Let $\pi:X\rightarrow S$ be a smooth scheme over $S$ and let's assume for simplicity that $S=\mathrm{Spec} R$ is affine, Noetherian and regular. Let $\mathcal F$ be a coherent sheaf on $X$ and let's assume as well that $\mathcal F$ is torsion-free. Then I can take the derived global sections $R^\bullet\pi_* \mathcal F$, taking further its cohomology I get some $R$-modules $R^i\pi_*\mathcal F$.

Now the question is:

Given $R^i\pi_* \mathcal F$ can I say anything about torsion in it (as in an $R$-module)? In particular is there an open dense $U\subset S$ such that $R^i\pi_*\mathcal F$ is torsion-free restricted to $U$?

If $\pi$ is proper this is automatic since then $R^i\pi_* \mathcal F$ is finitely generated over $R$. If $\pi$ is affine then $R^\bullet\pi_* \mathcal F=\pi_*\mathcal F$ and we just consider some torsion-free module $M$ over $R'$ as a module over $R$ under an injection $R\subset R'$, so the statement holds as well. For the general case I tried to use a Cech cover of $X$ by affine $S$-schemes, but what you get is a complex whose terms are infinitely generated $R$-modules (though torsion-free) and I'm not sure what I could use to bound the torsion in the cohomology there.

Also, not sure if this makes life any simplier, but under the assumptions above $\mathcal F$ is a perfect complex on $X$ and it would be enough to take $\mathcal F=\mathcal O_X$.


1 Answer 1


No, this isn't true even when $S = \text{Spec}(\mathbf{Z})$ because of an example of Anurag Singh. Set $$ R = \mathbf{Z}[X, Y, Z, U, V, W]/(XU + YV + ZW) $$ and let $\mathfrak a \subset R$ be the ideal generated by $X, Y, Z$ in $R$. Set $$ X = \text{Spec}(R) \setminus V(\mathfrak a) $$ Then $H^2(X, \mathcal{O}_X) = H^3_\mathfrak a(R)$ has a $p$-torsion element for infinitely many primes $p$ (see link given above).

  • $\begingroup$ By the way, do you know if there is still a counterexamle for $S$ of characteristic 0? $\endgroup$
    – user42024
    Commented Aug 31, 2019 at 19:01

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.