pressure at which b and c

are both equal to zero. New trial values are generated by Newton-Raphson
iteration, solving the equations

(17)

(18)
where b and c represent the coefficients at the m th iteration and the
subscripts T and P denote their partial derivatives with respect to tempera-
ture and pressure. These partial derivatives can, of course, be determined by
numerical differentiation, but a much more inexpensive evaluation is possi-
ble.
Differentiation of eqn. (9) with respect to pressure yields
Bu, + Bpu = h,,u, + Xy (19)
 61

and

urup = 0 (20)
where X, and up are respectively the pressure derivative of Xmin and the
corresponding eigenvector, and BP is the pressure derivative of the matrix B.
Multiplication of eqn. (19) by uT yields A, directly:

uTBu, + uTBpu = X,i,,uTup + &uTu (21)

or

X, = uTB+ (22)
since uTB = Ami,uT and uTu, = 0. Equation (19) can now be solved for up:
(B - XminI)up = Y (23)
with Y = h,u - Bpll. The matrix of coefficients for this set of N linear
equations is of rank N - 1, and the augmented matrix (B - AminI, u) is also
of rank N - 1. Hence the general solution of eqn. (23) is
u,=h+Qu (24)
where h is any particular solution to eqn. (23) and Q is an arbitrary scalar
multiplier. Solving for h, Q is determined from the condition given as eqn.
(20) yielding Q = - uTh. The numerical procedure for solving the set of
homogeneous linear equations is described in Appendix A.
Explicit numerical evaluation of BP is not required, since only the vector
B,u is needed. The elements of this vector can be found from

(By); = u~~u,u~(# In c#+/&,W)~ = ~;$(a In &(Y)/aP),=,

= Uii/e)[(a ln cpi/aP),__, -(a ln +/aP),=,] + O(c) (25)

or, alternatively, by taking central differences between the pressure deriva-
tives at s = c and s = -e. The derivative b, equals X,/2. The derivative of c
with respect to P is found using eqn. (7):
aF,/aP = s’b, + s%, + s4d, + . . . (26)
Pressure differentiation using eqn. (1) yields

aF,/aP=C ~[(a In #i/aP)v-(a ln&/aP),] +CgJaF/aP) (27)

I I
where i3lpP = (aYpu,)ih,/aP = Sag. Evaluation of eqn. (27) at s = c
then yields
cp = (l/C3)[(aF,/aP),=, - b,c2] + O(c) (28)
62

or, using an additional evaluation at s = -c,

cp = (I/2e3)[(aF,/aP),,, -(aF,/ap),=_,] + W) (29)

The temperature derivatives of b and c are found similarly.

In conclusion, note that all of the quantities needed for a single
Newton-Raphson iteration can be evaluated from a calculation of the
partial derivatives of the fugacity coefficients with respect to composition,
temperature and pressure at the mixture composition z, supplemented by
one (or, if desired, two) additional calculations of the fugacity coefficients
and their partial derivatives with respect to temperature and pressure at a
perturbed composition. The additional calculation using central differences
may well be advisable, since very small values of c will result in excessive
round-off errors (e.g., in the division by r3). On the other hand, only
approximate values of the pressure and temperature derivatives of b and c
are needed. The parameter 6 is found exactly, and c can be found from eqn.
(9) with an error of c2, using only a single fugacity-coefficient calculation. In
the author’s opinion, central differences do provide more safety, and t = 10e3
(on a 16-digit machine) is a reasonable choice.
The algebraic overhead is very modest. The inverse iteration for X,i, and
u requires a single triangular decomposition of B, and the triangular factors
can be utilized for the solution of eqn. (23).

LOCAL STABILITY OF THE CRITICAL POINT

A Taylor-series expansion of F in the deviation variables X including

terms up to fourth order gives

, i.j r.j,k

+(1/24) c Dljk,X,XjXkX,+ . . (30)

1.J.k.i

where Ai = (aF/dXi)x,o=
0,Bij=(a2F/aX,aXj),=,
(our previous B), etc.
X may be expanded as

X=su+s2w+s3q+ ... (31)

with u’u = 1, and where w and q are yet undetermined vectors which are
orthogonal to u, i.e.,

urw = urq = 0 (32)

 63

Substituting eqn. (31) in the expansion and combining equal powers of s

yields

1[ +s3 ~Bijwiuj+(1/6)
1.J
c
r.J.k
Cijkuiujuk
1
CBtj(qzuj+ w,wj/2) +(1/2) c ‘,,kwlUjUk
i.j r.j.k

+ (i/24) c
i. j.k.1
D,Jk~u,u~ukul (33)
1
Selecting u to minimize the coefficient of s2 yields the previous result,
Bu = hmi,,u. Furthermore, the (i, j) cross-products are zero, i.e.,

CBtJWtui
= CBiJqr
UJ = 0 (34)

since C,Biiuj = Ati,,ui, and II is orthogonal to w and q. The terms in eqn.

(33) which contain only elements of u are identical to the coefficients b, c
and d defined earlier. Finally, let r;. = (1/2)x C,jk~i~k. Then
Jk

F= bs2 + cs3 + s4 (1/2)~Bi,w,wj + cr,w, + d (35)

r.j i 1
w is selected to minimize the coefficient of s4, subject to the constraint
uTw = 0. This coefficient is quadratic in w, and provided that B has at most a
single nonpositive eigenvalue (A,,,), there exists a unique minimizer given
(Fletcher, 1981, Chap. 10) by
Bw* + [r - (u’c>u] = 0, u*w* = 0 (36)
The elements of r can be evaluated from quantities already calculated. From
eqn. (30) where A, = 0,
aFlax,= ~~,~~,+(1~2)~c~~,x,x, + ... (37)
j i,k
or, using X = cu,
(aF/axJ, = drninUi + Ai + o(~‘) (38)
that is,
ri=(1/e2)[u,g,(s=e)-Aminui] +0(e) (39)
or

r,= (1/2e2)ui[gi(s=<)-gi(s= --E)] +O(r*) (40)

These minimizations of first the s* coefficient and subsequently the s4
64

coefficient reduce the tangent-plane distance to a function of the single

distance parameter s, i.e.,
F(s)=bs*+C.r3+d*S4+ ... (41)
with

d* = d + (1,‘2)(~*)~r (42)
At the critical point b = c = 0, and the critical point is locally stable
provided that d* > 0. If d* equals zero, higher-order terms must be evaluated
to determine whether or not the system is stable. Stability when d* = 0
requires that the coefficient of s’, which depends on u and w* only, must
equal zero, and that the minimal value of the coefficient of s6 is at least
nonnegative. This coefficient depends on q, and its minimal value is found
by solving a set of linear equations similar to eqn. (36) for q. Global stability
requires that F(Y) be nonnegative for arbitrary Y, and the question of global
stability cannot be resolved by expansions from Y = z. It is worthwhile
noticing that d* < d, where the equality sign holds only provided that r is
parallel to u (in which case w* equals 0). Positiveness of d is therefore not a
sufficient condition for local stability.

PHASE BOUNDARY IN THE CRITICAL REGION

In the immediate vicinity of a locally stable critical point, b and c are both
small relative to d*. At a point (T, P) = (T, + AT, P, + AP) on the phase
boundary close to the critical point, the coefficients b and c must have values
such that F(s) has a minimum at, say s = 6, with F( 8) = 0. This requires that
(dF/ds),,s = 6(2b + 3~8 + 4d*S2) = 0 (43)
and
F(6) = S2( b + CS + d*6*) = 0 (4)
from which,
c= -2d*6
b=d*S2 (45)
In addition, for b and c,
b(T, + AT, PC+ AP) = b(T,, PC) + b,AT+ b,AP = b,AT+ b,AP (46)
c(T, + AT, PC+ AP) = c(T,, P,)+c,AT+ c,AP = c,AT+ c,AP (47)
Combination of eqns. (45)-(47) then yields
AT = (~~8
AP = a,8 (48)
 65

with

01~= 2d*b,/( b,c, - b,c,)

lxp = - 2d*b,/( b,c, - b,c,) (49)

for 6 sufficiently small. The composition of the equilibrium phase is given by

y = zi -I- v,lQ

Of

In Ki = ln( y/z,) = SU,/V, (50)

Equations (48)-(50) provide a parametric relation between first-order varia-

tions in temperature, pressure and composition along the phase boundary in
the vicinity of the critical point.
It can finally be mentioned that alternative transformations of the inde-
pendent variables lead to identical results. Michelsen (1982a) used

yi = (v, -!- xi/2)2 (51)

and another possibility is

y = zi exp( X,/v,) (52)

