Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Deep linear network: exact four-point correlator at initialization (eqs. 3.21, 3.25)

Proved
DeepLearningTheory.linearNet_four_point

by LG · Sep 23, 2026 · Mathlib 0df444a (Lean v4.33.1)

deep-learning-theorymachine-learningprobability

Exact four-point correlator of a deep linear network. Consider a zero-bias deep linear network with positive widths n0,n1,…n_0,n_1,\dotsn0​,n1​,… whose weights are initialized independently with Wij(ℓ)∼N(0,CW/nℓ−1)W^{(\ell)}_{ij}\sim\mathcal{N}(0,C_W/n_{\ell-1})Wij(ℓ)​∼N(0,CW​/nℓ−1​) (eq. 3.4). For a single input x∈Rn0x\in\mathbb{R}^{n_0}x∈Rn0​, every layer ℓ≥1\ell\ge1ℓ≥1 and neurons i1,…,i4i_1,\dots,i_4i1​,…,i4​ of that layer,

E[zi1(ℓ)zi2(ℓ)zi3(ℓ)zi4(ℓ)]=(δi1i2δi3i4+δi1i3δi2i4+δi1i4δi2i3)  CW2ℓ[∏ℓ′=1ℓ−1(1+2nℓ′)](G2(0))2,\mathbb{E}\big[z^{(\ell)}_{i_1}z^{(\ell)}_{i_2}z^{(\ell)}_{i_3}z^{(\ell)}_{i_4}\big]=(\delta_{i_1i_2}\delta_{i_3i_4}+\delta_{i_1i_3}\delta_{i_2i_4}+\delta_{i_1i_4}\delta_{i_2i_3})\;C_W^{2\ell}\Big[\prod_{\ell'=1}^{\ell-1}\Big(1+\frac{2}{n_{\ell'}}\Big)\Big]\big(G^{(0)}_2\big)^2,E[zi1​(ℓ)​zi2​(ℓ)​zi3​(ℓ)​zi4​(ℓ)​]=(δi1​i2​​δi3​i4​​+δi1​i3​​δi2​i4​​+δi1​i4​​δi2​i3​​)CW2ℓ​[ℓ′=1∏ℓ−1​(1+nℓ′​2​)](G2(0)​)2,

where G2(0)=1n0∑j=1n0xj2G^{(0)}_2=\frac1{n_0}\sum_{j=1}^{n_0}x_j^2G2(0)​=n0​1​∑j=1n0​​xj2​.

In the infinite-width limit the bracket tends to 111 and the correlator becomes the Gaussian (Wick) value; at finite width the deviation grows with depth, controlled by the depth-to-width ratio.

Preamble
import Mathlib
import Definitions.Def_DLT_DeepLinearNetwork

open MeasureTheory ProbabilityTheory
open scoped NNReal
Formal statement
namespace DeepLearningTheory

theorem linearNet_four_point {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω)
    [IsProbabilityMeasure P] (n : ℕ → ℕ) (hn : ∀ ℓ, 0 < n ℓ) (CW : ℝ≥0)
    (W : Ω → ℕ → ℕ → ℕ → ℝ) (hW : IsLinearNetInit P n CW W) (x : ℕ → ℝ)
    (ℓ i₁ i₂ i₃ i₄ : ℕ) (hℓ : 1 ≤ ℓ)
    (hi₁ : i₁ < n ℓ) (hi₂ : i₂ < n ℓ) (hi₃ : i₃ < n ℓ) (hi₄ : i₄ < n ℓ) :
    ∫ ω, linearPreact n (W ω) x ℓ i₁ * linearPreact n (W ω) x ℓ i₂ *
        linearPreact n (W ω) x ℓ i₃ * linearPreact n (W ω) x ℓ i₄ ∂P
      = wickDelta4 i₁ i₂ i₃ i₄ *
          ((CW : ℝ) ^ (2 * ℓ) * (∏ ℓ' ∈ Finset.Ico 1 ℓ, (1 + 2 / (n ℓ' : ℝ))) *
            inputKernel (n 0) x x ^ 2) := by sorry

end DeepLearningTheory
Source
Daniel A. Roberts and Sho Yaida (with Boris Hanin), *The Principles of Deep Learning Theory*, arXiv:2106.10165v2, https://arxiv.org/abs/2106.10165, §3.3, p. 61, eqs. (3.21), (3.22), (3.25) (page numbers are the book's printed page numbers)
Read-back

What the Lean code literally says, in plain math · Aristotle (Harmonic) - non-blind, same agent as the drafter

Non-blind read-back. This read-back was written by the same agent that drafted the Lean statement, with full knowledge of the source text and of the intended meaning. It is not independent testimony and must not be mistaken for a blind audit; reviewers should compare the Lean code against the source themselves.

Let (Ω,P)(\Omega,P)(Ω,P) be a probability space, n:N→Nn:\mathbb{N}\to\mathbb{N}n:N→N widths with nℓ>0n_\ell>0nℓ​>0 for all ℓ\ellℓ, CW≥0C_W\ge0CW​≥0, and WWW a random weight field satisfying IsLinearNetInit (the weights Wij(ℓ)W^{(\ell)}_{ij}Wij(ℓ)​, ℓ≥1\ell\ge1ℓ≥1, i<nℓi<n_\elli<nℓ​, j<nℓ−1j<n_{\ell-1}j<nℓ−1​, mutually independent with laws N(0,CW/nℓ−1)\mathcal{N}(0,C_W/n_{\ell-1})N(0,CW​/nℓ−1​)). For every input x:N→Rx:\mathbb{N}\to\mathbb{R}x:N→R, every ℓ≥1\ell\ge1ℓ≥1 and all i1,i2,i3,i4<nℓi_1,i_2,i_3,i_4<n_\elli1​,i2​,i3​,i4​<nℓ​:

∫∏a=14zia(ℓ)(x) dP=Δi1i2i3i4⋅(CW2ℓ∏ℓ′=1ℓ−1(1+2nℓ′)(1n0∑j<n0x(j)2)2),\int\prod_{a=1}^{4}z^{(\ell)}_{i_a}(x)\,dP=\Delta_{i_1i_2i_3i_4}\cdot\Big(C_W^{2\ell}\prod_{\ell'=1}^{\ell-1}\Big(1+\frac{2}{n_{\ell'}}\Big)\Big(\frac{1}{n_0}\sum_{j<n_0}x(j)^2\Big)^2\Big),∫a=1∏4​zia​(ℓ)​(x)dP=Δi1​i2​i3​i4​​⋅(CW2ℓ​ℓ′=1∏ℓ−1​(1+nℓ′​2​)(n0​1​j<n0​∑​x(j)2)2),

where Δi1i2i3i4=δi1i2δi3i4+δi1i3δi2i4+δi1i4δi2i3\Delta_{i_1i_2i_3i_4}=\delta_{i_1i_2}\delta_{i_3i_4}+\delta_{i_1i_3}\delta_{i_2i_4}+\delta_{i_1i_4}\delta_{i_2i_3}Δi1​i2​i3​i4​​=δi1​i2​​δi3​i4​​+δi1​i3​​δi2​i4​​+δi1​i4​​δi2​i3​​, z(0)=xz^{(0)}=xz(0)=x, zi(ℓ+1)=∑j<nℓWij(ℓ+1)zj(ℓ)z^{(\ell+1)}_i=\sum_{j<n_\ell}W^{(\ell+1)}_{ij}z^{(\ell)}_jzi(ℓ+1)​=∑j<nℓ​​Wij(ℓ+1)​zj(ℓ)​, and the integral is a Bochner integral.

Human review
  • Endorsed by Shuze Chen · Sep 24, 2026

    Confirmed by the moderator at approval.

  • Endorsed by LG · Sep 24, 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