The Berezinskii–Kosterlitz–Thouless transition is a phase transition in which bound pairs of topological defects separate and proliferate, destroying a form of order sustained by correlations over large distances. Its standard realization occurs in two-dimensional systems with a continuous angular degree of freedom, such as planar spins or the phase of a superfluid. Below the transition, correlations decay algebraically rather than approaching a nonzero constant; above it, they decay exponentially. The transition therefore does not require the conventional appearance or disappearance of a nonzero local order parameter.
Named after Vadim Berezinskii, J. Michael Kosterlitz, and David J. Thouless, the transition established that topology can govern collective behavior in statistical mechanics. Its characteristic features include vortex–antivortex unbinding, an exponentially diverging correlation length, and a universal discontinuity in phase stiffness. These ideas describe thin helium films, two-dimensional atomic gases, superconducting films, and several lattice models, while also informing theories of two-dimensional melting and other defect-mediated transitions.
Historical development and conceptual significance
Conventional theories of phase transitions commonly distinguish phases by symmetry and a local order parameter. A ferromagnet, for example, develops a spontaneous magnetization below its ordering temperature. Applying this picture to two-dimensional systems with continuous rotational symmetry creates a difficulty: long-wavelength fluctuations are strong enough to prevent conventional long-range order under broad conditions. The Mermin–Wagner theorem, established in 1966, made this restriction precise for important classes of systems with sufficiently short-range interactions. Related arguments by Pierre Hohenberg addressed the absence of ordinary Bose condensation in uniform low-dimensional fluids at nonzero temperature.
The absence of conventional order did not, however, imply the absence of a thermodynamic transition. In papers published in 1971 and 1972, Berezinskii argued that certain two-dimensional systems could possess a low-temperature phase with algebraically decaying correlations. This phase would differ sharply from the high-temperature phase even though neither had the conventional long-range order associated with spontaneous alignment. His work identified the importance of defects and of a finite response to a slowly imposed twist.
Kosterlitz and Thouless developed the defect-unbinding account in work published in 1972 and especially in their 1973 paper on ordering, metastability, and phase transitions in two dimensions. Kosterlitz’s 1974 analysis supplied a systematic scaling description of the transition. In this picture, small vortex pairs are present below the transition, but isolated vortices become thermodynamically favorable only when thermal fluctuations overcome their long-range binding. The decisive distinction between phases is therefore not whether local angles fluctuate, but whether topological defects remain bound on large scales.
Further developments connected this theory to measurable quantities. In 1977, David R. Nelson and Kosterlitz derived the universal jump in superfluid density, giving experiments a particularly sharp test. Thin-helium-film measurements subsequently provided important evidence for this prediction. The 2016 Nobel Prize in Physics was awarded to Thouless, F. Duncan M. Haldane, and Kosterlitz for theoretical discoveries of topological phase transitions and topological phases of matter; Berezinskii had died in 1980.
The BKT transition does not invalidate symmetry-based descriptions. Rather, it shows that the absence of ordinary spontaneous symmetry breaking does not exclude distinct thermodynamic phases. Local fluctuations, long-distance rigidity, and the topology of defects must all be considered.
The planar model and algebraic order
The canonical microscopic example is the two-dimensional XY model. Each lattice site carries a unit vector specified by an angle , and neighboring vectors favor alignment. On a square lattice its Hamiltonian is
where is the nearest-neighbor coupling and denotes neighboring pairs. The energy is unchanged if every angle is shifted by the same amount, expressing a global continuous symmetry. Unlike the discrete spins of the Ising model, XY spins can rotate through arbitrarily small angles. Those low-energy rotations are crucial to the model’s long-distance behavior.
At low temperature, neighboring angles usually differ only slightly. Expanding the cosine gives a continuum free energy,
Here is the phase stiffness, with dimensions of energy in two dimensions. It measures the cost of spatially twisting the angle field. The subscript “sw” denotes the spin-wave approximation, which retains smooth fluctuations but omits vortices. In a superfluid, the corresponding angle is the phase of the macroscopic complex field, and stiffness measures resistance to a spatial phase twist rather than resistance to mechanical deformation.
The central diagnostic is the correlation function
For the Gaussian spin-wave theory, long-wavelength fluctuations produce
where is a microscopic cutoff and is the Boltzmann constant. Consequently,
In the interacting system, the long-distance stiffness must be used in this relation. The logarithmically increasing phase fluctuations make the thermodynamic-limit magnetization vanish, but correlations decay much more slowly than they do in a conventional disordered phase.
This behavior is called quasi-long-range order. There is no finite correlation length in the usual exponential-decay sense, and the exponent varies continuously with temperature. Thus the entire low-temperature phase is scale-free, rather than scale invariance occurring only at a single critical temperature. The standard short-range XY model satisfies the assumptions behind the prohibition of continuous-symmetry breaking; its algebraic order is compatible with that prohibition.
The compact nature of the angle, , is essential. A purely noncompact Gaussian field reproduces spin waves but lacks the integer vortices that terminate the algebraically ordered phase. The transition cannot therefore be understood from the quadratic approximation alone.
Vortices and the mechanism of unbinding
A vortex is a topological defect around which the phase winds by an integer multiple of . For a closed contour enclosing the defect,
is an integer called the winding number or vortex charge. A vortex with and an antivortex with have opposite circulation. Smooth local changes cannot eliminate a single winding while preserving a well-defined phase everywhere along the surrounding contour. Annihilation with an opposite defect, escape through a boundary, or suppression of the field amplitude is required.
Far from its core, a vortex has . Substituting this into the continuum energy gives
where is a core energy and is the system size. The logarithm results from integrating an energy density proportional to over a two-dimensional area. Defects of larger winding are especially expensive because the energy grows as ; elementary charges therefore control the standard transition.
The energetic penalty competes with entropy. A vortex core can occupy approximately positions, giving a positional entropy . The leading size-dependent contribution to its Helmholtz free energy is consequently
This argument suggests that free vortices become favorable when thermal disorder overwhelms stiffness. Taken literally with a microscopic stiffness, it is only an estimate: other vortex pairs screen the interaction and modify the stiffness observed at large distances. Using the properly renormalized stiffness leads to the universal transition condition.
At low temperatures, opposite charges form bound pairs whose interaction grows logarithmically with their separation. Small pairs occur at any nonzero temperature, so the transition is not the first appearance of vortices. Instead, the distribution of pair sizes changes until pairs can separate on arbitrarily large scales. Above the transition, unbound vortices screen one another’s long-range fields and destroy algebraic phase coherence.
This defect ensemble can be mapped to a neutral two-dimensional Coulomb gas, in which vortex charges interact through a logarithmic potential. Bound dipoles polarize in response to distant defects and reduce the effective interaction between them. The feedback between increased screening and easier pair separation drives unbinding. “Free vortex” in this context means unbound on the relevant observation scale, not a vortex with literally zero interaction energy.
Renormalization-group description
The renormalization group describes how the effective theory changes as progressively larger length scales are examined. For the dilute vortex gas, useful variables are the dimensionless stiffness
and the vortex fugacity , which is approximately proportional to at the microscopic scale. If is the logarithm of the length rescaling factor, a conventional normalization gives the leading flow equations
Numerical coefficients involving fugacity depend on its definition, but the physical structure of these equations does not.
The first equation states that vortex pairs reduce the effective stiffness through screening. The second compares the two-dimensional positional scaling of a defect with the energetic cost of its winding. When , a sufficiently small fugacity decreases under coarse-graining: isolated vortices become less important at larger distances. When , fugacity grows and the dilute description eventually fails as the system enters the vortex-disordered regime. It is the coupled flow, rather than either equation separately, that locates the boundary between the phases.
Below the transition, the flow approaches a line of fixed points with and finite . Different points along this line have different algebraic correlation exponents. At the transition, the critical trajectory approaches the endpoint , . Above it, the flow heads toward strong vortex fluctuations and vanishing long-distance stiffness. The trajectory separating these behaviors is called the separatrix. Microscopic models can begin with different couplings and core energies yet approach the same critical endpoint.
The flow is unusually slow close to the transition because the vortex perturbation is marginal at the endpoint: its leading scaling exponent vanishes there. As a result, finite systems can remain near critical behavior over many decades of length before their eventual phase becomes apparent. Logarithmic corrections are therefore intrinsic to BKT scaling, not merely experimental imperfections. They are particularly important when extracting transition temperatures from numerical data.
A complementary representation is the sine-Gordon model, in which a periodic cosine term in a dual field represents vortex creation. The BKT transition corresponds to a change in whether that term is relevant at long distances. The equivalence between a compact phase field, a logarithmically interacting defect gas, and a sine-Gordon field theory connects lattice statistical mechanics with continuum field theory. These descriptions share their long-distance content, although their microscopic parameters and normalization conventions need not coincide.
Universal jump and essential singularities
The most distinctive quantitative prediction is the universal discontinuity of the renormalized stiffness. In the thermodynamic limit,
where the superscript minus denotes approach from below. Just above the transition, the equilibrium stiffness vanishes at asymptotically long distances. The quantity is also called the helicity modulus when defined as the second derivative of free energy with respect to an imposed twist. The jump is universal in the sense that its ratio to is fixed for the standard transition driven by unit vortices.
For a neutral fluid whose mobile particles have mass , the stiffness can be written
where is the superfluid number per unit area and is the reduced Planck constant. The jump condition becomes
Equivalently, with thermal wavelength , it reads . This is a statement about superfluid density, not the total density or a condensate fraction.
On the disordered side, the correlation length has the asymptotic form
The amplitude and constant are nonuniversal. This essential singularity differs from the power-law divergence characteristic of many ordinary critical points. In the low-temperature phase, the correlation exponent approaches at the transition; the leading critical correlation is , modified by multiplicative logarithmic factors.
Despite the stiffness discontinuity, the standard BKT transition has no latent heat. Its singular free-energy density is of order , up to scaling factors, so the singularity is exponentially weak. Temperature derivatives of this contribution do not behave like those at an ordinary first-order transition. The description “infinite-order transition” refers to this essential thermodynamic singularity, not to an infinite number of phases.
The heat capacity therefore need not display a sharp divergent peak at the transition. In the square-lattice XY model, a broad maximum occurs above the BKT temperature and reflects substantial short-distance fluctuations. Stiffness, correlations, and vortex statistics provide more direct critical diagnostics than the position of that maximum.
Superfluid helium films
Thin films of helium-4 provided an early physical realization of BKT behavior. Bulk helium undergoes a three-dimensional superfluid transition, but a sufficiently thin film has effectively two-dimensional long-wavelength dynamics. The relevant field is its superfluid phase, and the defects are quantized vortices circulating within the film. Their interaction is logarithmic over the range where two-dimensional hydrodynamics applies. Film thickness controls both the available areal superfluid density and the importance of microscopic substrate effects.
The connection between phase and motion follows from
where is the mass of a helium-4 atom. Integrating around a unit vortex gives circulation . Combining this relation with the kinetic energy per unit area reproduces the phase-gradient free energy and identifies the stiffness with . Thus the abstract angular rigidity of the XY model becomes a directly meaningful hydrodynamic response in superfluidity.
Experiments by D. J. Bishop and J. D. Reppy, reported in 1978, used torsional-oscillator measurements of helium films to test the predicted universal jump. A torsional oscillator detects changes in the mass participating in the substrate’s motion: the superfluid component can partly decouple from that motion, shifting the oscillation frequency. Dissipation measurements supply complementary information about vortex dynamics. These observations gave strong support to vortex unbinding and to the Nelson–Kosterlitz relation.
The experimental response is not identical to the ideal equilibrium stiffness at infinite wavelength. At a finite oscillation frequency, vortices can move only over a limited distance during one cycle. The apparatus therefore probes a frequency-dependent effective stiffness. Near the transition, very large pairs may respond too slowly to be fully sampled, smoothing or shifting the apparent onset. Surface roughness, film nonuniformity, and finite lateral size introduce additional scales.
Areal superfluid density must also be distinguished from the total amount of adsorbed helium. Some helium close to the substrate may be effectively immobilized, and thermal excitations produce a normal component even in the mobile fluid. The universal jump applies to the renormalized superfluid response, not simply to coverage multiplied by atomic mass. This distinction is essential when comparing measurements from different substrates or film thicknesses.
Helium films illustrate an important feature of the theory: a system can sustain a superfluid response without a nonzero conventional condensate density in the uniform two-dimensional thermodynamic limit. Long-range phase rigidity and true long-range order are related but distinct properties.
Ultracold two-dimensional atomic gases
Ultracold atomic gases permit controlled studies of BKT physics with tunable confinement, density, and interaction strength. A gas becomes effectively two-dimensional when motion along one direction is frozen into a tightly confined state while motion in the remaining plane remains active. For harmonic transverse confinement of angular frequency , this usually requires the relevant thermal and interaction energy scales to be small compared with . Geometric thinness alone is insufficient if excited transverse modes are substantially occupied.
An experimental milestone was the 2006 work of Zoran Hadzibabic and collaborators on two-dimensional rubidium gases. Two parallel clouds were released and allowed to overlap, producing an interference pattern sensitive to their relative phases. Spatial fluctuations of the fringes provided access to phase correlations, while dislocations in the pattern revealed vortices in the original clouds. The measurements connected the loss of long-distance coherence with vortex proliferation, rather than relying solely on a density profile.
Interactions are important because the ideal uniform two-dimensional Bose gas does not undergo the same finite-temperature superfluid transition. For a weakly interacting gas, a commonly used dimensionless coupling is
where is the effective two-dimensional interaction constant. In the weak-coupling regime, the critical total phase-space density has the approximate form
with approximately in the standard convention. This interaction-dependent critical total density should not be confused with the universal superfluid phase-space density of four.
A Bose–Einstein condensate, a quasicondensate, and a superfluid describe different properties. Condensation concerns macroscopic occupation of a single quantum state; quasicondensation generally denotes suppressed local density fluctuations despite substantial phase fluctuations; superfluidity concerns phase rigidity and related transport responses. A uniform interacting two-dimensional gas can have BKT superfluidity without true long-range condensate order at nonzero temperature. Finite traps can nevertheless produce a large coherent fraction, so experimental interpretation must account for confinement.
Later experiments extended these investigations to flatter trapping potentials, spatially resolved coherence measurements, controlled vortex dynamics, and strongly interacting two-dimensional Fermi gases. In the fermionic case, pairing produces composite bosonic degrees of freedom, but pair formation and establishment of global superfluid coherence are not identical events. The BKT mechanism governs the latter when the effective long-distance theory has a compact phase and logarithmically interacting vortices.
Atomic systems also expose the importance of equilibration. Vortices left by preparation or rapid cooling may persist out of equilibrium. Their observation alone does not establish equilibrium unbinding; correlation measurements, response functions, and temperature dependence are needed to identify the transition.
Superconducting films and Josephson arrays
In a thin superconductor, the relevant angle is the phase of the Cooper-pair field. Establishing a substantial pairing amplitude does not automatically establish long-distance phase coherence. If amplitude fluctuations are comparatively weak while phase fluctuations remain strong, vortex unbinding can control the loss of superconductivity. This makes BKT theory especially useful for effectively two-dimensional films and networks, though not every broadened superconducting transition is a BKT transition.
A superconducting vortex couples to the electromagnetic field and carries a flux quantum
For a film of thickness , where is the London penetration depth, the characteristic magnetic screening distance is the Pearl length, conventionally
Vortex interactions are approximately logarithmic at distances smaller than , but change at larger distances. Consequently, a finite film most closely displays neutral-system BKT behavior when its relevant dimensions and observation scales lie within the logarithmic-interaction regime. Electromagnetic screening complicates a literal thermodynamic-limit identification.
