You are on page 1of 4

Solutions for Problem Set 7

1. 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 (because x 7→ 2x is a bijective map
here). 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.

2. 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 .

3. 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

1
n
Y n
Y 1/(k+1)
Therefore, P (k) = (k + ri ) ≥ (k + 1)n 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 .

Remark. We can in fact prove that if P (x) = nj=0 aj xj then aj ≥ nj . To show this, use
P 

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 <···<i1 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
product above equals just ( i=1 ri )m/n = 1.

4. Let f : [a, b] → R be twice differentiable, satisfy f (x) = f 0 (x) + f 00 (x) for each x ∈ [a, b]
and f (a) = f (b) = 0. Find all such function f.

Solution. First assume that f is not identically 0. Then there exists p, q such that f (p) =
f (q) = 0 and f (x) 6= 0 for x ∈ (p, q). Since f is continuous and f 6= 0 on (p, q), f must
be either positive or negative throughout (p, q).

Suppose f > 0 on (p, q). By Rolle’s theorem there exists c ∈ (p, q) such that f 0 (c) = 0.
Then f 00 (c) = f (c) > 0, which tells that f has a minimum at c. But f > 0 on (p, q), so
for a minimum to occur at c, there must be a maximum preceding it, say at d (where
p < d < c). But then f (d) = f 0 (d)+f 00 (d) = f 00 (d) < 0 which contradicts our assumption
that f > 0 on (p, q). One can deal with the case f < 0 in a similar manner.

5. 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.

2
Pn
Thus we have shown that an+1 + an+2 ≤ i=1 ai + 2 holds for every n ≥ 1. Starting
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).

6. (a) If α is a root of the polynomial p(x) = a0 + a1 x + · · · + an xn with real coefficients


a
k
(an 6= 0), then show that |α| ≤ 1 + max .
0≤k≤n−1 an

(b) Let a0 + 10a1 + · · · + 10n an be the decimal representation of a prime number such
that an ≥ 2, n > 1. Prove that the polynomial p(x) = a0 + a1 x + · · · + an xn cannot
be written as a product of two non-constant polynomials with integer coefficients.

Solution. (a) Note that the bound is trivial if |α| = 0 or 1. Assume now that |α| =
6 0, 1.
Pn
It is given that k=0 ak αk = 0. We transfer the term an αn to the RHS and use triangle
inequality to obtain

n−1 n−1 n−1  n 


n
X X
k k
X
k
 |α| − 1
|an α | = ak α ≤ |ak ||α| ≤ max |ak | |α| = max |ak | .

k=0 k=0
0≤k≤n−1
k=0
0≤k≤n−1 |α| − 1

The second inequality above comes from the fact that each |ai | is less than max |ak |.
0≤k≤n−1
Now just cross-multiply to arrive at
 n 
ak |α| − 1 ak
|α| − 1 ≤ max
≤ max .
0≤k≤n−1 an |α|n 0≤k≤n−1 an

This completes the proof of part (a).

(b) Let, if possible, q(x) and r(x) be non-constant polynomials with integer coeffi-
cients such that p(x) = q(x)r(x). Let x1 , . . . , xk be the zeros of q(x) and xk+1 , . . . , xn
be the zeros of r(x). Since q(10)r(10) = p(10) which is a prime, and q(10), r(10) both
are integers, one of them must be ±1. We can assume w.l.o.g that q(10) = ±1. Then
|q(10)| = |(10 − x1 )(10 − x2 ) · · · (10 − xk )| = 1. On the other hand, the inequality in
part (a) tells us that each zero xi has a modulus less than 1 + 9/2 = 11/2 < 9, hence
|10 − xi | > 1 for each 1 ≤ i ≤ k, which contradicts the above equality.

3
7. Suppose f : [0, a] → R is a continuous function. Define another function g : (0, a) → R
Z a Z a Z a
f (t)
as g(x) = dt, x ∈ (0, a). Prove that lim+ g(t)dt = f (t)dt.
x t δ→0 δ 0
Solution. First set g(a) = 0, as per the usual convention. The fundamental theorem of
Calculus tells us1 that g 0 (x) = − f (x)
x
for each x ∈ (0, a). Now we can apply integration
by parts on g to write
Z a a Z a Z a
0
tg (t)dt = −δg(δ) +
g(t)dt = tg(t) − f (t)dt.

δ δ δ δ
Z a
Again by fundamental theorem of Calculus f (t)dt is a continuous function of x,
x Z a
therefore the second summand above converges to f (t)dt when δ → 0+ . It only
0
remains to show that δg(δ) → 0 when δ → 0+ .

Since f is continuous on [0, a], it must be bounded, say |f (x)| ≤ M for every x ∈ [0, a].
Hence note that
a Z a Z a
Z
f (t) f (t) M
|δg(δ)| = δ dt ≤ δ dt ≤ δ dt = M δ(log a − log δ).
δ t δ
t
δ t

Now lim+ δ log δ = 0, so the above RHS converges to 0 and we are through.
δ→0

8. 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.

1
To be more rigorous, we fix an x ∈ (0, a), take an x0 such that 0 < x0 < x and apply the fundamental
theorem to f (x)/x on the interval [x0 , a].

You might also like