For vector spaces, compactness and dualizability both mean finite-dimensionality. In general, we show that dualizable objects are compact whenever the unit is compact, and give a criterion for the converse. For modules over a commutative ring spectrum, both conditions characterize the perfect modules.

Lemma 11.2.1. Let \(C\) be a symmetric monoidal \(\infty \)-category admitting filtered colimits. If its unit \(\unit \) is compact, then every dualizable object is compact.

Proof. Let \(X\) be a dualizable object in \(C\), and let \(X^{\vee }\) denote its dual. Then the functor \(X \otimes -\colon C \to C\) admits a right adjoint of the form \(X^{\vee } \otimes -\colon C \to C\). In particular, we have \[ \Hom _C(X,-) \simeq \Hom _C(X \otimes \unit ,-) \simeq \Hom _C(\unit ,X^{\vee } \otimes -). \] Since \(X\) is also a dual of \(X^{\vee }\), the functor \(X^{\vee }\otimes -\) is itself a left adjoint and preserves filtered colimits. Compactness of \(\unit \) therefore implies that the displayed functor preserves filtered colimits. □

For the converse, we use that dualizable objects form a thick subcategory. It therefore suffices to generate the compact objects from dualizable ones.

Lemma 11.2.2 (cf. Hovey et al. (1997)). Let \(C\) be a presentably symmetric monoidal stable \(\infty \)-category in the sense of Definition 16.2.8, and let \(\Gg \) be a small collection of compact generators of \(C\). If every object of \(\Gg \) is dualizable, then every compact object of \(C\) is dualizable.

Proof. The tensor product admits internal homs by Theorem 23.2.5. By the compact-generation theorem Proposition 23.3.4, the compact objects form the thick subcategory generated by \(\Gg \). The claim now follows from Lemma 11.1.17. □

Corollary 11.2.3. Let \(C\) be a presentably symmetric monoidal stable \(\infty \)-category, and let \(\Gg \) be a small collection of compact generators. Then the compact objects of \(C\) are precisely its dualizable objects if and only if the unit \(\unit \) is compact and every object of \(\Gg \) is dualizable.

Proof. If the unit is compact and every object of \(\Gg \) is dualizable, the claim follows from Lemma 11.2.1, Lemma 11.2.2. Conversely, if the compact and dualizable objects coincide, then the unit is compact because it is always dualizable, while every object of \(\Gg \) is dualizable because it is compact. □

We apply this criterion to modules over a commutative ring spectrum \(R\). Recall from Definition 8.1.10 that the perfect modules form the thick subcategory generated by \(R\).

Proposition 11.2.4. Let \(R\) be a commutative ring spectrum. An \(R\)-module is perfect if and only if it is dualizable in the symmetric monoidal \(\infty \)-category \((\Mod _R,\otimes _R)\).

Proof. By Proposition 8.1.5, the \(\infty \)-category \(\Mod _R\) is presentably symmetric monoidal and stable. Its unit \(R\) is a compact generator by Lemma 8.1.11, and is dualizable. Thus Corollary 11.2.3 identifies the dualizable modules with the compact modules, which are precisely the perfect modules. □

11.2.1 Spanier–Whitehead duality

Taking \(R=\S \) in Proposition 11.2.4, we see that a spectrum is dualizable if and only if it is compact. We relate this to the following notion of finiteness.

Definition 11.2.5 (Finite spectrum). A spectrum is finite if it lies in the smallest full stable subcategory of \(\Sp \) containing the sphere spectrum \(\S \).

One can show that the finite spectra are precisely the spectra of the form \(\Sigma ^{\infty }Y[-n]\) for a finite pointed cell complex \(Y\) and some \(n\geq 0\). By Proposition 23.3.4, the compact spectra are precisely the retracts of finite spectra. For spectra, taking retracts does not enlarge this class:

Lemma 11.2.6. A retract of a finite spectrum is finite.

Proof. A finite spectrum is bounded below and has finitely generated integral homology concentrated in finitely many degrees. These properties pass to retracts. We show that any spectrum \(X\) with these properties is finite.

Choose \(a\leq b\) such that \(X\) is \(a\)-connective and \(H_i(X)=0\) for \(i>b\). Set \(X_a=X\). For \(i=a,\ldots ,b\), choose generators of \(H_i(X_i)\) and lift them, by Theorem 6.2.17, to a map \(\S [i]^{\oplus r_i}\to X_i\). Its cofiber \(X_{i+1}\) is \((i+1)\)-connective, still has finitely generated homology, and satisfies \(H_j(X_{i+1})\cong H_j(X_i)\) for \(j>i+1\).

Thus \(X_{b+1}\) has homology only in degree \(b+1\), where it is the kernel of \(\Z ^{r_b}\to H_b(X_b)\) and hence free of finite rank. A basis lifts to a homology isomorphism \(\S [b+1]^{\oplus r}\to X_{b+1}\), which is an isomorphism by Corollary 6.2.18. Reversing the finitely many cofiber sequences shows that \(X\) is finite. □

Corollary 11.2.7. A spectrum is finite if and only if it is compact, and if and only if it is dualizable.

Proof. The compact spectra are the retracts of finite spectra, so the first equivalence follows from Lemma 11.2.6. The second is the case \(R=\S \) of Proposition 11.2.4, Lemma 8.1.11. □

Corollary 11.2.8. Let \(X\) be a finite anima. Then the spectrum \(\S [X]\) is dualizable.

Proof. The collection of animae \(X\) for which \(\S [X]\) is dualizable is closed under finite colimits and contains \(* \in \An \); consequently it contains all finite animae. □

The dual of a finite spectrum is called its Spanier–Whitehead dual. Tensoring with this dual computes the internal hom out of the original spectrum, so duality interchanges homology and cohomology. For spectra \(X\) and \(E\), we use the notation \[ E_k(X):=\pi _k(E\otimes X) \qquadtext {and} E^k(X):=\pi _{-k}\hom (X,E). \] For \(X=\S [A]\), this recovers the conventions for the \(E\)-(co)homology of an anima \(A\).

Lemma 11.2.9. Let \(X\) and \(E\) be spectra and assume that \(X\) is dualizable. Then there are natural isomorphisms \[ E_*(X^{\vee }) \cong E^{-*}(X) \qquadtext {and} E^*(X^{\vee }) \cong E_{-*}(X). \]

Proof. The first isomorphism follows from \[ E_*(X^{\vee }) \cong \pi _*(E \otimes X^{\vee }) \cong \pi _*(\hom (X,E)) \cong E^{-*}(X). \] The second follows by applying the first to \(X^{\vee }\), whose dual is \(X\). □