Chemical Engineering Science 99 (2013) 203214

Contents lists available at SciVerse ScienceDirect

Chemical Engineering Science

journal homepage: www.elsevier.com/locate/ces

Application of attainable region theory to batch reactors

David Ming n, David Glasser, Diane Hildebrandt
School of Chemical and Metallurgical Engineering, University of the Witwatersrand, 1 Jan Smuts Avenue, Braamfontein 2000, Johannesburg, South Africa

H I G H L I G H T S

Continuous attainable region theory can be applied directly to batch reactors.

Fundamental reactor types are interpreted as a particular fed-batch reactor.
This helps to suggest a recommended sequencing policy in the batch.
Both batch and continuous reactors can be improved by a single AR construction.
CSTR solutions may be hard to obtain depending on reaction kinetics and stability.

art ic l e i nf o

a b s t r a c t

Article history:
Received 1 March 2013
Received in revised form
10 May 2013
Accepted 1 June 2013
Available online 10 June 2013

The application of attainable regions (AR) to batch reactor congurations is presented. In this work, we
consider the problem of constructing an equivalent batch structure given an AR generated for a
continuous system from a specied feed composition and associated kinetics. The authors show that
by the appropriate transformation, results developed from a continuous AR may be used to form a
related batch structure, and thus improvement of batch reactor structures is also possible by use of the
attainable region.
& 2013 Elsevier Ltd. All rights reserved.

Keywords:
Attainable regions
Batch
Chemical reactors
Mathematical modeling
Optimization
Reaction engineering

1. Introduction
The problem of determining optimal operating policies in batch
operations is a popular one, and has been given a great deal of
attention in the literature, particularly due to the importance of
batch reaction in a large number of industrially relevant processes.
Batch reactors are used extensively in the production of a variety
of pharmaceutical (Davies and Gloor, 1971) and biological(Cheong
et al., 2007; Modak and Lim, 1992; Najafpour et al., 2004;
Senthuran et al., 1997) products, as well as in the wastewater
(Woolard and Irvine, 1995; Zwiener et al., 2002) and polymerization (Zeman and Amundson, 1965) industries, and are typically
considered to be rather more versatile than that of equivalent
continuous processes (Bonvin, 1998). Batch reactors also nd use
in the small-scale production of highly specialized products,
negating the benet otherwise obtained by large-scale production.

Corresponding author.
E-mail address: [email protected] (D. Ming).

0009-2509/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ces.2013.06.001

Moreover, batch reaction may be the only viable method of

producing certain products, such as when experimental work is
performed at lab-scale and pilot-scale.
It follows that optimization plays an important function in the
design and operation of these processes. Much attention has been
placed on the development of methods that seek to improve batch
performance, particularly with regards to the determination of
optimal operating conditions and scheduling strategies. Although
research into determining efcient batch structures exists (Allgor
et al., 1996; Aziz and Mujtaba, 2002; Capon-Garcia et al., 2011;
Mendez et al., 2006), a brief search of the current literature would
suggest that this area may not appear to be as popular as
traditional batch optimization. Even then, current methods for
determining optimal batch structures often rely on traditional
optimization methods (Allgor and Barton, 1999; Allgor et al., 1999,
1996).
For the past two decades, the papers of Feinberg and
Hildebrandt (1997), Feinberg (1999, 2000a, 2000b, Glasser et al.
(1987), Glasser and Hildebrandt (1997) have motivated the use of a
novel method of identifying optimal reactor congurations,

204

D. Ming et al. / Chemical Engineering Science 99 (2013) 203214

termed attainable region (AR) analysis. Determination of optimal

reactor structures using the AR is unique since synthesis is
achieved via a geometric interpretation of reactor congurations.
This approach has shown to be an effective alternative method to
optimal reactor synthesis, particularly with regard to systems
involving multiple side reactions or reactions with complicated
kinetics. Nearly all applications of the AR have been performed on
continuous reactors however; the single paper of Davis et al.
(2008) currently remains the only use of AR theory to batch
systems.
Our aim in this work will thus be towards improving batch
reactor structures with specic use of the AR and the associated
benets of this approach. This is done as follows: for a given set of
feed conditions and kinetics, a candidate AR may be generated and
interpreted, initially, in the form of a continuous reactor structure.
Once generated, the appropriate conversions to the batch setting
may be performed, providing an equivalent recommended batch
reactor structure. A key objective in this work has been towards
reusing the results and insight discovered from a continuous AR
structure, such that the same outputs might be obtained with an
equivalent batch system.
We will begin by rst providing a brief overview of AR theory,
and drawing a relevant set of results associated with continuous
equipment; this is done in Section 2. Section 3 investigates how
equivalent structures in batch may be derived from continuous
equipment. The correct choice and exact combination of continuous equipment is guided by AR analysis, and thus the development of the AR is central to the approach presented in this work.
Finally, examples and concluding remarks are given in Sections
4 and 5, respectively.

N is formed; we call this the rate eld. Reaction rates are typically
a unique function of the state of the system (temperature,
pressure, composition etc.). Thus, for each point in state space, a
unique direction vector may be determined with components
corresponding to the rate of formation of each species. Although
many reactor types may be formed that each separately utilize
r(C), movement through the rate eld will be different for each
type.
2.2.1.1. Mixing and convexity. Consider now two distinct
compositions C1 and C2. It is easy to show that any mixture
concentration, Cn, formed from a combination of the two must lie
on the straight line segment C1 C2 in concentration space Cn is then
a linear combination of C1 and C2, and mixing is a linear process as
a result:
Cn C2 1C1
C1 C2 C1
where 01.
The above reasoning can be generalized to a set of k compositions fC1 ; C2 ; ; Ck gX (Feinberg and Hildebrandt, 1997). A set X is
considered convex if and only if for any two points C1 ; C2 X, the
line segment C1 C2 is completely contained in X. Cn is considered a
convex combination of the set if there exists a set of k scalars {1,
2k} such that i0, i1,2,,k
k

i 1

i1

and
k

Cn i Ci
2. Overview of AR theory in continuous equipment
2.1. Motivation
The purpose of this section will be towards providing a brief
overview of AR theory for continuous equipment. The results
obtained will allow for the development of optimal batch structures. The geometric nature of AR analysis has been discussed in
numerous papers before (Davis et al., 2008; Feinberg and
Hildebrandt, 1997; Feinberg, 1999, 2000a, 2000b; Glasser and
Hildebrandt, 1997; Glasser et al., 1987; Godorr et al., 1994, 1999;
Hildebrandt and Glasser, 1990; Hildebrandt et al., 1990; Khumalo
et al., 2006; Nicol et al., 1997, 2001; Nisoli et al., 1997), and only a
brief overview of the theory shall be provided here as a result. For
the reader who may already be familiar with the existing theory
and results, this section can be skipped. We shall begin with a
basic understanding of reaction and mixing.
2.2. Both reaction and mixing are processes that may be interpreted
geometrically
2.2.1. Concentration, reaction and mixing
Let us assume that we have a set containing N species with
molar concentrations c1,c2,,cN. We call the N-tuple of species
concentrations the concentration vector, denoted by C [c1,c2,,
cN]T. C thus represents a unique coordinate in concentration space
that may indicate the instantaneous concentration within a
reactor, or the resulting composition formed from a combination
of mixing many efuent streams. Assume that we are now given a
system of reactions with a known set of kinetic expressions for
each component. In the same manner that a composition vector
may be formed by C, the rate of formation for each species is also
expressible by a rate vector, r(C). For each point in concentration
space, the rate vector can be computed and thus a vector eld in

i1

2.2.1.2. Convex hull. The convex hull of a set X is the smallest

convex set that contains X (Berg et al., 2008). Note that although
the assumption of constant density has been applied here to arrive
at a linear mixing law, the same result is achievable by use of mass
fractions instead. Mixing is therefore always viewed to be a linear
process, and thus the above analysis is equally applicable to nonconstant density systems.
2.2.2. The attainable region (AR)
The AR is the geometric representation of the set of all
achievable points for all possible reactor congurations obtained
by reaction and mixing alone. The AR is commonly represented in
concentration space, however, constructions involving residence
time are also found in the literature (Glasser et al., 1994; Godorr
et al., 1994; Hildebrandt and Glasser, 1990; Hildebrandt et al.,
1990). With the exception of unique conditions containing irregular kinetics, the AR is typically composed of surfaces generated
from both reaction and mixing processes. Reaction surfaces are the
result of distinct combinations of three specic reactor types
(described below). The particular combination of reactors and
their arrangement with respect to one another is termed a reactor
structure. Whereas reaction operations may be viewed to enlarge
the AR, the linear nature of mixing operations ensures that any set
of compositions may always be formed by a convex combination of
achievable points. The duty of mixing therefore serves to ll in
concave regions in space. Consequently, it is sufcient to dene
the AR only in terms of its boundary points, as mixing between
these points allows one to achieve all interior points of the AR.
It also follows that points located within a convex region may be
formed by arbitrary combinations of the boundary points, and
thus there is generally no unique combination of boundary points
that may be used to achieve points located within the region.

D. Ming et al. / Chemical Engineering Science 99 (2013) 203214

2.3. AR construction from three fundamental reactor types

2.3.1. Plug ow reactor (PFR)
It may be shown that no more than three fundamental reactor
types, together with mixing between structures, are needed to
construct the AR (Feinberg and Hildebrandt, 1997). These are the
plug ow reactor (PFR), the continuously stirred tank reactor
(CSTR) and the differential side-stream reactor (DSR).
PFRs are dened by the following expression.
dC
rC
d

Geometrically, the set of concentrations achieved by integration

of Eq. (1) produces a solution trajectory in concentration space. All
rate vectors evaluated along this trajectory are tangent to it, and
thus PFR solutions move along in the direction of the vector eld.
It is shown (Feinberg and Hildebrandt, 1997) that PFR trajectories
serve to act as highways to which the nal most states of the AR
boundary are achieved and therefore they assume an important role
in the overall optimal AR structure. In view of this, we nd that
optimal AR structures usually terminate in a PFR. PFRs allow one to
traverse along the outer most extremities of the AR boundary, and
as a result, play an integral role in optimal AR structures.
A collection of PFR trajectories that form part of the AR boundary
is often termed a manifold (Feinberg and Hildebrandt, 1997).
2.3.2. Continuously-stirred tank reactor (CSTR)
The CSTR is one in which the reactor contents is assumed to be
perfectly mixed, and therefore the exit stream of a CSTR is
assumed to be identical to that found inside the reactor. The
dening equation is given as
C C0 rC

CSTRs possess a unique geometric interpretation. If the mixing

vector is taken as CC0, then clearly the set of concentrations
satisfying Eq. (2) must be those points which are collinear to r(C).
Whereas trajectories corresponding to the PFR are obtained by
integration of Eq. (1), the locus of points associated with the CSTR
are generally found by the solution of a system of nonlinear
equations. The CSTR therefore operates on discrete points in space,
and hence there exists an opportunity for jumps and discontinuities to also occur as a result.
Interpretation of CSTR behavior in this manner allows for a
convenient geometric method for the determination of CSTR
solutions. This is useful for systems involving multiple steady
states. Critical CSTR points are those which reside on the AR
boundary.
2.3.3. Differential side-stream reactor (DSR)
A DSR may be viewed as a PFR with side-stream addition along
its length. The rate of material addition, , as well as the
concentration of the side-stream, C, are allowed to vary along
the reactor length. We will assumed here that C remains at a xed
value in this work however
dC
rC CCC
d

where 0. Geometrically, the rate of formation of species

achieved in a DSR is determined by the rate of reaction and the
mixing vector (CC). The resultant vector is then a linear combination of the two, and the net direction must therefore lie within
the space spanned by r(C) and (CC). The overall direction may be
steered between the two by appropriate manipulation of the
parameter, and consequently, its magnitude must be varied in a
manner that produces a trajectory that is able to maximize the set
of achievable concentrations. The unique policy that allows the

205

DSR trajectory to traverse along the AR boundary is termed a

critical policy, whilst the DSR corresponding to this policy is
termed a critical DSR.
Construction of the AR using structures containing DSRs is only
useful for systems involving three or more independent reactions,
and thus it is sufcient to describe the AR, involving two or one
independent reactions, with combinations of PFRs and CSTRs only.
2.4. Necessary conditions of the AR
From the above discussion, it is possible to establish a set of
necessary conditions that the AR must satisfy (Glasser and
Hildebrandt, 1997; Glasser et al., 1987); these are:

The AR must contain the feedThe feed point is attainable, and

therefore it must be contained with the set of all achievable

points.
The AR must be convexA set of achievable points that is not
convex can always be made convex by mixing. Mixing can ll in
concave regions and spaces between two separate, yet achievable, regions.
All rate vectors on the boundary of the AR must point into, or
be tangent, to the ARA PFR may be used to extend the region
otherwise.
All rate vectors (or linear combinations of rate vectors) located
in the compliment of the AR may not be extended backwards to
intersect the boundaryA CSTR may be used to achieve the
given point and therefore extend the region otherwise.

Currently, there exists no sufciency condition for the AR.

Regions that are produced that satisfy the above necessary conditions are therefore termed candidate attainable regions.
A number of AR construction methods, suitable for implementation on computer, have been devised. The interested reader
should refer to the works of Rooney et al. (2000), Burri et al.
(2002), Kauchali et al. (2002), Abraham and Feinberg (2004),
Manousiouthakis et al. (2004), Zhou and Manousiouthakis
(2006), Seodigeng et al. (2009), Ming et al. (2010) for further
details.
2.5. Dimensional considerations
2.5.1. Dimension of the AR
Although it is natural to describe the dimension of the AR in
terms of the total number of species in the system, these are
generally not independent, and hence this approach may be
unnecessary. Feinberg (2000a, 2000b) demonstrated that it is
usually possible to recast the AR in a lower dimensional subspace
in L . Here, L is the number of linearly independent reactions
taking part in the system. In addition to this, given an initial feed
concentration and system stoichiometry, the set of attainable
compositions achieved by reaction and mixing may not violate
the conservation of mass. The AR must therefore be bounded by
stoichiometric constraints. This region is termed the stoichiometric
subspace. We denote by S the set of concentrations stoichiometrically compatible with the feed.
2.5.2. Maximum number of parallel structures
There also exists an upper bound on the maximum number of
parallel structures required by the AR in order to achieve a given
state, both within the AR, and on its boundary (Feinberg, 2000a).
The results are summarized below for an AR constructed in n .

The maximum number of parallel structures required to

achieve any concentration within the interior of the region is n.

206

D. Ming et al. / Chemical Engineering Science 99 (2013) 203214

If concentrations on the boundary are to be achieved, the

number of parallel structures is, at most, equal to n1.
It follows that the maximum number of parallel structures is
equal to the number of independent reactions in the system.
Although it is generally not possible to determine, a priori, the
number or sequence of reactor units that make up a branch of an
optimal AR structure, it is still feasible to obtain an upper bound
on the maximum number of unique parallel structures required by
the system. Unless the objective function intersects the AR
boundary at multiple points, it is generally not a requirement that
all optimal reactor structures be used in the optimization. This
result shall become more meaningful in the examples that are to
follow.

3. Similarities between batch and continuous reactive

equipment
3.1. A note on continuous and batch operation
In the upcoming discussion, we will want to show how the
states realized in the three fundamental continuous reactors used
in AR theory can be achieved by batch equipment. Moreover, we
would like to demonstrate that the necessary transformations
from continuous to batch can be used in conjunction with the AR
to help guide the design of optimal batch structures. Before we
begin our discussion on the similarities between batch and
continuous reactors however, we would like to highlight a clear
difference between the two operating regimes.
Thus, even though we wish to show that it is possible to attain
steady state concentrations in a batch, it is not assumed that steady
state itself can ever be attained. By the very nature of batch
processes, this is clearly not possible. Hence, our outlook here will
not be of the view that batch reactors, specically fed-batch
reactors, can ever be operated in any particular way that is able
to reach steady state. We would, however, like to demonstrate that
it is still possible to achieve the specic concentration associated
with that steady state. Certainly, this will require a special
operating regime to achieve (which shall be detailed in the
following sections), but this will always be with the idea that
the reactor is operated under batch conditionsthat is, with a
distinct cycle time in which the state variables of the batch reactor
(volume, concentration, etc.) do, in fact, vary for the duration of
this period.
With this in mind, let us begin our discussion starting with the
simplest of cases.

running a standard batch reactor with the appropriate reaction

time.
3.3. Fed-batch reaction
Fed-batch or semi-batch reactors are reactors that allow the
feed of additional material into the reaction vessel during the
reaction period, and may be used to bias the selectivity of
components in competing reactions. The general form of the
fed-batch reactor is given as follows



dC
Ft
rC
CC
dt
Vt
rC CC
5
where F(t)/v(t). F(t) is the rate of material addition into the
reactor, and V(t) the reactor volume. It is assumed here that the
side-stream feed concentration, C, is constant in the derivation of
the above expression. Such an assumption makes no difference to
the nal result however (Feinberg, 2000a). The properties of fedbatch reactors are such that they may assume several functions
corresponding to the continuous equipment in Section 2.2. This is
more easily understood by way of an illustrative example.
Consider then Fig. 1, which represents a hypothetical scenario
for a number of fed-batch trajectories that have been plotted in
concentration space. We assume that each trajectory is obtained
by the integration of Eq. (5) for a unique policy. The feed
conditions for each are the same and given here by point O, that
is, point O represents Cf. The behavior for each policy is then given
as follows:
3.3.1. Case 1: 0
The solid curve OP is that of a solution trajectory corresponding
to a fed-batch reactor when 0. Point P is therefore the
equilibrium point. Clearly, the form of the fed-batch for this
situation reduces to that of Eq. (4) for the standard batch, and
thus PQ is also a solution for a PFR with feed composition given by
point O. The analogy between standard batch reaction and PFRs
has already been discussed above, and therefore this case will not
be considered further.
3.3.2. Case 2: (C(t))
The most familiar method of operation is one in which the
feed addition policy is adjusted throughout the reaction period.

3.2. The standard batch

By standard batch, we have in mind a batch reactor where
there is no inow or outow of material during the reaction
period. The reactor contents is well mixed, although, the concentration of species evolves with respect to time. Given an initial
concentration of species, the concentration prole within a standard batch reactor may be determined by integration of the
appropriate rate expression. Thus, we have for the standard batch
dC
rC
dt

Thus, the solution trajectory obtained by integration of Eq. (4)

is identical to that obtained by integration of Eq. (1) for the same
initial conditions. The reaction time of the batch replaces residence time of the PFR as the integration parameter in this regard.
As a result, optimal reactor structures where PFRs form the
boundary of the AR may be achieved in batch processes by

Fig. 1. Graphical interpretation for various fed-batch reactor trajectories. Each

trajectory corresponds to a different policy as follows: Solid curve OP corresponds
to a typical fed-batch trajectory for when 0 (standard batch reaction). Curve OT
is representative of a trajectory obtained by a varying policy (t). Curves RV,
RU, SV and SU represent fed-batch trajectories associated with constant alpha
policies 1 and 2. Thus RU and SU belong to 1 whilst RV and SV belong to 2. The
locus of points given by curve OVUQ is obtained by determining the equilibrium
concentration for all constant fed-batch reactors between the range 0. The
side-stream composition of all trajectories is given by point O.

D. Ming et al. / Chemical Engineering Science 99 (2013) 203214

Hence, consider now the case when is given by a varying prole.

Again, Eq. (5) may be integrated if an initial condition, side-stream
composition (given by C) and policy are supplied. The curve OT
is representative of such a solution where point T is the equilibrium point for this trajectory. Observe once more that the form of
the fed-batch equation is identical to Eq. (3) for the DSR, and
identical behavior may be achieved in the batch by appropriate
interpretation of the variable as a result.
The set of concentrations belonging to the trajectory OT is thus
equivalent to that obtained in a DSR with feed composition equal
to point O, and side-stream composition given by C (not shown in
the gure). With similar reasoning to the standard batch/PFR
analogy, residence time in the DSR is replaced by reaction time
in the fed-batch. The value of in the DSR represents the feed rate
with respect to DSR length, whereas in the batch, F/V (the ratio
of side-stream addition feed rate to reactor volume). F(t) and V(t)
are determined by integration of the policy with respect to time
and the differential expression for the volume change, usually
given by dV/dt F(t). Nevertheless, notice that there is a direct
analogue between fed-batch reactors with varying policies and
DSRs. AR boundary structures that contain DSR segments are thus
easily convertible to an equivalent fed-batch system as a result.

3.3.3. Case 3: constant

Consider now the particular case when is maintained at a
constant value throughout the reaction period. Again, if an initial
condition and fed-batch addition composition are specied, Eq. (5)
can be integrated for any arbitrary reaction time. The pair of curves
represented by RU and SU are representative solutions for when
is kept at a constant value of 1, whereas the pair RV and SV is that
which is obtained when is kept at a different value given by 2.
Clearly, RU and RV share the same initial point, given by point R,
and points SU and SV share the same initial condition given by S.
For 1, the fed-batch trajectories both terminate at point U,
even though the starting conditions differ, the same behavior is
observed for 2. All trajectories however (RU, SU, RV and SV),
are fed with material of concentration given by point O.
For a sufciently long reaction time, the concentration of
species within the vessel approaches equilibrium, and Eq. (5)
reduces to
0 rC CC

End points U and V therefore represent equilibrium points in

the fed-batch for a constant policy, and are thus also solutions to
Eq. (6). For all values of in the range 0 o o, integration of Eq.
(5) to equilibrium may be carried out with the fed-batch feed
concentration C set at point O. The locus of Xs given by the curve
OVUQ represents the set of equilibrium concentrations achieved in
the system as a result.
Clearly, equilibrium concentrations for the fed-batch are
equivalent to CSTR concentrations when the CSTR feed composition,
C0, is equal to point O. This may be more easily seen by comparing
Eq. (6) to the form of the CSTR expression given by Eq. (2). The value
of at equilibrium is thus related to the CSTR residence time by

1
At equilibirum
CSTR

It follows that all values of in the range 0 for fed-batch

reactors operating at equilibrium correspond to the set of CSTRs
with residence times between the range CSTR0.
The curve OVUQ in Fig. 1 also corresponds to solutions for a
CSTR (with feed point O), and therefore curve OVUQ is also
representative of a CSTR locus. CSTR compositions can thus be
obtained in a batch by operating a fed-batch reactor at its
equilibrium concentration. The particular equilibrium point is

207

dened by the specic constant value chosen, which corresponds

to a particular CSTR solution with residence time given by 1/.
Several observations may be noted for the constant fed-batch
reactor:
1. Initial conditions: Notice that the equilibrium point in the fedbatch is dened by the value of and the fed-batch feed
concentration, and is not a function of the initial condition at
t0. The choice of initial condition may still be somewhat
constrained, as certain specic instances prevent the designer
from choosing the starting point in a completely arbitrary
manner. Details of these are discussed under point 3. It follows
that equivalent CSTR behavior may only be achieved in the fedbatch at equilibrium, and therefore, in order to obtain CSTR
concentrations in a batch system, the reactor must be initiated
at the equilibrium conditions. In practice, this may be achieved
by several methods:

Initially run a small CSTR at the appropriate residence time
to produce the desired equilibrium composition. Then
transfer the contents of the CSTR into a fed-batch.

