Let $A$ be a simple abelian variety of dimension $g$. Let $K$ be an irreducible perverse sheaf on $A$. We know that $\chi(A,K)\geq 0$. (Corollary 1.4 of Franecki and Kapranov.) How small can $\chi(A,K)$ be?

There are some trivial examples. If $K$ is a lisse sheaf on all of $A$ placed in degree $g$, then $\chi(A,K)=0$. If $K$ is a skyscraper sheaf, then $\chi(A,K)=1$.

Beyond these trivial examples, if $A$ is the Jacobian of a curve $C$ of genus $g$ and $K$ is the constant sheaf on $C$, placed in degree $-1$ and pushed forward to $A$, then $\chi(A,K)=2g-2$.

I want to know if there is any nontrivial example with smaller Euler characteristic:

Does there exist a simple abelian variety $A$ of dimension $g$ and an irreducible perverse sheaf $K$ on $A$ such that $K$ is not lisse, $K$ is not a skyscraper sheaf, and $\chi(A,K)<2g-2$?

We know every irreducible perverse sheaf of Euler characteristic $0$ or $1$ is trivial in this sense. (Proposition 10.1 of Kramer and Weissauer.) So it is equivalent to ask:

Does there exist a simple abelian variety $A$ of dimension $g$ and an irreducible perverse sheaf $K$ on $A$ such that $2 \leq \chi(A,K) <2g-2?$

One way to construct a perverse sheaf on an abelian variety is to take a smooth variety $X$ and a finite morphism $f: X\to A$ and pushforward the constant sheaf in degree $-\dim(X)$ along $f$. So a subquestion is:

Does there exist a simple abelian variety $A$ of dimension $g$, a smooth non-abelian variety $X$ of dimension $d$, and a finite map $f:X \to A$, such that $(-1)^{d} \chi(X) < 2g-2$?

  • $\begingroup$ Isn't $d=g$ in your last statement ? $\endgroup$ Commented Apr 1, 2014 at 7:11
  • $\begingroup$ @DamianRössler - not if $f$ is a closed immersion, say. $\endgroup$
    – Will Sawin
    Commented Apr 1, 2014 at 14:35

1 Answer 1


Mihnea Popa explained this lower bound to me:

If $f: X \to A$ is a finite morphism (or weaker conditions are possible), then we have generic vanishing for every Hodge number of $X$ but $h^{p,q}$ for $p+q=d$. So $(-1)^d \chi(X)$ is just the sum of the generic values of $h^{p,q}$ for $p+q=d$. In particular, it is at least twice the generic value of $h^{0,d}$, which is equal to the arithmetic Euler characteristic of $\omega_X$. A lower bound for this is the codimension of $X$ as long as $A$ is simple or $X$ has no irrational pencils (slight variant on Theorem A of Pareschi and Popa). This gives a lower bound of $2g-2d$.

As far as I know this should hold for any perverse sheaf that arises from any geometric construction, because we should always be able to put a Hodge structure on it (making it a mixed Hodge module) and use the same argument.

I would be interested in further lower bounding arguments or upper bounding examples. The largest gap between the lower bound and the examples I know how to construct occurs for hypersurfaces, where the lower bound is $2$ and the best example I know about it is the theta divisor of a Jacobian, which has Euler characteristic $ \left( \begin{array}{c} 2g-2 \\ g-1\end{array}\right)$, or about $4^g$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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