Theorem 7.1 — stable finite-dimensional calibration
ProvedInfoLossLAR.WorstCase.theorem_7_1functional-analysisneumann-seriesp2o-gran-per-chapterp2o-plan-bookp2o-v1
Let be a finite-dimensional real normed space, a probability space, a real vector space, and , , linear. Suppose a linear operator satisfies
For every there is a unique with , and satisfies . For every reference , . If the constraints include total mass and the target mass is one, then .
The principle is applied twice in §§8–9 (local and joint calibration) to construct the transport of Theorem 9.9.
Differences from the source:
- reading: "the constraints include total mass and the target mass is one" → for every linear functional on with for all and ; reason: makes the informal phrase precise
- encoding: linear → a continuous linear map with the operator norm; reason: every linear map of a finite-dimensional normed space is continuous
Preamble
import Mathlib set_option autoImplicit false open MeasureTheory
Formal statement
namespace InfoLossLAR.WorstCase
/-- Theorem 7.1 (Stable finite-dimensional calibration), p. 30. `V` is a finite-dimensional real
normed space, `(Ω, μ)` a probability space, `Ψ : V → L¹(μ)` (synthesis), `M : L¹(μ) → Y`
(constraints, `Y` a real vector space) and `E : V → Y` (reference response) linear, and
`A : V → V` linear with `MΨ = EA` and `‖A − I‖ ≤ κ < 1` (operator norm). For every `τ ∈ V` there
is a unique `z` with `Az = τ`; `w = Ψz` satisfies `M(w) = Eτ`; `‖z − z_0‖ ≤ ‖τ − Az_0‖/(1 − κ)`
for every `z_0`; and if the constraints include total mass (a linear functional `m` of `Y` with
`m(M(w')) = E_μ w'` for every weight `w'`) and the target mass is one (`m(Eτ) = 1`), then
`E_μ w = 1`. -/
theorem theorem_7_1 {V Y Ω : Type*} [NormedAddCommGroup V] [NormedSpace ℝ V]
[FiniteDimensional ℝ V] [AddCommGroup Y] [Module ℝ Y] [MeasurableSpace Ω]
(μ : Measure Ω) [IsProbabilityMeasure μ]
(Ψ : V →ₗ[ℝ] Lp ℝ 1 μ) (M : Lp ℝ 1 μ →ₗ[ℝ] Y) (E : V →ₗ[ℝ] Y) (A : V →L[ℝ] V) (κ : ℝ)
(hfac : M ∘ₗ Ψ = E ∘ₗ (A : V →ₗ[ℝ] V)) (hA : ‖A - ContinuousLinearMap.id ℝ V‖ ≤ κ)
(hκ : κ < 1) (τ : V) :
∃ z : V, (A z = τ ∧ ∀ z' : V, A z' = τ → z' = z) ∧
M (Ψ z) = E τ ∧
(∀ z₀ : V, ‖z - z₀‖ ≤ ‖τ - A z₀‖ / (1 - κ)) ∧
(∀ m : Y →ₗ[ℝ] ℝ, (∀ w' : Lp ℝ 1 μ, m (M w') = ∫ ω, w' ω ∂μ) → m (E τ) = 1 →
∫ ω, (Ψ z) ω ∂μ = 1) := by sorry
end InfoLossLAR.WorstCase
Source
Chen Qian, Information Loss Is All You Need: TC⁰ Pseudorandomness from Worst-Case Lattices, arXiv:2611.05136 (held version, sha256 2ebdfbf3…), p. 30, Theorem 7.1
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.