TOPICS IN NUMBER THEORY:

An Olympiad-Oriented Approach

Masum Billal Amir Hossein Parvardi

August 28, 2018

 Dedicated to

Our regular studies, without which, we could have finished this book long time ago.
Fermat, the father of modern number theory.
Euler, without whom number theory probably wouldn’t be so rich today.
Riemann, who made huge contributions to analytic number theory.
Ramanujan, the mathematician of mathematicians.
Paul Erdős, the man who loved only numbers.
Links, Contact
• The website of the book always contains the updates, errata, and new free ma-
terials that we regularly post on social media. Exclusive olympiad problem-sets
will be released in the website in the near fuure: TopicsInNumberTheory.com

• Facebook page: @TopicsInNumberTheory. This is our main page for feedback

and reviews on the book. Feel free to post any kind of review, comment, solution,
or suggestion you have about this book. Amir Hossein will create a forum for the
book on AoPS so everyone can post their solutions to be added to the book later
on.

• If you find any mistake or typo in the book, we would be happy to hear it at
Facebook or [email protected].

3
First Words
I would like to have a few words before diving into the discussion. First of all, from my
personal experience, I have found that there is a common practice to learn1 by learning
a lot of theories and then investigating how those theories are used to solve problems.
As our primary audience would be students who are looking to get into mathematical
olympiads, I highly discourage this. Please do not take number theory for a collection
of theories just because the word theory is literally juxtaposed with it. That being said,
one could say that our book itself is a collection of a lot of theorems as well. Sadly,
that is partially true for multiple reasons even though it was not our intention at all.
When I first thought about writing this book, my intention was to make students
realize that they do not need to know a lot of theorems in order to be able to solve
problems. My primary target was to build confidence about this claim. But as we kept
writing we had to increase the pace since we had to cover a lot (and that was discarding
a lot of contents which we thought would ask for even discussion or we just felt lazy
about it), we had to increase the pace.
I would like to address one more concern. Originally, my plan was to make this book
a series of 5 volumes, this being the first one. In those volumes I wanted to discuss a
lot of topics such as special numbers like egyptian fraction or even interesting numbers
like abundant number or deficient number and their properties etc or crucial topics such
as Diophantine equations. You will notice that we have left a lot of important topics
like those out of this book. The reason is, I quickly realized I can hardly finish writing
this first volume, and if I wanted to complete the series we will probably have to keep
writing my whole life. So, I had to discard a lot of contents and make the book concise
which resulted in squeezing in a lot of contents in a few hundreds of pages.
Finally, I would like to thank Amir to join me in this project. At one point I stopped
writing the book. If he had not agreed to be a co-author, this book would have probably
not been completed at all. More so because he agreed to follow the style I wanted to
write in even when we had objection from reputed publisher like Springer.

Masum Billal

4
 In the past three years, we have always been worried about this book. It’s been a
long and tedious job to manage everything and edit all we had written a very long time
ago. After I studied higher level number theory concepts, there were times that I found
errors/typos in my previous drafts for the book. And that sometimes happened two or
more times in a short period, and so, it was getting annoying. Anyhow, we managed
to finish it at this time of the summer. It is now 5:36 AM, Tuesday, July 17, 2018 that
I’m writing this. You can imagine how crazy this process has made me!
BUT! a wise man once said “high quality math books are written in a series of
editions in long term,” and we are happy to hear that. Many of our friends helped us
during the way to finish this book, as mentioned in the Acknowledgement, and we are
so proud to have such friends.
I wouldn’t be able to finish my part in this book if I didn’t have the support of
my wonderful, beautiful, and lovely wife Nadia Ghobadipasha. She gave me hope
to choose mathematics and always believed in me. She was my only one in the hardest
days of life.
Professor Peyman Nasehpour, whom I knew from the very first semester of my
undergraduate studies in elelctrical englineering at University of Tehran, helped me
a lot in the process of enhancing my mathematical abilities to change my field to
mathematics (number theory) for my master’s. He is an inspiration to me, and a great
colleague when it comes to teamwork. I’m looking forward to working with him more
ofter.
The idea behind the lattices in the cover is a geometric proof of law of quadratic
reciprocity. Our friends found other interpretations of it such as Pick’s theorem (which
is not the case here) or the sum of positive integers up to n, which you will realize is
also not always true (for all primes p, q). Section ( ) is dedicated to this proof and
investigates why it is true using counting the lattice (grid) points. When we picked
this idea for the cover, we chose quadratic reciprocity because its proof is geometrically
visual and indeed very beautiful. Our hope was to make the reader curious because
the design looks familiar. Stay excited until you read that proof! Or, maybe, go read
it if you have the prerequisite knowledge and the puzzle is boggling your mind. I
remember I found the idea of the proof in one of Kenneth H. Rosen’s books on discrete
mathematics, but I’m not sure which one, as it was a long time ago when I wrote it. I
used TikZ package for LATEXto write the codes and generate the graphs.
There are so many wonderful things to learn in this book. I hope you enjoy it!

Amir Hossein Parvardi,

University of British Columbia,
Vancouver, BC, Canada,
July 2018.

5
Acknowledgement
Here is a list of all the kind people who helped us review, edit, and improve this book.
We could not decide who to thank more, hence an alphabetically (last name) ordered
list:
1. Thanks to Ali Amiri, a kind friend who helped us in the cover design.
2. Thanks to AnréC from TeX.StackExchange who wrote the code for figure ( )
in base conversion.
3. We are thankful to Arta Khanalizadeh for designing the cover of the book.
4. Thanks to Kave Eskandari for reading the whole book and commenting on the
general points. He caught a good mistake in chapter .
5. Cheers to our mathematical friend Leonard Mihai C. Giugiuc from Romania
who gave us positive and constructive feedback on the book. He also wrote us a
wonderful review in the website.
6. Thanks to Valentio Iverson for proof-reading chapters and , and pointing
out the typo in figure ( ).
7. Thanks to Aditya Khurmi for reading the book and giving us positive feedback.
8. We appreciate Hesam Korki’s comments on chapter . He mentioned a few
very important typos, including grammatical. He also mentioned a mathematical
change in that chapter, which was very helpful.
9. Professor Peyman Nasehpour sent us the beautiful problem ( ) and an
amazing solution using Prime Number Theorem. He also gave us pretty use-
ful comments on chapter . He also introduced in section ( ) the amicable
numbers to us with a brief historical note on it.
10. We appreciate Kenji Nakagawa’s comments on chapters and . He was one
of the first people who read and reviewed these two chapters when we put them
on the website of the book.
11. We are thankful to Mohammadamin Nejatbakhshesfahani, Iran National
Olympiad gold medalist (2010) and winner of gold medals at IMS and IMC, who
honored us toread and reviewe the whole book and gave us really instructive
comments.
12. We would like to thank Nur Muhammad Shafiullah Mahi for his efforts to
make this book better.
13. We are honored to thank Professor Greg Martin, a faculty member at the
Mathematics Department of University of British Columbia. He happens to be
Amir Hossein’s Master’s supervisor. He kindly reviewed a printed draft of the
book and emailed us over 10 major points to correct in the book. We do appreciate
his advice on improving the whole context of the book.

6
 14. We are thankful to Sohrab Mohtat for his comments on chapter . Thanks to
him, we avoided a fatal mistake in the beginning of the book. He also wrote a
very useful and detailed review for our book in the website.

15. We are thankful to Aditya Guha Roy who reviewed the whole book and caught
a few LaTeX typos, generalized lemma ( ), fixing problems in chapter . Aditya
wrote an amazing, educative review in our website.

16. Navneel Singhal carefully reviewed and proof-read the whole first part of the
book (chapters to ) and gave us very constructive comments. We are thankful
to him.

17. Thanks to Amin Soofiani, who is a Master’s student of mathematics at Univer-

sity of British Columbia, we noticed there was a mistake in theorem ( ). He
did a perfect, precise, and detailed review on chapter .

18. We are thankful to Sepehr Yadegarzadeh for informing us about the correct
umlaut for Möbius among other grammatical and vocabulary points.

1
umlaut: a mark (¨) used over a vowel, as in German or Hungarian, to indicate a different vowel
quality, usually fronting or rounding.

7
Contents

15
17
. . . . . . . . . . . . . . . . . . . . . . . . 17
. . . . . . . . . . . . . . . . . . 24
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
. . . . . . . . . . . . . . . . . . . . . . . . . . . 36
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
. . . . . . . . . . . . . . . . . . . . . . . . . . 44
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

69
. . . . . . . . . . . . . . . . . . . . . . . . . 70
. . . . . . . . . . . . . . . . . . . . . . . . . . 76
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
. . . . . . . . . . . . . 83
. . . . . . . . . . . . 85
. . . . . . . . . . . . . . . . . . . . . . . . 88
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
. . . . . . . . . . . . . . . . . . . . . . . 94
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
. . . . . . . . . . . . . . . . . . . . . . . . . . 101
. . . . . . . . . . . . . . . . . . . . . . . 106
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
. . . . . . . . . . . . . . . . . . . . . . . . . . 109
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
. . . . . . . . . . . . . . . . . . . . . . . . . . . 124
. . . . . . . . . . . . . . . . . . 138
. . . . . . . . . . . . . . . . . . . . . . . 138
. . . . . . . . . . . . . . . . . . . . . . . . . . 140
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

9
 . . . . . . . . . . . 142
. . . . . . . . . . . . . . . 145
. . . . . . . . . . . . . . . . . . . . . . . 145
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150

159
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
. . . . . . . . . . . . . . . . 167
. . . . . . . . . . . . . . . . . . . 169
. . . . . . . . . . . . . . . . . . . . . . . . . 171
. . . . . . . . . . . . . . . . . . . . . . 174
. . . . . . . . . . . . . . . . . . . . . . . . . 174
. . . . . . . . . . . . . . . . . . . . . . . . . . . 176
. . . . . . . . . . . . . . 179
. . . . . . . . . . . . . . . . . . 183
. . . . . . . . . . . . . . . . . . 185
. . . . . . . . . . . . . . . . . . . . . 190
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
. . . . . . . . . . . . . . . . . 200
. . . . . . . . . . . . . . . . . . . . . 206
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221

229
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
. . . . . . . . . . . . . . . . . . . . . 242
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
. . . . . . . . . . . . . . . . . . . . . . 253
. . . . . . . . . . . . . . . . . . . . . . . . 254
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
. . . . . . . . . 268
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
. . . . . . . . . . . . . . . . . 273
. . . . . . . . . . . . . . . . . . . . 274
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
. . . . . . . . . . . . . . . . . . . . . . . . . 279

283
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
. . . . . . . . . . . . . . . . . . . . . . . . 289
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293

10
 11

. . . . . . . . . . . . . .
. . . . . . . . . . . . . 296
. . . . . .
. . . . . . . . . . . . . 298
. . . . . . . . . .
. . . . . . . . . . . . . 301
. . .
. . . . . . . . . . . . . 301
. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . 302
. . . . . . . . . . . . . .
. . . . . . . . . . . . . 304
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . 305
. . . . . . . . . . . . .
. . . . . . . . . . . . . 306
. . . . . . . . . . . . . .
. . . . . . . . . . . . . 308
. . . . . . . . . . . . . .
. . . . . . . . . . . . . 312
. .
. . . . . . . . . . . . . 318
. . .
. . . . . . . . . . . . . 320
. . . . . . . . . .
. . . . . . . . . . . . . 324
. . . . . . . . . . . . . 329
. . . . . . . . . . . . . . . . . . . . 331
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336

347

353
355

385
 13

Notations
• N, N0 , Z, Q, R, P, and C are the sets of positive integers, non-negative integers,
integers, rational numbers, real numbers, primes and complex numbers, respec-
tively.

• |a| denotes the absolute value of a for any real number a.

• min(a, b) is the minimum of a and b and max(a, b) is the maximum of a and b.

• (a, b) and [a, b] are greatest common divisor and least common multiple of a and
b respectively.

• a|b means b is divisible by a without any remainder.

• a ⊥ b means (a, b) = 1.

• pn is the nth prime.

• π(x) is the number of primes less than or equal to the real number x.

• ϕ(n) is the number of positive integers less than n which are coprime to n.

• d(n) is the number of divisors of n.

• σ(n) is the sum of divisors of n.

• φ(n) is the Euler function.

• µ(n) is the Mőbius function.

• λ(n) is the Carmichael function

• vn (a) is the largest non-negative integer α so that nα |a but nα+1 ∤ a.

• n! = 1 · 2 . . . n.
 
n
• is the binomial coefficient indexed by non-negative integers n and k.
k
 
a
• is the Legendre symbol for integer a and prime p.
p
• ⌊x⌋ is the largest integer not greater than x.

• ⌈x⌉ is the smallest integer integer not less than x.

 Part I

Fundamentals

15
Chapter 1

Divisibility

Contents
. . . . . . . . . . . . . . . . . . . 17
. . . . . . . . . . . . . . . . . 24
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
. . . . . . . . . . . . . . . . . . . . . . . . . 35
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
. . . . . . . . . . . . . . . . . . . . . . . . . 36
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
. . . . . . . . . . . . . . . . . . . . . . . . . 44
. . . . . . . . . . . . . . . . . . . . . . . . . 45
. . . . . . . . . . . . . . . . . . . . . . . . . . 54
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

1.1 Definitions and Propositions

Let us start from the very basics. Whenever you encounter a definition, try to make
sense of it with examples. Try to get used to these terms and divisibility notation. It
will not only make your life a lot easier, but also keep your solutions or ideas concise.
So try not to get bored with the definitions. This PDF was created
for Amazon.

Definition 1.1.1. When an integer a is divided by another integer b, we can write

a = bq + r for some integers q and r. In this general case, we call r the remainder of
the division.
However, if we choose q such that 0 ≤ r < |b|, then we say r is the minimum
remainder of the division .
At times, it is convenient to carry out the division so that b is as close as possible to
an integral multiple of a. Then we can write a = bq + r, with |r| ≤ | 2b | for some integer
q. In this case, we call r the minimum absolute remainder.
1
|x| is the absolute value of x.

17
18 1.1. DEFINITIONS AND PROPOSITIONS

Note (1). We will prove later that the minimum remainder is unique. That is, there
exists exactly one value for r such that a = bq + r and 0 ≤ r < b. Because of its
uniqueness and positivity, remainder in this book will mean minimum remainder unless
stated otherwise.
Note (2). When we talk about the the minimum absolute remainder in the division
a = bq + r, note that if b is odd, we will have a unique r. That is, there exist only one
possible value for r such that |r| ≤ | 2b |. However, if b and a both are even, there can be
two possible values for r. For example, if b = 20 and a = 8, the division can be done as
both 20 = 8 · 2 + 4 and 20 = 3 · 8 − 4 and we would have two values for r: 4 and −4. To
force r to be unique in this case, we will take the positive value of r as the minimum
absolute remainder. Therefore, in our example, 20 = 2 · 8 + 4 would be accepted as the
division equation and r = 4 is the minimum absolute remainder.
In the next definition, we have swapped a and b and written b = aq + r instead of
a = bq + r. So, read carefully.
Definition 1.1.2. If b = aq + r is the proper division of b by a (that is, a division
in which r is the minimum remainder), b is the dividend, a is the divisor, and q is the
quotient.
Example. Let b = 23 and a = 5. The division of b by a can be done in many ways.
Take 23 = 5 · 2 + 13, so we can say 13 is a remainder of 23 upon division by 5, but not
a minimum remainder. If we write the division as 23 = 5 · 4 + 3, then 3 is the minimum
remainder. And if we write it as 23 = 5 · 5 + (−2), then −2 is the minimum absolute
remainder.
If we divide 4 by 2, we get a zero remainder. In this case, we say that 4 is divisible
by 2 and write it as 2 | 4.
Definition 1.1.3. Let a and b be two integers. If b leaves a zero remainder upon
division by a, or equivalently, if there is an integer k for which b = ak, we write a | b
and say that
• b is divisible by a,

• a divides b,

• a is a divisor (or a factor ) of b, or

• b is a multiple of a.
Likewise, a ∤ b denotes that b is not divisible by a.
.
Remark. In some countries, authors use the notation b .. a instead of a|b, but it is not
as common.
Definition 1.1.4. If a divides b and |a| < |b|, the number a is called a proper divisor
of b. Here, |a| denotes the absolute value of a. For example, | − 5| = 5 and |5| = 5.
Example. 4 | 20 and 5 | 20 but 11 ∤ 20.
CHAPTER 1. DIVISIBILITY 19

Example. 1 is a divisor of all integers.

You can also try to make sense of division in this way: 8 divides 40 because 40 has
every factor that 8 has in it. In other words, if 8 had a factor which was not a factor of
40, 40 would not be divisible by 8. For example, 42 does not have the factor 4, which
is a factor of 8, therefore 8 ∤ 42.

Definition 1.1.5 (Prime and Composite Numbers). An integer n > 1 is called

prime if it has exactly two distinct (positive) divisors, namely 1 and n itself. A number
greater than 1 which is not a prime is composite. In other words, an integer n is
composite if it has a proper divisor other than 1.

Note. When we say a is a divisor of b, unless otherwise stated, we usually mean a is a

positive divisor of b. This is distinguished because negative divisors exist as well.

Question 1.1.6. Is 1 a prime number? If so, why? If not, how is it a composite number?
You may be already familiar with prime numbers and in that case, the definition may
seem a little different. Try to understand why we chose to stick with this definition
rather than a positive integer n that is not divisible by any positive integer other than
1 and n is a prime.

Definition 1.1.7 (Parity). Parity is the property of an integer being even or odd. An
even number is one which is divisible by 2. Odd numbers, on the other hand, leave a
remainder of 1 when divided by 2.

Example. 2 and 4 have the same parity, they are both even. 5 and 10 are of different
parity, 5 is odd and 10 is even.
Example. 11 is a prime because no positive integer greater than 1 and less than 11
divides it. Similarly 2, 3, and 29 are primes, but 169 (divisible by 13) and 1001 (why?)
are composites. If a number is divisible by 2, it is composite. Thus, the only even prime
is 2.
If we add or subtract two numbers of the same parity, the answer will be even.
Conversely, the result of addition or subtraction of two numbers with different parities
is always an odd number. Using these two facts, you can easily find many properties of
integers related to parity. For instance, we can say that if we add or subtract an even
number to or from a positive integer n, the parity does not change; i.e., parity remains
invariant in this case. Moreover, any odd multiple of n has the same parity as n.
We usually deal only with positive integers in divisibility relations. However, some-
times negative integers or zero also come into play.

Proposition 1.1.8 (Basic Properties of Divisibility). For any three integers a, b,

and c, the following statements are true.

1. a | 0.

2. a | a.

3. 1 | a and −1 | a.
20 1.1. DEFINITIONS AND PROPOSITIONS

4. If a is non-zero, then 0 ∤ a. In the case when a = 0, since 0/0 is undefined, 0 | 0

is also meaningless.

5. If a | b, then a | −b, −a | b, and −a | −b.

6. If a | b, then a | bk for all integers k.

7. If a | b, then ak | bk for all integers k.

8. If ak | bk for some non-zero integer k, then a | b.

9. If a | b and b | c, then a | c.
Keith Conrad states this as a mantra: A factor of a factor is a factor.

10. If a | b and b is non-zero, then |b| ≥ |a|. In other words, if a | b and |a| > |b|, then
we have b = 0.

11. If a | b, then an | bn for all non-negative integers n.

12. If an | bn for some positive integer n, then a | b.

Proof. Most of the parts are trivial and we prove only the important ones. A general
approach to solve this kind of problems is simply transforming them into equations.
That is, when x | y, write y = kx for some integer k.
4. 0 | a means that a = 0k for some integer k. So a = 0.

6. a|b means that b = aq for some q. Multiply both sides of this equation by k to
get bk = akq = aq ′ . Therefore a|bk.

9. a | b and b|c, so b = aq1 and c = bq2 for some integers q1 , q2 . Combine these two
equations to get c = aq1 q2 = aq and thus a|c.

10. a | b, so b = ak for some integer k. Rewrite this equation in absolute value terms:
|b| = |ak| = |a| · |k|. Since k is a non-zero integer, the smallest value for |k| is 1,
so |b| = |a| · |k| ≥ |a|.

Let us talk about (12) a bit. You may think that it is straightforward. It kind of

is,
bn b n
however, not that straightforward. The reason is we need to prove that if an = a is
an integer, so is ab . This is not obvious (try to make sense why).
Proposition 1.1.9. If a | b and b | a, then |a| = |b|. In other words, a = ±b.
Proof. The proof is quite simple using part 10 of the previous proposition. In fact, we
get |b| ≥ |a| and |a| ≥ |b|, which means |a| = |b| or a = ±b.
The above proposition often comes handy when you want to prove that two expres-
sions are equal. If you can show that each expressions divides the other one and that
they have the same sign (both positive or both negative), you can imply that their
values are equal.
CHAPTER 1. DIVISIBILITY 21

Proposition 1.1.10. For fixed positive integers a and b, there are unique integers q
and r so that b = aq + r with 0 ≤ r < a. In other words, the quotient and the minimum
remainder of the division are unique.
Proof. We can easily rule out the case r = 0. Just notice that a|b.
So, we can now assume that a ∤ b. So b must have a nonzero remainder upon division
by a, say r. One can rewrite the equation b = aq + r as aq = b − r or q = (b − r)/a.
This simply means that uniqueness of r implies the uniqueness of q. Therefore, we only
need to prove that r is unique. Assume that there exist integers q ′ and r′ so that
b = aq + r = aq ′ + r′ .
This implies a(q − q ′ ) = r′ − r, which shows that r′ − r is divisible by a. Is it possible?
The answer is clearly no. Since r′ and r are both less than a, we have |r′ − r| < a.
By part 10 of proposition ( ), we must have |r′ − r| = 0, which gives r′ = r. This
means two remainders are the same and the minimum remainder is unique. The proof
is complete.
Question 1.1.11. We know that 4 | 20 and 4 | 16. Do they imply 4 | 20 + 16? Again,
this is not too obvious. Try to prove it first.
Proposition 1.1.12. If a | b and a | c then a | bx + cy where x, y are arbitrary integers.
In particular, a | b ± ak, a | b + c, and a | b − c.
Proof. Since a | b, there is an integer k so that b = ak. Similarly, there is an integer ℓ
so that c = aℓ. Therefore,
bx + cy = akx + aℓy = a(kx + ℓy),
which is certainly divisible by a since it has a factor a in it.
Note. Here, x and y can be negative integers as well. That is how negative numbers
may come into play even when we start off with positive integers.
Question 1.1.13. Let a, b, and n be positive integers such that a | n and b|n. Do we
have ab | n?
This is a very common mistake among new problem solvers. Consider an integer
that is divisible by both 6 and 4. Is that integer divisible by 4 · 6 = 24? If you think
the answer is yes, think again! How about 12?
In general, the answer is a big NO. Try to find a few more counterexamples and
then find the condition when we can be certain that ab | n if a | n and b | n. We will
now focus on prime divisors.
Proposition 1.1.14. Any integer n greater than 1 has a prime divisor.
Proof. If n itself is a prime, we are done. So, assume n is composite. By definition,
n has a proper divisor, i.e., there is an integer greater than 1 and less than n which
divides n. Call this divisor d. Now, if d is not a prime, then d has a divisor too. We can
continue like this until we reach a point where d does not have a proper divisor greater
than 1. We know by definition that only a prime number does not have a proper divisor
other than 1. Therefore, that divisor must be a prime.
22 1.1. DEFINITIONS AND PROPOSITIONS

Corollary 1.1.15. Let n be a positive integer larger than 1. The smallest divisor of n
is a prime.
You should be able to prove the following proposition by yourself. Even if you can
not prove it formally, at least make sense of why this is true. We will not prove it here.
Try not to skip it and move on. Can you use induction to prove it? How about using
the facts you already know?
Proposition 1.1.16 (Euclid’s Lemma). Let a and b be two integers. If p is a prime
and p | ab, then p | a or p | b.
Corollary 1.1.17. If a prime divides the product of some integers, it divides at least
one of them.
√
Proposition 1.1.18. Every composite n has a prime factor less than or equal to n.
Proof. Since n is composite, it has at least one proper prime divisor. Consider the
smallest prime factor p of n and write n = pk for some integer k. Of course p ≤ k,
because otherwise according to corollary ( ), the smallest prime factor would be k
or one of its divisors. Therefore,
n = pk ≥ p2 ,
√
which in turn implies p ≤ n.
This proposition is quite useful to test if a number is prime or not. It implies
√ that
it suffices to check if n is divisible by any of the primes less than or equal to n. If
it is, then n is not a prime. Check this with some simulations by hand, say for 11, 25,
and 479. However, the test can be quite lengthy and tedious if n is too large and its
smallest prime factor is not small. If you are curious how lengthy this test can be, take
the number 357879581 and try finding its smallest prime divisor.
Definition 1.1.19 (Prime Factorization). Prime factorization is the process of find-
ing all the prime factors of a positive integer.
The fact that every positive integer greater than 1 has a prime factorization gives
birth to the Fundamental Theorem of Arithmetic.
Theorem 1.1.20 (Fundamental Theorem of Arithmetic). Every positive integer
n larger than 1 can be written as a product of primes in a unique way. We write this
factorization as
(1.1) n = pe11 pe22 · · · pekk ,
where p1 , p2 , . . . pk are different primes and e1 , e2 , . . . ek are positive integers. Using
product notation, we can write equation ( ) as
k
Y
n= pei i , or
i=1
Y
n= pe .
p|n

2
A traditional induction process. Prove the claim for a base case, say n0 . Assume the claim is true
for n = m, then prove it for n = m + 1.
CHAPTER 1. DIVISIBILITY 23

In the second equation above, p runs through different primes dividing n and e is the
maximum power of p that divides n.
Proof. We first prove by induction that the factorization indeed exists. Suppose that all
numbers k such that 1 < k < n have a factorization. If n is a prime, its factorization is
obvious. Otherwise, there exist positive integers a and b both less than n such that ab =
n. By induction hypothesis, a and b both have factorizations. Let a = pα1 1 pα2 2 . . . pαk k
and b = q1β1 q2β2 . . . qlβl be prime factorizations of a and b. Then,

n = ab = pα1 1 pα2 2 · · · pαk k q1β1 q2β2 · · · qlβl .

This is a prime factorization for n and induction is complete.

Now, let us prove that the prime factorization of n is unique. Suppose that there
are two factorizations for n. That is,

(1.2) pα1 1 pα2 2 · · · pαs s = q1β1 q2β2 · · · qtβt

are two factorizations for n, where pi and qj are primes and αi and βj are positive
integers (1 ≤ i ≤ s, 1 ≤ j ≤ t). The above equation implies that p1 divides q1β1 q2β2 . . . qtβt .
By generalization of Euclid’s lemma (Corollary ( )), p1 must divide at least one of
β1 β2 βt
q1 , q2 , . . . , or qt . Without loss of generality, suppose that p1 |q1β1 . Since q1 is a prime,
the only prime divisor of q1β1 is q1 . This means that p1 = q1 and α1 = β1 . Divide both
sides of equation ( ) by p1 to obtain

pα2 2 pα3 3 · · · pαs s = q2β2 q3β3 · · · qtβt .

β
With similar reasoning, we can deduce that pα2 2 is equal to some other qj j , say q2β2 .
Continuing this process, one soon realizes that t = s and all prime powers in the left
side of equation ( ) appear in the right side, but maybe in a different order. In other
words, the two factorizations of n are the same and the uniqueness of factorization is
implied.
Example. Try to understand the following examples and match them with the factor-
ization formula:
• 12 = 22 · 3. Here, p1 = 2, e1 = 2, p2 = 3, e2 = 1.

• 180 = 22 · 32 · 5. So, p1 = 2, e1 = 2, p2 = 3, e2 = 2, p3 = 5, e3 = 1.
The factorization of n is unique no matter in what order you factor out the primes.
Also, note that all the powers are positive. We could bring primes with power zero but
that does not make any sense (because any number to the power of zero yields 1 and
the product would not be changed).
k
Q
Note. In the product n = pei i , k is the number of distinct prime factors of n. For
i=1
example, if n = 12, then k = 2 since 12 has only two distinct prime factors: 2 and 3. If
n = 180, then k = 3 because 2, 3, and 5 are the only prime factors of n.
3
P Q
Make sure you understand the notations and well.
24 1.1. DEFINITIONS AND PROPOSITIONS

As√explained above, one way to factorize a number n is to divide it by all primes less
than n. Dividing a number by another would be pretty boring for large numbers. In
order to simplify the process, next section provides some rules for divisibility by some
specific numbers like 3, 5, 7, 11, etc.

1.1.1 Divisibility by Certain Numbers

As soon as one is introduced to the concept of natural numbers, one can make sense
of odd and even numbers. We all know that an even number, namely a number that is
divisible by 2, must have an even rightmost digit (ones digit). We also know the rules
of divisibility for 5: a number is divisible by 5 if and only if its ones digit is. Let us
make our knowledge concrete and write a proof for these two simple facts.
Proposition 1.1.21 (Divisibility by 2). A number is divisible by 2 if and only if its
last digit is even (one of 0, 2, 4, 6, 8).
Proof. Suppose that one has the number x = xk−1 xk−2 . . . x0 , where xi (0 ≤ i ≤ k − 1)
are its digits (so the number has k digits). One can write

x = xk−1 xk−2 . . . x1 0 + x0 = 10 · xk−1 xk−2 . . . x1 + x0 .

It is now clear that the first term in the right-hand side of the latest equation is divisible
by 2 (because 2 | 10). So, x is divisible by 2 if and only if x0 is even.
The proof of the rule of divisibility by 5 is very similar and left as an exercise for
the reader.
Proposition 1.1.22 (Divisibility by 5). A number is divisible by 5 if and only if it
has 0 or 5 as last digit.
So far, we discovered and proved the rules of divisibility by 2 and 5. Is it always
possible to find a divisibility rule for division by a given arbitrary number n? Maybe
the question is somehow unclear at the moment. We need to define a divisibility rule
first:
Definition 1.1.23. Let n > 1 be any integer. A divisibility rule for n is defined to be
a process which leads to determining whether a given natural number is divisible by n.
Let’s think about the question again. Can we find a divisibility rule for any n > 1?
The answer is obviously yes. With our definition of a divisibility rule, you can consider
division by n as a divisibility rule. One can always divide a natural number by n and
find the remainder of the division. The number is divisible by n if and only if this
remainder is equal to zero.
Well, of course, that is not what we had in mind. We are looking for divisibility
rules that make our life simpler, not the ones which ask you to do it by brute force. So,
it would be wise to refine our definition:
Definition 1.1.24. Let n > 1 be any integer. A proper divisibility rule for n is a
divisibility rule which uses a recursive algorithm. From now on, by a divisibility rule,
we mean a proper one.
CHAPTER 1. DIVISIBILITY 25

As an example, recall the rule of divisibility by 3. To find out the remainder of a

number upon division by 3, we add up all the digits to create a smaller number. This is
the recursive step: we are doing a simple process (adding the digits) to transform our
initial given number (which we suppose is large) into a new number which is smaller in
size. Then, we find the remainder of the new smaller number by 3. We know from high
school (or elementary school) that the remainder of the new number when divided by
3 is the same as that of the original number.
In general, we are looking for divisibility rules which explain a method for creating
a smaller number from an initially large given number. This new number has one thing
in common property with the initial number: its remainder when dived by n. We can
repeat this algorithm as many times as we wish until we reach numbers small enough
for us to do the division by hand. We know that our iterative algorithm will terminate
at some point because you cannot go smaller than 1.
It is quite easy to see that the algorithm for checking divisibility by 3 is indeed
very fast and terminates quickly. That is, the algorithm is time-efficient. Suppose that
you have a 100 digit number and you want to check its divisibility by 3. Dividing the
number by 3 to find the remainder is the worst thing to do because there is probably
no calculator which can do this task for you and you would need to do the division by
hand. Now, if you use the algorithm provided above, you would need to sum up all
the digits of the number, which would be at most 100 × 9 = 900 (in case where all the
digits are the largest, 9), and then find the remainder of this sum upon division by 3. If
you are reading this text right now, you must be able to divide any three digit number
by 3 without using a calculator. So, this algorithm terminates after one iteration in
this case because you will find the remainder right after finding the sum of the digits.
Considering we started with a 100 digit number, that is assumed to be a pretty fast
algorithm.
Let’s see the mathematical idea behind the divisibility rule for 3.
Proposition 1.1.25 (Divisibility by 3). A number is divisible by 3 if and only if the
sum of digits of the number is divisible by 3.
Example. Take the number 951 which has a sum of digits 9 + 5 + 1 = 15, divisible by
3. According to our claim, 951 should be divisible by 3 and indeed it is: 951 = 3 · 317.
Proof. The secret lies behind the fact that 10 = 3 · 3 + 1. Write your initial number
x = xk−1 xk−2 . . . x0 as
x = 10k−1 xk−1 + 10k−2 xk−2 + · · · + 10x1 + x0
= (3 · 3 + 1)k−1 xk−1 + (3 · 3 + 1)k−2 xk−2 + · · · + (3 · 3 + 1)x1 + x0
Try to show as an exercise that the remainder of division of (3·3+1)i (for 1 ≤ i ≤ k −1)
is always equal to 1 (hint: you might want to use theorem ( ) (binomial theorem)).
In other words, there exist integers qi (for 1 ≤ i ≤ k − 1) such that (3 · 3 + 1)i = 3qi + 1.
Therefore,
x = (3 · 3 + 1)k−1 xk−1 + (3 · 3 + 1)k−2 xk−2 + · · · + (3 · 3 + 1)x1 + x0
= (3qk−1 + 1)xk−1 + (3qk−2 + 1)xk−2 + · · · + (3q1 + 1)x1 + x0
= 3(qk−1 + qk−2 + · · · + q1 ) + xk−1 + xk−2 + · · · + x1 + x0
26 1.1. DEFINITIONS AND PROPOSITIONS

It is now clear that 3(qk−1 + qk−2 + · · · + q1 ) is always divisible by 3. This means that
the remainder of x upon division by 3 is the same as that of xk−1 + xk−2 + · · · + x1 + x0 ,
which is the sum of digits of x. It remains to verify the other direction, but all we did
can be reversed and so the other direction also follows. The proof is complete.
You might be curious if we can find such simple proper divisibility rules for other
numbers. Unfortunately, the recursive algorithms we find for (most of) other numbers
are not always time-efficient like that of 3 and it is possible that we need to wait for
several iterations for the algorithm to terminate.
Another issue that we need to discuss is the following: for which numbers n do we
really need a divisibility rule? For instance, now that we know the divisibility rules
for 2 and 3, do we really need another rule for divisibility by 6? No! We know that a
number is divisible by 6 if and only if it is divisible by both 2 and 3. If the previous
sentence is not clear for you, think about it for a few minutes and investigate some
examples to see why it is true. You don’t need to know a rigorous proof of this fact;
just convince yourself that it is true.
Proposition 1.1.26 (Divisibility by 6). A number is divisible by 6 if and only if it
is even and divisible by 3.
We got that there is no need for us to define a divisibility rule for 6 because we
already know rules for 2 and 3. There is nothing special about 6 = 2 · 3. In general, we
do not need a divisibility rule for pq, where p and q are distinct primes, if we already
have rules for p and q.
Proposition 1.1.27 (Divisibility by pq). Let p and q be distinct primes. A number
is divisible by pq if and only if it is divisible by both p and q.
Proof. The only if part is pretty obvious: if n is divisible by pq, then it is divisible by
both p and q. To prove the if part, assume that a positive integer n is divisible by both
p and q. From divisibility by p, we can write n = pk for some integer k. Since p and q
are different, p is not divisible by q. Since n = pk is divisible by q, we must have that
q divides k. This means that k = ql for an integer l. Finally, n = pql, which implies
that n is divisible by pq.
You have probably figured out where this is going: we only need divisibility rules
for n when n is either a prime or a power of a prime. The other cases would follow
from the next corollary of proposition ( ) (can you explain why?).
Corollary 1.1.28. Let m and n be two positive integers (not necessarily primes) which
do not share any common divisors. Then, a number is divisible by mn if and only if it
is divisible by both m and n.
We said that in order to find divisibility rules for all natural numbers, we need a
divisibility rule for primes as well as their powers. Finding divisibility rules for powers
of primes other than 2, 3, and 5 would not be something of interest for our book. We
will discuss only divisibility rules for powers of 2. Let’s see the most basic case, 4 = 22 .
Proposition 1.1.29 (Divisibility by 4). A number is divisible by 4 if and only if the
number formed by its last two digits is divisible by 4.
CHAPTER 1. DIVISIBILITY 27

Example. 202 390 2348 has the last two digits 4 and 8 which make the number formed
by its last two digits 48. Since 48 is divisible by 4, the number 202 390 2348 is divisible
by 4.
What about 23 = 8?
Proposition 1.1.30 (Divisibility by 8). A number is divisible by 8 if and only if the
number formed by its last three digits is divisible by 8.
If you have a curious mind, you should already notice a pattern in the divisibility
rules for 2, 4, and 8. For 2, we only check the last digit. For 4, we check the last two
digits and for 8, the last three digits. Do you see the pattern now? 2 = 21 , 4 = 22
and 8 = 23 . You can easily check that the same is true if we take 16 = 24 . To check
divisibility by 16, it suffices to test the number formed by the last 4 digits. We can
generalize this result for 2k .
Theorem 1.1.31 (Divisibility by 2k ). Let k be a positive integer. A number x is
divisible by 2k if and only if the number formed by the last k digits of x is divisible by
2k .
Proof. To prove this one, suppose that you have an n digit number x, represented as
x = xn−1 . . . xk xk−1 . . . x1 x0 .
Then, note that
x = xn−1 . . . xk+1 000
| {z. . . 0} + xk xk−1 . . . x1 x0
k times
k
= 10 · xn−1 . . . xk+1 + xk xk−1 . . . x1 x0 .
Since 10k = 2k · 5k is divisible by 2k , we get the conclusion.
Let’s find a divisibility rule for 7.
Proposition 1.1.32 (Divisibility by 7). A number is divisible by 7 if and only if the
difference of the number formed by the last three digits and the rest of digits is divisible
by 7.
Proof. Notice that if x = xn−1 . . . x1 x0 , then
x = xn−1 . . . x1 x0
= xn−1 . . . x3 000 + x2 x1 x0
= 103 · xn−1 . . . x3 + x2 x1 x0 .
The remainder of division of 1000 by 7 is 6. So, there exists a positive integer q such
that 1000 = 7q − 1 (why is that? find q). Hence,
x = (7q − 1) · xn−1 . . . x3 + x2 x1 x0
= 7q · xn−1 . . . x3 + (x2 x1 x0 − xn−1 . . . x3 ) .
Now, if x is divisible by 7, then so must be x2 x1 x0 − xn−1 . . . x3 and vice versa. The
proof is complete.
28 1.1. DEFINITIONS AND PROPOSITIONS

Example. Take the number 13 111. To see if it is divisible by 7 or not, first separate
the number into two parts: Form a number with last three digits and another with the
other part. In this case, we have 111 and 13. Their difference is 98, which is divisible
by 7. According to the rule of divisibility by 7, this number is divisible by 7.
There are other rules for divisibility as well, and we are just suggesting a selected
one for each number. You might have seen other divisibility rules for 7 (and other
numbers). Another famous divisibility rule for 7 is the following: take the last digit of
the number, double it, and then subtract the result from the number formed by the
rest of the digits. The resulting number would have the same remainder upon division
by 7. For instance, the steps for 13 111 would be
13 111 =⇒ 1 311 − 2 = 1 309
1 309 =⇒ 130 − 18 = 112
112 =⇒ 11 − 4 = 7,
and this verifies that 13 111 is divisible by 7. Try to prove this new divisibility rule for
7.
It is very important to bear in mind that each divisibility rule is time-efficient when
applied to a number with proper number of digits. For instance, the divisibility rule
for 7 given in proposition ( ) works best for numbers with a few digits (say, 6 to
10 digit numbers) and if your number has, say, 100 digits, you would need to wait for
a long time for the algorithm to terminate because you are removing three digits from
your number at each step and a 97 digit number is not much different from a 100 digit
number when it comes to computation: they are both huge! Try to figure out for what
range of numbers the other divisibility rule for 7 works best.
Proposition 1.1.33 (Divisibility by 9). A number is divisible by 9 if and only if the
sum of its digits is divisible by 9.
Proposition 1.1.34 (Divisibility by 11). A number is divisible by 11 if and only if
the difference of sums of alternating digits is divisible by 11.
Example. Take 12 047 816. The sum of digits in even places is 2 + 4 + 8 + 6 = 20 and
sum of digits in odd places 1 + 0 + 7 + 1 = 9. Their difference is 20 − 9 = 11, which is
divisible by 11. You can easily check that
12 047 816 = 11 · 1 095 256,
which verifies our test.
Proposition 1.1.35 (Divisibility by 13). A number is divisible by 13 if and only if
the result of addition of four times the last digit and the the number formed by rest of
the digits is divisible by 13.
Example.
8658 =⇒ 865 + 4 · 8 = 897
897 =⇒ 89 + 4 · 7 = 117
117 =⇒ 11 + 4 · 7 = 39.
And 39 = 13 · 3, so 8658 is divisible by 13.
CHAPTER 1. DIVISIBILITY 29

Proposition 1.1.36 (Divisibility by 17). A number is divisible by 17 if and only if

the result of subtraction of five times the last digit from the number formed by rest of
the digits is divisible by 17.
Example.
11 322 =⇒ 1 132 − 5 · 2 = 1 122
1 122 =⇒ 112 − 5 · 2 = 102
102 =⇒ 10 − 5 · 2 = 0.
So 11 322 is divisible by 17.
Proposition 1.1.37 (Divisibility by 19). A number is divisible by 19 if and only if
the result of addition of two times the last digit and the the number formed by rest of
the digits is divisible by 19.
Example.
12 654 =⇒ 1 265 + 2 · 4 = 1 273
1 273 =⇒ 127 + 2 · 3 = 133
133 =⇒ 13 + 2 · 3 = 19.
Therefore 12 654 is divisible by 19.
Although the above propositions can be easily proved by modular arithmetic, it
would be a great exercise for the reader to prove them with elementary tools we have
discussed so far.
Problem 1.1.38. Find a divisibility rule for 37.
Hint. Use the fact that 999 = 1000 − 1 is divisible by 37.

1.2 gcd and lcm

Consider 18 and list all its divisors in your mind. Do the same for 27. What do these
two lists have in common? If you have done the calculations correctly, you should come
up with 1, 3, and 9. Among these common divisors of 18 and 27, 9 is the largest. So,
we say that 9 is the greatest common divisor of 18 and 27.
Consider 18 and 27 again. This time, list all their integral multiples mentally. The
list of multiples of 18 is
18, 36, 54, 72, 90, 108, . . . ,
whereas the list of multiples of 27 is
27, 54, 81, 108, . . . .
There are infinitely many common numbers shared by these two lists: 54, 108, . . . . The
smallest number in the latter list is 54, which we call the least common multiple of 18
and 27.
You may ask, what makes you think there will even be a common element in both
sets? Well, we will have at least 1 common as a divisor since it divides all the integers.
30 1.2. GCD AND LCM

Definition 1.2.1 (gcd and lcm). For two integers a and b which are not zero at the
same time, the greatest common divisor of a and b, denoted by gcd(a, b), is the greatest
positive integer which divides both a and b. For brevity, we denote this by (a, b) in this
book.
The least common multiple of a and b, denoted by lcm(a, b), is the smallest positive
integer that is divisible by both a and b. Again, for brevity, we denote this by [a, b] in
this book.
The concept of gcd and lcm is the same for more than two integers. The greatest
common divisor of a1 , a2 , . . . , an is the largest positive integer which divides them all.
We denote this by (a1 , a2 , . . . , an ). One can define [a1 , a2 , . . . , an ] in a similar way.
Example. (18, 27) = 9 and [18, 27] = 54. (18, 27, 36) = 9 and [18, 27, 36] = 108.
Note. The above definition of gcd is equivalent to the following: if (a, b) equals g, then
g divides both a and b. Furthermore, if there is a positive integer c for which c|a and
c|b, then g ≥ c. Analogously, if [a, b] equals ℓ, then ℓ is divisible by both a and b.
Moreover, if there is a positive integer c for which a|c and b|c, then c ≥ ℓ.
Proposition 1.2.2 (Properties of gcd and lcm). Let a and b be two positive inte-
gers. The following statements are true.
1. (a, b) = (b, a) = (a, −b) = (−a, −b) and [a, b] = [b, a] = [a, −b] = [−a, −b].

2. (a, 0) = a, [a, 0] = 0, (a, 1) = 1 and [a, 1] = a.

3. (a, a) = [a, a] = a.

4. [a, b] ≥ (a, b).

5. a|b if and only if (a, b) = a. Similarly, a|b if and only if [a, b] = b

6. For any integer k, (a, b + ak) = (a, b) and (ka, kb) = k(a, b). Furthermore,
[ka, kb] = k[a, b].

7. For any non-negative integer n, we have (an , bn ) = ((a, b))n and [an , bn ] = ([a, b])n .

8. For any two integers x and y, we have (a, b)|ax + by.

9. If p is a prime divisor of a or b, then p|[a, b].

10. For any prime divisor p of (a, b), we have p | a and p | b.

11. If p is a prime, then

(
p, if p | a
(a, p) =
1, otherwise

The proofs are pretty obvious and the reader is encouraged to prove them as an
exercise. We only provide a hint for part 4: use part 10 of proposition ( ).
We will now prove part 12 of proposition ( ) as a problem.
CHAPTER 1. DIVISIBILITY 31

Problem 1.2.3. For integers a and b, if an | bn for some positive integer n, then a | b.

Solution. Let g = (a, b). By part 7 of proposition ( ), g n = (an , bn ). On the

other hand, an |bn implies (an , bn ) = an by part 5 of the same proposition. Hence,
(a, b)n = an = g n . Taking root n, we find (a, b) = a = g. Using part 5 again, one
obtains a|b. The proof is complete.

Proposition 1.2.4. For positive integers a, b, and c, if c | a and c | b, then c | (a, b).
Analogously, if a | c and b | c then [a, b] | c.

Definition 1.2.5. Two positive integers are coprime or relatively prime to each other
if their greatest common divisor is 1. We shall use a ⊥ b to denote that (a, b) = 1, i.e.,
that a and b are coprime.

Example. 3 ⊥ 4, but 4 6⊥ 6 since 2 divides both 4 and 6.

Proposition 1.2.6. Let a, b, and c be three integers. The following statements hold
true.

1. If a ⊥ b, a | c, and b | c, then ab | c.

2. If a⊥b and a | bc, then a | c.

3. If a ⊥ b and a ⊥ c, then a ⊥ bc.

4. If a ⊥ b, then am ⊥ bn for all non-negative integers m, n.

5. If a ⊥ b, then [a, b] = ab.

6. If a ⊥ b, then (a, bc) = (a, c).

7. If p and q are distinct primes, then p ⊥ q.

Note (1). Part 2 is really useful in solving problems. It is actually the general form
of proposition ( ). Be careful not to use this part incorrectly. To be precise, one
cannot imply from a | bc that a | c unless a⊥b. We state a proof to this part in the
solution of problem ( ).

Note (2). If a ⊥ b and a ⊥ c, it is not necessary that b⊥c. Can you think of an
example?

Given integers a and b, how can we calculate (a, b)? Chances are you already know
how to do this from elementary school. There are two common ways to do that. The
first way is to factorize both numbers and find their common factors. Although we use
this method a lot, it may not be wise to factorize very large numbers just to find their
gcd. In such cases, the second way comes helpful: the Euclidean algorithm.

Theorem 1.2.7 (Euclidean Algorithm). Let a and b be two positive integers. Divide
b by a and write b = aq + r, where q is an integer and 0 ≤ r < a. Then (a, b) = (a, r).
32 1.2. GCD AND LCM

Proof. Let g = (a, b). We already know that g|a and g|b. Since b = aq + r, it follows
that g|aq + r. On the other hand, g | a implies g|aq. As established in proposition
( ), we can subtract these two divisibility relations and find g | aq + r − aq, or g | r.
This means that g divides r too. That is, the greatest common divisor of a and b is a
divisor of r too. In order to show that g = (a, r), there is only one thing remained to
prove: if there exists some c for which c | a and c|r, then g ≥ c. We will prove a stronger
argument: if such a c exists, then c|g (since c and g are both positive, c | g implies
g ≥ c). Note that c|a gives c|aq. Adding the latter relation with c|r yields c | aq +r = b.
From proposition ( ), we see that c | (a, b) = g. The proof is complete.
Euclidean algorithm is pretty useful because it helps us find the gcd only by a
series of divisions. Suppose that you have two extremely large numbers a and b such
that b > a. First, find the remainder r of b upon division by a. The remainder is
strictly less than a, and there is a chance that it would be much smaller than b because
b > a > r. According to Euclidean algorithm, instead of finding gcd of a and b, we can
calculate (a, r). In the next step, find the remainder r1 of a upon division by r. Then
(a, b) = (a, r) = (r, r1 ). As you might have noticed, the numbers are becoming smaller
and smaller. Continuing the divisions, you will reach a point where the numbers are
small enough to compute the gcd by hand.
Corollary 1.2.8. Let n, a, and b be positive integers such that n | a − b. Then (n, a) =
(n, b).
The proof of this corollary is exactly the same as the previous theorem. If you look
closely, you will find that we did not make use of the given inequality 0 ≤ r < a in the
process of proving Euclidean algorithm. In fact, that condition is given only to make
the algorithm more efficient.
To make sense of Euclidean algorithm, follow the next example.
Example. Let’s find (112, 20). The first division 112 = 20 · 5 + 12 gives (112, 20) =
(20, 12). By performing the second division, 20 = 12 · 1 + 8, we obtain(20, 12) = (12, 8).
The next steps result in (12, 8) = (8, 4) and (8, 4) = (4, 0). The gcd of any number and
zero is the number itself. So, (20, 112) = 4. If you understood the process correctly,
you will know that the algorithm terminates when one of a or b becomes 0.
The other way of finding the greatest common divisor relies on factorization. Before
stating the related proposition for this method, we solve the previous example again by
factorizing.
Example. First, factorize both 20 and 112:

20 = 22 · 5,
112 = 24 · 7.

You can easily find that since the prime factor 5 does not appear in the factoring of
112, 5 cannot appear in (20, 112). For the same reason, 7 will not appear in the gcd as
well. In other words, we just have to consider the primes that are in both 112 and 20.
We are left with the only prime 2. The point is to pay attention to the power of 2 that
4
An algorithm means a set of operations in a certain process to solve a problem or to find something.
CHAPTER 1. DIVISIBILITY 33

divides the numbers. 112 is divisible by 24 . However, 20 is divisible by 22 but not by

any higher power of 2. Therefore, (20, 112) cannot have a power of 2 greater than 22
(otherwise it will not divide 20). Since we are looking for the greatest common divisor,
we will take 22 as proper power of 2 in the gcd. Since there is no other prime to take
care of, we have (112, 20) = 22 .
This method works for the lcm as well:
Example. Take 180 = 22 · 32 · 51 and 105 = 31 · 51 · 71 . The lcm of these two numbers is
divisible by both of them, so it must contain all their prime factors. That is, [180, 105] =
2a 3b 5c 7d for some positive integers a, b, c, and d. Since we are looking for the least
common multiple, we need a, b, c, and d to be as small as possible. The smallest value
for a is 2 because otherwise a = 1, then the lcm would not be divisible by 4 and
consequently not divisible by 180 either. Similarly, the smallest values of b, c, and d
are the largest power of 3, 5, and 7, respectively, which divide 180 and 105. Finally,
b = 2, c = 1, d = 1, and [180, 105] = 22 · 32 · 51 · 71 = 1260.
Do the simulation for a few more examples to convince yourself. The following
proposition formalizes this method.

Proposition 1.2.9. Let a and b be two positive integers. If a = pe11 pe22 · · · pekk and
b = pf11 · pf22 · · · pfkk , where pi are primes and ei , fi ≥ 0 are integers for 1 ≤ i ≤ k, then:
min(e1 ,f1 ) min(e2 ,f2 ) min(e ,f )
(a, b) = p1 p2 · · · pk k k ,
max(e1 ,f1 ) max(e2 ,f2 ) max(ek ,fk )
[a, b] = p1 p2 · · · pk .

In other words, using the product notation,

k
Y min(ei ,fi )
(a, b) = pi ,
i=1
Yk
max(ei ,fi )
[a, b] = pi .
i=1

Note. This idea can be generalized for finding gcd or lcm of n integers. Just factorize
the numbers and select the proper powers of primes.

Proposition 1.2.10. For two integers a and b with g = (a, b), there exist integers m
and n such that a = gm and b = gn. Moreover, m and n are relatively prime.

Proof. Integers m and n exist because g|a and g|b, but why are m and n coprime? This
is true because if there were any other common factor between m and n, that would
have been included in g too. Otherwise, g could not remain the greatest common divisor
since we can make a bigger one multiplying that common factor with g. We can think
of m and n as the uncommon factors between a and b.
5
Try to find out why ei ≥ 0 whereas we consider only ei ≥ 1 when we first introduced prime
factorization.
34 1.2. GCD AND LCM

Example. Consider 18 and 27. (18, 27) = 9 and we can write 18 = 9 · 2 and 27 = 9 · 3.
Now, 9 is the largest common factor. Here, 2 is the uncommon factor from 18 which
27 does not have besides 9, and 3 is the uncommon factor of 27 which 18 does not have
apart from 9. Thinking about m and n in this way may make more sense to you.

Proposition 1.2.11. Let g and ℓ be the greatest common divisor and the least common
multiple of positive integers a and b, respectively. If we write a = gm and b = gn with
(m, n) = 1, then ℓ = gmn.

Proposition 1.2.12. Let g = (a, b) and ℓ = [a, b]. Then ab = gℓ. In words, the product
of two positive integers is equal to the product of their gcd and lcm.

First proof. By proposition ( ), there exist coprime positive integers m and n such
that a = gm and b = gn. Therefore ab = gm · gn = g 2 mn. On the other hand, by
proposition ( ), gl = g · gmn = g 2 mn.

The following proof uses prime factorization and is somewhat more rigorous. But
the previous one makes more sense. Even though it may look uglier, it shows how to
invoke prime factorization.

Second Proof (Using Prime Factorization). Assume that we have the prime factoriza-
tion of a and b (as in proposition ( )):

a = pe11 · · · pekk and b = pf11 · · · pfkk .

Now, can you understand the simple fact that min(x, y) + max(x, y) = x + y? If so,
then the proof should be clear to you. In fact,

ab = pe11 . . . pekk · pf11 . . . pfkk = p1e1 +f1 . . . pkek +fk .

On the other hand, from proposition ( ),

min(e1 ) min(ek ) max(e1 ,f1 ) max(ek ,fk )

(a, b) · [a, b] = p1 · · · pk · p1 · · · pk
min(e1 ,f1 )+max(e1 +f1 ) min(ek ,fk )+max(ek ,fk )
= p1 · · · pk
= p1e1 +f1 · · · pkek +fk
= ab.

Question 1.2.13. Let a, b, and c be positive integers. Does the equation

abc = (a, b, c) · [a, b, c]

hold? If not, why?

CHAPTER 1. DIVISIBILITY 35

1.3 Numeral Systems

1.3.1 Introduction
There is a rumor that Pascal once promised 1000000 dollars to anyone who marries
his daughter. Later, when the son-in-law asked for the money, Pascal gave him only 64
dollars. Poor guy! If you got the joke, you are probably good with bases. If not, do
not worry, keep reading and you will get the point.
In daily mathematics, meaning the math you face in real life, you unconsciously
express numbers in decimal system. That is, when you say you have 15 apples, you
are using base 10 without ever realizing it. The question here is that what is this base
actually? Let us start with a simple example.
Example. Consider the number 573. Have you ever thought why we write digits in this
way to denote a number? The reason is that every digit in the number represents the
coefficient of a power of ten. That is,
573 = 5 · 102 + 7 · 101 + 3 · 100 .
Rigorously talking, each integer has to be written in a base for it to make sense. For
example, the number 15 has different values when expressed in base 6 and in base 10.
Actually, 15 in base 6 is equal to 11 in base 10. Probably it still does not make sense.
If so, just keep reading.
All of our calculations in daily life are done in base 10, which is called the decimal
system. However, this does not mean that we cannot present numbers in any other
bases other than 10. There are a limited number of digits in each base (to be precise,
the number of digits is the same as the value of base). These digits run from 0 to the
largest integer smaller than the base. Therefore, in base 10 the digits are 0, 1, . . . , 9 .
Similarly, in base b the digits will be 0, 1, . . . , b − 1. Observe the base-10 representation
of 573 in the above example again. The rightmost digit is multiplied by 1. The middle
digit, 7, is multiplied by 10 (hence the name tenths digit). Finally, the leftmost digit
is multiplied by 100 (the hundredths digit). In simpler words, each time we go left, we
multiply the multiplier by 10. If it were base b instead of base 10, we would multiply by
b each time. So, the multipliers would be 1, b, b2 , . . . and so on. You should understand
the formal representation of an integer in base b (for b > 1 an integer). Note that it is
pointless to take base 1 since we do not have any meaningful digit (recall that 0 is not
a meaningful digit). You can think of these multipliers as weights (or contribution) of
those digits.
Definition 1.3.1 (Base b Representation). Let x and b be positive integers such
that b > 1. In a numeral system , the number x and its base b are written together as
(x)b . If the digits of an n-digits number x are represented as xn−1 xn−2 . . . x0 in base b,
then
(x)b = (xn−1 xn−2 . . . x1 x0 )b = bn−1 xn−1 + bn−2 xn−2 + · · · + b1 x1 + x0 ,
6
Take a wild guess why we use base 10 and not any other base.
7
A numeral system (or system of numeration) is a writing system for expressing numbers using
digits or symbols.
36 1.3. NUMERAL SYSTEMS

Base System Name

2 Binary
3 Ternary
4 Quaternary
8 Octal
10 Decimal
16 Hexadecimal
36 Hexatrigesimal
60 Sexagesimal

Table 1.1: Common numeral systems and their names.

where x0 , x1 , . . . , xn−1 are non-negative integers less than b.

The base is written in the subscript, while the digits are inside the bracket (or a
line is drawn over the digits) to clarify it is not a product. An important note to be
mentioned is the fact that xn−1 is always non-zero, because if it is zero, then there is
no point in keeping it as the leftmost digit. That is, we can remove this digit until we
get a nonzero digit as the leftmost digit. The rightmost digit of x is denoted by x0 ,
and it is called the least significant digit of x. Analogously, the leftmost digit, xn , is
called the most significant digit of x. You should already be able to guess where these
names come from! For instance, in decimal system, observe that the rightmost digit has
multiplier 1, and so contributes the least. The largest digit 9 contributes 9 if it is in the
rightmost position. On the other hand, if 1 is the leftmost digit (say thousandth), then
it has multiplier (you can think of it as a weight in this regard as well) 1000, which is
a lot more than 9. So, 1 contributes the most, and naturally it is the most significant
digit. Same explanation applies for the least significant digit.
Example. The number (327)8 is calculated as

(327)8 = 3 · 82 + 2 · 8 + 7 = 215.

Note. When the base is absent in a representation, it is regarded to be 10 by default.

Different bases maybe used in different systems. For example, in computer science,
it is conventional to represent the numbers in base 16 or 8. Another example is base
60 which was used by ancient Summerians in the 3rd millennium BC. In order to avoid
repeating the base numbers, we use a specific name for popular bases. You can find a
list of such names in the following table.

1.3.2 Base Conversion

After defining bases, the first problem that we face is defining the relationship between
the numbers in different bases. For many of us, we can only make sense of numbers
when they are represented in decimal system. For example, you may have no clue about
the real value of (1234)5 , but you surely know what is 1234 (if we do not write the base,
it is 10 by default).
CHAPTER 1. DIVISIBILITY 37

1.3.2.1 Conversion from Base b to Base 10

In order to understand the meaning of numbers (their value) in a particular base b, we
need to convert them to a number in base 10. In the previous given example, we showed
how to convert (327)8 to base 10. The process of converting base b < 10 to base 10 is
directly resulted from definition ( ). However, we have an issue when the base is
larger than 10. Before stating the process of conversion, think about this: what happens
if the base, b, is greater than 10? Then we have a problem in representing the digits.
For example, in the hexadecimal system with base of 16, the digits must be less than 16.
Therefore, we should be able to represent digits 10 to 15 in hexadecimal system. But
how is it possible to have a two-digits number as one digit in an hexadecimal number?
In order to avoid the confusion, we use the following notation for digits bigger than ten:

10 = A, 11 = B, 12 = C, 13 = D, 14 = E, 15 = F.

Example.

(2FE05)16 = 2 · 164 + 15 · 163 + 14 · 162 + 0 · 16 + 5

= 131, 072 + 61, 440 + 3, 584 + 5
= 196, 101.

We know how to convert numbers in any base to a number in base 10. The next
step is exactly the opposite: converting numbers in decimal system to other bases. This
is quite interesting as well. Even it is possible that you have learned the method from
school. However, instead of focusing on the method, let us focus on finding a way to
do it.

1.3.2.2 Conversion from Base 10 to Base b

Suppose that we want to convert the number y in decimal system to a number in base
b. In other words, we want to write y as (xn xn−1 . . . x1 x0 )b so that x0 , x1 , . . . xn are the
digits of y when represented in base b. By definition, we know that y can be written as

(1.3) y = bn xn + bn−1 xn−1 + · · · + b1 x1 + x0 ,

where 0 ≤ xi < b for 0 ≤ i ≤ n. Our aim is to find the value of xi ’s. Look at the
equation carefully and try to understand how we can retrieve the digits in base b. If
you are stuck, then go forward. Rewrite equation ( ) as


y = b · bn−1 xn + bn−2 xn−1 + · · · + x1 +x0 .
| {z }
y1

We have written y as y = by1 + x0 , which means that the remainder of y when divided
by b is x0 . In fact, the rightmost digit of y in base b (x0 here) is the remainder of y
when divided by b. As you can see, we are simply using the division theorem here (does
it make sense now?). Bases are related to divisibility after all!
38 1.3. NUMERAL SYSTEMS

To find the next digit, x1 , we have to divide the quotient of the above division, y1 ,
by b. The reason is the same as above: rewrite y1 as


y1 = b · bn−2 xn + bn−3 xn−1 + · · · + x2 +x1 .
| {z }
y2

Therefore, x1 is the remainder of y1 when divided by b. We can find the next digit x2
by finding the remainder of y2 when divided by b. The digits x3 , x4 , . . . , xn−1 can be
found similarly by continuing this process. The leftmost digit, xn , is the quotient of
the last division because we can no longer divide it by b (since the digits are all less
than b).
Example. Let us find the representation of 215 in base 8 and base 2. Start with base
8 first. We initialize the process by dividing the given number by 8. The remainder of
this division is the rightmost digit of 215 in base 8. Then we divide the quotient by 8
again. The remainder of this division is the digit before the rightmost one. Since the
quotient of this division is less than 8, the process is over. Using long division:

26
3
8 215 8 26
160 24
55 2
48
7

Figure 1.1: Conversion process of 215 from decimal system to octal (base 8).

In different countries, people use different ways to demonstrate long division. The
above way is usually used in English-speaking countries such as USA or Canada. In
European and Asian countries, people (mostly) use the following way of doing long
division:

215 8 26 8
− −
16 26 24 3
55 2
−
48
7

Figure 1.2: Conversion process of 215 from decimal to octal with divisions done in a
different way (common in Europe and Asia).

Thus, 215 = (327)8 , which matches the result in previous section. To find 215 in
base 2, we do the divisions as follows. If this is not how you do long divisions, just do
them in your own way.
CHAPTER 1. DIVISIBILITY 39

107
53
26
13
6
3
1
2 215 2 107 2 53 2 26 2 13 2 6 2 3
200 100 40 20 12 6 2
15 7 13 6 1 0 1
14 6 12 6
1 1 1 0

Figure 1.3: Conversion process of 215 from decimal to binary (base 2).

Reading from right to left, the quotient of the first division is the leftmost digit, and
the remainders of divisions form the other digits. So, the result is (11010111)2 .
Example. Assume we want to convert the number 196 101 to base 16.

1961 01 16 12256 16 7 6 6 16 47 16
− − − −
16 12256 112 766 6 4 47 32 2
36 105 1 2 6 15
− − −
32 96 1 1 2
41 96 1 4
− −
32 96
9 0 0
−
8 0
1 01
−
96
5

Figure 1.4: Conversion process of 196 101 from decimal to hexadecimal (base 16).

If you write down the divisions like we did in figure ( ), you can see that the last
quotient is the most significant digit and the remainders of the divisions (from right to
left) form the other digits (as you move from the last remainder to the first one, the
significance of the digits decreases). We have drawn circles around those digits. Finally,
writing 15 as F and 14 as E, we have

196 101 = (2F E05)16 .

1.3.2.3 Conversion Rules for Certain Bases

Assume that we want to convert a number in base 2 to base 8. How should it be done?
One approach is to convert the number to base 10 first, and then convert it to base
8
Thanks to AndréC for writing the LaTeX code for the figure.
40 1.3. NUMERAL SYSTEMS

8. You should be able to do this procedure by now. However, we prefer to do the

conversion in one step, if possible, rather than two steps.
Let x = (an an−1 . . . a1 a0 )2 be our binary number, where ai is either 0 or 1 for
0 ≤ i ≤ n. Assume that we have converted this number to base 8 and the result is
x = (bm bm−1 . . . b1 b0 )8 , where 0 ≤ bj ≤ 7 for 0 ≤ j ≤ m. Our aim is to find the relation
between ai ’s and bj ’s. It might seem a bit difficult to find a relation, but it will be clear
if you write the expansion of x in both bases. For base 2, we can represent x as

x = 2n an + 2n−1 an−1 + · · · + 21 a1 + a0 .

Start clustering ai digits in groups of 3, starting at the right. For convenience, assume
that n + 1 is divisible by 3 (because there are n + 1 digits a0 , a1 , . . . , an and we are
putting them in groups of size 3). Other cases when number of digits is not divisible
by 3 will be discussed later. Then, x can be represented as


2n an + 2n−1 an−1 + 2n−2 an−2 +



2n−3 an−3 + 2n−4 an−4 + 2n−5 an−5 + · · · + 22 a2 + 21 a1 + a0 ,

which is equal to


2n−2 22 an + 2an−1 + an−2 +



2n−5 22 an−3 + 2an−4 + an−5 + · · · + 20 22 a2 + 2a1 + a0 .

Since 2a = 8a/3 for all real numbers a, we can write the above as
n−2

8 3 22 an + 2an−1 + an−2 +
| {z }
bm
n−5


8 3 22 an−3 + 2an−4 + an−5 + · · · + 80 22 a2 + 2a1 + a0 .
| {z } | {z }
bm−1 b0

The power of 8 in all of the terms in the above sum is an integer (why?). Note that the
number 22 ai +2ai−1 +ai−2 is actually (ai ai−1 ai−2 )2 , and so it is a non-negative integer less
than or equal to (111)2 = 7. This means that 22 ai +2ai−1 +ai−2 is acceptable as a digit in
base 8 (remember that digits in base b should be less than b and non-negative). Now look
at the last line of the above equations. It is of the form 8m bm +8m−1 bm−1 +· · ·+8b1 +b0 .
We have therefore found a relation between digits in base 2 and base 8:
n−2
m= , bj = 22 a3j+2 + 2a3j+1 + a3j = (a3j+2 a3j+1 a3j )2 for 0 ≤ j ≤ m.
3
There is only one point remaining: we first assumed the number of digits of the binary
number is divisible by 3 so that we can group them. What if number of digits is not
divisible by 3? It is not actually a problem. Put one or two zeros at the left side of the
binary number and make the number of digits divisible by 3, then continue the process.
The above result looks a bit scary, but it is really simple in plain English, explained
in the following theorem.
CHAPTER 1. DIVISIBILITY 41

Theorem 1.3.2 (Base 2 to 8 Conversion Rule). To convert a binary number to

base 8 directly, start grouping the 0 and 1 digits of the number in groups of 3. If number
of digits of the binary number is not divisible by 3, put one or two zeros at the left side
of the number to make it so, and then group the digits. Then convert each of these
groups into one octal digit (a digit in base 8)and rewrite the number. Conversion is
done.
Example. Let us convert (1 010 011 010)2 to base 8. Number of digits is 10, which is not
divisible by 3. So we add two zeros to the left and start the process with the number
(001 010 011 010)2 . Now,
(001)2 = (1)8 , (010)2 = (2)8 , (011)2 = (3)8 , (010)2 = (2)8 .
Hence, (1 010 011 010)2 = (1232)8 .
To convert a number from base 8 to base 2 one can do the converse process. For
the sake of completeness, we include it here.
Theorem 1.3.3 (Base 8 to 2 Conversion Rule). To convert an octal number to
base 2 directly, convert each digit to a three digit number in base 2 and put all these
three digit numbers together.
Example. Let us convert (2051)8 to a binary number. We have
(2)8 = (010)2 , (0)8 = (000)2 , (5)8 = (101)2 , (1)8 = (001)2 ,
and the result is
(2051)8 = (10 000 101 001)2 .
Notice that the zero as the most significant digit is meaningless and must be removed
after the conversion is done. We suggest you to do the conversion in a two-step method
(from base 8 to base 10 and then to base 2) to verify the answer.
You can convert a binary number to base 16 just by clustering the digits into groups
of 4 and convert each group to a hexadecimal number. In general, one can use a similar
approach to convert a binary number to base 2n (and vice versa). However, the cases
where n is larger than 4 are rarely used.
Similar approaches can be used for conversion between other numbers as well. For
example, to convert base 3 to base 9, one should start grouping the digits in groups of
2 and then do the conversion. The theorems are pretty similar to the above and we do
not mention them here.

1.3.3 Logarithms
1.3.3.1 Introduction
You are already familiar with the concept of exponentiation, which means raising a
number to an exponent. Take for example 16 = 24 . Here, 2 is called the base and 4 is
called the exponent or power.
The logarithm, first introduced by John Napier in 1614, is the inverse operation of
exponentiation. For instance, the logarithm of 64 in base 2 is 6 because 64 = 26 .
42 1.3. NUMERAL SYSTEMS

Definition 1.3.4. Let x and b be positive real numbers such that b 6= 1. The logarithm
of x to base b is the exponent by which b must be raised to yield x. In other words, the
logarithm of x to base b is the solution y to the equation

by = x.

We denote this by y = logb x.

Remark.

1. The base of the logarithm cannot be equal to 1. The reason is simple: the equation
1x = 1y holds for all the values of x and y, and the definition of logarithm to base
1 would be pointless. This PDF was created
for Amazon.

2. Some books use the notation log x (no base presented) to mean common loga-
rithm . That is, logarithm to base 10. For example, log 1000 = 3.

3. The logarithm to the base of mathematical constant e is called the natural loga-
rithm. The natural logarithm of x is generally written as ln x.

Example.

1. Let a 6= 1 be a positive real number. Then loga a = 1 and loga 1 = 0.

1
2. log5 = −3.
125
3. log1/7 49 = −2.
√
4. log49 7 = 0.25.

5. log 0.0001 = −4.

1.3.3.2 Laws of Logarithms

As in exponentiation where you can write ax ·ay = ax+y , a similar relation for logarithms
can be written. This, along with a few other properties of logarithms, are introduced
in the following proposition.

Proposition 1.3.5 (Laws of Logarithms). Let a and b be positive reals not equal
to 1. For any two positive reals x and y, we have

1. loga xy = loga x + loga y.

x
2. loga = loga x − loga y.
y
m
3. logan xm = loga x for any real m and n (with n 6= 0).
n
9
or decimal logarithm.
10
e ≈ 2.71828.
CHAPTER 1. DIVISIBILITY 43
1
4. loga b = .
logb a
logb x
5. loga x = .
logb a
Proof. The proofs are rather straightforward. We prove the first and the last parts and
leave others as exercises for the reader. For the first part, let loga x = m and loga y = n.
Then by definition, x = am and y = an . Thus,

xy = am+n ,

which in turn gives loga xy = m+n. For the last part, assume logb x = m and logb a = n.
Then x = bm and a = bn and so by part 3,
m m
loga x = logbn bm = logb b =
n n
logb x
= ,
logb a
since logb b = 1.
Note. Both loga (x + y) = loga x + loga y and loga (x + y) = loga x · loga y are false. Do
not use them.
Example. Suppose that you know the logarithm of every number to base e. How would
you find log3 5? The answer would be to use the last part of proposition ( ):
ln 5 1.6094
log3 5 = ≈ ≈ 1.4649.
ln 3 1.0986
You can easily find the values of ln 5 and ln 3 using a simple scientific calculator.
Example. Given that log2 3 ≈ 1.5849, let’s find log0.125 36. Notice that 0.125 = 1/8 =
2−3 and 36 = 22 · 32 . Therefore,


log0.125 36 = log2−3 22 · 32 = log2−3 22 + log2−3 32
2 2 2
= log2 2 + log2 3 ≈ − (1 + 1.5849)
−3 −3 3
≈ −1.7232.

Proposition 1.3.6. Let b, x1 , x2 be positive real numbers such that b 6= 1. Then

1. if b > 1, x1 > x2 if and only if logb x1 > logb x2 ,

2. if b < 1, x1 > x2 if and only if logb x1 < logb x2 .

Proof. We illustrate the proof with two sample cases. In Figure , you can see the
plot of y = log2 x for x ∈ [0, 10]. As seen in the figure, logarithm to the base 2 is
strictly increasing. That is, if x1 > x2 , then log2 x1 > log2 x2 . Now, let’s prove that if
log2 x1 > log2 x2 , then x1 > x2 . In the same figure, you can see the plot of y = 2x , which
44 1.3. NUMERAL SYSTEMS

y = 0.5x

y = log0.5 (x)

Figure 1.5: Plot of functions y = log0.5 x and y = 0.5x .

is, again, strictly increasing. It means that for any two reals x1 and x2 , if x1 > x2 , then
2x1 > 2x2 . By definition, we can find positive reals y1 and y2 such that x1 = log2 y1 and
x2 = log2 y2 . Now, the relation

x1 > x2 =⇒ 2x1 > 2x2 ,

transforms to

log2 y1 > log2 y2 =⇒ y1 > y2 ,

which is what we wanted. Analogously, as shown in Figure , the functions y = log0.5 x

and y = 0.5x are both strictly decreasing and the proof of part 2 is almost the same as
the previous part.

1.3.4 Number of Digits

There is a reason why we have explained logarithms in details even though they are
part of elementary school teachings. A good number of students can not understand
the meaning of logarithm properly. Let’s show a problem so you realize what we mean.
Question 1.3.7. How many digits does 57 have in base 2?
Solution. First, think of a much simpler question: how many digits does 57 have in
base 10? The answer is obviously 2 digits. The reason is that any number between 10
and 100 has two digits in decimal system. In other words, the inequality 10 < 57 < 102
tells us that 57 has two digits in base 10.
CHAPTER 1. DIVISIBILITY 45

y = 2x

y = log2 (x)

Figure 1.6: Plot of functions y = log2 x and y = 2x .

In general, in order to find number of digits of an integer n in base b, we must find

two consecutive powers of b such that n lies between them. In fact, if bm < n < bm+1 ,
then n has m + 1 digits in base b.
In case where n = 57 and b = 2, the required inequality is 25 < 57 < 26 . Therefore,
57 has six digits when written in binary. Indeed, 57 = (111001)2 .
We can now generalize this problem to the following proposition.
Proposition 1.3.8. Let b and n be integers bigger than 1. Number of digits of n when
presented in base b is ⌊logb n⌋ + 1.
Proof. As stated in the above solution, suppose that m is the unique integer for which
bm < n < bm+1 . Then, the number of digits of n in base b is m + 1. That is, it suffices
to prove that m = ⌊logb n⌋. Since b is larger than 1, we know from proposition ( )
that logarithm to base b is a strictly increasing function. This means that we can take
logarithm to base b from bm < n < bm+1 and the direction of the inequalities will not
change. In other words,
logb bm < logb n < logb bm+1 ,
which gives m < logb n < m + 1. Since m is an integer, by definition of floor function,
m = ⌊logb n⌋.

1.4 Some Useful Facts

In this section, we provide some theorems which may help you solve problems more
easily. Though you will find that we have emphasized not to be dependent on theorems
46 1.4. SOME USEFUL FACTS

or making them the core of solving problems, they help us a great deal and make our
lives a lot easier. Therefore, you will find a lot of theorems in this book. But it does
not mean in any way that theorems are the best way to improve in number theory.
They merely help us speed up the process of solving problems. Nothing more. One
could rediscover all the theorems while solving a problem and the end result would still
be the same. The point is: you can get to the top of a mountain in many ways and
the view is same. But the pleasure might be different based on the approach you take.
If you use a helicopter to reach the peak of the mountain or climb all the way to the
top. But definitely the latter approach brings you more pleasure. Anyway, we hope
you understand our primary intention.
Theorem 1.4.1. Two integers a and b are of the same parity if and only if their sum and
difference is even. Equivalently, they are of different parity if their sum and difference
is odd.
Corollary 1.4.2. Exactly one of the two integers a and b is even if and only if a ± b is
odd.
Theorem 1.4.3. Every positive integer n can be written in the form n = 2k s, where k
is a non-negative integer and s is an odd positive integer.
Note.
1. Here, k is the largest power of 2 that divides n. Therefore, s must be odd.
Moreover, s is the largest odd divisor of n. Notice that if n is odd, then k = 0
and s = n.

2. One can write any positive integer n as n = pk s, where p is a prime, k is a

non-negative, and s is an integer coprime to p. The case p = 2 (above theorem)
is usually used in problem solving (and sometimes even in some combinatorics
problems).
Problem 1.4.4. Let n be a positive integer and let S = {1, 2, . . . , 2n}. Choose n + 1
numbers a1 , a2 , . . . , an+1 out of S so that

a1 < a2 < · · · < an+1 .

Prove that ai | aj for some integers 1 ≤ i, j ≤ n + 1 with i 6= j.

Solution. Write all members of S in the form 2α β, where α is non-negative and β is
the odd factor. There are exactly n odd numbers among 1, 2, . . . , 2n. Therefore, by
pigeonhole principle , among n + 1 selected numbers a1 , a2 , . . . , an+1 , there exist at
least two numbers ai and aj (with i 6= j) which have the same odd factor. In other
words,

ai = 2 α 1 β and aj = 2α2 β.

Now, ai | aj if α1 ≤ α2 , and aj | aj otherwise.

11
The pigeonhole principle states that if n items are put into m containers, with n > m, then at
least one container must contain more than one item.
CHAPTER 1. DIVISIBILITY 47

Theorem 1.4.5. Every composite positive integer n can be written as ab where a and
b are coprime positive integers larger than 1.

Theorem 1.4.6. (n − 1)! is divisible by n if and only if n > 4 is a composite integer

or n = 1.

Proof. The case n = 1 is trivial. For n = 2, 3, 4 we can check by hand. For n > 4, if n
is a prime, n does not share a factor with any of 1, 2, . . . , n − 1. Consequently, n does
not divide their product (n − 1)!. If n is composite, we can write n = ab with a and b
larger than n and a 6= b. Since a 6= b and (n − 1)! contains both a and b in the product,
n = ab|(n − 1)!.

Theorem 1.4.7 (Four Numbers Theorem). Let a, b, c, and d be four positive inte-
gers such that ab = cd. Then there exist four positive integers r, s, t and u so that

a = rs, b = tu, c = rt, and d = su.

Proof. Let (a, c) = g1 . By proposition ( ), there exist integers x1 and y1 such that
a = g1 x1 and c = g1 y1 with x1 ⊥ y1 . Also, let (b, d) = g2 , then there exist integers x2
and y2 such that b = g2 x2 and d = g2 y2 with x2 ⊥ y2 . Substitute these equations into
the given equation ab = cd to get

g 1 g 2 x1 x2 = g 1 g 2 y1 y2 ,

or simply x1 x2 = y1 y2 . We claim that x1 = y2 and x2 = y1 . To prove this claim, we

use a trick. Observe that x1 x2 = y1 y2 means x1 |y1 y2 . The second part of Proposition
( ) tells us that since x1 ⊥ y1 , we must have x1 |y2 . With the same reasoning, one
can show that y2 |x1 . It follows by proposition ( ) that x1 = y2 . One can similarly
show that x2 = y1 . Compare the parameters in the theorem with the ones we just used
and take r = g1 , s = x1 , t = x2 , and u = g2 . The proof is complete.

Theorem 1.4.8. For positive integers a, b, n, and e, suppose that a ⊥ b and ab = ne .

Then there exist positive integers x and y such that a = xe and b = y e . In other words,
if product of two coprime positive integers is a perfect eth power, then both of them
should be perfect eth powers.

Proof. Let us consider the prime factorization of a and b as the following.

a = pe11 . . . pekk and b = q1f1 . . . qℓfℓ .

Then,

ab = pe11 . . . pekk q1f1 . . . qℓfℓ .

From uniqueness of prime factorization, the only prime factors of n must be p1 , . . . , pk

and q1 , . . . , qℓ . Let n = p1a1 . . . pakk q1b1 . . . qℓbℓ . Then

ne = pa11 e . . . pakk e q1b1 e . . . qℓbℓ e .

48 1.4. SOME USEFUL FACTS

Since ab = ne , we get

pe11 . . . pekk q1f1 . . . qℓfℓ = pa11 e . . . pkak e q1b1 e . . . qℓbℓ e .

The exponents of the primes in both sides must be equal. Therefore, ei = ai e (for
i = 1, 2, . . . , k) and fj = bj e (for j = 1, 2, . . . , ℓ). Therefore,
 e  e
a = pa11 . . . pakk and b = q1b1 . . . qℓbℓ ,

which proves the claim.

Note. You should think about why we considered the prime factorization and try to
understand what led us into that way of thinking.

Corollary 1.4.9. If a and b are relatively prime positive integers such that ab is a
perfect square, then a and b both are perfect squares.

Theorem 1.4.10. Let a, b, and c be positive integers such that ab = c2 . Then there
exist integers g, u, and v with u ⊥ v so that

a = gu2 , b = gv 2 , and c = guv.

Proof. Let g = (a, b). Then there exist coprime integers x and y such that a = gx and
b = gy. Then g 2 xy = c2 so that g 2 |c2 , or g|c (see the example provided right after
proposition ( ) for a proof). This implies that there exists some positive integer k
for which c = gk. Substituting this into the equation, we find k 2 = xy with x ⊥ y.
From corollary ( ), there exist integers u and v such that x = u2 and y = v 2 . Thus,
a = gu , b = gv , and c = guv.
2 2

Now, we introduce a simple but really useful method for factorization: Simon’s
Favorite Factorization Trick , or SFFT in brief.

Proposition 1.4.11 (SFFT). For any real numbers x, y, j, and k, the following rela-
tion holds:

xy + xk + yj + jk = (x + j)(y + k).

Two special common cases are

xy + x + y + 1 = (x + 1)(y + 1),

and
xy − x − y + 1 = (x − 1)(y − 1).

Let’s see the motivation behind this trick. Once Mr. Simon was studying number
theory, he found this problem: find all positive integers x and y such that xy − x + y =
49. Simon probably hates expressions of this form because he cannot factorize them.
However, if he adds −1 to both sides of his equation, he finds the nice and factored
12
Named after Simon Rubinstein-Salzedo, a member of AoPS.
CHAPTER 1. DIVISIBILITY 49

form (x + 1)(y − 1) = 48, which is much easier to solve than the original equation.
In fact, to solve the factorized equation, he only needs to find the divisors of 48 (see
theorem ( ) for more details). If you look closely, SFFT is inspired by the so-called
Completing the Square Method:
 2
2 k2 k
x + kx + = x+ .
4 2
In fact, the act of adding jk to xy + xk + yj in order to be able to factor it could
be called completing the rectangle in analogy to the famous completing the square
trick.
Theorem 1.4.12. If N is the least common multiple of positive integers upto n, that
is, N = [1, 2, . . . , n], then for a prime p, the maximum integer α for which pα |N is
the unique non-negative integer α so that pα ≤ n < pα+1 . In other words, if pi is a
prime less than or equal to n, and for that prime, αi is the unique integer such that
pαi i ≤ n < pαi i +1 then,

N = pα1 1 . . . pαk i .

We can write αi in terms of n and pi using logarithm. Actually, αi is the greatest

non-negative integer less than or equal to logpi (n).
The proof in fact follows from the definition of least common multiple. Try some
examples yourself and prove it. Also, we often use this kind of argument using logarithm
to find some boundaries in some problems.
Theorem 1.4.13. The square of every odd integer leaves a remainder of 1 when divided
by 8.
Proof. We can write each odd integer n in the form n = 2k − 1 for some k ∈ Z. Then

n2 = (2k − 1)2 = 4k 2 − 4k + 1 = 4k(k − 1) + 1.

Since one of k or k − 1 must be even (why?), we can write k(k − 1) = 2ℓ for some
integer ℓ. Then n2 = 8ℓ + 1, and so n2 leaves a remainder of 1 when divided by 8.
Theorem 1.4.14. Every number of the form 4k + 3 has at least one prime factor of
the form 4k + 3.
Proof. The idea comes from the fact that if we multiply two numbers of the form 4k +1,
say 4a + 1 and 4b + 1, the result will be

(4a + 1)(4b + 1) = 16ab + 4a + 4b + 1 = 4(4ab + a + b) + 1,

which is, again, of the form 4k + 1. Clearly, all prime factors of a number n of the
form 4k + 3 are odd, and therefore are either of the form 4k + 1 or 4k + 3. If all prime
factors of n are of the form 4k + 1, then by the logic represented in above lines, any
product of powers of these primes, including n, should be of the form 4k + 1. So, we
get a contradiction and there exists at least one prime factor of n which is of the form
4k + 3.
50 1.4. SOME USEFUL FACTS

Theorem 1.4.15. Let n be a positive integer. The number of pairs (a, b) of positive
integers which satisfy the equation

ab = n

is τ (n), where τ (n) is the number of positive divisors of n.

The following example gives you an idea for the proof.
Example. Take n = 6. The only positive divisors of n are 1, 2, 3, and 6. There-
fore, τ (6) = 4. On the other hand, the equation ab = 6 has exactly four solutions
(1, 6), (2, 3), (3, 2), and (6, 1).
We will study τ (n) in details in Chapter .
Theorem 1.4.16. Every prime greater than 3 is either of the form 6k + 1 or of the
form 6k − 1.
Proof. Consider an integer n. By definition ( ), the minimum absolute remainder r
of n upon division by 6 has the property that |r| ≤ 6/2 = 3. In other words, r can have
the values 1, 2, 3, −1, −2, and −3. For this reason, we can write an integer in exactly
one of the following forms:

6k − 2, 6k − 1, 6k, 6k + 1, 6k + 2, 6k + 3.

Numbers of the form 6k − 2, 6k + 2, 6k, 6k + 3 cannot be prime because the first two
are divisible by 2 and the last two are divisible by 3. Thus, if n is a prime, it must be
either 6k − 1 or 6k + 1.
Using the above theorem, we can prove the following.
Theorem 1.4.17. For a prime p > 3, 24|p2 − 1.
Proof. We can assume p = 6k ± 1 for some integer k. So

p2 = (6k ± 1)2 = 36k 2 ± 12k + 1 = 12k(3k ± 1) + 1.

Note that (k)+(3k ±1) is odd. From corollary ( ), k and 3k ±1 have different parity
and k(3k ± 1) is divisible by 2. Let k(3k ± 1) = 2ℓ, then p2 = 24ℓ + 1, or p2 − 1 = 24ℓ,
which proves the theorem.
Theorem 1.4.18. If the sum of two positive integers is a prime, they are coprime to
each other.
Proof. Assume that a + b = p, where p is a prime. If (a, b) = g then there exist coprime
positive integers x and y for which a = gx and b = gy. So,

p = a + b = g(x + y),

which means g divides p. Since p is a prime, its only divisors are itself and 1. Hence, the
only possible values for g are 1 and p. However, g = p cannot happen since otherwise
x + y = 1 would lead to one of x or y being zero. Thus, (a, b) = g = 1.
CHAPTER 1. DIVISIBILITY 51

Theorem 1.4.19. For any positive integer n,

 
an − bn = (a − b) an−1 + an−2 b + · · · + bn−1 .

Furthermore, if n is odd, then

 
n n n−1 n−2 n−1
a + b = (a + b) a −a b + ··· + b .

Proof. We can just use induction on n, but we will try to avoid using induction as much
as possible. For the first identity, define

S = an−1 + an−2 b + · · · + bn−1 .

Then, aS = an + an−1 b + · · · + bn−1 a and bS = an−1 b + an−2 b2 + · · · + bn . Subtracting,

we obtain

(a − b)S = an − bn ,

which finishes the proof. The second identity can be proved by using the first identity
and the fact that when n is odd, we have an + bn = an − (−b)n .

Corollary 1.4.20. Let a and b be any two integers. Then a − b divides an − bn for all
positive integers n. Also, a + b divides an + bn for odd n.

Theorem 1.4.21. Let a, m, and n be positive integers. Then, am − 1|an − 1 if and only
if m|n.

Proof. We will first show that if m|n, then am − 1|an − 1. Notice that if m|n, that there
exists a positive integer h such that n = mh, and so,

an − 1 = amh − 1 = (am )h − 1.

Let x = am . By corollary ( ), x − 1 = am − 1 divides xh − 1 = an − 1, as desired.

Let us now show the other side. If am − 1|an − 1, then,

am − 1 | an − 1 − (am − 1) = an − am = am (an−m − 1).

Since an − 1 ⊥ am (why?), we have an − 1 | an−m − 1 by second part of proposition

( ). Repeating the same process, we find an − 1 | an−2m − 1. This suggests us to
take n = mq + r so that 0 ≤ r < m. Then, we will have an − 1 | an−mq − 1 = ar − 1.
Evidently, ar − 1 < am − 1 ≤ an − 1, which forces ar − 1 = 0. So, r = 0 and n = mq.
The proof is complete.

Theorem 1.4.22. If ak − 1 is a prime for positive integers a and k > 1, then a = 2

and k must be a prime.
52 1.4. SOME USEFUL FACTS

Proof. As we already know,

ak − 1 = (a − 1)(ak−1 + ak−2 + · · · + a + 1).

If a > 2, then a − 1 will divide ak − 1 which is absurd because ak − 1 is a prime. So,

a must be 2. Suppose that k is composite. Then, k = pℓ for some prime p and integer
ℓ > 1. Therefore,

ak − 1 = (ap )ℓ − 1


= (ap − 1) ap(ℓ−1) + ap(ℓ−2) + · · · + 1 ,

which is impossible. Hence, k must be a prime.

Remark. The numbers of the form 2n − 1 are called Mersenne numbers and denoted
by Mn . According to the theorem, if Mn is a prime, then n must also be a prime.
Theorem 1.4.23. Let a and k be positive integers. If ak + 1 is an odd prime, then a
is even and k is a power of 2.
See Problem for a proof.
n
Remark. For the particular case when a = 2, the numbers 22 + 1 are called Fermat
numbers and denoted by Fn . Notice that the reverse of the above theorem does not
hold. That is, not all Fermat numbers are primes. For instance,
5
F5 = 22 + 1 = 4 294 967 297 = 641 × 6 700 417.

Theorem 1.4.24. For two relatively prime positive integers a and b,

(am − bm , an − bn ) = a(m,n) − b(m,n) .

Proof. Here, we will use Euclidean algorithm as the idea is applicable to many similar
problems. If m = n, the result is trivial since both sides are am − bm . So, we can take
m > n without loss of generality . Then we can write m = n + k for some k ∈ N.
Therefore,

am − bm = an+k − bn+k
= an (ak − bk ) + bk (an − bn )

Thus, by Euclidean algorithm,

(am − bm , an − bn ) = (an (ak − bk ) + bk (an − bn ), an − bn )

= (an (ak − bk ), an − bn ).

Since a and b are coprime, an ⊥an − bn (why?). This gives us

(am − bm , an − bn ) = (an − bn , ak − bk ).

Note that, this is the descending step of Euclidean algorithm! If we repeat the same
process a couple of times, we would eventually reach (m, n) in the exponent.
13
Wherever there is symmetry, don’t forget to use this trick!
CHAPTER 1. DIVISIBILITY 53

As a conclusion to this chapter, we define factorial and binomial coefficient.

Definition 1.4.25 (Factorial). Let n be a positive integer. The factorial of n, denoted

by n!, is the product of all positive integers less than or equal to n. That is,

n! = 1 × 2 × · · · × n.

For convenience, we define 0! to be one .

Definition 1.4.26 (Binomial Coefficient). Let n and k be non-negative integers  

n
such that k ≤ n. The Binomial Coefficient indexed by n and k is denoted by
k
(read “n choose k”) and is equal to
 
n n!
= .
k k!(n − k)!


Note. It can be proved combinatorially that there are nk ways to choose k elements,
disregarding their order, from a set of n elements. Although we may use this com-
binatorial fact in some problems, we only deal with the number theoric definition of
binomial coefficients in this book.

Proposition 1.4.27. The binomial coefficient is an integer.

Hint. Use Problem in chapter .

You are probably familiar with the binomial theorem. We state this theorem here
because of its high importance but do not prove it. Interested reader may see a proof
of binomial theorem along with several useful related identities in Appendix .

Theorem 1.4.28 (Binomial Theorem). For any positive integer n,

     
n n n n−1 n n−2 2 n
(a + b) = a + a b+ a b + ··· + abn−1 + bn .
1 2 1

Theorem
1.4.29. Let n be a positive integer. Then, n is a prime if and only if n

divides nk for all 0 < k < n.

For now, we will be proving only the first part.

Proof. From identity number of theorem ( ) in Appendix , we know that

   
p p−1
k =p .
k k−1

Obviously, the right side is divisible by p, so must be the left side. From the definition
of prime numbers,
p⊥k. So, by the second part of the previous proposition, we can say
p must divide kp .
14
We can argue combinatorially as well, but let us put that aside for now.
54 1.5. SOLVED PROBLEMS

1.5 Solved Problems

Well, now that we have a basic understanding of how divisibility relations work, it is
the time to see how to apply those propositions to solve problems. Let us start with
some really easy problems so that the reader has a firm idea about how to approach a
problem. We will gradually discuss more difficult problems. Most importantly, we will
not focus on how to use theorems to solve problems or implement them. We want to
improve the intuitive ability of the reader instead.

Problem 1.5.1. Find all positive integers n so that n divides n + 3.

This can be solved in a variety of ways. Before we solve this, it would be pleasant
to share some experience. This may not directly help readers understand how to solve
the problem, but it may help them understand how different people think when they
encounter a problem.
Masum was coaching a group of average students who barely had the idea of problem
solving back in 2014. After showing them some basic facts about divisibility, he threw
this problem at them. Here is how they approached it: most of them started trial and
error to find the solution. They were checking different values for n to see if they satisfy
the condition. Not all student could realize that this method will not work in general
for all numbers even if you find some initial solutions. One or two of the students went
for the idea that if n divides n + 3, then n + 3 is at least twice n. However, when
increasing n, 2n becomes larger than n + 3 after some n, and this leads to a solution.
Another student solved the problem in the old fashioned way: by division process. That
is, he divided n + 3 by n and found out that n must divide 3. Other than those few
students, most of them got that n = 1 and n = 3 work but could not prove that these
are the only solutions. In fact, that is the most difficult part of solving a number theory
problem.
When you encounter a problem in mathematics, you probably come up with an
assumption. For instance, in this problem, you may assume that the only solutions
are 1 and 3. Most of the times our observations are true, but in many cases there are
counter examples or some cases where our assumption fails. In our problem, if we just
say the only solutions to the problem are 1 and 3 without proving it, we must tolerate
comments such as “How do you know there is not any large n which is also a solution?”
That is why we have to prove everything we claim is true. Of course, if our claim is
implied from a well-known theorem in mathematics, we can skip the proof and refer to
the theorem. However, if the claim is totally new or relies on theorems which are not
well known, we have to provide a proof. Obviously, the reader of your proof defines the
famousness of the theorems you are allowed to use. There must be a huge difference in
the way you solve a problem when you are explaining it to a math teacher or to a sixth
grade student. By the way, when writing a solution to a problem at an olympiad test
(or any other sophisticated exam), be careful not to consider every theorem obvious or
well known. Whatever the problem is, try to prove every claim that you make.
A suggestion: while reading a solution, try to find the motivation behind its idea
rather than understanding it or learning the technique used in the solution.
CHAPTER 1. DIVISIBILITY 55
n+3 3
Solution (Old-fashioned Way). Since n|n + 3, the fraction = 1 + is a positive
n n
3
integer. This shows that must be an integer. Therefore, n divides 3 and the only
n
possible values for n are 1 and 3.

Solution (A Smarter Way). We know that if a|b and a|c, then a|b − c. But how can
we know what to subtract? If m is a given constant, we can easily find all numbers n
such that n|m just as in the previous solution. See that the right side of n|n + 3 has
a variable n which is not constant. So, we need to remove it somehow. This is where
divisibility rules come into play. We can now subtract n|n from n|n + 3 from to obtain
n|(n + 3) − n = 3 or n|3. In words, n is a divisor of 3 which has namely two divisors 1
and 3. Thus, n can be 1 or 3.

Solution (Another Smart Solution). For positive integers a and b, if a|b, b ≥ a. Now,
since n + 3 cannot be equal to n, if it has to be divisible by n, it must be at least twice
n, i.e., n + 3 ≥ 2n. This forces 3 ≥ n and so n is one of 1, 2, or 3. There are only three
values so we can check them by hand and get the answer.

Note the difference between the first and the second solution. There are pretty
much no differences other than the fact that the latter is more systematic. You will
eventually find that solving problems this way is particularly useful in Olympiads.
For now, we will keep the use of inequality in store and solve a few problems by
twisting up left and right sides of divisibility relations.

Problem 1.5.2. Find all n ∈ N so that n|2n + 3.

Solution. This time we have to remove n like before, but notice that there is an extra
2 attached.
We can overcome this easily: just see that n|2n and then it is the same as the
previous problem: n|2n + 3 − 2n = 3.

Problem 1.5.3. Find all n ∈ N so that n + 1 | 3n + 4.

Solution. The left side contains more than just a n, but it does not make any difference.
We have n + 1 | 3(n + 1) = 3n + 3 and it is given in the problem that n + 1 | 3n + 4.
Subtracting,

n + 1|3n + 4 − (3n + 3) = 1,

which means n + 1 must be 1. This is impossible because n + 1 is a positive integer

greater than 1.

Note. When dealing with a divisibility relation k|m in which m is constant, there are
τ (m) possible values for k, where τ (m) is the number of positive divisors of m (why?).

Try to do the following problem yourself before reading the solution.

Problem 1.5.4. Find all n ∈ N which satisfies 4n + 2|6n + 5.

56 1.5. SOLVED PROBLEMS

Solution. There are coefficients on both sides of divisibility now. Still it can be
overridden but it is a bit tricky. See that 4n + 2|3(4n + 2) = 12n + 6 and also
4n + 2 | 2(6n + 5) = 12n + 10. Now that the coefficients are matched, we can subtract
and find

4n + 2|12n + 10 − (12n + 6) = 4.

So, 4n + 2 must be one of 1, 2, or 4. None of these values give a valid solution for n.
If you have not noticed already, this idea can be generalized to find all n satisfying
an + b|cn + d, where a, b, c, and d are integers. Then what shall we multiply both sides
with? Here is a hint: consider the lcm of a and c. Why a and c?
Problem 1.5.5. Find all positive integers n for which 8n + 9|12n + 5.
Solution. Our working principle is: we want to eliminate the variables on the right
side so we get a constant value. That is, we want the right side to be a number, not
a variable. In order to do this, we must construct two divisibility relations to subtract
from each other and get a constant value. In other words, we have to find a and b such
that the difference of right sides of 8n + 9 | a(8n + 9) and 8n + 9 | b(12n + 5) does not
include n. So, the coefficients of n must be equal in the two. The minimum value for
this common coefficient will be the lcm of 8 and 12, which is [8, 12] = 24. Therefore,
a = 3 and b = 2 and hence,

8n + 9 | 3(8n + 9) = 24n + 27,

8n + 9 | 2(12n + 5) = 24n + 10.

Thus, 8n + 9 | 24n + 27 − (24n + 10) = 17 and 8n + 9 is either 1 or 17. The equation

8n+9 = 1 does give a valid solution. So, the only solution is obtained from 8n+9 = 17,
which gives n = 1.
Problem 1.5.6 (Dhaka Divisional Olympiad, 2010). Find all positive integers n
greater for which divides n + 4 and n + 12 for some positive integer n.
Solution. We are asked to find positive integers which would divide n + 4 and n + 12
both, no matter what. If d is such a positive integer then

d|n + 4, d|n + 12
⇐⇒ d|(n + 12) − (n + 4) = 8

We immediately get that d ∈ {1, 2, 4, 8}.

Problem 1.5.7. Find all primes p so that 9p + 1 is also a prime.
Solution. This can be done easily considering parity. When you encounter a problem
with primes (especially if you are asked to find a prime), it may be helpful to separate
the problem into two cases. The first case is p = 2 which works here. The second case
is when p is odd. It is trivial that as p is odd, 9p + 1 is even. However, we want 9p + 1
be a prime. This is a contradiction because 9p + 1 is larger than 2. Therefore, the only
solution is p = 2 for which 9p + 1 = 19 is also a prime.
CHAPTER 1. DIVISIBILITY 57

Problem 1.5.8. Find all positive integers n so that 7n + 1 | 8n + 55.

Solution. Since [8, 7] = 56, the two required relations would be

7n + 1 | 8(7n + 1) = 56n + 8,
7n + 1 | 7(8n + 55) = 56n + 385.

By subtracting, we find 7n+1 | 377. Now, we have to find divisors of 377. When you are
stuck with finding divisors, instead of trying random numbers or testing the numbers
2, 3, 4, 5, . . . serially to find if they divide 377, try a more efficient approach which reduces
your effort significantly. By proposition √ ( √), if n is composite, then it must have
a prime factor less than or equal to n. Since 377 ≈ 19.4, it suffices to check which
primes less than or equal to 19 divide 377. These primes are 2, 3, 5, 7, 11, 13, 17, 19.

• 2 ∤ 377 since the rightmost digit is odd,

• 3 ∤ 377 since the sum of digits is 17,

• 5 ∤ 377 because the number ends in 7,

• 7 ∤ 377 because 377 = 7 · 53 + 6,

• 11 ∤ 377 as the difference of sum of altering digits is (3 + 7) − 7 = 3,

• 13 | 377 because 37 + 4 · 7 = 65 = 13 · 5.

So, we find the factorization of 377 to be 377 = 13 · 29. This means that 7n + 1 is a
divisor of 13 · 29. Notice that 7n + 1 is a number which leaves remainder 1 when divided
by 7. Therefore, look for numbers among 13, 29, 13, and 377 which leave a remainder
of 1 when divided by 7. The only possibility is 7n + 1 = 29, which gives n = 4.

Problem 1.5.9. Find all n ∈ N for which n + 3 divides n2 + 2.

We should again remove the variables from the rights side of n + 3 | n2 + 2. Here
are two solutions for the problem.

Solution (First Solution). First, multiply n + 3 by n to get

n + 3 | n2 + 3n.

Now, subtract this from the given relation to find

n + 3 | 3n − 2.

The problem is now similar to the ones we solved earlier. In fact, subtracting

n + 3 | 3n + 9 and n + 3 | 3n − 2,

we obtain n + 3|11 which immediately gives n = 8 as the only solution.

58 1.5. SOLVED PROBLEMS

Solution (Second Solution). This one is more elegant. In order to avoid the second
step, we can directly multiply n + 3 by n − 3 to obtain
n + 3|n2 − 9.
Subtracting from the original relation, n + 3 | n2 + 2 − (n2 − 9) or n + 3|11, and we find
the same solution as before.
Problem 1.5.10. Find the greatest positive integer x for which
x + 10|x3 + 10.
Solution. Since 3 is odd, we can use corollary ( ) to write x + 10 | x3 + 103 . So,
x + 10|(x3 + 1000) − (x3 + 10) = 990,
and the answer is x = 980.
Problem 1.5.11. Find all positive integers n so that n6 + n4 is divisible by 7n + 1.
Solution. How do we approach this problem? One idea is to go ahead like we did in
the first solution of problem ( ), by eliminating powers of n one by one. Here is a
better solution.
First write it as, 7n + 1 | n4 (n2 + 1). The problematic part is n4 . Many beginners
make mistakes in such situations claiming that 7n + 1 does not divide n4 , and so
7n + 1 | n2 + 1. This is wrong. In general, the following statement is wrong:
For positive integers a, b, and c, if a|bc and b is not divisible by a, then a|c.
This is another common mistake new problem solvers make. You can check this using
an example: 14 divides 28 = 7 × 4 but 14 does not divide 7. But in this problem, we
can still take n4 off because n4 is coprime to 7n + 1. Just notice that 7n + 1 leaves a
remainder of 1 when divided by n, so n ⊥ 7n + 1. Evidently, n4 ⊥ 7n + 1 by part 4 of
proposition ( ). Using part 2 of the same proposition, we can cancel out n4 and get
7n + 1 | n2 + 1.
Multiply n2 + 1 by 49 to find 7n + 1 | 49n2 + 49. On the other hand, 7n + 1 divides
(7n + 1)(7n − 1) = 49n2 − 1, and so,
7n + 1 | (49n2 + 49) − (49n2 − 1) = 50.
Note that 7n + 1 leaves a remainder of 1 when divided by 7. Therefore, we look for
divisors of 50 which leaves a remainder 1 when divided by 7. The divisors of 50 are
1, 2, 5, 10, 25, and 50. Only 1 and 50 leave the desired property among these numbers.
Since n is a positive integer, 7n + 1 ≥ 7 · 1 + 1 = 8. Therefore, 7n + 1 = 50 or n = 7.
Remark. We could handle the last part in another way. Write 7n+1 | (n2 +1)−(7n+1),
so that
7n + 1 | n2 − 7n =⇒ 7n + 1 | n(n − 7).
Again, n ⊥ 7n + 1 implies 7n + 1|n − 7. If n = 7, we have a solution. Otherwise, 7n + 1
would have a larger value than |n − 7| and we get a contradiction.
CHAPTER 1. DIVISIBILITY 59

Problem 1.5.12. If ax + by = 1, find (a, b).

Solution. Assume that (a, b) = g. Then we can find two coprime integers m and
n so that a = gm and b = gn. Setting these into the given equation, we observe
g(mx + ny) = 1. This implies that g divides 1, and so g = 1.
Note. We can similarly show that (x, y) = (x, b) = (a, y) = 1.
Problem 1.5.13. Find the number of solutions to the equation
1 1 1
+ =
x y 2015
in positive integers.
Solution. We can rewrite the equation as
x+y 1
= .
xy 2015
Multiplying both sides by 2015xy, we find xy = 2015(x + y). This can be represented
as

(x − 2015)(y − 2015) = 20152 .

According to theorem ( ), we get that the number of solutions is the number of

positive divisors of 2015 , τ (20152 ).
2

Note. Did you realize that this was actually SFFT? And what do you think the result
would be if we considered solutions in integers (not necessarily positive)?
Problem 1.5.14. Find all n ∈ N such that 2n + n|8n + n.
Solution. There is 8 on the right side and 2 on the left side. Since 8 = 23 , this should
certainly provoke us to use the fact that a + b|a3 + b3 . In this case,

2n + n | (2n )3 + n3
=⇒ 2n + n | 8n + n3
=⇒ 2n + n | (8n + n3 ) − (8n + n)
=⇒ 2n + n | n3 − n.

Now we need to find all n such that 2n + n | n3 − n. If you play around with the small
values of n, you will clearly see that 2n grows a lot faster than n3 . Since 2n + n|n3 − n,
we must have

2n + n ≤ n3 − n.

The following lemma gives us the limit we need for n. Once again, you have to prove
it. You can use induction to do so (can you use logarithm?).
Lemma 1.5.15. For n > 9, the inequality 2n + n > n3 − n holds.
60 1.5. SOLVED PROBLEMS

By the lemma, if n > 9, there are no solutions. Now we are left with a few possi-
bilities for n because n must be less than or equal to 9. We can easily check by hand
that there are no solutions for this case as well.

Problem 1.5.16 (IMO 1959, Problem 1). Prove that for any integer n, the fraction
14n + 3
21n + 4
is irreducible.

Solution. We must make sense of the problem first. It asks to prove that a fraction is
irreducible. That means the fraction cannot be simplified anymore. In other words, we
must prove that the numerator (14n + 3) and denominator (21n + 4) of the fraction are
coprime to each other. How do we prove this? We will show this in two ways:

1. Let g = (14n + 3, 21n + 4). Then,

g | 14n + 3,
g | 21n + 4.

We already know we have to prove that g = 1. So, let us try to remove the n on
the right side. Since [14, 21] = 42 and 42 = 14 · 3 = 21 · 2, we can make use of the
following two relations:

g | 3(14n + 3) = 42n + 9,
g | 2(21n + 4) = 42n + 8.

Thus, g|(42n + 9) − (42n + 8) = 1 and g = 1.

2. Take 14n + 3 = gx and 21n + 4 = gy for some integers x and y. Note that
3(14n + 3) − 2(21n + 4) = 1, and so

3gx − 2gy = 1,

or g(3x − 2y) = 1. So, g|1 and this gives us the same result g = 1.

Remark. You should understand that both solutions are essentially the same, but with
different approaches or thinking styles.

Problem 1.5.17. Show that if a prime is of the form 2n + 1, then n = 2m for some
integer m.

Solution. According to theorem ( ), write n = 2m s, where s is an odd positive

integer. By theorem ( ), we can write
m m
s
p = 2n + 1 = 22 s + 1 = 2 2 +1
m
m m m

= 22 + 1 22 (s−1) + 22 (s−2) + · · · + 22 + 1 .

Clearly, as p is a prime, this is impossible unless s = 1. So n = 2m .

CHAPTER 1. DIVISIBILITY 61

Problem 1.5.18. Find all integer solutions to the equation

a(b + 1) + b(c + 1) + c(a + 1) + abc = 0.
Solution. Yes, this looks similar to SFFT. It has three variables instead of two, so
we cannot directly use Simon’s trick. However, if you already learned the motivation
behind that trick, the form of the problem does not matter. Here, 1 is missing! Add it
to both sides to get
abc + ab + bc + ca + a + b + c + 1 = 1.
The left hand side shows you how SFFT looks like for three variables. Try to compute
(a + j)(b + k)(c + ℓ) and remember it. You will soon realize that the left hand side of
the above equation factors as (a + 1)(b + 1)(c + 1). Therefore,
(a + 1)(b + 1)(c + 1) = 1.
The rest of the solution is just case work. If the product of three integers equals 1,
what can they be? The solutions are
(a, b, c) = (0, 0, 0), (−2, −2, 0), (−2, 0, −2), (0, −2, −2).
Problem 1.5.19 (Turkey TST 2014, Day 2, Problem 4). Find all odd positive
integers m and n such that
n | 3m + 1 and m | n2 + 3.
Solution. Do not panic! Even some TST problems can be solved using simple tricks.
The general idea is to first see if there are infinitely many solutions. Check some
numbers and some special values for m and n. For instance, put m = n to see if it fits
in. In this problem, we cannot find a pattern to construct infinitely many solutions,
so we guess that there are finite solutions and continue. To be honest, many of such
divisibility problems (with finite solutions) rely heavily on case working, and you need
to know how to start the case work. One of the best ways to handle this situation is
to find a limit for m and n. Remember the properties of divisibility from proposition
( ). Limit means boundary, and boundary means inequality (see part 10 of that
proposition). Actually, we will use the fact that if a|b for positive integers a and b, then
a ≤ b. After we have found the limit, the case work begins.
Write n|3m + 1 as 3m + 1 = nk or m = nk−1 3
. Now try to rewrite the second
divisibility relation as

nk − 1  2
n +3
3 
×3
=⇒ nk − 1 | 3n2 + 9
×k
=⇒ nk − 1 | 3n2 k + 9k.
On the other hand, we know that nk − 1|3n(nk − 1) = 3n2 k − 3n. Subtract these two
last divisibility relations to find
nk − 1 | 3n + 9k,
62 1.5. SOLVED PROBLEMS

which in turn implies nk − 1 ≤ 3n + 9k. This inequality can be simplified using SFFT
in this way:
add 27
nk − 3n − 9k ≤ 1 =⇒ (n − 9)(k − 3) ≤ 28.

Notice that we have found the limit we wanted! The solution is almost obvious from
here on, because one can manually put all possible values for n and k such that (n −
9)(k − 3) ≤ 28 and pick those which satisfy the problem conditions (n|3m + 1 and
m|n2 + 3). In order to avoid this tedious endeavor, we start a proper case work.
We start case work on the parameter k (you will know why). Since m is odd and
3m + 1 = nk, k is even. We check some possible values of k:

(a) If k = 2, then from nk − 1 | 3n + 9k we have 2n − 1 | 3n + 18, and thus

2n − 1|2(3n + 18) − 3(2n − 1) = 39.

This means that 2n − 1 = 1, 3, 13, or 39. None of these values make a solution for
the problem.

(b) If k = 4, then from nk − 1|3n + 9k we have 4n − 1|3n + 36, and thus

4n − 1|4(3n + 36) − 3(4n − 1) = 147.

This means that 4n − 1 = 1, 3, 7, 21, 49, or 147. Checking the values of n, we

find that n = 1 and n = 37 satisfy the conditions of the problem and give us the
solutions (m, n) = (1, 1) and (m, n) = (49, 37).

(c) If k ≥ 6, then the inequality (n − 9)(k − 3) ≤ 28 can be transformed to

28 28
(n − 9) ≤ ≤ ,
k−3 3
which means n ≤ 9 + 28/3. Since n is a positive integer, the latter inequality
simplifies to n ≤ 18 (why?). Also, n is odd, so we have to check only the numbers
n = 1, 3, 5, 7, 9, 11, 13, 15, and 17. In this case, only n = 13 gives a valid solution
(check it) and it is (m, n) = (43, 13).

Finally, the solutions are

(m, n) = (1, 1), (49, 37), (43, 13).

CHAPTER 1. DIVISIBILITY 63

1.6 Exercises
Problem 1.6.1. Find all pairs of positive integers (x, y) for which

x2 + y 2
x−y
is an integer that divides 1995.
Problem 1.6.2. Let x, y, and z be integers such that

(x − y)2 + (y − z)2 + (z − x)2 = xyz.

Prove that (x + y + z + 6) divides (x3 + y 3 + z 3 ).

Problem 1.6.3. Let k and m, with k > m, be positive integers such that the number
km(k 2 − m2 ) is divisible by k 3 − m3 . Prove that (k − m)3 > 3km.
Problem 1.6.4. Let F (n) = 136n+1 + 306n+1 + 1006n+1 + 2006n+1 and define

G(n) = 2F (n) + 2n(n − 2)F (1) − n(n − 1)F (2).

Prove by induction that for all integers n ≥ 0, G(n) is divisible by 73 .

Problem 1.6.5. Prove that for every integer n ≥ 2, there exist n different positive
integers such that for any two of these integers a and b, a + b is divisible by a − b.
Problem 1.6.6 (Iran Second Round 1994). Let a1 a2 a3 . . . an be the representa-
tion of a n−digits number in base 10. Prove that there exists a one-to-one function
like f : {0, 1, 2, 3, . . . , 9} → {0, 1, 2, 3, . . . , 9} such that f (a1 ) 6= 0 and the number
f (a1 )f (a2 )f (a3 ) . . . f (an ) is divisible by 3.
Problem 1.6.7 (Slovenia 2010). Let a, b, and c be three positive integers. Prove
that a2 + b2 + c2 is divisible by 4 if and only if a, b, and c are all even.
Problem 1.6.8 (ILL 1990, THA1). Let m be an odd positive integer not divisible
by 3. Prove that
 m √ 
4 − (2 + 2)m

is divisible by 112.
Note: for any real number x, ⌊x⌋ is the largest integer not exceeding x.
√ √
Hint. Show that each term is equal to 4m − (2 + 2)m − (2 − 2)m .
Problem 1.6.9. Let Pn = (19 + 92)(192 + 922 ) · · · (19n + 92n ) for each positive integer
n. Determine, with proof, the least positive integer m, if it exists, for which Pm is
divisible by 3333 .
Problem 1.6.10. Prove that we can choose 2n numbers from 2n+1 positive integers
such that their sum is divisible by 2n .
64 1.6. EXERCISES

Problem 1.6.11 (Baltic Way 1993). Prove that for any odd positive integer n,
n12 − n8 − n4 + 1 is divisible by 29 .

Problem 1.6.12. Let m and n be two positive integers. Does there exist positive
integers a, b, and c all greater than m such that abc is divisible by a + n, b + n, and
c + n?

Problem 1.6.13 (Baltic Way 1997). Prove that in every sequence of 79 consecutive
positive integers written in the decimal system, there is a positive integer whose sum
of digits is divisible by 13.

Problem 1.6.14 (Baltic Way 2006). A 12−digit positive integer consisting only of
digits 1, 5, and 9 is divisible by 37. Prove that the sum of its digits is not equal to 76.
This PDF was created
for Amazon.

Problem 1.6.15. Prove that the number 191976 + 761976

4
1. is divisible by the (Fermat) prime number F4 = 22 + 1, and

2. is divisible by at least four distinct primes other than F4 .

Problem 1.6.16. Show that 5n − 4n + 15 is divisible by 16 for all positive integers n.

Problem 1.6.17. Prove, without using mathematical induction, that 52n+2 − 24n − 25
is divisible by 576.

Problem 1.6.18 (IMO 1979, Problem 1). If p and q are positive integers so that

p 1 1 1 1 1
= 1 − + − + ··· − + ,
q 2 3 4 1318 1319

prove that p is divisible with 1979.

Problem 1.6.19 (ILL 1970–1973, FIN1). Prove that 2147 − 1 is divisible by 343.

Problem 1.6.20 (ISL 1998, NT7). Prove that for each positive integer n, there
exists a positive integer with the following properties:

• It has exactly n digits.

• None of the digits is zero.

• It is divisible by sum of its digits.

Problem 1.6.21 (ISL 1998, NT1). Determine all pairs (x, y) of positive integers
such that x2 y + x + y is divisible by xy 2 + y + 7.

Problem 1.6.22 (Ukrain TST‌2008, P6). Prove that there exist infinitely many
pairs (a, b) of natural numbers not equal to 1 such that bb + a is divisible by aa + b.

Problem 1.6.23 (Japan Final Olympiads 2009, P1). Find all positive integers n
such that 8n + n is divisible by 2n + n.
CHAPTER 1. DIVISIBILITY 65

Problem 1.6.24 (Romania JBMO TST‌2001). Determine all positive integers

a, b, c, and d in the order a < b < c < d with the property that each of them di-
vides the sum of the other three.
Problem 1.6.25 (BrMO 2003, Modified). Determine all triples (x, y, z) of integers
greater than 1 with the property that x divides yz − 1, y divides zx − 1 and z divides
xy − 1.
Problem 1.6.26 (All Russian Olympiads, P10.7). Positive integers x > 1 and y
satisfy the equation 2x2 − 1 = y 15 . Prove that 5 divides x.
Problem 1.6.27. Find all pairs (a, b) of positive integers such that 2a − 1 and 2b + 1
are coprime and a + b divides 4ab + 1.
Problem 1.6.28. Determine all pairs (m, n) of positive integers such that m > n and

1 = gcd(n + 1, m + 1) = gcd(n + 2, m + 2) = · · · = gcd(m, 2m − n).

Problem 1.6.29. Let a1 , a2 , . . . , an , . . . be any permutation of all positive integers.

Prove that there exist infinitely many positive integers i such that gcd(ai , ai+1 ) ≤ 43 i.
Problem 1.6.30. Let x ≥ 1 be a real number. Prove or disprove that there exists a
positive integer n such that gcd (⌊x⌋, ⌊nx⌋) = 1.

Note. Here ⌊x⌋ denotes the greatest integer less than or equal to x.
Problem 1.6.31. Find all pairs of positive integers (a, b) such that

ab = 160 + 90 gcd(a, b).

Problem 1.6.32. Let a, b, and c be positive integers satisfying gcd(a, b) = gcd(b, c) =

gcd(c, a) = 1. Show that 2abc − ab − bc − ca cannot be represented as bcx + cay + abz
with nonnegative integers x, y, z.
Problem 1.6.33. Let {an }n≥1 be a sequence of positive integers such that if i 6= j,
then

gcd(ai , aj ) = gcd(i, j).

Prove that an = n for all positive integers n.

Problem 1.6.34. Let a and b be two positive integers such that a > b. We know that
gcd(a − b, ab + 1) = 1 and gcd(a + b, ab − 1) = 1. Prove that (a − b)2 + (ab + 1)2 is not
a perfect square.
Problem 1.6.35. Let m and n be positive integers with (m, n) = 1. Find (5m +
7m , 5n + 7n ).
Problem 1.6.36. Define the function f : N → N as follows:
n
X
f (n) = gcd (i, n) .
i=1
66 1.6. EXERCISES

• Prove that the function f is multiplicative.

• Prove that for every positive integer a, the equation f (x) = ax has a solution
x ∈ N.

• Prove that, for a positive integer a, the equation f (x) = ax has exactly one
solution x ∈ N if and only if a is a power of 2 (where 1 = 20 is also considered as
a power of 2).

Problem 1.6.37. Let a0 , a1 , a2 , . . . be a sequence of positive integers such that the

greatest common divisor of any two consecutive terms is greater than the preceding
term; in symbols, gcd(ai , ai+1 ) > ai−1 . Prove that an ≥ 2n for all n ≥ 0.

Problem 1.6.38. Let p be prime, let n be a positive integer, show that

           
p−1 p+1 p p p−1 p+1
gcd , , = gcd , , .
n−1 n n+1 n−1 n n+1

Problem 1.6.39. Prove that

     
n n n
gcd , ,...,
1 2 n−1

is a prime if n is a power of a prime, and 1 otherwise.

Problem 1.6.40. Let a, b, p, and q be positive integers such that

p2 + q 2 pq
=
a b
and gcd(a, b) = 1.

1. Show that pq is divisible by b.

√
2. Show that a + 2b is an integer.

Problem 1.6.41. Prove that if a and b are positive integers such that a + b = (a, b) +
[a, b], then one of a or b divides the other.

Problem 1.6.42 (British Mathematical Olympiad 2005). The integer N is pos-

itive. There are exactly 2005 ordered pairs (x, y) of positive integers satisfying

1 1 1
+ = .
x y N

Prove that N is a perfect square.

Problem 1.6.43. Determine all positive integer numbers m and n such that
1 1 1 2
+ − = .
m n mn 5
CHAPTER 1. DIVISIBILITY 67

Problem 1.6.44. Find all triples (a, b, c) of positive integers such that
1 1 1
+ + = 1.
a b c
Problem 1.6.45. Find all triples (x, y, z) of positive integers which satisfy
1 2 3
+ + = 1.
x y z
Hint. Assume without loss of generality that x is the minimum of the three numbers
and then apply SFFT.
Chapter 2

Modular Arithmetic

Contents
. . . . . . . . . . . . . . . . . . . . 70
. . . . . . . . . . . . . . . . . . . . . 76
. . . . . . . . . . . . . . . . . . . . . . . . . . 79
. . . . . . . . . . . . . . . . . . . . . . . . . . 82
. . . . . . . . . . . . 83
. . . . . . . . . . . 85
. . . . . . . . . . . . . . . . . . 88
. . . . . . . . . . . . . . . . . . . . . . . . . 92
. . . . . . . . . . . . . . . . . . 94
. . . . . . . . . . . . . . . . . . . . . . . . 98
. . . . . . . . . . . . . . . . . . . . . . . . . 101
. . . . . . . . . . . . . . . . . . . . . . 106
. . . . . . . . . . . . . . . . . . . . . . . . . . 107
. . . . . . . . . . . . . . . . . . . . . 109
. . . . . . . . . . . . . . . . . . . . . . . . . . 117
. . . . . . . . . . . . . . . . . . . . . . . . 120
. . . . . . . . . . . . . . . . . . . . . . 124
. . . . . . . . . . . . 138
. . . . . . . . . . . . . . . . . . . . . . 138
. . . . . . . . . . . . . . . . . . . . . . . . . 140
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
. . . . . . . . . . 142
. . . . . . . . . 145
. . . . . . . . . . . . . . . . . . . . . 145
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150

69
70 2.1. BASIC MODULAR ARITHMETIC

2.1 Basic Modular Arithmetic

Consider the timestamp we use in our daily life. Certainly, there was a point when
people started counting time. Then, why is it not something like, 2147483647? Rather
we say something like 12.09 am (and there is a date of course, that separates two 12.09
am). The reason is, each time the hour hand in a clock crosses 12, it starts from 1
again, not 13. If the numbers kept going large, we would have a hard time realizing
what time we are living in. Similarly, when the second hand ticks 60 times, it starts
from 1 again (meaning it has been 1 minute, letting the minute hand tick once). Here,
intentionally or inadvertently, we have been using what number theorists call modular
arithmetic. The idea is, we keep the integers that leave the same remainder (when
divided by a certain integer) in the same class. It will be clear afterwards what exactly
we mean by class here when we discuss complete set of residue class. This PDF was created
for Amazon.

Definition 2.1.1. For a non-zero integer m, integers a and b are congruent modulo m
if and only if m|a − b. We show this by the notation

a≡b (mod m).

If m does not divide a − b, we say that a and b are not congruent modulo m and denote
it by a 6≡ b (mod m).

Note.

1. It is clear that if m|a − b, then −m|a − b. So from now on, we assume that m is
a positive integer.

2. If a is divisible by m, then a ≡ 0 (mod m). So, for example, an integer a is even

if and only if a ≡ 0 (mod 2).

Example. 3 ≡ 7 (mod 4), 52 ≡ −1 (mod 13), n2 − 1 ≡ 0 (mod n + 1).

Proposition 2.1.2. Assume that a and b are two integers and m is a positive integer.
Then the following propositions are correct.

i. If a is divided by b with remainder r, then a is congruent to r modulo b.

ii. If a ≡ b (mod m), then for any divisor d of m, a ≡ b (mod d).

iii. a ≡ a (mod m). We call this the reflexivity property of modular congruences.

iv. If a ≡ b (mod m), then b ≡ a (mod m). We call this the symmetry property.

v. If a ≡ b (mod m) and b ≡ c (mod m), then a ≡ c (mod m). We call this the
transitivity property.

vi. If a ≡ b (mod m) and c ≡ d (mod m), then a ± c ≡ b ± d (mod m) and ac ≡ bd

(mod m).

vii. If a ≡ b (mod m), then for any integer k, ka ≡ kb (mod m).

CHAPTER 2. MODULAR ARITHMETIC 71

Proposition 2.1.3. If n is a positive integer and a ≡ b (mod m), then an ≡ bn

(mod m).

Proof. From the definition, a ≡ b (mod m) means m|a − b. We know from Theorem
that
 
n n n−1 n−2 n−1
a − b = (a − b) a + a b + ··· + b .

This gives a − b|an − bn . So m|an − bn , or an ≡ bn (mod m).

Proposition 2.1.4. If f (x) is a polynomial with integer coefficients and a ≡ b (mod m),
then f (a) ≡ f (b) (mod m).

Proof. Assume f (x) = an xn + an−1 xn−1 + · · · + a1 x + a0 . Use Proposition ( ) to get

ai ai ≡ ai bi (mod m) and add up all the terms.

Proposition 2.1.5. If a is an integer and n is a positive integer, then exactly one of

the following relations holds.

a ≡ 0 (mod n),
a ≡ 1 (mod n),
..
.
a ≡ n − 1 (mod n).

Theorem 2.1.6. Let m be a positive integer and a, b, and c be integers. Then

(a) If ac ≡ bc (mod m) and gcd(c, m) = d, then

 m
a≡b mod .
d
We will call this the cancellation rule for congruence.

(b) If b ≡ c (mod m), then gcd(b, m) = gcd(c, m).

Before we prove it, let us see some examples. Usually, ac = bc implies a = c in

equations and so far we have seen congruences maintain equation relations. However,
is this any different in division? This is another trap you may fall into.
Since 15 divides 35 − 20, 5 · 7 ≡ 5 · 4 (mod 15). If we could just do division, this
would give us

7≡4 (mod 15),

which is clearly false. But, why? Here is the reason: 15 = 5 · 3. And when we canceled
5 without thinking where that 5 came from in 15, we accidentally took out the only
portion where 5 came from. So we can not do that recklessly. However, this also means
that if we took out 5 from all sides, it would be true:

7≡4 (mod 3).

72 2.1. BASIC MODULAR ARITHMETIC

Proof.

(a) The greatest common factor of c and m is d, so there exist integers c1 and m1 such
that

c = c1 d, m = m1 d, and gcd(c1 , m1 ) = 1.

Since ac ≡ bc (mod m), we have m = m1 d|(a − b)c = (a − b)c1 d. Canceling d from

both sides, we get m1 |(a − b)c1 . But gcd(c1 , m1 ) = 1, and so by Proposition ,
we get m1 |a − b. Thus,

a≡b (mod m1 ),

as desired.

(b) Because b ≡ c (mod m), there exists an integer k for which b − c = mk. So
gcd(b, m)|c. On the other hand, from definition of gcd, it is clear that gcd(b, m)|m.
Now by Proposition we have gcd(b, m)| gcd(c, m). Similarly, one can show
that gcd(c, m)| gcd(b, m). Using Proposition , we get gcd(c, m) = gcd(b, m).

Definition 2.1.7 (Arithmetic Progression). A sequence a1 , a2 , a3 , . . . of real num-

bers is called an arithmetic sequence, arithmetic progression, or AP if each new term of
the sequence is obtained by adding a constant real number d, called the common dif-
ference, to the preceding term. In other words, the terms of an arithmetic progression
are of the form

a, a + d, a + 2d, a + 3d, . . .

where a is the initial term of the sequence.

Example. The sequence of odd numbers is an AP. The following sequence

−3, 2, 7, 12, 17, . . .

which includes numbers of the form 5k + 2 for k = −1, 0, 1, . . . , is also an arithmetic

sequence with initial term −3 and common difference 5.

Corollary 2.1.8. All terms of an arithmetic progression are equivalent modulo the
common difference.

What is the sum of the terms of an arithmetic progression? Obviously, if the se-
quence has infinite number of terms, that is, if it has infinitely many terms, then the
sum is not a finite number as the common difference is constant . However, when the
arithmetic progression is finite (such as, a portion of the sequence), the sum of all its
elements is finite as well. Often we consider partial sum of such a series.
1
we say that it diverges.
CHAPTER 2. MODULAR ARITHMETIC 73

Theorem 2.1.9. Let (a1 , a2 , . . . , an ) be a finite arithmetic sequence such that

a1 = a,
a2 = a + d,
..
.
an = a + (n − 1)d,

where n is a positive integer and a and d are reals. The sum of all elements of this AP
is
n
X n n

ai = (a1 + an ) = 2a1 + (n − 1)d .
i=1
2 2

Proof. We have
n
X
ai = a1 + a2 + · · · + an
i=1
 
= a + (a + d) + · · · + a + (n − 1)d
 
= na + d 1 + 2 + · · · + (n − 1) .

From Identity of Appendix ( ), 1 + 2 + · · · + n = n(n − 1)/2. Therefore,

n
X n

ai = 2a1 + (n − 1)d .
i=1
2

Definition 2.1.10 (Geometric Progression). A sequence a1 , a2 , a3 , . . . of real num-

bers is called an geometric sequence, geometric progression, or GP if each new term of
the sequence is obtained by multiplying the previous term by a constant real number
r, called the common ratio. In other words, the terms of an arithmetic progression are
of the form

a, ar, ar2 , ar3 , · · ·

where a is the initial term of the sequence.

Example. The sequence of powers of 2 is a geometric progression. The sequence

1 1 1 1
, , , ,...
2 6 18 54
is a GP with initial term 1/2 and common ratio 1/3.
Similar to arithmetic progressions, the sum of terms of a finite geometric progression
is always possible to find. An interesting question would be what happens if we add
74 2.1. BASIC MODULAR ARITHMETIC

all the terms of an infinite geometric sequence? For example, what is the value of the
following sum?
1 1 1
1+ + + + ···
2 4 8
This is not a finite sequence. But is the sum divergent or convergent? The terms
of the above sequence are gradually decreasing and approach zero. To see this, notice
that the ninth term is
1
= 0.00390625,
256
which is very close to zero. So, on a second thought, we can guess that the given sum
has a finite value. In general, when the absolute value of common ratio of a geometric
progression is less than one, that is, when the absolute value of each term of the sequence
is smaller that its preceding term, then the geometric series (either finite or infinite)
has a finite value . We will see this from a different point of view. This is due to
Chamok Hasan, a senior brother of Masum.
Consider a pumpkin. Let us assume that it is totally symmetrical. Now, divide it
in half and put aside half of it. You have half of the pumpkin to yourself. Divide it
in half again. Keep one to yourself and discard the other half. So now you have one
fourth of the pumpkin. Again, cut it in half. Keep one, discard one. Now you have
one eighth. See that if you keep going this way, you end up getting 21 , 14 , 81 , 16
1
, . . . and
dividing them. And the fun fact is, you can keep doing this for as many times as you
want. Obviously, if we put together all the parts again, we get the whole pumpkin.
That is, if we take all the discarded portions and put them back, the pumpkin becomes
whole again. This shows us without any rigorous proof that
1 1 1 1
+ + + + · · · = 1.
2 4 8 16
Now you should be able to make sense how a sequence with infinite terms can have a
finite sum.

Theorem 2.1.11. Let be given the finite geometric sequence

a1 = a,
a2 = ar,
..
.
an = arn−1 ,

where n is a positive integer and a and r 6= 1 are reals. The sum of all elements of this
GP is
n
X a (rn − 1)
ai = .
i=1
r−1
2
to put it differently, it converges to a fixed value.
CHAPTER 2. MODULAR ARITHMETIC 75

Proof. Multiply the sum by (r − 1) to obtain

n
X
(r − 1) ai = (r − 1)(ar0 + ar1 + · · · + arn−1 )
i=1
= a(r − 1)(1 + r + · · · + rn−1 )
= a(rn − 1).
We have used Theorem to write the last line. Since r 6= 1, we can divide both
sides by r − 1 to get the desired result.
Did you notice anything? In fact, this is a special case of what we encountered in
divisibility. Recall the expansion of an − bn and try to find a correlation between the
two.
Corollary 2.1.12. Take the geometric progression in Theorem . If |r| < 1, then
the sum converges. More precisely,
X a
ai = .
i≥1
1−r

Proof. If |r| < 1, then |rn | decreases as we increase n. Therefore, when n is very large,
|rn | is almost zero. Here, we should borrow the idea of limit from calculus but for now,
let us convince ourselves that as n → ∞, rn → 0. This gives us
X a (0 − 1) a
ai = = ,
i≥1
r−1 1−r

which is what we wanted.

Note. In the footnote below this page, we used r∞ to mean that r is raised to a very
large power and thus is almost zero. The use of notation should not be misleading.
Example.
1 1 1
• We can now compute 1 + + + + · · · . In fact, this is an infinite geometric
2 4 8
series with initial term a = 1 and common ratio r = 21 < 1. Thus,
X  1 i 1
= 1 = 2.
i≥0
2 1 − 2

• Suppose that we want to find

2 + (−6) + 18 + (−54) + · · · + (−39366) + 118098
This is a geometric sequence with common ratio −3 and initial term 2. The last
term equals 2(−3)10 . So,
10
X
i 2 ((−3)11 − 1)
2(−3) = = 88, 573.
i=0
(−3) − 1
3
We are trying to avoid situations such as r∞ = 0 since that is a wrong concept. Because infinity
is not a number.
76 2.2. MODULAR EXPONENTIATION

2.2 Modular Exponentiation

In the early stage of problem solving, we all calculate big integers modulo an integer.
For example, consider the next problem.

Problem 2.2.1. Define an = 6n + 8n . Find the remainder of a49 when divided by 49.

The first idea that crosses your mind might be calculating 649 and finding the re-
mainder when divided by 49. This would be a large integer and the calculation is really
tedious, not to mention, pointless. A slight improvement would be multiplying 6 with
6 and taking modulo 49 each time. We need to do this for 49 times but at least, now
we do not have to deal with that large numbers anymore. Let us call this iterative
exponentiation method.
Suppose we want to find c ≡ ak mod n. The iterative exponentiation method
computes the values 1 = a0 , a1 , a2 , . . . , ak = c modulo n instead of directly calculating
ak modulo n. Suppose that we have computed ai modulo n for some i < k and the
result is b. According to the above theorem, to calculate ai+1 , all we need to do is to
compute a · b mod n. Obviously, a · b is much smaller than ai when i is large. This
is why this method takes less time for computations. Iterative exponentiation may be
expressed as an algorithm as shown below.

Iterative Exponentiation Algorithm

1. Set k1 ←− 0 and c ←− 1.

2. Increase k1 by 1.

3. Set c ←− a · c (mod n).

4. If k1 < k, go to step 2. Otherwise return c.

Example. Let’s compute 520 modulo 751 by iterative exponentiation algorithm. Table
shows 5i mod 751 for i = 1 to 20. As obtained from the table, 520 ≡ 200 (mod 751),
which is in agreement with what we previously found.

Remark. In the above example where a is small (compared to modulus n), we can
increase k1 more than one unit in each iteration of the algorithm. For example, in
above example, we could increase k1 two units each time to compute 52 , 54 , . . . , 520 . In
this case, the number of calculations is divided by two and therefore there will be less
time needed to find the result. This is done in general case but one must notice that
when one increases k1 , say, two units at each step, he is in fact computing a2 · c mod n
instead of a · c mod n in step 3 of the algorithm to reduce the number of iterations of
the algorithm. If a is small, there will be no difference in computation time. But if a
is (too) large, computing a2 · c mod n may will be more time consuming and it may
reduce the time efficiency of algorithm.

A more efficient method to do this is modular exponentiation algorithm. The beauty

of this idea is that you can use it to compute big integers modulo n by hand. The idea
CHAPTER 2. MODULAR ARITHMETIC 77

k1 5k1 (mod 751) k1 5k1 (mod 751)

1 5 11 358
2 25 12 288
3 125 13 689
4 625 14 441
5 121 15 703
6 605 16 511
7 21 17 302
8 105 18 8
9 525 19 40
10 372 20 200

Table 2.1: Applying iterative exponentiation method to find 520 modulo 751.

actually inherits from binary representation. Consider the binary number (101101)2 .
We discussed how to convert it to a decimal integer in base conversion. However,
Masum uses a variation for faster mental calculation. Start from the left most digit
(which always will be 1 if there is no leading 0). Initially, the decimal integer is 1. Now,
go to the next digit. If it is 0, double the current value. If it is 1, double and add 1.
Since the next digit is 0, we have 2. Next digit is 1. So it will become 2 · 2 + 1 = 5.
Next digit is 1 as well. It will be 5 · 2 + 1 = 11. Next digit is 0, so we have 11 · 2 = 22.
Next digit is 1, so it will be 22 · 2 + 1 = 45. There is no more digits left, so this is the
desired value in decimal. You can verify that this indeed is the intended result. And
more importantly, think why this works if you have not figured it out already!
We just saw a way of converting binary numbers into decimal. How does that help
us in modular exponentiation? Assume that we want ak (mod n). We will not compute
it directly or iteratively. Instead, represent k in binary. Then, initially, the result is 1.
Divide k by 2 and keep the remainder. If it is 1, multiply the current result by a and
do the modulo operation, that is, r → ra (mod n). Also, set a → a2 (mod n). Keep
doing this until k = 0. In the end we will see r ≡ ak (mod n). Again, make sense why
this works. Do the example above this way and see if the result matches. Algorithm to
find ak (mod n).

Modular Exponentiation Algorithm (ak (mod n))

1. Set R ←− 1.

2. If k = 0, stop and return the value R. Otherwise, continue.

3. Divide k by 2, take the remainder r. That is, set k ←− ⌊k/2⌋.

4. Set a ←− a2 (mod n).

5. If r = 1, set R ←− Ra (mod n).

6. Go to step 2.
78 2.2. MODULAR EXPONENTIATION

However, we face another concern here. What if Ra is very large? We can take care of
it the same way. Express a in binary and take modulo from there. Algorithm to find
ab (mod n) for large b.
1. Set R = 1.
2. If b = 0, return R, otherwise continue.
3. Set b ←− ⌊b/2⌋ and r = b (mod 2).
4. If r = 1, set R ←− R + a (mod n).
5. Set a = (2 · a) (mod n).
6. Go to step 2.
We can call this modular multiplication. This way, we will not have to actually multiply
two numbers to get the remainder. The proofs for the last two ideas were not shown
deliberately. We expect that you can do it easily. By the way, did you notice something
else too? In modular exponentiation, we do not have to iterate k times. The number of
times we need to iterate is actually ⌊log2 (k)⌋ + 1 (again, why?). Same goes for modular
multiplication. Therefore, it is a very desirable improvement. In fact, these methods are
highly used in primality tests or similar fields (we will discuss about primes in Chapter
).
Notice that, we can write modular exponentiation algorithm in a better fashion.

Modular Exponentiation Algorithm - Cleaner Version (ak (mod n))

1. Set R = 1.
2. Represent k in binary. Assume k = (x0 x1 · · · xl )2 .
3. If k = 0, return R.
4. Find r = (k (mod 2)).
5. If r = 1, set R ←− Ra (mod n).
6. Set k ←− ⌊k/2⌋.
7. Set a ←− a2 (mod n).
8. Go to step 3.
Example. Let us calculate 520 (mod 751) this way. First, we need to find the binary
representation of 20, which is (10100)2 . Then, we can write
R
z }|3 {
R2
 z }| { 2 ! 2

2
520 ≡ 52
|{z} ·5 (mod 751).
R1
| {z }
R4
| {z }
R5
CHAPTER 2. MODULAR ARITHMETIC 79

This is how we proceed: we want to construct 520 . The rightmost digit is zero. What
happens if we remove this digit? The number gets divided by 2. This is identical to
writing 520 = 52 · 510 . Therefore, we first compute R1 = 52 . We now need to construct
510 . The binary representation of 10 is (1010)2 . Again, divide it by two to remove the
rightmost zero. This time, we are doing this operation:

520 = R1 · 52 · 55 .

We must compute R2 = R1 · 52 at this stage. Now, how do we construct 55 given its

binary representation (101)2 , which does not end in zero? It’s easy. We just have to
write it as 1 + (100)2 . Now, we have (100)2 which ends in zero. In other words,

520 = R2 · 5 · 54

So, we calculate R3 = R2 · 5 at this stage and try to construct 54 . The rest of the
solution is similar and we expect the reader to finish it. Try to find R4 and R5 for
yourself. In case you want to check your answers, you can consult table .

i 1 2 3 4 5
Ri 25 625 121 372 200

Table 2.2: Applying modular exponentiation method to find 520 modulo 751.

2.3 Residue Systems

Residue systems are very simple definitions which will help you make a good sense of
some later-explained theorems such as Fermat’s and Euler’s.

2.3.0.1 Complete Residue Systems

The definitions and theorems below assume that m is a positive integer.

Definition 2.3.1. Two integers a and b are said to be members of the same residue
class modulo m, if and only if a ≡ b (mod m).

Clearly, there are m distinct residues modulo m.

Definition 2.3.2. Let m be a positive integer. The set A is called a complete residue
system modulo m if and only if every number is congruent to a unique element of A
modulo m. In other words, A should be representing all the residue classes modulo m.

Example. A = {0, 1, . . . , m − 1} is a complete residue system modulo m. So is B =

{15, 36, −7, 27, 94} modulo 5.
We will state two simple propositions without proof. The reader should be able to
prove them by his/her own.
80 2.3. RESIDUE SYSTEMS

Proposition 2.3.3. The set A = {a1 , a2 , . . . , ak } is a complete residue set (or system)
modulo m if and only if k = m and ai 6≡ aj (mod m) for i 6= j.
Proposition 2.3.4. Let A = {a1 , a2 , . . . , am } be a complete residue set modulo m and
let a, b be integers such that a⊥m. Then the set

B = {aa1 + b, aa2 + b, . . . , aam + b}

is also a complete residue set modulo m.

2.3.0.2 Reduced Residue Systems and Euler’s Totient Function

We really wish that you have a firm grasp of function. However, if you are in 10th grade
or below, there is a good chance, you are not familiar with the concept of functions
very well. Since that is entirely a different topic, we restrain ourselves from discussing
it. Make sure you at least realize what function actually is. Here, we will say a thing
about function or two but it is not nearly enough for covering the fundamentals.
A function is like a machine. It takes a number as its input, functions on the number,
and gives another number as its output with the property that each input is related to
exactly one output. This property seems logical. Consider a weighing scale designed to
measure the weight of people. Obviously, a person cannot be both 70 and 75 kilograms
at the same time. The weight of a person (in a specific time) is a constant number, and
hence the weighing scale actually works as a function: it takes a person as its input,
measures his weight, and then shows the person’s weight as its input.
Another example would be a function that takes a real number x as its input and
gives x2 as its output. For convenience, we can call this function f : R → R and write
its relation as f (x) = x2 for all x ∈ R. The notation f : S → T means that the function
f takes its inputs from the domain S (the set of inputs) and assigns them an output
from the codomain T (the set of outputs and maybe some additional elements). For
the previous example, we see that both domain and codomain of f are R. However,
the range (or image) of f , which is the set containing only outputs of f , is R+ , the set
of all positive real numbers.
Definition 2.3.5 (Euler’s Totient Function). For every positive integer n > 1, ϕ(n)
is the number of positive integers less than or equal to n which are relatively prime to
n. We call this function Euler’s phi function (or totient function).
Example. ϕ(5) = 4 and ϕ(10) = 4.
We will investigate properties of this function in details in Chapter . For now let’s
just assume the following claims are true.
Proposition 2.3.6 (Properties of Euler’s Totient Function). Let m and n be two
positive integers.
(a) ϕ is a multiplicative function. That is, if m⊥n, then

ϕ(mn) = ϕ(m) · ϕ(n).

4
You might be wondering what totient means. In Latin, tot means so many. The suffix of iens is
probably from the Sanskrit.
CHAPTER 2. MODULAR ARITHMETIC 81

(b) For all n ≥ 3, ϕ(n) is even.

(c) ϕ is neither increasing , injective nor surjective .

(d) If n is factorized as n = pα1 1 pα2 2 · · · pαk k , then

    
1 1 1
ϕ(n) = n 1 − 1− ··· 1 −
p1 p2 pk
α1 −1 α2 −1 αk −1
= p1 p2 · · · pk (p1 − 1) · · · (pk − 1) .

Why do we require function in congruence? Moreover, Euler’s totient function?

Before you decide it sounds irrelevant, take a look at the following example.
Consider the integers {1, 2, 3, 4, 5, 6} (complete set of residue class modulo 7 except
0). Now, take an integer, say 3. Multiply the whole set with 3 and find the residues
again:

{3, 6, 9, 12, 15, 18} ≡ {3, 6, 2, 5, 1, 4} (mod 7).

Does this look interesting? If not, take a look again. And try to understand what
happened and why. Firstly, the products forms a residue class as well. Alternatively, it
is a permutation of the residue class. Why? What if we multiplied by 5? Check it out
yourself and see if the conclusion holds. Check for some more integers like 10, 13, 14
etc. You will see the same is true for all integers except 0, 7, 14, . . . i.e. multiples of 7.
Again, why? 7 is a prime. So we know if an integer is not divisible by 7, it is co-prime
to 7. Take a such that a⊥7. Now, what does it mean that the set of products is a
permutation of the original? We could state it this way: no two products leave the
same remainder when divided by 7. And you can see, if this is true, everything makes
sense. If we can show that for 0 < i < j < 7, ia and ja are not congruent modulo 7,
we are done! That is indeed the case. For the sake of contradiction, assume that,

ia ≡ja (mod 7)
⇐⇒ 7|ia − ja = a(i − j)

Here a⊥7, so we have 7|i − j. But remember that, i < j < 7 so |i − j| < 7. This yields
the contradiction we were looking for. This claim was true mainly because a⊥7. What
if we did not take a prime 7? Well, we could still do something similar. And that is
why Euler’s totient function comes to the play. This is more valuable than you may
realize.

Definition 2.3.7. Let m be a positive integer. The set A is called a reduced residue
system modulo m if all elements of A are coprime to m, and also every integer which
is coprime to m is congruent to a unique element of A modulo m.
5
The function f is increasing if for a1 > a2 , we have f (a1 ) > f (a2 ).
6
The function f is injective if for a1 6= a2 , we have f (a1 ) 6= f (a2 ).
7
The function f : X → Y is surjective if for every y ∈ Y , there exists x ∈ X such that f (x) = y.
82 2.4. BÉZOUT’S LEMMA

Example. The set A = {1, 2, . . . , p − 1} is a reduced residue system modulo a prime p.

The set {7, 17} is also a reduced residue system mod 6.
You can clearly sense how the Euler’s phi function is related to reduced residue
systems: the number of elements of A is φ(m). So we can express the above definition
in a better way:

Proposition 2.3.8. The set A = {a1 , a2 , . . . , ak } is a reduced residue set (or system)
modulo m if and only if

• ai ⊥ m for all i,

• k = ϕ(m), and

• ai 6≡ aj (mod m) for i 6= j.

An important aspect of the reduced systems is stated in the next proposition. It

will help us prove the Euler’s theorem later.

Proposition 2.3.9. Let A = {a1 , a2 , . . . , aϕ(m) } be a reduced residue set modulo m

and let a be an integer such that a⊥m. Then the set

B = {aa1 , aa2 , . . . , aaϕ(m) }

is also a reduced residue set modulo m.

The proof of this theorem is pretty easy, try it for yourself. Pay attention to the
difference between this proposition and the similar Proposition for complete sys-
tems.
The latest theorem says that there are infinitely many reduced residue systems for
any m. So, it makes sense to define a set as the original reduced residue system for any
positive integer m. We call this set Um .

Definition 2.3.10. Let m be a positive integer. The set of units modulo m, Um , is

the set of positive integers g1 , . . . , gϕ(m) less than m which are coprime to m.

Example. U8 = {1, 3, 5, 7}, and U15 = {1, 2, 4, 7, 8, 11, 13, 14}. If p is a prime, then
Up = {1, 2, . . . , p − 1}.
You might be wondering why we call Um the set of units. In algebraic structures, a
unit is an element a for which there exists some element b such that ab = 1. In our case,
the number a is a unit if there exists some b such that ab ≡ 1 (mod m). As proved
before, a is a unit if and only if it is coprime to m, and this shows us why Um is called
the set of units.

2.4 Bézout’s Lemma

In this section, we are going to explain the simple but useful Bézout’s lemma and then
introduce modular multipliccative inverses.
CHAPTER 2. MODULAR ARITHMETIC 83

2.4.1 Bézout’s Identity and Its Generalization

Before representing this lemma, we would like you define the linear combination of two
integers.
Definition 2.4.1. For two integers a and b, every number of the form

ax + by

is called a linear combination of a and b, where x and y are integers.

For example, 2a + 3b and −4a are both linear combinations of a and b, but a2 − b is
not. The Bézout’s lemma (sometimes called Bézout’s identity states that for every two
integers a and b, there exists a linear combination of a and b which is equal to (a, b).
For example if a = 18 and b = 27, then 18 · (−1) + 27 · 1 = 9 = (18, 27).
Theorem 2.4.2 (Bézout’s Identity). For two nonzero integers a and b, there exists
x, y ∈ Z such that

ax + by = (a, b)

You are probably familiar with this theorem. A simple proof uses Euclidean division,
but it doesn’t show you where exactly to use this identity. So, we prove a stronger
theorem and the proof of Bézout’s identity is immediately implied from it.
Theorem 2.4.3. Let a, b, m be integers such that a, b are not zero at the same time.
Then the equation

ax + by = m

has solutions for x and y in positive integers if and only if (a, b)|m.
Proof. The first part is easy. Suppose that there exist integers x0 and y0 such that

ax0 + by0 = m.

We know that (a, b)|a and also (a, b)|b, thus (a, b)|m and we are done.
Conversely, if (a, b) = d and m is divisible by d, then we want to prove that there
exist some positive integers x and y for which ax + by = m. First, we show that it’s
sufficient to show that there exist x and y such that

ax + by = d.

The reason is simple: if there exist x and y such that ax + by = d, then

 m  m
a x +b y = m.
d d
Assume that A is the set of all positive integer linear combinations of a and b. A is
non-empty because if a 6= 0, then
|a|
0 < |a| = a + b · 0,
a
84 2.4. BÉZOUT’S LEMMA

and if b 6= 0, then
|b|
0 < |b| = a · 0 + b .
b
Because of well-ordering principle , A contains a least element. Let this smallest element
be t. So there exist integers x0 and y0 such that

ax0 + by0 = t.

We claim that t|a and t|b. Using division theorem, divide a by t:

a = tq + r, 0 ≤ r ≤ t,

and thus

a (1 − qx0 ) +b (−qy0 ) = a − tq = r < t.

| {z } | {z }
=x1 =y1

If r 6= 0, then r is a positive integer written in the form ax1 + by1 , which is a positive
integer linear combination of a and b, so r ∈ A. But r < t, which is in contradiction
with minimality of t. Therefore r = 0 and t|a. We can prove that t|b in a similar way.
By Proposition , we find that t|d. Also, according to the first part of the proof, we
have d|t. Following Proposition , t = d. This means that d ∈ A and there exist
integers x and y such that

ax + by = d.

Bézout’s Identity has many interesting applications. We will see one such application
in Chapter , to prove Chicken McNugget Theorem.
We are now ready to represent a stronger version and also a generalization of Bé-
zout’s lemma.
Corollary 2.4.4. [Stronger Form of Bézout’s Identity] The smallest positive integer
linear combination of a and b is (a, b).
Corollary 2.4.5. If a ⊥ b for non-zero integers a and b, then there exist integers x and
y such that

ax + by = 1.

Theorem 2.4.6 (Generalization of Bézout’s Identity). If a1 , a2 , . . . , an are integers

with (a1 , a2 , . . . , an ) = d, then the equation

a1 x1 + a2 x2 + · · · + an xn = m

has a solution (x1 , x2 , . . . , xn ) in integers if and only if d|m.

8
The well-ordering principle states that every non-empty set of positive integers contains a least
element.
CHAPTER 2. MODULAR ARITHMETIC 85

Theorem 2.4.7. Let m be a positive integer and let a and b be positive integers. Then
the modular arithmetic equation

ax ≡ b (mod m)

has a solution for x in integers if and only if (m, a)|b.

Proof. Rewrite the congruence equation as ax − my = b for some integer y. Now it

is the same as Theorem . The equation ax − by = m has solutions if and only if
(m, a)|b, which is what we want.

Problem 2.4.8. Let a, b, and c be non-zero integers such that (a, c) = (b, c) = 1. Prove
that (ab, c) = 1.

Solution. By Corollary , there exist integers x, y, z, and t such that

ax + cy = 1, and
bz + ct = 1.

Multiply these two equations to get

1 = (ax + cy)(bz + ct)

= ab(xz) + c(axt + byz + cyt).

This means that we have found a linear combination of c and ab which is equal to 1.
From Corollary it follows that (ab, c) = 1 (why?).

Let’s prove the second part of proposition ( ) in section ( ).

Problem 2.4.9. Let a, b, and c be integers. If a|bc and (a, b) = 1, prove that a|c.

Solution. The problem is obvious for c = 0. Assume that c 6= 0. Since (a, b) = 1, there
exist integers x and y such that ax + by = 1. Multiply both sides of this equation by c
to obtain acx + bcy = c. Because a divides both acx and bcy, it must also divide their
sum, which is equal to c.

2.4.2 Modular Arithmetic Multiplicative Inverse

When speaking of real numbers, the multiplicative inverse of x – usually named recip-
rocal of x – is x1 . This is because x · x1 = 1 for non-zero x.
The definition of a multiplicative inverse in modular arithmetic must be more clear
for you now.

Definition 2.4.10. Let a be an integer and let m be a positive integer. The modular
multiplicative inverse of a modulo m is an integer x such that

ax ≡ 1 (mod m).

Once defined, x may be denoted by a−1 and simply called inverse of a.

86 2.4. BÉZOUT’S LEMMA

Note. Unlike real numbers which have a unique reciprocal, an integer a has either no
inverse, or infinitely many inverses modulo m.

Example. An inverse for 3 modulo 7 is 5:

3·5≡1 (mod 7).

We can easily generate other inverses of 3 modulo 7 by adding various multiples of 7 to

5. Thus, the numbers in the set {. . . , −2, 5, 12, 19, . . .} are all inverses of 3 modulo 7.
Example. An inverse for 216 + 1 modulo 231 − 1 is 216 − 1. In fact,

(216 − 1)(216 + 1) = 232 − 1 = 2(231 − 1) + 1 ≡ 1 (mod 231 − 1).

Theorem 2.4.11. Let a be an integer and let m be a positive integer such that a ⊥ m.
Then a has an inverse modulo m. Also, every two inverses of a are congruent modulo
m.

Proof. The proof is a straightforward result of corollary ( ). Since a ⊥ m, the

equation ax + my = 1 has solutions. Now take modulo m from both sides to complete
the proof of the first part. For the second part, assume that x1 and x2 are inverses of
a modulo m. Then,
(a,m)=1
ax1 ≡ ax2 ≡ 1 (mod m) =⇒ x1 ≡ x2 (mod m),

as desired.

The uniqueness of inverse of an integer a modulo m gives us the following corollary.

Corollary 2.4.12. For a positive integer m, let {a1 , a2 , . . . , aϕ(m) } be a reduced residue
system modulo m. Then {a−1 −1 −1
1 , a2 , . . . , aϕ(m) } is also a reduced residue system modulo
m.

Problem 2.4.13. Find the unique odd integer t such that 0 < t < 23 and t + 2 is the
modular inverse of t modulo 23.

Solution. This means that t(t+2) ≡ 1 (mod 23). Add 1 to both sides of this congruence
relation to get (t+1)2 ≡ 2 ≡ 25 (mod 23). Therefore, 23|(t+1)2 −25 or 23|(t−4)(t+6).
By Euclid’s lemma (Proposition ), 23|t − 4 or 23|t + 6, which give t = 4 and t = 17
as solutions. Since we want t to be odd, the answer is t = 17.

We are going to prove a very simple fact which will be very useful later (for instance,
in the next theorem or in the proof of Wolstenholme’s theorem, where we re-state the
same result as Lemma ).

Proposition 2.4.14. For a prime p ≥ 3 and any positive integer a coprime to p,

(a−1 )n ≡ (an )−1 (mod p),

for all positive integers n.

CHAPTER 2. MODULAR ARITHMETIC 87

Proof. Since a is coprime to p, a−1 exists. Therefore,

a · a−1 ≡ 1 (mod p) =⇒ an · (a−1 )n ≡ 1 (mod p)

=⇒ (a−1 )n ≡ (an )−1 (mod p),

as desired.

Theorem 2.4.15. Let a, b be integers and x, y, and n be positive integers such that
(a, n) = (b, n) = 1, ax ≡ bx (mod n), and ay ≡ by (mod n). Then,

a(x,y) ≡ b(x,y) (mod n).

Proof. By Bézout’s identity, we know there are integers u and v so that ux+vy = (x, y).
Therefore,

a(x,y) ≡ aux+vy
(2.1) ≡ (ax )u · (ay )v ≡ (bx )u · (by )v
≡ bux+vy ≡ b(x,y) (mod n).

Remark. Thanks to Professor Greg Martin, we should point out a very important
detail here. In the computations above, we used the fact that there exist integers u
and v such that ux + by = 1. One must notice that these integers u and v need
not be positive. In fact, if x and y are both positive, then u and v cannot be both
positive (why?). But that doesn’t make our calculations wrong, due to Proposition
. If it’s not clear to you yet, think of it in this way: suppose that, say, u is
negative. For instance, consider the example when x = 3 and y = 15. Then, since
3 · (−4) + 15 · 1 = (3, 15), we have u = −4 and v = 1 . Then, equation ( ), would look
like
(a3 )−4 · (a15 )1 ≡ (b3 )−4 · (b15 )1 (mod n).
This might not seem normal because we have a −4 in the exponents. So, using Propo-
sition , we can write the above congruence equation as

3 4 
3 4
a−1 · (a15 )1 ≡ b−1 · (b15 )1 (mod n).

Notice that we need (a, n) = 1 and (b, n) = 1 to imply a−1 and b−1 exist modulo n.

Problem 2.4.16. Prove that,  

(m, n) m
m n
is an integer.

Since this problem is juxtaposed with this section, it is obvious we are going to use
this theorem. But in a real contest, that is not the case at all.
88 2.5. CHINESE REMAINDER THEOREM

Solution. Since there are integers x, y with (m, n) = mx + ny, it is easy to deduce that:
   
(m, n) m mx + ny m
=
m n m n
   
m ny m
=x +
n m n
   
m ny m m − 1
=x + ·
n m n n−1
   
m m−1
=x +y .
n n−1

Now we can say this is an integer. See how tactfully we tackled this problem.

2.5 Chinese Remainder Theorem

Chinese Remainder Theorem –usually called CRT – is a very old principle in mathe-
matics. It was first introduced by a Chinese mathematician Sun Tzu almost 1700 years
ago. Consider the following example.

Problem 2.5.1. A positive integer n leaves remainder 2 when divided by 7 but has a
remainder 4 when divided by 9. Find the smallest value of n.

You might have encountered similar problems when you were in 4th or 5th grade.
May be even more basic ones. But the idea is essentially the same. If the problem was
a bit different, like

Problem 2.5.2. A positive integer n leaves remainder 2 when divided by 7 or 9. Find

the smallest value of n.

Then it would be easier. Because then we would have that n − 2 is divisible by

both 7 and 9. That means n − 2 has to be divisible by their least common multiple,
63. Obviously, the minimum such n is n = 2. Let’s see what happens if we want n > 2.
Then all such positive integers would be n = 2 + 63k. Now, as for this problem, we
can’t do this directly when the remainders are different. So we go back to the original
problem and see how we can tackle the new one. Let’s write them using congruence.

n≡2 (mod 7),

n≡4 (mod 9).

In other words, using divisibility notation, 7|n − 2 and 9|n − 4. We can not do the same
now. But if these two remainders were same, we could do that. Probably we should
focus on that. That is, we want it to be something like

n ≡ a (mod 7),
n ≡ a (mod 9).
CHAPTER 2. MODULAR ARITHMETIC 89

The only thing we can do here is

n ≡ 2 + 7k (mod 7),
n ≡ 4 + 9l (mod 9),
for some suitable integer k and l. Our aim is to find their values. Since both 2 + 7k
and 4 + 9l must be the same modulo 7 and 9, if we can find a way to keep 9 in 2 + 7k
and 7 is 4 + 9l, that could work! One way around it is to do the following:
n ≡ 2 · 1 + 7 · 4 (mod 7),
n ≡ 2 · 9 · 9−1 + 7 · 4 (mod 7).
Let’s do the same for the other congruence.
n ≡ 4 · 1 + 9 · 2 (mod 9),
n ≡ 4 · 7 · 7−1 + 2 · 9 (mod 9).
Now you should understand what we can do to make those two equal. In the first
congruence, no matter what we multiply with 7 · 4, the remainder won’t change modulo
7. The same for 9 in the second congruence. We will exploit this fact. We need to
rearrange it just a bit more. But here is a warning. We wouldn’t be able to do it if 7
and 9 were not coprime, since then they would not have any multiplicative inverse. We
could do that trick writing 1 = 7 · 7−1 only because 7−1 modulo 9 exists. For simplicity,
let’s assume 7−1 ≡ u (mod 9) and 9−1 ≡ v (mod 7).
n≡2·9·v+4·7·u (mod 7)
n≡4·7·u+2·9·v (mod 9)
And now, we have what we want! We can say,
n≡2·9·v+4·7·u (mod 7 · 9),
since 7 ⊥ 9. We have our solution! Think more on our approach and what led us to do
this. Question is, is this n the smallest solution? If we take r with 0 ≤ r ≤ mn so that
n ≡ r ≡ 18v + 28u (mod 63),
where u ≡ 7−1 ≡ 4 (mod 9) and v ≡ 9−1 ≡ 4 (mod 7). Therefore,
n = (18 · 4 + 28 · 4) (mod 63) = 184 (mod 63) = 58.
Since 58 < 63, such a solution will be unique! Mathematically, we can write it this way.
Let the inverse of a modulo n be a−1n .

Theorem 2.5.3 (Chinese Remainder Theorem for Two Integers). For two pos-
itive integers a⊥b,
x ≡ m (mod a),
x ≡ n (mod b),
has a solution
−1
x0 ≡ (mbb−1
a + naab ) (mod ab),
and all the solutions are given by x = x0 + abk.
90 2.5. CHINESE REMAINDER THEOREM

But this form is not that convenient. We will give it a better shape. Let M = ab,
then Ma ⊥b and Mb ⊥b. Rewrite the theorem in the following form.

Theorem 2.5.4 (Refined CRT). If a1 ⊥a2 and M = a1 a2 , then the congruences

x ≡ r1 (mod a1 ),
x ≡ r2 (mod a2 ),

has the smallest solution

   −1    −1 !
M M M M
x 0 ≡ r1 + r2 (mod M ).
a1 a1 a2 a2 a2 a 1

If we take n relatively prime integers instead of two, the same process will work! So
we can generalize this for n variables.

Theorem 2.5.5 (CRT). For n pairwise coprime integers a1 , a2 , . . . , an there exists a

solution to the congruences

x ≡ r1 (mod a1 ),
x ≡ r2 (mod a2 ),
..
.
x ≡ rn (mod an ).

M
If M = a1 a2 · · · an and Mi = and Mi ei ≡ 1 (mod ai ), then the smallest modulo M
ai
is given by
n
!
X
x0 ≡ (r1 M1 ei + . . . + rn Mn en ) ≡ ri Mi e i (mod M )
i=1

Proof. Note that, for a fixed i, Mj is divisible by ai if i 6= j. Therefore,

n
X
x0 = r i M i e i ≡ ri Mi e i ≡ ri (mod ai )
i=1

So x0 is a solution to those congruences. Since Mi ⊥ai , there is a multiplicative inverse

of Mi modulo ai due to Bézout’s identity. We leave it to the reader to prove that if
x, y are two solutions, then x ≡ y (mod M ). That would prove its uniqueness modulo
M.

We want to mention a particular use of CRT. When you are facing some problems
related to congruence equation, if you can not solve for some n, instead show a solution
for pei i where n = pe11 · · · pekk . Then you can say that such a solution modulo n exists as
well. In short, we could reduce the congruences to prime powers because p1 , . . . , pk are
pairwise coprime integers. By the way, we could generalize CRT the following way.
CHAPTER 2. MODULAR ARITHMETIC 91

Theorem 2.5.6 (General CRT). For n integers a1 , . . . , an the system of congruences

x ≡ r1 (mod a1 ),
x ≡ r2 (mod a2 ),
..
.
x ≡ rn (mod an ).

has a solution if and only if

ri ≡ rj (mod (ai , aj )),

for all i and j. Any two solutions x, y are congruent modulo the least common multiple
of all ai . That is, if M = [a1 , . . . , an ] and x, y are two solutions, then x ≡ y (mod M ).
Problem 2.5.7. Prove that, for any n there are n consecutive integers such that all of
them are composite.
Solution. We will use CRT here forcibly, even though it has a much easier solution.
Consider the following congruences:

x ≡ −1 (mod p1 p2 ),
x ≡ −2 (mod p3 p4 ),
..
.
x ≡ −n (mod p2n−1 p2n ).

Here, p1 , · · · , p2n are distinct primes. Therefore, M1 = p1 p2 , · · · , Mn = p2n−1 p2n are

pair-wisely coprime. So, by CRT, there is indeed such an x which satisfies all of the
congruences above. And our problem is solved. Notice that, x + 1 is divisible by at
least two primes p1 , p2 . Similarly, x + i is divisible by p2i−1 p2i .
Note. A common idea in such problems is to bring factorial into the play. Here,
(n + 1)! + 2, (n + 1)! + 3, · · · , (n + 1)! + (n + 1) are such n consecutive integers. But
the motivation behind the solution above is that, we making use of the fact: primes are
coprime to each other. And to make an integer composite, we can just use two or more
primes instead of one.
Problem 2.5.8. Suppose that {s1 , s2 . . . , sφ(m) } is a reduced residue set modulo m.
Find all positive integers a for which {s1 + a, s2 + a . . . , sφ(m) + a} is also a reduced
residue set modulo m.
Solution. We claim that the given set is a reduced residue system modulo m if and
only if a is divisible by each prime factor of m.
First, suppose that m has a factor p and a is not divisible by p. Let m = pα n for
some positive integer n coprime to p. Since n⊥p, by CRT, there exists some integer k
such that

k ≡ −a (mod p),
k ≡ 1 (mod n).
92 2.6. WILSON’S THEOREM

Since k ≡ −a 6≡ 0 (mod p), we have k⊥p. Also, let (k, n) = d. Then d | n | k − 1 and
d|k, meaning d|1 and so d = 1. It follows that k⊥m. So, k ∈ {s1 , s2 . . . , sϕ(m) }. But
k+a is divisible by p, and therefore not coprime to n, forcing {s1 +a, s2 +a . . . , sϕ(m) +a}
not a reduced residue system.
For the converse, suppose that a is an integer which is divisible by all prime factors
of m. Obviously, s1 + a, s2 + a . . . , sϕ(m) + a are all distinct modulo m. We just need
to show that if s is coprime to m, then so is s + a. For any prime p which divides m,
we have s + a ≡ s (mod p) because as assumed, a is divisible by p. Since s is coprime
to p, so is s + a. Thus s + a is coprime to all prime factors of m, making it relatively
prime to m as well.
Problem 2.5.9 (1997 Czech and Slovak Mathematical Olympiad). Show that
there exists an increasing sequence {an }∞
n=1 of natural numbers such that for any k ≥ 0,
the sequence {k + an } contains only finitely many primes.
It is a standard example of CRT because it is not obvious how CRT comes into the
play here.
Solution. Let pk be the kth prime number. Set a1 = 2. For n ≥ 1, let an+1 be the
least integer greater than an that is congruent to −k modulo pk+1 for all k ≤ n. Such
an integer exists by the Chinese Remainder Theorem. Thus, for all k ≥ 0, k + an ≡ 0
(mod pk+1 ) for n ≥ k + 1. Then at most k + 1 values in the sequence {k + an } can be
prime since the ith term onward for i ≥ k + 2, the values are nontrivial multiples of
pk+1 and must be composite. This completes the proof.
Note. We could deal with this using an = (pn − 1)! as well, combining with Wilson’s
theorem. Because if k > 1 then pn − 1 > k for sufficiently large n so it will be composite
from that n. Otherwise (p − 1)! + 1 is divisible by p, so it is composite as well.

2.6 Wilson’s Theorem

We have probably discussed that if n > 4 is a composite integer, then (n−1)! is divisible
by n. What if n is a prime? Take n = 3, then (n − 1)! = 2, not divisible by 3. Take
n = 5. (n − 1)! = 24, which is not divisible by 5. Take n = 7, then (n − 1)! = 120
which is not divisible by 7. Take n = 11, (n − 1)! = 3 628 800 and this is not divisible
by 11 either. Since they are not divisible by the primes (as expected), we should check
for remainders. 2! leaves a remainder 2 when divided by 3. 4! leaves 4 when divided by
5, 6! leaves 6 when divided by 7. If you calculate further, you will see the pattern goes
on. So it suggests us to conjecture that the remainder of (p − 1)! when divided by the
prime p is p − 1. In fact, this is what we call Wilson’s theorem.
Theorem 2.6.1. Let p be a prime number and a is a positive integer. Show that if the
inverse of a modulo p is equal to a, then a ≡ 1 or p − 1 (mod p).
Proof. The proof for p = 2 is obvious, so assume that p > 2. The inverse of a is itself,
so a2 ≡ 1 (mod p). This means that p | a2 − 1. So, p divides (a − 1)(a + 1). We know
that 2|(a − 1, a + 1), so p divides either a + 1 or a − 1, which results in a ≡ 1 or p − 1
(mod p).
CHAPTER 2. MODULAR ARITHMETIC 93

Now, why are we concerned about such a situation at all? The inverse of a being
a is not something of interest, at least not obviously. In fact, there is a reason behind
it. And you should have thought of it before reaching this point. Notice that, in the
congruence

(p − 1)! ≡ (p − 1) (mod p),

(p − 1) is co-prime to p. So we can cancel it from both sides and get (p − 2)! ≡ 1

(mod p). Now, this is interesting and it asks us to reach 1 from a product. Notice the
following rearrangement for p = 11.

9! = 1 · 2 · · · 9
= (2 · 6) · (3 · 4) · (5 · 9) · (7 · 8)
= 12 · 12 · 45 · 56
≡ 1 · 1 · 1 · 1 (mod 15).

Does it make sense now why the theorem above is necessary? If not, think a little bit
more. Then proceed and you will realize we have actually found the crucial step to
prove Wilson’s theorem.
According to this theorem, the only two numbers in the set {1, 2, . . . , p − 1} which
have their inverse equal themselves are 1 and p − 1. This will help us to prove the
Wilson’s theorem.

Theorem 2.6.2 (Wilson’s Theorem). The positive integer p > 1 is a prime if and
only if (p − 1)! ≡ −1 (mod p).

Proof. We divide the proof of this theorem into two parts. First, we show that if p
is a prime, then (p − 1)! ≡ −1 (mod p). The result is obvious for p = 2 and 3. So
we assume that p ≥ 5. According to the Theorem , the only two numbers among
{1, 2, . . . , p − 1} which have their inverse equal themselves are 1 and p − 1. Put them
away and consider the set A = {2, 3 . . . , p − 2}. There are p − 3 elements in this set,
and each of them has an inverse modulo p, as proved in Theorem . Furthermore,
inverse of each number is congruent to p. This means that the inverses of elements of
A are distinct. So we can divide the elements of A into (p − 3)/2 inverse pairs. Thus

2 · 3 · · · (p − 2) ≡ 1 (mod p).

Multiply 1 and p − 1 to both sides of the above equation and the proof is complete.
Now, we should show that if n is not a prime number, then (n − 1)! 6≡ −1 (mod n).
The theorem is obviously true for n = 3 and n = 4. So assume that n ≥ 5 is a composite
number. We can write n = pq where p and q are integers greater than 1. If p 6= q, then
both p and q appear in (n − 1)!, which means (n − 1)! ≡ 0 (mod n). In case p = q, we
have n = p2 . Note that n > 2p and so both p and 2p appear in (n − 1)!, which again
yields to (n − 1)! ≡ 0 (mod n).

Corollary 2.6.3. For a prime p, (p − 2)! ≡ 1 (mod p).

Problem 2.6.4. What is the remainder of 24! when divided by 29?

94 2.7. EULER’S AND FERMAT’S THEOREM

Solution. By Wilson’s theorem, 28! ≡ −1 (mod 29). Also,

−1 ≡ 28! ≡ 24! · 25 · 26 · 27 · 28
≡ 24! · (−4) · (−3) · (−2) · (−1)
≡ 24! · 24
≡ 24! · (−5) (mod 29).

The last congruence equation can be written as 24! · 5 ≡ 1 (mod 29). In other words,
24! is the modular inverse of 5 modulo 29. The problem now reduces to finding the
inverse of 5 modulo 29, which is 6.
Problem 2.6.5. Let n be a positive integer such that
1 1 1 1 n
1+ + + + ··· + = .
2 3 4 23 23!
Find the remainder of n modulo 13.
Solution. Multiply both sides by 23!. The right side will be n and the left side includes
23 terms all of which are divisible by 13 except 23!
13
. Thus, we must find the remainder
of this term modulo 13. Note that
23!
= 12! · 14 · 15 · · · 23
13
≡ −1 · 1 · 2 · · · 10
≡ −10! (mod 13).

This means that we only need to find the remainder of −10! modulo 13. We will use
the same trick as in the previous problem:

−1 ≡ 12! ≡ 10! · 11 · 12
≡ 10! · (−2) · (−1)
≡ 10! · 2 (mod 13).

Rewriting the last equation, we find that 2 is the modular inverse of −10! modulo 13.
Therefore, the answer is the modular inverse of 2 mod 13, which is 7.
Try the next problem yourself!
Problem 2.6.6. Let p be a prime such that p ≡ 1 (mod 4). Prove that
  2
p−1
! ≡ −1 (mod p).
2

2.7 Euler’s and Fermat’s Theorem

Find the remainder of 20162016 when divided by 2017. You might think we are joking
but we are not. Without explicitly calculating this BIG integer, number theorists will
CHAPTER 2. MODULAR ARITHMETIC 95

tell you the remainder is 1. They will even consider this trivial. This demonstrates
another important aspect of numbers. If you know how numbers dance, you know
numbers. Anyway, as you may have already guessed, there is a theorem for it. But
we want to focus on the intuition part. What could lead you to find this remainder
without actually calculating it? If you have been attentive so far, you should already
understand that you should focus on finding a 1 when you are multiplying (obviously,
since we want to reduce the work we do!). We can try this in couple of ways, but let
us start with the most obvious one. Even before that, have you noticed that 2017 is
a prime? This may not be of much importance right now, but keep going. Instead of
such big numbers, try a smaller example first. Find 66 modulo 7

61 ≡ −1 (mod 7),
62 ≡ 1 (mod 7).

We already have 1. Is it clear to you that we will reach 1 in 64 and 66 as well? If

not, just calculate them by hand and see if this is true or false. We will come back to
this topic later. But it seems we have the result 1. Now, do this for 44 (mod 5) and
1010 (mod 11). After you have done all the work, you should realize, like in Wilson’s
theorem, we are getting 1 again. This should encourage you to experiment with some
further values such as 24 (mod 5), 36 (mod 7) etc. Surprisingly, the result is always 1
when the exponent is 1. Pierre De Fermat was the first one to observe and propose
this.

Theorem 2.7.1 (Fermat’s Little Theorem). If p is a prime and a is a positive

integer such that a⊥p. Then

ap−1 ≡ 1 (mod p).

Again, let’s see an example. Take a = 3 and p = 7. And consider the numbers
3 · 1, 3 · 2, 3 · 3, 3 · 4, 3 · 5, 3 · 6 modulo 7. They are respectively 3, 6, 2, 5, 1, 4. Notice
anything? It’s just a rearrangement of 1, 2, . . . , 6. In fact we already proved it before!
Now we will just multiply them all to get

3·1×3·2×3·3×3·4×3·5×3·6≡1·2·3·4·5·6 (mod 7).

Collecting all 3’s in the left-hand side, we will have

36 × 1 · 2 · 3 · 4 · 5 · 6 ≡ 1 · 2 · 3 · 4 · 5 · 6 (mod 7).

Since 1, 2, . . . , 6 are all co-prime to 7, we can divide both sides of the above equation
by 1 · 2 · · · 6 to obtain

36 ≡ 1 (mod 7).

It is now clear that the same argument works for the general case. This PDF was created
for Amazon.

9
Fermat proposed (but did not prove) another theorem in number theory which is much more
difficult than this one. So they call this theorem the “little" one. The other theorem is called Fermat’s
Last Theorem.
96 2.7. EULER’S AND FERMAT’S THEOREM

Proof. From the Definition , it is clear that the set A = {0, 1, . . . , p − 1} is a

complete residue system modulo p. We know that a⊥p, so from Proposition the
set A = {0 · a, 1 · a, . . . , (p − 1) · a} is also a complete residue system modulo p. Putting
′

aside the first element, 0, it is clear that the product of the elements of A and A′ are
congruent modulo p:

1 × a · 2 × a · · · (p − 1) × a ≡ 1 · 2 · · · (p − 1) (mod p)y

So, we find that

(2.2) ap−1 · (p − 1)! ≡ (p − 1)! (mod p).

In congruence equation ( ), we can use the fact that (p, (p − 1)!) = 1 to divide both
sides by (p − 1)! and obtain ap−1 ≡ 1 (mod p).
You can find some other proofs for Fermat’s Little Theorem which use either num-
ber theoretic techniques or even combinatorial approaches. For a very strange yet
interesting proof of this theorem, read the proof by counting necklaces in [ ].
Corollary 2.7.2. If p is a prime and a is an arbitrary positive integer (not necessarily
co-prime with p), then

ap ≡ a (mod p).

Fermat’s little theorem comes handy in so many situations, but it only handles
prime numbers. So we present the Euler’s theorem which is a more general form of
Fermat’s little theorem. The proof is similar as well. Let’s take the following example.
Problem 2.7.3. Show that 420 + 640 + 1260 is divisible by 13.
Solution. Obviously, 1260 ≡ (−1)60 ≡ 1 (mod 13). By Fermat’s little theorem, 612 ≡ 1
(mod 13), hence

3
640 ≡ 636 · 64 ≡ 612 · 64 ≡ 64 ≡ 9 (mod 13).

We can apply the same method to find 420 ≡ 3 (mod 13). Finally,

420 + 640 + 1260 ≡ 3 + 9 + 1 ≡ 0 (mod 13).

Problem 2.7.4 (Taken from [ ]). For integers a, b, prove that ap b − abp is divisible
by p.
Solution. This problem is kind of a direct consequence of Fermat’s little theorem. Write
ap b − abp = ab(ap−1 − bp−1 ). If one of a or b is divisible by p, we are done. If neither of
them is divisible by p,
ap−1 ≡ 1 ≡ bp−1 (mod p).
Thus, p divides ap−1 − bp−1 .
Problem 2.7.5 (Problem 124 in [ ]). Prove that there exist infinitely many com-
posite numbers of the form (22n + 1)2 + 4, where n is a positive integer.
CHAPTER 2. MODULAR ARITHMETIC 97

Solution. A common approach for this kind of problems is to take different moduli (the
first ones would be primes, obviously). Here, we will make use of modulo 29. We will
show that for any n of the form 28k + 1 (for k ≥ 1), the number (22n + 1)2 + 4 will be
divisible by 29. By Fermat’s theorem, 22·28k ≡ 1 (mod 29). Therefore, for n = 28k + 1,

(22n + 1)2 + 4 ≡ (22·(28k+1) + 1)2 + 4

≡ (22 + 1)2 + 4
≡ 0 (mod 29).

Theorem 2.7.6 (Euler’s Theorem). If a and n are positive integers such that
a⊥n. Then

aϕ(n) ≡ 1 (mod n),

where ϕ(.) is the Euler’s totient function.

Proof. Before we start the proof, remember Definition where we defined Euler’s
totient function. The proof is very similar to the proof of Fermat’s theorem and you only
need to apply Proposition once. Let A = {a1 , a2 , . . . , aϕ(m) } be a reduced residue
set mod m. Then so is B = {aa1 , aa2 , . . . , aaϕ(m) }. From the definition of reduced
systems, any number which is coprime to m is congruent to exactly one element of A
and exactly one element of B. Thus, the product of all elements of A must be congruent
to that of B, modulo B. Therefore

(aa1 )(aa2 ) · · · (aaϕ(m) ) ≡ a1 a2 · · · aφ(m) (mod m)

=⇒ aϕ(m) · a1 a2 · · · aϕ(m) ≡ a1 a2 · · · aϕ(m) (mod m)
(2.3) =⇒ aϕ(m) ≡ 1 (mod m).

Note that in equation ( ) we have used the fact that ai ⊥ m for 1 ≤ i ≤ ϕ(m), which
results in a1 a2 · · · aϕ(m) ⊥ m.

We can easily conclude the Fermat’s little theorem from Euler’s theorem: for a
prime p, we have ϕ(p) = p − 1 and so aϕ(p) ≡ ap−1 ≡ 1 (mod p) for any integer a not
divisible by p.

Problem 2.7.7 (Problem 68 in [ ]). Let a and b be positive integers. Prove that
in the arithmetic progression ak + b (for k ≥ 0 an integer), there exist infinitely many
terms with the same prime divisors.

Solution. Let d = (a, a + b). There exist positive integers a1 and c such that

a = da1 and a + b = dc.

Also, (a1 , c) = 1 and c > 1 (why?). Using Euler’s theorem, one can write cnϕ(a1 ) ≡ 1
(mod a1 ) for any positive integer n. This means that there exists a positive integer tn
10
Sometimes called Fermat-Euler theorem or Euler’s totient theorem, proposed by Euler in 1763.
98 2.8. QUADRATIC RESIDUES

such that cnϕ(a1 ) − 1 = tn a1 . Now,

a(ctn + 1) + b = da1 (ctn + 1) + (dc − da1 )

= dc(tn a1 + 1)


= dc cnϕ(a1 )
= dcnϕ(a1 )+1 .

Therefore, the only prime divisors of the term a(ctn + 1) + b in the progression are
prime divisors of dc, which are fixed (because d and c depend only on a and b, which
are fixed). This means that there exist infinitely many terms in the sequence which
have the same prime divisors and we are done.

Here is an exercise for you.

2
Problem 2.7.8 (Taken from [ ]). Find all primes p such that p2 divides 5p + 1.

2.8 Quadratic Residues

Let n be a fixed positive integer. There are many cases when we are interested in integers
a coprime to n for which there exists another integer x such that a ≡ x2 (mod n). As
an example, assume that we want to solve the quadratic congruence relation

ax2 + bx + c ≡ 0 (mod n)

for x. Multiply both sides of the above relation by 4a to obtain

4a2 x2 + 4abx + c ≡ 0 (mod n).

Rewriting the left side of the last relation as (2ax + b)2 − b2 + c, we have

(2ax + b)2 ≡ b2 − c (mod n),

which is of the form y 2 ≡ z (mod n). Therefore, solving any quadratic congruence rela-
tion is equivalent to solving x2 ≡ a (mod n) for some a. We call such an a a quadratic
residue modulo n. Quadratic residues play an important role in cryptography. They
are even used in acoustical engineering.
In this section, we will discuss different aspects of quadratic residues in number
theory.

Definition 2.8.1 (Quadratic Residue). Let m > 1 be a positive integer and let a
be an integer such that (a, n) = 1. Then a is a quadratic residue of n if there exists an
integer x such that
x2 ≡ a (mod n).
If there is no such x, then a is a quadratic non-residue of n.
CHAPTER 2. MODULAR ARITHMETIC 99

Example. 2 is a quadratic residue modulo 7 because 32 ≡ 2 (mod 7). However, 3 is a

non-residue modulo 7. In fact, modulo 7,

12 ≡ 1, 22 ≡ 4,
32 ≡ 2, 42 ≡ 2,
52 ≡ 4, 62 ≡ 1.

This means that the only quadratic residues modulo 7 are 1, 2, and 4.

Corollary 2.8.2. If a ≡ b (mod n), then a is a quadratic residue (non-residue) modulo

n, then so is b.

Note. Whenever we say that an integer a is a quadratic residue (non-residue) modulo

n, it is clear that a + kn is also a quadratic residue(non-residue) modulo n. Therefore,
in order to find which numbers are quadratic residues modulo n, we only need to check
the numbers 1, 2, . . . , n − 1. Obviously, a = 0 is a quadratic residue modulo any n, and
we omit this case in our calculations.
p−1
Theorem 2.8.3. Let p be an odd prime number. There are exactly quadratic
2
residues modulo p (excluding zero). Furthermore, the residues come from the numbers
 2
2 2 p−1
1 ,2 ,..., .
2
Proof. Clearly, the quadratic residues modulo p are

12 , 22 , . . . , (p − 1)2 (mod p).

Note that x2 ≡ (p − x)2 (mod p) for x = 1, 2, . . . , p − 1. So we only need to go halfway,

i.e., we should only consider the numbers
 2
2 2 p−1
1 ,2 ,..., (mod p).
2

These numbers are distinct modulo p, because otherwise if x2 ≡ y 2 (mod p) for some
x, y ∈ {1, 2, . . . , p−1
2
}, then

p|x2 − y 2 =⇒ p|(x − y)(x + y).

But note that x + y < p−1

2
+ p−1
2
= p − 1, and so p ∤ x + y, which means p|x − y. Finally,
since x and y are less than p, we should have x = y. So we have proved that there are
exactly (p − 1)/2 quadratic residues and they are
 2
2 2 p−1
1 ,2 ,..., .
2

Theorem 2.8.4. Let p be an odd prime. Then,

(i) the product of two quadratic residues is also a quadratic residue,

100 2.8. QUADRATIC RESIDUES

(ii) the product of two quadratic non-residues is also a quadratic residue, and

(iii) the product of a quadratic residue and a quadratic non-residue is a quadratic

non-residue.

Proof. The first one is obvious. If a ≡ x2 (mod p) and b ≡ y 2 (mod p), then ab ≡ (xy)2
(mod p). Let’s prove (iii) now. Assume that a ≡ x2 is a residue and b is a non-residue
modulo p and suppose to the contrary that ab is a residue and ab ≡ y 2 (mod p). Then

ab ≡ x2 b ≡ y 2 (mod p).

Note that since (x2 , p) = (a, p) = 1, the multiplicative inverse of x2 exists. Therefore

b ≡ (x2 )−1 · y 2 ≡ (x−1 )2 · y 2 ≡ (x−1 · y)2 (mod p),

which contradicts the assumption that b is a non-residue. So ab is a non-residue. In

order to prove (ii), we use the fact that if a is an integer coprime to p, then

{a, 2a, . . . , (p − 1)a} = {1, 2, . . . , p − 1}.

(The proof is easy, try it yourself). From Theorem , we see that there are exactly
(p − 1)/2 quadratic residues among {a, 2a, . . . , (p − 1)a}. Assume that a is a fixed
quadratic non-residue modulo p. From the proof of (iii), we can say that whenever
p−1
a is multiplied by one of quadratic residues of the set {1, 2, . . . , p − 1}, the
2
result is a non-residue. Therefore, each non-residue element of {a, 2a, . . . , (p − 1)a} is
multiplication of a by a residue in the same set. This means that the multiplication of
a by any non-residue element of the set {a, 2a, . . . , (p − 1)a} is a residue, and we are
done.

Theorem gives us a nice result. Quadratic residues and quadratic non-residues

act just like 1 and −1. How? Notice that 1×1 = 1, (−1)×1 = −1, and (−1)×(−1) = 1,
and this guides us to a point that quadratic residues behave like 1, and quadratic non-
residues behave like −1. We can represent this result using Legendre’s notation.
 
Definition 2.8.5 (Legendre Symbol). We call ap the Legendre symbol for a prime
p. It is defined by:

   0, if p|a,
a
= 1, if a is a qr of p,
p 

−1, otherwise.

Using this notation, Theorem becomes

Theorem 2.8.6. Let p be an odd prime and let a, b be two integers. Then
    
ab a b
= .
p p p
CHAPTER 2. MODULAR ARITHMETIC 101

Remark. Clearly, the same relation holds for the product of any n integers, that is,
      
a1 a2 · · · an a1 a2 an
= ··· .
p p p p

Example. Theorem ( ) is helpful specially when dealing with big numbers. For
instance, let’s see if 18 is a quadratic residue modulo 73. According to the theorem,
     
18 3 3 2
= .
73 73 73 73
3


Note that we don’t need to calculate 73 because whatever it is (−1 or 1), it has
appeared twice and 12 = (−1)2 = 1. So
   
18 2
= .
73 73
2


Now, how can we find 73 ? It would be a pain to check all the values 1, 2, . . . , 73−1
2
to
see if square of any of them is equal to 2 modulo 73. There are two ways to trick this.
One is to use the general formula for p2 which will be discussed next. The second (and
better) idea is to construct the solution. Assume that you have a prime p and an integer
a coprime to p. We know from Corollary that if a is a residue (non-residue), then
a + kp is also a residue (non-residue). So we will add multiples of p to a and check if
we have reached a perfect square. If yes, then a is a residue. Otherwise, we factorize
the new number into perfect squares times some other number b. Continue this process
until you reach either a non-residue b (which means a was a non-residue) or a perfect
square factorization (which means a was a residue). This whole process might seem a
little confusing to you, but applying it to our case (a = 2, p = 73) will make it clear:

2 ≡ 75 ≡ 148 ≡ 22 · 37
≡ 22 · (37 + 73) ≡ 22 · 110
≡ 22 · (110 + 73) ≡ 22 · 183
≡ 22 · (183 + 73) ≡ 22 · 256
≡ 22 · 162
≡ 322 .

So 2 is a quadratic residue modulo 73 and finally

   
18 2
= = 1.
73 73

2.8.1 Euler’s Criterion

 
In the last example of previous section, we provided a method for computing ap . How-
ever, this method only works when p and a are small enough to make the calculations.
Fortunately, Euler developed a criteria to find out whether an integer is a quadratic
102 2.8. QUADRATIC RESIDUES

residue or a quadratic non-residue. After we explain and prove

  Euler’s
 criterion,
 we will
2 −1
explore some special cases, e.g., we will find the value of p and p for all primes
p.
Theorem 2.8.7 (Euler’s Criterion). Let p be an odd prime and let a be an integer
coprime to p. Then
 
p−1 a
a 2 ≡ (mod p).
p
Proof. First notice that from Fermat’s theorem, ap−1 ≡ 1 (mod p). Since p − 1 is even,
we can write this as
 p−1   p−1 
ap−1 − 1 = a 2 − 1 a 2 + 1 ≡ 0 (mod p).
p−1 p−1
So either a 
2 ≡ 1 or a 2 ≡ −1 (mod p). Assume that a is a quadratic residue modulo
 p−1
p. That is, ap = 1. We should prove that a 2 ≡ 1. Since a is a residue, there exists
some integer x for which a ≡ x2 (mod p). So, from Fermat’s theorem,
p−1
a 2 ≡ xp−1 ≡ 1 (mod p).
Now assume the case where a is a quadratic non-residue modulo p. Then we should
p−1
prove that a 2 ≡ −1. We will use an interesting approach here. Let b ∈ {1, 2, . . . , p −
1}. Since (b, p) = 1, the congruence equation bx ≡ a has a unique solution x ≡
a · b−1 (mod p). Also, x 6≡ b (mod p) because otherwise b2 ≡ a (mod p) which is in
contradiction with a being a non-residue. This means that the set {1, 2, . . . , p − 1} can
p−1
be divided into pairs (b, x) such that bx ≡ a (mod p). So
2
p−1
(p − 1)! = 1 × 2 × · · · × (p − 1) ≡ a × a × · · · × a ≡ a 2 (mod p).
| {z }
p−1
2
times

p−1
By Wilson’s theorem, (p − 1)! ≡ −1 (mod p) and therefore a 2 ≡ −1 (mod p), as
desired. The proof is complete.
 
Let’s find ap for some small values of a.
 
Problem 2.8.8. What is −1 p
for a prime p?

Solution. This is a simple case. By Euler’s criterion, we get

 
p−1 −1
(−1) 2 ≡ (mod p).
p
p−1
If p ≡ 1 (mod 4), that is,  = 4k + 1, then p−1
if p  2
is even and so (−1) 2 = 1. On the
−1
other side, we know that p is either 1 or −1. So in this case it must equal one.
This means that
 
−1
p ≡ 1 (mod 4) ⇐⇒ = 1.
p
CHAPTER 2. MODULAR ARITHMETIC 103
p−1
Note that the only if part of the above statement is true because if (−1) 2 = 1, then
p−1
2
= 2k for some integer k. So p = 4k + 1, or p ≡ 1 (mod 4). You can easily check
that the following statement is also true
 
−1
p ≡ 3 (mod 4) ⇐⇒ = −1.
p
Each prime has either the form p ≡ 1 (mod 4) or p ≡ 3 (mod 4), so these primes
together make all primes. All in all,
Theorem 2.8.9. For all primes p,
  (
−1 1, if p ≡ 1 (mod 4) or p = 2,
=
p −1, if p ≡ 3 (mod 4).

This infers the following theorem and the next.

Theorem 2.8.10. −1 is a quadratic residue of a prime p if and only if p ≡ 1 (mod 4).
Theorem 2.8.11. Let a and b be coprime positive integers. Then every prime divisor
of a2 + b2 is either 2 or of the form 4k + 1.
Proof. Let p be a prime divisor of a2 + b2 . If a and b both are odd then p can be 2.
Now assume p is larger than 2. Then

p−1
p−1
a2 ≡ −b2 (mod p) =⇒ a2 2 ≡ −b2 2 (mod p).
p−1
If p is of the form 4k + 3, then 2
= 2k + 1 is odd, hence,

p−1
p−1
a2 2
≡ −b2 2
(mod p) =⇒ ap−1 ≡ −bp−1 (mod p).

Clearly, since p|a2 +b2 , if p divides one of a or b, it should divide the other one. But this is
impossible because a⊥b. So p⊥a and p⊥b. By Fermat’s little theorem, ap−1 ≡ bp−1 ≡ 1
(mod p), which is in contradiction with the above equation since p is odd. So, p cannot
be of the form 4k + 3 and therefore every odd prime divisor of a2 + b2 is of the form
4k + 1.
Note. We could just say,

a2 ≡ −b2 (mod p) =⇒ (ab−1 )2 ≡ −1 (mod p),

which means that −1 is a quadratic residue of p, and therefore p ≡ 1 (mod 4).

We get the following corollary, which can directly solve an IMO problem.
Corollary 2.8.12. Let k be a positive integer. Every divisor of 4k 2 + 1 is of the form
4n + 1 for some integer n.
Proof. Thanks to the previous theorem, we know that all the prime divisors of 4k 2 + 1
are of the form 4t + 1. Every divisor of 4k 2 + 1 is a multiplication of its prime divisors.
And if we multiply two numbers of the form 4t+1, then the number is again of the same
form (multiply 4x + 1 and 4y + 1 and see the result yourself). Hence, conclusion.
104 2.8. QUADRATIC RESIDUES

Problem 2.8.13 (Iran, Third Round Olympiad, 2007). Can 4xy − x − y be a

square for integers x and y?

Solution. Assume that 4xy − x − y = t2 . We can rearrange it as x(4y − 1) = t2 + y, so

4y − 1 | t2 + y
=⇒ 4y − 1 | 4t2 + 4y
=⇒ 4y − 1 | 4t2 + 4y − (4y − 1)
=⇒ 4y − 1 | 4t2 + 1

Since 4y − 1 is of the form 4k + 3, it must have at least one prime factor of the form
4j + 3. Then we have a prime factor of 4t2 + 1 which is of this form, a contradiction.
Thus, it can’t be a square.

Back to quadratic residues, the next step is to determine if 2 is a quadratic residue

modulo a prime. Unfortunately, we cannot apply Euler’s criterion directly in this case
p−1
because we do not know what 2 2 (mod p) would be for different values of p. So our
p−1
challenge is to find 2 2 modulo p.
 
Problem 2.8.14. What is p2 for a prime p?

p−1
Solution. As explained just above, we only need to find a way to calculate 2 2
(mod p). The idea is similar to what we did in the proof of Fermat’s little theorem.
Remember that in Fermat’s theorem, we needed to somehow construct ap−1 and we did
that by multiplying elements of the set {a, 2a, . . . , (p − 1)a}. Now how can we construct
p−1
2 2 ? The idea is to find a set with p−1
2
elements such that the product of all elements
p−1
has the factor 2 2 . One possibility is to consider the set A = {2, 4, . . . , p − 1}. Then
the product of elements of A is
p−1 p−1
2 × 4 × · · · × (p − 1) = 2 2 × 1 × 2 × ··· ×
  2
p−1 p−1
(2.4) =2 2 × !.
2


In order to get rid of the term p−1
2
!, we have to compute the product of elements of
A in some other way. Notice that we are
looking for 2 × 4 × · · · × (p − 1) and we want
p−1
to make it as close as possible to 2 ! so that we can cancel out this term and find
p−1
the value of 2 2 (mod p). To construct this factorial, we need all the numbers in the
set  
p−1
B = 1, 2, . . . , .
2
However, we only have even integers in A. For example, take p = 11. Then A =
{2, 4, 6, 8, 10} and B = {1, 2, 3, 4, 5}. Now, we want to construct 5! using the product
of elements of A. Clearly, the elements 2 and 4 are directly chosen from A. Now it
remains to somehow construct the product 1 × 3 × 5 with the elements 6, 8, 10. The
CHAPTER 2. MODULAR ARITHMETIC 105

trick is pretty simple: just notice that 10 ≡ −1, 8 ≡ −3, and 6 ≡ −5 (mod 11). This
means that 6 × 8 × 10 ≡ (−1)3 · 1 × 3 × 5, and so

(2.5) 2 × 4 × 6 × 8 × 10 ≡ (−1)3 · 5!.

Comparing equations ( ) (for p = 11) and ( ), we find that

25 · 5! ≡ (−1)3 · 5! (mod 11) =⇒ 25 ≡ (−1)3 ≡ −1 (mod 11).

Let’s go back to the solution of the general problem. As in the example of p = 11, we
are searching for the power of (−1) appeared in the congruence relation. In fact, this
power equals the number of even elements bigger than p−1 2
and less than or equal to
p − 1. Depending on the remainder of p modulo 8, this power of (−1) can be even or
odd. Consider the case when p ≡ 1 (mod 8). Then p − 1 = 8k for some positive integer
k and so the even numbers less than or equal to p−1 2
= 4k in the set A = {2, 4, . . . , 8k}
are 2, 4, . . . , 4k, which are 2k + 1 numbers. Therefore
2k+1 items 2k items
z }| { z }| {
2 × 4 × · · · × (p − 1) = (2 × 4 × · · · × 4k) · ((4k + 2) × (4k + 4) × · · · × 8k)
≡ (2 × 4 × · · · × 4k) · ((−(4k − 1)) × (−(4k − 3)) × · · · × (−1))
≡ (−1)2k · (4k)!
 
2k p−1
≡ (−1) · ! (mod p).
2
Compare this result with ( ), you see that
   
p−1 p−1 2k p−1 p−1
2 2 × ! ≡ (−1) · ! (mod p) =⇒ 2 2 ≡ (−1)2k ≡ 1 (mod p).
2 2
 
So if p ≡ 1 (mod 8), then p2 = 1.
The process is similar for p ≡ 3, 5, 7 (mod 8) and we put it as an exercise for the
reader.  
After all this work, we are finally done computing p2 . The final result is stated in
the following theorem.
Theorem 2.8.15.
  (
2 1, if p ≡ 1 or 7 (mod 8),
=
p −1, if p ≡ 3 or 5 (mod 8).
 
a
We can generalize the method used in the solution of Problem to find p
for all integers a and primes p.
Theorem 2.8.16 (Gauss’ Criterion). Let p be a prime number and let a be an
integer coprime to p. Let µ(a, p) denote the number of integers x among
p−1
a, 2a, . . . , a
2
106 2.8. QUADRATIC RESIDUES

such that x > p/2 (mod p). Then

 
a
= (−1)µ(a,p) .
p
The proof is left as an exercise for the reader.
Theorem 2.8.17. The smallest quadratic non-residue of a prime p is a prime less than
√
p + 1.
Proof. Let r be the smallest quadratic non-residue of prime p. Then, any i < r is a
quadratic residue of p. If r = kl for k, l > 1, using Legendre’s symbol, we have
     
r k l
= · ,
p p p
which gives −1 = 1 · 1 (since r is a quadratic non-residue modulo p) and we get a
contradiction. So, r can’t be a prime. Let’s move on to the next part of the theorem.
We have that r, · · · , (r − 1)r are quadratic non-residues mosulo p. If any of them is
greater than p, say ri, we have r(i − 1) < p < ri for some i. But then,

ri < p + r,

so that ri = p + s with s < r. Thus,

ri = p + s
≡ s (mod p).

Since s < r, s is a quadratic residue of p, which in turn means ri is a quadratic

√
residue of p as well. Another contradiction, so r(r − 1) < p. If r ≥ p + 1, we have
√ √ √
r(r − 1) ≥ p( p + 1) = p + p > p, yet another contradiction. The claim is therefore
true.

2.8.2 Quadratic Reciprocity

 
Assume we want to compute ap for some integer a and a prime p. The remark after
Theorem says that it’s enough to find
     
a1 a2 an
, ,..., ,
p p p
 
where a1 , a2 , . . . , an are divisors of a. This means that if we know the value of pq for
 
a prime q, we can find the values of ap for any a.
 
So let’s discuss on the value of pq . If q is a big prime number, then by Corollary
 
q
, we can reduce q modulo p until we reach some c < p and find by some
p
method (Euler’s or Gauss’s criteria). So we can handle the case when q is a big prime.
CHAPTER 2. MODULAR ARITHMETIC 107
 
Now, what about the case when p is a big prime? In this case, finding pq would
be very hard with theorems and methods stated by now. There is a very nice property
of prime numbers which helps us to handle
  this   This property is called the Law
case.
q p
of Quadratic Reciprocity which relates p and q . So, in case q is big, we can first
   
calculate pq and then use this law to find pq .

Theorem 2.8.18 (Law of Quadratic Reciprocity). Let p and q be different odd

primes. Then
  
p q p−1 q−1
= (−1) 2 · 2 .
q p
11


Example. Let’s find 6661 . We have
  
11 6661
(2.6) = (−1)5·3330 = −1.
6661 11
Now, since 6661 ≡ 6 (mod 11), we obtain
      
6661 6 3 2
= = = 1 × (−1) = −1.
11 11 11 11
Replacing in equation ( ), we finally find
 
11
= 1.
6661
The proof of this theorem is a bit complicated and it would make you lose the
continuity of the context. For this reason, we will provide the proof in section ( ).

2.8.3 Jacobi Symbol

 
In previous sections, whenever we used the Legendre symbol pq , we needed p to be a
prime number. We are now interested in cases where p can be a composite number. In
1837, Jacobi generalized the symbol used by Legendre in this way:
Definition 2.8.19 (Jacobi Symbol). Let a be an integer and n = pα1 1 pα2 2 · · · pαk k ,
where pi are odd primes and αi are non-negative integers (1 ≤ i ≤ k). The Jacobi
symbol is defined as the product of the Legendre symbols corresponding to the prime
factors of n:
 a   a α 1  a α 2  α k
a
= ··· .
n p1 p2 pk


Also, we define a1 to be 1.
Remark.

The immediate result of the above definition is that if gcd(a, n) = 1, then
a
n
is either +1 or −1. Otherwise, it equals zero.
108 2.8. QUADRATIC RESIDUES

Example.
       2
14 14 14 14
=
2535 3 5 13
     2
2 4 1
=
3 5 13
= (−1) · 1 · 1
= −1.

Example.
    
2 2 2
= = (−1) · (−1) = 1.
15 3 5


Note. If na = −1 for some a and n, then a is a quadratic non-residue modulo n.
However, the converse is not necessarily true. The above
example shows you a simple
case when a is a quadratic non-residue modulo n but na = 1.

Theorem 2.8.20. Let a be an integer and n = pα1 1 pα2 2 · · · pαk k , where pi are odd primes
and αi are non-negative integers (1 ≤ i ≤ k). Then a is a quadratic residue modulo n
if and only if a is a quadratic residue modulo every pαi i (1 ≤ i ≤ k).

Proof. The if part is easy to prove. Assume that a is a quadratic residue modulo n
and a ≡ x2 (mod n) for some integer x. Then a ≡ x2 (mod pαi i ) since pi s are relatively
prime to each other. Now the only if part: assume that a ≡ x2i (mod pαi i ) for all i,
where xi are integers. According to Chinese Remainder Theorem, since the numbers
pαi i are pairwise coprime, the system of congruence equations

x ≡ x1 (mod pα1 1 )
x ≡ x2 (mod pα2 2 )
..
.
x ≡ xk (mod pαk k )

has a solution for x. Now x2 ≡ x2i ≡ a (mod pαi i ), and therefore x2 ≡ a (mod n), which
means that a is a quadratic residue modulo n.
The following theorem sums up almost everything explained in quadratic residues.
The proofs are simple and straightforward, so we leave them as exercises for the reader.

Theorem 2.8.21. Let a and b be any two integers. Then for every two odd integers
m and n, we have
    
ab a b
i. = ,
n n n
   
a + bn a
ii. = ,
n n
CHAPTER 2. MODULAR ARITHMETIC 109
m  n  m−1 n−1
iii. = (−1) 2 · 2 ,
n m
 a  a a 
iv. = ,
mn n m
 
−1 n−1
v. = (−1) 2 ,
n
 
2 n2 −1
vi. = (−1) 8 .
n
Theorem 2.8.22. If a positive integer is a quadratic residue modulo every prime, then
it is a perfect square.

This is a really cool theorem. However, proving this might be challenging! Give
it a try.

Theorem 2.8.23. 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.

Proof. Assume that b has a prime factor, p, so that pxn+r ||b with 0 < r < n. Then,
we can let k = pxn+n . It follows that b ≡ ank mod pxn+n . Since pxn+r ||b, we see that
r
pxn+r ||ank . Then, px+ n ||ank , which is a contradiction since nr is not an integer. Thus, we
have a contradiction, so r = 0, which means that only nth powers of primes fully divide
b, so b is an nth power.

2.9 Wolstenholme’s Theorem

The purpose of this section is to discuss the sum
p−1
X 1 1 1 1
=1+ + + ··· + .
k=1
k 2 3 p−1

Well, not exactly. We are more interested in this sum modulo p where p is a prime.
But how do we calculate fractions modulo p? The answer should be obvious by now. ab
(mod p) is actually ab−1 (mod p) where b−1 (mod p) is the inverse of b modulo p. So
if be ≡ 1 (mod p), then
a
≡ ae (mod p).
b
However, from modular cancellation property, we can take fractions modulo p if p ∤ b.
Moreover, we can introduce some sort of divisibility here.
Let ab be a fraction and let n be an integer such that (n, b) = 1. If a is divisible
by n, we say that ab is divisible by n. Following this convention, the congruence ab ≡ 0
(mod n) makes sense.
11
A popular proof for this uses Dirichlet’s theorem: For two co-prime positive integers a and b, there
are infinitely many primes in the sequence {an + b}n≥1 . This theorem is very famous for being difficult
to prove and it is well beyond our scope. How you are going to use it, that is entirely up to you.
110 2.9. WOLSTENHOLME’S THEOREM
100
Example. Since 25 | 100 and (3, 25) = 1, we have 3
≡ 0 (mod 3). We can also
calculate it this way:
100
≡ 100 · (3)−1
3
≡ 100 · 17
≡ 0 (mod 25),
Now, let’s compute a non-zero fraction modulo 7:
840 120
= ≡ 120 · (11)−1
77 11
≡ 120 · 2
≡ 2 (mod 7).
Theorem 2.9.1 (Wolstenholme’s Theorem). Let p > 3 be a prime. Then
p−1
X 1 1 1 1
S= =1+ + + ··· + ≡ 0 (mod p2 ).
k=1
k 2 3 p−1
Pp−1
Note. According to our assumption, the sum k=1 1/k has been written in lowest
terms, that is, as a fraction a/b such that (a, b) = 1.
Remark. Theorem ( ) is not the original theorem stated by Wolstenholme. Actu-
ally the theorem was as stated below.
Theorem 2.9.2. If p is a prime bigger than 3, then
 
2p
≡ 2 (mod p3 ), and
p
 
2p − 1
≡ 1 (mod p3 ).
p−1
This theorem is equivalent to Theorem .
This seems to be a very interesting theorem, however the proof is not straightfor-
ward. Let us tackle this theorem step by step (these steps are really intuitive and very
useful in olympiad problems). But first we will show a weaker version of the Theorem
.
Theorem 2.9.3. For any prime p,
 
2p
≡2 (mod p2 ).
p
Proof. We make use of the identity ( ) to write
    2  2  2  2
2p p p p p
= + + ··· + +
p 0 1 p−1 p
 2  2
p p
=2+ + ··· +
1 p−1
2
≡ 2 (mod p ).
CHAPTER 2. MODULAR ARITHMETIC 111

The last line is true because from Theorem , for 0 < i < p, we have
   2
p p
≡0 (mod p) and ≡0 (mod p2 ).
i i

Lemma 2.9.4. Let p > 3 be a prime and S be defined as in Theorem . Then,

S≡0 (mod p).

Proof. The proof is straightforward. There are p − 1 terms in the sum and since p > 3
is an odd prime, the number of terms is even. So we can write S as sum of pairs of the
1
form k1 + ˚ p−k , for k = 1, 2, . . . , p−1
2
. Thus

1 1 1
S =1+ + + ··· +
2 3 p−1
    !
1 1 1 1 1
= 1+ +˚ + + ··· + p−1 + p−1
p−1 2 p−2 2 2
+1
p−1 p−1 p−1
2   2 2
X 1 1 X (k) + (p − k) X p
= + = =
k=1
k p−k k=1
k(p − k) k=1
k(p − k)
p−1
2
X 1
=p· ≡0 (mod p).
k=1
k(p − k)

a
In the last line of above equations, the sum can be written as (p−1)! , where a is some
integer. Note that (p, (p − 1)!) = 1 and that’s why we can conclude
p−1
2
X 1
p· ≡0 (mod p).
k=1
k(p − k)

Lemma 2.9.5. For a prime p > 3,

(1−1 )2 + (2−1 )2 + · · · + ((p − 1)−1 )2 ≡ 0 (mod p),

where i−1 is the multiplicative inverse of i modulo p for i = 1, 2, . . . , p − 1.

Proof. We recommend you re-read section ( ) if you have forgotten the definition
of multiplicative inverse. We already know that

(p − 1)(p)(2p − 1)
12 + 22 + · · · + (p − 1)2 = .
6
Clearly, the sum is an integer. Therefore (p − 1)(p)(2p − 1) is divisible by 6. Now since
p > 3, we have (p, 6) = 1 and thus p divides (p − 1)(p)(2p − 1)/6. Therefore,

12 + 22 + · · · + (p − 1)2 ≡ 0 (mod p).

112 2.9. WOLSTENHOLME’S THEOREM

In order to prove the lemma we should show that

(1−1 )2 + (2−1 )2 + · · · + ((p − 1)−1 )2 ≡ 12 + 22 + · · · + (p − 1)2 (mod p).

We shall show that the sets A = {1, 2, . . . , p − 1} and B = {1−1 , 2−1 , . . . , (p − 1)−1 } are
equal. A proof is as follows: from Theorem , for any x ∈ A, there exists some
y ∈ B such that xy ≡ 1 (mod p). This y is unique, because if there exists some other
z ∈ B for which xz ≡ 1 (mod p), then xy ≡ xz (mod p), and since (x, p) = 1, we have
y ≡ z (mod p) which means y = z (why?). So there exists a unique y ∈ B for each
x ∈ A, and thus A = B since A and B have equal number of elements. Finally,

(1−1 )2 + (2−1 )2 + · · · + ((p − 1)−1 )2 ≡ 12 + 22 + · · · + (p − 1)2 ≡ 0 (mod p).

We are going to re-state Proposition because, as we already mentioned, it is

very useful:
Lemma 2.9.6. For a prime p ≥ 3 and any positive integer a coprime to p,

(a−1 )n ≡ (an )−1 (mod p),

for all positive integers n.

Proof.

a · a−1 ≡ 1 (mod p) =⇒ an · (a−1 )n ≡ 1 (mod p)

=⇒ (a−1 )n ≡ (an )−1 (mod p).

Lemma 2.9.7. For a prime p > 3,

p−1
2
X 1
≡0 (mod p).
i=1
i(p − i)

Proof. Let’s write the sum as

p−1 p−1 p−1
2 2 (p−1)! 2
X 1 X i(p−i) 1 X (p − 1)!
= = · .
i=1
i(p − i) i=1
(p − 1)! (p − 1)! i=1 i(p − i)

Since (p − 1)! is relatively prime to p, we only need to show that

p−1
2
X (p − 1)!
≡0 (mod p).
i=1
i(p − i)

Define
(p − 1)! p−1
ai = for i = 1, 2, . . . , .
i(p − i) 2
CHAPTER 2. MODULAR ARITHMETIC 113

From Wilson’s theorem, we know that (p − 1)! ≡ −1 (mod p). Observe that
i · (p − i) · ai = (p − 1)! ≡ −1 (mod p).
Replacing p − i ≡ −i (mod p) in the above equation, we have
(2.7) −i2 · ai = −1 (mod p) =⇒ i2 · ai ≡ 1 (mod p).
Notice that the above equations are true for i = 1, 2, . . . , (p − 1)/2. Now, ( ) means
that ai is the multiplicative inverse of i2 modulo p. So we have proved that
(p − 1)! p−1
(2.8) ai = ≡ (i2 )−1 (mod p) for i = 1, 2, . . . , ,
i(p − i) 2
where (i2 )−1 means the multiplicative inverse of i2 modulo p. We should now prove
(p−1)/2
X
that the sum of all ai s is divisible by p. Let a = ai . According to ( ),
i=1
p−1 p−1
2
X 2
X
a= ai = (i2 )−1 .
i=1 i=1

From Lemma , (i2 )−1 ≡ (i−1 )2 (mod p), and so

p−1
2
X
(2.9) a≡ (i−1 )2 (mod p).
i=1

We want to show that a ≡ 0 (mod p). The trick is to convert ( ) to what we proved
in Lemma , using the fact that −(a−1 ) ≡ (−a)−1 (mod p):
p−1 p−1 p−1 p−1
2
X 2
X 2
X 2
X
2a ≡ a + a ≡ (i−1 )2 + (i−1 )2 ≡ (i−1 )2 + (−(i)−1 )2
i=1 i=1 i=1 i=1
p−1 p−1 p−1 p−1
X2 X2 X2 X2

≡ (i−1 )2 + ((−i)−1 )2 ≡ (i−1 )2 + ((p − i)−1 )2

i=1 i=1 i=1 i=1
p−1
2 p−1 p−1
X X X
−1 2 −1 2
≡ (i ) + (i ) ≡ (i−1 )2 ≡ 0 (mod p).
i=1 i= p+1 i=1
2

Thus a ≡ 0 (mod p) and we are done.

We are ready to prove Wolstenholme’s theorem now.
Proof of Wolstenholme’s Theorem. According to Lemma , we can write S as
p−1
2
X 1
S =p· .
i=1
i(p − i)

From Lemma , we know that the above sum is divisible by p, so S is divisible by

p2 .
114 2.9. WOLSTENHOLME’S THEOREM

Problem 2.9.8. Let p ≥ 5 be a prime number, and

1 1 a
1+ + ··· + = ,
2 p b
where a and b are two coprime integers. Show that p4 |ap − b.
Solution. From Wolstenholme’s theorem, we have
1 1 x
1+ + ··· + = p2 · ,
2 p−1 y
for some integers x and y such that y⊥p. Replacing this in the given equation,
x 1 a x
p2 · + = =⇒ ap − b = p3 b · .
y p b y
Since b is divisible by p, we have p4 |ap − b.
Problem 2.9.9 (Masum Billal). Let p ≥ 5 be a prime and
1 2 p−1 a
+ + ··· + = ,
p−1 p−2 1 b
where a and b are two coprime integers. Show that p3 |a − b + bp.
Solution. Note that
p−1 p−1 p−1 p−1
!
a X p − i X p  X p X 1
= = −1 = − (p − 1) = p · − (p − 1)
b i=1
i i=1
i i=1
i i=1
i
2x
= p · p − (p − 1),
y
where x and y are coprime integers with y⊥p (we have used Wolstenholme’s theorem
in the last line). Now,
a p2 x
+p−1=p· =⇒ (a − b + bp)y = p3 xb,
b y
and since y⊥p, we have p3 |a − b + bp.
Problem 2.9.10. For any prime p and a positive integer k such that 1 ≤ k ≤ p − 1,
prove that
 
p−1
≡ (−1)k (mod p).
k
Solution. We use the fact that p − i ≡ −i (mod p) and that (i, p) = 1 for 0 < i < p.
 
p−1 (p − 1)(p − 2) · · · (p − 1 − k + 1)
=
k 1 × 2 × ··· × k
(−1) × (−2) × · · · × (−k)
≡
1 × 2 × ··· × k
(−1)k · 1 × 2 × · · · × k
≡
1 × 2 × ··· × k
≡ (−1)k (mod p).
CHAPTER 2. MODULAR ARITHMETIC 115

Problem 2.9.11. For an odd prime p, show that

2p − 2 1 1
≡ 1 − + ··· − (mod p).
p 2 p−1

Solution. The approach is not obvious here unless one knows the above theorem. In
problems like this, it is usually hard to pin down how to approach the problem. However,
one should of course try to make use of the fact that
   
p p p p
2 = (1 + 1) = 1 + + ··· + +1
1 p−1
   
p p−1 p p−1
=1+ + ··· + + 1.
1 0 p−1 p−2

So,
      
p 1 p−1 1 p−1 1 p−1
2 −2=p + + ··· + .
1 0 2 1 p−1 p−2

Now, the problem is in a suitable shape and we can use the theorem above to write
     
2p − 2 1 p−1 1 p−1 1 p−1
= + + ··· +
p 1 0 2 1 p−1 p−2
1 1
≡ (−1)0 + (−1)1 + · · · + (−1)p−3
2 p−1
1 1
≡ 1 − + ··· − (mod p).
2 p−1

Corollary 2.9.12. For an odd prime p,

2p−1 − 1 1 1
≡ 1 − + ··· − (mod p).
p 2 p−2

Problem 2.9.13. Let p ≥ 5 be a prime. Prove that

 2
p
≡ p (mod p5 ).
p

Solution. Notice that

 2
p p2 (p2 − 1)(p2 − 2) · · · (p2 − (p − 1))
−p= −p
p p!
p  2 
= (p − 1)(p2 − 2) · · · (p2 − (p − 1)) − (p − 1)! .
(p − 1)!

Since (p, (p − 1)!) = 1, it suffices to show that

(p2 − 1)(p2 − 2) · · · (p2 − (p − 1)) ≡ (p − 1)! (mod p4 ).

116 2.9. WOLSTENHOLME’S THEOREM

Expand the left side to obtain

(p2 − 1)(p2 − 2) · · · (p2 − (p − 1))

1  1 
2
= (p − 1)! + p 1 + + · · · + (p − 1)! + p4 · x,
2 p−1

where x is some integer. By Wolstenholme’s theorem, the second term in the above
expansion is divisible by p4 and we are done.

Corollary 2.9.14. Let p ≥ 5 be a prime and n ≥ 1 be an integer. Then

 n+1 
p
≡ pn (mod p2n+3 ).
p

Problem 2.9.15 (Taken from [ ]). Let k be a non-negative integer and p ≥ 5 be a

prime. Prove that

1 1 1
+ + ··· + ≡0 (mod p2 ).
kp + 1 kp + 2 kp + (p − 1)

Hint. Use the following:

p−1 p−1
X 1 1 X 1 1 
= + .
i=1
kp + i 2 i=1 kp + i kp + p − i

Problem 2.9.16. For a prime p ≥ 5, show that

 3  2
p p
2
≡ (mod p8 ).
p p

Problem 2.9.17 (PEN [ ], D23). Let p be an odd prime of the form p = 4n + 1.

• Show that n is a quadratic residue (mod p).

• Calculate the value nn (mod p).

Problem 2.9.18. Let p ≥ 7 be a prime and let s be a positive integer such that p−1 ∤ s.
Prove that
1 1 1
1+ s
+ s + ··· + ≡0 (mod p).
2 3 (p − 1)s

Problem 2.9.19. Let n be a positive integer not divisible by 6. Also, let S be a reduced
residue system modulo n such that 1 ≤ a < n for all a ∈ S. Prove that
X1
≡0 (mod n2 ).
a∈S
a
CHAPTER 2. MODULAR ARITHMETIC 117

2.10 Lucas’ Theorem

In 2010, Masum encountered the following problem:

Problem 2.10.1. Find the number of odd binomial coefficients in the expansion of
(a + b)2010 .

Try to solve it yourself. If you do not
have any idea how to proceed, then here is a
2010
hint: you need to find the value of i (mod
2) for 0 ≤ i ≤ 2010. One idea for doing
2010
that is to count the exponent of 2 in i using Legendre’s theorem. If look for the
condition when a coefficient can be odd.
Here, we will focus on a generalized version of such problems. In problems like this,

it happens that we need to find the remainder of division of the binomial coefficient m n
by a prime number p. Edouard Lucas found patterns in Pascal triangle which resulted
in the following theorem.

Theorem 2.10.2 (Lucas’s Theorem). Let p be a prime and let m and n be non-
negative integers. Then
  Y k  
m mi
≡ (mod p),
n i=0
ni

where

m = mk pk + mk−1 pk−1 + · · · + m1 p + m0 , and

n = nk pk + nk−1 pk−1 + · · · + n1 p + n0

are
the base p expansions of m and n respectively. This uses the convention that
m
n
= 0 if m < n.

Example. For p = 7, m = 67, and n = 10. Now

67 = 1 · 72 + 2 · 7 + 4, 10 = 0 · 72 + 1 · 7 + 3,

and therefore
     
67 1 2 4
≡ ≡1·2·4≡1 (mod 7).
10 0 1 3


Note that 67
10
= 247, 994, 680, 648 ≡ 1 (mod 7), which is a huge number and it would
be a tedious work to find the remainder modulo 7 without Lucas’s theorem.
In order to prove Lucas’s theorem, we need to state a lemma first.

Lemma 2.10.3. For a prime p, an integer x, and a positive integer r, we have

r r
(1 + x)p ≡ 1 + xp (mod p).
118 2.10. LUCAS’ THEOREM

Proof. We will use induction on r to prove them lemma. The
base case r = 1 is easy:
for any integer k such that 1 ≤ k ≤ p − 1, we know that kp ≡ 0 (mod p). Now
     
p p p 2 p
(1 + x) ≡ 1 + x+ x + ··· + xp−1 + xp
1 2 p−1
≡ 1 + xp (mod p).
r r
Now suppose that (1 + x)p ≡ 1 + xp (mod p) is true for some integer r ≥ 1. Then
r+1 r
p r
p
(1 + x)p ≡ (1 + x)p ≡ 1 + xp
         
p p pr p 2pr p (p−1)pr p pr+1
≡ + x + x + ··· + x + x
0 1 2 p−1 p
r+1
≡ 1 + xp (mod p).

So the congruence relation holds for all r ≥ 1.

Proof of Lucas’s Theorem. The idea is to find the coefficient of xn in the expansion of
(1 + x)m . We have
k +m k−1 +···+m
(1 + x)m = (1 + x)mk p k−1 p 1 p+m0

k k−1
= [(1 + x)p ]mk [(1 + x)p ]mk−1 · · · [(1 + x)p ]m1 (1 + x)m0
k k−1
≡ (1 + xp )mk (1 + xp )mk−1 · · · (1 + xp )m1 (1 + x)m0 (mod p).

We want the coefficient of xn in (1 + x)m . Since n = nk pk + nk−1 pk−1 + · · · + n1 p + n0 ,

k k−1 i
we want the coefficient of (xp )nk (xp )nk−1 · · · (xp )n1 xn0 . The coefficient

of each (xp )ni
i
comes from the binomial expansion of (1 + xp )mi , which is m i
ni
. Therefore we take the
mi


product of all such ni , and thus we have
  Y k  
m mi
≡ (mod p).
n i=0
ni

Corollary 2.10.4. Let s, t, q, r be non-negative integers and p p be a prime such that

0 ≤ q, r ≤ p − 1. Then
    
sp + q s q
≡ (mod p).
tp + r t r

Problem 2.10.5. How many ordered triples (a, b, c) of positive integers satisfy a + b +
c = 94 and 3 does not divide
94!
?
a!b!c!
Solution. Write c = 94 − a − b, and hence
   
94! 94 94 − a
= · .
a!b!(94 − a − b)! a b
CHAPTER 2. MODULAR ARITHMETIC 119
94


By Lucas’ theorem, since 94 = (10111)3 , 3 does not divide a
only when a is an
element of the set
S = {1, 3, 4, 9, 10, 12, 13, 81, 82, 84, 85, 90, 91, 93, 94}.
By symmetry, we only need to find a, b, c which are elements of S. There exist six such
triples (a, b, c) which sum to 94:
(1, 3, 90), (1, 9, 84), (1, 12, 81), (3, 9, 82), (3, 10, 81), (4, 9, 81).
Problem 2.10.6. Let p be a prime. Prove that
 n 
p −1
≡ (−1)sp (k) (mod p),
k
where sp (k) is the sum of digits of k when represented in base p.
Hint. Use Problem and apply Lucas’ theorem.
Problem 2.10.7 (Taken from [ ]). Let p and q be distinct odd primes. Prove that
 
2pq − 1
≡ 1 (mod pq)
pq − 1
if and only if
 
2p − 1
≡1 (mod q), and
p−1
 
2q − 1
≡1 (mod p).
q−1
Solution. Since 2pq − 1 = (2q − 1)p + p − 1, the rightmost digit of 2pq − 1 when
represented in base p is p − 1 and the other digits form 2q − 1. Analogously, the
rightmost digit of pq − 1 when represented in base p is p − 1 and the other digits form
q − 1. Applying corollary ( ), we find
      
2pq − 1 2q − 1 p − 1 2q − 1
(2.10) ≡ ≡ (mod p).
pq − 1 q−1 p−1 q−1
The if part is obvious since p and q are different primes. We will prove the only if part
now.


Suppose that 2pq−1
pq−1
≡ 1 (mod pq). This means that 2pq−1 pq−1
≡ 1 (mod p). By
2q−1

2p−1


equation ( ), q−1 ≡ 1 (mod p) as desired. Proving p−1 ≡ 1 (mod q) is similar.
Problem 2.10.8 (Taken from [ ]). Let n and k be arbitrary positive integers and
let p be an odd prime p. Prove that
   k 
2  pk
p − .
pm m
Hint. Induct on n and equate the coefficients of apm bp(n−m) in both sides of
(a + b)pn = (a + b)p(n−1) (a + b)p .
120 2.11. LAGRANGE’S THEOREM

2.11 Lagrange’s Theorem

Lagrange’s theorem in polynomial congruency is a really influential result in number
theory. It has many implications and applications. And it is so important that we
decided to keep it in this book even though we are not discussing polynomials or poly-
nomial congruency in the current book. We just need the following simple definition.
Definition 2.11.1. A polynomial is an expression consisting of variables and coeffi-
cients which only employs the operations of addition, subtraction, multiplication, and
non-negative integer exponents.
Example. An example of a polynomial of a single variable x and with integer coefficients
is P (x) = x4 + 3x2 + x − 8. An example in three variables and rational coefficients is
4
P (x, y, z) = 2x3 + xy − 7xyz + 3zy 2 − 6.
5
Note. We only work with polynomials of a single variable and with integer coefficients
in this book.
Definition 2.11.2. Consider a polynomial P (x) with integer coefficients. The degree
of P (x) is the largest exponent of x in P (x). That is, if
P (x) = an xn + an−1 xn−1 + · · · + a1 x + a0 ,
where ai , 0 ≤ i ≤ n are integers and an 6= 0, then the degree of P (x) is n. We show
this by deg P (x) = n.
Definition 2.11.3. Let
P (x) = an xn + an−1 xn−1 + · · · + a1 x + a0
be a polynomial with integer coefficients. Assume that at least one coefficient of P (x)
is not divisible by p. For any prime p, the degree of P (x) modulo p is the largest integer
k, 0 ≤ k ≤ n, for which p ∤ ak . We denote this by degp P (x) = k.
Example. The degree of P (x) = 7x4 + 14x3 − 5x2 + 5x + 3 is 4. However, the degree of
P (x) is 2 modulo 7.
Theorem 2.11.4 (Lagrange’s Theorem). Let p be a prime and let P (x) be a poly-
nomial with integer coefficients not all divisible by p. Also, let degp P (x) = k. The
congruence equation P (x) ≡ 0 (mod p) has at most k incongruent solutions modulo p.
Note. Assume that
P (x) = an xn + an−1 xn−1 + · · · + a1 x + a0 .
If P (x) ≡ 0 (mod p) for some x, then since
(x + kp)i ≡ xi + (kp)i ≡ xi (mod p), ∀i, k ∈ N,
we have P (x + kp) ≡ 0 (mod p) as well. This means that we only need to search
for solutions in the set {0, 1, . . . , p − 1}. The term incongruent solutions in the above
theorem is there just for the same reason.
CHAPTER 2. MODULAR ARITHMETIC 121

Example. Let P (x) = 10x3 +3x2 +12x+17. The degree of P (x) modulo 5 is 2. According
to Lagrange’s theorem, the equation P (x) ≡ 0 (mod 5) has at most 2 solutions modulo
5. To check this, note that

P (x) = 10x3 + 3x2 + 12x + 17 ≡ 3x2 + 12x + 12 ≡ 3(x + 2)2 ≡ 0 (mod 5),

has only one solution x ≡ −2 modulo 5.

We will prove Lagrange’s theorem in the following.

Proof. We induct on k. Since degp P (x) = k, we can write

P (x) ≡ ak xk + ak−1 xk−1 + · · · + a1 x + a0 ,

where ak , ak−1 , . . . , a0 are coefficients of P (x). It is clear that for k = 0, the equation
f (x) = a0 has no solutions modulo p because p ∤ a0 . Assume that the claim is true
for all polynomials of degree up to k − 1 modulo p. Assume that P (x) ≡ 0 (mod p)
has d solutions. If d < k, we are done. Otherwise, if d ≥ k, take x1 , x2 , . . . , xk to be k
arbitrary incongruent solutions of P (x) ≡ 0 (mod p). Define

Q(x) = P (x) − ak (x − x1 )(x − x2 ) · · · (x − xk ).

Clearly, degp Q(x) < degp P (x) = k. However,

Q(x1 ) ≡ Q(x2 ) ≡ · · · ≡ Q(xk ) ≡ 0 (mod p),

which means Q(x) ≡ 0 (mod p) has at least k solutions. The induction hypothesis
forces that Q(x) ≡ 0 (mod p) for all x. It follows that

P (x) ≡ ak (x − x1 )(x − x2 ) · · · (x − xk ) (mod p).

This means that P (x) ≡ 0 (mod p) if and only if x − xi ≡ 0 (mod p) for some i ∈
{1, 2, . . . , k}. So, x1 , x2 , . . . , xk are the only solutions to P (x) ≡ 0 (mod p). The
induction is complete.

In this section, we will discuss only the following result and see how to apply it to
prove some other theorems.

Theorem 2.11.5 (Lagrange). If p is a prime and

(2.11) (x + 1)(x + 2) · · · (x + p − 1) = xp−1 + a1 xp−2 + · · · + ap−2 x + (p − 1)!

then the coefficients a1 , a2 , . . . , ap−2 are divisible by p where p is an odd prime.

The term xp−1 is produced by multiplying all x terms. Multiplying all the constant
terms, we get 1 · 2 · · · (p − 1) = (p − 1)!, which explains the reasoning behind the terms
on the right side of the equation. The proof is an intuitive one. Though there maybe
other proofs, we prefer this one.
122 2.11. LAGRANGE’S THEOREM

Proof. Assume that f (x) = (x + 1)(x + 2) · · · (x + p − 1). We start by noticing that

f (x + 1) = (x + 2)(x + 3) · · · (x + p). We can write
(x + p)f (x) = (x + 1)f (x + 1),
or equivalently,
(2.12) pf (x) = (x + 1)f (x + 1) − xf (x).
Substituting the expressions for f (x) and f (x + 1) in equation ( ), we see that
pf (x) = pxp−1 + pa1 xp−2 + · · · + pap−2 x + p!,
(x + 1)f (x + 1) = (x + 1)p + a1 (x + 1)p−1
+ · · · + ap−2 (x + 1)2 + (x + 1)(p − 1)!,
xf (x) = xp + a1 xp−1 + · · · + ap−2 x2 + x(p − 1)!.
Replace these values into ( ),
(x + 1)f (x + 1) − xf (x) = (x + 1)p − xp + a1 ((x + 1)p−1 − xp−1 )
(2.13) + · · · + ap−2 ((x + 1)2 − x2 ) + (x + 1 − x)(p − 1)!.
We need to expand the terms (x + 1)i − xi (for 1 ≤ i ≤ p) using binomial theorem so
we can collect the terms with same degree (exponent).
       
i i i i i−1 i i−2 i
(x + 1) − x = x + x + x + ··· + x + 1 − xi
1 2 i−1
     
i i−1 i i−2 i
(2.14) = x + x + ··· + x + 1.
1 2 i−1
Since pf (x) = (x + 1)f (x + 1) − xf (x), the coefficients of same exponents of x should
be the same for both sides. The coefficient of xp−2 in pf (x) is pa1 , while that of
(x + 1)f (x + 1) − xf (x) comes from the first two terms of ( ) (that is, (x + 1)p − xp
and a1 ((x + 1)p−1 − xp−1 )). Using ( ) to calculate these two terms, we get
   
p p−1
pa1 = + a1 .
2 1



From Theorem , we know that p divides kp for any 0 < k < p. So, p divides p2 ,
therefore p divides a1 .
Equating coefficient of xp−3 , we find
     
p p−1 p−2
pa2 = + a1 + a2 .
3 2 1


Here, p divides p3 and a1 , so p divides a2 . Continuing this process in a similar way, we
find that a1 , a2 , . . . , ap−2 are divisible by p. To check correctness of this, we can equate
the coefficient of x and find
     
p p−1 2
pap−2 = + a1 + · · · + ap−2 .
p−1 p−2 1
This equation implies p divides ap−2 , as claimed. The proof is complete.
CHAPTER 2. MODULAR ARITHMETIC 123

Before we describe some applications, let’s try to understand the coefficients a1 , a2 ,

. . . , ap−2 in a better way. By investigating ( ), one can easily obtain

a1 = 1 + 2 + · · · + p − 1
a2 = 1 · 2 + · · · + 1 · (p − 1) + 2 · 3 + · · · + 2 · (p − 1) + · · ·
..
.

You should already guess what a1 , . . . , ap−2 are. a1 is the sum of all 1,
· · · , p − 1. a2 is
the sum of products of two numbers from 1, · · · , p − 1 (all possible p−12
combinations).
Similarly, ap−2 is the sum of products of p − 2 numbers taken at a time. In general ai
the sum of all possible products of i numbers taken from 1, 2, · · · , p − 1. Therefore, we
can state Theorem as

Theorem 2.11.6. If p is an odd prime and 0 < k < p − 1, then the sum of all possible
products of k numbers taken at a time from 1, 2, . . . , p − 1 is divisible by p.

Let’s see just how powerful this theorem can be, if used properly. We can take
advantage of the fact that the theorem is actually an identity, so we can choose x freely
as we wish.

Proof of Wilson’s Theorem. The theorem is true when p = 2. Therefore, it is safe to

assume that p is odd. Put x = 1 in Theorem to obtain

2 × 3 × · · · × p = 1 + (a1 + · · · + ap−2 ) + (p − 1)!,

and so,

p! = 1 + a1 + · · · + ap−2 + (p − 1)!.

Clearly, p! is divisible by p, and so are a1 , · · · , ap−2 . Thus, 1 + (p − 1)! must be divisible

by p too, which is exactly what we want.

We will use this as an intermediary to prove Fermat’s theorem. We want to prove

xp−1 − 1 is divisible by p when x⊥p.

Proof of Fermat’s Theorem. Since x is co-prime to p, one of x + 1, · · · , x + p − 1 is

divisible by p because they are p − 1 consecutive integers. Therefore, their product is
divisible by p too. Thus,

(x + 1) · · · (x + p − 1) = xp−1 + a1 xp−2 + · · · + ap−2 x + (p − 1)!

Here, left side is divisible by p so must be right side. Again, since a1 , . . . , ap−2 are
multiples of p, we have xp−1 + (p − 1)! is a multiple of p.

xp−1 ≡ −(p − 1)! (mod p).

Hence, by Wilson’s theorem, xp−1 ≡ 1 (mod p), which finishes the proof.
124 2.12. ORDER, PRIMITIVE ROOTS

As for the last demonstration, we will use it to prove Wolstenholme’s theorem, which
we also proved before. The theorem requires us to show that for p > 3 a prime, the
numerator of
1 1
1 + + ··· +
2 p−1
is divisible by p2 in its reduced form.
Proof of Wolstenholme’s Theorem. The numerator is the sum of products of p − 2 num-
bers taken from 1, 2, . . . , p − 1. So, it is ap−2 . Since the denominator of the fraction is
(p − 1)!, which is not divisible by p, we only need to show that p2 |ap−2 .
Set x = −p in Theorem to obtain

(−p + 1) · · · (−p + p − 1) = pp−1 − a1 pp−2 + · · · − ap−2 p + (p − 1)!.

The left hand side of the above equation equals (p − 1)!. So

pp−1 − a1 pp−2 + · · · − ap−2 p = 0,

which gives

ap−2 = pp−2 + a1 pp−3 + · · · + ap−3 p2 .

If p > 3, then p − 2 ≥ 2 and all the terms on the right side are divisible by p2 .
Note. You should try to guess what motivates us to set exactly those values of x to
get nice results.

2.12 Order, Primitive Roots

Recall the examples we took while discussing Fermat’s little theorem. We were working
with something like 26 (mod 7) or 66 (mod 7). While calculating, we found that 62 ≡ 1
(mod 7) or 23 ≡ 1 (mod 7) which eventually led to 66 ≡ 1 (mod 7) and 26 (mod 7).
Along with the ideas we used there, did you conjecture anything else? We left a hint
when we said that since 62 ≡ 1 (mod 7), 64 ≡ 1 (mod 7) and 66 ≡ 1 (mod 7) as well.
We hope that this is sort of obvious by now. But it should also trigger you to think
of something. If we can find the smallest exponent for which 2x ≡ 1 (mod 7), then
we can say 2y ≡ 1 (mod 7) for all multiples of x (y here). We will shortly prove this
formally. Moreover, it also encourages us to study these smallest values for which we
get 1. The motivation is obvious. Whenever we get 1, we get a cycle of remainders
from which point, the remainders repeat. Just finish the examples above if you did not
entirely understand what we meant. We call this smallest integer order. And it should
be clear to you why the study of order is important.
Definition 2.12.1 (Order Modulo Integers). Let a and n be co-prime positive
integers. If x is the smallest positive integer such that

ax ≡ 1 (mod n),

then x is called the order of a modulo n. We denote this by ordn (a) = x.

CHAPTER 2. MODULAR ARITHMETIC 125

Example. ord8 (3) = 2 i.e. 2 is the smallest positive integer such that 32 ≡ 1 (mod 8).
Theorem 2.12.2. Let a and n be positive integers. If ordn (a) = d and ax ≡ 1 (mod n),
then d|x.
Proof. If x < d, it would contradict the fact that, d is such smallest positive integer
that ad ≡ 1 (mod n). We are left with the case x > d. Assume that x = dq + r with
0 ≤ r < d.
ax ≡ adq · ar ≡ (ad )q · ar ≡ 1 · ar ≡ ar (mod n).
So ar ≡ ax ≡ 1 (mod n). Since 0 ≤ r < d and d is the order of a, this is impossible
unless r = 0. Thus x = dq and we are done.
Corollary 2.12.3. If a ⊥ n, then ordn (a)|ϕ(n).
Proof. If d = ordn (a), then ad ≡ 1 (mod n). From Euler’s theorem, aϕ(n) ≡ 1 (mod n).
Then using Theorem , we can say that d|ϕ(n).
We can use this result to find orders in practice. We only need to check for divisors
of ϕ(n) and find the smallest divisor for which the relation ad ≡ 1 (mod n) holds.
Corollary 2.12.4. ak ≡ al (mod n) if and only if k ≡ l (mod ordn (a)).
Proof. ak ≡ al (mod n) implies ak−l ≡ 1 (mod n). By Theorem , we have
ordn (a)|k − l. The reverse of this approach can be applied to prove the other part
of the corollary.
One could ask if we know the order of a modulo n, how do we find the order of
other powers of a. Or, if we know order of a modulo two positive integers m and n,
then what would be the order of a modulo mn?
Theorem 2.12.5. If m and n are coprime positive integers such that ordm (a) = d and
ordn (a) = e, then ordmn (a) = [d, e].
Proof. Let ordmn (a) = h, so
ah ≡ 1 (mod mn),
which gives ah ≡ 1 (mod m) and ah ≡ 1 (mod n) as well. By Theorem , since d
and e are order of a modulo m and n, respectively, we have d|h and e|h. Therefore, for
the minimum h, we must have h = [d, e] to satisfy the conditions.
Theorem 2.12.6. Let a, b, and n be positive integers such that ordn (a) = k and
ordn (b) = l, where k⊥l. Then ordn (ab) = kl.
Proof. Let ordn (ab) = h. First, note that

l
alh ≡ alh · blh ≡ (ab)lh ≡ (ab)h ≡1 (mod n).
So, by Theorem , we have k|lh and since (k, l) = 1, it follows that k|h. We can
similarly prove that l|h. So kl|h. On the other hand,
(ab)kl ≡ (ak )l · (bl )k ≡ 1 (mod n).
Again, by Theorem ( ), we have h|kl. This finishes the proof.
126 2.12. ORDER, PRIMITIVE ROOTS

Theorem 2.12.7. If the order of a modulo n is d, then the order of ak modulo n is

d/(d, k).

h
Proof. Let the order of ak modulo n be h. Then ak ≡ akh ≡ 1 (mod n). Theorem
says that d must divide kh. Assume that (k, d) = g, so there exist coprime
positive integers l and e such that k = gl and d = ge. Rewriting d|kh implies

ge|glh =⇒ e|lh,

and since l⊥e, e must divide h. Since dl = ke = gel,

e
l
ak ≡ ad ≡ 1 (mod n).

This means that the order of ak modulo n must divide e. So, h divides e as well. We
get that
d
h=e=
(d, k)
must hold.
The previous theorem also implies the following one.
Theorem 2.12.8. The order of a modulo n is the same as the order of ak modulo n if
and only if (k, n) = 1.
Here is a very useful theorem, often used to solve Diophantine equations.
Theorem 2.12.9. Let q be a prime and x be a positive integer. Every prime divisor
of the number

1 + x + · · · + xq−1

is either q or congruent to 1 modulo q.

Proof. The sum can be written as

q−1 xq − 1
S = 1 + x + ··· + x = .
x−1
Let p be any prime factor of S. Then

xq ≡ 1 (mod p).

If the order of x modulo p is d, we have d|q. Since q is a prime, either d = 1 or d = q.

If d = 1, then x ≡ 1 (mod p). In that case,

S ≡ 1 + 1 + ··· + 1 (mod p),

which gives 0 ≡ q (mod p). So, p = q. Now assume the case that d = q. Because of
Fermat’s little theorem,

xp−1 ≡ 1 (mod p).

This implies that d = q divides p − 1. So p ≡ 1 (mod q), as claimed.

CHAPTER 2. MODULAR ARITHMETIC 127

Definition 2.12.10 (Primitive Root). Let n be a given positive integer. An integer

g which is relatively prime to n is called a primitive root modulo n if ordn (g) = ϕ(n).
That is, if g x 6≡ 1 (mod n) for any positive integer x < ϕ(n).

Note. Using this definition, we can say that:

1. Let g be a positive integer coprime to n. It is clear that g m is also coprime to n

for any m ∈ N.

2. If g is a primitive root modulo n and if g a ≡ g b (mod n) for some positive integers

a and b less than ϕ(n), then a = b. The reason is simple: if a 6= b, then g a−b ≡ 1
(mod n), which is absurd since a − b ≤ ϕ(n) and g is a primitive root.

These two notes tell us that if g is a primitive root of n, then the set {g, g 2 , · · · , g ϕ(n) }
is equal to Un , where Un is the set of units modulo n (as defined in Definition ).
Notice that equality of these two sets is considered modulo n. Actually, the set
{g, g 2 , · · · , g ϕ(n) } may contain some elements larger than n. We reduce those elements
modulo n so that we have all elements less than n. This new set is now equal to Un .
We may denote this by the notation {g, g 2 , · · · , g ϕ(n) } ≡ Un (mod n).
In algebraic words, g is a generator of Un . Moreover, the generators of Un are
exactly the primitive roots of n (if there is any). We will summarize this result in the
following theorem.

Theorem 2.12.11. A primitive root g modulo n (if existing) is a generator of Un . That

is, for any a ∈ U, there is a unique k with 0 < k ≤ ϕ(n) such that g k ≡ a (mod n).

Proof. Consider the powers g, g 2 , . . . , g ϕ(n) modulo n. Now assume g u ≡ g v (mod n)

for some u, v ∈ {1, 2, . . . , ϕ(n)}, then since (g, n) = 1, we obtain

g u−v ≡ 1 (mod n).

This is not possible unless u = v. The reason is simple: if u 6= v, then we have found
some x = u − v such that 0 < x ≤ ϕ(n) and g x ≡ 1 (mod n), which is in contradiction
with g being a primitive root modulo n.

Example.

1. 3 is a primitive root modulo 7 since ϕ(7) = 6 and 3i 6≡ 1 (mod 7) for i ∈

{1, 2, 3, 4, 5}. Notice that powers of 3 create the whole set U7 :

31 ≡ 3, 32 ≡ 2, 33 ≡ 6, 34 ≡ 4, 35 ≡ 5, 36 ≡ 1,

where all the congruences are taken modulo 7.

2. Let’s see if there exists a primitive root modulo 15. To show this, a possible way
is to start from a = 2 and compute all the powers ai for i = 2, . . . , ϕ(15) − 1 = 7
modulo 15 one by one:

22 ≡ 4, 23 ≡ 8, 24 ≡ 1.
128 2.12. ORDER, PRIMITIVE ROOTS

We stop at 24 because we got 1 mod 15, and this shows that 2 is not a primitive
root modulo 15. Then, we should do the same process, but this time for a = 4
(we don’t check 3 because it’s not coprime to 15). Now you should be able to do
the math much faster, and come up with 42 ≡ 1 (mod 15), which shows a = 4 is
not a primitivie root modulo 15. Fortunately, we don’t need to check a = 5 and
a = 6. For a = 7, the computations are not as easy as a = 2 and a = 4, but still
not hard
72 ≡ 4, 73 ≡ 13, 74 ≡ 1.
So, 7 is not a primitive root mod 15 either. Now, we don’t need to do the
computations for a = 8 because in this case, a−1 is 2 and we showed that 2 is not
a primitive root (why is that enough?). The next values for a to check are 11, 13,
and 14. Since 13 = 7−1 , we don’t need to worry about 13. Check 11 and 14
for yourself and verify that neither of them are primitive roots mod 15 (we can’t
do the modular arithmetic invers trick here because 11 and 14 are the inverse of
themselves modulo 15). This shows that there is no primitive root mod 15.

3. Let’s assume that elements gi of Un are sorted in ascending order. That is,
g1 < · · · < gϕ(n) . Then we have g1 = 1 and gϕ(n) = n − 1.

Corollary 2.12.12. If g is a primitive root of p then

G = {g 1 , g 2 , . . . , g p−1 }

forms a complete residue system modulo p.

Theorem 2.12.13. Let n be a positive integer and let a be a quadratic non-residue

modulo n such that a⊥n. Assume that Un = {g1 , g2 , . . . , gϕ(n) }. Then
ϕ(n)
g1 g2 · · · gϕ(n) ≡ a 2 (mod n).

Proof. According to Theorem , for any gi , there exists some x such that

(2.15) gi x ≡ a (mod n).

It is clear that x⊥n because if (x, n) = d, then

d|n|gi x − a =⇒ d|a =⇒ d|(a, n) = 1 =⇒ d = 1.

So x = gj for some j. We have gi 6= gj since a is a quadratic non-residue. Moreover, gj

is unique because if gi gk ≡ a (mod n) for some k, then gk ≡ gj (mod n) and since gk
and gj are both less than n, this forces gk = gj . Thus, we can pair up the ϕ(n) elements
of Un into ϕ(n)/2 pairs (gi , gj ), such that gi gj ≡ a (mod n). Hence,
ϕ(n)
g1 g2 · · · gϕ(n) ≡ a 2 (mod n).

Here is a nice theorem which relates primitive roots and quadratic residues modulo
a prime p.
CHAPTER 2. MODULAR ARITHMETIC 129

Theorem 2.12.14. If g is a primitive root modulo a prime p, then the quadratic

residues of p are g 2 , g 4 , . . . , g p−1 .

Proof. By Euler’s criterion, we know that if a is a quadratic residue modulo p, then

p−1
a 2 ≡1 (mod p).

Using Fermat’s little theorem, it follows that g 2 , g 4 , . . . , g p−1 are all quadratic residues
modulo p. Since the set {g 1 , g 2 , . . . , g p−1 } is a complete residue set modulo p and
we know by Theorem that there are exactly p−1 2
incongruent quadratic residues
modulo p, we find that g , g , . . . , g
2 4 p−1
are the only quadratic residues.

It is a natural question whether there exists a primitive root g modulo an arbitrary

positive integer n. If the answer is negative, one might ask for which n there exists a
primitive root. We will answer these questions shortly.

Theorem 2.12.15. A positive integer g is a primitive root modulo n if and only if

ϕ(n)
g p 6≡ 1 (mod n)

for any prime p which divides ϕ(n).

Proof. It is straightforward to check the truth of the “if” part. For the sake of contra-
diction, assume ϕ(n) = pk and

gk ≡ 1 (mod n).

But this would contradict the minimality of ϕ(n) since k is less than ϕ(n), meaning
that g is not a primitive root.
For the “only if” part, assume that for every prime divisor p of ϕ(n), we have
ϕ(n)
g p 6≡ 1 (mod n).

We want to show that g is a primitive root modulo n. Let d = ordn (g), so that d | ϕ(n).
If d < ϕ(n), then we must have d | ϕ(n)/p for some prime p dividing ϕ(n). Letting
ϕ(n) = pdl,
ϕ(n)
l
g p ≡ g dl ≡ g d ≡1 (mod n),

which is a contradiction. Hence, d = ϕ(n) must hold.

Note. In the proof above, we could just take p to be the smallest prime divisor of ϕ(n).
Then we must have that d is a divisor of ϕ(n)/p. This is because the greatest divisor
of n less than n is n/p, where p is the smallest prime divisor of n (can you sense why?).

Corollary 2.12.16. Let m be a positive integer. If g is a primitive root of n, then g m

is also a primitive root modulo n if and only if m⊥ϕ(m).
130 2.12. ORDER, PRIMITIVE ROOTS

Proof. Let (m, ϕ(n)) = d, so that mϕ(n) = d·[m, ϕ(n)]. According to Theorem ,
g m is a primitive root modulo n if and only if

(2.16) g d·[m,ϕ(n)]/p 6≡ 1 (mod n),

for all prime divisors p of ϕ(n). Now, if d 6= 1, there exists a prime q which divides d.
In that case, write d = qk for some integer k. But then,

g d·[m,ϕ(n)]/q ≡ g k·[m,ϕ(n)] (mod n),

and since [m, ϕ(n)] is divisible by ϕ(n), we have g [m,ϕ(n)] ≡ 1 (mod n). Thus,

k
g d·[m,ϕ(n)]/q ≡ g [m,ϕ(n)] ≡1 (mod n),

which is in contradiction with equation ( ) since q is a prime divisor of ϕ(n) (why?).

So, we must have d = 1, and the proof is complete.
Assume that some positive integer n has a primitive root. An interesting question
is to find the number of primitive roots which are incongruent modulo n. The next
theorem answers this question.
Theorem 2.12.17. For any positive integer n, if there exists a primitive root modulo
n, then there are exactly ϕ(ϕ(n)) incongruent primitive roots modulo n.
Note. In case the word incongruent is somewhat unclear to you: two integers a and b
are called incongruent modulo a natural number n if and only if a 6≡ b (mod n).
Proof. Assume that g is a primitive root modulo n. We aim to find all primitive roots
of n. Since we are looking for incongruent primitive roots modulo n, it suffices to
search in the set Un . Theorem tells us that {g, g 2 , · · · , g ϕ(n) } ≡ Un (mod n)
and so we should search for primitive roots in the set {g, g 2 , · · · , g ϕ(n) }. On the other
hand, Corollary implies tha we should only investigate powers g m of g for which
(m, ϕ(n)) = 1. The number of such elements is ϕ(ϕ(n)).
We are back to the first question: for which integers n do we have a primitive root?
The process of finding such n is long, and we will break it into smaller parts. The first
step is to see if there exist primitive roots modulo primes. We will soon prove that
there always exists a primitive root modulo any prime. We need the following lemma
to prove our claim.
Lemma 2.12.18. Let p be a prime and d be a positive integer such that d | p − 1.
Then xd − 1 ≡ 0 (mod p) has exactly d incongruent solutions modulo p.
Proof. Let p − 1 = dk for some integer k. Consider the polynomial

2
k−1
P (x) = 1 + xd + xd + · · · + xd .

Then,

(2.17) xp−1 − 1 = (xd − 1)P (x).

CHAPTER 2. MODULAR ARITHMETIC 131

By Fermat’s theorem, all integers 1, 2, . . . , p − 1 are solutions to xp−1 − 1 ≡ 0 (mod p).

So, this equation has exactly p − 1 = dk solutions. From ( ), each of these dk
solutions is either a solution of P (x) ≡ 0 (mod p) or a solution of xd − 1 ≡ 0 (mod p).
However, Lagrange’s theorem says that P (x) ≡ 0 (mod p) has at most d(k−1) solutions
and that xd − 1 ≡ 0 (mod p) has at most d solutions. Since dk = d(k − 1) + d, this
is only possible when xd − 1 ≡ 0 (mod p) has exactly d solutions and also P (x) ≡ 0
(mod p) has exactly d(k − 1) solutions.
Theorem 2.12.19. Let p be a prime. There are exactly ϕ(p − 1) incongruent primitive
roots modulo p.
Proof. The case p = 2 is obvious. If there exists one primitive root of p, then by
Theorem there are exactly ϕ(ϕ(p)) = ϕ(p − 1) incongruent primitive roots of p.
So we just need to construct a primitive root for p. The trick is to factorize ϕ(p) =
p − 1 into product of prime powers. Let q be a prime such that q k | p − 1 for some
integer k ≥ 1. We want to show that there exists some integer a for which ordp (a) = q k .
k
By previous lemma, the equation xq − 1 ≡ 0 (mod p) has exactly q k solutions. Take
k
a to be one of these solutions. Then aq ≡ 1 (mod p), and so by Theorem , it
follows that ordp (a) | q . So ordp (a) = q for some integer 0 ≤ j ≤ k. This means that
k j
j
a is a solution to the equation xq − 1 ≡ 0 (mod p). If j = k, we have found such an a.
j
Otherwise, suppose that j < k. Let i = k − j ≥ 1. Note that if xq ≡ 1 (mod p), then,
 j k−j−1  j i−1
q k−1
x ≡ xq ≡ xq ≡ 1 (mod p).
j k−1
That is, every solution to xq ≡ 1 (mod p) is also a solution to xq ≡ 1 (mod p).
q k−1
According to the preceding lemma, number of solutions of x ≡ 1 (mod p) is exactly
qk
q . So there are exactly q − q
k−1 k k−1
integers x which satisfy x ≡ 1 (mod p) but not
k−1
xq ≡ 1 (mod p). If we select a from these solutions, we will have ordp (a) = q k .
To finish the proof, let
p − 1 = q1α1 q2α2 · · · qtαt ,
be the prime factorization of p − 1. From what we have just proved, there exists some
integer ai for each qi such that ordp (ai ) = qiαi . According to Theorem ( ), since qi
are coprime,
t
!
Y
degp ai = degp (a1 ) degp (a2 ) · · · degp (at ) = q1α1 q2α2 · · · qtαt = p − 1 = ϕ(p),
i=1
Qt
and so i=1 ai is a primitive root modulo p.
The next step is to find what other numbers have a primitive root. We will show,
by the help of the following lemma, that all powers of an odd prime number have a
primitive root.
Lemma 2.12.20. Let p be an odd prime and let g be a primitive root modulo p such
that g p−1 6≡ 1 (mod p2 ). Then,
k
g ϕ(p ) ≡
6 1 (mod pk+1 )
for any integer k ≥ 1.
132 2.12. ORDER, PRIMITIVE ROOTS

Proof. We will induct on k. The base case k = 1 is immediately followed from the
k
assumption that g p−1 6≡ 1 (mod p2 ). As the induction hypothesis, consider that g ϕ(p ) 6≡
k
1 (mod pk+1 ) for some k ≥ 1. From Euler’s theorem, g ϕ(p ) ≡ 1 (mod pk ), which means
k
g ϕ(p ) = 1 + mpk

for some m. The induction hypothesis implies that p ∤ m. By Proposition , we

know that


ϕ(pk+1 ) = pk+1 − pk = p pk − pk−1 = p · ϕ(pk ).

Hence,
k+1 )
g ϕ(p = (1 + mpk )p
     
p k p k 2 p
=1+ mp + (mp ) + · · · + (mpk )p−1 + (mpk )p
1 2 p−1
| {z }
divisible by pk+2
k+1 k+2
≡ 1 + mp (mod p ).

As m is not divisible by p, mpk+1 is not divisible by pk+2 . So,

k+1 )
g ϕ(p 6≡ 1 (mod pk+2 ),

as desired.

Theorem 2.12.21. Let p be an odd prime and let g be a primitive root modulo p (as
we know exists from Theorem ). Then, either g or g + p is a primitive root
k
modulo p for any integer k ≥ 1.

Proof. We break the proof into two parts:

1. If g p−1 6≡ 1 (mod p2 ), then we will show that g is a primitive root of pk . That is,
we will prove that

(2.18) ordpk (g) = ϕ(pk ) = pk−1 (p − 1).

This is obviously true for k = 1. Suppose that equation ( ) holds for some
k ≥ 1. We will prove that it also holds for k + 1. Let ordpk+1 (g) = m. Then

gm ≡ 1 (mod pk+1 ) =⇒ g m ≡ 1 (mod pk ).

Since we know that order of g modulo pk is ϕ(pk ), we should have ϕ(pk ) =

pk−1 (p − 1)|m. On the other hand, since m is the order of g modulo pk+1 , by
corollary ( ), we get m|ϕ(pk+1 ) = pk (p − 1). Therefore, m equals either
ϕ(pk ) = pk−1 (p − 1) or ϕ(pk+1 ) = pk (p − 1). Previous lemma states that it is
impossible to have m = ordpk+1 (g) = ϕ(pk ). So, m = ϕ(pk+1 ) and we are done.
CHAPTER 2. MODULAR ARITHMETIC 133

2. If g p−1 ≡ 1 (mod p2 ), then we will show that g + p is a primitive root modulo pk

for any integer k ≥ 1. Note that
 
p−1 p−1 p − 1 p−2
(g + p) =g + g p
1
   
p − 1 p−3 2 p−1
+ g p + ··· + gpp−2 + pp−1
2 p−2
| {z }
divisible byp2

Taking modulo p2 , the above is

≡ g p−1 + (p − 1)g p−2 p ≡ 1 − g p−2 p 6≡ 1 (mod p2 ),

because p ∤ g. We can now apply the same approach we followed in the first case,
but now with g + p instead of g. So g + p is a primitive root modulo all powers
of p and the proof is complete.

Finally, we are ready to answer our question.

Theorem 2.12.22 (Primitive Root Theorem). Let n > 1 be a positive integer.

There exists a primitive root modulo n if and only if n ∈ {2, 4, pk , 2pk } for some odd
prime p and a positive integer k.

Proof. Obviously, g = 1 and g = 3 are primitive roots modulo 2 and 4, respectively. So,
n = 2 and n = 4 are off the list. Let’s consider the “if” part first. If n has a primitive
root, we will prove n must be of the form pk or 2pk , where p is an odd prime. First,
let us show that 2k does not have a primitive root for k > 2. It is obvious that if a is
a primitive root modulo 2k , then a is odd. We leave it as an exercise for the reader to
prove by induction that for any odd a and k > 2,
k−2
2 k | a2 − 1.

Since ϕ(2k ) = 2k−1 , a is never a primitive root modulo 2k .

Now, if n is not of the form pk or 2pk , we can write n = ab with gcd(a, b) = 1 and
a > b > 2. So, ϕ(b) and ϕ(a) are larger than 1, and by Proposition , they are both
even. Let g be a primitive root modulo n. This means that ordn (g) = ϕ(ab). We will
show that this cannot happen. Let orda (g) = d and ordb (g) = e. Since

g ϕ(a) ≡ 1 (mod a), and

ϕ(b)
g ≡1 (mod b),

by Corollary ( ), we find that d | ϕ(a) and e | ϕ(b). Hence, by Theorem ( ),

ϕ(a)ϕ(b) ϕ(ab) ϕ(ab)

ordab (g) = [d, e] ≤ [ϕ(a), ϕ(b)] = = ≤ ,
(ϕ(a), ϕ(b)) (ϕ(a), ϕ(b)) 2
134 2.12. ORDER, PRIMITIVE ROOTS

where we have used the fact that (ϕ(a), ϕ(b)) is at least 2. This gives us the contradic-
tion we were looking for. So, n must be of the form pk or 2pk .
The only remaining part is to prove that for an odd prime p and k ≥ 1, there exist
primitive roots modulo the numbers pk and 2pk . According to Theorem , p has
a primitive root, say g. It now follows from Theorem that either g or g + p is a
primitive root modulo pk . Since Euler’s totient function is multiplicative, we have
ϕ(2pk ) = ϕ(2) · ϕ(pk ) = ϕ(pk ).
Let g be a primitive root modulo pk .
1. If g is odd, then
ga ≡ 1 (mod pk ) ⇐⇒ g a ≡ 1 (mod 2pk ).
Let m = ord2pk (g). If m < ϕ(2pk ) = ϕ(pk ), then g m ≡ 1 (mod 2pk ) implies
g m ≡ 1 (mod pk ), which contradicts the fact that g is a primitive root modulo
pk . Therefore, g is also a primitive root modulo 2pk .
2. If g is even, then g ′ = g + pk is an odd number and it is also a primitive root
modulo pk . Applying the same approach used in the first case, we find that g ′ is
a primitive root modulo 2pk .
We have shown that 2pk always has a primitive root and the proof is complete.
Here is a generalization of Wilson’s theorem, though it can be generalized even
further. We refer the reader to section ( ) of the book to see another generalization
of Wilson’s theorem.
Problem 2.12.23. Let n be a positive integer and let Un = {g1 , g2 , . . . , gϕ(n) }. Prove
that if there exists a primitive root modulo n, then
g1 g2 · · · gϕ(n) ≡ −1 (mod n).
Otherwise,
g1 g2 · · · gϕ(n) ≡ 1 (mod n).
Hint. Combine Theorems and along with the fact that if p is an odd
prime and k is a positive integer, then p | a − 1 implies pk | a + 1 or pk | a − 1.
k 2

Theorem 2.12.24. Let g be a primitive root modulo n. Then n − g is a primitive root

modulo n as well if 4 divides ϕ(n).
Proof. We have a criteria to see if x is a primitive root modulo n. We need to check if
xϕ(n)/p 6≡ 1 (mod n) for any prime p which divides ϕ(n). Therefore, to check if n − g is
a primitive root of n, we just need to prove the following holds
(n − g)ϕ(n)/p 6≡ 1 (mod n)
for any prime divisor p of ϕ(n). Now, since 4 | ϕ(n), we have 2 | ϕ(n)/2. So, ϕ(n)/p is
even for any proper p. Using the fact that g 2 ≡ (n − g)2 (mod n), we get
ϕ(n)
ϕ(n)
ϕ(n) ϕ(n)
(n − g) p ≡ (n − g)2 2p ≡ g 2 2p ≡ g p 6≡ 1 (mod n).
Thus, (n − g) is a primitive root modulo n as well.
CHAPTER 2. MODULAR ARITHMETIC 135

The use of primitive roots is usually not obvious in problems. There is hardly any
hint on why you should use it. Best if you see its use through problems.

Problem 2.12.25. Let p be odd prime number. Prove that equation xp−1 ≡ 1
(mod pn ) has exactly p − 1 different solution modulo pn .

Solution (1). Let g be a primitive root modulo pn (which exists by Theorem ).

Now, take x = g , so every solution x maps to a certain k. The number of different k
k

is the number of solutions of this congruence equation. Since

ordpn (g) = pn−1 (p − 1)andg k(p−1) ≡ 1 (mod pn ),

we either have
pn−1 (p − 1) | k(p − 1)
or
pn−1 | k
for any such k. Take k = pn−1 ℓ. If ℓ = sp + r with 1 ≤ r < p, then we have that
n−1 (p−1)(sp+r) n−1 (p−1)r
gk = gp ≡ gp (mod pn ).

Therefore, for two incongruent solutions, we must have 1 ≤ r ≤ p − 1, giving us exactly

p − 1 solutions.

Solution (2). This is a special case of Lemma , where d = p − 1.

Problem 2.12.26. Prove that 3 is a primitive root modulo p, where p is any prime of
the form 2n + 1 for some integer n > 1.

Solution (1). p = 2n + 1 in particular means p ≡ 1 (mod 4). According to Problem

r
, we find that p is of the form 22 + 1 for some positive integer r. Therefore,
r r
p = 22 + 1 ≡ (−1)2 + 1 ≡ 2 (mod 3),

which is not a quadratic residue modulo 3. Using the law of quadratic reciprocity and
the fact that p ≡ 1 (mod 4),
  
3 p
= (−1)(p−1)/2 = 1.
p 3

 
From the above discussion, we know that p3 = −1. Therefore, p3 = −1 and 3 is a
quadratic non-residue modulo p.
r
We will now prove that for a prime of the form p = 22 +1 every quadratic non-residue
modulo p is a primitive root modulo p. Since p is a prime, we know that there exists
a primitive root modulo p, say g. By Theorem , we know that g 2 , g 4 , . . . , g p−1
are p − 1/2 different nonzero residues modulo p and they are all quadratic residues.
Therefore, all the quadratic non-residues are given by

g, g 3 , g 5 , . . . , g p−2 .
136 2.12. ORDER, PRIMITIVE ROOTS

We will now take one of these residues, say g 2k+1 , and show that it is a primitive root
mod p. This means we want to show that

g 2k+1 , g 2(2k+1) , g 3(2k+1) , . . . , g (p−1)(2k+1)

are incongruent modulo p, which happens if and only if

2k + 1, 2(2k + 1), 3(2k + 1), . . . , (p − 1)(2k + 1)

are all different modulo p − 1. This happens if and only if (2k + 1, p − 1) = 1, or

r r
(2k + 1, 22 ) = 1, which is clearly true since 2k + 1 is odd and 22 is a power of 2.
Therefore, all quadratic-non residues are primitive roots modulo p, and as we have
shown 3 is among them, we are done.

Solution (2). Just like the previous solution, we will use the fact that 3 is not a
quadratic residue modulo p. Therefore, by Euler’s criterion,
 
p−1 3 m−1
(2.19) 3 2 ≡ = −1 (mod p) =⇒ 32 ≡ −1 (mod p).
p

Let d be the order of 3 modulo p. Since d | p − 1 = 2n , we must have d = 2α for some

integer α. If α < n then
α n−1
32 ≡ 1 (mod p) =⇒ 32 ≡1 (mod p),

which is in contradiction with equation ( ). So, d = 2n , and this means that 3 is

primitive root modulo p = 2n + 1.

Problem 2.12.27. Let p and q be prime numbers such that p = 2q + 1. Let a be

an integer coprime to p and incongruent to −1, 0, and 1 modulo p. Show that −a2 is
primitive root modulo p.

Solution. Check q = 2 for yourself. Assume q ≥ 3 is an odd prime, say q = 2k + 1.

Hence, p = 4k + 3, or p ≡ 3 (mod 4). According to Theorem , −a2 is not a
quadratic residue modulo p. Suppose that −a is not a primitive root modulo p. Let g
2

be a primitive root modulo p. Theorem states that there exists an l ≥ 1 such

that
g 2l+1 ≡ −a2 (mod p).
Since −a2 is not a primitive root, there exists an integer k with k < p − 1 such that

g (2l+1)k ≡ (−a2 )k ≡ 1 (mod p).

This, together with Fermat’s little theorem, implies (2l + 1)k | p − 1 = 2q and hence
k = 2. Therefore, a4 ≡ 1 (mod p) and by Fermat’s little theorem 4 | 2q, which leads to
a contradiction as q is prime. Hence, −a2 is a primitive root modulo p.

Problem 2.12.28. Let q be a prime such that q ≡ 1 (mod 4) and that p = 2q + 1 is

also prime. Prove that 2 is a primitive root mod p.
CHAPTER 2. MODULAR ARITHMETIC 137

Solution. By Euler’s criterion, we have

p−1
2q ≡ 2 2 ≡ ±1 (mod p).

We analyze both cases now:

• Assume that 2q ≡ −1 (mod p). Let ordp (2) = d. Then, d | ϕ(p) = p − 1 = 2q.
Since q is a prime, we must have

d ∈ {1, 2, q, 2q}.

Since q ≡ 1 (mod 4), we have q ≥ 5 and the cases d = 1 and d = 2 cannot

happen. Also, if d = q, then 2q ≡ 1 (mod p), which is in contradiction with
2q ≡ −1 (mod p). Thus d = 2q = p − 1 and 2 is a primitive root modulo p.

• Assume that 2q ≡ 1 (mod p). Multiply both sides of this equation by 2 to get
2q+1 ≡ 2 (mod p). Since q + 1 is even, the latter equation means that 2 is a
quadratic residue modulo p. Therefore, by Theorem , p must be congruent
to either 1 or 7 modulo 8. However, problem says that q ≡ 1 (mod 4) which
results in q ≡ 1 or 5 (mod 8). Now,

p ≡ 2q + 1 ≡ 3 or 5 (mod 8),

which is a quick contradiction. Hence, 2q ≡ 1 (mod p) is not possible.

Problem 2.12.29. Suppose that p is an odd prime number. Prove that there exists a
positive integer x such that x and 4x are both primitive roots modulo p.

Solution. We will prove a stronger claim: there exists some x such that both x and
d2 x are primitive roots mod p for any integer d. Let g be a primitive root modulo p.
Since d2 is a quadratic residue mod p, it follows by Theorem that

d2 ≡ g 2k (mod p)

for some integer k. We then find by Corollary that any power g n of g is a

primitive root modulo p if and only if (n, p − 1) = 1.
Now, it suffices to show there exist two integers a and b such that

b − a = 2k and gcd(b, p − 1) = gcd(a, p − 1) = 1,

because then x = g a would be a solution. This is luckily easy. Let 2, q1 , q2 , . . . , qz be

the prime divisors of p − 1. Suppose that a1 , a2 , . . . , az are integers such that 2k ≡ ai
(mod qi ) for each 1 ≤ i ≤ z. By CRT, there exists an a such that

a≡1 (mod 2) and a ≡ −ai + pi (mod qi ) for all 1 ≤ i ≤ z,

where pi is some prime, not equal to qi , and ai 6≡ pi (mod qi ). It is easy to see

gcd(a, p − 1) = 1 and gcd(a + 2k, p − 1) = 1. Thus, g a and g a+2k are primitive roots
modulo p and g a+2k ≡ d2 g a (mod p), done.
138 2.13. CARMICHAEL FUNCTION, PRIMITIVE λ-ROOTS

2.13 Carmichael Function, Primitive λ-roots

2.13.1 Carmichael λ Function
In this section, we discuss a very important function, first introduced by Robert Daniel
Carmicheal in [ ]. Consider the following scenario: coprime positive integers a and n
are given, and ordn (a) = d. Now, fix n. Consider the case when ad ≡ 1 (mod n) holds
for any positive integer a coprime to n. This brings up some questions.

Problem 2.13.1. Does there exists an a such that ordn (a) = d?

Problem 2.13.2. How do we find the minimum d such that ad ≡ 1 (mod n) holds for
any a coprime to n?

Let’s proceed slowly. We will develop the theories that can solve these problems.
For doing that, we have to use properties of order and primitive roots we discussed in
previous sections.

Definition 2.13.3 (Carmichael Function). For a positive integer n, λ(n) is the

smallest positive integer for which aλ(n) ≡ 1 (mod n) holds for any positive integer
a relatively prime to n. This number λ(n) is called the Carmichael function of n.
Sometimes, it is also called the lambda function of n.

Note that Theorem ( ) implies the following theorem.

Theorem 2.13.4. If ad ≡ 1 (mod n) holds for all a coprime to n, then λ(n) | d.

Corollary 2.13.5. For any positive integer n, λ(n) | ϕ(n).

The following theorem is self-implicating and solves the first problem, if we can
prove that λ(n) exists. For now, let’s assume it does.

Theorem 2.13.6. Let n be a positive integer. There exists a positive integer a coprime
to n such that ordn (a) = λ(n).

Let’s focus on finding λ(n). First, consider the case n = 2k .

Theorem 2.13.7. If k > 2 then λ(2k ) = 2k−2 .

Proof. The integers coprime to 2k are all odd numbers. We will prove by induction
k−2
that x2 ≡ 1 (mod 2k ) holds for all odd positive integers x. The base case k = 3 is
obvious. Assume that for some k ≥ 3, we have
k−2
x2 ≡1 (mod 2k ),
k−2
or equivalently, x2 − 1 = 2k t for some t. Using the identity a2 − b2 = (a − b)(a + b),
we can write
k−1
 k−2   k−2 


(2.20) x2 − 1 = x2 − 1 x2 + 1 = 2k t 2k t + 2 = 2k+1 t 2k−1 t + 1 .
CHAPTER 2. MODULAR ARITHMETIC 139
k−1
This gives x2 ≡ 1 (mod 2k+1 ), and the induction is complete.
Now, we should prove that 2k−2 indeed is the smallest such integer. Again, by
induction, the base case is to find an x for which ord8 (x) = 2. Obviously, any x =
8j ± 3 satisfies this condition. Assume that for all numbers t from 1 up to k, we have
k−1
λ(2l ) = 2l−2 . Let λ(2k+1 ) = λ. Since we proved that x2 ≡ 1 (mod 2k+1 ) for all odd
x, it follows from Theorem that λ | 2k−1 . So λ is a power of 2. If λ = 2k−1 , we
are done. Otherwise, let λ = 2 , where 1 ≤ α < k − 1. Then for every x, one can write
α

α
(2.21) x2 ≡ 1 (mod 2k+1 ).

However, similarly as in ( ), for some t,

α
(2.22) x2 − 1 = 2α+2 t (2α t + 1) .
α
In ( ), the highest power of 2 which divides x2 − 1 is 2α+2 (since 2α t + 1 is odd).
But
α + 2 < (k − 1) + 2 = k + 1
, which contradicts ( ). The induction is complete.

Theorem 2.13.8. For any prime p and any positive integer k,

λ(pk ) = λ(2pk ) = ϕ(pk ).

Proof. Consider the congruence equation xd ≡ 1 (mod pk ) and let d = λ(pk ). By

Corollary , d | ϕ(pk ). Take x = g where g is a primitive root modulo pk . Then,
ordpk (g) = ϕ(pk ) and we immediately have ϕ(pk )|d. Thus, d = ϕ(pk ). A very similar
proof can be stated to show that λ(2pk ) = ϕ(2pk ) = ϕ(pk ).

Theorem 2.13.9. Let a and b be coprime positive integers. Then

λ(ab) = lcm(λ(a), λ(b)).

Proof. Suppose that

λ(a) = d, λ(b) = e, and λ(ab) = h.

Then,
xd ≡ 1 (mod a), xe ≡ 1 (mod b), and xh ≡ 1 (mod ab).
We also have xh ≡ 1 (mod a) and xh ≡ 1 (mod b) as well. Hence, d | h and e | h. This
means that [d, e] = h since [d, e] is the smallest positive integer that is divisible by both
d and e.
Generalization of this theorem is as follows.

Theorem 2.13.10. For any two positive integers a and b,

lcm(λ(a), λ(b)) = λ(lcm(a, b)).

The next theorem combines the above results and finds λ(n) for all n.
140 2.13. CARMICHAEL FUNCTION, PRIMITIVE λ-ROOTS

Theorem 2.13.11. Let n be a positive integer with prime factorization n = pe11 pe22 · · · perr .
Also, let p be a prime and k be a positive integer. Then


 ϕ(n), if n = 2, 4, pk , or 2pk ,

ϕ(n)
λ(n) = , if n = 2k with k > 2,

 2
lcm(λ(pe11 ), . . . , λ(perr )), otherwise.


Theorem 2.13.12. For positive integers a and b, if a | b, then λ(a) | λ(b).

The proof is left as an exercise for the reader. We are now ready to fully solve
Problem .

Theorem 2.13.13. For fixed positive integers n and d, there exists a positive integer
a coprime to n so that ordn (a) = d if and only if d | λ(n).

Proof. The “if” part is true by Theorem . For the “only if” part, assume that g is
an integer with ordn (g) = λ(n) and λ(n) = de. Then ordn (g e ) = d, as desired.

We finish this section by proposing a theorem. We will leave the proof for the reader
as an exercise.

Theorem 2.13.14. If λ(n) is coprime to n, then n is square-free.

Recall that n is square-free if it is not divisible by any perfect square other than 1.

2.13.2 Primitive λ-roots

Carmichael defined a generalization of primitive roots as follows using his function. As
you will see, this section generalizes everything related to primitive roots.

Definition 2.13.15 (Primitive λ-root). Let a and n be coprime positive integers. If

ordn (a) = λ(n), then a is a primitive λ-root modulo n. That is, aλ(n) is the smallest
power of a which is congruent to 1 modulo n.

Definition 2.13.16. Let n be a positive integer. Define

ϕ(n)
ξ(n) = .
λ(n)

(Read ξ as “ksi”). According to Corollary , ξ(n) is an integer.

Proposition 2.13.17. There is a primitive root (defined in the previous section) mod-
ulo n if and only if ξ(n) = 1. Carmichael calls a primitive root a ϕ-primitive root, and
they are in fact a special case of λ-primitive roots.

Now, the existence of primitive root is generalized to the following theorem, cited
from Carmichael’s paper [ ].
CHAPTER 2. MODULAR ARITHMETIC 141

Theorem 2.13.18 (Carmicahel). For any positive integer n, the congruence equation

xλ(n) ≡ 1 (mod n)

has a solution a which is a primitive λ-root, and for any such a, there are ϕ(λ(n))
primitive roots congruent to powers of a.

We can show that this theorem is true in a similar fashion to what we did in last
section, and we leave it as an exercise.
As we mentioned earlier in Proposition , ϕ(n) is always even for n > 2. As it
turns out, λ and ϕ share more common properties than we think. The following next
two problems, which were taken from [ ], demonstrate this.

Problem 2.13.19. For any integer n ≥ 1, either ξ(n) = 1 or ξ(n) is even.

Hint. Use the formula for λ(n) in Theorem .

Problem 2.13.20. If λ(n) > 2, the number of primitive λ-roots modulo n is even.

The next theorem generalizes Theorem , which itself was a generalization to

Wilson’s theorem.

Theorem 2.13.21. Let n be a positive integer such that λ(n) > 2. Also, suppose that
g is a primitive λ-root modulo n. The product of primitive λ-roots of n is congtuent to
1 modulo n.

Proof. Since λ(n) > 2, we can easily argue that it must be even. If g is a primitive
λ-root modulo n, all the primitive λ-roots are

{g e1 , g e2 , . . . , g ek },

where ei (for 1 ≤ i ≤ k) are all (distinct) positive integers with (ei , λ(n)) = 1. Also,
note that we can pair them up since λ(n) is even if n > 2. In fact, we can pair g ei with
g λ(n)−ei for all i. Then,

g e1 · g e2 · · · g ek ≡ g λ(n) · · · g λ(n) ≡ 1 (mod n).

Corollary 2.13.22. For any n, there are ϕ(λ(n)) primitive λ-roots modulo n.

2.14 Pseudoprimes
In general, a Pseudoprime is an integer which shares a common property with all prime
numbers but is not actually a prime. Pseudoprimes are classified according to which
property of primes they satisfy. We will investigate a few types of pseudoprimes in this
section.
142 2.14. PSEUDOPRIMES

2.14.1 Fermat Pseudoprimes, Carmichael Numbers

The most important class of pseudoprimes are Fermat pseudoprimes which come from
Fermat’s little theorem.
Definition 2.14.1 (Fermat Pseudoprime to Base a). For an integer a > 1, if a
composite integer n satisfies an−1 ≡ 1 (mod n), then n is said to be a Fermat pseudo-
prime to base a and is denoted by psp(a).
Suppose a > 1 is an integer. It can be shown that the number of Fermat pseudo-
primes to base a is small compared to the number of primes. Therefore, any number n
that passes Fermat’s little theorem (i.e., an−1 ≡ 1 (mod n)) could be considered to be
probably a prime and that is why it is called pseudoprime.
Example. Fermat pseudoprime to base 2 are called Poulet numbers. 341 = 11 × 31 is
the smallest Poulet number. The reason is that

68
2340 ≡ 25 ≡ (32)68 ≡ 168 ≡ 1 (mod 31),
and

34
2340 ≡ 210 ≡ (1024)34 ≡ (1)34 ≡ 1 (mod 11),
which yields 2340 ≡ 1 (mod 341).
Theorem 2.14.2. For any integer a > 1, there are infinitely many Fermat pseudoprime
to base a.
Proof. Let p ≥ 3 be any prime number such that p ∤ a2 − 1. We show that
a2p − 1
n=
a2 − 1
is a Fermat pseudoprime to base a. First, n is composite because
ap − 1 ap + 1
n= · .
a−1 a+1
By Fermat’s little theorem, a2p ≡ a2 (mod p) and therefore p | a2p − a2 . Since p does
not divide a2 − 1, it divides
a2p − a2
n−1= = a2p−2 + a2p−4 + · · · + a4 + a2 ,
a2 − 1
which is an even integer. We can now deduce that 2p | n − 1 because p is odd. Now,
a2p − 1 = n (a2 − 1) which means a2p ≡ 1 (mod n). Thus an−1 ≡ 1 (mod n) and n is a
Fermat pseudoprime to base a.
When you first encountered Fermat’s little theorem, you may have wondered if the
reverse is true. That is, if an−1 ≡ 1 (mod n) for all integers a coprime to n, then n
is prime or not. If you try some examples by hand, you may convince yourself that
n must be a prime in order to hold the condition true. Unfortunately, that is not the
case. There are infinitely many composite integers n with the given property and the
are called Carmichael numbers.
CHAPTER 2. MODULAR ARITHMETIC 143

Note. Do not be mistaken by this simple statement. It took a long time for number
theorists to prove that there indeed exist infinitely many Carmichael numbers.
With the above definition of Fermat pseudoprimes, we may provide another defini-
tion for Carmichael numbers.
Definition 2.14.3 (Carmichael Number). Let n be a positive integer. If n is a
Fermat pseudoprime for all values of a that are coprime to n, then it is a Carmichael
number or Fermat pseudoprime (and sometimes absolute Fermat pseudoprime).
The first few Carmichael numbers are 561, 1105, 1729, . . . .
The following theorem shows us a way to determine if an integer is a Carmichael
number.
Theorem 2.14.4 (Korselt’s Criterion). A positive integer n is a Carmichael number
if and only if all of the following conditions meet.
i. n is composite.

ii. n is squarefree.

iii. For any prime p|n, we also have p − 1 | n − 1.

Proof. Let’s prove the second proposition first. For the sake of contradiction, let p be
a prime factor of n such that p2 divides n. Then for all a, p2 | n | an − a. Choose a = p
and we have p2 | pn − p or p2 | p, which is impossible. So, n is square-free.
Now we will prove the third one. To prove this, we will use a classical technique.
Let p be a prime divisor of n. Since an ≡ a (mod n), we can say an ≡ a (mod p) for
all a. Choose a so that a⊥p. Then p divides an − a = a(an−1 − 1), thus p | an−1 − 1.
Also from Fermat’s little theorem, p | ap−1 − 1.
Here is the crucial part. From Theorem , we know that there is a primitive
root for all primes p, i.e., there is a positive integer g with ordp (g) = p − 1. For that g,

g n−1 ≡ 1 (mod p), and

g p−1 ≡ 1 (mod p).

Since p − 1 is the order, by Theorem , p − 1 | n − 1 must hold.

Note. The connection of Carmichael numbers with Carmichael function is obvious. We
could just do it in the following way:
It is evident that we need λ(n) | n − 1. For n > 2, λ(n) is even so n − 1 is even too.
This means n is odd. Next, λ(n) is coprime to n, so n is square-free.
Definition 2.14.5 (Euler Pseudoprime to Base a). For an integer a > 1, if an odd
composite integer n which is coprime to a satisfies the congruence relation
a
≡ a(n−1)/2 (mod n),
n


where na is the Jacobi symbol, then n is called an Euler pseudoprime to base a and
denoted by epsp(a).
144 2.14. PSEUDOPRIMES

Corollary 2.14.6. Let a > 1 be an odd integer. Then every Euler pseudoprime to
base a is also a Fermat pseudoprime to base a.

There are infinitely many epsp(a) for any integer a > 1. Actually, even more is
true: there exist infinitely many Euler pseudoprimes to base a which are product of k
distinct primes and are congruent to 1 modulo d, where k, d ≥ 2 are arbitrary integers.
You may wonder if there exist absolute Euler pseudoprimes, numbers which are
Euler pseudoprimes to every base coprime to themselves. The answer is negative. In
fact, it can be shown that an odd composite integer n can be Euler pseudoprime for at
most 12 ϕ(n) bases a, where 1 < a < n and (a, n) = 1. The proof needs some algebraic
background and we do not include it in this book.
Example. 121 is an epsp(3). To see why, note that
   2
3 3
= = 1,
121 11

by the definition of Jacobi symbol (Definition ). Now,

12
360 = 35 = (243)12 ≡ 112 ≡ 1 (mod 121).

As the last class of pseudoprimes, we mention strong pseudoprimes.

Definition 2.14.7 (Strong Pseudoprime to Base a). Let n = 2s d + 1 where s and

d are positive integers and d is odd. Also, let a > 1 be a positive integer coprime to n
such that one of the following conditions holds:

ad ≡ 1 (mod n), or
2r d
a ≡ −1 (mod n), for some integer 0 ≤ r < s.

Then n is called a strong pseudoprime to base a and is denoted by spsp(a).

It can be proved that every spsp(a) is also a epsp(a) (and hence a psp(a)). There
exist infinitely many strong pseudoprimes to base a for every integer a ≥ 1. We show
a special case of this where a = 2 in the following proposition.

Proposition 2.14.8. There are infinitely many strong pseudoprimes to base 2.

Proof. If n is a Fermat pseudoprime to base 2, then 2n−1 ≡ 1 (mod n) and so 2n−1 −1 =

nk for some integer k. Choose m = 2n −1. We will show that m is a strong pseudoprime
to base 2. To proceed, notice that m − 1 = 2n − 2 = 2 (2n−1 − 1) and 2n−1 − 1 is an
n−1
odd integer. So it suffices to show that 22 −1 ≡ 1 (mod m). Now,
n−1 −1
22 = 2nk = (2n )k ≡ 1k ≡ 1 (mod m).

The proof is complete.

CHAPTER 2. MODULAR ARITHMETIC 145

2.15 Using Congruence in Diophantine Equa-

tions
Diophantine equations are an especial kind of equations which allow solutions only in
integers. They have been studied for a really long time. The name is taken after the
mathematician Diophantus of Alexandria. We have avoided discussing such equations
in this book because this area is too huge for us to include right now, and for the same
reason we had to ax a lot of topics. However, it is compulsory that we discuss how
to use modular arithmetic to solve some particular Diophantine equations. And even
if the whole equation can not be solved, we can say a lot about the solutions using
modular properties.

2.15.1 Some Useful Properties

There are some modular arithmetic properties that usually come handy. But before
showing them, we intend to pose a question.
Question 2.15.1. Find two positive integer whose sum of squares is 123.
Since there does not exist many squares below 123, you may try to do it by hand.
And after exhausting all possible cases, you must conclude there are no such integers.
But if you are clever, you don’t have to go through trial and error. Let’s write a2 + b2 =
123 and notice the following. Exactly one of a or b must be odd since 123 is odd.
Without loss of generality, assume a is even (you can take b if you want). Then b is
odd, and we know b2 ≡ 1 (mod 4). Thus, a2 + b2 ≡ 1 (mod 4), whereas 123 ≡ 3
(mod 4). This is a straight contradiction implying there are no such positive integers a
and b. The idea seems simple enough, yet powerful to be of great use.
For reaching such a contradiction (it is often the case, Diophantine equations usually
do not have any solutions), we use some common facts. The main idea is the same:
find a proper n so that the two sides of the equation leave different remainders modulo
n.
You might ask what happens if the equation actually does have a solution in integers?
Let us explain this with an example. Suppose that you are given the simple linear
Diophantine equation 6x + 5y = 82 and you want to solve it over non-negative integers.
Let’s solve this problem by trial and error. First, notice that x ≤ 13 (otherwise 6x
would exceed 82). We can draw a table to find the solutions.

x 0 1 2 3 4 5 6
y none none 14 none none none none
x 7 8 9 10 11 12 13
y 8 none none none none 2 none

Table 2.3: Solving 6x + 5y = 82 by trial and error.

As seen in Table , we need to do 13 calculations to find the solutions

(x, y) = (2, 14), (7, 8), (12, 2).
146 2.15. USING CONGRUENCE IN DIOPHANTINE EQUATIONS

Now, consider the same linear equation 6x + 5y = 82 again. We are going to solve
it using modular arithmetic this time. Take modulo 5 from both sides of the equation.
The left side would be x while the right side is 2, giving us the relation x ≡ 2 (mod 5).
Although this does not give us the solution directly, it helps us find the solutions much
faster. Just notice that we already know x must be less than or equal to 13, and it must
have a remainder of 2 when divided by 5. The only choices for x then are 2, 7, and 12.
We can now plug these values of x into the equation and find the solutions with only
three calculations (instead of thirteen).
Sometimes we need to use some theorems such as Fermat’s little theorem or Wilson’s
theorem and pair them up with some modular arithmetic. Here are some highly useful
congruences:
Theorem 2.15.2. Let x be an integer (not necessarily positive). Then
x2 ≡ 0, 1 (mod 3),
x2 ≡ 0, 1 (mod 4),
x2 ≡ 0, 1, 4 (mod 8),
x2 ≡ 0, 1, 4, 9 (mod 16),
x3 ≡ 0, ±1 (mod 7),
x3 ≡ 0, ±1 (mod 9),
x4 ≡ 0, 1 (mod 16),
x4 ≡ 0, ±1, ±4 (mod 17),
x5 ≡ 0, ±1 (mod 11),
x6 ≡ 0, 1, 4 (mod 13).
Most of above congruences can be proved easily. Some are direct consequence of
Fermat’s or Euler’s theorem. Or you can just consider the complete set of residue of
the modulus and then investigate their powers. Whatever the case, we will leave the
proofs as exercises. Sometimes you may notice that Fermat’s little theorem or Euler’s
theorem is disguised in the equation.
Problem 2.15.3. The sum of two squares is divisible by 3. Prove that both of them
are divisible by 3.
Solution. Assume that a2 + b2 is divisible by 3. If a is divisible by 3, so must be b. So,
take a not divisible by 3. Then, from the properties above, we have a2 ≡ 1 (mod 3) and
b2 ≡ 1 (mod 3). And this immediately gives us a contradiction that a2 + b2 ≡ 1 + 1 ≡ 2
(mod 3).
Remark. We could just use Theorem which shows that every prime factor of
a + b is of the form 4k + 1 if a and b are coprime.
2 2

Problem 2.15.4. Show that there are no integers a, b, c for which a2 + b2 − 8c = 6.

Solution. The term −8c guides us to choose the right modulo. Consider the equation
modulo 8. We have a2 + b2 ≡ 6 (mod 8). By Theorem , a2 ≡ 0, 1, or 4 (mod 8).
Now you may check the possible combinations to see that a2 + b2 ≡ 6 (mod 8) is
impossible.
CHAPTER 2. MODULAR ARITHMETIC 147

Problem 2.15.5. Solve the Diophantine equation x4 − 6x2 + 1 = 7 · 2y in integers.

Solution. There are no solutions for y < 0. So assume y ≥ 0. Add 8 to both sides of
the equation to get

(x2 − 3)2 = 7 · 2y + 8.

Note that if y ≥ 3, the right hand side of above equation is divisible by 8. So taking
modulo 8 may seem reasonable. However, it leads to (x2 − 3)2 ≡ 0 (mod 8) and no
further results are included. We should look for another modulo. If y ≥ 4, then the
right hand side is congruent to 8 modulo 16. However, the left hand side, (x2 − 3)2 is
a square and so it’s 0, 1, 4, or 9 modulo 16. The only left cases are y = 0, 1, 2, and 3
which imply no solutions. Hence, no solutions at all.
Let’s see another problem in which we will also see an application of Fermat’s method
of infinite descent. This is a technique for solving Diophantine equations but we briefly
use the idea here.
Problem 2.15.6. Find all integer solutions to the equation

x2 + y 2 = 7(z 2 + t2 ).

Solution. First of all, using the same approach as in previous problem, we can prove
that 7 divides both x and y. Let x = 7a and y = 7b and substitute them in the equation.
After dividing by 7,

z 2 + t2 = 7(a2 + b2 ).

Note that this equation looks like the original one. However, z and t in the latter
equation are strictly smaller than x and y in the original equation. We can continue
this process by noting the fact that z and t are divisible by 7. So, assume that z = 7u
and t = 7v and rewrite the equation as

u2 + v 2 = 7(a2 + b2 ).

This process can be done infinitely many times. Thus, we get that x and y are divisible
by 7i for all positive integers i, which is not possible. So, the equation does not have
any solutions. The process of finding new equations similar to the original one is called
the method of infinite descent.
Problem 2.15.7. Show that the following equation does not have any solutions in
positive integers:

5n − 7m = 1374.

Solution. The most important thing in solving a Diophantine equation is to take the
right modulo. In this case, it’s obvious that the easiest mods to take are 5 and 7. Let’s
take modulo 5 from both sides of the equation. Since 7m ≡ 2m (mod 5),

−2m ≡ −1 (mod 5) =⇒ 2m ≡ 1 (mod 5).

148 2.15. USING CONGRUENCE IN DIOPHANTINE EQUATIONS

4
Since ord5 (2) = 4, we have 4 | m. Let m = 4k for some integer k. So 7m = 74k = 7k .
This reminds us of the fact that x2 (and thus x4 ) is either 0 or 1 modulo 4. So, 7m ≡ 1
(mod 4). Taking modulo 4 from the original equation, we get
5n − 74k ≡ 2 (mod 4) =⇒ 1 − 1 ≡ 2 (mod 4),
which is a contradiction. Thus, there are no solutions.
Problem 2.15.8 (Kazakhstan 2016). Solve in positive integers the equation
n! + 102014 = m4
Solution. You can usually use modular arithmetic to solve the problem when there
is a factorial term in the given equation. The interesting property of n! is that it is
divisible by all integers less than or equal to n. In this problem, if we find the right
modulo k, we can assume n ≥ k and take modulo k from the equation (we will check
the cases when n < k later). It will be 102014 ≡ m4 (mod k). As said before, we guess
the equation does not have any solutions. So, we are searching for a modulo k for
which m4 cannot be congruent to 102014 . We should first try the simplest values for
k, i.e., values of k for which m4 can have a few values. For k = 16, we have m4 ≡ 0
(mod 16), no contradiction. For k = 17, we have m4 ≡ 8 (mod 17), which is impossible
because m4 can only have the values 0, ±1, or ±4 modulo 17. We have found our
desired contradiction, and we just have to check the values of n < 17. This is easy.
Obviously, n! + 102014 is bigger than 102014 . However, the smallest perfect square bigger
than 102014 is
2
101007 + 1 = 102014 + 2 · 101007 + 1,
which is way bigger than 102014 + 16!. So, no solutions in this case as well.
Problem 2.15.9. Prove that the equation x2 + 5 = y 3 has no integer solutions.
Solution. Taking modulo 4, since x2 + 5 is congruent to either 1 or 2 modulo 5, but y 3
is never congruent to 2 modulo 4, we have x2 + 5 ≡ y 3 ≡ 1 (mod 4), and so x is even,
y ≡ 1 (mod 4). Rewrite the equation as
x2 + 4 = (y − 1)(y 2 + y + 1).
Note that since y ≡ 1 (mod 4), we have y 2 + y + 1 ≡ 3 (mod 4). According to theorem
( ), we know that every number congruent to 3 modulo 4 has a prime divisor also
congruent to 3 modulo 4. Let p ≡ 3 (mod 4) be that prime divisor of y 2 + y + 1. Then
x2 + 4 ≡ 0 (mod p)
p−1
If we raise both sides of the congruence equation x2 ≡ −4 (mod p) to the power of 2
(which is an odd integer since p ≡ 3 (mod 4)), we have

p−1 p−1
x2 2 ≡ − (4) 2 (mod p),
or, by Fermat’s little theorem,
1 ≡ xp−1 ≡ −4p−1 ≡ −1 (mod p).
This is the contradiction we were looking for and the equation does not have integer
solutions.
CHAPTER 2. MODULAR ARITHMETIC 149

Note. The idea of taking a prime p ≡ 3 (mod 4) of a number n ≡ 3 (mod 4) comes

handy in solving Diophantine equations pretty a lot. Keep it in mind.

Problem 2.15.10 (Romania JBMO TST 2015). Solve in nonnegative integers the
equation
21x + 4y = z 2 .

Solution. First, let us consider the case x = 0. Then z 2 − 1 = (z − 1)(z + 1) = 4y . If

z − 1 = 1 and z + 1 = 4y , we have no solutions. Otherwise, both z − 1 and z + 1 should
be perfect powers of 4, which is impossible.
We can show, in a similar way, that the case y = 0 gives no solutions as well. So,
suppose that x and y are positive integers.
Rewrite the original equation as

3x · 7x = (z − 2y )(z + 2y ).

There are a few cases to check:

• If z − 2y = 1, then z + 2y = z − 2y + 2y+1 = 1 + 2y+1 = 21x . This implies

2y+1 = 21x − 1. But the right hand side of the latter equation is divisible by 20,
contradiction. So no solutions in this case.

• If both z − 2y and z + 2y are divisible by 21, then 21|(z − 2y , z + 2y ). This is

impossible because if d = (z − 2y , z + 2y ), then d|(z + 2y ) − (z − 2y ) = 2y+1 , which
means d is a power of 2.

• If z − 2y = 3x and z + 2y = 7x , then

(2.23) 7x − 3x = 2y+1 .

y = 1 gives the solution (x, y, z) = (1, 1, 5). Assume y ≥ 2. Take modulo 8 from
equation ( ). 2y+1 is divisible by 8 and so 7x − 3x ≡ 0 (mod 8). But this does
not happen for any x (just consider two cases when x is even or odd). So the only
solution in this case is (x, y, z) = (1, 1, 5).

Note that we have used the fact that z + 2y > z − 2y to omit some cases (like when
z − 2y = 21x and z + 2y = 1). So, (x, y, z) = (1, 1, 5) is the only solution to the given
equation.
150 2.16. EXERCISES

2.16 Exercises
Problem 2.16.1. Consider the following progression:
1
u0 = ,
2
un
un+1 = for n = 1, 2, 3, . . .
3 − 2un
Let a be a real number. We define the series {wn } as
un
wn = for n = 1, 2, 3, . . .
un + a
Find all values of a such that wn is a geometric progression.
Problem 2.16.2. Let p be an odd prime number and consider the following sequence
of integers: a1 , a2 . . . ap−1 , ap . Prove that this sequence is an arithmetic progression if
and only if there exists a partition of the set of natural numbers N into p disjoint sets
A1 , A2 . . ., Ap−1 , Ap such that the sets {ai + n|n ∈ Ai } (for i = 1, 2, . . . , p) are identical.
Problem 2.16.3. Assume we have 15 prime numbers which are elements of some
arithmetic sequence with common difference d. Prove that d > 30000.
Problem 2.16.4 (Vietnam Pre-Olympiad 2012). Determine all values of n for
which there exists a permutation (a1 , a2 , a3 , . . . , an ) of (1, 2, 3, . . . , n) such that

{a1 , a1 a2 , a1 a2 a3 , . . . , a1 a2 . . . an }

is a complete residue system modulo n.

Problem 2.16.5. Prove that for any two positive integers m and n, there exists a
positive integer x, such that
 x
2 ≡ 1999 (mod 3m ),
2x ≡ 2009 (mod 5n ).

Problem 2.16.6. Let f (x) = 5x13 + 13x5 + 9ax. Find the least positive integer a such
that 65 divides f (x) for every integer x.
Problem 2.16.7 (Romanian Mathematical Olympiad 1994). Find the remainder
when 21990 is divided by 1990.
Problem 2.16.8. Prove that a2 + b5 = 201517 has no solutions in Z.
Problem 2.16.9 (Middle European Mathematical Olympiad 2009). Determine
all integers k ≥ 2 such that for all pairs (m, n) of different positive integers not greater
than k, the number nn−1 − mm−1 is not divisible by k.
Problem 2.16.10 (ELMO 2000). Let a be a positive integer and let p be a prime.
Prove that there exists an integer m such that
m
mm ≡ a (mod p).
CHAPTER 2. MODULAR ARITHMETIC 151

Problem 2.16.11. Find all pairs of prime numbers (p, q) for which

7pq 2 + p = q 3 + 43p3 + 1.

Problem 2.16.12 (IMO 1996). The positive integers a and b are such that the
numbers 15a + 16b and 16a − 15b are both squares of positive integers. What is the
least possible value that can be taken on by the smaller of these two squares?
Problem 2.16.13. 2017 prime numbers p1 , . . . , p2017 are given. Prove that
Y p
(pi j − ppj i )
i<j

is divisible by 5777.
Problem 2.16.14 (Ukraine 2014). Find all pairs of prime numbers (p, q) that satisfy
the equation
3pq − 2q p−1 = 19.
Problem 2.16.15. Let p be an odd prime and let ω be the pth root of unity (that is,
ω is some complex number such that ω p = 1). Let
X X
X= ω i and Y = ωj ,

where i in the first sum runs through quadratic residues and j in the second sum runs
over quadratic non-residues modulo p and 0 < i, j < p. Prove that XY is an integer.
Problem 2.16.16. Let p > 2 be a prime number. Prove that in the set
√
{1, 2, . . . , ⌊ p⌋ + 1} ,

there exists an element which is not a quadratic residue mod p.

Problem 2.16.17 (APMO 2014). Find all positive integers n such that for any
integer k there exists an integer a for which a3 + a − k is divisible by n.
Problem 2.16.18. 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.
Problem 2.16.19. 16 is an eighth power modulo every prime.
Problem 2.16.20. Let m and n be integers greater than 1 with n odd. Suppose that
n is a quadratic residue modulo p for any sufficiently large prime number p ≡ −1
(mod 2m ). Prove that n is a perfect square.
Problem 2.16.21. Form the infinite graph A by taking the set of primes p congruent
to 1 (mod 4), and connecting p and q if they are quadratic residues modulo each other.
Do the same for a graph B with the primes 1 (mod 8). Show A and B are isomorphic
to each other.
Problem 2.16.22. Find all positive integers n that are quadratic residues modulo all
primes greater than n.
152 2.16. EXERCISES

Problem 2.16.23. Let k be an even positive integer and k ≥ 3. Define

2k − 1
n= .
3
Find all k such that (−1) is a quadratic residue modulo n.
Problem 2.16.24. Let n and k be given positive integers. Then prove that
• there are infinitely many prime numbers p such that ±1, ±2, ±3, . . . , ±n are
quadratic residue of p, and
• there infinitely many prime numbers p > n such that ±i/j are k th -power residue
of p, where i and j are integers between 1 and n (inclusive).
Problem 2.16.25. Let p > 5 be a prime number and
A = {b1 , b2 , . . . , b p−1 }
2

be the set of all quadratic residues modulo p, excluding zero. Prove that there doesn’t
exist positive integers a and c satisfying (ac, p) = 1 such that set
B = {ab1 + c, ab2 + c, . . . , ab p−1 + c}
2

and A are disjoint modulo p.

Problem 2.16.26. Let p be a prime number such that p = a2 + 5b2 , where a and b are
positive integers and a is odd. Prove that a is a quadratic residue modulo p if and only
if p ≡ 1 (mod 5).
Problem 2.16.27. Find all primes p such that 5, 7, and −7 are quadratic residues
modulo p.
Problem 2.16.28. Prove that there are no positive integers k such that for the first k
odd prime numbers p1 , p2 , . . . , pk , there are a, n ∈ Z+ (n > 1) satisfying
p1 p2 · · · pk = an + 1.
Problem 2.16.29. Find all the pairs of positive integers (x, p) such that p is a prime,
x ≤ 2p, and xp−1 is a divisor of (p − 1)x + 1.
Problem 2.16.30. Determine all positive integers n such that 3n + 1 is divisible by n2 .
Problem 2.16.31. Find a condition for a ∈ N such that there are infinitely many
natural x such that a2x ≡ a2a (mod p) implies ax ≡ −aa (mod p), where p is any
positive integer.
Problem 2.16.32. Show that n does not divide 2n − 1 for n > 1.
Problem 2.16.33 (China 2006). Find all positive integer pairs (a, n) such that
(a + 1)n − an
n
is an integer.
CHAPTER 2. MODULAR ARITHMETIC 153

Problem 2.16.34. Prove that for any integer n ≥ 2 the number

3n − 2n
n
is not an integer.

Problem 2.16.35 (China 2009). Find all the pairs of prime numbers (p, q) such that

pq | 5p + 5q .

Problem 2.16.36. Prove that any two different Fermat numbers are relatively prime
with each other.
n
Note. The nth Fermat number is Fn = 22 + 1.

Problem 2.16.37. Prove that for all positive integers n, gcd(n, Fn ) = 1, where Fn is
the nth Fermat number.

Problem 2.16.38. Let a and b be coprime integers and let d be an odd prime that
k k
divides a2 + b2 . Prove that d − 1 is divisible by 2k+1 .

Problem 2.16.39. Prove that if p is a prime, then pp − 1 has a prime factor greater
than p.

Problem 2.16.40.

1. Show that if p is a prime and ordp (a) = 3, then

2
!2
X
j2
a ≡ −3 (mod p).
j=0

2. Show that if p is a prime and ordp (a) = 4, then

3
!2
X 2
aj ≡ 8a (mod p).
j=0

3. Show that if p is a prime and ordp (a) = 6, then

5
X 2
aj ≡ 0 (mod p).
j=0

Problem 2.16.41 (Poland 2016). Let k and n be odd positive integers greater than
1. Prove that if there a exists positive integer a such that k | 2a + 1 and n | 2a − 1, then
there is no positive integer b satisfying k | 2b − 1 and n | 2b + 1.

Problem 2.16.42. Let n > 9 be a positive integer such that gcd(n, 2014) = 1. Show
that if n | 2n + 1, then 27 | n.
154 2.16. EXERCISES

Problem 2.16.43. Find all primes p and q that satisfy

 2
p + 1 | 2003q + 1,
q 2 + 1 | 2003p + 1.

Problem 2.16.44. Prove that there do not exist non-negative integers a, b, and c such
that
c
(2a − 1)(2b − 1) = 22 + 1.

Problem 2.16.45. Find all triples (x, y, z) of positive integers which satisfy the equa-
tion
2x + 1 = z(2y − 1).

Problem 2.16.46. Let p be a prime number of the form 3k + 2 that divides a2 + ab + b2

for two positive integers a and b. Prove that p divides both a and b.

Problem 2.16.47. Prove Wilson’s theorem using primitive roots.

Problem 2.16.48. If p is a prime, show that the product of the primitive roots of p is
congruent to to (−1)ϕ(p−1) modulo p.

Problem 2.16.49. Let g be a primitive root modulo a prime p. Find ordpr (g).

Problem 2.16.50. Prove that if r is a primitive root modulo m, then so is the multi-
plicative inverse of r modulo m.

Problem 2.16.51. Prove that 3 is a primitive root modulo p for any prime p of the
form 2n + 1.

Problem 2.16.52. Suppose q ≡ 1 (mod 4) is a prime, and that p = 2q + 1 is also

prime. Prove that 2 is a primitive root modulo p.

Problem 2.16.53. Find all Fermat primes Fn such that 7 is a primitive root modulo
Fn .
m
Problem 2.16.54. Prove that if Fm = 22 + 1 is a prime with m ≥ 1, then 3 is a
primitive root of Fm .

Problem 2.16.55. For a given prime p > 2 and a positive integer k, let

Sk = 1k + 2k + · · · + (p − 1)k .

Find those values of k for which p | Sk .

Problem 2.16.56. Show that for each odd prime p, there is an integer g such that
1 < g < p and g is a primitive root modulo pn for every positive integer n.

Problem 2.16.57. Show that if p = 8k + 1 is a prime for some positive integer k, then
p|x4 + 1 for some integer x.
CHAPTER 2. MODULAR ARITHMETIC 155

Problem 2.16.58. Let n be a positive integer. Prove that

n ≤ 4λ(n) 2λ(n) − 1 ,

where λ(n) denotes the Carmichael function of n.

Problem 2.16.59. Find λ(1080).

Problem 2.16.60 (RMO 1990). Find the remainder when 21990 is divided by 1990.

Hint. Use Carmichael’s function.

Problem 2.16.61. Given three integers a, b, c satisfying a · b · c = 20152016 . Find the

remainder when we divide A by 24, knowing that

A = 19a2 + 5b2 + 1890c2 .

Problem 2.16.62 (APMO 2006). Let p ≥ 5 be a prime and let r be the number of
ways of placing p checkers on a p × p checkerboard so that not all checkers are in the
same row (but they may all be in the same column). Show that r is divisible by p5 .
Here, we assume that all the checkers are identical.

Problem 2.16.63 (Putnam 1996). Let p be a prime greater than 3. Prove that

 ⌊X 3 ⌋ 
2p

2  p
p  .
i=1
i

Problem 2.16.64. Let a and b be two positive integers satisfying 0 < b ≤ a. Let p be
any prime number. Show that
   
pa a
≡ mod p3 .
pb b

Problem 2.16.65. The sequence an is defined as follows: a1 = 0 and

(4n + 2).n3 3n + 1
an+1 = 4
an + for n = 1, 2, 3, . . .
(n + 1) (n + 1)4

Prove that there are infinitely many positive integers n such that an is an integer.

Hint. Find an explicit formula for an and then try Wolstenholme’s theorem.

Problem 2.16.66. Let p be an odd prime. Define

1 1 1
Hn = 1 + + + ··· +
2 3 n
to be the nth harmonic number for any positive integer n. Prove that p divides the
numerator of both Hp(p−1) and Hp2 −1 .
156 2.16. EXERCISES
3p−5
Problem 2.16.67. Let p be an odd prime number. Define q = 2
and
1 1 1
Sq = + + ··· + .
2·3·4 5·6·7 q(q + 1)(q + 2)
If we write p1 −2Sq as an irreducible fraction, prove that p divides the difference between
numerator and denominator of this fraction.
Problem 2.16.68. Find the largest power of a prime p which divides
 n+1   n 
p p
Sp = n
− n−1 .
p p
Problem 2.16.69. Let p ≥ 5 be a prime. Prove that
p−1 k
X 2 (2p−1 − 1)2
≡− (mod p)
k=1
k2 p2
Problem 2.16.70. Let p ≥ 3 be a prime number and let
p−1  
X (−1)j p − 1 a
= ,
j=1
j j b

where a and b are coprime integers. Prove that p2 | a.

n −k

Problem 2.16.71. Prove that 2k−1 is even for all positive integers n and k such that
2 ≤ k ≤ 2n−1 .
Problem 2.16.72. How many of the following numbers are divisible by 3?
       
200 200 200 200
, , ,..., .
0 1 2 200
Problem 2.16.73. Find all pairs (p, q) prime numbers such that
7p3 − q 3 = 64.
Problem 2.16.74 (BMO 2009). Solve the equation
3x − 5y = z 2
in positive integers.
Problem 2.16.75. Solve the equation 7x = 3y + 4 in integers.
Problem 2.16.76. Solve the equation 2x + 3 = 11y in positive integers.
Problem 2.16.77. Solve the Diophantine equation
2x (1 + 2y ) = 5z − 1
in positive integers.
Hint. Take modulo 16.
Problem 2.16.78 (Putnam 2001). Prove that there are unique positive integers a
and n such that
an+1 − (a + 1)n = 2001.
Bibliography

[1] Andreescu, T., Andrica, D., and Feng, Z. 104 Number Theory Problems: From the
Training of the USA IMO Team. Springer Science & Business Media. 2007.

[2] Cai, T. X. and Granville, A. On the Residues of Binomial Coefficients and Their
Products Modulo Prime Powers. Acta Mathematica Sinica. Volume 18.2 (2002):
pp. 277 − 288.

[3] Cameron, P. J. and Preece, D. A. Notes on primitive lambda-roots. Preprint (2003).

[4] Carlitz, L. A Note on Wolstenholme’s Theorem. The American Mathematical

Monthly. Volume 61.3 (1954). pp. 174 − 176.

[5] Carmichael, R. D. Note on a New Number Theory Function, Bulletin of the Amer-
ican Mathematical Society. Volume 16 (1909 − 10). pp. 232 − 238.

[6] Engel, A. Problem Solving Strategies, Springer (1998).

[7] Granville, A. Arithmetic Properties of Binomial Coefficients I: Binomial Coef-

ficients Modulo Prime Powers. Organic Mathematics (Burnaby, BC, 1995) 20
(1997). pp. 253 − 276.

[8] Herman, J., Kucera, R., and Simsa, J. Equations and Inequalities: Elementary
Problems and Theorems in Algebra and Number Theory. Vol. 1. Springer Science
& Business Media. 2000.

[9] Kokhas, K., Mogileva, A. Amazing Properties of Binomial Coefficients. Summer

Conference International Mathematics Tournament of Towns, Russia, Teberda.
Aug. 2012.

[10] Lee, H. et al. Problems in Elementary Number Theory (PEN), Art of Problem
Solving.

[11] Sierpinski, W. 250 Problems in Elementary Number Theory. American Elsevier

(New York). 1970.

157
Chapter 3

Arithmetic Functions

Contents
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
. . . . . . . . . . . . . . . . . . . . . . . . . 163
. . . . . . . . . . . . . . . 167
. . . . . . . . . . . . . . . . . 169
. . . . . . . . . . . . . . . . . . . . . . . 171
. . . . . . . . . . . . . . . . . 174
. . . . . . . . . . . . . . . . . . . . . . . . 174
. . . . . . . . . . . . . . . . . . . . . . . . . . 176
. . . . . . . . . . . . . 179
. . . . . . . . . . . . 183
. . . . . . . . . . . . 185
. . . . . . . . . . . . . . . . 190
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
. . . . . . . . . . . . . . . . 200
. . . . . . . . . . . . . . . . . . . . 206
. . . . . . . . . . . . . . . . . . . . . . . . . . 214
. . . . . . . . . . . . . . . . . . . . . . . . . 217
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221

In chapter ( ), we encountered cases where we had to iterate through the divisors

of a number. For example, in problem ( ), we were looking for divisors of 17 which
satisfied a certain condition. In the next problem, we checked for divisors of 8. This is
a pattern which encourages us to study the number of divisors in details. It is natural
that we would encounter more such functions repeatedly. Certainly, this calls for a
separate study of such functions. We call them arithmetic functions, that is, functions
which take positive integers as input. Apart form the need to study these functions for
common purpose, when you have dived into number theory, you will find that arithmetic

159
160 3.1. DEFINITIONS

functions form the nucleus of modern number theory (more precisely, analytical number
theory, which is out of our scope). This PDF was created
for Amazon.

3.1 Definitions
Definition 3.1.1 (Arithmetic Function). An arithmetic function f is a function
f : N → C with domain N = {1, 2, . . . , n, . . .}√and codomain C = {a + bi : (a, b) ∈ R2 },
where i is the imaginary unit defined by i = −1.

Remark. The original definition of arithmetic functions states that the output of an
arithmetic function can be any complex number. For example, consider the function
f : N → C with

ne e−n
f (n) = .
en + n + 1 + ln n
With the above definition, f (n) is an arithmetic function although it does not represent
any specific property of n. In this book, however, we are looking for functions which
represent some number theoric property of their input. One example of such a function
would be ϕ(n). As you already know, the Euler’s totient function takes a positive
integer n as its input and gives the number of positive integers less than n and coprime
to n as its output.
This being said, we are interested in integer-valued functions only. In fact, arithmetic
functions are introduced to express some arithmetical properties of positive integers and
it is logical to expect such functions to have integer values. Therefore, we only study
functions with codomain Z in this book.

Here are some examples of arithmetic functions.

Definition 3.1.2 (Unit Function). The unit function is defined as below for all
positive integers n:
  (
1 1 if n = 1,
ε(n) = =
n 0 if n > 1.

The first values of ε(n) are illustrated in figure ( ).

Note. Here, we call ε(n) the unit function because it acts like 1 when multiplying
arithmetic functions. That is, ε is the multiplicative identity of arithmetic functions.
In section ( ), we will discuss the process of finding the result of multiplication of
two arithmetic functions (called the Dirichlet product) and explain why ε(n) has this
property.

Definition 3.1.3 (Identity Function). The identity function, id, takes every positive
integer to itself. In other words, id(n) = n for all positive integers n. You can see the
plot of this function in figure ( ).
CHAPTER 3. ARITHMETIC FUNCTIONS 161

ε(n)

n
1 2 3 4 5

Figure 3.1: The unit function ε(n) for n = 1, 2, . . . , 5.

id(n)

n
1 2 3 4 5

Figure 3.2: The identity function id(n) for n = 1, 2, . . . , 5.

162 3.1. DEFINITIONS

f (n)

n
1 2 3 4 5

Figure 3.3: The constant 4 function (f (n) = 4) for n = 1, 2, . . . , 5.

Definition 3.1.4 (Constant Function). Let c be a fixed positive integer.Consider

the function f with the property that f (n) = c for all positive integers n. We call f
the constant c function. A plot for constant 4 function can be seen in figure ( ).
Definition 3.1.5 (Additive Function). An arithmetic function f is called an additive
function if and only if
(3.1) f (mn) = f (m) + f (n)
for all m, n ∈ N such that m ⊥ n. The condition m ⊥ n is imposed to make it work on
some special cases (we will find out why later). The case where equation ( ) is true
for all m, n ∈ N, f is called a completely additive function.
Definition 3.1.6 (Multiplicative Function). An arithmetic function f is called a
multiplicative function if and only if
(3.2) f (mn) = f (m)f (n)
for all m, n ∈ N such that m ⊥ n. Again, the condition m ⊥ n is imposed to make it
work on some special cases. In the case where equation ( ) is true for all m, n ∈ N, f
is called a completely multiplicative function.
Example. The identity function and the constant 1 function g(n) = 1 are completely
multiplicative because for any two positive integers m and n,
id(mn) = mn
=m·n
= id(m)id(n),

g(mn) = 1
=1·1
= g(m)g(n).
CHAPTER 3. ARITHMETIC FUNCTIONS 163

⌊x⌋

1
x
−2 −1 1 2

−1

−2

Figure 3.4: The floor function ⌊x⌋ for −2 ≤ x ≤ 2.

3.2 Floor and Ceiling

Let us discuss a classical problem.
Problem 3.2.1. How many integers between 10 and 100 are divisible by 7?
First, we count the multiples of 7 less than 100: 7, 14, 21, 28, 35, 42, 49, 56, 63, 70, 77,
84, 91, and 98. We don’t need go any further since it will make the multiple larger than
100. Amongst these multiples, we should not consider 7 since we are asked between 10
and 100. This gives us the intuition behind the problem. We first find the number of
multiples of 7 between 1 and 100, which is 14. Then, subtract the number of multiples
that are less than 10, i.e., those which are between 1 and 9. In this case, only 7 itself
would be such a multiple. Therefore, the answer to the problem is 14 − 1 = 13. In fact,
the problem is converted into two sub-problems:
1. how many integers from 1 to 100 are divisible by 7?

2. how many integers from 1 to 10 are divisible by 7?

These two sub-problems are basically the same. Therefore, we only need to answer
the general question that how many integers in the interval [1 : x] are divisible by n.
The result is actually the quotient of the division x/n (why?). Consequently, rises the
definition of the floor function.
Definition 3.2.2 (Floor Function and Fractional Part). We call the function
⌊x⌋ : R → Z the floor function and for every real x, ⌊x⌋ (read floor of x) is the largest
integer less than or equal to x. Moreover, we denote by {x} the fractional part of x,
which is equal to x − ⌊x⌋.
Example. ⌊3.1415⌋ = 3, ⌊−2⌋ = −2, and ⌊5⌋ = 5. Figure ( ) shows the value of ⌊x⌋
for x between −2 and 2. For positive real numbers, the fractional part is simply the
the non-integer part of the number. For example, the fractional part of 3.14 is 0.14.
164 3.2. FLOOR AND CEILING

However, the definition of fractional part for negative real numbers may be deceptive.
For instance, the fractional part of −3.14 is not 0.14, but

{−3.14} = −3.14 − (−4)

= 0.86.

We can generalize the above problem now. The number of integers between two
natural numbers a and b (inclusive) which are divisible by n is
   
b a−1
− +1
n n

(Can you see why we are using a − 1 in the second fraction?) Here, we have assumed
that b ≥ a.
Note. Floor and ceiling functions are two of the crossover functions in number theory.
If we consider these two functions as arithmetic functions (i.e., functions which act
only on positive integers), they would seem useless (we already know that the floor and
ceiling of an integer is equal to itself). However, when we are working generally with
real numbers, floor and ceiling are pretty useful and pop up in various problems (like
problem ( )).
Proposition 3.2.3 (Properties of Floor Function). For any two reals x and y and
any two integers m and n,
1. x ≥ ⌊x⌋ > x − 1,

2. if n ≤ x, then n ≤ ⌊x⌋,

3. ⌊x + n⌋ = ⌊x⌋ + n,

4. if x < y, then ⌊x⌋ ≤ ⌊y⌋,

5. ⌊x⌋ + ⌊y⌋ ≤ ⌊x + y⌋ ≤ ⌊x⌋ + ⌊y⌋ + 1,

6. ⌊⌊x⌋⌋ = ⌊x⌋,

7. if x is an integer, ⌊x⌋ + ⌊−x⌋ = 0. Otherwise, it equals 1.

Example. The inequality ⌊x⌋ + ⌊y⌋ ≤ ⌊x + y⌋ (as seen in part 5) is sometimes referred
to as the triangle inequality of floor function. You can easily check why this inequality
holds and find the condition in which it becomes an equality. Put x = 3.6 and y = 2.5.
Then, ⌊x⌋ + ⌊y⌋ = 5, which is strictly less than ⌊x + y⌋ = 6. The reason why we have
a strict inequality here is somewhat obvious now: 0.6 and 0.5 add up to 1.1, which is
more than one. You can easily check that only when {x} + {y} is less than one, the
equality case occurs. Otherwise, the inequality is strict.
Proof. 1. Let ⌊x⌋ = k. By definition, k is the greatest integer not exceeding x, so x ≥ k
and x < k + 1.

2. n ≤ x = ⌊x⌋ + {x} < ⌊x⌋ + 1, and since n is an integer, n ≤ ⌊x⌋.

CHAPTER 3. ARITHMETIC FUNCTIONS 165

3. By definition, ⌊x⌋ ≤ x and so ⌊x⌋ + n ≤ x + n. On the other hand, again by

definition, x < ⌊x⌋ + 1. Therefore

⌊x⌋ + n ≤ x + n < ⌊x⌋ + n + 1.

This means that ⌊x⌋ + n is the largest integer less than or equal to x + n and hence
⌊x + n⌋ = ⌊x⌋ + n.

4. Write x < y as ⌊x⌋ + {x} < ⌊y⌋ + {y} and then ⌊x⌋ < ⌊y⌋ + {y} − {x}. Now by
parts 2 and 3,
 
⌊x⌋ ≤ ⌊y⌋ + {y} − {x}
 z <1
}| { 
= ⌊y⌋ + {y} − {x}

= ⌊y⌋.

5. Let ⌊x⌋ = a, ⌊y⌋ = b, {x} = a1 , and {y} = b1 . By 2, a1 , b1 ≥ 0 and so

a + b ≤ a + a1 + b + b1 .

Sine a + b is an integer, applying part 2 we find

⌊x⌋ + ⌊y⌋ = a + b
≤ ⌊a + a1 + b + b1 ⌋
= ⌊x + y⌋.

On the other hand, by part 3 we observe that

⌊x + y⌋ = ⌊a + a1 + b + b1 ⌋
= a + b + ⌊a1 + b1 ⌋
<2
z }| {
= ⌊x⌋ + ⌊y⌋ + ⌊a1 + b1 ⌋
≤ ⌊x⌋ + ⌊y⌋ + 1.

6. ⌊x⌋ is an integer and the floor of every integer is equal to itself.

7. This is immediately implied from definition.

Proposition 3.2.4 (Properties of Fractional Part). For any real x,

1. 0 ≤ {x} < 1,

2. {x + n} = {x},

3. {{x}} = {x},
166 3.2. FLOOR AND CEILING

⌈x⌉

1
x
−2 −1 1 2

−1

−2

Figure 3.5: The ceiling function ⌈x⌉ for −2 ≤ x ≤ 2.

4. if x is an integer, {x} + {−x} is zero, otherwise it equals 1.

Proofs are straightforward and we leave them for the reader as an exercise.
Now, we get to the other way of solving problem ( ). First, we subtracted the
number of multiples less than 10. This time we find out the first multiple of 7 that is
greater than or equal to 10, and the greatest multiple less than or equal to 100 (this
part is same as before). 14 = 7 · 2 is the first multiple of 7 greater than 10. Since
98 = 7 · 14 is the largest multiple of 7 less than 100, the answer to the problem would
be the number of integers between 2 and 14 (inclusive). There are 14 − 2 + 1 = 13 such
integers. Consequently, this makes us define ceiling. Try to make sense how this relates
to the properties of this function.
Definition 3.2.5 (Ceiling Function). We call the function ⌈x⌉ : R → Z the ceiling
function and for every real x, ⌈x⌉ (read ceiling of x) is the smallest integer greater than
or equal to x.
Example. ⌈3.14⌉ = 4 and ⌈4⌉ = 4. Also ⌈−4.1⌉ = −4 whereas ⌈4.1⌉ = 5. As seen
in figure ( ), similar to the floor function, the plot of the ceiling is like a chain of
steps. Every two real numbers which lie between two consecutive integers have the
same ceiling and floor value. Compare the plots of these two functions.
Proposition 3.2.6 (Properties of Ceiling Function). For any two reals x, y and
any two integers m, n,
1. x ≤ ⌈x⌉ < x + 1,

2. if n ≥ x, then n ≥ ⌈x⌉,

3. ⌈x + n⌉ = ⌈x⌉ + n,

4. if x < y, then ⌈x⌉ ≤ ⌈y⌉,

5. ⌈x⌉ + ⌈y⌉ − 1 ≤ ⌈x + y⌉ ≤ ⌈x⌉ + ⌈y⌉,

CHAPTER 3. ARITHMETIC FUNCTIONS 167

6. ⌈⌈x⌉⌉ = ⌈x⌉,

7. if x is an integer, ⌈x⌉ + ⌈−x⌉ = 0. Otherwise, it equals 1,

8. ⌈x⌉ ≥ x ≥ ⌊x⌋.

The proofs are pretty much the same as those of floor function and we do not provide
them here.

Problem 3.2.7. For any two positive integers n and k, the following equation holds
  j k (
n+1 n 1 if k|n + 1,
− =
k k 0 otherwise.
 2 j k (
n+1 n 2 2 n+1
k
− 1 if k|n + 1,
− =
k k 0 otherwise.

Solution. We will prove the first identity. The second one can be proved in a similar
way. Let n + 1 = kq + r for some positive integers q and r such that 0 ≤ r < k. Then,
  j k    
n+1 n kq + r kq + r − 1
− = −
k k k k
 
j r k r−1
= q+ − q+
k k
 j r k  
r−1

= q+ − q+
k k
j r k r − 1
= − .
k k

Now, since r < k, if r 6= 0, both ⌊r/k⌋ and ⌊(r −1)/k⌋ are zero and so is their difference.
However, when r = 0, we have ⌊r/k⌋ = 0 and ⌊(r − 1)/k⌋ = −1 and in this case,
  j k
n+1 n
− = 1.
k k

The proof is complete.

3.2.1 Fractions and Increasing Functions

Theorem 3.2.8. Let x be a real number. Then for every positive integers n and any
integer m,
       
x+n ⌊x⌋ + n x+n ⌈x⌉ + n
= and = .
m m m m

Proof. We only show the equality occurs for floor function. The proof for ceiling func-
tion is almost the same and is left as an exercise to the reader. Let f : R → R be a
function defined by f (x) = (x + n)/m. We want to show that ⌊f (x)⌋ = ⌊f (⌊x⌋)⌋. We
168 3.2. FLOOR AND CEILING

know that ⌊x⌋ ≤ x. If ⌊x⌋ = x (i.e., if x is an integer) there is nothing to prove. So,
suppose that ⌊x⌋ < x. The function f (x) is strictly increasing. That is, if x1 < x2 for
two reals x1 and x2 , then f (x1 ) < f (x2 ). Therefore ⌊x⌋ < x implies f (⌊x⌋) < f (x).
Now by part 4 of proposition ( ),
   
f (⌊x⌋) ≤ f (x) .

We will show that ⌊f (⌊x⌋)⌋ < ⌊f (x)⌋ does not happen. Suppose on the contrary that
it does happen. Let k = ⌊f (x)⌋ and y = km − n. Then,

y+n
f (y) =
m
(km − n) + n
=
m
=k
= ⌊f (x)⌋.

So f (y) = ⌊f (x)⌋ ≤ f (x), which implies y ≤ x (because f is strictly increasing). Also,

y > ⌊x⌋ because otherwise ⌊f (x)⌋ = f (y) ≤ f (⌊x⌋) and by part 2 of proposition ( ),
⌊f (x)⌋ ≤ ⌊f (⌊x⌋)⌋, which is in contradiction with our assumption. This means that we
have found an integer y such that ⌊x⌋ < y ≤ x. This is obviously in contradiction with
the definition of ⌊x⌋ (since x is non-integer). Therefore, ⌊f (x)⌋ = ⌊f (⌊x⌋)⌋ and we are
done.

Corollary 3.2.9. For any given real numbers x, a1 , a2 , . . . , an , we have

$ % % $
j  k x
. . . x/a1 /a2 . . . /an = , and
a1 a2 . . . an
& ' & '
l  m x
. . . x/a1 /a2 . . . /an = .
a1 a2 . . . an

The above theorem can be generalized in this way:

Theorem 3.2.10. Let f be any continuous and strictly increasing function with the
property that if f (x) is an integer, then so is x. Then,
       
f (⌊x⌋) = f (x) and f (⌈x⌉) = f (x) .

Problem 3.2.11. Prove that for every positive real x, the following relations hold
p 
⌊x⌋ = ⌊x⌋, and
p 
⌈x⌉ = ⌈x⌉.
CHAPTER 3. ARITHMETIC FUNCTIONS 169

3.2.2 Power of a Prime in a Number

In a lot of cases, it just happens that we need to calculate the highest power of a prime
p that divides an integer n. We denote this by vp (n). Determining how many zeros
there are at the end of n! is a famous problem. Clearly, the number of zeros at the end
of any number equals the highest power of 10 which divides that number, and this is
an example of where we need to use this function.
Definition 3.2.12. We define vp (x) to be the greatest power in which a prime p divides
x. In particular, if vp (x) = α, then pα | x but pα+1 ∤ x. We also write pα kx, if and only
if vp (x) = α.
Example. The greatest power of 3 that divides 63 is 32 . because 32 = 9 | 63 but
33 = 27 ∤ 63. in particular, 32 k63 or v3 (63) = 2.
Example. If p and q are two different prime numbers, then vp (pα q β ) = α. This can also
be shown as pα kpα q β .
Proposition 3.2.13. For any two positive integers x and y, and any prime p, we have

vp (xy) = vp (x) + vp (y) and vp (x + y) ≥ min {vp (x), vp (y)} .

Proof. The first equation follows from the product rule of exponentiation (as ·at = as+t ).
To make sense the other equation, you can think of a simple example: take x = 9 and
y = 18. Obviously, 33 does not divide any of x or y, but it divides their sum.
Note. vp (0) = ∞ for all primes p.
Theorem 3.2.14 (Legendre’s Theorem). Let p be a prime and n be a positive
integer. Then
∞  
X n
vp (n!) =
i=1
pi

Proof. By definition, n! is the product of first n positive integers. Therefore, by propo-

sition ( ),

vp (n!) = vp (1)vp (2) . . . vp (n).

This means that we have to find the largest power of p in each of integers 1, 2, . . . , n. As
we were finding a solution for problem ( ), we found out the answer to the question
that ”how many integers between 1 and n are divisible by p¿‘ The answer is ⌊n/p⌋, the
first term in the sum. Similarly, there are ⌊n/p2 ⌋ numbers among these n numbers which
are divisible by p2 . It is now clear that vp (n!) is the sum of ⌊n/p⌋, ⌊n/p2 ⌋, ⌊n/p3 ⌋, . . . .
We have chosen the upper bound of infinity since the value these floors will be zero
after somewhere. If you want to write a sum which includes only non-zero terms, the
upper bound would be vp (n) (why?).
There is another way for finding vp (n!) using bases. Before we state the alternative
version of Legendre’s theorem, we need a definition.
170 3.2. FLOOR AND CEILING

Definition 3.2.15. Let n be a positive integer and let p be a prime number. We denote
by sp (n) the sum of the standard base p digits of n. That is, if n = (nk nk−1 . . . n1 n0 )p ,
then sp (n) = nk + nk−1 + · · · + n1 + n0 .

Example. 10 = (20)5 , and so s5 (10) = 2.

Theorem 3.2.16. Let p be a prime number and let n be a positive integer. Then

n − sp (n)
vp (n!) = .
p−1

Proof. Assume that the base p representation of n is (nk nk−1 . . . n1 n0 )p . So n = nk pk +

nk−1 pk−1 + · · · + n1 p + n0 and
   
n nk pk + nk−1 pk−1 + · · · + n1 p + n0
=
pi pi
= nk pk−i + nk−1 pk−i−1 + · · · + ni+1 p + ni ,

for any integer 1 ≤ i ≤ k. Now by Legendre’s theorem,

k  
X n
vp (n!) =
pi
i=1     
n n n
= + 2 + ··· + k
p p p
k−1 k−2
= nk p + nk−1 p + · · · + n2 p + n1
+ nk pk−2 + nk−1 pk−3 + · · · + n2
.. ..
. .
+ nk p + nk−1
+ nk
k
X
= ni (pi−1 + pi−2 + · · · + p + 1)
i=1
k
X pi − 1
= ni
i=1
p−1
Xk k
X
i
ni p − ni
i=0 i=0
=
p−1
n − sp (n)
= .
p−1
CHAPTER 3. ARITHMETIC FUNCTIONS 171

3.2.3 Kummer’s Theorem

It may seem difficult to
find a relation for the largest power of a prime p that divides a
binomial coefficient m n
. In 1852,

Kummer published an article in which he introduced
a very simple way to find vp m n
.
Theorem 3.2.17 (Kummer’s Theorem). Let p be a prime and let m and n be


s to be the number of carries when n is added
positive integers such that n ≤ m. Take
m
to m − n in base p. Then s = vp n .
To make a good sense of this theorem, we provide an example and then move to the
proof.



Example. Let us find v3 2811
. First, find the base-3 representation of 11 and 28 − 11 =
17:

11 = (102)3 , and 17 = (122)3 .

Now we do the columnar addition to see how many carries happen when we add (1001)3
and (122)3 :
1 1 1
102
+ 122
1001


We see that three carries happen. Therefore, Kummer’s theorem tells us that 28 11
is
divisible by 3 but not by 3 . If we calculate the value of this binomial coefficient, we
3 4

obtain
 
28
= 21474180
11
= 22 × 33 × 5 × 7 × 13 × 19 × 23,

which verifies our result.

Let us prove Kummer’s theorem now.
Proof. Let the representation of m, n, and m − n in base p be

m = mk pk + mk−1 pk−1 + · · · + m1 p + m0 ,
n = nk pk + nk−1 pk−1 + · · · + n1 p + n0 ,
m − n = qk pk + qk−1 pk−1 + · · · + q1 p + q0 .

Let’s find the number of carriers in addition of m and m − n in base p. We can write

q0 + n0 = m0 + x1 p,
q1 + n1 + x1 = m1 + x2 p,
..
.
qk + n k + x k = m k ,
172 3.2. FLOOR AND CEILING

where xi denotes the carry at the (i − 1)th digit from the right (it’s either one or zero).
Adding all the above equations, we find

(qk + qk−1 + · · · + q0 ) + (nk + nk−1 + · · · + n0 ) = (mk + mk−1 + · · · + m0 )

+ (p − 1)(xk + xk−1 + · · · + x0 ).

Therefore, the number of carries in the addition is

s = xk + xk−1 + · · · + x0
(qk + qk−1 + · · · + q0 ) + (nk + nk−1 + · · · + n0 ) − (mk + mk−1 + · · · + m0 )
= .
p−1
We will use the fact that
 
m
vp = vp (m!) − vp (n!) − vp ((m − n)!)
n



along with theorem ( ) to show that s = vp m n
. By theorem ( ), we know
that for every positive integer a,
a − sp (a)
vp (a!) = ,
p−1
where sp (a) denotes the sum of the standard base p digits of a. Therefore,
 
m m − sp (m) n − sp (n) (m − n) − sp (m − n)
vp = − −
n p−1 p−1 p−1
m − (mk + mk−1 + · · · + m0 )
=
p−1
n − (nk + nk−1 + · · · + n0 )
−
p−1
(m − n) − (qk + qk−1 + · · · + q0 )
−
p−1
(qk + qk−1 + · · · + q0 ) + (nk + nk−1 + · · · + n0 ) − (mk + mk−1 + · · · + m0 )
=
p−1
= s.

The proof is complete.

Corollary 3.2.18. Let p be a prime and let n be a positive integer. For any positive
integer k such that 0 < k ≤ pn ,
 n 
p
vp = n − vp (k).
k
Proof. When you add pn − k and k in base p, you get a 1 followed by n zeros. Starting
the addition from the rightmost digits and moving to the left, you get a carry as soon
as both digits in the column are not zero. In other words, the first carry happens at
CHAPTER 3. ARITHMETIC FUNCTIONS 173

the first (rightmost) non-zero of k in base p and you will have a carry for all digits
after that. So, we are searching for the rightmost non-zero digit of k in base p. This is
exactly the (vp (k) + 1)th digit (why?). Thus, the number of carries would be

(n + 1) − (vp (k) + 1) = n − vp (k),

which is what we wanted.

Problem 3.2.19.

Let p be a prime and let m, n be positive integers. If pm < n < pm+1 ,
prove that vp nk ≤ m for every positive integer k ≤ n.

Solution. The condition pm < n < pm+1

means that n has m+1 digits when represented
in base p. By Kummer’s theorem, vp nk is the number of carries in addition k+(n−k)
in base p. Obviously, the number of carries must be less than number of digits of n in
base p (why?). The conclusion follows.

We will provide another solution for problem ( ) stated in section ( ).

Problem 3.2.20 (China 2015). Determine all integers k such that there exists in-
finitely many positive integers n satisfying
2n

n+k 6 .
n


Solution. We already know that n + 1 2n n
(see solution of problem ( )). We will
prove that all integers except k = 1 satisfy the given condition. First, let us handle the
case k


= 0. In this case, choose n = 2m for any integer

m ≥ 2. By Kummer’s Theorem,
2n 2n
v2 n = 1. On the other hand, 4|n, thus n 6 n . 
For k 6= 0, 1,

take any m ≥ 3 + log2 |k|, and choose n = 2α −
k. From Kummer’s
theorem, v2 2n n
≤ m − 1, but n + k = 2 m
. That is, n + k 6  2n for our choice of n.
n

n


Problem 3.2.21. Let n be a positive integer. Show that n divides k
for all k such
that 1 ≤ k ≤ n − 1 if and only if n is a prime.

Solution. We will first prove that n must be a power of a prime. Suppose the contrary.
Take
 nany

prime divisor p of n and let v = vp (n). By
Lucas’ theorem, it follows that
n
p 6 pv , which contradicts the assumption that n pv . Thus, n must be a power of a
 
prime.
Write n = pr for some prime p and positive integer r. if r > 1, then from corollary
( ),
 n 
2 
p 6  r−1 ,
p

which is a contradiction. Therefore, r = 1 and n must be a prime.

174 3.3. COMMON ARITHMETIC FUNCTIONS

3.3 Common Arithmetic Functions

3.3.1 Number of Divisors
We first introduced τ (n) in chapter ( ). To be precise, in theorem ( ), we explained
that there are τ (n) positive integer solutions (a, b) to the equation ab = n. For exam-
ple, ab = 12 has τ (12) = 6 solutions (1, 12), (2, 6), (3, 4), (4, 3), (6, 2), (12, 1) in positive
integers.
Definition 3.3.1 (Number-of-divisors Function). Let n be a positive integer. The
number of positive divisors of n is denoted by τ (n). In other words, the number-of-
divisors function is defined as
X
τ (n) = 1.
d|n

What if the number is so huge that we can not compute all the divisors by hand? Or
what if we need to tell a computer how to compute the number of divisors in general?
If you need more motivation to find a way to compute number of divisors easily, consider
the following nice problem!
Problem 3.3.2 (Bangladesh National Mathematical Olympiad, 2010). You
have a regular 2016-gon. You have to choose some points from that polygon so that the
resulting polygon you get from those points is a regular polygon. How many distinct
regular polygons can you construct this way? Order of vertex is not important. And
the polygon will be constructed joining a point with the next one in an order that does
not result in a self intersecting polygon or a polygon which has an angle larger than
180 (convex simple polygon in other words).
Solution. First, try to make sense of what this problem is asking for. This is one of
those problems that makes you think in a nice way while also teaching that theorems
are not everything to solve problems. A very important point to keep in mind here is
that, the order in which we pick the points does not matter.
The next step should be finding out how we should pick those points if we want
to make a regular polygon. And of course, how many point for that matter. Let us
label the points of the 2016-gon as P1 , P2 , . . . , P2016 . So, the regular polygon would be
P1 P2 . . . P2016 . Here, P1 is connected to P2 , P2 to P3 , and so on. However, P2016 would
be connected to P1 to complete the cycle of the polygon.
An important observation: we can always let P1 be the first vertex of the regular
polygon we want to construct. Because no matter where we start from, if we rotate, it
would be identical to the one that can be created starting from P1 . So, P1 is the first
vertex of our regular polygon. Also, we fix the number of points in the new regular
polygon we want to create. If it has k points, all k sides must be equal.
Assume that we want to construct a regular quadrilateral P1 Pi Pj Pk . What can we
say about i, j or k? The length P1 Pi must be equal to Pi Pj , and Pj Pk . This implies
1
Well, you can use a brute force method if you are familiar with programming. But certainly
iterating through integers from 1 to n or a solution of that complexity is not a good idea.
CHAPTER 3. ARITHMETIC FUNCTIONS 175

that j must be i + i. Because if j < i then Pi Pj would be less than P1 Pi (if you are
confused about it, just rotate Pi Pj back to P1 Pi ). Then we also must have k = 3i. This
in turn implies that if there are m points we must have 2016 = mq for some integer
q. That is, m must be a divisor of 2016. Therefore, the number of distinct polygons is
the number of distinct divisors of 2016. But we must consider divisors greater than 2
only. Because we can not create a polygon with one or two points. We have found the
solution, but it would be a lot nicer if we actually knew the value of that.
We will see how to compute number of divisors for a positive integer n. Take n = 12.
If d is a divisor of n, then it is evident that d can not have any prime factor that is
not in n. But the opposite might be true. n might have some prime factor that is
not in d. In this case, a divisor of 12 must have prime factor of 2 and 3 only. Second
observation is: we have n = 22 · 31 . Thus, d can not have the exponent of 2 greater
than 2 or the exponent of 3 greater than 1. If d = 2a 3b then a ≤ 2 and b ≤ 1. The
most interesting part is that, we can actually generate all the divisors this way. Set
a = 0, 1, 2 and b = 0, 1. Consider the sets A = {20 , 21 , 22 } and B = {30 , 31 }. If
we multiply an element of A with an element of B, we get a divisor of 12. And for
each multiplication, we get a distinct divisor each time. And certainly, this can be
generalized. If the prime factorization of n is pe11 pe22 · · · pekk , then the divisors can be
generated multiplying elements from sets A1 = {p01 , . . . , pe11 }, A2 = {p02 , . . . , pe22 }, Ak =
{p0k , . . . , pekk }. Ai has ei + 1 elements because the exponent ranges from 0 to e1 . Then
clearly, the number of divisors of n would be the product of number of elements in all
Ai , that is, (e1 + 1) · · · (ek + 1). However, we will have to prove something even though
it is clear from the example. That is, we will not get a duplicate divisor in this process.
This will be left as an exercise for the readers.
Theorem 3.3.3. Let n be a positive integer with prime factorization n = pe11 pe22 . . . pekk .
Then,
k
Y
(3.3) τ (n) = (1 + ei ).
i=1

Proof. Every positive divisor of n must be of the form

d = pα1 1 pα2 2 . . . pαk k ,

where 0 ≤ ai ≤ ei for i = 1, 2, . . . , k. There are ei + 1 possible values {0, 1, 2, . . . , ei } for

the power of prime pi in d. The number of divisors of n, τ (n), is therefore the product
of e1 + 1, e2 + 1, . . . , and ek + 1. We are done.
√
Theorem 3.3.4. Let n be a positive integer. Prove that τ (n) ≤ 2 n.
Proof. We can find two positive√ integers a and b such that n = ab. At least one of a
and b is less than or equal to n (otherwise their product will be larger than n). This
√
means that τ (n)
2
can take at most ⌊ n⌋ values. So,
√ √
τ (n) ≤ 2⌊ n⌋ ≤ 2 n.
176 3.3. COMMON ARITHMETIC FUNCTIONS

Theorem 3.3.5. τ (n) is odd if and only if n is a perfect square.

Proof. Let n be a positive integer with factorization n = pe11 pe22 . . . pekk and suppose that
τ (n) is odd. This means that all terms in the right hand side of equation ( ) must be
odd. Therefore, ei is even for i = 1, 2, . . . , k. Take ei = 2fi for some integer fi . Then,

n = pe11 pe22 . . . pekk

= p2f 1 2f2
1 p2 . . . pk
2fk

 2
= pf11 pf22 . . . pfkk ,

which is a perfect square. The proof for the converse is easy.

Problem 3.3.6. Show that for any integer n > 1, in the infinite sequence

n, τ (n), τ (τ (n)), τ (τ (τ (n))), . . .

all the terms after a certain point onwards are equal to 2.

Solution.
For n = 2 the statement is obvious, so assume n > 2. Note that τ (n) counts the
number of positive integers in the set {1, 2, . . . , n} that divide n. So, it is at most n.
For n > 2, we know that n − 1 does not divide n, hence τ (n) < n for all n > 2. This
means that the sequence is strictly decreasing as far as its terms are greater than 2.
The proof is complete.

3.3.2 Sum of Divisors

The sum of divisors of a number is another important property of each positive integer.
Definition 3.3.7 (Sum-of-divisors Function). Let n be a positive integer. The
sum of positive divisors of n is denoted by σ(n). In other words, the sum-of-divisors
function is defined as
X
σ(n) = d.
d|n

Note. One might rephrase this definition as: the sum-of-divisors is the summatory
function of id.
Example.
• σ(p2 ) = p2 + p + 1 for any prime p.

• σ(pq) = pq + p + q + 1 = (p + 1)(q + 1), for any two primes p and q.

Note. Just like the number-of-divisors function, σ is not completely multiplicative.
Can you think of an example which shows this?
2
Thanks to Professor Greg Martin for his suggestion about proving τ (n) < n.
3
Read “sigma of n”.
CHAPTER 3. ARITHMETIC FUNCTIONS 177

Theorem 3.3.8. For any positive integer n,

σ(n) X 1
=
n d
d|n

P
Proof. Beginning from σ(n) = d, you can simply realize that the equation can be
d|n
written in the form
Xn
σ(n) =
d
d|n

because when for every divisor d of n, the number n/d also divides n. We can now take
n out of the sum in the right side of above equation and finish the proof.

Note. The idea of employing n/d instead of d, where d is a divisor of n, is a good trick
for solving such problems.

As for the case of τ (n), one can express σ(n) explicitly as well.

Theorem 3.3.9. Let n be a positive integer with prime factorization n = pe11 pe22 . . . pekk .
Then,
k
Y pei +1 − 1
i
σ(n) = .
i=1
p−1

Proof. Later in proposition ( ), we will show that σ is multiplicative. Just accept

it for now. So, it suffices to prove the assertion for the case when n is a power of a
prime. Let n = pα , where p is a prime and α is a positive integer. In this case, the only
divisors of n are 1, p, p2 , . . . , pα . Therefore, their sum, σ(n), is

σ(pα ) = 1 + p + p2 + · · · + pα .

Now, by theorem ( ), we can write the above equation as

α pα+1 − 1
σ(p ) = .
p−1

To finish the proof, consider any positive integer n with factorization n = pe11 pe22 . . . pekk .
Then,

σ(n) = σ(pe11 pe22 . . . pekk )

= σ(pe11 )σ(pe22 ) . . . σ(pekk )
k
Y pei +1 − 1
i
= .
i=1
p−1
178 3.3. COMMON ARITHMETIC FUNCTIONS

We are going to define amicable numbers as an application of the sum of divisors

function.
Definition 3.3.10. Two positive integers are called amicable numbers if the sum of
proper divisors of each of them is equal to the other. Recall that a proper divisor of n
is a divisor of n which is not equal to n. In other words, (m, n) is a pair of amicable
numbers if m = σ(n) − n and n = σ(m) − m.
Example. The smallest pair of amicable numbers is (220, 284). One can easily check
this:
220 = 1 + 2 + 4 + 71 + 142,
284 = 1 + 2 + 4 + 5 + 10 + 11 + 20 + 22 + 4 + 55 + 110.
Amicable numbers have been studied since a very long time ago, and there is evi-
dence that they were known to Pythagoreans (followers of Pythagoras) which originated
in the sixth century BC. Thābit ibn Qurra (826–901) found a formula that generates
amicable numbers (which we will state in the following) [ ]. The Iranian mathematician
Muhammad Baqir Yazdi, who lived in 16th century, found the pair (9363584, 9437056)
of amicable numbers. You can imagaine how difficult it is to find such numbers without
using computers. Over a billion pairs of amicable numbers have been found so far (July
2018).
Theorem 3.3.11 (Thābit ibn Qurra’s Rule). Let n ≥ 2 be a positive integer such
that
p = 3 · 2n−1 − 1,
q = 3 · 2n − 1,
r = 9 · 22n−1 − 1,
are all primes. Then, (2n · p · q, 2n · r) is a pair of amicable numbers.
Proof. Let a = 2n · p · q and b = 2n · r. Since p, q, and r are all odd primes, the
divisors of a are of the form 2α · pβ · q γ , where α, β, γ are integers with 0 ≤ α ≤ n
and β, γ ∈ {0, 1}. Similarly, the divisors of b are of the form 2δ · rǫ , where δ and ǫ are
integers with 0 ≤ δ ≤ n and ǫ ∈ {0, 1}. Hence,
n X
X 1 X
1 n X
X 1 1
X n X
X 1
σ(a) = 2 α pβ q γ = 2 α pβ qγ = 2α pβ (q + 1)
α=0 β=0 γ=0 α=0 β=0 γ=0 α=0 β=0
n
X 1
X n
X n
X
α β α
= 2 (q + 1) p = 2 (p + 1)(q + 1) = (p + 1)(q + 1) 2α
α=0 β=0 α=0 α=0
n+1

2n−1 n+1


= (p + 1)(q + 1) 2 −1 =9·2 2 −1 ,
and so,




σ(a) − a = 9 · 22n−1 2n+1 − 1 − 2n 3 · 2n−1 − 1 (3 · 2n − 1) = 2n 9 · 22n−1 − 1 = b.
We expect the reader to be able to prove σ(b) − b = a in a similar way on their own.
This shows that (a, b) is a pair of amicable numbers.
4
Amicable: having a spirit of friendliness.
CHAPTER 3. ARITHMETIC FUNCTIONS 179

The above theorem is pretty interesting, but you might wonder if it gives us all the
amicable numbers. The answer is negative. You can compute the values of p, q, r, and
a, b to see how large they get even when n is as small as 10. For the above theorem
to work, we require p, q, r to be primes, and determining whether those numbers are
prime is a difficult task for large n.
The mighty Euler found a more general formula for amicable numbers and discovered
61 pairs of amicable numbers at his time [ ]. We state Euler’s generalization of Thābit
ibn Qurra’s rule below, usually called Euler’s rule, and leave the proof to the interested
reader.
Theorem 3.3.12 (Euler’s Rule). Let m and n be positive integers with m < n such
that


p = 2m · 2n−m + 1 − 1,


q = 2n · 2n−m + 1 − 1,

2
r = 2n+m · 2n−m + 1 − 1,

are all primes. Then, (2n · p · q, 2n · r) is a pair of amicable numbers.

3.3.3 Euler’s and Jordan’s Totient Functions

We have already defined Euler’s totient function and used some of its properties to
solve problems. In this section, we will prove those properties and provide some more
features of ϕ. We already know that ϕ(n) is the number of positive integers less than
n which are equal to n. Suppose that n = pα1 1 pα2 2 · · · pkαk is the prime factorization of n.
Then

(3.4) ϕ(n) = p1α1 −1 p2α2 −1 · · · pkαk −1 (p1 − 1) · · · (pk − 1) .

In order to find ϕ(n), we will first find the number of positive integers less than or equal
to n which are not co-prime to n. Then, ϕ(n) would be the difference of this number
and n. Let us consider the simple case n = pq, where p and q are primes. A positive
integer less than or equal to n is not co-prime to n if it is divisible by either p or q.
Let ψ(n) be the number of such positive integers. There are n/p and n/q numbers less
than or equal to n which are divisible by p and q, respectively. So, we might guess that
ψ(n) = n/p + n/q. However, notice that we are counting the number pq in both the
numbers divisible by p and divisible by q. So, the true value for ψ(n) is n/p + n/q − 1.
Thus,

ϕ(n) = n − k(n)
= n − n/p − n/q + 1
= pq − q − p + 1
= (p − 1)(q − 1),

which is in agreement with equation ( ). Take another example when n = p1 , p2 , . . . , pk ,

where p1 , p2 , . . . , pk are different primes. Our first guess for ψ(n) would be n/p1 +
180 3.3. COMMON ARITHMETIC FUNCTIONS

˚n/p2 + · · · + n/pk . However, we are counting the numbers divisible by pi pj (for

1 ≤ i < j ≤ k) twice. So, our next guess for ψ(n) is
X n X n
ψ(n) = − .
1≤i≤k
p i
1≤i<j≤k
p i pj

But this is still not true. For instance, the number p1 p2 p3 is counted in the first sum
and we remove it by subtracting the second sum. So, we would have to add another
sum
X n
1≤i<j<t≤k
pi pj pt

to include all products such as p1 p2 p3 . This process of adding and subtracting sums
is called the inclusion–exclusion principle. We will continue the process until we count
every number less than or equal to n which is not co-prime to n exactly once. Then,
the totient function may be calculated from ϕ(n) = n − ψ(n). The reader is encouraged
to calculate ϕ(n) = (p1 − 1)(p2 − 1) . . . (pk − 1) using the given approach and check that
it is compatible with the formula ( ). The very same approach may be applied to the
general case n = pα1 1 pα2 2 · · · pαk k to finally prove formula ( ). We strongly recommend
the reader try this method for several examples (e.g., n = p2 q or n = p3 q 4 r5 ) and then
prove the whole thing.
Using the same method, we can prove a stronger result.
Theorem 3.3.13 (Generalization of ϕ). Let m and n be positive integers and let
n = pα1 1 pα2 2 · · · pαk k be the prime factorization of n. The number of positive integers less
than or equal to m which are co-prime to n is m − Ψ(m), where
X m X  m  
m

k+1
Ψ(m) = − + · · · + (−1) .
1≤i ≤k
p i1
1≤i <i ≤k
p i1 pi2 p 1 p 2 . . . pk
1 1 2

We already stated (but did not prove) some properties of Euler’s totient function in
proposition ( ). Here, we are going to prove them beside a few more.
Theorem 3.3.14 (Properties of ϕ). Let m and n be positive integers. Then,
(a) ϕ is a multiplicative function. That is, if m⊥n, then

ϕ(mn) = ϕ(m) · ϕ(n).

(b) For all n ≥ 3, ϕ(n) is even.

(c) ϕ is neither increasing, injective, nor surjective.
(d) If n is factorized as n = pα1 1 pα2 2 · · · pαk k , then
    
1 1 1
ϕ(n) = n 1 − 1− ··· 1 −
p1 p2 pk
α1 −1 α2 −1 αk −1
= p1 p2 · · · pk (p1 − 1) · · · (pk − 1) .
5
Notice the indexes under the summation notation
CHAPTER 3. ARITHMETIC FUNCTIONS 181

Proof.
(a) Write the numbers 1, 2, . . . , mn in a table with m rows and n columns as below:
1 m+1 2m + 1 ... (n − 1)m + 1
2 m+2 2m + 2 ... (n − 1)m + 2
.. .. .. .. ..
. . . . .
m 2m 3m ... nm
The numbers in the rth row of the table are of the form km + r, where k =
0, 1, 2, . . . , m − 1. Since (km + r, m) = (r, m), one of these two cases happen: either
all numbers in a row are co-prime to m or all of them are not co-prime to m. As
we are looking for numbers co-prime to mn, which are obviously those co-prime to
both m and n, we consider the rows with all numbers co-prime to m. There are
ϕ(m) such rows. Consider one such row in th table:
r m+r 2m + r ... (n − 1)m + r
The set {0, 1, 2, . . . , n} is a complete residue system modulo n. Since (m, n) = 1,
by proposition ( ) of chapter ( ), the numbers in the above row of table also
form a complete residue system modulo n. This means that all the remainders
0, 1, 2, . . . , n−1 happen when you take the numbers in this row modulo n. Now, how
many numbers in this row are relatively prime to n? The answer is the same number
of integers co-prime to n in the set {0, 1, 2, . . . , n}, which is ϕ(n) by definition (try to
figure out why). Therefore, there are totally ϕ(m)·ϕ(n) numbers in the table which
are co-prime to both m and n, and hence mn. On the other hand, by definition,
there are ϕ(mn) numbers co-prime to mn. The conclusion follows.
(b) ϕ(n) is the number of positive integers k such that k ≤ n and k⊥n. The point is
that if k is co-prime to n, then so is n − k. Therefore, for n ≥ 3, one can match all
numbers co-prime to n in pairs of (k, n − k), which means that ϕ(n) must be even.
(c) Take ϕ(5) > ϕ(6), ϕ(n) = ϕ(2n) for all odd n ≥ 1, and ϕ(n) ≡ 1 (mod 2) as
counterexamples.
(d) We have already provided the sketch of a proof in the beginning of this section.
Here is another proof using multiplicativity of ϕ. Since ϕ(n) is multiplicative, it
suffices to prove the result when n is a power of a prime. Let p be a prime and α ≥ 1
be an integer. The only numbers which are not co-prime to pα between 1, 2, . . . , pα
are multiples of p. How many multiples of p are there among these numbers? The
answer is pα /p = pα−1 (why?). Therefore,
ϕ(pα ) = pα − pα−1
= pα−1 (p − 1).
Now, if we factorize n as pα1 1 pα2 2 · · · pαk k , we can write
ϕ(n) = ϕ(pα1 1 pα2 2 · · · pαk k )
= ϕ(pα1 1 )ϕ(pα2 2 ) · · · ϕ(pαk k )
= p1α1 −1 p2α2 −1 · · · pαk k −1 (p1 − 1) · · · (pk − 1) .
182 3.3. COMMON ARITHMETIC FUNCTIONS

Corollary 3.3.15. If a|b, then ϕ(a)|ϕ(b).

Theorem 3.3.16. Prove that
X
ϕ(d) = n
d|n

holds for all positive integers n

Instead of a full proof, we provide an example and the readers are encouraged to
complete the proof for themselves. Let n = 15 and consider the following 15 fractions:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
, , , , , , , , , , , , , , .
15 15 15 15 15 15 15 15 15 15 15 15 15 15 15
Put all these fractions into lowest terms:
1 2 1 4 1 2 7 8 3 2 11 4 13 14 1
, , , , , , , , , , , , , , .
15 15 5 15 3 5 15 15 5 3 15 5 15 15 1
Obviously, the denominators are all the divisors of 15. Also, there are 15 fractions.
Now, note that only those fractions whose numerator is relatively prime to 15 have
denominator equal to 15. These are
1 2 4 7 8 11 13 14
, , , , , , , ,
15 15 15 15 15 15 15 15
which are ϕ(15) = 8 fractions. Similarly, there are ϕ(5) = 4 fractions
1 2 3 4
, , ,
5 5 5 5
which have denominator equal to 5, and ϕ(3) = 2 fractions
1 2
,
3 3
which have denominator equal to 3. Therefore, we find that the number of fractions is
15 on one hand and ϕ(15) + ϕ(5) + ϕ(3) on the other hand. These two must be equal,
hence ϕ(15) + ϕ(5) + ϕ(3) = 15.
Camille Jordan generalized the definition of Euler’s totient function and introduced
Jordan’s totient functions.
Definition 3.3.17. Let n and k be positive integers. The number of sequences

a1 , a2 , . . . , ak

of positive integers less than or equal to n such that (a1 , a2 , . . . , ak , n) = 1 is called the
k th Jordan’s totient function and is denoted by Jk (n).
Note. Here, (a1 , a2 , . . . , ak , n) is the notation for greatest common divisor of all numbers
a1 , a2 , . . . , ak , and n.
CHAPTER 3. ARITHMETIC FUNCTIONS 183

Example.

• Jk (1) = 1 for all positive integers k.

• Obviously, J1 (n) = ϕ(n). That is, the first Jordan totient function is Euler’s
totient function.

• Jk (p) = pk − 1 because all positive integers less than p are co-prime to p.

• Let us write all pairs (a, b) of positive integers less than 6 such that (a, b, 6) = 1.

(1, 1), (1, 2), (1, 3), (1, 4), (1, 5), (1, 6),
(2, 1), (2, 3), (2, 5),
(3, 1), (3, 2), (3, 4), (3, 5),
(4, 1), (4, 3), (4, 5),
(5, 1), (5, 2), (5, 3), (5, 4), (5, 5), (5, 6),
(6, 1), (6, 5).

There are 24 such pairs. Hence, J2 (6) = 24.

We will state and prove the properties of Jordan’s totient function later in section
( ).

3.4 Characterizing Multiplicative Functions

Multiplicative functions are of high interest. All multiplicative functions possess some
generic properties which we can use to characterize them. It is obvious that if a result
is true for multiplicative functions, it is true for completely multiplicative functions as
well.

Proposition 3.4.1. Let f be a multiplicative function. Let n be a positive integer and

n = pe11 pe22 · · · pekk be its prime factorization. Then,

i. f (1) = 1,
k
Y
ii. f (n) = f (pei i ),
i=1

iii. If f is completely multiplicative, then

k
Y
f (n) = f (pi )ei .
i=1

Proof. For the first part, just set m = 1 in f (mn) = f (m)f (n). We have f (m) =
f (1)f (m) or f (1) = 1 since f (n) > 0.
184 3.4. CHARACTERIZING MULTIPLICATIVE FUNCTIONS

For the second part, we just need to extend it as the definition says.

f (n) = f (pe11 · pe22 · · · pekk )

= f (pe11 ) · f (pe22 · · · pekk )
= f (pe11 ) · f (pe22 ) · f (pe33 · · · pekk )
= f (pe11 )f (pe22 ) · · · f (pekk )
Yk
= f (pei i )
i=1

We can prove the third part in a similar way:

f (pe11 ) = f (p1 · · · p1 )
| {z }
e1 times
= f (p1 ) · · · f (p1 )
| {z }
e1 times
= f (p1 )e1 .
Qk
The same goes for f (pe22 ), . . . , f (pekk ). Thus, f (n) = i=1 f (pi )ei .
Proposition 3.4.2. Let f be a (completely) additive function. If g(n) = cf (n) for some
positive integer constant c, then g is a (completely) multiplicative function.
Proof. The proof is rather straight.

g(mn) = cf (mn)
= cf (m)+f (n)
= cf (m) cf (n)
= g(m)g(n).

And the same goes for completely multiplicative functions as well.

An example for demonstrating this would be the Liouville Function. We will discuss
more about this function later in this chapter. It is noteworthy to mention that Liouville
function plays a huge role in analytic number theory.
Definition 3.4.3 (Big Omega Function). Let n be a positive integer. The number
of prime factors of n (with multiplicity) is denoted by Ω(n).
Definition 3.4.4 (Small Omega Function). Let n be a positive integer. The number
of distinct prime factors of n is denoted by ω(n).
Example. ω(45) = 2 but Ω(45) = 3 since 3 and 5 are the only prime divisors of 45 and
3 is repeated twice (that is, 32 |45). On the other hand, ω(35) = Ω(35) = 2 since the
only prime factors of 35 are 5 and 7, each repeated exactly once.
Definition 3.4.5 (Liouville Function). For a natural number n, the Liouville func-
tion λ(n) is defined as

λ(n) = (−1)Ω(n) .
CHAPTER 3. ARITHMETIC FUNCTIONS 185

Example. λ(1) = 1 and λ(p) = −1 for all primes p. Furthermore, λ(a2 ) = 1 for all
integers a.
Theorem 3.4.6. λ is a completely multiplicative function.
Proof. By proposition ( ), Ω is a completely additive function. The conclusion
follows directly from proposition ( ).

3.5 Dirichlet Product and Möbius Inversion

When working with arithmetic functions, we are mostly interested in the Dirichlet
product of two functions rather than their normal product. As you keep reading this
section, you find out why Dirichlet products are important when studying arithmetic
functions.
Definition 3.5.1 (Dirichlet Product or Dirichlet Convolution). For two arith-
metic functions f and g, we define their Dirichlet product as
X n
h(n) = f (d)g ,
d
d|n

where the sum extends over all positive divisors d of n. We denote this by h = f ∗ g.
In other words,
X
(f ∗ g)(n) = f (a)g(b).
ab=n

Example. Let f be the constant 1 function. Let us find (f ∗ f )(n). By definition,

X n
(f ∗ f )(n) = f (d)f
d
d|n
X
= 1
d|n
= 1| + 1 +{z· · · + 1}
repeated τ (n) times

= τ (n),

where τ (n) is the number of divisors of n. More details about this function will be
explained in section ( ).
Example. This time, we aim to find the convolution of constant 1 function f with the
identity function. Again, by definition,
X n
(id ∗ f )(n) = id(d)f
d
d|n
X
= d
d|n

= σ(n),
186 3.5. DIRICHLET PRODUCT AND MÖBIUS INVERSION

where σ(n) is the sum of divisors of n. If we show the constant 1 function by 1(n), we
can write (id ∗ 1)(n) = σ(n) or simply id ∗ 1 = σ.
Proposition 3.5.2 (Properties of Dirichlet Convolution). Let f, g, and h be any
arithmetic functions. The Dirichlet product is
1. commutative, that is, f ∗ g = g ∗ f ,

2. associative. This means that (f ∗ g) ∗ h = f ∗ (g ∗ h),

3. distributive, which means that f ∗ (g + h) = f ∗ g + f ∗ h.

Furthermore, the unit function ε(n) acts as a unit element for Dirichlet product. That
is, ε ∗ f = f ∗ ε = f for all arithmetic functions f .
Proof. The proof for commutativity
Xof Dirichlet convolution is pretty straightforward
using the definition (f ∗ g)(n) = f (a)g(b). Obviously, since there is no difference
ab=n
between a and b in the latter formula, we can write it as
X X
(f ∗ g)(n) = f (a)g(b) = f (b)g(a)
ab=n ab=n
X
= g(a)f (b)
ab=n
= (g ∗ f )(n).

Let us prove the associativity property now. By definition,

X
((f ∗ g) ∗ h)(n) = (f ∗ g)(d)h(c)
cd=n
!
X X
= f (a)g(b) h(c)
cd=n ab=d
!
X X
= f (a)g(b)h(c)
cd=n ab=d
X
= f (a)g(b)h(c).
abc=n

Again, since there is a symmetry between a, b, and c, we find that the order of f, g, and
h in the expression (f ∗ g) ∗ h does not change the result. Hence, the Dirichlet product
is associative. The proof for distributivity is similar and is left for the reader. It is also
trivial by definition that ε is the unit element of Dirichlet convolution.
We are going define the Mőbius function µ(n), which doesn’t seem to have a natural
definition. As we go on in this chapter and later in chapter on primes, we will find
out why this definition is in fact very natural and it will help us a lot when solving
problems concerning inclusion–exclusion. In other words, the definition of the Mőbius
function is natural in the way that it’s trying to alternate positive and negative signs
in an inclusion–exclusion argument.
CHAPTER 3. ARITHMETIC FUNCTIONS 187

n µ(n) n µ(n) n µ(n) n µ(n)

1 1 26 1 51 1 76 0
2 −1 27 0 52 0 77 1
3 −1 28 0 53 −1 78 −1
4 0 29 −1 54 0 79 −1
5 −1 30 −1 55 1 80 0
6 1 31 −1 56 0 81 0
7 −1 32 0 57 1 82 1
8 0 33 1 58 1 83 −1
9 0 34 1 59 −1 84 0
10 1 35 1 60 0 85 1
11 −1 36 0 61 −1 86 1
12 0 37 −1 62 1 87 1
13 −1 38 1 63 0 88 0
14 1 39 1 64 0 89 −1
15 1 40 0 65 1 90 0
16 0 41 −1 66 −1 91 1
17 −1 42 −1 67 −1 92 0
18 0 43 −1 68 0 93 1
19 −1 44 0 69 1 94 1
20 0 45 0 70 −1 95 1
21 1 46 1 71 −1 96 0
22 1 47 −1 72 0 97 −1
23 −1 48 0 73 −1 98 0
24 0 49 0 74 1 99 0
25 0 50 0 75 0 100 0

Table 3.1: The Mőbius function values for the first 100 positive integers.

Definition 3.5.3 (Mőbius Function). The Mőbius function µ(n) is defined as


1
 if n = 1,
µ(n) = 0 if n is divisible by p2 for some prime p,


(−1)k if n = p1 p2 · · · pk .

Example. µ(20) = 0 because 22 |20. Also, µ(105) = (−1)3 = −1 since 105 = 3 · 5 · 7.

The values of µ(n) for n = 1, 2, . . . , 100 can be found in table ( ).

Note. Why do we care about Mőbius? We can express long, ugly-looking, tedious
sums into a very brief expression including sums over µ(n). If you feel comfortable with
arithmetic functions so far, you might like to see an application of the Mőbius function
µ(n) in theorem ( ). You would be able to make sense of a beautiful application of
this function.

Mőbius function has amazing properties. The first is that it is multiplicative.

188 3.5. DIRICHLET PRODUCT AND MÖBIUS INVERSION

Proposition 3.5.4. The Mőbius function is multiplicative.

Hint. Take m = pα1 1 pα2 2 . . . pαk k and n = pβ1 1 pβ2 2 . . . pβkk and consider the factorization of
the product mn. Use the definition of Mőbius function to finish the proof.
You may be wonder why would someone define such a not-very-nice-looking func-
tion? What is the motivation behind Mőbius function? In fact, this function is the
inverse of a specific arithmetic function. Before going further, we must define the
Dirichlet inverse of an arithmetic function.
Definition 3.5.5 (Dirichlet Inverse Function). Let f be an arithmetic function.
The arithmetic function g for which
f ∗g =ε
is called the Dirichlet inverse of f . The function g is often denoted by f −1 .
You now see why we call ε the unit function. Back to our discussion: let f be an
arithmetic function. Compute the Dirichlet product of f and constant 1 function and
let the result be F , which is also an arithmetic function. By definition of convolution,
we have
F (n) = (f ∗ 1)(n)
X
= f (d).
d|n

This function F is called the summatory function of f . In number theory, there are
times when we know what F = f ∗ 1 is and we want to know what f is. This is where
Mőbius function comes in. Suppose that you do not know Mőbius function, and you
just define µ(n) as the Dirichlet inverse of 1(n). That is, you define µ(n) be a function
such that µ ∗ 1 = ε. Then, multiply F by µ and use proposition ( ) to get
F ∗ µ = (f ∗ 1) ∗ µ
= f ∗ (1 ∗ µ)
=f ∗ε
= f.
In other words, you just need to multiply F by µ to find f . So, we just need to find
out what our defined function µ is. A few computations help us find µ, the Dirichlet
inverse of 1. First, (µ ∗ 1)(1) = ε(1) = 1, which gives µ(1) = 1. Let p be a prime and α
be a positive integer. Then,
(µ ∗ 1)(pα ) = ε(pα )
= 0.
On the other hand, by definition of Dirichlet product and using the fact that the only
divisors of pα are 1, p, . . . , pα ,
X
(µ ∗ 1)(pα ) = µ(d)
d|pα

= µ(1) + µ(p) + · · · + µ(pα )

= 0.
CHAPTER 3. ARITHMETIC FUNCTIONS 189

Since we chose α arbitrarily, the equation µ(1) + µ(p) + · · · + µ(pα ) = 0 must be true
for all values of α. For α = 1, we find 1 + µ(p) = 0, hence µ(p) = −1. Similarly, for
α ≥ 2, we find that µ(pα ) = 0. Finally,

1
 if α = 0,
α
µ(p ) = −1 if α = 1,


0 if α ≥ 2.

The general definition of µ, as stated in definition ( ), can be inferred from the

above result using the fact that µ is a multiplicative function.
As you can observe, the µ function is defined to help us find an arithmetic function
f given its summatory function F . In fact, f = F ∗ µ. This was first introduced by
Mőbius in 19th century and is known as the Mőbius Inversion Formula.
Theorem 3.5.6 (Mőbius Inversion X Formula). If the summatory function of an
arithmetic function f is F (n) = f (d), then
d|n
X n
f (n) = µ(d)F .
d
d|n

In other words, f = µ ∗ F .
When we first discussed Dirichlet inverse of a function, you might have wondered if
this inverse always exists. The following theorem sheds light on this issue.
Theorem 3.5.7. For an arithmetic function f with f (1) 6= 0, the Dirichlet inverse f −1
exists and is unique. In fact, f −1 can be defined recursively as

 1

 if n = 1,


 f (1)

−1
f (n) =
 −1 X −1 n


 f (1) f (d)f if n > 1.

 d
 d|n
d<n

As discussed above, µ(n) is the Dirichlet inverse of the simplest arithmetic function,
1(n). The next theorem follows.
Theorem 3.5.8. The Dirichlet inverse of the Mőbius function is the constant 1 func-
tion. In other words, 1 = µ−1 or
X
µ(n) = ε(n).
d|n

Theorem 3.5.9. Let f be an arithmetic function and denote by F its summatory

function. That is, for all positive integers n,
X
F (n) = f (d).
d|n
190 3.6. MORE ON MULTIPLICATIVE FUNCTIONS

Then,
n
X n
X jnk
F (k) = f (k) .
k=1 k=1
k

Proof. By definition of F , we can write

n
X n X
X
F (k) = f (d)
k=1 k=1 d|k
X X
= f (d) + f (d)
d|1 d|2
X X
+ ··· + f (d) + f (d).
d|n−1 d|n

In the last formula, f (1) is repeated n times because 1 divides all integers. Also,
f (2) is repeated ⌊n/2⌋ times because there are exactly ⌊n/2⌋ numbers between 1 and
n (inclusive) which are even. Analogously, there are exactly ⌊n/k⌋ multiples of k
(1 ≤ k ≤ n) in this interval and thus f (k) is repeated ⌊n/k⌋ times. The proof is
complete.

Example. According to theorem ( ), the summatory function of µ is ε. Therefore,

n
X n
X jnk
ε(k) = µ(k) .
k=1 k=1
k

Since ε(k) equals one when k = 1 and zero otherwise, the sum on the left side equals
one. Consequently,
n
X jnk
µ(k) = 1.
k=1
k

You can check the correctness of this formula by trying some small values for n.

3.6 More on Multiplicative Functions

The most interesting feature of Dirichlet product is that it preserves multiplicativity.
That is, if two arithmetic functions are multiplicative, their Dirichlet product is also
multiplicative. Before we prove this, we show a simple case. As we saw in previous
sections, the summatory function F of an arithmetic function f is defined by F = f ∗ 1.
So, if f is multiplicative, so is F .
CHAPTER 3. ARITHMETIC FUNCTIONS 191
e1 e2 ek
Theorem 3.6.1. Let n be a positive integer with prime factorization
P n = p 1 p 2 . . . pk .
If f is a multiplicative function with summatory function F (n) = f (d), then
d|n

   
(3.5) F (n) = 1 + f (p1 ) + . . . + f (pe11 ) · · · 1 + f (pk ) + . . . + f (pekk )
ei
k X
Y

= f pji
i=1 j=0
k
Y
= F (pei i ) .
i=1

Proof. The proof somewhat follows from unique prime factorization. Assume T is
the expansion of the right side of equation ( ). If d is a divisor of n, then d =
wk
pw 1 w2
1 p2 . . . pk , where 0 ≤ wi ≤ ei for i = 1, 2, . . . , k. Therefore,

wk
f (d) = f (pw 1 w2
1 p2 . . . pk )
wk
= f (pw w2
1 ) f (p2 ) . . . f (pk ) .
1

which isP
a term that is present in T . Thus, we conclude that each term in the sum
F (n) = f (d) appears in T . We can easily find that the converse is also true. After
d|n
expanding the right side of equation ( ), we see that every term in T is of the form
wk
f (pw w2
1 ) f (p2 ) . . . f (pk ) ,
1

where wi are integers with 0 ≤ wi ≤ ei for all i. Since f is multiplicative, we can write
this as
wk
f (pw 1 w2
1 p 2 . . . pk ) ,

which equals f (d). Therefore, every term in T is also in F (n). Combining these two,
T = F (n).

Corollary 3.6.2. If f is a non-zero multiplicative arithmetic function and n = pe11 pe22 . . . pekk
is the prime factorization of a positive integer n,

X k
Y
µ(d)f (d) = (1 − f (pi )).
d|n i=1

Theorem 3.6.3 (Multiplicative Function Theorem (MFT)). Let f, g, and h be

arithmetic functions.

1. If f and g are both multiplicative, then so is f ∗ g.

2. If f is multiplicative, then so is its Dirichlet inverse.

3. If f ∗ g and f are both multiplicative, then so is g.

192 3.6. MORE ON MULTIPLICATIVE FUNCTIONS

Proof.

1. Let f ∗ g = h. We need to prove that if a and b are co-prime positive integers,

h(ab) = h(a)h(b). We have

h(a)h(b) = [(f ∗ g)(a)] · [(f ∗ g)(b)]

  X  
X a b
= f (d1 )g · f (d2 )g
d1 d2
d1 |a d2 |b
   
X a b
= f (d1 )f (d2 )g g
d1 d2
d1 |a,d2 |b
 
X ab
= f (d1 d2 )g
d1 d2
d1 |a,d2 |b
 
X ab
= f (d)g
d
d|ab

= (f ∗ g)(ab)
= h(ab),

which is what we wanted.

2. f is multiplicative, hence f (mn) = f (m)f (n) for all co-prime positive integers m
and n. Put m = n = 1 in this equation to get f (1) = f (1)2 , which immediately
gives f (1) = 1 because f only accepts positive integer values. Let g be the
Dirichlet inverse of f . By theorem ( ), g(1) = 1. We will prove by induction
that g is multiplicative. The base case n = 1 is true. Suppose that n > 1 and
whenever ab < n with (a, b) = 1, then g(ab) = g(a)g(b). Also, suppose that
xy = n with (x, y) = 1. We will show that g(xy) = g(x)g(y). Since g is the
inverse of f , we have (f ∗ g)(n) = ε(n). Also, ε(n) = 0 for n > 1. Therefore,

X n
0 = f (d)g
d
d|n
X  xy 
= f (d)g
d
d|xy
X  xy 
(3.6) = g(xy) + f (d)g .
d
d|xy
d>1

The sum in equation ( ) includes terms of te form f (d)g(xy/d), and since d is

larger than 1, xy/d is less than xy = n. Since x⊥y, there exist positive integers
CHAPTER 3. ARITHMETIC FUNCTIONS 193

d1 and d2 such that d = d1 d2 , d1 |x, and d2 |y. As a result,

X  xy 
g(xy) = − f (d)g
d
d|xy
d>1
 
X x y
= − f (d1 d2 )g
d1 d2
d1 |x,d2 |y
d1 d2 >1
   
X x y
= − f (d1 )f (d2 )g g
d1 d2
d1 |x,d2 |y
d1 d2 >1
 X 
  
x y
= − f (d1 )f (d2 )g g + g(x)g(y)
d1d2
d1 |x,d2 |y
X  X  
x y
= − f (d1 )g f (d2 )g + g(x)g(y)
d1 d2
d1 |x d2 |y
| {z }| {z }
=(f ∗g)(x) =(f ∗g)(y)

= − [(f ∗ g)(x)] [(f ∗ g)(y)] + g(x)g(y).

So, g(xy) = −(f ∗ g)(x) · (f ∗ g)(y) + g(x)g(y). Note that since xy > 1, at least
one of x or y is larger than 1. Therefore, since g is the Dirichlet inverse of f , at
least one of (f ∗ g)(x) or (f ∗ g)(y) is equal to zero. Finally, g(xy) = g(x)g(y) and
we are done.

3. Let f ∗ g = h. Multiply both sides by f −1 , the Dirichlet inverse of f , to get

g = h ∗ f −1 (why?). Multiplicativity of f implies the multiplicativity of f −1 by
part 2. Now, from part 1, since both h and f −1 are multiplicative, their product,
g must also be multiplicative.

Note. If two functions f and g are completely multiplicative, their Dirichlet product
is multiplicative, but not necessarily completely multiplicative. An example would be
(1 ∗ 1)(n) = τ (n). Although the constant 1 function is completely multiplicative, the
number of divisors function d is not.
The same is true for part 2: if f is completely multiplicative, then f −1 is multiplica-
tive, but not in general completely multiplicative.
In proposition ( ), we stated that Euler’s totient function ϕ is multiplicative.
We can now prove this result.
Proposition 3.6.4. Prove that ϕ is multiplicative.
Proof. As proved in theorem ( ), we know that ϕ ∗ 1 = id. Obviously, both 1(n)
and id(n) are multiplicative. The conclusion follows from part 3 of theorem ( ).
Although it is not always easy to find the Dirichlet inverse of an arithmetic function,
there is an exception for completely multiplicative functions.
194 3.6. MORE ON MULTIPLICATIVE FUNCTIONS

Theorem 3.6.5 (Inverse of Completely Multiplicative Functions). Let f be a

multiplicative function. Then f is completely multiplicative if and only if

f −1 (n) = µ(n)f (n) for all n ≥ 1.

Note. Here, µ(n)f (n) is the point-wise multiplication of µ and f at point n. It is not
to be confused with Dirichlet product (µ ∗ f )(n).

Proof. First, if f is completely multiplicative, define g(n) = µ(n)f (n). Then, by defi-
nition of Dirichlet product,
X n
(g ∗ f )(n) = g(d)f
d
d|n
X n
= µ(d)f (d)f
d
d|n
X  n
= µ(d)f d · since f is multiplicative
d
d|n
X
= f (n) µ(d)
d|n

= f (n)ε(n) by theorem ( )
= ε(n).

The last equality holds because f (1) = ε(1) = 1 (why?) and ε(n) = 0 for all n ≥ 2.
This means that g is the Dirichlet inverse of f .
For the converse, suppose that f −1 (n) = µ(n)f (n) and we want to prove that f
is completely multiplicative. It suffices to prove that f (pα ) = f (p)α for every prime
p and any positive integer α (why?). We will proceed by induction. The base case
α = 1 is obvious. Suppose that we know f (pα−1 ) = f (p)α−1 and we want to show that
f (pα ) = f (p)α . For any n ≥ 2, (f ∗ f −1 )(n) = ε(n) = 0. Therefore,
X n
µ(d)f (d)f = 0.
d
d|n

Put n = pα in the above equation to obtain

X n
µ(d)f (d)f = µ(1)f (1)f (pα ) + µ(p)f (p)f (pα−1 ) + · · · + µ(pα )f (pα )f (1)
d
d|pα

= µ(1)f (1)f (pα ) + µ(p)f (p)f (pα−1 )

= f (pα ) − f (p)f (pα−1 )
= 0.

That is, f (pα ) = f (p)f (pα−1 ). Using induction hypothesis, we immediately get the
result.
CHAPTER 3. ARITHMETIC FUNCTIONS 195

3.6.1 More on τ
We provide two proofs for this theorem. The first one uses the concept of multiplicative
functions discussed in previous sections. The second solution has a combinatorial spirit.

Proof. Since d is multiplicative, it suffices to find τ (n) only when n is a power of a

prime. Let n = pα for some prime p and a positive integer α. There exist α + 1 divisors
1, p, p2 , . . . , pα for n. Therefore, d(pα ) = α + 1. To prove the theorem in the general
case, take n = pe11 pe22 . . . pekk and write

d(n) = d(pe11 pe22 . . . pekk )

= d(pe11 )d(pe22 ) . . . d(pekk )
= (e1 + 1)(e2 + 1) . . . (ek + 1)
Yk
= (1 + ei ).
i=1

Example.

• τ (p) = 2 for all primes p.

• τ (100) = τ (22 · 52 ) = (2 + 1)(2 + 1) = 9.

Problem 3.6.6. Show that 1 = τ ∗ µ.

Solution. By theorem ( ), 1 ∗ µ = ε. Multiply both sides of this equation by 1 and

use the fact that 1 ∗ 1 = τ , we find

τ ∗µ = 1∗1∗µ
= 1∗ε
= ε.

As mentioned in section ( ), the number of solutions to ab = n in positive

integers is τ (n). For the sake of completeness, we will state this result as a theorem.

Theorem 3.6.7. Let n be a positive integer. The number of pairs (a, b) of positive
integers which satisfy

ab = n

is τ (n).

Proof. First, if for some positive integers a and b the equation ab = n holds, then a is
a divisor of n. Second, for any divisor a of n, the number b = n/a is an integer that
satisfies ab = n. Therefore, τ (n) counts exactly all the solutions of ab = n.

Here is a more interesting question:

196 3.6. MORE ON MULTIPLICATIVE FUNCTIONS

n τ (n)
1 1
2 2
3 2
4 3
5 2
6 4
7 2
8 4
9 3
10 4

Table 3.2: Value of τ (n) for n = 1, 2, . . . , 10.

Question 3.6.8. Let x ≥ 1 be a real number. Find the number of pairs (a, b) of
positive integers which satisfy

(3.7) ab ≤ x.

One way to answer this question is to break the condition ab ≤ x into smaller
conditions. Obviously, the equation ab = y does not have any solutions in integers if y
is not an integer. So, in order to find the number of solutions to the inequality ( ),
we only need to find the number of solutions to each of ab = 1, ab = 2, . . . , ab = ⌊x⌋
and the answer would be the sum of these numbers. According to theorem ( ), the
answer is

τ (1) + τ (2) + · · · + τ (⌊x⌋).

We are seeking a way to determine this sum.

Definition 3.6.9. Let x ≥ 1 be a real number. Denote by T (x) the number of positive
integer solutions to ab ≤ x.
From what we have already found,
⌊x⌋
X
(3.8) T (x) = τ (k).
k=1

Example. Let us find T (10.2). We need to compute τ (1), τ (2), . . . , τ (10). According to
table ( ) and equation ( ), T (10.2) is equal to the sum of numbers in the second
column, which is 27.
As you observe in the above example, it would be a tedious job to calculate T (x)
by summing up the τ values. There is another way to compute T (x). Let a be a fixed
positive integer. The inequality ab ≤ x is equivalent to
x
b≤ .
a
CHAPTER 3. ARITHMETIC FUNCTIONS 197

Since b is a positive integer, it can take the values 1, 2, . . . , ⌊ xa ⌋. These are ⌊ xa ⌋ numbers.
Since a can take all the values 1, 2, . . . , ⌊x⌋, we have
jxk jxk  
x
T (x) = + + ··· + .
1 2 ⌊x⌋

Example. Let us calculate T (10.2) with the new approach. We have

     
10.2 10.2 10.2
T (10.2) = + + ··· +
1 2 10
= 10 + 5 + 3 + 2 + 2 + 1 + 1 + 1 + 1 + 1
= 27,

which is in agreement with our previous result.

The formula
⌊x⌋ j k
X x
(3.9) T (x) =
k=1
k

is more convenient than ( ). That is, we can find T (x) using equation ( ) with less
number of calculations. In fact, to compute T (x) from ( ), we need to calculate the
number of divisors of ⌊x⌋ numbers, which obviously needs more computations than ⌊x⌋
computations needed in formula ( ). However, this can be refined even more.
We will divide the √solutions to ab ≤ x into two classes. The first class includes
√
solutions in which a ≤ x and the second one includes all solutions such that a > x.
Let T1 (x) be the number of solutions in the first class and let T2 (x) be that of the
second class. Therefore,

T (x) = T1 (x) + T2 (x).

√
For the first class of solutions, let us consider 1 ≤ a ≤ x is a fixed integer. The
inequality ab ≤ x, can be written as b ≤ xa . Therefore, in order to satisfy ab ≤ x, b can
take only the values 1, 2, . . . , ⌊ xa ⌋, which x
√ are ⌊ a ⌋ different numbers. Keeping in mind
that a can take the values 1, 2, . . . , ⌊ x⌋, the number of solutions in the first class is
√
⌊ x⌋ j
X xk
T1 (x) = .
a=1
a

For the second class, we must find how many pairs (a, b) of integers satisfy the following
inequalities:
√
ab ≤ x and a > x.

This can be written as

√ x
x<a≤ .
b
198 3.6. MORE ON MULTIPLICATIVE FUNCTIONS
√
Obviously,√since the product of a and b is less than or equal to x and a > x, we must
have b ≤ x (otherwise ab would be larger than x). Fix b. The only possible values for
a are
√ √ jxk
⌊ x⌋ + 1, ⌊ x⌋ + 2, . . . , .
b
√ √
These are ⌊ xb ⌋ − ⌊ x⌋ numbers. So, since b can take all values 1, 2, . . . , ⌊ x⌋, we have
√
⌊ x⌋ j
X xk √ 
(3.10) T2 (x) = − ⌊ x⌋ .
b=1
b

In order to add T1 (x) and T2 (x) to obtain T (x), we have to make their summation
parameter the same. We already know that changing the summation parameter does
not change the value of the sum. Therefore, we can change b to a in equation ( )
and write T2 (x) as
√
⌊ x⌋ j
X xk √ 
T2 (x) = − ⌊ x⌋ .
a=1
a

We can now add up T1 (x) and T2 (x) and write

T (x) = T1 (x) + T2 (x)

√ √ 
⌊ x⌋ j k ⌊ x⌋ j k
X x X x √ 
= + − ⌊ x⌋ 
a=1
a a=1
a
√ √
⌊ x⌋ j ⌊ x⌋
X xk X √
= 2 −2 ⌊ x⌋
a=1
a
a=1 
√
⌊ x⌋ j k
X x  √ √ √ 
= 2 − ⌊ x⌋ + ⌊ x⌋ + · · · + ˚⌊ x⌋
a=1
a | √
{z }
⌊ x⌋ times
√
⌊ x⌋ j
X xk √
= 2 − ⌊ x⌋2 .
a=1
a

Finally, we have the formula

√
⌊ x⌋ j
X xk √
(3.11) T (x) = 2 − ⌊ x⌋2 .
a=1
a
√
Equation ( ) needs ⌊ x⌋ computations to calculate T (x), which is obviously less
than ⌊x⌋ computations needed to find T (x) from ( ).
Example. We will calculate the number of positive integer solutions to ab ≤ 100.2. Since
x = 100.2 is large in this case, the best way to calculate T (x) is from formula ( ).
CHAPTER 3. ARITHMETIC FUNCTIONS 199

We find
10  
X 100.2
T (100.2) = 2 − 100
a=1
a


= 2 100 + 50 + 33 + 25 + 20 + 16 + 14 + 12 + 11 + 10 − 100
= 482.

Theorem 3.6.10 (An Approximation for Average Value of τ (n)). Let n be a

positive integer. Prove that
τ (1) + τ (2) + · · · + τ (n) 1 1 1
= + + · · · + − e,
n 1 2 n
where e is a number between zero and one.
Proof. Assuming x = n, we employ equations ( ) and ( ) to write
n n j k
X X n
τ (k) =
k=1 k=1
k

The right hand side of the above equation can be written as

n j k n  n n o
X n X n
= − ,
k=1
k k=1
k k

where {n/k} is the fractional part of n/k. Therefore,

n n n n o
X X n X n
(3.12) τ (k) = − .
k=1 k=1
k k=1
k

Dividing both sides of this last equation by n, we find

n
P n 
P n
τ (k) n k
k=1
X 1 k=1
(3.13) = − .
n k=1
k n

We already know that 0 ≤ {n/k} < 1. Therefore, the second sum in the right hand
side of equation ( ) is between zero and n. Therefore,
Pn
{n/k}
0 < k=1 <1
n
(it is never equal to zero, can you see why?). The proof is complete.
Note. This theorem states that the average value of number-of-divisors function (i.e.,
[τ (1) + ˚τ (2) + · · · + τ (n)]/n) is approximately equal to sum of reciprocals of positive
integers up to n. We represent this as
τ (1) + τ (2) + · · · + τ (n) 1 1 1
∼ + + ··· + .
n 1 2 n
200 3.6. MORE ON MULTIPLICATIVE FUNCTIONS

3.6.2 More on σ and its Generalization

Let us prove that σ is multiplicative.

Proposition 3.6.11. The sum-of-divisors function σ is a multiplicative arithmetic

function.

Proof. If you look more closely, you will observe that

X
σ(n) = d
d|n
X n
= d·1
d
d|n

= (id ∗ 1)(n).

Now, since both id and constant 1 function are multiplicative, part 1 of theorem ( )
implies that σ is also multiplicative.

Problem 3.6.12. Prove that for two positive integers a and b, if a | b, then

σ(a) σ(b)
≤ .
a b
Q
Solution. Let b = pe be the prime factorization of b. Then, since a|b, the factorization
p|b
Q
of a would be a = pf with 0 ≤ f ≤ e for each prime p dividing b. From theorem
p|b
( ),
Y pe+1 − 1 Y pf +1 − 1
σ(b) = and σ(a) = .
p−1 p−1
p|b p|b

Thus, we need to show that

Y pf +1 − 1 Y pe+1 − 1
≤ .
pf (p − 1) pe (p − 1)
p|b p|b

Now it suffices to prove that

pe+1 − 1 pf +1 − 1
≥
pe (p − 1) pf (p − 1)
⇐⇒ pe+f +1 (p − 1) − pf (p − 1) ≥ pe+f +1 (p − 1) − pe (p − 1)
⇐⇒ pe (p − 1) ≥ pf (p − 1),

which is true since e ≥ f .

√
Proposition 3.6.13. Let n be a composite positive integer. Prove that σ(n) > n+ n.
CHAPTER 3. ARITHMETIC FUNCTIONS 201

Proof. Since n is composite, √ we know from proposition ( ) that n √

has a prime
factor p less than or equal to n. Then, n/p is a divisor of n and n/p ≥ n. Since n
has at least three divisors n, n/p, and 1,
n
σ(n) ≥ n + +1
p
√
≥ n+ n+1
√
> n + n.

Theorem 3.6.14. Let k > 1 be a given positive integer. Show that the equation

σ(n) = n + k

has a finite number of solutions for n.

Proof. Suppose that σ(n) = n+k holds for some positive integer n. Since k > 1, n must√
be composite (otherwise
√ σ(n) = n + 1). According to theorem ( ), σ(n) > n + n.
As a result, we find n < k or n < k . Obviously, there are at most k values 1, 2, . . . , k 2
2 2

possible for n. This finishes the proof.

Theorem 3.6.15. Let k be an arbitrary positive integer. Show that there exists a
positive integer n0 such that for any n > n0 ,

σ(n!)
> k.
n!
σ(n!)
In other words, prove that lim n!
= ∞.
n→∞

Proof. According to theorem ( ),

σ(n!) X 1
= .
n! d
d|n!

Since 1, 2, . . . , n all divide n!, we can write

σ(n!) X1
=
n! d
d|n!
1 1 1
> + + ··· + .
1 2 n
It suffices to find an n such that
1 1 1
+ + · · · + > k.
1 2 n
202 3.6. MORE ON MULTIPLICATIVE FUNCTIONS

Take n = 22k . Then,

   
1 1 1 1 1 1 1 1 1 1
+ + · · · + 2k > 1 + + + + + + +
1 2 2 2 4 4 8 8 8 8
 
 1 1 1 
+··· +  +
 22k 22k + · · · + 
2k 
| {z 2 }
22k−1 times
 
1
> 1 + 2k
2
= 1+k
> k,

as desired.
Theorem 3.6.16. σ(n) is odd if and only if n = k 2 or n = 2k 2 for some integer k.
Proof. Let n = pe11 pe22 . . . pekk be the factorization of n and suppose that σ(n) is odd.
We will prove that n is either a perfect square or twice a perfect square. According to
k e +1
Q pi i −1
theorem ( ), since σ(n) is odd, all the terms in the product p−1
must be odd.
i=1
That is,

(3.14) 1 + pi + p2i + · · · + pei i

must be odd for i = 1, 2, . . . , k. There are two cases: if pi 6= 2 for all i, then the number
of terms in the expression ( ) must be odd so that their sum is also odd (they are
all odd). That is, ei must be even for i = 1, 2, . . . , k. Hence, n is a perfect square.
The other case is when some pi , say p1 is equal to 2. The expression ( ) is odd for
p1 = 2. Now, in order for expression ( ) to be odd, all ei must be even for 2 ≤ i ≤ k
(exactly like the previous case). Now, if the power of 2 that divides n is even, n is of
the form k 2 and if it is odd, n is of the form 2k 2 .
The proof of the converse is now obvious.
Corollary 3.6.17. If n is an odd positive integer, then n is a square if and only if σ(n)
is odd.
Proposition 3.6.18. Let n be a positive integer. The number of positive integers less
than or equal to n with even sum of divisors is
r 
√  n
n− n − .
2
Proof. There are n positive integers less than or equal to n. We know from theorem
( ) that numbers which are a perfect square or twice a perfect
√ square have an even
σ. The number of perfect squares less than or equal to n is ⌊ n⌋ (why?). Similarly,
p the
number of integers less than or equal to n which are twice a perfect square is ⌊ n/2⌋.
The conclusion follows.
CHAPTER 3. ARITHMETIC FUNCTIONS 203

A family of numbers which are related to sum-of-divisors function are perfect num-
bers.

Definition 3.6.19. A positive integer n is a perfect number if σ(n) = 2n.

Example. The only perfect numbers less than 104 are 6, 28, 496, and 8128.

Theorem 3.6.20. Let k > 1 be a positive integer. If 2k − 1 is a prime number, then

2k−1 (2k − 1) is a perfect number. Also, every even perfect number has this form.

Proof. First, suppose that p = 2k − 1 is a prime. We will prove that n = 2k−1 (2k − 1)
is perfect. Notice that σ(p) = p + 1 = 2k and since σ is multiplicative,


σ(n) = σ 2k−1 (2k − 1) = σ(2k−1 )σ(p)



= 1 + 2 + 22 + · · · + 2k−1 2k


= 2k − 1 2k
= 2n.

Now, suppose that n is an even perfect number. We want to show that it is of the form
2k−1 (2k − 1). We will use the trick stated in theorem ( ). One can write n as 2k s
for a positive integer k and an odd integer s. Again by multiplicativity of σ,

σ(n) = σ(2k s) = σ(2k )σ(s)

= 2k+1 − 1 σ(s).

On the other hand, as n is a perfect number, σ(n) = 2n = 2k+1 s. Therefore,

(3.15) 2k+1 s = 2k+1 − 1 σ(s).

This means that 2k+1 −1 divides 2k+1 s, but since (2k+1 , 2k+1 −1)
= 1, we have 2k+1 −1|s.
Consequently, there exists an integer t such that s = 2k+1 − 1 t. Substituting this in
equation ( ), one realizes that

2k+1 t = σ(s) ≥ s + t = 2k+1 t.

Therefore, σ(s) = s + t. It means that s must be a prime and its only divisors are itself
(s) and one (t) (why?). Hence n = 2k (2k+1 − 1), 2k+1 − 1 is a prime, and n has the
desired form.

Remark. We do not know whether there exist infinitely many perfect numbers. In
fact, even perfect numbers are formed by Mersenne primes (see theorem ( ) and
its remark), and it has not yet been proved that there are infinitely many Mersenne
primes. Furthermore, it is not known whether there are any odd perfect numbers.
Mathematicians have tested first 101500 integers but did not find any odd perfect number
among them, so they guess there probably does not exist any.

Theorem 3.6.21. Let n be a positive integer. Then, σ(n) is a power of 2 if and only
if n is a product of some Mersenne primes.
204 3.6. MORE ON MULTIPLICATIVE FUNCTIONS

Proof. It is obvious that sum of divisors of the product of some Mersenne primes is a
power of 2 (prove it yourself). We will now show that if σ(n) = 2k for some natural
k, then n is a product of Mersenne primes. Let n = pα1 1 pα2 2 . . . pαmm be the prime
factorization of n. Since σ is multiplicative, σ(pαi i ) must be a power of 2 for i =
1, 2, . . . , m. Therefore, it suffices to prove that if for a prime p and a positive integer
α, σ(pα ) = 2k , then p is a Mersenne prime and α = 1. The only divisors of pα are
1, p, p2 , . . . , pα , so,

2 k = 1 + p + p2 + · · · + pα .

This means that α is odd. Take α = 2β + 1. Then we can factorize the right hand side
as
 
(3.16) 2k = (1 + p) 1 + p2 + p4 + · · · + p2β

This leads us to the conclusion that 1 + p is a power of 2, which means p is a Mersenne

prime. Now suppose that α > 1. From equation ( ), we observe that β must be
odd. Take β = 2γ + 1 and write

2t = 1 + p2 + p4 + · · · + p2β

 
= 1 + p2 1 + p4 + p8 + · · · + p4γ ,

for some natural t. This means that 1 + p2 is a power of 2. However, since p is a

Mersenne prime, there exists some s for which p = 2s − 1. Therefore,

1 + p2 = 1 + (2s − 1)2
= 2 + 22s − 2s+1 .

But this last expression is not divisible by 4 and cannot be a power of 2. This is a
contradiction. Thus, a = 1 and we are done.

Problem 3.6.22. Prove that for any positive integer n,

 
 
X µ(d)σ(d)  1
 ≥ .

 d|n d  n


Solution. Let f (n) = µ(n)σ(n)/n. Sine all three functions µ, σ, and id are multi-
plicative, so is f . Theorem ( ) implies that the summatory function of f is also
multiplicative. Therefore, it suffices to prove the given inequality only when n = pα ,
where p is a prime and α is a positive integer. Since µ(px ) is zero for x ≥ 2, µ(1) = 1,
CHAPTER 3. ARITHMETIC FUNCTIONS 205

and µ(p) = −1,

 
X   α a 

 µ(d)σ(d)   µ(1)σ(1) µ(p)σ(p) µ(p )σ(p )
  =  + + ··· + 

 d|pα d 

 1 p pa 
 
 µ(1)σ(1) µ(p)σ(p) 
= 
 + 
1 p 
 
1 p + 1
=  − 
1 p 
1
=
p
1
≥ α.
p
The reader is supposed to draw the conclusion and finish the proof.
We can generalize the concept of number-of-divisors and sum-of-divisors functions
to define divisor functions.
Definition 3.6.23. Let n be a positive integer and α ≥ 0 be an integer . The sum of
αth powers of positive divisors of n is denoted by σα (n). In other words,
X
σα (n) = dα .
d|n

The functions σα are called divisor functions.

According to this definition, the number-of-divisors function τ is equal to σ0 and
the sum-of-divisors function σ is equal to σ1 .
Theorem 3.6.24 (Properties of Divisor Functions). For every non-negative inte-
ger α and positive integer n,
1. σα is multiplicative (but not completely multiplicative),
2. σα = 1 ∗ idα ,
3. if α is non-zero and n = pe11 pe22 . . . pekk is the prime factorization of n,
k α(e +1)
Y p i −1
i
σα (n) =
i=1
p−1
Yk
ai α


= 1 + pαi + p2α
i + · · · + pi .
i=1

4. the Dirichlet inverse of σα is obtained by

X n
−1 α
σα (n) = d µ(d)µ .
d
d|n
6
More generally, one can assume that α is any non-negative real number.
206 3.6. MORE ON MULTIPLICATIVE FUNCTIONS

Proof. We only prove parts 2 and 4.

2. Notice that
X n
α α α
(1 ∗ id )(n) = (id ∗ 1)(n) = id (d) · 1
d
d|n
X
= dα
d|n

= σα (n).

4. From theorem ( ), sine the function idα is completely multiplicative, its Dirich-
α
let inverse is µ id . Hence, from part 2 of this theorem,

(σα−1 )(n) = (1 ∗ idα )−1 (n)

= 1−1 ∗ (idα )−1 (n)
= (µ ∗ µ idα ) (n)
X n
α
= d µ(d)µ .
d
d|n

3.6.3 More on ϕ(n) and Jk (n)

Theorem 3.6.25. Let n be a positive integer. Prove that
X µ(d)
ϕ(n) = n .
d
d|n
P
Proof. Let f (n) = ϕ(n). By theorem ( ), we know that F (n) = ϕ(d) = n.
d|n
Therefore, by Mőbius inversion formula,
X n
ϕ(n) = f (n) = µ(d)F
d
d|n
X n
= µ(d)
d
d|n
X µ(d)
= n ,
d
d|n

as desired.
Corollary 3.6.26. ϕ = µ ∗ id.
P
In theorem ( ), we proved that ϕ(d) = n. We will now prove that the
d|n
converse of this theorem is also true. That is, ϕ is the only arithmetic function with
this property.
CHAPTER 3. ARITHMETIC FUNCTIONS 207

Theorem 3.6.27. If f is an arithmetic function such that for all n ∈ N,

X
f (d) = n,
d|n

then f (n) = ϕ(n) for all n ∈ N.

P
Proof. By Mőbius inversion formula, f = µ ∗ F , where F (n) = f (d). So, in this case
d|n
f = µ ∗ id. On the other hand, we already know from corollary ( ) that ϕ = µ ∗ id.
Thus, f = ϕ.
Theorem 3.6.28. Let n be a positive integer. Prove that
X Y
1. µ(d)ϕ(d) = (2 − p),
d|n p|n
X Y
2. µ(d)2 ϕ(d)2 = (1 + (p − 1)2 ),
d|n p|n

X µ(d) Y 1

3. = 1− .
ϕ(d) p−1
d|n p|n

Proof. We will only prove 1 here. The other two parts are similar. Let f (n) = µ(n)ϕ(n).
Since µ and ϕ are both multiplicative, their product f is also multiplicative. By theorem
( ), the summatory
P function of f is multiplicative as well. Therefore, it suffices to
find the sum F (n) = µ(d)ϕ(d) only when n is a power of a prime. Let n = p . Then,
α
d|n
since µ(p ) = 0 for integer β > 1,
β

X
µ(d)ϕ(d) = µ(1)ϕ(1) + µ(p)ϕ(p) + · · · + µ(pα )ϕ(pα )
d|pα

= µ(1)ϕ(1) + µ(p)ϕ(p)
= 1 − (p − 1)
= 2 − p.

Suppose that n = pe11 pe22 . . . pekk is the prime factorization of n. Then, again by theorem
( ),

X Y ei
k X
µ(d)ϕ(d) = µ(d)ϕ(d)
d|n i=1 j=0
k
Y
= (2 − pi ).
i=1

Theorem 3.6.29. Let n be a composite integer larger than 6. Then,

√
ϕ(n) ≤ n − n.
208 3.6. MORE ON MULTIPLICATIVE FUNCTIONS

Proof. Let n = pα1 1 pα2 2 · · · pαk k be the prime factorization of n. Since n is composite, we
get from proposition
√ ( ) that it must have a prime factor, say p1 , which is less than
or equal to n. Then, using the formula in part (d) of theorem ( ),
    
1 1 1
ϕ(n) = n 1 − 1− ··· 1 −
p1 p2 pk
 
1
≤ n 1−
p1
 
1
≤ n 1− √
n
√
= n − n.

Theorem 3.6.30. Let k be an arbitrary positive integer. Show that there exists a
positive integer n0 such that for any n > n0 ,
ϕ(n) > k.
In other words, prove that lim ϕ(n) = ∞.
n→∞
√
Proof. We will prove that ϕ(n) ≥ 2n and the result immediately follows. First, suppose
that n is odd. Let n = pα1 1 pα2 2 · · · pαk k be the prime factorization of n. Then, because
of the fact that pi are primes larger than 2, 1 ≤ i ≤ k, we can use the inequality
1/2
pi − 1 ≥ pi (prove it yourself) to write
ϕ(n) = p1α1 −1 p2α2 −1 · · · pαk k −1 (p1 − 1) · · · (pk − 1)
1/2 1/2 1/2
≥ p1α1 −1 p2α2 −1 · · · pαk k −1 · p1 p2 · · · pk
α −1/2 α2 −1/2 α −1/2
≥ p1 1 p2 · · · pk k .
Now, since αi is a positive integer for all i, the inequality αi − 1/2 ≥ αi /2 holds true.
Hence,
α −1/2 α2 −1/2 α −1/2
ϕ(n) ≥ p1 1 p2 · · · pk k
α /2 α /2 α /2
≥ p1 1 p2 2 · · · pk k
√
= n.
For even n, let n = 2α t for some positive integer
√ α and odd t. Then, ϕ(n) = ϕ(2 )ϕ(t).
α

Since t is odd, we already know that ϕ(t) ≥ t. Thus,

ϕ(n) = ϕ(2α )ϕ(t)
= 2α−1 ϕ(t)
√
≥ 2α−1 t
√
≥ 2α/2−1 t
1 α/2 √
= 2 t
2
1√
= n.
2
CHAPTER 3. ARITHMETIC FUNCTIONS 209
√
Consequently, ϕ(n) ≥ 2n for all positive integers n. To finish the proof, select n0 = 4k 2
so that for any n > n0 ,

1√
ϕ(n) ≥ n
2
1√
> n0
2
= k.

Corollary 3.6.31. Let k be an arbitrary integer. The number of positive integers n

for which ϕ(n) = k is finite.

Theorem 3.6.32. Let n be a positive integer. Then,

n nj n k2
X 1 1X
ϕ(k) = + µ(k) .
k=1
2 2 k=1 k

Proof. Define

n j n k2
1X
f (n) = µ(k) .
2 k=1 k

Then, from theorem ( ),

n+1  2 n j n k2
1X n+1 1X
f (n + 1) − f (n) = µ(k) − µ(k)
2 k=1 k 2 k=1 k
n+1   !
2 j n k2
1X n+1
= µ(k) −
2 k=1 k k
 
1 X n+1
= µ(k) 2 −1
2 k
k|(n+1)
X µ(k) 1 X
= (n + 1) − µ(k).
k 2
k|(n+1) k|(n+1)

By theorem ( ), the first term in the last line of above equations equals ϕ(n + 1),
and by theorem ( ), the second term is 12 ε(n + 1). Therefore,

1
f (n + 1) − f (n) = ϕ(n + 1) − ε(n + 1).
2
210 3.6. MORE ON MULTIPLICATIVE FUNCTIONS

Now, by identity ( ) in appendix ( ),

n j n k2 n−1
1X X
µ(k) = f (n) = f (1) + (f (k + 1) − f (k))
2 k=1 k k=1
n−1  
1 X 1
= + ϕ(k + 1) − ε(k + 1)
2 k=1 2
n−1 n−1
1 X 1X
= + ϕ(k + 1) − ε(k + 1)
2 k=1 2 k=1
| {z }
=0
n−1
1 X
= + ϕ(k + 1)
2 k=1
n
1 X
= + ϕ(k) − ϕ(1)
2 k=1
n
X 1
= ϕ(k) − .
k=1
2

The proof is complete.

Theorem 3.6.33. For all positive integers n,

n n
X ϕ(k) X µ(k) j n k
= .
k=1
k k=1
k k

First Proof. In an analogous way to the proof of previous theorem, we use theorem
n
P  
µ(k) n
( ). Define f (n) = k k
. In this case,
k=1

n+1   n
X µ(k) n + 1 X µ(k) j n k
f (n + 1) − f (n) = −
k=1
k k kk=1
k
n+1  j n k

X µ(k) n+1
= −
k=1
k k k
X µ(k)
=
k
k|(n+1)

ϕ(n + 1)
= by theorem ( ).
n+1
CHAPTER 3. ARITHMETIC FUNCTIONS 211

Consequently,
n n−1
X µ(k) j n k X
= f (n) = f (1) + (f (k + 1) − f (k))
k=1
k k k=1
n−1
X ϕ(k + 1)
= 1+
k=1
k+1
n
X ϕ(k) ϕ(1)
= 1+ −
k=1
k 1
n
X ϕ(k)
= ,
k=1
k

as desired.
µ(n)
Second Proof. Take f (n) = . Then, F (n), the summatory function of f , would be
n
X µ(d)
F (n) =
d
d|n
 
1  X µ(d) 
= n .
n d
d|n

According to theorem ( ), the expression in the parenthesis equals ϕ(n). Therefore,

ϕ(n)
F (n) = .
n
The conclusion follows from theorem ( ).

Theorem 3.6.34 (Properties of Jordan’s Totient Function). Let n and k be

positive integers. Then,
Y 1

k
1. Jk (n) = n 1− k ,
p
p|n

2. Jk is multiplicative,
X
3. Jk (d) = nk ,
d|n

4. Jk = idk ∗ µ,

5. Jk is even if and only if n ≥ 3.

Proof.
212 3.6. MORE ON MULTIPLICATIVE FUNCTIONS

1. Let p1 , p2 , . . . , pm be the prime factors of n. Consider all k−tuples (x1 , x2 , . . . , xk )

of positive integers so that xi ≤ n for i = 1, 2, . . . , n. The number of such
k−tuples is nk (why?). In order to find Jk (n), we must find out how many of
these nk k−tuples have the property that (x1 , x2 , . . . , xk , n) = 1. In other words,
we need to omit the k−tuples in which all xi (1 ≤ i ≤ k) are divisible by at least
one of pj (1 ≤ j ≤ m). There are exactly (n/p1 )k k−tuples in which all xi are
divisible by p1 . So, the number of k−tuples in which not all xi are divisible by p1
is
 
k 1
n1 = n 1 − k .
p1
 
Among these n1 k−tuples, there are n2 = n1 1 − p1k ones in which not all xi
2
are divisible by p2 . Continuing this way, assuming nj (1 ≤ j ≤ m) to be the
number of k−tuples (x1 , x2 , . . . , xk ) in which not all xi (1 ≤ i ≤ k) are divisible
by p1 , p2 , . . . , or pj ,
 
1
n3 = n2 1 − k ,
p3
 
1
n4 = n3 1 − k ,
p4
..
.
 
1
nm = nm−1 1 − k .
pm
It is now obvious that Jk (n) = nm , thus
    
k 1 1 1
Jk (n) = n 1 − k 1 − k ... 1 − k
p1 p2 pm
Y 1

= nk 1− k .
p
p|n

2. Let m and n be co-prime positive integers. We will show that Jk (mn) = Jk (m)Jk (n).
Suppose that
m = pα1 1 pα2 2 . . . pαt t and n = q1β1 q2β2 . . . qsβs
are the prime factorizations of m and n, respectively. Then, according to part 1
of this theorem,
Y 1

k
Jk (mn) = (mn) 1− k
p
p|mn

Y !  !
1 Y 1
= mk 1− k nk 1− k
p p
p|m p|n

= Jk (m)Jk (n).
CHAPTER 3. ARITHMETIC FUNCTIONS 213

3. Since Jordan’s totient function is multiplicative, it suffices to prove that

X
Jk (d) = (pα )k ,
d|pα

where p is a prime and α is a positive integer. According to part 2 of this theorem,

for any positive integer i,

Jk (pi ) = pki − pk(i−1) .

Therefore,
X
Jk (d) = Jk (1) + Jk (p) + · · · + Jk (pα )
d|pα
α
X

= 1+ pki − pk(i−1) .
i=1

Now, from identity ( ) in appendix ( ),

X α
X

Jk (d) = 1 + pki − pk(i−1)
d|pα i=1


= 1 + pkα − 1
= (pα )k .

X
4. The equation Jk (d) = nk can be represented as 1 ∗ Jk = idk . Then, since the
d|n
inverse of function 1 is µ (theorem ( )),

Jk = 1−1 ∗ (1 ∗ Jk )
= 1−1 ∗ idk
= µ ∗ idk .

5. Let n = pα1 1 pα2 2 . . . pαmm be the prime factorization of n. One can write the formula
in part 1 of this theorem as
    
k 1 1 1
Jk (n) = n 1− k 1 − k ... 1 − k
p1 p2 pm
α1 −1 α2 −1


αm −1 k

k


= p1 p2 . . . pm p1 − 1 p2 − 1 . . . pkm − 1 .
k

The proof is obvious now.

214 3.7. MENON’S IDENTITY

3.7 Menon’s Identity

Menon’s Identity is a very nice theorem in number theory, though not popular. Here,
we prove it using a powerful theorem. First, we need the following lemma.
Lemma 3.7.1. Let m and n be positive integers such that (m, n) = d. Prove that
d
ϕ(mn) = ϕ(m)ϕ(n) · .
ϕ(d)
Q 1

Proof. Using part (d) of theorem ( ), we know that ϕ(x) = x 1− p
, where x
p|x
is a positive integer and p ranges over all prime divisors of x. So,
Y 1

ϕ(mn) = mn 1−
p
p|mn
Q   Q 
1 1
1− p · 1− p
p|m p|n
= mn · Q 
1
1− p
p|d
Q 1
 Q 1

m 1− p
·n 1− p
·d
p|m p|n
= Q 
1
d 1− p
p|d
d
= ϕ(m)ϕ(n) · .
ϕ(d)

Theorem 3.7.2. Given integers r, d, k, and n so that n = dk, d, k > 0, n ≥ 1, and

gcd(r, d) = 1. Then the number of integers of the form r + id (for 1 ≤ i ≤ k) which are
ϕ(n)
co-prime to n is .
ϕ(d)
ϕ(n)
Proof. First, let us write in another way. Let s = (d, k). Then, from lemma
ϕ(d)
( ),
ϕ(n) ϕ(kd)
=
ϕ(d) ϕ(d)
ϕ(k)ϕ(d)s
ϕ(s)
=
ϕ(d)
ϕ(k)
= s .
ϕ(s)
Notice that if (r + id, n) = 1 for some i, then (r + id, k) = 1. The reason is that if
(r+id, k) = g, then g|k, which in turn means g|n. Now, g divides r+id and n and hence
CHAPTER 3. ARITHMETIC FUNCTIONS 215

it also divides their gcd, which is 1. So, g = 1. On the other hand, if (r + id, k) = 1 for
some i, then (r + id, n) = 1 since r ⊥ d.
Therefore, the problem reduces to showing that the number of integers of the form
ϕ(k)
r + id (1 ≤ i ≤ k) which are co-prime to k is s .
ϕ(s)
Since s = (d, k), there exist positive integers k1 and d1 such that k = k1 s and
d = d1 s. So, the numbers r + id (for 1 ≤ i ≤ k) will become r + id1 s (for 1 ≤ i ≤ k1 s).
Classify these numbers in the following manner:
r + d1 s r + 2d1 s ... r + k1 d 1 s
r + (k1 + 1)d1 s r + (k1 + 2)d1 s ... r + 2k1 d1 s
.. .. .. ..
. . . .
r + [(s − 1)k1 + 1]d1 s r + [(s − 1)k1 + 2]d1 s . . . r + sk1 d1 s
We want to prove that the number of integers among these numbers which are co-prime
ϕ(k)
to k is s . Let us prove the case when s = 1 first. In this case, k ⊥ d, k = k1 , and
ϕ(s)
d = d1 . Therefore, the numbers in the first row are r + d, r + 2d, . . . , r + kd, which are
all different modulo k (why?). This means that they form a complete residue system
modulo k. Obviously, there are
ϕ(k)
ϕ(k) =
ϕ(s)
numbers co-prime to k among these integers. So, the theorem is proved for the case
s = 1. Now assume that s > 1. There are s rows of numbers. So, if we prove that
ϕ(k)
there are numbers co-prime to k in each row, we are done. Similar to the above,
ϕ(s)
for any i, we have (r + id1 s, k) = 1 if and only if (r + id1 s, k1 ) = 1. Moreover, the
numbers in each column are congruent modulo k1 . We know that if a ≡ b (mod k1 ),
then (a, k1 ) = (b, k1 ). This means that the number of integers co-prime to k1 (and thus
k) in all the rows are equal.
ϕ(k)
So, it suffices to prove that there are exactly numbers co-prime to k1 in the
ϕ(s)
first row of numbers, namely, r + is for i = d1 , 2d1 , . . . , k1 d1 . If you look closely, this
is exactly the same thing as the original theorem when we replace n by k, k by k1 ,
and d by s (the only difference is that here, the possible values for i are k1 numbers
d1 , 2d1 , . . . , k1 d1 , while possible values for i in the original theorem are 1, 2, . . . , k).
In the new version of the theorem, which we want to prove, there are k1 numbers of
the form r + is, while this number is k = k1 s in the original theorem. Since s > 1, we
have k1 < k. Take s1 = (s, k1 ) and let k2 and d2 be integers such that k1 = k2 s1 and s =
d2 s1 . With the same method as above, classify the numbers r+d1 s, r+2d1 s, . . . , r+k1 d1 s
as:
r + d 1 d 2 s1 r + 2d1 d2 s1 ... r + k2 d 1 d 2 s1
r + (k2 + 1)d1 d2 s1 r + (k2 + 2)d1 d2 s1 ... r + 2k2 d1 d2 s1
.. .. .. ..
. . . .
r + [(s − 1)k2 + 1]d1 d2 s1 r + [(s − 1)k2 + 2]d1 d2 s1 . . . r + s1 k2 d 1 d 2 s1
216 3.7. MENON’S IDENTITY

With the exact same reasoning as above, we can reduce the problem to showing that
ϕ(k2 )
there are integers co-prime to k2 in the first row of above numbers. If s1 = 1,
ϕ(s1 )
we are done. Otherwise, we have k2 < k1 and we can take s2 = (s1 , k2 ), k2 = k3 s2 ,
and s1 = d3 s2 . If we continue this process, we will finally reach a point where the gcd
becomes one. That is, the sequence s, s1 , s2 , . . . will eventually reach 1. The reason is
that the sequence k, k1 , k2 , . . . is strictly decreasing and contains only positive integers.
This means that for some m, we will have km = 1 and then the sequence stops. In
that point, km−1 = km , which in turn means sm−1 , the gcd of km and sm−1 , equals one.
We have already showed that when this gcd is one, the theorem is true. The proof is
complete.

Theorem 3.7.3 (Menon’s Identity). Let n be a positive integer. Then,

n
X
(i − 1, n) = τ (n)ϕ(n),
i=1
(i,n)=1

where τ (n) is the number of divisors of n.

Proof. We can write the sum as

n
X n
X X
(i − 1, n) = ϕ(d)
i=1 i=1 d|(i−1,n)
(i,n)=1 (i,n)=1
X n
X
= ϕ(d) 1.
d|n i=1
(i,n)=1
i≡1 (mod d)

According to theorem ( ), the second sum in the last line equals ϕ(n)/ϕ(d). There-
fore,
n
X X ϕ(n)
(i − 1, n) = ϕ(d)
i=1
ϕ(d)
d|n
(i,n)=1
X
= ϕ(n)
d|n

Since ϕ(n) is independent from d in the last summation, we have

n
X X
(i − 1, n) = ϕ(n) 1
i=1 d|n
(i,n)=1

= τ (n)ϕ(n),

which is what we wanted.

CHAPTER 3. ARITHMETIC FUNCTIONS 217

3.8 Liouville Function

Liouville function is of high interest in many sections of number theory. As a reminder,
we are going to state the definitions of the functions ω(n), Ω(n), and λ(n) once again.
For more examples, read section ( ).
Definition 3.8.1 (Big Omega Function). Let n be a positive integer. The number
of prime factors of n (with multiplicity) is denoted by Ω(n).
Definition 3.8.2 (Small Omega Function). Let n be a positive integer. The number
of distinct prime factors of n is denoted by ω(n).
Definition 3.8.3 (Liouville Function). For a natural number n, the Liouville func-
tion λ(n) is defined as

λ(n) = (−1)Ω(n) .

Proposition 3.8.4. Let n = pα1 1 pα2 2 · · · pαk k be the prime factorization of a positive
integer n. Then, ω(n) = k and Ω(n) = α1 + α2 + · · · + αk . In other words,
k
X
ω(n) = 1, and
i=1
k
X
Ω(n) = αi .
i=1

Note. ω(1) = Ω(1) = 0, as 1 does not have any prime factors.

Proposition 3.8.5. For a natural number n, Ω(n) = ω(n) if and only if n is square-free.
Proof. If n is square-free, then it is obvious that ω(n) = Ω(n). Now, suppose that
ω(n) = Ω(n) holds for some positive integer n with prime factorization n = pα1 1 pα2 2 · · · pαk k .
In this case,

α1 + α2 + · · · + αk = k.

Notice that all αi are positive integers (1 ≤ i ≤ k) and the above equation only occurs
when all of them are equal to 1. In other words, n is not divisible by square of any
prime and is therefore square-free.
Proposition 3.8.6.
• ω is an additive function,

• Ω is a completely additive function.

The reader should have a decent idea of what to use in order to prove it!
Theorem 3.8.7. Prove that the Dirichlet inverse of Liouville function is the absolute
value of Mőbius function.
218 3.8. LIOUVILLE FUNCTION

Proof. We want to prove that λ ∗ |µ| = ε. From theorem ( ), since both λ and |µ|
are multiplicative, their Dirichlet product is also multiplicative. Therefore, it suffices to
prove that (λ ∗ |µ|)(pα ) = ε(pα ), where p is a prime and α is a positive integer. Notice
that ε(pα ) = 0, so we have to prove that (λ ∗ |µ|)(pα ) = 0. We have
 
α
X pα
(λ ∗ |µ|)(p ) = |µ(d)|λ
d
d|pα

= |µ(1)|λ(pα ) + |µ(p)|λ(pα−1 ) + · · · + |µ(pα )|λ(1)

= |µ(1)|λ(pα−1 ) + |µ(p)|λ(pα−2 )
= (−1)α + (−1)α−1
= 0,

as desired.

The following theorem is due to [ ].

Theorem 3.8.8. Let s be an arbitrary integer. Then, for all integers n > 1 with prime
factorization n = pα1 1 pα2 2 · · · pαk k ,

k (α +1)s
X
s
Y (−1)αi p i +1i
λ(d)d = .
i=1
psi + 1
d|n

Proof. Since both λ(d) and ds are completely multiplicative functions,

P their product is
a multiplicative function. We find from theorem ( ) that λ(d)ds is also multi-
d|n
plicative. Therefore, it suffices to prove the theorem for the case when n is a power of
a prime. Take n = pα for some prime p and positive integer α. Then,
X
s
s
λ(d)ds = λ(1)1s + λ(p)ps + λ(p2 ) p2 + λ(p3 ) p3 + · · · + λ(pα ) (pα )s
d|pα

= 1s − ps + p2s − p3s + · · · + (−1)α pαs .

This is a geometric progression with initial term 1 and common ratio −ps . Therefore,
by theorem ( ) which gives the formula for geometric series,

X α
X
s
λ(d)d = (−1)si psi
d|pα i=0

(−ps )α+1 − 1
=
−ps − 1
(−1)α p(α+1)s + 1
= .
ps + 1

Applying this result to n = pα1 1 pα2 2 · · · pαk k , using multiplicativity of the functions as
CHAPTER 3. ARITHMETIC FUNCTIONS 219

discussed,

X k X
Y
s
λ(d)d = λ(d)ds
d|n i=1 d|pαi
i
(α +1)s
(−1)αi pi i +1
= ,
psi + 1

as desired.

The previous theorem is pretty strong. The following theorem, which is an example
case of the preceding theorem, is useful and important.

Theorem 3.8.9. Let κ be the summatory function of λ. That is, let κ be an arithmetic
function such that
X
κ(n) = λ(d).
d|n

Then,
(
1 if n is a perfect square,
κ(n) =
0 otherwise.

Proof. Take s = 0 in theorem ( ). Then, assuming n = pα1 1 pα2 2 · · · pαk k as the prime
factorization of n,

k
X Y 1 + (−1)αi
λ(d) = .
i=1
2
d|n

If n is a perfect square, that is, if all the αi (for 1 ≤ i ≤ k) are even, the above product
equals one. Otherwise, at least one of αi is odd and the thus 1 + (−1)αi = 0, making
the product equal zero.
n
P
Remark. The sum L(n) = λ(k) has been investigated for a long time. It was first
k=1
conjectured that L(n) < 0 for all integers n > 1. You can check and see that it is true
for small integers. However, in 1980, Minoru Tanaka found that n = 906150257 is the
smallest counter-example to this conjecture.

Theorem 3.8.10. Let n be a positive integer. Then,

X n
λ(n) = µ 2 .
2
k
k |n
220 3.8. LIOUVILLE FUNCTION
X
Proof. Assume that κ(n) = λ(d). Then, by Mőbius inversion formula (theorem
d|n
( )),
X n
λ(n) = µ(d)κ
d
d|n
X n
= κ(d)µ
d
d|n
X n
= µ 2 ,
2
k
k |n

because κ(d) is 1 whenever d is a perfect square (i.e., d = k 2 for some k), and zero
otherwise.

Theorem 3.8.11. Prove that for every positive integer n,

n
X jnk √ 
λ(k) = n .
k=1
k

Proof. The summatory function of λ is κ. Therefore, from theorem ( ),

n
X jnk n
X
µ(k) = κ(k).
k=1
k k=1

We know by theorem ( ) that κ(k) is 1 if k is a perfect square and 0 otherwise. So,

n
P √
κ(k) is exactly ⌊ n⌋.
k=1
CHAPTER 3. ARITHMETIC FUNCTIONS 221

3.9 Exercises
Problem 3.9.1. How many integers between a and b are divisible by n? Note that, a
and b are not necessarily positive.

Problem 3.9.2. Let n be a positive integer. Prove that for any prime p,
∞    
X n n−1
vp (n) = − .
k=1
pk pk

Problem 3.9.3 (Taken from [ ]). Let n be a positive integer. Show for every prime
p that

n ln(n + 1) n−1
− ≤ vp (n!) ≤ .
p−1 ln p p−1

Problem 3.9.4. Let n > 1 be an integer. Prove that (|µ| ∗ 1)(n) = 2n . Here, | · |
denotes the absolute value.

Problem 3.9.5. Prove for any positive integer n that

X µ(d)2 n
= .
ϕ(d) ϕ(n)
d|n

Problem 3.9.6. Let n be a positive integer. Show that

X n Y 1

2
dµ = nϕ(n) 1+ .
d p
d|n p|n

Problem 3.9.7. Prove that for any even positive integer n,

X
µ(d)ϕ(d) = 0.
d|n

Problem 3.9.8. Prove that τ 3 ∗ 1 = (τ ∗ 1)2 .

Problem 3.9.9. Let s be an arbitrary integer. Prove for all integers n > 1 with prime
factorization n = pα1 1 pα2 2 · · · pαk k that

X k
Y
s
µ(d)d = (1 − psi ).
d|n i=1

Problem 3.9.10. For every integer n ≥ 1, show that

X
µ2 (n) = µ(d).
d2 |n
222 3.9. EXERCISES

Problem 3.9.11. Let m and n be positive integers such that n ≥ 2. Prove that
X
µ(d)
d|n
ω(d)≤m

is non-negative if m is even and non-positive if m is odd.

Problem 3.9.12. Prove that the product of all divisors of a positive integer n is nτ (n)/2 .
Problem 3.9.13. Let n be a positive integer. Show that
σ(n)2
≥ τ (n)2 .
n
Hint. Use the fact that
X Xn
σ(n) = d= .
d
d|n d|n

Then exploit the Cauchy–Schwarz inequality.

Problem 3.9.14 (Romania TST 2010). Given a positive integer a, prove that
σ(am) < σ(am + 1) for infinitely many positive integers m.
Problem 3.9.15. Prove that there exist infinitely many positive integers n such that
σ(n) = 2n + 12.
Problem 3.9.16. Let k ≥ 2 be an integer. Prove that there exist no positive integer
n such that σ(n) = nk .
Problem 3.9.17. Show for any three positive integers α, m, and n that
X  mn 
2
σα (m)σα (n) = d σα .
d2
d|(m,n)

Problem 3.9.18 (From [ ]). Let p be a prime that generates the even perfect number
Ep = 2p−1 (2p − 1). Then Ep is expressible as the sum of the cubes of the first n
consecutive positive integers, where n = 2(p−1)/2 .
Problem 3.9.19. Prove that if existing, an odd perfect number would be of the form
12n + 1.
Problem 3.9.20 (Putnam 1976). A positive integer n is called quasi-perfect if σ(n) =
2n + 1. Prove that any quasi-perfect number is the square of an odd integer.
Problem 3.9.21 (Romania TST 2014). Show that a positive integer n, which has
at most two distinct prime factors, satisfies the condition σ(n) = 2n − 2 if and only if
n = 2k (2k+1 + 1), where k is a non-negative integer and 2k+1 + 1 is prime.
Problem 3.9.22 (IMO Longlist 1979). If n has at most 5 distinct prime divisors,
77
prove that σ(n) < 16 n. Also prove that there exists a positive integer n for which
σ(n) < 76
16
n holds.
CHAPTER 3. ARITHMETIC FUNCTIONS 223

Problem 3.9.23. Let n be a positive integer. Show that

n−1  
X 1
ϕ(n) = .
k=1
(n, k)

Problem 3.9.24. For a positive integer n, find the value of the following sum
X
(−1)n/d ϕ(d).
d|n

Problem 3.9.25. For n ≥ 2, show that

σ(n) n π 2 σ(n)
< < .
n ϕ(n) 6 n
Hint. For the right side, use the fact that
Y 1

6
1 − 2 = 2,
p∈P
p π

where the product extends over all primes p.

Problem 3.9.26. Prove that for any integer n ≥ 2,
X 1
k = nϕ(n).
1≤k≤n
2
(k,n)=1

Problem 3.9.27. Prove that for any integer n larger than 6,

√
ϕ(n) ≥ n.

Problem 3.9.28. For all positive integers n, show that

n j k
X n n(n + 1)
ϕ(k) = .
k=1
k 2

Problem 3.9.29. Let a be an arbitrary integer. Prove that for every positive integer
n,
X n
ϕ ad ≡ 0 (mod n).
d
d|n

Problem 3.9.30. Prove that for any integer n ≥ 3,

n(n − 1)
ϕ(2) + ϕ(3) + ϕ(4) + · · · + ϕ(n) ≥ + 1.
4
Problem 3.9.31. Let x be a positive real number. Prove that
r
X  x X jxk
= 2
.
n≤x
n √ n
n≤ x
224 3.9. EXERCISES

Hint. Use theorem ( ).

Problem 3.9.32. Find all positive integers n such that

ϕ(n) + σ(n) = 2n.

Problem 3.9.33. Prove that if σ(n) = 5n than n has more than 5 distinct prime
divisors.
Problem 3.9.34 (Belarus 2010). Find the greatest real number a such that for all
n > 1 the following inequality is true
σ(n) √
≥ a n.
τ (n)
Problem 3.9.35. Show that a positive integer n is prime if and only if σ(n) + ϕ(n) =
n · τ (n).
Problem 3.9.36 (China Western
Mathematical Olympiad 2015). Let k be a
positive integer and take n = 2k !. Prove that σ(n) has at least a prime divisor larger
than 2k .
Problem 3.9.37 (Cono Sur Shortlist 2012). Find all integers 1 < n < 2012 for
which (p(n))2 = σ(n) + 423, where p(n) is the product of all prime divisors of n.
Problem 3.9.38. Prove that for every positive integer n the following inequality holds
n n
X X ln k
τ (k) ln k ≤ 2n · .
k=1 k=1
k

Problem 3.9.39 (Russia 2011). Show that if σ(n) = 5n/3, then σ(5n) = 10n.
Problem 3.9.40. Find all positive integers k and n such that

ϕ(n) · τ (n) · σ(n) = k · (nk − 1)

n
Problem 3.9.41 (Ukraine 2006). Let n be a positive integer and define N = 22 + 1.
Prove that
σ(N )
< 2.
N
√
Problem 3.9.42. Let n be a positive integer such that σ(n) = 2n + 1. Show that n
is an odd number.
Problem 3.9.43. Prove that a positive integer n is prime if and only if ϕ(n)|n − 1 and
n + 1|σ(n).
Problem 3.9.44. Prove that 6|σ(6n + 5) for all positive integers n.
Problem 3.9.45 (China TST 2014). Prove that there exist finitely many positive
integers n satisfying the following two conditions:
CHAPTER 3. ARITHMETIC FUNCTIONS 225

1. τ (n) = a, and

2. n|φ(n) + σ(n).

Problem 3.9.46 (Serbia Additional TST 2012). For every n ∈ N define f (n) as
number of natural numbers m, m ≤ n, for which σ(m) is odd number. Prove that there
are infinitely many natural numbers n, such that f (n)|n.

Problem 3.9.47 (Iran Third Round 2009). We call a permutation π on the set
An = {1, 2, . . . , n} steply constant if the set {π(k) − k|k = 1, 2, . . . , n} has exactly two
elements. Prove that the number of steply constant permutations of An is σ(n) − τ (n).

Problem 3.9.48 (Belarus 1999). For any integer n ≥ 2, prove that

p
σ(n) < n 2τ (n).

Problem 3.9.49 (All Russian Olympiads 2000). A perfect number, greater than
28 is divisible by 7. Prove that it is also divisible by 49.

Problem 3.9.50. Find all positive integers n such that

τ (n)ϕ(n)σ(n) + 2ω(n)
= ϕ(n) + ω(n).
ϕ(n) + σ(n)

Problem 3.9.51 (IMS 2008). Find all natural numbers such that

nσ(n) ≡ 2 (mod ϕ(n))

Bibliography

[1] Masum Billal, A Nice Theorem in Multiplicative Functions, Eureka, 2014, Issue
63.

[2] Kahan, Steven. "Perfectly odd cubes." Mathematics magazine 71.2 (1998): 131.

[3] Sierpinski, Waclaw. Elementary Theory of Numbers: Second English Edition

(edited by A. Schinzel). Vol. 31. Elsevier, 1988.

[4] Panaitopol, Laurenţiu. "Some properties of Liouville’s function." Bulletin mathé-

matique de la Société des Sciences Mathématiques de Roumanie (2001): 365 − 370.

[5] Moll, Victor H. Numbers and functions: from a classical-experimental mathemati-

cian’s point of view. Vol. 65. American Mathematical Soc., 2012.

[6] Dickson, Leonard Eugene. History of the Theory of Numbers. Vol. 256. Chelsea
Pub. Co., 1952.

227
Chapter 4

Primes

Contents
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
. . . . . . . . . . . . . . . . . . . . . . . . 231
. . . . . . . . . . . . . . . . . . . . . . . . 238
. . . . . . . . . . . . . . . 242
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
. . . . . . . . . . . . . . . . . 253
. . . . . . . . . . . . . . . . . . . . . . . 254
. . . . . . . . . . . . . . . . . . . . . . . 260
. . . . . . . . . . . . . . . . . . . . . . . . . 264
. . . . . . . . 268
. . . . . . . . . . . . . . . . . . . . . . . . 272
. . . . . . . . . . . . . . . . 273
. . . . . . . . . . . . . . . . . . . 274
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
. . . . . . . . . . . . . . . . . . . . 279

4.1 Introduction
Prime numbers just might be the most mysterious topic in mathematics. There are
already countless books on the topic. We have already defined prime numbers in chapter
( ). Recall that a positive integer p is a prime if and only if it has exactly two positive
divisors: only 1 and p itself (recall how this lets us deal with the case of 1 automatically).
In this chapter, we are going to explore the properties of primes. This PDF was created
for Amazon.

We will start with a widely discussed topic: Infinitude of primes. Besides pro-
viding Euclid’s proof of the theorem, we will show some other proofs. Many of them
are not very common these days. We try to provide precise history such as who the

229
230 4.1. INTRODUCTION

proof should be accredited to and when etc. As we go on, we will encounter the famous
ingenious proof by Erdős, an elementary proof of Bertrand’s postulate. We will also
discuss primality testing and some relevant theorems. Most of them will be intercon-
nected. But you may be surprised when you see that we have discussed some things at
the end of this chapter which are not quite elementary or Olympiad style topics. Still,
we decided to include them because they let us understand why numbers dance the
way they do. Probably this will not make sense to many people, but we think so. We
would feel really good if we could provide the elementary proof of Prime Number
Theorem by Erdős and Selberg as well but had we done so, we would be way off
topic. Therefore, we will keep the analytic stuff as limited as possible, yet giving an
insight to what make number theorists think that way or what drives them study so
hard.
In section ( ), we will discuss how to list primes efficiently or decide whether an
integer is prime or not, and how to factorize an integer quickly. Now, there are a few
points to clear out in the last statement.
1. Why do we need a list of primes?

2. Why do we need a way to detect primes?

3. Why do we care how quickly we can factorize an integer? Because we can just
factorize 12 as 22 · 3 by hand, right?
If you remember, we asked you to factorize 357879581 (yes, we used this number because
it had larger prime factors than we usually deal with and because we knew the factors ).
If you actually tried doing that without using a computer, you must have cursed us all
the way. Why? Because the smallest prime factor of 357879581 is 479, and the other
one is 747139. As you can see, as the numbers get bigger, there prime factors get bigger
as well. And of course we don’t want to do all that by hand. We shouldn’t do that
either. Computers help us in the computing part, we just need to tell the computer
how to do that. Now, this is where we introduce the idea of algorithm. We used this
word back in chapter ( ), where we first used Euclidean Algorithm. A funny way to
say what algorithm is: the word used by programmers when they don’t want to explain
what they did. People sometimes say that because often algorithms are complex and
not very understandable at first glance. However, algorithm actually means a set of
operations which can define an entire process to do something, and this process is used
for computer programs. Another question may strike you again. Why should we care
about large numbers and determine if they are prime or not? The answer is not directly
related to Olympiad or problem solving. This is necessary for programming purposes
primarily, but they rely on number theoretic results to perform these factorizations or
similar tasks. And they are used in a lot of area such as security. For example, every
1
We accredit both of them, not only for avoiding any controversy, also because we believe they both
had contributions. Specially, after reading some papers (such as [ ]) from a close colleague of both of
them, the authors are convinced they both had their parts in this proof.
2
In fact, this is an integer the first author once thought was prime because he was using trial and
error to determine if it is prime. After he exhausted a lot options, he thought it was actually a prime.
But finally, it was revealed using a computer that it is not. And unless you like a lot of tedious
calculation, you won’t like the trial and error process by hand either.
CHAPTER 4. PRIMES 231

time you log ito Facebook, you use your password and this password is encrypted .
Now, for this encryption, often large integers with large prime factors are used . And
often encryption systems rely on the fact that, some integer that has been used in
the process of encryption, can not be factorized. If they can be factorized the secret
data that was used to turn your password into the code, would be revealed to the
third party and thus, they would know your password. So from this perspective, it is
pretty important. But even if you ignore this practical fact, you can just think about
contributing to the literature of mathematics and enriching it, providing better ways
to factorize so the process is not so tedious anymore. For this reason, we have decided
to include some really nice results and algorithms for prime factorization or primality
testing.
It would be appropriate to mention that, P is the set of primes and pi is the ith
prime, starting with p1 = 2, p2 = 3 and so on unless mentioned otherwise . Also, τ (n)
is the number of positive divisors of n. Riemann’s Zeta function, also simply called the
Zeta function, is a very important subject of interest and has a long interesting history
behind it. The usage and applications of Zeta function is beyond the scope of this book,
but as it is a very useful tool in inspecting number theory problems, we will introduce
a very simple definition of it. This definition needs precision, otherwise it may lead
to confusing conclusions. Therefore, we will only assume the following definition solely
for the use of this book, and not go into any complex details about the validity of the
definition or similar stuff. And in this text, we do not need any such discussion either.
Definition 4.1.1 (Riemann’s Zeta Function). Let s be a a real number larger than
1. The Zeta function of s is defined as
1 1 X1
ζ(s) = + + · · · =
1s 2s i≥1
is

It is one of the most well known functions in number theory. Euler defined it first in
1737 but Riemann is known for his works on this function.

4.2 Infinitude Of Primes

Euclid first proved that the number of primes is infinite. Here we provide some proofs
of this theorem, including a number theoretic version of Euclid. The idea is pretty
interesting and the same thought works for similar types of problems.
Theorem 4.2.1. The number of primes is infinite.
Euclid’s proof. Consider the converse: assume that number of primes is finite. Let
P = {p1 , p2 , . . . , pk } be the set of all primes. Euclid’s idea was to construct a number
3
meaning it is turned into a code so others don’t recognize this if they ever see this data containing
your password, so you should understand why it is so important.
4
we are not going to discuss anything in deep since this is not a computer science or cryptography
book, rather just a short note on why you should care about fast prime factorization
5
Sometimes we may denote the canonical prime factorization as pe11 · · · pekk . It’s important to dis-
tinguish between them.
232 4.2. INFINITUDE OF PRIMES

which has a prime divisor not in P . Consider the number:

N = p1 p2 · · · pk + 1.

N is not a prime, because it is clearly bigger than all elements of P . So, N is composite
and it has a divisor p in P (because P is the set of all primes). However,

(4.1) (N, p) = (p1 · · · pk + 1, pi ) = (1, pi ) = 1

for some pi ∈ P , which is in contradiction with p|N . Therefore, the set of primes is
infinite.

Note. The idea of Euclid was actually to construct a larger prime knowing previous
ones. As you see in the above proof, the product of primes p1 , p2 , . . . , pk plus one is
relatively prime to all of those primes, meaning that it is a prime itself.

Kummer’s proof . Again, it suffices to prove that for any n, there is a larger prime than
n. Consider N = n! + 1. Any prime less than n is coprime to N . Therefore, it must
have a prime divisor greater than n.
Goldbach’s proof. We are done if we can show that there is a strictly increasing infinite
sequence of positive integers a1 , a2 , a3 . . . so that they are pair-wisely coprime. Since
no prime can divide two terms of the sequence, each time a new term appears it will
produce a new prime factor. So, all we have to do is find such a sequence. One way to
n
do it is using Fermat numbers. The nth Fermat number, Fn , is defined as Fn = 22 + 1.
In the following lemma, we will show that any two Fermat numbers are coprime to each
other.

Lemma 4.2.2. If m 6= n, then (Fm , Fn ) = 1.

Proof. Note the identity:

n
F n − 2 = 22 − 1
 n−1   n−2 

= 22 + 1 22 + 1 · · · 22 + 1 (21 + 1)(2 − 1)
= Fn−1 Fn−2 · · · F0

Therefore, if n > m, then Fm |Fn − 2. If p is a prime so that p|Fm and p|Fn , then
p|Fn − 2 and so p|2, which is a contradiction since p has to be an odd prime.
There are other proofs that use the same idea of coprime integers.
Schorn’s Proof. First we will prove the following:

(j(n!) + 1, i(n!) + 1) = 1

for 1 ≤ i < j < n + 1. We can write j = i + k, so 1 ≤ k < n. By Euclidean algorithm,

((i + k)(n!) + 1, i(n!) + 1) = (i(n!) + 1 + k(n!), i(n!) + 1)

= (k(n!), i(n!) + 1).
CHAPTER 4. PRIMES 233

We also know from proposition ( ) that if (a, b) = 1, then (a, bc) = (a, c). Clearly
(n!, i(n!) + 1) = 1 since i(n!) + 1 leaves a remainder of 1 when divided by n!. Therefore,
((i + k)(n!) + 1, i(n!) + 1) = (k(n!), i(n!) + 1)
= (k, i(n!) + 1).
Since k < n, we also have that k divides n!, so i(n!) + 1 leaves a remainder of 1 when
divided by k too. Finally, we have
(j(n!) + 1, i(n!) + 1) = ((i + k)(n!) + 1, i(n!) + 1) = (k, i(n!) + 1) = 1.
From this we can say, the integers i(n!) + 1 for 1 ≤ i ≤ n are coprime. And so, we are
done.
This elegant proof is due to J. Braun (1896).
Proof by Braun. Assume that primes are finite, and p1 , p2 . . . , pk are all of them. Let
P = p1 p2 · · · pk and set
1 1 a
(4.2) + ··· + = .
p1 pk P
Note that
a 1 1 1
> + +
P 2 3 5
31
= > 1.
30
So a > P . Obviously, a has a prime divisor p. Since P is the product of all primes, p
must divide P . Let p = pi for some 1 ≤ i ≤ k and rewrite equation ( ) to obtain
P P P
(4.3) a = + ··· + + ··· + .
p1 pi pk
Obviously, pi | pPj for all j 6= i. On the other hand, p divides a. Equation ( ) now forces
pi | pPi , which is a contradiction.
Here is a combinatorial proof by Perott, which dates back to almost 1801 − 1900.
Perott’s proof. We will use the fact that if a > b then a1 < 1b , specially, 1
n+1
< 1
n
for
n ≥ 1. Now
X1 X1
= 1 +
i≥1
i2 i≥2
i2
X 1
< 1+
i≥2
i(i − 1)
 
X 1 1
= 1+ −
i≥2
i − 1 i
     
1 1 1 1 1
= 1+ 1− + − + − + ···
2 2 3 3 4
= 1 + 1 = 2.
234 4.2. INFINITUDE OF PRIMES

Therefore,
X1
(4.4) = 2 − m,
i≥1
i2

for some positive real m. Let’s get to the proof. Like before, we assume there are only
k primes p1 , p2 . . . , pk . Take n = p1 p2 · · · pk and any integer N > n. Since there are
no primes besides these k, any square-free number must be a divisor of n. Therefore,
there are 2k square-free numbers. Let p be a prime. The number of positive integers
less than or equal to N which are divisible by p2 is ⌊N/p2 ⌋. So, the number of positive
integers less than or equal to N which are divisible by any of p21 , p22 , . . ., or p2k is less
than
      k  
N N N X N
+ 2 + ··· + 2 = .
p21 p2 pk i=1
p 2
i

Since any number is either square-free or non-square-free, we have

k  
k
X N
N ≤ 2 +
i=1
p2i
k
X N
< 2k +
i=1
p2i
k
k
X 1
(4.5) = 2 +N .
p2
i=1 i

From equation ( ), we get

k
X 1 1 1 1
2
= 2 + 2 + ··· + 2
p
i=1 i
2 3 pk
pk
X 1 X1
< <
i=2
i2 i≥2
i2
X1
= −1
i≥1
i2
= 1 − m.
Substitute this into equation ( ),
k
X
k 1
N < 2 +N
p2
i=1 i
< 2k + N (1 − m).
Rewriting the above inequality, we get N m < 2k . Note that 2k is fixed, whereas we can
make N m as large as we want since N can be any integer larger than n. So, for those
N , we get a contradiction, W 5 (Which Was What We Wanted).
7
These are actually non-square-free integers up to N .
CHAPTER 4. PRIMES 235

The next proof uses Zeta function. But we need some more theorems to state it.
The following theorem is due to Euler.

Theorem 4.2.3.
1 1 Y p
ζ(1) = 1 + + + ··· = .
2 3 p∈P
p − 1

Proof. Euler investigated the sum (which is known as the Harmonic Series):

1 1 1
S= + + ··· + + ··· .
1 2 n
1 1
Consider the sum in terms of prime factorization. Obviously, 1, , 2 , . . . are part of
2 2
1 1 1 1
the series. So are , 2 , . . . and , 2 , . . . and so on. If you understood the fact we
3 3 5 5
1 1 1 1 1 1 1 1 1
showed above, note that · gives . Similarly, 2 · = and · = and so on.
2 3 6 2 3 12 3 5 15
We know that any number can be written as a product of primes in a unique
way. Therefore, when we are multiplying some powers of primes, we will get a unique
number. In other words, the same number won’t appear twice. As an example, notice
the following sum:
   
1 1 1 1
S1 = 1 + + 2 + · · · · 1 + + 2 + · · ·
2 2 3 3
1 1 1 1 1 1 1 1
=1+ + 2 + + + 2 + 2+ 2
+ 2 2 + ···
2 2 3 2·3 2 ·3 3 2·3 2 ·3
1 1 1 1 1 1 1 1
=1+ + + + + + + + + ...
2 4 3 6 12 9 18 36
Unique prime factorization guarantees that none of 2, 4, 6 or 18 will appear anywhere
in the series again. That is, any number of the form 2i 3j will appear exactly in this
series. Similarly, if we considered all the numbers generated by 2i 3j 5k , we would have
numbers like 30, 60 or 90 exactly once in the series. So, going this way, we can see that
1 1
the sum S is nothing but the product of sums 1 + + 2 + · · · for all primes p. So
p p

1 1 1
S = 1 + + + + ···
 2 3 4     
1 1 1 1 1 1
= 1 + + 2 + ··· · 1 + + 2 + ··· · 1 + + 2 + ··· ···
2 2 3 3 5 5
 
Y 1 1
(4.6) = 1 + + 2 + ··· ,
p∈P
p p

where P is the set of all primes. Back in high school, we learnt that the infinite geometric
series 1 + r + r2 + · · · where the ratio r has absolute value less than 1, has a finite sum
236 4.2. INFINITUDE OF PRIMES
1 1
. Here, r = < 1, and hence
1−r p
1 1 1
1+ + 2 + ··· =
p p 1 − p1
p
= .
p−1
Replacing this in equation ( ), we get the desired result
1 1
S = 1 + + + ···
Y 2p 3
= .
p∈P
p − 1

Euler found a general result for ζ(s) for any positive integer s. We have stated this
result in the following theorem. The proof is analogous to the proof of the previous
theorem.
Theorem 4.2.4.
1 1 Y ps
ζ(s) = 1 + + + · · · =
2s 3s p∈P
ps − 1

1 1
Theorem 4.2.5. The series S = 1 + + + · · · diverges, i.e., it does not have a finite
2 3
sum.
Proof. We can write S as
   
1 1 1 1 1 1 1 1
S= + + + + + + + + ···
1 2 3 4 5 6 7 8
   
1 1 1 1 1 1 1 1
> + + + + + + + + ···
1 2 4 4 8 8 8 8
1 1 1 1
= 1 + + + + + ···
2 2 2 2
= 1 + 1 + ···

So the sum diverges.

You may think that ζ(2), ζ(3), . . . all diverge too. Wrong! Using calculus Euler also
proved the following theorem:
π2
Theorem 4.2.6 (Euler’s ζ(2) theorem). ζ(2) = 6
. In other words,

1 1 π2
1+ + + · · · = .
22 32 6
We are now ready to prove the infinitude of primes using Zeta function.
CHAPTER 4. PRIMES 237

Proof using Zeta Function. In theorem ( ), it is already proved that S is infinite.

A series diverges if it has an infinite sum. If the number of primes is finite, then the
p
product of all would be finite too. But it gives us a contradiction. Thus, the
p−1
number of primes must be infinite.
We provide yet another proof due to Euler. The proof was published after his death.
The proof uses multiplicative property of Euler’s Totient function.
Proof using Euler function. Let P be the product of all primes (since they are finite,
P is finite too). Assume that the primes are p1 , p2 , . . . , pk and they are sorted, i.e.,
2 = p1 < 3 = p1 < · · · . Then P = p1 · p2 · · · pk and P is square-free as well. Using the
formula of Euler function,

ϕ(P ) = (p1 − 1) · (p2 − 1) · · · (pk − 1)

≥ 1 · 2···2
≥ 2k−1
≥2

if k > 1. Since 2, 3 are primes, evidently, k ≥ 2. So the last line holds true. This
implies that ϕ(P ) is at least 2, and there are at least two positive integers less than or
equal to P which are coprime to P . If we discard 1, there is at least one other positive
integer which is coprime to P . That positive integer must have another prime divisor
which does not divide P . Now the claim follows.
In the previous discussion, we have shown that there are infinitely many primes in
several different ways.
Theorem 4.2.7. There are infinitely many primes of the form 4m + 3.
Proof. We proceed the same way as Euclid did. Let p1 , p2 , . . . , pk be all the primes of
the form 4m + 3. Consider the number N = 4p1 p2 · · · pk − 1. Clearly, N ≡ 3 (mod 4).
According to theorem ( ) in chapter ( ), N has at least one prime factor p which
is of the form 4m + 3. This prime p divides N , so it is coprime to N − 1 = 4p1 p2 · · · pk ,
which means that p is none of those p1 , p2 , . . . , pk . Therefore, another prime p of the
form 4m+3 exists. This is a contradiction. So, the number of such primes is infinite.
Theorem 4.2.8. There are infinitely many primes of the form 4n + 1.
Proof. Let’s say the number of primes of this form is finite. Call these primes p1 , p2 , . . . , pk .
Consider the number N = 4p21 · · · p2k + 1. Using corollary ( ) of chapter ( ), we get
that every divisor of N is of the form 4t + 1. Thus, a prime divisor p of N must be of
the same form. The contradiction follows.
Theorem 4.2.9. Let p be a prime. There are infinitely many primes of the form pn+1.
Proof. The theorem is obvious for p = 2 since all primes are odd. Assume that p is
odd. Let us rephrase the theorem: for each prime p, there are infinitely many primes
q such that q ≡ 1 mod p. Let X ≥ 2 be an integer. We know from theorem ( )
p −1
that any prime divisor q 6= p of XX−1 is either p or 1 (mod p).
238 4.3. FORMULA FOR PRIMES

For the sake of argument, suppose that q1 , q2 , . . . , qn are the only primes which are
1 mod p. Set X = pq1 q2 · · · qn and consider the number

Xp − 1
N=
X −1
(pq1 q2 · · · qn )p − 1
=
pq1 q2 · · · qn − 1

N is an integer which is not divisible by any of the qi or p and is greater than 1. So N

has a prime divisor, say r. This r must be congruent to 1 modulo p. Contradiction!

Theorem 4.2.10. let p > 2 is a prime, then there are infinitely many primes q such
that q is a quadratic residue modulo p.

Proof. According to previous theorem, there are infinitely many primes q such that
q ≡ 1 (mod p). So, all these primes are quadratic residues modulo p and we are
done.

You might have already conjectured that there are infinitely many primes of the
form an + 1. Even more generally an + b, where a and b are coprime positive integers.
And luckily, this is true and Dirichlet was the first one to prove it. Though the proof
of this theorem is way beyond the scope of this book. It is even accepted in many
mathematics competitions. You should still try to avoid using it. Use it only if you find
no other way. For most of the problems, there is a solution to that does not require
a high level theorem like this. Readers are highly encouraged to try for a different
solution even if it makes their lives a lot harder.

Theorem 4.2.11 (Dirichlet’s Theorem on Arithmetic Progressions). If a and b

are two coprime positive integers, then there are infinitely many primes the arithmetic
progression

a + b, 2a + b, 3a + b, . . .

In other words, there are infinitely many primes of the form an + b.

4.3 Formula For Primes

Mathematicians have been trying to find a closed form for primes for a long time. But
this was such a mystery that many mathematicians thought it is not possible to find
a formula for primes. You may have thought so too! Whenever someone tries to find
a formula for primes, they tend to go for polynomials first. Our sympathies for them.
Because the following theorem tells us that we can not find a non-constant polynomial
which will always output a prime (for positive integer inputs of course).

Theorem 4.3.1. There is no non-constant polynomial P (x) with integer coefficients

such that P (n) is a prime for all integers n.
CHAPTER 4. PRIMES 239

Proof. Let P be a polynomial that generates only primes. Then P (0) = p for some
prime p. That is, P (x) looks like

P (x) = an xn + an−1 xn−1 + · · · + a1 x + p.

Put x = kp in the above equation. We find that p|P (kp) for all integers k. However,
since P (kp) is a prime, we must have P (kp) = p for all integers k. Now consider the
polynomial Q(x) = P (px) − p. All integers are roots of Q(x), which is impossible unless
Q(x) is the zero polynomial. So P (px) = p for all real numbers x. Hence, P (x) is
constant.

Back to the original question: are there any closed form for pn ? Is there an expression
which always generates a prime? The answer is yes for both of them. However, here
we will concentrate on the number of primes mostly . It makes sense that the formula
for primes and the prime counting function are close to each other. Because if you
get a good formula for π(n), the number of primes less than or equal to n, you can
replace n with pk for some k and get a recursive formula for primes, that is in terms of
previous primes or positive integers less than pk . However, recursive formulas do not
look pretty at all. Therefore, we focus on finding π(n) rather than a direct formula for
pn . The following theorem is one of those ideas which can give us a formula for the
prime counting function, due to the first author (unpublished). Let’s see the theorem
and then explain the idea behind it.

Theorem 4.3.2 (Masum Billal). The number of primes less than or equal to n can
be obtained as
 
 
n  
X  2 
π(n) =  
 Pi  i   i − 1  
i=2
j=1 −
j j

Don’t frown just because it looks ugly! It is actually very simple. Let us slowly
proceed how we can get to this expression.

Proof. First idea: assume f (i) = 1 if i is prime, otherwise 0. Then we will have
n
X
π(n) = f (i)
i=2

This is pretty obvious. Each time we get a prime we are just adding 1 to the sum.
All we have to do is find a good expression for f (i) that is computable in terms of i.
Remember that a prime has exactly 2 divisors. And any positive integer greater than
1 has at least two divisors. Therefore, if τ (i) is the number of divisors of i, τ (i) ≥ 2 for
8
We would like to discuss the second question as well but it requires some analytical number theory,
which of course is out of our scope. If you are an interested reader, you can study Ingham’s theorem,[ ]
Mill’s theorem[ ] and Niven’s theorem[ ].
240 4.3. FORMULA FOR PRIMES

i > 1. This gives us ⌊2/τ (i)⌋ = 0 if i is composite, otherwise 1. Since for composite i,
τ (i) > 2. Now, the formula for f (i) becomes
 
2
f (i) =
τ (i)
But this is still not computable in terms of i. We employ the same idea again, we add
1 to τ (n) each time we get a divisor of n. How do we do that? Assume that if i is a
divisor of n then tn (i) = 1, otherwise 0. Then,
n
X
τ (n) = tn (i)
i=1

Finding tn (i) can be easy. For i < n, we need to add 0 when i doesn’t divide n, otherwise
1. Assume that n = ik + r with r < i and n − 1 = il + s with s < i. If i divides n
then r = 0 and we would have that n − 1 = il + s = ik − 1. Thus, ik − il = s + 1
with s + 1 ≤ i. But i(k − l) = s + 1 gives us s + 1 ≥ i since i|s + 1 and s + 1 is a
positive integer, k > l (why?). This forces s + 1 = i and k − l = 1. The nicer news is
that k − l = 1. And if i didn’t divide n, we would have k = l (prove it) or k − l = 0.
Okay, that’s good news. We have found our characteristic function tn (i). What is the
meaning of k and l in terms of i and n? k = ⌊n/i⌋ and l = ⌊(n − 1)/i⌋, so we get
jnk n − 1
tn (i) = − .
i i
This completes the proof.
Have you ever thought about finding the number of primes not exceeding n yourself?
This is actually a very intriguing question for most of the people interested in number
theory, even for curious school students. At first it seems impossible to find a closed form
in such a case. However, as you think more, you can find different ways to proceed.
The above one is an example. This should enable you to find one as well. Here is
another example, and you may be surprised at this approach. In fact, we have used it
before when we tried to find the number of coprime integers less than or equal to n.
The idea is similar, in a sense that it is recursive in a way. Since it is troublesome to
directly find the number of primes, we will do exactly the opposite. We will find the
number of non-primes not exceeding n. Then we can just subtract it from n. Now,
we intend to find the number of positive integers m such that m = ab with√a, b > 1.
More specifically, we can say that the smallest prime divisor does not exceed √ n (recall
this from chapter ( )). Let p1 , p2 , . . . , pk be the primes not exceeding
√ n in increasing
order. That is, pk is the largest prime less than or equal to n (this is why we said
this approach is recursive). Any composite positive integer not exceeding n must have
a prime divisor from this set {p1 , p2 , . . . , pk }.
Again, this is a repetitive problem we encountered before. How many positive
integers not exceeding n are divisible by p1 ? The number is ⌊n/p1 ⌋. The same goes for
p2 , . . . , pk . So, the total number of non-prime positive integers should be
   
n n
+ + ··· .
p1 p2
CHAPTER 4. PRIMES 241

However, the positive integers that are multiple of both p1 and p2 were counted twice
in this sum. So, we need to subtract them. Now it becomes
       
n n n n
+ + ··· − − − · · · all possible pairs
p1 p2 p1 p2 p1 p3
Again, when we subtracted them all, the multiples of p1 p2 p3 or p3 p4 pk all vanished from
the calculation. To rectify that mistake, we need to add the number of multiples of
three primes (all possible combinations of course). Now it looks like
           
n n n n n n
+ + ··· − − − ··· + + + ···
p1 p2 p1 p2 p1 p3 p1 p2 p3 p1 p3 p4
Going this way, we see that, if the number of primes taken into account is even, we add
it, subtract otherwise. Hence, we get the following theorem .
Theorem 4.3.3.
√ Let n be a positive integer and p1 , p2 , . . . , pk be the primes less
√ than
or equal to n. If the number of primes not exceeding n is π(n), then π(n) − π( n) + 1
is
(4.7)           
n n n n n k n
n− − − + ··· + + + · · · + (−1) .
p1 p2 p3 p1 p2 p1 p3 p1 p2 · · · pk
In other words, and more generally, if π(x) for any positive real x ≥ 2 is the number of
primes not exceeding x, then,
√ Xx X  x  X  x 
(4.8) π(x) − π( x) + 1 = ⌊x⌋ − + − + ··· ,
p
pi p <p
pi pj p <p <p
pi pj pk
i i j i j k

√
where the sums are taken over all primes less than or equal to x.
√
√Notice that there is a π( n) here. Because when
√ we used primes less than or equal
to n, we missed all √ the primes that
√ are below n. So we should subtract the number
of primes less than n, which is π( n) − 1. Let us discuss this a bit further. We claim
that
P
√ X jnk
(4.9) π(n) − π( n) + 1 = µ(i) ,
i=1
i

where P = p1 p2 · · · pk . This is probably not obvious to you, so we explain how the above
fomula is obtained . It is now a very good time to mention a point the importance of
the Mőbius µ function defined in defintion ( ) as:

1
 if n = 1,
µ(n) = 0 if n is divisible by p2 for some prime p,


(−1)k if n = p1 p2 · · · pk .
9
In the final examination of senior year, Masum faced the following question: Provide an approx-
imation formula for the number of primes not exceeding N . He proved theorem ( ) instead as an
exact formula in the exam.
10
Thanks to Amin Soofiani for reminding us to add some more explanation here.
242 4.4. BERTRAND’S POSTULATE AND A PROOF

We stated in the previous chapter that Mőbius comes handy when dealing with inclusion–
exclusion arguments. Here, we have a nice one. We will investigate equation ( ) term
by term. The first term is n, obtained from i = 1. For i = pj , the contribution is
µ(pj )⌊n/pj ⌋ = −⌊n/pj ⌋ (here, 1 ≤ j ≤ k). That’s exactly the first group of terms in
equation ( ). But what about the case when i is not a prime? Well, if it’s a square-free
number, i.e., if it is of the form i = pi1 pi2 · · · pis , where {i1 , i2 , . . . , is } ⊆ {1, 2, . . . , k},
then, we get a contribution of
   
n s n
µ(pi1 pi2 · · · pis ) = (−1) .
pi1 pi2 · · · pis pi1 pi 2 · · · pis
Otherwise, if i is not square-free, then µ(i) = 0 and there would be no contribution
from that term. Therefore, we get exactly the same sum as in the long ( ).
While on the topic, we should mention the Meissel-Lehmer Method. It is a sieve
√
method based on this theorem, that ultimately provides a way√to compute pn in O( n)
complexity i.e. a function that does not grow faster than n. This is a very good
improvement for computing π(x) and pn .
Note. The study of finding number of primes has gradually morphed into a new branch
called Sieve theory. This is an interesting part of analytical number theory but not only
it is out of our scope, the authors have yet to study a lot in that area. Regardless, it
makes an interesting point for the readers to be interested in. Curious minds should
try going deeper in analytic number theory.

4.4 Bertrand’s Postulate and A Proof

Bertrand’s postulate is a very nice and influential theorem in number theory. Joseph
Bertrand first conjectured it, but he couldn’t prove it entirely. Later, Chebyshev proved
it, using analytic number theory tools. S. Ramanujan proved it (see [ ]) using properties
of Gamma function, which is beyond the scope of this book. However, Erdős proved it
in an elementary way. We will see that proof here. On another note, this elementary
proof given by Erdős (see [ ]) was the first paper he published! The proof is so beautiful
that we could not keep this theorem along with others, instead we gave it a separate
section!
There are many formulations of this theorem. All of them are equivalent.
Theorem 4.4.1 (Bertrand’s Postulate). For all integers n > 1, there is a prime p
so that n < p < 2n.
Theorem 4.4.2 (Alternative Formulations of Bertrand’s Postulate).
• Let pn denote the nth prime number, starting from p1 = 2. Then

pn+1 < 2pn .

• For any integer n > 1, we have

n
π(n) − π ≥ 1.
2
CHAPTER 4. PRIMES 243

We need to show some lemmas in order to prove Bertrand’s postulate. At first they
all seem to be unrelated. Keep reading back and forth until you see the motivation
behind all these lemmas. When you realize, see if you feel thunderstruck. You should!
If this does not make you wonder how a human being can think that far, we do not know
what will. And then you will probably understand why Erdős is our most favorite.

Lemma 4.4.3. For any positive integer n,

 
2n 4n
≥ .
n 2n + 1

Proof. From binomial the theorem, we already know that

     
2n 2n 2n 2n
(1 + 1) = 1+ + ··· + + ··· + .
1 n 2n
2n

2n


Since the binomial coefficients exhibit a symmetry , i.e., since k
= 2n−k
, all other


terms in the above sum are smaller than 2nn
. Therefore
 
2n 2n
2 ≤ (2n + 1) ,
n

which is what we wanted.

In section ( ) of previous chapter, we defined vp (n) to be the highest power of a

prime p which divides n.

Lemma
4.4.4. Let n be a positive integer and let 2n/3 < p ≤ n be a prime. Then
p ∤ 2n
n
.

Proof. We have
   
2n (2n)!
vp = vp = vp ((2n)!) − vp ((n!)2 ) = vp ((2n)!) − 2vp ((n!)).
n (n!)2

Note that 2n/3 < p means 2n < 3p, and so the only multiples of p which appear in
(2n)! are p and 2p. Hence vp ((2n)!) = 2. Also, p < n immediately gives vp ((n!)) = 1.
Therefore
 
2n
vp = vp ((2n)!) − 2vp ((n!)) = 2 − 2 · 1 = 0.
n

2n


Lemma 4.4.5. Let n be a positive integer. Let p be any prime divisor of N = n
.
Then pvp (N ) ≤ 2n.
244 4.4. BERTRAND’S POSTULATE AND A PROOF

Proof. Let α be the positive integer for which pα ≤ 2n < pα+1 . Then using theorem
( ) of chapter ( ),
α   α  
X 2n X n
vp (N ) = vp ((2n)!) − 2vp (n!) = i
−2
i=1
p i=1
pi
α     X α
X 2n n
= i
−2 i ≤ 1 = α.
i=1
p p i=1

The last line is true because for a rational x, ⌊2x⌋ − 2⌊x⌋ ∈ {0, 1}. Therefore, pvp (N ) ≤
pα ≤ 2n.
Lemma 4.4.6.
2n+1

Let n be a positive integer. Any prime p with n + 2 ≤ p ≤ 2n + 1
divides n .
Proof. Since p > n + 1,
   
2n + 1 (2n + 1)!
νp = vp = vp ((2n + 1)!) − vp (n!) − vp ((n + 1)!) = 1.
n (n!)(n + 1)!

Lemma 4.4.7. For any positive integer n,

 
2n + 1
≤ 22n .
n



Proof. From binomial theorem and the fact that 2n+1 n
= 2n+1
n+1
,
       
2n+1 2n + 1 2n + 1 2n + 1 2n + 1
(1 + 1) =1+ + ··· + + + ··· +
1 n n+1 2n + 1
     
2n + 1 2n + 1 2n + 1
≥ + =2 .
n n+1 n
This finishes the proof.
The following lemma is really a nice one, and the proof requires a good insight.
Lemma 4.4.8. The product of all primes less than or equal to n is less than or equal
to 4n .
Proof. We will use induction. The proof is trivial for n = 1 and n = 2. Assume it is
true for all positive integers up to n − 1. We will show that it is also true for n.
If n is even and greater than 2, n is definitely not a prime. Thus,
Y Y
p= p ≤ 4n−1 < 4n .
p≤n p≤n−1

Now, assume that n is odd. Take n = 2m + 1. We have

! !
Y Y Y Y
p= p= p p .
p≤n p≤2m+1 p≤m+1 m+2≤p≤2m+1
CHAPTER 4. PRIMES 245
Q
By induction hypothesis, the first product, p, is less than or equal to 4m+1 . From
p≤m+1
lemma
( ), we know that any prime p such
Q that m + 2 ≤ p ≤ 2m + 1 divides

2m+1
m
. Therefore, the second product, p, is less than or equal to 2m+1
m
.
m+2≤p≤2m+1
Combining these results with lemma ( ), we get
 
Y 2m + 1
p ≤ 4m+1 ≤ 4m+1 22m = 42m+1 = 4n .
p≤n
m

Hence, the lemma is also true for n and we are done.

We are ready to prove Bertrand’s postulate.
Proof of Bertrand’s postulate. We want to show that for any positive integer n, there
exists a prime p such that n < p ≤ 2n. Assume the converse, i.e., suppose that there
exists some n for which

there is no prime p with n < p ≤ 2n. We will find an upper
bound for N = 2n n
and seek for a contradiction. Let us divide the prime divisors of N
into two groups:
√
• Consider all prime divisors of N , say p, such that p ≤ 2n. Let p1 , p2 , . . . , pk be
√ vp (N )
such primes. Clearly, k ≤ 2n. According to lemma ( ), pi i ≤ 2n (for
1 ≤ i ≤ k). Therefore,
k
Y √
vpi (N ) 2n
pi ≤ (2n) .
i=1
√
• Consider all prime divisors of N which are larger than 2n. Let q1 , q2 , . . . , qm
vq (N )
be such primes. Again, by lemma √ ( ), we must have qi i ≤ 2n (where
1 ≤ i ≤ m). However, since qi > 2n, we find that vqi (N ) = 1 for all i.
Now, by our hypothesis, there are no primes p such that n < p ≤ 2n. On the
other hand, lemma ( ) says that there
√ are no prime divisors of N such that
2n/3 < p ≤ n. Altogether, we find that 2n < qi ≤ 2n/3 for 1 ≤ i ≤ m. Hence,
m
Y m
Y Y
vqi (N )
qi = qi = p.
√
i=1 i=1 2n<p≤2n/3
p|N

k m
Q vpi (N ) Q vqi (N )
We now use the fact that N = pi · qi , where pi and qi are as defined
i=1 i=1
above. According to what we have found,
k
Y m
Y √ Y
vp (N ) vqi (N ) 2n
N= pi i · qi ≤ (2n) · p
√
i=1 i=1 2n<p≤2n/3
p|N
√ Y
2n
≤ (2n) · p
p≤2n/3
√
2n
≤ (2n) · 42n/3 .
246 4.4. BERTRAND’S POSTULATE AND A PROOF

Note that we have used lemma ( ) for writing the last line.
Combining this with the result of lemma ( ), we see that

4n √
(4.10) ≤ (2n) 2n · 42n/3 .
2n + 1
However, this inequality can hold only for small values of n. Actually, one can check
that the inequality fails for n ≥ 468. For n < 468, one can check that

2, 3, 5, 7, 13, 23, 43, 83, 163, 317, 631

is a sequence of primes, each term of which is less than twice the term preceding it.
Therefore, any interval {n + 1, n + 2, . . . , 2n} with n < 468 contains one of the primes
in this sequence.
Hence, we have reached the contradiction we were looking for. This means that
there always exist a prime p such that n < p ≤ 2n for any positive integer n. The proof
is complete.

Theorem 4.4.9. For any positive integer n, the set S = {1, 2, . . . , 2n} can be parti-
tioned into n pairs (ai , bi ) so that ai + bi is a prime.

Before we show the proof, readers are highly encouraged to prove it themselves.
This is the kind of theorem that shows how good human thinking can be.

Proof. We will proceed by induction. The theorem is clearly true for n = 1 since
1 + 2 = 3, a prime. Assume that the theorem is true for all k < n and we can split the
set {1, 2, . . . , 2k} into pairs with a prime sum. By Bertrand’s postulate, there is a prime
p with 2n < p < 4n. Let p = 2n + m, where m must be odd since p is odd. Consider
the set {m, m + 1, . . . , 2n}. It has an even number of elements. Also, we can make
pairs of (m, 2n), (m + 1, 2n − 1), . . . with sum p, which is a prime. Now we only have
to prove that the set {1, 2, . . . , m − 1} can be paired into elements with a prime sum.
This is true by induction hypothesis because m − 1 < 2n. The proof is complete.

Problem 4.4.10. Let n > 5 be an integer and let p1 , p2 , . . . , pk be all the primes smaller
than n. Show that p1 + p2 + · · · + pk > n.
P
Solution. We first show by induction that ki=1 pi > pk+1 for k ≥ 3. The base case,
P
k = 3 is true because 2 + 3 + 5 > 7. Assume that ki=1 pi > pk+1 , then by the first
alternative form of Bertrand’s postulate stated in theorem ( ),
k+1
X k
X
pi = pk+1 + pi > 2pk+1 > pk+2 ,
i=1 i=1

and the induction is complete. Now, since pk < n ≤ pk+1 , we have

k
X
pi > pk+1 ≥ n.
i=1
CHAPTER 4. PRIMES 247

Problem 4.4.11 (China 2015). Determine all integers k such that there exists in-
finitely many positive integers n satisfying
2n

n+k 6 .
n

Solution. We will show that the problem statement holds for all integers k 6= 1. Note
that for k = 1, we have
     
2n 2n 1 2n
− = ,
n n+1 n+1 n


and therefore n + 1| 2n
n
. Assume that k > 1. By Bertrand’s postulate, there exists an
odd prime p such that k < p < 2k. Choose n = (p − k) + pm for any positive integer
m. From theorem ( ), we can write
 
2n
vp = vp ((2n)!) − 2vp ((n!))
n
2n − sp (2n) n − sp (n)
= −2·
p−1 p−1
2sp (n) − sp (2n)
= .
p−1
Since 2n = 2(p − k) + 2pm and 2(p − k) < p, it follows that sp (2n) = 2(p − k) + 2 =
(try to write the base p representation of n and 2n to
2sp (n)


see why). Consequently,
vp 2n n
= 0. However, p | n + k, so we have n + k ∤ 2n
n
for infinitely many n, as
desired.
For negative k, one can choose n = −k + pm for an odd prime p > |2k| (which exists
by Bertrand’s postulate)


and any positive integer m. In a similar

manner as above, one
2n 2n
obtains vp n = 0, but p | n + k. Consequently, n + k ∤ n . The proof is complete.

After the theorem was proved, number theorists tried to tighten the interval. Also,
a question was raised regarding the general case.

Problem 4.4.12. Let c be a real number. What is the minimum value of c such that,
there is always a prime between n and n + cn for positive integers n > 1?

Nagura [ ] proved the case for c = 1/5.

Theorem 4.4.13 (Nagura). For x ≥ 25, there is always a prime number between x
and 6x/5.

The proof uses a property of gamma function (a function involving the gamma
function turns out to be a prime counting function). We will not be proving the
improvements or generalizations, but they are worth mentioning. The general case
of this theorem would be like this:

Problem 4.4.14. Let k be a positive integer. Does there always exist a prime between
kn and (k + 1)n?
248 4.4. BERTRAND’S POSTULATE AND A PROOF

Bachraoui [ , Thm 1.3] proved the case k = 2. The idea is a variation of Erdős’s
proof.
Theorem 4.4.15 (Bachraoui). For a positive integer n > 1, there is always a prime
in the interval [2n, 3n].
Andy Loo [ ] proved the case for k = 3 without using prime number theorem or
any deep analytical method.
Theorem 4.4.16 (Loo). For a positive integer n ≥ 2, there is always a prime in the
interval (3n, 4n).
Conjecture 4.1 (General Bertrand’s Postulate). For a positive integer n ≥ 2 and
a positive integer k ≤ n, there always exists a prime p in the interval [kn, (k + 1)n].
If conjecture ( ) can be proven, we can prove some other conjectures using it. Let
us assume that the conjecture is true and it is in fact a theorem. Let us see how we
could use it to prove other conjectures. The ideas are due to Sambasivarao [ ].
Theorem 4.4.17 (Legendre’s theorem). Assuming general Bertrand’s postulate is
true, there always exists a prime in the interval [n2 , (n + 1)2 ].
As we stated before, it has an improvement and we will prove that instead.
Theorem 4.4.18 (Mitra’s theorem). Assume the general Bertrand’s postulate. There
exists at least two primes in the interval [n2 , (n + 1)2 ].
Proof. From theorem ( ), we know that there exists a prime in the interval [kn, (k +
1)n]. Set k = n and we get a prime p such that n2 ≤ p ≤ n2 +n. Again, there is a prime
in the interval [k(n + 1), (k + 1)(n + 1)]. Let q be a prime such that k(n + 1) ≤ q ≤
(k + 1)(n + 1) and set k = n. Now we have n2 ≤ p ≤ n2 + n and n2 + n ≤ q ≤ (n + 1)2 .
Combining them we have n2 ≤ p ≤ q ≤ (n + 1)2 since n2 + n ≤ (n + 1)2 . There will
always be two but not one. The reason is p = q can not occur. That would only be
possible if n2 + n = p = q which is not prime for n > 1 (why?). For n = 1, we can
check manually that 2, 3 are between 1 and 4. Therefore, the theorem holds true for all
n.
Theorem 4.4.19 (Brocard theorem). Assume the general Bertrand’s postulate. For
each n > 1, there are at least 4 primes in the interval [p2n , p2n+1 ].
Proof. We will keep using theorem ( ) repeatedly. There are primes q, r, s, t such that
p2n ≤ q ≤ pn (pn + 1),
pn (pn + 1) ≤ r ≤ (pn + 1)2 ,
(pn + 1)2 ≤ s ≤ (pn + 1)(pn + 2),
(pn + 1)(pn + 2) ≤ t ≤ (pn + 2)(pn + 2).
See the clever usage of the theorem. Now,
(pn + 2)2 ≤ p2n+1
because for n > 1, pn+1 ≥ pn + 2. By the same argument as above, no two of q, r, s, t
are equal. Therefore, we are done.
CHAPTER 4. PRIMES 249

Theorem 4.4.20 (Andrica theorem). Assume the general Bertrand’s postulate holds
√ √
true. For any positive integer n, pn+1 − pn < 1.

Proof. Fix the primes pn , pn+1 and think about the intervals they belong to. Like we
did in the proof of Mitra’s conjecture, assume that km ≤ pn ≤ (k + 1)m. Obviously,
we are looking for suitable values of k and m which fulfill our purpose. Consider the
following proof. We have

km ≤ pn ≤ (k + 1)m,
(k + 1)m ≤ pn+1 ≤ (k + 1)(m + 1).

Now, set k = m and we have k 2 ≤ pn < pn+1 ≤ (k + 1)2 . Therefore,

√ √
k≤ pn < pn+1 < k + 1,
√ √
which means pn+1 − pn < 1. Is this proof correct or faulty? We will leave it to
the reader. Here is the original proof. Consider the intervals A = [k(k − 1), k 2 ], B =
[k 2 , k(k + 1)], C = [k(k + 1), (k + 1)2 ]. We can certainly pick k such that pn belongs to
at least one of A, B or C. So we have three cases here.

i. pn is in A. Then pn+1 is in B (or A possibly) . Whichever it is, we have the

conclusion.

ii. pn is in B. Again, pn+1 is either in B or C. Same argument.

iii. pn is in C. If pn+1 is in C too, we are done. Otherwise, pn+1 is not in any of A, B, C.

However, since pn , pn+1 are consecutive primes and (k + 1)2 < pn+1 < (k + 1)(k + 2),
√ √
we have pn+1 − pn < 1. If you are not sure why, take a look at this:

1 1
k2 + k + < pn < pn+1 < k 2 + 3k + 2 +
4 4
2 2
⇐⇒ (2k + 1) < 4pn < 4pn+1 < (2k + 3)
√ √
⇐⇒ 2k + 1 < 2 pn < 2 pn+1 < 2k + 3
√ p
⇐⇒ 2 pn+1 − 2 4pn < 2
√ √
⇐⇒ pn+1 − pn < 1.

There are still many open questions left regarding prime numbers. We will describe
some later. You don’t have to necessarily find an answer to them, just try them! You
may learn something new by yourself, or even find new theorems. In many cases,
mathematicians develop theories this way.
11
Because A might contain more than one primes.
250 4.5. MISCELLANEOUS

4.5 Miscellaneous
Theorem 4.5.1. For any positive integer n, there are n consecutive integers none of
which are prime. In other words, there are arbitrarily large gaps in the sequence of
primes.
Proof. Let’s just look at the numbers (n + 1)! + 2, . . . , (n + 1)! + n + 1. These are
(n + 1) − (2) + 1 = n consecutive integers and none of them are prime since (n + 1)! + i
is divisible by i for 1 < i < n + 2.
Theorem 4.5.2. For any positive integer n, there are n consecutive integers so that
none of them are prime powers (not necessarily the power of same prime).
Proof. We will use Chinese Remainder Theorem to proceed. But how do we understand
we need CRT here? A basic idea is to show that n consecutive integers have at least two
different prime factors. That way, we can guarantee none of them is a prime power. So
we need x to be divisible by p1 p2 , x+1 to be divisible by p3 p4 and likewise, x+(n−1) to
be divisible by p2n−1 p2n . In other words, we need a solution to the system of congruences

x ≡ 0 (mod p1 p2 ),
x ≡ −1 (mod p3 p4 ),
..
.
x ≡ −(n − 1) (mod p2n−1 p2n ).

If we write shortly, we need x ≡ −i (mod p2i−1 p2i ) for 0 ≤ i ≤ n − 1. By CRT, we do

have such an x as a solution to those congruences. So, none of n consecutive integers
x, x + 1, . . . , x + (n − 1) are prime powers and the claim is proved.
Theorem 4.5.3. Let a, n, and d be positive integers so that a, a + d, . . . , a + (n − 1)d
are all primes. Then any prime p less than n divides d.
Proof. If p < n and p does not divide d, then (d, p) = 1. Therefore, by theorem ( ),
d has a unique inverse modulo p, say e. So de ≡ 1 (mod p), where 0 < e < p < n. Let
−ae ≡ i (mod p). Then

−a ≡ −ade ≡ id (mod p).

Note that i < p < n. Thus p|a + id for some i < n. This now gives p|a + (p − i)d and
p|a + (i − p)d. It is clear that either 0 < p − i < n or 0 < i − p < n. In either case,
p divides two terms of the sequence. Since all terms of the sequence are primes, those
two terms which are divisible by p must equal p. But this is a contradiction since the
sequence is strictly increasing. Hence, p must divide d.
Remark. The sequence a, a+d, a+2d, . . . is called an arithmetic sequence or arithmetic
progression (and briefly, AP) with initial term a and common difference d. The nth term
of the sequence is a + (n − 1)d. The above theorem shows that if all terms of an AP
with n terms and common difference d are primes, then d is divisible by any prime less
than d.
CHAPTER 4. PRIMES 251

We are going to explain and prove some inequalities about primes. In 1907, Bonse
found and proved the following two theorems:
Theorem 4.5.4. For n ≥ 4, p1 · · · pn > p2n+1 .
Theorem 4.5.5. For n ≥ 5, p1 · · · pn > p3n+1 .
In 1960, Pósa proved a more general form of Bonse’s theorems:
Theorem 4.5.6 (Pósa’s Inequality on Primes). For any integer k, there is a con-
stant m so that
p1 · · · pn > pkn+1
for all n > m.
We need some lemmas to prove this theorem.
Lemma 4.5.7. For n ≥ 5, pn > 2n.
Proof. We proceed by induction. For n = 5, p5 = 11 > 2 × 5. Assume pn > 2n is true
for some n and now we prove it for n + 1. Since n > 5, pn is odd and hence pn + 1 is
even, and is not a prime. So

pn+1 ≥ pn + 2 > 2n + 2 = 2(n + 1).

Lemma 4.5.8. For n ≥ 1, p1 · · · pn > 2n−1 n!.

Proof. Check the truth for n = 1, 2, 3, and 4. Note that p1 p2 p3 p4 = 2 · 3 · 5 · 7 = 210.
From lemma ( ), pi > 2i for i ≥ 5. Thus
n n Qn
Y Y 2i
pi > 2i > Q4i=1
i=5 i=5 i=1 2i
n
2 · n!
=
24 · 4!
n!
= 2n−7 · .
3
Now, we have
n
Y
p1 · · · pn = p1 p2 p3 p4 · pi
i=5
n!
> 210 · 2n−7
3
= 35 · 2n−6 n!
> 32 · 2n−6 n!
= 2n−1 n!.
252 4.5. MISCELLANEOUS
r
n! n
Lemma 4.5.9. The sequence un = , for n = 1, 2, . . ., is strictly increasing.
2
r r
n n! n+1 (n + 1)!
Proof. We are done if we can prove < , which is equivalent to
2 2
 n+1  n
n! (n + 1)!
< .
2 2

Simplifying, we find

(n!)n n! < 2(n!)n (n + 1)n ,

or n! < 2(n + 1)n , which is evident!

Lemma 4.5.10. For all positive integers n, pn ≤ 2n . Equality occurs if and only if
n = 1, otherwise pn < 2n .

Proof. If n = 1, p1 = 2 = 21 . We know that p2 = 3 < 22 . Thus we induct on n using

the first alternative of Bertrand’s postulate stated in theorem ( ). Let’s assume
that pn < 2 . Since p is odd, we have pn+1 < 2pn < 2 .
n n+1

We are ready to prove Pósa’s theorem.

Proof of Pósa’s Theorem. The case k ≤ 0 is trivially true. So we focus on k > 0.
Note that using lemma ( ), we find pkn+1 < 2(n+1)k . So we need to show that

p1 · · · pn > 2(n+1)k .

On the other hand, using lemma ( ), we have p1 · · · pn > 2n−1 n!, we are done if we
can prove that there is a n0 so that

2n−1 n! > 2(n+1)k

holds for all n ≥ n0 . We can write this as

n! 2(n+1)k
> = 2n(k−1) · 2k ,
2 2n
and so,
r
n! k
> 2k−1 · 2 n .
n

Note thatr2
k−1
is a constant and 2k/n decreases as n increases. However, by lemma
n!
( ), n increases when n gets larger. This means that the expression on the left
2
hand side of above inequality is a strictly increasing sequence, however the right hand
side sequence is strictly decreasing. It is obvious there is a smallest n0 so that the left
hand side gets bigger than the right hand side for all n ≥ n0 . The proof is complete.
CHAPTER 4. PRIMES 253

Theorem 4.5.11. The probability of two random positive integers being coprime is
6
.
π2
Proof. Two positive integers are coprime if they do not share any prime divisor. So we
can do just the opposite. We will find out the probability of them not being coprime.
Fix a prime p. What is the probability that both a and b are divisible by p? Think on
this for a bit.
Let us focus on what a and b leave as remainders when divided by p. There can be p
remainders (0, 1, . . . , p − 1). Both for a and b, there are p possibilities. The probability
that a leaves remainder 0 when divided by p is p1 . Similarly, the probability that b leaves
remainder 0 when divided by p is p1 as well. Therefore , both a and b leave remainder
0 when divided by p is p1 · p1 . Thus, the probability of a and b not being divisible by p is
1 − p12 . Now, this is only for a fixed prime p. Since p can be any prime, the probability
should be multiplied for all primes. The probability is
    Y  Y 2  Y
1 1 1 pi − 1 1
1 − 2 · 1 − 2 ··· = 1− 2 = 2
= 2
p1 p2 i≥1
pi i≥1
pi p
i≥1 2
i
pi −1
1 1 6
=  = = 2.
Q p2i ζ(2) π
i≥1
p2i − 1

We used Euler’s ζ(2) theorem ( ).

4.6 Distribution of Prime Numbers

Distribution of prime numbers is the topic which encouraged number theorist to start
a new branch called Analytic Number Theory. We have in fact discussed a little bit
about distribution of prime numbers already when we proved Bertrand’s theorem. Let
us focus on it a bit more.
There are 4 primes less than 10, 25 primes less than 100, 168 less than 1000 and so
on. And finding a formula for the number of primes less than n has always fascinated
mathematicians. Well, Gauss did not exactly provide a formula for the number of
n
primes, but he noticed that the value of and the number primes less than n, π(n),
ln n
gets closer as n tends to infinity. This gave birth to the Prime Number Theorem or PNT,
conjectured by Gauss. It was unproven for about 100 years. Then Chebyshev provided
a partial proof using his functions (which were later known as, Chebyshev function of
type 1 and 2), and that was only the start of analytical number theory. There is a huge
underlying significance here. Gauss did not conjecture any exact formula for π(n). But
since he was unable to provide one, he estimated instead. Analytical number theory
does not provide exact formulas like elementary number theory, rather it shows some
12
We assume you know that, the probability of two independent events is the product of the proba-
bility of those events. That is if A and B are independent, then P (A ∩ B) = P (A)P (B). And certainly
a being divisible by p has nothing to do with b being divisible by p. So they are independent.
254 4.6. DISTRIBUTION OF PRIME NUMBERS

estimation, and mathematicians tend to prove the estimates or improve them. This is
because, most of the times providing exact formulas for the functions are either very
hard or not so pretty. For example, one can find an exact formula for finding the nth
prime number, but one will not like it.
We will start with some functions and analyzing their properties. The obvious
question is, why do mathematicians define such functions? In this case, why are these
functions and their properties important? The reason is simple. If you can not under-
stand primes directly, understand some functions that can characterize them or tell us
something about them, some function that we can analyze. At first, they can be intim-
idating. So, we will try to show examples in order to make sense why these functions
have something to do with primes.

4.6.1 Chebyshev Functions

Definition 4.6.1. Let x > 0 be a real number. We define Chebyshev’s ϑ-function as
X
ϑ(x) = ln p,
p≤x

where the sum extends over all primes p less than or equal to x.
Example. Take x = 142.61. The primes less than x are 2, 3, 5, . . . , 137, 139. So
X
ϑ(142.61) = ln p = ln 2 + ln 3 + · · · + ln 137 + ln 139
p≤142

Corollary 4.6.2. If p1 , p2 , . . . , pk are primes less than or equal to x, then

ϑ(x) = ln (p1 p2 · · · pk ) .

Lemma 4.6.3. Let k and n be positive integers such that k < n < 2k + 1. Then
 
n Y
≥ p,
k k<p≤n

where the product extends over all primes p between k and n.

Proof. Write nk as
 
n n(n − 1)(n − 2) · · · (k + 2)(k + 1)
(4.11) = .
k (n − k)!

Let p1 , p2 , . . . , pm be the primes between k + 1 and n (including). Since n ≤ 2k + 1 can

be represented as n − k < k + 1, we have

n − k < k + 1 ≤ pi ≤ n

for i = 1, 2, . . . , m. So n − k < pi , which means that pi is relatively prime to all positive

integers less than or equal to n − k. In other words, (pi , (n − k)!) = 1 for all i. The
CHAPTER 4. PRIMES 255

rest is easy: the numerator of (
) can be regarded as the product of p1 p2 · · · pm and

another integer, say, s. Since nk is an integer and also (pi , (n − k)!) = 1 for all i, we
conclude that s must be divisible by (n − k)!. Thus,
 
n p1 p2 · · · pm · s s Y
= = p1 p2 · · · pm · ≥ p1 p2 · · · pm = p,
k (n − k)! (n − k)! k<p≤n

as claimed.

Proposition 4.6.4. Let x > 0 be a real number. Then

(4.12) ϑ(x) ≤ 2x ln 2.

Proof. We induct on ⌊x⌋. For our base cases, we note that for 0 ≤ x < 2, we have
ϑ(x) = 0 ≤ 2x ln 2.
Now suppose that x ≥ 2. Let n = ⌊x⌋ and suppose that the inequality holds for all
reals y such that ⌊y⌋ < n. Note that
         
x n n nn n n n
2 ≥ 2 = (1 + 1) = + + ··· + + ··· + +
0 1 ⌊n/2⌋ n−1 n
 
n Y
≥ ≥ p,
⌊n/2⌋
⌊n/2⌋<p≤n

where we have used lemma ( ) to write the last line. Taking logarithms from the
above inequality, we find
X
x ln 2 ≥ ln p
⌊n/2⌋<p≤n

= ϑ(x) − ϑ(⌊n/2⌋)
≥ ϑ(x) − 2⌊n/2⌋ ln 2
≥ ϑ(x) − x ln 2,

by the inductive hypothesis. Therefore

2x ln 2 ≥ ϑ(x),

as desired.

Definition 4.6.5. Let x > 0 be a real number. We define Chebyshev’s ψ-function as

X
ψ(x) = ln p,
pa ≤x

where pa ranges over all the powers of primes p1 , p2 , . . . , pk which do not exceed x. In
other words, ln p appears in the sum each time a power of p is less than or equal to x.
256 4.6. DISTRIBUTION OF PRIME NUMBERS

Example. Let’s find ψ(10.5). The primes which do not exceed 10.5 are 2, 3, 5, and
7. The powers of these primes which do not exceed 10.5 are 2, 22 , 23 , 3, 32 , 5, and 7.
Therefore

ψ(10.5) = ln 2 + ln 2 + ln 2 + ln 3 + ln 3 + ln 5 + ln 7
= ln(23 × 32 × 5 × 7)
= ln(2520)
≈ 7.83.

Corollary 4.6.6. Let p1 , p2 , . . . , pk be primes not exceeding a positive real number x.

Also, assume that pa11 , pa22 , . . . , pakk are the largest powers of these primes which do not
exceed x. Then

ψ(x) = ln (pa11 pa22 · · · pakk ) .

Corollary 4.6.7. Let x be a positive real number. Then

ψ(x) = lcm([1, 2, . . . , ⌊x⌋]).

Proof. Let p1 , p2 , . . . , pk be primes which do not exceed x. Let pa11 , pa22 , . . . , pakk be the
largest powers of these primes which do not exceed x. Each number in the set A =
{1, 2, . . . , ⌊x⌋} is of the form pb11 pb22 · · · pbkk , where bi is an integer and 0 ≤ bi ≤ ai (for
i = 1, 2, . . . , k). It is easy to check (see proposition ( )) that the least common
a1 a2 ak
multiple of all such integers is p1 p2 · · · pk . The previous corollary proves the claim
now.
Proposition 4.6.8. For any real x > 0, we have
X j ln x k
ψ(x) = ln p.
p≤x
ln p

Proof. Let p be a prime not exceeding x. We just need to show that the power of p
appearing in ψ(x) equals ⌊ln x/ ln p⌋. This is rather obvious. Let a be the power of p
we are searching for. Then pa ≤ x < pa+1 . Taking logarithms and dividing by ln p, we
find the desired result.
Chebyshev attempted to prove the Prime Number Theorem, and he succeeded in
proving a slightly weaker version of the theorem. In fact, he proved that if the limit
π(x) ln(x)/x as x goes to infinity exists at all, then it is equal to one. He showed that
this ratio is bounded above and below by two explicitly given constants near 1, for
all sufficiently large x. Although Chebyshev was unable to prove PNT completely, his
estimates for π(x), ϑ(x), and ψ(x) were strong enough to prove Bertrand’s postulate
at his time. We will state these estimations but hesitate to provide the proofs as they
need some calculus background.
Theorem 4.6.9 (Chebyshev Estimates). If the following limits exist, they are all
equal to 1.
ϑ(x) ψ(x) π(x) ln(x)
(4.13) lim , lim , and lim .
x→∞ x x→∞ x x→∞ x
CHAPTER 4. PRIMES 257

The inequalities in the next theorem show that n/ ln(n) is the correct order of
magnitude for π(n). In fact, these inequalities are pretty weak and better inequalities
can be obtained with greater effort but the following theorem is of our interest because
of its elementary proof.
Theorem 4.6.10. For all integers m ≥ 2,
1 m m
(4.14) < π(m) < 4 .
6 ln m ln m
Proof. Let’s prove the leftmost inequality first. Assume that n ≥ 1 is an intger. One
can easily show by induction that
 
n 2n
(4.15) 2 ≤ < 4n .
n


Using the fact that 2n
n
= (2n)!/(n!)2 , we can take logarithms from ( ) to obtain

(4.16) n ln 2 ≤ ln(2n)! − 2 ln n! < n ln 4.

We must now find a way to compute ln(2n)! and ln n!. Let k be a positive integer. By
theorem ( ), we have
α  
X k
(4.17) vp (k!) = ,
i=1
pi

where α is some positive integer. Here, we need to find α. See proof of theorem ( )
to realize that α + 1 is actually the number of digits of k in base p. On the other hand,
we know that the number of digits of a positive integer x in base y is ⌊logy x⌋ + 1 (prove
this as an exercise). So, in our case, α + 1 = ⌊logp k⌋ + 1, or simply α = ⌊logp k⌋. Since
we are working with natural logarithms (i.e., logarithms in base e), it would be better
to write a = ⌊ ln k
ln p
⌋. Finally, substituting n and 2n for k in equation ( ), we get

⌊X
ln p ⌋ 
ln n
 ⌊Xln p ⌋ 
ln 2n

n 2n
vp (n!) = , vp ((2n)!) = .
i=1
pi i=1
pi
Q
It is clear that n! = pvp (n!) , where the product is extended over all primes p less than
p≤n
or equal to n. After taking logarithms in the latter equation, the product turns into a
sum:

X ⌊X
ln p ⌋ 
ln n

X n
ln n! = vp (n!) ln p = ln p.
p≤n p≤n i=1
pi

Similarly,

X ⌊X
ln p ⌋ 
ln 2n

X 2n
ln(2n)! = vp ((2n)!) ln p = ln p.
p≤n p≤2n i=1
pi
258 4.6. DISTRIBUTION OF PRIME NUMBERS

Hence, the left hand side of inequality ( ) becomes

X ⌊X
ln p ⌋  ln p ⌋ 
X ⌊X
ln 2n ln n
 
2n n
n ln 2 ≤ ln(2n)! − 2 ln n! = ln p − 2 ln p
p≤2n i=1
pi p≤n i=1
pi

ln p ⌋ 
X ⌊X X ⌊X ln p ⌋
ln 2n ln 2n
  
2n n
= ln p − 2 i ln p
p≤2n i=1
pi p≤2n i=1
p

X ⌊X
ln p ⌋
ln 2n
   !
2n n
(4.18) = i
−2 i ln p.
p≤2n i=1
p p

Since for all rationals x, ⌊2x⌋ − 2⌊x⌋ is either 0 or 1, we can write

X ⌊Xln p ⌋
ln 2n !
n ln 2 ≤ 1 ln p
p≤2n i=1
X
≤ ln 2n
p≤2n

= π(2n) ln 2n.

The proof is almost finished. Note that ln 2 ≈ 0.6931 > 1/2 and therefore

n ln 2 1 n 1 2n 1 2n
π(2n) ≥ > = > .
ln 2n 2 ln 2n 4 ln 2n 6 ln 2n
So the left side inequality of ( ) is proved for even positive integers m = 2n. We
now prove it for m = 2n + 1. Since 2n/(2n + 1) ≥ 2/3, we get

1 2n 1 2n 2n + 1 1 2n + 1
π(2n + 1) ≥ π(2n) > > ≥ ,
4 ln 2n 4 2n + 1 ln(2n + 1) 6 ln(2n + 1)

and this proves the left hand inequality of ( ) for all m ≥ 2.

We will now prove the other inequality of ( ). We shall make use of proposition
( ). Let α be an arbitrary real number such that 0 < α < 1. Then n > nα and so
π(n) ≥ π(nα ). Using equation ( ), one can write
  X X 
α α
π(n) − π(n ) ln n = 1− 1 ln nα
p≤n p≤nα
X
= ln nα
nα ≤p≤n
X
≤ p
nα ≤p≤n

≤ ϑ(n)
< 2n ln 2.
CHAPTER 4. PRIMES 259

This already means that

2n ln 2
π(n) < + π(nα )
α ln n
2n ln 2
< + nα
α ln n
n  2 ln 2 ln n 
= + 1−α .
ln n α n
We use a bit calculus to finish the proof. Let f (x) = xln1−α x
. You can easily calculate
the derivative f (x) of f and find that it equals zero for x = e1/1−α . Putting this value
′

into f , you will see that nln1−α

n
≤ 1/e(1 − α). Since α is an arbitrary number, choosing
α = 2/3 helps us finish the proof:
n  3 n
π(n) < 3 ln 2 + <4 .
ln n e ln n

Here is an interesting problem which combines several concepts. This problem

appeared in Theorem 137 of [ ], and we are going to provide an elegant solution by
Professor Peyman Nasehpour [ ].

Problem 4.6.11. Let r be a real number whose decimal representation is

r = 0.r1 r2 . . . rn . . . = 0.011010100010 . . . ,

where rn = 1 if when n is prime and rn = 0 otherwise. Show that r is irrational.

Solution. We will prove a more general statement: such a number r in any base b > 1
is irrational. First, we define the average of digits of r in base b, denoted by Avb (r), as
r1 + r2 + · · · + rn
Avb (r) = lim .
n→∞ n
It is clear that Avb (r) is well-defined if the above limit exists. It is a good exercise for
you to prove that if Avb (r) = 0, then r is irrational. The latter result is true because if
r is rational, Avb (r) exists and is positive. If you have doubts about it, see [ ]). Now,
with the definition of r and by Prime Number Theorem,

π(n) n/ log n 1
Avb (r) = lim = lim = lim = 0,
n→∞ n n→∞ n n→∞ log n

and so r is irrational!
1 1 1 P1
Theorem 4.6.12 (Euler). The sum S = + + + ... = diverges i.e. does
2 3 5 p∈P p
not have a finite summation.

The proof is due to Dustin J. Mixon, which appeared at American Mathematical

Monthly [ ].
260 4.7. THE SELBERG IDENTITY

Proof. Let pi be the ith prime number and the sum does not diverge. Then there must
be a k such that
∞
X 1
<1
i=k+1
p i

We let A be the set of positive integers which has all prime factors less than or equal
to pk , and B be the set of positive integers with all prime factors greater than or equal
to pk+1 . From the fundamental theorem of arithmetic, each positive integer can be
uniquely expressed as a product ab where a ∈ A and b ∈ B. We have
X1 ∞ ∞
X X 1
= ···
a∈A
a x1 =0
px1
xk =0 1
· · · pxk k
∞
! ∞
X 1 X 1
= ···
p x1
x =0 1
p xk
x =0 k
1 k

<∞

Moreover, assume that Bi is the set of positive integers with exactly i distinct prime
factors. This yields,
X1 ∞ X
X 1
=
b∈B
b i=0 b∈B
b
i

∞ ∞
!j
X X 1
≤ <∞
i=1 j=k+1
pj

Since every positive integer greater than 1 belongs to exactly one of A or B, we have
∞
1 1 X1
+ ··· + + ··· =
2 n n=2
n
XX 1
=
a∈A b∈B
ab
X1X1
=
a∈A
a b∈B b
< ∞.

The claim follows from this (how?).

4.7 The Selberg Identity

Alte Selberg and Paul Erdős together first proved the Prime Number Theorem in
an elementary way. Selberg found an interesting identity in the process of the proof,
CHAPTER 4. PRIMES 261

known as the Selberg Identity. As we stated before, there was a dispute regarding who
proved prime number theorem elementarily. After we read the paper The elementary
proof of the prime number theorem: An historical perspective [ ] by D. Goldfeld, we
have decided that we trust what Goldfeld said in this paper and conclude that both
Erdős and Selberg had contributions in this proof. To be more precise, we believe
that Selberg proved the fundamental identity but could not proved PNT at the time.
Later Erdős proved PNT with the help of fundamental identity, and probably Selberg
proved PNT on his own as well (possibly afterwards) . However, as mentioned in
paper [ ], one must recognize that Erdős could immediately catch the fact that the
fundamental identity implies lim pn+1 = 1. This alone demonstrates the magnitude of
n→∞ pn
Erdős’s thinking ability. While we are on this point, we would also like to clarify the
magnitude of proving PNT elementarily. G. H. Hardy said this about PNT in 1921 [ ]:

No elementary proof of the prime number theorem is known, and one may
ask whether it is reasonable to expect one. Now we know that the theorem
is roughly equivalent to a theorem about an analytic function, the theorem
that Riemann’s zeta function has no roots on a certain line. A proof of such
a theorem, not fundamentally dependent on the theory of functions, seems
to me extraordinarily unlikely. It is rash to assert that a mathematical
theorem cannot be proved in a particular way; but one thing seems quite
clear. We have certain views about the logic of the theory; we think that
some theorems, as we say ‘lie deep’ and others nearer to the surface. If
anyone produces an elementary proof of the prime number theorem, he will
show that these views are wrong, that the subject does not hang together
in the way we have supposed, and that it is time for the books to be cast
aside and for the theory to be rewritten.

We need some definitions before stating the Selberg identity. We will use functions
defined in chapter ( ). The following theorem is almost trivial.

Theorem 4.7.1 (Invariance Theorem). Let f be an arithmetic function and I be

the identity function. Then

f ∗ I = I ∗ f = f.

Theorem 4.7.2. Let f be an arithmetic function and F is its summation function.

Then

f = µ ∗ F.

Proof. This is immediately resulted from Mőbius inversion theorem (theorem ( )).

13
The identity we are discussing here can be thought of as a basis of what Selberg calls the funda-
mental identity.
14
We are not making any assertion here, only expressing our thoughts. One should not draw any
conclusion from ours. You can read the papers and think for yourself what you want to decide.
262 4.7. THE SELBERG IDENTITY

Definition 4.7.3 (Dirichlet Derivative). For an arithmetic function f , we define its

Dirichlet derivative as

f ′ (n) = f (n) ln n.

Example. I ′ (n) = I(n) ln n = 0 for all positive integers n. Also, u′ (n) = ln n and
u′′ (n) = ln n · ln n = ln2 n.
We can easily check that some usual properties of differentiation hold true for Dirich-
let derivative as well. For instance:

Proposition 4.7.4. Let f and g be arithmetic functions. Then

(f + g)′ = f ′ + g ′ ,
(f ∗ g)′ = f ′ ∗ g + f ∗ g ′ .

Proof. The first one is obvious. For the second one, we can write
 
X n
(f ∗ g)′ (n) =  f (d)g  · ln n
d
d|n
 
n  
 · ln d + ln n
X
= f (d)g
d d
d|n
   
n n
 · ln n
X X
= f (d)g  · ln d +  f (d)g
d d d
d|n d|n
X  n  X  n  n
= f (d) · ln d · g + f (d)g · ln
d d d
d|n d|n
X  n  X  n 
= f ′ (d) · g + f (d)g ′
d d
d|n d|n
′ ′
= (f ∗ g)(n) + (f ∗ g )(n).

Definition 4.7.5 (Von Mangoldt Function). For any positive integer n, the von
Mangoldt function, denoted by Λ(n) is defined as
(
ln p if n = pm for some prime p and positive integer m,
Λ(n) =
0 otherwise.
P
Theorem 4.7.6. Let n be a positive integer. Then ln n = d|n Λ(d).
15
You can see it follows some properties of the usual derivative (if you are familiar with calculus,
you should know what derivative is. However, for this purpose you do not need any calculus.)
16
Λ is the upper case of the symbol lambda (λ) in Greek.
CHAPTER 4. PRIMES 263
Qk
Proof. Let n = i=1 pei i , where pi are primes (1 ≤ i ≤ k). Then,
k
!
Y
ln n = ln pei i
i=1
k
X
= ln pei i
i=1
Xk
= ei ln pi .
i=1

On the other hand, if p is a prime, for d 6= pm we have P Λ(d) = 0 by definition. Therefore,

only prime powers p contribute a ln p to the sum d|n Λ(d). So, if pi is a prime divisor
e

of n, p1i , . . . , pei i contribute ei · ln pi to the sum. Thus,

X k
X
Λ(d) = ei ln pi .
d|n i=1

Now we are ready to state and prove the Selberg’s identity.

Theorem 4.7.7 (Selberg’s Identity). Let n be a positive integer. Then
X n X n
Λ(n) ln n + Λ(d)Λ = µ(d) ln2 .
d d
d|n d|n
P
Proof. We proved in theorem ( ) that ln n = d|n Λ(d). We also found that u′ (n) =
ln n. This can be written as
Λ ∗ u = u′ .
Take derivative of both sides of the above equation to obtain
Λ′ ∗ u + Λ ∗ u′ = u′′ .
Using Λ ∗ u = u′ again,
Λ′ ∗ u + Λ ∗ (Λ ∗ u) = u′′ .
Now multiply both side by u−1 = µ (as proved in theorem ( )) to get
Λ′ ∗ (u ∗ u−1 ) + Λ ∗ (Λ ∗ (u ∗ u−1 )) = u′′ ∗ u′ .
Now, since u ∗ u−1 = I and f ∗ I = f for any arithmetic function f , we have
Λ′ + Λ ∗ Λ = u′′ ∗ µ.
Replacing the functions with their definitions, one easily finds
X n X n
Λ(n) log n + Λ(d)Λ = µ(d) log2 ,
d d
d|n d|n

as desired.
264 4.8. PRIMALITY TESTING

4.8 Primality Testing

Depending on the guarantee whether our algorithm can say for sure if a number is a
prime or not, we can divide the algorithms for prime testing in two:
1. Deterministic Primality Test

2. Non-deterministic or Probabilistic or Randomized Primality Test

First we will discuss some deterministic approach, then some probabilistic approach.
The first one is based on a theorem we have already established in chapter ( ).
√
Theorem 4.8.1 (Trial Division√until n ). If n is a composite number, it has at
least one prime factor q with q ≤ n.
This is the simplest way to check whether a positive
√ integer n is a prime. That is,
given n, you check weather any prime 2 ≤ p ≤ n divides n. If n is not divisible by
any such p, it is a prime. We take this opportunity to introduce a notion of runtime,
which will roughly mean the number of operations someone or a computer will have to
do in order to determine whether n is prime or not using a particular algorithm.
√ In
this algorithm, you can see that we are dividing n by primes less than n and so, if
the number of such primes is k then we could say, runtime is R(k). Here, assume that
R(k) denotes the runtime of the whole operation, though it is not rigorous at all. But
it will do for our purpose very nicely. Let’s look at the following theorem now.
Theorem 4.8.2 (Lucas Test). Let n > 1 be a positive integer. Then n is a prime if
and only if there is an integer 1 < a < n for which

an−1 ≡ 1 (mod n),

and for every prime factor p of n − 1,

a(n−1)/p 6≡ 1 (mod n).

Proof. We will show the if part first. If n is a prime, then by theorem ( ), it has a
primitive root. That is, there exists some integer a such that a ϕ(n)
=a n−1
≡ 1 (mod n)
and ad 6≡ 1 (mod n) for all d < n.
On the other hand, assume that given conditions hold for a positive integer n. The
first condition asserts that (a, n) = 1. Let d be the order of a modulo n. That is,
d is the smallest positive integer less than n such that ad ≡ 1 (mod n). By theorem
( ), d|n − 1. This means that dx = n − 1 for some x. Choose a prime q which
divides x so that x = qy for some integer y. Therefore, n − 1 = dqy or (n − 1)/q = dy.
But then

y
a(n−1)/q ≡ ady ≡ ad ≡ 1 (mod n),

which is in contradiction with the second condition since q is a prime such that q|x|n−1.
Thus the order of a modulo n is n − 1. So ϕ(n) = n − 1 which implies that n is a
prime.
CHAPTER 4. PRIMES 265

The next theorem is taken from [ ].

Theorem 4.8.3 (Pocklington’s Theorem). Let n > 1 be an integer and suppose

that there exist an integer a and a prime q such that the following conditions hold:
√
1. q|n − 1 and q > n − 1,

2. an−1 ≡ 1 (mod n), and

 n−1

3. a q − 1, n = 1.

Then n is a prime.
√
Proof. Assume n is not prime. Then n has a prime divisor p such that p ≤ n. By first
condition, q > p−1 and so (q, p−1) = 1. We can deduce by theorem ( ) that there
exists an integer x such that qx ≡ 1 (mod p − 1). This means that qx − 1 = (p − 1)k or
qx = (p − 1)k + 1 for some k. Since p|n, the second condition gives an−1 ≡ 1 (mod p)
and so

1 ≡ an−1

x
≡ an−1

qx
≡ a(n−1)/q

(p−1)k+1
≡ a(n−1)/q

k p−1

≡ a(n−1)/q · a(n−1)/q
| {z }
≡1
(n−1)/q
≡a (mod p).

This gives p|a(n−1)/q − 1. Combining the latter result with p|n, we have
  n−1 
p a q − 1, n = 1,

a contradiction. Hence n is prime.

Note. Depending on the implementation of a result, a deterministic test can be con-

verted into a non-deterministic one. For example, the above theorem can be both
deterministic and probabilistic. Because you can iterate over all possible a modulo a.
Or you could use some random a that are coprime to n. For a randomized test, we
would check only for some random a⊥n because an−1 ≡ 1 (mod n) must hold. If the
result was in favor for all a, we would say, n is a probable prime. Otherwise, n is a
definite composite.

In 1977, Robert Martin Solovay and Volker Strassen developed a method called
Solovay–Strassen primality testing which is based on Euler’s criterion.
266 4.8. PRIMALITY TESTING

Definition 4.8.4. Let n > 1 be an odd integer. Assume that a > 1 is a positive integer
such that (a, n) = 1 and
a
(n−1)/2
a 6≡ (mod n),
n
a
where is the Jacobi symbol defined in ( ). Then a is called an Euler witness
n
for compositeness of n, or simply an Euler witness for n.

Theorem 4.8.5 (Solovay–Strassen Primality Test). Let n > 1 be an odd integer.

Then n is composite if it has an Euler witness.

Proof. By Euler’s criterion, we know that if n is a prime, then for every integer a
coprime to n,
a
(4.19) a(n−1)/2 ≡ (mod n).
n

So, if for some a coprime to n, the above congruence equation does not hold, n cannot
be a prime. Thus it is composite.

Remark. We discussed different classes of pseudoprimes, integers which share a com-

mon property with all primes but are composite, in section ( ). If equation ( )
holds true for a composite integer n and an integer a > 1 coprime to it, then n is called
an Euler pseudoprime to base a, abbreviated as epsp(a). The Solovay–Strassen test is
closely related to Euler pseudoprimes. In fact, if an odd integer n is composite and an
integer a such that 1 < a < n and (a, n) = 1 is not an Euler witness for n, then n is
an epsp(a). On the other hand, if an odd n is an epsp(a) for some a, then a is not an
Euler witness for n.

As explained in that section, a well-known class of pseudoprimes are strong pseudo-

primes. Gary Lee Miller developed Miller’s primality test which involves the congru-
ences used in the definition of strong pseudoprimes. Michael Oser Rabin later modified
Miller’s primality test and obtained Miller–Rabin test which we will now explain. To
formulate Miller-Rabin primality test, it would be convenient to use the terminology
introduced by Rabin as below:

Definition 4.8.6. Let n = 2s d + 1 where s and d are positive integers and d is odd.
Let a > 1 be an integer coprime to n. Then a is said to be a witness for compositeness
of n, or simply a witness for n when

ad 6≡ 1 (mod n), and

2r d
a 6≡ −1 (mod n), for every integer 0 ≤ r < s.

Theorem 4.8.7 (Miller–Rabin Primality Test). Let n > 1 be an odd integer.

Then n is composite if it has a witness.
CHAPTER 4. PRIMES 267

Proof. Assume that n has a witness a. Then by definition (a, n) = 1 and

ad 6≡ 1 (mod n),
d
a 6≡ −1 (mod n),
2d
a 6≡ −1 (mod n),
a4d 6≡ −1 (mod n),
..
.
s−1 d
a2 6≡ −1 (mod n).

It follows that the following product is not divisible by n:

 s−1  s
ad − 1 ad + 1 a2d + 1 · · · a2 d + 1 = a2 d − 1.
s
But a2 d − 1 = an−1 − 1 and so n ∤ an−1 − 1. We know by Fermat’s little theorem that
if p is a prime, then p|ap−1 − 1 for any a such that (a, p) = 1. So, n cannot be a prime
and is therefore a composite number.
Note. If an odd integer n is composite and an integer a such that 1 < a < n and
(a, n) = 1 is not a witness for n, then n is a spsp(a). On the other hand, if an odd n is
a spsp(a) for some a, then a is not a witness.
The most well known deterministic algorithm known for primality testing is AKS
primality test. It was introduced by Manindra Agrawal, Neeraj Kayal, and Nitin Saxena
in 2002. The core idea of AKS primality test is the following theorem.
Theorem 4.8.8. Let a be an integer and n be a positive integer such that (a, n) = 1.
Then n is prime if and only if

(x + a)n ≡ xn + a (mod n),

for all integers x.

Proof. Let P (x) = (x + a)n − (xn + a). Then

P (x) = (x + a)n − (xn + a)

n  
X n i n−i
= x a − (xn + a)
i=0
i
n−1  
X n i n−i
= x a − (a − an ) .
i=1
i


If n is prime, then n divides ni for all 0 < i < n by theorem ( ) and also an ≡ a
(mod n) by Fermat’s little theorem. So P (x) ≡ 0 (mod n) and the condition holds.
If n is composite, take a prime divisor p of n. Let v be the greatest

power of p that
n
divides n. That is, p |n but p
∤ n. Then q does not divide i (why?) and therefore
v v+1 i

n does not divide the term np xq an−p in P (x). This means that P (x) 6≡ 0 (mod n) and
the proof is complete.
268 4.8. PRIMALITY TESTING

As you can see, the runtime of deterministic primality tests are not that great. Even
the best of them, AKS test has a runtime around (log2 n)12 , which was later reduced
to (log2 n)6 by mathematicians such as C. Pomerance. But it still is not very good for
running as a program. This runtime means, if n = 2100 , we would have to do around
1006 = 1012 operations, which is really costly. If we assume the best case scenario, an
average computer may perform 109 operations per second (in fact it is far less effectively
when it’s down to computing because there are many related calculations as well), so
it would require around 1000 seconds to test primality of a number of that magnitude.
But in practice, numbers around 1024 bits are used which are as large as 21023 − 1.
This makes this test obsolete. In turn, this gives rise to probabilistic primality test. In
a probabilistic test, one can not guarantee that the input n is definitely a prime. But
it can say if it is a probable prime or not. And if we use a good enough algorithm
the probability of having a false prime is really small, of the magnitude 2−k where k
is some iteration number or something else depending on the algorithm. But if k is
around 100, you can see how small this gets. This means the chances of getting a false
result is really really slim. Let’s first use Fermat’s little theorem as a probabilistic test.
We already know that for a prime and a positive integer x, we must have xp−1 ≡ 1
(mod p). Using this, we can make the test for input n this way.

i. Choose a random number x.

ii. Compute r as xn−1 ≡ r (mod n) (this needs to be done efficiently since n is large).

iii. If r 6= 1 then n is surely composite.

iv. Otherwise n is probably a prime. Probably, because the reverse of Fermat’s little
theorem is not true, as we discussed on chapter ( ) before.

But this doesn’t make a very reliable test. To make it a bit more reliable, we can iterate
this process for k times. And each time we have to choose another random x. The more
we iterate, the more the accuracy is.
The most popular and used method for probabilistic testing is Rabin-Miller primal-
ity test. This makes clever use of Fermat’s little theorem.

4.8.1 Primality Testing for Famous Classes of Primes

We have explained theorems which help us find out whether a number is prime. For
numbers having special forms, we can develop much better methods to test their pri-
mality. The first special type of numbers where 2k + 1 for an integer k ≥ 0.
n
Definition 4.8.9. Let n ≥ 0 be an integer. The numbers of the form Fn = 22 + 1 are
called Fermat numbers. If Fn is prime for some n, it is called a Fermat prime.

Proposition 4.8.10. If 2k + 1 is prime for an integer k ≥ 0, then it is a Fermat prime.

Proof. This is a special case of theorem ( ).

CHAPTER 4. PRIMES 269

Fermat conjectured that all Fermat numbers are actually primes. He computed Fn
for n = 0, 1, 2, 3, and 4 and found out they are all primes. However, he was unable to
show that F5 is prime. Euler later showed that for n ≥ 2, every factor of Fn should be
of the form m · 2n+2 + 1 and thus found 641 to be a divisor of F5 and factorized it as
F5 = 641 · 6700417.
Since Fn increases too rapidly with n, it is too difficult to check its primality. In
1877, Pepin developed a test for checking the primality of Fermat numbers:
Theorem 4.8.11 (Pepin’s Primality Test for Fermat Numbers). Let n ≥ 2 be
an integer and assume Fn denotes the nth Fermat number. Also, let k ≥ 2 be any
integer. Then the following conditions are equivalent:
 
k
1. Fn is prime and = −1.
Fn
2. k (Fn −1)/2 ≡ −1 (mod Fn ).
Proof. Assume that condition 1 holds. Then by Euler’s criterion (theorem ( )),
 
(Fn −1)/2 k
k ≡ ≡ −1 (mod Fn ).
Fn
To prove the other side of the theorem, assume that k (Fn −1)/2 ≡ −1 (mod Fn ). Choose
an integer m such that 1 ≤ m < Fn and m ≡ k (mod Fn ). Then
m(Fn −1)/2 ≡ k (Fn −1)/2 ≡ −1 (mod Fn ).
2n
Note that the only prime divisor of Fn − 1 = 2 is 2. Hence we can use Lucas test
(with a = m and p = 2, using notation of theorem ( )) and deduce that Fn is a
prime. Furthermore, we have by Euler’s criterion that
 
k
≡ k (Fn −1)/2 ≡ −1 (mod Fn ).
Fn
The proof is complete.
Pepin’s test is usually done with k = 3, 5, or 10. In practice, mathematicians have
not been able to show that any Fermat number Fn for n > 4 is a prime using Pepin’s
test. On the other hand, nobody has yet proved that all Fermat numbers larger than
F4 are composite.
Another type of numbers are Mersenne numbers, named after Marin Mersenne who
studies them back in 17th century.
Definition 4.8.12. Let n be an integer. The numbers of the form Mn = 2n − 1 are
called Mersenne numbers. If Mn is prime for some n, it is called a Mersenne prime.
Proposition 4.8.13. If Mn is a prime for an integer n > 1, then n is a prime.
Proof. See theorem ( ).
Mersenne stated that
M2 , M3 , M5 , M7 , M13 , M17 , M19 , M31 , M67 , M127 , M257
are the only Mersenne primes less than M258 . Although he was wrong about M67 and
17
M67 = 193707721 × 761838257287.
270 4.8. PRIMALITY TESTING

M257 and he missed M61 , M89 , and M107 in the list, his work is considered astonishing
because these numbers are astronomically large. Interested readers may study [ ] and
[ ] for more details about Mersenne primes.

Theorem 4.8.14. Let q > 2 be a prime. For every divisor n of Mq , we have

n ≡ ±1 (mod 8), and

n ≡ 1 (mod q).

Proof. It suffices to prove the theorem for prime n (why?). Let p be a prime divisor of
Mq = 2q − 1. Then 2q ≡ 1 (mod p) and so ordp (2)|q, which means that ordp (2) = q
since q is a prime. By corollary ( ), q = ordp (2)|ϕ(p) = p−1. Thus p ≡ 1 (mod q).
Since p and q are both odd, we can write the latter relation as p − 1 = 2kq. By Euler’s
criterion,
 
2
≡ 2(p−1)/2
p
≡ 2kq
≡ (2q )k
≡ 1 (mod p).

Theorem ( ) now verifies that p ≡ ±1 (mod 8).

The above theorem can be used to factorize Mersenne numbers. It will be tough to
use this theorem for large Mersenne number though. To realize whether a Mersenne
number is prime or composite, one may use the so-called Lucas–Lehmer primality test,
introduced by Édouard Lucas in 1856 and improved by Derrick Henry Lehmer later in
1930s. The proof is a bit difficult and we refuse to write it. The reader may see [ ] for
a proof if interested.

Theorem 4.8.15 (Lucas–Lehmer Primality Test for Mersenne Numbers). De-

fine the recursive sequence S(n) by S(1) = 4 and S(n + 1) = S(n)2 − 2 for any integer
n ≥ 1. Also, let p > 2 be a prime. Then Mp is prime if and only if it divides S(p − 1).

Example. We will apply Lucas–Lehmer test to factorize M11 = 211 − 1 = 2047. We

must check whether S(10) is divisible by 2047. Table shows values of S(n) for
n = 1, 2, . . . , 10. As seen in the table, S(10) is not zero modulo 2047 which means that
M11 is not a prime. In fact, 2047 = 23 · 89.
As the last class of primes, we will mention Proth numbers.

Definition 4.8.16. Let k and h be positive integers such that k is odd and k < 2h . A
number of the form n = k · 2h + 1 is called Proth number and if it is a prime, it is said
to be a Proth prime.

The following primality test for Proth numbers was published by François Proth
around 1878 and is known as Proth’s theorem.
18
Lehmer and Kraïtchik showed that M257 is composite.
CHAPTER 4. PRIMES 271

n S(n) (mod 2047)

1 4
2 14
3 194
4 788
5 701
6 119
7 1877
8 240
9 282
10 1736

Table 4.1: Applying Lucas–Lehmer test to test the primality of 2047.

Theorem 4.8.17 (Proth’s Primality Test for Proth Numbers). Let n be a Proth
number. Then n is prime if an integer a for which

a(n−1)/2 ≡ −1 (mod n).

We will prove a stronger result in the following lemma which was proposed by
Pocklington.
Lemma 4.8.18. Let a, b and n be positive integers such that 0 < a ≤ b + 1 and
n = ab + 1. Assume that for every prime divisor p of b there exists an integer x for
which

xn−1 ≡ 1 (mod n), and

x(n−1)/p 6≡ 1 (mod n).

Then n is prime.
Proof. Assume on the contrary that n is composite
√ and take the smallest prime factor
q of n. Theorem ( ) implies that q ≤ n. Let p be a prime factor of b. Write
b = p s, where k ≥ 1 and s are positive integer such that (s, p) = 1. Let x be an integer
k

which satisfies the given conditions. Then

(4.20) xn−1 ≡ 1 (mod q), and

(n−1)/p
(4.21) x 6≡ 1 (mod q).

because q|n. We claim that ordq (xa ) = b. To prove the claim, we notice that

(xa )b = xab
= xn−1
≡ 1 (mod q).

We must now show that (xa )m 6≡ 1 (mod q) for any integer 0 < m < b. Assume on
the contrary that there exists a positive integer m < b such that xma ≡ 1 (mod q). If
272 4.9. PRIME FACTORIZATION

d = ordq (x), then d|ma. On the other hand, the congruence relation ( ) shows that
d|n − 1 = pk sa. This implies d|(ma, pk sa). Suppose that m = pl t. Then



ma, pk sa = pl ta, pk sa


= a pl t, pk s
= apmin(l,k) (t, s).
(We have used propositions ( ) and ( ) in writing second and third lines.)
Now, the congruence equation ( ) implies that d ∤ n−1
p
= pk−1 as, which is in con-
tradiction with d|apmin(l,k) (t, s). We have thus shown that ordq (xa ) = b. It follows that
b ≤ ϕ(q) = q − 1 and hence,
q 2 ≥ (b + 1)2
≥ a(b + 1)
= ab + a
≥ n.
√
Since we first chose q so that q ≤ n, all inequalities above must be equalities. In
particular, n = q 2 , a = 1, and a = b + 1, which is a contradiction.
Remark. The Proth’s theorem is now a special case of above lemma where a = k and
b = 2n . The converse of Poth’s theorem is also true if x is a quadratic non-residue
modulo n.

4.9 Prime Factorization

Finding prime numbers has been a challenge for mathematicians since very long time
ago. Consider the following question:
Question 4.9.1. Given a real number X, find all primes less than X.
The very first answer to this question dates back to 200 B.C., when Eratosthenes
developed the Sieve method. This method is very simple but it is still used, after 2000
years of its birth! To apply sieve method of finding primes to an integer n, we write
down all the positive integers less than or equal to n. Put aside 1. The first number in
the list is 2, which we know is a prime. We start by erasing the multiples of 2 from the
list. Let’s simulate the process for X = 40:
2 3 4✁ 5 6✁ 7 8✁ ✚
9 ✚
10
✚ 13 ✚
12
11 ✚ ✚ 15 ✚
14 ✚ 17 ✚
16 ✚ 19 ✚
18 ✚
20
✚ 23 ✚
21 ✚
22 ✚ 25 ✚
24 ✚ 27 ✚
26 ✚ 29 ✚
28 ✚
30
✚
31 ✚ ✚ ✚
34 35 ✚
32 33 ✚ ✚
36 37 ✚ ✚
40
38 39 ✚
Now, choose the first number after 2 in the list. It is 3, a prime. Erase all multiples
of 3 from the list and choose the next number. The point is that the next number we
choose is always a prime because it is not divisible by any integer less than it in the
list (otherwise it would have been erased). We continue this method until we find the
largest prime less than or equal to n. The final list for n = 40 would look like this:
CHAPTER 4. PRIMES 273

2 3 4✁ 5 6✁ 7 8✁ 9✁ ✚
10
✚
11 ✚
12
✚ 13 ✚
14
✚ ✚
15
✚ ✚ 17 ✚
16
✚ ✚
18 19 ✚
20
✚
✚
21
✚ ✚
22
✚ 23 ✚
24
✚ ✚
25
✚ ✚
26 ✚
✚ ✚ ✚
27 ✚
28 29 ✚
30
✚
31 ✚
32
✚ ✚
33
✚ ✚
34
✚ ✚
35
✚ ✚
36
✚ 37 ✚
38
✚ ✚
39
✚ ✚
40
✚

The remaining numbers are primes less than or equal to n. In fact, we have sieved all
the primes in the list, hence the name sieve method. The sieve method is not time
efficient specially when n is large (can you find the reason?).

An algorithm for Sieve method of prime factorization

√
1. Generate the sieve up to m = n.

2. Assume that, the primes generated from the sieve are p1 , · · · , pk .

n′
3. For a prime pi where 1 ≤ i ≤ k, as long as pi |n, set n′ ←− . The number of times
pi
you could divide n by pi is the exponent of pi in n.

4. When i = k, stop and check if n > 1.

5. If n′ > 1 then this n′ itself is a prime factor of the original n (which we used for
factoring at the first step, and it will be decreasing since we keep dividing by a prime
if n is composite). And in this case, the exponent will be 1 (why?).

The last statement needs a bit clarification. In the steps before that, we divided n by
all prime factors of n less than or equal to m. Therefore, if all prime factors of n are less
than or equal to m, the n we will have after all these divisions is 1, since it is all divided
up by p ≤ m. But if n′ > 1 then we have this n′ > m, so n can not have two such n′ .
Because the product of two integer greater than m is greater than m2 , so greater than
m2 = n. That would be impossible. Therefore, for the case n′ > 1, n′ must be a prime
and it would be the largest prime factor of n. And if n′ = 1 then the largest i ≤ k for
which pi |n, pi would be the largest prime factor of n.

4.9.1 Fermat’s Method of Factorization

Fermat found a method for factorizing odd numbers. The idea behind his method is
very simple. Suppose that we want to factorize an integer n > 1. If we find positive
integers a and b such that

n = a2 − b 2 ,

and a − b > 1, then n = (a − b)(a + b) is a proper factorization of n. Remember that a

proper factorization is one in which neither of the factors are trivial (1 or n).
We already know a factorization for even integers n = 2m because 2 is a factor
of n in that case. Given an odd positive integer n, we try some value of a, hoping
that a2 − n is a perfect square. If this condition holds, we have found a factorization
274 4.9. PRIME FACTORIZATION

Step 1 2 3 4 5 6 7 8
a 60 61 62 63 64 65 66 67
x 11 132 255 380 507 636 767 900
√
x 3.31 11.48 15.96 19.49 22.51 25.21 27.69 30

Table 4.2: Applying Fermat’s method of factorization to 3589.

for n. Otherwise, increment a and check again. The point here is that if an odd n is
composite, i.e. if n = cd for some odd positive integers c and d, then
 2  2
c+d c−d
n= − .
2 2
This means that Fermat’s method of factorization always works when n is composite.
Fermat’s factorization is generally more time efficient than trial division. However, it
might be even slower than trial division in some cases.

An algorithm for Fermat’s method of factorization

1. Choose a = ⌈n⌉, and put x = a2 − n.
2. While x is not a perfect square, set a ←− a + 1 and compute x = a2 − n for the new
a.
√ √
3. If x is a perfect square, n = (a − x)(a + x) is a factorization for x.
Example. We will use Fermat’s method of factorization to factorize√n = 3589. Table
shows the steps of the algorithm. We have started from a = ⌈ 3589⌉ = 60 and
increase a by 1 at each step. When a = 67, we find x = 900, which is a perfect square.
The algorithm stops here and we have
√ √
n = (a − x)(a + x)
= (67 − 30)(67 + 30)
= 37 · 67,
which is a non-trivial factorization.

4.9.2 Pollard’s Rho Factorization

As already mentioned, prime factorization by sieve method is not time efficient. In fact,
most deterministic factorization methods are not. Therefore, we again use probabilistic
method. There are two crucial steps for probabilistic factorization methods.
• Finding a non-trivial factor of n (that is, a factor other than 1 and n).
• Using a time efficient primality test in order to check if the non-trivial factor d
n
is prime or not. If d is prime, we can just factorize only. Otherwise, we can
d
n
repeat the same process for d and . Mostly Rabin-Miller test is used widely
d
these days.
CHAPTER 4. PRIMES 275

Randomized tests vary mainly on the first step. Finding the non-trivial factor is the
crucial step here. Here we discuss Pollard’s method to find such a factor.
Let n > 1 be the composite integer which we want to factorize. Consider the
following sequence:

x0 = c, and
xi+1 ≡ g(xi ) (mod n) for i = 0, 1, 2, . . .

Here, g(x) is a polynomial with integer coefficients. Notice that this sequence will
eventually become periodic. That is, there exists a positive integer T such that xi ≡
xi+T (mod n) for all i ≥ i0 ≥ 0, where i0 is some integer. The reason is that there are
exactly n residues modulo n and the sequence is infinite, so by pigeonhole principle,
there are two terms xi and xj (with j > i) of the sequence for which xi ≡ xj (mod n).
Suppose that g(x) = an xn + an−1 xn−1 + · · · + a1 x + a0 , where a0 , a1 , . . . , an are integers.
Then

xi+1 ≡ g(xi )
= an xni + an−1 xn−1
i + · · · + a1 xi + a0
n n−1
≡ an xj + an−1 xj + · · · + a1 xj + a0
= g(xj )
≡ xj+1 (mod n).

therefore xi+k ≡ xj+k (mod n) for any k. Thus T = j − i is the period of the sequence.
It is clear now that T ≤ n.
The only thing that needs clarification is how to choose g(x) and x0 . It has been
practically shown that taking g(x) = x2 + a (for some integer a) is a good choice
for finding a non-trivial factor quickly. In 1975, John Pollard developed Pollard’s rho
method of factorization which takes g(x) = x2 − 1 and x0 = 2.
Here is how Pollard’s rho method of factorization work: assume that n = st, where
s and t are unknown factors of n such that t > s > 1. Suppose that we have found
integers j > i ≥ 0 such that xi ≡ xj (mod s) but xi 6≡ xj (mod n). Since s divides
both n and xi − xj , it must also divide (xi − xj , n). So (xi − xj , n) ≥ s > 1. On the
other hand, (xi − xj , n) is a factor of n and since it is larger than 1, it is a proper factor
of n (it is not equal to 1 or n). This means that we have found (xi − xj , n) to be a
factor of n.
So the problem now reduces to find indices j > i ≥ 0 such that xi ≡ xj (mod s) but
xi 6≡ xj (mod n). Pollard suggested that we take i = k and j = 2k for k = 1, 2, . . . , n.
You will see why in the following lines.
When we first discussed the periodicity of the sequence, we showed that the sequence
is periodic modulo n. However, one can show using Chinese Remainder Theorem that
the sequence is also periodic modulo s (why?). Assume that the sequence will be
periodic modulo s after xi0 with period T . Select an index k ≥ i0 such that T |k.
Then obviously T |2k and because of the periodicity, xk ≡ x2k (mod s). But now how
do we know that xk 6≡ x2k (mod n). We don’t know that for sure. There is just a
19
Mathematicians later found out that g(x) = x2 + 1 work better for almost all the cases.
276 4.9. PRIME FACTORIZATION

likelihood that it will happen. The reason for this is that the sequence {xi (mod s)}∞ i=0
is periodic modulo T , and as proved above, we have T ≤ s. Similarly, the sequence {xi
(mod n)}∞ i=0 is periodic with a period T ≤ n. Now, since s is a divisor of n, we have
′

s ≤ n so that the maximum value of period of the first sequence is smaller than that
of the second sequence. Because of this, it is likely that T < T ′ . If this latter condition
holds and we have xk ≡ x2k (mod s), then we can deduce that xk 6≡ x2k (mod n),
which is what we were searching for.
You might ask now what happens if the given condition, T < T ′ , does not hold?
Well, in that case, you cannot factorize n using Pollard’s method. In such cases, it is
usual to change the polynomial g(x) or the initial value x0 and then apply the method.
To summarize, in Pollard’s rho factorization method, starting with i = 1, we check
if gcd(x2i − xi , n) is a factor of n. If it is, we have found a factor for n. If not, increment
i and repeat the process. It is possible that we do not find any factor for n (even if n
is composite) and the process does not terminate in such cases, as explained above.

An algorithm for Pollard’s Rho method of prime factorization

1. Set x0 = 2 and form the sequence xi+1 ≡ x2i − 1 (mod n) for i = 0, 1, 2, . . . , n.

2. Compute dk = (xi − x2i , n) for i = 1, 2, . . . , n. If di 6= 1 and n, stop. Now di is a

factor of n.

3. If di is either 1 or n for all k, the algorithm does not work.

Example. Let us factorize n = 391 using Pollard’s Rho algorithm. The process is shown
in table . In 10th step, where (|x2i − xi |, 391) is 23, we find that 23 is a factor of
n (and indeed it is: 391 = 23 × 17). We just stopped the algorithm after that step
because we have factorized 391. However, we have written the value of xi for 11th step
so that you can observe the periodicity of xi modulo 391. As illustrated in the table,
x7 ≡ x11 ≡ 46 (mod 391). Now, if you look at the computed values of xk modulo 23,
you will see that for j ≥ 7, xj ≡ xj+2 (mod 23). In terms of our previous definitions,
T = 2, i0 = 7, and k = 10. Observe that we cannot choose k = 8 because then x2k − xk
is zero, which is divisible by both 23 and 391. Therefore we choose k = 10 so that
xk − x2k ≡ 0 (mod 23) but xk − x2k ≡ 69 6≡ 0 (mod 391).
The next example, taken from Patrick Stein’s website (see [ ]), takes a different
polynomial g(x) and initial value x0 in Pollard’s rho method.
Example. We will factorize a much larger integer n = 16843009. This time, we take
g(x) = 1024x2 + 32767 and x0 = 1. Table shows the steps. As you see in the table,
at 9 step we find 257 to be a factor of n and the factorization is done:
th

16843009 = 257 · 65537.

4 n
65537 is a prime number and it equals 22 + 1. Primes of the form 22 + 1 are called
Fermat primes, and the largest known such prime is 65537.
CHAPTER 4. PRIMES 277

i xi |x2i − xi | (mod 391) (|x2i − xi |, 391)

1 3 5 1
2 8 50 1
3 63 30 1
4 58 102 1
5 235 6 1
6 93 67 1
7 46 160 1
8 160 0 0
9 184 45 1
10 229 69 23
11 46 Whatever Whatever

Table 4.3: Applying Pollard’s rho method to factorize 391.

i xi |x2i − xi | (mod 16843009) (|x2i − xi |, 16843009)

1 33791 10798549 1
2 10832340 6592485 1
3 12473782 508279 1
4 4239855 893857 1
5 309274 5203404 1
6 11965503 7424857 1
7 15903688 1657047 1
8 3345998 15737239 1
9 2476108 15298182 257

Table 4.4: Applying Pollard’s rho method to factorize 16843009.

278 4.10. EXERCISES

4.10 Exercises
Problem 4.10.1. Let n ≥ 1 be an integer. Show that ψ(2n) > n ln 2.
Hint. Use proposition ( ) and the fact that
 
2n Y
≤ p⌊ln 2n/ ln p⌋ .
n p≤2n

Problem 4.10.2. Find a formula for the number of square-free numbers less than x
for a real number x. Recall that, a natural number n is square-free if n does not have
any factor that is perfect square other than 1. Can you represent this formula using
Mőbius function as well?
Problem 4.10.3. Show that 8081, 31627, and 65537 are all primes.
Hint. Take a = 2 or 3 and use Pocklington’s theorem.
Problem 4.10.4.
1. Let m, n, and k be non-negative integers such that m > 1. Prove that at least
one of the numbers
     
n n+1 n+k
, ,...,
k k k
is not divisible by m.

2. Let k and m be positive integers such that m > 1. Show that there are infinitely
many positive integers n such that
     
n n+1 n+k−1
, ,...,
k k k
are all divisible by m.
Hint. Use Kummer’s theorem to construct the solution.
Problem 4.10.5. Let k and n be positive integers

such that
0 < k ≤ n and let p be a
 n
prime such that p ∤ n + 1. Prove that if p 6 k , then p 6  n+1
k
.
Problem 4.10.6. Let m and n be positive integers. Suppose that the binary represen-
tation of m and n is

m = 2k mk + 2k−1 mk−1 + · · · + 2m1 + m0 , and

n = 2k nk + 2k−1 nk−1 + · · · + 2n1 + n0 .
m


Show that if n
is odd, then
  Y k
m
≡ (−1)ni−1 mi +ni mi−1 (mod 4).
n i=1
CHAPTER 4. PRIMES 279

4.11 Open Questions In Primes

Conjecture 4.2 (Twin Prime Conjecture). There exists infinitely many primes p
so that p + 2 is a prime too.

Conjecture 4.3 (Goldbach’s Conjecture). For all even number n greater than 4, n
is a sum of two primes.

Conjecture 4.4 (Legendre’s conjecture). There exists a prime between n2 and

(n + 1)2 .

Adway Mitra conjectured an improvement over this, which is known as the improved
version of Legendre’s conjecture.

