Uniform tail bound for the empirical spectral distribution.
ProvedRademacherWigner.expect_frac_large_eigenvalues_leaether-catalogprobability
Uniform tail bound for the empirical spectral distribution. For every order
k ≥ 1 and every threshold t > 0, the expected fraction of eigenvalues of W/√N
of modulus at least t is at most (k+1)^(2k) / t^(2k), a bound that does not
depend on the dimension N.
theorem RademacherWigner.expect_frac_large_eigenvalues_le{k : ℕ} (hk : 1 ≤ k) (hN : 0 < N) {t : ℝ}
(ht : 0 < t) :
expect (fun g : Config N =>
(((univ.filter fun i =>
t ≤ |(W_isHermitian g).eigenvalues i| / Real.sqrt (N : ℝ)).card : ℝ) / (N : ℝ)))
≤ ((k : ℝ) + 1) ^ (2 * k) / t ^ (2 * k) := by sorry
Formalization Note Transplanted verbatim from the Aether Catalog source Probability/WignerBulkTightness.lean; the statement is byte-identical to the source declaration, elaborated with autoImplicit disabled in the platform environment.
Preamble
-- Thm stub generated from Probability/WignerBulkTightness.lean
import Mathlib
import Definitions.Def_Probability_WignerMomentGrowth
import Definitions.Def_Probability_WignerRademacherEnsemble
import Theorems.Thm_RademacherWigner_W_isHermitian
/-
Copyright (c) 2026 Harmonic. All rights reserved.
Released under Apache 2.0 license.
# Tightness of the empirical spectral distribution
`Probability.WignerSpectralEdge` bounds the probability that *some* eigenvalue of
`W/√N` is large. This file records the complementary — and for the semicircle law
more fundamental — statement about the *bulk*: the empirical spectral distribution
(ESD) of `W/√N` puts almost no mass far from the origin, uniformly in the dimension.
* `WignerBridge.frac_large_eigenvalues_le` is deterministic and holds for an
arbitrary real symmetric matrix: the fraction of eigenvalues of `A/√N` of modulus
at least `t` is at most `(2k)`-th normalised moment divided by `t^(2k)` — Markov's
inequality applied to the ESD itself rather than to the ensemble.
* `RademacherWigner.expect_frac_large_eigenvalues_le` combines this with the uniform
moment bound `expect_normalizedMoment_two_mul_le` to give, for every `k ≥ 1`,
`E [ #{ i : |λᵢ|/√N ≥ t } / N ] ≤ (k+1)^(2k) / t^(2k)`,
a bound independent of `N`.
* `RademacherWigner.esd_tight` is the resulting **tightness** statement: for every
`ε > 0` there is a threshold `t` such that, in *every* dimension, the expected
fraction of eigenvalues of `W/√N` outside `[-t, t]` is at most `ε`. Tightness is
exactly the compactness hypothesis under which convergence of all moments upgrades
to weak convergence of the ESD, i.e. it is the missing analytic half of the moment
method (Conjecture 4 of `FUTURE_DIRECTIONS.md`).
-/
open Matrix BigOperators Finset
open WignerBridge
open RademacherWigner
variable {N : ℕ}Formal statement
theorem RademacherWigner.expect_frac_large_eigenvalues_le{k : ℕ} (hk : 1 ≤ k) (hN : 0 < N) {t : ℝ}
(ht : 0 < t) :
expect (fun g : Config N =>
(((univ.filter fun i =>
t ≤ |(W_isHermitian g).eigenvalues i| / Real.sqrt (N : ℝ)).card : ℝ) / (N : ℝ)))
≤ ((k : ℝ) + 1) ^ (2 * k) / t ^ (2 * k) := by sorrySource