A new parallel algorithm for simulation of spin-glasses in external fields

Spin glasses are prototypical models for disordered systems which provide a rich source for investigations of a number of important and difficult applied problems of physics, chemistry, material science, biology, evolution, organization dynamics, hard-optimization, environmental and social structures, human logic systems, financial mathematics etc. Numerical studies of spin-glass systems are difficult to accomplish and in general only small moderate system sizes can be accessed. An effective algorithm for parallel simulation of spin-glass system is developed. In contrast to well known algorithms of Metropolis and others which are based on the Monte Carlo simulations method, the developed algorithm allows with high efficacy to construct stable spin-chains of arbitrary length in parallel and calculate all statistical parameters of spin-glass system of large sizes. We have implemented software using cluster computation (MPI technology) and GPU technology (CUDA programming language is used) as well. Since the ideology of GPU technology is SIMD (Single Instruction Multiple Data) and our implemented algorithm is from that class of problems, we obtained fully parallel implementation. We have tested the developed code on example of simulation 1D spin-glass in external fields and we were convinced of reliability and efficiency of calculations.