Example 15.3.3. Let \(C\) be a small category and \(X\colon C\catop\to\An\) a presheaf. We consider two embeddings \(U\hookrightarrow X\) and \(V\hookrightarrow X\). We denote by \[ U\cap V:=U\times_XV \hookrightarrow X \] the embedding of the fiber product of \(U\) and \(V\) over \(X\). The union of \(U\) and \(V\) in \(X\), denoted by \(U\cup V\), is the colimit of the diagram of embeddings \[ U\hookrightarrow X\, , \ V\hookrightarrow X\ \text{and\ } U\cap V\hookrightarrow X\, . \] The previous lemma tells us that we have a canonical embedding \[ U\cup V\hookrightarrow X\, . \] In fact, one can check that the induced commutative square Commutative diagram; see the PDF for its LaTeX source is both cartesian and cocartesian. Indeed, since \(U\cup V\) embeds in \(X\), to check that it is a pullback square, it suffices to do it after replacing \(U\cup V\) by \(X\), and this holds then by definition. To prove that it is cocartesian, let \(I\) be the full subcategory of all embeddings with codomain \(X\) spanned by the embeddings with domain \(U\), \(V\) or \(U\cap V\). For every presheaf \(Y\), a cocone from the diagram indexed by \(I\) to \(Y\) is precisely a pair of maps \(U\to Y\) and \(V\to Y\) whose restrictions to \(U\cap V\) agree. Indeed, any additional morphism in \(I\) comes from a containment among \(U\) and \(V\), and its compatibility is already forced by this agreement. Thus the presentation \[ U\cup V=\underset{W\hookrightarrow X : I}{\colim}\ W \] has the universal property of the displayed pushout square.

From Section 15.3, Digression: unions of subpresheaves.