Begin at a condition sufciently close to the desired CSTR
concentration so that equilibrium may be approached during the reaction period.

Articially form the desired CSTR concentration by combining pure chemicals in the appropriate amounts to achieve
the correct equilibrium composition.
Irrespective of the method employed, this procedure need only
be carried out once at the beginning of the rst batch cycle.
Once product of the correct CSTR composition is produced, a
portion may be used to seed further fed-batch cycles; the
sequence is self-sustaining as a result.
2. Initial volume: At equilibrium, the concentration of species
within the vessel is maintained at a constant value of C
throughout the entire reaction period. Reactor volume therefore varies without a change in concentration in this instance.
In order to achieve a production rate comparable to that of an
equivalent CSTR, the correct initial volume of material must be
chosen. The nal batch volume must allow for an additional
small volume V0 for seeding of subsequent batches. Therefore,
the nal volume produced in the batch at t CSTR must be
equal to the equivalent CSTR volume (V0) and the initial volume
required for successive batches (VCSTR). We thus have
dV
V
dt
with boundary conditions given by V(t0)V0 and V(CSTR)
V0+VCSTR
3. Multiple steady states: Multiple steady states are often observed
in CSTRs (Gray and Scott, 1984, 1983; Schlosser and Feinberg,
1994). Indeed, even for fairly simple chemistry, there may exist
opportunities to operate at more than one steady state for a
given feed composition and residence time. To this end, fedbatch reactors may exhibit the same behavior for an equivalent
xed policy and side-stream feed composition. Thus, care
must be taken in the choice of equilibrium concentration
applied, as rather different outcomes in performance may arise,
even when the correct optimal structure belonging to the AR is
enforced.
4. Dynamic behavior of DSRs and fed-batch reactors: Observe that
the system of differential equations described by Eq. (5)
contains equilibrium points associated with solutions to Eq. (6).
These points, often termed xed points, are time-independent
solutions that provide meaningful information into the dynamical behavior of the system. The stability of xed points is
often distinguished by examination of the eigenvalues of the

208

D. Ming et al. / Chemical Engineering Science 99 (2013) 203214

with k1 1 L2/mol2/h and k2 1 h1. From the discussion of Section

2, the system contains two independent reactions and hence the
entire system can be constructed in 2 alone. The resulting rate
vector is then expressed in terms of components A and B, and
given by the following kinetics
#
"
# "
r A C
k1 C A C 2B
rC

r B C
k1 C A C 2B k2 C B
Our objective here will be towards nding the appropriate
batch sequence that maximizes the concentration of component B
in a batch cycle. This shall be achieved by rst constructing the AR
for a continuous system, and then carrying out the necessary
conversions to establish an optimal batch structure.
This system has previously been studied by Brooks (1988), who
noted two observations:

Fig. 2. Comparison between continuous and batch reactive equipment. All three
continuous reactors required to form the AR may be likened to an equivalent batch
structure.

Table 1
Summary of fed-batch operating parameters.
Continuous
structure

Initial
Policy
condition

PFR
DSR

C(0) Cf
C(0) Cf

0
(t)

CSTR

C(0)
CCSTR

Constant Cf

Side-stream
composition

Volume

Reaction
time

0
C

V constant tf PFR
V(t) F(t)/ tf DSR
(t)
V(t) F(t)/ tf free

Jacobian matrix of the system (Strogatz, 2001). These are

classied as either stable, if all real parts of the eigenvalues
are negative, or unstable, if at least one eigenvalue has a
positive real part. The stability and position of the xed points
in state space inuence the dynamical behavior of the system,
and thus also inuence the choice of initial conditionnodes
associated with unstable equilibrium points may exhibit chaotic behavior and unstable operation. Furthermore, depending
upon the feed condition and kinetics specied, multiple steady
states may arise, and thus the stability of each equilibrium
point should be considered in turn. Again, the initial condition
of the fed-batch should be considered carefully to ensure that
the desired equivalent CSTR behavior is maintained.

1. Multiple steady states are achievable depending on the feed

concentration specied and, in particular, occur when
C 0A 4 1 4 C 0B .
2. A higher concentration in B may be obtained if a period of
standard batch reaction is employed directly after the initial
fed-batch period.
The resulting maximum concentration achieved is then found
to be larger than both standard batch and fed-batch operation
alone. The above recommendations suggested correspond to those
obtained by AR analysis also.
We begin by rst computing the AR for a specied feed

T
condition. A composition of Cf 5:0 0:25
has been chosen
here so that it is consistent with the condition C 0A 4 1 4 C 0B
suggested by Brooks (1988), and as a result, the system is expected
to contain multiple steady states. Fig. 3 demonstrates this behavior
more clearly. The PFR trajectory and CSTR locus beginning at the
feed point have been plotted. The locus of CSTR points is of
particular interest, as it forms an almost completely closed loop
within the space. The residence times corresponding to these
points are easily computed from Eq. (2), and although the results
are not shown here, more than one CSTR exit concentration is
indeed achievable for a given residence time. Fig. 3 also establishes
all physically realizable concentrations achieved by mixing and

A comparison between continuous and batch equipment is

given in Fig. 2, with dening parameters required by the fed-batch
reactor summarized in Table 1. These will provide for a convenient
point of reference for the examples below.

4. Examples
4.1. 2D autocatalytic reaction
4.1.1. Identication of optimal structure
Consider the following autocatalytic system (Brooks, 1988)
k1

A 2B-3B
B-C
k2

Fig. 3. PFR trajectories () and the CSTR locus (  ) for the autocatalytic reaction

T
given by Brooks (1988). A feed vector of 5:0 0:25
has been used. The CSTR
locus produces a separate region is space that may be used to achieve higher
concentrations of component B. Point R corresponds to the point of maximum B for
a PFR whereas Q is obtained by a CSTR-PFR structure that is part of the AR
boundary.

D. Ming et al. / Chemical Engineering Science 99 (2013) 203214

209

reaction; that is, the shaded region represents the AR for the
chosen feed point. This is obtained by computing the convex hull
of all points resulting from a CSTR with feed concentration (point
O), followed by a PFR to equilibrium. The CSTR-PFR sequence is
thus the optimal reactor structure for the proposed AR. It follows
that this is the only structure required in order to achieve all other
physically realizable concentrations in the space. Of particular
interest is point P, which provides the unique CSTR concentration
that, when followed by a PFR, maximizes the concentration of
component B in the system; this occurs at point Q (CB  0.1 mol/L).
The corresponding CSTR residence time is then found to be
0.18367 h accordingly. All concentrations on the line segment OP
are thus obtained by a CSTR with bypass of a portion of the feed.
These results are in agreement with those suggested by Brooks
(1988) for the batch case, where it is observed that a higher
concentration in component B may be achieved if a standard batch
is employed directly after the fed-batch period. This is displayed in
the gure by the PFR trajectory PQ. Point R by comparison, is the
point of maximum B for a single PFR initiated at the feed
composition found at point O.

4.1.2. Conversion from continuous to batch

Once the AR and its associated continuous structure have been
formed, the task of converting to an equivalent batch system is
straightforward. The CSTR-PFR structure required for continuous
reaction is interpreted as follows:
1. A fed-batch reactor with constant policy for the CSTR portion
of the structure
2. Followed by a period of standard batch reaction for the PFR.
The distinct value for is given by the associated residence
time and equilibrium concentration of the CSTR. Hence, for a
residence time of 0.18367 h, 1/0.18367 h1. Prior to performing
the reaction however, the desired concentration at point P must
rst be prepared. This is carried out using the suggested procedures discussed in Section 3.3.3. Since concentrations within the
vessel do not vary with time during the fed-batch portion of the
cycle, reaction time is somewhat arbitrary, although, it is clear that
the rate of lling (governed by the value of ) and the nal desired
volume are important parameters when deciding on the reaction
time. We have used the familiar boundary condition when t CSTR,
V VCSTR+V0 as in Section 3.3.3.
For the standard batch component of this structure, reaction
time is determined by the residence time of the PFR. It is found
that a PFR residence time of 0.2140 h is required in order to react
from point P to point Q and therefore the same period is required
by the standard batch. Reaction time is determined directly by the
residence time of the PFR in this case.
Conversion from a continuous structure to an equivalent batch
structure is now complete. The resulting sequence of batch
operations is given by Fig. 4.
It is also possible to operate the entire sequence within a single
reaction vessel. By comparison, an equivalent approach would be
difcult to achieve in a single continuous reactor. Hence, from an
initial volume and CSTR concentration, the cycle might begin, say,
with a period of fed-batch operation and allow for the reaction to
proceed until the desired product volume has been obtained.
Batch reaction could then be initiated afterward by terminating
the side-stream feed to the reactor. The policy and concentration
prole of component B for this process is given by Fig. 5(a) and (b),
respectively. The recommended policy is fairly simple in this
instance, and involves a period of constant side-stream addition
for approximately 0.2 h, followed immediately by a longer period
with no additional side-stream feeding (the point of maximum B

Fig. 4. Comparison of continuous (a) and batch structures (b) for the autocatalytic
reaction in 2 . Only a single structure is required for this system. Notice that the
structure terminates with a PFR in the continuous case, and a standard batch period
for the batch. This is in agreement with the suggested nal batch period
recommended by Brooks(1988). (a) Recommended continuous structure. A CSTR
from the feed (point O) followed by a PFR. (b) Batch structure. The cycle begins with
a constant fed-batch period initiated at the critical CSTR efuent composition
(given by point P), followed by a standard batch period.

occurs at t 0.3977 h). The total cycle time is then CSTR+PFR.

In addition to this, the concentration prole corresponding to a
standard batch for the same feed point is also provided for
comparison. It is observed that a higher concentration is achieved
in the fed-batch when compared to the standard batch. Moreover,
the time required to achieve maximum concentration in the
standard batch is noticeably longer when compared to that of
the fed-batch-standard batch sequence (2.128 h compared to
0.3977 h); the production rate of the recommended structure is
approximately ve times faster compared to the standard batch as
a result.
The correct sequencing and timing of individual batch operations is important, and reaction time is a central parameter in this
regard. In contrast to batch equipment, continuous structures
require the adequate ordering of equipment through space. Hence
for continuous equipment, reactor volume (or residence time) is
the important parameter.

4.1.3. Stability and dynamic behavior

The appropriate choice of initial condition for the CSTR concentration is an important consideration when dealing with the
dynamic behavior of both continuous and batch operation. To see
this, consider again the CSTR locus given in Fig. 6.
As in Fig. 3, the AR and CSTR concentrations have been plotted
in CACB space. The graph also shows points on the locus
corresponding to solutions to Eq. (2) for when 0.18367 h (the
CSTR residence time associated with the critical CSTR), given here
by points P, Y and Z. Note that points O and P are the same point as
those given in Fig. 3. DSR trajectories corresponding to this
residence time for a variety of feed points have also been plotted,
allowing for the determination of the DSR map for the specied
in spacefor each value of specied, a unique DSR map may be
generated, providing information regarding the dynamic behavior
of DSRs in space. The path traced out by the CSTR locus therefore
represents the set of equilibrium points achieved by the DSR for
various values of . Points P, Y and Z are thus also the distinct

210

D. Ming et al. / Chemical Engineering Science 99 (2013) 203214

Fig. 6. DSR map for the autocatalytic system generated for an value corresponding to the critical CSTR residence time ( 1/0.18367 h1). Three equilibrium points
associated with this value of are observed (given by white triangles). Points P and
Z lie on the CSTR locus representing stable nodes (x). Equilibrium point Y lies on the
section of the locus associated with unstable nodes (unlled circles). The solid
circles are locus points associated with a stable focus. The trajectory ZUV is
obtained by a PFR initiated with a concentration from equilibrium point Z.

3. Unlled circles correspond to saddle points (real eigenvalues,

differing in sign).

Fig. 5. Recommended policy (a) and associated concentration prole of component B (b) for the autocatalytic system. (a) policy for the combined fed-batch to
standard batch system. The prole is composed of two sections, a constant value
( 1/0.18367 h1) corresponding to the critical CSTR portion of the structure, and a
period of no lling ( 0) corresponding to the nal PFR section. (b) Concentration
of component B achieved by the batch cycle obtained from the AR compared to that
achieved by a standard batch initiated at the feed concentration. The initial
constant concentration prole for the fed-batch corresponds to the CSTR concentration given by the continuous AR structure. The reaction time for the AR structure
is shorter and achieves a higher concentration in B compared to the standard batch.

equilibrium points associated with the unique fed-batch reactor

for when 1/0.18367 h1.
The CSTR locus is represented differently in this gure. Here, it
is generated by examination of the eigenvalues corresponding to
the Jacobian matrix, J, for this system
"
J

k1 C 2B

2k1 C A C B

k1 C 2B

2k1 C A C B k2

The CSTR locus may be classied according to the following

three stability criteria:
1. Sections of the locus displayed by lled circles are those which
represent a stable focus (complex eigenvalues, negative
real part).
2. Crosses indicate stable nodes (real eigenvalues both negative
in sign).

Notice that points P and Z lie on sections of the locus that are
stable, and are therefore stable equilibrium points, whereas point
Y is a saddle point. It is clear from the gure that depending on the
initial concentration chosen for a fed-batch, one may achieve
signicantly different outcomes. Although three fed-batch equilibrium solutions exist for the intended residence time, only two are
physically realizable.
If an initial concentration equal to point S is chosen, then the
fed-batch trajectory follows the curve SP and the equilibrium point
achieved is that of point P (the critical CSTR point for the AR).
Accordingly, a period of standard batch operation following from
point P results in the full set of concentrations achieved corresponding to the AR. By comparison, if an initial concentration
equal to point T is used instead, then the fed-batch trajectory
follows curve TYZ. For this initial condition, the trajectory begins
at point T, approaches unstable node Y, and then diverges sharply
towards stable node Z. Point Z is thus the equilibrium point for the
initial concentration given by point T and corresponds to an
alternate CSTR solution for the chosen residence time. Consequently, a period of standard batch operation following from point
Z results in the smaller achievable region, given by curve ZUV
instead. The set of achievable concentrations obtained by this
outcome is noticeably smaller than that given by the full AR. This
is in effect despite selecting conditions that correspond to the
correct, and optimal, CSTR-PFR structure for identical residence
times. If the feed point were to be used as the initial concentration
for this problem, then the fed-batch would have converged to the
incorrect equilibrium concentration, and the actual product
achieved by the batch sequence would differ to that suggested
by the continuous structure. Again, care must be taken when
selecting the initial condition for constant fed-batch operations.
Although it may seem convenient to use the feed concentration for
example, multiple steady states and instability might interfere
with the fed-batch equilibrium achieved, even for situations in
which the structure has been determined correctly.

