Extremes of the discrete two-dimensional Gaussian free field (Q2497167): Difference between revisions

A two-dimensional Gaussian free field (GFF) is a family of centered Gaussian random variables \(\Phi =\{\Phi_x; x\in V_N\}\) with covariance given by \(G_N(x,y)\), the Green function of the two-dimensional simple random walk in the square \(V_N= [ 1,N]^2\cap \mathbb Z^2\) stopped at the boundary. The conditioned free field (CFF) is the law of \(\Phi\) conditioned on being nonnegative. The first part of the paper contains a precise description of the number of high (resp. low) points of a GFF and the fractal properties of their repartition in \(V_N\). The proofs use a multiscale decomposition introduced by \textit{E. Bolthausen, J.-D. Deuschel} and \textit{G. Giacomin} [Ann. Probab. 29, No. 4, 1670--1692 (2001; Zbl 1034.82018)] who first computed the maximum of a two-dimensional GFF and, in particular, showed that the mean of the CFF is of the order \(\log N\). The proofs also rely on a close analogy between the GFF and \(\tau(x)\), the first time \(x\) is visited by a simple random walk on the torus \(\mathbb Z^2/N\mathbb Z^2\) starting at 0, which is carefully studied by \textit{A. Dembo, Y. Peres, J. Rosen} and \textit{O. Zeitouni} [ibid. 34, No. 1, 219--263 (2006; Zbl 1100.60057)]. These results are then used to study the CFF. One sees that, contrary to the situation in dimensions \(\geq 3\), the extrema of the shifted GFF and the CFF look alike.