Maths Olympiad Prep

Library / /40 of 45

Algebra Difficulty 9.1 IMO level Prove it United States

Let nn be a positive integer and let a1a2ana_1 \le a_2 \le \dots \le a_n and b1b2bnb_1 \le b_2 \le \dots \le b_n be two nondecreasing sequences of real numbers such that
a1++aib1++bifor every i=1,,n1 a_1 + \dots + a_i \le b_1 + \dots + b_i \quad \text{for every } i = 1, \dots, n-1
and
a1++an=b1++bn. a_1 + \dots + a_n = b_1 + \dots + b_n.

Suppose that for any real number mm, the number of pairs (i,j)(i, j) with aiaj=ma_i - a_j = m equals the number of pairs (k,)(k, \ell) with bkb=mb_k - b_\ell = m. Prove that ai=bia_i = b_i for i=1,,ni = 1, \dots, n.

Note: It is important to interpret the condition that for any real number mm, the number of pairs (i,j)(i, j) with aiaj=ma_i - a_j = m equals the number of pairs (k,)(k, \ell) with bkb=mb_k - b_\ell = m. It means that we have two identical multi-sets (a multi-set is a set that allows repeated elements)

Solution

First Solution: Put sn=a1++an=b1++bns_n = a_1 + \cdots + a_n = b_1 + \cdots + b_n. Then
2i=1n1(a1++ai)=2(n1)a1+2(n2)a2++2(1)an1=(n1)a1+(n3)a2++(1n)an+(n1)sn=(n1)sn+1i<jn(aiaj) \begin{aligned} 2 \sum_{i=1}^{n-1} (a_1 + \cdots + a_i) &= 2(n-1)a_1 + 2(n-2)a_2 + \cdots + 2(1)a_{n-1} \\ &= (n-1)a_1 + (n-3)a_2 + \cdots + (1-n)a_n + (n-1)s_n \\ &= (n-1)s_n + \sum_{1 \le i < j \le n} (a_i - a_j) \end{aligned}
and similarly
2i=1n1(b1++bi)=(n1)sn+1k<n(bkb). 2 \sum_{i=1}^{n-1} (b_1 + \cdots + b_i) = (n-1)s_n + \sum_{1 \le k < \ell \le n} (b_k - b_\ell).
By (*), these two quantities are equal, so
2i=1n1(a1++ai)=2i=1n1(b1++bi). 2 \sum_{i=1}^{n-1} (a_1 + \cdots + a_i) = 2 \sum_{i=1}^{n-1} (b_1 + \cdots + b_i).
Consequently, each of the inequalities a1++aib1++bia_1 + \cdots + a_i \le b_1 + \cdots + b_i for i=1,,n1i = 1, \dots, n-1 must be an equality. Since we also have equality for i=ni = n by assumption, we deduce that ai=bia_i = b_i for i=1,,ni = 1, \dots, n, as desired.

Second Solution: Expanding both sides of (**) yields
(n1)i=1nai2+21i<jnaiaj=(n1)i=1nbi2+21k<nbkb. (n-1) \sum_{i=1}^{n} a_i^2 + 2 \sum_{1 \le i < j \le n} a_i a_j = (n-1) \sum_{i=1}^{n} b_i^2 + 2 \sum_{1 \le k < \ell \le n} b_k b_\ell.
Squaring both sides of the given equation a1++an=b1++bna_1 + \cdots + a_n = b_1 + \cdots + b_n gives
i=1nai2+21i<jnaiaj=i=1nbi2+21k<nbkb. \sum_{i=1}^{n} a_i^2 + 2 \sum_{1 \le i < j \le n} a_i a_j = \sum_{i=1}^{n} b_i^2 + 2 \sum_{1 \le k < \ell \le n} b_k b_\ell.
From the above relations we easily deduce that
i=1nai2=i=1nbi2. \sum_{i=1}^{n} a_i^2 = \sum_{i=1}^{n} b_i^2.
By the Cauchy-Schwarz inequality, we obtain that
(i=1nbi2)2=(i=1nai2)(i=1nbi2)(i=1naibi)2 \left(\sum_{i=1}^{n} b_i^2\right)^2 = \left(\sum_{i=1}^{n} a_i^2\right) \left(\sum_{i=1}^{n} b_i^2\right) \ge \left(\sum_{i=1}^{n} a_i b_i\right)^2
or
i=1nbi2i=1naibii=1naibi.() \sum_{i=1}^{n} b_i^2 \geq \left| \sum_{i=1}^{n} a_i b_i \right| \geq \sum_{i=1}^{n} a_i b_i. \qquad (\dagger)
We set si=a1++ais_i = a_1 + \dots + a_i and ti=b1++bit_i = b_1 + \dots + b_i for every 1in1 \le i \le n. By Abel's summation formula, we have
i=1naibi=s1b1+[s2s1]b2+[s3s2]b3++[snsn1]bn=s1(b1b2)+s2(b2b3)++sn1(bn1bn)+snbn. \begin{aligned} \sum_{i=1}^{n} a_i b_i &= s_1 b_1 + [s_2 - s_1] b_2 + [s_3 - s_2] b_3 + \dots + [s_n - s_{n-1}] b_n \\ &= s_1(b_1 - b_2) + s_2(b_2 - b_3) + \dots + s_{n-1}(b_{n-1} - b_n) + s_n b_n. \end{aligned}
By the given conditions, sitis_i \le t_i and bibi+10b_i - b_{i+1} \le 0 for every 1in11 \le i \le n-1 and sn=tns_n = t_n. It follows that
i=1naibit1(b1b2)+t2(b2b3)++tn1(bn1bn)+tnbn=t1b1+[t2t1]b2+[t3t2]b3++[tntn1]bn=i=1nbi2. \begin{aligned} \sum_{i=1}^{n} a_i b_i &\ge t_1(b_1 - b_2) + t_2(b_2 - b_3) + \dots + t_{n-1}(b_{n-1} - b_n) + t_n b_n \\ &= t_1 b_1 + [t_2 - t_1] b_2 + [t_3 - t_2] b_3 + \dots + [t_n - t_{n-1}] b_n = \sum_{i=1}^{n} b_i^2. \end{aligned}
Combining the last inequality and (),(\dagger), we conclude that the equality case holds for every inequality we discussed above. In particular, si=tis_i = t_i for i=1,,ni = 1, \dots, n. These inequalities immediately give us an=bna_n = b_n, an1=bn1a_{n-1} = b_{n-1}, \dots, a1=b1a_1 = b_1 and the problem is solved.

Third Solution: If u=(u1,u2,,un)u = (u_1, u_2, \dots, u_n) and v=(v1,v2,,vn)v = (v_1, v_2, \dots, v_n) are two nonincreasing sequences, we say that uu majorizes vv if u1++un=v1++vnu_1 + \dots + u_n = v_1 + \dots + v_n and u1++uiv1++viu_1 + \dots + u_i \ge v_1 + \dots + v_i for i=1,2,,n1i = 1, 2, \dots, n-1. It is not difficult to see that (an,,a1)(a_n, \dots, a_1) majorizes (bn,,b1)(b_n, \dots, b_1). By a theorem of Birkhoff, it follows that there are constants cσ(0,1]c_\sigma \in (0, 1], where σ\sigma runs over some set SS of permutations of {1,,n}\{1, \dots, n\}, with σScσ=1\sum_{\sigma \in S} c_\sigma = 1 and
σScσaσ(i)=bifor i=1,2,,n. \sum_{\sigma \in S} c_{\sigma} a_{\sigma(i)} = b_i \quad \text{for } i = 1, 2, \dots, n.
We will prove the inequality
i,j=1naiaji,j=1nbibj, \sum_{i,j=1}^{n} |a_i - a_j| \ge \sum_{i,j=1}^{n} |b_i - b_j|,
with equality if and only if ai=bia_i = b_i for i=1,,ni = 1, \dots, n. With this result, we complete our proof by noting (***).
We have
i,j=1naiaj=σScσi,j=1naiaj=σScσi,j=1naσ(i)aσ(j)=i,j=1nσScσaσ(i)aσ(j)i,j=1nσScσ(aσ(i)aσ(j))=i,j=1nbibj, \begin{aligned} \sum_{i,j=1}^{n} |a_i - a_j| &= \sum_{\sigma \in S} c_{\sigma} \sum_{i,j=1}^{n} |a_i - a_j| = \sum_{\sigma \in S} c_{\sigma} \sum_{i,j=1}^{n} |a_{\sigma(i)} - a_{\sigma(j)}| \\ &= \sum_{i,j=1}^{n} \sum_{\sigma \in S} c_{\sigma} |a_{\sigma(i)} - a_{\sigma(j)}| \ge \sum_{i,j=1}^{n} \left| \sum_{\sigma \in S} c_{\sigma} (a_{\sigma(i)} - a_{\sigma(j)}) \right| = \sum_{i,j=1}^{n} |b_i - b_j|, \end{aligned}
using the fact that x1++xmx1++xm|x_1| + \dots + |x_m| \ge |x_1| + \dots + |x_m| for all real numbers x1,,xmx_1, \dots, x_m.
This establishes the desired inequality; it remains to check the equality condition. For this, we must have
σScσaσ(i)aσ(j)=σScσ(aσ(i)aσ(j)) \sum_{\sigma \in S} c_{\sigma} |a_{\sigma(i)} - a_{\sigma(j)}| = \left| \sum_{\sigma \in S} c_{\sigma} (a_{\sigma(i)} - a_{\sigma(j)}) \right|
for each pair ii and jj; in particular, for each pair ii and jj, the sign of aσ(i)aσ(j)a_{\sigma(i)} - a_{\sigma(j)} must be the same for all σS\sigma \in S for which aσ(i)aσ(j)a_{\sigma(i)} \neq a_{\sigma(j)}. It follows by the lemma below that the sequence aσ(1),,aσ(n)a_{\sigma(1)}, \dots, a_{\sigma(n)} itself must be the same for all σS\sigma \in S, yielding ai=bia_i = b_i.

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.