1: \begin{abstract}

2: It is known that the simple slice sampler has very robust convergence

3: properties, however the class of problems where it can be implemented

4: is limited.

5: In contrast, we consider hybrid slice samplers which are easily implementable and 

6: where another Markov chain approximately samples the 

7: uniform distribution on each slice. 

8: Under appropriate assumptions on the Markov chain on the slice we 

9: show a lower bound and an upper bound of the spectral gap of the hybrid slice sampler 

10: in terms of the spectral gap of the simple

11: slice sampler. An immediate consequence of this is that spectral gap and geometric ergodicity

12: of the hybrid slice sampler can be concluded from spectral gap and geometric ergodicity

13: of its simple version which is very well understood.

14: These results indicate that robustness properties of the

15: simple slice sampler are inherited by (appropriately designed) easily

16: implementable hybrid versions and provide the first theoretical underpinning of

17: their use in applications.

18: We apply the developed theory and analyse a number of specific algorithms 

19: such as the stepping-out shrinkage slice sampling, hit-and-run slice

20: sampling on very general multidimensional targets

21: and an easily implementable combination of both procedures on fairly general and realistic multidimensional bimodal

22: densities.

23: \end{abstract}

24: