Explain the empirical acceleration of dueling Q-learning

Explain why dueling Q-learning empirically accelerates learning relative to standard Q-learning, beyond the theoretical guarantees established for the residual-preconditioned advantage-based algorithm analyzed in the paper.

Background

The paper develops an advantage-based value-learning algorithm with convergence and finite-sample guarantees under Markovian sampling. Although the analysis establishes contraction improvements for the proposed residual-preconditioned method, it does not account for the empirically observed acceleration of dueling Q-learning over ordinary asynchronous Q-learning. Identifying the mechanism behind that empirical advantage is explicitly listed as an unresolved issue.