A converse to the Grace--Walsh--SzegH{o} theorem.pdf

a r X i v : 0 8 0 9 .3 2 2 5 v 1 [ m a t h .C V ] 1 8 S e p 2 0 0 8 A CONVERSE TO THE GRACE-WALSH-SZEGO? THEOREM PETTER BRA?NDE?N AND DAVID G. WAGNER Abstract. We prove that the symmetrizer of a permutation group preserves stability if and only if the group is orbit homogeneous. A consequence is that the hypothesis of permutation invariance in the Grace-Walsh-Szego? Coinci- dence Theorem cannot be relaxed. In the process we obtain a new characteri- zation of the Grace-like polynomials introduced by D. Ruelle, and prove that the class of such polynomials can be endowed with a natural multiplication. 1. Introduction and main result The Grace-Walsh-Szego? Coincidence Theorem is undoubtably one of the most useful phenomena governing the location of zeros of multivariate complex polyno- mials – see [7]. It is also very robust – after a full century of investigation, no significant improvement on it has been made. (Although significant generalizations to other settings have been obtained [6].) The main purpose of this note is to show that, in one sense at least, no such improvement is possible. A circular region is a proper subset of the complex plane that is bounded by either a circle or a straight line, and is either open or closed. A polynomial is multiaffine provided that each variable occurs at most to the first power. Proposition 1 (Grace-Walsh-Szego? [5, 9, 10]). Let f ∈ C[z1, . . . , zn] be a mul- tiaffine polynomial that is invariant under all permutations of the variables. Let A ? C be a circular region. Assume that either A is convex or that the degree of f is n. For any ζ1, . . . , ζn ∈ A there is a ζ ∈ A such that f(ζ1, ..., ζn) = f(ζ, . . . , ζ). The question we address is: for which permutation groups G ≤ Sn does an analogue of the Grace-Walsh-Szego? Theorem hold? That is, can one relax the hypothesis that f is invariant under all permutations of the variables by requiring only that f be invariant under the permutations in a subgroup G? (As usual, t