The Manin constant. (With an appendix by J. Cremona) (Q851483)

Let \(E/\mathbb Q\) be an elliptic curve. Let \(N\) be the conductor of \(E\). By Brueil, Conrad, Diamond and Taylor we can consider \(E\) as a quotient of \(J_0(N)\). After possibly replacing \(E\) by an isogenous curve we may assume that the kernel of \(J_0(N)\to E\) is connected. Let \(\Omega\) be the generator of the free rank-1 \(\mathbb Z\)-module of invariant differentials on the Néron model of \(E\). Let \(f_E\) be the normalized new cuspidal eigenform associated to \(E\). Then the pullback of \(\omega\) to \(J_0(N)\) equals \(c 2\pi i f(\tau)\,d\tau\). The Manin constant \(c_E\) of \(E\) is the absolute value of \(c\). It is conjectured by Manin that this constant is always 1. The authors give a survey of known results on the Manin constant such as: It is known that \(c_E\) is an integer, the only primes that divide \(c_E\) are \(2,3,5,7\) and that the valuation at these primes can be bounded independently of \(E\). The authors generalize the notion of the Manin constant \(c_A\) to quotients \(A\) of \(J_0(N)\) of arbitrary dimension. Thy show that \(c_A\) is always an integer. The authors prove some divisibility results for \(c_A\). In particular, if \(J_0(N) \to A\) factors through \(J_0(N) \to J_0(N)/J_0(N)_{\text{old}}\) then a prime \(\ell\) divides \(c_A\) only if \(\ell=2\) or \(\ell^2\mid N\). The authors improve this result by showing that if \(2\) divides \(c_A\) and \(4\) does not divide \(N\) then \(2\) divides the modular exponent of \(A\). Finally, the authors conjecture that \(c_A\) is always \(1\) if \(A=A_f\) is a so-called optimal newform quotient. In contrast, the authors cite \textit{L. J. P. Kilford} [J. Number Theory 97, No. 1, 157--164 (2002; Zbl 1110.11301)] and \textit{A. Joyce} [J. Number Theory 110, No. 2, 325--330 (2005; Zbl 1161.11355)], who constructed abelian varieties such that \(c_A>1\). In an appendix by J. Cremona it is verified that \(c_E=1\) for all optimal quotients of \(J_0(N)\) for \(N<130000\). Cremona's method uses the existence of a global minimal Weierstrass equation for \(E/\mathbb Q\), and for that reason it seems difficult to apply his approach for quotients of \(J_0(N)\) of dimension \(>1\).