Group convolutional networks with built-in lattice symmetries outperform fully connected networks for learned contour deformations in small Hubbard-model sign-problem simulations, but transfer learning across parameters is only sometimes useful.
Fermions at Finite Density in (2+1)d with Sign-Optimized Manifolds
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
abstract
We present Monte Carlo calculations of the thermodynamics of the (2+1) dimensional Thirring model at finite density. We bypass the sign problem by deforming the domain of integration of the path integral into complex space in such a way as to maximize the average sign within a parameterized family of manifolds. We present results for lattice sizes up to $10^3$ and we find that at high densities and/or temperatures the chiral condensate is abruptly reduced.
citation-role summary
baseline 1
citation-polarity summary
fields
cond-mat.dis-nn 1years
2025 1verdicts
CONDITIONAL 1roles
baseline 1polarities
unclear 1representative citing papers
citing papers explorer
-
Exploring Group Convolutional Networks for Sign Problem Mitigation via Contour Deformation
Group convolutional networks with built-in lattice symmetries outperform fully connected networks for learned contour deformations in small Hubbard-model sign-problem simulations, but transfer learning across parameters is only sometimes useful.