616 Ind. Eng. Chem. Res.

2003, 42, 616-629

Minimum Energy Consumption in Multicomponent Distillation. 3.

More Than Three Products and Generalized Petlyuk Arrangements
Ivar J. Halvorsen† and Sigurd Skogestad*
Department of Chemical Engineering, Norwegian University of Science and Technology,
N-7491 Trondheim, Norway

We consider separation of ideal multicomponent mixtures with constant relative volatilities and
constant molar flows and at constant pressure. The exact analytical solution of minimum energy
in a generalized Petlyuk arrangement for separation of N-component feed into M products has
been derived. Interestingly, the minimum-energy solution in a complex integrated Petlyuk
arrangement is equal to the most difficult split between any pair of the products, as if each
single split was to be carried out in an ordinary two-product column. This extends the results
for the three-product Petlyuk arrangement from part 2 of the series to a generalized arrangement
with any number of products and feed components. The solution is very simple to visualize in
the Vmin diagram (part 1 of the series), simply as the highest peak. In addition, we obtain detailed
flow rates and component distribution inside the arrangement. We also conjecture that the
minimum energy requirement for the generalized extended Petlyuk arrangement is lower than
the minimum-energy requirement for any distillation configuration when we consider conven-
tional adiabatic sections and no internal heat exchange. The Vmin diagram may thus be used to
obtain a target value for the energy requirements.

1. Introduction
What is the minimum energy requirement in multi-
component, multiproduct distillation? In this paper we
present an analytical expression for the minimum
energy requirement for the separation of N feed com-
ponents into M products (where normally M e N). We
derive the expressions for a generalized extended Pet-
lyuk arrangement, where all columns are directly (fully
thermally) coupled. The assumptions are constant rela-
tive volatilities, constant molar flows, constant pressure,
and an infinite number of stages. We focus on a
standard configuration shown in Figure 1. This config-
uration can be extended to any number of products by
adding more arrays of directly coupled columns.
Analytical expressions for minimum energy in a
ternary Petlyuk arrangement have been available for
some time.1,2 Carlberg and Westerberg3,4 presented
solutions for an arbitrary number of intermediate
components. However, as mentioned by Christiansen,5
the general analytical solution of minimum energy for
distillation of a multicomponent feed into multiple
products has not been given in the literature for more
than three products.
The extension to any number of products is the main
result of this paper. This is a direct extension of the Figure 1. Petlyuk arrangement extended to four products. Vapor
results for a three-product Petlyuk column presented and liquid flow rates can be set individually in each internal two-
in part 2 of the series.6,7 The derivation is based on the product column.
Underwood equations,8-11 and as in part 2 of the series,
we use the Vmin diagram to effectively visualize the solution. We obtain the detailed vapor flow requirement
minimum-energy solution also for the generalized Pet- in all column sections for general multicomponent feeds
lyuk column with more than three products. The Vmin and arbitrary product specifications. A review of the
diagram was presented in part 1 of the series12 and gives basic tools is given in section 2.
us a very simple tool to assess the properties of the
The derivation of the minimum-energy expression is
divided into two parts. First, in section 3, we deduce an
* To whom correspondence should be addressed. Phone:
analytical vapor flow rate expression for separation of
+47 73594030. Fax: +47 73594080. E-mail: Sigurd.Skogestad@
chemeng.ntnu.no. N feed components into N pure products for the case
†
Current address: SINTEF Electronics and Cybernetics, when all internal columns are operated at their respec-
N-7465 Trondheim, Norway. E-mail: Ivar.J.Halvorsen@ tive preferred splits,13 and we discuss some of its
sintef.no. properties. Second, in section 5 we verify that this
10.1021/ie0108651 CCC: $25.00 © 2003 American Chemical Society
Published on Web 01/01/2003
 Ind. Eng. Chem. Res., Vol. 42, No. 3, 2003 617

Figure 2. Vmin diagram for a given four-component feed (ABCD) to the prefractionator. The set of distributed components and corresponding
active Underwood roots are indicated in each distribution region. The preferred split is at PAD.

solution is a minimum-energy solution for the arrange- what we denote as the feed equation, which gives us
ment. Analytical expressions are only shown for sharp the set of possible common roots θ:
M ) N sharp split products. It is straightforward to
apply the same approach for nonsharp splits too. N Ri(wi,T - wi,B) N RiziF
However, instead of presenting the more complicated
analytical expressions, we illustrate the general M
VT - VB ) ∑
i)1 Ri - θ
) ∑
i)1R -θ
) (1 - q)F
i
product case (M > N) with both sharp and nonsharp (2)
product split specifications by an example in section 4.
Finally, in section 6, we conjecture that the minimum Underwood showed that, for ordinary columns, the
vapor flow expression for the system in Figure 1 number of each set of roots is equal to the number of
represents the minimum energy for any possible distil- components (N), and they obey Ri g φi g θi g ψi+1 g
lation arrangement when we apply adiabatic column Ri+1, and there are N - 1 possible common roots.
sections and no internal heat exchange in the system. Furthermore, for the case with an infinite number of
The term adiabatic refers to a typical column section stages, minimum vapor flow solutions correspond to the
where there is no heat exchange along the section. fact that pairs of Underwood roots in the top and bottom
sections coincide with the common roots (φi ) θi ) ψi+1).
2. Underwood Equations and the Vmin Diagram This occurs only for the roots in the range between the
relative volatility of the distributing components, and
The Underwood equations, and in particular how the we denote these as active roots. Observe that all of the
Underwood roots carry over to succeeding directly possible common roots from (2) depend only on feed
coupled columns,3 are the main keys to the analytical composition and quality and not on how the column is
solution. In addition, we use the Vmin diagram to operated.
effectively visualize the exact analytical solutions. A 2.1. Vmin Diagram. A two-product column has only
brief description of the basic equations is given below. 2 degrees of freedom in operation (e.g., D and V), and
Consider a two-product distillation column with a all possible operating points can be visualized in the
multicomponent feed (F) with liquid fraction q and D-V plane. This is the basic idea behind the Vmin
composition vector z of N components. The net flow of diagram presented in part 1 of the series. For a given
components (wi) are defined positive upward and into four-component feed, an example is shown in Figure 2.
feed junctions. The defining equation for the Underwood Each peak and knot (Pij) represent minimum vapor flow
roots (φ) in the top and (ψ) in the bottom are for a sharp split between components i and j. The
straight lines are distribution boundaries where one
N Riwi,T component is at the limit of being distributing. Inside
Top: VT ) ∑
i)1R -φ
each region, a particular given set of components is
i distributing and there is a corresponding set of active
Underwood roots (θ). Thus, the exact component distri-
N Riwi,B bution can then be calculated from the equation set
Bottom: VB ) ∑
i)1R -ψ
(1) obtained by applying the active roots in (1). All of the
i possible minimum-energy solutions is found below the
Vmin boundary (bold), and there is a unique solution for
There will be N solutions for each root, and the sets every feasible pair of key component recoveries. The
from the top and bottom equations are generally dif- peaks represent minimum-energy operation for a sharp
ferent. By subtracting the equations above, we obtain split between adjacent components. For a sharp split
618 Ind. Eng. Chem. Res., Vol. 42, No. 3, 2003

between components j and j + 1, only one common

Underwood (which obeys Rj < θj < Rj+1) is active, and
the peak (Pj,j+1) can be expressed by
j RiziF j
j/j+1
Pj,j+1: VT,min ) ∑
i)1R - θj
, Dj/j+1 ) ∑
i)1
ziF (3)
i

The preferred split (PAD in Figure 2) is particularly

interesting for directly coupled arrangements. Then all
possible common roots are active. The characteristic of
the preferred split is that it is the minimum-energy
solution when the heaviest component is removed from
the top and the lightest component is removed from the
bottom.
2.2. Carryover of Underwood Roots in Directly
Coupled Columns. In part 2 of the series, we showed
that the possible common Underwood roots in a directly
coupled succeeding column are equal to the actual roots
in the preceding column. For the arrangement in Figure
1, we have for column C21

θC21
i ) φC1
i (4)

In C21 we only need to consider the roots between

the relative volatilities of the components actually Figure 3. Extended four-product Petlyuk arrangement showing
appearing at the feed junction. When column C1 is the active Underwood roots for preferred split operation of all
internal columns. The intersection lines represent the product
operated at its preferred split, all common roots in C1 splits (I1, A/BCD; I2, AB/CD; I3, ABC/D).
are active and we simply obtain θC21 i ) θC1i . The
possible common roots (θ) in C1 are given by the feed extended to any number of products by adding more sets
equation (2). (In the following, we omit the superscript of directly coupled columns. There is only one reboiler
C1 for the roots of column C1.) and condenser, always at the outlets for the final bottom
For the arrangement in Figure 1, with the four feed and top products, respectively.
components ABCD, we have three common roots in For an M-product arrangement, there are M - 1 cross
column C1 (θA, θB, and θC). For C1 operated at its sections that may have independent total vapor flow
preferred split, all of these roots will be active. Compo- requirements through all intersected columns. These
nents ABC and the corresponding roots θA and θB will intersections represent the product splits in the system.
carry over to C21, and components BCD and the roots We have chosen to use the particular sets I1, I2, and I3
θB and θC will carry over to C22. When both C21 and for M ) 4, which intersect all internal top sections as
C22 are operated at their respective preferred splits, shown in Figure 3. Note that only the A product passes
components AB will appear in the feed to C31 and θA through intersection I1; thus, I1 represents the A/BCD
will carry over from C21. Similarly, components BC and split. I2 represents the AB/BC split because all of A and
the root θB will appear in C32 and components CD and B but none of C and D pass here. Finally, I3 represents
the root θC will appear in C33. This is indicated in the ABC/D split. This can easily be extended to the
Figure 3. The mixing of flows from C21 and C22 at the general M-product case.
feed junction to C32 does not give any problems. It is When each internal column operates at its preferred
shown in work by Halvorsen7 that the liquid and vapor split, all of the common Underwood roots (θA, θB, and
compositions are identical in the top of C22 and bottom θC for N ) 4) given by the feed equation (2) for the
of C21, and this is also confirmed by the simulation prefractionator feed will carry over to the succeeding
example in section 3.2.2. columns as indicated in Figure 3.
In part 2 of this series,6 it was shown that the Then, note that in each column section, cut by each
carryover of the common Underwood roots from column intersection line (I1, I2, or I3), there is one common
C1 to the succeeding columns also implicates that the active Underwood root (e.g., θB is active in columns C21
Vmin diagrams for the succeeding columns overlap the and C32 intersected by I2). We can apply this root in
diagram for the feed into the first column. Thus, this the defining equation for each column cross section and
Vmin diagram contains information about all minimum find the total vapor flow through the intersections. For
flows in the succeeding columns, provided that all a sharp product split, the net product flows are simply
preceding columns are operated at the preferred split. the amount of the main product component in the feed.
This is illustrated in Figure 4. The minimum vapor flow through I1 is trivially

3. Minimum Energy for N Components and N RAwC31

A,T RAzAF
Products VI1
min ) ) C1,A/BCD
) VT,min (5)
RA - θA RA - θA
We are now in position to compute the minimum
vapor flow in a general extended Petlyuk arrangement We recognize that this is the same minimum energy
with any number of feed components (N) and any as if the separation A/BCD was to be performed in a
number of products (M). We will start from the basic single column. At intersection I2 we know that all of
four-product arrangement in Figure 3, which can be the light A component passes through the top of C21
 Ind. Eng. Chem. Res., Vol. 42, No. 3, 2003 619

Figure 4. Vmin diagram showing the minimum vapor flows and product splits for every section in the Petlyuk arrangement in Figure 1
when each column C1, C21, and C22 operates at its preferred split (note that the subscript min should be on every vapor flow).

too, and for the B component, we have wC21 C32

B,T + wB,T )
nents and M ) N pure products is found directly as the
zBF. None of the heavier C and D components are highest peak in the Vmin diagram for the feed.
present. The middle Underwood root (θB) is active in The expression for a peak is given in (3), so if we
both C21 and C32; thus, we have relate the vapor to the top of the Petlyuk arrangement,
the minimum vapor flow is given by
C21 C32
Ri(wi,T ) wi,T RiziF

( )
)
VI2 ) ∑ ) ∑ AB/CD
) VT,min Petlyuk
VT,min j Rizi
∑
min
i)A,B Ri - θB i)A,BR i - θB ) max for j ∈ {1, 2, ..., N - 1}
i)1R - θj
(6) F j i
(8)
At I3 we know that all of components A, B, and C are
passing, but none of the heavy D is. The root (θC) is where the N - 1 common roots (θ1, ..., θN-1) are found
active in all columns (C1, C22, and C33), and we get by the feed equation (2).
Conclusion: The minimum-energy solution for a
C1 C22 C33
Ri(wi,T + wi,T + wi,T ) generalized Petlyuk arrangement for N products and N
VI3
min ) ∑ Ri - θC
) feed components is given by the highest peak in the Vmin
i)A,B,C diagram.
RiziF This is exactly the same as the most difficult binary
∑
i)A,B,CR
ABC/D
) VT,min (7) split between two adjacent component groups in an
i - θC ordinary two-product distillation column.
The result directly generalizes what was shown for
Again we recognize these expressions as the vapor the three-product Petlyuk column in part 2 of the series.
flow at the three peaks in the Vmin diagram for the In section 4, we will show how to generalize this to any
prefractionator feed equation (3). number of feed components (N) and products (M e N)
3.1. Vmin for N Feed Components and N Pure with possibly nonsharp specifications. However, before
Products. The results for the four-component and four- we move on the general case, let us discuss some more
product system given above is easily extended to any properties of the solution by the following example.
number of products. On the basis of the same procedure, 3.2. Example: Four Components and Four Pure
we obtain that the maximum minimum vapor flow Products. Feed data for this example are given as F )
requirement through any horizontal cross section in a 1, q ) 0.8, z ) [0.25, 0.25, 0.25, 0.25], and R ) [14, 7, 3,
generalized Petlyuk arrangement with N feed compo- 1]. The feed composition (zi), relative volatilities (Ri), and
620 Ind. Eng. Chem. Res., Vol. 42, No. 3, 2003

Table 1. Data for Peaks and Knots in the Vmin Diagram

PAB sharp A/B PBC sharp B/C PCD sharp C/D PAC B distributing PBD C distributing PAD preferred split
VT,min 0.8975 0.9585 1.0248 0.6350 0.7311 0.5501
D 0.2500 0.5000 0.7500 0.3663 0.5839 0.4490
ri,T 1, 0, 0, 0 1, 1, 0, 0 1, 1, 1, 0 1, 0.47, 0, 0 1, 1, 0.34, 0 1, 0.57, 0.22, 0

Figure 5. Composition profiles for the Petlyuk arrangement in Figure 1. Each column is operated at its preferred split with vapor flows
and product splits taken from Table 1 data as shown in the Vmin diagram in Figure 4. Observe the pinch zones in all junctions and how
one component is removed in each column end.

recoveries (ri,T in the table) are given for components identical to the feed stage composition in the succeeding
A-D, respectively. column. In each section, the compositions of the remain-
3.2.1. Visualization in the Vmin Diagram. We have ing components increase monotonically from the feed
applied the general procedure from part 1 of the series12 pinch to the end pinch without any remixing. Note that
for computing the numerical values for minimum energy if a column had its own reboiler and condenser, remixing
for a sharp split between each possible pair of key at the end is inevitable.
components (peaks and knots), and the results are given
in Table 1.
Observe in Figure 4 how the vapor flow in each 4. General Vmin for N Feed Components and M
individual column in Figure 1 appears as a difference Products
between the peaks and knots. Thus, for preferred split
operation in each column, all internal flows and com- For each extra product, we have to add another array
ponent recovery can be found from the data in Table 1. of columns to the structure in Figure 3. The total
The relations are quite trivial and come from the number of internal directly coupled two-product col-
material balance equations at the column junctions. umns to separate M products is (M - 1) + (M - 2) + ...
3.2.2. Composition Profile. A composition profile + 2 + 1 ) M(M - 1)/2. There are M - 1 product splits,
from a simulation example is shown in Figure 5. There and these can be related to M - 1 minimum-energy
are 30 stages in each column section (N ) 60 in each operating points (peaks) in the Vmin diagram.
column), and in practice, this is close to an infinite However, we have often more components (N) in the
number of stages for this case (with purity requirements feed than the number of products (M). Thus, we have
of around 99.9%). The flow rates are taken from the Vmin to consider a split between products, which may be
diagram in Figure 4 and are applied directly in the specified as an aggregate of components. Fortunately,
simulator. This simulation is a practical confirmation the characteristic of minimum-energy operation is
of the analytical expressions for flows and pinch zones unchanged. Each internal two-product column should
and for the minimum-energy behavior. only separate the components belonging to the most
Observe the characteristics of a preferred split pinch extreme products in its feed (in terms of relative
zone at all feed junctions and that one component is volatility).
completely removed in the end of each column. Note also For example, in the case of nonsharp separation of
that the pinch zone composition in each column end is the light A in column C21, the expression for the vapor
 Ind. Eng. Chem. Res., Vol. 42, No. 3, 2003 621

Figure 6. Assessment of minimum vapor flow for separation of an eight-component feed (ABCDEFGH) into four products (WXYZ). The
plot shows the Vmin diagram for the feed components (solid), and the equivalent diagram for the products (bold dashed) is easily obtained
from the product split specifications given in Table 2 or 3.

Table 2. Specification of Feed Component Recoveries in

flow through intersection I1 becomes [refer to (5]: Products W, X, Y, and Z
light key heavy key
RAwC31
A,T aBwC31
B,T impurity impurity
VI1 ) + (9) product specification components specification comment
min
RA - θA RB - θA
W A, B 0% C all of A, any
amount of B
Here we need to express the product specifications in X 0% A B, C, D, E <10% E the rest of B
terms of net flows (w) for the components appearing in Y <10.0% D D, E, F 0% G
each product stream. It is possible to continue with the Z 0% F G, H sharp F/G split
other intersections and deduce the exact minimum-
energy expression equivalent to (8), but it is much more The highest peak determines the maximum minimum
simple to illustrate the solution in a Vmin diagram as in vapor flow requirement in the arrangement. In this
the example below. example this is the middle peak PXY, which is directly
4.1. Example: Four Products and Eight Feed related to column C32 (note that VT,min values in Table
Components. A Vmin diagram for M composite products 3 are for the given split in a two-product column and
can easily be drawn into the general N-component that the required flow in the individual columns appears
diagram. The procedure is similar; we compute the as we have shown in Figure 4). With a single reboiler,
peaks and knots in the diagram from the minimum- all of the heat for vaporization has to be supplied in the
energy operation given by a sharp split between each bottom, and because the other peaks are lower, columns
possible pair of products. Note that this does not mean C33 and C31 will get a higher vapor load than required.
a sharp split between individual components if some However, with heat exchangers at the side-stream
components are allowed in more than one product. stage, we only have to supply heating for the require-
In Figure 6 we illustrate for a given example how to ment given by PYZ in the bottom reboiler and heating
use the Vmin diagram to assess minimum-energy opera- for the difference between PYX and PYZ in the bottom of
tion when M < N. The diagram (solid) is drawn for a C32, which is at a lower temperature. We may also take
given eight-component feed (ABCDEFGH), which shall heat out because of the difference between PXY and PWX
be separated into four products (WXYZ) in an extended
above C32.
four-product Petlyuk arrangement (Figure 3). The prod-
uct specifications are given in Table 2. On this basis, Observe that PXY and PABare of similar height. This
we can specify the required 2 degrees of freedom for each implies that we are able to separate the light component
possible pair of product splits in a single two-product A as a pure product in the top with a vapor flow
column. The resulting split specifications are given in requirement similar to that given by PXY. Thus, we can
Table 3, and the minimum-energy solution for each split see directly from the diagram that we may change the
(I/J) gives us the peaks and knots (PIJ) in the Vmin specification of product W to be pure A without consum-
diagram for the M products shown (bold dashed) in ing any more energy (but then we cannot take out any
Figure 6. heat above C32 of course).
622 Ind. Eng. Chem. Res., Vol. 42, No. 3, 2003

Table 3. All Possible Product Split Specifications, by Two Key Recoveries

light key heavy key
split co1umn in the top in the top VT,min comment
W/X C31 100% A 0% C 0.9632 sharp A/C split, B distributes
X/Y C32 >90% D <10% E 1.3944 nonsharp D/E split
Y/Z C33 100% F 0% G 1.2093 sharp F/G split
W/Y C21 100% A <10% E 0.5569 sharp bottom, nonsharp top
X/Z C22 <90% D 0% G 0.7477 nonsharp top, sharp bottom
W/Z C1 100% A 0% G 0.4782 “preferred split” A/G, not A/H

