the nonlinear pendulum
Everyone meets the pendulum first in the convenient lie that it swings in a perfect sine wave. That is only true for tiny swings. The honest pendulum — released from a wide angle, or whirled all the way over the top — does something richer, and capturing that richness requires keeping the full nonlinear equation instead of its small-angle simplification. The nonlinear pendulum is the canonical first example of nonlinear dynamics.
Newton's law for a bob on a rigid rod gives theta'' + (g/L) sin(theta) = 0, where theta is the angle from straight down. The nonlinearity is the sin(theta): for small angles sin(theta) is approximately theta, which is the linear simple-harmonic model with its constant period, but for large angles the restoring force grows more slowly than the angle, so the swing takes longer. Written as a planar system with omega = theta', it becomes theta' = omega, omega' = -(g/L) sin(theta). Without friction it is Hamiltonian — total energy is conserved — so the phase portrait is just the contour map of the energy: closed loops around the downward equilibrium (back-and-forth oscillation) and wavy open curves above a threshold (the bob going over the top, round and round).
Through the lens of this field, the pendulum is a gallery of the key ideas. The downward rest (theta = 0) is a center of the conservative system — Lyapunov stable but not asymptotically stable, since energy never drains. The inverted balance point (theta = pi) is a saddle — unstable, exactly the pencil-on-its-tip. The dividing trajectory between swinging and spinning is the separatrix through the saddle. Add damping (a -b omega term) and energy now decays: the center becomes an asymptotically stable spiral, and LaSalle's principle proves the bob must finally come to rest hanging straight down. Honest caveat: the small-angle 'constant period' is only an approximation; a real large-amplitude swing has a longer, amplitude-dependent period.
Undamped, theta'' + sin(theta) = 0 has a center at theta = 0 (oscillations) and a saddle at theta = pi (the upright balance); with damping, theta'' + 0.3 theta' + sin(theta) = 0, the center becomes a stable spiral and every swing eventually decays to hanging at rest.
One small example holds a center, a saddle, a separatrix, and (with damping) an asymptotically stable spiral.
The famous 'period independent of amplitude' belongs only to the linear small-angle approximation; the true nonlinear period grows with amplitude and diverges as the swing approaches going over the top. And the downward center is stable but never asymptotically stable until you add damping.