1062 II / DISTILLATION / Modelling and Simulation

some organic chemicals. In fact, some thermal de- helpful to consider properties other than the boiling
compositions have been shown to be autocatalytic. It point. For example, if a mixture of intermolecularly
therefore may be well to consider Sash distillation of bound substances is to be separated by distillation,
the crude product prior to subjecting it to fractional their partition is likely to be more difRcult than the
distillation to remove trace non-volatiles and/or non- differentials between their boiling points would indi-
volatiles which would accelerate decomposition or cate. On the other hand, a mixture of alkanes may
lead to excessively high pot temperatures. In some well be more easily separable than comparison of
cases one might consider the addition of a stabilizing their boiling points would otherwise indicate. In any
agent to the pot to retard decomposition. case practice is necessary, both conducting distilla-
tions and selecting systems for distillation. Once ex-
perience has been gained it is satisfying to be able to
Closing Remarks rationalize the results of a fractionation in terms of
With all of the above having been stated, fractional physico-chemical principles. One positive note: since
distillation, particularly at reduced pressure, can be distillation does not result in loss of product, in the
viewed as an opportunity to see physical chemistry at worst case one can recombine all the fractions and
work. When selecting a system one hopes will result redistill using different conditions and, if necessary,
in satisfactory partition of components it will be a different system.

Modelling and Simulation

J. R. Haas, UOP LLC, Des Plaines, Illinois, USA (moles of any component in the feeds)
Copyright ^ 2000 Academic Press
" (moles of the component in the products)

Introduction
Feed enthalpy#Heat added
Rigorous computer modelling of all types of frac-
tionation columns has become a necessary part of the "Product enthalpy#Heat removed
development and design process. There are numerous
software products available to do these calculations. Figure 1 shows a complex column with one feed
An understanding of the basic mathematics used in and one side product. The top stage of the column is
these programmes is helpful to select, use and a partial condenser, with a vapour product, D, and
troubleshoot a column model. Explained here are the a liquid product, d. The reSux is the liquid, L0, and
basic equations, numerical and solution methods the reSux ratio is L0/(D#d). The bottoms product,
commonly used. B, leaves stage N#1, the reboiler. The stages are
numbered from the top, with the condenser as stage
0, the top tray in the column, stage 1, the bottom tray,
Stage and Column Models stage N, and the reboiler, as stage N#1.
A rigorous method describes a column as a group of An ideal or equilibrium stage is where vapour and
equations and is the mathematical engine to solve and liquid entering and leaving the stage are perfectly
satisfy these equations to calculate the operating con- mixed and there are no inhibitions to material trans-
ditions of the column. fer between the phases. The material and energy Sows
Column design and performance calculations pres- in and out of a simple stage, with no feeds or side
ent the column at steady state, that is, what enters the products, is stage j depicted in Figure 2, and i repres-
column matches what leaves it (material and energy ents the component number. Components are num-
balances), i.e.: bered from 1 to the last, C.
The enthalpy terms, Hj and hj, are molar enthalpies
(molar feed Sow rates) of the vapour and liquid leaving the stage, respective-
ly. These molar enthalpies are multiplied by the total
" (molar product Sow rates)
Sow rates, Vj and Lj, leaving the stage to give the total
energy leaving the stage in each phase.
(mass feed Sow rates)
The feed stage model (stage f in Figure 2) for an
" (mass product Sow rates) equilibrium stage assumes that the feed liquid mixes
 II / DISTILLATION / Modelling and Simulation 1063

Figure 1 Overall column model with external variables.

with the liquid entering the feed stage while feed feed stage pressure before the feed enters the column.
vapour mixes with vapour leaving the stage (though Regardless of whether the feed is subcooled liquid or
special consideration is made for the vapour feed at superheated vapour, or if true mixing occurs, the
the bottom of absorber/stripper columns). The distri- assumption of an equilibrium stage is maintained in
bution is found by an adiabatic Sash of the feed at the most rigorous methods.
1064 II / DISTILLATION / Modelling and Simulation

Figure 2 Model of stage variables.

Similar models are drawn for the bottom and top ables. Each stage is assumed to be at equilibrium (a
stages of any column, plus other equipment such as theoretical stage), though an efRciency can be applied
product withdrawal stages (stage p of Figure 2), in the equations.
pump-around returns and draws, and inter-reboilers The equations were Rrst referred to as the MESH
and inter-condensers. Since a reSux, reboiler vapour, equations by Wang and Henke (1966). The MESH
feeds, or returns are often subcooled, superheated, or acronym stands for:
very different in composition from the material on the
stage, the assumption of an equilibrium stage rapidly Material or Sow rate balance equations, both com-
becomes invalid. ponent and total.
Equilibrium equations including the bubble and
Equations of Distillation Modelling dew point equations.
Summation or Stoichiometric equations or com-
The basic equations below fully describe a distillation position constraints.
column. These equations deRne the overall column Heat or enthalpy or energy balance equations.
total material balances, energy balances, and product
compositions. Internal to the column, they describe
equilibrium conditions, internal (stage-to-stage) com- The MESH variables are referred to as state variables.
ponent and total material balances, and internal These are:
energy balances. The independent variables of a col-
umn are the product rates and compositions, internal E Stage temperatures, Tj
vapour and liquid rates and compositions, and stage E Internal total vapour and liquid rates, Vj and Lj
temperatures. Equilibrium constants, also called E Stage compositions, yji and xji, or instead, compon-
K values, and mixture enthalpies are dependent vari- ent vapour and liquid rates, vji and lji
 II / DISTILLATION / Modelling and Simulation 1065

The equilibrium equation is: material withdrawn, wpi, is subtracted from the com-
ponent material balance. By convention, material
yji"Kjixji or vji/Vj"Kjilji/Lj leaving a tray has a negative value and material enter-
ing a tray has a positive value.
The equilibrium constant or K-value, Kji, can be The total material balances are organized in the
a complex function itself, dependent on the composi- same manner as the component balances. The total
tions, xji and yji material balance for the simple stage of Figure 2 is:

Kji"Kji(Tj, Pj, xji, yji) Vj#1#Lj 1!Vj!Lj"0

\
The dependence of Kji on xji and yji often appears in The same convention applies to feed and product
the MESH equations. The component rates can also trays where the total Sow rate of a feed, Ff, is positive
be expressed in the terms of each other, giving: and the product, Wp, is negative.
The equilibrium equation and the composition
vji"lji(KjiVj/Lj)"ljiSji constraint are combined to get the bubble point equa-
tion:
and 1 C
Kjilji!1"0
Ci"1lji * i"1
lji"vji(Lj/KjiVj)"vjiAji
and the dew point equation:
KjiVj/Lj is termed the stripping factor, Sji, while
Lj/KjiVj is termed the absorption factor, Aji. 1 C
vji
The summation equation or composition con- * !1"0
C
i"1 ji
v i"1 Kji
straints simply states that the sum of the mole frac-
tions on each stage is equal to unity. For the liquid These, or some variation, are important in some
phase: methods to Rnd the stage temperature, especially for
C C
more narrow boiling mixtures.
xji!1"0 or lji/Lj!1"0 or The energy balance equations are required in any
i"1 i"1 rigorous method. In narrow-boiling mixtures, they
C inSuence the internal total Sow rates. In wide-boiling
yji/Kji!1"0 mixtures and in columns where there are great
i"1
heat effects (e.g. oil reRnery fractionators) they also
strongly inSuence stage temperatures. The overall
and for the vapour phase:
energy balance for a column with one feed and side
C C
product is:
yji!1"0 or vji/Vj!1"0 or
i"1 i"1 FHF!DHD!BhB!WHW#QR!QC"0
C
Kji xji!1"0 The enthalpy terms, H and h, are per mole of mixture.
i"1
Note that the enthalpies of the top and side products
For a simple column (single feed, no side products), are written so that a vapour or liquid enthalpy can be
the overall component balance equation is: substituted, depending on the phase of the product.
The energy balance for the simple stage, j, of Figure 2
fi!di!bi"0 is:

The component balance for the simple stage (no feed vj#1Hj#1#Lj h !VjHj!Ljhj"0
\1 j\1
or side product), j, of Figure 2, is:
The enthalpies (energy per mole) for each phase
vji#1#lji 1!vji!lji"0 are functions of temperature, pressure and com-
\ position:
The component balance for feed stage, f, of Figure 2
will add the liquid portion of the feed, lFi, while the Hj"Hj(Tj, Pj, yji)
vapour portion, vFi, is added to the component bal-
ance for stage f!1. For the product stage, p, the hj"hj(Tj, Pj, xji)
1066 II / DISTILLATION / Modelling and Simulation

For feed stages, side product stages, and stages with inefRciency in separation. The bypass method cannot
inter-condensers or inter-reboilers, additional terms be used on trays that have material leaving or enter-
are included in the energy balance equations. The ing from outside the column such as a feed tray,
energy balance for the reboiler is: product draw tray, pump-around return or draw tray,
or side-stripper return or draw tray. The bypass
LNhN!VN#1HN#1!BhN#1#QR"0 method will cause one of these trays to be out of mass
balance. Some of the trays adjacent to these trays are
and for a partial condenser with both vapour and also affected by these actions. In some columns, this
liquid products: eliminates a large number of trays and makes results
difRcult to apply.
V1H1!L0h0!dh0!DH0!QC"0 Caution then should be used in any choice of efR-
ciency. More often, it is usually best to perform the
Subcooling is accounted for in h0 (the enthalpy of the rigorous calculation using ideal stages and then apply
reSux, L0, and the liquid distillate, d). an overall column efRciency based on sound engineer-
Most computer simulations work with ideal stages ing judgement and experience to account for stage
but to characterize a stage for the deviation from nonideality, and calculate the number of actual trays
ideality or equilibrium, stage efRciencies are often or packing height.
used in some software. Commonly, a Murphree
vapour efRciency is used for each component, given
Rigorous Computational Methods
as:
Classi\cation of the Methods
yji!yji 1
EMVji" \ The rigorous methods can be divided into four basic
yHji !yji 1
\ classes. These are:

where yHji is what the vapour composition would be if

E The bubble point methods (BP)
the vapour were in equilibrium with the actual liquid
E The sum-rates methods (SR)
on the stage and yji and yji 1 are actual vapour com-
\ E The 2N Newton methods
positions. If the absorption factor is used, the vapour
E The global Newton or simultaneous correction
efRciency can be expressed in terms of variables al-
(SC) methods.
ready presented:
The BP methods get their name because the stage
vji!vji#1(Vj/Vj#1)
EMVji" temperatures are found by directly solving the bubble
(KjiVj/Lj)lji!vji#1(Vj/Vj#1) point equation. The BP methods generally work best
for narrow-boiling, ideal or nearly ideal systems;
A vaporization efRciency, Eji, based on the Murphree where composition has a greater effect on temper-
efRciency is deRned as: ature than the latent heat of vaporization.
The sum-rates (SR) method is suitable for model-
yji#1 ling absorbers and strippers with extremely wide-
Eji"EMVji#(1!EMVji)
Kjixji boiling systems, especially those with non-conden-
sables. In these columns, temperatures are the domi-
This can be used in the MESH equations to account nant variables and are found by a solution of the stage
for stage nonideality. This vaporization efRciency is energy balances. Compositions do not have as great
applied to the equilibrium constant, Kji, and appears an inSuence in calculating the temperatures as do
as the product EjiKji. The vaporization efRciency does heat effects or latent heats of vaporization.
solve a computational problem in placing an efRcien- The 2N Newton methods calculate temperatures
cy in the MESH equations. A major disadvantage of and total Sow rates together but compositions are
the vaporization efRciency is that it does vary with still calculated in a separate, dependent step. The
composition. Near the top of a high purity column, as name 2N Newton means that there are two equations
yji#1 and xji approach unity, Eji also approaches per stage for a total of 2;N functions and variables
unity, and so a vaporization efRciency does not truly per column solved simultaneously by a Newton}
reSect stage nonidealities. Raphson method. The 2N Newton methods have
Another efRciency method is the bypass method been shown to work well for wide-boiling mixtures
where some of the vapour Sow of a component enter- including reRnery fractionators, absorber-stripper
ing the stage is sent to the next stage to account for its columns and reboiled absorbers.
 II / DISTILLATION / Modelling and Simulation 1067

The Rrst three classes are referred to as equation Global Newton Methods
tearing or decoupling methods because the MESH
One group of methods that is very popular is the
equations are divided and grouped or partitioned and
global Newton methods, also called the simultaneous
paired with MESH variables to be solved in a series of
correction (SC) methods. A common one is that of
steps. The SC methods attempt to solve all of the
Naphtali and Sandholm (1971), but there are numer-
MESH equations and variables together. Additional
ous applications in the literature and global Newton
classes are:
methods have been extended to include additional
equations and variables for solving three-phase and
E Inside-out methods
reactive distillation columns.
E Relaxation methods
In the global Newton methods, all of the equations
E Homotopy}continuation methods
are solved together in a Newton}Raphson technique.
E Nonequilibrium models.
The methods vary in their choice of variables and
MESH equations for the Newton}Raphson calcu-
The relaxation, inside-out and homotopy}con- lation but none of the MESH equations are solved in
tinuation methods are extensions of whole or part of any separate step. In the BP, SR and 2N Newton
the Rrst four methods in order to expand the range of methods, the component balances and compositions
columns, and to solve difRcult systems or columns. lag the other MESH calculations (since K values and
The nonequilibrium models are rate-based or trans- enthalpies are generated using the compositions from
port phenomena-based methods that do away alto- the previous trial) and compositions of each compon-
gether with the ideal stage concept and eliminate any ent are calculated independently of the others MESH
use of efRciencies. They are best suited for columns variables. These are major disadvantages with highly
where a theoretical stage is difRcult to deRne and nonideal systems, where K values (especially activity
efRciencies are difRcult to predict or apply by any coefRcients
ji) and enthalpies are highly composition
means. dependent and where the composition of one com-
ponent cannot be readily decoupled from those of
Numerical Methods ^ The Newton^Raphson others. The global Newton method includes the com-
Technique ponent balances among the Newton}Raphson inde-
pendent functions and compositions join other
The MESH equations form a large system of inter-
MESH variables as independent variables.
related, nonlinear, algebraic equations. The mathe-
The global Newton methods are the most sensitive
matical method used to solve all or part of these
of the rigorous methods to the quality of the initial
equations as a group is the Newton}Raphson
values and often require initial values near the
method. An understanding of the numerical method
answer. This, and applying the methods to a full
is needed to understand the performance of all
range of column equipment and speciRcations, is
column methods. Detailed discussion of the Newton}
their greatest problem. Variations on global Newton
Raphson method and its variations can be found in
methods are used in the inside-out, relaxation,
Holland’s (1981) text.
homotopy and nonequilibrium methods, where their
The Newton}Raphson is an approximation tech-
power and reliability is extended.
nique. It assumes in the derivatives that the MESH
equations are linear over short distances and the
Inside-out Methods
slopes will point towards the answers. The MESH
equations can be far from linear and the predictions The inside-out algorithm has become one of the most
can take the next trial well off the curves, and move popular methods because of its robustness and its
away from the solution. In some rigorous methods ability to be applied to the solution of a wide variety
based on Newton}Raphson, a poor set of starting of columns. The inside-out concept was developed by
values can cause the calculation never to approach Boston (1980). Russell (1983) presented an inside-out
a solution. Also, the calculation can oscillate, with method that works well for many reRnery frac-
values swinging to either side of the solution. The tionators. The inside-out methods are now the
independent variables calculated in a trial need to methods of choice for mainstream column simulation
move the column to a solution. The software should and have displaced other methods.
include means to prevent or detect these problems In older methods, the MESH variables of temper-
and improve stability, e.g. by damping or limiting the atures, total Sow rates and component Sow rates are
change to the next set of variables. A Newton}Raph- the primary solution variables and are used to gener-
son method will normally take even steps toward the ate the K values and enthalpies from complex correla-
solution. tions. These methods update the MESH variables in
1068 II / DISTILLATION / Modelling and Simulation

an outer loop with the K values and enthalpies The simple K values used in the inside loop are
updated whenever the MESH variables change. The easily determined from:
inside-out concept reverses this by using the complex
K value and enthalpy correlations to generate para- Kji(simple)"Kbjji
Hji
meters for simple K value and enthalpy models. These
parameters are unique for each stage and become Simple models for the enthalpy of a phase are also
the variables for the outside loop. The inside loop used to reduce effects such as that caused by compo-
consists of the MESH equations and is a variation on nents moving past their critical conditions. Thus, the
other methods. In every step through the outside outside loop calculation consists of updating the
loop, the simple models are updated using MESH terms of the simple K value, activity and enthalpy
variables from the inside loop. This sets up the next models which are updated after each inside loop
pass through the inside loop. Since the K values and solution using the latest temperatures and composi-
enthalpies are simple, the inside loop works well for tions from the inside loop.
a wide range of mixtures and is little affected by the The inside loop consists of the actual calculation of
nonideality of mixtures or the quality of the initial the MESH variables using the simple K value and
values. enthalpy models. Boston initially used an inside loop
The outer loop K value model is based on a simple solution method similar to a bubble point method
composition-independent K method: and from that it may appear that the Boston method
is most appropriate for narrow-boiling mixtures.
ln Kbj"Aj#Bj(1/Tj!1/TH) However, the forcing style of the method also allows
it to work well for wide-boiling mixtures. The Boston
where TH is a reference temperature for the K value method works well for tall, high purity (superfrac-
correlation. Outer loop variables, Aj and Bj, are gen- tionator) type columns, but has been extended to
erated for each stage from a reference KbjRef of a com- absorbers, to three-phase distillation, and to reactive
posite component: distillation by using other arrangements of the MESH
equations.
C The Boston method includes a middle loop to allow
ln KbjRef" wi ln Kji(actual) for column speciRcations and constraints. The ar-
i"1
rangement of equations in the inner loop, where the
where the wi are weight factors. The temperatures solution of the MESH variables occur, may allow for
and compositions used to get the Kji(actual) are the only a few control or speciRed variables, such as Rxed
latest from the inside loop. Simple relative volatil- reSux ratio and product rates. The middle loop ad-
ities are among the outside loop variables, and justs the control variables to meet the speciRcations.
are used in the Kb method to calculate the temper- The middle loop can be built as an optimization
atures and whenever K values are needed in the inside method with process speciRcation equations and eco-
loop: nomic objectives and constraints.
Russell’s (1983) method differs from Boston’s in
the inside loop by a solution method of the MESH
ji"Kji(actual)/KbjRef
equations that includes speciRcations for product
quality, stage temperatures, internal Sow rates, etc.,
These simple relative volatilities change little over
without the use of a middle loop to solve these. Here,
the range of temperatures that is seen on a given stage
for each heat exchanger in the column, plus each
and greatly simplify temperature and composition
additional side product, an additional speciRcation
calculations in the inside loop. For nonideal mixtures,
and operating variable is added to the problem. Rus-
an activity coefRcient for each component accounts
sell’s method has been found to work well for reRnery
for composition effects in the inside loop. This activ-
fractionators with side strippers and other similar
ity coefRcient has a simple model, similar to the
columns.
Kb model:

Relaxation Methods
ln
Hji "aji#bjixji
A relaxation method Rnds a steady-state solution of
where the new outer loop variables, aji and bji, for a column as if it were an operating column changing
each component are determined from the actual ac- with time. The column is initialized using some realis-
tivity coefRcient model at the current stage temper- tic condition and then makes steps to the steady-state
ature and stage composition. conditions by successive approximations of the
 II / DISTILLATION / Modelling and Simulation 1069

unsteady-state distillation equations. These unsteady- The correlations for the mass and heat transfer
state equations are modiRcations to the MESH equa- coefRcients and interface also take into account pack-
tions to include changes in the MESH variables with ing or tray geometries for the actual column. The
respect to time. This mimics the physical start-up total mass and energy rates are calculated from inte-
of the column, but the objective is not to follow grating the mass and energy Suxes across the total
the dynamic operation but to seek the steady-state interface surface.
solution. Krishnamurthy and Taylor (1986) present and test
a nonequilibrium model which includes rate equa-
tions among the traditional MESH equations. These
Homotopy^Continuation Methods include individual mass and energy balances in the
vapour and the liquid and across the interface. An
Homotopy or continuation methods are applied to
equilibrium equation exists for the interface only. The
difRcult-to-solve columns, and are a simple means of
solution methods for these equations are the same as
forcing a solution. The MESH equations can be difR-
the global Newton methods.
cult to solve, due either to the nature of the column
The total mass transfer rates are added to an ex-
(many feeds or side products, side strippers, near
panded set of the MESH equations called the
minimum reSux, etc.) or to the nonidealities of the
MERQ equations. The new MERQ acronym stands
K values or enthalpies. For three-phase systems, azeo-
for:
tropic systems or systems of columns with two or
more feed/recycle stream combinations, there may be
Material balances for each component } one for the
more than one calculated solution. The method must
bulk vapour, one for the bulk liquid and one across
be forced to reach the desired solution. Homotopy
the interface.
methods begin with a known solution of the column
Energy balance equations } one for the bulk va-
and from there follow a path to the desired solution.
pour, one for the bulk liquid and one across the
The known solution can be at different conditions or
interface.
with much simpler K value and enthalpy methods and
Rate equations for mass transfer for all but one
stepped changes are made from there, solving the
component } one from the interface to the bulk
column equations at each step, until the Rnal solution
vapour and one from the bulk liquid to the inter-
is reached.
face, plus one energy transfer rate equation from
the liquid to the vapour.
eQuilibrium equation at the interface only.
Nonequilibrium or Rate-based Methods
Stage efRciency prediction and scale-up from ideal or
equilibrium stages to the actual design can be difRcult
Outlook
and unreliable for many columns. For highly New rigorous methods for fractionation modelling
nonideal, polar and reactive systems, such as amine may no longer be forthcoming and most enhance-
absorbers and strippers, prediction and use of ef- ments will be driven by greater acceptance of
Rciencies is particularly difRcult. In such mixtures, nonequilibrium methods, and to other methods by
mass transfer and not equilibrium often limits the their application to more complex fractionators and
separation. difRcult systems of components. Teaching concepts of
Nonequilibrium methods attempt to get around the equations and solution may be limited to what is
difRculty of predicting efRciencies by replacing the necessary to understand a programme’s options, di-
equilibrium stage concept. Instead, they apply agnostics and why a programme acts in a certain
a transport phenomena approach for predicting mass manner. There should be greater emphasis on know-
transfer rates. Here, the bulk vapour and liquid ledge of the physical reality of a column and where
phases are not at equilibrium with each other, but the actual process is sensitive, to help set up
there is equilibrium at the interface between phases a problem. Software improvements are needed more
with a movement from the bulk phase through the in analysis and troubleshooting thought processes,
interface (Figure 3). The net loss or gain of material tools and reports. Some of these tools may be a return
and energy at the interface is expressed as transfer to use of pre-computer tools such as x-y,
rates. The mass and energy transfer rates are depen- McCabe}Thiele, and Hengstebeck diagrams and
dent on the mass and energy transfer coefRcients for shortcut methods. While computers continue to be-
each phase which are in turn dependent on composi- come more common, faster and easier to use, they
tion and conditions of each bulk phase and at the should never be a substitute for sound engineering
interface. experience and judgement.
1070 II / DISTILLATION / Modelling and Simulation

Figure 3 Model of a nonequilibrium separation and mass transfer.

See also: II/Distillation: Historical Development; Theory Kister HZ (1992) Distillation Design. New York:
of Distillation; Vapour-Liquid Equilibrium: Correlation and McGraw-Hill.
Prediction; Vapour-Liquid Equilibrium: Theory. Kister HZ (1995) Troubleshooting distillation simulation.
Chemical Engineering Progress 16(6): 63.
Further Reading Krishnamurthy R and Taylor R (1986) Multicomponent
mass transfer theory and applications. In Cheremisinoff
Boston JF (1980) Inside-out algorithms for multicompo- NP (ed.) Handbook of Heat and Mass Transfer. Gulf
nent separation process calculations. American Chem- Publishing Company.
ical Society Symposium Series No. 124: 135. Lockett MJ (1986) Distillation Tray Fundamentals. Cam-
Brierley RJP and Smith RI (1979) DISTPACK } Using a bridge, UK: Cambridge University Press.
combination of algorithms to solve difRcult distillation Naphtali L and Sandholm DS (1971) Multicomponent sep-
and absorption problems. Chemical Engineering Sym- arations calculations by linearization. American Insti-
posium Series No. 56: 89. tute of Chemical Engineers Journal 17: 148.
Friday JR and Smith BD (1964) An analysis of the equilib- Russell RA (1983) A Sexible and reliable method solves
rium stage separations problem}formulation and con- single-tower and crude-distillation-column problems.
vergence. American Institute of Chemical Engineers Chemical Engineering 90(20): 53.
Journal 10: 689. Taylor R, Wayburn TL and Vickery DJ (1987) The Devel-
Holland CD (1981) Fundamentals of Multicomponent Dis- opment of Homotopy methods for the solution of se-
tillation. New York: McGraw-Hill. paration process problems. International Chemical
Ketchum RG (1979) A combined relaxation}Newton Engineering Symposium Series No. 104: B305.
method as a new global approach to the computation of Wang JC and Henke GE (1996) Tridiagonal matrix
thermal separation processes. Chemical Engineering for distillation. Hydrocarbon Processing 45(8):
Science 34: 387. 155.