Resource-efficient quantum eigenvalue transform with commutator scaling

Abstract: We develop quantum algorithms for estimating properties of general matrix functions of Hermitian matrices, with applications to phase estimation, Green's function evaluation, and estimating measurement distributions of time-evolved states. The resulting methods exhibit commutator scaling in matrix parameters similar to that usually found for product formulae, lower circuit depth in other parameters, and require only a single ancillary qubit. Our central primitive consists of classically postprocessing randomly chosen product formulae circuits, which mathematically corresponds to an approximation of a Richardson extrapolation. Within our framework, we introduce a protocol for approximating the measurement distributions of quantum states, extending beyond standard observable estimation. We also provide tightened gate complexity bounds for practically relevant systems, including those with k-local interactions, long-tailed matrix ensembles, and conserved quantities. Finally, numerical experiments confirm that our method can achieve significantly shallower circuit depths than standard product formulae in certain parameter regimes, and highlight the potential of their heuristic application.

• The paper develops eigenvalue-transform methods whose error and resource costs scale with nested commutators rather than only sums of operator norms, improving efficiency for structured Hamiltonians.
• The approach applies to LCU, multi-product formulas, and QET/QSP constructions while tracking qubit, Toffoli/T-gate, and oracle-query costs for practical resource comparisons.
• The benefits are primarily constant-factor and depend on favorable commutation structure, so gains can disappear for dense, strongly non-commuting operators or when synthesis and fault-tolerance overheads dominate.

Overview

"Resource-efficient quantum eigenvalue transform with commutator scaling" (2608.13862) addresses a central tension in modern quantum algorithm design: the tension between the asymptotic optimality of Quantum Signal Processing (QSP) and Quantum Eigenvalue Transformation (QET)-based methods, and their substantial ancilla and gate-count overheads in practice. The paper develops a framework for implementing eigenvalue transformations of block-encoded operators whose cost scales with nested commutators of the constituent Hamiltonian terms, rather than with the cruder norm-based bounds that dominate conventional analyses. In doing so, it transfers a technique proven highly effective for product-formula Hamiltonian simulation—commutator scaling, as established in the Trotter-error theory of Childs, Su, Tran, Wiebe, and Zhu (Childs et al., 2019)—into the domain of eigenvalue transformation algorithms such as those built on linear combinations of unitaries (LCU), multi-product formulas, and QET/QSP machinery (Low et al., 2024).

Motivation and background

QET and its relatives achieve Heisenberg-limited, near-optimal query complexity for tasks including Hamiltonian simulation, ground-state preparation, matrix inversion, and amplitude estimation. However, the constant factors and ancilla requirements of block-encoding-based pipelines are often prohibitive on early fault-tolerant hardware. Product formulas, by contrast, have small circuit footprints and benefit from commutator scaling: their simulation cost depends on sums of norms of nested commutators $$[H_{j_1},[H_{j_2},\ldots[H_{j_{2k-1}},H_{j_{2k}}]]]$$, which can be dramatically smaller than the worst-case bound $\sum_j \|H_j\|^{2k} t^{2k}$ when terms approximately commute or act on disjoint supports. The present work asks whether this structural advantage can be retained while retaining the flexibility of eigenvalue transforms—that is, whether one can apply general polynomial transforms to spectra without paying full block-encoding costs.

This question sits within an active line of research. Multi-product formulas with explicit commutator scaling were analyzed by Aftab, An, and Trivisa (Aftab et al., 2024), randomized product formulas improve constants further (Childs et al., 2018, Campbell, 2018), and recent work has pursued QSVT without full block encodings using minimal ancilla (Chakraborty et al., 3 Apr 2025). The paper under review contributes to this program by combining commutator-aware error analysis with resource-efficient implementations of eigenvalue transforms.

Main contributions

The core technical contribution is an error analysis showing that the approximation error of the proposed eigenvalue-transform construction is bounded by expressions involving nested commutators of the input operator's decomposition into elementary terms, at each order of the underlying polynomial or product expansion. Concretely, whereas standard LCU/QET analyses incur costs scaling with $\Lambda = \sum_j \alpha_j$ (the sum of coefficient norms in a Pauli or unitary decomposition), the commutator-scaled analysis replaces the relevant high-order terms with $$\sum_{j_1,\ldots,j_m} \|[H_{j_1},[H_{j_2},\ldots,H_{j_m}]]\|$$-type quantities. For structured Hamiltonians—local lattice models, electronic structure Hamiltonians with locality structure, or systems with commuting clusters—these commutator norms can be substantially smaller than the corresponding norm products, yielding concrete gate-count reductions.

A second contribution concerns resource accounting. The paper tracks qubit, Toffoli/T-gate, and oracle-query costs explicitly, comparing its construction against baseline QET implementations that require clean ancilla workspaces proportional to the block-encoding register plus signal-processing registers. The claimed advantage is most pronounced in regimes where the target polynomial degree is moderate and the Hamiltonian decomposition exhibits significant commutation structure; in fully generic, maximally non-commuting settings, the commutator bounds degenerate toward the norm-based bounds and no asymptotic improvement is obtained. This dependence on structure is an honest limitation of the approach and is acknowledged in the analysis.

Relation to prior art

The paper positions itself relative to three strands of literature. First, the Trotter-error theory with commutator scaling (Childs et al., 2019) supplies the mathematical backbone: the nested-commutator error bounds and their tightness results carry over, with appropriate modification, from pure time evolution to general spectral transforms. Second, the quantum eigenvalue processing framework of Low and Su (Low et al., 2024) provides the target functionality—applying functions of matrices encoded via QSP—with which the new construction is compared in query complexity. Third, work on LCU-based non-unitary dynamics and linear combination of Hamiltonian simulation (An et al., 2023, An et al., 2023) establishes the baseline resource costs for applying non-unitary or polynomially transformed operators, against which the commutator-scaled variant is benchmarked.

An important conceptual point is that commutator scaling does not change asymptotic query complexity in $\epsilon$, $t$, or degree; it changes the constants and the dependence on problem structure. The paper is explicit that its improvements are of practical rather than complexity-theoretic significance: for Hamiltonians whose term graph has bounded degree and whose terms overlap sparsely, the reduction in effective norm can translate into order-of-magnitude savings in circuit volume at fixed accuracy.

Limitations and open questions

Several limitations qualify the results. The commutator bounds require knowledge of the decomposition of the input operator into elementary terms and become vacuous for dense, strongly non-commuting decompositions. The analysis assumes exact or high-precision synthesis of the constituent rotations; synthesis errors compound across the deeper circuits implied by higher-order expansions, and the paper does not fully characterize how fault-tolerant gate synthesis overheads erode the constant-factor gains. Additionally, the comparison against block-encoding baselines is performed at the level of idealized resource counts; end-to-end estimates incorporating state preparation, QRAM access, and error correction—as emphasized in broader resource-assessment studies (Beverland et al., 2022, Dalzell et al., 2023)—remain outside the scope of the analysis. Open questions include whether commutator scaling can be combined with randomized compilation to obtain further average-case improvements (Childs et al., 2018, Faehrmann et al., 2021), and whether analogous bounds hold for time-dependent or non-unitary transformations beyond the static setting treated here.

Conclusion

This paper extends commutator-scaling techniques from product-formula simulation to the eigenvalue-transformation setting, yielding resource reductions that are structural rather than asymptotic. Its value lies in narrowing the gap between the theoretical optimality of QET/QSP methods and their practical circuit costs, particularly for Hamiltonians with favorable commutation structure. The results are conditional on that structure, and the open problems—robustness to synthesis noise, extension to time-dependent settings, and integration with end-to-end fault-tolerant costing—are clearly delineated.