Theorem 3.7 — KL upper bounds in both orientations
ProvedInventoryBounds.stationary_pair_klinventoryprobabilitysample-complexity
On the stated horizon and accuracy range, both Bernoulli KL orientations between rhoPlus and rhoMinus are at most 22T(128epsilon/T^2)^2.
Preamble
import Definitions.Def_InventoryBounds_StationaryValuation set_option autoImplicit false open MeasureTheory ProbabilityTheory open scoped BigOperators
Formal statement
namespace InventoryBounds
theorem stationary_pair_kl (T : ℕ) (hT : 8 ≤ T) (ε : ℝ)
(hε : 0 < ε) (hεT : ε ≤ (T : ℝ) / 1024) :
max (binaryKL (rhoPlus T ε) (rhoMinus T ε))
(binaryKL (rhoMinus T ε) (rhoPlus T ε)) ≤
22 * (T : ℝ) * (128 * ε / (T : ℝ) ^ 2) ^ 2 := by sorry
end InventoryBoundsSource
Hanzhang Qin, David Simchi-Levi, Ruihao Zhu, Information Limits of Multistage Inventory Control: Learning, Valuation, and Censoring, arXiv:2609.37380v1 (29 September 2026), https://arxiv.org/abs/2609.37380v1, Theorem 3.7, equation (8)
Read-back
What the Lean code literally says, in plain math · gpt-6
For every natural number and every real satisfying , let and define . Then . Thus the stated same upper bound holds for both displayed orders of the two arguments. All occurrences of in arithmetic are real coercions. The logarithm and division are the total real operations, but these hypotheses place both and strictly between and , so all logarithm arguments in this assertion are positive and all displayed denominators are nonzero. The case is included and the case is excluded.
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.