Second moment.
ProvedScaleSmoothness.sum_structureCorrection_sqaether-catalognumbertheory
Second moment. ∑_N C(N)² = ∏_p (p + 1/(p−1)).
theorem ScaleSmoothness.sum_structureCorrection_sq:
∑ N : (∀ i, ZMod (a i)), (structureCorrection a N) ^ 2 =
∏ i, ((a i : ℚ) + 1 / ((a i : ℚ) - 1)) := by sorry
Formalization Note Transplanted verbatim from the Aether Catalog source NumberTheory/ScaleSmoothnessDispersion.lean; the statement is byte-identical to the source declaration, elaborated with autoImplicit disabled in the platform environment.
Preamble
-- Thm stub generated from NumberTheory/ScaleSmoothnessDispersion.lean
import Mathlib
import Definitions.Def_NumberTheory_QRDialLocalStatistics
import Definitions.Def_NumberTheory_ScaleSmoothnessDispersion
/-!
# Global structure correction for `x² − N`: mean one, and uniformly bounded dispersion
This is the global half of the round-73 #4 (exp 562) **RANDOM-AT-SCALE** finding.
`Catalog.NumberTheory.QRDialLocalStatistics` computed, prime by prime, the exact
distribution of the local correction
`localFactor p N = (p − dial p N)/(p − 1)`
by which the `p`-part of the smoothness density of `x² − N` differs from that of
a random integer. Multiplying over the primes `p ≤ B` gives the *structure
correction*
`structureCorrection a N = ∏_{i} localFactor (a i) (N i)`,
the full multiplicative factor relating the (Dickman-type) heuristic smoothness
probability of `x² − N` to that of a size-matched random integer.
## Main results
* `sum_structureCorrection` — **mean exactly one**. Averaged over the residue
data `N`, the structure correction is exactly `1`, for *every* finite family
of odd primes. Formalises the experimental null `r(u) = 1`: quadratic-sieve
polynomials carry no ensemble-level smoothness edge, at any `u`.
* `sum_structureCorrection_crt` — the same statement genuinely averaged over
`N mod ∏ p` via the Chinese remainder theorem.
* `sum_structureCorrection_sq`, `sum_structureCorrection_centred_sq` — the
variance is exactly `dispersionBound a − 1` with
`dispersionBound a = ∏_p (1 + 1/(p(p−1)))`.
* `dispersionBound_le_two` — **uniform dispersion bound**: for any finite set of
distinct odd primes, `dispersionBound a ≤ 2`, however large the smoothness
bound `B`. The measured per-`N` overdispersion `D = 1.61 [1.50,1.73]` sits
inside this a-priori ceiling; the numerical value of the infinite product over
all odd primes is ≈ `1.30`.
* `chebyshev_structureCorrection` — a finite Chebyshev inequality: the fraction
of `N` whose structure correction deviates from `1` by at least `t` is at most
`(dispersionBound a − 1)/t² ≤ 1/t²`.
* `no_first_order_smoothness_edge` — the packaged null statement: mean exactly
one *and* deviation mass controlled uniformly in the prime family.
Everything is exact and rational; no analytic estimate is used anywhere.
-/
open ScaleSmoothness
open Finset
variable {ι : Type*} [Fintype ι] [DecidableEq ι]
variable (a : ι → ℕ) [∀ i, Fact (a i).Prime] (hodd : ∀ i, a i ≠ 2)
include hoddFormal statement
theorem ScaleSmoothness.sum_structureCorrection_sq:
∑ N : (∀ i, ZMod (a i)), (structureCorrection a N) ^ 2 =
∏ i, ((a i : ℚ) + 1 / ((a i : ℚ) - 1)) := by sorrySource