EasyDeepLearn

Why do Bellman operators converge?

hard

Answer

  • Bellman operator T is a γ-contraction in the max norm: ||TV1    TV2\mathrm{TV}_{1}\; - \;\mathrm{TV}_{2}||_∞ ≤ γ ||V1    V2V_{1}\; - \;V_{2}||_∞.
  • By Banach fixed-point theorem, iterating T converges to unique fixed point V* at rate γk{\gamma}^{k}.
  • Underlies value iteration, TD(0) convergence in tabular case, policy iteration.
  • Function approximation can break the contraction (projected Bellman operator may not contract) → divergence.
Check yourself — multiple choice
  • Random
  • T is a γ-contraction in max-norm → Banach fixed-point → unique V* + γk{\gamma}^{k} convergence rate; function approximation can break contraction (projected T may not contract)
  • Not real
  • Same as GD

Bellman: γ-contraction → Banach → convergence; FA can break it.

#theory

Practise Reinforcement Learning

214 interview questions in this topic.

Related questions