In Chapter 1 we saw that compact generation is too restrictive for many purposes: for instance, the category \(\Shv (\R ^n;\Sp )\), for \(n \geq 1\), has no nonzero compact objects at all. The goal of this chapter is to introduce the class of categories that replaces the compactly generated ones: the dualizable categories, or, in the unstable setting, the compactly assembled categories.

To motivate the definition, let us first ask what makes compact generation so useful. Let \(C\) be a category with filtered colimits, and let us call an ind-object \(X \in \Ind (C)\) together with an equivalence \(\colim X \simeq x\) a presentation of \(x \in C\). Every object \(x\) has a trivial presentation, namely \(y(x)\); by the adjunction \(\colim \dashv y\), every presentation of \(x\) maps uniquely to \(y(x)\), compatibly with the identifications of the colimits with \(x\). So \(y(x)\) is the terminal, or coarsest, presentation of \(x\). If \(C = \Ind (C_0)\) is compactly generated, then \(x\) has a second canonical presentation \[ \hat {y}(x) = \quotcolim _{(c \to x) \in (C_0)_{/x}} c, \] and its usefulness comes from the fact that it maps into every other presentation: since every map \(c \to \colim _j z_j\) from a compact object factors through some stage \(z_j\), we have \[ \Map _C\big (x,\colim _j z_j\big ) \simeq \Map _{\Ind (C)}\big (\hat {y}(x), \quotcolim _j z_j\big ) \] for every filtered diagram \((z_j)_j\) in \(C\). In other words, \(\hat {y}(x)\) is the initial, or finest, presentation of \(x\). This formula says that maps out of \(x\) into filtered colimits are computed by the formal object \(\hat {y}(x)\), even if \(x\) itself is not compact.

A category is compactly assembled if every object admits such a finest presentation, i.e. if \(\colim \colon \Ind (C) \to C\) admits a left adjoint \(\hat {y}\). Unlike in the compactly generated case, the terms of the ind-object \(\hat {y}(x)\) are no longer required to be compact. A presentable stable category is compactly assembled if and only if it is dualizable (Theorem 3.16), and this is the class of categories to which K-theory will be extended.

Sections

Section 3.2

Dualizable categories

Duality for presentable stable categories and Lurie’s characterization.

Section 3.4

ω₁-presentability

ω₁-compact objects as sequential colimits along compact morphisms.