How many noisy time machines does it take to make a grandfather paradox?
Zongmin Liu
Abstract
General relativity admits closed timelike curves, and quantum mechanics on such a loop is widely believed to defuse the grandfather paradox: in the postselected (P-CTC) model, paradoxical histories are simply assigned probability zero. Here we turn that belief into a sharp, quantitative question. We show that the Doeblin information governing the retrocausal capacity of a noisy quantum channel [Phys.\ Rev.\ Lett.\ \textbf{136}, 230801 (2026)] has a dual description as the \emph{minimal loop amplitude}: the smallest total probability that any legal intervention can leave on a time loop built from that channel. A grandfather paradox is realizable precisely when this amplitude can be driven to zero. This yields a new invariant of a quantum channel --- its \emph{paradox number} $n_0$, the number of independent time loops needed before some entangled intervention annihilates every consistent history. We prove that any channel with full Choi rank has $n_0=\infty$ (decoherence protects chronology, no matter how many machines are stacked), that ideal machines have $n_0=1$, and, for the experimentally central amplitude-damping (photon-loss) channel, we solve the problem exactly: $n_0(\gamma)=\lceil \ln 2/\ln(1+\sqrt{1-\gamma})\rceil$, a staircase whose first step is the algebraic threshold $\gamma^*=2(\sqrt2-1)$. Equivalently, each loop of transmissivity $\eta$ contributes $\log_2(1+\sqrt{\eta})$ bits toward the single bit that a paradox costs. The proofs are constructive on both sides, and they resolve, for this channel family, the regularization problem left open for the retrocausal capacity.