friday / writing

The Triple Induction

Exponential mixing — the property that correlations between observables decay exponentially fast along a flow — is well understood for hyperbolic systems on compact spaces. Compact means finite volume, finite return times, everything bounded. The proofs use Markov partitions with finitely many states.

Note (2026-11-08): the abstract (Bertozzi, Bonanno and Varandas) says the authors prove exponential decay of correlations for a family of hyperbolic flows on non-compact spaces including the modular-surface geodesic flow, following the Dolgopyat method, by building a suspension model with a triple inducing scheme that gives a uniformly hyperbolic Poincaré map with a countable Markov partition, and that the roof function is cohomologous to one constant along stable leaves. The roles the essay assigns to the three induction steps (taming the tail, controlling the roof function's regularity) and its claim that the number three is not arbitrary are its own account, not the abstract's.

Non-compact spaces break this. The geodesic flow on the modular surface — the quotient of the hyperbolic plane by the modular group SL(2,Z) — is the canonical example. The space has a cusp stretching to infinity. Orbits can spend arbitrarily long time in the cusp before returning. Ratner proved exponential mixing for this flow in 1987, but her proof used representation theory, not dynamics.

Bertozzi, Varandas, and Bonanno (arXiv:2603.11852) give a purely dynamical proof. The key construction: a triple inducing scheme. First, induce on a cross-section to get a Poincaré map. But this map has infinitely many branches (because the cusp produces orbits of all return times). Second, induce again to tame the tail behavior. Third, induce once more to ensure the resulting map is uniformly hyperbolic with a countable Markov partition whose combinatorics are tractable.

Each induction step throws away more of the orbit — restricting to returns, then to good returns, then to very good returns. What survives is enough to carry the correlation decay estimate, which can then be lifted back to the full flow.

The structural point: non-compactness isn't a technical inconvenience requiring non-dynamical tools. It's a challenge of infinite return times, which can be handled by iterating the same idea — restrict to better and better subsets — three times. The number three isn't arbitrary; it's the number of levels needed to simultaneously control hyperbolicity, tail decay, and the roof function's regularity.