Isoperimetry in the infinite cluster Julian Gold UCLA May 9th Frontier Probability Days 2016 Outline I Percolation I Isoperimetric problems in the discrete setting I A conjecture of Benjamini I Attack via “shape theorem” in d = 2 I The Wulff construction in lattice models I d ≥3 Percolation: setup Bernoulli (bond) percolation: I Zd with standard graph structure, d ≥ 2 I p ∈ [0, 1] I (Xe )e∈E (Zd ) iid Bernoulli(p) I e is open if Xe = 1 and is closed otherwise Zd with only open edges: I random subgraph of the original lattice I connected components are open clusters Percolation: the infinite cluster Notation: I C(0) is the open cluster containing 0 ∈ Zd I θp (d) := Pp (|C(0)| = ∞) d ≥ 2: There is a critical probability pc (d) ∈ (0, 1) so that I p > pc (d) implies θp (d) > 0 I p < pc (d) implies θp (d) = 0 “supercritical” p > pc (d): ∃! infinite (open) cluster Pp -a.s., denoted C∞ . Goal: explore the geometry of C∞ Figure: Percolation on Z2 with p = .511 1 source: wikipedia. The Cheeger constant: bottlenecks and robustness Definition Let G = (V , E ) be a finite, connected graph. The Cheeger constant of G is |∂H| : H ⊂ G , 0 < |H| ≤ |G |/2 ΦG := min |H| where ∂H is the edge boundary of H in G . The Cheeger constant of the giant component Notation: I Cn := C∞ ∩ [−n, n]d I C̃n is the largest connected component of Cn Thanks to the work of I Benjamini-Mossel ’03 I Mathieu-Remy ’04 I Rau ’07 I Berger-Biskup-Hoffman-Kozma ’07 I Pete ’08 ...it’s known that ΦC̃n n−1 A conjecture Conjecture (Benjamini) The limit lim nΦC̃n n→∞ exists Pp -almost surely. Progress Theorem (Procaccia, Rosenthal ’11) Let d ≥ 2, and let p > pc (d). var nΦC̃n ≤ n2−d And in 2012... ...Benjamini’s conjecture was settled in d = 2 for a modification of the Cheeger constant. A natural modification of the Cheeger constant Definition In d = 2, the modified Cheeger constant of C̃n is ω |∂ H| Φ̃n := min : H ⊂ C̃n , 0 < |H| ≤ |C̃n |/2 |H| where ∂ ω H is the open edge boundary of H in C∞ . Figure: We treat subgraphs of C̃n as living in C∞ . Asymptotics in d = 2 through a shape theorem Gn := |∂ ω H| H ⊂ C̃n : = Φ̃n |H| Theorem (Biskup, Louidor, Procaccia, Rosenthal ’12) Let d = 2, and let p > pc (2). There is Wp ⊂ [−1, 1]2 convex so that Pp -a.s., max inf dH n−1 Gn , Wp + x −−−→ 0 Gn ∈Gn x∈R2 n→∞ Here dH is the `∞ -Hausdorff distance on compact sets. Asymptotics in d = 2 through a shape theorem Theorem (Biskup, Louidor, Procaccia, Rosenthal ’12) Let d = 2, and let p > pc (d). There is a norm βp on R2 so that Pp -a.s., lim nΦ̃n = n→∞ lengthβp (∂Wp ) θp (2)L2 (Wp ) where Wp is the limit shape from the previous theorem. I L2 is two-dimensional Lebesgue measure I L2 (Wp ) = 22 /2 Scaling limit I Wp is called the Wulff shape I Wp is the optimizer of minimize: lengthβp (E ) L2 (E ) subject to: L2 (E ) ≤ 22 /2 Discrete, random isoperimetric problems scale to a continuous, deterministic isoperimetric problem. The Wulff construction in isoperimetric problems Given a norm τ on R2 , there is a standard way (due to Wulff) to construct a solution to minimize: I lengthτ (E ) L2 (E ) subject to: L2 (E ) ≤ (const.) Define: cτ := W \ {x ∈ R2 : n · x ≤ τ (n)} n∈S 1 I cτ so that L2 (Wτ ) = (const.) Let Wτ be a dilate of W I Wτ is the solution, unique up to translations2 2 This works in higher dimensions (see work of Taylor from the ’70s). The Wulff construction in isoperimetric problems βp gives Wp Q: How do we extract a norm from our lattice model? A: There is an established theory (of analysis of equilibrium crystal shapes), due to work of I Dobrushin-Kotecký-Schlosman ’90 I Alexander-Chayes-Chayes ’90 I Ioffe-Schonmann ’98 I Cerf, Cerf-Pisztora ’98, ’00 I Bodineau ’99 I + more The Wulff construction in lattice models Setup: we have a material with two distinct “phases” I open versus closed in percolation I + versus - in Ising Goal: study the shape of a “droplet” of one phase immersed within another Postulate: “surface energy” between droplet phase and ambient phase should be minimized I surface energy gives rise to a norm by examining direction dependence The Wulff construction in lattice models In our context: Gn ∈ Gn is a droplet in C∞ \ Gn . Q: What kind of surface energy is being minimized at the boundary of this droplet? The Wulff construction in lattice models In a given direction, the surface energy is the “cost” of forming an interface between droplet and ambient region At the microscopic scale, this minimal cost is a random variable depending on the orientation and size of the box. Extracting a surface energy Consider a square Sn (v ) in R2 of side-length 2n, oriented so that its “top” and “bottom” faces are orthogonal to v ∈ S 1 . Xn (v ) := the (random) minimum size of a cutset separating the top and bottom faces of Sn (v ) Subadditivity: Ep Xn (v ) n→∞ n βp (v ) := lim This defines βp as a function on S 1 . Extend to a function on R2 . βp is a norm βp is well-studied (Kesten, Zhang, Cerf, Théret, Garet), and it is known that βp is a norm on R2 . I am lying: I In [BLPR], βp is defined using paths, not cutsets I Cutsets need to be “anchored” in order to use subadditivity Upshot: I This definition generalizes to d ≥ 3 Summary of d=2 I βp gives Wp through Wulff construction I Must show that βp is the “correct” norm One more modification Definition In d ≥ 3, the modified Cheeger constant of C̃n is ω |∂ H| Φ̃n := min : H ⊂ C̃n , 0 < |H| ≤ |C̃n |/d! |H| where ∂ ω H is the open edge boundary of H in C∞ . d ≥3 I βp,d defined analogously I Wp,d the associated Wulff shape I Ld (Wp,d ) = 2d /d! Theorem (G ’16) Let d ≥ 3, and let p > pc (d). There is Wp,d ⊂ [−1, 1]d convex so that Pp -a.s., max inf Gn ∈Gn x∈Rd ||1Gn − 1n(Wp,d +x)∩Cn ||`1 nd −−−→ 0 n→∞ d ≥3 The shape theorem gives Cheeger asymptotics. Theorem (G ’16) Let d ≥ 3, and let p > pc (d). There is a norm βp,d on Rd so that Pp -a.s., lim nΦ̃n = n→∞ perimeterβp,d (∂Wp,d ) θp (d)Ld (Wp,d ) where Wp,d is the limit shape from the previous theorem, and has volume 2d /d! Remarks Figure: Why (a priori) we use `1 convergence Summary Challenges in d ≥ 3: I finding a suitable norm I we lose graph duality I we lose theory of plane curves I vacant percolation Ingredients: I concentration estimates I (enhanced) renormalization argument of Zhang I geometric measure theory I + more Open questions The Wulff shape: I regularity? I facets? Boundary conditions: I original conjecture? I on a torus? `∞ in d ≥ 3? Figure: from Crystal Saga, Jean Giraud 1989 Thank you!