The Kosterlitz-Thouless Transition

The Mermin-Wagner theorem seems to close the book on two-dimensional order: a continuous symmetry cannot break spontaneously at finite temperature in 2D, so the XY model — spins free to rotate in a plane — has no magnetization, ever. And yet the model unmistakably has two phases: power-law correlations at low temperature, exponential collapse at high. Something sharp happens in between, and it cannot be described by an order parameter, symmetry breaking, or Landau theory. What changes is topology: the behaviour of vortices — point defects around which the spin angle winds by $2\pi$ . Below $T_{KT}$ , vortices and antivortices are bound in neutral pairs that merely soften the spin-wave phase. Above $T_{KT}$ they unbind, proliferate as a free plasma, and shred the power-law order. Kosterlitz and Thouless located the transition with an argument of startling economy: a single vortex costs energy $\pi J\ln(L/a)$ but gains positional entropy $2k_B\ln(L/a)$ , so the free energy of one free vortex flips sign at $k_BT_{KT}=\pi J/2$ . Both terms diverge logarithmically with system size — that is why the competition is sharp. The resulting transition is like nothing in the Landau catalogue: the free energy is infinitely smooth at $T_{KT}$ , the correlation length diverges with an essential singularity $\xi\sim e^{b/\sqrt{T-T_{KT}}}$ rather than a power law, and the superfluid stiffness drops to zero discontinuously by the universal amount $\rho_s(T_{KT}^-)/k_BT_{KT}=2/\pi$ — the same number in every film ever measured. This article derives the vortex energy, runs the free-energy argument to $T_{KT}$ , and extracts the universal jump from the RG relevance of vortices. The XY model and the Mermin-Wagner theorem are owned by The XY Model and Continuous Spins and the Mermin-Wagner Theorem and Lower Critical Dimension — recalled here, not re-derived.

Key takeaways

The 2D XY model has no long-range order (Mermin-Wagner), yet it has a sharp topological transition — vortex unbinding, with no order parameter and no symmetry breaking.
Below $T_{KT}$ : bound vortex-antivortex pairs; power-law correlations $\sim(a/r)^{\eta(T)}$ , $\eta=k_BT/2\pi J$ — a line of critical points.
A single vortex costs $E=\pi J\ln(L/a)$ and gains entropy $S=2k_B\ln(L/a)$ ; $F=(\pi J-2k_BT)\ln(L/a)$ changes sign at $k_BT_{KT}=\pi J/2$ .
Above $T_{KT}$ free vortices proliferate, screen the stiffness to zero, and force exponential decay.
The transition is infinite-order: $\xi\sim\exp(b/\sqrt{T-T_{KT}})$ , an essential singularity; the specific heat stays smooth.
Key idea: the stiffness does not go to zero continuously — it jumps from the universal value $\rho_s(T_{KT}^-)=(2/\pi)\,k_BT_{KT}$ to zero, a parameter-free prediction confirmed in every superfluid film.

The core insight

Both the energy of an isolated vortex and its positional entropy grow as $\ln(L/a)$ , so their competition is decided by coefficients alone: $F=(\pi J-2k_BT)\ln(L/a)$ flips sign at $k_BT_{KT}=\pi J/2$ . Below, topological defects are confined in neutral pairs and the phase is critical with continuously varying exponents; above, they unbind and destroy quasi-long-range order — a phase transition carried entirely by topology.

Where this lives today

The KT transition organizes every quasi-two-dimensional system with a $U(1)$ phase. Superfluid helium films show the universal jump in torsional-oscillator and third-sound measurements. Superconducting films and Josephson-junction arrays undergo vortex unbinding with real flux vortices — the jump surfaces as the $V\sim I^3\to V\sim I$ switch of the current-voltage exponent. Trapped 2D Bose gases realize the transition with cold atoms, matter-wave interference making the proliferating defects individually visible. Two-dimensional crystals melt by the KTHNY two-step scenario — dislocation then disclination unbinding, with an intervening hexatic phase, seen in colloidal monolayers. The vocabulary invented here — topological defects, quasi-long-range order, essential singularities — seeded the topological revolution recognized by the 2016 Nobel Prize. The deeper machinery (Coulomb-gas and sine-Gordon mappings, the full RG flow diagram) belongs to the renormalization-group chapter; the essentials — two logarithms, one competition — are all here.

