In Birkenhake and Lange's book, they prove a version of the Nakai-Moishezon theorem for complex abelian varieties that says that if $L_0$ is an ample line bundle on a complex abelian variety $X$ of dimension $g$, then a line bundle $L$ is ample if and only if $(L^\nu\cdot L_0^{g-\nu})>0$ for $\nu=1,\ldots,g$ (Corollary 4.3.3 on page 77 of the second edition).

I would love for this to be true in arbitrary characteristic. The problem is that the proof over $\mathbb{C}$ explicitly uses the Hermitian forms associated to the line bundles, and I don't see how this could be adapted to arbitrary characteristic.

Does anyone know if this is true in arbitrary characteristic? Are there any references or at least some idea for a proof?


1 Answer 1


On an abelian variety (regardless of the characteristic), an effective divisor with positive self-intersection is ample. To be more precise, it suffices here to recall that on any simple abelian variety, all non-zero effective divisors are ample; apply this to the factors in a decomposition (mod isogenies) of an arbitrary abelian variety into irreducibles.

It is therefore enough to show that a multiple of $L$ is effective. By Riemann-Roch and the theorem on page 32 (Example 1.2.31: Higher cohomology of nef divisors, I) in Lazarsfeld's "Positivity in alg. geom.," this will follow if we show that $L$ is nef, and Kleiman's theorem thus reduces the problem to checking $L.C \geq 0$ on every curve $C$ in $X$. Let me restrict to the case $\dim{X} = 2$ "for simplicity;" this should given an idea for the general case. Curves in abelian varieties are in any case nef, and if $a := -(L.C) > 0$, the divisor $H := aL_0 + (L_0.L)C$ would be ample. But $L$ has a positive self-intersection and is orthogonal to $H$, contradicting the Hodge index theorem on the surface $X$.

  • $\begingroup$ Ok, nice. Now how would this argument go in higher dimensions? We would want to intersect with curves to use Kleiman's Theorem, but I don't see how we can generalize what you wrote. If $(L\cdot C)<0$, we could take the 1-class $-(L\cdot C)L_0^{n-1}+(L_0^{n-1}\cdot L)C$ and this intersected with $L$ would be 0. But I'm not sure what else could be done... $\endgroup$
    – rfauffar
    Aug 28, 2013 at 18:57
  • $\begingroup$ In higher dimension, since $L_0$ is ample, you can take a surface $Y \subset X$ through $C$ in the linear equivalence class $L_0^{g-2} \in \mathrm{CH}^2(X)$ (i.e., linearly equivalent to a multiple of a $2$-plane in the appropriate projective embedding). Since $L_0^{g-2}.L^2 > 0$ we still have $L_{|Y}^2 > 0$, and the same argument shows that $L.C < 0$ contradicts the Hodge index theorem on the surface $Y$. $\endgroup$ Aug 28, 2013 at 23:31
  • $\begingroup$ Awesome answer! Thank you! It's really clear to me now. $\endgroup$
    – rfauffar
    Aug 29, 2013 at 0:54
  • $\begingroup$ One last question: Why can we assume that $C$ is contained in $L_0^{g-2}$? This doesn't seem obvious to me... $\endgroup$
    – rfauffar
    Aug 29, 2013 at 13:53
  • 1
    $\begingroup$ I'm sorry, I didn't write it cleanly in my last comment. I meant to say that $C$ lies on a surface which is linearly equivalent to a multiple of $c_1(L_0)^{g-2}$ in the Chow group $\mathrm{CH}_2$ of $2$-cycles mod rational equivalence. To fix ideas, say $g = \dim{X} = 3$. Because $L_0$ is ample, the sheaf $L_0^{\otimes n}(-C)$ has non-zero sections for $n \gg 0$, whose zero loci give surfaces through $C$ linearly equivalent to $nL_0$. $\endgroup$ Aug 29, 2013 at 14:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.