1: \begin{abstract}

2: %BEGIN update 20110213

3: We study poor behavior of Markov chain

4: Monte Carlo methods in large sample framework.

5: We define weak consistency to measure the convergence rate 

6: of Monte Carlo procedure.

7: This property is studied by convergence of step Markov process 

8: to a diffusion process. 

9: We apply weak consistency to a 

10: popular data augmentation for simple mixture model.

11: The Monte Carlo method is known to work poorly when one of a  

12: mixture proportion is close to $0$. 

13: We show that it is not (local) consistent but (local) weak consistent. 

14: As an alternative, we propose a Metropolis-Hastings algorithm

15: which is local consistent for the same model. 

16: These results come from a weak convergence property of 

17: Monte Carlo procedures which is difficult to obtain from Harris recurrence approach.

18: %END update 20110213

19: \\

20: Keyword: Markov chain Monte Carlo; Asymptotic Normality; Diffusion process

21: \end{abstract}

22: