当前位置: X-MOL 学术J. Theor. Probab. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
On Hitting Time, Mixing Time and Geometric Interpretations of Metropolis–Hastings Reversiblizations
Journal of Theoretical Probability ( IF 0.8 ) Pub Date : 2019-04-16 , DOI: 10.1007/s10959-019-00903-2
Michael C. H. Choi , Lu-Jing Huang

Given a target distribution $$\mu $$ μ and a proposal chain with generator Q on a finite state space, in this paper, we study two types of Metropolis–Hastings (MH) generator $$M_1(Q,\mu )$$ M 1 ( Q , μ ) and $$M_2(Q,\mu )$$ M 2 ( Q , μ ) in a continuous-time setting. While $$M_1$$ M 1 is the classical MH generator, we define a new generator $$M_2$$ M 2 that captures the opposite movement of $$M_1$$ M 1 and provide a comprehensive suite of comparison results ranging from hitting time and mixing time to asymptotic variance, large deviations and capacity, which demonstrate that $$M_2$$ M 2 enjoys superior mixing properties than $$M_1$$ M 1 . To see that $$M_1$$ M 1 and $$M_2$$ M 2 are natural transformations, we offer an interesting geometric interpretation of $$M_1$$ M 1 , $$M_2$$ M 2 and their convex combinations as $$\ell ^1$$ ℓ 1 minimizers between Q and the set of $$\mu $$ μ -reversible generators, extending the results by Billera and Diaconis (Stat Sci 16(4):335–339, 2001 ). We provide two examples as illustrations. In the first one, we give explicit spectral analysis of $$M_1$$ M 1 and $$M_2$$ M 2 for Metropolized independent sampling, while in the second example, we prove a Laplace transform order of the fastest strong stationary time between birth–death $$M_1$$ M 1 and $$M_2$$ M 2 .

中文翻译:

关于大都会-黑斯廷斯可逆化的命中时间、混合时间和几何解释

给定目标分布 $$\mu $$ μ 和在有限状态空间上带有生成器 Q 的提议链,在本文中,我们研究了两种类型的 Metropolis–Hastings (MH) 生成器 $$M_1(Q,\mu )$连续时间设置中的 $M 1 ( Q , μ ) 和 $$M_2(Q,\mu )$$ M 2 ( Q , μ )。虽然 $$M_1$$ M 1 是经典的 MH 生成器,但我们定义了一个新的生成器 $$M_2$$ M 2 ,它捕获了 $$M_1$$ M 1 的相反运动,并提供了一套综合的比较结果,范围从击中时间和混合时间到渐近方差、大偏差和容量,这表明 $$M_2$$ M 2 比 $$M_1$$ M 1 具有更好的混合特性。为了看到 $$M_1$$ M 1 和 $$M_2$$ M 2 是自然变换,我们提供了 $$M_1$$ M 1 的有趣几何解释,$$M_2$$ M 2 及其凸组合为 $$\ell ^1$$ ℓ 1 Q 和 $$\mu $$ μ 可逆生成器集合之间的最小化器,扩展了 Billera 和 Diaconis (Stat Sci 16(4):335–339, 2001 )。我们提供两个例子作为说明。在第一个例子中,我们给出了用于 Metropolized 独立采样的 $$M_1$$ M 1 和 $$M_2$$ M 2 的显式谱分析,而在第二个例子中,我们证明了之间最快强平稳时间的拉普拉斯变换阶生-死 $$M_1$$ M 1 和 $$M_2$$ M 2 。
更新日期:2019-04-16
down
wechat
bug