Chapter 3

Two species models

In this chapter we will study two species N1 and N2 occupying the same habitat and interacting, e.g.
competing for resources, predation, symbiosis, etc. The general form of the models we will consider is

dN1
= F1 (t, N1 , N2 )
dt
(3.1)
dN2
= F2 (t, N1 , N2 ).
dt

The model (3.1) is a coupled pair of ordinary differential equations, which given initial data N1 (0) −
N10 , N2 (0) = N20 and reasonable assumptions on the regularity of F1 and F2 have a unique solution for
all t ≥ 0.
In many cases, F1 , F2 will not depend explicitly on time, i.e. F1 = F1 (N1 , N2 ) and F2 = F2 (N1 , N2 ). Such
systems are called autonomous. Thus the vector field (F1 , F2 ) giving the velocity of a point (N1 , N2 ) is
fixed in time. To study such systems we can construct the phase plane, i.e. a picture of the solution
trajectories mapped out by points (N1 (t), N2 (t)) as t varies over (−∞, +∞). In particular we identify the
steady state populations:

Definition 3 The steady states of the system (3.1) are the solutions of dN1 /dt = 0 = dN2 /dt:

F1 (N1 , N2 ) = 0
(3.2)
F2 (N1 , N2 ) = 0.

The lines upon which trajectories are horizontal or vertical have special names:

Definition 4 We call F1 (N1 , N2 ) = 0 and F2 (N1 , N2 ) = 0 the nullclines of (3.1), i.e. they are the curves upon
which either dN1 /dt = 0 or dN2 /dt = 0.

The points where the nullclines (but not different branches of the same nullcline) cross are precisely the
steady states. Each phase plane sketch includes the nullcllines, the steady states, and a collection of

21
CHAPTER 3. TWO SPECIES MODELS

trajectories that start in varies parts of the plane. The individual trajectories are solutions of

dN1 F1 (N1 , N2 )
= , N1 (0) = N10 , N2 (0) = N20 . (3.3)
dN2 F2 (N1 , N2 )

In some cases the complete picture of the solutions of (3.1) can be established just by considering the
nullclines, steady states and how the sign of dN1 /dN2 changes as we go between regions demarked by
nullclines. In some cases, however, this level of detail is insufficient, and we must study more carefully
how (3.1) behaves near a steady state by considering its linearised form about that steady state.

1 Rules aiding construction of the phase plane

We list the following set of rules that help with the construction of the phase plane of the system (3.1):

1. trajectories cross vertically the nullcline

F1 (N1 , N2 ) = 0

since here dN1 /dt = 0;

2. trajectories cross horizontally the nullcline

F2 (N1 , N2 ) = 0

since here dN2 /dt = 0;

3. in regions enclosed by nullclines dN1 /dN2 has constant sign, i.e. trajectories are either rising or
falling;

4. trajectories can only go flat or vertical across nullclines;

5. steady states are where any branches of nullclines F1 (N1 , N2 ) = 0 and F2 (N1 , N2 ) = 0 cross.

In some cases further analysis (linear stability analysis of (3.1)) is required to further characterise the
detailed behaviour near a steady state, such as, for example, to distinguish between simple non-oscillatory
behaviour (e.g. tending directly to a steady state) or oscillatory behaviour (spiralling in to the steady
state).

Example: Construction of the phase plane I

Suppose we are given

dN1 /dt = F1 (N1 , N2 ) = N2

dN2 /dt = F2 (N1 , N2 ) = −N1 .

22
This is not a particularly realistic 2 species population model (it is, in fact, a model for a simple harmonic
oscillator), but it serves as a simple example of how the phase plane can be reconstructed. The nullclines
are
N2 = 0, N1 = 0,

i.e. the two axes of the plane. The nullclines cross at the unique steady state (0, 0). One may put a series
of arrows at various points in the plane that point in the direction of (F1 , F2 ) at those points. On N1 = 0
where Ṅ2 = 0 these arrows point horizontally to the left when N2 > 0 and horizontally to the right when
N2 < 0. Similarly on N2 = 0 where Ṅ1 = 0 the arrows point vertically upwards when N1 > 0 and vertically
downwards when N1 < 0. In the interior of the first quadrant Ṅ1 < 0, Ṅ2 > 0 so that the trajectory is moving
upwards to the left, and so on. Notice that there is not enough information to know whether the actual
trajectories spiral inwards, outwards or otherwise. However, for this simple example, the trajectories
satisfy
dN1 N2
=− ,
dN2 N1
which is a separable ode and can be integrated immediately to yield

N1 (t)2 + N2 (t)2 = constant = N10

2 2
+ N20 .