The diagram also illustrates that nonsharp product single liquid stream, then g(u) ) VC21 B - VC22
T ) 0
specifications can be handled quite easily. Note how the expresses the constraint.
peak PXY follows the contour lines for rE,T ) 0.1 and rD,T With given feed properties, (F, R, z, q), and sharp
) 0.9. product split specification, the optimization criterion can
The same example could be used for cases where M be expressed as
) N too.
As a last comment on our example, observe that the Jopt ) min J(u)
“preferred” split is at PWZ. We put “preferred” in quotes u

( )
because we have earlier defined the preferred split at
minimum energy for the most extreme component split rPi
i ) 1
which would be A/H here. However, because H never subject to constraints rPi j ) 0 ∀ i * j (12)
needs to be separated from the other components, we
do not need that split. Instead, we only separate g(u) ) 0
products W and Z in the prefractionator (C1), which Here Pi denotes product number i.
really is a split between components A and G. Thus, we 5.2. Requirement for Feasibility. The feasible
may say that PWZ represents the preferred split for our region is the operation region where we have fulfilled
four aggregate products. the operational constraints in (12). Here we only con-
sider the pure product specifications and no additional
5. Verification of the Minimum-Energy Solution constraints (no g(u) ) 0).
In section 3 we found the analytical expression for Then feasible operation requires operation on or above
the vapor flow requirement for the generalized Petlyuk the V-shaped boundary in the Vmin diagram for each
arrangement when all internal columns are operated column. For example, in the four-component example,
at their respective preferred splits, but we have not yet the feasible region for the prefractionator is on or above
proved that this expression really represents the mini- PAB-PAC-PAD-PBD-PCD. Note that the Vmin diagram
mum-energy solution for the extended Petlyuk arrange- for the succeeding columns only overlaps the prefrac-
ment. tionator diagram when this is operated at its preferred
Here we formulate minimization of energy as an split. In other cases, we must find the new Vmin diagram
optimization problem and verify that the solution given for each column, given by the actual Underwood roots
in equitation (8) (the highest peak) really is optimal for for the preceding columns.12
the extended Petlyuk arrangement. We will do this by This is easy to show by the following argumentation
two steps: first by determining the feasible region of for the four-product column:
operation for the given product specifications and second Assume first close to preferred split operation in all
by showing that no changes in any degrees of freedom columns. Then change the operation of C1 so we allow
within the feasible region may reduce the minimum some light A to be transported downward in C1 and into
vapor flow requirement. C22. This A have to be transported upward in C22
We will limit the presentation to N components and because it is more volatile than B, which also is
M ) N pure products. However, the result will also be transported upward, and then some amount of A has
valid for the general case, e.g., the example in section to be present at the feed junction to C32. A portion will
4. have to enter C32, and because A is still more volatile
5.1. Minimum Vapor Flow as an Optimization than B, it will also be transported upward in C32 and
Problem. We formulate the criterion function as the will appear in the product stream from the junction C31/
maximum of the minimum vapor flow requirements C32, where we have specified a pure B product.
through any of the intersections I1, I2, ..., I(M-1). We may do this “experiment” with a sloppy split for
any of C1, C21, and C22. In all of these columns, the
J(u) ) max (VI1, VI2, ..., VI(M-1)) (10) most heavy feed component for every column has to be
fully removed in the top, and the most volatile has to
Here u represents our degrees of freedom in operation, be fully removed from the bottom in order to obtain
and we have, in general, 2 degrees of freedom for every sharp product splits in the final columns of the se-
column, e.g., expressed by (D, V) for each. Thus quence.
5.3. Verification of the Optimal Solution. In the
dim(u) ) M(M - 1) (11) following we will show that it is not possible to obtain
any reduction in minimum vapor flow by changing the
The main constraints are given as the final product operation in a way from the preferred split (inside the
(Pi) specifications. We may also treat arrangements with feasible region).
a lower number of degrees of freedom, by specification An important characteristic of the direct coupling is
of a set of flow constraints, expressed as the equality that the actual Underwood roots in a column section (φ
g(u) ) 0. An example is if we restrict the feed to column in tops and ψ in bottoms) carry over as a common root
C32, in the four-product column in Figure 1, to be a (θ) to the succeeding column.4,6 We combine this with
 Ind. Eng. Chem. Res., Vol. 42, No. 3, 2003 623

Underwood’s minimum-energy results which states that and C are distributed to both products. However, it is a
for a given column φi g θi g ψi+1. bit more complicated if C21 or C22 is operated outside
Consider now the top of the four-product arrange- the region where θB is active. Then, the resulting root
ment. It is clear that the first roots in the columns C1, in C32 will be different from the corresponding root in
C21, and C33 have to obey C21, and the expression for the total flow through
intersection I2 will be more complicated than that for
φC31
A g θC31
A ) φC21
A g θC21
A ) φC1 C1
A g θA ) θA (13)
the case in (6).
Assume now that we keep the vapor flows and product
The vapor flow in the top of C31 is generally ex- splits constant in columns C1 and C22. Thus, any
pressed by φC31 change in vapor flow through intersection I2 must come
A ; thus, we obtain
through the bottom of C32 so
R AzAF RAzAF R AzAF
VC31
T ) g g g ∆VI2 C32
min ) ∆VB,min (17)
RA - φC31
A RA - φC21
A RA - φC1A
R AzAF A/B This can be expressed by the common Underwood root
) VT,min (14) in C32 and the amount of the C component into the feed
RA - θA
junction of this column.
This expression shows that there is no way to operate
column C1, C21, or C31 so that the vapor flow require- C32
RCwC32
C,F
VB,min ) (18)
ment in the top of C31 is reduced below the minimum, θC32 - RC
B
which is given by the peak PAB in the Vmin diagram. The
minimum solution is only obtained when we operate When the product splits in C21 and C22 are kept
column C1 in a region where θA is active. This is only constant, this vapor rate depends only on the behavior
obtained along the curve PAD-PAC (really also along of the common Underwood root in C32, which is given
PAC-PAB, but then we remove component C and not only as the solution of its feed equation:
D in the top of C1, and then we might remove column
C21 completely). In addition, C21 must also keep θA
RBwC32 RCwC22 C21
C,T - wC,B
active, which is obtained along PAC-PAB, and at last, VC32 )
B,F
+ ) VC22 - VC21
F C32 T B
C31 must be operated exactly at PAB. This line of RB - θ RC - θC32
argumentation is easy to extend to the general N- (19)
component N-product case.
Operation of columns C22, C32, and C33 has no direct Note that the net component feed rate to C32 is given
impact on PAB; thus, there is no way to operate these directly from the material balance at the junction:
C21 C22 C21
columns to reduce the peak PAB. This shows that the wi,F ) wi,T - wi,B . We assume that C22 is operated at
peak PAB represents the absolute minimum vapor flow its preferred split. Thus, θB is active in C22. In C21, we
for the top of the Petlyuk arrangement also for other may have operation outside the active θB region; thus,
operation points than the preferred split for each we have to use the actual root ψC. The right-hand side
internal column. of (19) can now be written as
Similarly, in the bottom of columns C1, C22, and C33,
we have RBwC21
B,B RCwC21
C,B RBwC22
B,T
VC22 - VC21 ) + + +
T B C21
RB - ΨC RC - ΨCC21 RB - θB
ψC31 C33
D e θC ) ψC21 C21
D e θC ) ψC1 C1
D e θC ) θC (15)
RCwC22
C,T
which gives (20)
RC - θB
RDzDF RDzDF RDzDF
VC33
B ) C31
g C21
g g By careful inspection of the structure of the feed
ψD - RD ψD - RD ψC1
D - RD equations (19) and (20), we observe that we always have
RDzDF ABC/D
(∂/∂ψC21 C32
C ) θB > 0 and that the solution has to obey
) VB,min (16)
θC - RD
θB g θC32
B g ψC21
C (21)
Thus, all of the bottom columns have to be operated
with θC. For a sharp split, this is only obtained for C1 Thus, we have that, for suboptimal operation of C21,
along PAD-PBD, C22 along PBD-PCD, and C33 at PCD. the actual Underwood root ψC21 C decreases from its
Thus, PCD represents the minimum vapor flow in the original optimal value ψC21
C ) θB. Because of the struc-
bottom of the Petlyuk arrangement for any operation ture of (20), the important Underwood root θC32 B also
of the arrangement. decreases, and from (18) and (19), we see that the flow
It is important to note that we have to operate column through the intersection I2 must increase.
C1 exactly at its preferred split (PAD) to avoid increased We may similarly analyze the operation of C22
vapor requirements in C31 or C33. Thus, operation of outside the region where θB is active and get to the
C1 in the region above the preferred split will increase conclusion that this will also increase the vapor rate
the vapor requirement represented by the peak PAB or through the cross section I2.
PCD. It is clear that this result is independent of any
However, column C1 has no such direct impact on the changes in distribution of B and C components from
middle peak PBC. The only requirement is that the root columns C1, C21, and C32. For each distribution case,
θB is active because this root has to carry over to C33 we may start with θB active in both C21 and C22. Then
via both C21 and C22. This is trivial as long as both B any operation outside the active region in either C21
624 Ind. Eng. Chem. Res., Vol. 42, No. 3, 2003

