Proof of Theorem 3.3, pp. 590–591 — the Gaussian mixture P_m(x) has mean μ and second moment Q
ProvedMinimaxSLP.RhsExtremal.mixture_mem_momentClassgaussianmoment-problemp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1probability
Let be a feasible point of (16) in which every block has or is zero. Let be the mixed distribution that, with probability , draws
Then is a probability distribution with finite second moments and
i.e. .
This is the moment-matching step of the construction of the extremal distribution.
Formalization Note The normal components are Mathlib's multivariateGaussian (which accepts singular positive semidefinite covariances), transported from EuclideanSpace ℝ (Fin r) to Fin r → ℝ; weights are ENNReal.ofReal . The paper's "", "" and "with probability " are read with the superscript .
Preamble
import Mathlib import Definitions.Def_MomentDRO_Conf_Setting import Definitions.Def_MinimaxSLP_RhsExtremal_Model
Formal statement
namespace MinimaxSLP.RhsExtremal
open MeasureTheory Matrix
/-- Proof of Theorem 3.3, pp. 590–591: if `s` is feasible for (16) and every block has
`v_k0^i > 0` or is zero, the mixture `P_m(x)` that draws `h̃_k^i ∼ ℕ(v_k^i/v_k0^i,
(V_k^i v_k0^i − v_k^i (v_k^i)′)/(v_k0^i)²)` with probability `v_k0^i` lies in `𝒫`:
`E[h̃] = μ` and `E[h̃h̃′] = Q`. -/
theorem mixture_mem_momentClass {r K N : ℕ} (μ : Fin r → ℝ) (Q : Matrix (Fin r) (Fin r) ℝ)
(s : Fin K → Fin N → (Matrix (Fin r) (Fin r) ℝ × (Fin r → ℝ) × ℝ))
(hs : s ∈ feasible16 K N μ Q)
(hpos : ∀ k i, 0 < (s k i).2.2 ∨ s k i = 0) :
mixture s ∈ MinimaxSLP.ObjSDP.momentClass μ Q := by sorry
end MinimaxSLP.RhsExtremal
Source
Bertsimas, Doan, Natarajan & Teo, Models for Minimax Stochastic Linear Optimization Problems with Risk Aversion, Math. Oper. Res. 35(3), 2010, pp. 590–591, proof of Theorem 3.3
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.