A Counterexample to Cover's 2P Conjecture on Gaussian Feedback Capacity

We provide a counterexample to Cover's conjecture that the feedback capacity $C_\textrm{FB}$ of an additive Gaussian noise channel under power constraint $P$ be no greater than the nonfeedback capacity $C$ of the same channel under power constraint $2P$, i.e., $C_\textrm{FB}(P) \le C(2P)$.