Definable K\H{o}nig theorems
Extracted source text ยท arXiv 2112.10222
\documentclass[11pt]{amsart}
\usepackage{amssymb}
%\usepackage[magyar,angol]{babel}
%\usepackage{graphicx}
\usepackage{amsmath}
\usepackage{tikz}
%\theoremstyle{plain}
\newtheorem{theorem}{Theorem}[section]
\newtheorem{question}{Question}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{prop}[theorem]{Proposition}
\newtheorem{cor}[theorem]{Corollary}
\newtheorem{conj}[theorem]{Conjecture}
\newtheorem{claim}[theorem]{Claim}
\newtheorem{example}[theorem]{Example}
\theoremstyle{definition}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{remark}[theorem]{Remark}
%\newcommand{\zz}{\mathbb{Z}\slash2\Z *\mathbb{Z}\slash2\Z}
\newcommand{\und}{\underline}
\newcommand{\ovl}{\overline}
\newcommand{\oi}{\overset{i}}
\newcommand{\zz}{(\mathbb{Z}\slash2\mathbb{Z}) *(\mathbb{Z}\slash2\mathbb{Z})}
\newcommand{\e}{\mathrm{Ends}}
\newcommand{\Z}{\mathbb{Z}}
\newcommand{\F}{\mathbb{F}}
\DeclareMathOperator{\supp}{supp}
\def\acts{\curvearrowright}
\newcommand{\asi}{\textnormal{asi}}
\newcommand{\res}{\upharpoonright}
\title{Definable K\H{o}nig theorems}
\author{Matt Bowen, Felix Weilacher}
\email{matthewbowen2015@gmail.com, fweilach@andrew.cmu.edu}
\begin{document}
\maketitle
\begin{abstract}
Let $X$ be a Polish space with Borel probability measure $\mu,$ and let $G$ be a Borel graph on $X$ with no odd cycles and maximum degree $\Delta(G).$ We show that the Baire measurable edge chromatic number of $G$ is at most $\Delta(G)+1$, and if $G$ is $\mu$-hyperfinite then the $\mu$-measurable edge chromatic number obeys the same bound. More generally, we show that $G$ has Borel edge chromatic number at most $\Delta(G)$ plus its asymptotic separation index.
\end{abstract}
\section{Introduction}
A classic theorem of K\H{o}nig states that every bipartite graph $G$ admits a (proper) edge coloring using $\Delta(G)$ many colors. While this result is generally stated for finite graphs only, the compactness principle ensures that the same bound holds even when $G$ is infinite. However, as the compactness principle in this context is equivalent to the existence of non-pinciple ultrafilters, obtaining bounds on chromatic numbers this way produces colorings that are poorly behaved with respect to any topological or measure structure on the graph.
The field of descriptive combinatorics is roughly dedicated to the problem of obtaining graph theoretic results that do respect underlying topological and measure theoretic structure. The typical objects of study are Borel graphs, i.e. graphs whose vertex sets are a Polish space $(X,\tau)$ and whose edge sets are Borel subsets of $X^2.$ This field essentially originated with the work of Kechris, Solecki, and Todorcevic in \cite{kst}, and many applications have been found in areas such as descriptive set theory, group dynamics, and finite combinatorics.
Given a Borel graph $G$ with compatible Borel probability measure $\mu,$ the Borel edge chromatic number, $\chi'_{B}(G),$ is the smallest cardinal $\kappa$ so that $G$ admits Borel proper edge coloring using $\kappa$ colors. The Baire measurable edge chromatic number, $\chi'_{BM}(G),$ is the minimum of $\chi'_B(G \res C),$ where $C$ varies over $\tau$-comeagre $G$-invariant Borel sets. The $\mu$-measurable edge chromatic number, $\chi'_\mu(G),$ is defined analogously.
The study of these parameters was begun in \cite{kst}, where it was shown that the greedy upper bound $\chi'_B(G)\leq 2\Delta(G)-1$ holds, while the irrational rotation graph, $G_\alpha,$ gives a simple example of a $2$-regular acylic graph where $2=\chi'(G_\alpha)<\chi'_B(G_\alpha)=\chi_\mu(G_\alpha)=\chi_{BM}(G_\alpha)=3.$ Constructions of Marks in \cite{Marks} show that the greedy bound cannot always be improved, even in the acyclic case, in stark contrast with K\H{o}nig's bound. In the special case of $\mu$-preserving graphs better positive results are known: Cs\'{o}ka, Lippner, and Pikhurko showed that bipartite $\mu$-invariant graphs satisfy $\chi'_\mu(G)\leq \Delta(G)+1$ in \cite{csoka.lippner.pikhurko}, and this was recently improved by Pikhurko and Greb\'{\i}k to work for not necessarily bipartite graphs in \cite{pikhurko.grebik}, matching the optimal general bound in the finite case and improving on an earlier bound of Bernshteyn \cite{Bernshteyn.lll}. In the Baire and not necessarily measure preserving settings, Marks \cite{Marks} showed that graphs with $\Delta(G)\leq 3$ satisfy $\chi'_{BM}(G),\chi'_\mu(G)\leq 4,$ and Kechris and Marks \cite{kechris.marks} showed that acyclic, $d$-regular graphs satisfy $\chi'_{BM}=d$ whenever $d\geq 3.$ In the Borel setting the edge chromatic numbers of Schreier graphs of abelian groups were very recently characterized in \cite{bht, grebik.rozhon, weilacher.abelian}, the latter of whom also proved a $\Delta(G)+1$ bound for some actions of polynomial growth groups, but few other general results seem to be known.
