Solutions to Practice Problems 4

1. Let ABC be a triangle. Let A1 , A2 be points on AB and AC respectively such that

A1 A2 k BC and the circumcircle of 4AA1 A2 is tangent to BC at A3 . Define B3 , C3
similarly. Prove that AA3 , BB3 , and CC3 are concurrent.
Solution. By the alternate segment theorem, ∠BA3 A1 = ∠BAA3 . On the other hand,
since A1 A2 k BC, we have ∠BA3 A1 = ∠A3 A1 A2 = ∠A3 AC. Thus, AA3 is the angle
bisector of ∠BAC. Similarly BB3 and CC3 are the angle bisectors of ∠ABC and ∠ACB
respectively. Therefore, AA3 , BB3 , CC3 are concurrent at the incentre of 4ABC.

2. Let f : R → R be a function satisfying f (x)f (y) = f (x − y) for all x, y ∈ R. Find all

possible values of f (2020).
Solution. First note that f (0)2 = f (0) says f (0) = 0 or 1. If f (0) = 0, then we get
f (x) = f (x)f (0) = 0 for all x, which is indeed a valid solution. Now let f (0) = 1.
Then f (x)f (x) = f (0) = 1 implies that for each x ∈ R we must have f (x) = ±1. It
is easy to note that f (x) = 1 for all x is a valid solution. What if f (x) = −1 for some
x? This can be dealt with in various ways. For instance, f (2x)f (x) = f (x) implies
f (2x) = 1 for every x ∈ R, which essentially means f (x) ≡ 1. Another way is to note
that f (−x) = f (0)f (x) = f (x)f (0) = f (x) and hence f (x) = f ( x2 )f (− x2 ) = f ( x2 )2 = 1 for
all x ∈ R. Answer: either f (x) ≡ 0 or f (x) ≡ 1.

3. Let AD be the altitude from vertex A onto BC. From D, drop perpendiculars DD1
and DD2 onto CA and AB respectively. Let `a be the length of D1 D2 . Define `b and `c
analogously. Prove that `a = `b = `c .
Solution. We begin with the observation that AD1 DD2 is a cyclic quadrilateral and
AD is the diameter of its circumcircle. Hence we get from (extended) sine rule that
`a = D1 D2 = AD sin ∠D1 AD2 = AD sin A. Since BD = c cos B (in usual notation) and
AD = BD tan B, we have AD = c sin B = 2R sin B sin C, where R is the circumradius
of 4ABC. Hence `a = 2R sin A sin B sin C. It is immediate from the symmetry of this
expression that similar calculation would lead us to same values for `b and `c .

4. Let a, b, c be positive integers such that a2 − bc is a perfect square. Prove that 2a + b + c

is not a prime.
Solution. If a2 − bc = k 2 then (a − k)(a + k) = bc. Let gcd(a − k, b) = x, say b = xy
and a − k = xz with gcd(y, z) = 1. Then xz(a + k) = xyc, so (a + k)z = cy. The left
side is divisible by z, so the right side must be too. Hence c must be divisible by z
because y and z have no common factors. Hence c = wz for some w, which implies
(a + k)z = wyz, or, a + k = wy. Now note that, 2a + b + c = (a + k) + (a − k) + b + c =
wy + xz + xy + wz = (w + x)(y + z), which cannot be a prime, since w, x, y, z ≥ 1.

1
5. Find all triples (f, g, h) of injective functions from R to R which satisfy

f (x + f (y)) = g(x) + h(y), g(x + g(y)) = h(x) + f (y), h(x + h(y)) = f (x) + g(y)

for all real numbers x and y.

