The Signal
Study Finds Coherent Errors Can Drive Honeycomb Surface Code Into a Critical Regime
A study of the honeycomb surface code finds that coherent Z errors can drive a transition into a critical regime with long-range entanglement and a diverging Markov length. The paper also connects symmetry and surviving quantum information to entanglement, and relates conditional mutual information to conductance.

Coherent errors can drive a surface code into a critical regime
Coherent errors can drive the honeycomb surface code from its clean code phase into an extended regime that is both critical and long-range entangled, according to an arXiv preprint posted on November 2 by Amir-Reza Negari, Subhayan Sahu, Jan Behrends, Benjamin Béri and Timothy H. Hsieh. The paper describes the regime as having a diverging Markov length and as a quasi-long-ranged counterpart of 1-form strong-to-weak spontaneous symmetry breaking. Its summary places the transition at a coherent Z-error angle of approximately 0.09π.
The work addresses how errors affect not only a code’s ability to preserve quantum information but also the phase and entanglement structure of its noisy states. Coherent errors are systematic rotations of constituent qubits. The paper notes that they pose challenges for existing approaches: after syndrome measurements, the noisy states are not explicitly connected by even a one-way Lindbladian evolution.
Symmetry and coherent information imply long-range entanglement
The authors establish two general results. First, when noise acting on a topological code preserves a higher-form symmetry, positive coherent information implies that the noisy state is long-range entangled for every generic encoded input. In the paper’s introduction, this is framed in terms of the code’s anomalous higher-form symmetry. The result links a symmetry condition and surviving quantum information to the entanglement properties of the noisy mixed state.
The paper distinguishes long-range entangled mixed states from short-range entangled ones by whether they can be approximated by a convex mixture of short-range-entangled pure states. Its discussion of earlier work describes a different pattern for the toric code under bit-flip noise: there, the transition in quantum memory coincides with a change from long-range to short-range entanglement. The authors say the relation among error thresholds, phase structure and entanglement had not been understood in the same way for coherent errors.
A recovery result extends beyond Lindbladian evolution
The second general result extends a local recovery construction based on finite Markov length to any locally parameterized family of states. The paper says this includes local Lindbladian evolution, coherent errors followed by syndrome dephasing, and thermal states. The underlying idea is that when conditional correlations decay exponentially with the width of a buffer between regions, their characteristic scale—the Markov length—can be used to assess whether local recovery is possible.
For the honeycomb surface code, the authors apply these results to coherent errors and identify two regimes. In the code phase, the paper’s summary describes full quantum memory, local decoding and long-range entanglement for every logical input. In the critical regime, it describes partial quantum memory that requires global decoding, with long-range entanglement for generic logical inputs. The regime’s diverging Markov length distinguishes it from the code phase, where the Markov length is finite.
Conditional mutual information tracks conductance
The paper also relates conditional mutual information (CMI) to conductance in a Majorana network model. CMI measures correlations between regions that are not mediated by a buffer separating them. In the paper’s summary, CMI and conductance decay exponentially with system size in the code phase, which has finite Markov and localization lengths. In the critical regime, both are described as scaling logarithmically with system size, while the Markov and localization lengths diverge. This connection gives the authors a way to relate conditional correlations in the noisy code to transport in the network model.