13.18 Injective resolutions
In this section we prove some lemmas regarding the existence of injective resolutions in abelian categories having enough injectives.
Definition 13.18.1. Let $\mathcal{A}$ be an abelian category. Let $A \in \mathop{\mathrm{Ob}}\nolimits (\mathcal{A})$. An injective resolution of $A$ is a complex $I^\bullet $ together with a map $A \to I^0$ such that:
We have $I^ n = 0$ for $n < 0$.
Each $I^ n$ is an injective object of $\mathcal{A}$.
The map $A \to I^0$ is an isomorphism onto $\mathop{\mathrm{Ker}}(d^0)$.
We have $H^ i(I^\bullet ) = 0$ for $i > 0$.
Hence $A[0] \to I^\bullet $ is a quasi-isomorphism. In other words the complex
\[ \ldots \to 0 \to A \to I^0 \to I^1 \to \ldots \]
is acyclic. Let $K^\bullet $ be a complex in $\mathcal{A}$. An injective resolution of $K^\bullet $ is a complex $I^\bullet $ together with a map $\alpha : K^\bullet \to I^\bullet $ of complexes such that
We have $I^ n = 0$ for $n \ll 0$, i.e., $I^\bullet $ is bounded below.
Each $I^ n$ is an injective object of $\mathcal{A}$.
The map $\alpha : K^\bullet \to I^\bullet $ is a quasi-isomorphism.
In other words an injective resolution $K^\bullet \to I^\bullet $ gives rise to a diagram
\[ \xymatrix{ \ldots \ar[r] & K^{n - 1} \ar[d] \ar[r] & K^ n \ar[d] \ar[r] & K^{n + 1} \ar[d] \ar[r] & \ldots \\ \ldots \ar[r] & I^{n - 1} \ar[r] & I^ n \ar[r] & I^{n + 1} \ar[r] & \ldots } \]
which induces an isomorphism on cohomology objects in each degree. An injective resolution of an object $A$ of $\mathcal{A}$ is almost the same thing as an injective resolution of the complex $A[0]$.
Lemma 13.18.2. Let $\mathcal{A}$ be an abelian category. Let $K^\bullet $ be a complex of $\mathcal{A}$.
If $K^\bullet $ has an injective resolution then $H^ n(K^\bullet ) = 0$ for $n \ll 0$.
If $H^ n(K^\bullet ) = 0$ for all $n \ll 0$ then there exists a quasi-isomorphism $K^\bullet \to L^\bullet $ with $L^\bullet $ bounded below.
Proof.
Omitted. For the second statement use $L^\bullet = \tau _{\geq n}K^\bullet $ for some $n \ll 0$. See Homology, Section 12.15 for the definition of the truncation $\tau _{\geq n}$.
$\square$
Lemma 13.18.3. Let $\mathcal{A}$ be an abelian category. Assume $\mathcal{A}$ has enough injectives.
Any object of $\mathcal{A}$ has an injective resolution.
If $H^ n(K^\bullet ) = 0$ for all $n \ll 0$ then $K^\bullet $ has an injective resolution.
If $K^\bullet $ is a complex with $K^ n = 0$ for $n < a$, then there exists an injective resolution $\alpha : K^\bullet \to I^\bullet $ with $I^ n = 0$ for $n < a$ such that each $\alpha ^ n : K^ n \to I^ n$ is injective.
Proof.
Proof of (1). First choose an injection $A \to I^0$ of $A$ into an injective object of $\mathcal{A}$. Next, choose an injection $I_0/A \to I^1$ into an injective object of $\mathcal{A}$. Denote $d^0$ the induced map $I^0 \to I^1$. Next, choose an injection $I^1/\mathop{\mathrm{Im}}(d^0) \to I^2$ into an injective object of $\mathcal{A}$. Denote $d^1$ the induced map $I^1 \to I^2$. And so on. By Lemma 13.18.2 part (2) follows from part (3). Part (3) is a special case of Lemma 13.15.5.
$\square$
Lemma 13.18.4. Let $\mathcal{A}$ be an abelian category. Let $K^\bullet $ be an acyclic complex. Let $I^\bullet $ be bounded below and consisting of injective objects. Any morphism $K^\bullet \to I^\bullet $ is homotopic to zero.
Proof.
Let $\alpha : K^\bullet \to I^\bullet $ be a morphism of complexes. Note that $\alpha ^ j = 0$ for $j \ll 0$ as $I^\bullet $ is bounded below. In particular, we can find an $n$ such that there exist $h^ j : K^ j \to I^{j - 1}$ for $j \leq n$ such that $\alpha ^ j = d^{j - 1} \circ h^ j + h^{j + 1} \circ d^ j$ for $j < n$. We will show that there exists a morphism $h^{n + 1} : K^{n + 1} \to I^ n$ such that $\alpha ^ n = d^{n - 1} \circ h^ n + h^{n + 1} \circ d^ n$. Note that
\begin{align*} (\alpha ^ n - d^{n - 1} \circ h^{n - 1}) \circ d^{n - 1} & = \alpha ^{n - 1} \circ d^{n - 1} - d^{n - 1} \circ h^{n - 1} \circ d^{n - 1} \\ & = d^{n - 1} \circ \alpha ^{n - 1} - d^{n - 1} \circ h^{n - 1} \circ d^{n - 1} \\ & = d^{n - 1} \circ (d^{n - 2} \circ h^{n - 1} + h^ n \circ d^{n - 1}) - d^{n - 1} \circ h^{n - 1} \circ d^{n - 1} \\ & = 0 \end{align*}
Since $K^\bullet $ is acyclic we have $d^{n - 1}(K^{n - 1}) = \mathop{\mathrm{Ker}}(K^ n \to K^{n + 1})$. Hence we can think of $\alpha ^ n - d^{n - 1} \circ h^{n - 1}$ as a map into $I^ n$ defined on the subobject $\mathop{\mathrm{Im}}(K^ n \to K^{n + 1})$ of $K^{n + 1}$. By injectivity of the object $I^ n$ we can extend this to a map $h^{n + 1} : K^{n + 1} \to I^ n$. With this choice the reader checks that we indeed have $\alpha ^ n = d^{n - 1} \circ h^ n + h^{n + 1} \circ d^ n$.
By induction on $n$ we conclude we can find $h = (h^ j)_{j \in \mathbf{Z}}$ which forms a homotopy between $\alpha $ and $0$ as desired.
$\square$
Lemma 13.18.6. Let $\mathcal{A}$ be an abelian category. Consider a solid diagram
\[ \xymatrix{ K^\bullet \ar[r]_\alpha \ar[d]_\gamma & L^\bullet \ar@{-->}[dl]^\beta \\ I^\bullet } \]
where $I^\bullet $ is bounded below and consists of injective objects, and $\alpha $ is a quasi-isomorphism.
There exists a map of complexes $\beta $ making the diagram commute up to homotopy.
If $\alpha $ is injective in every degree then we can find a $\beta $ which makes the diagram commute.
Proof.
The “correct” proof of part (1) is explained in Remark 13.18.5. We also give a direct proof here.
We first show that (2) implies (1). Namely, let $\tilde\alpha : K \to \tilde L^\bullet $, $\pi $, $s$ be as in Lemma 13.9.6. Since $\tilde\alpha $ is injective by (2) there exists a morphism $\tilde\beta : \tilde L^\bullet \to I^\bullet $ such that $\gamma = \tilde\beta \circ \tilde\alpha $. Set $\beta = \tilde\beta \circ s$. Then we have
\[ \beta \circ \alpha = \tilde\beta \circ s \circ \pi \circ \tilde\alpha \sim \tilde\beta \circ \tilde\alpha = \gamma \]
as desired.
Assume that $\alpha : K^\bullet \to L^\bullet $ is injective. Suppose we have already defined $\beta $ in all degrees $\leq n - 1$ compatible with differentials and such that $\gamma ^ j = \beta ^ j \circ \alpha ^ j$ for all $j \leq n - 1$. Consider the commutative solid diagram
\[ \xymatrix{ K^{n - 1} \ar[r] \ar@/_2pc/[dd]_\gamma \ar[d]^\alpha & K^ n \ar@/^2pc/[dd]^\gamma \ar[d]^\alpha \\ L^{n - 1} \ar[r] \ar[d]^\beta & L^ n \ar@{-->}[d] \\ I^{n - 1} \ar[r] & I^ n } \]
Thus we see that the dotted arrow is prescribed on the subobjects $\alpha (K^ n)$ and $d^{n - 1}(L^{n - 1})$. Moreover, these two arrows agree on $\alpha (d^{n - 1}(K^{n - 1}))$. Hence if
13.18.6.1
\begin{equation} \label{derived-equation-qis} \alpha (d^{n - 1}(K^{n - 1})) = \alpha (K^ n) \cap d^{n - 1}(L^{n - 1}) \end{equation}
then these morphisms glue to a morphism $\alpha (K^ n) + d^{n - 1}(L^{n - 1}) \to I^ n$ and, using the injectivity of $I^ n$, we can extend this to a morphism from all of $L^ n$ into $I^ n$. After this by induction we get the morphism $\beta $ for all $n$ simultaneously (note that we can set $\beta ^ n = 0$ for all $n \ll 0$ since $I^\bullet $ is bounded below – in this way starting the induction).
It remains to prove the equality (13.18.6.1). The reader is encouraged to argue this for themselves with a suitable diagram chase. Nonetheless here is our argument. Note that the inclusion $\alpha (d^{n - 1}(K^{n - 1})) \subset \alpha (K^ n) \cap d^{n - 1}(L^{n - 1})$ is obvious. Take an object $T$ of $\mathcal{A}$ and a morphism $x : T \to L^ n$ whose image is contained in the subobject $\alpha (K^ n) \cap d^{n - 1}(L^{n - 1})$. Since $\alpha $ is injective we see that $x = \alpha \circ x'$ for some $x' : T \to K^ n$. Moreover, since $x$ lies in $d^{n - 1}(L^{n - 1})$ we see that $d^ n \circ x = 0$. Hence using injectivity of $\alpha $ again we see that $d^ n \circ x' = 0$. Thus $x'$ gives a morphism $[x'] : T \to H^ n(K^\bullet )$. On the other hand the corresponding map $[x] : T \to H^ n(L^\bullet )$ induced by $x$ is zero by assumption. Since $\alpha $ is a quasi-isomorphism we conclude that $[x'] = 0$. This of course means exactly that the image of $x'$ is contained in $d^{n - 1}(K^{n - 1})$ and we win.
$\square$
Lemma 13.18.7. Let $\mathcal{A}$ be an abelian category. Consider a solid diagram
\[ \xymatrix{ K^\bullet \ar[r]_\alpha \ar[d]_\gamma & L^\bullet \ar@{-->}[dl]^{\beta _ i} \\ I^\bullet } \]
where $I^\bullet $ is bounded below and consists of injective objects, and $\alpha $ is a quasi-isomorphism. Any two morphisms $\beta _1, \beta _2$ making the diagram commute up to homotopy are homotopic.
Proof.
This follows from Remark 13.18.5. We also give a direct argument here.
Let $\tilde\alpha : K \to \tilde L^\bullet $, $\pi $, $s$ be as in Lemma 13.9.6. If we can show that $\beta _1 \circ \pi $ is homotopic to $\beta _2 \circ \pi $, then we deduce that $\beta _1 \sim \beta _2$ because $\pi \circ s$ is the identity. Hence we may assume $\alpha ^ n : K^ n \to L^ n$ is the inclusion of a direct summand for all $n$. Thus we get a short exact sequence of complexes
\[ 0 \to K^\bullet \to L^\bullet \to M^\bullet \to 0 \]
which is termwise split and such that $M^\bullet $ is acyclic. We choose splittings $L^ n = K^ n \oplus M^ n$, so we have $\beta _ i^ n : K^ n \oplus M^ n \to I^ n$ and $\gamma ^ n : K^ n \to I^ n$. In this case the condition on $\beta _ i$ is that there are morphisms $h_ i^ n : K^ n \to I^{n - 1}$ such that
\[ \gamma ^ n - \beta _ i^ n|_{K^ n} = d \circ h_ i^ n + h_ i^{n + 1} \circ d \]
Thus we see that
\[ \beta _1^ n|_{K^ n} - \beta _2^ n|_{K^ n} = d \circ (h_1^ n - h_2^ n) + (h_1^{n + 1} - h_2^{n + 1}) \circ d \]
Consider the map $h^ n : K^ n \oplus M^ n \to I^{n - 1}$ which equals $h_1^ n - h_2^ n$ on the first summand and zero on the second. Then we see that
\[ \beta _1^ n - \beta _2^ n - (d \circ h^ n + h^{n + 1}) \circ d \]
is a morphism of complexes $L^\bullet \to I^\bullet $ which is identically zero on the subcomplex $K^\bullet $. Hence it factors as $L^\bullet \to M^\bullet \to I^\bullet $. Thus the result of the lemma follows from Lemma 13.18.4.
$\square$
Lemma 13.18.8. Let $\mathcal{A}$ be an abelian category. Let $I^\bullet $ be bounded below complex consisting of injective objects. Let $L^\bullet \in K(\mathcal{A})$. Then
\[ \mathop{\mathrm{Mor}}\nolimits _{K(\mathcal{A})}(L^\bullet , I^\bullet ) = \mathop{\mathrm{Mor}}\nolimits _{D(\mathcal{A})}(L^\bullet , I^\bullet ). \]
Proof.
