The Potts Model

Ask the Ising model a slightly harder question and you get the Potts model. Ising spins choose between two states, up or down, related by a single flip. But most orderings in nature offer more than two equivalent options with no natural "distance" between them: an atom adsorbed on graphite can sit in one of three equivalent registries, a crystal can order into several symmetry-related sublattices, a graph can be colored with $q$ colors. The $q$ -state Potts model captures exactly this: a spin $\sigma_i$ takes one of $q$ labels, and neighbors are rewarded — through a bare Kronecker delta $\delta_{\sigma_i,\sigma_j}$ — only for being identical, never for being "close." That one change from $\mathbb{Z}_2$ to the full permutation symmetry $S_q$ transforms the physics. The central question the model answers is beautiful and sharp: how does the number of competing ordered phases decide whether a transition is smooth or abrupt? With only two phases (Ising, $q=2$ ) the transition is continuous. Add more degenerate ordered states and, past a threshold, the system prefers to jump discontinuously into one of them — a first-order transition with latent heat. In two dimensions this threshold is exactly $q_c=4$ : continuous for $q\le 4$ , first-order for $q\ge 5$ . Two exact structures make the model a jewel. First, in $d=2$ it is self-dual under Kramers–Wannier duality, which pins the critical temperature exactly at $e^{K_c}=1+\sqrt q$ . Second, the Fortuin–Kasteleyn random-cluster representation rewrites the partition function as a sum over bond graphs weighted by $q^{(\text{clusters})}$ , analytically continuing $q$ to non-integer values and linking Potts criticality to percolation ( $q=1$ ) and spanning trees ( $q\to 0$ ). Far from a toy, the Potts model describes adsorbed monolayers, grain growth, foam coarsening, and graph coloring.

Key takeaways

The $q$ -state Potts model places a label $\sigma_i\in\{1,\dots,q\}$ on each site and couples neighbors through $-J\,\delta_{\sigma_i,\sigma_j}$ : energy drops only when neighbors share a state, with no metric between states.
Its symmetry is the full permutation group $S_q$ , far richer than the $\mathbb{Z}_2$ of Ising; for $q=2$ it reduces to the Ising model up to a constant and a factor-of-2 coupling rescale.
In two dimensions the transition is continuous for $q\le 4$ and first-order (latent heat) for $q\ge 5$ ; the exact critical coupling $e^{K_c}=1+\sqrt q$ follows from Kramers–Wannier self-duality.
The symmetry-allowed cubic invariant in the Landau free energy is what drives the transition first-order once the number of phases is large enough.
The Fortuin–Kasteleyn random-cluster form maps the partition function onto correlated bond percolation and extends $q$ to real values.
Key idea: replacing the Ising $\mathbb{Z}_2$ symmetry by the permutation symmetry $S_q$ of $q$ degenerate states turns the number of competing ordered phases into a control parameter for the order of the transition — a cubic invariant drives it first-order once $q$ exceeds a dimension-dependent threshold ( $q_c=4$ in 2D).

The core insight

Replacing the Ising $\mathbb{Z}_2$ symmetry by the permutation symmetry $S_q$ of $q$ degenerate states turns the number of competing ordered phases into a control parameter for the order of the transition: a symmetry-allowed cubic invariant drives the transition first order once $q$ exceeds a dimension-dependent threshold ( $q_c=4$ in 2D).

Where this lives today

The Potts model is a working tool, not a museum piece. Its large- $q$ kinetic version — the "cellular Potts" or Glazier–Graner model — is a standard simulation engine for the coarsening of polycrystalline grains, soap froths, and even biological tissues, where each cell or grain is one of many ( $q\gg1$ ) states and curvature-driven boundary motion emerges from Monte Carlo dynamics. Adsorbed monolayers realize small- $q$ Potts criticality directly: krypton on graphite is a 3-state Potts system (three equivalent registries of the $\sqrt3\times\sqrt3$ structure), and its continuous order–disorder transition matches the predicted 3-state exponents. At the theory frontier, the Fortuin–Kasteleyn map keeps the Potts model central to probability and combinatorics: the $q\to1$ limit is percolation, the antiferromagnetic $T\to\infty$ limit counts graph colorings (the chromatic polynomial), and the complex- $q$ zeros of the partition function connect Lee–Yang theory to hard questions about when a graph can be colored. Conformal field theory and the Coulomb gas give the exact 2D exponents for every $q\le4$ , and Schramm–Loewner evolution describes the fractal cluster boundaries — a modern synthesis in which the humble "same-or-different" rule anchors some of the deepest results in two-dimensional statistical mechanics.

The physical picture

Imagine $q$ political parties and a lattice of voters. Each voter is happier the more neighbors share their party, but all parties are interchangeable — there is no "left–right" axis, only same-or-different. At high temperature opinions are random ( $S_q$ -symmetric, disordered). As you cool, one party comes to dominate. Here is the twist: with many parties competing, the system can find it cheaper to snap into a polarized state all at once — a first-order jump — rather than tilt gradually. The more parties $q$ , the stronger this discontinuous tendency. Contrast this with an Ising magnet, which distinguishes "up" from "down" along a line — a metric. The Potts spin distinguishes only "same" from "different," a graph with no geometry. That is why the natural order parameter is not a simple scalar but a vector recording the population imbalance among the $q$ states, and why the symmetry is the full permutation group, not $\mathbb{Z}_2$ . A second, powerful way to see the model is the random-cluster picture. Each ferromagnetic bond is either "occupied" — gluing its two endpoints into one cluster — or not, and every independent cluster contributes a factor $q$ (its number of color choices). Large $q$ rewards having many small clusters at high $T$ and one giant cluster at low $T$ ; the competition between cluster entropy and bond energy produces the abrupt switch we read thermodynamically as latent heat.

Figure

The Potts spin: same-or-different, not near-or-far — q states with no metric

The punchline is the phase diagram in the plane of state-number $q$ and dimension. The exact 2D critical temperature slides down smoothly as $q$ grows, but the order of the transition changes discontinuously at a knife-edge: everything at $q\le 4$ is continuous, and the instant $q$ reaches $5$ the latent heat jumps to a nonzero value. That vertical line at $q_c=4$ — separating "smooth" from "abrupt" — is the thesis of the whole subject drawn as a single boundary.

Figure

Exact 2D critical coupling kᵦTc/J = 1/ln(1+√q) vs q, with the qc = 4 order boundary

The core equations

The Potts Hamiltonian expresses the "same-or-different" rule directly. Neighbors lower their energy by $J$ only when their labels coincide; an optional field $h$ favors one distinguished state. There is no term measuring "how far apart" two labels are, which is precisely what gives the model its full permutation symmetry.

\mathcal{H}=-J\sum_{\langle ij\rangle}\delta_{\sigma_i,\sigma_j}-h\sum_i\delta_{\sigma_i,1} \qquad (1)

where: $\sigma_i$ — state label at site $i$ , an integer in $\{1,\dots,q\}$ ; $\delta_{\sigma_i,\sigma_j}$ — Kronecker delta, $1$ if neighbors share a state, $0$ otherwise; $J>0$ — ferromagnetic coupling favoring agreement; $\langle ij\rangle$ — sum over nearest-neighbor pairs; $h$ — field favoring state $1$ ; $q$ — number of states per site. Because only matches count, every relabeling permutation of the $q$ states that acts uniformly leaves $\mathcal{H}$ invariant: the symmetry group is $S_q$ . The partition function sums the Boltzmann weight over all $q^N$ configurations. Writing $K=\beta J$ for the dimensionless coupling makes the whole thermodynamics a function of that single combination in zero field.

Z=\sum_{\{\sigma\}}\exp\!\Big(K\sum_{\langle ij\rangle}\delta_{\sigma_i,\sigma_j}\Big),\qquad K=\beta J \qquad (2)

where: $\beta=1/(k_BT)$ ; $\sum_{\{\sigma\}}$ — sum over all $q^N$ spin configurations on $N$ sites. The exponential of a sum of $0/1$ deltas is what makes the Fortuin–Kasteleyn expansion below exact rather than approximate. The self-dual point of Kramers–Wannier duality pins the exact 2D critical coupling. This is a remarkable statement of symmetry over spectrum: without diagonalizing anything, duality alone locates $T_c$ .

e^{K_c}=1+\sqrt{q}\qquad (d=2,\ \text{square lattice}) \qquad (3)

where: $K_c=\beta_c J$ — critical coupling. Read physically, the critical temperature $k_BT_c/J=1/\ln(1+\sqrt q)$ slides steadily downward as $q$ grows: more competing states makes ordering harder. At $q=2$ this gives the Ising value in Potts normalization, and its rescale to standard Ising units recovers $\tfrac12\ln(1+\sqrt2)$ — the check performed in the derivation. The Fortuin–Kasteleyn (random-cluster) representation is the model's second exact face. It converts the sum over spins into a sum over bond graphs, each cluster carrying a factor of $q$ .

Z=\sum_{G\subseteq E}v^{\,b(G)}\,q^{\,c(G)},\qquad v=e^{K}-1 \qquad (4)

where: $G$ — subset of occupied bonds of the edge set $E$ ; $b(G)$ — number of occupied bonds; $c(G)$ — number of connected clusters (isolated sites count); $v=e^{K}-1$ — bond fugacity. Because $Z$ is now a polynomial in $q$ , the parameter can be continued to any real value: $q\to 1$ is bond percolation, $q\to 0$ is spanning trees. Landau theory shows why the transition turns first-order. For $q>2$ the coarse-grained free energy acquires a symmetry-allowed cubic term — forbidden by $\mathbb{Z}_2$ in Ising — and a cubic term generically forces a discontinuous jump.

f_{\text{MF}}(m)=f_0+\tfrac12\,a\,m^2-\tfrac13\,w\,m^3+\tfrac14\,u\,m^4,\qquad w\neq 0\ \ (q>2) \qquad (5)

where: $m$ — scalar order parameter (population excess of the favored state); $a\propto(T-T_0)$ ; $w$ — cubic coefficient, nonzero for $q>2$ ; $u>0$ — quartic stabilizer. The mere presence of the cubic invariant is the structural reason Potts differs from Ising: no fine-tuning is needed to make the transition first order in mean field.

The physical picture in equations

Derivation AKramers-Wannier self-duality and the exact Tc

We want the exact critical coupling of the 2D Potts model without solving the model — locating $T_c$ purely from a symmetry that maps high temperature to low temperature. The strategy is to start from the random-cluster form (4), use the Euler relation for planar graphs to rewrite the sum over bond subsets as a sum over dual-lattice bonds, and read off a duality relation $v\,v^*=q$ between the fugacity and its dual. The square lattice is self-dual, so criticality (assuming a single transition) must sit at the fixed point $v=v^*$ . Watch that "assuming a single transition" — it is an assumption, correct here but not automatic.

Step 1. Start from the random-cluster partition function, $Z(v)=\sum_G v^{b(G)}\,q^{c(G)}$ , summing over bond subsets $G$ of the lattice. Why: this form is exact — it came from expanding $e^{K\delta}=1+v\delta$ bond by bond and tracing over spins, giving one factor $q$ per monochromatic cluster.

Step 2. Apply the planar Euler relation to $G$ : with $N$ sites, $b(G)$ occupied bonds and $\ell(G)$ independent loops, the number of clusters is

c(G)=N-b(G)+\ell(G).

Why: this is Euler's formula for the graph $G$ embedded in the plane; loops are what reconnect the graph after bonds are added.

Step 3. On the dual lattice, occupied bonds of $G$ and the loops exchange roles with dual bonds: $\ell(G)=b(G^*)$ where $G^*$ is the complementary dual graph. Substituting the Euler relation reorganizes the sum into a Potts partition function on the dual with fugacity $v^*=q/v$ . Why: adding a bond on the direct lattice cuts a bond on the dual; loops of $G$ correspond exactly to occupied dual bonds.

Step 4. The reorganized sum has the same form as the original with $v\to v^*$ , giving the duality relation

v\,v^{*}=q.

Why: the square lattice is its own dual, so $Z(v)$ and $Z(v^*)$ are the same function up to a smooth prefactor that cannot move a singularity.

Step 5. Set $v=v^*$ (the self-dual point). Then $v_c^2=q$ , so $v_c=\sqrt q$ and

e^{K_c}-1=\sqrt q\ \Longrightarrow\ e^{K_c}=1+\sqrt q.

Why: a single phase transition can only sit where the duality map has its fixed point; anywhere else it would predict two singularities related by duality.

Step 6. Check $q=2$ . The Potts coupling relates to Ising by $\delta_{\sigma\sigma'}=\tfrac12(1+s s')$ , so $K_{\text{Potts}}=2K_{\text{Ising}}$ . Then $e^{2K_{\text{Ising}}^c}=1+\sqrt2$ gives $2K_{\text{Ising}}^c=\ln(1+\sqrt2)$ , i.e. $K_{\text{Ising}}^c=\tfrac12\ln(1+\sqrt2)=0.4407$ — exactly the Onsager–Kramers–Wannier Ising value. Why: the factor-of-2 rescale is the whole content of "Potts $q=2$ is Ising"; it must reproduce the known Ising $T_c$ , and it does. *Verified: verify_aq-sm-potts-model.py reproduces $e^{K_c}=1+\sqrt q$ , the Ising rescale $K_{\text{Ising}}^c=\tfrac12\ln(1+\sqrt2)=0.44069$ , and $k_BT_c/J=1/\ln(1+\sqrt q)$ for $q=2,3,4$ .*

Go deeper · Derivation B · 2 minMean-field theory: why q > 2 is first order

Derivation BMean-field theory: why q > 2 is first order

We want to show that mean-field (Bragg–Williams) theory predicts a finite jump in the order parameter — hence latent heat — for every $q>2$ , and that this jump smoothly vanishes as $q\to 2$ , recovering the continuous Ising limit. The strategy is a single-site variational free energy in which state $1$ is occupied with probability $x$ and the other $q-1$ states share the rest equally; we then find where the ordered and disordered minima have equal free energy (coexistence) and measure the discontinuity between them. The key structural fact to watch is a cubic term $\propto(q-2)$ that appears when we expand near the symmetric point.

Step 1. Ansatz: single-site probabilities $p_1=x$ and $p_a=(1-x)/(q-1)$ for $a\neq 1$ . The symmetric (disordered) point is $x=1/q$ ; ordering means $x>1/q$ . Why: this is the most general uniform single-site distribution consistent with the residual $S_{q-1}$ symmetry of the non-favored states.

Step 2. The mean-field free energy per site combines a bond energy and a mixing entropy,

f(x)=-\tfrac12 Jz\Big[x^2+\tfrac{(1-x)^2}{q-1}\Big]+k_BT\Big[x\ln x+(1-x)\ln\tfrac{1-x}{q-1}\Big],

with $z$ the coordination number. Why: the energy is $-\tfrac12 Jz$ times the probability that two independent neighbors match; the entropy is the single-site mixing entropy.

Step 3. Stationarity $df/dx=0$ gives the self-consistency equation

\ln\frac{x(q-1)}{1-x}=Kz\,\frac{qx-1}{q-1},

which $x=1/q$ always solves. Why: the disordered state is always a stationary point; the question is whether a distinct ordered root appears continuously or discontinuously.

Step 4. Expand near $x=1/q$ . The cubic coefficient of $f$ about the symmetric point is proportional to $(q-2)$ : it vanishes only at $q=2$ (pure quartic, pitchfork, second order) and is nonzero for $q>2$ (a first-order branch). Why: a nonzero cubic invariant is the Landau signature of a first-order transition — this is the microscopic origin of eq. (5)'s $w\neq 0$ .

Step 5. Impose coexistence $f(x_*)=f(1/q)$ with a common tangent. The ordered root and transition coupling are

x_*=\frac{q-1}{q},\qquad K_c z=\frac{2(q-1)}{q-2}\ln(q-1).

Why: a first-order transition happens where the two competing minima are degenerate, not where one becomes unstable.

Step 6. The order-parameter jump is $\Delta x=x_*-1/q=(q-2)/q$ , finite for every $q>2$ and $\to 0$ as $q\to 2^+$ . Why: latent heat is proportional to this discontinuity; its vanishing at $q=2$ is exactly the statement that Ising is continuous. *Verified: verify_aq-sm-potts-model.py confirms $x_*=(q-1)/q$ , $\Delta x=(q-2)/q$ (e.g. $1/3$ at $q=3$ , $1/2$ at $q=4$ , $3/5$ at $q=5$ ), $K_c z=2(q-1)\ln(q-1)/(q-2)$ , and $\Delta x\to0$ as $q\to2$ .*

Limiting cases

The special values of $q$ are a map of the model's connections to the rest of statistical mechanics.

Case	Result	Meaning
$q=2$ (Ising)	$-J\delta_{\sigma_i\sigma_j}=-\tfrac{J}{2}(s_i s_j+1)$ , $s_i=\pm1$	Reduces to Ising with $J_{\text{Ising}}=J/2$ plus a constant; continuous transition in $d\ge2$ .
$q=1$	$Z\to Z_{\text{percolation}}$	Fortuin–Kasteleyn limit: correlated bond percolation; derivatives of $Z$ at $q=1$ give percolation observables.
$q\to 0$	$Z\sim\sum_{\text{spanning trees}}$	Uniform spanning trees / resistor networks (with appropriate scaling of $v$ ).
$q\to\infty$	strongly first order, $\Delta E\to J$	Latent heat saturates; the large- $q$ kinetic Potts model models grain growth and foam coarsening.
2D threshold	$q_c=4$	Continuous for $q\le4$ , first order for $q\ge5$ ; $q=4$ is marginal with logarithmic corrections.

Worked examples

Example 1 — Exact critical temperature of the 3-state Potts model in 2D.

Step 1. For $q=3$ the self-dual point gives $e^{K_c}=1+\sqrt3$ . Numerically $1+\sqrt3=2.7321$ .

Step 2. Take the logarithm: $K_c=\beta_c J=\ln(2.7321)=1.0051$ .

Step 3. Invert for the reduced temperature: $k_BT_c/J=1/K_c=1/1.0051=0.9950$ .

Step 4. Since $q=3<4$ , the transition is continuous (second order), with known 3-state Potts exponents in 2D ( $\nu=5/6$ , $\alpha=1/3$ ). The exact 2D critical temperature of the 3-state Potts model is $k_BT_c/J=1/\ln(1+\sqrt3)\approx0.9950$ , a continuous transition.

Example 2 — The mean-field order-parameter jump for the 5-state model.

Step 1. Mean-field theory predicts a first-order transition for all $q>2$ , with the favored-state probability jumping from the disordered value $1/q$ to the ordered value $x_*=(q-1)/q$ .

Step 2. For $q=5$ : disordered $x=1/5=0.20$ , ordered $x_*=4/5=0.80$ .

Step 3. The jump is $\Delta x=x_*-1/q=(q-2)/q=3/5=0.60$ .

Step 4. Compare $q=3$ : $\Delta x=1/3\approx0.333$ , and $q\to2^+$ : $\Delta x\to0$ . The discontinuity — and hence the latent heat — grows with the number of competing phases, exactly as the political-parties picture suggested. The mean-field order-parameter discontinuity for the 5-state Potts model is $\Delta x=(q-2)/q=3/5=0.60$ , versus $1/3$ for $q=3$ and $0$ for the Ising limit $q=2$ .

Origins · 1 minPotts model and how this was discovered

Potts model

introduced 1952 · The q-state generalization of the Ising model, proposed in Renfrey B. Potts's 1952 Cambridge PhD work under Cyril Domb. Domb had suggested the problem; Potts worked out the 2D self-dual critical points. Ashkin and Teller had studied a related four-state model in 1943. The name "Potts model" stuck for the permutation-symmetric version, which became a central testbed for how symmetry controls the order of phase transitions and, via Fortuin–Kasteleyn (1972), a bridge to percolation and graph theory.

History

1943

— Ashkin and Teller study a four-component generalization of the Ising model

1952

— Renfrey Potts (PhD under Cyril Domb) introduces the q-state model and its 2D self-dual critical points

1954

— Domb and collaborators develop series expansions distinguishing continuous from first-order Potts transitions

1972

— Fortuin and Kasteleyn establish the random-cluster representation, extending q to real values

1982

— F. Y. Wu's Reviews of Modern Physics survey consolidates the field, including the qc = 4 threshold in 2D

Going further

References. Wu, F. Y., "The Potts model," Rev. Mod. Phys. 54, 235 (1982) · Pathria & Beale, Statistical Mechanics, 3rd ed., Ch. 13 · Kardar, Statistical Physics of Fields, Ch. 8 · Baxter, Exactly Solved Models in Statistical Mechanics, Ch. 12 · Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity.

Provenance. Hamiltonian, $S_q$ symmetry, and the $q=2\to$ Ising reduction are standard (Wu 1982; Pathria & Beale Ch. 13). The Fortuin–Kasteleyn expansion $e^{K\delta}=1+v\delta$ and the cluster trace giving $q^{c(G)}$ , the Kramers–Wannier relation $vv^*=q$ via the planar Euler formula yielding $e^{K_c}=1+\sqrt q$ (Baxter Ch. 12), and the Bragg–Williams coexistence condition giving $K_c z=2(q-1)\ln(q-1)/(q-2)$ with jump $\Delta x=(q-2)/q$ were all recomputed in verify_aq-sm-potts-model.py. The 2D threshold $q_c=4$ (continuous $q\le4$ , first order $q\ge5$ ) and exponents (e.g. $q=3$ : $\nu=5/6$ , $\alpha=1/3$ ) are established results (Baxter; Wu 1982).

Builds on: The Ising Model: Definition and Hamiltonian · Order Parameter and Spontaneous Magnetization · Mean-Field (Weiss) Theory of the Ising Model · Kramers-Wannier Duality. Leads to: High- and Low-Temperature Series Expansions · Domain Walls and Interfacial Excitations · Bethe-Peierls Approximation · Lattice Gas Model and Ising Mapping.

If the Potts and Ising models are "the same" at q=2, why do their critical temperatures look different?

They are the same physics, but written in different coupling units. The Potts energy $-J\delta_{\sigma_i\sigma_j}$ equals the Ising energy $-\tfrac{J}{2}s_i s_j$ plus a constant, so $K_{\text{Potts}}=2K_{\text{Ising}}$ . Plug $q=2$ into $e^{K_c^{\text{Potts}}}=1+\sqrt2$ : you get $K_c^{\text{Potts}}=\ln(1+\sqrt2)$ , which halves to $K_c^{\text{Ising}}=\tfrac12\ln(1+\sqrt2)\approx0.4407$ — the familiar Onsager value. Different numbers, identical physics; the factor of two is pure bookkeeping.

Why does adding more states make the transition abrupt rather than just shifting Tc?

The number of ordered phases is a genuinely new degree of freedom, not a rescaling. With $q$ degenerate ordered states, the coarse-grained free energy is allowed a cubic term (the trace of a tensor order parameter), which $\mathbb{Z}_2$ Ising symmetry forbids. A cubic term makes the ordered minimum appear at a finite distance from the disordered one rather than growing continuously from it, so the order parameter jumps — a first-order transition with latent heat. More states means a larger cubic coefficient and a bigger jump.

What does q=1 have to do with percolation — there is only one state?

The random-cluster form $Z=\sum_G v^{b(G)}q^{c(G)}$ is a polynomial in $q$ , so we can analytically continue to $q=1$ even though a "1-state spin" is trivial. At $q=1$ the cluster factor is $1$ and $Z=\sum_G v^{b(G)}=\prod_{\text{bonds}}(1+v)$ — pure independent bond percolation with occupation probability $p=v/(1+v)$ . Percolation quantities (cluster-size distribution, percolation probability) come out as derivatives of the Potts free energy at $q=1$ . So percolation is the $q\to1$ shadow of the Potts model.

Is the 2D threshold qc = 4 an exact result or an approximation?

It is exact. In two dimensions, Baxter's solution of the random-cluster model shows the transition is continuous for $q\le4$ and first order for $q\ge5$ , with $q=4$ marginal (logarithmic corrections). This is not a mean-field statement — mean-field theory wrongly predicts first order for every $q>2$ . The threshold $q_c=4$ is special to $d=2$ ; the mean-field and higher-dimensional pictures differ, which is exactly why the exact 2D result is celebrated.

How is the "order parameter" defined when there is no up/down axis?

Because the states are unordered, the order parameter cannot be a single number like Ising magnetization. Instead it is a set of population imbalances: how much more probable state $1$ is than the average $1/q$ , and so on for a basis of the $(q-1)$ -dimensional space orthogonal to the uniform vector. In mean-field practice we track one scalar $x=\langle\delta_{\sigma,1}\rangle$ , the probability of the favored state; $x=1/q$ is disordered and $x>1/q$ is ordered. The full order parameter transforms as a $(q-1)$ -component representation of $S_q$ .

Why can large-q Potts models describe grain growth in metals?

A polycrystal is a patchwork of grains, each with a different crystallographic orientation and no energetic preference among orientations — exactly $q\gg1$ equivalent Potts states. Grain boundaries are the domain walls between unlike states, and at large $q$ the transition is strongly first order, giving sharp boundaries with a well-defined surface tension. Simulating Monte Carlo dynamics of the large- $q$ Potts model reproduces curvature-driven boundary motion and the statistical laws of grain coarsening, which is why metallurgists and foam physicists use it as a standard model.