Differential Equations › Section E · Systems of differential equations › Topic E.12
Limit cycles, energy, and Lyapunov functions
Nonlinear systems can sustain rhythms that no linear system can, and their stability can often be proved without solving anything. This topic covers limit cycles, the theorem that guarantees them in the plane, and the energy and Lyapunov arguments that settle the cases linearisation leaves open.
Rhythms with a fixed size
Around a linear centre every closed orbit is possible: start further out and the oscillation is larger, for ever. A heartbeat or an electronic oscillator instead settles into the same rhythm whatever its start. Linear systems cannot do this, because any multiple of a periodic solution is another one. Such a rhythm is a limit cycle: an isolated closed trajectory, with nearby trajectories spiralling towards it or away from it. It is stable if trajectories on both sides approach it, unstable if they leave on both sides, and semi-stable otherwise. A stable limit cycle has an amplitude and period set by the system, not by the start, so it models a self-sustained oscillation.
Worked example: a cycle in polar form
Analyse \(x' = x - y - x(x^2 + y^2)\), \(y' = x + y - y(x^2 + y^2)\). 1. Using \(rr' = xx' + yy'\) and \(r^2\theta' = xy' - yx'\), the system becomes \(r' = r(1 - r^2)\), \(\theta' = 1\), so every trajectory turns anticlockwise at unit rate. 2. On the circle \(r = 1\), \(r' = 0\): a closed orbit of period \(2\pi\). 3. For \(0 < r < 1\), \(r' > 0\), and for \(r > 1\), \(r' < 0\), so trajectories approach the circle from both sides: \(r = 1\) is a stable limit cycle. 4. Check: at the origin the Jacobian has rows \((1, -1)\) and \((1, 1)\), so \(\tau = 2\) and \(\Delta = 2\), an unstable spiral, consistent with trajectories spiralling outwards.
The van der Pol oscillator
Balthasar van der Pol studied this equation in the 1920s as a model of vacuum-tube circuits:
The middle term is damping whose sign depends on \(x\): negative for \(|x| < 1\), feeding energy in so small oscillations grow, and positive for \(|x| > 1\), so large ones shrink. The balance is a unique stable limit cycle for every \(\mu > 0\), with an amplitude in \(x\) of about 2. The origin is unstable, since its Jacobian has \(\tau = \mu > 0\) and \(\Delta = 1\).
The Poincaré-Bendixson theorem
Let \(f\) and \(g\) have continuous partial derivatives. If a trajectory of \(x' = f\), \(y' = g\) stays for all \(t \ge 0\) inside a closed, bounded region of the plane that contains no equilibrium points, then it is a closed orbit or it approaches one. In practice, look for a trapping region, often a ring, that trajectories enter but cannot leave. In the polar example the ring \(\frac{1}{2} \le r \le 2\) works: \(r' = 0.375 > 0\) on the inner circle and \(r' = -6 < 0\) on the outer one, and the only equilibrium is outside the ring. A consequence is that planar autonomous systems cannot be chaotic; chaos needs at least three dimensions.
Conservative systems and energy
For \(x'' + P'(x) = 0\), where \(P\) is a potential energy, define
Along a solution,
so energy is conserved and every trajectory lies on a level curve of \(E\). The frictionless pendulum \(\theta'' + 4\sin\theta = 0\) has \(P = 4(1 - \cos\theta)\), with \(y = \theta'\). Near the minimum of \(E\) at the origin the level curves are closed loops, so the linear centre is a true centre. \(P\) peaks at 8 when \(\theta = \pi\), so energies below 8 give swings and energies above 8 give full rotations.
Lyapunov's direct method
Suppose the origin is an equilibrium and \(V(x, y)\) is continuously differentiable with \(V(0, 0) = 0\) and \(V > 0\) at every other point near the origin. Along trajectories,
computed straight from the equations, without solving them. If \(\dot V \le 0\) near the origin, the origin is stable. If \(\dot V < 0\) at every point near it except the origin itself, the origin is asymptotically stable. If instead \(\dot V > 0\) at every such point, the origin is unstable. Such a \(V\) is a Lyapunov function. There is no recipe for one; energies and sums like \(ax^2 + by^2\) are the usual first tries.
Worked example: where linearisation fails
Decide the stability of the origin for \(x' = y - x^3\), \(y' = -x - y^3\). 1. The Jacobian at the origin has rows \((0, 1)\) and \((-1, 0)\): a linear centre, so linearisation is inconclusive. 2. Try \(V = x^2 + y^2\), which is zero at the origin and positive elsewhere. 3. Compute \(\dot V = 2x(y - x^3) + 2y(-x - y^3)\). 4. The \(xy\) terms cancel, leaving \(\dot V = -2x^4 - 2y^4\). 5. This is negative except at the origin, so the origin is asymptotically stable. 6. Contrast: for \(x' = y + x^3\), \(y' = -x + y^3\), with the same linearisation, the same \(V\) gives \(\dot V = 2x^4 + 2y^4 > 0\), so that origin is unstable.
Worked example: energy with friction
Add friction: \(x' = y\), \(y' = -4\sin x - cy\), with \(c > 0\). 1. Take \(V = E = \frac{1}{2}y^2 + 4(1 - \cos x)\), zero at the origin and positive at other points with \(|x| < 2\pi\). 2. Then
so the origin is stable. 3. But \(\dot E = 0\) all along \(y = 0\), so this test alone does not give asymptotic stability. 4. The linearisation fills the gap: \(\tau = -c < 0\) and \(\Delta = 4 > 0\), a hyperbolic stable equilibrium, so the swings die away.
Hopf bifurcations and chaos
Consider the family \(r' = r(\mu - r^2)\), \(\theta' = 1\). For \(\mu < 0\), \(r'\) is negative for all \(r > 0\) and the origin is a stable spiral. For \(\mu > 0\), the origin becomes an unstable spiral, with Jacobian eigenvalues \(\mu \pm i\), and a stable limit cycle of radius \(\sqrt{\mu}\) appears. A steady state giving way to an oscillation as a parameter crosses a threshold is a Hopf bifurcation; it is used to model the onset of aircraft-wing flutter and of some chemical oscillations. In three or more dimensions there is a further possibility, chaos, taken up in the next topic.
Common mistakes to avoid
First, calling every closed orbit a limit cycle: the orbits around a centre are not isolated. Second, applying Poincaré-Bendixson to a region that contains an equilibrium, or in three dimensions. Third, concluding asymptotic stability from \(\dot V \le 0\); that needs \(\dot V < 0\) away from the equilibrium, or another argument. Fourth, deciding a point is unstable because one guess for \(V\) failed: a different \(V\) may work. Fifth, using a \(V\) that is not positive near the equilibrium.
Key takeaways
A limit cycle is an isolated closed orbit; a stable one, as in the van der Pol oscillator, is a self-sustained oscillation. Poincaré-Bendixson guarantees a closed orbit in a closed, bounded planar region that traps trajectories and holds no equilibria. Conservative systems keep energy constant, so trajectories follow its level curves. A Lyapunov function \(V\) proves stability if \(\dot V \le 0\) and asymptotic stability if \(\dot V < 0\) away from the equilibrium, even where linearisation is silent. Hopf bifurcations turn steady states into oscillations.
Flashcards
Flashcard
An isolated closed trajectory
Test your understanding
Question 1 of 5
A system in polar form is r′ = r(4 − r²), θ′ = 1. What closed orbit does it have?
Question 2 of 5
For x′ = −x³ + y, y′ = −x − 2y³, take V = x² + y². What does V̇ show?
Question 3 of 5
Which situation lets you apply the Poincaré-Bendixson theorem to conclude that a closed orbit exists?
Question 4 of 5
A frictionless pendulum obeys θ″ + 4 sin θ = 0. Starting from the bottom, θ = 0, what angular speed must it exceed to go over the top?
Question 5 of 5
For r′ = r(μ − r²), θ′ = 1, what happens when μ = 0.25?
