Compact graphings (Q2003771): Difference between revisions

A graphing \(G=(I,\mathcal{A},E,\lambda)\) is a graph on vertex set a standard Borel space \((I, \mathcal{A})\) with \(\lambda\) a probability measure on \((I,\mathcal{A})\). All degrees are assumed bounded by some fixed (and suppressed) number \(D\). The edge set \(E\subseteq I\times I\) is a Borel set avoiding the diagonal and invariant under interchanging the coordinates which has the following additional symmetry property. Defining the edge measure of the graphing for \(A,B\in \mathcal{A}\) by \(\eta(A\times B)=\int_{A}\text{deg}_{B}(x) d\lambda(x)\) and extending to Borel sets in the usual way, we have that \(\eta(A\times B)=\eta(B\times A)\). Graphings represent limits of certain graph sequences in a local or local-global sense, and are in a loose sense analogous to graphons, corresponding limit objects for dense graphs. \par In the theory of graphons, an important process is \lq purification\rq\, of a graphon, i.e. converting it into an essentially equivalent graphon which additionally is defined on a compact metric space. The main result of the article under review is to define a compact graphing and to show that every graphing embeds in a well-behaved way in a compact graphing. In detail, if \(d\) is a metric such that \(\mathcal{A}\) is the Borel subsets of \((I,d)\) we write \((I,d,E,\lambda)\) in lieu of \((I,\mathcal{A},E,\lambda)\) and say that this graphing is compact if \((I,d)\) is a compact metric space and additionally \(E\) is a closed subset of \(I\times I\) and for all \(\epsilon>0\) and non-negative integer \(r\) there is \(\delta>0\) such that whenever \(x,y\in I\) with \(d(x,y)<\delta\) there is an isomorphism \(\phi\) from \(B(x,r)\) to \(B(y,r)\) with \(\phi(x)=y\) and such that \(\max_{z\in B(x,r)}d(z,\phi(z))<\epsilon\). \par The appropriate notion of embedding in a nice way is that of being a full subgraphing: graphing \((I,\mathcal{A},E,\lambda)\) is a full subgraphing of \((I',\mathcal{A}', E',\lambda')\) if \(G\) is a union of connected components of \(G'\), \(I\) is a Borel subset of \(I'\), \(\mathcal{A}=\mathcal{A}' \cap I\) and \(\lambda '(X)=\lambda(X\cap I)\) for all Borel sets \(X\). \par The result is now that every graphing is a full subgraphing of a compact graphing; the proof is quite technical, but the dependence of \(\delta\) on \(\epsilon\) and \(r\) is fairly mild. The author notes that in the loosely analogous situation of graphons, there is a ``canonical'' compact graphon locally equivalent to the graphon we started from and asks is an analogous result can be proved here (the argument at present depends on the metric on the underlying set we started with).