1: \begin{abstract}

2: The multiple-try Metropolis (MTM) algorithm is an extension of the Metropolis-Hastings (MH) algorithm by selecting the proposed state among multiple trials according to some weight function.

3: Although MTM has gained great popularity owing to its faster empirical convergence and mixing than the standard MH algorithm, 

4: its theoretical mixing property is rarely studied in the literature due to its complex proposal scheme. 

5: We prove that MTM can achieve a mixing time bound smaller than that of MH by a factor of the number of trials under a general setting applicable to high-dimensional model selection problems with discrete state spaces.  

6: Our theoretical results motivate a new class of weight functions called \textit{locally balanced weight functions} and guide the choice of the number of trials, 

7: which leads to improved performance over standard MTM algorithms.

8: We support our theoretical results by extensive simulation studies and real data applications with several Bayesian model selection~problems.

9: \end{abstract}

10: