Marchenko–Pastur distribution explained

In the mathematical theory of random matrices, the Marchenko–Pastur distribution, or Marchenko–Pastur law, describes the asymptotic behavior of singular values of large rectangular random matrices. The theorem is named after soviet mathematicians Volodymyr Marchenko and Leonid Pastur who proved this result in 1967.

If

X denotes a m x n random matrix whose entries are independent identically distributed random variables with mean 0 and variance \sigma ^{2} < inf ty , letY _{n} = (1)/(n) X X ^{T}

and let

λ _{1,} λ _{2,} ... , λ _{m} be the eigenvalues of Y _{n} (viewed as random variables). Finally, consider the random measure\mu _{m} (A) = (1)/(m) \# \left\{ λ _{j} \in A \right\}, A \subset R . counting the number of eigenvalues in the subset A included in R .

Theorem. Assume that

m,n \to inf ty so that the ratio m/n \to λ \in ( 0 , +inf ty) . Then \mu _{m} \to \mu (in weak* topology in distribution), where\mu(A) = \begin{cases} ( 1- (1)/(λ) ) 1 _{0\in} + \nu(A),& if λ > 1 \\ \nu(A),& if 0 \leq λ \leq 1 , \end{cases} andd\nu(x) = (1)/(2 \pi \sigma ^{2}) (\sqrt{( λ _{+} - x)(x -λ _{-} ))/() } \,\mathbf_\, dxwithλ _{\pm} = \sigma ^{2(1} \pm \sqrt{ λ }) ^{2.}

The Marchenko–Pastur law also arises as the free Poisson law in free probability theory, having rate

1 / λ and jump size \sigma ^{2} .

Moments

For each

k \geq 1 , its k -th moment is

$$\backslash sum\_^\backslash frac\backslash binom\backslash binom\; \backslash lambda^\; =\; \backslash frac\backslash sum\_^\backslash binom\backslash binom\; \backslash lambda^$$

Some transforms of this law

The Stieltjes transform is given by

s(z) = (\sigma ^{2} ( 1-λ ) - z + \sqrt{(z - \sigma ^{2(λ} +1 )) ^{2-4λ} \sigma ^{4})/() }for complex numbers of positive imaginary part, where the complex square root is also taken to have positive imaginary part. The Stieltjes transform can be repackaged in the form of the R-transform, which is given byR(z) = (\sigma ^{2})/(1- \sigma ^{2} λ z)

The S-transform is given by

S(z) = (1)/(\sigma ^{2} ( 1+λ z)) .

Application to correlation matrices

For the special case of correlation matrices, we know that

\sigma ^{2=1} and λ= m/n . This bounds the probability mass over the interval defined by λ _{\pm} = \left( 1 \pm \sqrt{ (m)/(n}\right)) ^{2.}

Since this distribution describes the spectrum of random matrices with mean 0, the eigenvalues of correlation matrices that fall inside of the aforementioned interval could be considered spurious or noise. For instance, obtaining a correlation matrix of 10 stock returns calculated over a 252 trading days period would render

λ _{+= \left( 1+ \sqrt{ (10)/(252)} }\right)^2\approx 1.43. Thus, out of 10 eigenvalues of said correlation matrix, only the values higher than 1.43 would be considered significantly different from random.

See also

• Wigner semicircle distribution
• Tracy–Widom distribution

References

• Book: 2567175. Bai. Zhidong. Silverstein. Jack W.. Spectral analysis of large dimensional random matrices. Second edition of 2006 original. 2010. Springer. New York. Springer Series in Statistics. 978-1-4419-0660-1. 10.1007/978-1-4419-0661-8. 1301.60002.
• Epps . Brenden . Krivitzky . Eric M. . 2019 . Singular value decomposition of noisy data: mode corruption . Experiments in Fluids . 60 . 8 . 1–30 . 10.1007/s00348-019-2761-y . 2019ExFl...60..121E . 198436243 .
• Götze . F. . Tikhomirov . A. . 2004 . Rate of convergence in probability to the Marchenko–Pastur law . Bernoulli . 10 . 3 . 503–548 . 10.3150/bj/1089206408 . free .
• Marchenko . V. A. . Pastur . L. A. . 1967 . Распределение собственных значений в некоторых ансамблях случайных матриц . Distribution of eigenvalues for some sets of random matrices . ru . . N.S. . 72 . 114:4 . 507 - 536 . 10.1070/SM1967v001n04ABEH001994 . 1967SbMat...1..457M . Link to free-access pdf of Russian version
• Book: Nica . A. . Roland Speicher . Speicher . R. . 2006 . Lectures on the Combinatorics of Free probability theory . limited . Cambridge Univ. Press . 0-521-85852-6 . 204, 368 . Link to free download Another free access site
• Tulino. Antonia M.. Verdú. Sergio. Random matrix theory and wireless communications. Foundations and Trends in Communications and Information Theory. 1. 1. 1–182. 10.1561/0100000001. 2004. 1143.94303. Sergio Verdú.
• Zhang . W. . Abreu . G. . Inamori . M. . Sanada . Y. . 2011 . Spectrum sensing algorithms via finite random matrices . IEEE Transactions on Communications . 60 . 1 . 164–175 . 10.1109/TCOMM.2011.112311.100721 . 206642535 .