Since the polynomials P1,…,Pn do not have integer roots then Pi(0)=0 for any i=1,n and Pi(0)=±1. Let us consider all pairs of polynomials Pk and Pj, such that k=j and Pk(0)=Pj(0). For each pair of polynomials we determine Qk,j=Pk−Pj. If a and b are the number of polynomials such that Pi(0)=1 and Pi(0)=−1 respectively, then the number of polynomials Qk,j will be
2Ca2+2Cb2=a(a−1)+b(b−1)=a2+b2−n≥21(a+b)2−n=21n(n−2).
Note that coefficients of polynomials Qk,j may be only equal to −2,−1,0,1,2. Hence, the coefficients of Qk,j−Qi,l cannot be bigger than 4 by absolute value. Therefore Qk,j(5)=Qi,l(5) if and only if Qk,j and Qi,l are the same, because ∀n∈N, 5n>4(5n−1+⋯+5+1).
It is easy to see that Qk,j(5) is divisible by 5. Moreover, Qk,j(5)=0 (otherwise Qk,j is the zero polynomial which is impossible) and ∣Qk,j(5)∣≤∣Pi(5)∣+∣Pj(5)∣≤n2. Therefore, Qk,j(5) can take at most 52n2 different values. But the overall number of polynomials Qk,j is bigger than 52n2, because
21n(n−2)>52n2⇔5(n−2)>4n⇔n>10.
This implies that Qk,j(5)=Qi,l(5)⇒Qk,j=Qi,l for some i,j,k,l. Consequently, Pi+Pj=Pk+Pl for some 1≤i,j,k,l≤n, {i,j}={k,l} which finishes the proof.