Asymptotic behavior of a system of interacting nuclear-space-valued stochastic differential equations driven by Poisson random measures (Q1334927)

The authors consider a countable Hilbertian nuclear space \(\Phi\), with the increasing sequence of norms \(\| \;\|_ n\), \(n \geq 0\), such that there exists a sequence \((h_ i) \subset \Phi\), complete orthogonal for every \(\| \|_ n\); denote by \(H_ n\) the \(\| \|_ n\) completion and extend, using the duals, this notation for \(n \leq 0\). They consider a measurable space \(U\) and a system \((t \in [0,T]\)) \[ \begin{multlined} X^ n_ j (t) = X^ n_ j (t) + \int^ t_ 0 \left(a (s,X^ n_ j (s)) + n^{-1} \sum^ n_{i = 1} b(s, X^ n_ j (s), X^ n_ i(s))\right)ds + \\ + \int^ t_ 0 \int_ U \left(g (s,X^ n_ j (s-), u) + n^{-1} \sum^ n_{i = 1} c(s, X^ n_ j (s-), X^ n_ i (s-), u)\right) \widetilde{N}_ j (du ds), j = 1,\dots, n,\end{multlined} \] \(N_ j\) being independent Poisson random measures with the same characteristic measure \(\mu\) on \(U\) and \(X^ n_ j (t)\) being \(\Phi'\)-valued random, and prove, under assumptions of ``continuity, coercivity, growth and monotonicity'' involving an index \(p_ 0 (T)\), together with \(b(t, v, w) = 0\) and \(c(t, v,w) = 0\) for \(\| v - w \|_{-p} > M(p)\) and all \(p\), and with \(E\| X^ n_ j (0)\|^ 2_{-r_ 0} < \infty\), the existence and unicity of the strong solution in \(H_{-p_ 1(T)}\) (\(H_{-\max(r_ 0, p_ 0(T))} \to H_{-p_ 1(T)}\) being Hilbert-Schmidt), by reducing to the case ``without \(b\) and \(c\)'' and with \(n = 1\). Having in mind \(n = 1, 2,\dots,\) they prove that if the initial conditions are exchangeable and \(\sup_ n E\| X^ n_ j (t) \|^ 2_{-r_ 0} < \infty\), then \[ \sup_ n E\biggl(\sup_{t \in [0,T]} \| X^ n_ j(t) \|^ 2_{-p(T)}\biggr) < \infty. \] If furthermore \(n^{-1} \sum^ n_{j = 1} \delta_{X^ n_ j (0)} \to \mu_ 0\) on \(\Phi'\), then the distribution \(\eta_ n\) (on the set of probabilities on \(D([0,T],H_{- p_ 1(T)}))\) of \(\zeta_ n = n^{-1} \sum^ n_{j = 1} \delta_{X^ n_ j (\cdot)}\) tends to \(\delta_{\lambda_ 0}\), where \(\lambda_ 0\) is the distribution of the unique solution \((X_ s)_{s \in [0,T]}\) of the (McKean-Vlasov) equation \[ \begin{multlined} X_ t = X_ 0 + \int^ t_ 0 \Bigl(a (s,X_ s) + \int b(s, X_ s, y) d{\mathcal D} (X_ s)(y)\Bigr) ds +\\ + \int^ t_ 0 \Bigl(g (s, X_{s-}, u) + \int c (s, X_{s-}, y, u) d{\mathcal D} (X_ s)(y)\Bigr)\widetilde{N}(duds)\end{multlined} \] (\(\mathcal D\) meaning distribution of), with \(\mu_ 0\) as ``initial condition''. Before that they prove an existence and unicity result for the McKean-Vlasov equation, under analogous conditions. The motivation comes from a model of random behavior of voltage potentials of spatially extended neurons described in a special section of the paper. The last section shows how in the particular case \(\Phi = R^ d\) one gets improvements of some known results.