Flat vs. filtered colimits in the enriched context (Q2671886)

A functor \(M:\mathcal{C}^{\mathrm{op}}\rightarrow\boldsymbol{Set}\) is called \textit{flat} if its left Kan extension \(\mathrm{Lan}_{Y}M:[\mathcal{C},\boldsymbol{Set}]\rightarrow\boldsymbol{Set}\) along the Yoneda embedding preserves finite limits, Equivalently, \(M\) is flat iff it is a filtered colimit of representable functors, which suggests a deep connection between flatness and filtered colimits.The connection plays a key role in the theory of accessible categories. A category \(\mathcal{A}\) is \textit{finitely accessible} by definition if it is the free cocompletion of a small category \(\mathcal{C}\) under filtered colimits, which is tantamount to saying that \(\mathcal{A}\) is equivalent to the category \(\mathrm{Flat}(\mathcal{C}^{\mathrm{op}},\boldsymbol{Set})\) of flat presheaves on a small category \(\mathcal{C}\). The situation becomes rather more complicated in enriched category theory. Let us fix a base of enrichment \(\mathcal{V}=(\mathcal{V}_{0} ,\otimes,I)\), which is symmetric monoidal closed and locally finitely presentable as a closed category. A \(\mathcal{V}\)-functor is called \textit{flat} if \(\mathrm{Lan}_{Y}M:[\mathcal{C},\mathcal{V}]\rightarrow\mathcal{V}\) preserves all finite weighted limits. A \(\mathcal{V}\)-category \(\mathcal{A}\) is called \textit{finitely accessible} if it is the free cocompletion of a small \(\mathcal{V}\)-category under flat-weighted colimits. A \(\mathcal{V}\)-category \(\mathcal{A}\) is called \textit{conically finitely accessible} if it is the free cocompletion of a small \(\mathcal{V}\)-category under conical filteredd colimits. \textit{G. M. Kelly} [The basic concepts of enriched category theory. Seminarber. Fachber. Math., Fernuniv. Hagen 9 (1981; Zbl 0709.18501); Basic concepts of enriched category theory. Cambridge: Cambridge University Press. London: London Mathematical Society (1982; Zbl 0478.18005); Repr. Theory Appl. Categ. 2005, No. 10, 1--136 (2005; Zbl 1086.18001)], \S 6.4] posed the question whether for any locally finitely presentable base, every flat presheaf is a filtered colimit of representables. This is actually false for the case \(\mathcal{V} =\boldsymbol{Ab}\) [\textit{F. Borceux} et al., Theory Appl. Categ. 4, 47--72 (1998; Zbl 0981.18006)]. This paper gives a positive answer to the question in a large number of examples, including \(\mathcal{V} =\boldsymbol{Cat}\) and \(\mathcal{V}=\boldsymbol{SSet}\), in which the two notions of accessibility agree. The authors can give a full characterization of flat \(\mathcal{V}\)-functors. \textit{J. Bourke} [J. Pure Appl. Algebra 225, No. 3, Article ID 106519, 43 p. (2021; Zbl 1451.18015)] gave some powerful techniques for proving that a wide range of \(2\)-categories of categories with structure are conically accessible as \(\boldsymbol{Cat}\)-enriched categories. It was shown in [\textit{J. Bourke} and \textit{S. Lack}, ``Accessible \(\infty\)-cosmoi'', Preprint, \url{arXiv:2111.00147}] how to adapt these techniques to the simplicially enriched case. It is shown in this paper that these conically accessible \(\boldsymbol{Cat}\)-enriched or \(\boldsymbol{SSet}\)-enriched categories are in fact accessible, so that they are also sketchable, and the whole theory of enriched accessible categories applies. There is an adjunction \[ [\mathcal{C}^{\mathrm{op}},\mathcal{V}]_{0} \begin{array} [c]{c} \underleftarrow{\quad\mathcal{F\quad}}\\ \bot\\ \overrightarrow{\quad\mathcal{U\quad}} \end{array} [\mathcal{C}_{0}^{\mathrm{op}},\boldsymbol{Set}] \] induced by the underlying functor \[ Y_{0}:\mathcal{C}_{0}\rightarrow[\mathcal{C}^{\mathrm{op}} ,\mathcal{V}]_{0} \] of the enriched Yoneda embedding. It is true in general that \(\mathcal{F}\) sends a flat (in the ordinary sense) presheaf on \(\mathcal{C}_{0}\) to a flat (in the extended sense) presheaf on \(\mathcal{C}\), while it is not necessarily true that \(\mathcal{U}\) preserves flatness. The authors' basic strategy is to show in particular cases that \begin{itemize} \item[(I)] \(\mathcal{U}\) preserves flatness; \item[(II)] if \(M\) is a flat presheaf on , then the component \(\varepsilon _{M}:\mathcal{FU}M\rightarrow M\) of the counit is invertible. \end{itemize} A synopsis of the paper goes as follows. \begin{itemize} \item[\S 2] is concerned with basic notions. \item[\S 3] gives conditions on \(\mathcal{V}\) under which this is the case for all small \(\mathcal{C}\), deducing that for such , the existence and preservation of flat weighted colimits is equivalent to that of filtered colimits (Theorem 3.13). Consequently, the notions of \(\alpha\)-accessibility and conical \(\alpha\)-accessibility\ agree (Theorem 3.14). Examples of \(\mathcal{V}\) abiding by these conditions include the cartesian closed categories \(\boldsymbol{Set}\) of sets, \(\boldsymbol{Cat}\) of small categories, \(\boldsymbol{SSet}\) of simplicial sets, \(\boldsymbol{2}\) of the free-living arrow, \(Pos\) of partially ordered sets, and so on. \item[\S 4] gives conditions on \(\mathcal{V}\) under which (I) and (II) hold provided that \(\mathcal{C}\) has certain \(\mathcal{V}\)-enriched absolute colimits (Proposition 4.14), from which it follows easily that a \(\mathcal{V}\)-category is \(\mathcal{V}\)-accessible iff it is conically \(\alpha\)-accessible\ and has these absolute colimits (Theorem 4.19). Examples of \(\mathcal{V}\) abiding by these conditions include the monoidal categories \(\boldsymbol{CMon}\) of commutative monoids, \(\boldsymbol{Ab}\) of abelian groups, \(R\)-\(\boldsymbol{Mod}\) of \(R\)-modules for a commmutative ring \(R\), \(\boldsymbol{GAb}\) of graded abelian groups and so on. \item[\S 5] investigates the case where \(\mathcal{V}\) is the cartesian closed category \(\boldsymbol{Set}^{G}\) of -sets for a non-trivial finite group \(G\), showing that in this case \(\alpha\)-accessibility is strictly stronger than conical \(\alpha\)-accessibility and the existence of absolute colimits (Corollary 5.4). \item[\S 6] investigates one further class of examples related to those in \S 4, where (II) still holds when \(\mathcal{C}\) has some finite direct sums and copowers by dualizable objects, while (I) doesn't seem to be true even with this further condition. What is true is that, when \(\mathcal{C}\) has those absolute colimits and \(M\) is flat, ordinary functor \(\mathcal{U}M\) is part of what is called a protofiltered diagram. It is shown that flat colimits are generated by the absolute ones and plus these protofiltered colimits (Theorem 6.16). \end{itemize}