On eigenvalues of symmetric (+1, −1) matrices

Abstract

To every symmetric matrixA with entries ±1, we associate a graph G(A), and ask (for two different definitions of distance) for the distance ofG(A) to the nearest complete bipartite graph (cbg). Letλ ₁(A),λ ¹ (A) be respectively the algebraically largest and least eigenvalues ofA. The Frobenius distance (see Section 4) to the nearest cbg is bounded above and below by functions ofn −λ ₁ (A), wheren=ord A. The ordinary distance (see Section 1) to the nearest cbg is shown to be bounded above and below by functions ofλ ¹ (A). A curious corollary is: there exists a functionf (independent ofn, and given by (1.1)), such that |λ _i (A) | ≦f(λ ¹(A), whereλ _i (A) is any eigenvalue ofA other thanλ _i (A).