Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proposition 4.3: linear images of Gaussian vectors are Gaussian

Proved
LesHouchesWidth.gaussian_linear_image

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

machine-learningneural-networksprobability

Let W∼N(μ,Σ)W\sim\mathcal N(\mu,\Sigma)W∼N(μ,Σ) be a Gaussian vector in Rd\mathbb R^dRd with mean μ\muμ and positive semidefinite covariance Σ\SigmaΣ, and let A∈Rk×dA\in\mathbb R^{k\times d}A∈Rk×d be a fixed matrix. Then

AW∼N(Aμ, AΣAT).AW\sim\mathcal N\big(A\mu,\ A\Sigma A^{T}\big).AW∼N(Aμ, AΣAT).

This is the step that makes the next layer of a network conditionally Gaussian given the previous one (Lemma 4.4).

Formalization Note The notes allow AAA to be random and independent of WWW. The formal statement treats a deterministic AAA, from which the independent case follows by conditioning.

Preamble
import Mathlib
import Definitions.Def_LesHouchesWidth_GaussianMLP
import Definitions.Def_LesHouchesWidth_FiniteWidth
Formal statement
namespace LesHouchesWidth

open MeasureTheory ProbabilityTheory

theorem gaussian_linear_image {k d : ℕ} (μ : EuclideanSpace ℝ (Fin d))
    (S : Matrix (Fin d) (Fin d) ℝ) (hS : S.PosSemidef) (A : Matrix (Fin k) (Fin d) ℝ) :
    (multivariateGaussian μ S).map (Matrix.toEuclideanLin A) =
      multivariateGaussian (Matrix.toEuclideanLin A μ) (A * S * A.transpose) := by sorry

end LesHouchesWidth
Source
Bahri, Hanin, Brossollet, Erba, Keup, Pacelli, Simon, *Les Houches Lectures on Deep Learning at Large & Infinite Width*, arXiv:2309.01592v3 (https://arxiv.org/abs/2309.01592), p. 34, Proposition 4.3 (Section 4.9.1).
Read-back

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

Non-blind read-back. This read-back was written by the same agent (Aristotle, by Harmonic) that drafted the Lean statement, with full knowledge of the source and of the intended meaning. It is not independent testimony and should not be treated as a blind audit; an independent read-back is still recommended before launch.

Let k,d∈Nk,d\in\mathbb Nk,d∈N, μ∈Rd\mu\in\mathbb R^dμ∈Rd (Euclidean space), SSS a real d×dd\times dd×d positive semidefinite matrix, and AAA a real k×dk\times dk×d matrix. The theorem asserts that the pushforward of Mathlib's multivariate Gaussian N(μ,S)\mathcal N(\mu,S)N(μ,S) on Rd\mathbb R^dRd under the linear map v↦Avv\mapsto Avv↦Av (into Rk\mathbb R^kRk) equals the multivariate Gaussian N(Aμ, ASAT)\mathcal N(A\mu,\,ASA^{T})N(Aμ,ASAT) on Rk\mathbb R^kRk. AAA is a fixed deterministic matrix, and no rank condition is imposed. ASATASA^{T}ASAT is automatically positive semidefinite.

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