arXiv:2611.03148
Correlated Catalysis Removes the Strong Converse for Entanglement Distillation
Piotr Masajada, Aby Philip, Alexander Streltsov
Abstract
Entanglement distillation converts noisy shared quantum states into maximally entangled pairs using local operations and classical communication. The distillable entanglement $E_D$ determines the maximum asymptotic rate at which this process can be performed reliably. For several classes of quantum states, distillation above this rate obeys a strong converse: the fidelity with the target state vanishes asymptotically. Here we show that correlated catalysis removes the strong converse for every distillable bipartite state. Specifically, at any target rate $R>E_D$, the optimal asymptotic fidelity is given by $E_D/R$. Thus, although catalysis does not increase the maximum reliable distillation rate, it allows nonvanishing fidelity at every finite rate. Remarkably, this advantage persists even when the catalyst is returned exactly at every finite blocklength and the correlations between the output and catalyst vanish asymptotically. We further characterize the catalyst size required to sustain nonvanishing fidelity above the reliable distillation rate.
AI Summary
The paper asks whether entanglement distillation retains a strong converse under correlated catalysis: a finite-dimensional bipartite catalyst must be returned exactly, but may become correlated with the output. For every finite-dimensional source with $E_D(\rho)>0$, it establishes the optimal asymptotic squared fidelity at target rate $R>0$ as $\min\{1,E_D(\rho)/R\}$. This holds under any prescribed positive mutual-information bounds $I(S:C)\leq\delta_n\to0$. Thus the reliable rate remains $E_D(\rho)$, while the nonvanishing-fidelity threshold becomes infinite. Achievability uses a batch-to-catalyst construction that converts $n$ source copies into a mixture containing a larger source block with probability proportional to the copy-number ratio, followed by ordinary distillation. An entropy identity controls output–catalyst correlations. The converse combines known catalytic rate invariance with a filtering-and-hashing bound relating target overlap to the output’s distillable entanglement. The paper also characterizes catalyst memory for fixed fidelity $0<f_0<E_D(\rho)/R$. A truncated-binomial construction achieves $O(n\log_2(1/\delta_n))$ qubits on each side. When non-catalytic distillation satisfies an exponential converse at a threshold below $R$, repeated catalyst reuse yields a matching $\Omega(n\log_2(1/\delta_n))$ lower bound on catalyst max-relative entropy of entanglement and local memory. This assumption holds for pure and maximally correlated sources at every $R>E_D$, and generally for $R$ above the Rains relative entropy. The matching memory bounds require a fixed gap below optimal fidelity; memory scaling when fidelity approaches $E_D(\rho)/R$ remains open. The results allow unrestricted catalyst resources and do not settle the strong converse for general unassisted distillation.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 13 pages