Renormalization of the two-dimensional Lotka-Volterra model (Q2426611)

\textit{C. Neuhauser} and \textit{St. W.~Pacalla} [Ann. Appl. Probab. 9, 1226--1259 (1999; Zbl 0948.92022)] introduced the so-called stochastic Lotka-Volterra model as a spatial ecological model for the competition of two types of plants, say \(0\) and \(1\). At each site \(x\) of the \(d\)-dimensional integer lattice there is exactly one plant, either of type \(0\) or \(1\); hence a typical state of the model is \(\xi\in\{0,1\}^{\mathbb{Z}^d}\). Let \(f_i=f_i(x,\xi)\) be the proportion of neighboring sites of type \(i\). More precisely, \(f_i=\sum_{y}p(x,y)\mathbf{1}_{\{i\}}(\xi(y))\). Here \(p\) is the kernel of a symmetric random walk with covariance matrix \(\sigma^2\) times the unit matrix and with \(p(x,x)=0\). Furthermore, let \(\alpha_1,\alpha_2\geq0\) be parameters. At each site the present plant of type \(i\) dies at rate \(f_i+\alpha_{i}f_{1-i}\) and is immediately replaced by a plant of type \(j\) with probability \(f_j\). That is, the \textit{flip rate} for site \(x\) is \[ c(x,\xi)=f_{1-i}(f_i+\alpha_if_{1-i}). \] The parameters \(\alpha_1\) and \(\alpha_2\) can be interpreted as measuring the interspecies competition while the parameters for intraspecies competition equal \(1\). The corresponding strong Markov process \((\xi_t)_{t\geq0}\) is called the stochastic Lotka-Volterra model. Note that the special case \(\alpha_1=\alpha_2=0\) is the celebrated voter model. In an earlier paper [Ann. Probab. 33, 904-947 (2005; Zbl 1078.60082)], the authors showed that a properly space-time-mass rescaled version of the model converges to super-Brownian motion if \(d\geq3\). Later [Probab. Theory Relat. Fields 139, No. 1--2, 89--142 (2007; Zbl 1129.60089)] they used this result to identify a region of parameters \((\alpha_0,\alpha_1)\) where the system \textit{survives}: If started with only one plant of type \(1\), with positive probability, the population of type \(1\) will persist for all times. The aim of the present paper is to establish both of these results for the case of dimension \(d=2\). For \(N=1,2,\ldots\) define the scale \(N'=N'(N)=N/\log N\) (for \(d\geq3\) the appropriate scale is \(N'=N\)). Let \(\xi^N\) the Lotka-Volterra model with flip rates \(c_N(x,\xi)=Nf_{1-i}(f_i+\alpha_i^Nf_{1-i})\) (if \(\xi(x)=i\)), where the parameters \(\alpha_i^N\) are assumed to fulfill \(N'(\alpha_i^N-1)\to \theta_i\) as \(N\to\infty\) for some \(\theta_i\in\mathbb{R}\). Define the measure-valued process \(X^N\) by \[ X^N_t=\frac{1}{N'}\sum_{x\in\mathbb{Z}^d}\xi^N_t(x)\delta_{x/\sqrt N} \] and assume \(X^N_0\to X_0\) weakly for some finite measure \(X_0\) on \(\mathbb{R}^2\). In Theorem 1.2 the authors show that (in the Skorohod space of paths taking values in the finite measures on \(\mathbb{R}^2\)), the sequence \((X^N)_{N=1,2,\ldots}\) converges to super-Brownian motion with branching rate \(4\pi\sigma^2\), drift \(\theta=\gamma_0\theta_0-\gamma_1\theta_1\) and variance \(\sigma^2\) for the motion. Here \(\gamma_0\) and \(\gamma_1\) are quantities described as certain hitting probabilities of coalescing random walks. Recall that this super-Brownian motion can be characterized as the solution of the following martingale problem: For any bounded smooth function \(\phi:\mathbb{R}^2\to\mathbb{R}\) \[ M_t(\phi):=X_t(\phi)-X_0(\phi)-\int_0^tX_s((\sigma^2/2)\Delta\phi)\,ds -\theta\int_0^tX_s(\phi)\,ds \] is a continuous martingale with quadratic variation \[ \langle M(\phi)\rangle_t=\int_0^t X_s(\gamma\phi^2)\,ds. \] Theorem 1.3 describes a condition on \(\alpha_0\) and \(\alpha_1\) that ensures survival of \((\xi_t)_{t\geq0}\): For \(\eta\in(0,1)\) there exists an \(r(\eta)>0\) such that \((\xi_t)_{t\geq0}\) survives if \(\alpha_0>1-r(\eta)\), \(\alpha_1<1+r(\eta)\) and \[ \alpha_1-1<\begin{cases} (1-\eta)(\alpha_0-1),\qquad\text{if }\alpha_0\geq 1,\\ (1+\eta)(\alpha_0-1),\qquad\text{if }\alpha_0< 1. \end{cases} \]