Small-time Gaussian behavior of symmetric diffusion semigroups (Q1431487)

The version \(\lim_{t\downarrow 0} t \log P_t(A,B)=-d(A,B)^2/2\) of Varadhan's formula for heat kernels (\(A,B\) being measurable sets) is generalized to arbitrary Dirichlet spaces for which the form \((\mathcal E,\mathbb D)\) is local and conservative. By definition, \(P_t(A,B):=\int_A T_t 1_B\,d\mu\), \((T_t)\) being the associated Markov semigroup and \((\Omega,\mathcal B,\mu)\) the underlying probability space. Moreover, \(d(A,B)\) denotes the supremum of all differences \(\text{ essinf}_{x\in B} f(x)-\text{ essinf}_{y\in A} f(y)\) (intrinsic distance), where \(f\in \mathbb D_0\), that is \(f\in \mathbb D\cap L^\infty\) such that \(2\mathcal E (fh,f)-\mathcal E(f^2,h)\leq \| h\| _{L^1}\) for any \(h\in \mathbb D\cap L^\infty\). Assuming \(\mu(A)>0\), it is shown that there is a largest positive function~\(d_A\) satisfying \(d_A=0\) a.e.\ on \(A\) and \(d_A\wedge N\in\mathbb D_0\) for any \(N\geq 0\) and that \(d(A,B)= \text{ essinf}_{x\in B} d_A(x)\). Further, the authors prove a convergence of \(t\log T_t1_A\) to \(-d_A^2/2\). Finally, the intrinsic distance is characterized by pointwise distance for canonical Dirichlet spaces on loop groups.