§2.1, p. 210 — for x > 0, ∇ₓv(x, Y) exists iff the optimal dual price vector is unique
ProvedRLPBidPrice.Unbiased.gradient_iff_unique_dualLet be a 0/1 incidence matrix with nonzero columns, , , and let be a capacity vector with every component . Then the perfect-information LP value is (Fréchet) differentiable at if and only if the problem
has exactly one solution, i.e. the optimal vector of dual prices of the capacity constraints is unique.
This is the reduction of the paper's differentiability question to uniqueness of the dual prices.
Formalization Note. The paper's "" is at the fixed capacity . Differentiability is taken at an interior capacity , where a neighbourhood of consists of feasible capacities; at a boundary point of the two-sided derivative is not meaningful. Dual optimality is minimization of the Lagrangian (13) over (Eq. (14)). The time index is dropped.
import Mathlib import Definitions.Def_RLPBidPrice_Unbiased_Model open MeasureTheory Matrix
namespace RLPBidPrice.Unbiased
theorem gradient_iff_unique_dual {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 : ∀ i, 0 < x i)
(y₀ : Fin n → ℝ) (hy₀ : 0 ≤ y₀) :
DifferentiableAt ℝ (fun x' => piValue A r x' y₀) x ↔ ∃! μ, IsDualOpt A r x y₀ μ := by sorry
end RLPBidPrice.Unbiased
Read-back
What the Lean code literally says, in plain math · claude-opus-5-5
Let be a real matrix whose entries are all or and whose every column has a nonzero entry. Assume further:
- with ;
- with every coordinate strictly positive, ;
- with .
Define, for any ,
This is a real supremum, when the set is empty. Since , that happens exactly when some .
Define
Call dual-optimal if and for all .
The statement asserts an equivalence:
Degenerate cases.
- If , the incidence condition forces . Then is a single point, is trivially differentiable, and the unique vector is dual-optimal. Both sides hold.
- If (with ), near and . The asserted equivalence covers this case too.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.