\vspace{2mm}
In this paper we prove a Baire measurable version of K\H{o}nig's theorem.
\vspace{2.5mm}
\begin{theorem}\label{BM konig}
Let $G$ be a bipartite Borel graph. Then $\chi'_{BM}(G)\leq \Delta(G)+1.$
\end{theorem}
\vspace{2mm}
We can also obtain the same result in the (not necessarily pmp) measure setting, so long as the graph is hyperfinite $\mu$-a.e., i.e. when some invariant conull induced subgraph of $G$ is a countable increasing union of Borel graphs with finite connected components (component finite). Note that every locally countable Borel graph is hyperfinite on an invariant comeagre set, and that many naturally arising graphs, such as the Schreier graphs of free Borel actions of amenable groups, are hyperfinite a.e.
\vspace{2mm}
\begin{theorem}\label{mu konig}
Let $X$ be a Polish space with probability measure $\mu,$ and let $G$ be a $\mu$-hyperfinite bipartite Borel graph on $X$. Then $\chi'_\mu(G)\leq \Delta(G)+1.$
\end{theorem}
\vspace{2mm}
Unfortunately there is no hope of proving the same bound in the Borel setting, as a result of Conley, Jackson, Marks, Seward, and Tucker-Drob \cite{CJMSTD} shows that there are hyperfinite, $d$-regular acyclic graphs with $\chi'_B(G)=2d-1$ for any $d.$ Still, in the special case of $2$-ended graphs we get the following strengthening:
\vspace{2mm}
\begin{theorem}\label{2 bip}
Let $G$ be a bipartite Borel graph whose connected components all have 2 ends. Then $\chi'_B(G)\leq \Delta(G)+1.$
\end{theorem}
\vspace{2mm}
This problem was considered by Weilacher in \cite{weilacher}, where a bound of $\Delta(G)+1$ was proven for actions of $2$-ended groups and where it was conjectured that this bound suffices in general. This confirms the conjecture for bipartite graphs.
The previous three theorems will follow from a single more general one which is stated in terms of asymptotic separation index, a parameter recently defined in \cite{dimension}. We say a locally finite Borel graph $G$ has asymptotic separation index $\leq s$, written $\asi(G) \leq s$, where $s \in \omega$, if for any $r \in \omega$ we can find Borel sets $U_0,\ldots,U_s$ which cover $X$ such that for each $i$, the graph $G^{\leq r} \res U_i$ is component finite, where $G^{\leq r}$ is the Borel graph on $X$ for which two points are adjacent if the path distance between them in $G$ is at most $r$. Our general theorem is:
\begin{theorem}\label{asi alt}
Let $G$ be a bipartite Borel graph with finite asymptotic separation index. Then $\chi_B'(G) \leq \Delta(G) + \asi(G)$.
\end{theorem}
In \cite{weilacher.abelian}, it was asked whether a bound of $\Delta(G)+1$ holds for general Borel graphs $G$ with $\asi(G) = 1$. Again, this confirms a positive answer for bipartite graphs.
The parameter $\asi$ has been used in similar results in \cite{dimension} (Theorem 8.1). The authors in that paper also note that it is open whether there is any $G$ with $1 <\asi(G) < \infty$. Theorem \ref{asi alt} provides further motivation to study this question.
Note that if $\Delta(G)$ is infinite, Theorem \ref{asi alt} is automatic by the Lusin-Novikov uniformization theorem. Thus for the rest of the paper we assume $\Delta(G)$ is finite. Note also that Theorem \ref{asi alt} gives the classical K\H{o}nig's theorem when $\asi(G) = 0$. The following results show that Theorem \ref{asi alt} immediately implies Theorems \ref{BM konig}, \ref{mu konig}, and \ref{2 bip}:
\begin{theorem}[Essentially \cite{conley.miller.toast},\cite{miller.2end}]
Let $G$ be a locally finite Borel graph on a Polish space $X$, and $\mu$ a Borel probability measure on $X$
\begin{enumerate}
\item There is a $G$-invariant comeager Borel set $C \subset X$ such that $\asi(G \res C) \leq 1$.
\item If $G$ is $\mu$-hyperfinite, there is a $G$-invariant $\mu$-conull Borel set $C \subset X$ such that $\asi(G \res C) \leq 1$.
\item If each connected component of $G$ has 2 ends, $\asi(G) = 1$.
\end{enumerate}
\end{theorem}
Finally, we mention that determining if the measurable K\H{o}nig bound holds for bipartite Borel graphs that are not hyperfinite or pmp is still open. By applying results on measurable fractional perfect matchings from \cite{bowen.kun.sabok} we may reduce the problem to determining if acyclic, leafless graphs admit measurable matchings that cover all but a sparse set of vertices. In the hyperfinite case this reduces further to the $2$-ended case by applying Theorem A from \cite{conley.miller.pm}, which motivated the results in the present paper. However, a recent counterexample of Kun \cite{gabor.new} shows that no analogue of \cite{conley.miller.pm} Theorem A holds for graphs that are not hyperfinite, so a different approach must be used.
