In Section 3.4 we saw that the \(\omega _1\)-compact objects of a compactly assembled category are precisely the sequential colimits along compact morphisms (Theorem 3.20 and Remark 3.21). We now show that, for presentable stable categories, the existence of enough such objects characterizes dualizability.

Definition 4.13. Let \(C\) be a category with filtered colimits. A compact exhaustion of an object \(x \in C\) is a sequence \(x_0 \to x_1 \to x_2 \to \cdots \) of weakly compact morphisms together with an equivalence \(\colim _n x_n \simeq x\). An object is called compactly exhaustible if it admits a compact exhaustion.

Proposition 4.14 ([Krause et al. (2024), Theorem 2.2.15]). Let \(C \in \PrL _{\st }\). Then \(C\) is dualizable if and only if \(C\) is generated under colimits by compactly exhaustible objects.

Proof.

If \(C\) is dualizable, then \(C\) is generated under colimits by its \(\omega _1\)-compact objects, which are compactly exhaustible by Theorem 3.20 together with Proposition 4.4.

Conversely, assume that \(C\) is generated under colimits by compactly exhaustible objects. By Theorem 3.16, it suffices to show that \(\colim \colon \Ind (C) \to C\) admits a left adjoint, i.e. that for every \(x \in C\) the functor \(\Map _C(x,\colim (-))\colon \Ind (C) \to \An \) is corepresentable. The category \(\Ind (C)\) is stable and admits small colimits (any small diagram lies in some \(\Ind (C^{\lambda })\), and the inclusions \(\Ind (C^{\lambda }) \subseteq \Ind (C^{\mu })\) are exact and preserve filtered colimits, see Construction A.21). Hence the class of \(x\) for which \(\Map _C(x,\colim (-))\) is corepresentable is closed under small colimits: if \(x \simeq \colim _a x_a\) and \(\Map _C(x_a,\colim (-)) \simeq \Map _{\Ind (C)}(X_a,-)\) naturally, then \begin {align*} \Map _C(x,\colim (-)) &\simeq \lim _a \Map _C(x_a,\colim (-)) \\ &\simeq \lim _a \Map _{\Ind (C)}(X_a,-) \simeq \Map _{\Ind (C)}(\colim _a X_a,-). \end {align*}

So it suffices to treat a compactly exhaustible object \(x \simeq \colim _n x_n\), with \(x_n \to x_{n+1}\) weakly compact. Put \(X := \colim _n y(x_n) \in \Ind (C)\), the colimit being taken in \(\Ind (C)\). Then \(\colim X \simeq \colim _n x_n \simeq x\), as \(\colim \) is a left adjoint and \(\colim \circ y \simeq \id \). We claim that the map \[ \Map _{\Ind (C)}(X,Z) \to \Map _C(\colim X, \colim Z) \simeq \Map _C(x,\colim Z) \] induced by \(\colim \) is an equivalence, naturally in \(Z \in \Ind (C)\). Under the adjunction \(\colim \dashv y\), it is identified with the map \((\eta _Z)_*\colon \Map (X,Z) \to \Map (X,y(\colim Z))\). Since everything is stable, it suffices to show that \(\hom _{\Ind (C)}(X,W) \simeq 0\) for \(W = \fib (\eta _Z)\), which satisfies \(\colim W \simeq 0\) (as \(\colim (\eta _Z)\) is an equivalence).

Put \(h_n := \hom _{\Ind (C)}(y(x_n),W)\). We have \(\hom (X,W) \simeq \lim _n h_n\), so there are Milnor short exact sequences \[ 0 \to {\lim _n}^1 \pi _{m+1}h_n \to \pi _m\hom (X,W) \to \lim _n \pi _m h_n \to 0. \] Now \(\pi _m h_n \simeq \pi _0\Map (y(x_n),\Omega ^mW)\), and \(\colim \Omega ^mW \simeq 0\). Since \(x_n \to x_{n+1}\) is weakly compact, Proposition 4.9 shows that all transition maps in the tower \((\pi _m h_n)_n\) are zero, for every \(m \in \Z \). A tower of abelian groups with zero transition maps has vanishing \(\lim \) and \(\lim ^1\), so \(\pi _m\hom (X,W) = 0\) for all \(m\), as desired. □

Remark 4.15. In Proposition 4.14, stability is used to reduce to ind-objects with vanishing colimit. For general presentable categories, the analogous statement holds provided filtered colimits in \(C\) are exact, with “dualizable” replaced by “compactly assembled” [Krause et al. (2024), Theorem 2.2.15].