Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Promotion and covariance of sums and outer products

Proved
QAlgorithms.Papers.Independent.Intakenext100.P94.promotion_aggregation

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

prove2quantum

Let χ:[m]↪[n]\chi:[m]\hookrightarrow[n]χ:[m]↪[n] be the injective child-to-parent correspondence and let FFF be an order-kkk child tensor. Define promotion by embedding each tensor index along χ\chiχ and assigning zero off the image. For parent permutation σ\sigmaσ and child permutation π\piπ, relabel the correspondence as χ′=σ∘χ∘π−1\chi'=\sigma\circ\chi\circ\pi^{-1}χ′=σ∘χ∘π−1. Then

promote⁡χ′(πF)=σ(promote⁡χF).\operatorname{promote}_{\chi'}(\pi F)=\sigma\bigl(\operatorname{promote}_{\chi}F\bigr).promoteχ′​(πF)=σ(promoteχ​F).

Finite sums of promoted activations transform covariantly, and the outer product of tensors of orders k,lk,lk,l transforms covariantly as a tensor of order k+lk+lk+l:

∑cσFc=σ ⁣(∑cFc),(σF)⊗(σG)=σ(F⊗G).\sum_c\sigma F_c=\sigma\!\left(\sum_cF_c\right),\qquad (\sigma F)\otimes(\sigma G)=\sigma(F\otimes G).c∑​σFc​=σ(c∑​Fc​),(σF)⊗(σG)=σ(F⊗G).

These operations provide the source’s sum and product aggregation rules on the parent domain. The correspondence must move with the relabelling. The zero extension fixes how contributions from overlapping child domains are combined.

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

/-- Promotion and covariant sums/products, source PDF pp.5–6, Eq.(6).
Parent and child relabellings move chi to s∘chi∘p⁻¹. Empty domains/families
use the same finite-sum zero-extension convention. -/
theorem promotion_aggregation :
    (∀ {m n k : ℕ} (chi : Fin m ↪ Fin n) (F : Tensor m k) u,
      promote chi F (fun a => chi (u a)) = F u) ∧
    (∀ {m n k : ℕ} (chi : Fin m ↪ Fin n) (s : Perm n) (p : Perm m) (F : Tensor m k),
      promote (relabelEmbedding chi s p) (permutationTensorAction p F) =
        permutationTensorAction s (promote chi F)) ∧
    (∀ {n k : ℕ} {C : Type*} [Fintype C] (s : Perm n) (F : C → Tensor n k),
      (∑ a, permutationTensorAction s (F a)) = permutationTensorAction s (∑ a, F a)) ∧
    (∀ {n k l : ℕ} (s : Perm n) (F : Tensor n k) (G : Tensor n l) t,
      tensorOuterProduct (permutationTensorAction s F) (permutationTensorAction s G) t =
        tensorOuterProduct F G (fun a => s.symm (t a))) := 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. 6, §5, unnumbered: promotion and covariant sums/products
Read-back

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

Writing [r]={0,…,r−1}[r]=\{0,\ldots,r-1\}[r]={0,…,r−1}, a real order-kkk tensor on [n][n][n] is an arbitrary function F:([k]→[n])→RF:([k]\to[n])\to\mathbb RF:([k]→[n])→R; a permutation sss acts by AsF(t)=F(s−1∘t)A_sF(t)=F(s^{-1}\circ t)As​F(t)=F(s−1∘t); and promotion along an injection χ:[m]↪[n]\chi:[m]\hookrightarrow[n]χ:[m]↪[n] is defined by promote⁡χF(t)=∑v:[k]→[m], χ∘v=tF(v)\operatorname{promote}_{\chi}F(t)=\sum_{v:[k]\to[m],\,\chi\circ v=t}F(v)promoteχ​F(t)=∑v:[k]→[m],χ∘v=t​F(v), with an empty sum equal to 000. The theorem asserts the conjunction of four universally quantified identities. First, for all natural numbers m,n,km,n,km,n,k, injections χ:[m]↪[n]\chi:[m]\hookrightarrow[n]χ:[m]↪[n], tensors F:([k]→[m])→RF:([k]\to[m])\to\mathbb RF:([k]→[m])→R, and tuples u:[k]→[m]u:[k]\to[m]u:[k]→[m], one has promote⁡χF(χ∘u)=F(u)\operatorname{promote}_{\chi}F(\chi\circ u)=F(u)promoteχ​F(χ∘u)=F(u). Second, for all such m,n,k,χ,Fm,n,k,\chi,Fm,n,k,χ,F, every permutation sss of [n][n][n], and every permutation ppp of [m][m][m], one has

promote⁡ s∘χ∘p−1(ApF)=As(promote⁡χF).\operatorname{promote}_{\,s\circ\chi\circ p^{-1}}(A_pF) =A_s(\operatorname{promote}_{\chi}F).promotes∘χ∘p−1​(Ap​F)=As​(promoteχ​F).

Third, for all natural numbers n,kn,kn,k, every type CCC equipped with a finite enumeration, every permutation sss of [n][n][n], and every family of tensors FaF_aFa​ indexed by a∈Ca\in Ca∈C, one has ∑a∈CAsFa=As(∑a∈CFa)\sum_{a\in C}A_sF_a=A_s(\sum_{a\in C}F_a)∑a∈C​As​Fa​=As​(∑a∈C​Fa​). Fourth, for all natural numbers n,k,ln,k,ln,k,l, permutations sss of [n][n][n], tensors F:([k]→[n])→RF:([k]\to[n])\to\mathbb RF:([k]→[n])→R and G:([l]→[n])→RG:([l]\to[n])\to\mathbb RG:([l]→[n])→R, and tuples t:[k]⊔[l]→[n]t:[k]\sqcup[l]\to[n]t:[k]⊔[l]→[n], one has ((AsF)⊠(AsG))(t)=(F⊠G)(s−1∘t)((A_sF)\boxtimes(A_sG))(t)=(F\boxtimes G)(s^{-1}\circ t)((As​F)⊠(As​G))(t)=(F⊠G)(s−1∘t), where (F⊠G)(t)=F(a↦t(inl⁡(a)))G(b↦t(inr⁡(b)))(F\boxtimes G)(t)=F(a\mapsto t(\operatorname{inl}(a)))G(b\mapsto t(\operatorname{inr}(b)))(F⊠G)(t)=F(a↦t(inl(a)))G(b↦t(inr(b))). All natural-number parameters may be zero, subject to the existence of the specified injections and tuples; zero-length tuples are unique, and the third identity also includes an empty indexing type CCC, in which case both sides are the zero tensor.

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