Large deviation principle at speed $n$ for the random metric in first-passage percolation

Julien Verges
2024-12-04
Abstract:Consider standard first-passage percolation on $\mathbb Z^d$. We study the lower-tail large deviations of the rescaled random metric $\widehat{\mathbf T}_n$ restricted to a box. If all exponential moments are finite, we prove that $\widehat{\mathbf T}_n$ follows the large deviation principle at speed $n$ with a rate function $J$, in a suitable space of metrics. Moreover, we give three expressions for $J(D)$. The first two involve the metric derivative with respect to $D$ of Lipschitz paths and the lower-tail rate function for the point-point passage time. The third is an integral against the $1$-dimensional Hausdorff measure of a local cost. Under a much weaker moment assumption, we give an estimate for the probability of events of the type $\{\widehat{\mathbf T}_n \le D \}$.
Probability
What problem does this paper attempt to address?
The problem that this paper attempts to solve is about **the large deviation principle for the lower tail of the random metric in the First - Passage Percolation (FPP) model**. Specifically, the author studied the large - deviation behavior of the lower tail of the rescaled random metric $\hat{T}_n$ restricted in a box in the standard First - Passage Percolation model. ### Problem Background The First - Passage Percolation model was introduced by Hammersley and Welsh in 1966 to describe the random propagation process on the lattice $Z^d$. Each edge $e$ is assigned an independent and identically distributed random variable $\tau_e$, which represents the time required to pass through this edge. For any two points $x, y\in Z^d$, the passage time $T(x, y)$ between them is defined as the minimum total passage time among all paths connecting these two points. ### Research Objectives The author's main objective is to study, under certain conditions, the large - deviation behavior of the probability that the rescaled random metric $\hat{T}_n$ takes a value lower than a certain target pseudo - metric $D$ in the box. Specifically: 1. **Large - Deviation Principle**: When all exponential moments are finite, prove that $\hat{T}_n$ satisfies the large - deviation principle at speed $n$ and give the corresponding rate function $J$. 2. **Rate Function Expression**: Provide three different expressions of the rate function $J(D)$, involving the metric derivative of Lipschitz paths, the lower - tail rate function of point - to - point passage times, and the one - dimensional Hausdorff measure integral of local costs. 3. **Estimation under Weak Moment Assumptions**: Under weaker moment assumptions, give the probability estimate of the event $\{\hat{T}_n\leq D\}$. ### Main Contributions - **Extended Existing Results**: Extended Kesten's results on the large - deviation of the lower tail of point - to - point passage times to arbitrary directions and more general metric spaces. - **Provided New Rate Function Expressions**: By introducing concepts such as highway networks and path gradients, multiple expressions of the rate function are given. - **Wide Application**: These results are not only applicable to the First - Passage Percolation model but can also be generalized to other similar random geometric models. ### Formula Summary - **Rescaled Random Metric**: \[ \hat{T}_n(x, y)=\frac{1}{n}T_{[0,n]^d}(\lfloor nx\rfloor,\lfloor ny\rfloor) \] - **Rate Function**: \[ J^{-}(D)=\lim_{\varepsilon\rightarrow0}\lim_{n\rightarrow\infty}-\frac{1}{n}\log P(LD_n^{-}(D,\varepsilon)) \] - **Highway Network Expression**: \[ J^{-}(D)=\sum_{k\geq1}\int_0^LJ_{pp}(\sigma_k'(t),|\dot{\sigma}_k|_D(t))dt \] - **Path Gradient Expression**: \[ J^{-}(D)=\int_X\max_{u\in S^2}J_{pp}(u,(P - \text{grad }D)_z(u))H^1(dz) \] Through these studies, the author provides new tools and methods for understanding the large - deviation behavior in the First - Passage Percolation model.