Lift an optimal-policy expectation subsequence limit to the full sequence
ProvedPandoraBO.Private20261023.optimal_policy_subsequence_limitsoptimal-searchpandoras-box
Consider optimal policies chosen consistently at equal nonnegative multipliers. Along a monotone multiplier sequence, convergence of expected cost and reward on a strictly increasing subsequence implies convergence of the full expected cost and reward sequences to the same limits. The equal-multiplier consistency condition prevents arbitrary switching among ties on plateaus.
Preamble
import Definitions.Def_PandoraBO_Model open PandoraBO Filter Topology set_option autoImplicit false
Formal statement
theorem PandoraBO.Private20261023.optimal_policy_subsequence_limits {n : ℕ} (M : Instance n)
(multipliers : ℕ → ℝ) (hnonneg : ∀ k, 0 ≤ multipliers k)
(policies : ℕ → Policy n)
(hopt : ∀ k, optimal M (multipliers k) (policies k))
(hsame : ∀ j k, multipliers j = multipliers k → policies j = policies k)
(hmono : Monotone multipliers ∨ Antitone multipliers)
(φ : ℕ → ℕ) (hφ : StrictMono φ) (costLimit rewardLimit : ℝ)
(hcost : Tendsto (fun k => expectedCost M (policies (φ k))) atTop (𝓝 costLimit))
(hreward : Tendsto (fun k => expectedReward M (policies (φ k))) atTop (𝓝 rewardLimit)) :
Tendsto (fun k => expectedCost M (policies k)) atTop (𝓝 costLimit) ∧
Tendsto (fun k => expectedReward M (policies k)) atTop (𝓝 rewardLimit) := by sorry
Source
Auxiliary scalar convergence argument for the private formalization of finite-policy multiplier limits in Xie et al., author revised manuscript reviewed 23 October 2026.