Harmonic pinnacles in the discrete Gaussian model (Q2630720)

The \( 2D \) discrete Gaussian model gives each height function \(\eta:{Z^2}\rightarrow {Z}\) a probability proportional to \(\exp(-\beta H(\eta)), \) where \( \beta\) is the inverse temperature and \(H(\eta)=\sum_{\sim}(\eta_x-\eta_y)^2\) sums over nearest-neighbor bonds. The authors consider the discrete Gaussian model at large fixed \(\beta,\) where it is flat unlike its continuous analogue. They also establish the following main results: Theorem 1. Fix \(\beta >0\) large enough and let \( X_L \) be the maximum of the discrete Gaussian model on an \( L\times L\) box in \( Z^2 \) the inverse temperature \(\beta\). Then there exists some \( M=M(L) \) with \[ M\sim\sqrt{(1/2\pi\beta)\log L\log\log L} \] such that \( X_L\in\{M,M+1\}\) with probability going to \(1\) as \(L\rightarrow\infty\). Theorem 2. Fix \( \beta>0\) sufficiently large and consider the discrete Gaussian model on an \( L\times L\) box in \( Z^2\) at inverse temperature \(\beta\) with a floor at \(0\). Then there exists some \(H=H(L)\) with \( M\sim\sqrt{(1/4\pi\beta)\log L \log\log L }\) such that w.h.p. \[ \# \{v:\eta_v\in\{H,H+1\}\}\geq(1-\varepsilon_\beta)L^2, \] where \(\varepsilon_\beta \) can be made arbitrary small as \( \beta \) increases. Theorem 3. Fix \( \beta>0 \) large enough and let \( X^*_L \) be the maximum of the discrete Gaussian on an \(L\times L\) box at inverse temperature \(\beta\) with a floor at \(0\). Let \( M^*\equiv M^{(L)}=M(L)+H(L),\) where \(M(L), H(L)\) are given in Theorem 1 and 2. In particular, \[ M^*\sim\frac{1+\sqrt{2}}{2\sqrt{\pi\beta}}\sqrt{\log L \log\log L}. \] Then w.h.p. \( X^*_L\in\{M^*,M^*+1,M^*+2\}\).