1: \begin{abstract}

2:   Selection among alternative theoretical models given an observed

3:   data set is an important challenge in many areas of physics and

4:   astronomy.

5:   %ranging from gravitational-wave data analysis to

6:   %measurements of black-hole masses in X-ray binaries.

7:   Reversible-jump Markov chain Monte Carlo (RJMCMC) is an extremely

8:   powerful technique for performing Bayesian model selection, but it

9:   suffers from a fundamental difficulty: it requires jumps between

10:   model parameter spaces, but cannot retain a memory of the favored

11:   locations in more than one parameter space at a time.  Thus, a naive

12:   jump between parameter spaces is unlikely to be accepted in the MCMC

13:   algorithm and convergence is correspondingly slow.  Here we

14:   demonstrate an interpolation technique that uses samples from

15:   single-model MCMCs to propose inter-model jumps from an

16:   approximation to the single-model posterior of the target parameter

17:   space.  The interpolation technique, based on a kD-tree data

18:   structure, is adaptive and efficient in arbitrary dimensions.  We

19:   show that our technique leads to dramatically improved convergence

20:   over naive jumps in an RJMCMC, and compare it to other proposals in

21:   the literature to improve the convergence of RJMCMCs.  We also

22:   discuss the use of the same interpolation technique in two other

23:   contexts: as a convergence test for a single-model MCMC and as a way

24:   to construct efficient ``global'' proposal distributions for

25:   single-model MCMCs without prior knowledge of the structure of the

26:   posterior distribution.

27: \end{abstract}

28: