Efficiency and acceptance rates in the application of the Metropolis algorithm to lattice gauge theories

We study the efficiency of trial distributions in the Metropolis algorithm test for the SU(2) Wilson action. The acceptance rate of trial matrices in the Metropolis test and the correlations of results are measured for distributions peaked around the identity. Away from the strong coupling region, we observe a frame where the acceptance rate decreases exponentially with the distance of the trial matrices from the identity, as well as a correlation minimum just at the beginning of this frame. Suggestions for optimizing application of the Metropolis algorithm are also presented. © 1988.