1: \begin{abstract}

2:   Finding roots of univariate polynomials is one of the fundamental tasks

3:   of numerics, and there is still a wide gap between root finders that are

4:   well understood in theory and those that perform well in practice.

5:   We investigate the root finding method of Weierstrass, a root finder

6:   that tries to approximate all roots of a given polynomial in parallel

7:   (in the Jacobi version, i.e., with parallel updates). This method has a

8:   good reputation for finding all roots in practice except in obvious cases

9:   of symmetry, but very little is known about its global dynamics and

10:   convergence properties.

11: 

12:   We show that the Weierstrass method, like the well known Newton method,

13:   is not generally convergent: there are

14:   open sets of polynomials~$p$ of every degree $d \ge 3$ such that

15:   the dynamics of the Weierstrass method applied to~$p$ exhibits attracting

16:   periodic orbits.  Specifically, all polynomials

17:   sufficiently close to $Z^3 + Z + 180$ have attracting cycles of period~$4$.

18:   Here, period $4$ is minimal: we show that for cubic polynomials, there are no

19:   periodic orbits of length $2$ or~$3$ that attract open sets of starting

20:   points.

21: 

22:   We also establish another convergence problem for the Weierstrass method:

23:   for almost every polynomial of degree $d\ge 3$ there are orbits

24:   that are defined for all iterates but converge

25:   to~$\infty$; this is a problem that does not occur for Newton's method.

26: 

27:   Our results are obtained by first interpreting the original problem

28:   coming from numerical mathematics in terms of higher-dimensional

29:   complex dynamics, then phrasing the question in algebraic terms in such a

30:   way that we could finally answer it by applying methods from computer

31:   algebra.

32: \end{abstract}

33: