Computers and Chemical Engineering 24 (2000) 2043 – 2054

www.elsevier.com/locate/compchemeng

Synthesis of distillation-based processes for non-ideal mixtures

Arthur W. Westerberg *, Jae Woo Lee, Steinar Hauan
Department of Chemical Engineering, Carnegie Mellon Uni6ersity, Pittsburgh, PA 15213, USA

Abstract

Our understanding of how to design distillation-based processes to separate mixtures displaying azeotropic behavior has grown
enormously in the past quarter century. We elegantly sketch distillation column behavior on a composition diagram where we
display VLE and column behavior using residue and distillation curves. In the presence of azeotropes, these curves partition
composition space into distillation regions that trap the performance of individual columns. We can view liquid – liquid behavior
as a tearing of composition space, a tear that always spawns from a minimum binary azeotrope. A material balance across a
column section leads to a difference point, which we can use to understand column tray-by-tray and limiting behavior for
ordinary, extractive and even reactive distillation. We demonstrate how to synthesize alternative separation processes, concentrat-
ing on those that produce pure products. We use boundary curvature, solvent addition, extractive agents, liquid – liquid behavior
and strategically placed reaction to step across distillation boundaries. We show that these processes always contain recycles to
gain feasibility but also to have economically attractive processes. © 2000 Elsevier Science Ltd. All rights reserved.

Keywords: Distillation based processes; Azeotropic behavior; Non-ideal mixtures; Reactive distillation; Difference points

1. Introduction ex-Soviet Union literature (see the references in Pet-

lyuk, 1998) and papers by Doherty and Perkins (1978)
We were invited to prepare a paper on the synthesis and then a large number of papers by Doherty’s group
of distillation-based separation processes for the at the University of Massachusetts.
CEPAC Workshop held September 2 and 3, 1999, at The theme for this paper will be visual insights,
INTEC in Santa Fe, Argentina. We elected to prepare which we strongly believe aid designers to be innova-
a tutorial paper that looks at the series of improve- tive. Very powerful insights are often based on simple
ments we have had in chemical engineering of our approximate sketches that expose the essence of a de-
understanding of these processes in the last 15–20 sign problem. The grand composite curve for heat
years. 20 years ago we had reasonable methods to aid exchanger network synthesis is one example (described
engineers to synthesize processes that separated rela- in Chapter 10 in Biegler, Grossmann & Westerberg,
tively ideal mixtures, using a series of conventional two 1997). Another classic example is the McCabe–Thiele
product columns. We were just beginning to develop diagram for binary distillation. We often teach our
visualization methods to aid us to find the better heat students the mechanics for constructing a McCabe–
integrated sequences, including the use of more com- Thiele diagram and show them how to determine the
plex configurations such as side strippers and enrichers number of trays for a column. However, determining
and so-called Petlyuk configurations. the number of trays is not the primary use for such a
There were already at this time some very notable diagram. We would almost always use a readily avail-
publications guiding us to understand the behavior of able computer program instead. Perhaps the most im-
azeotropic mixtures — the book by Hoffman (1964), portant use is when experienced engineers use it to
visualize how columns behave. For example, knowing

Paper presented at CEPEC, Santa Fe, Argentina, September 2 – 3, how the operating lines on a McCabe–Thiele diagram
1999. shift when one preheats the feed allows one to decide if
* Corresponding author. Tel.: +1-412-2682344; fax: + 1-412-
2687139.
preheating the feed is likely to useful for a given
E-mail address: [email protected] (A.W. Westerberg). column.

0098-1354/00/$ - see front matter © 2000 Elsevier Science Ltd. All rights reserved.
PII: S0098-1354(00)00575-5
2044 A.W. Westerberg et al. / Computers and Chemical Engineering 24 (2000) 2043–2054

In this paper we shall focus on the development of 2.2. Residue cur6es

insights that have become available for separating mix-
tures that display azeotropic behavior. We shall end It is useful to sketch the vapor–liquid (VL) equi-
this paper by presenting some of the recent insights that librium behavior of a set of species on a composition
are becoming available for reactive distillation. If these diagram. There are two characterizations commonly
are successful, we may at times be able to use sketches used: residue curves and distillation curves. The former
to decide if we should use reactive distillation and, if so, relates to boiling a liquid in a pot while the latter
even where we should place the reaction within the corresponds to liquid compositions we would see in a
column. staged distillation column.
We first present a number of fundamental concepts For residue curves, consider having the pot of boiling
to aid in the design of distillation-based separation liquid shown in Fig. 1. We can determine the composi-
processes. We shall generally limit ourselves to ternary tion of the liquid remaining in the pot. We write the
mixtures so we can visualize the geometry. We shall use following component material balance
a fairly informal writing style to be consistent with our
dxi M dM dxi
intent that this paper be tutorial and, hopefully, easy to =xi · + M· = −yi V (1)
dt dt dt
read.
We note that dM/dt is −V. We also define a dimen-
sion-less time t= tV/M. Eq. (1) then becomes
2. Basics dx
=x6 − y6 (2)
dt
In this section we shall review the use of composition
diagrams to display VL(L) behavior and to allow for Assuming the vapor compositions y6 are in equi-
the geometric construction of the tray-by-tray behavior librium with the liquid compositions x6 , and if we plot
of columns. Very recent developments allow us to the trajectory for the liquid compositions x6 on a com-
consider columns in which reaction is also occurring. position diagram, we get what is termed a residue
curve, i.e. the composition trajectory for the residue
2.1. Composition diagrams that is left in a boiling pot as we boil away the liquid
versus time. Looking at this last equation as a vector
Composition diagrams are particularly useful to visu- equation, we see that the direction for the vapor com-
alize the performance of distillation columns, especially position y6 in equilibrium with the liquid composition in
ternary diagrams as we can readily sketch them on a the pot x6 lies on the tangent line that touches the
sheet of paper. A ternary diagram is triangular in residue curve at x6 .
shape, and it lies entirely within a plane. A four-compo-
nent diagram forms a tetrahedron in three-space — a 2.3. Distillation cur6es
bit difficult for visualization — while a binary diagram
is simply a line which does not convey too much The other type of curve we can draw, which gener-
insight. A composition diagram for nc components has ates a very similar diagram, is the curve passing
nc − 1 degrees of freedom as the compositions all must through the tray-by-tray liquid compositions in a
add to unity, which is, for example, why a three compo- staged distillation column operating at total reflux.
nent diagram lies in a two-dimensional plane. When we Looking at Fig. 2, material balance requires that the
examine difference points later, we shall find individual two streams are identical that flow counter to each
component compositions that lie outside the range 0–1; other between any two stages. Assume the vapor leav-
these add some interesting geometry to these diagrams. ing a tray is in equilibrium with the liquid leaving. To
step from tray to tray, we start with any arbitrary
composition xn. We compute yn in equilibrium with xn,
a bubble point computation. We then set xn − 1 =yn. We
compute yn − 1 in equilibrium with xn − 1, set xn − 2 =
yn − 1, etc. Thus a series of bubble point calculations will
move us up the column. In a similar fashion, we can
use a series of dew point calculations to move down the
column.
Just where will these curves head that we produce?
With increasing time, a residue curve will move upward
in temperature, terminating ultimately at a local maxi-
mum temperature. Such a point is called a stable node
Fig. 1. Batch still. as the time trajectory ends there as time goes to infinity.
 A.W. Westerberg et al. / Computers and Chemical Engineering 24 (2000) 2043–2054 2045

from this previous bubble point computation. Repeat-

ing enough times will generate a set of points leading to
the unstable node from which the curve emanates.
When close enough to this unstable node, return to the
original liquid composition and carry out a dew point
calculation for it. Repeat this step by carrying out a
dew point computation for the liquid resulting from
this previous dew point computation. Eventually the
points will approach the stable node towards which this
curve is moving. When close enough to the stable node,
this curve is complete. Develop another curve by plac-
ing another first composition point in the composition
diagram and develop, as above, the curve for it. Repeat
until curves are developed throughout the composition
diagram.

2.4. Distillation regions

Fig. 2. Total reflux column. Fig. 3 illustrates the shape of residue curves for a
ternary mixture of ethanol, water and glycol. All trajec-
tories start at the unstable node — the minimum
boiling azeotrope between water and ethanol, move
toward the components having intermediate boiling
points (toward water or ethanol, depending on which
direction we move away from the minimum boiling
azeotrope) and then towards the stable node corre-
sponding to pure glycol where they terminate. Points
like those for pure water and ethanol are called saddle
points as each has a trajectory that enters it, turns, and
leaves in another direction. For water for example, the
trajectory enters from the right along the lower edge,
turns and leaves along the left edge toward glycol.
Of interest are topologies where there are two or
more stable and/or unstable nodes. Fig. 4 illustrates a
rather complicated composition diagram. Here there
are two unstable nodes (at local minimum temperatures
of 120°C at the top and 155°C along the bottom edge)

Fig. 3. Trajectories for composition in still pot.

Stepping down a column will create a set of points that

move upward in temperature, also ending at a stable
node. Integrating the differential equations for the
residue curve backwards in time or heading upward in
a column will lead downward in temperature, ulti-
mately to a local minimum point in temperature. Such
a point is known as an unstable node for the residue
curve as all curves will leave such a point as time
increases.
We can develop distillation curves by repeatedly us-
ing bubble and dew point flash computations, such as
available in commercial simulators. Pick an arbitrary
composition somewhere in the composition diagram.
Do a bubble point calculation for it. Repeat this step
by finding the bubble point for the vapor composition Fig. 4. Complicated distillation or residue curve diagram.
2046 A.W. Westerberg et al. / Computers and Chemical Engineering 24 (2000) 2043–2054

liquid–liquid behavior to be computed. Also attach two

liquid product steams to the flash unit.
Select an arbitrary overall liquid composition (point
x). If it lies in the VLL (vapor–liquid–liquid) equi-
librium region, a bubble point flash computation will
produce two liquid products with differing composi-
tions (points a and b) and a vapor composition (point
c) in equilibrium with both of them. The two liquid
compositions must lie at the end of a tie-line that passes
through the overall liquid composition x (a bubble
point produces no vapor so the original mixture parti-
tions into the two liquid phases). If we choose any
other liquid composition on that same tie-line, say x*,
we will again produce the exact same compositions a, b,
and c. Only the amount of the liquids having composi-
tions a and b will differ to satisfy the lever rule. Thus to
discover the structure of the VLL region, we need only
Fig. 5. Vapor – liquid–liquid behavior on a ternary diagram.
place one liquid composition on each tie-line of interest.
If the vapor composition happens to lie on precisely the
and two stable nodes at local maximum temperatures same tie-line as the overall liquid composition that
of 160 and 170°C at the lower right and left. This produced it, then it corresponds to a ternary azeotrope.
diagram has four different distillation regions. In region We could have selected an overall liquid composition
I, residue curves emanate from the upper unstable node equal to that vapor composition and produced a calcu-
and terminate at the lower left stable node. Region II lation in which the overall liquid composition is the
has the same unstable node, but all of its residue curves same as the vapor composition produced from it-which
head to the lower right stable node. Regions III and IV is the definition of an azeotrope.
have trajectories that start from the second unstable All the vapor compositions in the VLL region will lie
node. along a single line. If we select our overall liquid
The ternary azeotrope in the middle is a saddle point compositions closer and closer to the lower edge, the
with the trajectories coming from the unstable nodes, vapor composition must also approach that edge as in
turning, and leaving toward the stable nodes. the limit we will have only a binary mixture of species
Discovering distillation regions by plotting residue or A and B. That vapor composition will be an azeotrope
distillation curves is one of the most important geomet- as it lies on the tie-line that produced it. It will, in fact,
ric insights we can get from them. Experience with such be the minimum boiling azeotrope that must occur
diagrams has told us that we cannot design conven- between species A and B. If we start with a liquid
tional distillation columns to operate such that their composition outside the VLL region, then we may
distillate and bottoms products are in two different produce a vapor that lies within that region if it were
regions (Doherty & Perkins, 1978). These regions essen- condensed. Thus the distillation curve we are discover-
tially trap where the products can lie as a result. The ing using bubble point computations would move inside
trick to designing separation processes for species dis- the VLL region and immediately jump to be on the
playing this type of complex behavior is to figure out vapor line we described above. We can approach the
ways to cross these boundaries. VLL region from either side of it, jumping to the vapor
line once we enter it. Thus one can think of the VLL
2.5. Liquid–liquid beha6ior region as a tear — i.e. a fissure — in the composition
space for the components.
It is interesting to think about what will happen As an exercise, think what the geometry will be if
when a mixture of three components breaks into two or there are three liquid phases in equilibrium with a single
more liquid phases that are in equilibrium with a single vapor, i.e. a VLLL problem. Also think about how
vapor phase. We will use Fig. 5 to illustrate. Liquid– many liquid compositions are needed to establish its
liquid behavior occurs when the Gibbs free energy of a geometry.
mixture decreases if the material breaks into two liquid
phases. (A discussion of this behavior is in most chem- 2.6. Reachable products for a simple distillation column
ical engineering thermodynamics textbooks — see for
example, Smith and Van Ness (1987). It also appears in Once we know the structure of the VL(L(L)) behav-
Biegler et al. (1997)). When generating a distillation ior as displayed on a ternary composition diagram, we
curve, select a physical property option that allows can approximate the behavior of conventional distilla-
 A.W. Westerberg et al. / Computers and Chemical Engineering 24 (2000) 2043–2054 2047

tion columns, i.e. a column with a single feed and a top one option, we place the distillate product at the un-
(distillate) product and a bottoms product (Westerberg stable node and draw a straight line from it through the
& Wahnschafft, 1996; Biegler et al., 1997; Stichlmair & feed to the opposite edge of the region containing the
Fair, 1998). Because the feed is the sum of the two distillate top product to locate approximately the corre-
products by material balance, we know that the feed sponding bottom product. For the other option, we
composition must lie on a straight line between the top place the bottom product at the stable node and draw
and bottom product compositions. Also the lever rule a straight line from it through the feed and onward to
tells us just where along this line the feed composition the opposite side of the region containing the bottom
lies. In Fig. 6, the line lengths a, b, and a + b are product to locate approximately the corresponding top
proportional to the flowrates for B, D and F, product. We get a bow-tie shaped region as shown in
respectively. Fig. 7. It is possible to argue that any products lying in
If we assume the column operates at total reflux, the the shaded region and along a straight line through the
product compositions must also both lie on the same feed can correspond to a feasible distillation column.
distillation curve. We normally would also like to get The usual way to assess quickly the possible products
fairly pure products from a column. We would get fairly is to assume either (1) the stable node is the bottom
pure products if the column has a large number of product (b1) or the unstable node the top product (d2).
stages, which is well approximated by assuming an The other product for each of these two cases (d1 or b2)
infinite number of stages. Thus the bottom end will be is then that which is most distant along the straight line
close to a stable node or the top to an unstable node. through the feed and in the same distillation region.
Fig. 7 illustrates the products regions we can expect. For The feed does not have to lie in the same region as the
two products. When it does not, the feed tray composi-
tion must be different from the feed composition. The
feed tray composition will lie in the same region as the
products — as will all the liquid tray composition —
while the feed composition lies in the other region. One
can always have the feed and feed tray composition
more or less the same for a binary column but not in
general for columns having three or more components
in them.
Fig. 6. Lever rule on composition diagrams.
2.7. Difference points — the geometry of component
material balances

The next interesting geometric insight we can observe

for a distillation column is the notion of a difference
point (Westerberg & Wahnschafft, 1996; Biegler et al.,
1997; Hauan, Ciric, Westerberg & Lien, 2000). In Fig.
8 we write the overall material balance for the top
section of an ordinary distillation column. We see the
form for the balance suggests that the entering vapor,
Vn + 1, is the sum of the distillate product and the liquid
flow leaving, Ln. Thus the composition for Vn + 1 lies on
a straight line between the compositions for D and Ln.
Fig. 7. Bow-tie shaped reachable region for a two-product column Assume the species are A, B and C, with A being the
operating at total reflux. most volatile and that we want our distillate product to
be relatively pure A. We sketch this geometry for an
arbitrary composition for Ln. The level rule says we
place Vn + 1 such that the line segments D−Vn + 1,
Vn + 1 − Ln and D−Ln are proportional to D, Ln and
Vn + 1, respectively. Ln /D is the reflux ratio for the
column so, if we set the reflux ratio, we know exactly
where to place Vn + 1 along this line. This line represents
a material balance. It allows us to find the composition
for the vapor flowing counter to the liquid between two
stages — just as the operating line does on a McCabe–
Fig. 8. Difference point for top part of ordinary distillation column. Thiele diagram for a binary column.
2048 A.W. Westerberg et al. / Computers and Chemical Engineering 24 (2000) 2043–2054

exactly where as the line lengths D−D, D−S and

D− S, must correspond to the ratio of the flows of S, D
and D, respectively. If the solvent flow is nearly zero,
the composition of D is very close to D. As the flow of
solvent increases, it will eventually be equal to D. At
that point the composition for D will have moved out
along the line to infinity. When S increases past D, the
flow D becomes negative, and its composition jumps to
infinity at the other end of the line. As S increases, the
composition for D moves towards that for S, which it
equals when S has an infinite flow.
In Fig. 9 we construct a material balance line to find
Vn + 1; only this time we use the compositions for Ln
Fig. 9. Difference point for an extractive distillation column.
and D.

2.8. The reaction difference point for a column (Hauan,

Westerberg & Lien, 1999)

We are now ready to examine the impact of reaction

on a column. Reaction can be viewed as two flows. The
reactants flow out from the stage while the products
flow into it. If we were to do our material balances
using mass flows rather than molar flows, there would
be no loss or gain in material because of reaction. In
other words, 1 kg of reactants will always produce 1 kg
of products (disallowing nuclear reactions). However,
we can easily gain or lose moles. For example, the
reaction A+BUC converts two moles into one mole
Fig. 10. Difference point when reaction occurs. each time the reaction turns over as written. We charac-
terize the rate of reaction as a turnover, which is a
The point D is called a difference point. The distillate flowrate (e.g. mol s − 1) corresponding to the flowrate at
D is the fixed difference (Vn + 1 −Ln ) for all stages n which the reaction occurs as written. We designate
above the feed in the column. Thus all material balance reaction turnover by the symbol j.
lines such as we have constructed are straight lines that Fig. 10 aids in understanding the following ideas. We
pass through this single point for the top section of our can write the reaction A+BUC in the form −A−
column. B+ C= 0. The vector of stoichiometric coefficients n
Let us now add a solvent as an extractive agent to for this reaction is the transpose of [− 1, −1, 1]. If the
our column (see Fig. 9). Let us assume it is the heaviest turnover of the reaction as written is j mol s − 1, then
species in the mixture, component C, which lies at the the net flow due to reaction into the column is ntotj=
− 1j, i.e. there will be two moles leaving and one
upper corner of our composition diagram. The solvent
entering for a net of − 1j mol s − 1 entering. The
is a second feed to the column. We examine the mate-
composition of this flow is − 1 mol s − 1 of A, −1 mol
rial balance for trays just below this feed.
s − 1 of B and + 1mol s − 1 of C, rescaled to add to
Again we write a material balance around the top
unity, i.e. n/ntot = [1, 1, −1]T. We illustrate this compo-
section, but this time the constant difference point is sition in Fig. 10. Note it is on the line where the
D −S rather than D. As is routinely done in the compositions for A and B are one and C is negative
textbooks, we call this difference D. We see that D is the one. (To see this, note that the composition of A is 1 in
difference between D and S rather than the sum. If the the lower left corner of the composition triangle and
solvent flow S is very small relative to D, then D is zero all along the right edge of the triangle. Lines of
positive, and we can move S to the other side, giving us constant A are parallel to each other so along the line
D = S +D, with all flows being positive. Thus the com- joining D and the reaction difference point the compo-
position for D should lie between the compositions for sition of A is one.)
S and D. However, D lies in the lower left corner, and There is an interesting significance to this reaction
S lies at the upper corner. How can D lie between S composition. It is a difference point for reaction. Select
and D? The composition for D must lie outside the any arbitrary composition within the composition tri-
feasible composition diagram. The lever rule tells us angle and draw a straight line through it and the
 A.W. Westerberg et al. / Computers and Chemical Engineering 24 (2000) 2043–2054 2049