% \section{Proofs}
% Graphs with asymptotic separation index 1 can be approximated well by component finite subgraphs. A typical strategy for coloring them, therefore, is to define colorings on those components just using finite combinatorics, then somehow address global coherence. We will use the classical K\H{o}nig's theorem for the finite combinatorial part. The next lemma helps with coherence. It is an alternate version of Lemma 1 from \cite{weilacher.abelian}.
% For $U$ a set of vertices and $r \in \omega$, define $B(U,r)$ to be the set of vertices with path distance from $U$ at most $r$, and $E(U,r)$ to be the set of vertices with path distance from $U$ exactly $r$.
% \begin{lemma}\label{barriers}
% Let $G$ be a locally finite Borel graph on $X$ with $\asi(G) = 1$. For any $\Delta \in \omega$, we can find Borel sets $S_1,\ldots,S_\Delta \subset X$ such that for each $i$, the restrictions of $G$ to $B(S_i,2)$ and $X \setminus S_i$ are component finite, and for $i \neq j$, the path distance from $S_i$ to $S_j$ is at least 5.
% \end{lemma}
% \begin{proof}
% Since we may clearly assume $\Delta \geq 2$, and by the distance requirement $B(S_i,2) \subset X \setminus S_j$ whenever $i \neq j$, we only need to worry about the component finiteness of each $X \setminus S_i$.
% Let $X = U \sqcup V$ be a Borel partition of $X$ witnessing $\asi(G) = 1$ for $r = 10\Delta+1$. That is, such that the graphs $G^{\leq r} \res U$ and $G^{\leq r} \res V$ are component finite. For each $1 \leq i \leq \Delta$, set $S_i = E(U,5i)$. Also set $W = X \setminus B(U,5\Delta)$. The $S_i$'s are Borel since $U$ is, and clearly the path distance between any two distinct ones is at least 5.
% Fix $i$. We want to show $G \res (X \setminus S_i)$ is component finite. By K\H{o}nig's lemma (not theorem!), It suffices to show that any injective path $x_0,x_1,\ldots$ through $G$ has some point in $S_i$. Fix such a path. We first show it contains some point in $W$. If not, then for all $n$, there is $y_n \in U$ such that the distance from $x_n$ to $y_n$ is at most $5\Delta$. Then for all $n$, the distance from $y_n$ to $y_{n+1}$ is at most $10\Delta + 1 = r$. Thus the $y_n$'s all lie in the same $G^{\leq r} \res U$-component, so there are finitely many of them. But $\{x_n\} \subset B(\{y_n\},5\Delta)$ is infinite, contradicting local finiteness.
% Similarly, our path must contain some point from $U$. Suppose $x_n \in U$, $x_{n'} \in W$. Consider the distance from $U$ to $x_m$ for $m$ between $n$ and $n'$. When $m = n$, it is 0. When $m = n'$, it is greater than $5\Delta$. When $n$ changes by 1, it can change by at most 1. Thus since $0 \leq 5i \leq 5\Delta$, there must be some such $n$ for which $x_n \in E(U,5i)$.
% \end{proof}
% The proof of the classical K\H{o}nig's theorem proceeds by successively removing matchings from the initial graph, each time lowering the maximum degree. Our approach is similar, except that the edge sets removed are not quite matchings. The following lemma is our inductive step.
% We say a vertex is covered by a set of edges (e.g. a matching) if it is contained in some edge in the set.
% \begin{lemma}\label{inductive}
% Let $G$ be a bipartite Borel graph on $X$ with $\Delta(G) \leq \Delta \in \omega$. Let $S \subset X$ Borel such that the restrictions of $G$ to $B(S,2)$ and $X \setminus S$ are component finite. Then there are Borel matchings $M,N \subset G$ such that each $x \in X$ of degree $\Delta$ is covered by $M \cup N$, and each edge of $N$ is contained in $B(S,2)$.
% \end{lemma}
% \begin{proof}
% Consider the component finite bipartite Borel graphs $G \res B(S,2)$ and $G \res (X \setminus S)$. By the classical K\H{o}nig's theorem, each component of each of these graphs admits a matching which covers its vertices of maximal degree. By the Lusin-Novkikov uniformization theorem, there is a Borel way of picking such a matching for each component. Thus, let $M \subset G \res (X \setminus S)$ be a Borel matching covering each vertex of $G \res (X \setminus S)$-degree $\Delta$, and likewise for $N$ and $G \res B(S,2)$.
% Let $x$ have $G$-degree $\Delta$; we need to show $x$ is covered by $M \cup N$. If $x \in B(S,1)$, then $x$ still has degree $\Delta$ in $G \res B(S,2)$, so it is covered by $N$. If not, then $x$ still has degree $\Delta$ in $G \res (X \setminus S)$, so it is covered by $M$.
