Takens' theorem (1981): a scalar time series x(t) from a chaotic system can reconstruct the full attractor via delay vectors (x(t), x(t+τ), x(t+2τ)). The embedding recovers the topology of the original attractor from a single observable.