Numerical Study of Random Correlation Matrices : Finite-Size Effects

Numerical Study of Random Correlation Matrices : Finite-Size Effects
复制标题

随机相关矩阵的数值研究:有限尺寸效应

DOI:
10.1007/978-3-642-22194-1_55
复制
发表时间:
2011
期刊:
Smart Innovation, Systems and Technologies (Springer)
影响因子:
--
通讯作者:
Yuta Arai
Yuta Arai
中科院分区:
--
文献类型:
--
作者:
日比谷孟俊;佐藤悟;内田保廣;Yuta Arai

文献摘要

相似文献

本文给出了随机相关矩阵的最大特征值分布的数值计算。这样一个广泛的研究,使我们能够制定出经验公式的平均值和标准偏差的最大特征值,这是准确的参数范围很广。作为这些公式的应用,我们提出了一个标准,挑选出统计上有意义的相关性的主成分分析。新准则将有限尺寸效应引入到基于随机矩阵理论的现有方法中,在无限尺寸极限下给出了精确结果。
We report the numerical calculations of the distribution of maximal eigenvalue for various size of random correlation matrices. Such an extensive study enables us to work out empirical formulas for the average and standard deviation of the maximal eigenvalue, which are accurate in a wide range of parameters. As an application of those formulas, we propose a criterion to single out statistically meaningful correlations in the principal component analysis. The new criterion incorporates finite-size effects into the current method based on the random matrix theory, which gives the exact results in the infinite-size limit.