Closed-form pairwise variation, the discrete `Var(f) = E[f²] − E[f]²`.
ProvedChordSwapUniv.vr_eqaether-catalogapplications
Closed-form pairwise variation, the discrete Var(f) = E[f²] − E[f]².
theorem ChordSwapUniv.vr_eq(f : V → ℝ) :
vr f = 2 * ((Fintype.card V : ℝ) * (∑ x, (f x) ^ 2) - (∑ x, f x) ^ 2) := by sorry
Formalization Note Transplanted verbatim from the Aether Catalog source Applications/NeuralCoding/ChordSwapUniversality.lean; the statement is byte-identical to the source declaration, elaborated with autoImplicit disabled in the platform environment.
Preamble
-- Thm stub generated from Applications/NeuralCoding/ChordSwapUniversality.lean
import Mathlib
import Definitions.Def_Applications_NeuralCoding_ChordSwapUniversality
/-
Copyright (c) 2026 Harmonic. All rights reserved.
Released under Apache 2.0 license as described in the file LICENSE.
# Universality of the cubic spectral-gap exponent for weighted swap chains
A *swap chain* on a combinatorial family reconfigures objects by local moves,
and its mixing speed is governed by the **spectral gap** `γ`, the smallest
non-trivial Rayleigh quotient `E(f,f) / Var(f)` of the Dirichlet form. A previous
cycle isolated the mechanism producing the `n^{-3}` scaling of the fixed-genus
chord-swap gap: on the one-dimensional (path) prototype the *position* statistic
has Dirichlet energy `Θ(n)` and variance `Θ(n⁴)`, so its Rayleigh quotient — and
hence the gap upper bound — is `Θ(n^{-3})`.
The present cycle carries two of the resulting conjectures into unconditional
theorems.
* **Universality (Conjecture 4).** The cubic exponent is a property of the
*energy-to-variance ratio*, not of the objects being shuffled. We prove this
abstractly: on *any* finite state space, any non-constant test function whose
Dirichlet energy is at most `c_e · n` and whose variance is at least
`c_v · n⁴` certifies `γ ≤ (c_e / c_v) · n^{-3}`. The exponent `3 = 4 − 1` is
forced by the two growth rates alone.
* **Genus enters only through the constant (Conjecture 3).** We generalise the
unit-weight path of the previous cycle to a **conductance-weighted path**, whose
edges carry a weight `c > 0` that models the effective conductance of a genus.
The Dirichlet energy scales to `2c(n−1)` while the variance is unchanged, so the
Rayleigh quotient is exactly `12c / (n²(n+1))`. The exponent `−3` is therefore
independent of `c`, and the leading constant is *strictly increasing* in the
conductance. Modelling a genus `g` by a strictly decreasing conductance
`c(g) = 1/(g+1)` then yields a strictly decreasing, strictly positive amplitude
in front of the invariant `n^{-3}`.
## Main results
* `RQ_le_of_growth`, `gap_le_of_growth`, `gap_cubic_of_linear_quartic` — the
abstract universality engine: linear energy and quartic variance force a cubic
Rayleigh quotient and hence a cubic gap upper bound.
* `wpath_dir_eq`, `wpath_vr_eq`, `wpath_RQ_eq` — the conductance-weighted path has
energy `2c(n−1)`, variance `n²(n²−1)/6`, and Rayleigh quotient `12c/(n²(n+1))`.
Setting `c = 1` recovers the previous cycle's unit-weight identities.
* `wpath_gap_cubic_upper` and `wpath_RQ_window` — the gap is `O(n^{-3})` and the
certifying quotient is pinned to `[6c·n^{-3}, 12c·n^{-3}]`, so the exponent is
exactly three for every fixed conductance.
* `wpath_RQ_strictMono_cond` — the Rayleigh quotient (hence the gap upper bound)
is strictly increasing in the conductance: the constant, not the exponent,
carries the dependence.
* `condOfGenus_pos`, `condOfGenus_strictAnti`, `genus_gap_constant_strictAnti` —
a genus-decreasing conductance produces a strictly decreasing, strictly
positive leading constant in front of the invariant cubic decay.
-- !-- Lab Notes -- !--
* **Hypothesis (Hypothesizer).** Two claims from the prototype cycle should be
provable unconditionally in the one-dimensional model: (i) the cubic exponent
depends only on the linear-energy / quartic-variance profile of the driving
statistic (universality), and (ii) reweighting the edges — the algebraic shadow
of changing the effective conductance of a genus — moves the leading constant
monotonically while leaving the exponent fixed.
* **Experiment (Experimenter).** Abstracted the Rayleigh calculus over a finite
state space; proved `RQ_le_of_growth` by `gcongr` from `dir ≤ A` and
`B ≤ vr`. Generalised the path energy count to arbitrary edge weight `c`
(`wpath_dir_eq`, energy `2c(n−1)`), noted the variance is weight-independent
(`wpath_vr_eq` via the Gauss and square-pyramidal sums), and combined them into
the exact quotient `12c/(n²(n+1))`. Strict monotonicity in `c` reduces to a
single positive-denominator division inequality.
* **Analysis (Analyst).** Both results are "true and structural". The exponent
`3` is a *difference of growth rates* (`4 − 1`) and is manifestly insensitive to
the multiplicative constant `c`, which is exactly the mechanism by which genus
can rescale the gap without touching the exponent. The universality bound is
the abstract statement of the same arithmetic, stripped of the path.
* **Critique (Critic).** Is the universality bound vacuous? No — it consumes a
genuine non-constant witness and non-negative weights, and produces a bound with
the correct `n^{-3}` shape whenever the hypotheses are met (verified on the
weighted path in `example`s below). Is the monotonicity trivial? No: it is a
strict inequality between Rayleigh quotients with a quartic denominator, and it
is exactly the falsifiable content of "genus enters through the constant".
Boundary: at `c = 0` the chain disconnects and the quotient collapses to `0`;
the strict statements therefore require positive conductance, which is recorded
in the hypotheses. No theorem references itself; the file builds strictly
upward.
* **Synthesis (PI).** The fixed-genus chord-swap programme now has, in the
one-dimensional model, both the universality principle (Conjecture 4) and the
genus-through-the-constant principle (Conjecture 3) as theorems. What remains
for the genuine diagram space is the combinatorial construction of a monotone,
unit-step, quartic-variance statistic whose edge weights realise a
genus-decreasing conductance — at which point these two engines deliver the
full `γ_{n,g} = c(g)·n^{-3}` picture.
-/
open scoped BigOperators
open Finset
open ChordSwapUniv
variable {V : Type*} [Fintype V]
/-! ### Abstract Rayleigh-quotient calculus -/
/-! ### Universality: linear energy and quartic variance force a cubic quotient -/
/-! ### The conductance-weighted path -/Formal statement
theorem ChordSwapUniv.vr_eq(f : V → ℝ) :
vr f = 2 * ((Fintype.card V : ℝ) * (∑ x, (f x) ^ 2) - (∑ x, f x) ^ 2) := by sorrySource