Transport measurements offer practical signatures because moving vortices generate voltage. Above the transition, unbound vortices can produce resistance even when a local pairing amplitude remains. A frequently used form near the transition is
with a nonuniversal coefficient , reflecting the rapidly growing vortex correlation scale. Current–voltage measurements are also used: under standard diffusive vortex-dynamics assumptions, the voltage obeys , with at the transition. This exponent is a dynamical prediction under specified conditions, not a universal consequence independent of vortex motion and measurement protocol.
Arrays of superconducting islands joined by Josephson junctions provide a more explicitly designed phase model. The Josephson effect gives an interaction energy
where is a junction coupling and represents the applied electromagnetic vector potential integrated along the link, with the appropriate charge factor. With negligible charging effects, approximately uniform couplings, and no imposed frustration, the classical array closely resembles the XY model.
Disorder, electron heating, contact effects, finite measurement currents, and limited sample size can imitate or distort expected signatures. A convincing identification therefore combines stiffness or inductive response with transport scaling and an assessment of the available length scales. Strong charging effects introduce quantum phase fluctuations and may instead drive a superconductor–insulator transition. That transition is distinct from the thermal BKT transition, although BKT-type scaling can also arise in certain quantum models.
Related transitions, simulations, and boundaries of applicability
BKT behavior defines a distinctive universality class, but its microscopic requirements matter. The standard case involves an effectively two-dimensional compact phase, sufficiently short-range microscopic couplings, and elementary vortices with logarithmic interactions. Changing the allowed defects, introducing anisotropy, or coupling additional fluctuating fields can alter the phase diagram. A two-dimensional system is not automatically a BKT system, and observing a rounded transition is not by itself evidence for vortex unbinding.
The clock model illustrates how discrete anisotropy changes the picture. Its spins occupy only equally spaced angular orientations. For sufficiently large , the standard ferromagnetic model has an intermediate algebraically ordered phase between a high-temperature disordered phase and a low-temperature phase with true discrete order. For integer , the boundaries of this intermediate phase are BKT-type transitions. Their correlation exponents differ: the upper boundary has , while the lower boundary has . Smaller values of have different critical structures.
Two-dimensional melting provides another extension. In the Kosterlitz–Thouless–Halperin–Nelson–Young theory, dislocation unbinding destroys quasi-long-range translational order and produces a hexatic phase; subsequent disclination unbinding destroys its quasi-long-range bond-orientational order. The defects are not vortices of a superfluid phase, but analogous logarithmic interactions and screening mechanisms organize the transitions. The theory does not require every two-dimensional material to melt by this route: first-order melting and mixed scenarios also occur.
Numerical work commonly employs the Monte Carlo method, including cluster and worm algorithms, to measure correlations, vortex statistics, and winding fluctuations. For the nearest-neighbor square-lattice XY model, the transition occurs at approximately . That number is model-specific, whereas the limiting stiffness ratio and critical correlation exponent are universal. Accurate extraction requires treatment of logarithmic finite-size corrections; near criticality, the stiffness approaches its limiting jump only slowly as system size increases.
Several physical changes move a model outside the simplest BKT framework. Long-range interactions can modify the growth of phase fluctuations and permit different forms of order. Coupling between layers can turn the asymptotic transition into a three-dimensional one, even if substantial two-dimensional crossover behavior remains. Fractional vortices in multicomponent systems can change stiffness-jump relations because the elementary winding differs from the unit vortex assumed in the standard formula.
Finally, driven and dissipative systems may exhibit algebraic correlations without satisfying equilibrium BKT theory. Their long-distance dynamics can contain fluctuations and nonlinearities absent from an equilibrium phase-only free energy. Establishing a BKT transition in such settings requires identifying the effective symmetry, defect interactions, steady-state statistics, and relevant observation scales, rather than inferring universality from spatial dimensionality alone.