University of Wollongong

Research Online
University of Wollongong Thesis Collection

University of Wollongong Thesis Collections

2013

Modelling of biological wastewater treatment

Rubayyi Turki Alqahtani
University of Wollongong

Recommended Citation
Alqahtani, Rubayyi Turki, Modelling of biological wastewater treatment, Doctor of Philosophy thesis, School of Mathematics and
Applied Statistics, University of Wollongong, 2013. http://ro.uow.edu.au/theses/3951

Research Online is the open access institutional repository for the

University of Wollongong. For further information contact the UOW
Library: research-pubs@uow.edu.au

Modelling of Biological
Wastewater Treatment

A thesis submitted in fullment of the

requirements for the award of the degree

Doctor of Philosophy
from

University of Wollongong

by

Rubayyi Turki Alqahtani

B.Sc., M.Sc. Mathematics

School of Mathematics and Applied Statistics

July, 2013

Dedicated to

My parents, wife and children

Certication
I, Rubayyi Turki Alqahtani, declare that this thesis, submitted in fullment of requirements for the award of Doctor of Philosophy, in the School of Mathematics
and Applied Statistics, University of Wollongong, is wholly my own work unless
otherwise referenced or acknowledged. The document has not been submitted for
qualications at any other academic institution.

Rubayyi Turki Alqahtani

iii

Acknowledgements
In the Name of Allah, the Most Gracious, the Most Merciful.
First of all, I thank God for helping me and sending people who have been inuence
in my life.
I express my deep sense of gratitude to my supervisor Associate Professor Mark
Neslon, for his eort and guidance in helping me throughout my research. As a supervisor, he gave me the opportunity to explore challenging mathematical research
problems and he has been a constant source of guidance.
I am deeply thankful to my co-supervisor, Associate Professor Annette Worthy for
her guidance and encouragement through my Journey.
I would like to express my deepest thanks and gratitude to my father Turki (Oh
God, forgive him and Rahma and insert it into the Prog bliss God), mother Sara,
my wife; Fatimah and my daughters; Sara, Yara and Dana as the gift of unbounded
love and support has no equal.
I would also like to extend my thanks to committee in school of mathematics and
applied statistics in UOW in particular Prof.Timothy Marchant, Dean of research
and Prof.Jacqui Ramagge, the head of the school.
I would also like to extend my thanks to my brothers and my sisters.
iv

Abstract
Many models have been proposed in the literature for the specic growth rate of
biomass, including those of Tessier, Monod, Moser and Contois kinetics. Of these
models the Monod model has become the default model used in bioreactor engineering. However, there is growing experimental evidence showing that in a number of
industrial processes, in particular wastewater treatment, the specic growth rate is
more accurately described by Contois kinetics rather than other models. Thus in
this thesis, we construct and analyze models for wastewater treatment where the
specic growth rate of biomass on biodegradable organic matter is assumed to be
given by the Contois growth rate.
This thesis is organized into four parts. Part 1 consists of three chapters (1, 2 and 3)
which contain general information on the wastewaster treatment and the pertinent
modeling studies.
For remaining parts, specic issues related to wastewater treatment modelling are
investigated. In part 2, which contain three chapters (4, 5 and 6), we develop a
model for wastewater treatment. In chapter (4), the analysis of wastewater treatment in a single reactor with recycle is presented. In chapter (5), we consider a
n-cascade reactor with recycle around each reactor. We investigate how to tune
vi

vii
the parameters of a settling unit to minimize the euent concentration leaving the
reactor cascade. If only one settling unit is to be used, we ask the question, Where
should it be placed in order to optimized the performance of the reactor cascade?
The chapter (6) deals with the scenario in which the settling unit is placed after the
nal cascade reactor and the euent stream from the settling unit is recycled back
into the rst reactor. We show that there is a critical value of the total residence
time. If the total residence time is below the critical value then the settling unit
improves the performance of the reactor cascade whereas if the residence time is
above the critical value the performance of the cascade is reduced compared to that
of a cascade without a settling unit. We conclude by noting that the conguration of
the cascade where recycle occurs around each reactor (chapter (5)), outperforms the
conguration of the cascade where recycle is present around the whole cascade at
high total residence time. This is noteworthy as the latter is often used in industry.
A further three chapters (7, 8 and 9) are contained in part 3. In this part, we
extend the standard Contois expression to include both substrate inhibition and a
variable yield coecient. These extensions are important in both the theoretical
and practical application of wastewater treatment. We investigate parameter regions in which either natural oscillations or bistable behaviour can occur. In this
rst of these chapters, we consider the case when the yield coecient is variable
with decay coecient whilst in the second of these chapter, the substrate inhibition
is analyzed. In the nal chapter of this part, the combination of substrate inhibition
and a variable yield coecient is investigated.
The nal part of this thesis investigates the use of a sludge disintegration unit to

viii
minimize the sludge production inside the reactor, represented by chapters (10 and
11). This is important to reduce the operation cost associated with the wastewater process. In chapter (10), we investigate Yoons model with innite reaction
rate by replacing Monod kinetic with Contois Kinetic. The operation condition for
two congurations of reactor, continuous ow reactor and membrane reactor under
which a sludge disintegration unit is required are investigated. In chapter (11), we
extend Yoons model to include nite reaction rates and Contois kinetics and establish that the innite reaction rate assumption of Yoons model is only correct for
specic value of the sludge solubilization eciency but a viable approximation in
only certain practical cases. The eect of the nite reaction rate and the relative
volume of the sludge disintegration unit (n) upon the production of activate sludge
inside the reactor are investigated. Finally, chapter (12) will summarize the thesis
ndings and give future research directions.

List of Publications
The following publications have been submitted by the author during his study.

Journal article
[1] R.T. Alqahtani , M.I. Nelson and A. Worthy. A fundamental analysis of
continuous ow bioreactor models governed by Contois kinetics. IV. Recycle around
the whole reactor cascade. Chemical Engineering Journal, 218, 99-107, 2013.
[2] R.T. Alqahtani. M.I. Nelson and A.L. Worthy. A fundamental analysis
of continuous ow bioreactor models with recycle around each reactor governed
by Contois kinetics. III. Two and three reactor cascades. Chemical Engineering
Journal, 183, 422-432, 2012.
[3] R.T. Alqahtani p, M.I. Nelson and A.L. Worthy. Analysis of a chemostat
model with variable yield coecient: Contois kinetics. In M. Nelson, M. Coupland,
H. Sidhu, T. Hamilton and A.J. Roberts, editors, Proceedings of the 10th Biennial
Engineering Mathematics and Applications Conference, EMAC 2011, ANZIAM J,
53, pages C155-C171. 2012.
[4] R.T. Alqahtani. M.I. Nelson and A.L. Worthy. Analysis of a chemostat model
with variable yield coecient and substrate inhibition: Contois growth kinetics.
chemical engineering communications, accepted.
ix

Conference paper
[1] R.T. Alqahtani , M.I. Nelson and A.L. Worthy. A mathematical analysis
of continuous ow bioreactor models governed by contois kinetics: A two reactor cascade. In Proceedings of the Australasian Chemical Engineering Conference,
CHEMECA 2011, pages 111. Engineers Australia, 2011. On CDROM. ISBN 978
085 825 9225.
[2] R.T. Alqahtani , M.I. Nelson and A.L. Worthy. A mathematical model
for the biological treatment of industrial wastewaters in a cascade of four reactors. In F. Chan, D. Marinova, and R.S. Anderssen, editors, 19th International
Congress on Modelling and Simulation, MODSIM 2011, pages 256-262. Modelling
and Simulation Society of Australia and New Zealand, 2011. On CDROM. ISBN
978-0-9872143-1-7.

Table of Contents
Certication

iii

Acknowledgements

iv

Abstract

vi

List of Publications

ix

Table of Contents

xi

1 Introduction

2 Literature review
2.1 Biological Wastewater Treatment . . . . . .
2.2 Anaerobic and aerobic wastewater treatment
2.3 Microbial growth . . . . . . . . . . . . . . .
2.4 Substrate inhibition . . . . . . . . . . . . . .
2.5 Models with a variable yield coecient . . .
2.6 Endogenous processes . . . . . . . . . . . . .
2.7 Cascade reactors . . . . . . . . . . . . . . .

.
.
.
.
.
.
.

3 Literature review of Contois growth kinetics

3.1 Contois growth kinetics . . . . . . . . . . . . .
3.2 Contois growth kinetic with inhibition . . . .
3.3 Interpretations of the Contois kinetics function
3.4 Modeling studies . . . . . . . . . . . . . . .
3.5 Experimental studies . . . . . . . . . . . . . .
4 Analysis of single reactor with recycle
4.1 Introduction . . . . . . . . . . . . . .
4.2 Model equations and assumptions . . .
4.3 Results . . . . . . . . . . . . . . . . . .
4.4 Conclusion . . . . . . . . . . . . . . . .

xi

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

. . . . . . . . .
. . . . . . . . .
in microbiology
. . . . . . . . .
. . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.

.
.
.
.
.
.
.

8
8
10
15
21
23
25
25

.
.
.
.
.

31
31
33
33
34
36

.
.
.
.

46
46
47
51
71

CONTENTS

xii

5 Analysis of continuous ow bioreactor models with recycle around

72
each reactor
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.2 Model equations and assumptions . . . . . . . . . . . . . . . . . . . . 74
5.3 Mathematical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6 Analysis of continuous ow bioreactor models with recycle around
the whole reactor cascade
122
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
6.2 Model equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
6.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
6.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
7 Analysis of a chemostat model with
7.1 Introduction . . . . . . . . . . . . .
7.2 Results . . . . . . . . . . . . . . . .
7.3 Conclusion . . . . . . . . . . . . . .

variable yield
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .

8 Analysis of a chemostat model with

8.1 Introduction . . . . . . . . . . . . .
8.2 Results . . . . . . . . . . . . . . . .
8.3 Conclusion . . . . . . . . . . . . . .

substrate inhibition
171
. . . . . . . . . . . . . . . . . . . 171
. . . . . . . . . . . . . . . . . . . 173
. . . . . . . . . . . . . . . . . . . 183

9 Analysis of a chemostat model

substrate inhibition
9.1 Introduction . . . . . . . . . .
9.2 Results . . . . . . . . . . . . .
9.3 Conclusion . . . . . . . . . . .
10 Analysis of a model for the
vated sludge process
10.1 Introduction . . . . . . . .
10.2 Model equations . . . . . .
10.3 Biochemical processes . . .
10.4 Results . . . . . . . . . . .
10.5 Discussion . . . . . . . . .

coecient
157
. . . . . . . . . . 157
. . . . . . . . . . 159
. . . . . . . . . . 169

with variable yield coecient and

186
. . . . . . . . . . . . . . . . . . . . . . 186
. . . . . . . . . . . . . . . . . . . . . . 189
. . . . . . . . . . . . . . . . . . . . . . 208

treatment of wastewater by the acti210

. . . . . . . . . . . . . . . . . . . . . . . . 210
. . . . . . . . . . . . . . . . . . . . . . . . 212
. . . . . . . . . . . . . . . . . . . . . . . . 213
. . . . . . . . . . . . . . . . . . . . . . . . 217
. . . . . . . . . . . . . . . . . . . . . . . . 230

11 A nite rate of sludge disintegration unit model

11.1 Introduction . . . . . . . . . . . . . . . . . . . . . .
11.2 Yoon model (Innite rate SDA model) . . . . . . .
11.3 Revised Yoon model (nite rate SDA model) . . . .
11.4 Result . . . . . . . . . . . . . . . . . . . . . . . . .
11.5 Stability of the steady-state solutions . . . . . . . .
11.6 Discussion . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

242
242
243
244
249
250
253

CONTENTS

xiii

12 Conclusion
263
12.1 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
A The analysis for Chapter 6
268
A.1 Parameter value for Monod kinetics . . . . . . . . . . . . . . . . . . . 269
A.2 Stability of the washout solution . . . . . . . . . . . . . . . . . . . . . 272
B The
B.1
B.2
B.3

analysis for Chapter 9

Analysis of the quadratic equation . . . . . . . . . . . . . . . . . . .
Stability analysis for
the case a=0. . . . . . . . . . . . . . . . . . .
Analysis of + b . . . . . . . . . . . . . . . . . . . . . . . . . .

276
. 276
. 277
. 278

C The analysis for Chapter 8

C.1 Claim 1. The steady state solution is not realist when 2 . . .
1K

C.2 Claim 2. BR
1 , For any value of e on [a, K d ). . . . . . . . .
d

C.3 Analysis of + b . . . . . . . . . . . . . . . . . . . . . . . . . .

280
. 280
. 281
. 281

D List of Symbols

283

Bibliography

287

Chapter 1
Introduction
Biochemical and bioprocess engineering is swiftly becoming important areas of applied mathematics as the use of living cells to produce marketable chemical products
becomes increasingly important. There have been many powerful advances that are
transforming the eld, including applications to genetic sequencing, the production
of pharmaceuticals, biologics, commodities and medical applications such as tissue
engineering and gene therapy. It is a broad eld drawing its fundamentals from
biochemistry, microbiology, molecular biology and mathematical modeling.
The mathematical modeling of the bioreactor dynamic, specially the treatment
plant model includes clarication, aeration, settling (gravity thickening) and aerobic/anaerobic digestion, has attracted a lot of attention since the last decades.
The study of the biological treatment of wastewater stands at the crossroads of
several elds including biochemistry, microbiology, molecular biology, ecology, engineering, applied mathematics. The method of biological treatment of wastewater
has been extensively employed to reduce slurries and wastewaters pollutants that
include high concentrations of biodegradable organic materials before discharged in
1

Chapter 1

order to protect the environment. The bioreactor is a vessel in which wastewater

process is carried out which contain biomass grows through consumption of the
wastewater pollutants.
There are various dierent classications of bioreactors including batch, fed batch
or continuous, which generally refer to when and how much substrate is fed into the
bioreactor over time. The bioreactors environmental conditions, such as gas (i.e. air,
oxygen, nitrogen, carbon dioxide), ow rates, temperature, pH and dissolved oxygen
levels, and agitation speed/circulation rate, need to be closely monitored and controlled to ensure that the process continues as desired. Attracted by these tempting
mathematical challenges and encouraged by new tools developed in dynamical systems such as Lyapunov functions for stability, control, robust control [1], optimal
control, stochastic and software tools as Matlab and Maple, mathematicians have
studied the dynamic behavior, the steady state, the stability [18] and the control of
wastewater models [1,911]. However, in order to understand the evolution of mathematic modeling in wastewater treatment, we need to go back in history. Various
models have been used to model the wastewater processes. The most notable models
in this area are based upon the Monod specic growth rate expression [12]. These
models have been popular due to their ability to explain empirical data, and have
found wide application in many industries. However, extensive experimental work
has shown that the anaerobic and the aerobic degradation of wastewater originating
from industrial processes is often better described by the Contois specic growth
rate [13]. Interestingly, there has been only a limited number of modelling studies
employing the Contois model. Hence, this thesis begins by a presenting a case for

Chapter 1

Contois growth models utility followed by developing and discussing mathematical

models where the degradation of a biodegradable organic material is represented by
the Contois expression.
This thesis contain four parts each of which is divided into several chapters. In part
1 we introduce a background and explain the principles behind wastewater treatment and its mathematical modeling in chapter (2 followed by a survey of earlier
studies which have employed Contois model for specic growth rate in 3). In part
2, three dierent reactor conguration for wastewater treatment are studied across
three chapters: chapter (4), chapter (5) and chapter (6). In chapter (4), we analyze the steady-state operation of a generalized reactor model that includes both
a continuous-ow and a membrane reactor as limiting cases. We analyze the performance of the reactor by nding the steady-state solutions and determining their
stability as a function of the residence time. We show that periodic solution do
not occur. As an application of our results, we show that the recycle improves the
performance of the reactor at moderate values of the residence time. However, when
the value of the dimensionless residence time is suciently large, the performance
of the reactor is independent of the recycle ratio.
In chapter (5), the steady-state treatment of industrial wastewaters in a cascade
reactor with recycle is analyzed. A number of cascades with alternative arrangements of the settling units are considered. Specically, we consider the case when
the recycle stream leaving a settling unit placed around a reactor goes back into
the feed stream for that reactor. The steady-states for the model are found and
their stability determined as a function of the total residence time in the cascade.

Chapter 1

Asymptotic solutions in the limit of large total residence time are obtained for the
euent concentration leaving a cascade. This analysis is used to determine the reactor conguration that minimizes the euent concentration leaving the nal reactor.
It is found that, the optimised reactor cascade is obtained by using perfect recycle
around the nal reactor and imperfect recycle around the preceding reactors. When
only one settling unit is used, the performance of the reactor cascade is optimized
at short residence times by placing it around the rst reactor whilst at large total
residence times the performance is optimized by placing it around the nal reactor.
However, at suciently large total residence times there is a little benet gained by
using any settling units.
A common reactor cascade conguration employs a settling unit to recycle biomass
from the nal cascade reactor back to the rst. In chapter (6), we use steady-state
analyse to examine the process eciency of such a reactor conguration. It is found
that there is a critical value of the total residence time which identies a turning
point in the performance of the reactor cascade. If the total residence time is below the critical value then the settling unit improves the performance of cascade,
whereas, if the residence time is above the critical value then the performance of
the reactor cascade with the settling unit is inferior to that of a cascade without
one. It is shown that the critical values of residence time depends upon the values
of the recycle ratio and the concentration factor. We compare the performance of
a reactor conguration employing recycle around the whole cascade with that of a
cascade in which the settling unit recycles the euent stream leaving the ith reactor
into the feed stream for the ith reactor.

Chapter 1

In part 3, we extend the standard Contois growth-rate model allowing for either a
variable yield coecient and/or substrate inhibition which containing three chapters, i.e. chapter (7), chapter (8) and chapter (9). In chapter (7), we investigate the
eect of a variable yield coecient including the death rate of the microorganisms.
Analysis shows that the system has natural oscillations for some range of the parameters. We also investigate the eects of the death rate parameter on the region of
periodic behaviour. In chapter (8), we investigate eects of the substrate inhibition
on the performance of the reactor. In chapter (9), the biochemical reaction kinetics is governed by Contois growth model subject to both noncompetitive substrate
inhibition and a variable substrate yield coecient. The steady-state performance
of the reactor is predicted and stability of the steady-state solutions as a function
of dimensionless residence time reported. Our results identify two feature of practical interest. The rst feature corresponds to the case where the no-washout and
washout solutions are bistable. The second feature identies the parameter region in
which periodic solutions can occur when the yield coecient is not constant. These
features are often undesirable in practical applications and must be avoided. Scaling
of the model equations reveals that both the secondly bifurcation parameters are
functions of the inuent concentration. Our results show how the reactor behaviour
varies as a function of inuent concentration and identify the range of inuent concentration where the reactor displays neither periodic nor bistable behaviour.
A pressing problem associated with the operation of the activate sludge process is
the disposal of excess sludge. This can signicantly increase the operation cost associated with the process. A promising method to reduce the amount of the sludge

Chapter 1

is to destroy it in situ through the use of a sludge disintegration unit. In part

4, two models for sludge disintegration are considered across two chapters, namely
chapter (10), chapter (11).
In chapter (10), we analyze a model for the activated sludge process coupled to a
sludge disintegration unit. This model was proposed by Yoon [14], through in his
work the Monod specic growth rate law was used. Using steady state analysis we
nd a formula which identies a critical value of the inuent concentration. If the
inuent concentration is below criticality then the reactor is guaranteed to operate in a state of negative excess sludge production even in the absence of a sludge
disintegration unit. If the inuent concentration is higher than criticality then a
sludge disintegration unit must be used to ensure that the reactor is in a state of the
negative excess sludge production. For a continuous ow reactor the critical value
is relatively large and consequently a sludge disintegration unit is benecial for only
heavily contaminated industrial wastewaters. On other hand, for a membrane reactor the critical value is very low and in practice a sludge disintegration unit is
required to ensure that the reactor in state of the negative excess sludge production.
In chapter (11), we investigate a model for an activated sludge process consisting of a
membrane reactor coupled to sludge disintegration unit. This work extends a model
due to Yoon [14], which assumes that the processes inside the sludge disintegration
unit occur innitely quickly, to consider the processes in the sludge disintegration
unit occur at a nite rate. We investigate how the reaction rate in the sludge disintegration unit aects the formation of sludge in the reactor. Steady state analysis
is used to study this system and compare it with Yoons model. We consider two

Chapter 1

scenarios. In the rst scenario, we study the case in which the processes in the
sludge disintegration unit occur at an innite rate. As in chapter (10) there is a
critical value of the sludge disintegration factor above which the reactor system is
guaranteed to be in a state of negative excess sludge production. We show that, the
Yoons model is only correct if the sludge solubilization eciency is equal to one but
in some cases suciently accurate for practical purposes.
For the second scenario, we investigate the system with a nite reaction rate. We
nd that there is a critical value of the reaction rate which depends upon the size
of the sludge disintegration unit. If the reaction rate is above the critical value
then the error in assuming an innite rate is less than 10% of the exact value using
nite rate. For such value for the reaction rate, the behaviour of the reactor can
be estimated, within experimental error, by assuming an innite rate. Finally, we
show that there is a second critical value of the reaction rate. If the reaction rate
in sludge disintegration unit is suciently small, then there is no longer a critical
value of the sludge disintegration factor above which the reactor operate in a state
of negative excess sludge production for all residence times. Instead negative excess
sludge production can be only achieved only when the residence time is suciently
large.

Chapter 2
Literature review

2.1

Biological Wastewater Treatment

Biological wastewater treatment is a natural process in which bacteria are used to

degrade waste organic matter to a mixture of carbon dioxide and methane. It is
widely used to minimize the volume of contaminated wastewaters that are produced
by industrial and municipal sources.
Biological wastewater treatment consists of several stages, including primary and
secondary treatments of the wastewater. In the primary treatment, particulate
solids are mostly removed by sedimentation. This is followed by secondary treatment in which microorganisms, in the form of activated sludge, are used to remove
substrates (suspended organic solids and dissolved organic compounds) from the
wastewater. In the activated sludge process, microorganisms are suspended in the
wastewater, and this mixture is called mixed liquor.
The secondary treatment process is generally conducted in two steps. In the rst
step, bacteria degrades organic pollutants in an aerated biological reactor as food
8

Chapter 2

to sustain their life processes, including energy production, reproduction, digestion,

and movement. These processes inevitably result in byproducts that include water
and carbon dioxide.
The second step involves a settling unit (or clarier). The purpose of this unit is to
allow the sludge to settle to the bottom. Then, the substances at the bottom of the
clarier, which include activated sludge and mixed liquor, are recycled into the reactor, where they provide the nutrients required to keep the microorganisms alive and
active. The liquids that remain in the clarier are pumped to disinfection facilities
and discharged to receiving waters or sent for additional treatment (referred to as
tertiary treatment). A schematic of two stages of treatments of the wastewater is
shown in gure (2.1). The design of the activated sludge processes resolves around

Figure 2.1: Wastewater treatment stages

two main variables related to the treatment of sludge. These are: 1) The quantity
of suspended solids in the mixed liquor, often referred to as MLSS, and 2) the biochemical oxygen demand (BOD) or the chemical oxygen demand (COD), which are
indicative of the extent to which the wastewater is loaded with organic materials.
The microbial growth that occurs when the microorganisms are utilizing the nutrients in the wastewater can be used to dene the interaction between the concen-

Chapter 2

10

tration of microorganisms and the substrate. Microbial growth depends upon the
concentration levels of the substrate and the microorganisms in the bioreactor, these
levels in turn depend on the type of process and the operational mode. Thus, microbial growth depends on various types of processes which can be simple or complex
to determine and control [15].
There are several types of bioreactors. They are classied generally as either open
or closed systems. For an open system, the ow through the system is continuous.
In a closed system, the rate of growth of the biomass decreases with time, eventually approaching zero, which is the inevitable result when insucient nutrients are
available to sustain growth [16].
In our research, we consider the secondary treatment of wastewater in a continuous
ow reactor. We assume that the substrate and microorganism concentrations in the
reactor are well mixed i.e. the substrate and microorganism concentrations leaving
the reactor are the same as those in the reactor.

2.2

Anaerobic and aerobic wastewater treatment

Organic matter is removed from wastewater primarily by two biological processes,

i.e., anaerobic and aerobic processes. They have been applied successfully for treating waste euents from various types of domestic and industrial wastewaters. The
anaerobic process does not require oxygen for bacteria to break down organic matter, whereas oxygen is required in the aerobic process. A major disadvantage of
the aerobic treatment process is the amount of energy that it requires [17]. Thus,
in most cases, anaerobic treatment processes are preferable for treating waste ef-

Chapter 2

11

uents that have high organic content [1720]. The advantages of using anaerobic
treatment processes are [21],
1. Low cost. As oxygen is not required an anaerobic reactor is less expensive than
an aerobic reactor and it is simpler to operate and maintain.
2. Enhanced energy production. Anaerobic processes produce useful products in
the form of biogas that can be used to provide energy for other systems.
3. Very good stability. Anaerobic processes are stable under operating conditions
that would make other processes unstable.
4. Lower nutrient requirements. The requirement of nitrogen in the anaerobic
process is lower than that for an aerobic system.
5. Very high organic loading rate. The eciency of anaerobic systems for COD
removal is signicantly greater than that of aerobic systems.
6. Low area space.
7. No requirement for aeration.
The disadvantages of using anaerobic treatment processes,
1. The heavy metals and contaminants in sludge are accumulated.
2. The range of temperature control is narrowed .
3. Installing and managing an interrelated group of system to safely hand heating
of the tank.
4. Environmental Sensitivities

Chapter 2

12

The anaerobic process consists of three biological reaction stages in series. These are
the hydrolysis stage, the acidogenesis stage, and the methanogenesis stage. These
processes convert the organic matter to methane (CH4 ) and carbon dioxide (CO2 )
[22, 23]. Some researchers have proposed dierent numbers of stages, including two,
four, and nine stages [23, 24].
In the rst stage, insoluble organic polymers are broken down by bacterial hydrolysis
to soluble substrate that the bacteria can utilize. In the second stage, acidogenic
bacteria convert the products from the rst stage into carbon, hydrogen and other
materials, i.e, CO2 , H2 , NH3 , and organic acids such as acetates, propionates, and
butyrates. In the third stage, the products from the second stage are converted to
methane by methanogenic bacteria [2527]. The processes in all three stages are
generally dependent on dierent bacterial species.
In the past few years there has been an increased interest in the kinetics associated
with the three stages of the anaerobic digestion process. As a result, variations in
the anaerobic digestion process have been described extensively by models that have
been developed to represent the process [24,28]. Mathematical models that are used
to describe anaerobic processes to predict the reactor behavior are normally divided
into four categories,
1. Simple basic models such as Tessier [29], Monod [12], Moser [30] and Contois
[13].
2. Un-structured non-segregated model [31].
3. Un-structured segregated models [32].

Chapter 2

13

4. Structured kinetic models such as IWA models [3338].

The last three groups of models provide more comprehensive descriptions of the
anaerobic processes and are normally complicated than the simple basic models.
However, the simple basic models are often sucient for technical purposes and
are an accepted method to predict the behavior of biological wastewater treatment [3942]. The study [39] is of interest because all the numerical simulation
reported in this thesis use the kinetics value from this study.
Several accurate descriptive structured models have been internationally accepted
to simulate activated sludge processes such as the ASM-1, ASM-2, ASM-3 [37, 43].
The ASM-1 model includes 13 dierential equations and involves eight chemical
processes including the growth and decay of heterotrophic and autotrophic biomass,
ammonication of soluble organic nitrogen and hydrolysis. The model can predict
nitrogen, chemical oxygen demand, nitrication and denitrication processes including aerobic, anaerobic processes. Henze et al. [37] developed the ASM-1 model
by taking into account the biological nutrient removal in activated sludge systems,
leading to the ASM-2 model. This was achieved by incorporating the biological
uptake of phosphorus into the basic ASM1 framework. Biological nutrient removal
is handled by the addition of a module to model enhanced biological phosphorus
removal, and also by including chemical removal of phosphorus via precipitation.
Enhanced biological phosphorus removal refers to the biological uptake and removal
of phosphorus by the activated sludge system in excess of the amount removed by
normal aerobic activated sludge system.
Henze et al. [43] further extended ASM-1 by allowing the storage of organic sub-

Chapter 2

14

strates as a new process and improving the modeling of lysis (decay) of endogenous
respiration process. These models are analyzed by direct integration of the governing equations [43]. Nelson and Sidhu [38] investigated the steady states of the
ASM-1 model without recycle considering how the performance of the reactor change
as function of control parameters including the oxygen transfer coecient and the
residence time. The result of this study showed that there is a critical bifurcation
point in residence time which determines the performance of reactor.
Mosey [24] studied four groups of bacteria during the degradation of organic matter
to carbon dioxide and methane. These groups are acid-forming, acetogenic, acetoclastic methane, and hydrogen-utilising methane bacteria. A mathematical model
was developed to describe each of the steps of the kinetics processes where the Monod
model is used to simulate the reaction. Kalyuzhnyi [44] developed a mathematical
model of the batch anaerobic digestion of glucose, which involved ve dierent types
of bacteria, i.e., acidogenic, ethanol-degrading acetogenic, butyrate-degrading acetogenic, acetoclastic methanogenic, and hydrogenotrophic methanogenic bacteria.
This model also considered several factors and potential inhibitors that are dependent on bacterial action. A kinetic model was developed earlier for use in simulating
the various steps that occur when given substrate materials are degraded in an anaerobic sequencing batch reactor (ASBR) [45].
Munch et al. [46] developed a volatile acid production model and validated it against
the experimental data reported in the literature. Knobel and Lewis [47] developed a
mathematical model that can be used to assess the process of anaerobic digestion in
wastewater in which sulfate species exist in high concentrations. Their model incor-

Chapter 2

15

porated the Debye-Huckle theory, and it can be used to calculate activity coecients
and several other process parameters includes sulphate reduction, hydrolysis, acidogenesis, beta oxidation of long chain fatty acids, acetogenesis, and methanogenesis.
In this thesis, we represent the biological processes by considering a single substrate
concentration consumed by a single microorganism species in which the reaction rate
is described by the Contois model (to be described later, see chapter (3)). Thus, the
three stages at lined above are replaced by a single rate-determining step.

2.3

Microbial growth

Microbial growth increases the number of cells and the mass of the bacteria as
a consequence of cell division. Several factors can aect microbial growth, and
these are classied as intracellular and extracellular factors. Intracellular factors
include the internal structure, metabolic mechanisms, and genetic material of the
cell. Extracellular factors are external environmental conditions that aect the cells,
including pH, temperature, oxygen concentration, water, and food. The growth of
biomass in a culture requires a suitable environment in which the microorganisms
can live and grow. This environment must satisfy several conditions, including viable
inoculums, an energy source, nutrients, absence of inhibitors which prevent growth,
and suitable physicochemical conditions.
In the investigation of the kinetic behaviour of such a system, it is important to
understand how the concentrations of representative components of the system, i.e.,
cells, substrate, products, and byproduct, change with time. The study of microbial
growth in terms of various variables (various growth parameters) is useful for many

16

Chapter 2

purposes and is required to predict and control the behavior of the system. These
growth parameters are dened to describe the growth of both simple and complex
cultures. These parameters include specic growth rate, growth yield, metabolic
quotients for substrate utilization and product formation, and maximum biomass.

2.3.1

Specic growth rate

If all the requirements for growth are satised, then the biomass concentration
increases exponentially with time, since the overall rate of change of biomass is
proportional to the mass of biomass:
dX
= X.
dt

(2.3.1)

The dierential coecient ( dX

) is the population growth rate. X is the cell concendt
tration (kg cell/m3 ) and the parameter , which represents the rate of growth per
unit of biomass, is termed the specic growth rate.

2.3.2

Growth yield

Growth yield is a biological variable that allows us to assess the rates of production
and consumption of energy and mass in a biological system. From the standpoint of
modeling and describing such systems, the measured growth rate is quite benecial.
The growth yield is dened by,

Yx/s =

rx
,
rs

(2.3.2)

where rx is the amount of biomass produced and rs is the amount of substrate

consumed. When the reaction rates are equal to accumulation rates, which occurs

17

Chapter 2

in batch systems, the growth yield becomes,

dX
dt
Yx/s = dS
=
dt

dX
,
dS

(2.3.3)

For a continuous system, the growth rate is given by,

Yx/s =

X
X X0
=
,
S
S S0

(2.3.4)

where X0 and S0 are the initial biomass and substrate concentrations, respectively,
and X and S are the corresponding concentrations during the growth of the culture.
The growth rate of X and S have opposite signs, so the use of the negative sign
is required to reect that reality. For a growth-limiting substrate when the culture reaches its maximum biomass (Xm ) and is approximately 0, the growth yield
becomes,
Yx/s =

Xm X0
.
S0

(2.3.5)

Growth yield is an important variable because it is a quantitative expression of

the nutrient requirements of an organisms. As early as 1869, Raulin expressed the
nutrient requirements of fungus in term of growth yield. In some bacterial cultures,
the growth yield is constant when the conditions are maintained constant [48]. In
other studies, it is considered to be variable [49].

2.3.3

Metabolic quotient

The rate at which the substrate is utilized by organisms is called the rate of consumption of the substrate. This rate is given by
dS
= qX,
dt

(2.3.6)

18

Chapter 2

where X is the biomass, and the coecient q is known as the metabolic quotient or
specic metabolic rate. If the biomass concentration is constant and the environmental factors are constant, then q must be constant. In terms of growth, the rate
of consumption of a substrate is given by:
dS
X
=
,
dt
Yx/s

2.3.4

(2.3.7)

Eect of substrate concentration on growth rate

The microorganism growth rate (and therefore the microorganism concentration) is

related directly to the concentration of the substrate that is normally consumed by
the microorganisms. In some bacterial cultures, however, the growth rate is virtually
unaected by substrate concentration, i.e., zero order kinetics exists. When the
substrate consumption follows enzyme kinetics, the metabolic quotient is given by:
(
q = qmax

S
Ks + S

)
,

(2.3.8)

where Ks is the saturation constant, which is equivalent to the Michaelis-Menten

constant [50], and qmax is the maximum value of q. If we make the substitutions,
q=

and qmax =

max
Y

, then it follows that:

(
= max

S
Ks + S

)
.

(2.3.9)

This equation is known as the Monod equation.

2.3.5

Kinetic models of microbial growth

Two variables that may determine the rate at which microbial growth occurs are
the microbial populations specic growth rate () and the concentration of the sub-

19

Chapter 2

strate (S) such as in equation (2.3.9). This relationship is used to great benet in
many dierent specialties, including biotechnology, ecology, genetics, microbiology,
and physiology. During the last two centuries, extensive studies of microbial cultivation were conducted, and, as early as the 1830s, Cagniard de Latour, Kutzing,
and Schwann revealed that the growth of yeasts and other protists is responsible
for fermentation. An overview of the historical development of knowledge concerning microbial growth is presented in [16]. The understanding of microbial growth
has been improved by the understanding of principle of metabolic uxes and by
a number of mathematical models that have been proposed [51, 52]. In the 19th
century, various classical models were used to characterize the growth of microbial
populations, such as the Verhulst and Gompertz function [5355].
In 1912, the rst kinetic approach that was associated with the growth of microbes
was posited by Penfold and Norris [56], who proposed a saturation type of curve
to express the relationship between and S. This curve indicated that the maximum
rate of growth of the organisms (max ) is independent of high levels of substrate
concentrations [56]. Monods model is consistent with the Penfold hypothesis, but
some have been critical of it due to the fact that it includes the determination of
at a wide range of substrate concentrations, ranging from high to low [57,58]. Later,
Monod introduced a model that incorporated the concept that the substrate could
limit the rate of growth of the microbes,
(
= max

S
Ks + S

)
,

(2.3.10)

where = specic growth rate, max = maximum specic growth rate, S = substrate concentration, and Ks = substrate saturation constant. Monod used several

20

Chapter 2

parameters to dene the relation between growth rate and the utilization of the
substrate [48], i.e, Ks , max and the yield coecient, Yx/s [58],
dX
dS
Yx/s dS
=
.
Yx/s q.
X dt

Yx/s =

(2.3.11)

Many researchers have attempted to improve Monods representation for the growth
kinetics of cells by using three approaches. These are 1) investigating the eect of
physicochemical factors on the Monod growth parameters [5963] ; 2) adding a
new parameter in Monods model to account for the inhibition of the substrate or
product, the diusion of the substrate, maintenance, or the eects of cell density
on max [13, 16, 28, 57, 6366] ; and 3) the development of innovative kinetic theories
that support the mechanistic and empirical models [65,6772]. Contois adapted the
Monod model after discovering evidence that showed the specic microbial growth
rate may also depend on the microorganism concentration. The Contois expression
is given by
(
(S, X) = max

S
Ks X + S

)
,

(2.3.12)

where = specic growth rate, max = maximum specic growth rate, X= microorganism concentration, S = substrate concentration, and Ks = substrate saturation
constant. In this thesis, we use the Contois model to describe the growth rate.
The choice of this model is motivated by the increasing number of the experimental
studies showing the Contois model giving excellent t with experimental data. For
more detail see chapter 3.5.

Chapter 2

2.4

21

Substrate inhibition

Monod proposed the dependence of growth rate on the concentration of the substrate, and the concept was explored further by additional researchers in the eld.
These models t experimental data for biological processes in wastewater treatment
quite well, but Powells [73] showed that they are not valid at low concentrations of
some substrates or when high concentrations inhibit the growth of organisms.
Since high concentrations of a nutrient inhibit the growth of microorganisms, it
could be concluded that the Monod models only represent a special case rather
than a general relationship. High concentrations that inhibit growth may occur
during start-up or transient conditions. Models that describe the inhibitory eect
of substrate concentrations are very useful for modeling the biological treatment of
waste euents from industrial operations. For example, such euents may contain
ammonia, nitrates, phenols, thiocyanates, and volatile acids, all of which are known
to inhibit the growth of microbes [28].
Several empirical models that describe the inhibitory eect of substrates on microbial growth have been proposed. Haldane [74] proposed a popular homologous
equation that represents the uncompetitive inhibition of enzymes. Andrews [28]
later applied Haldanes equation to substrate-inhibited microbial growth. Other researchers used Haldanes equation to describe the kinetics involved in the inhibition
of microbial growth by various substrates [75, 76]. The equation describes some experimental data quite well, but it has been shown to be less than adequate for other
sets of data [77,78]. Andrews equation is the most widely used substrate inhibition

22

Chapter 2

model [7981].The Andrew/or Haldanes equation is given by,

(
= m

KM

S
+S+

)
S2
KI

(2.4.13)

The rate increases to its maximum value and then decreases as S increases, with
the shape of the curve depending on the values of KM and KI . At low substrate
concentrations, we have:
KM

S2
S+
KI

then
= m

S
KM

)
.

(2.4.14)

In this case the rate increases linearly as substrate concentration increases. At high
substrate concentrations we have :
KM

S2
+S
KI

then
= m

KI
S

)
.

(2.4.15)

In this case the rate decreases as the substrate concentration increases. There is a
maximum value of that is achieved at an intermediate value of S . This maximum
value of that can be obtained from [15] is given by,
S=

KM KI ,

=
2(

(2.4.16)
.

Ks
)
KI

(2.4.17)

+1

Han and Levenspiel [82] proposed a more general expression that has been used
to describe substrate inhibition,
(
= m

(1

S n
)
Sm

Ks + S (1

S m
)
Sm

(2.4.18)

where Sm is critical inhibitor concentration, and n and m are constants.

Most of the current information concerning the role of the substrate in inhibiting
biodegradation has been derived from studies using phenol. High concentrations

Chapter 2

23

of phenol inhibit the growth of microorganisms, and the kinetics of inhibition have
been described by several models [78, 8385]. Among these models, most of them
are empirical models, but they are quite eective for modeling at high phenol concentrations. Rozich et al. [86] reported that, among ve models that they evaluated, Andrews equation described the experimental data most accurately. However,
there is disagreement among various researchers concerning the viability of Andrews
equation for the intended purpose. Pawlowsky and Howell [84] observed that the
dierences between ve inhibition models were statistically insignicant. Yang and
Humphrey [78] made a similar observation for three models for describing the degradation of phenol by Pseudomonas putida and Trichosporon cutaneum. Tan et al. [87]
generalized a model for substrate inhibition by describing the inhibition of microbial growth associated with the substrate using statistical thermodynamics. The
selection of an appropriate model for representing the inhibition of the substrate
based on theoretical considerations was discussed by Goudar et al. [88] who found
Andrews equation to be suitable for describing the biodegradation of phenol.

2.5

Models with a variable yield coecient

In many models of bioreactors it is assumed that the yield coecient is a constant

during the course of the reaction. However, some experimental evidence has shown
that the yield coecient can be a function of substrate concentration. In this section,
we provide a brief overview of models using a variable substrate yield by extending
the literature review appearing in [89]. An example of Contois model with variable

24

Chapter 2

yield coecient can be presented as follows, see [49]

V

dS
(S, X)
= F (S0 S) V X
,
dt
Y (S)

(2.5.19)

dX
= F X(R(C 1) 1) + V X(S, X) Kd V X.
dt

(2.5.20)

where Y (S) = + S, with , > 0.

Note that the study based upon Contois kinetics is related to chapter (3).
Ajbar [2] analysed a continuous bioreactor using a general dependence for the growth
rate and a variable yield coecient upon the substrate concentration. A non-ideal
bioreactor is considered in which a bioreactor is divided into two regions, a wellmixed region and an unreacted region. This study showed that the variability of
the substrate yield is a necessary condition for the existence of periodic behavior.
This extends earlier nding of Crooke et al [90], who showed that periodic solutions
cannot occur for systems with a constant yield, from an ideal reactor to a non-ideal
reactor.
Zhu et al. [91] investigated a bioreactor in which the growth rate was assumed to
depend only on the substrate concentration. They claimed that limit cycles can
occur in a chemostat with two competitors for a single nutrient when there is a constant substrate yield. Sari [92] re-investigated the reactor model of Zhu et al [91]
and showed that there was an error in the computation of the eigenvalues in [91]
and that consequently the result in [91] is false. In this study the growth rate was
assumed to depend only on the substrate concentration.
A number of authors have studied the behavior of a chemostat with two microorganisms and a single substrate with variable yields [9395]. Song and Li [93] assumed
that the growth rate follows Monod kinetics with a variable substrate yield. Huang

Chapter 2

25

and Zhu [94] extended the study [93] by considering quadratic substrate yield. Huang
and Zhu [95] furthermore generalised the two linear yield functions to nth and mth
order-polynomials and replaced the standard Monod specic growth rate by a general non deceasing function. Sari [96] studied a chemostat model in which n species
compete for a single growth-limiting substrate by extending the results of [97] to
the case when yields are variable. The growth rate includes both monotone and
non-monotone response functions (which also includes Monod kinetics).

2.6

Endogenous processes

The endogenous processes, acting to reduce the amount of biomass, incorporates a

number of mechanisms including endogenous respiration, predation, cell death and
lysis [36]. The most important of theses is usually cell death. Decay has a signicant
inuence on the performance of biological reactors. Such eects include reducing the
sludge yield, an increase in electron acceptor utilization and a deterioration of the
performance of the reactor [98100]. Several investigators [38, 101] have combined
these endogenous processes into a single parameter as a rst-order processes due to
the dicultly of separation between these processes experimentally. Therefore, the
endogenous processes are included in all-but-one of the modeling chapter.

2.7

Cascade reactors

In this section we provide a brief historical discussion of the literature relating to the
mathematical analysis of cascade reactors. Cascade reactors using Contois kinetic

Chapter 2

26

are discussed separately in chapter (3).

The optimal design of a reactor cascade has been receiving attention since the early
1960s. A reactor cascade is optimized by adjusting the physical properties of the
reactor cascade, such as volume, interconnection structure of dierent tanks, recirculation rates, tanks shapes, with respect to the design objective such as the residence
time.
Harmand et al. [10] analyzed the optimal design of a two-reactor cascade including
two feed streams (multi-stream ow) and/or a recirculation loop in which a single
reaction occurs. The model assumed that the microorganism does not decay. The
objective of this study is to minimize the total volume of the cascade given a specied conversion. The steady-state design problem was analyzed for a generalized
growth rate law that included Monod kinetics subject to substrate inhibition.
The obtained result was that the appropriate allocation of the inuent into the reactor cascade in order to minimize the total volume depend on the ratio rate. The
ratio rate depends upon the euent concentration leaving the cascade. They showed
that the distribution of the inuent and introducing a recycle into the reactor cascade is only required to minimize the total volume when the ratio rate is less than
one. Harmand et al. [9] revisited the optimal design of a two-reactor cascade whilst
investigating enzymatic biological reactions. Latter Rapaport et al. [102] extended
the study on optimal design of two-reactor cascade to includes several microbial
species, instead of a single specie. The important nding of their study was that
the optimal design does not allow the coexistence of several species.
Powell and Lowe [103] studied the behaviour of a cascade of N reactors of equal vol-

Chapter 2

27

ume with recycle and with no death rate. Recycle was considered to occur between
the last reactor of the cascade and the rst reactor. As the number of the reactors
increased, they found that the behaviour of the system approached that of an ideal
tubular fermenter with plug ow.
Erickson and Fan [104] studied the optimum hydraulic regime for several activated
sludge systems composed of N reactors (N = 2 or 3) with recycle and a non-zero
death rate. They considered two points of view: the minimum total reactor volume
of the cascade required to produce the desired euent concentration; and an estimate of the cost of the organic waste being discharged and the total cost for the
volume of the total reactor cascade. They compared the results from the reactor
cascades with these obtained using a single tank.
Erickson et al. [105] examined the optimisation of a multi-stream cascade with N
reactor (N = 2, 3, 4, and 5), recycle, and a zero death rate. They considered two
stages in each reactor, a mixing stage and an aeration stage. The inuent entered
the reactor cascade in the mixing stage, where no reactions occurred. The objective
function was to minimize the total volume of the reactor cascade by determining
the appropriate allocation of the inuent into the mixing tanks and the distribution
of reactor volumes for specic concentrations of the organisms and substrate in the
recycle stream. The parameter value of the concentration factor (C) was 2 which is
also commonly used in practice.
The optimal cascade designed can be divided to two groups which are static [106]
or dynamical [107]. Several investigators have sought to determine the optimum
design of two-reactor cascade in series [107110]. Yang and Su [110] conducted their

Chapter 2

28

experiment in each reactor whilst residence times in each were equivalent. In their
ndings, they observed a superior performance for the two reactor cascade as compared to a single reactor of the same dilution rate. Basing their arguments on the
later observation, they inferred that enormous output will be experienced when two
reactor cascade are connected in series.
Chen et al. [109] examined the possibility of increasing performance in a single reactor and two-reactor cascade in series with a variable yield coecient as a result of
natural oscillations generation. For a single reactor using Monod, Tessier and Moser
growth models, they found out that the improvement of the reactor performance can
be achieved by operating at the optimal steady-state residence time which always
above the oscillatory region. Basing their study on the later observations, they compared the cascade reactor which has equal residence time against the single reactor
that has substrate inhibition in its growth kinetics. While analyzing two more complex microbial systems, Balakrishnan and Yang [108] visited again Monod-growth
model. In general, the output of a single chemostat system is poor as compared to
a two-chemostat-in-series system with the same total residence time.
Nelson and Sidhu [107] re-investigated the biological system of a two-reactor cascade
in series with no death rate that has been originally considered by Balakrishnan and
Yang [108], Chen et al. [109] and Yang and Su [110] in which the growth rate is
given by a Monod growth kinetics with a variable yield coecient. The criteria they
used to compare between the two congurations were dierent from that proposed
in [108110] wherein the performance of optimal single reactor compared with a
cascade performance when the residence time of former is the same, or smaller than

Chapter 2

29

that of the latter.

They found that the improvement of performance is obtained by using two-reactor
cascade than a single reactor for the choice of specic parameter. The nding of the
comparison with respect of the reactor productivity, is the single reactor surprisely
is superior than the cascade. It also showed that an insignicant improvement of
the cell mass eciency of the cascade which is not as the rst nding that reported
in [108] and [110].
Grady and Lim [80] studied biological wastewater treatment using reactor cascades
with and without recycle. Their study included the use of a single feed stream and
multiple feed streams. They investigated the optimisation of the reactor design from
two points of view, i.e., 1) the minimum euent concentration that can be obtained
for a specied total volume and 2) the minimum reactor volume that is required to
deliver a specied euent concentration. They found that the recycle ratio had no
signicant eect on either the substrate concentration or the microorganism concentration in each reactor. They concluded that recycle is not a signicant tool to
reduce the substrate concentration.
Using several and dierentiated equipments, Mantzaris, et al. [111] explained how
the growth process in multiple bioreactors can be modelled and also addressed the
following issues: (1) the chemical environment is likely to vary in diverse reactors
and (2) the biomass arises as distinct cells which are in reality separate and remain
separated from each other and thus a constituent of the biomass that go into the
reactors will not mix up with the biomass that was previously there. They developed dierent numerical algorithms for dierent models to solve the steady-state

Chapter 2

30

and transient problems and concluded that the heterogeneity in the biomass can be
ignored for the purposes of calculating the concentrations of biomass, nutrients, and
products in the abiotic environment if the structured model is linear in the state
vector.

Chapter 3
Literature review of Contois
growth kinetics

3.1

Contois growth kinetics

In many biological processes, the rate at which a population of microorganisms (X)

grows and the rate at which the substrate (S) is consumed are essential part of the
kinetic models. Often researchers have assumed that the specic growth rate is not
dependent on the concentration of the microorganisms [30,48,112115]. However, it
has been found experimentally that in some systems, the specic growth rate model
should include the concentrations of both the substrate and the microorganisms.
Such a growth kinetics model was introduced by Contois [13], who presented evidence
to show that the specic growth rate can also depend upon the microorganism
concentration from studies of batch cultures growing under conditions of nutrient

31

32

Chapter 3

limitation. The Contois growth model is given by,

(
(S, X) = max

S
Ks X + S

)
(3.1.1)

where the specic growth rate (S, X) is a function of microorganism concentration (X) and the concentration of the substrate (S). The parameters max and Ks
are the maximum specic growth rate and the saturation constant respectively.
These are constants under dened conditions.
Contoiss model shows that as the microorganism concentration increases, the growth
rate ((S, X)) decreases. The growth rate is given by,
(
G = max

S
Ks X + S

(
X = max

S/X
Ks + S/X

)
X.

