Let m be a positive integer and A be a finite set. Let A1,A2,…,Am be subsets of A (not necessarily distinct). It is known that for any nonempty set I⊆{1,2,…,m}, i∈I⋃Ai≥∣I∣+1.
Prove: the elements of A can be coloured black or white, such that every of A1,A2,…,Am contains both black and white elements.
Solution
We give three solutions as follows.
Solution 1
Construct a bipartite graph G whose two parts are X={A1,A2,…,Am} and Y=A: for 1≤i≤m and a∈A, Ai∈X and a∈Y are adjacent if and only if a∈Ai. Since for each nonempty I⊆{1,2,…,m}, ∣⋃i∈IAi∣≥∣I∣+1, it follows that for any k vertices of X, the number of vertices of Y that are neighbouring to one or more of them, is at least k+1. According to Hall's theorem, there exists a transversal f from X to Y. For 1≤i≤m, let ai=f(Ai). Clearly, ai∈Ai, and a1,a2,…,am are distinct elements of A. Colour all the other elements A∖{a1,…,am} white, and determine the colours of a1,a2,…,am as follows.
Every time, choose an i such that ai is uncoloured and Ai has a coloured neighbour, say b. Colour ai the opposite colour to b. Suppose, during the process, some elements say a1,a2,…,ak (1≤k≤m) are uncoloured, but we cannot find another i and colour ai. Based on the algorithm, this means that all the neighbours of A1,A2,…,Ak are in {a1,…,ak}, yet it contradicts ∣A1∪A2∪⋯∪Ak∣≥k+1. Hence, this process can continue until all of a1,a2,…,am are coloured. Moreover, for 1≤i≤m, when ai is coloured, Ai is guaranteed to have black and white neighbours, that is, Ai contains both black and white elements. This colouring satisfies the problem requirements. □
Solution 2
We begin with a lemma.
Lemma The finite sets A1,A2,…,Am are called "nice", if the union of any k≤m (k is arbitrary) of them contains at least k+1 elements. If A1,A2,…,Am are nice, then there exist 2-element sets B1,B2,…,Bm, such that Bi⊆Ai, i=1,2,…,m, and B1,B2,…,Bm are nice.
Proof of lemma Assume the lemma is untrue. Let A1,A2,…,Am be a counterexample with the smallest m value and the smallest ∣A1∣+∣A2∣+⋯+∣Am∣ for such m. There are three situations.
Case 1, if the union of any k≤m (k is arbitrary) sets does not contain exactly k+1 elements. Take k=1, and apparently every set contains 3 or more elements. Choose an arbitrary set and remove any element from it. The sets are still nice (as a counterexample), yet ∣A1∣+∣A2∣+⋯+∣Am∣ is smaller. A contradiction.
Case 2, if the union of some k≤m−1 sets contains exactly k+1 elements. Assume ∣A1∪A2∪⋯∪Ak∣=k+1. Since k<m, there exist 2-element sets B1,B2,…,Bk such that Bi⊆Ai, i=1,2,…,k, and B1,B2,…,Bk are nice.
Let Ci:=Ai∖(A1∪A2∪⋯∪Ak), i=k+1,…,m. It is easy to see that, among Ck+1,Ck+2,…,Cm, the union of any d (1≤d≤m−k, d is arbitrary) sets contains d or more elements. By Hall's theorem, there exist xi∈Ci (i=k+1,…,m), such that xk+1,xk+2,…,xm are all distinct. We construct Bk+1,Bk+2,…,Bm in the following way: if for some i∈{k+1,k+2,…,m}, Bi has not been made yet, but Ai contains an element yi that belongs to some B set already constructed, then let Bi={xi,yi}. Clearly, xi is a new element in the constructed B sets, hence these B sets are still nice. Suppose, after the construction of several Bi's, no more set can be constructed in the above way. Then the remaining A sets do not contain any element in the constructed Bi's. Since m is minimal, we may construct Bj's from those A sets such that they are nice. Furthermore, Bj's and Bi's are disjoint, when combined, B1,B2,…,Bm are nice. So, case 2 is not possible.
Case 3, if the union of any k (1≤k≤m−1, k is arbitrary) sets contains k+2 or more elements, while the union of all m sets contains exactly m+1 elements. Observe that any m−1 sets have the same union as that of all m sets. Hence, every element must belong to at least 2 sets. Meanwhile, each set contains at least 3 elements. Choose any set and remove any element from it. The sets are still nice, but they have a smaller ∣A1∣+∣A2∣+⋯+∣Am∣, contradiction.
This verifies the lemma.
Return to the original problem. According to the lemma, we may find 2-element subsets B1,B2,…,Bm of A1,A2,…,Am, respectively, such that the union of any k of them has k+1 or more elements. Treat every element of A as a vertex, and then for every 1≤i≤m, connect the two elements in Bi by an edge. Note that this graph has no cycle (otherwise, the edges in a cycle correspond to B sets the number of which is the same as the number of elements in their union), and thus it is a forest, which is a bipartite graph. Colour all elements in one part black, and all elements in the other part white. This colouring satisfies the problem condition. □
Solution 3
Use induction on the number of sets m. When m=1, ∣A1∣≥2, choose two elements of A1 and colour one black and one white. Suppose the conclusion holds for all m≤n−1. Consider m=n.
Case 1. If for any I⊆{1,…,n}, ∣∪i∈IAi∣≥∣I∣+2. Take a∈An. By the induction hypothesis, there is a colouring of A1∖{a},A2∖{a},…,An−1∖{a} for which each Ai∖{a} contains black and white elements. As An∖{a} has been coloured, we can colour a such that An has elements of both colours.
Case 2. If for some I={1,2,…,n}, ∣∪i∈IAi∣=∣I∣+1. Let I be the largest, J={1,2,…,n}∖I, and B=∪i∈IAi, C=∪j∈JAj.
Case 2a. If B∩C=∅, then colour B and C separately. By induction, there is a colouring that guarantees each Ai contains black and white elements.
Case 2b. If B∩C=∅, take b∈B∩C, and define Aj′=Aj∖(B∖{b}) (j∈J). For any T⊆J, since I is the largest, it follows that t∈T⋃At′≥t∈I∪T⋃At−i∈I⋃Ai≥(∣I∪T∣+2)−(∣I∣+1)=∣T∣+1. Similarly, j∈J⋃Aj′=i=1⋃Ai−∣B∖{b}∣≥n+1−∣I∣=∣J∣+1. By induction, we may colour C∖(B∖{b})=∪j∈JAj′ and B=∪i∈IAi such that Aj′(j∈J) and Ai(i∈I) contain elements of both colours. If b has different colours in the two colourings, then reverse the colouring of B. Now they are compatible and give a colouring of A which meets the problem requirements. □
Want a route through all this instead of an archive? The track
puts 2,000 problems in a working order, from AMC 10 level to the IMO shortlist.
Source: MathNet,
licensed CC-BY-4.0.
Statement reproduced verbatim; metadata (topic, difficulty) added by this project.