Solutions Notes for DNY-NTCONSTRUCT

Evan Chen∗

January 17, 2018

§1 Solution Notes to TSTST 2015/5

Let ϕ(n) denote the number of positive integers less than n that are relatively prime to

least 2015 solutions in n.

h n
n. Prove that there exists a positive integer m for which the equation ϕ(n) = m has at

e e
C U s
I consider the following ELEVEN PRIME NUMBERS:

a n a l
S = {11, 13, 17, 19, 29, 31, 37, 41, 43, 61, 71} .

It has the property that for any p ∈ S, all prime factors of p − 1 are one digit.

Ev e n
Let N = (210)billion , and consider M = φ (N ). For any subset T ⊂ S, we have

r 
N


t
Y
M = φ Q p .
(p − 1)

y n
p∈T p∈T

B S, I
Since 2|T | > 2015 we’re done.
This solution was motivated by the deep fact that ϕ(11 · 1000) = ϕ(10 · 1000), for
example.

I
§2 Solution Notes to IMO 2003/6

T
O
Let p be a prime number. Prove that there exists a prime number q such that for every
integer n, the number np − p is not divisible by q.

By orders, we must have q = pk + 1 for this to be possible. So we just need np 6≡

p ⇐⇒ pk 6≡ 1 (mod q).
So we need a prime q ≡ 1 (mod p) such that pk 6≡ 1 (mod q). Wishfully we hope the
order of p is p and k - p. One way to do this is extract a prime factor from the cyclotomic
polynomial
pp − 1
p−1
which does not happen to be 1 (mod p2 ).

∗
Internal use: Olympiad Training for Individual Study (OTIS). Last update January 17, 2018.

1
Evan Chen (January 17, 2018) Solutions Notes for DNY-NTCONSTRUCT (OTIS)

§3 Solution Notes to December TST 2015/2

Prove that for every positive integer n, there exists a set S of n positive integers such
that for any two distinct a, b ∈ S, a − b divides a and b but none of the other elements of
S.

The idea is to look for a sequence d1 , . . . , dn−1 of “differences” such that the following
two conditions hold. Let si = d1 + · · · + di−1 , and ti,j = di + · · · + dj−1 for i ≤ j.

(i) No two of the ti,j divide each other.

(ii) There exists an integer a satisfying the CRT equivalences

a ≡ −si (mod ti,j ) ∀i ≤ j

n
Then the sequence a + s1 , a + s2 , . . . , a + sn will work. For example, when n = 3 we can

e
take (d1 , d2 ) = (2, 3) giving

h s e z
5
}| {
10 |{z} 12 |{z} 15

C U
2 3

because the only conditions we need satisfy are

a n a l a≡0 (mod 2)

n
a≡0 (mod 5)

Ev e r a ≡ −2 (mod 3).

t
But with this setup we can just construct the di inductively. To go from n to n + Q1, take
n−1

y
a d1 , . . . , dn−1 and let p be a prime not dividing any of the di . Moreover, let M = i=1 di .

In
Then we claim that d1 M, d2 M, . . . , dn−1 M, p is such a difference sequence. For example,

B S,
the previous example extends as follows.

z
1507
}| {

I
907
z }| {
a |{z} b |{z} c |{z} d

T
600 900 7

The new numbers p, p + M dn−1 , p + M dn−2 , . . . are all relatively prime to everything else.

O
Hence (i) still holds. To see that (ii) still holds, just note that we can still get a family of
solutions for the first n terms, and then the last n + 1th term can be made to work by
Chinese Remainder Theorem since all the new p + M dk are coprime to everything.

§4 Solution Notes to USAMO 2017/1

Prove that there exist infinitely many pairs of relatively prime positive integers a, b > 1
for which a + b divides ab + ba .

dd +2d
One construction: let d ≡ 1 (mod 4), d > 1. Let x = d+2 . Then set

x+d x−d
a= , b= .
2 2

2
Evan Chen (January 17, 2018) Solutions Notes for DNY-NTCONSTRUCT (OTIS)

To see this works, first check that b is odd and a is even. Let d = a − b be odd. Then:

a + b | ab + ba ⇐⇒ (−b)b + ba ≡ 0 (mod a + b)
a−b
⇐⇒ b ≡1 (mod a + b)
⇐⇒ bd ≡ 1 (mod d + 2b)
⇐⇒ (−2) ≡ dd d
(mod d + 2b)
⇐⇒ d + 2b | dd + 2d .

So it would be enough that

dd + 2d dd + 2d
 
1
d + 2b = =⇒ b = −d
d+2 2 d+2

which is what we constructed. Also, since gcd(x, d) = 1 it follows gcd(a, b) = gcd(d, b) = 1.

h en
Remark. Ryan Kim points out that in fact, (a, b) = (2n − 1, 2n + 1) is always a solution.

e
C U s
§5 Solution Notes to JMO 2016/2
Prove that there exists a positive integer n < 106 such that 5n has six consecutive zeros

a n a l
in its decimal representation.

Ev e rn
One answer is n = 20 + 219 = 524308.
First, observe that

t
5n ≡ 520 (mod 520 )

y
B S, In 5n ≡ 520 (mod 220 )

the former being immediate and the latter since ϕ(220 ) = 219 . Hence 5n ≡ 520 (mod 1020 ).
Moreover, we have

I
1 1
520 = 20
· 1020 < · 1020 = 10−6 · 1020 .
2 10002

O T
Thus the last 20 digits of 5n will begin with six zeros.

§6 Solution Notes to Shortlist 2007 N2

Let b, n > 1 be integers. Suppose that for each k > 1 there exists an integer ak such that
b − ank is divisible by k. Prove that b = An for some integer A.

Just let k = b2 , so b ≡ C n (mod b2 ). Hence C n = b(bx + 1), but gcd(b, bx + 1) = 1 so

b = An for some A.

§7 Solution Notes to IMO 2000/5

Does there exist a positive integer n such that n has exactly 2000 prime divisors and n
divides 2n + 1?

3
Evan Chen (January 17, 2018) Solutions Notes for DNY-NTCONSTRUCT (OTIS)

Answer: Yes.
We say that n is Korean if n | 2n + 1. First, observe that n = 9 is Korean. Now, the
problem is solved upon the following claim:

Claim. If n > 3 is Korean, there exists a prime p not dividing n such that np is Korean
too.

Proof. I claim that one can take any primitive prime divisor p of 22n − 1, which exists by
Zsigmondy theorem. Obviously p 6= 2. Then:

• Since p - 2ϕ(n) − 1 it follows then that p - n.

• Moreover, p | 2n + 1 since p - 2n − 1.

Hence np | 2n + 1 | 2np + 1 by Chinese Theorem, since gcd(n, p) = 1.

n
§8 Solution Notes to BAMO 2011/5

e e
Decide whether there exists a row of Pascal’s triangle containing four pairwise distinct

h
C
An example is 203
U s
numbers a, b, c, d such that a = 2b and c = 2d.

203 203 203

l





68 = 2 67 and 85 = 2 83 .

va n
To get this, the idea is to look for two adjacent entries and two entries off by one, and

a
solving the corresponding equations. The first one is simple:

n
r
   
n n
=2 =⇒ n = 3j − 1.

y E n t e
j

The second one is more involved:

j−1

I
   
n n

B S,
=2
k k−2
=⇒ (n − k + 1)(n − k + 2) = 2k(k − 1)

I
=⇒ 4(n − k + 1)(n − k + 2) = 8k(k − 1)
=⇒ (2n − 2k + 3)2 − 1 = 2 (2k − 1)2 − 1

O T √
=⇒ (2n − 2k + 3)2 − 2(2k − 1)2 = −1

Using standard methods for the Pell equation:

√ √
• (7 + 5 2)(3 + 2 2) = 41 + 29 2. So k = 15, n = 34, doesn’t work.
√ √ √
• (41 + 29 2)(3 + 2 2) = 239 + 169 2. Then k = 85, n = 203.

§9 Solution Notes to TSTST 2012/5

A rational number x is given. Prove that there exists a sequence x0 , x1 , x2 , . . . of rational
numbers with the following properties:

(a) x0 = x;

(b) for every n ≥ 1, either xn = 2xn−1 or xn = 2xn−1 + n1 ;

(c) xn is an integer for some n.

4
Evan Chen (January 17, 2018) Solutions Notes for DNY-NTCONSTRUCT (OTIS)

Think of the sequence as a process over time. We’ll show that:

Claim. At any given time t, if the denominator of xt is some odd prime power q = pe ,
then we can delete a factor of p from the denominator, while only adding powers of two
to the denominator.

(Thus we can just delete off all the odd primes one by one and then double appropriately
many times.)

Proof. The idea is to add only fractions of the form (2k q)−1 .
Indeed, let n be large, and suppose t < 2r+1 q < 2r+2 q < · · · < 2r+m q < n. For some
binary variables εi ∈ {0, 1} we can have
ε1 ε2 εm
xn = 2n−t xt + c1 · + c2 · · · · + cs ·
q q q

en e
n−2r+i q
where ci is some power of 2 (to be exact, ci = 2 2r+1 , but the exact value doesn’t

h s
matter).
If m is large enough the set {0, c1 } + {0, c2 } + · · · + {0, cm } spans everything modulo

C U
p. (Actually, Cauchy-Davenport implies m = p is enough, but one can also just use
Pigeonhole to notice some residue appears more than p times, for m = O(p2 ).) Thus we

n l
can eliminate one factor of p from the denominator, as desired.

va n a
§10 Solution Notes to Shortlist 2014 N4

r
y E n e
Let n > 1 be an integer. Prove that there are infinitely many integers k ≥ 1 such that

t  k
n

B S,
is odd.
I k

T I
If n is odd, then we can pick any prime p dividing n, and select k = pm for sufficiently
large integers m.
Suppose n is even now. Then by Kobayashi’s Theorem, there exist infinitely many

O
primes p dividing some number of the form

nn
r −1
− 1.

for some integer r. Let p > n be such a prime, with corresponding integer r. It then
follows that
r
nn p ≡ nr (mod nr p)
since this is clearly correct mod nr , and also correct modulo p. If we select k = nr p, we
have  k
n nr p − nr
=
k nr p
which is odd.

5
Evan Chen (January 17, 2018) Solutions Notes for DNY-NTCONSTRUCT (OTIS)

§11 Solution Notes to USAMO 2006/3

For integral m, let p(m) be the greatest prime divisor of m. By convention, we set
p(±1) = 1 and p(0) = ∞. Find all polynomials f with integer coefficients such that the
sequence
{p f n2 − 2n}n≥0

is bounded above. (In particular, this requires f (n2 ) 6= 0 for n ≥ 0.)

If f is the (possibly empty) product of linear factors of the form 4n − a2 , then it

satisfies the condition. We will prove no other polynomials work. In what follows, assume
f is irreducible and nonconstant.
It suffices to show for every positive integer c, there exists a prime p and a nonnegative
integer n such that n ≤ p−1 2
2 − c and p divides f (n ).
Firstly, recall there are infinitely many odd primes p, with p > c, such that p divides

p−1 2

en
some f (n2 ), by Schur’s Theorem. Looking mod such a p we can find n between 0 and
2
2 (since n ≡ (−n) (mod p)). We claim that only finitely many p from this set can

e
fail now. For if a p fails, then its n must be between p−1

h
p−1

s
2 − c and 2 . That means for
some 0 ≤ k ≤ c we have

C U
2 !  !
1 2
 
p−1
0≡f −k ≡f k+ (mod p).

l
2 2

va n a
There are only finitely many p dividing

n c 
1
2 !

r
Y
f k+

E
2

t e k=1

unless one of the terms in the product is zero; this means that 4n − (2k + 1)2 divides

y n
f (n). This establishes the claim and finishes the problem.

B S, I
§12 Solution Notes to USAMO 2013/5

I
Let m and n be positive integers. Prove that there exists an integer c such that cm and
cn have the same nonzero decimal digits.

O T
One-line spoiler: 142857.
To work out the details, there exist arbitrarily large primes p such that
p | 10e m − n
for some positive integer e, say by Kobayashi theorem (or other more mundane means).
In that case, the periodic decimal expansions of m n
p and p are cyclic shifts of each other.
Thus if one looks at p1 the repeating block of decimals, one may take c to be that resulting
integer.
Remark. The official USAMO solutions propose using the fact that 10 is a primitive
root modulo 7e for each e ≥ 1, by Hensel lifting lemma. This argument is incorrect,
because it breaks if either m or n are divisible by 7.
One may be tempted to resort to using large primes rather than powers of 7 to deal
with this issue. However it is an open conjecture (a special case of Artin’s primitive root
conjecture) whether or not 10 (mod p) is primitive infinitely often, which is the condition
necessary for this argument to work.

6
Evan Chen (January 17, 2018) Solutions Notes for DNY-NTCONSTRUCT (OTIS)

§13 Solution Notes to RMM 2012/4

Prove there are infinitely many integers n such that n does not divide 2n + 1, but divides
n
22 +1 + 1.

Zsig hammer! Define the sequence n0 , n1 , . . . as follows. Set n0 = 3, and then for k ≥ 1
nk−1
we let nk = pnk−1 where p is a primitive prime divisor of 22 +1 + 1 (by Zsigmondy).

For example, n1 = 57.

This sequence of nk ’s works for k ≥ 1, by construction.
It’s very similar to IMO 2000 Problem 5.

§14 Solution Notes to Shortlist 2013 N4

Determine whether there exists an infinite sequence of nonzero digits a1 , a2 , a3 , . . . such

h en
that the number ak ak−1 . . . a1 is a perfect square for all sufficiently large k.

e
The answer is no.

s √
Assume for contradiction such a sequence exists, and let xk = ak ak−1 . . . a1 for k

C U
large enough. Difference of squares gives

a n a l def
Ak · Bk = (xk+1 − xk )(xk + xk+1 ) = ak · 10k

v n
with gcd(Ak , Bk ) = 2 gcd(xk , xk−1 ) since xk and xk−1 have the same parity. Note that

r
we have the inequalities

E e √

t
Ak ≤ Bk < 2xk+1 < 2 · 10k+1 .

y
B S, n
The idea will be that divisibility issues will force one of Ak and Bk to be too large.

I
We now split the proof in two cases:

• First, assume ν5 (x2k ) ≥ k for all k. Then in particular a1 = 5, so all xk are always
odd. So one of Ak and Bk is divisible by 2k−1 . Moreover, both divisible by at least

T I
5k/2 . So for each k,
min(Ak , Bk ) ≥ 2k−1 · 5k/2
which is impossible for large enough k.

O
• Next assume ν5 (x2m ) = 2e < m for some m. Then since x2k+1 ≡ x2k (mod 10k ), we
obtain ν5 (x2k ) = 2e for all k > m. Now,

min(Ak , Bk ) ≥ 5k−e

which again is impossible for k large enough.

§15 Solution Notes to EGMO 2014/3

We denote the number of positive divisors of a positive integer m by d(m) and the number
of distinct prime divisors of m by ω(m). Let k be a positive integer. Prove that there
exist infinitely many positive integers n such that ω(n) = k and d(n) does not divide
d(a2 + b2 ) for any positive integers a, b satisfying a + b = n.

7
Evan Chen (January 17, 2018) Solutions Notes for DNY-NTCONSTRUCT (OTIS)

Weird problem. The condition is very artificial, although the construction is kind
of fun. I’m guessing the low scores during the actual contest were actually due to an
unusually tricky P2.
Let n = 2p−1 t, where t ≡ 5 (mod 6), ω(t) = k − 1, and p  t is a sufficiently large
prime. Let a + b = n and a2 + b2 = c. We claim that p - d(c), which solves the problem
since p | 2(n).
First, note that 3 - a2 + b2 , since 3 - n. Next, note that c < 2n2 < 5p−1 (since p  t)
so no exponent of an odd prime in c exceeds p − 2. Moreover, c < 23p−1 .
So, it remains to check that ν2 (c) ∈/ {p − 1, 2p − 1}. On the one hand, if ν2 (a) < ν2 (b),
then ν2 (a) = p − 1 and ν2 (c) = 2ν2 (a) = 2p − 2. On the other hand, if ν2 (a) = ν2 (b) then
ν2 (a) ≤ p − 2, and ν2 (c) = 2ν2 (a) + 1 is odd and less than 2p − 1.

§16 Solution Notes to USAMO 2012/3

n
Determine which integers n > 1 have the property that there exists an infinite sequence
a1 , a2 , a3 , . . . of nonzero integers such that the equality

h e s e ak + 2a2k + · · · + nank = 0

C U
holds for every positive integer k.

n l
a
Answer: all n > 2.

for all k.

va
For n = 2, we have ak + 2a2k = 0, which is clearly not possible, since it implies a2k = 2ak1

rn
For n ≥ 3 we will construct a completely multiplicative sequence (meaning aij = ai aj

E e
for all i and j). Thus (ai ) is determined by its value on primes, and satisfies the condition

t
as long as a1 + 2a2 + · · · + nan = 0. The idea is to take two large primes and use Bezout’s

y n
theorem, but the details require significant care.

B S, I
We start by solving the case where n ≥ 9. In that case, by Bertrand postulate there
exists primes p and q such that
1

I
dn/2e < q < 2 dn/2e . and (q − 1) < p < q − 1
2

T
Clearly p 6= q, and q ≥ 7, so p > 3. Also, p < q < n but 2q > n, and 4p ≥ 4 12 (q + 1) > n.

