arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv math/0703513math.PR
收藏

对称稳定概率分布的离差测度

Dispersion measure for symmetric, stable probability distributions

Jussi I. Tyhtila

摘要:

稳定分布是概率分布中一类有趣而重要的分布。它们是由Paul Lévy在1925年\cite{lk}明确发现的。它们具有许多有趣的性质,最重要的是,它们在加法下是定义不变的。值得注意的是,它们具有幂律类型的衰减,因此它们是模拟许多自然现象的优秀模型,例如地震,金融回报,以及众多的社会现象,例如城市和公司的规模分布\cite{scaling}。关于它们的主要问题是它们具有无穷的方差\cite{GK},因此它们的实际适用性有些有限。此外,它们通常不具有可以用解析形式表示的密度。这项研究提出了一个分散措施,从费舍尔信息,微分几何,最重要的是,傅立叶变换对\cite{Weyl}的不确定性原则的想法。本文首先简要讨论了特征函数及其与傅里叶变换的关系和性质,接着简要介绍了稳定分布,并定义了\textit{characteristic curvature}的概念,提出了它作为一类稳定分布的色散度量。特征曲率满足常见的稳定过程的标度性质,在高斯的情况下,减少到标准差,并明确相关的欧拉伽玛函数,赫斯特指数H和分形维数2-H。

英文摘要:

Stable distributions is an interesting and important class of probability distributions. They were discovered explicitly by Paul Lévy in 1925 \cite{lk}. They possess many interesting properties, most importantly they are by definiton invariant under addition, up to a scale. Noteworthly they have power-law type of decay and therefore they are an excellent model for modelling many natural phenomena, such as earthquakes, financial returns, and a multitude of social phenomena such as size distributions of cities and firms \cite{scaling}. The major problem concerning them is that they have an infinite variance \cite{GK} and therefore their practical applicability is somewhat limited. Also they generally do not possess a density expressible in an analytic form. This study proposes a dispersion measure for them, drawing ideas from Fisher information, differential geometry and most importantly, the uncertainty principle for Fourier transform pairs \cite{Weyl}. The study begins with a brief discussion on characteristic functions and their relation to Fourier transforms and their properties, proceeds to a brief presentation of stable distributions and accumulates in defining a concept of \textit{characteristic curvature}, which is proposed as a suitable measure of dispersion for class of stable distributions. Characteristic curvature satisfies the familiar scaling property for stable processes, reduces to that of standard deviation in the Gaussian case and is explicitly related to the Euler Gamma function, the Hurst exponent H and the fractal dimension 2-H.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