Brownian bridges to submanifolds (Q1787163): Difference between revisions

Let \(M\) be a complete, connected Riemannian manifold, \(N\subset M\) be a closed embedded submanifold and \(p^M_t(x,y)\) be the heat kernel of Brownian motion on \(M\). Denote by \(m\) and \(n<m\) the dimensions of \(M\) and \(N\), respectively. Set \[ p_t^M(x,N) := \int_N p_t^M(x,y) dv_N(y) \] whrere \(v_N\) is the canonical volume measure on \(N\) induced by the embedding. Let \(X(x)\) be Brownian motion on \(M\) starting at \(x\), fix \(T>0\) and set \[ \mathbb E\left[F\mid X_T(x)\in N\right] := \frac{\left[p_{T_t}^M(X_t(x),N) F\right]}{p_T^M(x,N)} \] (\(F\) is any bounded, adapted random variable); this induces a diffusion with generator \(\frac 12\Delta + \nabla\log p_{T_t}^M(\cdot,N)\) which is a Brownian bridge from \(x\) to the submanifold \(N\). It is shown that this is a semimartingale. Denote by \(r_N(x)\) the distance from \(x\) to \(N\) and consider the diffusion on \(M\) starting at \(x\) and ending at \(N\) by time \(T\) having the generator \(\frac 12 \Delta - \frac{r_N}{T-t}\frac{\partial}{\partial r_N}\) (the latter means differentiation in radial direction). This is a special case of a Brownian bridge from \(x\) to \(N\) called the Fermi bridge \(\hat X(x)\). The main result of this paper establishes the following formula \[ \begin{multlined} p_T^M(x,N) = (2\pi T)^{(m-n)/2} \exp\left[-r_N^2(x)/(2T)\right]\times\\ \times\lim_{t\uparrow T}\mathbb E\left[\exp\left\{\int_0^t \frac{r_N(\hat X_s(x))}{T-s}\big(d\mathsf A_s + d\mathsf L_s\big)\right\}\right]. \end{multlined} \] In this formula \(d\mathsf A\) is an absolutely continuous random measure which takes into account the geometry of \(M\) in between \(N\) and the cut locus, and \(d\mathsf L\) is a singular continuous random measure taking into account just the cut locus itself.