Maths Olympiad Prep

Library / IMO Shortlist /363 of 365

IMO Shortlist, 2016

Combinatorics Difficulty 9.2 IMO level Prove it IMO

Find all integers n⩾3n \geqslant 3 with the following property: for all real numbers a1,a2,…,ana_{1}, a_{2}, \ldots, a_{n} and b1,b2,…,bnb_{1}, b_{2}, \ldots, b_{n} satisfying ∣ak∣+∣bk∣=1\left|a_{k}\right|+\left|b_{k}\right|=1 for 1⩽k⩽n1 \leqslant k \leqslant n, there exist x1,x2,…,xnx_{1}, x_{2}, \ldots, x_{n}, each of which is either −1-1 or 11, such that
∣∑k=1nxkak∣+∣∑k=1nxkbk∣⩽1 \left|\sum_{k=1}^{n} x_{k} a_{k}\right|+\left|\sum_{k=1}^{n} x_{k} b_{k}\right| \leqslant 1

Solutions — 2

Solution 1

Answer. nn can be any odd integer greater than or equal to 33.

For any even integer n⩾4n \geqslant 4, we consider the case
a1=a2=⋯=an−1=bn=0andb1=b2=⋯=bn−1=an=1. a_{1}=a_{2}=\cdots=a_{n-1}=b_{n}=0 \quad \text{and} \quad b_{1}=b_{2}=\cdots=b_{n-1}=a_{n}=1.
The condition ∣ak∣+∣bk∣=1\left|a_{k}\right|+\left|b_{k}\right|=1 is satisfied for each 1⩽k⩽n1 \leqslant k \leqslant n. No matter how we choose each xkx_{k}, both sums ∑k=1nxkak\sum_{k=1}^{n} x_{k} a_{k} and ∑k=1nxkbk\sum_{k=1}^{n} x_{k} b_{k} are odd integers. This implies ∣∑k=1nxkak∣⩾1\left|\sum_{k=1}^{n} x_{k} a_{k}\right| \geqslant 1 and ∣∑k=1nxkbk∣⩾1\left|\sum_{k=1}^{n} x_{k} b_{k}\right| \geqslant 1, which shows (1) cannot hold.

For any odd integer n⩾3n \geqslant 3, we may assume without loss of generality bk⩾0b_{k} \geqslant 0 for 1⩽k⩽n1 \leqslant k \leqslant n (this can be done by flipping the pair (ak,bk)\left(a_{k}, b_{k}\right) to (−ak,−bk)\left(-a_{k},-b_{k}\right) and xkx_{k} to −xk-x_{k} if necessary) and a1⩾a2⩾⋯⩾am⩾0>am+1⩾⋯⩾ana_{1} \geqslant a_{2} \geqslant \cdots \geqslant a_{m} \geqslant 0>a_{m+1} \geqslant \cdots \geqslant a_{n}. We claim that the choice xk=(−1)k+1x_{k}=(-1)^{k+1} for 1⩽k⩽n1 \leqslant k \leqslant n will work. Define
s=∑k=1mxkakandt=−∑k=m+1nxkak s=\sum_{k=1}^{m} x_{k} a_{k} \quad \text{and} \quad t=-\sum_{k=m+1}^{n} x_{k} a_{k}
Note that
s=(a1−a2)+(a3−a4)+⋯⩾0 s=\left(a_{1}-a_{2}\right)+\left(a_{3}-a_{4}\right)+\cdots \geqslant 0
by the assumption a1⩾a2⩾⋯⩾ama_{1} \geqslant a_{2} \geqslant \cdots \geqslant a_{m} (when mm is odd, there is a single term ama_{m} at the end, which is also positive). Next, we have
s=a1−(a2−a3)−(a4−a5)−⋯⩽a1⩽1 s=a_{1}-\left(a_{2}-a_{3}\right)-\left(a_{4}-a_{5}\right)-\cdots \leqslant a_{1} \leqslant 1
Similarly,
t=(−an+an−1)+(−an−2+an−3)+⋯⩾0 t=\left(-a_{n}+a_{n-1}\right)+\left(-a_{n-2}+a_{n-3}\right)+\cdots \geqslant 0
and
t=−an+(an−1−an−2)+(an−3−an−4)+⋯⩽−an⩽1. t=-a_{n}+\left(a_{n-1}-a_{n-2}\right)+\left(a_{n-3}-a_{n-4}\right)+\cdots \leqslant-a_{n} \leqslant 1.
From the condition, we have ak+bk=1a_{k}+b_{k}=1 for 1⩽k⩽m1 \leqslant k \leqslant m and −ak+bk=1-a_{k}+b_{k}=1 for m+1⩽k⩽nm+1 \leqslant k \leqslant n. It follows that ∑k=1nxkak=s−t\sum_{k=1}^{n} x_{k} a_{k}=s-t and ∑k=1nxkbk=1−s−t\sum_{k=1}^{n} x_{k} b_{k}=1-s-t. Hence it remains to prove
∣s−t∣+∣1−s−t∣⩽1 |s-t|+|1-s-t| \leqslant 1
under the constraint 0⩽s,t⩽10 \leqslant s, t \leqslant 1. By symmetry, we may assume s⩾ts \geqslant t. If 1−s−t⩾01-s-t \geqslant 0, then we have
∣s−t∣+∣1−s−t∣=s−t+1−s−t=1−2t⩽1 |s-t|+|1-s-t|=s-t+1-s-t=1-2 t \leqslant 1
If 1−s−t⩽01-s-t \leqslant 0, then we have
∣s−t∣+∣1−s−t∣=s−t−1+s+t=2s−1⩽1 |s-t|+|1-s-t|=s-t-1+s+t=2 s-1 \leqslant 1
Hence, the inequality is true in both cases.

