pith. sign in

arxiv: 2511.05058 · v2 · pith:TRRMQVYQnew · submitted 2025-11-07 · ✦ hep-lat

Unbiased Krylov subspace method for the extraction of ground state from lattice correlators

Ryutaro Tsuji , Shoji Hashimoto , Ryan Kellermann This is my paper

Pith reviewed 2026-05-21 20:04 UTC · model grok-4.3

classification ✦ hep-lat
keywords lattice QCDKrylov subspaceground statesingular value decompositionstatistical noisecorrelatorextrapolationmeson
0
0 comments X

The pith

Low-rank SVD approximation followed by extrapolation to zero eigenvalue variance removes bias from noisy lattice correlators in ground-state extraction.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a way to pull ground-state energies and matrix elements out of lattice QCD correlators even when statistical noise corrupts the usual Krylov-subspace diagonalization of the transfer matrix. A low-rank SVD is first applied to the matrix of measured correlators to stabilize the procedure. The remaining bias is then removed by extrapolating the extracted energies to the limit where their statistical variance vanishes. Tests on both artificial mock data and real K and Ds meson correlators show that the combined steps recover the correct ground-state values.

Core claim

Ground-state energy and matrix element are reconstructed from correlators in lattice QCD by diagonalizing transfer matrix T within the Krylov subspace spanned by T^n |chi>, where |chi> is a state generated by an interpolating field on the lattice. In numerical applications this strategy is spoiled by statistical noise. A low-rank approximation based on singular-value decomposition of a matrix made of the correlators is introduced, and the associated bias is eliminated by an extrapolation to the limit of vanishing variance of energy eigenvalue.

What carries the argument

low-rank singular-value decomposition of the correlator matrix, followed by extrapolation of extracted energies to the zero-variance limit

If this is right

  • Ground-state energies extracted from noisy lattice data become free of the systematic shift that normally appears in finite-variance Krylov analyses.
  • Matrix elements between the ground state and the interpolating field can be obtained alongside the energies without additional bias.
  • The method works for both pseudoscalar meson correlators such as those of the K and Ds mesons.
  • Only a single extra extrapolation step is required after the SVD truncation, keeping the computational overhead modest.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same SVD-plus-extrapolation pattern could be applied to other noisy observables that rely on subspace projections in lattice field theory.
  • If the variance-to-bias relation remains linear at higher statistics, the extrapolation slope itself may become a diagnostic for residual excited-state contamination.
  • Combining the method with improved interpolating operators that already suppress excited states would reduce the extrapolation distance and tighten the final uncertainty.

Load-bearing premise

The low-rank SVD approximation preserves the ground-state signal sufficiently that the subsequent extrapolation to zero variance recovers the unbiased value without residual systematic distortion from the rank truncation.

What would settle it

Apply the full procedure to a set of high-statistics mock correlators whose exact ground-state energy is known in advance; if the extrapolated result deviates from the known value by more than the final statistical error, the unbiasedness claim fails.

Figures

Figures reproduced from arXiv: 2511.05058 by Ryan Kellermann, Ryutaro Tsuji, Shoji Hashimoto.

Figure 1
Figure 1. Figure 1: FIG. 1. Normalized singular values [PITH_FULL_IMAGE:figures/full_fig_p013_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Estimated ground-state mass [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Largest eigenvalue [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Ground-state matrix element [PITH_FULL_IMAGE:figures/full_fig_p016_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. Same as Fig. 1 but with noisy data sets. Each symbol represents a result for a given [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. Largest eigenvalue [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7. Estimate of the largeset eigenvalue obtained by an extrapolation [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8. Histogram of [PITH_FULL_IMAGE:figures/full_fig_p019_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9. Quantile-Quantile plot for the distribution of the ground-state eigenvalues obtained with [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10. Distribution of the ground-state eigenvalues and a comparison of three approaches to [PITH_FULL_IMAGE:figures/full_fig_p021_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: FIG. 11. Ground-state matrix element [PITH_FULL_IMAGE:figures/full_fig_p022_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: FIG. 12. Ground-state matrix element [PITH_FULL_IMAGE:figures/full_fig_p023_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: FIG. 13. Normalized singular values of the matrix [PITH_FULL_IMAGE:figures/full_fig_p025_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: FIG. 14. Eigenvalue [PITH_FULL_IMAGE:figures/full_fig_p026_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: FIG. 15. Estimate of [PITH_FULL_IMAGE:figures/full_fig_p027_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: FIG. 16. Same as Fig. 15 for [PITH_FULL_IMAGE:figures/full_fig_p027_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: shows the matrix element ⟨Ds|V4|Ds⟩ with the ground state reconstructed from correlators themselves. The plot is obtained with r = 0, 1 and 2 for m = 2, · · · , 6. We find a rather strong dependence on r, and an extrapolation to ∆λ (m) 0 → 0 is important. In [PITH_FULL_IMAGE:figures/full_fig_p028_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: FIG. 18. Matrix element [PITH_FULL_IMAGE:figures/full_fig_p029_18.png] view at source ↗
read the original abstract

Ground-state energy and matrix element are reconstructed from correlators in lattice QCD by diagonalizing transfer matrix $\hat{T}$ within the Krylov subspace spanned by $\hat{T}^n|\chi\rangle$, where $|\chi\rangle$ is a state generated by an interpolating field on the lattice. In numerical applications, this strategy is spoiled by statistical noise. To circumvent the problem, we introduce a low-rank approximation based on a singular-value decomposition of a matrix made of the correlators. The associated bias is eliminated by an extrapolation to the limit of vanishing variance of energy eigenvalue. The strategy is tested using a set of mock data as well as real data of $K$ and $D_s$ meson correlators.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 3 minor

Summary. The manuscript proposes an unbiased Krylov subspace method for extracting ground-state energies and matrix elements from lattice QCD correlators. It constructs the Krylov basis from powers of the transfer matrix applied to an interpolating field, applies a low-rank SVD approximation to the correlator matrix to suppress statistical noise, and removes the resulting truncation bias via extrapolation of the extracted energy eigenvalue to the limit of vanishing variance. The procedure is validated on mock data (where the true ground state is known) and applied to real K and D_s meson correlators.

Significance. If the central claim holds, the approach provides a practical route to noise-robust ground-state extraction that corrects for SVD truncation bias through extrapolation rather than ad-hoc regularization. The mock-data tests directly confront the method with a known truth value, which is a clear strength. The method is relatively parameter-light (SVD rank and extrapolation form), reducing the risk of hidden tuning. This could be useful for precision spectroscopy in channels where statistical noise currently limits the reach of standard effective-mass or GEVP analyses.

major comments (1)
  1. [Validation / mock-data section] The validation on mock data is described only qualitatively in the abstract and summary sections. Quantitative results—such as the difference between the extrapolated energy and the known input value, its dependence on SVD rank, and the stability of the extrapolation functional form—are not reported with error bars or tables. Because the central claim rests on the extrapolation recovering the unbiased ground state without residual distortion from rank truncation, this omission is load-bearing for assessing whether the method succeeds.
minor comments (3)
  1. [Method description] Clarify the precise definition of the variance used in the extrapolation (e.g., whether it is the variance of the eigenvalue estimator or a proxy derived from the SVD singular values).
  2. [Extrapolation procedure] Specify the functional form(s) adopted for the extrapolation to zero variance and justify the choice (linear, quadratic, etc.).
  3. [Numerical implementation] Add a brief discussion of how the method scales with the number of time slices or the condition number of the correlator matrix.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their positive evaluation of the manuscript and for the constructive comment on the mock-data validation. We address the point below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Validation / mock-data section] The validation on mock data is described only qualitatively in the abstract and summary sections. Quantitative results—such as the difference between the extrapolated energy and the known input value, its dependence on SVD rank, and the stability of the extrapolation functional form—are not reported with error bars or tables. Because the central claim rests on the extrapolation recovering the unbiased ground state without residual distortion from rank truncation, this omission is load-bearing for assessing whether the method succeeds.

    Authors: We agree that a more quantitative presentation of the mock-data tests is needed to substantiate the central claim. In the revised version we will add a dedicated table (or figure panel) that reports, for several SVD ranks, the difference between the extrapolated energy and the known input value together with its statistical uncertainty. We will also include a brief discussion of the stability of the chosen extrapolation form and show the dependence of the extrapolated result on the SVD cutoff. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper constructs the ground-state extraction via a low-rank SVD approximation applied to the correlator matrix within the Krylov subspace, followed by extrapolation of the extracted energy eigenvalue to the limit of vanishing variance in order to remove truncation bias. This chain is tested directly against mock data sets in which the true ground-state energy is known by construction, as well as against real K and D_s meson correlators; the mock-data validation supplies an external benchmark that is independent of the paper's own fitted or extrapolated quantities. No step reduces by the paper's equations to a self-definition, a renamed fit presented as a prediction, or a load-bearing self-citation; the central claim therefore remains self-contained against external numerical checks rather than internally forced.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The approach rests on standard lattice QCD transfer-matrix formalism and the assumption that statistical noise can be modeled through variance extrapolation after SVD truncation.

free parameters (2)
  • SVD truncation rank
    Chosen to balance noise suppression and signal retention; value not specified in abstract.
  • extrapolation functional form
    Form of the fit to vanishing variance is not detailed in abstract.
axioms (1)
  • domain assumption Correlators are generated by the transfer matrix acting on interpolating fields
    Standard assumption in lattice QCD stated in the abstract.

pith-pipeline@v0.9.0 · 5649 in / 1084 out tokens · 31311 ms · 2026-05-21T20:04:34.297218+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

24 extracted references · 24 canonical work pages · 2 internal anchors

  1. [1]

