
Random Perturbation of Toeplitz Matrices
Charles Bordenave of the Institute for Advanced Study examines why computers produce large numerical errors when calculating eigenvalues of badly conditioned matrices. He traces the idea back to a 1947 paper by John von Neumann and Herman Goldstine, written while they developed the IAS computing machines, which proposed modeling rounding and computational noise as random perturbations. Bordenave applies this probabilistic framework to Toeplitz matrices, a widely used class of matrices in mathematics, physics, and engineering known for extreme sensitivity to small perturbations. Drawing on nearly a century of spectral theory for these matrices, he shows how random perturbation models yield precise predictions about their eigenvalue behavior. The talk, part of the IAS Members' Colloquium, is based on joint work with Mireille Capitaine and François Chapon and gives a technical but accessible account of a problem connecting numerical analysis, random matrix theory, and computational practice.