You are on page 1of 43

LMAO SL 2021

Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas


May 11, 2021

§1 Problems
Algebra
A1 Find all functions f : Z → Z such that

f (x + y) + f (x)2 f (y) = f (y)3 + f (x + y)f (x)2

for all x, y ∈ Z
-Anish Kulkarni

A2 On a board finite sequences of real numbers are written . In a move we write a new
sequence on the board leaving the original one untouched in one of the following
things-
For a given sequence (a1 , a2 , · · · , an ) written on the board write a new sequence
a1 a2 an
(0, , , · · · , ) .
1 2 n
Choose two sequence < ai >ni=1 and < bi >m i=1 where n ≥ m written on
the board and two arbitrary non-negative real number α, β and write a new
sequence < ci >ni=1 on the board where ci = αai + βbi for i ≤ m and ci = αai
for i > m.
At the start the following sequences are written on the board-

(1), (1, −1, 1), (1, −1, 1, −1, 1), · · ·

And
(4, −2, 1), (9, −3, 1), ...., (k 2 , −k, 1) · · ·

Is it possible to reach (2021, 0, 0, · · · , 0, −2021, 1) for some finite number of zeroes?


-Anish Kulkarni

A3 A function g : [0, 1] → [0, 1] of the type g(x) = (c − x)−1 is called simplistic for
any real number c > 2. Alice multiplies and takes the composition of simplistic
functions in any combination, to get some function f : [0, 1] → [0, 1]. Prove that
for any real numbers 0 ≤ a, b ≤ 1, Alice has that f (a2 ) + f (b2 ) ≥ 2f (ab)
For example, if g1 , g2 , g3 , g4 are simplistic, one possibility is that f (x) := g1 (g2 (x)g3 (g4 (x)))

-Pranjal Srivastava

1
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

A4 Find all functions f : R → R and Polynomials P such that

f (P (x + y) + yP (f (x) + y)) = f (x) + y(1 + x + P (y))

-Ojas Mittal and Tejas Mittal

A5 Let x, y, z be positive real numbers, with x + y + z = 3, prove that


Y √ √ √
(1 + x)(10 + 5x + yz) ≥ 64( 3 x + 3 y + 3 z − 1)9

-Pranjal Srivastava

2
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

Combinatorics
C1 There are N boxes labelled B1 , B2 , · · · , BN which are filled with balls of N different
colours C1 , C2 , · · · , CN . Further, it is known that for each colour we can partition
the boxes into two sets, such that the total number of balls of that colour in both
sets is the same.
Prove that one can choose a box Bi , such that at most half of the balls in box Bi
are of colour Ci .
-Mahavir Gandhi

C2 Jerry the mouse traces a non-intersecting polygonal path on a plane, always facing
one of the four cardinal directions, and ends up where it started, facing the same
direction. Prove that the number of times Jerry turned from North to East, and
the number of times Jerry turned from East to North differ by 1.
-Ojas Mittal

C3 Cat-keeper Pranjal places cats C1 , C2 , · · · Cn on the plane. The cats wear magic
helmets that can restrict their field of vision to an angle of their choosing. They
now tilt their heads to restrict their vision so that no cat can see another cat. If
n
P
the field of vision of Ci is αi , then the score of the cats is defined to be αi . Show
i=1
that Pranjal can prevent the cats from scoring more than 1000 ln n and that the
cats can always score at least π ln n.
-Pranjal Srivastava and Rohan Goyal

C4 A game is played on a rectangular board as follows. Initially, there is one stone


on the bottom-left corner of the board. On each turn, the player may pick up any
stone, remove it, and do one of the following:
Put one stone on each square to the right of the selected square.
Put one stone on each square above the selected square.
Reverse the order of the stones to the right of the selected square.
Reverse the order of the stones above the selected square.
For example, if the board is
2 1 3 2
1 4 0 1
0 6 3 4
3 2 0 3
4 0 0 1
where the numbers represent the number of stones on the square, and the player
chooses the square highlighted in bold, then each of the above operations will
respectively result in the boards

2 1 3 2
1 4 0 1
0 6 3 4
3 1 1 4
4 0 0 1

3
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

2 2 3 2
1 5 0 1
0 7 3 4
3 1 0 3
4 0 0 1

2 1 3 2
1 4 0 1
0 6 3 4
3 1 3 0
4 0 0 1

2 6 3 2
1 4 0 1
0 1 3 4
3 1 0 3
4 0 0 1
The player may play as many turns as they wish. The goal of the player is to
maximize the number of stones they have at the end of the game. If two players
play optimally on boards of dimensions 2021 × 2021 and 2022 × 2020 respectively,
which player is going to finish with more stones?
-Ritam Nag

4
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

Geometry
G1 Let H be the orthocentre of acute angled triangle ABC. Suppose that E and F lie
on AB and AC and that circles (BHE) and (CHF ) intersect again at P . Prove
that HP passes through the circumcentre of EAF .
-Pranjal Srivastava

G2 Let M be the midpoint of side BC of triangle ABC. Suppose that K and T are
the intersection of the reflection of AM over bisector of angle BAC with (ABC)
and BC respectively. Let ABV C be a parallelogram, and let S be the reflection of
K in BC. If U is the reflection of A in S, prove that line T V passes through the
midpoint of KU .
-Pranjal Srivastava

G3 Let I be incentre of 4ABC and let Γ be its circumcircle. Let D, E and F be


tangency points of the incircle of 4ABC with BC, CA and AB respectively. Let
the circle with centre A passing through I meet Γ at points A1 and A2 . Similarly
define B1 , B2 and C1 , C2 . Let A0 , B 0 and C 0 be feet of perpendiculars from I onto
A1 A2 , B1 B2 and C1 C2 respectively. Prove that A0 D, B 0 E and C 0 F are concurrent.
-Mahavir Gandhi
Note. Finally, an easier version was asked. The concurrency instead demanded
was AA0 , BB 0 , CC 0 .

G4 Let ABC be a scalene triangle with circumcircle Ω and circumcenter O. Let the A
symmedian intersect the circle Ω at D. Let K be the midpoint of AD. Let ω be a
circle through A and K tangent to the circumcircle of B, O and C. Let ω intersect
AC, AB and Ω at X, Y, Z such that X, Y, Z 6= A. Prove that AZ, XY concur on
BC.
-Rohan Goyal

G5 Let I and IA be incentre and excentre opposite to A of 4ABC respectively and


let L be midpoint of arc BAC. Let LI and LIA meet (ABC) again at points
Y and Z respectively. Prove that if AX is internal angle bisector of ∠BAC with
X ∈ BC then one of the external common tangents of (ABC) and (XY Z)
passes through antipode of A in (ABC).
-Mahavir Gandhi

G6 Let P, Q be chosen inside triangle ABC such that ∠P BC = ∠ABQ and ∠P CB =


∠ACQ. The tangent to (BP C) at P intersects the tangent to (BQC) at Q at R.
Prove that the circles (ABC) and (P QR), and line AR share a common point.
-Mahavir Gandhi

5
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

G7 Bob marks three special points A, B, C on the plane, such that the euler line of
ABC is not parallel to BC. Given a special point P (ie P is one of A, B, C), and
an arbitrary point Q, Alice can draw in line P Q and the circle centred at P that
passes through Q. Prove that Alice can construct the intersection of the euler line
of ABC with BC.
-Pranjal Srivastava
Note. We found this too similar to a pre-existing AoPS thread, so while I’ve left
it in the shortlist, it was withdrawn from selection for the actual mock.

6
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

Number Theory
N1 Let p1 < p2 < p3 · · · be the sequence of all prime numbers. Suppose we have
a set S ⊆ N such that for all natural k, S contains a number in the interval
[p2021k , p2021(k+1) ). Prove that infinitely many primes divide at least one element
of S
-Pranjal Srivastava

N2 Let k be a positive integer such that p = 2k − 1 is a prime. Let P be an integer


polynomial satisfying
π
P (sin(θ + 2k−1 )) = P (sin(θ))
for all θ ∈ R. Define the sequence (ai )i≥0 by a0 = 1, a1 = 2, and

an+1 = 4an − an−1

Show that for any i, j we have p | P (ai ) − P (aj ).


