1: \begin{abstract}

2:   We consider a continuous analogue of the simulated annealing algorithm in $\R^d$, namely the solution of the SDE $dX_t = \sigma(t) dB_t - \grad V(X_t)dt$, where $V$ is a function called potential. We prove a convergence result, similar to the one in \cite{Mic91}, under weaker hypotheses on the potential function. In particular, we cover cases where the gradient of the potential goes to zero at infinity. The main idea is to replace the Poincaré and log-Sobolev inequalities used in \cite{Mic91,HCS87} by weak Poincaré inequalities (introduced in \cite{RW01}), and to estimate constants with measure-capacity criteria. We show that the convergence still holds for the ``classical'' schedule $\sigma(t) = c/\ln(t)$, where $c$ is bigger than a constant related to $V$.

3: 

4:   Keywords: simulated annealing, diffusion process, weak Poincaré inequality

5: 

6:   MSC 2000: 90C59, %

7:   60J60, %

8:   60F99 %

9: \end{abstract}

10: