arXiv:2609.20708
Proof of a positive coherent-error threshold for topological quantum codes
Shiro Tamiya, Masato Koashi
Abstract
Threshold analyses of quantum error-correcting codes are well established for stochastic error models, in which errors occur randomly with given probabilities. However, errors in actual devices can also be coherent, such as unwanted $Z$ rotations due to imperfect control, which are not captured by stochastic error models. For the surface code, numerical studies have indicated threshold behavior even under coherent errors, but a rigorous proof of threshold existence is lacking. Here we prove that a positive threshold for coherent $Z$-rotation errors exists for quantum low-density parity-check codes with a bounded number of logical qubits, including the surface code and other topological codes. Specifically, we show that the maximum-likelihood Pauli recovery suppresses the entanglement infidelity exponentially in the code distance up to a prefactor linear in the number of physical qubits whenever the rotation angles lie below a constant that is independent of the code size. The proof combines Fourier analysis to retain the interference among the amplitudes of coherent errors with the cluster expansion of abstract polymer models. Our results expand the theoretical foundation of quantum error correction and offer a statistical-mechanical description of quantum error correction beyond stochastic errors.
AI Summary
Does syndrome measurement followed by Pauli recovery give a nonzero threshold for coherent errors without twirling? This paper addresses that question in the code-capacity setting for product errors $U=\bigotimes_j e^{i\theta_j Z_j}$ on CSS QLDPC codes, with noiseless syndrome extraction and maximum-likelihood Pauli recovery determined using the rotation angles. For an $(r,c)$-QLDPC family encoding $1\leq k\leq k_0$ logical qubits, define $\Delta_0=c(r-1)$ and $\delta_*=\sup_{i,j}|\tan(2\theta_{i,j})|$. If $0<\delta_*<\delta_{\mathrm{th}}=1/\{e[(e+1)\Delta_0+1]\}$, the recovered entanglement infidelity satisfies $1-F_e\leq(2^{k_0}-1)^2n e^{-bd_Z}$, where $b=\tfrac12\log(\delta_{\mathrm{th}}/\delta_*)$ and $d_Z\geq d$. Thus infidelity, and also logical diamond distance, vanish when $d/\log n\to\infty$. For rotated planar surface codes, the sufficient angle bound is $\max_j|\theta_j|<\tfrac12\arctan[1/(e(6e+7))]\approx0.0025\pi$. The proof bounds decoding failure by pairwise amplitude overlaps between distinct logical classes. Fourier analysis converts these overlaps into Fourier $1$-norm estimates while retaining interference. A polymer representation and convergent cluster expansion then cancel logically trivial contributions; surviving clusters have total size at least $d_Z$, producing exponential suppression. The result does not cover noisy syndrome extraction, circuit-level coherent faults, or practical decoders such as minimum-weight decoding. The threshold bound is substantially below numerical estimates. Before imposing bounded $k$, the prefactor is $(2^k-1)^2$, limiting applicability to high-rate families with sublinear distance.
Error found: The abstract omits the CSS restriction required by the theorem and proof when describing the covered QLDPC codes. The abstract omits the assumption of ideal syndrome measurement, so the stated threshold should explicitly be identified as a code-capacity threshold. The abstract's general threshold-existence statement omits a distance-growth qualification needed to obtain vanishing infidelity from the proved bound C n exp(-b d).
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 21 pages, 1 figure