arXiv:2609.19310
Quantum inflation and source-transfer hierarchies are complete for tree networks
Xiangling Xu, Igor Klep, Marc-Olivier Renou
Abstract
Understanding the capabilities and limits of quantum systems is central to quantum information processing. This includes determining which correlations can arise in networks of quantum systems distributed by independent sources. Yet for general quantum networks, no systematic, dimension-free method is known to bound these correlations arbitrarily well. We resolve this problem for all quantum tree networks through a new source-transfer model, which we prove equivalent to a mixed quantum model. The mixed model places no restriction on dimension and reduces to the standard tensor-product model in finite dimensions; the source-transfer model instead describes these correlations through states on $C^*$-algebras and completely positive maps composed recursively along the tree. We prove the equivalence by reconstructing local measurements as Radon-Nikodym derivatives in the spatial tensor products prescribed by the network. Building on this characterization, we establish two convergent outer hierarchies of semidefinite programs (SDPs): a new source-transfer hierarchy and the inflation-NPA hierarchy. For the source-transfer hierarchy, we extend noncommutative real algebraic geometry to multi-state polynomials with completely positive maps, prove a reconstruction theorem establishing its convergence, and give a sufficient stopping criterion for extracting finite-dimensional realizations. For the inflation-NPA hierarchy, we reconstruct source-transfer realizations from weak-operator limits of averages over source copies produced by quantum inflation. Consequently, the game-value promise problem for quantum tree networks in the mixed model is $\mathsf{coRE}$-complete.
AI Summary
Quantum inflation gives outer bounds on correlations compatible with independent quantum sources, but its convergence is unresolved for general networks. This paper proves completeness for every finite network whose bipartite source–party graph is a tree, with finite measurement-setting and outcome sets. The target is the dimension-unrestricted mixed model: independent sources have tensor-product states, while parties sharing a source have commuting observable algebras. A new source-transfer model represents the same correlations recursively using source $C^*$-algebras, states, and completely positive maps. Source independence becomes a source-replacer condition: summing a party’s outcome-conditioned maps evaluates the incoming product state and returns a scalar identity. Equivalence with the mixed model follows by reconstructing local POVMs as Radon–Nikodym derivatives in the prescribed spatial tensor products. Two SDP hierarchies characterize this correlation set: inflation-NPA and a new source-transfer hierarchy based on multi-state polynomials with completely positive maps. Every correlation outside the set is rejected at a finite level, and bounds converge for polynomial objectives. Inflation convergence uses de Finetti factorization and weak-operator limits of source-copy averages. The source-transfer hierarchy also has a sufficient finite-level stopping criterion for extracting finite-dimensional realizations. In finite dimensions, both models agree with the standard tensor-product model. The mixed-model game-value promise problem, distinguishing optimum $1$ from optimum at most $1/2$, is $\mathsf{coRE}$-complete. Completeness remains open for fully quantum cyclic networks, including the triangle. Practical implementation and comparison of hierarchy costs are deferred; a finite-level extraction criterion for inflation-NPA is also unresolved.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 84 pages, 7 figures. v2: more informative title, typos corrected. Comments welcome!