Rather than working with the modified tangent-plane distance, the original
tangent-plane distance

f(Y) =C.Yj[ln.Yi + ln @i(y) - ln Zi- ln +ilz>] = F(Y) - 1 + C_Yj (53)

I i

can also be used. The modifications required, which of course do not affect
the constants of eqn. (41), are described in Appendix B.
To illustrate the nature of tangent-plane-distance surfaces in the critical
region, a contour plot of f(y) is given in Fig. l(a) for a mixture containing
70% CH,, 15% CO, and 15% H,S at T = 225.02 K, P = 70.49 atm., using the
Soave-Redlich-Kwong equation of state (Soave, 1972). This corresponds to
a point very close to the phase boundary, the nontrivial minimum having a
value close to zero. The contours are replotted in Fig. l(b) using “eigenvec-
tor coordinates” as described in Appendix B, where s1 corresponds to the
distance s along the eigenvector u. The curve X = su + s*w* is also shown in
Fig. l(b). Notice that this curve passes the saddle point and the nontrivial
minimum at very close distances, while the variation of f along the s1 axis
just barely reveals the presence of an additional minimum. The critical point
for this mixture is at T = 232.15 K, P = 77.81 atm.
 (A)

Fig. 1. Tangent-plane distance for a ternary mixture of CH,, CO, and H,S for a point close
to the phase boundary: (a) contour plot of f(y); (b) plot in “eigenvector coordinates”
(Appendix B).
 67

STABILITY ANALYSIS

The stability investigation used by Michelsen (1982a) for nearly critical

conditions closely resembles the approach described here. In particular, the
suggestion of supplementing the search in the primary direction (along u) by
a correction step in an orthogonal direction corresponds to the present
calculation of w*. A “stability indicator” can be deduced from eqns. (47)
and (48). The existence of a nontrivial minimum requires that

9c2 - 32bd* a 0 (54

and a negative value at the minimum is found provided that

c= - 4bdx > 0 (55)

Local stability can thus be expected (for positive b and d*) if eqn. (55) is not
violated. This result is stronger than the usual criterion, b > 0, but it is only
approximate, since it requires that higher-order terms are negligible.

CRITICAL-REGION PHASE BEHAVIOUR IN THE TWO-PHASE REGION

Inside the phase boundary, two phases of compositions y(l) and y”’ and in
amounts of /3 and (1 - p) moles per mole of feed will form such that the
total Gibbs energy of the system is at a minimum. Michelsen (1982a, eqn. 8)
showed that the change in Gibbs energy is related to the tangent-plane
distances at compositions y(r) and yC2)by

AG/RT= pf(y”‘) + (1 - P)f(y”‘)

= #OF(y”‘) + (1 - /3)F(y’2’) (56)

We define a composition-difference vector by

A=y (1)_ YCO, CA,=0

(57)
I
yielding

Y(‘)=z+(l -/3)A

Y(=)= z - PA (58)
to satisfy the material-balance constraints. The objective is to select /? and A
such that eqn. (56) is minimized.
Again we select new variables Xi = A;/ui corresponding to the transforma-
tion used in eqn. (4). In the critical region the concentration difference will
68

be small, and F(y “)) and F(yC2’) can be expanded in power series in X from
X = 0, in analogy with eqn. (30):

~=(1/2)8(‘-P)CB,jX,X,+(‘/6)P(1-8)(I-28) C CjkX,XjXk
1.j i.j.k
+ (l/24)/3(1 - @)(l - 3/3+ 3p2) c Dijk,X;XjXkX,+ . . . (59)
i.j.k.1

The Xi must now satisfy xuiXi = 0. X may be expanded in a power series in

the distance parameter s, as
x=su+s2w+ . . (60)
with
UrU=1,UrW=O,Urv=WTv=O (61)
in order to satisfy the summation constraint for X. Substituting eqn. (60)
into eqn. (59) and collecting terms in equal powers of s yields
AG/RT=s~K,(~~,~)+s~K~(~~,~~+s~K~(/~,u,w)+ ... (62)
Next, u is selected to minimize ~~ and subsequently w to minimize K~. The
development follows closely that for the stability analysis, the essential dif-
ferences being the presence of the parameter /3 and the need to satisfy the
summation constraint. As shown in Appendix B, this constraint is without
notable influence on the final result given below. It is again found that u is
the eigenvector of B corresponding to Xminr and that w can be taken as
w = (1 - 2p)w* (63)
where w* is the solution of eqn. (36).
Substituting u, w and the coefficients b, c, d and d* from the stability
analysis finally yields

AG/RT= /3(1- ,8){ bs2 + ~(1 - 2p)s3

+~“[(1-2/3)~(d*-d)+(l-3/?+3/3~)d])+ ... (64)

that is, an expression which involves only s and /I as independent variables.
These are next determined from the requirement that AG is at a minimum, or

(65)
Assume that the minimum is found at s = 6. Then, the above conditions
(eqn. (65)) yield
c = -2d*(l- 2/?)6 (66)
b = [ d*(l - 2p)’ - 2dP(l- /3)]S2 (67)
 69

Combining these relations with the temperature and pressure derivatives of b

and c at the critical point (eqns. (46) and (47)), the temperature and pressure
variations are obtained as
AT= T- T, = ‘~~(1 - 2b)6 (68)
AP=P-P,=c+(l--228)s (69)
together with eqn. (50) for the composition variation.
For the phase line p = 0.5, AT and AP become proportional to 62 when
the critical point is approached. The proportionality constants can be
derived, including the s5 term in the expansion represented by eqn. (4), the
result being

AT
i AP !
j3=0.5
(70)

with

e = (1/120)(d5F,/ds5) + (l/2) c D,jk,~~~j~k~, (71)

i,j.k.l
Finally, eqns. (66) and (67) can be used to calculate approximate values of
the phase split for a point (T, P) in the two-phase region and close to the
critical point. Evaluating b, c, d and d* at (T, P, z), elimination of 6 from
eqns. (66) and (67) yields

(1 - z/3-‘= 1 + 2(d*/d)(l+ 4bd*/c2) (72)

Subsequently, 8 can be calculated from eqn. (66) and the K factors from eqn.
(50).
Note that, in accordance with the stability criterion given as eqn. (57), a
value of p in the range 0 < /3 < 1 requires that c2 - 4bd* > 0. The results for
the two-phase region are thus fully consistent with the results for the phase
boundary.

NUMERICAL EXAMPLE

The first-order approximations derived in this work are here compared

with the results of direct phase equilibrium calculations for a seven-compo-
nent natural-gas mixture described by Michelsen (1982a). The relevant
thermodynamic properties are evaluated from the Soave-Redlich-Kwong
equation of state (Soave, 1972). The mixture composition and the elements
u/u; describing the K-factor variation at the critical point (T, = 203.12 K,
P, = 58.11 atm.) are given in Table 1. The relevant expansion coefficients at
the critical point are listed in Table 2.
70

TABLE 1
Composition and principal eigenvector u at critical point for a seven-component mixture

Component Mole fraction u;z,- “2

C, 0.9430 -0.16
C* 0.0270 2.66
C, 0.0074 4.79
n-c, 0.0049 6.87
n -c, 0.0027 8.99
n -c, 0.0010 11.04
N* 0.0140 - 1.78

TABLE 2
Expansion coefficients and their temperature and pressure derivatives at the critical point for
mixture given in Table 1

Coefficient Value
4 - 0.0618
b, 0.0504
CT 0.1537
CP - 0.2105
d 2.058
d* 0.943

TABLE 3
Predicted and calculated temperature and pressure variations along the p = 0.5 line for
mixture given in Table 1

6 CT- T,)/S’
-. IP-R)/S2
_..
0.10 - 38.6 - 66.7
0.05 - 37.3 -65.9
0.02 - 36.8 - 65.6

Predicted - 37.0 - 65.7

TABLE 4
Predicted and calculated phase distributions in the critical region for mixture given in Table 1

Specified vahtes Predicted from eqn. (71)

B 6 B 6
0.20 0.05 0.30 0.039
0.20 0.02 0.24 0.017
0.20 0.01 0.22 0.0093
0.20 0.005 0.21 0.0048
0.5 0.05 0.39 0.054
0.5 0.02 0.46 0.0202
0.5 0.01 0.48 0.010
0.5 0.005 0.49 0.005
 71

,T-Tc (K)
-- 1.

