Proof of Theorem 3, p. 30 — is convex in the uniform service duration
ProvedPriceQualityService.yValue_convexmultinomial-logitoperations-managementpricing
The products are . Product has quality sensitivity , service utility per unit time , production-cost coefficient and service-cost coefficients ; the firm chooses its price , quality and service duration .
Fix a quality vector . The function
is convex on .
Convexity of in the common duration is what places the optimal uniform duration at an end of .
Formalization Note The claim is convexity, not strict convexity: the paper's computation "" fails when does not depend on (e.g. all and ).
Differences from the source:
- reading: y evaluated at the unconstrained optimal quality q_i = (α_i + t b_i)/(2c_i) → y at a fixed quality vector q; reason: the form the argument needs under a quality constraint; it is the paper's y when q is that point
- generalization: convex in the service duration t (durations in [t_s, t_l]); α_i, s_i, c_i > 0 → convex on all of ℝ; arbitrary real parameters; reason: each exponent is affine in t for fixed q; restricting to [t_s, t_l] gives the paper's claim
- encoding: products 𝒩 = {1, …, N} → Fin N (N = 0 allowed); reason: Lean convention; N = 0 gives empty sums and changes no instance with N ≥ 1
Preamble
import Mathlib import Definitions.Def_PriceQualityService_yValue
Formal statement
namespace PriceQualityService
/-- **`y` is convex in the uniform service duration.**
Wang, Ke & Cui, *Product Price, Quality and Service Decisions under Consumer Choice Models*,
accepted manuscript, SSRN 3766191 (MSOM 2020, doi:10.1287/msom.2020.0947), proof of Theorem 3 (App. B), unnumbered, p. 30: "Let
`y = Σ_{i∈𝒩} exp(b_i²t²/(4c_i) + (s_i − a_i + α_i b_i/(2c_i))t + α_i²/(4c_i))`. … Thus, y is a convex
function in service duration t."
For every fixed quality vector `q`, the map `t ↦ y(q, t) = Σ_i exp(α_i q_i − c_i q_i² + t(s_i − (a_i − b_i q_i)))`
is convex on all of `ℝ`.
**Formalization Note.** Convexity, not strict convexity: the paper's "`∂²y/∂t² > 0`" fails when
`y` is constant in `t` (e.g. all `b_i = 0` and `s_i = a_i`). `y` is taken at a fixed quality
vector rather than at the unconstrained optimal quality `(α_i + t b_i)/(2c_i)`. -/
theorem yValue_convex {N : ℕ} (α s a b c q : Fin N → ℝ) :
ConvexOn ℝ Set.univ (fun t : ℝ => yValue α s a b c q (fun _ => t)) := by sorry
end PriceQualityService
Source
Wang, Ke & Cui, Product Price, Quality and Service Decisions under Consumer Choice Models, accepted manuscript, SSRN 3766191 (MSOM 2020, doi:10.1287/msom.2020.0947), proof of Theorem 3 (App. B), p. 30 (unnumbered)