Spectral Decomposition

In dynamical systems, a spectral decomposition is the decomposition of a system's invariant set into a finite number of basic pieces, each of which is topologically transitive and dynamically irreducible. The spectral decomposition theorem for Axiom A systems, proved by Stephen Smale, states that the non-wandering set can be broken into finitely many basic sets — attractors, repellers, and saddle-like sets — each of which is either an isolated periodic orbit or a topologically transitive set with dense periodic points. This decomposition is the dynamical analogue of the prime factorization in algebra: it breaks complexity into irreducible components that can be analyzed separately.

The spectral decomposition is not just a classification tool. It is a structural guarantee that the global dynamics is built from local pieces that are well-behaved. For hyperbolic systems, the basic sets are the atoms of chaos: each one is a Markov partition away from a shift of finite type, and the global dynamics is a concatenation of these symbolic atoms. The decomposition also reveals the connectivity structure of the system: which basic sets can be reached from which, and which are trapped in basins of attraction.

The spectral decomposition theorem does not apply to non-hyperbolic systems, and its failure is one of the signatures of the Newhouse phenomenon. In systems with infinitely many periodic attractors, there is no finite decomposition into irreducible pieces; the dynamics is a tangle of interleaved attractors whose basins are fractal and intermingled. The absence of a spectral decomposition is not a technical inconvenience — it is a sign that the system has escaped the grasp of finite symbolic coding and entered a regime of genuine, irreducible complexity.

Spectral decomposition is the dream of reductionism made real: the belief that any complex system can be broken into simple parts. The dream dies at the Newhouse phenomenon, where infinity breaks the finiteness. But the dream was always too narrow. A system that cannot be decomposed is not a failure of analysis; it is a system that has achieved a level of integration that no finite list of parts can capture.