6
$\begingroup$

Suppose that $\pi: X \to S$ is a proper morphism of schemes, and $G \to X$ is a flat affine algebraic group scheme over $X$. Is the push-forward $\pi_* G$ also affine?

I'd be happy to assume more -- say that $\pi$ is smooth, and that $G$ is smooth over $X$, or even a torus...

$\endgroup$
7
  • 1
    $\begingroup$ No, it is not affine. Already for $X\to S$ a smooth, projective morphism whose geometric fibers are connected curves of genus $g>0$, for $G$ the multiplicative group, the pushforward $\pi_*G$ is an extension of an etale group scheme (basically $\mathbb{Z}$) by the relative Jacobian of $X/S,$ and this is an Abelian scheme over $S.$ $\endgroup$ Nov 5, 2017 at 21:00
  • 4
    $\begingroup$ @JasonStarr: did you mean to refer to ${\rm{R}}^1\pi_{\ast}(\mathbf{G}_m)$? For the class of $\pi$ you mention (or more generally smooth proper with geometrically connected fibers of any dimension), $\pi_{\ast}(\mathbf{G}_{m,X})$ is identified with $\mathbf{G}_{m,S}$, as you know. Likewise, $\pi_{\ast}({\rm{GL}}_{n,X}) = {\rm{GL}}_{n,S}$ for such $\pi$ for any $n > 0$. $\endgroup$
    – nfdc23
    Nov 5, 2017 at 21:13
  • $\begingroup$ right -- to clarify, I didn't mean $R^1\pi_*$, but just $\pi_*$. $\endgroup$
    – Danny
    Nov 5, 2017 at 21:16
  • $\begingroup$ @nfdc23. You are completely correct. I was thinking of $\pi_*(B \mathbb{G}_m)$, the pushforward of the classifying stack, not $\pi_* \mathbb{G}_m,$ the pushforward of the group scheme. Sorry about that! $\endgroup$ Nov 5, 2017 at 21:19
  • 3
    $\begingroup$ @R.vanDobbendeBruyn. Representability follows, for instance, from Lemma 2.3.3 of Max Lieblich's article, "Remarks on the stack of coherent algebras": arxiv.org/pdf/math/0603034.pdf I believe that it can also be deduced from Cor. 7.7.8 of EGA $\textrm{III}_2,$ but I cannot quite make that work . . . $\endgroup$ Nov 5, 2017 at 22:15

2 Answers 2

9
$\begingroup$

Sorry about the mistake in the comment. First of all, for a proper, locally finitely presented morphism of schemes, $\pi:X\to S,$ with $S$ excellent (e.g., a finite type scheme over a field or over $\text{Spec}\ \mathbb{Z}$), for a finitely presented, flat, affine group scheme $\rho:G\to X$, the set-valued functor $\pi_* G$ is representable by a group scheme algebraic space that is locally finitely presented over $S$. (Ed. Thanks to @nfdc23 for pointing out that Lieblich's result only gives an algebraic space.) This follows, for instance, from Lemma 2.3.3 of the following article.

MR2233719 (2008c:14022)
Lieblich, Max(1-PRIN)
Remarks on the stack of coherent algebras.
Int. Math. Res. Not. 2006, Art. ID 75273, 12 pp.
https://arxiv.org/pdf/math/0603034.pdf

However, it is not true that $\pi_*G$ is always affine for $\pi$ a proper, locally finitely presented morphism and $G/X$ a flat, affine group scheme.
Let $S$ be $\mathbb{A}^2_k,$ the affine plane. The open complement, $j:V\hookrightarrow \mathbb{A}^2_k,$ is a flat morphism, but it is not affine.

Let $f:X\to S$ be the blowing up of $\mathbb{A}^2_k$ at the origin. This is a proper morphism. Denote by $E$ the exceptional divisor of $f$. Denote by $U$ the open complement of $E$. As the complement of a Cartier divisor in a smooth scheme, the open immersion $i:U\hookrightarrow X$ is an affine morphism.

Let $\rho:G\to X$ denote an $X$-scheme with two connected components, one of which maps isomorphically to $X$, $$\rho_e:G_e \xrightarrow{\cong} X,$$ and the second of which maps isomorphically to $U$, $$\rho_{\sigma}:G_{\sigma} \xrightarrow{\cong} U.$$ There is a unique structure of $X$-group scheme on $G$: the identity section is the inverse isomorphism of $\rho_e,$ and the multiplication morphism, $$G_\sigma\times_X G_\sigma \to G_e,$$ is the unique open immersion of $X$-schemes.

The $X$-group scheme $G$ is flat and affine. Yet the pushforward $\pi_*G$ is a disjoint union of a copy of $S$ and a copy of the open immersion $j:V\to S.$ This open immersion is not affine.

$\endgroup$
3
  • $\begingroup$ What I was imagining was that if $G$ sits inside of an affine bundle $A \to X$, where $A$ is the relative $Spec$ of the symmetric algebra of the dual to a vector bundle $V/X$, then the pushforward $\pi_* G$ should sit inside the affine bundle associated to $\pi_* V$. I guess the thing that broke in my intuition was that $\pi_* G$ didn't need to sit inside this as a closed subscheme, but just as something constructible...? $\endgroup$
    – Danny
    Nov 6, 2017 at 13:53
  • $\begingroup$ How does Lemma 2.3.3 of the given link imply the representability? (Note that we don't know if $G$ is a closed $X$-subgroup of some ${\rm{GL}}_n$, say even fppf-locally on $S$.) Olsson's paper on Weil restriction for proper flat maps gives that the pushforward is at least a quasi-separated algebraic space locally of finite type. $\endgroup$
    – nfdc23
    Nov 6, 2017 at 15:43
  • $\begingroup$ @nfdc23. Thank you for the correction. I changed "group scheme" to "group algebraic space." $\endgroup$ Nov 6, 2017 at 15:50
3
$\begingroup$

If $\pi \colon X \to S$ is proper, flat and of finite presentation and $W$ is an affine $X$-scheme, then $\pi_*W \to S$ is affine and of finite presentation. Unless $W$ is etale over $X$, it is difficult to deduce much about the smoothness of $\pi_*W \to S$.

To see this, you can combine an affine representability result for modules, such as https://stacks.math.columbia.edu/tag/08JY or [EGAIII-2, 7.7.8], with the ideas from Proposition 2.5 of:

Lieblich, Max, Remarks on the stack of coherent algebras, Int. Math. Res. Not. 2006, No. 11, Article ID 75273, 12 p. (2006). ZBL1108.14003.

A precise reference is Theorem 2.3 of

Hall, Jack; Rydh, David, General Hilbert stacks and Quot schemes, Mich. Math. J. 64, No. 2, 335-347 (2015). ZBL1349.14013.1434731927. https://projecteuclid.org/euclid.mmj/1434731927

$\endgroup$
2
  • $\begingroup$ If $W$ is smooth over $S$, and if $\pi$ is proper, finitely presented, and flat of relative dimension $\leq 1,$ then it should be possible to prove smoothness of $\pi_*W\to S$ by infinitesimal deformation theory. For a geometric point $\text{Spec}\ k\to S$ and for a section $\sigma:X_k\to W_k$, the obstruction space is $H^2(X_k,\sigma^*(\Omega_{W/K})^\vee).$ This is zero if the relative dimension is $\leq 1.$ $\endgroup$ Nov 7, 2017 at 11:27
  • $\begingroup$ The previous comment was wrong. Again I was thinking of the pushforward of the classifying stack $BG$, not the pushforward of $G.$ For the pushforward of $G,$ the obstruction space is $H^1(X_k,\sigma^*(\Omega_{W/k})^\vee).$ If $W$ is a smooth, linear group scheme of $X$, then there exists a locally free sheaf $\mathfrak{w}$ on $X$ such that $\Omega_{W/k}$ is the pullback of $\mathfrak{w}^\vee.$ Thus, the obstruction space is $H^1(X_k,\mathfrak{w}^\vee).$ This is typically nonzero. So you were correct: it seems difficult to deduce smoothness (unless $X/S$ is a family of genus $0$ curves). $\endgroup$ Nov 7, 2017 at 19:48

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.