Coalescing and annihilating random walk with `action at a distance' (Q1580489)

Two models of simple interacting particle systems are defined as follows. Let the particles randomly and independently move like continuous-time random walks. In the so-called annihilating model each particle that decides to jump to a site that is already occupied is removed together with the particle present at that site: they annihilate each other. In the coalescing model only the jumping particle is removed from the system: they coalesce. Asymptotics for the probability \(p_t\) that a given site is occupied at time \(t\) as \(t\to\infty\) have earlier been obtained by Arratia, and variants have been considered by van den Berg and Kesten and by Stephenson. The present paper considers another variant of the model in which the interaction between the particles is present at a certain distance already. This is a model for chemical processes (coagulation, e.g.) and has been proposed by Smoluchowski. The mechanism is as follows. Assume that a particle present at the site \(x\) decides to jump to a site \(y\). In the annihilating model, if \(y\) is occupied, both particles are removed. If \(y\) is vacant, but one of the neighbors of \(y\) other than \(x\) is occupied, then the jumping particle is removed, together with one particle that is randomly chosen among these neihboring particles. If \(y\) and all its neighbors are vacant, then the particle just jumps from \(x\) to \(y\). In the coalescing model, if \(y\) is occupied, then the jumping particle is removed. If \(y\) is vacant, but precisely \(k\in\{0,\dots,2d\}\) neighbors of \(y\) other than \(x\) are occupied, then the jumping particle is removed with probability \(k/(k+1)\), or one of the neighbors is removed, each with probability \(1/(k+1)\). This defines a Markov process on the space of subsets of \(\mathbb Z^d\) which do not contain any pair of neighboring sites. The underlying random walk is assumed to be genuinely \(d\)-dimensional and to have finite range and zero mean. The initial state is assumed to be translation invariant and to be constructable from i.i.d. random variables in a certain way. The main result of the paper is the proof of the asymptotics for the occupation probability \(p_t\) of, say, the origin at time \(t\) as \(t\to\infty\): For a certain constant \(C_0>0\) and some \(\zeta>0\), it is shown that \(p_t=1/(C_0 t)+{O}(t^{-1-\zeta})\) in the annihilating model, and \(p_t=2/(C_0 t)+{O}(t^{-1-\zeta})\) in the coalescing model. The results have been proven in dimensions \(d\geq 9\) only but are conjectured to be true for any \(d\geq 3\). One reason for putting the mechanism in the way specified above is that a certain nice super martingale property needed in the proof seems to be ensured only by this mechanism.