Section 2.2 Problems
2.3 More Population Models
2.3.2 The Discrete Logistic Equation
The most popular discrete-time single-species model is the discrete logistic equation, whose recursion is given by
Nt+1= Nt
1+R
1− Nt K
(2.13) where Rand K are positive constants. R is called thegrowth parameterand K is called thecarrying capacity. The analysis that folllows will explain the terminology.
This model of population growth exhibits very complicated dynamics, described in an influential review paper by Robert May (1976).
Before we illustrate its behavior, we will rewrite the model in what is called thecanonical form. The advantage of this form is that the resulting recursion will be simpler. The algebraic steps presented next are not obvious, but will lead to the canonical form of the discrete logistic equation. We write
Nt+1 =Nt
1+R
1− Nt K
=Nt
1+R− R K Nt
=Nt(1+R)
1− R
K(1+R)Nt
Now, dividing by 1+Ryields 1
1+RNt+1 =Nt
1− R
K(1+R)Nt
Let’s multiply both sides byR/K (you'll see why in a moment):
R
K(1+R)Nt+1 = R K Nt
1− R
K(1+R)Nt
(2.14) If we define the new variable as
xt = R
K(1+R)Nt (2.15)
then
R
K(1+R)Nt+1=xt+1 and R
K Nt =(1+R)xt and (2.14) becomes
xt+1=(1+R)xt(1−xt)
At this point, the new parameterr = 1+ Ris customarily introduced. Note that r >1, since R>0. We thus arrive at the canonical form of the logistic recursion:
xt+1 =r xt(1−xt) (2.16)
The advantage of this form is threefold: (1) The recursion (2.16) looks simpler than the original recursion (2.13); (2) instead of two parameters (RandK), there is just one (r); and (3) the quantityxt isdimensionless. The last point needs some expla- nation. The original variableNt has units (or dimension) of number of individuals;
the parameter K has the same units. Dividing Nt by K in (2.15), we see that the units cancel and we say that the quantityxtis dimensionless. [The parameterRdoes not have a dimension, so multiplying Nt/K by R/(1+ R)does not introduce any additional units.] A dimensionless variable has the advantage that it has the same numerical value regardless of what the units of measurement are in the original variable. (See Problems 31–34.) The process of making a quantity dimensionless is callednondimensionalization.
Let’s go back to the discrete logistic equation in its canonical form (2.16) and see what its behavior is. The function f(x)=r x(1−x)is an upside-down parabola, sincer > 1 (Figure 2.16). We see from the figure that ifx is outside of the interval (0,1), f(x)is nonpositive. Sincext = K(1+R)R Nt [see (2.15)], and we wantNt to be positive (it is a population size, after all), we requirextto be positive. This means that we need to ensure thatxt+1 = f(xt)stays within the interval(0,1). The maximum value of f(x)occurs atx = 1/2, and f(1/2)= r/4, so, in order to make sure that f(xt) ∈ (0,1), we require thatr/4 < 1, orr < 4. We already require thatr > 1, sinceR>0. To summarize, if 1<r <4, thenxtstays within the interval(0,1)for all t =1,2,3, . . ., provided thatx0 ∈(0,1). In what follows, we will therefore assume that 1<r <4 andx0∈(0,1).
2 1.5 1 0.5 0 0.5 1
0 0.5 1 1.5 2 x
rx(1 x)
Figure 2.16 A graph of the discrete logistic equation in its canonical form. (Here,r =2.5.)
We first compute fixed points of (2.16). We need to solve x =r x(1−x)
Solving immediately yields the solutionx = 0. Ifx =0, we divide both sides byx and find that
1=r(1−x), or x =1− 1 r
(See Figure 2.17.) Provided thatr >1, both fixed points are in[0,1).
1 1 1/r
0 x
0.5 0.5
rx(1 x) y f(x)
y x
Figure 2.17 A graphical illustration of the fixed points of the discrete logistic equation in its canonical form. The fixed points are where the parabola and the liney=x
intersect.
We return to the original variableNt for a moment to see whatx = 0 andx = 1−1/r mean in terms ofN. Sincex = K(1+R)R N, the fixed pointx =0 corresponds to the fixed point N = 0, which is why we callx = 0 a trivial equilibrium. When x =1−1/r, then, usingr =1+R, we obtain
N = K(1+R)
R x= K(1+R) R
1− 1
1+R
= K(1+R) R
1+R−1 1+R = K soN =K is the other fixed point.
The long-term behavior of the discrete logistic equation is very complicated. We will go through the different cases by simply listing them. Later, in Section 5.6, we will be able, at least to some extent, to understand why this equation has such complicated behavior.
When 1<r <3 andx0∈(0,1),xt converges to the fixed point 1−1/r(Figure 2.18). Increasingr to a value between 3 and 3.449. . ., we learn thatxt settles into a cycle of period 2 (Figure 2.19). This means that, for large enough times, xt will oscillate back and forth between a larger and a smaller value. Forrbetween 3.449. . . and 3.544. . ., the period doubles: A cycle of period 4 appears for large enough times. The population size now oscillates between the same four values (Figure 2.20).
Increasingr continues to double the period: A cycle of period 8 is born whenr = 3.544. . ., a cycle of period 16 whenr = 3.564. . ., and a cycle of period 32 when r =3.567. . .. This doubling of the period continues untilrreaches a value of about 3.57, when the population pattern becomeschaotic(Figure 2.21). The population dynamics seem to be random, although the rules are entirely deterministic! There is no regular pattern we can discern:xtno longer oscillates between the same values; the dynamics areaperiodic. Furthermore, starting from ever so slightly different initial conditions quickly produces very different trajectories (Figure 2.22). This sensitivity to initial conditions is characteristic of chaotic behavior.
0.6 0.5 0.4 0.3 0.2 0.1 0
0 5 10 15
xt
t
Figure 2.18 A graph ofxtas a function oftwhenr =2.
0.6 0.5 0.4 0.3 0.2 0.1 0 0.9 0.8 0.7
0 5 10 15
xt
t
Figure 2.19 A graph ofxtas a function oftwhenr =3.2.
0 5 0
0.2 0.4 0.6 0.8 1
10 15
t xt
20 25 30
Figure 2.20 A graph ofxtas a function oftwhenr =3.52.
0 5
0 0.2 0.4 0.6 0.8 1
10 15
t xt
20 25 30
Figure 2.21 A graph ofxtas a function oftwhenr =3.8.
0 1 0.8 0.6 0.4 0.2 0
5 10
t xt
15 20
x0 0.15 x0 0.2
Figure 2.22 Graphs ofxtas a function oftwhenr =3.8 for two different initial values ofx0.
To obtain biologically sensible results, we needed to restrict bothr andx0. The reason was that ifxt >1, thenxt+1is negative. This situation can be easily remedied by changing the dynamics slightly. We discuss such a model in the next subsection.