Theorem 6.B.16(b) — closure of the usual multivariate stochastic order under conjunctions (corrected: Borel , general dimensions)
ProvedStochasticOrders.Multivariate.multivariate_order_conjunction_v2Let be independent random vectors on , of dimension , and let be another set of independent random vectors on , also of dimension . Denote . If for every , then for any increasing Borel function ,
i.e. for every real (the univariate usual stochastic order, since is real-valued). That is, the usual multivariate stochastic order is closed under conjunctions; in particular (taking a sum) it is closed under convolutions. The book notes that parts (a) and (c) of Theorem 6.B.16 are special cases of this part.
Formalization Note. The retired version assumed monotone but not measurable: for an increasing function need not be Borel, and for a non-Borel the outer measure depends on the σ-algebra of rather than on the law of , which the order cannot control (refuted with the indicator of the upper set generated by a Vitali antichain). The new statement requires Measurable ψ, as the book's " is a random variable" does, and uses the corrected MultivariateOrder_v2 (Borel test functions) for the hypotheses . The book's general dimensions are kept: the concatenated vector lives in (i : Fin m) → Fin (k i) → ℝ with its coordinatewise order (order- and Borel-isomorphic to ), whereas the retired version fixed a common dimension. Independence of each family is Mathlib's iIndepFun; the conclusion is stated as the univariate order's defining inequality rather than by importing Chapter I's UsualOrder. Random vectors are measurable functions on probability spaces (Measurable, IsProbabilityMeasure). No correction to the printed source.
import Mathlib import Definitions.Def_StochasticOrders_Multivariate_MultivariateOrder_v2
namespace StochasticOrders.Multivariate
open MeasureTheory ProbabilityTheory
/-- Theorem 6.B.16(b) (Shaked & Shanthikumar, *Stochastic Orders*, Springer 2007, p. 273): let
`X 1, …, X m` be independent random vectors, `X i` of dimension `k i`, and `Y 1, …, Y m` another
independent family with `Y i` of dimension `k i`. If `X i ≤st Y i` for every `i`, then for any
increasing function `ψ : ℝ^k → ℝ` (`k = k 1 + ⋯ + k m`, the concatenated vector living in
`(i : Fin m) → Fin (k i) → ℝ` with its coordinatewise order), `ψ(X 1, …, X m) ≤st ψ(Y 1, …, Y m)`
in the *univariate* usual stochastic order (`ψ` has codomain `ℝ`), stated directly as the
univariate order's defining inequality since drafts cannot import another mission's `UsualOrder`.
Corrected version (`_v2`) of `multivariate_order_conjunction`: (1) `ψ` is Borel measurable, as
the book's "`ψ(X)` is a random variable" requires — an increasing function on `ℝ^k`, `k ≥ 2`,
need not be Borel, and for a non-Borel `ψ` the outer measure `μ {x < ψ ∘ X}` is not determined by
the law of `X`, which is how the retired statement was refuted. (2) `MultivariateOrder` is the
corrected `_v2` definition (Borel test functions). (3) The book's general dimensions `k i` are
kept rather than a common dimension `n`. -/
theorem multivariate_order_conjunction_v2 {Ω Ω' : Type*} [MeasurableSpace Ω] [MeasurableSpace Ω']
(μ : Measure Ω) (ν : Measure Ω') [IsProbabilityMeasure μ] [IsProbabilityMeasure ν]
{m : ℕ} (k : Fin m → ℕ) (X : (i : Fin m) → Ω → Fin (k i) → ℝ)
(Y : (i : Fin m) → Ω' → Fin (k i) → ℝ)
(hX : ∀ i, Measurable (X i)) (hY : ∀ i, Measurable (Y i))
(hXindep : iIndepFun X μ) (hYindep : iIndepFun Y ν)
(hord : ∀ i, MultivariateOrder μ ν (X i) (Y i))
(ψ : ((i : Fin m) → Fin (k i) → ℝ) → ℝ) (hψmeas : Measurable ψ) (hψ : Monotone ψ) :
∀ x : ℝ, μ {ω | x < ψ (fun i => X i ω)} ≤ ν {ω | x < ψ (fun i => Y i ω)} := by sorry
end StochasticOrders.Multivariate