The Grothendieck-Verdier duality: $$ Rf_*\big(R\mathcal{H}\textit{om}_X^\bullet(\mathcal{E}^\bullet,f^!\mathcal{F}^\bullet)\big) \cong R\mathcal{H}\textit{om}^\bullet_Y(Rf_*\mathcal{E}^\bullet,\mathcal{G}^\bullet) $$ is known to hold for $f:X\to Y$ being a proper map of noetherian schemes.

Is there a way to get rid of this requirement on schemes to be noetherian, possibly by putting some extra conditions on the morphism $f$?

For example: let $X$ be smooth (projective in necessary) and consider the projection to the second factor $f:X\times U\to U$, where $U$ is some affine scheme. Does one have the duality in this setting?

Edit: After reading about the topic in Neeman's and Lipman's work, I have not managed to find an explicit construction of the right adjoint $f^\times$. In what cases is the explicit construction of $f^\times$ known? What would it be in the above example?

  • 2
    $\begingroup$ The name Grothendieck-Verdier is wrong. Grothendieck studied duality for complexes of coherent sheaves on schemes while Verdier developed duality for constructible sheaves on locally compact spaces. I understand they are different theorems and should be named appropriately. $\endgroup$
    – Leo Alonso
    Commented Oct 14, 2019 at 8:44
  • 1
    $\begingroup$ I just used the name that Huybrecht has given the standard formulation of the theorem. $\endgroup$
    – Arkadij
    Commented Oct 14, 2019 at 10:17

1 Answer 1


Does one have the duality in this setting?

Yes, we do have duality in a very general setting. Your question is equivalent to asking for the existence of a right adjoint to the derived pushforward functor $\mathbf{R}f_*\colon \mathbf{D}_{qc}(X) \to \mathbf{D}_{qc}(Y)$. It turns out that it exists for any morphism $f\colon X \to Y$ of qcqs schemes. Look at tag 0A9D for more details.

P. S. It is not a good idea to call this adjoint functor by $f^!$ unless $f$ is proper. Usually, another functor is denoted by $f^!$ that actually differs from the right adjoint $f^\times$ (SP denotes this functor by $a(f)$).

UPD: If $Y$ is qcqs and $f\colon X \to Y$ is proper and smooth of (pure) relative dimension $d$, then $f^!(-)\cong f^{\times}(-)\cong \Omega^d_{X/Y}[d]\otimes_{\mathcal O_X} f^*(-)$. Look at tag 0BRT and tag 0A9U for a proof in the noetherian case. The proof in the non-noetherian setup is essentially the same. Just note that noetherianness of $Y$ is used only to show that $\mathbf{R}f_*$ preserves perfect objects. However, it is true without this assumption as it is briefly explained in Example 2.2 of the paper Quasi-perfect scheme-maps and boundedness of the twisted inverse image functor.

  • $\begingroup$ In the example I mention, is the functor $a(f)$ given in the usual way that one would expect? That is, by derived pullback and tensoring with the canonical bundle on $X$ shifted by $\text{dim}(X)$? $\endgroup$
    – Arkadij
    Commented Oct 13, 2019 at 8:06
  • 1
    $\begingroup$ @Arkadij Bojko That's correct if we assume that $f$ is proper. I've updated my answer. $\endgroup$
    – gdb
    Commented Oct 14, 2019 at 4:45

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.