Slow and fast scales for superprocess limits of age-structured populations (Q655325)

The authors are interested in asexual population models in which the survival probability and reproduction rate of each individual are characterized by a quantitative trait. They study the joint effects of age and trait structures in the interplay between ecology and evolution, and in particular, evolution, acting on the distribution of traits in the population, is the consequence of three basic mechanisms: (i) \textit{heredity} which transmits traits to new offspring, (ii) \textit{mutation} which drives the variation in the trait values, and (iii) \textit{selection} between these different trait values which is due to ecological interactions. This paper treats the model which relies on an individual-based birth and death process with age and trait introduced in author' previous paper [J.\ Math.\ Biol.\ 58, No. 6, 881--921 (2009; Zbl 1204.92053)]. It generalizes the trait-structured case developed in [\textit{N. Champagnat, R. Ferrière} and \textit{S. Méléard}, Stoch.\ Models 24, Suppl. 1, 2--44 (2008; Zbl 1157.60339)] and the age-structured case in [\textit{P. Jagers} and \textit{F. Klebaner}, Stochastic Processes Appl. 87, No. 2, 235--254 (2000; Zbl 1045.60090)]. The authors describe the dynamics of large populations composed of small individuals with short lives. Life-lengths and durations between reproductions are assumed to be proportional to the individuals' weights, and these weights are inversely proportional to the population size. When the population size tends to infinity, their main results show that two qualitatively different asymptotic behaviors arise from the separation of the fast age and show trait time scales. While the trait marginals converge in a pathwise sense to a superprocess, the age distributions stabilize into deterministic equilibria that depends on the traits. Assume that (a) the birth and death rates \(b\) and \(d\) are continuous on \({\mathcal X} \times {\mathbb R}_+\) and bounded respectively by \(\bar{b}\) and \(\bar{d}\) (where \({\mathcal X}\) is a closed subset of \({\mathbb R}^d\)); (b) the function \(r\) is continuous on \({\mathcal X} \times {\mathbb R}_+\), and there exist a positive constant \(\bar{r}\) and a non-negative real function \(\underline{r}\) such that for any \((x,a) \in\) \({\mathcal X} \times {\mathbb R}_+\), \(\underline{r}(a)\) \(\leqslant\) \(| r(x,a) |\) \(\leqslant \bar{r}\) with \[ \int_0^{\infty} \underline{r}(a) da = + \infty \,\, ; \tag{1} \] (c) the competition kernel \(U\) is continuous on \(( {\mathcal X} \times {\mathbb R}_+)^2\) and is bounded by \(\bar{U}\); (d) for any \(x \in {\mathcal X}\), the mutation kernel \(\pi^n(x, d h)\) has its support in \({\mathcal X} \setminus \{ x \}\) \(=\) \(\{ h \in {\mathbb R}^d\) \(|\) \(x + h \in {\mathcal X} \}\). The authors consider the following two cases: (Case I) the trait space \({\mathcal X}\) is a compact subset of \({\mathbb R}^d\) and there exists a generator \(A\) of a Feller semigroup on \(C_b({\mathcal X}, {\mathbb R})\) with domain \(D(A)\) dense in \(C_b({\mathcal X}, {\mathbb R})\) such that \[ \lim_{n \to \infty} \sup_{x \in {\mathcal X} } \left| n \int_{ {\mathcal X} \setminus \{ x \} } \{ f(x + h) - f(x) \} \pi^n(x, d h) - A f(x) \right| =0\tag{2} \] holds for any \(f \in D(A)\); (Case II) the trait space \({\mathcal X}\) is a closed subset of \({\mathbb R}^d\) and assume, in addition, that there exists \(\ell_1 \geq \ell_2 \geq 2\) such that \(C_b^{\ell_1} ({\mathcal X}, {\mathbb R} ) \subset D(A)\) and such that for any \(f \in C_b^{\ell_1}( {\mathcal X}, {\mathbb R})\) and any \(x \in {\mathcal X}\), \[ | A f(x) | \leqslant C \sum_{ _{\substack{ | k | \leqslant \ell_0 \\ k = ( k_1, \dots, k_d) }} } | D^k f(x) | \tag{3} \] and \[ \sup_{x \in {\mathcal X} } \left| n \int_{ {\mathcal X} \setminus \{ x \} } \{ f(x + h) - f(x) \} \pi^n(x, d h) - A f(x) \right| \leqslant \varepsilon_n \sum_{ _{\substack{ | k | \leqslant \ell_1 \\ k = ( k_1, \dots, k_d) }} } \| D^k f \|_{\infty} \tag{4} \] where \(D^k f(x) = \partial_{x_1}^{k_1} \dots \partial_{x_d}^{k_d} f(x)\), \(\varepsilon_n\) is a sequence tending to 0 as \(n\) tends to infinity and \(C\) is a constant. The population at time \(t\) is represented by a point measure \[ X_t^n = \frac{1}{n} \sum_{i=1}^{ N_t^n} \delta_{ ( X_i(t), A_i(t) )} \tag{5} \] where \(N_t^n= \langle n X_t^n, 1 \rangle\) is the number of individuals alive at time \(t\), and \(X_i(t)\) (resp. \(A_i(t)\)) denotes respectively the trait and age of individual at time \(t\). For any \(x \in {\mathcal X}\), the age probability density is given by \[ \hat{m}(x,a) = \frac{ \displaystyle{ \exp \left\{ - \int_0^a r(x,\alpha) d \alpha \right\} }}{ \displaystyle{ \int_0^{\infty} \exp \left\{ - \int_0^a r(x, \alpha ) d \alpha \right\} d a }}. \tag{6} \] The marginal \(\bar{X}_t^n( dx)\) of \(X_t^n(dx, da)\) is defined for any bounded and measurable function \(f\) on \({\mathcal X}\) and for any \(t \in {\mathbb R}_+\) by \[ \int_{ {\mathcal X} } f(x) \bar{X}_t^n ( dx) = \int_{ {\mathcal X} \times {\mathbb R}_+ } f(x) X_t^n ( dx, da). \tag{7} \] Here is the main result. Theorem. Assume that \(\sup_{n \in {\mathbb N} } E \langle X_0^n, 1 \rangle < \infty\), and assume that there exists a finite measure \(X_0 \in\) \(M_F( {\mathcal X} \times {\mathbb R}_+ )\) such that \(\lim_{n \to \infty} X_0^n\) \(=\) \(X_0\) in \(M_F({\mathcal X} \times {\mathbb R}_+ )\) endowed by the weak topology. Then, for each \(T > 0\), the sequence \(( \bar{X}^n )\), \(n \in {\mathbb N}\), converges in law in the Skorokhod space \({\mathbb D}( [0,T]\), \(M_F({\mathcal X}))\) to the unique superprocess \(\bar{X} \in\) \(C( [0, T]\), \(M_F( {\mathcal X}) )\) such that for any function \(f \in D(A)\), \[ \begin{aligned} M_t^f = & \langle \bar{X}_t, f \rangle - \langle \bar{X}_0, f \rangle \tag{8} \\ & - \int_0^t \int_{ {\mathcal X} } \left\{ ( \hat{pr})(x) \cdot A f(x) - [ \hat{b}(x) - ( \hat{d}(x) + \bar{X}_s \hat{U}(x) ) ] f(x) \right\} \bar{X}_s(dx) ds \end{aligned} \] is a square integrable with quadratic variation \[ \langle M^f \rangle_t = \int_0^t \int_{ {\mathcal X} } 2 \hat{r}(x) f^2(x) \bar{X}_s(dx) ds. \tag{9} \] Here \(M_F(E)\) denotes the set of finite measures on \(E\), and any \(\hat{\psi}(x)\) denotes \(\hat{\psi}(x)\) \(=\) \(\int_{ {\mathbb R}_+ } \psi(x,a) \hat{m}(x,a) da\) for a bounded function \(\psi(x,a)\). Moreover, \(\bar{X}_t \hat{U}(x)\) is given by \[ \bar{X}_t \hat{U}(x) = \int_{ {\mathcal X} } \left( \int_{ {\mathbb R}_+ } \int_{ {\mathbb R}_+ } U( (x,a), (y, \alpha)) \hat{m}(y, \alpha) d \alpha \cdot \hat{m}(x,a) da \right) \bar{X}_t ( dy). \tag{10} \] Their results generalizes the work done by \textit{K. Athreya, S. Athreya} and \textit{S. Iyer} [Bernoulli 17, No. 1, 138--154 (2011; Zbl 1284.60155)]. The techniques they use are based on martingale properties and generalize to this infinite dimensional setting the treatment of the slow-fast scales for diffusion processes, developed by \textit{K. Ball, T. Kurtz, L. Popovic} and \textit{G. Rempala} [Ann.\ Appl.\ Probab.\ 16, No. 4, 1925--1961 (2006; Zbl 1118.92031)]. Lastly, two typical examples ( (i) logistic physical-age and size-structured population model, and (ii) logistic biological-age and size-structured population model) illustrate the general theory discussed and developed in this paper, and some simulation results for each model are presented as well.