gpt 6 astra pro attempt on Math-180

Conformal Invariance of the Critical Random-Cluster Model with q <= 4

Definitions and mathematical statement

On a finite planar graph with boundary condition $\xi$, the FK measure gives a configuration $\omega\subseteq E$ weight proportional to $p^{|\omega|}(1-p)^{|E|-|\omega|}q^{k_\xi(\omega)}$, where $k_\xi$ counts open clusters after boundary identifications. In a simply connected lattice approximation with marked boundary points, Dobrushin conditions produce an interface between wired primal and dual arcs. The intended statement is convergence of that interface at criticality to chordal $\mathrm{SLE}_{\kappa(q)}$ under mesh refinement. SLE is the Loewner chain driven by $\sqrt\kappa B_t$, and the customary FK relation is $\sqrt q=-2\cos(4\pi/\kappa)$, $4\le\kappa<8$, for $0<q\le4$; see [E1]. The catalogue omits the lower range of $q$, lattice, boundary setup, and curve topology. These must be taken from the cited formulation; the formula does not define a probability model for arbitrary negative $q$.

Formulation and concept references: [S1], [E1].

Short English statement

Do critical planar random-cluster interfaces approach the conformally invariant SLE curve predicted by their cluster-weight parameter, with the model and boundary conventions properly fixed?

Research attempt

Current frontier.

Editorial scope note. The original quotation is retained, including its omissions. The consulted sources formulate planar FK interfaces with specific lattice approximations and Dobrushin boundary arcs; a statement for arbitrary negative $q$ is not a probability assertion. The discussion below uses the positive-$q$ Dobrushin setup of [S1] and [E1]; it does not silently complete every omitted lattice or topology convention. In particular, FKG-based arguments in the range $q\ge1$ cannot simply be applied to $0<q<1$.

For $q=2$ the FK-Ising interface and its scaling limit are established; Kemppainen--Smirnov obtain the full FK-Ising scaling limit in the setting of [E2]. That is not a theorem for all $q\le4$. Nor does triangular site percolation settle every square-lattice FK model at $q=1$. The exact identification step below is conditional on analytic limits that are not supplied by a lattice contour identity alone.

Attack route.

One route proves compactness from crossing estimates and then identifies every subsequential limit. Another seeks a convergent parafermionic observable. The first lacks a characterization of the limit; the second lacks enough regularity and boundary control for general $q$. We combine them at a concrete point: two continuum martingales, if obtained with the correct covariance and sufficient integrability, force the entire Loewner driver. This derives the predicted parameter while making the missing lattice-to-continuum input explicit.

Attempt: normalization of the candidate observable.

For $0<q\le4$, choose $0<\sigma\le1$ by \[ \sin(\pi\sigma/2)=\sqrt q/2, \qquad \kappa=\frac8{1+\sigma}. \] Then $4\le\kappa<8$ and \[ -2\cos(4\pi/\kappa) =-2\cos\left(\pi(1+\sigma)/2\right)=\sqrt q. \] Thus $q=1,2,4$ correspond respectively to $(\sigma,\kappa)=(1/3,6),(1/2,16/3),(1,4)$, as in the FK parameter conventions of [E1]. This is a normalization check, not a convergence theorem.

Suppose a subsequential interface limit is generated in the upper half-plane by continuous chordal Loewner evolution \[ \partial_tg_t(z)=\frac2{g_t(z)-U_t},\qquad U_0=0. \] For the calculation assume additionally that $U$ is a continuous semimartingale, with decomposition $U=A+N$, where $A$ is continuous of finite variation and $N$ is a continuous local martingale. Set $K=[N]$. These properties are hypotheses here; a continuous driver need not be a semimartingale merely because it arose as a subsequential limit. For an interior probe $z$, put \begin{equation}\label{wsref-0-65-71-31-37-33-2d-6f-62-73-65-72-76-61-62-6c-65} Z_t(z)=g_t(z)-U_t, \qquad M_t(z)=\left(\frac{g_t'(z)}{Z_t(z)}\right)^\sigma. \end{equation} Choose analytic branches locally and stop before the probe is swallowed or the displayed quantities approach zero or infinity. Such stopping makes every differentiation below legitimate.

Attempt: two martingales force the driver.

The equations for $Z$ and $g'$ are \[ dZ=2Z^{-1}dt-dU,\qquad d\log g'=-2Z^{-2}dt. \] Itô's formula, including the quadratic variation of $Z$, gives \begin{equation}\label{wsref-0-65-71-31-37-33-2d-69-74-6f} \frac{dM_t(z)}{M_t(z)} =\frac{\sigma}{Z_t(z)}\,dU_t +\frac{-4\sigma\,dt+\sigma(1+\sigma)dK_t/2}{Z_t(z)^2}. \end{equation} The coefficient $-4\sigma$ includes both the derivative of $g'$ and the drift in $Z$; omitting either would predict the wrong value of $\kappa$.

