MathLabs

Problem 3

For an integer-coefficient polynomial P(x)=a0+a1x+⋯+akxkP(x)=a_0+a_1x+\cdots+a_kx^k, let w(P)w(P) be the number of odd coefficients. Put Qi(x)=(1+x)iQ_i(x)=(1+x)^i for i=0,1,2,…i=0,1,2,\ldots. Prove that if 0≤i1<i2<⋯<in0\le i_1<i_2<\cdots<i_n, then w(Qi1+Qi2+⋯+Qin)≥w(Qi1)w(Q_{i_1}+Q_{i_2}+\cdots+Q_{i_n})\ge w(Q_{i_1}).
Step 2 of 4: Shift every index when the smallest index is high
In plain words

When every index is at least mm, remove the same power-of-two layer from every term; the induction problem is the same picture at a smaller scale.

Qi1+⋯+Qin=(1+x)m(B)(i1≥m)Q_{i_1}+\cdots+Q_{i_n}=(1+x)^m(B)\quad(i_1\ge m)
Detailed analysis

Use strong induction on ini_n. Let mm be the largest power of two with m≤in<2mm\le i_n<2m. If i1≥mi_1\ge m, write Qij=(1+x)mQij−mQ_{i_j}=(1+x)^mQ_{i_j-m} and let B=∑jQij−mB=\sum_jQ_{i_j-m}. All shifted indices have largest index less than ini_n, so induction gives w(Qi1−m)≤w(B)w(Q_{i_1-m})\le w(B). Step 1 doubles both sides: w(Qi1)=2w(Qi1−m)≤2w(B)=w(∑jQij)w(Q_{i_1})=2w(Q_{i_1-m})\le2w(B)=w(\sum_jQ_{i_j}).