D. Ming et al. / Chemical Engineering Science 99 (2013) 203214

4.2. 3D van de Vusse kinetics

4.2.1. Optimal continuous structure
Next, we shall examine the nature of the optimal batch
structure for the well-known van de Vusse reaction scheme.
Consideration of this system will demonstrate the role that DSRs,
and by extension, the role that fed-batch reactors with varying
policies play in the formation the AR boundary. The system is
composed of three independent reactions, that is,
k1

k2

A-B-C
2A-D
k3

Although four species are present, only three are independent,

as the fourth may be determined by mass balance. The rate
expression follows mass-action kinetics and may be written in
terms of components A and B by the following rate vector
2

3
k1 C A 2k3 C 2A
6
7
7
rC 6
4 k1 C A k2 C B
5
2
k3 C A
the associated rate constants are then given as k1 1 h1,
k2 1 h1, k3 10 L/mol/h.
Historically, the associated AR has been determined in CACBCD
space, so that the concentration vector may be expressed as

T
C C A C B C D . We
 specify that
T the feed is pure in component
A, in which case Cf 1 0 0 . Construction of the AR for the
above problem has been discussed in previous work, and the
optimal structure is well understood (Feinberg and Hildebrandt,
1997; Hildebrandt and Glasser, 1990). As a result, we are able to
utilize the previously developed results and apply them here with
very little added effort, so that the development of an equivalent
batch structure may be carried out.
The kinetics exhibit three independent reactions and hence, at
most, two parallel reaction schemes are required to achieve all
concentrations on the AR boundary. With the aid of geometric and
optimal control concepts, it may be shown(Feinberg, 2000a) that a
unique critical DSR policy exists that must be followed in order to

211

maintain a path on the AR boundary. The resulting critical policy

is given by Eq. (7)
"
#
1 C A 20C 3A C B 80C 2A C B C 2A 37C A C B C A 2C B
C
7
2
C B C 2A 2C A 1
The form of the policy is expressed in terms of concentration of
species rather than explicitly in terms of residence time, and is a
function of the mixing point and the feed vector. This is done as it is
more convenient to express the addition policy is terms of the state
space variables. Nevertheless, an explicit function for in terms of
residence time may be found once the optimal DSR concentration
prole is determined. By consideration of Eq. (7), together with the
rate expressions and feed point provided above, the AR may be
constructed. The results of which are given in Fig. 7. The optimal
continuous structure is composed of two parallel congurations,
denoted here by paths ABEC and DBEC, and a distinct choice must
be made between the two depending upon the desired nal state.
The rst is a DSR from the feed vector A to equilibrium point B
followed by a PFR to equilibrium point C. The second structure
forms the underside of the AR, and its corresponding structure is
slightly more subtle. Here, a CSTR-DSR-PFR structure is required.
The concentration used to feed along the DSR requires a concentration equal to that of a CSTR operating near the equilibrium
concentration (point D). Thus, although it appears in the gure
that the direction of movement occurs from point B to point D, the
true direction of the critical DSR is that from point D to point B.
The unique PFR trajectory BC that passes through point E is the
point of maximum concentration in component B. Therefore, both
structures may be used to achieve this point.
4.2.2. Conversion to batch
For this example, the AR is dened by two parallel continuous
structures. It follows that we require, at most, two batch structures
to achieve all points on the AR boundary. Fig. 8 provides the
necessary conversion from continuous to batch. Again, a one-toone correspondence between batch and continuous equipment is
observed. In both cases, fed-batch reactors with varying policies
are seen to contribute signicantly to the overall set of achievable
concentrations.

Fig. 7. AR for the van de Vusse system in 3 . The mixing points for both policies have been approximated arbitrarily closely. Mixing lines have been excluded from the
construction so that the optimal continuous structures are more clearly visible. DSR trajectories are given by solid green lines, the CSTR locus from the feed point (point A) by
blue circles, and PFR trajectories by dashed black lines.

212

D. Ming et al. / Chemical Engineering Science 99 (2013) 203214

Fig. 8. Continuous (a) and batch (b) structure comparison for the van de Vusse
system in 3 . (a) Optimal continuous structures. Structure 1 is given by a critical DSR
to equilibrium (point B) followed by a PFR to point C. Structure 2 is given by a CSTR
operating at the equilibrium concentration (point D), followed by a critical DSR to
point B, and completed with a PFR. For both structures, the DSR side-stream
composition is given by the feed point concentration. (b) Optimal batch structures.
The batch cycle for structure 1 begins with a varying fed-batch initiated at the feed
concentration and ends with a standard batch period. Structure 2 begins with a
constant fed-batch initiated at the CSTR equilibrium concentration given by point
D, followed by a varying policy corresponding to a critical DSR. This is terminated
with a standard batch period. Similar to the continuous case, all side-stream
compositions used for fed-batch operations are equal the feed point concentration.

The following batch cycles are required (Fig. 8b)):

For structure 1, a fed-batch reactor followed by a period of

standard batch reaction. This is provided by path ABEC in Fig. 7.

For structure 2, three distinct reaction intervals are needed, two

of which are fed-batch operations. The cycle is initiated with a
period of constant , followed by a non-constant interval, and
completed with a period of standard batch reaction. This is
given in Fig. 7 by the path DBEC.
Again, as these structures are analogous to the original continuous equipment, both may be used to achieve point E. Unless
one is interested in achieving concentrations along the unique PFR
trajectory BC, reduced segments belonging to structures 1 and
2 such as those of AFG and DHI can be used. For concentrations
occurring within the region however, combinations of these
structures along with mixing between them may be used.

4.2.3. Effect of different objective functions

Now that the AR and its associated batch structures have been
determined, optimization for a number of scenarios may be

Fig. 9. The AR (a) and associated batch structures and policies for two different
outcomes (b) and (c). When it is desired to limit the amount of component D to
0.3 mol/L, structure 2 given by path DHI is used. If however, we require CD 0.4 mol/L,
path AFG corresponding to structure 1 is employed instead. (a) Example of the paths
followed corresponding to two separate objective functions (CD 0.3 mol/L and
CD 0.4 mol/L). Path AFG corresponds to structure 1, whereas DHI corresponds to
structure 2. In both cases, the structures result in a traversal on the AR boundary and
not within the region. Points A and D are the same as those given in Fig. 7. (b) policy
and associated concentration prole for structure 1 of the van de Vusse problem. The
value of decreases throughout this cycle. The size of initially begins at a large value
and drops rapidly until approximately 1.5 h (corresponding to point F). The sidestream feed is then terminated resulting in a standard batch period that ends at
CD 0.4 mol/L. (c) Concentration prole and policy for structure 2 of the van de
Vusse problem. For this structure, the value of initially increases from zero to
 1.3 h1 near a reaction time of 7.8 h (corresponding to point H). Beyond this point,
the side-stream feed is again terminated resulting in a nal concentration of 0.3 mol/L
in component D.

D. Ming et al. / Chemical Engineering Science 99 (2013) 203214

investigated. Suppose that it is desired to achieve the maximum

amount of intermediate component B. By the above discussion,
both structures in Fig. 8 pass through point E and therefore both
may be used to achieve maximum B. Suppose, instead, that we
wish to limit the formation of component D to a maximum value
of 0.3 mol/L whilst still achieving maximum B. This point must lie
on the intersection of the AR boundary and the plane described by
CD 0.3 mol/L; this is obtainable by structure 2 (given here by
Fig. 9(a)). By comparison, assume that it is decided to nd the
maximum amount of B for CD 0.4 mol/L instead. We nd that
structure 1 is now required, and thus the nature of the batch cycle
is different. Both of these scenarios however, are contained within
the structure of the AR. The AR is composed of at most two unique
structures, and as a result, all concentrations on the boundary may
be achieved by these two structures alone together with mixing.
The associated policies and concentration prole of component D
for a generalized single fed-batch conguration associated with
structures 1 and 2 are given by Fig. 9(b) and (c), respectively. For
structure 1, the policy begins with an exponential decline in sidestream addition to a time of approximately 1.536 h, followed by a
period of standard batch reaction to termination. The total batch
cycle time for this structure is thus 3.5 h. Theoretically, the correct
policy for this structure starts at a value of , however this has been
approximated by utilizing a starting concentration that gives an

T
initial value of 200 h1 (C 0:9885 0:0006 0:0055 ). The
recommended operating policy corresponding to structure 2 is
slightly different when compared to that of structure 1, as it
undergoes three reaction periods. The cycle begins with a fedbatch period of constant in accordance with the equilibrium CSTR
concentration given by point D. The cycle is then brought into a
period of varying , where the side-stream addition is observed to
increase sharply to a maximum value of  1.356 h1, and then
completed with a standard batch period lasting approximately
9.24 h; the total batch cycle time for this structure is thus roughly
15 h.
An observation on parallel structures should be made here with
regards to the differences in operating requirements between
batch and continuous equipment. If the reaction is to be performed in a continuous setting, two separate sets of equipment are
required and must be run simultaneously. Compositions are then
obtained by mixing the appropriate streams belonging to each
structure in the appropriate amounts. By comparison, if it is
desired to perform the same task under batch conditions, an
identical outcome may be achieved by two serial operations
together with intermediate storage of material used in mixing
processes. Hence, it may turn out that a single set of reactive
equipment be required. Again, whereas in continuous systems one
is concerned with the arrangement of equipment through space,
in batch systems, one may also utilize time. The order of batch
operations is important in developing an efcient reactive
structure.

5. Conclusions
At the outset, an important goal of this paper has been towards
developing a systematic method of converting optimal continuous
structures, obtained via conventional AR analysis, to an equivalent
batch structure. Due to the one-to-one nature between continuous
and batch reactive equipment, it is a fairly simple task to convert
continuous structures to batch. Indeed, it is possible to derive an
equivalent batch structure by rst computing the AR in the
continuous setting, and then applying the necessary transformations to batch.
Previous work (Feinberg, 2000a) has demonstrated that three
unique reactor types are needed to construct the AR, these are the

213

PFR, DSR and CSTR. The transformations to batch for these types
are as follows:

Standard batch reactors replace PFRs. It is already well under-

stood that standard batch reactors have identical performance

to PFRs, and so they may be used in place of them in batch
processes. Reaction time then replaces residence time of
the PFR.
Fed-batch reactors may be operated to function as both a DSR
or CSTR depending on the policy chosen. The variable in this
case, is the ratio of F/V, and if an expression for the policy
exists, then both the side-stream feed rate and reactor volume
may be determined. Thus, if the fed-batch is operated with a
varying policy, DSR performance is obtained. If, instead, a
constant policy is employed, then CSTR behavior is possible,
although, only under the correct operating conditions.

For each value of , there exists an associated equilibrium

concentration that must be used. Possibly many equilibrium points
may be obtained for kinetics involving multiple steady states. This
is related to the equivalence found between CSTRs and constant
DSRs operating at equilibrium. Constant fed-batch reactors thus
mimic CSTRs at the equilibrium point. This results in interesting
operational consequences:
1. Fed-batch reactors must be initiated and operated at a concentration equal to that of the equivalent CSTR efuent composition. The concentration of species within the fed-batch is
constant during this period of reaction as a result. Reaction
time is also determined by the lling rate and vessel volume.
2. Various methods may be used to achieve the initial CSTR
composition, although these methods may be ineffective in
obtaining the desired CSTR composition if the corresponding
DSR equilibrium point is unstable.
Thus, care must be taken when choosing an initial concentration for the fed-batch when converting structures containing
CSTR's to an equivalent batch. It is generally not sufcient to
choose an initial concentration in an arbitrary manner in the hope
to reach a desired CSTR concentration. The dynamic behavior of
the system, in the form of multiple steady states and instability,
may not allow one to do so. Furthermore, it might not be possible
to achieve the desired CSTR concentration if the associated
equilibrium point corresponding to Eq. (6) proves to be unstable.
The authors have yet to encounter systems in which the desired
equilibrium point belonging to a critical CSTR is also unstable,
however the possibility should not be overlooked. Such a situation
could prove to be troublesome for both continuous and batch
structures, as both the CSTR and fed-batch may not be controllable. One would then need to work within a constrained region of
the full AR utilizing points that are known to be stable.
Finally, although many similarities between batch and continuous equipment exist, a number of differences are also noted.
It may be possible to achieve all concentrations belonging to the
AR by a single reaction vessel in batch systems, whereas for
continuous equipment, a physically separate set of reactors is
required for each branch of the recommended AR structure. By the
same reasoning, one should also be mindful of the order that batch
operations are carried out, as different batch operating policies
may be required for different points desired in or on the AR
boundary.
The above discussion suggests that improvement in batch
reactors is possible by a geometric approach. Interpretation of
the problem rst as a continuous system allows for the determination of the AR, which provides insight into the limits of
achievability for all reactor structures. This, in turn, guides the

214

D. Ming et al. / Chemical Engineering Science 99 (2013) 203214

choice of reactor structure employed. Consideration of the AR thus

offers a different method of potential improvement in batch
reactors. The AR and its corresponding structures therefore serve
the dual purpose of establishing a set of recommended reactor
congurations that may be used to improve both batch and
continuous equipment from a single construction.
References
Abraham, T.K., Feinberg, M., 2004. Kinetic bounds on attainability in the reactor
synthesis problem. Ind. Eng. Chem. Res. 43, 449457.
Allgor, R.J., Barrera, M.D., Barton, P.I., Evans, L.B., 1996. Optimal batch process
development. Comput. Chem. Eng. 20, 885896.
Allgor, R.J., Barton, P.I., 1999. Screening models for batch process development: Part
II. Case studies. Chem. Eng. Sci. 54, 40654087.
Allgor, R.J., Evans, L.B., Barton, P.I., 1999. Screening models for batch process
development: Part I. Design targets for reaction/distillation networks. Chem.
Eng. Sci. 54, 41454164.
Aziz, N., Mujtaba, I.M., 2002. Optimal operation policies in batch reactors. Chem.
Eng. J. 85, 313325.
Berg, M., de, Cheong, O., Kreveld, M., van, Overmars, M, 2008. Computational
Geometry: Algorithms and Applications. Springer, pp. 243244.
Bonvin, D., 1998. Optimal operation of batch reactorsa personal view. J. Process
Control 8, 355368.
Brooks, B.W., 1988. Semi-batch reactors and their use for autocatalytic reactions.
Chem. Eng. Sci. 43, 12871290.
Burri, J.F., Wilson, S.D., Manousiouthakis, V.I., 2002. Innite DimEnsionAl Statespace approach to reactor network synthesis: application to attainable region
construction. Comput. Chem. Eng. 26, 849862.
Capon-Garcia, E., Bojarski, A.D., Espuna, A., Puigjaner, L., 2011. Multiobjective
optimization of multiproduct batch plants scheduling under environmental
and economic concerns. AIChE J. 57, 27662782.
Cheong, D.-Y., Hansen, C.L., Stevens, D.K., 2007. Production of bio-hydrogen by
mesophilic anaerobic fermentation in an acid-phase sequencing batch reactor.
Biotechnol. Bioeng. 96, 421432.
Davies, W.L., Gloor, W.T., 1971. Batch production of pharmaceutical granulations in
a uidized bed I: Effects of process variables on physical properties of nal
granulation. J. Pharm. Sci. 60, 18691874.
Davis, B.J., Taylor, L.A., Manousiouthakis, V.I., 2008. Identication of the attainable
region for batch reactor networks. Ind. Eng. Chem. Res. 47, 33883400.
Feinberg, M., 1999. Recent results in optimal reactor synthesis via attainable region
theory. Chem. Eng. Sci. 54, 25352543.
Feinberg, M., 2000a. Optimal reactor design from a geometric viewpoint. Part II.
Critical sidestream reactors. Chem. Eng. Sci. 55, 24552479.
Feinberg, M., 2000b. Optimal reactor design from a geometric viewpointIII.
Critical CFSTRs. Chem. Eng. Sci. 55, 35533565.
Feinberg, M., Hildebrandt, D., 1997. Optimal reactor design from a geometric
viewpointI. Universal properties of the attainable region. Chem. Eng. Sci.
52, 16371665.
Glasser, D., Hildebrandt, D., 1997. Reactor and process synthesis. Comput. Chem.
Eng. 21, S775S783.
Glasser, D., Hildebrandt, D., Crowe, C., 1987. A geometric approach to steady ow
reactors: the attainable region and optimization in concentration space. Ind.
Eng. Chem. Res. 26, 18031810.
Glasser, D., Hildebrandt, D., Godorr, S., 1994. The attainable region for segregated,
maximum mixed, and other reactor models. Ind. Eng. Chem. Res. 33, 11361144.
Godorr, S., Hildebrandt, D., Glasser, D., McGregor, C., 1999. Choosing optimal
control policies using the attainable region approach. Ind. Eng. Chem. Res. 38,
639651.
Godorr, S.A., Hildebrandt, D., Glasser, D., 1994. The attainable region for systems
with mixing and multiple-rate processes: nding optimal reactor structures.
Chem. Eng. J. Biochemical. Eng. J. 54, 175186.

Gray, P., Scott, S.K., 1983. Autocatalytic reactions in the isothermal, continuous
stirred tank reactor: Isolas and other forms of multistability. Chem. Eng. Sci. 38,
2943.
Gray, P., Scott, S.K., 1984. Autocatalytic reactions in the isothermal, continuous
stirred tank reactor: oscillations and instabilities in the system A+2B-3B; BC. Chem. Eng. Sci. 39, 10871097.
Hildebrandt, D., Glasser, D., 1990. The attainable region and optimal reactor
structures. Chem. Eng. Sci. 45, 21612168.
Hildebrandt, D., Glasser, D., Crowe, C.M., 1990. Geometry of the attainable region
generated by reaction and mixing: with and without constraints. Ind. Eng.
Chem. Res. 29, 4958.
Kauchali, S., Rooney, W.C., Biegler, L.T., Glasser, D., Hildebrandt, D., 2002. Linear
programming formulations for attainable region analysis. Chem. Eng. Sci. 57,
20152028.
Khumalo, N., Glasser, D., Hildebrandt, D., Hausberger, B., Kauchali, S., 2006. The
application of the attainable region analysis to comminution. Chem. Eng. Sci.
61, 59695980.
Manousiouthakis, V.I., Justanieah, A.M., Taylor, L.A., 2004. The Shrink-Wrap algorithm for the construction of the attainable region: an application of the IDEAS
framework. Comput. Chem. Eng. 28, 15631575.
Mendez, C.A., Cerda, J., Grossmann, I.E., Harjunkoski, I., Fahl, M., 2006. State-of-theart review of optimization methods for short-term scheduling of batch
processes. Comput. Chem. Eng. 30, 913946.
Ming, D., Hildebrandt, D., Glasser, D., 2010. A revised method of attainable region
construction utilizing rotated bounding hyperplanes. Ind. Eng. Chem. Res. 49,
1054910557.
Modak, J.M., Lim, H.C., 1992. Optimal mode of operation of bioreactor for
fermentation processes. Chem. Eng. Sci. 47, 38693884.
Najafpour, G., Younesi, H., Ismail, K.S.K., 2004. Ethanol fermentation in an
immobilized cell reactor using Saccharomyces cerevisiae. Bioresour. Technol.
92, 251260.
Nicol, W., Hernier, M., Hildebrant, D., Glasser, D., 2001. The attainable region and
process synthesis: reaction systems with external cooling and heating. The
effect of relative cost of reactor volume to heat exchange area on the optimal
process layout. Chem. Eng. Sci. 56, 173191.
Nicol, W., Hildebrandt, D., Glasser, D., 1997. Process synthesis for reaction systems
with cooling via nding the attainable region. Comput. Chem. Eng. 21, S35S40.
Nisoli, A., Malone, M.F., Doherty, M.F., 1997. Attainable regions for reaction with
separation. AICHE J. 43, 374386.
Rooney, W.C., Hausberger, B.P., Biegler, L.T., Glasser, D., 2000. Convex attainable
region projections for reactor network synthesis. Comput. Chem. Eng. 24,
225229.
Schlosser, P.M., Feinberg, M., 1994. A theory of multiple steady states in isothermal
homogeneous CFSTRs with many reactions. Chem. Eng. Sci. 49, 17491767.
Senthuran, A., Senthuran, V., Mattiasson, B., Kaul, R., 1997. Lactic acid fermentation
in a recycle batch reactor using immobilized Lactobacillus casei. Biotechnol.
Bioeng. 55, 841853.
Seodigeng, T., Hausberger, B., Hildebrandt, D., Glasser, D., 2009. Recursive constant
control policy algorithm for attainable regions analysis. Comput. Chem. Eng. 33,
309320.
Strogatz, S.H., 2001. Nonlinear Dynamics and Chaos: With Applications to Physics,
Biology, Chemistry, and Engineering. Westview Press.
Woolard, C.R., Irvine, R.L., 1995. Treatment of hypersaline wastewater in the
sequencing batch reactor. Water Res. 29, 11591168.
Zeman, R.J., Amundson, N.R., 1965. Continuous polymerization modelsPart II:
Batch reactor polymerization. Chem. Eng. Sci. 20, 637664.
Zhou, W., Manousiouthakis, V.I., 2006. Non-ideal reactor network synthesis
through IDEAS: attainable region construction. Chem. Eng. Sci. 61, 69366945.
Zwiener, C.Z., Seeger, S.S., Glauner, T.G., Frimmel, F.F., 2002. Metabolites from the
biodegradation of pharmaceutical residues of ibuprofen in biolm reactors and
batch experiments. Anal. Bioanal. Chem. 372, 569575.