3. Reiner, Markus, Am. 1. Math., 67,350 ( 1945).

4. Rivlin, R. S., Nature, 160,611 (1947).

5.
, Proc. Roy. SOC. (London) Ser. A, 193, 260
(1948).
6. Serrin, James, Handbuch der Physik, vol. 8/1, p. 231,
S. Flugge, ed. Springer-Verlag, Berlin, Germany ( 1959).
7. Reiner, Markus, Deformation, Strain, and Flow, p. 246,
Interscience, New York ( 1960).
8. Sadowski, T. J., Trans. SOC. of Rheol., 9, No. 2, 251
(1965).
9. Ashare, Edward, R. B. Bird, and J. A. Lescarboura,
AIChE J., 11,910 (1965).
10. Sutterby, J. L., Ph.D. thesis, Univ. Wisc., Madison (1964).
11. Turian, R. M., AIChE I., 13,999 (1967).
12. Slattery, J. C., and R. B. Bird, Chem. Eng. Sci., 16, 231
(1961).
13. Tomita, Yukio, Bull. Japan. SOC. Mech. Engrs., 2, 469
( 1959).
14. Wallick, G. C., J. G. Savins, and D. R. Arterburn, Phys.
Fluids,5,367 ( 1962).
15. Pawlowski, J., Kolbid Z., 138,6 (1954).

16. Slattery, J. C., AIChE I., 8, 663 (1962).

17. Hill, R., 1. Mech. Phys. Solids, 5, 66 (1956).
18. Hill, R., and G. Power, Quart. J . Mech. Appl. Math., 9,
313 (1956).
19. Johnson, M. W., Jr., Phys. Fluids,3,871 (1960).
20.
, Trans. SOC. Rheol., 5 , 9 (1961).
21. Bird, R. B., Phys. Fluids, 3,539 (1960).
22. Ehrlich, Robert, and John C. Slattery, Ind. Eng. Chem.
Fundamentals, 7,239 ( 1968 ) .
23. Bird, R. B., Can. I. Chem. Eng., 8,161 (1965).
,Chem. Eng. Sci., 6,123 (1957).
24.
25. Wilde, D. J., Optimum Seeking Methods, PrenticeHall, Englewood Cliffs, N. J. ( 1964).
26. Slattery, J. C., Ph.D. thesis, Univ. Wisc., Madison (1959).
27. Turian, R. M., Ph.D. thesis, Univ. Wisc., Madison (1963).
28. Rabinowitsch, B., 2.physik. Chem., A145, 1 (1929).
29. Coleman, B. D., H. Markovitz, and W. Noll, Viscometric
Flows of Non-Newtonian Fluids, Springer-Verlag, New
York (1966).

Manuscript received December 26, 1967; revision receiued August 19,

1968; paper accepted August 21, 1968.

A New Simulation Method for Equilibrium

Stage Processes
JOHN

F. TOMICH

Esso Production Research Company, Houston, Texas

This paper presents a new, general method for mathematical simulation of equilibrium stage
operations. The procedure solves component material balance equations with a tridiagonal matrix algorithm. Heat balances and summation equations are handled with Broydens method.
The unique feature of this procedure is that, in a mathematical sense, all equations are solved
simultaneously. Therefore, the method can be used for all types of equilibrium stage processes.
Additionally, the use of Broyden iteration insures solutions which are both stable and more
rapid than current techniques. An exact solution for a twenty tray column with twenty components takes approximately 30 sec. on an IBM 360/65 computer. Successful simulations have been
made for both absorption and distillation type of operations which have included complex columns with multiple feeds and side product streams. Design applications of the method cover a
variety of equilibrium stage processes in the chemical and petroleum industries.

Countercurrent flow, mass transfer columns are primary

unit operations used for separation in the chemical and
petroleum industries. Methods for solving the equations
simulating specific types of these column processes are
known (1 to 7 ) . There is, however, need for a more efficient, general calculation method which will predict component distributions and temperature profiles in all types
of towers, even complex units with multiple feeds and
widely varying temperatures and vapor flow rates. Procedures in current use as well as those reported in the
literature work well for either absorption or distillation
type of problems, but in general cannot be applied to

Vol. 16, No. 2

both. Although many of the column processes used are of

these two types, many do not conveniently fit into either
category. Rich oil demethanizers in natural gas processing
plants, for example, usually have absorption characteristics
over a certain part of the column and distillation characteristics over another.
This paper describes a new method which provides
stable, rapid solutions to the mass and energy balance
equations associated with all kinds of tower processes. The
procedure is highly numerical, but with digital computers
it can be applied to many tower design problems. The
method is generally applicable to all countercurrent flow

AlChE Journal

Page 229

f--l

VI

yi,j

- Ki,j ~ i , =
j
0,
f o r l 4 i "M,

lLjLN

(2)

Component material balances

+ Lj-1
+ Fj +
- (Vj + Wj)
- (Lj +

Vj+l yi,j+l

Xi,j-l

~'i,j

Fj+1zui,j+l

R j ) xi,j = 0,
for I l i L M , 1 L j l N

yi,j

(3)

Summation equations
M

i=l

i=l

Heat balances
Ej = Vj+l H j + l
Lj-1 hj-1
Fj aj Ii"j
Fj+l ( 1 - a j + l ) H'j+l - (Vj Wj) Hj
- ( L j Rj)hj - Qj= 0, for 1 L j l N

+
+

(5)

There are N ( 2 M
3) equations. The N(2M 3) unknowns are vapor and liquid internal flow rates, vapor
and liquid mole fractions, and tray temperatures. Quantities which must be given to obtain a solution are: rates,
compositions, and temperatures of all feed streams; number of theoretical trays; rates of all side streams; amount
of heat added to or removed from each tray; equilibrium
and enthalpy data; and initial estimates of vapor flow rates,
tray temperatures, and tray compositions (if nonideal
solutions are used). Different variables may be selected as
unknowns, or additional equations and unknowns may be
iiicluded in the general equations to be solved. Thus, the
method is considerably more versatile than as presented
here.
METHOD OF SOLUTION

VN+I

Equation set ( 2 ) can be substituted into the set ( 3 )

to eliminate yi,j and reduce the component material balance equations to the tridiagonal matrix form

Fig. 1. Model for mathematical simulation.

equilibrium stage operations.

DESCRIPTION OF MODEL COLUMN

The model column for mathematical simulation is shown

in Figure 1. The standard assumptions of constant column
pressure and theoretical equilibrium trays connected by
internal vapor and liquid flows are made. Because of the
general form, the model represents all types of stage-bystage separation processes. Feed streams can be introduced
on each tray of the model, with the vapor portion being
added to the tray above the feed tray; vapor and liquid
side streams can be removed from each tray. Heat -can be
added to or removed from each tray to simulate reboilers,
side heaters, interchillers, and condensers. For simplicity,
only the feed, liquid and vapor side streams, and external
heat flow for a typical tray, tray j, are shown in Figure 1.

for 1 A i

where

EQUATIONS TO BE SOLVED

The basic equations for column simulation are derived

by making material and heat balances around the jth tray
of the model. Heat and mass balances plus equilibrium
relationships comprise the five sets of equations to be
solved:
Overall material balances
Lo

+ Vj+l +

(Fk

-w

Rk)

k=l
+Fj+l

(l-a,+l)

-Vl-Lj=O,

forlLjAN

Vapor-liquid equilibrium

P a g e 230

(1)

These equations are a linear set only when the equilibrium ratios, the &, are not functions of compositional
changes which occur as the solution progresses. Equilibrium ratios can reflect the variation of composition with
position in the column if estimates of this variation can be
made prior to computation. In this case, the Ki,j's at any
position in the tower, while only functions of temperature, are different functions of temperature than those at
some other position.
Equation sets (l), ( 2 ) , (4), ( S ) , arid ( 6 ) are solved
by using an iterative procedure which makes use of

AlChE Journal

March, 1970

Broyden's method (8) for obtaining the solution of simultaneous, nonlinear, algebraic equations. The procedure
may be summarized as follows.
1. Some initial set of Vj and Tj is assumed.
2. The Lj are computed from Equation set (1).
3. Equation set (6) is solved by using the tridiagonal
matrix algorithm (9) to give the xi,+
4. Equation set (2) is solved for the yi,j.
5. These results are substituted into Equation sets (4)
and (5) which are then solved simultaneously by using
the procedure of Broyden for a new set of Vj and Tj.
6. By using this new set of Vj and Ti, steps 2 through
5 are repeated until an appropriate convergence criterion
is satisfied.
Although it is not obvious, all equations are solved
simultaneously by numerically substituting for Lj, xij, and
yi,j in Equation sets ( 4 ) and (5). These equation sets
then become functions of the Vj and Tj only and can be
solved simultaneously for these quantities.
The Broyden method is actually an improved NewtonRaphson iteration (10).Expanding SJ and EJ (the EJ are
normalized by dividing the total heat inputs to the trays
so that they are of the order of magnitude of 1) as functions of Vj and Tj using Taylor's approximation (11) , and
truncating after the first derivative term, we get

TABLE
1. REDUCTIONOF SUMOF RESIDUALS
WITH
ITERATION
NUMBER
J=1

[(EJ12+

(sJI21

Iteration
Absorber

No.

1
2
3
4
5
6
7
8
9
10
11

0.223
0.353 x 10-1
0.355 x 10-2
0.268 x 10-2
0.131 x 10-2
0.333 x 10-3
0.147 x 10-4
0.413 x 10-5
0.511 x 10-6"

Stabilizer
0.483
0.293
0.129
0.783 x
0.297 x
0.261 x
0.331 x
0.463 x
0.106 x
0.288 x
0.348 x

10-1
10-1
10-1
10-2
10-3
10-3
10-4
10-5'

Demethanizer
0.133
0.122
0.109
0.927 X 10-1
0.588 X 10-1
0.210 x 10-1
0.378 X 10-2
0.108 x 10-2
0.129 x
0.284 x 10-4
0.176 x 10-50

Converged to prescribed tolerance.

is fulfilled when a solution is reached, the Broyden procedure (8) as used here systematically searches between
-1 and +1 for a value of t which will meet the following condition:
N

2 [ ( S J ~ " ) ' + ( E J ' " ) ~ ] < 2' [ ( S J s ) ' +

(EJ')']
(13)
Although this value may not be optimum from the standpoint of maximizing the reduction of the residuals, it gives
assurance that the latest values of Vj and Tj result in
movement toward the solution.
J=l

J=l

At this poiat, if
N

Since the SJ and EJ are errors or residuals in the summation and heat balance equations and are equal to zero
when a solution is reached, these residuals are set equal
to zero at the new iteration level. Equation sets (7) and
(8) can also be written in matrix form:

If the S J ~and EJS are calculated by using Equation

sets ( 4 ) and ( 5 ) , and if the partial derivatives are evaluated by finite difference methods, Equation (9) can be
solved by matrix inversion for the AVj and 4Tj. Then,
new Vj and Ti can be computed from the following equations:
Vjstl = VjS t aVi, for 1 6 i 4N
(10)
Since

+
Tj5+l = TjB + t ATj,

for 1L j A N

S J = EJ = 0, for 1A J
Vol. 16, No. 2

(11)
(12)

[sJB+I)~

+ ( E J " ' ) ~ ] < convergence tolerance

(14)

J=1

the problem is considered solved. Where desired, each SJ

and EJ may also be checked for convergence.
If the above test is not met, the entire procedure is
repeated with the new set of Vj and Tj. However, for
succeeding iterations, Broyden's procedure (8) for updating the inverse of the approximate Jacobian matrix [the
matrix of partial derivatives in Equation ( 9 ) ] using residuals calculated at the preceding iteration level is employed. Thus, there is no need for more than one matrix
inversion per problem solution, and computation time is
greatly reduced; in fact, the computation effort is not
much greater than that required when the approximate
Jacobian matrix is not improved.
To obtain a simultaneous solution of all the equations,
new values of Lj, xi,j, and yi,j must be calculated each
time any of the Vj or Tj vary. This is necessary both in
evaluation of partial derivatives and during the iterative
solution.
ADVANTAGES OF THIS METHOD

Several types of iterative methods are available for

solving column process equations. One (1, 2 ) uses trayto-tray procedures, which are often numerically unstable,
mainly due to buildup of truncation error (3).Another
( 4 , 5 ) solves the component material balance equations
for composition and then, in sequence, solves the summation equations for tray temperatures and the heat balance

AlChE Journal

Page 231

equations for vapor rates. These are referred to as bubblepoint methods and are acceptable for distillation processes,
but not absorption (3). A third type (6, 7 ) also solves
the component material balance equations for composition
but sequentially solves the summation equations for vapor
rates and the heat balance equations for temperature.
These are called sum rate methods and have usually
worked for absorption problems, but not distillation ( 3 ) .
The present method overcomes the limitations of both
the sum rate and bubble-point techniques by solving all
equations simultaneously. Successful calculations have
been made for absorption and distillation towers as well
as complex demethanizers. Some aspects of this technique
have been suggested by Friday and Smith (3) and Tierney and Bruno ( 1 2 ) , but applications of the total concept
have not been reported. Both of the suggested approaches
would make use of the Newton-Raphson iteration (10)
to solve the summation and heat balance equations. The
recommended Broyden procedure (8),however, has two
advantages over classical Newton-Raphson iteration. First,
numerical stability is improved because the Broyden
technique guarantees that each iterative step moves
toward the solution. Second, since Broydens procedure
uses only one matrix inversion per problem, it significantly
reduces the amount of computation required.

NOTATION
= coefficient in tridiagonal matrix equation
= coefficient in tridiagonal matrix equation
= coefficient in tridiagonal matrix equation
= coefficient in tridiagonal matrix equation
= error in tray heat balance equation for iteration
= flow rate of feed stream to tray
= enthalpy of liquid stream leaving tray
= enthalpy of liquid portion of feed stream to tray
= enthalpy of vapor stream leaving tray
= enthalpy of vapor portion of feed stream to tray
= equilibrium K value of component on tray (de-

fined as y i , j / q j )
= flow rate of liquid stream leaving tray
=

number of components

= number of theoretical trays

= heat removed from tray
= flow rate

of liquid side stream leaving tray

= error in tray summation equation for iteration

= weighting factor in Broydens iterative procedure
= temperature of tray
= flow rate of vapor stream leaving tray
= flow rate of vapor side stream leaving

tray

= liquid mole fraction of component on tray

= vapor

mole fraction of component on tray

= mole fraction of component in liquid portion of

feed stream to tray

EXAMPLE CALCULATIONS

Example calculations for an absorber, stabilizer, and

complex demethanizer were made. The absorber is a
column of five theoretical trays with a liquid feed introduced at the top tray and a vapor stream fed to the
bottom tray. The solution required 3 sec. on an IBM
360/65 computer for the nine iterations needed to converge. The stabilizer is a ten tray tower with liquid feed
to the top tray and heat added to the bottom tray to
simulate a reboiler. The stabilizer computation took 7 sec.
and eleven iterations. The complex rich oil demethanizer
column has eighteen theoretical trays. A liquid is fed to
the top tray, and two-phase feed streams are introduced
on trays seven and eight. There is a side heater at tray
eleven, and the bottom tray serves a s a reboiler. The
eleven iterations used 20 sec. of computer time.
The time and number of iterations given for each
solution depend upon the accuracy of the initial estimates
of tray temperatures and vapor flow rates. The sum of
the residuals of the heat balance and summation equations at each iteration level are shown in Table l and are
a better indication of the methods efficiency.

= mole fraction of component in vapor portion of

=

feed stream to tray

liquid mole fraction of feed stream to tray

Su bscripts

i
j,J

= component number
= tray number

= dummy subscript
= bottom tray in column

Superscripts

= iteration number

F
1
u

= liquid portion of stream

feed stream

= vapor

portion of stream

LITERATURE CITED

1. Holland, C. D., Multi-Component Distillation, PrenticeHall, Englewood Cliffs, N. J. ( 1952).

2. Thiele. E. W., and R. L. Geddes, Ind. Eng.
- Chena., 25,
289 (1933).
3. Friday, J. R., and B. D. Smith, AIChE J . , 10, 698 (1964).
Process., 45,
4. Wane. 1. C.. and G. E. Henke,. Hqdrocarbon

155 r1$66).
5. Admundson, N. R., and A. 5. Pontinen, Ind. Eng. Chent.,
50,730 ( 1958 ) .
6. McNeese, C. R., Chena. Eng. Prog. Symposium Ser. NO. 37,
58,43 ( 1962).
7. Smith, B. D., Design of Equilibrium Stage Processes,
McGraw-Hill, New York ( 1963).
8. Broyden, C. G., Mathematics of Computations, 19, 577
(1965).
9. Lapidus, Leon, Digital Computation for Chemical Engineers, p. 254, McGraw-Hill, New York ( 1962).
10. Scarborough, J. B., Numerical Mathematical Analysis,
p. 203, John Hopkins Press, Baltimore, Md. (1958).
11. Sokolnikoff, I. S., and R. M. Red:effer,
Mathematics of
Physics and Modern Engineering, p. 257, McGraw-Hill,
New York ( 1958).
12. Tierney, J. W., and J. A. Bruno, AZChE J., 13, 556 (1967).

CONCLUSIONS

A general method, based upon matrix operations, has

been developed for giving stable, rapid solutions to column
material and energy balance equations. The calculation
solves all equations simultaneously with Broyden iteratloll
methods.
This technique has the advantage of yielding solutions
to all different types of tower processes, including both
absorption and distillation operations. Furthermore, the
iteration methods, which require only one matrix inversion for each column simulation, give numerically stable
solutions with a minimum of computing time.
ACKNOWLEDGMENT

The author wishes to thank D. L. Guenzel, H. A. Lesser,

and J. F. Wolfe for their valuable suggestions. He also would
like to express appreciation to the management of Esso Production Research Company for permission to publish this work.

Page 232

Mnrrrrscript receitied M a y 5 , 1968; reaisioti received A u g u s t 15, 1968;

i x r p e r occepted August 19, 1968. P a p e r presented at AIChE Tnmpo meeting.

AlChE Journal

March, 1970