% \end{proof}
% We now prove Theorem \ref{asi1}:
% \begin{proof}
% Let $\Delta = \Delta(G)$, and $X = V(G)$. Let $S_1,\ldots,S_\Delta \subset X$ be given by lemma \ref{barriers}. Let $G_1 = G$.
% Suppose $1 \leq i \leq \Delta$ and we have Borel $G_i \subset G$ with maximum degree $ \leq \Delta - i + 1$. Apply lemma \ref{inductive} to $G_i$ and $S_i$ to get Borel matchings $M_i,N_i \subset G_i$ whose union covers each vertex of $G_i$-degree $\Delta-i+1$ and with each edge of $N_i$ contained in $B(S_i,2)$. Let $G_{i+1} = G_i \setminus (M_i \cup N_i)$. Then $G_{i+1}$ is Borel and has maximum degree $\leq \Delta - i$, so we can repeat this process.
% At the end, $G_{\Delta+1}$ has maximum degree $\leq 0$, and so is empty. Thus, the matchings we removed along the way, $M_1,\ldots,M_\Delta$ and $N_1,\ldots,N_\Delta$, union to all of $G$. Furthermore, since $B(S_i,2)$ and $B(S_j,2)$ are disjoint for $i \neq j$, $\bigcup_i N_i$ is still a Borel matching, call it $N$. Now $N$ and the $M_i$'s give a Borel $(\Delta+1)$-edge coloring of $G$.
% \end{proof}
\section{Proofs}
Graphs with finite asymptotic separation index can be approximated well by component finite subgraphs. A typical strategy for coloring them, therefore, is to define colorings on those components just using finite combinatorics, then somehow address global coherence. We will use the classical K\H{o}nig's theorem for the finite combinatorical part. The next lemma helps with coherence.
For $U$ a set of vertices and $r \in \omega$, define $B(U,r)$ to be the set of vertices with path distance from $U$ at most $r$, and $E(U,r)$ to be the set of vertices with path distance from $U$ exactly $r$.
\begin{lemma}\label{barriers alt}
Let $G$ be a locally finite Borel graph on $X$ with $\asi(G) \leq s \in \omega$. For any $\Delta \in \omega$, we can find Borel sets $S_i^j \subset X$ for $j < s$ and $i < \Delta$ such that, letting $S_i = \bigcup_j S_i^j$,
\begin{enumerate}
\item The restrictions of $G$ to $B(S_{i}^j,2)$ for each $i,j$ and to $(X \setminus S_i)$ for each $i$ are component finite.
\item For each $j$ and $i \neq i'$, the path distance from $S_i^j$ to $S_{i'}^j$ is at least 5.
\end{enumerate}
\end{lemma}
\begin{proof}
Let $X = U_0 \sqcup \cdots \sqcup U_{s-1} \sqcup V$ be a Borel partition of $X$ witnessing $\asi(G) \leq s$ for $r = 10\Delta+1$. That is, such that the graphs $G^{\leq r} \res U_j$ for each $j$ and $G^{\leq r} \res V$ are component finite. For each $i < \Delta$ and $j < s$, set $S_{i}^j = E(U_j,5i)$. Each $S_i^j$ is Borel since each $U_j$ is, and the distance requirement (2) is clearly satisfied.
Fix $i$. We want to show $G \res (X \setminus S_i)$ is component finite. By K\H{o}nig's lemma (not theorem!), it suffices to show that if $x_0,x_1,\ldots$ is an injective path through $G$, then it contains a point of $S_i$. We first show that for all $j$, it contains a point of $X \setminus B(U_j,5\Delta)$. If not, then for all $n$, there is a $y_n \in U_j$ with distance at most $5\Delta$ from $x_n$. Then for all $n$, the distance between $y_n$ and $y_{n+1}$ is at most $10\Delta+1 = r$, so the $y_n$'s are all in the same $G^{\leq r} \res U_j$-component, so there are only finitely many of them. But $\{x_n\} \subset B(\{y_n\},5\Delta)$ is infinite, contradicting local finiteness.
By the same proof, our path contains some point of $X \setminus V$, say $x_{n_0} \in U_j$. Let $n$ be such that $x_n \not\in B(U_j,5\Delta)$. Consider the distance from $U_j$ to $x_m$ for $m$ between $n_0$ and $n$. When $m = n_0$, it is 0. When $m = n$, it is more than $5\Delta$. When $m$ changes by 1, it can change by at most 1. Thus since $0 \leq 5i < 5\Delta$, there must be some $m$ for which $x_m \in E(U_j,5i) = S_i^j \subset S_i$.
Fix $i,j$. We want to show that $G \res B(S_i^j,2)$ is component finite. Above, we saw that any infinite injective path contains a point of distance more than $5\Delta$ from $U_j$. All points of $B(S_i^j,2)$ have distance at most $5i+2 < 5\Delta$ from $U_j$, though, so we are again done by K\H{o}nig's lemma.
\end{proof}
The proof of the classical K\H{o}nig's theorem proceeds by successively removing matchings from the initial graph, each time lowering the maximum degree. Our approach is similar, except that the edge sets removed are not quite matchings. The following lemma is our inductive step.
We say a vertex is covered by a set of edges (e.g. a matching) if it is contained in some edge in the set.
\begin{lemma}\label{inductive alt}
Let $G$ be a bipartite Borel graph on $X$ with $\Delta(G) \leq \Delta \in \omega$. Let $s \in \omega$. Let $S^j \subset X$ for $j < s$ such that, letting $S = \bigcup_j S^j$, the restrictions of $G$ to $X \setminus S$ and to $B(S^j,2)$ for each $j$ are component finite. Then there are Borel matchings $M,N^0,\ldots,N^{s-1} \subset G$ such that each vertex of degree $\Delta$ is covered by $M \cup \bigcup_j N^j$, and for each $j$, each edge of $N^j$ is contained in $B(S^j,2)$.
\end{lemma}
\begin{proof}
Consider the component finite bipartite Borel graphs $G \res (X \setminus S)$ and $G \res B(S^j,2)$ for $j < s$. By the classical K\H{o}nig's theorem, each component of each of these graphs admits a matching which covers its vertices of maximal degree. By the Lusin-Novikov uniformization theorem, there is a Borel way of picking such a matching for each component. Thus, let $M \subset G \res (X \setminus S)$ be a Borel matching covering each vertex of $G \res (X \setminus S)$-degree $\Delta$, and likewise for $N^j$ and $G \res B(S^j,2)$ for each $j$.
Let $x$ have $G$-degree $\Delta$; we need to show $x$ is covered by $M \cup \bigcup_j N^j$. If $x \in B(S^j,1)$ for some $j$, then $x$ still has degree $\Delta$ in $G \res B(S^j,2)$, so it is covered by $N^j$. Else, $x$ has degree $\Delta$ in $G \res (X \setminus S)$, so it is covered by $M$.
\end{proof}
We now prove Theorem \ref{asi alt}:
\begin{proof}
Let $\Delta = \Delta(G) \in \omega$, $s = \asi(G) \in \omega$, and $X = V(G)$. Let $S_i^j \subset X$ for $i < \Delta$, $j < s$ be as given by Lemma \ref{barriers alt}. Let $G_0 = G$.
Suppose $i < \Delta$ and we have Borel $G_i \subset G$ with maximum degree $\leq \Delta - i$. Apply Lemma \ref{inductive alt} to $G_i$ and the $S_i^j$'s to get Borel matchings $M_i,N_i^0,\ldots,N_i^{s-1}$ whose union covers each vertex of $G_i$-degree $\Delta - i$ and with each edge of $N_i^j$ contained in $B(S_i^j,2)$ for each $j$. Let $G_{i+1} = G_i \setminus (M_i \cup \bigcup_j N_i^j)$. Then $G_{i+1}$ is Borel and has maximum degree $\leq \Delta - i - 1$, so we can repeat this process.
At the end, $G_\Delta$ has maximum degree $\leq 0$, and so is empty. Thus, the matchings we removed along the way, $M_i$ and $N_i^j$ for $i < \Delta$, $j < s$, union to all of $G$. Furthermore, for each $j$, since $B(S_i^j,2)$ and $B(S_{i'}^j,2)$ are disjoint for $i \neq i'$, $\bigcup_i N_i^j$ is still a Borel matching, call it $N_j$. Now the $M_i$'s and $N_j$'s give a Borel $(\Delta + s)$-edge coloring of $G$.
\end{proof}
\section*{Acknowledgements}
The second author was partially supported by the ARCS foundation, Pittsburgh chapter.
% The finite version of K\H{o}nig's thoerem is typically proven by embedding a bipartite graph $G$ in a larger $\Delta(G)$-regular bipartite graph, and successively using the fact that such graphs admit perfect matchings to color the edges of the original graph $G.$ Our strategy is essentially the same, but now we must deal with the fact that regular bipartite graphs do not necessarily have measurable perfect matchings. Still, we will show that such graphs have perfect matchings besides on unions of biinfinite lines, and use the extra $\Delta(G)+1$'st color to fix any problems caused by these lines.
% Throughout this section we fix a Borel graph $G$ that is $\mu$-hyperfinite. We proceed in a few steps:
% \begin{lemma}\label{reg}
% We may assume that $G$ is $\Delta(G)$-regular.
% \end{lemma}
% \begin{proof}
% Given $G,$ let $G'$ be the graph with vertex set $G\times\{0,1\}$ and $E(G)=\{\{(x,i),(y,i)\}: xGy\}\cup \{\{(x,i),(x,j)\}: i\neq j \textnormal{ and }deg(x)<\Delta(G)\}.$ $G'$ is still bipartite, hyperfinite, and has max degree $\Delta(G),$ and $\delta(G')=\Delta(G)+1.$ Iterating this until $\delta(G')=\Delta(G')$ gives the desired regular graph.
% \end{proof}
% Recall that a \textit{fractional perfect matching} is a function $f:E(G)\rightarrow [0,1]$ such that $\sum_{e: v\in e}f(e)=1$ for every vertex $v\in V(G),$ and that given an even length cycle $C$ in a graph with a distinguished edge and $\varepsilon \in \mathbb{R}$, an
% \textit{alternating $\varepsilon$-circuit} on $C$ is a function that is equal to $\varepsilon$ on the even edges and $-\varepsilon$ on the odd edges (even and odd edges with respect to their distance from the distinguished one).
