Computers and Chemical Engineering 25 (2001) 1141– 1152

www.elsevier.com/locate/compchemeng

A direct method for incorporation of tray-efficiency matrix in

simulation of multicomponent separation processes
D.P. Rao *, C.V. Goutami, S. Jain
Department of Chemical Engineering, Indian Institute of Technology, Kanpur 208016, India

Received 14 July 2000; received in revised form 5 February 2001; accepted 6 February 2001

Abstract

Both experimental and estimated multicomponent tray efficiencies are known to become unbounded and exhibit bizarre
variation from tray to tray in a distillation column even for ideal systems. Besides lacking the physical significance, they pose
numerical difficulties in computation of the component tray efficiencies from the point efficiencies whenever the latter assume
negative values (Lockett, M. J. (1986). Distillation tray fundamentals. Cambridge: Cambridge University Press). The use of the
Stefan–Maxwell approach for estimation of point efficiency leads to a matrix form. Recently, the tray-efficiency matrix for
well-mixed, dispersed and plug flow of liquid has been obtained in terms of point-efficiency matrix similar to binary systems.
However, there are no method to use directly the tray-efficiency matrix in simulation. Here, we present a method for direct
incorporation of the efficiency matrix in the well-known Naphtali– Sandholm method. We have studied the convergence
characteristics of the modified and the original methods for distillation, absorption and extraction columns. The results show the
modified one is superior to the original method. © 2001 Elsevier Science Ltd. All rights reserved.

Keywords: Simulation; Distillation; Absorption; Extraction; Separations; Tray-efficiency

Nomenclature

A matrix of order (2C + 1) × (2C + 1)

B matrix of order (2C + 1) × (2C + 1)
C matrix of order (2C + 1) × (2C + 1)
C number of components
EP point-efficiency matrix
Ep,ik element of point-efficiency matrix
Epi individual point efficiency of component i
Ej tray-efficiency matrix
Eji individual tray efficiency of component i
Ej,ik element of tray-efficiency matrix for jth tray
E%j,ik element of tray-efficiency matrix for jth tray with entrainment
e entrainment, kmol of liquid per kmol of vapor
H vapor enthalpy (kJ/kmol)
h liquid enthalpy (kJ/kmol)
I identity matrix of order (C −1)
K matrix of vapor – liquid equilibrium ratio
K distribution coefficient
K% modified distribution coefficient

* Corresponding author. Fax: + 91-512-590260.

E-mail address: dprao@iitk.ac.in (D.P. Rao).

0098-1354/01/$ - see front matter © 2001 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 9 8 - 1 3 5 4 ( 0 1 ) 0 0 6 8 6 - X
1142 D.P. Rao et al. / Computers and Chemical Engineering 25 (2001) 1141–1152

Kj,ik element of vapor– liquid equilibrium matrix for jth tray

L liquid flow rate (kmol/s)
li flow rate of component i in liquid stream (kmol/s)
Pe Peclet number
V vapor flow rate (kmol/s)
6i flow rate of component i in vapor stream (kmol/s)
6%i flow rate of component i in wet vapor stream (kmol/s)
x vector of mole fractions in the liquid phase
xji mole fractions in liquid phase leaving the tray j
y vector of mole fractions in the vapor phase
y* mole fraction of vapor in equilibrium with liquid at any point on the tray
yji mole fractions in vapor phase leaving the tray j
Y vector of wet vapor mole fractions
ki ratio of driving forces of component i and component C− 1 on a tray
ii defined by Eq. (28)

Subscripts
I for component i
ik for the pair i and k
j leaving the tray j

Superscripts
* equilibrium composition
% modified to account for entrainment

1. Introduction On the other hand, the multicomponent mass-trans-

fer theory suggests that the natural extension of the
In simulation of multicomponent distillation Murphree tray efficiency from binary to multicompo-
columns, the component tray efficiencies are used to nent system would take a matrix form (Taylor and
account for the departure from the equilibrium between Krishna, 1993). The diagonal elements reflect the con-
the phases. The component tray efficiencies, Ej,i, are tacting efficiency of the phases on the tray and the
defined as off-diagonal elements the multicomponent interaction
effects. For a C-component mixture, the tray efficiency
yj,i −yj + 1,i matrix, Ej of stage j may be defined as
Ej,i = for i =1, 2, …, C (1)
y*j,i −yj + 1,i
yj − yj + 1 = Ej (y*−
j yj + 1) (2)
where y is mole fraction of vapor and the subscripts j
and i denote tray number and component, respectively. where
But, these tray efficiencies suffer from two serious
drawbacks. First, they may become unbounded and y= [y1, y2, …, yC − 1]T (3)
exhibit bizarre behavior even for ideal systems (Kr- and
ishna, Martineez, Sreedhar and Standart, 1977 Hol-
land, 1981; Chan and Fair, 1984; Rao, Prem Kumar,
Æ E11 E12 … E1C − 1 Ç
Pandit and Das, 1995). As a consequence, the magni- Ã E21 Ã
tudes of Ej,i do not always reflect the effectiveness of E22 … E2C − 1
Ej = Ã Ã (4)
contact and flow pattern of the phases on a stage unlike à … … … … Ã
the tray efficiencies in the case of binary systems. ÈEC − 11 EC − 12 … EC − 1C − 1 É
Second, the methods — for evaluation of Ej,i from the
corresponding component point efficiencies — involve Note that Ej is of the order of (C− 1)×(C−1) and
very restrictive assumptions and, even more serious, fail the subscript j on its elements is omitted for clarity. If
whenever the latter assume negative values (Lockett, the multicomponent effects are absent, Ej reduces to a
1986). diagonal matrix and to a scalar for the binary case.
 D.P. Rao et al. / Computers and Chemical Engineering 25 (2001) 1141–1152 1143

The well-known Colburn method, used to account Mj,i = lj − 1,i + 6%j + 1,i − (1+ sj )lj,i − (1+ Sj )6%j,i + fj,i =0
for entrainment for binary systems, is valid only for a
for j=1, 2, …, N; i= 1, 2, …C −1 (5)
well-mixed liquid flow and it overestimates the deleteri-
ous effect of entrainment for the dispersed and plug where l and 6% are the component flow rates of the
flow of liquid. Mohan, Rao and Rao (1983) have liquid and wet-vapor streams and sj and Sj are the
modified the AIChE method for estimating tray effi- fraction of liquid and wet-vapor as side streams drawn
ciencies to account for entrainment for binary systems from the corresponding streams leaving the jth tray,
(AIChE, 1958). Recently, Rao et al. (1995) presented a and fj,i is component feed rate. The equilibrium rela-
method, based on the Stefan– Maxwell approach, to tions can be expressed as (Rao et. al., 1995)
estimate tray-efficiency matrix from point-efficiency
E%j (K%j − Yj )= (Yj − Yj + 1) (6)
matrix accounting for different liquid flow patterns as
well as entrainment. This method may be regarded as a where
proper generalization of the ‘AIChE’ method (devel-
oped based on the binary systems) to multicomponent Y= [Y1, Y2, …, YC − 1]T (7)
systems. The diagonal elements are positive bounded (Kj,i + ej )
and reflect the rate of interfacial transport on the tray K%j,i = (8)
1+ ej
as in the case binary systems. The off-diagonal elements
reflect the diffusional and thermodynamic interactions yi + exi
Yi = (9)
of the components (Rao et al.,1995). Thus efficiency 1+ e
matrix overcomes the drawbacks mentioned earlier. and x is mole fraction of liquid and y is that of dry
It is obvious that the use of the component point and vapor; e is the moles of liquid entrained per mole of dry
tray efficiencies is redundant if the tray-efficiency ma- vapor; K and K% are the distribution coefficients based
trix can be used directly in the algorithm employed for on mole fractions of dry vapor and wet vapor, respec-
the simulation. The algorithm based on the Naphtali– tively. Alternatively, Eq. (6) can be written in terms of
Sandholm method (Naphtali and Sandholm, 1971) or a discrepancy function, Q, as
one of its variants makes use of the component tray
efficiencies. At present, there is no method that incor- Qj,i =
6%j,i C − 1
− %
 C−1
% (E%j,imK%j,ml )
lj,i C − 1
− % (li,l − E%j,il )
porates directly the tray-efficiency matrix. Vj l = 1 m=1 Lj l = 1
However, one can use the ‘Rate-Based’ method pro- 6%j + 1,l
posed by Krishnamurthy and Taylor (1985), which =0 for i =1, 2, …C − 1; j =1, 2, …, N
Vj + 1
eliminates the need for the concept of tray efficiency. (10)
Recently, Kooijman and Taylor (1995) extended the
method to account for the cross flow of liquid on the The Cth equilibrium relation is required to get the
tray. However, the computational overheads are very requisite number of discrepancy functions to apply the
high. The latter work discusses at length the strengths Naphtali–Sandholm method. We have chosen to for-
and weaknesses of the methods used in the simulation mulate the discrepancy function Qj,C using the relation,
of multicomponent separation processes. Y*j,C = K%j,Cxj,C (11)
In this paper, we present a modified Naphtali – Sand-
holm method in which tray-efficiency matrix is incorpo- so as to include K%j,C in the model equations. Expressing
rated retaining the block tridiagonal structure of the the mole fraction of Cth component in terms of mole
Jacobian and the same sparsity structure of the block fractions of (C− 1) components and on manipulation,
we get
!    "
matrices of the original method.
C−1 C−1
6%j,m 6%j + 1,m 6%j + 1,i
Qj,C = 1− % % E%j,im
−1
− +

 
i=1 m=1 Vj Vj + 1 Vj + 1
C−1
lj,i
− K%j,C 1− % =0 (12)
2. Modified Naphtali – Sandholm method i=1 Lj
The enthalpy balance can be written as
The modified method makes use of mole fraction and
C C C
flow rate of wet-vapor instead of those of dry vapor
mj = % hj − 1,i lj − 1,i − (1+ sj ) % H%j,i 6%j,i + % hfj,i fj,i
(free from entrainment) in mass, energy balances and i=1 i=1 i=1
equilibrium relations. The total flow rate of wet vapor C
is denoted by V% and its mole fraction by Y. Consider a + % H%j + 1,i 6%j + 1,i + qj = 0 for j =1, 2, …, N
i=1
column consisting of N stages. The model equations for
(13)
the process can be formulated as follows. For the jth
tray, the component material balances are where
1144 D.P. Rao et al. / Computers and Chemical Engineering 25 (2001) 1141–1152

C
1 non-zero elements; 0C is null vector of order C and XC
H%j = % (yj,i Hj,i + ej xj,i hj,i ) (14)
i=1 1 +ej is a vector of order C with non-zero elements.
A close examination reveals that the fill-in of subma-
Eqs. (5), (10), (12) and (13) form a set of N(2C + 1) trices Aj, Bj and Cj are identical to those of the Naph-
equations in terms of N(2C +1) variables. The above tali–Sandholm method if we account for the reordering
equations can be written as of the independent variables. The elements of Aj are
F(X) = 0 (15) identical to those of the Naphtali–Sandholm method,
whereas the elements of Bj and Cj are different. The
where elements of these matrices are given in Appendix A.
F = (F1, F2, … FN )T (16) The non-zero elements of the submatrices were evalu-
ated from the analytical derivatives.
Fj = (Mj,1, …, Mj,C, Qj,1, …, Qj,C, xj ) T
(17)
X = (X1, X2, …, Xj, …, XN )T (18)
3. Simulation
Xj = (lj,1, …, lj,C, 6j,1, …, 6j,C, Tj )T (19)
3.1. Code
On lineralization we obtain
DX = − J − 1F (20) A computer code based on the Naphtali–Sandholm
method was available, which has been tested extensively
where and used to simulate binary and multicomponent distil-
(F lation and absorption columns. In this code, we have
J= (21) taken the advantage of the sparsity of the submatrices
(X
in evaluating the correction vector, DX. The results of
The Jacobian is a block tridiagonal matrix and can simulations obtained using this code have been reported
be represented as earlier (Rao et al., 1995). This code was adopted for the
proposed method. Only the subroutines for evaluation
of the discrepancy functions of the equilibrium relations
and the elements of the Jacobians were appended to the
existing code. This method of simulation is hereafter
(22) referred to as modified method. To estimate the effi-
ciency, the column diameter and liquid path length are
required. The column diameter and liquid path length
are found as outlined by Fair (1963). The details of the
method of estimation of efficiencies are available else-
where (Pandit, 1993; Rao et al., 1995).

where A, B and C are submatrices of the order (2C + 3.2. Initialization

1) × (2C + 1). The rest of the blocks, not shown in Eqn
22, are null matrices. The submatrices Aj, Bj and Cj can We have employed a simple method to initialize the
be represented as given below. iteration variables. These variables were initialized by
setting both the vapor and liquid compositions and
Æ ICC 0CC 0C Ç
à à temperature on all trays to feed composition and its
Aj = Ã 0CC 0CC 0C Ã (23) saturation temperature, respectively, in the case of dis-
à T Ã
È XC 0TC XÉ tillation. The phase flow rates were set assuming
equimolar overflow on the trays. For absorption and
Æ −(1 +sj )ICC −(1 +sj )ICC 0CC Ç extraction the gas- or raffinate-phase flow rates, compo-
à à sitions and temperatures were set to that of the feed.
Bj = Ã XCC XCC XC Ã (24)
à à Those of liquid or extract phase were set to that of the
È XTc XTc X É
solvent. This procedure of initialization gave conver-
gence in all cases.
Æ0CC ICC 0C Ç
à Ã
Cj = Ã 0CC XCC 0C Ã (25) 3.3. Method of con6ergence
à T Ã
È 0c XTc XÉ
First, the convergence was achieved considering the
where ICC and 0CC are identity and null matrices of trays are ideal i.e. by setting Eij equal to one for i=j,
order C×C; XCC is a matrix of order C × C with otherwise zero. The profiles, thus obtained, were used
 D.P. Rao et al. / Computers and Chemical Engineering 25 (2001) 1141–1152
to initialize the iteration variables for the well-mixed
liquid flow as well as to estimate the column diameter
and tray efficiencies. The tray efficiencies were esti-
mated in each iteration. On achieving convergence,
these profiles were similarly used for the entrainment of
0.1 and for other entrainment rates. The similar proce-
dure was for the dispersed and plug flow.
We have employed the convergence criterion recom-
mended by Seader and Henley (1998) to terminate the
iterations. The enthalpy balances were scaled by divid-
ing the sum of the enthalpies of incoming stream on the
trays. In addition, an additional constraint that the
temperature difference between consecutive iteration be
less than 0.01 K was imposed. A step limiting factor of
1.0 and 0.5 was used in the Newton– Raphson method
to examine the convergence characteristics.
4. Results and discussion
To validate and to examine convergence characteris-
tics of the proposed method, we have selected the
following systems.
Distillation
1. Ideal mixtures
– n-pentane, n-hexane and n-heptane;
– methane, ethane, propane, butane and
pentane.
2. Nonideal mixtures
– methanol, ethanol and water;
– methanol, iso-propanol and water;
– methyl acetate, chloroform and benzene;
– ethanol, tert-butanol and water; and
– benzene, n-hexane, methyl cyclopentane and
ethanol.
Absorption
1.
– methane, ethane, propane, butane and pentane
(feed gas); n-tridecane (solvent).
Liquid extraction
1.
– Benzene, n-hexane-cyclohexane (feed); sul-
folane (solvent).
The specifications of the problems employed in the
simulations are listed in Table 1. We have carried out
the simulation by the original Naphtali– Sandholm
method (using component-tray efficiencies) and the
proposed method (using tray-efficiency matrix) for the
well-mixed, dispersed and plug flow of liquid and well-
mixed vapor phase on the trays. For extraction, we
have not estimated the efficiencies; rather we have set
the component efficiencies and diagonal elements of
efficiency matrix to arbitrary values and off-diagonal
elements to zero to examine the performance of the
methods.
Since the original and proposed methods differ only
1145
in their solution procedure, it is expected that the
results obtained by them should be identical though the
elements of Jacobians in these methods are different.
For a few typical cases, the product compositions ob-
tained are given in Table 2. It can be seen that the
compositions of products are the same. The differences
in the mole fractions are in the fourth decimal places
and product temperatures are within 0.01 K. Typical
temperature profiles obtained with different entrain-
ment rates for plug flow case is shown in Fig. 1. It
shows the product temperatures lower, and hence the
extent of separation is less, in the presence of
entrainment
The convergence characteristics of the modified and
original methods for a step size of 1.0 are given in
Table 3. For the case of ideal trays, the number itera-
tions required for both methods are the same. A mo-
ment of reflection shows that both methods are
identical for the ideal trays. The convergence was
achieved for all except for problem 7. The convergence
is rather fast. Problem 7 deals with distillation of wide
boiling 5-component mixture of hydrocarbons. The
variation of k-values with temperature was large. This
led to divergence. In the case of problem 8 (absorption
of light hydrocarbons into tridecane), the modified
method diverged for the plug flow whereas the original
method diverged for all cases.
There are two cases worth noting — plug flow of
Problem 1 and of Problem 3. These are identified by *a
and *b in Table 2. In the first case (*a), we have
examined the cause for as the slow convergence (63
iterations). The component efficiency of hexane on the
seventh tray widely fluctuated during the iterations as
shown in Fig. 1. In fact, it was not a true convergence.
At this iteration, the sum of the discrepancy functions
happened to satisfy the convergence criteria. In the
other case, at the iteration from where it started diverg-
ing, we found that the component efficiencies on the
ninth tray assumed very large positive and negative
values. These efficiencies are shown Fig. 2. These obser-
vations indicate how the use of component efficiencies
can pose problems.
The behavior of those component efficiencies shown
in Figs. 1 and 2 warrants a comment. Even though the
variation of the elements of tray-efficiency matrix with
tray number is smooth, the component efficiencies ex-
hibit abrupt changes on some trays (Rao et al., 1995).
A similar behavior has also been reported by Chan and
Fair (1984) among others. To understand the cause for
this trend, we need to know how the component effi-
ciencies are deduced from the tray-efficiency matrix.
They are deduced as given below.
C−1
(y*jk −yj + 1,k )
Eji = % Ej,ik for i= 1, 2, …, C−1
k=1 (y*−
ji yj + 1,i )
(26)
1146 D.P. Rao et al. / Computers and Chemical Engineering 25 (2001) 1141–1152

and It was found that, in some cases, the term in the

C−2 denominator of Eq. (26) became very small as y*ji and
EjC − 1 + % ii Eji yj + 1i are nearly equal. It is well known that such a
i=1
Ejc = (27) subtraction leads to large numerical error. This led to a
C−2
1+ % ii large negative or positive component efficiencies, which,
i=1 in turn, led to divergence.
where We employed a step size of 0.5 and studied the
convergence characteristics. The results are given in
(y*−y j + 1,i )
ii = ji
(28) Table 4. The convergence was obtained for ideal trays
(y*jC − 1 − yj + 1,C − 1) for all problems though it took more iterations (1.5–3
Table 1
Specifications of problems employed in simulation

Problem number Number of Feed stage Specifications Feed condition, Method of Diameter (m)
stages flow rates estimation of K
(kmol/h)

1 10 5 R= 1.6 Liquid 493.9 K Polynomial of K 0.92

B= 50.0 Liquid 493.9 K Polynomial of K
P= 20.4 atm n-Pentane 30.0
Distillation n-Hexane 35.0
n-Heptane 35.0
2 22 12 R= 3.0 Liquid 340.0 K Wilson 1.22
B= 50.0 Methanol 30.0
P= 1.0 atm Ethanol 50.0
Distillation Water 20.0
3 22 10 R= 5.0 Liquid 347.0 K Wilson 0.92
B= 50.0 Methanol 33.33
P= 1.0 atm Iso-propanol 16.67
Distillation Water 33.33
4 20 11 R= 2.0 Liquid 335.0 K UNIQUAC 1.22
B= 50.0 Methyl acetate 20.0
P= 1.0 atm Chloroform 50.0
Distillation Benzene 30.0
5 20 11 R= 2.0 Liquid 352.25 K Wilson 1.22
B= 50.0 Ethanol 40.0
P= 1.0 atm tert-Butanol 20.0
Distillation Water 40.0
6 11 8 R= 2.0 Liquid 347.0 K UNIQUAC 1.52
B= 50.0 Bezene 40.0
P= 1.0 atm n-Hexane 20.0
Distillation Methyl 20.0
cyclo-pentane
Ethanol 20.0
7 13 7 R= 1.89 Liquid 317.55 K Polynomial of k 2.13
B= 270.0 Methane 160.0
P= 27.0 atm Ethane 370.0
Distillation Propane 240.0
Butane 25.0
Pentane 5.0
8 6 1, 6 P= 27.0 atm Liquid (l) 305.0 K Polynomial of k 1.52
Absorption Vapor (v) 313.55 K
Oil Methane 0.0 (l), 160 (v)
Ethane 0.0 (l), 370 (v)
Propane 0.0 (l), 240 (v)
Butane 0.05 (l), 25 (v)
Pentane 0.78 (l), 5 (v)
Tridecane oil 164.2 (l), 0 (v)
9 10 1, 10 P= 1.0 atm Liquid (l) 348.0 K UNIQUAC –
Liquid–liquid Liquid (v) 348.0 K
Extraction Benzene 0.0 (l), 26.52 (v)
Hexane 0.0 (l), 31.22 (v)
Cyclohexane 0.0 (l), 42.26 (v)
Sulfolane 166.6(l), 0.0 (v)
 D.P. Rao et al. / Computers and Chemical Engineering 25 (2001) 1141–1152 1147

Table 2
Product mole fractions and temperatures obtained by modified and original methodsa

Problem number Stage e Flow pattern i Top product composition Bottom product composition TC, TB (K)

