Kosterlitz-Thouless Transition: Topological Phase Transitions in Two Dimensions

Intuition Beginner

The Mermin-Wagner theorem delivers a stark verdict for two-dimensional systems: if your system has a continuous symmetry (like rotational symmetry), that symmetry cannot be spontaneously broken. No matter how cold you make a 2D magnet with continuous spin directions, the spins will never align. The thermal fluctuations in 2D are just too strong -- they grow without bound as the system size increases, washing out any tendency toward long-range order.

For decades this seemed to settle the question: no symmetry breaking means no phase transition in 2D systems with continuous symmetry. But in 1973, J. Michael Kosterlitz and David Thouless discovered that 2D systems can have phase transitions -- just not the kind Landau described. The key insight was that the Mermin-Wagner theorem forbids symmetry breaking, but it says nothing about topological defects.

Think of a 2D system of arrows (spins) that want to point in the same direction as their neighbours. In a smooth, well-behaved configuration, all arrows point roughly the same way and the system looks ordered locally. But there is another kind of excitation that cannot be smoothed away: a vortex. At the centre of a vortex, the arrows rotate through a full 360 degrees as you walk in a circle around it. You cannot remove a vortex by locally adjusting the arrows -- the "twist" is locked in by the topology of the configuration.

A single free vortex costs an energy that grows with the logarithm of the system size. In a small system, this energy is modest; in a large system, it is enormous. At low temperatures, free vortices are suppressed by this energy cost. But vortices can also appear in bound pairs: a vortex (counter-clockwise twist) paired with an antivortex (clockwise twist). The pair looks like a dipole -- the far-field disturbance cancels, and the pair's energy depends only on the distance between the two defects, not on the system size. At low temperature, the system is filled with tightly bound vortex-antivortex pairs that have little effect on the large-scale physics.

As you raise the temperature, thermal fluctuations pull these pairs apart. At a critical temperature $T_{\mathrm{KT}}$, the pairs unbind: vortices and antivortices break free and wander independently through the system. These free vortices disrupt the local order so severely that the correlation length -- the distance over which spins still "talk" to each other -- becomes finite. Below $T_{\mathrm{KT}}$, the correlation length is infinite (quasi-long-range order). Above $T_{\mathrm{KT}}$, it is finite (disordered). The system has undergone a phase transition without any symmetry being broken.

This is the Kosterlitz-Thouless (KT) transition. It was the first example of a topological phase transition: a transition driven not by the appearance of an order parameter but by the unbinding of topological defects. Kosterlitz and Thouless were awarded the 2016 Nobel Prize in Physics (shared with Duncan Haldane) for this discovery. The independent contribution of Vadim Berezinskii in 1971 is acknowledged in the frequent name "BKT transition."

Visual Beginner

The KT transition is best visualised through the behaviour of vortices in a 2D spin field.

Low temperature ($T<T_{\mathrm{KT}}$): The spin field is mostly smooth. Occasional vortex-antivortex pairs appear as tiny dipoles -- the spin twists clockwise around one defect and counter-clockwise around the other, and the two twists cancel at large distances. These pairs are tightly bound.

At $T_{\mathrm{KT}}$: The pairs begin to unbind. Vortices and antivortices separate to ever larger distances. The spin field becomes rough on progressively larger scales.

High temperature ($T>T_{\mathrm{KT}}$): Free vortices proliferate. The spin direction randomises over a finite distance. The correlation length $\xi$ is finite and decreases as temperature increases.

Worked example Beginner

A 2D XY magnet on a square lattice has spins that can point in any direction in the plane. Each spin interacts with its nearest neighbours, preferring alignment. The coupling strength is $J$.

The energy of a single isolated vortex at the centre of a system of linear size $L$ (with lattice spacing $a$) is approximately:

$$E_{\mathrm{vortex}}\approx \pi J\ln \left(\frac{L}{a}\right).$$

For a system with $L/a=1000$ (a thousand spins on a side) and $J=1$ (in units where $k_B=1$):

$E_{\mathrm{vortex}}\approx 3.14\times \ln (1000)\approx 3.14\times 6.91\approx 21.7$.

The entropy of a single vortex (the number of places it can sit) is $S_{\mathrm{vortex}}\approx \ln (L^2/a^2)=2\ln (L/a)$. For $L/a=1000$: $S\approx 2\times 6.91=13.8$.

The free energy of a single vortex is $F=E-TS=\pi J\ln (L/a)-T\cdot 2\ln (L/a)=(\pi J-2T)\ln (L/a)$. This is negative (favouring vortex proliferation) when $T>\pi J/2$. This crude estimate gives a transition temperature $T_{\mathrm{KT}}\approx \pi J/2$. The actual RG calculation gives $T_{\mathrm{KT}}\approx 0.893J$ for the square lattice, which is close to $\pi J/2\approx 1.57J$ in spirit but differs in detail because the RG accounts for the screening of vortex interactions by other vortex pairs.