Fig. 2. T- T, versus 6 for seven-component mixture given in Table 1: full lines, calculated
results; dotted lines, results predicted using eqn. (48).

Figure 2 shows T - T, as a function of the distance parameter S (caicu-

lated as S = &uivi ln Ki) for phase fractions p = 0 and B = 0.2. The agree-
ment with eqn. (48) (dotted line) is perfect near 6 = 0 and quite good over a
fairly wide range. Table 3 shows values of T - T, and P - PCas a function of
S along the line /I = 0.5. The agreement with the prediction is excellent.
Finally, a comparison of the phase splits predicted by eqn. (72) with
actual phase splits at the specified conditions is given in Table 4 for a
number of points in the two-phase region. Although the predictions are
qualitatively correct, substantial deviations are noted unless 6 is very small.

CONCLUSION AND DISCUSSION

The present procedure for the calculation of critical points and of the
phase boundary in the vicinity of the critical point is of particular use in
connection with complete phase-boundary calculations as described by
Michelsen (1980). This procedure for constructing the phase boundary is
capable of passing the critical point, but numerical problems can be severe in
12

the vicinity of the latter, since the determinant of the Jacobian matrix is
inversely proportional to the fourth power of the distance from the critical
point.
Combination of Michelsen’s marching procedure with the present direct
calculation of the critical point and the associated derivatives required for
phase-envelope construction obviates the need to calculate points on the
phase boundary close to the critical point. The present procedure also allows
calculation of the phase boundary starting from a critical point. Finally, the
procedure provides a first estimate of phase composition and phase amounts
for flash calculations at specified T and P in the critical region, based on
properties derived for the feed composition only.
For stand-alone calculations of critical points, good initial estimates of the
critical temperature and pressure are needed. A procedure similar to that
described here using temperature and total volume as the independent
variables and based on a tangent plane in Helmholtz energy is easily derived,
and preliminary investigations indicate that such a procedure combines the
economy of the present method with the reliability and insensitivity to initial
estimates of the method of Heidemann and Khalil.

ACKNOWLEDGEMENT

The author thanks Prof. Warren D. Seider, University of Pennsylvania,

for many suggestions and helpful discussions.

LIST OF SYMBOLS

a, b, c, d, e expansion coefficients for tangent-plane distance

A, B, C, D coefficients of matrices of partial derivatives for tangent-plane
distance
matrix of second derivatives
original tangent-plane distance
F modified tangent-plane distance
g
hG, G
gradient vector (eqn. (2))
molar Gibbs energy
h
i,j,
4
k, f
solution vector (eqn. (24))
component indices
Equilibrium factor for component i
mole number
lb number of components in mixture
P, AP pressure, pressure difference
P, critical pressure
q expansion vector (eqn. (33))
 73

Q scalar multiplier (eqn. (24))

r vector defined in eqn. (35)
R gas constant
s distance parameter
T, AT temperature, temperature difference
U eigenvector of B
V vector with ith element zf”
w, w* expansion vectors (eqns. (31) and (36))
X vector of deviation variables
(1) phase mole fractions
Y 3 YC2’
Y trial phase mole numbers
Z vector of feed mole fractions
ffT’ aP coefficients of temperature and pressure variation (eqn. (49))
P phase fraction
Y vector defined in eqn. (23)
s distance parameter (value of s at minimum)
f increment for numerical differentiation
3 K4 coefficients of Gibbs-energy expansion
y;3 eigenvalue of B
9i
fugacity coefficient for component i

Superscripts

T vector transpose

Subscripts

P partial derivative with respect to pressure

T partial derivative with respect to temperature

APPENDIX A: NUMERICAL SOLUTION OF HOMOGENEOUS ALGEBRAIC EQUA-

TIONS

This appendix describes briefly the solution procedure for a set of N linear
algebraic equations of the form
Bx=a (Al)
where the matrix of coefficients B and the augmented matrix (B, a) are both
of rank N - 1, the objective being to determine a particular solution, x = h.
Such equations are encountered in the calculation of the eigenvector deriva-
tives up and uT (eqn. (23)) and of the correction vector w* (eqn. (36)).
Since the N equations are known to be linearly dependent, a simple
74

procedure is simply to take x,,, = 0 and delete the last equation from the set.
This yields the reduced system
wx’ = a’ (A2)
where B’ is the upper (N - 1). (N - 1) submatrix of B, and a’ contains the
first N - 1 elements of a. Provided that a triangular decomposition of B is
available (B = LU with L unit lower triangular), an even simpler procedure is
available, where
Ux=L-‘a=b (A31
where (in exact arithmetic) U,, and b, are both equal to zero. Taking xN
equal to zero results in a reduced system where the matrix of coefficients U’
is triangular.
In solving eqn. (23),
(B - XminI)up = Y (A41
the triangular decomposition of B is available from the eigenvalue calcula-
tion, while exact evaluation of up would require the triangular factors of
B - h,,I. To avoid renewed factorization, the small term X,i,I in eqn. (A4)
is neglected, the equation to be solved becoming
Bu,=v (As)
or
uu, = L- lY (A@
where the procedure above (taking (u~)~ = 0) is used even though U,, and
the last element of L-‘V are both nonzero. It can be shown that the
approximations introduced in this calculation of up (and ur) do not degrade
the performance of the Newton-Raphson iteration.

APPENDIX B: RELATION BETWEEN MODIFIED AND ORIGINAL TANGENT-

PLANE CRITERIA

The advantage of using the modified tangent-plane criterion (eqn. (1)) is

the convenience that the composition variables q can be treated as fully
independent. The original tanget-plane criterion (eqn. (53)) requires that the
y, satisfy the constraint C,J+ = 1. In terms of the new independent variables
Xi, this requirement yields
x’v=o @I)
with u = zl/*. The transformation to X as the vector of independent varia-
bles consisis of a translation of the coordinate system followed by a scaling
such that contours of constant F close to X = 0 become hyperspheres for an
ideal mixture.
 15

We shall in addition perform a rotation of the coordinate system using an

orthogonal transformation
x = us (B2)
where U is the matrix of eigenvectors of B,
UTBU = A, UT = U-’ (B3)
with A being the diagonal eigenvalue matrix of B. Discarding terms of the
third and higher orders in X, the contours are now given by

F(X) = (l/2)XTBx = (l/2)sTUTBUs = (1/2)sTAs

or

(B4)

i.e., contours of constant F become hyperellipsoids in the s coordinate

system provided that all Xi are positive.
Let the ith eigenvector of B, that is, the ith column of U, be u,. Of the N
eigenvectors, two are of particular interest. The first is the eigenvector u1
corresponding to the smallest eigenvalue A, = X,, (that is, our previous u),
and the second is the eigenvector uN = v with eigenvalue A, = 1. That v is an
eigenvector with an eigenvalue of 1 follows from

or Bv = v. From the orthogonality of eigenvectors for a symmetric matrix,

vTui = 0, i # N (JW
and hence vectors X which are selected as linear combinations of the first
N - 1 columns of u all satisfy eqn. (Bl). The function F,(s) uses X = su, and
hence is identical for the modified and original criteria.
A minor difference is noted in the minimization of the coefficient of s4 in
eqn. (35), where the minimizer for the original tangent-plane criterion, which
we shall call %, must satisfy the additional constraint vTfi = 0. Straightfor-
ward but fairly extensive derivations reveal that
c%=w*-bv 036)
and that the corresponding coefficient d is given by
cj=d*+b2/2 037)
At the critical point b equals zero, and within the region where the present
expansions are valid, b is small. The differences between the modified and
original tangent-plane criteria are thus without significance for the results
given here.

REFERENCES

Fletcher, R., 1981. Practical methods of Optimization. Vol. 2. Constrained Optimization.

Wiley, New York.
Heidemann, R.A. and Khalil, A.M., 1980. The calculation of critical points. Am. Inst. Chem.
Eng. J., 25: 769-779.
Michelsen, M.L., 1980. Calculation of pase envelopes and critical points for multicomponent
mixtures. Fluid Phase Equilibria, 4: l-10.
Michelsen, M.L., 1982a. The isothermal flash problem. I. Stability analysis. Fluid Phase
Fquilibria, 8: l-19.
Michelsen, M.L., 1982b. The isothermal flash problem. II. Phase-split calculation. Fluid
Phase Equilibria, 8: 21-40.
Soave, G., 1972. Equilibrium constants from a modified Redlich-Kwong equation of state.
Chem. Eng. Sci., 27: 1197-1203.
Wilkinson, J.H., 1965. The Algebraic Eigenvalue Problem. Clarendon, Oxford.