Machine learning renormalisation group transformations for QCD
Video
Presenters
Poster

Details
While renormalisation-group transformations provide a framework for improved discretisations that remove of significantly suppress these artifacts, they are notoriously difficult to implement for practical simulations. In this talk I show how gauge-equivariant convolutional neural networks and machine learning can bridge the gap between the theoretical concept and the practical application for Quantum Chromodynamics (QCD), the SU(3) gauge theory of the strong nuclear force. I demonstrate that this approach achieves a "classically perfect" discretisation free of tree-level lattice artifacts to all orders, enabling the extraction of continuum physics with high precision even on very coarse lattices. Our results on gradient-flow scales, topology and the deconfinement phase transition establish machine learning as a transformative tool for overcoming computational barriers in lattice gauge theory.