-Ritam Nag

N3 Find the least positive integer k that satisfies the following:


For any monic polynomial P (x) of degree 2021 with integer coefficients, there exists
a set T (dependent on P ) of k integers, such that there is no set S which satisfies
the following three conditions simultaneously:
a) T ⊆ S
b) S is proper subset of Z
c) If u, v ∈ S, then P (u) + v ∈ S
-Anish Kulkarni

N4 Given an odd
j prime
k q, find all integer polynomials P such that every prime p > q 2021 ,
p
divides P (q q
), and P (1) = 0.
-Rohan Goyal

Remark. The condition P (1) = 0 is not necessary but helps avoid some heavy
machinery.

7
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

§2 Solutions
Algebra
A1 Find all functions f : Z → Z such that

f (x + y) + f (x)2 f (y) = f (y)3 + f (x + y)f (x)2

for all x, y ∈ Z
-Anish Kulkarni
Solution. Put x = 0 in the given expression to get f (y) = f (y)3 for all y ∈ Z
And hence range of f is a subset of {0, −1, 1}.
The new expression is f (x + y) + f (x)2 f (y) = f (y) + f (x + y)f (x)2 which on
rearranging gives

P (x, y) : (f (x)2 − 1)(f (x + y) − f (y)) = 0


Now we do case work
Case I - 0 has no pre-image
Define f arbitrarily on Z to take values from {−1, 1}
Case II- f (x) = 0 ⇐⇒ x = 0
Again we define f arbitrarily on Z/{0} to take values from {−1, 1}
Case III - There exists a non zero integer a such that f (a) = 0
Let S be the set of pre-images of 0, so a ∈ S
P (a, 0) =⇒ f (0) = 0 =⇒ 0 ∈ S
Let s, t ∈ S, P (s, t) =⇒ s + t ∈ S
Let s ∈ S, P (s, −s) =⇒ −s ∈ S
And hence S is a subgroup of Z and hence of form kZ for some positive integer k
Now P (k, y) =⇒ f (y + k) = f (y) and hence f is periodic with period k
Now again we arbitrarily define f on {1, 2, · · · , k − 1} to take values from {−1, 1}

Further we see that all the solutions described above indeed satisfy given equation.

8
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

A2 On a board finite sequences of real numbers are written . In a move we write a new
sequence on the board leaving the original one untouched in one of the following
things-
For a given sequence (a1 , a2 , · · · , an ) written on the board write a new sequence
a1 a2 an
(0, , , · · · , ) .
1 2 n
Choose two sequence < ai >ni=1 and < bi >m i=1 where n ≥ m written on
the board and two arbitrary non-negative real number α, β and write a new
sequence < ci >ni=1 on the board where ci = αai + βbi for i ≤ m and ci = αai
for i > m.
At the start the following sequences are written on the board-

(1), (1, −1, 1), (1, −1, 1, −1, 1), · · ·

And
(4, −2, 1), (9, −3, 1), · · · ., (k 2 , −k, 1) · · ·

Is it possible to reach (2021, 0, 0, · · · , 0, −2021, 1) for some finite number of zeroes?


-Anish Kulkarni
Solution. For each sequence (ai )ni=1 consider P (x) = a1 + a2 x + · · · + an xn−1 .
The first move corresponds to getting Pc (x) = αPa (x) + RβPb (x).
x
Second corresponds to writing new polynomial equal to 0 P (t) dt.
Then for all starting sequences we have P (x) ≥ 0 for x ≥ 0
Moreover this property is invariant under all the operations.
But it is not true for any sequence of form (2021, 0, 0, · · · , 0, −2021, 1).
Hence the answer is NO.

9
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

A3 A function g : [0, 1] → [0, 1] of the type g(x) = (c − x)−1 is called simplistic for
any real number c > 2 . Alice multiplies and takes the composition of simplistic
functions in any combination, to get some function f : [0, 1] → [0, 1]. Prove that
for any real numbers 0 ≤ a, b < 1, Alice has that f (a2 ) + f (b2 ) ≥ 2f (ab)
For example, if g1 , g2 , g3 , g4 are simplistic, one possibility is that f (x) := g1 (g2 (x)g3 (g4 (x)))

-Pranjal Srivastava
Solution 1. Let S be the set of increasing convex functions from [0, 1] → [0, 1].

Lemma 2.1
If f, g ∈ S, then so are f g and f ◦ g

Proof. Suppose 0 ≤ a ≤ b ≤ 1.
f (a)g(a) ≤ f (b)g(b), so f g is increasing, and f ( a+b a+b 1
2 )g( 2 ) ≤ 4 (f (a) +
1
f (b))(g(a) + g(b)) ≤ 2 (f (a)g(a) + f (b)g(b)) by the rearrangement inequality
noting that f and g are both increasing, so f g is convex.
a ≤ b =⇒ g(a) ≤ g(b) =⇒ f (g(a)) ≤ f (g(b)), so f ◦ g is increasing.
g(a)+g(b)
f (g( a+b
2 )) ≤ f ( 2 ) ≤ f (g(a))+f
2
(g(b))
, so f ◦ g is convex.

It is clear that the simplistic functions all lie in S, so it follows that f ∈ S. Then, we
2 2
have f (a2 ) + f (b2 ) ≥ 2f ( a +b
2 ) ≥ 2f (ab), by fact that f is convex and increasing,
respectively.

Solution 2. If f, g have positive first and second derivative then so do f ◦ g and f g


because of the identities -

(f g)0 = f 0 g + f g 0 ; (f g)00 = f 00 g + 2f 0 g 0 + f g 00

(f ◦ g)0 (x) = (f 0 (g(x)))g 0 (x) ; (f ◦ g)00 = (f 00 (g(x))g 0 (x)2 + (f 0 (g(x))g 00 (x)


Now for any simplistic function g(x) = (c − x)−1 , g 0 (x) = (c − x)−2 and g 00 (x) =
2(c − x)−3 , both of which are positive for x in [0, 1].
Hence for any function f which Alice constructs,f 0 (x), f 00 (x) > 0 for x in [0, 1].
Now let h(x) = f (ex ), so h00 (x) = f 00 (ex )e2x + f 0 (x)ex > 0 when ex ∈ [0, 1].
Hence h is convex and by Jensen we have h(u) + h(v) ≥ 2h( u+v 2 ).
2 2 2 2
Take u = ln a , v = ln b =⇒ f (a ) + f (b ) ≥ 2f (ab).
Hence proved.

Remark. The set of decreasing convex functions (or the set of functions with deriva-
tives of alternating signs) is also closed under the given operations. So the problem
1
also holds if the domain is changed to R+ and the starting functions changed to c+x
with c ≥ 0, even if we include addition.

10
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

A4 Find all functions f : R → R and Polynomials P such that

f (P (x + y) + yP (f (x) + y)) = f (x) + y(1 + x + P (y))

-Ojas Mittal and Tejas Mittal


Solution. Plug y = 0 ⇒ f (P (x)) = f (x) hence f (P (P · · · (P (x)) · · · ) = f (x)

Lemma:Suppose degreeP > 1 then ∃N so that for fixed x > N P (P · · · (P (x) · · · )


for some finitely many iterations attains large enough values or small enough values
if leading coefficient is positive or negative respectively.

Proof :

Case1-P has positive leading coefficient.

Degree P ≥ 2 hence P (x) − x has same degree,leading coefficient as P and thus be-
comes large enough for large enough x thus ∃N so that for x ≥ N we get P (x) > x
hence P i (x) > P i−1 (x) for x ≥ N thus x, P (x), · · · is an increasing sequence hence
P i (x) attains large enough values.

Case2-P has negative leading coefficient.

As degree P > 1 ∃N so that |P (x)| > |x| for all|x| > N and P (x)x is negative(as P
has negative leading coefficient) hence for fix x > N |x|, |P (x)|, · · · is an increasing
sequence hence for any M we have ∃i so that |P i (x)| > M now if P i (x) is neg-
ative then we are done. If not then P i+1 (x) is negative and |P i+1 (x)| > |P i (x)| > M

Now fix x > N as in lemma then consider P (x + y) + yP (f (x) + y) = Q(y) as a


polynomial in y then clearly leading coefficient of this polynomial is same as that
of P hence the polynomial attains all values large enough/small enough for degree
P positive/negative respectively(by IVT).

Hence ∃ infinitely many y : y1 , y2 , · · · so that Q(yi ) = P Ji (x)

⇒ yi (1 + x + P (yi )) = 0 for all i

⇒ P (yi ) = P (yj ) for infinitely many j taking yi 6= 0(this can be done as yi are
distinct.)

Hence the polynomial P is a constant polynomial contradiction as degP > 2.

therefore we get P linear!!. Assume P (x) = ax + b now we plug P in the original


equation.

Case1-a > 0

11
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

Fix x. Consider the polynomial P (x + y) + yP (f (x) + y) − P (x) = R(y).


Clearly R(y) = a(x + y) + b + y(a(f (x) + y) + b) − (ax + b) = y(a + b + ay + af (x))
plugging y = −1 − f (x) − b/a we see that R(y) = 0.

Hence f (P (x + y) + yP (f (x) + y)) = f (P (x)) = f (x)

Hence y(1 + x + ay + b) = 0

⇒ either y = 0 or y = −(1 + x + b)/a

⇒ either f (x) = −1 − b/a or f (x) = (1 + x)/a − 1

Now let −1 − b/a = c then we further make 2 cases

Subcase1:∃N so that f (x) 6= c for all x > N

Then consider x so that x > N and P (x) = ax + b > N (This can be done as a > 0).
Clearly there are infinitely many such x
Then we have 1 + (ax + b)/a − 1 = f (ax + b) = f (x) = 1 + x/a − 1 so ax + b = x
for infinitely many x.

Hence a = 1,b = 0 ie f (x) = x for allx

Subcase2:∃x1 < x2 < · · · so that f (xi ) = c∀i and the sequence is unbounded.

Then consider Q(y) = P (x1 + y) + yP (f (x1 ) + y) then clearly leading coefficient of


Q(y) > 0.
Degree Q(y) > 0 hence ∃y1 , y2 · · · so that Q(yi ) = xj for some j(as the sequence xi
is unbounded and range of Q is [T, ∞]).
But then yi (1 + x1 + ayi + b) = 0∀i This is a contradiction.

Case2:a < 0

Fix x. Consider the polynomial P (x + y) + yP (f (x) + y) − P (x) = R(y).


Clearly R(y) = a(x + y) + b + y(a(f (x) + y) + b) − (ax + b) = y(a + b + ay + af (x))
plugging y = −1 − f (x) − b/a we see that R(y) = 0.

Hence f (P (x + y) + yP (f (x) + y)) = f (P (x)) = f (x)

Hence y(1 + x + ay + b) = 0

⇒ either y = 0 or y = −(1 + x + b)/a

⇒ either f (x) = −1 − b/a or f (x) = (1 + x)/a − 1

12
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

Now let −1 − b/a = c then we further make 2 cases

Subcase1:∃N so that f (x) 6= c for all |x| > N

Then consider x so that |x| > N and |P (x)| = |ax + b| > N (This can be done as
a < 0).
Clearly there are infinitely many such x
Then we have 1 + (ax + b)/a − 1 = f (ax + b) = f (x) = 1 + x/a − 1 so ax + b = x
for infinitely many x.

Hence a = 1,b = 0 ie f (x) = x for allx

Subcase2:∃x1 , x2 , · · · so that f (xi = c)∀i and the sequence is unbounded.


Then note that range of Q(y) = P (x1 + y) + yP (f (x1 ) + y) is [−∞, T ] (as a < 0)
now note that as the sequence is unbounded we have infinitely many xi so that
|xi | > |T | and |axi + b| > |T | but note that xi , axi + b have opposite signs. so one
of them belongs to [−∞, T ].Thus we get y1 , y2 · · · so that Q(yi ) = xj oraxj + b for
some j Thus f (Q(yi )) = f (xj ) = f (axj + b)
⇒ yi (1 + x1 + ayi + b)) = 0
⇒ yi = 0 or (−1 − x1 − b)/a But this is a contradiction as there were infinitely
many distinct such yi
Case3:a = 0
Thus P (x) = b
⇒ f (P (x)) = f (x) ⇒ f (x) = b ⇒ y(1 + x + b) = 0∀x, y a contradiction.
Hence the only function satisfying is f (x) = x∀x!!! Clearly these satisfy.

13
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

A5 Let x, y, z be positive real numbers, with x + y + z = 3, prove that


Y √ √ √
(1 + x)(10 + 5x + yz) ≥ 64( 3 x + 3 y + 3 z − 1)9

-Pranjal Srivastava
Solution. First, observe that

(2 + x + y)(2 + x + z) = (4 + 4x + 2y + 2z + x2 + xy + xz + yz)
= 4 + 6 + 2x + x(x + y + z) + yz
= 10 + 5x + yz

Now, let a3 = x, b3 = y, c3 = z
We have that
Y Y
(1 + x)(10 + 5x + yz) ≥ (1 + x)(2 + x + y)(2 + x + z)
Y
≥ (a3 + 1)(2 + b3 + c3 )(2 + b3 + c3 )
1Y 3
≥ (2a + 1 + 1)(2 + b3 + c3 )(2 + b3 + c3 )
8
1Y
≥ (2a + b2 + c2 )3
8
1Y
≥ (2a + 2b − 1 + 2c − 1)3
8
1
≥ (2a + 2b + 2c − 2)9
8
≥ 64(a + b + c − 1)9
√ √ √
≥ 64( 3 x + 3 y + 3 z − 1)9

14
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

Combinatorics
Problem 2.2.

C1 There are N boxes labelled B1 , B2 , · · · , BN which are filled with balls of N different
colours C1 , C2 , · · · , CN . Further, it is known that for each colour we can partition
the boxes into two sets, such that the total number of balls of that colour in both
sets is the same.
Prove that one can choose a box Bi , such that at most half of the balls in box Bi
are of colour Ci .
-Mahavir Gandhi
Solution. Let f (i, j) be the number of balls of colour Ci in box Bj and g(i) be the
number of balls in box Bi .
Now, observe that
n
X
2f (i, i) ≤ f (i, j)
j=1

by the partitioning rule. Now, summing over all elements, we get that
n
X n X
X n n
X
2 f (i, i) ≤ f (i, j) = g(i)
i=1 i=1 j=1 i=1

Now, if f (i, i) > 12 g(i) for all i then


n
X n
X
2f (i, i) > g(i)
i=1 i=1

which contradicts our previous claim and we are done!

15
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

C2 A mouse named Jerry traces a non-intersecting polygonal path on a plane, always


facing one of the four cardinal directions, and ends up where it started, facing the
same direction. Prove that the number of times Jerry turned from North to East,
and the number of times Jerry turned from East to North differ by 1.
-Ojas Mittal
Solution. Let E, N, W, S denote the east, north, west and south directions respec-
tively.
Consider an infinite line consisting of the text . . . N ESW N ESW . . . .
We keep track of a pointer, that always points to the letter which is the direction
the mouse is moving in. Every time the mouse turns, the pointer either slides
forward or backward by 1.

Note that as the polygon isn’t self-intersecting, the polygon will always be to the
same side of Jerry’s path. WLOG let the polygon always be to the Jerry’s right.
This means, that the pointer moving forward by 1, corresponds to Jerry turning
right, and the polygon having an interior angle of π2 . Similarly, the pointer moving
backward by 1 corresponds to Jerry turning left, and the polygon having an interior
angle of 3π
2

However, since the sum of exterior angles is 2π, there are precisely 4 more interior
angles of size π2 than 3π
2 . In other words, this means that the number of times
the pointer moves forward is 4 more that the number of times the pointer moves
backward, ie it ends up 4 positions ahead of where it started.

Now consider any letter N in the infinite line. If this letter does not lie between
where the pointer started and eventually ended, the pointer crosses from it to the
neighbouring E, and from the neighbouring E to it an equal number of times. This
corresponds to turning east from north, and north from east an equal number of
times. However if the letter was the unique N lying between where the pointer
started and ended, it is crossed one time more in the forward direction, this means
that the pointer moves forward once more than backward from this N . Thus overall,
the pointer moves once more from an N to an E than from any E to an N . Put
in Jerry’s terms, Jerry turns once more from the North to East than from east to
north. Note that if we had assumed that the polygon was always to the left of
Jerry’s path, we would have arrived at the conclusion that Jerry turns once more
from east to north as compared with to north to east.

Remark: We can also generalise this to the directions of a regular n-gon so that
adjacent sides have consecutive directions.

16
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

C3 Cat-keeper Pranjal places cats C1 , C2 , · · · Cn on the plane. The cats wear magic
helmets that can restrict their field of vision to an angle of their choosing. They
now tilt their heads to restrict their vision so that no cat can see another cat. If
n
P
the field of vision of Ci is αi , then the score of the cats is defined to be αi . Show
i=1
that Pranjal can prevent the cats from scoring more than 1000 ln n, but that the
cats can always score at least π ln n.
-Pranjal Srivastava and Rohan Goyal
Solution 1. We first explicitly create a configuration where the cats cannot guar-
antee an angle sum of 1000 ln n. To do this, pick an extremely small positive real
number t
Pranjal asks the cats to arrange themselves as follows on the complex plane; In
k
the k’th step, he asks 2k−1 cats to stand on the primitive 2k ’th roots of tk2 . All

cats placed on the k’th step can each only wear a helmet of angle at most 2k−1 +
k−1
[the previous placed cats make an regular 2 −gon viewed from the perspective
of such a cat]. Thus, the cats from the kth step can overall see an angle of at most
4π. Since there are log2 (n) + 1 steps, the cats cannot ensure angle sum more than
4π(log2 (n) + 1) ≤ 16 ln n ≤ π 3 ln n < 1000 ln n.
We now show that f (n) > π ln n.
Orient the xy axis such that no two cats share a co-ordinate.
WLOG their x co-ordinates are in increasing order.
Since there are i − 1 cats to the left of Ci , she can pick an helmet confining her
vision to the left plane with αi ≥ πi . Thus the cats can ensure a score of at least
P π
i > π ln n, regardless of how they are positioned.
i

Solution 2. The first part, f (n) < 1000 ln n, is same as above .


For other part, we consider the convex hull C1 = C11 C12 · · · C1t1 .
Now remove these points and consider the convex hull of remaining points to be
C2 = C21 C22 · · · C2t2 .
Continue this process until you partition all points into such convex hulls C1 , .., Ck .
Now consider Ci = Ci1 Ci2 · · · Citi , let ∠Ci(u−1) Ciu Ci(u+1) = θu .
In reflex angle ∠Ci(u−1) Ciu Ci(u+1) there are at most t1 + t2 + · · · + ti−1 points
2π−θi
Hence we can give Ciu a helmet of size at least t1 +t2 +···+t i−1 +1
.
Adding over Ci we have total helmet angle of at least t1 +t2π(ti +2)
+···+ti−1 +1 >
π(ti )
t1 +t2 +···+ti−1 +1 .
Now all that remains is to prove the following lemma.

Lemma- Given positive integers t1 , t2 , · · · , tk we have


t1 t2 t3 tk
+ + + ··· + ≥ ln(t1 + t2 + · · · tk + 1)
1 1 + t1 1 + t1 + t2 1 + t1 + · · · + tk−1
Proof - Let X denote the LHS.
t1 + 1 t2 + t1 + 1 1 + t1 + · · · + tk−1 + tk 1
X+k = + +· · ·+ ≥ k(1+t1 +· · ·+tk−1 +tk ) k
1 t1 + 1 1 + t1 + · · · + tk−1
Now let t = ln(t1 + t2 + · · · tk + 1) , so we have
t t
X ≥ k(e k − 1) ≥ k( ) = t
k

17
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

Where in the last step we used the inequality ex − 1 ≥ x, for x ≥ 0

Now notice that t1 + t2 + · · · + tk = n


So for any configuration of points we can guarantee at least an angle of π ln(n + 1) >
π ln n
And hence f (n) > π ln n.

18
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

C4 A game is played on a rectangular board as follows. Initially, there is one stone


on the bottom-left corner of the board. On each turn, the player may pick up any
stone, remove it, and do one of the following:
Put one stone on each square to the right of the selected square.
Put one stone on each square above the selected square.
Reverse the order of the stones to the right of the selected square.
Reverse the order of the stones above the selected square.
For example, if the board is
2 1 3 2
1 4 0 1
0 6 3 4
3 2 0 3
4 0 0 1
where the numbers represent the number of stones on the square, and the player
chooses the square highlighted in bold, then each of the above operations will
respectively result in the boards

2 1 3 2
1 4 0 1
0 6 3 4
3 1 1 4
4 0 0 1

2 2 3 2
1 5 0 1
0 7 3 4
3 1 0 3
4 0 0 1

2 1 3 2
1 4 0 1
0 6 3 4
3 1 3 0
4 0 0 1

2 6 3 2
1 4 0 1
0 1 3 4
3 1 0 3
4 0 0 1
The player may play as many turns as they wish. The goal of the player is to
maximize the number of stones they have at the end of the game. If two players
play optimally on boards of dimensions 2021 × 2021 and 2022 × 2020 respectively,
which player is going to finish with more stones?
-Ritam Nag
Sketch. The following is only a sketch and the details at the end are left for the
reader to fill out.

19
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

Instead of the given game, we play the related game where instead of putting one
stone on every square to the right/above the chosen stone, the player gets to put
two stones on the square just to the right/above the given square, unless this square
is on the edge, in which case they may put only one.

Lemma 2.3 (The new game is more powerful)


If position Y can be obtained from position X in the original game, it can be
done in the new game.

Proof. Replace every expansion in the original game by repeated expansions in the
new until the expansion reaches the edge of the board. 
Note 2.4. The m in g m (x) denotes that the function g is iterated m times:
g 0 (x) = x, g m+1 (x) = g m (g(x)).
Definition 2.5 (Family of hyperoperations). The sequence of functions fk : N0 →
N0 with k ∈ N is defined recursively by f1 (x) = n + 1, f2 (x) = n + 2,
(
n + 4 if n ≤ 3
f3 (n) =
2n otherwise

and (
fk2 (n) if n ≤ 2
fk+1 (n) =
fkn (0) otherwise

Lemma 2.6 (1D case)


Consider a 1 × m + 1 board, with the squares labelled 0 to m from right to
left, with sk stones on the square k. Then the optimal number of stones is
s
fksk (fk−1
k−1
(. . . (f1s1 (s0 ))))

in both the original and the new game.

Proof. The proof for the new game is trivial by induction and definition of f . For
the original game, note that whenever an expansion is performed, it is performed
until the edge of the board is reached. 
Assign indices to the squares, with the top-right corner being (0, 0) and (i, j) being
the i-th row and j-th column from the top and right respectively.

Lemma 2.7
Both the games are finite.

P i+j
Proof. Assign the ordinal ω s(i, j) to the position with s(i, j) stones at (i, j).
Now observe the ordinal assigned to the position decreases every time a move is
made (in either game), concluding the proof by the well-foundedness of ordinals. 
Definition 2.8 (Obtainability relation). Write A  B for positions A and B if
there exists a sequence of moves starting from A and ending at B and A 6= B.

20
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

Note 2.9.  defines a well-founded partial-order, since the games are finite and
obtainability is transitive.
Definition 2.10 (Best squares). Define v(0) = ω, v(n) = n. Then the best
square is defined to be the smallest square according the lexicographic ordering
on (v(max(i, j), v(min(i, j), i).
Definition 2.11 (Squishing). For a given stone at (i, j), squishing the stone up
refers to performing the optimal sequence of moves on the sub-board given by
{(k, j)|k ≤ i}. Squishing right is defined analogously.

Lemma 2.12 (Optimal Strategy)


Picking a stone on the best square (i, j), and squishing it up if 0 < i < j or
j = 0 or i = j and s(0, j) ≥ s(i, 0), and squishing it right otherwise leads to
an optimal sequence of moves.

Proof. Assume this is not the case. Since  is a well-founded partial order, there
should be a maximal position P for which this is not the case. Suppose the best
stone is at square (i, j). Whatever any optimal move is, it must land some stone at a
square better than (i, j) or at (0, j) or (i, 0), since otherwise by the maximality of P
the optimal move next move would be to move the stone at (i, j) anyway, in which
case the two moves will commute. Let z = s(i, j), x = s(0, 0), a = s(0, j), b = s(i, 0).
Now we make cases:
Case i = 0. Then the only possible form of interference is if (k, 0) squishes up
first, for some k ≥ j. Since fk (fj (x)) ≥ fj (fk (x)), squishing (i, j) right first is
optimal anyway.
Case j = 0. Identical to the above case.
Case wrong direction:
– Sub-case i = j. Assume without loss of generality that a > b. Then the
final value after squishing the top right k × k square in the two cases are
f z (a)+b f z (b)+a) f (b)+f z−1 (a)
fk k (x) and max(fk k , fk k k
(x) respectively. The former
is larger than the latter, so the optimal squishing must be in the stated
direction.
– Sub-case i < j. If the move made is expanding to the right, then after
following the optimal moves after that, we get that the new stone count at
fi2 (0)
(0, 0) is fj−1 (x). But this can be beaten by simply spending one stone
f (0)
upward and ignoring the stones at (0, j), getting the value to fj i (x).
– Sub-case i > j. Identical to the above case.
Case i ≤ j:
– Sub-case increase at (0, j) by swapping. Squishing a and z right, squishing
the result of z up, and then swapping is at least as good.
– Sub-case increase at (0, j) by expansion. Squishing up, then right, then
f z (a) f z (a+2)
swapping yields fj+1 (fj i (x)), while expansion yields fj i (x), the
former of which is better.
– Sub-case swapping/expansion at a. Squishing z to a first and then doing
this is better.
– Sub-case right expansion at (k, j + 1), k < i. Expanding later gives
f (0)
k f z (a) f z (fk2 (a))
fj+1 (fj i (x)), while doing so first gives fj i (x), the former of
which is at least as big.

21
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

– Sub-case up expansion at (i + 1, j − 1) = (j, i). Follows by wrong direction


at (j, i).
Case j ≤ i: Identical to above cases.

By the 1D lemma, squishing can also be performed in the original game, so this
strategy carries over to the original game. Since the new game is dominant over
the original game, this strategy must also be optimal for the original game.
Once we know the optimal strategy, it follows that the optimal number of stones in
the 2021 × 2021 board is f2020 2 (0), while the number in the 2022 × 2020 board is
f2021 (f2019 (0)), the latter of which is obviously larger.

22
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

Geometry
G1 Let H be the orthocentre of acute angled triangle ABC. Suppose that E and F lie
on AB and AC and that circles (BHE) and (CHF ) intersect again at P . Prove
that HP passes through the circumcentre of EAF .
-Pranjal Srivastava
Solution. Let O be the circumcentre of EAF . We will show that both P and O lie
on the angle bisector of EP H. This will prove that HP O are collinear.

O
E

P
B C

First, observe that

∠EP F = 2π − ∠EP H − ∠F P H
= ∠EBH + ∠ECH
π π
= − ∠BAC + − ∠BAC
2 2
= π − 2∠EAF
= π − ∠EOF

which proves that EP F O is a cyclic quadrilateral. Since EO = F O, it follows that


O is the midpoint of arc EF , and lies on the angle bisector of EP F .
Also note that ∠EP H = π − ∠EBH = π − ( π2 − ∠BAC) = π − ∠F CH =
∠F P H.

23
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

G2 Let M be the midpoint of side BC of triangle ABC. Suppose that K and T are
the intersection of the reflection of AM over bisector of angle BAC with (ABC)
and BC respectively. Let ABV C be a parallelogram, and let S be the reflection of
K in BC. If U is the reflection of A in S, prove that line T V passes through the
midpoint of KU .
-Pranjal Srivastava
Solution 1. Let the midpoint of KU be R and AM ∩ ABC = X Now, we will show
by Menelaus on AKU that T, R, V are collinear.

S
O

M
B T C
U

K X

Observe that since KR AT VU


RU = 1, we just want that T K = − V A but observe that
VU MS MX AM AT
V A = M A = − M A . So, we just want AX = AK which is equivalent to XK k BC.

But, observe that AM, AT are isogonal and thus, KX k BC and we are done.

Solution 2. This is a fancy unreasonable solution given just for fun, and also how
the proposer created the problem.
Let midpoint of AK be K 0 . Now, we want the midpoints of BC, AT and SK 0 to
be collinear. Note that K 0 is the A− dumpty point and S is the A− humpty or
HM point.
Observe that ∠K 0 T B = ∠KT C = ∠CT S. Thus, by optical property the inellipse of

24
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

ABC with focii K 0 , S is an inellipse of quadrilateral ABT C. Since the centre of any
inellipse of a quadrilateral is known to lie on the newton-gauss line, the midpoints
of the diagonals i.e. BC and AT and the midpoint of K 0 S are collinear.

25
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

G3 Let I be incentre of 4ABC and let Γ be its circumcircle. Let D, E and F be


tangency points of the incircle of 4ABC with BC, CA and AB respectively. Let
the circle with centre A passing through I meet Γ at points A1 and A2 . Similarly
define B1 , B2 and C1 , C2 . Let A0 , B 0 and C 0 be feet of perpendiculars from I onto
A1 A2 , B1 B2 and C1 C2 respectively. Prove that A0 D, B 0 E and C 0 F are concurrent.
-Mahavir Gandhi
Note. Finally, an easier version was asked. The concurrency instead demanded
was AA0 , BB 0 , CC 0 .
Solution. Let ω be the incircle.
The following claim is central to the problem and we present two proofs of this
fact.

Claim — A1 A2 is tangent to ω.

Proof 1. Let MA be the midpoint of arc BC not containing A. Since, A is midpoint


of arc A1 A2 , by trident lemma or fact 5, I is the incenter of A1 A2 MA . We wish
to show that the inradii of ABC and MA A1 A2 are equal. For this, observe that
MA B AA1
sin ∠MA AB = 2R = sin ∠AMA A1 , so AI sin ∠IAB = MA I sin ∠AMA A1 which is
equivalent to the two inradii being equal.
Proof 2. Perform inversion from A with radius AI. Observe that this sends A1 A2
to (ABC) and the ω to the A− mixtilinear incircle and the claim follows.
Proof 3. As above, let MA be the midpoint of arc BC not containing A.
By Poncelet’s porism, there is a triangle with vertex MA and incircle ω and
circumcircle Γ. Now, let it’s two tangents to ω intersect Γ at A01 , A02 . Now, the
midpoint of arc A01 , A02 must be A. Thus, by fact 5, we have AA01 = AA02 = AI.
Thus, A01 ≡ A1 , A02 ≡ A2 and A1 A2 is tangent to ω.
Thus, A0 is the point at which A1 A2 is tangent to ω. Let O be the center of Γ, we
have OA ⊥ A1 A2 since both lie on its perpendicular bisector. Thus, we also have
OA k IA0 .

Now, let X be the ex-similicenter of Γ and ω. The homothety taking ω to Γ takes


I to O.
Observe that it must take A0 to a point Xa on Γ such that Xa O k IA0 but this
point is A for the positive homothety. Similarly, this homothety takes B to B 0 and
C to C 0 . It also takes D to MA since the tangent at D to ω and the tangent to Γ
at Ma are parallel.