These show nn can be any odd integer greater than or equal to 33.

Solution 2

The even case can be handled in the same way as Solution 1. For the odd case, we prove by induction on nn.

Firstly, for n=3n=3, we may assume without loss of generality a1⩾a2⩾a3⩾0a_{1} \geqslant a_{2} \geqslant a_{3} \geqslant 0 and b1=a1−1b_{1}=a_{1}-1 (if b1=1−a1b_{1}=1-a_{1}, we may replace each bkb_{k} by −bk-b_{k}).

- Case 1. b2=a2−1b_{2}=a_{2}-1 and b3=a3−1b_{3}=a_{3}-1, in which case we take (x1,x2,x3)=(1,−1,1)(x_{1}, x_{2}, x_{3})=(1,-1,1).
Let c=a1−a2+a3c=a_{1}-a_{2}+a_{3} so that 0⩽c⩽10 \leqslant c \leqslant 1. Then ∣b1−b2+b3∣=∣a1−a2+a3−1∣=1−c|b_{1}-b_{2}+b_{3}|=|a_{1}-a_{2}+a_{3}-1|=1-c and hence ∣c∣+∣b1−b2+b3∣=1|c|+|b_{1}-b_{2}+b_{3}|=1.

- Case 2. b2=1−a2b_{2}=1-a_{2} and b3=1−a3b_{3}=1-a_{3}, in which case we take (x1,x2,x3)=(1,−1,1)(x_{1}, x_{2}, x_{3})=(1,-1,1).
Let c=a1−a2+a3c=a_{1}-a_{2}+a_{3} so that 0⩽c⩽10 \leqslant c \leqslant 1. Since a3⩽a2a_{3} \leqslant a_{2} and a1⩽1a_{1} \leqslant 1, we have
c−1⩽b1−b2+b3=a1+a2−a3−1⩽1−c. c-1 \leqslant b_{1}-b_{2}+b_{3}=a_{1}+a_{2}-a_{3}-1 \leqslant 1-c.
This gives ∣b1−b2+b3∣⩽1−c|b_{1}-b_{2}+b_{3}| \leqslant 1-c and hence ∣c∣+∣b1−b2+b3∣⩽1|c|+|b_{1}-b_{2}+b_{3}| \leqslant 1.

- Case 3. b2=a2−1b_{2}=a_{2}-1 and b3=1−a3b_{3}=1-a_{3}, in which case we take (x1,x2,x3)=(−1,1,1)(x_{1}, x_{2}, x_{3})=(-1,1,1).
Let c=−a1+a2+a3c=-a_{1}+a_{2}+a_{3}. If c⩾0c \geqslant 0, then a3⩽1a_{3} \leqslant 1 and a2⩽a1a_{2} \leqslant a_{1} imply
c−1⩽−b1+b2+b3=−a1+a2−a3+1⩽1−c c-1 \leqslant-b_{1}+b_{2}+b_{3}=-a_{1}+a_{2}-a_{3}+1 \leqslant 1-c
If c<0c<0, then a1⩽a2+1a_{1} \leqslant a_{2}+1 and a3⩾0a_{3} \geqslant 0 imply
−c−1⩽−b1+b2+b3=−a1+a2−a3+1⩽1+c. -c-1 \leqslant-b_{1}+b_{2}+b_{3}=-a_{1}+a_{2}-a_{3}+1 \leqslant 1+c.
In both cases, we get ∣−b1+b2+b3∣⩽1−∣c∣|-b_{1}+b_{2}+b_{3}| \leqslant 1-|c| and hence ∣c∣+∣−b1+b2+b3∣⩽1|c|+|-b_{1}+b_{2}+b_{3}| \leqslant 1.

- Case 4. b2=1−a2b_{2}=1-a_{2} and b3=a3−1b_{3}=a_{3}-1, in which case we take (x1,x2,x3)=(−1,1,1)(x_{1}, x_{2}, x_{3})=(-1,1,1).
Let c=−a1+a2+a3c=-a_{1}+a_{2}+a_{3}. If c⩾0c \geqslant 0, then a2⩽1a_{2} \leqslant 1 and a3⩽a1a_{3} \leqslant a_{1} imply
c−1⩽−b1+b2+b3=−a1−a2+a3+1⩽1−c. c-1 \leqslant-b_{1}+b_{2}+b_{3}=-a_{1}-a_{2}+a_{3}+1 \leqslant 1-c.
If c<0c<0, then a1⩽a3+1a_{1} \leqslant a_{3}+1 and a2⩾0a_{2} \geqslant 0 imply
−c−1⩽−b1+b2+b3=−a1−a2+a3+1⩽1+c. -c-1 \leqslant-b_{1}+b_{2}+b_{3}=-a_{1}-a_{2}+a_{3}+1 \leqslant 1+c.
In both cases, we get ∣−b1+b2+b3∣⩽1−∣c∣|-b_{1}+b_{2}+b_{3}| \leqslant 1-|c| and hence ∣c∣+∣−b1+b2+b3∣⩽1|c|+|-b_{1}+b_{2}+b_{3}| \leqslant 1.

We have found x1,x2,x3x_{1}, x_{2}, x_{3} satisfying (1) in each case for n=3n=3.

Now, let n⩾5n \geqslant 5 be odd and suppose the result holds for any smaller odd cases. Again we may assume ak⩾0a_{k} \geqslant 0 for each 1⩽k⩽n1 \leqslant k \leqslant n. By the Pigeonhole Principle, there are at least three indices kk for which bk=ak−1b_{k}=a_{k}-1 or bk=1−akb_{k}=1-a_{k}. Without loss of generality, suppose bk=ak−1b_{k}=a_{k}-1 for k=1,2,3k=1,2,3. Again by the Pigeonhole Principle, as a1,a2,a3a_{1}, a_{2}, a_{3} lies between 00 and 11, the difference of two of them is at most 12\frac{1}{2}. By changing indices if necessary, we may assume 0⩽d=a1−a2⩽120 \leqslant d=a_{1}-a_{2} \leqslant \frac{1}{2}.

By the inductive hypothesis, we can choose x3,x4,…,xnx_{3}, x_{4}, \ldots, x_{n} such that a′=∑k=3nxkaka' = \sum_{k=3}^{n} x_{k} a_{k} and b′=∑k=3nxkbkb' = \sum_{k=3}^{n} x_{k} b_{k} satisfy ∣a′∣+∣b′∣⩽1|a'|+|b'| \leqslant 1. We may further assume a′⩾0a' \geqslant 0.

- Case 1. b′⩾0b' \geqslant 0, in which case we take (x1,x2)=(−1,1)(x_{1}, x_{2})=(-1,1).
We have ∣−a1+a2+a′∣+∣−(a1−1)+(a2−1)+b′∣=∣−d+a′∣+∣−d+b′∣⩽max⁡{a′+b′−2d,a′−b′,b′−a′,2d−a′−b′}⩽1|-a_{1}+a_{2}+a'|+|-(a_{1}-1)+(a_{2}-1)+b'|=|-d+a'|+|-d+b'| \leqslant \max \{a'+b'-2d, a'-b', b'-a', 2d-a'-b'\} \leqslant 1 since 0⩽a′,b′,a′+b′⩽10 \leqslant a', b', a'+b' \leqslant 1 and 0⩽d⩽120 \leqslant d \leqslant \frac{1}{2}.

- Case 2. 0>b′⩾−a′0 > b' \geqslant -a', in which case we take (x1,x2)=(−1,1)(x_{1}, x_{2})=(-1,1).
We have ∣−a1+a2+a′∣+∣−(a1−1)+(a2−1)+b′∣=∣−d+a′∣+∣−d+b′∣|-a_{1}+a_{2}+a'|+|-(a_{1}-1)+(a_{2}-1)+b'|=|-d+a'|+|-d+b'|. If −d+a′⩾0-d+a' \geqslant 0, this equals a′−b′=∣a′∣+∣b′∣⩽1a'-b'=|a'|+|b'| \leqslant 1. If −d+a′<0-d+a'<0, this equals 2d−a′−b′⩽2d⩽12d-a'-b' \leqslant 2d \leqslant 1.

- Case 3. b′<−a′b'<-a', in which case we take (x1,x2)=(1,−1)(x_{1}, x_{2})=(1,-1).
We have ∣a1−a2+a′∣+∣(a1−1)−(a2−1)+b′∣=∣d+a′∣+∣d+b′∣|a_{1}-a_{2}+a'|+|(a_{1}-1)-(a_{2}-1)+b'|=|d+a'|+|d+b'|. If d+b′⩾0d+b' \geqslant 0, this equals 2d+a′+b′<2d⩽12d+a'+b'<2d \leqslant 1. If d+b′<0d+b'<0, this equals a′−b′=∣a′∣+∣b′∣⩽1a'-b'=|a'|+|b'| \leqslant 1.

Therefore, we have found x1,x2,…,xnx_{1}, x_{2}, \ldots, x_{n} satisfying (1) in each case. By induction, the property holds for all odd integers n⩾3n \geqslant 3.

Want a route through all this instead of an archive? The track puts 2,604 problems in a working order, from Junior Challenge level to the IMO shortlist.

Source: MathNet, licensed CC-BY-4.0. Statement reproduced verbatim; metadata (topic, difficulty) added by this project.