Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/256023197
READS
343
2 authors:
Anand V. P. Gurumoorthy
Jason R. Picardo
VIT University
12 PUBLICATIONS 88 CITATIONS
16 PUBLICATIONS 33 CITATIONS
SEE PROFILE
SEE PROFILE
Abstract
As the study of nonlinear system dynamics and bifurcation theory percolate into applied
engineering and biological disciplines, the need for a user friendly, flexible and robust
bifurcation analysis software is evident. MATCONT is a MATLAB toolbox with a graphical
user interface for the continuation of static and dynamic equilibria of nonlinear systems. This
software is relatively recent and has certain advantages over existing continuation software
which include its MATLAB foundation, in addition to new features. In this paper, this
toolbox is evaluated via a well studied illustrative example of an ethanol fermentor which
shows rich nonlinear behavior. Some computed results are displayed and a discussion of the
basic workflow in MATCONT is provided. A few difficulties encountered while studying
nonlinear systems with MATCONT are highlighted. Nevertheless, due to its merits and future
scope, as is discussed, this software is recommended for bifurcation analysis of nonlinear
systems in research as well as in the classroom.
Keywords
MATCONT, bifurcation diagrams, nonlinear systems.
1. Introduction
The advent of chaos and the science of nonlinear dynamic systems has caused the scientific
community to rethink the way it understands nature and its patterns as well as the manner in
which engineering systems are modelled, designed, operated and controlled. In chemical and
process systems engineering, much progress has been made in understanding the complex
nonlinear behavior (multiplicity, oscillations and chaos) of a variety of systems ranging from
the ideal lumped model CSTR to the distributed model packed bed reactor; from the type IV
FCCU units to more complex configurations involving autothermal and reverse flow reactors.
Distillation columns and absorbers have also been studied. Extensive reviews on this subject
are available (Elnashaie and Grace, 2007, Elnashaie et al., 2006, Elnashaie and Elshishini,
1996, Marquardt and Mnnigmann, 2005 and Lee and Chang, 1996).
Industrial interest in these phenomena remains at a low. This is partially due to the
philosophy of avoiding the troublesome operating regions altogether (Elnashaie and Grace,
2007). With the great advancements made in dynamic modelling, computational power and
digital control, the time is ripe to explore the possible advantages of higher yield and
efficiency which may be achieved in these non-conventional dynamic operating regimes. The
implications of these studies on the design, operation and control of chemical/biological
reactors cannot be ignored in a world where the cost of energy and time is on the rise. Among
the life sciences, it is becoming increasingly obvious that dynamic systems theory has a lot to
contribute to our understanding of life and nature. The inherent non-monotonic dependencies
of biological factors on environmental conditions along with the synergetic coupling between
reaction and diffusion phenomena lead to complex dynamics in many biological systems
(Murray, 2002 and Murray, 2003).
In order to facilitate the application of the much advanced mathematical theory of bifurcation
and nonlinear dynamics by practically oriented engineers and biologists, a robust, flexible
and user friendly software tool is necessary. Such software would require the integration of
advanced numerical algorithms for solution of differential and difference equations,
bifurcation routines and algorithms for computing various dynamical parameters of interest
(eigenvalues, periods of oscillations, floquet multipliers etc.) as well as an advanced,
interactive visualization package. It could have a similar impact to that of the now hugely
popular computational fluid dynamics (CFD) packages (ANSYS, COMSOL etc.) which have
allowed engineers studying fluid systems to focus on the engineering problem at hand and
experiment with various design and operational possibilities on a computer. This has made
the optimization of processes so much easier and brought the complex mathematics behind
CFD within the reach of non-mathematical workers.
2. Software for Bifurcation Analysis of Nonlinear Systems
Over the past decade and a half, various software packages have emerged with the objective
of aiding investigators in the exploration of dynamical systems via bifurcation analysis. The
initial versions were simple codes made available under the names of AUTO86, LINLBF,
BIFOR2, PATH and LOCA. The next generation saw interactive programs like AUTO97
(Doedel et al., 1997) XPPAUT and LOCBIF. AUTO97 was a very popular tool used by many
workers. Its latest version is AUTO07. XPPAUT is a dynamical systems analysis tool which
interfaces with AUTO. It is popular amongst mathematical biologists (Ermentrout, 2002). For
a windows version, XPP adopted LOCBIF, another continuation toolbox. LOCBIF eventually
evolved into CONTENT (Kuznetsov and Levitin, 1997). The focus of this paper is a
relatively new software package, which has been actively supported and continually
advanced since its introduction (2003), called MATCONT (Dhooge et al., 2003).
MATCONT is an improved, MATLAB version of CONTENT. It is a MATLAB toolbox with
a graphical user interface or GUI (without the GUI it is called MATCONT_CL). MATLAB
is a numerical computing environment which is now widely used in academia and industry
and is popular for its strong matrix based computational capabilities and visualization tools.
