arXiv:2609.21886
Ensemble Dependence of the Critical Exponent at a Quantum Error Correction Threshold
Idan Dror, Moshe Goldstein
Abstract
In thermodynamics it is common to assume that the choice of ensemble (e.g., micro-canonical, canonical, or grand-canonical) should not affect the underlying physics in the thermodynamics limit. We show that this does not necessarily hold for critical exponents. Examining a simplified model of quantum error correction (single step encoding and decoding by a random unitary) and the behavior of both the fidelity and magic at the corresponding threshold, we find different exponents when using generic channels or supposedly equivalent quantum trajectories. Interestingly, the obtained exponents saturate a recently derived information theoretic bound by Feldman et al. (2026), which we extend from the grand-canonical to the canonical case, including intermediate ensembles which we define. Moreover, even the existence of the transition is shown to be ensemble-dependent.
AI Summary
The paper asks whether finite-size critical exponents at an error-correction threshold depend on fixing error counts versus allowing them to fluctuate, and on averaging channels versus averaging trajectory-resolved fidelities. It studies Haar-random encoding of $k=rN$ logical qudits into $N$ qudits, followed by product noise and inverse-unitary decoding. The observable is fidelity conditioned on zero ancillary syndrome, used as a proxy for decoding success. For fidelity averaged separately over encodings in numerator and denominator, the calculation reduces to a multiplicative channel parameter $R$. Writing the mean single-site strength as $\bar E=\langle-\log_d R\rangle$, the threshold is $\bar E_c=1-r$. Fixed error counts give a transition width proportional to $N^{-1}$, hence $\nu=1$; independent fluctuating counts broaden it through the central limit theorem to $N^{-1/2}$, hence $\nu=2$, when the strength variance is finite and nonzero. Averaging channels before conditioning instead gives $\nu=1$ and can shift the threshold. A depolarizing-channel example shows that trajectory decomposition can even remove the nonzero-noise threshold. Information-theoretic sampling arguments give bounds $\nu\geq1$ and $\nu\geq2$ for canonical and independent-error ensembles, respectively, and an interpolating reservoir-dependent bound. Related stabilizer Rényi entropy scaling is also discussed. Brick-wall-circuit simulations support the predicted fidelity scaling for averages taken after conditioning. The analytic results concern encoding-annealed fidelity; agreement with fully quenched encoding averages is supported numerically rather than generally proved. The decoding-success interpretation assumes comparable performance in nonzero-syndrome sectors. The model uses global random encoding, with local circuits tested numerically.
Central claim may not hold: The exact exponent calculations concern encoding-annealed fidelity; extending them generally to encoding-quenched fidelity relies on an unproved self-averaging expectation and numerical evidence. The calculated postselected fidelity is only an assumed proxy for full error-correction performance, not an established threshold for a complete recovery procedure. The magic derivation replaces an average of negative logarithms by the negative logarithm of average fidelity, expressions that differ parametrically at the grand-canonical critical point. The claimed vanishing trajectory-fidelity variance contradicts the paper's own critical scaling formulas, which imply a limiting variance of 1/4 for nonzero error-strength variance.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.