Appendix C — Finite candidate prices suffice
ProvedFBMSP.discrete_candidate_sufficiencyFor a valid finite rational market with prediction values and error atoms, define
Then every candidate is nonnegative, , and every policy using arbitrary nonnegative real prices is weakly dominated by a policy using only prices from , with the same customer assignment and at most the same number of price slots.
This is the finite-support link needed to apply the finite-candidate greedy guarantee to unrestricted pricing. The statement proves a cardinality bound; a machine-level cost of exact arithmetic is not asserted.
import Definitions.Def_FBMSP_DiscreteCandidates
namespace FBMSP
/-- Finite-support bridge for Remark 2 / Appendix C. Every real-price policy
is dominated by a policy using realized nonnegative valuation breakpoints,
without changing its assignment. Thus a finite-candidate optimization loses
no revenue in this discrete setting. Draft statement; proof remains open.
The cardinality bound is explicit. No computational runtime bound is
asserted here; exact-arithmetic evaluation costs require a separate model. -/
theorem discrete_candidate_sufficiency (I : RationalMarketInput)
(hvalid : ValidRationalMarket I) :
0 ∈ rationalCandidatePrices I ∧
(∀ p ∈ rationalCandidatePrices I, 0 ≤ p) ∧
(rationalCandidatePrices I).card ≤ I.means.length * I.shocks.length + 1 ∧
∀ (prices : Fin I.segments → ℝ)
(assignment : Fin I.means.length → Fin I.segments),
(∀ j, 0 ≤ prices j) →
∃ candidatePrices : Fin I.segments → ℝ,
(∀ j, candidatePrices j ∈ rationalCandidatePrices I) ∧
rationalMarketRevenue I prices assignment ≤
rationalMarketRevenue I candidatePrices assignment := by sorry
end FBMSPRead-back
What the Lean code literally says, in plain math · gpt-6-astra
Let consist of rational mean and mean-weight lists and with , distinct nonnegative means, strictly positive weights, and ; rational shock and shock-weight lists and with , all , , and ; a natural number of price slots ; and an arbitrary rational threshold. Each value list has the same length as its weight list. Define the finite real set , with rational sums embedded in and repeated values identified. The assertion is that , all its elements are nonnegative, , and for every nonnegative real vector and every assignment there is a vector with every such that . The assignment is the same on both sides, and purchase at equality is included. The original prices may be arbitrary real numbers; the replacement prices are drawn from this finite set and may repeat or be assigned to no mean. Means need not be sorted; shocks may repeat and have zero weight, and all supplied mean-shock pairs contribute candidate sums regardless of weight. Negative realized values are allowed but do not become candidate prices. The threshold value is unrestricted and has no role in the conclusion; there is no bound on in this theorem, no assertion that the replacement prices maximize revenue, and no computation-time conclusion.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.