Preview

PDF, English Download (4MB) | Terms of use

Abstract

Progress in answering some of the most interesting open questions about the nature of reality is currently stalled by hard computational barriers. Research into artificial intelligence may provide solutions to these challenges. In this thesis, the applicability of modern machine learning algorithms to three longstanding problems in lattice quantum chromodynamics is investigated. First, normalizing flow architectures for the generative neural sampling of lattice field theories with dynamical fermions are developed and demonstrated to solve topological freezing in the Schwinger model at criticality. Flows are then applied to the density-of-states approach to complex action problems, showing that the Lee-Yang zeroes of the partition function of a scalar field theory with an imaginary external field can be successfully located. Finally, the problem of extracting real-time physics from imaginary-time data via spectral reconstruction is approached from the perspective of probabilistic inverse theory with Gaussian processes to compute ghost and gluon spectral functions. Future research directions are outlined for the application of the present work to state-of-the-art lattice and phenomenological calculations.