Flow-based approaches for Monte Carlo acceleration and variance reduction

by Alessandro Nada (Universita' degli studi di Torino)

Monday 6 Jul 2026, 14:30 → 16:30 Europe/Rome

U2-04 (Building U2 - Quantum)

Flow-based methods represent a promising approach to accelerating Monte Carlo calculations in lattice field theory. The first part of this talk reviews the formulation of Stochastic Normalizing Flows within the framework of non-equilibrium Monte Carlo simulations, in particular as a strategy to address critical slowing down. We outline a method to mitigate topological freezing in SU(3) pure gauge theory and present preliminary results at fine lattice spacings. The second part instead focuses on a novel variance reduction technique: we introduce a flow-based framework that recasts the notorious signal-to-noise ratio problem in terms of infinitesimal flows, together with some early results in $\phi^4$ scalar field theory.

Alessandro_Nada.pdf