1: \begin{abstract}

2: While statisticians are well-accustomed to performing exploratory analysis in

3: the modeling stage of an analysis, the notion of conducting preliminary

4: general-purpose exploratory analysis in the Monte Carlo stage (or more

5: generally, the model-fitting stage) of an analysis is an area which we feel

6: deserves much further attention.  Towards this aim, this paper proposes a

7: general-purpose algorithm for automatic density exploration.  The proposed

8: exploration algorithm combines and expands upon components from various

9: adaptive Markov chain Monte Carlo methods, with the Wang-Landau algorithm at

10: its heart.  Additionally, the algorithm is run on interacting parallel chains

11: -- a feature which both decreases computational cost as well as stabilizes the

12: algorithm, improving its ability to explore the density.  Performance of this

13: new parallel adaptive Wang-Landau (PAWL) algorithm is studied in several

14: applications.  Through a Bayesian variable selection example, the authors

15: demonstrate the convergence gains obtained with interacting chains.  The

16: ability of the algorithm's adaptive proposal to induce mode-jumping is

17: illustrated through a Bayesian mixture modeling application.  Lastly, through a

18: 2D Ising model, the authors demonstrate the ability of the algorithm to

19: overcome the high correlations encountered in spatial models. The appendices

20: contain the full algorithmic description in pseudo-code, a tri-modal toy example and 

21: remarks on the convergence of the proposed algorithm.

22: \end{abstract}

23: