Cyclotomic Diophantine problems (Hilbert irreducibility and invariant sets for polynomial maps) (Q2382010)

For an algebraic number field \(k\) denote by \(k^c\) the field generated over \(k\) by all roots of unity, and for \(d\geq1\) let \(U_d\) be the set of torsion points in \(G^d_m\) (\(G_d\) being the variety \(A^1\setminus\{0\}\) with the multiplicative group law). Let \(V/k\) be an affine variety irreducible over \(k^c\) and let \(\pi:\;V\to G_m^d\) be a morphism of finite degree defined over \(k\). In Theorem 1 the authors consider the case, when the set \(\{\eta\in V(k^c):\;\pi(\eta)\in U_d\}\) is Zariski dense in \(V\), and describe in that case the Zariski closure of \(\pi(V)\). This leads to a new proof of the fact that the field \(k^c\) is Hilbertian, and has interesting consequences concerning Hilbert's Irreducibility Theorem, e.g., if \(f\in k^c(X,Y)\) and for every \(1\leq m\leq \deg_Yf\) the polynomial \(f(X^m,Y)\) is irreducible over \(k^c\), then \(f(\zeta,Y)\) can be reducible over \(k^c[Y]\) only for finitely many roots of unity \(\zeta\). In Theorem 2 the authors prove the rather unexpected result that if a polynomial \(f\in k^c[X]\) of degree \(d\geq2\) has infinitely many pre-periodic points in \(k^c\), then \(f\) is linearly conjugated to either \((\pm x)^d\) or \(T_d(\pm x)\), \(T_d\) being the \(d\)th Chebyshev polynomial. Then they show (Theorem \(2^*\)) that the assumption in Theorem 2 can be replaced by the existence of an infinite subset \(\Gamma\) of \(k^c\) with \(f(\Gamma)=\Gamma\). The last result, Theorem 3, deals with the field property (P) defined in the following way: a field \(K\) has the property (P) if every \(f\in K[X]\) for which there exists an infinite set \(X\subset K\) with \(f(X)=X\) must be linear. The authors show that if \(p\) is an odd prime, which is not a Mersenne prime, \(l\) is an odd prime divisor of \(p-1\), the field \(K'\) is generated over the rationals by all roots of unity of orders \(p,p^2,\dots\), and \(K\) is the subfield of \(K\) of index \(l\), then \(K\) has property (P), but \(K'\) does not have it (this solves problems XIV, XV and a part of problem XIX in the reviewer's [Polynomial Mappings. Lect. Notes Math. 1600, Springer (1995; Zbl 0829.11002)]. This theorem implies also that \(K\) does not have the property (N) (Northcott property), defined as follows: a subfield \(K\) of the algebraic closure \(\overline {\mathbb Q}\) of the rationals has property (N) if it contains only finitely many elements with bounded absolute logarithmic Weil height. One obtains thus that the properties (N) and (P) are not equivalent for subfields of \(\overline {\mathbb Q}\) (one knows that for such fields (P) is implied by (N)).