AI-Assisted Channel Separates Zero Capacity from PPT and Antidegradability
Chengkai Zhu and Xin Wang report an AI-assisted explicit qutrit channel with zero private and quantum capacities that is neither PPT nor antidegradable, a claimed counterexample to two proposed characterizations; the model versions are unspecified, and the retained sources do not establish peer review.
Chengkai Zhu and Xin Wang report an explicit qutrit quantum channel whose private and quantum capacities both vanish although the channel is neither positive under partial transposition nor antidegradable, providing a claimed counterexample to two proposed characterizations of zero-capacity channels.
Their preprint proves vanishing optimized coherent and private information at every blocklength for the qutrit noisy Werner–Holevo channel. Zhu's thread says frontier LLM models helped identify the result and that a harnessed AI model suggested the key signed-lift proof idea; the authors then checked and formalized the AI-produced arguments and developed the complete analysis.
The evidence consists of the authors' announcement and an arXiv preprint. The thread does not identify the model products or versions, and the retained sources do not establish peer review, so both the mathematical claim and the scope of AI assistance remain attributed to the authors.
Sources: Chengkai Zhu's announcement thread
Related Materials: Quantum Incapacity beyond No-Cloning and PPT Mechanisms