Title: Markov Chain Monte Carlo without Detailed Balance and Bounce-free Worm Algorithm
Author: Hidemaro Suwa
Affiliation: Department of Applied Physics, the University of Tokyo, Japan
Abstract:
In the Markov chain Monte Carlo method, the detailed balance is usually
imposed as a sufficient condition for the balance condition. If the
Markov sequence goes beyond the detailed balance, however, rejection
rates can get minimized. Although it has been considered to be difficult
to generally construct a transition matrix beyond the detailed balance,
we have invented a new Monte Carlo algorithm that surely makes it
possible with minimized rejection rates. As a benchmark, we have
confirmed that our algorithm significantly boosts up the relaxation
speed in 4-state Potts model to nearly 7 times more than the heat bath
algorithm and 30 times more than the Metropolis algorithm. In the same
manner, we have also developed a bounce-free worm algorithm in the
quantum Monte Carlo method. In one dimensional Heisenberg model with
magnetic field, it is confirmed that the relaxation by the bounce-free
worms gets about 50 times faster than by the conventional method called
the generalized directed-loop.