IMO 2017 Solution Notes

Evan Chen《陳誼廷》
17 July 2024

This is a compilation of solutions for the 2017 IMO. The ideas of the
solution are a mix of my own work, the solutions provided by the competition
organizers, and solutions found by the community. However, all the writing
is maintained by me.
These notes will tend to be a bit more advanced and terse than the “official”
solutions from the organizers. In particular, if a theorem or technique is not
known to beginners but is still considered “standard”, then I often prefer to
use this theory anyways, rather than try to work around or conceal it. For
example, in geometry problems I typically use directed angles without further
comment, rather than awkwardly work around configuration issues. Similarly,
sentences like “let R denote the set of real numbers” are typically omitted
entirely.
Corrections and comments are welcome!

Contents
0 Problems 2

1 Solutions to Day 1 3
1.1 IMO 2017/1, proposed by Stephan Wagner (SAF) . . . . . . . . . . . . . 3
1.2 IMO 2017/2, proposed by Dorlir Ahmeti (ALB) . . . . . . . . . . . . . . . 5
1.3 IMO 2017/3, proposed by Gerhard Woeginger (AUT) . . . . . . . . . . . 7

2 Solutions to Day 2 9
2.1 IMO 2017/4, proposed by Charles Leytem (LUX) . . . . . . . . . . . . . . 9
2.2 IMO 2017/5, proposed by Grigory Chelnokov (RUS) . . . . . . . . . . . . 11
2.3 IMO 2017/6, proposed by John Berman (USA) . . . . . . . . . . . . . . . 12

1
IMO 2017 Solution Notes web.evanchen.cc, updated 17 July 2024

§0 Problems
1. For each integer a0 > 1, define the sequence a0 , a1 , a2 , . . . , by
(√ √
an if an is an integer,
an+1 =
an + 3 otherwise

for each n ≥ 0. Determine all values of a0 for which there is a number A such that
an = A for infinitely many values of n.

2. Solve over R the functional equation

f (f (x)f (y)) + f (x + y) = f (xy).

3. A hunter and an invisible rabbit play a game in the plane. The rabbit and hunter
start at points A0 = B0 . In the nth round of the game (n ≥ 1), three things occur
in order:
(i) The rabbit moves invisibly from An−1 to a point An such that An−1 An = 1.
(ii) The hunter has a tracking device (e.g. dog) which reports an approximate
location Pn of the rabbit, such that Pn An ≤ 1.
(iii) The hunter moves visibly from Bn−1 to a point Bn such that Bn−1 Bn = 1.
Let N = 109 . Can the hunter guarantee that AN BN < 100?

4. Let R and S be different points on a circle Ω such that RS is not a diameter. Let
` be the tangent line to Ω at R. Point T is such that S is the midpoint of RT .
Point J is chosen on minor arc RS of Ω so that the circumcircle Γ of triangle JST
intersects ` at two distinct points. Let A be the common point of Γ and ` closer to
R. Line AJ meets Ω again at K. Prove that line KT is tangent to Γ.

5. Fix N ≥ 1. A collection of N (N + 1) soccer players of distinct heights stand in a

row. Sir Alex Song wishes to remove N (N − 1) players from this row to obtain a
new row of 2N players in which the following N conditions hold: no one stands
between the two tallest players, no one stands between the third and fourth tallest
players, . . . , no one stands between the two shortest players. Prove that this is
possible.

6. An irreducible lattice point is an ordered pair of integers (x, y) satisfying gcd(x, y) =

1. Prove that if S is a finite set of irreducible lattice points then there exists a
nonconstant homogeneous polynomial f (x, y) with integer coefficients such that
f (x, y) = 1 for each (x, y) ∈ S.

2
IMO 2017 Solution Notes web.evanchen.cc, updated 17 July 2024

§1 Solutions to Day 1
§1.1 IMO 2017/1, proposed by Stephan Wagner (SAF)
Available online at https://aops.com/community/p8633268.

Problem statement

For each integer a0 > 1, define the sequence a0 , a1 , a2 , . . . , by

(√ √
an if an is an integer,
an+1 =
an + 3 otherwise

for each n ≥ 0. Determine all values of a0 for which there is a number A such that
an = A for infinitely many values of n.

The answer is a0 ≡ 0 (mod 3) only.

¶ First solution. We first compute the minimal term of any sequence, periodic or not.

Lemma
Let c be the smallest term in an . Then either c ≡ 2 (mod 3) or c = 3.

Proof. Clearly c 6= 1, 4. Assume c 6≡ 2 (mod 3). As c is not itself a square, the next
√ √ √
perfect square after c in the sequence is one of (b cc + 1) , (b cc + 2) , or (b cc + 3) .
2 2 2

So by minimality we require
√  √
c≤ c +3≤ c+3

which requires c ≤ 5. Since c 6= 1, 2, 4, 5 we conclude c = 3.

Now we split the problem into two cases:

• If a0 ≡ 0 (mod 3), then all terms of the sequence are 0 (mod 3). The smallest term
of the sequence is thus 3 by the lemma and we have

3→6→9→3

so A = 3 works fine.

• If a0 6≡ 0 (mod 3), then no term of the sequence is 0 (mod 3), and so in particular 3
does not appear in the sequence. So the smallest term of the sequence is 2 (mod 3)
by lemma. But since no squares are 2 (mod 3), the sequence ak grows without
bound forever after, so no such A can exist.

Hence the answer is a0 ≡ 0 (mod 3) only.

3
IMO 2017 Solution Notes web.evanchen.cc, updated 17 July 2024

¶ Second solution. We clean up the argument by proving the following lemma.

Lemma
If an is constant modulo 3 and not 2 (mod 3), then an must eventually cycle in the
form (m, m + 3, m + 6, . . . , m2 ), with no squares inside the cycle except m2 .

Proof. Observe that an must eventually hit a square, say ak = c2 ; the next term is
ak+1 = c. Then it is forever impossible to exceed c2 again, by what is essentially discrete
intermediate value theorem. Indeed, suppose a` > c2 and take ` > k minimal (in
√
particular a` 6= a`−1 ). Thus a`−1 ∈ {c2 − 2, c2 − 1, c2 } and thus for modulo 3 reasons
we have a`−1 = c2 . But that should imply a` = c < c2 , contradiction.
We therefore conclude sup{an , an+1 , . . . } is a decreasing integer sequence in n. It must
eventually stabilize, say at m2 . Now we can’t hit a square between m and m2 , and so we
are done.

Now, we contend that all a0 ≡ 0 (mod 3) work. Indeed, for such a0 we have an ≡ 0
(mod 3) for all n, so the lemma implies that the problem statement is valid.
Next, we observe that if ai ≡ 2 (mod 3), then the sequence grows without bound
afterwards since no squares are 2 (mod 3). In particular, if a0 ≡ 2 (mod 3) the answer
is no.
Finally, we claim that if a0 ≡ 1 (mod 3), then eventually some term is 2 (mod 3).
Assume for contradiction this is not so; then an ≡ 1 (mod 3) must hold forever, and the
lemma applies to give us a cycle of the form (m, m + 3, . . . , m2 ) where m ≡ 1 (mod 3).
In particular m ≥ 4 and
m ≤ (m − 2)2 < m2
but (m − 2)2 ≡ 1 (mod 3) which is a contradiction.

4
IMO 2017 Solution Notes web.evanchen.cc, updated 17 July 2024

§1.2 IMO 2017/2, proposed by Dorlir Ahmeti (ALB)

Available online at https://aops.com/community/p8633190.

Problem statement

Solve over R the functional equation

f (f (x)f (y)) + f (x + y) = f (xy).

The only solutions are f (x) = 0, f (x) = x − 1 and f (x) = 1 − x, which clearly work.
Note that

• If f is a solution, so is −f .

• Moreover, if f (0) = 0 then setting y = 0 gives f ≡ 0. So henceforth we assume

f (0) > 0.

Claim — We have f (z) = 0 ⇐⇒ z = 1. Also, f (0) = 1 and f (1) = 0.

Proof. For the
forwards direction, if f (z) = 0 and z 6= 1 one may put (x, y) =
z, z(z − 1)−1 (so that x + y = xy) we deduce f (0) = 0 which is a contradiction.
For the reverse, f (f (0)2 ) = 0 by setting x = y = 0, and use the previous part. We also
conclude f (1) = 0, f (0) = 1.

Claim — If f is injective, we are done.

Proof. Setting y = 0 in the original equation gives f (f (x)) = 1 − f (x). We apply this
three times on the expression f 3 (x):

f (1 − f (x)) = f (f (f (x))) = 1 − f (f (x)) = f (x).

