When a random surface becomes a tree and how a particle moves on it

← All digests

Family 211 · Probability and statistical mechanics
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-11

1. The question

A very large network drawn on a sphere can be viewed as a discrete random surface. Its vertices are locations, its edges provide routes, and counting edges along a shortest route gives a distance. If the network is sampled randomly and its size grows, does a definite kind of continuum geometry emerge?

The answer can depend on the statistical mechanics used to weight the network. In the Fortuin–Kasteleyn model, abbreviated FK, a parameter $q$ changes the relative likelihood of cluster configurations and therefore changes the geometry of the sampled map itself.

This family claims a complete surface-to-tree transition for specified finite spherical FK maps. For $0<q\le4$, the limit is a random sphere described by Liouville quantum gravity. For $q>4$, the large-scale distances instead approach those of a continuum random tree.

Three further manuscripts ask how a particle moves on such a random surface. In the FK–Ising and spanning-tree cases, they identify a limiting diffusion and its exact time normalization. The FK–Ising results also describe the convergence of the graph’s relaxation frequencies. These are related questions, but a limit for distances alone does not determine the answer to any of the dynamical ones.

2. Background and essential definitions

Maps and their probability law

A planar map is a connected graph embedded in an oriented sphere, where the way edges surround vertices and faces is part of the data. A distinguished directed edge serves as a root. Loops and multiple edges are allowed here.

For a map $M$, choose an edge subset $A$. It gives a cluster configuration on the original graph. The complementary edges define clusters on the dual graph, whose vertices represent the faces of $M$. Let $\ell(M,A)$ count the interfaces separating the two kinds of clusters.

Among rooted maps with exactly $n$ edges, the joint law gives $(M,A)$ weight proportional to

$$q^{\ell(M,A)/2}.$$

A normalizing sum makes these weights into probabilities. The choice is at the FK critical point for every fixed $q$; varying $q$ is not the same as varying temperature in one fixed model.

The graph distance $d_n$ uses all original map edges, not only the selected edges of $A$. The natural vertex probability measure is

$$\mu_n(v)=\frac{\deg(v)}{2n},$$

where a loop contributes twice to the degree. This measures mass by incident edge ends and is the stationary distribution of the usual random walk. It is generally different from choosing a vertex uniformly.

Continuum geometry and motion

A Liouville quantum gravity, or LQG, sphere is a random surface with a random area measure and an intrinsic random distance. Its construction starts from a Gaussian random field, a continuum analogue of correlated Gaussian variables. The field is too irregular to be evaluated pointwise in the ordinary way, so defining its area and distance requires limiting constructions.

A parameter $\gamma$ determines the surface law. In this family, for $0<q<4$,

$$q=2+2\cos\left(\frac{\pi\gamma^2}{2}\right), \qquad \sqrt2<\gamma<2.$$

The endpoint $q=4$ corresponds to the critical parameter $\gamma=2$. Each limiting sphere has total quantum area one.

A conformal loop ensemble, written $\mathrm{CLE}_\kappa$, is a distribution of random loops whose law respects angle-preserving changes of coordinates. It describes continuum cluster boundaries. The parameter here is $\kappa=16/\gamma^2$, reaching four at the critical endpoint.

A Brownian continuum random tree is a compact random metric tree, encoded by a Brownian excursion. An excursion is a continuous Brownian-type path that starts and ends at zero while staying positive between them. A metric tree has a unique simple route between any two points. It can be highly branched even though it has no cycles.

Finally, Liouville Brownian motion is diffusion on an LQG surface. In a conformal coordinate its spatial motion is Brownian, with the clock changed according to quantum area. Its usual energy convention contains the factor $1/2$; that convention matters when stating an exact discrete time scale.

3. Development history and previous partial results

The discrete and continuum theories developed through several layers. Mullin’s enumeration and later bijections of Bernardi encoded maps decorated by spanning trees using pairs of tree contours. Sheffield’s inventory construction extended the encoding to FK-decorated maps and identified a transition in the limiting contour fluctuations at $q=4$. His paper is a starting point for the connection between cluster weights and random geometry.

Duplantier, Miller, and Sheffield’s mating of trees theory reconstructs LQG surfaces from correlated Brownian contour data. Miller and Sheffield developed the finite-area sphere version. Gwynne, Mao, and Sun, and Gwynne and Sun, established increasingly detailed convergence results for discrete FK encodings, including the finite-volume setting. The continuum construction explains why Brownian data can encode a random surface.

Convergence of an encoding is weaker than convergence of every graph distance in a prescribed drawing. Separate work constructed and characterized LQG metrics. For uniform random maps, corresponding to $q=1$ in the present ensemble, Brownian-sphere metric limits were already known through work of Le Gall, Miermont, and others. M1–M7 do not present that special case as a new discovery.

Embedded limits also had important precedents. Gwynne, Miller, and Sheffield treated the Tutte embedding of maps built directly from continuum trees; Holden and Sun treated the Cardy embedding of triangulations. These are different ensembles or embedding prescriptions from the finite spherical FK maps considered here.

For diffusion, Berestycki and Garban–Rhodes–Vargas constructed Liouville Brownian motion. Berestycki and Gwynne later proved time-parametrized walk convergence for mated-CRT maps, with a deterministic exit-time normalization comparable to inverse cell area. The exact asymptotic constant for that normalization was not identified there. Their theorem provides an important predecessor to the present finite-map claims.

Above $q=4$, an infinite-volume tree limit was established by Feng. The finite compact limit claimed here requires separate control of conditioning on the total size and of distances using every map edge.

4. What the new manuscripts claim

M4–M7: the geometric transition

M6 states that, for each fixed $0<q<4$, there is a deterministic distance scale $a_n\to0$ such that

$$\bigl(V(M_n),a_nd_n,\mu_n\bigr) \Longrightarrow \bigl(S,D_h,\mu_h\bigr).$$

Here $V(M_n)$ is the vertex set, while $S$, $D_h$, and $\mu_h$ are the limiting sphere, distance, and area. The arrow means convergence in distribution. The comparison simultaneously matches distances and probability measures, in what is called Gromov–Hausdorff–Prokhorov topology.

M6 also treats maps weighted by their numbers of spanning trees, with $\gamma=\sqrt2$. That is a separate argument, not a substitution of $q=0$ into the finite FK law.

M4 strengthens the FK conclusion by placing the maps in a specified conformal coordinate. Each edge is divided into incidence triangles, these are made equilateral, and the resulting sphere is conformally mapped to a standard sphere. Three independently sampled area points fix the coordinate ambiguity. Area, the full distance function, and all nested interfaces then converge together. The limiting loops have the law of an independent $\mathrm{CLE}_{16/\gamma^2}$.

M5 constructs the corresponding critical sphere at $q=4$ and states the joint conformal, area, distance, and nested-loop limit with $\mathrm{CLE}_4$. It treats this endpoint directly; the result is not obtained merely by letting $q$ approach four in a fixed-parameter theorem.

For every fixed $q>4$, M7 instead states

$$\bigl(V(M_n),c(q)n^{-1/2}d_n,\mu_n\bigr) \Longrightarrow \bigl(\mathcal T,d_{\mathcal T},\mu_{\mathcal T}\bigr),$$

where $c(q)>0$ is deterministic and the right side is the Brownian continuum random tree. This theorem forgets the root and cluster decoration. All these geometric limits run through every positive integer edge count.

M1–M3: motion and relaxation frequencies

At $q=2$, the FK–Ising sphere has $\gamma=\sqrt3$. M1 states that stationary random walks on the finite maps converge to Liouville Brownian motion on that same limiting sphere when time is accelerated by exactly $n$, the number of edges.

The unaccelerated walk makes one attempted move per unit time, choosing uniformly among incident half-edges. Loops can cause an attempt with no displacement. Starting vertices have law $\mu_n$. M1 retains the time parameter and any fixed finite number of conditionally independent walks on the same random map.

M3 gives the corresponding result for spanning-tree-weighted maps, retaining the conditional path law jointly with geometry and area, again with acceleration exactly $n$.

M2 treats FK–Ising spectra. If $\lambda_{n,j}$ are the ordered eigenvalues of the negative rate-one walk generator, then its claim includes

$$n\lambda_{n,j}\Longrightarrow\Lambda_j$$

jointly for all fixed mode indices, with multiplicities and the same limiting surface. It also states convergence of the heat traces

$$\sum_j e^{-nt\lambda_{n,j}} \Longrightarrow \sum_j e^{-t\Lambda_j}, \qquad t>0,$$

locally uniformly away from time zero. A heat trace adds the contributions of all relaxation modes, so this goes beyond convergence of a finite list of eigenvalues.

5. Why the advance matters

Importance: 5/5. The combined claims would identify a broad geometric phase diagram for a central random-surface ensemble and connect that geometry to precisely normalized diffusion and spectra.

The transition at four changes the type of limiting space: a random sphere on one side and a tree on the other. Finite maps above four still have cycles, but those cycles do not survive as macroscopic loops in the limiting metric space.

