Skip to content
Tutorials Center
My learning

Topics

Lesson 28 of 39 · Phase Transitions and the Ising Model

The Ising Model in One Dimension

Define the Ising model, solve the one-dimensional chain exactly with the transfer-matrix method, check it against brute-force enumeration, and see why there is no phase transition.

The Ising model is the hydrogen atom of phase transitions: simple enough to solve in some cases, rich enough to show magnetism, critical points and universality. Each site of a lattice carries a spin s_i = +1 or -1, and neighboring spins prefer to align. In this lesson you define the model, solve the one-dimensional chain exactly with the transfer-matrix method, and discover that it has no phase transition at any positive temperature.

The model

For a lattice of N sites with a coupling J between neighbors and an external field h, the energy of a spin configuration is

Text
E({s}) = -J Σ_<ij> s_i s_j  -  h Σ_i s_i

The sum Σ_<ij> runs over nearest-neighbor pairs. With J > 0 aligned neighbors have lower energy (ferromagnet). The field h favors s = +1. The model is a deliberate cartoon: it keeps only the up/down degree of freedom and a short-range interaction. The canonical partition function is a sum over all 2^N configurations:

Text
Z = Σ_{s} exp( -β E({s}) ),      β = 1/(kT)

Everything we want follows from Z: the free energy per spin f = -(kT/N) ln Z, the magnetization m = ⟨s⟩ = -∂f/∂h, and correlations such as ⟨s_i s_j⟩.

Exact solution of the periodic chain

Arrange the spins in a ring of N sites with s_{N+1} = s_1. Write the Boltzmann weight as a product of local factors and define the 2×2 transfer matrix

Text
T(s, s') = exp( βJ s s' + βh (s + s')/2 )

with rows and columns indexed by s, s' ∈ {+1, -1}. Then Z = Tr(T^N) = λ₊^N + λ₋^N, where λ± are the eigenvalues of T:

Text
λ± = e^(βJ) cosh(βh) ± √( e^(2βJ) sinh²(βh) + e^(-2βJ) )

For large N the larger eigenvalue dominates, so the free energy per spin is f = -kT ln λ₊. This is analytic in T for all T > 0 (the two eigenvalues never become degenerate), so there is no phase transition at finite temperature in one dimension. The reason is intuitive: a single "domain wall" between up and down regions costs energy 2J but gains entropy k ln N (it can be anywhere), so for large N walls always proliferate and destroy long-range order.

At h = 0: λ₊ = 2 cosh βJ and λ₋ = 2 sinh βJ, so Z = (2 cosh βJ)^N + (2 sinh βJ)^N.

Check against brute force

Enumerating 2^N states for a small ring tests the formula. Here N = 12 (4096 states), for which the λ₋^N term is still visible.

Python Transfer matrix versus brute-force enumeration

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

All ratios equal 1 to numerical precision.

Magnetization and correlations

Differentiating ln λ₊ with respect to βh gives the magnetization of the infinite chain:

Text
m(h, T) = sinh(βh) / √( sinh²(βh) + e^(-4βJ) )

As h → 0, m → 0 for every T > 0: no spontaneous magnetization. Only at T = 0 (β → ∞) does m jump to ±1 with any field.

At zero field, the correlation between spins a distance r apart is

Text
⟨s_i s_{i+r}⟩ = tanh^r(βJ) = exp(-r/ξ),     ξ = -1 / ln tanh(βJ)

The correlation length ξ grows rapidly as T → 0 (roughly like e^(2J/kT)/2) but stays finite for all T > 0. This is a one-dimensional hint of critical behavior: a diverging correlation length is the signature of a phase transition, which in 1D appears only at T = 0.

Python Magnetization curves and correlation length in 1D

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

At low temperature the curve m(h) becomes a very steep step; the approximation ξ ≈ e^(2J/kT)/2 agrees closely once kT is well below J.

What 1D teaches us

  • Short-range interactions in 1D cannot sustain order at T > 0: fluctuations always win.
  • The transfer-matrix method is general: it solves any 1D chain with finite-range interactions, and is the starting point for the exact 2D solution.
  • In 2D and above the picture changes completely, which is the subject of the next lessons.

Check your understanding

Quick check · 4 questions

of 4 answered You scored / 4 · perfect!

  1. 1. What does the transfer-matrix method give for the partition function of a periodic chain?

    The trace of T^N equals λ₊^N + λ₋^N.
  2. 2. Why is there no finite-temperature phase transition in the 1D Ising model?

    The largest eigenvalue never becomes degenerate with the other, so ln λ₊ is smooth for all T > 0.
  3. 3. At zero field, the spin-spin correlation at distance r is

    It decays exponentially with correlation length ξ = -1/ln tanh(βJ).
  4. 4. As T approaches 0 in 1D, the correlation length

    ξ grows like e^(2J/kT)/2; the transition occurs only at T = 0.

Summary

  • The Ising model has spins ±1 with energy E = -J Σ s_i s_j - h Σ s_i.
  • The periodic 1D chain has Z = λ₊^N + λ₋^N with λ± = e^(βJ)cosh βh ± √(e^(2βJ) sinh² βh + e^(-2βJ)).
  • m(h) = sinh βh / √(sinh² βh + e^(-4βJ)); correlations decay as tanh^r(βJ).
  • There is no phase transition at T > 0 in 1D.

Next: Mean-Field Theory of the Ising Model