Baker domains for Newton's method (Q2372815)

The authors construct a transcendental entire function~\(f\) without finite asymptotic values such that its associating Newton function \(N\) has an invariant Baker domain. Their function~\(f\) can have any preassigned order strictly between~\(1/2\) and~\(1\). This answers a question asked by A. Douady in 2003 about any relation between the existence of an invariant Baker domain for~\(N\) and the possibility of~\(f\) to have the asymptotic value~\(0\). Newton's method is a way to find zeros of a nonlinear entire function~\(f(z)\) as the limit of the iterates of the Newton function \(N(z)=z-f(z)/f'(z)\) associated to~\(f(z)\) for some starting value. This is based on the fact that, for \(\xi\in\mathbb{C}\), \(N(\xi)=\xi\) if and only if \(f(\xi)=0\) or more concretely that each zero~\(\xi\) of~\(f\) is an attractive fixed point of~\(N\) and is contained in an \(N\)-invariant domain~\(U\), \(N(U)\subset U\), of the Fatou set \({\mathcal F}(N)\) of~\(N\), in which the iterates \(N^k\) converge to~\(\xi\). This \(U\) is called the immediate basin of~\(\xi\). On the other hand, \(N^k\) may also tend to~\(\infty\) in an \(N\)-invariant component of \({\mathcal F}(N)\). If \(N\) is rational, the point at~\(\infty\) is a fixed point of~\(N\) of multiplier~\(1\), and the \(N^k\) tend to~\(\infty\) in the immediate basin of this fixed point. When \(N\) has the essential singularity at~\(\infty\), a possible maximal invariant domain where the iterates \(N^k\) tend to~\(\infty\) is called an invariant Baker domain of~\(N\), which we may also call a virtual immediate basin in a sense weaker than that of \textit{X. Buff} and \textit{J. Rückert} in [Int. Math. Res. Not. 2006, No. 3, Article ID 65498, 18 p. (2006; Zbl 1161.37327)]. Douady's question is about the existence of the virtual root of \(f\) at \(\infty\), `\(f(\infty)=0\)', that is, whether the virtual immediate basin of \(N\) is related to \(0\) being an asymptotic value of~\(f\). \textit{W. Bergweiler, F. v. Haeseler, H. Kriete, H.-G. Meier}, and \textit{N. Terglane} in [Pitman Res. Notes Math. Ser. 305, 147--158 (1994; Zbl 0810.30017)] study `functions without zero as asymptotic value' and `functions tending to zero in a sector' and also give an example of functions~\(f\) with zero as asymptotic value such that \(N^k\to\infty\) for some \(N\)-invariant domain in a sector. Buff and Rückert show that a logarithmic singularity over~\(0\) implies a virtual immediate basin for~\(N\), too. Douady asks indeed in the opposite direction, that is, whether an entire function~\(f\) must always have~\(0\) as asymptotic value if~\(N\) has an invariant Baker domain. In the paper by Buff and Rückert, it is shown that the answer is positive in a large family of entire functions. As mentioned at this beginning, the authors of the paper under review show that this is not always the case by giving an example of entire functions of positive finite order. Buff and Rückert also point out that these results are parted according to types of their virtual immediate basins. The construction by Bergweiler, Drasin and Langley goes along the following steps: In \(\S 2\), using functions of the form introduced by \textit{S. K. Balashov} [Math. USSR, Izv. 7(1973), 601--627 (1974); translation from Izv. Akad. Nauk SSSR, Ser. Mat. 37, 603--629 (1973; Zbl 0272.30030)], they ingeniously construct an entire function~\(f\) of any preassigned order in \((1/2, 1)\) such that \(f(z)\sim z^{1/q}\) or \(N(z)\sim (1-q) z\) holds for some integer~\(q\) as \(z\to\infty\) in a logarithmically spiralling sector~\(S\). In \(\S 3\) they give an explicit error estimate in the asymptotic equality, which yields that \(S\) contains an invariant Baker domain of~\(N\). Finally in \(\S 4\) they apply the Denjoy-Carleman-Ahlfors Theorem to \(f(z)^q/z\) and conclude that \(f\) cannot have any finite asymptotic value. Then it seems crucial to assume that \(f\) has order strictly less than~\(1\), but they remark how to modify their method to produce both examples of~\(f\) of any preassigned non-integer order and examples with more than one invariant Baker domains. They also mention the possibility of treating the case of integer orders along the same lines. One might note that \textit{W. Bergweiler} [Newton's method and Baker domains, preprint] also proves the existence of entire functions with the asymptotic value~\(0\) and without invariant Baker domains.