We now stipulate that ar = 1 for any prime r = 6 p, q (in particular including r = 2 and

O
r = 3). There are now three cases, identical in substance.

• If p, 2p, 3p ∈ [1, n] then we would like to choose nonzero ap and aq such that

1
6p · ap + q · aq = 6p + q − n(n + 1)
2
which is possible by Bézout lemma, since gcd(6p, q) = 1.

• Else if p, 2p ∈ [1, n] then we would like to choose nonzero ap and aq such that

1
3p · ap + q · aq = 3p + q − n(n + 1)
2
which is possible by Bézout lemma, since gcd(3p, q) = 1.

8
Evan Chen (January 17, 2018) Solutions Notes for DNY-NTCONSTRUCT (OTIS)

• Else if p ∈ [1, n] then we would like to choose nonzero ap and aq such that

1
p · ap + q · aq = p + q − n(n + 1)
2
which is possible by Bézout lemma, since gcd(p, q) = 1. (This case is actually
possible in a few edge cases, for example when n = 9, q = 7, p = 5.)

It remains to resolve the cases where 3 ≤ n ≤ 8. We enumerate these cases manually:

• For n = 3, let an = (−1)ν3 (n) .

• For n = 4, let an = (−1)ν2 (n)+ν3 (n) .

• For n = 5, let an = (−2)ν5 (n) .

• For n = 6, let an = 5ν2 (n) · 3ν3 (n) · (−42)ν5 (n) .

n
• For n = 7, let an = (−3)ν7 (n) .

e e
• For n = 8, we can choose (p, q) = (5, 7) in the prior construction.

h
C s
This completes the constructions for all n > 2.

U
n l
§17 Solution Notes to TSTST 2016/3

a a
n
Decide whether or not there exists a nonconstant polynomial Q(x) with integer coefficients

v r
with the following property: for every positive integer n > 2, the numbers

E e
t
Q(0), Q(1), Q(2), . . . , Q(n − 1)

y
B S, In
produce at most 0.499n distinct residues when taken modulo n.

We claim that
Q(x) = 420(x2 − 1)2

T I
works. Clearly, it suffices to prove the result when n = 4 and when n is an odd prime p.
The case n = 4 is trivial, so assume now n = p is an odd prime.
First, we prove the following easy claim.

O
 2

Claim. For any odd prime p, there are at least 12 (p−3) values of a for which 1−a
p = +1.

Proof. Note that if k 6= 0, k 6= ±1, k 2 6= −1, then a = 2(k + k −1 )−1 works. Also a = 0
works.

Let F (x) = (x2 − 1)2 . The range of F modulo p is contained within the 12 (p + 1)
quadratic residues modulo p. On the other hand, if for some t neither of 1 ± t is a
quadratic residue, then t2 is omitted from the range of F as well. Call such a value of t
useful, and let N be the number of useful residues. We aim to show N ≥ 41 p − 2.

9
Evan Chen (January 17, 2018) Solutions Notes for DNY-NTCONSTRUCT (OTIS)

We compute a lower bound on the number N of useful t by writing

X            !
1 1−t 1+t 2 −2
N= 1− 1− − 1− − 1−
4 t
p p p p
     
1X 1−t 1+t
≥ 1− 1− −1
4 t p p
!
1 X  1 − t2 
= p+ −1
4 t
p
1
p + (+1) · 21 (p − 3) + 0 · 2 + (−1) · ((p − 2) − 12 (p − 3)) − 1


≥
4
1
≥ (p − 5) .
4
Thus, the range of F has size at most

h en 1

e
2
1 3
(p + 1) − N ≤ (p + 3).
2 8

C U s
This is less than 0.499p for any p ≥ 11.

Remark. In fact, the computation above is essentially an equality. There are only two

a l
points where terms are dropped: one, when p ≡ 3 (mod 4) there are no k 2 = −1 in

n
the lemma, and secondly, the terms 1 − (2/p) and 1 − (−2/p) are dropped in the initial

a n
estimate for N . With suitable modifications, one can show that in fact, the range of F is

v
exactly equal to

E e r 
1
8 (3p + 5) p ≡ 1 (mod 8)

t



1

y

1 1 (3p + 7) p ≡ 3 (mod 8)

n

(p + 1) − N = 18

I
2 2 8 (3p + 9) p ≡ 5 (mod 8)

B S,



 1 (3p + 3) p ≡ 7 (mod 8).

8

T I
O

10