pith. machine review for the scientific record. sign in

arxiv: 2605.00643 · v1 · submitted 2026-05-01 · ✦ hep-lat

Recognition: unknown

Variance reduction strategies for lattice QCD

Tim Harris
Authors on Pith no claims yet

Pith reviewed 2026-05-09 15:00 UTC · model grok-4.3

classification ✦ hep-lat
keywords lattice QCDvariance reductionquark propagatorscorrelation functionsWick contractionscomputational costprecision observables
0
0 comments X

The pith

Decompositions of quark propagators can reduce the variance of correlation function estimators in lattice QCD without introducing bias.

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

Lattice QCD predictions require heavy computation of correlation functions on gauge field ensembles, and this cost grows with the observable and its statistical variance. The variance often scales strongly with lattice spacing, volume, and quark separation, which limits access to the physical regime. The paper reviews observables built from quark propagators, including both connected and disconnected Wick contractions. It examines variance-reduction schemes that decompose those propagators to produce lower-noise estimators. If the decompositions work as described, they cut the resources needed for precision results and for simulations at large volumes.

Core claim

The paper states that variance-reduction schemes based on decompositions of the quark propagators already improve precision observables and offer a route to lower the computational cost of reaching large volumes, while preserving the exact physics content of the correlation functions.

What carries the argument

Variance-reduction schemes based on decompositions of the quark propagators, which split the propagator to form estimators whose statistical fluctuations are smaller than those of the undecomposed version.

If this is right

  • Reduced statistical error for the same number of gauge configurations in observables involving quark propagators.
  • Lower overall computational cost when targeting larger lattice volumes.
  • Better scaling of effort with decreasing lattice spacing for precision calculations.
  • Direct applicability to both quark-line connected and disconnected Wick contractions.

Where Pith is reading between the lines

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

  • The same decomposition approach could be combined with multilevel or multi-grid techniques to compound the cost savings.
  • If variance reduction proves robust across different fermion discretizations, it may become a standard ingredient in ensemble generation pipelines.
  • The scaling of variance with volume and separation supplies a diagnostic for choosing optimal lattice parameters before large-scale runs.

Load-bearing premise

The variance of the estimators depends strongly on physical and kinematical parameters, and the decompositions reduce this variance without introducing uncontrolled biases or approximations that affect the final physics results.

What would settle it

Measure the same set of correlation functions on an identical ensemble of gauge configurations once with the standard propagator and once with the decomposed version, then check whether the mean value remains unchanged while the variance drops by a measurable factor.

Figures

Figures reproduced from arXiv: 2605.00643 by Tim Harris.

Figure 1
Figure 1. Figure 1: Illustration of the Wick contractions which appear in the variances for a quark-line connected (left) and disconnected (right) primary observable. estimator. In general, however, the cost to achieve a reduced variance may not be so favourable and one should keep in mind that the relevant metric to be optimized is, for example the computational cost for a fixed precision, 𝜀 = √︁ 𝜎2/𝑁, which is total cost = … view at source ↗
Figure 2
Figure 2. Figure 2: The variance of the Hutchinson estimator for the single-propagator trace 𝑇 (1) 𝜇, 𝑓 for 𝜇 = 𝑘 (blue squares) and 𝜇 = 5 (red circles). The dashed lines indicate the gauge variance which differs by many orders of magnitude for the two cases unlike the stochastic variance which is identical if the reality of observable is ignored. The first line corresponds to the variance associated with the gauge fields (of… view at source ↗
Figure 3
Figure 3. Figure 3: One minus the deficit of the lowest modes in subspaces built from just a few 𝑁c = 20 (blue or yellow) or 100 (green or red) low modes. Decreasing the block sizes (indicated in lattice units) increases the subspace size and reduces the deficits further. cost. The use of such blocked low modes has been instrumental in the definition of efficient deflated or multigrid preconditioned solvers [33–35], and is cr… view at source ↗
Figure 4
Figure 4. Figure 4: Volume dependence of the variance of the levels of the improved estimators without using block-projection of the low modes (LMA, left) and with block projection (MG LMA, right) for a separation 𝑥0 − 𝑦0 ≈ 1.3 fm. The variance of the undeflated estimator is also shown with blue points, while the grey band indicates an estimate of the gauge variance. All variances for the stochastic estimators are computed wi… view at source ↗
Figure 5
Figure 5. Figure 5: The variance of the estimator for the trace of the difference of propagators S0 (left) or the single￾propagator trace 𝑇 (1) 𝑘, 𝑓 (right). In both cases the plain Hutchinson estimator (red points) is compared with the improved estimators, the split-even estimator for the difference (open squares, left) and two variants of the frequency-splitting estimator including the hopping expansion (blue circles, right… view at source ↗
Figure 6
Figure 6. Figure 6: Variance of two-point functions with two-level integration schemes. The left-hand panel shows the (square root of the) variance of the spatial correlator of the action density for a simulation of SU(3) pure gauge theory at high temperature. The right-hand panel shows the variance of the isovector vector correlator using the multi-boson domain decomposed HMC [64], compared with a standard HMC algorithm comp… view at source ↗
read the original abstract

A significant component of the cost of making predictions from lattice QCD stems from the computation of correlation functions on a given ensemble of gauge fields. This cost depends on the observable of interest and the details of its representation, including any approximation needed to estimate it. Moreover, the variance of such estimators may depend strongly on physical and kinematical parameters such as the lattice spacing, volume or separation, which gives an important insight into the costs of reaching the relevant physical limits. In these proceedings, I review some observables involving quark propagators, including both quark-line connected and disconnected Wick contractions, and discuss variance-reduction schemes based on decompositions of the quark propagators. Such strategies have already proven useful for precision physics observables and in future may help reduce the computational cost of reaching large volumes.

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 / 2 minor

Summary. The manuscript reviews variance reduction strategies for lattice QCD correlation functions involving quark propagators, covering both connected and disconnected Wick contractions. It motivates the topic by noting that estimator variance depends strongly on parameters such as lattice spacing, volume, and separation, then discusses propagator decomposition methods as a means to reduce this variance without introducing biases. The central claim is that these established strategies have already proven useful for precision observables and hold promise for lowering the computational cost of large-volume simulations.

Significance. If the review accurately compiles the literature, it provides a useful synthesis for lattice QCD practitioners facing high computational costs in precision calculations. The focus on practical, bias-free variance reduction via propagator decompositions addresses a key bottleneck, and the forward-looking discussion on large volumes is relevant to ongoing efforts in the field. As a proceedings-style review rather than a new derivation, its value lies in consolidation and motivation rather than novel predictions or machine-checked results.

major comments (1)
  1. Abstract: the claim that 'such strategies have already proven useful for precision physics observables' is load-bearing for the paper's significance but is stated without a specific example, quantitative improvement factor, or citation to a landmark application; the main text must supply at least one concrete case (e.g., a referenced calculation of a matrix element or form factor) to substantiate the assertion.
minor comments (2)
  1. The abstract introduces 'decompositions of the quark propagators' without naming the principal classes (e.g., low-mode averaging, all-mode averaging, or deflation); a short taxonomy or equation in the introduction would improve accessibility.
  2. Ensure that every referenced variance-reduction technique is accompanied by a full bibliographic citation; the current abstract contains none.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their positive assessment of the manuscript and for the constructive suggestion regarding the abstract. We address the major comment below.

read point-by-point responses

  1. Referee: Abstract: the claim that 'such strategies have already proven useful for precision physics observables' is load-bearing for the paper's significance but is stated without a specific example, quantitative improvement factor, or citation to a landmark application; the main text must supply at least one concrete case (e.g., a referenced calculation of a matrix element or form factor) to substantiate the assertion.

    Authors: We agree that the claim in the abstract benefits from explicit substantiation in the main text. Although the manuscript reviews a range of applications of propagator decomposition methods for both connected and disconnected contractions, we have revised the introduction to include a concrete example of a precision observable. The updated text now references a specific lattice QCD calculation of a matrix element where these variance reduction strategies yielded a documented improvement in statistical precision, together with the corresponding citation. This addition directly addresses the referee's request without altering the overall scope of the review. revision: yes

Circularity Check

0 steps flagged

No significant circularity: review of prior literature

full rationale

The manuscript is explicitly a review/proceedings summary of established variance-reduction techniques for quark-propagator observables in lattice QCD. It states background facts about variance dependence on lattice parameters and summarizes existing decomposition methods from the literature without introducing new derivations, quantitative predictions, or first-principles results that could reduce to fitted inputs or self-citations by construction. No equations or claims are presented that equate outputs to inputs via definition or internal fitting; the forward-looking statement about future cost reduction is a qualitative summary rather than a testable prediction internal to the paper.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is a review paper; the central claim rests on summaries of prior literature rather than new derivations, so no free parameters, axioms, or invented entities are introduced in the provided abstract.

pith-pipeline@v0.9.0 · 5410 in / 1082 out tokens · 38209 ms · 2026-05-09T15:00:25.338166+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

68 extracted references · 50 canonical work pages · 2 internal anchors

  1. [1]

    K. G. Wilson. In:Phys. Rev. D10 (1974). Ed. by J. C. Taylor, pp. 2445–2459

    1974

  2. [2]

    Boyd et al

    G. Boyd et al. In:Nucl. Phys. B469 (1996), pp. 419–444. arXiv:hep-lat/9602007. 15 Variance reduction strategies for lattice QCDTim Harris

    work page arXiv 1996

  3. [3]

    L.Altenkortetal.In:Phys.Rev.D105.9(2022),p.094505.arXiv:2112.02282 [hep-lat]

    work page arXiv 2022

  4. [4]

    G. Parisi. In:Phys. Rept.103 (1984), pp. 203–211

    1984

  5. [5]

    G. P. Lepage. In:Boulder ASI 1989:97-120. 1989, pp. 97–120

    1989

  6. [6]

    M. Lüscher. In:Les Houches Summer School: Session 93: Modern perspectives in lattice QCD: Quantum field theory and high performance computing. Feb. 2010, pp. 331–399. arXiv:1002.4232 [hep-lat]

    work page arXiv 2010

  7. [7]

    G.Balietal.In:PoSLAT2009(2009).Ed.byC.LiuandY.Zhu,p.149.arXiv:0911.2407 [hep-lat]

    work page arXiv 2009

  8. [8]

    Blum et al

    T. Blum et al. In:Phys. Rev. D93.1 (2016), p. 014503. arXiv:1510.07100 [hep-lat]

    work page arXiv 2016

  9. [9]

    Liu et al

    K.-F. Liu et al. In:Phys. Rev. D97.3 (2018), p. 034507. arXiv:1705.06358 [hep-lat]

    work page arXiv 2018

  10. [10]

    H. B. Meyer. In:Eur. Phys. J. C77.9 (2017), p. 616. arXiv:1706.01139 [hep-lat]

    work page arXiv 2017

  11. [11]

    M. Lüscher. In:EPJ Web Conf.175 (2018). Ed. by M. Della Morte et al., p. 01002. arXiv: 1707.09758 [hep-lat]

    work page arXiv 2018

  12. [12]

    Giusti and M

    L. Giusti and M. Lüscher. In:Eur. Phys. J. C79.3 (2019), p. 207. arXiv:1812 . 02062 [hep-lat]

    2019

  13. [13]

    Bruno et al

    M. Bruno et al. In:JHEP11 (2023), p. 167. arXiv:2307.15674 [hep-lat]

    work page arXiv 2023

  14. [14]

    Bernecker and H

    D. Bernecker and H. B. Meyer. In:Eur. Phys. J. A47 (2011), p. 148. arXiv:1107.4388 [hep-lat]

    work page arXiv 2011

  15. [15]

    Della Morte et al

    M. Della Morte et al. In:JHEP08 (2005), p. 051. arXiv:hep-lat/0506008

    work page arXiv 2005

  16. [16]

    M.DellaMorteandA.Juttner.In:JHEP11(2010),p.154.arXiv:1009.3783 [hep-lat]

    work page arXiv 2010

  17. [17]

    Giusti et al

    L. Giusti et al. In:Eur. Phys. J. C79.7 (2019), p. 586. arXiv:1903.10447 [hep-lat]

    work page arXiv 2019

  18. [18]

    T. Harris. In:PoSEuroPLEx2023 (2024), p. 011

    2024

  19. [19]

    Altherr et al

    A. Altherr et al. In:PoSLATTICE2024 (2025), p. 116. arXiv:2502.03145 [hep-lat]

    work page arXiv 2025

  20. [20]

    Altherr et al

    A. Altherr et al. In:JHEP10 (2025), p. 158. arXiv:2506.19770 [hep-lat]

    work page arXiv 2025

  21. [21]

    G. M. de Divitiis et al. In:Phys. Rev. D87.11 (2013), p. 114505. arXiv:1303 . 4896 [hep-lat]

    2013

  22. [22]

    Bitar et al

    K. Bitar et al. In:Nucl. Phys. B313 (1989), pp. 348–376

    1989

  23. [23]

    M.Hutchinson.In:CommunicationsinStatistics-SimulationandComputation19.2(1990), pp. 433–450

    1990

  24. [24]

    Dong and K.-F

    S.-J. Dong and K.-F. Liu. In:Phys. Lett. B328 (1994), pp. 130–136. arXiv:hep - lat / 9308015

    1994

  25. [25]

    G.M.deDivitiisetal.In:Phys.Lett.B382(1996),pp.393–397.arXiv:hep-lat/9603020

    work page arXiv 1996

  26. [26]

    Michael and J

    C. Michael and J. Peisa. In:Phys. Rev. D58 (1998), p. 034506. arXiv:hep-lat/9802015

    work page arXiv 1998

  27. [27]

    Bernardson et al

    S. Bernardson et al. In:Computer Physics Communications78.3 (1994), pp. 256–264.issn: 0010-4655. 16 Variance reduction strategies for lattice QCDTim Harris

    1994

  28. [28]

    Gruber et al

    R. Gruber et al. In:PoSLATTICE2023 (2024), p. 153. arXiv:2401.14724 [hep-lat]

    work page arXiv 2024

  29. [29]

    Giusti et al

    L. Giusti et al. In:JHEP04 (2004), p. 013. arXiv:hep-lat/0402002

    work page arXiv 2004

  30. [30]

    T.A.DeGrandandS.Schaefer.In:Comput.Phys.Commun.159(2004),pp.185–191.arXiv: hep-lat/0401011

    work page arXiv 2004

  31. [31]

    Foley et al

    J. Foley et al. In:Comput. Phys. Commun.172 (2005), pp. 145–162. arXiv:hep - lat / 0505023

    2005

  32. [32]

    T.BanksandA.Casher.In:NuclearPhysicsB169.1(1980),pp.103–125.issn:0550-3213

    1980

  33. [33]

    M. Lüscher. In:JHEP07 (2007), p. 081. arXiv:0706.2298 [hep-lat]

    work page arXiv 2007

  34. [34]

    Babich, J

    R. Babich et al. In:Phys. Rev. Lett.105 (2010), p. 201602. arXiv:1005.3043 [hep-lat]

    work page arXiv 2010

  35. [35]

    Frommer et al

    A. Frommer et al. In:SIAM J. Sci. Comput.36.4 (2014), A1581–A1608. arXiv:1303.1377 [hep-lat]

    work page arXiv 2014

  36. [36]

    M. A. Clark et al. In:EPJ Web Conf.175 (2018). Ed. by M. Della Morte et al., p. 14023. arXiv:1710.06884 [hep-lat]

    work page arXiv 2018

  37. [37]

    Gruber, T

    R. Gruber et al. In:Phys. Rev. D111.7 (2025), p. 074508. arXiv:2412.06347 [hep-lat]

    work page arXiv 2025

  38. [38]

    Frommer et al

    A. Frommer et al. In: (Sept. 2025). arXiv:2509.11424 [hep-lat]

    work page arXiv 2025

  39. [39]

    K.JansenandR.Sommer.In:Nucl.Phys.B530(1998).[Erratum:Nucl.Phys.B643,517–518 (2002)], pp. 185–203. arXiv:hep-lat/9803017

    work page arXiv 1998

  40. [40]

    Fritzsch et al

    P. Fritzsch et al. In:Nucl. Phys. B865 (2012), pp. 397–429. arXiv:1205.5380 [hep-lat]

    work page arXiv 2012

  41. [41]

    Lüscher et al.https://luscher.web.cern.ch/luscher/openQCD

    M. Lüscher et al.https://luscher.web.cern.ch/luscher/openQCD. Accessed: 2024

    2024

  42. [42]

    Stathopoulos and J

    A. Stathopoulos and J. R. McCombs. In:ACM Transactions on Mathematical Software37.2 (2010), 21:1–21:30

    2010

  43. [43]

    Dynamical Twisted Mass Fermions with Light Quarks: Simulation and Analysis Details,

    P. Boucaud et al. In:Comput. Phys. Commun.179 (2008), pp. 695–715. arXiv:0803.0224 [hep-lat]

    work page arXiv 2008

  44. [44]

    R. Gruber. PhD thesis. Zurich, ETH, 2025

    2025

  45. [45]

    Peardon, J

    M. Peardon et al. In:Phys. Rev. D80 (2009), p. 054506. arXiv:0905.2160 [hep-lat]

    work page arXiv 2009

  46. [46]

    Gülpers et al

    V. Gülpers et al. In:PoSLATTICE2014 (2014), p. 128. arXiv:1411.7592 [hep-lat]

    work page arXiv 2014

  47. [47]

    Dinter et al

    S. Dinter et al. In:JHEP08 (2012), p. 037. arXiv:1202.1480 [hep-lat]

    work page arXiv 2012

  48. [48]

    Ceet al.,hep-lat/2203.08676

    M. Cè et al. In:JHEP08 (2022), p. 220. arXiv:2203.08676 [hep-lat]

    work page arXiv 2022

  49. [49]

    Hybrid calculation of hadronic vacuum polarization in muon g-2 to 0.48\%

    A. Boccaletti et al. In: (July 2024). arXiv:2407.10913 [hep-lat]

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  50. [50]

    Djukanovic et al

    D. Djukanovic et al. In:Phys. Rev. D109.9 (2024), p. 094510. arXiv:2309 . 06590 [hep-lat]

    2024

  51. [51]

    Djukanovic, G

    D. Djukanovic et al. In:JHEP04 (2025), p. 098. arXiv:2411.07969 [hep-lat]

    work page arXiv 2025

  52. [52]

    Efficiently unquenching QCD+QED at O($\alpha$)

    T. Harris et al. In:PoSLATTICE2022 (2023), p. 013. arXiv:2301.03995 [hep-lat]

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  53. [53]

    Hodgson et al

    R. Hodgson et al. In:PoSLATTICE2024 (2025), p. 258. arXiv:2501.18358 [hep-lat]

    work page arXiv 2025

  54. [54]

    Thron et al

    C. Thron et al. In:Phys. Rev. D57 (1998), pp. 1642–1653. arXiv:hep-lat/9707001. 17 Variance reduction strategies for lattice QCDTim Harris

    work page arXiv 1998

  55. [55]

    G.S.Balietal.In:Comput.Phys.Commun.181(2010),pp.1570–1583.arXiv:0910.3970 [hep-lat]

    work page arXiv 2010

  56. [56]

    Gülpers et al

    V. Gülpers et al. In:Phys. Rev. D89.9 (2014), p. 094503. arXiv:1309.2104 [hep-lat]

    work page arXiv 2014

  57. [57]

    J.M.TangandY.Saad.In:NumericalLinearAlgebrawithApplications19.3(2012),pp.485– 501

    2012

  58. [58]

    A.Stathopoulosetal.In:SIAMJ.Sci.Comput.35.5(2013),S299–S322.arXiv:1302.4018 [hep-lat]

    work page arXiv 2013

  59. [59]

    Whyte et al

    T. Whyte et al. In:Comput. Phys. Commun.294 (2024), p. 108928. arXiv:2212.04430 [hep-lat]

    work page arXiv 2024

  60. [60]

    M.LüscherandP.Weisz.In:JHEP09(2001),p.010.arXiv:hep-lat/0108014 [hep-lat]

    work page arXiv 2001

  61. [61]

    H. B. Meyer. In:JHEP01 (2003), p. 048. arXiv:hep-lat/0209145

    work page arXiv 2003

  62. [62]

    Parisi et al

    G. Parisi et al. In:Phys. Lett. B128 (1983), pp. 418–420

    1983

  63. [63]

    García Vera and S

    M. García Vera and S. Schaefer. In:Phys. Rev. D93 (2016), p. 074502. arXiv:1601.07155 [hep-lat]

    work page arXiv 2016

  64. [64]

    M.DallaBridaetal.In:Phys.Lett.B816(2021),p.136191.arXiv:2007.02973 [hep-lat]

    work page arXiv 2021

  65. [65]

    Cè et al

    M. Cè et al. In:Phys. Rev.D93.9 (2016), p. 094507. arXiv:1601.04587 [hep-lat]

    work page arXiv 2016

  66. [66]

    Cè et al

    M. Cè et al. In:Phys. Rev.D95.3 (2017), p. 034503. arXiv:1609.02419 [hep-lat]

    work page arXiv 2017

  67. [67]

    Lüscher, Martin. In:Nucl. Phys. B418 (1994), pp. 637–648. arXiv:hep-lat/9311007

    work page arXiv 1994

  68. [68]

    Barca et al

    L. Barca et al. In:Phys. Rev. D113.3 (2026), p. 034505. arXiv:2512.10644 [hep-lat]. 18

    work page arXiv 2026