In classical algebra, projective modules are the ‘well-behaved’ building blocks essential for homological algebra. Their definition makes sense in an arbitrary abelian category \(\Aa \): an object \(P\) is called projective if the Hom functor \(\Hom _{\Aa }(P,-) \colon \Aa \to \Ab \) is an exact functor of abelian categories. This definition has an immediate analogue in an arbitrary t-structure:
Definition 6.4.1. Let \(C\) be a stable \(\infty \)-category equipped with a t-structure. An object \(P \in C\) is called t-projective if the mapping spectrum functor \(\hom _C(P,-)\colon C \to \Sp \) is t-exact.
Observation 6.4.2. Every t-projective object \(P\) is connective. To see this, consider the truncation \(\tau _{\leq -1}P\). By t-exactness, the mapping spectrum \(\hom _C(P,\tau _{\leq -1} P)\) lies in \(\Sp _{\leq -1}\), and it follows that \[ 0 = \pi _0 \hom _C(P,\tau _{\leq -1} P) \cong \pi _0 \Hom _C(\tau _{\leq -1}P, \tau _{\leq -1}P), \] where the last identification holds by adjunction. In particular, the identity on \(\tau _{\leq -1}P\) can be identified with the zero map, forcing \(\tau _{\leq -1}P = 0\), hence \(P \in C_{\geq 0}\).
Observation 6.4.3. Conversely, if \(P \in C_{\geq 0}\) is a connective object, then \(\hom _C(P,-)\colon C \to \Sp \) is always left t-exact: if \(X \in C_{\leq 0}\), then for \(n > 0\) we have \[ \pi _n(\hom _C(P,X)) \cong \pi _0 \Hom _C(P,X[-n]) = 0, \] since \(P \in C_{\geq 0}\) while \(X[-n] \in C_{\leq -1}\). So \(P\) is t-projective if and only if \(\hom _C(P,-)\) is also right t-exact.
In the case \(C = \D (\Aa )\), this definition does indeed recover the usual notion of projectivity in \(\Aa \):
Lemma 6.4.4. Let \(\Aa \) be an abelian category and let \(M \in \Aa \). Then \(M\) is projective in \(\Aa \) if and only if the complex \(M[0] \in \D (\Aa )\) is t-projective.
Proof. Suppose first that \(M[0]\) is t-projective, so that \(\hom _{\D (\Aa )}(M[0],-)\colon \D (\Aa ) \to \Sp \) is t-exact. By Corollary 6.3.17, this functor restricts to an exact functor on hearts \(\D (\Aa )^{\heartsuit } \to \Sp ^{\heartsuit }\), i.e., \(\Aa \to \Ab \). This restricted functor is \(\Hom _{\Aa }(M,-)\), showing that \(M\) is projective in \(\Aa \).
Conversely, assume that \(\Hom _{\Aa }(M,-)\colon \Aa \to \Ab \) is exact. Just as in Corollary 6.2.9, this implies that for any chain complex \(C_{\bullet }\) in \(\Aa \) we have an isomorphism \[ \pi _n \hom _{\Kk (\Aa )}(M[0], C_{\bullet }) \cong \Hom _{\Aa }(M,H_n(C)). \] In particular, the functor \(\hom _{\Kk (\Aa )}(M[0],-)\colon \Kk (\Aa ) \to \Sp \) inverts quasi-isomorphisms. The same argument as in Lemma 6.2.10 shows that it is corepresented by \(M[0] \in \D (\Aa )\), giving a natural isomorphism \(\hom _{\D (\Aa )}(M[0],-) \cong \hom _{\Kk (\Aa )}(M[0],-)\). It follows that \(\pi _n\hom _{\D (\Aa )}(M[0],X) \cong \Hom _{\Aa }(M,H_n(X))\) for all complexes \(X \in \D (\Aa )\), showing that \(\hom _{\D (\Aa )}(M[0],X)\) is connective whenever \(X\) is. □
Corollary 6.4.5. Let \(\Aa \) be an abelian category and let \(M\in \Aa \) be projective. Then for every \(X\in \D (\Aa )\) and every \(n\in \Z \), there is a natural isomorphism \[ \pi _n\hom _{\D (\Aa )}(M[0],X)\cong \Hom _{\Aa }(M,H_n(X)). \]
Proof. This is the computation in the second half of the proof of Lemma 6.4.4. □
6.4.1 Ext-groups
We will next give a formulation of t-projectivity in terms of the Ext-groups in \(C\).
Definition 6.4.6 (Ext-groups). Let \(C\) be a stable \(\infty \)-category. For objects \(X,Y \in C\) and \(n \in \Z \), we define the \(n\)-th Ext-group to be \[ \Ext ^n_C(X,Y) := \pi _{-n}\hom _C(X,Y) \simeq \pi _0 \Hom _C(X,Y[n]). \]
For objects \(M,N\in \Aa \), regarded as complexes concentrated in degree zero, we abbreviate \(\Ext ^n_{\D (\Aa )}(M,N)\) to \(\Ext ^n_{\Aa }(M,N)\).
Assume now that \(C\) comes equipped with a t-structure. For objects \(M,N \in C^{\heartsuit }\), the Ext-groups \(\Ext ^n_C(M,N)\) admit a direct interpretation in small degrees:
Observation 6.4.7. For \(n < 0\), we have \(\Ext ^n_C(M,N) = 0\), since \(M \in C_{\geq 0}\) while \(N[n] \in C_{\leq n} \subseteq C_{\leq -1}\).
Observation 6.4.8. For \(n = 0\), we have \(\Ext ^0_C(M,N) = \pi _0\Hom _C(M,N) = \Hom _{C^{\heartsuit }}(M,N)\), the set of morphisms in the heart.
The interpretation of \(\Ext ^1_C(M,N)\) in terms of extensions is developed in Chapterexercise 6.5.
Observation 6.4.9 (Two-stage Postnikov extensions). For \(X\in C_{\geq 0}\cap C_{\leq 1}\), the truncation sequence takes the form \[ \pi _1(X)[1]\longrightarrow X\longrightarrow \pi _0(X) \xrightarrow {\ \kappa _X\ }\pi _1(X)[2]. \] The class \(\kappa _X\in \Ext ^2_C(\pi _0(X),\pi _1(X))\) is called the first Postnikov \(k\)-invariant of \(X\). Conversely, for \(M,N\in C^{\heartsuit }\) and a morphism \(\kappa \colon M\to N[2]\), the object \(X:=\fib (\kappa )\) lies in \(C_{\geq 0}\cap C_{\leq 1}\), with \(\pi _0(X)\cong M\) and \(\pi _1(X)\cong N\). Its truncation sequence is \(N[1]\to X\to M\xrightarrow {\kappa }N[2]\), so it recovers \(\kappa \). Thus taking fibers gives a bijection between \(\Ext ^2_C(M,N)\) and isomorphism classes of objects \(X\in C_{\geq 0}\cap C_{\leq 1}\) equipped with identifications \(\pi _0(X)\cong M\) and \(\pi _1(X)\cong N\), where isomorphisms must respect these identifications. The zero class corresponds to \(M\oplus N[1]\). Thus the \(k\)-invariant records the information needed to reconstruct \(X\) from its two homotopy group objects.
Exercise 6.4.10. Let \(\Aa \) be an abelian category with enough projectives, meaning that every object admits an epimorphism from a projective object. For \(M,N\in \Aa \) and \(n\geq 2\), show that \(\Ext ^n_{\Aa }(M,N)=\Ext ^n_{\D (\Aa )}(M,N)\) can be described in terms of degree \(n\) Yoneda extensions: exact sequences in \(\Aa \) of the form \[ 0 \to N \to E_n \to E_{n-1} \to \dots \to E_2 \to E_1 \to M \to 0. \] Here extensions are identified under the equivalence relation generated by morphisms of exact sequences that are the identity on \(M\) and \(N\). Hint: use Proposition 6.4.16 below.
We now record the fundamental long exact sequences for Ext-groups.
Observation 6.4.11 (Ext long exact sequence). Every exact sequence \(Y' \to Y \to Y''\) in \(C\) induces an exact sequence of mapping spectra \(\hom _C(X,Y') \to \hom _C(X,Y) \to \hom _C(X,Y'')\), thus giving rise to a long exact sequence of homotopy groups of the form
Similarly, every exact sequence \(X' \to X \to X''\) in \(C\) induces a long exact sequence of the form \[ \begin {aligned} \dots \to \Ext ^{n-1}_C(X',Y) \to \Ext ^n_C(X'',Y) \to \Ext ^n_C(X,Y) \to \Ext ^n_C(X',Y) \to \Ext ^{n+1}_C(X'',Y) \to \dots . \end {aligned} \] As a special case, given an object \(N \in C^{\heartsuit }\) and a short exact sequence \(M' \hookrightarrow P \twoheadrightarrow M\) in \(C^{\heartsuit }\), the negative Ext-groups vanish, giving a one-sided long exact sequence of the form \[ \begin {aligned} 0 \to \Hom _{C^{\heartsuit }}(&M,N) \to \Hom _{C^{\heartsuit }}(P,N) \to \Hom _{C^{\heartsuit }}(M',N) \\ &\to \Ext ^1_C(M,N) \to \Ext ^1_C(P,N) \to \Ext ^1_C(M',N) \\ &\to \Ext ^2_C(M,N) \to \dots . \end {aligned} \]
Using Ext-groups, we can give several equivalent characterizations of t-projectivity.
Proposition 6.4.12 ([Lurie (2017), Proposition 7.2.2.6]). Let \(C\) be a stable \(\infty \)-category equipped with a t-structure, and let \(P \in C_{\geq 0}\) be a connective object. Then the following conditions are equivalent:
- (1)
- (2)
-
For every \(Q \in C_{\geq 0}\) and every \(i > 0\), the abelian group \(\Ext ^i_C(P,Q)\) is zero;
- (3)
-
For every \(Q \in C_{\geq 0}\), the abelian group \(\Ext ^1_C(P,Q)\) is zero;
- (4)
-
Given an exact sequence \[ N' \to N \to N'' \] in \(C\) with \(N',N,N'' \in C_{\geq 0}\), the map \(\Ext ^0_C(P,N) \to \Ext ^0_C(P,N'')\) is surjective.
If \(C\) is left complete, this is further equivalent to:
Proof. Recall from Observation 6.4.3 that \(P\) is t-projective if and only if the mapping spectrum functor \(\hom _C(P,-)\colon C \to \Sp \) is right t-exact. By definition, this means that for any object \(Q \in C_{\geq 0}\), the mapping spectrum \(\hom _C(P,Q)\) has vanishing negative homotopy groups \(\pi _{-i}\hom _C(P,Q) = \Ext ^i_C(P,Q)\) for \(i > 0\). This shows the equivalence \(\text{(1)} \Leftrightarrow \text{(2)}\).
It is clear that (2) implies (3), and the implication \(\text{(3)} \Rightarrow \text{(2)}\) follows by replacing \(Q\) by \(Q[i-1]\) for all \(i \geq 1\).
We next show that \(\text{(3)} \Leftrightarrow \text{(4)}\). It is clear that (3) implies (4): the exact sequence \(N' \to N \to N''\) induces a long exact sequence of the form \[ \dots \to \Ext ^0_C(P,N) \to \Ext ^0_C(P,N'') \to \Ext ^1_C(P,N') \to \dots , \] so the vanishing of \(\Ext ^1_C(P,N')\) implies that the first map must be surjective. Assume now that (4) holds. Given a class \(\eta \in \Ext ^1(P,Q) = \pi _0\Hom _C(P,Q[1])\), we obtain an exact sequence \[ P' \to P \xrightarrow {\eta } Q[1] \] and thus a long exact sequence \[ \dots \to \Ext ^0_C(P,P') \to \Ext ^0_C(P,P) \xrightarrow {\eta \circ -} \Ext ^1_C(P,Q) \to \dots \] As \(P,P',Q \in C_{\geq 0}\), the first map is surjective, hence the second map is the zero map. Since it sends \(\id _P\) to \(\eta \), we conclude that \(\eta = 0\). This shows that \(\text{(3)} \Leftrightarrow \text{(4)}\).
Finally, suppose that \(C\) is left complete. It is clear that (2) implies (5). Conversely, assume (5). The exact sequences \[ (\pi _{n+1}Q)[n+1] \to \tau _{\leq n+1}Q \to \tau _{\leq n}Q \] show that the towers of Ext-groups in each positive degree are eventually constant. Since left completeness gives \(Q \simeq \lim _n \tau _{\leq n}Q\), the corresponding tower of mapping spectra computes \(\hom _C(P,Q)\). Eventual constancy in adjacent degrees eliminates the derived-limit term and gives \[ \Ext ^i_C(P,Q) \cong \Ext ^i_C(P,\tau _{\leq 0}Q)=0 \] for \(i>0\). This is the Postnikov-tower argument of [Lurie (2017), Proposition 7.2.2.6]. □
Remark 6.4.13. More generally, an object \(P\) of an \(\infty \)-category \(D\) with geometric realizations is called projective if \(\Hom _D(P,-)\) preserves geometric realizations [Lurie (2009b), Definition 5.5.8.18]. If \(C\) is a stable \(\infty \)-category with a left-complete t-structure, this notion of projectivity in \(C_{\geq 0}\) agrees with t-projectivity by [Lurie (2017), Proposition 7.2.2.6] and Proposition 6.4.12.
6.4.2 Ext-groups via projective resolutions
As a consequence of Proposition 6.4.12, we may compute Ext-groups in terms of projective resolutions.
Definition 6.4.14. Let \(C\) be a stable \(\infty \)-category equipped with a t-structure, and let \(M \in C^{\heartsuit }\) be an object in the heart of \(C\). A projective resolution of \(M\) is a long exact sequence \[ \dots \to P_2 \to P_1 \to P_0 \twoheadrightarrow M \to 0 \] in \(C^{\heartsuit }\) such that each object \(P_n\) is t-projective.
Remark 6.4.15. When \(C = \D (\Aa )\) is the derived \(\infty \)-category of some abelian category \(\Aa \), this recovers the usual notion of projective resolutions in \(\Aa \) from homological algebra by Lemma 6.4.4.
Proposition 6.4.16 (Ext-groups via projective resolutions). Let \(C\) be a stable \(\infty \)-category equipped with a t-structure, and let \(P_{\bullet }\) be a projective resolution of \(M \in C^{\heartsuit }\). Then for every other object \(N \in C^{\heartsuit }\) and every \(n \in \Z \) there is an isomorphism of abelian groups \[ \Ext ^n_C(M,N) \quad \cong \quad H^n(\Hom _{C^{\heartsuit }}(P_{\bullet },N)) \] between the \(n\)-th Ext-group and the \(n\)-th cohomology group of the cochain complex \(\Hom _{C^{\heartsuit }}(P_{\bullet },N)\).
Proof. The claim is clear for \(n < 0\) as both sides are zero: \(\Ext ^n_C(M,N) = \pi _0 \Hom _{C}(M,N[n]) = 0\) since \(M \in C_{\geq 0}\) and \(N[n] \in C_{\leq n} \subseteq C_{\leq -1}\).
For the remaining cases, we break the resolution into short exact sequences and use the vanishing of the higher Ext-groups of the \(P_k\) to splice the resulting long exact sequences together. Define auxiliary objects \(M_{-1} := M\) and \(M_k := \ker (P_k \twoheadrightarrow M_{k-1})\) for \(k \geq 0\). The projective resolution then admits the following factorization in \(C^{\heartsuit }\), where each of the diagonal sequences are exact:
For each of these short exact sequences, we obtain a long exact sequence on Ext-groups from Observation 6.4.11. Since the higher Ext-groups of the t-projective objects \(P_k\) vanish by Proposition 6.4.12, its initial part is \[ \begin {aligned} 0 \to \Hom _{C^{\heartsuit }}(M_{k-1},N) &\to \Hom _{C^{\heartsuit }}(P_k,N) \to \Hom _{C^{\heartsuit }}(M_k,N) \\ &\to \Ext ^1_C(M_{k-1},N) \to 0, \end {aligned} \] and its remaining terms give isomorphisms \[ \Ext ^{i+1}_C(M_{k-1},N) \cong \Ext ^i_C(M_k,N) \qquad (i\geq 1). \]
