Chaos is aperiodic long-term behavior in a deterministic system that exhibits sensitive dependence on initial conditions. "Deterministic" means the future is uniquely determined by the present state — there are no random inputs. "Aperiodic" means trajectories never repeat exactly. "Sensitive dependence" means that nearby initial conditions diverge exponentially fast, making long-term prediction impossible despite perfect determinism. Chaos requires at least three dimensions for continuous flows (by Poincare-Bendixson) or can occur in one-dimensional discrete maps.
Everything in nonlinear dynamics so far has been, in a sense, well-behaved. Fixed points sit still. Limit cycles repeat periodically. The Poincare-Bendixson theorem guarantees that in two dimensions, nothing more exotic can happen. Chaos shatters this picture: deterministic systems can produce behavior that looks random, never repeats, and defies long-term prediction. It is not randomness, not noise, not complexity — it is the intrinsic unpredictability of certain deterministic systems.
The three defining properties, formalized by Devaney, capture different aspects of the phenomenon. Sensitive dependence on initial conditions is the most famous: two initial conditions that are arbitrarily close will eventually diverge to become completely different. This is not gradual drift — the divergence is exponential, measured by Lyapunov exponents. A butterfly's wing-flap doesn't cause a hurricane through some chain of force; rather, the atmosphere is a chaotic system where any perturbation, no matter how tiny, eventually grows to dominate the forecast. Topological transitivity ensures the chaos is indecomposable — the system explores its entire attractor and can't be split into separate non-interacting parts. Dense periodic orbits mean that arbitrarily close to any chaotic trajectory, there is a periodic orbit — chaos is organized around an infinite skeleton of unstable periodic orbits.
The mechanism of chaos is stretching and folding. Consider a small blob of initial conditions in phase space. Under the dynamics, this blob gets stretched in some directions (divergence of nearby trajectories) and compressed in others (the system is dissipative — volumes contract). But the stretched blob can't extend to infinity if the attractor is bounded. So it gets folded back on itself, like a baker kneading dough. This stretch-fold process, repeated indefinitely, creates an infinitely layered, self-similar structure — the strange attractor. Points that were far apart get folded close together; points that were close get stretched far apart. The result is sensitive dependence (stretching) combined with bounded, recurrent behavior (folding).
Why does chaos require three continuous dimensions? The Poincare-Bendixson theorem explains: in 2D, the non-crossing property of trajectories prevents the folding required for chaos. Trajectories can stretch apart, but they can't fold back past each other — any closed curve that a trajectory tries to weave through becomes a barrier. In three dimensions, trajectories can pass over and under each other, enabling the stretching-and-folding that generates chaotic behavior. This is why the Lorenz system (3D) can be chaotic while the van der Pol oscillator (2D) cannot. For discrete maps, the situation is different: the logistic map is one-dimensional and chaotic, because discrete maps don't have the continuity constraints that prevent crossing.