Comp* Matrix** Comp Matrix Comp Matrix

2 Real 0.1 w 1 0.5577 0.5577 0.0423 0.0423 345.54 345.54

2 0.3860 0.3857 0.6140 0.6143 351.36 351.36
4 Real 0.2 w 1 0.2981 0.2980 0.1019 0.1020 340.16 340.17
2 0.6191 0.6188 0.3809 0.3812 345.21 345.2
6 Ideal 0.0 1 0.2967 0.2967 0.5033 0.5033 347.76 347.76
2 0.2849 0.2849 0.1151 0.1151 349.66 349.66
3 0.2558 0.2558 0.1442 0.1442
6 Real 0.3 p 1 0.2917 0.2920 0.5083 0.5080 347.69 347.69
2 0.2892 0.2892 0.1108 0.1108 349.73 349.73
3 0.2593 0.2593 0.1407 0.1407
7 Ideal 0.0 1 0.0002 0.0002 0.0923 0.0923 262.66 262.66
2 0.0000 0.0000 0.0184 0.0184 344.35 344.35
3 0.0054 0.0054 0.8784 0.8784
4 0.3013 0.3013 0.0012 0.0012
7 Real 0.2 d 1 0.0000 0.0000 0.0926 0.0926 262.23 263.23
2 0.0000 0.0000 0.01852 0.01852 343.81 343.81
3 0.0072 0.0073 0.8748 0.8746
4 0.3019 0.3019 0.0000 0.0000
8 Ideal 0.0 1 0.2762 0.2762 0.0284 0.0284 338.62 338.62
2 0.0001 0.0001 0.3816 0.3816 334.62 334.62
3 0.2022 0.2022 0.3066 0.3066
4 0.0024 0.0024 0.0553 0.0553
5 0.0004 0.0004 0.0129 0.0129
8 Real 0.0 p 1 0.2768 0.2768 0.0285 0.0285 338.65 338.69
2 0.0001 0.0001 0.3804 0.3804 334.66 334.65
3 0.2011 0.2012 0.3076 0.3074
4 0.0022 0.0022 0.5534 0.0553
5 0.0004 0.0004 0.0129 0.0129

a
*, Using component tray efficiencies; **, using tray efficiency matrix.

times) compared with the step size of 1.0. The proposed proposed method, the CPU times for Problem 1 are
method converged for all cases except for the plug-flow 0.019, 0.028, 0.027, 0.028 s for equilibrium, well-mixed,
of Problem 7 and 8. The original method fared better dispersed and plug flow, respectively. The rest of the
than that proposed for Problem 8, but it was worse off
for Problem 7.
We have further explored to improve the conver-
gence characteristics. In obtaining the matrix efficien-
cies of (C− 1)×(C −1) order, the mole fraction of last
component was treated as a dependent variable. The
components were ordered with decreasing volatility.
Therefore, the mole fraction of the last component
became the one, which was treated as the dependent
variable. We interchanged the last component with the
one of the middle components. Such interchange did
not affect the final component and temperature profiles
as should be. However, it improved the convergence
characteristics. These results are given in Table 5 for
problems 7 and 8. It could be seen that the improve-
ment was marginal for step size of 1.0. But, there was a
marked improvement with step size of 0.5. Both meth-
ods converged for all cases except for one case marked
*c.
We ran these codes on a Silicon Graphics Machine:
Model O200-R10000 CPU with 256 MB RAM. For the Fig. 1. Variation component efficiencies of hexane with iteration.
 1148
D.P. Rao et al. / Computers and Chemical Engineering 25 (2001) 1141–1152
Table 3
Convergence Characteristics of the modified and original methods with step size of 1.0a

Problem c Number of iterations

Matrix Component

Ideal Well-mixed flow Dispersed flow Plug flow Well-mixed flow Dispersed flow Plug flow

e e e e e e e

0.0 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3

1 5 4 3 3 3 3 3 4 4 4 5 7 23 3 4 2 2 2 3 3 3 4 63*a 7 13
2 4 2 3 3 3 2 2 3 3 3 3 3 5 2 3 3 3 2 3 3 3 2 3 3 3
3 7 2 2 3 4 2 2 3 4 3 4 4 5 3 3 3 4 2 3 3 4 4 F*b 4 10
4 4 2 4 4 4 2 3 3 3 3 4 4 4 2 2 2 2 2 2 2 2 2 3 3 2
5 3 1 3 3 3 1 3 2 2 1 3 3 4 1 2 2 2 2 2 2 2 2 2 2 2
6 3 2 4 4 4 2 3 3 2 2 5 6 7 2 2 2 2 2 2 2 2 2 2 2 2
7 F F F F F F F F F F F F F F F F F F F F F F F F F
8 12 3 18 17 16 7 17 16 15 F F F F F F F F F F F F F F F F
9 4 2 (Efficiency 0.6) 2 (Efficiency 0.3)

a
F, diverged.
 D.P. Rao et al. / Computers and Chemical Engineering 25 (2001) 1141–1152
Table 4
Convergence Characteristics of the modified and original method for step size of 0.5a

Problem c Number of iterations

Matrix Component

Ideal Well-mixed flow Dispersed flow Plug flow Well-mixed flow Dispersed flow Plug flow

E e e e e e e

0.0 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3

1 13 9 6 6 5 8 8 9 9 11 12 15 22 7 6 6 5 8 7 7 6 9 15 7 26
2 11 4 5 5 5 3 5 5 5 5 8 8 8 4 6 5 5 3 6 5 5 5 5 5 5
3 15 4 4 5 7 4 5 6 7 8 8 8 9 4 6 4 6 5 6 5 7 9 13 12 21
4 11 3 8 8 8 2 5 5 5 4 9 9 10 3 5 5 5 2 5 5 5 4 8 5 5
5 7 2 3 3 3 1 3 4 3 2 4 6 7 2 5 3 3 2 3 4 3 2 4 5 3
6 9 3 8 8 8 2 8 7 7 3 10 12 14 3 8 7 7 2 8 7 7 3 8 7 7
7 7 6 4 5 5 5 5 6 7 F F F F 13 F F F F F F F F F F F
8 9 12 6 18 17 3 18 5 16 F F F F 16 34 20 11 2 29 11 16 20 17 F F
9 8 3 (Efficiency 0.6) 3 (Efficiency 0.3) 3 (Efficiency 0.6) 3 (Efficiency 0.3)

a
F, diverged.

1149
 1150
D.P. Rao et al. / Computers and Chemical Engineering 25 (2001) 1141–1152
Table 5
Effect of interchange of components on the convergencea

Problem c Number of iterations

Matrix Component

Ideal Well-mixed flow Dispersed flow Plug flow Well-mixed flow Dispersed flow Plug flow

E e e e e e e

0.0 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3

Step size 1.0

7 F F F F F F F F F F F F F F F F F F F F F F F F F
8 12 10 17 7 19 10 17 F 20 10 17 19 17 F F F F F F F F F F F F
Step size 0.5
7 7 7 7 7 7 5 4 6 6 9 10 11 11 6 F*c 7 8 3 7 3 7 10 6 7 8
8 9 3 10 9 9 3 10 9 9 4 13 18 12 37 13 28 13 5 18 14 15 6 13 28 6

a
F, diverged.
 D.P. Rao et al. / Computers and Chemical Engineering 25 (2001) 1141–1152 1151

(Mj,i
= (1+ sj )li,k for i, k= 1, 2, …, C (A.1)
(Mj,k
(Mj,i
= (1+ Sj )li,k for i, k= 1, 2, …, C (A.2)
(6j,k
(Mj,i
=0 for i =1, 2, …, C (A.3)
(Tj

  
Equilibrium relations.
(Qj,i C−1 C−1
Ljli,k − lj,l
=− % % E%j,im
−1
K%j,ml
(lj,k
 
l=1 m=1 L 2j
C−1 C−1
(K%j,ml lj,l
− % % E%j,im for
l=1 m=1 (lj,k Lj
i= 1, 2, …, C− 1; k=1, 2, …, C
(A.4)
(Qj,i (Vjli,k − 6%j,i )
= for i= 1, 2, …, C− 1;
(6%j,k V 2j

 
k= 1, 2, …, C (A.5)
(Qj,i C−1 C−1
(K%j,ml lj,i
Fig. 2. Variation of component efficiencies with tray number. =− % % E%j,im
−1
for
(Tj l=1 m=1 (Tj Lj
problems are within 90.005. For the original method i=1, 2, …, C− 1 (A.6)
too, the CPU times are in the same range. It seems that,

    
The Cth equilibrium relation.
with the present day’s machines, the robustness of the
method rather than CPU time is an important (Qj,C C−1
Ljli,k − lj,l (K%j,C C−1
l
= − K%j,C % − 1− % j,i
consideration. (lj,k i=1 L 2j (lj,k i = 1 Lj

 
for k= 1, 2, …, C (A.7)
(Qj,C C−1 C−1
(6jlk,m − 6%j,m )
5. Conclusions =− % % E%j,im
−1
for
(6%j,k i=1 m=1 V 2j

 
The Naphtali–Sandholm method has been modified k=1, 2, …, C (A.8)
to incorporate directly tray-efficiency matrix, which has (Qj,C (K% l C−1
a firmer theoretical base than the component tray effi- = − j,C 1− % j,i (A.9)
(Tj (Tj i = 1 Lj
ciencies. The convergence characteristics of the pro-
posed method are superior to the original method. If Enthalpy balances.
divergence is encountered, it can be overcome by inter- (mj
change of the order of the components. The modified = − (1+ sj )hj,k for k= 1, 2, …, C (A.10)
(lj,k
method renders the use of component efficiencies
redundant. (mj
= − (1+ Sj )H%j,k for k= 1, 2, …, C (A.11)
(6%j,k
(mj C

(h C

= − (1+ sj ) % lj,i j,i − (1+ Sj ) % 6%j,i
(H%j,i  
Appendix A. Elements of the Jacobian (lj,k i=1 (Tj i=1 (Tj
(A.12)
The elements of the submatrices Aj, Bj, and Cj of the
Elements of matrix Cj
Naphtali– Sandholm method are available elsewhere
Component material balances.
(Henley and Seader, 1981). Note that a slight reorder-
ing of variables was done in this work. The submatrices (Mj,i
=0 for i =1, 2, …, C (A.13)
of Aj of the proposed method are identical to those of (lj + 1,k
the original method. Therefore, they are not given here.
(Mj,i
The elements of Bj, and Cj of the proposed method = li,k for i= 1, 2, …, C (A.14)
differ from the original method and are given below. (6%j + 1,k
Elements of matrix Bj (Mj,i
=0 for i =1, 2, …, C (A.15)
Component balances. (Tj + 1,k
1152 D.P. Rao et al. / Computers and Chemical Engineering 25 (2001) 1141–1152

References
Equilibrium relations.
AIChE (1958). Bubble tray design manual. New York: American
(Qj,i Institute of Chemical Engineers.
=0 for i =1, 2, …, C − 1;
(lj + 1,k Chan, H., & Fair, J. R. (1984). Prediction of point efficiencies on sieve
trays. 2. Multicomponent system. Industrial Engineering and Chem-


k =1, 2, …, C (A.16) ical Process Design De6elopment, 23, 820 – 827.
(Qj,i C−1
= − % (li,l −E%j,il )
Vj + 1ll,k −6%j + 1,l n for
Fair, J. R. (1963). Tray hydraulics — perforated trays. In B. D. Smith,
Design of equilibrium stage processes. New York: McGraw-Hill.
(6%j + 1,k l=1 V 2j + 1 Henley, J. E., & Seader, J. D. (1981). Equilibrium stage separation
operations in chemical engineering. New York: Wiley.
i= 1, 2, …, C−1; k =1, 2, …, C (A.17) Holland, C. D. (1981). Fundamentals of multicomponent distillation.
New York: McGraw-Hill.
The Cth equilibrium relation. Kooijman, H. A., & Taylor, R. (1995). Modelling mass transfer in
multicomponent distillation. The Chemical Engineering Journal, 57,
(Qj,C 177 – 188.
Krishna, R., Martineez, H. F., Sreedhar, R., & Standart, G. L. (1977).
(6%j + 1,k
   
Murphree point efficiencies in multicomponent mixtures. Transac-
C−1 C−1
6j + 1lk,m −6%j + 1,m ) tions of Institute of Chemical Engineers, 55, 178 – 183.
= % % E%j,im
−1
Krishnamurthy, R., & Taylor, R. (1985). A nonequilibrium model of

 n
i=1 m=1 V 2j + 1 multicomponent separation processes. I. Model development and
6j + 1lk,i −6%j + 1,i ) method of solution. American Institute of Chemical Engineering
× for k =1, 2, …, C Journal, 31, 449 – 456.
V 2j + 1 Lockett, M. J. (1986). Distillation tray fundamentals. Cambridge:
(A.18) Cambridge University Press.
Mohan, T., Rao, K. K., & Rao, D. P. (1983). Effect of vapor
(Qj,i maldistribution and entrainment on tray efficiency. Industrial
=0 for i =1, 2, …, C (A.19) Engineering and Chemical Process Design De6elopment, 22, 380–
(Tj + 1 385.
Naphtali, L. M., & Sandholm, D. P. (1971). Multicomponent separa-
Enthalpy balances. tion calculations by linearization. American Institute of Chemical
Engineering Journal, 17, 148 – 153.
(mj Pandit, P.( 1993). Tray efficiencies in multicomponent distillation and
=0 for k =1, 2, …, C (A.20) absorption. M.Tech. Thesis, Indian Institute of Technology, Kan-
(lj + 1,k pur, India.
Rao, D. P., Prem Kumar, R. S., Pandit, P., & Das, T. C. T. (1995).
(mj Multicomponent tray efficiencies accounting for entrainment. The
= H%j + 1,k for k = 1, 2, …, C (A.21)
(6%j + 1,k Chemical Engineering Journal, 57, 237 – 246.
Seader, J. D., & Henley, J. E. (1998). Separation process principles. New
York: Wiley.
(mj C
(H%j + 1,i
= % 6%j + 1, j (A.22) Taylor, R., & Krishna, R. (1993). Multicomponent mass transfer. New
(Tj + 1 i = 1 (Tj + 1 York: Wiley.