1: \begin{abstract}

2: 		Component-wise MCMC algorithms, including Gibbs and conditional

3: 		Metropolis-Hastings samplers, are commonly used for sampling from

4: 		multivariate probability distributions.  A long-standing question

5: 		regarding Gibbs algorithms is whether a deterministic-scan

6: 		(systematic-scan) sampler converges faster than its random-scan

7: 		counterpart.  We answer this question when the samplers involve two

8: 		components by establishing an exact quantitative relationship between

9: 		the $L^2$ convergence rates of the two samplers.  The relationship

10: 		shows that the deterministic-scan sampler converges faster.  We also

11: 		establish qualitative relations among the convergence rates of

12: 		two-component Gibbs samplers and some conditional Metropolis-Hastings

13: 		variants.  For instance, it is shown that if some two-component conditional

14: 		Metropolis-Hastings samplers are geometrically ergodic, then so are

15: 		the associated Gibbs samplers.

16: 	\end{abstract}