A-posteriori shadowing certificate for a binary64 logistic execution.
ProvedNovelty.FloatBackwardError.logistic_aposteriori_shadowingaether-catalognovelty
A-posteriori shadowing certificate for a binary64 logistic execution.
The exponential factor of logistic_binary64_shadowing is replaced by the
product of the observed local expansion factors 4 · max(xₖ, 1 - xₖ) ∈ [2,4],
a quantity computable from the execution itself.
theorem Novelty.FloatBackwardError.logistic_aposteriori_shadowing(M : RoundingModel)
(hu : M.u ≤ (2:ℝ) ^ (-53 : ℤ)) (x₀ : ℝ) (hx₀ : x₀ ∈ Set.Icc (0:ℝ) 1) (N : ℕ)
(hstay : ∀ n ≤ N, flOrbit M logisticCoeffs x₀ n ∈ Set.Icc (0:ℝ) 1) :
∀ n ≤ N, |flOrbit M logisticCoeffs x₀ n - trueOrbit logistic x₀ n|
≤ errBound ((2:ℝ) ^ (-46 : ℤ))
(fun k => 4 * max (flOrbit M logisticCoeffs x₀ k)
(1 - flOrbit M logisticCoeffs x₀ k)) n := by sorry
Formalization Note Transplanted verbatim from the Aether Catalog source Novelty/FloatShadowingSharpness.lean; the statement is byte-identical to the source declaration, elaborated with autoImplicit disabled in the platform environment.
Preamble
-- Thm stub generated from Novelty/FloatShadowingSharpness.lean
import Mathlib
import Definitions.Def_Novelty_FloatBackwardErrorHorner
import Definitions.Def_Novelty_FloatPseudoOrbitShadowing
import Definitions.Def_Novelty_FloatShadowingSharpness
/-!
# Nonautonomous shadowing bounds and their sharpness
Third cycle of the programme. Two questions left open by the previous cycles
are settled here.
1. *Is the uniform Lipschitz constant necessary?* No: `variable_shadowing`
replaces `L` by the observed, step-dependent local expansion factors `L n`,
producing the a-posteriori error recursion `E 0 = 0`,
`E (n+1) = δ + L n · E n`. Specialising to a constant sequence recovers the
geometric bound (`errBound_const`), so the nonautonomous statement is a
strict refinement.
2. *Is the exponential growth of the forward bound an artifact of the proof?*
No: `finite_shadowing_sharp` exhibits, for every `L ≥ 0` and `δ ≥ 0`, an
`L`-Lipschitz map and a `δ`-pseudo-orbit for which the distance to the true
orbit **equals** `δ (1 + L + ⋯ + L^{n-1})` at every step. Hence the forward
shadowing theorem of the first cycle cannot be improved, and the improvement
obtained in the second cycle (`expanding_backward_shadowing`) genuinely
requires moving the initial condition.
-- !-- Lab Notes -- !--
Hypothesis (Hypothesizer): the geometric factor in forward shadowing is exactly
attained by the linear map `z ↦ L z` driven by a constant defect `δ`, so no
proof technique can remove it while keeping the initial point fixed.
Experiment (Experimenter): the witness is the affine recursion
`x_{n+1} = L xₙ + δ` started at `0`, whose true orbit is identically `0`; the
distance is the geometric sum, verified by induction (`sharpWitness_eq`).
Analysis (Analyst): the three cycles now separate cleanly:
semantics (`O(u)` defect) → forward dynamics (`O(u · Lⁿ)`, sharp) →
backward dynamics under expansivity (`O(u)`, uniform in n).
Critique (Critic): the sharpness witness is an honest pseudo-orbit — its defect
is exactly `δ` at every step, not merely bounded by `δ` — and the map is exactly
`L`-Lipschitz, so no slack is hidden in the hypotheses.
-- !-- End Lab Notes -- !--
-/
open Novelty.FloatBackwardError
open scoped BigOperators
/-! ### Sharpness of the forward shadowing bound -/
/-! ### The a-posteriori bound for a binary64 logistic execution -/Formal statement
theorem Novelty.FloatBackwardError.logistic_aposteriori_shadowing(M : RoundingModel)
(hu : M.u ≤ (2:ℝ) ^ (-53 : ℤ)) (x₀ : ℝ) (hx₀ : x₀ ∈ Set.Icc (0:ℝ) 1) (N : ℕ)
(hstay : ∀ n ≤ N, flOrbit M logisticCoeffs x₀ n ∈ Set.Icc (0:ℝ) 1) :
∀ n ≤ N, |flOrbit M logisticCoeffs x₀ n - trueOrbit logistic x₀ n|
≤ errBound ((2:ℝ) ^ (-46 : ℤ))
(fun k => 4 * max (flOrbit M logisticCoeffs x₀ k)
(1 - flOrbit M logisticCoeffs x₀ k)) n := by sorrySource