26
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

A1
A
MB

A0
E

MC
A2 F X

I
B0 C0

B D C

MA

Thus, this homothety creates the following maps-


I 7→ O
A0 7→ A
B 0 7→ B
C 0 7→ C
D 7→ MA
E 7→ MB
F 7→ MC
Thus, AA0 , BB 0 , CC 0 concur at X and A0 D, B 0 E, C 0 F concur iff AMA , BMB , CMC
concur, but these lines concur at I and we are done.

27
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

G4 Let ABC be a scalene triangle with circumcircle Ω and circumcenter O. Let the A
symmedian1 intersect the circle Ω at D. Let K be the midpoint of AD. Let ω be a
circle through A and K tangent to the circumcircle of B, O and C. Let ω intersect
AC, AB and Ω at X, Y, Z such that X, Y, Z 6= A. Prove that AZ, XY concur on
BC.
-Rohan Goyal
Solution. Observe that the problem is equivalent to showing that BCXY is cyclic(by
radical axes on Ω, ω, (BCXY )).

O
Y0
K
Y
A0

B C

D
X0

Claim — The center of ω lies on the A altitude.

Proof. There is a homothety that takes (BOC) to ω from K. But, T = AD∩(BOC)


is the intersection of the tangents from B, C to Ω. But OT is a diameter of (BOC).
Thus, the homothety from K takes OT to a diameter of ω but this line is parallel
to OT and passes through A, thus it is altitude and we get the claim.
Now, let A0 be the A antipode in ω. Now, let C 6= Y 0 = (BOC) ∩ AC. Now,

∠OT Y 0 = ∠OCY 0 = ∠BAA0 = ∠Y AA0

and we have AA0 and OT as diameters of the two circles. So, we get that the
homothety taking ω to (BOC) takes Y to Y 0 and similarly X to X 0 . Thus,
1
The A symmedian is the line joining A and the intersection of the B and C tangents to Ω

28
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

XY k X 0 Y 0 .
Now, by Reim’s theorem as B, C, X 0 , Y 0 are concyclic, we get that B, C, X, Y are
concyclic and we are done.

29
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

G5 Let I and IA be incentre and excentre opposite to A of 4ABC respectively and


let L be midpoint of arc BAC. Let LI and LIA meet (ABC) again at points
Y and Z respectively. Prove that if AX is internal angle bisector of ∠BAC with
X ∈ BC then one of the external common tangents of (ABC) and (XY Z)
passes through antipode of A in (ABC).
-Mahavir Gandhi
Solution. Let AL ∩ BC = L0 . D be the incircle touch point on BC and D0 be the
A extouch point and let D00 be the diametrically opposite point of D of the incircle.
MA be the arc midpoint of BC not containing A. Let U be the foot of A perpen-
√ BC.
dicular on
Perform bc inversion and reflection.

L0 C
B D X D0
Y
Z

IA

Observe that the following maps happen-


(ABC) 7→ BC
I 7→ IA
L 7→ L0
IA L 7→ (AIL0 ) =⇒ Z 7→ D.
Y 7→ D0
X 7→ IA
Tangent to (ABC) at antipode of A maps to (AU )

So, we just want to show that (AU ) is tangent to (DD0 MA ).


Let S be the foot of perpendicular from U to AI and T be the intersection of
perpendicular bisector of BC and U S.

30
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

D00

T
S

B U D X M D0 C

MA

Observe that U S · U T = U M · U X

Claim — (U X; DD0 ) = −1

Proof.
A
(U X; DD0 ) = (∞DI I; DD00 ) = −1

Thus, U S · U T = U X · U M = U D · U D0 . Thus, ST DD0 is cyclic.

But since, MA is on perpendicular bisector of DD0 and ASMA collinear. We get


that ST DD0 MA cyclic and MA is the antipode of T . Thus, there is a homothety
that takes (AU ) to (T DD0 MA ) and thus both circles are tangent.

31
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

G6 Let P, Q be chosen inside triangle ABC such that ∠P BC = ∠ABQ and ∠P CB =


∠ACQ. The tangent to (BP C) at P and the tangent to (BQC) at Q intersect at
R. Prove that circles (P QR) and (ABC), and line AR share a common point.
-Mahavir Gandhi
Solution. Let BP, CQ intersect at M and BQ, CP intersect at N .
Let L be the miquel point of P M QN , that is, the point of concurrency of circles
(BP N ), (BQM ), (CP M ), (CQN ).
Then we will show that this point lies on (ABC) and (P QR).

K
P

M N
Q
T
R
B C

First we show for (ABC),

∠BLC = ∠BLM +∠M LC = ∠BQM +180−∠M P C = 180−∠BQC+180−∠BP C

= ∠QBC + ∠QCB + ∠P BC + ∠P CB = ∠QBC + ∠QCB + ∠ABQ + ∠ACQ


= ∠ABC + ∠ACB = 180 − ∠BAC.
Hence L lies on (ABC).
Now for (P QR),

∠P LQ = ∠P LC − ∠CLQ = ∠P M C − ∠QN P

= ∠P BC + ∠QCB − ∠QBC − ∠P CB
Now if P B ∩ QR = T

∠P RQ = ∠P T Q − ∠BP R = ∠P BQ + ∠RQB − ∠RP B

32
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

As R lies on those tangents ∠RQB = ∠QCB, ∠RP B = ∠P CB

=⇒ ∠P BQ + ∠RQB − ∠RP B = ∠P BC − ∠QBC + ∠QCB − ∠P CB

And hence we get ∠P LQ = ∠P RQ, which gives that L lies on (P QR).


Now let AR intersect (ABC) at K, we want to show that K lies on (P QR), that
is ∠RKL = ∠RP L

∠RKL = 180 − ∠AKL = ∠ACL = ∠ACQ + ∠QCL

= ∠P CB + ∠BP L = ∠RP B + ∠BP L = ∠RP L


Hence Proved .

33
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

G7 Bob marks three special points A, B, C on the plane, such that the euler line of
ABC is not parallel to BC. Given a special point P (ie P is one of A, B, C), and
an arbitrary point Q, Alice can draw in line P Q and the circle centred at P that
passes through Q. Prove that Alice can construct the intersection of the euler line
of ABC with BC.
-Pranjal Srivastava
Note. We found this too similar to a pre-existing AoPS thread, so while I’ve left
it in the shortlist, it was withdrawn from selection for the actual mock.
Solution. We claim the Alice can use the following construction. Mark points
X and Y on AB and AC such that CX = CB and BY = BC. Mark K, the
intersection of CX and BY . Then, KA ∩ BC is the required point.
Y
A

K X

P B D M C

O0

T
We now give a proof that this construction works.
Let I be the incentre of BKC, and let O be the circumcentre of ABC. Observe
that I is also the orthocentre of ABC. Let the feet of perpendiculars from I and O
to BC be D and M respectively. We will show that the respective sides of triangles
KID and AOM are parallel to each other. This will imply that lines AK, OI and
M D concurr, which is what we want to show.
Since OM ⊥ BC and ID ⊥ BC, we have OM ||ID
∠BKI + ∠KBA = (π − ∠BAC) + ( π2 − ∠ACB − ∠ABC) = π
2 − ∠ACB = ∠BAO,
which implies that OA||KI

34
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

Let O0 be the reflection of O in BC, and let T be the reflection of I across O0 .


Observe that AI = 2M O = M O0 , so T is also the reflection of A across M . Also,
since ∠BO0 C +∠BKC = 2∠BAC +(2π −2∠BAC) = π which implies that BKCO0
is concyclic. O0 is the midpoint of arc BC in (KBC), so T is the excentre of KBC.
By a well known result, T M ||DK, but since T was simply the reflection of A across
M , AM ||T K. This proves the desired conclusion.

35
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

Number Theory
N1 Let p1 < p2 < p3 · · · be the sequence of all prime numbers. Suppose we have
a set S ⊆ N such that for all natural i, S constains a number in the interval
[p2021i , p2021(i+1) ). Prove that infinitely many primes divide at least one element of
S
-Pranjal Srivastava
Solution 1. Assume for the sake of contradiction, that only finitely many primes
divide some element of S. Let these primes be q1 , . . . , qn , and let T denote the set
of natural numbers, all of whose prime factors lie in {q1 , . . . , qn } Observe that
 
X1 X1 Y X
≤ =  qj−i 
s t
s∈S t∈T j i≥0

P 1
In particular, s∈S s doesn’t diverge.
On the other hand,
X 1 X 1 X 1 X1
≤ + 2021 ≤ 2022 +
pi pi p2021i+1 s
i≤2022 s∈S

This means that the sum of reciprocals of primes converges, which is a contradiction
- a result of Euler states the sum of reciprocals of primes diverges.

