Skip to content

Proposal: finite fiber-average conditional expectation API #494

@AlyciaBHZ

Description

@AlyciaBHZ

Finite fiber-average conditional expectation: combinatorial API proposal

We propose a self-contained combinatorial API for finite conditional expectation by fiber averaging, filling a gap between Mathlib's abstract measure-theoretic condExp and explicit finite-sum computation.

Mathlib's condExp provides the standard a.e. laws (condExp_const, condExp_of_aestronglyMeasurable', condExp_condExp_of_le, pull-out lemmas) but does not identify condExp for counting measure on a finite quotient with the explicit pointwise fiber-average formula. The API below supplies this concrete bridge.

Definitions

Given finite types Ω, X, and a surjective map π : Ω → X:

  • fiberAverage f a = (∑ b ∈ fiber(π(a)), f b) / |fiber(π(a))|
  • finiteMean f = (∑ a, f a) / |Ω|
  • finiteVariance f = ∑ a, f(a)² − |Ω| · (finiteMean f)²
  • FiberInvariant f iff f is constant on each fiber

Proved properties (over any ordered field 𝕜)

Operator-algebraic package:

  1. Positivity: (∀ a, 0 ≤ f a) → ∀ a, 0 ≤ fiberAverage f a
  2. Unitality: fiberAverage 1 = 1
  3. Idempotence: fiberAverage (fiberAverage f) = fiberAverage f
  4. Identity on invariants: FiberInvariant f → fiberAverage f = f
  5. Bimodule law: fiberAverage (B₁ · A · B₂) = B₁ · fiberAverage(A) · B₂ for fiber-invariant B₁, B₂

Analytic package:
6. Mean preservation: finiteMean (fiberAverage f) = finiteMean f
7. L² contraction: ∑ a, (fiberAverage f a)² ≤ ∑ a, f(a)²
8. Variance dissipation: finiteVariance (fiberAverage f) ≤ finiteVariance f
9. One-atom annihilation: [Subsingleton X] → finiteVariance (fiberAverage f) = 0

All proofs use only finite sums, sq_sum_le_card_mul_sum_sq, and field_simp/ring. The file has no sorry, admit, or axiom, and typechecks against Lean 4.29.0-rc8 + Mathlib v4.29.0-rc8 (the versions pinned by this repository).

Non-redundancy with Mathlib

Mathlib's MeasureTheory.Function.ConditionalExpectation provides the five operator-algebraic laws at the abstract a.e. level (with integrability/measurability hypotheses). It does not contain a theorem identifying condExp for counting measure and a comap sigma-algebra with the explicit pointwise fiber-average formula above. The analytic package (L² contraction, variance dissipation) is also absent in this finite combinatorial form.

A prototype Lean file (332 lines) is available and builds cleanly as Analysis.Misc.FiniteConditionalExpectation in a local checkout of this repo. If you think this fits the scope of the project, we would be glad to submit a PR adapted to the repository's style and review conventions.

Repo: https://github.com/the-omega-institute/automath

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions