USEMO 2024 Sol and Mark Scheme

Evan Chen

October 2024
1 Problem 1 (captain: Mihir Singhal)
§1.1 Current statement
There are 1001 stacks of coins S1 , S2 , . . . , S1001 . Initially, stack Sk has k coins for each
k = 1, 2, . . . , 1001. In an operation, one selects an ordered pair (i, j) of indices i and j
satisfying 1 ≤ i < j ≤ 1001 subject to two conditions:

• The stacks Si and Sj must each have at least one coin.

• The ordered pair (i, j) must not have been selected in any previous operation.

Then, if Si and Sj have a coins and b coins, respectively, one removes gcd(a, b) coins
from each stack.
What is the maximum number of times this operation could be performed?

§1.2 Solution
The answer is 500 · 501 = 250500 . Our solution is split into two parts. Firstly, we will
give a valid construction.
We start by performing operations (1001, 1000), (1001, 999), . . . , (1001, 1), in order. By
induction, at each step (1001, j), S1001 will have j+1 coins and thus, since gcd(j+1, j) = 1,
one coin will be removed from each stack. At the end of this process, 1000 operations will
have been performed. S1001 will have one coin; we discard it. The remaining (nonempty)
stacks will have 1, 2, . . . , 999 coins, and no operation will have been performed between
any of them. Thus we can repeat this process, performing operations with the 999-coin
stack and the rest of the stacks in descending order. Repeating this process until all the
stacks have been discarded, we perform 1000 + 998 + · · · + 2 = 500 · 501 operations, as
desired.
To prove this is the maximum number of operations we can perform, we bound the
total number of operations. The stacks S1 , . . . , S500 can only participate in at most
1 + · · · + 500 = 500 · 501/2 operations (since each operation
removes at least one coin
from them). The remaining 501 stacks can only perform 501 2 = 500 · 501/2 operations
between themselves, since each pair can only perform the operation once. Thus, in total,
we can perform at most 500 · 501 operations.

§1.3 Marking scheme

The solution is split into two parts: the lower bound (construction), worth 4 points,
and the upper bound, worth 3 points. These parts are completely additive.
In general, minor errors will be worth a deduction, but please message the channel
when you find any that are not included in the rubric so that we can add them to the
rubric for a deduction. Errors purely in arithmetic (even in the final answer), including
incorrect summation of an arithmetic series, will not merit any deduction.

2
Evan Chen — October 2024 USEMO 2024 Sol and Mark Scheme

§1.3.1 Lower bound

The following items are available (here n = 1001):

• 1 point for any construction that achieves Ω(n2 ) moves.

• 2 points for any construction that achieves n2 /4 − O(n) moves.

• 4 points for a correct lower bound construction (achieving 500 · 501 = (n2 − 1)/4
moves).

Up to 1 point may be deducted for constructions that are correct as stated but for which
insufficient justification is provided that the construction works. (For example, in the
case of the construction of the official solution, no justification would be required, since
the fact that it works is obvious enough not to require justification.)

§1.3.2 Upper bound

The following items are available:

• 1 point for any correct upper bound that is strictly less than b1001 · 1002/4c =
250750.

• 3 points for a correct upper bound (of 500 · 501 = (n2 − 1)/4).

3
2 Problem 2 (captain: Alec Sun)
§2.1 Current statement
Let k be a fixed positive integer. For each integer 1 ≤ i ≤ 4, let xi and yi be positive
integers such that their least common multiple is k. Suppose that the four points (x1 , y1 ),
(x2 , y2 ), (x3 , y3 ), (x4 , y4 ) are the vertices of a non-degenerate rectangle in the Cartesian
plane. Prove that x1 x2 x3 x4 is a perfect square.

§2.2 Solution
It suffices to prove that 2 | vp (x1 x2 x3 x4 ) for each prime p | k. Since the four points form
a rectangle, we have
x1 + x3 = x2 + x4 (2.1)
y1 + y3 = y2 + y4 (2.2)
x2 x4 − x1 x3 = y1 y3 − y2 y4 (2.3)
Let vp (k) = m. For each 1 ≤ i ≤ 4, we have vp (xi ), vp (yi ) ≤ m and equality holds for
at least one coordinate. We split into cases.
Case 1. If vp (xi ) = m for three i then vp (xi ) = m for the fourth by eq. (2.1), so
2 | vp (x1 x2 x3 x4 ) = 4m.
Case 2. If vp (yi ) = m for three i then vp (yi ) = m for the fourth by eq. (2.2). By
eq. (2.3) we have p2m | x2 x4 −x1 x3 . We now use the fact that vp (x±y) = min(vp (x), vp (y))
whenever vp (x) 6= vp (y). Since vp (x1 x3 ), vp (x2 x4 ) ≤ 2m, we have vp (x1 x3 ) = vp (x2 x4 )
and hence 2 | vp (x1 x2 x3 x4 ).
Case 3. Otherwise, vp (xi ) = m for exactly two i. If these i are consecutive (cyclically),
for example i = 1, 2 without loss of generality, then from eq. (2.1) we have pm | x3 − x4 .
Since vp (x3 ), vp (x4 ) ≤ m, we have vp (x3 ) = vp (x4 ) and hence 2 | vp (x1 x2 x3 x4 ). If these i
are not consecutive, for example i = 1, 3 without loss of generality, then from eq. (2.1) we
have pm | x2 + x4 , and we can finish using the same argument as in the consecutive case.
Remark. There are rectangles which satisfy the hypothesis, for instance (a, b), (a, ab),
(ab, ab), (ab, b) where gcd(a, b) = 1.

§2.3 Marking scheme

The following are awarded marks, with items being non-additive:

• 2 points for writing down eq. (2.1), eq. (2.2), and eq. (2.3). The condition
(y2 − y1 )(y4 − y1 ) = −(x2 − x1 )(x4 − x1 ) or similar in lieu of eq. (2.3) is accepted
but the condition (x1 − x3 )2 + (y1 − y3 )2 = (x2 − x4 )2 + (y2 − y4 )2 or similar is not.
• 2 points for solving the problem in at least one of Case 2 and Case 3, irrespective
of whether the conditions eq. (2.1), eq. (2.2), and eq. (2.3) are written down.
• 5 points for both of the above items.
• 7 points for a complete solution.

4
3 Problem 3 (captain: Hans Yu)
§3.1 Current statement
Let ABC be a triangle with incenter I. Two distinct points P and Q are chosen on the
circumcircle of ABC such that

∠AP I = ∠AQI = 45◦ .

Lines P Q and BC meet at S. Let H denote the foot of the altitude from A to BC.
Prove that ∠AHI = ∠ISH.

§3.2 Solution
§3.2.1 Solution via Tebault circles
Construct the Tebault circles ω1 and ω2 which are tangent to Ω = (ABC), side BC, and
cevian AH.

Lemma 3.2.1
P and Q are the tangency points of ωi and Ω.

Proof. Let ω1 touches BC at T , and Ω at P 0 . Our aim is to show that P 0 coincides with
P . First of all, it is well-known that the line P 0 T goes through the midpoint W of the
arc BC of Ω which doesn’t contain A (by angle chasing or inversion). Because of the
trivial similarities (a.k.a. Shooting Lemma) we have

W T · W P 0 = W B2 = W I 2
where the last equality holds because of the incircle-excircle lemma. Therefore, if
L = AW ∩ BC then

]IT H = ]IT W − ]LT W = −]P 0 IW − ]LT W.

As the tangent line of (ABC) at W is parallel to BC, we see that AP 0 T L are concyclic.
Using this:

−]P 0 IW − ]LT W = −]P 0 IW − ]IAP 0 = ]AP 0 I.

It is left to remember that ]IT H = ±45◦ because IT must be parallel to an angle
bisector of ]AHB by the properties of Tebault circles. In addition to the same fact for
Q0 it finishes the proof since there are only two such points P and Q on Ω.

Having proven the lemma, we apply it to deduce the next one:

Lemma 3.2.2
S is the center of the positive homothety which maps ω1 to ω2 .

5
Evan Chen — October 2024 USEMO 2024 Sol and Mark Scheme

Proof. It is trivial because of lemma 3.2.1 and Monge’s theorem for circles ω1 , ω2 , and Ω
since BC is already the common external tangent to ω1 and ω2 .

