The evaluation pairing between a finite-dimensional vector space and its linear dual has an analogue in any symmetric monoidal \(\infty \)-category. Throughout this section, we fix a symmetric monoidal \(\infty \)-category \((C,\otimes ,\unit )\).

Definition 11.1.1 (Dualizable object). Consider objects \(X\) and \(Y\) of \(C\). We say that a morphism \(\ev \colon Y \otimes X \to \unit \) exhibits \(Y\) as a dual to \(X\) if there exists a morphism \(\coev \colon \unit \to X \otimes Y\) such that the following triangle diagrams commute:

Commutative diagram involving X tensor Y tensor X, X, with arrows labeled identity tensor ev, coev tensor identity.
Commutative diagram involving Y tensor X tensor Y, Y, with arrows labeled ev tensor identity, identity tensor coev.

In this case we refer to \(\ev \) as the evaluation map and to \(\coev \) as the coevaluation map, and call \((Y,\ev )\) a duality datum for \(X\). We say that an object \(X\) is dualizable if it admits such a duality datum.

Example 11.1.2 (Linear duality). Let \(V\) be a finite-dimensional vector space over a field \(k\), and write \(V^*=\Hom _k(V,k)\). In the symmetric monoidal 1-category \((\Vect _k,\otimes _k,k)\), the evaluation pairing \(\ev \colon V^*\otimes _k V\to k\) exhibits \(V^*\) as a dual to \(V\). For a basis \(v_1,\ldots ,v_r\) with dual basis \(v_1^*,\ldots ,v_r^*\), the coevaluation is given by \[ \coev \colon k\to V\otimes _k V^*,\qquad 1\mapsto \sum _{i=1}^r v_i\otimes v_i^*. \] This tensor corresponds to \(\id _V\) under the natural isomorphism \(V\otimes _k V^*\cong \Hom _k(V,V)\), so it is independent of the basis. The triangle identities follow from the dual-basis identities.

Example 11.1.3 (Finitely generated projective modules). Let \(R\) be an ordinary commutative ring. In the category of ordinary \(R\)-modules with tensor product \(\otimes _R\), the dualizable objects are precisely the finitely generated projective modules.

Indeed, such a module \(P\) is a retract of \(R^r\), so there are elements \(p_i\in P\) and linear maps \(\lambda _i\colon P\to R\) with \(p=\sum _{i=1}^r\lambda _i(p)p_i\) for every \(p\in P\). The evaluation pairing with \(P^*=\Hom _R(P,R)\) and the coevaluation \(1\mapsto \sum _i p_i\otimes \lambda _i\) satisfy the triangle identities.

Conversely, if \(Q\) is a dual of \(P\), write \(\coev (1)=\sum _{i=1}^r p_i\otimes q_i\) and set \(\lambda _i(p)=\ev (q_i\otimes p)\). The triangle identity gives \(p=\sum _i\lambda _i(p)p_i\). Thus the maps \(P\to R^r\), \(p\mapsto (\lambda _i(p))_i\), and \(R^r\to P\), \((a_i)_i\mapsto \sum _i a_i p_i\), exhibit \(P\) as a retract of \(R^r\).

The following proposition characterizes duality by an adjunction: the evaluation map exhibits \(-\otimes Y\) as right adjoint to \(-\otimes X\). In particular, it implies that the duality datum \((Y,\ev )\) is unique if it exists.

Proposition 11.1.4. Consider objects \(X\) and \(Y\) of \(C\) and consider a morphism \(\ev \colon Y \otimes X \to \unit \) in \(C\). The following conditions are equivalent:

(1)

The morphism \(\ev \colon Y \otimes X \to \unit \) exhibits \(Y\) as a dual to \(X\).

(2)

For all objects \(Z\) and \(W\) of \(C\), the composite \[ \Hom _{C}(W,Z \otimes Y) \xrightarrow {- \otimes X} \Hom _{C}(W \otimes X, Z \otimes Y \otimes X) \xrightarrow {\ev \circ -} \Hom _{C}(W \otimes X, Z) \] is an equivalence.

(3)

The map in (2) is an equivalence for \((W,Z) = (\unit ,X)\) and for \((W,Z) = (Y,\unit )\).

Proof. We first show that (1) implies (2). Let \(\coev \colon \unit \to X \otimes Y\) be the corresponding coevaluation map. It then follows directly from the triangle identities that an inverse to the map in (2) is given by the composite \[ \Hom _{C}(W \otimes X, Z) \xrightarrow {- \otimes Y} \Hom _{C}(W \otimes X \otimes Y, Z \otimes Y) \xrightarrow {- \circ \coev } \Hom _{C}(W,Z \otimes Y). \] The implication (2)\(\Rightarrow \)(3) is immediate. For (3)\(\Rightarrow \)(1), the first test produces the coevaluation and one triangle identity; the second detects the other triangle identity. Indeed, the test \((W,Z)=(\unit ,X)\) gives a map \(\coev \colon \unit \to X\otimes Y\) whose image is \(\id _X\), so the composite \[ X \xrightarrow {\coev \otimes \id } X \otimes Y \otimes X \xrightarrow {\id \otimes \ev } X \] is identified with \(\id _X\). The test \((W,Z)=(Y,\unit )\) gives an equivalence \[ \Hom _C(Y,Y) \xrightarrow {- \otimes X} \Hom _C(Y \otimes X, Y \otimes X) \xrightarrow {\ev \circ -} \Hom _C(Y \otimes X, \unit ). \] Both \(\id _Y\) and \((\ev \otimes \id )\circ (\id \otimes \coev )\) map to \(\ev \): for the latter, this follows from the first triangle identity and the commutative diagram

Commutative diagram involving Y tensor X, Y tensor X tensor Y tensor X, Ytensor X, unit, with arrows labeled identity tensor coev tensor identity, identity tensor identity tensor ev, ev tensor identity tensor identity, ev.

The two maps are therefore identified, proving the second triangle identity. □

Thus being a duality datum is a property of the evaluation map \(\ev \): once condition (2) holds, the required coevaluation and triangle identities follow. We will use this criterion below to establish duality by constructing and analyzing an evaluation map.

Convention 11.1.5. Let \(X\) be a dualizable object of \(C\). By condition (2) of Proposition 11.1.4, its dual represents the functor \(\Hom _C(-\otimes X,\unit )\colon C\catop \to \An \). In particular, the dual is unique, and we denote it by \(X^{\vee }\). An object representing this functor need not be a dual when \(X\) is not known to be dualizable; see the discussion of weak duals in Subsection 11.1.1.

Definition 11.1.6 (Invertible object). An object \(X\) of \(C\) is called invertible if there exists another object \(X^{-1}\) and an isomorphism \(X \otimes X^{-1} \cong \unit \). Equivalently, \(X\) is invertible if the functor \(X \otimes -\colon C \to C\) is an equivalence.

The full subanima \(\Pic (C) \subseteq C^{\simeq }\) spanned by the invertible objects is called the Picard anima of \(C\).

Lemma 11.1.7. Every invertible object is dualizable.

Proof. Let \(X\) be an invertible object and let \(X^{-1}\) be an inverse with isomorphism \(\ev \colon X^{-1} \otimes X \iso \unit \). We claim that \(\ev \) exhibits \(X^{-1}\) as a dual to \(X\). Using Proposition 11.1.4 we must show that for all objects \(Z\) and \(W\) of \(C\) the composite \[ \Hom _{C}(W,Z \otimes X^{-1}) \xrightarrow {- \otimes X} \Hom _{C}(W \otimes X, Z \otimes X^{-1} \otimes X) \xrightarrow {\ev \circ -} \Hom _{C}(W \otimes X, Z) \] is an equivalence. But both maps are equivalences: the first because the functor \(- \otimes X\colon C \to C\) is an equivalence, and the second because \(\ev \) is an isomorphism. □

In particular, the unit is dualizable, with itself as dual.

Lemma 11.1.8. Symmetric monoidal functors preserve dualizable objects and their duals. Moreover, if \(X,Y\in C\) are dualizable, then \(X\otimes Y\) is dualizable with dual \(X^{\vee }\otimes Y^{\vee }\).

Proof. A symmetric monoidal functor preserves evaluation and coevaluation maps and the triangle identities. For the second claim, apply this observation to the tensor-product functor \(C\times C\to C\), which is symmetric monoidal for the componentwise tensor product on \(C\times C\), and to the dualizable object \((X,Y)\). □

11.1.1 Internal homs

When \(C\) admits internal homs, every object \(X\) has a candidate dual \(D(X)=\iHom (X,\unit )\). Dualizability requires that the natural maps \(Z\otimes D(X)\to \iHom (X,Z)\) be isomorphisms for all \(Z\). We make this precise below and use it to prove that dualizable objects form a thick subcategory.

Definition 11.1.9. Given two objects \(X,Z \in C\), an object \(\iHom (X,Z) \in C\) equipped with a map \(\ev \colon \iHom (X,Z) \otimes X \to Z\) is said to exhibit \(\iHom (X,Z)\) as an internal hom from \(X\) to \(Z\) if for every third object \(W\) the composite \[ \Hom _{C}(W,\iHom (X,Z)) \xrightarrow {- \otimes X} \Hom _{C}(W \otimes X, \iHom (X,Z) \otimes X) \xrightarrow {\ev \circ -} \Hom _{C}(W \otimes X, Z) \] is an equivalence.

