You are on page 1of 24

Rev. 2.

ON THE DISTINCTION BETWEEN LAG AND


DELAY IN POPULATION GROWTH

by

Peter Vadasz 
and Alisa S. Vadasz

College of Engineering, Forestry and Natural Sciences, Northern Arizona University


P.O.Box 15600, Flagstaff AZ 86011-5600, U.S.A.
(also of the Faculty of Engineering, University of KZ Natal, Durban, South Africa)

(Submitted August 6, 2009)


Corresponding Author: Phone: (+1-928) 523-5843 Fax: (+1-928) 523-8951,
e-mail: peter.vadasz@nau.edu
ON THE DISTINCTION BETWEEN LAG AND
DELAY IN POPULATION GROWTH

by

Peter Vadasz 
and Alisa S. Vadasz

College of Engineering, Forestry and Natural Sciences, Northern Arizona University


P.O.Box 15600, Flagstaff AZ 86011-5600, U.S.A.
(also of the Faculty of Engineering, University of KZ Natal, Durban, South Africa)

(Submitted August 6, 2009)

ABSTRACT
The analysis and results presented in this paper provide conclusive evidence to distinguish
between the delay effect and the LAG as two biologically distinct phenomena. It therefore
dispels the incorrect notion that delay effects represented by delay differential equations are
the biological reason behind the LAG phase in microorganism growth. The resulting
consequence so far is that the only other reason for the LAG phase is the existence of unstable
stationary states. The latter are a result of accounting for the microbial metabolic mass transfer
in the population growth process.

Keywords: Population Dynamics, Microorganism Growth, LAG, Delay.


Corresponding Author: Phone: (+1-928) 523-5843 Fax: (+1-928) 523-8951,
e-mail: peter.vadasz@nau.edu
2
1 . I nt r o du c ti o n
Delay models were suggested for population growth models in ecology to explain among others
the LAG phenomenon in microbial growth. The most common delay mechanism was introduced in
population growth models in ecology by Hutchinson (1948) and Wangersky and Cunningham
(1957) and was extensively reviewed (see May 1973, 1978) as a delay differential equation
amending the Logistic growth model (Pearl 1927, Verhulst 1838) in the form
⎛ x (t − τ ) ⎞
x ( t ) = µmax ⎜ 1 − ⎟⎠ x ( t ) (1)
⎝ δ

where, x is the population number (or cell concentration), t is time, µmax is the maximum specific

growth rate, δ is the carrying capacity of the environment, and τ is the delay time. May (1973,
1978, 1981) indicated that “… more realistically, the regulatory term is likely to depend not on the
population at a time exactly τ earlier, but rather on some smooth average over past populations”
leading to an integro-differential equation of the form
⎛ t ⎞
x ( t ) = µmax ⎜ 1 − ∫ Q ( t − t ′ ) x ( t ′ ) dt ′ ⎟ x ( t ) (2)
⎝ −∞ ⎠
A correction to this common delay model (1) introduced a more realistic delay effect as presented by
May (1981), by accounting for a delay in “recruitment into the adult population” the latter being
proportional to the population levels at a time τ earlier, while the mortality depends simply on the
population at time t , as follows
⎛ x 2 (t ) ⎞
x ( t ) = µmax ⎜ x ( t − τ ) −        ∀ t ≥ 0 (3)
⎝ δ ⎟⎠
subject to a history function as the initial conditions in the form

()
x t =ϕ t () ∀ t ∈ ⎡⎣ −τ , 0 ⎤⎦ (4)

Actually, May (1981) presented in his equation (2-5) a slightly different form of equation (3) where
the inhibition term ( second term in the brackets on the right hand side of equation (3) ) was linear

( )
and the delay term was represented implicitly via a function R ⎡⎣ x t − τ ⎤⎦ that was not explicitly

defined but rather allowed to take specific forms in different applications. Equation (3) uses May
(1981) principle of introducing the delay (i.e. x ( t − τ ) ) in the first term that promotes growth
(recruitment term) and accounting for the inhibition mechanism (mortality) via a quadratic term
3
x 2 δ . This kind of delay mechanism can be derived systematically from more detailed age-
structured models based on McKendrick-von Foerster model (McKendrick, 1926, von Foerster,
1959) such as Lebowitz and Rubinow (1974), Vance et al. (1988) and many others. An additional
delay model uses the Gompertz (1825) model rather than the Logistic as its kernel in the form
x ( t )
= α − β ln ⎡⎣ x ( t − τ ) ⎤⎦ (5)
x (t )
There are additional models that include delays such as models that account for microbial
metabolism explicitly (Wood et al. 1994, 1995). In such models the delays are introduced in the
metabolic process components and not explicitly in the population size. As a result the dynamics of
population growth is affected by delayed availability and utilization of resources, which can
introduce lag effects even without explicitly introducing delays into the model. Examples of the
latter are the non-delay models presented by Baranyi and Roberts (1994), and Vadasz and Vadasz
(2002a,b, 2005, 2007). It was demonstrated (Vadasz and Vadasz 2007) that a LAG appears in
Baranyi and Roberts (1994) model as well as in the Vadasz and Vadasz (2002a,b) neoclassical
model due to one and the same reason, i.e. the existence of unstable stationary points. The latter will
be discussed in more detail towards the end of the present paper.
Delay effects of the type presented in the model represented by equations (3) and (4) (and
possibly (1) or (2)) are being considered to be the biological reason behind the LAG phase in
microorganism growth. The objective of the present paper is to investigate the latter and present
results that either confirm or deny the notion that LAG and delay effects are essentially the same
phenomenon. In order to undertake this kind of analysis we present first the factual description of
microbial growth and the conditions for LAG appearance, which are experimentally undisputed
facts (Maier 2000, Baranyi and Roberts 1994, Pirt 1975, Lodge and Hinshelwood 1943).
A qualitative description of a typical microbial growth curve is presented in Figure 1 where five
distinct phases are present (and possibly 6 if a decay/decline phase is also present). The classical
way of describing a typical microbial growth curve is via four phases (and possibly 5 if a
decay/decline phase is also present), i.e. the LAG phase, which may or may not be present, an
exponential growth phase, and inhibition phase and a stationary phase. We decided to introduce a
further subdivision of the exponential growth phase as it becomes convenient for the purpose of
identifying different stages of growth that result in our study. As such we subdivided the exponential
phase into a Logarithmic Exponential phase (LogEx) and a Regular Exponential growth phase
4
(REx). We will define each one of these phases in what follows. The first is a LAG phase identified
by an approximately constant value of population number at the initial growth stage on a logarithm
of the cell concentration versus time graph ( ln x vs. t ). Therefore by its very definition the LAG
phase is characterized by the condition
d ln x(t)
≅0 ∀ 0≤t≤λ (6)
dt

where λ ≡ λcl or λ ≡ λrd is the LAG duration that will be defined and discussed later. The LAG
phase is followed by a Logarithmic Exponential Phase (LogEx) identified by an exponential growth
on a logarithmic versus time graph ( ln x vs. t ). We introduce the terminology of Logarithmic
Exponential Phase (LogEx) in order to distinguish it from the Regular Exponential growth phase
(REx) that follows, the latter being identified by an approximately straight line on the ln x vs. t
graph. The Inhibition (Inhib) phase is next which expresses itself as an exponential consolidation
towards the stationary constant value on the ln x vs. t graph. The stationary (Stat) phase is usually
the last phase (if decay/decline is absent) and can be identified by an approximately constant value
of the population number at the final growth stage. The growth curve, which starts with an almost
constant value during the LAG phase becomes concave-up during the LogEx and beginning of the
REx phases, and converts to a concave-down curve at the end of the REx phase and during the
Inhibition and Stationary phases. The point where the curve shifts from concave-up to concave-
down is the Logarithmic Inflection Point (LIP) which is substantially distinct from the Regular
Inflection Point (RIP) which appears on typical sigmoid curves produced by the Logistic model for
example on an x vs. t graph. The salient features of microbial growth that include the LAG, LogEx
and the beginning of the REx phase (prior to the LIP point) can only be observed on ln x vs. t
graph. The Logistic model for example cannot reproduce a LIP but it may reproduce a RIP,
depending on the initial conditions.
The classical definition of the LAG duration λcl is due to Pirt (1975) as the point where the line

parallel to the t axis at the value of the initial condition yo intersects the tangent line to the growth
curve at the Logarithmic Inflection Point (LIP) (the black dot on Figure 1). This classical definition
has the tendency to overestimate the LAG duration as observed in Figure 1 where λcl is clearly
positioned within the LogEx phase, while in reality it should be the border point between the LAG
and LogEx phases. A redefined LAG duration introduced by Vadasz and Vadasz (2005, 2007) is

5
presented in Figure 1, although the latter is linked with different models such as Baranyi and
Roberts (1994) and Vadasz and Vadasz (2002a,b). A definition independent of the models although
desired and useful was not yet derived.
Qualitative Growth Curve
y Inhib
Stat

RIP Stationary
Phase
Inhibition
Phase
y = ln (x)

LIP t stat

Logarithmic Regular
Exponential Exponential
Phase Phase
(Logarithmic
LAG phase REx Linear Growth)

yo
LAG LogEx

λ rd λ cl time , t t

Figure 1: A typical microbial growth curve.

While in presenting the phases in Figure 1 it is more consistent to use the redefined LAG duration
based on Vadasz and Vadasz (2005, 2007) it becomes easier to identify the existence of a LAG
phase geometrically by using the classical definition of the LAG or by observing the existence of a
concave-up part of the curve during the LogEx phase. Clearly if LIP does not exist, implying a
concave-down curve from t = 0 onwards it is an indication for lack of the LAG phase. Indeed, the
LAG phase is not obtained in all cases of microbial growth. As a matter of fact there are quite clear
conditions under which a LAG phase may be anticipated and these conditions depend on the pre-
inoculation conditions of the inoculum. A typical experiment (Pirt 1975, Lodge and Hinshelwood
1943, Maier 2000) that aims at testing the conditions under which a LAG phase will be present
consists of sampling an inoculum from a prior microbial growth medium and re-inoculating it in a
new medium as presented in Figure 2. In this type of experiment a sample is taken from a pre-
inoculation medium at different phases of growth. Based on experimental evidence (Maier 2000,
Baranyi and Roberts 1994, Pirt 1975, Lodge and Hinshelwood 1943) it turns out that subject to the
inoculum size being smaller than a threshold value (Meier 2000, indicates this threshold value to be
6
about 106 cells/ml) and the re-inoculation medium being not substantially different than the pre-
inoculation growth medium then a LAG phase will be present under the following conditions. If the
sample was taken from the stationary phase or a prior LAG phase in the pre-inoculation medium a
LAG will be present following re-inoculation, as presented in Figure 2(a) and 2(c). However, a LAG
phase will never appear if the sample was taken from the exponential growth phase (LogEx or REx)
of the pre-inoculation medium as presented in Figure 2(b). In the latter case exponential post-
inoculation growth is observed. We will attempt to test precisely these conditions of re-inoculation
growth by using delay models and test whether such delay effects may cause a LAG phase as
observed experimentally. Since any inoculum is a result of previous growth in another medium
testing the cases specified above (inoculum taken from each growth phase in the pre-inoculation
medium) covers all possibilities. A LAG phase should not appear if the inoculum is taken from a
pre-inoculation exponential growth phase and it should appear only if it is taken from a pre-
inoculation LAG or stationary phase (subject also to a sufficiently small size of the inoculum below
the threshold value).

ln x ln x
Pre-Inoculation Growth Post-Inoculation Growth ln x ln x ln x ln x
Pre-Inoculation
Pre-Inoculation Growth Post-Inoculation Growth Post-Inoculation Growth
A
[cells/ml] Growth
Re-Inoculation

8 Inoculum from [cells/ml]


10
Stat Phase

Re-Inoculation
8 8
10 10
LAG
(on a Log scale)
(on a Log scale)
(on a Log scale)

LAG
10
7
B 7 Inoculum from
10 NO LAG 10
7

B LAG Phase
Inoculum from REx Phase
Re-Inoculation

6
10 6
10 106 C
x
x

C
LAG
5
10 5
10 105
t 0 2 4 6 8 10 12 14 16
t
0 4 8 12 16 20 24 28 0 2 4 6 8 10 12 14 16
time, t [hours] time, t [hours]
time, t [hours]

(a) (b) (c)

Figure 2: A typical re-inoculation experiment for identifying the conditions for a LAG phase presence.

2 . T he o r e t i c a l a na l y s i s a n d c o ndi t i o ns fo r L A G a pp e a r a nc e
This section is focused on deriving conditions for LAG appearance theoretically based on the delay
models that were introduced in the previous section. By using condition (6) that is by definition a
necessary characteristic of the LAG phase , i.e. d ln x(t) d t ≅ 0 ∀ 0 ≤ t ≤ λ , that can be presented

in the alternative form x x ≅ 0 ∀ 0 ≤ t ≤ λ (because d ln x(t) d t = x x ), one may test the delay

7
models and develop the conditions that are to be imposed on the history function ϕ t in the pre- ()
inoculation medium, i.e. for times within the range −τ ≤ t ≤ 0 , in order to allow a LAG phase to be
present in the post-inoculation medium, i.e. for t ≥ 0 . A necessary condition based on (6) would be
that
⎛ d ln x ⎞ ⎛ x ⎞
⎜⎝ d t ⎟⎠ =⎜ ⎟ ≅0 (7)
t =0
⎝ x ⎠ t =0

This condition implies that x x ( ) t =0


is either exactly zero or is approximately zero. Let us check

each one of these possibilities. What the exact equality implies in terms of its impact on the possible

()
types of history functions ϕ t that will be then compatible with the existence of a LAG in the post-

inoculation medium is assessed first. All delay models introduced in the previous section are subject
to initial conditions in terms of this history function in the form presented in equation (4), i.e.

() ()
x t = ϕ t , ∀ t ∈ ⎡⎣ −τ , 0 ⎤⎦ . The Hutchinson delay model represented by equation (1), which is valid

for times t ≥ 0 , can be therefore used to evaluate the term x x at t = 0 leading to

⎡ x ( t ) ⎤ ⎛ x (t − τ ) ⎞ ⎛ x ( −τ ) ⎞
⎢ ⎥ = µmax ⎜⎝ 1 − ⎟⎠ = µmax ⎜⎝ 1 − δ ⎟⎠ (8)
⎣ x ( t ) ⎦t = 0 δ t =0

Applying now equation (7) in its exact sense, i.e. x x ( ) t =0


= 0 exactly implies by using (8) that

⎛ x ( −τ ) ⎞
⎜⎝ 1 − δ ⎟⎠ = 0    →    x ( −τ ) = ϕ ( −τ ) = δ (9)

Therefore, the conclusion is that a necessary condition for obtaining a LAG phase in the Hutchinson
delay model (1) is that a constant history function be equal to the carrying capacity of the post-
inoculation medium, δ . This removes all dynamics from the solution resulting in a constant solution

() )
x t = δ ∀ t ∈ ⎡⎣ −τ ,∞ . All other history functions will not create a LAG if the equality in equation

(7) is imposed accurately. Performing the same process for May’s (1981) delay model (3) for any

() ()
constant history function x t = ϕ t = xo , ∀ t ∈ ⎡⎣ −τ , 0 ⎤⎦ that is applicable to stationary (or LAG)

pre-inoculation conditions leads to


xo2
x ( −τ ) = ϕ ( −τ ) = = xo      →   xo = δ (10)
δ

8
implying again that a necessary condition for obtaining a LAG phase in the May’s (1981) delay
model (3) is that a constant history function be equal to the carrying capacity of the post-inoculation
medium, δ . Repeating the same procedure for the Gompertz delay model (5) yields
x ( −τ ) = ϕ ( −τ ) = eα β (11)
which is the stationary post inoculation solution of the Gompertz delay model (5), i.e. the post-
inoculation carrying capacity. Last but not least let us derive the corresponding condition for the
integro-differential delay system (2). The result leads to the following normalization constraint on
the delay integral
t

∫ Q ( t − t ′ ) x ( t ′ ) dt ′ = 1
−∞
(12)

More explicit conclusions from (12) seem difficult to derive, however this normalization condition
(12) seems also extremely restricting the dynamics of the post-inoculation growth. In conclusion,
one may indicate that applying the equality condition (7) exactly leads to impossible conditions for
post-inoculation growth that includes a LAG phase.
Therefore, we may try to allow the condition for LAG appearance to apply approximately
only, i.e.

( x x ) t =0
( )
≈ 0 or x x t =0
<< 1 (13)

( )
The Hutchinson as well as May’s (1981) delay models (1) and (3) imply that x −τ ≈ δ which again

is unrealistic for post-inoculation growth, but taking the Malthus (1798) form of May’s (1981) delay
model, a correct representation of the initial growth stages when the LAG is expected to appear, i.e.
x ( t ) = µmax x ( t − τ )    ∀ t ≥ 0 (14)

produces the condition for LAG appearance in the form


x ( −τ )
<< 1     →     ϕ ( −τ ) = x ( −τ ) << xo (15)
xo

( )
where xo = x t = 0 . In deriving (15) we did not assume yet as previously that the history function

()
ϕ t is constant. The result implies that if the history function starts from a population value that is

much smaller than the initial post-inoculation population value xo a post-inoculation LAG phase

() ()
followed by growth would be possible. A constant history function x t = ϕ t = xo , ∀ t ∈ ⎡⎣ −τ , 0 ⎤⎦

9
that is applicable to stationary (or LAG) pre-inoculation conditions is not compatible with (15) and
cannot therefore produce a post-inoculation LAG. An exponential pre-inoculation history function

() ()
such as x t = ϕ t = a e ( ) , ∀ t ∈ ⎡⎣ −τ , 0 ⎤⎦ can, on the other hand, comply with condition (15) and
σ t +τ

produce a post-inoculation LAG phase followed by growth. However, this is contrary to


experimental evidence for LAG phase occurrence as presented in the previous section (see Figure 2)
that indicated that a pre-inoculation constant population (stationary or LAG) will produce a post-
inoculation LAG, while a pre-inoculation exponentially growing population cannot produce a post-
inoculation LAG, just opposite from what we concluded here based on an approximation to the
May’s (1981) delay model. All other models cannot produce a LAG at all that is followed by
growth. To test these theoretical results and conclusions we present solutions to these different delay
models and attempt to simulate pre-inoculation conditions expressed via the history function ϕ t ()
consistent with the ones pertinent to the experimental evidence of LAG occurrence or lack of LAG
occurrence.

3 . D e l a y m o de l s a n d me tho d o f s o l u ti o n
3.1 – Malthus model with delay
As the main objective of this investigation is testing whether a LAG phase can be obtained from
a delay model subject to conditions similar to the ones pertaining to experimental evidence as
described in the introduction a simple delay model that accounts only for the short times following
re-inoculation is introduced. The delay model used is based on Malthus (1798) kernel, i.e.
x ( t ) = µmax x ( t − τ )        ∀ t ≥ 0 (16)
subject to a history function as the initial conditions in the form

()
x t =ϕ t () ∀ t ∈ ⎡⎣ −τ , 0 ⎤⎦ (17)

It is a short time approximation to May’s (1981) delay model presented in equations (3) and (4).
This is the simplest retarded type Delay Differential Equation (DDE) (Bellman and Cooke 1963).
Clearly this model is the Malthus (Malthus, 1798) equivalent delay model, which breaks down at
large times because x → ∞ as t → ∞ . The inhibition phase is absent in this model. Nevertheless,
for the purpose of testing whether a LAG phase consistent with experimental observations is present

10
or not this model is sufficient, and the way by which the solution stabilizes via an inhibition phase
towards a stationary phase is irrelevant to the present test.
The solution to the problem of the delay model presented by equations (16) and (17) was
obtained numerically. Originally, a tested Fortran DDE solver developed by Thompson and
Shampine (2006), based on a higher order Runge-Kutta method aimed for the solution of systems of
Delay Differential Equations (DDE), was used by choosing its default numerical parameters for
simplicity. The results obtained by using this model were compared with a solver developed by us
using the first order explicit Euler method of solution. It turns out that for the simple problem
defined by equations (16) and (17) it is redundant to invoke a solver that is designed to handle
systems of DDE’s with multiple delays and multiple history functions. The maximum relative
difference between the two solutions turned out to be less than 0.03%, i.e. x Eu − x RK x Eu < 3⋅10−4 ,

where x Eu is our Euler method solution and x RK is the Runge-Kutta solution obtained by using
Thompson and Shampine’s (2006) Fortran DDE solver subject to its default parameters. Therefore,
having confirmed the accuracy and validating our model’s results with an independently developed
solver we will present from now on the results obtained by using our Euler method solution. The
discretization of the selected time domain −τ ≤ t ≤ 4τ into 5N equal time intervals of size Δ t , i.e.

