2
$\begingroup$

Does the Grothendieck class behave well under normalization? Here by Grothendieck class, I mean the class of a variety in $K_0(\mathcal{var/\mathbb{C}})$

$\endgroup$
3
  • 4
    $\begingroup$ If $X$ is a nodal cubic curve and $\tilde X$ is its normalization, then $[\tilde X] = [X] + [pt]$. Is that "behaving well"? $\endgroup$ Apr 16, 2016 at 5:26
  • $\begingroup$ @AllenKnutson Yeah, that behaves well. Sorry it is a general question. I guess it depends on what we normalize. But I still hope that there is some general result about it. $\endgroup$
    – Feng Hao
    Apr 16, 2016 at 12:49
  • 2
    $\begingroup$ Maybe you should consider seminormalizing first. I expect that shouldn't change the Grothendieck class at all. $\endgroup$ Apr 16, 2016 at 23:54

1 Answer 1

2
$\begingroup$

If $\tilde{X}$ is the normalization of $X$, then $\tilde{X}$ and $X$ are birational (see here for example).

So there exists open sets $U\subset X$ and $\tilde{U}\subset \tilde{X}$ that are isomorphic via the normalization map. Let $D=X-U$ and $\tilde{D}=\tilde{X}-\tilde{U}$. Then $[\tilde{X}]=[X]+([\tilde{D}]-[D])$.

This generalizes the example given by Allen Knutson in the comments since in the case of a nodal cubic curve, normalization is desingularization (as with any such curve) and so over the node, the normalization has two points; thus, $[\tilde{D}]-[D]=[pt]$.

Of course this observation is true for any birational map, and really says nothing about the normalization itself (aside from that property).

Here is another example. The cusped cubic curve has both $[\tilde{D}]=[pt]$ and $[D]=[pt]$, so $[\tilde{D}]-[D]=0$. In that case the normalization map is injective.

More generally, now suppose that the normalization map is injective.

Lemma: $X \to Y$ is a dominant morphism of irreducible algebraic varieties. If there exists a dense subset $Z \subset Y$ such that for any $y \in Z$ the fiber consists of a single point, then the map is birational.

So the restriction of the normalization map to $\tilde{D}$ is bijective and finite onto $D$ and hence is birational (by the lemma). So on each component of $\tilde{D}$ we get an isomorphism on open sets to open sets in components of $D$. By induction on dimension, we eventually get down to an isomorphism. This shows that $[\tilde{D}]-[D]=0$ whenever the normalization map is injective.

More generally, radical surjective maps $f:X\to Y$ give equality $[X]=[Y]$ in characteristic $0$, and if the field is algebraically closed then $f$ is radicial if it is injective. So over $\mathbb{C}$ bijections $f:X\to Y$ imply $[X]=[Y]$.

Now, if one considers instead of $\tilde{X}$, the seminormalization $X^{sn}$, then there is $\tilde{X}\to X^{sn}\to X$ where the map $X^{sn}\to X$ is finite and bijective (hence birational by the above lemma). So the same argument as above shows that $[X]=[X^{sn}]$, as commented by Allen Knutson.

In general, $[\tilde{D}]-[D]$ has to be handled case-by-case however, but one can WLOG assume that $X$ is seminormal.

$\endgroup$
0

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.