For an object \(x\) of a compactly assembled category \(C\), the functor \(\Map _{\Ind (C)}(y(x),\hat {y}(-))\) is a “filtered-colimit-preserving approximation” of \(\Map _C(x,-)\): if \(x\) is compact, the two agree, and in general the former is the best approximation of the latter by a functor that preserves filtered colimits. To make this precise, for categories \(C\) and \(D\) with filtered colimits we write \(\Fun ^{\mathrm {filt}}(C,D) \subseteq \Fun (C,D)\) for the full subcategory of functors preserving filtered colimits.
Proposition 4.16 ([Krause et al. (2024), Proposition 2.4.8]). Let \(C\) be compactly assembled, and let \(D\) be a category with filtered colimits. Then the inclusion \(\Fun ^{\mathrm {filt}}(C,D) \hookrightarrow \Fun (C,D)\) admits a right adjoint, given by \[ F \mapsto \colim _D \circ \Ind (F) \circ \hat {y}, \] where \(\Ind (F)\colon \Ind (C) \to \Ind (D)\) is the extension of \(F\) preserving filtered colimits.
Proof.
Write \(\colim = \colim _C\). The adjunction \(\hat {y} \dashv \colim \) induces an adjunction \[ \colim ^*\colon \Fun (C,D) \rightleftarrows \Fun (\Ind (C),D)\noloc \hat {y}^* \] given by precomposition: a transformation \(G\colim \to H\) corresponds to the composite \(G \to G\colim \hat {y} \to H\hat {y}\), using the unit \(\id \to \colim \hat {y}\). Since \(\colim \) and \(\hat {y}\) preserve filtered colimits, this restricts to an adjunction between \(\Fun ^{\mathrm {filt}}(C,D)\) and \(\Fun ^{\mathrm {filt}}(\Ind (C),D)\). By the universal property of \(\Ind (C)\) (the analogue of Theorem A.15(2) for large \(C\), see Construction A.21), restriction along \(y\) is an equivalence \(\Fun ^{\mathrm {filt}}(\Ind (C),D) \simeq \Fun (C,D)\), with inverse \(F \mapsto \colim _D \circ \Ind (F)\). Under this equivalence, the left adjoint \(\colim ^*\) becomes \(G \mapsto G\colim y \simeq G\), i.e. the inclusion, and the right adjoint \(\hat {y}^*\) becomes \(F \mapsto \colim _D \circ \Ind (F) \circ \hat {y}\). □
Corollary 4.17. Let \(C\) be compactly assembled and \(x \in C\). Then the natural transformation \[ \Map _{\Ind (C)}\big (y(x),\hat {y}(-)\big ) \xrightarrow {(\varepsilon )_*} \Map _{\Ind (C)}\big (y(x),y(-)\big ) \simeq \Map _C(x,-) \] exhibits \(\Map _{\Ind (C)}(y(x),\hat {y}(-))\) as the terminal object of \(\Fun ^{\mathrm {filt}}(C,\An )_{/\Map _C(x,-)}\). If \(C\) is moreover stable, the same holds for the mapping spectra \(\hom _{\Ind (C)}(y(x),\hat {y}(-)) \to \hom _C(x,-)\) in \(\Fun ^{\mathrm {filt}}(C,\Sp )_{/\hom _C(x,-)}\).
Proof.
For a right adjoint \(R\) of a fully faithful inclusion \(i\), the counit \(iR(F) \to F\) is a terminal object of the slice over \(F\), since \(\Map (G,R(F)) \simeq \Map (iG,F)\). Apply this to Proposition 4.16 with \(D = \An \) and \(F = \Map _C(x,-)\). For \(Z = \quotcolim _j z_j \in \Ind (C)\) we have \[ \colim _{\An }\Ind (F)(Z) \simeq \colim _j \Map _C(x,z_j) \simeq \Map _{\Ind (C)}(y(x),Z), \] as \(y(x)\) is compact in \(\Ind (C)\). Hence the right adjoint sends \(F\) to \(\Map _{\Ind (C)}(y(x),\hat {y}(-))\). To identify the counit, write \(\tilde {F} = \colim _{\An } \circ \Ind (F)\). By construction, the counit at \(F\) is obtained by restricting the counit \(\tilde {F}\hat {y}\colim \to \tilde {F}\) of the precomposition adjunction along \(y\), and the latter is \(\tilde {F}\) applied to the counit \(\hat {y}\colim \to \id \) of \(\hat {y} \dashv \colim \). At \(y(c)\), this counit is the map \(\hat {y}(c) \simeq \hat {y}(\colim y(c)) \to y(c)\), which is \(\varepsilon _c\) by definition of \(\varepsilon \). So the counit is \(\tilde {F}(\varepsilon )\), i.e. the map induced by \(\varepsilon \). The stable case is identical, with \(D = \Sp \). □