Cone of No Escape

A live simulation of competing first-passage percolation on the square lattice: a lone seed at the origin races an adversary that already owns an infinite wedge pointed straight at it. Drag θ and watch the exact phase transition proved in arXiv:2607.07589 play out cell by cell.

SURVIVES
growing… t = 0.00
Your territory Opponent's cone Unclaimed

Controls

π/2
Theory: escapes to infinity (θ < π/2)
Your cells
0
Opponent cells
0
Your share
0%
Grid claimed
0%
Each edge of the lattice gets an independent Exp(1) "clock" — a standard first-passage-percolation / exponential-race growth rule. Both colors expand along a shared random weight field computed once via a single multi-source Dijkstra pass, then replayed in time order.

Space play/pause   ←/→ nudge θ

Territory over simulated time

What the paper actually proves

Daniel Ahlberg, Maria Deijfen, Matteo Sfragara — "Surviving from the Tip of a Cone in Competing First-Passage Percolation" — arXiv:2607.07589 (math.PR)

The paper studies two-type first-passage percolation on ℤ²: every edge of the grid gets an independent random passage time, and two "infections" spread outward, each claiming any site it reaches before the other. One infection starts as a single point at the origin. The other starts already occupying an infinite cone (wedge) with its tip at the origin, opening at angle θ directly toward the lone seed — both use the exact same passage-time distribution, so it's a fair race over a shared random landscape.

The dichotomy

The authors confirm a conjecture of Itai Benjamini:

"The type starting at the origin can grow unboundedly if and only if the slope [angle] of the cone is strictly smaller than π/2."

In other words: a wedge narrower than a right angle always leaves the lone seed a corridor to infinity — it survives forever, no matter how the randomness falls. Widen the wedge to a right angle or beyond and the seed is eventually pinched off: it can only ever claim a bounded region — it is swallowed. That threshold, exactly θ = π/2, is the phase transition this demo animates.

How it's proved

The proof doesn't track individual growth fronts directly. Instead it uses Busemann functions — asymptotic time-difference "compass directions" for the percolation metric that describe infinite time-minimizing paths — combined with local modification arguments on the percolation configuration. The stated theorem applies when θ is a direction along which the model's asymptotic (limit) shape has a tangent line, a regularity condition tied to the classical FPP shape theorem.

Simplifications made for this interactive demo: the lattice here is a finite 161×161 window standing in for ℤ², so "reaches the edge of the visible grid" is used as a finite-time proxy for "grows unboundedly" — the real theorem is an asymptotic (t → ∞) statement on the infinite lattice. Edge weights are drawn i.i.d. from an Exponential(1) distribution, a canonical choice for continuous-time ("exponential clock") growth models; the paper's dichotomy is about the θ = π/2 threshold itself and is not tied to one specific passage-time law. The θ ≥ π/2 side of the slider intentionally extends well past π/2 (up to ~350°) purely so you can watch the opponent's advantage grow — the paper's proven threshold is exactly π/2, not any larger angle shown here.