reaction composition point. All points along this line s − 1) varies. We locate the composition for the flow
within the composition triangle can be converted into D− ntotj by drawing a line between the composition for
each other by carrying out the reaction to different D and for the reaction (both fixed). We split this line
extents. The projections developed by Doherty (e.g. using the lever rule to give line length ratios that
Barbosa & Doherty, 1987) and his students map all the correspond to the ratio for the two flows D and −ntotj.
points on one of these lines onto a single point using his As both of these flows are positive for this example, the
projection equations (Stichlmair & Fair, 1998). The resulting difference point here lies between their two
projection for A+ BUC +D results in a two-dimen- compositions. The composition for Vn + 1 must then lie
sional square (see Fig. 11). To understand why it on a straight line connecting this composition to that
becomes a square, construct a tetrahedron out of paper. for Ln. The point can move as we step from tray to tray
The reaction difference point is at infinity for this because the reaction flow, ntotj, is the total of all the
reaction occurring within the stages around which we
reaction, as there are moles that are neither created nor
have written our material balance.
destroyed. The lines passing through this infinite differ-
We now point out one reason this construction is
ence point are parallel to the line going from a 50–50
interesting. Examine Fig. 12. Pick an arbitrary point for
mixture of A and B that reacts to become a 50–50
the composition of Ln in the column. The composition
mixture of C and D. Viewed from the difference point
for Vn is on the line tangent to the residue curve passing
at infinity, the tetrahedron projects into a square. though Ln (or along the distillation line passing through
We now return to the material balance equations Ln ). For an ordinary column, Vn + 1 will be along a line
written at the top of Fig. 10. We see that D for passing through the composition for the distillate, D. If
extractive distillation is replaced by D −ntotj for extrac- we introduce an extractive agent and/or reaction, we
tive distillation with reaction. D has a fixed composition move the difference point. We show it moving as for a
and flow. However, the reaction extent can change case in which we have reaction occurring. We see that
from tray to tray. The flow ntotj varies as j (in mol Vn + 1 is to the left of Vn for an ordinary column and to
the right for a reactive column. We have caused the
direction the compositions change from tray-to-tray to
reverse. The trays below this tray continue to use this
same difference point if they do not have any reaction
occurring on them. Thus we can have a non-reactive
tray in which the compositions seem to change in a
direction that is the reverse of what we expect. The
temperature could actually decrease as we move down
the column.

Fig. 11. A square projection for A + BUC+ D. 3. Using geometric insights to aid design

The above insights can help us to design separation

processes. We will emphasize separation tasks that in-
volve mixtures displaying azeotropic behavior. We will
present this section using examples. The purpose is not
to be all-inclusive but rather to explain why the geome-
try is suggesting these solutions. Thus it is hoped the
reader will see other alternatives based on similar rea-
soning for these and other problems.

3.1. Example: sol6ent selection to break a binary

azeotrope
Fig. 12. Strategic use of difference point placement to steer plate-by-
plate composition trajectories in a column. Our goal here is to find a third component that will
allow us to break a binary azeotrope. The approach we
shall investigate is to find a third component that will
allow us to step around the azeotrope using conven-
tional distillation. Fig. 13 illustrates the situation when
the binary species A and B form a minimum boiling
Fig. 13. Node behavior for a minimum binary azeotrope. azeotrope. We note that A has the lower normal boiling
2050 A.W. Westerberg et al. / Computers and Chemical Engineering 24 (2000) 2043–2054

is no B in this top product; we distill it to separate A

from the entrainer. We recycle the entrainer. It should
not be difficult now to describe the characteristics of an
entrainer to break a maximum boiling azeotrope. The
entrainer can form azeotropes with A and B; we need
only to have a single region in which A becomes a
saddle.
Doherty and Caldarola (1985) and Stichlmair and
Fair (1998) present an extensive set of rules for picking
a suitable entrainer to break binary azeotropes. They
involve this type of thinking.

Fig. 14. Breaking a binary azeotrope with an intermediate boiler. 3.2. Example: crossing distillation boundaries to break
an azeotrope

Another way to think about breaking an azeotrope is

to add a third component that converts the azeotrope
into a saddle. If the azeotrope is a minimum boiling
azeotrope, we need to select a light entrainer (Fig. 15).
This species will cause a distillation boundary to exist
that connects the entrainer with the azeotrope. We
want this boundary to be very curved, so we have a
second requirement: namely, that in regions where the
entrainer concentration is high (near the entrainer cor-
ner of the triangle), the two species have a very differ-
ent volatility. Here species A is much more volatile that
Fig. 15. Breaking a minimum boiling azeotrope with a low boiling B. When that happens distillation curves leave the
entrainer that forms a strongly curved distillation boundary. entrainer (a stable node) and head toward the species
that is acting like the intermediate boiler-here A. They
point for the two components. Because the azeotrope is then turn toward the high boiler as shown. As a result,
a minimum one, it will typically lie closer to the lower the distillation boundary is very curved as shown.
boiling component, as shown. We first mix the entrainer and the azeotrope, moving
If we can alter our problem so A, B and the the feed for the first column well into the three compo-
azeotrope are all in the same distillation region, we will nent part of the diagram so as to take maximum
be able to break the azeotrope using distillation. To be advantage of the curvature of the boundary. We distill,
in the same distillation region, we must have only one getting species B as the bottom product. The distillate is
stable and one unstable node. We have to convert at close to the distillation boundary. This distillate be-
least one of these nodes to a saddle. For example, we comes the feed to the second column. While the feed is
can convert A into a saddle in the three component in the right region, we can have both products in the
composition diagram by picking a third component left region at the same time that both lie on a straight
that forms no azeotrope with them and that has a line through the feed. As noted earlier, only the prod-
boiling point that is between that for A and B. We ucts have to be in the same distillation region. A third
would get the behavior shown in Fig. 14. In this column recovers A and azeotrope. We recycle the
diagram, the binary azeotrope remains the unstable azeotrope back to the feed.
node. B remains a stable node, while A and C are This type of reasoning is involved in inventing sepa-
saddles. All distillation curves will emanate from the ration processes for the acetone, chloroform and ben-
azeotrope and end up at pure B. zene system described in Westerberg and Wahnschafft
To invent a separation process, we could first mix (1996) and Biegler et al. (1997).
our entrainer with the azeotropic mixture, forming a
mixture that is away from the lower edge of the compo- 3.3. Decantation
sition diagram. In the three component region, we can
have a column whose bottom product is the stable If we have a three component mixture to separate
node, species B, the highest boiling component in the where the species have a region in which there is
distillation region. The top product is along the straight liquid–liquid behavior, we often can readily develop a
line passing through the feed and as far as the furthest separation process. Separating pyridine, toluene and
edge of the distillation region, here the left edge. There water is an example. Fig. 16 shows the structure of the
 A.W. Westerberg et al. / Computers and Chemical Engineering 24 (2000) 2043–2054 2051

distillation (or residue) curve map. The two phase the other two species. We also want a solvent in which
region boundary can either correspond to having a the two species will have very different volatilities; i.e.
vapor phase in equilibrium, or it can be the result of we want a solvent that is increases the activity coeffi-
cooling the liquid well below this point and having only cient of one of the species much more than it does for
two liquid phases present. The more one cools, the the other. The species with an increased activity coeffi-
larger the two liquid phase region typically is. cient becomes relatively much more volatile. We can
We start by decanting the feed, to form water- and detect this behavior by examining infinite dilution K-
toluene-rich phases. We can then distill each of the values for the water and IPA in the presence of a
phases to get pure water and pure toluene bottoms number of candidate solvents. If volatilities in a high
products, respectively. Mixing the distillate products, concentration of solvent are very different, we have a
which lie along the distillation boundaries shown, pro- candidate solvent.
duces a mixture in the distillation region where we can In a matter of a few seconds, we can readily examine
distill to recover pyridine, which we do. The distillate thousands of components using a computer, provided
from this distillation lies somewhere in the two-phase we have the appropriate data on each to allow us to
region; we recycle it to the decanter. Here the appear- estimate physical properties. We should find ethylene
ance of a VLL region allows us to use decanting to glycol to be a good candidate here. It is readily avail-
cross distillation boundaries and thus to develop a able, fairly safe to handle, boils at a much higher
process fairly easily. Using this type of reasoning, we temperature than either water or IPA, and it has a large
can invent other possible processes for this feed. infinite dilution activity coefficient with IPA. The com-
position diagram in Fig. 17 illustrates the shape of the
3.4. Extracti6e distillation distillation curves for these components. Examine the
portion of the composition diagram to the right of the
We can also use extractive distillation to effect a curve ab where EG is in high concentration. The shape
separation. What are the geometric clues that would of the distillation curves in this region suggests that
suggest this approach? Consider trying to break the water is the intermediate component while IPA is the
water–isopropylalcohol (IPA) binary azeotrope. Fig. 17 light or most volatile component. Only when the curves
shows a typical extractive distillation column. In such a are near the minimum azeotrope do they bend to move
column, we usually select a very heavy solvent — one toward the minimum boiling azeotrope, which here is
that boils at a much higher temperature than either of the unstable node.
The shape on the right suggests that water is the
intermediate and IPA is the light component. If this
shape appears in high concentrations of a proposed
solvent, then extractive distillation will allow us to
break the azeotrope.
We note that the IPA corner is a saddle yet, with
extractive distillation, we can have it as the top
product. To understand this, return to our earlier dis-
cussion on difference points for extractive distillation.
The D point (see Fig. 9) for extractive distillation will
lie on the straight lines that pass through the IPA
product and EG feed compositions. It will lie outside
the composition triangle. If the solvent flow is less than
Fig. 16. A process to separate species having a region with VLL the distillate product flow, it will be in the direction
behavior. shown. If the solvent flow is larger than the distillate, it
will wrap around and be below and to the right of the
EG corner along this same straight line. In either case,
if we take a tray-by-tray step, as shown in Fig. 12,
somewhere along the curve ab, we would find move-
ment up the extractive column just below the EG feed
would be toward the IPA/EG edge of the composition
space. We could get arbitrarily close to that edge,
effectively eliminating all the water from the top part of
the column. The top of the column above the EG feed
is then separating IPA from EG, a binary separation
that is done in one or two trays as EG is much heavier
Fig. 17. Extractive distillation example. than IPA.
2052 A.W. Westerberg et al. / Computers and Chemical Engineering 24 (2000) 2043–2054

we have − B+ A= 0 and the stoichiometric coeffi-

cients are [−1, 1]. Here ntot = − 1+1= 0 as there are
no net moles lost or produced as this reaction proceeds.
The reaction difference point is at infinity.
We shall construct a McCabe–Thiele diagram for
this system (Lee, Hauan & Westerberg, 2000a). Our
goal is to discover how reaction alters this diagram. If
we write a material balance around the top of a column
(as in Fig. 10) for the component A, we get
Vn + 1yn + 1, A = Lnxn, A + DxD, A − j

= Lnxn + D xD, A −
j  (3)
D
Since there is no net production or loss of moles,
Vn + 1 = Ln + D. If we let yn + 1 = xn to find where this
operating line intersects the 45° line, we find that it
occurs when
j
Fig. 18. McCabe – Thiele diagram for column having reaction on
xA = xD, A − (4)
D
stages 2 and 3 from the top.
Thus the intersection point shifts down by the fraction
the reaction turnover (flowrate) is of the distillation top
product flowrate. For example, if we feed 10 mol s − 1 of
B along with some A and run the column to convert
50% of the B in the feed into A, the turnover will be 5
mol s − 1 for the entire column.
Fig. 18 is a McCabe–Thiele diagram (McCabe &
Thiele, 1925), where we have allowed reaction to occur.
We start at the top tray where we specify that the
composition of A to be recovered in the distillate to be
0.95. The original top operating line intersects the 45°
line at this composition. We (arbitrarily) specify that we
shall have no reaction on the top stage; we step off this
stage using this original operating line. We place cata-
lyst on stage 2 so that we can get a reaction turnover of
j2. The operating line shifts downward. We step to this
operating line below stage 2. Let us suppose that we
also place catalyst on stage 3. We again shift the
operating line down as shown. Note that j3 as shown is
the sum of the turnover rates on and above stage 3 (the
material balance has to account for all stages above the
Fig. 19. McCabe – Thiele diagram when placing reaction in bottom tray of interest). If we choose to put catalyst on no
section. other stages, we should then use this last operating line
for all stages below stage 4 and above the feed tray. We
4. Reactive distillation analyze the bottom of the column as we would for a
non-reactive column.
We will complete this paper by looking at interesting We note that each time we allow reaction, the operat-
geometry for reactive distillation. In particular we shall ing line moves down. Having reaction in the top of the
look at the geometry for binary columns. We have column makes our separation task easier. We ask next
already examined the use of a difference point for what happens to our diagram if we allow reaction for
reaction in ternary systems. We start by supposing we this case to occur in the bottom of the column.
have one component that can rearrange to form the We carry out a similar analysis, but this time we
other; thus our system can be reactive and binary. We place reaction in on a bottom stage of a column. We
shall assume the reactant is the less volatile species and construct the resulting McCabe–Thiele diagram in Fig.
write our reaction as B “ A. Written in standard form, 19 for the same case for which we constructed Fig. 18.
 A.W. Westerberg et al. / Computers and Chemical Engineering 24 (2000) 2043–2054 2053

This time we shall number from the bottom of the operating line is also well below the equilibrium line
column. Assume reaction occurs only on the second even though the equilibrium line is below the 45° line.
stage from the bottom. When we allow reaction and We can observe several interesting things about this
because of the sign changes that occur in the analysis, diagram. First, if we had used total reflux to decide
the intersection point for the operating line and the 45° how to design this column, the operating line would
line moves downward here also. However, doing so coincide with the 45° line, and we would not be able to
now shifts the new operating line upward, leading to a step past the azeotrope. Thus, it is our use of a finite
more difficult separation; indeed, we can make the reflux ratio that gives us this possibility. Second, we
operating line move above the equilibrium curve, which step past the azeotrope below the trays where reaction
would cause us to reverse the direction we separate the occurs. Once past the azeotrope, the more volatile
mixture, i.e. we would start to step back to the left. The species, B, is enriching as we go down the column. This
second dashed operating line illustrates a shifted oper- behavior is similar to the behavior we discussed earlier
ating line that just touches the equilibrium curve lead- for a ternary diagram when we add reaction and possi-
ing to a pinch situation. Thus we can conclude that, for bly an extractive agent flow to the column. As before
this case of the heavy species forming the light species, this enrichment says the temperature is decreasing on
we make the separation task easier by having the non-reactive trays as we step down the column. This
reaction occur in the top section of the column (Lee, behavior is very counter-intuitive for most chemical
Hauan, Lien & Westerberg, 2000b). engineers. It is occurring because reaction has altered
the material balance for the column, here dramatically.

4.1. Stepping past a binary azeotrope (Lee, Hauan &

Westerberg, 2000c) 5. Conclusion

Again consider the reaction B “ A. Only this time We have examined some of the very interesting ge-
we have two species that form a maximum boiling ometry associated with analyzing distillation processes.
azeotrope as shown in Fig. 20. Note that the equi- In particular we first looked at residue and distillation
librium curve crosses the 45° line. This shape is indica- curves to expose VL(L) equilibrium phase behavior. We
tive of a maximum boiling azeotrope. In the upper part next showed that difference points are how we account
of the diagram A is more volatile while in the bottom, geometrically for the material balances of a column. We
B is. We start at the top where we again ask for 95% then looked at altering material balances for a column
pure A as the top product. Arbitrarily, we do not allow by adding a solvent feed to and/or allowing reaction
reaction to occur on the top stage. We place catalyst on within the column and showed we could make the
stages 2 and 3, each time causing the indicated reaction tray-by-tray behavior of a column change directions
turnover and shifting downward of the operating line. and even reverse directions.
The operating line is, by this time, well below the 45° We also showed how to use these insights to discover
line. We find we can now step past the azeotrope as the useful column designs that can break binary
azeotropes. Finally we looked at binary reactive distil-
lation and showed that we can construct and use a
McCabe–Thiele diagram for this situation. Again we
showed that altering the material balance by having
reaction occur within the column can lead to some very
interesting and we think not so intuitive behavior. We
could also derive a rule of thumb from just a sketch
that tells us where to place the reaction in the column.

References

Barbosa, D., & Doherty, M. F. (1987). A new set of composition

variables for the representation of reactive phase diagram. Pro-
ceedings of the Royal Society of London, A413, 459.
Biegler, L. T., Grossmann, I. E., & Westerberg, A. W. (1997).
Systematic methods of chemical process design (pp. 455–494).
Upper Saddle River, NJ: Prentice-Hall Chapter 14.
Doherty, M. F., & Perkins, J. D. (1978). On the dynamics of
distillation processes — I (the simple distillation of multicompo-
Fig. 20. Stepping past a binary azeotrope in a reactive distillation nent non-reacting homogeneous liquid mixtures). Chemical Engi-
column. neering Science, 33, 281 – 301.
2054 A.W. Westerberg et al. / Computers and Chemical Engineering 24 (2000) 2043–2054

Doherty, M. F., & Caldarola, G. A. (1985). Design and synthesis of The McCabe – Thiele diagram. Proceedings: Mathematical, Physi-
homogeneous azeotropic distillations. 3. The sequencing of cal and Engineering Sciences (2000b) accepted.
columns for azeotropic and extractive distillations. Industrial En- Lee, J. W., Hauan, S., & Westerberg, A. W. (2000c). Graphical
gineering Chemistry Fundamentals, 24, 474–485. methods for reaction distribution in a reactive distillation column.
Hauan, S., Westerberg, A. W., & Lien, K. M. (1999). Phenomena Proceedings: Mathematical, Physical and Engineering Sciences
based analysis of fixed points in reactive separation systems. (2000c) accepted.
Chemical Engineering Science, 55 (6), 1053–1075. McCabe, W. L., & Thiele, E. W. (1925). Graphical design of fraction-
ating columns. Industrial Engineering Chemistry, 17, 605.
Hauan, S., Ciric, A. R., Westerberg, A. W., & Lien, K. M. (2000).
Petlyuk, F. B. (1998). Simple methods for predicting feasible sharp
Difference points in extractive and reactive cascades. I — Basic
separations of azeotropic mixtures. Theoretical Foundations of
properties and analysis. Chemical Engineering Science, 55(16),
Chemical Engineering, 32 (3), 245 – 253.
3145 – 3159. Smith, J. M., & Van Ness, H. C. (1987). Introduction to chemical
Hoffman, E. J. (1964). Azeotropic and extracti6e distillation. New engineering thermodynamics (4th ed.). New York: McGraw Hill.
York: Interscience-Wiley. Stichlmair, J. G., & Fair, J. R. (1998). Distillation: principle and
Lee, J. W., Hauan, S., & Westerberg A. W. (2000a). Circumventing practice. New York: Wiley-VCH.
an azeotrope in reactive distillation. Ind Eng. Chem. Res., 39, Westerberg, A. W., & Wahnschafft, O. (1996). Synthesis of distilla-
1061 – 1063. tion-based separation processes. In J. L. Anderson, Ad6ances in
Lee J. W., Hauan, S., Lien, K. M., & Westerberg A. W. (2000b). A chemical engineering. In: Process synthesis, vol. 23 (pp. 63–170).
graphical method for designing reactive distillation columns. II — New York: Academic.