5N + 1 grid points in the form ti = i Δ t − τ , ∀ i = 0,1,2, ...., N ,....,5N , allows the finite difference
representation of equations (16) and (17) in the form
xi +1 = xi + ( µmax Δ t ) xi − N        ∀ i = N, N + 1, N + 2, ...., 5N (18)

()
xi = ϕ t i ∀ i = 0,1,2, ...., N (19)

()
Since the values of xi − N are known from the history function ϕ ti for the time range −τ ≤ t ≤ 0 ,

i.e. for i = 0,1, 2, ...., N the values of xi +1 can be evaluated recursively for all future times. This is in
essence the explicit Euler method of solution that we used for the solution of the problem (16), (17).

3.2 – May (1981) model with delay


The May (1981) delay model represented by equations (3) and (4) is solved by using the explicit
Euler method of solution having established its accuracy and validity. The discretization of the
selected time domain −τ ≤ t ≤ 4τ into 5N equal time intervals of size Δ t , i.e. 5N + 1 grid points in

11
the form ti = i Δ t − τ , ∀ i = 0,1,2, ...., N ,....,5N , allows the finite difference representation of
equations (3) and (4) in the form
µmax Δ t 2
xi +1 = xi + ( µmax Δ t ) xi − N −  xi       ∀ i = N, N + 1, N + 2, ...., 5N (20)
δ

()
xi = ϕ t i ∀ i = 0,1,2, ...., N (21)

()
Since the values of xi − N are known from the history function ϕ ti for the time range −τ ≤ t ≤ 0 ,

i.e. for i = 0,1, 2, ...., N the values of xi +1 can be evaluated recursively for all future times.

3.3 – Hutchinson (1948) delay model


The Hutchinson (1948) delay model represented by equation (1) subject to initial conditions (4)
is solved also by using the explicit Euler method of solution. The discretization of the selected time
domain −τ ≤ t ≤ 4τ into 5N equal time intervals of size Δ t , i.e. 5N + 1 grid points in the form

ti = i Δ t − τ , ∀ i = 0,1,2, ...., N ,....,5N , allows the finite difference representation of equations (1)
and (4) in the form
µmax Δ t
xi +1 = (1 + µmax Δ t ) xi −  xi xi − N       ∀ i = N, N + 1, N + 2, ...., 5N (22)
δ

()
xi = ϕ t i ∀ i = 0,1,2, ...., N (23)

()
Since the values of xi − N are known from the history function ϕ ti for the time range −τ ≤ t ≤ 0 ,

i.e. for i = 0,1, 2, ...., N the values of xi +1 can be evaluated recursively for all future times.

3.4 – Gompertz (1825) model with delay


The Gompertz (1825) delay model represented by equation (5) subject to initial conditions (4) is
solved also by using the explicit Euler method of solution. The discretization of the selected time
domain −τ ≤ t ≤ 4τ into 5N equal time intervals of size Δ t , i.e. 5N + 1 grid points in the form

ti = i Δ t − τ , ∀ i = 0,1,2, ...., N ,....,5N , allows the finite difference representation of equations (5)
and (4) by introducing the definition of y = ln x
yi = ln xi
(24)
yi +1 = yi + α Δ t − β Δ t yi − N       ∀ i = N, N + 1, N + 2, ...., 5N

12
()
yi = ln ⎡⎣ϕ ti ⎤⎦ ∀ i = 0,1,2, ...., N (25)

()
Since the values of yi − N are known from the history function ϕ ti for the time range −τ ≤ t ≤ 0 ,

i.e. for i = 0,1, 2, ...., N the values of yi +1 can be evaluated recursively for all future times.

4 . N um e r i c a l R e s ul ts
4.1 – Malthus model with delay
Three cases were considered consisting of initial conditions for the history function consistent
with the three experimental cases described in the introduction and in Figure 2 regarding re-
inoculation conditions that may or may not produce a LAG phase. A constant time step of

Δ t = 8 × 10−4 h was used in all calculations.


The first case consisted of constant pre-inoculation conditions attempting to approximate the re-
inoculation sample taken from a previous LAG phase. The parameters used were closely selected to
be consistent with experimental results presented by O’Donovan and Brooker (2001) in their Figure
2, which demonstrates LAG phases of varying durations. As such the value used for the maximum
specific growth rate was µmax = 1.548 h −1 and a delay time of τ = 8 h . The constant initial value for

() ()
the history function at the time interval −8 ≤ t ≤ 0 used was x t =ϕ t =1.15 × 105 cfu/ml (cfu =

colony forming units). The results presented on a ln x versus t graph are shown in Figure 3, where
it is evident from the inset (Fig. 3b) that contrary to the experimental evidence no LAG phase is
present. The latter result is obvious from the concave-down shape of the solution for t > 0 indicating
the lack of a LIP and consequently the lack of a LAG.
The second case considered consisted of pre-inoculation conditions consistent with the
stationary phase of a Logistic solution. The latter has a simple closed form analytical solution that
we used as the history function in the time interval −8 ≤ t ≤ 0 . The results presented on a ln x
versus t graph are shown in Figure 4, where it is evident from the inset (Fig. 4b) that contrary to the
experimental evidence no LAG phase is present. The latter result is obvious from the concave-down
shape of the solution for t > 0 indicating the lack of a LIP and consequently the lack of a LAG.
The third case considered consisted of pre-inoculation conditions consistent with an exponential
phase growth. The latter has a linear shape when presented on a ln x versus t graph. This
exponential function was used as the history function in the time interval −8 ≤ t ≤ 0 . The results
13
presented on a ln x versus t graph are shown in Figure 5, where it is evident from the inset (Fig. 5b)
that contrary to the experimental evidence an apparent LAG phase is present this time.

Constant pre-inoculation conditions Stationary Logistic pre-inoculation conditions

5 5
10 x in 10 cfu/ml y = ln x 10 x in 10 cfu/ml y = ln x

8 8

y = ln x
y = ln x

6 6

NO LAG
NO LAG
pre-inoculation

pre-inoculation
conditions

conditions
4
4

see inset see inset


2
for details 2
for details
Inoculation
Inoculation
0
(a) 0
-8 0 8 16 24 32 40
(a) -8 0 8 16 24 32 40
t [hours]
t [hours]

DETAIL : Constant pre-inoculation conditions DETAIL :


3 Stationary Logistic pre-inoculation conditions
3.5
5
x in 10 cfu/ml y = ln x 5
x in 10 cfu/ml y = ln x
2.5 3

2.5
2
y = ln x

2
y = ln x

1.5
1.5

1
pre-inoculation 1 pre-inoculation
conditions conditions
0.5 0.5

NO LAG NO LAG
0
(b)
0
-8 -4 0 4 8
(b) -8 -4 0 4 8

t [hours] t [hours]

Figure 3: Malthus delay model solution starting from Figure 4: Malthus delay model solution starting from
constant pre-inoculation conditions. A LAG phase stationary logistic pre-inoculation conditions. A LAG
is not present contrary to experimental evidence. phase is not present contrary to experimental evidence.

4.2 – May (1981) model with delay


Two cases were considered consisting of initial conditions for the history function consistent
with two experimental cases described in the introduction and in Figure 2 regarding re-inoculation
conditions that may or may not produce a LAG phase. A constant time step of Δ t = 8 × 10−4 h was
used in all calculations.

14
The first case consisted of constant pre-inoculation conditions attempting to approximate the re-
inoculation sample taken from a previous LAG or stationary phase. The parameters used were
µmax = 0.1875 h −1 and a delay time of τ = 8 h . The constant initial value for the history function at

the time interval −8 ≤ t ≤ 0 used was x t =ϕ t =1.15 × 105 cfu/ml (cfu = colony forming units). () ()
The results presented on a ln x versus t graph are shown in Figure 6, where it is evident from the
inset (Fig. 6b) that contrary to the experimental evidence no LAG phase is present. The latter result
is obvious from the concave-down shape of the solution for t > 0 indicating the lack of a LIP and
consequently the lack of a LAG.

Exponential pre-inoculation conditions Constant pre-inoculation conditions


0.9
16 5 5
x in 10 cfu/ml y = ln x x in 10 cfu/ml y = ln x
0.8
14
0.7
pre-inoculation

12
conditions

0.6 x 2 (t ) ⎞

NO LAG
10 ⎛

y = ln x
Inoculation
x ( t ) = µ max ⎜ x ( t − τ ) −
y = ln x

see inset ⎝ δ ⎟⎠
8 for details 0.5

pre-inoculation
conditions
6 0.4

4 Apparent LAG, 0.3 see inset


but it results from
2 an exponential for details
0.2 Inoculation
pre-inoculation growth
0
(a) -8 0 8 16 24 32 40 (a) 0.1
-8 0 8 16 24 32 40
t [hours]
t [hours]
DETAIL : Exponential pre-inoculation conditions
10
DETAIL: Constant pre-inoculation conditions
5
x in 10 cfu/ml y = ln x 0.5
9.5 5
y = ln x
x in 10 cfu/ml
0.45

9 0.4 ⎛ x 2 (t ) ⎞
x ( t ) = µmax ⎜ x ( t − τ ) −
Apparent LAG, ⎝ δ ⎟⎠
y = ln x

0.35
but it results from
y = ln x

8.5 LIP
an exponential 0.3
pre-inoculation growth
8
0.25
pre-inoculation
7.5
0.2
conditions
apparent LAG NO LAG
0.15
(b) 7 Inoculation
0 2 4 6 8 10 12 (b) 0.1
-8 -4 0 4 8
t [hours]
t [hours]

Figure 5: Malthus delay model solution starting from an Figure 6: May (1981) delay model solution starting from
exponential phase pre-inoculation conditions. A LAG constant pre-inoculation conditions. A LAG
phase is present contrary to the experimental evidence. phase is not present contrary to experimental
evidence.

15
The second case considered consisted of pre-inoculation conditions consistent with an
exponential phase growth. The latter has a linear shape when presented on a ln x versus t graph.
This exponential function was used as the history function in the time interval −8 ≤ t ≤ 0 . The
results presented on a ln x versus t graph are shown in Figure 7. Two values of µmax were used, i.e.
µmax = 1.548 h −1 presented in Figure 7(a) and, µmax = 0.1875 h −1 presented in Figure 7(b). While
the results show no LAG in both cases, they also show an oscillatory decaying solution (Figure 7a)
and a decay followed by growth (Figure 7b) both contrary to experimental evidence that predicted
an exponential growth.
Exponential pre-inoculation conditions Constant pre-inoculation conditions
5 1.4
y = ln x
Inoculation
y = ln x
1.2 5
4 x in 10 cfu/ml
5
NO LAG ; x in 10 cfu/ml
1
Oscillatory Decay
3
y = ln x

⎛ x 2 (t ) ⎞
x ( t ) = µmax ⎜ x ( t − τ ) − 0.8
δ ⎠⎟

y = ln x

2
pre-inoculation

0.6
⎛ x (t − τ ) ⎞
conditions

x ( t ) = µmax ⎜ 1 − ⎟⎠ x ( t )
⎝ δ
0.4
1
NO LAG
0.2 Inoculation

(a) 0 pre-inoculation
conditions
-8 0 8 16 24 32 40 (a) 0
0 50 100 150 200 250 300 350 400
t [hours]
t [hours]

Exponential pre-inoculation conditions


DETAIL: Constant pre-inoculation conditions
0.9 1.4
0.8 y = ln x
1.2
Inoculation y = ln x 5
0.7 x in 10 cfu/ml
pre-inoculation

5 1
conditions

x in 10 cfu/ml ⎛
x ( t ) = µ max ⎜ 1 −
x (t − τ ) ⎞
x (t )
0.6 ⎝ δ ⎠⎟
NO LAG ;
y = ln x

0.8
y = ln x

0.5 Decay followed by growth


0.6
0.4
⎛ x 2 (t ) ⎞
x ( t ) = µmax ⎜ x ( t − τ ) −
⎝ δ ⎠⎟ 0.4
0.3 pre-inoculation
conditions NO LAG
0.2 0.2
Inoculation
(b) 0.1 (b) 0
-8 0 8 16 24 32 40 -5 0 5 10 15 20 25 30
t [hours] t [hours]

Figure 7: May (1981) delay model solution starting from Figure 8: Hutchinson (1948) delay model solution starting
an exponential phase pre-inoculation conditions. Non- from constant pre-inoculation conditions. A
monotonic oscillatory growth is observed contrary to LAG phase is not present contrary to
experimental evidence. (a) under-damped oscillations experimental evidence. Under-damped decaying
oscillations are evident.
µ max = 1.548 h ; (b) over-damped oscillations appearing
-1

as decay followed by growth µ max = 0.1875 h .


-1

16
4.3 – Hutchinson (1948) delay model
The first case consisted of constant pre-inoculation conditions attempting to approximate the re-
inoculation sample taken from a previous LAG or stationary phase. The parameters used were
µmax = 0.1875 h −1 and a delay time of τ = 8 h . The constant initial value for the history function at

() ()
the time interval −8 ≤ t ≤ 0 used was x t =ϕ t =1.15 × 105 cfu/ml (cfu = colony forming units).

The results presented on a ln x versus t graph are shown in Figure 8 indicating a decaying
oscillatory solution, where it is evident from the inset (Fig. 8b) that contrary to the experimental
evidence no LAG phase is present. The latter result is obvious from the concave-down shape of the
solution for t > 0 indicating the lack of a LIP and consequently the lack of a LAG.
The second case considered consisted of pre-inoculation conditions consistent with an
exponential phase growth. The latter has a linear shape when presented on a ln x versus t graph.
This exponential function was used as the history function in the time interval −8 ≤ t ≤ 0 . The
results presented on a ln x versus t graph are shown in Figure 9 indicating a decaying oscillatory
solution. It is evident from the inset (Fig. 9b) that an apparent exponential growth is present however
this is only a portion of a decaying under-damped oscillatory growth as observed in Figure 9a. This
result is also contrary to the experimental evidence that suggested exponential montonic growth,
rather than oscillations.

4.4 – Gompertz (1825) model with delay


The first case consisted of constant pre-inoculation conditions attempting to approximate the re-
inoculation sample taken from a previous LAG or stationary phase. The parameters used were
µmax = 0.1875 h −1 and a delay time of τ = 8 h . The constant initial value for the history function at

() ()
the time interval −8 ≤ t ≤ 0 used was x t =ϕ t =1.15 × 105 cfu/ml (cfu = colony forming units).

The results presented on a ln x versus t graph are shown in Figure 10 indicating a diverging
oscillatory solution, where it is evident from the inset (Fig. 10b) that contrary to the experimental
evidence no LAG phase is present.
The second case considered consisted of pre-inoculation conditions consistent with an
exponential phase growth. The latter has a linear shape when presented on a ln x versus t graph.
This exponential function was used as the history function in the time interval −8 ≤ t ≤ 0 . The
results presented on a ln x versus t graph are shown in Figure 11 indicating a diverging oscillatory
17
solution. It is evident from the inset (Fig. 11b) that an apparent exponential growth is present
however this is only a portion of a diverging under-damped oscillatory growth as observed in Figure
11a. This result is also contrary to the experimental evidence that suggested exponential montonic
growth, rather than oscillations.

Exponential pre-inoculation conditions Constant pre-inoculation conditions


1.4 5
x
y = ln x 4
y = ln x
x
( )
= α − β ln ⎡⎣ x t − τ ⎤⎦
1.2
5 5
x in 10 cfu/ml x in 10 cfu/ml
3
1
2

y = ln x
0.8
y = ln x

1
0.6
⎛ x (t − τ ) ⎞ 0
x ( t ) = µ max ⎜ 1 − ⎟⎠ x ( t )
⎝ δ Inoculation
0.4 -1
Oscillatory - pre-inoculation NO LAG
Inoculation conditions
0.2 non montonic growth -2
pre-inoculation
conditions -3
0 0 50 100 150 200 250 300 350 400
(a) 0 50 100 150 200 250 300 350 400 (a)
t [hours]
t [hours]

DETAIL: Constant pre-inoculation conditions


2
DETAIL: Exponential pre-inoculation conditions y = ln x
1.4
5
y = ln x x in 10 cfu/ml
1.2 1.5
x
5
x in 10 cfu/ml
x
(
= α − β ln ⎡⎣ x t − τ ⎤⎦)
1
y = ln x

pre-inoculation 1
0.8
conditions
Apparent exponential:
y = ln x

Oscillatory -
0.6 non montonic growth
0.5
pre-inoculation
0.4
conditions
⎛ x (t − τ ) ⎞ NO LAG
x ( t ) = µmax ⎜⎝ 1 − ⎟⎠ x ( t )
0.2 δ
0 Inoculation

0
-8 -4 0 4 8 12 16 20
(b) -5 0 5 10 15 20 25 (b)
t [hours] t [hours]

Figure 9: Hutchinson (1948) delay model solution starting Figure 10: Gompertz delay model solution starting from
from an exponential phase pre-inoculation constant pre-inoculation conditions. A LAG
conditions. Oscillatory non-monotonic growth is phase is not present contrary to experimental
obtained contrary to experimental evidence. evidence. Under-damped diverging oscillations
Apparent exponential growth is observed in the are evident.
detail (b) but this is only a portion of a decaying
under-damped oscillatory growth observed in
(a).

18
Exponential pre-inoculation conditions DETAIL: Exponential pre-inoculation conditions
4 2
x
y = ln x x
( )
= α − β ln ⎡⎣ x t − τ ⎤⎦ y = ln x x
( )
= α − β ln ⎡⎣ x t − τ ⎤⎦
x
3 5 5
x in 10 cfu/ml x in 10 cfu/ml
1.5
Inoculation pre-inoculation
2 conditions

y = ln x
1
y = ln x

1
Apparent exponential:
Oscillatory -
0 non montonic growth
0.5
pre-inoculation
conditions Oscillatory -
-1 non montonic growth
0
-2 (b)
(a) 0 50 100 150 200 250 300 350 400 -8 -4 0 4 8 12 16 20 24
t [hours] t [hours]

Figure 11: Gompertz delay model solution starting from an exponential phase pre-inoculation conditions. Oscillatory
non-monotonic growth is obtained contrary to experimental evidence. Apparent exponential growth is observed in the
detail (b) but this is only a portion of a diverging under-damped oscillatory growth observed in (a).

5 . A l t e r na ti v e mo d e l s tha t c a p tur e the L A G pha s e c o ns i s te n t


w i th e x pe r i me n ta l e v i d e nc e
Delay models that introduce explicit delays in the population growth consistently produce
results that are contrary to experimental evidence. The latter conclusion was reached both from the
theoretical analysis results as well as from the numerical solutions of the different delay models
presented. As a result we will present shortly the type of models without delays, which can capture
the LAG phase in a manner that is consistent to experimental evidence. The main feature that these
non-delay models have in common is the existence of unstable stationary points, and these form the
reason for the appearance of the LAG phase. In such models the LAG duration is a RESULT of the
model’s solution subject to given parameter values and initial conditions, rather than an apriori
imposed DELAY that is set upfront as input into the model, rather as part of the solution.
Vadasz and Vadasz (2002a,b) derived their neoclassical model from first biological as well as
physical principles. The monotonic version of this model takes the form
⎡ x
x = µmax ⎢ 1 − +
( zo − µmax ) ⎤ x
⎥ (26)
⎣ δ µmax (1 + rm x ) ⎦

where rm = m1 mo is the ratio between the viable biomass constant coefficients m1 and mo
corresponding to a linear constitutive relationship between the total viable biomass and the cell
concentration in the form M ( x ) = mo + m1 x , and z is related (but not identical) to the nutrient (or

19
resource) consumption/utilization rate and for monotonic growth z = zo = const. , where zo
represents the initial condition of z at t = 0 , defined in the form
⎡ x ⎛x ⎞⎤
zo = µmax + (1 + rm xo ) ⎢ o + µmax ⎜ o − 1⎟ ⎥ (27)
⎣ xo ⎝δ ⎠⎦

It is straightforward to observe that when zo = µmax , corresponding to a specific particular set of


initial conditions, eq. (26) produces the particular case of Pearl’s (1927) Logistic Growth Model
(LGM). Vadasz and Vadasz (2005) presented an analysis of this model for monotonic growth and
demonstrated its capability to capture correctly the LAG phase. An example of comparing the
model’s analytical solution results with experimental data based on O’Dnovan and Brooker (2001)
for four different growth curves differing mainly by different LAG durations is presented in Figure
12 where it is evident that the model fitted excellently the experimental data.

Neoclassical Model (Vadasz & Vadasz, 2002a,b) .


results compared to experimental data of x
O'Donovan & Brooker (2001) x Phase Diagram
0.00035
10
8
δ ( rmδ - 1 )
xA =
µ max = 0.0003 2 rm δ
Cell Count, x (cfu/ml)

Locus of LIP
d
µ max ( rm δ - 1 )

c 0.00025 LGM
7
10
b
rmδ

(a) Model 0.0002


a (a) Experimental
(b) Model
6 0.00015
10 (b) Experimental
(c) Model
(c) Experimental 0.0001
x/x

(d) Model
.

(d) Experimental
1.10
5
10 -5
x =δ =
7
5 10
0 5 10 15 20 25
time (hours) 0 xA x
6 6 6 6 7 7
0 2 10 4 10 6 10 8 10 1 10 1.2 10
Unstable Stationary Points Stable Stationary Points

( ) ( )
2
zo = − µ max rmδ − 1 4rmδ zo = 0 zo = µ max rmδ − 1 rmδ zo = µmax

Figure 12: Comparison of the Neoclassical Model Figure 13: Phase diagram for the solution of monotonic growth
analytical solution for monotonic growth based corresponding to rmδ > 1 , in terms of the specific
on Vadasz & Vadasz (2002a,b) with the
experimental data based on O’Donovan and growth rate x x versus the cell concentration x , for
Brooker (2001) (here redrawn from published −7 −1
rm = 5 ⋅ 10 (Cell/ml) , δ = 10 Cell/ml
7
data). The initial cell concentration is and

xo = 1.1458 × 10 (cfu/ml). (Vadasz and Vadasz


5 µ max = 3 ⋅ 10 s . (Vadasz and Vadasz 2005)
−4 −1

2005)

20
An analysis of the model can be performed by drawing a phase diagram in terms of x x versus
x . The latter is done without solving the model but rather by presenting equation (26) in the form

x ⎡ x
= µmax ⎢ 1 − +
( zo − µmax ) ⎤
⎥ (28)
x ⎣ δ µmax (1 + rm x ) ⎦

and plotting the left hand side of (28) versus x by using the function on the right hand side of (28)
for different values of the parameters δ , µmax and rm , and initial conditions zo (or xo and xo ). An

example of such a phase diagram is presented in Figure 13 for rm = 5 ⋅10−7 (Cell/ml)−1 ,

δ = 107 Cell/ml and µmax = 3 ⋅10 −4 s −1 . From the phase diagram presented in Figure 13 it can be
observed that the positive x axis consists of a continuous distribution of stationary points where
( x˙ x) = 0 . The stationary points to the right of the point x A represented by the continuous thick line
are stable, while the stationary points to the left of x A represented by the dotted line are unstable as
observed in Figure 13 by following the direction of the arrows representing the solution change in
the positive time direction. Any point on the phase plane represents a possible initial condition.
Once such a point is set (i.e. an initial condition for both x o as well as x˙ o ), the solution follows the
corresponding curve that passes through that point in the direction of the arrows towards a stationary
point. From equation (28) it is easy to observe that the value of zo is identical to the value of the
specific growth rate at x = 0 , i.e. zo = ( x˙ x ) x =0 . The reason for the possibility of a LAG is the

existence of positive but unstable stationary points to the left of point x A represented by the dotted
line on the x axis in Figure 13. When the initial conditions are sufficiently close to one of these
unstable stationary points the solution spends a relatively long time to escape from its neighborhood.
This implies that if 0 < xo << µmaxδ , i.e. x˙ o is positive but sufficiently small, and x o < x A , then a
LAG exists. The value of x A was evaluated by Vadasz and Vadasz (2005) to be x A = (rmδ − 1) 2rm .
The threshold value x A for which x o < x A is one of the conditions for the possibility of a LAG phase

and for the parameter values used in Figure 13 its value is x A = 4 × 106 Cell/ml . The threshold value

indicated by Maier (2000) based on experimental evidence was about 106 Cell/ml . The analysis
presented by using the phase diagram is completely consistent with experimental evidence as almost
constant pre-inoculation conditions imply initial conditions for x˙ o to be sufficiently small as
required above as a condition to obtain a LAG. On the other hand pre-inoculation conditions

