pith. sign in

arxiv: 2605.15674 · v1 · pith:AGIGYW3Unew · submitted 2026-05-15 · ✦ hep-lat

Spectral reconstruction from Euclidean lattice correlators through singular value decomposition

Ryutaro Tsuji , Shoji Hashimoto This is my paper

Pith reviewed 2026-05-19 19:13 UTC · model grok-4.3

classification ✦ hep-lat
keywords spectral reconstructionsingular value decompositionEuclidean correlatorslattice QCDinverse Laplace transformsmeared spectral functionsMellin transform
0
0 comments X

The pith

Truncating the singular value decomposition of the kernel function reconstructs smeared spectral densities from Euclidean lattice correlators with controlled uncertainties.

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

The paper establishes that inverting the Laplace transform to obtain spectral densities from imaginary-time lattice correlators is ill-conditioned due to statistical noise. The authors apply singular value decomposition to the kernel exp(-ωt) on discrete t and ω grids, identifying orthogonal bases where the largest singular values correspond to statistically reliable data components. Retaining only those components produces smeared spectral functions whose uncertainties remain under control. The truncation introduces a systematic error that can be bounded when the underlying spectral function satisfies reasonable positivity or smoothness assumptions. In the infinite continuous limit the SVD basis approaches the Mellin transform, yielding a representation independent of specific lattice spacings and volumes.

Core claim

By retaining only the components associated with the largest singular values, for which the correlator data remain statistically significant, one can reconstruct smeared spectral functions with controlled uncertainties. The systematic error arising from the truncation can also be bounded under reasonable assumptions. In the limit where the ranges of t and ω become infinitely large and continuous, the SVD basis approaches the Mellin transform, allowing a representation of the smeared spectrum that is independent of the details of the lattice parameters.

What carries the argument

Singular value decomposition of the kernel function exp(-ω t) on discrete imaginary-time t and energy ω sets, which supplies orthogonal bases in both spaces and ranks their contributions by singular-value magnitude.

If this is right

  • Smeared spectra can be extracted without performing a full ill-conditioned inversion.
  • Statistical significance of each SVD mode directly limits the number of retained terms.
  • The method supplies an explicit bound on truncation systematics once positivity or smoothness of the spectrum is assumed.
  • In the continuum limit the reconstruction becomes independent of lattice spacing and volume through the Mellin-transform connection.

Where Pith is reading between the lines

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

  • The same SVD truncation strategy could be tested on other ill-posed inverse problems that share an exponential kernel, such as certain imaging or relaxation-spectrum inversions.
  • Combining the SVD basis with positivity constraints might further reduce the required number of retained modes while preserving the error bound.
  • The Mellin-transform limit suggests that the method could be formulated directly in momentum space for continuum theories, bypassing lattice discretization details.

Load-bearing premise

The systematic error from SVD truncation can be bounded under reasonable assumptions about the form or positivity properties of the underlying spectral function.

What would settle it

A numerical test in which a known positive spectral function is used to generate exact correlators, noise is added at realistic levels, and the SVD-truncated reconstruction deviates from the true smeared spectrum by more than the claimed bounded systematic error.

Figures

Figures reproduced from arXiv: 2605.15674 by Ryutaro Tsuji, Shoji Hashimoto.

Figure 1
Figure 1. Figure 1: FIG. 1. The singular values [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Rescaled basis functions of [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Left [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. Left [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. Left [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7. Left [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8. SVD components [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9 [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10. A comparison between the true [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: FIG. 11 [PITH_FULL_IMAGE:figures/full_fig_p017_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: FIG. 12 [PITH_FULL_IMAGE:figures/full_fig_p018_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: FIG. 13. Exponent [PITH_FULL_IMAGE:figures/full_fig_p019_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: FIG. 14 [PITH_FULL_IMAGE:figures/full_fig_p020_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: FIG. 15 [PITH_FULL_IMAGE:figures/full_fig_p021_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: FIG. 16. Singular values [PITH_FULL_IMAGE:figures/full_fig_p024_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: FIG. 17 [PITH_FULL_IMAGE:figures/full_fig_p025_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: FIG. 18. Same as Fig. 17 but for [PITH_FULL_IMAGE:figures/full_fig_p026_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: FIG. 19. Same as Fig. 17 but for [PITH_FULL_IMAGE:figures/full_fig_p027_19.png] view at source ↗
read the original abstract

Reconstructing spectral densities from Euclidean lattice correlators requires an inverse Laplace transform, which is inherently ill-conditioned when applied to numerical data with statistical uncertainties. The maximum amount of information that can be extracted from the imaginary-time dependence of correlators can be characterized by the singular value decomposition (SVD) of the kernel function $\exp(-\omega t)$ defined on discrete sets of imaginary times $t$ and energies $\omega$. The SVD provides orthogonal basis functions in both the $t$- and $\omega$-spaces, while the singular values determine the magnitude of their contributions to the correlators. By retaining only the components associated with the largest singular values, for which the correlator data remain statistically significant, one can reconstruct smeared spectral functions with controlled uncertainties. The systematic error arising from the truncation can also be bounded under reasonable assumptions. In the limit where the ranges of $t$ and $\omega$ become infinitely large and continuous, the SVD basis approaches the Mellin transform, allowing a representation of the smeared spectrum that is independent of the details of the lattice parameters.

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

2 major / 2 minor

Summary. The manuscript proposes reconstructing smeared spectral functions from Euclidean lattice correlators via singular value decomposition (SVD) of the kernel exp(−ωt) on discrete t and ω grids. By retaining only the largest singular-value components for which the correlator data remain statistically significant, the method claims to yield smeared spectra with controlled uncertainties; the truncation systematic error is asserted to be boundable under reasonable assumptions on the spectral function. In the infinite-range continuous limit the SVD basis approaches the Mellin transform, furnishing a representation independent of lattice details.

Significance. If the truncation-error bound can be placed on a rigorous, assumption-light footing and validated numerically on realistic QCD spectra, the approach would supply a linear-algebraic, largely parameter-free route to controlled spectral information from noisy Euclidean data, complementing existing inversion techniques. The explicit link to the Mellin transform in the continuum limit is a conceptual strength that could aid analytic understanding.

major comments (2)
  1. [Abstract] Abstract: the statement that 'the systematic error arising from the truncation can also be bounded under reasonable assumptions' is load-bearing for the central claim of 'controlled uncertainties,' yet the manuscript supplies neither an explicit error formula nor numerical tests against known spectra (resonances plus continuum) that would demonstrate the bound holds for typical lattice correlators.
  2. [Method / truncation analysis] The SVD truncation procedure retains a free parameter (the singular-value cutoff threshold). The manuscript does not show how this cutoff is chosen in a manner that keeps the omitted contributions small in the relevant norm without additional assumptions on ρ(ω) whose validity for general lattice data is not demonstrated.
minor comments (2)
  1. [Notation] Clarify the precise definition of the smeared spectral function that is reconstructed and how the retained SVD components map onto it.
  2. [Results] Provide at least one concrete numerical example (even a toy model) showing the reconstructed spectrum, the retained singular values, and a quantitative estimate of the truncation error.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading of our manuscript and the constructive comments. We address each major point below and have revised the manuscript to incorporate additional derivations, explanations, and numerical tests where appropriate.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the statement that 'the systematic error arising from the truncation can also be bounded under reasonable assumptions' is load-bearing for the central claim of 'controlled uncertainties,' yet the manuscript supplies neither an explicit error formula nor numerical tests against known spectra (resonances plus continuum) that would demonstrate the bound holds for typical lattice correlators.

