IMO 2021 Solution Notes

Evan Chen《陳誼廷》
15 December 2024

This is a compilation of solutions for the 2021 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 2021/1, proposed by Australia . . . . . . . . . . . . . . . . . . . . . 3
1.2 IMO 2021/2, proposed by Calvin Deng . . . . . . . . . . . . . . . . . . . . 4
1.3 IMO 2021/3, proposed by Mykhalio Shtandenko (UKR) . . . . . . . . . . 5

2 Solutions to Day 2 7
2.1 IMO 2021/4, proposed by Dominik Burek (POL) and Tomasz Ciesla (POL) 7
2.2 IMO 2021/5, proposed by Spain . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 IMO 2021/6, proposed by Austria . . . . . . . . . . . . . . . . . . . . . . 11

1
IMO 2021 Solution Notes web.evanchen.cc, updated 15 December 2024

§0 Problems
1. Let n ≥ 100 be an integer. Ivan writes the numbers n, n+1, . . . , 2n each on different
cards. He then shuffles these n + 1 cards, and divides them into two piles. Prove
that at least one of the piles contains two cards such that the sum of their numbers
is a perfect square.

2. Show that the inequality

n X
X n q n X
X n q
|xi − xj | ≤ |xi + xj |
i=1 j=1 i=1 j=1

holds for all real numbers x1 , x2 , . . . , xn .

3. Let D be an interior point of the acute triangle ABC with AB > AC so that
∠DAB = ∠CAD. The point E on the segment AC satisfies ∠ADE = ∠BCD, the
point F on the segment AB satisfies ∠F DA = ∠DBC, and the point X on the line
AC satisfies CX = BX. Let O1 and O2 be the circumcenters of the triangles ADC
and EXD, respectively. Prove that the lines BC, EF , and O1 O2 are concurrent.

4. Let Γ be a circle with center I, and ABCD a convex quadrilateral such that each of
the segments AB, BC, CD and DA is tangent to Γ. Let Ω be the circumcircle of
the triangle AIC. The extension of BA beyond A meets Ω at X, and the extension
of BC beyond C meets Ω at Z. The extensions of AD and CD beyond D meet Ω
at Y and T , respectively. Prove that

AD + DT + T X + XA = CD + DY + Y Z + ZC.

5. Two squirrels, Bushy and Jumpy, have collected 2021 walnuts for the winter. Jumpy
numbers the walnuts from 1 through 2021, and digs 2021 little holes in a circular
pattern in the ground around their favourite tree. The next morning Jumpy notices
that Bushy had placed one walnut into each hole, but had paid no attention to
the numbering. Unhappy, Jumpy decides to reorder the walnuts by performing a
sequence of 2021 moves. In the kth move, Jumpy swaps the positions of the two
walnuts adjacent to walnut k.
Prove that there exists a value of k such that, on the kth move, Jumpy swaps some
walnuts a and b such that a < k < b.

6. Let m ≥ 2 be an integer, A a finite set of integers (not necessarily positive) and

B1 , B2 , . . . , Bm subsets of A. Suppose that, for every k = 1, 2, . . . , m, the sum of
the elements of Bk is mk . Prove that A contains at least m 2 elements.

2
IMO 2021 Solution Notes web.evanchen.cc, updated 15 December 2024

§1 Solutions to Day 1
§1.1 IMO 2021/1, proposed by Australia
Available online at https://aops.com/community/p22698392.

Problem statement

Let n ≥ 100 be an integer. Ivan writes the numbers n, n + 1, . . . , 2n each on different

cards. He then shuffles these n + 1 cards, and divides them into two piles. Prove
that at least one of the piles contains two cards such that the sum of their numbers
is a perfect square.