Solution. We start with the first equation f (x + f (y)) = g(x) + h(y). We replace x
by x + g(z) so that in the LHS we can have g(x + g(z)), which can be substituted by
h(x) + f (z) using the second equation. Thus we obtain

f (x + g(z) + f (y)) = g(x + g(z)) + h(y) = h(x) + f (z) + h(y). (1)

Now we switch the roles of x and y here, which gives

f (y + g(z) + f (x)) = g(y + g(z)) + h(x) = h(y) + f (z) + h(x). (2)

Since the RHS of equations (1) and (2) are same, we equate their LHS and use the
injectivity of f to conclude that x + g(z) + f (y) = y + g(z) + f (x). This implies f (x) − x =
f (y) − y for every x, y ∈ R, which tells us that f (x) − x is a constant function. Say,
f (x) = x + a for all x ∈ R.
Next observe that by the symmetry of the problem, we can apply similar arguments
for g and h to conclude that g(x) − x and h(x) − x are also constant functions. Let
g(x) = x + b and h(x) = x + c for all x ∈ R. Plugging these into the original three
equations we get 2a = b + c, 2b = c + a, 2c = a + b. These clearly gives a = b = c. It
is straightforward to check that f (x) = g(x) = h(x) = x + a indeed satisfy the given
equations, for any value of the constant a.

6. A polynomial P (x) with non-negative real coefficients has n real roots. Moreover, let
the leading coefficient and constant term both be 1. Find the minimum value of P (k)
for any positive integer k.
Solution. First note that all the roots must be negative. Let the roots be −r1 , . . . , −rn
where r1 , . . . , rn > 0. We can write P (x) = (x + r1 ) · · · (x + rn ) and observe that
r1 r2 · · · rn = 1. Now fix any k ∈ N. Applying weighted AM-GM inequality (with 1
taken k times and ri taken once) we get

(1 × k + ri × 1) 1/(k+1)
≥ (1 × · · · × 1 × ri )1/(k+1) = ri for each 1 ≤ i ≤ n.
k+1
n
Y n
Y 1/(k+1)
n
Therefore, P (k) = (k + ri ) ≥ (k + 1) ri = (k + 1)n . This maximum value
i=1 i=1
is attained when all the roots are equal, i.e. when P (x) = (x + 1)n .

2
 Remark. We can in fact prove that if P (x) = nj=0 aj xj then aj ≥ n
P

j
. To show this, use
Vieta’s formulae along with AM-GM inequality:
 −1  −1 1/(mn )
n n X  Y
am = ri1 ri2 · · · rim ≥ ri1 ri2 · · · rim = 1.
m m 1≤i <···<i
1 m ≤n 1≤i1 <···<im ≤n

The last equality can be justified either using the symmetry, or by observing that the
n−1

n

exponent of ri in the big product in the RHS equals m−1 / m = m/n, so that big
Qn m/n
product above equals just ( i=1 ri ) = 1.

7. Suppose that P is a polynomial with integer coefficients such that n divides P (2n ) for
every positive integer n. Prove that P must be the zero polynomial.
Solution. Let p and q be two odd primes. It is given that pq | P (2pq ) . We have by
Fermat’s little theorem 2pq ≡ (2q )p ≡ 2q (mod p). And we know the following property
of integer polynomials: a − b | P (a) − P (b). Combining these we get

P (2pq ) ≡ P (2q ) (mod p) =⇒ p | P (2q ) (since p | P (2pq )).

Now in the last divisibility relation, taking p large enough forces P (2q ) = 0. But this
is true for arbitrary odd prime q, so we get P (x) = 0 for infinitely many x. This forces
that P is identically zero, as desired.

