arXiv:2611.02227
Quantum Temporal Marginal Problem
Dingqian Wang, Jiayi Zhao, Ge Bai
Abstract
Experiments on a multi-time quantum process may reveal its responses only on selected, possibly overlapping, sets of times. We ask whether these partial descriptions admit a common underlying process, and formulate this process tensor marginal problem as a semidefinite feasibility test. Agreement on overlaps is necessary but insufficient: three locally consistent qubit marginals can have no common extension. Another family is incompatible as a time-ordered process even though a single quantum channel satisfies both its channel and temporal marginal constraints. This isolates temporal causality as the obstruction. We quantify incompatibility by the mixing needed to restore compatibility and derive finite tests using separate interventions at each time and final measurements. We further show that incompatibility is equivalent to an advantage over every compatible family in a multi-time process-discrimination task. Together, these results test whether process descriptions obtained in different experiments can arise from one causally ordered quantum process.
AI Summary
The paper asks whether process descriptions reconstructed on selected, possibly overlapping sets of times admit one common causally ordered process tensor. Temporal marginalization retains selected intervention slots, closes omitted earlier slots with identity channels, and truncates at the final selected time; it is not ordinary partial tracing of the Choi operator. Compatibility is formulated exactly as semidefinite feasibility using positivity, recursive deterministic-comb normalization, and linear marginal constraints. With unrestricted finite-dimensional memory, marginals on pairwise-disjoint selected time sets always have an extension, even when their time intervals interleave. Agreement on overlaps is insufficient: three qubit marginals prescribing nearest-neighbor identity propagation and endpoint bit-flip propagation are incompatible despite identical maximally mixed one-time marginals. Their numerically computed retained-weight robustness is $R\approx0.75$, where $R=1$ denotes compatibility. A second, classically encoded qubit construction admits a common channel satisfying both channel and temporal marginal constraints but no causal comb extension; an explicit positive-operator certificate establishes the obstruction. The paper derives an SDP and strong duality for robustness, finite intervention-and-measurement witnesses, and an equivalence between incompatibility and advantage over every compatible family in some multi-time process-discrimination task. These follow from convex separation, control-space spanning, and deterministic completion of CP controls; the discrimination comparison fixes ensembles and measurements. The framework assumes finite-dimensional, fully specified marginal processes and unrestricted memory. Extensions to incomplete or noisy data, classification of overlap patterns permitting extensions, and testing for Markovian extensions remain open. The discrimination result guarantees existence of a suitable task, not advantage for arbitrary ensembles.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 29 pages including appendices, 5 figures