14 · Bifurcations
Chapter 14

Bifurcations

In chapter 12 we learned to read the flow near an equilibrium from its linearization. Every system there had fixed numbers in it. But models come with parameters: a harvesting rate, a drag, a growth rate. In section 11.3 we saw that moving a parameter can change what a system does completely. There, we could only watch it happen in pictures. In this chapter we study how and where it happens: how equilibria appear, disappear, and change type as a parameter moves, and why linearization fails at exactly those parameter values.

14.1Bifurcation Diagrams

In section 4.2.2 we harvested fish from a pond at a constant rate of 𝐻 fish per year:

𝑃′=𝑃(1−𝑃200)−𝐻.

For 𝐻 =32, the population had a threshold at 40 and a stable equilibrium at 160. As we raised 𝐻, the two moved toward each other. At 𝐻 =50 they met at 100, and for any larger harvest there were no equilibria at all.

Each value of 𝐻 gives a different equation. Let's look at its solutions.

Figure 14.1 The slope field and solutions of 𝑃′ =𝑃(1 −𝑃/200) −𝐻, with the equilibria as red lines: solid where nearby solutions approach, dashed where they move away. Move the slider to choose the harvest 𝐻.

For a harvest below 50, solutions that start above the lower equilibrium level off at the upper one, and solutions that start below it fall to zero. Raise the harvest past 50 and the red lines are gone: every solution falls.

For the long run, this picture holds more than we need. All that matters is where the equilibria are and which way solutions move between them. So we can collapse the picture onto a single vertical line: mark the equilibria, and draw arrows toward or away from them, pointing up where 𝑃′ >0 and down where 𝑃′ <0. This is the phase line of section 3.5.

Figure 14.2 The phase line of 𝑃′ =𝑃(1 −𝑃/200) −𝐻: the equilibria, filled where they attract and open where they repel, with arrows showing which way the population moves. Move the slider to choose the harvest 𝐻.

As the harvest grows, the two equilibria on the phase line move toward each other, meet at 𝐻 =50, and disappear. Instead of watching one phase line change, we can stand the phase lines for every harvest side by side.

Figure 14.3 The equilibria of 𝑃′ =𝑃(1 −𝑃/200) −𝐻 for every harvest 𝐻, solid red where they attract and dashed where they repel. The gold line is the phase line for the harvest chosen with the slider.

To draw the curve, we solve 𝑃 −𝑃2/200 =𝐻 for 𝑃. Multiplying by 200 and rearranging gives 𝑃2 −200𝑃 +200𝐻 =0, so

𝑃=100±√10000−200𝐻.

For 𝐻 <50 there are two equilibria, the upper one stable and the lower one a threshold. At 𝐻 =50 the square root is zero and they meet. For 𝐻 >50 there are none. A picture like this, showing the equilibria and their stability for every value of a parameter, is called a bifurcation diagram. A parameter value where the phase portrait changes in a basic way, like 𝐻 =50 here, is called a bifurcation. This kind, where a stable equilibrium and an unstable one move together, collide, and disappear, is a saddle-node bifurcation. The name comes from its version in the plane, where the two equilibria that collide are a saddle and a sink or source.

Linearization tells us which branch is which. For a single equation 𝑃′ =𝑓(𝑃), the Jacobian is just the derivative 𝑓′. So near an equilibrium 𝑃∗, writing 𝑃 =𝑃∗ +𝑢, the linearization is

𝑢′=𝑓′(𝑃∗)𝑢.

The equilibrium attracts when 𝑓′(𝑃∗) <0 and repels when 𝑓′(𝑃∗) >0. Here 𝑓(𝑃) =𝑃 −𝑃2/200 −𝐻, so 𝑓′(𝑃) =1 −𝑃/100. This is negative above 𝑃 =100 and positive below it. So the upper branch attracts and the lower branch repels, just as the phase lines show.

At the bifurcation itself, 𝑃 =100, and 𝑓′(100) =0: the linearization fails exactly there. That is no coincidence. Picture the graph of 𝑓, the parabola from section 4.2.2 shifted down by 𝐻. An equilibrium is a place where the graph crosses zero. If it crosses with nonzero slope, then moving the graph up or down a little still leaves a crossing nearby, so the equilibrium survives a small change in 𝐻. An equilibrium can only appear or disappear where the graph touches zero without crossing, that is, where

𝑓(𝑃)=0and𝑓′(𝑃)=0.

Solving these together finds the bifurcation. Here 𝑓′(𝑃) =0 gives 𝑃 =100, and then 𝑓(100) =0 gives 𝐻 =50.

So the largest harvest the pond can sustain, which we found in section 4.2.2, is a bifurcation value. For any harvest above 50 fish per year, the pond has no equilibrium at all, and every population eventually collapses.

14.2Transcritical and Pitchfork Bifurcations

Equilibria don't always appear and disappear. Sometimes they stay, but trade stability.

In section 4.2.2 we noted that a constant quota keeps removing fish from an empty pond. Suppose instead that we remove a fixed fraction 𝑞 of the fish each year:

𝑃′=𝑃(1−𝑃200)−𝑞𝑃=𝑃(1−𝑞−𝑃200).

Now there are always two equilibria: the empty pond 𝑃 =0, and 𝑃 =200(1 −𝑞).

Figure 14.4 Equilibria of 𝑃′ =𝑃(1 −𝑞 −𝑃/200) for each fishing fraction 𝑞, solid red where they attract and dashed where they repel. Move the slider to choose 𝑞; the gold line is the phase line for that 𝑞.

As 𝑞 rises toward 1, the second equilibrium slides down to 0. Past 𝑞 =1 it is negative, which has no meaning for a pond, but the mathematics continues. Linearization shows what changes. Here 𝑓(𝑃) =𝑃(1 −𝑞) −𝑃2/200, so 𝑓′(𝑃) =1 −𝑞 −𝑃/100. At the empty pond, 𝑓′(0) =1 −𝑞. At the other equilibrium, 𝑓′(200(1 −𝑞)) = −(1 −𝑞). For 𝑞 <1, the fish population attracts and the empty pond repels. For 𝑞 >1, they swap. At 𝑞 =1 the two equilibria meet, both derivatives are zero, and the linearization fails again. This is a transcritical bifurcation: two equilibria pass through each other and exchange stability.

The two harvesting policies end very differently. With proportional harvesting, the population shrinks steadily as 𝑞 grows, reaching zero only at 𝑞 =1. With a constant quota, the stable population is still 100 at 𝐻 =50, and then any larger quota sends it to collapse.

Finally, an equilibrium can split into three. Consider

𝑥′=𝜇𝑥−𝑥3,

where 𝜇 is a parameter. The origin is always an equilibrium. When 𝜇 >0 there are two more, 𝑥 = ±√𝜇.

Figure 14.5 Equilibria of 𝑥′ =𝜇𝑥 −𝑥3 for each 𝜇, solid red where they attract and dashed where they repel. Move the slider to choose 𝜇; the gold line is the phase line for that 𝜇.

Here 𝑓′(𝑥) =𝜇 −3𝑥2. At the origin, 𝑓′(0) =𝜇, so the origin attracts when 𝜇 <0 and repels when 𝜇 >0. At 𝑥 = ±√𝜇, 𝑓′ =𝜇 −3𝜇 = −2𝜇 <0, so both new equilibria attract. As 𝜇 increases through 0, the stable origin becomes unstable and hands its stability to a pair of new equilibria, one on each side. Once more, the linearization fails at the bifurcation: at 𝜇 =0, 𝑓′(0) =0. The diagram looks like a pitchfork, and this is called a pitchfork bifurcation.

The pair appears together because of a symmetry. Replacing 𝑥 by −𝑥 turns the equation into itself, since 𝑓( −𝑥) = −𝑓(𝑥). So whenever 𝑥 is an equilibrium, so is −𝑥. A ruler squeezed from both ends behaves this way: past a certain force, it buckles to the left or to the right, and by symmetry neither side is preferred.