Critical Brownian Loop-Soup

• Critical Brownian loop-soup is a conformally invariant ensemble of Brownian loops defined by a Poisson point process with specific intensity parameters in 2D and 3D domains.
• At critical intensity, the model exhibits power-law cluster-size distributions and fractal geometries, with boundaries forming conformal loop ensembles (CLE₄) in two dimensions.
• The loop-soup framework links statistical lattice models to quantum field theory by coupling with the Gaussian free field and revealing universal scaling limits.

A critical Brownian loop-soup is a random, conformally invariant ensemble of Brownian loops in Euclidean space or a planar domain, parameterized by an intensity. In two dimensions, at certain critical intensities, the loop-soup model exhibits a robust geometric and probabilistic structure tied to phase transitions, conformal field theory (CFT), and probabilistic representations of quantum fields. It provides a canonical scaling limit for statistical lattice models and underpins key correspondences with Schramm–Loewner Evolution (SLE), Conformal Loop Ensembles (CLE), and the Gaussian Free Field (GFF) in the plane.

1. Construction and Definition

The Brownian loop-soup in a domain $D$ is a Poisson point process on the space of unrooted Brownian loops with intensity $\lambda>0$, governed by the Brownian loop measure $\mu_D$. The loop measure is constructed as

$$\mu_D(d\ell) = \int_D \int_{0}^\infty (2\pi t^2)^{-1} \mathbb{P}^{(t)}_{z,z}(d\ell) \, dt \, d^2z$$

where $\mathbb{P}^{(t)}_{z,z}$ is the law of a planar Brownian bridge of duration $t$ from $z$ to $z$ staying in $D$ (Sheffield et al., 2010, Camia et al., 2015).

For $\lambda>0$, the Brownian loop-soup $\lambda>0$0 is the Poisson process of loops in $\lambda>0$1 with intensity $\lambda>0$2. In $\lambda>0$3, the analogous construction is with

$\lambda>0$4

where $\lambda>0$5 is Brownian bridge law in $\lambda>0$6 (Jego et al., 8 Jan 2026).

Clusters are defined by the intersection graph: two loops are adjacent if they intersect, and clusters are maximal connected sets.

2. Critical Intensity and Phase Transition

A nontrivial critical intensity $\lambda>0$7 governs the macroscopic geometry of the loop-soup. In two dimensions, the phase transition occurs at $\lambda>0$8, or equivalently $\lambda>0$9 (with a change of conventions in the literature) (Sheffield et al., 2010, Camia, 2015). At criticality:

• For $\mu_D$0, every cluster is almost surely finite and the collection of clusters remains disconnected.
• At $\mu_D$1, one is at a sharp threshold: cluster-size distributions follow power laws, the outer cluster boundaries become Conformal Loop Ensembles CLE$\mu_D$2, and the “carpet” (the complement of all loops) forms a nontrivial fractal set.
• For $\mu_D$3, there is almost surely a unique giant cluster (“full-packing” regime).

In three dimensions, the critical intensity $\mu_D$4 satisfies $\mu_D$5. For $\mu_D$6, all clusters are finite, while for $\mu_D$7, there is at least one unbounded cluster; for large $\mu_D$8, all loops join the same, space-filling cluster. This feature distinguishes the critical Brownian loop-soup in $\mu_D$9 from the $$\mu_D(d\ell) = \int_D \int_{0}^\infty (2\pi t^2)^{-1} \mathbb{P}^{(t)}_{z,z}(d\ell) \, dt \, d^2z$$0 case: true percolation (infinite cluster formation) only appears in $$\mu_D(d\ell) = \int_D \int_{0}^\infty (2\pi t^2)^{-1} \mathbb{P}^{(t)}_{z,z}(d\ell) \, dt \, d^2z$$1 (Jego et al., 8 Jan 2026).

Table: Main Phase Transitions in Brownian Loop-Soups

Dimension	Critical Intensity	Subcritical ($$\mu_D(d\ell) = \int_D \int_{0}^\infty (2\pi t^2)^{-1} \mathbb{P}^{(t)}_{z,z}(d\ell) \, dt \, d^2z$$2 critical)	Critical	Supercritical ($$\mu_D(d\ell) = \int_D \int_{0}^\infty (2\pi t^2)^{-1} \mathbb{P}^{(t)}_{z,z}(d\ell) \, dt \, d^2z$$3 critical)
$$\mu_D(d\ell) = \int_D \int_{0}^\infty (2\pi t^2)^{-1} \mathbb{P}^{(t)}_{z,z}(d\ell) \, dt \, d^2z$$4	N/A	N/A	Trivial	Trivial
$$\mu_D(d\ell) = \int_D \int_{0}^\infty (2\pi t^2)^{-1} \mathbb{P}^{(t)}_{z,z}(d\ell) \, dt \, d^2z$$5	$$\mu_D(d\ell) = \int_D \int_{0}^\infty (2\pi t^2)^{-1} \mathbb{P}^{(t)}_{z,z}(d\ell) \, dt \, d^2z$$6	Finitely many clusters	Power-law	Unique giant cluster
$$\mu_D(d\ell) = \int_D \int_{0}^\infty (2\pi t^2)^{-1} \mathbb{P}^{(t)}_{z,z}(d\ell) \, dt \, d^2z$$7	$$\mu_D(d\ell) = \int_D \int_{0}^\infty (2\pi t^2)^{-1} \mathbb{P}^{(t)}_{z,z}(d\ell) \, dt \, d^2z$$8	Only finite clusters	Percolation threshold	Unique dense cluster for $$\mu_D(d\ell) = \int_D \int_{0}^\infty (2\pi t^2)^{-1} \mathbb{P}^{(t)}_{z,z}(d\ell) \, dt \, d^2z$$9
$\mathbb{P}^{(t)}_{z,z}$0	No percolation	No intersection	No percolation	No percolation

3. Conformal Invariance, Restriction, and Loop-Soup–CLE Correspondence

The Brownian loop measure and the loop-soup law are exactly conformally invariant in $\mathbb{P}^{(t)}_{z,z}$1. For simply connected domains $\mathbb{P}^{(t)}_{z,z}$2 and conformal $\mathbb{P}^{(t)}_{z,z}$3, loops are mapped in law to those in $\mathbb{P}^{(t)}_{z,z}$4 via $\mathbb{P}^{(t)}_{z,z}$5 (Camia et al., 2015, Camia et al., 2021). This gives rise to the restriction property: conditioning on loops remaining in subdomains produces a statistically identical loop-soup in the smaller domain (Sheffield et al., 2010).

At criticality in two dimensions ($\mathbb{P}^{(t)}_{z,z}$6), the outermost boundaries of clusters in the loop-soup are distributed as CLE$\mathbb{P}^{(t)}_{z,z}$7—simple, non-nested, disjoint continuous loops with the full conformal restriction law. This establishes an equivalence between:

• Branching SLE$\mathbb{P}^{(t)}_{z,z}$8 loop ensembles for $\mathbb{P}^{(t)}_{z,z}$9,
• Outermost cluster boundaries in the loop-soup at intensity $t$0,
• Ensembles satisfying the conformal restriction axioms (Sheffield et al., 2010).

4. Critical Exponents, Fractal Geometry, and Cluster Structure

At and near criticality, the model exhibits explicit critical exponents and a well-characterized fractal geometry:

• The Hausdorff dimension of the “carpet” for $t$1 is $t$2; the CLE$t$3 cluster boundary loops have dimension $t$4.
• Crossing/connectivity probabilities exhibit nontrivial exponents: the probability that an annulus is traversed decays as $t$5 at $t$6, in line with SLE$t$7 and CLE$t$8 one-arm events (Sheffield et al., 2010).
• In $t$9, there exists an algebraic one-arm exponent $z$0 for the connection probability between $z$1 and $z$2, with $z$3 as $z$4 at criticality (Jego et al., 8 Jan 2026).

Importantly, at criticality in two dimensions, the cluster boundaries have no double points almost surely: the set of double points on critical cluster boundaries is empty with probability one (Gao et al., 27 Jul 2025). This feature confirms that loop-soup CLE$z$5 boundaries are simple.

5. Conformal Field Theory, Primary Operators, and Central Charge

The critical Brownian loop-soup gives rise to a nontrivial CFT with central charge $z$6 (Camia et al., 2015, Camia et al., 2021):

• Primary operators in the loop-soup CFT are given by exponentiated statistics of the loops (“layering” and “winding” operators), $z$7, with either $z$8 the signed number of loops covering $z$9, or the total winding number (Camia et al., 2015, Foit et al., 2020).
• Their scaling dimensions are computable: for the layering operator, $z$0; for winding, $z$1, each showing $z$2-periodicity in $z$3.
• The CFT admits an explicit stress-energy tensor $z$4 (expressed both in terms of loop edge numbers and vertex operators), which satisfies standard OPEs and conformal Ward identities (Camia et al., 2021).
• The loop-soup CFT is non-minimal, non-unitary for generic $z$5, and allows a continuous operator spectrum. At $z$6, the central charge is $z$7, matching the free (Gaussian) boson theory.

The field of occupation times (local time fields) of the critical loop-soup can be rigorously identified in law with the Wick square $z$8 of the Gaussian free field $z$9 in the domain (Lehmkuehler et al., 2024, Aïdékon et al., 2021).

6. Loop-Soup Cluster Decomposition, Excursions, and CLE–GFF Couplings

A pivotal structural feature at criticality is the exact decomposition of loop-soup clusters conditioned on their boundaries. For the $D$0 loop-soup in $D$1 and a CLE$D$2 outer boundary $D$3, the set of boundary-touching loops in the soup is distributed as an independent Poisson point process of Brownian excursions of intensity $D$4 in the domain enclosed by $D$5 (Qian et al., 2015, Qian, 2016). This allows:

• A Markovian resampling property and a domain Markov structure for CLEs and loop-soups,
• The identification between loop-soup occupation times and the GFF squared,
• The equivalence and commutativity of the three central couplings: GFF $D$6 CLE$D$7 level-lines (Miller–Sheffield), GFF$D$8 $D$9 occupation fields (Le Jan), loop-soup $\lambda>0$0 CLE$\lambda>0$1 via cluster boundaries (Sheffield–Werner) (Qian et al., 2015).

This decomposition is unique to $\lambda>0$2: at other intensities, the set of loops touching the boundary or a portion thereof fails to be a Poisson point process of excursions, and the Markovian structure is strictly weaker (Qian, 2016).

7. Scaling Limits, Universality, and Higher Dimensions

The critical Brownian loop-soup arises as the universal scaling limit of many lattice loop models: discrete random-walk loop soups on planar graphs (or tori) converge in the metric of unrooted loops to the Brownian loop-soup, provided suitable invariance and RSW-type properties hold (Pang, 13 Mar 2026, Camia, 2015). At criticality, the cluster boundaries become conformal loop ensembles CLE$\lambda>0$3, and the critical exponents, central charge, and CFT data are universal.

In $\lambda>0$4, the critical loop-soup defines the only nontrivial continuum percolation scenario for Brownian loop-soups: above $\lambda>0$5, there is a unique, possibly dense, cluster spanning $\lambda>0$6. For $\lambda>0$7, Brownian loops almost surely do not intersect, so clusters are single loops and percolation is absent (Jego et al., 8 Jan 2026).

In high dimensions ($\lambda>0$8), loop-soup clusters containing macroscopic cycles split into two asymptotically independent families (“intensity doubling”): cycles made by a single large loop and cycles built from chainings of small loops (ghost cycles), both scaling to independent critical loop-soups (Lupu et al., 26 Nov 2025).

---

References by arXiv id:

• (Sheffield et al., 2010): Conformal Loop Ensembles: Construction via Loop-soups
• (Camia, 2015): Brownian Loops and Conformal Fields
• (Camia et al., 2015): Conformal Correlation Functions in the Brownian Loop Soup
• (Qian et al., 2015): Decomposition of Brownian loop-soup clusters
• (Qian, 2016): Conditioning a Brownian loop-soup cluster on a portion of its boundary
• (Foit et al., 2020): New Recipes for Brownian Loop Soups
• (Camia et al., 2021): The Brownian loop soup stress-energy tensor
• (Aïdékon et al., 2021): Multiplicative chaos of the Brownian loop soup
• (Lehmkuehler et al., 2024): Parity questions in critical planar Brownian loop-soups
• (Gao et al., 27 Jul 2025): Non-existence of several random fractals in the Brownian motion and the Brownian loop soup
• (Lupu et al., 26 Nov 2025): Intensity doubling for Brownian loop-soups in high dimensions
• (Jego et al., 8 Jan 2026): Three-dimensional Brownian loop soup clusters
• (Pang, 13 Mar 2026): Universality for the 2D Random Walk Loop Soup