Let $f\colon X \to Y$ be a dominant morphism between normal projective algebraic varieties; assume $f_*\mathcal {O}_X=\mathcal{O}_Y $; let’s also assume that it is defined over the complex numbers, but I do not think this is relevant for the question. I do NOT assume that $f$ is flat.

Let $L$ be a line bundle on $X$ which is trivial along all fibres of $f$. Is $L$ the pull-back of a line bundle from $Y$?

(I would expect that $f_*L$ is a line bundle, and the natural map $$ f^*f_*L\to L $$ is an isomorphism. When $f$ is flat, this is the seesaw principle.)

I am very interested in the case where $f$ is a birational map.


  • 3
    $\begingroup$ You should assume that $Y$ is normal and that $f$ is dominant with connected fibers, otherwise there are easy counterexamples. $\endgroup$ Nov 23 '18 at 20:56
  • 1
    $\begingroup$ It seems likely that there should be counterexamples even with $X$ a surface and $f$ the contraction of a curve $C$: it suffices to find a line bundle on $X$ which is trivial on $C$ but not on some infinitesimal neighbourhood. (I have not worked out an explicit example...) $\endgroup$
    – naf
    Nov 24 '18 at 9:20
  • $\begingroup$ @ulrich Sounds convincing. Should the answer be positive if we assume $R^1 f_* \mathcal{O}_X = 0$? In this case if $L$ is trivial on the fiber, it should be trivial on all of its infinitesimal neighborhoods. $\endgroup$ Nov 24 '18 at 10:09
  • 1
    $\begingroup$ @PiotrAchinger: Indeed: working over $\mathbb{C}$ in the analytic topology, this follows by using the exponential sequence (and the Leray spectral sequence) to compute the Picard group. $\endgroup$
    – naf
    Nov 24 '18 at 12:24

I have received the following great answer from Vlad Lazić:

In the numerical setup, what you ask (with some assumptions on the base, such as Q-factoriality) is well-understood, see the paper of Lehmann "Numerical triviality and pullbacks" and Nakayama's book on Zariski decomposition. I suppose this should give you the answer under additional hypothesis, such as the vanishing of the first direct image of the structure sheaf. I doubt you can get away without the assumptions on pseudoeffectivity of L: an example should be something like E\times E (for a general elliptic curve) and L=diagonal minus a fibre of one projection, and the map is the second projection (I didn't check the details, although you probably have to take something like E\times P^1 or something similar if you're only interested in linear equivalence). You might also find interesting Lemma 3.1 in my paper with Peternell, "On Generalised Abundance, I".


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.