% For bipartite graphs the existence of a fractional perfect matching is equivalent to the existence of a perfect matching, and recently measurable fractional perfect matchings were used by the author, Kun, and Sabok \cite{bowen.kun.sabok} to show that one-ended graphings have perfect matchings a.e. The following lemma is similar to Lemma 5.1 in \cite{bowen.kun.sabok}, but the proof is somewhat different as we are not only working in the measure setting and cannot assume that acyclic graphs have at most two ends.
% \begin{lemma}\label{acyclic bip}
% Any $d$-regular hyperfinite bipartite Borel graph $G$ admits an acyclic leafless Borel subgraph $H$ such that the induced graph on $V(G)\setminus V(H)$ has a Borel perfect matching.
% \end{lemma}
% \begin{proof}
% First note that the constant $\frac{1}{\Delta(G)}$ function $f:E(G)\rightarrow [0,1]$ is a Borel fractional perfect matching since $G$ is degree regular. Now, let $(G_n)_{n\geq 1}$ be a sequence of component finite Borel graphs witnessing that $G$ is hyperfinite. We inductively construct a sequence of Borel fractional perfect matchings $f=f_0,f_1,...$ such that there are no cycles $C$ in a $G_n$ component such that $0<f_n(e)<1$ for all $e\in C,$ and so that if $f_n(e)\in \{0,1\}$ then $f_n(e)=f_{n+1}(e).$ Towards this, assume that we have such an $f_n.$ Given a cycle $C$ in a $G_{n+1}$ component not satisfying the restriction we may apply an alternating $\varepsilon$-circuit to $C,$ where the specified edge and $\varepsilon$ are chosen so that at least one edge in $C$ becomes integral after applying the circuit and $\varepsilon=\min_e\{d_\mathbb{R}(f_n(e),1), d_\mathbb{R}(f_n(e),0)\}.$ Note that the result of applying such a circuit to a fractional perfect matching is still a fractional perfect matching. Let $f_{n+1}$ be the function obtained by iteratively applying this process to cycles in $G_{n+1}$ components.
% Let $M=\bigcup_n\{x: \exists y \textnormal{ with } f_n(x,y)=1\}$ and note that the set of edges with $f(e)=1$ is a Borel matching of $M.$ Now, let $H'$ be the induced subgraph on $V(G)\setminus M$ and let $H=H'\setminus\bigcup_nf_n^{-1}(0).$ We claim that this set satisfies the desired properties. To see this, first suppose that $C$ is a cycle in $H.$ Then there is some $n$ such that $C$ is contained in a $G_n$ component. Since $C\in H$ we know $f_n(e)\notin \{0,1\}$ for each $e\in C,$ but this contradicts our construction of $f_n.$ Similarly, suppose that $v$ is a degree $1$ vertex in $H,$ say with neighbor $u$. Let $n$ be the first integer such that $f_n(v,w)\in \{0,1\}$ for all $w\in N(v)\setminus u.$ Then we have that $\sum_{e: v\in e}f(e)\in \mathbb{R}\setminus \mathbb{N},$ contradicting the construction of $f_n.$
% \end{proof}
% Our next goal is to reduce from acyclic leafless graphs to biinfinite lines. This can be accomplished with the help of the following already known results on perfect matchings:
% \begin{theorem}[\cite{conley.miller.pm}\label{CM} Theorems A and B]
% Let $G$ be an acyclic leafless $\mu$-hyperfinite graph. $G$ admits an invariant conull set which has a perfect matching off of a union of biinfinite lines.
% \end{theorem}
% It is interesting to observe that in the measure setting we do not need hyperfiniteness to prove Lemma \ref{acyclic bip} (the proof of Theorem 5.1 in \cite{bowen.kun.sabok} works for this), so it is tempting to hope that Theorem \ref{CM} can be carried out without hyperfiniteness as well, which would allow one to deduce that $\chi'_\mu(G)\leq \Delta(G)+1$ for all bipartite graphs. Kun \cite{gabor.new} recently found acyclic, $d$-regular, pmp counterexamples to the non-hyperfinite version of Theorem \ref{CM} for all $d,$ so this approach can not work.
% \begin{theorem}[Essentially \cite{marks-unger} Theorem 1.3]\label{mu}
% Suppose that is $G$ is a locally finite bipartite Borel graph with bipartition $\{A_0,A_1\},$ and $S$ is a Borel subgraph of $G$ such that for some $\epsilon>0$ any finite $F\subset A_i\cap S$ satisfies $|N(F)|\geq (1+\epsilon)|F|.$ Then there is a Borel matching which covers $S$ up to an invariant comeagre set.
% \end{theorem}
% We note that the previous theorem was only stated for the case $S=G$ in Marks and Unger's paper, but the proof of the stated result is identical to their proof up to considering unmatched vertices in $S$ rather than $G$ during the construction.
% \begin{lemma}\label{biinf}
% Let $G$ be a bipartite $d$-regular $\mu$-hyperphinite Borel graph. $G$ admits a comeagre Borel set $A$ and a Borel matching $M$ such that the set of unmatched vertices in $G|A$ is a union of biinfinite lines. Similarly for an invariant conull Borel set.
