Fluctuations of the linear functionals for supercritical non-local branching superprocesses

Abstract

Suppose $\{X_{t}:t\ge 0\}$ is a supercritical superprocess on a Luzin space $E$, with a non-local branching mechanism and probabilities $\mathbb{P}_{\delta_{x}}$, when initiated from a unit mass at $x\in E$. By ``supercritical", we mean that the first moment semigroup of $X_{t}$ exhibits a Perron-Frobenius type behaviour characterized by an eigentriplet $(\lambda_{1},\varphi,\widetilde{\varphi})$, where the principal eigenvalue $\lambda_{1}$ is greater than $0$. Under a second moment condition, we prove that $X_{t}$ satisfies a law of large numbers. The main purpose of this paper is to further investigate the fluctuations of the linear functional $\mathrm{e}^{-\lambda_{1}t}\langle f,X_{t}\rangle$ around the limit given by the law of large numbers. To this end, we introduce a parameter $\epsilon(f)$ for a bounded measurable function $f$, which determines the exponent term of the decay rate for the first moment of the fluctuation. Qualitatively, the second-order behaviour of $\langle f,X_{t}\rangle$ depends on the sign of $\epsilon(f)-\lambda_{1}/2$. We prove that, for a suitable test function $f$, the fluctuation of the associated linear functional exhibits distinct asymptotic behaviours depending on the magnitude of $\epsilon(f)$: If $\epsilon(f)\ge \lambda_{1}/2$, the fluctuation converges in distribution to a Gaussian limit under appropriate normalization; If $\epsilon(f)<\lambda_{1}/2$, the fluctuation converges to an $L^{2}$ limit with a larger normalization factor. In particular, when the test function is chosen as the right eigenfunction $\varphi$, we establish a functional central limit theorem. As an application, we consider a multitype superdiffusion in a bounded domain. For this model, we derive limit theorems for the fluctuations of arbitrary linear functionals.