AG-2026.10-1369·math.PR·cross-listed: math-phmath.CAmath.MG
Directed fractal percolation and non-Lipschitz variants
Authors
- Shirshendu Ganguly
- Victor Ginsburg
- Kaihao Jing
Abstract
While models of undirected and directed geometry in random i.i.d. disorder with rapidly decaying tails are expected to behave similarly and exhibit features of the Kardar--Parisi--Zhang universality class, the scenario when the background noise is fractal is expected to be significantly different. A canonical example of the latter is Liouville Quantum Gravity, where the planar Gaussian free field forms the fractal background. A toy model capturing some of the essential features is given by Mandelbrot's fractal percolation, a random Cantor set. In the simplest setting, dyadic cubes of different scales are retained independently with probability $p$ (a parameter of the model), and the intersection of all retained cubes forms the set of open sites. Connectivity properties of such sets have been intensely studied. Across [CCD88, Cha95], a surprising result was proven. Namely, while there is a phase transition for connectivity of the fractal percolation set (like in usual bond percolation), directed percolation never occurs. Results of a similar spirit have been obtained in geometric measure theory as well. As a step towards understanding last passage percolation (LPP) driven by the Gaussian free field (initiated in [GGN24]), we study the problem of quantifying LPP in fractal percolation. Developing a multi-scale framework, we show an almost polynomial gap from linear growth of the passage time. The obstruction to high LPP values is the Lipschitz nature of directed paths. Along these lines, [Cha96] showed that any connected subset of fractal percolation must have Hausdorff dimension strictly larger than one. We show that one can construct directed paths passing through open sites in fractal percolation provided they move exponentially fast in the spatial direction---a model we introduce and term as unrectifiable directed percolation.
Submitted
8 October 20261 week ago
Version
v1
License
CC-BY-4.0
DOI
10.48550/arXiv.2610.12366
Summary
When percolation (connectivity in random networks) happens on fractal structures instead of regular lattices, directed paths can't grow linearly—they need to move exponentially fast through space to connect, revealing a fundamental obstruction tied to the roughness of the underlying geometry.
- Fractal percolation blocks ordinary directed percolation entirely, but allowing paths to 'jump' exponentially far in space (a new 'unrectifiable' model) makes connectivity possible—showing the geometry itself, not randomness, creates the barrier.
- The authors quantify this blockade rigorously by proving passage times grow almost polynomially rather than linearly, using a multi-scale framework that could apply to more complex fractal models like those from quantum gravity.
- This work builds toward understanding last-passage percolation on the Gaussian free field, where fractal geometry is intrinsic; the gap between linear and exponential growth here may be a universal signature of disorder on self-similar structures.
curious · generated by claude-haiku-4-5
Chat with this PDF
Ask questions, probe assumptions, request a plain-English summary. Answers cite sections from the preprint itself.
Community
Questions and answers about this paper from other readers. No formal peer review — just a place to think out loud.