Figure 7. Vmin diagram for four-component feed ABCD with optimality regions for operation of columns C1, C21, and C22. The contour
lines for constant φA and a given constant θC21
B which makes VA/BCD ) VAB/CD ) VABC/D
min are shown (dashed). These boundaries are the
upper bounds for the optimality regions.

or C22 or both will lead to an increase in the required because both affect the cross section I2 through the
flow through intersection I2. Underwood root (θC32 B ) given by (19). In Figure 7 we
We have not carried out a detailed proof for the have indicated the operation at PBD. Then we may find
general N-component M-product case for other than the the optimality region for C21 in the marked region
far left and right peaks. However, we expect that this above PAC. Note how the operation of C1 limits the lower
can be done by the same line of argumentation as that part of the optimality region for C21 through the contour
in which we used to state that the middle peak cannot for constant φA through X.
be reduced for any feasible operation of C21 and C22.
Numerical evidence also supports this. 6. Minimum Heat Supply for All Adiabatic
If any part in the sequence of columns is operated Distillation Arrangements without Internal
away from the preferred split, the vapor flow require- Heat Exchange
ment in some of the cross sections has to increase; in
other words, one or more of the peaks related to the Petlyuk et al.14 showed that it is possible to devise a
specified product splits has to increase. In general, if a reversible Petlyuk arrangement with zero lost separa-
column has its preferred split at PXY and is operated tion work and thus requiring minimum separation work
above this point, all succeeding columns with knots and compared to any other separation process.
peaks related to either X or Y will, in general, be However, it has also been conjectured that the adia-
affected. Any suboptimal operation somewhere in the batic Petlyuk arrangement, where all of the heat is
arrangement cannot be recovered in the succeeding supplied in the bottom reboiler at the maximum tem-
columns. perature, requires minimum energy for vaporization
5.4. Flat Optimality Region. When the peaks are (Vmin) compared to any other adiabatic distillation
of different height, we may operate some of the columns arrangement (without internal heat exchange). (We
away from the preferred split as long as the highest apply the term adiabatic column section, as used by
peak is not affected, and the other peaks do not grow Petlyuk et al.,15 to denote a column section with
above this one. This gives rise to “flat” regions in the constant molar flows and no heat exchange along the
plot for the overall energy requirement, VPetlyuk
B , as a section. Thus, the directly coupled columns in Figure 1
function of the degrees of freedom. and also typical conventional arrangements contain
We illustrate this by an example in Figure 7. Because adiabatic sections.) However, no proof has been found
PCD is the highest peak, the optimality region for C1 is in the literature,16 except for the ternary case. For the
along PAD-PBD. However, somewhere the actual Un- ternary case Fidkowski and Krolikowski17 showed that
derwood root in the top of C1 related to the AB split the three-product Petlyuk arrangement always has a
will get a value which makes the peak PAB′ given by smaller vapor flow than any arrangements with side
φA,Bal equal the peak PCD. This line segment limits the strippers or rectifiers, and they showed that these also
optimality region for both columns C1 and C21, and this performed better than the conventional direct and
is very similar to the result from the ternary case indirect split sequences.
discussed in part 2 of the series. For the generalized adiabatic Petlyuk arrangement
Similarly, C22 has to be operated along PBD-PCD. in Figure 1, the minimum energy requirement for
This optimality region is limited by how C21 is operated separation of a feed mixture of Nc components is given
 Ind. Eng. Chem. Res., Vol. 42, No. 3, 2003 625

Figure 8. General column interconnection junction. The direct (full thermal) coupling gives θC21
A ) φA, which implies min [max (V , V )]
I1 I2

and a zero external heat exchange at the interconnection (Q ) 0).