Since the map \(P_{k+1} \twoheadrightarrow M_k\) is an epimorphism, the induced map \(\Hom _{C^{\heartsuit }}(M_k,N) \hookrightarrow \Hom _{C^{\heartsuit }}(P_{k+1},N)\) is injective, and so the initial part of this sequence provides for each \(k \geq 0\) an isomorphism \[ \Hom _{C^{\heartsuit }}(M_{k-1},N) \quad \cong \quad \ker (\Hom _{C^{\heartsuit }}(P_k,N) \to \Hom _{C^{\heartsuit }}(P_{k+1},N)). \] The next part of the sequence then shows that \(\Ext ^1_C(M_{k-1},N)\) is the cokernel of the map \(\Hom _{C^{\heartsuit }}(P_k,N) \to \Hom _{C^{\heartsuit }}(M_k,N)\). Combining this with the previous identification of \(\Hom _{C^{\heartsuit }}(M_k,N)\) as a kernel, this shows that \[ \Ext ^1_C(M_{k-1},N) \quad \cong \quad \frac {\ker (\Hom _{C^{\heartsuit }}(P_{k+1}, N) \to \Hom _{C^{\heartsuit }}(P_{k+2}, N))}{\im (\Hom _{C^{\heartsuit }}(P_{k}, N) \to \Hom _{C^{\heartsuit }}(P_{k+1}, N))} \quad = \quad H^{k+1}(\Hom _{C^{\heartsuit }}(P_{\bullet }, N)). \] Observe that setting \(k = 0\) in these two isomorphisms proves the statement of the proposition for \(n = 0\) and \(n = 1\). Iterating the displayed dimension-shifting isomorphisms gives \(\Ext ^n_C(M,N) \cong \Ext ^1_C(M_{n-2},N)\) for all \(n \geq 1\). Setting \(k = n-1\) in the previous isomorphism thus gives \[ \Ext ^n_C(M,N) \quad \cong \quad H^{n}(\Hom _{C^{\heartsuit }}(P_{\bullet }, N)) \] for all \(n \geq 1\), finishing the proof. □
Example 6.4.17 (A nontrivial Postnikov extension). Homology objects need not determine an object of a derived category. To see this, set \(R:=\Z /4\) and \(K:=R/(2)\cong \Z /2\), and consider the complex \[ A_{\bullet }:=\big (0\longrightarrow R\xrightarrow {\ 2\ }R\longrightarrow 0\big ) \] in \(\D (R)\), with the two copies of \(R\) in degrees \(1\) and \(0\). Its only nonzero homology groups are \[ H_1(A_{\bullet })\cong K \qquad \text {and}\qquad H_0(A_{\bullet })\cong K. \] The periodic projective resolution \(P_{\bullet }\to K\) \[ \cdots \xrightarrow {\ 2\ }R\xrightarrow {\ 2\ }R \xrightarrow {\ 2\ }R\twoheadrightarrow K\longrightarrow 0 \] and Proposition 6.4.16 show that \(\Ext ^n_R(K,K)\cong K\) for every \(n\geq 0\): after applying \(\Hom _R(-,K)\), all differentials vanish. The truncation sequence of \(A_{\bullet }\) identifies its \(k\)-invariant with the Yoneda \(2\)-extension \[ 0\longrightarrow K\xrightarrow {\ j\ }R\xrightarrow {\ 2\ }R \xrightarrow {\ q\ }K\longrightarrow 0, \] where \(q(r)=\bar r\) and \(j(\bar r)=2r\); see Exercise 6.4.10. This extension is obtained by splicing the short exact sequences \[ 0\longrightarrow K\xrightarrow {j}R\xrightarrow {2}2R\longrightarrow 0, \qquad 0\longrightarrow 2R\longrightarrow R\xrightarrow {q}K\longrightarrow 0, \] coming from the kernel and cokernel of the differential of \(A_{\bullet }\). Their connecting morphisms compose to the truncation map \(K\to K[2]\). To compute its class, lift the resolution to this extension by the identity maps in degrees \(0\) and \(1\). Since \(j q=2\), the resulting degree-two cocycle is \(q\colon P_2=R\to K\). It is nonzero and cannot be a coboundary, so \(\kappa _{A_{\bullet }}\) is the nonzero element of \(\Ext ^2_R(K,K)\).
By Observation 6.4.9, there are therefore exactly two isomorphism classes of objects in \(\D (R)\) with homology \(K\) in degrees \(0\) and \(1\) and zero elsewhere: \(K[0]\oplus K[1]\) and \(A_{\bullet }\). Indeed, the two classes in \(\Ext ^2_R(K,K)\) give these objects, and \(K\) has no nontrivial \(R\)-module automorphisms, so forgetting the chosen identifications of their homology does not identify them.