A spin model is a probability model assigning a state, called a spin, to each site and weighting configurations by their interactions. For
a graph 
, a state space 
, a single-site reference probability 
, and pair log-weights 
, a finite-volume form is
 |
The constant 
normalizes the Gibbs measure.
External fields give additional single-site terms, and multispin models permit interactions
involving more than two sites. Binary spins give the Ising
model, spherical unit-vector spins give the O(n)
model, and random couplings give spin-glass models.
Different state spaces and interaction laws specify different models.
