The Ising Model
CSIR NET Part C: Ising model by exact enumeration, the 1D transfer matrix, correlation length, mean-field theory, Peierls argument, Onsager's 2D result.
- 17 min
On this page · 16 sections
- Prerequisites
- The Hook: A Model Its Inventor Thought Was a Failure
- The Model
- Small Systems by Hand
- The 1D Chain: Transfer Matrix
- Two Dimensions: Onsager’s Exact Solution
- Mean-Field Theory of the Ising Model
- The Ising Model in Disguise
- Monte Carlo in One Paragraph
- Memory Tricks
- Don’t Confuse
- Common Mistakes
- Practice
- Quick Revision Table
- Key Takeaways
- What’s Next
Prerequisites
- Canonical and Grand Canonical Ensembles: computing \( Z \), \( U = -\partial\ln Z/\partial\beta \) and \( C \) for small discrete systems.
- Linear Algebra: Matrices and Eigenvalues: the transfer-matrix method is \( Z = \text{Tr}\,T^N \) and an eigenvalue problem.
- Diamagnetism, Paramagnetism and Ferromagnetism: the Weiss mean field, which reappears here in Ising form.
The Hook: A Model Its Inventor Thought Was a Failure
The Model
Spins \( s_i = \pm1 \) sit on the sites of a lattice:
$$\boxed{H = -J\sum_{\langle ij\rangle}s_is_j - h\sum_is_i}$$- \( \langle ij\rangle \) means each nearest-neighbour pair counted once.
- \( J > 0 \): ferromagnetic (aligned pairs cost \( -J \), anti-aligned \( +J \)). \( J < 0 \): antiferromagnetic.
- \( h \) is the field times the moment (an energy). \( z \) denotes the coordination number.
The partition function is \( Z = \sum_{\{s\}}e^{-\beta H} \), a sum over \( 2^N \) configurations. Write \( K = \beta J \).
Useful identity: since \( s_is_j = \pm1 \),
$$\boxed{e^{Ks_is_j} = \cosh K\,(1 + s_is_j\tanh K)}$$Small Systems by Hand
Two spins, no field
Configurations: \( \uparrow\uparrow, \downarrow\downarrow \) (energy \( -J \)) and \( \uparrow\downarrow, \downarrow\uparrow \) (energy \( +J \)).
$$Z = 2e^{K} + 2e^{-K} = 4\cosh K, \qquad \langle s_1s_2\rangle = \tanh K, \qquad U = -J\tanh K$$Open chain of N spins, no field
Change variables to bond variables \( \tau_i = s_is_{i+1} \). The first spin is free (2 choices) and each of the \( N - 1 \) bonds is independent:
$$\boxed{Z_{\text{open}} = 2(2\cosh K)^{N-1}}, \qquad U = -(N - 1)J\tanh K$$So an open chain behaves exactly like \( N - 1 \) independent two-level systems: a smooth Schottky-like heat capacity, no singularity.
A ring of three spins: frustration
For a triangle with \( J > 0 \), the two fully aligned states have energy \( -3J \); the other six states have exactly one spin disagreeing with the other two, giving one satisfied and two broken bonds, energy \( +J \):
$$Z_3 = 2e^{3K} + 6e^{-K}$$For \( J < 0 \) (antiferromagnet) the roles reverse: the six states with energy \( -\lvert J\rvert \) are now the ground states. No arrangement can satisfy all three bonds, a situation called frustration. The ground state is six-fold degenerate, so the system keeps a residual entropy \( k_B\ln6 \) at \( T = 0 \).
A ring of four spins
Counting broken bonds (each costs \( 2J \) relative to aligned):
| Configurations | Number | Broken bonds | Energy |
|---|---|---|---|
| all aligned | 2 | 0 | \( -4J \) |
| one spin flipped | 8 | 2 | 0 |
| two adjacent flipped | 4 | 2 | 0 |
| two opposite flipped (alternating) | 2 | 4 | \( +4J \) |
A ring of four is bipartite, so the antiferromagnet has the same \( Z \) (flip every other spin to map \( J \to -J \)). The triangle is not bipartite, which is why it frustrates.
