Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

§2.1, p. 211 — strong duality v(x, Y) = L(x, μ_*), μ_* is a supergradient, and a unique μ_* is the gradient

Proved
RLPBidPrice.Unbiased.strong_duality_subgradient

by mikedeng1 · Oct 3, 2026 · Mathlib 0df444a (Lean v4.33.1)

linear-programminglp-dualityp2o-batch-rm-cafterp2o-gran-per-chapterp2o-plan-paperp2o-v1revenue-management

Let AAA be a 0/1 incidence matrix with nonzero columns, r≥0r \ge 0r≥0, x≥0x \ge 0x≥0 and Y≥0Y \ge 0Y≥0, with vvv and LLL as above. Then:

  1. an optimal dual price vector exists: some μ∗≥0\mu_* \ge 0μ∗​≥0 minimizes L(x,⋅)L(x, \cdot)L(x,⋅) over μ≥0\mu \ge 0μ≥0;
  2. (strong duality and the supergradient inequality) every such minimizer μ∗\mu_*μ∗​ satisfies v(x,Y)=L(x,μ∗)v(x, Y) = L(x, \mu_*)v(x,Y)=L(x,μ∗​) and
v(z,Y)≤v(x,Y)+μ∗⊤(z−x)for all capacity vectors z≥0;v(z, Y) \le v(x, Y) + \mu_*^\top (z - x) \qquad \text{for all capacity vectors } z \ge 0;v(z,Y)≤v(x,Y)+μ∗⊤​(z−x)for all capacity vectors z≥0;
  1. if the minimizer μ∗\mu_*μ∗​ is unique and xi>0x^i > 0xi>0 for every iii, then x′↦v(x′,Y)x' \mapsto v(x', Y)x′↦v(x′,Y) is differentiable at xxx with gradient μ∗\mu_*μ∗​.

These are the facts from LP duality that turn the question of differentiability of vvv into the uniqueness of μ∗\mu_*μ∗​.

Formalization Note. The paper says "μ∗\mu_*μ∗​ is always a subgradient of vk(x,Y)v_k(x, Y)vk​(x,Y)"; vvv is concave in xxx, so this is meant in the concave sense (a supergradient), stated here as the displayed inequality, over the feasible capacities z≥0z \ge 0z≥0. The gradient in part 3 is expressed as HasFDerivAt with the linear functional h↦μ∗⊤hh \mapsto \mu_*^\top hh↦μ∗⊤​h; it is taken at an interior point x>0x > 0x>0. The time index kkk is dropped.

Preamble
import Mathlib
import Definitions.Def_RLPBidPrice_Unbiased_Model

open MeasureTheory Matrix
Formal statement
namespace RLPBidPrice.Unbiased

theorem strong_duality_subgradient {m n : ℕ} (A : Matrix (Fin m) (Fin n) ℝ) (r : Fin n → ℝ)
    (hA : IsIncidence A) (hr : 0 ≤ r) (x : Fin m → ℝ) (hx : 0 ≤ x)
    (y₀ : Fin n → ℝ) (hy₀ : 0 ≤ y₀) :
    (∃ μ, IsDualOpt A r x y₀ μ) ∧
    (∀ μ, IsDualOpt A r x y₀ μ →
      piValue A r x y₀ = lagr A r x y₀ μ ∧
      ∀ z : Fin m → ℝ, 0 ≤ z → piValue A r z y₀ ≤ piValue A r x y₀ + μ ⬝ᵥ (z - x)) ∧
    (∀ μ, IsDualOpt A r x y₀ μ → (∀ μ', IsDualOpt A r x y₀ μ' → μ' = μ) →
      (∀ i, 0 < x i) → HasFDerivAt (fun x' => piValue A r x' y₀) (gradCLM μ) x) := by sorry

end RLPBidPrice.Unbiased
Source
Talluri, van Ryzin, A Randomized Linear Programming Method for Computing Network Bid Prices, Transp. Sci. 33(2), 1999, p. 211, §2.1, first sentences (after Eq. (14))
Read-back

What the Lean code literally says, in plain math · claude-opus-5-5

Let AAA be a real m×nm \times nm×n matrix whose entries are all 000 or 111 and whose every column has a nonzero entry. Assume further:

  • r∈Rnr \in \mathbb{R}^nr∈Rn with r≥0r \ge 0r≥0;
  • x∈Rmx \in \mathbb{R}^mx∈Rm with x≥0x \ge 0x≥0 (zero coordinates allowed);
  • y0∈Rny_0 \in \mathbb{R}^ny0​∈Rn with y0≥0y_0 \ge 0y0​≥0.

Define, for z∈Rmz \in \mathbb{R}^mz∈Rm,

v(z)=sup⁡{r⋅y:Ay≤z, 0≤y≤y0},v(z) = \sup\{r \cdot y : Ay \le z,\ 0 \le y \le y_0\},v(z)=sup{r⋅y:Ay≤z, 0≤y≤y0​},

a real supremum, 000 if empty. Define

L(μ)=∑j(y0)jmax⁡(0,rj−∑iμiAij)+μ⋅x.L(\mu) = \sum_j (y_0)_j \max\Big(0, r_j - \sum_i \mu_i A_{ij}\Big) + \mu \cdot x.L(μ)=j∑​(y0​)j​max(0,rj​−i∑​μi​Aij​)+μ⋅x.

Call μ\muμ dual-optimal if μ≥0\mu \ge 0μ≥0 and L(μ)≤L(μ′)L(\mu) \le L(\mu')L(μ)≤L(μ′) for all μ′≥0\mu' \ge 0μ′≥0.

The statement asserts all three of the following:

  1. Existence. A dual-optimal μ\muμ exists.
  2. Strong duality and supergradient. Every dual-optimal μ\muμ satisfies v(x)=L(μ)v(x) = L(\mu)v(x)=L(μ). It also satisfies, for every z∈Rmz \in \mathbb{R}^mz∈Rm with z≥0z \ge 0z≥0 (and only such zzz),
v(z)≤v(x)+μ⋅(z−x).v(z) \le v(x) + \mu \cdot (z - x).v(z)≤v(x)+μ⋅(z−x).
  1. Gradient. Suppose μ\muμ is dual-optimal, every dual-optimal μ′\mu'μ′ equals μ\muμ, and additionally every xi>0x_i > 0xi​>0. Then vvv, as a function on all of Rm\mathbb{R}^mRm, is Fréchet differentiable at xxx with derivative h↦∑iμihih \mapsto \sum_i \mu_i h_ih↦∑i​μi​hi​, i.e. gradient μ\muμ.

Degenerate cases.

  • If m=0m = 0m=0, the incidence condition forces n=0n = 0n=0. All three parts then hold trivially, with v=L=0v = L = 0v=L=0.
  • If n=0n = 0n=0, v(z)=0v(z) = 0v(z)=0 for z≥0z \ge 0z≥0 and L(μ)=μ⋅xL(\mu) = \mu \cdot xL(μ)=μ⋅x.
  • Part 3 applies only when xxx is strictly positive. For xxx with a zero coordinate it asserts nothing.
Human review
  • Endorsed by Shuze Chen · Oct 21, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 21, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me