8. The numbers from 1 through 2020 are written on a blackboard. Every second, Dr. Math
erases four numbers of the form a, b, c, a+b+c (of course if such numbers are available)
and replaces them with the numbers a + b, b + c, c + a. Prove that this cannot continue
for more than 10 minutes.
Solution. First observe that a + b + c + (a + b + c) = (a + b) + (b + c) + (c + a) and
a2 + b2 + c2 + (a + b + c)2 = (a + b)2 + (b + c)2 + (c + a)2 . So the sum of the numbers on the
blackboard as well as the sum of their squares remain invariant throughout the process.
But the number of numbers remaining on the blackboard reduces by 1 at each step.
Initially there was N = 2020 numbers, after m moves there will be n = N − m many
numbers remaining. At this stage let us label the remaining numbers as a1 , a2 , . . . , an .
It follows from the above observations that
n N n N
X X N (N + 1) X
2
X N (N + 1)(2N + 1)
ai = j= and ai = j2 = . (∗)
i=1 j=1
2 i=1 j=1
6

ai and a2i . How can we do that?

P P
Next, we want to connect the two sums
We have by the Cauchy-Schwarz inequality n ni=1 a2i ≥ ( ni=1 ai )2 , which combines
P P

3
 with (∗) to produce

N (N + 1)(2N + 1)  N (N + 1) 2 3N (N + 1)
n· ≥ =⇒ N − m = n ≥ .
6 2 2(2N + 1)
N (N −1)
This provides an upper bound on the no. of moves: m ≤ 2(2N +1)
= 504.6. Thus the
number of moves cannot be more than 504. Hence Dr. Math cannot play the game for
more than 10 minutes (since it requires at least 600 moves).

9. Consider a 6 × 6 grid. Define a diagonal to be the six squares whose coordinates (i, j)
(1 ≤ i, j ≤ 6) satisfy i − j ≡ k (mod 6) for some k = 0, 1, . . . , 5. Hence there are six
diagonals. Determine if it is possible to fill it with the numbers 1, 2, . . . , 36 (each exactly
once) such that each row, each column, and each of the six diagonals has the same sum.
Solution. (due to Evan Chen) Let us assume that the grid has been
filled with the numbers in the desired manner. Observe that the
common sum must be 111. Color the grid as shown. Let A, B, C
be the sum of the numbers at squares colored red, blue and green
respectively. Note that adding up the 1st, 3rd and 5th rows we get
A + B = 333. Similarly, adding the 2nd, 4th and 6th columns we get B + C = 333,
and adding 1st, 3rd and 5th diagonals we get C + A = 333. But this set of equations
does not admit a solution for A, B, C ∈ N. Therefore the grid cannot be filled up in the
desired manner.

10. Consider infinite sequences a1 , a2 , . . . of positive integers beginning with a1 = 1 and

such that an divides ak + ak+1 + · · · + ak+n−1 for every k, n ∈ N. For a given positive
integer m, find the maximum possible value of a2m .
Solution. We have a2 | a1 + a2 = 1 + a2 =⇒ a2 = 1. Next, fix a positive integer n, and
denote sn = a1 +· · ·+an . Then an+1 | sn , an+1 | sn −a1 +an+2 =⇒ an+2 ≡ 1 (mod an+1 ).
This implies gcd(an+2 , an+1 ) = 1. Again, an+2 | sn + an+1 . Since an+1 and an+2 are
coprime, we can combine this with an+1 | sn to arrive at an+2 an+1 | sn + an+1 . Now
this provides us a bound: an+2 an+1 ≤ sn + an+1 , or, an+1 (an+2 − 1) ≤ sn . Since ai ’s are
positive integers, we get an+1 + an+2 − 1 ≤ an+1 (an+2 − 1) + 1 ≤ sn + 1.
Thus we have shown that an+1 + an+2 ≤ ni=1 ai + 2 holds for every n ≥ 1. Starting
P

with a1 + a2 = 2, we inductively get that a2m−1 + a2m ≤ 2m for every m ≥ 1. This gives
an upper bound on a2m , namely a2m ≤ 2m − a2m−1 ≤ 2m − 1. And we can easily achieve
this upper bound, by setting a2m−1 = 1, a2m = 2m −1 for each m ≥ 1. It remains to show
that the sequence an defined in this manner satisfy the divisibility condition, which is
easy to check (left an exercise for you).