Tackling instabilities of quantum Krylov subspace methods: an analysis of the numerical and statistical errors
Maria Gabriela Jordão Oliveira, Karl Michael Ziems, Nina Glaser
Abstract
Krylov subspace methods are among the most extensively studied early fault-tolerant quantum algorithms for estimating ground-state energies of quantum systems. However, the rapid onset of ill-conditioning might make accurate energies difficult or even impossible to retrieve. In this communication, we analyse the numerical stability and statistical problems of these methods using numerical simulations both in the presence and absence of sampling noise. While in ideal numerical simulations the generalized eigenvalue problem indeed becomes unstable with increased Krylov subspace size, we find that, in realistic noisy settings, these methods do not primarily suffer from ill-conditioning. Instead, statistical fluctuations dominate and can prevent reliable solution extraction unless appropriate regularization or filtering techniques are employed. We consequently introduce two new metrics, the imaginary and unitary filters, that successfully assess the reliability of the obtained solutions without any knowledge of the true eigenspectrum.
AI Impact Assessments
(3 models)Scientific Impact Assessment
Core Contribution
This paper addresses a widely cited concern in the quantum computing community: the ill-conditioning of the generalized eigenvalue problem (GEVP) in quantum Krylov subspace methods. The central insight is a reframing of the problem — the authors demonstrate that in realistic noisy settings, ill-conditioning is not the primary bottleneck. Instead, statistical fluctuations from finite sampling dominate the error and prevent reliable solution extraction. This distinction, while perhaps intuitive in hindsight, has been somewhat conflated in the prior literature, where condition number κ(S) has been used as the primary diagnostic for algorithm stability.
The paper introduces two filtering metrics — the imaginary filter (for QBKS-H, exploiting the fact that eigenvalues should be real) and the unitary filter (for QBKS-U, exploiting the fact that eigenvalues of a unitary operator should have unit norm) — that assess solution reliability without requiring knowledge of the true eigenspectrum. These are conceptually simple but practically useful diagnostics that exploit algebraic properties that must hold for correctly solved GEVPs.
Methodological Rigor
The numerical analysis is systematic and thorough. The authors examine:
The experimental design with equicost lines (connecting parameter combinations requiring equal quantum resources) is a nice touch that enables fair comparisons across different algorithm configurations.
However, there are methodological limitations. The test systems are small (STO-3G basis, freeze-core approximation), which is standard for quantum algorithm benchmarking but limits the strength of conclusions about scalability. The paper does not provide formal error bounds connecting the proposed metrics to actual energy errors — the filters are presented as heuristic diagnostics, and the authors appropriately acknowledge they indicate reliability rather than accuracy. The assumption of error-free Hamiltonian implementations (no gate noise, no Trotter error) simplifies the analysis but omits important practical error sources.
Potential Impact
The practical impact is moderate but useful for the quantum computing algorithms community. The key findings that:
1. κ(S) is a poor standalone diagnostic for solution quality
2. Shot noise actually reduces condition numbers but introduces its own instabilities
3. Single-reference Krylov methods with proper regularization are more cost-effective than multi-reference approaches
...provide actionable guidance for practitioners implementing quantum Krylov algorithms. The imaginary and unitary filters are trivially implementable (requiring only inspection of already-computed eigenvalues) and add essentially zero computational overhead.
The comparison of regularization strategies (fixed threshold, elbow method, two adaptive literature methods) serves as a useful practical guide, with the finding that the adaptive method from Ref. [16] performs best across the tested scenarios.
Timeliness & Relevance
The paper is timely. Quantum Krylov methods are among the most actively studied algorithms for the early fault-tolerant era, and questions about their practical viability — particularly whether ill-conditioning fundamentally limits their utility — are central to the field's trajectory. Multiple recent works (Refs. [12-17] in the paper) have grappled with these stability questions, making this systematic analysis and the clarification that statistical noise rather than ill-conditioning is the primary practical challenge a relevant contribution to an active debate.
Strengths
1. Clear reframing: The distinction between numerical ill-conditioning and statistical noise as separate failure modes, with the demonstration that the latter dominates in practice, is valuable.
2. Practical diagnostics: The imaginary and unitary filters are simple, zero-cost, reference-free metrics that can be immediately adopted by the community.
