Optimal expected cost and reward decrease with the cost multiplier
ProvedPandoraBO.Private20261023.optimal_policy_cost_reward_monotonicityexpected-budgetmonotonicitypandoras-box
For two real cost multipliers 0≤s<t, let π be globally optimal at s and ρ globally optimal at t. Then the expected cost of ρ is at most that of π, and its expected reward is also at most that of π. The claim holds for any choices of these optimizers.
Preamble
import Definitions.Def_PandoraBO_Model open PandoraBO set_option autoImplicit false
Formal statement
theorem PandoraBO.Private20261023.optimal_policy_cost_reward_monotonicity {n : ℕ} (M : Instance n) (s t : ℝ)
(hs : 0 ≤ s) (hst : s < t) (π ρ : Policy n)
(hπ : optimal M s π) (hρ : optimal M t ρ) :
expectedCost M ρ ≤ expectedCost M π ∧
expectedReward M ρ ≤ expectedReward M π := by sorry
Source
Supporting monotonicity argument in Xie et al., author revised manuscript reviewed 23 October 2026, Appendix B.5, proof of Lemma 7 (equation (43) and the final subsequence argument on page 22). Isolated algebraic/analytic auxiliary lemma for formalization.