Domain Walls and Interfacial Excitations

Domain-wall thermodynamics is the working language of an enormous range of modern physics. In materials science the surface tension sets the Wulff shape of equilibrium crystals and the nucleation barriers that control everything from raindrop formation to the crystallization of alloys and pharmaceuticals; classical nucleation theory, $\Delta F^{*} \propto \sigma^{d}/(\Delta f)^{d-1}$, is exactly the droplet balance derived above. After a temperature quench, the curvature-driven motion of domain walls produces the universal coarsening law $\ell(t) \sim t^{1/2}$, seen in everything from magnetic thin films to phase-separating polymer blends and observed directly in cold-atom and liquid-crystal experiments. The wall is also a probe. In Monte Carlo simulation the standard, high-accuracy way to measure surface tension is to detect the exponentially small free-energy dip between the two magnetization peaks of a finite box, $\sim e^{-2\sigma L^{d-1}/k_B T}$ — precisely the finite-size splitting derived above. And the roughening transition of interfaces, a Kosterlitz–Thouless transition of the solid-on-solid model, connects lattice interfacial physics to the topological phase transitions of two-dimensional systems, from superfluid films to the crystal facets that engineers care about when growing semiconductor surfaces atom by atom.

Picture a long bar of magnet, half its spins up and half down. The seam between them is the domain wall. Every bond the seam cuts flips from satisfied ($-J$) to broken ($+J$), costing $2J$, so the total energetic price is proportional to the area of the seam — a length in 2D, a sheet in 3D, but just a single point in 1D. This geometric scaling of cost with dimension is the entire story of whether order survives. Now let the wall wiggle. A wandering wall samples many shapes, and shapes are entropy. In 1D the wall is a free point that roams the whole length — unbounded entropy, guaranteed disorder. In 2D the wall is a fluctuating line whose transverse excursions are capillary waves; the entropy it gains lowers its effective tension but cannot overcome the linear-in-$L$ energy until $T$ reaches $T_c$, where the tension finally vanishes.

Read the figure as the fight, refereed by dimension. In one dimension (blue) the wall's free energy $F = 2J - k_B T\ln L$ starts positive but crosses zero and plunges to $-\infty$: for any $T > 0$ a large enough chain always favors making a wall, so walls proliferate and the order dies — the 1D Ising model has $T_c = 0$. In two dimensions (green) the cheapest wall costs $\sigma L$, growing with the system rather than sinking, so below $T_c$ it stays positive and order is stable. The single sign of $\partial F/\partial(\text{size})$ decides the phase.

Think of surface tension as the free-energy price per unit area of an interface, exactly analogous to a liquid's. It is what makes droplets round (minimizing area), sets nucleation barriers ($\sim \sigma R^{d-1}$ for a droplet of radius $R$), and controls the rate of domain coarsening after a quench. The same $\sigma$ governs finite-size scaling: a box of size $L$ at coexistence has a free-energy splitting between its two phases of order $e^{-2\sigma L^{d-1}/k_B T}$ — the cost of the pair of walls needed to tunnel between them. There is also a phase transition of the wall itself. Below the roughening temperature in $d = 3$ an interface is locally flat (faceted); above it, thermal capillary waves make it macroscopically rough. In $d = 2$ the interface is rough at all $T > 0$. Roughening is a genuine transition — a Kosterlitz–Thouless transition for the solid-on-solid model — a reminder that a domain wall is a thermodynamic system in its own right.

The first quantity is the energy cost of a wall, which is pure bond-counting. Each bond the wall cuts flips from satisfied to broken, and the total is proportional to the wall's area:

where: $J$ is the Ising coupling (energy per aligned bond); $A_{\rm wall}$ is the number of broken bonds, i.e. the wall area in lattice units; and $2J$ is the energy to flip one bond from $-J$ to $+J$. Physically, the wall is a sheet of unsatisfied bonds, and its energy is set by how many it contains — a length in 2D, an area in 3D. In one dimension the wall is a single point of fixed energy but free position, and its free energy is a competition of that energy against configurational entropy:

where: $L$ is the chain length; $L-1$ the number of bonds where the wall can sit; $k_B$ the Boltzmann constant. For any $T > 0$, the logarithm eventually dominates and $F \to -\infty$ as $L \to \infty$ — walls always appear, destroying order. Setting $F = 0$ gives the crossover length $L^* = 1 + e^{2J/k_B T}$, below which order is metastable and above which it is gone. The general definition of surface tension isolates the interfacial free energy by comparing an ensemble forced to contain a wall with one that is not:

where: $F_s$ is the excess free energy of the interface; $Z_{\rm a}$ the partition function with antiperiodic (wall-forcing) boundary conditions; $Z_{\rm p}$ that with periodic boundaries (no forced wall); $L^{d-1}$ the interface's cross-sectional area. The bulk free energies cancel exactly in the ratio, leaving only the cost of the seam — the clean, unambiguous definition used in exact solutions and simulations alike. For the 2D square lattice this can be evaluated in closed form, and here the standard textbook derivation hides a subtle factor-of-two trap that we correct. The exact axis-direction result is a single logarithm:

where: $K$ is the dimensionless coupling; $K^{*}$ its Kramers–Wannier dual, satisfying $\tanh K^{*} = e^{-2K}$; and $K_c = \tfrac12\ln(1+\sqrt2) \approx 0.4407$ is the self-dual critical coupling where $\sigma = 0$. Reading it physically: the surface tension is twice the gap between the direct coupling and its dual, positive in the ordered phase ($K > K^{*}$), vanishing exactly at the self-dual point, and approaching $2K$ (the bare $2J$ per lattice spacing) as $T \to 0$. At low energies a long, smooth interface behaves like a tensioned membrane; its undulations are capillary waves with an energy quadratic in the local slope:

where: $h(x)$ is the local height (transverse displacement) of the interface; $\sigma$ the surface tension (stiffness against area increase); $\nabla h$ the local slope. The energy is $\sigma$ times the excess area $\tfrac12(\nabla h)^2$ of a gently sloped surface, so each Fourier mode obeys equipartition $\langle |h_q|^2\rangle = k_B T/(\sigma q^2)$ — soft, long-wavelength undulations dominate. Integrating those mode fluctuations gives the interface an intrinsic width whose growth with system size depends sharply on dimension:

where: $w^2$ is the mean-square roughness; $q$ the in-plane wavevector, cut off at $q_{\min}\sim 1/L$ and $q_{\max}\sim 1/a$; $a$ the lattice spacing. In $d = 2$ (a 1D interface) the integral diverges as $\ln(L/a)$ — the interface is always rough; in $d = 3$ it diverges logarithmically above the roughening temperature and is finite below; for $d > 3$ it converges and interfaces are smooth. The surface tension also sets the nucleation barrier for a droplet of a competing phase, balancing surface cost against bulk gain:

where: $R$ is the droplet radius; $\Delta f$ the bulk free-energy density favoring the new phase; $c_d, V_d$ are $d$-ball surface/volume geometric factors. Maximizing over $R$ gives the critical radius $R^{*} = (d-1)c_d\sigma/(d\,V_d\Delta f)$ and a barrier $\Delta F^{*} \propto \sigma^{d}/(\Delta f)^{d-1}$ that controls the exponential nucleation rate $\Gamma \sim e^{-\Delta F^{*}/k_B T}$. Finally, the same $\sigma$ quantifies the exponentially small symmetry breaking of a finite box, where tunneling between the two ordered states requires a pair of walls:

where: $\Delta F_{\pm}$ is the free-energy splitting between the two ordered states in a finite periodic box, and $2\sigma L^{d-1}$ is the cost of the two interfaces of a slab configuration. The splitting vanishes only as $L \to \infty$ — which is how spontaneous symmetry breaking emerges in the thermodynamic limit.

The fate of domain walls and the form of $\sigma$ depend sharply on dimension and on proximity to $T_c$:

1 · Where the 1D chain loses its order (confidence builder)

In one dimension a single domain wall has free energy $F = 2J - k_B T\ln(L-1)$. For $J = 1$ (in units of $k_B$) at temperature $T = 0.5$, find the crossover length $L^{*}$ where $F = 0$, and confirm $F < 0$ for a longer chain and $F > 0$ for a shorter one.

Step 1. Set $F = 0$: $2J = k_B T\ln(L^{*}-1)$, so $\ln(L^{*}-1) = 2J/(k_B T) = 2(1)/0.5 = 4$.

Step 2. Solve: $L^{*} - 1 = e^{4} = 54.598$, so $L^{*} = \boxed{55.6}$. For any chain longer than this the wall is thermodynamically favored.

Step 3. Check a longer chain, $L = 100$: $F = 2 - 0.5\ln(99) = 2 - 0.5(4.595) = 2 - 2.298 = \boxed{-0.30 < 0}$. The wall lowers the free energy — order is gone.

Step 4. Check a shorter chain, $L = 10$: $F = 2 - 0.5\ln(9) = 2 - 0.5(2.197) = 2 - 1.099 = \boxed{+0.90 > 0}$. Here the wall costs free energy — order is (metastably) intact. The lesson is stark: in 1D there is always a crossover length beyond which walls win, so the thermodynamic limit ($L \to \infty$) always disorders. This is why the 1D Ising model has $T_c = 0$, no matter how strong the coupling.

2 · The exact 2D surface tension and its linear vanishing (exam level)

Using the corrected Onsager form $\sigma/k_B T = 2K + \ln\tanh K$, evaluate the reduced surface tension at $K = 0.6$ (below $T_c$), confirm it vanishes at $K_c = \tfrac12\ln(1+\sqrt2)$, and show it vanishes linearly in $t = (T - T_c)/T_c$.

Step 1. At $K = 0.6$: $\tanh(0.6) = 0.53705$, $\ln(0.53705) = -0.62175$, so $\sigma/k_B T = 2(0.6) + (-0.62175) = 1.2 - 0.622 = \boxed{0.578 > 0}$. Positive, as an ordered phase demands.

Step 2. Contrast the wrong (double-log) form: $2K + 2\ln\tanh K = 1.2 - 1.244 = -0.044 < 0$ — an unphysical negative surface tension. This is the factor-of-two trap; the single-log form is correct.

Step 3. At the critical point $K_c = \tfrac12\ln(1+\sqrt2) = 0.44069$: $\tanh(0.44069) = 0.41421 = \sqrt2 - 1$, and $2K_c + \ln(\sqrt2-1) = 0.88137 + \ln(0.41421) = 0.88137 - 0.88137 = \boxed{0}$. The tension vanishes exactly at $K_c$, since $\ln(\sqrt2-1) = -2K_c$.

Step 4. Linear vanishing: expand near $K_c$. The derivative $d(\sigma/k_B T)/dK = 2 + \operatorname{csch}(K)\operatorname{sech}(K) = 2 + 2/\sinh(2K)$; at $K_c$ where $\sinh 2K_c = 1$ this is $2 + 2 = 4$, finite and nonzero. So $\sigma/k_B T \approx 4(K - K_c)$ near the transition.

Step 5. Since $K - K_c \propto (T_c - T)/T \propto |t|$ to leading order, $\sigma \propto |t|^{1}$: the exponent is $\mu = 1$, matching $\mu = (d-1)\nu = 1\cdot 1 = 1$ in 2D. The surface tension of the 2D Ising interface vanishes linearly as the critical point is approached — a direct measurement of the correlation-length exponent $\nu = 1$.

References. Pathria & Beale, Statistical Mechanics, 3rd ed., Ch. 12–13 (Ising model, phase transitions) · Kardar, Statistical Physics of Fields, Ch. 2–3 (interfaces, surface tension, capillary waves) · Reif, Fundamentals of Statistical and Thermal Physics, Ch. 10 · Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity, Ch. 11 · Landau & Lifshitz, Statistical Physics Part 1, §146–148 · D. B. Abraham & M. E. Fisher, works on the 2D Ising interfacial tension (exact $\sigma(T)$).

Provenance. All equations are standard textbook results. A verifier should recompute: (1) bond energy $2J$ and the 1D free energy $2J - k_B T\ln(L-1)$ with crossover $L^{*} = 1 + e^{2J/k_B T}$; (2) the Peierls contour bound with $3^L$ counting and convergence threshold $k_B T < 2J/\ln 3$; (3) the 1D correlation length $\xi = 1/|\ln\tanh\beta J| \to \tfrac12 e^{2\beta J}$ from the transfer matrix; (4) the corrected exact 2D surface tension $\sigma/k_B T = 2K + \ln\tanh K = 2(K - K^{*})$ with $K^{*} = \operatorname{atanh}(e^{-2K})$ and the self-dual $K_c = \tfrac12\ln(1+\sqrt2) \approx 0.4407$ where it vanishes linearly; (5) the exponent $\mu = (d-1)\nu$ with $\nu = 1$ (2D). Base-error note: the AllQFT source wrote the surface tension as $\sigma/k_B T = 2K + 2\ln\tanh K$ (using $K^{*} = -\ln\tanh K$), a factor-of-two slip that yields a negative surface tension just below $T_c$; the corrected single-log form is used throughout. The droplet barrier $\Delta F^{*} \propto \sigma^{d}/(\Delta f)^{d-1}$ is classical nucleation theory. All numbers are reproduced in `verify_aq-sm-domain-walls-interfacial-excitations.py`.

Prerequisites. The Ising Model: Definition and Hamiltonian · Order Parameter and Spontaneous Magnetization · Peierls Argument for Long-Range Order · One-Dimensional Ising Model and Transfer Matrix Method. Leads to. Onsager Solution of the Two-Dimensional Ising Model · Kramers-Wannier Duality · High- and Low-Temperature Series Expansions · Lattice Gas Model and Ising Mapping.