Theorem 4: is quasi-convex in on
ProvedAssortSearch.MNLQC.theorem_4Consider the traditional multinomial logit (MNL) model of consumer choice, without search. There are variants with preferences and a no-purchase option with preference . Variant has margin . Including a variant with demand costs , where the operational cost is concave and increasing on . With assortment , variant 's demand is and its profit is .
Fix an assortment and a variant , and regard 's preference as a variable. Let be variant 's profit in the assortment , and let be the profit lost on the variants already in . Then the profit change from adding ,
is quasi-convex in on . That is, for every the set is an interval.
Because is quasi-convex, it attains its maximum over any interval at an endpoint. If adding a less popular variant is profitable, adding a more popular one is even better. This is how the paper shows that the optimal no-search assortment consists of the most popular variants. It extends van Ryzin and Mahajan's (1999) result from their newsvendor cost to any concave increasing cost.
Formalization Note Variants are Fin n, 0-based, and the no-purchase preference is a separate positive real. is assumed, as the paper's "adding variant to an assortment " requires. is assumed concave and increasing only on , where demands lie, and nothing is assumed about , so need not be . No differentiability of is assumed. Margins may differ across variants, as in the paper's model. The paper's proof writes a common margin, but the theorem does not need one.
import Mathlib import Definitions.Def_RetailVariety_Structure_Model import Definitions.Def_AssortSearch_MNLQC_Model
namespace AssortSearch.MNLQC
/-- **Theorem 4** (p. 13): in the no-search MNL model, with a concave increasing operational
cost `c` on `[0, 1]` and arbitrary margins `m_i`, the function
`h^m(v_j) = π_j^m(v_j) − L^m(v_j)` is quasi-convex in the added variant's preference `v_j`
on `[0, ∞)`. -/
theorem theorem_4
{n : ℕ} (m : Fin n → ℝ) (c : ℝ → ℝ) (v : Fin n → ℝ) (v0 : ℝ)
(S : Finset (Fin n)) (j : Fin n)
(hv : ∀ i, 0 < v i) (hv0 : 0 < v0)
(hc_concave : ConcaveOn ℝ (Set.Icc 0 1) c) (hc_mono : MonotoneOn c (Set.Icc 0 1))
(hj : j ∉ S) :
QuasiconvexOn ℝ (Set.Ici 0) (hm m c v v0 S j) := by sorry
end AssortSearch.MNLQC
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.