Consistent families of Brownian motions and stochastic flows of kernels (Q837999)

The paper has 8 paragraphs, the first is an introduction, containing also an intuitive description of the concepts and results. 2-7 are devoted to the proof of the existence and uniqueness in law of the following martingale problem. Let \(N\geq 1\), \(\theta (k,r)\geq 0\), \(k,r\geq 0\), \(\theta (k,r)=\theta (k+1,r)+\theta (k,r+1)\). \({\mathbb R}^{N}\) is split into ``cells'', each being defined by a weak (\(i\prec j\), \(j\prec i\) do not imply \(i=j\)) total ordering \(\prec\) of \(\{1,\dots,N\} \) as the set of all \((x_{i})\) with \(x_{i}\leq x_{j}\) for \(i\prec j\). The diagonal \(D\) is one of the cells. For every \(I,J\subset \{1,\dots,N\}\), \(I\cap J=\emptyset\), \(I\cup J\neq \emptyset\), consider \(v_{IJ}\in {\mathbb R}^{N}\), \((v_{IJ})_{i}=0,1,-1\) for \(i\notin I\cup J\), \(i\in I\), \(i\in J\) respectively. Every \(x\in {\mathbb R}^{N}\) determines a partition \(\pi (x)\) of \(\{1,\dots,N\} \), via the equivalence relation \(x_{i}=x_{j}\) . Let \({\mathcal V}(x)\) be the set of all \(v_{IJ}\) for which \(I\cup J\) is an element of \(\pi (x)\). If \(v_{IJ}\in {\mathcal V}(x)\), \(I,J\neq \emptyset\), then \(v_{IJ}\) ``points'' to a cell \(E(v_{IJ})\), depending only on \(\pi (x)\), let \({\mathcal V}_{+}(x)\) be the set of such \(v_{IJ}\). Let \(\theta (v_{IJ})=\theta (|I|,|J|)\). Let \(L_{N}\) be the set of all continuous \(f:{\mathbb R}^{N}\rightarrow {\mathbb R}\), linear on each cell. For every \(f\in L_{N}\) let \((A_{N}^{\theta } f)(x)= \sum_{v\in {\mathcal V}(x)}\theta (v)\nabla_{v} f(x)\). A solution of the martingale problem is a continuous \({\mathbb R}^{N}\)-valueed process \(X(t)=(X_{i}(t))\), \(t\geq 0\), such that, for every \(f\in L_{N}\), \(f(X(t))-\int_{_{0}}^{^{t}}(A_{N}^{\theta }f)(X(s))ds\) is a martingale and the brackets \(<X_{i},X_{j}>(t)\) are \(\int_{_{0}}^{^{t}}1_{(X_{i}(s)=X_{j}(s))}ds\). The proof of the unicity relies on some inequalities about the brownian motion on \({\mathbb R}\times [0,\infty )\), established in 7 and starts with \(\lim_{\varepsilon \rightarrow 0}\varepsilon^{-1}E(T_{\varepsilon })=(2\sum_{v\in {\mathcal V}_{+}(D)}\theta (v))^{-1}\), \(\lim_{\varepsilon \rightarrow 0}P(X(T_{\varepsilon })\in E(u))= \theta (u)(\sum_{v\in {\mathcal V}_{+}(D)}\theta (v))^{-1}\), where \(x\in D\), \(X\) is a solution of the martingale problem and \(T_{\varepsilon }=inf\) of \(t\) with \(|X_{i}(t)-X_{j}(t)|\geq \varepsilon\) for some \(i,j\) and \(u\in {\mathcal V}_{+}(D)\). First the case \(N=3\) is treated. It continues with an induction, the induction step showing that the distribution of the process until reaching \(D\) is determined, then the same for its projection on \(\sum x_{i}=0\) and then considering a conditional distribution. For the existence, \(\theta (k,r)\) is represented as \(\lim n^{1/2}(p_{n}(k,r)-(1_{(k=0)}+1_{(r=0)} )/2)\), \(p_{n}\) having the properties of \(\theta\), the solution is the limit in law of \(n^{- 1/2} Y^{n}(n t)\), \(Y^{n}(t)\) being Markov, \({\mathbb Z}^{N}\)-valued, with generator \(({\mathcal G}_{p_{n}}^{N} f)(x)= \sum_{v\in {\mathcal V}(x)}p_{n}(v)(f(x+v)-f(x))\). The topic of the last paragraph (8) starts with the following construction. Let \(\mu\) be a probability on \([0,1]\), \(\Lambda\) a Poisson point process on \({\mathbb R}\times {\mathbb Z}\) corresponding to the uniform measure and an indepenednt family \(R(t,x)\), \((t,x)\in \Lambda\), of \(\mu\)-distributed random variables. Consider a particle moving on \({\mathbb Z}\), jumping, when in \(x\), only at times \(t\) with \((t,x)\in \Lambda\), into \(x+1\) with probability \(R(t,x)\) or into \(x-1\). Let, for \(s\leq t\), \(K_{st}^{\mu }(x,\cdot )\) be the distribution of the position of the particle at time \(t\), when starting from \(x\) at time \(s\). Then \(K_{st}^{\mu }(x,A)\) are random variables, \(K_{st}^{\mu } K_{tu}^{\mu }= \)K\(_{su}^{\mu }\) a.s, \(K_{t_{i},t_{i+1}}^{\mu }\) are independent for \(t_{i}<t_{i+1}\), \(i=1,\dots,n\), \(K_{s,t}^{\mu }\) and K\(_{s+u,t+u}^{\mu }\) having the same distribution (shortly, \(K_{st}^{\mu }\) is a stochastic flow of kernels on \({\mathbb Z}\)). For every stochastic flow of kernels \(K\) on \(F\) consider the semigroup on \(F^{N}\) defined by \(P_{t}^{N}((x_{i}),\prod A_{i})=E(\prod K_{0,t}(x_{i},A_{i}))\). The authors prove that the semigroup \(P^{N}\) of \(K^{\mu }\) has as generator the \({\mathcal G}_{p}^{N}\) constructed with \(p(k,r)=\int x^{k}(1-x)^{r}d\mu (x)\). If \(\int xd\mu_{n}(x)=1/2\) and \(n^{1/2}x(1-x)\mu_{n}(dx)\rightarrow \nu\) weakly then \(K_{ns,nt}^{^{\mu }n}(n^{1/2}x,n^{1/2}A)\) converge in distribution to a flow \(K^{\theta }\) on \({\mathbb R}\) for which the \(P_{t}^{N}\) is the semigroup of the solution of the martingale problem with \(\theta (k,r)= \int x^{k-1}(1-x)^{r-1}d\nu (x)\), \(\theta (1,0)=\theta (0,1)=0\).