Percolation theory: Stochastic models of porous media (Q1345444)

This well written and informative expository paper gives to readers (of German) a gentle introduction to the thorny area of percolation theory. My only criticism is the author's omission of the obvious connections to random graph theory. Thus, let \(G\) be a graph with vertex set \(V\) and edge set \(E\). Thinking of \(G\) as a network of roads connecting cities, we independently ``bomb'' each edge with fixed probability \(q\) of demolishing it. With probability \(p = 1 - q\) it remains usable. Percolation theory asks the question: What journeys remain possible within the (random) subgraph of \(G\) produced by this ``bombing raid''? The sample space of the ``experiment'' is \(\Omega = \{0, 1\}^ E\), with probability \(P =\) the infinite product of the probability on each copy of \(\{0, 1\}\) which assigns the value \(q\) to \(\{0\}\) and \(p\) to \(\{1\}\). When \(V = \mathbb{Z}^ 2\) and \(E =\) the set of all line segments of length one which join points of \(V\), the resulting pair \((\Omega, P)\) is called the standard model. Most results are stated for this model with generalizations to other ``lattice models'' noted. While results are mainly quoted without proof, a major exception is the following: Declare \(v,w \in V\) to be equivalent if the random subgraph of \(G\) contains a path from \(v\) to \(w\). Equivalence classes of this relation are called clusters. Let \(\theta (p)\) denote the probability that \((0,0)\) belongs to an infinite cluster. Then, in the standard model and some others, there exists a number \(p_ c\), \(0 < p_ c < 1\), such that \(\theta (p) = 0\) when \(p < p_ c\), and \(\theta (p) > 0\) when \(p > p_ c\). After giving an elementary proof of this, the author uses ergodic theory to show, when \(p > p_ c\), that with probability one there exists exactly one infinite cluster. Also, \(\theta (p_ c) = 0\), but this does not generalize to the case \(V = \mathbb{Z}^ d\), \(d > 2\). A good deal of attention is paid to the question: Given (disjoint) subsets \(A\) and \(B\) of \(V\), what is the probability of a ``crossover'' from \(A\) to \(B\), i.e. a path in the random subgraph from a point of \(A\) to a point of \(B\)? This is of particular interest in the standard model when \(p = p_ c\). When we consider corresponding results for the shrunken lattice with \(V = V_ N = ({1 \over N} \mathbb{Z})^ 2\), limits may be taken to form a ``percolation structure'' on all of \(\mathbb{R}^ 2\), or on an open subset of \(\mathbb{R}^ 2\). Different percolation structures are linked by structure-preserving morphisms, and these may also be used to extend the idea of percolation structure to Riemann surfaces and other manifolds. The eventual involvement of Hausdorff dimension, conformal mappings, homology groups, and the Dedekind eta-function, inter alia, attests to the depth and profundity of results which may be obtained in this area.