betaFeedback

WorksMathematics

Generic arm separation for planar percolation models

Abstract

Arm separation is an important technical tool in the study of critical and near-critical planar percolation systems. When combined with RSW estimates, it proves quasi-multiplicativity of arm-event probabilities, links the derivatives of different events and allows the identification of universal arm exponents. We provide a very generic statement of arm separation that applies to arbitrary planar percolation models. Our argument is based on exploring configurations conditioned on the existence of arms in such a way that "exposed" arms are revealed with positive probability. The exploration algorithm is independent of the percolation model and applies to all polychromatic arm events. As an application, we prove a strong arm-separation result for critical planar FK-percolation.

Introduction

Arm separation is a fundamental tool in the study of planar critical and near-critical percolation. An arm event denotes the existence of disjoint paths of prescribed colours joining the two boundaries of an annulus. Informally, arm separation states that, conditional on such an event, the arms can be chosen so that their endpoints on the outer boundary are separated by distances comparable to the outer radius, with uniformly positive probability. An analogous statement applies at the inner boundary.

Arm separation was introduced by Kesten [23] for planar Bernoulli percolation; see [27] for a detailed exposition. Its usefulness lies in the room it provides to extend arms and join them to other paths using Russo–Seymour–Welsh (RSW) estimates. Keeping the endpoints sufficiently far apart allows these connections to be constructed with uniformly positive probability, which is crucial when comparing events across different scales. Arm separation is thus a technical but essential ingredient in a wide range of applications, including quasi-multiplicativity [23, 27, 7, 11], the derivation of scaling relations [22, 23, 27, 17, 18, 10], the construction of incipient infinite clusters [21 Jár03, 14 BS17], noise sensitivity [30, 16 TV25] and identification and comparison of arm exponents [31, 5 KSL25, GM25, 28] to name just a few.

In this paper, we define a related geometric property, which we call exposition. An exploration algorithm will reveal the configuration step by step and, conditional on the arm event, will produce exposed paths of the prescribed colours with uniformly positive probability. These paths will have adjacent unexplored regions through which suitable crossings may be glued to them. The algorithm is the same for every percolation measure and requires no assumptions such as RSW estimates, positive association or independence.

Unlike classical proofs, which intertwine exploration and RSW techniques, our approach completely separates the geometric and probabilistic aspects of arm separation. We show how to explore the configuration in order to produce exposed arms. One may then use RSW estimates to glue these arms to desired destinations.

As an illustration, we use our approach to produce a strong arm-separation result for critical FK-percolation. This separation property applies to any polychromatic arm event (even non-alternating) in an annulus ΛR∖Λr\Lambda_R \setminus\Lambda_r (with r<R/2r < R/2) and will separate the arms on the outer boundary. It holds even conditionally on the configuration in ΛR/2\Lambda_{R/2}. A precise statement is provided in § 5.

Definitions and main result

A tiling of the plane is a family G\mathcal{G} of closed, bounded sets, called cells, with pairwise disjoint interiors and whose union is R2\mathbb{R}^2. We assume that every cell is simply connected and bounded by a simple closed curve of finite length. Furthermore, we restrict our attention to locally finite tilings, meaning that every compact set of the plane intersects only a finite number of cells. Finally, we require that no point belongs to the boundaries of more than three cells.

A percolation configuration on G\mathcal{G} is an element of {0,1}G\{0,1\}^{\mathcal{G}}, viewed as an assignment of colours to the cells of G\mathcal{G}. Cells of colour 0 are called closed, while those of colour 1 are called open; these will be represented by yellow and blue1, respectively, in all figures. A percolation model P\mathbb{P} is a probability measure on the set {0,1}G\{0,1\}^{\mathcal{G}} equipped with the standard cylinder σ\sigma-algebra. No further assumptions are imposed on G\mathcal{G} or P\mathbb{P} at this stage.

Fix 0<r<R/20 < r < R/2 and z∈R2z \in\mathbb{R}^2. Let Λr(z)=z+[−r,r]2\Lambda_r(z) = z + [-r,r]^2 denote the box of radius rr centred at zz, and set Λr=Λr(0)\Lambda_r = \Lambda_r(0). We write ∂Λr\partial\Lambda_r for its topological boundary and Ann⁡(r,R)=ΛR∖Λr\operatorname{Ann}(r,R) = \Lambda_R \setminus\Lambda_r for the closed annulus of inner radius rr and outer radius RR. Finally, for any subset A⊂R2A \subset\mathbb{R}^2, we write G(A)\mathcal{G}(A) for the union of cells of G\mathcal{G} intersecting AA.

A colour sequence σ\sigma is an element of {0,1}N\{0,1\}^{N} for some N≥1N \ge1. We call it polychromatic if it contains both values 0 and 1, and monochromatic otherwise.

A cell-path γ\gamma is a finite sequence of distinct cells in which consecutive cells are adjacent. Two cell-paths are disjoint if they have no cells in common. For a configuration ω∈{0,1}G\omega\in\{0,1\}^{\mathcal{G}}, a cell-path is called open (resp. closed) in ω\omega if all its cells are open (resp. closed) in ω\omega.

Furthermore, a cell-path γ\gamma in G(Ann⁡(r,R))\mathcal{G}(\operatorname{Ann}(r,R)) is said to be a crossing of the annulus Ann⁡(r,R)\operatorname{Ann}(r,R) if it contains a cell intersecting ∂Λr\partial\Lambda_r and a cell intersecting ∂ΛR\partial\Lambda_R. We say that Ann⁡(r,R)\operatorname{Ann}(r,R) contains arms of colours σ\sigma if there exist disjoint cell-paths γ1,…,γN\gamma_1,\ldots,\gamma_N crossing Ann⁡(r,R)\operatorname{Ann}(r,R), indexed in counterclockwise order, such that γi\gamma_i is open (resp. closed) if σi=1\sigma_i = 1 (resp. σi=0\sigma_i = 0). The paths γi\gamma_i are called arms and we write Aσ(r,R)A_\sigma(r,R) for the event corresponding to their existence. See Figure 1 (left) for an example.

Figure 1. Left: A percolation configuration on a tiling G\mathcal{G}. The annulus Ann⁡(r,R)\operatorname{Ann}(r,R) is depicted in black. The configuration contains six arms corresponding to the colour sequence σ=(1,0,0,1,1,0)\sigma=(1,0,0,1,1,0). Right: An exploration that exposes each arm. The exposing tubes are marked in red; the exposition zones are the unexplored regions surrounded by bold red lines.

Our approach crucially relies on explored sets. Loosely speaking, an explored set Exp⁡\operatorname{Exp} is a union of cells of G\mathcal{G} that is measurable in terms of the colours of the cells it contains, along with a source of randomness independent of the percolation configuration2. We will actually use the term “explored” in a more restrictive sense — see Definition 4 below.

A tube of quality η>0\eta> 0 at scale RR is the union T=⋃i=1mΛηR(xi)T = \bigcup_{i=1}^m \Lambda_{\eta R}(x_i) where x1,…,xmx_1,\ldots,x_m are distinct points belonging to a common translate of 2ηRZ22\eta R\mathbb{Z}^2 with ∥xi−xi−1∥2=2ηR\lVert x_i-x_{i-1}\rVert_2 = 2\eta R for all 2≤i≤m2 \le i \le m and with m≤η−2m \le\eta^{-2}. The starting box of TT is ΛηR(x1)\Lambda_{\eta R}(x_1) and its ending box is ΛηR(xm)\Lambda_{\eta R}(x_m). A crossing of TT is any continuous curve in TT starting in ΛηR(x1)\Lambda_{\eta R}(x_1) and ending in ΛηR(xm)\Lambda_{\eta R}(x_m).

Given an explored set Exp⁡⊂G(Ann⁡(r,R))\operatorname{Exp} \subset\mathcal{G}(\operatorname{Ann}(r,R)), a cell-path γ\gamma is said to be η\eta-exposed towards the outside if

  1. γ⊂Exp⁡\gamma\subset\operatorname{Exp},

  1. there exists a tube TT of quality η\eta at scale RR contained in Ann⁡(R/2,(1+2η)R)\operatorname{Ann}(R/2,(1+2\eta)R), whose starting box lies in ΛRc\Lambda_R^c and such that

  1. any crossing of TT necessarily intersects γ\gamma and

  1. the connected component of the start of TT in T∖γT \setminus\gamma — henceforth referred to as the exposition zone of γ\gamma in TT — is disjoint from Exp⁡\operatorname{Exp}.

We will also say that γ\gamma is exposed by TT in Exp⁡c\operatorname{Exp}^{c}. One can define a similar notion of η\eta-exposition towards the inside by considering tubes of quality η\eta at scale rr, contained in Ann⁡((1−2η)r,2r)\operatorname{Ann}((1-2\eta)r,2r) with their starting box contained in Λr\Lambda_r.

Finally, we say Exp⁡\operatorname{Exp} η\eta-exposes paths of colours σ\sigma towards the outside if

  1. Exp⁡⊂G(Ann⁡(r,R))\operatorname{Exp}\subset\mathcal{G}(\operatorname{Ann}(r,R)),

  2. Exp⁡\operatorname{Exp} contains disjoint cell-paths γ1,…,γN\gamma_1,\ldots,\gamma_N starting on ∂Λr\partial\Lambda_r, with each γi\gamma_i of colour σi\sigma_i, indexed in counterclockwise order and

  3. the paths γ1,…,γN\gamma_1,\ldots,\gamma_N are η\eta-exposed towards the outside, with pairwise disjoint exposition zones.

See Figure 1 (right) for an illustration. A similar definition applies towards the inside.

Theorem 1 (Arm exposition). For any polychromatic colour sequence σ∈{0,1}N\sigma\in\{0,1\}^{N}, there exist constants η=η(σ)>0\eta=\eta(\sigma)>0 and c=c(σ)>0c=c(\sigma)>0 such that the following holds. For any 0<r<R/20<r<R/2, any tiling of the plane G\mathcal{G} such that all cells intersecting ΛR\Lambda_R have diameter at most ηR\eta R, and any percolation measure P\mathbb{P} on G\mathcal{G}, there exists an explored set Exp⁡\operatorname{Exp} such that

P[Exp⁡ η-exposes paths of colours σ towards the outside∣Aσ(r,R)]≥c,(1)\mathbb{P}\left[\operatorname{Exp}\ \eta\text{-exposes paths of colours }\sigma\text{ towards the outside}\mid A_\sigma(r,R)\right]\ge c, \tag*{(1)}

as long as the event in the conditioning has positive probability. An analogous result applies for exposition towards the inside.

Remark 2. A similar statement for monochromatic arm events is only available in a simply connected setting — see Theorem 3 below. We were unable to prove such a statement for the annulus. However, in many applications involving monochromatic arm events, separation may be replaced by applications of the FKG inequality.

Comments and applications

Minimal assumptions. It may be surprising that Theorem 1 applies to any percolation measure, with no assumption of properties of criticality, most notably the RSW property. Note however that the exposition of the different arms is generally only useful when combined with versions of the RSW, Spatial Markov and FKG properties.

Uniformity over percolation models. It will be apparent from the proofs that the set Exp is revealed by an exploration algorithm (formalised as a decision tree in § 2) that queries the colours of cells one by one. When successful, the algorithm produces an explored set that η\eta-exposes the desired paths. The same algorithm is used for all percolation models and its probability of success is bounded below by a positive constant depending only on the colour sequence σ\sigma. This allows us to state the theorem uniformly over percolation models on tilings.

Probability of success. The probability of success in (1) may be rendered arbitrarily close to 1 by shrinking η>0\eta> 0. Indeed, for fixed δ>0\delta> 0, taking η>0\eta> 0 small enough, one may force the exploration in Theorem 1 to satisfy

Exp⁡⊂G(Ann⁡(r,(α+δ)R))\operatorname{Exp} \subset\mathcal{G}(\operatorname{Ann}(r,(\alpha+\delta)R))

for any 1/2≤α≤1−δ1/2 \le\alpha\le1-\delta, without affecting the probability of success cc. Repeating the procedure in successive annuli of “thickness” δR\delta R yields independent attempts at exposition, so that the overall probability of success is at least 1−(1−c)1/(2δ)1-(1-c)^{1/(2\delta)}. Taking δ\delta small enough renders the probability of success arbitrarily close to one.

Robustness. The interpretation of percolation as colourings of cells allows one to incorporate site or bond percolation models on any bi-periodic planar lattice, and much more. Bond percolation is recovered by viewing the colour 0 as the dual percolation model, while site percolation is recovered by viewing closed cells as the closed vertices, with the connectivity notion given by the matching graph. See Figure 2 for examples of cell representations of bond and site percolation on Z2\mathbb{Z}^{2}.

Figure 2. Left: A bond percolation on the square lattice and its interpretation as a colouring of cells. The cells corresponding to vertices are all blue (open), while those corresponding to faces are all yellow (closed). The cells corresponding to edges are coloured depending on the state of the corresponding edge. Right: A site percolation on the square lattice and its interpretation on the same graph of cells G\mathcal{G}. Here the cells corresponding to edges are all blue and those corresponding to faces are all yellow. The cells corresponding to vertices carry all the randomness.

Theorem 1 also applies to less classical settings, such as Voronoi percolation or percolation on random planar maps, and, with a natural adaptation, to Boolean percolation, where balls of random radius are placed in the plane according to a Poisson point process. In the latter setting, arm events for alternating colour sequences are defined exactly as above. A separate definition is needed to make sense of several consecutive arms in the occupied set, while the corresponding notion for the vacant set does not make sense.

We believe that those interested in applying Theorem 1 in these non-standard settings will find it easily adaptable.

One caveat applies to models such as Voronoi or Boolean percolation, where the tiling itself is random. Theorem 1 may be generalised to this case with Exp revealing both the colour and the shape of each explored cell. We will not detail this here, as the adaptation is immediate. We do however warn the interested reader that, in such contexts, the knowledge of the cell shapes in Exp⁡\operatorname{Exp} may affect the RSW estimates in the unexplored regions. For more background on RSW estimates in Voronoi and Boolean percolation, see [32, 1] and [2], respectively. It is worth mentioning that the quenched result of [1] may actually allow one to perform arm exposition by first revealing the Voronoi tessellation and then applying the algorithm of Theorem 1 on it.

Motivation. The present work is mainly motivated by applications to critical FK-percolation, where RSW techniques give lower bounds on the probability of open and closed crossings of tubes of given quality. Combined with the FKG inequality and the spatial Markov property, these estimates allow one to construct crossings in unexplored regions and glue them to exposed paths. This is the basic mechanism underlying arm separation.

The classical proofs of arm separation for Bernoulli percolation [23, 27] extend to FK-percolation when suitable RSW estimates are available. For alternating arm events, the usual box-crossing estimate (see (RSW)) is sufficient. For general colour sequences, a complete proof for FK-Ising percolation is given in [7], using stronger RSW estimates for arbitrary topological rectangles. The same approach extends to 1≤q<41 \le q < 4 using [11], but does not cover q=4q = 4, where these stronger estimates are expected to fail.

Our argument treats arbitrary polychromatic colour sequences using only RSW estimates uniform over boundary conditions at macroscopic distance. These estimates remain valid at q=4q = 4, so our approach provides a unifying framework. Moreover, it yields a stronger arm-separation result which we describe next.

Strong arm separation under conditioning. An important consequence of our approach is that arm separation remains valid under arbitrary compatible conditioning in the inner annulus Ann⁡(r,R/2)\operatorname{Ann}(r, R/2). Indeed, fix a configuration in Ann⁡(r,R/2)\operatorname{Ann}(r, R/2) under which Aσ(r,R)A_{\sigma}(r, R) has positive conditional probability. Since Theorem 1 applies uniformly over percolation measures, it also applies to the FK-percolation measure conditioned on the fixed configuration in Ann⁡(r,R/2)\operatorname{Ann}(r, R/2).

To turn the exposed paths into a more standard notion of well-separated arms (see Definition 7), we use RSW constructions in the exposition zones. As we will show in the proof, these zones lie at macroscopic distance from the conditioned region. The spatial Markov property and the uniformity of RSW estimates over boundary conditions at such distances therefore give a positive lower bound on the probability of these constructions, independently of the prescribed configuration in Ann⁡(r,R/2)\operatorname{Ann}(r, R/2). This yields arm separation uniformly under this conditioning; see Theorem 8 for the precise statement.

In particular, our approach implies the “super-strong” separation statement conjectured by Garban, Pete and Schramm in [17 Conjecture A.4]. A half-plane version was previously obtained in [13] for critical site percolation on the triangular lattice. Our result covers the full-plane case and extends this form of separation to critical FK-percolation.

Exposed crossings in rectangles

An intermediate step in our proof is an exposition result similar to Theorem 1, but for crossings of quads (that is, topological rectangles). We state it in the context of Euclidean rectangles for simplicity.

For H≥2R>0H \ge2R > 0, define Rect⁡(R,H)=[−R,R]×[0,H]\operatorname{Rect}(R,H) = [-R,R] \times[0,H], with its four sides denoted ∂TRect⁡(R,H)\partial_{\mathrm{T}}\operatorname{Rect}(R,H), ∂LRect⁡(R,H)\partial_{\mathrm{L}}\operatorname{Rect}(R,H), ∂BRect⁡(R,H)\partial_{\mathrm{B}}\operatorname{Rect}(R,H) and ∂RRect⁡(R,H)\partial_{\mathrm{R}}\operatorname{Rect}(R,H), in counterclockwise order, starting from the top. For a colour sequence σ\sigma, let Aσ(Rect⁡(R,H))A_{\sigma}(\operatorname{Rect}(R,H)) be the event that Rect⁡(R,H)\operatorname{Rect}(R,H) contains disjoint cell-paths from ∂BRect⁡\partial_{\mathrm{B}}\operatorname{Rect} to ∂TRect⁡\partial_{\mathrm{T}}\operatorname{Rect} of colours σ1,…,σN\sigma_1,\ldots,\sigma_N when indexed from left to right.

An explored set Exp⁡⊂G(Rect⁡(R,H))\operatorname{Exp} \subset\mathcal{G}(\operatorname{Rect}(R,H)) is said to η\eta-expose a cell-path γ\gamma towards the top if

  1. γ⊂Exp⁡\gamma\subset\operatorname{Exp},

  2. there exists a tube TT of quality η\eta at scale RR whose starting box is contained in [−R,R]×[H,+∞)[-R,R] \times[H,+\infty),

  3. any crossing of TT necessarily intersects γ\gamma and

  4. the connected component of the start of TT in T∖γT \setminus\gamma — henceforth referred to as the exposition zone of γ\gamma in TT — is entirely contained in the unexplored part of [−R,R]×[H−R,+∞)[-R,R] \times[H-R,+\infty).

As before, we will also say that γ\gamma is exposed by TT in Exp⁡c\operatorname{Exp}^{c}. That the exposition zones are required to be above the height H−RH-R is similar to them being required to be outside of ΛR/2\Lambda_{R/2} in Theorem 1. It ensures that the exposition actually occurs near the top boundary of the rectangle.

Finally, we say Exp⁡\operatorname{Exp} η\eta-exposes crossings of colours σ\sigma towards the top if

  1. Exp⁡⊂G(Rect⁡(R,H))\operatorname{Exp} \subset\mathcal{G}(\operatorname{Rect}(R,H)),

  2. Exp⁡\operatorname{Exp} contains disjoint cell-paths γ1,…,γN\gamma_1,\ldots,\gamma_N starting on ∂BRect⁡(R,H)\partial_B\operatorname{Rect}(R,H) with each γi\gamma_i of colour σi\sigma_i, indexed from left to right and

  3. the cell-paths γ1,…,γN\gamma_1,\ldots,\gamma_N are η\eta-exposed towards the top, with their exposition zones being disjoint.

Theorem 3. For any colour sequence σ∈{0,1}N\sigma\in\{0,1\}^{N}, there exist constants η=η(σ)>0\eta=\eta(\sigma)>0 and c=c(σ)>0c=c(\sigma)>0 such that the following holds. For any H≥2R>0H \ge2R>0, any tiling G\mathcal{G} such that all cells intersecting Rect⁡(R,H)\operatorname{Rect}(R,H) have diameter at most ηR\eta R and any percolation measure P\mathbb{P} on G\mathcal{G}, there exists an explored set Exp⁡\operatorname{Exp} such that

P[Exp⁡ η-exposes crossings of colours σ towards the top∣Aσ(Rect⁡(R,H))]≥c,\mathbb{P}\left[\operatorname{Exp}\ \eta\text{-exposes crossings of colours }\sigma\text{ towards the top}\mid A_{\sigma}(\operatorname{Rect}(R,H))\right] \ge c,

as long as the event in the conditioning has positive probability.

Unlike Theorem 1, this result also applies to monochromatic colour sequences. Indeed, the fact that the rectangle has natural notions of left and right boundaries — which boils down to it being simply connected — renders the proof simpler. Actually, we deduce Theorem 1 by cutting the annulus so that it becomes simply connected without altering the arm event, then applying Theorem 3. In the cutting step, we use that the colour sequence is polychromatic to explore an interface.

Structure of the paper. In § 2, we formally define the notion of explored set. § 3 contains the proof of Theorem 3, which is then used in § 4 to show Theorem 1. Finally, § 5 contains consequences for FK-percolation.

Exploration algorithms

As stated in the introduction, the explored set Exp⁡\operatorname{Exp} will be the result of an exploration procedure. Such a procedure consists of revealing a random configuration ω\omega cell by cell, with the choice of the next cell to be revealed depending on all previously revealed information and, potentially, on an external source of randomness that is independent of ω\omega. We formalise this notion using decision trees.

Decision trees

Fix a tiling G\mathcal{G} and a configuration ω\omega. For a sequence of cells e=(e1,e2,…)e=(e_1,e_2,\ldots) and t≥0t \ge0, we write e[t]=(e1,…,et)e_{[t]}=(e_1,\ldots,e_t) and ω[t]=(ωe1,…,ωet)\omega_{[t]}=(\omega_{e_1},\ldots,\omega_{e_t}). Write G[t]\mathcal{G}_{[t]} for the collection of ordered tt-tuples of cells, so that e[t]∈G[t]e_{[t]} \in\mathcal{G}_{[t]}. We will sometimes slightly abuse notation and also write e[t]e_{[t]} for the set {e1,…,et}\{e_1,\ldots,e_t\} whenever it is convenient.

A decision tree T\mathbf{T} is an initial cell e1e_1 and a set of functions ψt:G[t]×{0,1}t→G\psi_t:\mathcal{G}_{[t]} \times\{0,1\}^{t} \to\mathcal{G} for t≥1t \ge1. It should be seen as associating to each pair (e[t],ω[t])(e_{[t]},\omega_{[t]}) of revealed cells the next cell to reveal. In particular, we ask that ψt(e[t],ω[t])∉{e1,…,et}\psi_t(e_{[t]},\omega_{[t]}) \notin\{e_1,\ldots,e_t\}.

When the decision tree is fixed, we write (e[t],ωe[t])t≥1(e_{[t]},\omega_{e_{[t]}})_{t\geq1} for the unique sequence starting with (e1,ωe1)(e_1,\omega_{e_1}) and such that et+1=ψt(e[t],ωe[t])e_{t+1}=\psi_t(e_{[t]},\omega_{e_{[t]}}) for all t≥1t\geq1. In the rest of the paper, we will interpret decision trees as exploration algorithms and refer to them as such.

A stopping time for T\mathbf{T} is a random variable τ\tau taking values in N∪{∞}\mathbb{N}\cup\{\infty\} which is such that {τ≤t}\{\tau\leq t\} is measurable with respect to (e[t],ωe[t])(e_{[t]},\omega_{e_{[t]}}). The set of cells explored up to the stopping time τ\tau (including the order in which they were revealed) is e[τ]e_{[\tau]} and the configuration revealed at that time is ωe[τ]\omega_{e_{[\tau]}}.

We call a decision tree random if both the choice of e1e_1 and the functions ψt\psi_t also depend on an external source of randomness, which may be encoded by a sequence (Ut)t∈N(U_t)_{t\in\mathbb{N}} of independent uniform variables on [0,1][0,1] which is independent of ω\omega. The notion of stopping time adapts readily, by imposing that {τ≤t}\{\tau\leq t\} is measurable with respect to (e[t],ωe[t],(Us)s≤t)(e_{[t]},\omega_{e_{[t]}},(U_s)_{s\leq t}).

Definition 4. An explored set Exp⁡\operatorname{Exp} is a random set of cells of G\mathcal{G} for which there exists a random decision tree and a stopping time τ\tau such that Exp⁡=e[τ]\operatorname{Exp}=e_{[\tau]}.

Interface exploration

An important example is the exploration of an interface between open and closed cells. See Figure 3 for a helpful illustration.

Figure 3. Interface exploration: A domain Ω\Omega surrounded by the black curve, with three points a,b,ca,b,c on the boundary. The exploration path starting at aa and ending on the arc (b,c)(b,c) is depicted in red. The cells of Ω\Omega adjacent to the arcs (a,b)(a,b) and (c,a)(c,a) are considered open and closed, respectively. Only the cells in Ω\Omega adjacent to the exploration path are explored (brighter colours).

Let Ω\Omega be a finite union of cells whose boundary is a simple closed curve. Let a,ba,b and cc be three distinct points on the boundary of Ω\Omega, placed in counterclockwise order, each at the interface of three cells, with exactly one of the three cells belonging to Ω\Omega. Write (a,b)(a,b), (b,c)(b,c) and (c,a)(c,a) for the three boundary arcs going in counterclockwise direction.

The exploration curve Γ\Gamma will be a path of edges of G\mathcal{G} started from aa such that all cells on one side of Γ\Gamma are of one colour and all cells on the other side of Γ\Gamma are of the opposite colour. For definiteness, we describe an exploration following edges with closed on the left and open on the right. To ensure that Γ\Gamma remains inside Ω\Omega, we impose open boundary conditions on the arc (a,b)(a,b) and closed boundary conditions on (c,a)(c,a). This will be made precise below. The remaining arc, including its endpoints bb and cc, is the target at which the exploration stops.

We write (Γn)n≥0(\Gamma_n)_{n\geq0} and (Γ(n,n+1))n≥0(\Gamma_{(n,n+1)})_{n\geq0} for the sequence of vertices and oriented edges3, respectively, visited by Γ\Gamma. Set Γ0=a\Gamma_0=a. To properly initialise the procedure, we also define Γ(−1,0)\Gamma_{(-1,0)} as the unique edge between the two exterior cells incident to aa, oriented towards aa.

At each step n≥0n \ge0, first check whether Γn\Gamma_n lies on the target arc (b,c)∪{b,c}(b,c) \cup\{b,c\}. If so, stop the exploration. Otherwise, proceed as follows.

For n=0n=0, let ee be the unique cell in Ω\Omega incident to aa. For n≥1n \ge1, let ee be the unique cell incident to Γn\Gamma_n that is not incident to the incoming edge Γ(n−1,n)\Gamma_{(n-1,n)}. If ee belongs to Ω\Omega, reveal its colour (unless it has already been revealed) and let Γ(n,n+1)\Gamma_{(n,n+1)} be the unique outgoing edge adjacent to ee that preserves closed on the left and open on the right. If ee lies outside Ω\Omega, use the boundary condition imposed above. More precisely, the following rules should be applied for n≥0n \ge0:

  • (i) if e⊂Ωe \subset\Omega and ee is open, let Γ(n,n+1)\Gamma_{(n,n+1)} be the outgoing edge to the left as seen from Γ(n−1,n)\Gamma_{(n-1,n)};

  • (ii) if e⊂Ωe \subset\Omega and ee is closed, let Γ(n,n+1)\Gamma_{(n,n+1)} be the outgoing edge to the right as seen from Γ(n−1,n)\Gamma_{(n-1,n)};

  • (iii) if e⊄Ωe \not\subset\Omega and Γn\Gamma_n lies on (a,b)(a,b), pretend that ee is open and follow the same procedure as in (i);

  • (iv) if e⊄Ωe \not\subset\Omega and Γn\Gamma_n lies on (c,a)(c,a), pretend that ee is closed and follow the same procedure as in (ii).

After choosing the outgoing edge Γ(n,n+1)\Gamma_{(n,n+1)}, denote its endpoint by Γn+1\Gamma_{n+1}. Recording the cells revealed during this procedure defines a decision-tree exploration.

Due to our imposition of boundary conditions and Ω\Omega containing only finitely many cells, the exploration curve will reach the target after finitely many steps. In particular, the number τ\tau of cells that are revealed before the exploration stops is finite and is a stopping time for the associated decision tree. Formally, the exploration is only defined up to τ\tau, but one may continue the exploration after τ\tau using arbitrary rules.

Exposition in a rectangle: Proof of Theorem 3

Before delving into the proof of Theorem 3, let us make some useful observations about the notion of tubes.

Cutting tubes

Recall that a tube TT of quality η\eta at scale RR is T=⋃i=1mΛηR(xi)T = \bigcup_{i=1}^{m} \Lambda_{\eta R}(x_i) for a sequence x1,…,xm∈v+2ηRZ2x_1,\ldots,x_m \in v + 2\eta R\mathbb{Z}^2 of distinct points with ∥xi−xi−1∥2=2ηR\lVert x_i-x_{i-1}\rVert_2 = 2\eta R for all i=2,…,mi=2,\ldots,m and 1≤m≤η−21 \le m \le\eta^{-2} and some fixed translation vector v∈R2v \in\mathbb{R}^2.

We may cut any tube of quality η\eta at scale RR lengthwise into two tubes T1,T2T_1,T_2 of quality η/2\eta/2 such that T1∪T2=TT_1 \cup T_2 = T and T1T_1 and T2T_2 have disjoint interiors. This is done as follows (see Figure 4 (right) for an example).

Figure 4. Left: A tube TT of quality η\eta at scale RR (in green) with the start and end marked by shaded squares. An exploration curve (in black) with the open side marked by a solid line and the closed side marked by a dashed line. Its open side is exposed in TT. The red crossing of the tube hits the open side of the exploration curve first. Right: The lengthwise cutting of TT into two tubes of quality η/2\eta/2. Each square ΛηR(⋅)\Lambda_{\eta R}(\cdot) is cut depending on the position of the previous and following box.

Denote by ζ\zeta the polygonal path obtained by joining each consecutive pair of points xix_i and xi+1x_{i+1} by a line segment for all i=1,…,m−1i=1,\ldots,m-1. Extend the first and last segments to the boundaries of the corresponding starting and ending boxes such that they are cut in half. Note that ζ\zeta naturally cuts each box ΛηR(xi)\Lambda_{\eta R}(x_i) of TT in two parts P1(xi)P_1(x_i) and P2(xi)P_2(x_i) that lie at its right and at its left, respectively. For each ii, P1(xi)P_1(x_i) and P2(xi)P_2(x_i) are each a union of one, two or three boxes ΛηR/2(z)\Lambda_{\eta R/2}(z) for z∈v+(ηR2,ηR2)+ηRZ2z \in v + (\frac{\eta R}{2},\frac{\eta R}{2}) + \eta R\mathbb{Z}^2. On each side of ζ\zeta, the subsquares can be ordered along the path to form a nearest-neighbour chain. Each chain contains at most 3m3m squares and 3m≤(η/2)−23m \le(\eta/2)^{-2}. Hence, we obtain tubes of quality η/2\eta/2 by taking T1=⋃P1(xi)T_1=\bigcup P_1(x_i) and T2=⋃P2(xi)T_2=\bigcup P_2(x_i).

An immediate consequence of this construction is that if a cell-path γ\gamma is exposed by TT in Exp⁡c\operatorname{Exp}^{c}, then γ\gamma is also exposed by T1T_1 and T2T_2 in Exp⁡c\operatorname{Exp}^{c}. Indeed, any crossing of T1T_1 is in particular a crossing of TT, and hence intersects γ\gamma; and the connected component of the start of T1T_1 in T1∖γT_1\setminus\gamma is contained in the connected component of the start of TT in T∖γT\setminus\gamma, which is unexplored by assumption. The same arguments apply to T2T_2.

Proof of Theorem 3 for favourable tilings

Fix a tiling G\mathcal{G} and a percolation measure P\mathbb{P}. Also, fix RR, HH and σ\sigma as in the statement of the theorem. We start by proving the result in the case where all cells of G\mathcal{G} have diameter at most 11 and where RR is assumed to be large enough. Below, we will define different tubes of quality η\eta at scale RR. Since RR and even ηR\eta R will be taken large, we may ignore cells crossing the boundary of the tubes. More precisely, we will suppose that Rect⁡(R,H)\operatorname{Rect}(R,H) and all tubes considered hereafter can be approximated by unions of cells. We will discuss these approximations and the case of more general tilings G\mathcal{G} at the end of the proof.

A quad is a simply connected domain QQ with four marked points aa, bb, cc, dd on its boundary in counterclockwise order. This defines four boundary arcs (a,b)(a,b), (b,c)(b,c), (c,d)(c,d) and (d,a)(d,a), which we also call the bottom, right, top and left of ∂Q\partial Q, respectively. Below, all quads will be unions of cells, so one can perform the interface exploration described in the previous section in these domains.

Proof of Theorem 3. Recall that NN is the length of the colour sequence σ\sigma. Since the scale RR at which we work is fixed, we will omit it from notation. In particular we henceforth write η\eta-exposed and tube of quality η\eta without referring to the scale RR.

Since the result is stated with no assumptions on the percolation measure, we may work under P[⋅∣Aσ(Rect⁡(R,H))]\mathbb{P}[\cdot\mid A_\sigma(\operatorname{Rect}(R,H))].

We will construct Exp⁡\operatorname{Exp} by progressively revealing interfaces up to different stopping times τ1≤⋯≤τN\tau_1\leq\cdots\leq\tau_N. At each stage, we may have τk=∞\tau_k=\infty, which means that the exploration procedure failed. The procedure is successful if τN<∞\tau_N<\infty and we set Exp⁡=e[τN]\operatorname{Exp}=e_{[\tau_N]}.

Set ηk=2−k−3\eta_k=2^{-k-3} for 0≤k≤N0\leq k\leq N. At the end of stage k≤Nk\leq N, if τk<∞\tau_k<\infty then the following conditions are satisfied.

  1. There exist indices ℓ\ell and rr with ℓ+N+1−r=k\ell+N+1-r=k such that the explored region contains cell-paths γ1,…,γℓ\gamma_1,\ldots,\gamma_\ell and γr,…,γN\gamma_r,\ldots,\gamma_N of colours σ1,…,σℓ\sigma_1,\ldots,\sigma_\ell and σr,…,σN\sigma_r,\ldots,\sigma_N, respectively, that are ηk\eta_k-exposed. Write T1(k),…,Tℓ(k),Tr(k),…,TN(k)T_1^{(k)},\ldots,T_\ell^{(k)},T_r^{(k)},\ldots,T_N^{(k)} for the tubes of quality ηk\eta_k exposing the paths γi\gamma_i; they will all be defined on a common translated grid.

If ℓ=0\ell= 0, we will also have a tube T0(k)T_0^{(k)} that is disjoint from the above and which exposes4 the segment {−R}×[0,H−4ηkR]\{-R\} \times[0, H - 4\eta_kR]. Similarly, when r=N+1r = N + 1, we have an additional tube TN+1(k)T_{N+1}^{(k)}, disjoint from the above, and which exposes the segment {R}×[0,H−4ηkR]\{R\} \times[0, H - 4\eta_kR].

(2) The explored part at this stage satisfies e[τk]⊂Rect⁡(R,H−4ηkR)e_{[\tau_k]} \subset\operatorname{Rect}(R, H - 4\eta_kR) and the unexplored part e[τk]c:=Rect⁡(R,H)∖e[τk]e_{[\tau_k]}^c := \operatorname{Rect}(R, H) \setminus e_{[\tau_k]} is a simply connected domain. Moreover, when k<Nk < N, the intersection of ∂e[τk]c\partial e_{[\tau_k]}^c with the bottom of Rect⁡(R,H)\operatorname{Rect}(R, H) is exactly the segment of the latter contained between the starting cells of γℓ\gamma_\ell and γr\gamma_r. We call this the bottom of e[τk]ce_{[\tau_k]}^c.

(3) There exist a.s. disjoint cell-paths from the bottom of e[τk]ce_{[\tau_k]}^c to the top of Rect⁡(R,H)\operatorname{Rect}(R, H), avoiding γℓ\gamma_\ell and γr\gamma_r, ordered from left to right, of colours σℓ+1,…,σr−1\sigma_{\ell+1}, \ldots, \sigma_{r-1}. These paths may use both unexplored and explored cells, but no cells of γℓ\gamma_\ell and γr\gamma_r.

See Figure 5 and Figure 6 for illustrations.

Figure 5. The green tube is T0(0)T_0^{(0)}; its start and end are shaded. This tube exposes the left side of the rectangle. On the right, a primal path has been explored. The explored region at this stage is hatched; the red tube is TN(1)T_N^{(1)}.

Figure 6. An example of an exploration (hatched) at stage kk assuming τk<∞\tau_k < \infty. The black paths are γℓ\gamma_\ell and γr\gamma_r; they are exposed through the tubes Tℓ(k)T_\ell^{(k)} and Tr(k)T_r^{(k)}, respectively. These tubes are cut lengthwise into Tℓ(k+1)T_\ell^{(k+1)} and Tr(k+1)T_r^{(k+1)} (green) and SLS_L and SRS_R. The quad Qk+1Q_{k+1} is depicted in yellow.

We will construct the exploration inductively and simultaneously check the points (1), (2) and (3) above. At the end, we will bound the probability that τN<∞\tau_N < \infty. As mentioned before, an external source of randomness will be used during the procedure. This will be encoded by a series of i.i.d. uniform variables X1,…,XNX_1, \ldots, X_N in {L,R}\{\mathrm{L}, \mathrm{R}\}.

Initial step: To start, set τ0=0\tau_0 = 0, e[τ0]=∅e_{[\tau_0]} = \varnothing and let

T0(0)=[−(1−η0)R,−(1−3η0)R]×[H−7η0R,H+3η0R]∪[−(1+3η0)R,−(1−η0)R]×[H−7η0R,H−5η0R]T_0^{(0)} = [-(1-\eta_0)R, -(1-3\eta_0)R] \times[H-7\eta_0R, H+3\eta_0R] \cup[-(1+3\eta_0)R, -(1-\eta_0)R] \times[H-7\eta_0R, H-5\eta_0R]

and

TN+1(0)=[(1−3η0)R,(1−η0)R]×[H−7η0R,H+3η0R]∪[(1−η0)R,(1+3η0)R]×[H−7η0R,H−5η0R],T_{N+1}^{(0)} = [(1-3\eta_0)R, (1-\eta_0)R] \times[H-7\eta_0R, H+3\eta_0R] \cup[(1-\eta_0)R, (1+3\eta_0)R] \times[H-7\eta_0R, H-5\eta_0R],

be LL-shaped tubes; see Figure 5 for an example. These two tubes expose the left and right side of the rectangle, respectively. They only serve to initialise the induction process. Set ℓ=0\ell= 0 and r=N+1r = N + 1. All points are clearly satisfied when k=0k = 0. For completeness, we set γ0={−R}×[0,H−4ηkR]\gamma_0 = \{-R\} \times[0, H - 4\eta_kR] and γN+1={R}×[0,H−4ηkR]\gamma_{N+1} = \{R\} \times[0, H - 4\eta_kR]. This is a slight abuse of notation since the other γi\gamma_i will refer to cell-paths.

Induction step: construction. Let 0≤k<N0 \le k < N and consider the exploration up to stage kk, assuming that τk<∞\tau_k < \infty. In particular, this means that conditions (1), (2) and (3) are satisfied at step kk. Proceed as follows for step k+1k+1 of the algorithm.

Cut the tubes Tℓ(k)T_\ell^{(k)} and Tr(k)T_r^{(k)} lengthwise. Write Tℓ(k+1)T_\ell^{(k+1)} for the left half of the former and SLS_L for its right half. Similarly, write Tr(k+1)T_r^{(k+1)} for the right half of the latter and SRS_R for its left half. For j<ℓj < \ell and j>rj > r, cut the tube Tj(k)T_j^{(k)} lengthwise and set Tj(k+1)T_j^{(k+1)} to be the left half if j<ℓj < \ell and the right half if j>rj > r.5 Observe that the tubes Tj(k+1)T_j^{(k+1)} for j≤ℓj \le\ell and j≥rj \ge r are indeed of quality ηk+1=ηk/2\eta_{k+1} = \eta_k/2.

We now define a quad Qk+1Q_{k+1}; see Figure 6 for an illustration. Consider the curve χL\chi_L obtained by following the left boundary of SLS_L from the top of Rect⁡(R,H)\operatorname{Rect}(R,H) up to the first time it reaches the boundary of a cell of γℓ\gamma_\ell — denote this point by dd. Then, continue extending χL\chi_L by following the boundary of γℓ\gamma_\ell down to the first time it reaches the bottom of Rect⁡(R,H)\operatorname{Rect}(R,H). Observe that the curve χL\chi_L connects top and bottom of Rect⁡(R,H)\operatorname{Rect}(R,H). We write aa for its endpoint on the bottom of Rect⁡(R,H)\operatorname{Rect}(R,H).

Let χR\chi_R be defined similarly, using the right boundary of SRS_R and γr\gamma_r instead of γℓ\gamma_\ell. Denote by cc the point where the boundary of SRS_R meets the boundary of a cell of γr\gamma_r for the first time and by bb the endpoint of χR\chi_R on the bottom of Rect⁡(R,H)\operatorname{Rect}(R,H). The fact that the boundaries of the tubes intersect γℓ\gamma_\ell and γr\gamma_r, respectively, is guaranteed by their respective exposition.

By point (2), the horizontal segment (a,b)(a,b) is exactly the bottom part of ∂C[τk]\partial\mathcal{C}_{[\tau_k]}. Let xx be the last point of χL\chi_L on [−R,R]×{H−4ηk+1R}[-R,R] \times\{H - 4\eta_{k+1}R\} when χL\chi_L is considered as a curve oriented towards aa. Similarly, let yy be the last intersection of χR\chi_R with this horizontal line. Write (x,y)(x,y) for the horizontal segment between xx and yy. Let Qk+1Q_{k+1} be the region delimited by the horizontal segment (a,b)(a,b), the sub-arc (b,y)(b,y) of χR\chi_R, the horizontal segment (y,x)(y,x) and the sub-arc (x,a)(x,a) of χL\chi_L; see Figure 6. By closely following the definitions of χL\chi_L and χR\chi_R, one may notice that the cells of γℓ\gamma_\ell and γr\gamma_r are outside of Qk+1Q_{k+1}; some of them being adjacent to the arcs (d,a)(d,a) and (b,c)(b,c) of the quad ∂Qk+1\partial Q_{k+1}.

Note that (x,y)(x,y) has length at least 2ηk+1R=ηkR2\eta_{k+1}R=\eta_kR. Write zz for its midpoint. The points a,b,c,da,b,c,d defined above lie on the boundary of Qk+1Q_{k+1} and are distributed in counterclockwise order. The arc (c,d)(c,d) of ∂Qk+1\partial Q_{k+1} contains the segment (y,x)(y,x).

As a consequence of point (3), the quad Qk+1Q_{k+1} is crossed from (a,b)(a,b) to (c,d)(c,d) by disjoint paths of colours σℓ+1,…,σr−1\sigma_{\ell+1},\ldots,\sigma_{r-1} ordered from left to right.

If Xk+1=LX_{k+1}=L, we set γℓ+1\gamma_{\ell+1} to be the leftmost6 cell-path in Qk+1Q_{k+1} of colour σℓ+1\sigma_{\ell+1} from (a,b)(a,b) to (c,d)(c,d). This path is revealed by using the interface exploration in the domain Qk+1Q_{k+1} with marked points a,c,da,c,d, started from aa and with σℓ+1\sigma_{\ell+1} on its right and 1−σℓ+11-\sigma_{\ell+1} on its left. More precisely, γℓ+1\gamma_{\ell+1} is a subset of the cells on the right side of this interface. See Figure 4 (left). We then say that step k+1k+1 is successful if the endpoint of γℓ+1\gamma_{\ell+1} is to the left of zz, i.e., on the arc (z,d)(z,d).

Similarly, if Xk+1=RX_{k+1}=R, let γr−1\gamma_{r-1} be the rightmost path in Qk+1Q_{k+1} of colour σr−1\sigma_{r-1} from (a,b)(a,b) to (c,d)(c,d). Then, we say that step k+1k+1 is successful if the endpoint of γr−1\gamma_{r-1} is to the right of zz, i.e., on the arc (c,z)(c,z).

At this stage, if the exploration is successful, further reveal the state of all unexplored cells that are disconnected from the top of Rect⁡(R,H)\operatorname{Rect}(R,H) by the explored part. Set τk+1\tau_{k+1} to be the end of this exploration step. If the exploration is not successful, reveal the state of all cells in arbitrary order and set τk+1=∞\tau_{k+1}=\infty.

Induction step: checking conditions. We now prove that, if the exploration is successful at step k+1k+1, then the points (1), (2) and (3) are satisfied at time τk+1\tau_{k+1}. Assume we are in the first case Xk+1=LX_{k+1}=L, where we explored γℓ+1\gamma_{\ell+1} from the left. The case Xk+1=RX_{k+1}=R is handled analogously, with the roles of left and right interchanged. Assuming that step k+1k+1 is successful, we have that the endpoint of γℓ+1\gamma_{\ell+1} is on the arc (z,d)(z,d).

We start with point (1). Recall the tubes Tj(k+1)T_j^{(k+1)} defined above for j≤ℓj\leq\ell and j≥rj\geq r. Their exposition zones are indeed disjoint and do not intersect Qk+1Q_{k+1}. As such, they are not affected by the exploration between τk\tau_k and τk+1\tau_{k+1}, and e[τk+1]ce_{[\tau_{k+1}]}^c exposes the corresponding paths γj\gamma_j for j≤ℓj\leq\ell and j≥rj\geq r.

Define the tube Tℓ+1(k+1)T_{\ell+1}^{(k+1)} as follows — red tube in Figure 7. Follow SRS_R from its start up to the first box contained in Qk+1∩[−R,R]×[H−4ηkR,H−4ηk+1R]Q_{k+1}\cap[-R,R]\times[H-4\eta_kR,H-4\eta_{k+1}R] — such an intersection necessarily occurs since SRS_R exposes γr\gamma_r in e[τk]ce_{[\tau_k]}^c, which is entirely below the level H−4ηkRH-4\eta_kR. Then use boxes contained in Qk+1∩[−R,R]×[H−4ηkR,H−4ηk+1R]Q_{k+1}\cap[-R,R]\times[H-4\eta_kR,H-4\eta_{k+1}R] to connect SRS_R to SLS_L. Finally, follow SLS_L to its end. By the fact that γℓ+1\gamma_{\ell+1} exits Qk+1Q_{k+1} on the arc (z,d)(z,d), any crossing of Tℓ+1(k+1)T_{\ell+1}^{(k+1)} necessarily intersects γℓ+1\gamma_{\ell+1}. The construction of Tℓ+1(k+1)T_{\ell+1}^{(k+1)} guarantees that the connected component of the start of Tℓ+1(k+1)T_{\ell+1}^{(k+1)} in Tℓ+1(k+1)∖γℓ+1T_{\ell+1}^{(k+1)}\setminus\gamma_{\ell+1} is entirely unexplored. Thus γℓ+1\gamma_{\ell+1} is exposed by Tℓ+1(k+1)T_{\ell+1}^{(k+1)} in e[τk+1]ce_{[\tau_{k+1}]}^c.

Figure 7. A case where Xk+1=LX_{k+1}=L and γℓ+1\gamma_{\ell+1} (in thick black) lands left of zz. The cells explored between τk\tau_k and τk+1\tau_{k+1} are all in Qk+1Q_{k+1}, left of γℓ+1\gamma_{\ell+1}. Then γℓ+1\gamma_{\ell+1} is exposed in the unexplored part e[τk+1]e^{[\tau_{k+1}]} by the red tube Tℓ+1(k+1)T_{\ell+1}^{(k+1)}.

Furthermore, the exposition zone of γℓ+1\gamma_{\ell+1} in Tℓ+1(k+1)T_{\ell+1}^{(k+1)} is disjoint from the exposition zones of γj\gamma_j in Tj(k+1)T_j^{(k+1)} with j≤ℓj\leq\ell and j≥rj\geq r. Finally, all Tj(k+1)T_j^{(k+1)} are of quality ηk+1=ηk/2\eta_{k+1}=\eta_k/2 by definition.

Point (2) is clear from the construction. Indeed, the exploration guarantees that e[τk+1]ce_{[\tau_{k+1}]}^c is simply connected. Furthermore, since Qk+1Q_{k+1} is contained below the line [−R,R]×{H−4ηk+1R}[-R,R]\times\{H-4\eta_{k+1}R\}, all explored cells are also contained in this part.

Finally, for point (3), we observe that, since γℓ+1\gamma_{\ell+1} was the leftmost path of colour σℓ+1\sigma_{\ell+1} crossing Qk+1Q_{k+1}, the other paths of colours σℓ+2,…,σr−1\sigma_{\ell+2},\ldots,\sigma_{r-1} are contained between γℓ+1\gamma_{\ell+1} and γr\gamma_r.

Conclusion: Set Exp⁡=e[τN]\operatorname{Exp}=e_{[\tau_N]}. Observe that, if τN<∞\tau_N<\infty, then Exp⁡\operatorname{Exp} indeed η\eta-exposes crossings of colours σ\sigma towards the top for η=ηN\eta=\eta_N.

Probability of success: We turn to estimating the probability of success of the process. Observe that if τk<∞\tau_k<\infty then, at step kk, for any possible configuration in the unexplored region, the leftmost crossing of Qk+1Q_{k+1} from (a,b)(a,b) to (c,d)(c,d) of colour σℓ+1\sigma_{\ell+1} lies to the left of the rightmost crossing of colour σr−1\sigma_{r-1}. Recall that the existence of these two crossings is a consequence of point (3). As such, at least one of the following holds:

  • the leftmost crossing of Qk+1Q_{k+1} from (a,b)(a,b) to (c,d)(c,d) of colour σℓ+1\sigma_{\ell+1} exits left of zz or

  • the rightmost crossing of Qk+1Q_{k+1} from (a,b)(a,b) to (c,d)(c,d) of colour σr−1\sigma_{r-1} exits right of zz.

We conclude that for any (undiscovered) configuration, τk+1<∞\tau_{k+1}<\infty for at least one of the cases Xk+1=LX_{k+1}=L or Xk+1=RX_{k+1}=R. Thus

P[τk+1<∞∣τk<∞,e[τk],ωe[τk],Aσ(Rect⁡(R,H))]≥12,\mathbb{P}\left[\tau_{k+1}<\infty\mid\tau_k<\infty, e^{[\tau_k]}, \omega_{e^{[\tau_k]}}, A_\sigma(\operatorname{Rect}(R,H))\right]\geq\frac{1}{2},

and by repeated applications of the above, we obtain the desired result:

P[τN<∞∣Aσ(Rect⁡(R,H))]≥(12)N>0.\mathbb{P}\left[\tau_N<\infty\mid A_\sigma(\operatorname{Rect}(R,H))\right]\geq\left(\frac{1}{2}\right)^N>0.

Proof of Theorem 3 for general tilings

We turn to the case of general tilings. Fix such a tiling G\mathcal{G} and a percolation measure P\mathbb{P} on G\mathcal{G}. Moreover, fix RR, HH and σ\sigma as in the statement of the theorem. Let η\eta be the quantity obtained in Section 3.2. We will prove the theorem for η~=η/4\widetilde{\eta}=\eta/4.

Applying the same algorithm as in Section 3.2, we obtain explorations and tubes of quality η\eta. Explored cells intersecting the tubes T1,…,TNT_1,\ldots,T_N that are used to expose certain paths γ1,…,γN\gamma_1,\ldots,\gamma_N may however overlap. As such, these tubes do not satisfy the requirements of the theorem. To accommodate this, we replace each tube by a narrower tube lying strictly inside it.

The hypothesis of the theorem requires that all cells of G\mathcal{G} intersecting Rect⁡(R,H)\operatorname{Rect}(R,H) have diameter at most η~R=ηR/4\widetilde{\eta}R=\eta R/4. In particular, this bound applies to every cell revealed by the exploration. As such, each tube TiT_i contains a tube T~i\widetilde{T}_i of quality η~\widetilde{\eta} at scale RR, whose start is contained in the start of TiT_i, whose end is contained in the end of TiT_i and such that the distance between T~i\widetilde{T}_i and TicT_i^c is at least η~R\widetilde{\eta}R.

The algorithm above, together with the paths γ1,…,γN\gamma_1,\ldots,\gamma_N and the tubes T~1,…,T~N\widetilde{T}_1,\ldots,\widetilde{T}_N satisfies the conclusion of the theorem.

Exposition in an annulus: Proof of Theorem 1

Fix all quantities as in the statement of Theorem 1. As in the proof of Theorem 3, we will assume that our measure is already conditioned on Aσ(r,R)A_\sigma(r,R), and thus that P[Aσ(r,R)]=1\mathbb{P}[A_\sigma(r,R)]=1. We will see later how to choose 0<η<180<\eta<\frac{1}{8} and c>0c>0, depending only on the length NN of the colour sequence. As stated in the theorem, we assume that all cells of G\mathcal{G} intersecting ΛR\Lambda_R are of diameter at most ηR\eta R.

Since σ\sigma is polychromatic, there exists at least one open/closed interface crossing the annulus Ann⁡(r,R)\operatorname{Ann}(r,R) from ∂Λr\partial\Lambda_r to ∂ΛR\partial\Lambda_R. Our algorithm will consist of two steps. First, we explore an interface starting from ∂Λr\partial\Lambda_r up to ∂Λ3R/4\partial\Lambda_{3R/4}. This will allow us to cut the annulus in order to reduce the problem to a quad as in Theorem 3. In the second step, we apply a slight variation of the algorithm of Theorem 3.

Interface exploration in annuli

We start by describing how to explore an interface from ∂Λr\partial\Lambda_r up to the first time it exits Λ3R/4\Lambda_{3R/4}. Note that, since Aσ(r,R)A_\sigma(r,R) is assumed to occur, at least NN cells intersect Λr\Lambda_r. After a small perturbation of Λr\Lambda_r, we may also assume that ∂Λr\partial\Lambda_r contains no triple point of G\mathcal{G}.

Fix an arbitrary ordering of the edges of G\mathcal{G} crossing ∂Λr\partial\Lambda_r and orient each of them outwards. Examine these edges in the chosen order, revealing their adjacent cells if they have not been revealed previously. Do so until an edge separating opposite colours is found. Starting from this edge, follow the interface, preserving the colours on either side of the interface, until it either exits Λ3R/4\Lambda_{3R/4} or returns to Λr\Lambda_r. In the former case, stop the exploration. In the latter case, consider the next unexplored edge crossing ∂Λr\partial\Lambda_r and proceed as before. Repeat the procedure until the exploration exits Λ3R/4\Lambda_{3R/4}.

Due to our assumption, at least one interface exists between Λr\Lambda_r and ∂ΛR\partial\Lambda_R. As such, the exploration above does eventually reach ∂Λ3R/4\partial\Lambda_{3R/4}, and stops. We denote by Γ\Gamma the set of all edges examined during this procedure, excluding those intersecting ∂Λr\partial\Lambda_r which separate cells of same colour. See Figure 8 for an illustration.

Figure 8. The explored edges, starting at the marked point on ∂Λr\partial\Lambda_r, with the ordering being given by the counterclockwise order. The exploration process stops when reaching ∂Λ3R/4\partial\Lambda_{3R/4}.

The purpose of the exploration above is to represent the arms of Aσ(r,R)A_\sigma(r,R) as crossings of a suitable quad to which we can adapt the argument of Theorem 3. The important property of the set of edges Γ\Gamma is that no arm of Aσ(r,R)A_\sigma(r,R) can cross Γ\Gamma since these edges are between cells of opposite colour. So one may view Γ\Gamma as a union of walls that cuts the annulus into disjoint connected components. There is a unique connected component of Ann⁡(r,R)∖Γ\operatorname{Ann}(r,R)\setminus\Gamma that contains ∂ΛR\partial\Lambda_R — we denote this component by DD. Since the arms producing Aσ(r,R)A_\sigma(r,R) may not cross Γ\Gamma, they all occur in DD. Note, however, that they may use revealed cells of DD on either side of Γ\Gamma.

At this stage DD is topologically still an annulus, and not a rectangle as desired. It does however have long stretches on its inner boundary where arms may not originate from: the part of the inner boundary formed by Γ\Gamma. We exploit this feature in the next section.

Remark 5. It may be tempting to actually explore the interface until it reaches ∂ΛR\partial\Lambda_R, rather than stopping at ∂Λ3R/4\partial\Lambda_{3R/4}. This would produce an unexplored outer region which is a topological rectangle. The problem with doing this is that the exploration may come arbitrarily close to ∂ΛR\partial\Lambda_R several times before finally reaching ∂ΛR\partial\Lambda_R. As such, the topological rectangle obtained above may degenerate, which would prevent the application of the algorithm of Theorem 3.

Producing a good quad

Recall that DD is the connected component of Ann⁡(r,R)∖Γ\operatorname{Ann}(r,R)\setminus\Gamma containing ∂ΛR\partial\Lambda_R and that all arms occur in this topological annulus. After rotating the picture, we may consider that Γ\Gamma ends on the top side of Λ3R/4\Lambda_{3R/4}. We now describe how to obtain a domain D′D' in which we may apply the algorithm of Theorem 3. Figure 9 provides a useful illustration.

Figure 9. A sketch of the procedure in Section 4.2. Left: The exploration of Γ\Gamma is depicted in blue and yellow. To be precise, Γ\Gamma is the edge-path between the blue and yellow sides. The horizontal lines ℓi\ell_i are dashed. A domain C1C_1 is depicted in grey. We assume here that J=2J=2 is good; the line ℓJ−1/2\ell_{J-1/2} is dotted and the points a,b,c,da,b,c,d are marked. The exposition zones contained in T0(0)T_0^{(0)} and TN+1(0)T_{N+1}^{(0)} are green. Right: The grey region is the simply connected domain D′D', with the marked points. We separated the two sides of ℓJ−1/2\ell_{J-1/2} for clarity. The target arc (c,d)(c,d) (which corresponds to the top of the quad) is depicted by a slightly bolder black line. Observe that the arcs (b,c)(b,c) and (d,a)(d,a) of ∂D′\partial D' correspond to the inside of CJC_J, which is assumed to be good, and hence no arm originates from these arcs. Depicted in deeper colours are five arms, one open (dark blue) and four closed (orange). They are allowed to use the revealed cells on the boundary of D′D'. The purple lines play the role of the horizontal levels H−4ηkRH-4\eta_kR of Figure 6 and Figure 7.

Consider N+2N+2 equally spaced points on the right side of ∂ΛR\partial\Lambda_R:

xj=(R,R2+jR4(N+1)),j=0,…,N+1.x_j=\left(R,\frac{R}{2}+\frac{jR}{4(N+1)}\right),\qquad j=0,\ldots,N+1.

Let ℓj\ell_j be the line segment starting from xjx_j and travelling horizontally to the left until it meets ∂D\partial D. We will actually approximate ℓj\ell_j by paths of edges of G\mathcal{G}; by taking η\eta sufficiently small, we may assume that such approximate disjoint lines exist.

For j=1,…,N+1j=1,\ldots,N+1, let CjC_j be the connected component of D∖(ℓj−1∪ℓj)D\setminus(\ell_{j-1}\cup\ell_j) that contains the part of ∂ΛR\partial\Lambda_R between xj−1x_{j-1} and xjx_j. The regions C1,…,CN+1C_1,\ldots,C_{N+1} are pairwise disjoint.

We now make a crucial observation. Fix a configuration such that Aσ(r,R)A_\sigma(r,R) occurs and choose NN arms realising the event. Each arm has a starting point on ∂Λr\partial\Lambda_r. Since there are N+1N+1 regions C1,…,CN+1C_1,\ldots,C_{N+1} and only NN starting points for the arms, at least one region CjC_j contains no starting point of any of the NN chosen arms. We say that CjC_j is good (for this configuration) if Aσ(r,R)A_\sigma(r,R) may be realised with no arms originating in CjC_j.

In the final algorithm, one of the regions CjC_j will be chosen uniformly at random. For now, we describe how to construct a quad in order to apply the algorithm of Theorem 3 for a fixed index JJ, under the assumption that CJC_J is good.

Let ℓJ−1/2\ell_{J-1/2} be the horizontal line segment starting from xJ−1/2=xJ−(0,R8(N+1))x_{J-1/2}=x_J-\left(0,\frac{R}{8(N+1)}\right) and running left until the first time it meets ∂D\partial D. As before, we use a discrete approximation using edges of G\mathcal{G}.

We now set D′=D∖ℓJ−1/2D'=D\setminus\ell_{J-1/2}; this is a simply connected domain. Define the four points (or rather prime ends to be more precise) a,b,c,da,b,c,d in counterclockwise order on ∂D′\partial D' as follows. The left endpoints of ℓJ−1\ell_{J-1} and ℓJ\ell_J are aa and bb respectively. The left endpoint ℓJ−1/2\ell_{J-1/2} corresponds to both cc and dd; cc is part of the boundary as seen from above, while dd is seen as part of the boundary of the lower part. We denote by (a,b)(a,b), (b,c)(b,c), (c,d)(c,d) and (d,a)(d,a) the four arcs of the boundary of D′D'. See Figure 9.

Let S=(a,b)∩∂ΛrS=(a,b)\cap\partial\Lambda_r be the subset of (a,b)(a,b) lying on the inner boundary ∂Λr\partial\Lambda_r. Observe that SS does not intersect CJC_J. Thus, in any configuration for which CJC_J is good, there exist arms of colours prescribed by σ\sigma starting from SS and ending on the arc (c,d)(c,d). To be precise, there exists a cyclic permutation σ′\sigma' of σ\sigma such that paths of colours σ′\sigma' cross D′D' between SS and (c,d)(c,d), ordered from left to right. We denote this event by Aσ′(D′)A_{\sigma'}(D'). The parts of (a,b)(a,b) created by the exploration of Γ\Gamma are not admissible as starting points for the arms but the arms may subsequently use cells adjacent to Γ\Gamma. The event Aσ′(D′)A_{\sigma'}(D') will serve as the starting point for the induction in the proof of Theorem 3.

Let us also define the tubes appearing in condition (1) of the induction for the initial step. Set hj=R2+jR4(N+1)h_j=\frac{R}{2}+\frac{jR}{4(N+1)},

T0(0)=[−(1+2η)R,(1+2η)R]×[hJ−1,hJ−1/2]T_0^{(0)}=[-(1+2\eta)R,(1+2\eta)R]\times[h_{J-1},h_{J-1/2}]

and

TN+1(0)=[−(1+2η)R,(1+2η)R]×[hJ−1/2,hJ],T_{N+1}^{(0)}=[-(1+2\eta)R,(1+2\eta)R]\times[h_{J-1/2},h_J],

and view them as tubes starting at their rightmost squares and ending at their leftmost squares. These are indeed tubes of quality η0=η0(N)=1/(8(N+1))\eta_0=\eta_0(N)=1/(8(N+1)). Observe that the tube T0(0)T_0^{(0)} exposes7 the boundary arc (da)(da) of the quad D′D' and the tube TN+1(0)T_{N+1}^{(0)} exposes the arc (bc)(bc). See again Figure 9.

Assume that we know the cyclic permutation σ′\sigma' for which Aσ′(D′)A_{\sigma'}(D') occurs. Then we may adapt the algorithm of Theorem 3 in the domain D′D' with the colour sequence σ′\sigma'. Indeed, at each induction step, we select the leftmost or rightmost crossing in D′D' among those starting on SS. The final additional detail needed is to define an equivalent for the horizontal strips [−R,R]×[H−4ηkR,H−4ηk+1R][-R,R]\times[H-4\eta_kR,H-4\eta_{k+1}R] used to switch between the left and right tube when constructing Tℓ+1(k+1)T_{\ell+1}^{(k+1)} or Tr−1(k+1)T_{r-1}^{(k+1)} in the induction step. In our case, these strips are defined as the intersections of D′D' with the annuli Ann⁡(R(1−η0/2k),R(1−η0/2k+1))\operatorname{Ann}(R(1-\eta_0/2^k),R(1-\eta_0/2^{k+1})). Thus, the adaptation of the algorithm, if successful, produces an exploration in D′D' and disjoint tubes of quality η(N)=η0/2N\eta(N)=\eta_0/2^N that expose paths of colours σ\sigma towards the outside.

Conclusion

In conclusion, the algorithm for Theorem 1 is the following.

  1. Explore Γ\Gamma starting from an arbitrary point on ∂Λr\partial\Lambda_r until it reaches ∂Λ3R/4\partial\Lambda_{3R/4}.

  2. Define the regions CjC_j and choose JJ uniformly in {1,…,N+1}\{1,\ldots,N+1\}. Also, choose a cyclic permutation σ′\sigma' of σ\sigma uniformly at random.

  3. Finally, attempt the algorithm of Theorem 3 in D′D' for arms of colours σ′\sigma'.

For any fixed configuration, there exists a chance of at least 1N(N+1)\frac{1}{N(N+1)} that step (2) correctly identifies a good CJC_J and that σ′\sigma' is such that Aσ′(D′)A_{\sigma'}(D') occurs. Thus, using the estimate of Theorem 3, we obtain a uniform lower bound on the probability of success of this algorithm. □

Application to FK-percolation

Theorems 1 and 3 expose arms without imposing any assumptions on the percolation measure. In this section, we show how exposition of arms can be combined with crossing estimates to obtain arm separation. We work with FK-percolation for concreteness. The argument relies on the spatial Markov property, positive association and RSW estimates (all of which will be recalled below). As explained in § 1.2, the notion of exploration and the result of Theorem 1 can readily be adapted to the case of bond percolation on Z2\mathbb{Z}^2. We shall use such an adaptation implicitly.

FK-percolation and the necessary tools

Let us start by briefly recalling the definition of FK-percolation and the properties that will be used below. For more detailed statements and proofs, see [19, 9, 26]. Let G=(V,E)G=(V,E) be a finite subgraph of Z2\mathbb{Z}^2 with vertex set VV and edge set EE. A percolation configuration ω∈{0,1}E\omega\in\{0,1\}^E is the spanning subgraph of GG obtained by retaining the edges e∈Ee\in E for which ωe=1\omega_e=1 and discarding the others. Retained edges are said to be open while the discarded edges are said to be closed. A boundary condition ξ\xi is a partition of the boundary vertices of GG; vertices belonging to the same partition element are identified.

For p∈(0,1)p\in(0,1) and q>0q>0, the FK-measure on GG with boundary condition ξ\xi is

ϕG,p,qξ[ω]=1ZG,p,qξpo(ω)(1−p)∣E∣−o(ω)qkξ(ω),\phi_{G,p,q}^{\xi}[\omega]=\frac{1}{Z_{G,p,q}^{\xi}}p^{o(\omega)}(1-p)^{|E|-o(\omega)}q^{k_{\xi}(\omega)},

where o(ω)o(\omega) is the number of open edges in ω\omega, kξ(ω)k_{\xi}(\omega) is the number of connected components after the identifications prescribed by ξ\xi, and ZG,p,qξZ_{G,p,q}^{\xi} is a normalising constant. The case q=1q=1 corresponds to Bernoulli bond percolation.

The dual configuration ω∗\omega^* is defined by declaring a dual edge open precisely when the corresponding primal edge is closed. In the terminology of the previous sections, primal open paths have colour 1 and dual open paths have colour 0. See also the discussion in § 1.2 and Figure 2.

Throughout this section, we fix q∈[1,4]q\in[1,4] and p=pc(q)=q1+qp=p_c(q)=\frac{\sqrt{q}}{1+\sqrt{q}}, the critical parameter [3]. We suppress these from notation. Also write ϕ\phi for the unique infinite-volume FK-measure on Z2\mathbb{Z}^2 [12].

We will use the following form of the RSW estimates deduced readily from [12]. Recall the notion of tube of quality η\eta at scale RR and that of crossing of such a tube.

Proposition 6. For any η>0\eta>0 there exists c>0c>0 such that, for any RR large enough, any tube TT of quality η\eta at scale RR and any boundary conditions ξ\xi on TT,

ϕTξ[T contains a primal open crossing]≥cand\phi_{T}^{\xi}[T\text{ contains a primal open crossing}] \ge c \quad\text{and}
ϕTξ[T contains a dual open crossing]≥c.(2)\phi_{T}^{\xi}[T\text{ contains a dual open crossing}] \ge c. \tag*{(2)}

Note that the crossings in the above are not required to touch the boundary of TT; they only need to connect the starting box with the ending box of TT. Thus, the inequalities above are indeed valid for all q∈[1,4]q\in[1,4].

When working with (RSW) estimates, a particularly convenient property of FK-percolation is positive association: for any increasing events AA and BB and any graph GG and boundary conditions ξ\xi

ϕGξ[A∩B]≥ϕGξ[A]ϕGξ[B].(3)\phi_{G}^{\xi}[A\cap B]\ge\phi_{G}^{\xi}[A]\phi_{G}^{\xi}[B]. \tag*{(3)}

Here, an event is increasing if opening additional edges preserves its occurrence. The same inequality holds for two decreasing events.

Finally, after an exploration, the spatial Markov property identifies the conditional law in the unexplored region as an FK-measure with the same parameters p,qp,q and with the boundary conditions induced by the revealed configuration.

Combining these properties allows us to construct crossings in the exposition zones and attach them to the exposed arms.

Flower domains and separated arms

We formulate separation of arms using flower domains, following [10]. A particularly convenient feature of flower domains is that they can be explored from one side while leaving the region on the other side unexplored.

For r<Rr<R, the inner flower domain explored from ΛR\Lambda_R to Λr\Lambda_r is obtained by exploring the interfaces of Ann⁡(r,R)\operatorname{Ann}(r,R) starting on ∂ΛR\partial\Lambda_R and retaining the unexplored domain containing the inner box Λr\Lambda_r. Similarly, the outer flower domain explored from Λr\Lambda_r to ΛR\Lambda_R is obtained by exploring the interfaces of Ann⁡(r,R)\operatorname{Ann}(r,R) starting on ∂Λr\partial\Lambda_r and retaining the domain containing the exterior of ΛR\Lambda_R. See Figure 10 (left) for an illustration.

Figure 10. Left: The outer flower domain between scales ΛR/2\Lambda_{R/2} and ΛR\Lambda_R is η\eta-well-separated. This flower domain is revealed by exploring part of the hatched region. Some of the interfaces revealed during the process are drawn. Right: A realisation of the event A0,η(r,R)A_{0,\eta}(r,R). This event is revealed by exploring the outer flower domain between ΛR/2\Lambda_{R/2} and ΛR\Lambda_R and the entire configuration in Ann⁡(r,R/2)\operatorname{Ann}(r,R/2). This correspond to the hatched region. In particular, the arms can be extended in the unexplored region by using the properties of FK-percolation.

The boundaries of these domains consist of alternating primal and dual arcs, called petals, or of a single primal or dual circuit. For the inner flower domain, the endpoints of petals lie on ∂Λr\partial\Lambda_r and for the outer flower domain, they lie on ∂ΛR\partial\Lambda_R. An inner flower domain on Λr\Lambda_r is η\eta-well-separated if it contains at least two petals and any two distinct petal endpoints are at distance greater than ηr\eta r. For an outer flower domain on ΛR\Lambda_R to be η\eta-well-separated, the corresponding requirement is that the endpoints lie at a distance greater than ηR\eta R.

Definition 7 (Separated arm events). Fix σ∈{0,1}N\sigma\in\{0,1\}^{\mathbb{N}} and R≥4rR \ge4r. Let Fin\mathcal{F}_{\mathrm{in}} be the inner flower domain explored from Λ2r\Lambda_{2r} to Λr\Lambda_r, and let Fout\mathcal{F}_{\mathrm{out}} be the outer flower domain explored from ΛR/2\Lambda_{R/2} to ΛR\Lambda_R. Let Aση,η′(r,R)A_{\sigma}^{\eta,\eta'}(r,R) be the event that Fin\mathcal{F}_{\mathrm{in}} and Fout\mathcal{F}_{\mathrm{out}} are η\eta- and η′\eta'-well-separated, respectively, and that Aσ(r,R)A_\sigma(r,R) can be realised by arms connecting distinct petals of Fin\mathcal{F}_{\mathrm{in}} to distinct petals of Fout\mathcal{F}_{\mathrm{out}}, with each arm attached to petals of its own colour.

We write Aση,0A_{\sigma}^{\eta,0} when only the inner separation is imposed and Aσ0,η′A_{\sigma}^{0,\eta'} when only the outer separation is imposed. See Figure 10 (right).

Requiring distinct arms to use distinct petals also separates consecutive arms of the same colour; unused petals may lie between them. The well-separated petals provide enough room to continue the arms through additional crossings. The following theorem shows that imposing this separation costs only a uniformly positive factor, even when the configuration away from the target boundary is fixed.

Theorem 8 (Arm separation). Fix a polychromatic sequence σ∈{0,1}N\sigma\in\{0,1\}^{\mathbb{N}}. For sufficiently small η,η′>0\eta,\eta' > 0, there exists c=c(q,σ,η,η′)>0c=c(q,\sigma,\eta,\eta')>0 such that the following holds for all R≥4rR \ge4r:

  • separation towards the inside:

ϕ[Aση,0(r,R)∣Aσ(r,R),ωout]≥c;\phi[A_{\sigma}^{\eta,0}(r,R)\mid A_{\sigma}(r,R),\omega_{\mathrm{out}}] \ge c;
  • separation towards the outside:

ϕ[Aσ0,η′(r,R)∣Aσ(r,R),ωin]≥c;\phi[A_{\sigma}^{0,\eta'}(r,R)\mid A_{\sigma}(r,R),\omega_{\mathrm{in}}] \ge c;
  • separation in both directions:

ϕ[Aση,η′(r,R)∣Aσ(r,R),ωmid]≥c.\phi[A_{\sigma}^{\eta,\eta'}(r,R)\mid A_{\sigma}(r,R),\omega_{\mathrm{mid}}] \ge c.

Here, ωout,ωin\omega_{\mathrm{out}},\omega_{\mathrm{in}} and ωmid\omega_{\mathrm{mid}} denote fixed configurations on the respective annuli Ann⁡(2r,R)\operatorname{Ann}(2r,R), Ann⁡(r,R/2)\operatorname{Ann}(r,R/2) and Ann⁡(2r,R/2)\operatorname{Ann}(2r,R/2). Each inequality holds for every such conditioning of positive probability.

To experts, the above will appear as a direct consequence of Theorem 1 and the properties of FK-percolation, particularly (RSW). We prefer stating separation for FK-percolation in terms of flower domains, as these are explored using a standard algorithm, rather than the bespoke algorithm appearing in Theorem 1. Furthermore, flower domains and their exploration already appeared in [10 GM25] and seem particularly adapted to surgeries for FK-percolation and similar models.

Remark 9. In certain contexts, it may be useful to consider half-plane arm events, i.e., arm events where the arms are restricted to lie in a given half-plane. The notion of flower domain adapts naturally to this context, so one can prove an analogue of Theorem 8 for arm events in the half-plane by using the same method. This construction uses Theorem 3 directly, rather than Theorem 1. Thus, in the case of half-plane arm events, one is not restricted to polychromatic arm events.

From exposition to separation

We now explain how exposed arms can be extended to well-separated arms with uniformly positive probability. Several variants of this construction will be used below, depending on the direction of exposition and the configurations already revealed. For clarity, we state the outward version with the configuration in the inner part of the annulus revealed. The particular radii chosen below only serve to provide a macroscopic buffer between the revealed configuration and the regions used to construct the connections.

Lemma 10 (Extending exposed arms). Fix a sequence σ∈{0,1}N\sigma\in\{0,1\}^{N} and η>0\eta>0 sufficiently small. Then there exists c=c(σ,η)>0c=c(\sigma,\eta)>0 such that the following holds for all R≥4rR\ge4r. Fix a configuration ωin\omega_{\mathrm{in}} on $\operatorname{Ann}(r,R/2) and consider the FK-measure conditioned on ωin\omega_{\mathrm{in}}. Suppose that an exploration Exp⁡\operatorname{Exp} η\eta-exposes paths of colours $\sigma towards the outside by only revealing edges in Ann⁡(r,5R/8)\operatorname{Ann}(r,5R/8). Write $\omega_{\mathrm{exp}} for the revealed configuration. Then,

ϕ[Aσ0,η(r,R)∣ωin,ωexp]≥c,\phi\left[A_{\sigma}^{0,\eta}(r,R)\mid\omega_{\mathrm{in}},\omega_{\mathrm{exp}}\right]\ge c,

as long as the conditioning is non-degenerate.

Proof sketch. The lemma follows by a standard (RSW) construction which we sketch here. Fix a realisation of the configuration ωin\omega_{\mathrm{in}}, and a realisation Exp⁡\operatorname{Exp} of the exploration; we write ωexp\omega_{\mathrm{exp}} for the explored configuration and Exp⁡\operatorname{Exp} for the explored set. Write γ1,…,γN\gamma_1,\ldots,\gamma_N for the arms exposed by the exploration.

First, we explore an η\eta-well-separated double 2N2N-petal flower domain between Λ3R/4\Lambda_{3R/4} and ΛR\Lambda_R. This consists of the inner flower domain Fin\mathcal{F}_{\mathrm{in}} explored from Λ7R/8\Lambda_{7R/8} to Λ3R/4\Lambda_{3R/4} and the outer flower domain Fout\mathcal{F}_{\mathrm{out}} explored from Λ7R/8\Lambda_{7R/8} to ΛR\Lambda_R. We say that they form an η\eta-well-separated double flower domain with 2N2N petals if each has 2N2N petals, with the corresponding petals connected in Finc∩Foutc\mathcal{F}_{\mathrm{in}}^{c}\cap\mathcal{F}_{\mathrm{out}}^{c} by disjoint paths of the appropriate colours, and both are η\eta-well-separated. See Figure 11 for an example. It is a direct generalisation of [10 Lemma 3.4] that, for any NN, there exist η>0\eta>0 and c>0c>0 such that

Figure 11. A double 2N2N-petal flower domain between Λ3R/4\Lambda_{3R/4} and ΛR\Lambda_R for N=3N=3. This domain is revealed by exploring the corresponding inner and outer flower domains Fin\mathcal{F}_{\mathrm{in}} and Fout\mathcal{F}_{\mathrm{out}}; the resulting explored region is hatched. Note that the petals of Fout\mathcal{F}_{\mathrm{out}} and Fin\mathcal{F}_{\mathrm{in}} are connected as required by the definition.

ϕ[(Fin,Fout) is an η-well-separated double 2N-petal flower domain∣ωin,ωexp]≥c,\phi\left[(\mathcal{F}_{\mathrm{in}},\mathcal{F}_{\mathrm{out}})\text{ is an }\eta\text{-well-separated double }2N\text{-petal flower domain}\mid\omega_{\mathrm{in}},\omega_{\mathrm{exp}}\right]\ge c,

for any configuration ωin,ωexp\omega_{\mathrm{in}},\omega_{\mathrm{exp}} as above.

Assume that the exploration of (Fin,Fout)(\mathcal{F}_{\mathrm{in}},\mathcal{F}_{\mathrm{out}}) did produce an η\eta-well-separated double 2N2N-petal flower domain. Select NN distinct petals P1,…,PNP_1,\ldots,P_N of Fin\mathcal{F}_{\mathrm{in}} whose colours, in cyclic order, agree with σ\sigma.

By the exposition condition, there exist disjoint tubes T1,…,TNT_1,\ldots,T_N of quality η′\eta' at scale RR such that, for each kk, any crossing of TkT_k induces a connection in Fin∩Exp⁡c\mathcal{F}_{\mathrm{in}}\cap\operatorname{Exp}^{c} between γk\gamma_k and PkP_k. Here η′>0\eta'>0 is a constant depending only on η\eta and NN.

Assume now that kk is such that σk=1\sigma_k=1. Conditionally on ωin,ωexp\omega_{\mathrm{in}},\omega_{\mathrm{exp}}, the exploration producing Fin\mathcal{F}_{\mathrm{in}}, and any configuration outside of TkT_k, the measure inside Tk∩Fin∩Exp⁡cT_k\cap\mathcal{F}_{\mathrm{in}}\cap\operatorname{Exp}^{c} dominates ϕTk0\phi^0_{T_k}. Indeed, the only edges on the boundary of this domain that are contained in TkT_k are those of γk\gamma_k or PkP_k, which are open. As such, we find

ϕ[each γk connects to Pk by a σk-type path in Fin∩Exp⁡c∣ωin,ωexp,Fin,Fout]≥c0N\phi\left[\text{each }\gamma_k\text{ connects to }P_k\text{ by a }\sigma_k\text{-type path in }\mathcal{F}_{\mathrm{in}}\cap\operatorname{Exp}^{c}\mid\omega_{\mathrm{in}},\omega_{\mathrm{exp}},\mathcal{F}_{\mathrm{in}},\mathcal{F}_{\mathrm{out}}\right]\ge c_0^N

with c0c_0 given by (RSW). Combining the two inequalities above yields

ϕ[Aσ0,η(r,R)∣ωin,ωexp]≥c⋅c0N\phi\left[A_\sigma^{0,\eta}(r,R)\mid\omega_{\mathrm{in}},\omega_{\mathrm{exp}}\right]\ge c\cdot c_0^N

which concludes the proof. □\square

Proof of Theorem 8. We first prove separation towards the outside; separation towards the inside can be treated similarly. Fix a configuration ωin\omega_{\mathrm{in}} in Ann⁡(r,R/2)\operatorname{Ann}(r,R/2) and assume that ϕ[Aσ(r,R)∣ωin]>0\phi[A_\sigma(r,R)\mid\omega_{\mathrm{in}}]>0. Apply Theorem 1 in the annulus Ann⁡(r,5R/8)\operatorname{Ann}(r,5R/8) to the arm event corresponding to σ\sigma. This yields an exploration Exp⁡\operatorname{Exp} that reveals no edges outside Λ5R/8\Lambda_{5R/8} and, on success, exposes paths towards the outside of Ann⁡(r,5R/8)\operatorname{Ann}(r,5R/8) with some quality η>0\eta>0. Write ωexp\omega_{\mathrm{exp}} for the configuration revealed by the algorithm. Since the exposition occurs with positive probability uniformly over the conditioning on ωin\omega_{\mathrm{in}}, we have

ϕ[Exp⁡ η-exposes paths of colours σ∣Aσ(r,R),ωin]≥cexp,\phi\left[\operatorname{Exp}\ \eta\text{-exposes paths of colours }\sigma\mid A_\sigma(r,R),\omega_{\mathrm{in}}\right]\ge c_{\mathrm{exp}},

with cexp=cexp(η,N)>0c_{\mathrm{exp}}=c_{\mathrm{exp}}(\eta,N)>0 being the same constant as in Theorem 1.

