Speaker
Description
Score-based diffusion models are applied to two-dimensional SU(2) lattice pure gauge theory with the Wilson action, using a quaternion parameterization of the group manifold. Trained at a single coupling on an 8×8 lattice, the model generates configurations at different couplings via physics-conditioned sampling and generalizes to larger lattice sizes through a fully convolutional U-Net with periodic boundary conditions, reproducing exact analytical predictions with biases |Δ| ≤ 0.001 near the training point. The extension to four-dimensional non-Abelian gauge theories and the prospects for applying the diffusion model framework to the reconstruction of transverse-momentum-dependent parton distribution functions (TMD PDFs) and fragmentation functions (FFs) are discussed.