arXiv:2611.02339
Achievability of CSS codes performance limits over Pauli channels under separate ML decoding with bounded graphical complexity
Gulnihal Mercanoglu, Achilleas Anastasopoulos, S. Sandeep Pradhan
Abstract
Capacity-approaching quantum error correction faces a fundamental tension: the required parity-check matrices need to become increasingly dense, while practical decoding and syndrome extraction favor sparse, bounded-degree structures. Reconciling these competing requirements remains an important open problem in quantum code design. We construct Calderbank-Shor-Steane (CSS) code ensembles that approach the CSS rate over memoryless Pauli channels under maximum-likelihood decoding while retaining sparse, bounded-degree graphical representations. The construction separates parity-check density from graphical complexity: although capacity-approaching components may require dense parity-check matrices, they are represented by sparse concatenated factor graphs. We first study a construction combining low-density parity-check (LDPC) and low-density generator-matrix (LDGM) component codes and numerically analyze the average weight distributions of the resulting codes. Motivated by these results, we introduce a second construction based on concatenated LDPC and LDGM codes and generalized irregular repeat-accumulate (GIRA) codes. We derive bounds on the average weight distribution of the resulting sum ensemble and use the Miller-Burshtein framework to prove simultaneous reliability of the two CSS decoding components. Consequently, the proposed ensemble achieves rates arbitrarily close to the CSS hashing bound under separate bit- and phase-error correction, while maintaining bounded graphical complexity, thus providing a foundation for future iterative decoders and indirect syndrome-extraction architectures.
AI Summary
The paper asks whether nested CSS ensembles can approach the rate attainable with separate bit- and phase-error decoding while retaining bounded-degree graphical representations. For a memoryless Pauli channel, the benchmark is $R_{\mathrm{CSS}}=1-h_2(p_X+p_Y)-h_2(p_Z+p_Y)$, which can lie below the unrestricted hashing bound because separate decoding ignores bit–phase correlations. Both constructions enforce nesting through $\mathcal C_2^\perp=\mathcal C_1^\perp+\widetilde{\mathcal C}$. Construction A combines an LDPC first component with an auxiliary LDPC code. Numerical weight-spectrum calculations suggest that the sum component closely follows the random-code spectrum, but the regular-LDPC component remains limiting; the small-weight contribution for the sum component is not established in this construction. Construction B instead uses a concatenated LDPC/LDGM first component, whose dual has a generalized repeat–accumulate representation. The stated main theorem establishes simultaneous vanishing ML block-error probabilities when both component design rates lie below their respective marginal BSC capacities, using fixed finite graph degrees independent of blocklength and capacity gap. The analysis derives ensemble-average weight-enumerator bounds for the dual and sum ensembles, controls small-weight codewords, and applies the Miller–Burshtein framework together with an existing LDPC/LDGM achievability result. Bounded graphical complexity concerns extended factor graphs with auxiliary variables, not sparse stabilizers. ML decoding remains generally exponential; capacity-achieving belief propagation and fault-tolerant syndrome extraction are not demonstrated. Indirect measurements exploiting the sparse constituent structure are left for future work.
Central claim may not hold: The sum-spectrum proof takes boundary tolerances to zero while its interior degree requirements depend on the shrinking boundary-strip width, leaving the claimed fixed-degree achievability guarantee without a uniform parameter-selection argument. The small-weight appendix subtracts rather than adds polynomial decay exponents when multiplying two bounds, although correcting that arithmetic removes the associated objection to its final minimum exponent. Construction A's table reports sum-code threshold lower bounds without establishing the small-weight condition required to certify those bounds. The sum-spectrum Hessian proof uses a false numerical inequality to obtain its stated curvature bound, although a weaker valid curvature bound still permits strict concavity.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 45 pages, 9 figures