The 1D Chain: Transfer Matrix
For a ring of \( N \) spins (\( s_{N+1} = s_1 \)), write the Boltzmann weight as a product of factors linking neighbours:
$$Z = \sum_{\{s\}}\prod_{i=1}^{N}T_{s_is_{i+1}} = \text{Tr}\,T^N, \qquad T = \begin{pmatrix}e^{K + \beta h} & e^{-K}\\ e^{-K} & e^{K - \beta h}\end{pmatrix}$$with the field split equally between the two spins of each bond. The eigenvalues are
$$\boxed{\lambda_\pm = e^{K}\cosh\beta h \pm \sqrt{e^{2K}\sinh^2\beta h + e^{-2K}}}$$and \( Z = \lambda_+^N + \lambda_-^N \). For large \( N \) only the larger eigenvalue survives:
$$\boxed{f = -k_BT\ln\lambda_+ \ \text{per spin}}$$Results
| Quantity | Exact 1D result |
|---|---|
| Zero-field eigenvalues | \( \lambda_+ = 2\cosh K \), \( \lambda_- = 2\sinh K \) |
| Free energy per spin (\( h = 0 \)) | \( f = -k_BT\ln(2\cosh K) \) |
| Energy per spin | \( u = -J\tanh K \) |
| Heat capacity per spin | \( c = k_BK^2\,\text{sech}^2K \) |
| Magnetization | \( m = \dfrac{\sinh\beta h}{\sqrt{\sinh^2\beta h + e^{-4K}}} \) |
| Zero-field susceptibility per spin | \( \chi = \beta e^{2K} \) (moment set to 1) |
| Correlation function | \( \langle s_is_{i+r}\rangle = (\tanh K)^r = e^{-r/\xi} \) |
| Correlation length | \( \xi = -1/\ln\tanh K \approx \frac{1}{2}e^{2K} \) at low \( T \) |
Check of the ring formula: \( \lambda_+^4 + \lambda_-^4 = 16(\cosh^4K + \sinh^4K) \), which equals \( 2e^{4K} + 12 + 2e^{-4K} \), the enumeration above.
Why: the domain-wall argument
Start from the fully ordered chain and insert one domain wall (a broken bond). Energy cost \( 2J \); the wall can sit at any of about \( N \) bonds, gaining entropy \( k_B\ln N \):
$$\Delta F = 2J - k_BT\ln N < 0 \quad \text{for large } N \text{ at any } T > 0$$Walls proliferate and destroy long-range order. In 2D, a domain of perimeter \( L \) costs \( 2JL \), while the number of such boundaries grows only as roughly \( (z - 1)^L \), so energy wins at low enough \( T \). This Peierls argument proves that the 2D Ising model orders at a finite temperature.
Two Dimensions: Onsager’s Exact Solution
Onsager solved the square-lattice model in zero field exactly. The critical point is given by
$$\boxed{\sinh\frac{2J}{k_BT_c} = 1 \ \Rightarrow\ \frac{k_BT_c}{J} = \frac{2}{\ln(1 + \sqrt2)} \approx 2.269}$$Key features:
- The heat capacity diverges logarithmically, \( C \sim -\ln\lvert T - T_c\rvert \) (\( \alpha = 0 \), but not a mere jump).
- The spontaneous magnetization (later derived exactly) is \( m = [1 - \sinh^{-4}(2J/k_BT)]^{1/8} \) below \( T_c \), so \( \beta = \frac{1}{8} \).
- Other exponents: \( \gamma = \frac{7}{4} \), \( \nu = 1 \), \( \eta = \frac{1}{4} \), \( \delta = 15 \).
- The 2D model in a non-zero field has no known exact solution.
Exact critical points for other 2D lattices (zero field) are also closed-form: triangular (\( z = 6 \)) \( k_BT_c/J = 4/\ln3 \approx 3.641 \); honeycomb (\( z = 3 \)) \( k_BT_c/J = 2/\ln(2 + \sqrt3) \approx 1.519 \). The 3D model has no exact solution; for the simple cubic lattice, numerical work gives \( k_BT_c/J \approx 4.51 \).