3. Comprehensive comparison: The systematic evaluation across multiple regularization strategies, algorithm variants, time steps, and subspace sizes provides a useful reference.
4. Reproducibility: Code is made publicly available, and the simulation methodology is clearly described.
5. Cost-effectiveness analysis: The equicost comparison between single and multiple references provides practical guidance.
Limitations
1. Small system sizes: Both test molecules are very small, and it remains unclear whether findings generalize to larger, more chemically interesting systems where the spectral gap structure and Krylov space properties may differ significantly.
2. No formal guarantees: The filters are heuristic — no rigorous bounds connect the filter metrics to actual energy errors. The relationship between imaginary components/unitarity deviations and energy accuracy is empirical.
3. Idealized noise model: Only shot noise is considered; gate errors, Trotter errors, and decoherence are absent. Real early-FT devices will have residual errors from imperfect error correction.
4. Limited novelty of individual components: SVD regularization and the idea of using eigenvalue properties as sanity checks are not entirely new. The contribution is more in the systematic analysis and specific formulation of the metrics.
5. Incomplete analysis of the filters as standalone tools: The authors acknowledge that in the shot noise regime, the filters alone are insufficient for chemical accuracy and must be combined with regularization, somewhat limiting their independent utility.
6. Reference state selection: The scheme requires knowledge of the true ground state for ordering references by overlap, which is unrealistic in practice.
Overall Assessment
This is a solid, well-executed numerical study that provides useful practical insights and simple diagnostic tools for the quantum Krylov methods community. The main contribution — clarifying that statistical noise rather than ill-conditioning is the dominant practical challenge — is important for setting correct expectations and directing future algorithmic development efforts. However, the novelty is incremental rather than transformative, the test systems are small, and the proposed metrics, while useful, lack formal theoretical backing. The paper reads more as a thorough practical guide than a fundamental methodological advance.
Generated Apr 19, 2026
Comparison History (42)
Paper 1 addresses a fundamental algorithmic challenge (instability in quantum Krylov subspace methods) relevant to the broader field of early fault-tolerant quantum computing. By introducing novel filtering metrics to assess solution reliability, it offers methodological advancements applicable across various quantum systems. In contrast, Paper 2 focuses on benchmarking existing algorithms for a specific surface chemistry application, which, while valuable, presents a narrower scope of impact compared to the foundational algorithmic improvements in Paper 1.
Paper 1 demonstrates an exponential quantum-classical separation for a natural, practically relevant problem (Shannon entropy estimation in data streams), which is a fundamental complexity-theoretic result. This type of separation result is rare and significant, bridging quantum computing theory with practical applications in computer networking. Paper 2 makes valuable practical contributions to quantum Krylov methods by analyzing instabilities and proposing new filtering metrics, but it is more incremental—addressing known issues in an existing algorithmic framework rather than establishing a new fundamental result. The exponential separation in Paper 1 is likely to have broader and longer-lasting impact across quantum complexity theory and quantum algorithms.
Paper 1 likely has higher impact because it addresses a timely, practical bottleneck for early fault-tolerant quantum algorithms: stability and noise in quantum Krylov subspace methods for ground-state estimation. It combines simulations with realistic sampling noise, clarifies that statistical errors—not just ill-conditioning—dominate, and proposes actionable, spectrum-agnostic reliability metrics (imaginary/unitary filters) that can be adopted broadly across Krylov-style quantum algorithms. Paper 2 is elegant and broadly applicable in principle, but state-to-state unitary constructions are a well-trodden area and the real-world methodological/experimental leverage may be less immediate.
Paper 2 likely has higher impact: it proposes a new architectural concept (quantum actuators) that could broaden the design space for scalable quantum processors under global-control constraints, with clear implications for hardware connectivity, compilation, and long-range entangling operations, plus links to quantum thermodynamics (quantum batteries) that may attract cross-field attention. Paper 1 is timely and rigorous for near-term fault-tolerant algorithms, but is more specialized (stability/metrics for Krylov methods) and primarily improves reliability rather than introducing a broadly enabling paradigm.
Paper 1 targets a key bottleneck for early fault-tolerant quantum algorithms: practical reliability of quantum Krylov subspace methods under sampling noise. Its diagnosis (statistical error dominating over ill-conditioning in realistic regimes) and introduction of spectrum-agnostic reliability metrics/filters are broadly actionable for quantum algorithm practitioners and can influence benchmarking, error mitigation, and algorithm design across quantum chemistry and materials. Paper 2 offers a solid, more specialized advance in nonequilibrium open spin-chain dynamics and scaling, with narrower immediate applicability beyond quantum many-body theory.