Let D be the tangency point of incircle ω of triangle ABC with BC; DD0 be the
diameter of ω, and ` be another tangent form S to ω1 , ω2 (which is a tangent to ω as
well because of the Tebault circles properties). Also let K = ` ∩ AH.

Lemma 3.2.3
SHIK is cyclic

Proof. Because of the Tebault circles property, the intersection of cevian and the second
tangent is always lying on the tangent to ω at D0 . In our case, it follows KD0 is tangent
to ω and thus parallel to the line BC. As a consequence, SI is parallel to another angle
bisector of ]SKD0 . Hence, ]KIS = 90◦ = ]KHS as desired.

From the third lemma we may conclude that

]KHI = ]KSI = ]ISH,

as desired.

Remark. For basic properties about Tebault circles you may see, e.g., the following article
having all the needed facts either as proven statements or solvable exercises to the reader with
hints (regrettably, it is in Russian only) https://geometry.ru/articles/protasovtebo.
pdf.

6
Evan Chen — October 2024 USEMO 2024 Sol and Mark Scheme

§3.2.2 Solution by Nikolai Beluhov

Solution.

O0
Q
V
Y
K
U

P
X
S
H
L

◦ A∗
MA90

IA

Let A∗ be the antipodal point of A on (ABC).

Lemma 3.2.4
]AHI = ]A∗ IA.

Proof. Llet IA be the excenter opposite A. Since ABH ∼ AA∗ C and ACH ∼ AA∗ B,
we get that AA∗ · AH = AB · AC. Since ABI ∼ AIA C and ACI ∼ AIA B, similarly
AI · AIA = AB · AC. But also ]A∗ AI = ]HAI, and we conclude that AIH ∼ AA∗ IA .
So ]AHI = ]AIA A∗ Let MA be the midpoint of IIA . Then as MA is on (ABC), we
know that AMA ⊥ MA A∗ , and so A∗ is on the perpendicular bisector of IIA . Thus
]AIA A∗ = ]A∗ IA,as desired. √
Note that AIH ∼ AA∗ IA can be seen simply by taking the AB · AC-inversion at A
as well.

From now on, we will be proving ]A∗ IA = ](SI, BC).

Let K, L be the second intersections of P I, QI and (ABC), respectively. Then the
angle condition is equivalent to that KL is the perpendicular bisector of AA∗ .

Lemma 3.2.5
SI passes through the circumcenter O0 of triangle AIA∗ .

7
Evan Chen — October 2024 USEMO 2024 Sol and Mark Scheme

Proof. Let U and V be the midpoints of the arcs AB and AC of (ABC). Then as U, V
are both on the perpendicular bisector of AI, we know that O0 = U V ∩ KL. It thus
suffices to show that U V, KL, SI are concurrent.
Let IS intersect (ABC) at X, Y . Then
C V
(S, I; X, Y ) = (B, U ; X, Y ) = (I, U V ∩ SI; X, Y )

and
P L
(S, I; X, Y ) = (Q, K; X, Y ) = (I, KL ∩ SI; X, Y ).
Thus U V ∩ SI = KL ∩ SI, as desired.

The rest is a straightforward angle chase. We know that

](SI, BC) = ]SO0 K + ](KL, BC) = ]IA∗ A + ]A∗ AI = ]A∗ IA,

as desired.

§3.2.3 Solutions by Hans Yu

Let IA be the A-excenter. Now let X be the second intersection of (P IQ) and (BIC).
Since S is the radical center of the circles (P IQ), (BIC) and (ABC), we see that SXI
are collinear.

I
X
P

S H

IA

Claim 3.2.6 — It suffices to show that IA HX are collinear.

Proof of the claim. Suppose that IA HX are collinear. It is well-known that BC bisects
IA HI (say, by harmonicity of (I, IA ; A, BC ∩ AI)). Therefore ]IHA = ]AHX =
90◦ − ](XI, AH) = 90◦ − ](SI, AH) = ]HSI, as desired. Here we used that ]HXI =
]IA XI = 90◦ .

