We recall the basic notions; see [Lurie (2009), Chapter 5] for proofs of the results quoted in this section.

Definition A.14. For a small category \(C_0\), we let \(\Ind _{\kappa }(C_0) \subseteq \PSh (C_0)\) denote the full subcategory spanned by the \(\kappa \)-filtered colimits of representable presheaves. We write \(\Ind = \Ind _{\omega }\).

Theorem A.15 ([Lurie (2009), Section 5.3.5]). Let \(C_0\) be a small category.

(1)

The category \(\Ind _{\kappa }(C_0)\) admits \(\kappa \)-filtered colimits, computed in \(\PSh (C_0)\), and the Yoneda embedding \(y\colon C_0 \hookrightarrow \Ind _{\kappa }(C_0)\) lands in \(\kappa \)-compact objects.

(2)

For every category \(D\) admitting \(\kappa \)-filtered colimits, restriction along \(y\) induces an equivalence between the category of functors \(\Ind _{\kappa }(C_0) \to D\) preserving \(\kappa \)-filtered colimits and \(\Fun (C_0,D)\). The inverse is given by left Kan extension along \(y\).

(3)

If \(C_0\) admits \(\kappa \)-small colimits, then \(\Ind _{\kappa }(C_0)\) admits all small colimits, \(y\) preserves \(\kappa \)-small colimits, and for every category \(D\) admitting small colimits, restriction along \(y\) induces an equivalence between colimit-preserving functors \(\Ind _{\kappa }(C_0) \to D\) and functors \(C_0 \to D\) preserving \(\kappa \)-small colimits.

For the equivalence in (3) when \(D\) is \(\kappa \)-presentable, see [SCT, Proposition 17.3.5].

Definition A.16. A category \(C\) is \(\kappa \)-accessible if it is locally small, admits \(\kappa \)-filtered colimits, the full subcategory \(C^{\kappa }\) is essentially small, and every object of \(C\) is a \(\kappa \)-filtered colimit of \(\kappa \)-compact objects. It is accessible if it is \(\kappa \)-accessible for some \(\kappa \). A functor between accessible categories is accessible if it preserves \(\kappa \)-filtered colimits for some \(\kappa \). A category is presentable if it is accessible and admits all small colimits.

Theorem A.17 ([Lurie (2009), Section 5.4.2]).

(1)

A category \(C\) is \(\kappa \)-accessible if and only if \(C \simeq \Ind _{\kappa }(C_0)\) for some small category \(C_0\). In this case, the inclusion \(C^{\kappa } \subseteq C\) induces an equivalence \(\Ind _{\kappa }(C^{\kappa }) \simeq C\), and \(C^{\kappa }\) is the idempotent completion of \(C_0\).

(2)

If \(C\) is \(\kappa \)-accessible, then \(C^{\lambda }\) is essentially small for every regular cardinal \(\lambda \), and \(C\) is \(\lambda \)-accessible for arbitrarily large \(\lambda \).

Theorem A.18 ([Lurie (2009), Proposition 5.4.7.3]). The subcategory \(\mathrm {Acc} \subseteq \widehat {\Cat }\) of accessible categories and accessible functors admits small limits, and the inclusion \(\mathrm {Acc} \subseteq \widehat {\Cat }\) preserves them.

Theorem A.19 (Adjoint functor theorem, [Lurie (2009), Corollary 5.5.2.9]). Let \(F\colon C \to D\) be a functor between presentable categories. Then \(F\) admits a right adjoint if and only if it preserves small colimits, and \(F\) admits a left adjoint if and only if it is accessible and preserves small limits. For the right adjoint criterion, the target need only be locally small and admit small colimits, see [SCT, Proposition 21.2.16].

Lemma A.20. Let \(C\) be an accessible category. Then every object of \(C\) is \(\lambda \)-compact for some regular cardinal \(\lambda \). Consequently, \(C = \bigcup _{\lambda } C^{\lambda }\), where each \(C^{\lambda }\) is essentially small.

Proof.

Let \(C\) be \(\kappa \)-accessible and \(x \in C\). Write \(x = \colim _{i \in I} x_i\) with \(I\) small and \(\kappa \)-filtered and each \(x_i\) \(\kappa \)-compact. Choose a regular cardinal \(\lambda \geq \kappa \) such that \(I\) is \(\lambda \)-small. Each \(x_i\) is \(\lambda \)-compact by Corollary A.4(1). Moreover \(\Map (x,-) \simeq \lim _{i \in I\catop }\Map (x_i,-)\) is a \(\lambda \)-small limit of functors preserving \(\lambda \)-filtered colimits, hence preserves them by Theorem A.3. The last statement follows from Theorem A.17(2). □

Construction A.21. Let \(C\) be a locally small, possibly large, category. We let \(\Ind (C) \subseteq \Fun (C\catop ,\widehat {\An })\) denote the full subcategory spanned by the small filtered colimits of representable functors, where \(\widehat {\An }\) denotes the category of large animae. If \(C' \subseteq C\) is a small full subcategory, then left Kan extension along \(C' \subseteq C\) restricts to a fully faithful functor \(\Ind (C') \hookrightarrow \Ind (C)\) which preserves filtered colimits: indeed, for filtered diagrams \((c_i)\) and \((d_j)\) in \(C'\) both categories compute \[ \Map (\quotcolim _i c_i, \quotcolim _j d_j) \simeq \lim _i \colim _j \Map _C(c_i,d_j). \] If \(C\) is accessible, then Lemma A.20 shows that \[ \Ind (C) = \bigcup _{\lambda } \Ind (C^{\lambda }), \] since a small filtered diagram in \(C\) takes values in \(C^{\lambda }\) for \(\lambda \) large enough.

If \(C\) admits filtered colimits, then the functor \(\colim \colon \Ind (C) \to C\) sending \(\quotcolim _i c_i\) to \(\colim _i c_i\) (the unique filtered-colimit-preserving extension of \(\id _C\), by the analogue of Theorem A.15(2)) is left adjoint to the Yoneda embedding \(y\colon C \hookrightarrow \Ind (C)\), since \[ \Map _C(\colim _i c_i, c) \simeq \lim _i \Map _C(c_i,c) \simeq \Map _{\Ind (C)}(\quotcolim _i c_i, y(c)). \]