Title: Entanglement Confers Strategic Benefits in Competitive Scenarios

Original: arXiv:2603.10289v2 Announce Type: replace-cross Abstract: The question of whether exclusively quantum resources offer benefits in purely classical, competitive settings remains unresolved. Fully classical, competitive zero-sum reinforcement learning presents significant hurdles, as victory depends on understanding dynamic interactions between rival agents instead of relying on static state-action relationships. To address this, we performed a controlled experiment to isolate the specific impact of quantum entanglement on a quantum-classical hybrid agent learning to play Pong, a competitive Markov game. We utilized an 8-qubit parameterized quantum circuit as a feature extractor within a proximal policy optimization structure. This setup enabled a direct comparison between separable circuits and those featuring either fixed (CZ) or trainable (IsingZZ) entangling gates. Our results indicate that entangled circuits consistently surpass separable ones with similar parameter numbers. Furthermore, in scenarios with limited capacity, they perform on par with or better than classical multilayer perceptron benchmarks. Analysis of representation similarity reveals that entangled circuits acquire structurally unique features, which aligns with their enhanced ability to model interacting state variables. Consequently, these results identify entanglement as a functional resource for representation learning within competitive reinforcement learning contexts.

Source: arXiv Generated at: 2026-06-04 00:00:00 UTC