Let \(R\) be a commutative ring. For bounded below complexes of \(R\)-modules, we computed the derived tensor product using projective resolutions. For unbounded complexes we use the projective cofibrations on \(\Ch (R)\). The unit and tensor products of cofibrant complexes are again cofibrant, and tensoring with a cofibrant complex preserves quasi-isomorphisms.
Proposition 21.2.1 (Unbounded projective cofibrations, [Lurie (2017), Propositions 7.1.2.8 and 7.1.2.11]). Let \(W\) be the class of quasi-isomorphisms in \(\Ch (R)\). There is a class \(I_{\proj }\) of projective cofibrations with the following properties.
- (1)
-
The triple \((\Ch (R),W,I_{\proj })\) is a homotopy cocomplete \(\infty \)-category with weak equivalences and cofibrations.
- (2)
-
The unit \(R[0]\) is cofibrant, and a finite tensor product of cofibrant complexes is again cofibrant.
- (3)
-
If \(P\) is cofibrant, then \(P\otimes _R-\) is homotopy cocontinuous and preserves quasi-isomorphisms.
To see why \(P\otimes _R-\) preserves all quasi-isomorphisms in (3), recall from the proof of Lurie (2017), Proposition 7.1.2.8 that every cofibrant complex is a retract of one built from zero by successively adjoining free generators. Tensor this construction with an acyclic complex \(K\). Each inclusion at a successor stage is degreewise split with quotient a shift of \(K\), and at limit stages filtered colimits of \(R\)-modules are exact. It follows that \(P\otimes _R K\) is acyclic. Since tensoring commutes with mapping cones, the claim follows from Lemma 6.1.15.
Proposition 21.2.2 (The symmetric monoidal derived category). The derived \(\infty \)-category \(\D (R)\) admits a canonical symmetric monoidal structure with tensor product \(\otimes _R^{\bL }\) and unit \(R[0]\). The derived tensor product preserves small colimits separately in each variable, and the localization functor \(\gamma \colon \Ch (R)\to \D (R)\) admits a canonical lax symmetric monoidal refinement.
Proof. By Proposition 21.2.1, Remark 21.1.4, the tensor product on \(\Ch (R)\) is left derivable. Hence Theorem 21.1.6 equips \(\D (R)\) with a symmetric monoidal structure and \(\gamma \) with a lax symmetric monoidal refinement. Applying part (3) of Proposition 21.1.13 to the binary tensor product identifies the tensor product on \(\D (R)\) with the total left derived functor of \(\otimes _R\). Applied to the empty tensor product, it identifies the unit with \(\gamma (R[0])\).
For a cofibrant complex \(P\), the binary lax structure map gives a transformation \[ \gamma P\otimes _R^{\bL }\gamma Y\longrightarrow \gamma (P\otimes _R Y) \] natural in \(Y\). Choose a cofibrant replacement \(Q\to Y\). By the description of derived transport above, the structure map is represented by \(P\otimes _R Q\to P\otimes _R Y\), a quasi-isomorphism by part (3) of Proposition 21.2.1. Thus \(\gamma P\otimes _R^{\bL }-\) is the functor on localizations induced by \(P\otimes _R-\). Since \(P\otimes _R-\) is homotopy cocontinuous, its induced functor preserves small colimits by the dual of Theorem 2.2.11. Every object of \(\D (R)\) is represented by a cofibrant complex by part (1) of Corollary 2.2.9, and symmetry gives the assertion in the other variable. □
Corollary 21.2.3 (DGAs as derived algebras). The localization \(\gamma \) induces functors \[ \Alg (\Ch (R))\longrightarrow \Alg (\D (R)) \quad \qquadtext {and}\quad \CAlg (\Ch (R))\longrightarrow \CAlg (\D (R)). \] For an associative or commutative DGA \(A\) with multiplication \(\mu \), the multiplication on \(\gamma A\) is the composite \(\gamma A\otimes _R^{\bL }\gamma A\to \gamma (A\otimes _R A)\xrightarrow {\gamma (\mu )}\gamma A\).
Proof. Apply Proposition 21.2.2 and the fact that lax symmetric monoidal functors preserve associative and commutative algebras. □
Proposition 21.2.4. The full subcategory \(\D ^-(R)\subseteq \D (R)\) contains the unit and is closed under the derived tensor product. Its induced symmetric monoidal structure agrees with the bounded below construction of Example 21.1.7.
Proof. A complex is cofibrant in the unbounded projective structure if maps out of it lift against every degreewise surjective quasi-isomorphism [Lurie (2017), Proposition 7.1.2.8]. For a bounded below complex of projectives, the induction in Lemma 6.6.14 constructs these lifts starting at the lower bound of the source, regardless of bounds on the other two complexes. Thus if \(P\) and \(Q\) are bounded below projective resolutions, their ordinary tensor product \(P\otimes _R Q\) computes both derived tensor products. It is again bounded below, as is the unit \(R[0]\).
By part (4) of Theorem 21.1.6, the composite \(\Ch ^-(R)\hookrightarrow \Ch (R)\xrightarrow {\gamma }\D (R)\) induces a lax symmetric monoidal functor \(\D ^-(R)\hookrightarrow \D (R)\). On \(P,Q\) its binary structure map is represented by the identity of \(P\otimes _R Q\), and its unit map is the identity of \(R[0]\). Hence the inclusion is symmetric monoidal, giving the claimed identification by Lemma 14.7.1. □
We next describe the derived category in terms of perfect complexes. To include the tensor product in this description, we first explain how a tensor product on a small stable \(\infty \)-category extends to its Ind-completion.
Theorem 21.2.5 (Symmetric monoidal Ind-completion, [Lurie (2017), Corollary 4.8.1.14]). Let \(C\) be a small stable symmetric monoidal \(\infty \)-category, and assume that the tensor product on \(C\) is exact separately in both variables. Then \(\Ind (C)\) admits a unique symmetric monoidal structure with the following properties:
- (1)
-
The Yoneda embedding \(C\hookrightarrow \Ind (C)\) is symmetric monoidal.
- (2)
-
The tensor product on \(\Ind (C)\) preserves colimits separately in both variables.
For filtered-colimit presentations \(X\simeq \colim _iX_i\) and \(Y\simeq \colim _jY_j\) with \(X_i,Y_j\in C\), its underlying tensor product is characterized by \[ X\otimes Y\simeq \colim _{(i,j)}(X_i\otimes Y_j). \]
Proof. Equip \(\PSh (C)\) with Day convolution. Its tensor product preserves colimits separately in each variable, and the Yoneda embedding \(Y\colon C\to \PSh (C)\) is symmetric monoidal by Proposition 16.2.10, Proposition 19.5.2. Thus \[ \bigl (\colim _iY(c_i)\bigr )\otimes _{\Day }\bigl (\colim _jY(d_j)\bigr ) \simeq \colim _{(i,j)}Y(c_i\otimes d_j). \] Products of filtered \(\infty \)-categories are filtered, so \(\Ind (C)\) contains the unit and is closed under Day convolution. By Lemma 14.7.1, it inherits a symmetric monoidal structure with the asserted formula.
For \(c\in C\), the right adjoint to \(Y(c)\otimes _{\Day }-\) on \(\PSh (C)\) sends a presheaf \(F\) to \(F(c\otimes -)\), by Proposition 16.2.7 and Yoneda. This functor preserves \(\Ind (C)\): by Proposition 23.3.6, its objects are precisely the presheaves sending finite colimits in \(C\) to limits, and \(c\otimes -\) preserves finite colimits. The adjunction therefore restricts to \(\Ind (C)\), so \(Y(c)\otimes -\) preserves all colimits there. Every object of \(\Ind (C)\) is a filtered colimit of representables, and filtered colimits are computed in \(\PSh (C)\). Tensoring with an arbitrary object is consequently a filtered colimit of these colimit-preserving functors, and hence preserves all colimits. Symmetry gives the assertion in the other variable.
Finally, repeated application of the universal property of \(\Ind (C)\) from Proposition 23.1.2 shows that restriction to \(C^n\) is fully faithful on functors \(\Ind (C)^n\to \Ind (C)\) preserving filtered colimits separately in each variable. Thus the finite tensor products and all their coherence data are uniquely determined by their restrictions to \(C\). This proves uniqueness. □
We now return to the derived category of \(R\).
Corollary 21.2.6 (Perfect complexes). The unit \(R[0]\) is a compact generator of \(\D (R)\), and the inclusion of compact objects induces a symmetric monoidal equivalence \[ \Ind (\D (R)^\omega )\iso \D (R). \]
Proof. Since \(R\) is projective as an \(R\)-module, Corollary 6.4.5 gives \[ \pi _n\hom _{\D (R)}(R[0],X)\cong H_n(X) \] for every \(X\in \D (R)\). The canonical t-structure on \(\D (R)\) is compatible with filtered colimits by [Lurie (2017), Proposition 1.3.5.21], so each homology functor preserves filtered colimits. Since homotopy groups of spectra preserve filtered colimits by Lemma 4.4.28, the displayed formula shows that \(R[0]\) is compact. Its shifts jointly detect zero objects, since a complex is zero in \(\D (R)\) precisely when all its homology groups vanish. It follows from Proposition 23.3.4 that \(R[0]\) is a compact generator.
By the same proposition, \(\D (R)^\omega \) is small and is the smallest thick subcategory containing \(R[0]\). For each compact \(P\), the exact functor \(P\otimes _R^{\bL }-\) sends \(R[0]\) to \(P\), so it preserves this subcategory. By Proposition 23.3.7, the inclusion induces an equivalence \(\Ind (\D (R)^\omega )\iso \D (R)\). Since the derived tensor product preserves colimits separately in each variable, the uniqueness in Theorem 21.2.5 makes this equivalence symmetric monoidal. □
The compact objects of \(\D (R)\) are precisely the retracts of bounded complexes of finitely generated free \(R\)-modules. Indeed, filtering such a complex by degree builds it from shifts of finite sums of \(R[0]\), so it is compact. Conversely, the bounded complexes of finitely generated free modules are closed under shifts and cofibers. Any map between them in \(\D (R)\) is represented by a chain map by Lemma 6.6.32, and its mapping cone is again bounded and finite free in each degree. A map between retracts extends to a map between the original complexes by composing with the retraction from the source and the inclusion into the target. The original map is a retract of its extension, and the same holds for their cofibers. Thus the retracts form a thick subcategory containing \(R[0]\), so they include every compact object.
By Proposition 23.3.6, \(\D (R)\) is presentable. In Corollary 8.2.4 we apply symmetric monoidal Morita theory to its compact generator \(R[0]\) and identify its endomorphism ring spectrum with the Eilenberg–MacLane ring spectrum of \(R\). The resulting equivalence \(\D (R)\simeq \Mod _R(\Sp )\) sends \(R[0]\) to \(R\), so it identifies the subcategory \(\D (R)^\omega \) of perfect complexes with the perfect \(R\)-modules \(\Perf (R)\) of Definition 8.1.10.
Exercises
Exercise 21.1 (A tensor-stable localization). Let \(k\) be a field and let \(X\) be any (possibly unbounded) complex of \(k\)-vector spaces. Show that \(X\otimes _k-\) preserves quasi-isomorphisms and deduce that the localization functor \(\Ch (k)\to \D (k)\) is symmetric monoidal.
Exercise 21.2 (Derived transport over \([1]\)). Let \(E\to [1]\) be the cocartesian fibration classified by \(F\colon C_0\to C_1\). Suppose it is left derivable with respect to classes \(W_0\) and \(W_1\) in its fibers. Write \(\gamma \colon E\to E[(W_0\cup W_1)^{-1}]\) for the localization, \(\gamma _i\) for its restriction to \(C_i\), and \(\bL F\) for the total left derived functor of \(\gamma _1\circ F\).
- (1)
-
For \(X\in C_0\) and \(Y\in C_1\), identify the anima of maps from \(\gamma _0X\) to \(\gamma _1Y\) over \(0\to 1\) with \(\Hom _{C_1[W_1^{-1}]}((\bL F)(\gamma _0 X),\gamma _1Y)\).
- (2)
-
Under this identification, which morphism corresponds to the image of a cocartesian lift \(X\to F(X)\)? Deduce a criterion for this lift to remain cocartesian after localization.
Exercise 21.3 (The lax structure map and self-intersection). Let \(R\) be a commutative ring and let \(\gamma \colon \Ch (R)\to \D (R)\) be the localization functor.
- (1)
-
For \(C,D\in \Ch (R)\), describe the lax structure map \(\gamma C\otimes _R^{\bL }\gamma D\to \gamma (C\otimes _R D)\) using cofibrant replacements. Show that it is an isomorphism if either \(C\otimes _R-\) or \(-\otimes _R D\) preserves quasi-isomorphisms.
- (2)
-
Let \(x\in R\) be a non-zero-divisor and put \(A=R/(x)\). Use the Koszul resolution of Example 8.2.7 to compute \(A\otimes _R^{\bL }A\) and identify the lax structure map \(A\otimes _R^{\bL }A\to A\otimes _R A\simeq A\). What does it do on \(H_0\) and \(H_1\)? Specialize to \(R=\Z \) and \(x=p\) prime to obtain an example where the map fails to be an isomorphism.
Exercise 21.4 (Duality for a perfect complex). Let \(P\) be a bounded complex of finitely generated projective modules over a commutative ring \(R\). Show that \(P^\vee :=\iHom _R(P,R)\) is perfect and that evaluation induces an isomorphism of complexes \(P^\vee \otimes _R M\to \iHom _R(P,M)\) for every complex \(M\). Deduce that \(P^\vee \otimes _R^{\bL }M\to \bR \iHom _R(P,M)\) is an isomorphism for every \(M\in \D (R)\) and that \(P^\vee \) is a dual of \(P\).