36
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

N2 Let k be a positive integer such that p = 2k − 1 is a prime. Let P be an integer


polynomial satisfying
π
P (sin(θ + 2k−1 )) = P (sin(θ))
for all θ ∈ R. Define the sequence (ai )i≥0 by a0 = 1, a1 = 2, and

an+1 = 4an − an−1

Show that for any i, j we have p | P (ai ) − P (aj ).


-Ritam Nag
Solution. Let T0 (X) = 1, T1 (X) = X, Tn+1 (X) = 2XTn (X) − Tn−1 (X) be the
Chebyshev polynomials. It is well known (follows from induction) that Tn (cos θ) =
cos nθ. From this, it follows that the roots of Tn (X) − 1 are given precisely by
the numbers cosk 2πn , including multiplicity (differentiate the last equation with
respect to θ to get the multiplicity. It also follows that Tm (Tn (X)) = Tmn (X), by
subtracting one from both sides and comparing roots.
Now observe that T2k (X)−1|P (X)−P (1) by comparing roots [roots with multiplicity
two can be seen by differentiating the given equation with respect to θ]. Also
observe that an = Tn (2) by induction. Note that 12 ((2 + X)n + (2 − X)n ) is an even
polynomial, so let gn (X 2 ) = 12 ((2 + X)n + (2 − X)n ), and observe that an =
 gn (3).
Now, g2k (X 2 ) = 12 ((2 + X)p+1 + (2 − X)p+1 ) ≡ 4 + X p+1 (mod p). As 3
p = −1,
we have a2k ≡ 1 (mod 2k ), so T2k (2) − 1 ≡ 0 (mod p). Noting that Tn (1) = 1,
we have 1 ≡ Tn (T2k (2)) ≡ T2k (Tn (2)) ≡ T2k (an ) (mod p) =⇒ P (an ) ≡ P (1)
(mod p), concluding the proof.

37
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

N3 Find the least positive integer k that satisfies the following:


For any monic polynomial P (x) of degree 2021 with integer coefficients, there exists
a set T (dependent on P ) of k integers, such that there is no set S which satisfies
the following three conditions simultaneously:
a) T ⊆ S
b) S is proper subset of Z
c) If u, v ∈ S, then P (u) + v ∈ S
-Anish Kulkarni
Solution 1. We claim that the answer is (2021)! .
For P (x) ∈ Z[x] let gcd(P ) denote the largest positive integer g such that g divides
P (a) for all a ∈ Z.
Call a subset S of Z , P-nice if u, v ∈ S =⇒ P (u) + v ∈ S.
For each polynomial P , let Λ(P ) be the smallest possible size of set T such that
there does not exist a P −nice set S which contains T and is a proper subset of Z.
Then one needs to find maximum of Λ(P ) over all monic polynomials P of degree
2021.

We will complete the problem in two steps .


First we will show for any polynomial of odd degree Λ(P ) = max(gcd(P ), 2).
And next we will show that gcd(P ) ≤ deg(P )!.

Now for first step fix a polynomial P (x) with gcd(P ) = g.


Let Pg be the set of primes dividing g.

Claim 1- There exist u with νp (P (u)) = νp (g) for all p ∈ P .


Proof - Let p ∈ P, νp (g) = α.
Now if P (x) is divisible by pα+1 for all x, then pα+1 divides g which is a contradiction.
Hence there exists some up such that P (up ) is not divisible by pα+1 .
Now define u such that u ≡ up (mod pα+1 ) =⇒ p(u) ≡ p(up )(mod pα+1 ) for all
p ∈ Pg .
We see that this u indeed satisfies our requirement.

Claim 2-If g > 1, there exist u, v with u 6≡ v(mod g) and gcd(P (u).P (v)) = g.
Proof - Chose u from claim 1. Let Pu be set of primes dividing P (u) but not g.
Let p ∈ P , now p does not divide g implies there exist a vp such that p does not
divide P (vp ).
Now chose a v such that v ≡ vp (mod p) for each p ∈ Pu and v ≡ u + 1(mod g).
Now note that any prime p dividing P (u) but not g does not divide P (v).
Hence gcd(P (u), P (v)) is only divisible by primes which divide g.
But we have chosen u such that νp (P (u)) = νp (g) for all p ∈ P .
Hence gcd(P (u), P (v)) divides g and hence is equal to g, as g divides both.

Claim 3- There exist u, v satisfying condition of Claim 2 and P (u), P (v) have
different sign.
Proof - We take v 0 = v + rP (u), so v 0 ≡ v(mod g).
And P (v 0 ) ≡ P (v)(mod P (u)) and hence gcd(P (u), P (v 0 )) = gcd(P (u), P (v)) = g.
Now as P has odd degree we can chose a r, so that P (u), P (v 0 ) have diff sign.
(Note if P (u) = 0, then u cannot satisfy claim 1.And hence that’s not possible.)

Claim 4- If we have integers p, q of opposite signs and gcd(p, q) = g then for any

38
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

integer n divisible by g we have positive integers a, b such that ap + bq = n.


Proof - By Bezout’s theorem there exits integers r, s such that rp + sq = n.
Now let p < 0, q > 0, so we have (r + ql)p + (s − pl)q = n.
And we get required a, b by taking large enough l.

Claim 5- If t, u, v ∈ S where S is P −nice then for any positive integers a, b, we


have t + aP (u) + bP (v) ∈ S
Proof - This follows straightaway from repeatedly applying definition of a P −nice
set.

Claim 6-Λ(P ) = max(g, 2)


Proof -If g = 1, we find u, v with relatively prime P (u) and P (v) and apply idea
of claim 3 to get them such that P (u) and P (v) have opposite sign.
Now let S be a P −nice set which has u, v.
Now by claim 4 , for any n we have positive integers a, b such that n − u =
aP (u) + bP (v) and hence by claim 5 n ∈ S.
And hence S = Z is not a proper subset of Z, so the set set T = {u, v} works.
It is easy to see that no singleton set T works and hence Λ(P ) = 2.
Now let g > 1.We have u, v satisfying condition of Claim 2 and 3.
Now take a1 = u, a2 = v, a3 · · · ag such that {a1 , a2 , · · · ag } is a complete residue set
modulo g.
Now let T = {a1 , a2 , · · · ag } and let S be a P −nice set containing T .
Now let n be any integer, let n ≡ at (mod g).
Then by Claim 4 there exist positive integers y, z such that n − at = yP (u) + zP (v).
And hence n = at + yP (u) + zP (v) =⇒ n ∈ S and hence S = Z. Hence this T
works.
If T had less than g elements we may take S to be set of all integers congruent
to some element of T modulo g, and we would have a P −nice proper subset of Z
containing T .
Hence Λ(P ) = g
Hence proved .

And finally we show the following lemma


Lemma - For any non-constant monic polynomial P (x), gcd(P ) ≤ deg(P )!
Proof - Let deg(P )=n, gcd(P )=g ,[x]k = (x − 1)(x − 2) · · · (x − k).
Let P (x) = [x]n + an−1 [x]n−1 + · · · + a1 [x]1 + a0
Now we prove by induction that g divides ak k! for 0 ≤ k ≤ n.
Basis step - g divides P (1) = a0 = a0 0!.
Assume it is true for all c < t.
As c! divides [x]c for all x, we have g divides ac [x]c for all x and c < t.
Now P (t + 1) = at t! + at−1 [t + 1]t−1 + · · · + a1 [t + 1]1 + a0 and g divides this.
Hence g divides at t!.
Continuing we get that g divides n!. and hence g ≤ n!.
Hence proved.

From this we get that k ≤ (2021)!


Now to end the problem take P (x) = (x − 1)(x − 2) · · · (x − 2020)(x − 2021).
Note that gcd(P ) = 2021! and hence we have Λ(P ) = 2021! =⇒ k ≥ 2021!.
Hence k = 2021!.
QED.

39
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

Solution 2. We use method of finite differences to show that the answer is k = 2021!.
Lemma - Let P (x) be a monic polynomial of degree n then for any integer x,
gcd(P (x), P (x + 1), · · · P (x + n)) divides n!.
Proof - Let ∆0 (x) = P (x) and ∆i (x) = ∆i−1 (x + 1) − ∆i−1 (x).
Let di , ai denote the degree and leading coefficient of ∆i (x)
Then we have di = di−1 − 1 and ai = ai−1 di−1 .
Hence we have that ∆n (x) is a constant polynomial equal to n!.
But also note that ∆n (x) can be written as a linear combination of P (x), P (x +
1), · · · P (x + n).
Hence gcd(P (x), P (x + 1), · · · P (x + n)) divides n!

Now let P (x) be monic polynomial of degree 2021.


As P (x) has an odd degree we can chose a s such that there is a positive as well as
negative number in P (s), P (s + 1), · · · P (s + 2021) .
Let g = gcd(P (s), P (s + 1), · · · P (s + 2021)), so g divides 2021!.
Now by generalized Bezout’s Theorem for any integer n divisible by g we have
integers a0 , a1 , · · · , a2021 such that

n = a0 P (s) + a1 P (s + 1) + · · · + a2021 P (s + 2021)

Now we give a procedure to keep the above condition true while making all ai
positive .
Let ai < 0, if P (s + i) = 0 we can just set ai = 1.
If P (s + i) > 0, let j be such that P (s + j) < 0.
Replace ai by ai − yP (s + j) and aj by aj + yP (s + i).
Note that by choosing a big enough y we can make ai positive while increasing aj .
Similarly if P (s + i) < 0, we find a j such that P (s + j) > 0.
And do the following ai → ai + yP (s + j), aj → aj − yP (s + i). Hence we would
eventually get all ai ≥ 0 .
Now chose T = {s, s + 1, s + 2, · · · , s + 2021! − 1}.
Let S be a set containing T such that if u, v ∈ S then P (u) + v ∈ S.
Now consider any integer n, let n ≡ s + t(mod 2021!), where 0 ≤ t < 2021!.
So n − s − t is divisible by 2021! and hence by g and hence there exist positive
integers a0 , a1 , · · · , a2021 such that

n − s − t = a0 P (s) + a1 P (s + 1) + · · · + a2021 P (s + 2021)

=⇒ n = s + t + a0 P (s) + a1 P (s + 1) + · · · + a2021 P (s + 2021)


Now s + t, s ∈ S =⇒ s + t + P (s) ∈ S
=⇒ s + t + P (s) + P (s) = s + t + 2P (s) ∈ S =⇒ ..... =⇒ s + t + a0 P (s) ∈ S.
And we do same for s + 1, · · · , s + 2021, till we get n ∈ S
Hence S = Z. So we must have k ≤ 2021!
Now consider P (x) = (x − 1)(x − 2)(x − 3) · · · (x − 2021), so 2021!|P (x) for all x.
Let T be a set having less than 2021!, we consider S to be set of all integers
congruent to some element of T modulo 2021!.
Now for any u, v ∈ S, P (u) + v ≡ v(mod 2021!) and hence lies in S too. This
contradicts the condition of the problem.
Hence k ≥ 2021! and so we have k = 2021!.
Hence Proved.

40
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

N4 Given an odd
j prime
k q, find all integer polynomials P such that every prime p > q 2021 ,
p
divides P (q q
), and P (1) = 0.
-Rohan Goyal

Remark. The condition P (1) = 0 is not necessary but helps avoid some heavy
machinery.

Solution. For simplicity, we consider where P is non constant, else it would be 0


polynomial.
Let α1 , · · · αi be the roots of P .
n
Now, let Q(x) = k (xq − αiq ). Observe that this is also an integer polynomial
Q
i=1
divisible by P when k is a large enough constant.

Now, let p jbek a prime congruent to a (mod q). Let Q(x) = S(xq ). Now, we have
p
that p|P (q q )|S(q p−a ) =⇒ p|S(q 1−a ) but by Dirichlet’s theorem, we can find
arbitrarily large p ≡ a (mod q). Thus, S(q 1−a ) = 0. Thus, if we consider a from 2
to q − 1. We get that 1, 1q , q12 , · · · qq−2
1
are roots of S.
q−2
Thus, Q(x) = R(xq ) (xq − q −i ).
Q
i=0

Now, since factors of Q are written in the form, (xq − αiq ), thus, q −i is of the form
αiq . Thus, atleast one of the roots of xq − αiq is a root of P . But, all the roots
of this polynomial has minimal polynomial xq − q −i for i > 0. Thus, we get that
(x − q −i )|P =⇒ (q i x − 1)|P for i in {1, 2, · · · q − 2}.

Now, we are concerned with the part where i = 0. Now we have xq − 1|Q.
Observe that Φqr (q) ≡ 1−q (mod q 2 ) for r prime. Let, s be a prime factor of Φqr (q)
not congruent to 1 (mod q 2 ). Now, ords (q) = qr. But since q 2 6 |s−1, we have that q
is not a q th power (mod s). We can now get infinitely many s by picking different r.

Thus, we get that there are infinitely many primes p such that q is not a q th power
(mod p). Thus, for infinitely many primes we have different factors of xq − 1 other
than 1 divide P . Thus, some root of Φq (x) is also a root of P . Thus, Φq (x)|P .

Now, we are given that P (1) = 0. Thus, x − 1|P . Thus, we get that
q−2
Y
(q i xq − 1)|P
i=0

q−2
(q i xq − 1) for any integer polynomial
Q
We can now claim that any P (x) = H(x)
i=0
H works.

j kLet p be an arbitrary
We can now check that all such polynomials work. j k large
p p
q
prime such that p − 1 ≡ a (mod q). Thus, p|(q a q q
− 1). Thus, p|P (q q
) and
we are done.

41
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

Note. We can now even show that P (1) = 0 is unnecessary. For that we need to
show that for infinitely many primes p ≡ 1 (mod q), and q is a q th power.
For that, we can use the following theorem.

For any polynomial P ∈ Z[x], let S(P ) = {p|p prime, ∃n ∈ N s.t. p | P (n)}

Theorem
For any P, Q ∈ Z[x], we have |S(P ) ∩ S(Q)| is infinite.

Assuming this theorem, observe that we can conclude by letting P = xq − q and


Q = Φq (x).

Now, we provide the proof of the theorem.


Proof. Let α be a root of P and β be a root of Q.

Consider Q(α, β). Now, by Primitive element theorem, we have Q(α, β) = Q(γ).
Now, let P1 (γ) = α, Q1 (γ) = β.

Let the minimal polynomial of γ be R. Thus, we have that if p ∈ S(R).

Consider P (P1 (x)) and Q(Q1 (x)). Observe that γ is a root of both these polynomials.
Thus, R(x)|P (P1 (x)) and R(x)|Q(Q1 (x)).
Thus, S(R) ⊂ S(P ) ∩ S(Q) and we are done and since by Schur’s we have |S(R)|
is infinite, we are done.

42
Anish, Mahavir, Ojas, Pranjal, Ritam, Rohan and Tejas (May 11, 2021) LMAO SL 2021

§3 Acknowledgements
The exam and the shortlist were compiled and all problems were written by our team, M
Raptor consisting of the following 7 members.

Mahavir Gandhi
Rohan Goyal
Anish Kulkarni
Pranjal Srivastava
Tejas Mittal
Ojas Mittal
Ritam Nag

We would like to thank the testsolvers Bhavya Agrawalla, Shantanu Nene and
Shourya Pandey.

Also, we would like to thank all who are helping us grade the tests.

And finally we thank everyone who took out the time to take the tests xD.

43

You might also like