Theorem 3.16 (Lurie, [Lurie (2018), Proposition D.7.3.1]). Let \(C \in \PrL _{\st }\). Then the following are equivalent:
- (1)
-
\(C\) is dualizable;
- (2)
-
\(C\) is compactly assembled;
- (3)
-
There exists a localization \(D \to E\) in \(\PrL _{\omega ,\st }\) such that \(C = \ker (D \to E)\);
- (4)
-
\(C\) has the left lifting property with respect to any localization in \(\PrL _{\st }\): for every localization \(D \to E\) and any \(C \to E\), there exists a lift \(C \to D\):
Proof.
(2) \(\implies \) (1): Every compactly assembled category is a retract of a compactly generated one. Remark 3.14 shows that compactly generated categories are dualizable, and since dualizable objects are closed under retracts we are done.
(1) \(\implies \) (4): Assume \(C\) is dualizable, and consider a cofiber sequence \[ K \hookrightarrow D \to E \] in \(\PrL _{\st }\). Tensoring with \(C^{\vee }\) preserves colimits, so gives another cofiber sequence \[ C^{\vee } \otimes K \to C^{\vee } \otimes D \to C^{\vee } \otimes E. \] In particular, the functor \(C^{\vee } \otimes D \to C^{\vee } \otimes E\) is again a localization, hence essentially surjective. It follows that the functor \(\Sp \to C^{\vee } \otimes E\) classifying \(C \to E\) lifts to a functor \(\Sp \to C^{\vee } \otimes D\) classifying a lift \(C \to D\).
(4) \(\implies \) (2): Let \(C_0 \subseteq C\) be a small dense stable subcategory, and consider the adjunction \[ y\colon C \rightleftarrows \Ind (C_0) \noloc \colim . \] Since \(y\) is fully faithful, it follows that \(\colim \colon \Ind (C_0) \to C\) is a localization, and so by (4) this admits a section in \(\PrL _{\st }\). It follows that \(C\) is a retract of \(\Ind (C_0)\), hence it is compactly assembled.
(3) \(\implies \) (2): Consider a fiber sequence \[ C \hookrightarrow D \to E. \] Since \(D \to E\) preserves compact objects, its right adjoint preserves colimits. By Lemma 2.13, the kernel inclusion \(C \to D\) also has a colimit-preserving right adjoint \(D \to C\). This right adjoint is a retraction, so \(C\) is a retract of \(D\) in \(\PrL _{\st }\).
(2) \(\implies \) (3): As before, we may find some \(\lambda \) such that \(\colim \colon \Ind (C^{\lambda }) \to C\) admits a left adjoint \(C \to \Ind (C^{\lambda })\). Then consider the cofiber sequence \[ C \xhookrightarrow {\hat {y}} \Ind (C^{\lambda }) \to \cofib (\hat {y}) \simeq \fib (\colim ). \] The localization functor \(\Ind (C^{\lambda }) \to \cofib (\hat {y})\) admits a right adjoint in \(\PrL _{\st }\), and so it preserves compact objects. It in particular follows that \(\cofib (\hat {y})\) is compactly generated, i.e. \(\cofib (\hat {y}) \simeq \Ind (\Calk ^{\cont }_{\lambda }(C))\) where \(\Calk ^{\cont }_{\lambda }(C) := \cofib (\hat {y})^{\omega }\) is the continuous \(\lambda \)-Calkin category. So this gives (3). □