Pauli-structured preconditioning for quantum linear system solvers

Abstract: Preconditioning is a fundamental technique for accelerating classical linear system solvers, and understanding when its benefits persist in quantum linear system (QLS) solvers is important for assessing the practical resource requirements of quantum linear algebra. In QLS algorithms, however, the potential advantage of preconditioning may be offset by the normalization overhead incurred by composing separate block-encodings of the system matrix and the preconditioner, as observed in recent work. This limitation leaves open whether additional algebraic structure can make preconditioning effective in quantum access models. Motivated by this question, we show that Pauli-structured representations of both the system matrix and the preconditioner allow the preconditioned operator to be accessed through regrouped Pauli expansions. In this setting, algebraic regrouping of Pauli products can reduce the Pauli coefficient weight of the preconditioned operator, thereby altering the normalization parameters relevant to quantum algorithms. We derive explicit size and coefficient-weight bounds for the regrouped Pauli representations, and we trace their consequences for both direct block-encoding constructions and randomized Pauli linear system solvers. These results identify when Pauli-structured preconditioning can reduce the effective complexity parameters of QLS algorithms, rather than merely improving the classical condition number. Numerical experiments on a finite-dimensional synthetic benchmark show reductions in norm-aware direct block-encoding diagnostics and in the randomized QLS per-sample depth proxy.

• The paper demonstrates that classical preconditioning via separate block-encoding fails to lower quantum query complexity due to combined normalization factors.
• The paper shows that regrouping Pauli terms in direct block-encoding can significantly reduce effective normalization weight and structural depth proxies.
• The paper supports its theoretical findings with numerical experiments that highlight order-of-magnitude reductions in per-sample depth for randomized solvers.

Pauli-Structured Preconditioning in Quantum Linear System Solvers

Introduction and Motivation

The study addresses quantum preconditioning within the Pauli expansion input model for quantum linear system solvers (QLS). Preconditioning is a core classical technique to improve the conditioning of a linear system, thereby accelerating iterative solvers. However, its quantum analog is nontrivial: prior analyses revealed that the classical condition number reduction does not automatically translate to improved query complexity in QLS algorithms, due to normalization overhead when multiplying independent block-encodings of matrices and preconditioners. This work questions whether algebraic structure—specifically Pauli expansions—can circumvent these overheads in quantum access models.

The authors formulate the problem within two oracle models: the block-encoding model, where the system matrix and preconditioner are encoded as unitaries, and the randomized Pauli access model, where matrices are input as Pauli expansions. They prove that regrouping Pauli terms after preconditioning can alter effective normalization and complexity parameters, rigorously characterizing the setting in which preconditioning offers genuine quantum advantage.

Limitations of Separate Block-Encoding Composition

A fundamental result of the paper is a limitation theorem regarding preconditioning via separate block-encoding composition. If access to the system matrix $A$ and preconditioner $P$ is given via separate block-encodings, the normalization factor of the block-encoded product $PA$ is at least the product of the normalization factors, i.e., $\alpha_P \alpha_A$, and the condition number of the preconditioned system cannot be lower than $\alpha_A \kappa_A$. Thus, classical improvement in condition number does not reduce quantum query complexity; normalization overhead eliminates the benefit. This generalizes recent negative results in the literature and provides a unified lower bound for both left and symmetric preconditioning (2606.01733).

Direct Pauli-Structured Block-Encoding

The paper then considers a direct access model enabled by Pauli structure, where the preconditioned operator is formed classically and block-encoded via an LCU circuit or Hamiltonian simulation. If both $A$ and $P$ have Pauli expansions, their product admits a regrouped Pauli expansion, possibly with significant cancellation and reduction in coefficient weight. The normalization parameter for the block-encoding is then determined by the regrouped Pauli coefficient weight $w(Q)$ (where $Q = PA$ or $Q = PAP^\dagger$), not the product of the normalizations.

Explicit bounds are derived: $P$0 and $P$1, but the actual weight can be much lower after regrouping due to cancellations. The quantum cost for Hamiltonian simulation is detailed in terms of commutator scaling, Pauli word counts, and block-encoding precision. The direct model avoids the separate block-encoding obstruction and can offer query complexity reduction, but this must be balanced against the increased complexity of implementing the regrouped Pauli expansion and associated classical preprocessing.

Randomized Pauli Access and Structural Complexity

The study analyzes randomized QLS solvers that estimate functionals of the solution vector using classical Pauli access, as in [wang2024qubit]. Here, the dominant parameters governing circuit depth are the Pauli weight $P$2 and the smallest singular value $P$3. The structural depth proxy is $P$4. Preconditioners in the Pauli model transform this proxy to $P$5, allowing targeted design: a Pauli-structured preconditioner is beneficial if regrouping lowers $P$6 sufficiently without sacrificing stability in $P$7.

A precise trade-off exists; the improvement is certified by comparing

$P$8

as the criterion for per-sample depth reduction. The adjustment also affects sample complexity, but Pauli weight remains the central gate-depth parameter.

Numerical Experiments

Concrete numerical evidence is presented using a synthetic diagonal-term-dominated Hamiltonian family. For $P$9 to $PA$0 qubits, a diagonal Pauli ansatz preconditioner is constructed by least-squares fitting. For each instance, regrouped Pauli weight and singular values are directly computed, and the proxy ratio $PA$1 is evaluated.

Results show substantial proxy reductions, e.g., $PA$2, indicating that the randomized QLS per-sample depth can decrease by more than an order of magnitude. The diagnostic ratios for symmetric direct block-encoding also confirm reduced effective QLS condition parameters with preconditioning throughout the tested range.

Implications and Future Directions

The theoretical and numerical results establish when Pauli-structured preconditioning can benefit quantum algorithms: only in models leveraging direct access to regrouped Pauli expansions. The findings inform design principles for quantum preconditioners in both randomized and block-encoding-based solvers. Practical implications include improved quantum resource efficiency for linear system tasks in quantum chemistry, condensed matter, and discretized PDEs, where Pauli expansion models are natural.

Future work should systematically optimize preconditioner ansatz families for low Pauli coefficient weight, explore scalable constructions retaining cancellation structure, and bridge the gap between proxy-level complexity and hardware-level gate resource cost. Further investigation is warranted into classical preprocessing overheads, polynomial-size ansatz choices, and their impact on both query and end-to-end gate complexity in large problem settings.

Conclusion

The paper rigorously characterizes the limitations and possible advantages of quantum preconditioning in linear system solvers under Pauli expansion access. It clarifies that separate block-encoding composition does not lower quantum query complexity, while regrouped Pauli-structured preconditioners can reduce structural depth proxies in randomized QLS and effective QLS condition parameters via direct block-encoding. Numerical evidence supports these claims within current ansatz families. The results provide foundational guidance for resource-aware quantum linear algebra and suggest new directions in ansatz design, classical preprocessing, and end-to-end quantum algorithm benchmarking.