I am reading A. Yekutieli's original article on dualizing complexes for noncommutative algebras and I found a problem I cannot solve.

First, some background. We start with a field $k$ and a noetherian $\mathbb N$-graded algebra $A$ and assume that $A_0 = k$. All objects are graded, and we write $\mathcal A$ for the category of graded $A$-bimodules. A dualizing complex is an object $R^\bullet$ in the derived category $\mathcal D(\mathcal A)$ satisfying the following:

  1. Its cohomology modules are finitely generated as left and right $A$-modules.

  2. If we look at $R^\bullet$ as a complex of left $A$-modules, then it has finite injective dimension, and the same holds exchangng left for right.

  3. The obvious morphisms $A \to \mathcal R Hom_{A^\circ}(R^\bullet, R^\bullet)$ and $A^\circ \to \mathcal R Hom_{A}(R^\bullet, R^\bullet)$ are isomorphisms in the category $\mathcal D (\mathcal A)$.

These "obvious morphism" come from the following setup: By 2 we may replace $R^\bullet$ with a bounded complex $I^\bullet$ such that each $I^i$ is injective as left and right $A$-module (but not necessarily as bimodule). Thus for every complex of left $A$-modules $M^\bullet$ there is a map, defined at the level of chain complexes \begin{align*} M^\bullet \to Hom_{A^\circ}(Hom_A(M^\bullet, I^\bullet), I^\bullet) \end{align*} given by simple evaluation. Passing to the derived category, the complex on the r.h.s. is isomorphic to $\mathcal R Hom_{A^\circ}(\mathcal R Hom_{A}(M^\bullet, R^\bullet), R^\bullet)$. Putting $M^\bullet = A$ gives us the morphism of point 3 of the above definition (if $M^\bullet$ is a complex of bimodules then the map respects this structure), and in the paper Yekutieli proves that this map is an isomorphism for $A$ if and only if it is an isomorphism whenever $M^\bullet$ is a bounded complex with finitely generated cohomology modules. So far, so good.

Dualizing complexes, when they exists, are not unique. In order to fix a special class of isomorphisms Yekutieli introduces local cohomology. Set $\Gamma_m(-) = \varinjlim Hom_A(A/A_{\geq n},-)$, where $A_{\geq n}$ denotes the ideal of elements of $A$ of degree at least $n$. This is a left exact functor, with a right derived functor $\mathcal R \Gamma_m$ defined over left bounded complexes of left $A$-modules. Then he states that there is a map similar to that above, \begin{align*} \mathcal R \Gamma_m(M^\bullet) \to \mathcal R Hom_{A^\circ}(\mathcal R Hom_{A}(M^\bullet, R^\bullet), \mathcal R \Gamma_m(R^\bullet)) \end{align*} It is this map I am trying to make sense of. By looking at it simply at the level of complexes, we might feel tempted to replace $M^\bullet$ by an injective resolution $J^\bullet$, reuse our old resolution of $R^\bullet$ called $I^\bullet$, and thus obtain a map \begin{align*} \Gamma_m(J^\bullet) \to Hom_{A^\circ}(Hom_{A}(J^\bullet, I^\bullet), \mathcal \Gamma_m(I^\bullet)) \end{align*} given again by evaluation. This is indeed well-defined, but the problem is that passing to the derived category, it is not immediate, at least to me, that the complex on the r.h.s. is isomorphic to $\mathcal R Hom_{A^\circ}(\mathcal R Hom_{A}(M^\bullet, R^\bullet), \mathcal R \Gamma_m(R^\bullet))$. In order for that to happen, $\Gamma_m(I^\bullet)$ should consist of $A$-module which are injective as right $A$-modules, and I see no reason for that to be true (in the article it is proven that these modules are injective as left $A$-modules).

Any ideas on how to fix this? Have I made a mistake with the definition of this last morphism?


2 Answers 2


I happened to be looking at this very issue a few days ago! Here is my explanation to myself...

Indeed, in the notation above, there is a homomorphism of complexes of graded modules (not derived) $$ \zeta : \Gamma_{\mathfrak{m}} (J) \to \mathrm{Hom}_{A^{\mathrm{op}}} ( \mathrm{Hom}_A ( J, I ), \Gamma_{\mathfrak{m}} (I) ) . $$ And $$ \Gamma_{\mathfrak{m}} (J) \cong \mathrm{R} \Gamma_{\mathfrak{m}} (M) . $$ But now we can choose a quasi-isomorphism $ \Gamma_{\mathfrak{m}} (I) \to K $ where $K$ is a complex of bimodules, injective on the right (and bounded below). We then get a morphism $$ \eta : \mathrm{Hom}_{A^{\mathrm{op}}} ( \mathrm{Hom}_A ( J, I ), \Gamma_{\mathfrak{m}} (I) ) \to \mathrm{Hom}_{A^{\mathrm{op}}} ( \mathrm{Hom}_A ( J, I ), K) \cong \mathrm{RHom}_{A^{\mathrm{op}}} ( \mathrm{RHom}_A ( M, R ), \mathrm{R}\Gamma_{\mathfrak{m}} (R) ) . $$ The morphism we want is $\eta \circ \zeta$. It is functorial.

In case $A$ is noetherian, we know that $ \Gamma_{\mathfrak{m}} (I) $ is already injective on the right (as mentioned in the previous answer), and then $\eta$ is an isomorphism.

I hope this makes sense!

Added 27 Nov 2018:

  1. When $A$ is noetherian connected, and $R$ is a balanced DC, then $\mathrm{R} \Gamma_{\mathfrak{m}}(R)$ is isomorphic to $A^{*}$, and this is graded-injective on both sides.
  2. In the book (in preparation) Derived Categories (prepulication ver 3) there is an improved treatment of local duality in the NC connected graded case -- see Thm 17.2.7.

You can consult [Michel Van den Bergh, Existence theorems for dualizing complexes over noncommutative graded and filtered rings, Lemma 4.1]. Your problem can be solved when $A$ is assumed to be Noetherian.

  • $\begingroup$ Just to clarify, said lemma states that if $E$ is an injective $A$-bimodule then $\Gamma_m(E)$ is a direct limit of injective right modules, which is also injective when $A$ is noetherian. Still, I am curious about Yekutieli's construction, which is done in general for coherent algebras. $\endgroup$ Oct 2, 2013 at 11:38

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.