Check your understanding Beginner

Formal definition Intermediate+

The 2D XY model

The XY model in two dimensions is defined by the Hamiltonian:

$$H=-J\sum _{⟨i,j⟩}\cos ({\theta }_i-{\theta }_j),$$

where ${\theta }_i\in [0,2\pi )$ is the angle of the spin at site $i$, $J>0$ is the ferromagnetic coupling, and the sum runs over nearest-neighbour pairs. Each spin is a unit vector in the plane, and the interaction energy is minimised when neighbouring spins are parallel.

In the continuum limit (valid for slowly varying configurations), the Hamiltonian becomes:

$$H=\frac{K}{2}\int d^{2}r|\nabla \theta (\mathbf{r}){|}^2,$$

where $K=J/a^2$ is the spin-wave stiffness (with $a$ the lattice spacing). This is the 2D classical elastic free energy for a phase field $\theta (\mathbf{r})$.

Spin waves and the Mermin-Wagner theorem

For configurations with no vortices, the angle field $\theta (\mathbf{r})$ is single-valued and can be expanded in Fourier modes. Each mode contributes a quadratic free energy $F_q=(K/2)q^2|{\theta }_q{|}^2$, and by equipartition $⟨|{\theta }_q{|}^2⟩=k_{B}T/(K{q}^2)$. The mean-square angular fluctuation between two points separated by distance $r$ is:

$$⟨(\theta (\mathbf{r})-\theta (0){)}^2⟩=\frac{k_{B}T}{\pi K}\ln \left(\frac{r}{a}\right).$$

The spin-spin correlation function is:

$$⟨\mathbf{s}(0)\cdot \mathbf{s}(r)⟩=⟨\cos (\theta (r)-\theta (0))⟩=\exp \left(-\frac{1}{2}⟨(\Delta \theta {)}^2⟩\right)={\left(\frac{r}{a}\right)}^{-\eta },$$

where the exponent is $\eta =k_{B}T/(2\pi K)$. This is a power-law decay, not long-range order (which would require the correlation to approach a constant at large $r$). The Mermin-Wagner theorem is satisfied: the fluctuations grow logarithmically, which is just enough to destroy true long-range order but not enough to produce exponential decay. This intermediate state is called quasi-long-range order (QLRO).

Vortex excitations

The angle field $\theta (\mathbf{r})$ need not be single-valued. A configuration contains a vortex at position ${\mathbf{r}}_0$ if the angle winds by $2\pi n$ ($n\in \mathbb{Z}$, the vorticity) around any closed loop enclosing ${\mathbf{r}}_0$:

$$\oint \nabla \theta \cdot d\mathbf{l}=2\pi n.$$

For a single vortex of unit vorticity ($n=\pm 1$) at the centre of a disc of radius $R$ (with core radius $a$), the angle field is $\theta (\mathbf{r})=n\varphi$ (where $\varphi$ is the polar angle) and the gradient is $\nabla \theta =n\hat{\varphi }/r$. The energy is:

$$E_{\mathrm{vortex}}=\frac{K}{2}{\int }_a^R\frac{1}{r^2}2\pi rdr=\pi K\ln \left(\frac{R}{a}\right).$$

The energy diverges logarithmically with system size, but so does the entropy: the vortex can sit at any of $(R/a{)}^2$ locations, giving $S=2\ln (R/a)$. The free energy is:

$$F_{\mathrm{vortex}}=(\pi K-2k_{B}T)\ln \left(\frac{R}{a}\right).$$

A single free vortex becomes favourable when $T>\pi K/(2k_B)$ -- this is the single-vortex estimate of $T_{\mathrm{KT}}$.

Vortex-antivortex pairs

A vortex at ${\mathbf{r}}_{+}$ and an antivortex at ${\mathbf{r}}_{-}$ separated by distance $d=|{\mathbf{r}}_{+}-{\mathbf{r}}_{-}|$ have a combined angle field $\theta ={\varphi }_{+}-{\varphi }_{-}$ (where ${\varphi }_{\pm }$ are the polar angles measured from the two defect positions). The interaction energy is:

$$E_{\mathrm{pair}}=2\pi K\ln \left(\frac{d}{a}\right)+E_{\mathrm{core}}^{+}+E_{\mathrm{core}}^{-},$$

where the logarithmic term is the mutual interaction (attractive, falling off with separation) and the core energies are finite constants of order $K$. The pair energy is independent of system size for $d\ll R$, so pairs can be created without the divergent energy cost of free vortices. At low temperature, the Boltzmann weight for pair creation at separation $d$ is:

$$\exp \left(-\frac{2\pi K}{k_{B}T}\ln \frac{d}{a}\right)={\left(\frac{d}{a}\right)}^{-2\pi K/(k_{B}T)}.$$

For $T$ well below $T_{\mathrm{KT}}$, the exponent is large, pairs are tightly bound, and their density is low. As $T$ increases toward $T_{\mathrm{KT}}$, pairs grow larger and more numerous. At $T_{\mathrm{KT}}$, the largest pairs unbind.

The Coulomb gas mapping

The partition function for vortices in the 2D XY model can be mapped onto the grand canonical partition function of a 2D Coulomb gas. Each vortex of charge $n_i=\pm 1$ at position ${\mathbf{r}}_i$ interacts with every other vortex through a logarithmic potential:

$$H_{\mathrm{Coulomb}}=-2\pi K\sum _{i<j}{n}_i{n}_j\ln \left(\frac{|{\mathbf{r}}_i-{\mathbf{r}}_j|}{a}\right)+E_{\mathrm{core}}\sum _i{n}_i^2.$$

This is the Hamiltonian of a 2D plasma of unit charges interacting via a $-\ln r$ potential (equivalent to the 2D Coulomb interaction, since the Green's function of the 2D Laplacian is $\ln r$). The vortex fugacity $y=e^{-E_{\mathrm{core}}/(k_{B}T)}$ controls the chemical potential for vortex creation.

At low temperature (large $K$, small $y$), the Coulomb gas is in its dielectric phase: charges form bound dipole pairs that screen weakly. At high temperature (small $K$, large $y$), the gas is in its plasma phase: charges are free and screen strongly. The KT transition is the dielectric-to-plasma transition of this Coulomb gas.

RG flow equations

Kosterlitz and Thouless derived the renormalization group equations for the Coulomb gas by integrating out vortex pairs of sizes between $a$ and $a{e}^{dl}$ (where $dl$ is the differential RG step). The two coupling constants are the inverse stiffness $K^{-1}$ (or equivalently the temperature in units of $J$) and the vortex fugacity $y$. The RG equations to leading order in $y$ are:

$$\frac{d{K}^{-1}}{dl}=4{\pi }^3{y}^2+O(y^4),$$ $$\frac{dy}{dl}=\left(2-\pi K\right)y+O(y^3).$$

The RG flow has the following structure:

For $K>K_c=2/\pi$ (i.e., $T<T_{\mathrm{KT}}$): The fugacity $y$ is irrelevant -- it flows to zero under coarse-graining. The stiffness $K$ flows to a finite fixed-point value $K^{\ast }$ on a line of fixed points. The system is in the quasi-long-range-ordered phase with power-law correlations.

For $K<K_c=2/\pi$ (i.e., $T>T_{\mathrm{KT}}$): The fugacity $y$ is relevant -- it flows to large values. Free vortices proliferate and screen the spin stiffness to zero. The system is in the disordered phase with exponential correlations.

At $K=K_c$: The fugacity is marginal. The RG equations predict a line of fixed points at $y=0$ for all $K\ge K_c$. This line of fixed points is the hallmark of the KT transition and is responsible for its unusual critical behaviour.

The essential singularity

Unlike conventional second-order phase transitions where the correlation length diverges as a power law $\xi \sim |T-T_c{|}^{-\nu }$, the KT transition produces an essential singularity:

$$\xi \sim a\exp \left(\frac{b}{\sqrt{T/T_{\mathrm{KT}}-1}}\right),T\to T_{\mathrm{KT}}^{+},$$

where $b$ is a constant of order unity. The correlation length diverges faster than any power of $(T-T_{\mathrm{KT}})$ as $T\to T_{\mathrm{KT}}$ from above. This is one of the most distinctive predictions of the KT theory: there are no power-law critical exponents. All derivatives of the free energy are finite at $T_{\mathrm{KT}}$ -- the transition is sometimes described as "infinite order" because every derivative of the free energy is continuous.

Below $T_{\mathrm{KT}}$, the correlation length is infinite (power-law correlations persist to arbitrarily large distances), but the power-law exponent $\eta =k_{B}T/(2\pi K)$ varies continuously with temperature along the line of fixed points. This continuous variation of the exponent is another unique feature of the KT transition, impossible in conventional critical phenomena where exponents are universal constants.

The Nelson-Kosterlitz universal jump

Nelson and Kosterlitz (1977) showed that the spin-wave stiffness (equivalent to the superfluid density in a 2D superfluid) has a universal discontinuous jump at $T_{\mathrm{KT}}$:

$$\frac{{\rho }_s(T_{\mathrm{KT}}^{-})}{T_{\mathrm{KT}}}=\frac{2m{k}_B}{\pi {\hslash }^2},$$

where ${\rho }_s$ is the superfluid mass density and $m$ is the particle mass. In the spin language, this becomes $\pi K(T_{\mathrm{KT}}^{-})=2$, i.e., the stiffness jumps from $K_c=2/\pi$ just below $T_{\mathrm{KT}}$ to zero just above. The jump is universal: its magnitude depends only on fundamental constants, not on microscopic details. This provides a sharp, experimentally testable prediction.

Key derivation: Vortex free energy and the single-vortex estimate of $T_{\mathrm{KT}}$ Intermediate+

Proposition. The free energy of a single vortex in a 2D system of linear size $L$ with spin-wave stiffness $K$ is $F=(\pi K-2k_{B}T)\ln (L/a)$. A single free vortex becomes thermodynamically favourable when $k_{B}T>\pi K/2$.

Proof. The energy of a single vortex of unit vorticity in a disc of radius $R\approx L/2$ with core radius $a$ is $E=\pi K\ln (R/a)$. Up to an additive constant of order unity, $E\approx \pi K\ln (L/a)$.

The entropy: the vortex can be placed at any lattice site, giving $\Omega \approx (L/a{)}^2$ positions. The entropy is $S=k_B\ln \Omega =2k_B\ln (L/a)$.

The free energy is:

$$F=E-TS=\pi K\ln \left(\frac{L}{a}\right)-2k_{B}T\ln \left(\frac{L}{a}\right)=(\pi K-2k_{B}T)\ln \left(\frac{L}{a}\right).$$

For $F>0$ ($T<\pi K/(2k_B)$): free vortices are suppressed. The Boltzmann factor $e^{-F/(k_{B}T)}$ vanishes as $L\to \infty$, so free vortices are absent in the thermodynamic limit.

For $F<0$ ($T>\pi K/(2k_B)$): free vortices are favoured. The Boltzmann factor diverges as $L\to \infty$, and the single-vortex approximation breaks down because vortices interact and screen each other.

The transition estimate is $k_B{T}_{\mathrm{KT}}^{(0)}=\pi K/2$. This estimate neglects the screening of vortex interactions by other pairs, which renormalises $K$ and shifts $T_{\mathrm{KT}}$ downward. The full RG treatment gives $k_B{T}_{\mathrm{KT}}=(\pi /2)K_R(T_{\mathrm{KT}}^{-})$ where $K_R$ is the renormalised stiffness satisfying $\pi K_R/2=k_B{T}_{\mathrm{KT}}$, consistent with the Nelson-Kosterlitz condition. $\square$

Bridge. This single-vortex estimate is the KT analogue of the mean-field estimate for conventional transitions 11.06.02: it captures the correct physics (vortex unbinding) but gives an imprecise value of $T_{\mathrm{KT}}$ because it neglects the interactions between vortices. The full RG treatment 11.07.01 corrects this by treating vortex pairs systematically, just as the Wilson RG corrects mean-field theory by treating fluctuations systematically.

Exercises Intermediate+

Lean formalization Intermediate+

A formal treatment of the KT transition requires: (1) the 2D XY model Hamiltonian and its continuum limit; (2) the homotopy classification of maps $S^1\to S^1$ via the winding number (Mathlib has the fundamental group of the circle, ${\pi }_1(S^1)\cong \mathbb{Z}$, which classifies the vortices); (3) the logarithmic vortex energy derived from the 2D Laplacian Green's function; (4) the Coulomb gas partition function as a grand canonical ensemble of interacting point charges; (5) the RG flow equations and their fixed-point structure; (6) the essential singularity of the correlation length. Steps (1)-(3) are within reach given Mathlib's existing topology and analysis, but the RG derivation (steps 4-6) requires a formal perturbative RG framework that does not exist. The gap is substantial but well-defined.

Advanced results Master

Detailed RG derivation: the Coulomb gas

The full derivation of the KT RG equations proceeds by mapping the vortex degrees of freedom onto a neutral Coulomb gas and then applying the renormalization group to the gas. The starting point is the XY partition function in the Villain approximation, which replaces $\cos ({\theta }_i-{\theta }_j)$ by a Gaussian:

$$Z=\int \mathcal{D}\theta \exp \left(-\frac{K}{2}\int d^{2}r|\nabla \theta {|}^2\right)\times \sum _{\{n(\mathbf{r})\}}\exp \left(-\frac{\pi K}{2}\int d^{2}r{d}^2{r}^{\prime }n(\mathbf{r})G(\mathbf{r}-{\mathbf{r}}^{\prime })n({\mathbf{r}}^{\prime })\right),$$

where $n(\mathbf{r})\in \mathbb{Z}$ is the vorticity at position $\mathbf{r}$ (zero almost everywhere, $\pm 1$ at vortex/antivortex positions), and $G(\mathbf{r})=-\ln (|\mathbf{r}|/a)/(2\pi )$ is the 2D Green's function satisfying ${\nabla }^{2}G=-\delta (\mathbf{r})$. The neutrality constraint $\int n(\mathbf{r})d^{2}r=0$ ensures the angular field remains well-defined at infinity.

The Coulomb gas Hamiltonian is:

$$\frac{H_{\mathrm{CG}}}{k_{B}T}=-2\pi K\sum _{i<j}{n}_i{n}_j\ln \left(\frac{|{\mathbf{r}}_i-{\mathbf{r}}_j|}{a}\right)+\mu \sum _i{n}_i^2,$$

where $\mu =E_{\mathrm{core}}/(k_{B}T)$ is the vortex core energy and the fugacity is $y=e^{-\mu }$.

The RG proceeds by splitting the vortex pair-size distribution into shells. Pairs with separation between $a$ and $a{e}^{dl}$ are integrated out, producing a differential change in the effective coupling constants. The calculation yields:

$$\frac{d{K}^{-1}}{dl}=4{\pi }^3{y}^2{K}^{-2}+O(y^4),$$ $$\frac{dy}{dl}=\left(2-\pi K\right)y+O(y^3).$$

The first equation says that vortex pairs reduce the effective stiffness (they screen the spin-wave interaction, just as dielectric dipoles screen the electric field). The second equation says that vortices are relevant (grow under RG) when the stiffness is below the critical value $\pi K_c=2$, and irrelevant above it.

The Nelson-Kosterlitz universal jump: detailed derivation

The superfluid density ${\rho }_s$ in a 2D superfluid is related to the spin-wave stiffness $K$ by ${\rho }_s=(m^2/{\hslash }^2)k_{B}TK$ (in appropriate units). The Nelson-Kosterlitz condition $\pi K(T_{\mathrm{KT}}^{-})=2$ translates to:

$$\frac{{\rho }_s(T_{\mathrm{KT}}^{-})}{T_{\mathrm{KT}}}=\frac{2m{k}_B}{\pi {\hslash }^2}.$$

The jump can be derived by noting that the RG equations conserve a quantity related to the helicity modulus. The helicity modulus ${\rm Y}$ (the response of the free energy to a twist in the boundary conditions) equals ${\rho }_s$ for the superfluid. At $T_{\mathrm{KT}}$, the RG flows along the separatrix that connects the critical fixed point to the high-temperature phase, and the helicity modulus at the transition satisfies ${\rm Y}(T_{\mathrm{KT}}^{-})=(2/\pi )k_B{T}_{\mathrm{KT}}$. Above $T_{\mathrm{KT}}$, the renormalised helicity modulus drops to zero because free vortices screen the spin stiffness completely.

The universality of the jump follows from the universality of the RG equations: all 2D systems with the same topological defect structure (logarithmically interacting vortices in 2D) share the same RG equations and hence the same jump condition.

Connections to 2D superconductivity and superfluid films

The KT transition is directly relevant to thin superconducting films. A 2D superconducting film has a complex order parameter $\psi =|\psi |e^{i\theta }$, and the phase fluctuations are described by the same XY model. The vortices in the superconducting film are magnetic flux quanta (fluxoids), each carrying a flux ${\Phi }_0=h/(2e)$. The logarithmic interaction between fluxoids is the same as between XY vortices, with the stiffness replaced by the superfluid sheet density.

A thin superconducting film above its mean-field $T_c$ but below the KT temperature has no long-range phase coherence (no true superconductivity) but retains local pairing. The KT transition in such films has been observed through measurements of the sheet resistance, which drops to zero at $T_{\mathrm{KT}}$.

Josephson junction arrays provide another physical realisation. A regular array of Josephson junctions on a 2D lattice has a Hamiltonian $H=-E_J{\sum }_{⟨i,j⟩}\cos ({\theta }_i-{\theta }_j)$, which is precisely the XY model with $J=E_J$. The KT transition in these arrays has been extensively studied, with the vortex-unbinding transition observed through transport measurements.

Experimental verification: Bishop and Reppy (1978)

The definitive experimental test of the KT theory was performed by Bishop and Reppy using torsional oscillators containing thin films of liquid helium-4 adsorbed on substrates. The superfluid fraction of the film contributes an inertial anomaly to the oscillator: the superfluid component does not rotate with the substrate, reducing the moment of inertia.

Bishop and Reppy measured the superfluid density ${\rho }_s(T)$ as a function of temperature for films of various thicknesses. Their key findings were:

1. ${\rho }_s(T)$ decreased continuously as $T$ increased toward $T_{\mathrm{KT}}$, then dropped abruptly to zero at $T_{\mathrm{KT}}$.
2. The ratio ${\rho }_s(T_{\mathrm{KT}}^{-})/T_{\mathrm{KT}}$ was the same for all film thicknesses, within experimental error.
3. The universal value agreed with the Nelson-Kosterlitz prediction ${\rho }_s(T_{\mathrm{KT}}^{-})/T_{\mathrm{KT}}=2m_4{k}_B/(\pi {\hslash }^2)$, where $m_4$ is the helium-4 atomic mass.

This agreement between theory and experiment is one of the most convincing verifications of a critical-phenomena prediction in condensed matter physics.

Nobel Prize 2016 and historical context

The Nobel Prize in Physics 2016 was awarded to David Thouless, Duncan Haldane, and Michael Kosterlitz "for theoretical discoveries of topological phase transitions and topological phases of matter." The prize citation specifically highlighted the 1973 Kosterlitz-Thouless paper as the discovery of the first topological phase transition.

The historical context is important. Berezinskii's 1971 paper (published in Soviet physics journals) independently established the vortex-unbinding mechanism, though without the full RG treatment that Kosterlitz and Thouless provided. Berezinskii's untimely death in 1980 at age 44 prevented his participation in the later developments. The transition is frequently called the "BKT transition" in recognition of both contributions.

The KT transition opened the door to the broader concept of topological order: phases of matter characterised not by local order parameters and symmetry breaking, but by global topological properties. This concept has since found applications in the quantum Hall effect, topological insulators, topological superconductors, and quantum spin liquids. Haldane's contribution to the 2016 Nobel (the Haldane gap in 1D spin chains) represents another facet of topological physics in condensed matter.

Limitations and extensions

The KT theory has several limitations and has been extended in important directions:

1. Higher-order corrections. The RG equations given above are the leading-order results in an expansion in the vortex fugacity $y$. Higher-order terms (from multi-vortex processes) have been computed and produce small corrections to $T_{\mathrm{KT}}$ and the correlation-length exponent $b$.

2. Lattice effects. The continuum Coulomb gas mapping neglects the lattice structure, which produces a sixfold anisotropy (on a triangular lattice) or fourfold anisotropy (on a square lattice) in the spin stiffness. These anisotropies are irrelevant at the KT fixed point but affect the approach to the asymptotic regime.

3. Finite-size effects. In a finite system of size $L$, the essential singularity is cut off at $\xi =L$, and the transition is rounded. The finite-size scaling of the KT transition is unusual: $T_{\mathrm{KT}}(L)-T_{\mathrm{KT}}(\infty )\sim 1/(\ln L{)}^2$.

4. Three dimensions and beyond. The KT mechanism does not apply in $d\ge 3$ because the vortex energy is finite (not logarithmic) and vortices are always present as finite-energy excitations. The 3D XY model has a conventional second-order transition with power-law critical behaviour.

5. Quantum KT transitions. The KT mechanism can appear in quantum phase transitions at zero temperature in 2+1 dimensions, where the imaginary time direction plays the role of a third spatial dimension. These quantum KT transitions share the essential singularity and the line of fixed points with the classical case.

Synthesis. The Kosterlitz-Thouless transition is the paradigmatic example of a topological phase transition. It demonstrates that the Mermin-Wagner theorem, while forbidding symmetry breaking in 2D, does not forbid all phase transitions. The mechanism -- unbinding of topological defects -- is fundamentally different from the Landau symmetry-breaking paradigm 11.06.03. The RG treatment 11.07.01 reveals a line of fixed points, an essential singularity, and a universal jump in the superfluid density -- phenomena that have no counterpart in conventional critical phenomena. The experimental confirmation by Bishop and Reppy stands as a landmark test of theoretical physics, and the 2016 Nobel Prize recognised the profound influence of this work on the subsequent development of topological phases of matter.

Full proof set Master

Proposition. The Kosterlitz-Thouless RG equations $d{K}^{-1}/dl=4{\pi }^3{y}^2$ and $dy/dl=(2-\pi K)y$ possess a line of fixed points at $y=0$, $K\ge 2/\pi$. The RG eigenvalue $\lambda =2-\pi K$ changes sign at $K=2/\pi$, making the fugacity relevant for $K<2/\pi$ and irrelevant for $K>2/\pi$. The correlation length above $T_{\mathrm{KT}}$ diverges with an essential singularity $\xi \sim \exp (b/\sqrt{T-T_{\mathrm{KT}}})$.

Proof. The fixed-point equations $d{K}^{-1}/dl=0$ and $dy/dl=0$ are satisfied by $y=0$ for any $K$. This is a line of fixed points parametrised by $K$.

Linearise around the fixed point $(K^{\ast },0)$ by writing $K=K^{\ast }+\delta K$, $y=\delta y$. The linearised equations are:

$$\frac{d(\delta K^{-1})}{dl}=0,\frac{d(\delta y)}{dl}=(2-\pi K^{\ast })\delta y.$$

The first equation says $\delta K$ is marginal (to linear order). The eigenvalue for the $y$ direction is ${\lambda }_y=2-\pi K^{\ast }$. For $K^{\ast }>2/\pi$: ${\lambda }_y<0$ (irrelevant), and $y$ flows back to zero. For $K^{\ast }<2/\pi$: ${\lambda }_y>0$ (relevant), and $y$ grows.

For the essential singularity, rescale variables: let $x=\pi K/2-1$ (so $x=0$ at the critical point) and absorb constants into $y$. The RG equations become $dx/dl=-C{y}^2$ and $dy/dl=-xy$ (with $C>0$). The constant of motion is $x^2-C{y}^2=\text{const}$ (verify by computing $d(x^2-C{y}^2)/dl=2x(-C{y}^2)-2Cy(-xy)=-2Cx{y}^2+2Cx{y}^2=0$).

For $T$ slightly above $T_{\mathrm{KT}}$: the initial condition is $x_0\approx -\alpha (T-T_{\mathrm{KT}})$ (small, negative) and $y_0$ small. The trajectory follows the hyperbola $x^2-C{y}^2=x_0^2-C{y}_0^2\approx x_0^2<0$, i.e., $C{y}^2-x^2=|x_0^2|$.

The RG time $l^{\ast }$ to reach $y\sim 1$ is:

$$l^{\ast }\sim {\int }_{y_0}^1\frac{dy}{|x|y}={\int }_{y_0}^1\frac{dy}{y\sqrt{C{y}^2-|x_0^2|}}.$$

For $|x_0|\ll 1$, the integral is dominated by the region near $y=0$ where $|x|\approx |x_0|$, giving $l^{\ast }\sim |\ln y_0|/|x_0|$. More precisely, for the full integral, the substitution $u=C{y}^2$ gives:

$$l^{\ast }\sim \frac{1}{\sqrt{C}}\int \frac{du}{2u\sqrt{u-|x_0^2|}}\sim \frac{1}{|x_0|}\sim \frac{1}{\sqrt{T-T_{\mathrm{KT}}}}.$$

Since $\xi =a{e}^{l^{\ast }}$:

$$\xi \sim a\exp \left(\frac{b}{\sqrt{T/T_{\mathrm{KT}}-1}}\right),$$

with $b$ a non-universal constant of order unity. $\square$

Proposition (Nelson-Kosterlitz universal jump). At the KT transition, the helicity modulus ${\rm Y}$ (equivalent to the spin-wave stiffness in the spin language and to the superfluid density in the superfluid language) jumps discontinuously from ${\rm Y}(T_{\mathrm{KT}}^{-})=(2/\pi )k_B{T}_{\mathrm{KT}}$ to ${\rm Y}(T_{\mathrm{KT}}^{+})=0$.

Proof. The helicity modulus is defined as the change in free energy when a twist $\Delta \theta$ is applied across the system: ${\rm Y}={\partial }^{2}F/\partial (\Delta \theta {)}^2{|}_{\Delta \theta =0}$. For the XY model without vortices (spin-wave approximation), ${\rm Y}=K$.

Including vortex pairs, the helicity modulus is renormalised by the screening effect of bound pairs. A pair of separation $d$ contributes a polarisability that reduces the effective stiffness. The RG equation for the stiffness $d{K}^{-1}/dl=4{\pi }^3{y}^2$ precisely captures this screening.

At $T_{\mathrm{KT}}$, the RG flow starts at the critical fixed point $(K_c,0)$ with $K_c=2/\pi$. Since $y$ flows to zero for $K\ge K_c$, the renormalised stiffness just below $T_{\mathrm{KT}}$ is:

$$K_R(T_{\mathrm{KT}}^{-})=K_c=\frac{2}{\pi }.$$

Just above $T_{\mathrm{KT}}$, the RG flows to the high-temperature fixed point where $y$ is large and the stiffness is fully screened: $K_R(T_{\mathrm{KT}}^{+})=0$. The jump is:

$$\Delta K_R=K_R(T_{\mathrm{KT}}^{-})-K_R(T_{\mathrm{KT}}^{+})=\frac{2}{\pi }.$$

In superfluid units, ${\rho }_s=(m^2{k}_{B}T/{\hslash }^2)K$, so the jump condition $\pi K(T_{\mathrm{KT}}^{-})=2$ becomes ${\rho }_s(T_{\mathrm{KT}}^{-})/T_{\mathrm{KT}}=2m{k}_B/(\pi {\hslash }^2)$. $\square$

Connections Master

• 11.06.01 The Ising model in 2D has a conventional phase transition with discrete symmetry breaking (${\mathbb{Z}}_2$), which is not forbidden by Mermin-Wagner. The XY model in 2D has continuous $O(2)$ symmetry, which Mermin-Wagner forbids from breaking -- the KT transition is the alternative.
• 11.06.03 Landau theory describes phase transitions through symmetry breaking and an order parameter. The KT transition has no local order parameter and no symmetry breaking -- it lies entirely outside the Landau paradigm. This was the first example of a phase transition that Landau theory cannot describe.
• 11.06.04 The Ginzburg criterion identifies $d=4$ as the upper critical dimension for Ising-type transitions. The Mermin-Wagner theorem is an even stronger constraint for continuous symmetries: in $d\le 2$, Landau theory does not just fail near $T_c$ -- the ordered phase does not exist at all. The KT transition circumvents this constraint through topology.
• 11.07.01 The Wilson RG framework 11.07.01 describes conventional critical points with isolated fixed points and power-law scaling. The KT transition introduces a new RG phenomenology: a line of fixed points, a marginal operator that becomes relevant at a critical stiffness, and an essential singularity. This expands the taxonomy of RG fixed points beyond the Wilson-Fisher type.
• 11.07.02 Block-spin renormalization of the 2D XY model in the spin-wave sector reproduces the line of fixed points. The full RG treatment requires including the vortex sector (the Coulomb gas mapping), which goes beyond the spin-wave block-spin analysis.

Historical and philosophical context Master

The Kosterlitz-Thouless transition occupies a unique position in the history of statistical mechanics: it was the first phase transition that could not be understood within the Landau symmetry-breaking framework, and it introduced the concept of topological phase transitions that has since become central to condensed matter physics.

Mermin and Wagner (1966). The Mermin-Wagner theorem proved that continuous symmetries cannot be spontaneously broken in one- and two-dimensional systems at finite temperature. The proof relies on the infrared divergence of Goldstone mode fluctuations in $d\le 2$: the fluctuation $⟨(\theta (r)-\theta (0){)}^2⟩\sim \ln (r)$ grows without bound, preventing the establishment of long-range order. For a decade after this result, it was widely assumed that 2D systems with continuous symmetry had no phase transitions at all.

Berezinskii (1971). Vadim L'vovich Berezinskii, working at the Landau Institute for Theoretical Physics in Moscow, was the first to recognise that vortex-like excitations could mediate a phase transition in 2D systems. His 1971 paper ("Destruction of long-range order in one-dimensional and two-dimensional systems with a continuous symmetry group," Sov. Phys. JETP 34, 610) analysed the role of topological defects and identified the unbinding mechanism. Berezinskii's treatment was more physical than mathematical; he identified the essential physics but did not derive the full RG equations.

