Documentation

Oseledets.Lyapunov.Extensions.SingularExponentGenLog

The genuine-log forward singular exponent γ_k^log (EReal-valued) #

For a possibly-singular matrix cocycle generator A : X → Matrix (Fin d) (Fin d) ℝ — no det A ≠ 0, no inverse integrability, only forward integrability — this module packages the cumulative genuine-log forward singular exponent

γ_k^log(x) = limsup_n ((1/n) log sprod_k(A⁽ⁿ⁾ x) : EReal),

built from the honest logarithm Real.log (NOT the positive part Real.posLog used in Oseledets.forwardSingularExponent). Here sprod_k = Oseledets.sprod A T k is the top-k singular-value product (the k-volume growth).

This is the right object for the −∞ kernel / volume-collapse stratum of the Raghunathan / Quas non-invertible multiplicative ergodic theorem. When the cocycle collapses k-volume (sprod_k → 0), the genuine log sprod_k → −∞, so γ_k^log can attain the bottom value of EReal — a regime the log⁺ exponent γ_k cannot see (it is pinned at 0 there). The log⁺ exponent records only the expanding part of the spectrum; this genuine-log exponent is what detects the singular −∞ exponent on the kernel stratum.

The non-invertible MET via exterior algebra, the singular value decomposition, and Kingman's subadditive ergodic theorem (the Raghunathan approach) is the structure followed here; see A. Quas, Multiplicative Ergodic Theorems and Applications (Theorem 2, §3.1; method due to M. S. Raghunathan, A proof of Oseledec's multiplicative ergodic theorem, Israel J. Math. 32 (1979), 356–362).

Main definitions #

Main results #

Implementation notes #

References #

noncomputable def Oseledets.forwardSingularExponentLog {X : Type u_1} {d : } (A : XMatrix (Fin d) (Fin d) ) (T : XX) (k : ) (x : X) :

The genuine-log forward singular exponent γ_k^log of a possibly-singular cocycle generator, as an EReal-valued limsup:

γ_k^log(x) = limsup_n ((1/n) log sprod_k(A⁽ⁿ⁾ x) : EReal),

where sprod_k = Oseledets.sprod A T k is the top-k singular-value product. Unlike the log⁺ exponent Oseledets.forwardSingularExponent, this uses the genuine Real.log, so it can equal when the k-volume collapses (sprod_k → 0, the kernel stratum of the non-invertible MET).

Equations
Instances For
    theorem Oseledets.measurable_forwardSingularExponentLog {X : Type u_1} [MeasurableSpace X] {T : XX} {d : } [NeZero d] {A : XMatrix (Fin d) (Fin d) } (hAmeas : Measurable A) (hTmeas : Measurable T) (k : ) :

    γ_k^log is measurable. Each x ↦ (1/n) log sprod_k(A⁽ⁿ⁾ x) is measurable: sprod is measurable (measurable_sprod, which carries [NeZero d]), Real.log is measurable (Real.measurable_log), and the scalar multiply is too; its ℝ → EReal coercion is measurable (measurable_coe_real_ereal), and the -limsup of measurable EReal-valued functions is measurable (Measurable.limsup). Mirrors measurable_forwardSingularExponent.

    theorem Oseledets.forwardSingularExponentLog_le {X : Type u_1} {d : } (A : XMatrix (Fin d) (Fin d) ) (T : XX) (k : ) (x : X) :

    γ_k^log(x) ≤ γ_k(x) for every x (deterministic, no hypotheses), where γ_k is the log⁺ exponent Oseledets.forwardSingularExponent. Termwise Real.log t ≤ Real.posLog t, so (1/n) log sprod_k ≤ (1/n) log⁺ sprod_k for every n (the factor (n:ℝ)⁻¹ ≥ 0); coercing to EReal (EReal.coe_le_coe_iff) and using monotonicity of the EReal-limsup (Filter.limsup_le_limsup with the everywhere-) yields the bound. The genuine-log exponent is thus always dominated by the log⁺ one; the gap is the collapse −∞ stratum log⁺ cannot see.

    theorem Oseledets.forwardSingularExponentLog_eq_bot_of_tendsto {X : Type u_1} {d : } {A : XMatrix (Fin d) (Fin d) } {T : XX} {k : } {x : X} (h : Filter.Tendsto (fun (n : ) => (↑n)⁻¹ * Real.log (sprod A T k n x)) Filter.atTop Filter.atBot) :

    The −∞ kernel / volume-collapse stratum hook. If the normalized genuine log-volume (1/n) log sprod_k(A⁽ⁿ⁾ x) tends to −∞ (the top-k volume collapses super-exponentially — the kernel stratum of the non-invertible Raghunathan/Quas MET), then the genuine-log exponent attains the bottom value: γ_k^log(x) = ⊥. The real sequence tending to atBot makes its ℝ → EReal coercion converge to 𝓝 ⊥ (EReal.tendsto_coe_nhds_bot_iff), and the EReal-limsup of a convergent sequence is its limit (Filter.Tendsto.limsup_eq). This is the exponent the log⁺ packaging cannot record (forwardSingularExponent ≥ 0 always).