Mean-Field Theory of the Ising Model
Replace each neighbour by its average: \( s_is_j \approx m(s_i + s_j) - m^2 \). Each spin then feels an effective field \( zJm + h \):
$$\boxed{m = \tanh[\beta(zJm + h)], \qquad k_BT_c^{\text{MF}} = zJ}$$The free energy per spin is
$$f(m) = \frac{zJ}{2}m^2 - k_BT\ln\{2\cosh[\beta(zJm + h)]\}$$Expanding at \( h = 0 \) gives the Landau form \( f \approx \frac{zJ}{2}\left(1 - \frac{T_c}{T}\right)m^2 + b\,m^4 \) with \( b > 0 \), so all the mean-field exponents follow (\( \beta = \frac{1}{2} \), \( \gamma = 1 \), \( \delta = 3 \), \( \alpha = 0 \)). Above \( T_c \),
$$\chi = \frac{\partial m}{\partial h} = \frac{1}{k_B(T - T_c)}$$the Curie–Weiss law.
How good is mean field?
| Lattice | \( z \) | Mean-field \( k_BT_c/J \) | Exact or best value |
|---|---|---|---|
| 1D chain | 2 | 2 | 0 (no transition) |
| Honeycomb | 3 | 3 | 1.519 |
| Square | 4 | 4 | 2.269 |
| Triangular | 6 | 6 | 3.641 |
| Simple cubic | 6 | 6 | about 4.51 |
Mean-field theory always overestimates \( T_c \), because ignoring fluctuations overstates order. It improves as \( z \) and \( d \) grow, and its exponents become exact at and above \( d = 4 \), the upper critical dimension. The lower critical dimension (below which no order survives) is 1 for the Ising model. For spins with a continuous symmetry (Heisenberg, XY) and short-range interactions, the Mermin–Wagner theorem forbids spontaneous order at \( T > 0 \) already in \( d = 2 \).
The Ising Model in Disguise
graph TD
I["Ising model: s = +1 or -1"] --> A["Uniaxial ferromagnet: s = spin up/down"]
I --> B["Lattice gas: n = (1 + s)/2 = occupied/empty"]
I --> C["Binary alloy: s = atom A or B on a site"]
I --> D["Antiferromagnet on a bipartite lattice: flip one sublattice"]
B --> E["Liquid-gas critical point: Ising universality class"]Lattice gas. Let \( n_i = (1 + s_i)/2 \in \{0, 1\} \) mark occupied sites, with attraction \( -\varepsilon \) between occupied neighbours and chemical potential \( \mu \): \( H_{\text{LG}} = -\varepsilon\sum_{\langle ij\rangle}n_in_j - \mu\sum_in_i \). Substituting gives an Ising model with
$$J = \frac{\varepsilon}{4}, \qquad h = \frac{\mu}{2} + \frac{z\varepsilon}{4}$$plus a constant. Magnetization maps to density, the field to chemical potential, and the Ising critical point to the liquid–gas critical point. That is why fluids and uniaxial magnets share critical exponents.
Binary alloy. In an AB alloy (the standard example is \( \beta \)-brass, CuZn), \( s_i = \pm1 \) labels which atom sits on site \( i \). Order–disorder on two sublattices is an antiferromagnetic Ising transition.
Monte Carlo in One Paragraph
For lattices with no exact solution, one samples configurations with probability \( e^{-\beta E}/Z \) using the Metropolis rule: pick a spin, compute the energy change \( \Delta E \) if it flips, and accept with probability \( \min(1, e^{-\beta\Delta E}) \). This rule obeys detailed balance, so the Boltzmann distribution is its stationary state. The idea of a master equation relaxing to equilibrium is taken up in Non-Equilibrium Processes.
Memory Tricks
Don’t Confuse
| This | vs | That |
|---|---|---|
| 1D Ising: no order at \( T > 0 \) | 2D Ising: order below \( k_BT_c = 2.269J \) | |
| Ising (discrete symmetry) orders in 2D | Heisenberg/XY (continuous) cannot order at \( T > 0 \) in 2D (Mermin–Wagner) | |
| Open chain \( Z = 2(2\cosh K)^{N-1} \) | Ring \( Z = (2\cosh K)^N + (2\sinh K)^N \) | |
| Mean-field \( T_c = zJ/k_B \) | Exact \( T_c \): lower, lattice dependent | |
| \( \langle ij\rangle \) counts each pair once | \( \sum_{i,j} \) over all \( i, j \) counts each pair twice | |
| 2D heat capacity: logarithmic divergence | Mean-field heat capacity: finite jump | |
| Frustration (triangle, \( J < 0 \)): degenerate ground state | Bipartite antiferromagnet: maps exactly onto the ferromagnet at \( h = 0 \) |
Common Mistakes
Practice
All questions below are original practice questions written for this page. They are not from any CSIR NET paper. \( K = J/k_BT \) throughout.
Level 1: Foundation
- What is the critical temperature of the 1D Ising model with nearest-neighbour coupling?
- What is the mean-field critical temperature of an Ising model with coordination number \( z \)?
- What is the exact value of \( k_BT_c/J \) for the square-lattice Ising model (to three significant figures)?
Answers
- \( T_c = 0 \): no transition at any positive temperature.
- \( k_BT_c = zJ \).
- 2.27, from \( \sinh(2J/k_BT_c) = 1 \).
Level 2: Part B Style
Practice Q1. Two Ising spins interact through \( H = -Js_1s_2 \) with \( J > 0 \), at \( k_BT = J \). The correlation \( \langle s_1s_2\rangle \) is closest to:
(a) 0.46 (b) 0.76 (c) 0.88 (d) 1.00
Answer: (b) 0.76
\( \langle s_1s_2\rangle = (2e^K - 2e^{-K})/(2e^K + 2e^{-K}) = \tanh K = \tanh1 = 0.762 \).
Practice Q2. An open chain of three Ising spins has \( H = -J(s_1s_2 + s_2s_3) \), \( h = 0 \). Its partition function and mean energy are:
(a) \( 8\cosh^2K \), \( -2J\tanh K \) (b) \( 8\cosh^3K \), \( -3J\tanh K \) (c) \( 4\cosh^2K \), \( -2J\tanh K \) (d) \( 8\cosh^2K + 8\sinh^2K \), \( -2J\tanh K \)
Answer: (a)
\( Z = 2(2\cosh K)^{2} = 8\cosh^2K \). Two independent bonds, each contributing \( -J\tanh K \). Check: 8 configurations: 2 with energy \( -2J \), 4 with 0, 2 with \( +2J \); \( 2e^{2K} + 4 + 2e^{-2K} = 2(e^K + e^{-K})^2 = 8\cosh^2K \). Option (d) adds a \( \sinh^2K \) term that belongs to no configuration count; the \( \lambda_- \) term appears only for a closed ring.
Practice Q3. Three Ising spins on a triangle interact antiferromagnetically, \( H = \lvert J\rvert(s_1s_2 + s_2s_3 + s_3s_1) \). The ground-state degeneracy and the entropy as \( T \to 0 \) are:
(a) 2, \( k_B\ln2 \) (b) 3, \( k_B\ln3 \) (c) 6, \( k_B\ln6 \) (d) 8, \( k_B\ln8 \)
Answer: (c) 6, k_B ln 6
All-aligned states (2) have energy \( +3\lvert J\rvert \). Every other state (6) has two anti-aligned bonds and one aligned: \( -\lvert J\rvert \), the minimum. Degeneracy 6, \( S(0) = k_B\ln6 \).
Practice Q4. For the square-lattice Ising model, the ratio of the exact critical temperature to the mean-field estimate is closest to:
(a) 0.38 (b) 0.57 (c) 0.76 (d) 1.13
Answer: (b) 0.57
Exact \( 2.269J/k_B \); mean-field \( zJ/k_B = 4J/k_B \). Ratio \( 2.269/4 = 0.567 \). A ratio above 1 is impossible: mean field always overestimates \( T_c \).
Level 3: Part C Style
Practice Q5. An infinite 1D Ising chain in zero field is held at \( k_BT = J/2 \). Consider:
- The correlation length is about 27 lattice spacings.
- \( \langle s_0s_{10}\rangle \approx 0.69 \).
- The spontaneous magnetization is non-zero at this temperature.
- The zero-field susceptibility per spin, in units of \( 1/J \) with unit moment, is about 109.
Which are correct?
(a) 1 and 2 only (b) 1, 2 and 4 only (c) 2 and 3 only (d) 1, 2, 3 and 4
Answer: (b) 1, 2 and 4 only
\( K = 2 \), \( \tanh2 = 0.9640 \).
- \( \xi = -1/\ln0.9640 = 27.3 \) (compare \( \frac{1}{2}e^{4} = 27.3 \)). True.
- \( 0.9640^{10} = 0.693 \). True.
- No spontaneous order in 1D at \( T > 0 \). False.
- \( \chi = \beta e^{2K} = (2/J)e^{4} = 109.2/J \). True.
Practice Q6. Four Ising spins form a ring, \( H = -J(s_1s_2 + s_2s_3 + s_3s_4 + s_4s_1) \), \( J > 0 \), \( h = 0 \), at \( k_BT = J \). Consider:
- Twelve of the sixteen configurations have energy zero.
- \( Z = (2\cosh1)^4 + (2\sinh1)^4 \approx 121.2 \).
- The probability of finding the ring in one of its two ground states is about 0.90.
- The mean energy equals \( -4J\tanh1 \approx -3.05J \).
Which are correct?
(a) 1 and 2 only (b) 1, 2 and 3 only (c) 2, 3 and 4 only (d) 1, 2, 3 and 4
Answer: (b) 1, 2 and 3 only
- Energies: \( -4J \) (2 states), 0 (8 + 4 = 12 states), \( +4J \) (2 states). True.
- \( Z = 2e^4 + 12 + 2e^{-4} = 121.23 \), and \( 16(\cosh^41 + \sinh^41) \) gives the same. True.
- \( 2e^4/121.23 = 0.901 \). True.
- \( U = -\partial\ln Z/\partial\beta = -(8Je^{4} - 8Je^{-4})/Z = -3.60J \). The formula \( -NJ\tanh K \) ignores the \( \lambda_-^N \) term, which matters for \( N = 4 \). False.
Practice Q7. Consider mean-field theory for an Ising ferromagnet with coordination number \( z \) and moment set to 1. Consider:
- It predicts \( k_BT_c = zJ \).
- Above \( T_c \) it predicts \( \chi = 1/k_B(T - T_c) \).
- Applied to a 1D chain it predicts a finite-temperature transition, which the exact solution does not have.
- For the square lattice it predicts a \( T_c \) lower than Onsager’s exact value.
Which are correct?
(a) 1 and 2 only (b) 1, 2 and 3 only (c) 1, 3 and 4 only (d) 1, 2, 3 and 4
Answer: (b) 1, 2 and 3 only
- Linearising \( m = \tanh(\beta zJm) \) gives the condition \( \beta zJ = 1 \). True.
- Linearising with a field: \( m = \beta(zJm + h) \Rightarrow m = h/k_B(T - T_c) \). True.
- Mean field gives \( k_BT_c = 2J \) for \( z = 2 \); exact is zero. True.
- Mean field gives \( 4J \), higher than \( 2.269J \). False.
Quick Revision Table
| Idea | Formula or fact |
|---|---|
| Hamiltonian | \( H = -J\sum_{\langle ij\rangle}s_is_j - h\sum s_i \) |
| Two spins | \( Z = 4\cosh K \), \( \langle s_1s_2\rangle = \tanh K \) |
| Open chain | \( Z = 2(2\cosh K)^{N-1} \) |
| Ring | \( Z = \lambda_+^N + \lambda_-^N \); \( h = 0 \): \( 2\cosh K \), \( 2\sinh K \) |
| Triangle | \( Z = 2e^{3K} + 6e^{-K} \); antiferro ground degeneracy 6 |
| Square ring of 4 | \( Z = 2e^{4K} + 12 + 2e^{-4K} \) |
| 1D free energy | \( f = -k_BT\ln(2\cosh K) \) |
| 1D correlations | \( (\tanh K)^r \); \( \xi \approx \frac{1}{2}e^{2K} \) |
| 1D susceptibility | \( \chi = \beta e^{2K} \) |
| 1D heat capacity | \( k_BK^2\text{sech}^2K \), no singularity |
| Onsager | \( k_BT_c/J = 2/\ln(1 + \sqrt2) = 2.269 \); \( \beta = \frac{1}{8} \) |
| Mean field | \( m = \tanh[\beta(zJm + h)] \), \( k_BT_c = zJ \) |
| Lattice gas | \( n = (1 + s)/2 \), \( J = \varepsilon/4 \) |
| Metropolis | accept with \( \min(1, e^{-\beta\Delta E}) \) |
Key Takeaways
What’s Next
- Bose–Einstein Condensation: a transition driven by quantum statistics rather than interactions.
- First- and Second-Order Phase Transitions: exponents, scaling relations and universality behind the numbers here.
- Non-Equilibrium Processes: master equations and detailed balance behind Monte Carlo.
- Small-system partition functions in chemistry: Model System Calculations.