An asymptotically optimal heuristic for general nonstationary finite-horizon restless multi-armed, multi-action bandits: Corrigendum

Abstract

The above lemma is used to prove Theorems 1–2 and Propositions 1–3 in Sections 4 and 6 of [1]. It has been graciously pointed out to us that the bound in the lemma may not be correct in general. The original proof of this lemma uses a combination of linear program (LP) duality and sensitivity analysis results. The mistake is in the application of a known sensitivity analysis result under a certain assumption that happens to be not necessarily satisfied by our LP. Fortunately, it is possible to correct the bound in the above lemma. The new bound that we will prove in this correction note is as follows: