1: \begin{abstract}

2: Particle Markov Chain Monte Carlo methods (PMCMC) \citep{andrieuetal2010}

3: are used to carry out inference in non-linear and non-Gaussian state space

4: models, where the posterior density of the states is approximated using

5: particles. Current approaches usually carry out Bayesian inference using a

6: particle Marginal Metropolis-Hastings algorithm, a particle Gibbs sampler,

7: or a particle Metropolis within Gibbs sampler. Our article gives a general

8: approach for constructing sampling schemes that converge to the target

9: distributions given in \cite{andrieuetal2010} and \cite{olssonryden2011}.

10: Our approach shows how the three ways of generating variables mentioned

11: above can be combined flexibly. The advantage of our general approach is

12: that the sampling scheme can be tailored to obtain good results for

13: different applications. We investigate the properties of the general

14: sampling scheme, including conditions for uniform convergence to the

15: posterior. We illustrate our methods with examples of state space models

16: where one group of parameters can be generated in a straightforward manner

17: in a particle Gibbs step by conditioning on the states, but where it is

18: cumbersome and inefficient to generate such parameters when the states are

19: integrated out. Conversely, it may be necessary to generate a second group

20: of parameters without conditioning on the states because of the high

21: dependence between such parameters and the states. A particularly important

22: case that is illustrated by our examples occurs when there are many unknown

23: parameters. In this case it is usually highly inefficient to generate all

24: the parameters without conditioning on the states and it is preferable to

25: generate as many parameters as possible using particle Gibbs and only

26: generate those parameters that are highly correlated with the states using

27: particle Marginal Metropolis-Hastings.

28: \end{abstract}