% \end{lemma}
% \begin{proof}
% By Lemma \ref{acyclic bip} we may assume that $G$ is acyclic and leafless. For the measure result, we may apply Theorem \ref{CM} directly. To apply the Baire result, we must modify the graph slightly. This modification is precisely the same as in the comment following Theorem 12.2 in \cite{kechris.marks}, but we sketch it here for convenience. Let $T$ be the set of vertices of degree at least $3$ in $G,$ and let $F$ be the set of vertices in finite connected components of $G\setminus T.$ As any vertex in $F$ has $G$ degree $2$, we know any perfect matching in $G$ must alternate along edges in $F,$ and so we can contract along such paths to obtain a graph $G'$ so that any two vertices in $F'$ have $G'$ distance at least $2$ and so that any matching covering a comeagre set of vertices in $T\cup F'$ extends to a matching covering a comeagre set of vertices in $T\cup F,$ where $F'$ is the set of vertices in finite connected components of $G'\setminus T.$ Then $S=T\cup F'$ satisfies the conditions of Theorem \ref{mu}, and after removing $S$ we are left with a union of infinite rays (which are smooth and thus admit a Borel matching) and biinfinite lines.
% \end{proof}
% We are now in a position to prove our main results.
% \vspace{2mm}
% \begin{proof}[Proof of Theorems \ref{BM konig} and \ref{mu konig}]
% We inductively define color classes $C_i$ and $C_{\Delta(G)+1,i}$ so that generically many/almost every vertex sees at least one edge in $C_i\cup C_{\Delta(G)+1,i}$ and the $G$ distance between any two edges in $\bigcup_iC_{\Delta(G)+1,i}$ is very large. Once this is done $C_1,...,C_{\Delta(G)}$ together with $C_{\Delta(G)+1}=\bigcup_iC_{\Delta(G)+1,i}$ will be the desired edge coloring.
% Towards this, suppose that we have defined $C_1,...,C_{i-1}$ and their associated $C_{\Delta(G)+1,j}$ as stated. We may assume that $G'=G\setminus \bigcup_{j\leq i-1}(C_j\cup C_{\Delta(G)+1,j})$ is degree regular by Lemma \ref{reg}. Let $C_i'$ be all of the edges in the matching of $G'$ given by Lemma \ref{biinf} and let $L$ be the remaining biinfinite lines. By \cite[Proposition 4.5]{kst} we may find a Borel maximal independent set in the distance $n$ graph of $L$, and from this find a Borel tiling of $L$ with connected sets of length $n$ or $n+1$ for some very large $n$ depending only on $\Delta(G).$ Since the edges in $\bigcup_{j\leq i-1} C_{\Delta(G)+1,j}$ have large pairwise distance we may shift tiles to the left or right several spaces to ensure that cut edges between tiles have large distance with edges in $\bigcup C_{\Delta(G)+1,j}.$ These tiles admit perfect matchings besides on at most an endpoint. Let $C_i$ be the union of $C_i'$ and the edges in these matchings, and let $C_{\Delta+1,i}$ be edges connecting unmatched endpoints to adjacent tiles.
% \end{proof}
% \vspace{1mm}
% The proof Theorem \ref{2 bip} is similar.
% \vspace{2mm}
% \begin{proof}[Proof of Theorem \ref{2 bip}]
% Let $\mathcal{S}$ be the family of all finite connected sets such that the induced graph on $V(G)\setminus V(S)$ has exactly $2$ infinite components for each $G$ component and the $G$ distance between these components is at least $3$ for all $S\in\mathcal{S}.$ Since $G$ is locally finite we have a countable Borel coloring of the intersection graph of $G$ by \cite[Proposition 4.5]{kst}, and so we may find a Borel family of such cutsets that are pairwise disjoint and form a complete section for $G.$ We may further partition this into $\chi'(G)$ many families $\mathcal{S}_i$ of cutsets with the same properties.
% We inductively define color classes $C_i$ and $C_{\Delta(G)+1,i}$ so that every vertex of degree $\Delta(G)$ sees at least one edge in $C_i\cup C_{\Delta(G)+1,i}$ and edges in $C_{\Delta(G)+1,i}$ are only between vertices in elements of $\mathcal{S}_i.$
% Assume that we have defined $C_1,...,C_{i-1}$ and their associated $C_{\Delta(G)+1,j}.$ From our assumptions we know that $G'=G\setminus \bigcup_{j\leq i-1}(C_j\cup C_{\Delta(G)+1,j})$ has max degree $\Delta(G)-i+1,$ and so each finite component of $G'\setminus V(\mathcal{S}_i)$ admits a $\Delta(G)-i+1$ coloring that extends to the rest of the graph. Let $C_i'$ be all of the edges recieving color $1$ in each finite component for such colorings along with edges colored $1$ that leave a component in that component's extension. Now, for each cut set $S\in S_i$ let $S'=S\setminus \{v: v \textnormal{ sees an edge colored } 1\}$ and consider edges in $S'$ that would be colored $1$ in either the finite component to the left's or right's extension of their colorings. The induced graph has maximum degree at most $2$ and is bipartite. Therefore, we may $2$-edge color the graph induced on $S'$ and let $C_i$ be the edges in $C_i'$ together with edges colored $1$ in cutsets $S',$ and $C_{\chi'(G)+1,i}$ be the edges colored $2$ in the cutsets $S'.$
% \end{proof}
\bibliographystyle{amsalpha}
\bibliography{main}
\end{document}
\newcommand{\etalchar}[1]{$^{#1}$}
\providecommand{\bysame}{\leavevmode\hbox to3em{\hrulefill}\thinspace}
\providecommand{\MR}{\relax\ifhmode\unskip\space\fi MR }
% \MRhref is called by the amsart/book/proc definition of \MR.
\providecommand{\MRhref}[2]{%
\href{http://www.ams.org/mathscinet-getitem?mr=#1}{#2}
}
\providecommand{\href}[2]{#2}
\begin{thebibliography}{CJM{\etalchar{+}}20b}
\bibitem[Ber19]{Bernshteyn.lll}
Anton Bernshteyn, \emph{Measurable versions of the {L}ov\'{a}sz local lemma and
measurable graph colorings}, Adv. Math. \textbf{353} (2019), 153--223.
\MR{3979016}
\bibitem[BHT]{bht}
Ferenc Bencs, Aranka Hru\v{s}kov\'{a}, and L\'{a}szl\'{o}~M\'{a}rton T\'{o}th,
\emph{Factor-of-iid schreier decorations of lattices in {E}uclidean spaces},
preprint, arXiv:2101.12577.
\bibitem[BKS]{bowen.kun.sabok}
Matthew Bowen, G\'abor Kun, and Marcin Sabok, \emph{Perfect matchings in
hyperfinite graphings}, preprint, arxiv:2106.01988.
\bibitem[CJM{\etalchar{+}}20a]{dimension}
Clinton Conley, Steve Jackson, Andrew Marks, Brandon Seward, and Robin
Tucker-Drob, \emph{Borel asymptotic dimension and hyperfinite equivalence
relations}, 2020, preprint, arXiv:2009.06721.
\bibitem[CJM{\etalchar{+}}20b]{CJMSTD}
Clinton Conley, Steve Jackson, Andrew Marks, Brandon Seward, and Robin
Tucker-Drob, \emph{Hyperfiniteness and {B}orel combinatorics}, J. Eur. Math.
Soc. (JEMS) \textbf{22} (2020), no.~3, 877--892. \MR{4055991}
\bibitem[CLP16]{csoka.lippner.pikhurko}
Endre Cs\'{o}ka, G\'{a}bor Lippner, and Oleg Pikhurko, \emph{K{\H o}nig's line
coloring and {V}izing's theorems for graphings}, Forum Math. Sigma \textbf{4}
(2016), Paper No. e27, 40. \MR{3569061}
\bibitem[CM16]{conley.miller.toast}
Clinton~T. Conley and Benjamin~D. Miller, \emph{A bound on measurable chromatic
numbers of locally finite {B}orel graphs}, Math. Res. Lett. \textbf{23}
(2016), no.~6, 1633--1644.
\bibitem[CM17]{conley.miller.pm}
\bysame, \emph{Measurable perfect matchings for acyclic locally countable
{B}orel graphs}, J. Symb. Log. \textbf{82} (2017), no.~1, 258--271.
\bibitem[GP20]{pikhurko.grebik}
Jan Greb\'{\i}k and Oleg Pikhurko, \emph{Measurable versions of {V}izing's
theorem}, Adv. Math. \textbf{374} (2020), 107378, 40. \MR{4157578}
\bibitem[GR]{grebik.rozhon}
Jan Greb\'{\i}k and V\'aclav Rozho\v{n}, \emph{Local problems on grids from the
perspective of distributed algorithms, finitary factors, and descriptive
combinatorics}, preprint, arXiv:2103.08394.
\bibitem[KM16]{kechris.marks}
Alexander~S. Kechris and Andrew~S. Marks, \emph{Descriptive graph
combinatorics}, 2016, preprint.
\bibitem[KST99]{kst}
A.~S. Kechris, S.~Solecki, and S.~Todorcevic, \emph{Borel chromatic numbers},
Adv. Math. \textbf{141} (1999), no.~1, 1--44.
\bibitem[Kun]{gabor.new}
G\'abor Kun, \emph{The measurable {H}all theorem fails for treeings}, preprint,
arXiv:2106.02013.
\bibitem[Mar16]{Marks}
Andrew~S. Marks, \emph{A determinacy approach to {B}orel combinatorics}, J.
Amer. Math. Soc. \textbf{29} (2016), no.~2, 579--600. \MR{3454384}
\bibitem[Mil09]{miller.2end}
Benjamin Miller, \emph{Ends of graphed equivalence relations, i}, Israel
Journal of Mathematics \textbf{169} (2009), 375--392.
\bibitem[Weia]{weilacher.abelian}
Felix Weilacher, \emph{Borel edge colorings for finite dimensional groups},
preprint, arXiv:2104.14646.
\bibitem[Weib]{weilacher}
\bysame, \emph{Borel vizing's theorem for 2-ended groups},
https://arxiv.org/abs/2101.12740v1.
\end{thebibliography}