The joint nature of the results is equally significant. Knowing an abstract limiting distance does not locate cluster boundaries in a chosen conformal coordinate. Knowing the shape of a walk’s trace does not tell us how fast it moves. The family claims these additional identifications in compatible settings, while preserving which conclusions apply to which ensembles.

6. The main idea

Upgrade a contour encoding to geometry

The starting point is an exact combinatorial encoding by words or pairs of contour paths. Brownian limits describe the large-scale fluctuations of those encodings. The geometric challenge is to recover adjacency, paths, and separating loops without overlooking unusual vertices or small regions that create shortcuts.

The subcritical papers keep detailed local records of the encoded map: which pieces meet, how paths enter and leave, and which parts of the surrounding exploration must be retained. Conditional laws of these records connect discrete passage distances with the local continuum field.

Comparisons across many surrounding annuli then control distances and conformal distortion. A final uniform step extends the comparison from typical sampled points to every vertex. The critical paper performs the analogous work at $q=4$, where the field and area normalization have additional endpoint features.

Above four, a different description becomes effective. M7 decomposes the map into blocks joined through a branching tree. It states that the relevant offspring law has finite variance and that accumulated distances through the blocks are uniformly comparable to tree distances. A conditioned branching-tree limit then yields the Brownian continuum random tree. Controlling the block contribution is what connects the abstract encoding to the all-edge metric.

Electrical energy determines the clock

Treat every graph edge as a unit conductor. A voltage function $f$ has energy proportional to

$$\sum_{uv\in E(M_n)}(f(u)-f(v))^2,$$

counting parallel edges separately. This energy determines the random-walk generator together with the stationary vertex masses.

The diffusion papers identify the continuum electrical energy, using the geometric companions and their stated Brownian and LQG inputs. Planar duality helps fix the effective conductivity to one in their conventions. Controlling functions concentrated on very small-mass regions then prevents hidden spectral contributions or unexplained delays.

M2 uses these estimates to pass to inverse operators, eigenvalues, and heat traces. M1 additionally controls whole trajectories, ruling out rapid macroscopic excursions that finite-time observations might miss. M3 performs the energy and path analysis for the tree-weighted ensemble. These steps explain why an exact walk clock requires more than a metric scaling limit.

7. Scope, limitations, and open questions

The map laws, spherical setting, area normalization, and embeddings are specific. The results do not claim universality for every random planar-map ensemble or every way of drawing a map.

The geometric statements use the degree measure, not the uniform vertex measure. For $q\le4$, the claimed distance normalizers are deterministic, but the main statements do not give explicit power laws or convergence rates. No uniformity as $q$ approaches four is asserted.

The walk statements start in stationarity. They do not establish convergence from arbitrary prescribed vertices. The exact factor $n$ uses the rate-one half-edge walk and the continuum energy factor $1/2$; changing either convention changes the clock.

The spectral theorem concerns FK–Ising maps, not every $q$, and retains its stated geometric and Brownian/LQG inputs. Further directions include other initial distributions, explicit metric exponents, quantitative convergence rates, and extensions beyond these finite spherical ensembles.

8. Sources and further reading

• Sheffield, Quantum gravity and inventory accumulation, for the FK encoding and its transition.
• Duplantier, Miller, and Sheffield, Liouville quantum gravity as a mating of trees, for the continuum construction.
• Gwynne and Sun, Scaling limits for the critical Fortuin–Kasteleyn model on a random planar map III, for finite-volume contour limits.
• Berestycki and Gwynne, Random walks on mated-CRT planar maps and Liouville Brownian motion, for earlier time-parametrized diffusion limits in a related ensemble.

M4–M7 provide the geometric claims; M1–M3 provide the diffusion and spectral claims with the conventions described above.

Source manuscripts

• M1. OpenAI, Random Walks on Critical FK–Ising Maps and Liouville Brownian Motion (October 5, 2026).
• M2. OpenAI, Spectral convergence for critical FK–Ising planar maps (October 5, 2026).
• M3. OpenAI, A Linear Clock for Random Walk on Tree-Weighted Planar Maps (October 5, 2026).
• M4. OpenAI, Canonical conformal limits of subcritical FK planar maps (September 24, 2026).
• M5. OpenAI, The critical Liouville quantum sphere and geometric limits of FK maps at q=4 (September 24, 2026).
• M6. OpenAI, Metric-measure limits of subcritical FK and spanning-tree planar maps (September 24, 2026).
• M7. OpenAI, Brownian continuum random tree limits of finite Fortuin–Kasteleyn maps above four (September 24, 2026).

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