3
$\begingroup$

Let $\mathbf A$ be a dg-category. Denote by $\mathsf{C}_{\mathrm{dg}}(\mathbf A)$ the dg-category of right $\mathbf A$-modules, and by $\mathsf{C}(\mathbf A) = Z^0(\mathsf{C}_{\mathrm{dg}}(\mathbf A))$, its underlying category (which is endowed with the projective model category structure). Moreover, set $\mathsf{K}(\mathbf A) = H^0(\mathsf{C}_{\mathrm{dg}}(\mathbf A))$, the homotopy category of modules, and let $\mathsf{D}(\mathbf A)$ be the derived category of $\mathbf A$, that is, the localisation of $\mathsf{K}(\mathbf A)$ (or, equivalently, of $\mathsf{C}(\mathbf A)$) along quasi-isomorphisms. Denote by $\delta \colon \mathsf{K}(\mathbf A) \to \mathsf{D}(\mathbf A)$ the localisation functor. The machinery of model categories tells us that, if $P \in \mathsf{C}(\mathbf A)$ is a cofibrant dg-module, then, for any dg-module $M$, the localisation functor induces an isomorphism \begin{equation} \mathsf{K}(\mathbf A)(P, M) \xrightarrow{\sim} \mathsf{D}(\mathbf A)(P,M). \end{equation}

The question is: is the above result true if $P$ is a h-projective dg-module? By definition, $P$ is h-projective if, whenever $A$ is an acyclic dg-module, the hom-complex $\mathsf{C}_{\mathrm{dg}}(\mathbf A)(P, A)$ is acyclic. Notice that a cofibrant dg-module is also h-projective.

$\endgroup$

1 Answer 1

3
$\begingroup$

The answer is: yes, if $P$ is h-projective, then the map \begin{equation} \delta_{P,M} \colon \mathsf{K}(\mathbf A)(P, M) \to \mathsf{D}(\mathbf A)(P,M) \end{equation} is an isomorphism, for any dg-module $M$.

The key argument in the proof is the following: assuming $P$ h-projective, then any quasi-isomorphism $u \colon X \to P$ has a right inverse in $\mathsf{K}(\mathbf A)$. In fact, consider the distinguished triangle in $\mathsf{K}(\mathbf A)$: \begin{equation} X \xrightarrow{u} P \xrightarrow{j} C(u) \to X[1]. \end{equation} Since $u$ is quasi-isomorphism, the cone $C(u)$ is acyclic. $P$ is h-projective, so we have \begin{equation} \mathsf{K}(\mathbf A)(P, C(u)) = H^0(\mathsf{C}_{\mathrm{dg}}(\mathbf A)(P, C(u))) = 0; \end{equation} in particular, $j=0$ in $\mathsf{K}(\mathbf A)$. Applying the cohomological functor $\mathsf{K}(\mathbf A)(P,-)$ to the above triangle, we immediately find $u' \colon P \to X$ such that $u u' = 1$.

Now, let us prove that $\delta_{P,M} = \delta$ is injective. Let $f, g \colon P \to M$ such that $\delta(f) = \delta(g)$. This means that there exists $u \colon X \to P$ such that $fu=gu$ in $\mathsf{K}(\mathbf A)$. By the above remark, $u$ has a right inverse $u'$ in $\mathsf{K}(\mathbf A)$, and so $f=g$. Finally, let us prove that $\delta$ is surjective. Let $\overline{f} \in \mathsf{D}(\mathbf A)(P, M)$ We may write $\overline{f} = \delta(f)\delta(s)^{-1}$, with $s \colon X \to P$ being a quasi-isomorphism, and $f \colon X \to M$. The above remark gives us a right inverse $s' \colon P \to X$ of $s$. Hence, $\delta(s')$ is the (two-sided) inverse of $\delta(s)$, and we have \begin{equation} \overline{f} = \delta(f) \delta(s') = \delta(fs'), \end{equation} and we are done.

$\endgroup$

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.