Chapter 3

Continuous Models: Ordinary differential equations

3.1 Introduction
In contrast to the last chapter, we will now use continuous models, i.e. mainly models
which include a continuous time course. This kind of modelling is done by differential
equations. In this chapter, we consider only ordinary differential equations and systems
thereof. This allows to consider only one independent variable, e.g. time, but for a lot of
problems, this is sufficient. Roughly speaking, for spatial modelling we will need also
partial differential models (which is done later).

Example: Spread of a rumour

In a human population of fixed size N , a rumour is distributed by word-of-mouth
advertising, i.e. a member of the population comes to know about the rumour by
narration of another person. Let I(t) be the number of people who know already about
the rumour at time t. Model assumption: Each informed person has k > 0 contacts per
time unit with members of the population, and - very human of course :-)tells the
rumour each time. If q is the fraction of non-informed people, then each informed
person has
qk contacts with non-informed people. At time t, it is

q = N − I(t) .
N
Thus, in the very short time interval dt, each informed person had contact to
N − I(t)
qk dt = k dt
N
to non-informed people, which corresponds to the number of newly informed persons
(by this certain information source). Thus, including all informed persons at time t,
there are
N − I(t)
dI = I(t) k dt,
N
newly informed persons in the time interval dt, which leads to the differential equation
dI = I(t) N − I(t) k.
dt N
This can be reformulated to
dI k k
= kI − I2 = I(N − I).
dt N N

It has exactly the form of the logistic differential equation, a very famous ODE,
which we will study in detail later. Also, the solution will be determined later, here it
is only mentioned to be of the form
N
I(t) = .
1 + (N − 1)e−kt
If t → ∞, then I(t) N , which means, that sooner or later each member of the
population
→ knows the rumour. This maybe shows that our assumptions were not too
realistic ...

The appearance of the logistic differential equation is a typical example that a certain
differential equa- tion (or difference equation or whatever) may describe very different
processes. So, one solution or model analysis may help for understanding the
behaviour of many different models!

3.2 Stability and attractiveness

An interesting and important topic for models is the stability and attractiveness of
solutions, which should be defined carefully.

Let f ∈ C0,1−(R × Rn, Rn), x˙ = f (t, x), let x(t, t0, x0) be the maximal solution with x(t0) = x0.
Let
x̄ ∈ C 1 ([t0 ,∞ ), Rn ) be a fixed solution of the ODE with x̄(t0 ) = x̄ 0 . All solutions
exist globally in a neighbourhood of x̄, for given initial data.
Deftnition 16 (Stability, Lyapunov) x̄ is called stable, if

∀ t1 ≥ t0 ∀ ε > 0 ∃ δ > 0 : ("x0 − x̄ 0 " < δ ⇒ "x(t, t1 , x0 ) − x̄(t)" < ε in [t1 , ∞))

K(x)

K(x)

x̄ is called uniformly stable, if δ does not depend on t1 (which is always the case if f is
not dependent on or periodic in t).

Deftnition:(Attractiveness) x̄ is called attractive, if

∀ t1 ≥ t0 ∃ δ0 > 0 : ("x0 − x̄ 0 " < δ0 ⇒

t→∞
lim "x(t, t1 , x0 ) − x̄(t)" = 0).

K (x)

In this case, all solutions starting in the δ0 neighbourhood of x̄(t1 ) are

by“attracted” x̄. attractor for that trajectory.Domain of attraction:x̄ is called

A(x̄) = {x0 ∈ Rn | ∃ t1 ≥ t0 : lim "x(t, t1 , x0 ) − x̄(t)" = 0}

x̄ is called uniformly attractive, if δ0 does not depend on t1 (which

is the case if f is not dependent on or periodic in t).x̄ is called
asymptotically stable, if x̄ is stable and attractive.

3.1 Verhulst equation

In the last section, it was assumed that there are unlimited
resources for population growth, that means that the increase is
proportional to the present population (leading to the
exponential growth), or the per capita birth rate is constant and
density-independent. This assumption is justified only for
certain populations at the beginning of their development. To
describe a more realistic situation, it is necessary to incorporate
limited resources into the model. This was done by Verhulst (see
by the following assumptions:
A biotope has a capacity K(t), where K(t) > 0 is a continuous function
• For small densities of populations, there is approximately exponential growth with an
exponent
a(t), where a(t) > 0 is continuous.
The increase u˙of the population density u is additionally proportional to the remaining living
space ((K − u) resp.
u (1 − )).
K

Solving the time-dependent Verhulst equation can easily be done by the so-called
“trick of Riccati”: We introduce a new variable v(t) = 1 . W.l.o.g. let t0 = 0, v0 = 1 .
Hence we get: u(t) u0
u̇(t)
v̇(t) = − 2
u (t)
u(t) )
a(t)u(t)(1 − K(t)
= −
u2(t)
1 1
= −a(t)( − )
u(t) K(t)
a(t)
= −a(t)v(t) +
K(t)

Apply the variation of ∫t

e−(A(t)−A(s)) a(s)ds, (A(t) − A(s) = a(τ ) dτ ),
constants:
K(s) s
⇒ v(t) = v0e−A(t) +
0

∫ t

0 K(s) s

∫ t
where A(t) = t0a(s). Return to the variable u(t) = 1v(t)
and we get the solution:

u(t) u0
=
∫ t −(A(t)−A(s)) a(s)
e−A(t) + 0 u0 e K(s) ds
For the special case a, K = const we get the following explicit solution of the time-
independent Verhulst equation:
u(t) = u0 ∫
e−at + u a e−at t eas ds

0K 0
= u0 = u0 K
u0
=
e−at + u0 (1
K − e−at)
e−at(1 −
K u0 )
K e−at(K − u0) + u0
u0
+

12

10

0
-4 -2 0 2
4 6 8 10
t

Here, we can consider the stationary solutions and their behaviour. The time-
independent Verhulst equation can be written in the form
u˙ = f (u) = αu − βu2.
A short insight can be gained by the following simple procedure:
u’=f(u)



Zeros of f (u)
stationary
points f (u) positive:
Arrow to the
 right f (u) negative:
Arrow to the
left.
Hence we get: Let ȳ a stationary point.
f ′ (ȳ ) < 0: ȳ stable stationary point
f ′ (ȳ ) > 0: ȳ unstable stationary point.
Further special cases of the Verhulst
equations:
for t → ∞.
1. Let the function K tend to a positive
limit K̄ ,
K(t) → K̄

Furthermore, for the positive function a it is assumed that

0
∫∞
a(t) dt = ∞.
Then, the rule of de l’Hospital yields the limit for equation (3.6) with u0 > 0:
R
e t0a(s) dsa(t)
lim u(t) = lim = K̄ .
t→∞ R t a(s) ds
t→∞ a(t) 0 e
K(t)
 If K has a positive limit, then u(t) tends to this limit (if it is positive at all). For
condition it is sufficient that a has a positive bound.
These conclusions are true in particular if K(t) = K̄ is constant, which seems to be the
biologically
most relevant case. If u0 > 0 and a has a positive lower bound (no further
conditions), then u(t) tends to K̄ .
2. The parameters are assumed to be periodic functions with period T . Assume the period
to be
T = 1, hence
a(t + 1) = a(t), K(t + 1) = K(t).
A solution with period 1 and initial value u0 satisfies u1 = u(1) = u0. By formula we get
u0 )
R 1 ∫ R 1
e− a(s) ds + u 1 e− a(τ ) dτ a(s) ds
which leads to
R
1− e 0 1 a(s) ds
u = .
∫ 01 − R 1 a(τ ) dτ a(s)
e0 s
K(s) ds
For the Verhulst equation with periodic parameters, there is exactly one periodic
solution (except for u = 0). The right hand side of (3.8) defines a function g : R
→ R with g(u0) = u1 which has the following properties: g(0) = 0, g(u0) < u0
for large u0, g is strictly monotone increasing and concave. Hence, all solutions
with initial data u0 > 0 converge towards the unique non-trivial periodic
solution.
Assume a to be constant (just for simplification). Then the periodic solutions reads
u(t) = −at ∫u0
e + u t e−a(t−s) 1 ds
00 K(
with the initial s)
value u0 = ea −
1
.
∫ 1 a(1−s) 1
a e 0 dsK(s)
Using this expression for u0, we get after several steps
Using this expression for u0, we get after several steps
∫ 1 −a(1−s) 1 ∫ t −a(t−2) 1 −1
e ds e
ds
0 K(s 0 K(s
)
u(t) = ) e−at ∫
(1 − e−at) .
∫1 + 0
t
0 −a(t−s)
e−a(1−s) e
ds ds

