Let $k$ be a field of characteristic $0$. Let $R$ be a Noetherian local normal domain containing $k$. Also assume that $R$ is the homomorphic image of a Gorenstein ring of finite dimension, hence $R$ admits a Dualizing complex. Let $Y$ be a Gorenstein normal scheme over Spec$(k)$ of finite Krull-dimension.

If there exists a proper birational map $f: Y \to \text{Spec}(R)$ such that $R^i f_* \mathcal O_Y=0, \forall i>0$, then is it true that $R$ is Cohen-Macaulay? If this is not true in general, what if I also assume $Y$ is regular?

As $R$ is normal, $R$ is of course Cohen-Macaulay when $\dim R \leq 2$. So we only have to think about $\dim R \geq 3$.

  • $\begingroup$ I believe that this is a theorem of Elkik when $R$ is the localization of a finitely generated $k$-algebra. Thus, it also holds if $R$ sits "between" such a localization and its completion (since you can check the Cohen-Macaulay property after passage to the localization). So probably you can deduce your case from the Cohen structure theorem. $\endgroup$ Jul 6 at 11:33
  • $\begingroup$ @JasonStarr: Thank you for your comment. Could you please give a reference for the case of when $R$ is essentially of finite type over $k$? Also, Is Spec$(R)$ necessary, or can we just do it for any normal scheme essentially of finite type over $k$? $\endgroup$
    – Snake Eyes
    Jul 7 at 0:27
  • $\begingroup$ I learned about this in Koll'ar's paper, "Singularities of pairs," in the Santa Cruz volume, but he references an earlier paper of Elkik. I will try to track down the original article. $\endgroup$ Jul 7 at 11:09
  • $\begingroup$ Elkik explains the proof in the discussion leading up to Definition 1 in the article, "Rationalit'e des singularit'es canoniques": eudml.org/doc/142810 $\endgroup$ Jul 7 at 11:16
  • $\begingroup$ I read the proof, and I think it probably directly applies to your situation, without using the completion. I will try to write more soon. $\endgroup$ Jul 7 at 12:48

2 Answers 2


I started writing this last night, but didn't finish. In addition to Jason Starr's answer, you can use my new vanishing theorems to remove the assumption that $R$ is essentially of finite type over a field. Note that the special case when $R$ is essentially of finite type over a field of characteristic zero also follows from Kempf's criterion for rational singularities [Kempf 1973, p. 50].

In short, if $Y$ is assumed to be regular, then $R$ will be Cohen–Macaulay. In fact, we have the following stronger result.

Theorem. Let $R$ be a Noetherian ring and let $f\colon Y \to \operatorname{Spec}(R)$ be a proper birational morphism such that $Y$ is regular and $R^if_*\mathcal{O}_Y = 0$ for all $i > 0$. Suppose $R$ contains $\mathbf{Q}$. Then, every localization of $R$ is pseudo-rational, and in particular $R$ is Cohen–Macaulay.

Proof. Since the conclusion can be checked locally, it suffices to consider the case when $(R,\mathfrak{m})$ is local. Set $\hat{Y} := Y \times_{\operatorname{Spec}(R)} \operatorname{Spec}(\hat{R})$, and consider the Cartesian diagram $$\require{AMScd}\begin{CD} \hat{Y} @>>> Y\\ @V\hat{f}VV @VVfV\\ \operatorname{Spec}(\hat{R}) @>>> \operatorname{Spec}(R) \end{CD}$$ By base change, $\hat{f}$ is proper, and by flat base change, $\hat{f}$ is birational [EGAI$_\text{new}$, Proposition 3.9.9] and $R^i\hat{f}_*\mathcal{O}_{\hat{Y}} = 0$ for all $i > 0$. Note that $\hat{R}$ is normal by [Lipman 1969, Remark 16.2] and that the pseudo-rationality of $\hat{R}$ would imply that $R$ is pseudo-rational [Murayama 2022, Proposition 4.20]. Since $\hat{Y}$ is regular by [EGAIV$_2$, Lemme], we may replace $R$ by $\hat{R}$ and $Y$ by $\hat{Y}$ to assume that $R$ is complete local.

We now prove the special case when $(R,\mathfrak{m})$ is complete local. First, by my version of Grauert-Riemenschneider vanishing [Murayama, Theorem B(i)], we know that $R$ has rational singularities (see, e.g., [Kollár 2013, Definition 2.76] for the definition). Since $R$ having rational singularities is equivalent to saying that $R$ is pseudo-rational in our setting [Murayama, Remark 7.4], we are done.

We can also show that $R$ is Cohen–Macaulay directly. Since Cohen–Macaulayness can be checked locally, it suffices to consider the case when $(R,\mathfrak{m})$ is local. Since $R$ is normal, the map $R \overset{\sim}{\to} f_*\mathcal{O}_Y$ is an isomorphism [EGAIII$_1$, Corollaire 4.3.12]. Combined with the assumption that $R^if_*\mathcal{O}_Y = 0$ for all $i > 0$, we have the quasi-isomorphism $R \overset{\sim}{\to} \mathbf{R}f_*\mathcal{O}_Y$. Applying $\mathbf{R}\Gamma_{\mathfrak{m}}(-)$, we obtain $$\mathbf{R}\Gamma_{\mathfrak{m}}(R) \overset{\sim}{\longrightarrow} \mathbf{R}\Gamma_{\mathfrak{m}}(\mathbf{R}f_*\mathcal{O}_Y) \cong \mathbf{R}\Gamma_{f^{-1}(\{\mathfrak{m}\})}(\mathcal{O}_Y).$$ Now taking $i$-th cohomology modules, we obtain the isomorphisms $$H^i_{\mathfrak{m}}(R) \overset{\sim}{\longrightarrow} H^i_{\mathfrak{m}}(\mathbf{R}f_*\mathcal{O}_Y) \cong H^i_{f^{-1}(\{\mathfrak{m}\})}(\mathcal{O}_Y).$$ Since the local cohomology modules on the right vanish for all $i < \dim(R)$ by my version of Grauert–Riemenschneider vanishing (in its dual formulation) [Murayama, Theorem B*(i)], we see that $R$ is Cohen–Macaulay. $\blacksquare$

  • $\begingroup$ Thank you for your answer. I will read it carefully, and will surely have many questions, but I quickly want to ask two things first: $Y$ is just a regular scheme with no other assumption? By $\widehat Y$, do you actually mean the fiber-product $Y \times_{Spec(R)} Spec(\widehat R)$? $\endgroup$
    – Snake Eyes
    Jul 8 at 2:25
  • $\begingroup$ I like your second argument... could you please spell out the Leray spectral sequence isomorphism ? $\endgroup$
    – Snake Eyes
    Jul 8 at 3:59
  • $\begingroup$ This is definitely better than the original version. $\endgroup$ Jul 8 at 10:10
  • $\begingroup$ @SnakeEyes I edited my answer. For the regularity of $\hat{Y}$, see [Stacks Project, Tag 0BG6] for another reference. $\endgroup$ Jul 9 at 15:36

Edit. Thanks to user @Johan for pointing out the indexing mistake. It is now corrected.

I read more of Elkik's article. I am just expanding my comments into one answer. Edit. Assume that $R$ is essentially of finite type over $\text{Spec}\ k$. By hypothesis, the scheme $X=\text{Spec}\ R$ has a dualizing complex $\omega^\bullet_R$ that has nonzero homology sheaf $\mathcal{H}^{\text{dim}\ R}(\omega^\bullet_R)$, and has vanishing homology sheaves $\mathcal{H}^i(\omega^\bullet_R)$ for $i>0$ and for $i<-\text{dim}\ R$. The ring $R$ is Cohen-Macaulay if and only if the only integer $i$ with $\mathcal{H}^i(\omega^\bullet_R)$ nonzero is $i=-\text{dim}\ R$. It is Gorenstein if, further, this homology sheaf is a projective $R$-module of rank $1$.

Also by hypothesis, the scheme $Y$ has a dualizing complex $\omega_Y^\bullet$ such that the homology sheaf $\mathcal{H}^i(\omega_Y^\bullet)$ is nonzero if and only if $i=-\text{dim}\ Y = -\text{dim}\ R$. By relative duality, we have a natural quasi-isomorphism, $$\alpha:Rf_* \omega_Y^\bullet \cong R\text{Hom}_{\mathcal{O}_X}(Rf_*\mathcal{O}_Y,\omega_X^\bullet).$$ By hypothesis, the natural map $\mathcal{O}_X\to Rf_*\mathcal{O}_Y$ is a quasi-isomorphism. Thus, $\alpha$ is equivalent to a quasi-isomorphism, $$Rf_*\omega^\bullet_Y \cong \omega_X^\bullet.$$

Since the only integer $i$ with a nonvanishing homology sheaf $\mathcal{H}^i(Rf_*\omega^\bullet_Y)$ is $i=-\text{dim}\ Y$, the complex $Rf_*\omega^\bullet_Y$ has nonvanishing homology sheaves $R^if_*\omega^\bullet_Y$ only for $ -\text{dim}\ Y \leq i \leq 0$.

Edit. Since $R$ is essentially of finite type and $Y$ is regular, by Grauert-Riemenschneider Vanishing the derived direct image $R^if_*\omega^\bullet_Y$ is nonzero only for $i=-\text{dim}\ Y = -\text{dim}\ R$, in which degree it equals $f_*\mathcal{H}^{-\text{dim}\ Y}(\omega^\bullet_Y)$.

Thus the complex $\omega_X^\bullet$ has nonvanishing homology sheaves only for $i =- \text{dim}\ R = \text{dim}\ Y$. Therefore $R$ is Cohen-Macaulay.

  • $\begingroup$ Sorry, Jason, but there is something wrong with the numbering. Usually, people normalize the dualizing complex of R such that it has nonzero cohomology in degrees $[-\dim R, 0]$. When I use that normalization I have to use Grauert-Riemenschneider vanishing as in the article of Elkik. Cheers! $\endgroup$
    – Johan
    Jul 7 at 23:14
  • $\begingroup$ You are right: I misread the indexing convention. I will correct it now. $\endgroup$ Jul 8 at 0:57
  • $\begingroup$ I know that $\omega_Y^{\bullet}$ is only supported at $\dim Y$, but could you please explain why $\mathcal H^i(Rf_* \omega_Y^{\bullet })=0$ for $i\neq -\dim Y$? If we can just say this, we would be done at this point, no? Because $X$ is Cohen-Macaulay if and only if $\omega_X^{\bullet}$ is supported in a single degree ... $\endgroup$
    – Snake Eyes
    Jul 8 at 1:36
  • $\begingroup$ As pointed out by user @Johan (and explained in the article by Elkik), this follows from Grauert-Riemenschneider Vanishing. $\endgroup$ Jul 8 at 1:43
  • $\begingroup$ @JasonStarr: Thank you for your edits ... I do not have a reference in front of me right now, but for Grauert-Riemenschneider, don't you need both the schemes to be actually a variety over a characteristic $0$ field? $\endgroup$
    – Snake Eyes
    Jul 8 at 1:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.