The physical picture

Take the two-dimensional XY model — planar spins $\mathbf{s}_i=(\cos\theta_i,\sin\theta_i)$ with nearest-neighbour coupling $-J\cos(\theta_i-\theta_j)$ , as developed in The XY Model and Continuous Spins. At low temperature the angles vary slowly and the energy reduces to the continuum "elastic" form $\tfrac{J}{2}\int(\nabla\theta)^2\,d^2r$ . The Mermin-Wagner Theorem and Lower Critical Dimension shows that gapless spin waves on this sheet destroy true long-range order at any $T>0$ — $\langle\mathbf{s}\rangle=0$ always. But the same spin-wave calculation leaves behind something remarkable: correlations that decay not exponentially but as a power law, $\langle\mathbf{s}(0)\cdot\mathbf{s}(r)\rangle\sim(a/r)^{\eta}$ , with temperature-dependent exponent $\eta=k_BT/2\pi J$ . Every low temperature is critical — a line of critical points, each with its own exponent. This is quasi-long-range order (QLRO): no magnetization, but orientational memory that fades only algebraically. Spin waves alone would keep that power law to arbitrarily high temperature, which cannot be right — at high $T$ the lattice model manifestly has exponentially short correlations. What the smooth spin-wave picture misses are vortices: point defects around which $\theta$ winds by $2\pi n$ . A vortex cannot be smoothed away — its winding number is a topological invariant. Around a $+1$ vortex the spins circulate; around a $-1$ antivortex they form the hyperbolic pattern in the figure below. A bound $+1,-1$ pair perturbs the spin field only out to the pair size (the windings cancel far away), costing finite energy $\approx 2\pi J\ln(d/a)$ for separation $d$ . At low temperature the vortex gas is a dilute set of bound neutral "molecules", and QLRO survives, merely renormalized. At high temperature the pairs ionize: free vortices roam, each scrambling the phase of every spin around it — correlations collapse exponentially.

Figure

Vortex-antivortex pair bound below T_KT (left); above T_KT the pair ionizes into free vortices that scramble the phase (right)

Why does the balance tip at a sharp temperature? Because for a single vortex, energy and entropy scale with system size identically. The winding forces $|\nabla\theta|=1/r$ out to the edge, so the elastic energy integrates to $\pi J\ln(L/a)$ . But the core can sit on any of $(L/a)^2$ plaquettes — entropy $2k_B\ln(L/a)$ , the same logarithm. The free energy of one free vortex is $(\pi J-2k_BT)\ln(L/a)$ : for $T<\pi J/2k_B$ it is $+\infty$ in the thermodynamic limit (free vortices absolutely forbidden; only bound pairs occur), while for $T>\pi J/2k_B$ it is $-\infty$ (free vortices nucleate without bound until their mutual screening stabilizes a plasma). A logarithm competing with a logarithm produces a knife-edge — that is the Kosterlitz-Thouless transition, and its signature observable is the spin stiffness, which the free-vortex plasma screens to zero discontinuously, as the second figure shows.

Figure

The universal jump: the stiffness curve meets the line (2/π)T and drops discontinuously to zero at T_KT

Two consequences deserve emphasis. First, because the low-temperature phase is a line of critical points, $\eta$ varies continuously with temperature, from $0$ at $T\to0$ up to exactly $1/4$ at the transition — universality fixes the endpoint, not a single exponent. Second, no local quantity becomes singular: the free energy is smooth to all orders at $T_{KT}$ , and the specific heat shows only an analytic bump above the transition, where the pairs actually ionize. Everything sharp about this transition lives in the stiffness and the correlations.

The core equations

The stage is the continuum elastic description recalled from The XY Model and Continuous Spins.

H \;=\; -J\sum_{\langle ij\rangle}\cos(\theta_i-\theta_j)\;\;\longrightarrow\;\; H[\theta]\;=\;\frac{J}{2}\int |\nabla\theta|^2\, d^2r \qquad (1)

where: $\theta(\mathbf{r})$ is the local spin angle and $J$ the exchange coupling; the continuum form holds for slowly varying configurations above the lattice scale $a$ . The stiffness $J$ is the single coupling controlling everything below. Gaussian spin waves on (1) give the low-temperature correlations — power-law, not exponential.

\langle \mathbf{s}(0)\cdot\mathbf{s}(\mathbf{r})\rangle \;=\; \Big(\frac{a}{r}\Big)^{\eta(T)}, \qquad \eta(T)\;=\;\frac{k_BT}{2\pi J} \qquad (2)

where: $a$ is the lattice (core) cutoff and $\eta$ the temperature-dependent decay exponent. Each $T$ below the transition is a critical point with its own $\eta$ ; at the transition itself $\eta(T_{KT})=1/4$ , a universal endpoint. A single vortex — the configuration spin waves cannot make — carries a logarithmically divergent energy.

E_{\text{vortex}} \;=\; \pi J \ln\!\Big(\frac{L}{a}\Big) + E_c \qquad (3)

where: $L$ is the linear system size, $a$ the core radius, and $E_c$ the finite, non-universal core energy from $r\lesssim a$ where the elastic description fails. For winding $n$ the coefficient becomes $\pi J n^2$ — why only $n=\pm1$ defects matter. Balancing (3) against the core's positional entropy gives the transition temperature.

F \;=\; E - TS \;=\; \big(\pi J - 2k_BT\big)\ln\!\Big(\frac{L}{a}\Big), \qquad k_BT_{KT} \;=\; \frac{\pi J}{2} \qquad (4)

where: $S=2k_B\ln(L/a)$ counts the $(L/a)^2$ core positions. Below $T_{KT}$ , $F\to+\infty$ as $L\to\infty$ : isolated vortices forbidden. Above, $F\to-\infty$ : proliferation. The estimate uses the bare stiffness; pair screening renormalizes $J$ down to $\rho_s(T)$ and lowers the actual transition (Monte Carlo: $k_BT_{KT}\approx0.893\,J$ on the square lattice), but the criterion's form survives exactly. From above, the correlation length diverges faster than any power — an essential singularity.

\xi(T) \;\sim\; a\,\exp\!\Big(\frac{b}{\sqrt{T-T_{KT}}}\Big), \qquad T\to T_{KT}^{+} \qquad (5)

where: $b$ is a non-universal constant. There is no exponent $\nu$ : $\xi$ outgrows every power law $|T-T_{KT}|^{-\nu}$ , and the singular free energy $\sim\xi^{-2}$ vanishes with all its derivatives — an infinite-order transition, invisible in the specific heat. The renormalized stiffness obeys the exact endpoint condition — the universal jump.

\frac{\rho_s(T_{KT}^{-})}{k_B T_{KT}} \;=\; \frac{2}{\pi}\;=\;0.6366\ldots, \qquad \rho_s(T_{KT}^{+})\;=\;0 \qquad (6)

where: $\rho_s(T)$ is the renormalized spin stiffness (helicity modulus; for a superfluid film $\rho_s=\hbar^2\sigma_s/m^2$ , $\sigma_s$ the areal superfluid mass density). At $T_{KT}$ the stiffness sits exactly on the line $(2/\pi)k_BT$ and then drops discontinuously to zero. The ratio $2/\pi$ contains no material parameters — the same for helium films, superconducting films, and cold-atom gases.

Derivation AThe energy of a single vortex

The goal is the coefficient in equation (3): a winding-number $+1$ defect in the elastic theory (1) costs $E=\pi J\ln(L/a)$ . The strategy is to translate the topological constraint — the angle winds by $2\pi$ around the core — into the gradient field it forces at every radius, then integrate the elastic energy density over the annulus from the core cutoff $a$ to the system edge $L$ . The subtlety: the logarithm is generated by the whole annulus, every decade of scale contributing equally — neither a core artefact nor an edge effect, which is why it competes cleanly with positional entropy at every scale.

Step 1. Impose the topology. A vortex of winding number $n$ at the origin means $\oint_C \nabla\theta\cdot d\boldsymbol{\ell} = 2\pi n$ for every loop $C$ encircling the core. The minimal-energy field with this constraint is the symmetric one, $\theta(\mathbf{r})=n\,\varphi$ , with $\varphi$ the polar angle.

Step 2. Read off the gradient. On a circle of radius $r$ the angle changes by $2\pi n$ over circumference $2\pi r$ , so $|\nabla\theta| = n/r$ , azimuthal. The gradient decays only as $1/r$ — the topological charge is felt at all distances, unlike a spin wave, whose gradients can be made arbitrarily small everywhere.

Step 3. Integrate the elastic energy density. With $n=1$ , $E=\frac{J}{2}\int |\nabla\theta|^2\,d^2r = \frac{J}{2}\int_a^{L} \frac{1}{r^2}\,(2\pi r)\,dr = \pi J\int_a^{L}\frac{dr}{r} = \pi J\,\ln\!\Big(\frac{L}{a}\Big).$ The $1/r^2$ from the squared gradient times the $2\pi r$ area element gives $1/r$ , integrating to the logarithm — equal energy per decade of scale.

Step 4. Note the corrections. The region $r<a$ contributes the finite, non-universal core energy $E_c$ (absorbed into the fugacity $y=e^{-E_c/k_BT}$ later). General winding gives $E=\pi J n^2\ln(L/a)$ : doubly charged vortices cost four times the energy with no extra positional entropy, so the transition is carried by $n=\pm1$ . A $+1,-1$ pair at separation $d$ has its far fields cancel, cutting the integral off at $d$ : $E_{\text{pair}}\approx 2\pi J\ln(d/a)+2E_c$ — finite, which is why pairs exist at every temperature. Verified: the companion verifier performs the Step-3 integral symbolically with sympy — $\tfrac{J}{2}\int_a^L r^{-2}\,2\pi r\,dr = \pi J\ln(L/a)$ exactly.

Go deeper · Derivation B · 1 minThe Kosterlitz-Thouless free-energy argument

Derivation BThe Kosterlitz-Thouless free-energy argument

The goal is equation (4): the sharp temperature $k_BT_{KT}=\pi J/2$ above which isolated vortices proliferate. The strategy — the original Kosterlitz-Thouless argument, one of the great back-of-envelope calculations of condensed-matter physics — is to write the free energy of a single free vortex in a system of size $L$ and watch its sign in the thermodynamic limit. It works only because energy and entropy carry the same $\ln(L/a)$ : for any other functional form one term would win at all temperatures and there would be no finite-temperature transition.

Step 1. Energy. From the first derivation, a single $n=\pm1$ vortex costs $E=\pi J\ln(L/a)+E_c$ ; as $L\to\infty$ the core term is negligible against the logarithm.

Step 2. Entropy. The core can occupy any of $(L/a)^2$ plaquettes, so $S=k_B\ln[(L/a)^2]=2k_B\ln(L/a)$ — the same logarithm, different coefficient.

Step 3. Compete. $F = E-TS = \big(\pi J-2k_BT\big)\ln(L/a)$ . For $k_BT<\pi J/2$ the coefficient is positive and $F\to+\infty$ : the Boltzmann weight of a free vortex vanishes as the power $(a/L)^{(\pi J-2k_BT)/k_BT}$ — free vortices are absent in the thermodynamic limit, and vorticity exists only in bound neutral pairs. For $k_BT>\pi J/2$ the coefficient is negative, $F\to-\infty$ : proliferation, halted only when the growing plasma screens the logarithmic interaction itself.

Step 4. Locate the transition and interpret. The sign change defines $k_BT_{KT}=\pi J/2$ . Two refinements sharpen but do not alter the result. First, bound pairs present below $T_{KT}$ polarize in the field of a test vortex like molecules in a dielectric, replacing $J$ by the renormalized stiffness $\rho_s(T)<J$ ; run self-consistently, the criterion becomes $k_BT_{KT}=\tfrac{\pi}{2}\rho_s(T_{KT})$ — the exact renormalized statement, equation (6) read backwards. The screening is substantial: Monte Carlo places the square-lattice transition at $k_BT_{KT}\approx0.893\,J$ , well below the bare $\pi/2\approx1.571\,J$ . Second, the same balance read scale-by-scale generates the flow equations of the next derivation. Verified: the verifier confirms symbolically that $F=(\pi J-2k_BT)\ln(L/a)$ changes sign precisely at $T=\pi J/2k_B$ , evaluating the coefficient at $T_{KT}/2$ (positive, bound) and $2T_{KT}$ (negative, free).

Go deeper · Derivation C · 1 minThe universal stiffness jump from vortex relevance

Derivation CThe universal stiffness jump from vortex relevance

The goal is equation (6): at the transition the renormalized stiffness takes the parameter-free value $\rho_s/k_BT_{KT}=2/\pi$ , then drops to zero. The strategy is renormalization-group in miniature: ask how the vortex fugacity — the statistical weight per unit area — changes under coarse-graining, using only the energy-entropy logarithms already derived. The subtlety is to phrase the KT criterion scale by scale: what matters is the dimensionless stiffness $K(\ell)=\rho_s(\ell)/k_BT$ at the scale examined, because pairs smaller than that scale are already absorbed into $K$ .

Step 1. Define the players. Let $K=\rho_s/k_BT$ be the dimensionless renormalized stiffness and $y=e^{-E_c/k_BT}$ the vortex fugacity at the core scale. A vortex at scale $r=ae^{\ell}$ carries Boltzmann suppression $(r/a)^{-\pi K}$ from its elastic energy and phase-space gain $(r/a)^{2}$ from its positional freedom.

Step 2. Coarse-grain the fugacity. Combining the two factors, $y(\ell) = y\, e^{(2-\pi K)\ell}$ , i.e. $\frac{dy}{d\ell} = \big(2-\pi K\big)\,y .$ Vortices are irrelevant (bound) when $\pi K>2$ and relevant (proliferating) when $\pi K<2$ . Marginality $\pi K=2$ is the scale-invariant restatement of the free-energy argument.

Step 3. Close the system. Pairs at scale $\ell$ screen the stiffness felt at larger scales, as bound charges screen a 2D Coulomb plasma: to leading order, $\frac{dK^{-1}}{d\ell} = 4\pi^3 y^2 + \mathcal{O}(y^4).$ These are the Kosterlitz recursion relations. On the critical separatrix the flow terminates on the fixed line $y^*=0$ with terminal stiffness $\pi K^*=2$ ; every low-temperature point flows to some $K^*(T)>2/\pi$ , and integrating the flow just above the separatrix yields the essential singularity of equation (5).

Step 4. Read off the jump. As $T\to T_{KT}^{-}$ the terminal stiffness approaches the marginal value, so $\lim_{T\to T_{KT}^-}\rho_s(T)/k_BT=K^*(T_{KT})=2/\pi$ , while for any $T>T_{KT}$ the fugacity runs away and screening drives $\rho_s\to0$ . Hence the discontinuity, with no material constant anywhere — the Nelson-Kosterlitz universal jump. Consistency: inserting $K=2/\pi$ into $\eta=1/(2\pi K)$ gives $\eta(T_{KT})=1/4$ , the universal endpoint of the critical line. Verified: the verifier checks $2/\pi=0.63662$ , the endpoint $\eta=1/(2\pi K)=1/4$ at $K=2/\pi$ , and $\eta(T_{KT})=k_BT_{KT}/2\pi J=1/4$ in the bare theory as well.

Limiting cases

Each limit is a consistency check on the vortex picture.

