An Optimal Tightness Bound for the Simulation Lemma

TL;DR

This paper derives a tight, constant-factor bound for value-prediction errors in reinforcement learning, improving the classical simulation lemma by better handling probability errors and broadening its applicability.

Contribution

It introduces a novel, tight bound for the simulation lemma that accounts for model misspecification more accurately and applies to hierarchical abstraction.

Findings

The new bound is tight and includes constant factors.

Existing bounds are shown to be loose and vacuous for large discount factors.

The technique improves bounds in hierarchical abstraction.

Abstract

We present a bound for value-prediction error with respect to model misspecification that is tight, including constant factors. This is a direct improvement of the "simulation lemma," a foundational result in reinforcement learning. We demonstrate that existing bounds are quite loose, becoming vacuous for large discount factors, due to the suboptimal treatment of compounding probability errors. By carefully considering this quantity on its own, instead of as a subcomponent of value error, we derive a bound that is sub-linear with respect to transition function misspecification. We then demonstrate broader applicability of this technique, improving a similar bound in the related subfield of hierarchical abstraction.