Let us write E\mathcal{E} for the event that Exp⁡\operatorname{Exp} η\eta-exposes paths of colours σ\sigma. We now work under the measure ϕ[⋅∣E,ωin]\phi[\cdot\mid\mathcal{E},\omega_{\mathrm{in}}], without conditioning on the arm event. Lemma 10 gives

ϕ[Aσ0,η(r,R)∣E,ωexp,ωin]≥cext\phi\left[A_\sigma^{0,\eta}(r,R)\mid\mathcal{E},\omega_{\mathrm{exp}},\omega_{\mathrm{in}}\right]\ge c_{\mathrm{ext}}

for any possible realisation of the explored configuration ωexp\omega_{\mathrm{exp}}. Averaging over successful explorations, it follows from the previous inequalities that

ϕ[Aσ0,η(r,R)∣ωin]≥cext ϕ[E∣ωin]≥cextcexp ϕ[Aσ(r,R)∣ωin].\phi\left[A_\sigma^{0,\eta}(r,R)\mid\omega_{\mathrm{in}}\right]\ge c_{\mathrm{ext}}\,\phi[\mathcal{E}\mid\omega_{\mathrm{in}}]\ge c_{\mathrm{ext}}c_{\mathrm{exp}}\,\phi[A_\sigma(r,R)\mid\omega_{\mathrm{in}}].

Internal separation follows by the same argument using exposition and extension towards the inside instead.

We finally treat the case of separation in both directions. There is a slight subtlety here in that the outward exploration must preserve the distinct attachments to the inner flower domain obtained by the inward separation or vice versa. The following argument shows how the exploration algorithm can be adapted to deal with this issue.

Fix a configuration ωmid\omega_{\mathrm{mid}} in Ann⁡(2r,R/2)\operatorname{Ann}(2r,R/2). Averaging the separation estimate towards the inside over configurations outside Λ2r\Lambda_{2r} gives

ϕ[Aση,0(r,R)∣ωmid]≥cin ϕ[Aσ(r,R)∣ωmid].\phi\left[A_\sigma^{\eta,0}(r,R)\mid\omega_{\mathrm{mid}}\right]\ge c_{\mathrm{in}}\,\phi[A_\sigma(r,R)\mid\omega_{\mathrm{mid}}].

Fix now a configuration ω~\widetilde{\omega} in Ann⁡(r,R/2)\operatorname{Ann}(r,R/2) such that ϕ[ω~∣Aση,0(r,R),ωmid]>0\phi[\widetilde{\omega}\mid A_\sigma^{\eta,0}(r,R),\omega_{\mathrm{mid}}]>0. We now wish to apply the algorithm of Theorem 1 for the measure ϕ[⋅∣Aση,0(r,R),ω~]\phi[\cdot\mid A_\sigma^{\eta,0}(r,R),\widetilde{\omega}]. While this measure is conditioned on having arms separated towards the inside, it may still be possible to find a family of arms that are not separated towards the inside. Thus, we have no guarantee that the algorithm of Theorem 1 exposes toward the outside a family of arms that are already separated toward the inside.

To overcome this small difficulty, we adapt the algorithm of Theorem 1 to only search for arms originating from distinct petals of Fin\mathcal{F}_{\mathrm{in}}. More precisely, change the definitions of γℓ+1\gamma_{\ell+1} and γr−1\gamma_{r-1} by taking γℓ+1\gamma_{\ell+1} (resp. γr−1\gamma_{r-1}) to be the leftmost (resp. rightmost) path crossing the quad Qk+1Q_{k+1} among those starting from an unused petal of Fin\mathcal{F}_{\mathrm{in}}.

Using this adapted algorithm and the same argument as for separation towards the outside gives

ϕ[Aση,η′(r,R)∣Aση,0(r,R),ω~,ωmid]≥cout.\phi\left[A_\sigma^{\eta,\eta'}(r,R)\mid A_\sigma^{\eta,0}(r,R),\widetilde{\omega},\omega_{\mathrm{mid}}\right]\ge c_{\mathrm{out}}.

Averaging over ω~\widetilde{\omega} and combining the two inequalities gives

ϕ[Aση,η′(r,R)∣ωmid]≥coutcin ϕ[Aσ(r,R)∣ωmid],\phi\left[A_\sigma^{\eta,\eta'}(r,R)\mid\omega_{\mathrm{mid}}\right]\ge c_{\mathrm{out}}c_{\mathrm{in}}\,\phi[A_\sigma(r,R)\mid\omega_{\mathrm{mid}}],

which concludes the proof. □\square

Statement on the use of AI

This work was entirely developed without the use of AI. AI was then used in proofreading selected passages, which were then rewritten manually.

Acknowledgements

This work was financially supported by the Swiss National Science Foundation (SNSF). The third author further acknowledges support from the swissuniversities “cotutelle de thèse” grant.

References

University of Fribourg, Switzerland Email address: [email protected]

University of Fribourg, Switzerland Email address: [email protected]

University of Fribourg, Switzerland and University of Innsbruck, Austria Email address: [email protected]

References

  1. [1]Daniel Ahlberg, Simon Griffiths, Robert Morris, and Vincent Tassion. Quenched Voronoi percolation. Adv. Math., 286:889–911, 2016.
  2. [2]Daniel Ahlberg, Vincent Tassion, and Augusto Teixeira. Sharpness of the phase transition for continuum percolation in R². Probab. Theory Related Fields, 172(1-2):525–581, 2018.
  3. [3]Vincent Beffara and Hugo Duminil-Copin. The self-dual point of the two-dimensional random-cluster model is critical for q ≥ 1. Probab. Theory Related Fields, 153(3-4):511–542, 2012.
  4. [4]Vincent Beffara and Hugo Duminil-Copin. Planar percolation with a glimpse of Schramm-Loewner evolution. Probab. Surv., 10:1–50, 2013.arxiv.org/abs/1107.0158
  5. [5]Vincent Beffara and Pierre Nolin. On monochromatic arm exponents for 2D critical percolation. Ann. Probab., 39(4):1286–1304, 2011.DOI
  6. [7]Dmitry Chelkak, Hugo Duminil-Copin, and Clément Hongler. Crossing probabilities in topological rectangles for the critical planar FK-Ising model. Electron. J. Probab., 21:Paper No. 5, 28, 2016.arxiv.org/abs/1312.7785
  7. [8]Federico Camia and Charles M. Newman. Two-dimensional critical percolation: the full scaling limit. Comm. Math. Phys., 268(1):1–38, 2006.
  8. [9]Hugo Duminil-Copin. Lectures on the Ising and Potts models on the hypercubic lattice. In Random Graphs, Phase Transitions, and the Gaussian Free Field, pages 35–161. Springer International Publishing, 2020.
  9. [10]Hugo Duminil-Copin and Ioan Manolescu. Planar random-cluster model: scaling relations. Forum of Mathematics, Pi, 10:e23, 2022.
  10. [11]Hugo Duminil-Copin, Ioan Manolescu, and Vincent Tassion. Planar random-cluster model: fractal properties of the critical phase. Probab. Theory Related Fields, 181(1-3):401–449, 2021.
  11. [12]Hugo Duminil-Copin, Vladas Sidoravicius, and Vincent Tassion. Continuity of the phase transition for planar random-cluster and Potts models with 1 ≤ q ≤ 4. Comm. Math. Phys., 349(1):47–107, 2017.
  12. [13]Hang Du, Yifan Gao, Xinyi Li, and Zijie Zhuang. Sharp asymptotics for arm probabilities in critical planar percolation. Comm. Math. Phys., 405(8):Paper No. 182, 51, 2024.arxiv.org/abs/2205.15901
  13. [14]Michael Damron and Artëm Sapozhnikov. Outlets of 2D invasion percolation and multiple-armed incipient infinite clusters. Probab. Theory Related Fields, 150(1-2):257–294, 2011.DOI
  14. [16]Christophe Garban, Gábor Pete, and Oded Schramm. The Fourier spectrum of critical percolation. Acta Math., 205(1):19–104, 2010.
  15. [17]Christophe Garban, Gábor Pete, and Oded Schramm. Pivotal, cluster, and interface measures for critical planar percolation. J. Amer. Math. Soc., 26(4):939–1024, 2013.arxiv.org/abs/1008.1378
  16. [18]Christophe Garban, Gábor Pete, and Oded Schramm. The scaling limits of near-critical and dynamical percolation. J. Eur. Math. Soc. (JEMS), 20(5):1195–1268, 2018.arxiv.org/abs/1305.5526
  17. [19]Geoffrey Grimmett. The random-cluster model, volume 333 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer-Verlag, Berlin, 2006.
  18. [21]Harry Kesten. The incipient infinite cluster in two-dimensional percolation. Probab. Theory Related Fields, 73(3):369–394, 1986.DOI
  19. [22]Harry Kesten. A scaling relation at criticality for 2D-percolation. In Percolation theory and ergodic theory of infinite particle systems (Minneapolis, Minn., 1984–1985), volume 8 of IMA Vol. Math. Appl., pages 203–212. Springer, New York, 1987.DOI
  20. [23]Harry Kesten. Scaling relations for 2D-percolation. Comm. Math. Phys., 109(1):109–156, 1987.DOI
  21. [25]Mikhail Khristoforov, Mikhail Skopenkov, and Stanislav Smirnov. A generalization of Cardy’s and Schramm’s formulae. Comm. Math. Phys., 406(5):Paper No. 105, 17, 2025.DOI
  22. [26]Ioan Manolescu. Exploring the phase transition of planar FK-percolation. Preprint arXiv:2502.08394, 2025.arxiv.org/abs/2502.08394
  23. [27]Pierre Nolin. Near-critical percolation in two dimensions. Electron. J. Probab., 13:no. 55, 1562–1623, 2008.
  24. [28]Ritvik Ramanan Radhakrishnan and Vincent Tassion. Strict inequalities for arm exponents in planar percolation. Probability Theory and Related Fields, 195:659–688, 2026.
  25. [29]Stanislav Smirnov. Critical percolation in the plane: conformal invariance, Cardy’s formula, scaling limits. C. R. Acad. Sci. Paris Sér. I Math., 333(3):239–244, 2001.DOI
  26. [30]Oded Schramm and Jeffrey E. Steif. Quantitative noise sensitivity and exceptional times for percolation. Annals of Mathematics, 171(2):619–672, 2010.
  27. [31]Stanislav Smirnov and Wendelin Werner. Critical exponents for two-dimensional percolation. Math. Res. Lett., 8(5-6):729–744, 2001.arxiv.org/abs/math/0109120
  28. [32]Vincent Tassion. Crossing probabilities for Voronoi percolation. Ann. Probab., 44(5):3385–3398, 2016.arxiv.org/abs/1410.6773

Paper details

vlnc:W_qhcfbjr2
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 11

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

AI

View statement
Looking for related papers…
Contents