The solution trajectories are therefore all circles centred on the origin. The origin is neutrally stable, also
known as a centre.

Figure 3.1: The phase plane for dN1 /dt = −N2 , dN2 /dt = N1 . The inequalities in each quadrant are for
the interior of each quadrant, and indicate how N1 and N2 are changing along an orbit.
Note that without knowing that N12 + N22 is conserved, we would not know that the orbits
are circles - there is nothing in the sign of the velocities that guarantees that the orbits
are not spirals.

23
CHAPTER 3. TWO SPECIES MODELS

2 Behaviour on the boundary of the first quadrant

The kind of ode population models considered in this course are part of a larger class of systems called
Kolmogorov systems. Such systems take the form ẋi = xi fi (x1 , . . . , xn ) for i = 1, . . . , n where n is the number
of species and the smooth functions fi describe the per capita growth rate for the ith species. One of the
key properties of such systems is that if at some time t ∗ we have xi (t ∗ ) = 0 for i ∈ J (where J ⊂ {1, 2, . . . , n}
is some nonempty set) then xi (t) = 0 for all t and i ∈ J. In our planar models this means that trajectories
starting on the axes stay on the axes, and interior trajectories cannot reach the axes in finite time. Hence
to find what happens to a trajectory starting at x1 = 0 we simply solve

ẋ2 = x2 f2 (0, x2 ), x2 (0) given,

which is a ode in one variable, as for the single species models of the first 2 chapters. Hence drawing the
trajectories on each axes in the phase plane is a relatively simple task for planar Kolmogorov systems.

Example: The Lotka-Volterra competition equations

Recall the Logistic equation for a single speces:

! "
dN N
= ρN 1 − .
dt K

Here ρ is the linear birth rate, and K the carrying capacity. For two species N1 , N2 living in the same
habitat, but not interacting, we simply have
! "
dN1 N1
= ρ1 N1 1 −
dt K1
! "
dN2 N2
= ρ2 N2 1 − .
dt K2

The competition in these equations intraspecific (i.e. between the same species). When the species com-
pete with each other (for nesting sites, food, etc.), the interspecific competition is detrimental to both
specie’s per capita growth rates. The simplest model is to say that the per capital growth rates decrease
linearly with the density of the other species. The competition equations then become
! "
dN1 N1 c1
= ρ1 N1 1 − − N2
dt K1 ρ1
! " (3.4)
dN2 N2 c2
= ρ2 N2 1 − − N1 ,
dt K2 ρ2

where c1 , c2 > 0 measure the strength of the interspecific competition. To ease calculations, we first set
ui = Ni /Ki for i = 1, 2 and a12 = c1 K2 /ρ1 , a21 = c2 K1 /ρ2 . We also introduce a dimensionless time τ = ρ1t and

24
set ρ = ρ2 /ρ1 . This gives the simpler set of equations (fewer parameters)

du1
= u1 (1 − u1 − a12 u2 )
dτ
(3.5)
du2
= ρu2 (1 − u2 − a21 u1 ) .
dτ

Our first step is to locate the nullclines: These are

u1 = 0 and 1 − u1 − a12 u2 = 0 (3.6)

u2 = 0 and 1 − u2 − a21 u1 = 0. (3.7)

Hence steady states occur at points

! "
1 − a12 1 − a21
(u∗1 , u∗2 ) = (0, 0), (1, 0), (0, 1), P = , .
1 − a12 a21 1 − a12 a21

This last steady state is only feasible (non-negative populations!) when either1

1. a12 > 1 and a21 > 1, since then also 1 − a12 a21 < 0; OR

2. a12 < 1 and a21 < 1, since then also 1 − a12 a21 > 0;

Hence we have either 3 or 4 steady states. As we indicate in Figure 3.2 there are 4 cases to consider:

Case I a12 < 1 and a21 < 1;

Case II a12 > 1 and a21 > 1;

Case III a12 < 1 and a21 > 1;

Case IV a12 > 1 and a21 < 1.

Now let us determine what happens on the axes. Suppose first that initially u2 = 0, so that the evolution
is on the u1 axis. We find u1 (t) by solving the first equaton in (3.5) with u2 = 0:

u̇1 = u1 (1 − u1 ).

This is just the Logistic equation with ρ = 1, K = 1. Provided u1 (0) > 0 we have u1 (t) → 1 as t → ∞. Similarly
we find when u1 (0) = 0 then u̇2 = ρu2 (1 − u2 ) and hence u2 (t) → 1 as t → ∞ if u2 (0) > 0 (see Figure 3.3).
Let us consider the case I in detail (see figure 3.4). We have already dealt with the boundary behaviour.
Consider an interior trajectory A1. It starts (as drawn, at least) in region R1 that lies below both nullclines
1 − u1 − a12 u2 = 0 and 1 − u2 − a21 u1 = 0 so that here u̇1 > 0, u˙2 > 0 and the trajectory therefore advances in
the direction shown. This trajectory has positive gradient provided that it does not cross a nullcline. In
fact, all trajectories in R1 have positive gradient. Following A1 we see that it cannot turn back on itself,
and so must cross the nullcline where u̇1 = 0, whereby its gradient becomes negative since then, in R3 we
1 For simplicity here we do not consider the cases where a12 = 1 and/or a21 = 1.

25
CHAPTER 3. TWO SPECIES MODELS

N2 N2
steady state

1/a 1
12

1
1/a
12
P=(u1*, u2*) P=(N1*, N2*)

N1 N1
1 1/a 1/a 1
21 21

case I case II

N2 N2

1/a 1
12

1 1/a
12

N1 N1
1/a 1 1 1/a
21 21

case III case IV

Figure 3.2: The possible nullcline crossings for the Lokta-Volterra model (3.4)

have u̇1 < 0, u˙2 > 0. The trajectory thus goes vertical across u̇1 = 0 and continues upwards to the left. It
cannot leave R3, since to re-enter R1 is needs to cross it vertically and thus must go horizontal first, and
it cannot enter R2 since trajectories cross the boundary between R2 and R3 downwards. Hence A1 ends at
the interior steady state. A similar argument works for A2. In R3 the trajectories are above the nullclines
1 − u1 − a12 u2 = 0 and 1 − u2 − a21 u1 = 0 so that here u̇1 < 0, u˙2 < 0. A2 enters from R2 into R3 where it is
then trapped and must end at the interior steady state. After some practice, it is possible to draw the
trajectory directions by noting their directions on the nearby boundary. We can thus construct sketches
for the phase planes in each of these 4 cases:

Case I a12 < 1 and a21 < 1;

The steady state P attracts all interior trajectories. The remaining 3 steady states are unstable.

Case II a12 > 1 and a21 > 1;

The steady state P is unstable. The steady state (0, 0) is unstable, and both (1, 0) and (0, 1) are stable.
A separatrix splits the phase plane into two regions; above the seperatrix interior trajectories go to
the steady state (1, 0) and below they go to the steady state (0, 1)

26
 Figure 3.3: The trajectories on the boundary for the phase plane of the Lokta-Volterra model (3.4).
These trajectories stay on the axes and tend to the normalised carrying capacities if the
initial population is not at the origin.

Case III a12 < 1 and a21 > 1;

There is no steady state P. The steady states (0, 0) and (0, 1) are unstable, but (1, 0) is stable and
interior trajectories go to this steady state.

Case IV a12 > 1 and a21 < 1

There is no steady state P. The steady states (0, 0) and (1, 0) are unstable, but (0, 1) is stable and
interior trajectories go to this steady state.

Considering all these possibilities, we see that whatever the parameter values, the population always
tends (globally, i.e. whatever the initial populations are) to a finite steady state. In particular there can
be no population explosion or total extinction, nor oscillations.

Ecological considerations

In terms of the ecology, we understand the 4 cases as follows:

Case I a12 < 1 and a21 < 1;

If the interspecific competition is not too strong the two populations can coexist stably, but at lower
populations than their respective carrying capacities. Thus although the species may coexist, the
price that they pay for competing with each other is that they do not reach the population density
that they would have achieved (i.e. their carrying capacity) with the other species absent;

Case II a12 > 1 and a21 > 1;

Interspecific competition is aggressive and ultimately one population wins, while the other is driven
to extinction. The winner depends upon which has the starting advantage;

27
CHAPTER 3. TWO SPECIES MODELS

Figure 3.4: Building the phase plane for case I of the competition model (3.4). The (non-zero) null-
clines divide the first quadrant into 4 regions R1 − R4. In each region, the trajectories are
curves where the gradient has constant sign. The direction so trajectories can be deter-
mined by continuation from the boundaries.

Case III, IV a12 < 1 and a21 > 1 or a12 > 1 and a21 < 1 ;
Interspecific competition of one species dominates the other and, since the stable node in each case
is globally stable, the species with the strongest competition always drives the other to extinction.

3 Predator-Prey systems
In 1926 Volterra came up with a model to describe the evolution of predator and prey fish populations in
the Adriatic. Let N(t) denote the prey population and P(t) the predators. He assumed that

1. in the absence of predators the per capita prey growth rate was constant, but fell of linearly as a
function of predator population when predation was present;

2. in the absence of prey the per capita growth rate of the predator was contsant (and -ve), and in-
creased linearly with the prey population when prey was present.

Thus
1 dN
= a − bP
N dt
(3.8)
1 dP
= cN − d
P dt
where a, b, c, d > 0 are constants. It turns out that this model has an explicit integration. We find that

(−d + cN) dN (a − bP) dP

− = 0,
N dt P dt

or
d
{cN + bP − d log N − a log P} = 0.
dt

28
 Figure 3.5: The 4 topologically distinct phase planes for the Lokta-Volterra model (3.4)

If we set
H(N, P) = cN + bP − d log N − a log P,

then all trajectories (N(t), P(t)) evolve so that H(N(t), P(t)) = H(N(0), P(0)) = constant , i.e they are the
projections of the the level curves of H. Now we claim that H is a concave function. Note that

HNN = d/N 2 , HPP = a/P2 , HNP = HPN = 0,

2 = ad/(NP)2 > 0 and H

so that HNN HPP − HNP NN > 0, HPP > 0, and hence H is convex. It is strictly convex for
NP > 0. The minimum of H occurs where ∇H = 0, i.e. where
! "
d a d a
c− = 0 = b − ⇒ (N, P) = , .
N P c b

Notice that this corresponds to the unique steady state of the system (3.8). Since H is strictly convex with
a unique minimum in the positive quadrant, every trajectory must be a closed curve. Thus the orbits are
a one-parameter (the value of H) set of closed curves starting at the steady state.
The Lokta-Volterra equations are actually canonically Hamiltonian (in appropriate coordinates) and so
are not structurally stable. This means that a small change in the model will not typically lead to another
Hamiltonian system, nor one where the phase plane is topologically equivalent. In general we will end up

29
CHAPTER 3. TWO SPECIES MODELS

Figure 3.6: Orbits for the classic predator-prey model. The orbits are the level sets of H, i.e. each
orbit is a projection of the intersection of a plane H = H0 with the surface H = H(N, P). NB:
This figure, and many of the figures in the notes were produced using the CurvesGraph-
ics Mathematica package by Gianluca Gorni, Department of Mathematics and Computer
Science, University of Udine, Italy.

with a dissipative system. For a biological system life is the constant struggle against increasing entropy
where irreversible processes play an important role. In this light, a Hamiltonian system cannot model a
‘living’ system. There are several points of criticism worth noting for the Volterra-Lokta model:

1. There is no possibility of either population being driven to extinction;

2. Changing the birth and death rates does nothing but change the period of the oscillation - i.e. noone
can dominate;

3. For certain ecological conditions (fitness of species, etc.) one would expect one species to win regard-
less of initial conditions, but this does not happen.

We will now take a digression into Linear stability analysis for pairs of autonomous ODEs.

30
 3.1. DIGRESSION: LINEAR STABILITY ANALYSIS OF PLANAR ODES

0 1 2 3 4

Figure 3.7: Nested periodic orbits for the classic predator-prey model. Boundary trajectories are not
shown. All interior trajectories a periodic and enclose the unique interior steady state at
(1, 1).

3.1 Digression: Linear stability analysis of planar odes

In many of the models we meet in the course, we will need to study the behaviour of trajectories close to
steady states for the system
dx
= f (x, y)
dt
(3.9)
dy
= g(x, y).
dt
We assume that P = (a, b) is a steady state of (3.9). Thus

f (a, b) = 0 = g(a, b). (3.10)

Consider the evolution of a trajectory close to P, and write x(t) = a+X(t) and y(t) = b+Y (t). We may expand
in (3.9) using the Taylor expansion:

dX
= f (a, b) + fx (a, b)X + fy (a, b)Y + · · ·
dt

dY
= g(a, b) + gx (a, b)X + gy (a, b)Y + · · · .
dt

31
CHAPTER 3. TWO SPECIES MODELS

Using (3.10), this becomes

dX
= fx (a, b)X + fy (a, b)Y + · · ·
dt

dY
= gx (a, b)X + gy (a, b)Y + · · · .
dt

Close to P, |X(t)|, |Y (t)| ( 1 this system is well-approximated by the linearised version obtained by neglect-
ing second order terms in X,Y :
dX
= fx (a, b)X + fy (a, b)Y
dt
(3.11)
dY
= gx (a, b)X + gy (a, b)Y.
dt
Notice that since we are neglecting higher order than linear terms, the linear approximation will only
potentially give a good indication of the full nonlinear system while X(t),Y (t) remain small.
Now let X(t) = (X(t),Y (t))T and # $
fx (a, b) fy (a, b)
M= .
gx (a, b) gy (a, b)

Then (6.12) can be rewritten in matrix form:

dX(t)
= MX(t). (3.12)
dt

This has steady state (0, 0) (which corresponds to (x, y) = (a, b)). For a trajectory of the linearised system
(3.12) starting at X(0) = X0 ,
X(t) = exp(Mt)X0 .

If M has two distinct eigenvalues λ1 , λ2 with corresponding eigenvectors v1 , v2 then

X(t) = αeλ1 t v1 + βeλ2 t v2 , (3.13)

where α, β are defined by the decomposition X0 = αv1 + βv2 (using linear independence of v1 , v2 ). When the
two eigenvalues are equal (and therefore real) we have

X(t) = eλt (X0 + ctv) (3.14)

for some real c (which may be zero).

To find the (local) stability of the steady state (a, b) we examine the dynamics of (3.12) which has
solution (3.13). If (a, b) is stable then for small X0 , the solution X(t) will eventually decay to the origin
(0, 0) and this happens, according to (3.13), when both eigenvalues have negative real parts.
Let us list the various possibilities for behaviour near a steady state
1. λ1 )= λ2 ∈ R

(a) λ1 < λ2 < 0 - stable node.

32
 3.1. DIGRESSION: LINEAR STABILITY ANALYSIS OF PLANAR ODES

0 0

0 0 0 0

0 0

Figure 3.8: Linear stability for real eigenvalues: (i) λ1 , λ2 < 0 (stable node) and (ii) λ1 λ2 < 0 (saddle).
The thick (red) lines are in the direction of the eigenvectors.

(b) λ1 > λ2 > 0 - unstable node.

(c) λ1 λ2 < 0 - saddle (unstable)

(The reader may wish to ask themselves what happens when λ1 λ2 = 0)

2. Complex eigenvalues
Then λ1 = µ + iw, λ2 = µ − iw and

X(t) = ℜ{eµt (αv1 eiwt + βv2 e−iwt )}

Hence we have spirals. If µ < 0 the trajectories go to the steady state (stable spiral), for µ > 0 they
leave (unstable spiral). When µ = 0 the linearised system gives concentric ellipses (but caution: we
need to consider the nonlinear terms to get true picture as this linear system is not structurally
stable).

3. λ1 = λ2 Now the Jordan form of M is # $

a 0
.
b a

There are two cases:

(a) b = 0
Then we have what is known as a ‘star’ - all trajectories approach the origin along straight lines
(all lines through origin are eigenvectors) when a < 0, and leave origin when a > 0.
(b) b )= 0

33
CHAPTER 3. TWO SPECIES MODELS

0 0

0 0 0 0

0 0

Figure 3.9: Linear stability for complex eigenvalues: (i) λ = µ ± iw, µ > 0 (unstable spiral) and (ii) λ =
±iw (centre).

Then there is only one linearly independent eignevector and we have behaviour as shown in
Figure 4.12).

The following lemma is very useful when determining the nature of eigenvalues. Recall that the trace of
a matrix is the sum of its diagonal elements.

Lemma 1 Let M be a 2 × 2 matrix with eigenvalues λ1 , λ2 . Then

λ1 + λ2 = Trace M, (3.15)
λ1 λ2 = det M. (3.16)
# $
a b
Proof: Let M = . Then the characteristic equation for the eigenvalues λ reads det(M − λI) = 0
c d
which becomes
λ2 − (a + d)λ + (ad − bc) = 0,

or equivalently
λ2 − Trace Mλ + det M = 0.

On the other hand, if λ2 − Trace Mλ + det M has roots λ1 , λ2 then λ2 − Trace Mλ + det M = (λ − λ1 )(λ − λ2 ).
Multiplying out the last product and comparing coefficients in λ gives the desired formulae.

Corollary 1 Let S = (a, b) be a steady state of some planar ode and M the 2 × 2 real stability matrix at
(a, b).

1. If det M < 0 then M has eigenvalues of opposite sign and S is a saddle;

34
 3.1. DIGRESSION: LINEAR STABILITY ANALYSIS OF PLANAR ODES

0 0

0 0 0 0

0 0
! "
a 0
Figure 3.10: Linear stability for equal eigenvalues: (i) Jordan form and (ii) Jordan form
0 a
! "
a 0
(b )= 0).
b a

2. If det M > 0 and Trace M < 0 then the real parts of the eigenvalues of M are negative and S is locally
stable.

3. If det M > 0 and Trace M > 0 then the real parts of the eigenvalues of M are positive and S is unstable.

4. If det M > 0 and (Trace M)2 ≥ 4 det M then S is a node and if (Trace M)2 < 4 det M then S is a spiral.

1 Linear stability analysis of the LV competition model

We will now demonstrate how the above linear stability analysis works with the competition model:

du1
= f (u1 , u2 ) = u1 (1 − u1 − a12 u2 )
dτ
(3.17)
du2
= g(u1 , u2 ) = ρu2 (1 − u2 − a21 u1 )
dτ

The first thing we have to do is compute the matrix

# $
fu1 fu2
M= .
gu1 gu2

We find that at the point (u1 , u2 ) that

# $
1 − u1 − a12 u2 − u1 −a12 u1
M(u1 ,u2 ) = . (3.18)
−ρa21 u2 ρ(1 − u2 − a21 u1 ) − ρu2

35
CHAPTER 3. TWO SPECIES MODELS

There are always the 3 steady states (0, 0), (1, 0) and (0, 1). There may be a fourth and interior steady
state.

1. (u1 , u2 ) = (0, 0). Here # $

1 0
M(0,0) = .
0 ρ

Since the eigenvalues of a triangular matrix are its diagonal elements, we see that the eigenvalues
of the linear stability matrix at the origin are 1, ρ. Since these are both positive we conclude that
(0, 0) is an unstable node.

2. (u1 , u2 ) = (1, 0). Here # $

−1 −a12
M(1,0) = .
0 ρ(1 − a21 )

Thus the eigenvalues are −1, ρ(1 − a21 ) and hence (1, 0) is a stable node if a21 > 1 and a saddle if
a21 < 1.

3. (u1 , u2 ) = (0, 1). Here # $

1 − a12 0
M(0,1) = .
−ρa21 −ρ

Thus the eigenvalues are 1 − a21 , −ρ and hence (0, 1) is a stable node if a12 > 1 and a saddle if a12 < 1.

Finally, when the interior steady state (u∗1 , u∗2 ) exists, so that a12 , a21 > 1 or a12 , a21 < 1, we obtain
# $
(1 − u∗1 − a12 u∗2 ) − u∗1 −a12 u∗1
M(u∗1 ,u∗2 ) = .
−ρa21 u∗2 ρ(1 − u2 − a21 u∗1 ) − ρu∗2
∗

Now since (u∗1 , u∗2 ) is an interior steady state 1 − u∗1 − a12 u∗2 = 0 = 1 − u∗2 − a21 u∗1 and hence the bracketed
expressions in the last matrix vanish and we have
# $
−u∗1 −a12 u∗1
M(u∗1 ,u∗2 ) = . (3.19)
−ρa21 u∗2 −ρu∗2

Notice that we left M in the form (3.18) in order to obtain the simple form of the stability matrix at the
the interior steady state in (3.19). In order to determine the nature of the eigenvalues of M(u∗1 ,u∗2 ) we use
Corollary 1. We see that Trace M = −u∗1 − ρu∗2 < 0 and det M = ρu∗1 u∗2 (1 − a12 a21 ). Hence if a12 < 1, a21 < 1 then
S is locally stable (we do not bother to distinguish between a focus and a spiral), and if a12 > 1, a21 > 1 then
S is a saddle.
These calculations can be checked by referring back to the phase plane plots in Figure 3.5.

36
 3.1. DIGRESSION: LINEAR STABILITY ANALYSIS OF PLANAR ODES

Figure 3.11: The possible nullcline crossings for the predator-prey Lokta-Volterra model with in-
traspecific competition (Equations (3.20), (3.21)).

Example: Predator-Prey with intraspecific competition

Now we consider the classic Predator-Prey model with additional interactions, namely intraspecific com-
petition.

Ṅ = N(a − eN − bP) = F(N, P) (3.20)

Ṗ = P(−d − f P + cN) = G(N, P). (3.21)

Thus we have the original predator-prey model with the intraspecific competition terms −eN 2 and − f P2
added. Recall that in the original model e = 0, f = 0 and we obtain a continuum of periodic orbits around
the interior steady state. What happens when e > 0, f > 0?
Let us consider the nullclines. These are solutions to:

Ṅ = 0 : N = 0 or a − eN − bP = 0,

Ṗ = 0 : P = 0 or − d + cN − f P = 0.

The two possible cases are plotted in Figure 3.11. In case 1, there are only two steady states (0, 0) and
(a/e, 0), but when ac > de there is a third non-zero steady state, say (N ∗ , P∗ ).
On the boundary P = 0 we have Ṅ = N(a − eN) so the trajectories tend to N = a/e if N(0) )= 0. On N = 0,
we have Ṗ = −P(d + f P) ≤ 0 with equality if and only if P = 0, so that trajectories tend to P = 0.
In case 1, it is possible to fill in the trajectories to complete the phase space plot. However, in case 2,
while it is clear that the trajectories near the non-zero steady state cycle around the steady state, it is not

37
CHAPTER 3. TWO SPECIES MODELS

obvious whether these trajectories are closed paths or spirals (or something else?). To complete the phase

Figure 3.12: The possible phase space plots for the predator-prey Lokta-Volterra model with intraspe-
cific competition (Equations (3.20), (3.21)).

plots we need to determine the correct behaviour of the trajectories near to steady states, i.e. perform the
linear stability analysis.

For the stability matrix we obtain

# $ # $
FN FP (a − eN − bP) − eN −bN
M= = .
GN GP cP (−d − f P + cN) − f P

Hence at (0, 0) we have # $

a 0
M(0,0) = ,
0 −d

so that the eigenvalues a, −d are of opposite sign showing that the origin is a saddle. At (a/e, 0) we have
# $
−a −ba/e
M(a/e,0) = ,
0 −d + ca/e

The eigenvalues of M(a/e,0) are thus −a < 0 and −d + ca/e. In the case ca < de, so that there is no interior
steady state, (a/e, 0) is a stable node, whereas when ca > de, so that the interior steady state exists, (a/e, 0)
is a saddle.

38
 3.1. DIGRESSION: LINEAR STABILITY ANALYSIS OF PLANAR ODES

Finally when ca > de we consider the linear stability of (N ∗ , P∗ ). We have

# $
−eN ∗ −bN ∗
M(N ∗ ,P∗ ) = ,
cP∗ − f P∗

Trace M(N ∗ ,P∗ ) = −eN ∗ − f P∗ < 0 and det M(N ∗ ,P∗ ) = N ∗ P∗ (e f + bc) > 0, so that the eigenvalues of M(N ∗ ,P∗ ) have
negative real parts. Hence when it exists (N ∗ , P∗ ) is locally stable.

1.4 2

1.2
1.5
1

0.8
1
0.6

0.4
0.5
0.2

0.5 1 1.5 2 0.5 1 1.5 2

Figure 3.13: Phase space plots for the predator-prey Lokta-Volterra model with intraspecific competi-
tion (Equations (3.20), (3.21)). Left ca < de where there are just two steady states, right
ca > de where the interior steady state now exists and is locally stable with a spiral.

39