Mirror descent actor-critic methods for entropy regularised MDPs in general spaces: stability and convergence

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zorba, Denis, Šiška, David, Szpruch, Lukasz
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914322323603456
author Zorba, Denis
Šiška, David
Szpruch, Lukasz
author_facet Zorba, Denis
Šiška, David
Szpruch, Lukasz
contents We provide theoretical guarantees for convergence of discrete-time policy mirror descent with inexact advantage functions updated using temporal difference (TD) learning for entropy regularised MDPs in Polish state and action spaces. We rigorously derive sufficient conditions under which the single-loop actor-critic scheme is stable and convergent. To weaken these conditions, we introduce a variant that performs multiple TD steps per policy update and derive an explicit lower bound on the number of TD steps required to ensure stability. Finally, we establish sub-linear convergence when the number of TD steps grows logarithmically with the number of policy updates, and linear convergence when it grows linearly under a concentrability assumption.
format Preprint
id arxiv_https___arxiv_org_abs_2602_10838
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Mirror descent actor-critic methods for entropy regularised MDPs in general spaces: stability and convergence
Zorba, Denis
Šiška, David
Szpruch, Lukasz
Optimization and Control
We provide theoretical guarantees for convergence of discrete-time policy mirror descent with inexact advantage functions updated using temporal difference (TD) learning for entropy regularised MDPs in Polish state and action spaces. We rigorously derive sufficient conditions under which the single-loop actor-critic scheme is stable and convergent. To weaken these conditions, we introduce a variant that performs multiple TD steps per policy update and derive an explicit lower bound on the number of TD steps required to ensure stability. Finally, we establish sub-linear convergence when the number of TD steps grows logarithmically with the number of policy updates, and linear convergence when it grows linearly under a concentrability assumption.
title Mirror descent actor-critic methods for entropy regularised MDPs in general spaces: stability and convergence
topic Optimization and Control
url https://arxiv.org/abs/2602.10838