General algorithm for improved lattice actions on parallel computing architectures

Abstract

Quantum field theories underlie all of our understanding of the fundamental forces of nature. There are relatively few first-principles approaches to the study of quantum field theories (such as quantum chromodynamics [QCD] relevant to the strong interaction) apart from the perturbative (i.e., weak-coupling) regime. Currently, the most commonly used method is the Monte Carlo method on a hypercubic space–time lattice. These methods consume enormous computing power for large lattices, and it is essential that increasingly efficient algorithms be developed to perform standard tasks in these lattice calculations. Here we present a general algorithm for QCD that allows one to put any planar improved gluonic lattice action onto a parallel computing architecture. High performance masks for specific actions (including nonplanar actions) are also presented. These algorithms have been successfully employed by us in a variety of lattice QCD calculations using improved lattice actions on a 128 node Thinking Machines CM-5.