Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Corollary 3.5 — ‖Aᴺ‖ = (1 + o(1))·2√(2d−1) with high probability, for any d = d(N)

Proved
StrongConv.tendsto_prob_norm_adjPerp_div_sub_one_le

by dbenbenn · Nov 4, 2026 · Mathlib 0df444a (Lean v4.33.1)

random-matricesrandom-permutationsrandom-regular-graphsstrong-convergence

Let d:N→Nd : \mathbb N \to \mathbb Nd:N→N be any function with d(N)≥1d(N) \ge 1d(N)≥1. For every δ>0\delta > 0δ>0,

P[∣∥AN∥22d(N)−1−1∣≤δ]→1(N→∞),\mathbf P\Big[\Big|\frac{\|A^N\|}{2\sqrt{2d(N)-1}} - 1\Big| \le \delta\Big] \to 1 \quad (N \to \infty),P[​22d(N)−1​∥AN∥​−1​≤δ]→1(N→∞),

where ANA^NAN is the adjacency matrix of the random 2d(N)2d(N)2d(N)-regular graph on NNN vertices (permutation model) restricted to {1N}⊥\{1_N\}^\perp{1N​}⊥. That is, ∥AN∥=(1+o(1))22d−1\|A^N\| = (1 + o(1))2\sqrt{2d-1}∥AN∥=(1+o(1))22d−1​ with probability 1−o(1)1 - o(1)1−o(1). The paper does not restrict d(N)d(N)d(N) beyond "arbitrary"; d(N)≥1d(N) \ge 1d(N)≥1 excludes the empty tuple, for which 22d−12\sqrt{2d-1}22d−1​ is not defined.

Preamble
import Mathlib
import Definitions.Def_StrongConv

open Filter Topology
Formal statement
namespace StrongConv

theorem tendsto_prob_norm_adjPerp_div_sub_one_le (d : ℕ → ℕ) (hd : ∀ N, 1 ≤ d N) (δ : ℝ)
    (hδ : 0 < δ) :
    Tendsto (fun N : ℕ => uniformProb (fun σ : Fin (d N) → Equiv.Perm (Fin N) =>
      |‖adjPerp σ‖ / (2 * Real.sqrt (2 * d N - 1)) - 1| ≤ δ)) atTop (𝓝 1) := by
  sorry

end StrongConv
Source
Chen, C.-F., Garza-Vargas, J., Tropp, J. A. and van Handel, R., A new approach to strong convergence, Ann. of Math. (2) 203 (2026) 555–602, https://doi.org/10.4007/annals.2026.203.2.4 (arXiv:2405.16026v3, whose page numbers are used), p. 10, Corollary 3.5
Read-back

What the Lean code literally says, in plain math · claude-opus-5-5

What the statement asserts

Let d:N→Nd : \mathbb{N} \to \mathbb{N}d:N→N be any sequence of natural numbers with d(N)≥1d(N) \ge 1d(N)≥1 for every NNN, and let δ>0\delta > 0δ>0 be a real number. Then

lim⁡N→∞  Pr⁡σ[  ∣∥Aσ∥22d(N)−1−1∣≤δ  ]  =  1,\lim_{N \to \infty} \; \Pr_{\sigma}\left[\; \left| \frac{\|A_\sigma\|}{2\sqrt{2d(N) - 1}} - 1 \right| \le \delta \;\right] \;=\; 1,N→∞lim​σPr​[​22d(N)−1​∥Aσ​∥​−1​≤δ]=1,

where for each NNN the probability is over σ=(σ1,…,σd(N))\sigma = (\sigma_1, \dots, \sigma_{d(N)})σ=(σ1​,…,σd(N)​), a d(N)d(N)d(N)-tuple of permutations of the NNN-element set {0,1,…,N−1}\{0, 1, \dots, N-1\}{0,1,…,N−1}.

Spelled out: for every η>0\eta > 0η>0 there is N0N_0N0​ such that for all N≥N0N \ge N_0N≥N0​ the probability above lies within η\etaη of 111. The limit is the ordinary limit of a real sequence as N→∞N \to \inftyN→∞ through the natural numbers. The sequence ddd and the number δ\deltaδ are fixed before the limit is taken; nothing restricts how d(N)d(N)d(N) grows with NNN (it may be constant, bounded, or grow arbitrarily fast), beyond d(N)≥1d(N) \ge 1d(N)≥1.

Here d(N)d(N)d(N) is read as a real number and 2d(N)−12d(N) - 12d(N)−1 is ordinary real subtraction; since d(N)≥1d(N) \ge 1d(N)≥1, 2d(N)−1≥12d(N) - 1 \ge 12d(N)−1≥1 and the denominator 22d(N)−1≥22\sqrt{2d(N)-1} \ge 222d(N)−1​≥2 is positive. ∣⋅∣|\cdot|∣⋅∣ is the real absolute value, and the event uses the non-strict inequality ≤δ\le \delta≤δ. Equivalently, the event is

(1−δ) 22d(N)−1  ≤  ∥Aσ∥  ≤  (1+δ) 22d(N)−1.(1 - \delta)\, 2\sqrt{2d(N)-1} \;\le\; \|A_\sigma\| \;\le\; (1 + \delta)\, 2\sqrt{2d(N)-1}.(1−δ)22d(N)−1​≤∥Aσ​∥≤(1+δ)22d(N)−1​.

The probability

For each NNN, Pr⁡σ\Pr_\sigmaPrσ​ is the uniform probability on the finite set of all (N!)d(N)(N!)^{d(N)}(N!)d(N) tuples (σ1,…,σd(N))(\sigma_1, \dots, \sigma_{d(N)})(σ1​,…,σd(N)​): the number of tuples for which the event holds divided by (N!)d(N)(N!)^{d(N)}(N!)d(N). Equivalently, the σi\sigma_iσi​ are independent uniformly random permutations.

The operator AσA_\sigmaAσ​ and its norm

Give CN\mathbb{C}^NCN the standard Euclidean (Hilbert) norm ∥x∥=(∑j∣xj∣2)1/2\|x\| = \big(\sum_j |x_j|^2\big)^{1/2}∥x∥=(∑j​∣xj​∣2)1/2 and let

VN={ x∈CN:⟨1,x⟩=0 }={x:∑jxj=0},V_N = \{\, x \in \mathbb{C}^N : \langle \mathbf 1, x \rangle = 0 \,\} = \Big\{ x : \textstyle\sum_j x_j = 0 \Big\},VN​={x∈CN:⟨1,x⟩=0}={x:∑j​xj​=0},

the orthogonal complement of the all-ones vector 1\mathbf 11, with the inherited norm.

For a permutation π\piπ, let PπP_\piPπ​ be the operator on CN\mathbb{C}^NCN with (Pπx)j=xπ−1(j)(P_\pi x)_j = x_{\pi^{-1}(j)}(Pπ​x)j​=xπ−1(j)​ (so Pπei=eπ(i)P_\pi e_i = e_{\pi(i)}Pπ​ei​=eπ(i)​). It fixes 1\mathbf 11 and preserves inner products, hence maps VNV_NVN​ into itself; let Sπ:VN→VNS_\pi : V_N \to V_NSπ​:VN​→VN​ be its restriction and Sπ∗S_\pi^*Sπ∗​ the Hilbert-space adjoint of SπS_\piSπ​ as an operator on VNV_NVN​ (equal to the restriction of Pπ−1P_{\pi^{-1}}Pπ−1​). Then

Aσ=∑i=1d(N)(Sσi+Sσi∗):VN→VN,A_\sigma = \sum_{i=1}^{d(N)} \big( S_{\sigma_i} + S_{\sigma_i}^* \big) : V_N \to V_N,Aσ​=i=1∑d(N)​(Sσi​​+Sσi​∗​):VN​→VN​,

and ∥Aσ∥\|A_\sigma\|∥Aσ​∥ is its operator norm on VNV_NVN​:

∥Aσ∥=inf⁡{ c≥0:∥Aσx∥≤c ∥x∥ for all x∈VN }.\|A_\sigma\| = \inf\{\, c \ge 0 : \|A_\sigma x\| \le c\,\|x\| \ \text{for all } x \in V_N \,\}.∥Aσ​∥=inf{c≥0:∥Aσ​x∥≤c∥x∥ for all x∈VN​}.

Edge cases

  • Only large NNN matter for the limit. The sequence is nonetheless defined for every NNN, including N=0N = 0N=0 and N=1N = 1N=1: there VN={0}V_N = \{0\}VN​={0}, so ∥Aσ∥=0\|A_\sigma\| = 0∥Aσ​∥=0 and the event becomes 1≤δ1 \le \delta1≤δ.
  • δ\deltaδ may be any positive real; for δ≥1\delta \ge 1δ≥1 the lower half of the two-sided event, (1−δ)22d(N)−1≤∥Aσ∥(1-\delta)2\sqrt{2d(N)-1} \le \|A_\sigma\|(1−δ)22d(N)−1​≤∥Aσ​∥, is automatic.
  • The hypothesis d(N)≥1d(N) \ge 1d(N)≥1 is required for every NNN, not only for large NNN.
Human review
  • Endorsed by Shuze Chen · Nov 4, 2026

    Confirmed by the moderator at approval.

  • Endorsed by dbenbenn · Nov 4, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me