    Authors: We agree that the truncation error bound is central to the claim of controlled uncertainties and that the original manuscript would be strengthened by an explicit formula and validation. In the revised version we derive the bound explicitly: the truncation error in the L2 norm is at most the sum of the neglected singular values multiplied by an a priori bound on ||ρ||_2, under the standard assumptions that ρ(ω) ≥ 0 and is normalized. We have added this derivation in a new subsection and included numerical tests on synthetic correlators generated from Breit-Wigner resonances plus a continuum threshold, demonstrating that the observed reconstruction error remains within the derived bound at noise levels representative of lattice data. These additions appear in the updated Sections 3.2 and 4. revision: yes

  2. Referee: [Method / truncation analysis] The SVD truncation procedure retains a free parameter (the singular-value cutoff threshold). The manuscript does not show how this cutoff is chosen in a manner that keeps the omitted contributions small in the relevant norm without additional assumptions on ρ(ω) whose validity for general lattice data is not demonstrated.

    Authors: The cutoff is chosen by retaining only those singular values for which the corresponding projection of the correlator exceeds the statistical uncertainty by a factor of three or more; this ensures that omitted terms lie below the noise floor. In the revision we have added a quantitative analysis showing that, when ||ρ||_2 is bounded (a mild assumption satisfied by normalized spectral functions), the contribution of the tail is controlled by the decay of the singular-value spectrum. We discuss the validity of this assumption for QCD spectra and note that it is comparable to assumptions used in other reconstruction methods; we also describe a posteriori consistency checks. While the procedure is not entirely assumption-free, the added material clarifies the rationale and the regime of applicability. revision: partial

Circularity Check

0 steps flagged

SVD truncation is a direct linear-algebra regularization with no self-referential reduction

full rationale

The derivation applies the singular value decomposition to the kernel matrix K(t,ω) = exp(−ω t) on the discrete lattice grids, retains the dominant singular components where the correlator data exceed statistical noise, and reconstructs a smeared spectral function. This procedure is a standard, assumption-light regularization of the ill-posed inverse Laplace transform; the retained coefficients are fixed by the data and the kernel itself rather than by any fitted parameter or redefinition of the target spectrum. The statement that truncation error “can also be bounded under reasonable assumptions” invokes external positivity or smoothness properties of ρ(ω) and does not close a loop back onto the SVD coefficients or the lattice correlators. The continuous Mellin-transform limit is a separate mathematical observation about the infinite-range case and does not alter the finite-data construction. No equation or claim reduces to its own input by construction, and no load-bearing step relies on a self-citation whose content is unverified outside the present work.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central procedure rests on the ability to bound truncation error once singular values are dropped; this bound requires an assumption about the spectral function that is not derived inside the paper.

free parameters (1)
  • singular-value cutoff threshold
    Chosen according to where correlator data remain statistically significant; the precise numerical value is data-dependent.
axioms (1)
  • domain assumption Reasonable assumptions on the spectral function allow the truncation systematic error to be bounded.
    Invoked to guarantee controlled uncertainties after discarding small singular values.

pith-pipeline@v0.9.0 · 5709 in / 1228 out tokens · 43663 ms · 2026-05-19T19:13:58.368333+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages · 7 internal anchors

  1. [1]

    Bertero, C

    M. Bertero, C. D. Mol, and E. R. Pike, Inverse Problems1, 301 (1985), URLhttps://doi. org/10.1088/0266-5611/1/4/004

    work page doi:10.1088/0266-5611/1/4/004 1985

  2. [2]

    Bertero, C

    M. Bertero, C. D. Mol, and E. R. Pike, Inverse Problems4, 573 (1988), URLhttps://doi. org/10.1088/0266-5611/4/3/004

    work page doi:10.1088/0266-5611/4/3/004 1988

  3. [3]

    Hadronic Spectral Functions in Lattice QCD

    Y. Nakahara, M. Asakawa, and T. Hatsuda, Phys. Rev. D60, 091503 (1999), hep-lat/9905034

    work page internal anchor Pith review Pith/arXiv arXiv 1999

  4. [4]

    Asakawa, T

    M. Asakawa, T. Hatsuda, and Y. Nakahara, Prog. Part. Nucl. Phys.46, 459 (2001), hep- lat/0011040

    work page arXiv 2001

  5. [5]

    Otsuki, M

    J. Otsuki, M. Ohzeki, H. Shinaoka, and K. Yoshimi, Journal of the Physical Society of Japan 24 10 1 100 101 102 t 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4U0(t) tmax = max = 50.1 tmax = max = 75.1 tmax = max = 100.1 10 1 100 101 102 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4V0( ) tmax = max = 50.1 tmax = max = 75.1 tmax = max = 100.1 10 1 100 101 102 t 2.0 1.5 1.0 0.5 0.0 U1(...

    work page arXiv 2020

  6. [6]

    Itou and Y

    E. Itou and Y. Nagai, JHEP07, 007 (2020), 2004.02426

    work page arXiv 2020

  7. [7]

    A novel Bayesian approach to spectral function reconstruction

    Y. Burnier and A. Rothkopf, Phys. Rev. Lett.111, 182003 (2013), 1307.6106

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  8. [8]

    Del Debbio, A

    L. Del Debbio, A. Lupo, M. Panero, and N. Tantalo, Eur. Phys. J. C85, 185 (2025), 25 10 1 100 101 102 t 1.5 1.0 0.5 0.0 0.5 1.0 U8(t) tmax = max = 50.1 tmax = max = 75.1 tmax = max = 100.1 10 1 100 101 102 1.5 1.0 0.5 0.0 0.5 1.0 V8( ) tmax = max = 50.1 tmax = max = 75.1 tmax = max = 100.1 10 1 100 101 102 t 1.0 0.5 0.0 0.5 1.0 1.5 U9(t) tmax = max = 50.1...

    work page arXiv 2025

  9. [9]

    Bergamaschi, W

    T. Bergamaschi, W. I. Jay, and P. R. Oare, Phys. Rev. D108, 074516 (2023), 2305.16190

    work page arXiv 2023

  10. [10]

    Nevanlinna-Pick interpolation from uncertain data

    S. Fields and N. Christ, Phys. Rev. D113, 054502 (2026), 2510.12136

    work page internal anchor Pith review arXiv 2026

  11. [11]

    Kades, J

    L. Kades, J. M. Pawlowski, A. Rothkopf, M. Scherzer, J. M. Urban, S. J. Wetzel, N. Wink, and F. P. G. Ziegler, Phys. Rev. D102, 096001 (2020), 1905.04305. 26 10 1 100 101 102 t 1.0 0.5 0.0 0.5 1.0 U19(t) tmax = max = 50.1 tmax = max = 75.1 tmax = max = 100.1 10 1 100 101 102 1.0 0.5 0.0 0.5 1.0 V19( ) tmax = max = 50.1 tmax = max = 75.1 tmax = max = 100.1...

    work page arXiv 2020

  12. [12]

    Horak, J

    J. Horak, J. M. Pawlowski, J. Rodr´ ıguez-Quintero, J. Turnwald, J. M. Urban, N. Wink, and S. Zafeiropoulos, Phys. Rev. D105, 036014 (2022), 2107.13464

    work page arXiv 2022

  13. [13]

    Buzzicotti, A

    M. Buzzicotti, A. De Santis, and N. Tantalo, Eur. Phys. J. C84, 32 (2024), 2307.00808

    work page arXiv 2024

  14. [14]

    Aarts, K

    G. Aarts, K. Fukushima, T. Hatsuda, A. Ipp, S. Shi, L. Wang, and K. Zhou, Nature Rev. Phys.7, 154 (2025), 2501.05580. 27

    work page arXiv 2025

  15. [15]

    On the extraction of spectral densities from lattice correlators

    M. Hansen, A. Lupo, and N. Tantalo, Phys. Rev. D99, 094508 (2019), 1903.06476

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  16. [16]

    Bailas, S

    G. Bailas, S. Hashimoto, and T. Ishikawa, PTEP2020, 043B07 (2020), 2001.11779

    work page arXiv 2020

  17. [17]

    Alexandrou et al

    C. Alexandrou et al. (Extended Twisted Mass Collaboration (ETMC)), Phys. Rev. Lett.130, 241901 (2023), 2212.08467

    work page arXiv 2023

  18. [18]

    Barone, S

    A. Barone, S. Hashimoto, A. J¨ uttner, T. Kaneko, and R. Kellermann, JHEP07, 145 (2023), 2305.14092

    work page arXiv 2023

  19. [19]

    C. E. Creffield, E. G. Klepfish, E. R. Pike, and S. Sarkar, Phys. Rev. Lett.75, 517 (1995), URLhttps://link.aps.org/doi/10.1103/PhysRevLett.75.517

    work page doi:10.1103/physrevlett.75.517 1995

  20. [20]

    Shinaoka, J

    H. Shinaoka, J. Otsuki, M. Ohzeki, and K. Yoshimi, Phys. Rev. B96, 035147 (2017), URL https://link.aps.org/doi/10.1103/PhysRevB.96.035147

    work page doi:10.1103/physrevb.96.035147 2017

  21. [21]

    Chikano, J

    N. Chikano, J. Otsuki, and H. Shinaoka, Phys. Rev. B98, 035104 (2018), URLhttps: //link.aps.org/doi/10.1103/PhysRevB.98.035104

    work page doi:10.1103/physrevb.98.035104 2018

  22. [22]

    S. Tominaga, Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment243, 530 (1986), ISSN 0168-9002, URL https://www.sciencedirect.com/science/article/pii/0168900286909927

    work page arXiv 1986

  23. [23]

    Bruno, L

    M. Bruno, L. Giusti, and M. Saccardi, Phys. Rev. D111, 094515 (2025), 2407.04141

    work page arXiv 2025

  24. [24]

    Bellman and R

    R. Bellman and R. Roth, The Laplace Transform, Modern Applied Mathematics Series (World Scientific, 1984), ISBN 9789971966737, URLhttps://books.google.co.jp/books? id=roFjH53691wC

    work page 1984

  25. [25]

    J. G. McWhirter and E. R. Pike, Journal of Physics A: Mathematical and General11, 1729 (1978), URLhttps://doi.org/10.1088/0305-4470/11/9/007

    work page doi:10.1088/0305-4470/11/9/007 1978

  26. [26]

    C. L. Epstein and J. Schotland, SIAM Review50, 504 (2008), https://doi.org/10.1137/060657273, URLhttps://doi.org/10.1137/060657273

    work page doi:10.1137/060657273 2008

  27. [27]

    Kellermann, Z

    R. Kellermann, Z. Hu, A. Barone, A. Elgaziari, S. Hashimoto, T. Kaneko, and A. J¨ uttner, Phys. Rev. D112, 014501 (2025), 2504.03358

    work page arXiv 2025

  28. [28]

    Inclusive semileptonic $D_s\to X_s\ell\bar\nu$ decays from lattice QCD: continuum and chiral extrapolation

    R. Kellermann, A. Barone, A. Elgaziari, S. Hashimoto, Z. Hu, A. J¨ uttner, and T. Kaneko (2026), 2604.22201

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  29. [29]

    Gambino and S

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

    work page arXiv 2020

  30. [30]

    Bruno and M

    M. Bruno and M. T. Hansen, JHEP06, 043 (2021), 2012.11488

    work page arXiv 2021

  31. [31]

    Bulava and M

    J. Bulava and M. T. Hansen, Phys. Rev. D100, 034521 (2019), 1903.11735

    work page arXiv 2019

  32. [32]

    Frezzotti, N

    R. Frezzotti, N. Tantalo, G. Gagliardi, F. Sanfilippo, S. Simula, and V. Lubicz, Phys. Rev. D 28 108, 074510 (2023), 2306.07228

    work page arXiv 2023

  33. [33]

    Hosoya, M.-a

    A. Hosoya, M.-a. Sakagami, and M. Takao, Annals Phys.154, 229 (1984)

    work page 1984

  34. [34]

    E.-k. Wang, U. W. Heinz, and X.-f. Zhang, Phys. Rev. D53, 5978 (1996), hep-ph/9509331

    work page internal anchor Pith review Pith/arXiv arXiv 1996

  35. [35]

    H. B. Meyer, Phys. Rev. D76, 101701 (2007), 0704.1801. 29

    work page internal anchor Pith review Pith/arXiv arXiv 2007