Getting my Institute of Mathematical Statistics
Fellows plaque from Jim Pitman, at the big statistics
meeting in Salt Lake City, July 30, 2007.

My fellow ``Fellows''.

Office: 4214 SAS Hall

Phone: (919)515-7864

## Interests

### Probability and mathematical statistics

### Spectral properties of large dimensional random matrices

I'll try to explain what the above graphs represent. The histogram in
the first one is that of the eigenvalues of a sample covariance matrix
(s.c.m.) formed from 4000 samples of a 200 dimensional random vector whose
population covariance matrix has 3 distinct eigenvalues:
1, 3, and 10, with respective multiplicities 40, 80, and 80. The
assumptions on the random vector are general enough to include multivariate
normals. The function you see in both graphs is the limiting density
(whose existence and properties are part of my work) of the empirical
distribution of the eigenvalues of s.c.m. as the dimension
and sample size approach infinity, the population eigenvalues remain at
(1,3,10) with relative proportions (.2,.4,.4), and the ratio of dimension
to sample size is .05. The closeness of the limiting density to the
histogram displays several things: 1) the validity of the above mentioned
``limit theorem'', 2) the ``curse of high dimensionality'', namely that, even
though 4000 samples were used to estimate the population matrix, because
of the dimension of the random vector, the estimation is not so great, and
3) since the histogram of eigenvalues is predictable from the limit
theorem and because of an understanding of how the limiting
density should look, due to the relatively small ratio of dimension
to sample size (.05), the histogram does reveal quite a bit of information
on the eigenvalues of the population matrix (after all, there are 3 clumps
centering around the population eigenvalues).

The second graph also contains a scatter plot of all 200 eigenvalues,
in clumps of (from left to right) 40, 80, and 80, demonstrating
Z.D. Bai and my previous results (see references 32 and 33 in publications): no
eigenvalues where they shouldn't be, and the
right number where they should be. These two facts,
although thoroughly reasonable from the above limit theorem, are a bit
more complicated to prove. The limit theorem only guarantees proportion
of eigenvalues are converging to the right numbers.

# Announcing

## of my latest paper with Wlodek Bryc

*Singular Values of Large Non-central Random Matrices
*

## of my paper with Katarzyna Bryc and Wlodek Bryc

*Separation of the Largest Eigenvalues in Eigenanalysis of
Genotype Data from Discrete Subpopulations*

## of my paper with Romain Couillet and
Frédéric Pascal

*Robust M-Estimation for Array Processing: A Random Matrix
Approach*

## of my paper with Walid Hachem, Malika Kharouf, and Jamal
Najim

*A CLT for Information-theoretic Statistics of Non-centered
Gram Random Matrices*

## of my paper with Zhidong Bai

*No Eigenvalues Outside the Support of the Limiting Spectral
Distribution of Information-Plus Noise Type Matrices*

## of my paper with Romain
Couillet, Zhidong Bai, and Merouane Debbah

*Eigen-inference for Energy Estimation of Multiple Sources*

## and

## the book Zhidong Bai and I have written:

## Second Edition

## Springer Series in Statistics, 2009

### MA 544: The course I offer every spring

**My favorite group**

**The best children's book series **

Write to me at jack at math ncsu edu

Department of Mathematics, Box 8205

North Carolina State University

Raleigh, NC 27695-8205
(February, 2018)
You are visitor number: