Two machines, one toolkit
You arrive at the last guide of this rung holding four tools, and it is time to use all of them at once on two systems worth knowing for life. The first is the nonlinear pendulum — a weight on a rod, swinging under gravity, with that stubborn sin(theta) that no linear trick can wish away. The second is the van der Pol oscillator, a circuit Balthasar van der Pol built in the 1920s whose damping changes sign: it pumps energy in when the swing is small and bleeds it out when the swing is large. The pendulum will show us a conservative world of nested loops; van der Pol will show us a single isolated loop that attracts everything. Same toolkit, two opposite fates.
Both are second-order equations, so the first move is the one from the systems rung: convert each into a planar first-order system on the phase plane, with position on one axis and velocity on the other. Then every tool in this rung speaks: find the equilibria, linearize and read the Jacobian, decide where Hartman-Grobman lets you trust the linear verdict, reach for a Lyapunov function where it does not, and call on Poincare-Bendixson when a closed loop is at stake.
The pendulum: down a center, up a saddle
Write the undamped pendulum theta'' + sin(theta) = 0 as x = theta, y = theta', giving x' = y and y' = -sin(x). The equilibria need y = 0 and sin(x) = 0, so the rest points sit at x = 0 (hanging straight down) and x = pi (balanced bolt upright), repeating every 2*pi. The Jacobian has top row (0, 1) and bottom row (-cos(x), 0). At the bottom, cos(0) = 1 gives eigenvalues +/- i — a center. At the top, cos(pi) = -1 gives real eigenvalues +/- 1 of opposite sign — a saddle. The standing-up state is a knife-edge: lean a hair either way and you fall. Exactly the verdict your eye expects.
Energy settles the borderline case
The frictionless pendulum conserves total energy: kinetic plus potential, E = (1/2) y^2 - cos(x), stays fixed along every trajectory. You can check it directly — differentiate E along the motion, dE/dt = y*y' + sin(x)*x' = y*(-sin(x)) + sin(x)*y = 0 — so each orbit is trapped on a level curve E = constant. That is a conserved quantity, and a conservative system like this is the cleanest instance of a Hamiltonian system. Around the bottom the level curves are genuine closed loops, so the 'center' the Jacobian guessed is the real thing after all: small swings circle forever, undecaying.
This is also a lesson in what each tool is for. Energy here plays the role of a Lyapunov-style function, but a special one: it is exactly conserved, dE/dt = 0, not strictly decreasing. So it certifies stability — orbits stay on closed curves and never run away — but never asymptotic stability; nothing is pulled inward, because no friction removes energy. The note about Lyapunov's direct method from earlier in this rung lands precisely here: dE/dt <= 0 buys stability, while dE/dt < 0 strictly is what you would need for the trajectory to actually settle to the rest point. The pendulum gives the first but not the second.
Add a little friction, theta'' + b*theta' + sin(theta) = 0 with b > 0, and the energy tips downward: dE/dt = -b*y^2 <= 0, dropping whenever the bob is moving. Now energy is a true Lyapunov function, the closed loops unwind into inward spirals, and the bottom becomes asymptotically stable — the pendulum finally swings down to rest. The Jacobian agrees: friction nudges those +/- i eigenvalues just left of the imaginary axis into a stable spiral. Honest small print: when b is large the bottom is an overdamped node rather than a spiral, and on the full circle the global picture is richer than any single rest point — high-energy orbits whirl over the top instead of swinging.
Van der Pol: a loop that attracts everything
Now the opposite character. The van der Pol equation is x'' - mu*(1 - x^2)*x' + x = 0 with mu > 0, a sign-changing damping. When x is small (|x| < 1) the factor (1 - x^2) is positive, so the term acts as negative damping — it feeds energy in and pushes the swing to grow. When x is large (|x| > 1) the factor goes negative, ordinary damping returns, and the swing is bled back down. Caught between growing-when-small and shrinking-when-large, the system cannot rest at any fixed amplitude except one special closed orbit. This push-pull is the signature of a self-sustained oscillator: it manufactures its own rhythm with no external forcing at all.
Write it as a system, x' = y, y' = mu*(1 - x^2)*y - x. The only equilibrium is the origin. Its Jacobian there has top row (0, 1) and bottom row (-1, mu), with trace mu > 0 and determinant 1 > 0: both eigenvalues have positive real part, an unstable spiral. So every trajectory is thrown outward from the center. Yet large-amplitude motion is damped back inward. Trajectories are squeezed from both sides — repelled from the origin, herded in from far away — onto a single closed curve trapped in between.
Why the two fates differ
Put the two side by side and the deep contrast clicks. The pendulum is conservative: its phase plane is filled with a continuous family of nested closed loops, one for each energy level, and which one you ride depends entirely on where you start. The van der Pol oscillator is dissipative-with-a-pump: out of that whole plane it selects exactly one closed orbit, and it forgets your initial condition entirely. A true limit cycle is isolated — no other closed orbits nearby — whereas the pendulum's loops come in a thick continuum and so are not limit cycles at all. Telling these apart is precisely the conceptual reward of this rung.
pendulum (undamped) van der Pol (mu > 0)
equilibria down: center origin: unstable spiral
up: saddle
energy conserved, dE/dt = 0 pumped low, damped high
closed orbits continuum of nested loops one isolated limit cycle
start matters? yes (picks the loop) no (all roads -> same loop)
key tool conserved quantity Poincare-Bendixson
Keep two honest caveats in mind. First, Poincare-Bendixson is a strictly two-dimensional theorem — its trapping argument relies on the plane's topology, and in three or more dimensions trajectories can wander forever onto strange attractors and chaos, which the next rung explores. Second, the van der Pol limit cycle has no closed-form formula: for small mu it is a near-circle of radius about 2, while for large mu it becomes a sharp, spiky relaxation oscillation of slow buildups and fast snaps — and we know it exists, is unique, and attracts, all without ever writing x(t). That is the whole spirit of qualitative theory: we read a system's destiny straight off its structure, precisely because we cannot solve it.