The XY transition: a quarter-power law, logarithms, and an exponential length scale

← All digests

Family 216 · Probability and statistical mechanics
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-11

1. The question

A two-dimensional magnet can change phase without acquiring a preferred direction. In the planar XY model, each spin is an arrow constrained to a circle. At high temperature, distant arrows rapidly forget one another. At low temperature, their correlations persist as a power of distance, even though there is no ordinary spontaneous magnetization.

The Berezinskii–Kosterlitz–Thouless, or BKT, picture predicts unusually precise behavior at the boundary between these phases. The critical correlation should decay with exponent $1/4$, multiplied by a small logarithmic correction. Approaching the transition from the high-temperature side, the correlation length should grow faster than any power of the distance to criticality.

The six manuscripts in this family claim these critical asymptotics for the ordinary cosine XY model and identify its critical continuum spin field. A broader companion also treats rough height surfaces and low-temperature spin fields. Together they describe both individual correlations and a random field built from many spins at once.

2. Background and essential definitions

Write the spin at $x\in\mathbb Z^2$ as $e^{i\theta_x}$, where angles are taken modulo $2\pi$. In a square box, the probability density is proportional to

$$\exp\!\left(b\sum_{\{x,y\}}\cos(\theta_x-\theta_y)\right),$$

relative to independent uniform angles. Every nearest-neighbor edge occurs once. The parameter $b$ is inverse temperature: heating decreases it.

For free boundary conditions, first send the box size to infinity while keeping two sites fixed. Let $C_b(r)$ be their expected cosine difference when they are $r$ steps apart along a coordinate axis. Then define

$$m(b)=\lim_{r\to\infty}-\frac1r\log C_b(r), \qquad b_c=\inf\{b>0:m(b)=0\}.$$

A positive mass $m(b)$ means exponential decay. Its reciprocal $\xi(b)=1/m(b)$ is the axis correlation length. The infinite-volume limit comes before the separation limit. The threshold is defined through this observable, rather than by presupposing the desired critical exponent.

A vortex is a configuration in which the angle winds by a nonzero multiple of $2\pi$ around a closed circuit. Smooth angle fluctuations are called spin waves. In the BKT picture, vortex pairs remain bound at low temperature; unbinding destroys the slowly decaying phase.

A second description uses integer-valued heights on faces of the lattice. Fourier expansion of the spin interaction turns angle integration into conservation of integer currents. Rotating these currents gives height differences. For the cosine model the edge weights involve modified Bessel functions, so this dual height model is not simply a discrete Gaussian model.

The continuum Gaussian free field, or GFF, is a random distribution whose covariance is the inverse Laplacian. In a domain $D$ with zero boundary values, its covariance kernel has local behavior

$$G_D(z,w)=\frac{1}{2\pi}\log\frac1{|z-w|} +\text{a smooth remainder}.$$

Pointwise variance diverges, but averages against smooth functions are well defined. A spin field is related to an imaginary exponential of this Gaussian field, rather than to the Gaussian field itself.

3. Development history and previous partial results

Berezinskii’s early analysis predicted a low-temperature phase with algebraic correlations. Kosterlitz and Thouless explained the transition through vortex binding and unbinding, and Kosterlitz’s renormalization calculation predicted both the critical exponent and the essential singularity of the correlation length. The important variable is the effective stiffness seen after averaging over small-scale fluctuations, which need not equal the original coupling. Kosterlitz–Thouless, Kosterlitz

McBryan and Spencer obtained algebraic upper bounds. Fröhlich and Spencer’s 1981 multiscale analysis proved a low-temperature algebraic regime for planar rotators and related models. This established a central part of the phase picture without fixing the exact critical correlation asymptotic. Fröhlich–Spencer

Van Engelenburg and Lis later gave a current-and-loop approach yielding an exponential-versus-polynomial dichotomy for the ordinary XY model. Their polynomial lower bound also holds at the threshold. This is an essential endpoint input in the present manuscripts, although its exponent is not the predicted $1/4$. Van Engelenburg–Lis

Height methods supplied another route. Aizenman, Harel, Peled, and Shapiro connected height depinning with BKT behavior, using discrete Gaussian comparison inequalities. Lammers related the XY and dual-height correlation lengths exactly and developed a localization dichotomy. These results identify the relation between the two transitions; determining the effective coefficient precisely at the endpoint remains additional work. Depinning and BKT phases, Bijecting the BKT transition

Rigorous renormalization also established detailed Gaussian limits and logarithmic corrections in related settings. Bauerschmidt, Park, and Rodriguez treated high-temperature discrete Gaussian heights, while Falco analyzed a dilute Coulomb gas at criticality. Such results are important methodological predecessors. Their hypotheses and observables differ from those of the fixed nearest-neighbor cosine model at its physical transition, so their conclusions cannot simply be transferred unchanged.

4. What the new manuscripts claim

Critical bulk and boundary correlations

M5 first states

$$C_{b_c}(r)=r^{-1/4+o(1)}.$$

This fixes the exponent but allows slowly varying multipliers. M1 sharpens it to

$$C_{b_c}(r)=B\,r^{-1/4}(\log r)^{1/8}(1+o(1)), \qquad 0<B<\infty.$$

The amplitude $B$ is model-specific. The logarithm grows more slowly than every positive power of $r$, but it means that $r^{1/4}C_{b_c}(r)$ does not approach a finite nonzero constant.

For a different observable, fix every boundary angle of $[-n,n]^2$ to zero and let

$$a_n=\mathbb E^0_{n,b_c}\cos\theta_0.$$

M2 states, using its specified companion inputs,

$$a_n=A\,n^{-1/8}(\log n)^{1/16}(1+o(1)), \qquad 0<A<\infty.$$

This is magnetization at the center caused by an aligned boundary. It is not the root-mean-square spatially averaged magnetization in a periodic box. The two amplitudes $A$ and $B$ also need not be equal.

The near-critical correlation length

M4 claims that, approaching from the massive side $b<b_c$,

$$\sqrt{b_c-b}\,\log\xi(b)\longrightarrow A_\xi, \qquad 0<A_\xi<\infty.$$

Equivalently,

$$\xi(b)=\exp\!\left(\frac{A_\xi+o(1)}{\sqrt{b_c-b}}\right).$$

The constant is not evaluated in closed form and depends on the temperature convention. This logarithmic asymptotic does not assert a multiplicative equivalent with a fixed prefactor outside the exponential.

The critical continuum spin field

M3 works in $D=(-1,1)^2$ with mesh $1/n$ and boundary angles zero. Set $K_*=2/\pi$. Let $G_n$ be the inverse of the unscaled nearest-neighbor Dirichlet graph Laplacian. The normalized spin distribution is

$$S_n=\frac1{n^2}\sum_{x\in D_n^\circ} \frac{\exp\!\left((G_n(x,x)-G_n(0,0))/(2K_*)\right)}{a_n} e^{i\theta_x}\,\delta_x.$$

Here $D_n^\circ$ is the set of interior mesh sites and $\delta_x$ evaluates a test function at $x$. The exact center magnetization removes the microscopic amplitude; the Green-function correction adjusts its spatial dependence.

The claimed limit is

$$V_* =\lim_{\varepsilon\downarrow0} \exp\!\left( \frac{i\Phi_{D,\varepsilon}}{\sqrt{K_*}} +\frac{\mathbb E\Phi_{D,\varepsilon}^2}{2K_*} \right),$$

interpreted after integration against smooth test functions. The regularized GFF $\Phi_{D,\varepsilon}$ averages over small circles. The positive variance term compensates for the vanishing mean of the unnormalized imaginary exponential. It uses the entire variance, including its finite position-dependent part.

M3 claims convergence along all integer meshes in a local negative Sobolev topology, a topology for distributions, together with all mixed smeared moments. Both spins and their complex conjugates are included, without requiring equal numbers. The theorem uses the exact $a_n$, so its formulation does not depend on first knowing M2’s asymptotic.

Rough heights and low-temperature universality

M6 treats discrete Gaussian heights with values in $2\pi\mathbb Z$, for any finite interaction set $J$ invariant under square symmetries and containing nearest neighbors. If

$$v_J^2=\frac{1}{2|J|}\sum_{y\in J}y_1^2,$$

its effective height temperature at the physical roughening threshold satisfies

$$\beta_{\mathrm{eff}}(J,\beta_c(J))=8\pi v_J^2.$$

Mean-subtracted macroscopic torus heights converge to a GFF with covariance coefficient $\beta_{\mathrm{eff}}/v_J^2$, including at the threshold. The paper also gives differentiability above the threshold, an infinite right slope there, and asymptotics for interactions extending over increasingly large neighborhoods. The height temperature $\beta$ is a separate parameter from the spin inverse temperature $b$.

For the ordinary XY and Villain spin models, M6 proves imaginary-GFF field limits at sufficiently low fixed temperatures, with model-dependent stiffness and amplitude. The Villain interaction is a periodic sum of Gaussian angle weights. The low-temperature theorem does not itself reach either spin model’s transition; M3 supplies the additional critical XY result.

5. Why the advance matters

Importance: 5/5. The family addresses several signature predictions of one of statistical mechanics’ central unconventional phase transitions. It connects the critical exponent, its logarithmic refinement, the divergence of the correlation length, and a full continuum spin field for the original cosine interaction.

These are complementary achievements. An exponent is a coarse summary of decay. A positive finite amplitude with a logarithmic correction requires finer control. A field limit organizes arbitrary collections of spin averages. A near-critical theorem explains how the finite correlation length becomes infinite as the transition is approached.

The height results add a broader universality statement across finite-range interactions. Their normalization also shows why numerical constants such as $8\pi$ and $2/\pi$ must always be tied to the precise field and Laplacian convention.

6. The main idea

Identify the critical Gaussian coefficient

The first stage uses finite comparison inequalities and changes to the height graph to construct an effective covariance coefficient $a(b)$. For the dual XY heights, the manuscripts establish the alternative $a=0$ or $a\geq8\pi$.

The known critical polynomial lower bound rules out zero at $b_c$. A separate openness argument says that a coefficient strictly above $8\pi$ would keep the model massless in a neighborhood extending below $b_c$, contradicting the threshold definition. Thus the endpoint is $a(b_c)=8\pi$.

Spin insertions create prescribed windings in the height description. Their energy then gives exponent $2\pi/a=1/4$. Controlling ordinary height averages is insufficient: one must also compare the partition functions with these winding defects.

Accumulate a small correction across scales

At criticality, the renormalized interaction approaches its limiting Gaussian form slowly. If $j$ counts geometric length scales, the relevant deviation has size proportional to $1/j$. Summing that deviation gives $\sum_{j\leq N}1/j\sim\log N$. Since physical size grows exponentially with $N$, exponentiating the accumulated correction produces a power of $\log n$.

The center observable acquires power $1/16$; the bulk two-point function acquires power $1/8$. To obtain a positive finite amplitude, the remaining errors must be summable, not merely smaller than the leading term. Comparing nested geometries with identical microscopic insertions lets otherwise troublesome normalization factors cancel exactly.

Turn a departure scale into a physical length

Near criticality, two coupled coordinates track stiffness and the first periodic height interaction. Their leading recursion has the form

$$x_{j+1}-x_j\approx-y_j^2, \qquad y_{j+1}-y_j\approx-x_jy_j.$$

M4 controls how changing the original temperature moves the system away from the critical trajectory. The exit occurs after order $(b_c-b)^{-1/2}$ scale steps. Exponentially growing block sizes then suggest the essential singularity.

The manuscript must additionally show that this analytic exit scale is the actual correlation length. It does so through height fluctuations, annular winding estimates, and a finite-size criterion for spin decay. Without that connection, a recursion time would not yet be an observable of the original magnet.

7. Scope, limitations, and open questions

The critical XY statements concern the square lattice with its ordinary nearest-neighbor cosine interaction. They do not give a closed-form value of $b_c$, the amplitudes, or $A_\xi$. Universality of exponents does not mean equality of microscopic constants.

The boundary conventions differ across conclusions. Bulk correlations use a free thermodynamic limit; center magnetization and the continuum spin field use aligned boundary spins on a square. The transfer between these geometries requires additional estimates and is not automatic from a field limit.

Similarly, the discrete Gaussian height theorem reaches its physical roughening threshold, while M6’s spin theorem covers only a sufficiently low-temperature range. The critical XY result relies on the companion’s specific critical-height inputs. This family does not silently extend that conclusion to the Villain endpoint or to arbitrary planar-spin interactions.

The claims concern distributions and integrated moments, not pointwise convergence of the angle field. They also leave more detailed questions about interfaces, finite-size corrections, and other domains beyond the particular assertions stated here.

8. Sources and further reading

M5 introduces the critical exponent and height coefficient; M1 and M2 refine the bulk and center asymptotics. M3 constructs the critical spin field, M4 treats the massive approach to criticality, and M6 supplies the broader height and low-temperature framework.

For the development of the physical picture, see Kosterlitz’s 1974 analysis. For established rigorous phase information, see Fröhlich–Spencer, van Engelenburg–Lis, and Lammers’ spin–height relation. The six source manuscripts identify precisely which companion inputs each argument uses.

Source manuscripts

• M1. OpenAI, The critical logarithmic correction for the planar XY model (October 5, 2026).
• M2. OpenAI, Critical Center Magnetization in the Planar XY Model (October 5, 2026).
• M3. OpenAI, The Critical Spin Field of the Planar XY Model (October 5, 2026).
• M4. OpenAI, Essential Singularity of the Correlation Length in the Planar XY Model (October 5, 2026).
• M5. OpenAI, The critical correlation exponent of the planar XY model (September 24, 2026).
• M6. OpenAI, BKT universality for height and planar spin fields (September 24, 2026).

Sources consulted on 11 October 2026 at repository revision fd4aeeb. Download the manuscript citations.