arxiv: 2604.01287 · v1 · submitted 2026-04-01 · ✦ hep-lat

Recognition: 1 theorem link

· Lean Theorem

Enhanced Sampling Techniques for Lattice Gauge Theory

Christian Hoelbling, Gianluca Fuwa, Lukas Varnhorst, Timo Eichhorn

Authors on Pith no claims yet

Pith reviewed 2026-05-13 21:57 UTC · model grok-4.3

classification ✦ hep-lat

keywords metadynamicsenhanced samplingtopological freezinglattice QCDautocorrelation timebias potentialSU(N) gauge theoryHMC

0 comments

The pith

Metadynamics with bias potentials reduces autocorrelation times for topological charge in lattice gauge theories.

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

The paper shows that conventional Monte Carlo updates in lattice QCD and similar theories become trapped in fixed topological sectors because action barriers grow with volume. By adding bias potentials that flatten the distribution of the topological charge, metadynamics and related methods lower these barriers and let the simulation cross between sectors much more often. The authors test how to build the bias potentials faster and whether potentials learned on small volumes can be reused on large ones. They also check whether longer molecular-dynamics trajectories or other HMC variants help on their own. If the bias approach works, large-volume simulations that were previously impractical become feasible while preserving correct equilibrium statistics.

Core claim

With appropriately constructed bias potentials, metadynamics and related enhanced sampling techniques can mitigate topological freezing and significantly reduce the integrated autocorrelation times of the topological charge and associated observables in four-dimensional SU(N) gauge theories with periodic boundary conditions.

What carries the argument

bias potentials that flatten the topological charge distribution to accelerate sector transitions

If this is right

Integrated autocorrelation times for the topological charge drop by a large factor once the bias potential is active.
Bias potentials learned on modest volumes can be transferred to production runs at larger volumes.
Longer HMC trajectories and related algorithmic tweaks provide an orthogonal improvement that can be combined with the bias method.
The same bias construction applies to both QCD and pure SU(N) gauge theories.
Faster sampling of topological sectors makes observables that depend on topology accessible at volumes where conventional runs freeze.

Where Pith is reading between the lines

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

If extrapolation of bias potentials succeeds, the computational cost of generating independent topological samples scales much more favorably with volume than with standard algorithms.
The method could be combined with other volume-independent sampling strategies such as parallel tempering in the gauge coupling.
Successful reuse of small-volume biases would allow systematic studies of theta dependence at physical volumes without prohibitive autocorrelation.

Load-bearing premise

Bias potentials constructed on small volumes or during short runs can be extrapolated or reused on large volumes without introducing uncontrolled systematic errors.

What would settle it

A direct comparison in which the histogram of topological charge obtained with an extrapolated bias potential deviates from the histogram obtained in an unbiased long run on the same large volume.

Figures

Figures reproduced from arXiv: 2604.01287 by Christian Hoelbling, Gianluca Fuwa, Lukas Varnhorst, Timo Eichhorn.

**Figure 1.** Figure 1: shows the VES parameter evolution and CV time series in a four-dimensional SU(3) simulation with the DBW2 action at 𝐿/𝑎 = 16 and 𝛽 = 1.25, using averaged SGD with heavy ball momentum, batches of 50 HMC trajectories of length 4, and a flat target distribution. 0 10 20 30 40 50 60 SGD iteration −20 −10 0 Parameter values 0 500 1000 1500 2000 2500 3000 Trajectory −1 0 1 Collective variable α1 α2 α¯1 α¯2 Refer… view at source ↗

**Figure 2.** Figure 2: Convolution-based volume extrapolation of bias potentials at 𝛽 = 6.1912 in four-dimensional SU(3) gauge theory. The 𝐿/𝑎 = 24 reference potential was reconstructed from an unbiased simulation from [33]. The extrapolation from 𝐿/𝑎 = 12 to 𝐿/𝑎 = 24 is inaccurate due to sizeable finite volume effects. Since the volume ratio 244 /204 = 2.0736 is close to 2, a good estimate for the bias potential at the larger v… view at source ↗

**Figure 3.** Figure 3: Effect of the HMC trajectory length on the sampling efficiency for the energy density 𝐸 and the squared topological charge 𝑄 2 . The additional computational overhead of longer trajectories has already been taken into account. 1 2 4 8 Trajectory length T 0 10 20 30 τint E 1 2 4 8 Trajectory length T Q2 L/a = 12 β = 5.885707 L/a = 14 β = 5.969994 L/a = 16 β = 6.049472 L/a = 18 β = 6.123853 [PITH_FULL_IMAGE… view at source ↗

**Figure 4.** Figure 4: Effect of the HMC trajectory length on the integrated autocorrelation times of the energy density 𝐸 and the squared topological charge 𝑄 2 . The additional computational overhead of longer trajectories has already been taken into account. The solid lines are fits of the form 𝑐/ √ 𝑇 + 0.5 to all data points, while the dashed lines correspond to the same functional form with 𝑐 = 𝜏int,𝑇=1. 6 [PITH_FULL_IMAGE… view at source ↗

read the original abstract

In theories with topological sectors, such as lattice QCD and four-dimensional SU(N) gauge theories with periodic boundary conditions, conventional update algorithms suffer from topological freezing due to large action barriers separating distinct sectors. With appropriately constructed bias potentials, Metadynamics and related enhanced sampling techniques can mitigate this problem and significantly reduce the integrated autocorrelation times of the topological charge and associated observables. We test strategies to accelerate the buildup of bias potentials and the possibility of extrapolating potentials from small to large volumes. We also investigate the effectiveness of orthogonal algorithmic improvements, such as longer HMC trajectories and HMC variants, which may benefit conventional simulations as well.

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.

Desk Editor's Note private letter to a colleague

The paper tests metadynamics bias potentials and small-to-large volume extrapolation on lattice gauge theories to cut topological autocorrelation, but the extrapolation step still needs direct validation against unbiased large-volume runs.

read the letter

The main thing to know is that this work takes metadynamics and related bias-potential techniques and runs concrete tests on them for the topological freezing problem in four-dimensional gauge theories. They focus on speeding up bias buildup and on extrapolating the potentials from small volumes to large ones, while also checking longer HMC trajectories as a separate improvement. That targeted testing is the actual new piece; the methods themselves are imported from molecular dynamics, but the volume-extrapolation angle and the lattice-specific checks are not standard prior art here. The authors lay out the problem clearly and show they have tried practical strategies rather than just proposing an idea in the abstract. For anyone running Monte Carlo on topological observables, seeing how they handle the bias construction and the extrapolation attempts is useful information even if the gains turn out modest. The soft spot is exactly the one the stress-test note flags. Without an explicit cross-check—comparing reweighted topological charge or plaquette distributions from the extrapolated potential against fresh, unbiased HMC on the same large volume—there is no direct evidence that the extrapolated bias preserves the correct equilibrium distribution. Any volume-dependent mismatch would feed straight into the observables and could make the reported autocorrelation reductions look better than they are. The abstract gives no numbers or error estimates, so the strength of the central claim is still hard to judge from the visible material. This paper is for lattice gauge theorists who need better sampling tools for topology. A reader already working on algorithmic improvements or on topological observables would get practical value from the test strategies. It is coherent enough and addresses a real bottleneck, so it deserves a serious referee rather than a desk reject; the extrapolation method is worth having someone check in detail.

Referee Report

1 major / 1 minor

Summary. The manuscript investigates the application of Metadynamics and related enhanced sampling techniques to lattice gauge theories with topological sectors, where conventional algorithms suffer from topological freezing. It claims that appropriately constructed bias potentials can mitigate this issue and significantly reduce integrated autocorrelation times for the topological charge and associated observables. The authors test strategies for accelerating bias potential buildup, extrapolating potentials from small to large volumes, and orthogonal improvements such as longer HMC trajectories and HMC variants.

Significance. If the extrapolation and reweighting procedures preserve the correct equilibrium ensemble without uncontrolled systematics, the work could enable more efficient simulations of topological observables on larger volumes in lattice QCD and SU(N) gauge theories, addressing a persistent barrier in the field. The empirical testing of established enhanced-sampling algorithms on a well-known lattice problem is a methodological strength.

major comments (1)

[volume extrapolation section] Volume extrapolation tests: the manuscript examines extrapolation of bias potentials from small to large volumes but reports no explicit cross-validation, such as direct comparison of reweighted topological charge histograms or plaquette expectation values against independent, unbiased HMC runs performed on the same large volume. This validation is required to confirm that the target equilibrium distribution is recovered and that reported reductions in autocorrelation times are not artifacts of potential mismatch.

minor comments (1)

[Abstract] The abstract states that the techniques 'significantly reduce' autocorrelation times yet supplies no numerical values, error estimates, or baseline comparisons, which weakens the reader's ability to gauge the practical improvement.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading of the manuscript and for highlighting the need for stronger validation of the volume-extrapolation procedure. We address this point below and have revised the manuscript to include the requested cross-checks on volumes where they are computationally feasible.

read point-by-point responses

Referee: [volume extrapolation section] Volume extrapolation tests: the manuscript examines extrapolation of bias potentials from small to large volumes but reports no explicit cross-validation, such as direct comparison of reweighted topological charge histograms or plaquette expectation values against independent, unbiased HMC runs performed on the same large volume. This validation is required to confirm that the target equilibrium distribution is recovered and that reported reductions in autocorrelation times are not artifacts of potential mismatch.

Authors: We agree that direct cross-validation against unbiased HMC on the target large volumes would provide the strongest confirmation. In the original submission we presented consistency checks based on plaquette values and the convergence of the bias potential with volume, but we did not include explicit histogram comparisons on the largest volumes. We have now added such comparisons for all volumes where unbiased HMC runs could be completed within available resources. On these intermediate volumes the reweighted topological-charge histograms and plaquette expectation values agree with the direct HMC results within statistical errors. For the very largest volumes, where topological freezing renders unbiased sampling impractical, we have added a quantitative discussion of the residual extrapolation uncertainty and demonstrate that the bias potential approaches a volume-independent limiting form. These additions are included in the revised Section on volume extrapolation and in a new appendix containing the histogram overlays. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical testing of external algorithms on lattice observables

full rationale

The manuscript reports numerical experiments applying Metadynamics and related bias-potential methods to mitigate topological freezing in lattice gauge theories. All reported improvements are measured directly from autocorrelation times and histogram comparisons on the simulated ensembles; no equation or central claim is obtained by fitting a parameter to a subset of the same data and then relabeling the fit as a prediction, nor does any load-bearing step reduce to a self-citation whose content is itself unverified. The extrapolation tests from small to large volumes are presented as empirical checks rather than derivations, leaving the overall chain self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the standard domain assumption that periodic-boundary lattice gauge theories possess distinct topological sectors separated by large action barriers, plus the transferability of Metadynamics bias potentials from other fields. No free parameters or new entities are introduced in the abstract.

axioms (1)

domain assumption Lattice gauge theories with periodic boundary conditions possess topological sectors separated by large action barriers that cause conventional Monte Carlo updates to freeze.
Explicitly stated in the abstract as the source of the problem being solved.

pith-pipeline@v0.9.0 · 5397 in / 1235 out tokens · 30694 ms · 2026-05-13T21:57:04.826323+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

38 extracted references · 38 canonical work pages · 8 internal anchors

[1]

Alles, G

B. Alles, G. Boyd, M. D’Elia, A. Di Giacomo and E. Vicari,Hybrid Monte Carlo and topological modes of full QCD,Phys. Lett. B389(1996) 107 [hep-lat/9607049]. [2]RBC Collaborationcollaboration,Chiral properties of domain wall fermions with improved gauge actions,Nucl. Phys. B Proc. Suppl.106(2002) 721 [hep-lat/0110074]. 8 Enhanced Sampling Techniques for La...

work page arXiv 1996
[2]

DeGrand, A

T.A. DeGrand, A. Hasenfratz and T.G. Kovacs,Improving the chiral properties of lattice fermions,Phys. Rev. D67(2003) 054501 [hep-lat/0211006]. [4]RBC Collaborationcollaboration,Calculation of weak matrix elements in domain wall QCD with the DBW2 gauge action,Nucl. Phys. B Proc. Suppl.119(2003) 362 [hep-lat/0211013]

work page arXiv 2003
[3]

Aoki et al.,The Kaon B-parameter from quenched domain-wall QCD,Phys

Y. Aoki et al.,The Kaon B-parameter from quenched domain-wall QCD,Phys. Rev. D73 (2006) 094507 [hep-lat/0508011]

work page arXiv 2006
[4]

Investigating the critical slowing down of QCD simulations

S. Schaefer, R. Sommer and F. Virotta,Investigating the critical slowing down of QCD simulations,PoSLAT2009(2010) 032 [0910.1465]. [7]ALPHA Collaborationcollaboration,Critical slowing down and error analysis in lattice QCD simulations,Nucl. Phys. B845(2011) 93 [1009.5228]

work page internal anchor Pith review arXiv 2010
[5]

Kanwar and O

G. Kanwar and O. Vega,Spectral Diffusion for Sampling onSU(𝑁),2512.19877

work page arXiv
[6]

Bonanno, A

C. Bonanno, A. Bulgarelli, E. Cellini, A. Nada, D. Panfalone, D. Vadacchino et al.,A scalable flow-based approach to mitigate topological freezing,2601.20708

work page arXiv
[7]

Matsumoto, M

A. Matsumoto, M. Hirasawa, J. Nishimura and A. Yosprakob,Phase diagram of 4D SU(3) Yang-Mills theory at𝜃=𝜋via imaginary theta simulations,2603.09604

work page arXiv
[8]

Singh and L

S. Singh and L. Funcke,Normalizing-flow-based density of states for (1+1)D U(1) lattice gauge theory with a𝜃-term,2603.12501

work page arXiv
[9]

Finkenrath,Review on Algorithms for dynamical fermions,PoSLATTICE2022(2023) 227 [2402.11704]

J. Finkenrath,Review on Algorithms for dynamical fermions,PoSLATTICE2022(2023) 227 [2402.11704]

work page arXiv 2023
[10]

Kanwar,Flow-based sampling for lattice field theories,PoSLATTICE2023(2024) 114 [2401.01297]

G. Kanwar,Flow-based sampling for lattice field theories,PoSLATTICE2023(2024) 114 [2401.01297]

work page arXiv 2024
[11]

Boyle,Advances in algorithms for solvers and gauge generation,PoSLATTICE2023 (2024) 122 [2401.16620]

P.A. Boyle,Advances in algorithms for solvers and gauge generation,PoSLATTICE2023 (2024) 122 [2401.16620]

work page arXiv 2024
[12]

Finkenrath,Future trends in lattice QCD simulations,PoSEuroPLEx2023(2024) 009

J. Finkenrath,Future trends in lattice QCD simulations,PoSEuroPLEx2023(2024) 009

work page 2024
[13]

Trivializing maps, the Wilson flow and the HMC algorithm

M. Luscher,Trivializing maps, the Wilson flow and the HMC algorithm,Commun. Math. Phys.293(2010) 899 [0907.5491]

work page internal anchor Pith review Pith/arXiv arXiv 2010
[14]

Albergo, G

M.S. Albergo, G. Kanwar and P.E. Shanahan,Flow-based generative models for Markov chain Monte Carlo in lattice field theory,Phys. Rev. D100(2019) 034515 [1904.12072]

work page arXiv 2019
[15]

L. Wang, G. Aarts and K. Zhou,Diffusion models as stochastic quantization in lattice field theory,JHEP05(2024) 060 [2309.17082]

work page arXiv 2024
[16]

Hasenbusch,Fighting topological freezing in the two-dimensional𝐶𝑃𝑁−1 model,Phys

M. Hasenbusch,Fighting topological freezing in the two-dimensional𝐶𝑃𝑁−1 model,Phys. Rev. D96(2017) 054504 [1706.04443]

work page arXiv 2017
[17]

Bonanno, G

C. Bonanno, G. Clemente, M. D’Elia, L. Maio and L. Parente,Full QCD with milder topological freezing,JHEP08(2024) 236 [2404.14151]

work page arXiv 2024
[18]

Eichhorn, G

T. Eichhorn, G. Fuwa, C. Hoelbling and L. Varnhorst,Parallel tempered metadynamics: Overcoming potential barriers without surfing or tunneling,Phys. Rev. D109(2024) 114504 [2307.04742]

work page arXiv 2024
[19]

Eichhorn, G

T. Eichhorn, G. Fuwa, C. Hoelbling and L. Varnhorst,Parallel Tempered Metadynamics,PoS LATTICE2024(2025) 61 [2503.09747]

work page arXiv 2025
[20]

Duane, A.D

S. Duane, A.D. Kennedy, B.J. Pendleton and D. Roweth,Hybrid Monte Carlo,Phys. Lett. B 195(1987) 216. 9 Enhanced Sampling Techniques for Lattice Gauge TheoryTimo Eichhorn

work page 1987
[21]

Analytic Smearing of SU(3) Link Variables in Lattice QCD

C. Morningstar and M.J. Peardon,Analytic smearing of SU(3) link variables in lattice QCD, Phys. Rev. D69(2004) 054501 [hep-lat/0311018]

work page internal anchor Pith review Pith/arXiv arXiv 2004
[22]

Instantons from over-improved cooling

M. Garcia Perez, A. Gonzalez-Arroyo, J.R. Snippe and P. van Baal,Instantons from over - improved cooling,Nucl. Phys. B413(1994) 535 [hep-lat/9309009]

work page internal anchor Pith review Pith/arXiv arXiv 1994
[23]

Topology of the SU(2) vacuum: a lattice study using improved cooling

P. de Forcrand, M. Garcia Perez and I.-O. Stamatescu,Topology of the SU(2) vacuum: A Lattice study using improved cooling,Nucl. Phys. B499(1997) 409 [hep-lat/9701012]

work page internal anchor Pith review Pith/arXiv arXiv 1997
[24]

Tanizaki, A

Y. Tanizaki, A. Tomiya and H. Watanabe,Lattice gradient flows (de-)stabilizing topological sectors,JHEP04(2025) 123 [2411.14812]

work page arXiv 2025
[25]

Butti, M

P. Butti, M. Della Morte, B. Jäger, S. Martins and J.T. Tsang,Comparison of smoothening flows for the topological charge in QCD-like theories,Phys. Rev. D112(2025) 014504 [2504.10197]

work page arXiv 2025
[26]

Laio and M

A. Laio and M. Parrinello,Escaping free-energy minima,Proc. Natl. Acad. Sci.99(2002) 12562

work page 2002
[27]

Valsson and M

O. Valsson and M. Parrinello,Variational approach to enhanced sampling and free energy calculations,Phys. Rev. Lett.113(2014) 090601

work page 2014
[28]

A. Laio, G. Martinelli and F. Sanfilippo,Metadynamics surfing on topology barriers: the 𝐶𝑃 𝑁−1 case,JHEP07(2016) 089 [1508.07270]

work page arXiv 2016
[29]

Raiteri, A

P. Raiteri, A. Laio, F.L. Gervasio, C. Micheletti and M. Parrinello,Efficient reconstruction of complex free energy landscapes by multiple walkers metadynamics,J. Phys. Chem. B110 (2006) 3533

work page 2006
[30]

Topological susceptibility and excess kurtosis in SU(3) Yang-Mills theory

S. Durr and G. Fuwa,Topological susceptibility and excess kurtosis in SU(3) Yang-Mills theory,Phys. Rev. D113(2026) 054508 [2501.08217]

work page internal anchor Pith review Pith/arXiv arXiv 2026
[31]

Cost of the Generalised Hybrid Monte Carlo Algorithm for Free Field Theory

A.D. Kennedy and B. Pendleton,Cost of the generalized hybrid Monte Carlo algorithm for free field theory,Nucl. Phys. B607(2001) 456 [hep-lat/0008020]

work page internal anchor Pith review Pith/arXiv arXiv 2001
[32]

Exploring the HMC trajectory-length dependence of autocorrelation times in lattice QCD

H.B. Meyer, H. Simma, R. Sommer, M. Della Morte, O. Witzel and U. Wolff,Exploring the HMC trajectory-length dependence of autocorrelation times in lattice QCD,Comput. Phys. Commun.176(2007) 91 [hep-lat/0606004]

work page internal anchor Pith review Pith/arXiv arXiv 2007
[33]

J.OstmeyerandP.Buividovich,MinimalAutocorrelationinHybridMonteCarlosimulations using Exact Fourier Acceleration,2404.09723

work page arXiv
[34]

Yamamoto, P

S. Yamamoto, P. Boyle, T. Izubuchi, L. Jin, C. Lehner and N. Matsumoto,Improvement in Autocorrelation Times Measured by the Master-Field Technique using Field Transformation HMC in 2+1 Domain Wall Fermion Simulations,PoSLATTICE2024(2025) 034 [2502.05452]. [38]ALPHA Collaborationcollaboration,Monte Carlo errors with less errors,Comput. Phys. Commun.156(200...

work page arXiv 2025
[35]

Hoffman, A

M.D. Hoffman, A. Gelman et al.,The no-u-turn sampler: adaptively setting path lengths in hamiltonian monte carlo.,J. Mach. Learn. Res.15(2014) 1593

work page 2014
[36]

Betancourt,Identifying the optimal integration time in hamiltonian monte carlo, 1601.00225

M. Betancourt,Identifying the optimal integration time in hamiltonian monte carlo, 1601.00225

work page arXiv
[37]

Nishimura and D

A. Nishimura and D. Dunson,Recycling intermediate steps to improve hamiltonian monte carlo,Bayesian Analysis15(2020)

work page 2020
[38]

Vishwanath and H

S. Vishwanath and H. Tak,Repelling-attracting hamiltonian monte carlo,2403.04607. 10

work page arXiv