Without loss of generality, we can assume that a1≤a2≤⋯≤a2n. We distinguish two cases:
- If an+a2n−1≥0 then we have ai+a2n−1≥0 for i=n,⋯,2n−2, and ai+a2n≥0 for i=n⋯,2n−1. This provides at least 2n−1 non-negative sums.
- If an+a2n−1<0
0>an+a2n−1≥an−1+a2n−2≥⋯≥a2+an+1,
then
a2+a3+⋯+an−1+an+1+⋯+a2n−2<0.
Let ℓ be the number of non-negative terms among a1,…,a2n. Then there are at least 2ℓ(ℓ−1)≥2n(n+1) pairs (ai,aj) with i<j,ai⩾0 and aj⩾0. Since 2n(n+1)−(2n−1)=2n2−3n+2=2(n−1)(n−2)⩾0, there are at least 2n−1 pairs (ai,aj) with i<j such that ai+aj⩾0.
First case: ℓ>n. Then there are at least 2ℓ(ℓ−1)≥2n(n+1) pairs (ai,aj) with i<j,ai⩾0 and aj⩾0. Since 2n(n+1)−(2n−1)=2n2−3n+2=2(n−1)(n−2)⩾0, there are at least 2n−1 pairs (ai,aj) with i<j such that ai+aj⩾0.
Second case: ℓ⩽n. Let c1⩽⋯⩽cℓ be the smallest integers among a1,…,a2n. Since ℓ⩽n, we have cℓ<0. Moreover, ∑i=12nai is equal to the sum of ∑i=1ℓ(bi+ci) and negative terms, so ∑i=1ℓ(bi+ci)⩾0. Since bℓ+cℓ⩾bi+ci for all i, we have bℓ+cℓ⩾0.
Thus, we already have 2n−ℓ pairs (ai,aj) by taking aj=bℓ and ai other than c1,…,cℓ−1,bℓ.
Furthermore, for all k=1,…,ℓ−1, we have ∑i=1ℓ(bi+ci+k)⩾0 (where by convention cℓ+1=c1, cℓ+2=c2, etc.), so for all k there exists i such that bi+ci+k⩾0. This provides another ℓ−1 pairs.