Strategies for improved sampling in lattice field theory

在数值格点场论中获得可靠的结果需要对路径积分进行有效的采样。这种努力经常受到阻碍，随着晶格变得越来越大，越来越精细，受到临界减速和拓扑冻结等效应的影响。我正在研究的技术从一个新的角度来处理生成代表性配置的挑战;使用生成神经网络的耦合层。它代表了一个有前途的采样策略，特别是当与多级算法相结合时，可以缓解上述突出的长期存在的问题。

Obtaining reliable results in numerical lattice field theory requires efficient sampling of the path integral. Such efforts are often hampered, with increasing severity as lattices are made larger and finer, by effects such as critical slowing down and topological freezing. The technique that I am working on approaches the challenge of generating representative configurations from a novel angle; using coupled layers of generative neural networks. It represents a promising sampling strategy which, particularly when combined with multilevel algorithms, may alleviate the longstanding issues highlighted above.