On a Predator-Prey System with Random Switching that Never Converges to its Equilibrium

Source abstract

We study the dynamics of a predator-prey system in a random environment. The dynamics evolves according to a deterministic Lotka--Volterra system for an exponential random time after which it switches to a different deterministic Lotka--Volterra system. This switching procedure is then repeated. The resulting process is a piecewise deterministic Markov process (PDMP). In the case when the equilibrium points of the two deterministic Lotka--Volterra systems coincide we show that almost surely the trajectory does not converge to the common deterministic equilibrium. Instead, with probability one, the densities of the prey and the predator oscillate between 0 and $\infty$. This proves a conjecture of Takeuchi, Du, Hieu, and Sato [ J. Math. Anal. Appl., 323 (2006), pp. 938--957]. The proof of the conjecture is a corollary of a result we prove about linear switched systems. Assume $(Y_t, I_t)$ is a PDMP that evolves according to $\frac{dY_t}{dt}=A_{I_t} Y_t$, where $A_0,A_1$ are $2\times2$ matrices and $I_t$ is a Markov chain on $\{0,1\}$ with transition rates $k_0,k_1>0$. If the matrices $A_0$ and $A_1$ are not proportional and are of the form $A_i := (\begin{smallmatrix} \alpha_i & \beta_i \\ \gamma_i & -\alpha_i \end{smallmatrix})$ with $\alpha_i^2 + \beta_i \gamma_i < 0$, then there exists $\lambda >0$ such that $\lim_{t \to \infty} \frac{\log \| Y_t \|}{t} = \lambda.$