The Abelian sandpile model is a simple mathematical system defined on a grid where each site holds a number of sand grains. When a site accumulates too many grains, it topples, distributing grains to its neighbors, which can trigger further topplings in a cascade. If you run this process long enough, the system reaches a statistical steady state. One natural question is: how often is a given site completely empty (height zero) in that steady state? This paper studies how that probability behaves when you zoom out, replacing the discrete grid with finer and finer lattices approximating a smooth region of the plane.
The main finding is that, under a symmetry assumption on the lattice, the height-zero probability has a clean two-part structure as the lattice spacing shrinks. There is a constant background probability that depends only on which type of lattice you use, plus a small correction term that depends on the geometry of the surrounding region in a conformally covariant way. Conformal covariance is a strong geometric property meaning the correction transforms predictably under angle-preserving maps of the plane, which is a hallmark of many two-dimensional critical systems. This result holds across a broad class of symmetric lattices, generalizing earlier work that was limited to the square lattice and a specific simple domain.
However, the paper also shows that the symmetry assumption is genuinely necessary. The authors construct explicit counterexamples, including certain isoradial graphs (a well-studied family of lattices), where the clean conformally covariant behavior breaks down. Interestingly, conformal covariance can sometimes be rescued by using a different, less obvious way of measuring distances on those grids. The work is partly motivated by a 2006 conjecture that results computed for a specific case should hold much more broadly, and this paper represents concrete mathematical progress toward proving that conjecture while also carefully mapping out where and why it can fail.