POPULATION DYNAMICS: TWO SPECIES MODELS; If they co-exist in the same environment:

Susceptible–Infected–Recovered (SIR) MODEL

rate of change of u = + growth − effect of predator-prey
Next logical step: consider dynamics of more than one encounters
species. We start with models of 2 interacting species. We
rate of change of v = − decay + effect of predator-prey
consider, so-called, “box models” where species are
assumed to be “well mixed”. Models that include spatial encounters
movement of species are no discussed in this course. How to describe the effect of encounters? Law of Mass
(Ordinary vs. partial diff. equations.) Action!
1. Lotka-Volterra predator-prey model: “heuristic”
derivation. Consider 2 species, prey u, and predator v.
Population of prey without predator grows (a > 0 is a
const.):
du
= au; u(0) = u0 ;
dt
population of predator without prey decays (b > 0 is a
According to Law of Mass Action the probability of
const.):
encounters of 2 species is proportional to the product of
dv population densities of these species. Proportionality
= −bv; v(0) = v0 .
dt coefficients depend of various factors. Thus, we arrive at
24 25
the system of equations (a, b, n, m > 0 are const.):
rate of change of u = + growth − competition between u
du
= au − nuv,
dt − competition between u and v
dv rate of change of v = + growth − competition between v
= −bv + muv;
dt
− competition between u and v
u(0) = v(0) =
u0 , v0 .
Thus, using once again the Law of Mass Action, we write:
Possible model modifications: logistic growth for prey du
(instead of exponential), etc. = k1 u − α1 u2 − β1 uv,
dt
dv
2. Competition model: derivation. “Heuristic” = k2 v − α2 v 2 − β2 uv;
dt
derivation: consider two species that consume the same
u(0) = u0 , v(0) = v0 .
resource. Assume that each species population in the
absence of the other is described by the logistic equation
(here u and v are population densities of the two species): Another way to derive: Consider a well stirred batch
reactor. Let u and v be the population densities of two
du types of bacteria, and c be the food concentration. The
= k1 u − α1 u2 ,
dt same food is consumed by both types of bacteria. Assume
dv that growth rate coefficient for each bacteria type is a
= k2 v − α 2 v 2 .
dt linear function of c: Ki = Ki (c) = κi c (i = 1, 2).
When the other species is present: Then we have a system of equations
26 27
 Conservation of mass! From the above we express
du
= κ1 cu,
dt c(t) = a1 u0 + a2 v0 + c0 − a1 u(t) − a2 v(t) = A − a1 u − a2 v,
dv
= κ2 cv, and substitute this expression in the original equations for
dt
dc u and v to obtain the system:
= −a1 κ1 cu − a2 κ2 cv,
dt du
where 1/ai (i = 1, 2) are, so-called, yield factors. Initial = κ1 cu = κ1 u(A − a1 u − a2 v)
dt
conditions are
= k1 u − α1 u2 − β1 uv;
u(0) = u0 , v(0) = v0 , c(0) = c0 .
dv
= κ2 cv = κ2 v(A − a1 u − a2 v)
The above system can be reduced to two equations as dt
follows. Let us multiply the first equation by a1 , the second = k2 v − α2 v 2 − β2 uv;
— by a2 , and add the three equations. We obtain, where ki = κi A, α1 = κ1 a1 , α2 = κ2 a2 , β1 = κ1 a2 , and
du dv dc d(a1 u + a2 v + c) β2 = κ2 a1 .
a1 + a2 + = = 0.
dt dt dt dt In the future, for the analysis, we will write this system in
Integrating this equation, we get for any t: yet another form.
a1 u(t) + a2 v(t) + c(t) = const, 3. SIR model. It is common to start with a schematic
representation:
and from the initial conditions:
a1 u(t) + a2 v(t) + c(t) = a1 u0 + a2 v0 + c0 .
28 29
 S(0) = S0 , I(0) = I0 , R(0) = 0 without immunization.
S I R
Remark 1. It can be easily checked that in this system the
total population is conserved: at any instant of time
Which processes affect the rates of change of respective
S(t) + I(t) + R(t) = S0 + I0 = Nt ot al = const.
populations?
What are the assumptions? Remark 2. If deaths are included in the model, we get the
system:
Law of Mass Action! dS
= −αIS − δ1 S,
dt
For “box models” it does not matter whether we use the
dI
population densities or actual populations: numerical = +αIS − β I − δ2 I,
dt
values of coefficients will be different but qualitative dR
behavior is going to be the same! Let us use notations S, I, = +β I − δ1 R,
dt
and R for susceptible, infected, and recovered. Then, S(0) = S0 , I(0) = I0 , R(0) = 0.
dS We note that the SIR system is, in fact, a combination of a
= −αIS,
dt closed system of 2 equations for S and I (this system can
dI be solved independently of R) and the differential relation
= +αIS − β I,
dt for R (i.e., when I is known, R is obtained by simple
dR integration). So, we actually have to analyze the system of
= +β I,
dt 2 equations:
30 31
 dR
dS = +β I − γR,
= −αIS, dt
dt S(0) = S0 , I(0) = I0 , R(0) = R0 in general case.
dI
= +αIS − β I. Using the fact that in this model (without deaths) the total
dt
population, N , is once again conserved, we can reduce the
above 3-dimensional system to a system of 2 equations.
If recovered can become susceptible again we arrive at the,
We express R in terms of S and I as follows. For
so-called, SIRS model.
N = S0 + I0 + R0 = const known, we have:
4. SIRS model. Schematic representation:
R(t) = N − I(t) − S(t).
Substituting this into equations for S and I we, finally,
obtain:
dS
S I R = −αIS + γ(N − I − S),
dt
dI
= +αIS − β I,
dt
Corresponding model system will now be in the form:
dS S(0) = S0 , I(0) = I0 .
= −αIS + γR,
dt
dI 4. General systems of two nonlinear differential
= +αIS − β I,
dt equations. It can be easily seen that all the models
32 33
introduced above can be written in the following general somehow known: u = u(t), v = v(t). Consider a
form: (u, v)–plane, which we call a phase plane. Then the point
du with coordinates (u(t), v(t)) (where time t is changing)
= f (u, v),
dt will trace a curve on this plane.
dv
= g(u, v),
dt
u(0) = u0 , v(0) = v0 known.
Functions f (u, v) and g(u, v) describe the “rules” of
species interactions and behavior.
Our goal is to describe possible types of solutions: we want
to know when the populations will grow to certain values,
go extinct, oscillate, etc., and how will these types of
solutions depend on numerical values of model
parameters?
Let us extend the approach that worked previously for
scalar (single) nonlinear differential equations to systems
of two (and later, to systems of three and more)
differential equations.
5. Model solutions and phase plane. Geometry of the
model system. Assume that the solution of a system is
34 35
If the solution goes to a steady state (i.e., as t → ∞,
(u(t), v(t)) → (ū, v̄)) then the point on the phase plane
corresponding to a solution will eventually stop.
If the solution oscillates (i.e., the values of (u(t), v(t))
periodically repeat themselves with certain period T ) then
the point on the phase plane will trace a closed loop.
The velocity vector w  related to motion of a point
(u(t), v(t)) on the plane (this vector is tangential to the, Sometimes it is convenient to show only the direction of
so-called, trajectory of the moving point, it shows the motion at every point of the phase plane and not how fast
direction of point’s motion and its length shows how fast the solution changes along the phase trajectory. Then
the point is moving) is defined as follows: instead of velocity vector field one may show the,
  so-called, direction field: all the vectors associated with
du d v
 =
w , = ( f (u, v), g(u, v)). different locations in the phase plane will have the same
dt dt
length and will possibly differ only by direction. To
The above expression means that the system of differential normalize the vectors with coordinates ( f (u, v), g(u, v))
equations specifies the, so-called, vector field on the phase (i.e., to make them all be of the same length L) the
plane: with every point (u, v) we associate a vector following formula may be used: the new “normalized”
( f (u, v), g(u, v)). These vectors show the direction and vectors will have coordinates
speed of motion of a solution point, that is currently
located at position (u, v), as time increases. (F (u, v), G(u, v))
36 37
 ⎛ ⎞
L f (u, v) L g(u, v) The prescribed initial condition hits one of the trajectories
= ⎝ , ⎠.
on the phase plane and follows it as time increases. For us
f 2 (u, v) + g 2 (u, v) f 2 (u, v) + g 2 (u, v)
it is important to know where it will go. If the initial
condition corresponds to a steady state, then the solution
will stay at this steady state forever. It turns out that if the
initial condition is not at a steady state, only a few
possibilities may occur on the phase plane: (a) solution
tends to a stable steady state, (b) moves away (to infinity),
(c) belongs to a limit cycle, (d) tends to a limit cycle, (e)
moves away from a limit cycle, (f) what else?
Same as in the case of the scalar equation, steady states
(ū, v̄) are the constant solutions that satisfy the following
system of equations (since d ū/d t = 0 and d v̄/d t = 0):
0 = f (ū, v̄), 0 = g(ū, v̄).
In terms of behavior on the phase plane we have that if the
the point corresponds to a steady state, it will not move
since the velocity vector at this point has zero entries: Important! No chaos for continuous systems of 2
( f (ū, v̄), g(ū, v̄)) = (0, 0) (and thus, no direction of motion autonomous differential equations, i.e., no chaos on the
is defined, and the speed of motion is zero). phase plane! We need 3 equations to produce chaos.
38 39
Our goal will be to characterize possible types of solution
behavior in the vicinity of the steady states, which will lead
to identification of several distinct types of steady states
(also called equilibrium points). Then, using some
additional information on general behavior of phase
trajectories away from steady states (e.g., the fact that
trajectories can only intersect at steady states), we will be
able to qualitatively characterize the global behavior of
solutions (not only near the steady states).
In what follows we will extensively use the idea of a
6. Linearization procedure for systems of two
null-cline. The curves in the phase plane whose (u, v)
nonlinear differential equations. Given a system of
coordinates satisfy the equation f (u, v) = 0 are called u
equations
null-clines. Special feature: solution trajectories intersect
these null-clines vertically (since on these curves the du
= f (u, v),
u-component of velocity vectors is zero). Similarly, curves dt
in the phase plane whose (u, v) coordinates satisfy the dv
= g(u, v),
equation g(u, v) = 0 are called v null-clines. Their special dt
feature: solution trajectories intersect these null-clines and a steady state (ū, v̄) satisfying
horizontally (since on these curves the v-component of
velocity vectors is zero). Evidently, the steady states 0 = f (ū, v̄), 0 = g(ū, v̄).
correspond to points of intersection of u null-clines and v
null-clines. Let us perturb this steady state, i.e., we consider initial
40 41
conditions
dβ
= gu (ū, v̄)α + g v (ū, v̄)β,
u(0) = ū + α(0), v(0) = v̄ + β(0) dt
and α = 0, β = 0 correspond to the steady state solution of
with |α(0)|, |β(0)|  1. We want to find out what happens
the above linear system.
to α(t), β(t) at t increases.
=⇒ Importance of linear equations! To find out how
We substitute
perturbations α(t) and β(t) behave, we first need to
u(t) = ū + α(t), v(t) = v̄ + β(t), discuss solutions of systems of 2 linear constant coefficient
differential equations.
into the system to obtain:
d(ū + α) dα 7. Linear systems of two equations with constant
= = f (ū + α, v̄ + β)
dt dt coefficients and equivalent second order equation. Let
= f (ū, v̄) + fu (ū, v̄)α + f v (ū, v̄)β + ...; us write down such system in a special form (here we
assume that x and y are the dependent variables):
d(v̄ + β) dβ
= = g(ū + α, v̄ + β) dx
dt dt = a11 x + a12 y,
dt
= g(ū, v̄) + gu (ū, v̄)α + g v (ū, v̄)β + . . . .
Here fu , gu are the derivatives of f and g with respect to dy
= a21 x + a22 y.
u, f v , g v are the derivatives of f and g with respect to v. dt
Thus, approximately, we obtain Evidently, the point in the (x, y)-phase plane with
coordinates x = 0, y = 0 is an equilibrium solution of this
dα
= fu (ū, v̄)α + f v (ū, v̄)β; system.
dt
42 43
Constant coefficients ai j may be written as a matrix (called differentiating the first equation of the system with respect
Jacobian matrix): to t:
⎛ ⎞
d2 x dx dy
⎜ a11 a12 ⎟ = a11 + a12 .
A=⎝ ⎠. d t2 dt dt
a21 a22
Then we substitute here d y/d t from the second equation:
For the future classification of steady states the following d2 x dx
= a11 + a12 (a21 x + a22 y),
quantities are important: d t2 dt
and finally, we eliminate y using the first equation of the
trace of matrix A: trA = a11 + a22 ; original system: y = (d x/d t − a11 x)/a12 .
Collecting the terms, we obtain the following equation for
determinant of matrix A: det A = a11 a22 − a12 a21 . x:
There are several approaches to finding solutions x(t), d2 x dx
− (a11 + a22 ) + (a11 a22 − a12 a21 )x = 0.
y(t). Here, for simplicity, we use the fact that every system d t2 dt
of 2 first order equations (i.e., system of 2 equations We note that this equation may be re-written as:
containing only first order derivatives) is equivalent to
some scalar second order equation (i.e., one differential d2 x dx
− (trA) + (det A) x = 0.
equation that contains both first and second order d t2 dt
derivatives). The initial conditions for the original system are
Let us derive an equivalent second order equation, e.g., for x(0) = x 0 , y(0) = y0 . The second order equation requires
x (equation for y will be the same). We begin with 2 conditions for x. Substituting x 0 and y0 into the first
44 45
equation of the original system we see that the condition
for y may be converted into the second condition for x:
dx
(0) = a11 x 0 + a12 y0 known.
dt
If x(t) is known (as a solution of the second order
equation with corresponding 2 initial conditions), then
y(t) is obtained from
d x/d t − a11 x
y= .
a12
The original system and the above second order equation
are equivalent in the following sense: if x is the solution of
one, then it is the solution of the other (the same is true
for y).
8. Possible solutions: characterization of steady states.
We start with the observation that x = 0 (and thus, y = 0)
is one of the solutions of the second order equation (which
was expected since (0, 0) is the steady state of the original
system). Assume that the initial conditions are chosen
close to (0, 0): x(0) = x 0 , y(0) = y0 , with x 0 and y0 not
46
equal to zero simultaneously. What will happen to x(t)
and y(t)?
Based on our previous experience with scalar first order
equations, we seek solution of the second order equation
in the form:
x = eλt ,
where λ is yet unknown. Substituting this into the
equation, we obtain:
λ2 eλt − (trA)λeλt + (det A)eλt = 0.
Canceling eλt = 0, we arrive at the
CHARACTERISTIC EQUATION for λ:
λ2 − (trA)λ + det A = 0.
This is a quadratic equation, and thus, it has 2 solutions.
So, as a result, we will have not one x(t), but two: x 1 (t)
and x 2 (t). Since the second order differential equation is
linear, the linear combination of these two solutions,
C1 · x 1 (t) + C2 · x 2 (t), where C1 and C2 are arbitrary
47
constants of integration, is also a solution. Thus, the (a) For det A > 0, and tr A > 0: λ1 > 0, λ2 > 0, and the
so-called general solution of the linear second order steady state (0, 0) is unstable.
equation can be written as follows:
x(t) = C1 · x 1 (t) + C2 · x 2 (t).
Two constants of integration are defined by 2 known initial
conditions x(0) and d x/d t(0).
Let us find λ1,2 and corresponding x 1,2 for different
possible cases. From equation for λ, we obtain: Steady state of this type is called an unstable node.

 2 (b) For det A > 0, and tr A < 0: λ1 < 0, λ2 < 0, and the
trA trA
λ1,2 = ± − det A. steady state (0, 0) is (asymptotically) stable.
2 2
Case 1: If
 2
trA
− det A > 0,
2
we have 2 distinct real λ’s; x 1 (t) = exp(λ1 t),
x 2 (t) = exp(λ2 t), and Steady state of this type is called a stable node.
x(t) = C1 · eλ1 t + C2 · eλ2 t . (c) For det A < 0, and tr A either positive or negative:
48 49
λ1 < 0, λ2 > 0, and and the steady state (0, 0) is unstable.
Figures are similar to those presented earlier for stable and
unstable node.
Case 3: If
 2
trA
− det A < 0,
2
Steady state of this type is called a saddle.

we have 2 distinct complex λ’s: λ1,2 = σ ± iδ, i = −1,
Case 2: If where 
 2
 2 trA trA
trA σ= , δ= det A − ;
− det A = 0, 2 2
2
and x 1 (t) = exp(σt) sin(δt), x 2 (t) = exp(σt) cos(δt). So,
we have 2 identical real λ’s: λ = trA/2; x 1 (t) = exp(λt),
x 2 (t) = t exp(λt), and x(t) = C1 exp(σt) sin(δt) + C2 exp(σt) cos(δt).
x(t) = C1 · eλt + C2 · t · eλt .
(a) For tr A = 0 we have pure oscillations: σ = 0, and

x(t) = C1 sin(δt) + C2 cos(δt), where δ = det A. The
(a) For tr A > 0 =⇒ λ > 0: unstable node.
steady state (0, 0) is stable (but not asymptotically stable).
(b) For tr A < 0 =⇒ λ < 0: stable node.
50 51
Steady state of this type is called a center.
(b) For tr A > 0 we have oscillations with growing
amplitude. The steady state (0, 0) is unstable.
Steady state of this type is called an unstable focus.
(c) For tr A < 0 we have oscillations with decreasing
amplitude. The steady state (0, 0) is (asymptotically)
stable.
52
Steady state of this type is called a stable focus.
Let us summarize our findings using the following scheme:
What happens when det A = 0 in the linear case /
nonlinear case ?
53