Proof of Theorem 3.8, p. 12 — for a box, max_{|S|≤K} min_v (1/v₀) Σ v_i (r_i − λ) = max_{|S|≤K} (1/u₀) Σ l_i (r_i − λ)
ProvedRobustMNL.SizeCon.rectangular_reductionLet be a rectangular uncertainty set: and for every product , the no-purchase weight ranges over and the weight of product over . Let be real revenues and . Then for every real ,
The right-hand side is the same parametric quantity evaluated at the single parameter vector , so over a box the robust parametric problem collapses to a problem with known weights. The equality holds only after the maximum over assortments: for a fixed containing a product with the two inner quantities differ.
Formalization Note The inner minimum is a real infimum over the box. The hypotheses and (componentwise) say that each interval is nonempty, the reading of "the interval represents the range of the likely values"; and express . Revenues are arbitrary reals. Products are 0-indexed (Fin n).
import Mathlib import Definitions.Def_ChoiceCDLP_MNL_mnlObjective import Definitions.Def_RobustMNL_SizeCon_Model
namespace RobustMNL.SizeCon
/-- Proof of Theorem 3.8, p. 12, second display: over the rectangular set
`∏_{i=0}^n [l_i, u_i] ⊆ ℝⁿ⁺¹₊₊`, for every real `λ`,
`max_{|S| ≤ K} min_{v ∈ V} (1/v₀) ∑_{i ∈ S} v_i (r_i − λ) = max_{|S| ≤ K} (1/u₀) ∑_{i ∈ S} l_i (r_i − λ)`. -/
theorem rectangular_reduction {n : ℕ} (l₀ u₀ : ℝ) (l u : Fin n → ℝ) (hl₀ : 0 < l₀)
(hlu₀ : l₀ ≤ u₀) (hl : ∀ i, 0 < l i) (hlu : l ≤ u) (r : Fin n → ℝ) (K : ℕ) (lam : ℝ) :
paramValue (box l₀ u₀ l u) r K lam =
(sizeFeasible n K).sup' (sizeFeasible_nonempty n K)
(fun S => (1 / u₀) * ∑ i ∈ S, l i * (r i - lam)) := by sorry
end RobustMNL.SizeCon
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.