We will find three cards a < b < c such that

b + c = (2k + 1)2
c + a = (2k)2
a + b = (2k − 1)2

for some integer k. Solving for a, b, c gives

(2k)2 + (2k − 1)2 − (2k + 1)2

a= = 2k 2 − 4k
2
(2k + 1)2 + (2k − 1)2 − (2k)2
b= = 2k 2 + 1
2
(2k + 1)2 + (2k)2 − (2k − 1)2
c= = 2k 2 + 4k
2
We need to show that when n ≥ 100, one can find a suitable k.
Let

Ik := {n ∈ Z | n ≤ a < b < c ≤ 2n}

= {n ∈ Z | k 2 + 2k ≤ n ≤ 2k 2 − 4k}

be the interval such that when n ∈ Ik , the problem dies for that choice of k. It would
be sufficient to show these intervals Ik cover all the integers ≥ 100. Starting from
I9 = {99 ≤ n ≤ 126}, we have

k ≥ 9 =⇒ 2k 2 − 4k ≥ (k + 1)2 + 2(k + 1)

which means the right endpoint of Ik exceeds the left endpoint of Ik+1 . Hence for n ≥ 99
in fact the problem is true.

Remark. The problem turns out to be false for n = 98, surprisingly. The counterexample
is for one pile to be

{98, 100, 102, . . . , 126} ∪ {129, 131, 135, . . . , 161} ∪ {162, 164, . . . , 196}.

3
IMO 2021 Solution Notes web.evanchen.cc, updated 15 December 2024

§1.2 IMO 2021/2, proposed by Calvin Deng

Available online at https://aops.com/community/p22697952.

Problem statement

Show that the inequality

n X
X n q n X
X n q
|xi − xj | ≤ |xi + xj |
i=1 j=1 i=1 j=1

holds for all real numbers x1 , x2 , . . . , xn .

The proof is by induction on n ≥ 1 with the base cases n = 1 and n = 2 being easy to
verify by hand.
In the general situation, consider replacing the tuple (xi )i with (xi + t)i for some
parameter t ∈ R. The inequality becomes
n X
X n q n X
X n q
|xi − xj | ≤ |xi + xj + 2t|.
i=1 j=1 i=1 j=1

The left-hand side is independent of t.

Claim — The right-hand side, viewed as a function F (t) of t, is minimized when

2t = −(xi + xj ) for some i and j.

Proof. Since F (t) is the sum of piecewise concave functions, it is hence itself piecewise
concave. Moreover F increases without bound if |t| → ∞.
On each of the finitely many intervals on which F (t) is concave, the function is
minimized at its endpoints. Hence the minimum value must occur at one of the endpoints.

If t = −xi for some i, this is the same as shifting all the variables so that xi = 0. In
that case, we may apply induction on n − 1 variables, deleting the variable xi .
If t = − i 2 j , then notice
x +x

xi + t = −(xj + t)
so it’s the same as shifting all the variables such that xi = −xj . In that case, we may
apply induction on n − 2 variables, after deleting xi and xj .

4
IMO 2021 Solution Notes web.evanchen.cc, updated 15 December 2024

§1.3 IMO 2021/3, proposed by Mykhalio Shtandenko (UKR)

Available online at https://aops.com/community/p22698068.

Problem statement

Let D be an interior point of the acute triangle ABC with AB > AC so that
∠DAB = ∠CAD. The point E on the segment AC satisfies ∠ADE = ∠BCD, the
point F on the segment AB satisfies ∠F DA = ∠DBC, and the point X on the line
AC satisfies CX = BX. Let O1 and O2 be the circumcenters of the triangles ADC
and EXD, respectively. Prove that the lines BC, EF , and O1 O2 are concurrent.

This solution was contributed by Abdullahil Kafi.

Claim — Quadrilateral BCEF is cyclic.

Proof. Let D0 be the isogonal conjugate of the point D. The angle condition implies
quadrilateral CEDD0 and BF DD0 are cyclic. By power of point we have

AE · AC = AD · AD0 = AF · AB

So BCEF is cyclic.

Claim — Line ZD is tangent to the circles (BCD) and (DEF ) where Z = EF ∩BC.

Proof. Let ∠CAD = ∠BAD = α, ∠BCD = β, ∠DBC = γ, ∠ACD = φ, ∠ABD = .

From 4ABC we have 2α + β + γ + φ +  = 180◦ . Let ` be a line tangent to (BCD) and
K be a point on it in the same side of AD as C and L = AD ∩ BC. From our labeling
we have,

∠AF E = β + φ ∠BF D = α + γ ∠DF E = α + φ ∠CDL = α + φ

Now ∠CDJ = 180◦ − γ − β − (α + φ) = α + . So ∠DF E = ∠EDK = α + , which

means ` is also tangent to (DEF ). Now by the radical center theorem we have ` passes
through Z.

Let M be the Miquel point of the cyclic quadrilateral BCEF . From the Miquel
configuration we have A, M , Z are collinear and (AF EM ), (ZCEM ) are cyclic.

Claim — Points B, X, M , E are cyclic.

Proof. Notice that ∠EM B = 180◦ − ∠AM B − ∠EM Z = 180◦ − 2∠ACB = ∠EXB.

Let N be the other intersection of circles (ACD) and (DEX) and let R be the
intersection of AC and BM .

5
IMO 2021 Solution Notes web.evanchen.cc, updated 15 December 2024

N
A
M
R F
E

Z C B

Claim — Points B, D, M , N are cyclic.

Proof. By power of point we have

Pow(R, (ACD)) = RC · RA = RM · RB = RE · RX = Pow(R, (DEX)).

Hence R lies on the radical axis of (ACD) and (DEX), so N , R, D are collinear. Also

RN · RD = RA · RC = RM · RB

So BDM N is cyclic.

Notice that (ACD), (BDM N ), (DEX) are coaxial so their centers are collinear. Now
we just need to prove the centers of (ACD), (BDM N ) and Z are collinear. To prove
this, take a circle ω with radius ZD centered at Z. Notice that by power of point

ZC · ZB = ZD2 = ZE · ZF = ZM · ZA

which means inversion circle ω swaps (ACD) and (BDM N ). So the centers of (ACD)
and (BDM N ) must have to be collinear with the center of inversion circle, as desired.

6
IMO 2021 Solution Notes web.evanchen.cc, updated 15 December 2024

§2 Solutions to Day 2
§2.1 IMO 2021/4, proposed by Dominik Burek (POL) and Tomasz Ciesla
(POL)
Available online at https://aops.com/community/p22698001.

Problem statement

Let Γ be a circle with center I, and ABCD a convex quadrilateral such that each of
the segments AB, BC, CD and DA is tangent to Γ. Let Ω be the circumcircle of
the triangle AIC. The extension of BA beyond A meets Ω at X, and the extension
of BC beyond C meets Ω at Z. The extensions of AD and CD beyond D meet Ω
at Y and T , respectively. Prove that

AD + DT + T X + XA = CD + DY + Y Z + ZC.

Let P QRS be the contact points of Γ an AB, BC, CD, DA.

X
T

E
A
P
S
D

B I R

Q
Y
C
Z

Claim — We have 4IQZ ∼

= 4IRT . Similarly, 4IP X ∼
= 4ISY .

Proof. By considering (CQIR) and (CIT Z), there is a spiral similarity similarity mapping
4IQZ to 4IRT . Since IQ = IR, it is in fact a congruence.

This congruence essentially solves the problem. First, it implies:

Claim — T X = Y Z.

Proof. Because we saw IX = IY and IT = IZ.

7
IMO 2021 Solution Notes web.evanchen.cc, updated 15 December 2024

Then, we can compute

AD + DT + XA = AD + (RT − RD) + (XP − AP )

= (AD − RD − AP ) + RT + XP = RT + XP

and

CD + DY + ZC = CD + (SY − SD) + (ZQ − QC)

= (CD − SD − QC) + SY + ZQ = SY + ZQ

but ZQ = RT and XP = SY , as needed.

8
IMO 2021 Solution Notes web.evanchen.cc, updated 15 December 2024

§2.2 IMO 2021/5, proposed by Spain

Available online at https://aops.com/community/p22697921.

Problem statement

Two squirrels, Bushy and Jumpy, have collected 2021 walnuts for the winter. Jumpy
numbers the walnuts from 1 through 2021, and digs 2021 little holes in a circular
pattern in the ground around their favourite tree. The next morning Jumpy notices
that Bushy had placed one walnut into each hole, but had paid no attention to
the numbering. Unhappy, Jumpy decides to reorder the walnuts by performing a
sequence of 2021 moves. In the kth move, Jumpy swaps the positions of the two
walnuts adjacent to walnut k.
Prove that there exists a value of k such that, on the kth move, Jumpy swaps
some walnuts a and b such that a < k < b.

Assume for contradiction no such k exists. We will use a so-called “threshold trick”.
This process takes exactly 2021 steps. Right after the kth move, we consider a situation
where we color walnut k red as well, so at the kth step there are k ones. For brevity, a
non-red walnut is called black. An example is illustrated below with 2021 replaced by 6.

2 4 2 6 2 5 2 5

5 Initial 1 5 1 6 1 4 1

3 6 3 4 3 4 3 6

3 5 3 5 3 5

4 1 4 1 4 1

2 6 2 6 2 6

Claim — At each step, the walnut that becomes red is between two non-red or two
red walnuts.

Proof. By definition.

On the other hand, if there are 2021 walnuts, one obtains a parity obstruction to this
simplified process:

Claim — After the first step, there is always a consecutive block of black walnuts
positive even length.

Proof. After the first step, there is a block of 2020 black walnuts.

9
IMO 2021 Solution Notes web.evanchen.cc, updated 15 December 2024

Thereafter, note that a length 2 block of black walnuts can never be changed. Meanwhile
for even lengths at least 4, if one places a red walnut inside it, the even length block
splits into an odd length block and an even length block.

Remark. The statement is true with 2021 replaced by any odd number, and false for any
even number.
The motivation comes from the following rephrasing of the problem:
Start with all 0’s and at each step change a 0 between two matching numbers
from a 0 to a 1.
Although the coloring (or 0/1) argument may appear to lose information at first, I think
it should be equivalent to the original process; the “extra” information comes down to the
choice of which walnut to color red at each step.

10
IMO 2021 Solution Notes web.evanchen.cc, updated 15 December 2024

§2.3 IMO 2021/6, proposed by Austria

Available online at https://aops.com/community/p22698082.

Problem statement

Let m ≥ 2 be an integer, A a finite set of integers (not necessarily positive) and B1 ,

B2 , . . . , Bm subsets of A. Suppose that, for every k = 1, 2, . . . , m, the sum of the
elements of Bk is mk . Prove that A contains at least m 2 elements.

If 0 ≤ X < mm+1 is a multiple of m, then write it in base m as

m
X
X= ci mi ci ∈ {0, 1, 2, . . . , m − 1}
i=1

Then swapping the summation to over A through the Bi ’s gives

 
n
fa (X)a where fa (X) :=
X X X X
X=  b ci = ci .
i=1 b∈Bi a∈A i:a∈Bi

Evidently, 0 ≤ fa (X) ≤ n(m − 1) for any a and X. So, setting |A| = n, the right-hand
side of the display takes on at most (n(m − 1) + 1)n distinct values. This means

mm ≤ (n(m − 1))n

which implies n ≥ m/2.

Remark (Motivation comments from USJL). In linear algebra terms, we have some n-
dimensional 0/1 vectors v~1 , . . . , v~m and an n-dimensional vector ~a such that v~i · ~a = mi
for i = 1, . . . , m. The intuition is that if n is too small, then there should be lots of linear
dependences between v~i .
In fact, Siegel’s lemma is a result that says, if there are many more vectors than the
dimension of the ambient space, there exist linear dependences whose coefficients are not-
too-big integers. On the other hand, any linear dependence between m, m2 , . . . , mm is
going to have coefficients that are pretty big; at least one of them needs to exceed m.
Applying Siegel’s lemma turns out to solve the problem (and is roughly equivalent to the
solution above).

Remark.
 In https://aops.com/community/p23185192,
 dgrozev shows the stronger bound
n ≥ 3 + log m m elements, for some absolute constant c > 0.
2 c

11