Bifurcations

So far a system's parameters have been fixed and we have asked where its equilibria sit and whether they are stable. Now turn a knob. As a parameter r drifts, the fixed points can slide, collide, appear, vanish, or swap stability. A value of r at which the qualitative structure of the phase portrait suddenly changes is a bifurcation — a tipping point where the number or type of equilibria is reborn.

Bifurcations are how smooth, gradual change produces abrupt, catastrophic effect: a slowly warming climate that flips a stable state, a beam that buckles the instant a load crosses a threshold, a laser that switches on. And there is a beautiful economy to it — near the tipping point the dynamics of any such system collapse onto one of a tiny handful of universal normal forms. Learn four little polynomials and you have met almost every bifurcation there is.

Where a bifurcation must happen

The signal is always the same. A fixed point can only change its character when the linear stability goes borderline — when an eigenvalue's real part crosses zero. In one dimension that means f'(x^\*) = 0 at the same time as f(x^\*) = 0: the curve is tangent to the axis. Away from that condition the fixed point is hyperbolic and (by Hartman–Grobman) robust; nothing dramatic can happen. So the hunt for bifurcations is the hunt for parameter values where

f(x^\*, r) = 0 \quad\text{and}\quad \frac{\partial f}{\partial x}(x^\*, r) = 0 \quad\text{simultaneously.}

The three one-dimensional normal forms

Slide the graph of f(x) and watch its axis-crossings — the fixed points — appear and vanish as r passes through zero.

Saddle-node — \dot{x} = r + x^2. For r < 0 two fixed points x^\* = \pm\sqrt{-r} (one stable, one unstable); they collide and annihilate at r = 0; for r > 0 none. Equilibria created out of nothing.
Transcritical — \dot{x} = rx - x^2. Two fixed points x^\* = 0 and x^\* = r for all r; at r = 0 they cross and exchange stability. Nothing is created — the roles swap.
Pitchfork (supercritical) — \dot{x} = rx - x^3. One stable fixed point x^\* = 0 for r \le 0; at r = 0 it loses stability and two new stable branches x^\* = \pm\sqrt{r} split off. Symmetry-breaking: the system must choose a side.

The pitchfork's picture — one branch splitting into two as r increases past zero — is the bifurcation diagram: fixed point x^\* plotted against the parameter r, solid for stable branches and dashed for unstable ones. It is the single most useful summary in the whole subject.

Into two dimensions: the Hopf bifurcation

The three forms above all involve fixed points on a line. The genuinely two-dimensional event is the Hopf bifurcation, and it is the birthplace of oscillation. Here a spiral fixed point's eigenvalues \alpha(r) \pm i\beta cross the imaginary axis: as \alpha passes from negative to positive the fixed point changes from a stable spiral to an unstable one, and a small limit cycle — an isolated closed orbit — is born around it. In polar normal form,

\dot{r} = \mu r - r^3, \qquad \dot{\theta} = \omega.

The radial equation is exactly a pitchfork: for \mu > 0 a stable radius r = \sqrt{\mu} appears, and because \theta keeps turning, that stable radius is a circular limit cycle. This is how a system that was settling to rest suddenly starts to hum: a heartbeat, a firing neuron, the flutter of an aircraft wing, the onset of a chemical clock. A single eigenvalue crossing the imaginary axis and steady rest gives way to rhythm.

The cubic's sign matters enormously. In the supercritical pitchfork \dot{x} = rx - x^3 the new branches grow gently as \sqrt{r} — a soft, reversible transition. Flip the cubic to subcritical, \dot{x} = rx + x^3, and the branches are unstable and bend backward: the state jumps discontinuously to a far-away attractor and does not return when you reverse r. That backward-bending hysteresis is the mathematics of catastrophe — the buckling beam that snaps, the ecosystem that collapses and won't recover just by nudging the parameter back. Same bifurcation, one sign flipped, utterly different consequence.

Worked example: find the bifurcation

For \dot{x} = r + x^2, locate the bifurcation and classify it.

Step 1 — fixed points. r + x^2 = 0 \Rightarrow x^\* = \pm\sqrt{-r}, which is real only for r \le 0.

Step 2 — tangency condition. f'(x) = 2x vanishes at x = 0, and x^\* = 0 is a fixed point only when r = 0. So the bifurcation is at r_c = 0.

Step 3 — stability of the two branches (for r < 0). f'(\sqrt{-r}) = 2\sqrt{-r} > 0 — the upper branch is unstable; f'(-\sqrt{-r}) = -2\sqrt{-r} < 0 — the lower branch is stable.

Step 4 — verdict. Two fixed points for r < 0, they merge at r = 0, and vanish for r > 0: a textbook saddle-node bifurcation. Past r = 0 there is no equilibrium at all and every trajectory runs off to +\infty.

Don't confuse "a fixed point moves" with "a bifurcation." As r varies, a hyperbolic fixed point slides around continuously and its stability doesn't change — that is not a bifurcation. A bifurcation requires the qualitative structure to change: a point created or destroyed, or its stability flipped, which only happens when an eigenvalue's real part hits zero. And note the counting: a saddle-node changes the number of fixed points (2 ↔ 0), whereas a transcritical or pitchfork keeps or splits them while changing stability. Always check both conditions — f = 0 and f' = 0 — not just the first.