Maths Olympiad Prep

Library / /33 of 106

, 2022

Combinatorics Difficulty 7.7 National Olympiad, round 2 Prove it China

Let nn be a positive integer. Let x1,x2,,x2nx_1, x_2, \dots, x_{2n} be 2n2n nonnegative real numbers such that x1+x2++x2n=4x_1 + x_2 + \dots + x_{2n} = 4. Prove that there exist nonnegative integers p,qp, q such that qn1q \le n - 1, and that
i=1qxp+2i11,i=q+1n1xp+2i1. \sum_{i=1}^{q} x_{p+2i-1} \le 1, \quad \sum_{i=q+1}^{n-1} x_{p+2i} \le 1.

*Remark 1 : the subscripts are understood modulo 2n2n, i.e. if k(mod2n)k \equiv \ell \pmod{2n}, then xk=xx_k = x_\ell.*
*Remark 2 : If q=0q = 0, then the first sum is considered as 0; if q=n1q = n - 1, then the second sum is considered as 0.*

Solution

Proof: Set A=x1+x3++x2n1A = x_1 + x_3 + \dots + x_{2n-1} and B=x2+x4++x2nB = x_2 + x_4 + \dots + x_{2n}.
If one of A,BA, B is less than or equal to 1, the problem is obvious. If A>1A > 1 and B>1B > 1, for 0kn10 \le k \le n-1, let m(k){1,2,,n1}m(k) \in \{1, 2, \dots, n-1\} be the unique integer such that
i=0m(k)x2k+2i+1>1(1) \sum_{i=0}^{m(k)} x_{2k+2i+1} > 1 \quad (1)
i=0m(k)1x2k+2i+11.(2) \sum_{i=0}^{m(k)-1} x_{2k+2i+1} \le 1. \quad (2)
Note that, if x2k+2m(k)+2+x2k+2m(k)+4++x2k+2n21x_{2k+2m(k)+2} + x_{2k+2m(k)+4} + \dots + x_{2k+2n-2} \le 1, then this equation and (2) gives what we needed with p=2kp = 2k and q=m(k)q = m(k).
Now suppose that
x2k+2m(k)+2+x2k+2m(k)+4++x2k+2n2>1.(3) x_{2k+2m(k)+2} + x_{2k+2m(k)+4} + \dots + x_{2k+2n-2} > 1. \quad (3)
Now, we construct an oriented graph with 0,1,2,,n10, 1, 2, \dots, n-1 as vertices, and connect an edge from each kk to k+m(k)+1k+m(k)+1 (with indices modulo nn). It is easy to see that there are loops in this graph. We may assume that k1k2ktk1k_1 \to k_2 \to \dots \to k_t \to k_1 is a minimal loop.
If t=1t=1, then i=0n2x2k1+2i+11\sum_{i=0}^{n-2} x_{2k_1+2i+1} \le 1. Setting p=2k1p=2k_1 and q=n1q=n-1 gives what we want.
If t>1t>1, then set
{k2k1n}+{k3k2n}++{ktkt1n}+{k1ktn}=s. \left\{ \frac{k_2 - k_1}{n} \right\} + \left\{ \frac{k_3 - k_2}{n} \right\} + \dots + \left\{ \frac{k_t - k_{t-1}}{n} \right\} + \left\{ \frac{k_1 - k_t}{n} \right\} = s.
In other words, when k=k1,k2,,ktk = k_1, k_2, \dots, k_t, there are totally sns \cdot n elements in the sums of (1). Moreover, using the condition for defining the edges, we know that every term in x1,x3,,x2n1x_1, x_3, \dots, x_{2n-1} appeared in the sum (1) exactly ss times. On the other hand, when k=k1,k2,,ktk = k_1, k_2, \dots, k_t, there are in total (ts)n(t-s) \cdot n elements in the sums in (3). The condition for linking an edge shows that each of x2,x4,,x2nx_2, x_4, \dots, x_{2n} appeared exactly tst-s times in the above sum; this is because x2jx_{2j} appeared in (3) if and only if x2j+1x_{2j+1} did not appear in (1).
Now take the total sum of (1) when k=k1,k2,,ktk = k_1, k_2, \dots, k_t, we get sA>ts \cdot A > t. Similarly, taking the total sum of (3) when k=k1,k2,,ktk = k_1, k_2, \dots, k_t gives (ts)B>t(t-s) \cdot B > t. But A+B>t(1s+1ts)4A + B > t(\frac{1}{s} + \frac{1}{t-s}) \ge 4. This leads to a contradiction. The problem is proved.

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.