On the cost of the bubble set for random interlacements (Q6107882)

The paper treats random interlacements of \(\mathbb{Z}^d\) (for \(d\geq 3\)) in the strongly percolative regime. Informally, the random interlacement at level \(u>0\) is a random subset \(\mathcal {I}^u\) of \(\mathbb{Z}^d\) that describes the picture left by the trace of a random walk on a large discrete torus, run up to a number of steps that is proportional to the size of the torus. As \(u\) increases, the complement \(\mathcal V^u\) of \(\mathcal {I}^u\) in \(\mathbb{Z}^d\) thins out, and there is a value \(u_\ast>0\), called the \emph{critical level for percolation} such that \(\mathcal V^u\) has an infinite component when \(u< u_\ast\) and all connected components of \(\mathcal V^u\) are finite when \(u>u_\ast\). The assumption of \emph{strong criticality} made in the present paper is stronger: informally strongly percolative means local presence and local uniqueness of an infinite cluster of \(\mathcal{V}^u\). There is another threshold \(\overline u\leq u_\ast\), called the critical level for strong percolation, such that strong criticality holds whenever \(u<\overline u\). Thus, the paper assumes that \(u<\overline u\). It is conjectured, but not proved, that \(u_\ast = \overline u\). The present article studies the occurrence of the fraction of sites in a discrete box that get disconnected by the interlacements from the boundary of a box twice the size. An upper bound for the exponential decay of the probability of such sites exceeding a given threshold is the main result. This continues previous work by the author on this subject. In particular, if the conjecture that \(u_\ast=\overline u\) turns out to be true (as expected), the upper bound derived in this paper matches to first order a previously known lower bound. More formally, let \(D_N=[-N,N]^d\cap \mathbb{Z}\) and \(S_N\) be the set of \(x\in\mathbb{Z}^d\) with sup-norm equal to \(N\). For \(r>0\) define \(\mathcal{C}_r^u\) to be the connected component of \(S_r\) in \(\mathcal {V}^u\cup S_r\), so that the set of points in \(D_N=[-N,N]^d\cap \mathbb{Z}\) that are disconnected from \(S_{2N}\)by \(\mathcal {I}^u\) may be written \(D_N\setminus C_{2N}^u\). The main result of the paper is a large deviations upper bound of speed \(N^{d-2}\) for the probability of the event \(\mathcal {A}_N\) that \(D_N\setminus C_{2N}^u\) contains more than \(v |D_N|\) elements, for a suitably chosen parameter \(v\): \[ \limsup_{N\to\infty} \frac1{N^{d-2}} \log \mathbb{P}(A_N) \leq -\widehat{J}_{u,v}, \] where the function \(\widehat{J}_{u,v}\) is given by the minimum of \(\frac1{2d} \int_{\mathbb{R}^d} |\nabla \phi(z)|^2 dz\) over suitable positive, locally integrable functions \(\phi\) on \(\mathbb{R}^d\) with finite Dirichlet energy. The results significantly improve the findings in the author's previous work (see [Probab. Theory Relat. Fields 167, No. 1--2, 1--44 (2017; Zbl 1365.60080); Probab. Math. Phys. 2, No. 3, 563--611 (2021; Zbl 1483.60149)]). In particular, they match up with a previously known large deviations lower bound with the same speed and function \(\widehat{J}_{u,v}\) to give a full large deviations principle if the conjecture \(\overline u=u_\ast\) holds.