Theorem 3.2 — Revenue-ordered assortments are robust: S*(𝒱) = {i : rᵢ > Z*(𝒱)}
ProvedRobustMNL.Static.revenue_ordered_robustRevenue-ordered assortments are robust. Let be a compact nonempty uncertainty set for the parameters of the multinomial logit model, and let be the product revenues. An assortment is an optimal solution of the Robust Logit problem
with the smallest cardinality among all optimal solutions if and only if
In particular the smallest optimal robust assortment is unique, and it consists of the products whose revenue strictly exceeds the optimal worst-case revenue. When the revenues are ordered, , it is therefore one of the revenue-ordered assortments , so the robust problem needs only these to be searched.
Formalization Note The theorem is stated as an "if and only if" for every assortment : the forward direction is the paper's identity for any choice of , the backward direction says that the threshold set itself is an optimal assortment of smallest cardinality. The page writes ""; the parameter vector has components (p. 5) and is compact by the standing assumption of Sec. 3 (p. 6), which the proof uses, so the statement is made for compact nonempty . Revenues are arbitrary reals rather than : the proof uses neither the order nor the sign, and Sec. 4 of the paper applies the theorem to revenues that may be negative. Products are Fin n (Lean index is the paper's product ).
import Mathlib import Definitions.Def_RobustMNL_Static_Model
namespace RobustMNL.Static
theorem revenue_ordered_robust {n : ℕ} (r : Fin n → ℝ) (V : Set (ℝ × (Fin n → ℝ)))
(hV : IsCompact V) (hne : V.Nonempty) (hpos : ∀ p ∈ V, IsPos p) (S : Finset (Fin n)) :
IsSmallestOptimal V r S ↔ S = Finset.univ.filter (fun i => Zstar V r < r i) := by sorry
end RobustMNL.Static
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.