Remark 11.1.10. Internal homs are unique if they exist. When they exist for every \(Z\), the assignment \(Z\mapsto \iHom (X,Z)\) defines a right adjoint to \(-\otimes X\colon C\to C\). A symmetric monoidal \(\infty \)-category admitting all internal homs is also called closed.

We now provide a formulation of duality in terms of internal homs.

Lemma 11.1.11. Let \(X\) and \(Y\) be objects and let \(\ev \colon Y \otimes X \to \unit \) be a morphism in \(C\). Then the following conditions are equivalent:

(1)

The morphism \(\ev \) exhibits \(Y\) as a dual to \(X\).

(2)

For every object \(Z\) of \(C\), the morphism \(\id _Z \otimes \ev \colon Z \otimes Y \otimes X \to Z\) exhibits \(Z \otimes Y\) as an internal hom \(\iHom (X,Z)\).

(3)

The functor \(X \otimes -\colon C \to C\) admits a right adjoint \(\iHom (X,-)\colon C \to C\), and for every object \(Z\) of \(C\) the map \[ Z \otimes Y \to \iHom (X,Z) \] induced by the morphism \(\id \otimes \ev \colon Z \otimes Y \otimes X \to Z\) is an isomorphism in \(C\).

Proof. The equivalence between (1) and (2) is a reformulation of Proposition 11.1.4. The equivalence between (2) and (3) is clear. □

There is often an a priori candidate for the dual of \(X\), sometimes called the weak dual:

Definition 11.1.12. Assume that \(C\) admits internal homs. We define the internal duality functor \(D\colon C\catop \to C\) as \begin {align*} D(-) := \iHom (-,\unit )\colon C\catop \to C. \end {align*}

If \(X\) is dualizable, the evaluation map \(\ev \colon X^{\vee } \otimes X \to \unit \) induces an isomorphism \(X^{\vee } \iso \iHom (X,\unit ) = D(X)\).

Corollary 11.1.13. Let \(X\) be an object of \(C\) for which the internal hom functor \(\iHom (X,-)\colon C \to C\) exists. Then \(X\) is dualizable if and only if for every \(Z \in C\) the natural map \(Z \otimes D(X) \to \iHom (X,Z)\) adjoint to the composite \(Z \otimes D(X) \otimes X \xrightarrow {Z \otimes \ev } Z \otimes \unit \simeq Z\) is an isomorphism.

Lemma 11.1.14. Assume that \(C\) admits internal homs. For \(X,Y,Z \in C\), there is a natural currying isomorphism \begin {align*} \iHom (X\otimes Y, Z) \simeq \iHom (X, \iHom (Y,Z)). \end {align*}

Proof. This follows from the Yoneda lemma, and the observation that for every \(W \in C\) there are natural equivalences \begin {align*} \Hom _{C}(W,\iHom (X\otimes Y, Z)) &\simeq \Hom _{C}(W \otimes X \otimes Y, Z) \\ &\simeq \Hom _{C}(W \otimes X, \iHom (Y,Z)) \\ &\simeq \Hom _{C}(W,\iHom (X, \iHom (Y,Z))). \qedhere \end {align*} □

Remark 11.1.15. The objects called dualizable here are also known as strongly dualizable. A useful criterion is given in Lewis et al. (1986), Chapter III, Definition 1.1: if \(C\) is closed, then \(X\) is dualizable precisely when the morphism \(\unit \to \iHom (X,X)\) adjoint to \(\id _X\) factors through the natural map \(X\otimes D(X)\to \iHom (X,X)\).

Definition 11.1.16. A stably symmetric monoidal \(\infty \)-category is a symmetric monoidal \(\infty \)-category \(C\) which is stable and whose tensor product \(- \otimes - \colon C \times C \to C\) is exact in both variables.

Lemma 11.1.17. Let \(C\) be a stably symmetric monoidal \(\infty \)-category which admits internal homs. Then the collection of dualizable objects forms a thick subcategory of \(C\), i.e. it is a stable subcategory closed under retracts.

Proof. By Corollary 11.1.13, an object \(X\) is dualizable if and only if the natural map \[ Z \otimes D(X) \to \iHom (X,Z) \] is an isomorphism for every \(Z \in C\). By the universal property of internal homs and the exactness of tensor products, \(\iHom (-,Z)\) carries finite colimits to finite limits. Thus both sides are exact functors \(C\catop \to C\), and the objects on which this natural map is an isomorphism form a stable subcategory. This subcategory is also closed under retracts, since isomorphisms are closed under retracts. □