Ising Model · aiwiki.page

The Ising model is a mathematical model in statistical mechanics consisting of interacting variables that each take one of two values, usually $+1$ and $-1$ . Originally developed to investigate ferromagnetism, it demonstrates how simple local interactions can produce large-scale order and phase transitions. Its standard form places these variables on a lattice, although generalizations allow other interaction networks. Despite its simplicity, it captures important features of collective behavior in physical systems. (damtp.cam.ac.uk)

Historical development

Wilhelm Lenz proposed the model in 1920. His student Ernst Ising solved its one-dimensional version in his 1924 doctoral thesis and published the result in 1925. Ising found that the nearest-neighbor chain does not exhibit ferromagnetic ordering at any positive temperature. This result applies to the one-dimensional short-range model, not to every possible arrangement of interacting two-state variables. (arxiv.org)

Rudolf Peierls demonstrated in 1936 that the two-dimensional model can support order at sufficiently low temperature. In 1944, Lars Onsager obtained the exact partition function for the two-dimensional square-lattice model without an external field. This established a precisely solvable example of a system with an order–disorder transition and singular thermodynamic behavior. (arxiv.org)

Mathematical definition

For $N$ sites, a configuration is a collection $s=(s_1,\ldots,s_N)$ , where $s_i=\pm1$ . A common energy function is

H(s)=-\sum_{\langle i,j\rangle}J_{ij}s_is_j-\sum_i h_i s_i.

The first sum counts each interacting pair once; in the standard lattice model, these are nearest neighbors. The coupling $J_{ij}$ specifies interaction strength. With this sign convention, positive couplings favor equal neighboring spins, while negative couplings favor opposite spins. The parameter $h_i$ is a local field measured in energy units; for a magnetic interpretation, it incorporates the magnetic moment and applied magnetic field. (weizmann.ac.il)

The variables are conventionally called “spins,” but the classical model does not require the full machinery of quantum spin. They are ordinary discrete variables, rather than noncommuting operators or quantum superpositions. For uniform interactions and field, one writes $J_{ij}=J$ and $h_i=h$ . (damtp.cam.ac.uk)

At thermal equilibrium, configurations follow the Boltzmann distribution:

P(s)=\frac{e^{-\beta H(s)}}{Z}, \qquad Z=\sum_s e^{-\beta H(s)}, \qquad \beta=\frac{1}{k_{\mathrm B}T}.

Here $k_{\mathrm B}$ is the Boltzmann constant, and $Z$ is the partition function, summed over all $2^N$ configurations. The Helmholtz free energy is $F=-k_{\mathrm B}T\ln Z$ ; thermodynamic observables can be obtained from its derivatives or from equilibrium averages. (damtp.cam.ac.uk)

Ordering and dimensionality

For ferromagnetic couplings, low energy favors aligned spins, whereas thermal fluctuations favor disorder. The magnetization per site,

m=\frac{1}{N}\sum_i\langle s_i\rangle,

serves as an order parameter. At zero field, the energy is unchanged when every spin reverses sign. An ordered equilibrium phase selects one of two opposite magnetizations, illustrating spontaneous symmetry breaking. (damtp.cam.ac.uk)

For a finite system at zero field, an equilibrium average that samples both orientations gives zero magnetization. Spontaneous magnetization is therefore defined using the thermodynamic limit, with system size taken to infinity before a symmetry-selecting field is removed. Finite lattices display rounded changes rather than true thermodynamic singularities. (damtp.cam.ac.uk)

Dimensionality changes the outcome. The uniform nearest-neighbor chain has no positive-temperature transition, while the square-lattice ferromagnet has

k_{\mathrm B}T_c=\frac{2J}{\ln(1+\sqrt2)} \approx2.269185J.

This expression assumes equal horizontal and vertical couplings and zero external field. Three-dimensional models also support a transition, but their properties are generally investigated through approximations and numerical calculations rather than an Onsager-type exact solution. (damtp.cam.ac.uk)

Critical behavior and universality

Near the critical point, fluctuations become correlated over increasingly large distances. The correlation length diverges in an infinite system, and observables exhibit scaling described by critical exponents. For the two-dimensional square-lattice model, spontaneous magnetization vanishes as $(T_c-T)^{1/8}$ , susceptibility diverges with exponent $7/4$ , and heat capacity has a logarithmic singularity. (damtp.cam.ac.uk)

These features help define the Ising universality class: microscopically different systems can share critical behavior when their relevant dimensionality, symmetries, and interaction characteristics agree. The liquid–gas critical point provides an important physical example. Renormalization-group methods explain this connection by examining how descriptions change when short-distance details are averaged out. (damtp.cam.ac.uk)

Computational methods

Direct enumeration becomes impractical as the number of configurations grows exponentially. Markov chain Monte Carlo methods instead generate representative equilibrium samples. In a single-spin Metropolis update, a proposed flip changing the energy by $\Delta E$ is accepted with probability

A=\min\{1,e^{-\beta\Delta E}\}.

With appropriate proposals, this rule satisfies detailed balance for the equilibrium distribution. (arxiv.org)

Near criticality, local updates can suffer from long correlation times, termed critical slowing down. Cluster methods, including Wolff’s algorithm, update groups of spins collectively and can greatly improve sampling efficiency. Simulation results must account for equilibration and correlations between successive measurements. (journals.aps.org)

Related models and applications

Replacing spins by occupation variables $n_i=(s_i+1)/2$ produces a lattice-gas description, connecting magnetic ordering with density changes and fluid phase transitions. In mathematical optimization, the same substitution relates Ising energy minimization to quadratic unconstrained binary optimization, allowing binary decision problems to be represented through pairwise couplings and biases. (journals.aps.org)

A distinct extension, the transverse-field Ising model, replaces classical variables with quantum operators and introduces a field that does not commute with the interaction term. It supports quantum phase transitions driven by changing the field relative to the coupling, including transitions in one-dimensional chains at zero temperature. These quantum transitions differ from the temperature-driven ordering of the classical model. (arxiv.org)