Limits of approximately good, approximately invariant measures #
Paper I, the proof of Main Theorem I (2), second half (I:1204–1217): a weak-* limit of points of
the polytopes Q̃_{B_t} is an IRS. Here: if probability measures ν t on sets of words give
eventually no mass to each bad set and are eventually invariant, up to any η > 0, on each
cylinder restrSet J q under conjugation by each letter below s, then their values are
eventually at most valErg + ε (eventually_valΩ_le). The probability measures on the compact
space WordSpace form a compact space (Prokhorov, in Mathlib), so a frequently bad subsequence
would have a cluster point; it is carried by Good s and invariant (the cylinders are a π-system
generating the σ-algebra), so it is the image of an IRS (irs_of_good), whose value is at most
the ergodic value.
The cylinders.
Equations
Instances For
Two probability measures on sets of words agreeing on the cylinders are equal.
A closed set eventually containing a sequence along a filter contains its cluster points.
Limits of approximately good, approximately invariant measures: their values are
eventually at most the ergodic value plus any ε > 0.