Suppose the processes (eq173-observable) for two distinct probes $z,z'$ are complex local martingales on a common stopped interval. Uniqueness of the semimartingale decomposition says their finite-variation parts vanish. Multiplication by the bounded nonzero factors gives equalities of complex measures \begin{equation}\label{wsref-0-65-71-31-37-33-2d-6d-65-61-73-75-72-65-73} Z_t(z)\,dA_t+\frac{1+\sigma}{2}\,dK_t-4dt=0, \qquad Z_t(z')\,dA_t+\frac{1+\sigma}{2}\,dK_t-4dt=0. \end{equation} Before either probe is swallowed, conformality implies $g_t(z)\ne g_t(z')$. Subtracting the equations yields $dA=0$; locally one may divide by their nonzero continuous difference. Substitution yields $dK=8dt/(1+\sigma)=\kappa dt$. Lévy's characterization now gives $U=\sqrt\kappa B$ on that interval. An exhaustion by stopped intervals, with suitable probes covering every finite time, gives the chordal $\mathrm{SLE}_\kappa$ driver.

This is a proved stochastic identification lemma under its stated hypotheses. It does not assert that the particular FK observable has already converged to (eq173-observable). Two fixed probes can be swallowed; the exhaustion or a sufficiently rich family of probes is necessary for a conclusion about all times.

Attempt: the lattice-to-limit inputs that would close the route.

Here is a sufficient package for a fixed admissible lattice, domain, and parameter $q$. First obtain tightness of the interfaces in the chosen curve topology and enough control to pass to Loewner chains parametrized by half-plane capacity. Next show that the normalized lattice observable, with its prescribed boundary values, converges to (eq173-observable) along every subsequence, jointly with the explored hulls. Finally justify passage of its conditional martingale identities and establish a semimartingale driver, or derive that property by an independent characterization theorem. The preceding lemma then identifies every subsequential law. Tightness and uniqueness of all limit laws imply convergence of the full sequence.

A usable martingale-passage argument must be quantitative. On each stopped interval, joint distributional convergence together with uniform integrability allows passage of ${\mathbb{E}}[(M_t-M_s)F]=0$ for bounded continuous functions $F$ of the past. It then extends to the limiting past sigma-field by approximation. Without uniform integrability, convergence in distribution of martingales need not preserve their conditional means. Stopping a proposed continuum observable does not automatically bound its lattice precursors, so a separate lattice estimate is still required.

The discrete parafermionic contour relation described in [E1] is a motivation for this package, not a substitute for it. To identify a holomorphic boundary-value solution, one needs precompactness, the correct boundary behavior, and uniqueness in a suitable analytic class. For general $q$, a local linear contour identity alone has not supplied those inputs in this attempt. At $q=4$ the stochastic calculation is nonsingular, but that observation does not remove possible logarithmic-scale difficulties in the lattice estimates.

Stress test.

No use was made of independent increments before deriving Brownian motion. The semimartingale assumption, the uniform-integrability requirement, and the availability of probes after successive swallowing times have not been hidden in the word "martingale." The computations are stopped away from singularities, so fractional powers and Itô derivatives are well defined.

The limit lemma works for each positive $q$ with the displayed parameter, but it does not manufacture a critical FK model for omitted or inadmissible conventions. In particular, results proved only for FK-Ising [E2] are not invoked for other cluster weights. The extra geometric control needed for trace convergence is not supplied solely by convergence of driving functions, and the topology in the original source must remain part of any completed statement.

Outcome.

Outcome: Conditional reduction.

The attempt proves a two-probe continuum identification lemma and derives the exact FK--SLE parameter relation from its drift cancellation. The missing input is a lattice convergence and tightness package strong enough to realize those martingales, with the specified boundary and topology conventions. The argument is not a new general conformal-invariance theorem; its standard Loewner and stochastic ingredients are used to make the gap explicit rather than to assume it away.

Sources

• S1 H. Duminil-Copin, 'Lectures on the Ising and Potts models on the hypercubic lattice', arXiv:1707.00520 $2017$, in Random Graphs, Phase Transitions and the Gaussian Free Field, Springer PIMS 123 $2020$. https://ar5iv.labs.arxiv.org/html/1707.00520. Catalogue formulation source.
• E1 Hugo Duminil-Copin and Stanislav Smirnov, Conformal invariance of lattice models, arXiv:1109.1549v4, FK interface and SLE discussion. https://arxiv.org/html/1109.1549v4. Added primary source: the random-cluster interface setup and parameter conventions. Consulted 24 September 2026.
• E2 Antti Kemppainen and Stanislav Smirnov, Conformal invariance in random cluster models. II. Full scaling limit as a branching SLE, arXiv:1609.08527, 2016, abstract and FK-Ising scaling-limit theorem. https://arxiv.org/abs/1609.08527. Consulted 27 September 2026. The cited result concerns $q=2$, not arbitrary $q\le4$.