by (8). Note that all of the heat can be supplied in the The most important effect of the direct coupling is
bottom reboiler and can be removed in the top con- that the Underwood roots in the top of C1 “carry over”
denser, but in some cases, some of the heat may be as the common (minimum-energy) Underwood roots for
supplied or removed at the product outlets. C21. Thus, θC21
A ) φC1
A , which is vital in the following
In the following we consider adiabatic column sec- analysis.
tions, and we conjecture that the adiabatic Petlyuk For a given operation of the first column (not neces-
arrangement is indeed the best distillation arrangement sarily at minimum energy), the vapor flow and net
when we regard the total requirement for vaporization component flows in the top can be related to a certain
at constant pressure and when we do not consider any Underwood root (φ), here given by the defining equation
internal heat exchange within the arrangement. in column C1 (we omit the superscript C1 on w and φ):
6.1. Direct Coupling Giving Minimum Vapor
Flow. First we will show that the direct (fully thermal) R A wA RBwB R C wC
coupling minimizes the vapor flow requirement through VC1
T ) + +
RA - φA RB - φA RC - φA
any column junction.
Let us consider a general junction at the top of the (note DC1 ) ∑wi) (23)
prefractionator (C1) and the succeeding column (C21)
as illustrated in Figure 8. To simplify, we assume a Consider now any type of interconnection to the
ternary feed, but similar results can be obtained for any succeeding column (C21). At minimum-energy operation
number of components and at any junction in an in C21, the flow rates are determined by the component
arrangement. distribution and the common Underwood roots. Thus
We assume that the 2 degrees of freedom in column
C1 (e.g., DC1 and VC1
T ) are fixed. In work by Halvorsen,
7
C21
RAwC21
A RBwC21
B RCwC21
C
we showed that the composition in the recycle flow VT,min ) C21
+ C21
+
RA - θA RB - θA RC - θC21
(LC1
T ) from C21 to C1 normally has no effect on the net
A
component flows from C1 to C21. This is so unless a
component which would have been removed in an C21
RA(wC21
A - w A) RB(wC21
B - wB)
ordinary column (with a condenser) is not introduced VB,min ) + +
RA - θC21
A RB - θC21
A
in the recycle flow to the directly coupled column. For
reasonable operation of the system, this will normally RC(wC21
C - w C)
not be a problem. (24)
At the interconnection to C21, we allow for supply or RC - θC21
A

removal of heat (still with fixed DC1 and VC1T ). This will
The common Underwood roots can be found from the
then only affect the effective liquid fraction (qC21) to
feed equation of C21 (25) and will depend on the
column C21 and have no impact on the component flows
external heat through the feed quality. The net compo-
(wC1
i ). Recall that direct coupling implies that the nent flow and net distillate flow in C1 are constants.
reflux in C1 is taken directly as a side draw from C21
and that the vapor flow from C1 is fed directly to C21. R A wA R B wB RCwC
In this case the external heat exchange is zero, and we
C21
+ C21
+ ) (1 - qC21)DC1
obtain an equivalent liquid fraction given by RA - θ RB - θ RC - θC21
(25)
qC21
dc )1- VC1
T /D
C1
(22) Note that, for any reasonable operation of columns,
all net component flows are positive in the top sections
Note that we always have qC21 < 0 with direct and negative in the bottom sections. This implies that
coupling, which is equivalent to a superheated vapor the minimum vapor flow in the top section will increase
feed. Heat removal (e.g., a condenser) will increase qC21, as the common Underwood root increases and the vapor
and heat supply (superheater) will decrease its value. flow in the bottom section will decrease.
626 Ind. Eng. Chem. Res., Vol. 42, No. 3, 2003

In the following we fix the operation of column C1 tion. For example, when a side-stripper configuration
such that VC1 T and all wi, and thereby all φj, are is used instead of an indirect split configuration, the
constant, and we want to find the value of the common direct coupling replaces a condenser (which, in practice,
Underwood root in C21 (θ21 has an inevitable loss). This is probably the background
A ) which minimizes the
maximum vapor flow rates through any of the intersec- for the term “full thermal coupling” used by many
tions above or below the feed junction (see Figure 8): authors. However, here we use the term “direct cou-
pling”, which relates to the fact that both the vapor and
min [max (VI1, VI2)] (26) liquid flows are coupled directly between two columns.
θC21 In addition, we obtain reversible mixing at the junctions
A
when we keep the vapor and liquid flows in the
where junctions at equilibrium.

C21
VI1 ) VT,min and VI2 ) VC1 C21
T + VB,min (27) 7. Discussion

A typical dependency of VI1 and VI2 as a function of 7.1. Alternative Configurations. There exists a
θC21 is shown in Figure 8, and we see that the analyti- very large variety of possible realizations for extended
A
cal solution is given by Petlyuk arrangements,18 which are equivalent in terms
of energy requirement. For example, in a recent paper,19

{
arg min
θC21
A
}
[max (VI1, VI2)] ) φA (28) it is shown that, for a four-product column, sections can
be arranged together in 32 different configurations. For
the five-product column, the number is 448 configura-
Proof. For normal operating conditions, we have tions which are equivalent in terms of the minimum
energy requirement. There are, of course, many impor-
dVI1/dθC21
A > 0 and dVI2/dθC21A < 0. tant differences, i.e., in how easy it is to set individual
This implies that min [max (VI1, VI2)] is found when vapor and liquid flow rates in practice, how the column
VI1 ) VI2. arrangement behaves for nonoptimal operation, how
By applying θC21
A ) φA in (23)-(27), we obtain VI1 ) easy it is to control, the possibility for operation at more
I2
V . than one pressure level, practical construction issues,
Q.E.D. etc.
In conclusion, minimization of the vapor rate through 7.2. Practical Petlyuk Arrangements (Four-
any intersection (I1 or I2) is found when the common Product Dividing Wall Column, DWC). In the Pet-
Underwood roots in column C21 equal the actual roots lyuk arrangement in Figure 1, we assume that we can
in the top section of C1. This is exactly what we obtain adjust the vapor and liquid flows individually in all
with a direct coupling. Note that the proof does not columns. The more practical arrangement in Figure 9
require the first column to be operated at minimum is a bit less flexible because all of the vapor flow has to
energy and that it is valid for any distribution of come from the bottom reboiler, and similarly, the liquid
components in C1. flow comes from the top condenser. It will generally
6.2. Implications for Side Strippers and Recti- have a higher energy requirement, although it may be
fiers. A direct implication of the result in section 6.1 is the same in some cases (see example). Because we
that arrangements with side strippers (like in Figure 8 extract only liquid side-stream products, also in the
with a direct coupling) or side rectifiers will always have junction into the feed of C32, we get a simpler config-
a lower total need for vaporization than the correspond- uration, which also may be implemented as a DWC in
ing indirect split or direct split configurations. This was a single shell, as indicated in Figure 9b.
also shown by Fidkowski and Krolikowski17 for the
ternary case, but it is straightforward to extend the However, operation is by no means simple, and we
result in section 6.1 to the general multicomponent case. still have nine manipulated inputs left, and when four
6.3. Adiabatic Petlyuk Arrangement as Optimal. are used for product purity, there are five left. These
The result in section 6.1 gives rise to the following must be set properly in order to achieve the optimal
conclusion: operation given by the highest peak in the Vmin diagram.
We assume constant relative volatilities, constant The cross-sectional area is usually designed for a
molar flows, constant pressure, and no internal heat maximum vapor load. We know that there may be large
integration. Then the generalized adiabatic Petlyuk differences between each section, e.g., in C31 from
arrangement has the lowest need for vaporization com- Figure 2. However, in cases where the peaks are similar,
pared to any other adiabatic distillation arrangement we know that the total vapor requirement is similar in
for separation of an arbitrary feed mixture into its pure any cross section (I1, I2, or I3). Thus, as indicated in
components. Figure 9b, the DWC can be implemented in a single
This result is based on the simple argument that, at shell with a constant diameter and with quite different,
any junction where we might consider another type of but suitable, cross-sectional areas for the internal
connection than the direct coupling, the required vapor columns. This is one issue which makes DWC imple-
flow through the junction, and thereby through a cross mentations attractive.
section of the whole arrangement, will increase. We assume the same feed and Vmin diagram as in
We have not presented a complete proof, so the above Figure 4. We start by determining the requirement of
conclusion is a conjecture. However, for the ternary case, the prefractionator (C1). The original diagram is, of
it has been proved by Fidkowski and Krolikowski,17 course, valid for C1, and we chose to operate C1 at its
when considering conventional arrangements and side preferred split, which is at PAD. Then all of the common
strippers as alternative configurations. roots from C1 carry over to C21 and C22. However, in
A qualitative explanation is that the direct (full Figure 9 we have the restriction V21 22
B ) VT . Here
thermal) coupling can be regarded as ideal heat integra- column C22 controls the vapor requirement because
 Ind. Eng. Chem. Res., Vol. 42, No. 3, 2003 627

Figure 9. Practical four-product Petlyuk arrangements with some flow restrictions. We allow only liquid feed to C32 and liquid intermediate
side products B and C.

Figure 10. Vmin diagram for four-component feed ABCD with the less flexible Petlyuk arrangement in Figure 9. Vertical arrows are
vapor flow requirements in each column section. Feed data: z ) [0.25, 0.25, 0.25, 0.25], R ) [14, 7, 3, 1], and q ) 0.8.

21
VB,min < V22
T,min. Thus, the minimum vapor for column
but as illustrated in the figure, this gives a higher vapor
C21 is somewhere on the line between the points X and flow requirement than PCD, which was our original
Y in Figure 10. First we try to operate C21 in X. Then highest peak. However, here we may increase the net
the root θA carries over all the way to C31, and the vapor product flow from C21 and move operation to Z. In this
flow requirement will be given by PAB. However, θB will case V >Vmin in C21, and none of the common roots are
not carry over to C22. Instead, a larger root will carry active. Both C31 and C32 will be affected, and the new
over, and the requirement for C32 will be given by P′′BC, minimum vapor requirements are given by P′AB and
628 Ind. Eng. Chem. Res., Vol. 42, No. 3, 2003

at the preferred split but at a sharp B/C split, which is

given by the middle peak (PBC). If this is the highest
peak, it is obvious that the Kaibel column requires a
higher reboiler vapor rate because it requires this vapor
rate for C1, and we must, in addition, have some vapor
flow for the separation of C/D in the top of C22. If PBC
is not the highest peak, we observe that when C1 is
operated at PBC, none of the common roots θA and θB
are active in C1 and cannot carry over to C21 or C22.
Then, as shown in section 5.3, the expressions for
minimum vapor in each of C21 and C22 have to be
higher than the peaks PAB and PCD.
However, in cases when the B/C split is simple (PBC
is much lower than PAB and PCD), the difference in
energy requirement can be small, and the simpler
configuration may be preferable with respect to capital
cost and more simple operation.
7.4. Required Number of Stages: Simple Design
Rule. The proposed stage design for ternary Petlyuk
arrangements given in part 2 of this series can be
applied for the extended arrangements too. We can
calculate the pinch zones in all junctions for all columns
Figure 11. Kaibel arrangement for separation of a four- at the preferred split. This is trivial when we know all
component feed. flow rates and component distributions from the Vmin
diagram. Then a minimum number of stages (Nmin) can
P′BC, respectively. In this example, we get a resulting be found from the Fenske equation for each section for
diagram where PCD still is the highest peak, and the a given impurity of the component to be removed in that
minimum vapor flow requirement for this less flexible section. This impurity can be set according to the
Petlyuk arrangement is the same as the fully flexible impurity requirement in the products. The simple
arrangement. It is quite clear, however, that we may design rule N ≈ 2Nmin will typically give a real mini-
use another feed and find cases where the less flexible mum vapor flow (VR,min) in the range between 5 and 10%
arrangement can never reach the minimum require- above Vmin found for an infinite number of stages, for
ment of the fully flexible configuration. For example, if the same separation.
the peak PCD were at the same height as PBC in Figure
This simple design rule may, of course, be adjusted
10, then either of the peaks P′AB or P′BC would be higher
by more rigorous column computations and cost func-
than the original three peaks for any operation of C21
tions.
along the line Y-Z.
In summary, the solution is still simple to find by the 7.5. Control. M-Product columns will, of course, be
Vmin diagram, but we get new peaks for the columns more complicated than the more familiar ternary Pet-
where the preceding column cannot operate at its lyuk arrangements. However, the characteristic of
preferred split. This can be done accurately by Under- optimal operation is similar and is given by keeping
wood’s equations, but we can also look directly at the each individual column at its preferred split.
diagram and find an approximate solution graphically. By keeping the impurities of the components to be
Note how the peak P′AB rises and P′BC falls as the removed in each section at setpoints fixed at small
operation of C21 is moved on the line from X toward Y. values, we ensure that the operation is at the preferred
Another important lesson is that we may change split, even if we do not know the feed. The magnitude
operation in some parts of the arrangement within the of the allowed impurity setpoints in intermediate col-
optimality region without affecting the highest peak. umns should be set according to the allowed impurities
The extent of this region is dependent on how different in the final products.
the peaks are and the practical impact is that some of 7.6. Real Mixtures. The characteristic of the optimal
our degrees of freedom do not need to be set accurately, solution, with the ideal assumptions about mixture and
only within a certain range. properties used in this paper, is that every internal
7.3. Kaibel Column or the “|- Column”. The column has to be operated at its preferred split. How-
Kaibel column20 is a directly coupled arrangement for ever, this is most likely to be the characteristic of the
separating four components as shown in Figure 11. The optimal solution for real mixtures too. Then there will
interesting part is the extra column section (C2x where be pinch zones across every feed junction and in the end
L ) V) for separating B/C in the main column. However, of the preceding column, as shown in the simulation
the sharp B/C split is performed already in the prefrac- example for the ideal case in Figure 5. The actual pinch-
tionator (C1), so section C2x is really not needed and zone compositions and the flow rates have to be calcu-
can be replaced by heat exchange between the bottom lated numerically because the relative volatilities and
of C21 and the top of C22, denoted the “3/4|- column” molar flows may change along the column sections. At
by Christiansen.5 the internal mixing junctions, e.g., at the feed junction
The minimum vapor flow requirement in the Kaibel to column C32 in Figure 1, we may get some losses
column is always outperformed by the full Petlyuk because there may be different compositions of the flows
arrangement in Figure 1. This is simple to see from the into the junctions even for preferred split operation of
Vmin diagram (e.g., in Figure 2). In the Petlyuk arrange- every column. Thus, the characteristic of the real
ment, the overall vapor requirement is given by the minimum-energy solution may deviate a little from the
highest peak. In the Kaibel column, C1 is not operated ideal case, but it is expected to be close. It is also possible
 Ind. Eng. Chem. Res., Vol. 42, No. 3, 2003 629

