Hierarchical quantum decoders

Decoders are a critical component of fault-tolerant quantum computing. They must identify errors based on syndrome measurements to correct quantum states. While finding the optimal correction is NP-hard and thus extremely difficult, approximate decoders with faster runtime often rely on uncontrolled heuristics. In this work, we propose a family of hierarchical quantum decoders with a tunable trade-off between speed and accuracy while retaining guarantees of optimality. We use the Lasserre Sum-of-Squares (SOS) hierarchy from optimization theory to relax the decoding problem. This approach creates a sequence of Semidefinite Programs (SDPs). Lower levels of the hierarchy are faster but approximate, while higher levels are slower but more accurate. We demonstrate that even low levels of this hierarchy significantly outperform standard Linear Programming relaxations. Our results on rotated surface codes and honeycomb color codes show that the SOS decoder approaches the performance of exact decoding. We find that Levels 2 and 3 of our hierarchy perform nearly as well as the exact solver. We analyze the convergence using rank-loop criteria and compare the method against other relaxation schemes. This work bridges the gap between fast heuristics and rigorous optimal decoding.

💡 Research Summary

The paper introduces a novel family of quantum decoders based on the Lasserre Sum‑of‑Squares (SOS) hierarchy, offering a systematic trade‑off between decoding speed and accuracy while retaining provable optimality guarantees. The authors start by formulating the decoding problem for stabilizer codes as a Maximum‑Likelihood Decoding (MLD) task: given a syndrome s and parity‑check matrix H, find a binary error vector e that satisfies H e = s and maximizes the likelihood under independent error probabilities. This combinatorial problem is NP‑hard, and existing practical decoders (e.g., Minimum‑Weight Perfect Matching, Union‑Find) are fast but sub‑optimal, whereas exact solvers such as Mixed‑Integer Programming (MIP) are accurate but computationally prohibitive.

To bridge this gap, the authors rewrite MLD as a polynomial optimization problem with a linear objective f(e)=∑ γ_i e_i and polynomial constraints enforcing binary variables (e_i² − e_i = 0) and syndrome consistency (H_j e − s_j = 0). They then apply the Lasserre SOS hierarchy: at level ℓ, a Lagrangian is constructed with SOS multipliers of degree ℓ + 1 for the binary constraints and degree ℓ for the syndrome constraints. The resulting condition “L(x,q,q′) − λ ≥ 0” is expressed as a semidefinite program (SDP) that maximizes λ. Theory guarantees that as ℓ → ∞ the SDP optimum converges to the true MLD optimum, and the rank‑loop condition (equality of the ranks of consecutive moment matrices) provides a certificate of exactness at finite ℓ.

Numerical experiments focus on rotated surface codes and honeycomb color codes under independent bit‑flip noise (syndrome measurements assumed perfect). The authors use the MATLAB‑based SparsePOP package to exploit problem sparsity and solve the SDPs for hierarchy levels ℓ = 1 … 5. Results show that level 1 behaves like a linear‑programming relaxation, suffering from “pseudo‑codewords” (fractional solutions) and delivering poor logical error rates. Starting from level 2, performance improves dramatically: logical error rates p_L at physical error rate p = 0.05 drop from ~0.023 (ℓ = 2) to ~0.025 (ℓ = 3) and become virtually indistinguishable from the exact MIP decoder (p_L ≈ 0.0248). Threshold estimates obtained by finite‑size scaling indicate thresholds p_th ≈ 0.083 (ℓ = 2), 0.085 (ℓ = 3), and 0.093 (ℓ = 4), closely matching the MIP threshold (~0.084). The moment matrix rank stabilizes at ℓ = 5, confirming exactness via the rank‑loop condition.

The discussion compares the SOS hierarchy with other lift‑and‑project relaxations such as Sherali‑Adams and Lovász‑Schrijver, noting that Lasserre yields the tightest bound at each level but at higher computational cost. The authors acknowledge that current SDP solvers are slower than graph‑matching heuristics, limiting immediate applicability to large‑scale codes. However, they argue that the method’s ability to provide a rigorous optimality certificate makes it valuable for benchmarking medium‑size codes, exploring new code families, and possibly guiding the design of hardware‑accelerated SDP solvers. They also highlight connections to QUBO formulations used in quantum annealing, emphasizing that unlike heuristic QUBO solvers, the SOS approach guarantees convergence to the exact solution.

Future directions include exploiting additional sparsity patterns, developing customized SDP algorithms (e.g., chordal decomposition), parallel GPU implementations, and extending the hierarchy to compute code distances, another NP‑hard problem. Overall, the work establishes a principled, tunable framework that unifies fast heuristics and exact solvers, offering a clear path toward high‑performance, provably near‑optimal quantum decoding.