Short-time geometry of random heat kernels

From MaRDI portal





Let us consider the stochastic partial differential equation \[ du= \textstyle {{1\over 2}} \Delta udt+ (\sigma,\nabla u) \circ dW_t, \] where \(M\) is a \(C^\infty\), compact, and connected \(d\)-dimensional Riemannian manifold with Riemannian metric \((\cdot, \cdot)\); \(\Delta\) and \(\nabla\) are the associated Laplace-Beltrami and gradient operators, resp., and \(\sigma\) is a \(C^\infty\) vector field on \(M\); \(W_t\) is a Brownian motion which is defined on probability space \((\Omega, F,P)\) and \(\circ dW_t\) is the Stratonovich stochastic differential. The heat kernel is the solution of \[ dp_y =\textstyle {{1\over 2}} \Delta p_ydt+ (\sigma, \nabla p_y) \circ dW_t, \quad p_y (0,\cdot) =\delta_y, \] where \(\delta_y\) is the Dirac measure concentrated at \(y\in M\). The main goal of this monograph is to find a geometric expansion for \(p_y\). Let \(\text{T}M\) be the tangent manifold of \(M\), and let \(\| \cdot \|\) be the fibre metric on \(\text{T}M\) defined by \(( \cdot, \cdot)\). Let \(d:M^2 \to R_+\) be the Riemannian distance function, \(\text{Cut} (y)\) be the cut-locus of \(y\), and set \(O_y= M\sim \text{Cut} (y)\). For every \(t>0\) and \(x\in O_y\), let \(\gamma_y^{t,x} :[0,t]\to M\) be the shortest geodesic which runs from \(y\) to \(x\) in time \(t\), and let \(\{\dot \gamma_y^{t,x} (s):0 \leq s\leq t\}\) be its velocity. Let \(\Theta_y: O_y\to R\) be the square root of the reciprocal of the Ruse invariant at \(y\), \(\overline \nabla\) be the Levi-Civita connection, \(R_{ \text{op}}\) be the curvature operator and let \(R\) be the curvature tensor. Finally, let \(\overline\nabla^* \sigma\) be the adjoint of the map \(X\to\overline \nabla_x \sigma\) from \(\text{T} M\) to itself. Let us define \[ \begin{multlined} G_y(t,x)= \textstyle{ {1\over 2}}\int^t_0\bigl \|\Xi_y^{2, t,x} (s)\bigr\|^2 ds+\\ +\textstyle {{1 \over 2}} \int^t_0 R\bigl(\Xi_y^{1,t,x} (s),\;\dot \gamma_y^{t,x} (s)\bigr)ds+\int^t_0 \bigl(\overline \nabla_{\gamma_y^{t,x} (s)}\sigma ,\Xi_y^{1,t,x} (s)\bigr) \circ dW_s, \end{multlined} \] where \(\Xi_y^{1,t,x} (s)\) and \(\Xi_y^{2,t,x} (s)\) satisfy the coupled linear stochastic differential equations \[ D_s \Xi_y^{1,t,x} (s) =\Xi_y^{2,t,x} (s)ds- \sigma \bigl( \gamma_y^{t,x} (s)\bigr) \circ dW_s, \] \[ D_s\Xi_y^{2,t,x}(s)=R_{\text{op}}\bigl(\dot\gamma_y^{t,x}(s),\;\Xi_y^{1,t,x}(s)\bigr)\dot\gamma_y^{t,x}(s)ds+\overline\nabla_{\dot\gamma_y^{t,x} (s)}\sigma\circ dW_s,\;\Xi_y^{1,t,x}(0)=0_y,\;\Xi_y^{1,t,x}(t)=0_x, \] here \(0_x\) and \(0_y\) are respectively the zero elements of tangent space over \(x\) and \(y\), \(D_s\) denotes the stochastic covariant derivative along the deterministic curve \(\gamma_y^{t, x}\). The main result of this monograph is the following. For each \(\kappa \in (0,1/2)\) we have \[ p_y(t,x) =\widehat P_y (t,x) \bigl(1+O (t^{1/2-\kappa}) \bigr)\;P\text{-a.s. as } t\to 0, \] for \(y\in M\) and \(x\in O_y\). Here \[ \widehat P_y (t,x)= \exp\left[ -{d^2(x,y) \over 2t}-\int^t_0 \biggl(\dot \gamma_y^{t,x} (s), \sigma\bigl( \gamma_y^{t,x} (s)\bigr) \biggr)\circ dW_s-G_y(t,x) \right] \Theta_y(x) (2\pi t)^{-d/2}. \] Finally, the author finds an analog of the Minakshisundaram-Pleijel expansion of \(p_y\) in iterated stochastic integrals.











This page was built for publication: Short-time geometry of random heat kernels

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3840052)