Now let AH intersects (ABC) again at D.

Claim 3.2.7 — It suffices to show that AIA XD are concyclic.

8
Evan Chen — October 2024 USEMO 2024 Sol and Mark Scheme

Proof of the claim. Suppose that AIA XD are concyclic, then H = BC ∩ AD is the
radical center of (ABCD), (AIA XD) and (BICIA ). Hence H is also on IA X, and we
are done by claim 3.2.7.

Let P I, QI intersect the circumcircle of ABC again at P 0 , Q0 , and let A∗ be the

antipodal point of A on the circumcircle of ABC. Then since ∠AP P 0 = ∠AP I = 45◦
and so is ∠AQQ0 , we see that AP 0 A∗ Q0 is a square.
Let MA be the midpoint of IIA . Let O be the circumcenter of ABC. Then we can see
that MA is the midpoint of arc A∗ D as well: to see this, note that ]DAA∗ = ]OMA A =
]MA AO = ]MA AA∗ .
Let O0 be the circumcenter of AIA∗ , and let DI intersects (ABC) again at D0 .

O0
D0

Q
N
P0
I
X
P
Q0

D A∗
MA

IA

Claim 3.2.8 — It suffices to show that MA D0 ⊥ O0 D0 .

Proof of the claim. Suppose that MA D0 ⊥ O0 D0 . Let N be the midpoint of AI. Then
MA N D0 O0 are concyclic.
Consider the inversion at I sending A to MA . This inversion fixes the circumcircle
of ABC. It sends D0 to D and N to IA as IA I = 2MA I. Now to see where X is sent
to, note that (P QI) is sent to the line P 0 Q0 , which is the perpendicular bisector of AA∗ .
Moreover, (BIA C) is sent to the line through N perpendicular to AI, which is just the
perpendicular bisector of AI. Thus X is sent to the circumcenter of AIA∗ , which is
O0 . As a consequence, O0 is sent to X, and so the circle MA N D0 O0 is sent to the circle
AIA DX, and we are done by Claim 2.

9
Evan Chen — October 2024 USEMO 2024 Sol and Mark Scheme

D0O0

D
A∗
MA

To finish off, we will show that MA D0 ⊥ O0 D0 . Let Z be on ID0 such that A∗ Z k D0 MA .

Then ]IZA∗ = ]ID0 MA = ]DD0 MA = ]MA D0 A∗ = ]MA AA∗ = ]IAA∗ , showing
that Z is on the circumcircle of AIA∗ . As a consequence, O0 is on the perpendicular
bisector of A∗ Z. However, since ]D0 ZA∗ = ]DZA∗ = ]DD0 MA = ]MA D0 A∗ =
]ZA∗ D0 , we have that D0 is on the perpendicular bisector of A∗ Z as well. This shows
that O0 D0 ⊥ A∗ Z k MA D0 , as desired.

§3.3 Marking scheme

For all solutions, the following are not awarded marks:

• Rephrasing the angle condition in terms of P I ∩ (ABC) and QI ∩ (ABC).

• Swapping ]AHI with some other angles, even if they are used in the official
solutions.

For solutions not using Tebault circles, the following items are not additive:

• 2 points Showing that SI passes through the circumcenter of AIA∗ . Alternatively,

show that SI, the perpendicular bisector of AI and the perpendicular bisector of
AA∗ are concurrent.
Note: Points are still awarded if SI is replaced by some other two points that clearly
lie on S, I, the perpendicular bisector of AI is replaced by the line connecting two
points that are clearly on the perpendicular bisector, or the perpendicular bisector
of AA∗ is similarly replaced.

• 2 points Show that if SI passes through the circumcenter of AIA∗ , then the
statement holds true.

• 7 points Complete solution.

For solutions using Tebault circles, the following items are additive.

• +1 point Show that P, Q are tangency points of the Tebault circles to the circum-
circle.

• +1 point Show that S is the center of homothety of the two Tebault circles.

• +3 points Construct K and show that SHIK are concyclic.

• +2 points Finishing the solution.

10
4 Problem 4 (captain: Kevin Zhao)
§4.1 Current statement
Find all sequences a1 , a2 , . . .of nonnegative integers such that for all positive integers n,
the polynomial
1 + xa1 + xa2 + · · · + xan
has at least one integer root. (Here x0 = 1.)

§4.2 Solution
Note that letting Pn (x) = ni=0 xai with a0 = 0 means that plugging in any x ≥ 0 gets
P
since xk≥0 > 0, then Pn (x) ≥ 1 > 0 and so our root must be negative. We also note that
Pk (1) = k + 1, trivially.

Now, letting k = p − 1 where p is prime, we aim to find an integer root of Pk (x). Let
this integer root be −r where r > 0 - then, since Pp−1 (1) = p and Pp−1 (−r) = 0, we see
that setting Pp−1 (x) = (x + r)Qp−1 (x) for polynomial Q results in the fact that plugging
in x = 1 achieves

Pp−1 (1) = (1 + r)Qp−1 (x) → p = (1 + r)Qp−1 (x).

Since this means 1 + r | p and r > 0, then r = p − 1 is our only solution. This means
for any prime p, Pp−1 (−p + 1) = 0. Now, we use the Extremal Principle - consider the
largest ai , letting the value of it be M , and assume it occurs k ≥ 1 times - the sum of
the remaining p − k copies must total absolute value at least k(p − 1)M −1 but each can
contribute at most (p − 1)M −1 meaning we need at least

k(p − 1)M
= k(p − 1) ≥ p − 1
(p − 1)M −1

copies of ai , but since we have only p − k ≤ p − 1 remaining copies, all values of ai must
be equal to M − 1 besides one which is equal to M in order for our condition to even be
possible.

Note that a given value of ai is 0, so M − 1 = 0 or M = 0. Since all ai are nonnegative,

the latter is not valid and thus, M = 1. Now, if n = 1, 1 + xa1 must have a root meaning
a1 = 0 is impossible and thus a1 = 1, rendering all other P ai>1 = 0 fixed. We verify the
solution is correct - given any n, the polynomial 1 + x + ni=2 x0 = x + n indeed has a
solution of x = −n, so we are done.

§4.3 Marking scheme

For all solutions, the following are not awarded marks:

• Getting the correct answer with no explanation.

• Showing that all roots are negative.

11
Evan Chen — October 2024 USEMO 2024 Sol and Mark Scheme

For correct solutions:

• 7 points for a complete solution that shows a1 = 1 and ai>1 = 0 is the only
possible solution.

Solutions that are not complete may get the partial and the following, which are all
additive unless otherwise said:

• +2 points for noting that if n is one less than a prime, 1 − p must be the integer
root.

• +2 points for using Extremal Principle and taking the largest value of ai into
account.

• +2 points for showing that Pp−1 (0) = rQp−1 (0) so p − 1 | Pp−1 (0) - not additive
with the Extremal Principle bullet point.

• +1 point for setting the absolute value equation up in either of the two previous
cases.

For all solutions which are incomplete with errors, the following deductions apply and
are all additive:

• -1 point for not checking that a1 = 1 and ai>1 = 0 is indeed valid.

12
5 Problem 5 (captain: Hans Yu)
§5.1 Current statement
Let ABC be a scalene triangle whose incircle is tangent to BC, CA, AB at D, E, F
respectively. Lines BE and CF meet at G. Prove that there exists a point X on the
circumcircle of triangle EF G such that the circumcircles of triangles BCX and EF G
are tangent, and
∠BGC = ∠BXC + ∠EDF.

§5.2 Solution
Let (BDE) and (CDF ) intersect at X. We will show that this is the desired point.

X
A

E
F

B
D
C

Claim 5.2.1 — X is on (EF G).

This is evident by Miquel’s theorem on the triangle BGC with the points D, E, F on sides
BC, BG and CG. Alternatively, angle chasing suffices: ]EXF = ]EXD + ]DXF =
]EBD + ]DCF = ]EGF . Thus X, E, G, F are concyclic.

Claim 5.2.2 — ]BGC = ]BXC + ]EDF .

We can compute

]BXC = ]BXD + ]DXC

= ]BED + ]DF C
= ](BE, DE) + ](DF, F C)
= ](BE, F C) − ](DE, DF )
= ]BGC − ]EDF.

13
Evan Chen — October 2024 USEMO 2024 Sol and Mark Scheme

Claim 5.2.3 — (BXC) is tangent to (EF X) at X.

We will show this by showing that ]BCX − ]F EX = ]BXF . We do this by chasing
the angles as follows.

]BCX = ]DCX
= ]DF X
= ]DF E + ]EF X
= ]BDE + ]EF X
= ]EGF − (]EGF − ]BDE) + ]EF X
= ]EGF − (]EXF − ]BDE) + ]EF X
= ]EGF + (]BXE − ]EXF ) + ]EF X
= ]EGF + ]BXE + ]F EX
= ]EXF + ]BXE + ]F EX
= ]F EX + ]BXF.

§5.3 Alternative solution

Here we give an alternative solution after showing the first two claims. Take the inversion
with respect to the incircle. For any point Y , let Y 0 be its image under the inversion.

X0 E

C0
B0

Then D = D0 , E = E 0 and F = F 0 . Moreover, B 0 is the midpoint of DF and C 0 is the

midpoint of DE. Since X = (BDE) ∩ (CDF ), we have X 0 = (B 0 DE) ∩ (C 0 DF ). Note
that since (B 0 DE) and (C 0 DF ) intersect at D, X 0 and also B 0 F, C 0 E pass through D,
we have 4X 0 C 0 E ∼ 4X 0 F B 0 . Thus, as B 0 C 0 k EF , we have

]B 0 C 0 X 0 = ]F EX 0 + ]EX 0 C 0 = ]F EX 0 + ]B 0 X 0 F.

showing that (B 0 C 0 X 0 ) and (EF X 0 ) are tangent at X 0 . Inverting back, we get (BCX)
and (EF X) are tangent at X, as desired.

§5.4 Marking scheme

For all solutions, the following items are additive:

14
Evan Chen — October 2024 USEMO 2024 Sol and Mark Scheme

• +1 point for a correct description of X that allows a ruler-compass con-

struction determining a unique X. Most common examples are (1) X =
(BDE) ∩ (CDF ), and (2) Y = DF ∩ (EF G), Z = DE ∩ (EF G) and X = BY ∩ CZ.

• +1 point for showing that (BDE), (CDF ), (EF G) are concurrent either by angle
chasing or stating Miquel’s theorem.

• +1 points for proving that X satisfies the angle condition.

• +2 points for proving that X satisfies that (BCX) is tangent to (EF X).

• +2 points for getting all of the above items.

No partial points are awarded to non-synthetic solutions unless a synthetic statement

clearly equivalent to one of the items above is stated and proved explicitly in the solution.
No deductions are made for configuration issues.

15
6 Problem 6 (captain: Mihir Singhal)
§6.1 Current statement
Let n be an odd positive integer and consider an n × n chessboard of n2 unit squares.
In some of the cells of the chessboard, we place a knight. A knight in √ a cell c is said to
attack a cell c0 if the distance between the centers of c and c0 is exactly 5 (in particular,
a knight does not attack the cell which it occupies).
Suppose each cell of the board is attacked by an even number of knights (possibly
zero). Show that the configuration of knights is symmetric with respect to all four axes of
symmetry of the board (i.e. the configuration of knights is both horizontally and vertically
symmetric, and also unchanged by reflection along either diagonal of the chessboard).

