The dynamics of the spin autocorrelation function and the relaxation of the magnetization in the Griffiths phase of the two-dimensional bond-diluted Ising model at the percolation threshold are studied using Monte Carlo techniques. The results resolve a previous ambiguity about the decay law.
The authors examine the dynamical properties of the 2D+or-J Edwards-Anderson Ising spin glass in the anomalous phase which corresponds to the Griffiths phase in the diluted ferromagnetic system. By using Monte Carlo simulation they find that the autocorrelation function can be described as a function of a power of scaled time. This new scaling includes both the stretched exponential law for the intermediate stage and the logarithmic law for the early stage of relaxation. The relaxation time obtained from the scaling is found to diverge exponentially with exponent proportional to T -2 .
Relaxation phenomena are studied for the ± J Ising spin glass in two dimensions by Monte Carlo method. For each spin, a representative relaxation time of the spin is estimated from the equilibrium autocorrelation function of the spin. Clusters of slowly relaxing spins are found to exist below the critical temperature of the nonrandom Ising ferromagnet. It is also found that the distribution of the relaxation times of spins is consistent with the prediction of the theory which assumes independent motion of clusters. This distribution explains the nonexponential decay of the averaged spin autocorrelation function as in the case of the diluted Ising ferromagnets.
The critical dynamics of the diluted two-dimensional Ising model below the percolation threshold is studied. The dynamics exhibits a crossover from a concentration-dependent behaviour for xi T >> xi p to a percolating lattice type behaviour for xi T << xi p , where xi T and xi p are the thermal and percolation correlation lengths, respectively. As a consequence, the dynamical exponents A and B are shown to be the limits of dynamical functions A(x) and B(x), respectively; here, x=ln xi p /ln xi T . Extensive Monte Carlo simulations support our predictions.