Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Multiplicities of first- and second-order activation spaces

Open
QAlgorithms.Papers.Independent.Intakenext100.P94.activation_multiplicities

by Gohan · Oct 31, 2026 · Mathlib 0df444a (Lean v4.33.1)

prove2quantum

Use the real Specht models indexed by partitions in the paper’s order. For n≥2n\ge2n≥2, the first-order permutation representation has multiplicities 1,11,11,1 on (n),(n−1,1)(n),(n-1,1)(n),(n−1,1). For n≥4n\ge4n≥4, use the four shapes (n),(n−1,1),(n−2,2),(n−2,1,1)(n),(n-1,1),(n-2,2),(n-2,1,1)(n),(n−1,1),(n−2,2),(n−2,1,1). The ordered off-diagonal and full matrix representations have types

τoff=(1,2,1,1),τfull=(2,3,1,1).\tau_{\rm off}=(1,2,1,1),\qquad \tau_{\rm full}=(2,3,1,1).τoff​=(1,2,1,1),τfull​=(2,3,1,1).

All other multiplicities are zero. Here a type means the existence of an actual invertible equivariant real linear identification with the corresponding copies of the concrete Specht modules, not just equality of dimensions. These decompositions explain the parameter counts of the second-order layers.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Intakenext100_P94
Formal statement
noncomputable section

namespace QAlgorithms.Papers.Independent.Intakenext100.P94

open scoped BigOperators

attribute [local instance] Classical.propDecidable

/-- Unnumbered activation multiplicities, source PDF p.4 and p.15.
First-order range n≥2; the four-shape stable range is n≥4. Trailing copies are zero. -/
theorem activation_multiplicities :
    (∀ (n : ℕ), 2 ≤ n →
      ∀ lambda : Fin 2 → QAlgorithms.Schur.Partition n n,
        (∀ a, (lambda a).val.rowLens = (![[n], [n-1,1]] : Fin 2 → List ℕ) a) →
        HasTensorType (⊤ : Submodule ℝ (Tensor n 1)) lambda ![1,1]) ∧
    (∀ (n : ℕ), 4 ≤ n →
      ∀ lambda : Fin 4 → QAlgorithms.Schur.Partition n n,
        (∀ a, (lambda a).val.rowLens = secondShapes n a) →
        HasTensorType (DistinctSpace n 2) lambda ![1,2,1,1] ∧
        HasTensorType (⊤ : Submodule ℝ (Tensor n 2)) lambda ![2,3,1,1]) := by sorry

end QAlgorithms.Papers.Independent.Intakenext100.P94

end
Source
The general theory of permutation equivarant neural networks and higher order graph variational encoders, arXiv:2004.03990v1, PDF p. 4, §4 and §10.1, unnumbered: activation multiplicities
Read-back

What the Lean code literally says, in plain math · gpt-6.1-sol

Write [m]={0,…,m−1}[m]=\{0,\ldots,m-1\}[m]={0,…,m−1} and let Tn,kT_{n,k}Tn,k​ be the real vector space of all functions F:[n]k→RF:[n]^k\to\mathbb RF:[n]k→R, where a tuple in [n]k[n]^k[n]k is a function [k]→[n][k]\to[n][k]→[n]. Permutations sss of [n][n][n] act by (AsF)(t)=F(s−1∘t)(A_sF)(t)=F(s^{-1}\circ t)(As​F)(t)=F(s−1∘t). For a Young diagram λ\lambdaλ with nnn cells and first-column length at most nnn, fix the cell numbering by [n][n][n] used in the definition, let RλR_\lambdaRλ​ and CλC_\lambdaCλ​ be the permutations preserving, respectively, the row and column of every numbered cell, and let Vλ⊆RSnV_\lambda\subseteq\mathbb R^{S_n}Vλ​⊆RSn​ be the range of the matrix

Yλ(g,h)=∑a∈Rλ∑b∈Cλ{sgn⁡(b),g=hab,0,g≠hab.Y_\lambda(g,h)=\sum_{a\in R_\lambda}\sum_{b\in C_\lambda} \begin{cases}\operatorname{sgn}(b),&g=hab,\\0,&g\ne hab.\end{cases}Yλ​(g,h)=a∈Rλ​∑​b∈Cλ​∑​{sgn(b),0,​g=hab,g=hab.​

For a subspace S⊆Tn,kS\subseteq T_{n,k}S⊆Tn,k​, a family (λa)a∈[r](\lambda_a)_{a\in[r]}(λa​)a∈[r]​, and multiplicities (τa)a∈[r](\tau_a)_{a\in[r]}(τa​)a∈[r]​, the decomposition assertion below means that there exists a real-linear isomorphism U:S→∏a∈[r]VλaτaU:S\to\prod_{a\in[r]}V_{\lambda_a}^{\tau_a}U:S→∏a∈[r]​Vλa​τa​​ such that, for every s∈Sns\in S_ns∈Sn​ and F∈SF\in SF∈S, there exists G∈SG\in SG∈S satisfying G(t)=F(s−1∘t)G(t)=F(s^{-1}\circ t)G(t)=F(s−1∘t) for every ttt, and (UG)a,c(g)=(UF)a,c(s−1g)(UG)_{a,c}(g)=(UF)_{a,c}(s^{-1}g)(UG)a,c​(g)=(UF)a,c​(s−1g) for every a∈[r]a\in[r]a∈[r], c∈[τa]c\in[\tau_a]c∈[τa​], and g∈Sng\in S_ng∈Sn​. The theorem asserts both of the following: for every natural number n≥2n\ge2n≥2 and every family λ0,λ1\lambda_0,\lambda_1λ0​,λ1​ of such Young diagrams whose row-length lists are respectively [n][n][n] and [n−1,1][n-1,1][n−1,1], this decomposition assertion holds for S=Tn,1S=T_{n,1}S=Tn,1​ with multiplicities (1,1)(1,1)(1,1); and, for every natural number n≥4n\ge4n≥4 and every family λ0,…,λ3\lambda_0,\ldots,\lambda_3λ0​,…,λ3​ whose row-length lists are respectively [n][n][n], [n−1,1][n-1,1][n−1,1], [n−2,2][n-2,2][n−2,2], and [n−2,1,1][n-2,1,1][n−2,1,1], it holds both for the subspace of Tn,2T_{n,2}Tn,2​ consisting of functions that vanish whenever the two tuple entries coincide, with multiplicities (1,2,1,1)(1,2,1,1)(1,2,1,1), and for all of Tn,2T_{n,2}Tn,2​, with multiplicities (2,3,1,1)(2,3,1,1)(2,3,1,1).

Human review
  • Endorsed by Shuze Chen · Oct 31, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Gohan · Oct 31, 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