that some of the alternative configurations for the same assumed constant pressure and that we have not
number of products (refer to section 7.4) may be a little considered any internal heat exchange inside the sys-
better than the others for a particular case. tem.
The Underwood roots will still carry over at the The latter may, as shown in work by Halvorsen,7 give
junctions, but because of the changes in relative volatili- some further energy savings.
ties and molar flows, the actual Underwood roots at any
cross section will also vary along the column, even for Literature Cited
minimum-energy operation. This implies that the Vmin
diagram for the succeeding columns will not exactly (1) Fidkowski, Z.; Krolikowski, L. Thermally Coupled System
of Distillation Columns: Optimization Procedure. AIChE J. 1986,
overlap the Vmin diagram for the feed to the first column. 32 (No. 4).
However, we expect that the final solution for the (2) Glinos, K. N.; Nikolaides, I. P.; Malone, M. F. New complex
internal flows will not be very far from the ideal case, column arrangements for ideal distillation. Ind. Eng. Chem.
as in the example illustrated in Figure 4. Process Des. Dev. 1986, 25 (No. 3), 694-699.
(3) Carlberg, N. A.; Westerberg, A. W. Temperature-Heat
Diagrams for Complex Columns. 3. Underwood’s Method for the
8. Conclusion Petlyuk Configuration. Ind. Eng. Chem. Res. 1989, 28, 1386-1397.
We have shown that the minimum-energy results for (4) Carlberg, N. A.; Westerberg, A. W. Temperature-Heat
Diagrams for Complex Columns. 2. Underwood’s Method for Side-
the three-product Petlyuk arrangement can indeed be strippers and Enrichers. Ind. Eng. Chem. Res. 1989, 28, 1379-
extended to general multicomponent, multiproduct Pet- 1386.
lyuk arrangements. An exact analytical expression, (5) Christiansen, A. C. Studies on optimal design and operation
which is based on the Underwood equations, has been of integrated distillation arrangements. Ph.D. Thesis, Norwegian
derived. The solution is very simple to visualize in the University of Science and Technology (NTNU), Trondheim, Nor-
Vmin diagram for the feed, given by the following rule: way, 1997 http://www.chemeng.ntnu.no/∼skoge/publications/thesis/
1998/christiansen/.
The minimum total vapor flow requirement in a (6) Halvorsen, I. J.; Skogestad, S. Minimum Energy Consump-
multicomponent, multiproduct Petlyuk arrangement is tion in Multicomponent Distillation. 2. Three-Product Petlyuk
determined by the highest peak in the Vmin diagram. Arrangements. Ind. Eng. Chem. Res. 2003, 42, 596-604.
Alternatively, because the Vmin diagram originally just (7) Halvorsen, I. J. Minimum Energy Requirements in Complex
characterizes a two-product column with a multicom- Distillation Arrangements. Dr. ing. Thesis, Norwegian University
ponent feed, this may also be expressed as follows: of Science and Technology (NTNU), Trondheim, Norway, 2001.
Available from the web page of Sigurd Skogestad, Department of
The minimum total vapor flow requirement in a Chemical Engineering at NTNU (May 2002: http://www.chem-
multicomponent, multiproduct Petlyuk arrangement is eng.ntnu.no/∼skoge/publications/thesis/2001/halvorsen/).
the same as the required vapor flow for the most difficult (8) Underwood, A. J. V.; et al. Fractional Distillation of Ternary
split between two of the specified products if that Mixtures. Part I. J. Inst. Pet. 1945, 31, 111-118.
separation is to be carried out in a single conventional (9) Underwood, A. J. V.; et al. Fractional Distillation of Ternary
two-product column. Mixtures. Part II. J. Inst. Pet. 1946, 32, 598-613.
(10) Underwood, A. J. V. Fractional Distillation of Multi-
Note that the rule above applies to any feasible Component MixturessCalculation of Minimum reflux Ratio. J.
product specifications, both in cases with an equal Inst. Pet. 1946, 32, 614-626.
number of feed components and products and for any (11) Underwood, A. J. V. Fractional Distillation of Multi-
possible component grouping in the products in cases Component Mixtures. Chem. Eng. Prog. 1948, 44 (No. 8).
where the number of products is less than the number (12) Halvorsen, I. J.; Skogestad, S. Minimum Energy Consump-
of feed components. tion in Multicomponent Distillation. 1. Vmin diagram for a two-
product column. Ind. Eng. Chem. Res. 2003, 42, 605-615.
In addition to the overall vapor flow requirement, we (13) Stichlmair, J. Distillation and Rectification. Ullmann’s
find the individual vapor flow requirement for every Encyclopedia of Industrial Chemistry; VCH: Weinheim, Germany,
column section, directly from the same diagram. The 1988; Vol. B3, pp 4-1-4-94.
Vmin diagram is based on feed data only and was (14) Petlyuk, F. B.; Platonov, V. M.; Slavinskii, D. M. Thermo-
originally intended to visualize minimum-energy re- dynamically optimal method for separating multicomponent mix-
gions and distribution regions for all possible operating tures. Int. Chem. Eng. 1965, 5 (No. 3), 555-561.
(15) Petlyuk, F. B.; Platonov, V. M.; Girsanov, I. V. The design
points, in an ordinary two-product distillation column of optimal rectification cascades. Khim. Prom. 1964, No. 6, 45
with a multicomponent feed. (445)-53 (453).
The characteristic of the optimal solution is that all (16) Petlyuk, F. B. Private communication, 2000.
internal columns are operated on their respective (17) Fidkowski, Z.; Krolikowski, L. Minimum Energy Require-
preferred splits. In general, this requires that we can ments of Thermally Coupled Distillation Systems. AIChE J. 1987,
adjust 2 degrees of freedom in each internal column. 33 (No. 4), 643-653.
(18) Sargent, R. W. H.; Gaminibandara, K. Optimum Design
However, practical arrangements with less degrees of of Plate Distillation. Optimization in Action; Academic Press:
freedom may also reach the same minimum vapor flow. London, 1976.
The results have been derived for ideal assumptions, (19) Agrawal, R. A method to Draw Fully Thermally Coupled
but the main characteristics of the solution will be valid Distillation Column Configurations for Multicomponent Distilla-
for real mixtures too. tion. Trans. Inst. Chem. Eng. 2000, 78, Part A, 454-464.
Although arrangements with more than three prod- (20) Kaibel, G. Distillation Columns with Vertical Partitions.
Chem. Eng. Technol. 1987, 10, 92-98.
ucts may be feasible, the results for general M product
systems have mainly theoretical interest. The most Received for review October 19, 2001
important result is that we can find the minimum target Revised manuscript received June 3, 2002
value for the vapor flow required for separation of a Accepted June 24, 2002
multicomponent feed by distillation in directly coupled
arrangements. It is important to note that we have IE0108651