(3.1.2)

There are two extreme cases for the Contois model as the population density of
biomass increases. In these cases, the Contois model reduces to rst-order kinetics
for either biomass or substrate concentration. These approximations are [116],
1. First-order kinetics for biomass growth.
S
Ks = G
= max X.
X

(3.1.3)

2. First-order kinetics for substrate consumption.

S
Ks = G
= max
X

S
Ks

)
.

(3.1.4)

The Contois growth rate has been used as a surface limiting model to explain the
mass transfer limitations due to the limited surface area when the biomass concentration is high [46, 117120]. In this case, the specic growth rate is expressed
as,
(
= max

S/X
Ks + S/X

)
.

(3.1.5)

33

Chapter 3

Thus, if the population density of biomass increases, leading to increased obstruction

to substrate uptake and growth of any particular microbe, then the Contois rate law
reduces to,
(
G max

3.2

S
Ks

)
.

(3.1.6)

Contois growth kinetic with inhibition

Substrate inhibition is a vital component (see chapter (2)) in microbial degradation

processes. Monod model does not include substrate inhibition. Andrew extended the
Monod model to include substrate inhibition and his extended form has been applied
to many biochemical processes. In the following sections, we establish that there
are many experimental situations where the Contois model is the most appropriate.
Hence we believe that extending the Contois model to include substrate inhibition,
could improve its applicability greatly over many domains. This is achieved by
modifying the growth rate model as follows,
(
= m

3.3

S
KM X + S +

)
S2
KI

(3.2.7)

Interpretations of the Contois kinetics function in microbiology

The Contois model is sometimes found to t experimental data better than models that are independent of the biomass concentration (such as the Monod model).
Researchers in several scientic elds have attempted to determine why this phenomenon occurs. Both Fujimoto [121] and Characklis [122] reported that the use of

34

Chapter 3

the Contois model can be derived by mechanistic means on the basis of saturation
kinetics when mass transfer is growth-limiting or when enzyme kinetics is growthlimiting. Furthermore, the availability of a local medium that is heterogeneous and
that surrounds the cell, producing restricted access to the substrate by the biomass,
is one of the most important features in the use of the Contois model [123]. Arditi
and Ginzburg, [124] evaluated the role of predator-prey in the procedure that determines growth rate. They found that describing population growth-rate processes
using both predator-prey ratios was more appropriate than considering only the prey
in the growth process. Other evidence that supports this hypothesis were provided
by Bail [125], Greenleaf [126] and Pearl and Parker [127]. They specically studied spatial heterogeneity and its role the growth process. Bail studied cell space,
and Greenleaf evaluated the reproduction rate of infusoria and how cell proximity
aected it.

3.4

Modeling studies

In this section we summarize the literature regarding reactor models that use Contois kinetics. Although the Monod expression has been the most popular model
used to describe microbial growth rate, several researchers have used Contois kinetics to simulate the behavior of the growth rate in wastewater treatment systems [3, 49, 101, 128131].
Nelson et al. [101] were the rst to use the Contois model to simulate growth rate
in a single reactor with recycle. They showed that the euent concentration can
be decreased to any desired level by operating the reactor at a suciently large

35

Chapter 3

residence time. This is not true for processes controlled by Monod kinetics in which
the euent concentration has a limiting value [132].
Nelson and Holder [128] studied the behavior of a cascade of N reactors without
recycle and reported the steady state solutions with associated stability analysis.
Their objective was to minimize the euent concentration as a function of residence
time and as the number of tanks in series. For the majority of cases considered,
a signicant decrease in the euent concentration was achieved by increasing the
number of reactors in the reactor cascade. They found that the value of the substrate concentration leaving a N reactor cascade (Sn ) reduces by

1
n

as the number

of the reactor increase where t is the total residence time.

Several authors have investigated the behavior of reactors in which the specic
growth rate follows the Contois model with a linear yield coecient [3, 49, 129131].
Nelson and Sidhu [49] analyzed a well-mixed continuously stirred reactor for both
a single reactor and a double reactor cascade. Assuming that the death rate was
zero, they considered two cases. In the rst case they analyzed the model for a
single reactor by nding the steady state solutions and identifying the condition for
washout. They also found a parameter region in which self -sustained oscillation was
generated. After a Hopf bifurcation, the time-averaged euent concentration is an
increasing function of the residence time so they concluded that periodic behaviour
is undesirable.
In the second case, they examined how the performance of a two-reactor cascade
changes as the residence time in the rst reactor is varied by assuming that the total
residence time in the cascade is xed. Operating at the same residence time, the

Chapter 3

36

performance of the cascade can be inferior to that of the single reactor. However,
the optimal performance of the cascade was always better than that of the single
reactor. Nelson et al. [129] further investigated a membrane reactor with a non-zero
death rate of the microorganism. They studied the model using the same analysis
approach as in [49] and investigated how the death rate aected the performance of
a single reactor and a double reactor cascade as well as the Hopf bifurcation region.
Ajbar et al. [3] extended Nelson et als study [49] to take into account oxygen transfer
limitations. The steady state solutions and their stability were determined and the
performance of the reactor was carried out by analyzing the process eciency and
the productivity. It was found that a decrease in the inuent concentration resulted
in an increase in the eciency and a decrease in the productivity of the reactor.
The eciency of the reactor was found to decrease as the oxygen transfer coecient
decreased. It was found that a variable yield coecient is necessary to produce a
periodic solution but it has a minor eect on the performance of the reactor.
Ajbar [131] considered a chemostat model in which two microbial populations competed for the same substrate with two dierent variable yield coecient. In this
study, the specic growth rates of both species are assumed to follow the Contois
model. Analysis of this model showed that the assumption of non-constant yield
may lead to complex behavior such as chaotic behavior.

3.5

Experimental studies

In this section we provide an overview of experimental research reported in the

literature which have employed the Contois growth model. We provide specic (and

Chapter 3

37

many) instances where the Contois model has proved to be the best choice for tting
experimental data when compared with alternate models like Monod. Specically,
in section 3.5.1, applications of this model in the processing of industrial wastewater
are discussed. Other types of applications are discussed in section 3.5.2.

3.5.1

Industrial wastewater

The Contois growth model has been applied to both aerobic and anaerobic industrial wastewater treatment processes in recent years. It has been shown that the
Contois model is suitable for tting various experimental data from a broad range
of organic materials [133135].
Beltran et al. [136] presented a study of the oxidation treatment of black olive
wastewater which was performed by an aerobic biological degradation process and
biological ozonation. In the aerobic biological treatment, the Contois kinetic was
used to describe the specic decomposition rate. It was found that the Contois
model ts the experimental data very well which conrmed that the Contois model
is suitable for the present experimental system.
Bhattacharya and Pham [137] studied how Monod and Contois kinetics t experimental data from the anaerobic treatment of cow dung digesters. The result of the
comparison indicated that the Contois kinetic model is more suitable for tting the
experimental data of cow dung digesters than the Monod kinetic model.
Hu et al. [39] investigated the process kinetics of the anaerobic digestion of ice-cream
wastewater using two kinetic models, Monod and Contois. It was found that the root
mean square for the Contois kinetic is much greater than that of the Monod kinetic.

Chapter 3

38

Thus, the comparisons of experimental data and predicted values obtained from
Monod and Contois models suggested that the Contois model is more appropriate
than the Monod model for modeling the process kinetics of the anaerobic digestion
of ice-cream wastewater, i.e., predicting the performance of the anaerobic digester
reactor, with the highest correlation coecient of 0.918. Both the performance of
the anaerobic digester reactor and microbial growth were predicted better by the
Contois model than by the Monod model. This was because the Contois kinetic
model considered the eect of inuent substrate in making its prediction [39].
Hu et al. [42] also studied the process kinetics for the anaerobic digestion of sulphaterich wastewater using models based on Monod and Contois kinetics at continuously
stirred tank reactor. The results of the kinetic studies indicated that the Contois
model predicted the kinetic reactions of the process very well, showing good agreement with the experimental data and having a greater correlation coecient of 0.989
than Monod model. Both the performance of the anaerobic digester reactor and the
microbial growth were predicted better (very well) by the Contois kinetic model
than by the Monod kinetic model. Hu et al. [42] suggested that the reason for the
good prediction of Contois kinetic model was due to the attachment of biomass at
the walls of the reactor which can provide a source of inoculum.
Isik and and Sponza [41] used several models to study the process kinetics of the
anaerobic treatment of textile wastewater in a lab-scale upow anaerobic sludge
blanket reactor, including Monod, Contois, Grau second order, modied StoverKincannon, and rst order kinetic. The results of their kinetic studies showed that
the Contois kinetic model, with a correlation coecient of 0.967, is more appropri-

Chapter 3

39

ate than the Monod and rst order kinetic models for describing microbial kinetics.
The results further emphasized that Contois model is the best model and is more
suitable for predicting the performance of anaerobic digester reactors compared to
the other applied models with a correlation coecient of 0.97. Krzystek et al. [138]
showed that the Contois kinetic model had good agreement with experimental data
from the aerobic biodegradation of solid municipal organic waste with a correlation
coecient of 0.985.
Moosa et al. [139] investigated the kinetics of the anaerobic reduction of sulphate
using several kinetic expressions including the Monod, Chen & Hashimoto and Contois models at continuous bioreactors. Among the models that were tested, the
Contois kinetic model provided the best agreement with the experimental data of
the anaerobic reduction of sulphate for describing the bacterial specic growth rate
on sulphate concentration with the highest accuracy. This process has several applications such as the cleaning of sulphate containing industrial euents and in the
cleaning of acid mine drainage. In their study, the kinetic coecients of the Contois
model were determined for dierent sulphate concentrations in the inuent. It was
found that the death rate coecient is not aected by variations in initial concentration of sulphate, and it remained constant.However, as the initial concentration
of sulphate increased, the saturation constant (Ks ) increased signicantly while the
change in the value of the maximum specic growth rate (max ) was insignicant.
These values are stated in table (3.1).
Sun et al. [140] studied the biological treatment of pharmaceutical wastewater by a
functional strain Xhhh employing three ions, Mn+2 , Cu+2 and Zn+2 . Three dierent

Chapter 3

40

biodegradation kinetic models, Tessier, Monod and Contois models, were used to
simulate the process. Each of the models gave a good t to the experimental data.
However, the Contois kinetic model was found to be the most appropriate for describing the microbial growth of Xhhh growth with the highest correlation coecient
of 0.984. Similarly, using three kinetic models, i.e., the Monod, Contois and Chen
& Hashimoto kinetic models, Abdurahman et al. [141, 142] reviewed the treatment
of palm oil mill euent using membrane anaerobic system. In both studies, they
found that the Contois model provided an excellent t with experimental data with
correlation coecients of 0.962 and 0.997, respectively. In the latter case the Contois
model gave the highest value for correlation coecient. Poh et al. [143] cultivated
a thermophilic mixed culture for treating the palm oil mill euent at thermophilic
conditions using a continuous stirred tank reactor. In this study, the Contois model
provided a good t with the experimental data with R2 values ranging between 0.82
and 0.97 depending on the euent concentration.
Pinna et al. [144] investigated caeic aside biodegradation using a mixed culture in
a batch reactor. The Contois model, with an R-squared value of 0.851, was found to
give a minor improvement in tting experimental data over the Monod model, which
had an R-squared value of 0.848. Zhang et al. [145] developed a kinetic model to
describe simultaneous saccharication and co-fermentation of paper sludge using the
xylose-utilizing yeast Saccharomyces cerevisiae RWB222 and the commercial cellulase preparation Spezyme CP. Their results showed that the Contois model gave the
best description of the specic rate of growth on xylose.
Paulo et al. [146] used a bioreactor to treat mining inuenced water that had been

Chapter 3

41

discharged from abandoned mining sites. The Contois model and the rst-order
model were used to simulate the decomposition process. They compared their results with the predictions based on the Contois and the rst-order models to conclude
that the Contois model stayed closer to the experimental data. Emerald et al. [147]
investigated the kinetics of the activated sludge process treating synthesized dairy
wastewater using four kinetic models, i.e. the Monod, Moser, Contois and Chen &
Hashimoto models, at an organic loading rate of 1200 mg L1 . Among the tested
models, the Contois kinetic model, with a correlation coecient of 0.95, was found
to be an appropriate model and the best kinetic model for describing the kinetic
reactions of the activated sludge process due to tting the data very well.
Karim et al. [148] proposed a new model, including the Contois kinetic model and an
endogenous decay model to predict the methane production rate from the anaerobic
digestion of cow manure in bench-scale gas-lift digesters. The proposed model and
two other well-known kinetic models, the Chen & Hashimoto [149] and Hill [150]
models, were used to t the experimental data obtained for methane production
rate. It was found that the tting of the Chen & Hashimoto and Hill models with
the experimental data was poor with the values of the correlation coecient being
0.86 and 0.51 respectively whereas their new model featuring Contois kinetics tted the experimental data of observed methane production rate with a correlation
coecient of 0.99. Thus, the comparisons of experimental values with the results
produced by the three models showed that the proposed model that included Contois model is more suitable for modeling the methane production rate.
The rst-order kinetic model has been traditionally used to simulate hydrolysis re-

Chapter 3

42

action in anaerobic digestion which is independent of the hydrolytic microorganism

concentrations. However, some researchers have shown that the hydrolysis step depends on the microorganism concentration and activity [151]. The rst-order kinetic
may not describe the hydrolysis steps accurately [135] due to the complexity of the
multi-step process involved in the hydrolysis of carbohydrase, proteins and lipids
hydrolysis while the Contois model was found to t the experimental data very
well [135]. This result has been supported by several studies in which the Contois
model was used to illustrate the hydrolysis of particulate wastes [116118, 152].
Ramirez et al. [116] developed anaerobic digestion model No.1 (ADM1) to describe
the thermophilic anaerobic digestion of thermally pre-treated waste activated sludge
using a batch reactor that involved three disintegration biochemical parameters, nine
hydrolytic biochemical parameters and four stoichiometric parameter values. The
Contois model was used to simulate the disintegration and hydrolysis processes. It
was found that the Contois model gave a better t to the experimental data than the
standard ADM1 model which uses a rst-order kinetic expression. Vavilin et al. [118]
discussed four kinetic models, including Monod, rst-order, two-phase and Contois
kinetics, for the hydrolysis of particulate organic material in anaerobic digestion.
The obtained results showed that using the Contois kinetic for the hydrolysis kinetics of swine waste, sewage sludge, cattle manure and cellulose provided excellent
ts to the experimental data. Gawande et al. [153] presented the development
of a generalized biochemical process model. The hydrolysis of particulate matter
was modeled using Contois kinetics which also described the microbiologically mediated reaction. Gawande et al. [152] used their model to simulate the anaerobic

Chapter 3

43

reduction of municipal solid waste. The use of Contois kinetics was found to give
an excellent prediction in the reduction of solid concentrations. Myint and Nirmalakhandan [117] compared three models, i.e., the rst-order, the second-order
and the surface-limiting reaction models (also known as Contois kinetic model) for
simulating hydrolysis-acidogenesis in the digestion of cattle manure residues using a
batch reactor. They found that the Contois kinetic model t the experimental data
very well, with a correlation coecient of 0.914, which was better than the other
applied models. Therefore, the results discussed above provide convincing evidence
that the substrate-microorganism ratio (S/X) used in the Contois model could be
a better limiting factor in the hydrolysis of particulate substrate, rather than the
substrate concentration (S) as modeled by the rst-order reaction model.
Several researchers have conducted theoretical studies of a continuous ow bioreactor
model using Contois kinetics [39,41,118,137,139,154]. One should note that at high
feed substrate concentrations, the growth rate as a function of residual substrate
concentration has been predicted successfully by the Contois kinetic model [31].

3.5.2

Other applications of the Contois model

In this section we review other experimental applications of the Contois model.

Hidaka et al. [155] developed a model to describe the lactate fermentation characteristics of B. Coagulans in a batch reactor. The model included the inhibitory
eects of the substrate, lactate (product), NaCl, and bacterial growth. The Contois
kinetic model was used to simulate the degradation of particulate carbohydrates.
Their results indicated that the Contois model is more suitable for simulating the

Chapter 3

44

hydrolysis of particulate kitchen garbage than rst order reaction. This result is in
agreement with other studies that used the Contois model to describe the anaerobic
hydrolysis of particulate organic wastes [116, 156].
Several studies have indicated that the Contois equation can be used to describe
fungal growth kinetics [157]. Zhou et al. [158] studied the growth kinetics of Rhizopus nigricans fungal on glucose using the Contois model. The results showed that
the Contois model is useful for simulating the kinetics of cell growth due to the high
value of the correlation coecient of 0.99.
Mazutti et al. [159] developed a phenomenological model containing 19 kinetic models for microbial growth to simulate inulinase production in a batch reactor by the
yeast Kluyveromyces marxianus NRRL Y 7571, employing a medium containing
agroindustrial residues as the substrate. The results of this study showed that the
Contois model was the most appropriate model for use in representing inulinase
production.
Hernalsteens and Maugeri [160] investigated the Rhodotorula red yeast strains which
produce extracellular enzyme. The growth kinetics of the microorganisms was described by the Contois model. It was found that the Contois model gave a good
agreement with the experimental data, with a high correlation coecient of 0.95
while the classic Monod model did not t this process.
The Contois model has been used to simulate the algae growth [161] and to model
biological reactions in composting [162, 163]. In addition, it has been receiving increasing attention in ecology [164166].

45

Chapter 3

Table 3.1: Kinetic constants obtained using the Contois kinetic

Substrate

Initial

concen-

max

Ks

0.9297 day1

0.4818

Kd

R2

Reference

0.0131 day1

0.918

Hu et al. [39]

0.0531 day1

0.989

Hu et al. [42]

0.0065 day1

0.967

Isik

tration
Ice-cream wastewater

0.2116
1

Sulphate-rich wastew-

0.6796 day

ater
0.105 day1

Textile wastewater

(g COD)(g VSS)

(gVSS)(g COD)

0.8850

0.2659

(g COD)(g VSS)1

(gVSS)(g COD)1

0.465

0.125

(mg COD)(mg VSS)1 (mg VSS)(mg COD)1

Solid

municipal

or-

ranged from 0.18

ganic waste
Sulphate

to 0.249 h

and

Sponza

[41]

289 g O2 dm3

0.985

Krzystek

et al.

[138]

concentra- 1.0 kg m3

0.058 h1

0.027 kg m3

0.035 h1

Moosa et al. [139]

concentra- 2.5 kg m3

0.061 h1

0.038 kg m3

0.035 h1

Moosa et al. [139]

concentra- 5.0 kg m3

0.063 h1

0.071 kg m3

0.035 h1

Moosa et al. [139]

concentra- 10.0 kg m3

0.065 h1

0.125 kg m3

0.035 h1

Moosa et al. [139]

0.43 day1

K=0.89,K=Ks

0.03 day1

0.99

Karim et al. [148]

17.6 day1

0.517

3.2 day1

0.978

Guerrero et al [167]

0.86

Ardestani [168]

0.53 0.55

0.851

Pinna et al. [144]

tion
Sulphate
tion
Sulphate
tion
Sulphate
tion
Cow manure
Synthetic

domestic

wastewater

(hy-

drolytic

two

stage,

(g COD)(g VSS)

reaction)
Penicillium brevicom-

0.04 h1

0.31 g L1

pactum
Caeic Acid

0.1200316

5.6 33

0.068
Pharmaceutical

(Monod=0.848)

1.264(before),

wastewater

2.276(after)

558 (before) , 558 (after) mg L

0.984

chryso-

Sun et al. [140]

fore)(best,m.g)

day1
Penicillium

(be-

0.810 (after)

0.103 h1

0.145(-)

0.445 g g1

0.984(best,m.g)

Goudar et al. [157]

0.133 h1

6.293g L1

0.154(-)

2.503(SSR)best

Mazutti et al. [159]

genum on glucose
Inulinase production

(Monod=2.521)
palm oil mill euent

0.344 day

0.111 (-)

0.344 day

0.111 (-)

dairy

0.62

0.21 day

(g VSS)(g COD)1

(POME) wastewater
Synthesised

0.24 day

(g VSS)(g COD)1

(POME) wastewater
palm oil mill euent

0.67

2.13 day1

0.20 mg(COD)L1

0.68(-)

0.070 day1

0.8 h1

22.4 g L1

0.5 g g1

wastewater
Extracellular enzyme

99.7

best

Abdurahman et al.

(Monod=99.4)

[142]

96.2

Abdurahman et al.

(Monod=97.1)

[141]

0.95

Emerald et al. [147]

(Monod=0.93)
0.95

Hernalsteens
Maugeri [160]

Note: 1-Before and after optimization with Mn2 , Cu2 and Zn2 .

and

Chapter 4
Analysis of single reactor with
recycle

4.1

Introduction

In this chapter, the treatment of industrial wastewater in single reactor with recycle
is studied. A very similar model was investigated earlier by Nelson et al [101] for a
slightly dierent reactor conguration. In this earlier study the reactor membrane
was assumed to be placed after the settling unit. In this chapter, a more natural
conguration is used in which the reactor membrane is placed before the settling
unit. These two models are identical when there is no membrane present. However,
even for this case we present several new results which did not appear in [101].
Figure (4.1) shows a bioreactor with recycle. In such a reactor, the euent emerges
from the reactor and ows into the settling unit. Microorganisms settle to the
bottom of the tank, and they are recycled into the reactor vessel. The remaining
microorganisms, the substrate, and the reaction products ow out the reactor. One
46

Chapter 4

47

advantage of using a setting unit is that it increases the concentration of the microorganisms inside the reactor, thereby increasing the eciency of the process.
The objectives of the current chapter are:
1. to provide a more detailed investigation of the model.
2. to investigate the conditions that minimize the concentrations of the pollutants
in the euent.

4.2
4.2.1

Model equations and assumptions

Model assumptions

The assumption that the substrates and microorganisms in the biological reactor can
be represented as a single substrate species and a single microorganism species is an
acceptable method to predict the complex process of behavior of biological wastewater treatment [3942]. Thus, in this thesis, we consider the single substrate species
and a single microorganism species where the overall process kinetics is controlled
by a rate limiting step. The endogenous respiration, predation, cell death, and lysis
are represented as rst-order processes by the term, Kd X in equation (4.2.2) [169].
Three assumptions are important in the process model. These are:
- Substrate is not concentrated in the settling,
- The utilization of the substrate only occurs in the reactor and
- The reaction does not occur in either the settling tank or the return line.

48

Chapter 4

Reactor

FS0 , FX0

S,X

F!X + F!RX

Settling unit

F Xe

Recycle, RFC!X

Figure 4.1: Schematic diagram for bioreactor with recycle. F: Flow rate. V: reactor
volume. R: recycle ratio. C: recycle concentration factor.
In this study, a system of nonlinear ordinary dierential equations are used to model
the biodegradation of organic substances in wastewaters. The microorganism (X)
grows through consumption of a substrate species (S). The specic growth rate,
equation (4.2.3), is given by a Contois expression. The dependent variables are the
concentration of organic substrate and the concentration of the degrader (microorganism), the independent variable is time. The dimensional and dimensionless forms
of our model are stated in sections (4.2.2) and (4.2.4), respectively.

4.2.2

The dimensional model

The model equations for reactor with recycle around the reactor are given by,
dS
(S, X)
= F (S0 S) V X
,
dt

dX
= F (X0 X) + RF (C 1)X + V X(S, X) Kd V X,
V
dt
V

(4.2.1)
(4.2.2)

The specic growth rate is

(
(S, X) =max

S
Ks X + S

)
.

(4.2.3)

49

Chapter 4

The residence time is

V
= .
F

(4.2.4)

All parameters in the model are strictly non-negative. For details of the denition
of model parameters, we refer readers to the appendix (D).
The term (F (R[C 1]1)X) in equation (4.2.2) is dierent than the corresponding
term (F (R[C 1])X) considered in [101]. In the latter expression the parameter
can take value of 0 or 1 [101]. As noted in introduction, these two terms are equal
when = = 1. The dierences in these expression when = arise from the
dierent assumptions regarding where the reactor membrane is located. Thus, the
conguration considered in this chapter is more widely used than that used in [101],
See Mark I Neslon, private communication.
We consider a general reactor model where the choice of value for parameter determines the conguration of the reactor. When = 1 the bioreactor is a continuous
ow reactor. When = 0 it is an idealized membrane reactor, in which all of the
microorganisms is forced to remain in the reactor vessel. The choice 0 < < 1
gives a non-idealised membrane reactor which is equivalent to a reactor with recycle
because the parameter is related to the recycle parameter. In this case some of
the microorganisms leave the reactor vessel and some of them stay at the reactor
vessel. For a non-idealised membrane reactor to be operationally eective we require 0 < 1.
For a specic wastewater, a given biological community and a particular set of environmental conditions the parameters Ks , Kd , and max are xed. The parameters
that can be varied are C, R , S0 , X0 and . In our numerical simulations we

50

Chapter 4

use parameter values for the anaerobic digestion of ice-cream wastewater proposed
by Hu et al. [39]. These are: = 0.2116 (gVSS)(g COD)1 , max = 0.9297 (day1 ),
Kd =0.0131 (day1 ) and Ks =0.4818 (COD)(g VSS)1 . It should be noted that the
previous information is hold for all following chapters.

4.2.3

Settling unit

A settling unit improves the performance of the reactor by increasing the concentration of microorganisms and is modeled by the concentration factor C. This concentration factor C depends upon the design and operation of the settling unit and
sludge properties such as settling, thickening and compressibility behaviour [101].
The microorganism concentration leaving the separating unit back to the reactor is
given by CX in equations (4.2.2).
By taking a mass balance over the settling unit it can be shown that maximum value
of the concentrating factor Cmax is given by
Cmax = (

4.2.4

1
+ 1).
R

(4.2.5)

The dimensionless model

To aid in the analysis, equations (4.2.1 and 4.2.2) are written in dimensionless form
by introducing dimensionless variables for the substrate concentration [S = S/S0 ],
the microorganism concentration [X = Ks X/S0 ], and time [t = max t]. This generates the following dimensionless equations:
1
S X
dS

=
(1

S
)

,
dt

(S + X )

(4.2.6)

S X
(R 1)
dX
=
+
X Kd X .
dt
(S + X )

(4.2.7)

51

Chapter 4

In equations (4.2.6, 4.2.7), all dimensionless parameters in the model are strictly
non-negative. For details of the denition of model parameters, we refer readers to
the appendix (D). The model given in equations (4.2.6, 4.2.7) is investigated with
the assumptions that the concentration of microorganisms owing into the reactor is
zero [X0 = X0 = 0]. Also, it is assumed that the concentration of substrate entering
into the reactor is greater than zero [S0 > 0] and the death rate is strictly positive
[Kd > 0]. Once should be noted that this assumptions are hold for all following
chapters.
Using the same values for the anaerobic digestion of ice-cream wastewater [39],
we have nd that = 0.1019 and Kd = 0.0141. Using the formula of R =
(C 1)R, the maximum value of R is obtained by using the maximum value of the
concentration factor (Cmax = (1 + R1 )) and is given by

Rmax
= 1.

From equation (4.2.7), it follows that a ow reactor model with idealized recycle
(R = = 1 ) is identical to idealized membrane reactor model without recycle
( R = = 0). In the following analysis of equations (4.2.6, 4.2.7) all terms
are considered to be dimensionless, therefore, all reference to dimensionless will be
dropped.

4.3

Results

We analysis the generalized reactor models by consider the cases 0 1 and

0 R 1. In section (4.3.1), an invariant region is determined. In section (4.3.2),
the steady state solutions branches are found and the conditions of no-washout

52

Chapter 4

solution branch (The no-washout solution refers to when the microorganism concentration is positive) to be physically meaningful are stated. In section (4.3.3) and
(4.3.4) the stability of the steady state solutions are determined and, the washout
solution (The washout solution refers to when the microorganism concentration is
equal to zero) is shown to be globally stable if it is locally stable. In section (4.3.5)
the no-washout solution (4.3.14) is globally asymptotically stable if it is locally stable. In section (4.3.6) the transcritical bifurcation points are presented. Asymptotic
solutions for large residence times are determined in section (4.3.7) while, in section (4.3.8), these solutions are presented for residence times just higher than the
washout point. In section (4.3.9), we discuss steady-state diagrams for the euent concentration and microorganism concentration leaving the reactor. In section
(4.3.4) we show that the washout solution is globally stable when Kd 1. Hence,
we assume in the following that 0 < Kd < 1.

4.3.1

Invariant region

In this section our aim is to establish the basic properties of invariance for the
region = {(S , X ) R2+ , 0 S 1, X 0}.
On the boundary of , if X = 0 with S > 0 then
X = 0 is invariant. If S = 0 with X > 0 then

dS
dt

dX
dt

= 0. Thus, the plane

> 0. Thus, the vector

eld of S is pointed to inside the region . However, the point (S , X ) = (1, 0) is

a steady-state solution and solution trajectories can not leave through it.
The model (4.2.6, 4.2.7) is not dened at the point (S , X ) = (0, 0). We show
that no solution can enter this point. If X = 0, then

dS
dt

> 0, thus S (t ) is

53

Chapter 4

an increasing function. Now suppose X > 0. Then

1
S X
dS

=
(1

S
)

,
dt

(S + X )

1
S

(1

S
)

For S suciently small. Hence,

dS
dt

(4.3.8)

> 0 and the value of S increases away from 0.

Hence the region is (positively) invariant.

For the boundness of S and X , we have that,

dS
1
S X

=
(1

S
)

,
dt

(S + X )

1
(1 S ),

S > 0, X 0.

(4.3.9)

It follows that
[

t
S 1 + C exp

(4.3.10)

Where C is an arbitrary constant. i.e. the solution trajectory is attracted into the
invariant region. Hence, the substrate concentration is bounded. We now show that
the concentration of microorganisms is bounded.
Let Z = S + X , Adding equations (4.2.6) and (4.2.7) we have,

(1 R )
dZ

=
(1

S
)

(
+ Kd )X ,
dt

(1 R )
(1 S )
X ,

(1 R ) S (1 (1 R )) S (1 R )

X ,

(1 R )( S + X )

,

(1 R )Z
=
(4.3.11)

54

Chapter 4

It follows that
[
]

(1 R )t
Z
+ C1 exp
(1 R )

(4.3.12)

Where C1 is an arbitrary constant. Hence the function Z = S + X is bounded

and as S is bounded so is X .

4.3.2

Steady State Solutions Branches

Steady state solutions of the system for the equation (4.2.6, 4.2.7) are found by
setting the derivatives terms equal to 0 and given by,
Washout branch:
(S , X ) = (1, 0).

(4.3.13)

No-washout branch:
(S , X ) = (

((R 1) + (1 Kd ))
,
)
a
a((1 R ) + Kd )

(4.3.14)

where
a = (R 1) + (1 Kd ) + .
The no-washout branch in equation (4.3.14) is physically meaningful only when the
substrate and microorganism concentrations are positive (ie. where S > 0 and
X > 0). Upon analysing both components of (4.3.14) we nd that
>

(1 R )
1 Kd

(4.3.15)

is required for S > 0 and X > 0.

By dierentiating the S component of equation (4.3.14), the eect of recycle on the

55

Chapter 4

substrate concentration is determined. We obtain that

dS

=
< 0.
dR
[ + (1 R ) + (Kd 1)]2

(4.3.16)

Now the denominator in equation (4.3.16) is equal to zero when

=

(1 R )

(1 R )

<
.
(1 Kd )
1 Kd
(1 Kd )

It is shown latter that this condition is the washout point. Hence, for physically
meaningful solution

dS
< 0 which means that substrate concentration is a decreasdR

ing function of the eective recycle parameter (R ). Thus, at xed residence time,
the performance of reactor reaches the highest level of eciency at the maximum
value of the eective recycle parameter (i.e, R = 1).
From equation (4.3.14), it can be shown that,
dS
(1 Kd )
=
< 0.
d
[ + (1 R ) + (Kd 1) ]2

(4.3.17)

Since 0 < Kd < 1 then the substrate concentration is a decreasing function of

residence time along the no-washout branch.

4.3.3

Stability of the steady state solutions

The Jacobian matrix for the system (4.2.6) and (4.2.7) is given by

J(S , X ) =

X 2
(X +S )2

X 2
(X +S )2

S
(X +S )2
(R 1) Kd

S
(X +S )2

(4.3.18)

Evaluating the Jacobian matrix (4.3.18) at the washout steady state solution
produces

J(1, 0) =

(R 1) Kd

+1

(4.3.19)

56

Chapter 4

The eigenvalues of this matrix are given by

1 =

1
<0

and

2 =

(R 1) Kd
+ 1.

Hence, the washout branch is stable when 2 < 0 which is when

(1 R ) > (1 Kd ).
It can be seen that when
-

Kd > 1 then 2 < 0 for all ,

Kd = 1 then 2 < 0 provided that 0 R < 1,

0 < Kd < 1 then 2 < 0 for <

Kd = 1 and R = 1 then 2 = 0 and

(1 R )
=
then 2 = 0
(1 Kd )

(1 R )
,
(1 Kd )

The Jacobian matrix along the no-washout branch can be written as

J(S , X ) =

X 2
(X +S )2

X 2
(X +S )2

S
(X +S )2
X S
(X +S )2

(4.3.20)

For physically meaningful solutions to the system of equations (4.2.6, 4.2.7), we

require that X > 0 and S > 0. From the Jacobian matrix (4.3.20), the trace and
determinant are:
(

)(
)
X S
1
X 2
det(J) =
+
(X + S )2
(X + S )2
)(
)
(
S22
X 2
> 0.
+
(X + S )2
(X2 + S2 )2
(
)
X S
1
X 2
trace(J) =
+
+
< 0.
(X + S )2 (X + S )2

(4.3.21)

(4.3.22)

57

Chapter 4

Since the trace is negative and the determinant is positive for the Jacobian matrix then the no-washout branch is stable for all physically meaningful solutions of
equations (4.2.6, 4.2.7).
Theorem 4.3.1 The washout solution (1,0) is globally asymptotically stable pro
vided < cr
on .

Proof. To prove global stability of the washout solution (1,0) whenever < cr , we
consider the following Lyapunov function V (S , X ) = X ,
V = X [

S
(R 1)
+
Kd ],
(S + X )

X [1 +

(R 1)
Kd ] = X [ cr ].

By hypothesis we have < cr , this leads to V < 0. Since the region is positively
invariant, the washout solution (1,0) is globally asymptotically stable in . The
analysis shows that if Kd > 1 then the washout solution is globally stable. Thus no
practical biochemical plane can be operated when Kd > 1.

4.3.4

Global stability for washout solution

In this section, we show that when the washout solution is locally stable, it is also
globally stable. We do this by using the argument from [170].
There are three cases to consider:
1. Kd > 1,
2. Kd = 1 provided that R = 1 and
3. 0 < Kd < 1 provided that <

(1 R )
.
1 Kd

58

Chapter 4

Proof for the rst case can be done by assuming that X > 0, then from equation
(4.2.7) we have that
dX
S X
(R 1)
=
+
X Kd X

dt
(S + X )

(R 1)
X Kd X + X

as

S
1
(S + X )
(4.3.23)

((R 1) + (1

X (1 Kd ),

since

Kd ))

Rmax
=1

It follows that X is decreasing function of t . We know that the region X 0 is a

positively invariant region. Hence, the microorganism concentration X approaches
zero as t and therefore, the washout solution is globally stable.
The cases ((2) and (3)) use a similar argument to the above. Furthermore, the case
when linearized stability is not dened (ie. when 2 = 0) can be also be considered.
Two more cases need to be consider for this circumstance. These are:
4. R = 1 provided that Kd = 1 and
5. =

(1 R )
.
1 Kd

By using a similar argument as above it can be shown that the washout solution is
globally stable for these specic cases.

4.3.5

Global stability of no-washout solution

Theorem 4.3.2 The no-washout solution (4.3.14) is globally asymptotically stable

on .
provided > cr

Chapter 4

59

Proof. Since the system (4.2.6 and 4.2.7) is a plane dynamical system dened on
invariant domain and since the trajectories are bounded, we can apply the theorem
of Poincare-Bendixson [171] assuring that there are three cases, for a trajectory
which are,
1. It is a periodic solution,
2. It converges to a periodic solution,
3. It converges to an equilibrium point.
On the other hand, since the bounded domain containing trajectories is connected,
we can apply Dulac-Bendixson criteria to prove that there is no periodic orbits.
Using Dulacs test [172] it is shown that there are no periodic solution for the
system of equations (4.2.6 and 4.2.7).
1
then the Dulacs test yields that
X
[
]
d
(X + S )2 + X ( S + X )
dS
d
dX
[
]+
[
]=
< 0.(4.3.24)
dS dt
dX dt
X (X + S )2

Let =

Equation (4.3.24) is negative when the components of (S , X , , ) are all positive.

This is true for physical meaningful solutions therefore, by Dulacs test, the system
of equation (4.2.6, 4.2.7) has no periodic solutions. So since there is no periodic
solution and since > cr then the washout solution is not stable then the only
possibility that a trajectory has, is to converge into the no-washout solution. Thus,
the no-washout solution (4.3.14) is globally asymptotically stable provided > cr
on .

60

Chapter 4

4.3.6

Transcritical bifurcation

A transcritical bifurcation occurs at the residence time value when

cr =

(1 R )
.
1 Kd

(4.3.25)

The value of cr represents the maximum residence time at which treatment process
fails. At lower residence times microorganisms are removed from the reactor at a
rate greater than their maximum growth rate, resulting in process failure. At the
residence times lower (higher) than the transcritical value, the washout (no-washout)
solution is the only stable solution. The critical value of the residence time is zero
when either R = 1 or = 0. Thus, increasing the recycle parameter R or decreasing the reactor parameter allows the reactor to operate at lower residence times.
Equation (4.3.25) is governed by Contois growth kinetics. It shows that the washout
condition does not depend upon the euent pollutant concentration (S0 ) for wastewater treatment processes. In vast contrast in which the treatment process is governed
by Monod kinetics, the washout condition depends upon the euent pollutant concentration (S0 ) [132]. The practical implication of this is that, for processes governed
by Monod kinetics, the wastewater that enters the reactor needs to be concentrated.

4.3.7

Residence time approximations near the washout point

When residence times is slightly higher than the washout point, the approximate
values of the microorganism concentration and the substrate concentration are given
by
(1 Kd )2
X =
+ O(2 ),

(1 R )

(4.3.26)

61

Chapter 4

S = 1
where =

(1 Kd )
+ O(2 ).

(4.3.27)

(1 R )
<< 1.
1 Kd

From equations (4.3.26 and 4.3.27), increasing leads to, respectively, an increase
and decrease in the microorganism and substrate concentrations. It is observe that
for xed , the increasing the value of expression (1 R ) leads to a decrease in
the microorganism concentration.

4.3.8

Steady-State diagrams

Steady-state diagrams showing the physical meaningful solutions for the variation
of substrate concentration (S ) and microorganism concentration (X ) as a function
of the residence time ( ) time are demonstrated in gures (4.2), (4.3) and (4.4).
The Maple package [173] was used to obtain the steady-state diagrams.
In section (4.3.2), it was shown that the no washout branch is only physically meaningful when > cr . These physical meaningful solutions (i.e 0 < S 1 and
X > 0) are plotted in gure (4.2). From this gure, it can be seen that for suciently low values of the residence time ( < cr ) the stable solution is the washout
solution (given by the solid lines S = 1 and X = 0) and the unstable solution is
the no-washout solution. For > cr , only the no-washout solution is stable (presented by the solid curve for both S and X ) and the washout solution is unstable
(presented by a dotted line for both S and X ).
Using these same parameter values as for gure (4.2), a general pattern of substrate
S and micro-organism X behaviour with respect to low value residence time is
shown gures (4.3 and 4.4). From these gures, it can be seen as the value of

Chapter 4

62

(1 R ) decreases to 0, the critical residence time (cr ) also decreases. This means
that the length of the residence time for where the washout solution is stable diminishes too.
In the following discussions we assume that the residence time is suciently high

(ie cr
<< ) so that the no-washout solution branch is the stable solution. Figure

(4.3) shows that for both xed residence time ( ) and eective recycle parameter
(R ), the performance of the reactor increases as the value of the reactor parameter
decreases (from curve (a) to (e)). It also shows that if the values of and R are
xed, the performance of the reactor improves as increases. In general, as the
value of the expression (1R ) decreases, the performance of the reactor improves.
That is there is eective removal of pollutant or substrate from the bioreactor.
Figure (4.4) shows that the microorganism concentration (X ) is an increasing function of the residence time ( ), except for the case where (1 R ) = 0. It also
shows that if the residence time is xed, then the microorganisms concentration
(X ) decreases as the value of (1 R ) increases (from curve (a) to (e)).
In the following section, we show that the steady state of microorganism concentration (X ) decreases towards zero when the residence time increases to innity (i.e.
). Thus, the microorganism concentration must increase to a maximum
before it decreases towards zero.

63

Dimensionless microorganism concentration ( X )

Chapter 4

0.1
0.08
0.06
0.04
0.02
0
0

10

Dimensionless residence time ( )

Dimensionless substrate concentration ( S )

(a)

0.1

0.01
0

10

Dimensionless residence time ( )

(b)

Figure 4.2: Dimensionless microorganism (a) substrate concentrations (b) as a function of dimensionless residence time. Parameter value: dimensionless decay rate
Kd = 0.014091, dimensionless yield coecient = 0.10194 and (1 R )= 1 .

64

Dimensionless substrate concentration ( S )

Chapter 4

a
b
c
d
e
f

0.1

0.01
0

10

Dimensionless residence time ( )

Figure 4.3: Dimensionless substrate concentration as a function of dimensionless

residence time. Parameter value: dimensionless decay rate Kd = 0.014091, dimensionless yield coecient = 0.10194 and (1 R )= 0 (a), 0.25 (b), 0.5( c), 0.75
(d) , 1 (e) and washout solution (f)

65

Dimensionless microorganism concentration ( X )

Chapter 4

a
b
c
d
e
f

0.8
0.6
0.4
0.2
0
0

10

Dimensionless residence time ( )

Figure 4.4: Dimensionless microorganism concentration as a function of dimensionless residence time. Parameter value: dimensionless decay rate Kd = 0.014091,
dimensionless yield coecient = 0.10194 and (1 R )= 0 (a), 0.25 (b), 0.5( c),
0.75 (d) , 1 (e) and washout solution (f)

66

Chapter 4

4.3.9

Large residence time approximations

At large residence times, the approximate value of the substrate concentration is

obtained by using Taylor series,
S

1
+ O( 2 ),

(1 Kd )

0 < Kd < 1.

(4.3.28)

Equation (4.3.28) demonstrates that the substrate concentration decreases to zero

at large value of the residence times. Interestingly, this equation also shows that
the substrate concentration (S ) is independent of the recycle parameter R and the
reactor parameter . This means that equation (4.3.28) shows that, for processes
controlled by Contois kinetics, the substrate concentration can be decreased to any
desired level by operating the reactor at a suciently large the residence time. This
is not the case for processes controlled by Monod kinetics whereby the substrate
concentration has a limiting value,

Kd
, [132].
1 + Kd

At large residence times, the approximate value of the microorganism concentration

is given by,
X

1
+ O( 2 ),

Kd

0 < Kd < 1.

(4.3.29)

Equation (4.3.29) shows that, for large values of the residence time, the microorganism concentration (X ) is a decreasing function the residence time . Thus, in
the limiting case, that is where , the microorganism concentration tends to
0. The physical meaning of this is that the microorganism concentration becomes
depleted due to the limited availability of substrate concentration.

67

Chapter 4

4.3.10

Maximizing the microorganism concentration

In some bioreactor processes it is important to maximise the microorganism concentration, for instance if the bioreactor is being used to grow the microorganism. The
microorganism concentration is zero (i.e. X = 0) along the washout solution and
in the limit that the residence time increase to innity. Thus, there is a maximum
value of microorganism concentration (X ). Dierentiating the expression of the
microorganism concentration from equation (4.3.14), with respect to the resident
time, we nd that the maximum value occur when,

max

4.3.11

(1 R )
=
+
1 Kd

Kd (1 R )
Kd (1 Kd )

Discussion

The performance of a bioreactor processing industrial wastewaters can be characterized by using a number of criterions such as the specic utilization, the eciency
and the rate of waste treatment. Here, we will consider the process eciency of
bioreactor characteristic along the no-washout branch, that is when
>

(1 R )
.
1 Kd

(4.3.30)

The treatment eciency is the percentage of substrate that has been removed by
the reactor. It is dened by
(
= 100

S0 S
S0

)
.

(4.3.31)

and therefore, the eciency is given by

= 100(1 S ).

(4.3.32)

68

Chapter 4

From equation (4.3.32), it is clear that when S < 1 along the no-washout branch,
the eciency is positive ( > 0). Substituting for S from equation (4.3.14) into
equation (4.3.32), we obtain the eciency along the no-washout branch as
(
= 100 1

(R 1) + (1 Kd ) +

(
= 100 1

0
0 + ( cr )

)
.

(4.3.33)
where 0 =

, 0 R 1 and 0 < Kd < 1.

1 Kd

The eciency increases when

-

becomes large or

cr becomes small. This is when (1 R ) is small.

The eciency (equation (4.3.33)) as a function of the residence time ( ) are shown
in gure (4.5). This gure shows that at xed values for both the residence time and
the parameter reactor , the eciency of the reactor increases as the value of the
recycle parameter R increases. The gure also shows that when the value of the
parameters and R are given, then the eciency characteristic of the bioreactor
improves as the residence time increases. Furthermore, as the value of the expression
((1 R )) decreases, the eciency of the bioreactor improves.
When 0 < cr < 1. the values of the reactor parameter and the recycle parameter R play a signicant role in the eciency of waste reduction. By expanding
about cr in equation (4.3.25), we nd that the eciency along the no-washout
branch becomes
= 100

(1 Kd )

2
( cr
) + O(( cr
) ),

(4.3.34)

69

Chapter 4

At large residence times, the eciency becomes

1
= 100(1
) + O( 2 ),

(1 Kd )

(4.3.35)

As the residence time approaches innity, the eciency of the process approaches
100 % and is independent of the inuent pollutant concentration. In contrast for
processes controlled by Monod kinetics, the maximum eciency is bounded below
100 and depends upon the inuent pollutant concentration (S0 ) [132].
Finally, at high values of the residence time processes eciency is independent of
the reactor parameter () and the recycle parameter (R ). To study the eciency as
a function of the parameter (1 R ), table (4.1) shows the residence time required
to achieve eciencies of 99%, 99.9% and 99.99% for varies value of (1 R ). It
shows that the residence time to achieve a given eciency increases as the value of
(1 R ) increases. However, the dependence upon the parameter (1 R ) is very
week.
Table 4.1: The residence time in single reactor to achieve an eciency of 99.9%,
99.99% and 99.999% for varies value of (1 R ) = 0, 0.25, 0.5, 0.75 and 1.
(1 R )

Eciency

0.25

0.5

0.75

percentage increase.

99%

11.01

11.28

11.55

11.82

12.10

9.9%

99.9% 111.104

111.37

111.64

111.92

112.19

0.977%

1112.58 1112.86

1113.13

0.098%

99.99%

1112.04 1112.31

70

Chapter 4

100

a
b
c
d
e

Dimensionless Efficiency

80

60

40

20

0
0

Dimensionless residence time

Figure 4.5: The dimensionless eciency ( ) in an ideal reactor as a function of

dimensionless residence time. The parameter values: dimensionless decay rate Kd =
0.014091, dimensionless yield coecient = 0.10194 and (1 R )= 0 (e), 0.25
(d), 0.5( c), 0.75 (b) and 1 (a)

71

Chapter 4

4.4

Conclusion

We have investigated a bioreactor model for the interaction between a microorganism and a rate-controlling substrate. We have modeled the specic growth rate
using the Contois model. An examination of the literature reveals that there are
a signicant number of processes where it is more appropriate to use this growthrate expression than the more commonly used Monod expression. We considered a
generalized reactor model in which the idealized membrane reactor and well-stirred
ow reactor with or without recycle are limiting cases. The steady-state solutions
and the stability were found as a function of the residence time. A transcritical
bifurcation occur when,
(1 R )
=
.
1 Kd

(4.4.36)

The washout solution is only stable solution when the residence times is lower than
the transcritical value and the no-washout solution is the only stable solution when
the residence times is higher than transcritical value. The washout and no-washout
solutions are globally asymptotically stable if it is locally stable.
We showed that the substrate concentration decreases if either the reactor parameter () decreases or the recycle parameter (R ) increases. At large values for the
residence times the substrate concentration is independent of the reactor parameter
() and the recycle parameter (R ). One way to characterize the performance of
a bioreactors is the treatment eciency. Increasing the residence time increase the
treatment eciency.

Chapter 5
Analysis of continuous ow
bioreactor models with recycle
around each reactor

5.1

Introduction

In previous chapter, the treatment of industrial wastewater in a single reactor with

recycle was studied. We provided more details of the assumptions of model parameters and the analysis of the single reactor, such as the periodic solution, invariant
region and global stability are presented. The Contois growth kinetics was used to
model the degradation of the biodegradable organic material. We now extend our
model to consider the cascade of n reactors with recycle around each reactor.
Cascade reactors are frequently used in the industrial treatment of wastewaters. The
main advantages of using a cascade is an increased stability for the treatment plant

72

Chapter 5

73

and also an improved degree of degradation by an adopting recycle. A number of

investigators have studied the behavior of the reactor cascade and an overview of cascade reactors was presented in chapter (2). The only previous mathematical study
of a reactor cascade using the Contois model was that of Nelson and Holder [128]
who studied the behavior of a reactor cascade of n reactors without recycle. They
found that at high total residence times the euent concentration is proportional
to

1
,
(t )n

where t is the total residence time through a cascade containing n reactors.

To our knowledge a similar study based upon other kinetics models has not been
published. It is likely that similar results hold for other kinetics models such as
the Monod model. However it is not possible to obtain large total residence time
asymptotic for the Monod model. This is a mathematical advantage of the study
considering the Contois Model.
In this chapter, we will study the behavior of a cascade with n reactors with recycle.
We consider the case in which the output from a settling unit placed on the ith
reactor enters the inuent stream for the ith reactor. We then nd the steady-state
solutions, determine their stability and obtain asymptotic solutions in the limit of
high total residence times. Hence, Steady state analysis is used to investigate how
recycle aects the performance of the reactor cascade. Our major concern is the
improvement in performance that can be achieved by the use of either one, two or
n recycling units. Even with one settling unit various congurations can be utilized. For instance, in a 3-reactor cascade, the settling unit can be placed after
the rst, second or third reactor. It is noted that the geometry of the ith reactor
cascade with recycle is demonstrated in gure (5.1). A major part of this chapter

74

Chapter 5

has been published as a journal article [174] and has been published as conference
papers [175, 176].
Reactor i

FSi 1 , FX i 1 (1  Ri 1 (1  Ci 1 ))

Settling unit i

FX i (1  Ri ), FSi

FX i (1  Ri (1  Ci )) , FSi

FX i RiCi

Figure 5.1: A schematic diagram of the ow around reactor i in a n reactor cascade.

5.2

Model equations and assumptions

In this study, using the Contois expression as the growth rate, a system of nonlinear ordinary dierential equations are used to model the biodegradation of organic
substances in wastewaters for an n reactor cascade. The dependent variables are
the concentration of organic substrate and the degrader (microorganism) while the
independent variable is time. A more detailed model assumptions was dened in
chapter (4).
The dimensional and dimensionless forms of our new model for an n reactor cascade
using the Contois growth rate are given in sections (5.2.1) and (5.2.2), respectively.

75

Chapter 5

5.2.1

The dimensional model

Given that i denotes the ith reactor in a n reactor cascade with recycle around ith
reactor, the model equations are,
dSi
(Si , Xi )
= F (Si1 Si ) Xi
,
(5.2.1)
dt

dXi
V
= F [i1 (1 + Ri1 (1 Ci1 ))Xi1 i Xi ] + Xi (Si , Xi )
dt
V

Kd Xi + F Ri i (Ci 1)Xi .

(5.2.2)

For i = 1, ..., n.

Specic growth rate (Contois model)

(
(Si , Xi ) = max

Si
Ks Xi + Si

)
.

(5.2.3)

The total residence time

t =

nV
.
F

(5.2.4)

Note that by convention 0 = R0 = C0 = 0. The parameter i is dened in chapter (4). All parameters in the model are strictly non-negative. For more details of
the assumptions and denition of model parameters, we refer readers to chapter (4)
and the appendix (D) respectively.
Based on the discussion in chapter (4), the value of the maximum concentrating
factor, denoted by Ci,max , is given by
Ci,max = (1 +

1
), 1 i n.
Ri

We refer to the case C = Ci,max as perfect recycle.

(5.2.5)

Chapter 5

5.2.2

76

The dimensionless model

The system of dierential equations, (5.2.1-5.2.2) can be written in dimensionless form by introducing dimensionless variables for the substrate concentration
[Si = Si /S0 ], the microorganism concentration [X = Ks Xi /S0 ], and time [t = max t].
We obtain
dSi
n
Si Xi

=
(S

S
)

,
i
dt
t i1
(Si + Xi )

(5.2.6)

n[i1 (1 Ri1
)Xi1
i Xi ]
dXi
Si Xi
ni Ri Xi
=
+
+
Kd Xi . (5.2.7)
dt
t
(Si + Xi )
t

All dimensionless parameters in the model are strictly non-negative. For more details of the assumptions and denition of model parameters, we refer readers to the
chapters (4) and the appendix (D) respectively.
In section (5.3.3) we show that the global washout solution is globally stable when
Kd 1. Thus, the process failure occurs for any value of the residence time when
Kd 1 consequently the condition 0 < Kd < 1 is required.
The maximum value of Ri is obtained by using the maximum value of the concen
tration factor Ci,max and is given by Ri,max
= 1. The cases (Ri = 0), (0 < Ri < 1)

and (Ri = 1) represent no recycle around the ith reactor, imperfect recycle around
the ith reactor and perfect recycle around the ith reactor. A feature of our dimensionless scheme is that there is a one to one relationship between our dimensionless
variable and their dimensional counterparts. Hence, we write often residence time,
rather than dimensionless residence time.

77

Chapter 5

5.3

Mathematical Analysis

In this section we analysis the ow-reactor with recycle (0 Ri 1) for the cases
of n reactor cascade. In section (5.3.1), an invariant region is determined. In section (5.3.2) the steady state solutions branches are found and the conditions for the
no-washout solution branch to be physically meaningful are stated. In section (5.3.3)
the stability of the steady-state solutions are determined. In section (5.3.4) the
transcritical bifurcation points are presented. In all our analysis we assume that
0 < Kd < 1.
In our calculations we use parameter values for the anaerobic digestion of ice-cream
wastewater [39]. These are: = 0.2116 (gVSS)(g COD)1 , max = 0.9297 (day1 ),
Kd =0.0131 (day1 ) and Ks =0.4818 (COD)(g VSS)1 . These lead to = 0.1019
and Kd = 0.0141.

5.3.1

The positive and bounded solution

In this section our aim is to establish the basic properties of invariance for the

region = {(S1 , X1 , S2 , X2 , ..., Sn , Xn ) R2n

+ , (Si , Xi ) 0, i = 1, ..., n.}.

On the boundary of , if Xi = 0 then

)Xi1
ni1 (1 Ri1
dXi
=
,
dt
t

(5.3.8)

= 0 then the plane

> 0 then the vector eld of Xi point inside and if Xi1
If Xi1

Xi = 0 is invariant. If S1 = 0 then
dS1
n
=
,
dt
t

(5.3.9)

78

Chapter 5

then the vector eld of S1 point inside . Now, if Si = 0 for i = 2, ..., n then
dSi
Si1
=
,
dt
t

(5.3.10)

If Si1
> 0 then the vector eld of Si point inside and if Si1
= 0 then the plane

Si = 0 is invariant. We conclude that is positively invariant. This means that

if the initial concentrations are in then the solution of system (5.2.6) and (5.2.7)
remain in there after for all time t > 0.
The boundness of Si and Xi (i = 1, ..., n) will be proved by mathematical induction.
When i=1, we have that,
dS1
n
S1 X1

=
(1

S
)

,
1
dt
t
(S1 + X1 )

(5.3.11)

dS1
n
(1 S1 ),

dt
t

(5.3.12)

By solving the equality in (5.3.12), we obtain,

S1 1 + C1 exp[

nt
]

(5.3.13)

where C1 is constant and hence that S1 is bounded.

Assume that Sn1

is bounded, we now need to prove that Sn is also bounded. We

have that,
n
Sn Xn
dSn

,
= (Sn1 Sn )
dt
t
(Sn + Xn )

(5.3.14)

n
dSn

(S
Sn ),
dt
t n1

(5.3.15)

dSn
n
(M Sn ).

dt
t

(5.3.16)

(t) is bounded by assumption, it follows from

(t). Since Sn1
where M = max Sn1

solving the equality (5.3.16) that,

Sn C2 + C3 exp[

nt
].

(5.3.17)

79

Chapter 5

where C2 and C3 are constant. Therefore, Sn is also bounded. Hence by mathematical induction, Si is bounded for all i = 1, .., n as required.
Let us now prove this fact for Xi (i = 1, ..., n), using again mathematical induction.
When i=1, we have that,
S1 X1
dX1
n
n

X
=

+ 1 R1 X1 Kd X1
1 1

dt
t
(S1 + X1 ) t

(5.3.18)

S1 X1
dX1
n

+
)X

(1

R
.
1
1
1
dt
t
(S1 + X1 )

(5.3.19)

We have proved that S1 is bounded, then

S1 X1
(S1 +X1 )

is also bounded. Hence

dX1
n
1 (1 R1 )X1 + L,

dt
t

(5.3.20)

dX1
L HX1 ,
dt

(5.3.21)

S1 X1
n
where L = max
and H = 1 (1 R1 ) 0. By solving the equality

(S1 + X1 )
t
in (5.3.21), we obtain,
X1 C4 + C5 exp[Ht ]

(5.3.22)

where C4 and C5 are constant. Assume that Xn1

is bounded, we now need to prove

that Xn is bounded. Now,

n
n
n
Sn Xn
dXn

+ n Rn Xn Kd Xn .
= n1 (1 Rn1 )Xn1 n Xn +

dt
t
t
(Sn + Xn ) t
(5.3.23)

and
We know that Xn1

Sn Xn
are bounded. Therefore,
(Sn + Xn )
dXn
n
L1 n (1 Rn )Xn ,

dt
t

(5.3.24)

dXn
L1 H1 Xn
dt

(5.3.25)

80

Chapter 5

where L1 =

max[Xn1
;

Si+1
Xn
n
] and H1 = n (1 Rn ). Therefore, Xn is also

(Sn + Xn )
t

bounded. Hence by mathematical induction, Xi is bounded for all i = 1, .., n as

required.

5.3.2

Steady state solutions

5.3.2.1

A single reactor

The steady-state solutions in the single reactor, (i = 1, n = 1) in (5.2.6 and 5.2.7),

are given by,
The washout branch is
(S1 , X1 ) = (1, 0).

(5.3.26)

The no-washout branch is

(
(S1 , X1 )

(1 S1 )
,
1 (1 Kd ) + 1 (R1 1) + (1 Kd + 1 (1 R1 ))

)
.

(5.3.27)

The no-washout branch is physically meaningful when,

t > cr =

1 (1 R1 )
,
(1 Kd )

The washout branch is stable when the no-washout branch is not physically meaningful (t < cr =

1 (1R1 )
).
(1Kd )

The transcritical bifurcation occurs when

t = cr =

5.3.2.2

1 (1 R1 )
.
(1 Kd )

A two reactor cascade

Here, we will rst consider the mathematical analysis of a two reactor cascade model.
It is found that there are three steady state solution branches of the system of

Chapter 5

81

equations (5.2.6 and 5.2.7) which can be described in the following way:
In the rst reactor two steady state solutions can occur, one being the washout
solution and the other being a no-washout solution. When the washout solution
occurs in the rst reactor, there are two possible outcomes in the second reactor.
These outcomes are either washout (global washout branch) and no-washout (partial
washout branch) solutions.
On the other hand, when the no-washout solution occurs in the rst reactor then
there are two possible cases that should be considered in the second reactor which
correspond to either perfect and imperfect recycle around the rst reactor. For
the case of the perfect recycle around the rst reactor (R1 = 1), two steady state
solutions are possible in the second reactor. These are the no-washout solution or
washout solution. In this case, washout solution corresponds to the absent of the
microorganism concentration in the second reactor and the euent concentration
leaving the rst reactor is equal to that leaving the second reactor. When, there is
imperfect recycle around the rst reactor (R1 = 1), then only one outcome in the
second reactor which corresponds to the no-washout solution.
Along the washout solution the euent concentration is identical to the inuent
concentration; ie no cleaning has occurred. The partial washout branch solution is
not desirable from a practical perspective due to process failure in the rst reactor.
Thus, the rst reactor does not contribute to the performance of the reactor cascade.
The no-washout solution is of main interest due to the process does not fail in both
reactors. Once should noted that equation (5.2.6) has washout solution when the
euent leaving the i th reactor is less/or equal to that leaving the (i+1) th reactor.

82

Chapter 5

Thus, along the no-washout solution, the euent leaving the i th reactor is always
higher than that leaving the (i+1) th reactor.
These three steady state solutions are mathematically described below:
(a)- The global washout branch is
(S1 , X1 , S2 , X2 ) = (1, 0, 1, 0) .

(5.3.28)

(b)- The partial washout branch is

(S1 , X1 , S2 , X2 )

(
= 1, 0,

n
n (1 S2 )
,
t (1 Kd ) + n2 (R2 1) + n t Kd + n2 (1 R2 )

)
.

(5.3.29)
The partial washout branch is physically meaningful when,
t

n2 (1 R2 )
>
.
(1 Kd )

(5.3.30)

(c)- The no-washout branch has two cases to consider. These are: R1 = 1 and
R1 = 1.
For the case with perfect recycle in the rst reactor (R1 = 1) the steady state
solutions for the second reactor of the reactor cascade are considerably simplied.
Substituting R1 = 1 in equation (5.2.7) when i = 2, we obtain,
(
X2

2 t (R2 1)
S
Kd + 2
n
(S2 + X2 )

)
= 0.

(5.3.31)

It can be seen from equation (5.3.31) that, there are two steady state solutions,
(S2 , X2 ) = (S1 , 0),
(S2 , X2 ) = (

(5.3.32)

S1
(S1 S2 )
,
). (5.3.33)
+ 2 (R2 1) + (t /n)(1 Kd ) 2 (1 R2 ) + Kd (t /n)

83

Chapter 5

The steady state solution (5.3.32) is not desirable as there is washout solution in
the second reactor. The steady state solution (5.3.33) is realistic when
t >

n2 (1 R2 )
.
1 Kd

(5.3.34)

The case of perfect recycle in both reactors is of interest. In this case, equation (5.3.33) simplies to
(
(S2 , X2 )

S1
S1 (1 Kd )
,
+ (t /n)(1 Kd ) [ + (t /n)(1 Kd )] Kd

)
.

(5.3.35)

Providing the washout solution does not occur in the rst reactor and R1 = 1, the
steady state solutions in the second reactor is obtained by setting the system of
equations (5.2.6, 5.2.7) when i = 2, equal to zero for i = 2. We have that,
f (S2 , X2 ) =

n
S2 X2

(S

S
)

= 0,
2
t 1
(S2 + X2 )

g(S2 , X2 ) =

n[(1 R1 )1 X1 2 X2 ]
S2 X2
n2 R2 X2
+
+
Kd X2 = 0.
t
(S2 + X2 )
t

(5.3.36)

(5.3.37)
where S1 and X1 are the steady state solution found in equation (5.3.27). To obtain
the steady state solution from equations (5.2.6, 5.2.7), we consider the following,
f (S2 , X2 ) +

g(S2 , X2 )
= 0,

(5.3.38)

After some algebraic manipulations, it is found that

S2 = A2 X2 B2 ,

(5.3.39)

where the coecient A2 and B2 are dened by

A2 = S1 +

(1 R1 )1 X1
,

(5.3.40)

84

Chapter 5

B2 =

2 (1 R2 ) + (t /n)Kd
.

(5.3.41)

Since, 0 R2 1 and 0 R1 1 then A2 > 0 and B2 > 0. Now, we have,

S2 = A2 X2 B2 X2 =
Substituting X2 =

A2 S2
B2

A2 S2
,
B2

(5.3.42)

in equation (5.3.36) gives,

Q(S2 ) = aS22 + bS2 + c = 0,

(5.3.43)

where
a = (t /n) (B2 1),
b = (B2 1)S1 A2 [ + (t /n)],
c = A2 S1 .
The solutions of equation (5.3.43) are
b
S2 =

2a

We dene that S2,

=

b
2a

b2 4ac
2a

b2 4ac
.
2a

and S2,+
=

b
2a

(5.3.44)
b2 4ac
.
2a

Note that the constant coecient of the quadratic function (5.3.43) is strictly positive. Thus, we consider the sign of the coecient of S22 , The coecient of S22
is,
a =(t /n) (B2 1) = (1 Kd )(t /n) + (R2 2 ) + .

(5.3.45)

which may be positive or negative. This leads to study three cases for the coecient (a) to determine the sign of the roots of equation (5.3.43).
1) The case a < 0.

85

Chapter 5

Since (t /n) (B2 1) < 0 and Q(0) = A2 S1 > 0, the quadratic func
tion (5.3.43) has a positive and a negative root. When a < 0, we have S2,
>

0 > S2,+
. Thus S2,
is positive.

We will show that 0 < S2,

< A2 which will leads to X2 > 0.

Since A2 > 0, then substituting S2 = A2 into the quadratic function (5.3.43) gives,
Q(A2 ) = B2 A2 [S1 A2 ],
= (1 R1 )B2 A2 1 X1

Hence, Q(A2 ) < 0 for 0 R1 < 1. Thus, we conclude that the positive root, S2,
,

satises 0 < S2,

< A2 and hence X2 =

A2 S2,
B2

> 0.

2) The case a > 0.

We have the following properties,
(i) a = (t /n) (B2 1) > 0 (t /n) > (B2 1).
(ii) Q(0) = A2 S1 > 0.
(iii) b = (B2 1)S1 A2 [ + (t /n)],
< (t /n)S1 A2 [ + (t /n)]
because
(t /n) > (B2 1),
[
]
(1 R1 )1 X1
(1 R1 )1 X1

= (t /n)S1 S1 +
[
+
(
/n)],
because
A
=
S
+
,
2
t
1

)
(

= S1 1 + t (1 R1 ) 1 X1 < 0.
n
Properties (i, ii and iii) demonstrate that the quadratic function (5.3.43) has either
two positive roots or two imaginary roots. Since, Q(A2 ) = B2 A2 1 X1 (1 R1 ) < 0
and A2 > 0, the quadratic function (5.3.43) has two positive roots. One root is less

. Thus S2,
< S2,+
than A2 and others is bigger than A2 . When a > 0, we have S2,

< A2 and hence X2 > 0.

satises 0 < S2,

86

Chapter 5

Hence, in both cases a < 0 and a > 0, we conclude that the solution of the quadratic

function (5.3.43) corresponding to the choice of the negative square root (S2,
)

is realistic solution as both the substrate concentration and the microorganisms

concentration are positive. The choice of the positive square root (S2,+
) is not

realistic solution.
We conclude that the no washout solution is given by
(S1 , X1 , S2 , X2 )

(
)
A2 S2,

c1 , X
c1 , S ,
= S
.
2,
B2

(5.3.46)

where
A2 =

S1

(1 R1 )X1
+
,

a = t (B2 1) ,
n

(1 R2 ) +
B2 =

b = (B2

Kd t
n

1) S1

)
(
t

A2 +
,
n

c = A2 S1 ,
c1 =
S

n
,
t (1 Kd ) + n(R1 1) +

c1 =
X

n (1 S1 )
.
t Kd + n(1 R1 )

3) The case a = 0.
In this case equation (5.3.43) becomes,
Q(S2 ) = { (B2 1)S1 A2 [ + (t /n)]}S2 + A2 S1 .

(5.3.47)

The solution of this equation is,

S2 =

A2 S1
(B2 1)S1 A2 [ + (t /n)]

We have,
(t /n) (B2 1) = 0.

(5.3.48)

87

Chapter 5

This gives,

B2 =
By substituting, B2 =

t
n

t
+1
n

+ 1, into equation (5.3.48), we obtain,

A2 S1
,
(t /n)S1 A2 [ + (t /n)]
A2
=
< A2 ,

(1 R1 ) ( + t ) 1 X1
1+
nS1 2

S2 =

as 0 R1 < 1.

Thus, we conclude that the root, S2 , satises 0 < S2 < A2 and hence X2 =

A2 S2
B2

>

0. If t n2 (1 R2 ) / (1 Kd ) then S2 = S1 and washout occurs in the second

reactor.

5.3.2.3

N reactor cascade

Equations (5.3.33 and 5.3.46) give the substrate and microorganisms concentrations
in the rst two reactors of a n-reactor cascade. These expressions can be extended
for any value of n. For instance, the steady state solution for the third reactor
of cascade containing the three reactors is obtaining by an increasing all relevant
indices of equations (5.3.33 and 5.3.46) by one which given by,
(S3 , X3 ) = (

S2
(S2 S3 )
,
), for R2 = 1,
+ 3 (R3 1) + (t /n)(1 Kd ) 3 (1 R3 ) + Kd (t /n)
(5.3.49)

(S3 , X3 )

S3,

(
)
A3 S3,

= S3, ,
, for R2 = 1,
B3

b23 4a3 c3
b3
=

,
2a3
2a3

(1 R2 )X2
A3 = S2 +
,

a3 = t (B3 1) ,
n

(5.3.50)

(1 R3 ) +
B3 =

b3 = (B3

Kd t
n

1) S2

(
)
t

A3 +
,
n

88

Chapter 5

c3 = A3 S2 .

5.3.3

Stability of the steady state solutions

The study of the stability of the solution is important because only the stable solution can be observed experimentally. The Jacobian matrix for the n-reactor cascade,
the system of equations (5.2.6, 5.2.7), is given by

0
J1 0 0

B J
0 0
2
2
M =

0 B J
0

3
3

... ... ...

0

. . .

where 0 represents the zero matrix of size 2 2. The 2 2 matrix blocks Bi and Ji
are dened by,

n
t

Ji =

2
Xi

(Xi +Si )2

Xi

(Xi +Si )2

Si

(Xi +Si )2
ni (Ri 1)t Kd
t

2
Si

(Xi +Si )2

t
Bi =

ni1 (1Ri1
t

where i = 1, ..., n.

Evaluating the Jacobian matrix (M) for a n-reactor cascade at the global washout
gives

n
t

Ji =

ni (Ri 1)
t
n

t
Bi =

+ 1 Kd

ni1 (1Ri1
t

89

Chapter 5

The eigenvalues of the matrix M are given by

i =

ni (Ri 1)
+ 1 Kd ,
t

n
< 0,
t

2ni =

(i = 1, ..., 2n/2).

The global washout branch is stable when

(
t

< min

)
ni (1 Ri )
,
1 Kd

(i = 1, ..., 2n/2).

Evaluating the Jacobian matrix, J, for a cascade of two reactors at the partial
washout steady state solution gives

J(1, 0, S2 , X2 ) =

n
t

n1 (R1 1)
t

+ 1 Kd
n
t

2
X2

(X2 +S2 )2

2
X2

(X2 +S2 )2

n1 (1R1 )
t

2
S2

(X2 +S2 )2

X S

2 2
(X +S
)2
2

(5.3.51)
Here, we have used the result that along the no-washout branch
t Kd + n2 (1 R2 )
S2
n1 (1 R1 )X1
=

.
t
(X2 + S2 )
t X2
After elementary row operations are performed on the Jacobian matrix J, (5.3.51),
we nd that the eigenvalues of this matrix satisfy the relationship
(

and

n
t

n1 (R1 1)
+ 1 Kd

2
X2
(X2 +S2 )2
2

X2
(X2 +S2 )2

)(

2
S2
(X2 +S2 )2

n (1R )X
1 X 1 1
t
2

X2 S2
(X2 +S2 )2

)
= 0,





 = 0.



(5.3.52)

(5.3.53)

The eigenvalues in equation (5.3.52) are:

1 =

n1 (R1 1)
+ 1 Kd ,
t

2 =

n
.
t

(5.3.54)

90

Chapter 5

Here 2 is always negative. The eigenvalue (1 ) is negative when

t

n1 (1 R1 )
<
.
1 Kd

Physically meaningful solutions only occur when t >

n2 (1R2 )
1Kd

and 0 < Kd < 1.

For physically meaningful solutions, the determinant and trace of matrix (5.3.53)
are, respectively,
(

)(
)
n1 (1 R1 )X1
X2 S2
n
X22
det(J) =
+
+
t X2
(X2 + S2 )2
t (X2 + S2 )2
)(
)
(
S22
X22
> 0,
+
(X2 + S2 )2
(X2 + S2 )2
tr(J) =

n1 (1 R1 )X1
X2 S2
n
X22

< 0.
t X2
(X2 + S2 )2 t (X2 + S2 )2

(5.3.55)
(5.3.56)

It follows from equations (5.3.55 and 5.3.56) that the real part of the eigenvalues
in matrix (5.3.53) are negative. Hence, the partial washout steady state solution is
stable when
n2 (1 R2 )
n1 (1 R1 )

<

<
.
t
1 Kd
1 Kd

(5.3.57)

Note that the partial washout branch is never stable if R1 R2 . Equation (5.3.57)
establishes conditions under which the partial washout branch is stable.

91

Chapter 5

The eigenvalues of the matrix (M) satised the following equation,



J (1, 1)
J1 (1, 2)
0
0
0
0
0
 1


 J (2, 1)
J1 (2, 2)
0
0
0
0
0
 1



n
0
J2 (1, 1)
J2 (1, 2)
0
0
0

t


n1 (1R1 )

0
J2 (2, 1)
J2 (2, 2)
0
0
0

t



n
0
0
0
J3 (1, 1)
J3 (1, 2)
0

t


n2 (1R2 )

0
0
0
J3 (2, 1)
J3 (2, 2) 0

t


..
..
..
..
..
..

.
.
.
.
.
.
0













 = 0.









.. 
.

(5.3.58)
Expanding along the rst row, we obtain,


J (2, 2)
0
0
0
0
0
 1



0
J2 (1, 1)
J2 (1, 2)
0
0
0



 n1 (1R1 )
J2 (2, 1)
J2 (2, 2)
0
0
0

t
(J1 (1, 1))

n
0
0
J3 (1, 1)
J3 (1, 2)
0

t


n2 (1R2 )

0
0
J3 (2, 1)
J3 (2, 2) 0

t


..
..
..
..
..
..

.
.
.
.
.
.



J (2, 1)
0
0
0
0
0
 1


 n
J2 (1, 1)
J2 (1, 2)
0
0
0
 t


 0
J2 (2, 1)
J2 (2, 2)
0
0
0

J1 (1, 2)
 0
n
0
J3 (1, 1)
J3 (1, 2)
0

t


n2 (1R2 )

J3 (2, 1)
J3 (2, 2) 0
0
 0
t


..
..
..
..
..

.
.
.
.
.
 0

..



















.. 
.

92

Chapter 5

Expanding the both determinants in the previous equation along the rst row, we
obtain,
0 = [(J1 (1, 1) )(J1 (2, 2) ) J1 (1, 2)J1 (2, 1)]


J (1, 1)
J2 (1, 2)
0
0
0
 2


 J (2, 1)
J2 (2, 2)
0
0
0
 2


n

0
J3 (1, 1)
J3 (1, 2)
0
t


n2 (1R2 )

J3 (2, 1)
J3 (2, 2) 0
0

t


...
...
...
...
...



(5.3.59)









 .






. . .


(5.3.60)

We note that equation (5.3.59) is the elements for the Jacobian block J1 and the
structure of the determinant in (5.3.60) is similar with the structure in determinant (5.3.58) except all relevant indices have increased by one. It follows by induction that the eigenvalues of the Jacobian matrix M are those of the constituent
blocks Ji , i = 1, ..., n.
Along the no-washout branch, the Jacobian block Ji can be re-written as,

n
t

Ji =

2
Xi

(Xi +Si )2
2

Xi
(Xi +Si )2

2
Si

(Xi +Si )2

)X
ni1 (1Ri1
i1
t Xi

Xi Si
(Xi +Si )2

We have used the fact that along the no-washout branch,

)Xi1
ni1 (1 Ri1
t Kd + nn (Ri 1)
Si

=
.
t
(Xi + Si )
t Xi

(5.3.61)

The eigenvalues of the matrix Ji have negative real parts when the determinant and
trace are positive and negative respectively.
(
det(Ji ) =

)i1 Xi1
n(1 Ri1
Xi Si
+
(Xi + Si )2
t Xi

)(

n
Xi
+
t (Xi + Si )2
2

)
+

93

Chapter 5

)(
)
2
2
Xi
Si
> 0,
(5.3.62)
(Xi + Si )2
(Xi + Si )2
)
) (
(

2

n(1

R
)
X
Xi Si
n
X
i1 i1 i1

+ i 2 < 0.
tr(Ji ) =
2

(Xi + Si )
t Xi
t
(Xi + Si )
(5.3.63)
As, the trace of the jacobian (Ji ) is negative and the determinant of the Jacobian (Ji )
is positive, the real part of the eigenvalues are negative. Consequently, provided that
washout does not occur in the rst reactor, the steady-state solution of the cascade
is locally stable.
Theorem 5.3.1 If Kd 1 then the washout solution is globally asymptotically
stable on .
Proof. To prove global stability of the globally washout solution whenever Kd 1,
we consider the argument from [170]. By assuming that y = X1 + ... + Xn , we
obtain,

n
n1
n

dy
n
n
Si

[R

1]X
+

[1

R
]X
+
X
[
Kd ],
i
i
i
i
i
i
i

dt
t i=1
t i=1
X i + Si
i=1

y(1

Kd )

nn (Rn 1)Xn
Si Xi
Xi ,
+
, because

t
X i + Si

y(1 Kd ).
Equation (5.3.64) shows that

(5.3.64)
dy
dt

< 0 when Kd 1. Since, the solution Xi , i =

1, 2, ..., n are a positive and bounded (see section (5.3.1)). Thus y is a decreasing
function of t , reaching zero when t increases to innity. It follows that X1 , ..., Xn
are decreasing functions of t reaching zero when t increases to innity. Thus,

94

Chapter 5

the scaled microorganism concentrations X1 , ..., Xn approaches zero and the global
washout solution is globally stable when Kd 1.

5.3.4

Transcritical bifurcations

In this section, we nd the value of the total residence time at which transcritical
bifurcations occur. We note that the behavior of the ith reactor is independent of
the (i + 1)th reactor. In the ith reactor, the critical value of the total residence time
is given by
i =

i,cr
(1 Ri )
=
.
n
(1 Kd )

(5.3.65)

We consider in detail a cascade contains two reactors . When R2 > R1 , we have the
following possibilities,
t <

22 (1 R2 )
(1 Kd )

t
1 (1 R1 )
2 (1 R2 )
<
<
(1 Kd )
2
(1 Kd )
t
1 (1 R1 )
>
2
(1 Kd )

Global washout branch stable,

(5.3.66)

Partial washout branch stable,

(5.3.67)

No-washout branch stable.

(5.3.68)

Hence, there are two transcritical bifurcations: t,1

= 21 (1 R1 )/(1 Kd ) and

= 22 (1 R2 )/(1 Kd ).
t,2

When R1 R2 , we have the following possibilities,

t <

21 (1 R1 )
(1 Kd )

t
1 (1 R1 )
>
2
(1 Kd )

Global washout branch stable,

No-washout branch stable.

(5.3.69)
(5.3.70)

In this case there is one transcritical bifurcation: t = 21 (1 R1 )/(1 Kd ). The

value of the residence time at the transcritical bifurcation is the highest residence

Chapter 5

95

time at which the treatment process fails. For the three reactor cascade, the analysis
of the transcritical bifurcations generalises in the obvious way.

5.4

Discussion

In this section, we discuss the steady-state diagrams for the euent concentration
leaving the cascade. In particular, we discuss how recycle aects the performance
of the reactor cascade as a function of the total residence time inside the cascade.
In section (5.4.1), we discuss steady-state diagrams for the euent concentration
leaving the single reactor and a cascade of two reactors with either perfect recycle
or no recycle.
In section (5.4.2), we establish that the euent concentration leaving a n-reactor
cascade at any total residence time is minimized by setting Rn = 1 (perfect recycle
around the nal reactor). In section (5.4.3), we investigate how the use of recycle
around the rst reactor aects the euent concentration leaving a cascade of two
reactors.
Further, in section (5.4.4) we analysis the performance of a cascade of two reactors
as a function of the total residence time inside the cascade by considering six reactor
congurations. In sections (5.4.5) and (5.4.6), we investigate the performance of a
cascade of three and four reactors as a function of the total residence time inside
the cascade by considering six reactor congurations respectively. In section (5.4.7)
asymptotic solutions for the substrate and cell-mass concentrations at large total
residence times are stated.

Chapter 5

5.4.1

96

Steady-State diagrams

Steady state diagrams showing the variation of the euent concentration leaving
both a single reactor and a cascade of two reactors as a function of the total residence
time (t ) are shown in gure (5.2). It should be noted that only the stable physical
meaningful solutions are presented.
Figure (5.2(a)) shows the case when there is a settling unit around each reactor
with perfect recycle (R1 = R2 = 1). In both the single reactor and the two reactors
cascade, the washout solution is always unstable. For both reactor conguration,
the euent concentration leaving the reactor conguration decrease rapidly with an
increase in the total residence time. However, the euent concentration leaving the
reactor cascade decreases more rapidly than the euent concentration leaving the
single reactor.
Figure (5.2(b)) shows the case when there are no settling units There are three
distinct regions. In the rst region (0 t 1.01), washout occurs in each reactor
in the cascade. Thus, the euent concentration leaving the system is the same as the
pollutant concentration entering it (Se = 1). In the second region (1.01 t 2.1),
the euent concentration leaving the single reactor is lower than that leaving the
cascade reactor of two reactors. In this region, the performance of a single reactor
is superior to that of the two reactor cascade. In the third region (t > 2.1), the
euent concentration leaving the two reactor cascade is lower than that leaving the
single reactor.
We now investigate how to minimize the euent concentration by optimizing the
values of the recycle parameters R1 and R2 .

97

Two reactor
Single reactor

Dimensionless substrate concentration ( S e)

Chapter 5

0.1

0.01

0.001

0.0001
0

Dimensionless total residence time ()

Two reactor
Single reactor

Dimensionless substrate concentration ( S e)

(a) Ri = 1

0.1

0.01
Region 1 Region 2

Region 3

0.001

0.0001
0

Dimensionless total residence time ()

(b) Ri = 0

Figure 5.2: euent concentration in a cascade of two reactors (solid line) and in a
single reactor (dashed line).

98

Chapter 5

5.4.2

The eect of the recycle around the nth reactor

In this section we show that the euent concentration leaving a n-reactor cascade
is minimized by employing perfect recycle around the nal reactor (Rn = 1) for any
value of the total residence time.
Dierentiating the steady state equations (5.2.6-5.2.7) in the nal reactor of the
cascade with respect to Rn , we obtain
(

(
)
)
Xn2
n
dSn
Sn2
dXn

= 0, (5.4.71)
t
(Xn + Sn )2 dRn
(Xn + Sn )2 dRn
(
)
(
)
2
Xn
nn (Rn 1)
Sn2
nn Xn
dSn
dXn

K
+
=
.
d
(Xn + Sn )2 dRn
t
(Xn + Sn )2 dRn
t
(5.4.72)
We will use the fact that along the no-washout branch,

nn1 (1 Rn1
)Xn1
t Kd + nn (Rn 1)
Sn
=

t
(Xn + Sn )
t Xn

(5.4.73)

Applying Cramers rule to this system we obtain

(

)(

)
nn Xn
< 0,
n = 0
t
(
)
)(

dXn
1 n
Xn2
nn Xn
=
+
> 0,
n = 0
dRn
D t (Xn + Sn )2
t
dSn
1
=

dRn
D

S
n 2
(Xn + Sn )

(5.4.74)
(5.4.75)

where
(
D=

(
+

Xn 2
n
+
t (Xn + Sn )2

Sn

(Xn + Sn )2

)(

)(

)Xn1
nn1 (1 Rn1
Xn Sn
+
t Xn
(Xn + Sn )2

Xn

(Xn + Sn )2

(5.4.76)

)
> 0.

Hence, the euent concentration is minimized by taking Rn = 1.

Chapter 5

5.4.3

99

The eect of the recycle around the rst reactor

The euent concentration leaving a single reactor is minimized by having perfect

recycle (R1 = 1) [101]. We show that, in general, perfect recycle around any reactor
other than nal reactor does not minimize the euent concentration.
We consider the performance of a cascade of two reactors whereby the recycle parameter in the rst reactor (R1 ) is varied for a xed value of the recycle parameter
in the second reactor (R2 ). Figure (5.3) shows the performance of the cascade
of two reactors as the value of R1 is varied for two values of R2 . In both cases

there is an optimal value of R1 , denoted by R1,min

, at which the euent concen-

tration is minimized. This optimal value of R1 is less than the value of perfect

recycle (i.e. R1,min

< 1).

The natural question that arises is how does the optimal value of R1 , i.e R1,min
,

change if either the recycle around the second reactor (R2 ) or the total residence
time is varied. For this purpose, Figure (5.4(a)) shows the optimal value of R1 ,

i.e R1,min
, as a function of R2 with the total residence time xed at three values:

t = 6, t = 20 and t = 60. Clearly, in each case the optimal value of R1 , i.e

R1,min
, is an increasing function of the recycle around the second reactor (R2 ) with

dierent xed total residence time. Moreover, for xed recycle around the second
reactor (R2 ), gure (5.4(a)) shows that an increase in the total residence time leads

. This indicates that at high

to a decrease in the optimal value of R1 , i.e R1,min

total residence time the optimal value of R1 is achieved by setting R1 = 0, while

at suciently small total residence time, it is R1 = 1. Therefore at higher total
residence time, there is less needs of using the settling unit around the rst reactor

Chapter 5

100

in order to get its benet in the improvement of the reactor cascade performance.
This nding is in agreement with our results in section (5.4.7) which show that in
the asymptotic limit of large residence times the optimal value of R1 is zero.
The euent concentration that corresponds to the optimal value of R1 , donated by

S2 (R1 = R1,min
), as a function of the recycle around the second reactor (R2 ) with

dierent xed total residence time is shown in gures (5.4(b), 5.4(c) and 5.4(d)).

As observed in gure (5.3), the euent concentration (S2 (R1 = R1,min

)) is a slightly

decreasing function with the recycle around the second reactor (R2 ). One should

note that the change in the euent concentration (S2 (R1 = R1,min
)) is insensitively

small as the recycle around the second reactor is varied. This suggests that the
recycle around the second reactor has a minor eect on the performance of the two
reactor cascade.
A second feature of the diagram (5.3) is that for the case R2 = 0 there is a crit

ical value of R1 , denoted by (R1,max

). When 0 < R1 < R1,max
recycle has a posi
tive eect upon the euent concentration i.e. S2 (R1 ) < S2 (R1 = 0). If R1,max
<

R1 1 then recycle has a negative eect upon the euent concentration, i.e.

< R1 1) that
S2 (R1 ) > S2 (R1 = 0). We deduce from the previous region (R1,max

the performance of a cascade with perfect recycle around the rst reactor (R1 = 1)
is inferior to that of a cascade with no recycle in the rst reactor (R1 = 0) for the

only exists for particular values of the

case R2 = 0. It should be noticed that R1,max

is only
recycle around the second reactor (R2 ). For example, when t = 20, R1,max

found when the value of the recycle around the second reactor (R2 ) ranges between
zero to 0.1. From now and on we donate the maximum value of recycle around the

Chapter 5

101

second reactor (R2 ) in which R1,max

exist by R2,cr
.

To better understand the eect of the recycle around the second reactor upon R1,max
,

table (5.1) shows the value of R1,max

corresponding to two values of R2 , R2 = 0 and

R2 = R2,cr
for three total residence time: t = 6.58, t = 20 and t = 60. The

smallest total residence time given by t = 6.58 represents the minimum value of

the total residence time for which R1,max

occur. Thus, for any value of the recycle

around the second reactor, the performance of the reactor cascade is surly improved
by an operating the reactor cascade under t < 6.58.
It can be observed from table (5.1) that as the value of the total residence time

range between t = 6.58 and t = 60 the value of R2,cr

increases from 0.0002 to

increases
0.40. This indicates that at high total residence time, the value of R2,cr

to one in which the value of R1,max

exist for any value of recycle around the second

reactor.
As pointed out previously, gure (5.3) shows a surprising result, namely that the
optimal performance of a cascade with recycle around each reactor is not obtained
with perfect recycle around each reactor, i.e. when R1 = R2 = 1. It is given instead

by R1 = R1,min
and R2 = 1. The value of R1,min
must be determined for each value

of t . This surprising outcome is a result of two competing processes. As R1 increases, the microorganism concentrations (X1 ) entering the second reactor, due to
the term (1 R1 )X1 in equation (5.2.7), decreases. At the same time, the substrate
concentration (S1 ) entering the second reactor also decreases. These processes have
opposing eects upon the euent concentrations leaving the second reactor. Over
most values of R1 the later eect dominates and the substrate concentration de-

102

Chapter 5

creases in the second reactor. However, when R1 is close to one the former eect
dominates and the euent concentration increases in the second reactor. Thus there
is a competition in the second reactor between the benets of reducing the substrate
concentrations entering it and the disadvantages of reducing the microorganism concentrations entering it.
This analysis can be extended to a cascade of n reactors. Generally the performance
of a reactor cascade of n reactors is not minimized by having perfect recycle around
any of the preceding reactors (Ri , i = 1, 2, ..., (n 1)).

Dimensionless substrate concentration ( S 2)

0.00012
R*1,max
0.000115

0.00011

0.000105

0.0001
0

Figure 5.3:

0.2

0.4

0.6

Recycle parameter (R*1)

0.8

euent concentrations as a function of the recycle parameter R1 .

Parameter values: R2 = 0 (solid line), R2 = 1 (dashed line) and t = 20.

5.4.4

Reactor congurations of a two reactor cascade

In this section, six reactor congurations are considered for a two reactor cascade to
determine the conguration that produces the most ecient system for the removal

103

Chapter 5

0.001135

=6
=20
=60

0.00113
*
1,min)

0.99
0.98

*
S*2(R 1=R

Recycle parameter(R*1,min)

0.97
0.96

0.001125
0.00112
0.001115
0.00111
0.001105
0.0011

0.95

0.001095
0

0.2

0.4
0.6
0.8
*
Recycle parameter(R 2)

0.2
0.4
0.6
0.8
*
Recycle parameter(R 2)

(b) t = 6

(a)
0.0001042

*
1,min)

0.000104
0.0001039

*
S*2(R 1=R

*
*
S*2(R 1=R 1,min)

0.0001041

0.0001038
0.0001037
0.0001036
0.0001035
0.0001034
0

0.2
0.4
0.6
0.8
*
Recycle parameter(R 2)

(c) t = 20

1.166e-005
1.1658e-005
1.1656e-005
1.1654e-005
1.1652e-005
1.165e-005
1.1648e-005
1.1646e-005
1.1644e-005
1.1642e-005
1.164e-005
0

0.2
0.4
0.6
0.8
*
Recycle parameter(R 2)

(d) t = 60

Figure 5.4: The minimum point R1,min

(5.4(a)) as a function of the recycle parameter

R2 . The euent concentration that correspond to R1,min

, (S2 (R1 = R1,min
)) (5.4(b),

5.4(c), 5.4(d)) as a function of the recycle parameter R2 . The parameter value:

i = 1 where i = 1, 2.

Table 5.1: The maximum point R1,max

for the cases R2 = 0 and R2 = R2,cr
. The

parameter value: i = 1 where i = 1, 2.

, R2 = 0)
(R1,max

)
, R2 = R2,cr
(R1,max

=6.58 (0.999999721, 0)
t,cr2

(0.999999979, 0.0002)

20

(0.9998401, 0)

(0.9999930, 0.11)

60

(0.9994266, 0)

(0.9999946, 0.40)

104

Chapter 5

of wastewater. These are:

(a) No recycle around either reactor (Ri = 0),

i = 1, 2.

(b) Optimized recycle around the rst reactor and no recycle around the second

reactor (R1 = R1,min

, R2 = 0).

(c) Perfect recycle around the rst reactor and no recycle around the second reactor
(R1 = 1, R2 = 0).
(d) No recycle around the rst reactor and perfect recycle around the second reactor
(R1 = 0, R2 = 1).

(e) The optimized cascade (R1 = R1,min

, R2 = 1).

(f) Perfect recycle around each reactor (R1 = 1, R2 = 1).

A comparison of these reactor congurations is provided in gure (5.5) and table (5.2). The solution curves in gure (5.5) split into two components. The component with the lower euent concentration contains the optimized cascade (e), the
cascade with optimized recycle around the rst reactor (b) and the cascade with
perfect recycle around each reactor (f). The performance of the optimized cascade
(e) is marginally superior to that of the other two cascade (b and f).
The component with the higher euent concentration contains the congurations
where there is either one settling unit with perfect recycle (c and d) or no settling
units (a). The performance of the cascade with perfect recycle around the second
reactor (d) is slightly superior to the performance of the reactor cascade without
recycle (a).

Chapter 5

105

For suciently small values of the total residence time (t < 2/(1 Kd )), the performance of the optimized reactor cascade with one settling unit (b and d) and
perfect recycle around the rst reactor (c) is signicantly better than that of a cascade without recycle (a).
For large values of the total residence time (t 2/(1 Kd )), we have the following
results. The performance of a cascade with no recycle (a) converges towards the
performance of the optimized cascade with recycle around the second reactor (d).
Furthermore, the performance of the reactor cascade with perfect recycle around
the rst reactor (c) is worst than that without recycle (a). This outcome is not
surprising because we show in section (5.4.7) that in the asymptotic limit of large
residence times the euent concentration is minimized by taking R1 = 0. The performance of a cascade with optimized recycle around the rst reactor (b) converges
towards the performance of the optimized cascade (e). These result is explained by
the large residence time asymptotic (see section (5.4.7)).
We conclude that at suciently high total residence time there is little benet in
using a cascade with one and two settling units compared to using a cascade with
no settling units.
Table (5.2) shows the total residence time required to achieve a specied euent
concentrations for various reactor congurations. This table demonstrates a surprising result. With the euent concentration set to 0.1 the single reactor with
perfect recycle outperforms any cascade of two reactors using a single recycle unit
with perfect recycle (conguration c and d). The reason for this is as follows. For
the case (R1 = 1, R2 = 0), there are no microorganism entering the second reactor,

Chapter 5

106

due to the term (1 R1 )X1 in equation (5.2.7) being zero. Hence, when R1 = 1 the
second reactor of the cascade acts as a single reactor. Thus washout will occur in the
second reactor if the residence time in the second reactor is smaller than 1/(1 Kd ),
i.e t < 2/(1 Kd ). It follows that when t < 2/(1 Kd ), washout occurs in the
second reactor and only one reactor is working in the cascade.
Thus when < 2/(1 Kd ) the euent concentration emerging from conguration c (R1 = 1, R2 = 0) will be higher than that emerging from a single reactor
with perfect recycle with the same total residence time. Let T be the residence
time in a single reactor with perfect recycle required to obtain the desired euent
concentration Se = 0.1. It follow that if T < 2/(1 Kd ) then the single reactor
with perfect recycle will outperform the cascade of two reactor.
When the target euent concentration is 0.001, the performance of the two reactor cascade with perfect recycle around the rst reactor is slightly inferior to that
with no recycle. The reason for this is similarly to the case shown in gure (5.3)
in which the performance of a cascade with perfect recycle around the rst reactor
(R1 = 1, R2 = 0) is inferior to that of a cascade with no recycle in the rst reactor
(R1 = 0, R2 = 0). This kind of behavior does not occur when the euent concentration is either 0.1 or 0.01. This is because at these values of the euent concentration

), which was discussed earlier in relation to

there is no critical value for R1 , (R1,max

gure (5.3).
Furthermore, in agreement with gure (5.5), the data in table (5.2) shows that the
performance of the cascade with optimized recycle around the rst reactor (b) is
slightly worse than the performance of both the optimized cascade (e) and the cas-

Chapter 5

107

cade with perfect recycle around each reactor. Table (5.2) also demonstrates that
as the desired euent concentration decreases the performance of the cascade with
no recycle (a) converges to the performance of cascade with perfect recycle around
either the rst or the second reactor (c and d). For example, when the desired
euent concentration is Se = 0.1 we see from tables (5.2) that the residence time
required for a cascade of two reactors with congurations (c and d) is 86% of the
value without a settling unit. The corresponding values for euent concentrations
of Se = 0.001 for congurations (c and d) are 100% and 99% respectively.
Convergence in the total residence time is also seen in tables (5.2) between a cascade
with optimized recycle around the rst reactor (b) and the optimized cascade (e)
as the desired euent concentration decreases. We see from tables (5.2) when the
desired euent concentration is Se = 0.1 the total residence time required for the optimized cascade (e) is 65% of the value of a cascade with optimized recycle around the
rst reactor (b). The corresponding value for euent concentrations of Se = 0.001
is 98%. We observe in table (5.2) that the decrease in total residence time obtained
by switching from the cascade with perfect recycle around each reactor (f) to the
optimized cascade (e) is insignicant.
It is worth returning the question as to whether optimized recycle around the rst
or second reactor is a signicant tool to reduce the euent concentration. When the
desired euent concentration is Se = 0.1 we see from tables (5.2) that in the total
residence time required for a cascade of two reactors with optimized recycle around
the rst and second reactor are 31% and 86% of the value without a settling unit
respectively. The corresponding values for euent concentrations of Se = 0.001 are

108

Chapter 5

a
b
c
d
e
f

Dimensionless substrate concentration ( S 2)

(c and d)
0.1

0.01

0.001

(e and f)

0.0001
0

Figure 5.5:

Dimensionless total residence time ()

10

euent concentration S2 in a cascade of two reactors as function of

total residence time: (a) R1 = R2 = 0, (b) R1 = R1,min

, R2 = 0, (c) R1 = 1, R2 = 0,

(d) R1 = 0, R2 = 1, (e) R1 = R1,min

, R2 = 1 (the optimized cascade) and (f)

R1 = R2 = 1.
86% and 99% respectively. Thus it is better to replaced the recycle around the rst
reactor than the second reactor.
Finally, for target euent concentrations Se = 0.001, or higher, a signicant increase
in performance is obtained from using a reactor conguration employing optimized
recycle (e) compared against a reactor cascade without recycle (a).

5.4.5

Reactor congurations of a three reactor cascade

In this section, six reactor congurations are considered for a 3-reactor cascade to
determine the most ecient system. These are:
(a) No recycle around any reactor (Ri = 0),

i = 1, 2, 3.

109

Chapter 5

Table 5.2: total residence times in a cascade of n reactors (n = 1, 2) that is required

to reach a specic level of the euent concentration. The labels (af) refer to the
curves on gure (5.5).
R1 , n = 1
1

(R1 , R2 ), n = 2
0

Se

0.001

(R1,min
, 0)

(0, 1)

(1, 1) (R1,min
, 1)

0.6796 1.861

0.447

0.4468

1.944

2.165

1.861

12.10

3.185

3.110

2.005

3.100

1.861

1.860

111.10 112.19

7.414

7.425

6.395

7.382

6.333

6.305

0.1 0.9305
0.01

(0, 0) (1, 0)

11.01

(b) Optimized recycle around the rst reactor and no recycle around either the

second or third reactor (R1 = R1,min

, Ri = 0),

i = 2, 3.