Paper 2 has higher estimated impact due to broader relevance and applicability: it addresses a pervasive, practical bottleneck (encoding continuous/numeric data) in generative quantum ML, potentially affecting many models and datasets beyond QCBMs. The Gray-code remedy is simple, low-overhead, and readily adoptable, making translation to practice likely. Paper 1 is valuable and rigorous for quantum Krylov methods, but its impact is more specialized to a narrower algorithmic family in quantum chemistry/physics; its filters help diagnostics rather than expanding capability across domains.
Paper 1 combines a novel algorithmic advance (QMC for nonlinear Renyi-2 correlators in 2+1D under strong decoherence) with a coherent effective-field-theory mapping to 2D Ashkin–Teller, yielding a rich mixed-state phase diagram and validated universality predictions. This is methodologically rigorous, broadly relevant to open quantum many-body physics, mixed-state phases, and criticality, and enables concrete progress where standard QMC/sign-problem issues block alternatives. Paper 2 is timely for quantum algorithms but is more incremental (error/conditioning analysis plus practical diagnostics) with narrower immediate impact.
Paper 2 likely has higher impact due to broader relevance and timeliness: stabilizing quantum Krylov (subspace) algorithms directly affects near-term and early fault-tolerant quantum simulation across chemistry and materials. It addresses a widely encountered practical bottleneck (sampling-noise–dominated failure modes), supports claims with noisy/ideal simulations, and contributes actionable diagnostics (imaginary/unitary filters) that can be adopted across platforms and methods. Paper 1 is novel and elegant for photonic TF-GKP control, but its immediate impact is narrower, more specialized, and more experimentally contingent.
Paper 1 makes fundamental theoretical advances in quantum information theory by resolving long-standing questions about additivity of one-way distillable entanglement beyond known special cases. It introduces novel structural conditions (regularized less-noisy, informationally degradable) and a generalized spin-alignment principle with broad implications for quantum capacity and entanglement theory. Paper 2 provides useful practical analysis of Krylov subspace methods with new filtering metrics, but is more incremental—addressing known instability issues with diagnostic tools rather than establishing new theoretical foundations. Paper 1's results have deeper and broader implications across quantum information science.
Paper 2 addresses a critical practical bottleneck (instabilities) in quantum Krylov subspace methods, which are among the most promising early fault-tolerant quantum algorithms. By introducing actionable diagnostic metrics (imaginary and unitary filters) and clarifying that statistical noise—not ill-conditioning—is the primary challenge, it provides immediately useful tools for a broad community working on near-term quantum algorithms. Paper 1, while interesting in proposing a Rydberg-atom simulation of Motzkin spin chains, targets a more niche mathematical physics model with narrower immediate applicability. Paper 2's methodological contributions have broader impact across quantum chemistry and materials science.
Paper 2 introduces a fundamental generalization of the TEMPO method to handle non-commuting coupling operators in open quantum systems, addressing a significant limitation in the field. This broadens applicability to realistic physical scenarios (e.g., multi-site emitters coupled to bosonic lattices) and provides a methodological advance with wide-ranging applications across quantum optics, condensed matter, and quantum information. Paper 1, while valuable in analyzing stability issues of Krylov methods and proposing useful filters, is more incremental—addressing known numerical challenges rather than enabling fundamentally new capabilities.
Paper 2 introduces a novel algorithm (ADAPT-VMPE) with strong theoretical guarantees (polynomial scaling, controllable error bounds) and demonstrates scalability to 100 qubits on a clinically relevant molecule. It bridges quantum-inspired classical algorithms with practical drug discovery applications, offering broader real-world impact. Paper 1 provides valuable diagnostic tools for Krylov methods' stability issues, but is more incremental—analyzing known problems and proposing filtering metrics rather than a fundamentally new algorithmic paradigm. Paper 2's combination of theoretical rigor, scalability demonstration, and biomedical relevance gives it higher potential impact.
