For a positive integer n, and a non-empty subset A of {1,2,…,2n}, we call A good if the set {u±v∣u,v∈A} does not contain the set {1,2,…,n}. We aim to find the smallest real number c such that for any positive integer n, and any good subset A of {1,2,…,2n}, ∣A∣≤cn.
We will prove that the smallest constant is c=56.
First, let us show that c≥56. Consider n=10q+1 for q∈N0 and set k=54n+1. Observe that k is an odd positive integer with 1≤k≤n. Now, consider the set A=A1∪A2∪A3 where
A1={1,2,…,2k−1},A2={2k+1,2k+2,…,2k+2k−1},A3={k+2k+1,k+2k+3,…,2k}.
It is clear that A is a subset of {1,2,…,2n} with
∣A∣=2k−1+(k−2k−1)+2k−1=56n−51.
Hence, if we take the limit as q→∞, it follows that ∣A∣>ϵn for any ϵ<56. Therefore, to show that c≥56, it suffices to prove that A is good. In particular, we will show that the set B={u±v∣u,v∈A} does not contain the integer k.
First, it is clear that if u,v∈A satisfy u+v=k, then u,v∈A1 (since all elements of A2 and A3 are greater than k). However, this is impossible, since the greatest possible sum of two elements of A1 is 2k−1+2k−3<k. Meanwhile, if u,v∈A satisfy u−v=k, we must have u≡v(modk). By breaking up A into subsets modulo k, we find that
A={1,2k+1}∪{2,2k+2}∪⋯∪{2k−1,2k+2k−1}∪{k+2k+1}∪{k+2k+3}∪⋯∪{2k}.
It is then easy to see that no u,v∈A satisfy u−v=k. Hence, c≥56, as desired. ■
Now, we will show that ∣A∣≤56n for any good set A⊆{1,2,…,2n}. Suppose, by way of contradiction, that there exists a good set A⊆{1,2,…,2n} with ∣A∣>56n. Then there must exist some integer k∈{1,2,…,n} such that k∈B, where B={u±v∣u,v∈A}.
By the Division Algorithm, let us write 2n=mk+p where m∈N and 0≤p<k. In particular, notice that
2n=mk+p<(m+1)k≤(m+1)n⟹2<m+1⟹2≤m.
Now, consider the sets Si={b∈B∣b≡i(modk)} (i=1,2,…,k). We will examine these sets in pairs: (S1,Sk−1),(S2,Sk−2),…. First, observe that the only sets that are not part of a pair are Sk and Sk/2 (if k is even).
We begin by proving that at most 2m+1m+1 of the elements in Sk∪Sk/2 are in B (if k is odd, simply ignore the set Sk/2 in the following analysis; the same conclusion still holds). Observe that Sk has precisely m elements, and Sk/2 has either m or m+1 elements. Within each of these sets, no two consecutive elements can both be in B, since then the difference of these two consecutive elements would equal k, a contradiction. Hence, at most ⌈2m⌉ of the elements in Sk are in B, and at most ⌈2m+1⌉ of the elements in Sk/2 are in B. It is then easy to see that at most 2m+1m+1 of the elements in Sk∪Sk/2 are in B.
Now, we prove a similar bound for the pairs of sets described earlier: Consider any pair (Si,Sk−i). Notice that at most 21 of the elements of one of these sets can be in B. This is because if more than 21 of the elements of each of these sets are in B, then because no two consecutive elements in either of these sets can be in B, it would follow that i∈Si and k−i∈Sk−i must be in B. However, this is impossible, since then the sum of these two elements would equal k, a contradiction. Therefore, at most 21 of the elements in one of these two sets must be in B. Keeping in mind that ∣Si∣=m,m+1, it’s not hard to see that at most 2m+1m+1 of the elements in Si∪Sk−i are in B.
Therefore, since B⊆S1∪S2∪⋯∪Sk, it follows that
∣B∣≤2m+1m+1∣S1∪S2∪⋯∪Sk∣=2m+1m+1(2n).
Because 2m+1m+1=21+4m+21 is a decreasing function of n over N, it follows that 2m+1m+1 takes on its maximal value for m=2. Hence,
∣B∣≤4+12+1(2n)=56n.
This is a clear contradiction, since we assumed that ∣B∣>56n. Thus, the proof is complete. □
The answer is 56.