UUniqueness of the infinite
NICOLAS CURIEN, GADY KOZMA, VLADAS SIDORAVICIUS, AND LAURENT TOURNIER
Abstract.
Consider the graph obtained by superposition of an independent pair of uniform infinitenon-crossing perfect matchings of the set of integers. We prove that this graph contains at mostone infinite path. Several motivations are discussed.
Figure 1.
The graph obtained by the superposition of two indepen-dent planar perfect matchings on respectively 50, 70 and 150 points.Our main result concerns infinite clusters in the infinite version of thisconstruction. 1.
Introduction
Let us start by defining the model. On the integer line Z we sample indepen-dently for each site x P Z two independent variables ω ` x and ω ´ x uniformly in the set t` , ´ u . We interpret ω ` x (resp. ω ´ x ) as being a parenthesis: an opening parenthesis Mathematics Subject Classification. primary ; secondary .The research of G.K. was supported by the Israel Science Foundation. a r X i v : . [ m a t h . P R ] J a n N. CURIEN, G. KOZMA, V. SIDORAVICIUS, AND L. TOURNIER
Figure 2.
The cluster of 0 in a typical simulation of the infinitemodel. This cycle contains 2936 sites.for ` and closing one for ´ living on the upper half-plane (resp. lower half-plane).By pairing the parentheses, it is then standard that p ω ` x q x P Z yields a perfect match-ing on Z , i.e. an involution of Z without fixed points. This matching is furthermoreplanar, meaning that we can draw arcs between paired points so that the arcs are noncrossing. In most of the following drawings, these arcs will be tents or semi-circles.We repeat this construction twice, once with the parentheses of the upper half plane,and once in lower half plane. After gluing the top and bottom arches on points of Z we are left with a random (multi)graph G with n paths P t , , , . . . u Y t8u manyinfinite clusters (actually bi-infinite paths). Theorem 1.
Either n paths “ almost surely or n paths “ almost surely. Unfortunately, we have not been able to decide which of the two alternativesactually holds and leave this as an open question. Before moving to the proof, letus present a few motivations for studying this model, except of course of its intrinsicbeauty, see Figures 1 and 2.
Random planar maps and Liouville quantum gravity . In the last decades,the geometry of random uniform planar graphs (or maps) has been studied inten-sively and is now quite well understood [3]. The large scale structure of decorated planar maps is much less understood and virtually nothing is rigorously known onthe asymptotic geometry of these objects. One of the simplest model is that of trian-gulations given with a spanning tree. Upon cutting along the spanning tree, such atriangulation can be seen as a binary tree of triangles and an “outside” planar match-ing. If we further blow this tree of triangles we end up with a discrete cycle with twosystems of non-crossing arches (both counted by Catalan numbers), an outside oneidentifying pairs of edges and an inside one connecting points. See Figure 3.
Figure 3.
A triangulation with a spanning tree, cut along its span-ning tree and seen as a cycle with two systems of arches.Hence our model can be seen as a (infinite) simplified version of the above con-struction where the two systems of arches play a symmetric role. We hope that theinsight we get in studying our model will be useful to understand the geometry oftree-decorated maps. In the continuous setting, the idea to glue a pair of randomtrees (which can be seen as continuous limit of planar perfect matchings) appears inthe construction of the Brownian map [5, 4] and in Liouville quantum gravity [6, 1].
Meanders . A meander is a self-avoiding closed loop crossing a horizontal line, seenup to topological equivalence, i.e. up to a homeomorphism of the plane preserving thehorizontal line. In our model, this correspond to gluing two (finite) planar perfectmatchings so that the resulting graph G is connected, in other words, the finiteclusters in our graph G are meanders. The (asymptotic) enumeration of meandersis a notorious difficult open problem in combinatorics and in theoretical physics, see[2]. Our model in contrast allows the explicit computation of several quantities; letus mention for instance that the expected number of circles, i.e. length 2 clusters,going around or passing through a given vertex, turns out to have the surprisinglysimple value π . Figure 4.
Two random meanders over vertices. N. CURIEN, G. KOZMA, V. SIDORAVICIUS, AND L. TOURNIER
Proof sketch . The proof resembles the Burton-Keane argument for percolation.We first note that n paths is constant a.s. due to ergodicity. We then discard thecase n paths ą (infinity included) by a trifurcation argument. The argument doesnot apply to the paths themselves, of course, but to the space between them (hereplanarity is crucial). This argument is presented in §3. The argument that precludesthe case n paths “ is a local modification argument, but is complicated by the factthat, in fact, a local modification cannot change the number of infinite paths. Itcan, however, wire them differently i.e. make a new path by joining the tails of twoexisting paths (after some parity issues are resolved). In §4 we explain why this isenough.As already stated, we believe that n paths “ . In fact, we believe this holds in muchgreater generality, here is the precise formulation: Conjecture.
Let F ` be a random matching of Z (not necessarily planar), stationaryand ergodic with respect to the action of translation by x , for every x P Z . Let F ´ bea second matching of Z , also stationary and ergodic to the action of all translations.Assume F ` and F ´ are independent (but not necessarily identically distributed).Let G be the random graph whose vertex set is Z and edge set is the union of F ` and F ´ . Then G does not have a unique infinite cluster, almost surely.2. Finite configurations
Let Ω “ pt´ , ` u q Z be our set of configurations, and let us consider a randomvariable ω “ p ω ` , ω ´ q in Ω , where ω ` “ p ω ` n q n P Z and ω ´ “ p ω ´ n q n P Z are independentsequences of i.i.d. random variables with uniform distribution in t´ , ` u . We denoteby E ` (resp. E ´ ) the edges belonging to the upper (resp. lower) planar perfectmatching. For x P Z we write C p x q for the cluster of G containing x . Knowingthe restriction of the configuration ω to a finite interval amounts to knowing a finitesub-graph of G plus the orientation of edges exiting the inverval (i.e., the fact thatthey leave the interval toward the left or right). Let us introduce some notationand properties regarding this situation, which will be useful in order to make “local”modifications of ω . Note indeed that changing ω in such a finite interval may haveglobal consequences on G , unless the location of the out-going edges is preserved.Thanks to translation invariance, we consider here integer intervals of the form v , N w ,where N P N , without loss of generality.For S Ă Z , we shall denote the set of configurations on S by(1) Ω S “ pt´ , ` u q S . Let N P N , and η P Ω v ,N w . Let us complete η into ω P Ω by letting ω n “ p` , ` q forall n ă and ω n “ p´ , ´ q for all n ą N ; we may then define the graph G “ G p ω q N n + R n + L n − R n − L } } } } Figure 5.
Left: Notations for numbers of ends of paths going out of v , N w . These ends have a natural cyclic ordering viewed as points ofthe dotted loop. Right: Non-crossing matching realised by the leftconfiguration.as before, and define the numbers of edges outgoing v , N w through the top left andtop right (see Figure 5): n ` L “ n ` L p η q “ t x P v , N w : D y ă , t x, y u P E ` u ,n ` R “ n ` R p η q “ t x P v , N w : D y ą N, t x, y u P E ` u , and similarly n ´ L and n ´ R using E ´ , for the lower part, and the total number ofboundary edges: n tot “ n tot p η q “ n ` L ` n ` R ` n ´ L ` n ´ R . We shall sometimes call these edges “dangling ends”, or ends, of v , N w . These edgesare non-crossing and all of them start inside and end outside the circle of diameter r , N ` s (for any disjoint embedding). Thus, they have a natural cyclic ordering,given by the order on this circle of their last intersection points (or of their only intersection, for the embeddings mentioned in the beginning). Let us number themfrom to n tot starting for instance from the bottommost top left outgoing edgeand following the clockwise order. By associating the two ends of each connectedcomponent in G to each other, η defines a non-crossing matching σ of v , n tot w ; weshall say that η realises σ .Conversely, any non-crossing matching can be realised with any prescribed num-bers of ends provided some room is allowed and the necessary parity conditions hold: Lemma 1.
Let N ě be an integer. For any nonnegative integers a ` , b ` , a ´ , b ´ such that a ` ` b ` , a ´ ` b ´ and N have same parity, and such that n tot : “ a ` ` b ` ` a ´ ` b ´ ď N , and for any non-crossing matching σ : v , n tot w Ñ v , n tot w , there exists N. CURIEN, G. KOZMA, V. SIDORAVICIUS, AND L. TOURNIER N Figure 6.
Example of construction of local configuration realising agiven matching (top left scheme), see Lemma 1. Here, N has to beeven and ě . a configuration η P Ω v ,N w such that p n ` L , n ` R , n ´ L , n ´ R qp η q “ p a ` , b ` , a ´ , b ´ q and thatrealises σ . Note that the parity assumptions are necessary conditions for the conclusion tohold: since the N vertices, together with the n ` L p η q ` n ` R p η q upper ends are pairedby η ` , their total number N ` n ` L p η q ` n ` R p η q has to be even, and similarly for thelower ends. Proof.
Let us first construct a configuration that realises σ using at most two ver-tices for each matched pair, disregarding the value of N . We will now describe theconstruction verbally, but the reader is probably better served by simply checkingFigure 6. We realise every horizontal path (i.e. one that starts on the left and endson the right) with both ends above the line by a ^ shape, every horizontal path withboth ends below the line by a _ shape, and horizontal below-to-above and above-to-below (only one kind may exist, by planarity), by diagonals. Each ’V’ shape usestwo vertices, and each diagonal one. Paths starting and ending on the same side (leftor right) are then inserted between these, in order of containment, with above-to-above and below-to-below inserted as diagonal strips (each using two vertices) andabove-to-below as a ą or ă (each using one vertex).This construction uses a maximum total of n tot ď N vertices. Thus, this construc-tion can be fitted inside v , N w and leaves one free interval provided we used verticesthat are next to each other. Furthermore, the number of used vertices is easily seento have same parity as n ` L p η q ` n ` R p η q , hence the remaining number of vertices has to be even due to the assumptions, which enables making a series of short loops pp` , ` q , p´ , ´ qq to complete the configuration in v , N w . (cid:3) Lemma 2.
Almost surely, the number of ends going out of v , N w is negligible com-pared to N : (2) n tot p ω |v ,N w q N ÝÑ N Ñ8 a.s.Proof. If for k ě we introduce S ` k “ ω ` ` ¨ ¨ ¨ ` ω ´ k and similarly for S ´ then S ` and S ´ are independent simple random walks and we have n tot p ω |v ,N w q “ ` max ď k ď N S ` k ´ min ď k ď N S ` k ˘ ` ` max ď k ď N S ´ k ´ min ď k ď N S ´ k q hence the conclusion comes from the law of large numbers. (cid:3) Finally, the following classical lemma controls probabilities after a local modifica-tion ϕ . Lemma 3 (Finite energy property) . Let S be a finite subset of Z , and ϕ be a mapping Ω S c Ñ Ω S (recall the notation from (1)). For any event C , define r C “ t r ω : ω P C u ,where, for ω P Ω , r ω “ ϕ p ω | S c q S ` ω | S c S c . Then P p r C q ě ´| S | P p C q .Proof. Note indeed that C Ă p C where p C “ t ω P Ω : D η P Ω S , η S ` ω | S c S c P C u isindependent of ω | S , and ω | S is uniform in Ω S , hence P p r C q “ P p p C Xt ω | S “ ϕ p ω | S c quq “ ´| S | P p p C q ě ´| S | P p C q . (cid:3) Trifurcations
In this section we prove that n paths ď almost surely. The proof adapts Burtonand Keane’s classical argument based on an appropriate notion of multifurcation.Let us first, for x P Z , consider the first infinite cluster straddling over x , if it exists: C ` p x q “ C p x q , x “ min t y ą x : D z ă x, t y, z u P E ` , | C p y q| “ 8u , and C ` p x q “ H if there is no such cluster (it actually exists a.s. but we don’t needthis fact), and define similarly C ´ p x q below x , using E ´ .In the following definition d p x, A q is the distance in Z of the point x to a set A Ă Z . N. CURIEN, G. KOZMA, V. SIDORAVICIUS, AND L. TOURNIER
Definition . A point x P Z is called a trifurcation point if it belongs to thefollowing set : Tri “ t x P Z : p ω ` x , ω ´ x q P tp´ , ´ q , p` , ` qu , | C p x q| “ | C ` p x q| “ | C ´ p x q| “ 8 ,d p x ; C ` p x qq ď , d p x ; C ´ p x qq ď , and C p x q , C ` p x q , C ´ p x q are all different u Notice that our definition is a bit more subtle than asking that infinite clusterscome nearby x . Roughly speaking we do not want that one cluster separates thetwo-others. The need for our definition should become clearer in the proof of thefollowing lemma. Lemma 4.
For any N , p Tri Xv , N wq ď ` t C : | C | “ 8 , C X v , N w ‰ Hu . where C denotes a path of our graph G .Proof. Let us view the paths C as embedded in R and compactify R by adding apoint at infinity thus changing the bi-infinite lines into simple closed curves of thetopological sphere R Y t8u that are disjoint except for the point . The proofrevolves around an auxiliary planar graph T , constructed as follows. The verticesof T are the faces of this embedding i.e. regions of R Y t8u whose boundary isa union of paths C , and two vertices are connected in T if they share a path C (infinite, intersecting v , N w ) as a boundary. Jordan’s theorem and the fact that thepaths intersect only at ensures that every path C is incident to exactly two faces,i.e. that T is indeed a graph.We first note that T is a tree. This follows from the fact that its dual graph,which is a set of loops based on the same vertex, admits a spanning tree with onevertex and zero edges, and from [7, Theorem XI.6] (the spanning trees of the dual ofa planar graph G are the complements of the edge-duals of spanning trees of G ).Let now f be some vertex of T and let U be a collection of edges of T inci-dent to f , or in other words, infinite paths contained in the boundary of the face f (not necessarily all of them). Denote by Tri p U q the set of x P Tri such that t C p x q , C ` p x q , C ´ p x qu Ă U .We now claim that | U | ě Tri p U q ` . We prove this by induction on | U | . For | U | “ , , there is nothing to prove as a trifurcation requires | U | ě . Assumetherefore that | U | ě and that x is some trifurcation point that belongs to U .Consider the vertical segment S in R passing through x and whose upper and lowerextremities are the first intersections with C ` and C ´ respectively (here and belowwe write C ˘ as a short for C ˘ p x q ). See figure 7. By the definition of C ˘ , S intersectsno other infinite C , in particular no other C P U . Thus S dissects the face f intothree disjoint parts. For example, one of them has as boundary the part of C ` up x oxC C C C C C SS S R Figure 7.
Mapping of a trifurcation point into S . The vertical seg-ment S meets C , C and C only once. Any curve with that propertymust meet S , and thus have same abscissa x if it is also a verticalsegment. Cf. proof of Lemma 4.to the top of S , the part of S in the top half plane, and then the part of C p x q up to x (and possibly some paths in U zt C p x q , C ` , C ´ u ). Let U , , be the collection ofpaths of U which compose the boundaries of these three parts. By the construction,only C p x q and C ˘ may belong to more than one of U , , and each belongs to exactlytwo, so | U | ` | U | ` | U | “ | U | ` . Further, each y P Tri p U q , other than x itself,must belong to one of the Tri p U i q . Applying the claim inductively to U i we get | Tri p U q| ď ` | Tri p U q| ` | Tri p U q| ` | Tri p U q|ď ` | U | ´ ` | U | ´ ` | U | ´ “ | U | ´ as needed.To conclude, let e be the number of edges of T , and let n be the number ofvertices. Since T is a tree e “ n ´ and then e “ e ´ n ` “ ` ÿ f P T p deg p f q ´ q ě ` ÿ f P T p Tri p f q ` q where Tri p f q “ Tri p U q for U the collection of all paths forming the boundary of f ,and the last inequality is from the previous discussion. Since there is at least one f ,the lemma is proved. (cid:3) Let us justify that trifurcation points may indeed occur if there are at least 3infinite clusters.
Lemma 5.
Assume n paths ě almost surely. Then P p P Tri q ą .Proof. Assume n paths ě almost surely. For M P N , let us consider the event A p“ A M q that at least 3 infinite clusters meet v´ M, M w and that the edge-boundaryof v´ M, M w has size at most M { in G , i.e. n tot “ n tot p ω | v´ M,M w q ď M { . Theprobability of this event goes to 1 as M Ñ 8 due to Lemma 2, hence we may choose
Figure 8.
Pairing the vertices i , i , . . . , i and corresponding config-urations around such that P Tri (bottom). We don’t specify herewhether the vertices lie in the top or bottom half-plane (this is con-sidered in the next figure). The five “cut” symbols on the bottom linesstand for an unspecified distance. In the symmetric cases, applying ahorizontal symmetry leads to trifurcation points where the configura-tion at is p` , ` q . M such that P p A q ą . In order to conclude, let us show that, on A , by changingthe configuration inside v´ M, M w appropriately, we can obtain a trifurcation pointat 0 or at 1.Recall that ω | v´ M,M w encodes a perfect matching over v , n tot w , see Figure 5.Similarly the outside configuration ω | Z zv´ M,M w encodes on v , n tot w another (planar)matching. However this matching is not anymore perfect since any infinite cluster of G touching v´ M, M w will be split into exactly two semi-infinite paths correspondingto two vertices of v , n tot w that are unmatched (the breaking of the infinite path by re-striciting to Z zv´ M, M w might also create finite paths, but these correspond to pairsin the matching). On the event A , we thus have k points ď i ă ¨ ¨ ¨ ă i k ď n tot with k ě that correspond to the two semi-infinite lines coming for each clustertouching v´ M, M w . We first claim that i ´ i ´ , i ´ i ´ , . . . , i k ´ i k ´ ` and n tot ` i ´ i k ` are all even . Indeed, inbetween two consecutive vertices i (cid:96) and i (cid:96) ` the exterior configuration ω | Z zv´ M,M w realises a planar perfect matching and so thenumber of vertices involved must be even.We will now modify the inside configuration ω | v´ M,M w in order to induce a perfectmatching on v , n tot w which pairs i Ø i , i Ø i and i Ø i and all the other pointsto their neighbour. We want to do that in such a way that the three infinite clusters C , C and C coming from the pairing of i , i , . . . , i have a trifurcation point eitherat or at .Depending on the number of ends of infinite paths on the left and right of v´ M, M w ,we may reduce (up to horizontal symmetry) to one of the configurations in the toppart of Figure 8, where we only depict six ends of semi-infinite paths.Depending on the location of the infinite paths, the bottom part of Figure 8 depictshow one sets the configuration around 0 (notice in figure the numbers which indicate + + − + − − Figure 9.
Sketch of the ways to adapt the construction dependingwhether the ends of infinite paths lie in the top or bottom part.which external edge connects to which local point) — due a parity constraint, itmay in fact be later necessary to shift the picture by 1, see below. Then one canaccomodate for the finite clusters between the ends of infinite paths and for theother infinite clusters, in a way similar to Lemma 1 except that it may be necessaryto introduce turns to the semi-infinite paths (i.e. to choose the configuration atsome more vertices), depending whether each end lies on the top or bottom part,as sketched in Figure 9 in the case of two neighbouring infinite paths and readilyextended to a larger number (up to four may be needed).In this way, we set so far the configuration at some vertices around 0 so that,disregarding the other vertices, the ends at the boundary of v´ M, M w are connectedas wished, and P Tri . In this construction, we could choose to use vertices around0 without leaving empty space between them.This procedure uses a number of vertices that has same parity as n ` L ` n ` R (or equiv-alently as n ´ L ` n ´ R ). Since ` v´ M, M w ˘ ` n ` L ` n ` R is even (cf. Lemma 1 above),the number of yet unsettled vertices inside v´ M, M w is even. Up to a possible trans-lation of the previously set configuration by 1, which would produce a trifurcationpoint at 1, we may thus assume that the numbers of unsettled vertices on the rightand on the left of 0 are both even. We then complete the configuration in v´ M, M w by length-two loops ( p` , ` q , p´ , ´ q ) in the empty space. See Figure 10 for anexample.Matching two neighbouring ends of finite paths requires at most 2 vertices, andmatching two ends of semi-infinite paths according to the previous rule requires atmost 6 vertices. The fact that the number of ends is smaller that M { (from thedefinition of A ) implies that there is indeed room for the construction.Thus, on the event A , there is indeed a modification of the configuration within v´ M, M w that leads to a configuration in t P Tri u Y t P Tri u . Since P p A q ą , thefinite energy property (Lemma 3) enables to conclude that P pt P Tri u Y t P Tri uq ą , hence P p P Tri q ą by translation invariance, which proves the lemma. (cid:3) M − M Figure 10.
Example of construction of a trifurcation by local modi-fication. The top left scheme sketches the chosen matching of the endsof finite and infinite (thicker) paths. Since there are an even numberof unsettled vertices on the left of 0 after the boundary ends have beenmatched, the trifurcation can be put at 0 and the configuration com-pleted with short loops (the picture would otherwise have been shiftedby 1).Let us now prove that n paths ď almost surely. Assume by contradiction that n paths ě almost surely. Since the random subset Tri is translation invariant and P p P Tri q ą , it follows from the ergodic theorem that n ` Tri Xv , n w ˘ ÝÑ n Ñ8 P p P Tri q ą a.s.,where the lower bound is Lemma 5. However, Lemma 4 implies that, for all n P N , ` Tri Xv , n w ˘ is larger than the number of infinite clusters that are involved in thesetrifurcations points; and, by the definition of Tri , these clusters all meet v´ , n ` w ,and therefore contribute to at least twice as many edges in the boundary of v´ , n ` w in G . In particular, the size of the boundary of v´ , n ` w in G grows linearly in n .This contradicts Lemma 2. 4. n paths ‰ . In order to complete the proof of our main result, we have to rule out the possibilityof n paths “ which is the goal of this section. Proof.
Assume by contradiction that n paths “ almost surely. Let us denote the twoinfinite paths by C and C ; for instance, we can choose indices so that C is closerto than C (equality cannot happen for parity reasons). For any integers k ă l , define the event A v k,l w that both clusters meet v k, l w : A v k,l w “ t C X v k, l w ‰ H , C X v k, l w ‰ Hu , and for any integer N , let B N be the event that the boundary of v , N w in G has sizesmaller than N : B N “ t n tot p ω |v ,N w q ď N u . Since P p A v ,N w q “ P p A v´ N { ,N { w q Ò and P p B N q Ñ as N Ñ 8 (by Lemma 2), wemay take N to be such that P p A v ,N w q ě { and P p B N q ě { . With N established, let us pick M such that P p A v ,M w q ě ´
14 4 ´ N . For any integer k ă l , let R v k,l w be the maximum distance from v k, l w reached bya path in G starting in v k, l w before coming back to v k, l w for the first time (and R v k,l w “ if no path comes back). Since this distance is finite, there exists r suchthat P p R v ,M w ă r q ě { . Let us finally define the event C “ A v ,N w X B N X A v N ` r,N ` r ` M w X t R v ,N w ă r u X t R v N ` r,N ` r ` M w ă r u . and note already that(3) P p C q ě ´ P p A c v ,N w q ´ P p B cN q ´ P p R v ,N w ě r q ´ P p A c v ,M w qě ´ ´ ´ ´
14 4 ´ N ą . Let us justify that, on the event C , it is possible to modify the configuration inside v , N w in such a way that the event A v N ` r,N ` r ` M w is no longer satisfied.Assume C holds. Due to A v ,N w , the infinite cluster C decomposes into one finitepath that starts and ends in v , N w and two semi-infinite paths that start in v , N w and do not visit v , N w again (by cutting the arbitrarily oriented path C at itsvery first and last visits in v , N w ). Due to R v ,N w ă r the finite path does not visit v N ` r, N ` r ` M w and due to A v N ` r,N ` r ` M w Xt R v N ` r,N ` r ` M w ă r u , exactly one of thesemi-infinite paths does. The same holds for C . We now claim that, by modifyingthe configuration in v , N w it is possible to connect together the ends of these twosemi-infinite paths starting in v , N w and not visiting v N ` r, N ` r ` M w , withoutchanging these semi-infinite paths outside v , N w ; this results in A v N ` r,N ` r ` M w notholding anymore since one of the two infinite clusters is not visiting v N ` r, N ` r ` M w .The possibility of such a modification comes from Lemma 1, whose applicabilityis granted by the event B N and the combination of the following two facts: ‚ due to planarity of the paths, if exactly 4 semi-infinite paths exit v , N w , andonly two of them meet v N ` r, N ` r ` M w , then these two can’t separatethe other two (denoted x and y ) in R Y t8u and therefore, by planarityagain, each of these pairs is separated by an even number of ends along theboundary of v , N w ; ‚ if two ends x, y are separated by an even number of ends, then there existsclearly a non-crossing matching that matches x to y .Therefore Lemma 1 indeed applies to find a configuration that maintains the valuesof n ˘ L and n ˘ R and matches x to y .Let us now conclude. Applying Lemma 3 to the previously described modification,we get, by (3), P p r C q ą ´ N . This contradicts the fact that r C and A v N ` r,N ` r ` M w are disjoint, since that event has probability at least ´ ´ N . (cid:3) Acknowledgments
We would like to thank Omer Angel for fruitful discussions in particular aboutparity arguments in the proof of the last section.
References [1]
B. Duplantier, J. Miller, and S. Sheffield , Liouville quantum gravity as a mating oftrees , arXiv:1409.7055 .[2] E. Guitter , Physique statistique de modeles contraints sur reseaux reguliers et aleatoires : despliages aux meandres , archive-ouvertes.fr/00006781 .[3] J.-F. Le Gall , Random geometry on the sphere , Proceedings of the ICM 2014.[4] ,
The topological structure of scaling limits of large planar maps , Invent. Math., 169 (2007),pp. 621–670.[5]
J.-F. Marckert and A. Mokkadem , Limit of normalized quadrangulations: the Brownianmap , Ann. Probab., 34 (2006), pp. 2144–2202.[6]
S. Sheffield , Quantum gravity and inventory accumulation , arXiv:1108.2241 .[7] W. Tutte , Graph theory , vol. 21 of Encyclopedia of Mathematics and its Applications, Addison-Wesley, London, 1984.
Laboratoire de mathématique, Université Paris-Sud, 91400 Orsay, France
E-mail address : [email protected] Weizmann institute of Science, 76100, Rehovot, Israel
E-mail address : [email protected] Courant Institute of Mathematical Sciences, New York, NYU-ECNU Instituteof Mathematical Sciences at NYU Shanghai, Cemaden, São José dos Campos.
E-mail address : [email protected] LAGA, Université Paris 13, Sorbonne Paris Cité, CNRS, UMR 7539, 93430 Villeta-neuse, France.
E-mail address ::