No label defined (Q4955462)

Let \(\{x_t\}_{t\geq 0}\) be a Brownian motion with values in \(R^d\) and \(D\subset R^d\) a bounded Borel set. Consider \(u(t,x)=E_x\{f(x_t)e^{\int_0^tV(x_s)ds};1(T_D>t)\}\) where \(T_D\) is the exit time of \(D\). It is well known, Feynman-Kac formula, that \(u\) satisfies the partial differential equation \(\dot u=-\triangle u+Vu\) and \(u(0,\cdot)=f(\cdot)\) and \(u=0\) outside of \(D\). In this remarkable paper the author proves several large deviation results for the quantity \(P(\frac{1}{t}\int_0^tV(x_s)ds \in A)\). Two scales are needed, first the ``natural scaling'' \(\lambda_t=t\) and secondly scales verifying \(\lambda_t/t=o(1)\). In the first case the author proves that \(\frac{1}{t}\log P(\frac{1}{t}\int_0^tV(x_s) ds \in A)\approx -\inf_{a\in A}I(a)\). The entropy function \(I(a)\) is not necessarily convex. To prove this result the author must obtain bounds for \(\int_D E_x(e^{\lambda \int_0^tV(x_u) du};1(T_D>t))dx\) using the spectral properties of the self-adjoint operator \(D=\triangle+V\). In Theorem 3.1 it is shown a first large deviation result for open sets and in Theorem 3.2 he gets the reciprocal result for closed sets. As a bonus he obtains that \(\frac{1}{t}\int_0^tV(x_s)ds\) converges exponentially fast to \(h=c'(0)\). The function \(c(\lambda)\) is defined as \(c(\lambda)= \sup_{f\in C}\lambda(Vf,f)-I(f)\) where \(C\) is the set of infinitely differentiable functions vanishing at infinity and \(I(f)=(-\triangle f, f)\). The result holds when \(c\) has a derivative in zero, this is the case when \(V\) has a bounded support or when \(V\) is quasiperiodic. Furthermore the derivatives can be computed. The problem is different if \(V\) is a random ergodic process defined as \(V(x,\omega)=v(\tau_x\omega)\), where \((\Omega,{\mathcal B},\{\tau_x\}_{x \in R^d}, \mu)\) is a dynamical system and \(v\) is bounded and measurable. The scaling must be modified and the results are different for \(d=1\) and \(d\geq 2\). For the entire collection see [Zbl 0937.00066].