A Monte Carlo approach to estimating the spectrum of large positive semidefinite matrices.
About this article
Abstract
The method presented in this paper aims at furnishing a series of numerical values to approximate the matrix eigenvalues. The method is based on a statistical model that involves the first three central moments of the eigenvalue distribution, and it relies on the solution of a nonlinear system of equations that implements the moment-matching method and on a subsequent procedure of Monte Carlo simulations. The method is only applicable to real positive semidefinite matrices (PSD), and it is especially useful when other techniques lead to computational problems, e.g., when the matrices become too large to be processed or the required storage space sets heavy limits to the computational process.
References
[1] Y. Saad, Numerical Methods for Large Eigenvalue Problems, SIAM (2011); available online:
http://www-users.cs.umn.edu/~saad/eig_book_2ndEd.pdf
[2] B.C. Arnold, N. Balakrishnan, H.N. Nagaraja, A First Course in Order Statistics, SIAM (2008)
[3] F.R. Gantmacher, The Theory of Matrices, Chelsea Publ. Company (1959), page 84, vol. I
[4] E. Floyd, Handbook of Applicable Mathematics, vol II: Probability, John Wiley & S. (1980)
View more references (3)
[5] N. Cristianini, J. Shawe-Taylor, An Introduction to Support Vector Machines, Cambridge University Press (2000), pp. 42-44
[6] N. El Karoui, Spectrum Estimation for Large Dimensional Covariance Matrices Using Random Matrix Theory, Ann. Statist. (2008), Volume 36, Number 6, pp. 2757-2790
[7] J.S. Ramberg, E.J. Dudewicz, P.R. Tadikamalla, E.F. Mykytka, A probability distribution and its uses in fitting data, Technometrics (1979), vol. 21, No. 2, pp. 201-214.