Math 445: Statistical, Dynamical, and Computational Modeling

POPULATION DYNAMICS: ONE SPECIES MODELS

Fundamental concepts of population ecology and disease propagation dynamics rely heavily on the basic notions coming from the general qualitative theory of dierential equations. For example, in the book The Ecology of Wildlife Diseases we may nd such terms as equilibrium population that must be perturbed to estimate certain parameter values (p. 45), reproduction number (p. 48), structural stability/instability of a model (p. 50), bifurcation theory (p. 61), stability analysis for the equilibrium (steady) states (p. 69), Law of Mass Action (p. 87), etc. We will discuss (in simple terms) the underlying ideas related to these notions. Main goal of this part of the course: teach you how to read articles with models formulated in terms of dierential (and dierence) equations; also, teach you how to derive and analyze the models formulated in terms of dierential equations.

1. Introduction. Where do the models come from? What is the relation between the statistics and the dynamic models? When the data is collected, which curves do we want to t to the data and why? In what follows we will use notation t for independent variable (time), and notation u for dependent variable (concentration, population density, etc.)

2. Assume that the parameters are known exactly (in reality, they must be estimated from the experimental data). We want to know what types of behavior (qualitative and quantitative) to expect. It is convenient to associate numerical values with such quantities as birth/death rate coecients, mobility, etc., to compare the characteristics of healthy and infected individuals. 3. Derivation of the elementary model:

Simplistically, u = u(t + t) u(t) ku(t )t, where t [t, t + t], k = (birth rate coef - death rate coef).

Divide by t and take the limit as t 0: u du = = k lim u(t ). t0 t t0 dt lim Thus, du = ku(t). dt Initial condition must be specied: u(0) = u0 . Solution can be easily guessed: we recall that dekt = kekt . dt The equation can be multiplied by any nonzero constant! Thus, the general solution of this linear homogeneous equation is u(t) = Cekt ,
3

where C is an arbitrary constant of integration. Finally, substituting the general solution into the initial condition, we dene C : u(0) = C = u0 , so, u(t) = u0 ekt .

Here are some characteristic shapes of u(t) for various u0 :

4. Phase line. Phase line is the u-axis. Analysis of solution behavior on the phase line (and later, in the case of several dependent functions, on the phase plane) is a powerful tool that allows one to make conclusions on the qualitative behavior of the solution (e.g., whether the population is increasing/decreasing, going to a steady state, etc). We recall that the derivative du/dt represents the rate of change: if du/dt > 0, then the function u is increasing; if du/dt < 0, then the function u is decreasing. In the phase line we show the increase (decrease) of u for increasing t by the arrow pointing to the right (left):

For the equation du = ku(t) dt the phase line may be constructed as follows:
4

In general, we have du = f (u), dt and

For steady states u = u =const we have du /dt = 0, and thus f ( u) = 0. From this equation we nd u 1 , u 2 , etc.
5

5. Stability of the steady states. Geometric considerations; domains of attraction.

Thus, we conclude that if f ( ui ) > 0, then u i is unstable; if f ( ui ) < 0, then u i is stable; if f ( ui ) = 0, then further analysis is needed:

6. Linearization. Analytically the same results may be obtained as follows. Let the system have a steady state ui . We introduce a small perturbation (e.g., we add a small number of species to the system, or remove a small number of species). What is going to happen to the population?
6

We substitute u(t) = u i +(t) (|(0)| Since du i /dt 0, we have:

1) into the equation du/dt = f (u).

d = f ( ui + (t)). dt Taylor series expansion with the center at u i : f ( ui + ) = f ( ui ) + f ( ui ) + f ( ui ) 2 + .... 2!

At the steady state f ( ui ) = 0. Then, for small, we arrive at the linear equation that we already know how to solve: d = f ( ui ), dt We write:
ui )t (t) = 0 ef ( ,

(0) = 0

given.

and so if f ( ui ) > 0, then u i is unstable; if f ( ui ) < 0, then u i is stable; if f ( ui ) = 0, then further analysis is needed: higher order terms in the Taylor series expansion must be considered. 7. Bifurcations. Usually models contain parameters. Let us change some parameter and consider solutions of the models for dierent parameter values. Bifurcation is the abrupt change of the qualitative behavior of model solutions (e.g., change in the number of steady states, change in the stability properties of steady states, etc.) for certain, so-called, bifurcation values of the parameter. (Parameter itself then is often called the bifurcation parameter.) Consider a function f (u, ), where lambda is a parameter.

Geometry of the bifurcation ( = is the bifurcation value of the parameter):

8. Bifurcation diagram.

9. Logistic equation. Heuristic derivation: Population growth/decay coecient is population dependent. The simplest form of population dependent growth rate coecient that decays as population increases is k (u) = K 1 u , M

where M is called carrying capacity, and K = const. Then, du u = k (u(t))u(t) = Ku 1 dt M = Ku u2 = growth w.o. competition competition. Here = K/M . Another way to derive: Consider a well stirred batch reactor (closed tank). Let u be the population density of bacteria in the tank, and c be the food concentration. Assume k = k (c) = c, then we have a system of equations du = cu, dt dc = cu, dt

where 1/ is, so-called, yield factor. Initial conditions are u(0) = u0 , c(0) = c0 .

The above system can be reduced to one logistic equation as follows. Let us multiply the rst equation by and add the two equations. We obtain, du dc d(u + c) + = = 0. dt dt dt

Integrating this equation, we get for any t: u(t) + c(t) = const, and from the initial conditions:
9

u(t) + c(t) = u0 + c0 From the above we express

(conservation of mass!).

c(t) = u0 + c0 u(t) = A u, substitute in the original equation for u, to arrive at the logistic equation: du du u = cu = = u(A u) = Ku 1 , dt dt M where K = A and M = A/. Phase line analysis of the logistic equation.

Exact solution (derivation later!) of u du = Ku 1 , dt M is u(t) = M u0 . u0 + (M u0 ) exp(Kt)

10

u(0) = u0 > 0

10. Logistic equation with harvesting. Several ways to harvest! For constant rate of species removal (vs. constant eort): du u = Ku 1 h, dt M where h is the harvesting term (bifurcation parameter). For which value of h = h will the species go extinct?

h = max Ku 1
u[0,M ]

u M

=K

M 1 KM = . 2 2 4

11. Bifurcation diagram.

11

12. Other useful facts. (a) Non-dimensionalization/re-scaling. How many parameters are actually important? dimensional variable = non-dimensional variable dimensional constant Example 1. Consider du = ku, dt u(0) = u0 .

Let u = U a, t = b, where a, b = const, and U , are non-dimensional variables. Substituting into the equation and condition, we obtain: a dU = kaU, b d If we choose a = u0 we can write: dU = U, d U (0) = 1. (i.e., we measure u in the units of u0 ), b = 1/k, aU (0) = u0 .

NO PARAMETERS! After solving the above problem, we may return to the original variables by back substitution: U ( ) = e , or u(t) = et/b , a or u(t) = u0 ekt .

Example 2. It can be easily checked that for the logistic equation problem the change to non-dimensional variables U= converts to dU = U (1 U ), d
12

u , M

= Kt

du u = Ku 1 , dt M

u(0) = u0 , u0 . M

U (0) = U0 =

(b) Superposition principle. What if we want to solve the non-homogeneous linear equation (i.e., equation that also contains terms without u or du/dt)? Example 3. Consider du = ku h, dt where k , h, u0 are constants. Superposition principle: general solution of the non-homogeneous equation = general solution of corresponding homogeneous equation + particular solution of the non-homogeneous equation. In our case the general solution uG (t) of corresponding homogeneous equation, du/dt = ku, is uG (t) = C exp(kt), where C is an arbitrary constant of integration. A particular solution uP of the non-homogeneous equation may be sought in the form uP = A = const (yet unknown). Substituting uP into the original equation, we obtain: dA = 0 = kA h, dt Finally, we have h u(t) = uG (t) + uP (t) = Cekt + , k and, taking into account the initial condition, u(0) = C + we arrive at h = u0 , k h and thus, C = u0 , k and thus, A = h . k u(0) = u0 ,

h u(t) = u0 ekt + (1 ekt ). k

(c) Solution of the logistic equation. Consider a problem for the non-dimensionalized logistic equation: dU = U (1 U ), d U (0) = U0 > 0.

Let us divide both sides of the equation by U 2 : 1 1 dU = 1. U 2 d U

13

Next, we use the new variable Z = 1/U and take into account that dZ/dt = (1/U 2 )dU/dt. The equation will now have the form: dZ = Z + 1. d This is a linear non-homogeneous equation! Applying the procedure described earlier, we arrive at the general solution in the form: Z = Ce + 1, or U = 1 1 = . Z Ce + 1

Substituting this into the initial condition, we determine C = (1 U0 )/U0 . Finally, U ( ) = U0 . U0 + (1 U0 ) exp( )

Returning to the dimensional variables, we get the solution in the form presented earlier in the section on logistic equation. (d) Time dependent growth rate constant. The environmental conditions usually change in time inuencing the birth and death rates. (The time scales of the processes are important!) Consider It can be shown that u(t) = u0 exp
t 0

du = k (t)u, dt

u(0) = u0 .

k (s)ds .

14

Example 4. If du = cos(t)u, dt then we have an oscillating population: u(t) = u0 esin(t) .

Oscillating population 3 2.5 2 1.5 1 0.5 0 u

u(0) = u0 ,

10 t

15

20

15