arXiv:2609.40259
Self-testing the toric code against classical communication
Gustavo Fróes, Amanda Wei, Carlos de Gois, Denis Rochette, Marc-Olivier Renou
Abstract
Device-independent self-testing certifies a multipartite quantum state from the correlations it produces alone, without assumptions on the inner workings of the devices. Most existing self-tests, however, break down as soon as the parties can exchange messages. We consider a relaxed setting in a quantum network, where in the honest version of the protocol, the parties never communicate, yet certification remains sound even in an untrusted case where the untrusted devices secretly exchange classical messages within a bounded distance in the network. We show that the code space of the toric code, distributed over a network with the geometry of a torus, can be self-tested in this setting in a noise-robust way, against communication up to distances proportional to the network diameter. We also develop an algorithmic method for constructing such protocols and extend our results to the surface code and cluster states, robust to one round of nearest-neighbor classical communication
AI Summary
Can a quantum error-correcting code subspace be self-tested when devices exchange bounded-range classical information? This paper constructs protocols for the toric-code subspace that retain a separate extraction isometry at every party despite such communication. The communication model allows devices to learn all measurement inputs within graph distance $r$, but excludes messages depending on quantum-measurement outcomes; it therefore does not cover arbitrary LOCC. For an $L\times L$ toric-code lattice, an analytical protocol using single-qubit Pauli measurements tolerates $r$ communication rounds whenever $L\geq 6r+3$. The required lattice size scales optimally as $\Theta(r)$, although the constant is not established as optimal. If every tested correlator differs from its ideal value by at most $\epsilon$, the extracted state's norm distance from the code subspace is bounded by $(\sqrt{2}+8)L^2\sqrt{\epsilon}$. The construction combines four stabilizer-derived submeasurements to certify local $X$–$Z$ anticommutation, then separately certifies the code stabilizers. Enlarged primal and dual loops, together with inputs assigned to parties whose outputs are discarded, match the local contexts needed for operator cancellations despite communication. An integer-linear-programming formulation automates searches for these certificates. Restricted searches find toric-code constructions with $L=2r+2$ for all tested distances, conjectured minimal, and one-round-robust certificates for a $4\times4$ surface code and a $6\times6$ cluster state without auxiliary parties. The robustness bound deteriorates with system size. The numerical searches restrict contexts and certificate lengths, and no general direct construction for arbitrary graph states or protection against quantum communication is established.
Central claim may not hold: The abstract's communication-robust self-testing claim is established only when messages cannot depend on quantum-measurement outcomes, a substantive restriction absent from the abstract.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.