Lemma A.4 — Gaussian sum bounds
ProvedTentNoise.lemma_A_4analysisp2o-gran-per-chapterp2o-plan-paperp2o-v1probability
For every ,
These are the tail estimates for the spectral series of the reflected heat kernel used in Lemmas A.5 and A.6.
Formalization Note The summation index is written with . Both series are also asserted to be summable.
Differences from the source:
- encoding: ∑{n=1}^∞ a_n ≤ B → Summable (n ↦ a{n+1}) ∧ ∑' n : ℕ, a_{n+1} ≤ B; reason: index shift; summability is asserted so that Lean's default value 0 of a divergent tsum cannot satisfy the bound
Preamble
import Mathlib open Set
Formal statement
namespace TentNoise
/-- **Lemma A.4 (Gaussian sum bounds)** (Huan, Risteski, Wei, *Mitigating error propagation in chaotic dynamics via inference-time noising: a view from backward sampling*, STOC 2027 submission (manuscript sha256 67925b3e…), Lemma A.4, p.18, eqs. (26)–(27)): for every
`α > 0`, `∑_{n≥1} e^{−αn²} ≤ √π/(2√α)` and `∑_{n≥1} n e^{−αn²} ≤ 1/α`.
**Formalization Note** The index `n ≥ 1` is written `n + 1` with `n : ℕ`. Both series are asserted to
be summable, so the bounds are not about Lean's default value `0` of a divergent `tsum`. -/
theorem lemma_A_4 (α : ℝ) (hα : 0 < α) :
(Summable (fun n : ℕ => Real.exp (-α * ((n : ℝ) + 1) ^ 2)) ∧
∑' n : ℕ, Real.exp (-α * ((n : ℝ) + 1) ^ 2) ≤ Real.sqrt Real.pi / (2 * Real.sqrt α)) ∧
(Summable (fun n : ℕ => ((n : ℝ) + 1) * Real.exp (-α * ((n : ℝ) + 1) ^ 2)) ∧
∑' n : ℕ, ((n : ℝ) + 1) * Real.exp (-α * ((n : ℝ) + 1) ^ 2) ≤ 1 / α) := by sorry
end TentNoise
Source
Huan, Risteski, Wei, Mitigating error propagation in chaotic dynamics via inference-time noising: a view from backward sampling, STOC 2027 submission (manuscript sha256 67925b3e946da20fbfc5ba5a928c6f5743bf4b2ee96e526e8e738f0494f558e6), Lemma A.4, p.18 (PDF p.18), eqs. (26)–(27)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.