Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated


Source. Hall (1935), Theorem 3, printed pp. 29–30 (canonical PDF). It is the source's specialization of Theorem 2.

Statement. Let m≥0m\ge0 be an integer, and let (Pi)i∈[m](P_i)_{i\in[m]} and (Qj)j∈[m](Q_j)_{j\in[m]} be two partitions of the same set SS into mm classes. A set R⊆SR\subseteq S can contain exactly one point from each PiP_i and exactly one point from each QjQ_j if and only if

∣{i∈[m]:Pi∩⋃j∈JQj≠∅}∣≥∣J∣for every J⊆[m].(1)\left|\left\{i\in[m]: P_i\cap\bigcup_{j\in J}Q_j\ne\varnothing\right\}\right| \ge |J|\qquad\text{for every }J\subseteq[m]. \tag{1}

Such an RR has mm elements. Equivalently, there is a permutation σ\sigma of [m][m] and pairwise distinct points aj∈Qj∩Pσ(j)a_j\in Q_j\cap P_{\sigma(j)}. The classes and the ambient set may be infinite. The criterion is also equivalent to the condition obtained by interchanging the two partitions.

Proof. Suppose first that RR is a common representative set. For each j∈Jj\in J let aja_j be its unique point in QjQ_j. These ∣J∣|J| points lie in distinct PiP_i classes, since RR contains exactly one point from each such class. All those classes meet ⋃j∈JQj\bigcup_{j\in J}Q_j, so (1) follows.

Conversely, apply Theorem 2 to the family Tj=QjT_j=Q_j and the partition {Pi:i∈[m]}\{P_i:i\in[m]\}. Condition (1) supplies points aj∈Qja_j\in Q_j in pairwise distinct PiP_i classes. There are mm selected classes and only mm classes in that partition, so every PiP_i is selected. The set R={aj:j∈[m]}R=\{a_j:j\in[m]\} therefore contains exactly one point in each PiP_i. It also contains exactly one in each QjQ_j, because these classes are pairwise disjoint and there is one selected point for each index jj.

Writing σ(j)\sigma(j) for the index of the selected PP class gives an injective map from [m][m] to itself. Finiteness makes it a permutation, proving the equivalent indexed formulation. Interchanging PP and QQ in the necessity and sufficiency arguments proves the asserted symmetry. If m=0m=0, both partitions are empty and hence S=∅S=\varnothing; the empty set and empty permutation give all the conclusions. □\square

Source precision. Hall numbers the two partitions by SiS_i and Si′S'_i, and states the condition on the primed classes. Taking Pi=SiP_i=S_i and Qj=Sj′Q_j=S'_j gives exactly (1). The original describes permuting the primed suffixes; this is equivalent to the permutation form above by inversion and reindexing. Necessity, symmetry and the empty-family endpoint are made explicit here. No assumption that the two partitions are unequal is required: the identical-partition case satisfies the same argument.

Used by. The equal finite block corollary. The further Rado (1933) remark on p. 30 remains only the historical pointer described in the scope record.

Bears on. No problem. The Problem 126 argument cites only Theorem 1, not this partition form.