MATCONT with its GUI provides a user friendly experience with the added advantage of
generating visualizations directly without using any other software. Further, it can take full
advantage of the advanced capabilities of MATLAB including its robust integrators and
symbolic computation toolbox. A comparison of the features of MATCONT with AUTO and
CONTENT as of 2005 is presented in Table 1 (Kuznetsov, 2005). While most of these
features are not discussed here, the comparison is useful in gauging the merits of MATCONT
vis. a vis. other popular software. It is the goal of this paper to put this promising software to
the test and evaluate its performance as a tool for the non-mathematical worker who is
interested in exploring a dynamical system.
3. Isothermal Stirred Tank Ethanol Fermentor- an Illustrative Example
To evaluate the capabilities of MATCONT, an ethanol fermentor system which exhibits
multiplicity of steady states, oscillation, period doubling and chaos is adopted as an
illustrative example. In this isothermal fermentor system, the ethanol (P) is produced on
fermentation of a sugar based substrate (S) by the microorganism Zymomonas mobilis. The
dynamics of this system were studied in detail by Garhyan et al. (2003). Garhyan and
Elnashaie (2004) went on to demonstrate the new system dynamics when a selective
membrane to separate ethanol was included in the process. Ethanol acts as an inhibitor to the
microorganism (Z. mobilis). The incorporation of the membrane was shown to stabilize the
chaotic regime of operation allowing the achievement of higher yields. They also
experimentally verified the model and the predictions of bifurcation theory (Garhyan and
Elnashaie, 2005). They patented their work in 2005 (US 2005/0170483 A1), as an invention
entitled- Chaotic Fermentation of Ethanol. This system has been selected since it is well
studied and provides an example of the practical benefits which result from the application of
bifurcation theory to industrial systems.
Features
Time integration
Poincar maps
Continuation of equilibria
Detection of branch points and codimension 1 bifurcations of equilibria
(hopf and limit points)
Computation of normal forms for codimension 1 bifurcations of equilibria
Continuation of codimension 1 bifurcations of equilibria
Detection of codimension 2 equilibrium bifurcations
Continuation of limit cycles
Detection of branch points and codimension 1 bifurcations of cycles
Continuation of codimension 1 bifurcations of cycles
Branch switching at equilibrium and cycle bifurcations
Continuation of branching points of equilibria and cycles
Computation of normal forms for codimension 1 bifurcations of cycles
Detection of codimension 2 bifurcations of cycles
Continuation of orbits homoclinic to equilibria
A C M
Table 1: Comparison between AUTO, CONTENT and MATCONT. Symbols: A- AUTO, CCONTENT, M- MATCONT
The system model, without the selective membrane, consists of 4 nonlinear ordinary
differential equations. The model was developed by Jobses and coworkers who also verified
the same experimentally (Jobses et al., 1985, Jobses et al., 1986, Jobses, 1986). An
unsegregated structured two compartment representation was adopted in their work in which
the biomass is described in terms of two components- a key cellular component (e) which is
essential for growth and ethanol formation and the non-active component (X). The synthesis
of (e) has a nonlinear dependence on ethanol concentration. Thus, the inhibition of ethanol is
modelled as an indirect effect on specific growth rate via component (e). The four model
equations are given below and parameter values used in the simulation are given in Table 2
(Garhyan et al., 2003). The state variables are the concentrations of the active component (e),
the inactive biomass (X), the substrate (S) and ethanol (P).
dCe
2
k1 k 2 C P k 3C P
dt
KC CCs DC
s
e0
DCe
C C
dC X
p S e DC X 0 DC X
dt
K S CS
1 C S Ce
dC S
mS C X DCS 0 DCS
p
dt
Y
K
C
S
SX S
(1)
(2)
(3)
1
dC P
p
dt
YPX
C S Ce
K S C S
mP C X DCP 0 DCP
(4)
The dilution rate (D) and substrate concentration in the feed (Cso) were taken as bifurcation
parameters since they are most easily manipulated during design and manufacturing. A
detailed study involving several bifurcation diagrams was presented by Garhyan et al. (2003).
These results are reproduced using MATCONT. Three of these diagrams are presented below
to aid in the demonstration and discussion of MATCONTs capabilities.
