Optimization Using Locally-Quantum Decoders

Abstract: It was pointed out in [JSW+25] that widely-studied optimization problems such as D-regular max-k-XORSAT can be reduced to decoding of LDPC codes, using quantum algorithms related to Regev's reduction. LDPC codes have very good decoders, such as Belief Propagation (BP), and this therefore makes D-regular max-k-XORSAT an enticing target for this class of quantum algorithms. However, BP was found insufficient to achieve quantum advantage. Here, we develop an intrinsically quantum decoding technique, which decodes classical LDPC codes subject to coherent superpositions of bit flip errors. For average-case instances of D-regular max-k-XORSAT drawn from Gallager's ensemble, this quantum decoder strongly outperforms classical belief propagation at many values of k and D. For some (k,D) the approximate optima achievable using this decoder surpass both Prange's algorithm and simulated annealing. However, we stop short of achieving quantum advantage because we identify an enhancement to Prange's algorithm that recovers a precise tie, much as a precise tie was observed between the standard version of Prange's algorithm and a more limited version of locally-quantum decoding in [CT24].

• The paper introduces locally-quantum decoding that blends tailored quantum measurements with classical post-processing to tackle NP-hard max-k-XORSAT problems.
• The paper demonstrates that the Regev+FGUM decoder achieves higher satisfaction fractions than classical BP and simulated annealing for specific (k, D) pairs, though Turbo Prange matches its performance.
• The paper compares quantum decoders with QAOA, providing rigorous error bounds and performance analyses that inform future advancements in global quantum measurement strategies.

Optimization Using Locally-Quantum Decoders: A Technical Overview

Problem Setting and Reductions

The paper addresses the potential for quantum algorithms to provide advantages in combinatorial optimization, with a focus on average-case $D$-regular max-$k$-XORSAT. The authors relate max-$k$-XORSAT, an archetypal NP-hard constraint satisfaction problem, to the decoding of classical LDPC codes using quantum reductions inspired by Regev's approach. Specifically, they formalize a reduction whereby optimizing over binary variables subject to random parity constraints can be mapped to recovering codewords from quantum-corrupted channels, a transformation that is central to both Regev's reduction and Decoded Quantum Interferometry (DQI) [JSW25].

Max-$k$-XORSAT instances are constructed by drawing the constraint matrix $B$ from the Gallager ensemble, generating sparse parity check matrices canonical for testing decoders. Solutions involve finding $\mathbf{x} \in \mathbb{F}_2^n$ minimizing $|B\mathbf{x} - \mathbf{v}|$, which, when unsatisfiable, is computationally hard. Recent work shows that DQI can achieve exponential speedup for certain algebraic optimization problems, but the efficacy of quantum approaches for less structured problems like max-$k$-XORSAT is less established.

Quantum Decoding Problem Formulation

The reduction implies preparing quantum states that bias toward good solutions and subsequently pose a "quantum decoding problem": given a superposition of codewords corrupted by coherent superpositions of bit-flip errors, recover the original codeword efficiently. The paper elucidates the technical formulation whereby a suitable function $P$ biases toward low objective values, and its Hadamard transform $\widetilde{P}$ determines the error channel. For LDPC codes, classical decoding via belief propagation (BP) is often efficient, but the quantum analogue is non-trivial, especially under quantum noise models with coherent superpositions.

Locally-Quantum Decoding Schemes

The authors introduce the framework of "locally-quantum decoding," blending quantum measurements with code-specific strategies. Locally-quantum schemes operate by performing local quantum operations (e.g., change of basis or unambiguous state discrimination) on sub-blocks of qubits, exploiting code structure (especially sparsity and block partitions from Gallager ensembles), and then applying classical post-processing. Prior locally-quantum approaches have not yielded quantum advantage, as their performance could be matched by efficient classical algorithms such as Prange's algorithm.

Fine-Grained Unambiguous Measurements (FGUM)

A central technical contribution is the development of decoding techniques leveraging blockwise unambiguous measurements (FGUM) tailored to the codes arising from Gallager's ensemble. The authors construct block partitions of qubits aligned with check nodes and implement FGUM that allow recovery of the codeword bits with enhanced resilience to erasure-type errors. They provide a detailed combinatorial analysis bounding the fraction of blockwise erasures that are recoverable via Gaussian elimination, leading to tight estimates of the achievable satisfaction fraction.

Analytically, they derive formulas for the maximum correctable erasure rates and satisfaction fractions, showing that their quantum decoder (Regev+FGUM) outperforms both simulated annealing and classical Prange's algorithm on certain $k$0 pairs. For example, empirical scores show Regev+FGUM achieving 0.8930 for $k$1 compared to Prange's 0.875 and simulated annealing's 0.9366.

Classical Algorithmic Matching: Turbo Prange

Despite the performance gap over classical BP and simulated annealing, the authors observe that the satisfaction fractions achieved by Regev+FGUM are matched precisely by a classical heuristic adapted from Prange's algorithm (Turbo Prange), which combines structured variable selection and greedy updates. This negates claims of outright quantum advantage for max-$k$2-XORSAT in this regime, aligning with previous negative results on code-agnostic measurements [buzet2025fine], and highlighting the subtlety of quantum speedups when classical heuristics are carefully optimized.

Error Analysis and Robustness

The manuscript provides rigorous bounds on the effect of quantum decoding failures, relating the decoding error probability $k$3 to the degradation in constraint satisfaction. In particular, formal statements show that for max-XORSAT with uniformly random $k$4, a failure rate $k$5 reduces the expected satisfaction fraction by at most $k$6, allowing robust guarantees in practical settings.

Quantum Approximate Optimization Algorithm (QAOA) Comparison

The authors systematically benchmark their quantum decoder against QAOA, leveraging efficient high-girth tree contraction methods to compute QAOA energies for large $k$7 and finite depth $k$8. At depth $k$9, QAOA outperforms Regev+FGUM for certain $k$0 configurations and is predicted to be asymptotically superior for large $k$1 and fixed $k$2. Explicit scaling formulas are provided for both turbo Prange/Regev+FGUM and QAOA, showing, for example, that QAOA with $k$3 rounds achieves satisfaction fractions exceeding those of Regev+FGUM for all $k$4 considered.

Numerical Results and Claims

The comparison table and asymptotic analyses highlight several bold claims:

• Locally-quantum decoding using FGUM achieves superior satisfaction fractions compared to classical belief propagation and outperforms Prange and simulated annealing for some $k$5 pairs.
• No quantum advantage is claimed, as an enhanced classical algorithm (Turbo Prange) matches the quantum performance exactly, even with code-dependent FGUM measurements.
• QAOA at sufficient depth surpasses all classical and locally-quantum decoders for large $k$6 and fixed $k$7, and outperforms Regev+FGUM quantitatively for practical depths and several parameter choices.

Implications and Future Directions

Theoretical implications are multifold:

• Locally-quantum decoding, leveraging code structure via tailored quantum measurements, can achieve optima beyond classical belief propagation but is limited by the existence of efficiently matchable classical heuristics.
• The tight correspondence between fine-grained quantum measurements and classical decoding for LDPC codes drew attention to the robustness and limitations of quantum advantage in combinatorial optimization.
• The failure to achieve quantum advantage under locally-quantum strategies motivates investigations into more global, code-aware quantum measurements or adaptive strategies, potentially invoking belief propagation with quantum messages as in [mandal2026belief, piveteau2025efficient].

Practically, the results suggest that quantum algorithms are at the threshold of outperforming classical heuristics on average-case optimization over LDPC codes and constraint satisfaction, with further development in quantum decoding methods likely necessary to cross this boundary. The comparison with QAOA underscores that circuit depth and parameter optimization are critical levers; high-depth QAOA is predicted to surpass classical and quantum decoders in the large-$k$8 regime, with ongoing efforts to optimize QAOA and analyze its boundaries for different code structures.

Conclusion

This work systematically defines and analyzes the potential for quantum advantage in average-case max-$k$9-XORSAT via locally-quantum decoding, introducing code-dependent FGUM and elucidating rigorous performance bounds. While quantum decoders employing FGUM outperform classical BP and simulated annealing, enhanced classical algorithms (Turbo Prange) match their performance precisely, precluding quantum advantage on the testbed ensemble. The study provides a comprehensive framework for benchmarking quantum and classical decoders and sets the stage for future developments in both theory and practical quantum optimization, with the evolution of quantum message-passing and adaptive decoding seen as promising avenues.