Spectral-edge tail bound.
ProvedRademacherWigner.prob_exists_large_eigenvalue_leaether-catalogprobability
Spectral-edge tail bound. For every order k ≥ 1 and threshold t > 0, the
probability that the ensemble produces an eigenvalue of W/√N of modulus at least
t is at most N (k+1)^(2k) / t^(2k).
theorem RademacherWigner.prob_exists_large_eigenvalue_le{k : ℕ} (hk : 1 ≤ k) (hN : 0 < N) {t : ℝ}
(ht : 0 < t) :
prob (Finset.univ.filter fun g : Config N =>
∃ i, t * Real.sqrt (N : ℝ) ≤ |(W_isHermitian g).eigenvalues i|)
≤ (N : ℝ) * ((k : ℝ) + 1) ^ (2 * k) / t ^ (2 * k) := by sorry
Formalization Note Transplanted verbatim from the Aether Catalog source Probability/WignerSpectralEdge.lean; the statement is byte-identical to the source declaration, elaborated with autoImplicit disabled in the platform environment.
Preamble
-- Thm stub generated from Probability/WignerSpectralEdge.lean
import Mathlib
import Definitions.Def_Probability_WignerMomentGrowth
import Definitions.Def_Probability_WignerRademacherEnsemble
import Definitions.Def_Probability_WignerSpectralEdge
import Theorems.Thm_RademacherWigner_W_isHermitian
/-
Copyright (c) 2026 Harmonic. All rights reserved.
Released under Apache 2.0 license.
# A quantitative spectral-edge bound at every order
`Probability.WignerMomentGrowth` bounds every even trace moment of the symmetric
Rademacher ensemble by `N^(k+1) (k+1)^(2k)`. Because a single large eigenvalue
already forces a large `2k`-th trace moment, Markov's inequality turns that bound
into a tail estimate for the spectral radius: for every order `k` and every
threshold `t > 0`,
`P [ some eigenvalue of W/√N has modulus ≥ t ] ≤ N (k+1)^(2k) / t^(2k)`.
This is the classical moment route to the spectral edge; the constant `(k+1)^(2k)`
is the crude spanning-tree count rather than the sharp Catalan constant `4^k`, so
the bound becomes informative for `t` of order `k`, and combined with the
deterministic lower bound `√(1 - 1/N) ≤ ‖W/√N‖` of
`Probability.WignerSemicircleCapstone` it sandwiches the spectral radius from both
sides.
-/
open Matrix BigOperators Finset
open RademacherWigner
variable {N : ℕ}Formal statement
theorem RademacherWigner.prob_exists_large_eigenvalue_le{k : ℕ} (hk : 1 ≤ k) (hN : 0 < N) {t : ℝ}
(ht : 0 < t) :
prob (Finset.univ.filter fun g : Config N =>
∃ i, t * Real.sqrt (N : ℝ) ≤ |(W_isHermitian g).eigenvalues i|)
≤ (N : ℝ) * ((k : ℝ) + 1) ^ (2 * k) / t ^ (2 * k) := by sorrySource