Asymptotics for Rooted Bipartite Planar Maps and Scaling Limits of Two-Type Spatial Trees

We prove some asymptotic results for the radius and the profile of large random bipartite planar maps. Using a bijection due to Bouttier, Di Francesco and Guitter between rooted bipartite planar maps and certain two-type trees with positive labels, we derive our results from a conditional limit theorem for two-type spatial trees. Finally we apply our estimates to separating vertices of bipartite planar maps: with probability close to one when n tends to infinity, a random $2k$-angulation with n faces has a separating vertex whose removal disconnects the map into two components each with size greater that $n^{1/2 - \varepsilon}$.


Introduction
The main goal of the present work is to investigate asymptotic properties of large rooted bipartite planar maps under the so-called Boltzmann distributions. This setting includes as a special case the asymptotics as n → ∞ of the uniform distribution over rooted 2κ-angulations with n faces, for any fixed integer κ ≥ 2. Boltzmann distributions over planar maps that are both rooted and pointed have been considered recently by Marckert & Miermont (16) who discuss in particular the profile of distances from the distinguished point in the map. Here we deal with rooted maps and we investigate distances from the root vertex, so that our results do not follow from the ones in (16), although many statements look similar. The specific results that are obtained in the present work have found applications in the paper (13), which investigates scaling limits of large planar maps.
Let us briefly discuss Boltzmann distributions over rooted bipartite planar maps. We consider a sequence q = (q i ) i≥1 of weights (nonnegative real numbers) satisfying certain regularity properties. Then, for each integer n ≥ 2, we choose a random rooted bipartite map M n with n faces whose distribution is specified as follows : the probability that M n is equal to a given bipartite planar map m is proportional to where f 1 , . . . , f n are the faces of m and deg(f i ) is the degree (that is the number of adjacent edges) of the face f i . In particular we may take q κ = 1 and q i = 0 for i = κ, and we get the uniform distribution over rooted 2κ-angulations with n faces. Theorem 2.5 below provides asymptotics for the radius and the profile of distances from the root vertex in the random map M n when n → ∞. The limiting distributions are described in terms of the one-dimensional Brownian snake driven by a normalized excursion. In particular if R n denotes the radius of M n (that is the maximal distance from the root), then n −1/4 R n converges to a multiple of the range of the Brownian snake. In the special case of quadrangulations (q 2 = 1 and q i = 0 for i = 2), these results were obtained earlier by Chassaing & Schaeffer (4). As was mentioned above, very similar results have been obtained by Marckert & Miermont (16) in the setting of Boltzmann distributions over rooted pointed bipartite planar maps, but considering distances from the distinguished point rather than from the root.
Similarly as in (4) or (16), bijections between trees and maps serve as a major tool in our approach. In the case of quadrangulations, these bijections were studied by Cori & Vauquelin (5) and then by Schaeffer (19). They have been recently extended to bipartite planar maps by Bouttier, Di Francesco & Guitter (3). More precisely, Bouttier, Di Francesco & Guitter show that bipartite planar maps are in one-to-one correspondence with well-labelled mobiles, where a well-labelled mobile is a two-type spatial tree whose vertices are assigned positive labels satisfying certain compatibility conditions (see section 2.4 for a precise definition). This bijection has the nice feature that labels in the mobile correspond to distances from the root in the map. Then the above mentioned asymptotics for random maps reduce to a limit theorem for welllabelled mobiles, which is stated as Theorem 3.3 below. This statement can be viewed as a conditional version of Theorem 11 in (16). The fact that (16) deals with distances from the distinguished point in the map (rather than from the root) makes it possible there to drop the positivity constraint on labels. In the present work this constraint makes the proof significantly more difficult. We rely on some ideas from Le Gall (12) who established a similar conditional theorem for well-labelled trees. Although many arguments in Section 3 below are analogous to the ones in (12), there are significant additional difficulties because we deal with two-type trees and we condition on the number of vertices of type 1.
A key step in the proof of Theorem 3.3 consists in the derivation of estimates for the probability that a two-type spatial tree remains on the positive half-line. As another application of these estimates, we derive some information about separating vertices of uniform 2κ-angulations. We show that with a probability close to one when n → ∞ a random rooted 2κ-angulation with n faces will have a vertex whose removal disconnects the map into two components both having size greater that n 1/2−ε . Related combinatorial results are obtained in (2). More precisely, in a variety of different models, Proposition 5 in (2) asserts that the second largest nonseparable component of a random map of size n has size at most O(n 2/3 ). This suggests that n 1/2−ε in our result could be replaced by n 2/3−ε .
The paper is organized as follows. In section 2, we recall some preliminary results and we state our asymptotics for large random rooted planar maps. Section 3 is devoted to the proof of Theorem 3.3 and to the derivation of Theorem 2.5 from asymptotics for well-labelled mobiles. Finally Section 4 discusses the application to separating vertices of uniform 2κ-angulations.

Boltzmann laws on planar maps
A planar map is a proper embedding, without edge crossings, of a connected graph in the 2dimensional sphere S 2 . Loops and multiple edges are allowed. A planar map is said to be bipartite if all its faces have even degree. In this paper, we will only be concerned with bipartite maps. The set of vertices will always be equipped with the graph distance : if a and a ′ are two vertices, d(a, a ′ ) is the minimal number of edges on a path from a to a ′ . If M is a planar map, we write F M for the set of its faces, and V M for the set of its vertices.
A pointed planar map is a pair (M, τ ) where M is a planar map and τ is a distinguished vertex. Note that, since M is bipartite, if a and a ′ are two neighbouring vertices, then we have |d(τ, a) − d(τ, a ′ )| = 1. A rooted planar map is a pair (M, e ) where M is a planar map and e is a distinguished oriented edge. The origin of e is called the root vertex. At last, a rooted pointed planar map is a triple (M, e, τ ) where (M, τ ) is a pointed planar map and e is a distinguished non-oriented edge. We can always orient e in such a way that its origin a and its end point a ′ satisfy d(τ, a ′ ) = d(τ, a) + 1. Note that a rooted planar map can be interpreted as a rooted pointed planar map by choosing the root vertex as the distinguished point.
Two pointed maps (resp. two rooted maps, two rooted pointed maps) are identified if there exists an orientation-preserving homeomorphism of the sphere that sends the first map to the second one and preserves the distinguished point (resp. the root edge, the distinguished point and the root edge). Let us denote by M p (resp. M r , M r,p ) the set of all pointed bipartite maps (resp. the set of all rooted bipartite maps, the set of all rooted pointed bipartite maps) up to the preceding identification.
Let us recall some definitions and propositions that can be found in (16). Let q = (q i , i ≥ 1) be a sequence of nonnegative weights such that q i > 0 for at least one i > 1. For any planar map where we have written deg(f ) for the degree of the face f . We require q to be admissible that is For k ≥ 1, we set N (k) = 2k−1 k−1 . For every weight sequence q, we define Let R q be the radius of convergence of the power series f q . Consider the equation From Proposition 1 in (16), a sequence q is admissible if and only if equation (1) has at least one solution, and then Z q is the solution of (1) that satisfies Z 2 q f ′ q (Z q ) ≤ 1. An admissible weight sequence q is said to be critical if it satisfies which means that the graphs of the functions x → f q (x) and x → 1 − 1/x are tangent at the left of x = Z q . Furthermore, if Z q < R q , then q is said to be regular critical. This means that the graphs are tangent both at the left and at the right of Z q . In what follows, we will only be concerned with regular critical weight sequences.
Let q be a regular critical weight sequence. We define the Boltzmann distribution B r,p q on the set M r,p by Let us now define Z Note that the sum is over the set M r of all rooted bipartite planar maps. From the fact that Z q < ∞ it easily follows that Z (r) q < ∞. We then define the Boltzmann distribution B r q on the set M r by Let us turn to the special case of 2κ-angulations. A 2κ-angulation is a bipartite planar map such that all faces have a degree equal to 2κ. If κ = 2, we recognize the well-known quadrangulations. Let us set We denote by q κ the weight sequence defined by q κ = α κ and q i = 0 for every i = κ. It is proved in Section 1.5 of (16) that q κ is a regular critical weight sequence, and For every n ≥ 1, we denote by U n κ (resp. U n κ ) the uniform distribution on the set of all rooted pointed 2κ-angulations with n faces (resp. on the set of all rooted 2κ-angulations with n faces). We have

Two-type spatial Galton-Watson trees
We start with some formalism for discrete trees. Set where N = {1, 2, . . .} and by convention N 0 = {∅}. An element of U is a sequence u = u 1 . . . u n , and we set |u| = n so that |u| represents the generation of u. In particular |∅| = 0. If u = u 1 . . . u n and v = v 1 . . . v m belong to U, we write uv = u 1 . . . u n v 1 . . . v m for the concatenation of u and v. In particular ∅u = u∅ = u. If v is of the form v = uj for u ∈ U and j ∈ N, we say that v is a child of u, or that u is the father of v, and we write u =v. More generally if v is of the form v = uw for u, w ∈ U, we say that v is a descendant of u, or that u is an ancestor of v. The set U comes with the natural lexicographical order such that u v if either u is an ancestor of v, or if u = wa and v = wb with a ∈ U * and b ∈ U * satisfying a 1 < b 1 , where we have set U * = U \ {∅}. And we write u ≺ v if u v and u = v.
A plane tree T is a finite subset of U such that We denote by T the set of all plane trees.
Let T be a plane tree and let ζ = #T − 1. The search-depth sequence of T is the sequence u 0 , u 1 , . . . , u 2ζ of vertices of T which is obtained by induction as follows. First u 0 = ∅, and then for every i ∈ {0, 1, . . . , 2ζ − 1}, u i+1 is either the first child of u i that has not yet appeared in the sequence u 0 , u 1 , . . . , u i , or the father of u i if all children of u i already appear in the sequence u 0 , u 1 , . . . , u i . It is easy to verify that u 2ζ = ∅ and that all vertices of T appear in the sequence u 0 , u 1 , . . . , u 2ζ (of course some of them appear more that once). We can now define the contour function of T . For every k ∈ {0, 1, . . . , 2ζ}, we let C(k) denote the distance from the root of the vertex u k . We extend the definition of C to the line interval [0, 2ζ] by interpolating linearly between successive integers. Clearly T is uniquely determined by its contour function C.
A discrete spatial tree is a pair (T , U ) where T ∈ T and U = (U v , v ∈ T ) is a mapping from the set T into R. If v is a vertex of T , we say that U v is the label of v. We denote by Ω the set of all discrete spatial trees. If (T , U ) ∈ Ω we define the spatial contour function of (T , U ) as follows.
First if k is an integer, we put V (k) = U u k with the preceding notation. We then complete the definition of V by interpolating linearly between successive integers. Clearly (T , U ) is uniquely determined by the pair (C, V ).
Let (T , U ) ∈ Ω. We interpret (T , U ) as a two-type (spatial) tree by declaring that vertices of even generations are of type 0 and vertices of odd generations are of type 1. We then set T 0 = {u ∈ T : |u| is even}, Let us turn to random trees. We want to consider a particular family of two-type Galton-Watson trees, in which vertices of type 0 only give birth to vertices of type 1 and vice-versa. Let µ = (µ 0 , µ 1 ) be a pair of offspring distributions, that is a pair of probability distributions on Z + . If m 0 and m 1 are the respective means of µ 0 and µ 1 we assume that m 0 m 1 ≤ 1 and we exclude the trivial case µ 0 = µ 1 = δ 1 where δ 1 stands for the Dirac mass at 1. We denote by P µ the law of a two-type Galton-Watson tree with offspring distribution µ, meaning that for every where t 0 (resp. t 1 ) is as above the set of all vertices of t with even (resp. odd) generation. The fact that this formula defines a probability measure on T is justified in (16).
Let us now recall from (16) how one can couple plane trees with a spatial displacement in order to turn them into random elements of Ω. To this end, let ν k 0 , ν k 1 be probability distributions on R k for every k ≥ 1. We set ν = (ν k 0 , ν k 1 ) k≥1 . For every T ∈ T and x ∈ R, we denote by R ν,x (T , dU ) the probability measure on R T which is characterized as follows. Let (Y u , u ∈ T ) be a family of independent random variables such that for u ∈ T with k u (T ) = k, Y u = (Y u1 , . . . , Y uk ) is distributed according to ν k 0 if u ∈ T 0 and according to ν k 1 if u ∈ T 1 . We set X ∅ = x and for every v ∈ T \ {∅}, where ]∅, v] is the set of all ancestors of v distinct from the root ∅. Then R ν,x (T , dU ) is the law of (X v , v ∈ T ). We finally define for every x ∈ R a probability measure P µ,ν,x on Ω by setting P µ,ν,x (dT dU ) = P µ (dT )R ν,x (T , dU ).

The Brownian snake and the conditioned Brownian snake
Let x ∈ R. The Brownian snake with initial point x is a pair (e, r x ), where e = (e(s), 0 ≤ s ≤ 1) is a normalized Brownian excursion and r x = (r x (s), 0 ≤ s ≤ 1) is a real-valued process such that, conditionally given e, r x is Gaussian with mean and covariance given by We know from (10) that r x admits a continuous modification. From now on we consider only this modification. In the terminology of (10) r x is the terminal point process of the one-dimensional Brownian snake driven by the normalized Brownian excursion e and with initial point x.
Write P for the probability measure under which the collection (e, r x ) x∈R is defined. Let x > 0. From Proposition 4.9 in (18) we have for every s ∈ [0, 1], Since r 0 is continuous under the probability measure P, we get that We may then define for every x > 0 a pair (e x , r x ) which is distributed as the pair (e, r x ) under the conditioning that inf t∈[0,1] r x (t) ≥ 0.
We equip C([0, 1], R) 2 with the norm (f, g) = f u ∨ g u where f u stands for the supremum norm of f . The following theorem is a consequence of Theorem 1.1 in (15).
Theorem 2.1. There exists a pair (e 0 , r 0 ) such that (e x , r x ) converges in distribution as x ↓ 0 towards (e 0 , r 0 ).
The pair (e 0 , r 0 ) is the so-called conditioned Brownian snake with initial point 0.

The Bouttier-Di Francesco-Guitter bijection
We start with a definition. A (rooted) mobile is a two-type spatial tree (T , U ) whose labels U v only take integer values and such that the following properties hold : Furthermore, if U v ≥ 1 for every v ∈ T , then we say that (T , U ) is a well-labelled mobile.
Let T mob 1 denote the set of all mobiles such that U ∅ = 1. We will now describe the Bouttier-Di Francesco-Guitter bijection from T mob 1 onto M r,p . This bijection can be found in section 2 in (3). Note that (3) deals with pointed planar maps rather than with rooted pointed planar maps. It is however easy to verify that the results described below are simple consequences of (3).
Let (T , U ) ∈ T mob 1 . Recall that ζ = #T − 1. Let u 0 , u 1 , . . . , u 2ζ be the search-depth sequence of T . It is immediate to see that u k ∈ T 0 if k is even and that u k ∈ T 1 if k is odd. The search-depth sequence of T 0 is the sequence w 0 , w 1 , . . . , w ζ defined by w k = u 2k for every k ∈ {0, 1, . . . , ζ}. Notice that w 0 = w ζ = ∅. Although (T , U ) is not necessarily well labelled, we may set for every v ∈ T , U + v = U v − min{U w : w ∈ T } + 1, and then (T , U + ) is a well-labelled mobile. Notice that min{U + v : v ∈ T } = 1. Suppose that the tree T is drawn in the plane and add an extra vertex ∂. We associate with (T , U + ) a bipartite planar map whose set of vertices is and whose edges are obtained by the following device : for every k ∈ {0, 1, . . . , ζ}, • if U + w k = 1, draw an edge between w k and ∂ ; • if U + w k ≥ 2, draw an edge between w k and the first vertex in the sequence w k+1 , . . . , w ζ−1 , w 0 , w 1 , . . . , w k−1 whose label is U + w k − 1.
Notice that condition (b) in the definition of a mobile entails that U + w k+1 ≥ U + w k − 1 for every k ∈ {0, 1, . . . , ζ − 1} and recall that min{U + w 0 , U + w 1 , . . . , U + w ζ−1 } = 1. The preceding properties ensure that whenever U + w k ≥ 2 there is at least one vertex among w k+1 , . . . , w ζ−1 , w 0 , . . . , w k−1 with label U + w k − 1. The construction can be made in such a way that edges do not intersect (see section 2 in (3) for an example). The resulting planar graph is a bipartite planar map. We view this map as a rooted pointed planar map by declaring that the distinguished vertex is ∂ and that the root edge is the one corresponding to k = 0 in the preceding construction.
It follows from (3) that the preceding construction yields a bijection Ψ r,p between T mob 1 and M r,p . Furthermore it is not difficult to see that Ψ r,p satisfies the following two properties : let (T , U ) ∈ T mob 1 and let M = Ψ r,p ((T , U )), In particular U + ∅ = 1. This implies that the root edge of the planar map Ψ r,p ((T , U )) contains the distinguished point ∂. Then Ψ r,p ((T , U )) can be identified to a rooted planar map, whose root is an oriented edge between the root vertex ∂ and w 0 . Write T mob 1 for the set of all welllabelled mobiles such that U ∅ = 1. Thus Ψ r,p induces a bijection Ψ r from the set T mob 1 onto the set M r . Furthermore Ψ r satisfies the following two properties : let (T , U ) ∈ T mob 1 and let M = Ψ r ((T , U )),

Boltzmann distribution on two-type spatial trees
Let q be a regular critical weight sequence. We recall the following definitions from (16). Let µ q 0 be the geometric distribution with parameter f q (Z q ) that is and let µ q 1 be the probability measure defined by From (16), we know that µ q 1 has small exponential moments, and that the two-type Galton-Watson tree associated with µ q = (µ q 0 , µ q 1 ) is critical. Also, for every k ≥ 0, let ν k 0 be the Dirac mass at 0 ∈ R k and let ν k 1 be the uniform distribution on the set A k defined by We can say equivalently that ν k 1 is the law of (X 1 , . . . , X 1 + . . . + X k ) where (X 1 , . . . , X k+1 ) is uniformly distributed on the set B k defined by Notice that #A k = #B k = N (k + 1). We set ν = ((ν k 0 , ν k 1 )) k≥1 . The following result is Proposition 10 in (16). However, we provide a short proof for the sake of completeness.
q is the image of the probability measure P µ q ,ν,1 under the mapping Ψ r,p .
Proof : By construction, the probability measure P µ q ,ν,1 is supported on the set T mob We have by the choice of ν, We set m = Ψ r,p ((t, u)). We have from the property (ii) satisfied by Ψ r,p , which leads us to the desired result.
Let us introduce some notation. Since µ q 0 (1) > 0, we have P µ q (#T 1 = n) > 0 for every n ≥ 1. Then we may define, for every n ≥ 1 and x ∈ R, Furthermore, we set for every (T , U ) ∈ Ω, with the convention min ∅ = ∞. Finally we define for every n ≥ 1 and x ≥ 0, Proof : The first assertion is a simple consequence of Proposition 2.2 together with the property (i) satisfied by Ψ r,p . Recall from section 2.1 that we can identify the set M r to a subset of M r,p in the following way.
. We then verify that µ κ 0 is the geometric distribution with parameter 1/κ and that µ κ 1 is the Dirac mass at κ − 1. Recall the notation U n κ and U n κ .
Corollary 2.4. The probability measure U n κ is the image of P n µ κ ,ν,1 under the mapping Ψ r,p . The probability measure U n κ is the image of P n µ κ ,ν,1 under the mapping Ψ r .

Statement of the main result
We first need to introduce some notation. Let M ∈ M r . We denote by o its root vertex. The radius R M is the maximal distance between o and another vertex of M that is The normalized profile of M is the probability measure λ M on {0, 1, 2, . . .} defined by Note that R M is the supremum of the support of λ M . It is also convenient to introduce the rescaled profile. If M has n faces, this is the probability measure on R + defined by At last, if q is a regular critical weight sequence, we set Recall from section 2.3 that (e, r 0 ) denotes the Brownian snake with initial point 0.
Theorem 2.5. Let q be a regular critical weight sequence.
(i) The law of n −1/4 R M under the probability measure B r q (· | #F M = n) converges as n → ∞ to the law of the random variable (ii) The law of the random measure λ (n) M under the probability measure B r q (· | #F M = n) converges as n → ∞ to the law of the random probability measure I defined by where a is a vertex chosen uniformly at random among all vertices of M , under the probability measure B r q (· | #F M = n) converges as n → ∞ to the law of the random variable In the case q = q κ , the constant appearing in Theorem 2.5 is (4κ(κ − 1)/9) 1/4 . It is equal to (8/9) 1/4 when κ = 2. The results stated in Theorem 2.5 in the special case q = q 2 were obtained by Chassaing & Schaeffer (4) (see also Theorem 8.2 in (12)).
Obviously Theorem 2.5 is related to Theorem 3 proved by Marckert & Miermont (16). Note however that (16) deals with rooted pointed maps instead of rooted maps as we do and studies distances from the distinguished point of the map rather than from the root vertex.
3 A conditional limit theorem for two-type spatial trees Recall first some notation. Let q be a regular critical weight sequence, let µ q = (µ q 0 , µ q 1 ) be the pair of offspring distributions associated with q and let ν = (ν k 0 , ν k 1 ) k≥1 be the family of probability measures defined before Proposition 2.2.
If (T , U ) ∈ Ω, we denote by C its contour function and by V its spatial contour function.
Recall that C([0, 1], R) 2 is equipped with the norm (f, g) = f u ∨ g u . The following result is a special case of Theorem 11 in (16).
converges as n → ∞ to the law of (e, r 0 ). The convergence holds in the sense of weak convergence of probability measures on C([0, 1], R) 2 .
Note that Theorem 11 in (16) deals with the so-called height-process instead of the contour process. However, we can deduce Theorem 3.1 from (16) by classical arguments (see e.g. (11)).
In this section, we will prove a conditional version of Theorem 3.1. Before stating this result, we establish a corollary of Theorem 3.1. To this end we set Notice that this conditioning makes sense since µ q 0 (1) > 0. We may also define for every n ≥ 1, converges as n → ∞ to the law of (e, r 0 ). The convergence holds in the sense of weak convergence of probability measures on C([0, 1], R) 2 .
Proof : We first introduce some notation. If (T , U ) ∈ Ω and w 0 ∈ T , we define a spatial tree As a consequence of Theorem 11 in (16), the law under Q n µ q ,ν of  converges as n → ∞ to the law of (e, r 0 ). We then easily get the desired result.
Recall that (e 0 , r 0 ) denotes the conditioned Brownian snake with initial point 0.
Theorem 3.3. Let q be a regular critical weight sequence. For every x ≥ 0, the law under P converges as n → ∞ to the law of (e 0 , r 0 ). The convergence holds in the sense of weak convergence of probability measures on C([0, 1], R) 2 .
To prove Theorem 3.3, we will follow the lines of the proof of Theorem 2.2 in (12). From now on, we set µ = µ q to simplify notation.

Rerooting two-type spatial trees
If T ∈ T, we say that a vertex v ∈ T is a leaf of T if k v (T ) = 0 meaning that v has no child. We denote by ∂T the set of all leaves of T and we write ∂ 0 T = ∂T ∩ T 0 for the set of leaves of T which are of type 0.
Let us recall some notation that can be found in section 3 in (12). Recall that U * = U \ {∅}.
If v 0 ∈ U * and T ∈ T are such that v 0 ∈ T , we define k = k(v 0 , T ) and l = l(v 0 , T ) in the following way. Write ζ = #T − 1 and u 0 , u 1 , . . . , u 2ζ for the search-depth sequence of T . Then we set which means that k is the time of the first visit of v 0 in the evolution of the contour of T and that l is the time of the last visit of v 0 . Note that l ≥ k and that l = k if and only if v 0 ∈ ∂T .
Then there exists a unique plane tree T (v 0 ) ∈ T whose contour function is C (v 0 ) . Informally, T (v 0 ) is obtained from T by removing all vertices that are descendants of v 0 and by re-rooting the resulting tree at v 0 . Furthermore, if v 0 = u 1 . . . u n , then we see that v 0 = 1u n . . . u 2 belongs to T (v 0 ) . In fact, v 0 is the vertex of T (v 0 ) corresponding to the root of the initial tree. At last notice that k ∅ ( T (v 0 ) ) = 1.
If T ∈ T and w 0 ∈ T , we set The following lemma is an analogue of Lemma 3.1 in (12) for two-type Galton-Watson trees.
Note that in what follows, two-type trees will always be re-rooted at a vertex of type 0.
Recall the definition of the probability measure Q µ .
Proof : We first notice that In particular, both conditionings of Lemma 3.4 make sense. Let t be a two-type tree such that v 0 ∈ ∂ 0 t and k ∅ (t) = 1.
which implies the desired result.
Before stating a spatial version of Lemma 3.4, we establish a symmetry property of the collection of measures ν. To this end, we let ν k 1 be the image measure of ν k 1 under the mapping (x 1 , . . . , x k ) ∈ R k → (x k , . . . , x 1 ) and we set ν = ((ν k 0 , ν k 1 )) k≥1 . Lemma 3.5. For every k ≥ 1 and every j ∈ {1, . . . , k}, the measures ν k 1 and ν k 1 are invariant under the mapping φ j : R k → R k defined by Proof : Recall the definition of the sets A k and B k . Let ρ k be the uniform distribution on B k . Then ν k 1 is the image measure of ρ k under the mapping ϕ k : B k → A k defined by For every (x 1 , . . . , x k+1 ) ∈ R k+1 we set p j (x 1 , . . . , x k+1 ) = (x j+1 , . . . , x k+1 , x 1 , . . . , x j ). It is immediate that ρ k is invariant under the mapping p j . Furthermore φ j • ϕ k (x) = ϕ k • p j (x) for every x ∈ B k , which implies that ν k 1 is invariant under φ j . At last for every (x 1 , . . . , x k ) ∈ R k we set S(x 1 , . . . , x k ) = (x k , . . . , x 1 ). Then φ j •S = S •φ k−j+1 , which implies that ν k 1 is invariant under φ j .
If (T , U ) ∈ Ω and v 0 ∈ T 0 , the re-rooted spatial tree ( where v is the vertex of the initial tree T corresponding to v, and for every vertex v ∈ T (v 0 ),1 , we set If (T , U ) ∈ Ω and w 0 ∈ T , we also consider the spatial tree (T (w 0 ) , U (w 0 ) ) where U (w 0 ) is the restriction of U to the tree T (w 0 ) .
Recall the definition of the probability measure Q µ,ν .
Lemma 3.6. Let v 0 ∈ U * be of the form v 0 = 1u 2 . . . u 2p for some p ∈ N. Assume that Q µ (v 0 ∈ T ) > 0. Then the law of the re-rooted spatial tree ( T (v 0 ) , U (v 0 ) ) under Q µ,e ν (· | v 0 ∈ T ) coincides with the law of the spatial tree ( Lemma 3.6 is a consequence of Lemma 3.4 and Lemma 3.5. We leave details to the reader. If (T , U ) ∈ Ω, we denote by ∆ 0 = ∆ 0 (T , U ) the set of all vertices of type 0 with minimal spatial position : We also denote by v m the first element of ∆ 0 in the lexicographical order. The following two lemmas can be proved from Lemma 3.6 in the same way as Lemma 3.3 and Lemma 3.4 in (12).
Lemma 3.7. For any nonnegative measurable functional F on Ω, Lemma 3.8. For any nonnegative measurable functional F on Ω,

Estimates for the probability of staying on the positive side
In this section we will derive upper and lower bounds for the probability P n µ,ν,x (U > 0) as n → ∞. We first state a lemma which is a direct consequence of Lemma 17 in (16). Lemma 3.9. There exist constants c 0 > 0 and c 1 > 0 such that We now establish a preliminary estimate concerning the number of leaves of type 0 in a tree with n vertices of type 1. Recall that m 1 denotes the mean of µ 1 .
Thanks to the strong Markov property, the random variables X ′ j are independent and distributed according to µ 1 . A standard moderate deviations inequality ensures the existence of a positive constant β 1 > 0 such that for every n sufficiently large, In the same way as previously, we define another sequence of stopping times (θ ′ j ) j≥0 by θ ′ 0 = 0 and θ ′ j+1 = inf{n > θ ′ j : h ′ n is even} for every j ≥ 0 and we set for every j ≥ 0, Using the sequences (θ ′ j ) j≥0 and (Y ′ j ) j≥0 , an argument similar to the proof of (3) shows that there exists a positive constant β 2 > 0 such that for every n sufficiently large, From (3), we get for n sufficiently large, However, for n sufficiently large, At last, we use (4) to obtain for n sufficiently large, where C is a positive constant. The desired result follows by combining this last estimate with (3).
We will now state a lemma which plays a crucial role in the proof of the main result of this section. To this end, recall the definition of v m and the definition of the measure Q n µ,ν . Lemma 3.11. There exists a constant c > 0 such that for every n sufficiently large, Proof : The proof of this lemma is similar to the proof of Lemma 4.3 in (12). Nevertheless, we give a few details to explain how this proof can be adapted to our context.
We can now state the main result of this section.
Proposition 3.12. Let K > 0. There exist constants γ 1 > 0, γ 2 > 0, γ 1 > 0 and γ 2 > 0 such that for every n sufficiently large and for every x ∈ [0, K], Proof : The proof of Proposition 3.12 is similar to the proof of Proposition 4.2 in (12). The major difference comes from the fact that we cannot easily get an upper bound for #(∂ 0 T ) on the event {#T 1 = n}. In what follows, we will explain how to circumvent this difficulty.
We first use Lemma 3.
Using Lemma 3.9 it follows that lim sup which ensures the existence of γ 2 .

Asymptotic properties of conditioned trees
We first introduce a specific notation for rescaled contour and spatial contour processes. For every n ≥ 1 and every t ∈ [0, 1], we set In this section, we will get some information about asymptotic properties of the pair (C (n) , V (n) ) under P n µ,ν,x . We will consider the conditioned measure Q n µ,ν = Q n µ,ν (· | U > 0).
Berfore stating the main result of this section, we will establish three lemmas. The first one is the analogue of Lemma 6.2 in (12) for two-type spatial trees and can be proved in a very similar way.
Lemma 3.13. There exists a constant c > 0 such that, for every measurable function F on Ω with 0 ≤ F ≤ 1, where the constant β 0 is defined in Lemma 3.10 and the estimate O n 5/2 e −n β 0 for the remainder holds uniformly in F .
Recall the notationv for the "father" of the vertex v ∈ T \ {∅}.
Recall the definition of the re-rooted tree ( T (v 0 ) , U (v 0 ) ). Its contour and spatial contour functions At last, recall that (e 0 , r 0 ) denotes the conditioned Brownian snake with initial point 0.
converges to the law of (e 0 , r 0 ). The convergence holds in the sense of weak convergence of probability measures on the space C([0, 1], R) 2 .
Proof : From Corollary 3.2 and the Skorokhod representation theorem, we can construct on a suitable probability space a sequence a sequence (T n , U n ) and a Brownian snake ( , Ö 0 ), such that each pair (T n , U n ) is distributed according to Q n µ,ν , and such that if we write (C n , V n ) for the contour and spatial contour functions of (T n , U n ) and ζ n = #T n − 1, we have  uniformly on [0, 1], a.s.
Then if (T , U ) ∈ Ω and v 0 ∈ T , we introduce a new spatial tree ( T (v 0 ) , U (v 0 ) ) by setting for where w is the vertex corresponding to w in the initial tree (in contrast with the definition of U when v is of type 1). We denote by V (v 0 ) the spatial contour function of ( T (v 0 ) , U (v 0 ) ), and we set is either a vertex of type 0 or a vertex of type 1 which does not belong to the ancestral line of v 0 , then whereas if w ∈ T (v 0 ),1 belongs to the ancestral line of v 0 , then Then we have sup Write v n m for the first vertex realizing the minimal spatial position in T n . In the same way as in the derivation of (18) in the proof of Proposition 6.1 in (12), it follows from (16) uniformly on [0, 1], a.s., where the conditioned pair ( 0 , Ö 0 ) is constructed from the unconditioned one ( , Ö 0 ) as explained in section 2.3. Let ε > 0. We deduce from (17) that where we have written Q for the probability measure under which the sequence (T n , U n ) n≥1 and the Brownian snake ( , Ö 0 ) are defined. From (14) we get Q sup and the desired result follows.
The following proposition can be proved using Lemma 3.13 and Lemma 3.15 in the same way as Proposition 6.1 in (12).

Proof of Theorem 3.3
The proof below is similar to Section 7 in (12). We provide details since dealing with two-type trees creates nontrivial additional difficulties.
On a suitable probability space (Ω, P) we can define a collection of processes (e, r z ) z≥0 such that (e, r z ) is a Brownian snake with initial point z for every z ≥ 0. Recall from section 2.3 the definition of (e z , r z ) and the construction of the conditioned Brownian snake (e 0 , r 0 ).
Recall that C([0, 1], R) 2 is equipped with the norm (f, g) = f u ∨ g u . For every f ∈ C([0, 1], R) and r > 0, we set Let x ≥ 0 be fixed throughout this section and let F be a bounded Lipschitz function. We have to prove that E n µ,ν, We may and will assume that 0 ≤ F ≤ 1 and that the Lipschitz constant of F is less than 1.
The first lemma we have to prove gives a spatial Markov property for our spatial trees. We use the notation of section 5 in (12). Let recall briefly this notation. Let a > 0. If (T , U ) is a mobile and v ∈ T , we say that v is an exit vertex from (−∞, a) if U v ≥ a and U v ′ < a for every ancestor v ′ of v distinct from v. Notice that, since U v = Uv for every v ∈ T 1 , an exit vertex is necessarily of type 0. We denote by v 1 , . . . , v M the exit vertices from (−∞, a) listed in lexicographical order.
For v ∈ T , recall that At last, we denote by T a the subtree of T consisting of those vertices which are not strict descendants of v 1 , . . . , v M . Note in particular that v 1 , . . . , v M ∈ T a . We also write U a for the restriction of U to T a . The tree (T a , U a ) corresponds to the tree (T , U ) which has been truncated at the first exit time from (−∞, a). The following lemma is an easy application of classical properties of Galton-Watson trees. We leave details of the proof to the reader.
are independent and distributed respectively according to P The next lemma is analogous to Lemma 7.1 in (12) and can be proved in the same way using Theorem 3.1. where B = (9(Z q − 1)/(4ρ q )) 1/4 .
We can now follow the lines of section 7 in (12). Let b > 0. We will prove that for n sufficiently large, We can choose ε ∈ (0, b ∧ 1/10) in such a way that for every z ∈ (0, 2ε). By taking ε smaller if necessary, we may also assume that, For α, δ > 0, we denote by Γ n = Γ n (α, δ) the event From Proposition 3.16, we may fix α, δ ∈ (0, ε) such that, for all n sufficiently large, Recall that m 1 denotes the mean of µ 1 . We also require that δ satisfies the following bound Recall the notation ζ = #T − 1 and B = (9(Z q − 1)/(4ρ q )) 1/4 . On the event Γ n , we have for This incites us to apply Lemma 3.17 with a n = αn 1/4 , where α = αB −1 . Once again, we use the notation of (12). We write v n 1 , . . . , v n Mn for the exit vertices from (−∞, αn 1/4 ) of the spatial tree (T , U ), listed in lexicographical order. Consider the spatial trees The contour functions of these spatial trees can be obtained in the following way. Set k n 1 = inf k ≥ 0 : V (k) ≥ αn 1/4 , l n 1 = inf {k ≥ k n 1 : C(k + 1) < C(k n 1 )} , and by induction on i, Using (22), we see that on the event Γ n , all integer points of [2ζδ, 2ζ(1 − δ)] must be contained in a single interval [k n i , l n i ], so that for this particular interval we have if n is sufficiently large, P n µ,ν,x a.s. Hence if then, for all n sufficiently large, Γ n ⊂ E n so that As in (12), on the event E n , we denote by i n the unique integer i ∈ {1, . . . , M n } such that l n i − k n i > 2ζ(1 − 3δ). We also define ζ n = #T [v n in ] − 1 and Y n = U v n in . Note that ζ n = (l n i − k n i )/2 so that ζ n > ζ(1 − 3δ). Furthermore, we set We need to prove a lemma providing an estimate of the probability for p n to be close to n. Note that p n ≤ n = #T 1 , P n µ,ν,x a.s. Proof : In the same way as in the proof of the bound (3) we can verify that there exists a constant β 3 > 0 such that for all n sufficiently large, So Lemma 3.9 and Proposition 3.12 imply that for all n sufficiently large, Now, on the event Γ n , we have since we saw that Γ n ⊂ E n and that ζ n > (1 − 3δ)ζ on E n . If n is sufficiently large, we have 3δ(m 1 + 1)n + 3δn 3/4 ≤ 4δ(m 1 + 1)n so we obtain that Γ n ∩ ζ ≤ (m 1 + 1)n + n 3/4 ⊂ (Γ n ∩ {p n ≥ (1 − 4δ(m 1 + 1))n}) , for all n sufficiently large. The desired result then follows from (20) and (24).
Let us now define on the event E n , for every t ∈ [0, 1], Note that C (n) and V (n) are rescaled versions of the contour and the spatial contour functions of (T . On the event E c n , we take C (n) (t) = V (n) (t) = 0 for every t ∈ [0, 1]. Straightforward calculations show that on the event Γ n , for every t ∈ [0, 1], We then get that on the event Γ n , for every t ∈ [0, 1], Likewise, we set E n = E n ∩ {p n ≥ (1 − 4δ(m 1 + 1))n}.
Lemma 3.17 implies that, under the probability measure P n µ,ν,x (· | E n ) and conditionally on the σ-field G n defined by Note that E n ∈ G n , and that Y n and p n are G n -measurable. Thus we have From Lemma 3.18, we get for every p sufficiently large, which implies using (18), since 3αB/2 ≤ 2αB = 2α < 2ε, that for every p sufficiently large, Furthermore Lemma 3.14 implies that P n µ,ν,x So we get for every n sufficiently large, Thus we use (27), (28), (29) and the fact that p n ≥ 1 − 4δ(m 1 + 1)n on E n , to obtain that for every n sufficiently large, From Lemma 3.19, we have P n µ,ν,x ( E n ) ≥ 1 − 2b. Furthermore, 0 ≤ F ≤ 1 so that (30) gives On the other hand, since Γ n ⊂ E n and F is a Lipschitz function whose Lipschitz constant is less than 1, we have using (25) and (26), for n sufficiently large, By the same arguments we used to derive (31), we can bound the right-hand side of (32), for n sufficiently large, by b + 2ε + E 4δ(m 1 + 1) sup r 0 (s) ∧ 1 + ω r 0 (6δ) ∧ 1 .
From (19) together with (21), the latter quantity is bounded above by 7b. Since we get for all n sufficiently large, which implies together with (31) that for all n sufficiently large, This completes the proof of Theorem 3.3
Note that K T (k) is the number of vertices of type 0 in the search-depth sequence up to time k. As previously, we extend K T to the real interval [0, 2ζ] by setting K T (t) = K T (⌊t⌋) for every t ∈ [0, 2ζ], and we set for every t ∈ [0, 1] Consequently, the law under P Proof : For T ∈ T, we let v 0 (0) = ∅ ≺ v 0 (1) ≺ . . . ≺ v 0 (#T 0 − 1) be the list of vertices of T of type 0 in lexicographical order. We define as in (16) and we set G T (#T 0 ) = ζ. Note that v 0 (k) does not belong to the set {u ∈ T : u ≺ v 0 (k)}.
Recall that m 0 denotes the mean of the offspring distribution µ 0 . From the second assertion of Lemma 18 in (16), there exists a constant ε > 0 such that for n sufficiently large, Then Lemma 3.9 and Proposition 3.12 imply that there exists a constant ε ′ > 0 such that for n sufficiently large, From our definitions, we have for every 0 ≤ k ≤ #T 0 − 1 and 0 ≤ n ≤ ζ, It then follows from (35) that, for every η > 0 P n µ,ν,1 n −1 sup Also from the bound (3) of Lemma 3.10 we get for every η > 0, P n µ,ν,1 The first assertion of Lemma 3.20 follows from the last two convergences.
We are now able to complete the proof of Theorem 2.5. The proof of (i) is similar to the proof of the first part of Theorem 8.2 in (12), and is therefore omitted.
Let us turn to (ii). By Corollary 2.3 and properties of the Bouttier-Di Francesco-Guitter bijection, the law of λ It is more convenient for our purposes to replace I n by a new probability measure I ′ n defined by Let g be a bounded continuous function. Clearly, we have for every η > 0, P n µ,ν,1 Furthermore, we have from our definitions where the first term in the right-hand side corresponds to v = ∅ in the definition of I ′ n . Then from Theorem 3.3, (34) and the Skorokhod representation theorem, we can construct on a suitable probability space, a sequence (T n , U n ) n≥1 and a conditioned Brownian snake ( 0 , Ö 0 ), such that each pair (T n , U n ) is distributed according to P n µ,ν,1 , and such that if we write (C n , V n ) for the contour functions of (T n , U n ), ζ n = #T n − 1 and K n = K Tn , we have, uniformly in t ∈ [0, 1], a.s. Now g is Lipschitz, which implies that a.s., 1 0 g n −1/4 V n (2ζ n t) dK n (t) − 1 0 g 4ρ q 9(Z q − 1) Furthermore, the sequence of measures dK n converges weakly to the uniform measure dt on [0, 1] a.s., so that a.s.
Finally, the proof of (iii) from (ii) is similar to the proof of the third part of Theorem 8.2 in (12). This completes the proof of Theorem 2.5.

Separating vertices in a 2κ-angulation
In this section, we use the estimates of Proposition 3.12 to derive a result concerning separating vertices in rooted 2κ-angulations. Recall that in a 2κ-angulation, all faces have a degree equal to 2κ.
Let M be a planar map and let σ 0 ∈ V M . Let σ be a vertex of M different from σ 0 . We denote by S σ 0 ,σ Let M r,p be the set of all triples (M, e, τ ) where (M, e ) ∈ M r and τ is a distinguished vertex of the map M . We denote by s the canonical surjection from the set M r,p onto the set M r,p which is obtained by "forgetting" the orientation of e. We observe that for every (M, e, τ ) ∈ M r,p # s −1 ((M, e, τ )) = 2.
Denote by U n κ the uniform measure on the set of all triples (M, e, τ ) ∈ M r,p such that M is a 2κ-angulation with n faces. Then the image measure of the measure U n κ under the mapping s is the measure U n κ . Thus we obtain from Theorem 4.2 that On the other hand let p be the canonical projection from the set M r,p onto the set M r . If M is a 2κ-angulation with n faces, we have thanks to Euler formula Thus the image measure of the measure U n κ under the mapping p is the measure U n κ . This remark together with (43) implies Theorem 4.1.
The remainder of this section is devoted to the proof of Theorem 4.2. We first need to state a lemma. Recall the definition of the spatial tree (T [v] , U [v] ) for (T , U ) ∈ Ω and v ∈ T . Proof : Let (T , U ) ∈ T mob 1 . Write w 0 , w 1 , . . . , w ζ for the search-depth sequence of T 0 (see Section 2.4). Recall from Section 2.4 the definition of (U + v , v ∈ T ) and the construction of the planar map Ψ r,p ((T , U )). For every i ∈ {0, 1, . . . , ζ}, we set s i = ∂ if U + w i = 1, whereas if U + w i ≥ 2, we denote by s i the first vertex in the sequence w i+1 , . . . , w ζ−1 , w 0 , w 1 , . . . , w i−1 whose label is U + w i − 1. Suppose that there exists v ∈ T 0 such that T The vertices w k , w k+1 , . . . , w l are exactly the descendants of v in T 0 . The condition U [v] > 0 ensures that for every i ∈ {k + 1, . . . , l − 1}, we have This implies that s i is a descendant of v for every i ∈ {k + 1, . . . , l − 1}. Furthermore s k = s l and s i is not a strict descendant of v if i ∈ {0, 1, . . . , ζ} \ {k, k + 1 . . . , l}. From the construction of edges in the map Ψ r,p ((T , U )) we see that any path from ∂ to a vertex that is a descendant of v must go through v. It follows that v is a separating vertex of the map M = Ψ r,p ((T , U )) and that the set T [v],0 is in one-to-one correspondence with the set S v,τ M .
Thanks to Corollary 2.4 and Lemma 4.3, it suffices to prove the following proposition in order to get Theorem 4.2. Recall the definition of µ κ = (µ κ 0 , µ κ 1 ).
Proof : For n ≥ 1 and ε > 0, we denote by Λ n,ε the event Let ε > 0 and α > 0. We will prove that for all n sufficiently large, We first state a lemma. For T ∈ T and k ≥ 0, we set Z 0 (k) = #{v ∈ T : |v| = 2k}.
The first step is to prove that for all n sufficiently large, P n µ κ (E n,ε ) ≥ 1 − 2α.
From (49), we obtain that there exists a constant c ′ > 0 such that which together with (47) implies that P µ κ Γ n ∩ C c n,ε ≤ M K n √ ne −n ε .
Note that the total progeny of Z 0 is #T 0 and recall that under the probability measure P µ κ , we have #T 0 = 1 + (κ − 1)#T 1 a.s.
Denote by l(t) the total local time of e at level t that is, However we can find β > 0, γ > 0 and M > 0 such that