§6.2 Solution
Let n = 2k + 1, and coordinatise the cells of the board by (x, y) with 0 ≤ x, y ≤ 2k.
Consider the width-two outer frame F of the board formed by all cells (x, y) which
satisfy at least one of the four conditions x ≤ 1, x ≥ 2k − 1, y ≤ 1, and y ≥ 2k − 1.
Observe that, if two valid configurations agree on F , then they agree everywhere.
Indeed, suppose not, and consider the earliest cell (x, y) where they disagree, going
from left to right and from top to bottom. The number of knights which attack cell
(x − 1, y − 2) will then differ by one between the two configurations, and we arrive at a
contradiction.
Thus it suffices to show that F is fully symmetric.
Let f (x, y) = 0 when cell (x, y) is empty and f (x, y) = 1 when it is occupied by a
knight. We treat the values of f as remainders modulo two, with 1 + 1 = 0. We also set
f (x, y) = 0 for all x and y which are not the coordinates of a cell.
Consider any set of cells S (these must all be valid cells), and let T be the set of all
cells with an odd number of knight neighbours in S. Then, in a valid configuration, the
sum of f over T will always be zero. (A)
For convenience, given a cell (x, y), let fp,q (x, y) be the sum of all eight expressions of
the form either f (x ± p, y ± q) or f (x ± q, y ± p). (Notice that some of these expressions
might coincide, and the coinciding ones will cancel out.)
For some m ≥ 1, consider any subboard M of our board of size (2m + 1) × (2m + 1)
centered at (x, y) (such that m ≤ x, y ≤ 2k−m). By (A) applied to the set of all cells in M
with the same checkerboard colour as (x, y), we get that fm−1,m (x, y)+fm+1,m+2 (x, y) = 0
(see fig. 6.1(i)).
Let us add together these identities over m, m − 2, m − 4, . . ., all the way down to the
remainder of m modulo two. Then all corresponding expressions of the form f`,`+1 (x, y)
with ` < m will cancel out, and in the end we will arrive at fm+1,m+2 (x, y) = 0, for all
m ≤ x, y ≤ 2k − m. (B)
By (B) with (x, y) = (k, k + 1) and m = k − 1, we get that f (0, 0) = f (2k, 0). (C)
By (A) with S = {(0, y), (2, y), (4, y), . . . , (2k, y)}, we get that f (0, y−1)+f (2k, y−1) =
f (0, y + 1) + f (2k, y + 1) for all y with 0 ≤ y ≤ 2k (see fig. 6.1(ii)). (D)
By induction on y, with (C) for the base case and (D) for the induction step, we
conclude that f (0, y) = f (2k, y) for all y. (E)
By (B) with (x, y) = (k + 1, k + 1) and m = k − 1, we get that f (0, 1) = f (1, 0). (F)

16
Evan Chen — October 2024 USEMO 2024 Sol and Mark Scheme

(i) (ii)

Figure 6.1: Visual demonstration of (B) and (D)

Similarly to (F), also f (2k − 1, 0) = f (2k, 1). By (E) with y = 1, (F), and the previous
identity, we arrive at f (1, 0) = f (2k − 1, 0). (G)
By (A) with S = {(1, y), (3, y), (5, y), . . . , (2k − 1, y)}, we get that f (1, y − 1) + f (2k −
1, y − 1) + f (1, y + 1) + f (2k − 1, y + 1) + f (0, y − 2) + f (2k, y − 2) + f (0, y + 2) +
f (2k, y + 2) = 0. Using (E) where y is substituted with y ± 2, the latter simplifies to
f (1, y − 1) + f (2k − 1, y − 1) = f (1, y + 1) + f (2k − 1, y + 1) for all y with 0 ≤ y ≤ 2k.
(H)

By induction on y, with (G) for the base case and (H) for the induction step, we
conclude that f (1, y) = f (2k − 1, y) for all y. (I)
Similarly to (E) and (I), also f (x, 0) = f (x, 2k) and f (x, 1) = f (x, 2k − 1) for all x.
(J)
By (A) with S = {(x, 0), (x − 1, 1), (x − 2, 2), . . . , (0, x)}, we get that f (x − 1, 0) +
f (x, 1) + f (x + 1, 0) + f (x + 2, 1) = f (0, x − 1) + f (1, x) + f (0, x + 1) + f (1, x + 2) for
all x with 0 ≤ x ≤ 2k. (K)
Similarly to (K), also f (x − 2, 2k − 1) + f (x − 1, 2k) + f (x, 2k − 1) + f (x + 1, 2k) =
f (2k−1, x−2)+f (2k, x−1)+f (2k−1, x)+f (2k, x+1). By (E), (I), and (J), it follows that
f (x−2, 1)+f (x−1, 0)+f (x, 1)+f (x+1, 0) = f (1, x−2)+f (0, x−1)+f (1, x)+f (0, x+1)
for all x with 0 ≤ x ≤ 2k. (L)
By induction on x, with (F) for the base case and (K) and (L) for the induction step,
we conclude that f (x, 0) = f (0, x) and f (x, 1) = f (1, x) for all x. (M)
Similarly to (M), also f (x, 2k) = f (2k, x) and f (x, 2k − 1) = f (2k − 1, x) for all x.
(N)
By (M) and (N), we get that F is symmetric with respect to the unit-slope main
diagonal of the board. By the same reasoning, F is symmetric with respect to the other
main diagonal of the board as well.

17
Evan Chen — October 2024 USEMO 2024 Sol and Mark Scheme

Consider any cell (x, y) of F . When either x ≤ 1 or x ≥ 2k − 1, we get that

f (x, y) = f (2k − x, y) by (E) and (I). Otherwise, when either y ≤ 1 or y ≥ 2k − 1, we
get that f (x, y) = f (y, x) by diagonal symmetry, f (y, x) = f (2k − y, x) by (E) and (I),
and f (2k − y, x) = f (2k − x, y) by diagonal symmetry once again.
Therefore, F is symmetric with respect to the vertical midline of the board. By the
same reasoning, F is symmetric with respect to the horizontal midline of the board as
well. The solution is complete.

Remark 6.2.1. The number of knight configurations which satisfy the conditions of the
problem is 2n .
This can be verified as follows. Let F` be the set of all cells (x, y) with min{x, 2k −
x, y, 2k − y} = `. Thus F0 , F1 , . . ., Fk form a partitioning of the board into pairwise disjoint
concentric frames, with F = F0 ∪ F1 .
First we place some knights on the cells of F so that F is fully symmetric. There are 2n
ways to do that.
It is straightforward to see that we can now fill in F2 uniquely so that every cell of F0 is
attacked by an even number of knights. Thus F2 will be fully symmetric as well. After that,
we can similarly fill in F3 so that every cell of F1 is attacked by an even number of knights,
and so on and so forth. Therefore, every fully symmetric arrangement of knights within F
can be extended to a valid configuration over the complete board in a unique manner.

Remark 6.2.2. When n is even, the number of knight configurations which satisfy the
conditions of the problem is 22n .
Here follows a quick sketch of the proof.
First we place some knights on rows 0 and 1 in an arbitrary manner. There are 22n ways
to do that.
It is straightforward to see that we can now fill in row 2 uniquely so that every cell of
row 0 is attacked by an even number of knights. After that, we can similarly fill in row 3
uniquely so that every cell of row 1 is attacked by an even number of knights, and so on
and so forth.
It turns out that, at the end of this process, every cell of rows n − 2 and n − 1 will be
attacked by an even number of knights as well. One proof relies on a series of applications
of (A) to certain sets of cells S within rows 0, 1, . . ., n − 3. Therefore, every arrangement of
knights within the lowermost couple of rows can be extended to a valid configuration over
the complete board in a unique manner.

§6.3 Marking scheme

We will give partial credit on an ad hoc basis, since we expect that the number of solutions
worth partial credit will be very small. If you believe a solution might be worth partial
credit, please let me (Mihir) know so that I can make sure that partial credit is awarded
in a consistent manner across solutions.
Specifically, there are likely many different chains of reasoning that will lead to the
solution, so be wary of awarding partial credit based solely on the solution following
similar steps to the official solution.
Concretely, we give points for the following, though any other significant progress may
be awarded partial credit, as we will decide when it comes up. These point totals are
subject to adjustment for now, as we read more solutions.

• 2 points for proving symmetry by 180 degree rotation about the cell (k, k) (using
the notation of thes solution). This may be done on the entire board, only on F ,

18
Evan Chen — October 2024 USEMO 2024 Sol and Mark Scheme

or on any other set that determines the rest of the knight positions (without proof,
unless it is not reasonably easy to see).

• 3 points for proving any other type of symmetry, i.e., horizontal or vertical
symmetry, symmetry across a main diagonal, or symmetry under 90 degree rotation.

• 7 points for a complete solution.

Point deductions may be given for minor flaws, but we will be especially lenient with
omitted details, since it’s somewhat unreasonable to expect every step to be justified in
detail.

19