Bicyclotomic polynomials and impossible intersections (Q2448541)

This work is based on a previous result by the same two authors [C. R., Math., Acad. Sci. Paris 346, No. 9--10, 491--494 (2008; Zbl 1197.11066)] and the ``presenter'' of that paper (J.-P. Serre) is acknowledged for asking the question that is partially answered here. The previous result says: The set of complex numbers \(t\neq 0, 1\) such that the points \[ (2, \sqrt{2(2-t)}), (3, \sqrt{6(3-t)}) \] both have finite order on the elliptic curve \(E_t\) defined by \(y^2=x(x-1)(x-t)\) is finite. Serre asked the authors to consider replacing ``the abscissas 2,3 by any two distinct complex numbers \(u, v\)'', in fact more precisely, by a pair \(u, v\) such that the transcendence degree \(\delta(u,v)\) of \(\mathbb Q(u,v)\) over \(\mathbb Q\) ranges over the possibilities \(0, 1, 2\). It turns out that the problem requires finding whether the identity (holding for all \(t\)) \(qP=pQ\) has solutions (namely, integers \(p, q\) not both zero), where \[ P=(u,\sqrt{u(u-1)(u-t)}),\quad Q=(v,\sqrt{v(v-1)(v-t)}) \] are points on the ``Legendre elliptic curve'' \(y^2=x(x-1)(x-t)\). The set of pairs \(u, v\) such that both \(P\) and \(Q\) have finite order is called \(\mathcal T (u, v)\). Having to use the formula for adding points of the Legendre curve, which is a quotient of polynomials in the coordinates, specifically, multiplication by a positive integer \(n\) sends the point \((x, y)\) to \[ \left(\frac{A_n(x,t)}{B_n(x,t)},y_n\right),\quad A_n(X,T),B_n(X,T)\in\mathbb Q[X,T] \] with leading coefficient (in \(X\)) normalization \[ A_n(X,T)=X^{n^2}+\dots,\quad B_n(X,T)=n^2X^{n^2-1}+\dots, \] the authors introduce the definition of ``bicyclotomic polynomials'', in analogy to ``cyclotomic polynomials'', namely the irreducible factors of \(X^n-1\) over \(\mathbb Q\): the bicyclotomic are the irreducible factors of \(B_n(X,T)\). The authors find the factorization using standard techniques such as Möbius inversion and Galois theory, but this is already a lengthy analysis. Then they switch to computational methods to attack their question, concluding with the main result: For each positive integer \(d\) there is an effectively computable finite set \(\mathcal F_d\) of polynomials in \(\mathbb Q[U,V]\), irreducible over \(\mathbb Q\) and of degree \(d\), with the following property. Suppose \(u,v\) are complex numbers satisfying \(uv(u-1)(v-1)(u-v) \neq 0\), not both algebraic over \(\mathbb Q\) and algebraically dependent over \(\mathbb Q\) through a polynomial, which is defined and irreducible of degree \(d\) over \(\mathbb Q\). Then the set \(\mathcal T(u,v)\) is effectively computable. If further \(F(u,v)\neq 0\) for every \(F\) in \(\mathcal F_d\) , then the set \(\mathcal T(u,v)\) is empty. Geometrically, the ``impossible intersections'' of the title refers to the affine space \(\mathbb A^5\) where the elliptic curve, now viewed as defined over \(\mathbb Q(u, v)\), is parametrized by \[ (u,\sqrt{u(u-1)(u-t)}, v,\sqrt{v(v-1)(v-t)}, t). \] The proofs in this paper are hands-on, with the authors ``beg[ging] forgiveness for further tediousness''. However, to reach the conclusions, which they do only in the case \(\delta = 1\), the authors use ``no fewer than five techniques'', ultimately invoking previous results (e.g., Raynaud's theorem on torsion points, \(abc\) techniques, stability properties of Néron-Tate heights on an elliptic threefold, upper bounds for values of points on an elliptic surface). When previous results are used, details are intentionally omitted. This makes the organization of the paper hard to follow. As a bonus, this combination of computational techniques and appeals to ad hoc major results that give sophisticated growth estimates, yields some surprising, beautiful facts about numbers, such as: there are no complex numbers \(t\neq 0, 1\) such that \[ (2\pi,\sqrt{2\pi(2\pi-1)(2\pi-t)}),\quad (3\pi,\sqrt{3\pi(3\pi-1)(3\pi-t)}) \] both have finite order on \(E_t\) (something to celebrate on next March 14).