21
consistent with exponential growth lead to a very large value of x˙ o and consequently from the phase
diagram a LAG becomes impossible. Vadasz and Vadasz (2002a,b, 2005) used a definition for LAG
duration λrd different than the classical one λcl as presented in Figure 1. That definition is
consistent with their findings that the LAG is essentially related to the existence of unstable
stationary points and therefore define the LAG duration λrd as the amount of time that elapses until
the solution reaches a value, which is by a certain percentage above the corresponding unstable
stationary point x 3s (see Figure 13). This implies that xλ = b x 3s , where b > 1 is a constant that

specifies by how far is xλ from x 3s . We define the LAG duration, λrd , as the time needed for the
solution to reach the value xλ for any predetermined value of b . This definition is very similar to
the way one defines the time needed for a monotonic solution to reach a steady state. In most of the
computations a value of 3.5% above x 3s was used as the LAG threshold, i.e. b = 1.035 .
A different non-delay model was introduced by Baranyi and Roberts (1994) that has the ability
to capture a LAG phase. This model is non-autonomous, i.e. one of its parameters is explicitly time
dependent. However, Vadasz and Vadasz (2007) transformed Baranyi and Roberts (1994) model
into an equivalent autonomous form and demonstrated that its ability to capture the LAG phase is
due to the existence of unstable stationary points in a manner quite similar although not identical to
Vadasz and Vadasz (2002a,b, 2005) neoclassical model.
Both the neoclassical model (Vadasz and Vadasz 2002a,b, 2005) and the Baranyi and Roberts
model (Baranyi and Roberts 1994, Vadasz and Vadasz 2007) share three properties, namely the
existence of unstable stationary points that are the reason for the LAG, both include in some way the
metabolic mass transfer process via the resource consumption/utilization, and both are degenerated
second order systems. Therefore, models that account for a delay in the metabolic process explicitly
such as Wood et al. (1994, 1995) can be expected to capture a LAG phase even without introducing
delays in the metabolic process.

6 . C o n c l us i o ns
Since all the theoretical and numerical results presented for delay growth models are contrary to
the experimental evidence regarding the conditions for the occurrence of a LAG phase we may
conclude that delay effects and the LAG are two distinct biological phenomena. It therefore dispels
the incorrect notion that delay effects represented by delay differential equations are the biological
22
reason behind the LAG phase in microorganism growth. The resulting consequence so far is that the
only other reason for the LAG phase is the existence of unstable stationary states as proposed by
Vadasz and Vadasz (2002a,b), Vadasz and Vadasz (2005) and Vadasz and Vadasz (2007). The latter
are a result of accounting for the microbial metabolic mass transfer in the population growth
process.

Acknowledgments
One of the authors (PV) is grateful and sincerely expresses his gratitude to Professor Peter Chesson
from University of Arizona for an insightful discussion on the topic.
This material is based upon work supported by the National Science Foundation under Grant No.
CTS-0500466. The authors wish to thank the NSF for the funding support.

References
Baranyi, J., Roberts, T.A., (1994) A dynamic approach to predicting bacterial growth in food.
International Journal of Food Microbiology 23: 277-294.
Bellman, R., Cooke, K. L., (1963) Differential-Difference Equations, Academic Press.
Gompertz, B., (1825) On the nature of the function expressive of the law of human mortality, and a
new mode of determining the value of life contingencies. Philosophical Transactions of the
Royal Society of London 115: 513-583.
Hutchinson, G.E., (1948) Circular casual systems in ecology. Ann. N.Y. Acad. Sci. 50: 211-246.
Lebowitz, J.L. Rubinow, S.I., (1974) A Theory for the Age and Generation Time Distribution of a
Microbial Population. Journal of Mathematical Biology 1: 17-36.
Lodge R.M., Hinshelwood, C.N., (1943) Physiochemical Aspects of Microbial Growth. Part IX. The
Lag Phase of Bact. Lactis Aerogenes. Journal of the Chemical Society: 213-219.
Maier, R.M., (2000) Bacterial Growth. in Environmental Microbiology, Maier, R.M., Pepper, I.L.,
Gerba, C.P. (Ed.), pp. 43 – 59, Academic Press.
Malthus, T.R., (1798) An essay on the principle of population, Penguin, Harmondsworth, England.
May, M. Robert, (1973) Time-delay versus stability in population models with two and three trophic
levels. Ecology 54: 315-325.
May, M. Robert, (1978) Mathematical aspects of the dynamics of animal populations. in: Levin, S.
A. (Ed.), Studies in Mathematical Biology – Part II: Populations and Communities, Studies in
Mathematics Vol. 16, The Mathematical Association of America, pp. 317-366.
May, M. Robert, (1981) Models for single populations. in Theoretical Ecology, Robert M. May
(Ed.), Blackwell Scientific Publications, Oxford, pp. 5-29.

23
McKendrick, A.G., (1926) Application of Mathematics to Medical Problems. Proceedings of the
Edinburgh Mathematical Society 44 (1): 98-130.
O’Donovan, L., Brooker, J.D., (2001) Effect of hydrolysable and condensed tannins on growth,
morphology and metabolism of Streptococcus gallolyticus (S. caprinus) and Streptococcus
bovis. Microbiology 147: 1025-1033.
Pearl, R., (1927) The growth of populations. The Quarterly Review of Biology II (4): 532-548.
Pirt, S.J., (1975) Principles of Microbe and Cell Cultivation, p.11 & p.194, John Wiley & Sons.
Thompson, S., Shampine, L.F., (2006) A Friendly Fortran DDE Solver. Applied Numerical
Mathematics 56: 506-516.
Vadasz, P., Vadasz, A.S., (2002a) The Neoclassical Theory of population dynamics in spatially
homogeneous environments – Part I: Derivation of universal laws and monotonic growth.
Physica A 309 (3-4): 329-359.
Vadasz, P., Vadasz, A.S., (2002b) The Neoclassical Theory of population dynamics in spatially
homogeneous environments – Part II: Non-monotonic dynamics, overshooting and oscillations.
Physica A 309 (3-4): 360-380.
Vadasz, P., Vadasz, A.S., (2005) Predictive modeling of microorganisms: LAG and LIP in
monotonic growth. International Journal of Food Microbiology 102: 257-275.
Vadasz, P., Vadasz, A.S., (2007) Biological Implications from an Autonomous Version of Baranyi
and Roberts Growth Model. International Journal of Food Microbiology 114: 357-365.
Vance, R.R., Newman, W.I., Sulsky, D., (1988) The Demographic Meanings of the Classical
Population Growth Models of Ecology. Theoretical Population Biology 33: 199-225.
Verhulst, P.F., (1838) Notice sur la loi que la population suit dans son accroissement. Corr. Math. et
Phys. Publ. par A. Quetelet. T. X: 113-121.
von Foerster, H., (1959) Some remarks on Changing Populations. in The Kinetics of Cellular
Proliferation, F. Stohlman Jr. (Ed.), Grune & Stratton (publisher), pp. 382-407.
Wangersky, P.J. and Cunningham, W.J., (1957) Time lag in population models. Cold spring Harb.
Symp. Quant. Biol. 22: 329-338.
Wood, B.D., Dawson, C.N., Szecsody, J.E., Streile, G.P., (1994) Modeling contaminant transport
and biodegradation in a layered porous media system. Water Resources Research 30 (6): 1833-
1845.
Wood, B.D., Ginn, T.R., Dawson, C.N., (1995) Effects of microbial lag in contaminant transport and
biodegradation modeling. Water Resources Research 31 (3): 553-563.

24

You might also like