The limit a → ∞
yields u(t) = K(t).
That means: For a large, the population adapts fast to the changes of the →
capacity. For a Σ 1 0
the limit is
exist
3.2 Ecosystems Modelling

Example: We consider plankton in the ocean. It is limited by the element nitrogen

(which is essen- tial, i.e. plankton cannot grow without any nitrogen). There are two
kinds of plankton:
• Phytoplankton (plant plankton) : can perform photosynthesis and therefore
requires essential ele- ments
3.2.1 Zooplankton (animal plankton) : they feed on phytoplankton

Assumptions for the model approach:

3.2.2 Everything is well-mixed (so, we can deal with spatially independent concentrations)
3.2.3 The system is closed to nitrogen
• Nitrogen can be taken up from the ocean by phytoplankton (and then is
incorporated into the phytoplankton)
3.2.4 Zooplankton can incorporate nitrogen by consuming phytoplankton
• Death and excretion from the plankton recycle the nitrogen to the ocean - the
delaying process of decay and remineralisation is neglected here
The following notations are used: N describes the concentration of available nitrogen, P
the concentration of phytoplankton, and Z the concentration of zooplankton. Remark,
that we take P and Z to be in the unit of mass of nitrogen which is incorporated in the
plankton per unit surface area of the ocean.

The basic model (with linear functional responses) of these processes can be formulated as follows:
 Ṅ = aP + bZ − cN P
P˙ = cN P − dP Z − aP
Obviously, these equations Z˙ = dP Z − bZ.
satisfy
Ṅ + P˙ + Ż = 0,
 which leads immediately to N + P + Z = A. This means that the nitrogen is conserved, A is
the total concentration of nitrogen in the system.
If the population can adapt only slowly, it stays nearly constant, at a value of the
harmonic mean value of the capacity.

3.5Compartmental models
The idea of linear compartmental models is to define different states, together with a directed graph which
denotes possible transitions between the states. Each edge of such a graph is equipped with a rate. If there is a
connection to the “outer space” of the system (i.e. there may be an in- or outflow from the system), the system
is open. Speaking in terms of a population, the total number of particles is not preserved, there may be
emigration or immigration. Otherwise, the system is called closed. In the graph, this means: If there is an edge
going outside (or from outside to) of the system, it is not closed.
Here two examples for linear compartmental models:
State 1 State 2 State 1 State 2
x y x y

State 3 State 3
z z

The left one (a) shows a linear compartmental model in a closed system, the right one
(b) for a open system. Let x, y, z denote the the densities of the states 1, 2, 3. Then the
equation for the closed system reads

3.6Treatment of Hepatitis C

Hepatitis C is a viral infection which is quite prevalent in the population. About 2-15
% are infected, most of them without symptoms, but they have a higher risk to
develop cirrhosis and liver cancer.
Present treatment strategy: The patients get high doses of interferon-α (IFN) over a
period of about one year.
Not much is known about the viral dynamics and the mechanisms of the IFN therapy,
and the early time prediction about treatment success is quite poor (but would be
interesting since the adverse effects are pretty serious).
Questions to a model are:
• Is it possible to gain insight into the mechanisms of interferon-α?
• Is it possible to predict the success or failure of the treatment for a certain
individual short after the beginning of the treatment?
• Is it possible to enhance the treatment, if it fails?
These questions are approached in a series of articles (Neumann et al. [47]). In the
figure, there are data about the decline of the virus load during the treatment with
interferon-α (data of three patients). After the onset of treatment, these data show
three phases:
1. Delay (1-3 hours)

80
 2. Phase 1: sharp decrease of the total virus load (≈ 2 days)
3. Phase 2: slow decrease of the total virus load (during several months)

How can the rapid decline during the first day and the significantly slower rate later be explained?

Some background information: Target cells are infected by virons, free viral particles
of Hepatitis C. These infected cells can be detected by the immune system and thus
have a higher clearance rate. Addi- tionally, these infected cells “produce” new virons
which can infect more target cells ... This basic model has the following structure:

Target Cells
VT Cells

pI I

Virons V

cV Death

State of the system:

• Density of the target cells T
• Density of free virons V
• Density of infected cells I
Dynamics: The following assumption is made: The population of target cells is
constant, T (t) = T (only slightly influenced by infection). Since their number should
be quite large, one can assume that, even if the viral infection is quite prevalent, only
a small fraction of target cells is infected. This yields that the incidence (the number
of newly infected cells per time unit) is proportional to the number of free virons.
81
Additionally, the infection is assumed to be in an equilibrium when the therapy is
started. The death rate of infected cells is denoted by δ. The “production” of new
virons depends on the number of infected cells, thus there are pI new virons per time
unit. c denotes the so-called clearance rate for free virons. Taken together, this
yield the following model equations:
I˙ = βT V − δI
V˙ = pI − cV.

Assuming T to be constant, this is a linear system.

82
3.7 Epidemic models:

Another famous example for a compartmental system are epidemic models. There is a
broad collection of models of this type, depending e.g. on the properties of the given
disease and its pathogens. Here, we can only have a short look into this class of models,
we do this for the example of the so-called SIR-model.

The total population N is divided into distinct classes, one very common possibility is to take

S = Susceptibles (who can catch the disease)

I = Infectives (who have the disease and can transmit it
to others)
R = Removed/Recovered/Immunes (who cannot catch
the disease) ,

where N = S + I + R.

83
 S (death) b(S+R)
+ bI
(birth)
S
S
vaccination
SI/N (infection)
R
(loss of
immunity)

R I recovery
I

R (death)
(death)

A corresponding ODE system can read as follows:

(−ΨS) + γR
Ṡ = −µS +b(S + R) + b̄I −β
SI
N −αI
I˙ = −µ̄I+β SI N
Ṙ = −µR+αI(+ΨS) − γR
the meaning of the parameters is explained in
the figure above

This model can be extended by

modification of certain terms,
introduction of new classes, time
de- lay ... many possibilities!
In contrast to the Hepatitis model, we consider
here also nonlinear terms.

84
3.8 Kermack-McKendrick epidemic model:
It describes the dynamics resp. the persistence of a disease
which is transferred by a direct contact of an infective with
a susceptible individual. The model is well-suited to describe
typical diseases which are caused by bacteria or viruses, like
smallpox or measles.
Assumption: There is no population dynamics, the disease is
dispersed on a much faster time scale than a time scale on
which demographic changes are relevant - so this assumption is
justified in most of the cases.
Ṙ = αI − γR
If α is the recovery rate, then 1/α is the mean retention period
≤
−αx
in class I and P (t x) = e is the probability that an
individual which was infected at time 0 is still infected at
time x (it is the survival probability; the underlying
probability distribution is the exponential distribution). Such
a model is classically called a SIRS-model, indicating the
typical way of an individual through the different classes: A
susceptible becomes infected, then recovers and (maybe after a
certain time) becomes susceptible again.

Check, if the total population N stays constant:

Ṅ = Ṡ + I˙ + Ṙ
= −βSI + γR + βSI − αI + αI − γR
=0