Monte Carlo simulation of domain growth in the kinetic Ising model on the connection machine

Abstract

A fast multispin algorithm for the Monte Carlo simulation of the two-dimensional spin-exchange kinetic Ising model, previously described by Sullivan and Mountain and used by Amar et al. has been adapted for use on the Connection Machine and applied as a first test in a calculation of domain growth. Features of the code include: (a) the use of demon bits, (b) the simulation of several runs simultaneously to improve the efficiency of the code, (c) the use of virtual processors to simulate easily and efficiently a larger system size, (d) the use of the (NEWS) grid for last communication between neighbouring processors and updating of boundary layers, (e) the implementation of an efficient random number generator much faster than that provided by Thinking Machines Corp., and (f) the use of the LISP function “funcall” to select which processors to update. Overall speed of the code when run on a (128x128) processor machine is about 130 million attempted spin-exchanges per second, about 9 times faster than the comparable code, using hardware vectorised-logic operations and 64-bit multispin coding on the Cyber 205. The same code can be used on a larger machine (65 536 processors) and should produce speeds in excess of 500 million attempted spin-exchanges per second.