Access the full text.

Sign up today, get DeepDyve free for 14 days.

Mathematics
, Volume 2020 (1502) – Feb 24, 2015

/lp/arxiv-cornell-university/convergence-of-the-two-dimensional-random-walk-loop-soup-clusters-to-RcBbb1a7S6

- ISSN
- 1435-9855
- eISSN
- ARCH-3343
- DOI
- 10.4171/JEMS/859
- Publisher site
- See Article on Publisher Site

CONVERGENCE OF THE TWO-DIMENSIONAL RANDOM WALK LOOP-SOUP CLUSTERS TO CLE TITUS LUPU CNRS and LPSM, UMR 8001, Sorbonne Universit e, 4 place Jussieu, 75252 Paris cedex 05, France Abstract. We consider the random walk loop-soup of sub-critical intensity parameter on the discrete half-plane H := Z N. We look at the clusters of discrete loops and show that the scaling limit of the outer boundaries of outermost clusters is a CLE Conformal loop ensemble. 1. Introduction One can naturally associate to a wide class of Markov processes an in nite measure on time-parametrised loops. Roughly speaking, given a locally compact second-countable space S, a Markov process (X ) t 0t< on S, de ned up to a killing time 2 (0; +1], with transition densities p (x; y) with respect some - nite measure m(dy), incorporating the killing if there is one, and with bridge probability measures P (), x;y where the bridges are conditioned on > t, the loop measure associated to X is Z Z dt (1.1) () = P ()p (x; x) m(dx): x;x x2S t>0 See [8] for the precise setting and de nition. A Poisson ensemble of Markov loops or loop-soup of intensity parameter > 0 is a Poisson point process of loops of intensity . It is a random countable collection of loops. These loop-soups satisfy some universal properties, one of which is the relation to the Gaussian free eld at intensity parameter = 1=2 [3, 9]. We will deal with the clusters of loops. Two loops and in a loop-soup belong to the same cluster if there is a chain of loops ; : : : ; such that = , 0 j 0 = and and visit a common point in S. j i i1 We will consider loop-soups in three dierent settings. In the rst one, on the continuum half-plane H = f=(z) > 0g C, we will consider the loop-soups associated to the Brownian motion on H killed at the rst hitting time of the boundary R and denote them L . These two-dimensional Brownian loop- soups were introduced by Lawler and Werner in [5] and used by Sheeld and Werner in [14] to give a construction of Conformal loop ensembles (CLE). In (1.1) we use the same normalisation of the loop measure as in [5], [14], [3] or [4]. However, contrary to what is claimed in [14], the intensity parameter is not equal to the central charge c. The central charge is a notion that comes from Conformal Field Theory and representations of Virasoro algebra. Actually, = : The 1=2 factor was pointed out by Werner in a private communication. It also appears in Lawler's work [6]. The confusion originates from the article [5]. There the authors consider a Brownian loop soup in the half-plane and a continuous path cutting the half-plane, parametrised by the half-plane capacity. For such a path the half-plane capacity at time t equals 2t . It discovers progressively new Brownian loops and the authors map these loops conformally to the origin. In Theorem 1 they identify the processes of these conformally mapped Brownian loops to be a Poisson point process with intensity proportional to the Brownian bubble measure. In the identi cation of the intensity there is a factor 2 missing. Actually, in the article [5], Theorem 1 is inconsistent with Proposition 11. In the second setting, on the discrete rescaled half-plane 1 1 H := Z N ; n n E-mail address: titus.lupu@upmc.fr. 2010 Mathematics Subject Classi cation. Primary 60G15; 60J67; 60K35; 82B20; Secondary 82B27. Key words and phrases. Conformal loop ensemble; Gaussian free eld; loop-soup; metric graph; Poisson ensemble of Markov loops. arXiv:1502.06827v4 [math.PR] 2 Dec 2019 we will consider the loop-soups associated to the nearest neighbours Markov jump process with uniform transition rates and killed at the rst hitting time of the boundary Zf0g. We will denote these loop- soups L . The loop-soups associated to Markov jump processes on more general electrical networks were studied by Le Jan in [3]. If one forgets the parametrisation by continuous time and the "loops" that visit only one vertex, these are exactly the random walk loop-soups studied by Lawler and Trujillo-Ferreras in [4]. See also [7], Section 9. In the third setting, we will use the metric (or cable) graphs H associated to H : each "discrete" edge n n i j i+1 j i j i j+1 1 H f( ; ); ( ; )g or f( ; ); ( ; )g is replaced by a continuous line of length . Let (B ) be 0t< t n n n n n n n n n n the Brownian motion on H (cable process) killed at reaching the boundary, that is to say the vertices 1 i i+1 H Zf0g and all the lines joining ( ; 0) to ( ; 0). One can nd a construction of (B ) in [9]. t 0t< n n n n Inside each line segment, B evolves like a one-dimensional Brownian motion. After reaching a vertex, the process makes Brownian excursions in each of the four possible directions before hitting the next vertex. Each direction has an equal rate. (B ) converges in law to the Brownian motion on the 2t 0t< =2 H H n n half-plane H killed at reaching R. We will denote by L the loop-soups associated to (B ) . The 0t< t n loop-soups on metric graphs were rst considered in [9]. We will use metric graphs because at intensity parameter = 1=2 the probability that two points belong to the same cluster of loops can be explicitly expressed using a metric graph Gaussian free eld. Indeed, the clusters of loops are then exactly the sign clusters of the Gaussian free eld [9]. H H n n The discrete loops L can be deterministically recovered from the metric graph loops L . The rst are the trace on the vertices of the latter. In particular each cluster of L is contained in a cluster of e e H H n n L , but the clusters of L may be strictly larger [9]. c = 1 is the critical central charge for the Brownian loop percolation on H (or any other simply connected proper subset of C). This means that the critical intensity parameter is = 1=2. For > 1=2, L has only one cluster everywhere dense in H. If 2 (0; 1=2], there are in nitely many clusters and each is bounded [14]. = 1=2 is also the critical intensity parameter for the existence of an unbounded cluster of loops on discrete or metric graph half-plane H respectively H [10, 9]. In all n n three settings, for 2 (0; 1=2], we will consider the collection of outer boundaries of outermost clusters (not surrounded by any other cluster) and denote it F (L ), where S is H, H or H . Next we give ext n n S S the formal de nition of F (L ). We consider the set of all points in H visited by a loop in L and ext take its complement in H. This complement has only one unbounded connected component. We take the boundary in H of this connected component (by de nition it does not intersect R). The elements of S S F (L ) are the connected components of this boundary. We will call the elements of F (L ) contours. ext ext The contours are pairwise disjoint and non nested. See Figure 1 for a representation of F (L ). ext The contours in F (L ), 2 (0; 1=2], are non self-intersecting loops, and are equal in law to a ext Conformal loop ensemble CLE , 2 (8=3; 4] [14]. The relation between and is given by (3 8)(6 ) (1.2) 2 = c = : We will denote by () the value of corresponding to a particular intensity parameter . (d) H H H n n We will show that both F (L ) and F (L ) converge in law to F (L ) = CLE for 2 ext ext ext () (0; 1=2]. Observe that (1=2) = 4 and F (L ) and CLE are both related to the Gaussian free eld. ext 4 1=2 F (L ) is the collection of outer boundaries of outermost sign clusters of a GFF on the metric graph ext 1=2 H [9] and the CLE loops are in some sense zero level lines of the continuum GFF on H with zero n 4 boundary conditions on R [11, 15, 1, 12, 13]. Next we de ne the notion of convergence we will use. d will be Hausdor distance on the compact subsets of H. We introduce the distance d between nite collections of compact subsets of H: +1 if jKj 6= jK j; d (K;K ) = min 0 max d (K; (K )) otherwise; 2Bij(K;K ) K2K H 0 0 where K and K are nite collections of compact subsets and Bij(K;K ) is the set of all bijections from K to K . Given z 2 H, we will denote by F (L )(z) ext 2 e Figure 1. Illustration of three clusters (thin full lines) of L , two of them being external and one being surrounded. The thick lines represent the elements of F (L ). ext the contour of F (L ) that contains or surrounds z, whenever it exists. It exists a.s. in the case S = H. ext Given z ; : : : ; z 2 H, we will denote 1 j S S F (L )[z ; : : : ; z ] := fF (L )(z )j1 i jg: ext 1 j ext i H H H n n By the convergence in law of F (L ) and F (L ) to F (L ) we mean that for any z ; : : : ; z 2 H, ext ext ext 1 j H H H n n F (L )[z ; : : : ; z ] and F (L )[z ; : : : ; z ] converge in law to F (L )[z ; : : : ; z ] for the distance d . ext 1 j ext 1 j ext 1 j So, the main result in this article is the following. H H n n Theorem 1.1. Let 2 (0; 1=2]. F (L ) and F (L ) converge in law (in the above de ned sense) ext ext as n ! +1 to F (L ), that is to say to a CLE on H. ext () In the article [2] Van de Brug, Camia and Lis consider clusters of rescaled two-dimensional random H ;T H n n walk loops that are not too small. Given T > 0 let L be the subset of L consisting of random walk loops that do at least T jumps. In [2] it is almost shown that for 2 (16=9; 2) and 2 (0; 1=2], H ;n F (L ) converges in law to a CLE process in the sense described previously. The result uses ext () the approximation of "not too small" Brownian loops by "not too small" random walk loops obtained by Lawler and Trujillo-Ferreras in [4]. However the authors in [2] consider the loop-soups only on bounded domains. In the present paper, we will extend their result by removing the cuto on microscopic loops (and also consider the case of unbounded domains). Actually, the "microscopic" loops that are thrown away in [2] create additional connections and may merge large clusters. So the point is to show that this happens with a probability converging to 0 and the contribution of microscopic loops does not change the picture at macroscopic level. Observe that in [2] the authors use the same normalisation of the measure on loops as we do but with the widespread confusion about the factor 2 in the intensity of loop-soups. From above considerations one deduces that the contours obtained in the limit from F (L ) and ext a fortiori from F (L ) are "at least as big as" CLE loops. We thus have a "lower bound". To ext () conclude the convergence we need an "upper bound". We will prove Theorem 1.1 in two steps. First, e e H H n n we will construct an "upper bound" for F (L ) and deduce the convergence to CLE of F (L ) ext 4 ext 1=2 1=2 and F (L ). Then from this we will deduce the desired convergences for 2 (0; 1=2). For this, we ext 1=2 will divide the loop-soup of intensity 1=2 in two independent loop-soups of respective intensities and , with + = 1=2. If the scaling limit of F (L ) happens to contain contours "strictly larger" ext than CLE , then the additional independent contribution of L would give in the scaling limit of () F (L ) contours "strictly larger" than CLE , and this would contradict the rst step. ext 4 1=2 3 Next we explain how the "upper bound" in the critical case = 1=2 will be constructed. We additionally introduce two Poisson point processes of excursions on H and on H. First we consider 1 H H . Let x 2 Z f0g, where Z includes 0. Let (x ! (1; 0]) be the measure on excursions of exc the metric graph Brownian motion B from x to a point in Z f0g. It is de ned as follows: Let e e H H 1 e n n H P (; B 2 Z f0g) be the law of a sample path of B , started at x + i", restricted to the event x+i" H 1 B 2 Z f0g (we do not condition and the total mass is < 1). Then 1 1 e e e H H H n n n (x ! (1; 0]) = lim P ; B 2 Z f0g : exc x+i" "!0 " n 1 H Let q 2 (1; +1) and x 2 (( Z)\ [1; q])f0g. We will similarly denote by (x ! [1; q]) the measure exc on excursions from x to (( Z)\ [1; q])f0g. Let e e H H n n (1.3) ((1; 0]) := (x ! (1; 0]); exc exc x2 Z f0g e e H H n n (1.4) ([1; q]) := (x ! [1; q]): exc exc x2(( Z)\[1;q])f0g e e H 1 H n n ((1; 0]) is a measure on excursions from and to Z f0g. ([1; q]) is a measure on excursions exc exc from and to (( Z)\ [1; q])f0g. The above measures can be disintegrated over the starting and the endpoint. The measure induced over the couple starting and endpoint is X X 1 j 8 P i 1 B hits Z f0g on ; 0 i 0 j 0 ; ( ; ) (( ; );( ; )) n n n n n n n n i j 2 interval 2 interval where "interval" stands for either (1; 0] or [1; q], and denotes the Dirac mass. Let G (;) be the Green's function of the simple random walk (x ) on H = Z N, killed at the rst hitting time of k k0 Zf0g. Let i; j 2 Z. Then 1 j P i 1 B hits Z f0g on ; 0 ( ; ) n n n n = P (x ; : : : ; x 62 Zf0g; x = (j; 1); x = (j; 0)) (i;1) 1 k1 k k+1 k=0 (1.5) = P (x ; : : : ; x 62 Zf0g; x = (j; 1)) (i;1) 1 k1 k k=0 1 1 H H = G ((i; 1); (j; 1)) = G ((0; 1); (j i; 1)): 4 4 i 1 j j 1 Indeed, to go from ( ; ) to ( ; 0) the moving particle needs to reach ( ; ), possibly make excursions n n n n n 1 1 from and to this point without hitting Z f0g, and then with probability transition to ( ; 0). n 4 n Thus, the measure over the starting and endpoint is X X 2 G ((0; 1); (j i; 1)) i 0 j 0 : (( ; );( ; )) n n n n i j 2 interval 2 interval Observe that the above measure is invariant by permuting the starting and the endpoint. Moreover, the conditional probability measures on excursions where the both ends are xed are covariant with time reversal, that is to say the distribution on the unoriented excursion does not change. This means that e e H H n n the whole measures on excursions ((1; 0]) and ([1; q]) are invariant under time reversal. exc exc According to the asymptotic expansion given in [7], Section 8:1:1, 1 1 (1.6) G ((0; 1); (j; 1)) = + O : 2 3 j j 4 So, as n tends to in nity, the measure on the starting and endpoint converges to a measure with density with respect to Lebesgue: dxdy 2 1 : x;y2 interval (y x) The conditional probability measures on excursions of B with xed endpoints converge too. The limits are the probability measures on two-dimensional Brownian excursions from x to y in H, where x; y 2 R, and we will denote them P (). See [18], Section 1.2, for more on these normalised excursion probability x;y measures. e e H H n n Consequently, as n tends to in nity, ((1; 0]) and ([1; q]) have limits which are measures on exc exc Brownian excursions in H, from and to (1; 0]f0g respectively [1; q]f0g, and which disintegrate as follows: Z Z Z Z 0 0 q q dxdy dxdy H H H H ((1; 0]) = 2 P ; ([1; q]) = 2 P : exc x;y exc x;y 2 2 (y x) (y x) 1 1 1 1 In general, given a < b 2 R, we will use the notation Z Z b b dxdy H H ([a; b]) := 2 P : exc x;y (y x) a a See [18], Section 4.3, for more on these in nite mass excursion measures. We will consider on H three independent Poisson point processes: a loop-soup L , 1=2 e e H H n n a Poisson point process of excursions of intensity u ((1; 0]), u > 0, denoted byE ((1; 0]), exc u e e H H n n a Poisson point process of excursions of intensity v ([1; q]), v > 0, denoted by E ([1; q]). exc v We will consider the following event: either an excursion from E ((1; 0]) intersects an excursion from e e e H H H n n n E ([1; q]) or an excursion from E ((1; 0]) and one from E ([1; q]) intersect a common cluster of v u v e e H H n n L . We will denote by p (q) the probability of this event. The second condition of intersecting a 1=2 1=2;u;v common cluster is equivalent to intersecting a common contour in F (L ). ext 1=2 Similarly we will consider on H three independent Poisson point processes: a loop-soup L , 2 (0; 1=2], H H a Poisson point process of excursions of intensity u ((1; 0]), u > 0, denoted by E ((1; 0]), exc u H H a Poisson point process of excursions of intensity v ([1; q]), v > 0, denoted by E ([1; q]). exc v Then we will consider the event when either an excursion from E ((1; 0]) intersects an excursion from H H H H E ([1; q]) or an excursion from E ((1; 0]) and one from E ([1; q]) intersect a common cluster of L . v u v This event is schematically represented in Figure 2. We denote by p (q) its probability. ;u;v Figure 2. Two excursions (full lines) connected by a chain of two loops (doted lines). 5 e In Section 2 we will compute p (q) using the duality with the Gaussian free eld, and compute 1=2;u;v its limit as n tends to +1. In Section 3, for an arbitrary value of v and a particular value u () of u (depending on ) we will establish a dierential equation in q for 1 p (q). Using this we will show ;u;v that H H (1.7) lim p (q) = p (q): 1=2;u (1=2);v 1=2;u (1=2);v 0 n!+1 This convergence will provide the "upper bound" we need. Indeed, if the scaling limit of F (L ) ext 1=2 contains contours "strictly larger" than CLE , then the limit contours would connect E ((1; 0]) u (1=2) and E ([1; q]) with a probability strictly larger than p (q), which in (1.7) would give a strict inequality rather then an equality. In Section 4 we 1=2;u (1=2);v will prove the convergences to CLE out of (1.7) using the above argument. 2. Computations on metric graph Let G = (V; E) be a connected undirected graph. V is countable and each vertex is of nite degree. Each edge fx; yg is endowed with a positive conductance C (x; y) > 0. We also consider a metric graph G associated to G where each edge fx; yg is replaced by a continuous line of length (2.1) r(x; y) = C (x; y) : Let B be the Brownian motion on the metric graph G. Let F be a subset of V . Let be the rst e e G G;F time B hits F . Let be the measure on loops associated to (B ) , the Brownian motion killed 0t< t F G;F at reaching F . It is de ned according to (1.1). See [9] for details. Let L be the Poisson point process G;F of intensity . e e G z G B has a time-space continuous family of local times L (B ). The Green's function of the killed Brownian motion (B ) is de ned to be 0t< t F h i e e G;F 0 z G G (z; z ) = E L (B ) e e G G;F z and is symmetric. Just as B , a loop 2 L has a family of continuous local times L ( ). We will denote by t the total life-time of the loop . The occupation eld (L ) is de ned as z2GnF z z L = L ( ): G;F 2L G;F It is a continuous eld. The clusters of L are delimited by the zero set of the occupation eld. At intensity parameter = 1=2, the occupation eld (L ) is related to the Gaussian free eld z2GnF G;F z e b ( ) with zero mean and covariance function G . Given z 2 GnF such that L > 0, we denote z e z2GnF 1=2 G;F by C (z) the cluster of L that contains z. We introduce a countable family ((C (z))) of 1=2 1=2 e 1=2 z2GnF G;F i.i.d. random variables, independent of L conditional on the clusters, which equal 1 or 1 with equal 1=2 probability. There is an equality in law (see [9]): (d) (2.2) ( ) = (C (z)) 2L : z e 1=2 z2GnF 1=2 z2GnF eq eq eq Let x; y 2 V n F . Let C (x; y); (x); (y) be the quantities de ned by (x;y) (x;y) ! ! eq e e eq eq G;F G;F (x) + C (x; y) C (x; y) G (x; x) G (x; y) (x;y) = : eq eq eq e e G;F G;F C (x; y) (y) + C (x; y) G (x; y) G (y; y) (x;y) eq eq eq eq eq eq eq Then C (x; y) > 0, (x); (y) 0, ( (x) and (y)) 6= (0; 0). C (x; y), (x) and (x;y) (x;y) (x;y) (x;y) (x;y) eq (y) are the conductances of a network electrically equivalent to G, where all vertices in F are at the (x;y) same electrical potential. This equivalent network has three vertices, x, y and a vertex corresponding eq eq eq to the set F . C (x; y) is the conductance between x and y, (x) respectively (y) is the (x;y) (x;y) conductance between x and F respectively y and F . G;F Let N (x; y) the number of loops in L that visit both x and y. 1=2 1=2 6 Lemma 2.1. Let u; v > 0 and x; y 2 V n F . eq x 2C (x;y) uv b b (2.3) P C (x) 6= C (y) L = u;L = v;N (x; y) = 0 = e : 1=2 1=2 1=2 1=2 1=2 Proof. If N (x; y) > 0 then C (x) = C (y). Thus 1=2 1=2 1=2 x y b b (2.4) P C (x) 6= C (y) L = u;L = v;N (x; y) = 0 1=2 1=2 1=2 1=2 1=2 b b P C (x) 6= C (y) L = u;L = v 1=2 1=2 1=2 1=2 = : bx b P N (x; y) = 0 L = u;L = v 1=2 1=2 1=2 The value of the denominator b b P N (x; y) = 0 L = u;L = v 1=2 1=2 1=2 e e e G;F G;F G;F depends only on u; v and on G (x; x); G (y; y); G (x; y) (or equivalently on eq eq eq C (x; y), (x), (y)). This a general property of the loop-soups (see [3], especially chapter 7). (x;y) (x;y) As for the numerator, it can be computed using the duality with the Gaussian free eld (2.2). If C (x) = C (y), then and have same sign. Otherwise, and have same sign with conditional x y x y 1=2 1=2 probability 1=2. Thus h i p p b b P C (x) 6= C (y) L = u;L = v = 1 E sgn( ) sgn( ) j j = 2u;j j = 2v x y x y 1=2 1=2 1=2 1=2 p p eq eq 2C (x;y) uv 2C (x;y) uv e e = 1 p p eq eq 2C (x;y) uv 2C (x;y) uv e + e eq 2C (x;y) uv = : eq cosh(2C (x; y) uv) It follows that the probability (2.3) that we want to compute only depends on u; v and on eq eq eq C (x; y); (x); (y). Thus it is the same if we replace G by the interval (x;y) (x;y) 1 1 1 eq eq 1 eq 1 1 I = (x) ; C (x; y) + (y) ; (x;y) (x;y) 2 2 2 the Brownian motion on G by the Brownian motion on I killed at endpoints, and the points x and y by 1 eq 1 0 and C (x; y) respectively. According to Lemma 3:4 and 3:5 in [9], we get (2.3). By the way we also get that x y eq 1 b b P N (x; y) = 0 L = u;L = v = cosh(2C (x; y) uv) : 1=2 1=2 1=2 eq In [3], chapter 7, there is a combinatorial representation of C (x; y). Given z 2 V , we will denote (z) := C (z; z ); z 2V z z where the sum is over the neighbours of z in the (discrete) graph G. Then X X Y C (z ; z ) i1 i eq C (x; y) = (x) : (z ) i1 j+1 j1 i=1 (z ;:::;z )2(VnF ) 0 j z =x;z =y;z z 0 j i i1 z 6=x;y for 1ij1 The sum is over all the discrete nearest neighbour paths joining x to y, that avoid F and only visit x and y at endpoints. The above equality can be rewritten as eq G (2.5) C (x; y) = C (x; z)P (B hits y before F or x): z2V zx Next we return to the metric graph half-plane H . Let a > 0. Let G (q) be the metric graph obtained n n;a from H by identifying the following vertices: All the vertices in (( Z)\ [a; 0])f0g are identi ed into a single vertex C (a). All the vertices in (( Z)\ [1; q])f0g are identi ed into a single vertex B (q). 7 See Figure 3. We consider a nite value of a just to have a nite degree for the quotient vertex C (a), but eventually we will consider a ! +1. Figure 3. Illustration of points identi ed into C (a) and B (q). n n i j i+1 j i j i j+1 1 As the length of the line joining ( ; ) to ( ; ) or ( ; ) to ( ; ) is , the corresponding n n n n n n n n n n eq conductance is according to (2.1) equal to . Let C (q) be the equivalent (or eective) conductance n;a between C (a) and B (q) when all the points in ( )Z f0g other than those identi ed to C (a) or n n n B (q) have the same electrical potential. According to (2.5), bnqc n 1 eq H C (q) = P i 1 B hits Z f0g on [a; 0]f0g : n;a ( ; ) n n 2 n i=n eq As a tends to in nity, C (q) increases and converges to n;a bnqc n 1 eq H (2.6) C (q) = P i 1 B hits Z f0g on (1; 0]f0g : n ( ; ) n n 2 n i=n eq Lemma 2.2. For all n 2 N and q > 1, C (q) < +1. Moreover, 1 1 eq lim C (q) = log(q): n!+1 n 8 Proof. Using the computation (1.5) and the asymptotic expansion (1.6), we get that eq C (q) < +1 and that 0 1 bnqc bnqc +1 +1 X X X X 1 1 1 1 eq @ A C (q) = + O 2 3 n 8 (i + j) (i + j) i=n j=0 i=n j=0 0 1 bnqc bnqc X X 1 1 1 @ A = + O 8 i i i=n i=n 1 1 = log(q) + O : 8 n e e H H n n Let ([a; 0]) be the measure on excursions ((1; 0]) restricted to the excursions from and exc exc G (q) n;a to [a; 0]f0g. Let L be the loop-soup associated to the Brownian motion on the metric graph G (q), killed at the rst hitting time of ( )Zf0g outside the points identi ed to C (a) or B (q). Let n;a n n G (q) z n;a (L ) be the occupation eld of L . Let N (C (a);B (q)) be the number of loops in n n n;a;q; e z2G (q) n;a G (q) n;a L joining C (a) to B (q). n n G (q) n;a Lemma 2.3. Let a; ; u; v > 0. We consider L conditioned on C (a) B (q) n n b b L = u;L = v and N (C (a);B (q)) = 0: n n n;a;q; n;a;q; G (q) n;a Then L consists of three independent families of loops: 8 e The loops that visit neither C (a) nor B (q). These are the same as the loops in L . n n The loops that visit C (a). The excursions these loops make outside C (a) form a Poisson point n n n H process of intensity u ([a; 0]). exc The loops that visit B (q). The excursions these loops make outside B (q) form a Poisson point n n process of intensity v ([1; q]). exc Proof. This follows from universal properties of loop-soups. The subset of loops that do not visit a given set F is distributed like the loop-soup of the same Markov process, but with additional killing at hitting F (restriction property). The loops that visit a particular point z can be represented by a Poisson point process of Markovian excursions outside z. See for instance [3], Sections 2.2, 2.3, 7.1, 7.2, 7.3, and e e n n H n H n n [7], Propositions 9.3.1 and 9.4.1. The factor in u ([a; 0]) and v ([1; q]) comes from the exc exc 8 8 8 e e 8 H H n n normalisation factor in the de nition of ([a; 0]) ((1.3)) and ([1; q]) ((1.4)). exc exc Proposition 2.4. Let u; v > 0, q > 1 and n 1. 8 uv eq H 2C (q) n n (2.7) p (q) = 1 e : 1=2;u;v H 2 uv (2.8) lim p (q) = 1 q : 1=2;u;v n!+1 Proof. Let a > 0. Consider three independent Poisson point processes: a loop-soup L , 1=2 a P.p.p of excursions of intensity u ([a; 0]), exc a P.p.p of excursions of intensity v ([1; q]). exc The probability for the two P.p.p. of excursions to be connected either directly or through a cluster of L equals, according to Lemma 2.3, the probability for C (a) and B (q) to be in the same cluster of n n 1=2 G (q) n;a C (a) B (q) n 8 n 8 b b L conditional on L = u, L = v and N (C (a);B (q)) = 0. According to n n 1=2 1=2 n;a;q;1=2 n n;a;q;1=2 n Lemma 2.1 this probability equals 8 uv eq 2C (q) n;a n 1 e : Taking the limit as a tends to in nity we get (2.7). Using Lemma 2.2 we get the limit (2.8). 3. Computations on continuum half-plane On the continuum upper half plane H we consider two independent Poisson point processes: a Brownian loop-soup L , 0 < 1=2, a P.p.p. of Brownian excursions from and to (1; 0]f0g, E ((1; 0]), u > 0. H H We will consider the clusters made out of loops in L and excursions in E ((1; 0]). Among these clusters we only take the clusters that contain at least one excursion and consider the rightmost envelop of these clusters. This envelop is a non self-intersecting curve joining R to in nity. It can be formally de ned as follows. Take the clusters that contain at least one excursion. The curve minus its starting point on R is the right-most component of the boundary in H of the closure in H of the set of points visited by the above clusters. All the excursions E ((1; 0]) are located left to the curve and there are only clusters made of loops right to it. According to [16] and [19] this boundary curve is an SLE(; ) starting from 0, where is given by (1.2) and by ( + 2)( + 6 ) u = : We will de ne 6 () (3.1) u () := : 2() We will consider the particular case u = u () (and thus = 0), which is simpler to deal with. SLE(; ) is then a chordal SLE curve starting from 0. For a description of SLE processes see [17]. We will denote by ( ) this curve. = 0. It does not touch R at positive times. See Figure 4. t t0 0 There is only one conformal map g that sends Hn ([0; t]) (half-plane minus the curve up to time t) onto H and that is normalised at in nity z ! 1 as g (z) = z + + o(z ): 9 H Figure 4. Full lines represent Brownian excursions in E ((1; 0]). Dashed lines rep- resent contours in F (L ). The dotted line represents . ext Moreover, one parametrises the curve by half-plane capacity (a = 2t). The Loewner ow (g ) satis es t t t0 the dierential equation @g (z) 2 = p ; @t g (z) W t t where (W ) is a standard Brownian motion on R. t t0 H H Lemma 3.1. Let 2 (0; 1=2]. p (q) equals the probability that an excursion from E ([1; q]) ;u ();v v intersects an independent SLE curve. () H H H Proof. Let be the SLE curve constructed from L and E ((1; 0]), independent from E ([1; q]). () u () If no excursion from E ([1; q]) intersects , then these excursions are all on the right side of and by H H de nition of , can only intersects loops in L that are not connected to E ((1; 0]). u () Conversely, assume that an excursion from E ([1; q]) intersects at a point z . Then =(z ) > 0. 0 0 Since E ([1; q]) and are independent, by the properties of sample Brownian paths, there is " > 0 small enough such that makes a closed loop around the disc with center z and radius ", disconnecting it from in nity. Thus, any connected set that intersects both this disc and the real line, has to intersect . H H By the de nition of , there is either an excursion from E ((1; 0]), or a loop from L connected by u () a nite chain to an excursion from E ((1; 0]), that intersects the "-neighbourhood of z . Denote u () 0 0 this excursion or loop by . In the rst case, the excursion intersects . In the second case, an 0 H element from the chain connecting to E ((1; 0]) intersects . u () The excursions E ([1; q]) satisfy the one-sided conformal restriction property (see [18], Section 8, and [18], Section 4, in particular Section 4.3 ): if K is a compact subset of C that does not intersect [1; q]f0g and such that Hn K is simply connected, if f is a conformal map from Hn K onto H such that f (1) < f (q) 2 R, then the probability that E ([1; q]) does not intersect K equals 0 0 2 f (1)f (q)(q 1) (f (q) f (1)) H H Moreover, conditional on this event, the law of f (E ([1; q])) is E ([f (1); f (q)]), up to a change of v v parametrisation of the excursions. From this conformal restriction property, it immediately follows: Lemma 3.2. Let 2 (0; 4]. Let ( ) be an SLE with the driving Brownian motion ( W ) and t t0 t t0 Loewner ow (g ) . Denote by g the derivative of g with respect the complex variable: t t0 t @g (z) g (z) = : @z Denote by p (q) the probability that an independent family of excursions E ([1; q]) does not intersect . ;v Then the conditional probability of the event that E ([1; q]) does not intersect conditional on ( ) s 0st 10 (or equivalently conditional on (W ) ) and on not intersecting ( ) equals s 0st s 0st g (q) W t t (3.2) p p : ;v g (1) W t t The conditional probability of the event that E ([1; q]) does not intersect conditional on ( ) is s 0st 0 0 2 g (1)g (q)(q 1) g (q) W t t t t (3.3) p p : ;v (g (q) g (1)) g (1) W t t t t In particular, for all t 0, 0 0 2 g (1)g (q)(q 1) g (q) W t t t t (3.4) p (q) = E p p : ;v ;v (g (q) g (1)) g (1) W t t t t Proof. (3.2) is the conditional probability that g (E ([1; q])) does not intersect (g ( )) . To express it we used the fact that g (E ([1; q])) has same law as t t+s s0 t E ([g (1); g (q)]) and that (g ( )) is a chordal SLE starting from W . In (3.3) we multiplied t t t t+s s0 t the conditional probability that E ([1; q]) does not intersect ( ) and the conditional probability s 0st that g (E ([1; q])) does not intersect (g ( )) . t t t+s s0 Next we derive the dierential equation in q satis ed by p (q) on (1; +1), provided p isC -regular. ;v ;v Lemma 3.3. Let 2 (0; 4], v > 0 and q > 1. Let f be a bounded, C function on (1; +1). Then 0 0 2 g (1)g (q)(q 1) g (q) W t t t t (g (q) g (1)) g (1) W t t t t is a martingale if and only if f satis es the dierential equation 1 4 4 4v 00 0 (3.5) f + 2 q f f = 0: (q 1)q q Proof. Let 0 0 2 g (1)g (q)(q 1) g (q) W t t t t (3.6) R := ; q := : t t (g (q) g (1)) g (1) W t t t t R has bounded variation (in t). Let M := R f (q ): t t We apply It^ o's formula to (M ) . t t0 dR 1 v 0 00 dM = R vf (q ) + f (q )dq + f (q )dhqi : t t t t t t R 2 Denote H := f=(z) 0g. For z 2 Hn ([0; t]), 0 0 @g (z) @ @g (z) @ 2 2g (z) t t p p = = = : @t @z @t @z g (z) W (g (z) W ) t t t t Thus 0 0 2 0 0 2 2g (1)g (q)(q 1) 2g (1)g (q)(q 1) t t t t dR = p + p 2 2 2 2 (g (1) W ) (g (q) g (1)) (g (q) W ) (g (q) g (1)) t t t t t t t t 0 0 2 0 0 2 4g (1)g (q)(q 1) 4g (1)g (q)(q 1) t t t t + p + p dt 3 3 (g (1) W )(g (q) g (1)) (g (q) W )(g (q) g (1)) t t t t t t t t 1 1 = 2R p + p 2 2 (g (1) W ) (g (q) W ) t t t t p p dt (g (1) W )(g (q) W ) t t t t 1 1 = 2R p p dt g (1) W g (q) W t t t t (q 1) = 2R p dt: (g (q) W ) t t 11 Further 1 g (q) W t t dq = p + p dW t t g (1) W (g (1) W ) t t t t 2 g (q) W t t p p p + 2 (g (q) W )(g (1) W ) (g (1) W ) t t t t t t g (q) g (1) t t + p dt (g (1) W ) t t (q 1)q (q 1)q t t t t p p = dW + (( 2)q 2)dt: t t g (q) W (g (q) W ) t t t t 2 2 (q 1) q dhqi = dt: (g (q) W ) t t Finally, (q 1)q R (q 1) t t t v 0 p p dM =R f (q ) dW + t t t g (q) W (g (q) W ) t t t t 2 00 0 (q 1)q f (q ) + q (( 2)q 2)f (q ) 2v(q 1)f (q ) dt: t t t t t t t It follows that (M ) is a local martingale (hence a true one, f being bounded) if and only if t t0 2 00 0 (q 1)q f (q ) + q (( 2)q 2)f (q ) 2v(q 1)f (q ) 0; t t t t t t t which gives the equation (3.5). (3.5) is the dierential equation for p . However, we do not know a priori that p is C -regular. ;v ;v The idea is to show that both p and a solution of (3.5) with right boundary conditions are xed points ;v of a contracting operator, and thus coincide. We will do this for the case = 4 which interests us. Proposition 3.4. Let q > 1, v > 0. H H v lim p (q) = p (q) = 1 q : 1=2;u (1=2);v 1=2;u (1=2);v 0 n!+1 Proof. By de nition p (q) = 1 p (q): 4;v 1=2;u (1=2);v According to Proposition 2.4, H 2 u (1=2)v v n 0 lim p (q) = 1 q = 1 q : 1=2;u (1=2);v n!+1 Let f (q) := q . With = 4, the ODE (3.5) becomes 1 v 00 0 f + f f = 0 q q and it is satis ed by f . According to Lemma 3.3, (R f (q )) is a martingale (we use the notations v v t t0 (3.6) and = 4) for any initial value of q . In particular for any t > 0 f (q ) = E[R f (q )]: v 0 v t The same is true if we replace f by p ((3.4)). Thus, v 4;v (3.7) f (q ) p (q ) = E[R (f (q ) p (q ))] v 0 4;v 0 v t 4;v t for any starting value of q 2 (1; +1) and t > 0. p is non-increasing on (1; +1) with boundary limits 4;v p (1) = 1; p (+1) = 0: 4;v 4;v Moreover p is continuous. Indeed, let q 2 (1; +1). A.s. there is no excursion in E ([1; q]) with 4;v endpoint (q; 0). This means that p is left-continuous at q. Moreover, a.s. there is " > 0 such that there 4;v is no excursion in E ([1; q + ") with an endpoint in [q; q + ")f0g that intersects an independent SLE curve. This implies that p is right-continuous at q. From the continuity of p follows that there is 4;v 4;v q ^2 (1; +1) such that jf (q ^) p (q ^)j = max jf (q) p (q)j: v 4;v v 4;v q2(1;+1) 12 Let t > 0 and let q ^ be the initial value q of (q ) . From (3.7) we get that 0 s s0 jf (q ^) p (q ^)j E[R ]jf (q ^) p (q ^)j: v 4;v v 4;v But a.s. R < 1 and E[R ] < 1. This implies that jf (q ^) p (q ^)j = max jf (q) p (q)j = 0 v 4;v v 4;v q2(1;+1) and that p (q) q : 4;v 4. Convergence to CLE In this section we prove the convergence results. H \Q ;T H n l n Let Q := (l; l) (0; l). Let L be the loops in L that are contained in Q and do at least l l Q H T jumps. Let L be the Brownian loops in L that are contained in Q . From [2] follows that for H \Q ;n Q n l l 2 (0; 1=2], l > 0 and 2 (16=9; 2), F (L ) converges in law to F (L ). ext ext H ;n H Lemma 4.1. Let 2 (0; 1=2] and 2 (16=9; 2). F (L ) converges in law to F (L ). ext ext H ;n H Proof. Let z ; : : : ; z 2 H. To deduce that F (L )[z ; : : : ; z ] converges in law to F (L )[z ; : : : ; z ] 1 j ext 1 j ext 1 j from the result of [2] we need only to show that H ;n lim lim inf P(Contours of F (L )[z ; : : : ; z ] contained in Q ) = 1: ext 1 j l l!+1 n!+1 Let " 2 (0; 1=2). There is l > 0 such that P Contours of F (L )[z ; : : : ; z ] contained in Q 1 ": ext 1 j l Denote @ Q := (flg (0; l])[ (flg (0; l])[ ([l; l]flg): H l There is l > l such that 1 0 P(9 2 L ; \ Q 6= ;; \ @ Q 6= ;) ": l H l 0 1 Then H \Q ;n lim P(Contours of F (L )[z ; : : : ; z ] contained in Q ) ext 1 j l n!+1 = P(Contours of F (L )[z ; : : : ; z ] contained in Q ) ext 1 j l P Contours of F (L )[z ; : : : ; z ] contained in Q 1 ": ext 1 j l According to the approximation of [4], H ;n lim P(9 2 L ; \ Q 6= ;; \ @ Q 6= ;) l H l 0 1 n!+1 = P(9 2 L ; \ Q 6= ;; \ @ Q 6= ;) ": l H l 0 1 But H ;n P(Contours of F (L )[z ; : : : ; z ] contained in Q ) ext 1 j l H \Q ;n n l P(Contours of F (L )[z ; : : : ; z ] contained in Q ) ext 1 j l H ;n P(9 2 L ; \ Q 6= ;; \ @ Q 6= ;): l H l 0 1 Thus, H ;n lim inf P(Contours of F (L )[z ; : : : ; z ] contained in Q ) 1 2": ext 1 j l n!+1 From now on 2 (16=9; 2) will be xed. will belong to (0; 1=2]. For z 2 H, we de ne H ;n H n n (z ) := maxfd(z;F (L )(z ))jz 2 F (L )(z )g: ;n 0 ext 0 ext 0 e e H H n n By z 2 F (L )(z ) we mean that z is a point on the contour F (L )(z ). The random variable ext 0 ext 0 H ;n (z ) is de ned only when F (L )(z ) is de ned, which happens with probability converging to ;n 0 ext 0 H H Lemma 4.2. Assume that F (L ) does not converge in law to F (L ). Then there is z 2 H such ext ext ;0 that (z ) does not converge in law to 0. ;n ;0 13 e H H Proof. If F (L ) does not converge in law to F (L ) then by de nition there are z ; : : : ; z 2 H such ext ext 1 j that F (L )[z ; : : : ; z ] does not converge in law to ext 1 j H H ;n H F (L )[z ; : : : ; z ]. To the contrary F (L )[z ; : : : ; z ] does converge in law to F (L )[z ; : : : ; z ]. ext 1 j ext 1 j ext 1 j H ;n H n n Since each contour of F (L )[z ; : : : ; z ] is surrounded by a contour of F (L )[z ; : : : ; z ], one of ext 1 j ext 1 j (z ) must not converge in law to 0. ;n i Let z be de ned by the previous lemma under the non-convergence assumption. The set of points ;0 z on the metric graph contained in or surrounded by F (L )(z ) and not in the interior surrounded ext ;0 H ;n H ;n n n by F (L )(z ), such that d(z;F (L )(z )) = (z )^ 1, is non-empty (when (z ) ext ;0 ext ;0 ;n ;0 ;n ;0 is de ned). Indeed, F (L )(z ) plus the set of points it surrounds, minus the interior surrounded by ext ;0 H ;n F (L )(z ), is connected and compact. Let Z be a random point taking values in the above ext ;0 ;n set, for instance the maximum for the lexicographical order. H H Lemma 4.3. Assume that F (L ) does not converge in law to F (L ). Then there is a sub-sequence ext ext of indices n such that the joint law of ;0 H ;n ;0 ;0 (F (L )(z ); Z ) ext ;0 ;n ;0 has a limit when n ! +1. It is a law on ;0 (F (L )(z ); Z ) ext ;0 satisfying the property that with positive probability the point Z is not contained or surrounded by F (L )(z ). ext ;0 Proof. (z ) does not converge in law to 0. This means that there is " > 0 and a sub-sequence of ;n ;0 indices n such that 0 H 0;n (4.1) 8n ;P(d(Z ;F (L )(z )) ") ": ;n ext ;0 The sub-sequence of random variables 0 0 n ;n (F (L )(z ); Z 0 ) ext ;0 ;n is tight. Indeed the rst component of the couple converges in law and the second is by de nition at distance at most 1 from the rst. Thus there is a sub-sequence of indices n out of n such that there ;0 H ;n ;0 ;0 H is a convergence in law. F (L )(z ) converges in law F (L )(z ). Let Z be de ned as ext ;0 ext ;0 H ;n ;0 ;0 the second component of the limit in law of (F (L )(z ); Z ). (4.1) implies that ext ;0 ;n ;0 P(d(Z ;F (L )(z )) ") ": ext ;0 Moreover, a.s. Z cannot be in the interior surrounded by F (L )(z ) because Z is not surrounded ext ;0 ;n H ;n by F (L )(z ). ext ;0 From now on (z ) will be a xed everywhere dense sequence in H. j j1 H H Lemma 4.4. Assume that F (L ) does not converge in law to F (L ). Then there is a family of ext ext sub-sequences of indices n such that ;j n is given by Lemma 4.3. ;0 n is a sub-sequence of n . ;j+1 ;j The random variable H ;n ;j ;j (F (L )[z ; z ; : : : ; z ]; Z ) ext ;0 1 j ;n ;j converges in law as n ! +1 and the limit de nes the joint law of ;j (F (L )[z ; z ; : : : ; z ]; Z ): ext ;0 1 j The family of joint laws on (F (L )[z ; z ; : : : ; z ]; Z ) is consistent in the sense that the ext ;0 1 j j1 law on (F (L )[z ; z ; : : : ; z ]; Z ) induced by the law of ext ;0 1 j (F (L )[z ; z ; : : : ; z ]; Z ) is the same as the one given by the convergence. In particular ext ;0 1 j+1 the law on (F (L )(z ); Z ) is the one given by Lemma 4.3. ext ;0 H H The family of laws of (F (L )[z ; z ; : : : ; z ]; Z ) uniquely de nes a law on (F (L ); Z ). ext ;0 1 j j1 ext 14 Proof. The consistency of law follows from the fact that n is a sub-sequence of n . A contour ;j+1 ;j loop in F (L ) almost surely surrounds one of the z points. Thus the fact that a consistent family ext j H H of laws on (F (L )[z ; z ; : : : ; z ]; Z ) uniquely de nes a law on (F (L ); Z ) follows from the ext ;0 1 j j1 ext Kolmogorov extension theorem. Next we explain how we extract n out of n . By construction, the sub-sequence ;j+1 ;j H ;n ;j ;j (F (L )[z ; z ; : : : ; z ]; Z ) converges in law as n ! +1 and de nes a joint law on ext ;0 1 j ;n ;j ;j H ;n ;j ;j (F (L )[z ; z ; : : : ; z ]; Z ). Moreover we have the convergence in law of F (L )(z ) to ext ;0 1 j ext j+1 H ;n ;j ;j F (L )(z ). Thus the sub-sequence (F (L )[z ; z ; : : : ; z ]; Z ) is tight and one can ext j+1 ext ;0 1 j+1 ;n ;j extract a subset of indices n such that it converges in law. The limit law is a law on ;j+1 (F (L )[z ; z ; : : : ; z ]; Z ). ext ;0 1 j+1 H H n n H Theorem 4.5. F (L ) and F (L ) converge in law as n ! +1 to F (L ), that is to say to ext ext ext 1=2 1=2 1=2 a CLE on H. Proof. It is enough to prove the convergence of F (L ). Indeed we already have the convergence for ext 1=2 H ;n H H ;n n n n F (L ) and each contour F (L )(z) lies between the contour F (L )(z) and the contour ext ext ext 1=2 1=2 1=2 F (L )(z). ext 1=2 n H Assume that F (L ) does not converge in law to F (L ). Let z be the point de ned ext ext 1=2;0 1=2 1=2 by Lemma 4.2 and n the sub-sequences de ned by Lemma 4.4. We also consider the joint law of 1=2;j (F (L ); Z ) de ned by Lemma 4.4. ext 1=2 1=2 For u; v > 0 and q > 1 we consider additional independent Poisson point processes of excursions H H E ((1; 0]) and E ([1; q]). Let A (q) be the event that is satis ed if either an excursion from 1=2;u;v u v H H E ((1; 0]) and one from E ([1; q]) intersect each other or both intersect a common contour from u v F (L ). By de nition ext 1=2 P(A (q)) = p (q): 1=2;u;v 1=2;u;v Let A (q) be the event that is satis ed if one of the following conditions holds: 1=2;u;v H H An excursion from E ((1; 0]) and one from E ([1; q]) intersect each other. u v H H H An excursion fromE ((1; 0]) and one fromE ([1; q]) intersect a common contour fromF (L ). ext u v 1=2 H H H An excursion from E ((1; 0]) intersects F (L )(z ) and an excursion from E ([1; q]) ext u 1=2;0 v 1=2 hits or surrounds Z . 1=2 H H H An excursion from E ([1; q]) intersects F (L )(z ) and an excursion from E ((1; 0]) ext 1=2;0 v u 1=2 hits or surrounds Z . 1=2 We claim that + H P(A (q)) > P(A (q)) = p (q): 1=2;u;v 1=2;u;v 1=2;u;v To see that the strict inequity holds, consider the following: Restrict to the event when Z is not contained or surrounded by the contour F (L )(z ), 1=2 ext 1=2;0 1=2 which has a positive probability. Let K by a compact subset of f=(z) 0g that contains F (L )(z ) and Z , such that ext 1=2;0 1=2 1=2 Hn K is simply connected and such that K intersects the real line on (0; +1) only. H H Since E ((1; 0]) is independent from (F (L ); Z ; K ), there is a positive probability ext u 1=2 1=2 that no excursions in E ((1; 0]), except one, hits K, and one excursion hits the contour F (L )(z ) without surrounding Z . Then the point Z is to the right from the region ext 1=2;0 1=2 1=2 1=2 H H de ned by E ((1; 0]) and the contours in F (L ) it intersects. See Figure 4 again for a ext 1=2 representation of this region. H H H Since E ([1; q]) is independent from (F (L ); Z ;E ((1; 0])), there is a positive probabil- ext 1=2 v u 1=2 H H ity that no excursion from E ([1; q]) hits the region de ned by E ((1; 0])) and the contours v u H H H in F (L ) intersected by E ((1; 0])), but one excursion from E ([1; q]) surrounds the point ext u v 1=2 Z , which is to the right from this region. 1=2 See Figure 5 for the illustration of A (q)n A (q). 1=2;u;v 1=2;u;v Let j 1. The events A (q; j) respectively A (q; j) are de ned similarly to A (q) re- 1=2;u;v 1=2;u;v 1=2;u;v H H spectively A (q), where the condition ofE ((1; 0]) andE ([1; q]) intersecting a common contour of u v 1=2;u;v 15 + H Figure 5. Illustration of A (q) where an excursion from E ((1; 0]) surrounds 1=2;u;v u H H Z and an excursion from E ([1; q]) intersects F (L )(z ). 1=2 ext 1=2;0 1=2 H H F (L ) is replaced by the condition of intersecting a common contour of F (L )[z ; z ; : : : ; z ]. ext ext 1=2;0 1 j 1=2 1=2 Then + + lim P(A (q; j)) = P(A (q)); lim P(A (q; j)) = P(A (q)): 1=2;u;v 1=2;u;v 1=2;u;v 1=2;u;v j!+1 j!+1 n;+ We will denote by A (q; j) and A (q; j) the events de ned similarly to 1=2;u;v 1=2;u;v A (q; j) and A (q; j) by doing the following replacements: 1=2;u;v 1=2;u;v e e H H H H n n E ((1; 0]) replaced by E ((1; 0]) and E ([1; q]) replaced by E ([1; q]), u u v v Z replaced by Z , 1=2 1=2;n H ;n H n H F (L ) replaced by F (L ) and F (L )[z ; z ; : : : ; z ] replaced by ext ext ext 1=2;0 1 j 1=2 1=2 1=2 H ;n F (L )[z ; z ; : : : ; z ]. ext 1 j 1=2;0 1=2 H ;n e n H H F (L )[z ; z ; : : : ; z ] converges in law to F (L )[z ; z ; : : : ; z ], the P.p.p. E ((1; 0]) ext 1=2;0 1 n ext 1=2;0 1 j 1=2 1=2 H H H to E ((1; 0]) and E ([1; q]) to E ([1; q]). Moreover, in the limit, if an excursion intersects a contour u v v loop in F (L )[z ; z ; : : : ; z ], then a.s. it goes inside the interior surrounded by the loop. Thus ext 1=2;0 1 j 1=2 the intersection still holds for small deformations of the excursion and of the contour. Thus for all j 1 we have the convergence lim P(A (q; j)) = P(A (q; j)): 1=2;u;v 1=2;u;v n!+1 From Lemma 4.4 follows that n ;+ 1=2;j + lim P(A (q; j)) = P(A (q; j)): 1=2;u;v 1=2;u;v n !+1 1=2;j H ;n n H Each contour of F (L ) is surrounded by a contour of F (L ) and Z belongs to or is sur- ext ext 1=2;n 1=2 1=2 H n;+ e n H rounded byF (L )(z ; 0). Thus, on the event A (q; j), an excursion fromE ((1; 0]) and one ext 1=2 u 1=2 1=2;u;v e H H n fromE ([1; q]) either intersect each other or intersect a common contour fromF (L )[z ; z ; : : : ; z ]. ext 1 j v 1=2;0 1=2 Thus, H n;+ p (q) P(A (q; j)): 1=2;u;v 1=2;u;v Let u be equal to u (1=2). Then 1=2;j p (q) = lim p (q) 1=2;u (1=2);v 0 1=2;u (1=2);v n !+1 1=2;j n ;+ 1=2;j lim P(A (q; j)) = P(A (q; j)): 1=2;u (1=2);v 1=2;u (1=2);v 0 0 n !+1 1=2;j 16 Taking the limit as j ! +1 we get + + p (q) lim P(A (q; j)) = P(A (q)) > 1=2;u (1=2);v 0 1=2;u (1=2);v 1=2;u (1=2);v 0 0 j!+1 P(A (q)) = p (q); 1=2;u (1=2);v 0 1=2;u (1=2);v n H which is a contradiction. It follows that F (L ) converges in law to F (L ). ext ext 1=2 1=2 H H Lemma 4.6. Let 2 (0; 1=2). Let := 1=2 . Let L and L be independent and let H H H L = L [L : 1=2 H H H Let z 6= z ~2 H. Let F (L )(z ~), respectively F (L )(z), denote the region surrounded by F (L )(z ~), ext ext ext respectively F (L )(z), i.e. the complement in H of the unique unbounded connected component of ext H H HnF (L )(z ~), respectively HnF (L )(z). The conditional probability ext ext H H H H P(F (L )(z) 6= F (L )(z ~)jF (L );F (L )(z ~)) ext ext ext ext 1=2 1=2 is a.s. positive on the event H H F (L )(z ~)\F (L )(z) = ;: ext ext Proof. On the event that F (L )(z) does not surround z ~ one can choose a continuous path ~ joining z ~ to ext H H @H = Rf0g and avoiding F (L )(z) ( ~ is thus random and measurable with respect to F (L )(z)). ext ext H H Let K be the union of ~, F (L )(z ~) and all the contours in F (L ) that do intersect either ~ or ext ext e e F (L )(z ~). Let Hull(K ) be the hull of K , that is to say the complement in H of the unique unbounded ext connected component of Hn K . H H On the event that F (L )(z) does not intersect F (L )(z ~), z does not belong to Hull(K ). One ext ext can than choose a path that connects z to @H and avoids Hull(K ), being random measurable with respect to Hull(K ). Let K be the union of and all the contours in F (L ) that intersects . Let ext Hull(K ) be the hull of K . Figure 6 is an illustration of ~, , K and K . Figure 6. Illustration of ~, , K and K . ~, and F (L )(z ~) are drawn in full lines. ext Elements of F (L ) are drawn in dashed lines. ext By construction, on the event H H F (L )(z ~)\F (L )(z) = ;; ext ext we have Hull(K )\ Hull(K ) = ;, Hn (Hull(K )[ Hull(K )) is simply connected, no Brownian loop from L crosses the boundary of Hull(K ) or Hull(K ) and in particular a contour in F (L ) is either inside Hull(K ), Hull(K ) or inside the complement Hn (Hull(K )[ ext Hull(K )). 17 Conditional on H H F (L )(z ~)\F (L )(z) = ;; ext ext Hn(Hull(K)[Hull(K)) and on Hull(K ), Hull(K ), the law of the contours F (L ), created by the loops ext 1=2 Hn(Hull(K)[Hull(K)) L from L that stay inside Hn (Hull(K )[ Hull(K )), is a CLE inside Hn (Hull(K )[ 1=2 1=2 Hn(Hull(K)[Hull(K)) Hull(K )), and they are conditionally independent from L nL . 1=2 1=2 Conditional on the event H H F (L )(z ~)\F (L )(z) = ; ext ext H H and on Hull(K ), Hull(K ), F (L ), F (L )(z ~), the probability that ext ext H H F (L )(z) = F (L )(z ~) ext ext 1=2 1=2 is less or equal to the probability that Hull(K ) and Hull(K ) are connected by a cluster of L , which 1=2 Hn(Hull(K)[Hull(K)) is less or equal to the probability that given the contours F (L ) and an independent ext 1=2 loop-soup in H of parameter , there is a contour and two loops and in the loop-soup of intensity 1 2 such that intersects and Hull(K ), intersects and Hull(K ). The latter conditional probability is a.s. strictly smaller than 1. This is what we needed to prove. H H H n n Theorem 4.7. Let 2 (0; 1=2). F (L ) and F (L ) converge in law as n ! +1 to F (L ), ext ext ext that is to say to a CLE on H. () H H Proof. As for Theorem 4.5, it is enough to prove that F (L ) converges in law to F (L ). Let us ext ext assume that this is not the case. Let z be the point and n the sub-sequence de ned by Lemma 4.2. ;0 ;0 We also consider the joint law of (F (L ); Z ) de ned by Lemma 4.4. ext Since H H lim P F (L )(z ) = F (L )(z) = 0 ext ;0 ext z!1 and P d(Z ;F (L )(z )) > 0 > 0; ext ;0 we can choose z ~2 H such that H H H P F (L )(z ) = F (L )(z ~) < P d(Z ;F (L )(z )) > 0 : ext ;0 ext ext ;0 In that way H H H P d(Z ;F (L )(z )) > 0;F (L )(z ) 6= F (L )(z ~) > 0: ext ;0 ext ;0 ext H e H H n H Let := 1=2. We take L independent from (L ; Z ) and L independent from (L ; Z ). We ;n H n de ne L and L as unions of two independent Poisson point processes: 1=2 1=2 e e e H H H H H H n n n L = L [L ; L = L [L : 1=2 1=2 H H + Let A be the event de ned by F (L )(z ) = F (L )(z ~). Let A be the event which holds if ext ;0 ext 1=2 1=2 one of the below conditions is satis ed: H H F (L )(z ) = F (L )(z ~), ext ;0 ext 1=2 1=2 F (L )(z ~) surrounds Z . ext Figure 7 is an illustration of A n A . Let us show that P(A n A ) > 0. Let E be the event de ned by the following four conditions: d(Z ;F (L )(z )) > 0, ext ;0 H H F (L )(z ) 6= F (L )(z ~), ext ;0 ext F (L )(z ~) surrounds Z , ext H H F (L )(z ~)\F (L )(z ) = ;. ;0 ext ext H H It has positive probability because of our choice of z ~ and the independence ofF (L )(z ~) from (F (L ); Z ). ext ext H H Let A be the complement of A . A and E are independent conditional on (F (L );F (L )(z ~)). 4 ext ext Thus + H H P(A n A ) = P(E ; A ) = E[1 P(A jF (L );F (L )(z ~))]: 4 E ext ext 18 + Figure 7. Illustration of A n A . H H According to Lemma 4.6, P(A jF (L );F (L )(z ~)) is a.s. positive on the event E . It follows that ext ext 4 P(A n A ) > 0. n n;+ + H Let A and A be the events de ned similarly to A and A where the contours F (L )(z ), ext ;0 1=2 H H F (L )(z ~) and F (L )(z ~) are replaced by ext ext 1=2 e e e H H H n n n F (L )(z ), F (L )(z ~) and F (L )(z ~) respectively and Z is replaced by Z . Since Z is ext ;0 ext ext ;n ;n 1=2 1=2 H n;+ n on the contour F (L )(z ) we have the equality A = A . From Theorem 4.5 follows that ext ;0 lim P(A ) = P(A ): n!+1 On the other hand n ;+ + ;0 lim inf P(A ) P(A ) > P(A ); n !+1 ;0 H H which is a contradiction. It follows that F (L ) converges in law to F (L ). ext ext Acknowledgements This research was supported by Universit e Paris-Sud, Orsay. The author thanks Wendelin Werner for explaining the theory of restriction measures and pointing out the 1=2 factor in the relation between the loop-soup intensity parameter and the central charge. References [1] J. Aru, A. Sepulv eda, and W. Werner. On bounded-type thin local sets of the two-dimensional Gaussian free eld. J. Inst. Math. Jussieu:1{28, 2017. [2] T. Van de Brug, F. Camia, and M. Lis. Random walk loop soups and conformal loop ensembles. Probab. Theory Related Fields, 166:553{584, 2016. [3] Y. Le Jan. Markov paths, loops and elds. In 2008 St-Flour summer school, L.N. Math., volume 2026. Springer, [4] G. F. Lawler and J. A. Trujillo-Ferreras. Random walk loop soup. Trans. Amer. Math. Soc., 359(2):767{787, 2007. [5] G. F. Lawler and W. Werner. The Brownian loop-soup. Probab. Theory Related Fields, 128:565{588, 2004. [6] G.F. Lawler. Partition functions, loop measure, and versions of SLE. J. Stat. Phys., 134:813{837, 2009. [7] G.F. Lawler and V. Limic. Random walk: a modern introduction, volume 123 of Cambridge Stud. Adv. Math. Cambridge University Press, 1st edition, 2010. [8] Y. Le Jan, M.B. Marcus, and J. Rosen. Permanental elds, loop soups and continuous additive functionals. Ann. Probab., 43(1):44{84, 2015. [9] T. Lupu. From loop clusters and random interlacements to the free eld. Ann. Probab., 44(3):2117{2146, 2016. [10] T. Lupu. Loop percolation on discrete half-plane. Electron. Commun. Probab., 21(30), 2016. [11] J. Miller and S. Sheeld. CLE(4) and the Gaussian free eld. In preparation. [12] O. Schramm and S. Sheeld. Contour lines of the two-dimensional discrete Gaussian free eld. Acta Math., 202:21{ 137, 2009. [13] O. Schramm and S. Sheeld. A contour line of the continuum Gaussian free eld. Probab. Theory Related Fields, 157:47{80, 2013. 19 [14] S. Sheeld and W. Werner. Conformal loop ensembles: the Markovian characterization and the loop-soup construc- tion. Ann. of Math., 176(3):1827{1917, 2012. [15] M. Wang and H. Wu. Level lines of Gaussian free eld I: zero-boundary GFF. Stochastic Process. Appl., 127(4):1045{ 1124 , 2017. [16] W. Werner. SLEs as boundaries of clusters of Brownian loops. C.R. Acad. Sci. Paris, 337:481{486, 2003. [17] W. Werner. Random planar curves and Schramm-Loewner Evolutions. In 2002 St-Flour summer school, L.N. Math., volume 1840. Springer, 2004. [18] W. Werner. Conformal restriction and related questions. Probab. Surv., 2:145{190, 2005. [19] W. Werner and H. Wu. From CLE() to SLE(; )'s. Electron. J. Probab., 18:1{20, 2013.

Mathematics – arXiv (Cornell University)

**Published: ** Feb 24, 2015

Loading...

You can share this free article with as many people as you like with the url below! We hope you enjoy this feature!

Read and print from thousands of top scholarly journals.

System error. Please try again!

Already have an account? Log in

Bookmark this article. You can see your Bookmarks on your DeepDyve Library.

To save an article, **log in** first, or **sign up** for a DeepDyve account if you don’t already have one.

Copy and paste the desired citation format or use the link below to download a file formatted for EndNote

Access the full text.

Sign up today, get DeepDyve free for 14 days.

All DeepDyve websites use cookies to improve your online experience. They were placed on your computer when you launched this website. You can change your cookie settings through your browser.