2
$\begingroup$

Let $X$ be a smooth quasi-projective variety over $\mathbb{C}$ and $Z$ be a closed subvariety of codimension $k$.

Q1. How to define a cycle class $[Z]\in H^k(X,\Omega_X^{k})$ ?

Q2. More general, which are necessary conditions to have a "good" cycle class map of this type? The "good" means it is coincide with the usual cycle class map when $X$ is smooth projectve complex variety. I mean that if $X$ can be define over arbitrary field or in the case $X$ is a variety but maybe not smooth quasi-projective.

$\endgroup$
4
  • $\begingroup$ This looks rather strange; one usually defines cycle classes only for those cohomology theories that satisfy some sort of the homotopy invariance property (that fails for the cohomology you consider unless $X$ is also proper). $\endgroup$ May 19, 2015 at 20:39
  • $\begingroup$ I find this in subsection 1.1 of Variations de structure de Hodge et zéro-cycles sur les surfaces générales, where the author give a cycle class in $H^2(X,\Omega_X^2)$ $\endgroup$
    – mwZhang
    May 20, 2015 at 1:13
  • 1
    $\begingroup$ In that paper $X$ is proper, so this just the standard definition. $\endgroup$
    – abx
    May 20, 2015 at 5:13
  • 2
    $\begingroup$ There does indeed exist a cycle class map in Hodge cohomology for any smooth variety. See, for example, the article by El Zein "Complexe dualisant et applications à la classe fondamentale d'un cycle". $\endgroup$
    – naf
    May 20, 2015 at 9:08

2 Answers 2

1
$\begingroup$

Here's one way to obtain a cycle class, at least over a field $k$ of characteristic 0 where resolutions of singularities are available:

Let $Z \subset X$ be a subvariety of codimension $c$ (possibly singular), and let $\pi: \tilde{Z} \to Z$ be a resolution of singularities. This means $\pi$ is proper and birational, and $\tilde{Z}$ is smooth. Let $f = \iota \circ \pi : \tilde{Z} \to X$ where $\iota: Z \to X$ is the inclusion.

The (dual of the) differential $d f^\vee: f^*\Omega_X^* \to \Omega_{\tilde{Z}}^*$ induces homomorphisms of Hodge cohomology groups $$ f^*: H^q(X, \Omega_X^p) \to H^q(\tilde{Z}, \Omega_{\tilde{Z}}^p) \text{ for all } p, q $$ Now, $H^{\dim \tilde{Z}}(\tilde{Z}, \Omega_\tilde{Z}^{\dim \tilde{Z}}) \xrightarrow{\mathrm{Tr}, \simeq} k$ via the trace map, and the composition $$ H^{\dim \tilde{Z}}(X, \Omega_X^{\dim \tilde{Z}}) \xrightarrow{f^*} H^{\dim \tilde{Z}}(\tilde{Z}, \Omega_{\tilde{Z}}^{\dim \tilde{Z}}) \xrightarrow{\mathrm{Tr}, \simeq} k $$ is an element of $$ H^{\dim \tilde{Z}}(X, \Omega_X^{\dim \tilde{Z}})^{\vee} \simeq H^{\dim X -\dim \tilde{Z}}(X, \Omega_X^{\dim X -\dim \tilde{Z}}) $$ This isomorphism comes from Poincare duality -- since $\Omega_X^{\dim \tilde{Z}} \otimes \Omega_X^{\dim X - \dim \tilde{Z}} \xrightarrow{\wedge} \omega_X$ is a perfect pairing it induces an isomorphism $\Omega_X^{\dim X - \dim \tilde{Z} } \simeq \mathrm{Hom}(\Omega_X^{\dim \tilde{Z}}, \omega_X)$. Hence we've assigned a cohomology class, say $\eta(Z) \in H^c(X, \Omega_X^c)$ to $Z$. Well, strictly speaking we've assigned it to the morphism $f : \tilde{Z} \to X$; it'd take more work to prove $\eta(Z)$ is idependent of the resolution $ \pi : \tilde{Z} \to Z$ (idea: any 2 resolutions of $Z$ can be dominated by a third).

$\endgroup$
1
$\begingroup$

An alternative approach without using resolution of singularities is given in Lipman's "Blue Book":

Dualizing sheaves, differentials and residues on algebraic varieties. Astérisque No. 117 (1984)

See Chap. 3, Remark (ii) on page 39.

In short, for a $d$-dimensional variety $V$ there is a canonical map $c_V \colon \Omega^n_V \to \omega_V$ that provides an element in $$ H^{N-d}_V(X,\Omega^{N-d}_X) $$ where $V \hookrightarrow X$ is the embedding of $V$ into a regular variety $X$ of dimension $N$. Take $Z := V$ in your notation. Of course there is map from cohomology with supports to usual cohomology.

$\endgroup$

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.