Pith. Sign in

REVIEW

Schwarz-preconditioned HMC algorithm for two-flavour lattice QCD

PaperPDF from arXiv

data files

ReviewThe objections and verdict, as JSON Paper breakdownThe authors' claims, conditions and measurements, as JSON Signed receiptProof this review has not been altered
X Bluesky LinkedIn Reddit Hacker News
Martin Lüscher

High Energy Physics - Lattice · Submitted 17 Sep 2004 · 35 references

Pith's verdict: A careful, honest methods paper: the Schwarz solver preconditioner promoted to a determinant factorization for HMC, with real simulations behind it — the ancestor of the SAP/GDDHMC family.

T0 · AI review

How this review was made: An AI referee (selfhosted/deepseek-v4.1-flash) read the paper on November 11, 2026 and answered a fixed list of questions about it. T0 is the starting level every paper gets. The levels describe how deeply Pith has checked the paper, never the authors.

  1. T0 AI review An AI read the paper and answered a fixed list of questions.
  2. T1 Audit Automatic checks that the references and files the paper relies on exist and match.
  3. T2 Compute The paper's code or calculations re-run with the same results.
  4. T3 Formal statement The claims written out in a formal math language a computer can read.
  5. T4 Proof Those claims proved, and the proof checked by computer.

How Pith checks papers →

· 4 major objections · 8 minor · Ask about this paper · The compiled record · Read the full review →
Challenge this review Comment (0) Read on arXiv ↗
Cite ▾
arXiv IDhep-lat/0409106 (version 1)
Pith IDH3LCXQ3S, Pith's permanent record of this paper and its review
CitationMartin L\"uscher. "Schwarz-preconditioned HMC algorithm for two-flavour lattice QCD." arXiv:hep-lat/0409106 (2004). https://pith.science/paper/H3LCXQ3S
Cite this reviewBibTeX for Pith's review of this paper
Claim paper

Why Pith says this

Highlight any text to comment on it.

Referee report

How Pith checks papers → · Full audit of this paper →

The paper

The paper claims

Preconditioning the HMC algorithm itself by a non-overlapping Schwarz domain decomposition cuts the cost per statistically independent configuration in two-flavour lattice QCD, with the gain growing as the quark mass is lowered.

The reading

Simulating QCD on a computer means sampling gauge-field configurations with a Markov chain. The standard engine, Hybrid Monte Carlo (HMC), invents fictitious momenta, pushes the gauge field along a classical trajectory, and accepts or rejects the endpoint. The expensive part is the quark determinant, which enters through pseudo-fermion fields and produces forces that must be integrated with small step sizes, especially at light quark masses. The cost per trajectory therefore rises steeply as the quark mass falls, and this is the main obstacle to simulating realistic up and down quarks. Preconditioning usually means speeding up the linear solver inside HMC. This paper does something different: it factorizes the quark determinant itself. The lattice is tiled with non-overlapping blocks, coloured like a chessboard into two domains. The determinant splits into a product of block determinants and a Schur complement living on the domain boundaries. Each factor gets its own pseudo-fermion field and its own force term, integrated on its own time scale. A second ingredient is to update only 'active' links, those with both endpoints inside one block and at most one endpoint on the block's inner boundary. This decouples the blocks, lets them be integrated in parallel, and shrinks the most expensive force by roughly an order of magnitude near the block centres. A random translation of the gauge field after each cycle keeps all links visited equally. The author measures force magnitudes (roughly 32:8:1 across the three terms), tunes step numbers, checks reversibility to 1e-9, and reports a machine-independent cost figure: the number of full-lattice Dirac solves needed per independent plaquette measurement. Across the tested masses that figure stays between 0.28 and 0.86, against 1.0
Extended reading notes

Core claim

The abstract states the central claim directly: 'The combination of a non-overlapping Schwarz preconditioner and the Hybrid Monte Carlo (HMC) algorithm is shown to yield an efficient simulation algorithm for two-flavour lattice QCD with Wilson quarks' (abstract:s0). The quantitative form of the claim is the cost figure nu = 10^-3 (2N2+3) tau_int[P] in eq. (4.10), the number of full-lattice Dirac solves per statistically independent plaquette measurement. Table 4 reports nu between 0.28(9) and 0.86(27) across the tested masses, against plain-HMC values of 1.0(6) and 2.8(8) at matched parameters on the 32x16^3 lattice, and 1.8(8) and 5.1(5) on 40x24^3. If correct, the algorithm's cost does not degrade as the quark mass falls over the tested range.

Load-bearing premise

The efficiency claim rests on the integrated autocorrelation time of the average plaquette, tau_int[P], estimated by the truncation rule of appendix E.13 on time series of 800 to 9400 trajectories. The author states in sect. 4.5 that 'the available statistics is insufficient for a solid error of the error analysis' and that the quoted values 'should therefore be taken as first estimates only'. Table 2 shows errors of 30 to 40 percent (68(25), 168(42), 33(11)). Since nu in eq. (4.10) is linear in tau_int[P], a systematic bias in the summation-window rule would move the headline numbers directly. The claim that there is 'no indication that the autocorrelation times grow when the quark mass is lowered' is therefore a statement about three mass points with large error bars, not a measured scaling law.

Original abstract

The combination of a non-overlapping Schwarz preconditioner and the Hybrid Monte Carlo (HMC) algorithm is shown to yield an efficient simulation algorithm for two-flavour lattice QCD with Wilson quarks. Extensive tests are performed, on lattices of size up to 32x24x24x24, with lattice spacings a~0.08 fm and at bare current-quark masses as low as 21 MeV.

The compiled record

The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.

Compiled by classifier-1.13-20260917 via OpenRouter, question bank abstract-p1-20260923, 2026-09-23: 1 request, 2,390 input tokens, $0.0001 reported.

What the abstract itself states, as probabilities: claims first 0.03 · states limitation 0.09 · human participants 0.01 · controlled comparison 0.05 · randomized assignment 0.01 · states quantitative result 0.87 · compares with prior results 0.11 · mentions code or data availability 0.02.

The abstract, sentence by sentence

arXiv listing abstract v1. Results and the main claim are underlined; .

The combination of a non-overlapping Schwarz preconditioner and the Hybrid Monte Carlo (HMC) algorithm is shown to yield an efficient simulation algorithm for two-flavour lattice QCD with Wilson quarks. Extensive tests are performed, on lattices of size up to 32x24x24x24, with lattice spacings a~0.08 fm and at bare current-quark masses as low as 21 MeV.

background contribution method result interpretation main claim

What the classifier reports for each sentence. The number is the probability it gave the label.
#sentencerolethe authors' own workstates the main conclusionstrength
s0 The combination of a non-overlapping Schwarz preconditioner and the Hybrid Monte Carlo (HMC) algorithm is shown to yield an efficient simulation algorithm for two-flavour lattice QCD with Wilson quarks. contribution 0.50 0.89 0.92 asserted 1.00
s1 Extensive tests are performed, on lattices of size up to 32x24x24x24, with lattice spacings a~0.08 fm and at bare current-quark masses as low as 21 MeV. method 0.90 0.94 0.05 asserted 0.96

What kind of paper the classifier reports this is: method or tool 0.90 · evidence from simulation or computation 1.00.

What was examined, and what was not

sources read
  • arXiv listing abstract v1
not examined
  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers
levels run
  • P abstract
not run
  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. Every number in the measurements table is the authors', checked by code. Record as JSON

Figures
Figure 1. Two-dimensional cross-section of a 24 × 123 lattice covered by non-overlap￾ping 64 blocks Λ. The domains Ω and Ω∗ are the unions of the black and white blocks respectively, and their exterior boundaries ∂Ω and ∂Ω ∗ consist of all points in the complementary domain represented by open circles. It is often convenient to let these operators act on quark fields that are defined on the whole lattice rather than on Ω or Ω∗ only. The extension is done in the obvious way by padding with zeros so that eq. (3.1), for example, may be written as D = DΩ + DΩ∗ + D∂Ω + D∂Ω∗ . (3.2) Similarly the
Figure 2. Two-dimensional view of a 64 block showing the links on which the active link variables reside. Such links must have both endpoints in the same block and at most one endpoint on the interior boundary of the block (open circles). other field variables fixed (see fig. 2). At the beginning or the end of every update cycle, the gauge field should then be translated by a random vector v, U(x, µ) → U(x + v, µ) for all x, µ, (3.9) to ensure that all link variables are treated equally on average. Evidently, the same can also be achieved using several block coverings of the lattice…
Figure 3. Average magnitude of the forces Fk at the end of the molecular dynamics trajectories on the 32×163 lattice at distance d from the block boundary. Dotted lines are drawn to guide the eyes, and the three curves for F2 correspond to the different quark masses (the mass dependence of the other curves is negligible). X of the Lie algebra of SU(3), the convention kXk 2 = −2tr{X2} was adopted here, and an average was taken over all gauge configurations and all links at the specified distance from the block boundary. The magnitudes of the forces are thus strongly ordered, approximately like…
Figure 4. Histories of ∆H and of the deviation ∆P of the average plaquette from its mean value at κ = 0.15800 on the 32 × 243 lattice as a function of the trajectory number. Only the values after every 5th trajectory are plotted. the modification amounts to replaying the trajectories with values of |∆H| above a specified threshold, using a smaller step size and a somewhat complicated acceptance rule that preserves detailed balance. For illustration, a typical time series of ∆H (after including the replay trick) is shown in fig. 4. The replay threshold was set to 1.5 in this case and the…
Figure 5. Probability distribution of the average iteration n
Figure 6. Normalized autocorrelation functions of the average
Figure 7. Average number hNcfi of accepted gauge-field configurations generated per day by the Schwarz-preconditioned HMC algorithm on 8 nodes of a recent PC cluster (described in ref. [20], appendix B), as a function of the bare current-quark mass. The point at the smallest mass on the 32 × 243 lattice is from the last run in table 1 and the linear scaling curves (dotted lines) are drawn for comparison with the data. is not small on these lattices [5,6]. In physical units the quark and the pion masses quoted on the last line of table 3 are about 21 MeV and 386 MeV respectively (if the scale…
Figure 8. The hopping terms in D∂Ω∗ move the quark spinors from the interior to the exterior boundary points of the black blocks. Depending on the position of the inner point, the spinor is moved in one or more directions. link variables. In particular, two components of the spinors residing on the subset [∂Ω ∗ ] =  x ∈ ∂Ω ∗
Figure 9. Three copies of a block grid in two dimensions, two of t

Figures reproduced from arXiv:hep-lat/0409106 by the authors.

References
How later papers used it

51 later papers cite it

Evidence: substantial. 51 of the citing papers share no author with it, and 32 of those carry a confident label from Pith's citation classifier.

Pith read 20 of the 67 citing sentences, choosing only those whose citation markers point at this paper alone, and wrote the paraphrases from them. The summary, the counts, the evidence level, the links and the quotes come from code; the summary counts the papers listed below. A later paper is listed under a label when Pith's citation classifier gives one of its sentences that label at probability 0.8 or more (0.95 for support), Pith's reading finds that the citing paper itself did it, and a second check confirms the paraphrase against the sentence. This section lists uses only. It does not list criticism or disagreement, so their absence here says nothing about how the paper has been received. Papers that share an author with this one are listed apart, under "By the same authors". A quote is the citing authors' own sentence, shown when their paper's licence is CC BY, CC BY-SA or CC0. JSON

Uses its method or tool 12 of 30 papers with this label
How the literature uses it

49 arXiv papers cite this in 66 sentences

Each citing sentence is taken from the citing paper's LaTeX source and typed by Pith's classifier. A label appears at probability 0.8 or more (0.95 for support), a contest at 0.7 or more; the quoted words are the citing authors'. How each label was checked · JSON

Papers citing it · 24 sentences without a confident label

Uses its method or tool 30 papers
Compares with it 1 paper
Cites it as background 4 papers

Discussion

Disagree with the review, or know something it missed? Leave a comment, highlight any passage above to comment on it, or challenge this review formally.