Let $a$ be an element of the right hand side. We may represent $a = \gamma \alpha ^{-1}$ where $\alpha : K^\bullet \to L^\bullet $ is a quasi-isomorphism and $\gamma : K^\bullet \to I^\bullet $ is a map of complexes. By Lemma 13.18.6 we can find a morphism $\beta : L^\bullet \to I^\bullet $ such that $\beta \circ \alpha $ is homotopic to $\gamma $. This proves that the map is surjective. Let $b$ be an element of the left hand side which maps to zero in the right hand side. Then $b$ is the homotopy class of a morphism $\beta : L^\bullet \to I^\bullet $ such that there exists a quasi-isomorphism $\alpha : K^\bullet \to L^\bullet $ with $\beta \circ \alpha $ homotopic to zero. Then Lemma 13.18.7 shows that $\beta $ is homotopic to zero also, i.e., $b = 0$.
$\square$
Lemma 13.18.9. Let $\mathcal{A}$ be an abelian category. Assume $\mathcal{A}$ has enough injectives. For any short exact sequence $0 \to A^\bullet \to B^\bullet \to C^\bullet \to 0$ of $\text{Comp}^{+}(\mathcal{A})$ there exists a commutative diagram in $\text{Comp}^{+}(\mathcal{A})$
\[ \xymatrix{ 0 \ar[r] & A^\bullet \ar[r] \ar[d] & B^\bullet \ar[r] \ar[d] & C^\bullet \ar[r] \ar[d] & 0 \\ 0 \ar[r] & I_1^\bullet \ar[r] & I_2^\bullet \ar[r] & I_3^\bullet \ar[r] & 0 } \]
where the vertical arrows are injective resolutions and the rows are short exact sequences of complexes. In fact, given any injective resolution $A^\bullet \to I^\bullet $ we may assume $I_1^\bullet = I^\bullet $.
Proof.
Step 1. Choose an injective resolution $A^\bullet \to I^\bullet $ (see Lemma 13.18.3) or use the given one. Recall that $\text{Comp}^{+}(\mathcal{A})$ is an abelian category, see Homology, Lemma 12.13.9. Hence we may form the pushout along the map $A^\bullet \to I^\bullet $ to get
\[ \xymatrix{ 0 \ar[r] & A^\bullet \ar[r] \ar[d] & B^\bullet \ar[r] \ar[d] & C^\bullet \ar[r] \ar[d] & 0 \\ 0 \ar[r] & I^\bullet \ar[r] & E^\bullet \ar[r] & C^\bullet \ar[r] & 0 } \]
Because of the $5$-lemma and the last assertion of Homology, Lemma 12.13.12 the map $B^\bullet \to E^\bullet $ is a quasi-isomorphism. Note that the lower short exact sequence is termwise split, see Homology, Lemma 12.27.2. Hence it suffices to prove the lemma when $0 \to A^\bullet \to B^\bullet \to C^\bullet \to 0$ is termwise split.
Step 2. Choose splittings. In other words, write $B^ n = A^ n \oplus C^ n$. Denote $\delta : C^\bullet \to A^\bullet [1]$ the morphism as in Homology, Lemma 12.14.10. Choose injective resolutions $f_1 : A^\bullet \to I_1^\bullet $ and $f_3 : C^\bullet \to I_3^\bullet $. (If $A^\bullet $ is a complex of injectives, then use $I_1^\bullet = A^\bullet $.) We may assume $f_3$ is injective in every degree. By Lemma 13.18.6 we may find a morphism $\delta ' : I_3^\bullet \to I_1^\bullet [1]$ such that $\delta ' \circ f_3 = f_1[1] \circ \delta $ (equality of morphisms of complexes). Set $I_2^ n = I_1^ n \oplus I_3^ n$. Define
\[ d_{I_2}^ n = \left( \begin{matrix} d_{I_1}^ n
& (\delta ')^ n
\\ 0
& d_{I_3}^ n
\end{matrix} \right) \]
and define the maps $B^ n \to I_2^ n$ to be given as the sum of the maps $A^ n \to I_1^ n$ and $C^ n \to I_3^ n$. Everything is clear.
$\square$
Comments (3)
Comment #8418 by Elías Guisado on
Comment #8782 by Elías Guisado on
Comment #9042 by Stacks project on