Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

§3.1, p. 212 (corrected: ≤ for ≥) — an optimal dual price vector μ(x, Y) is a supergradient: v(z, Y) ≤ v(x, Y) + μᵀ(z − x) for all z ≥ 0

Proved
RLPBidPrice.Directional.dualOpt_supergradient

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

convex-analysislinear-programmingp2o-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 a capacity vector and Y≥0Y \ge 0Y≥0 a demand vector. Let μ\muμ be an optimal vector of dual prices for the capacity constraints, i.e. μ≥0\mu \ge 0μ≥0 minimizes L(x,μ)=∑jYj(rj−μ⊤Aj)++μ⊤xL(x, \mu) = \sum_j Y^j (r^j - \mu^\top A^j)^+ + \mu^\top xL(x,μ)=∑j​Yj(rj−μ⊤Aj)++μ⊤x over μ≥0\mu \ge 0μ≥0. Then for every capacity vector z≥0z \ge 0z≥0,

v(z,Y)≤v(x,Y)+μ⊤(z−x),v(z, Y) \le v(x, Y) + \mu^\top (z - x),v(z,Y)≤v(x,Y)+μ⊤(z−x),

where v(x,Y)=max⁡{r⊤y:Ay≤x, 0≤y≤Y}v(x, Y) = \max\{r^\top y : Ay \le x,\ 0 \le y \le Y\}v(x,Y)=max{r⊤y:Ay≤x, 0≤y≤Y}.

In words: for every realization of demand, the dual price vector is a supergradient of the concave function z↦v(z,Y)z \mapsto v(z, Y)z↦v(z,Y) (the paper says "subgradient").

Formalization Note. zzz ranges over the feasible capacities z≥0z \ge 0z≥0 (for other zzz the LP has no value). The time index kkk is dropped.

Correction. The paper prints the inequality as vk(z,Y)≥vk(x,Y)+μ⊤(x,Y)(z−x)v_k(z, Y) \ge v_k(x, Y) + \mu^\top(x, Y)(z - x)vk​(z,Y)≥vk​(x,Y)+μ⊤(x,Y)(z−x). Since v(⋅,Y)v(\cdot, Y)v(⋅,Y) is concave the inequality goes the other way: with one leg, one itinerary, r=1r = 1r=1, Y=5Y = 5Y=5, x=1x = 1x=1 (so μ=1\mu = 1μ=1) and z=10z = 10z=10, the printed inequality reads 5≥105 \ge 105≥10.

Preamble
import Mathlib
import Definitions.Def_RLPBidPrice_Directional_Model

open MeasureTheory Matrix
Formal statement
namespace RLPBidPrice.Directional

/-- §3.1, p. 212 (sign corrected): an optimal dual price vector `μ` of (14) at `x` is a
supergradient of the concave map `z ↦ v(z, y₀)` on the capacities `z ≥ 0`. -/
theorem dualOpt_supergradient {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 : Fin m → ℝ) (hμ : IsDualOpt A r x y₀ μ)
    (z : Fin m → ℝ) (hz : 0 ≤ z) :
    piValue A r z y₀ ≤ piValue A r x y₀ + μ ⬝ᵥ (z - x) := by sorry

end RLPBidPrice.Directional
Source
Talluri, van Ryzin, A Randomized Linear Programming Method for Computing Network Bid Prices, Transp. Sci. 33(2), 1999, p. 212, §3.1, first display
Read-back

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

Setting. Let m,n∈Nm, n \in \mathbb{N}m,n∈N, let AAA be a real m×nm \times nm×n matrix, and let r∈Rnr \in \mathbb{R}^nr∈Rn. All vector inequalities are componentwise.

Hypotheses.

  • AAA is an incidence matrix: every entry is 000 or 111, and every column has at least one nonzero entry.
  • r≥0r \ge 0r≥0.
  • x∈Rmx \in \mathbb{R}^mx∈Rm with x≥0x \ge 0x≥0.
  • y0∈Rny_0 \in \mathbb{R}^ny0​∈Rn with y0≥0y_0 \ge 0y0​≥0.
  • μ∈Rm\mu \in \mathbb{R}^mμ∈Rm is an optimal dual price at (x,y0)(x, y_0)(x,y0​). That is, μ≥0\mu \ge 0μ≥0 and L(x,y0,μ)≤L(x,y0,μ′)L(x,y_0,\mu) \le L(x,y_0,\mu')L(x,y0​,μ)≤L(x,y0​,μ′) for every μ′≥0\mu' \ge 0μ′≥0, where
L(x,y0,μ)=∑j=1n(y0)jmax⁡(0, rj−∑iμiAij)+μ⋅x.L(x,y_0,\mu) = \sum_{j=1}^n (y_0)_j \max\Big(0,\ r_j - \sum_{i} \mu_i A_{ij}\Big) + \mu \cdot x .L(x,y0​,μ)=j=1∑n​(y0​)j​max(0, rj​−i∑​μi​Aij​)+μ⋅x.
  • z∈Rmz \in \mathbb{R}^mz∈Rm with z≥0z \ge 0z≥0.

As before, v(x,y0)=sup⁡{ r⋅y:Ay≤x, 0≤y≤y0 }v(x,y_0) = \sup\{\, r\cdot y : Ay \le x,\ 0 \le y \le y_0 \,\}v(x,y0​)=sup{r⋅y:Ay≤x, 0≤y≤y0​}, which is 000 if that set is empty.

Conclusion.

v(z,y0)  ≤  v(x,y0)+μ⋅(z−x).v(z, y_0) \;\le\; v(x, y_0) + \mu \cdot (z - x).v(z,y0​)≤v(x,y0​)+μ⋅(z−x).

Degenerate cases.

  • Because x,z,y0≥0x, z, y_0 \ge 0x,z,y0​≥0, both LPs are feasible (y=0y = 0y=0 works), so neither value is the junk value.
  • The hypothesis on μ\muμ may be unsatisfiable for some data (for example, if L(x,y0,⋅)L(x,y_0,\cdot)L(x,y0​,⋅) has no minimiser on the orthant). In that case the statement is vacuous for those data.
  • If n=0n = 0n=0: v≡0v \equiv 0v≡0 and L=μ⋅xL = \mu\cdot xL=μ⋅x, so optimality forces μ⋅x=0\mu \cdot x = 0μ⋅x=0. The claim reduces to 0≤μ⋅z0 \le \mu\cdot z0≤μ⋅z.
  • If m=0m = 0m=0: the incidence hypothesis forces n=0n = 0n=0, and the claim is 0≤00 \le 00≤0.
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