Table 2: Parameters used in simulation
parameter
valuea
k1
regression coefficient (h-1)
16.0
k2
regression coefficient (m3/kgh)
4.9710-1
k3
regression coefficient (m6/kg2h)
3.8310-3
mS
maintenance coefficient (substrate utilization) (kg/kgh) 2.16
mP
maintenance coefficient (product formation) (kg/kgh)
1.1
YSX
yield coefficient of biomass from substrate (kg/kg)
2.4449810-2
YPX
yield coefficient of product from biomass (kg/kg)
5.2631510-2
KS
saturation constant (kg/m3)
0.5
-1
p
maximum growth rate (h )
1
D
dilution rate (h-1)
4.0
V
fermentor volume (m3)
0.003
3
CX0,P0,e0
concentrations in feed (kg/m )
0,0,0
Fig. 1. Bifurcation diagram with D as the bifurcation parameter and CS0 at 150.3 kg/m3. Hhopf point, LP- limit point and PD- period doubling. The dotted line denotes unstable
states while the solid line denotes stable states
Fig. 2. Bifurcation diagram with CS0 as the bifurcation parameter and D at 0.045 h-1.
Nomenclature same as Fig. 1
The major steps involved in plotting these diagrams in MATCONT are described below:
1. Firstly, the system of equations is defined in an interactive window, in which the state
variables and parameters are to be specified and the equations typed out in simple
MATLAB syntax. The MATLAB symbolic toolbox can be used to compute analytic
expressions of derivatives necessary for computation. This significantly improves
computation time especially when high accuracy is desired.
2. In order to begin the continuation of static equilibria, an initial equilibrium point
(steady state) is required. This can be obtained by time integration at the desired
parameter values. This task is made easy by MATCONT which allows the user to
integrate the equations in time using any of MATLABs ordinary differential equation
integrators. This operation can be carried out in the same GUI framework which
eliminates the necessity of writing a separate m-file (MATLAB code file) for the task.
Moreover, once a steady state is reached, the value of the last point can be
interactively selected and the continuation of static equilibria initiated. Garhyan et al.
(2003) have employed a separate set of codes using IMSL FORTRAN routines for
time integration.
Fig. 3. Two parameter bifurcation diagram with CS0 and D as bifurcation parameters.
5. Critical Comments
In this section the difficulties encountered while studying various nonlinear systems, using
MATCONT, are discussed.
1. In the analysis of an electrochemical oscillator (Koper and Sluyters, 1991) the model
definition required a conditional statement (if-else). This statement cannot be included
in the GUI system definition window. Hence, it was necessary to make suitable
modifications to the MATLAB m-file which defines the system equations after it was
generated by MATCONT.
2. A novel continuously stirred decanting reactor (CSDR) studied by Khinast et al
(1998) shows rich dynamic behavior. The model consists of 13 differential algebraic
equations (DAEs) with a state dependent mass matrix. Unfortunately, MATCONT
does not have the capability to work with DAE models and could not be used to study
this system.
3. MATCONT does not have option to modify plotting features such as color, number of
points plotted, markers etc. in the GUI. Thus, as mentioned previously, changes must
be made after the figure is plotted which can be tedious. Further, the limit cycles are
represented by a number of points along the cycle which is in contrast to the standard
practice of using markers for the extreme values of oscillation of the dependent
variable. This makes visualization difficult; moreover the number of markers on a
limit cycle cannot be modified even after generating the plot.
6. Conclusion
In this work, the utility of MATCONT as a tool for studying multiplicity and dynamic
oscillations of nonlinear systems has been demonstrated. It has proved to be user friendly
especially for users of MATLAB. Its strengths include convenient access to powerful
visualization tools and integrators and a large variety of features which encompass all the
capabilities of present bifurcation software, with some additions (Table 1). MATCONT has
several advanced features, not discussed in this paper, which are helpful in studying complex
dynamical behavior including quasi-periodicity and chaos. Apart from systems of differential
equations it can also be used to study discrete dynamical maps (systems of iterative
difference equations). The possibility of interfacing MATCONT with other MATLAB
toolboxes holds promise for the future. For e.g., a module which interfaces between
MATCONT and the MATLAB optimization toolbox would make an excellent tool for
optimizing systems while considering changes in output due to bifurcations.
The MATCONT team continues to work on adding new features and fixing bugs. This
software
is
freely
available
for
download
at
Source
Forge
(http://sourceforge.net/projects/matcont) and can be installed and run easily by a MATLAB
user. It is possible to use MATCONT to investigate nonlinear systems with a minimum of
mathematical knowledge, thus allowing practically oriented engineers and biologists to
analyze nonlinear systems with just an intuitive feel for the related mathematics and
knowledge of essential rules and theorems. Such a tool would be particularly useful in
applied engineering studies as well as in classroom courses on nonlinear dynamics and
bifurcation theory.
References
Dhooge A, Govaerts W and Kuznetsov Yu. A (2003), Matcont: A matlab package for
numerical bifurcation analysis of ODEs, ACM TOMS., Vol. 29, No. 2, pp. 141-164.
Lee J S and Chang K S (1996), Applications of chaos and fractals in process systems
engineering, J. Proc. Cont., Vol. 6, pp. 71-87.