Case	Result	Meaning
$T\to0$	$\eta\to0$ , $\rho_s\to J$ , pairs exponentially rare	Nearly ordered; pure spin-wave physics; bare stiffness.
$0<T<T_{KT}$	$\langle\mathbf{s}(0)\cdot\mathbf{s}(r)\rangle\sim r^{-\eta(T)}$ , $\eta=k_BT/2\pi\rho_s$	Line of critical points, continuously varying exponent — QLRO; magnetization still zero (Mermin-Wagner respected).
$T=T_{KT}$	$\eta=1/4$ , $\rho_s=(2/\pi)k_BT_{KT}$ , then jump to $0$	Universal endpoint: exponent and stiffness ratio fixed by no material parameter.
$T\to T_{KT}^{+}$	$\xi\sim e^{b/\sqrt{T-T_{KT}}}$ , $\rho_s=0$	Essential singularity, infinite order; free-vortex plasma screens all stiffness; smooth specific heat.

Common trapMermin-Wagner does not forbid this transition — and Landau theory cannot describe it

Two opposite confusions surround the KT transition. The first: "Mermin-Wagner proves there is no order in 2D, so there can be no phase transition." Wrong — the theorem (see Mermin-Wagner Theorem and Lower Critical Dimension) forbids a nonzero local order parameter for a continuous symmetry in 2D; it says nothing about transitions between two disordered phases that differ in correlation decay (power law versus exponential) and stiffness (finite versus zero). Both KT phases have $\langle\mathbf{s}\rangle=0$ . The second confusion: forcing the transition into the mold of Landau Theory of Phase Transitions. There is no order parameter turning on, no symmetry breaking, no power-law $\xi\sim|t|^{-\nu}$ , and no specific-heat singularity — the free energy is infinitely differentiable at $T_{KT}$ . The transition is diagnosed only by defects unbinding, the stiffness jump, and the change in correlation decay. Calorimetry sees only a smooth bump above $T_{KT}$ — famously, the transition is invisible at the transition.

Worked examples

Example 1 — A magnetic film with J/k_B = 1 K. A quasi-2D easy-plane magnet has exchange coupling $J/k_B=1$ K. Estimate the KT temperature, the stiffness jump, and compare with the honest (renormalized) numbers.

Step 1. Bare estimate, equation (4): $T_{KT}=(\pi/2)(1\ \mathrm{K})=\mathbf{1.571\ K}$ .

Step 2. The stiffness at the transition, equation (6): $\rho_s(T_{KT}^-)/k_B=(2/\pi)\,T_{KT}=(2/\pi)(\pi/2)(1\ \mathrm{K})=\mathbf{1.000\ K}$ — exactly the bare $J/k_B$ , as it must be: the naive estimate says the transition happens where the unrenormalized stiffness sits on the $2/\pi$ line.

Step 3. Screening in reality. Bound pairs reduce the stiffness before the transition arrives, so the true transition sits lower: Monte Carlo gives $k_BT_{KT}\approx0.893\,J$ , hence $T_{KT}\approx\mathbf{0.893\ K}$ and an actual jump height $\rho_s/k_B=(2/\pi)(0.893)\approx\mathbf{0.568\ K}$ . The bare estimate overshoots by $76\%$ , all of it vortex-pair screening. The takeaway: the bare argument nails the mechanism and the universal ratio; renormalization moves only the non-universal location.

Example 2 — The Nelson-Kosterlitz line for helium films. For a superfluid film the phase stiffness is $\rho_s=\hbar^2\sigma_s/m^2$ , with $\sigma_s$ the areal superfluid mass density and $m$ the $^4$ He atomic mass. Turn the universal jump into a measurable prediction.

Step 1. Insert into equation (6): $\hbar^2\sigma_s(T_c)/m^2 = (2/\pi)k_BT_c$ , so $\frac{\sigma_s(T_c)}{T_c} = \frac{2m^2k_B}{\pi\hbar^2}$ — every quantity on the right a fundamental constant, the same for every film, whatever its thickness or substrate.

Step 2. Evaluate with $m=6.6447\times10^{-27}$ kg: $\sigma_s(T_c)/T_c = \mathbf{3.49\times10^{-9}\ g\,cm^{-2}\,K^{-1}}.$

Step 3. Interpret. A film going superfluid at $T_c=1$ K must arrive carrying $3.5\times10^{-9}$ g/cm² of areal superfluid density, then lose it discontinuously. Bishop and Reppy's 1978 torsional-oscillator experiments traced $\sigma_s(T_c)/T_c$ across films of widely varying thickness — the data collapse onto exactly this line. The takeaways: the KT theory makes a zero-parameter quantitative prediction, $\sigma_s(T_c)/T_c=3.49\times10^{-9}$ g cm⁻² K⁻¹, and experiment confirms it.

Origins · 1 mintopological defect and how this was discovered

topological defect

1971–2016 · The field configurations whose winding number no smooth deformation can undo — and whose unbinding drives a phase transition beyond the Landau paradigm

A topological defect is a configuration — here the XY vortex — classified by an integer winding that no continuous deformation can remove; it can disappear only by annihilating with a partner of opposite charge. That robustness is why defects, not spin waves, control the fate of two-dimensional order. Vadim Berezinskii (1935–1980) in Moscow argued in 1971 that the low-temperature 2D XY phase is qualitatively distinct despite carrying no magnetization; John Kosterlitz (b. 1943) and David Thouless (1934–2019) in Birmingham independently identified the vortex-unbinding mechanism in 1972–73, and Kosterlitz's 1974 renormalization group made it quantitative. The 2016 Nobel Prize went to Thouless, Haldane and Kosterlitz — the recognition that topology, not symmetry, can organize phases of matter. Berezinskii did not live to see it; the transition carries the initials of all three (BKT).

History

1966

— Mermin and Wagner prove there is no spontaneous breaking of a continuous symmetry in two dimensions at finite temperature

1971

— Berezinskii's first JETP paper identifies the anomalous low-temperature phase of 2D systems with continuous symmetry

1972

— Berezinskii's second paper treats the discrete vortex degrees of freedom; Kosterlitz and Thouless publish their first short letter in J. Phys. C

1973

— Kosterlitz and Thouless, "Ordering, metastability and phase transitions in two-dimensional systems" (J. Phys. C 6, 1181): the vortex-unbinding transition

1974

— Kosterlitz derives the renormalization-group recursion relations and the essential singularity of the correlation length

1977

— Nelson and Kosterlitz predict the universal superfluid-density jump 2/π

1978

— Bishop and Reppy confirm the universal jump in torsional-oscillator experiments on superfluid helium films

1979

— Halperin, Nelson and Young extend the theory to two-stage 2D melting (KTHNY)

2006

— Hadzibabic and collaborators observe BKT physics interferometrically in a trapped 2D atomic Bose gas

2016

— Nobel Prize in Physics to Thouless, Haldane and Kosterlitz for topological phase transitions and topological phases of matter

Going further

References. Berezinskii, Sov. Phys. JETP 32, 493 (1971); 34, 610 (1972) · Kosterlitz & Thouless, J. Phys. C 6, 1181 (1973) · Kosterlitz, J. Phys. C 7, 1046 (1974) · Nelson & Kosterlitz, PRL 39, 1201 (1977) · Bishop & Reppy, PRL 40, 1727 (1978) · Kardar, Statistical Physics of Fields, Ch. 8 · Cardy, Scaling and Renormalization, Ch. 6.

Provenance. A verifier should reproduce: (1) the vortex-energy integral $=\pi J\ln(L/a)$ symbolically; (2) the entropy coefficient $2$ ; (3) the sign change of $(\pi J-2k_BT)\ln(L/a)$ at $k_BT_{KT}=\pi J/2$ ; (4) the ratio $2/\pi=0.63662$ and endpoint $\eta=1/4$ at $K=2/\pi$ ; (5) the helium line $\sigma_s/T_c=2m^2k_B/\pi\hbar^2=3.49\times10^{-9}$ g cm⁻² K⁻¹; (6) that $e^{b/\sqrt{t}}$ outgrows any power of $1/t$ ; (7) Example 1's $T_{KT}=\pi/2$ K and jump $1.0$ K; (8) the spin-wave power law with $\eta=k_BT/2\pi J$ .

Builds on: The XY Model and Continuous Spins · Mermin-Wagner Theorem and Lower Critical Dimension · Correlation Length and Its Divergence. Leads to: Universality and Universality Classes · superfluid films and the universal jump · KTHNY theory of two-dimensional melting.

If there is no order parameter, in what sense is this a phase transition at all?

In the sharpest sense: a non-analyticity of the free energy in the thermodynamic limit separating qualitatively different phases. Below $T_{KT}$ correlations decay as a power law and the stiffness is finite; above, correlations decay exponentially and $\rho_s=0$ — not smoothly connectable, since the stiffness jumps. What is absent is only the Landau signature: an order parameter and a broken symmetry. The KT transition showed the Landau checklist is sufficient but not necessary for a genuine transition — precisely why it was revolutionary.

How can the low-temperature phase be a line of critical points?

A critical point is a state with scale-invariant, power-law correlations. The 2D XY model has $\langle\mathbf{s}(0)\cdot\mathbf{s}(r)\rangle\sim r^{-\eta(T)}$ at every temperature below $T_{KT}$ , with $\eta$ growing from $0$ toward $1/4$ — an interval of critical points, in RG language a fixed line. Exponents in this phase are non-universal (they depend on $T$ ); the endpoint $\eta(T_{KT})=1/4$ is universal.

Why do vortices matter when spin waves already destroy the order?

They do different jobs. Spin waves kill true long-range order (Mermin-Wagner) but are too gentle to produce exponential decay: they leave a power law at any temperature. Vortices are what spin waves cannot build — a $2\pi$ winding is a topological invariant. Only when free vortices proliferate does phase-scrambling appear at all scales, converting power-law into exponential decay. The transition is exactly where the defect sector takes over.

Why does the energy-entropy argument give a sharp transition rather than a crossover?

Because both terms scale as the same logarithm of system size. In the thermodynamic limit $F=(\pi J-2k_BT)\ln(L/a)$ is $+\infty$ or $-\infty$ by the sign of the coefficient alone: a free vortex is strictly forbidden below $T_{KT}$ and infinitely favourable above. Energy growing like $L$ against entropy $\ln L$ would forbid defects at all temperatures; two finite terms would give only a crossover. The logarithm-versus-logarithm competition is the whole trick.

Is the universal jump really universal across lattices and materials?

The location is non-universal — $k_BT_{KT}\approx0.893\,J$ on the square lattice, thickness-dependent in films. But the ratio $2/\pi$ contains no microscopic parameter: it comes from the marginality condition $\pi K=2$ , which knows only the long-wavelength form $\tfrac{\rho_s}{2}\int(\nabla\theta)^2$ . Any system coarse-graining to a 2D $U(1)$ phase field — helium films, superconducting films, 2D Bose gases — must show the same $2/\pi$ . Bishop and Reppy's data collapsed films of many thicknesses onto the single Nelson-Kosterlitz line.

What would I actually see in the specific heat at T_KT?

Almost nothing — and that is a prediction. The singular free energy scales as $\xi^{-2}\sim e^{-2b/\sqrt{T-T_{KT}}}$ , vanishing at $T_{KT}$ with all its derivatives: infinite order. The specific heat has only a smooth bump above $T_{KT}$ , where pair-ionization entropy is released. The transition is diagnosed instead through the stiffness (torsional oscillators, third sound), transport (the $V\sim I^a$ exponent), or vortex imaging in cold-atom interference.

How does this relate to two-dimensional melting?

Halperin, Nelson and Young ported the mechanism to 2D crystals, which host two defect types: dislocations (translational order) and disclinations (orientational order). KTHNY theory predicts melting in two KT-like steps — dislocation unbinding gives a "hexatic" phase with quasi-long-range orientational but short-range translational order; disclination unbinding then gives the liquid. Colloidal monolayers show both steps — the same logarithm competition run twice.