An Improved Method for Numerical Solution of

Distillation Processes
JAMES W . GENTRY
University of Maryland, College Park, Maryland, U.S.A.

A new method for the numerical solution of a stagewire On prksente une nouvelle mkthode pour la solution
process is presented. The method consists of total linear- numerique des prockdes &tag&; elle consiste en la mise
ization of all the equations in the distillation process, using sous forme lineaire de toutes lea kquations relatives au pro-
the method of quasi-linearization. The resulting equations ckdk de distillation, en employant la mkthode dite de
form a block-band matrix which can be readily solved by "quasi-linkarisation". Les kquations rksultantes forment
a method developed by the author. The method used for une matrice H groupe unitaire que I'on peut rksoudre
obtaining convergence is presented. facilement au moyen d'une methode mise au point par
The method has been tested for both binary and multi- l'auteur. On expose la mkthode utilisee pour obtenir la
component syetems. The method has been compared with convergence. On a mis la dite mkthode H I'kpreuve dans
the Amundson-Pontinen method for a five component mix- le cas de systemes binaires et H constituants multiples et
ture. The method gives more rapid convergence. Numerical I'a comparke avec celle d'Amundson et Pontinen en rap-
experiments indicate that the vapor and liquid flow rates port avec un mklange H cinq constituants. La nouvelle
are more stringent requirements than the bubble point. methode permet I'obtention d'une convergence plus rapide.
Les expkrienees numkriques indiquent que les debits de la
vapeur et du liquide sont des exigences plus rigoureuses
T he description of equilibrium stages processes by the use of
matrix manipulation is well known. (Amundson and
Pontinen(I), Amundson, Pontinen and Tierneyc2), Tierney and
que le point de bulle.

Bruno@). These methods involve the assumption that the effects This approximation is linear in the variables zli.
of temperature, flow rates, and composition can be decoupled, For example the component mass balance about the reboiler is
so that the resulting matrix equations can be solved by short-cut LI + VI Y,j = L2
Xlj xzj.
methods of Gaussian elimination such as the Thomas algorithm, or
(Thomas(')). LIX1j + VI = LZX2,.Klj X l j
Recently the application of the Newton method to all non- T h e linearized form of this equation is
linear terms as suggested by Bellman, Juncosa, and Kalaba(6)
has been applied to tubular reactors and small particle Li0 Xij' + Xtj' [Li- Lo] L1' + [Xlj - X1,OI +
reactions (Gcntrycs)). The method was applied to a distillation
column by Greenstadt, Bard, and R . l ~ r s e ( ~however
), their
calculation procedure did not involve matrix operations.
In this paper, a recently developed algorithm for solving block
band matrices (a matrix in which the band elements are square
LI" xZjo .+ Xzj* [ ~ -
2 ~ 2 0 1 +.L ~ [O,X , - x,'].
blocks of m equations and m unknowns) by Gentry@)has been T h e superscript ( 0 ) indicates the previous trial. The formerly
applied to a distillation column. The application of the method non-linear equation is now linear in the variables L,, V,, T I ,
to an equilibrium stage process here is the first known application L2,X 2 j and xlk. (k = 1, 2 . . . n).,
of the algorithm to chemical engineering problems. T h e equations describing a distillation column, heat and
In the development the equations are linearized, solved by component mass balances as well as equilibrium relations, are
the a1 orithm, and then corrected to insure convergence. The non-linear. In this treatment all the equations are linearized.
B
use o the method in the treatment of three problems is discussed This approach was previously suggested by Greenstadt, Bard,
and Morsec9).
1) the binary n-butane - iso-butane In the sections below the linearized equations for each stage
2 ) the binary iso-butane - propane are developed.
3) the ternary n-butane, iso-butane, - propane.
Overall mass balance
T h e method is applied to a 26 stage (including the reboiler
and condenser) column for a wide variety of cases. The overall mass balance is given by
T h e paper gives a practical method for solving the equations Vi+L,= V,+Lk+Fi . . . . . . . . . . . . . _(1)
resulting from a distillation process. It involves complete where V, is the vapor flow leaving rhe i stage,
linearization of the distillation equations. It is the first use of Li is the liquid flow leaving the i stage.
the new computational procedures, and it suggests a method of
convergence. T h e subscript g indicates the i - 1 stage and the subscript
The method is applied to a wide variety of cases for the R indicates the i +
I stage.
three systems. The values were selected in order to give as The superscript o indicates the value of liquid and vapor
demanding a test of the program as possible. for the previous trial.
T h e linearized mass balance is given by
Quasilineariration
In the method of quasilincarization a non-linear function
Vi'yi + Li' Xi - v,"y, - Lh" Xt = Fi - v? - Li"f vg' + Lk"(2)
F (ui) is approximated by In the above equation
aP" vi - Li - LiO
.. . . . . . . . . . (3)
F (Ui) = F (Ui(0)) +-
aui . ( U i - %a). yi = ~

ViO
vio
and Xi = ____
Li"

The Canadian Journal of Chemical Engineering, Vol. 48, Augurt, 1970 451
Component mara balance In the mathematical derivation presented here, the equation
is derived for the equilibrium K values as general functions of
The component mass balance is given by composition and temperatures. Similarly the partial molar
Vn Yi, + Li Xij = Vg YIj + Lh XhJ + Fc Zij, enthalpies of the liquid and vapor phase are treated as general
or functions of composition and temperature. In the numerical
(Va Ki, + Li) Xtj = V; K;j X,j LI x h j -I-Fi Zij.. . . . (4) examples presented here the K values are assumed to be inde-
pendent of composition and the partial molar enthalpies are
T h e linearized equations are given by assumed to be independent of composition. T h e first assumption
Vi Kij Xijl' yi + [Li Killo hi - IV; KIj X;jlD7;- is frequently made for hydrocarbon mixtures. T h e assumption
[Lh K I ~ ]X "k f [ Vi Kij Li]"Xij - I VI KIjl' X;j - can be removed by using the Wilson equation to correct the
K values for composition. T h e assumption that the partial
molar enthalpy is independent of composition is equivalent to
assuming no heat of mixing.

Equilibrium conriderationr
It is assumed that the liquid on a tray is at its bubble point.
The basic assumption is that

If the distribution constants K are independent of composition z

j-1
Yij = z
j-1
Kij Xlj = 1.. . . . . . . , . . . . . .(9)
the last two terms can be neglected.
In the above equation T h e linearized form of this equation is
Ti - Ti0
7. = - .. . . . . . . . . . . . . . . . . . . (6)

Enthalpy balance
The enthalpy balance for each stage i is given by

or
a
Lk xkghk; = Fi h i . ... ., ... , . . . . . (7)
t-1

where g,, is the partial-molar enthalpy of the qth component for

I "-1 aK.. *
the vapor phase and h,, is the partial molar enthalpy of the z ( a2
liquid phase of the qth component at a temperature Ti. Ti is
2;
1-1 ,-I , ) ( X i , - Xi,") x,j -
x' 1 = 0.

the temperature at the ith stage.

The linearized equation for the enthalpy is If the equilibrium K values are independent of composition the
term with the double summation may be neglected and one has

[vi s 1-1 Kag xi, ~ i a ] ' yi + [

Li j , xi, hi,] -

In the numerical calculations presented in this paper the

equilibrium constant K,j was expressed as a polynomial in
temperatures. T h e extension to cases where the K value is a
factor of composition as well as temperature is a straight-
forward differentiation of the Wilson (or an analogous equation)
equation. I t is convenient to treat the derivatives of the K values
with composition and temperature as function subroutines.
This requirement is modified if a total condenser rather tliaii
a partial condenser is used. If the reflux is returned as a bubble
point liquid, the requirement remains the same. Otherwise this

[ z"
I.,
a-I
xi,
- a
2:
axi,I_ (hi,)
I ( X , p - XIP0)-
equation is replaced with one of the specifications for the total
condenser.
The equations describing mass and heat exchange between
stages can be expressed as a block-band matrix with the resulting
matrix of the coefficients being a tri-diagonal block-matrix. If
the additional assumption is made, that each variable within a
block, is independent of the other variables within the block,
one can uncouple the equations. Therefore one has (C+2) tri-
diagonal matrices. Assutning that one has N stages, the number
of operations required for an interaction is 6 (C4-2): N . This
is essentially the method of Amundson-Pontinen.
T h e method is comparable with the Amundson-Pontinen
method since in many respects the method is an extension of
their method. In order to be an improvement over the existing
methods, the method should be more reliable and not require
much more time. In the opinion of the author both requirements
are mer.

The Canadian Journal of Chemical Engineering, Vol. 48, Augur:, 1970

 The method described here does not uncouple the elements TABLE1
of the block, but solves the complete matrix. In this case the
required number of operations are approximately N (C+2)*. Storage Operations
Ideal Com- -
Stages ponents
Z s - 1 Algorithm Gauss Algorithm
Thus the use of the algorithm requires approximately
c+2 3 16 2 4.103 8.10' 8.10' 5.103
--
6 2
(3C 8) + + 3 times as many operations. 31
16
2
3
1.5.10'
6.4.105
1.6.103
1.2.103
6.3.10'
1 . 7 . lo6
1.104
9.10'
16 5 9.3.105 1.8.103 8.9.103 2.2.10'
This comparison is not complete, for it neglects the time 100 8 1.0.106 3.0.10' 3.3.108 4.3.10'
required to set up the matrix for the required manipulations. A s
a rough measure the algorithm here requires the determination
of (C+2)2 . A' . 3 coefficients. Although the Amundson-
Pontinen method requires only (C+2) . N . 3 coefficients, the The superscript (s) indicates the number of the iteration.
work involved in calculating the coefficients is of the same order T h e parameters e and 6 p are arbitrary positive numbers.
of magnitude. It is obvious that the method converges to a solution if
The time for structuring the matrix requires Equation (1 1) applies for arbitrary values of B p and e. The
solutions are further restricted so that Ti, Li, Vi, and Xji are
N(C+2)*.3$3fl non-negative. It may be possible to obtain a solution to Equation
(1 1) that is physically unrealistic since it includes negative
temperatures, flow rates, or compositions.
T h e convergence equation
the time required to solve the matrix. 4 is the percentage of 0
elements in each block. fl is the average number of operations Fij(*+') =
N

Fij(') aij + (1 - aij) ... . .

, , , (12)
required per element. If the k values are functions of composition N

where Fij(') is the solution of the matrix equations where the

I
then t$ 5 1 and s1 - (3C
d.
+ 8) + 1. In this case there is no coefficients are evaluated using Fij(').The superscript s indicates
the number of the iteration. T h e parameter aij is an arbitrary
value chosen for rapid convergence.
advantage using the Amundson-Pontinen method. If the k
aij is chosen so that
values and enthalpies are functions of the temperature alone,
cp 5 3/C+2. l(Fij('+') - Fij('))/(Fij(') + 6il Fij)l < e i j . . . . . . . (13)
Therefore, it seems reasonable to conclude that this method and
requires more time per iteration than the Amundson-Pontinen F$) Fij('+l) > 0 for all i a n d j . , . . . . . . , , . , (14)
method, but this time is probably less than C t*. (t* being the In Equation ( 1 3 ) , ei, is chosen to be much greater than zero,
time consumed by the Amundson-Pontinen procedure). where S,j is allowed to approach arbitrarily close to zero.
The time required for conventional Gaussian elimination is Superficially, it appears to be of advantage if all aij are chosen
of the order of [(C+2) NJ3/3. In the table below the number equal. Unfortunately, this is not the case. For very low con-
of operations and the storage capacity required for this method centrations Xij < lo-' or very high concentrations Xij > 0.999,
and conventional Gaussian elimination are given. the value of aij must be small in order to satisfy Equation (14).
It should be pointed out that there is no loss of accuracy or T h e use of such small values of a would make convergence
precision using the algorithm, indeed error due to round-off extremely time consuming.
is even less. The algorithm was tested with a quin-diagonal
matrix with 1800 variables. T h e answer obtained was the Results (discussion oE numerical studies)
correct result. In this section the numerical results for the binary and
T h e method is especially convenient for the use in stagewise
ternary mixtures are given. A 26 stage column with the feed
processes, in that similar equations can be written for each introduced at the 12th stage is used for most of the cases,
stage. T o use the method it is necessary that the same number T h e computations were carried out on a Univac 1108 digital
of equations be written about each stage. This necessitates some computer. Approximately .34 seconds were required per itera-
redefining. For example, if the reflux ratio and the amount of tion for the binary and .80 seconds for the ternary. I t is quite
distillate arc specified, it is necessary to specify the reflux and possible that most of the time was required for the convergence
the bottoms. The method is applicable to all the following procedure and determining the coefficients of the matrix.
specifications of variables:
I ) number of plates Binary (iso-butane -normal -butane)
2) feed plates For the binary systems, three different amounts of distillate
3) reflux ratio, purity of A in distillate, fraction of B in were examined (D/F = 0.5, 0.05, and 0.95). T h e reflux rate
distillate, heat duty in condenser, temperature of con- was 10. In most cases the initial reflux rate (assumed flow rates)
denser were many times the actual rate. In the table below the results
4) heat duty in reboiler, temperature of reboiler, amount of are shown for the iso-butane - propane binary. T h e feed for all
bottoms, fraction of A in bottoms. cases was introduced at 323'K and was a saturated liquid. T h e
In general, the enthalpy balances in the reboiler and condenser results are given in Table 2.
are replaced by the other specifications. If a total condenser is
used, the bubble point equation at the top stage (condenser) is Binary (propane -iso-butane)
replaced. T h e propane-iso-butane split is more difficult to calculate
Convergence due to the smaller compositions. T h e cases investigated, here,
are essentially the same as those for the iso-butane - n-butane
The criterion that is used for convergence is that Maximum system. The data here are less complete than for the iso-butane -
(AViJ, ALi) +
Maximum (AT,) +
Maximum (AXjj) 5 e n-butane system due to the more difficult calculation. (It is
where ironic that the easier the separation the more difficult the calcula-
tion.) T h e source of the difficulty may be due to the poor
initial estimate of the vapor-liquid flow rates. T h e results are
for all functions F . given in Table 3 .

The Canadian Journal of Chemical Engineering, VoI. 48, Augurt, 1970 453
 TABLE2 TABLE
3
--
- -- I_
__
I I I
Feed Bottoms Feed Bottoms
Composition Composition Composition Composition
Initial D / F Initial
__ ___ rrials D/F
^I 'Trials
Reflux Ratio
Rate i- n-
stage:
R{fFtRatio I

Butani 3utanr
__ - - __
4.2 0.95 0.5 0.5 26 14 8 0.50 0.50 0.50 0.10.10-3 0.9999 35'
4.2 0.95 0.95 0.05 26 8 8 0.50 0.95 0.05 0.9000 0.1000 12
4.2 0.95 0.05 0.95 0.31. 8 0.50 0.05 0.95 0.3 1.1 O-s 1.000 7
26 7 80 0.05 0.95 0.05 0.9474 0.0526 14
80.5 0.95 0.05 26 10 80 0.05 0.5 0.5 0.4737 0.5263 lh
80.5 0.05 0.032 0.997
0.95 26 7 80 0.05 0.05 0.95 0.0220 0.9780 19
8 0.5 0.5 0.123 0.877
0.5 26 7 4.2 0.95 0.05 0.95 0.70. 10V6 1.000 8
80 0.05 0.05 0.040 0.960
0.95 26 7
80 0.05 0.5 0.475 0.525
0.5 26 8
80 0.05 0.95 0.947 0.526
0.05 26 7
10 0.4 0.48 0.1567 0.8433
0.52 9 6
10 0.4 0.46 0.1295 0.8706
0.54 9 6 T h e results of the 1, 2, 3, 5 , 7, 9, and 10th iteration for tlic
10 0.4 0.44 0.1047 0.8953
0.56 9 7 15 stages are presented in Table 5*. Satisfactory convergence
10 0.4 0.42 0.0834 0.9166
0.58 9 6 occurs after 10 trials. In Table 6, the maximum error of the
10 0.4 0.30 0.0242 0.9759
0.70 9 6
bubble point check on the 1 5 stages is prcsented as a function of
___ __
__ __ __ _I __
~ __ -
__ __
the number of iterations. T h e maximum error occurs for the
check on the enthalpy balance. This value is also given in Table
+ This rcpresents an earlier experiment. In order to insure 6 as a function of the number of iterations. T h e bubble point
convergence, a change was adopted. This change resulted in
one more trail bcing neccssary for convergence.
I In this case the nicthod, was clearly converging. T h e bubble
deviation is defined as
(j:i

-
)
2 Kii Xi]- 1 and the liquid flow

rate check is defined as iL('+') L('))/L(').T h e superscript (5)

points chccked within 1 .O. 10-'6 for all plates. T h e flow rates
in the middle of the column were undergoing small changes indicates the numbers of the iteration.
(.I%) which prevented che formal requirement for convergence In most distillation methods (lJ0s11), the convergence is based
from being met. on a bubble point check. Here all variables are testcd for
comparison with their previous values. In this case a satis-
Ternary Syetem factory bubble check is obtained after 4 trials; however, the
Extensive numerical calculations were carried out for the estimates of the liquid flow rate are unsatisfactory. This numeri-
ternary system propane - iso-butane - n-butane. T h e reflux cal experiment as well as others (12)* show that a check on thc
ratio was 10 and the distillate was set at .5 times the feed. T w o bubble point is a thousand times less restrictive than a chcck oii
cases were investigated, the initial L I D ratios in the enriching the liquid composition for this method. If very bad assumptions
section of the column were 10, and 20. are made regarding the initial flow rates, convergence requites
The results were given in Table 4. one or two more iterations. T h e early iterations may givc
bizarre results ( Q . T h e temperature convergence is rapid.
The method used to insure Convergence is that no tempera- After four or five iterations, the tem erature profile is vcr)'
ture on a higher stage can be greater than the temperature on a
lower stage. If this condition is violated
ff
near the final profile. T h e initial pro le shows a 20-30' dif-
ference from the final profile.
(Ti+l > Ti) T h e composition is slower to converge, especially for very
small compositions. After seven to eight iterations convergencc
then thc i+ 1 is assigncd the flow rates and compositions appro-
priate to the itli stage. T h e approach is awkward, and results
is rapid. Only for very small concentrations 10-0) are -
the discrepancies significant. T h e result after the first few
in bizarre compositions in some of the early iterations, but it
iterations is especially inconsistent.
does converge.
In all the calculations difficulties were encountered for T h e problem was chosen, so that the method could be
systems in which the amount of propane in the feed was equal to compared with the Amundson-Pontinen method. In a similar
the amount of distillate. The above mentioned scheme seems to problem (the same compositions were used in a 1 5 stage column),
converge, but convergence is still difficult. It should be noted even after 1 S iterations the bubble point check was not within
that the initial value of the composition on each try was the feed -
1.0 lo-*. T h e same check is reached in four iterations with
composition. the current method. For the Amundson-Pontinen method, the
temperature checks within 0.5' between the 10th and 15th
Results (discussion of convergence) iterations. With the present method a similar check is obtaincd
In this section a special case of a 5-component mixture is after five iterations. For the Aniundson-Pontincn method the
considered. T h e purpose of this section is to indicate the details flow rates between the tenth and fifteen iterations agree within
of convergence. Such a study may indicate possible changes to 0.4%. W e obtain a similar result after 8 - 9 iterations. How-
make the convergence more rapid. ever, the flow rates predicted by the Amundson-Pontinen
In this calculation a mixture consisting of Ethane (0.03), method are very stable from the 5th iteration upwards in con-
Propane (0.20), Butane (0.37), Pentane (0.35), and Hexane trast to our result. T h e faster convergence for our method is to
(0.05) is separated in a 1 5 stage column such that the ratio of the be expected, since the essential difference between the methods
liquid to the distillate is 5: 1 and such that the ratio of D / F is is that here all equations are solved simultaneously.
0.226. The feed is introduced in the 8th stage at 225°F. T h e I t is difficult to compare computation time, since prcscnt
problem was deliberately chosen to be very similar to the day computers are faster than those previously used. Here, cach
problem worked by Amundson and Pontinent'). Feed composi- iteration took between 2 and 3 seconds. T h e bulk of the time
tion is chosen as the first estimate of the composition on the is spent in restructing the matrix. For a fifty plate, five coni-
plates. T h e flow rates are assigned values, assuming constant -
molar overflow. A linear temperature profile is used with the 'Table 5 of this paper has been d osited with the National Depository
of Unpublished Data, National g i e n c e s Library, National Research
top and bottom temperatures chosen at 110°F and 30S'F. Council, Ottawa 2, Om. Write NRC for copies.

454 The Canadian Journal of Chemical Engineering, Vol. 48, August, 1970
 TABLE
4 TABLE
6
- - - ~- ~

I
Composition Bottoms Composition
Initial -_ -1 I Iteration
Reflux Trials
Ratio Pro- iso- n-
patie utant iutanc I

-1-1 0.4999 - 1.09.10-1

1.1.10-1
10 0.50 0.25 0.25 0.90. lo-‘ 0.5000 2 0’ 5.6.10-8
10 0.25 0.50 0.25 0.45.10-1 0.5025 0.4975 9 1.3. lo-‘ 3.7.10-1
10 0.25 0.25 0.50 0.63.10-6 0.0501 0.9499 9 6.2.10-6 1.2.10-1
10 0.20 0.20 0.40 0.20.10-5 0.2065 0.7935 9 1 .2.10-5 8.0.10-8
10 0.30 0.40 0.30 0.40.10-5 0.4026 0.5971 9 2.0.10-8 2 .4.10-J
10 0.40 0.30 0.30 0.53.10-6 0.4014 0.5986 11 1 .0 .lo-’ 6.8.10-4
10 0.20 0.20 0.60 0.35.10-0 0.0182 0.9819 9 3 .O. 1. 7 . 1 k 4
20+ 0.50 0.25 0.25 Di- 4.5. 5.4.
verges
20 0.25 0.50 0.25 I
0.45.10-6 1 0.5025 0.4975 8
20 0.25 0.25 0.50 Same as Before 8
20 0.30 0.30 0.40 8
20 0.30 0.40 0.30 8 The calculations, which have been carried out, indicate that
20 0.40 0.30 0.30 8 the algorithm for the solution of block-band matrices is suffici-
20 0.20 0.20 0.60 8
20 0.20 0.60 0.20 0.48.10-6 0.6021 0.3979 8 ently rapid to allow the linearization of all the equations de-
__ __
scribing the distillation process.
_ _ __ __
__ __
__ Furthermore the method of convergence used in this paper,
has a high degree of success even with relatively insensitive
ponent distillation process, approximately 6 seconds are re- convergence coefficients and demanding initial estimates of the
quired for each iteration. flow rates, compositions, and temperature.
Extension of the method
Acknowledgments
The program was originally designed to solve a distillation Computer time was supported through the facilities of the Computer
column equipped with a partial condenser with the following set Science Center of the University of Maryland under NASA Grant 398.
of variables specified:
1) the number of plates Nomenclature
2) feed plate location C = number of components
3) reboiler duty D = amount of distillate
F = feed flow rate; dummy variable
4) condenser duty.
The method can be easily modified if instead of the reboiler
fi
h
= partial molar enthalpy (vapor)
= partial molar enthalpy (liquid)
duty, another specification pertaining to the bottoms is specified. K = equilibrium constant ( U / X )
Typical examples of such specifications would be: N = number of stages
1) reboiler temperature n = the number of components
T = temperature
2) bottoms amount (distillate amount) Y = vapor mole fraction
3) bottonis/fced ratio ( d / f , d / b ) X = liquid mole fraction
4) purity of a component in the bottoms. = convergence parameter
A more difficult problem is if the additional requirement is
the purity of a second component in the distillate. This requires
! = dimensionless variable associated with liquid flow rate
= dimensionless variable associated with vapor flow rate
y
a trial and error solution. Similar difficulties arise if conditions T = dimensionless variable associated with temperature
= fraction of zero elements in a block
(1) or (2) arc altered.
T h e method is easily extended to multi-feed situations. The
8 = number of computation operations per element
c, 6 = small positive number
introduction of side streams does introduce some problems.
Superscripts
If the amount of the side stream withdrawn and the location of
the sidestreani are specified, then there are no complications. * = the previous iteration
s = the sth iteration
Otherwise, the problem appears to be a trial and error cal-
culation. Subscripts
i , g , k = indicate the number of the stage
Concluswnr g = i - 1
T h e chief restriction on the method is that one is virtually k = i + l
i,q,p = the number of the component
committed to systems where the number of trays and the loca- n = the nth component
tion of the fced tray are specified. One has the additional
restriction that one must have C+2 equations for each stage. References
Almost all other s ecifications can be treated by substitution ( 1 ) Amundson, W. R. and Pontinen, A. J., IEC, SO, 730 (1958).
of the specification f!r the enthalpy equation in the reboiler (2) Amundson, W. R., Pontinen, A. J. and Tiemey, J. W., AIChE
Jnurnal 5. 295 l l O 5 R ) .
and the condenser. For a total condenser the bubble Doint ( 3 ) %in;? y.-W- &d-Bruno J. A AIChE Journal 13. 556.
requirement ( j: Vij = 1) can be replaced. I t is also necessary
( 4 ) Thoma;, L. H.. “Ellietic h o b l e k s in Linear Dikerence Esuationr
0LeL.a .Network”, Rept. Watson Sci. Computing Lab., New York
(1Y4Y).
( 5 ) Bellman, R., Juncosa, M. and Kalaba, R., Rand Corp. Rept. No.
P.-2200, Santa Monica, Cahfornia (1961).
to modify the mass balance equations for the top stage (con- ( 6 ) Lee, S. E AIChE Journal, 14, 977 (1968).
(7) Lee, S. E:: AIChE Journal,. 14, 490 (1968).
denser) somewhat. An exception to the rule is if two speci- ( 8 ) Gentry J. W. Ph.D. Theas Univ. of Texas at Austin (1969).
( 9 ) Greensiadt, J.,’Bard, Y. and’Morse, B., IEC, SO, 1644 (1958).
fications are given for the distillate (three in the case of a total (10) Bonner J. S Proc. Am. Petrol. Inst 36, Sect. 111 238 (1956).
condenser) which cannot be related to the bottoms. For example (11) Thiele ’E. W.’ and Geddes R. L IEt“, 23, 289 (1935).
(12) Gentj, J. W., Unpublishid Da& 1970.
if both X D , A and X D , B are specified, then one cannot directly
replace the enthalpy equation for the reboiler. Manuscript received July 28, 1969; accepted February 4, 1970.
The major advantages of the method are
1) the relative speed of the calculation
2) the relative insensitiveness to the initial estimates. * * *
The Canadian Journal of Chemical Engineering, Vol. 48, Augurt, 1970 455