    Parisi, Phys

    G. Parisi, Phys. Rept.103, 203 (1984)

    work page 1984

  2. [2]

    G. P. Lepage, in Theoretical Advanced Study Institute in Elementary Particle Physics (1989)

    work page 1989

  3. [3]

    Aoki, K.-I

    Y. Aoki, K.-I. Ishikawa, Y. Kuramashi, S. Sasaki, K. Sato, E. Shintani, R. Tsuji, H. Watanabe, and T. Yamazaki (PACS), Phys. Rev. D112, 074510 (2025), 2505.06854

    work page arXiv 2025

  4. [4]

    Gambino and S

    P. Gambino and S. Hashimoto, Phys. Rev. Lett.125, 032001 (2020), 2005.13730

    work page arXiv 2020

  5. [5]

    Michael, Nucl

    C. Michael, Nucl. Phys. B259, 58 (1985)

    work page 1985

  6. [6]

    Luscher and U

    M. Luscher and U. Wolff, Nucl. Phys. B339, 222 (1990)

    work page 1990

  7. [7]

    Blum et al

    T. Blum et al. (RBC, UKQCD), Phys. Rev. D107, 094512 (2023), [Erratum: Phys.Rev.D 108, 039902 (2023)], 2301.09286

    work page arXiv 2023

  8. [8]

    M. L. Wagman, Phys. Rev. Lett.134, 241901 (2025), 2406.20009

    work page arXiv 2025

  9. [9]

    D. C. Hackett and M. L. Wagman (2024), 2407.21777

    work page arXiv 2024

  10. [10]

    D. C. Hackett and M. L. Wagman, Phys. Rev. D112, 014514 (2025), 2412.04444

    work page arXiv 2025

  11. [11]

    Chakraborty, D

    D. Chakraborty, D. Sood, A. Radhakrishnan, and N. Mathur (2024), 2412.01900

    work page arXiv 2024

  12. [12]

    Ostmeyer, A

    J. Ostmeyer, A. Sen, and C. Urbach, Eur. Phys. J. A61, 26 (2025), 2411.14981

    work page arXiv 2025

  13. [13]

    Abbott, D

    R. Abbott, D. C. Hackett, G. T. Fleming, D. A. Pefkou, and M. L. Wagman (2025), 2503.17357

    work page arXiv 2025

  14. [14]

    Ostmeyer and C

    J. Ostmeyer and C. Urbach (2025), 2510.15500

    work page arXiv 2025

  15. [15]

    Cullum and R

    J. Cullum and R. A. Willoughby, Journal of Computational Physics44, 329 (1981), ISSN 0021- 9991, URLhttps://www.sciencedirect.com/science/article/pii/0021999181900565

    work page arXiv 1981

  16. [16]

    Cullum and R

    J. Cullum and R. Willoughby, Classics in Applied MathematicsI(2002)

    work page 2002

  17. [17]

    Imada and T

    M. Imada and T. Kashima, Journal of the Physical Society of Japan69, 2723 (2000), URL https://doi.org/10.1143/JPSJ.69.2723

    work page doi:10.1143/jpsj.69.2723 2000

  18. [18]

    Kashima and M

    T. Kashima and M. Imada, Journal of the Physical Society of Japan70, 2287 (2001), URL https://doi.org/10.1143/JPSJ.70.2287. 32

    work page doi:10.1143/jpsj.70.2287 2001

  19. [19]

    F. L. Bauer and C. T. Fike, Numerische Mathematik2, 137 (1960), URLhttps://api. semanticscholar.org/CorpusID:121278235

    work page 1960

  20. [20]

    Sorella, Phys

    S. Sorella, Phys. Rev. B64, 024512 (2001), URLhttps://link.aps.org/doi/10.1103/ PhysRevB.64.024512

    work page 2001

  21. [21]

    Colquhoun, S

    B. Colquhoun, S. Hashimoto, T. Kaneko, and J. Koponen (JLQCD), Phys. Rev. D106, 054502 (2022), 2203.04938

    work page arXiv 2022

  22. [22]

    Renormalization of domain-wall bilinear operators with short-distance current correlators

    M. Tomii, G. Cossu, B. Fahy, H. Fukaya, S. Hashimoto, T. Kaneko, and J. Noaki (JLQCD), Phys. Rev. D94, 054504 (2016), 1604.08702

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  23. [23]

    The nucleon axial charge from lattice QCD with controlled errors

    S. Capitani, M. Della Morte, G. von Hippel, B. Jager, A. Juttner, B. Knippschild, H. B. Meyer, and H. Wittig, Phys. Rev. D86, 074502 (2012), 1205.0180

    work page internal anchor Pith review Pith/arXiv arXiv 2012

  24. [24]

    Hashimoto, PTEP2017, 053B03 (2017), 1703.01881

    S. Hashimoto, PTEP2017, 053B03 (2017), 1703.01881. 33

    work page arXiv 2017