Quantile regularity and continuous regular instances
DefinitionMinBuying_RegularityDefines the lower generalized quantile by an infimum over nonnegative values. Revenue-quantile concavity is imposed on , equivalent for nonnegative curves to the zero-endpoint convention on . Continuous regular marginals are nonnegative probability laws absolutely continuous with respect to Lebesgue measure, with no strict positivity or smoothness condition on their densities. An instance has positive finite item count and the source's well-posed optimization conditions.
import Definitions.Def_MinBuying_Model
noncomputable section
open MeasureTheory Set
namespace MinBuying
/-- The lower generalized quantile at an interior probability. On the domain
used below, the defining set is nonempty and bounded below for a nonnegative
probability law. Values at u=0 and u=1 are not used for regularity. -/
def quantile (F : Measure ℝ) (u : ℝ) : ℝ :=
sInf {x : ℝ | 0 ≤ x ∧ u ≤ cdf F x}
def revenueQuantile (F : Measure ℝ) (q : ℝ) : ℝ :=
q * quantile F (1 - q)
/-- Interior concavity avoids the expression 0 times an infinite upper
quantile. For nonnegative curves it is equivalent to concavity on [0,1]
after assigning zero at both endpoints. -/
def Regular (F : Measure ℝ) : Prop :=
ConcaveOn ℝ (Ioo (0 : ℝ) 1) (revenueQuantile F)
/-- Absolute continuity is the source's standing meaning of continuous
valuation distributions; strict positivity or smoothness of density is not added. -/
def ContinuousRegular (F : Measure ℝ) : Prop :=
ValidMarginal F ∧ F ≪ volume ∧ Regular F
def RegularInstance {n : ℕ} (F : Fin n → Measure ℝ) : Prop :=
0 < n ∧ (∀ i, ContinuousRegular (F i)) ∧ WellPosed (productLaw F)
end MinBuying
Read-back
What the Lean code literally says, in plain math · gpt-6-astra
For every measure on and every real , define and . The conversion is the usual real value for finite nonnegative measures and is zero for ; the real infimum used here is zero for an empty set, and every nonempty set in this definition is bounded below by zero. Thus is a real-valued function for every real , even when the displayed set is empty; in particular for . Define the revenue-quantile function on all of by . A measure is called regular exactly when is concave on the open interval : for all and , . This condition tests neither endpoint and by itself assumes neither probability mass one, nonnegative support, nor absolute continuity. A measure is called continuously regular exactly when it has total mass one, is concentrated almost everywhere on , is absolutely continuous with respect to Lebesgue measure, and satisfies that concavity condition. For each natural number and each family of measures on , where , a regular instance means , continuous regularity of every , and the following condition on the finite product measure . For a vector let be the minimum of the prices over indices with , with value zero when there is no such index, and put . Let , , , and . The required condition is that both these value sets are bounded above, that some coordinatewise nonnegative vector attains , that some nonnegative scalar uniform price attains , and that . The integral is the real-valued Bochner integral, defined as zero when its integrand is not integrable, and the real supremum has default value zero for an empty or unbounded-above set; both value sets in fact contain zero, and boundedness is explicitly required here. The outer definition permits any natural as an argument, but its positivity clause excludes from being a regular instance. No equality among the marginal measures is required. For a probability measure, the set defining is empty when ; it can also be empty at , in which case the same zero default applies. No endpoint condition on is added by regularity.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.