Hence 1 − f (x) = x or f (x) = 1 − x.

Remark. The result f (f (x)) + f (x) = 1 also implies that surjectivity would solve the
problem.

Claim — f is injective.

Proof. Setting y = 1 in the original equation gives f (x + 1) = f (x) − 1, and by induction

f (x + n) = f (x) − n. (1)

Assume now f (a) = f (b). By using (1) we may shift a and b to be large enough that we
may find x and y obeying x + y = a + 1, xy = b. Setting these gives

f (f (x)f (y)) = f (xy) − f (x + y) = f (b) − f (a + 1)

= f (b) + 1 − f (a) = 1

5
IMO 2017 Solution Notes web.evanchen.cc, updated 17 July 2024

from which we conclude

f (f (x)f (y) + 1) = 0.
Hence by the first claim we have f (x)f (y) + 1 = 1, so f (x)f (y) = 0. Applying the first
claim again gives 1 ∈ {x, y}. But that implies a = b.

Remark. Jessica Wan points out that for any a 6= b, at least one of a2 > 4(b − 1) and
b2 > 4(a − 1) is true. So shifting via (1) is actually unnecessary for this proof.

Remark. One can solve the problem over Q using only (1) and the easy parts. Indeed,
that already implies f (n) = 1 − n for all n. Now we induct to show f (p/q) = 1 − p/q for all
0 < p < q (on q). By choosing x = 1 + p/q, y = 1 + q/p, we cause xy = x + y, and hence
0 = f (f (1 + p/q)f (1 + q/p)) or 1 = f (1 + p/q)f (1 + q/p).
By induction we compute f (1 + q/p) and this gives f (p/q + 1) = f (p/q) − 1.

6
IMO 2017 Solution Notes web.evanchen.cc, updated 17 July 2024

§1.3 IMO 2017/3, proposed by Gerhard Woeginger (AUT)

Available online at https://aops.com/community/p8633324.

Problem statement

A hunter and an invisible rabbit play a game in the plane. The rabbit and hunter
start at points A0 = B0 . In the nth round of the game (n ≥ 1), three things occur
in order:

(i) The rabbit moves invisibly from An−1 to a point An such that An−1 An = 1.

(ii) The hunter has a tracking device (e.g. dog) which reports an approximate
location Pn of the rabbit, such that Pn An ≤ 1.

(iii) The hunter moves visibly from Bn−1 to a point Bn such that Bn−1 Bn = 1.

Let N = 109 . Can the hunter guarantee that AN BN < 100?

No, the hunter cannot. We will show how to increase the distance in the following way:

Claim — Suppose the rabbit is at a distance d ≥ 1 q

from the hunter at some point
in time. Then it can increase its distance to at least d2 + 1
2 in 4d steps regardless
of what the hunter already knows about the rabbit.

Proof. Consider a positive integer n > d, to be chosen later. Let the hunter start at B
and the rabbit at A, as shown. Let ` denote line AB.
Now, we may assume the rabbit reveals its location A, so that all previous information
becomes irrelevant.
The rabbit chooses two points X and Y symmetric about ` such that XY = 2 and
AX = AY = n, as shown. The rabbit can then hop to either X or Y , pinging the point
Pn on the ` each time. This takes n hops.
n X
hunter rabbit
M
B A H
n
Y

Now among all points H the hunter can go to, min max{HX, HY } is clearly minimized
with H ∈ ` by symmetry. So the hunter moves to a point H such that BH = n as well.
In that case the new distance is HX = HY .
We now compute
p 2
HX 2 = 1 + HM 2 = 1 + AX 2 − 1 − AH
p 2
=1+ n2 − 1 − (n − d)
  2
1
≥1+ n− − (n − d)
n
= 1 + (d − 1/n)2

which exceeds d2 + 1
2 whenever n ≥ 4d.

7
IMO 2017 Solution Notes web.evanchen.cc, updated 17 July 2024

In particular we can always take n = 400 even very crudely; applying the lemma
2 · 1002 times, this gives a bound of 400 · 2 · 1002 < 109 , as desired.

Remark. The step of revealing the location of the rabbit seems critical because as far as I
am aware it is basically impossible to keep track of ping locations in the problem.

Remark. Reasons to believe the answer is “no”: the 109 constant, and also that “follow
the last ping” is losing for the hunter.

Remark. I think there are roughly two ways you can approach the problem once you
recognize the answer.

