The XY Model and Continuous Spins

Take the Ising model and let each spin rotate freely in a plane instead of choosing up or down. That single change — from a discrete arrow to a continuous angle $\theta_i\in[0,2\pi)$ , coupled as $-J\cos(\theta_i-\theta_j)$ — replaces the discrete $\mathbb{Z}_2$ symmetry with the continuous $O(2)\cong U(1)$ , and the physics transforms. Continuous symmetry means gapless Goldstone modes: long-wavelength twists of the spins that cost almost no energy. In low dimensions these soft spin waves accumulate so violently that they destroy ordinary long-range order altogether. In three dimensions the XY model still orders conventionally, defining the 3D $O(2)$ universality class that governs the superfluid $\lambda$ -transition of liquid $^4\mathrm{He}$ . In two dimensions, though, the Mermin–Wagner theorem forbids any spontaneous magnetization at $T>0$ — yet the system is not trivially disordered. At low temperature, spin waves alone produce algebraically decaying correlations: a "quasi-long-range-ordered" critical phase with a temperature-dependent exponent, ordered enough to be interesting but never truly ordered. What rescues a genuine phase transition from this marginal situation is topology. Because the order parameter lives on a circle, configurations carry an integer winding number, and point defects with nonzero winding are vortices. At low temperature vortices appear only as tightly bound neutral pairs; at the Berezinskii–Kosterlitz–Thouless (BKT) temperature these pairs unbind into a free plasma, shredding the quasi-order. The BKT transition is of infinite order, with an essential singularity in the correlation length, and is one of the very few exactly understood unconventional transitions — the physics of superfluid films, thin superconductors, Josephson-junction arrays, and two-dimensional melting.

Key takeaways

The XY model places planar unit spins $\mathbf{s}_i=(\cos\theta_i,\sin\theta_i)$ on a lattice with energy $-J\sum_{\langle ij\rangle}\cos(\theta_i-\theta_j)$ , giving a continuous $O(2)$ symmetry.
The Mermin–Wagner theorem forbids spontaneous breaking of this continuous symmetry in $d\le 2$ at any $T>0$ : no conventional long-range order in two dimensions.
Yet the 2D model has a finite-temperature Berezinskii–Kosterlitz–Thouless transition separating a low- $T$ quasi-long-range-ordered phase (power-law correlations) from a high- $T$ exponentially disordered phase.
The transition is driven by the unbinding of topological vortex–antivortex pairs; a single vortex costs energy $E=\pi J\ln(L/a)$ , and the energy–entropy balance gives $k_BT_{\mathrm{BKT}}\simeq\pi J/2$ .
Below $T_{\mathrm{BKT}}$ , the correlation exponent $\eta(T)$ varies continuously; in terms of the renormalized spin stiffness, $\eta(T)=k_B T/[2\pi\rho_s(T)]$ , reaching the universal value $\eta(T_{\mathrm{BKT}})=1/4$ at the transition.
In three dimensions the model undergoes a conventional continuous transition in the 3D O(2) universality class.
Key idea: continuous $O(2)$ symmetry forbids long-range order in 2D (Mermin–Wagner), but topology saves a transition — vortices whose thermal binding–unbinding (a Coulomb-gas plasma transition) drive the BKT transition with a universal spin-stiffness jump.

The core insight

Continuous $O(2)$ symmetry forbids long-range order in 2D (Mermin–Wagner), but the model still has a transition: the topology of the circular order parameter permits vortices, whose thermally driven binding–unbinding (a Coulomb-gas plasma transition) drives the Berezinskii–Kosterlitz–Thouless transition with a universal spin-stiffness jump.

