Last passage percolation is a mathematical model where you have a grid of points, each assigned a random value, and you want to find the path from one corner to the opposite corner that collects the largest total value along the way. This winning path is called a geodesic. The question this paper asks is: among all the possible paths the geodesic could take, which specific path is it most likely to be?
The authors argue, and partially prove, that the geodesic is most likely to follow what they call "corner paths," meaning routes that hug the edges of the grid rather than wandering through the interior. They identify three separate mathematical reasons why corner paths have this advantage over more central routes. Intuitively, a path along the boundary has less competition from alternative nearby paths, and the random values it collects tend to be more favorable in a statistical sense. For a specific version of the model where the random values follow an exponential distribution, they can make this precise using tools called moderate deviation estimates, showing that corner paths are nearly the most probable in a quantifiable way.
The paper also establishes a monotonicity result, meaning that paths closer to the corners are more likely than paths further from them, at least in a special simplified setting. The authors conjecture this ordering holds much more broadly throughout the model. Taken together, the results give a clearer picture of the geometry of randomness in these models, which matter beyond mathematics because last passage percolation connects to models of random growth, queuing theory, and the behavior of large random matrices.