Causal postselection and the Stein exponent for quantum superchannels
A proof candidate that general causal interleavings, nested strategies and parallel strategies have the same superchannel Stein exponent, with an exponential strong converse.
Yuxuan Zhang · 13 November 2026 · Version 1.0
A quantum superchannel transforms one channel into another. Physically, it has a first component, an accessible channel slot, and a second component connected to the first by a hidden memory. When many uses are available, a tester can nest them or interleave their components while respecting causality. Can this improve the asymptotic discrimination rate?
The proposed theorem says that it cannot in the Stein setting:
\[\zeta_{\mathrm{general}} =\zeta_{\mathrm{nested}} =\zeta_{\mathrm{parallel}} =D_{\mathrm{sc}}^\infty(\Theta_1\Vert\Theta_2).\]It covers every finite-dimensional pair, with arbitrary finite tester memory and no access to the superchannels’ internal memories. At a type-II rate strictly above the displayed value, the probability of accepting the null hypothesis tends to zero exponentially.
This is a complete proof candidate with sequential internal analytical review. Independent expert review and comprehensive priority assessment remain outstanding.
Full paper (PDF) · Source, checks and review (ZIP) · Version and SHA-256 metadata
Two catalogue questions, one result
The nested-versus-parallel question asks whether nesting improves the vanishing-type-I Stein exponent. The general-versus-nested question also allows the first and second components of different uses to be interleaved, provided each first component precedes its matched second.
The theorem answers both with the same equality. The nested result is a corollary, so these are overlapping answers rather than two independent discoveries. The paper uses precisely this causal access model; it does not grant the tester access to the hidden memories.
The causal constraint in the proof
For ordinary channels, positive-slack and postselection methods can turn a parallel approximation into a bound for more general testers. A superchannel adds a constraint: its second input cannot affect its first output. A generic channel approximation on the combined slots need not preserve that constraint.
The first step here gives a polynomial domination of a symmetric joint two-stage comb by a mixture of tensor powers of physical one-copy combs. Canonical dilations keep their memory dimensions in tensor-power form and make the required permutation covariance explicit. Fixed-marginal postselection is applied at each stage, followed by contraction of the internal memory legs.
The second step constructs one approximation that works for every parallel tester. A strategy norm measures its error. Tensor amplification makes the error exponential, while the causal postselection costs only a fixed-degree polynomial. Every permitted general tester is normalized on the physical iid combs in that mixture, which completes the converse.
What is inherited
Hirche develops the superchannel strategy classes and parallel rate. The amplification argument is adapted from Gao, Ji and Liu, and the positive-slack and postselection approach closely parallels Zhu and Wang’s ordinary-channel theorem.
The isometry domination uses the fixed-marginal result of Nahar and collaborators. Gutoski and Watrous supply the finite-strategy realization and positive domination duality. These ingredients are credited to their sources. The proposed extension is their combination with physical two-stage postselection and the strategy-amplitude norm.
The package distinguishes this finite-n statement from infinite exchangeability for causal processes and from ordinary-channel marginal de Finetti results. The specific comparisons support the stated difference; they do not establish exhaustive priority.
Review and a corrected detail
Review found that a fixed rank-deficient tester’s dual optimization need not attain its infimum. The package contains a physical qubit-channel counterexample, including one above the relevant regularized rate. The paper now keeps an infimum at that step and proves attainment only after optimizing uniformly over testers. A full-rank tester supplies the needed compactness, so the theorem’s scope is unchanged.
Exact controls also reconstruct the Haar normalization through a permutation Gram pseudoinverse at two and three qubit copies, including a singular Gram matrix. Incorrect constants are detected. The all-dimensional claim rests on the analytic proof, not these finite controls.
An alternative nested-only proof is included in the source package. It uses the external CP trace-nonincreasing approximation theorem with a success flag and a uniform limit argument; it is not counted again as a new result.
The author directed the research and workflow. AI systems assisted with derivation, literature comparison, computation and drafting. This website release is separate from journal review, archival submission and catalogue acceptance.