(i) Try and control the location of the pings

(ii) Abandon the notion of controlling
√ possible locations, and try to increase the distance
by a little bit, say from d to d2 + ε. This involves revealing the location of the
rabbit before each iteration of several jumps.

I think it’s clear that the difficulty of my approach is realizing that (ii) is possible; once you
do, the two-point approach is more or less the only one possible.
My opinion is that (ii) is not that magical; as I said it was the first idea I had. But I
am biased, because when I test-solved the problem at the IMO it was called “C5” and not
“IMO3”; this effectively told me it was unlikely that the official solution was along the lines
of (i), because otherwise it would have been placed much later in the shortlist.

8
IMO 2017 Solution Notes web.evanchen.cc, updated 17 July 2024

§2 Solutions to Day 2
§2.1 IMO 2017/4, proposed by Charles Leytem (LUX)
Available online at https://aops.com/community/p8639236.

Problem statement

Let R and S be different points on a circle Ω such that RS is not a diameter. Let `
be the tangent line to Ω at R. Point T is such that S is the midpoint of RT . Point J
is chosen on minor arc RS of Ω so that the circumcircle Γ of triangle JST intersects
` at two distinct points. Let A be the common point of Γ and ` closer to R. Line
AJ meets Ω again at K. Prove that line KT is tangent to Γ.

¶ First solution (elementary). First, note

]RKA = ]RKJ = ]RSJ = ]T SJ = ]T AJ = ]T AK

so RK k AT . Now,

• RA is tangent at R iff 4KRS ∼ 4RT A (oppositely), because both equate to

−]RKS = ]SKR = ]SRA = ]T RA.

• Similarly, T K is tangent at T iff 4KT S ∼ 4ART .

• The two similarities are equivalent because RS = ST the SAS gives KR · T A =

RS · RT = T S · T R.

K∗ T J∗

S
J

R A B

Remark. The problem is actually symmetric with respect to two circles; RA is tangent at
R if and only if T K at T .

¶ Second solution (inversion). Consider an inversion at R fixing the circumcircle Γ of

T SJA. Then:

• T and S swap,

9
IMO 2017 Solution Notes web.evanchen.cc, updated 17 July 2024

• A and B swap, where B is the second intersection of ` with Γ.

• Circle Ω inverts to the line through T parallel to RAB, call it `.

• J ∗ is the second intersection of ` with Γ.

• K ∗ is the intersection of ` with the circumcircle of RBJ ∗ ; this implies RK ∗ J ∗ B is

an isosceles trapezoid. In particular, one reads RK ∗ k AT from this, hence RK ∗ T A
is a parallelogram.

Thus we wish to show the circumcircle of RSK ∗ is tangent to Γ. But that follows from
the final parallelogram observed: S is the center of the parallelogram since it is the
midpoint of the diagonal.

Remark. This also implies RKT B is cyclic, from K ∗ SA collinear. Moreover, quadrilateral
KK ∗ T S is cyclic (by power of a point); this leads to the second official solution to the
problem.

10
IMO 2017 Solution Notes web.evanchen.cc, updated 17 July 2024

§2.2 IMO 2017/5, proposed by Grigory Chelnokov (RUS)

Available online at https://aops.com/community/p8639240.

Problem statement

Fix N ≥ 1. A collection of N (N + 1) soccer players of distinct heights stand in a

row. Sir Alex Song wishes to remove N (N − 1) players from this row to obtain a new
row of 2N players in which the following N conditions hold: no one stands between
the two tallest players, no one stands between the third and fourth tallest players,
. . . , no one stands between the two shortest players. Prove that this is possible.

Some opening remarks: location and height are symmetric to each other, if one
thinks about this problem as permutation pattern avoidance. So while officially there are
multiple solutions, they are basically isomorphic to one another, and I am not aware of
any solution otherwise.

12
11
10
9
8
7
6
5
4
3
2
1

Take a partition of N groups in order by height: G1 = {1, . . . , N + 1}, G2 = {N +

2, . . . , 2N + 2}, and so on. We will pick two people from each group Gk .
Scan from the left until we find two people in the same group Gk . Delete all people
scanned and also everyone in Gk . All the groups still have at least N people left, so we
can induct down with the non-deleted people; the chosen pair is to the far left anyways.

Remark. The important bit is to scan by position but group by height, and moreover not
change the groups as we scan. Dually, one can have a solution which scans by height but
groups by position.

