aiqbench analysis
The paper applies Stochastic Normalizing Flows, a generative machine-learning method, to a lattice gauge theory problem: sampling topology in SU(3) Yang-Mills near the continuum limit. The result is numerical evidence that flow-based sampling keeps autocorrelation times and topological susceptibility uncertainties under control at lattice spacings down to 0.03 fm, with the authors framing cost scaling as the key question. For ML readers, the takeaway is a concrete, scaling-focused test of normalizing flows on a hard sampling problem, not a general claim about AI capability.
Ergodic topology sampling toward the continuum with Stochastic Normalizing Flows
arXiv:2610.20747v1 Announce Type: cross Abstract: In this work we provide numerical evidence that topological freezing in $\mathrm{SU}(3)$ Yang-Mills theory can be mitigated with a flow-based approach, as tested at lattice spacings as fine as 0.03~fm. We do so by analyzing the scaling of the costs of a computational strategy based on the combination of open boundary conditions and Stochastic Normalizing Flows. We show that the integrated autocorrelation time of the squared topological charge and the statistical uncertainty on the topological susceptibility remain under control throughout the explored parameter range with currently available computational resources. This marks an important step toward the use of even more efficient flow-based architectures for ergodic sampling of topology in state-of-the-art lattice gauge theory calculations.
Abstract:In this work we provide numerical evidence that topological freezing in $\mathrm{SU}(3)$ Yang-Mills theory can be mitigated with a flow-based approach, as tested at lattice spacings as fine as 0.03~fm. We do so by analyzing the scaling of the costs of a computational strategy based on the combination of open boundary conditions and Stochastic Normalizing Flows. We show that the integrated autocorrelation time of the squared topological charge and the statistical uncertainty on the topological susceptibility remain under control throughout the explored parameter range with currently available computational resources. This marks an important step toward the use of even more efficient flow-based architectures for ergodic sampling of topology in state-of-the-art lattice gauge theory calculations.
High Energy Physics - Lattice (hep-lat); Statistical Mechanics (cond-mat.stat-mech); Machine Learning (cs.LG); High Energy Physics - Phenomenology (hep-ph)
arXiv:2610.20747 [hep-lat]
(or arXiv:2610.20747v1 [hep-lat] for this version)
Full text under the publisher's open licence · 140 words
Related coverage
GEM Training: How Meta Doubled the Efficiency of Its LLM-Scale Ads Foundation Model
Meta Engineering · Aug 3, 2026
Multilevel Plaquette-Space Sampling for Lattice Gauge Theories with Local Constraint Solves
arXiv · Sep 30, 2026
Neural Autoregressive Control Variates for the Quantum Monte Carlo Sign Problem
arXiv · Jun 5, 2026
Bridging Stochastic Flow Maps and Boltzmann Generators with Normalizing Flows
arXiv · Sep 30, 2026