Stab

Una l

BircaHon
IheJ an
Changa o StabillH
x
beueen
po'n b n ex cha ng stabilk
bl nel Ja adonita bi ca

Lec1 18tolao
* *****

P'tchok BiUY Cakien **

81r calon
There are plkhork
Supper CTik Ca pikhok
Suberiicel Pichftk

Super Crical poot

d+
poinH are
The criHeol

19 aluway a
ixod ponk,

ord an -

S he only Soluhbn
7

M-)o o ithe hly Souh

Phase pokrai- d
d
- where

Ad/ar
d M-A

when Mlo

galv
>(als1

unen

-2A

Etabl

unstal
*x
X* Lineou Stabiliky Areygir
***
LA
evalwale ak al theixed pointy

AF ,-

i unslabl
Stobl

SHable

(u-) o
94he ony
 A-X)o
4 h ony Sol
*GUrcc}dn diaq-am
13 weakly Stable weak Stabilly
inea
hstabildy is no caput h
tabikky anauy cir A phaje poaik

stabliy ad
fonslank Axod poîr
stablepon CuLcaled Supar
Crh o bErcalk
Noumal dX x-
d

**SubCuuih ce

Noumal

Psr Kpeinls :
A
aluway x pon
i s

(-1a (y+)

th Only Soluk
3

Ahe only Soluhn

i9
A o
3

Phage orart
 fx/4

uahan

*»
** Lineau
***
9tabilik Analysis
Lel
eualwod
all the xd poink
A
Ay i unsable
A .

Stab

o , weaKl) unjlab

Fou -2jJM

isunstable ot -¥-A
-24

So iunJtobla

A -

GiUr CoHdn di0gram

***

*-
+(onFQn poinl Jwitch
SHab ana wo unskala
Vaniiha SubciHcal pllthor brCo
epoh
NOuumo Ym u+

Lec-12 25|1o 20724

*

S t t -9ubraic mod +
Spuwte buduosm) mode
budwm o tMe t, 4no
Le be 9uus
densihy 4h
he

imiled Ca RAenkel

uuedaln -iMeperdeny roudth

d
dt
Capauy sheh s
whea Ko astuy m and he paramdr
Con S l a n
assumed hea Oc o budum.
in rin il qrouh
PuedafKeh
NouD we mClude
4ho 4h pRdaktn
ayafon O. T i8 asumod
bud uvm Proy
brdy depe nds on

bud uuy m
by SakroHs, value wne
and i aPproaches
deh Sttu

vat
i n appuLoYnala mall hile

4he puu dlation i ver

belod whi
abpu PucdoHon
PuundaiUDerm P(U1S ge by (Hollmg
Value he
Un cona tm B(o) is 4he
where
Jumoua saa
MAxi mum
P a) = B budwm due r
h
A+ and ACo) i9
Pusedot
+he thre ghod Value Ke

drou Mp hpudoion

A
Ahal Puuedain 1 Siq mo i tuw

hovaser
A pudaHon is Smal
F
i Switched or
while A
A the
Stnte anSa
PA)= ;we
h
whih maU Aha denJty
8o uroion Con Stan
ho uh P u d aHn
dmoual Tala pi

bud Wm a maimum val B

mlddle
JARadha

IncopoTakna Such Puudatonal

buduve m may pe npyens by

BC

neglect tho 9 r o u t
pdar
hese octr
bra) and hos rand Srve) af
Compaul son b ha
Sloue rme Scale in

bududmm
he ea ) hay 4 pauameterr BKB B ,A .
A and KB haue Ahe Sama dimen sion as

Bh0r dimenslon (me) ard 8 has di mon Jion

aime)
We intuodue no dimen siona auan kHes b
wtm

noh ler

d
d
Tonrom i implies

y ang -4))
)
 dt and du
A

Henes, om-
l - -

1+
has only tu
Ahal Aisquodon d L
Noe
which aue dimcn°rlsy
POamees Ja and o

Equilibvibn point
Ahe SbluHon h e Ra
RQui brfUm poiny
Th

dt

.. q-0
T4u an omr

Poin
an equ ltbrium
clealy u athe
poste

Donts 3 noh
equl' Briun Compik

datemin
Con
Cubil eawaH0. w Can be
The e" CS
quaphiCal
is}en

o to
Koue

pe capia if po apfta
L.HS
+he RHS
Eölad vaulabla
puaedakO
The no- trivia
dea due
in rfe po

e q u i l t b r i m
Pon
CuTVe

ne 'u C
whf meeh
lin
9(u 19 9haiht
Heu,
Mulualism: d e p °n d s On

Tn this Cas wO0) more SpeLic

Ahovefme
nd Suuv{vo
enhan cad) in

'Qal poplaHsh is

guth hRuaCHdn,

LotkaVolteu

Puy moda
dator
modo|> ( be 4h
umuloahon and
+he nd
Let
Paudal modo
equoHns

he
n udo to mulo assumpf
he olloush
Speie at.a gien
The ous)h
ak ay
PudpuH dnal ha no
time a that ima.
4ho Spei. Puy en)
a honmoge n o u

in
9peies
enviuucnment
and ha
atun
no toke
o pe dattr the Puay
abscen naturo
Popul aKoh wasl

he gsCento Puudalt , tha pauy

In ngksalTale
wl4
fopul aHo

ab Stena y,th Pudat

iv Tn tha
Pop ul aHoh wola d ecline Q natra«al

904
ans fuudaj«Y
In Puue sesnt
Spa by an amon
dim' nlshed
do den sit an
o t h uedak i nanesd an am
to h PLy dnJity .
Can3eaoenhy. 4h e Hh u dalr 2aing fuy
de dih td h Py in h e
an in kuacHbn uak
t ) . ard an inleaHohJa
Puy popul okbn
4h Puad ar PopulaHd yty
guh(Sny)
BOn Sbemy pili
COntan
wth

P u y - P u edaHdn 2qakbn
obta h

gouHh, uak o
anY Quth
PuLdak duys
deatuale
Py an
an B ar)d
bature tu spease
. injeuauM
ha h e iniH
uh
aln
Q Knosn

CondihD
LotKe -Vakyo 2wah
ab ent
pusd al
Caje-1 w h o n

abSonk
wnOn

an pud ay
both Pu
'whon
a Se-1 model
PuARSant
d (d-Pa)
d

&
dt

Dividm

Rpone
d-Py) dy S)d
 boh 9idey
ahn n

e? - lo lo
la-a lo
loq -loq + lo
log y

Kis(on tont
whe
mtcguuaya

. clodtung

ä) is omily
The
Wan t=o
IniH a
ther K

Ean i
e

fuuom
K

an e Ah
har
Le

and
tuah mah®n
and
we

Y Y=«|
 4u npe-tive
t N t hr oulalion 27
bh feu{'0 italy
e t e t ims t. uois

1ft tivdy.

i
HHen
+Nontmm +,

- N (t ,NN
maabu7e thu aenah
h4h Uompidiion tbitierd
hefert c pei) n ipeies
hBhng4h ttho ekbert o
nd tURe

rmsedion nodd.

trantnm ths ym in 4
t 2ydem we t
maion fe fnm -n anu it
 1 klak

2tal 7
Dnaldiy ha thngad i d
Myw, e

Thun

,stem beom),

1-

1-t
7
hs dinnionlem m (atler maladng
mtaHlonal 2inpheH; ->shüinnait im
 T49:
0
Naluu A inpia, Thad
h igon 7"
(00)in untable. nOd;
Ho
hove nahix T«dures : 14479
i (0h 12
H
0) 2 (1-0-d,1 0b(-,)t-0

77,80
O1ging egn
valun ove A =-I
Th
Thus he ia eguilimiun
A 101

do,
above matir rdueey
EÅ (0t) h
- 0
-"
A= 1-4n t
U7Ufohding eiqun valun
aTe
Th rodi it
i 8ahle
02ial eqilibnium
l0

E,l7*h isepbian mahix n«diu

&1-2y. a,,
-
186 MATHEMATICAL BIOLOGY

3.5 Basic SIRS model

STRS (susceptible infected > recovered-> susceptible)
epidemic model arises when the immunity is not permanent but

temporary Here we extend the basic Kermack-McKendric SIR

epidemic model to incorporate such temporary immunity.
Consider that a disease spreads horizontally in a homogeneous
mixing group of N populations. Let S(t), I(t) and R(t) be,
respectively, the number of susceptible, infectives and recovered
is
populations at timet so that the total population at any time t
is assumed
gven by N)=S()+1()+R(0).Disease incidence
to follow the mass action lawIndividuals are removed from I class
and join R ciass at a rate proportional to the size of I. If immunity
is lost at a rate g, thena simpleSIRS model can be represented by
the following system of differential equations:

dS -SI +gR,
dt

= ASI - u,
dt

dk-uI-gR.
dt
(3.5.8)
where 2 is the dise ase transmission coetficient and is, the

recovery rate. The initial values

are S(0), I (0) and R(0)) A
is given in Fig. 3.5.
flow diagram of the model (3.5.8)

OO SIR model.
Figure 3.5: Flow diagram of
Here

R) =0>S()+I(4)+ R()= constant C(say) =

S+I+
dt
ie, N() S()+1(0)+R() =C.
=

Thus, the total population is constant and the population is said

are the initial susceptible
to be closed. As S(0), I(0) and R(0)
infected and recovered populations, we have

C-S(0)+1 (0)+ R(0) and

N=S()+(t)+ R(")=S (0)+I(0) +R(0).
Since the total population is constant, we can reduce the dimension

ofthe system by one and write R(t)=N-S(t)-1(t).

The system (3.5.8) then reduces to

= -ASI +g ( N - S - I ) = gN- ASI - gS- gl,

dt

=SI
dt
-ul. (3.5.9)

Observe that
dt
0 if AS> u» >1. Initially, number of
infected population is too small and therefore S» N. Under this

assumption, the previous inequality gives 1.It implies that

grow if R, >1, where R, is the basic

the epidemic can
reproduction number of the disease.
The equilibrium points of the system (3.5.9) are the
simultaneous solutions of
gN-AST - 8 S - gl = 0,

I(2S-u)=0. (3.5.10)
 MATHEMATICAL BIOLOGY
188

0 then S= N (see the

first equation). Thus, there exists
If I =

a disease-free equilibrium point E, {N,0).

=

S=. From the first equation of (3.5.10), we

If I 0 then
then have

I=-
+0
exists an endemic equilibrium E' ={S',I),
Therefore, there
where
6 N-
S and I=
+ò

Note that I will be positive if N->0 R,>1, where

9R is the basic reproduction number.

To determine the stability of the equilibrium points, we compute
the Jacobian matrix of the system at an arbitrary equilibrium
point (S,) as
J(5,1)=(2l+8) -(2S+s)
AS-4
At the disease-free equilibrium point E,, the variationai

matzix is
(N.0)=8 -(2N+s)
0 AN- )
The two eigenvalues are given by = -g and 52 = AN - .

Therefore, E, is locally asymptotically stable (node) if

AN-H<0, i.e., if 9, <1( Fig. 3.6 ). The disease-free

equilibrium is unstable (a saddle point) when R,>1, ie., when
the endemic equilibrium exists.
 189
EPIDEMIC MODELS
variational
At the endemicequilibrium point E' = (S",I), the
matrix is
-(AS"+8)
*8) -2s+e)-r+8
J(Sr)- AS-4 2
and T are
Since AS- 4 =
0 at the endemic equilibrium. As S"

positive, we note that traceJ(S,r)=-{21r +8) <0

0
and etJ(S',I) =Ar[2S" 8)> +

Thus, both eigenvalues of the matrix J(S",I'} will have negative

realparts and the endemic equilibrium will be locally asymptotically

stable whenever it exists, i.e., if R,>1(Fig. 3.6 ).

0.5

20 40 80
Time

D.5

20 40 50 60
Time

Figure 3.6: Time series evolutions of the

system (3.5.9) with
different initial values. Upper: Disease-free
for <1. equilibrium is stable
Lower: Endemic equilibrium is stable for R, >1.
1 Nicholson-Bailey model

In the 1930s Nicholson and Baileya proposed a simple discrete-time model for the population
dynamics of insect hosts and their parasitoids that has since become one of the classical models
of population biology. Parasitoids, such as parasitic wasps, lay eggs into their hosts and thus the
completion of the parasitoid life cycle requires that their hosts be killed. Parasitoids resemble
parasites as they grow inside a host, but also resemble predators in that they are obligate killers
of their host. The Nicholson-Bailey model is a two-dimensional system of difference equations
given by

Ht+1 = RHt e−aPt (1)

−aPt
Pt+1 = cHt (1 − e ) (2)

Here Ht and Pt represent the densities of the host and parasitoid population at year t. R is
the number of offspring of an unparasitized host surviving to the next year. Assuming random
encounter between hosts and parasitoids the probability that a host escapes parasitism can be
approximated by e−aPt , where a is a proportionality constant. Similarly, the probability to
become infected is then given by (1 − e−aPt ). Finally, the parameter c describes the number of
parasitoids that hatch from an infected host.

The equilibrium of the Nicholson-Bailey model is obtained by setting Ht+1 = Ht and

Pt+1 = Pt and is given by

log(R) R log(R)
P = and H= (3)
a R − 1 ac

However, this equilibrium is unstableb . The simulation of the Nicholson-Bailey model (see figure
1) shows that that the dynamics are characterized by oscillations of increasing amplitude until
the population crashes.

1.1 Tips to develop an R script for the simple Nicholson-Bailey model

Write a function that iterates the Nicholson-Bailey model. A simple example for a function that
iterates a difference equation would be
a
(i) Bailey, Victor Albert, born in Alexandria, Egypt, 18 December 1895; died in Geneva, Switzerland, 7
December 1964. Education Queen’s College Oxford (B.A. 1919; D.Philosophy 1923). Demonstrator in physics at
Oxford University 1924-25. Associate Professor of physics at Sydney University 1926-36; Professor of experimental
physics 1936-53; Research Professor 1952-60. Worked for the WHO at the end of his career. (ii) Nicholson,
Alexander John (England-Australia 1895-1969, population biologist, entomologist, and science administrator.
Worked on mimicry and animal population regulation. President of the Royal Society of Australia.
b
The stability of difference equations can be determined mathematically in a procedure similar to the math-
ematical stability analysis for differential equations as discussed in section 1.2.5 of the reader to “Ecology and
Evolution II: Populations”. As for differential equations one needs to determine the eigenvalues of the Jacobian
(which in discrete time is often called the “next-generation matrix”). In contrast to differential equations, where
the stability depends on whether the largest eigenvalue is larger or smaller than zero, for difference equations the
stability depends on whether the largest eigenvalue is larger or smaller than 1.

1
 ●
●

1e−01
●●●●●
●● ●
● ●●
●●
● ● ●
● ●
●●●●

1e−07
P

1e−13
1e−19
1e−25

1e−23 1e−17 1e−11 1e−05 1e+01

Figure 1: Phase diagram of the Nicholson-Bailey model. Starting from a value near the equi-
librium the trajectory spirals outwards until the population crashes (i.e. reaches extremely low
values, such that realistically all individuals in the population die). Thus the Nicholson-Bailey
model is an example for an unstable oscillation. Parameters used for the simulation are R = 2
R log(R)
and c = a = 1. The initial conditions were P0 = log(R)/a + 0.3 and H0 = R−1 ac + 0.3.

model <- function(time) {

param1 <- 2
param2 <- 1
population <- 1
cc <- data.frame(time=0,pop=population)
for (t in c(1:time)) {
population <- population*param1 + param2
cc <- rbind(cc,c(t,population))
}
cc
}

### Run the model

out <- model(10)
### Plot the model
plot(out$time,out$population,xlab="time",ylab="population size")

Modify the above model to obtain the Nicholson-Bailey model and then plot the time course
and phase diagram of the Nicholson-Bailey model. Try out different parameters.

2
2 Spatial Nicholson-Bailey model

Although host-parasitoid interactions are often characterized by very strong fluctuations from
year to year, unlike in the simple Nicholson-Bailey model, they typically do not lead to the
complete extinction of either the host or the parasitoid. One important factor neglected in the
Nicholson-Bailey model is space. Space can be neglected if the populations can be considered to
be well mixed. However, this may often not be justified. Both the interactions between species
and the dispersal of offspring may be local.

The Nicholson-Bailey model can be extended to incorporate space. To this end we consider a
spatial grid on which the Nicholson-Bailey dynamics take place. At each site (i, j) in the grid the
dynamics are (more or less) given by the simple Nicholson-Bailey model of the previous section,
but in addition we allow that hosts and parasitoids disperse to all immediately neighbouring
sites (with dispersal rates dh and dp ). Mathematically, the spatial model is thus given by the
following equations
?
?
Hi,j (t + 1) = RHi,j (t)e−aPi,j (t) (4)
? (t)
−aPi,j
?
Pi,j (t + 1) = cHi,j (t)(1 −e ) (5)

(where the time dependence is no longer indicated by a subscript but by brackets.) Here
X
?
Hi,j (t) = (1 − dh )Hi,j + dh /8 Hk,l (t) (6)
X
?
Pi,j (t) = (1 − dp )Pi,j + dp /8 Pk,l (t) (7)

where the sum is over all 8 neighbouring fields (i.e. (i − 1, j − 1), . . . , (i + 1, j + 1)).

Importantly, it can be shown (see figure 2) that in the spatial Nicholson-Bailey model hosts
and parasitoids can coexist indefinitely (in contrast to the non-spatial model).

2.1 Tips to develop the spatial extension of the Nicholson-Bailey model

• Define h and p as matrices in R. To create matrices look up

help(matrix)

in R. The matrix indices correspond to points on a two dimensional lattice. Hence hi,j
would be the density of hosts at lattice point (i, j).