Paper 2 likely has higher impact: it targets a central bottleneck in near-term fault-tolerant quantum algorithms (quantum Krylov methods) with careful numerical/noise analysis, clarifying that sampling noise—not ill-conditioning—dominates in practice. It proposes actionable, spectrum-agnostic reliability metrics (imaginary/unitary filters) broadly usable across quantum chemistry and materials simulations, improving methodological rigor and adoption. Paper 1 is ambitious and application-driven, but hybrid Newton+HHL approaches depend on strong assumptions (conditioning, state prep, readout) and uncertain near-term feasibility, which may limit real-world impact.
Paper 2 likely has higher impact: it introduces a broadly applicable tensor-network framework enabling large-deviation analysis of monitored quantum many-body systems, and uses it to identify first-order dynamical phase transitions and characterize conditioned states. This is methodologically substantial, timely for measurement-induced/monitored dynamics, and should influence multiple communities (quantum information, many-body theory, nonequilibrium statistical mechanics, tensor networks). Paper 1 provides valuable diagnostics and mitigation for quantum Krylov instabilities, but is more specialized to a narrower algorithmic niche and primarily refines reliability rather than opening a new general toolbox.
Paper 1 offers a rigorous, broadly applicable scaling law linking neural quantum state expressivity to an information-theoretic quantity (mutual information), with analytical results for stabilizer families and implications across tomography and learning. This combination of theoretical novelty, principled capacity characterization, and cross-architecture insights is likely to influence both quantum many-body theory and machine learning for physics. Paper 2 provides valuable diagnostics and metrics for Krylov-method instabilities under noise, but its impact is more specialized to a specific algorithmic family and primarily methodological rather than foundational.
Paper 1 establishes a fundamentally new theoretical result—the first kinetic uncertainty relation for collective dissipative quantum many-body systems—extending precision bounds from single-body to many-body regimes. It identifies a cooperative enhancement mechanism with particle-number scaling, relevant to quantum technologies and boundary time crystals. Paper 2 provides valuable practical analysis of numerical stability in quantum Krylov methods and introduces useful filtering metrics, but is more incremental in nature, addressing known instability issues rather than establishing new fundamental bounds. Paper 1's broader theoretical significance and implications for quantum technology design give it higher impact potential.
Paper 1 likely has higher near-term scientific impact: it addresses a key practical bottleneck for early fault-tolerant quantum algorithms (Krylov methods), identifies the dominant error source under realistic sampling noise, and proposes actionable, spectrum-agnostic reliability metrics (imaginary/unitary filters) that can directly improve algorithm performance and benchmarking. This is timely and broadly relevant to quantum simulation and algorithm engineering. Paper 2 offers elegant, potentially general results on information retention and dissipation-induced transitions in random circuits, but its applications are more indirect and primarily foundational within quantum dynamics/thermalization.
Paper 2 addresses a critical practical bottleneck in fault-tolerant quantum computing—optimizing lattice surgery compilation with concrete, quantifiable improvements (28-46.7% space reduction, up to order-of-magnitude error rate suppression). This has broad, immediate impact as the field moves toward practical FTQC implementations. Paper 1 provides valuable analysis of Krylov subspace method instabilities and useful diagnostic filters, but is more incremental—addressing known numerical issues in an existing algorithm class. Paper 2's framework-level contribution with demonstrated significant resource savings is more likely to influence the broader quantum computing community.
Paper 1 addresses a fundamental theoretical and practical challenge in quantum Krylov subspace methods—a prominent class of early fault-tolerant quantum algorithms. It provides rigorous analysis of numerical stability and statistical errors, and introduces novel diagnostic metrics (imaginary and unitary filters) applicable broadly across quantum chemistry and physics. Paper 2, while useful, is primarily an engineering contribution—a transpilation tool between two specific quantum programming frameworks—with narrower scope and limited conceptual novelty. Paper 1's insights into algorithm reliability have broader and more lasting scientific impact.
Paper 1 addresses a fundamental and widely recognized challenge (ill-conditioning) in Krylov subspace methods, which are among the most promising early fault-tolerant quantum algorithms. It provides rigorous analysis distinguishing numerical from statistical errors and introduces practical diagnostic metrics (imaginary and unitary filters). This has broad impact across quantum chemistry, materials science, and quantum computing. Paper 2, while interesting, addresses a more niche topic (variational MBQC for generative modeling) with a narrower scope and smaller community of researchers working on this specific intersection.