Second-order layers on ordered distinct pairs
OpenQAlgorithms.Papers.Independent.Intakenext100.P94.second_order_zero_diagonalLet . Let be a real linear endomorphism of zero-diagonal matrices, with the simultaneous inverse permutation action on both indices, and fix a shared pointwise nonlinearity . No symmetry of the matrices is assumed. Write , and . Then is equivariant if and only if there are weights such that, for all inputs and all ,
The nonlinear layer applies to this expression on the ordered distinct-pair coordinates. This characterizes the linear map; independence of the seven weights is not claimed for .
import Definitions.Def_QAlgorithms_Papers_Independent_Intakenext100_P94
noncomputable section
namespace QAlgorithms.Papers.Independent.Intakenext100.P94
open scoped BigOperators
attribute [local instance] Classical.propDecidable
/-- Theorem 2, source PDF p.4 and pp.13–14. Output indices are distinct,
as in the proof. Ordered pairs, without symmetry; no independence claim for n<4. -/
theorem second_order_zero_diagonal (n : ℕ) (hn : 2 ≤ n) (xi : ℝ → ℝ)
(L : DistinctSpace n 2 →ₗ[ℝ] DistinctSpace n 2) :
(L ∈ equivariantLinearMaps (distinctAction (n := n) (k := 2)) distinctAction ↔
∃ w : Fin 7 → ℝ, ∀ F i j, i ≠ j →
(L F).val ![i,j] = sevenFormula w F.val i j) ∧
(L ∈ equivariantLinearMaps (distinctAction (n := n) (k := 2)) distinctAction →
∃ w : Fin 7 → ℝ, ∀ F i j, i ≠ j →
xi ((L F).val ![i,j]) = xi (sevenFormula w F.val i j)) := by sorry
end QAlgorithms.Papers.Independent.Intakenext100.P94
end
Read-back
What the Lean code literally says, in plain math · gpt-6.1-sol
For every natural number , every function , and every real-linear map from the space of real arrays satisfying for every to that same space, define simultaneous permutation of the indices by . For seven real weights , put
where all summation indices range from to . The theorem asserts both that
and that equivariance of implies the existence of seven real weights for which for every such and every pair . Neither formula is required at diagonal pairs; the domain and codomain already require the arrays themselves to vanish there. No condition is imposed on , no uniqueness of weights is asserted, and is included.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.