1
$\begingroup$

The following is a theorem of Elkik from 1978 describing how rational singularities behave under flat morphisms.

Let $f:X\rightarrow S$ be a flat morphism of schemes of finite type over $\mathbb{C}$ such that both $S$ and the fibres of $f$ have rational singularities.

I am confused about one step in the proof and I am afraid that it should be obvious.

Let $\varphi:S'\rightarrow S$ be a resolution of singularities and let $X'=X\times_S S'$ be the fiber product, with projections $\varphi'$ and $f'$ to $X$ and $S'$ respectively. Also, let $\psi:Y\rightarrow X'$ be a resolution of singularities.

$$\begin{array}{ccccc} Y & \stackrel{\psi}{\longrightarrow} & X'=X\times_S S' & \stackrel{\varphi'}{\longrightarrow} & X \\ & & \downarrow & & \downarrow \\ & & S' & \stackrel{\varphi}{\longrightarrow} & S \end{array} f$$

By assumption, we have that $R^{\bullet}\psi\mathcal{O}_Y=\mathcal{O}_{X'}$ and $R^{\bullet}\varphi\mathcal{O}_{S'}=\mathcal{O}_S$. I am trying to show that then $X$ also has rational singularities, namely $R^{\bullet}(\varphi'\circ \psi)\mathcal{O}_Y=\mathcal{O}_X$.

By Grothendieck spectral sequence, since $X'$ has rational singularities, we have that $R^i(\varphi'\circ \psi)\mathcal{O}_Y = R^i\varphi'\mathcal{O}_{X'}$ so we have to show that $R^{\bullet}\varphi'\mathcal{O}_{X'}=\mathcal{O}_X$.

I am not sure about how this follows: by flat base change the natural map $$f^{\ast}R^i\varphi_{\ast}\mathcal{O}_{S'}\rightarrow R^i\varphi'f'^{\ast}\mathcal{O}_{S'}$$ is an isomorphism.

Now for instance for $i=0$ flat base change yields $\varphi_{\ast}'f'^{\ast}\mathcal{O}_{S'}=f^{\ast}\mathcal{O}_S$. Pushing forward by $f$ on both sides yields $$f_{\ast}\varphi'_{\ast}f'^{\ast}\mathcal{O}_{S'}=f_{\ast}f^{\ast}\mathcal{O}_S=\mathcal{O}_S\otimes f_{\ast}\mathcal{O}_X=f_{\ast}\mathcal{O}_X$$

Commutativity on the LHS yields $$f_{\ast}\varphi'_{\ast}f'^{\ast}\mathcal{O}_{S'}=\varphi_{\ast}f'_{\ast}f'^{\ast}\mathcal{O}_{S'}=\varphi_{\ast}(\mathcal{O}_{S'}\otimes f'_{\ast}\mathcal{O}_{S'})=\varphi_{\ast}f'_{\ast}\mathcal{O}_{X'}$$

So $f_{\ast}\varphi'_{\ast}\mathcal{O}_{X'}=f_{\ast}\mathcal{O}_X$.

Does it follow that $\varphi'_{\ast}\mathcal{O}_{X'}=\mathcal{O}_X$? Am I making this more complicated than it actually is?

Added: For instance, if $f$ is injective, the claim would follow, since then $f_{\ast}\mathcal{F}\simeq f_{\ast}\mathcal{G} \Longrightarrow \mathcal{F}\simeq \mathcal{G}$. (by definition $\mathcal{F}(f^{-1}(U))\simeq \mathcal{G}(f^{-1}(U))$ for every $U\subseteq Y$ open and provided that $f$ is flat and injective, for any open set $V\subseteq X$, we have $V=f^{-1}(f(V))$ (since flat morphisms of finite type between Noetherian schemes are open). But this is more than we need.

$\endgroup$
3
  • $\begingroup$ I am a little bit confused : you've already posted this question on math.SE : math.stackexchange.com/questions/464449/…. Moreover, you had already asked it on Mathoverflow, before deleting the question. Why so ? $\endgroup$ Aug 12, 2013 at 13:30
  • $\begingroup$ What do you mean by $f$ injective in the "Added" section? $\endgroup$ Aug 12, 2013 at 13:37
  • $\begingroup$ I meant injective as a map between topological spaces. $\endgroup$
    – Marc
    Aug 12, 2013 at 14:08

1 Answer 1

2
$\begingroup$

I think you are. That is, making it more complicated than it is.

You were done right after saying that "by flat base change the natural map $$f^{\ast}R^i\varphi_{\ast}\mathcal{O}_{S'}\rightarrow R^i\varphi'_*f'^{\ast}\mathcal{O}_{S'}$$ is an isomorphism."

By definition $f^*\mathscr O_S=\mathscr O_X$ and ${f'}^*\mathscr O_{S'}=\mathscr O_{X'}$. In other words, you get that

$$R^i{\varphi'_*}\mathscr O_{X'}\simeq R^i\varphi'_*f'^{\ast}\mathscr{O}_{S'}\simeq f^{\ast}R^i\varphi_{\ast}\mathscr{O}_{S'}\simeq \begin{cases} 0 \qquad \text{if $i>0$} \\ \mathscr O_X \quad \text{if $i=0$} \end{cases},$$ exactly what you need.

$\endgroup$

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.