Accelerated Successive Substitution Schemes for Bubble-Point

and Dew-Point Calculations

DING- YU PENG

Department of Chemical Engineering, University of Saskatchewan, Saskatoon, Saskatchewan, Canada S7N OW0
The accelerated successive substitution (SS) algorithms developed by Mehra et al. for flash calculations have been
extended to the prediction of saturation points. A transformation matrix which is used to calculate the acceleration parameter
has been rewritten in a form that is applicable at the saturation conditions. Simple equations for estimating the initial
values and recursive formulae according to which the iterates can be updated are presented. The proposed schemes
were compared with the conventional SS method and a multivariate Newton’s method. The comparison suggests that
the accelerated SS schemes are more tolerant of poor initial values and sometimes more efficient than Newton’s method.

Les algorithmes de substitution successive accC16rCe (SS) mis au point par Mehra et al. pour les calculs de flash ont
ttC Ctendus a la praiction des points de saturation. Une rnatrice de transformation qui est utilisCe pour calculer le paramhe
d’accklkration a CtC rekcrite sous une forme qui peut s’appliquer aux conditions de saturation. On prCsente des equations
simples pour l’estimation des valeurs initiales et des formules rCcursives selon lesquelles les itkrations peuvent &re
mises a jour. Les schemas proposCs ont CtC cornparks B la mCthode SS traditionnelle et une mCthode de Newton mul-
tivariee. La comparaison suggkre que les schkmds SS accklCrCs tolkrent mieux les rnauvaises valeurs initiales et qu’ils
sont parfois plus efficaces que la mCthode de Newton.

Keywords: phase equilibria, accelerated successive substitution, bubble point, dew point.

P hase equilibrium calculations form an important part of

the process design operations in the hydrocarbon and
petroleum industry. In addition to the class of methods which
(1983). However, the initial K-values used by Nghiem et al.
(1985) were generated from a stability test calculation.
Michelsen (1985) has also proposed the use of stability-based
use Gibbs free energy minimization formulations, which are procedure for calculating the approximate values of satura-
generally applied to combined phase and chemical equilibria, tion temperatures or pressures. These stability-based methods
there are largely two classes of methods for phase equilibrium may show their greatest advantage in the critical region, but
calculations: the successive substitution (SS) method and they are not always economical to use in regions removed
Newton’s method. The SS method is simple to use but con- from the critical point. A simpler initialization scheme is
verges very slowly near the critical points of the mixtures. desirable. Moreover, in view of the success of the acceler-
However, the improved convergence of Newton’s method ated SS schemes in flash calculations, it is worthwhile to
is achieved at the expense of the increased function evalua- study the performances of the other two accelerated SS
tions per iteration. The function evaluations are necessary schemes of Mehra et al. (1983) in bubble-point and dew-
for the computation of the Jacobian. point calculations. This paper reports the findings of such
During the past decade or so, a number of researchers (for a study and presents a set of empirical equations for cal-
example, Risnes et al., 1981; Michelsen, 1982; Nghiem et culating the initial values.
al., 1983; Mehra et al., 1983;Nghiem and Heidemann, 1982;
Rijkers and Heidemann, 1986; Gupta et al., 1988) have pro- Basic equations
posed acceleration algorithms to promote the convergence
of the successive substitution method for flash calculations. The condition for vapor-liquid equilibrium (VLE) is given
Generally, the acceleration schemes retain the simplicity of by
the SS method but require substantially less computing effort
in comparison to the SS method. In particular, Mehra et al. f;(T,P,x,,...,xN-l)=fiv(r,P,YI , . . . ,Y N - I ) ,
(1983) have shown that there exists a rigorous relation
between the conventional successive substitution method for (1 5 i 5 N ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1)
flash calculation and the gradient method for Gibbs free
energy minimization. From the material balance relations, the mole fractions of
In spite of the progress made in the convergence promo- the components in the equilibrating phases are given by
tion of the SS method for flash calculations, it appears that
the accelerated SS method has not been extended to the xi = zi / [ 1 + (K; - 1 ) a ] . . . . . . . . . . . . . . . . . . . (2)
prediction of saturation pressures and temperatures. In the
open literature, the main algorithms used in the calculation yj = KiXj .................................. (3)
of the saturation points are still the conventional SS method
and Newton’s method. Nghiem et al. (1985) have recently where the vapor fraction a is obtained from the well-known
proposed two new Newton formulations and an accelerated Rachford and Rice (1952) equation
SS algorithm for calculating the bubble points and dew points.
These authors have also presented an initialization scheme C(K; - 1)z; / [ l + (K; - l ) a ] = 0 . . . . . . . . . . (4)
which leads to convergence to all saturation points for their
test mixture. Successive substitution method
The equation used by Nghiem et al. (1985) in their acceler-
ated SS algorithm to calculate the acceleration parameter is In the conventional SS method, a set of K-values is assumed
essentially the same as that of Algorithm 3 of Mehra et al. and Equation (4) is solved for a . The mole fractions obtained

978 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING, VOLUME 69, AUGUST, 1991
from Equations (2) and (3) are then used to calculate the where
component fugacities, which are dependent on the tempera-
ture, the pressure, and the composition. The K-values are 6~ = T(")/(5.373 C (1 + w i ) T,
revised and new values of the mole fractions are obtained.
The iterative process continues until Equation (1) is satisfied. Dew point temperature prediction:
The initial K-values are usually calculated from some
T ( " + ' )= T(")(1 - fD {1 - C z, l K / " + I ) ]) ) . . (13)
simple empirical equations. For example, Wilson (1968) pro-
posed the following equation for estimating the initial
I(-values for the hydrocarbon systems where

K p = Ps, lP'O' . . . . . . . . . . . . . . . . . . . . . . . . . . . . (5)

where Similar to the situation in which two dew-point pressures
correspond to a given temperature, retrograde behavior may
pS, = P,, exp 15.373 (1 + 0,) (I - T,, /T"')] . (5a) also involve two dew-point temperatures at a given pressure.
The algorithm of correcting the K-values is based on the In this case the iteration scheme given by Equations (13) and
(14) tends to yield the higher-temperature dew point. The
ratios of the fugacities of the components and has the form
lower-temperature dew point may be obtained if the eD cal-
K,("+')= K,'").r,'"), (1 I i IN ) . . . . . . . . . (6) culated from Equation (14) is multiplied by - l . Similar treat-
ment may be applied to f E if retrograde behavior extends to
where the bubble-point region.
It should be noted that in the bubble-point and dew-point
r, = j .lI / f iv . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (7) calculations one does not solve Equation (4) for the value
Prausnitz et al. (1967) were among the first researchers to of a . In such cases the composition of the finite phase is set
have used this algorithm in their work. Mehra et al. (1983) equal to that of the feed. The mole fractions of the compo-
have shown that Equation (6) can be derived from an appli- nents in the incipient phase are calculated according to
cation of the gradient method to the minimization of the Gibbs
free energy at constant T and P. y!") = ziKY'IC zi K,@) . . . . . . . . . . . . . . . . . . . . . (15)
for bubble-point predictions and
Bubble point and dew point calculations
x!"' = zi / K?) I C [zil K f " ] . . . . . . . . . . . . . . . (16)
In bubble-point and dew-point calculations, in addition to
the composition of the incipient phase, either the tempera- for dew-point predictions, respectively.
ture or the pressure is also unknown. An algorithm is needed
to guide the trial-and-error process of estimating this Initial pressure and initial temperature
unknown. Although interpolation/extrapolationtechniques
may be used, it is best to have some ideas of the dependency In bubble-point and dew-point predictions it is necessary
of the K-values on the iterates so that simple yet robust recur- to provide an initial value for either the pressure or the tem-
rence formulae may be developed. perature which is to be predicted. Due to the numerous differ-
The assumption that the K-values are inversely proportional ences among the real systems, it is very unlikely that a
to the pressure at constant temperature, such as indicated by non-iterative procedure can be developed to generate the ini-
Equation (5), leads to the following equations which are tial values that will lead to convergence for all kinds of mix-
generally satisfactory for the hydrocarbon systems: tures. However, Equation (5) may be used to yield the
following empirical equations which, in general, are satis-
Bubble point pressure prediction:
factory for hydrocarbon systems over a wide range of
P ( " + ' J= P ( r f )[2 - 1/C zi K!"+l'] . . . . . . . . . . . . (8) conditions:

Dew point pressure prediction: Bubble point prediction:

P("+')
= P'"'/C [Zj/K!"+l)] . . . . . . . . . . . . . . . . . . (9) P(O) - c xi pi = 0 ......................... (17)
Because the controlling K-values in dew point calculations Dew point prediction:
are those of the less-volatile components and because these
K-values increase with an increase in the pressure in the retro- PcO' C y; / p i - IIp:' = 0 . . . . . . . . . . . . . . . . . . (18)
grade region, an equation for the calculation of the retro-
grade dew-point pressure takes the form where
P("+l' = P c n J( 2 - 1 / C [zi/ K ( " + ' ) ] . . . . . . . . (10) p i = Psi when T(O) I T,.; . . . . . . . . . . . . . . . . (1 9a)
Empirical equations can also be obtained for the tempera-
ture predictions. Assuming Equation ( 5 ) provides the cor- = (Psi Pc;)0,5 when T'O' > T,, . . . . . . . . . (1 9b)
rect temperature dependency of the K-values, we obtain the
following recurrence relations: Accelerated successive substitution

Bubble point temperature prediction: Accelerated SS algorithms are generally characterized by

the use of the suitably calculated step length X such that the
K-values are updated according to the equation

THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING, VOLUME 69, AUGUST, 1991 979
 ) K!") . [r?)]',
K ! ~ + '= (1 ~i IN) . . . . . . . (20) TABLE1
Compositions of the Test Mixtures
For example, Risnes et al. (1981) have assumed that the r-
values given by Equation (7) approach monotone after a finite Mole Fraction
number of iterations and proposed that the step length h be Component Mixture 1 Mixture 2 Mixture 3
estimated from the ratio of the logarithms of the ri between
two successive iterations. These authors also stated that a Nitrogen 0.0140
constant value of 2 may be used for X after ten iterations. Methane 0.9430 0.6436
In their study of the relationship between the S S algorithm Ethane 0.39842 0.0270 0.0752
described by Equation (6) and the gradient method for Gibbs Propane 0.29313 0.0074 0.0474
free energy minimization at constant T and P, Mehra et al. n-Butane 0.20006 0.0049 0.0412
( 1983) have presented three different algorithms for calculat- n-Pentane 0.07143 0.0010 0.0297
ing the optimal or near-optimal step lengths. Specifically, n-Hexdne 0.03696 0.0027 0.0138
n-Heptane 0.0303
their Algorithm 1 has the form n-Octane 0.0371
n-Nonane 0.0415
n-Decane 0.0402
(g(n-l) - g ( 9 1 1 X ( n - 1 ) .................... (21)
for bubble-point calculations and
where the elements of the matrix U-' are given by
w;, = X ; ~j + 6 i j ~ i[ C ( X ; / Z ~-) 11 . . . . . . . . . . (29)
u.71
1, -
- . . . . . . . . . . . . . . . . . . . (22)
(an,!'/aln K,)T,P,K
for dew-point calculations, respectively.
The acceleration is initiated after the initial guess and one The three formulae developed by Mehra et al. (1983), with
ss step; i.e., A"' = 1. the formula for Algorithm 1 modified as outlined above, were
Mehra et al. (1983) and Rijkers and Heidemann (1986) applied to the calculation of bubble points and dew points.
have shown that The schemes are referred to as Accl, Acc2, and Acc3, res-
pectively. The accelerated S S scheme proposed by Rijkers
Ui,Tl = [LV/(L + v)] (XiYIlZi) [tiij+ X j Y j / ( Z j S ) ] and Heidemann (1986) is similar to the Algorithm 1 of Mehra
et al. (1983) but requires more computational effort. It was
....................................... (23) not tested in this study.
where S is given by
Numerical examples
S = 1 - C ......................
( ~ k ~ k / ~ k ) (24)
The features of the acceleration schemes and those of the
Because the leading product term LV vanishes at the empirical equations developed in this study can be illustra-
bubble-point and the dew-point conditions, it appears that ted using three hydrocarbon mixtures. The compositions of
Equation (23) has limited the application of Algorithm 1 to the test mixtures are given in Table 1. The first system is
only those cases where the amounts of liquid and vapor are a five-component mixture of n-alkanes. The experimental
both nontrivial. One also may erroneously conclude that data for this system were first reported by Etter and Kay
u;,: * vanishes when either V or L vanishes. However, it can (1961). The second mixture is a typical natural gas system
be shown that (Gonzalez and Lee, 1968). The third is a volatile oil (Nghiem
et al., 1983). The Peng-Robinson equation of state (Peng
LV/(L + v)* = S/C [(Yk - Xk)Z/Zk] ........ and Robinson, 1976a) was used to model all three mixtures.
and that Equation (23) may therefore be rewritten as The calculated phase envelopes are plotted in Figure 1 to
show the characteristically different two-phase regions of the
u,,-' = L y, y,/ [C ( Y Z l Z k ) - 11 . . . . . . . . . . . . (26) three mixtures.
A multivariate Newton's method (Peng and Robinson,
at the bubble-point conditions and as 1976b, 1977) was also used in this study to make the calcu-
lations. The Jacobian matrices were calculated from analy-
u,J' = V X , x,/ [ E ( X ~ / Z , ) - 11 . . . . . . . . . . . . . (27) tical expressions for the derivatives of the fugacity, and
Crout's LU decomposition algorithm was used to calculate
at the dew-point conditions, respectively. Consequently, the the correction vector, To guard against excessive number
Algorithm 1 of Mehra et al. (1983) is also valid when the of occurrences of early divergence, which may be due to
specified temperature and pressure for a flash calculation hap- poor initial values, one S S step was always applied before
pen to be exactly the same as the conditions of either a bub- Newton's method was used. This iteration is included in the
ble point or a dew point.
counting of the number of iterations for Newton's method.
Unfortunately, the performance of the accelerated bubble- The convergence criterion used in all schemes was
point and dew-point calculations using Equation (2 1) with
the u,;' given by Equations (26) and (27) was found to be gTg < 10-l0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (30)
inferior to that of the same calculations using the other two
algorithms of Mehra et al. (1983). However, an empirically which is the same as that used by Mehra et al. (1983).
obtained matrix W may be substituted for the matrix U-'
to improve the convergence behavior. The elements of the Mixture 1
matrix W are given by
As shown in Figure 1, the critical point of this mixture
w,, = y, y, + A,, y1 [ E ( Y ; / Z ~ ) - 11 . . . . . . . . . . (28) is almost at the extreme temperature and pressure of the
980 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING, VOLUME 69, AUGUST, 1991
 25 -
MIXTURE 1
CRITICAL POINT P = 3 MPa
0- T(O)= 326 K

-2- ----
____-_-_______
ss
n Acc 1
I-
CT, -+ - - - - - Acc3
W m Newton
0
6 -6-
t i 0

rY
n
U
1 MlXTURE 2
"r\ MIXTURE 1
0
-
-8-
\
\
\
\
\

4
\

5j ;j,n
\
-10- \

-12 I I 1
0100 180 340 450 5cio
T E M ~ ~ R A T U R EK, NUMBER OF ITERATIONS
Figure I - Phase envelopes for the test mixtures. Figure 2 - Iteration performance for Mixture 1 at 3 MPa.

TABLE2 MIXTURE 1
Performances of the Various Schemes for Mixture 1 at Selected P = 3 MPa
Temperatures 326 K
T pl01 p(n) Number of Iterations 331
(K) (M Pa) SS Accl Acc2 Acc3 Newton .
LLI
CK
Bubble-point prediction 3
L
270
280
1.044
1.339
0.925
1.166
6
7
4
4
5
5
5
5
3
4
2
t

329
290 1.690 1.446 8 6 6 6 4 W

I
300 2.101 1.765 9 5 6 6 4 a
310 2.487 2.124 10 6 7 7 4
320 2.804 2.524 11 6 7 9 3 -I 327

Y
330 3.149 2.961 12 7 8 8 4
340 3.523 3.433 12 7 8 10 4
350 3.930 3.931 15 7 8 8 4
9 5
360
370
4.369
4.843
4.443
4.945
22
37
8
9
9
13 14 5
325 ' 3 6
I
i
l
I
12
.I
15
380 5.239 5.384 100 25 19 * 12
NUMBER OF ITERATIONS
Dew-point prediction Figure 3 - Effect of iteration number on an iterated variable for
270 0.189 0.076 4 3 4 4 4 Mixture 1 at 3 MPa.
280 0.189 0.121 5 4 4 4 4
290 0.272 0.185 5 4 4 4 4 temperatures are given in Table 2. As expected, the SS
300 0.381 0.276 6 5 5 5 4 method is quite efficient in the region which is far from the
310 0.691 0.398 7 5 5 5 4 critical point but its performance gradually deteriorates as
320 0.889 0.560 7 5 5 5 4 the specified temperature gets closer and closer to the crit-
330 1.125 0.773 8 5 6 6 4 ical region where the K-values approach unity. It is also evi-
340 1.402 1.047 10 6 5 6 4 dent that all three accelerated SS schemes are faster than the
350 1.722 1.398 11 6 6 6 4 S S method; the improvement is especially noticeable from
360 2.089 1.848 13 7 7 7 4 the results for the critical region.
370 2.503 2.428 15 7 7 7 3 The convergence behavior of a bubble-point temperature
380 2.915 3.199 24 8 9 9 4 prediction is illustrated in Figure 2 which is a plot of error
390 3.362 4.365 70 16 13 15 5 versus iteration number for a pressure of 3 MPa. It is seen
~~

*The calculation diverged. that four iterations were required when Newton's method
was used but seven iterations were required when either the
Accl or the Acc3 was used. However, as shown in Figure 3,
phase envelope. The initial pressures or temperatures the temperature values calculated from the accelerated SS
calculated from Equations (17) and (18) were very close schemes after four iterations were essentially the same as
to the converged values. The initial and converged values the converged value. It can also be seen that eight SS
and the number of iterations required for the prediction iterations were required in order to attain the same level of
of the bubble-point and dew-point pressures at selected precision. For this mixture, the performance of Acc2 was
THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING, VOLUME 69, AUGUST, 1991 981
 TABLE
3
Performances of the Various Schemes for Mixture 2 at Selected

T
(K)
pro,
(MPa)
pm
Temperatures

SS
Number of lterations
Accl Acc2 Acc3 Newton
0 - ss
Accl
Newton
I
I
I
I
I

Bubble-point prediction
W
140 0.672 0.708 6 5 4 4 3 t 2
150 1.074 1.112 7 3 5 6 4 /
160
170
1.626
2.354
1.660
2.374
9
10
6
6
6
7
6
7
4
4
LL
0 Ll
180 3.278 3.273 I1 6 5 5 4 CIL
190 4.414 4.355 28 7 14 14 5 W
200 5.106 5.583 252 37 * 20 10 m
I /
Dew-point prediction 3
210 0.491 0.023 4 3 4 4 5 Z
220 0.759 0.058 5 4 4 4 6
230 1.124 0.133 6 4 5 5 6
240 1.603 0.287 7 5 5 5 7 2
I
250 2.209 0.601 10 6 6 6 8 0 2 4
I I
6 8 D
PRESSURE, MPa
260 2.945 1.292 15 8 8 9 9
265 3.362 2.038 24 10 10 8 9 Figure 4 - Performances of the various schemes in dew-point tem-
Retrograde dew-point prediction
perature predictions for Mixture 2.
210 0.491 6.984 88 36 28 37 *
220 0.759 8.052 43 24 23 29 * When divergence in the prediction of retrograde dew points
230 1.124 8.664 29 18 13 18 * is due to poor initial values, the problem may be somewhat
240 1.603 8.733 22 16 12 15 * alleviated by using the values calculated from Equation (17)
250 2.209 8.259 18 14 12 12 7 as the initial values. As shown in the last row of Table 3
260 2.945 7.023 19 14 12 14 6 for the prediction of the retrograde dew-point at 210 K, the
265 3.362 5.866 28 14 13 12 6 calculation based on Newton’s method converged in eight
210 5.774 6.984 84 25 19 29 8 iterations using initial values based on Pcot = 5.774 MPa.
The performances of the other schemes were also improved
*The calculation diverged. by using this initial value for the pressure.
The capabilities of the SS scheme, the Accl scheme, and
similar to that of Acc3 and was therefore omitted from the Newton’s method in dew-point temperature calculations for
figures. this mixture are shown in Figure 4. The dew points are those
whose temperatures are greater than the temperature cor-
Mixture 2 responding to the cricondenbar. The initial temperatures were
calculated from Equation (18). It is obvious that, as far as
This mixture exhibits retrograde behavior over a very the number of iterations is concerned, Newton’s method was
broad range of pressure and temperature conditions of its the fastest among the three schemes. However, more itera-
VLE region. As methane is the predominant component of tions were required with the Accl scheme (as well as the
this mixture, the bubble-point locus tends to assimilate with other two accelerated SS schemes) than with the conventional
the vapor pressure locus of pure methane. Therefore the ini- SS method over an extended range of pressures at low pres-
tial values calculated from Equation (17), which in general sure conditions.
provides good estimates for vapor pressures, were very close All schemes failed to yield the correct results when the
to the converged bubble-point values over a wide range of initial values based on Equation (18) were used in the predic-
temperatures. Unfortunately, because of the pronounced tion of the lower-value dew-point temperatures of the retro-
retrograde region of this mixture, Equation (18) can not be grade region where there are two dew-point temperatures
expected to yield good initial values over the entire dew-point for each pressure value. The initial values were simply too
region of this mixture. In particular, Equation (18) is meant poor to result in adequate corrections during the early stage
for only the low-pressure dew points. Nevertheless, this of the iterations. Convergence was generally obtained when
equation was used in both normal dew-point and retrograde the cricondenbar temperature was used as the initial temper-
dew-point calculations. ature for all dew-point temperature calculations.
Table 3 shows the initial values, the predicted results, and
the number of iterations required for selected bubble-point Mixture 3
and dew-point calculations. It is interesting to note that for
a number of temperature conditions the Accl scheme was This mixture was first used by Nghiem et al. (1983) to
faster than Newton’s method in the prediction of the normal illustrate the application of Powell’s method in flash
dew-point pressures. Moreover, for a wide range of tem- calculations. The critical pressure and critical temperature
peratures Newton’s method diverged when the initial values calculated in this study are 19.17 MPa and 425.14 K,
calculated from Equation (18) were used to predict the retro- respectively. These values are based on the assumption
grade dew-point pressures. It appears that the accelerated that the binary interaction parameters are all zero and are
SS schemes, which are based on the gradient method, were therefore different from those reported by Nghiem et al.
not so seriously affected by the same poor initial values. (1983).

