Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proof of Theorem 4.4, p. 8 — ∂f^(2)/∂A^(n) = T̄_(n)·V^{n} and ∂f̄^(2)/∂A^(n) = ∂f^(3)/∂A^(n) = 0

Proved
TensorBTD.Cogradient.cograd_f2_f3

by mikedeng1 · Oct 9, 2026 · Mathlib 0df444a (Lean v4.33.1)

khatri-rao-productp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1tensor-unfoldingwirtinger-derivative

Consider the unstructured CPD with factor matrices A(1),…,A(N)∈CIn×R′A^{(1)},\dots,A^{(N)}\in\mathbb C^{I_n\times R'}A(1),…,A(N)∈CIn​×R′, a tensor T\mathcal TT and a mode nnn. Let

f(2)=⟨T(n),A(n)V{n}T⟩,f(3)=∥T(n)∥2.f^{(2)}=\langle T_{(n)},A^{(n)}V^{\{n\}\mathrm T}\rangle,\qquad f^{(3)}=\|T_{(n)}\|^2 .f(2)=⟨T(n)​,A(n)V{n}T⟩,f(3)=∥T(n)​∥2.

Then, for the cogradients with respect to the entries of A(n)A^{(n)}A(n),

  1. ∂f(2)∂A(n)=T‾(n)⋅V{n}\dfrac{\partial f^{(2)}}{\partial A^{(n)}}=\overline{T}_{(n)}\cdot V^{\{n\}}∂A(n)∂f(2)​=T(n)​⋅V{n}, where T‾(n)\overline{T}_{(n)}T(n)​ is the entrywise conjugate of the mode-nnn unfolding;
  2. ∂f(2)‾∂A(n)=0\dfrac{\partial \overline{f^{(2)}}}{\partial A^{(n)}}=0∂A(n)∂f(2)​​=0;
  3. ∂f(3)∂A(n)=0\dfrac{\partial f^{(3)}}{\partial A^{(n)}}=0∂A(n)∂f(3)​=0.

Together with ∂f(1)/∂A(n)=A‾(n)W{n}\partial f^{(1)}/\partial A^{(n)}=\overline{A}^{(n)}W^{\{n\}}∂f(1)/∂A(n)=A(n)W{n} and the decomposition f=12(f(1)−f(2)−f(2)‾+f(3))f=\frac12(f^{(1)}-f^{(2)}-\overline{f^{(2)}}+f^{(3)})f=21​(f(1)−f(2)−f(2)​+f(3)) this gives (4.5a).

Formalization Note. The inner product is conjugate on the first argument (Definition 2.1), so f(2)f^{(2)}f(2) is holomorphic in A(n)A^{(n)}A(n) and f(2)‾\overline{f^{(2)}}f(2)​ is antiholomorphic. The mode nnn may be any of the NNN modes.

Preamble
import Mathlib
import Definitions.Def_TensorBTD_Cogradient_Setting
Formal statement
namespace TensorBTD.Cogradient

open Matrix

/-- Proof of Theorem 4.4, p. 8: for every mode `n` of the unstructured CPD,
`∂f^(2)/∂A^(n) = T̄_(n) · V^{n}` and `∂ conj(f^(2))/∂A^(n) = ∂f^(3)/∂A^(n) = 0`. -/
theorem cograd_f2_f3 {P Q R : ℕ} {I : Fin P ⊕ Fin Q → ℕ} {L : Fin R → ℕ} (T : TensorBTD.Gramian.Tensor I)
    (x : TensorBTD.Gramian.GIdx I L → ℂ) (n : Fin P ⊕ Fin Q) :
    cogradMatG (fTwo T n) x n = (unfolding T n).map (starRingEnd ℂ) * V {n} (TensorBTD.Gramian.cpdFactor x) ∧
    cogradMatG (fun y => starRingEnd ℂ (fTwo T n y)) x n = 0 ∧
    cogradMatG (fun _ : TensorBTD.Gramian.GIdx I L → ℂ => fThree T n) x n = 0 := by sorry

end TensorBTD.Cogradient
Source
Sorber, Van Barel & De Lathauwer, Optimization-Based Algorithms for Tensor Decompositions, SIAM J. Optim. 23(2) (2013), doi:10.1137/120868323, authors' manuscript (Lirias), p. 8, proof of Theorem 4.4, sentence after the first display
Human review
  • Endorsed by Shuze Chen · Oct 9, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 9, 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