A Neural Network Model of an Ising Spin Glass
K.F. Wilson and D.J. Goossens
Department of Physics, Monash University, Clayton 3168, Australia
Full Text PDF
The behaviour of an Ising spin glass (S=1/2) with infinite range interactions is modelled using a numerical simulation based on a neural network. Thermodynamic variables are defined on the network, and are found to obey the Thouless-Anderson-Palmer theory when the applied magnetic field is zero. When a magnetic field is applied along the spin direction, complex field-dependent behaviour appears, including a state in which the Edwards-Anderson order parameter is independent of temperature below the critical temperature.
DOI: 10.12693/APhysPolA.97.983
PACS numbers: 75.10.-b, 75.10.Nr, 84.35.+i