Where this lives today

The XY model's most beautiful confirmation is the universal jump. A thin film of $^4\mathrm{He}$ realizes the 2D XY model, with $\theta$ the superfluid phase and the stiffness playing the role of superfluid density; the Nelson–Kosterlitz prediction that $\rho_s/T$ jumps by a universal value at $T_{\mathrm{BKT}}$ was confirmed by Bishop and Reppy in 1978 with torsional oscillators — a landmark test of an "impossible" transition with no local order parameter. The same physics recurs in thin-film superconductors and Josephson-junction arrays, where vortex unbinding produces a current–voltage exponent that jumps from $3$ to $1$ at $T_{\mathrm{BKT}}$ . In three dimensions the XY model defines the $O(2)$ universality class, whose most precise realization is the superfluid $\lambda$ -transition of bulk helium: the specific-heat exponent $\alpha\approx-0.0127$ was measured aboard the Space Shuttle (the Lambda-Point Experiment, in microgravity to avoid gravitational rounding) to test 3D renormalization-group and conformal-bootstrap predictions to extraordinary accuracy. Meanwhile the topological ideas generalize: replacing vortices by dislocations and disclinations gives the Halperin–Nelson–Young theory of two-dimensional melting through an intermediate hexatic phase, observed in colloidal monolayers and electrons on helium — the reach of a model that began as "Ising with a continuous spin."

The physical picture

Picture a sheet of compass needles confined to a plane, each free to rotate. Aligning them all costs nothing in the ground state, and you can deform that alignment smoothly and cheaply: a long-wavelength twist — a spin wave — costs energy proportional to the square of its gradient, so very gentle twists are nearly free. These soft modes are why two dimensions sits exactly at the borderline where they accumulate just enough to wash out true alignment. Now wrap the needles once around a point as you walk in a circle. They cannot be smoothly unwound into a uniform configuration while keeping the order parameter well-defined everywhere, because the winding number is a topological invariant. That stubborn knot is a vortex; its energy grows logarithmically with system size. Place a vortex and an antivortex close together and their far fields cancel, so the pair costs only a logarithm of their separation — like a bound charge dipole. Heat the system and these pairs stretch and finally snap free. The moment they do, gentle algebraic order shatters into exponential disorder. That snapping is the BKT transition.

Figure

A vortex in a planar-spin field: the phase winds by 2π around the core

The whole drama is governed by one quantity, the spin stiffness $K=\beta J$ , the energetic price of a uniform gradient. Free energy is energy minus $T\times$ entropy for a single vortex: the energy is $\pi J\ln(L/a)$ and the entropy of placing the core is $2k_B\ln(L/a)$ , so a free vortex appears spontaneously precisely when $\pi J = 2k_BT$ , i.e. $k_BT_{\mathrm{BKT}}\simeq\pi J/2$ . This back-of-envelope balance is confirmed by experiments on helium films. The punchline is the shape of the spin–spin correlation function, plotted log–log. In three dimensions correlations flatten to a plateau at $m^2$ — true long-range order. In two dimensions below $T_{\mathrm{BKT}}$ they follow a straight line — a pure power law $r^{-\eta(T)}$ , quasi-long-range order with no genuine order parameter. Above $T_{\mathrm{BKT}}$ the same 2D system curves sharply downward into exponential decay. The marginal 2D power law, sitting between plateau and exponential, is the entire subject on one plot.

Figure

Spin correlation G(r) log–log: 2D power-law quasi-order vs 2D exponential vs 3D plateau

The core equations

The XY Hamiltonian rewards alignment of planar spins. Written with angles, it is a sum of cosines of angle differences — periodic, and invariant under a global rotation of every angle by the same amount, which is the continuous $O(2)$ symmetry.

H = -J\sum_{\langle ij\rangle}\mathbf{s}_i\cdot\mathbf{s}_j = -J\sum_{\langle ij\rangle}\cos(\theta_i-\theta_j) \qquad (1)

where: $\mathbf{s}_i$ — unit planar spin at site $i$ ; $\theta_i\in[0,2\pi)$ — its angle; $J>0$ — ferromagnetic coupling; $\langle ij\rangle$ — nearest-neighbor bonds. The single continuous parameter $\theta$ per site (versus Ising's $\pm1$ ) is precisely what admits gapless Goldstone modes. For slowly varying configurations the cosine can be expanded to quadratic order, giving a free-field (Gaussian) spin-wave theory. This harmonic Hamiltonian is what makes the low-temperature analysis tractable.

H_{\mathrm{sw}} \simeq \frac{J}{2}\sum_{\langle ij\rangle}(\theta_i-\theta_j)^2 \xrightarrow{\ \text{continuum}\ } \frac{J}{2}\int \mathrm{d}^d x\,(\nabla\theta)^2 \qquad (2)

where: $\nabla\theta$ — gradient of the smooth phase field. Read physically: the energy depends only on gradients of $\theta$ , the signature of a Goldstone field — a uniform shift costs nothing, so the softest modes are essentially free. In two dimensions the resulting phase fluctuations are logarithmically divergent, giving power-law (not constant) correlations. The spin correlation is the exponential of minus half the phase variance, and the variance grows as $\ln r$ .

\langle \mathbf{s}(0)\cdot\mathbf{s}(r)\rangle = e^{-\frac12\langle[\theta(0)-\theta(r)]^2\rangle} \sim r^{-\eta},\qquad \eta=\frac{k_BT}{2\pi J} \qquad (3)

where: $\eta$ — anomalous exponent, here temperature-dependent. Because $\eta$ varies continuously with $T$ , the whole low-temperature phase is a line of critical points — quasi-long-range order rather than a single critical temperature. A single vortex costs an energy that diverges logarithmically with system size, the hallmark of 2D topological defects. Integrating the gradient energy of the winding phase gives the logarithm.

E_v = \pi J \ln\!\left(\frac{L}{a}\right) + E_c \qquad (4)

where: $E_v$ — energy of one isolated vortex in a box of size $L$ ; $a$ — core radius (lattice cutoff); $E_c$ — non-universal core energy. Far from the core a vortex looks like a 2D Coulomb charge; a bound $+/-$ pair at separation $r$ costs only $2\pi J\ln(r/a)$ , finite and therefore allowed at low $T$ . The Kosterlitz–Thouless energy–entropy argument locates the transition by balancing the vortex cost against the entropy of its placement. Both scale as $\ln(L/a)$ , so their competition has a sharp crossover.

F_v = E_v - TS_v = (\pi J - 2k_BT)\ln\!\left(\frac{L}{a}\right),\qquad k_B T_{\mathrm{BKT}}^{(\mathrm{bare})} \sim \frac{\pi J}{2} \qquad (5)

where: $F_v$ — free energy of a single free vortex; $S_v=2k_B\ln(L/a)$ — entropy, since the core can sit at $\sim(L/a)^2$ positions. For $T<\pi J/2k_B$ the free energy diverges to $+\infty$ (free vortices forbidden, only bound pairs survive); for $T>\pi J/2k_B$ it diverges to $-\infty$ (free vortices proliferate). Renormalization lowers the true value to $k_BT_{\mathrm{BKT}}\approx0.893\,J$ . At the transition the renormalized stiffness takes a universal value, producing the Nelson–Kosterlitz jump and fixing $\eta=1/4$ there. This is a genuinely universal prediction, independent of lattice details.

K(T_{\mathrm{BKT}}^-)=\frac{2}{\pi} \quad\Longrightarrow\quad \eta(T_{\mathrm{BKT}}) =\frac{1}{2\pi K(T_{\mathrm{BKT}}^-)} =\frac{1}{4} \qquad (6)

where: $K(T)=\rho_s(T)/(k_B T)$ — renormalized dimensionless spin stiffness; $\rho_s$ — superfluid density $\propto K$ . The stiffness jumps discontinuously from $2/\pi$ to $0$ at the transition — the experimentally confirmed signature that distinguishes BKT from any conventional transition.

Vortices and the destruction of order

Derivation AMermin-Wagner: destruction of long-range order in 2D

We want to prove, within the spin-wave (Gaussian) approximation, that the 2D XY magnetization vanishes at any $T>0$ , and to see that the vanishing is slow — correlations decay as a power law, not exponentially. The strategy: treat the phase $\theta$ as a free Gaussian field, compute the mean-square phase difference between two points using equipartition, evaluate the resulting 2D integral (which grows logarithmically), and exponentiate to get the correlation function. Watch the infrared (small- $k$ ) end of the integral — that is where two dimensions is marginal.

Step 1. In the Gaussian theory each Fourier mode of the phase has variance set by equipartition,

\langle|\theta_q|^2\rangle=\frac{k_BT}{J q^2}.

Why: the quadratic action is $\frac{J}{2}\sum_q q^2|\theta_q|^2$ , and each mode gets $\tfrac12 k_BT$ ; the $1/q^2$ is the infrared enhancement that makes long wavelengths dangerous.

Step 2. The mean-square phase difference between origin and a point $r$ is

\langle[\theta(0)-\theta(r)]^2\rangle=\frac{2k_BT}{J}\int\frac{\mathrm{d}^2 q}{(2\pi)^2}\frac{1-\cos(\mathbf q\cdot\mathbf r)}{q^2}.

Why: subtracting the equal-point term inserts the factor $1-\cos$ , which cuts off the small- $q$ divergence just enough to make the integral grow only logarithmically.

Step 3. Evaluating the 2D integral gives a logarithm,

\int\frac{\mathrm{d}^2 q}{(2\pi)^2}\frac{1-\cos(\mathbf q\cdot\mathbf r)}{q^2}=\frac{1}{2\pi}\ln\frac{r}{a}.

Why: in $d=2$ the measure $\mathrm{d}^2q=q\,\mathrm{d}q\,\mathrm{d}\phi$ combines with $1/q^2$ to give $\mathrm{d}q/q$ , whose integral is $\ln$ ; the lower cutoff is $1/r$ , the upper $1/a$ .

Step 4. Because $\theta$ is Gaussian, the average of $e^{i\Delta\theta}$ exponentiates the variance:

\langle\mathbf s(0)\cdot\mathbf s(r)\rangle=e^{-\frac12\langle[\theta(0)-\theta(r)]^2\rangle}=\left(\frac{a}{r}\right)^{k_BT/2\pi J}.

Why: this is the standard Gaussian identity $\langle e^{iX}\rangle=e^{-\frac12\langle X^2\rangle}$ ; the logarithm in the exponent becomes a power law in $r$ .

Step 5. As $r\to\infty$ the correlation vanishes for any $T>0$ — no spontaneous magnetization, exactly Mermin–Wagner. But the decay is algebraic ( $r^{-\eta}$ ), unlike the exponential decay in $d=1$ : quasi-long-range order. Within the spin-wave approximation, $\eta(T)=k_B T/(2\pi J)$ . More generally, $\eta(T)=k_B T/[2\pi\rho_s(T)]$ , and the universal BKT jump $\rho_s(T_{\mathrm{BKT}}^-)=2k_B T_{\mathrm{BKT}}/\pi$ gives $\eta(T_{\mathrm{BKT}})=1/4$ . Why: a power law is the borderline between the plateau of true order ( $d\ge3$ ) and exponential disorder ( $d=1$ ) — the marginality of $d=2$ .

Go deeper · Derivation B · 2 minVortex energy and the Kosterlitz-Thouless unbinding

Derivation BVortex energy and the Kosterlitz-Thouless unbinding

We want the energy of a single vortex and the temperature at which vortices unbind. The strategy: compute the gradient energy of a configuration whose phase winds once around a core (giving a logarithm), then weigh that energy against the entropy of where the core can sit (also a logarithm), so their difference — the free energy — changes sign at a sharp temperature. The subtlety to watch is that both energy and entropy scale as $\ln(L/a)$ , which is exactly why a clean crossover exists.

Step 1. A unit vortex forces the phase to wind by $2\pi$ around its core, so the far-field gradient is fixed,

\nabla\theta = \frac{1}{r}\,\hat{\phi}.

Why: the winding is a topological constraint — you cannot smoothly remove it, so the gradient cannot vanish; it decays only as $1/r$ .

Step 2. Integrate the gradient energy over the plane, from core radius $a$ to system size $L$ ,

E_v=\frac{J}{2}\int_a^L \frac{1}{r^2}\,2\pi r\,\mathrm{d}r = \pi J \ln\frac{L}{a}.

Why: $(\nabla\theta)^2=1/r^2$ and the area element is $2\pi r\,\mathrm{d}r$ , so the integrand is $1/r$ — a logarithm, the hallmark of 2D defects.

Step 3. The entropy of the vortex is the log of the number of core positions,

S_v = k_B\ln\big[(L/a)^2\big]=2k_B\ln(L/a).

Why: the core can sit at any of $\sim(L/a)^2$ lattice cells, so the configurational entropy is logarithmic in $L$ too — matching the energy's scaling.

Step 4. The free energy of one free vortex is therefore

F_v = E_v - TS_v = (\pi J - 2k_BT)\ln\frac{L}{a}.

Why: energy and entropy carry the same $\ln(L/a)$ , so their coefficients compete directly and the sign of $F_v$ flips at a sharp temperature.

Step 5. Within this bare single-vortex energy–entropy argument, the sign change occurs at $k_B T \sim \pi J/2$ . This provides only a qualitative estimate of the transition: below this scale isolated vortices are energetically suppressed and bound vortex–antivortex pairs dominate, whereas above it free vortices proliferate and destroy quasi-long-range order. Vortex–antivortex fluctuations renormalize the spin stiffness, shifting the actual transition of the square-lattice XY model to $k_B T_{\mathrm{BKT}}\approx 0.893\,J$ . More generally, the BKT transition is characterized by the universal jump $\rho_s(T_{\mathrm{BKT}}^-)=2k_B T_{\mathrm{BKT}}/\pi$ . In the equivalent two-dimensional Coulomb-gas picture, the proliferation of free topological charges screens the logarithmic interaction, corresponding to a dielectric-to-plasma transition. Why: a proliferating gas of free charges screens the interaction and disorders the phase — exactly the metal–insulator transition of the 2D Coulomb gas.

Limiting cases

The XY model's behavior depends sharply on dimension and on the nature of the fluctuations.

Case	Result	Meaning
$d=1$	$\langle\mathbf{s}_0\cdot\mathbf{s}_r\rangle\sim e^{-r/\xi}$ , $\xi\propto\beta J$	No transition at $T>0$ ; correlations always exponential, $\xi$ diverges only as $T\to0$ .
$d=2$ , $T<T_{\mathrm{BKT}}$	$\langle\mathbf{s}_0\cdot\mathbf{s}_r\rangle\sim r^{-\eta(T)}$ , $\eta(T)=\dfrac{k_BT}{2\pi\rho_s(T)}\simeq\dfrac{k_BT}{2\pi J}$ (spin-wave limit)	Quasi-long-range order: power-law correlations, continuously varying exponent — a line of critical points.
$d=2$ , at $T_{\mathrm{BKT}}$	$\eta=\tfrac14$ , $K(T_{\mathrm{BKT}}^-)=\dfrac{2}{\pi}$ , $\xi\sim e^{b/\sqrt{T-T_{\mathrm{BKT}}}}$	Universal stiffness jump $2/\pi\to0$ ; infinite-order transition, essential singularity in $\xi$ ; no latent heat.
$d\ge3$	$m(T)\sim(T_c-T)^{\beta}$ , $\beta\approx0.349$	Conventional second-order transition in the 3D $O(2)$ class; finite magnetization below $T_c$ ( $\nu\approx0.672$ , $\eta\approx0.038$ ).

Worked examples

Example 1 — The Kosterlitz–Thouless unbinding temperature.

Step 1. A single free vortex has free energy $F_v=(\pi J-2k_BT)\ln(L/a)$ , combining energy $\pi J\ln(L/a)$ and entropy $2k_B\ln(L/a)$ .

Step 2. Within this bare single-vortex energy–entropy argument, the coefficient of $\ln(L/a)$ changes sign when $\pi J-2k_BT=0$ .

Step 3. This gives the bare estimate $k_BT_{\mathrm{BKT}}^{(\mathrm{bare})}\sim\pi J/2\approx1.571J$ .

Step 4. This estimate neglects vortex–antivortex fluctuations and the renormalization of the spin stiffness. For the square-lattice XY model, the actual transition occurs at $k_BT_{\mathrm{BKT}}\approx0.893J$ . The energy–entropy estimate of the BKT temperature is $k_BT_{\mathrm{BKT}}=\pi J/2\approx1.571\,J$ , lowered by screening to the measured $\approx0.893\,J$ .

Example 2 — The correlation exponent at the transition, and how far quasi-order reaches.

Step 1. In the quasi-ordered phase $\eta(T)=k_BT/(2\pi J)$ . At the transition the stiffness takes its universal value $K=2/\pi$ , i.e. $\beta_{\mathrm{BKT}}J=2/\pi$ , so $k_BT_{\mathrm{BKT}}=\pi J/2$ .

Step 2. Substitute: $\eta(T_{\mathrm{BKT}})=(\pi J/2)/(2\pi J)=1/4=0.25$ . This is universal, tied to the universal stiffness jump.

Step 3. Thus, exactly at the BKT transition, $\eta(T_{\mathrm{BKT}})=1/4$ . For $T<T_{\mathrm{BKT}}$ , $\eta(T)<1/4$ , and the correlations decay algebraically as $r^{-\eta(T)}$ , becoming progressively slower at lower temperatures.

Step 4. Compare the correlation at $r=10a$ : at $T_{\mathrm{BKT}}$ , $\eta=1/4$ , so $G\sim10^{-0.25}=0.562$ . At lower temperatures, $\eta(T)<1/4$ , so the correlations decay more slowly; the precise value of $\eta(T)$ requires the temperature-dependent renormalized spin stiffness $\rho_s(T)$ . The transition exponent is $\eta(T_{\mathrm{BKT}})=1/4$ , universal; at $r=10a$ this gives $G\sim0.56$ at $T_{\mathrm{BKT}}$ , and correlations decay still more slowly below the transition, with the precise rate set by the renormalized stiffness $\rho_s(T)$ .

Origins · 1 minvortex and how this was discovered

vortex

topological defect of a phase field · A vortex is a point (in 2D) around which the phase of the order parameter winds by an integer multiple of 2π and cannot be smoothly removed. In the XY model, superfluids, and superconductors, vortices carry quantized circulation and interact logarithmically like 2D Coulomb charges. Their thermal binding and unbinding is the mechanism of the Berezinskii–Kosterlitz–Thouless transition, recognized by the 2016 Nobel Prize in Physics to Kosterlitz and Thouless (with Haldane).

History

1966

— Mermin and Wagner prove that continuous symmetry cannot break in d ≤ 2 at finite temperature

1971

— Berezinskii identifies the role of topological vortices in the 2D XY model

1973

— Kosterlitz and Thouless describe the vortex-unbinding transition and its infinite-order character

1977

— Nelson and Kosterlitz predict the universal spin-stiffness jump K = 2/π

1978

— Bishop and Reppy confirm the universal superfluid-density jump in helium films

2016

— Kosterlitz and Thouless (with Haldane) receive the Nobel Prize for topological phase transitions

Going further

References. Pathria & Beale, Statistical Mechanics (3rd ed.), Ch. 12–13 · Kardar, Statistical Physics of Fields, Ch. 8–9 · Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity, Ch. 9 · Chaikin & Lubensky, Principles of Condensed Matter Physics, Ch. 6 & 9 · Kosterlitz & Thouless, J. Phys. C 6, 1181 (1973) · Berezinskii, Sov. Phys. JETP 32, 493 (1971) · Nelson & Kosterlitz, Phys. Rev. Lett. 39, 1201 (1977). Key numerical benchmarks: the vortex energy is $\pi J\ln(L/a)$ , the bare energy–entropy estimate gives $k_B T_{\mathrm{BKT}}^{(\mathrm{bare})}\sim\pi J/2$ , the universal stiffness jump is $K(T_{\mathrm{BKT}}^-)=2/\pi$ with $\eta(T_{\mathrm{BKT}})=1/4$ , the square-lattice Monte Carlo value is $k_B T_{\mathrm{BKT}}\approx0.893J$ , and the 3D XY exponents are standard conformal-bootstrap values.

Builds on: The Ising Model: Definition and Hamiltonian · Order Parameter and Spontaneous Magnetization · Peierls Argument for Long-Range Order · Domain Walls and Interfacial Excitations. Leads to: Goldstone Modes and Spontaneous Symmetry Breaking · Mermin-Wagner Theorem · Renormalization Group and the Coulomb Gas · Two-Dimensional Melting (KTHNY Theory).

If spin waves cost so little energy, why is any order possible in 3D at all?

It comes down to how the soft-mode fluctuations add up. The angular variance is $\langle\theta^2\rangle\sim\int\mathrm{d}^dk/k^2\sim\int k^{d-3}\mathrm{d}k$ . In three dimensions this integral converges at small $k$ : the spin waves reduce the magnetization but leave it finite. In two dimensions the same integral grows as $\ln L$ and in one dimension as $L$ , so the accumulated fluctuations diverge and wash out the order. Spin waves are cheap in every dimension; whether they destroy order depends on how many long-wavelength modes there are, which is a matter of phase space, i.e. dimension.

What exactly unbinds at T_BKT — the spins or the vortices?

The vortices. Below $T_{\mathrm{BKT}}$ vortices exist only as tightly bound neutral pairs (a vortex next to an antivortex), whose far fields cancel so they cost only a finite $2\pi J\ln(r/a)$ and barely disturb the phase. As you heat the system, these pairs stretch; at $T_{\mathrm{BKT}}$ the entropy of separating a pair finally beats the logarithmic binding energy and the pairs snap apart into a plasma of free vortices. Those free vortices screen the phase stiffness to zero, converting the quasi-ordered power-law phase into an exponentially disordered one.

Why is the BKT transition called "infinite order"?

Because every temperature derivative of the free energy is continuous through $T_{\mathrm{BKT}}$ — there is no discontinuity or divergence in the specific heat, unlike a first-order (latent heat) or second-order (power-law divergence) transition. The correlation length diverges not as a power law but with an essential singularity, $\xi\sim\exp(b/\sqrt{T-T_{\mathrm{BKT}}})$ , which is smoother than any power. Ehrenfest's classification would call a transition " $n$ -th order" if the $n$ -th derivative first jumps; here no finite derivative ever does, so it is "infinite order." A broad, non-singular specific-heat bump sits slightly above $T_{\mathrm{BKT}}$ .

The exponent η(T) changes with temperature — how can a "critical exponent" not be a fixed number?

In an ordinary critical point there is one temperature $T_c$ and fixed exponents. The 2D XY low-temperature phase is different: it is an entire line of critical points, each temperature critical in the sense of power-law correlations, but with its own exponent $\eta(T)=k_BT/2\pi J$ that grows with temperature. This is what "quasi-long-range order" means — scale invariance without a unique fixed point. Only at the end of the line, $T_{\mathrm{BKT}}$ , does $\eta$ reach a universal value $1/4$ , pinned by the universal stiffness jump.

Does the same physics explain superfluid helium films?

Yes — that is the model's triumph. In a thin $^4\mathrm{He}$ film the superfluid order parameter is a complex amplitude whose phase is exactly the XY angle, and the spin stiffness is the superfluid density $\rho_s$ . Vortices are quantized circulation defects. The Nelson–Kosterlitz theory predicts that $\rho_s/T$ jumps by a universal value $2mk_B/\pi\hbar^2$ (in appropriate units) at $T_{\mathrm{BKT}}$ , independent of the film. Bishop and Reppy measured exactly this jump with a torsional oscillator in 1978, one of the cleanest confirmations of an entire theoretical framework.

Why does a magnetic anisotropy let a real 2D magnet order despite Mermin–Wagner?

Mermin–Wagner requires an exactly continuous symmetry. Any easy-axis or easy-plane anisotropy breaks $O(2)$ down toward a discrete symmetry and opens a gap in the spin-wave spectrum, $\epsilon(k)\to\Delta+a^2k^2$ . That gap $\Delta$ removes the dangerous $1/k^2$ at small $k$ , so the fluctuation integral converges and order survives at finite $T$ . Real 2D magnets like monolayer CrI $_3$ order only because of such anisotropy — their Curie temperatures track the anisotropy gap, a direct fingerprint of the theorem's loophole.