Avg pairwise
ProvedLogic.PhaseRoute.avg_pairwiseaether-cataloglogic
Formal statement of Logic.PhaseRoute.avg_pairwise from the Aether Catalog (Logic). The mathematical content is given by the Lean statement below; a human-readable write-up is pending.
theorem Logic.PhaseRoute.avg_pairwise(F H₂ H₃ : G × G → ℝ) :
avg (pairwise F H₂ H₃)
= ((∑ a : G, ∑ b : G, F (a, b)) + (∑ b : G, ∑ c : G, H₂ (b, c))
+ ∑ a : G, ∑ c : G, H₃ (a, c)) / ((Fintype.card G : ℝ) ^ 2) := by sorry
Formalization Note Transplanted verbatim from the Aether Catalog source Logic/PhaseRouteTripleAlignment.lean; the statement is byte-identical to the source declaration, elaborated with autoImplicit disabled in the platform environment.
Preamble
-- Thm stub generated from Logic/PhaseRouteTripleAlignment.lean
import Mathlib
import Definitions.Def_Logic_PhaseRouteAlignment
import Definitions.Def_Logic_PhaseRouteLeastSquares
import Definitions.Def_Logic_PhaseRouteTripleAlignment
/-
# Degree hierarchy: three-way alignment defeats every pairwise encoding
Two further structural results for the phase-encoding programme.
## 1. Window stability is automatic (`Reindex` section)
Every quantity of the finite-sample calculus (`avg`, `cov`, `varr`, `msse`,
`Rsq`) is invariant under an arbitrary relabelling `e : κ ≃ ι` of the sample
space. Consequently an *exactly zero* degree-1 effect transfers across windows
with ratio exactly `1`; a measured cross/same ratio different from `1` is
therefore evidence about the estimator, never about a genuine degree-1 signal.
## 2. The degree hierarchy does not stop at `2`
On `G × G × G` (`G` any finite additive commutative group; take `G = ZMod p`)
consider the **three-way alignment** target
`triAlign (a,b,c) = if a + b + c = 0 then 1 else 0`.
Then `cov_triAlign_pairwise_eq_zero` : *every* predictor built from arbitrary
functions of **pairs** of coordinates,
`pairwise F G H (a,b,c) = F (a,b) + G (b,c) + H (a,c)`,
has covariance exactly `0` with the target — so the entire degree-`≤2` layer,
interaction encodings included, is blind to it, while the target is trivially
degree-`3` measurable (`Rsq_triAlign_self_eq_one`).
This is the sharp prediction for the next experimental round: if the residual
excess is a `k`-way joint alignment, then encodings of degree `< k` must return
*exactly* zero population gain, no matter how many primes are dialled in.
-/
open Logic.PhaseRoute
open Finset
/-! ### Relabelling invariance: window stability of exact statements -/
variable {ι κ : Type*} [Fintype ι] [Fintype κ] [Nonempty ι] [Nonempty κ]
/-! ### Three-way alignment on a finite abelian group -/
variable {G : Type*} [Fintype G] [DecidableEq G] [AddCommGroup G]
/-! Lifting functions of two coordinates to the triple product. -/
omit [DecidableEq G] inFormal statement
theorem Logic.PhaseRoute.avg_pairwise(F H₂ H₃ : G × G → ℝ) :
avg (pairwise F H₂ H₃)
= ((∑ a : G, ∑ b : G, F (a, b)) + (∑ b : G, ∑ c : G, H₂ (b, c))
+ ∑ a : G, ∑ c : G, H₃ (a, c)) / ((Fintype.card G : ℝ) ^ 2) := by sorrySource