• Multiplication of h∗p in R results in a new matrix, say m, which is given by mi,j = hi,j ∗pi,j .
Note that this is different from normal matrix multiplication, but is very convenient for
programming. Hence, you can update the entire playing field in a single line:

hn <- r* h * exp(-a p)

where h and p are matrices, a and r are numbers, and hn is the updated matrix of host
densities.

3
 1.0
0.8
0.6
0.4
0.2
0.0

Figure 2: Simulation of the spatial Nicholson-Bailey model. Depending on the dispersal rates of
hosts and parasitoids, dp and dh , parasitoids and hosts can coexist indefinitely. The plot shows
the logarithm of the host density (as levels of gray) on a quadratic grid. The graph corresponds
to a 100 × 100 lattice.

• Implement host and parasitoid dispersal. Use periodic boundary conditions such that
hosts or parasitoids that exit on one side of the quadratic ”playing field” enter again on
the opposite side. This is important because otherwise parasitoids or hosts are lost from
the field. Notice that one can conveniently change the order of indices in a matrix in R.
The command

h[c(2:10,1),]

for example shifts the rows 2 to 10 one up and places the first row last. The command

h[,c(2:10,1)]

does the same for columns. I recommend testing that dispersal equations (eqs 6 and 7)
work independently.

• Use R-functions image() or persp() to plot the matrices containing the host and para-
sitoid densities.

3 Basic exercises

Note: It is sufficient to solve one of the two basic exercises to qualify for a completed level 1
module.

4
Eb1. Why does the simple Nicholson-Bailey model show unstable oscillations? How could it be
stabilized? Try to implement realistic biological factors to achieve this. See the footnotec
for some possible stabilizing factors or experiment with your own ideas.

Eb2. How does the spatial dispersal rate of hosts and parasitoids affect the population dynamics
of the spatial model and why? Determine the probability of coexistence or extinction for
different combinations of the two dispersal rates and plot the results in a 2D plot of
the parameter space (i.e. in a coordinate system with axes corresponding to the two
parameters). In the regime of stable coexistence, plot the host and parasitoid densities as
a function of the two parameters. (You may also experiment with changing R, a or c).

4 Advanced/additional exercises

Note: Solving one or more advanced questions may qualify this module as level 2. Ask the course
instructor.

Ea1. Implement extinction in the models. Introduce a cut-off value below which the population
density is set to zero. This can be done in both the simple and the spatial model. Hints
for the latter: logical operations on matrices (e.g. h<cutoff) produce a matrix of logical
values; multiplication with logical values behaves as if TRUE=1 and FALSE=0.

Ea2. Vary lattice size and boundary conditions, and test their effect on the probability of coex-
istence. How is the stable size of the global population in the spatial model related to the
(unstable) equilibrium point in the non-spatial system?

Ea3. Given the instability of the system, initial conditions can greatly influence the outcome even
in the spatial model. Try out different initial conditions. Hint: use the runif() function
to generate random initial values for the population sizes in each cell. Alternatively, you
can start the system with non-zero populations in just one or a few cells (as in Hassel et
al ). Investigate Exercise Eb1 with both alternatives.

Ea4. Does the order of dispersal and reproduction matter?

Ea5. What would happen if either the hosts or the parasitoids spread equally over the entire
space?

Ea6. Introduce “long-lived” parasitoids (i.e. a fraction surviving even in the absence of hosts in
the lattice cell) and show how it affects the stability of the system.

Ea7. Introduce self-limitation in the host or other stabilizing factors and show how it affects the
spatial dynamics.
c
Some possible stabilizing factors: self-limitation in the host or the parasitoid (see the module on logistic
difference equation), constant immigration from the neighbourhood, sanctuary for a part of the host population,
etc.

5
Ea8. Measure the correlation between the dynamics at two sites as a function of their dis-
tance. How strong is the spatial correlation and how does that depend on the parameters
controlling parasitoid and host dispersal? A lattice can be regarded as an infinite two-
dimensional surface, if the correlation length is smaller than half of the side length of the
lattice. Check whether this criterion is fulfilled for the parameters that you use. Hint:
measuring all pairwise correlations between the cells of the lattice may require a lot of
computer time. Experiment with a small lattice first. (Note: Solving this exercise alone
qualifies this module as level 2. However, it is both difficult and computationally intensive.)

Ea9. Can parasitoids facilitate the coexistence of competing hosts? Introduce two host types
and either one parasitoid type affecting both hosts or two parasitoid types specialized to
the two hosts. Hint: to be able to simulate the system without parasitoids, introduce
logistic self-limitation in the hosts.