(c) Optimized recycle around the second reactor and no recycle around either the

rst or third reactor (R2 = R2,min

, Ri = 0),

i = 1, 3.

(d) Optimized recycle around the third reactor and no recycle around either the

rst or second reactor (R3 = R3,min

, Ri = 0),

i = 1, 2.

(e) The optimized cascade (Ri = Ri,min

, R3 = 1),

i = 1, 2.

(f) Perfect recycle around each reactor (Ri = 1),

i = 1, 2, 3.

A comparison of these reactor congurations is provided in gure (5.6) and table (5.3). The solution curves in gure (5.6) again splits into two components. The
component with the lower euent concentration contains the optimized cascade (e),
the cascade with optimized recycle around the rst reactor (b) and the cascade with

Chapter 5

110

perfect recycle around each reactor (f). The performance of the cascade with perfect
recycle around each reactor (f) is slightly superior to the performance of the reactor cascade with optimized recycle around the rst reactor (b) and slightly inferior
to the performance of the optimized cascade (e). The data in table (5.3) shows
that when three settling units are used, the improvement in performance that is
obtained by optimising the values of the recycle parameters in the rst and second
reactors (e), compared against the performance when perfect recycle is used (f), is
only marginal.
In fact, for suciently large total residence time (t 3/(1 Kd )) the performance
of the cascade with optimized recycle around the rst reactor (b) converges towards
the performance of the optimized cascade (e). Thus at suciently large residence
times the system with a single optimised settling unit around the rst reactor is
little dierent from that obtained by using three settling units. However, the data
in table (5.3) shows that the dierence can be signicant if the target euent concentration is of the order of Se = 0.001 or higher. For instance, when the desired
euent concentration is Se = 0.1 we see from tables (5.2) that in the residence
time required for a cascade of three reactors with optimized cascade (e) using three
settling units is 50% of the value with a single optimised settling unit around the
rst reactor (b). The corresponding values for euent concentrations of Se = 0.01,
Se = 0.001 and Se = 0.0001 are 80%, 94% and 98% respectively.
The component with the higher euent concentration contains the reactor congurations where there is one settling unit placed around either the second or third
reactor (c and d) or no settling units (a). For suciently small total residence time

Chapter 5

111

(t < 3/(1 Kd )), the congurations in which there is one settling unit signicantly
outperforms the cascade without recycle.
For large total residence time (t 3/1 Kd ), there is little dierence between the
performance of the cascade with a single optimized settling unit placed around either
the second or the third reactor (c and d) and that the performance of the cascade
without recycle (a). In fact we show in section (5.4.7) that in the asymptotic limit of
large residence times the euent concentration is minimized by taking R1 = R2 = 0.
Furthermore, although the euent concentration is minimized by taking R3 = 1 this
is a second-order eect which is therefore insignicant at suciently high residence
times.
Table (5.3) shows that for the specied euent concentrations (0.1, 0.01 & 0.001),
the performance of a cascade using one optimized settling unit, placed around any
reactor (b, c & d), is superior to a cascade without recycle (a). The performance
when a single settling unit is optimized around the rst reactor (conguration b)
is superior to the performance when a single settling unit is optimized around the
second reactor (conguration c) which, in turn, is superior to the performance when
a single settling unit is optimized around the third reactor (conguration d). Thus
when only one settling unit is employed it should be placed around the rst reactor.
The dierence in performance between a cascade with optimized recycle around the
rst reactor (conguration b) and the optimized cascade (conguration e) decreases
as the target euent concentration decreases.
At a euent concentration Se = 0.001, the performance of the cascade with perfect
recycle around either the rst or the second reactor is inferior to that of a cascade

Chapter 5

112

without recycle. Similar behavior does not occur when the euent concentration
is either 0.1 or 0.01. The reason for this is that when Se = 0.001 the steady-state
diagram is similarly to the case shown in gure (5.3) in which there is a critical

value of Ri , Ri,max
, (i = 1, 2). Evidently, the steady-state diagrams when Se = 0.01

or Se = 0.1 is similarly to the case shown in gure (5.3) in which there is no critical
value for Ri .
For target euent concentrations Se = 0.001, or higher, a signicant increase in performance is obtained from using a reactor conguration employing optimised recycle
compared against a reactor cascade without recycle.
Finally, it is interesting to observe the circumstances under which three reactors
with an optimised settling unit placed around the rst reactor outperforms a reactor cascade of four reactors without recycle. The decision to either add a third
reactor or a settling unit to a cascade of three reactors depends upon the desired
euent concentration. The total residence time required to reach an euent concentration (0.01, 0.001 and 0.0001) in a four reactor cascade without recycle are
4.07, 4.28 and 5.38 respectively. Comparing to the data in the table (5.3), we reach
the conclusion that when the designed euent concentration is greater than 0.001,
then a settling unit should be added to the cascade of three reactors. If the designed
euent concentration is smaller than 0.0001, then the choice of adding one reactor
to the cascade should be made.

113

Dimensionless substrate concentration ( S 3)

Chapter 5

a
b
c
d
e
f

0.5

0.1
0.05
(e and f)
0.01
0.005

0.001
0.0005

0.0001
5e-005

1e-005
0

4
6
Dimensionless total residence time ()

10

Figure 5.6: euent concentration S3 in a cascade of three reactors as function of

total residence time: (a) Ri = 0, (b) R1 = R1,min

, Ri = 0, (c) R2 = R2,min
, Ri = 0,

(d) R3 = 1, Ri = 0, (e) Ri = Ri,min

(the optimized cascade) and (f) Ri = 1.

Table 5.3:

total residence times that is required to reach a specic level of the

euent concentration, a cascade of three reactors. The labels (af) refer to curves
on gure (5.6).

(R1 , R2 , R3 ), Ki = Ri,min
,n=3

(1, 1, 1) (0, 0, 0) (K1 , 0, 0) (1, 0, 0) (0, K2 , 0) (0, 1, 0) (0, 0, 1) (K1 , K2 , 1)

Se

0.1

0.358

3.052

0.704

2.791

1.019

0.01

1.129

3.234

1.390

3.219

0.001

2.791

4.306

2.939

0.0001

6.373

7.555

6.436

1.861

2.791

0.357

3.008

3.100

3.216

1.125

4.309

4.196

5.073

4.298

2.776

7.582

7.505

8.512

7.553

6.330

Chapter 5

5.4.6

114

Reactor congurations of a four reactor cascade

Six reactor congurations are considered for a 4-reactor cascade, to determine the
most ecient system. These are:
(a) No recycling around each reactor.
(b) Optimized recycle around the rst reactor and no recycling around the subsequent reactors.
(c) Optimized recycle around the second reactor and no recycling around the other
reactors.
(d) Optimized recycle around the third reactor and no recycling around the other
reactors.
(e) Optimized recycle around the fourth reactor and no recycling around the preceeding reactor.
(f) The optimized cascade.
The comparison of these reactor congurations is demonstrated in gure (5.7). The
solution curves in gure (5.7) again split into two components. The component
with the lower euent concentration contains the optimized cascade (f) and the
cascade with optimized recycle around the rst reactor (b). The performance of
conguration (f) is only slightly superior to the performance of conguration (b).
For large total residence time (t 4/(1 Kd )) the performance of conguration
(b) converges to the performance of conguration (f). Thus, there may be little gain
from using the three additional settling units required for the optimized cascade.

115

Chapter 5

a
b
c
d
e
f

Dimensionless substrate concentration ( S e)

0.1

0.01

0.001

0.0001

1e-005

1e-006
0

Dimensionless total residence time ()

Figure 5.7: Euent concentration in a cascade of four reactors as a function of to

tal residence time where (a) Ri = 0, (b)R1 = R1,opt

, Ri = 0, (c) R2 = R2,opt
, Ri = 0

(d) R3 = R3,opt
, Ri = 0 and (e) R4 = 1, Ri = 0, (f) Ri = Ri,opt
, R4 = 1.

The component with higher euent concentration contains the congurations where
there are one settling unit (c, d and e) or no settling units (a).
For small total residence time (t < 4/(1 Kd )), the performance of the reactor
cascade with the optimized recycle around the ith reactor improves as the value for
i decreases. In this region, the cascade using a single setting unit around any reactor
is superior to the cascade with no recycle.
For large total residence time (t 4/(1 Kd )), the performance of the cascade
with optimized recycle around either the second, the third reactor or fourth reactor
(c, d and e) is superior to the performance of the cascade with no recycle (a).
However, the performance of conguration (a) converges to that of the performance
congurations (c, d and e) in the limit of innite total residence time.

116

Chapter 5

5.4.7

Large residence time approximations

In this section, we provide large total residence time approximations for the concentrations of substrate and microorganisms. This result will explain the interesting
behavior that observed in sections (5.4.2) to (5.4.5) regarding to the recycle eect
upon the reactor cascade. We nd the following approximations:
S1

+O
(1 Kd )t

1
t2

)
,

1
n
+ O( 2 ),

Kd t
t
(
)2
( )
n
1
e1
+O
,

(1 Kd )t
t3
)2
( )
(
1
n

+O
,
d1

Kd t
t3
)3
( )
(
1
n
e1 e2
+O
,

t (1 Kd )
t4
(
)3
( )
n
1

d1 d2
+O
,

t kd
t4
)4
(
)
(
n
1
e1 e2 e3
+ O ( 5 ,
(1 Kd )t
t
(
)4
(
)
n
1

d1 d2 d3
+ O ( 5 ,
Kd t
t
)
( )
(
5
1
n
e1 e2 e3 e4
+O
,

(1 Kd )t
t6
(
)5
( )
n
1

d1 d2 d3 d4
+O
,

Kd t
t6

(5.4.77)

X1

(5.4.78)

S2

(5.4.79)

X2
S3
X3
S4
X4
S5
X5

where,

e1 =

(1 Kd )(1 R1 ) + Kd
,
Kd + (1 R1 )

d1 =

(1 Kd )(1 R1 ) + Kd
,
1 Kd

2 Kd 2 + (1 R2 )(1 Kd )( Kd + 1 R1 )
,
e2 =
2 Kd 2 + (1 R2 )( Kd + 1 R1 )

(5.4.80)
(5.4.81)
(5.4.82)
(5.4.83)
(5.4.84)
(5.4.85)
(5.4.86)

117

Chapter 5

d2 =
e3 =
d3 =
e4 =
d4 =

2 Kd 2 + (1 R2 )(1 Kd )( Kd + 1 R1 )
,
(1 Kd )( Kd + 1 R1 )
[
]
(1 R3 ) (1 Kd ) 2 Kd 2 + (1 R2 )( Kd + 1 R1 ) + Kd 3 3
[
]
,
(1 R3 ) 2 Kd 2 + (1 R2 )( Kd + 1 R1 ) + Kd 3 3
[
]
(1 R3 ) (1 Kd ) 2 Kd 2 + (1 R2 )( Kd + 1 R1 ) + Kd 3 3
[ 2 2
]
,
Kd + (1 R2 )( Kd + 1 R1 ) (1 Kd )
]
[
(
)
1 R4 (1 Kd ) {(1 R3 ) 2 Kd 2 + (1 R2 )( Kd + 1 R1 ) + Kd 3 3 } + Kd 4 4
]
}
(
){
[
,
1 R4 (1 R3 ) 2 Kd 2 + (1 R2 )( Kd + 1 R1 ) + Kd 3 3 + Kd 4 4
]
(
[
)
1 R4 (1 Kd ) {(1 R3 ) 2 Kd 2 + (1 R2 )( Kd + 1 R1 ) + Kd 3 3 } + Kd 4 4
{
[
]
}
(1 R3 ) 2 Kd 2 + (1 R2 )( Kd + 1 R1 ) + Kd 3 3 (1 Kd )

Equations (5.4.77-5.4.86) give the substrate and microorganisms concentrations in

the rst ve reactors of a cascade. These expressions hold for any value of n. For
example, substituting n = 7 in these expressions gives the concentrations in the rst
through to the ve reactor of a seven-reactor cascade.
Equations (5.4.77, 5.4.79, 5.4.81, 5.4.83 and 5.4.85) show that the substrate concentration leaving a n reactor cascade (Sn ) is proportional to

1
(t )n

(n = 1, ..., 5). It is

conjectured that this holds for any n. Should this be true then, at suciently large
total residence times, the addition of an extra reactor reduces the euent concentration by a multiplicative factor of

1
.
t

Equations (5.4.77, 5.4.79, 5.4.81, 5.4.83 and 5.4.85) show that the euent concentration in each reactor decreases to zero at large values of the total residence times.
At large values of the total residence time and at leading order the performance of
the single reactor conguration is independent of the value of the recycle parameter
R1 . At large values of the total residence time the euent concentration leaving a

but
cascade of n rectors depends upon the recycle parameter values R1 , R2 , ..., Rn1

is independent of the value of the recycle parameter Rn .

We now investigate the eect of the recycle in the preceding reactor upon the per-

118

Chapter 5

formance of the reactor. We show a surprising result namely that at large total
residence times recycle around the preceding reactors adversely aects the performance of the reactor. Dierentiating equations (5.4.79 and 5.4.81) we obtain
dS2
n2 3 Kd 2
=
> 0,
dR1
[ Kd + (1 R1 )]2 (1 Kd )2 t 2

(5.4.87)

dS3
q1 q2 q3 Kd 2 4 n3
> 0,
=
dR1
(1 Kd )3 t 3 ( Kd + 1 R1 )2 q42

(5.4.88)

[(1 Kd ) (1 R1 ) + Kd ] n3 Kd 3
dS3
=[
> 0.
]2
dR2
2 Kd 2 + (1 R2 ) ( Kd + 1 R1 ) (1 Kd )3 t 3
5

(5.4.89)

where
[
]
q1 = (1 R2 ) (3 Kd ) 3 Kd 3 + (1 R2 ) (1 Kd ) (1 R1 )2 ,
[
]
q2 = Kd [(1 Kd ) (1 R1 ) (1 R2 )] [(1 R1 ) + 2 (1 R2 )] + 3 Kd 3 ,
q3 = (1 R2 ) 2 Kd 2 [(1 R2 ) (1 Kd ) + 2 (2 Kd ) (1 R1 )] ,
q4 = 2 Kd 2 + (1 R2 ) Kd + (1 R2 ) (1 R1 ) .
Hence at large total residence time, the euent concentration leaving the nal reactor of the cascade is an increasing function of the recycle parameters in the preceding
reactors (Ri , i = 1, ..., (n 1)). At large residence time, equations (5.4.77-5.4.82)
show the substrate concentration and the microorganism concentrations leaving
the reactor i is independent of the recycle parameter (Ri ). However from equation (5.2.7) the microorganism concentrations entering reactor (i+1) is (1 Ri )Xi
and this is a decreasing function of the recycle parameter. Accordingly at large
values of the total residence time, the use of a settling unit around the ith reactor (1 < i < n 1) reduces the performance of the reactor cascade. The euent
concentration is therefore minimized in the limit of large total residence time by

119

Chapter 5

eliminating recycle around all of the preceding reactors (Ri = 0, 1 i < n 1).
As noticed earlier in the chapter (4) when there is a single reactor the operation
of the settling unit has a second order eect at large residence time. We now investigate the behavior of a two reactor cascade and established a similar result. At
large total residence time the operation of a settling unit around the second reactor
is only a second order eect. The leading term of the asymptotic series for S2 does
not include the parameter R2 . Therefore, in order to investigate the eect of this
parameter upon the euent concentration, we take the rst and second non-zero
terms of the asymptotic series for a two reactor cascade to obtain
(
S2

e1

n
(1 Kd )t

)2

(
+ e5 (e6 + e7 )

n
(1 Kd ) ((1 R1 ) + Kd ) t

)3

(
+O

1
t4

(5.4.90)
where
e5 = Kd + (1 R1 ) (1 Kd ) ,
e6 = (1 R1 ) [ (1 R1 ) (1 Kd ) + Kd (1 R2 ) 2 Kd ] ,
e7 = Kd 2 [(1 R2 ) ] .

Dierentiating equation (5.4.90), we obtain

dS2
[(1 Kd ) (1 R1 ) + Kd ] Kd n3 3
< 0.
=
dR2
t 3 (1 Kd )3 [ Kd + (1 R1 )]2

(5.4.91)

Thus the euent concentration leaving the second reactor is a decreasing function
of the recycle parameter (R2 ) and at xed total residence time, the performance
of a cascade of two reactors reaches the highest level of eciency at the maximum
value of the recycle parameter R2 = 1. Consequently, at large total residence times,

)
,

120

Chapter 5

the euent concentration leaving a cascade of two reactor is minimized by setting

R1 = 0 and R2 = 1. However, the inuence of a settling unit around the second
reactor is only at second-order. Thus, at high values of total residence times the
performance of the optimized cascade (R1 = 0 and R2 = 1) will be only marginally
superior to that of a cascade with no settling units. The same conclusions hold for
dS

a cascade with three reactors ( dR3 > 0) but the calculation are omitted here.
3

5.5

Conclusion

We have investigated a bioreactor model for the interaction between a microorganism and a rate-controlling substrate. The specic growth rate model that we used
was the Contois model with the addition of a microorganism decay coecient. In
recent years this biochemical model has found diverse applications in describing the
treatment of wastewater from a variety of industrial processes. Using a cascade of
two, three and four reactors, we investigated various congurations in which the
recycle stream leaving a settling unit placed around particular reactor enters the
feed stream of that reactor.
The steady-state solutions of the non-dimensional model were found and their stability determined as a function of the total residence time. Asymptotic solutions were
found for the no-washout solution branch. It was shown that at high total residence
times the euent concentration (Sn ) is minimised by setting the recycle parameter in all reactors preceeding the nal one to zero (Ri = 0, i = 1, 2, . . . n 1). An
application of Cramers rule shows that at any total residence time the euent concentration is minimised by using perfect recycle around the nal reactor (Rn = 1).

Chapter 5

121

However, the asymptotic analysis shows that at high total residence times this is a
second order eect.
For moderate values of the total residence time the euent concentration is minimised when the reactor parameter on the nal reactor is maximised (Rn = 1, corresponding to perfect recycle). In most circumstances the performance of the reactor
cascade is not maximised by using perfect recycle on any of the preceeding settling
units but is instead optimised by selecting a value between zero and one. When only
one settling unit is employed we found that the performance of the reactor cascade
is optimised by placing it around the rst reactor.
For target euent concentrations Se = 0.001, or higher, a signicant increase in performance is obtained from using a reactor conguration employing optimised recycle
compared against a reactor cascade without recycle.

Chapter 6
Analysis of continuous ow
bioreactor models with recycle
around the whole reactor cascade

6.1

Introduction

The industrial treatment of wastewaters typically employs a reactor cascade. In a

reactor cascade of n reactors the euent stream from the ith reactor in the cascade
acts as the feed stream for the (i+1)th reactor, i.e. the next reactor. The eciency
of the reactor cascade may be improved by using a settling unit. The settling unit
captures and concentrates the microorganisms in the euent stream of reactor (i)
and recycles it into the inuent stream of reactor (j, j i). The benet of using
the settling unit is that it increases the concentration of microorganisms in reactor
j, hopefully leading to an improvement in the performance of the cascade.

122

Chapter 6

123

In the previous chapter, we consider a reactor cascade with n-reactor in which a

settling unit recycles around reactor. The main nding of the previous chapter is
that the optimized reactor cascade is obtained by using perfect recycle around the
nal reactor and imperfect recycle around the preceding reactors.
In this chapter we investigate the use of a settling unit to reduce the euent concentration leaving a reactor cascade. We consider a commonly used industrial conguration in which the settling unit is placed after the nal reactor of the cascade and
recycles a proportion of the euent stream into the feed stream of the rst reactor.
This process is illustrated for a two reactor cascade in gure (6.1). We call this
scenario conguration (1). This conguration diers from that considered in an earlier chapter (5), in which the euent stream leaving any reactor in the cascade was
recycled into its own feed stream. We call this scenario conguration (2). In conguration (2) a settling unit is characterised by a single number, the dimensionless
recycle parameter, which ranges between zero (no recycle) and one (perfect recycle)
whereas in conguration (1) it is characterised by two parameters: a concentrating
factor (C) and a recycle parameter (R).
The governing equations concerning the performance of an n-reactor cascade employing a conguration (1) cannot be solved analytically. Therefore, we illustrate
the results numerically for the case of 2 reactors. Subsequent studies using 3,4 and
5 reactors (not included) did not inuence the ndings reported here. Also Grady
and Lim [80] suggest that a practical reactor cascade has a maximum of 5 reactors.
Hence we restricted our studies to a maximum of 5 reactors.
Steady-state analysis is used to determine the performance of conguration (1) and

Chapter 6

124

to compare it against that of conguration (2). Employing conguration (1), we

nd that there is a critical value of the total residence time. If the total residence
time is below (above) the critical value then the settling unit improves (worsens)
the performance of the cascade relative to that of a cascade without a settling unit.
This is noteworthy because in conguration (2) the addition of a settling unit always improves the performance of a cascade, see chapter (5). The critical values
of total residence time depends upon the values of the recycle ratio (R) and the
concentration factor (C). We use our results to investigate whether it is better to
add a settling unit or an extra reactor to a pre-existing reactor cascade.
The literature relating to the mathematical analysis of cascade reactors was discussed in chapter (2). The most pertinent of these to the subject of this chapter is
the work Grady and Lim [177, 178] who investigated the use of settling units in the
biological treatment of wastewater. They found that the recycle ratio has no signicant eect on either the substrate concentration or the microorganism concentration
in each reactor and consequently concluded that recycle is not a signicant tool to
reduce the euent concentration.
The main contributions of the current chapter are the following major features:
1. to extend the Contois model in chapter (5) to include recycle around whole
cascade;
2. to provide a detailed investigation of the steady-state behaviour of this process
model;
3. to compare reactor congurations (1 and 2) based on the criteria that minimize
the euent concentration leaving the nal reactor;

125

Chapter 6

4. to study the eect of the recycle on the performance of the reactor cascades.
A part of this chapter has been published as a journal article in leading chemical
engineering journal [179] and has been published as conference papers [175, 176].

Figure 6.1:

A schematic diagram of the ow around a two reactor cascade (Con-

guration 1).

6.2

Model equations

6.2.1

The dimensional model

The model equations for a n reactor cascade with recycle around the whole cascade
are given by,
Equations in the rst reactor
(S1 , X1 )
dS1
= F (S0 S1 ) V X1
+ F R(Sn S1 ),
dt

dX1
V
= F X1 + RF (CXn X1 ) + V X1 (S1 , X1 ) Kd V X1 .
dt
V

(6.2.1)
(6.2.2)

Equations in the ith reactor, (1 < i n)

dSi
(Si , Xi )
= F (1 + R)(Si1 Si ) V Xi
,
dt

(6.2.3)

126

Chapter 6

dXi
= F (1 + R)(Xi1 Xi ) + V Xi (Si , Xi ) Kd V Xi .
dt

(6.2.4)

Specic growth rate (Contois model).

(
(Si , Xi ) = max

Si
Ks Xi + Si

)
.

(6.2.5)

In the following i (1 < i n) denotes the ith reactor in a cascade containing n reactors. All parameters in the model are strictly non-negative. For details of the
assumptions and denition of model parameters, we refer readers to chapter (4) and
appendix (D) respectively.
For a specic wastewater, a given biological community and a particular set of environmental conditions the parameters Ks , Kd , and max are xed. The parameters
that can be varied are C, R , S0 , X0 and t . In our numerical simulations we
use parameter values for the anaerobic digestion of ice-cream wastewater proposed
by Hu et al. [39]. These are: = 0.2116 (gVSS)(g COD)1 , max = 0.9297 (day1 ),
Kd =0.0131 (day1 ) and Ks =0.4818 (COD)(g VSS)1 .
The operation of the settling unit is characterized by two parameters: a concentrating factor (C) and a recycle parameter (R). The maximum value of the concentrating
factor that be achieved in a specic settling unit is related to the value of the recycle
parameter.
The value of the maximum concentrating factor Cmax is given by
Cmax = (1 +

1
).
R

(6.2.6)

Alternatively the value of the maximum recycle Rmax is given by

Rmax =

1
,
C 1

C 1.

(6.2.7)

127

Chapter 6

The cases (R = Rmax ) and (C = Cmax ) represent perfect recycle. The constraint on
the maximum concentration factor and maximum recycle ratio in (6.2.6 and 6.2.7)
is derived by considering a mass balance for the case when all the microorganism
concentration leaving the euent stream of reactor cascade is captured by the setting unit. This means that there are no microorganisms owing out of the reactor
cascade.
The value of the recycle ratio depends upon the type of the reactor. Typically, the
designed value can vary between 0.1 to 1 [80][pp.117, 446, 622-664]. Furthermore, it
is recommend that the recycle pumps should be capable of delivering ows between
one-half and twice the design value [80]. The concentration factor (C) typically
varies between one to three [80][pp.117, 446-447, 664].

6.2.2

The dimensionless model

By using the transformations introduced in chapter (5), the equations, (6.2.1-6.2.4)

can be written into following dimensionless form,
Equations in the rst reactor.
dS1
n
S1 X1
nR

= (1 S1 )
+ (Sn S1 ),

dt
t
(S1 + X1 )
t

(6.2.8)

dX1
nX1
S1 X1
nR
=

+
+
(CXn X1 ) Kd X1 .

dt
t
(S1 + X1 )
t

(6.2.9)

Equations in the ith reactor.

Si Xi
n(1 + R)
dSi

S
=
(S
,
i
i1
dt
t
(Si + Xi )

Xi )
n(1 + R)(Xi1
dXi
Si Xi
=
Kd Xi .
+
dt
t
(Si + Xi )

(6.2.10)
(6.2.11)

Chapter 6

128

All dimensionless parameters in the model are strictly non-negative. For more details of the assumptions and denition of model parameters, we refer readers to
chapter (4) and appendix (D) respectively.
A feature of our dimensionless scheme is that there is a one to one relationship
between our dimensionless variable and their dimensional counterparts. Hence, we
write often residence time, rather than dimensionless residence time.

6.3

Results

The steady-state solution of equations (6.2.8-6.2.11) and their stability are determined numerically. Steady state diagrams showing the variation of the euent
concentration (Sn ) as a function of the total residence (t ) time are plotted. It
should be noted that only the stable physical meaningful solutions are presented.
In section (6.3.1), the positive and boundness of the solution are determined and in
section (6.3.2) the steady state solutions branches are discussed. In section (6.3.3),
the stability of the washout solution for an n-reactor cascade (n = 2, ..., 5) is determined. In section (6.3.4) we compares the performance of a two-reactor cascade
with recycle around the whole cascade against a two reactor cascade without recycle.
We discuss the eect of changing the concentration factor (C) for a given recycle
ratio for a two-reactor cascade in section (6.3.5). In section (6.3.6) we compare the
performance of a two-reactor cascade with that of a three-reactor cascade and nally
in section (6.3.7) we provide asymptotic solutions for the substrate and cell-mass
concentrations for an n-reactor cascade (n = 1, ..., 5) at large total residence times.
We show that the washout solution is globally stable when Kd 1. Hence, we

129

Chapter 6

assume in the following that 0 < Kd < 1.

6.3.1

The positive and bounded of the solution

In this section the aim is to establish the basic properties of invariance for the

region = {(S1 , ..., Sn , X1 , ..., Xn ) R2n

+ , 0 Si , Xi 0, i = 1, ..., n.}.

On the boundary of , if X1 = 0 then

dX1
dt

nRC
Xn ,
t

if Xn > 0 then the vector

eld of X1 point inside and if Xn = 0 then the plane X1 = 0 is invariant. Likely,

if S1 = 0 then

dS1
dt

n
t

nR
(Sn ),
t

which is strictly positive, then the vector eld of

S1 point inside . By the same way, if Xi = 0 then

dXi
dt

n(1+R)Xi1
.,

if Xi1
>0

then the vector eld of Xi point inside and if Xi1

= 0 then the plane Xi = 0 is

invariant. On the other hand, if Si = 0 then

dSi
dt

n(1+R)
Si1 ,
t

if Si1
> 0 then the

vector eld of Si point inside and if Si1

= 0 then the plane Si = 0 is invariant.

We conclude that is positively invariant.

For the boundness of Si and Xi , suppose that Sn (t) t+ +, so we can extract
a subsequence (tj ) such that tj j+ + and Sn (tj ) j+ + (is increasing
to +).
Hence,
S n (tj ) 0
It follows that
n(1 + R)
0 S n (tj )
(Sn1 (tj ) Sn (tj ))
t
so

(tj )
Sn (tj ) Sn1

130

Chapter 6

By the way,

Sn1
(tj ) j+ +

By the same way we can extract a subsequence, noted also (tj ) to simplify the

redaction, such that tj j+ + and Sn1

(tj ) j+ +

hence

S n1
(tj ) 0

It follows that
n(1 + R)

(Sn2 (tj ) Sn1

(tj ))
0 S n1
(tj )

t
so

Sn1
(tj ) Sn2
(tj )

By the way,

Sn2
(tj ) j+ +

Using the same reasoning, we prove that

Si (tj ) j+ + i = 1, ..., n
Now, by summarizing in equations S 1 and S i i = 2, ..., n, on tj we obtain that
0

i=1

Si (tj )Xi (tj )

n
S i (tj ) = (1 Sn (tj ))
t
(Si (tj ) + Xi (tj ))
i=1
n

So,
n

i=1

n
Si (tj )Xi (tj )
(1 Sn (tj ))

(Si (tj ) + Xi (tj ))

t

since Sn (tj ) j+ + then there exist a j0 0 such that for all j j0 , (1

Sn (tj )) < 0 and hence
n

i=1

Si (tj )Xi (tj )

<0
(Si (tj ) + Xi (tj ))

131

Chapter 6

which is contradiction. So, Sn is bounded. The boundness of Si ,, i = 1, ..., n 1

follows immediately from this fact. Note that by convention S =

dS
.
dt

In order to prove Xi (i=1,2,...,n) is bounded, we introduce a new variable that is

Z=

n
i=1

Xi . Hence, equations (6.2.9-6.2.11) become,

X S
n(1 + R)
nR
n
dZ
i i

+
X

K
(Xi1 Xi ) + (CXn X1 ) X1 ,
=
i
d

dt
X i + Si
t
t
t
i=1
i=2
i=1
n

Xi Si
nXn
=
+
(CR (R + 1)) ZKd ,

X i + Si
t
i=1
n

Xi Si

ZKd ,

Xi + Si
i=1

Because CR (R + 1) < 0,

(6.3.12)

We have proved that Si is bounded for all i, so

Si Xi
Si Xi

Si
(Si + Xi )
Xi
then

Si Xi
(Si +Xi )

is also bounded.

It follows that

Xi Si
i=1 Xi +Si

is also bounded. Thus, we have,

dZ
= L1 ZKd
dt

(6.3.13)

Xi Si
where L1 = maxt
. It follows that
Xi + Si
i=1

Z C4 + C5 exp[Kd t ]

(6.3.14)

Obliviously this fact implies that Z is bounded and hence that Xi is bounded for
all i = 1, .., n.

6.3.2

Steady state solutions

Under washout conditions, the concentration of the euent is equal to the concentration of the inuent. This state of operation must be avoided. The washout,

132

Chapter 6

steady state solution is given by:

(Sj = 1, Xj = 0),

(1 j 5).

(6.3.15)

It is not possible to nd the no-washout steady state of the system (6.2.8-6.2.11)

analytically. However, some interest points can be derived from the system (6.2.86.2.11). After some algebra, we nd the following properties.
Xn
t
(RC

(1
+
R))

X K ,

n i=1 i d
n

Sn = 1 +

(6.3.16)

Equation (6.3.16) shows when the recycle around the whole a cascade is perfect (RC
(1 + R) = 0) with no death rate of the microorganism then only value of the steady
state solution is Sn = 1, i.e, washout solution. Since, the washout steady state
solution is unstable when the recycle around whole a cascade is perfect , see equation (6.3.20), then the solution of the microorganism concentration should move
out from the washout of steady state solution to innity. Once should noted that
equation (6.2.10) has washout solution when the substrate concentration leaving the
i th reactor is less/or equal to that leaving the (i+1) th reactor. Thus, along the
no-washout solution, the substrate concentration leaving the i th reactor is always
higher than that leaving the (i+1) th reactor.

6.3.3

Stability of the steady state solutions

6.3.3.1

Global stability of washout solution when Kd 1

Theorem 6.3.1 If Kd > 1 then the washout solution is globally asymptotically

stable on .

133

Chapter 6

Proof. To prove global stability of the globally washout solution whenever Kd > 1,
we consider the argument from [170]. We introduce a new variable Z =

n
i=1

Xi .

Hence, equations (6.2.9-6.2.11) become,

X S

dZ
n(1 + R)
nR
n
i i

+
X
(Xi1 Xi ) + (CXn X1 ) X1 ,
=

K
i
d

dt
X i + Si
t
t
t
i=1
i=2
i=1
n

n(1 + R)
nR
n
+
(Xi1 Xi ) + (CXn X1 ) X1 ,

t
t
t
i=2
n

Z(1
because

Kd )

Xi Si
Xi

Xi + Si

= Z(1

Kd )

nXn
+ (CR (R + 1)) ,
t

Z(1 Kd ),

because CR R + 1.

Equation (6.3.17) shows that

dZ
dt

(6.3.17)

< 0 when Kd 1. Since, the solution Xi , i =

1, 2, ..., n are positive and bounded (see section (6.3.1)). Thus as

Z(t ) Ae(1Kd )t

where A is a constant.
Hence, the variable Z is a decreasing function of t , reaching zero when t increases
to innity. It follows that X1 , ..., Xn are decreasing functions of t reaching zero
when t increases to innity. Thus, the washout solution is globally stable when
Kd 1.

6.3.3.2

Stability along the washout branch

The stability of the washout branch for an n-reactor cascade (n = 2, ..., 5) is determined by evaluating the Jacobian matrix at the washout steady state. The eigenvalues are determined for an n-reactor cascade (n = 2, ..., 5) and it is presented in

134

Chapter 6

appendix (A.2). It is found that the real parts of all the eigenvalues are negative,
i.e., the washout branch is stable, when the following formula hold,
{
[
]1/n }
n (1 + R) (1 + R)n1 RC
t < w =
, n = 2, ..., 5.
1 Kd

(6.3.18)

It is conjectured that this formula (6.3.18) holds for all n. This condition can never
be satised for a reactor employing perfect recycle as w (C = Cmax ) = 0.
In appendix (A.0.1) we show that the critical value of the residence time (w ) is a
decreasing function of the recycle ratio (R). It follows that its maximum value is,
w (R = 0) =

n
.
1 Kd

(6.3.19)

and its minimum value is,

w (R = Rmax ) = 0.

6.3.3.3

(6.3.20)

Stability along the no-washout branch

The Jacobian matrix for the system of equations (6.2.8-6.2.11) along the no-washout
branch is,

0
a1 a2 a3

a
a6
4 a5 0

J(S1 , X1 , S2 , X2 ) =

a
0 a8 a9
7

0
a10 a11 a12

Where,
a1 = k1 + k2 , a2 = k10 , a3 = k3 , a4 = k11 , a5 = k4 k5 + k6 , a6 = k4 , a7 =
k1 , a8 = k1 + k7 , a9 = k12 , a10 = k1 , a11 = k13 , a12 = k1 k8 + k9 ,

135

Chapter 6

k1 =

2(1+R)
,
t

k2 =

X2
,
(X2 +S2 )2
2
S2

(X2 +S2 )2

k8 =

, k13 =

X1
,
(X1 +S1 )2
X1
,
X2

k9 =

2
X2

(X2 +S2 )2

k3 =

2R
,
t

S2 X2
,
(X2 +S2 )2

k4 =

2RC
,
t

k5 =

X2
, k6
X1

k10 =

S1
,
(X1 +S1 )2

S1 X1
,
(X1 +S1 )2

k7 =

k11 =

X1
,
(X1 +S1 )2

k12 =

It should be noted that ai > 0, i = 1, ..., 12. In the matrix (J), we have used the
facts that along the no-washout branch,
1)
2)

t Kd +2(R+1)
t
t Kd +2(R+1)
t

=
=

2RCX2
t X1

2(R+1)X1
t X2

S1
.
(X1 +S1 )

S2

(X2 +S2 )

The polynomial characteristic associated with matrix (J) has been obtained by using
the maple package [173] and given by,
z() =4 + b1 3 + b2 2 + b3 + b4 ,

(6.3.21)

Where,
b1 = a12 + a8 + a5 + a1 ,
b2 = a10 a6 + a11 a9 + a12 a8 + a12 a5 + a12 a1 a7 a3 + a8 a5 + a8 a1 + a4 a2 + a5 a1 ,
b3 = a10 a6 a8 a10 a6 a1 + a11 a9 a5 + a11 a9 a1 a12 a7 a3 + a12 a8 a5 + a12 a8 a1 + a12 a4 a2 +
a12 a5 a1 a7 a3 a5 + a8 a4 a2 + a8 a5 a1 ,
b4 = a10 a4 a3 a9 +a11 a7 a2 a6 +a10 a6 a7 a3 a10 a6 a8 a1 +a11 a9 a4 a2 +a11 a9 a5 a1 a12 a7 a3 a5 +
a12 a8 a4 a2 + a12 a8 a5 a1 .
The Routh-Hurwitz Criterion is general technique that can be used to determine if
the eigenvalue of characteristic polynomial have negative real part. The explicitly
Routh-Hurwitz stability conditions for fourth degree of polynomial equation (6.3.21)
are given by [180],
b1 > 0, b3 > 0, b4 > 0 and b1 b2 b3 [b23 + b21 b4 ] > 0

(6.3.22)

136

Chapter 6

We consider each coecient of the Characteristic Polynomial. For rst coecient

b1 , It is clear that b1 > 0.
For second coecient b2 , it includes a two negative terms, a10 a6 and a7 a3 . Adding
the two negative terms to the following two positive terms from b2 , a12 a5 and a8 a1 ,
we have,
a12 a5 a10 a6 = k9 k4 k5 + k6 [k1 k8 + k9 ] > 0,
a8 a1 a7 a3 = k1 [

2
+ k7 + k2 ] + k2 k7 > 0.

Thus, the second coecient b2 is positive.

The third coecient b3 has four negative terms, a10 a6 a8 , a10 a6 a1 , a12 a7 a3 and
a7 a3 a5 . Adding these four negative terms to the following four positive terms from
the expression b3 , a12 a5 a8 , a12 a5 a1 , a12 a8 a1 and a5 a8 a1 , we obtain,
[a8 + a1 ][a12 a5 a10 a6 ] + [a12 + a5 ][a8 a1 a7 a3 ] = [a8 + a1 ][k9 k4 k5 + k6 (k1 k8 + k9 )] +
[a12 + a5 ][k1 ( 2 + k7 + k2 ) + k2 k7 ] > 0.
Thus, the third coecient b3 is also positive.
The fourth coecient b4 has two negative terms, a10 a6 a8 a1 and a12 a7 a3 a5 . Adding
these two negative terms to the following positive terms from the expression b4 ,
a12 a5 a8 a1 and a10 a6 a7 a3 we obtain,
[a12 a5 a10 a6 ][a8 a1 a7 a3 ] = [k9 k4 k5 + k6 (k1 k8 + k9 )][k1 ( 2 + k7 + k2 ) + k2 k7 ] > 0,
Thus, the fourth coecient b4 is positive. Since all coecients of equation (6.3.21)
are a positive then the roots of this equation must be a negative if they are real.
However, this result is not true if the eigenvalue is complex. In this case, we need
to satisfy the last condition of Routh-Hurwitz Criterion (b1 b2 b3 [b23 + b21 b4 ] > 0) to
conclude that all eigenvalues of the characteristic equation z() have negative real

Chapter 6

137

parts. We have been unable to establish this analytically. However, it appears that
this condition is satised numerically using our parameters in section (6.2). Thus,
we believe that the no-washout branch is stable whenever it is physically meaningful.
The complexity of using the Routh-Hurwitz Criterion prevent us from attempting
stability analysis for a cascade containing more than two reactors. This complexity
has been reported by May [180]. Thus, the stability of the no-washout branch for of
the three, four and ve reactor cascades, was studied numerically. We observe that,
for our parameters, all eigenvalues along the no-washout solution are negative. Thus,
the no-washout branch, if it exist, is stable using our parameter in section (6.2).

6.3.4

Reactor Cascade With Recycle Around the whole Cascade

In this section, we particularly consider an analysis of a two reactor cascade and

compare the performance of a two-reactor cascade with recycle, conguration (1),
against that without recycle for three dierent values (minimum, moderate and
maximum) of the concentration factor (C): C = Cmin = 1, C = 2 and C = Cmax . In
the previous section, we established that the critical value of the residence time (w )
for the two reactor cascade without recycle is greater than the critical value of the
residence time for the two reactor cascade with recycle. That is, only the washout
branch is stable below these critical values, and therefore process failure occurs.

138

Chapter 6

6.3.4.1

Case C = Cmin =1

Figure (6.2) shows the euent concentration leaving a two reactor cascade, as a
function of the total residence time when the concentration factor is at its minimum (C = Cmin = 1) whereby it is seen from equation (6.2.7) that in this case the
maximum value of the recycle parameter is innity (Rmax = ).
It can be seen from gure (6.2) that there are three distinct regions for each value
of R (R > 0). In the rst region, washout occurs in the two reactor cascade. As the
value of R increases from 0 to 1, the value of the washout point (w ) decreases from
2/(1 Kd ) to zero. In this region the performance of the cascade with recycle is
identical to that of a cascade without recycle. In the second region, the performance
of the two reactor cascade with recycle is superior to the performance of the two
reactor cascade without recycle. Thus in this region the euent concentration can
be minimised, at a specied total residence time, by selecting an optimal value of the
recycle ratio, denoted by ROpt . The dependence of the euent concentration upon
the recycle ratio for a residence time in this region is illustrated in gure (6.3(a)).
For a specied value of R the second region ends when the corresponding euent
concentration curve with recycle intersects the euent concentration curve without
recycle. Using the data in table (6.1) we see that the second region corresponding
to the curves R = 0.5 and R = 1 in gure (6.2) ends at t = 2.5 and t = 2.4
respectively. The detail of intersection points for the cases R = 0.5 and R = 1 are
given in table (6.1). For total residence times beyond the intersection points, the
eect of recycle ratio (R) on the reactor performance is negative and vice-versa.
In the third region, the performance of a two reactor cascade with recycle is inferior

139

Chapter 6

to the performance without recycle, i.e., recycle has a negative eect on the euent

Dimensionless substrate concentration ( S*e)

concentration. Figure (6.3(b)) illustrates that in the third region where ROpt = 0.

No recycle
R=0.5
R=1.0
R=ROpt

0.1

0.01

0.001

0.0001
0

Dimensionless total residence time ( )

Figure 6.2: Euent concentration in a cascade of a two reactors with recycle around
the whole cascade. Parameter value: C = Cmin = 1.

6.3.4.2

Case C =2

Figure (6.4) shows steady-state diagrams of a two reactor cascade when C = 2.

When C > 1, equation (6.2.7) shows that there is a restriction on the maximum
value of the recycle parameter.
When 0 < R < Rmax = 1, we observe the same three distinct regions that were
described in section (6.3.4.1). (When R = Rmax , the rst of these regions does not
exist.) It can be seen that in the second region the performance of the two reactor
cascade with recycle increases as the recycle ratio (R) increases and the optimal
value of the recycle is R = Rmax = 1. The intersection points for the cases R = 0.5

140

Dimensionless substrate concentration ( S*2)

Dimensionless substrate concentration ( S*2)

Chapter 6

0.1

0.01
0

4
6
Recycle Ratio (R)

10

80

100

(a) = 2 (Region 2)

0.014
0.012
0.01
0.008
0.006
0.004
0.002
0
0

20

40
60
Recycle Ratio(R)

(b) = 8 (Region 3)

Figure 6.3: Diagram showing the eect of the recycle ratio on the euent concentration (S2 ). In gure (6.3(a)), the lled-in circles is the optimum value of the
recycle ratio (S2 = 0.0638, ROpt = 1.08). In gure (6.3(b)), the optimum value of
the recycle ratio is ROpt = 0. Parameter value: C = Cmin = 1.

141

Chapter 6

and R = Rmax = 1 are shown in table (6.1). In the third region the optimum value
of the recycle ratio is zero, i.e. the use of a settling unit in this region increases the
euent concentration.
Figures (6.5(a)) demonstrates the eect of a settling unit upon the euent concentration (S2 ) leaving a two reactor cascade for xed total residence time in the second
region. For suciently small values of R (0 R Rcr = 0.21) washout occurs as
< w . For suciently large values of the recycle parameter (R > Rcr = 0.21) the
euent concentration decreases as the recycle ratio increases; the minimum euent
concentration is obtained when ROpt = 1.
Figure (6.5(b)) demonstrates the eect of a settling unit in the third region. In this
region the euent concentration is an increasing function of the recycle ratio, i.e
ROpt = 0 and the euent concentration is minimised by turning the settling unit
o.

6.3.4.3

Case C = Cmax = 1 + 1/R

Figure (6.6) shows steady-state diagrams of a two reactor cascade for the special
case in which the concentration factor is maximized (C = Cmax = 1 + 1/R) for two
values of the recycle ratio (R = 0.5 and R = 1). There are two distinct regions.
In the rst region the euent concentration in a cascade with recycle is lower than
that without recycle. The euent concentration leaving the cascade with R = 0.5
is lower than that leaving the cascade with R = 1.0. The intersection points for
the cases R = 0.5 and R = 1 are shown in table (6.1). In the second region, for
suciently large values of the total residence time, the euent concentration leaving
the reactor cascade with recycle is greater than that without recycle. In this region

142

Dimensionless substrate concentration ( S*e)

Chapter 6

No recycle
R=0.5
R=1.0
R=ROpt

0.1

0.01
R=ROpt=1
0.001
R=ROpt=0
0.0001
0

1
2
3
4
5
6
7
8
9
Dimensionless total residence time ()

Figure 6.4: Euent concentration in a cascade of a two reactors with recycles around
the whole cascade. Parameter value: C = 2, Rmax = 1.

the performance of the cascade is maximised by removing the settling unit.

6.3.4.4

Summary

For each of the three cases considered there are two regions of practical interest. In
the rst region, the use of a settling unit improves the performance of the reactor
cascade. In the second region the operation of a settling unit worsens the performance of the reactor cascade.
Extensive numerical investigations (not shown) an n-reactor cascade for (n = 3, 4
and 5) showed the presence of the same two regions of practical interest that arise
for specic case of the two reactor cascade. The same conclusions were also reached
by using Monod model as explained in appendix (A.1). This latter investigation
was suggested by a referee of our paper [179].

143

Dimensionless substrate concentration ( S*2)

Dimensionless substrate concentration ( S*2)

Chapter 6

0.1

0.01
0

0.2

0.4
0.6
Recycle Ratio(R)

0.8

0.8

(a) = 1 (Region 2)

0.00125
0.0012
0.00115
0.0011
0.00105
0.001
0.00095
0.0009
0.00085
0.0008
0

0.2

0.4

0.6

Recycle Ratio (R)

(b) = 8 (Region 3)

Figure 6.5:

The eect of the recycle ratio upon the euent concentration (S2 ).

Parameter value: C = 2, Rmax = 1. In gures (6.5(a) and 6.5(b)), the optimum

value of the recycle ratio is ROpt = 1 and ROpt = 0 respectively.

144

Dimensionless substrate concentration ( S*e)

Chapter 6

No recycle
R=0.5
R=1.0

0.1

0.01

0.001

0.0001
0

1
2
3
4
5
6
7
8
9
Dimensionless total residence time ()

Figure 6.6: Euent concentration in a cascade of a two reactors with recycles around
the whole cascade. Parameter value: C = Cmax = 1 + 1/R.

6.3.5

The eect of changing the concentration factor (C)

Figure (6.7) shows the eect of varying the concentration factor on the euent concentration leaving a two reactor cascade for a xed value of the recycle ratio (R =
0.5). For a value of R = 0.5, we have from equation (6.2.6) that Cmax = 3. At suciently low values of total residence time, it is apparent that the performance of the
reactor cascade improves continually as the concentration factor increases. However,
for each value of the concentration factor, there is a critical value for the residence
time at which the performances of the two reactor cascade with and without recycle
are identical. These values are given in table (6.1). This critical value increases
as the concentration factor increases. If the residence time is greater (lower) than
the critical value, the performance of the two reactor cascade with a settling unit is
worse (superior) than its performance without recycle.

145

Chapter 6

Furthermore, the eect of the concentration factor (C) on the performance of the
two-reactor cascade decreases for larger values of total residence time (t 9 )
as shown in gure (6.7). This agrees with our results presented in section (6.3.7)
showed that in the asymptotic limit of large residence times the euent concentration leaving the nal reactor is independent of the concentration factor (C) (a
second-order eect). This behavior was observed for an n-reactor cascade (n = 3, 4

Dimensionless substrate concentration ( S*e)

and 5) as well.

No recycle
C=1
C=2
C=3

0.1

0.01

0.001

0.0001
0

1
2
3
4
5
6
7
8
9

Dimensionless total residence time ( )

Figure 6.7: Euent concentration in a cascade of a two reactors with recycle around
the whole cascade. The parameter value: R = 0.5.

6.3.6

Comparison of a two-reactor cascade and a three-reactor

cascade

Figure (6.8) compares the performance of two-reactor and three-reactor cascades

with and without recycle. For the cascades with recycle, the performance of the

146

Chapter 6

Table 6.1: Intersection points in gure i of the recycle curve with the no recycle
curve.
i

R = 0.5

R=1

6.2

0.02

0.03

0.003

0.003

6.6

0.001

0.003

1+1/R

??

0.00022

0.00016

6.2

2.5

2.4

4.3

4.5

6.6

6.3

4.4

1+1/R

A.1

46.60

68.11

C=1

C=2

C=3

6.4
S2

6.4
t

6.7

S2

0.028

0.003

0.0014

6.7

2.5

4.4

6.4

147

Chapter 6

three-reactor cascade is superior to the performance of the two-reactor cascade.

However, comparing the euent concentration leaving a two-reactor cascade with
recycle (Line 2A) against that leaving the three-reactor cascade without recycle (Line
3B) shows that there is a critical value of the total residence time in which the
performances of these two congurations are identical. This value is (t = 3.4, Se =
0.004). If the desired euent concentration is higher (lower) than this value, then
the performance of the two-reactor cascade with recycle is superior (worse) to the

Dimensionless substrate concentration ( S*e)

performance of the three-reactor cascade without recycle.

Figure 6.8:

Two reactors.2A
Three reactors.3A
Two reactors.2B
Three reactors.3B

0.1
0.01
0.001
0.0001
1e-005
0

1
2
3
4
5
6
7
8
9

Dimensionless total residence time ( )

Comparison of performance in a two reactor cascade against a three

reactor cascade. The parameter value: (A) R = 0.5, C = Cmax and (B) No recycle .

6.3.7

Large residence time approximations

In this section we provide approximations at large total residence times for the
concentrations of the substrate and microorganisms. These explain the interesting

148

Chapter 6

behavior that was observed in sections (6.3.4-6.3.6) regarding the eect of recycle on
the reactor cascade. We obtain the following approximations by using Taylor series,
)i
(
)
n
1
i1
= Ai1
(1 + R) + O
, (i = 1, ..., 5).
(i+1)
t (1 Kd )
t
(
)i1 (
)i
(
)
(1 + R)
n
1

Xi = Bi1
+O
(i = 1, ..., 5).
(i+1)
(1 Kd )
t Kd
t
(

Si

(6.3.23)
(6.3.24)

where,
A1 =

d1
,
e1

A2 =

B1 = d1 ,

d1 d2
,
e1 e3

A3 =

d1 d2 d3
,
e21 e2 e3

A4 =

d1 d2 d3 d4
,
e21 e2 e3 e4

d1 d2
d1 d2 d3 d4
d1 d2 d3 d4
,
B3 =
,
B4 =
,
e1
e1 e3
e21 e2 e3
(
)
(
)
e2 = Kd 2 2 + 1 , e3 = Kd 2 2 + 1 + Kd ,

B2 =

e1 = (Kd + 1) ,

(
)
e4 = Kd 4 4 + Kd 3 3 + Kd 2 2 + Kd + 1 , d1 = (1 Kd + Kd ) ,
(
)
d2 = 1 Kd + Kd (1 Kd ) + Kd 2 2 ,
(

d3 = 1
(

d4 = 1

Kd
Kd

Kd

Kd

Kd )

Kd )

(1
(1

2
Kd

2
Kd

Kd )

Kd )

(1
(1

Kd 3 3

3
Kd

(1

Kd )

Kd 4 4

)
.

Equations (6.3.236.3.24) give the concentrations of the substrate and microorganisms in the rst ve reactors of a n cascade. These expressions hold for any value
of n.
From equation (6.3.23) the following assertions can be made:
The substrate concentration in each reactor decreases to zero at large values
of the total residence times.
The euent concentration leaving a n reactor cascade (n = 1, ..., 5) is independent of the concentration factor (C). This is due the fact that under high
total residence time the microorganism concentration in the nal reactor of

149

Chapter 6

a n reactor cascade (n=2,...,5) is proportional to

1
n
t

which is much smaller

than the microorganism concentration in the rst reactor that is proportional

to

1
t

(see equation (6.3.24)). Thus, the returned microorganism concentra-

tion from the nal reactor to the rst reactor multiplied by the concentration
factor (C), see the term CXn in equation (6.2.9), has a negligible eect on the
microorganism concentration in the rst reactor. This is only second orders
eect.
The euent concentration leaving an n reactor cascade (Sn ) is proportional to
(1+R)n1 for n = 2, .., 5. Therefore, the addition of a settling unit to a cascade
without a settling unit increases the euent concentration by a multiplicative
factor of (1+R) at suciently large total residence times.
At leading order the euent concentration leaving an n reactor cascade (Sn )
is proportional to (1/ n ) (n = 1,..,5). The addition of an extra reactor reduces
the euent concentration by a multiplicative factor of

1
t

for suciently large

total residence times.

(Note that for a single reactor, the euent concentration at high residence times is
independent of both R and C, as noted by [101]).
In section (6.3.4) we found the surprising result that recycle around the whole reactor cascade increases the euent concentration at suciently large total residence
times. This result is substantiated by the asymptotic formula (6.3.23) which shows
that the use of a settling unit around the whole reactor cascade at large values of
total residence time reduces its performance. Therefore, in this limit the euent
concentration is minimized by removing the settling unit.

Chapter 6

150

At large values of the total residence time it is interesting to compare the performances of the reactor cascade with recycle around the whole cascade with a reactor
cascade employing recycle only around the nal reactor, conguration (2). In chapter (5), published in [174], we showed that, for conguration (2), the performance of
the reactor cascade is optimized with perfect recycle at large values of total residence
time. Thus, we conclude that, at large values of total residence time, if a settling
unit is used, it should recycle around the nal reactor rather than around the whole
cascade. However, asymptotic analysis for conguration 2 shows that at high total
residence times this is a only second order eect, see equation (5.4.90).

6.4

Discussion

The main aim of this section is to compare the performance of a two-reactor cascade
in conguration (1) with the performance of a two-reactor cascade performance in
conguration (2). Figure (6.9) shows the euent concentration leaving both a cascade with perfect recycle around the ith reactor (i = 1, 2, conguration 2) and a
reactor cascade with recycle around the whole cascade (conguration 1) over a range
of total residence times. It should be noted that the curves (B and C) are identical
to graphical accuracy and the choice of the parameter values (R = 0.5, C = 3) gives
perfect recycle around the ith reactor (conguration 2). Throughout our comparison
we will ensure that the two congurations possess the same biochemical parameters.
In gure (6.9), there are three distinct regions. In the rst region (0 < t < 6.1), the
performance of conguration (1) is superior to the performance of congurations (2).
In the second region (6.1 < t < 6.31), we can see that for this small range of to-

Chapter 6

151

tal residence time, the performance of conguration (1) is inferior (superior) to the
performance of conguration (2) when the recycle unit is place around the second
(rst) reactor. In the third region (t > 6.31), the performance of conguration (1)
is inferior to the performance of congurations (2). The intersection points for the
performances of the congurations (1 and 2) are shown in table (6.2). This clearly
explains that the desired euent concentration must be considered when choosing
the reactor conguration. Keeping this is mind, conguration (1) is ideal for the
rst region, and conguration (2) for the third region.
Table (6.3) shows the total residence time required to achieve specied euent concentrations for a two reactor cascade employing either conguration (1) or (2). For
conguration (2) it is assumed that only one settling unit is deployed. Consequently
conguration (2) has two columns in table (6.3), representing the cases when the
settling unit is placed around either the rst or second reactor respectively.
Table (6.3) shows that if the desired euent concentration is either Se = 0.1 or
Se = 0.05 then a two reactor cascade with perfect recycle around the whole of
the cascade (conguration 1) outperforms either cascade setup utilising conguration (2). When the target euent concentration is Se = 0.01, the performance of
conguration (1) is inferior (superior) to the performance of conguration (2) when
an optimized recycle unit is place around the rst (second) reactor. When the target
euent concentration is Se = 0.001, the performance of conguration (1) is inferior
to both setups using conguration (2).
We conclude that for target euent concentrations Se = 0.05, or higher, the total
residence time is minimized using a reactor conguration employing optimized recy-

152

Chapter 6

Table 6.2: Intersection points in gure (6.9) of the conguration (1) with the conguration (2), R = 0.5 and C = 3, A=Curve A, B=Curve B, C=Curve C in gure (6.9).
A=B

A=C

S2

0.001462

0.0015

6.31

6.1

cle around the whole cascade (conguration 1). However, when the target euent
concentrations is Se = 0.01, or less, it is better to have optimized recycle either
around the rst or second reactor (conguration 2).
An interesting question to ask is given an existing cascade of two reactors whether
it is better to add an optimized settling unit or an additional reactor. The data in
tables (6.3) and (6.4) reveals that when the designed euent concentration is greater
than Se = 0.01 that it is better to deploy the settling unit. However, if the specied
euent concentration is smaller than Se = 0.001 then an additional reactor should
be added.
Finally, it is interesting to return to the question as to whether recycle is a signicant
tool to reduce the euent concentration. When the desired euent concentration is
Se = 0.1 we see from tables (6.3 & 6.4) that in the best case the residence time required for a cascade of two reactors with a settling unit is 25% of the value without a
settling unit. The corresponding values for euent concentrations of Se = 0.01 and
Se = 0.001 are 63% and 86% respectively. Thus as the desired euent concentration
reduces, the impact of a settling unit becomes weaker and weaker.

153

Dimensionless substrate concentration ( S*e)

Chapter 6

A
B
C

0.1

0.01
(B and C)
0.001

0.0001
0

Dimensionless total residence time ()

Figure 6.9: Comparison of a two reactor cascade performance employing either conguration (1) or conguration (2). (A) Recycle around whole cascade, (B) Recycle
around rst reactor and (C) Recycle around second reactor. The parameter value:
R = 0.5 C = Cmax .

154

Chapter 6

Table 6.3: Dimensionless total residence times required to reach a specic dimensionless euent concentration in a cascade of two reactors, C = 2. Parameter value:
= 0.1019, Kd = 0.0141
Conguration (1)

Conguration (2)
R1 = ROpt , R2 = 0

Se

ROpt

0.1

0.5561

0.05

0.9262

0.01

R1 = 0, R2 = ROpt
t

ROpt

0.6796

0.98 1.861

0.932

0.97 2.178

2.5245

2.005

0.97 3.100

0.001 7.41202

6.395

0.97 7.382

ROpt

Table 6.4: The total residence times in a cascade of n reactors without recycle
that is required to reach a specic euent concentration. Parameter value: =
0.1019, Kd = 0.0141
Se

n=1

n=2

n=3

0.1

1.944

2.16

3.05

0.01

12.10

3.18

3.23

0.001

112.19

7.41

4.30

Chapter 6

6.5

155

Conclusion

We have investigated how the use of a settling unit eects the behavior of an nreactor cascade (n = 2, ..., 5). The settling unit is placed after the nal reactor and
the euent stream from the settling unit is recycled back into the rst reactor (conguration 1). We used the Contois specic growth rate model with the addition
of a microorganism decay coecient. The results obtained here were illustrated by
considering the anaerobic digestion of ice-cream wastewater, using kinetic values
from [39].
The stability of the washout branch was determined as a function of the total residence time. Asymptotic solutions were found for the no-washout solution branch.
The analysis of the asymptotic solutions has shown that at high total residence
times, the euent concentration (Sn ) is minimised by setting the recycle parameter
to zero.
For a xed value of the recycle ratio R and the concentration factor C, there is a
critical value of the total residence time. If the total residence time is below the
critical value then the settling unit improves the performance of the n reactor cascade (n = 2, ..., 5) compared to that of a cascade with a settling unit whereas if the
total residence time is above the critical values the performance of the n reactor
cascade (n = 2, ..., 5) is reduced compared to that of a cascade without a settling
unit.
We compared the performance of the two reactor cascade using conguration (1)
with the performance of the two reactor cascade using conguration (2). We found
that for target euent concentrations Se = 0.05, or higher, the total residence time

Chapter 6

156

is minimised using a reactor conguration employing optimised recycle around the

whole of the cascade (conguration 1). However, when the target euent concentrations is Se = 0.01, or lower, it is better to have optimised recycle around the rst
or second reactor (conguration 2).

Chapter 7
Analysis of a chemostat model
with variable yield coecient

7.1

Introduction

It has been suggested that, the experimentally found oscillatory behaviour of microbial population can be explained by allowing the yield coecient to be dependent on
the substrate concentration [181,182] and not when the yield coecient is constant.
A number of theoretical studies have examined the consequences of this assumption, these are discussed in chapter (3). The most relevant studies to this chapter
are [3, 49, 129] which used the Contois specic growth rate without a decay coefcient. However, the microorganisms inside biological wastewater treatment must
eventually die as discussed in chapter (3). Therefore, in this chapter, we extend the
study [49] by taking into account the death rate of the microorganisms.
We nd the steady-state solutions, determine their stability and obtain asymptotic
solutions in the limit of high residence times. We determine the conditions for
157

158

Chapter 7

washout to occur and the regions of parameter values for which periodic behaviour
can be generated. The main contributions of this study are the following. 1) Studying the eect of the death rate with a variable yield coecient on the bioreactor. 2)
Studying a generalised reactor model, including a continuous ow bioreactor and an
idealised continuous ow membrane reactor as limiting cases. A major part of this
chapter has been published as a journal article [130].

7.1.1

The dimensional model

The model equations are given by:

dS
(S, X)
= F (S0 S) V X
,
dt
Y (S)

(7.1.1)

dX
= F X + V X(S, X) Kd V X.
dt

(7.1.2)

Specic growth rate (Contois),

(
(S, X) = max

S
Ks X + S

)
,

(7.1.3)

The residence time,

V
,
F

(7.1.4)

The yield coecient,

Y (S) = + S,

(, > 0).

(7.1.5)

All parameters in the model are strictly non-negative. For more details of the
assumptions and denition of model parameters, we refer to chapter (4) and appendix (D) respectively.

159

Chapter 7

7.1.2

The dimensionless model

Using the transformations introduced in (5), equations, (7.1.1, 7.1.2) can be written
into the following dimensionless form,
dS
1
S X

=
(1

S
)

,
dt
( + S )(S + X )

(7.1.6)

dX X
S X
=
+
Kd X .
dt

S + X

(7.1.7)

All dimensionless parameters in the model are strictly non-negative. For more details
of the assumptions and denition of model parameters, we refer to chapter (4) and
appendix (D) respectively.

7.2

Results

In section (7.2.1), we show that there are two physically meaningful steady state
solution branches. These are a washout branch and a no-washout branch. In section (7.2.2), the stability of the washout solution is determined. The Hopf bifurcation on the no-washout state is discussed in section (7.2.3). In section (7.2.4), the
steady state diagrams are presented. The XPP software package [183] was used to
obtain the steady-state diagrams. The standard representation is used: solid lines
are stable steady states; dotted lines are unstable steady states; lled-in squares are
Hopf bifurcation points; open circles are unstable periodic orbits and lled-in circles
are stable periodic solutions.

160

Chapter 7

7.2.1

Steady-state solution branches

The steady state solutions of the system of equations (7.1.6) and (7.1.7) are the
following.
The washout branch:
(S , X ) = (1, 0).

(7.2.8)

The no-washout branches:

(
(S , X ) =

c
c , ( + S )(1 S )
S
+ Kd

c = b +
S
2a

)
,

b2 4ac
,
2a

(7.2.9)

(7.2.10)

c 2 + bS
c + c = 0,
aS

(7.2.11)

where the coecient a, b, and c are dened by

a = ,

b = (1 Kd ) + ,

c = .

Note that the coecient a is strictly positive while the coecient c is strictly negative. This means that the solution (7.2.11) for S is always positive. From equation (7.2.9) the steady state solution is positive when 0 < S < 1. The residence
time at which we have S = 1 is given by,
= cr =

.
1 Kd

(7.2.12)

Thus, the no-washout steady state is physically meaningful when the residence time
satises
> cr =

1 Kd

(7.2.13)

161

Chapter 7

Dierentiating Equation (7.2.11), we obtain,

dS (1 Kd ) S
=
<0
d
2aS b

because

S >

b
.
2a

(7.2.14)

Thus, the steady-state euent concentration is a decreasing function of the residence time. At large residence times, the approximate value of the microorganism
concentration and substrate concentration are:

7.2.2

1
+ O( 2 ),

Kd

1
+ O( 2 ),

(1 Kd )

0 < Kd < 1,

(7.2.15)

0 < Kd < 1.

(7.2.16)

Stability along the washout branch

Under washout conditions the euent concentration is equal to the inow concentration. Thus this state of operation must be avoided. The Jacobian matrix for the
system (7.1.6 and 7.1.7) is given by,

X ( S 2 X )
1 + ( + S )2 (X +S )2

J(S , X ) =

2
X
(X +S )2

2
S

( + S ) (X +S )2
Kd +

S 2

(7.2.17)

(X +S )2

The eigenvalues of the Jacobian matrix (7.2.17) at the washout state are given by,
1 =

1
< 0,

2 =

+ (1 Kd )
.

Hence, the washout branch is stable when:

<

.
1 Kd

(7.2.18)

Equation (7.2.18) shows that a perfect membrane ( = 0) cannot exhibits washout,

a good membrane ( << 1) only exhibits washout at very small residence times.

162

Chapter 7

7.2.3

Hopf bifurcation on the no-washout state

In this section, we investigate the possibility of Hopf bifurcations occurring along

the no-washout state. Evaluating the Jacobian matrix along the no-washout state,

we nd that

J11 J12

J(S , X ) =

J21 J22

(7.2.19)

where,

J11

(
)
X S 2 X
1
= +
,

( + S )2 (X + S )2

J12

S
=
,
( + S ) (X + S )2

J21

X
=
,
(X + S )2

J22 =

X S
.
(X + S )2

The conditions for both eigenvalues to be zero are J11 J22 J21 J12 = 0 and J11 +J22 =
0 [184]. Some algebra shows that
J11 J22 J21 J12 =

f (S , X )X S
> 0,
(X + S )3 ( + S )2

(7.2.20)

f (S , X ) = 2 S 3 + ( X + 2 ) S 2 + (2 X + ) S
+ X ( + ).

Thus, the rst condition cannot be satised for a physically meaningful solution
and as consequence a double-zero eigenvalue degeneracy cannot occur. The physical
signicance of a Hopf bifurcation is that it is one of the two main mechanisms by
which periodic behaviour is generated in systems of nonlinear dierential equations,
the other being a double zero eigenvalue bifurcation. The conditions for a Hopf

163

Chapter 7

bifurcation to occur are J11 J22 J21 J12 > 0 and J11 + J22 = 0 [184]. We have shown
that the rst condition is always satised. The values of the residence time at which
Hopf bifurcations occur, where J11 + J22 = 0, are found to satisfy the equations,
= r1 X + r2 X + r3 = 0,
2

(7.2.21)

[
]
r1 = b1 2 2 (1 + b1 )2 + b1 ,
[
]
r2 = b1 2 (1 + b1 )2 + b1 (2 1) ,
[
]
r3 = (1 + b1 )2 + (b1 + 1) ,
b1 =

+ Kd
.
(1 Kd )

Note that when > cr then b1 > 0. Hence the coecients r1 and r3 are positive.
From these equations, the following results can be obtained.
1. When the yield coecient is constant (i.e. = 0), the trace is negative and the
Hopf bifurcation conditions (7.2.21) are not satised. Under these conditions,
the no-washout branch is always stable. Thus the variability of the yield
coecient is a signicant condition for producing a natural oscillation.
2. The coecient r2 is positive when 1/2 and as result equation (7.2.21) is
the sum of positive terms and cannot therefore have zero value. Thus the Hopf
bifurcation cannot occur when,
1
.
2

(7.2.22)

3. For a membrane reactor ( = 0), the condition given in (7.2.21) simplies

considerably. After some algebra the solution of equation (7.2.21) is given by
X =

r2

(1 Kd )[Kd (1 Kd ) (4 Kd ) 4 ]
.
2r1

(7.2.23)

164

Chapter 7

This solution is complex when Kd 4 . Thus, in a membrane reactor, Hopf

bifurcations cannot occur when,
Kd 4

(7.2.24)

A double Hopf bifurcation is where two Hopf points annihilate each other in an
unfolding diagram (known as H21 degeneracy). The conditions for the double Hopf
bifurcation to occur are [184],
J11 J22 J21 J12 > 0,

= 0,

d
= 0.
d

(7.2.25)

When = 0.01 and Kd = 0.01 then the H21 degenerate conditions are satised
when,
( , ) = (13.328, 14.741),

( , ) = (0.022, 1.077).

(7.2.26)

Natural oscillations cannot occur in the regions 0.022 > and > 13.328. They
do occur in the region 0.022 < < 13.328.

7.2.4

Numerical results

Figure (7.1) shows two steady-state diagrams. Each steady-state diagram contains
two curves for the washout (S = 1) and the no-washout states (S < 1).
Our choice of the values of the parameter in gure (7.1) are motivated by the
values in equation (7.2.26). In gure (7.1(a)) there are two Hopf bifurcation points
because the value of is in the region 0.022 < < 13.328. Between the two Hopf
bifurcation points, the no-washout branch is unstable and there are stable periodic
solutions. In gure (7.1(b)) there are no periodic solutions as the value of is

165

Dimensionless substrate concentration ( S*)

Chapter 7

1.2
1
0.8
0.6
0.4
0.2
0
-0.2
0

10

10

Dimensionless residence time ( )

Dimensionless substrate concentration ( S*)

(a) = 1.05

1.2
1
0.8
0.6
0.4
0.2
0
-0.2
0

Dimensionless residence time ()

(b) = 0.01

Figure 7.1: Dimensionless euent concentrations as a function of the dimensionless

residence time. Parameter values: =1, Kd = 0.01, = 0.01. All the symbols are
explained in section (7.2).

166

Chapter 7

outside the critical region and the no-washout solution is always stable provided
> cr . Figure (7.2) is an unfolding diagram for the two Hopf bifurcations in

16
14
12

10
8
6
4
2
0
0

10

12

14

Figure 7.2:

Unfolding diagram showing the Hopf bifurcation locus. Parameter

values: =1, Kd = 0.01, = 0.01.

gure (7.1(a)) in the plane. This gure shows the values of the residence
time at which a Hopf bifurcation occurs as a function of . It can be seen that
Hopf bifurcation points are only found in the region (0.022 < < 13.328) and
they destroy each other at the H21 degeneracy given by equation (7.2.26). In the
region between the H21 degeneracies, periodic oscillations can occur. For either
large or small values of the parameter , including the constant yield case ( = 0),
periodic behaviour does not occur. Periodic solutions are undesirable in this system
because the average euent concentration of a periodic solution is higher than the
corresponding value of the unstable steady-state solution.
Figure (7.3) shows the H21 degeneracy curves, dened by (7.2.25), in the

167

Chapter 7

100
10

1
0.1
K**d=0.00
K*d=0.01
K*d=0.20
K d=0.40

0.01
0.001
0

0.02

0.04

0.06

0.08

0.1

(a)

330

K*d=0.2

100
50
20
10
5
3
1
0.36
0.049

0.052

0.055

0.058

(b) Blow up of figure (7.3(a))

Figure 7.3: H21 degeneracy unfolding diagram. Parameter value: =1.

168

Chapter 7

plane for dierent value of Kd . Figure (7.3(a)) shows that there is a critical value
of the death rate. This value is found to be Kd = 0.2872. If 0 < Kd < 0.2872 then

the H21 curve has two limit points (1,Lp

> 2,Lp
). Depending upon the value for

there may be no, one, two or three H21 degeneracies. For example, gure (7.3(b))
shows that when Kd = 0.2 there are two limit points and accordingly, in the region
between 0.049 < < 0.05 there are three H21 degeneracies. When Kd > 0.2872 the
H21 degeneracy curve has no limit points and there is always one H21 degeneracy,
see the curve on gure (7.3(a)) when Kd = 0.4. If the value of is to the right
the H21 degeneracy curve then there are no Hopf bifurcation points on the steadystate diagram. If it is to the left, then the steady-state diagram contains two Hopf
bifurcations.
The conditions for the limit point occurring along the curve in the gure (7.3) which
is produced from the following two algebraic equations,

= 0,

(7.2.27)

d
= 0.
d

(7.2.28)

is given by [184],

d
d2
d
d2
=
=
0
with
=

0
and
= .
2
d
d
d
d 2

The last condition of nding the limit point,

d
d

(7.2.29)

= 0, can be found by regarding

and as function of and then take implicit dierentiation of the two system (7.2.27, 7.2.28). Figure (7.4) summarizes the information about the limit points
that occur along H21 unfolding curve in Kd plane. It can be seen that the limit
points are possible for all values of Kd less than the critical value Kd = 0.2872.

169

Chapter 7

0.07

Limit point locus

0.06

0.05

H22

0.04
0.03
0.02
0.01
0
0

0.05

0.1

0.15
K *d

0.2

0.25

0.3

Figure 7.4: Diagram showing the domain of the limit point locus on the H21 curve
unfolding in parameter spaces (Kd , ). Parameter value: =1.

7.3

Conclusion

We have analysed the behaviour of a bioreactor using the Contois specic growth rate
model with a variable yield coecient. We extended earlier work [49] by including
the death rates of the microorganisms and by using a generalized bioreactor model
which includes as special cases the continuous ow and membrane reactors.
We determined a critical value of the residence time at which the washout solution
loses stability. This critical value is the minimum residence time that can be used
to operate a bioreactor. Our investigation of the reactors periodic behaviour has
shown that there is a wide range of parameter values for which natural oscillations
occur. The analysis of the eect of the death rate on the oscillatory behaviour has
shown that there is a critical value of the death rate (Kd = 0.2872). If the death
rate is less than this critical value then there may be between zero and three H21

Chapter 7

170

degeneracies. If the value of the death rate is higher than the critical value then one
H21 degeneracy can occur. We have also shown that the range of the yield constant
coecient for which oscillations are possible increases for high values of the death
rate. However, the Hopf bifuractions are impossible for 0.5. We found that
in the region of periodic behaviour the stable periodic solution has a higher average
euent concentration than the unstable steady-state solution and consequently the
reactor performance decreases.

Chapter 8
Analysis of a chemostat model
with substrate inhibition

8.1

Introduction

As described in chapter (2), there are a number of industrial processes, in particular

wastewater treatment, which use Contois kinetics [13] model for cell-growth rate
since Contois model is more accurate than Monod kinetics. Thus, it is reasonable to
assume that the Contois model can be improved as the Monod model by an inclusion
of similar substrate inhibition term (S 2 /KI ).
In this chapter we extend the study [101] to include substrate inhibition by considering the behavior of an unstructured model in a generalized bioreactor model that
encompasses a continuous ow bioreactor and an idealised continuous ow membrane reactor as limiting cases. The specic growth rate of the biomass is assumed
to be given by a Contois expression with the inclusion of a substrate inhibition term.
We nd the steady-state solutions, determine their stability and obtain asymptotic
171

172

Chapter 8

solutions in the limit of high residence times. We determine the conditions for
washout to occur. The main contributions of this study is investigating the static
bifurcations of the Contois growth model with the inclusion of a substrate inhibition
term.
The dimensional and dimensionless forms of our model are stated in sections (8.1.1)
and (8.1.2), respectively.

8.1.1

The dimensional model

The model equations for a single reactor with substrate inhibition are given by,
dS
(S, X)
= F (S0 S) V X
,
dt

dX
V
= F (X0 X) + V X(S, X) Kd V X.
dt
V

(8.1.1)
(8.1.2)

The Contois expression with substrate inhibition is given by

(
= m

S
Ks X + S +

)
S2
KI

(8.1.3)

The residence time,

V
.
F

(8.1.4)

It can be noted that in the limit when KI that we obtain the standard Contois
model. All parameters in the model are strictly non-negative. For more details of
the assumptions and denition of model parameters, we refer readers to chapter (4)
and appendix (D) respectively.

173

Chapter 8

8.1.2

The dimensionless model

Using the transformations introduced in chapter (5), equations, (8.1.1, 8.1.2) can be
written into the following dimensionless form,
dS
1
S X

=
(1

S
)

,
dt
(S + X + eS 2 )

(8.1.5)

dX X0 X
S X
=
+
Kd X .
dt

(S + X + eS 2 )

(8.1.6)

All dimensionless parameters in the model are strictly non-negative. For more details of the assumptions and denition of model parameters, we refer readers to
chapter (4) and appendix (D) respectively.
We have = 0.1019, Kd = 0.0141 and e = 0 from the anaerobic digestion of
ice-cream wastewater [39]. Thus, in order to investigate the eect of the parameter
e we take = 0.1019 and Kd = 0.0141.

8.2

Results

Using identical arguments to those in chapter (4), it can be shown that the following
hold
1. The region bounded by X > 0, 1 > S >

and 1 S

= 0 is an

invariant region.
2. If Kd 1 then the microorganisms die-out. (This means that the washout
solution branch is globally stable). This follows from the fact
0

S
< 1.
(S + X + eS 2 )

Hence, we assume in the following that 0 < Kd < 1.

174

Chapter 8

In section (8.2.1), it is shown that there are three steady state solutions branches.
These are a washout branch and two no-washout branches. The conditions for
the no-washout solution branches to be physically meaningful are derived. In section (8.2.2), the stability of the steady state solutions is determined. The steady
state diagrams are presented in section (8.2.4). Asymptotic solutions for large residence times are stated in section (8.2.3).

8.2.1

Steady State Solutions Branches

The steady state solutions of the system of equations (8.1.5) and (8.1.6) are given
by,
The washout branch,
(S , X ) = (1, 0).
The no-washout branches,
)
(

c )

(1

S
i
c ,
,
(S , X ) = S
i
+ Kd

b
b2 4ac

c
S1 =

,
2a
2a

(8.2.7)

(8.2.8)
c2 = b +
S
2a

b2 4ac
.
2a

c are the roots of the quadratic equation

The values S
i
c 2 + bS
c + c = 0.
aS
i
i

(8.2.9)

The coecients a, b, and c are dened by

a = e( Kd + ),

b = ( (1 Kd ) + ),

c = .

We have the following restriction on parameter values: e > 0, 0 < Kd < 1, 0

1, > 0. The necessary condition for the physically meaningful solutions to

175

Chapter 8

equation (8.2.9) is

>

min

( )
=
,
(1 Kd )

(8.2.10)

We now turn out attention to the discriminant () of equation (8.2.9). We have,

=a1 ( 1 )( 2 ),

(8.2.11)

where the coecient are

a1 = (1 Kd )2 ,
1

2 e Kd + (1 Kd )( ) + 2
=
(1 Kd )2

e
K
+
(1

K
)(

2
d
d
,
2 =
2
(1 Kd )
(
)
2
= e Kd e + (1 Kd )( Kd ) .
As a1 > 0, we have that equation (8.2.9) has two, one and no real solutions
when > 0, = 0, and < 0 respectively. When < 0, the discriminant () is
always positive and consequently there are two positive solutions and no limit point
bifurcation. This happen when,

e<

(1 Kd )( Kd )
2
Kd

(8.2.12)

For the case when the roots of the discriminant () are real ( 0) they are labeled
1 and 2 with 1 2 . This happen when
e

(1 Kd )( Kd )
2
Kd

(8.2.13)

We show in appendix (C) that the equation (8.2.9) are only of interest when
1 .When this condition holds the solutions of the quadratic equation (8.2.9) are both

176

Chapter 8

positive but they may not be physically meaningful solutions.

In the following we apply singularity theory to the quadratic equation (8.2.9). The
only possible bifurcation point that can occur in the steady state solution is the limit
point. We establish later that for certain parameter values where the limit point
can occur, reactor becomes bistable which is undesirable in many applications. The
limit point occurs when [4],

F = FSc = 0.

(8.2.14)

The non-denegencay conditions,

FSc Sc = 0

and

F = 0.

(8.2.15)

The condition F = FSc = 0 implies that

(

(S , ) =

)
(1 Kd )1 + ( )
, 1 .
e(Kd 1 + )

(8.2.16)

The non-denegencay conditions are satised at the point (8.2.16) as,

FS S = 2 a = 0,
F = S [eKd S (1 Kd )] = 0, provided S = (1 Kd )/eKd .
We conclude that the limit point occur at the point (8.2.16). From now and on, we
denote by S+ the value of the substrate concentration at the limit point.
The limit point occurs on the washout line when S+ = 1. From the point (8.2.16),
this happen when,
e = ecr =

1 Kd
.
Kd +

(8.2.17)

177

Chapter 8

The residence time when the solution of equation (8.2.9) intersects the washout
line (S = 1) is given by,

BR
=

(1 + e)
,
1 Kd (1 + e)

(8.2.18)

Dierentiating S+ with respect to the inhibition parameter (e) we obtain

dS+
(1 Kd )

=
< 0.
de
2e

(8.2.19)

From equation (8.2.19), S+ is a decreasing function of the inhibition parameter (e).

Thus when e < ecr , the value of S+ is bigger than the washout line and when e > ecr ,
S+ is less than the washout line. Simple analysis of the steady state solution reveals
three dierent possible cases depending on the value of the limit point.
Case 1. S+ > 1. (Generic)
For values of the residence time in which either S2 > S+ > 1 or S+ > S1 > 1,
then from equation (8.2.8), we see that the corresponding value for the microorganism concentration is negative. Thus, the solution branch S2 is never physically

meaningful. The range of the residence time at which S+ > S1 > 1 is 1 < < BR
.

Over this parameter range, the solution branch S1 is not physically meaningful. The

solution branch S1 is physically meaningful for > BR

because S1 < 1. There is

a transcritical bifurcation at = BR
and a limit point bifurcation at = 1 .

Case 2. S+ = 1.(Non-generic)
The solution branch S2 is never physically meaningful because S2 > S+ > 1.

. There is a pitchfork
The solution branch S1 is physically meaningful for > BR

.
singularity at = BR
because 1 = BR

Case 3. S+ < 1.(Generic)

178

Chapter 8

The solution branch S2 is physically meaningful when 1 < < BR

and it

is not physically meaningful for > BR

. The solution branch S1 is physically

meaningful when > 1 . There is a transcritical bifurcation at = BR

and limit

point bifurcation at = 1 .

8.2.2

Stability of the steady state solutions

The stability of the steady state solutions for the system (8.1.5) and (8.1.6) are
determined by the eigenvalues of Jacobian matrix evaluated at the steady state
solutions. The Jacobian matrix for system (8.1.5) and (8.1.6) is given by

X +eS 2
2

X
(
)
S (1+eS )
(X
1 + (X +S +eS 2 )2

+S +eS 2 )2

J(S , X ) =
.

X (X +eS 2 )
)

K
+ (XS +S(1+eS

(X +S +eS 2 )2
2
+eS 2 )
d
8.2.2.1

(8.2.20)

Stability along the washout branch

Evaluating the Jacobian matrix (8.2.20) at the washout steady state solution gives,

(11 +e)
1

J(1, 0) =

1
Kd
0 + 1 +e

(8.2.21)

The eigenvalue of this matrix are given by,

1 =

1
< 0,

2 =

(1 + e) + (1 Kd (1 + e))
.
(1 + e)

It follows that the washout branch is always stable when

Kd > 1
If = 0
1+e

Kd
If Kd <

1
1+e

1
1+e

(8.2.22)

If 0 < 1

then the washout branch is stable provided

<

(1 + e)
.
1 Kd (1 + e)

(8.2.23)

179

Chapter 8

8.2.2.2

Stability along the no-washout branch

The Jacobian matrix along the no-washout branch is given by,

X (X +eS 2 )
S 2 (1+eS )
1 + (X +S +eS 2 )2 (X +S +eS 2 )2

J(S , X ) =

X (X +eS 2 )
X S
(X +S
(X +S +eS 2 )2
+eS 2 )2

(8.2.24)

where we have simplied the Jacobian matrix using the result that along the nowashout solution we have,
( Kd + )
S
=

(X + S + eS 2 )

(8.2.25)

The determinant and the trace of the Jacobian matrix (8.2.24) are
det(J) =

f1 (S )X S
.
(X + S + eS 2 )3

(8.2.26)

1
X (X + eS 2 )
X S
+

,
(X + S + eS 2 )2 (X + S + eS 2 )2
[

X f (S )
S 2 + 2 e S 3 + e2 S 4
=
+
(X + S + eS 2 )2
(X + S + eS 2 )2
]
X S
+
(8.2.27)
.
(X + S + eS 2 )2

tr(J) =

where,
f1 (S , X ) = X ( + ) + S + e S 2 ( ).
f (S , X ) = e (2 )S 2 + 2 S + X ( + ).
c and X
c into f1 (S , X ) we obtain,
Substituting the values for S
i
i
[
]
+ b
c1 , X
c1 ) =
,
f1 (S
2e( + Kd )2
[
]
b
c2 , X
c2 ) =
.
f1 (S
2e( + Kd )2

(8.2.28)

(8.2.29)

The expression () is dened by equation (8.2.11) and it is always non-negative (

0). For positive steady state solution, we have 0 and b < 0. Providing

180

Chapter 8

= 0, the expression ( b
that + b

> 0) is positive and we show in appendix (C.3)

c , X
c ) is negative when i=1 and positive when
< 0. Hence, f1 (S
i
i

i=2. Thus, we conclude that for physically meaningful solution, the determinant
of the matrix (8.2.24) is positive along the branch solution S1 and negative along
the branch solution S2 except when = 0. When = 0 the determinant of the
matrix (8.2.24) is equal to zero. This happen when = 1 .
We now show that when the determinant of the Jacobian matrix (8.2.24) is positive
then the trace is negative. This should be true when f (S ) > f1 (S ). To do that we
show that f (S ) > f1 (S ). We have,
f (S )f1 (S ) = (e S 2 + S ) > 0.
This gives f (S ) > f1 (S ). Thus, the trace of the Jacobian matrix (8.2.24) is negative when the determinant is positive. We conclude that for physically meaningful
solution S1 with = 0, the solution branches S1 and S2 are stable and unstable,
respectively.

8.2.3

Large residence time approximations

At large the dimensionless residence times, the asymptotic approximate of the substrate and microorganism concentrations (S1 , X1 ) is given by,
S1

((1 Kd ) + (Kd (1 + e) 1))

1
+
+ O( 3 ),
3 2

(1 Kd )
t
(1 Kd )

0 < Kd < 1,
(8.2.30)

X1

(Kd + (1 Kd ))
1

+ O( 3 ),
2

Kd
t
Kd (1 Kd )

0 < Kd < 1.

(8.2.31)

181

Chapter 8

Equation (8.2.30) demonstrates that the substrate concentration (S1 ) decreases to

zero at large values of the residence times. Interestingly, this equation also shows
that the substrate concentration (S1 ) is dependent on the substrate inhibition (e)
1

only at order

. It also shows that the substrate concentration is increasing func-

tion of the substrate inhibition (e). Thus, for small residence time, the substrate
concentration is eected negatively by the substrate inhibition (e). However, the
asymptotic analysis reveals that at high residence times this is a third order eect.
Equation (8.2.31) shows that, for large values of the residence time, the microorganism concentration (X1 ) is a decreasing function of the residence times and
independent of the substrate inhibition (e).

8.2.4

Steady-State diagrams

In this section we provide the steady-state diagrams for the substrate concentration (S ) as a function of the residence time ( ) using the parameter values given
in section (8.1.2). The analysis of the steady state solutions shows that there are
three solution branches; the washout solution and two no washout solutions. Thus
the pitchfork bifurcation occurs, at the value of the inhibition parameter given by
equation (8.2.17).
Figure (8.1) shows the value of the parameter (e) at the pitchfork singularity as
function of the parameter . This shows that for > 0 the value of pitchfork
bifurcation is a decreasing function of . When = 0, the pitchfork bifurcation
point does not exist. In the steady state diagram, pitchfork bifurcation only occur
along line H0 in gure (8.1), i.e. non-generic. Since it is non-generic bifurcation,

Chapter 8

182

the steady state diagram, corresponding to it, is not shown.

In the parameter region H1 (e < ecr ), the limit point moves to the unrealistic region

and the bifurcation point (BR

) occurs at the intersection of the solution branch (S1 )

and the washout solution. A steady state diagram showing a pitchfork singularity
is presented in gure (8.2(b)) for = 1.
In region H2 (e > ecr ), there is a region in which two physical meaningful solutions

occur (Si , i = 1, 2) which is bounded by 1 < < BR

. This region increases

as the value of the inhibition parameter (e) increases. In this region the solution
branch (S1 ) is stable while the solution branch (S2 ) is unstable. In addition to these
two branches in this region, there is a stable washout branch, i.e the model exhibits
bistability.
For values of the residence time larger than the critical value of residence time

( > BR
), the solution branch (S1 ) is the only physical meaningful solution. The

solution branch (S2 ) is now above the washout line and therefore it is not a physical
meaningful solution. An example of the steady state behavior in this case is presented in gures (8.2(c)) using = 1.
Figure (8.2(c)) to gure (8.2(d)), illustrate the eect of the inhibition parameter (e).
We concluded that the the substrate inhibition is of considerable practical importance. For xed a residence time the performance of the reactor slightly decreases
as the substrate inhibition increases.

183

Chapter 8

10
H2
1
H0
e

0.1

H1

0.01

0.001
0

0.2

0.4

0.6

0.8

Figure 8.1: Unfolding diagram showing the pitchfork singularity locus as a function
of . The three regions are: H0 (e = ecr ), H1 (e < ecr ) and H2 (e > ecr ).

8.3

Conclusion

We have analytically investigated a bioreactor model for the interaction between a

microorganism and a rate-controlling substrate. The specic growth rate model was
the Contois model including substrate inhibition term with the addition of a microorganism decay coecient. The membrane reactor and well-stirred ow reactor
are special cases of this reactor model.
Although the inclusion of inhibition with Monod kinetics is often mentioned in the
literature, the adaption of Contois kinetics to include inhibition has not been previously investigated. A scaling of the model shows that the degree of the inhibition
is controlled by
S0
.
KI

184

Dimensionless substrate concentration ( S*)

Dimensionless substrate concentration ( S*)

Dimensionless substrate concentration ( S*)

Dimensionless substrate concentration ( S*)

Chapter 8

0.1

0.01

0.001
0

10

Dimensionless residence time ()

(a) e = ecr = 0.10035, Region (H0 )

1

0.1

0.01

0.001
0

10

Dimensionless residence time ()

(b) e = 0.05 < ecr , Region (H1 )

1

0.1

0.01

0.001
0

10

Dimensionless residence time ( )

(c) e = 1 > ecr , Region (H2 )

1

0.1

0.01

0.001
0

10

12

14

Dimensionless residence time ( )

(d) e = 10 > ecr , Region (H2 )

Figure 8.2: Diagram showing the dierent steady state regions of gure (8.1). The
value of parameter: = 1.

Chapter 8

185

The steady-state solutions of the non-dimensional model were found and their stability determined as a function of the residence time. Asymptotic solutions were found
for the no-washout solution branch. It was shown that at high residence times
the euent concentration is aected negatively by the substrate inhibition (e). The
analysis of the model showed that the system has three steady state solutions including the washout branch. It was shown that the pitchfork bifurcation point occurs
only for the Contois model which includes the substrate inhibition (e).

Chapter 9
Analysis of a chemostat model
with variable yield coecient and
substrate inhibition

9.1

Introduction

In the previous chapter, we investigated the behavior of a reactor using the Contois
expression subject to substrate inhibition. In this chapter, we examine the most
general growth rate expression based on the Contois expression kinetics including
the eects of both variable yield and substrate inhibition. In order to simplify the
analysis we consider the model without decay rate.
Both the static and dynamic behaviours are investigated using this generalized
growth-rate expression. We determine the stability of steady-state solutions, report asymptotic solutions in the limit of high residence times and discuss conditions

186

187

Chapter 9

where the bioreactor shows bistable (i.e. where the no washout solution and the
washout solution are both stable at the same value of the residence time) and periodic behaviours. When the reactor is bistable, care must be taken to ensure that
random perturbations do not move the system from the basin of attraction of the
no-washout solution to the basin of attraction of the washout solution. Similarly
periodic behaviour is not desirable in many applications. Thus we believe, the main
contribution of this study is identifying the parametric space where such undesirable
behaviours occur during static and dynamic operations of a bioreactor and oering
recommendations on suitable parametric space where bioreactor could successfully
operate in. A major part of this chapter has been accepted for publication on chemical engineering communications as a journal article after it has been revised in
accordance with the reviewers comments.
The dimensional and dimensionless forms of our model are stated in sections (9.1.1)
and (9.1.2), respectively.

9.1.1

The dimensional model

The governing equations for a single reactor arise from a mass balance on the substrate and biomass and are given by the following equations

dS
(S, X)
= F (S0 S) V X
,
dt
Y (S)

(9.1.1)

dX
= F (X0 X) + V X(S, X).
dt

(9.1.2)

The Contois expression with substrate inhibition is given by

(
= m

S
Ks X + S +

)
S2
KI

(9.1.3)

188

Chapter 9

It can be noted that in the limit KI we obtain the standard Contois model.
The residence time is,

V
.
F

(9.1.4)

The yield coecient is

Y (S) = + S,

(, > 0).

(9.1.5)

Note that the standard constant yield coecient corresponds to the case = 0.
All parameters in the model are strictly non-negative. For more details of the
assumptions and denition of model parameters, we refer readers to chapter (4) and
appendix (D) respectively.
The model (9.1.1 and 9.1.2) does not include biomass decay and has previously been
considered for the case of a yield coecient ( = 0 and e = 0) in [101]. In [49, 129]
a variable yield was used but microbial inhibition was not included in model ( >
0, KI = ). Thus this study both extends previous work and provides a useful
rst step towards studying a more comprehensive model that includes microbial
inhibition.

9.1.2

The dimensionless model

Using the transformations introduced in chapter (5), equations, (9.1.1, 9.1.2) can be
written into the following dimensionless form,
dS
1
S X

=
(1

S
)

,
dt
( + S)(S + X + eS 2 )

(9.1.6)

S X
dX X0 X
=
+
.
dt

(S + X + eS 2 )

(9.1.7)

Chapter 9

189

All dimensionless parameters in the model are strictly non-negative. For more details of the assumptions and denition of model parameters, we refer readers to
chapter (4) and appendix (D) respectively.
We take the dimensionless residence time as the primary bifurcation parameter. The
dimensionless constant in the yield coecient ( ) and the dimensionless inhibitions
term (e) are the secondary bifurcation parameters. The values for are determined
by the choice of microbial system and is therefore not a tunable parameter. Note
that the scaled kinetics parameters and e, which control the nonlinear terms in
the model equation, are functions of the inuent concentration (S0 ). Thus varying
the inuent concentration, changes the signicance of the nonlinear terms. Since
there is a one to one relationship between our dimensionless variables and their dimensional counterparts, often we refer to either with the same name while discussing
results.

9.2

Results

In section (9.2.1) the steady state solutions branches are found. In section (9.2.2),
we study the stability of the steady state solutions. In section (9.2.3), we discuss
the occurrence of degenerate Hopf bifurcations along the no-washout state and the
steady state diagrams. In section (9.2.4) asymptotic solutions for large residence
times are stated.
The XPP software package [183] was used to obtain the steady-state diagrams. The
standard representation is used in these: solid lines are stable steady states; dotted
lines are unstable steady states; lled-in squares are Hopf bifurcation points; open

190

Chapter 9

circles are unstable periodic orbits; lled-in circles are stable periodic solutions. We
use the symbol lled-in red circles to denote a limit point, lled-in triangles for other
static bifurcation points and lled-in red squares for a double-zero eigenvalue. For a
periodic orbit the average that is used is the integral over the period of the solution.

9.2.1

Steady State Solutions Branches

The steady state solutions of the system of equations (9.1.6,9.1.7) are given by,
The washout branch,
(S , X ) = (1, 0).

(9.2.8)

The no-washout branch,

(

c
c , ( + Si )(1 Si )
(S , X ) = S
i

2
c1 = b b 4ac
S
and
2a
2a

)
(9.2.9)

,
c2
S

b
=
+
2a

b2 4ac
.
2a

c (i=1, 2) are the roots of the quadratic equation

The values S
i
c + bS
c + c = 0.
F = aS
2

(9.2.10)

The coecients a, b, and c are dened by

a = e ,

b = ( + ) + ,

c = .

We have the following restriction on parameter values: e > 0, > 0, 0

1, > 0. Equation (9.2.10) denes the substrate concentration (S ) as an implicit
function of the residence time ( ). We are only interested in positive values of
the substrate concentration (S ). Since, equation (9.2.10) is a quadratic in S , it

191

Chapter 9

has one, or more, positive solutions. Note that the coecient c strictly positive
consequently S = 0 is never to be a solution of equation (9.2.10). In order to
analysis equation (9.2.10), we consider three possible cases: a < 0, a = 0 and a > 0.
Case 1: a < 0. This case happens when
> e ,

(9.2.11)

or in the original variables when

( Ks KI ) < 0.

(9.2.12)

Inequality (9.2.11) is always satised for a perfect membrane reactor ( = 0). When
c2 < 0 < S
c1 . Restricting our
a < 0, equation (9.2.10) has two real solutions with S
attention to the physically meaningful situation we see that equation (9.2.10) has
c1 and only physically meaningful solution when > .
one positive real solution, S
BR
Case 2: a = 0. This case happens when
= e ,

(9.2.13)

or in the original variables when

( Ks KI ) = 0.

(9.2.14)

Equality (9.2.13) is never satised for a perfect membrane reactor ( = 0). When
a = 0, equation (9.2.10) has one real solution given by
c =
S

1
(e+1)

+1

(9.2.15)

which is only physically meaningful solution when > (1 + e).

Case 3: a > 0. This case happens when
< e ,

(9.2.16)

192

Chapter 9

or in the original variables when

( Ks KI ) > 0.

(9.2.17)

Inequality (9.2.16) is never satised for a perfect membrane reactor ( = 0). In

the appendix (B.1) we show that the solutions of the equation (9.2.10) are only of
interest when
1 = ( + ) + 2

(e ).

(9.2.18)

The solutions of the quadratic equation (9.2.10) are both positive (with S2 S1 )
when the condition (9.2.18) holds, but they may not correspond to physically meaningful solutions.
It is instructive to investigate how the euent concentration on the non-washout
state changes as the residence time varied. Dierentiating equation (9.2.10) we
obtain
dS
S
S

.
=
=

d
2aS + b
b2 4ac

(9.2.19)

In equation (9.2.19) the choice of the sign +/ corresponds to the choice of S2 or S1

respectively.
For physically meaningful solutions, equation (9.2.19) is negative along the solution
branch S1 and positive along the solution branch S2 . Thus, the solution branch S1
is an decreasing function of the residence time while the solution branch S2 is an
increasing function of the residence time.
In the following we apply singularity theory to the quadratic equation (9.2.10). The
only possible bifurcation point that can occur in the steady state solution is the limit
point. We establish later that for certain parameter values where the limit point

193

Chapter 9

can occur, the reactor becomes bistable which is undesirable in many applications.
The limit point occurs when [4],
F = FSc = 0.

(9.2.20)

The non-denegencay conditions are [4]:

FSc Sc = 0

and

F = 0.

(9.2.21)

The condition F = FSc = 0 implies that

(

(S , ) =

, .
e 1

(9.2.22)

This condition is satised only if e > 0. Thus there is no limit point bifurcation when a 0 (cases 1,2). The non-denegencay conditions are satised at
equation (9.2.22) as,
FS S = 2 a = 0

provided

e = 0,

F = S = 0.
We conclude that the necessary condition for the existence of the limit point in
equation (9.2.22) is a > 0 (Case 3). Henceforth we denote the value of the substrate

concentration at the limit point by SLP

. Note that the limit point occurs on the

washout line (SLP

= 1) and leads to the simple condition,

e = ecr,1 =

+
.

(9.2.23)

This condition denes the transition between two types of static behavior. If the
substrate inhibition parameter (e) is below the critical value (e < ecr,1 ) then the limit
point is physically unrealistic whereas, if the substrate inhibition parameter (e) is

194

Chapter 9

above the critical value (e > ecr,1 ) then the limit point is physically realistic. In
practical applications it is more useful to turn the dimensionless critical value, ecr,1
to the critical value of the inuent concentration. Returning to dimensional values,
we nd that,

S0,cr =

Ks KI
.
Ks KI

(9.2.24)

If the inuent concentration is below this critical value (S0 < S0,cr ) then the limit
point, if it is exists, is physically unrealistic, we are to the left of line B in gure (9.2(a)). On other hand, if the inuent concentration is above this critical
value (S0 > S0,cr ) then the limit point is physically realistic, we are to the right
of line B in gure (9.2(a)), consequently the bistability occurs in the steady state
diagram. Once should be noted from inequality (9.2.17) that the existence of the
limit point requires the following condition, Ks KI > 0. If inequality (9.2.17)
does not hold then there is no bistabile solution.
A transcritical bifurcation in which the solution of equation (9.2.10) intersects the
washout line (S = 1) is given by,

= BR
= (1 + e ).

(9.2.25)

A pitchfork bifurcation on other hand occur when BR

= 1 .

It is useful to draw the location of the possible steady state and bifurcation points
that may occur in the model. Figure (9.1) shows the dierent type of the steady
state solutions, the realistic and unrealistic limit point occur in the parameter
space ( , e). It can be seen from the gure (9.1) that the existence of the realistic
limit point and the bistable solution for the cases (1 and 2)(a 0) are impossible.

195

Chapter 9

Simple analysis of the case 3 (a > 0) reveals three dierent possible cases depending
on the value of the limit point.

Case 3 A. SLP
> 1.

This case corresponds to the condition

< e <

+
.

For values of the

residence time for which either S2 > SLP

> 1 or SLP
> S1 > 1 then, from equation

(9.2.9), the corresponding value for the microorganism concentration is negative.

Thus, the solution branch S2 is not physically realistic. The range of the residence

. Over this parameter range, the

time at which SLP
> S1 > 1 is 1 < < BR

solution branch S1 is also not physically realistic, but it is a physically realistic

when > BR
because here S1 < 1 and hence X1 > 0. Note that a transcritical

bifurcation and a physically unrealistic limit point bifurcation occur at = BR

and = 1 respectively. Thus the existence of the bistable solution is impossible.

Case 3 B. SLP
= 1.

This case corresponds to the condition e =

+
.

The solution branch S2 is not

physically realistic because S2 > SLP

> 1. However, the solution branch S1 is phys
ically realistic for > BR
. One notice that transcritical bifurcation is equal to the

limit point bifurcation consequently a pitchfork singularity occur at = BR

= 1 .

Thus the existence of the bistable solution is impossible.

Case 3 C. SLP
< 1.

This case corresponds to the condition e >

+
.

The solution branch S2 is only

. It is not physically
physically realistic for the residence time when 1 < < BR

realistic for > BR

. The solution branch S1 is a physically realistic when > 1 .

Thus equation (9.2.10) has two physically realistic solutions if 1 < < BR
and

196

Chapter 9

one if > BR
. Note that a transcritical bifurcation and physically realistic limit

point bifurcation occur at = BR

and = 1 respectively. Thus the bistable

solution corresponds to the range of the residence time 1 < < BR

.

In this case there is a range of the residence time (1 < < BR

) in which the two

solutions, the washout and the no washout, are stable. This implies that a small
random perturbation can cause the reactor operating at the no-washout branch
to move and operate and the washout branch, i.e. the process failure occurs in the
reactor. In practices it may be desirable to avoid the bistable region. This always can

be achieved by operating the reactor at suciently large residence time ( > BR

).

Case 2

J*

Case 1

No Limit Point

Case 3B

J
J

Case 3A

Case 3C

Limit Point: PU

Limit Point: PM

Figure 9.1: Schematic diagram showing the location of the dierent type of steady
state diagrams in the parameter space ( , e).

PU=physically unmeaningful,

PM=physically meaningful.

9.2.2

Stability of the steady state solutions

The study of the stability of the solution is important because only the stable solution maybe observed experimentally. The stability of the steady state solutions

197

Chapter 9

for the system of equations (9.1.6, 9.1.7) is determined by the eigenvalues of the
Jacobian matrix evaluated at the steady state solutions. The Jacobian matrix is
given by

J(S , X ) =

(
)
2
3
X 2 e S +S ( + e) X
( + S )2 (X +S +eS

2 2

(
)
2
X X +e S

(1+e S )S

( +S ) (X +S +e S

(X +S +e S 2 )

2 2

S 2 (1+eS )
(X +S +eS 2 )2

(9.2.26)
9.2.2.1

Stability along the washout branch

Under washout conditions the euent concentration is equal to the inow concentration. Thus this state of operation must be avoided. Evaluating the Jacobian
matrix (9.2.26) at the washout steady state solution gives,

1 ( + 1) (1+e)

J(1, 0) =

1
0
+ 1+e

(9.2.27)

The eigenvalue of this matrix are given by,

1 =

1
< 0,

2 =

(1 + e) +
.
(1 + e)

It follows that the washout branch is a stable provided that

< (1 + e) = BR
.

(9.2.28)

Equation (9.2.28) established a condition of the residence time at which the process
fails occur in the reactor. When the inequality (9.2.28) is hold then due to the
process failure the rate of microorganisms that leaving the reactor is greater than
their maximum growth rate. Equation (9.2.28) also shows that a perfect membrane
cannot have washout, a good membrane could only have washout at a very small

residence time as BR
= 0.

198

Chapter 9

9.2.2.2

Stability along the no-washout branch

In this section, we investigate the stability of the no-washout state by evaluating

the Jacobian matrix at the no-washout state. We nd that

J11 J12

J(S , X ) =

J21 J22

(9.2.29)

where

J11 =

1
+

(
)
3
2
X 2 e S + S ( + e) X
2

( + S )2 (X + S + eS 2 )
(1 + e S ) S

J12 =

J21

2,

( + S ) (X + S + e S 2 )
(
)
2
X X + e S
=
2.
(X + S + e S 2 )

J22 =

X S

2,

(X + S + e S 2 )

To obtain the afore-given expression, we have used the fact that along the nowashout branch,

S
=

(X + S + eS 2 )

(9.2.30)

The special case 2 (a=0) is considered in appendix (B.2). We evaluate the determinant of matrix (J) for the case when a = 0 (Cases 1 and 3). After some algebra we
nd,
]
[(
)
+ b (1 S1 )
c1 , X
c1 ) =
,
det(J)(S
(
)2
3

b+
[(
]
)
2 (e )
b (1 + S2 )
c2 , X
c2 ) =
det(J)(S
.
(
)2
3

b +
2 (e )

(9.2.31)

(9.2.32)

199

Chapter 9

The expression () is dened by equation (2.1.4) in appendix (B.1) and it is always

c2 , X
c2 ), is physically unnon-negative ( 0). For case (1) (a < 0), the solution, (S
c1 , X
c1 ),
realistic so we ignore it. However, along the physically meaningful solution, (S
the following expression, + b

, is positive. It follows that the determinant of

c1 , X
c1 ) is positive.
matrix (J) along the physically meaningful solution, (S
For case 3 (a > 0), both solutions must be considered. A necessary condition for
either of these solutions to be physically realistic is b < 0 (see appendix (B.1)).
c2 , X
c2 ), the following expression,
Thus, along the physically meaningful solution, (S

, is positive. It follows that the determinant of matrix (J) along the phys-

c2 , X
c2 ) is negative. However, the following expression,
ically meaningful solution, (S
+b

is negative for b < 0 (see appendix (B.3)). It follows that the determinant

c1 , X
c1 ) is positive.
of matrix (J) along the physically meaningful solution, (S
Therefore we conclude that for physically meaningful solution, the determinant of
c1 and negative along the branch
the matrix (J) is positive along the branch solution S
c2 except when = 0. Note that Hopf bifuraction can not occur along the
solution S
c2 (because det(J) < 0) and the determinant of the matrix (J) is
branch solution S
equal to zero, when = 1 .

9.2.3

Degenerate Hopf bifurcation Analysis

The conditions of a Hopf bifurcation are [184],

J11 J22 J21 J12 > 0,

= J11 + J22 = 0,

(9.2.33)

200

Chapter 9

The conditions for a double-zero eigenvalue degeneracy (DZE) to occur are given
by [184],
J11 J22 J12 J21 = 0,

= J11 + J22 = 0.

(9.2.34)

After some algebra the double-zero eigenvalue degeneracy is found to satisfy the
equation,
X S

)
3
2
e2 S + eS + X ( e) S + X ( 1) = 0.

(9.2.35)

Note that S = 0 does not correspond to a steady state solution of the model. The
case X = 0 correspond to the washout solution. We showed in section (9.2.2.1)
that the washout solution always has a negative eigenvalue and consequently does
not exhibit the double zero eigenvalue degeneracy. Thus the double zero eigenvalue
degeneracy must satisfy the equation,
e2 S + eS + X ( e) S + X ( 1) = 0.
3

(9.2.36)

At a double-Hopf bifurcation, two Hopf points annihilate each other in an unfolding

diagram (This is also known as H21 degeneracy). The conditions for the double-Hopf
bifurcation to occur are given by [184],
det(J) = J11 J22 J21 J12 > 0,

= J11 + J22 = 0,

d
= 0.
d

(9.2.37)

Since it is not simple to analyse the conditions (9.2.33, 9.2.36 and 9.2.37) analytically,
they have been solved numerically.

9.2.3.1

Results: steady-state diagram

Using a y-axis log scale in gure (9.2), important features are displayed in the
secondary bifurcation plane, i.e.

, e -plane. These features include the loci of

201

Chapter 9

the double-zero eigenvalue degeneracy (curve A), the line that is dened by equation (9.2.23) (line B) and two disjoint curves for the double-Hopf points (curve C).
Each one of these features and the 6 regions bounded by line B and curves A and C
will be discussed below. Note these regions are labeled from a to g and the steady
state diagram in region (g) is identical to that in region (a).
The steady state diagram shows Hopf bifurcation points only in regions (b, c, d)
and none in regions (a, g, f). Using various parameter values for e and , a more
detailed characteristic or behavioral prole of the steady state diagram for each of
the 6 regions, are presented in gure (9.3).
Above line (B) there may be a limit point, but it is not physically realistic. Conversely, below line (B), the limit point is realistic and consequently the reactor exhibits bistability for some values of the residence time. This suggests that bistable
condition of the reactor can be avoided by either using suciently low substrate
inhibition (e) or high substrate yield ( ) parameters values. These parameters (e,
) turnout to be a function of the inuent concentration (S0 ) and consequently
the bistability can only be avoided if the inuent concentration is suciently small.
Thus, we conclude that the inuent concentration must be controlled to ensure that
the reactor does not exhibit bistable behavior.
Note that the lower curve C in gure (9.2(b)) does not intersect with curve A and
the end point of the lower curve C is obtained by,

) (0.30364, 0.01711).
(ecr,2 , cr,2

(9.2.38)

If the value of the substrate inhibition parameter (e) is higher than the critical
value (ecr,2 ) then there is only one H21 degeneracy point. For example, when =

202

Chapter 9

0.01 and e = 4 the only solution satisfying the H21 degeneracy condition (9.2.36) is
( , ) = (14.88647, 16.10751).

(9.2.39)

Hence, the natural oscillations cannot occur on the steady state diagram when >
14.88647 (region a only) but they may occur in other regions when < 14.88647.
If the value of the substrate inhibition (e) is less than the critical value (ecr,2 ) then
there are two H21 degeneracy points. For example, in gure (9.2), when = 0.01
and e = 0.1 the H21 points areF
( , ) = (10.84049, 12.05051),

( , ) = (0.020926, 1.08110).

(9.2.40)

Therefore, natural oscillations cannot occur in the regions (0.02092 > and >
10.84049). They occur in the region (0.020926 < < 10.84049).
An example of region (a) is presented in gure (9.3(a)). The region (a) is above
the curves A,C and line B; thus neither a physically realistic limit point nor Hopf
bifurcation points exist along the no washout solution branch.
On entering the region (b) from region (a), the H21 degeneracy curve is crossed
consequently two Hopf bifurcations appear on the steady state diagram. Between
the two Hopf points, the no washout solution is unstable. An example of this case
is shown in gure (9.3(b)).
Moving from region (b) to region (c), by crossing line (B), in gure (9.2(b)) the limit
point bifurcation becomes physically realistic. The physically meaningful steady
state diagram thus contains two Hopf bifurcations and a limit point. The upper
branch of the no-washout solution is unstable while the lower branch is unstable
between the two Hopf bifurcation points. An example of this case is shown in

203

Chapter 9

gure (9.3(c)).
In gures (9.3(b) and 9.3(c)), the periodic solution is stable. Between the two Hopf
points, the average of the stable periodic solution is higher than the unstable steadystate solution. For this small region of operating space, the reactor operates at the
stable periodic solution which increase the average euent concentration leaving
the reactor. We therefore conclude that in gures (9.3(b) and 9.3(c)), the Hopf
bifurcations have decreased the reactor performance.
When the parameter value moves into region (d) from region (c) by crossing the DZE
locus (Curve A), one Hopf point is destroyed at a double-zero eigenvalue bifurcation
and only one Hopf point remains as shown in gure (9.3(d)). The lower branch is
unstable between the limit point and the Hopf bifurcation point. Furthermore the
periodic solution is unstable and terminated at a homoclinic bifurcation.
Moving from region (c) into region (f), the remaining Hopf bifurcation point is
destroyed and the steady state diagram contains a physically realistic limit point as
shown in gure (9.3(e)). Finally, moving from region (f) into region (g), we cross
curve (B) and the limit point ceases to be physically realistic. The steady-state
diagram in this case is identical to that in region (a), and is therefore not shown.

9.2.3.2

Unfolding diagram for Hopf bifurcation

Figure (9.4(a)) is an unfolding diagram for the Hopf bifurcation point in gure (9.3(c)).
It shows the value of the residence time at which a Hopf bifurcation point occurs as
a function of . There are two Hopf bifurcation points in the region (0.020926 <
< 10.84049). The end points correspond to the H21 degeneracy given by equation (9.2.40). In the region between the H21 degeneracies, periodic oscillations occur.

204

Chapter 9

For either large or small values of , including the constant yield case ( = 0),
there is no periodic behaviour. This example corresponds to the case e < ecr , (see
section (9.2.3.1)). Figure (9.4(b)) is an unfolding diagram for the Hopf bifurcation
points in gure (9.3(d)) for the case e > ecr . There is only one Hopf bifurcation
point in the region 0.01310 < < 3.89568 and two Hopf bifurcation points in the
range 3.89568 < < 14.88647. A H21 degeneracy occur at the point,
( , ) = (14.88647, 16.10751).

(9.2.41)

and a double zero eigenvalue bifurcation occurs at the point

( , ) = (0.01310, 1.40245), ( , ) = (3.89568, 4.95028),

9.2.4

(9.2.42)

Large residence time approximations

It is useful to study the stability of the system at large residence times since in a
such system with a biochemical reaction, the no-washout solution is unstable at large
residence time and the periodic solution may occur. For this purpose, we obtain the
asymptotic formula, at large values of the residence time,

1
+
O(
),

2
( )
1
X1
+
+
O(
),

2
S1

(9.2.43)
(9.2.44)

To study the stability of the solution at large of the residence time, we substitute
equations (9.2.43 and 9.2.44) onto the Jacobian matrix. We obtain,
+ 2( + )
, J12 = 0,

2

=
,
J
=

,
22

J11 =
J21

205

Chapter 9

10

0.1

0.01

A
B
C

d
f

0.001
0

e
(a)

0.1

A
B
C

0.08
0.06

0.04

d
c

0.02

g
f

0
0

0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4

e
(b) Blow up of figure (9.2(a))

Figure 9.2: Bifurcation diagram. Line (A) is the double-zero eigenvalue locus,
Line (B) corresponds to the case = e locus. Line (C) represents the
H21 locus. The value of parameters: = 1 and = 0.01.

206

1.1

Dimensionless substrate concentration (S*)

Dimensionless substrate concentration (S*)

Chapter 9

1
0.9
0.8
0.7
0.6
0.5
0

10

1.2
1
0.8
0.6
0.4
0.2
0
-0.2
0

1
0.8
0.6
0.4
0.2
0
-0.2
1.05

1.1

1.15

1.2

1.25

1.3

10

1.35

0.1

0.01

0.001
0

1.4

Dimensionless residence time ()

Dimensionless residence time ( )

(c) Region (c) (e, ) = (0.1, 0.03)

Dimensionless substrate concentration (S*)

(b) Region (b) (e, ) = (4, 7)

Dimensionless substrate concentration (S*)

Dimensionless substrate concentration (S*)

(a) Region (a) (e, ) = (4, 15)

Dimensionless residence time ()

Dimensionless residence time ( )

(d) Region (d) (e, ) = (4, 0.1)

1
0.1
0.01
0.001
0.0001
0

10

Dimensionless residence time ()

(e) Region (f) (e, ) = (4, 0.005)

Figure 9.3: Diagram showing the dierent steady state regions of gure (9.2). = 1
and = 0.01. All the symbols are explained in section (9.2).

207

Chapter 9

14
12
10

8
6
4
2
0
0

10

12

(a) e = 0.1

18
16
14

12
10
8
6
4
2
0
0

10

12

14

16

(b) e = 4

Figure 9.4: Unfolding diagram for a single reactor showing the Hopf bifurcation
locus. Parameter values: =1, = 0.01. All the symbols are explained in section (9.2).

208

Chapter 9

1
2( + ) + (1 + )
+
,

det(J) = > 0.

Trace(J) =

(9.2.45)
(9.2.46)

At suciently large of the residence time the second term in equation (9.2.45) approaches zero. This shows that the trace is negative and the determinant is positive.
Thus the steady state solution is stable.

9.3

Conclusion

We have analysed the behaviour of a bioreactor using a generalized Contois growth

model including substrate inhibition and a variable yield coecient. We consider
a generalized bioreactor model which includes as special cases the continuous ow
and membrane reactors.
The steady-state solutions of the non-dimensional model were found and their stability determined as a function of the residence time. The residence time, at which
the washout is stable, satises the inequality,
< (1 + e).
This shows that substrate inhibition (e) increases the residence time at washout.
The analysis showed that the system may have up to three steady state solutions
and revealed six physically meaningful regions having ve distinct steady state diagrams (a, b, c, d, f). In regions (c, d and f) of the steady state diagram, a physically
meaningful limit point was observed. In regions (b and c) two Hopf points were
observed whilst in the last region (d), one Hopf point was observed. The existence
of Hopf and limit points is not possible in regions (a, g).

Chapter 9

209

The inclusion of substrate inhibition in the Contois Model equation leads to a critical value of the inuent concentration above which the reactor would exhibit the
bistability over a region of the residence time, i.e. the no washout solution and the
washout solution are both stable at the same value of the residence time. On the
other hand if the inuent concentration is below the critical value then no bistability occur on the reactor. Consequently, heavy wastewater is more likely to cause
bistability.
Furthermore, the inclusion of a variable substrate yield allows the prediction of
possible periodic behaviour. Both bistable and periodic behaviour are often undesirable in practice. Our analysis identies the regions in parameter space where
these undesirable phenomena occur and explained how it depends upon the inuent
concentration. Thus by using an appropriate inuent concentration the regions of
undesirable behaviour can be avoided in a bioreactor.

Chapter 10
Analysis of a model for the
treatment of wastewater by the
activated sludge process

10.1

Introduction

The activated sludge process is used extensively worldwide for treating domestic and
industrial wastewaters. One major disadvantage of this process lies in the fact that
the disposal costs of excess sludge is responsible for a large portion of the operating
costs of a treatment facility. This proportion can be as much as 50 to 60% of the
total operating cost associated with a wastewater treatment plant [185].
With traditional disposal routes, such as incineration, the use of landll sites and
dumping at sea, there is a growing environmental and regulatory problems related
to the presence of toxic elements in the sludge and the limited amount of available

210

Chapter 10

211

land. Thus an alternate method of sludge disposal is needed. A favored option to

reduces excess sludge production is to increases the biodegradability of the sludge.
This approach involves the disintegration of the cell walls of the bacteria using
various methods such as the addition of anoxic/anaerobic stages [186], [14], enzyme
treatment [187], freezing and thawing [188], use of an inclined-plate membrane bioreactor [189], ozonation [190193], ozonation combined with chemical treatment [194],
thermal treatment [195], and ultrasound [196]. Combinations of these treatments
are also possible. Other techniques have been investigated to increase the decay rate
of the sludge, including alkaline treatment, ball milling, the use of microwaves [14]
and by growing controllable predators [197, 198].
A convenient way to measure the amount of the sludge inside the reactor is to
used the mixed liquor suspended solids (MLSS). A target value for the mixed liquor
suspended solids concentration (MLSStarget ) may be set. For values of the mixed
liquor suspended solids concentration below (higher) than the target value of the
mixed liquor suspended solids concentration, the bioreactor is said to produce negative (positive) excess sludge. If the value of the mixed liquor suspended solids
concentration is equal to the target value of the mixed liquor suspended solids concentration then the bioreactor is said to producing zero excess sludge.
Yoon [14] proposed a model of a membrane bioreactor connected to a sludge disintegration unit (SDU) in which the SDU was assumed to destroy the biochemical
activity of the sludge, converting a fraction of it, , directly into usable substrate
and the remainder, (1 ), into organic particulates. The sludge disintegration unit
itself was not modeled as a separate component. Rather the conversion of the sludge

Chapter 10

212

was accounted for by adding terms to the conventional activated sludge model. This
model was investigated using numerical integration. Yoon [14] assumed the mixed
liquor suspended solids concentration target to be 12,000 mgL1 .
In this chapter, we extend Yoons model from a membrane reactor to a general bioreactor model with a settling unit in which the membrane reactor and continuous ow
reactor are a limiting case. We also replace the Monod growth rate used in [14] by
the Contois growth rate. Our objectives are to nd the steady-state solutions of the
activated sludge system, to determine their stability as the residence time and to
obtain the conditions for zero excess sludge production.

10.2

Model equations

In this section we detail the biochemical processes, and provide the model equations.
A system of nonlinear ordinary dierential equations is used for the concentration
of microorganisms, substrate and particulate. This system describe a well-stirred,
well aerated continuously stirred bioreactor including a settling unit and a sludge
disintegration unit.
The specic growth rate, equation (10.3.9), is given by a Contois expression. It is
assumed that the biokinetic parameters (max and KS ) are the same for all soluble
organic materials whether they originated from the inuent or the disintegrated
sludge. This is reasonable as biomass growth that occurs on the original substrate
can not be distinguished from biomass growth that occurs on the disintegrated
sludge [197, Section 2]. It is also assumed that organic particulates are continuously
hydrolyzed to generate soluble substrate.

213

Chapter 10

10.3

Biochemical processes

In the model, it is assumed three biochemical processes occur within the bioreactor
and two biochemical processes occur within the sludge disintegration unit. Comparing with these models considering in this thesis, there is an extra processes occur
within the bioreactor. This corresponds to hydrolysis of particulates which is large
organic into smaller soluble organic materials that can be utilised by microorganism
for food. It is an important to notice in the model it is assumed that no particulates follow in the bioreactor there is only formed inside sludge disintegration unit.
The biochemical processes consists of several biochemical reactions which are summarised as follows,
Consumption of substrate
(S,X)

S X

(10.3.1)

Death of biomass in the bioreactor

K

d
X
stu.

(10.3.2)

Hydrolysis of particulates
k

h
P
S.

(10.3.3)

Disintegration of biomass in the SDU

k

X S + (1 )P.

(10.3.4)

Disintegration of particulates in the SDU

k

P S + (1 )P.

(10.3.5)

214

Chapter 10

10.3.1

The dimensional model

The governing equations for a single reactor arise from a mass balance on the system
and are given by the following equations
dS
(S, X)
= F (S0 S) V X
+ DF (Sr S) + kh V P,
(10.3.6)
dt

dX
V
= F (X0 X) + V X(S, X) + RF (C 1)X DF X Kd V X, (10.3.7)
dt
dP
V
(10.3.8)
= F 1 P kh V P + DF (Pr P ).
dt
V

The Contois expression is given by

(
= max

S
Ks X + S

)
.

(10.3.9)

Concentration of recycled substrate

Sr = S + (X + P ).

(10.3.10)

Concentration of recycled damaged biomass

Pr = (1 )(X + P ).

(10.3.11)

The residence time is,

=

V
.
F

(10.3.12)

All parameters in the model are strictly non-negative. For more details of the
assumptions and denition of model parameters, we refer to chapter (4) and appendix (D) respectively. The maximum of the sludge solubilization eciency () is
one (0 1). Thus, the concentration of recycled damaged biomass decreases
as the value of increases. In equations (10.3.6-10.3.8) the term DF represents the
ow of material from the reactor to the sludge disintegration unit.

215

Chapter 10

For a specic wastewater, a given biological community and a particular set of environmental conditions the parameters Ks , Kd , , , and max are xed. The
parameters that can be varied are C, D R, S0 , X0 and .
The following assumptions are made in our model for the sludge disintegration unit:
The concentrations of the soluble materials in the sludge remain unaected
No microorganisms are returned to the reactor,they are all converted into
either particles or substrate in the disintegrator.
The concentrations of the substrate and the particulates owing out of the sludge
disintegration unit and back into the bioreactor are given by equations (10.3.10 and
10.3.11). The term [(X +P )] in the equation (10.3.6), represents the disintegrated
sludge produced by the sludge disintegration unit where - the solubilization ratio
- denotes the eciency of the the sludge disintegration process.
It can be noted that for every unit of substrate consumed in the reactor, units
of biomass are produced. These units of biomass in turn produce units of
substrate and (1 ) units of particulate matter into a sludge disintegration unit.
When the particulate matter is fully hydrolyzed additional substrate ((1 )
units) is regenerated. Thus, the total amount of substrate produced for one unit of
substrate consumed is : + (1 ) = units. Conservation of mass requires
that
1.

(10.3.13)

216

Chapter 10

The concentration of sludge within the reactor is equivalent to the mixed liquor
suspended solids (MLSS) and is given by
MLSS = X + P.

(10.3.14)

The quantity of particular interest in this study is the steady-state value for the
mixed liquor suspended solids (MLSS). For the target value, we follow Yoon [14]
and take [MLSStarget =12,000 mgL1 .

10.3.2

The dimensionless model

The system of dierential equations (10.3.6-10.3.8) can be written in dimensionless

form by introducing dimensionless variables for substrate concentration [S = S/S0 ],
the microorganism concentration [X = Ks X/S0 ], [P = Ks P/S0 ] and time [t =
max t].
dS
1
S X
D

=
(1

S
)

+
(X + P ) + kh P ,

dt

(S + X )

(10.3.15)

dX X0 X
S X
R X
DX

=
+
+

K
X

,
d
dt

(S + X )

(10.3.16)

dP
1 P
D
= kh P + ([1 ]X P ) .
dt

(10.3.17)

All dimensionless parameters in the model are strictly non-negative. For more details
of the assumptions and denition of model parameters, we refer to chapter (4) and
appendix (D) respectively. The equality (10.3.13) becomes,
1.

(10.3.18)

The scaled target value of MLSS is given by

MLSStarget =

Ks
MLSStarget .
S0

(10.3.19)

Chapter 10

217

It is worth noting that the scaled steady-state concentrations of the euent and the
mixed liquor suspended solids are independent of the feed concentration. However,
the scaled target value of the mixed liquor suspended solids does depends upon the
feed concentration. Increasing the concentration of the feed decreases the scaled
target value of the mixed liquor suspended solids.
A feature of our dimensionless scheme is that there is a one to one relationship
between our dimensionless variable and their dimensional counterparts. Hence, in
following section, we write often for example the residence time, rather than dimensionless residence time.

10.4

Results

In section (10.4.1) the steady state solutions branches are found. In section (10.4.2),
we study the stability of the steady state solutions. In section (10.4.3), asymptotic solutions are presented for residence times just higher than the washout point.
Asymptotic solutions for large residence times are determined in section (10.4.4). In
section (10.4.5), we discuss steady-state diagrams for the euent concentration and
mixed liquor suspended solid concentrations leaving the reactor.
The constants and parameters used in our calculation are given in table (10.1). In
table (10.1), the parameter Ks , Kd , , kh , , and max are kinetic values and
cannot be changed. The parameter that can be varied is the inuent concentration (S0 ). The value 4000 mgL1 is given by Yoon [14] as a typical value. At the
other extreme, a high strength acidic wastewater can have a feed concentration as
high as 62000 mgL1 [199].

218

Chapter 10

Table 10.1: Parameters values.

Parameter

Unit

Value

Reference

Kd

day1

0.0131

Hu et al. [39]

kh

day1 ,

0.062.2

Yoon [14]

Ks

(g COD)(g VSS)1 , mgL1

0.4818

Hu et al. [39]

(gVSS)(g COD)1

0.2116

Hu et al. [39]

mg COD(mg MLSS)1

1.2

Yoon [14]

day1

0.9297

Hu et al. [39]

S0

mg COD L1

4000

Yoon [14]

S0

mg COD L1

62000

Chian and DeWalle [199]

0.2

Yoon [14]

max

10.4.1

Steady State Solutions Branches

The steady state solutions of the system of equations (10.3.15-10.3.17) are given by,
The washout branch,
(S , X , P ) = (1, 0, 0).

(10.4.20)

The no-washout branch,

(
c , A(1 S
c ),
(S , X , P ) = S

c =
S
A=

)
D(1 )

X ,
1 + D + kh

(10.4.21)

cA
,
1 + cA
(D + kh + 1 )
,
(1 ) D 2 + bD + ((1 R ) + Kd ) (1 + kh )
(10.4.22)

219

Chapter 10

b = (1 R ) + Kd + 1 (1 ) + kh (1 ) > 0,
c=

(1 R ) + D + Kd
.
(1 Kd ) [(1 R ) + D]

Note that we have A > 0 as 1, by equation (10.3.18). The no-washout

branch is only physically meaningful when the substrate, cell-mass, and particulate
concentrations are positive (0 < S < 1, X > 0, P > 0). This happens when
> cr =

(1 R) + D
,
1 Kd

0 < Kd < 1,

(10.4.23)

The eect of the sludge disintegration factor (D) on the substrate concentration
is determined by dierentiating the substrate concentration component of equation
(10.4.21). We obtain,
dS
B1
= 2 > 0,
dD
B2

(10.4.24)

where
B1 =b1 D 4 + b2 D 3 + (b3 + b4 )D 2 + (b5 + b6 )D + (b7 + b8 + b9 ),
b1 = 2 (1 ) > 0,
b2 =2 (1 ) [ Kd + kh + 1 + (1 R ) ] > 0,
b3 = Kd (1 ) [Kd + 2( (1 R ) + 21 + 2kh )] + 2 Kd > 0,
2

[
b4 = (1 ) [(kh + (1 R ))2 + kh (2 (1 R ) + 1 ) + 4 (1 R )1 ]
]
+ (1 ) [kh 1 + 12 ] + [ (1 R ) + 1 (1 )] > 0,
b5 =2 Kd (1 + kh ) [(1 ) (Kd + 2 (1 R ) + kh ) +
+1 [1 ]] > 0,
b6 =2 (1 R ) (1 + kh ) [(1 ) ( (1 R ) + kh ) +

220

Chapter 10

+1 [1 ]] > 0,
b7 = Kd (1 + kh ) [(1 ) kh + 1 [1 ]] > 0,
2

b8 = Kd (1 + kh ) [2kh (1 R ) (1 ) + 21 (1 R )[1 ]
+ [ kh + 1 ]] > 0,
b9 = (1 R ) (1 + kh ) [kh (1 R ) (1 ) + 1 (1 R )[1 ]
+ [ kh + 1 ]] > 0.
B2 =b10 D (D + kh + 1 ) (b10 + ) (b10 ) (D + kh + 1 ) ,
b10 =(1 R ) + D + [ Kd 1].
Equation (10.4.24) shows that the substrate concentration is a increasing function of
the sludge disintegration factor (D). Thus, at xed residence time, the performance
of reactor reaches the highest level of eciency at the minimum value of the sludge
disintegration factor (D).
The steady-state concentration in a membrane bioreactor with a sludge disintegrator
unit is obtained by setting 1 = = 0 in equation (10.4.21). This gives,
(
)
D(1 )

c
c
(S , X , P ) = S , A(1 S ),
X ,
D + kh

c =
S
A=

(10.4.25)

cA
,
1 + cA

D + Kd
,
c
=
.
(1 ) D + Kd
(1 Kd ) D

Equation (10.4.25) shows that in a membrane bioreactor the euent concentration

depends only on two sludge disintegration parameters: the sludge disintegration
factor (D) and the dimensionless particulate to substrate conversion factor ( ). The
euent concentration increases as a function of the sludge disintegration factor (D)

221

Chapter 10

increases. The mixed liquor suspended solid concentration depends upon all of the
sludge disintegration parameters (D, , and kh ).
In the absence of a sludge disintegration unit, the steady state solution (10.4.26)
simplies further to,
(
)
c , A(1 S
c ), 0 ,
(S , X , P ) = S
c =
S

(10.4.26)

Kd
cA
, A = , c =
.
1 + cA
Kd
1 Kd

It follows that the maximum value of the the mixed liquor suspended solid concentration in the absence of the sludge disintegrator unit is equal to its value when the
residence time is zero. This is given by,
MLSSmax (D = 0) =

1 Kd
.
Kd

(10.4.27)

For the parameter values given in table (10.1), this maximum value is MLSSmax (D =
0) = 69.96.

10.4.2

Stability of the steady-state solutions

The Jacobian matrix is given by

J(S , X , P ) =

2
X

(X +S )2
X

(X +S )2

)
(D+kh

2
S

(X +S )2

(R 1) Kd D

D(1)

(X +S )2

1 +D+kh

(10.4.28)

222

Chapter 10

The Jacobian matrix evaluated at the washout steady-state solution is given by

J(1, 0, 0) =
0

)
(D+kh

(R 1) Kd D

+1

D(1)

1 +D+kh

(10.4.29)

The eigenvalues of this matrix are

1
< 0,

1 + D + kh
2 =
< 0,

(1 R ) (1 Kd ) + D
3 =
.

1 =

It follows that the washout branch is always stable if

Kd 1

(10.4.30)

If Kd < 1 then the washout steady-state is stable provided

< cr =

(1 R) + D
.
1 Kd

(10.4.31)

The Jacobian matrix for the no-washout branch is given by

J(S , X , P ) =

A1

A2 +

A3

A4

A1

A3

A5

A 6

(10.4.32)

where
(1 R ) + Kd + D(1 )
X
,
A
=
,
A1 =
2
(X + S )2

2

A3 =

S X
, A4 = (D + kh ), A5 = D(1 ), A6 = 1 + D + kh .
(X + S )2

223

Chapter 10

The expressions at J(1, 2) and J(2, 2) have been simplied using the relationship
that along the no-washout branch
Kd (1 R ) D
S
=

(X + S )
Restricting our attention to the physically meaningful situation we have Ai > 0, i =
1, ..., 6. The characteristic polynomial of the Jacobian matrix (10.4.32), is given by
z() = 3 + a1 2 + a2 + a3 ,
(A6 + A3 + 1) + A1
a1 =
,

)
(
A6 A3 + A6 + A6 A1 + 2 A1 A2 + A3
a2 =
,
2
A1 (A6 A2 A5 A4 ) + A6 A3
a3 =
.
2
By the Routh Hurwitz theorem [180] it suces to show that a1 > 0, a3 > 0 and
a1 a2 a3 > 0 to conclude that all the eigenvalues of the characteristic equation z()
have negative real parts.
As there are no negative terms in the formula of a1 . It is immediately positive. To
prove that a3 > 0, it suces to prove that A6 A2 A5 A4 > 0. We have
A6 A2 A5 A4 =

[1 + D + kh ]
> 0.
A

where A is dened in equation (10.4.22). For physically meaningful solutions, we

have A > 0, Hence a3 > 0.
We now examine the expression a1 a2 a3 . After some algebra, this expression (a1 a2
a3 ) can be written as
a1 a2 a3 =

A5 A1 A4
( [A3 + 1] + A1 ) z1
+
> 0,
2
3 2

z1 = ( [A2 + A6 ] A1 + [1 + A6 ] [A6 + A3 ]) .
Thus the no-washout branch is stable whenever it is physically meaningful.

224

Chapter 10

10.4.3

Residence time approximations near the washout point

In this section, we study the approximate values of the substrate, cell-mass, particulate and MLSS concentrations, when the residence times is slightly higher than the
washout point (cr ). Thus we write the residence time as,
= cr + , > 0
For a continuous ow reactor(R = 0, 1 = = 1), the approximations are,
S 1

(1 Kd ) (z2 D + [kh + 1 Kd ])
+ O(2 ),
({[1 Kd ] + kh } D + kh + (1 Kd ))

(1 Kd )2
+ O(2 ),
X
1+D
(1 Kd )3 (1 ) D
P
+ O(2 ),
(1 + D) ({[1 Kd ] + kh } D + kh + 1 Kd )

(kh + 1 Kd ) (1 Kd )2
+ O(2 ),
MLSS
{[1 Kd ] + kh } D + kh + 1 Kd

(10.4.33)
(10.4.34)
(10.4.35)
(10.4.36)

z2 = { Kd + (1 )} {[1 Kd ] + kh } > 0.
As the residence time is increased away from the critical value, i.e as increases
from zero, equation (10.4.33 and 10.4.36) show that the euent concentration and
the mixed liquor suspended solids decreases and increases respectively. For a xed
value of the euent concentration and mixed liquor suspended solids are decreased
by increasing the sludge disintegration factor (D). This can be seen visually in
Figure (10.2).
For a membrane reactor (R = 0, 1 = = 0), the approximations are,
(1 Kd ) { Kd + (1 )}
+ O(2 ),

2
(1 Kd )
X
+ O(2 ),
D
(1 Kd )3 (1 )
+ O(2 ),
P

{[1 Kd ] + kh } D
S 1

(10.4.37)
(10.4.38)
(10.4.39)

225

Chapter 10

MLSS

(kh + 1 Kd ) (1 Kd )2
+ O(2 ).

{[1 Kd ] + kh } D

(10.4.40)

Equation (10.4.37) demonstrates that for a membrane reactor the euent concentration is, to leading order, independent of the sludge disintegrator factor (D) when
the residence time is slightly higher than the washout point. This is not the case
for a continuous ow reactor in which the euent concentration dependent on the
sludge disintegrator factor (D) as shown in equation (10.4.33). This suggests that
the sludge disintegration unit has more eect on the euent concentration for the
continuous ow reactor than the membrane reactor.
Comparing the mixed liquor suspended solid concentration in a membrane reactor (MLSSM R ) with a continuous ow reactor (MLSSF ), we have,
MLSSM R
D
< 1.

=
k
h +1Kd
MLSSF
D + k +(1K
)
h

(10.4.41)

Equation (10.4.41) shows that the mixed liquor suspended solids concentration inside
a continuous ow reactor (MLSSF ) is larger than the corresponding concentration
inside a membrane reactor.

10.4.4

Large residence time approximations

At large residence times we have the approximations

S

+ O( 2 ),

(1 Kd )Kd

(10.4.42)

1
+ O( 2 ),

Kd

(10.4.43)

(1 ) D
1
+ O( 3 ),
2
K d kh

(10.4.44)

+ O( 2 ).

Kd

(10.4.45)

P
MSLL

226

Chapter 10

Equation (10.4.42) demonstrates that the substrate concentration decreases to

zero at large value of the residence times. This shows that, for processes controlled
by Contois kinetics, the substrate concentration can be decreased to any desired
level by operating the reactor at a suciently large the residence time.
Interestingly, for the Contois growth model, the substrate and the mixed liquor
suspended solid concentrations are independent of the operation of the sludge disintegration unit (equations (10.4.42 and 10.4.45)). This is not true for the Monod
growth model [200].
Equation (10.4.43 and 10.4.45) shows that, for large values of the residence time,
the microorganism concentration and mixed liquor suspended solids are a decreasing
function of the residence time . In the limiting case where , they both
tends to zero.

10.4.5

Steady-State diagrams

In this section we discuss steady-state diagram for two reactor congurations : the
membrane reactor and the continuous ow reactor. We consider the steady-state
value of the euent concentration and the mixed liquor suspended solid concentration. Unless otherwise stated, when we discuss the performance of the reactor it is assumed that the the residence time is greater that the critical value for
washout ( > cr ).
Figures (10.1 and 10.2) show the steady-state concentrations of the substrate and
the mixed liquor suspended solids (MLSS ) as a function of the residence time and
the sludge disintegration factor (D) for the membrane reactor and the continuous

Chapter 10

227

ow reactor respectively.
These gures show that as the sludge disintegration factor increases, the value of
the residence time below which process failure occurs increases as demonstrated by
equation (10.4.23).
For a xed value of the sludge disintegration factor (D > 0), the euent concentration decreases as the residence time increases. The mixed liquor suspended solids
increases from zero rapidly over a very narrow range of residence times, reaching
a maximum value and then decreasing as the residence time approaches innity.
When D=0, the mixed liquor suspended solids is decreasing function of residence
time. The large residence time approximation given in section (10.4.4) shows that
the mixed liquor suspended solids decreases to zero as .
Figures (10.1 and 10.2) show that at xed residence time the euent concentration
increases and the concentration of the mixed liquor suspended solids decreases as
the sludge disintegration factor (D) increases. This reects the fact that the sludge
disintegration unit increases the food supply, while it decreases the amount of live
biomass. Note from equation (10.4.27), for the case of a membrane bioreactor without a sludge disintegrator unit, the maximum value of the mixed liquor suspended
solid concentrations occur when the residence time equal to zero, and is given by
MLSSmax (D = 0) = 69.96 using default parameter value. From gures (10.1 and
10.2) the eect of the sludge disintegration factor (D) upon the euent concentration is seen to be marginal. Thus, in practise the operation of a sludge disintegration
unit improves the performance of the reactor by reducing the mixed liquor suspended
solids as desired.

228

Dimensionless MLSS concentration

Chapter 10

2.5

D=0.00
D=0.05
D=0.08
D=2.00

2
1.5
1
0.5
0
0

10

12

14

16

18

Dimensionless substrate concentration ( S*)

Dimensionless residence time ()

(a)

D=0.00
D=0.05
D=0.08
D=2.00

0.1

0.01

0.001
0

10

12

14

16

18

Dimensionless residence time ( )

(b)

Figure 10.1: Steady-state diagrams for a membrane reactor showing the variation of
the substrate concentration (S ), and the mixed liquor suspended solid concentration (MLSS ) as a function of the residence time. Other parameter values are stated
in table (10.1), S0 = 62, 000mg l1 (Green line), 4, 000mg l1 (Red line). 1 = =
0 and R = 0.

229

Dimensionless MLSS concentration

Chapter 10

0.14

D=0.00
D=0.05
D=0.50
D=1.00

0.12
0.1
0.08
0.06
0.04
0.02
0
0

10

12

14

16

18

Dimensionless substrate concentration ( S*)

Dimensionless residence time ()

(a)

D=0.00
D=0.05
D=0.50
D=1.00

0.1

0.01

0.001
0

10

12

14

16

18

Dimensionless residence time ( )

(b)

Figure 10.2: Steady-state diagrams of continuous ow reactor showing the variation

of the substrate concentration (S ), and the mixed liquor suspended solid concentration (MLSS ) as a function of the residence time. Other parameter values are stated
in table (10.1), S0 = 62, 000mg l1 (Green line), 73, 100mg l1 (Blue line). 1 = =
1 and R = 0.

230

Chapter 10

10.5

Discussion

In this section we investigate the condition for the mixed liquor suspended solids
to equal the target value (zero excess sludge production). The target value of the
mixed liquor suspended solids is set to be 12,000 mgL1 [14]. The default feed
concentration is S0 = 4, 000mg l1 . From equation (10.3.19) the corresponding
value of the scaled mixed liquor suspended solids target is 1.4. From now on, we
denote the scaled mixed liquor suspended solids target by MLSStarget .
As noted in section (10.3.2) the steady-state value for the mixed liquor suspended
solid concentration is independent of the inuent concentration. However, the scaled
mixed liquor suspended solids target does depend on the inuent concentration.

10.5.1

Membrane reactor

We rst consider the case when the target value (MLSStarget ) is 1.4. For a membrane
bioreactor, we see from gure 10.1(a) that when D = 0.05 or 0.08, the steadystate value of the mixed liquor suspended solid concentration (MLSS ) intersects
the target value (MLSStarget ) at two values of the residence time (1 , 2 ). At these
values for the residence time the reactor is operating in a state of zero excess sludge
production. If either 0 < < 1 or 2 < then the reactor operates in a state
of negative excess sludge production. If 1 < < 2 then the bioreactor operates
in a state of positive excess sludge production. When D = 0 there is only one
intersection point. The values of intersection points are given in table (10.2). When
D=2, we see that there are no intersection points and the system is always in a
state of negative zero excess sludge production. We discuss this case in more detail

231

Chapter 10

shortly.
When the inuent concentration is increased to S0 = 62, 000mg l1 the target value
MLSStarget is reduced to 0.093 (the green line). In this case there are two intersection
points when D=0.05, 0.08 and 2 and one intersection point when D = 0. It is
apparent from the gure (10.1(a)) that a further increase in the value of sludge
disintegration factor (D) would eliminate the intersection points.

10.5.2

Continuous ow reactor

We now consider the operation of a continuous ow system without recycle. Figure (10.2(a)) show that when the inuent concentration is S0 = 4, 000mg l1 , the
mixed liquor suspended solid concentration (MLSS ) is always below the target value
of MLSStarget = 1.4. This indicates that the use of a sludge disintegration unit is
superuous for this inuent concentration. If the inuent concentration is increased
to S0 = 73, 100mg l1 then the target value MLSStarget is reduced to 0.079 (blue
line). In gure (10.2(a)) we see that when D=0 or 0.05, the steady state value
of the mixed liquor suspended solid concentration intersects the new target value
MLSStarget at two points. When the sludge disintegration factor (D) is increased
to D=0.5, there are no intersection points, i.e, the system is always in a state of
negative excess sludge production. Thus for a heavily contaminated wasterwater
with S0 = 73, 100mg l1 , the use of a sludge disintegration unit is benecial.
In gure (10.2(a)), the green line represents the case when the inuent concentration is S0 = 62, 000mg l1 giving a target value of MLSStarget =0.093. We see that
this case is very near a boundary at which the sludge disintegration unit becomes

232

Chapter 10

benecial. When D=0, there are intersection points which are shown in table (10.2).
However when the sludge disintegration factor (D) is increased to 0.05 there are no
intersection points.
We now show how to determine the critical value of the inuent concentration at
which the use of the sludge disintegration unit becomes desirable, denoted by S0,cr .
We note that the maximum values of the scaled steady-state mixed liquor suspended
solid concentration (MLSSmax (D = 0)) is independent of the inuent concentration.
However, the scaled mixed liquor suspended solids target does depend on the inuent concentration. Therefore, the critical value of the inuent concentration (S0,cr )
is obtained by solving the equation,
MLSSmax (D = 0) = MLSStarget .

(10.5.46)

MLSStarget
Ks .
MLSSmax (D = 0)

(10.5.47)

This gives,

S0,cr =

In gure (10.2(a)) we have MLSSmax (D = 0) = 0.0933055. This gives S0,cr =

61, 964mg l1 . Thus if the inuent concentration is below 61, 964mg l1 , the continuous ow system without recycle always operates in a state of negative excess
sludge production and consequently the use of the sludge disintegration unit is not
required. However, if the inuent concentration above 61, 964mg l1 then, in the
absence of a sludge disintegration unit, there will be a range of the residence time
over which positive excess sludge production is released. Therefore for the highly
contaminated wastewaters, a sludge disintegration unit may be required.
For the case of a membrane bioreactor, equation (10.5.46) can be used to obtain a

233

Chapter 10

formula for the critical value of inuent concentration,

MLSStarget Ks Kd
S0,cr (Membrane reactor) =
,
1 Kd

(10.5.48)

For our parameter values, we nd that S0,cr = 82.64 mg l1 . For practical applications we conclude that the sludge disintegration unit will be benecial for a
membrane reactor

10.5.3

Ensuring negative excess sludge production

In this section, we discuss the zero excess sludge production points for a membrane
bioreactor, taking S0 = 4000mg l1 , and for a continuous ow bioreactor, taking
S0 = 62000mg l1 . The value of some operation points are given in table (10.2).
Figures (10.3(a) and 10.3(b)) shows how the value of the residence time at the zero
excess sludge production points varies as a function of the sludge disintegration
factor (D). The locus of this curve is found by solving the equation,
G( , D) = MLSS MLSStarget = 0.

(10.5.49)

From gures (10.3(a) and 10.3(b)) we see that there is a critical value of the sludge
disintegration factor (D = Dcr ) with corresponding critical value for the residence

time cr,D
, at which two zero excess sludge points disappear at a limit point bifur-

cation (These critical values are given in table (10.2). The critical values are found
by solving the simultaneous equations,
G( , D) = 0,

(10.5.50)

dG( , D)
= 0,
d

(10.5.51)

234

Chapter 10

subject to the non-degeneracy condition,

dG( , D) d2 G( , D)

= 0.
dD
d 2

(10.5.52)

If the sludge disintegration factor is higher than criticality than the reactor system
is guaranteed to be in a state of negative excess sludge production for any residence
time. An example when D > Dcr , can be seen in gures (10.1(a) and 10.2(a)) for
D=2.
It is now natural to ask how the critical value for the sludge disintegration factor (D = Dcr ) depends upon the inuent concentration. This is shown in gures (10.4(a) and 10.4(b)). If the value of the sludge disintegration factor (D) is
above the line in these gures, then the reactor is guaranteed to be in a state of
negative excess sludge production for any value of the residence time. Conversely,
if the value of the sludge disintegration factor (D) is below the line the state of the
reactor depends upon the choice of the residence time.
To summarize the important nding of this section. Using the kinetic data from
table (10.1), we show that the use of a sludge disintegration unit is always useful in
practice for a membrane reactor. However for the case of continuous ow reactor,
the use of a sludge disintegration unit is only useful when the inuent concentration
is suciently high. For the particular kinetic values corresponding to the treatment
of ice-cream wastewater we found that S0,cr = 61, 964mg l1 . We now investigate
how this critical value varies for dierent type of wastewater. Naturally, for dierent type of wastewater, the kinetic values are dierent. Table (10.3) provides some
literature values for kinetic constants for the Contois model for ve wastewaters. In
table (10.4), we calculate the critical value of the inuent concentration using the

235

Chapter 10

kinetic parameters provided in table (10.3). The rst column of table (10.4) denes
the substrate. The second column of table (10.4) gives the maximum value of the
mixed liquor suspended solid concentration for this wastewater in a ow reactor,
when there is no sludge disintegration unit. This is found numerically using the
steady state equation (10.4.21). The third column of table (10.4) gives the target
value for the mixed liquor suspended solid concentration using the default value
of the inuent concentration (S0 = 4, 000mg l1 ). This value is determined using
equation (10.3.19), and taking the appropriate value for Ks from table (10.3). The
important feature to notice from this column is that the target value for the mixed
liquor suspended solid concentration is in each case higher than the maximum value
of the mixed liquor suspended solid concentration. Thus for all these wastewater,
a sludge disintegration unit is not required in order to achieves a state of negative
excess sludge production when S0 = 4, 000mg l1 . The nal column of table (10.4)
gives the critical value of the inuent concentration, S0,cr , calculated using equation (10.5.46). This shows the minimum value of inuent concentration for which
the use of sludge disintegration unit may be practical. It shows that the existence
of positive excess sludge production require high inuent concentrations.

10.5.4

Conclusion

In this chapter, we investigated the behavior of a generalized model for a bioreactor

attached to a sludge disintegration unit. The model assumes that the disintegrator
unit operates innitely quickly. The specic growth rate model was the Contois
model with the addition of a microorganism decay coecient. This model is based

236

Dimensionless residence time ()

Dimensionless residence time ()

Chapter 10

Negative excess sludge

4
3
2

Positive excess sludge

1
0
0

(a)

0.02
0.04
0.06
0.08
Sludge disintegration factor ( D )
Membrane reactor S0 = 4000 mgL1 ,

4.4
4.2

Negative excess sludge

4
Positive excess sludge

3.8
3.6
3.4
3.2
3

0.1

1 =

0.0005
0.001
0.0015
Sludge disintegration factor ( D )

0.002

(b) Flow reactor S0 = 62000 mgL1 , 1 = = 1

=0

Figure 10.3: The zero excess sludge production curve as a function of the sludge
disintegration rate (D) and the dimensionless residence time ( ). This divides
the plane into regions of positive excess and negative excess sludge production.

3
2.5

Guaranteed Negative excess sludge

2
1.5
1
0.5

Negative excess sludge possible

0
10000 20000 30000 40000 50000 60000
Influent concentration ( S0)

(a) Membrane reactor 1 = = 0

Crtical sludge disintegration factor ( Dcr )

Crtical sludge disintegration factor ( Dcr )

Parameter values are stated in table (10.1).

5
4

Guaranteed Negative excess sludge

3
2
1
Negative excess sludge possible
0
80000
100000 120000 140000
Influent concentration ( S0)

160000

(b) Flow reactor 1 = = 1

Figure 10.4: The critical value of the sludge disintegration factor (Dcr ) as a function
of the inuent concentration. Parameter values are stated in table (10.1).

237

Chapter 10

Table 10.2: The residence time required to achieve the target value MLSStarget (zero
excess sludge production) using the data from table (10.1).
Membrane reactor, S0 = 4000

Continuous ow reactor, S0 = 62000

4.9023

3.4146 4.1153

0.05

0.1066 2.3534

0.08

0.2309 0.9701

(cr,D
, Dcr )

(0.4394, 0.0904)

(3.7401, 0.0127)

Table 10.3: Kinetic constants in Contois model

Substrate

max

Ice-cream wastewater

0.9297 0.4818 0.2116 0.0131 Hu et al. [39]

Textile wastewater

0.105

0.465

0.125

0.0065 Isik and Sponza [41]

Palm oil mill euent (POME) wastewater 0.344

0.111

0.67

0.24

Abdurahman et al. [142]

Palm oil mill euent (POME) wastewater 0.344

0.111

0.62

0.21

Abdurahman et al. [141]

Synthesised dairy wastewater

0.20

0.68

0.070

Emerald et al. [147]

2.13

Ks

Kd

Reference

Chapter 10

238

Table 10.4:

Critical value of the inuent concentration in a ow-reactor. If S0 >

S0,cr the sludge disintegration unit may be benecial. S0 = 4000mg l1 , 1 = = 1,

s
MLSStarget = 12000K
.
4000

Substrate

MLSSmax (D=0) MLSStarget

S0,cr mg l1

Ice-cream wastewater [39]

0.0933055

1.445

61964

Textile wastewater [41]

0.0485300

1.395

114980

Palm oil mill euent (POME) wastewater [142]

0.1155623

0.600

50030

Palm oil mill euent (POME) wastewater [141]

0.0184633

0.333

72143

Synthesised dairy wastewater [147]

0.0149151

0.333

89305

upon one purposed by Yoon [14] for a membrane reactor. We replaced the Monod
model used by Yoon by the Contois model and considered a generalized bioreactor
model rather than considering a membrane reactor.
The steady-state solutions of the model were found and their stability determined
as a function of the residence time. They were used to study how the performance
of the reactor (the euent concentration and the mixed liquor suspended solid concentration) depends upon the process parameters of the sludge disintegration unit.
A transcritical bifurcation occur when,
= cr =

(1 R ) + D
.
1 Kd

(10.5.53)

The washout solution is the only stable solution when the residence times is lower
than the transcritical value and the no-washout solution is the only stable solution
when the residence times is higher than the transcritical value.
Asymptotic solutions were determined when the residence time is slightly larger

Chapter 10

239

than the transcritical bifurcation. These demonstrate that for a membrane reactor,
an increase in the sludge disintegration factor (D) leads to a reduction in the mixed
liquor suspended solids concentration and that the euent concentration is independent of the process parameters of the sludge disintegration unit except for the
dimensionless conversion eciency ( ). For the case of a continuous ow reactor,
increasing the sludge disintegration factor (D) decreases the mixed liquor suspended
solids concentration. The euent concentration only depends on the sludge disintegration factor (D).
Asymptotic solutions in the limit of high residence times reveal that the mixed
liquor suspended solids concentration reaches zero. Furthermore, this values is independent on the process parameters of the sludge disintegration unit. This show
that the sludge disintegration unit only enhance performance at suciently low residence times. Our results shows that the operation of the sludge disintegration unit
increases the euent concentrations by a small amount which in practise is insignificant.
The main purpose of using the sludge disintegration unit is to reduce the mount of
sludge inside or leaving the reactor. In practise, a target value is set. If the steady
state mixed liquor suspended solid concentration value is higher than the target
value then the reactor is said to operates in a state of positive excess sludge production. If the steady state value is lower than the target value then the reactor is said
to operates in a state of negative excess sludge production. The most important
result in our chapter is concerned when the reactor operates in a state of negative
excess sludge production.

Chapter 10

240

The steady state diagram for the mixed liquor suspended solid concentration shows
two generic behaviors. In the rst scenario, positive excess sludge production occurs
for a range of residence times. At suciently small and high residence time, the
system is guaranteed in a state of negative excess sludge production. In the second
scenario which is more useful in practical applications, the reactor is in a state of
negative excess sludge production for any residence time. Which of the two scenarios
occurs depends upon the value of the inuent concentration. If the inuent concentration is below a critical value then the reactor operates in the second scenario
even in the absence of a the sludge disintegration unit, i.e the sludge disintegration
unit is not required. Conversely, if the value of the inuent concentration is above
a critical value then the reactor operation in the rst scenario on the absence of
sludge disintegration unit. There is then a critical value of the sludge disintegration
factor (Dcr > 0). If D > Dcr then the system is guaranteed to be in a state of
negative excess sludge production.
For a membrane reactor, the critical value of the inuent concentration is very low.
Thus, for all practical applications, in the absence of a sludge disintegration unit, a
membrane reactor will always operates in the rst scenario. Thus the operation of
a sludge disintegration unit for a membrane reactor is desirable to move the operational state of system from the rst scenario to the second scenario. For a continuous
ow reactor, the critical value of the inuent concentration is much higher and it is
only for highly contaminated wastewaters where the use of a sludge disintegration
unit may be benecial. This distinction between a membrane reactor and a ow
reactor is one of our most important nding.

Chapter 10

241

The zero excess sludge production curve can be readily determined in the residence
time and the sludge disintegration factor plane. This curve divides the plane into
two regions corresponding to positive and negative excess sludge productions. There
is a critical value for the sludge disintegration factor. If the value of sludge disintegration factor is below the critical value then the system operate in scenario 1 and
above the critical value then the system always operate in scenario 2. We showed
how this critical value can be calculated by solving two simultaneous equations. This
critical value illustrates the utility of using a sludge disintegration unit to ensure
that the mixed liquor suspended solids remains below a specied target level.
Finally, we have used the theory developed in this chapter to determine the value of
the inuent concentration for ve type of wastewater above which the use of sludge
disintegration unit may be benecial for a ow reactor, see table (10.4).

Chapter 11
A nite rate of sludge
disintegration unit model

11.1

Introduction

In the previous chapter we analyzed a model for a sludge disintegration unit attached
to a bioreactor due to Yoon [14]. We adopted this model by using the Contois model
instead of the Monod model and considered a general bioreactor model including a
membrane reactor and continuous ow reactor. The main nding in the previous
chapter was that there is a critical value of the inuent concentration above which
a sludge disintegration unit is required to ensure negative excess sludge production.
We also determined a critical value of the sludge disintegration factor corresponding,
to a limit point along the zero excess sludge production curve above which, for any
value of the residence time, negative excess sludge is produced in the reactor.
A critical assumption in the Yoon model is that the processes in the sludge disintegration unit occur innitely quickly. A natural question to ask is how important
242

243

Chapter 11

is this assumption to the results obtained in the previous chapter? This motivates
the nal chapter of the thesis where a new sludge disintegration unit model is developed by relaxing this assumption and allowing the kinetic processes in the sludge
disintegration unit to occur at a nite rate.
In this chapter, we will study the behavior of a membrane bioreactor connected to
a sludge disintegration unit when the reaction rate on sludge disintegration unit is
nite rate. We then nd the steady-state solutions, determine their stability. The
steady state is used to compare between Yoon model and our model at innite rate
and evaluate how the process of sludge disintegration unit at nite rate eect upon
the performance of the reactor.

11.2

Yoon model (Innite rate SDA model)

In this section we recall the Yoon model for a membrane bioreactor that discussed
in chapter (10).
dS
(S, X)
= F (S0 S) X
+ F D(Sr S) + kh P,
dt

dX
V
= X(S, X) F DX Kd X,
dt
dP
= kh P + F D(Pr P ).
V
dt
V

(11.2.1)
(11.2.2)
(11.2.3)

Concentration of recycled substrate

Sr = S + (X + P ).

(11.2.4)

Concentration of recycled particulate

Pr = (1 )(X + P ).

(11.2.5)

244

Chapter 11

Concentration of recycled microorganism,

Xr = 0.

(11.2.6)

The processes in the sludge disintegration unit are assumed to occur innity quickly.
They destroy the biochemical activity of the sludge, converting a fraction of the
sludge, , directly into usable substrate and the remainder, (1 ), into organic
particulates. The following assumptions are made,
the disintegrator does not eect soluble material,
all microorganisms entering the disintegrator are converted into either particles
or substrate.
It is clear from equation (11.2.4 and 11.2.5) that in Yoon model some of particulates
must be converted innitely quickly to substrate. The question that arises is if this
occur innitely quickly why the remainder of the particulate is also not converted
to substrate.

11.3

Revised Yoon model (nite rate SDA model)

We wish to investigate how the assumption that the sludge disintegration process
occur innitely quickly eect the model predictions. In order to do this, we need to
extend the model in the previous section by assuming that the sludge disintegration
processes occur at a nite rate. Before doing this we must rst write down the biochemical processes that occur inside the sludge disintegration unit. These processes
are not mentioned by Yoon [14], but based on his equation, it is reasonable process

245

Chapter 11

to assume that there are two such precesses,

Disintegration of particulate in the sludge disintegration unit
kp

P (1 1 )P + 1 1 S.

(11.3.7)

Disintegration of microorganism in the sludge disintegration unit

k

x
X
(1 2 )P + 2 2 S.

(11.3.8)

There is no reason to assume that 1 = 2 and that 1 = 2 . Thus we retain these

terms to make the model general as possible. We now write down the model equation for a membrane bioreactor connected to a sludge disintegration unit assuming
that the kinetics processes in the sludge disintegration unit occur at nite rate but
still much quicker than the biochemical processes, so that we can ignore all other
biochemical processes inside the sludge disintegration unit.
It can be noted that for every unit of substrate consumed in the reactor, units
of biomass are produced. These units of biomass in turn produce 2 2 units of
substrate and (1 2 ) units of particulate matter into a sludge disintegration unit.
When the particulate matter is fully hydrolyzed additional substrate ((1 2 )
units) is regenerated. Thus, the total amount of substrate produced for one unit
of substrate consumed is : 2 2 + (1 2 ) = (2 2 + [1 2 ]) units. By the
same argument we obtain that 2 2 + 1 (1 2 )1 = (2 2 + 1 1 [1 2 ]) units.
Thus, conservation of mass requires that
(2 2 + 1 1 [1 2 ]) < 1,

(11.3.9)

(2 2 + [1 2 ]) < 1,

(11.3.10)

246

Chapter 11

11.3.1

The dimensional model

A mass balance on the system are given by the following equations,

Membrane bioreactor model

dS
(S, X)
= F (S0 S) V X
+ F D(Sr S) + V kh P,
dt

dX
V
= V X(S, X) V Kd X + F D(Xr X),
dt
dP
V
= V kh P + F D(Pr P ),
dt
V

(11.3.11)
(11.3.12)
(11.3.13)

Sludge disintegration unit model

dSr
= F D(S Sr ) + Vs 1 1 kp Pr + Vs 2 2 kx Xr ,
dt
dXr
Vs
= F D(X Xr ) Vs kx Xr ,
dt
dPr
Vs
= F D(P Pr ) Vs 1 kp Pr + (1 2 )Vs kx Xr .
dt
Vs

(11.3.14)
(11.3.15)
(11.3.16)

where,
(
(S, X) =max

S
Ks X + S

)
.

(11.3.17)

All parameters in the model are strictly non-negative. For more details of the
assumptions and denition of model parameters, we refer to chapter (4) and appendix (D) respectively.

11.3.2

Reduction of nite model to Yoon model

In this section we compare our steady state expression with those of Yoon expression
for the processes in the sludge disintegration unit. Inside the sludge disintegration

247

Chapter 11

unit we nd that,
Pr =

qnD P
qnD kx (1 2 ) X
,
+
qnD + kp 1
(qnD + kx ) (qnD + kp 1 )

Sr =S +
Xr =

1 1 kp P
kx (kp 1 1 [1 2 ] + 2 2 [qnD + kp 1 ]) X
,
+
qnD + kp 1
[qnD + kx ] [qnD + kp 1 ]

qnD X
.
qnD + kx

where, q =

F
V

(11.3.18)
(11.3.19)
(11.3.20)

. In order for equation (11.3.20) to agree with equation (11.2.6), we

must take the limit kx . Taking this limit in equations (11.3.18 and 11.3.19),
we nd that,
qD n
[P + (1 2 ) X],
kx +
qD n + 1 kp
(
)
1 kp (1 2 ) 1
1 1 kp P
lim Sr =S +
+
+ 2 2 X,
kx +
qD n + 1 kp
qD n + 1 kp
lim Pr =

(11.3.21)
(11.3.22)

The only way in which equation (11.3.21) can equal equation (11.2.5), is when both of
these equations are zero. This happen for equation (11.3.21) in the limit of kp
and for equation (11.2.5) when = 1. Taking this limit in equation (11.3.21), we
nd that,
lim

kx =kp +

Sr =S + 1 P + ([1 2 ] 1 + 2 2 ) X,

(11.3.23)

Equations (11.3.23) is equivalent to equation(11.2.4) only if = 1 = 2 . We see

that the Yoon model is only correct if = 1.
An alternative explanation for this case is as follow. If the process inside the sludge
disintegration unit happen innity quickly then equation (11.3.7) shows that one
unit of particulate (P) disintegrates to give (1 1 ) units of particulate (P). If
this process happens innity quickly, then these (1 1 ) units of particulate (P)
must innity quickly disintegrate to give (1 1 )2 units of particulate (P). Similarly,

248

Chapter 11

(1 1 )2 units of particulate (P) must innity quickly disintegrate to give (1 1 )3

of particulate (P). Thus, the concentration of particulate (P) must reach zero if the
process occur innity quickly. Yoon model [14] only gives the value Pr = 0 when
= 1. Consequently, his model can only be correct when = 1.

11.3.3

The dimensionless model

The system of dierential equations (11.3.11-11.3.16) can be written in dimensionless

form by introducing dimensionless variables for substrate concentration [S = S/S0 ],
the microorganism concentration [X = Ks X/S0 ], [P = Ks P/S0 ], [Sr = Sr /S0 ],
[Xr = Ks Xr /S0 ], [Pr = Ks Pr /S0 ] and time [t = max t].

Membrane bioreactor model

1
S X
D[Sr S ]
dS

=
(1

S
)

k
P
+
,
h
dt
(S + X )

(11.3.24)

dX
S X
D[Xr X ]

=

K
X
+
,
d
dt
(S + X )

(11.3.25)

dP
D[Pr P ]
= kh P +
,
dt

(11.3.26)

Sludge disintegration unit model

dSr nD[S Sr ]
=
+ 1 kp 1 Pr + 2 kx 2 Xr ,
dt

dXr nD[X Xr ]
=
kx Xr ,
dt

dPr nD[P Pr ]
=
1 kp Pr + (1 2 )kx Xr .
dt

(11.3.27)
(11.3.28)
(11.3.29)

All dimensionless parameters in the model are strictly non-negative. For more details
of the assumptions and denition of model parameters, we refer to the chapters (4)
and the appendix (D) respectively.

249

Chapter 11

11.4

Result

In this section we assume the process in sludge disintegration unit sucient fast
which it operate at the steady state so the system (11.3.24-11.3.29) reduce to equation (11.3.24-11.3.26) and algebraic relationship,
Sr = S + a4 X + a5 P , Xr = a6 X , Pr = a1 P + a2 X ,
nD
(1 2 ) nD kx
(
),
,
a
=
2
nD + 1 kp
(nD + kx ) nD + 1 kp
])
[
(
kx 1 kp 1 [1 2 ] + 2 2 nD + 1 kp
(
)
a4 =
,
(nD + kx ) nD + 1 kp

a1 =

a5 =

11.4.1

1 kp 1
nD
, a6 =
.

nD + 1 kp
nD + kx

(11.4.30)

Steady State Solutions Branches

The steady state solutions of the system of equations (11.3.24,11.3.26) are given by,
The washout branch,
(S , X , P ) = (1, 0, 0).

(11.4.31)

The no-washout branch,

)

c
c
(S , X , P ) = S , A(1 S ), c4 X ,

c3 A
,
1 + c3 A
)
]
([
kh n + 1 kp D + kp 1 kh (D n + kx )
)
(
,
A=
b1 D2 + b2 D + Kd kh kp kx 2 1

(11.4.32)

c =
S

(11.4.33)

b1 = (c1 kx + nKd ) 1 kp + kx kh nc2 + Kd kh n2 > 0,

b2 = (c1 kh kx + Kd [kx + kh n]) 1 kp + Kd kh kx n > 0,
c1 = {1 [2 1] 2 2 } + 1

(11.4.34)

250

Chapter 11

c2 = { [2 1] 2 2 } + 1

(11.4.35)

Kd + D(1 a6 )
.
(1 Kd ) + D( a6 1 )

(11.4.36)

c3 =
c4 =

kh

Da2
.
+ D[1 a1 ]

Note that we have A > 0 as c1 > 0 and c2 > 0, by equations (11.3.9 and 11.3.10).
The no-washout branch is only physically meaningful when the substrate, cell-mass,
and particulate concentrations are positive (0 < S < 1, X > 0, P > 0). This
happens when the denominator of the expression c3 is positive,
> cr = D[

11.5

1
n
],

1 Kd
kx

0 < Kd < 1,

(11.4.37)

Stability of the steady-state solutions

The Jacobian matrix is given by

J(S , X , P ) =

2
X

a4 D

(X +S )2
X

(X +S )2

(X +S )2

Kd +D(1a6 )

]
+

(X +S )2

Da2

+a D
kh
5

2
S

.
0

[
]

D(1a )+k
1

(11.5.38)
The eigenvalues of Jacobian matrix evaluated at the washout steady-state solution
is given by
1
< 0,

[
]
D(1 a1 ) + kh
2 =
< 0,

1 =

3 =

(1 Kd ) + D(a6 1)
.

251

Chapter 11

It follows that the washout branch is always stable if

Kd 1

(11.5.39)

Using the same argument as in chapter (4) we can show that the washout solution
is globally stable when Kd 1. If Kd < 1 then the washout steady-state is stable
provided
[

]
1
n
<D

,
1 Kd kx

(11.5.40)

The critical value of the residence time (11.5.40) in which the stability of the washout
solution is changed depending on the reaction rate of hydrolysis (kx ). Thus, there is
a new critical value of the reaction rate (kx ) where the washout solution is unstable,
giving by,
kx < n[1 Kd ],

(11.5.41)

The Jacobian matrix for the no-washout branch is given by

J(S , X , P ) =

A1

A2 +

A3

A4

A1

A3

A5

A 6

(11.5.42)

where
Kd + Dc5
S X
X
,
A
=
,
A
=
,
A1 =
2
3
(X + S )2

(X + S )2
2

A4 = kh + a5 D, A5 = Da2 , A6 = D(1 a1 ) + kh ,
)
(
kx 1 c1 kp + Dn [1 2 2 ]
)
(
c5 =
.
(Dn + kx ) Dn + 1 kp

252

Chapter 11

Note that we have c5 > 0 as (1 2 2 ) > 0, by equations (11.3.9 and 11.3.10).

The expressions at J(1, 2) and J(2, 2) have been simplied using the relationship
that along the no-washout branch
[
]
Kd + D(1 a6 )
S

=
.

(X + S )
Restricting our attention to the physically meaningful situation we have Ai > 0, i =
1, ..., 6. The characteristic polynomial of the Jacobian matrix (11.5.42), is given by
z() = 3 + a1 2 + a2 + a3 ,
(A6 + A3 + 1) + A1
,

(
)
A6 A3 + A6 + A6 A1 + 2 A1 A2 + A3
a2 =
,
2
A1 (A6 A2 A5 A4 ) + A6 A3
.
a3 =
2

a1 =

By the Routh Hurwitz theorem [180] it suces to show that a1 > 0, a3 > 0 and
a1 a2 a3 > 0 to conclude that all the eigenvalues of the characteristic equation z()
have negative real parts.
As there are no negative terms in the formula of a1 . It is immediately positive. To
prove that a3 > 0, it suces to prove that A6 A2 A5 A4 > 0. We have
A6 A2 A5 A4 =

c6
> 0,
A

c6 = kh +

D 1 kp
.
D n + 1 kp

where A is dened in equation (11.4.33). For physically meaningful solutions, we

have A > 0, Hence a3 > 0.
We now examine the expression a1 a2 a3 . After some algebra, this expression (a1 a2
a3 ) can be written as
a1 a2 a3 =

A5 A1 A4
( [A3 + 1] + A1 ) z1
+
> 0,
2
3 2

253

Chapter 11

z1 = ( [A2 + A6 ] A1 + [1 + A6 ] [A6 + A3 ]) .
Thus the no-washout branch is stable whenever it is physically meaningful.

11.6

Discussion

In this section we investigate the condition for the mixed liquor suspended solids
to equal the target value, this is known as zero excess sludge production. The
target value of the mixed liquor suspended solids and the default feed concentration
are as chosen in chapter (10). Figure (11.1) shows the steady-state value of the
mixed liquor suspended solid concentration (MLSS ) as function of the residence
time. The zero excess sludge production points correspond to the intersection of the
horizontal red line with steady state mixed liquor suspended solid curves. In the
following we analyze how the zero excess sludge production points depend upon the
value of the sludge disintegration factor for two cases. In rst case we compare our
model predictions with that of Yoonmodel by assuming that the processes inside
the sludge disintegration unit occur innitely quickly. In the second case, we relax
the assumption that these processes occur at nite rate.

11.6.1

Innite rate kinetic (kx = kp = k +)

In order to compare our model with Yoon model, we consider the limit in our model
when the reaction rates in the sludge disintegration unit reaches innity (kx = kp =
k ). Figure (11.2) is a unfolding diagram of the zero excess sludge production
points for a membrane bioreactor, using the parameter values from chapter (10)

254

Dimensionless MLSS concentration

Chapter 11

2.5

D=0.00
D=0.05
D=0.06
D=0.08

2
1.5
1
0.5
0
0

10

12

14

16

18

Dimensionless residence time ( )

Figure 11.1: Steady-state diagrams for a membrane reactor showing the variation
of the mixed liquor suspended solid concentration (MLSS ) as a function of the
residence time. Other parameter values are stated in table (10.1), MLSStarget =
1.4, kx = kp = 100.

Chapter 11

255

with S0 = 4000mg l1 and the target value of the mixed liquor suspended solids set
to be 12,000 mgL1 [14].
Figure (11.2) shows that the zero excess sludge production curve contains a limit
point (D = Dcr ). When the sludge disintegration factor is less than the limit
point (D < Dcr ), there are two values of the residence time corresponding to zero
excess sludge production points. Only the higher of these points is of practical
interest. We see from gure (11.2) that for suciently low values of the sludge
disintegration factor the prediction of the Yoon model is very close to that our
model. If the sludge disintegration factor is lower than 0.046 (D < 0.046) then the
error using Yoon model is smaller than 10% of our model value. Thus it might
argued that for suciently low value of the sludge disintegration factor, the Yoon
model is suciently accurate for practical calculation.
For values of the sludge disintegration unit is larger than the limit point, D > Dcr ,
then the reactor is in a state of negative excess sludge production for any residence
time. These limit points for Yoon model and our model are given by Dcr =0.090
and Dcr =0.065 respectively. Thus, Yoon model overpredicts the limit point by
38% of our model. If a value of the sludge disintegration factor (D) is chosen to
ensure negative excess sludge production based on Yoon predictions of the limit
point (Dcr ) then zero excess sludge production will be observed. However, the limit
point (Dcr ) predicted by Yoon is a quite a bit larger than the limit point of our
model.

256

Dimensionless residence time ()

Chapter 11

k=infinity
Yoon Model

5
4
3
2
1
0
0

0.02

0.04

0.06

0.08

0.1

Sludge disintegration factor (D)

Figure 11.2: The zero excess sludge production curve as a function of the sludge
disintegration rate (D) and the dimensionless residence time ( ). Parameter values
are: kx = kp = k , n=1. Other parameter values are stated in table (10.1)

257

Chapter 11

11.6.2

Finite rate kinetics

In this section we continue to assume that the two reaction rates (kx = kp = k) in
the sludge disintegration unit are equal and investigate how the unfolding diagram
of the zero excess sludge production curve depends upon the reaction rate (k), the
sludge disintegration factor (D) and the relative volume of the sludge disintegration
unit (n).
A practical question is how does the zero excess sludge production curve change
as the reaction rate in the sludge disintegration unit is varied away from innity
? Figure (11.3) is a unfolding diagram of the zero excess sludge production points
showing four curves corresponding to four values of the reaction rate in the sludge
disintegration unit.
Figure (11.3) shows that within graphical accuracy the unfolding diagram when
k = 100 is identical to that when k +. Therefore, for any value of the reaction
rate (k) larger than 100, we can assume that it is eectively innity large. When
the value of the reaction rate is reduced to one then the shape of the zero excess
sludge production curve is similar to that the previous cases but the critical value
of the sludge disintegration (D) corresponding to limit point increases.
When k = 1, the critical value of the sludge disintegration (D) corresponding to
limit point is 0.085. This is 31% higher than the value when the reaction rate
reaches innity (k +). In practical application it can be argued that due to
uncertainty in the model parameters and in the model, a variation of 10% within the
correct value is acceptable. We see that as the value of k decrease the value of the
limit point increases. Calculation shows that when the reaction rate is larger than

258

Chapter 11

k = 2.0288 the critical value the sludge disintegration (D) is in 10% of the innite
model. Therefore it can be argued that if the value of the reaction rate is higher
than k = 2.0288 the innite model is suciently accurate for practical purpose.
That is the eect of the nite rate kinetic is only appropriate if the reaction rate is
smaller than k = 2.0288.
In the limit that k + and when k = 1, 100, the zero excess sludge production
curves has a critical value corresponding to limit point above which the reactor is
in a state of negative excess sludge production for any value of the residence time.
However, when the reaction rate is decreased to 0.01 there is no indiction on the
steady state diagram that there is a limit point. Of course there may be a limit
point but it may occur at a larger value of the sludge disintegration factor. However,
numerical investigations suggests that there is never a limit point when k = 0.01.
This suggests that there is a critical value of the reaction rate (k) at which the limit
point is destroyed. This critical value is found by solving the system of equations,
G( , D) = MLSS MLSStarget = 0,
dG( , D)
= 0,
d
dG( , D)
=0
dD

(11.6.43)
(11.6.44)
(11.6.45)

Equation (11.6.43 and 11.6.44) dene the limit point and equation (11.6.45) is
the non-degeneracy condition of the limit point. Solving this system we nd that
k = kcr = 0.53. Thus there is no limit point if k < kcr = 0.53. It can be noted from
table (11.1) this critical value decreases as the relative volume of sludge disintegration unit decreases. For such a value of k the negative excess sludge production can
only achieved by having the residence time suciently high.

259

Chapter 11

In gure (11.3) we investigate the behavior of a membrane bioreactor when the volume of the sludge disintegration unit is equal to the volume of reactor, i.e. n = 1.
This is not likely to be the case in practise. Thus, in gure (11.4), we study how
the critical value of the sludge disintegration factor at a limit point depends on the
relative volume of the sludge disintegration unit (n) and the reaction rate (k). When
n=0.01, i.e, the volume of the sludge disintegration unit is 1% of the volume of reactor, the critical value of the sludge disintegration factor does not depend signicantly
on the value of reaction rate (k). As the relative volume of sludge disintegration

Dimensionless residence time ()

unit increases the dependency become more noticeable.

k= infinity
k=1020
k=10-2
k=10

4
3
2
1
0
0

0.02

0.04

0.06

0.08

0.1

Sludge disintegration factor (D)

Figure 11.3: The zero excess sludge production curve as a function of the sludge
disintegration rate (D) and the dimensionless residence time ( ). Parameter values
are: n=1. Other parameter values are stated in table (10.1)

260

Chapter 11

0.1

n=1
n=0.5
n=0.01

0.09

Dcr

0.08
0.07
0.06
0.05
0

0.2

0.4

0.6

0.8

1/k
Figure 11.4: The critical value of the sludge disintegration factor (Dcr ) as a function
of the reaction rate (k). Parameter values are stated in table (10.1).

Table 11.1: The critical value of reaction rate (kcr ) for dierent value of the relative
volume of the sludge disintegration unit (n)
n

kcr

n
kcr

0.531

1.88

0.5

0.265

1.88

0.1

0.0531

1.88

0.01

0.00531 1.88

261

Chapter 11

11.6.3

Conclusion

In this chapter we extended the Yoon model for a membrane reactor by allowing the
processes inside the sludge disintegration unit to occur at nite rate. The steadystate solutions of the model were found and their stability determined as a function
of the residence time. A transcritical bifurcation occurs when,
[

]
n
1
<D

,
1 Kd kx

(11.6.46)

The analysis of the stability of the steady state solutions have shown that the
washout solution is unstable when,
kx
< [1 Kd ],
n

(11.6.47)

The analysis of the model is carried out by using the steady state solution and considering two scenarios. The rst scenario is when the processes inside the sludge
disintegration unit occur innity quickly. It was found that when the sludge disintegration factor is less than D = 0.046 then the prediction of Yoon model is within
10% of the prediction of our model. However, the critical value of the sludge disintegration factor, corresponding to the limit point, predicted by Yoonmodel is 10%
higher than the value of our model.
In the second scenario which relates to nite rate kinetics, the eect of the reaction
rate and the relative volume of the sludge disintegration unit (n) were investigated.
It was found that there is a critical value of the reaction rate above which the prediction of the nite rate model is within 10% of the prediction of the innite rate
model. Thus, in such circumstance, the innite rate model is valid for practical
purpose.

Chapter 11

262

It was also found that there is a second critical value of the reaction rate (k). If the
reaction rate (k) is less than this critical value, then the sludge disintegration factor, corresponding to limit point above which the sludge produced in the reactor is
below the target value, is destroyed. Instead the state of the negative excess sludge
production can be only achieved when the residence time is very large.

Chapter 12
Conclusion
A standard assumption in bioprocess engineering is to model the specic growth rate
using Monod kinetics. However, as discussed in chapter (3), growing evidence shows
that Contois kinetic describes industrial process such as wastewater treatment very
well and in some case it is superior than Monod Kinetic. In this thesis, we presented mathematical models for waste-water treatment where specic growth rate
of microorganisms was given by Contois growth model and investigated multiple
scenarios involving positioning of recycle (using a settling unit) in a reactor cascade
identifying conditions suitable for negative excess sludge production. We obtained
steady-state solutions for various reactor models, performed stability analysis and
identied conditions which are unsuitable for reactor to operate. This thesis is organized into four parts.
In the rst part of this thesis, we presented a survey on existing literature about the
modeling of wastewaster treatment. Following that we presented a survey of many
studies in chapter (3) whose empirical data t very well with high correlation coecient when Contois model was preferred to Monod or other models and developed a
263

Chapter 12

264

case for employing Contois Kinetic model for modelling waste-water treatment. For
the remaining three parts, specic issues related to wastewater treatment modelling
are investigated.
In the second part which contains three chapters, we built our mathematical model
based on Contois method and analyzed reactor cascades with recycle. In chapter (4), we extended the study [101] and investigated a single reactor with a settling
unit followed by analysis of two dierent congurations for a n-reactor cascade in
chapters (5) and (6). In practice a reactor cascade is often used to improve the
performance. In chapters (5) and (6), we investigated two dierent congurations
for a reactor cascade. In chapter (5) we considered the case when a settling unit
placed around the reactor i recycles the euent stream from reactor i back to the
feed stream for the reactor i. We found a surprising result namely in order to optimized the performance of the reactor cascade when a settling unit placed around
the nal reactor, the perfect recycle should be used. However, if the settling unit is
placed around any proceeding reactor, the optimal reactor cascade performance is
generally not obtained using perfect recycle but is instead obtained with imperfect
recycle is used. This can be attributed to balancing the advantages of reducing the
substrate concentrations against the disadvantages of reducing the microorganism
concentrations entering the proceeding reactor.
In chapter (6), a dierent conguration cascade was considered in which the settling unit is placed after the nal reactor and the euent stream from the settling
unit is recycled back into the rst reactor. For a xed value of the recycle ratio R
and the concentration factor C, we found that there is a critical value of the total

Chapter 12

265

residence time. Below the critical value, the use of the settling unit improves the
performance of the reactor cascade. However, the performance of the cascade without recycle is better than having recycle if the residence time is above the criticality.
The comparison between the conguration in this chapter with the conguration in
chapter (5) was carried out. The obtained result was shown that the performance
of the cascade conguration where the recycle around each reactor is worst (better)
than the performance of the cascade conguration where recycle around the whole
cascade at low (high) total residence time.
In the third part we returned to the Contois model in single reactor by considering
a generalized Contois setting where either the yield coecient is allowed to be linear
function of the substrate concentration, and/or the Contois model was subject to
substrate inhibition. There is a great deal theoretical work analyzing such extension
for the Monod kinetics. The use of a variable yield coecient introduce the possibility of Hopf bifurcations and hence generate periodic solutions while the possibility of
substrate inhibition produces a limit point bifurcation on the steady state diagram.
In this part we applied singularity theory and Hopf bifurcation theory to identify the
range of the parameter region in which neither periodic or bistable behaviour can be
observed. We believe this is important from the practical standpoint of operating
the reactors where such region of operation is generally avoided.
In the nal part of this thesis, we investigated the used of sludge disintegration unit
to reduce sludge production inside a reactor. This is an important consideration as
in many applications the cost of sludge disposal can be 50-60% of the operating cost
of the system. In chapter (10), we adapted the model that originally proposed by

Chapter 12

266

Yoon [14] by replacing the Monod model by the Contois model. The result showed
that there is a critical value of the inuent concentration. Above the critical inuent
concentration a sludge disintegration unit is required to reduce the sludge production inside the reactor. This critical value is observed to be relatively large for ow
reactor whilst it is very low for membrane reactor.
The critical assumption in Yoons model is that the processes inside the sludge disintegration unit occur innitely quickly. It follows that the model consists of three
ordinary dierential equation for an activated sludge model coupled to a system of
two algebraic equations for the processes inside the sludge disintegration unit. A
natural equation to ask is what happens if the processes in the sludge disintegration
unit occur at nite rate. In formulating such a model we arrive at a surprising
conclusion, namely that the Yoons model [14] has limited applicability and that
too only when the sludge solubilization eciency is equal to one. This surprising
nding motivated chapter (10) of this thesis where we analyze a nite-rate sludge
disintegration unit model.

12.1

Future work

In this thesis, wastewater processes have been investigated using the Contois specic growth rate expression Further analysis of wastewater treatment using Contois
model is possible.
In chapter (9), we analyzed a system which featured substrate variable yield and the
substrate inhibition in the absence of the decay rate. A logical extension would be
to include decay rate in the model and analyze which could yield extra bifurcation

Chapter 12

267

patterns not observed here.

The second possible future work for the wastewater processes is to extend the models
studied in this thesis by allowing the model parameters to vary. Model parameters
such as max , Ks , and Kd , could be a function of the substrate concentration, the
biomass concentration or even the product concentration. Thus, it could be interesting to investigate a new bifurcation behavior by analyzing models with non-constant
parameters using singularity theory and dynamic methods.
The optimization of a reactor when a cascade is employed, could be investigated
when the model parameters are not constant. This has been done for the Monod
kinetic model by many researchers and may be extended to Contois kinetics with
and without non-constant parameters.
The authors in a recent book [4] analysis many reactor model such as DNA and
wall attachment models. However, in these models, it is mostly assumed the growth
rate is given by the Monod growth rate expression. Thus, it would be interesting to
analysis these models using the Contois growth rate expression.
Finally, we could replace the recycle ratio which is a linear function in this work,
with a more realistic recycle ratio which could emerge from an ordinary dierentiation equation as in [1]. This future directions could greatly supplement the work
presented here.

Appendix A
The analysis for Chapter 6
A.0.1

Investigation of w

We have,

w =

{
[
]1/n }
n1
n (1 + R) (1 + R)
RC
1 Kd

(1.0.1)

Dierentiating we obtain,
(
[
]1/n )
dw
1
n1
=
nR(1
+
R)

[1
+
nR]
(1
+
R)
RC
.
dR
R(1 + R)(1 Kd )

(1.0.2)

We rst consider the case n=1. We have,


1C
dw 
=
,

dR n=1 1 Kd

(1.0.3)

If C = 1, w is independent of R, whilst if C > 1, w is a decreasing function of R

as claimed.
We now show that for n > 1 that if C (1, Cmax ] then
dw
< 0.
dR
268

(1.0.4)

269

Appendix

We rst note that

1
dw
(C = Cmax ) =
< 0.
dR
R(1 Kd )

(1.0.5)

The function (1.0.2) is zero when,

z}|{ ( nR )n
1
(1 + ).
C= C =
1 + nR
R

(1.0.6)

z}|{
Our result will follow if we can show that C < 1.
Calculation shows that,
z}|{ (
)n
d C
nR
1
=
(
)(n 1) > 0.
dR
1 + nR
(1 + nR)R2

(1.0.7)

It follows that,
z }| {
z}|{
C(R < ) < lim C = 1.
R

A.1

(1.0.8)

Parameter value for Monod kinetics

Monod kinetics have been used extensively to study the growth rate of biomass in a
cascade reactor, such as [80]. Unlike the Contois kinetics, the Monod kinetics only
depends upon the concentration of the substrate. Grady and Lim [80] conducted a
comprehensive investigation of the behavior of a cascade reactor with recycle for the
biological treatment of wastewater in which the specic growth-rate law followed
the Monod model.
Here, we investigate the eect of the recycle ratio on the performance of a tworeactor system as a function of the total residence time replacing Contois kinetics
with Monod kinetics. Using the transformations introduced in [132], equations,

270

Appendix

(6.2.1-6.2.4) can be written into the following dimensionless form,

Equations in the rst reactor.
dS1
n
nR
S1 X1

(S
=

S
)

+ (Sn S1 ),
0
1

dt
t
(S1 + 1)
t
S1 X1
nX1
nR
dX1
+
=

+ (CXn X1 ) Kd X1 .

dt
t
(S1 + 1)
t

(1.1.9)
(1.1.10)

Equations in the ith reactor.

dSi
Si Xi
n(1 + R)

(Si1 Si )
=
,
dt
t
(Si + 1)

Xi )
n(1 + R)(Xi1
dXi
Si Xi
=
+
Kd Xi .
dt
t
(Si + 1)

(1.1.11)
(1.1.12)

For more details of the parameter denitions, we refer to reference [132].

The two-reactor system is operated at steady state for three dierent recycle ratios,
i.e., R = 0, 0.05, and 1.0. Figure (A.1) shows the three corresponding curves obtained using the parameter values reported by [14].
When 0 < R < Rmax = 1, we observe the same three distinct regions that were
described in section (6.3.4.1). When R = Rmax = 1, we note that as the value of the
total residence time range between t = 0.063 to t = 100.063 the euent concentrations increases insensitively from S2 = 0.00016 to S2 = 0.00017. Thus, the value
of the euent concentration is very insensitive to the variation in the total residence
time. This is not the case when the growth rate depends on both the biomass and
the substrate concentration (Contois kinetics, see section (6.3.4.2)).
A further interesting point is that, for a single reactor with perfect recycle using Monod kinetics, the euent concentration does not depend on the residence
time [132]. However, for the case in which the growth rate follows Contois kinetics,
the euent concentration does depend on the residence time [101].

271

Appendix

The value of the total residence time at the intersection points of the second and
third regions are higher in comparison to the condition when Contois kinetics is
used. This can be seen in table (6.1). In the third region, the optimum value of
the recycle ratio is zero, i.e., the use of a settling unit in this region increases the
euent concentration as for Contois kinetics.
We conclude that there is a signicant dierent in model prediction for the case R =
Rmax = 1. When 0 < R < Rmax = 1, the general behavior is the same but the tran-

*
Dimensionless substrate concentration ( S 2)

sition occur at higher value of total residence time using Monod Model.

100

No recycle
R=0.5
R=1.0

10
1
0.1
0.01
0.001
0.0001
1e-005
0

20
40
60
80
100

Dimensionless total residence time ( )

Figure A.1: Euent concentration in a cascade of a two reactors with recycles

around the whole cascade using Monod kinetics. Parameter value: C = 2, Rmax = 1,
S0 = 4000, max = 1, Ks =100, = 0.5, Kd =0.028.

272

Appendix

A.2

Stability of the washout solution

The eigenvalues of the Jacobian matrix for a n-reactor cascade at the washout steady
state solution are given by,
Two reactors
2

1 =
2 =
3 =
4 =

)
(1 + R) R (1 + R)

(1.2.13)

(1 + R) R + 1 + R

,
(
)
2
(1 + R) CR (1 + R) + t (1 Kd )

(1.2.14)

(1.2.15)

(1.2.16)

)
(1 + R) CR + 1 + R + t (1 Kd )

It can be noted that 3 > 4 , 3 > 1 > 2 . Thus, the necessary condition for the
eigenvalues to be negative is 3 < 0.
Three reactors
(
)
2
3
3
(1 + R) R (1 + R)
1 =
,
(1.2.17)
t
(
)
2
3

3
(1 + R) R + 2(1 + R)
i 3 3 3 (1 + R)2 R
2 =
+
,
(1.2.18)
2t
2t
(
)
2
3

3
(1 + R) R + 2(1 + R)
i 3 3 3 (1 + R)2 R
3 =
(1.2.19)

,
2t
2t
(
)
2
3
3
(1 + R) CR (1 + R) + t (1 Kd )
4 =
,
(1.2.20)
t
)
(
2
3

(1 + R) R + 2(1 + R) + t (1 Kd )
3
i 3 3 3 (1 + R)2 R
5 =
+
,
2t
2t
(1.2.21)

273

Appendix

(
)
2
3
3
(1 + R) R + 2(1 + R) + t (1 Kd )
6 =

2t

i 3 3 3 (1 + R)2 R
2t

,
(1.2.22)

It can be noted that 4 > Re(5 ) = Re(6 ), 4 > 1 > Re(2 ) = Re(3 ). Thus, the
necessary condition for the eigenvalues to be negative is 4 < 0.
Four reactors

)
(
3
4
(1 + R) R (1 + R)
4
1 =
(
4

(1.2.23)

(1 + R) R + 1 + R

2 =

(1.2.24)

4(1 + R) i 4
3 =
+
t

(1 + R)3 R
,

(1.2.25)

i 4 4 (1 + R)3 R

4 =

4(1 + R)

,
t
t
)
(
3
4
(1 + R) RC (1 + R) + t (1 Kd )
4

5 =
4

(
4

(1.2.27)

(1 + R) RC + 1 + R + t (1 Kd )
3

6 =

(1.2.28)

4(1 + R) + t (1 Kd ) i 4
7 =
+
t
8 =

(1.2.26)

4(1 + R) +
t

t (1

It can be noted that 5 > Re(i ),

Kd )

i4

(1 + R)3 R
t

(1.2.29)

(1.2.30)

4
(1 + R)3 R

i = 1, 2, ..., 8,

i = 5. Thus, the necessary

condition for the eigenvalues to be negative is 5 < 0.

Five reactors
(
)
4
5
5
(1 + R) R (1 + R)
1 =
,
t

(1.2.31)

274

Appendix

(
5

5/4

(1 + R) R( 5 1) 4(1 + R)

2 =

5i 5 (1 + R)4 R
+

t
(
5

5/4

(1 + R)4 R( 5 1) 4(1 + R)

3 =

5i 5 (1 + R)4 R 10 + 2 5

(
5
6 =

4t

5i 5 (1 + R)4 R 10 2 5
4t

)
(1 + R) R (1 + R) + t (1 Kd )

(1.2.36)

5i 5 (1 + R)4 R
+

t
(

(1 + R)4 R( 5 1) 4(1 + R) + t (1 Kd )

5i 5 (1 + R)4 R 10 + 2 5

4t

(1.2.38)

5i 5 (1 + R)4 R 10 2 5
4t

(1.2.39)

(
)

4
5
5/4
(1 + R) R( 5 + 1) + 4(1 + R) + t (1 Kd )
10 =

(1.2.37)

10 + 2 5

4t

(
)

4
5
5/4
(1 + R) R( 5 + 1) + 4(1 + R) + t (1 Kd )
9 =

(1.2.35)

,
t
(
)

4
5
5/4
(1 + R) R( 5 1) 4(1 + R) + t (1 Kd )

5/4

7 =

8 =

5i 5 (1 + R)4 R 10 2 5

(1.2.34)

)
(

4
5
(1 + R) R( 5 + 1) + 4(1 + R)
5/4
5 =

4t

(1.2.33)

(
)

4
5
5/4
(1 + R) R( 5 + 1) + 4(1 + R)
4 =

4t

(1.2.32)

10 + 2 5

5i 5 (1 + R)4 R 10 2 5
4t
(1.2.40)

275

Appendix

It can be noted that 6 > Re(i ),

i = 1, 2, ..., 10,

condition for the eigenvalues to be negative is 6 < 0.

i = 6. Thus, the necessary

Appendix B
The analysis for Chapter 9

B.1

Analysis of the quadratic equation

We have
c 2 + bS
c + c = 0.
aS
i
i

(2.1.1)

The coecients a, b, and c are given by

a = e ,

b = ( + ) + ,

c = .

We have the following restriction on parameter values: e > 0, > 0, 0

1, > 0. The coecient c strictly positive. We consider the case a > 0, which
happen when,
< e .

(2.1.2)

The coecients a and c, are strictly positive. If b 0, then the roots of equation (2.1.1), should they are exist, are both non-positive. Conversely, if b < 0,
276

277

Appendix

then the roots of equation (2.1.1), should they are exist, are both positive. Thus, a
necessary condition for positive solutions of equation (2.1.1) is,

b < 0 > ( + ) = min,1

.

(2.1.3)

We now turn out attention to the discriminant () of equation (2.1.1). We have

=b2 4ac = ( 1 )( 2 ),

(2.1.4)

where

1 = ( + ) + 2 ,

2 = ( + ) 2

= (e ) > 0.

When > 0, = 0, and < 0 then equation (2.1.1) has two, one and no real
solutions respectively. When the roots of the discriminant are real they are labelled
1 and 2 with 1 > 2 . For what follows it is useful to note that when the roots of
the discriminant are real we have

1 > ( + ) = min,1
,

(2.1.5)

2 < ( + ) = min,1
.

(2.1.6)

Recall that the condition b < 0 requires > min,1

. Hence when > 1 the solu-

tions of equation (2.1.1) are both positive and may be physically meaningful. There
is a limit point at = 1 . The solutions corresponding to the case < 2 are
not of interest because they are negative as b > 0.

B.2

Stability analysis for the case a=0.

Using the particular solution (9.2.15) for this case, we obtain,

det(J) =

(BR
) (BR
)
.
3

(2.2.7)

278

Appendix

From the condition for the solution (9.2.15) to be physically meaningful, it follows
that the determinant (2.2.7) is positive and as a result a double zero eigenvalue
degeneracy does not occur for case (2).
From equation (2.2.7), the rst condition of a Hopf bifurcation to occur is satised.
The conditions (9.2.33, 9.2.36 and 9.2.37) are solved numerically. Using = 0.01
and = 1, the location of the double-hopf bifurcation is given by,
( , ) = (0.02254, 1.06985).

(2.2.8)

Figure (B.1) is an unfolding diagram for the Hopf bifurcation points in the
plane. This gure shows the values of the residence time at which a Hopf bifurcation
point occurs as a function of . It can be observed that there are always two Hopf
bifurcation points for > 0.02254 as there is only one H21 degeneracy given by
equation (2.2.8).

B.3

Analysis of + b

In the case of physically meaningful solution ( 0), we study the behavior of

function (G = + b

). The function G is equal to zero when = 0, i.e = 1 .

It is a clearly positive when b 0. From now on we assume that b < 0. After some
algebra the function G can be written in the form,

G=+b

4 (e )

=
, = 0,
1 b 1/

negative
if (e ) > 0, (Case 3).
=

positive
if (e ) < 0, (Case 1).

(2.3.9)

(2.3.10)

279

Appendix

12
10

8
6
4
2
0
0

10

Figure B.1:

Unfolding diagram for case (2) (a=0) showing the Hopf bifurcation

locus. Parameter values: =1, = 0.01. All the symbols are explained in section (9.2).

Appendix C
The analysis for Chapter 8

C.1

Claim 1.

The steady state solution is not

realist when 2
The expression for 2 can be rewritten
2 e Kd

2 =min
+
(1 Kd )2

{
1

(1 Kd )( Kd )
1+
2
e Kd

}
(3.1.1)

It follows that if Kd then 2 min

. Thus we deduce that when Kd

and 2 then equation (8.2.9) has two negative solutions, i.e no physically
meaningful solutions.
For the case < Kd , we have,
Si (

2 )

Kd

=
+
> 1,
Kd Kd

as

< Kd ,

(3.1.2)

Thus Si (2 ) is not physically meaningful solution.

We show in appendix (C.2) that BR

1 . Thus we deduce that when < Kd and

2 then equation (8.2.9) has no realistic solutions. We conclude that equation

280

281

Appendix

(8.2.9) has no realistic solutions when 2 and no limit point bifurcation.

C.2

BR
1, For any value of e on [a,

Claim 2.
1Kd
Kd ).

Here we need to show that BR

1 . To do this we will consider when 1 (e) is

only real number. This happen when 0 thus we will study only the case when
0, i.e, e a =

(1Kd )( Kd )
Kd

The function 1 (e) and BR

(e) are continuous on the interval [a=
1Kd
)
Kd

(1Kd )( Kd )
Kd

and satisfy the following conditions,

Kd ( + ) 2
Kd ( + )

<

(a)
=
,
BR
Kd (1 Kd )
Kd (1 Kd )

)
+
2
/Kd
1

K
1 Kd
d

1 (
)
=
<

(
) = .
BR
Kd
1 Kd
Kd
1 (a) =

Since there is only one value of e on [a,

e = ecr =

1Kd
)
Kd

(3.2.3)
(3.2.4)

for which 1 (e)=BR

(e) given by

1 Kd
1 Kd
<
.
Kd
Kd +

(3.2.5)

Hence, according to the continuity of BR

and 1 (e), we conclude that BR
1

on [a,

C.3

1Kd
).
Kd

Analysis of + b

In the case of a physically meaningful solution ( 0), we study the behavior of

the function (G = + b

). The function G is equal to zero when = 0, i.e

= 1 . It is positive when b 0. From now on we assume that b < 0. After some

282

Appendix

algebra the function G can be written in the form,

G=+b

4e (Kd + )

=
< 0,
1 b 1/

= 0,

(3.3.6)

Appendix D
List of Symbols
Symbol

Units

Description

Ci

The recycle concentration factor for the settling unit on

reactor i.

dm3 day1 The ow rate through the reactor cascade.

Kd

day1

Kd

The death coecient.

The dimensionless death rate [Kd = Kd /max ].

Ks

|X||S|1

The saturation constant.

Ri

The recycle ratio for the settling unit on reactor i based on

volumetric ow rates.

Ri

The dimensionless recycle ratio for the settling unit on reactor i based on volumetric ow rates [Ri = Ri (Ci 1)].

Si

|S|

The substrate concentration within the ith reactor of the

cascade.

Si

The dimensionless substrate concentration within the ith

reactor of the cascade [Si = Si /S0 ].
283

284

Appendix

S0

|S|

The concentration of substrate owing into the rst reactor

of the cascade.

dm3

The volume of the reactor in the cascade.

Xi

|X|

The microorganisms concentration within the ith reactor

of the cascade.

Xi

The dimensionless microorganisms concentration within

the ith reactor of the cascade [Xi = Ks Xi /S0 ].

X0

|X|

The concentration of the microorganisms owing into the

rst reactor the cascade.

X0

The dimensionless concentration of the microorganisms

owing into the rst reactor the cascade [X0 = Ks X0 /S0 ].

day

The time.

The dimensionless time [t = max t].

|X||S|1

The yield factor.

The dimensionless yield factor [ = Ks ].

(S, X)

day1

The specic growth rate model.

max

day1

The maximum specic growth rate.

day1

The specic growth rate.

day

The total residence time.

The dimensionless total residence time [t = nV max /F ].

Reactor parameter model.

KI

|S|

The inhibition coecient.

The dimensionless inhibition coecient [e = S0 /KI ].

|X||S|2

The constant in the yield coecient (Related to chapters (7,

9)).

285

Appendix

The dimensionless constant in the yield coecient [ =

Ks S0 ] (Related to chapters (7, 9)).

The sludge disintegration factor.

|P |

The concentration of particulates .

The dimensionless particulates concentration [P

Ks P/S0 ].
kh

day1

The hydrolysis rate of the particulates.

kh

The dimensionless hydrolysis coecient [kh = kh /max ].

|S||X|1

The conversion eciency from non-biodegradable particulates to biodegradable (Related to chapter (10)).

The dimensionless particulate to substrate conversion factor [ = /Ks ] (Related to chapter (10)).

i , i

The sludge solubilization eciency (0 1) .

|S||X|1

The conversion eciency from non-biodegradable particu-

1, 2
i i

lates to biodegradable (Related to chapter (11)).

=

tor [ = /Ks ] (Related to chapter (11)).

1, 2
i i = 1, 2

The dimensionless particulate to substrate conversion fac-

The sludge solubilization eciency (0 1) (Related to

chapter (11)).

kx

The rate of the biomass disintegration in the sludge disintegration unit.

kx

The dimensionless rate of the biomass disintegration in the

sludge disintegration unit [kx = kx /max ].

kp

The rate of the particulates disintegration in the sludge

disintegration unit.

286

Appendix

kp

The dimensionless rate of the particulates disintegration in

the sludge disintegration unit [kx = kx /max ].

Vs

dm3

The volume of the sludge disintegration unit.

The relative volume of the sludge disintegration unit (n =

V
).
Vs

Bibliography
[1] M. Serhani, N. Raissi, and P. Cartigny, Robust feedback control design for
a nonlinear wastewater treatment model, Mathematical Modelling of Natural
Phenomena, vol. 4, no. 5, pp. 128 143, 2009.
[2] A. Ajbar, Study of the operability of nonideal continuous bioreactors, Chemical Engineering Communications, vol. 198, no. 3, pp. 385415, 2010.
[3] A. Ajbar, M. AlAhmad, and E. Ali, On the dynamics of biodegradation of
wastewater in aerated continuous bioreactors, Mathematical and Computer
Modelling, vol. 54, no. 9, pp. 19301942, 2011.
[4] A. Ajbar and K. Alhumaizi, Dynamics of the Chemostat: A Bifurcation Theory
Approach. CRC Press, 2011.
[5] K. Alhumaizi and A. Ajbar, Optimization of an unstructured rst-order kinetics model of cyclically operated bioreactors, Journal of environmental engineering, vol. 132, no. 5, pp. 453462, 2006.
[6] A. Ajbar, Stability analysis of the biodegradation of mixed wastes in a continuous bioreactor with cell recycle, Water research, vol. 35, no. 5, pp. 1201
1208, 2001.
287

Appendix

288

[7] A. Ajbar and I. Gamal, Stability and bifurcation of an unstructured model of

a bioreactor with cell recycle, Mathematical and Computer Modelling, vol. 25,
no. 2, pp. 3148, 1997.
[8] A. Ajbar and G. Ibrahim, Periodic and nonperiodic oscillatory behavior in a
model for activated sludge reactors, Mathematical and Computer Modelling,
vol. 25, no. 10, pp. 927, 1997.
[9] J. Harmand, A. Rapaport, and A. K. Drame, Optimal design of two interconnected enzymatic bioreactors, Journal of Process Control, vol. 14, no. 7,
pp. 785794, 2004.
[10] J. Harmand, A. Rapaport, and A. Trono, Optimal design of interconnected
bioreactors: New results, AIChE journal, vol. 49, no. 6, pp. 14331450, 2003.
[11] J. Harmand and D. Dochain, The optimal design of two interconnected
(bio)chemical reactors revisited, Computers and Chemical Engineering,
vol. 30, no. 1, pp. 7082, 2005.
[12] J. Monod, The growth of bacterial cultures, Annual Reviews in Microbiology,
vol. 3, no. 1, pp. 371394, 1949.
[13] D. Contois, Kinetics of bacterial growth: relationship between population
density and specic growth rate of continuous cultures, Journal of General
Microbiology, vol. 21, no. 1, pp. 4050, 1959.
[14] S.-H. Yoon, Important operational parameters of membrane bioreactorsludge disintegration (mbr-sd) system for zero excess sludge production, Water research, vol. 37, no. 8, pp. 19211931, 2003.

Appendix

289

[15] I. J. Dunn, E. Heinzle, J. Ingham, and J. E. Prenosil, Biological reaction engineering: dynamic modelling fundamentals with simulation examples, vol. 2.
Wiley-VCH, 2003.
[16] S. J. Pirt, Principles of microbe and cell cultivation. Blackwell Scientic Publications., 1975.
[17] B. Demirel, O. Yenigun, and T. T. Onay, Anaerobic treatment of dairy
wastewaters: a review, Process Biochemistry, vol. 40, no. 8, pp. 25832595,
2005.
[18] R. E. Speece, Anaerobic biotechnology for industrial wastewater treatment,
Environmental science & technology, vol. 17, no. 9, pp. 416427, 1983.
[19] H. Gavala, I. Skiadas, N. A. Bozinis, and G. Lyberatos, Anaerobic codigestion of agricultural industries wastewaters, Water Science and Technology,
vol. 34, no. 11, pp. 6775, 1996.
[20] K. Rajeshwari, M. Balakrishnan, A. Kansal, K. Lata, and V. Kishore, Stateof-the-art of anaerobic digestion technology for industrial wastewater treatment, Renewable and Sustainable Energy Reviews, vol. 4, no. 2, pp. 135156,
2000.
[21] G. Lettinga, Anaerobic digestion and wastewater treatment systems, Antonie van leeuwenhoek, vol. 67, no. 1, pp. 328, 1995.
[22] A. Wheatley, Anaerobic digestion: a waste treatment technology. Chapman &
Hall, 1990.

Appendix

290

[23] F. Garca-Ochoa, V. Santos, L. Naval, E. Guardiola, and B. Lopez, Kinetic

model for anaerobic digestion of livestock manure, Enzyme and microbial
technology, vol. 25, no. 1, pp. 5560, 1999.
[24] F. Mosey, Mathematical modelling of the anaerobic digestion process: regulatory mechanisms for the formation of short-chain volatile acids from glucose,
Water Science & Technology, vol. 15, no. 8-9, pp. 209232, 1983.
[25] A. Moser, Bioprocess technology: kinetics and reactors. Springer, 1988.
[26] D.-G. Mou and C. L. Cooney, Growth monitoring and control through
computer-aided on-line mass balancing in a fed-batch penicillin fermentation,
Biotechnology and bioengineering, vol. 25, no. 1, pp. 225255, 1983.
[27] K. Hanaki and T. Matsuo, Mathematical modelling of the anaerobic digestion process, in Mathematical Models in Biological Waste Water Treatment
(S. Jorgensen and M. Gromiec, eds.), (Amsterdam, Netherlands), ElsevierScience, 1985.
[28] J. F. Andrews, A mathematical model for the continuous culture of microorganisms utilizing inhibitory substrates, Biotechnology and Bioengineering,
vol. 10, no. 6, pp. 707723, 1968.
[29] G. Tessier, Les lois quantitatives de la croissance, Annales de Physiologie et
de Physiochimie Biologique, vol. 12, pp. 527573, 1936.
[30] H. Moser, The dynamics of bacterial populations maintained in the chemostat.
Carnegie Institution of Washington, 1958.

Appendix

291

[31] Y. Chen and A. Hashimoto, Substrate utilization kinetic model for biological treatment process, Biotechnology and Bioengineering, vol. 22, no. 10,
pp. 20812095, 1980.
[32] D. Hill, Simplied monod kinetics of methane fermentation of animal wastes,
Agricultural Wastes, vol. 5, p. 116, 1983.
[33] D. M. Bagley and T. S. Brodkorb, Modeling microbial kinetics in an anaerobic
sequencing batch reactor model development and experimental validation,
Water Environment Research, vol. 71, pp. 13201332, 1999.
[34] D. J. Batstone, J. Keller, I. Angelidaki, S. Kalyuzhnyi, S. Pavlostathis,
A. Rozzi, W. Sanders, H. Siegrist, and V. Vavilin, The IWA anaerobic digestion model no 1(ADM 1), Water Science & Technology, vol. 45, no. 10,
pp. 6573, 2002.
[35] H. Siegrist, D. Vogt, J. Garcia-Heras, and W. Gujer, Mathematical model
for meso and thermophilic anaerobic sewage sludge digestion, Environmental
Science and Technology, vol. 36, pp. 113123, 2002.
[36] M. Henze, C. Grady, W. Gujer, G. Marais, and T. Matsuo, A general model
for single-sludge wastewater treatment systems, Water Research, vol. 21,
no. 5, pp. 505515, 1987.
[37] M. Henze, W. Gujer, T. Mino, T. Matsuo, M. Wentzel, and G. Marais, Activated Sludge Model No. 2. IAWQ Scientic and Technical Report No.3. IAWQ,
1995.

Appendix

292

[38] M. Nelson and H. Sidhu, Analysis of the activated sludge model (number 1),
Applied Mathematics Letters, vol. 22, no. 5, pp. 629635, 2009.
[39] W. Hu, K. Thayanithy, and C. Forster, A kinetic study of the anaerobic digestion of ice-cream wastewater, Process Biochemistry, vol. 37, no. 9, pp. 965
971, 2002.
[40] M. Pandian, N. Huu-Hao, and S. Pazhaniappan, Substrate removal kinetics
of an anaerobic hybrid reactor treating pharmaceutical wastewater, Journal
of Water Sustainability, vol. 1, no. 3, pp. 301312, 2011.
[41] M. Isik and D. T. Sponza, Substrate removal kinetics in an upow anaerobic sludge blanket reactor decolorising simulated textile wastewater, Process
Biochemistry, vol. 40, no. 3, pp. 11891198, 2005.
[42] W. Hu, K. Thayanithy, and C. Forster, Kinetic study of anaerobic digestion
of sulphate-rich wastewaters from manufacturing food industries, in 7th International Conference on Environmental Science and Technology Ermoupolis,
(Syros island, Greece), Sept 2001.
[43] W. Gujer, M. Henze, T. Mino, and M. v. Loosdrecht, Activated sludge model
no. 3, Water Science and Technology, vol. 39, no. 1, pp. 183193, 1999.
[44] S. Kalyuzhnyi, Batch anaerobic digestion of glucose and its mathematical
modeling. ii. description, verication and application of model, Bioresource
technology, vol. 59, no. 2, pp. 249258, 1997.

293

Appendix

Gonzalez-Barcelo, and S. Gonzalez-Martnez,

[45] G. Cuevas-Rodrguez, O.
Wastewater fermentation and nutrient removal in sequencing batch reactors,
Water science and technology, vol. 38, no. 1, pp. 255264, 1998.
[46] V. Munch, J. Keller, P. Lant, and R. Newell, Mathematical modelling of
prefermentersi. model development and verication, Water research, vol. 33,
no. 12, pp. 27572768, 1999.
[47] A. N. Knobel and A. E. Lewis, A mathematical model of a high sulphate
wastewater anaerobic treatment system, Water research, vol. 36, no. 1,
pp. 257265, 2002.
[48] J. Monod,

Recherches mr la Croissance des Cultures Bactkriennes.

Pans,Hermann et Cie, 1942.

[49] M. Nelson and H. Sidhu, Reducing the emission of pollutants in food processing wastewaters, Chemical Engineering and Processing: Process Intensication, vol. 46, no. 5, pp. 429436, 2007.
[50] A. Lehninger, D. L. Nelson, and M. M. Cox, Lehninger Principles of Biochemistry. W. H. Freeman, fth edition ed., 2008.
[51] S. Simkins and M. Alexander, Models for mineralization kinetics with the
variables of substrate concentration and population density, Applied and Environmental Microbiology, vol. 47, no. 6, pp. 12991306, 1984.
[52] S. K. Schmidt, M. Alexander, and M. L. Shuler, Predicting threshold concentrations of organic substrates for bacterial growth, Journal of theoretical
biology, vol. 114, no. 1, pp. 18, 1985.

Appendix

294

[53] P. Verhulst, Recherches mathmatiques sur la loi daccroissement de la population, M e memories of Acad e mie Royale des Sciences, Letters and Fine
Arts of Belgium, vol. 18, pp. 138, 1845.
[54] P. Verhulst, Deuxime mmoire sur la loi daccroissement de la population,
M e memories of Acad e mie Royale des Sciences, Letters and Fine Arts of
Belgium, vol. 20, pp. 118, 1847.
[55] B. Gompertz, On the nature of the function expressive of the law of human
mortality, and on a new mode of determining the value of life contingencies,
Philosophical transactions of the Royal Society of London, vol. 115, pp. 513
583, 1825.
[56] W. Penfold and D. Norris, The relation of concentration of food supply to
generation time bacteria, Journal of Hygiene, vol. 12, pp. 527531, 1912.
[57] E. Powell, C. Evans, R. Strange, and D. Tempest, Microbial physiology and
continuous culture. HM Stationery Oce, 1967.
[58] G. Okpokwasili and C. Nweke, Microbial growth and substrate utilization
kinetics, African Journal of Biotechnology, vol. 4, pp. 305317, 2006.
[59] A. M. Gibson, N. Bratchell, and T. Roberts, The eect of sodium chloride
and temperature on the rate and extent of growth of clostridium botulinum
type a in pasteurized pork slurry, Journal of Applied microbiology, vol. 62,
no. 6, pp. 479490, 1987.

Appendix

295

[60] D. Ratkowsky, R. Lowry, T. McMeekin, A. Stokes, and R. Chandler, Model

for bacterial culture growth rate throughout the entire biokinetic temperature
range, Journal of Bacteriology, vol. 154, no. 3, pp. 12221226, 1983.
[61] W. G. Characklis and W. Gujer, Temperature dependency of microbial reactions, Prog. Water Technol. Suppl, vol. 1, pp. 111130, 1979.
[62] K. Kovarova, A. Zehnder, and T. Egli, Temperature-dependent growth kinetics of escherichia coli ml 30 in glucose-limited continuous culture, Journal
of bacteriology, vol. 178, no. 15, pp. 45304539, 1996.
[63] T. E. Shehata and A. G. Marr, Eect of nutrient concentration on the growth
of escherichia coli, Journal of bacteriology, vol. 107, no. 1, pp. 210216, 1971.
[64] J. Dabes, R. Finn, and C. Wilke, Equations of substrate-limited growth: The
case for blackman kinetics, Biotechnology and bioengineering, vol. 15, no. 6,
pp. 11591177, 1973.
[65] J. Heijnen and B. Romein, Derivation of kinetic equations for growth on
single substrates based on general properties of a simple metabolic network,
Biotechnology progress, vol. 11, no. 6, pp. 712716, 1995.
[66] A. Mulchandani and J. Luong, Microbial inhibition kinetic revised, Enzyme
and Microbial Technology, vol. 11, pp. 6673, 1989.
[67] Y. Tan, Z.-X. Wang, R. Schneider, and K. C. Marshall, Modelling microbial growth: a statistical thermodynamic approach, Journal of biotechnology,
vol. 32, no. 2, pp. 97106, 1994.

Appendix

296

[68] G. Teissier, Croissance des populations bacteriennes et quantitedaliment

disponible, Rev. Sci, vol. 80, p. 209, 1942.
[69] H. V. Westerho, J. S. Lolkema, R. Otto, and K. J. Hellingwerf, Thermodynamics of growth non-equilibrium thermodynamics of bacterial growth the
phenomenological and the mosaic approach, Biochimica et Biophysica Acta
(BBA)-Reviews on Bioenergetics, vol. 683, no. 3, pp. 181220, 1982.
[70] J. Nielsen and J. Villadsen, Modelling of microbial kinetics, Chemical Engineering Science, vol. 47, no. 17, pp. 42254270, 1992.
[71] S. Kooijman, E. Muller, and A. Stouthamer, Microbial growth dynamics on
the basis of individual budgets, Antonie Leeuwenhoek, vol. 60, pp. 159174,
1991.
[72] M. Droop, 25 years of algal growth kinetics a personal view, Botanica marina, vol. 26, no. 3, pp. 99112, 1983.
[73] E. Powell, Theory of the chemostat, Laboratory practice, vol. 14, p. 1145,
1965.
[74] J. Haldane, Enzymes. Longmans, London, 1930.
[75] Y.-S. Oh, Z. Shareefdeen, B. C. Baltzis, and R. Bartha, Interactions between
benzene, toluene, and p-xylene (btx) during their biodegradation, Biotechnology and Bioengineering, vol. 44, no. 4, pp. 533538, 1994.
[76] W. Sokol and J. Howell, Kinetics of phenol oxidation by washed cells,
Biotechnology and Bioengineering, vol. 23, no. 9, pp. 20392049, 1981.

Appendix

297

[77] V. H. Edwards, The inuence of high substrate concentrations on microbial

kinetics, Biotechnology and Bioengineering, vol. 12, no. 5, pp. 679712, 1970.
[78] R. Der Yang and A. E. Humphrey, Dynamic and steady state studies of
phenol biodegradation in pure and mixed cultures, Biotechnology and Bioengineering, vol. 17, no. 8, pp. 12111235, 1975.
[79] W. Sokol, Oxidation of an inhibitory substrate by washed cells (oxidation of
phenol by pseudomonas putida), Biotechnology and bioengineering, vol. 30,
no. 8, pp. 921927, 1987.
[80] C. Grady and H. Lim, Biological Wastewater Treatment. New York and Basel:
Marcel Dekker, Inc., 1999.
[81] W.-T. Tang and L.-S. Fan, Steady state phenol degradation in a drafttube, gas-liquid-solid uidized-bed bioreactor, AIChE journal, vol. 33, no. 2,
pp. 239249, 1987.
[82] K. Han and O. Levenspiel, Extended monod kinetics for substrate, product,
and cell inhibition, Biotechnology and Bioengineering, vol. 32, no. 4, pp. 430
447, 1988.
[83] A. G. Livingston and H. A. Chase, Modeling phenol degradation in a
uidized-bed bioreactor, AIChE journal, vol. 35, no. 12, pp. 19801992, 1989.
[84] U. Pawlowsky and J. Howell, Mixed culture biooxidation of phenol. i. determination of kinetic parameters, Biotechnology and Bioengineering, vol. 15,
no. 5, pp. 889896, 1973.

Appendix

298

[85] S. Seker, H. Beyenal, B. Salih, and A. Tanyolac, Multi-substrate growth

kinetics of pseudomonas putida for phenol removal, Applied microbiology and
biotechnology, vol. 47, no. 5, pp. 610614, 1997.
[86] A. Rozich and A. Gaudy Jr, Response of phenol-acclimated activated sludge
process to quantitative shock loading, Journal of Water Pollution Control
Federation, pp. 795804, 1985.
[87] Y. Tan, Z.-X. Wang, and K. C. Marshall, Modeling substrate inhibition of
microbial growth, Biotechnology and bioengineering, vol. 52, no. 5, pp. 602
608, 1996.
[88] C. T. Goudar, S. H. Ganji, B. G. Pujar, and K. A. Strevett, Substrate inhibition kinetics of phenol biodegradation, Water environment research, vol. 72,
no. 1, pp. 5055, 2000.
[89] M. Nelson and H. Sidhu, Analysis of a chemostat model with variable yield
coecient: Tessier kinetics, Journal of mathematical chemistry, vol. 46, no. 2,
pp. 303321, 2009.
[90] P. S. Crooke, C.-J. Wei, and R. D. Tanner, The eect of the specic growth
rate and yield expressions on the existence of oscillatory behavior of a continuous fermentation model, Chemical Engineering Communications, vol. 6,
no. 6, pp. 333347, 1980.
[91] L. Zhu, Limit cycles in chemostat with constant yields, Mathematical and
computer modelling, vol. 45, no. 7, pp. 927932, 2007.

Appendix

299

[92] T. Sari, Comments on limit cycles in the chemostat with constant yields
mathematical and computer modelling 45 (2007) 927932, Mathematical and
Computer Modelling, vol. 52, no. 9, pp. 18221824, 2010.
[93] G. Song and X. Li, Stability of solution to the chemostat system with nonconstant consuming rate, J. Biomath, vol. 14, no. 1, pp. 2427, 1999.
[94] X. Huang and L. Zhu, A three dimensional chemostat with quadratic yields,
Journal of mathematical chemistry, vol. 38, no. 4, pp. 575588, 2005.
[95] X. Huang and L. Zhu, Bifurcation in the stable manifold of the bioreactor with nth and mth th order polynomial yields, Journal of Mathematical
Chemistry, vol. 46, pp. 199213, 2009.
[96] T. Sari, A lyapunov function for the chemostat with variable yields, Comptes
Rendus Mathematique, vol. 348, no. 13, pp. 747751, 2010.
[97] G. S. Wolkowicz and Z. Lu, Global dynamics of a mathematical model of competition in the chemostat: general response functions and dierential death
rates, SIAM Journal on Applied Mathematics, vol. 52, no. 1, pp. 222233,
1992.
[98] C. Lopez, M. Pons, and E. Morgenroth, Endogenous processes during longterm starvation in activated sludge performing enhanced biological phosphorus
removal, Water Research, vol. 40, no. 8, pp. 15191530, 2006.
[99] R. Manser, W. Gujer, and H. Siegrist, Decay processes of nitrifying bacteria
in biological wastewater treatment systems, Water Research, vol. 40, no. 12,
pp. 24162426, 2006.

Appendix

300

[100] B.-J. Ni and H.-Q. Yu, An approach for modeling two-step denitrication
in activated sludge systems, Chemical Engineering Science, vol. 63, no. 6,
pp. 14491459, 2008.
[101] M. I. Nelson, E. Balakrishnan, H. Sidhu, and X. Chen, A fundamental analysis of continuous ow bioreactor models and membrane reactor models to process industrial wastewaters, Chemical Engineering Journal, vol. 140, no. 1,
pp. 521528, 2008.
[102] A. Rapaport, J. Harmand, and F. Mazenc, Coexistence in the design of
a series of two chemostats, Nonlinear Analysis: Real World Applications,
vol. 9, no. 3, pp. 10521067, 2008.
[103] C.K.Essajee and R.D.Tanner, The eect of extracelluluar variables on the
stability of the continuous bakers yeast-ethanol fermentation process, Process
Biochemistry, vol. 14, pp. 1625, 1979.
[104] L. E. Erickson and L.-t. Fan, Optimization of the hydraulic regime of activated sludge systems, Journal of Water Pollution Control Federation, vol. 40,
pp. 345362, 1968.
[105] L. E. Erickson, Y. Ho, and L. Fan, Modeling and optimization of step aeration
waste treatment systems, Journal of Water Pollution Control Federation,
vol. 40, pp. 717732, 1968.
[106] C. D. de Gooijer, W. A. Bakker, H. H. Beeftink, and J. Tramper, Bioreactors
in series: an overview of design procedures and practical applications, Enzyme
and microbial technology, vol. 18, no. 3, pp. 202219, 1996.

Appendix

301

[107] M. Nelson and H. Sidhu, Evaluating the performance of a cascade of two

bioreactors, Chemical Engineering Science, vol. 61, no. 10, pp. 31593166,
2006.
[108] A. Balakrishnan and R. Yang, Improvement of chemostat performance via
nonlinear oscillations, part 2. extension to other systems, ACH, models in
chemistry, vol. 135, no. 1-2, pp. 118, 1998.
[109] C.-C. Chen, C. Hwang, and R. Y. Yang, Performance enhancement and optimization of chemostat cascades, Chemical engineering science, vol. 50, no. 3,
pp. 485494, 1995.
[110] R. Yang and J. Su, Improvement of chemostat performance via nonlinear
oscillations. part 1: Operation strategy, Bioprocess Engineering, vol. 9, no. 23, pp. 97102, 1993.
[111] N. V. Mantzaris, P. Daoutidis, F. Srienc, and A. G. Fredrickson, Growth
processes in a cascade of bioreactors: Mathematical models, AIChE journal,
vol. 45, no. 1, pp. 164176, 1999.
[112] J. Monod, La technique de culture continue, thorie et applications, Ann Inst
Pasteur, vol. 79, p. 390410, 1950.
[113] D. Herbert, R. Elsworth, and R. Telling, The continuous culture of bacteria; a
theoretical and experimental study, Journal of General Microbiology, vol. 14,
no. 3, pp. 601622, 1956.
[114] A. Novick and L. Szilard, Description of the chemostat, Science, vol. 112,
pp. 715716, 1950.

Appendix

302

[115] C. Spicer, The theory of bacterial constant growth apparatus, Biometrics,

vol. 11, no. 2, pp. 225230, 1955.
[116] I. Ramirez, A. Mottet, H. Carr`ere, S. Deleris, F. Vedrenne, and J.-P. Steyer,
Modied adm1 disintegration/hydrolysis structures for modeling batch thermophilic anaerobic digestion of thermally pretreated waste activated sludge,
Water research, vol. 43, no. 14, pp. 34793492, 2009.
[117] M. Myint and N. Nirmalakhandan, Evaluation of rst-order, second-order,
and surface-limiting reactions in anaerobic hydrolysis of cattle manure, Environmental engineering science, vol. 23, no. 6, pp. 970980, 2006.
[118] V. Vavilin, S. Rytov, and L. Y. Lokshina, A description of hydrolysis kinetics in anaerobic degradation of particulate organic matter, Bioresource
technology, vol. 56, no. 2, pp. 229237, 1996.
[119] M. Myint, N. Nirmalakhandan, and R. Speece, Anaerobic fermentation of
cattle manure: modeling of hydrolysis and acidogenesis, Water Research,
vol. 41, no. 2, pp. 323332, 2007.
[120] S. M. Ruel, Y. Comeau, P. Ginestet, and A. Heduit, Modeling acidogenic
and sulfate-reducing processes for the determination of fermentable fractions
in wastewater, Biotechnology and Bioengineering, vol. 80, no. 5, pp. 525536,
2002.
[121] Y. Fujimoto, Kinetics of microbial growth and substrate consumption, Journal of theoretical biology, vol. 5, no. 2, pp. 171191, 1963.

Appendix

303

[122] W. G. Characklis, Microbial reaction rate expressions, Journal of the Environmental Engineering Division, vol. 104, no. 3, pp. 531534, 1978.
[123] C. Lobry and J. Harmand, A new hypothesis to explain the coexistence of
n species in the presence of a single resource, Comptes Rendus Biologies,
vol. 329, no. 1, pp. 4046, 2006.
[124] R. Arditi and L. R. Ginzburg, Coupling in predator-prey dynamics: ratiodependence, Journal of Theoretical Biology, vol. 139, no. 3, pp. 311326,
1989.
[125] O. Bail, Ergebnisse experimenteller populationsforschung, Immun. Forsch,
vol. 60, no. 1, 1929.
[126] W. Greenleaf, The inuence of volume of culture medium and cell proximity
on the rate of reproduction of infusona, Journal of Experimental Zoology,
vol. 46, p. 143, 1926.
[127] R. Pearl and S. L. Parker, On the inuence of density of population upon
the rate of reproduction in drosophila, Proceedings of the National Academy
of Sciences of the United States of America, vol. 8, no. 7, p. 212, 1922.
[128] M. I. Nelson and A. Holder, A fundamental analysis of continuous ow bioreactor models governed by contois kinetics. ii. reactor cascades, Chemical Engineering Journal, vol. 149, no. 1, pp. 406416, 2009.
[129] M. Nelson, X. Chen, and H. Sidhu, Reducing the emission of pollutants in
industrial wastewater through the use of membrane bioreactors, in Aspects

Appendix

304

of Mathematical Modelling (R. Hosking and E. Venturino, eds.), Mathematics

and Biosciences in Interaction, pp. 95107, 2008.
[130] R. T. Alqahtani, M. I. Nelson, and A. L. Worthy, Analysis of a chemostat
model with variable yield coecient: Contois kinetics, ANZIAM Journal,
vol. 53, pp. C155C171, 2012.
[131] A. Ajbar, Study of complex dynamics in pure and simple microbial competition, Chemical Engineering Science, vol. 80, pp. 188 194, 2012.
[132] M. Nelson, T. Kerr, and X. Chen, A fundamental analysis of continuous
ow bioreactor and membrane reactor models with death and maintenance
included, Asia-Pacic Journal of Chemical Engineering, vol. 3, p. 7080, 2008.
[133] S. Sotemann, P. Van Rensburg, N. Ristow, M. Wentzel, R. Loewenthal, and
G. Ekama, Integrated chemical, physical and biological processes modelling
of anaerobic digestion of sewage sludge, Water Science & Technology, vol. 54,
no. 5, pp. 109117, 2006.
[134] A. Nopharatana, P. C. Pullammanappallil, and W. P. Clarke, Kinetics and
dynamic modelling of batch anaerobic digestion of municipal solid waste in a
stirred reactor, Waste management, vol. 27, no. 5, pp. 595603, 2007.
[135] V. Vavilin, B. Fernandez, J. Palatsi, and X. Flotats, Hydrolysis kinetics in
anaerobic degradation of particulate organic material: an overview, Waste
management, vol. 28, no. 6, pp. 939951, 2008.
[136] J. Beltran-Heredia, J. Torregrosa, J. R. Dominguez, and J. Garcia, Ozonation of black-table-olive industrial wastewaters: eect of an aerobic biological

Appendix

305

pretreatment, Journal of Chemical Technology and Biotechnology, vol. 75,

no. 7, pp. 561568, 2000.
[137] S. Bhattacharya and P. Q. Khai, Kinetics of anaerobic cowdung digestion,
Energy, vol. 12, no. 6, pp. 497500, 1987.
[138] L. Krzystek, S. Ledakowicz, H.-J. Kahle, and K. Kaczorek, Degradation
of household biowaste in reactors, Journal of biotechnology, vol. 92, no. 2,
pp. 103112, 2001.
[139] S. Moosa, M. Nemati, and S. TL Harrison, A kinetic study on anaerobic
reduction of sulphate, part i: Eect of sulphate concentration, Chemical
Engineering Science, vol. 57, no. 14, pp. 27732780, 2002.
[140] S. Sun, B. Wu, D. Zhao, X. Zhang, Y. Zhang, W. Li, and S. Cheng, Optimization of xhhh strain biodegradation with metal ions for pharmaceautical
wastewater treatment, Journal Of Environmental Biology / Academy Of Environmental Biology, vol. 30, pp. 877 882, 2009.
[141] N. Abdurahman, Y. Rosli, N. Azhari, and S. Tam, Biomethanation of palm
oil mill euent (pome) by membrane anaerobic system (mas) using pome as a
substrate, World Academy of Science, Engineering and Technology, vol. 75,
pp. 419424, 2011.
[142] N. Abdurahman, Y. Rosli, and N. Azhari, Development of a membrane anaerobic system (mas) for palm oil mill euent (pome) treatment, Desalination,
vol. 266, no. 1, pp. 208212, 2011.

Appendix

306

[143] P. Poh and M. Chong, Biomethanation of palm oil mill euent (pome) with
a thermophilic mixed culture cultivated using pome as a substrate, Chemical
Engineering Journal, vol. 164, no. 1, pp. 146154, 2010.
[144] A. Pinna, A. Lallai, G. Mura, and M. Grosso, Comparison across dierent
models for the description of batch biodegradation processes, Chemical Engineering Transactions, pp. 12271232, 2009.
[145] J. Zhang, X. Shao, O. V. Townsend, and L. R. Lynd, Simultaneous saccharication and co-fermentation of paper sludge to ethanol by saccharomyces
cerevisiae rwb222part i: Kinetic modeling and parameters, Biotechnology
and bioengineering, vol. 104, no. 5, pp. 920931, 2009.
[146] P. S. Hemsi, C. D. Shackelford, and L. A. Figueroa, Calibration of reactive transport models for remediation of mine drainage in solid-substrate biocolumns, Journal of Environmental Engineering, vol. 136, no. 9, pp. 914925,
2010.
[147] F. Emerald, M. Eljeeva, D. S. Prasad, M. R. Ravindra, and H. A. Pushpadass, Performance and biomass kinetics of activated sludge system treating
dairy wastewater, International Journal of Dairy Technology, vol. 65, no. 4,
pp. 609615, 2012.
[148] K. Karim, K. T. Klasson, S. R. Drescher, W. Ridenour, A. P. Borole, and
M. H. Al-Dahhan, Mesophilic digestion kinetics of manure slurry, Applied
biochemistry and biotechnology, vol. 142, no. 3, pp. 231242, 2007.

Appendix

307

[149] Y. R. Chen and A. G. Hashimoto, Kinetics of methane fermentation,

Biotechnology and Bioengineering Symposium, vol. 8, pp. 269282, 1978.
[150] D. Hill, Steady state mesophilic design equations for methane production
from livestock wastes, Transactions of American Society of Agricultural Engineers, vol. 34, pp. 21572163, 1991.
[151] B. Fernndez, P. Porrier, and R. Chamy, Eect of inoculum-substrate ratio on
the start-up of solid waste anaerobic digesters, Water Science & Technology,
vol. 44, no. 4, pp. 103108, 2001.
[152] N. A. Gawande, D. R. Reinhart, and G.-T. Yeh, Modeling microbiological and
chemical processes in municipal solid waste bioreactor, part ii: Application of
numerical model biokemod-3p, Waste management, vol. 30, no. 2, pp. 211
218, 2010.
[153] N. A. Gawande, D. R. Reinhart, and G.-T. Yeh, Modeling microbiological and
chemical processes in municipal solid waste bioreactor, part i: Development
of a three-phase numerical model biokemod-3p, Waste management, vol. 30,
no. 2, pp. 202210, 2010.
[154] A. Ghaly, S. Sadaka, and A. Hazzaa, Kinetics of an intermittent-ow,
continuous-mix anaerobic reactor, Energy sources, vol. 22, no. 6, pp. 525
542, 2000.
[155] T. Hidaka, T. Horie, S. Akao, and H. Tsuno, Kinetic model of thermophilic
l-lactate fermentation by bacillus coagulans combined with real-time PCR
quantication, Water research, vol. 44, no. 8, pp. 25542562, 2010.

Appendix

308

[156] H. Yasui, R. Goel, Y. Li, and T. Noike, Modied adm1 structure for modelling
municipal primary sludge hydrolysis, Water research, vol. 42, no. 1, pp. 249
259, 2008.
[157] C. T. Goudar and K. A. Strevett, Estimating growth kinetics of penicillium
chrysogenum by nonlinear regression, Biochemical engineering journal, vol. 1,
no. 3, pp. 191199, 1998.
[158] H. Zhou, W. Lu, J. Wen, and L. Ma, Kinetic analysis of 11a-hydroxylation of
steroids by rhizopus nigricans, Journal of Molecular Catalysis B: Enzymatic,
vol. 56, pp. 136 141, 2009.
[159] M. A. Mazutti, M. L. Corazza, F. Maugeri Filho, M. I. Rodrigues, F. C.
Corazza, and H. Treichel, Inulinase production in a batch bioreactor using
agroindustrial residues as the substrate: experimental data and modeling,
Bioprocess and biosystems engineering, vol. 32, no. 1, pp. 8595, 2009.
[160] S. Hernalsteens and F. Maugert, Synthesis of fructooligosaccharides using
extracellular enzymes from rhodotorula sp., Journal of Food Biochemistry,
vol. 34, no. 3, pp. 520 534, 2010.
[161] V. Ilichev, Active and passive states in ecological models (stabilization, adaptation, and survivability), Journal of Computer and Systems Sciences International, vol. 47, no. 5, pp. 830839, 2008.
[162] N. Bongochgetsakul and T. Ishida, A new analytical approach to optimizing
the design of large-scale composting systems, Bioresource technology, vol. 99,
no. 6, pp. 16301641, 2008.

Appendix

309

[163] A. Nakayama, K. Nakasaki, F. Kuwahara, and Y. Sano, A lumped parameter

heat transfer analysis for composting processes with aeration, TransactionsAmerican Society Of Mechanical Engineers Journal Of Heat Transfer, vol. 129,
no. 7, pp. 902906, 2007.
[164] P. A. Abrams and L. R. Ginzburg, The nature of predation: prey dependent,
ratio dependent or neither?, Trends in Ecology & Evolution, vol. 15, no. 8,
pp. 337341, 2000.
[165] F. Cabrera and I. Maria, Deterministic approach to the study of the interaction predatorprey in a chemostat with predator mutual interference. implications for the paradox of enrichment, Ecological Modelling, vol. 222, no. 3,
pp. 598605, 2011.
[166] J. Harmand and J. Godon, Density-dependent kinetics models for a simple
description of complex phenomena in macroscopic mass-balance modeling of
bioreactors, Ecological modelling, vol. 200, no. 3, pp. 393402, 2007.
[167] L. Guerrero, S. Montalvo, E. Coronado, R. Chamy, P. Poirrier, D. Crutchik,
E. Sanchez, M. De La Rubia, and R. Borja, Performance evaluation of a twophase anaerobic digestion process of synthetic domestic wastewater at ambient
temperature, Journal of Environmental Science and Health Part A, vol. 44,
no. 7, pp. 673681, 2009.
[168] F. Ardestani, Survey of the nutrient utilization and cell growth kinetic with
verhulst, contois and exponential models for penicillium brevicompactum atcc

Appendix

310

16024 in batch bioreactor, World Applied Sciences Journal, vol. 16, no. 1,
pp. 135140, 2012.
[169] S. Pavlostathis and E. Giraldo-Gomez, Kinetics of anaerobic treatment,
Water Science and Technology, vol. 24, no. 8, pp. 3559, 1991.
[170] G. J. Butler, S. B. Hsu, and P. Waltman, Coexistence of competing predators
in a chemostat, Journal of Mathematical Biology, vol. 17, no. 2, pp. 133151,
1983.
[171] L. Edelstein-Keshet, Mathematical models in biology, vol. 46. Siam, 1988.
[172] H. Dulac, Points singuliers des equations dierentielles. Gauthier-Villars,
1934.
[173] S. Lynch, Dynamical Systems with Applications using Maple. Birkhuser, 2nd
edition ed., 2009.
[174] R. T. Alqahtani, M. I. Nelson, and A. L. Worthy, A fundamental analysis of
continuous ow bioreactor models with recycle around each reactor governed
by contois kinetics. iii. two and three reactor cascades, Chemical Engineering
Journal, vol. 183, pp. 422432, 2012.
[175] R. T. Alqahtani, M. I. Nelson, and A. L. Worthy, A mathematical model
for the biological treatment of industrial wastewaters in a cascade of four
reactors, in Modelling and Simulation Society of Australia and New Zealand,
MODSIM 2011 (F. Chan, D. Marinova, and R. Anderssen, eds.), pp. 256262,
2011.

Appendix

311

[176] R. T. Alqahtani, M. I. Nelson, and A. L. Worthy, A mathematical analysis

of continuous ow bioreactor models governed by contois kinetics: A two
reactor cascade, in In Proceedings of the Australasian Chemical Engineering
Conference, CHEMECA 2011, pp. 111, 2011.
[177] C. Grady and H. Lim, Biological Wastewater Treatment, ch. 12. In [80], 1999.
[178] C. Grady and H. Lim, Biological Wastewater Treatment, ch. 13. In [80], 1999.
[179] R. T. Alqahtani, M. I. Nelson, and A. L. Worthy, A fundamental analysis
of continuous ow bioreactor models governed by contois kinetics. iv. recycle
around the whole reactor cascade, Chemical Engineering Journal, vol. 218,
pp. 99 107, 2013.
[180] R. M. May, Stability and complexity in model ecosystems. Princeton University
Press, 1973.
[181] A. Dorofeev, M. Glagolev, T. Bondarenko, and N. Panikov, Observation
and explanation of the unusual growth-kinetics of arthrobacter-globiformis,
Microbiology, vol. 61, no. 1, pp. 2431, 1992.
[182] I. H. LEE, A. Fredrickson, and H. Tsuchiya, Damped oscillations in continuous culture of lactobacillus plantarum, Journal of General Microbiology,
vol. 93, no. 2, pp. 204208, 1976.
[183] B. Ermentrout, Simulating, analyzing, and animating dynamical systems: a
guide to XPPAUT for researchers and students, vol. 14. Siam, 2002.

Appendix

312

[184] B. Gray and M. Roberts, A method for the complete qualitative analysis of
two coupled ordinary dierential equations dependent on three parameters,
Proceedings of the Royal Society of London. A. Mathematical and Physical
Sciences, vol. 416, no. 1851, pp. 361389, 1988.
[185] O. Nowak, Optimizing the use of sludge treatment facilities at municipal
wwtps, Journal of Environmental Science and Health Part A, vol. 41, no. 9,
pp. 18071817, 2006.
[186] P. Barker and P. Dold, Sludge production and oxygen demand in nutrient
removal activated sludge systems, Water Science and Technology, vol. 34,
no. 5, pp. 4350, 1996.
[187] M. Barjenbruch and O. Kopplow, Enzymatic, mechanical and thermal pretreatment of surplus sludge, Advances in environmental research, vol. 7, no. 3,
pp. 715720, 2003.
[188] C. Chu, W. Feng, B.-V. Chang, C. Chou, and D. Lee, Reduction of microbial
density level in wastewater activated sludge via freezing and thawing, Water
Research, vol. 33, no. 16, pp. 35323535, 1999.
[189] C.-H. Xing, K. Yamamoto, and K. Fukushi, Performance of an inclined-plate
membrane bioreactor at zero excess sludge discharge, Journal of membrane
science, vol. 275, no. 1, pp. 175186, 2006.
[190] H. Yasui, K. Nakamura, S. Sakuma, M. Iwasaki, and Y. Sakai, A full-scale operation of a novel activated sludge process without excess sludge production,
Water Science and Technology, vol. 34, no. 3, pp. 395404, 1996.

Appendix

313

[191] Y. Sakai, T. Fukase, H. Yasui, and M. Shibata, An activated sludge process

without excess sludge production, Water Science and Technology, vol. 36,
no. 11, pp. 163170, 1997.
[192] K.-G. Song, Y.-K. Choung, K.-H. Ahn, J. Cho, and H. Yun, Performance of
membrane bioreactor system with sludge ozonation process for minimization
of excess sludge production, Desalination, vol. 157, no. 1, pp. 353359, 2003.
[193] Z. Wang, L. Wang, B. Z. Wang, Y. F. Jiang, and S. Liu, Bench-scale study on
zero excess activated sludge production process coupled with ozonation unit
in membrane bioreactor, Journal of Environmental Science and Health Part
A, vol. 43, no. 11, pp. 13251332, 2008.
[194] Y.-K. Oh, K.-R. Lee, K.-B. Ko, and I.-T. Yeom, Eects of chemical sludge disintegration on the performances of wastewater treatment by membrane bioreactor, Water research, vol. 41, no. 12, pp. 26652671, 2007.
[195] U. Kepp, I. Machenbach, N. Weisz, and O. Solheim, Enhanced stabilisation
of sewage sludge through thermalhydrolysis-three years of experience with full
scale plant, Water Science & Technology, vol. 42, no. 9, pp. 8996, 2000.
[196] S.-H. Yoon, H.-S. Kim, and S. Lee, Incorporation of ultrasonic cell disintegration into a membrane bioreactor for zero sludge production, Process
Biochemistry, vol. 39, no. 12, pp. 19231929, 2004.
[197] Y. Wei, R. T. Van Houten, A. R. Borger, D. H. Eikelboom, and Y. Fan, Minimization of excess sludge production for biological wastewater treatment,
Water Research, vol. 37, no. 18, pp. 44534467, 2003.

Appendix

314

[198] B. Zhang and K. Yamamoto, Seasonal change of microbial population and

activities in a building wastewater reuse system using a membrane separation activated sludge process, Water Science and Technology, vol. 34, no. 5,
pp. 295302, 1996.
[199] E. Chian and F. DeWalle, Treatment of high strength acidic wastewater with
a completely mixed anaerobic lter, Water Research, vol. 11, no. 3, pp. 295
304, 1977.
[200] M. Nelson and E. Balakrishnan, An analysis of an activated sludge process containing a sludge disintegration system, In Proceedings of the Australasian Chemical Engineering Conference, Chemeca, Engineers Australia,
vol. 37, no. 8, pp. 1921 1931, 2011.