1. The question
A perfect matching of a graph pairs every vertex with exactly one neighbor. Draw two independent perfect matchings on the same square grid. An edge selected twice becomes a doubled edge; the other selected edges arrange themselves into alternating loops.
These double-dimer loops are predicted to approach a conformally invariant continuum ensemble called nested CLE₄. Earlier work already identified extensive topological information about that limit. Family 226 asks for the stronger geometric statement: do the actual loop curves converge, including every nested generation and the order in which each loop traverses its trace?
The distinction is visible in a simple example. A loop can make a long, extremely thin excursion and return without enclosing any of a chosen finite set of test points. A record of which points lie inside the loop will miss the excursion. Even identifying the limiting trace does not exclude running backward along part of it before continuing forward.
The manuscript claims to rule out these possibilities for the standard half-plane Temperleyan double-dimer model, thereby upgrading existing topological convergence to convergence of complete loop collections as curves.
2. Background and essential definitions
Matchings and the half-plane law
A dimer is an edge occupying its two endpoints. A dimer covering, or perfect matching, uses exactly one incident edge at every vertex. On a finite graph that admits coverings, the uniform dimer law gives every perfect matching equal probability.
The infinite domain here is the upper half-plane
An infinite-volume law needs a boundary convention. The manuscript takes simple lattice polygons in the coarse grid , keeps the fine-grid graph in their closure, and deletes one coarse boundary vertex, called the root. These are the specified Temperleyan graphs. Their polygons exhaust the half-plane while their roots escape to infinity. The local limit of uniform matchings defines the law used in the theorem.
Sample two matchings independently with this same law. In their superposition, each vertex has degree two if a doubled edge is counted twice. Under this particular half-plane law, the components are almost surely doubled edges and finite simple even cycles. Remove the doubled edges and call the collection of cycles .
The loops alternate between edges of the first and second matchings. They are treated as unrooted and unoriented: neither a starting point nor a preferred direction is part of the final observable.
What nested CLE₄ describes
A conformal loop ensemble is a random collection of loops whose law transforms naturally under conformal maps. For parameter four, the individual loops are simple and disjoint. The nonnested ensemble gives the outermost layer. Nested CLE₄ repeats the same construction independently inside each loop, conditional on its boundary, and continues recursively.
The source uses the standard normalization whose outermost ensemble can be constructed from a Brownian loop soup of central charge one. This identifies the conventional CLE₄ law, rather than a new loop process with an adjustable intensity. A Brownian loop soup is a Poisson collection of Brownian loops; its cluster boundaries provide this continuum construction.
Every fixed positive size cutoff leaves only finitely many loops in the disk, even though there are loops at arbitrarily small scales. This local finiteness is what makes comparison of whole collections possible.
A topology that remembers curves
To compare the unbounded half-plane globally, use
which maps it to the unit disk. All distances and diameters in the theorem are measured after this map. Thus infinity is handled through a compact conformal coordinate.
Two loops are close when one can choose starting points and either orientation, then continuously change traversal speed so that corresponding points remain uniformly close. Pauses do not matter; a backward detour cannot generally be removed by a monotone change of speed.
Two loop collections are close if their macroscopic members can be matched one-to-one, with close matched curves, leaving only small loops unmatched. Distinct loops remain distinct entries even if their limiting images coincide. This retains multiplicity as well as geometry.
3. Development history and previous partial results
Kasteleyn and Temperley–Fisher developed exact dimer enumeration through determinant and Pfaffian methods. Lieb’s transfer-matrix approach offered another way to exploit the model’s algebra. These exact formulas made dimers one of the most tractable interacting planar systems.
At the field level, Kenyon proved Gaussian-free-field limits for dimer heights in suitable Temperleyan domains. This strongly motivates a connection to CLE₄, whose loops are related to Gaussian-field level lines. But a random distribution does not determine the convergence of discrete contours without extra geometric arguments.
Kenyon and Wilson identified limiting boundary pairing probabilities for double dimers and their agreement with Gaussian-field contour predictions. Kenyon’s loop observables then accessed conformally invariant topological data using matrix-valued weights.
Dubédat connected such observables to CLE₄ through isomonodromic tau-functions, analytical quantities encoding how a differential equation changes when punctures move. Basok and Chelkak recovered probabilities of cylindrical events from the corresponding expansion, subject to a continuum tail estimate. Bai and Wan supplied the needed tail control for the nested ensemble.
The source imports the precise half-plane formulation in Basok and Izyurov’s 2025 paper. It states convergence of reduced loop records relative to finitely many punctures, including all nested generations. That work also treats fluctuations of nesting counts and their limiting field.
The proposed advance therefore concerns the remaining geometric information. It does not claim that the CLE₄ candidate or its puncture-record law was previously unknown. The task is to retain entire paths while preventing features invisible to every fixed finite record.
4. What the new manuscripts claim
For the half-plane Temperleyan law just defined, the manuscript claims tightness and full convergence
where is standard nested CLE₄ in . The arrow denotes convergence in distribution in the loop-collection topology described above.
Several parts of the claim matter separately. It holds along the full mesh limit, rather than only a selected subsequence. It includes all nested generations, rather than just the outermost boundaries. Every macroscopic loop is matched as a curve, and no extra macroscopic limiting loop is left unidentified.
The target loops lie strictly inside the open disk after applying . The statement includes tightness in that interior collection space, so boundary contact is not simply ignored by first working in a larger closed-disk space.
For comparison, the imported topological theorem works as follows. Choose finitely many distinct interior punctures. Record the homotopy classes of loops relative to them, keeping multiplicities but deleting loops surrounding at most one puncture. Homotopy means deformation without crossing a puncture. This reduced record converges to its CLE₄ counterpart. The new theorem supplements that information with control at successively finer geometric resolutions.
The result concerns unparametrized curves, in the precise sense of allowing changes of speed. It does not attach a limiting travel time, arc-length normalization, or occupation measure to each loop. Those would be additional observables, analogous to the separate natural-measure question discussed in family 223.
5. Why the advance matters
Importance: 4/5. Double dimers are a central exactly solvable loop model, and full geometric convergence is a major strengthening of topological identification. The result would complete the stated half-plane Temperleyan version of a prominent CLE₄ scaling-limit prediction.
The stronger topology permits questions about where a loop goes and how it is traversed, not merely which selected points it surrounds. It also ensures that the many nested loops fit into one continuum ensemble with the correct multiplicities.
The rating reflects a substantial advance with potentially reusable methods, while retaining the theorem’s particular domain and boundary law. It is not a claim for every dimer graph or every possible finite-domain approximation.
An instructive general lesson is that exact algebraic observables and geometric compactness solve different parts of a scaling-limit problem. Here the earlier algebraic and topological information identifies the candidate; the new estimates are designed to make that identification valid in a stronger space.
6. The main idea
The manuscript starts in a finite rectangle with a deleted corner, chosen to match the Temperleyan convention. It counts matchings one column at a time. The state along a vertical seam records which dimers cross it; after a change of variables, these states behave like configurations of particles on transverse sites.
The transfer in a sector with a fixed number of particles is an exterior power of a one-particle matrix. This expresses the many-particle calculation through the spectrum of a simpler operator. Moving far from the typical particle number has a quadratic spectral cost.
That cost yields the first geometric estimate: the total number of disjoint crossings of a fixed-width slab is tight. Importantly, repeated crossings by one loop count separately. Controlling only the number of distinct loops would allow a single loop to oscillate arbitrarily often. Bounds of this kind give compactness after suitable reparametrization.
A second estimate controls signed crossing counts through a short transverse interval. The sign records which way an arc crosses. A local two-particle insertion has matrix coefficients that are small quadratically in the interval length. A separate planar positivity identity converts this signed algebra into a probability bound, with weights that decay down nested chains.
This estimate excludes a thin excursion canceling against a nearby return path. Wilson’s spanning-tree algorithm and the Temperley correspondence then transfer the rectangle estimates to bounded windows of the half-plane. These are localization tools; the theorem is not restricted to the auxiliary rectangles.
Finally, the argument couples a subsequential curve limit with its known finite-puncture records. A progressively finer puncture grid identifies each visible trace and rules out invisible macroscopic loops. Another refinement excludes backward traversal along an already identified trace. This separates three tasks that a winding number alone cannot settle: identifying the image, its multiplicity, and its traversal order.
7. Scope, limitations, and open questions
The theorem uses two independent uniform matchings with the specified half-plane Temperleyan law. Changing edge weights, boundary conventions, lattice, or domain requires a separate statement. Conformal invariance of the proposed continuum limit does not by itself transfer the discrete theorem to arbitrary lattice approximations of other domains.
The convergence is measured after the fixed map to the disk. It is consequently a precise compactified geometric statement, rather than uniform Euclidean control over all of the unbounded half-plane.
No convergence rate or natural parametrization is asserted. Nor does convergence of these loops automatically settle every joint observable involving the two original matchings or their heights. The source takes established topological convergence and standard CLE properties as inputs; its claim is the geometric upgrade built on them.
Broader domains and boundary laws, quantitative rates, and simultaneous convergence with naturally normalized loop measures are directions beyond the displayed result. The mathematical claims here are reported as stated in the source.
8. Sources and further reading
The historical papers linked above trace the progression from height fields and boundary pairings to topological loop records. Basok–Izyurov provides the exact half-plane input cited by the manuscript.
For the continuum ensemble and its normalization, see Sheffield and Werner, the Markovian characterization and loop-soup construction of CLE. The source below supplies the claimed geometric estimates and full curve-limit statement.
Source manuscripts
- M1. OpenAI, The Curve Scaling Limit of Half-Plane Double Dimers (September 23, 2026).
Sources consulted on 11 October 2026 at repository revision fd4aeeb. Download the manuscript citations.