Plamen Koev: Accurate Eigenvalues and SVDs of (singular) totally nonnegative matrices with applications to Computer Aided Geometric Design.

Abstract: In Computer Aided Geometric Design, totally nonnegative bases present good shape properties and B-bases correspond to the bases with optimal shape preserving properties. In this talk we will present the algorithms for performing all matrix computations with totally nonnegative matrices to high relative accuracy in floating point arithmetic. The complexity is comparable to that of the traditional LAPACK algorithms, which deliver no such accuracy. In particular, these algorithms compute all eigenvalues, SVDs, various decompositions, and totally-nonnegative preserving transformations, such as products, submatrices, etc. Most significantly, we are also able to compute exactly the Jordan blocks corresponding to zero eigenvalues--which we believe is the first example of a Jordan structure being computed accurately. Our algorithms are based on the fact that when the computations are performed in a way that preserved the total nonnegativity, no subtractions are encountered and the relative accuracy is preserved.

References: [1] Carnicer J.M., Pe??a J.M. (1996) Total positivity and optimal bases. In: Gasca M., Micchelli C.A. (eds) Total Positivity and Its Applications. Mathematics and Its Applications, vol 359. Springer, Dordrecht.
[2] Farin, G., Curves and Surfaces for CAGD, Elsevier, 2001.
[3] Goodman, T.N.T., Siad, H.B., Shape preserving properties of the generalized Ball basis, Computer-Aided geometric design 8 (1991), 115-121.
[4] Karlin, S., Total Positivity, Vol. I, Stanford, 1968.
[5] Koev, P., Accurate computations with totally nonnegative matrices SIAM J. Matrix Anal. Appl. 29 (2007), 731-751.
[6] Koev, P., Accurate eigenvalues and zero Jordan blocks of totally nonnegative matrices, Preprint.