The Wulff crystal in Ising and percolation models. École d'Été de Probabilités de Saint-Flour XXXIV -- 2004 (Q2500515)

This is a self-contained survey on recent mathematically rigorous works on the phase coexistence phenomenon in three fundamental and much-studied microscopical models: the Ising model, the Bernoulli percolation model and the random cluster model. The author is one of the most distinguished researchers in this field and gives a detailed and comprehensive account on the heuristics, the achieved results and their proofs. Those proofs that can be found in research papers only are included, while classical auxiliary results are only quoted. The proofs have been streamlined and shortened in comparison to the original literature. The structure of the text is as follows. Chapter 1 gives a non-technical introduction to the phenomenon, Chapters 2--4 introduce the three respective models, in Chapter 5 the main results on the Wulff crystal are formulated and commented and their historic development is surveyed, Chapter 6 gives a brief introduction to large-deviations theory, the surface-order respectively volume-order large-deviation results leading to the main results are presented in Chapters 7 and 8, Chapters 9--12 contain fundamental probabilistic estimates needed in the proofs, while Chapters 13--15 present the necessary geometric tools, and the final steps of the proofs are in Chapters 16--19. Some simulations round the presentation. The main result discussed in these lecture notes for the case of the Ising model is roughly formulated as follows (assuming the reader's familiarity with this model). Let the dimension \(d\geq 2\) and fix any temperature below the critical temperature. Hence, the Ising model exhibits a phase transition, and the spontaneous magnetization \(m^ *\) is positive. We are interested in the empirical magnetization \(\sigma_ n=n^{-d}\sum_{x\in \Lambda_ n}\sigma(x)\delta_{x/n}\), where \(\sigma(x)\in\{1,-1\}\) is the spin at site \(x\in\mathbb Z^ d\), \(\Lambda_ n\) is the centered box in \(\mathbb Z^ d\) with diameter \(n\), and \(\delta_z\) is the point measure at \(z\). Let \(b_ n=n^{-d-1}\sum_{x\in \Lambda_ n}\sigma(x)x\) be the mean of \(\sigma_ n\). Then there exists a bounded, closed, convex set \(W\) (called the \textit{Wulff crystal}) such that \(W^ \circ\) contains 0 and such that the following holds. Rescale \(W\) by putting \(W_m=(\frac {m^*-m}{2m^ *})^{1/d} W/| W| ^{1/d}\), where \(m<m^ *\) is chosen such that \(W_ m\subset [-\frac 12,\frac 12]^ d\). Let \(w_ n\) be the measure having density \(-m^ *\) on \(W_ m-b_ n/(m^*-m)\) and \(m^ *\) on the complement. Then, under the conditional Ising measure (with plus boundary condition) given that \(n^{-d}\sum_{x\in \Lambda_ n}\sigma(x)\leq m\), the difference between \(\sigma_ n\) and \(w_n\) converges towards zero with speed of order \(\exp\{-c n^{d-1}\}\). That is, with probability tending to one extremely fast, the rescaled minus region assumes a deterministic shape, and the two phases are separated from each other.