Replication Algorithms

Abstract

In Monte Carlo simulations as in other computational methods one typically needs to run an algorithm with several different parameters. Consider the example of the two-dimensional Ising model. To determine the order parameter m as a function of the lattice size L and temperature T, i.e., m(T, L), many simulations need to be carried out. Often the lattice sizes we are interested in are fairly small, and so it does not make sense to spread them over many processors. The computational requirement of a single run with respect to storage and time is small enough to allow each individual simulation to be run on a single processor. In this case it is most efficient to spread the entire program over many processors. This farm concept for parallelization is illustrated in Fig. 6.1.