%%%% THIS FILE IS AUTOMATICALLY GENERATED %%%% DO NOT EDIT THIS FILE DIRECTLY, %%%% ONLY EDIT THE SOURCE, tom-55/document.tex. %%%% %% Standard package list \documentclass[letterpaper]{article} \usepackage[utf8]{inputenc} \usepackage[T1]{fontenc} \usepackage[english]{babel} \usepackage[top=3cm, bottom=3cm, left=3.5cm, right=3.5cm]{geometry} \usepackage[onehalfspacing]{setspace} \usepackage{amsmath,amssymb,amsthm,wasysym} \usepackage{nicefrac,booktabs} \usepackage{mathptmx} \usepackage{cite} \usepackage[colorlinks=true]{hyperref} %% Various helpers for Tom's papers \newcommand{\gs}{\textnormal{gs}} \newcommand{\ord}{\textnormal{ord}} \newcommand{\Exp}{\textnormal{Exp}} \newcommand{\Log}{\textnormal{Log}} \newcommand{\lcm}{\textnormal{lcm}} \newcommand{\range}{\textnormal{range}} \newcommand{\NR}{\textnormal{NR}} \newcommand{\Mod}[1]{\left(\textnormal{mod}~#1\right)} \newcommand{\ap}[2]{\left\langle #1;#2 \right\rangle} \newcommand{\summ}[1]{\sum_{k=1}^m{#1}} \newcommand{\bt}[1]{{{#1}\mathbb{N}}} \newcommand{\fp}[1]{{\left\lbrace{#1}\right\rbrace}} \newcommand{\intv}[1]{{\left[1,{#1}\right]}} %% Lifted from http://stackoverflow.com/questions/2767389/referencing-a-theorem-like-environment-by-its-name %% This lets me do things like "Theorem A" and have the references work properly. \makeatletter \let\@old@begintheorem=\@begintheorem \def\@begintheorem#1#2[#3]{% \gdef\@thm@name{#3}% \@old@begintheorem{#1}{#2}[#3]% } \def\namedthmlabel#1{\begingroup \edef\@currentlabel{\@thm@name}% \label{#1}\endgroup } \makeatother % end lift \newtheoremstyle{namedthrm} {}{}{}{}{}{}{ } % This last space needs to be there {\bf\thmname{#1} \thmnote{#3}.} %% End reference hack %% Document start \date{} \begin{document} %% Content start \newtheorem{cor}{Corollary} \newtheorem*{lemma}{Lemma} \newtheorem{thm}{Theorem} \newtheorem{q}{Problem} \theoremstyle{definition} \newtheorem{defn}{Definition} \newtheorem*{remark}{Remark} \newtheorem*{fact}{Fact} \title{Common Transversals} \author{T. C. Brown \\ Communicated by P. Erd\H{o}s} \date{} \maketitle \begin{center}{\small {\bf Citation data:} T.C. {Brown}, \emph{Common transversals}, J. Combin. Theory Ser. A \textbf{21} (1976), 80--85.}\bigskip\end{center} \begin{abstract} Given $t$ families, each family consisting of $s$ finite sets, we show that if the families ``separate points" in a natural way, and if the union of all the sets in all the families contains more than $(s + 1)^t - s^{t-1} - 1$ elements, then a common transversal of the $t$ families exists. In case each family is a covering family, the bound is $s^t - s^{t-1}$. Both of these bounds are best possible. This work extends recent work of Longyear~\cite{longyear1977}. \end{abstract} \section{Introduction and statement of results} \label{sec: 1} Throughout this paper, the symbol $\mathcal{F}$ will always denote a family of $t$ families of sets, each of the $t$ families consisting of $s$ finite, but not necessarily distinct or nonempty, sets. The symbol $\Omega$ will always denote the union of all of the sets contained in all of the $t$ families. Thus $\mathcal{F} = (F_1,F_2,\dots,F_t)$, where for each $j$, $1 \leq j \leq t$, $F_j = (F_j(1),F_j(2),\dots,F_j(s))$ is a family of $s$ (finite, but not necessarily distinct or nonempty) sets, and $\Omega = \bigcup \{F_j(i): 1 \leq j \leq t, 1 \leq i \leq s\}$ (or more briefly, $\Omega = \bigcup \bigcup \mathcal{F}$). We always assume that $\mathcal{F}$ \emph{separates points} of $\Omega$ in the following sense. Letting $F_j(0) = \Omega \setminus \bigcup F_j$, $1 \leq j \leq t$, we require \[ \left| \bigcap \{F_j(a_j): 1 \leq j \leq t \} \right| \leq 1 \] for every $t$-tuple $(a_1,a_2,\dots,a_t)$, where $0 \leq a_j \leq s$, $1 \leq j \leq t$. Note that this immediately implies $|\Omega| \leq (s+1)^t - 1$ (since $|\bigcap \{F_j(0): 1 \leq j \leq t\}| = 0$), and that in the case where each $F_j$ covers $\Omega$ (so that $F_j(0) = \emptyset$) we have $|\Omega| \leq s^t$. Recall that the set $T$ is a \emph{transversal} (sometimes called a \emph{system of distinct representatives} or SDR) of the family $F_j$ if there is a bijection $\varphi: T \mapsto \{1,2,\dots,s\}$ such that $x \in F_j(\varphi(x))$ for all $x \in T$. The set $T$ is a \emph{common transversal} of $F_1,F_2,\dots,F_t$ if $T$ is simultaneously a transversal of each $F_j$, $1 \leq j \leq t$. In this paper the following results are proved, which extend recent results of Judith Q. Longyear~\cite{longyear1977}. Longyear proved, among other things, Theorem~\ref{thm: 1}(b) below in the case where each family $F_j$ is assumed to consist of mutually disjoint sets. Theorem~\ref{thm: 1}(b) can be obtained as a corollary to her result. We give an alternative proof. \begin{thm} \label{thm: 1} Let $\mathcal{F}$ and $\Omega$ be as in the first paragraph of this paper, and assume further that each family $F_j$ covers $\Omega$, that is, $\bigcup F_j = \Omega$, $1 \leq j \leq t$. (a) If $|\Omega| > s^t - s^{t-1}$ then each family $F_j$ has a transversal, and $s^t - s^{t-1}$ is best possible. (b) If $|\Omega| > s^t - s^{t - 1}$ then a common transversal of $F_1,F_2,\dots,F_t$ exists, and $s^t - s^{t-1}$ is best possible. \end{thm} \begin{thm} \label{thm: 2} Let $\mathcal{F}$ and $\Omega$ be as in the first paragraph of this paper. (a) If $|\Omega| > (s + 1)^t - (s + 1)^{t-1} - 1$ then each family $F_j$ has a transversal, and $(s + 1)^t - (s + 1)^{t-1} - 1$ is best possible. (b) If $|\Omega| > (s + 1)^t - s^{t-1} - 1$ then a common transversal of $F_1,F_2,\dots,F_t$ exists, and $(s + 1)^t - s^{t-1} - 1$ is best possible. \end{thm} \section{Proofs} \label{sec: 2} Let us show first of all that the bounds given in Theorems~\ref{thm: 1} and~\ref{thm: 2} are best possible. For Theorem~\ref{thm: 1}(a) and Theorem~\ref{thm: 1}(b) let \[ \Omega = \{(a_1,a_2,\dots,a_t): 1 \leq a_j \leq s, 1 \leq j \leq t-1, 1 \leq a_t \leq s-1\}. \] For all $j,i$, $1 \leq j \leq t$, $ 1\leq i \leq s$, let $F_j(i) = \{\omega \in \Omega:$ the $j$th coordinate of $\omega$ equals $i\}$. Note that $F_t(s) = \emptyset$, so that $F_t$ has no transversal. It is easy to see that $|\Omega| = s^t - s^{t-1}$, each $F_j$ covers $\Omega$, $1 \leq j \leq t$, and $\mathcal{F} = (F_1,F_2,\dots, F_t)$ separates points. For Theorem~\ref{thm: 2}(a), we let \[ \Omega = \{(a_1,a_2,\dots,a_t): 0 \leq a_j \leq s, 1 \leq j \leq t-1, 0 \leq a_t \leq s-1\} \setminus \{(0,0,\dots,0)\}. \] For all $j,i$, $1 \leq j \leq t$, $1 \leq i \leq s$, let \[ F_j(i) = \{\omega \in \Omega: \textup{the $j$th coordinate of $\omega$ equals $i$} \}. \] Again $F_t(s) = \emptyset$ so $F_t$ has no transversal, and it is easy to see that $|\Omega| = (s + 1)^t - (s + 1)^{t-1} - 1$, $\Omega = \bigcup \bigcup \mathcal{F}$, and $\mathcal{F} = (F_1,F_2,\dots,F_t)$ separates points. For Theorem~\ref{thm: 2}(b), let $\Omega$ be the set of all $t$-tuples $(a_1,a_2,\dots,a_t)$, $0 \leq a_j \leq s$, $1 \leq j \leq t$, \emph{excluding} the set $(\{(a_1,a_2,\dots,a_t): 1 \leq a_j \leq s, 1 \leq j \leq t-1, a_t = s\} \cup \{(0,0,\dots,0)\})$. For all $j,i,$ $1 \leq j \leq t$, $1 \leq i \leq s$, let \[ F_j(i) = \{\omega \in \Omega: \textup{the $j$th coordinate of $\omega$ equals $i$} \}. \] Then any element $\omega$ of $F_t(s)$ must have its $j_0$th coordinate equal to $0$ for some $j_0 \ne t$, and hence $\omega$ cannot represent any set in the family $F_{j_0}$, therefore $\omega$ cannot belong to any common transversal of $F_1,F_2,\dots,F_t$. Therefore no common transversal exists. Again it is easy to see that $|\Omega| = (s + 1)^t - s^{t-1} - 1$, $\Omega = \bigcup \bigcup \mathcal{F}$, and $\mathcal{F} = (F_1,F_2,\dots,F_t)$ separates points. Throughout the remaining proofs, the following notation will be used. It is therefore fixed once and for all. For $t \geq 2$, let $X$ be the set of all $(t-1)$-tuples $(a_1,a_2,\dots,a_{t-1})$, where each $a_j$, $1 \leq j \leq t- 1$ satisfies $1 \leq a_j \leq s$. Note that $|X| = s^{t-1}$. For each $x = (a_1,a_2,\dots,a_{t-1}) \in X$, we denote by $f(x)$ the set $\bigcap\{F_j(a_j): 1 \leq j \leq t-1\}$. Then since $\mathcal{F}$ distinguishes points and each $F_j$ covers $\Omega$ we have $|f(x) \cap F_t(i)| \leq 1$ for all $x \in X$ and all $i$, $1 \leq i \leq s$, and $\Omega = \bigcup \{f(x): x \in X\}$. \begin{proof}[Proof of Theorem~\ref{thm: 1}(a)] The case $t = 1$ follows from the various definitions, so we assume $t \geq 2$, and without loss of generality we restrict our attention to $F_t$. We shall make use of the classical result of P. Hall~\cite{hall1935} according to which $F_t$ has a transversal if and only if $|\bigcup \{F_t(i): i \in I\}| \geq |I|$ for all $I \subset \{1,2,\dots,s\}$. Suppose that $F_t$ does not have a transversal, and that $|F_t(i_1) \cup F_t(i_2) \cup \cdots \cup F_t(i_{k-1})$ $(F_t(i_k) = \emptyset$ if $k = 1$), hence $\Omega = \bigcup \{ F_t(i): 1 \leq i \leq s, i \ne i_k\}$. Then \begin{align*} |\Omega| &= \left| \left( \bigcup \{f(x): x \in X\} \right) \cap \left( \bigcup \{F_t(i): 1 \leq i \leq s, i \ne i_k\} \right) \right| \\ &= \left| \bigcup \{f(x) \cap F_t(i): x \in X, 1 \leq i \leq s, i \ne i_k\} \right| \\ &\leq |\{(x,i): x \in X, 1 \leq i \leq s, i \ne i_k\}| \\ &= s^{t-1}(s - 1) = s^t - s^{t-1}, \end{align*} contrary to the hypothesis of the Theorem. Hence $F_t$ and similarly each $F_j$, has a transversal. \end{proof} \begin{proof}[Proof of Theorem~\ref{thm: 1}(b)] Since the family $F_t = (F_t(1), F_t(2), \dots, F_t(s))$ has a transversal (by Theorem~\ref{thm: 1}(a)) and covers $\Omega$, we can replace $F_t$ by a \emph{partition} $G = (G(1),G(2),\dots,G(s))$ such that $G(i) \subset F_t(i)$ for all $i$, $1 \leq i \leq s$. (The partition $G$ can be constructed as follows. Let $\{\omega_1,\omega_2,\dots,\omega_s\}$ be a transversal of $F_t$, where $\omega_i \in F_t(i)$, $1 \leq i \leq s$. Let \begin{align*} G(1) &= F_t(1) \setminus\{\omega_2,\dots,\omega_s\}, \\ G(2) &= F_t(2) \setminus ( G(1) \cup \{\omega_3,\dots,\omega_s\}), \\ G(3) &= F_t(3) \setminus (G(1) \cup G(2) \cup \{\omega_4,\dots,\omega_s\}), \\ &\vdots \\ G(s) &= F_t(s) \setminus (G(1) \cup G(2) \cup \cdots \cup G(s-1)). \end{align*} Then $\mathcal{F}' = (F_1,F_2,\dots,F_{t-1},G)$ distinguishes points, hence $|f(x) \cap G(i)| \leq 1$ for all $x \in X$ and all $i$, $1 \leq i \leq s$, and any common transversal of $F_1,F_2,\dots,F_{t-1}$, $G$ is a common transversal of $F_1,F_2,\dots,F_t$. At this point we could in fact replace \emph{every} $F_j$ by a partition (since we know by Theorem~\ref{thm: 1}(a) that every $F_j$ has a transversal); however, it is not necessary, and so we do not. We now demonstrate the existence of a common trasversal of $F_1,F_2,\dots,F_{t-1},G$. To this end we define a \emph{diagonal} of $X$ to be a subset $D$ of $X$ such that $|D| = s$ and for each $j$, $1 \leq j \leq t - 1$, the $j$th coordinates of the elements of $D$ run through $\{1,2,\dots,s\}$ in some order. Note that whenever $D = \{x_1,x_2,\dots,x_s\}$ is a diagonal, $\omega_k \in f(x_k)$, $ 1 \leq k \leq s$, and $\omega_1,\omega_2,\dots,\omega_s$ are all distinct, then $\{\omega_1,\omega_2,\dots,\omega_s\}$ is a common transversal of $F_1,F_2,\dots,F_{t-1}$. Now we let $\mathcal{D}$ be some fixed collection of \emph{mutually disjoint} diagonals of $X$ whose union is $X$, $X = \bigcup \mathcal{D}$. (The existence of $\mathcal{D}$ can be shown by induction on $t$.) Since $|X| = s^{t-1}$ and every diagonal has $s$ elements, we have $|\mathcal{D}| = s^{t-2}$. For any diagonal $D$, let $f(D) = \bigcup \{f(x): x \in D\}$. Then \[ \Omega = \bigcup \{f(x): x \in X\} = \bigcup \{f(D): D \in \mathcal{D}\}. \] Now \[ s^t - s^{t-1} < |\Omega| \leq \sum_{D \in \mathcal{D}} |f(D)| \leq |\mathcal{D}| \max\{|f(D)| : D \in \mathcal{D}\}, \] hence $s^2 - s < \max\{|f(D)|: D \in \mathcal{D}\}$. We now fix a diagonal $D$ with $s^2 - s < |f(D)|$. Let $D = \{x_1,x_2,\dots,x_s\}$, and define, for $ 1\leq i \leq s$, $1 \leq j \leq s$, \[ e_{ij} = \begin{cases} 1 &\textup{if $f(x_i) \cap G(j) \ne \emptyset$} \\ 0 &\textup{otherwise}. \end{cases} \] Then since $|f(x_i) \cap G(j)| \leq 1$ for all $i,j$, and $G$ is a partition of $\Omega$, the $s \times s$ 0-1 matrix $(e_{ij})$ contains exactly $|f(D)|$, and hence more than $s^2 - s$, 1's. Therefore there exist (as can be shown by induction on $s$) indices $i_1j_1, i_2j_2, \dots, i_sj_s$ such that $e_{i_1j_1} = e_{i_2j_2} = \cdots = e_{i_sj_s} = 1$ and $\{i_1,i_2,\dots,i_s\} = \{j_1,j_2,\dots,j_s\} = \{1,2,\dots,s\}$. Now let $\{\omega_k\} = f(x_{i_k}) \cap G(j_k)$, $1 \leq k \leq s$. Then since $G$ is a partition, $\omega_1,\omega_2,\dots,\omega_s$ are all distinct, and hence $\{\omega_1,\omega_2,\dots,\omega_s\}$ is not only a transversal of $G$ but is also (since $\{x_1,x_2,\dots,x_s\}$ is a diagonal) a common transversal of $F_1,F_2,\dots,F_{t-1}$. Therefore $\{\omega_1,\omega_2,\dots,\omega_s\}$ is a common transversal of $F_1,F_2,\dots,F_{t-1}, G$, and hence of $F_1,F_2,\dots,F_t$. This completes the proof of Theorem~\ref{thm: 1}(b). \end{proof} \begin{proof}[Proof of Theorem~\ref{thm: 2}(a)] Recall that for each $j$, $1 \leq j \leq t$, $F_j(0)$ denotes the complement in $\Omega$ of the union of the family $F_j$, that is, $F_j(0) = \Omega \setminus \bigcup \{F_j(i): 1 \leq i \leq s\}$. If now for each $j$, $1 \leq j \leq t$, we let \[ G_j = (F_j(0), F_j(1), \dots, F_j(s)), \] and let $\mathcal{G} = (G_1,G_2,\dots,G_t)$, then $\mathcal{G}$ separates points and each family $G_j$ covers $\Omega$, therefore we may proceed exactly as in the proof of Theorem~\ref{thm: 1}(a), where now we have $t$ families with $s + 1$ sets in each family. Furthermore, since we know that $\bigcap \{F_j(0): 1 \leq j \leq t\} = \emptyset$ (this is so because $\bigcup \{F_j(i): 1 \leq j \leq t, 1 \leq i \leq s\} = \Omega$), the last inequality in the proof of Theorem~\ref{thm: 1}(a) can be replaced by \begin{align*} |\Omega| &\leq |\{(x,i): x \in X, 0 \leq i \leq s, i \ne i_k, (i,x) \ne (0,(0,0,\dots,0))\}| \\ &= (s + 1)^t - (s+1)^{t-1} - 1. \end{align*} This proves Theorem~\ref{thm: 2}(a). \end{proof} \begin{proof}[Proof of Theorem~\ref{thm: 2}(b)] For each $j$, $1 \leq j \leq t$, let $\Omega_j = \bigcup F_j$, and let \[ \Omega_0 = \bigcap \{\Omega_j: 1\leq j \leq t\}. \] For each $j$, $1 \leq j \leq t$, $1 \leq i \leq s$, let $G_j(i) = F_j(i) \cap \Omega_0$, $G_j = (G_j(1),G_j(2),\dots,G_j(s))$, and \[ \mathcal{G} = (G_1,G_2,\dots,G_t). \] Then for each $j$, $1 \leq j \leq t$, $\Omega_0 = \bigcup G_j$. Also, since $G_j(i) \subset F_j(i)$, for all $j,i,$ the family $\mathcal{G}$ separates points. Thus, by Theorem~\ref{thm: 1}(b), it suffices now to show that $|\Omega_0| > s^t - s^{t-1}$, since any common transversal of $G_1,G_2,\dots,G_t$ is also a common transversal of $F_1,F_2,\dots,F_t$. Since $\mathcal{F}$ separates points, the cardinal of $\Omega \setminus \Omega_0$ cannot exceed the cardinal of the set of all those $t$-tuples $(a_1,a_2,\dots,a_t)$, $0 \leq a_j \leq s$, having at least one coordinate equal to $0$ (excluding $(0,0,\dots,0)$). That is, $|\Omega \setminus \Omega_0| \leq (s + 1)^t - s^t - 1$. Hence \begin{align*} (s + 1)^t - s^{t-1} - 1 &< |\Omega| = |\Omega_0| + |\Omega\setminus \Omega_0| \\ &\leq |\Omega_0| + (s + 1)^t - s^t - 1, \end{align*} and therefore \[ |\Omega_0| > s^t - s^{t-1}. \] This completes the proof of Theorem~\ref{thm: 2}(b). \end{proof} \bibliographystyle{amsplain} \bibliography{tom-all} \end{document}