Kosterlitz and Thouless (1973). J. Michael Kosterlitz and David Thouless, working at the University of Birmingham, published their landmark paper in 1973 ("Ordering, metastability and phase transitions in two-dimensional systems," J. Phys. C 6, 1181). Their contribution was threefold: (1) they mapped the vortex problem onto a 2D Coulomb gas, making the connection to a well-studied statistical mechanics problem; (2) they derived the RG equations for the Coulomb gas, showing that the transition is governed by the interplay between the vortex fugacity and the spin stiffness; and (3) they identified the essential singularity of the correlation length, establishing that the transition has no power-law critical exponents.

The Coulomb gas mapping was the key technical insight. The logarithmic interaction between vortices is identical to the 2D Coulomb interaction between point charges. The dielectric-to-plasma transition in the Coulomb gas was already understood, and Kosterlitz and Thouless recognised that this transition maps onto the vortex-unbinding transition in the XY model. This mapping allowed them to import techniques from plasma physics into condensed matter.

Nelson and Kosterlitz (1977). The prediction of the universal jump in the superfluid density transformed the KT transition from a theoretical curiosity into an experimentally testable prediction. The universality of the jump -- its independence from microscopic details -- made it a sharp test: any deviation would falsify the theory.

Bishop and Reppy (1978). The experimental confirmation using torsional oscillators and thin helium-4 films was a triumph of precision measurement. The agreement between the measured and predicted values of ${\rho }_s(T_{\mathrm{KT}}^{-})/T_{\mathrm{KT}}$ was within a few percent, remarkable for a prediction involving only fundamental constants.

Nobel Prize (2016). The Nobel Prize citation recognised not just the KT transition itself but the broader concept it introduced: that topological defects can drive phase transitions independently of symmetry breaking. This idea has since been generalised to topological phases of matter (topological insulators, quantum Hall states, topological superconductors) and has become one of the organising principles of modern condensed matter physics. The 2016 prize was shared with Duncan Haldane for his work on topological phases in one-dimensional spin chains.

Philosophical significance. The KT transition challenged the deeply held assumption that phase transitions require symmetry breaking. It demonstrated that the space of possible phases is richer than the Landau classification suggests: there exist phases (quasi-long-range order) and transitions (topological unbinding) that have no description in terms of local order parameters and symmetry groups. This insight has driven the development of topological quantum field theory and the classification of topological phases of matter, which now form a major branch of theoretical physics.

Bibliography Master

@article{kosterlitz1973,
  author = {Kosterlitz, J. M. and Thouless, D. J.},
  title = {Ordering, metastability and phase transitions in two-dimensional systems},
  journal = {J. Phys. C: Solid State Phys.},
  volume = {6},
  pages = {1181--1203},
  year = {1973}
}

@article{berezinskii1971,
  author = {Berezinskii, V. L.},
  title = {Destruction of long-range order in one-dimensional and two-dimensional systems with a continuous symmetry group},
  journal = {Sov. Phys. JETP},
  volume = {34},
  pages = {610--616},
  year = {1971}
}

@article{berezinskii1972,
  author = {Berezinskii, V. L.},
  title = {Destruction of long-range order in one-dimensional and two-dimensional systems possessing a continuous symmetry group. {II}. {Q}uantum systems},
  journal = {Sov. Phys. JETP},
  volume = {34},
  pages = {617--623},
  year = {1972}
}

@article{nelson1977,
  author = {Nelson, David R. and Kosterlitz, J. M.},
  title = {Universal Jump in the Superfluid Density of a Two-Dimensional Superfluid},
  journal = {Phys. Rev. Lett.},
  volume = {39},
  pages = {1201--1205},
  year = {1977}
}

@article{bishop1978,
  author = {Bishop, D. J. and Reppy, J. D.},
  title = {Study of the Superfluid Transition in Two-Dimensional {He}-4 Films},
  journal = {Phys. Rev. Lett.},
  volume = {40},
  pages = {1727--1730},
  year = {1978}
}

@article{mermin1966,
  author = {Mermin, N. D. and Wagner, H.},
  title = {Absence of Ferromagnetism or Antiferromagnetism in One- or Two-Dimensional Isotropic {H}eisenberg Models},
  journal = {Phys. Rev. Lett.},
  volume = {17},
  pages = {1133--1136},
  year = {1966}
}

@article{kosterlitz1974,
  author = {Kosterlitz, J. M.},
  title = {The critical properties of the two-dimensional {xy} model},
  journal = {J. Phys. C: Solid State Phys.},
  volume = {7},
  pages = {1046--1060},
  year = {1974}
}

@book{kardar2007,
  author = {Kardar, Mehran},
  title = {Statistical Physics of Fields},
  publisher = {Cambridge University Press},
  year = {2007}
}

@book{chaikin1995,
  author = {Chaikin, P. M. and Lubensky, T. C.},
  title = {Principles of Condensed Matter Physics},
  publisher = {Cambridge University Press},
  year = {1995}
}

@article{jose1977,
  author = {Jos\'{e}, Jorge V. and Kadanoff, Leo P. and Kirkpatrick, Scott and Nelson, David R.},
  title = {Renormalization, vortices, and symmetry-breaking perturbations in the two-dimensional planar model},
  journal = {Phys. Rev. B},
  volume = {16},
  pages = {1217--1241},
  year = {1977}
}