11
IMO 2017 Solution Notes web.evanchen.cc, updated 17 July 2024

§2.3 IMO 2017/6, proposed by John Berman (USA)

Available online at https://aops.com/community/p8639242.

Problem statement

An irreducible lattice point is an ordered pair of integers (x, y) satisfying gcd(x, y) =

1. Prove that if S is a finite set of irreducible lattice points then there exists a
nonconstant homogeneous polynomial f (x, y) with integer coefficients such that
f (x, y) = 1 for each (x, y) ∈ S.

We present two solutions.

¶ First solution (Dan Carmon, Israel). We prove the result by induction on |S|, with
the base case being Bezout’s Lemma (n = 1). For the inductive step, suppose we want
to add a given pair (am+1 , bm+1 ) to {(a1 , . . . , am ), (b1 , . . . , bm )}.

Claim (Standard) — By a suitable linear transformation we may assume

(am+1 , bm+1 ) = (1, 0).

Outline of proof. It would be sufficient to show there exists a 2 × 2 matrix T = [ us vt ]

with integer entries such that det T = 1 and T · bm+1 = [ 10 ]. Then we could apply T
am+1

to all the ordered pairs in S (viewed as column vectors); if f was a polynomial that
works on the transformed ordered pairs, then f (ux + vy, sx + ty) works for the original
ordered pairs. (Here, the condition that det T = 1 ensures that T −1 has integer entries,
and hence that T maps irreducible lattice points
 u to irreducible lattice points.)
To generate such a matrix T , choose T = −bm+1 am+1 where u and v are chosen via
v 

Bézout lemma so that uam+1 + vbm+1 = 1. This matrix T is rigged so that det T = 1
and the leftmost column of T −1 is bm+1 .
am+1

Remark. This transformation claim is not necessary to proceed; the solution below can be
rewritten to avoid it with only cosmetic edits. However, it makes things a bit easier to read.

Let g(x, y) be a polynomial which works on the latter set. We claim we can choose the
new polynomial f of the form
m
deg g·M −m
Y
M
f (x, y) = g(x, y) − Cx (bi x − ai y).
i=1

where C and M are integer parameters we may adjust.

Since f (ai , bi ) = 1 by construction we just need
Y
1 = f (1, 0) = g(1, 0)M − C bi .

If bi = 0 we are done, since bi = 0 =⇒ ai = ±1 in that case and so g(1, 0) = ±1, thus

Q
take M = 2. So it suffices to prove:

12
IMO 2017 Solution Notes web.evanchen.cc, updated 17 July 2024

Claim — We have gcd (g(1, 0), bi ) = 1 when bi 6= 0.

Proof. Fix i. If bi = 0 then ai = ±1 and g(±1, 0) = ±1. Otherwise know

1 = g(ai , bi ) ≡ g(ai , 0) (mod bi )

and since the polynomial is homogeneous with gcd(ai , bi ) = 1 it follows g(1, 0) 6≡ 0

(mod bi ) as well.

Then take M a large multiple of ϕ( |bi |) and we’re done.

Q

¶ Second solution (Lagrange). The main claim is that:

Claim — For every positive integer N , there is a homogeneous polynomial P (x, y)

such that P (x, y) ≡ 1 (mod N ) whenever gcd(x, y) = 1.
(This claim is actually implied by the problem.)

Proof. For N = pe a prime take (xp−1 + y p−1 )ϕ(N ) when p is odd, and (x2 + xy + y 2 )ϕ(N )
for p = 2.
Now, if N is a product of primes, we can collate coefficient by coefficient using the
Chinese remainder theorem.

Let S = {(ai , bi ) | i = 1, . . . , m}. We have the natural homogeneous “Lagrange polyno-

mials” Y
Lk (x, y) := (bi x − ai y)
i6=k

Now let Y
N := Lk (xk , yk )
k

and take P as in the claim. Then we can take a large power of P , and for each i subtract
an appropriate multiple of Li (x, y); that is, choose
X
f (x, y) = P (x, y)C − Li (x, y) · Qi (x, y)
i

where C is large enough that C deg P > maxi deg Li , and Qi (x, y) is any homogeneous
C
polynomial of degree C deg P − deg Li such that Lk (xk , yk )Qk (xk , yk ) = P (xk ,yNk ) −1 ·
Lk (xk , yk ) (which is an integer).

13