1: \begin{abstract} %{{{

2: Parallel tempering is a generic Markov chain Monte Carlo sampling

3: method which allows good mixing with multimodal target distributions,

4: where conventional Metropolis-Hastings algorithms often fail. The mixing

5: properties of the sampler depend strongly on the choice of tuning

6: parameters, such as the temperature schedule and the proposal

7: distribution used for local exploration.  We propose an adaptive

8: algorithm which tunes both the temperature schedule and the parameters

9: of the random-walk Metropolis kernel automatically. We prove the

10: convergence of the adaptation and a strong law of large numbers for

11: the algorithm. We illustrate the performance of our method with

12: examples. Our empirical findings indicate that the algorithm can cope

13: well with different kind of scenarios without prior tuning.

14: \end{abstract}