On a coupling of solutions to the interface stochastic differential equation on a star graph (Q1721905)

From MaRDI portal
scientific article
Language Label Description Also known as
English
On a coupling of solutions to the interface stochastic differential equation on a star graph
scientific article

    Statements

    On a coupling of solutions to the interface stochastic differential equation on a star graph (English)
    0 references
    0 references
    0 references
    13 February 2019
    0 references
    A metric star graph $(G,d)$ with $N$ rays $E_k$ and origin $0$ is a metric space such that $E_j\cap E_k = \{0\}$ if $k\neq j$, and for each $k$ there is an isometry $e_k:[0,\infty)\to E_k$; $d$ is the geodesic distance on $G$ and $\epsilon : G\setminus\{0\} \to \{1,\dots, N\}$ tells us in which ray $E_k$ the value $0\neq x\in G$ can be found. This type of graph (as subset of $\mathbb{C}$) was suggested by \textit{J. B. Walsh} [``A diffusion with a discontinuous local time'', Astérisque 52--53, 37--45 (1978; \url{http://www.numdam.org/article/AST_1978__52-53__37_0.pdf})] and he considered a Brownian motion on this set. The present authors study the following stochastic equation \[ f(X_t)=f(x) + \sum_{k=1}^N \int_0^t f'(X_s) 1_{E_k}(X_s)\,dW_s^k + \frac 12\int_0^t f''(X_s)\,ds,\quad X_0=x\in G\tag{*} \] driven by a standard Brownian motion $W= (W^1_t,\dots,W_t^N)_{t\geq 0}$ on $\mathbb{R}^N$ and where $f$ are $C^2$-functions on $G$ satisfying, together with their first derivatives, an interface condition at the origin $x=0$. The solution of (*) is a pair $(X,W)$ where $X$ is an $G$-valued process and $W$ is the driving Brownian motion. Moreover, a Wiener coupling $(X,Y,W)$ of the solutions to (*) are two solutions $(X,W)$ and $(Y,W)$ on the same filtered probability space such that $X$ and $Y$ are conditionally independent given $W$. There are no such couplings if $N\leq 2$. The authors show that, if $N\geq 3$, any Wiener coupling with $X_0=Y_0=0$ has the following properties \begin{itemize} \item $\mathbb E d(\bar X_t,\bar Y_t) = 2\frac {(N-2)}N \sqrt{\frac{2t}{\pi}}$ where $\bar X_t^k = f^k(X_t)$ with $f^k(x)=\frac{|x|}{Np_k} 1_{E_k}(x)$; \item $\mathbb P(X_t=Y_t)=0$ and $\mathbb P(g_t^X = g_t^Y)=0$ where $g_t^X$ and $g_t^Y$ are the last zeros of $X$ and $Y$ before time $t$; \item $\epsilon(X_t)$ and $\epsilon(Y_t)$ are independent. \end{itemize} Couplings of this type have been introduced by the first author and \textit{O. Raimond} [Stochastic Processes Appl. 126, No. 1, 33--65 (2016; Zbl 1330.60072)] and the present work is inspired by \textit{B. Tsirelson}'s proof that the filtration of Walsh's Brownian motion on a metric graph with $N\geq 3$ rays is not a Brownian filtration coming from a standard Brownian motion [Geom. Funct. Anal. 7, No. 6, 1096--1142 (1997; Zbl 0902.31004)].
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    Tsirelson theorem
    0 references
    non-Brownian filtration
    0 references
    Walsh Brownian motion
    0 references
    stochastic equations on graphs
    0 references
    0 references
    0 references