982 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING, VOLUME 69, AUGUST. 1991
 TABLE4
Performances of the Various Schemes for Mixture 3 at Selected

T
(K)
p‘0,

(MPa)
p(n)
Temperatures

SS
Number of Iterations
Accl Acc2 Acc3 Newton
0 1 0 J:
Z
0
F
5-
I -
- - - ss
Accl
Newton /
/
1

Dew-point prediction
2w 2- /
/

370 0.780 0.127 5 4 4 4 5 t /

/
380 0.967 0.185 6 4 5 5 5
390 1.184 0.264 6 5 5 5 5
400 1 436 0.371 7 5 6 6 6

i t
410 1.722 0.512 8 6 6 6 6
420 2.047 0.699 9 6 6 6 6
430 2.408 0.946 10 6 1 6 6
440 2.807 1.272 12 6 1 7 6
450 3.247 1.708 14 7 7 7 6
460 3.127 2.304 18 8 7 8 6 10
410 4.247 3.164 24 9 9 1 5 5
480 4.800 4.576 35 9 9 9 4 2to 290
1
320 3dO do 4
Retrograde dew-point prediction TEMPERATURE, K
480 4.800 11.928 150 19 23 * 4+
410 4.247 14.134 171 * 20 22 8+ Figure 5 - Performances of the various schemes in bubble-point
460 3 727 15.721 260 43 21 22 * pressure predictions for Mixture 3.
450 3 241 16.975 475 50 32 31 *
440 2.801 17.994 1000 55 32 * *
points are presented in Tables 4 and 5 . The results show that
430 2.408 18.826 1043 * 13 85 * the accelerated SS schemes are generally superior to the con-
+The calculation converged to a normal dew-point. ventional SS scheme. We note again that Newton’s method
*The calculation diverged. not only sometimes required more iterations but also was
less reliable than the accelerated SS schemes.
TABLE5 The convergence behavior of different schemes in the
Performances of the Various Schemes for Mixture 3 at Selected prediction of bubble-point pressures is shown in Figure 5 .
Pressures
The performances of the Acc2 and Acc3 schemes were con-
p ~((1’ T(n) Number of Iterations siderably worse than that of the Accl scheme and, for clarity,
(MPa) (K) SS Act1 Acc2 Acc3 Newton are not shown. The SS method converged for all tempera-
tures up to the critical point, although the number of itera-
Dew-point prediction tions became very large as the specified temperature
3 400 468.4 14 11 10 11 5 approached the critical temperature. Newton’s method, on
6 400 485.3 24 21 13 13 5 the other hand, diverged when the specified temperature was
9 400 486.9 42 18 13 14 8 greater than 413 K; when convergence was obtained, it
12 400 479.7 87 16 23 18 * required between a low of nine iterations and a high of thir-
15 400 464.9 253 30 36 57 * teen iterations for the range of temperatures tested. The per-
18 400 439.9 >2000 55 86 81 * formance of the Accl scheme is seen to be very good for
*The calculation diverged. temperatures below 280 K. However, the performance
became unpredictable when the specified temperature was
The calculated phase envelope indicates that this mixture greater than about 350 K; the number of iterations varied
exhibits both isothermal retrograde behavior at temperatures irregularly with temperature and, in a number of cases, the
above the critical temperature and isobaric retrograde scheme failed.
behavior at temperatures below the critical point. This kind
of complex phase behavior tends to adversely affect the Discussion and conclusions
quality of the description of the pressure and temperature
dependency of the K-values as given by the empirical for- The overall performances of the five schemes are shown
mulae. It also makes the calculation of a single saturation in Table 6, which is a summary of the total CPU time
point in the high pressure region particularly challenging, expended on those iterative calculations for which all
although the construction of the complete VLE phase enve- algorithms were successful. The computation loads in terms
lope can be easily accomplished by using a marching of CPU time per iteration for the saturation-pressure predic-
algorithm such as that proposed by Michelsen (1980). tions are shown in Figure 6 for the SS method, the Accl
Because of the very high critical pressure and critical tem- scheme, and Newton’s method. It can be seen that the com-
perature, the initial values calculated from Equations (1 7) putation load of Newton’s method increases very rapidly as
and ( 18) were substantially different from the converged the number of components increases. The CPU time per
values. These initial values were, however, still used for dew- Newton iteration can be reduced by applying Broyden’s
point and bubble-point pressure predictions. But an arbitrarily algorithm (Broyden, 1965). However, the number of itera-
selected temperature (400 K) instead of the values calculated tions required will be more than that for the rigorous
from Equation (18) was used as the initial temperature for Newton’s method.
the prediction of all dew-point temperatures. The perfor- The above results (which were based on the use of
mances of the various schemes in the prediction of dew Microsoft Fortran and an AT & T PC6300, with 8087-2

THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING, VOLUME 69, AUGUST, 1991 983
 0.6 SS scheme and/or from an accelerated SS scheme to
m Newton’s method. Nghiem et al. (1983) have proposed a set
of criteria which must be satisfied before switching from the
2
0
0.5
SS method to Powell’s method (Powell, 1970) in flash cal-
culations. To adapt their criteria to the prediction of bubble
F points and dew points requires that the conditions placed on
Q the vapor mole fraction be replaced with comparable condi-
w
W
0.4 tions on the pressure or temperature, whichever is the iterate.
On the other hand, a set of general criteria must be based
t on the results of the calculations for a large number of
0.3 different systems. For these reasons, no attempt has been
W made in this study to determine the optimal conditions under
o_ which a calculation should be switched from the SS method
to an accelerated SS scheme or Newton’s. method. The
w 0.2 author’s experience suggests that in most cases it is advan-
I
- tageous to switch to Newton’s method if five SS iterations
-I followed by five accelerated SS iterations have not resulted
in convergence.
When attempting a saturation-point calculation, one may
have a priori information to decide the type of saturation point
that is to be calculated. Alternatively, one may base on empir-
ical rules to select the proper calculation. For hydrocarbon
systems, the empirical correlations for calculating the true
NUMBER OF COMPONENTS critical points may be used to help to decide whether the cal-
Figure 6 - Computation loads of three selected schemes. culation should be for a bubble point or for a dew point. For
example, the critical temperatures of Mixture 1 and Mix-
TABLE6 ture 2 calculated from the correlation presented by Peng
Performances of the Various Schemes at Conditions Where All (1986) are 388.7 K and 204.9 K, respectively. These values
Schemes Converged compare very well with the rigorous values of 388.9 K and
202.4 K. However, empirical correlations may fail and, for
CPU Time / Number of Iterations, seconds calculations in the near-critical region, one may have to resort
Scheme Mixture 1 Mixture 2 Mixture 3 Overall to stability-based initialization procedures.
This paper presents a set of initial-value formulae which
ss 9.82 / 334 11.23 / 291 16.57 I 234 37.62 I 859
may be used in bubble-point and dew-point predictions for
Accl 6.01 / 150 8.34 / 140 13.46 / 125 27.81 / 415
Acc2 5.88 / 162 6.21 / 139 8.72 / 114 20.80 / 415 mixtures of hydrocarbons and related fluids. Empirical for-
Acc3 6.10 / 168 6.72 / 150 8.80 / 114 21.63 / 432 mulae according to which the iterates may be efficiently cor-
Newton 5.88 I 96 11.31 I 101 18.30 / 84 35.49 / 281 rected are also given. Furthermore, it is shown that the
acceleration algorithms developed by Heidemann and
’The SS iterations used in initialization are included in all the tables. coworkers for use in flash calculations are all applicable to
The numbers shown here include 24 for Mixture 1, 17 for Mix- saturation point calculations. In particular, the transforma-
ture 2. and 15 for Mixture 3. tion matrix of their Algorithm 1 can be rewritten in a form
that has nontrivial value at a saturation point. In conclusion,
coprocessor) show that the proposed acceleration schemes
the accelerated SS schemes represent a set of useful alterna-
all gave noticeable improvement in the convergence speeds
tives to the multivariate Newton’s method when an efficient
of bubble-point/dew-point calculations. The improvement is
algorithm is needed in bubble-point and/or dew-point
similar to that found by Mehra et al. (1983) for their acceler-
calculations.
ated SS algorithms in flash calculations. The extra mathe-
matical operations required to carry out the acceleration
schemes for the prediction of saturation points are also com- Acknowledgement
parable to that for the flash calculations. The author gratefully acknowledges the financial assistance
Although the phase behavior of the example mixtures does received from the Natural Sciences and Engineering Research
not represent that of all the possible hydrocarbon systems, Council of Canada.
the problematic features that are generally encountered in
phase equilibrium calculations have been manifested by these Nomenclature
mixtures. In particular, the results show that the SS method
is the most reliable and that the accelerated SS schemes are f = fugacity
generally more reliable than Newton’s method in getting a g = gradient of Gibbs free energy
correct solution. The results also show that, while the SS K = equilibrium ratio
method usually takes the greatest number of iterations, L = liquid moles
n = mole number
Newton’s method may take more iterations than the acceler-
N = number of components
ated SS schemes and that the accelerated schemes may = a quantity defined in Equations (19a) and (19b)
become inferior to the conventional SS scheme under cer- p
P = pressure
tain unpredictable conditions. r = fugacity ratio
Because of the complexity of the phase behavior of natur- S = a quantity defined in Equation (24)
ally occurring systems, it is difficult to devise a general T = temperature
strategy for switching from the SS scheme to an accelerated u - ‘ = element of a transformation matrix

Y84 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING, VOLUME 69, AUGUST, 1991
V = vapor moles Michelsen, M. L., “The Isothermal Flash Problem, Parts I and
w = element of an empirically defined matrix II”, Fluid Phase Equilibria 9, 1-40 (1982).
x = liquid mole fraction Michelsen, M. L., “Saturation Point Calculations”, Fluid Phase
y == vapor mole fraction Equilibria 23, 181-192 (1985).
I = feed mole fraction Nghiem, L. X., K. Aziz and Y.-K. Li, “A Robust Iterative Method
for Flash Calculations Using the Soave-Redlich-Kwong or the
Greek Letters Peng-Robinson Equation of State”, SOC. Pet. Eng. J . 23,
521-530 (1983).
01 = vapor-to-feed ratio Nghiem, L. X. and R. A. Heidemann, “General Acceleration
6 = Kronecker delta Procedure for Multiphase Flash Calculation with Application to
t = scaling factor Oil-Gas-Water Systems”, paper presented at the 22nd European
h = acceleration parameter Symposium on Enhanced Oil Recovery, Paris, France, November
w = acentric factor 8-10, 1982.
Nghiem, L. X., Y.-K. Li and R. A. Heidemann, “Application of
Subscripts the Tangent Plane Criterion to Saturation Pressure and Temper-
ature Computations”, Fluid Phase Equilibria 21, 39-60 (1985).
B = bubble point Peng, D.-Y., “An Empirical Method for Calculating Vapor-Liquid
c = critical state Critical Points of Multicomponent Mixtures”, Can. J . Chem.
D = dew point Eng. 64, 827-830 (1986).
i , j , k = component indices Peng, D.-Y. and D. B. Robinson, “A New Two-Constant Equa-
S = saturated state tion of State”, Ind. Eng. Chem. Fund. 15, 59-64 (1976a).
Peng, D.-Y. and D. B. Robinson, “User’s Guide to GPA Peng-
Superscripts Robinson Program”, Gas Processors Association, Tulsa, Okla-
homa, U.S.A. (1976b).
L = liquid Peng, D.-Y. and D. B. Robinson, “A Rigorous Method for Vapor-
V == vapor Liquid Equilibrium Calculations”, paper presented at the AIChE
(n) = iteration number 1977 Annual Meeting, New York, N.Y., November 13-17,
1977.
References Powell, M. J . D., “A Hybrid Method for Nonlinear Equations”,
in “Numerical Methods for Nonlinear Algebraic Equations”,
Broyden, C. G.. “A Class of Methods for Solving Nonlinear Simul- P. Rabinowitz, Editor, Gordon and Breach, London (1970).
taneous Equations”, Math. Comp. 19, 577-593 (1965). Prausnitz, J. M., C. A. Eckert, R. V. Orye and J . P. O’Connell,
Etter, D.0. and W. B. Kay, “Critical Properties of Mixtures of “Computer Calculations for Multicomponent Vapor-Liquid
Normal Paraffin Hydrocarbons”, J. Chem. Eng. Data 6,409-414 Equilibria”, Prentice-Hall, Englewood Cliffs, N.J. (1967).
(196 I). Rijkers, M. P. W. and R. A. Heidemann, “Convergence Behavior
Gonzalez. M. H. and A. L. Lee, “Dew and Bubble Points of Simu- of Single-Stage Flash Calculations”, ACS Symposium Series 300,
lated Natural Gases”, J . Chem. Eng. Data 13, 172-176 (1968). 476-493 (1986).
Gupta, A . K., P. R. Bishnoi and N. Kalogerakis, “An Acceler- Risnes, R., V. Dalen and J . I . Jensen, “Phase Equilibrium Calcu-
ated Successive Substitution Method for Single Stage Flash Cal- lations in the Near-Critical Region”, paper presented at the 1st
culations”, Can. J. Chem. Eng. 66, 291-296 (1988). European Symposium on Enhanced Oil Recovery, Bournemouth,
Mehra. R. K.. R. A. Heidemann and K . Aziz, “An Accelerated Great Britain, September 21-23, 1981, Dev. Pet. Sci. 13,
Successive Substitution Algorithm”, Can. J. Chem. Eng. 61, 329-350 (1981).
590-596 (1483).
Michelsen, M. L., “Calculation of Phase Envelopes and Critical ~~

Points for Multicomponent Mixtures”, Fluid Phase Equilibria Manuscript received October 23, 1989; revised manuscript
4, I--10 (1980). received April 10, 1990; accepted for publication May 9, 1990.

THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING, VOLUME 69, AUGUST, 1991 985