Exchangeability, density-or-PMF affiliation, and threshold events
DefinitionMinBuying_SymmetryDefines exchangeability as invariance under every coordinate permutation. Affiliation is pointwise MTP2 of either a measurable nonnegative joint Lebesgue density or a probability mass function on real-valued vectors. This includes countably supported laws and does not replace affiliation by association or by the desired conditional comparison. Threshold events refer to the maximum valuation and one specified coordinate.
import Definitions.Def_MinBuying_Model
/-! Symmetry and affiliation from Section 3 of the source. Affiliation is
the density-or-mass-function MTP2 condition itself; no conditional monotonicity
or pricing conclusion is included in the definition. -/
noncomputable section
open MeasureTheory Set
namespace MinBuying
def permuteCoordinates {n : ℕ} (σ : Equiv.Perm (Fin n))
(v : Valuations n) : Valuations n := fun i => v (σ i)
def Exchangeable {n : ℕ} (μ : JointLaw n) : Prop :=
∀ σ : Equiv.Perm (Fin n), Measure.map (permuteCoordinates σ) μ = μ
/-- Pointwise log-supermodularity, without requiring strictly positive mass. -/
def MTP2 {n : ℕ} (f : Valuations n → ℝ) : Prop :=
∀ v w : Valuations n,
f v * f w ≤
f (fun i => max (v i) (w i)) * f (fun i => min (v i) (w i))
/-- Definition 1 permits either a joint density or a probability mass function.
The PMF alternative includes countably supported laws, with no artificial
assumption that valuations are integer-valued or have a finite support. -/
def Affiliated {n : ℕ} (μ : JointLaw n) : Prop :=
(∃ f : Valuations n → ℝ,
Measurable f ∧ (∀ v, 0 ≤ f v) ∧ MTP2 f ∧
μ = volume.withDensity (fun v => ENNReal.ofReal (f v))) ∨
(∃ w : PMF (Valuations n),
μ = w.toMeasure ∧ MTP2 (fun v => (w v).toReal))
/-- The maximum valuation exceeds the given threshold, expressed without
choosing a default maximum for an empty item set. -/
def upperMaximumEvent {n : ℕ} (t : ℝ) : Set (Valuations n) :=
{v | ∃ j : Fin n, t ≤ v j}
def coordinateBelowAndUpperMaximumEvent {n : ℕ}
(i : Fin n) (a t : ℝ) : Set (Valuations n) :=
{v | v i < a ∧ v ∈ upperMaximumEvent t}
end MinBuying
Read-back
What the Lean code literally says, in plain math · gpt-6-astra
For every natural number , including , valuations are vectors in indexed by , and a joint law in these definitions is an arbitrary measure on this space. For a permutation of the indices, coordinate permutation sends to the vector with th coordinate . A measure is defined to be exchangeable exactly when its image under every such coordinate permutation equals . For an arbitrary real function on , the predicate called MTP2 is the assertion for all , where and are coordinatewise maximum and minimum; this predicate by itself imposes neither nonnegativity nor measurability. A measure is defined to be affiliated exactly when at least one of two alternatives holds: there is a Borel-measurable function that is nonnegative at every point, satisfies the preceding product inequality at every pair of points, and gives as Lebesgue measure with density ; or there is a probability mass function on such that is its associated measure and its real point-mass function satisfies that same product inequality at every pair of points. The mass function may have countably infinite support. The density alternative imposes no requirement that the density integrate to , or even have finite integral; the mass-function alternative is a probability measure. For any real , the upper-maximum event is . For an index and arbitrary real , the coordinate-below-and-upper-maximum event is . At , coordinate permutation invariance holds automatically, is empty for every , and there is no index with which to instantiate .
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.