Conjecture 4.5. There always exists at least two primes in the interval [n2 , (n + 1)2 ].

Another variation was proposed by Oppermann.

Conjecture 4.6 (Oppermann’s Conjecture). For all integer x > 1, there exists at
least one prime between x(x − 1) and x2 and another prime between x2 and x(x + 1).

An improved version was conjectured by Brocard.

Conjecture 4.7 (Brocard’s conjecture). There exists at least 4 primes between p2n
and p2n+1 where pn is the nth prime number.

Andrica’s inequality is worthy of mentioning while we are on the subject.

Conjecture 4.8 (Andrica’s Inequality). For all n ≥ 1,

√ √
pn+1 − pn < 1.

Conjecture 4.9 (n2 + 1 Problem). Does there exist infinitely primes of the form
n2 + 1?

Conjecture 4.10 (Polignac Conjecture). For every even integer 2n are there in-
finitely many pairs of consecutive primes which differ by 2n.

Conjecture 4.11 (Sophie Germain Primes). A prime is called a Sophie Germain

prime if 2p + 1 is a prime too. Does there exist infinitely many Sophie Germain primes?

Conjecture 4.12 (Mersenne Prime Problem). Is the number of Mersenne primes

infinite?

Conjecture 4.13 (Rassias Conjecture). For a prime p > 2, there exists two primes
p1 , p2 such that,
p1 + p2 + 1
p= .
p1
Bibliography

[1] D. Goldfeld, The Elementary Proof of the Prime Number Theorem: An Historical
Perspective, Number Theory (New York Seminar) (Springer, 2004), pp. 179˘192.

[2] H. Bohr, Address of Professor Harold Bohr, Proc. Internat. Congr. Math. (Cam-
bridge, 1950) vol 1, Amer. Math. Soc., Providence, R.I., 1952, 127˘134.

[3] Srinavasa Ramanujan, (1919). A proof of Bertrand’s postulate. Journal of the In-
dian Mathematical Society 11 : 181˘182.

[4] Paul Erdős, Beweis eines Satzes von Tschebyschef, Acta Sci. Math. (Szeged) 5
(1930˘1932), 194˘198.

[5] Nils A. Baas and Christian F. Skau., The Lord of Numbers: Alte Selberg On His
Life and Mathematics, Bulletin of The American Mathematical Society, Volume
45, Number 4, October 2008, Pages 617˘649.

[6] A. E. Ingham, On the difference between consecutive primes, Quart. J. Math. Ox-
ford Ser. vol. 8(1937) pp. 255 − 266.

[7] W. H. Mills, A prime-representing function, Bull. Amer. Math. Soc. vol. 53(1947)
p. 604.

[8] I. Niven, Functions Which Represent Prime Numbers, Amer. Math. Soc., November
1950.

[9] Lawrence E. Greenfield, Stephen J. Greenfield, Some Problems of Combinatorial

Number Theory Related to Bertrand’s Postulate, Journal of Integer Sequences, Vol.
1 (1998), Article 98.1.2.

[10] Apostol T. M., Introduction to Analytic Number Theory, 2nd Ed, Springer, 1976.

[11] Koblitz N., A Course in Number Theory and Cryptography, 2nd Ed, Springer,
1994.

[12] J. C. Lagarias, V. S. Miller, and A. M. Odlyzko, Computing π(x): The Meissel-

Lehmer method, Math. Comp. 44(1985), 537 − 560. MR 86h : 11111

[13] Stein P., Pollard’s Rho Method, Stein Patrick’s personal website at
.

281
282 BIBLIOGRAPHY

[14] Archibald C. R., Mersenne’s Numbers, The Prime Pages at

.

[15] Ribenboim P., The Little Book of Bigger Primes, Springer Science & Business
Media, 2004.

[16] Bruce J. W., A Really Trivial Proof of the Lucas-Lehmer Primality Test, The
American Mathematical Monthly 100.4 (1993), pp. 370-371.

[17] Srinivasha Ramanujan, A proof of Bertrand’s postulate, Journal of the Indian

Mathematical Society, XI, 1919, 181 − 182.

[18] Jitusoru Nagura, On the interval containing at least one prime number, Proc.
Japan Acad. Volume 28, Number 4 (1952), 177 − 181.

[19] M. El Bachraoui, Primes in the Interval [2n, 3n], Int. J. Contemp. Math. Sci., Vol.
1, 2006, no. 13, 617 − 621.

[20] Andy Loo, Primes in the Interval [3n, 4n], Int. J. Contemp. Math. Sciences, Vol.
6, 2011, no. 38, 1871 − 1882.

[21] S. Sambasivarao, Primes in the Interval [kn, (k + 1)n].

[22] Andreescu, Titu, and Dorin Andrica. Number Theory: Structures, Examples, and
Problems. Springer Science & Business Media, 2009.

[23] Dustin G. Mixon, Another Simple Proof that the Sum of the Reciprocals of the
Primes Diverges, The American Mathematical Monthly Vol. 120, No. 11, p. 831.

[24] Hardy, Godfrey Harold, and Edward Maitland Wright. An introduction to the
theory of numbers. Oxford university press, 1979.

[25] Nasehpour, Peyman. "A Simple Criterion for Irrationality of Some Real Numbers."
arXiv preprint arXiv:1806.07560 (2018).
Chapter 5

Special Topics

Contents
. . . . . . . . . . . . . . . . . . . . . . . . . . . 284
. . . . . . . . . . . . . . . . . . . 289
. . . . . . . . . . . . . . . . . . . . . . . . . . . 293
. . . . . . . . . . . . . . . . . . . . . . . 296
. . . . . . . . . . . . . . 298
. . . . . . . . . . . . . . . . . . 301
. . . . . . . . . . . . . . . 301
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
. . . . . . . . . . . . . . . . . . . . . . . . . 304
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
. . . . . . . . . . . . . . . . . . . . . . . . . 306
. . . . . . . . . . . . . . . . . . . . . . . 308
. . . . . . . . . . . . . . . . . . . . . . . 312
. . . . . . . . . . . . . . 318
. . . . . . . . . . . 320
. . . . . . . . . . . . . . . . . . 324
. . . . . . . 329
. . . . . . . . . . . . . . . 331
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336

When you have a hammer in your hand, it’s hard refraining yourself
from treating everything as a nail.

The objective of this chapter is to provide with some very powerful tools and some
special topics, which are incredibly helpful. Some topics may not be very useful for
solving problems, but they are quite good for making someone think and thus they
encourage us to study more on them. Let’s start with a really nice lemma. This PDF was created
for Amazon.

283
284 5.1. THUE’S LEMMA

5.1 Thue’s Lemma

Thue’s Lemma is a wonderful theorem in modular arithmetic. It should have been
quite popular, but unfortunately, it is not as well-known as it should be. Here we will
see what a powerful tool it is.

Theorem 5.1.1 (Thue’s Lemma). Let n > 1 be an integer and a be an integer

coprime to n. Then, there are integers x and y so that
√
0 < |x|, |y| ≤ n, and
x ≡ ay (mod n).

We call such a solution (x, y) to the above congruence equation a small solution.
√
Proof. Let r = ⌊ n⌋. That means r is the unique integer for which r2 ≤ n < (r + 1)2 .
The number of pairs (x, y) of integers for which 0 ≤ x, y ≤ r, is (r + 1)2 . This number
is greater than n. Therefore, by pigeonhole principle, there must be two different pairs
(x1 , y1 ) and (x2 , y2 ) among these (r + 1)2 pairs so that

x1 − ay1 ≡ x2 − ay2 (mod n)

=⇒ x1 − x2 ≡ a(y1 − y2 ) (mod n).

Let x = x1 − x2 and y = y1 − y2 , so we get x ≡ ay (mod n). We only need to show√ that

x and y are non-zero (it is obvious that |x| and |y| are both less than or equal to n).
Certainly, if one of x or y is zero, the other is zero as well. If both x and y are zero,
that would mean that two pairs (x1 , y1 ) and (x2 , y2 ) are actually the same. That is not
the case because we first assumed that they are different pairs of integers. Therefore,
neither x nor y is zero and we are done.
√
Note. The condition 0 < x, y < n is important. Because of this condition, we can
rule out trivial cases and bound the small solutions as the problems require.

Corollary 5.1.2. For a prime p and an integer a coprime to p, there exist integers x
√
and y with 0 < |x|, |y| < p such that

a ≡ xy (mod p).

This lemma can be generalized even more with the same proof.

Theorem 5.1.3 (Generalization of Thue’s Lemma). Let p be a prime number and

let α and β be two real numbers so that αβ ≥ p. Then, for an integer x coprime to p,
there are integers a and b with 0 < |a| < α and 0 < |b| < β so that

a ≡ xb (mod p).

We can also generalize the latter theorem to a two-dimensional theorem.

CHAPTER 5. SPECIAL TOPICS 285

Theorem 5.1.4
√ (Two-dimensional Thue’s Lemma). Let n ≥ 2 be an integer and
define r = n. For arbitrary integers a, b, c, and d, there exist integers w, x, y, and z
with at least one of y or z non-zero such that

0 ≤ |w|, |x|, |y|, |z| ≤ r,

w ≡ ay + bz (mod n), and
x ≡ cy + dz (mod n).

Now we demonstrate some applications of the lemma. First, we show an elegant

proof of Fermat’s 4n + 1 theorem, restated in theorem ( ).

Theorem 5.1.5 (Fermat’s Theorem on Sum of Two Squares). Any prime of the
form 4n + 1 can be represented as sum of two squares.

Proof. We already know from theorem ( ) that for p ≡ 1 (mod 4), there is an x so
that

x2 ≡ −1 (mod p).
√
From Thue’s lemma, for such an x, there are integers a and b with 0 < |a|, |b| < p so
that

a ≡ xb (mod p) =⇒ a2 ≡ x2 b2 ≡ −b2 (mod p)

=⇒ a2 + b2 ≡ 0 (mod p).

The last congruence means that p|a2 + b2 , so

p ≤ a2 + b2 , but
a2 + b2 < p + p = 2p.

Therefore, a2 + b2 = p must occur.

Remark. We can prove a stronger result than that of Theorem using Fibonacci-
Brahmagupta Identity (see Identity in Appendix ). This identity states that

(a2 + nb2 )(c2 + nd2 ) = (ac − nbd)2 + n(ad + bc)2

= (ac + nbd)2 + n(ad − bc)2 .

Since we know that the product of any two numbers of the form 4k + 1 is again of the
form 4k + 1 (see the proof of Theorem ), the special case when n = 1 of the above
identity along with Theorem shows that all numbers which are comprised only of
prime divisors of the form 4k + 1 are representable as the sum of two squares.

In fact, we can use the same technique for generalizing theorem ( ).

Theorem 5.1.6. Let n ∈ {−1, −2, −3}. If n is a quadratic residue modulo a prime p,
then there are integers a and b so that a2 − nb2 = p.
286 5.1. THUE’S LEMMA

Proof. We have already proven the case n = −1. If n is a quadratic residue modulo p,
x2 ≡ n (mod p)
has a solution. Fix the integer x and take a and b as in Thue’s lemma so that
a ≡ xb (mod p) =⇒ a2 ≡ x2 b2 ≡ nb2 (mod p)
=⇒ p|a2 − nb2 .
1. If n = −2, then p ≤ a2 + 2b2 < p + 2p = 3p. This means either a2 + 2b2 = p or
a2 + 2b2 = 2p occurs. If the first equation holds, we are done. If a2 + 2b2 = 2p, we
see that a must be even. Replace a = 2a′ in the latter equation to get p = b2 +2a′2 ,
as desired.
2. If n = −3, we find p ≤ a2 + 3b2 < p + 3p = 4p. If a2 + 3b2 = 2p, then a and b are
both odd or both even. If both are even, then 2p is divisible by 4, a contradiction
since p is odd. Otherwise, a and b are both odd:
a2 + 3b2 ≡ 1 + 3 · 1 (mod 4)
=⇒ 2p ≡ 0 (mod 4).
This is, again, a contradiction. We are left with the case a2 + 3b2 = 3p. This
shows a is divisible by 3. If we take a = 3a′ , we easily observe that p = b2 + 3a′2 .

Question 5.1.7. Can you prove a similar result to that of the remark after Theorem
, but for the above theorem? Try using Fibonacci-Brahmagupta’s identity before
reading the next corollary.
Corollary 5.1.8. For a prime p and an integer n with p ∤ n the following two statements
are equivalent:
• There exist coprime integers x and y so that p divides x2 + ny 2 .
• −n is a quadratic residue modulo p.
Proof. First, assume that p|x2 + ny 2 . Then, y must be coprime to p. Therefore, y has
an inverse modulo p, say a. So, ay ≡ 1 (mod p). Then, a2 y 2 ≡ 1 (mod p), and
p|x2 + ny 2 =⇒ p|a2 x2 + na2 y 2
=⇒ p|a2 x2 + n
=⇒ (ax)2 ≡ −n (mod p).
Now, suppose that −n is a quadratic residue modulo p. Let k 2 ≡ −n (mod p). Clearly,
(k, p) = 1, otherwise p will divide n. From Thue’s lemma, there are integers x and y
such that
x ≡ ky (mod p) =⇒ x2 ≡ k 2 y 2 ≡ −ny 2 (mod p)
=⇒ p|x2 + ny 2 .
CHAPTER 5. SPECIAL TOPICS 287

We can use these results to imply the following theorem.

Theorem 5.1.9. For D ∈ {1, 2, 3}, if n = x2 + Dy 2 for some coprime integers x and y,
then every divisor d of n is of the same form as n.

Proof. According to the Fibonacci-Brahmagupta Identity (identity ( ) in appendix

( )),

(a2 + Db2 )(c2 + Dd2 ) = (ac − Dbd)2 + D(ad + bc)2

= (ac + Dbd)2 + D(ad − bc)2 .

This means that the product of two numbers of the form x2 + Dy 2 is of the same form.
From theorems above, if p is a divisor of x2 + Dy 2 , then p = a2 + Db2 for some integers
a and b. The identity clearly says that if m = a2 + Db2 , then any power of m, say, mk ,
is of this form again. Let’s assume that the prime factorization of n is

k
Y
n = pe11 · · · pekk = pei i .
i=1

Then, since d is a factor of n, the factorization of d is

k
Y
d= pfi i where 0 ≤ fi ≤ ei .
i=1

For any 1 ≤ i ≤ k, pi divides n = x2 + Dy 2 . Therefore, according to corollary ( ),

−D is a quadratic residue modulo pi . Now, by theorem ( ), each pi is of the form
x2 + Dy 2 . From our previous discussion, we find that pfi i is of the same form for all i.
As a consequence, the product pf11 · · · pfkk = d is of the same form and we are done.

Now we prove another theorem that demonstrates the power of Thue’s lemma. We
will use a theorem which we proved in section ( ). For convenience, we state the
theorem here again.

Theorem 5.1.10. −3 is a quadratic residue modulo p if and only if p is of the form

3k + 1.

Using this theorem, we will prove the following.

Theorem 5.1.11. If p is a prime of the form 3k + 1, there are integers a and b such
that p = a2 + ab + b2 .

Proof. Since p is of the form 3k + 1, −3 is a quadratic residue of p. Take y to be an

odd integer for which p|y 2 + 3 or,

y 2 ≡ −3 (mod p).
288 5.1. THUE’S LEMMA

Such an y exists since p is odd. Then, the congruence equation y ≡ 2x + 1 (mod p) has
an integer solution for x. For that x, we get

(2x + 1)2 ≡ −3 (mod p)

4x2 + 4x + 1 ≡ −3 (mod p)
4(x2 + x + 1) ≡ 0 (mod p)
x2 + x + 1 ≡ 0 (mod p).

The latter congruence equation holds because p is odd. From Thue’s lemma, there are
√
integers a and b with 0 < |a|, |b| < p such that

a ≡ xb (mod p).

Then,

a2 + ab + b2 ≡ (xb)2 + (xb) · b + b2
≡ b2 (x2 + x + 1)
≡ 0 (mod p).

Since p|a2 + ab + b2 , we have p ≤ a2 + ab + b2 . On the other hand,

p ≤ a2 + ab + b2
<p+p+p
= 3p.

Consequently, either a2 + ab + b2 = p or a2 + ab + b2 = 2p happens. We can easily check

that a2 + ab + b2 = 2p can not happen (try it yourself). Thus, a2 + ab + b2 = p, which
is what we wanted.

You have probably figured out by now that our focus should be on the small
solutions so that we can bound the necessary expressions like the problem asks for.
Let’s see more examples on this.

Theorem 5.1.12. Let p > 5 be a prime which divides k 2 + 5 for some integer k. Show
that there are integers x and y such that p2 = x2 + 5y 2 .

Hint. Try to find x such that x2 ≡ −5 (mod p2 ). Then from Thue’s lemma, there exist
a and b so that a2 , b2 < p and a ≡ kb (mod p2 ). This gives a2 ≡ k 2 b2 ≡ −5b2 (mod p2 ).
Now, check all the cases like we did before.

Problem 5.1.13. Let p be a prime for which there exists a positive integer a such that
p divides 2a2 − 1. Prove that there exist integers b and c so that p = 2b2 − c2 .

Solution. Let’s look for small solutions again for the purpose of bounding! We have
2a2 − 1 ≡ 0 (mod p). Since we want to bound 2b2 − c2 , it is obvious that we must
find b and c so that p divides 2b2 − c2 and then bound it. Fix the integer a, which
CHAPTER 5. SPECIAL TOPICS 289

is clearly coprime to p. Then from Thue’s lemma, we there are integers b and c with
√
0 < |b|, |c| < p so that

b ≡ ac (mod p).

This gives us what we need. Note that

2b2 − c2 ≡ 2(ac)2 − c2
≡ c2 (2a2 − 1)
≡ 0 (mod p).

Thus, p divides 2b2 − c2 , and now we get to use the fact that

p ≤ 2b2 − c2
< 2b2
< 2p.

We immediately get that p = 2b2 − c2 .

5.2 Chicken McNugget Theorem

You are probably wondering how come this can be the name of a theorem if you have
encountered it for the first time. The name just might be the weirdest of all names
a theorem can possibly assume! Here is the reason behind such a name: The story
goes that the Chicken McNugget Theorem got its name because in McDonalds, people
bought Chicken McNuggets in 9 and 20 piece packages. Somebody wondered what
the largest amount you could never buy was, assuming that you did not eat or take
away any McNuggets. They found the answer to be 151 McNuggets, thus creating the
Chicken McNugget Theorem. Actually it is Sylvester’s Theorem, now known as the
Chicken McNugget Theorem. The problem is known as Frobenius Coin Problem, which
is a generalization of this one. Have you ever wondered about the coin system of your
own country? It is designed in a way so that you should never face a situation where
you can not exchange a certain amount of money. But have you thought how it is
possible? In this section, we will deal with problems like this. First think for yourself
on the following two problems:

Problem 5.2.1. You are in a strange country where only two units are available for
exchange: 4 and 6. Can you pay any amount you want?

Problem 5.2.2. In another country, you see that only two units are available for
exchange: 3 and 10. Can you pay any amount you want?

If you have come to the right conclusions, you will see that you can not pay any
amount you want in the first case. But you can pay whatever you want with the second
one. Let’s say two units available value a and b. So if you use a unit x times and b unit
y times, the total amount of money you can pay is ax + by. Here, x, y can be negative
290 5.2. CHICKEN MCNUGGET THEOREM

or non-negative integers. If x > 0, it will mean you are paying, or if x < 0 it will mean
you are being paid (or getting the exchange). Therefore, if you need to pay exactly n
amount, you need integers x and y with
ax + by = n.
Play with some more values of a and b. You will understand that you can pay n amount
with units a and b if and only (a, b) divides n. Here is another intuitive fact: If we can
pay just 1, we can pay any amount we want with as many 1s needed. So we should
focus on when we can pay 1 by a and b. This tells us, a and b must be co-prime. And
from Bézout’s Identity, for any co-prime a and b, we will get integers x, y so that
ax + by = 1.
In the problems above, we can’t pay any amount with 4 and 6 because they are not
co-prime. But we can pay any amount that is a multiple of (4, 6) = 2. But we can pay
any amount with 3 and 10 because they are co-prime. This leads us to the following
theorem.
Theorem 5.2.3. Any integer can be written as a linear combination of a and b if and
only if a ⊥ b.
By linear combination, we mean using only a and b as many times as we want. Now
we see the same problem from another perspective. Consider the following problem. If n
can be written as ax+by for non-negative x, y, we will call n a good number. Otherwise,
n is bad. But to do that, we can’t change the values of a and b simultaneously. Therefore,
we fix two co-prime integers a and b. Next, let’s see why we are only considering a ⊥ b.
If (a, b) = g and g > 1, then we already know that only multiples of g can be good.
But we want as many integers to be good as possible, and not skipping some integers
is better.
Problem 5.2.4. A shop sells nuggets in packages of two sizes, 3 nuggets and 10 nuggets.
What is the maximum number of nuggets that cannot be expressed as a nonnegative
combination of these package sizes?
Definition 5.2.5 (Frobenius Number). For two integers a and b, the largest bad
integer is the Frobenius number. In fact, it can be generalized for n natural numbers.
If a1 , . . . , an are natural numbers so that (a1 , . . . , an ) = 1, the largest natural number
that can not be written as a1 x1 + . . . an xn for nonnegative x1 , . . . , xn is the Frobenius
number. It is denoted as Fn (a1 , . . . , an ). Here, we will deal with the case n = 2, F2 (a, b).
The following theorem answers this question.
Theorem 5.2.6 (Sylvester’s Theorem, 1882). Let a and b be two co-prime positive
integers greater than 1. Then the maximum integer that can not be expressed as ax+by
for non-negative integer x, y is ab − a − b.
If we can prove that for all N > ab − a − b, there are non-negative integers x, y such
that
N = ax + by,
and that for N ≤ ab − a − b, there are no such x and y, we are done. First, let’s prove
the next lemma.
CHAPTER 5. SPECIAL TOPICS 291

Lemma 5.2.7. ab − a − b is a bad number.

Proof. On the contrary, let’s assume that

ab − a − b = ax + by,

for some x, y ∈ N0 . We can rewrite it as

a(x − b + 1) = −b(y + 1).

From this equation, a|b(y + 1) but a ⊥ b. So, a|y + 1. Again, b|a(x − b + 1) but b ⊥ a
so b|x − b + 1 or b|x + 1. We get x + 1 ≥ b and y + 1 ≥ a, and so

x ≥ b − 1 and y ≥ a − 1.

Using these inequalities,

ax + by ≥ a(b − 1) + b(a − 1)
= ab − a + ab − b
=⇒ ab − a − b ≥ 2ab − a − b,

which is a contradiction.
The above lemma shows that F2 (a, b) ≥ ab − a − b. It only remains to prove the
following lemma:

Lemma 5.2.8. Any integer n > ab − a − b is good.

Proof. Since (a, b) = 1, by Bézout’s identity, there are integers u and v so that

au + bv = 1 =⇒ anu + bnv = n
(5.1) =⇒ ax0 + by0 = n.

We need to show that such x0 , y0 ≥ 0 exist. If (x0 , y0 ) is a solution of equation ( ), then

so is (x0 − bt, y0 + at) for any integer t. Here, one can choose t such that 0 ≤ x0 − bt < b.
In case you don’t understand how we can choose such t, just divide x0 by b. Then
x0 = bq + r, where 0 ≤ r < b. This means that 0 ≤ x0 − bq < b, so one choice for t is
q. So we know that there exists some x0 such that 0 ≤ x0 < b. We will show that y0 is
also positive. Note that

ax0 + by0 = n > ab − a − b

=⇒ b(y0 + 1) > a(b − x0 − 1).

Since we know that x0 < b, we get b − x0 − 1 ≥ 0. This means that b(y0 + 1) > 0,
so y0 + 1 > 0, i.e., y ≥ 0. Therefore, there is a valid solution (x0 , y0 ) and the proof is
complete.
Now, the proof is complete. The same proof can be used for generalizing the case
where (a, b) > 1.
292 5.2. CHICKEN MCNUGGET THEOREM

Theorem 5.2.9 (Generalization of Sylvester’s Theorem). Let a, b be positive

integers with (a, b) = g. Then every integer
(a − g)(b − g)
n≥
g
such that g|n is good. Also,
(a − g)(b − g)
F2 (a, b) = − g,
g
i.e., F2 (a, b) is the largest non-trivial bad integer.
We see some problems related to this theorem. A classical example would be the
following problem that appeared at the IMO 1983.
Problem 5.2.10 (IMO 1983). Let a, b, c ∈ N with (a, b) = (b, c) = (c, a) = 1. Prove
that, 2abc−ab−bc−ca is the largest integer that can not be expressed as xbc+yca+zab
for non-negative x, y, z.
Solution. Clearly, we need to invoke Sylvester’s theorem here. But the expression tells
us, it can not be done in one step. Note that
xbc + yca + zab = c(bx + ay) + zab.
Therefore, we should first focus only on bx + ay first. From McNugger theorem, any
integer greater than ab − a − b is good. So we substitute ab − a − b + 1 + t for some
non-negative t into the equation and get
xbc + yca + zab = c(bx + ay) + zab
= c(ab − a − b + 1 + t) + zab
= abc − bc − ca + c + ct + zab.
This again calls for using the theorem for c and ab. Again, every integer greater than
abc−ab−c is good. So we substitute ct+zab = abc−ab−c+1+w for some non-negative
w. Then
xbc + yca + zab = abc − bc − ca + c + ct + zab
= abc − bc − ca + c + abc − ab − c + 1 + w
= 2abc − ab − bc − ca + 1 + w
This shows that all integers greater than 2abc − ab − bc − ca are good. Finally, in
order to prove the claim, we just have to show that 2abc − ab − bc − ca is bad. To
the contrary, assume that 2abc − ab − bc − ca = xbc + yca + zab for some non-negative
x, y, z. We have
bc(x + 1) + ca(y + 1) + ab(z + 1) = 2abc
Clearly, a|x + 1 because a|bc(x + 1) but gcd(a, bc) = 1. Similarly, b|y + 1 and c|z + 1.
This gives us bc(x + 1) + ca(y + 1) + ab(z + 1) ≥ bca + cab + abc or 2abc ≥ 3abc which
is obviously wrong.
So, the problem is solved. As you can see, the theorem is fairly easy to understand
and use in problems. There will be some related problems in the problem column. See
if you can get how to solve those using this (first you have to understand that this
theorem will come to the rescue though).
CHAPTER 5. SPECIAL TOPICS 293

5.3 Vietta Jumping

By now, Vietta jumping has become a standard technique for solving some particular
type of olympiad number theory problems. It is also known as Root Jumping or Root
Flipping. Though it involves Diophantine equations and for now, it is out of our scope,
many divisibility or congruence problems can be turned into one that can be solved
using this tactic. Hence, this section. To understand just how popular it has been, let’s
just mention that there are at least two IMO problems that have standard solutions
using this particular technique. And surely, there are many other olympiad problems
that fall into the same category. Now, let’s see what it is and what it actually does.
Consider the following quadratic equation

ax2 + bx + c = 0.

According to Vietta’s formula, if two of its roots are x1 and x2 , then

b
x1 + x2 = − , and
a
c
x1 x2 = .
a
Vietta jumping relies on these two equations. It is in fact, a descent method in which
we usually prefer using one of the following two methods:

(i) Standard Descent: It is usually used to show that the equation doesn’t have
any solution or some sort of contradiction to prove a claim, like we do in Infinite
Descent. For a solution (x, y) of the equation, we define a function dependent on
x, y (x + y is such a common function, as we will see later). Then we consider
the solution that minimizes that function over all solutions possible. If there are
multiple solutions that can achieve this, we are free to choose any one depending
on the problem. But then, using Vietta’s formulas, we try to find another solution
that makes the function’s value smaller, which gives us the necessary contradiction.
So, this is a modified version of infinite descent.

(ii) Constant Descending: Sometimes, we take some constants, for example, an

integer k and fix it for the whole problem. For a solution (a, b), we fix b and k.
Then using those formulas, we find a solution x so that 0 < x < b so that it
produces a solution (b, x) smaller than (a, b). Note that, here, we have to take
b < a so that the new solution is guaranteed to be smaller. Repeating this, we
will reach a base case and those constants (k, for example) will remain constant
through the whole process. Thus, we will show what’s required.

(iii) Sometimes, there can be even geometric interpretations. For example, Arthur
Engel showed one in his book Problem Solving Strategies chapter 6, problem 15.

We will now demonstrate this using some example problems. Let’s start with the
classical problem from IMO 1988. Here is what Engel said about this problem in his
book:
294 5.3. VIETTA JUMPING

Nobody of the six members of the Australian problem committee could solve
it. Two of the members were George Szekeres and his wife Esther Klein,
both famous problem solvers and problem creators. Since it was a number
theoretic problem it was sent to the four most renowned Australian number
theorists. They were asked to work on it for six hours. None of them could
solve it in this time. The problem committee submitted it to the jury of
the XXIX IMO marked with a double asterisk, which meant a superhard
problem, possibly too hard to pose. After a long discussion, the jury finally
had the courage to choose it as the last problem of the competition. Eleven
students gave perfect solutions.

Problem 5.3.1 (IMO 1988, Problem 6). Let a and b be two positive integers such
a2 + b2
that ab + 1 divides a2 + b2 . Show that is a perfect square.
ab + 1

Solution. Let k be an integer so that

a2 + b2
=k
ab + 1
=⇒ a2 + b2 = kab + k
=⇒ a2 − kab + b2 − k = 0.

As we said in the process, we will fix k and consider all pairs of integers (a, b) that
gives us k as the quotient. And take a solution (a, b) in nonnegative integers so that
the sum a + b is minimum (and if there are multiple such (a, b), we take an arbitrary
one). Without loss of generality, we can assume a ≥ b > 0. Now, fix b and set a = x
which will be the variable. We get an equation which is quadratic in x with a root a:

x2 − kbx + b2 − k = 0.

Using Vietta, we get that x + a = kb or x = kb − a. From this, we infer x is integer.

Note that, we can write it in another way:

b2 − k
x= .
a
This equation will do the talking now! Firstly, if x = 0, we are done since that would
give us b2 − k = 0 or k = b2 , a perfect square. So, we can assume that x 6= 0. To
descend the solution, we will need x > 0. For the sake of contradiction, take x = −z
where z > 0. But that would give us

x2 − kbx + b2 − k = z 2 + kbz + b2 − k
≥ z 2 + k + b2 − k
= z 2 + b2 > 0.

This is impossible. Thus, x > 0 and now, if we can prove that 0 < x < a, then we will
CHAPTER 5. SPECIAL TOPICS 295

have a solution (x, b) smaller than (a, b). We actually have this already because

b2 − k
x=
a
b2
< since k is a positive integer
a
a2
≤ = a.
a

Therefore, we must have a solution (0, b) for the equation which gives us k = b2 .

Problem 5.3.2. Let a and b be positive integers such that ab divides a2 + b2 + 1. Prove
that a2 + b2 + 1 = 3ab.
2 2
Solution. Again, let k = a +b
ab
+1
and among all the solutions of the equation, consider
the solution that minimizes the sum a + b. We can also assume that, a ≥ b. Now for
applying Vietta, we rewrite it as

a2 − kab + b2 + 1 = 0.

Just like before, let’s fix b and make it quadratic in x, which already has a solution a:

x2 − kbx + b2 + 1 = 0.

For the other solution, we have

b2 + 1
(5.2) x=
a
(5.3) = kb − a.

Equation ( ) implies that x is positive and equation ( ) implies that x is an integer.

12 + 12 + 1
Now, if a = b, we already get that k = = 3. So we are left with a > b. But
1·1
then,

b2 + 1 b2 + 2b + 1 (b + 1)2 a2
x= < = ≤ = a,
a a a a
which again produces a smaller sum x + b < a + b. This is a contradiction, so a = b
must happen.

Problem 5.3.3 (Romanian TST 2004). Find all integer values the expression
a2 + b 2 + 1
can assume for ab 6= 1 where a and b are positive integers.
ab − 1

Solution. Take
a2 + ab + b2
k= ,
ab − 1
296 5.4. EXPONENT GCD LEMMA

or a2 − a(kb − b) + k + b2 = 0 and fix b, when we consider the smallest sum a + b for a

solution (a, b) where a ≥ b. Consider it as a quadratic in x again which has a solution
a:
x2 − x(kb − b) + b2 + k = 0
=⇒ x + a = kb − b =⇒ x = kb − a − b, and
b2 + k
xa = b2 + k =⇒ x = .
a
We have that x is a positive integer. Since a + b is minimal, we have x ≥ a. So
b2 + k
≥a
a
=⇒ k ≥ a2 − b2 .
a2 + ab + b2
But k = , so
ab − 1
a2 + ab + b2
≥ a2 − b2
ab − 1
=⇒ a2 + ab + b2 ≥ (a2 − b2 )(ab − 1) = ab(a + b)(a − b) − a2 + b2
=⇒ a(a + b) ≥ ab(a + b)(a − b) − a2
(5.4) =⇒ a ≥ (a + b)(ab − b2 − 1).
3a2
If a = b, then k = 2 . Since a2 ⊥ a2 − 1, we have a2 − 1 divides 3 or a = 2. In that
a −1
a2 + a + 1
case, k = 4. If b = 1, then k = so a − 1 divides a2 + a + 1.
a−1
a − 1|a2 − 1
=⇒ a − 1|a2 + a + 1 − (a2 − 1) = a + 2
=⇒ a − 1|(a + 2) − (a − 1) = 3.
We get that a = 2 or a = 4. If a = 2 or a = 4, then k = 7. If a > b > 1, then, a ≥ b + 1
and we have
(a + b)(ab − b2 − 1) > a,
which is in contradiction with equation ( ). Therefore, we have k = 4 or k = 7.
Problem 5.3.4 (Mathlinks Contest). Let a, b, c, d be four distinct positive integers
in arithmetic progression. Prove that abcd is not a perfect square.

5.4 Exponent gcd Lemma

For brevity assume
xn − y n
f (x, y, n) = .
x−y
Remember from definition ( ) that vp (n) = α means α is the greatest positive
integer so that pα |n. Alternatively, we can denote this by pα kn.
CHAPTER 5. SPECIAL TOPICS 297

Theorem 5.4.1 (Exponent gcd Lemma). If x ⊥ y, then


g = x − y, f (x, y, n) n.

Proof. Re-call the identity

xn − y n = (x − y) xn−1 + xn−2 y + · · · + xy n−2 + y n−1 .

This yields to

f (x, y, n) = xn−1 + xn−2 y + . . . + xy n−2 + y n−1

We know that for polynomials P and Q, if

P (x) = (x − a) · Q(x) + r,

then r = P (a) (the reason is simple, just plug in a into P ). So, in this case,

f (x, y, n) = (x − y) · Q(x, y, n) + r.

Hence, r = f (y, y, n) which equals

f (y, y, n) = y n−1 + y n−2 · y + . . . + y n−1 = ny n−1 .

From Euclidean algorithm, we can infer

x − y, f (x, y, n) = x − y, f (y, y, n) = x − y, ny n−1 .

Earlier we assumed x ⊥ y, and so x − y ⊥ y n−1 because (x − y, y) = (x, y) = 1. Thus

g = x − y, f (x, y, n) = x − y, n ,

which results in g|n.

Corollary 5.4.2. The following result is true for any odd positive integer n:
 
xn + y n 
x + y, n.
x+y

Corollary 5.4.3. For a prime p,

(x − y, f (x, y, p)) = 1 or p.

Let’s see how we can use this lemma to solve problems.

Problem 5.4.4 (Hungary 2000). Find all positive primes p for which there exist
positive integers n, x, and y such that

x3 + y 3 = pn .
298 5.5. A CONGRUENCE LEMMA INVOLVING GCD

Solution. For p = 2, x = y = 1 works. Assume p is greater than 2, and hence odd.

If (x, y) = d, then we have d|pn . So, d is a power of p. But in that case, we can
divide the whole equation by d and still it remains an equation of the same form. Let’s
therefore, consider (x, y) = 1. Factorizing,

(x + y)(x2 − xy + y 2 ) = pn .

According to the lemma,


g = (x + y, f (x, y, 3)) (x + y, 3)

This means g|3. If g = 3, then we have 3|p or p = 3. On the other hand, g = 1

shall mean that x + y = 1 or x2 − xy + y 2 = 1. Neither of them is true because
x, y > 0, x + y > 1 and (x − y)2 + xy > 1.

Problem 5.4.5. Find all primes p and positive integer x such that

px − 1 = (p − 1)!.

Solution. We know that if n ≥ 6 is a composite integer, then n divides (n − 1)!. Now,

px − 1  px − 1
= (p − 2)!. Assume p > 5, then p ≥ 7 so p − 1|(p − 2)!. Thus, p − 1 and
p−1 p−1
so from the lemma, p − 1|x or x ≥ p − 1. So

(p − 1)! = px − 1 ≥ pp−1 − 1,

which is not true since

n! < (n + 1)n − 1,

for n > 1. So we need to check for only p ∈ {2, 3, 5}. If p = 2, then 2x − 1 = 1, so

x = 1. If p = 3, then 3x − 1 = 2 so x = 1. If p = 5, 5x − 1 = 24 so x = 2.

5.5 A Congruence Lemma Involving gcd

In this section, we discuss yet another lemma, which involves gcd like the previous one.
The first author of this book finds it really useful for solving some types of problems.
The lemma was proved in Theorem ( ) of chapter ( ).

Lemma 5.5.1. Let a, b, and n be three positive integers such that (a, n) = (b, n) = 1
and

ax ≡ b x (mod n),
ay ≡ by (mod n),

then

a(x,y) ≡ b(x,y) (mod n).

CHAPTER 5. SPECIAL TOPICS 299

Corollary 5.5.2. Let p be a prime and let a and b be integers not divisible by p so
that

ak ≡ bk (mod p).

Then

a(k,p−1) ≡ b(k,p−1) (mod n).

The following corollary also proves theorem ( ) easily.

Corollary 5.5.3. Let a, b, and n be three positive integers such that (a, n) = (b, n) = 1.
If h is the smallest integer such that

ah ≡ bh (mod n),

and k is an integer such that

ak ≡ b k (mod n),

then h|k.

Proof. From the lemma, we have a(h,k) ≡ b(h,k) (mod n). We have (h, k) ≤ h and
(h, k)|k. Now, if (h, k) < h then (h, k) is smaller than h which satisfies the condition.
So we must have (h, k) = h, or h|k.

Corollary 5.5.4. Let p be a prime and let a be a positive integer. If ordp (a) = d and
ak ≡ 1 (mod p), then d|(p − 1, k).

Proof. From Fermat’s little theorem, ap−1 ≡ 1 (mod p). From the theorem, a(k,p−1) ≡ 1
(mod p) and from corollary above, d|(k, p − 1).
You should see that if a problem can be solved using the dividing property of order,
then we can solve it using this lemma as well. Let’s see some problems that this lemma
is useful with. Sometimes, we have to couple this lemma with some other techniques
such as the smallest prime factor trick.

Problem 5.5.5. Find all n ∈ N such that 2n − 1 is divisible by n.

A standard problem with a very nice idea. There are many ways to start working
on such problems. A common one is to find the prime factors of n first. That way,
we have some idea about n at first, from which we can understand the nature of the
problem. Sometimes we have to find special prime factors first. The special
prime factors can provide some extra information necessary.

Solution. Here, we consider the smallest prime divisor of n. Let’s call this prime p.
Since n divides 2n − 1, p divides it too. Because 2n − 1 is odd, both n and p must be
odd. So

2n ≡ 1 (mod p).
300 5.5. A CONGRUENCE LEMMA INVOLVING GCD

This equation alone does not say a lot, so we need more information. Remember
Fermat’s little theorem! This is another reason to find primes first. Only for primes
we can get the power ap−1 , otherwise from Euler’s Totient theorem, it would be aϕ(n)
which would bring troubles in this case. We have

2p−1 ≡ 1 (mod p)

Whenever you get two congruences like this, be sure to use theorem ( ). Using
this,

2(n,p−1) ≡ 1 (mod p)

Now you will see why we specifically chose the smallest prime divisor instead of an
arbitrary prime divisor. Since p is the smallest prime divisor of n, if a prime q divides
p − 1, it can not divide n. Because if q|n, then q ≤ p − 1 < p, which is a smaller
prime divisor than the smallest prime divisor of n, a contradiction! We must have
(n, p − 1) = 1. But then 21 ≡ 1 (mod p) or p|2 − 1 = 1. Another contradiction. This
means for no prime p, n is divisible by p. So n can not have any primes i.e. n = 1.
Note. Not just smallest prime divisor, depending on the problem we occasionally take
the greatest prime divisor or something that makes our job easier to do. See the
following problems for better understanding.
Problem 5.5.6. Determine all pairs of primes (p, q) such that pq|pp + q q + 1.
Solution. If (p, q) is a solution, so is (q, p). Without loss of generality, assume that
p < q since p = q implies p|1. Now, pq|pp + q q + 1 gives us two things: p|q q + 1
and q|pp + 1.. Consider p = 2, then q|pp + 1 = 5, so q = 5. Now, p is odd and so
q > p + 1. We can alternatively write them as q 2q ≡ 1 (mod p) and p2p ≡ 1 (mod q).
From Fermat’s theorem, we also have q p−1 ≡ 1 (mod p) and pq−1 ≡ 1 (mod q). Thus,
q gcd(2q,p−1) ≡ 1 (mod p) and pgcd(2p,q−1) ≡ 1 (mod q). Since q is odd and greater than
p − 1, gcd(q, p − 1) = 1. We have q 2 ≡ 1 (mod p) or p divides (q + 1)(q − 1). If
p divides q − 1, then p also divides q q − 1. But that would force the contradiction
p|q q + 1 − (q q − 1) = 2. So, p must divide q + 1. On the other hand, since p can’t divide
q − 1, we get gcd(2p, q − 1) = 2. This gives p2 ≡ 1 (mod q) or q|(p + 1)(p − 1). This is
impossible since q divides none of p ± 1. So no other solutions.
Problem 5.5.7. Find all primes p, q such that pq|(5p − 2p )(5q − 2q ).
Solution. If p|5p − 2p , from FLT (Fermat’s little theorem) we get

5p − 2p ≡ 5 − 2 ≡ 3 (mod p).

So, p must be 3. Then if p = q, q = 3. Otherwise, q|5p − 2p = 53 − 23 = 117 = 32 · 13

so q = 13. Now, we can assume p|5q − 2q and q|5p − 2p . It is obvious, none of p or q
can be 2 or 5. From the lemma,

5(p,q−1) ≡ 2(p,q−1) (mod q).

Here, p > q − 1, so p⊥q − 1. Therefore, 51 ≡ 2 (mod q) or q = 3.

CHAPTER 5. SPECIAL TOPICS 301

5.6 Lifting the Exponent Lemma

Lifting The Exponent Lemma is a powerful method for solving exponential Diophantine
equations. It is pretty well-known in the literature though its origins are hard to trace.
Mathematically, it is a close relative of Hensel’s lemma in number theory (in both the
statement and the idea of the proof). This is a technique that has been used a lot in
recent olympiad problems.
One can use the Lifting The Exponent Lemma (this is a long name, let’s call it
LTE!) in problems involving exponential equations, especially when there are some
prime numbers (and actually in some cases overkills the problems). This lemma shows
how to find the greatest power of a prime p – which is often ≥ 3 – that divides an ± bn
for some positive integers a and b. The advantage of this lemma is that, it is quite
simple to understand and if in some contest, it is refrained from being used, the proof
is not hard as well.
In section ( ) of chapter ( ), we defined vp (n). Recall that vp (n) is the highest
power of a prime p which divides a positive integer n. Here, we will make use of this
function to solve Diophantine equations.
Here is a problem which will explain the main idea behind LTE.

Problem 5.6.1. Show that there exist no positive integers x and y such that

26x+1 + 1 = 32y .

Solution. The idea is that the largest power of 3 which divides the right side of the
given equation, should be the same as that of the left side. Clearly, v3 (32y ) = 2y. Let’s
find v3 (26x+1 + 1). Since 6x + 1 = 2 · 3 · x + 1 is odd, we can write


26x+1 + 1 = (2 + 1) 26x − 26x−1 + 26x−2 − · · · − 2 + 1 .

(26x − 26x−1 + 26x−2 − · · · − 2 + 1) is not divisible by 3 (try to figure out why, using
induction on x). Therefore



v3 26x+1 + 1 = v3 (2 + 1) + v3 26x − 26x−1 + 26x−2 − · · · − 2 + 1
=1+0
= 1.

This means that 2y = 1, which is impossible since y is an integer.

5.6.1 Two Important and Useful Lemmas

Lemma 5.6.2. Let x and y be (not necessarily positive) integers and let n be a positive
integer. Given an arbitrary prime p (in particular, we can have p = 2) such that
gcd(n, p) = 1, p | x − y and neither x, nor y is divisible by p (i.e., p ∤ x and p ∤ y). We
have


vp xn − y n = vp (x − y).
302 5.6. LIFTING THE EXPONENT LEMMA

Proof. We use the fact that

xn − y n = (x − y) xn−1 + xn−2 y + xn−3 y 2 + · · · + y n−1 .

Now if we show that p ∤ xn−1 + xn−2 y + xn−3 y 2 + · · · + y n−1 , then we are done. In order
to show this, we use the assumption p | x − y. So we have x − y ≡ 0 (mod p), or x ≡ y
(mod p). Thus

xn−1 +xn−2 y + xn−3 y 2 + · · · + y n−1

≡ xn−1 + xn−2 · x + xn−3 · x2 + · · · + x · xn−2 + xn−1
≡ nxn−1
6≡ 0 (mod p).

This completes the proof.

Lemma 5.6.3. Let x and y be (not necessarily positive) integers and let n be an odd
positive integer. Given an arbitrary prime p (in particular, we can have p = 2) such
that gcd(n, p) = 1, p | x + y and neither x, nor y is divisible by p, we have


vp xn + y n = vp (x + y).

Proof. Since x and y can be negative in lemma ( ) we only need to put (−y)n
instead of y n in the formula to obtain



vp xn − (−y)n = vp (x − (−y)) =⇒ vp xn + y n = vp (x + y).

Note that since n is an odd positive integer we can replace (−y)n with −y n .

5.6.2 Main Result

Theorem 5.6.4 (First Form of LTE). Let x and y be (not necessarily positive)
integers, let n be a positive integer, and let p be an odd prime such that p | x − y and
none of x and y is divisible by p (i.e., p ∤ x and p ∤ y). We have

vp (xn − y n ) = vp (x − y) + vp (n).

Proof. We may use induction on vp (n). First, let us prove the following statement:

(5.5) vp (xp − y p ) = vp (x − y) + 1.

In order to prove this, we will show that

(5.6) p | xp−1 + xp−2 y + · · · + xy p−2 + y p−1

and

(5.7) p2 ∤ xp−1 + xp−2 y + · · · + xy p−2 + y p−1 .

CHAPTER 5. SPECIAL TOPICS 303

For ( ), we note that

xp−1 + xp−2 y + · · · + xy p−2 + y p−1 ≡ pxp−1 ≡ 0 (mod p).

Now, let y = x + kp, where k is an integer. For an integer 1 ≤ t < p we have

y t xp−1−t ≡ (x + kp)t xp−1−t

 
p−1−t t t−1 t(t − 1) 2 t−2
≡x x + t(kp)(x ) + (kp) (x ) + · · ·
2


≡ xp−1−t xt + t(kp)(xt−1 )
≡ xp−1 + tkpxp−2 (mod p2 ).

This means

y t xp−1−t ≡ xp−1 + tkpxp−2 (mod p2 ), t = 1, 2, . . . , p − 1.

Using this fact, we have

xp−1 + xp−2 y + · · · + xy p−2 + y p−1

≡ xp−1 + xp−1 + kpxp−2 + xp−1 + 2kpxp−2 + · · · + xp−1 + (p − 1)kpxp−2


≡ pxp−1 + 1 + 2 + · · · + p − 1 kpxp−2
 
p−1 p(p − 1)
≡ px + kpxp−2
2
 
p−1 p−1
≡ px + kp2 xp−1
2
p−1
≡ px 6≡ 0 (mod p2 ).

So we proved the relation ( ) and the proof of equation ( ) is complete. Now let us
return to our problem. We want to show that


vp xn − y n = vp (x − y) + vp (n).

Suppose that n = pα b where gcd(p, b) = 1. Then

α α

vp (xn − y n ) = vp (xp )b − (y p )b
α α
α−1 α−1

= vp xp − y p = vp (xp )p − (y p )p
α−1 α−1
α−2 α−2

= vp xp − y p + 1 = vp (xp )p − (y p )p + 1
α−2 α−2

= vp xp − y p +2
..
.
1 1


= vp (xp )1 − (y p )1 + α − 1 = vp x − y + α



= vp x − y + vp n .

Note that we used the fact that if p | x − y, then we have p | xk − y k , because we

have x − y | xk − y k for all positive integers k. The proof is complete.
304 5.6. LIFTING THE EXPONENT LEMMA

Theorem 5.6.5 (Second Form of LTE). Let x, y be two integers, n be an odd

positive integer, and p be an odd prime such that p | x + y and none of x and y is
divisible by p. We have
vp (xn + y n ) = vp (x + y) + vp (n).
Proof. This is obvious using theorem ( ). See the trick we used in proof of lemma
( ).
The following theorem is a special case of Zsigmondy’s theorem (discussed later in
( )), which can be proved using LTE and EGL (theorem ( )). And probably it is
the most important case of Zsigmondy’s theorem we use in problems. In case someone
considers the original theorem to be a sledgehammer, in that case this theorem should
work fine. We leave the proof as an exercise.
k k
Theorem 5.6.6. For a prime p > 3 and coprime integers x, y, xp − y p has a prime
k k i i
factor q such that q|xp − y p but q ∤ xp − y p for 0 ≤ i < k.

5.6.3 The Case p = 2

Question 5.6.7. Why did we assume that p is an odd prime, i.e., p 6= 2? Why can’t
we assume that p = 2 in our proofs?
p−1
Hint. Note that 2
is an integer only for p > 2.
Theorem 5.6.8 (LTE for p = 2). Let x and y be two odd integers such that 4 | x − y.
Then


v2 xn − y n = v2 (x − y) + v2 (n).
Proof. We showed that for any prime p such that gcd(p, n) = 1, p | x − y and none of
x and y is divisible by p, we have
vp (xn − y n ) = vp (x − y)
So it suffices to show that
n n
v2 (x2 − y 2 ) = v2 (x − y) + n.
Factorization gives
n n n−1 n−1 n−2 n−2
x2 − y 2 = (x2 + y2 )(x2 + y2 ) · · · (x2 + y 2 )(x + y)(x − y)
k k
Now since x ≡ y ≡ ±1 (mod 4) then we have x2 ≡ y 2 ≡ 1 (mod 4) for all positive
k k
integers k and so x2 + y 2 ≡ 2 (mod 4), k = 1, 2, 3, . . . . Also, since x and y are odd
and 4 | x − y, we have x + y ≡ 2 (mod 4). This means the power of 2 in all of the
factors in the above product (except x − y) is one. We are done.
Theorem 5.6.9. Let x and y be two odd integers and let n be an even positive integer.
Then


v2 xn − y n = v2 (x − y) + v2 (x + y) + v2 (n) − 1.
CHAPTER 5. SPECIAL TOPICS 305

Proof. We know that the square of an odd integer is of the form 4k + 1. So for odd x
and y we have 4 | x2 − y 2 . Now let m be an odd integer and k be a positive integer such
that n = m · 2k . Then

k k

v2 xn − y n = v2 xm·2 − y m·2
k−1 k−1

= v2 (x2 )2 − (y 2 )2
..
.


= v2 x2 − y 2 + k − 1
= v2 (x − y) + v2 (x + y) + v2 (n) − 1.

5.6.4 Summary
Let p be a prime number and let x and y be two (not necessarily positive) integers that
are not divisible by p. Then:

(a) For a positive integer n

• if p 6= 2 and p | x − y, then


vp xn − y n = vp (x − y) + vp (n).

• if p = 2 and 4 | x − y, then


v2 xn − y n = v2 (x − y) + v2 (n).

• if p = 2, n is even, and 2 | x − y, then

v2 xn − y n = v2 (x − y) + v2 (x + y) + v2 (n) − 1.

(b) For an odd positive integer n, if p | x + y, then

vp xn + y n = vp (x + y) + vp (n).

(c) For a positive integer n with gcd(p, n) = 1, if p | x − y, we have

vp xn − y n = vp (x − y).

If n is odd, gcd(p, n) = 1, and p | x + y, then we have

vp xn + y n = vp (x + y).

Note. The most common mistake in using LTE is when you do not check the p | x ± y
condition, so always remember to check it. Otherwise your solution will be completely
wrong.
306 5.6. LIFTING THE EXPONENT LEMMA

5.6.5 Solved Problems

Problem 5.6.10 (Russia 1996). Find all positive integers n for which there exist
positive integers x, y and k such that gcd(x, y) = 1, k > 1 and 3n = xk + y k .
Solution. k should be an odd integer (otherwise, if k is even, then xk and y k are perfect
squares, and it is well known that for integers a, b we have 3 | a2 + b2 if and only if 3 | a
and 3 | b, which is in contradiction with gcd(x, y) = 1.). Suppose that there exists a
prime p such that p | x + y. This prime should be odd. So vp (3n ) = vp (xk + y k ), and
using ( ) we have

vp (3n ) = vp (xk + y k ) = vp (k) + vp (x + y).

But p | x + y means that vp (x + y) ≥ 1 > 0 and so vp (3n ) = vp (k) + vp (x + y) > 0 and

so p | 3n . Thus p = 3. This means x + y = 3m for some positive integer m. Note that
n = v3 (k) + m. There are two cases:
1. m > 1. We can prove by induction that 3a ≥ a + 2 for all integers a ≥ 1, and so
we have v3 (k) ≤ k − 2 (why?). Let M = max(x, y). Since x + y = 3m ≥ 9, we
have M ≥ 5. Then

xk + y k ≥ M k = M
|{z} ·M k−1
| {z }
≥ x+y
2
= 12 ·3m ≥5k−1

1
> 3m · 5k−1
2
> 3m · 5k−2
≥ 3m+k−2
≥ 3m+v3 (k)
= 3n ,

which is a contradiction.
2. m = 1. Then x + y = 3, so x = 1, y = 2 (or x = 2, y = 1). Thus 31+v3 (k) = 1 + 2k .
But note that 3v3 (k) | k so 3v3 (k) ≤ k. Thus

1 + 2k = 3v3 (k)+1 = 3 · |3v{z

3 (k) k
} ≤ 3k =⇒ 2 + 1 ≤ 3k.
≤k

And one can check that the only odd value of k > 1 that satisfies the above
inequality is k = 3. So (x, y, n, k) = (1, 2, 2, 3), (2, 1, 2, 3) in this case.
Thus, the final answer is n = 2.
Problem 5.6.11 (Balkan 1993). Let p be a prime number and m > 1 be a positive
integer. Show that if for some positive integers x > 1, y > 1 we have
 m
xp + y p x+y
= ,
2 2
then m = p.
CHAPTER 5. SPECIAL TOPICS 307

Solution. One can prove by induction on p that

 p
xp + y p x+y
≥
2 2

for all positive integers p. Now since

 m
xp + y p x+y
= ,
2 2

we should have m ≥ p. Let d = gcd(x, y), so there exist positive integers x1 and y1
with gcd(x1 , y1 ) = 1 such that x = dx1 , y = dy1 , and

2m−1 (xp1 + y1p ) = dm−p (x1 + y1 )m .

There are two cases:

1. Assume that p is odd. Take any prime divisor q of x1 + y1 and let v = vq (x1 + y1 ).
If q is odd, we see that

vq (xp1 + y1p ) = v + vq (p) and vq (dm−p (x1 + y1 )m ) ≥ mv

(because q may also be a factor of d). Thus m ≤ 2 and p ≤ 2, giving an immediate

contradiction. If q = 2, then m − 1 + v ≥ mv, so v ≤ 1 and x1 + y1 = 2, i.e.,
x = y, which immediately implies m = p.

2. Assume that p = 2. We notice that for x + y ≥ 4 we have

 2  3
x2 + y 2 x+y x+y
<2 ≤ ,
2 2 2

so m = 2. It remains to check that the remaining cases (x, y) = (1, 2), (2, 1) are
impossible.

Problem 5.6.12. Find all positive integers a, b that are greater than 1 and satisfy

ba |ab − 1.

Solution. Let p be the least prime divisor of b. Let m be the least positive integer for
which p|am − 1. Then m|b and m | p − 1, so any prime divisor of m divides b and is
less than p. Thus, not to run into a contradiction, we must have m = 1. Now, if p is
odd, we have avp (b) ≤ vp (a − 1) + vp (b), so

a − 1 ≤ (a − 1)vp (b) ≤ vp (a − 1),

which is impossible. Thus p = 2, b is even, a is odd, and

av2 (b) ≤ v2 (a − 1) + v2 (a + 1) + v2 (b) − 1

308 5.7. ZSIGMONDY’S THEOREM

whence

a ≤ (a − 1)v2 (b) + 1 ≤ v2 (a − 1) + v2 (a + 1)

which is possible only if a = 3 and v2 (b) = 1. Put b = 2B with odd B and rewrite the
condition as 23 B 3 | 32B − 1. Let q be the least prime divisor of B (now, surely, odd).
Let n be the least positive integer such that q | 3n −1. Then n | 2B and n | q −1 whence
n must be 1 or 2 (or B has a smaller prime divisor), so q | 3 − 1 = 2 or q | 32 − 1 = 8,
which is impossible. Thus B = 1 and b = 2.
Problem 5.6.13. Find all positive integer solutions of the equation x2009 + y 2009 = 7z
Solution. Factor 2009. We have 2009 = 72 · 41. Since x + y | x2009 + y 2009 and x + y > 1,
we must have 7 | x + y. Removing the highest possible power of 7 from x, y, we get

v7 (x2009 + y 2009 ) = v7 (x + y) + v7 (2009) = v7 (x + y) + 2,

so x2009 + y 2009 = 49 · k · (x + y) where 7 ∤ k. But we have x2009 + y 2009 = 7z , which means

the only prime factor of x2009 +y 2009 is 7, so k = 1. Thus x2009 +y 2009 = 49(x+y). But in
this equation the left hand side is much larger than the right hand one if max(x, y) > 1,
and, clearly, (x, y) = (1, 1) is not a solution. Thus the given equation does not have
any solutions in the set of positive integers.

5.7 Zsigmondy’s Theorem

Zsigmondy’s theorem is one of the tactics that can easily tackle a good number of hard
problems in recent years. This is indeed a mighty theorem to be used in an olympiad.
But it seems everyone has accepted it as a tool for solving problems because the proof of
its theorem is quite elementary (still hard). At the IMO, often some problems appear
that can be solved very easily using some heavy theorems. But those theorems are
usually not accepted by everyone for not being as elementary as needed. But in this
case, everyone seems to like this theorem. The first author was really keen to provide
the proof of this theorem along with some other similar theorems . But that really is
beyond the scope of our book. Since it is a well established theorem, for now, we will
just assume it is true. Rather we focus on how to implement this theorem in solving
problems.
Definition 5.7.1 (Primitive Divisor). For a sequence of integers a1 , a2 , . . . , an , . . . a
prime number p is a primitive divisor of an if p divides an but p doesn’t divide ak for
any k < n. R. D. Carmichael called such a prime an intrinsic divisor.
Example. Consider the sequence ak = 2k − 1. a1 = 1, a2 = 3, a3 = 7, a4 = 15. Note
that, a3 has primitive divisor 7 and a4 has the primitive divisor 5.
Theorem 5.7.2 (Zsigmondy’s Theorem, 1882). Let a, b be co-prime integers and
n ≥ 1 be an integer.
1
according to the first author, there may be a difference in opinion
2
such as Carmichael Theorem
CHAPTER 5. SPECIAL TOPICS 309

• an − bn has a primitive divisor except when:

(a) a − b = 1, n = 1.
(b) a = 2, b = 1 and n = 6.
(c) a + b is a power of 2 and n = 2.
• an + bn has a primitive divisor for n ≥ 2 except for the case 23 + 13 .

This theorem can be extended even further.

Theorem 5.7.3 (First Extension). Let p be a primitive divisor of an + bn . Then p
does not divide ak + bk for n + 1 ≤ k ≤ 2n.
Proof. Since n + 1 ≤ k ≤ 2n, for k = n + l, we get 1 ≤ l ≤ n. p does not divide any of
a or b. For the sake of contradiction, let’s assume, p divides ak + bk .
p|al (an + bn ) = ak + al bn , and
p|bl (an + bn ) = an bl + bk .
Therefore
p|ak + al bn + an bl + bk .
We already know p|ak + bk , so if n = l + m (since l ≤ n), then
p|al bn + an bl = al bl (am + bm ).
Since p⊥ab, we have p⊥al bl . So p|am + bm where m < n, which is a contradiction.
In a similar fashion, we can prove the following theorem.
Theorem 5.7.4 (Second Extension). Let p be a primitive divisor of an + bn . Then
n
p does not divide ak − bk for 1 ≤ k < .
2
In this section, we will see some demonstration of its power in solving problems and
then develop a theorem that generalizes a problem from the IMO Shortlist. The main
idea is to find some contradictions using the fact that an − bn will have a prime factor
that won’t divide something else.
Problem 5.7.5 (Japanese Math Olympiad, 2011). Find all 5−tuples (a, n, x, y, z)
of positive integers so that
an − 1 = (ax − 1)(ay − 1)(az − 1).
Solution. If a, n ≥ 3 and n > x, y, z, we already know from the theorem that an − 1
has a prime divisor that none of ax − 1, ay − 1 or az − 1 has. Therefore, two sides can
never be equal. We are left with cases n ≤ 3. Note that, n ∈ / {x, y, z}. But ax − 1
divides an − 1, so x divides n. Thus, n > x, y, z and hence a, n ≤ 3, like we said before.
Now, either a < 3 or n < 3. If a < 3, then a = 2 and
2n − 1 = (2x − 1)(2y − 1)(2z − 1).
Here, the only exception is n = 6 and 26 − 1 = 63 = 3 · 3 · 7 = (22 − 1)(22 − 1)(23 − 1).
So, {x, y, z} = {2, 2, 3}. Only n < 3 is left to deal with and it is easy to check that
there are no solutions in this case.
310 5.7. ZSIGMONDY’S THEOREM

Problem 5.7.6 (Polish Math Olympiad). If p and q are distinct odd primes, show
that 2pq − 1 has at least three distinct prime divisors.
Solution. Without loss of generality, consider that 2 < q < p < pq. Then 2q − 1 has at
least one prime factor, 2p − 1 has a prime factor that is not in 2q − 1 and 2pq − 1 has a
prime factor that is not in any of 2p − 1 or 2q − 1. Since 2p − 1|2pq − 1 and 2q − 1|2pq − 1,
we have three distinct prime factors.
Problem 5.7.7 (Hungary 2000, Problem 1). Find all 4−tuples (a, b, p, n) of posi-
tive integers with p a prime number such that
a3 + b 3 = p n .
Solution. To apply the theorem, first we need to make a and b co-prime. If q is a prime
divisor of (a, b) = g, then q|p. Therefore, g = pr for some r. Let, a = pr x, b = pr y with
x⊥y. Then,
x3 + y 3 = pn−3r .
Assume that m = n − 3r. Since the power is three, we need to consider the exceptional
case first. The case x = 2 and y = 1 when p = 3 and n − 3r = 2 produces infinitely
many solutions. Otherwise, x3 + y 3 has a prime divisor that does not divide x + y.
Obviously x + y is divisible by p since x + y > 1. This is a contradiction. Therefore,
the only families of solutions are
(a, b, p, n) = (2 · 3r , 3r , 3, 3r + 2) and (3r , 2 · 3r , 3, 3r + 2),
for any positive integer r.
The next problem is from the IMO Shortlist.
Problem 5.7.8 (IMO Shortlist 2002, Problem 4). If p1 , p2 , . . . , pn are distinct
primes greater than 3, prove that, 2p1 p2 ...pn + 1 has at least 4n divisors.
Here, we will prove a much more generalized form of this problem. You can certainly
see how much the problem can be improved with this theorem.
Theorem 5.7.9. If p1 , p2 , . . . , pn are all primes greater than 3, then 2p1 p2 ...pn + 1 has at
n
least 22 divisors.
In order to prove this theorem, let’s first prove the following lemma.
Lemma 5.7.10. Let N = 2p1 ···pn + 1 where pi > 3 is a prime. Then N has at least 2n
distinct prime divisors.
Proof. The number M = p1 p2 . . . pn has (1 + 1)(1 + 1) . . . (1 + 1) = 2n divisors. Say
| {z }
n times
the divisors are
1 = d 1 < d 2 < . . . < d 2n = p 1 · · · p n .
Then first 2d1 + 1 has the prime divisor 3. 2d2 + 1 has a divisor that is not 3. Generally,
each di > di−1 gives us a new primitive divisor that was not in 2di−1 + 1. Therefore, we
have at least 2n distinct prime divisors.
CHAPTER 5. SPECIAL TOPICS 311

Now we prove the theorem.

Proof. Let’s assume that these 2n primes are q1 , q2 , . . . , q2n . Then,

N = q1 q2 . . . q2n K,

for some integer K ≥ 1. Thus, every divisor of D = q1 q2 . . . q2n is a divisor of N and

n
so, N has at least 22 divisors.

This theorem can be generalized even more.

Theorem 5.7.11. Let a, b, n be positive integers with a⊥b.

i. If 3 ∤ n, an + bn has at least 2τ (n) divisors. That is,

τ (an + bn ) ≥ 2τ (n) .

ii. If n is odd and a − b > 1, an − bn has at least 2τ (n) divisors. That is,

τ (an − bn ) ≥ 2τ (n) .

Problem 5.7.12 (Romanian TST 1994). Prove that the sequence an = 3n − 2n

contains no three numbers in geometric progression.

Solution. Assume to the contrary, a2n = am ak , since they are in geometric progression.
So

(3n − 2n )2 = (3k − 2k )(3m − 2m ).

Since k, m, n are distinct, we must have k < n < m. If not, we can not have n < k < m
or n > m > k because that would make one side larger. But due to the fact m > n, we
get that, 3m − 2m has a prime divisor that does not divide 3n − 2n .

The next problem is taken from the IMO Shortlist.

Problem 5.7.13 (IMO Shortlist 2000). Find all positive integers a, m, and n such
that

am + 1|(a + 1)n .

Solution. Note that (a, m, n) = (1, m, n) is a solution for all m, n. (a, m, n) = (2, 3, n)
is a solution for n > 1. If m 6= 3 and a, m ≥ 2, then am + 1 has a prime factor that is
not a prime factor of a + 1. Therefore, in such cases there are no solutions.
312 5.8. HOW TO USE MATRICES?

5.8 How to Use Matrices?

Matrices come to the rescue as a useful tool in many problems (for example, in Dio-
phantine equations or representation problems), and contribute a much better and more
elegant solution. But since we are not doing a linear algebra course, we will define and
discuss only what’s required here.

Definition 5.8.1 (Matrix). A matrix is a rectangular array which can consist of

numbers, variables, or anything. Like a grid, it can have m horizontal rows and n
vertical columns. So, there are mn cells in a matrix. Then the matrix is of the size
m × n. There are some common notations for denoting matrix. But we will use the
usual one:
 
a11 a12 . . . a1n
 a21 a22 . . . a2n 
 
Am×n =  . .. . ..  .
 .. . . . . 
am1 am2 . . . amn

Here, aij are the entries of the m × n matrix A. Notice how the index of each entry is
written. The entry aij belongs to the ith row and j th column of the matrix.

Example. Consider the matrices

 
  1 2
1 0 0  
3 4 −1 0 1 −1
A = 0 1 0  , B=
5
, C= .
6 0 0 0 1
0 0 1
7 8

A is 3 × 3, B is 4 × 2, and C is 2 × 4.

Definition 5.8.2 (Square Matrix). A matrix is square if the number of rows is equal
to the number of columns, i.e., m = n if the size is m × n. The example above is a
square matrix as well.

Definition 5.8.3 (Matrix Diagonals). Let Am×n be a matrix with entries aij . The
main diagonal of A is the collection of entries aij where i = j.

Definition 5.8.4 (Identity Matrix). An square matrix An×n is called identity matrix
and denoted by In if entries of its main diagonal equal to one, and all other entries are
zero.

Example. Matrix A in the previous example is a square matrix however B and C are
not. Moreover, I is an identity matrix of dimension 3, that is, A = I3 . The main
diagonal of matrix B is formed by the entries b11 = 1 and b22 = 4.
Matrices might seem a bit confusing. You might wonder why someone would create
matrices, what’s wrong with normal numbers? Before giving you an application of
where matrices are used, you should know how to do some basic matrix operations.
CHAPTER 5. SPECIAL TOPICS 313

Definition 5.8.5 (Matrix Addition). The matrix addition is the operation of adding
two matrices by adding the corresponding entries together. If A and B are m × n
matrices, then
   
a11 a12 . . . a1n b11 b12 . . . b1n
 a21 a22 . . . a2n   b21 b22 . . . b2n 
   
Am×n + Bm×n =  . .. . ..  +  .. .. . . .. 
 .. . . . .   . . . . 
am1 am2 . . . amn bm1 bm2 . . . bmn

 
a11 + b11 a12 + b12 ... a1n + b1n
 a21 + b21 a22 + b22 ... a2n + b2n 
 
= .. .. .. .. .
 . . . . 
am1 + bm1 am2 + bm2 . . . amn + bmn

Definition 5.8.6 (Matrix Multiplication). Let A be an m × n matrix and let B be

an n × P . The multiplication of A and B is an m × p matrix C, such that
   
a11 a12 . . . a1n b11 b12 . . . b1p
 a21 a22 . . . a2n   b21 b22 . . . b2p 
   
Am×n × Bn×p =  . .. . ..  ·  .. .. . . .. 
 .. . . . .   . . . . 
am1 am2 . . . amn bn1 bn2 . . . bnp

 
c11 c12 . . . c1p
 c21 c22 . . . c2p 
 
= . .. .. ..  ,
 .. . . . 
cm1 cm2 . . . cmp
n
X
where cij = aik bkj . We denote this by AB = C.
k=1

Note. The product AB is defined only if the number of columns in A equals the number
of rows in B.

Addition of matrices is easily done by adding corresponding entries. However, the

product of two matrices may seem difficult to understand. We will clarify it with an
example.
Example. Let
 
 
 1 1 5  3 1 0 2
A =  2 3 1 ,
 
B=
 5 1 0 1 .

 4 6 1 
  4 0 1 1
2 1 3
314 5.8. HOW TO USE MATRICES?

A is 4 × 3 and B is 3 × 4, so we can multiply them together and the result is a 4 × 4

matrix. Let’s start calculating the product of A and B. Let AB = C. We start by
finding c11 . Note that
   
1 1 5   c 11 c 12 c 13 c 14
 2 3 1   3 1 0 2  
 
 · 5 1 0 1 = c21 c22 c23 c24  .
 4 6 1   c c c c 
   31 32 33 34 
4 0 1 1 c41 c42 c43 c44
2 1 3

From the definition, the entry c11 of C is calculated by multiplying the corresponding
entries of the first row of A and the first column of B (and you can now see why number
of columns of A must be equal to the number of rows of B). That is,
c11 = 1 · 3 + 1 × 5 + 5 × 4 = 28.
In general, the entry cij is calculated by multiplying the corresponding entries of ith
row of A and j th column of B. Do the product yourself and check the result with the
following:
 
28 2 5 8
25 5 1 8 
C= 46 10 1 15 .


23 3 3 8
Note. Let A and B be square matrices of the same dimension. Then both AB and
BA are defined. However, they are not necessarily equal, i.e., matrix multiplication is
not commutative.
Definition 5.8.7 (Matrix Powers). For a square matrix An×n , we define A2 as the
multiplication of A by itself. The definition of all higher powers of A is followed. In
fact, Ak = A · Ak−1 , for any positive integer k. We also assume that A0 = In , where In
is the n−dimensional identity matrix.
Definition 5.8.8 (Matrix Determinant). A determinant is a real number associated
with every square matrix. For a square matrix A, its determinant is denoted by det(A)
or |A|.
For the simplest case when A is 1 × 1 (a single number), the determinant of A equals
A, which is sensible (what else would it be?). The definition above is not the exact
definition of the determinant. We will first explain how to calculate the determinant of
a 2 × 2 matrix and then move to the precise definition of determinants.
Definition 5.8.9. The determinant of a 2 × 2 matrix is the product of entries on its
main diagonal minus the product of the two other entries. That is, if
 
a b
A= ,
c d
then det(A) = ad − bc. This is also shown by
 
a b 
 c d = ad − bc.
 
CHAPTER 5. SPECIAL TOPICS 315
 
1 2 
Example. 
  = 4 − 6 = −2.
3 4
We will now generalize the definition of determinant to n × n matrices. For this,
you need to know what cofactors and minors are first.

Definition 5.8.10 (Minor). Let A be an n × n matrix. The minor for entry aij is
denoted by Mij and is the determinant that results when the ith row and the j th column
of A are deleted.

Example. Let’s find M21 for the matrix

 
1 1 5
A=
 2 3 1 .

4 6 1

The corresponding row and column (which should be  deleted

 in order to calculate the
1 5 
minor) are shown in the matrix. Therefore, M21 =   = 1 − 30 = −29.
6 1

Definition 5.8.11 (Matrix of Minors). Let A be an n × n matrix. The matrix of

minors is an n × n matrix in which each element is the minor for the corresponding
entry of A.

Example. The matrix of minors for matrix A in the previous example is

   
3 − 6 2 − 4 12 − 12 −3 −2 0
M = 1 − 30 1 − 20 6 − 4  = −29 −19 2 .
1 − 15 1 − 10 3 − 2 −14 −9 1

Definition 5.8.12 (Cofactor). The cofactor for any entry of a matrix is either the
minor or the opposite of the minor, depending on where the element is placed in the
original determinant. If the row and column of the entry add up to an even number,
then the cofactor is the same as the minor. If the row and column of the entry add up
to an odd number, then the cofactor is the opposite of the minor.
In other words, if we denote Cij to be the cofactor of the corresponding entry ai ,
then Cij = (−1)i+j Mij .

Example. You are now be able to make sense of the definition of matrix of cofactors.
The matrix of cofactors of matrix A in previous examples is
   
(−1)2 (3 − 6) (−1)3 (2 − 4) (−1)4 (12 − 12) −3 2 0
3 4 5
M = (−1) (1 − 30) (−1) (1 − 20) (−1) (6 − 4)
  =  29 −19 −2 .
(−1)4 (1 − 15) (−1)5 (1 − 10) (−1)6 (3 − 2) −14 9 1

See the difference between the matrix of minors and the matrix of cofactors of A.
Now you are ready to see a formula for determinant. Our method is computing
larger determinants in terms of smaller ones.
316 5.8. HOW TO USE MATRICES?

Definition 5.8.13. Given the n × n matrix A with entries aij , the determinant of A
can be written as the sum of the cofactors of any row or column of A multiplied by
the entries that generated them. In other words, the cofactor expansion along the j th
column gives
n
X
det(A) = a1j C1j + a2j C2j + a3j C3j + · · · + anj Cnj = aij Cij .
i=1

The cofactor expansion along the ith row gives:

n
X
det(A) = ai1 Ci1 + ai2 Ci2 + ai3 Ci3 + · · · + ain Cin = aij Cij .
j=1

Example. Consider the matrix A in the previous examples. If we use the cofactor
expansion along the second column, we get

det(A) = a12 C12 + a22 C22 + a32 C32 = 1 · 2 + 3 · (−19) + 6 · 9 = −1.

Also, if we use the cofactor expansion along the third row, we get

det(A) = a31 C31 + a32 C32 + a33 C33 = 4 · (−14) + 6 · 9 + 1 · 1 = −1.

Note that we used the matrix of cofactors found above for Cij . As you see, the result
of both calculations is the same.
If you carefully track what we explained until now, you see that we used the de-
terminant of 2 × 2 matrices when calculating the matrix of cofactors of A. Then, in
order to calculate det(A), we used some entries of the matrix of cofactors of A. All in
all, we have used the determinant of 2 × 2 matrices when calculating the determinant
of the 3 × 3 matrix A. This process is the same for larger matrices. For example, in
order to find determinant of a 4 × 4 matrix, you need to calculate four 3 × 3 matrices
determinants.
Finding the determinant of large matrices (larger than 4 × 4) is a really boring job
and we do not want you to calculate such determinants. You know the basics and you
can find the determinant of any n × n matrix. It’s just the matter of time it takes to
find it.
The definition of the determinant may seem useless to you, but it actually is im-
possible to find the inverse of a matrix without knowing its determinant. However, we
are not going to introduce inverse matrices. We want to use determinants in a number
theoretical approach. We will only use the formula for determinant of 2 × 2 matrices,
however we included the definition of the determinant so that you can find 3 × 3 (or
even larger) determinants easily.
We state two theorems without proof. If you are interested in seeing a proof, you
can read any book on linear algebra.
Theorem 5.8.14. Let A and B be n × n matrices. The product of the determinant of
A and B equals the determinant of their product, i.e.,

det(A · B) = det(A) det(B).

CHAPTER 5. SPECIAL TOPICS 317

Theorem 5.8.15. If A is a square matrix, then

Am+n = Am · An .

Now let’s see some of its applications.

Problem 5.8.16 (Fibonacci-Brahmagupta Identity). The sum of two squares is

called a bi-square. Prove that product of two bi-squares is also a bi-square.

Problem 5.8.17. Let x and y be two integers. Prove that the product of two number
of the form x2 + dy 2 is of the same form for a certain d.

Solution. We want to solve this problem using matrices. We already know that
 
a b 
 c d = ad − bc, .
 

so we try to represent x2 + dy 2 in the form ad − bc, which is the determinant of some

matrix. This is pretty simple. Assume the matrices
   
x yd u vd
M= , and N = .
−y x −v u

So det(M) = x2 + yd2 and det(N ) = u2 + dv 2 . Now, we multiply them as explained in

Definition ( ) to get
 
xu − dvy dvx + duy
M·N = .
−(vx + uy) xu − dvy

Thus, det(M · N ) = (xu − dvy)2 + d(vx + uy)2 . Therefore,

(x2 + dy 2 )(u2 + dv 2 ) = (xu − dvy)2 + d(vx + uy)2 ,

which is of the same form.

Problem 5.8.18. Prove that the product of two numbers of the form x2 − dy 2 is again
of the same form.

Solution. This is the same as previous one. The only difference is that the matrix
would be
 
x yd
M= .
y x

Problem 5.8.19. Prove that the following equation has infinitely many solutions for
integers a, b, c, d, e, and f :

(a2 + ab + b2 )(c2 + cd + d2 ) = (e2 + ef + f 2 ).

318 5.8. HOW TO USE MATRICES?

Solution. The following identity gives an infinite family of solutions:

(x2 + x + 1)(x2 − x + 1) = x4 + x2 + 1.

But we present a different solution using matrices. In fact, we can prove that for any
quartet (a, b, c, d) there are integers e and f such that

(a2 + ab + b2 )(c2 + cd + d2 ) = (e2 + ef + f 2 ).

Again, we need to choose a suitable matrix to prove our claim. We choose

   
a b c d
A= , and B = .
−b a + b −d c + d

After this, the process is analogous to previous problems.

Note. We could factorize a2 + ab + b2 as (a + ζb)(a + ζ 2 b), where ζ 3 = 1 is the third root
of unity (don’t worry if this is unfamiliar for you, it needs some knowledge in complex
numbers).

5.8.1 Proving Fibonacci Number Identities

The original Fibonacci sequence Fn is defined by F0 = 0, F1 = 1, and Fn+1 = Fn + Fn−1
for n > 1. You are familiar with this sequence as it’s used in so many cases:

0, 1, 1, 2, 3, 5, 8, 13, . . .

We define general Fibonacci numbers Gn by

G0 = a, G1 = b, and Gn = pGn−1 + qGn−2 for n > 1.

The matrix representation for this sequence is

    
p q Gn Gn−1 Gn+1 Gn
= .
1 0 Gn−1 Gn−2 Gn Gn−1

Special cases are:

1. Fibonacci sequence: a = 0, and b = p = q = 1. The nth term is denoted by Fn .

2. Lucas sequence: a = 2, and b = p = q = 1. The nth term is denoted by Ln .

Theorem 5.8.20.
 n−1    
p q G2 G1 Gn+1 Gn
(5.8) = .
1 0 G1 G0 Gn Gn−1

Corollary 5.8.21.
 n  
1 1 Fn+1 Fn
= .
1 0 Fn Fn−1
CHAPTER 5. SPECIAL TOPICS 319

Proof. We can use induction. It’s rather straight-forward.

Theorem 5.8.22.


Gn+1 Gn−1 − G2n = (−1)n−1 q n−1 a2 p + abq − b2 .

Proof. Take determinant of both sides of equation ( ).

Applying the above theorem for Fibonacci and Lucas sequences, we find the follow-
ing corollaries.
Corollary 5.8.23.

Fn+1 Fn−1 − Fn2 = (−1)n .

Corollary 5.8.24.

Ln+1 Ln−1 − L2n = 5 · (−1)n−1 .

Problem 5.8.25. Prove that

(5.9) Fm+n+1 = Fm+1 Fn+1 + Fm Fn .

 
1 1
Solution. Consider I = . Then, I m+n = I m I n . Note that
1 0
     
m Fm+1 Fm n Fn+1 Fn m+n Fm+n+1 Fm+n
I = ,I = , and I = .
Fm Fm−1 Fn Fn−1 Fm+n Fm+n−1
Thus
     
Fm+1 Fm Fn+1 Fn Fm+1 Fn+1 + Fm Fn Fm+1 Fn + Fm Fn−1
· = .
Fm Fm−1 Fn Fn−1 Fm Fn+1 + Fm−1 Fn Fm Fn + Fm−1 Fn−1
We finally find that
   
Fm+1 Fn+1 + Fm Fn Fm+1 Fn + Fm Fn−1 Fm+n+1 Fm+n
= .
Fm Fn+1 + Fm−1 Fn Fm Fn + Fm−1 Fn−1 Fm+n Fm+n−1
Equating the cells of these two matrices, we get

Fm+n+1 = Fm+1 Fn+1 + Fm Fn .

The following corollaries are immediately concluded.

Corollary 5.8.26.

Fmk+n = Fmk+1 Fn + Fmk Fn−1 .

Corollary 5.8.27. Setting m = n, we have

F2n+1 = Fn2 + Fn+1

2
.

We end the discussion here, but hopefully you have a better idea of how useful
matrices can actually be.
320 5.9. A PROOF FOR LAW OF QUADRATIC RECIPROCITY

5.9 A Proof for Law of Quadratic Reciprocity

Law of quadratic reciprocity (theorem ( )) states that for any two different odd
primes p and q, we have
  
p q p−1 q−1
= (−1) 2 · 2 .
q p
Mathematicians have provided many proofs for the law of quadratic reciprocity.
Gauss himself proved this theorem as well. However, we will be showing arguably the
most amazing proof of this theorem, which is due to Eisenstein . Before explaining the
proof, we should prove two lemmas.
Lemma 5.9.1. Let p be a prime and let a be an integer co-prime to p. When the
numbers a, 2a, . . . p−1
2
a are reduced modulo p into the range from − p−1 2
to p−1
2
, the
p−1
reduced values are ±1, ±2, . . . , ± 2 in some order, with each number appearing once
with either a plus sign or a minus sign.

Proof. We should prove that for any k, t ∈ 1, 2, . . . , p−1
2
, the numbers ka and ta are
p−1
different members of the set ±1, ±2, . . . , ± 2 when reduced modulo p into the range
− p−1
2
to p−1
2
. Assume that ka ≡ ta (mod p). Then a(k − t) ≡ 0 (mod p) and since
a⊥p, we get k − t ≡ 0 (mod p). But since k and t are both at most p−1 2
, we should have
k − t = 0 or k = t. On the other hand, if ak ≡ −at (mod p), then k + t ≡ 0 (mod p).
But
p−1 p−1
k+t≤ + = p − 1,
2 2
so it’s impossible to have k + t ≡ 0. This finishes the proof.
The second lemma uses the definition of µ(a, p) which we defined in Gauss’s Criterion
(theorem ( )).
Lemma 5.9.2. Let p be a prime and let a be an odd integer co-prime to p. Then
p−1
2 j
X ka k
≡ µ(a, p) (mod 2).
k=1
p

Proof. For each k ∈ 1, 2, . . . , p−1
2
, we can write ka as
p−1 p−1
ka = pqk + rk , − < rk < .
2 2
Notice that this is different from the normal division (try to see why we can write ka
like that). Now divide both sides by p to get
j ka k jr k 1 rk 1
k
= qk + , − < < .
p p 2 p 2
3
Do not confuse it with Einstein.
CHAPTER 5. SPECIAL TOPICS 321
j k
This means that ka is either qk (when rk > 0) or qk − 1 (when rk < 0). Adding all
j k p
the values of ka
p
, we see that

p−1 p−1
2 j 2
X ka k X
= qk − X,
k=1
p k=1

where X is the number of negative rk s. If you look more closely, you see that X = µ(a, p)
(why?), and so
p−1 p−1
2 j 2
X ka k X
= qk − µ(a, p).
k=1
p k=1

p−1
2
X
We just need to show that qk is an even integer, because then
k=1

p−1 p−1
2 j 2
X ka k X
≡ qk − µ(a, p) ≡ 0 − µ(a, p) ≡ µ(a, p) (mod 2),
k=1
p k=1

which is just what we want. The trick is to write the equation ka = pqk + rk modulo 2.
Since both a and p are odd, ka ≡ k (mod 2) and pqk ≡ qk (mod 2), and so

k ≡ qk + r k (mod 2).

Summing over k, we see that

p−1 p−1 p−1
2
X 2
X 2
X
k≡ qk + rk (mod 2).
k=1 k=1 k=1

From lemma ( ), we see that the numbers r1 , r2 , . . . , r p−1 are equal to the numbers
2
±1, ±2, . . . , ± p−1
2
in some order, with each number appearing once with either a plus
sign or a minus sign. We also know that x ≡ −x (mod 2) for any integer x. So, we can
say that
p−1 p−1 p−1
2
X 2
X 2
X
rk ≡ k (mod 2) =⇒ qk ≡ 0 (mod 2).
k=1 k=1 k=1

The proof is complete.

We are now ready to provide a proof for the law of quadratic reciprocity.
Proof of law of quadratic reciprocity. This proof is based on geometry, and that’s in-
teresting. Consider a triangle in the xy-plane with vertices on (0, 0), (p/2, 0), and
(p/2, q/2).
322 5.9. A PROOF FOR LAW OF QUADRATIC RECIPROCITY
p q 
• • • • • ,
2 2
• • • • • • •

• • • • • • •

• • • • • • •

• • • • • • •

• • • • • • •

• • • • • • •

• • • • • • •
(0, 0) p
,0
2

The number of points with integer coordinates inside this triangle equals
p−1
2 j
X kq k
.
k=1
p

You can easily verify this by using the fact that the hypotenuse of triangle lies on the
line y = pq x, and so the number of points with x = k (where 1 ≤ k ≤ p−1 2
) inside the
j k
triangle equals kq p
(actually, we are counting the points vertically).
Now consider the triangle with vertices on (0, 0), (0, q/2), and (p/2, q/2).
p q 
 • • • • • ,
q 2 2
0,
2 • • • • • • •

• • • • • • •

• • • • • • •

• • • • • • •

• • • • • • •

• • • • • • •

• • • • • • •
(0, 0)
CHAPTER 5. SPECIAL TOPICS 323

We can find the number of points with integer coordinates inside this triangle in a
similar way to the previous one. This time, count the points horizontally and sum up
the number of points with y = 1, y = 2, . . ., and y = p−1
2
. The result is
q−1
2 j
X kp k
.
k=1
q

Now put these two triangles together to form a rectangle with vertices on (0, 0),
(0, q/2), (p/2, 0), and (p/2, q/2).
Let x be number of the points with integer coordinates inside this rectangle. Obvi-
ously, x is equal to the sum of such points in triangles (notice that since p and q are
different, there is no point with integer coordinates on the hypotenuse of triangles). So
p−1 q−1
2 j 2 j
X kq k X kp k
x= + .
k=1
p k=1
q

According to lemma ( ), it follows that

(5.10) x ≡ µ(q, p) + µ(p, q) (mod 2).
Let’s count x in another way.
p q 
 • • • • • ,
q 2 2
0,
2 • • • • • • •

• • • • • • •

• • • • • • •

• • • • • • •

• • • • • • •

• • • • • • •

• • • • • • •
(0, 0) p
,0
2

Clearly, number of points with integer coordinates inside this rectangle is

jpk jq k p − 1 q − 1
(5.11) x= · = · .
2 2 2 2
Combining equations ( ) and ( ),
p−1 q−1
µ(q, p) + µ(p, q) ≡ · (mod 2).
2 2
324 5.10. DARIJ-WOLSTENHOLME THEOREM

Now apply Gauss’s criterion to finish the proof:

  
p q
= (−1)µ(p,q) · (−1)µ(q,p)
q p
= (−1)µ(p,q)+µ(q,p)
p−1 q−1
· 2
= (−1) 2 .

5.10 Darij-Wolstenholme Theorem

The following theorem is a generalization of Wolstenholme’s theorem. It was proposed
and proved by Darij Grinberg on the Art of Problem Solving website. Before stating
the theorem, we need to define vp (x) for a rational number x.
Recall section ( ) where we defined vp (x) for x being an integer as the greatest
power of p which divides x. Now, since we are working with fractions, we need to
generalize this concept to include rational numbers.

Definition 5.10.1. Let p be a prime and let x = ab 6= 0 be a rational number reduced

to lowest terms. Define the p-adic evaluation of x as vp (x) = vp (a) − vp (b), where vp (n)
is the notation defined in definition ( ).
9

34


Example. v3 16 = 2, and v5 25 = −2.

Note. We can easily check the sign of vp (x) for any rational number x = ab . If p|a,
then vp (x) > 0. If p divides none of a and b, then vp (x) = 0. And if p|b, then vp (x) < 0.
Also, vp (xy) = vp (x) + vp (y) and vp (x + y) ≥ min (vp (x) , vp (y)) for all rationals x
and y.

We can now generalize the concept of congruency to include rational numbers.

Definition 5.10.2. If x and y are two rational numbers such that vp (x) ≥ 0 and
vp (y) ≥ 0, then we say that x ≡ y (mod p) if and only if vp (x − y) > 0.

The following problem gives you a good sight of the above notation.

Problem 5.10.3. Let p ≥ 3 be a prime. Prove that p|2p−2 + 3p−2 + 6p−2 − 1.

1
Solution. Let a⊥p be an integer. We can write ap−2 ≡ a
(mod p) because
   
p 1 ap−1 − 1
vp a − = vp >0
a a

by Fermat’s little theorem. Now

1 1 1
2p−2 + 3p−2 + 6p−2 − 1 ≡ + + −1≡0 (mod p).
2 3 6
CHAPTER 5. SPECIAL TOPICS 325

Theorem 5.10.4 (Darij-Wolstenholme Theorem). Let p > 3 be a prime and let u

be a non-negative and odd integer such that p ≥ u + 3. Then
p−1
!
X 1
vp ≥ 2.
k=1
ku

The idea of the proof is similar to the proof of Wolstenholme’s theorem. We need
to prove a lemma first.

Lemma 5.10.5. Let p be a prime and let n be an integer such that 1 ≤ n ≤ p − 2.

Then
p−1
X
kn ≡ 0 (mod p).
k=1

Proof. There exists an integer a co-prime to p such that p ∤ an − 1. The set A =

{0, 1n , 2n , . . . , (p − 1)n } forms a compete residue system modulo p (why?). Proposition
( ) says that the set B = {0, an , (2a)n , . . . , ((p−1)a)n } also forms a complete residue
system modulo p. Therefore, the sum of elements of both sets are equivalent modulo
p. So
p−1 p−1 p−1
X X X
n n n
k ≡ (a · k) ≡ a kn (mod p).
k=1 k=1 k=1

This means that

p−1
k

X
a −1 kn ≡ 0 (mod p),
k=1

and since p ∤ an − 1, we should have

p−1
X
kn ≡ 0 (mod p).
k=1

If the proof seemed confusing to you, here is a potentially better version. Consider a
primitive root g of p (we already know there is one from modular arithmetic chapter).
Then we also know that {1, 2, . . . , p−1} can be generated by g (the set {1, g, g 2 , . . . , g p−2 }).
So,

n
1n + 2n + · · · + (p − 1)n = 1n + g n + g 2n + · · · + g p−2
(g n )p−1 − 1
=
gn − 1
g (p−1)n − 1
=
gn − 1

From Fermat’s little theorem, g p−1 ≡ 1 (mod p), so the conclusion follows.
326 5.10. DARIJ-WOLSTENHOLME THEOREM

Proof of Darij-Wolstenholme Theorem. The idea is to use the trick explained in lemma
( ). That is, we write the given sum as a sum of terms of the form k1u + (p−k)
1
u . We

have
p−1 p−1 p−1
X 1 X 1 X 1
2 = +
k=1
k u
k=1
k u
k=1
(p − k)u
p−1  
X 1 1
= +
k=1
k u (p − k)u
p−1 u
X k + (p − k)u
=
k=1
k u (p − k)u
p−1
X k u + (pu − upu−1 k + · · · + upk u−1 − k u )
=
k=1
k u (p − k)u
p−1
X pu − upu−1 k + · · · + upk u−1
=
k=1
k u (p − k)u
p−1
X pu−1 − upu−2 k + · · · + uk u−1
=p· .
k=1
k u (p − k)u

We have used the fact that u is an odd integer to expand (p − k)u in above lines. Now
since p > 3 is an odd prime, vp (2) = 0 and therefore
p−1
! p−1
! p−1
!
X 1 X 1 X 1
vp 2 u
= vp (2) + vp u
= vp
k=1
k k=1
k k=1
ku
p−1 u−1
!
X p − upu−2 k + · · · + uk u−1
= vp p ·
k=1
k u (p − k)u
p−1 u−1
!
X p − upu−2 k + · · · + uk u−1
= vp (p) + vp
k=1
k u (p − k)u
p−1 u−1
!
X p − upu−2 k + · · · + uk u−1
= 1 + vp u (p − k)u
.
k=1
k
Pp−1 1


So instead of showing vp 2 k=1 ku ≥ 2, it is enough to show that
p−1 u−1
!
X p − upu−2 k + · · · + uk u−1
vp ≥ 1,
k=1
k u (p − k)u

which is equivalent to showing that

p−1 u−1
X p − upu−2 k + · · · + uk u−1
≡0 (mod p).
k=1
k u (p − k)u
CHAPTER 5. SPECIAL TOPICS 327

Since pu−1 − upu−2 k + · · · + uk u−1 ≡ uk u−1 (mod p) and k u (p − k)u ≡ k u (−k)u ≡

(−1)u k 2u (mod p), we should prove that

p−1 p−1
X uk u−1 u X −u−1
≡ k ≡0 (mod p).
k=1
(−1)u k 2u (−1)u k=1

From Fermat’s little theorem, we have k p−1 ≡ 1 (mod p) for every k such that 1 ≤ k ≤
p − 1. So k −u−1 ≡ k −u−1 k p−1 = k p−u−2 (mod p) and we must prove that

p−1
u X p−u−2
k ≡0 (mod p),
(−1)u k=1

which follows directly from lemma ( ) because 1 ≤ p − u − 2 ≤ p − 2. The proof

is complete.

Problem 5.10.6. Let p > 3 be a prime. Prove that

⌊2p/3⌋
X (−1)i−1
≡0 (mod p).
i=1
i

Solution. First, let us prove that

(5.12) p − ⌊⌊2p/3⌋ /2⌋ = ⌊2p/3⌋ + 1,

where ⌊x⌋ is the largest integer not greater than x for any real x. Since p > 3, either
p ≡ 1 (mod 3) or p ≡ 2 (mod 3). Consider both cases:
p−1
• If p ≡ 1 (mod 3), then 3
is an integer and

p−1 2p p−1
2· ≤ <2· + 1,
3 3 3
which means
 
2p p−1
=2· ,
3 3
     
2p p−1 p−1
/2 = 2· /2 = .
3 3 3

Finally,
    
2p p−1 p−1 2p
p− /2 = p − =2· +1= + 1,
3 3 3 3

as desired.
328 5.10. DARIJ-WOLSTENHOLME THEOREM
p−2
• If p ≡ 2 (mod 3), then is an integer and
3
 
p−2 2p p−2
2· +1≤ < 2· + 1 + 1.
3 3 3
This gives

2p p−2
=2· + 1,
3 3
       
2p p−2 p−2 1 p−2
/2 = 2· + 1 /2 = + = .
3 3 3 2 3
And finally
      
2p p−2 p−2 2p
p− /2 = p − = 2· +1 +1= + 1.
3 3 3 3

The proof of equation ( ) is finished. We will now prove the problem. Obviously,
⌊2p/3⌋
X (−1)i−1 X 1 X −1
= +
i=1
i i i
1≤i≤⌊2p/3⌋; 1≤i≤⌊2p/3⌋;
i is odd i is even
X 1 X 1
= −
i i
1≤i≤⌊2p/3⌋; 1≤i≤⌊2p/3⌋;
i is odd i is even
 
 X 1 X 1 X 1
= + −2
i i i
1≤i≤⌊2p/3⌋; 1≤i≤⌊2p/3⌋; 1≤i≤⌊2p/3⌋;
i is odd i is even i is even
⌊2p/3⌋
X 1 X 1
= −2
i=1
i i
1≤i≤⌊2p/3⌋;
i is even
⌊2p/3⌋⌊⌊2p/3⌋/2⌋
X 1 X 1
= −2
i=1
i j=1
2j
⌊2p/3⌋
X 1 ⌊⌊2p/3⌋/2⌋
X 1
= +
i=1
i j=1
−j
⌊2p/3⌋
X 1 ⌊⌊2p/3⌋/2⌋
X 1
≡ + (mod p).
i=1
i j=1
p−j

In the second sum in the last line of above equations, we have used the fact that
−j ≡ p − j (mod p). Replacing j by p − i in the second sum, we have
⌊2p/3⌋
X (−1)i−1 ⌊2p/3⌋
X 1 p−1
X 1
≡ + .
i=1
i i=1
i i
i=p−⌊⌊2p/3⌋/2⌋
CHAPTER 5. SPECIAL TOPICS 329

Using ( ), we can write i = p − ⌊⌊2p/3⌋ /2⌋ = ⌊2p/3⌋ + 1 and so by Wolstenholme’s

theorem,
⌊2p/3⌋
X (−1)i−1 ⌊2p/3⌋
X 1 p−1
X 1
≡ +
i=1
i i=1
i i
i=⌊2p/3⌋+1
p−1
X 1
≡ (mod p)
i=1
i
≡0 (mod p).

This finishes the proof of the problem.

5.11 Generalization of Wilson’s and Lucas’

Theorem
Wilson’s theorem says that (p − 1)! ≡ −1 (mod p) for all primes p. Clearly, for any
integer n larger than p, we have n! ≡ 0 (mod p). Now, if we remove the multiples of p
from n! and then calculate the result modulo p, what would it be? We will state this
as a generalization for Wilson’s theorem. But first, some definitions and lemmas.
Definition 5.11.1. Let n be a positive integer and p a prime number. The p-reduced
factorial of n is the product of all positive integers less than or equal to n which are
not divisible by p. We denote this by (n!)p .
Example. The 5-reduced factorial of 10, (10!)5 , is

(10!)5 = 9 × 8 × 7 × 6 × 4 × 3 × 2 × 1 = 72, 576.

Theorem 5.11.2. Let p be a prime number and let (nk nk−1 . . . n1 n0 )p be a positive
integer. Then

(n!)p ≡ (−1)[ p ] · n0 !
n
(mod p).

Proof. The numbers not divisible by p among 1, 2, . . . , n are

1 2 ··· n0 ··· p−1
p+1 p+2 ··· p + n0 ··· 2p − 1
.. .. .. .. .. ..
. . . . . .
([ np ] − 1)p + 1 ([ np ] − 1)p + 2 · · · ([ np ]p − 1)p + n0 · · · [ np ]p − 1
[ np ]p + 1 [ np ]p + 2 ··· [ np ]p + n0

Product of these numbers, (n!)p is

[n
p
]−1 !
Y




(kp + 1) · (kp + 2) · · · (kp + p − 1) · [ np ]p + 1 [ np ]p + 2 . . . [ np ]p + n0 ,
k=0
330 5.11. GENERALIZATION OF WILSON’S AND LUCAS’ THEOREM

which is equal to
[n
p
]−1 !
Y




= 1 · 2 · · · (p − 1) · [ np ]p + 1 [ np ]p + 2 . . . [ np ]p + n0
k=0
[n
p
]−1 !
Y

≡ −1 · (1 · 2 · · · n0 )
k=0
[n ]
≡ (−1) p n0 ! (mod p).

Proposition 5.11.3. Let p ≥ 3 be a prime and n be a positive integer. Then

(pn !)p ≡ −1 (mod pn ).

Proof. This is exactly the same as the proof of Wilson’s theorem. All numbers in the
product (pn !)p have a multiplicative inverse modulo pn . If the inverse of a number a
among these numbers is b 6= a, then ab ≡ 1 (mod pn ) and we can remove a and b from
the product (pn !)p . Our only concern is when the inverse of a equals a itself. But if
that’s the case, we have

a2 ≡ 1 (mod pn ) =⇒ pn |(a − 1)(a + 1).

But since (a − 1, a + 1) = 2, we should have a ≡ ±1 (mod pn ), which means a is either

1 or pn − 1. All in all, we see that the product of all numbers in (pn !)p except pn − 1
equals 1 modulo pn , and if we multiply this number by pn − 1, the result will be −1
modulo pn .
Problem 5.11.4. Prove that (2n !)2 ≡ 1 (mod 2n ).
We are ready to prove the following theorem.
Theorem 5.11.5 (Generalization of Wilson’s Theorem). Let p be a prime number
and let (nk nk−1 . . . n1 n0 )p be the representation of a positive integer n in base p. Then
n!
(5.13) ≡ (−1)vp (n!) n0 !n1 ! . . . nk ! (mod p).
pvp (n!)
Proof. According to theorem ( ), one can write
 
⌊n/p⌋ n
n! = (n!)p · p !
p
 
⌊n/p⌋ ⌊n/p⌋ n
≡ (−1) n0 ! · p !.
p

Now write (⌊ np ⌋)! in the same way and continue this process. The result is concluded.

Note. If you are interested, you can find a (different) generalization of Wilson’s theorem
in Problem .
CHAPTER 5. SPECIAL TOPICS 331

Theorem 5.11.6 (Generalization of Lucas’ Theorem). Let p be a prime number

and m, n, and r be non-negative integers such that r = m − n and

m = mk pk + mk−1 pk−1 + · · · + m1 p + m0 ,
n = kk pk + nk−1 pk−1 + · · · + n1 p + n0 ,
r = rk pk + rk−1 pk−1 + · · · + r1 p + r0 .
m



Also, let ℓ = vp n
. Then
   m !  m !   m ! 
1 m ℓ 0 1 d
≡ (−1) ... (mod p).
pℓ n n0 !r0 ! n1 !r1 ! nd !rd !

Proof. Note that

   
m m!
ℓ = vp = vp = vp (n!) − vp (n!) − vp (r!)
n n!r!
k   X k   k  
X m n X r
= i
− i
−
i=1
p i=1
p i=1
pi
k      
X m n r
= i
− i − i .
i=1
p p p

Just like the proof of theorem ( ), we can write

 
m
  !
m (m!)p p⌊m/p⌋ p
= · ⌊n/p⌋ ⌊r/p⌋ ·   .
n (n!)p (r!)p p ·p n r
!· !
p p

Use induction and generalization of Wilson’s theorem to finish the proof.

5.12 Inverse of Euler’s Totient Function

For a given positive integer n, we can find ϕ(n) after factorizing n. What about the
reverse problem? That is, given ϕ(n), can you find n? A more interesting question is
whether this solution n is unique or there are other solutions. We can answer the latter
question pretty quickly using an example: ϕ(4) = 2 and ϕ(6) = 2. In other words, ϕ is
not a one to one function. Now, another question normally arises here:

Problem 5.12.1. Is there any n ∈ N such that ϕ(x) = n has a unique solution for x?

There are good results on this topic. It has also been studied how to find such x,
and the upper or lower bounds of x. Here we will discuss some of the results, which fits
into our book.
332 5.12. INVERSE OF EULER’S TOTIENT FUNCTION

Definition 5.12.2 (Inverse Phi). Let n be a positive integer. Assume that ϕ−1 (n)
is the set of all possible values of x ∈ N such that ϕ(x) = n. In other words,
ϕ−1 (n) = {x : ϕ(x) = n}.
We call ϕ−1 (n) the inverse of Euler’s totient function, or simply the inverse of phi
function. Moreover, for ever positive integer x, we define N (x) to be the number of
positive integers y such that ϕ(x) = ϕ(y).
Carmichael stated in [ ] that the cardinality (number of elements) of ϕ−1 (n) is
always greater than 1 but due to his proof being inadequate, this is a conjecture now:
Conjecture 5.1 (Carmichael’s Totient Conjecture). For a positive integer n, the
number of solutions to ϕ(x) = n is either 0 or at least 2.
After this statement, quite a lot of number theorists worked on it. There has been
no proof of the theorem to our knowledge, though there are some nice results on it.
And it is indeed a very interesting topic to work on. Even though it is a conjecture,
everything points this to be true. For example, Klee pointed out in [ ] that if N (x) = 1
then x and ϕ(x) are both larger than 10400 . Carmichael originally proved that x > 1037
must be true. Let’s start investigating N (x).
Theorem 5.12.3. Let x be a positive integer. If N (x) = 1, then x is divisible by 4.
Proof. For n > 2, ϕ(n) is always even. If x is odd, then 2⊥x so ϕ(2x) = ϕ(x) so y = 2x
is a solution, so contradiction. Again, if x = 2t with t odd then ϕ(x) = ϕ(t) by same
argument. Thus, x is divisible by 4.
The following theorem is due to Carmichael.
Theorem 5.12.4. Let x be a positive integer and let p = 2k + 1 be a prime divisor of
x, where k is some natural number. If N (x) = 1, then p2 |x.
Proof. To the contrary, assume that x = 2e ps for some positive integers e and s with
s⊥2p. Then,
ϕ(x) = ϕ(2e )ϕ(p)ϕ(s)
= 2e−1 2k ϕ(s)
= ϕ(2k+e )ϕ(s)
= ϕ(2k+e s).
Thus, y = 2k+e s 6= x satisfies the condition, so we must have p|s and hence p2 |x.
Here is a very nice result that provides us with a sufficient condition for N (x) = 1
to happen.. The result is due to C. Pomerance [ ].
Theorem 5.12.5 (Carl Pomerance). Let x be a positive integer. Suppose that the
following property holds for every prime p:
p − 1|ϕ(x) implies p2 |x.
Then N (x) = 1. That is, if ϕ(y) = ϕ(x) for some positive integer y, then y = x.
CHAPTER 5. SPECIAL TOPICS 333

Proof. For every positive integer

Q n, define S(n) to be the set of prime divisors of n. If
the prime factorization of n is ri=1 pei i , then
r
Y
ϕ(n) = piei −1 (pi − 1).
i=1

According to our assumption, x is a positive integer such that if p − 1|ϕ(x) then p2 |x.
We are required to prove that under this assumption, if ϕ(x) = ϕ(y) then x = y must
hold. If p ∈ S(y) then p − 1|ϕ(y) = ϕ(x). So, from the assumption, p2 |x for any prime
p in S(y). This gives us S(y) ⊆ S(x).
We will investigate the exponent of a prime p in ϕ(n). There are two cases:
1. p divides n. Suppose that pe kn. Then we have pe−1 |ϕ(n). But is this the highest
exponent possible? No. Because in the factorization of ϕ(n), there are factors of
the form (q − 1) for any other prime divisor q of n. If p|q − 1 for any such q, those
will contribute to vp (ϕ(n)) as well. That is,
X
vp (ϕ(n)) = vp (n) − 1 + vp (q − 1).
q∈S(n)

2. p does not divide n. In this case, only factors of the form (q − 1) for any prime
divisor q of n may contribute to vp (ϕ(n)). In other words,
X
vp (ϕ(n)) = vp (q − 1).
q∈S(n)

Combining these two results, we find out that for any prime p and any positive
integer n,
 X

 vp (q − 1), if p ∤ n,

q∈S(n)
vp (ϕ(n)) = X


 v p (n) − 1 + vp (q − 1), otherwise.
q∈S(n)

Let p be a prime factor of x. Since ϕ(x) = ϕ(y), for any prime p, we must have

vp (ϕ(x)) = vp (ϕ(y)).

There are two cases to consider.

1. p ∈
/ S(y) or p ∤ y. Then,
X X
vp (x) − 1 + vp (q − 1) = vp (q − 1)
q∈S(x) q∈S(y)
X
≤ vp (q − 1) since S(y) ⊆ S(x).
q∈S(x)

The latter result implies vp (x) ≤ 1. But this is impossible since vp (x) ≥ 2 due to the
fact that p2 |x.
334 5.12. INVERSE OF EULER’S TOTIENT FUNCTION

2. p ∈ S(y). That is, p|y, or S(x) = S(y). In this case we should expect to get x = y.
One way to prove this is to show that vp (x) = vp (y). Notice that
X
vp (x) = vp (ϕ(x)) + 1 − vp (q − 1)
q∈S(x)
X
= vp (ϕ(y)) + 1 − vp (q − 1) since ϕ(x) = ϕ(y) and S(x) = S(y)
q∈S(y)

= vp (y),

which was what we wanted.

Gupta [ ] found upper and lower bounds for ϕ−1 (n). For odd n, ϕ−1 (n) is empty.
Therefore, we only need to consider the case when n is even.
Theorem 5.12.6 (Gupta). Let m and n be two positive integers such that n ∈
ϕ−1 (m). Then,
Y p
m<n≤m .
p−1
p−1|m

Proof. For even n, m = ϕ(n) < n because ϕ(n) = n holds for n = 1 only. This proves
the lower bound. For the upper bound, we can write
n Y p
=
ϕ(n) p−1
p|n
Y p
≤ .
p−1
p−1|m

The last line is true because if p|n then p − 1|m must hold, but the converse is not true.
If p − 1|m, p may or may not divide n.
Now we will look at the elements of ϕ−1 (m).
Theorem 5.12.7. Let m be a positive integers and suppose that ϕ−1 (m) contains A
elements. Then, the number of odd elements of ϕ−1 (m) is less than or equal to A/2.
Proof. For a positive integer n, if ϕ(n) = m then ϕ(2n) = m is true as well. Thus, for
any odd n, there is an even x which belongs to ϕ−1 (m). This proves that the number
of odd elements is at most half of the number of elements in ϕ−1 (m).
Theorem 5.12.8. For a prime p, there exists a positive integer n such that n ∈ ϕ−1 (2p)
if and only if 2p + 1 is a prime.
Proof. The “only if” part is easy to prove. When q = 2p + 1 is a prime, ϕ(q) = 2p so
q ∈ ϕ−1 (2p).
Now we prove the “if” part. For a positive integer n ∈ ϕ−1 (m), consider that
ϕ(n) = 2p. In other words, suppose that n ∈ ϕ−1 (2p). If p = 2 we see that n = 5
works. We need to show it for odd p now.
Suppose that n = 2a pe11 pe22 . . . pekk , where p1 , p2 , . . . , pk are odd primes. Obviously,
both a and k cannot be zero at the same time. We have three cases here:
CHAPTER 5. SPECIAL TOPICS 335

1. If a and k are both non-zero, then

ϕ(n) = 2a−1 pe11 −1 pe22 −1 . . . pkek −1 (p1 − 1)(p2 − 1) . . . (pk − 1)

= 2p.

Notice that v2 (ϕ(n)) ≥ a − 1 + k and v2 (2p) = 1. Therefore, a + k − 1 ≤ 1 or

a + k ≤ 2. This gives a = k = 1, which means n = 2p1 . Then,

ϕ(n) = p1 − 1
= 2p
=⇒ p1 = 2p + 1,

implying 2p + 1 is a prime.

2. If a = 0, then

ϕ(n) = pe11 −1 pe22 −1 . . . pkek −1 (p1 − 1)(p2 − 1) . . . (pk − 1)

= 2p.

In this case, 1 = v2 (2p) = v2 (ϕ(n)) ≥ k, and hence k = 1 or n = p1 . So,

ϕ(n) = p1 − 1 = 2p, and 2p + 1 will be a prime in this case.

3. If k = 0, then

ϕ(n) = 2a−1
= 2p,

which is not possible.

The proof is complete.

We leave the following theorems as exercise for the reader.

Theorem 5.12.9. The number of odd elements in ϕ−1 (2k ) is 0 or 1.

Theorem 5.12.10. For an odd m, the number of odd elements in ϕ−1 (m) is equal to
the number of even elements.
336 5.13. EXERCISES

5.13 Exercises
Problem 5.13.1. Let p be a prime number. Prove that there exist integers x and y
such that p = 2x2 + 3y 2 if and only if p is congruent to 5 or 11 modulo 24.

Problem 5.13.2 (KöMaL). Prove that the equation x3 − x + 9 = 5y 2 has no solution

among the integers.

Problem 5.13.3 (India 1998). If an integer n is such that 7n is the form a2 + 3b2 ,
prove that n is also of that form.

Problem 5.13.4 (USA TST 2017). Prove that there are infinitely many triples
(a, b, p) of positive integers with p prime, a < p, and b < p, such that (a + b)p − ap − bp
is a multiple of p3 .

Problem 5.13.5 (Taken from [ ]). Let p be a prime other than 7. Prove that the
following conditions are equivalent:

1. There exist integers x and y such that x2 + 7y 2 = p.

 
−7
2. = 1.
p
3. p is congruent to 1, 2, or 4 modulo 7.

Problem 5.13.6 (Taken from [ ]). Let p be a prime larger than 5. Prove that the
following conditions are equivalent:

1. There exist integers x and y such that x2 + 6y 2 = p.

2. p is congruent to 1 or 7 modulo 24.

Problem 5.13.7 (Vietnam TST 1998). Let d be a positive divisor of 5 + 19981998 .

Prove that d = 2 · x2 + 2 · x · y + 3 · y 2 , where x, y are integers if and only if d is congruent
to 3 or 7 (mod 20).

Problem 5.13.8 (Romania TST 1997). Let A be the set of positive integers of the
form a2 + 2b2 , where a and b are integers and b 6= 0. Show that if p is a prime number
and p2 ∈ A, then p ∈ A.

Problem 5.13.9 (India TST 2003). On the real number line, paint red all points
that correspond to integers of the form 81x + 100y, where x and y are positive integers.
Paint the remaining integer point blue. Find a point P on the line such that, for every
integer point T , the reflection of T with respect to P is an integer point of a different
color than T .

Problem 5.13.10 (USAMO 2001). Let S be a set of integers (not necessarily posi-
tive) such that

1. there exist a, b ∈ S with gcd(a, b) = gcd(a − 2, b − 2) = 1;

CHAPTER 5. SPECIAL TOPICS 337

2. if x and y are elements of S (possibly equal), then x2 − y also belongs to S.

Prove that S is the set of all integers.
Problem 5.13.11. Let a, b, and n be positive integers such that n > 2. Prove that if
an + bn
k=
(ab)n−1 + 1

is an integer, then k is a perfect nth power.

Problem 5.13.12 (IZHO 2005). Solve the equation p2 − 6pq + (q 2 + 4) = 0 in prime
numbers less than 2005.
Problem 5.13.13. Let a, b, and k be positive integers such that
a2 + b2
k=
ab − 1
Prove that k = 5.
Problem 5.13.14 (IMO 2007). Let a and b be positive integers. Show that if 4ab − 1
divides (4a2 − 1)2 , then a = b.
Problem 5.13.15 (PEN). If a, b, c are positive integers such that

0 < a2 + b2 − abc ≤ c,

show that a2 + b2 − abc is a perfect square.

Problem 5.13.16 (IMO ShortList 2003). Determine all pairs of positive integers
(a, b) such that
a2
2ab2 − b3 + 1
is a positive integer.
Problem 5.13.17. Find all triples (x, y, z) of positive integers such that (x + y + z)2 =
7xyz.
Problem 5.13.18. Let a and b be positive integers such that ab divides a2 + b2 + 2.
2 2 +2
Prove that a +b
ab
= 4.
Problem 5.13.19. Find all positive integers x, y, and z such that x2 + y 2 + 2 = xyz.
Problem 5.13.20 (Ireland 2005). Let m, n be integers with the same parity such
that m2 − n2 + 1 divides n2 − 1. Prove that m2 − n2 + 1 is a perfect square.
Problem 5.13.21 (Mongolia 2000). For which positive integer k there exist positive
integers x, y, and z such that (x + y + z)2 = kxyz?
Problem 5.13.22. Prove that the following equation has no positive integer solution
(x, y, z)
x2 + y 2 + z 2 = xyz + 1.
338 5.13. EXERCISES

Problem 5.13.23. Prove that the equation

x2 + y 2 + z 2 = n(xyz + 1)

has a solution (x, y, z) in positive integers if and only if n can be represented as sum of
two perfect squares.
Problem 5.13.24. Let a and b are positive integers such that

a + 1 | b2 + 1,
b + 1 | a2 + 1.

Prove that a and b are odd numbers.

Problem 5.13.25. Find all positive integers a and b such that

a2 + b2 + a + b + 1
ab
is an integer
Problem 5.13.26. Let m and n be positive integers such that mn 6= 1. Let

m2 + mn + n2
k= .
mn − 1
If k is an integer, find all its possible values.
Problem 5.13.27. Find all pairs of integers (m, n) such that
m n
+
n m
is also an integer.
Problem 5.13.28 (Vietnam 2002). Find all integers n for which there exist infinitely
many integer solutions to
√
a + b + c + d = n abcd.

Problem 5.13.29 (Putnam 1933). Prove that for every real number N , the equation

a2 + b2 + c2 + d2 = abc + bcd + cda + dab

has a solution in which a, b, c, and d are all integers greater than N .

Problem 5.13.30 (UNESCO Competition 1995). Let a, n be two positive integers
and let p be an odd prime number such that

ap ≡ 1 (mod pn ).

Prove that
a≡1 (mod pn−1 ).
CHAPTER 5. SPECIAL TOPICS 339

Problem 5.13.31 (Iran Second Round 2008). Show that the only positive integer
value of a for which 4(an + 1) is a perfect cube for all positive integers n, is 1.

Problem 5.13.32. Let k > 1 be an integer. Show that there exists infinitely many
positive integers n such that

n|1n + 2n + 3n + · · · + k n .

Problem 5.13.33 (Ireland 1996). Let p be a prime number, and a and n positive
integers. Prove that if
2p + 3p = an ,
then n = 1.

Problem 5.13.34 (Russia 1996). Let x, y, p, n, k be positive integers such that n is

odd and p is an odd prime. Prove that if xn + y n = pk , then n is a power of p.

Problem 5.13.35. Find the sum of all the divisors d of N = 1988 − 1 which are of the
form d = 2a 3b with a, b ∈ N.

Problem 5.13.36. Let p be a prime number. Solve the equation ap − 1 = pk in the

set of positive integers.

Problem 5.13.37. Find all solutions of the equation

(n − 1)! + 1 = nm

in positive integers.

Problem 5.13.38 (Bulgaria 1997). For some positive integer n, the number 3n − 2n
is a perfect power of a prime. Prove that n is a prime.

Problem 5.13.39. Let m, n, b be three positive integers with m 6= n and b > 1. Show
that if prime divisors of the numbers bn − 1 and bm − 1 be the same, then b + 1 is a
perfect power of 2.

Problem 5.13.40 (IMO ShortList 1991). Find the highest degree k of 1991 for
which 1991k divides the number
1992 1990
19901991 + 19921991 .

Problem 5.13.41. Prove that the number aa−1 − 1 is never square-free for all integers
a > 2.

Problem 5.13.42 (Czech Slovakia 1996). Find all positive integers x, y such that
px − y p = 1, where p is a prime.

Problem 5.13.43. Let x and y be two positive rational numbers such that for infinitely
many positive integers n, the number xn − y n is a positive integer. Show that x and y
are both positive integers.
340 5.13. EXERCISES

Problem 5.13.44 (IMO 2000). Does there exist a positive integer n such that n has
exactly 2000 prime divisors and n divides 2n + 1?

Problem 5.13.45 (China Western Mathematical Olympiad 2010). Suppose

m
that m and k are non-negative integers, and p = 22 + 1 is a prime number. Prove that
m+1 pk
• 22 ≡ 1 (mod pk+1 );

• 2m+1 pk is the smallest positive integer n satisfying the congruence equation 2n ≡ 1

(mod pk+1 ).

Problem 5.13.46. Let p ≥ 5 be a prime. Find the maximum value of positive integer
k such that
pk |(p − 2)2(p−1) − (p − 4)p−1 .

Problem 5.13.47. Find all triples (x, y, z) of integers such that 3x + 11y = z 2 .

Problem 5.13.48. Find all positive integer solutions to pa − 1 = 2n (p − 1), where p is

prime.

Problem 5.13.49. Prove that there are no positive integers x, y, and z such that
x7 + y 7 = 1998z .

Problem 5.13.50 (Baltic Way 2012). Let d(n) denote the number of positive divi-
sors of n. Find all triples (n, k, p), where n and k are positive integers and p is a prime
number, such that
nd(n) − 1 = pk .

Problem 5.13.51 (IZHO 2017). For each positive integer k, denote by C(k) the sum
of the distinct prime divisors of number k. For example, C(1) = 0, C(2) = 2, C(45) = 8.
Determine all positive integers n such that C(2n + 1) = C(n).

Problem 5.13.52 (Hong Kong TST 2016). Find all triples (m, p, q) such that

2 m p2 + 1 = q 7 ,

where p and q are ptimes and m is a positive integer.

Problem 5.13.53 (Brazil 2016). Define the sequence of integers an (for n ≥ 0) such
that a0 is equal to an integer a > 1 and

an+1 = 2an − 1.

Let A be a set such that x belongs to A if and only if x is a prime divisor of an for
some n ≥ 0. Show that the number of elements of A is infinite.

Problem 5.13.54 (USAMO‌2017). Prove that there are infinitely many distinct pairs
(a, b) of relatively prime integers a > 1 and b > 1 such that ab + ba is divisible by a + b.

Problem 5.13.55 (Italy TST 2003). Let a and b be positive integers and p be a
prime. Find all solutions to the equation 2a + pb = 19a .
CHAPTER 5. SPECIAL TOPICS 341

Problem 5.13.56 (Turkey EGMO TST 2017). Determine all triples (m, k, n) of
positive integers satisfying the following equation

3m 5k = n3 + 125.

Problem 5.13.57 (Balkan 2013). Determine all positive integers x, y, and z such
that x5 + 4y = 2013z .

Problem 5.13.58. If pn is the nth prime then prove that the integer N = p1 p2 p3 . . . pn +
1 can not be a perfect power.

Problem 5.13.59. Find all ordered triplets (a, b, c) of positive integers such that

2a − 5b · 7c = 1.

Problem 5.13.60 (Vietnam TST 2016). Find all positive integers a and n with
2016
a > 2 such that each prime divisor of an − 1 is also prime divisor of a3 − 1.

Problem 5.13.61. Find all positive integers n, for which n and 2n + 1 have the same
set of prime divisors.

Problem 5.13.62. Find all triplets (x, y, z) of positive integers such that

(z + 1)x − z y = −1.
Bibliography

[1] Clarke, Pete L. Some Applications of Thue’s Lemma. (2009).

[2] R. D. Carmichael, Theory of Numbers, page 36.

[3] C. Pomerance, On Carmichael Conjecture, American Mathematical Society, Vol-

ume 43, Number 2, April 1974.

[4] V. L. Klee, Jr, On A Conjecture Of Carmichael, American Mathematical Society,

December, 1947.

[5] Hansraj Gupta, Euler’s Totient Function And Its Inverse, Indian J. Pure Appl.
Math, 12(1): 22 − 30, January 1981.

343
Glossary

345
Appendix A

Identities and Well-Known Theorems

We suppose that a, b are real numbers in the following results. This PDF was created
for Amazon.

Identity A.1.
(a + b)2 + (a − b)2 = 2(a2 + b2 ),
(a + b)2 − (a − b)2 = 4ab.
Identity A.2 (Sophie Germain Identity).
a4 + 4b4 = (a2 + 2ab + 2b2 )(a2 − 2ab + 2b2 ).
Identity A.3.
(a + b)3 − a3 − b3 = 3ay(a + b),
(a + b)5 − a5 − b5 = 5ay(a + b)(a2 + ab + b2 ),
(a + b)7 − a7 − b7 = 7ay(a + b)(a2 + ab + b2 )2 .
Theorem A.0.1 (Binomial Theorem). For any positive integer n,
     
n n n n−1 n n−2 2 n
(a + b) = a + a b+ a b + ··· + abn−1 + bn
1 2 1
n  
X n n−i i
= a b
i=0
i
n  
X n i n−i
= ab .
i=0
i

Proof. If we expand the left side, there are n times (a + b) multiplied together. Ob-
viously, each term of this expansion will be of the form ai bn−i . So, we can write it
as
(a + b)n = sn an + sn−1 an−1 b + · · · + s1 abn−1 + s0 bn .
Here, si is the coefficient of ai bn−i . We see that for a term ai bn−i to appear in the sum,
we must choose i of the n times (a + b) to contribute an a to the term, and then each
of the other n −
i terms of the product must

contribute a b. We can take i objects from
n objects in ni ways. Therefore, si = ni .

347
348

Theorem A.0.2. For positive integers n and k such that k ≤ n,

   
n n
1. = ,
k n−k
     
n n−1 n−1
2. = + (Pascal’s recurrence),
k k k−1
   
n n n−1
3. = (absorption property),
k k k−1
       
n n n n
4. + + ··· + + = 2n ,
0 1 n−1 n
         
0 1 n−1 n n+1
5. + + ··· + + = ,
k k k k k+1
  2  2  2  2  
n n n n 2n
6. + + ··· + + = .
0 1 n−1 n n



Corollary A.0.3. If n and k are coprime, then n divides nk and k divides n−1
k−1
.
Proof. Directly implied from part of theorem ( ).
Identity A.4 (Sum of Powers of Consecutive Integers). Let n be a positive
integer. Then,
n
X n(n + 1)
1. i= ,
i=1
2
n
X n(n + 1)(2n + 1)
2. i2 = ,
i=1
6
n  2
X
3 n(n + 1)
3. i = ,
i=1
2
n
X n(n + 1)(2n + 1)(3n2 + 3n − 1)
4. i4 = .
i=1
30
Identity A.5 (Sum of Differences). Let a1 , a2 , a3 , . . . be an infinite sequence of
numbers. Then, for any positive integer n,
n−1
X
an = a1 + (ak+1 − ak ) .
k=1

Proof. Expand the sum on the right side to obtain

n−1
X
(ak+1 − ak ) = (an − an−1 ) + (an−1 − an−2 ) + · · · + (a2 − a1 )
k=1
= an − a1 .
The conclusion follows.
APPENDIX A. IDENTITIES AND WELL-KNOWN THEOREMS 349

Identity A.6 (Fibonacci-Brahmagupta Identity). For any reals a, b, c, d, and any

integer n,

(a2 + nb2 )(c2 + nd2 ) = (ac − nbd)2 + n(ad + bc)2

= (ac + nbd)2 + n(ad − bc)2 .

In other words, the product of two numbers of the form a2 + nb2 is of the same form.
Particularly, for n = 1,

(a2 + b2 )(c2 + d2 ) = (ac + bd)2 + (ad − bc)2

= (ad + bc)2 + (ac − bd)2 .

The following identity is a generalization of Fibonacci-Brahmagupta Identity. La-

grange used this identity to prove the Sum of Four Squares Theorem.

Identity A.7 (Euler’s Four Square Identity). Let a1 , a2 , . . . , a4 and b1 , b2 , . . . , b4

be reals. Then,

(a21 + a22 + a23 + a24 )(b21 + b22 + b23 + b24 )

= (a1 b1 + a2 b2 + a3 b3 + a4 b4 )2 + (a1 b2 − a2 b1 + a3 b4 − a4 b3 )2
+ (a1 b3 − a2 b4 − a3 b1 + a4 b2 )2 + (a1 b4 + a2 b3 − a3 b2 − a4 b1 )2 .

An interested reader can see Degen’s eight-square identity [ ] or Pfister’s sixteen-

square identity [ ], but they do not look pretty at all and we do not include them
here.

Identity A.8 (Lebesgue Identity).

(a2 + b2 − c2 − d2 )2 + (2ac + 2bd)2 + (2ad − 2bc)2 = (a2 + b2 + c2 + d2 )2

Identity A.9 (Euler-Aida Ammei Identity). Let x1 , x2 , . . . , xn be reals. Then,

n
X
(x21 − x22 2
− · · · − xn ) + (2x1 xi )2 = (x21 + x22 + · · · + x2n )2 .
i=2

Identity A.10 (Bhaskara’s Lemma). Let m, x, y, n and k be integers such that

k 6= 0. If y 2 − nx2 = k, then
 2  2
mx + ny mx + y m2 − n
−n = .
k k k

This identity is used in Chakravala method to find the solutions to Pell-Fermat

equation.
Bibliography

[1] Piezas III, Titus. "The Degen-Graves-Cayley Eight-Square Identity." Web address
accessed 27 August 2016.
. (2005).

[2] Conrad, Keith. "Pfister’s Theorem on Sum of Squares." Web address accessed
27 August 2016.
.

351
 Part II

Problem Column

353
Chapter 6

Solving Challenge Problems

There will be plenty of time to rest

in the grave.

Paul Erdós

Problem 6.1. Prove divisibility criteria for 2, 3, 4, 5, 7, 9, 11, 13, 17, 19, as stated in
section ( ). This PDF was created
for Amazon.

Solution. Let n be a positive integer and it has base 10 representation ak . . ¯. a1 a0 .

Here we show the proof for 2, 3, 4 and then you should try the rest yourself in a similar
fashion.
2: n is divisible by 2 if and only if the last digit is even. First we will prove that if
n is divisible by 2, then the last digit must be even.

2|n = 10k ak + · · · + 10a1 + a0

Now, 2|10 so 2|10k , . . . , 10. In turn,

2|10k ak + · · · + 10a1
2|10k ak + · · · + 10a1 + a0 − (10k ak + · · · + 10a1 ) = a0

This implies that 2 must divide a0 . Since a0 is a digit, a0 ∈ {0, 2, 4, 6, 8}.

Now, assume that, a0 is even so a0 = 2b0 . Then we have

n = 10k ak + · · · + 10a1 + a0
= 2(2k−1 5k ak + · · · + 5a1 + b0 )

which is obviously divisible by 2.

An alternative approach would be using congruence.

n ≡ 10k ak + · · · + 10a1 + a0 (mod 2)

≡ a0 (mod 2)

since 10 ≡ 0 (mod 2).

355
356

3: n is divisible by 3 if and only if the sum of digits is divisible by 3.

n = 10k ak + · · · + 10a1 + a0
. . 9} +1)ak + · · · + (9 + 1)a1 + a0
= (9| .{z
k 9’s
= (3 · |3 .{z
. . 9} +1)ak + · · · + (3 · 3 + 1)a1 + a0
k 3’s
!
=3· . . 9} ak + · · · + 3a1
3| .{z + ak + · · · + a1 + a0
k 3’s

Notice that, when we divide n by 3, the term with 3 vanishes since it is divisible
by 3. This is now straightforward that 3 will divide n if and only if ak + · · · + a0
is divisible by 3. Since ak , . . . , a0 are digits of n in base 10, the claim is proven.

4: n is divisible by 4 if and only if the number formed by the last two digits of n is
divisible by 4.

n = 10k ak + · · · + 10a1 + a0
= 10k ak + · · · + 100a2 + 4l


= 4 25 · 10k−2 ak + · · · + l

This is definitely divisible by 4. The only if part is straightforward from this.

5: n is divisible by 5 if and only if the last digit of n is divisible by 5 i.e. 0 or 5. This

is again straightforward since 10 is divisible by 5.

Problem 6.2. Let n be a positive integer. Show that the product of n consecutive
integers is divisible by n!.

Before we solve this problem, let me tell you about my story when I first encountered
this problem. Since n! = 1 · 2 · · · n, the first thing that I thought was: since there are
n consecutive integers, one must leave the remainder 0 modulo n. And this has to be
true for all m ≤ n. But then I immediately realized that even if we proved that the
product of these n integers is divisible by all m ≤ n individually, we can not guarantee
that their 1 · 2 · · · n will divide their product too! It would be a common mistake to
think so. Beginners tend to make assumptions that are wrong. For example, a and b
both divide c, then ab divides c too. We will see some common mistakes as we solve
the problems. Be aware of them whenever you are thinking about a problem. So, to be
on the safe side, if you assume something, don’t think it’s true until you can prove it.
Appearances can be deceiving! Now, let’s see a correct solution.

Solution. Let the n consecutive integers be a, a − 1, . . . , a − n + 1 (verify that if you

take n consecutive integers decreasing from a, the last integer will be a − n + 1). See
1
I was a total beginner then -
CHAPTER 6. SOLVING CHALLENGE PROBLEMS 357

that,
a(a − 1) · · · (a − n + 1) · (a − n) · · · 1
S = a(a − 1) · · · (a − n + 1) =
(a − n) · · · 1
a!
=
(a − n)!
S
If we can prove that this product is an integer, we are done.
n!
S a!
=
n! (a − n)!n!
 
a
=
n
which is clearly an integer.
2p


Problem 6.3. Prove that, p2 divides p
− 2.
Solution. This problem can be dealt with very easily if you know Wolstenholme’s
theorem which says for p > 3,
   
2p 2
≡ (mod p3 )
p 1
≡ 2 (mod p3 )
 
3 2p
p| −2
p
For p ≤ 3, we can check manually.
Remark. Another way to do it, if you don’t know or can not remember the theorem.
Use the identity:
n  2  
X n 2n
=
i=0
i n
p  2  
X p 2p
=
i=0
i p



p 2
We already know that for 0 < i < p, p divides pi . So p2 divides i
for such i and we
have,
p  2  2  2
X p p p
≡ + (mod p2 )
i=0
i 0 p
 
2p
≡ 2 (mod p2 )
p
 
2 2p
p| −2
p
2 n


If k < n, then k = 0.
358

Problem 6.4 (Masum Billal). Find all functions f : N → N such that

(f (n + 1), f (n)) = [f (n), f (n − 1)]
holds for all n > 1.
Solution. Let gn = (f (n + 1), f (n)) and ln = [f (n + 1), f (n)]. We have gn+1 = ln . But
note that f (n)|ln and ln = gn |f (n + 1). This gives f (n)|f (n + 1).
gn = (f (n + 1), f (n)) = f (n)
ln = [f (n), f (n + 1)] = f (n + 1)
This holds because if a|b, then (a, b) = a and [a, b] = b. Since f (n)|f (n + 1), we can
f (n + 1)
define a sequence (bi )i≥1 with bn = and b1 = f (1), which is obviously an
f (n)
integer sequence.
f (n + 1) = bn f (n)
= bn bn−1 f (n − 1)
= bn bn−1 bn−2 f (n − 2)
..
.
= bn bn−1 · · · b1
Yn
= bi
i=1

Any integer sequence (bi )i≥1 works.

Sometimes we may find expression that’s symmetric with respect to some variables
(say a, b). When you find an expression that’s symmetric on a and b, you can assume
a ≥ b. That’s a great advantage in many problems.
Problem 6.5 (Russia, 2000). If a and b are positive integers such that a + b =
(a, b) + [a, b] then one of a, b divides the other.
This is a problem that has many solutions, being a relatively easier problem. We
will show two solutions here. Let g = (a, b) and l = [a, b] for brevity. Then the equation
becomes a + b = g + l. Without loss of generality, we can assume a ≤ b.
ab
Solution (First). We know ab = gl or l = . Substituting this into the equation:
g
ab
a+b=g+
g
2
g + ab = g(a + b)
(g − a)(g − b) = 0
Since a ≤ b, g − a = 0, then (a, b) = a implies a|b.
3
To see if an expression is symmetric on a, b switch places of a and b. If the expression remains the
same it is symmetric, otherwise it is not. In short, we must have f (a, b) = f (b, a)
CHAPTER 6. SOLVING CHALLENGE PROBLEMS 359

Solution (Second). Assume a = gx, b = gy with x⊥y, then l = gxy.

g(x + y) = g + gxy
xy + 1 = x + y
(x − 1)(y − 1) = 0

Again, since a ≤ b, x ≤ y so x − 1 = 0 or x = 1, we have a = g. Same conclusion.

Problem 6.6 (Slovenia 2010). Find all primes p, q, r with 15p + 7pq + qr = pqr.
Solution. We can write it as p(15 + 7q) + qr = pqr or

p(qr − 15 − 7q) = qr

Therefore, p|qr and since p, q, r are primes p|q or p|r. If p = q, then

pr − 15 − 7p = r
r(p − 1) = 7p + 15
p − 1|7p + 15

Since p − 1|7p − 7, p − 1|(7p + 15) − (7p − 7) = 22. So p − 1 ∈ {1, 2, 11, 22} which gives
7p + 15
p ∈ {2, 3, 23}. But we also need r = , a prime. We get that r is a prime when
p−1
p = 2 only when r = 29. The other case is p = r, so

pq − 15 − 7q = q
(p − 8)q = 15

We have q ∈ {3, 5}. If q = 3, then p = 13 which is valid. If q = 5, p = 11 and this is a

valid solution too.
Problem 6.7 (Serbia 2014). A special number is a positive integer n for which there
exist positive integers a, b, c and d with
a3 + 2b3
n=
c3 + 2d3
Prove that,
a. There are infinitely many special numbers.

b. 2014 is not a special number.

Solution. Proving (a) is easy since we just have to show an infinite such n. So we
can choose a, b, c, d however we want, as long as they serve our purpose. Let’s go with
a = ck, b = dk, then
a3 + 2b3 c3 k 3 + 2d3 k 3
=
c3 + 2d3 c3 + 2d3
3
=k
360

Since we are free to choose k here, we have infinitely many n.

For part (b), let’s assume

a3 + 2b3 = 2014(c3 + 2d3 )

= 2 · 19 · 53(c3 + 2d3 )

We can consider modulo 19 in this equation. We just have to check cubes modulo 19
and the reader can verify that if a3 ≡ −2b3 (mod 19), then we must have 19|a, b since
x3 ≡ 0, ±1, ±7, ±8 (mod 19). Say, a = 19x, b = 19y.

193 (x3 + 2y 3 ) = 2 · 19 · 53(c3 + 2d3 )

19|c3 + 2d3

This also shows that 19|c, d so let c = 19z, d = 19w. but then

x3 + 2y 3 = 2014(z 3 + w3 )

we get a smaller solution (x, y, z, w) which is actually infinite descent.

Problem 6.8 (Croatia 2015). Let n > 1 be a positive integer so that 2n − 1 and
3n − 2 are perfect squares. Prove that 10n − 7 is composite.

Solution. Take 2n − 1 = x2 and 3n − 2 = y 2 . We need to reach 10n somehow here,

and incidentally 10 = 12 − 2 = 3 · 4 − 2 · 1. So, we do this:

4(y 2 ) − 1(x2 ) = 4(3n − 2) − (2n − 1)

(2y + x)(2y − x) = 10n − 7

Since n > 1, y > 1 so 2y − 1 > 1. Thus 10n − 7 is not a prime.

Problem 6.9. Find all nonnegative integers m, n such that 3m − 5n is a perfect square.

Solution. Let 3m − 5n = a2 . Since there are squares, we should consider modulo 4, the
numbers 3, 5 also suggest us to take modulo 4. That way, we get to know about m and
n. Since both sides must leave the same remainder upon division by 4,

(−1)m − 1n ≡ a2 ≡ 0, 1 (mod 4)

If m is odd then a2 ≡ −1 − 1 ≡ 2 (mod 4), which is not possible. So m is even. If

m = 2l, can write the equation as

2
3l − a2 = 5n
(3l + a)(3l − a) = 5n

In the right side there is nothing but 5, so we must have 3l + a = 5x , 3l − a = 5y for

some nonnegative integer x, y. If we add them

2 · 3 l = 5x + 5 y
CHAPTER 6. SOLVING CHALLENGE PROBLEMS 361

If y is 0, then
2 · 3 l = 5x + 1
5 + 1 is divisible by 2 and 3, but according to Zsigmondy’s theorem 5x + 1 will have a
prime factor that is neither 2 nor 3 if x > 1. Clearly x > y so if y 6= 0, then 5 divides
2 · 3l , contradiction. So there is no such integers m, n except the trivial solutions when
x = 0 or x = 1.
Problem 6.10 (Croatia 2015). Prove that there does not exist a positive integer n
for which 7n − 1 is divisible by 6n − 1.
Solution. Assume to the contrary that, 6n − 1|7n − 1. Due to 6n − 1|6n − 1, subtraction
gives
6n − 1|7n − 1 − (6n − 1) = 7n − 6n
See that left side is divisible by 6 − 1 = 5, so right side is divisible by 5 too.
7n − 6n ≡ 2n − 1 (mod 5)
The smallest positive integer for which 2n − 1 is divisible by 5 is n = 4. So, n must
be divisible by 4. But then 64 − 1 = 37 · 5 · 7 divides 6n − 1. So 7 divides 7n − 6n , a
contradiction.
Problem 6.11. Find all positive integers n such that n2 − 1|2n + 1.
Solution. Clearly 2n + 1 is odd, so n2 − 1 must be odd as well. This means that n is
even. Let n = 2k for some positive integer k. Then
n2 − 1 ≡ 4k 2 − 1 ≡ 3 (mod 4).
By theorem ( ), every number of the form 4k + 3 has a prime divisor of that form.
Therefore, there is a prime p such that p|n2 − 1 and p ≡ 3 (mod 4). Now, according to

2
theorem ( ), every prime divisor of 2n + 1 = 2k + 1 is of the form 4k + 1. This
is a contradiction because p|n2 − 1|2n + 1. Thus no such n exists.
Note. Although some problems seem difficult at first sight, they are pretty easy if you
think in a proper way.
Problem 6.12. Find all m, n ∈ N such that 2n − 1|m2 + 9.
Solution. n = 1 is obviously a solution (which works for any m), so let’s look at n > 1
only. Note that m2 + 9 = m2 + 32 , if m ⊥ 3, then m2 + 32 is a bisquare. Therefore,
if n > 1 then 2n − 1 ≡ −1 (mod 4), so m2 + 9 will have a prime divisor of the form
4k + 3. But we know that, no bisquare has a prime divisor of this form. Therefore, m
must be divisible by 3. If m = 3k,
2n − 1|9(k 2 + 1)
Now, no matter what k is, k 2 + 1 is always a bisquare. Therefore, it can not have any
divisors of the form 4k + 3. So, 2n − 1|9, checking with 2n − 1 = 1, 3, 9, we get that
n = 1, 2 are the solutions.
362

You never know what’s coming next until you think clearly!

Problem 6.13. If n > 1, prove that n2 − 1 divides 2n! − 1 for even n.

Solution. If we set m = n + 1, then we need to prove m(m − 2) divides 2(m−1)! − 1.

Since n is even, m is odd so m is co-prime to 2.

2ϕ(m) ≡ 1 (mod m) and

2ϕ(m−2) ≡ 1 (mod m − 2)

Since ϕ(m) < m and ϕ(m−2) ≤ m−2, ϕ(m) and ϕ(m−2) divides (m−1)!. Therefore,

2(m−1)! ≡ 1 (mod m)
(m−1)!
2 ≡1 (mod m − 2)

This implies m|2(m−1)! − 1 and m − 2|2(m−1)! − 1. Since m ⊥ m − 2 for odd m, we have

m(m − 2)|2(m−1)! − 1.

Problem 6.14. Prove that n divides 2n + 1 for infinitely many n ∈ N.

Solution. One can easily observe that n = 3 works since n is odd, so we could take
n = 3k. Then we see that 23k + 1 = 8k + 1 is divisible by 9 since k is odd. This suggests
k
us to take n = 3k . Indeed, it works because due to LTE, 3k+1 ||23 + 1, so n = 3k gives
us infinitely such n.

Problem 6.15 (Croatia 2015). Determine all positive integers n for which there exists
a divisor d of n such that dn + 1|d2 + n2 .

Solution. Let n = dk where k ∈ N. The equation becomes

d2 k + 1|d2 + d2 k 2 and
d2 k + 1|d2 k 2 + k
d2 k + 1|d2 k 2 + d2 − (d2 k 2 + k)
= d2 − k

If d2 > k then d2 k + 1|d2 − k but clearly d2 − k < d2 < d2 k + 1, contradiction. If k > d2 ,

then k − d2 < k < kd2 + 1. Thus, d2 − k = 0 or k = d2 . We get n = dk = d3 .

Problem 6.16 (IMO Shortlist 2013, N1, Proposed by Malaysia). Find all func-
tions f : N → N such that

m2 + f (n)|mf (m) + n

Arithmetic functional equations or divisibility problems are really popular for IMO
or Shortlist. Anyway, let’s see how we can solve this one. We say it beforehand that, it
can be solved in many ways, being an easy problem. So if you try yourself you should
be able to do it.
CHAPTER 6. SOLVING CHALLENGE PROBLEMS 363

Solution (First). There are two variables in this divisibility. Sometimes reducing to
one variable and then working on it alone suffices for some problems. By the way, you
have probably guessed already, f (n) = n is the solution.
First let’s play with some values of m and n. To remove two variables, set m = n.
n2 + f (n)|nf (n) + n
nf (n) + n ≥ n2 + f (n)
(n − 1)(f (n) − n) ≥ 0
Since n ≥ 1, we have f (n) − n ≥ 0 so f (n) ≥ n. If we can prove now that f (n) ≤ n,
we will have f (n) = n. Set n = 2 in the divisibility.
4 + f (2)|2f (2) + 2
4 + f (2)|8 + 2f (2)
4 + f (2)|8 + 2f (2) − (2f (2) + 2)
4 + f (2)|6
Since 4 + f (2) ≥ 5, we must have 4 + f (2) = 6 or f (2) = 2. Now, using n = 2 in the
original divisibility,
m2 + f (2)|mf (m) + 2
m2 + 2|mf (m) + 2
m2 + 2 ≤ mf (m) + 2
m2 ≤ mf (m)
m ≤ f (m)
And we are done!
Solution (Second). This one uses another great idea. This kind of technique is useful
in many cases. Another example of choosing a special prime.
Set m = f (n) in the divisibility, and it gives us
f (n)2 + f (n)|f (n)f (f (n)) + n
f (n)(f (n) + 1)|f (n)f (f (n)) + n
Thus, f (n)|f (n)f (f (n)) + n or f (n)|n, so f (n) ≤ n. Now, we will explain our main
idea. We will make the right side a prime (think what the benefit of doing so).
But we need to understand if that’s achievable. It is, since we can take n as we please,
and there is a free n on the right side. Let p = mf (m) + n for a prime p > mf (m). In
fact, we take a prime p > m2 , that way we also ensure p > mf (m) since f (m) ≤ n. So,
for that m,
m2 + f (n)|mf (m) + n = p
This forces us to m2 + f (n) = p since m2 + f (n) ≥ 2.
p − m2 = f (n) ≤ n
= p − mf (m)
which gives m ≤ f (m). So f (n) = n.
364

Problem 6.17 (IMO 1990, Problem 3). Find all n ∈ N for which n2 |2n + 1.
Solution. Let’s see if we can determine the smallest prime factor again. If p is the
smallest prime divisor of n,

2n ≡ −1 (mod p)
22n ≡ 1 (mod p)

And from Fermat’s little theorem, 2p ≡ 1 (mod p). So we get

2(2n,p−1) ≡ 1 (mod p)

Using the same argument as before, n⊥p − 1, so (2n, p − 1) = (2, p − 1) = 2 because p

is odd. This gives us 22 ≡ 1 (mod p) or p = 3. What do we do now? We only have the
smallest prime. So, we can assume n = 3α k where k is not divisible by 3. Think yourself
about the reason to do this. It would be certainly fruitful to find out what values α can
assume. This is where Lifting the Exponent lemma comes to the rescue! But first
we need to make sure we can actually apply LTE. In this case, we can because
2 + 1 = 3, divisible by 3 and 2⊥1. We have

ν3 (n2 ) = ν3 (32α k 2 )
= ν3 (32α ) + ν3 (k 2 )
= 2α

On the other hand, from LTE,

α
ν3 (23 + 1) = ν3 (2 + 1) + ν3 (3α )
=1+α

n2 |2n + 1 implies that

ν3 (2n + 1) ≥ ν3 (n2 )
1 + α ≥ 2α
1≥α

Since α is a positive integer, we have α = 1. So n = 3k with 3 ∤ k. The problem is now

finding k with

k 2 |8k + 1

Here, we again try to determine the smallest prime divisor of k, we call it q. Then
8k ≡ −1 (mod q).

82k ≡ 1 (mod q)
8q−1 ≡ 1 (mod q)
8(2k,q−1) ≡ 1 (mod q)
82 ≡ 1 (mod q)
CHAPTER 6. SOLVING CHALLENGE PROBLEMS 365

We hope the lines above don’t need a second explanation. This way, q|63 = 32 · 7. Since
3⊥q, we can only have q = 7. But,

8k + 1 ≡ 1k + 1 ≡ 2 (mod 7)

This is impossible, which means k doesn’t have any prime divisor i.e. k = 1. The only
solution we have is n = 1, 3.
Note. This is a fantastic problem which uses couple of techniques at the same time.
Worthy of being a problem 3 at the IMO!
Problem 6.18 (All Russian Olympiad 2014, Day 2). Define m(n) to be the greatest
proper natural divisor of n ∈∈ N . Find all n such that n + m(n) is a power of 10.
Solution. Let n + m(n) = 10a . If n is a prime then m(n) is clearly 1. In that case,

p + 1 = 10a
p = 10a − 1

Right side is divisible by 10 − 1 = 9, so p can not be prime. Now, if n > 1 and not a
prime, then n has a smallest prime divisor. Then the greatest proper divisor of n will
n
be , let’s say n = pk. Be careful here, p is the smallest prime does not mean that k
p
is not divisible by p. For example, 12 = 22 · 3 so k = 6. Take k = pr l where all prime
factors of l must be greater than p and r ≥ 0.

n + m(n) = 10a
pk + k = 10a
pr l(p + 1) = 10a

If r ≥ 1 then p divides 10. So p = 2 or p = 5. If p = 2, then p + 1 = 3 divides 10a ,

contradiction. If p = 5, p + 1 = 6 divides 10a , again contradiction. So r = 0 and
l(p + 1) = 10a . It is clear that l is odd, otherwise 2|l and hence p < 2 since all prime
factors of l are greater than p. This also provides with p + 1 ≤ l. Since l is odd, l = 5x
for some 1 ≤ x ≤ a. Since p is less than all prime factors of l, we must have p = 3.

5 x · 4 = 2a 5 a

which immediately gives a = 2, so x = a = 2. Thus, n = pk = 3 · 52 = 75.

Problem 6.19 (Czech Slovakia 1996). Find all positive integers x, y such that px −
y p = 1 where p is a prime.
Solution. If p = 2, then 2x = y 2 + 1. If x ≥ 2, then x2 ≡ −1 (mod 4), which is
impossible. So x = 1 and now p is odd. Then px = y p + 1 = (y + 1)S for some integer
S. Obviously y +1 is divisible by p but y p +1 has a primitive divisor unless y = 2, p = 1.
Problem 6.20 (China 2001, Problem 4). We are given three integers a, b, c such
that a, b, c, a + b − c, a + c − b, b + c − a, and a + b + c are seven distinct primes. Let d
be the difference between the largest and smallest of these seven primes. Suppose that
800 ∈ {a + b, b + c, c + a}. Determine the maximum possible value of d.
366

Solution. Observation: all of a, b, c are odd prime. In cases like this, show a contradic-
tion that the other case can not happen. So let’s assume that a = 2 and b, c are odd.
Then a + b − c is even since b − c is even, so not a prime unless a + b − c = 2 but then
b = c which contradicts that b, c are distinct primes. We leave the other cases for the
reader.
From what we just proved, the smallest prime of a, b, c (namely c ≥ 3) must be at
least 3. What other information is there for us to use? 800 ∈ {a + b, b + c, c + a}, but
we don’t need to analyze every case since they are symmetric over a, b, c. Without loss
of generality, take a + b = 800. a + b − c > 0 is a prime too, so c < a + b = 800 or
c ≤ 799, a prime. We can check that 17 divides 799, so c ≤ 797, inferring a + b + c ≤
800 + 797 = 1597. And by luck, 1597 is a prime (well, that’s how the problem creator
created the problem). So if we can find a, b so that all are primes, we are done and in
that case, d = a + b + c − 3 = 1594 since a + b + c is the largest prime and 3 is the
smallest prime. a must be greater than or equal to 5, but since a + b = 800, a can not
be 5. Let’s check starting from a = 7, b = 793. With some tedious calculations, we can
find that a = 13, b = 787 satisfies all the conditions. Other primes would be 23, 1571.
Problem 6.21 (Masum Billal, Bangladesh TST 2015). Find the number of pos-
itive integers d so that for a given positive integer n, d divides an − a for all integer
a.
Solution. Let’s focus only on n > 1. How do we understand the nature of d? Surely we
should take a prime divisor of d, say p. If we can find the values of p and the exponent
of p in d, we can find d. So we get p|an − a for all p|d. What should we use to get a clue
on the exponents? We can set different values of a. And it seems wise to use a = p.
This shows that p|pn − p. Now, if p2 |d then we would have p2 |pn − p which would breed
a contradiction p2 |p because p2 |pn for n > 1. Therefore, for any prime p|d, p2 can’t
divide d and so d is square-free.
Now, we only need to find the valid values of p. This is where it gets tricky. For
any integer a, either p|a or a ⊥ p. It’s safe to work only for a ⊥ p since the problem
asks for all values of a. In that case, from Fermat’s little theorem, ap−1 ≡ 1 (mod p).
And from the problem statement, we have a(an−1 − 1) is divisible by p. Since a ⊥ p,
we get an−1 − 1 is divisible by p, or an−1 ≡ 1 (mod p). This almost tells us to infer
that we must have p − 1|n − 1. That is the case indeed, however, we have to prove that
p − 1 must be the order of a for some integer a. You should think on this more and get
to the point where you understand: we should set a = g where g is a primitive
root of p. So that, we can tell g p−1 ≡ 1 (mod p) and ordp (g) = p − 1. Therefore, from
g n−1 ≡ 1 (mod p), we get Q p − 1|n − 1.
Notice that, if D = p, then any d|D satisfies the condition of the problem.
p−1|n−1
Therefore the number of such positive integer d is τ (D). If the number of primes p|n
for which p − 1|n − 1 is t(n) i.e.
X
t(n) = 1
p|n
p−1|n−1

then τ (D) = 2t(n) .

CHAPTER 6. SOLVING CHALLENGE PROBLEMS 367

Note. The following approach works too. Consider a⊥d so we have d|an−1 − 1. Using
theorem ( ), we get λ(n)|n − 1.
Problem 6.22. For rational a, b and all prime p, ap − bp is an integer. Prove that, a
and b must be integer.
Solution. Since a, b are rational, we can assume that a = md , b = nd with m ⊥ d, n ⊥ d.
Otherwise, if m 6⊥ d we can divide by the common factor. Moreover, we can assume
m ⊥ n. Indeed, if not, say r is a prime factor of d. Then we must have r 6 | gcd(m, n).
Otherwise the condition m ⊥ d would be broken. Therefore, without loss of generality,
m ⊥ n. Let q be a prime factor of d. Thus,

q p |mp − np

for all p, and e be the smallest positive integer such that

me ≡ ne (mod q)

We can say that e|p for all prime p. But this impossible except for e = 1. Hence,
q|m − n. Now, take a prime p 6= q, and from Exponent GCD lemma we have

gcd (m − n, f (m, n, p)) |p

q 6 |f (m, n, p)

This gives us, q p |m − n for all prime p 6= q which leaves a contradiction inferring
that d can’t have a prime factor i.e. d must be 1. And then, a and b both are integers.
Problem 6.23. Prove that, σ(n) = n + k has a finite number of solutions for a fixed
positive integer k.
√ √
Solution. We can easily show that σ(n) > n + n which says k > n. In other words,
n is bounded above so it can not have arbitrary number of solutions.
Note. We can use some sharper inequalities to get a upper bound of k. Try to sharpen
the inequality as much as you can.
Problem 6.24. Prove that, for a positive integer n,
 
2n
|[1, 2, . . . , 2n]
n

Solution. When proving divisibilities like this, you should consider theorem ( ).
The idea is that if we can show the exponent in the left side of the divisibility is less
than or equal to the exponent in the right side for a prime p, we are done. It is clear
that no side of divisibility will have a prime p > 2n. So we should consider only primes
p < 2n, since 2n is not a prime for n > 1. Using theorem ( ), if for a prime p,
α = logp (2n),

νp ([1, 2, . . . , 2n]) = α
368
2n


On the other hand, from Legendre’s theorem, if N = n
,
∞    
X 2n n
νp (N ) = − 2
i=1
pi pi
logp (2n)    
X 2n n
= − 2
i=1
pi pi
a
X
≤ 1 since ⌊2x⌋ − 2⌊x⌋ ∈ {0, 1}
i=1
=α
νp (N ) ≤ νp ([1, 2, . . . , 2n])

This is exactly what we needed to prove.

Problem 6.25 (Italy TST 2003). Find all triples of positive integers (a, b, p) such
that 2a + pb = 19a .
Solution. Rewrite the equation as pb = 19a − 2a . Right side is divisible by 19 − 2 = 17
which is a prime. Therefore, p = 17, 17b = 19a − 2a . But if a > 1 then 19a − 2a has a
prime factor other than 17, contradiction. Thus, the only solution is (a, b) = (1, 1, ).
np + 1
Problem 6.26 (APMO 2012 - Problem 3). Find all pairs of (n, p) so that
pn + 1
is a positive integer where n is a positive integer and p is a prime number.
Solution. We can re-state the relation as

pn + 1|np + 1

Firstly, we exclude the case p = 2. In this case,

2n + 1|n2 + 1

Obviously, we need
n2 + 1 ≥ 2n + 1 ⇒ n2 ≥ 2n
But, using induction we can easily say that for n > 4, 2n > n2 giving a contradiction.
Checking n = 1, 2, 3, 4 we easily get the solutions:

(n, p) = (2, 2), (4, 2)

We are left with p odd. So, pn + 1 is even, and hence np + 1 as well. This forces n
to be odd. Say, q is an arbitrary prime factor of p + 1. If q = 2, then q|n + 1 and since

np + 1 = (n + 1)(np−1 − .... + 1)

and p odd, there are p terms in the right factor, therefore odd. So, we infer that 2k |n+1
where k is the maximum power of 2 in p + 1.
We will use the following theorem from elementary calculus, which can also be
proved elementarily.
CHAPTER 6. SOLVING CHALLENGE PROBLEMS 369

Theorem 6.27.  n
1
lim 1+ =e
n→∞ n
where e is the Euler constant.
Now, we prove the following lemmas.
Lemma 6.28. If p ≥ 3 is an odd number (not necessarily prime), then pn ≤ np for
p ≤ n.
Proof. We will proceed by induction. The result is true for n = 1. Suppose that
n > 1 is an integer such that pn ≤ np holds for all 3 ≤ p ≤ n. We want to show that
pn+1 ≤ (n + 1)p for all 3 ≤ p ≤ n + 1. If p = n + 1, we have the equality case. So,
suppose that 3 ≤ p < n + 1. Then, since p ≤ n, we have

(pn + p)p ≤ (pn + n)p ,

which gives (after dividing both sides by (np)p )

 p  p
1 1
1+ ≤ 1+ .
n p
Therefore, since we assumed np ≤ pn ,
p
p p 1
(n + 1) = n 1 +
n
 p
n 1
≤p 1+
p
n
≤p ·e
< pn+1 ,

since e < 3.
Back to the problem. Assume that q is odd.

q|pn + 1|np + 1

Write them using congruence. And we have,

np ≡ −1 (mod q)

⇒ n2p ≡ 1 (mod q)
Suppose, e = ordq (n) i.e. e is the smallest positive integer such that

ne ≡ 1 (mod q)

Then, e|2p and e|q − 1 from theorem ( ).

Also, from Fermat’s theorem,

nq−1 ≡ 1 (mod q)
370

Therefore,
n(2p,q−1) ≡ 1 (mod q)
From p odd and q|p + 1, p > q and so p and q − 1 are co-prime. Thus,

(2p, q − 1) = (2, q − 1) = 2

This gives us, e|(2p, q − 1) and so we must have e = 2. Again, since p odd, if p = 2r + 1,

n2r+1 ≡ n (mod q)
p
Hence, q|n + 1. If q| nn+1
+1
, then by the theorem ( ) we get
 
np + 1
q| gcd n + 1, |p
n+1

which would imply q = 1 or p. Both of the cases are impossible. So, if s is the maximum
power of q so that q s |p + 1, then we have q s |n + 1 too for every prime factor q of p + 1.
This leads us to the conclusion p + 1|n + 1 or p ≤ n which gives pn ≥ np by lemma 2.5.
But from the given relation,

pn + 1 ≤ np + 1 ⇒ pn ≤ np

Combining these two, p = n is the only possibility to happen. Thus, the solutions are
(n, p) = (4, 2), (p, p).

Problem 6.29. Prove that aϕ(n) (aϕ(n) − 1) is always divisible by n for all positive
integers a and n.

Solution. Let p be any prime divisor of n. Suppose that that pk |n but pk+1 ∤ n, where
k ≥ 1 is an integer. Consider two possible cases:

1. p ∤ a. Note that by Corollary ( ), since pk |n, we have ϕ(pk )|˚ϕ(n). Then by

Euler’s theorem and Theorem ( ),
k
pk | aϕ(p ) − 1 | aϕ(n) − 1.

2. p|a. We know that ϕ(pk ) = pk−1 (p − 1), which is clearly bigger than k. Thus
k) n
pk | ak | aϕ(p | aϕ(a ) .

Therefore in both cases we have pk |aϕ(n) (aϕ(n) − 1) which results in n|aϕ(n) (aϕ(n) − 1).

Problem 6.30 (Serbian Mathematical Olympiad 2014, Day 2). We call a natural
number n nutty if there exist natural numbers a > 1 and b > 1 such that n = ab + b.
Do there exist 2014 consecutive natural numbers, exactly 2012 of which are nutty?
CHAPTER 6. SOLVING CHALLENGE PROBLEMS 371

Solution. Let’s say n is nutty for (a, b) if n = ab + b for a, b > 1. We need 2014
consecutive natural numbers with b > 1 among which 2012 will be nutty. Let say those
numbers are a1 + 1, a2 + 2, . . . , a2014 + 2014. Crucial observation: If n is nutty for a, b
then n is nutty for ak , b where k is a positive integer. Since there are integers
1, 2, 3, . . . , 2014 associated and we are free to choose a as long as a > 1, it makes sense
to take the exponent 2014!. But we can not change b, therefore, we must introduce this
factorial within a. So, let’s take a positive integer x > 1 and look at the numbers

x2014 ! + 1, x2014! + 2, . . . , x2014! + 2014

 2014! 2  2014! 2014
x2014! + 1, x 2 + 2, . . . , x 2014 + 2014

All of them are nutty except the first one (probably). But we need exactly two to be
not nutty. A way to do that is to divide 2014! by a number and hope it doesn’t remain
nutty. Let’s say this number is k, that is, we are looking at the numbers
2014!
 2014! 2  2014! 2014
x k + 1, x 2k + 2, . . . , x 2014k + 2014

We can see that there are only two numbers which can be candidates to be not nutty.
2014!
 2014! k
x k + 1 and x k2 + k. Essentially we don’t want the case k 2 divides 2014!. This is
a hint to using primes! If we take a prime p > 1007, p2 ∤ 2014!. Now we that have 2012
numbers nutty for sure, we just need to find x, p so that those two numbers become not
nutty.
2014!
x p + 1 = ab + b
2014!
x p + p = ab + b

Take N = 2014!
p
. If we find x, p so that the equations xN + 1 = ab + b and xN + p = ab + b
don’t have solutions, we are done. Let’s try with x = 2 since a ≥ 2, we might be able
to use inequalities. For b ≥ N ,

2N + p = ab + b
≥ 2b + b
≥ 2N + N
> 2N + p

So b < N . This also means we are in the right track to solve the problem. And b > p
too must hold. Otherwise,

p − b = ab − 2N
 N b
= ab − 2 b
 N
 N (b−1)

= a−2b ab−1 + . . . + 2 b
>p
0>b
372

contradiction. Now , if a is even, let a = 2u.

b − p = 2N − 2 b ub

Since b < N , 2b |b − p but clearly b − p < 2b . Thus, a is odd. If b is even with b = 2v,
 N
2
b−p= 2 − (2v )2
2

 N  N 
= 2 2 + 2 v 2 2 − 2v
>b

again contradiction. Try to do the same with 2N + 1.

Problem 6.31 (Columbia 2010). Find the smallest n ∈ N such that n! is divisible
by n10 .

Solution. Let p be a prime divisor of n and p-base representation of n is n = ak pk +

. . . + a1 p + a0 . Then

n − (ak + . . . + a0 )
νp (n!) =
p−1

If νp (n) = α then the last α digits of n in base p is 0.

n − (ak + . . . + aα )
νp (n!) = and
p−1
νp (n10 ) = νp (p10α ) = 10α

We must have νp (n!) ≥ νp (n10 ) = 10α.

n − (ak + . . . + aα )
≥ 10α
p−1

where α ≤ logp (n). Remember that, we are looking for the smallest n, not all n.
Therefore, we should first consider the case n = pα first. Then In base p,

n = (1 0| .{z
. . 0})p
α zeros

pα − 1
νp (n!) = and so,
p−1
pα − 1
≥ 10α
p−1

We need to find such p, α such that pα is minimum. Therefore, p must be 2. Now we

need the smallest α for which 2α −1 ≥ 10α. See that α = 6 works. But can we minimize
n even further? Since n! must be divisible by n10 , so we need to look at primes for
CHAPTER 6. SOLVING CHALLENGE PROBLEMS 373

which we get the exponent at least 10. If p2 |n for some prime p, we need νp (n!) ≥ 20
for some n < 64. The minimum n for which n! is divisible by 310 is 24. For n < 64,
 
63 63 63
ν3 (n!) ≤ + +
3 9 27
= 21 + 7 + 2 = 30
 
63 63
And we see that n = 63 is in fact a solution because ν7 (63!) = + = 10. The
7 49
minimum n for which ν5 (n!) ≥ 10 is n = 45. Using the arguments we made already,
show that we can not minimize n further.
Problem 6.32 (Greece National Mathematical Olympiad, 2015). Find all triplets
(x, y, p) of positive integers such that p is a prime number and

xy 3
=p
x+y
Solution. This equation does not suggest it, but we will take (x, y) = g and x =
ga, y = gb where a⊥b. Because that way, we can reduce the equation or extract more
information.
gag 3 b3
=p
g(a + b)
g 3 ab3 = p(a + b)

Now this equation talks more than the previous one. Since a⊥b, a + b⊥a, a + b⊥b and
b3 ⊥a + b. Therefore, a|p(a + b) gives us a|p and b3 |p(a + b) gives b3 |p. From this, b = 1
since b = p is not possible. If a = p, then

g 3 b3 = b + p
p = b(g 3 b2 − 1)
= g3 − 1

This equation gives p is divisible by g − 1, which is not possible unless g = 2. Then

a = p = 7 and x = ga = 14, y = gb = 2. We are left with a = 1 which gives g 3 = 2p
but this is not possible since right side can not be a cube.
Problem 6.33 (Korea 2010). A prime p is called a nice prime if there exists a sequence
of positive integers (n1 , . . . , nk ) satisfying following conditions for infinitely positive
integers k, but not for k = 1.
p+1
• For 1 ≤ i ≤ k, ni ≥ .
2
pni − 1
• For 1 ≤ i ≤ k, pni − 1 is a multiple of ni+1 and is co-prime to ni+1 . Set
ni+1
nk+1 = n1 .
Show that 2 is not a nice prime, but any odd prime is.
374

Solution. Let’s deal with the case p = 2 first. We will show that there does not exist
infinitely many k for which there exist n2 , . . . , nk with ni+1 |2ni − 1. For i > 1, ni is
odd, and fix k. Consider all prime factors of ni for i > 1, let the smallest of them be q.
If q|nl Then 2nl−1 ≡ 1 (mod q) and 2q−1 ≡ 1 (mod q). If d = ordq (2), then we can say
d|q − 1 and d|nl−1 . If d 6= 1, then we get a smaller divisor than q since d ≤ q − 1. This
shows that d = 1 but then q|21 − 1 = 1, contradiction.
Now we consider p ≥ 3. We have to take care of the k = 1 case first, since it
was explicitly mentioned, and it seems the easier part. We need to prove that, for no
positive integer n,

n|pn − 1 and
 
pn − 1
n, =1
n

does not hold. Again, we assume that q is the smallest prime divisor of n. Clearly
q 6= p, and q⊥p.

pn ≡ 1 (mod q)
pq−1 ≡ 1 (mod q) from FLT
p(n,q−1) ≡ (mod q)
p ≡ 1 (mod q)

since n ⊥ q − 1 < q. From this, you should be able to tell, you can invoke LTE!
Because q|p − 1, from LTE,
 
pn − 1
νq = νq (pn − 1) − νq (n)
n
= νq (p − 1) + νq (n) − νq (n)
= νq (p − 1)
≥1

pn − 1
Therefore q divides and q|n, contradiction. Such n does not exist for k = 1.
n
We just have to finda construction for such (n1 , n2 , . . . , nk ). First think on the
ni
p −1
condition ni+1 , = 1. This clearly means for any odd prime divisor q of ni+1 ,
ni+1
pni − 1
q does not divide .
ni+1
 
pni − 1
νq =0
ni+1
νq (pni − 1) − νq (ni+1 ) = 0
4
we already argued the same way before
5
we won’t show the case q = 2 here, do that yourself in the same fashion
CHAPTER 6. SOLVING CHALLENGE PROBLEMS 375
p+1
This suggests us to take n1 so that n1 = p − 1 and p − 1 ≥ is obvious for odd
2
p. Then we have n2 |p p−1
− 1. If q ||p − 1, then q ||p
e 2e p−1
− 1 according to LTE. Then
p−1
p −1
q ∤ 2
for any q|p − 1. So we should include n2 = q 2e . You should understand
q
that we have to define n3 the same way. Let

ei = νq (pni − 1) then
ni+1 = q ei for q|p − 1
p+1
The only thing left to do is ensure that ni ≥ . Note that the sequence (ei )i≥1 is
2
p+1
increasing for a fixed q. Therefore, there will be an index r for which nr ≥ must
2
hold. We leave it to the reader to verify that they satisfy the conditions of the problem.
Problem 6.34 (IMO Shortlist 2013, Problem 4, Proposed by Belgium). Prove
that there exist infinitely many positive integers n such that the largest prime divisor
of n4 + n2 + 1 is equal to the largest prime divisor of (n + 1)4 + (n + 1)2 + 1.
Solution. Let an = n4 + n2 + 1, and pn be the largest prime divisor of an . The problem
asks to prove there are infinite n for which pn = pn+1 .

(6.1) (n4 + n2 + 1) = (n2 + n + 1)(n2 − n + 1)

This identity tells us to consider the numbers bn = n2 + n + 1, and the largest prime
divisor of bn is qn . Then an = bn2 and pn = qn2 . From equation ( ), pn = max(qn , qn−1 )
since bn−1 = (n − 1)2 + n − 1 + 1 = n2 − n + 1. By Euclidean Algorithm

(n2 + n + 1, n2 − n + 1) = (n2 − n + 1, 2n)

= (n2 − n + 1, 2) since n⊥n2 − n + 1
= 1 since n(n − 1) + 1 is odd

Therefore bn ⊥bn−1 , implying qn 6= qn−1 . Up until now, all we have done was self
implicating. But what will allow us to prove that pn = pn+1 happens for infinite n?
pn = max(qn , qn−1 ) and pn+1 = max(qn+1 , qn ). The problem requires pn = pn+1 or
max(qn−1 , qn ) = max(qn+1 , qn ). This provides us a hint what we need to do. We will
focus on qn . That means, we will try to prove that for infinite n, qn = max(qn , qn−1 ) and
qn = max(qn+1 , qn ). In short, we have to prove qn > qn+1 and qn > qn−1 holds true for
infinite n. First we need to check that at least one such qn exists. q2 = 7, q3 = 13, q4 = 7,
so n = 3 gives us such a qn .
Assume to the contrary that, only for finite n, qn > qn−1 and qn > qn+1 . Then
there is a largest value of n for which this condition holds true, say it is N . Let’s think
if it is possible that qi > qi+1 for all i ≥ N . But that would give us an infinite set
of decreasing positive integers, which is impossible. So there is an i > N for which
qi < qi+1 (remember that qi 6= qi+1 ). Then is it possible to have an infinite chain of
qi < qi+1 < qi+2 < . . .? No, because q(i+1)2 = pi+1 = max(qi+1 , qi ) = qi+1 . Therefore,
we must have an j for which qj > qj+1 . For that j, we have qj > qj−1 , so it is a
contradiction. Thus, there are infinite such n.
376

Problem 6.35. Let p be an odd prime. If g1 , · · · , gφ(p−1) are the primitive roots
Pφ(p−1)
(mod p) in the range 1 < g ≤ p − 1, prove that i=1 gi ≡ µ(p − 1) (mod p)

Solution.
(p−1)/d
X X X
gi = µ(d) g kd = µ(p − 1).
i d|p−1 k=1
P(p−1)/d
Because k=1 g kd = 0 mod p, when d < p − 1.

Problem 6.36 (IMO Shortlist 2014, N4, Proposed by Hong Kong, also used
at Bangladesh TST 2015). Let n be a given integer. Define the sequence (ak )k≥1 by:
 k
n
ak =
k

Prove that this sequence has infinitely many odd terms.

nk
Solution. If n is odd, then we set k = ni , so that k divides nk and ak = is an odd
k
integer. Now we concentrate on even n.
When n > 2 is even, a prime divisor p of n − 1 is odd. Take p|n − 1 then from LTE,
 k 
νp np − 1 = νp (n − 1) + νp (pk )
= νp (n − 1) + k
k
pk np − 1
Thus, p divides n
k
− 1, so is an integer. We have
pk
$ j%
np
apj =
pj
j
np − 1
=
pj
 k
2
which is an odd integer. If n = 2, then ak = . Note that, 2m − 1 is divisible by 3
k
i
for even m. Therefore, we should consider k = 2 · 3i . From LTE, 3i+1 ||22·3 − 1. But if
k is even it won’t divide any odd integer. No worries, we will just borrow a power of 2.
$ j
%
23·4
a3·4j =
3 · 4j
j
23·4 − 4j
=
3 · 4j
j
is an odd integer since 4j clearly divides 23·4 and 3 · 4j > j. Show why 3 divides the
numerator yourself.
CHAPTER 6. SOLVING CHALLENGE PROBLEMS 377

Problem 6.37 (IMO Shortlist 2014, N5, Proposed by Belgium, also used at
Bangladesh TST 2015). Find all triples (p, x, y) consisting of a prime number p and
two positive integers x and y such that xp−1 + y and x + y p−1 are both powers of p.

Solution. Let y + xp−1 = pa and x + y p−1 = pb and y > x without loss of generality (so
b > a too). If p = 2, then x+y = 2a so (x, y) = (x, 2a −x) works with any a ∈ N, x < 2a .
We need to deal with odd p now. Let g = (x, y) and x = gm, y = gn with m⊥n.
We intend to prove that x⊥y. If not, g = ps for some s.

ps n + ps(p−1) mp−1 = pa
ps m + ps(p−1) np−1 = pb

If s 6= 0, the dividing the equations by ps we get,

n + ps(p−2) mp−1 = pa−s

m + ps(p−2) np−2 = pb−s

Since p > 2, p divides ps(p−2) , pa−s and pb−s , hence p divides both m and n. But this
contradicts the fact that m⊥n. Therefore, s = 0, and (x, p) = (y, p) = (x, y) = 1. From
FLT,

x + y p−1 ≡ x + 1 (mod p)
xp−1 + y ≡ y + 1 (mod p)

Thus, p|x + 1 − (y + 1) = x − y. You must understand by now that we are just trying
to set this up for applying LTE. Without loss of generality, let’s take a < b.

pa |x + y p−1
pa |y + xp−1
pa |x(xp−1 + y) − y(x + y p−1 )
pa |xp − y p

Since x⊥y and p|x − y, if pα ||x − y, using LTE,

νp (xp − y p ) = νp (x − y) + νp (p)
=α+1

We get α + 1 ≥ a, so pa−1 |x − y. Since y > x, assume that y − x = pa−1 k.

xp−1 + y = pa
y − x = pa−1 k
xp−1 + y − (y + x)pa − pa−1 k
xp−1 + x = pa−1 (p − k)
x(xp−2 + 1) = pa−1 (p − k)
378

Because x⊥p, x|p − k which implies p − k ≥ x or p ≥ x + k ≥ x + 1. On the other

hand, we had p|x + 1 or x + 1 ≥ p. Combining these two, x + 1 = p or x = p − 1 and
k = 1. Finally,

y = x + pa−1
= p − 1 + pa−1

But from the given condition,

xp−1 + y = pa
(p − 1)p−1 + pa−1 + p − 1 = pa
(p − 1)p−1 + p − 1 = pa−1 (p − 1)
(p − 1)p−2 + 1 = pa−1
(p − 1)p−2 = pa−1 − 1

Since p is odd, if a − 1 > 1, then according to Zsigmondy’s theorem, pa−1 − 1 has a

prime divisor that does not divide p − 1. So a − 1 = 1 or a = 2 and p = 3. Thus, x = 2
and y = x + pa−1 = 5. By symmetry, (5, 2) is also a solution.

Problem 6.38 (Russia 2000). Do there exist three distinct pairwise co-prime integers
a, b, c such that a|2b + 1, b|2c + 1 and c|2a + 1?

Solution. Let p be the smallest prime divisor of a. Then p|2b + 1, so 2b ≡ −1 (mod p).

22b ≡ 1 (mod p)
2p−1 ≡ 1 (mod p)
2(2b,p−1) ≡ 1 (mod p)

Without loss of generality, we can assume that p < q, r where q and r are the smallest
prime divisors of b and c (even if it is not, we can just switch the places of p and q,
or r). Therefore, (2b, p − 1) = 2 and p|22 − 1 = 3 so p = 3. Let a = 3x. We can see
that 3 ∤ x, because otherwise 9 divides 2b + 1, which would be possible only if b = 3y
meaning that (a, b) ≥ 3.
Again we are out of information. In order to dig out some information, let’s take
the smallest prime divisor of x, b and c and call it q. If q|x, then q|2b + 1 and again q
is smaller than the smallest prime divisor of b and c. But this again gives that q = 3,
which gives the contradiction that x is divisible by 3. Thus, we can say q divides b or
c. If q|c, then

23x ≡ −1 (mod q)
82x ≡ 1 (mod q)
8q−1 ≡ 1 (mod q)
8(2x,q−1) ≡ 1 (mod q)

Here, x must have no smaller prime divisor than q, otherwise that would have been the
smallest prime divisor instead of q. So, (2x, q − 1) = 2 and q|82 − 1 = 32 · 7. Since q is
CHAPTER 6. SOLVING CHALLENGE PROBLEMS 379

co-prime to a = 3x, q 6= 3. So q = 7 but then 7|2a + 1 = 8x + 1, which is a contradiction

due to

8x + 1 ≡ 1 + 1 (mod 7)

Thus, q must be the smallest prime divisor of b, and so c does not have any prime
divisor less than or equal to q. Similarly,

2c ≡ −1 (mod q)
22c ≡ 1 (mod q)
2q−1 ≡ 1 (mod q)
2(2c,q−1) ≡ 1 (mod q)
22 ≡ 1 (mod q)

which gives us q = 3 but then 3|b, a contradiction. Therefore, b can not have any prime
divisor, b = 1. We also have c|23 + 1 = 9, so c ∈ {1, 3, 9}. But if c = 1, it would collide
with b = 1. If c = 3 or c = 9, it would not be co-prime with a. Therefore, no such
a, b, c exist.
Problem 6.39 (Austria 2010). Let

f (n) = 1 + n + . . . + n2010

For every integer m with 2 ≤ m ≤ 2010, there is no nonnegative integer n such that
f (n) is divisible by m.
Solution. We can write f (n) as

n2011 − 1
f (n) =
n−1
Let p be a prime factor of f (n). Then if p|n − 1, we have n ≡ 1 (mod p). Take
d = ordp (n). If p⊥(n − 1), then

n2011 ≡ 1 (mod p)
nd ≡ 1 (mod p)

We have d|2011, if d = 1.

1 + n + . . . + n2010 ≡ 1 + 1 + . . . + 1 (mod p)
f (n) ≡ 2011 (mod p)

This gives us p|2011. Moreover, from FLT, we also have

np−1 ≡ 1 (mod p)

so d|p − 1. If d = 2011, then p ≡ 1 (mod 2011). But for 1 < m < 2011, there is
no m for which m has a prime divisor p so that 2011|p or p ≡ 1 (mod 2011) because
p ≥ 2011 > m.
380

Remark. We can make a general result. Every prime divisor q of

f (n) = 1 + n + . . . + np−1

must be either p or q ≡ 1 (mod p), where p is a prime.

Problem 6.40 (APMO 2014, Problem 3). Find all positive integers n such that for
any integer k there exists an integer a for which a3 + a − k is divisible by n.
Solution. Make sense of the problem before you try it. It can be rephrased this way:
find all n such that the set a3 + a for a = 1, 2, . . . , n we get a complete residue class
modulo n. Or, for no two 1 ≤ a < b ≤ n,

a3 + a ≡ b3 + b (mod n)

In order to understand the values of n, see some examples with smaller values first. A
pattern follows, n = 1, 3, 9 works. So may be the condition holds if and only if n = 3k .
Checking the if part is easy.

a3 + a ≡ b3 + b (mod 3k )
a3 − b3 + a − b ≡ 0 (mod 3k )
(a − b)(a2 + ab + b2 + 1) ≡ 0 (mod 3k )

For a 6≡ b (mod 3), we can see that 3 ∤ a2 + ab + b2 + 1. So we must have a ≡ b

(mod 3k ). Now we need to prove that if n 6= 3k , the condition does not hold. If n has
a prime divisor p for which the condition doesn’t hold, then the same is true for n as
well, so let’s just look at the primes.
If p ≡ 1 (mod 4) is a prime, we know that −1 is a quadratic residue of p, so there
is an x for which

x2 ≡ −1 (mod p)
x3 + x ≡ 0 (mod p)

If we choose y = 0,

x3 + x ≡ y 3 + y (mod p)

So, n can not have any prime factor ≡ 1 (mod 4). Let p ≡ 3 (mod 4) be a prime. We
must have

(13 + 1) · (23 + 2) · · · ((p − 1)3 + p − 1) ≡ 1 · 2 · · · (p − 1) (mod p)

p−1
Y
1 · 2 · · · (p − 1) (k 2 + 1) ≡ 1 · 2 · · · (p − 1) (mod p)
k=1
p−1
Y
(k 2 + 1) ≡ 1 (mod p)
k=1
p−1
Y
(k + i)(k − i) ≡ 1 (mod p)
k=1
CHAPTER 6. SOLVING CHALLENGE PROBLEMS 381

Note that, using Lagrange’s theorem, we can imply,

p−1 p−1
Y Y
(k + i) (k − i) ≡ 1 (mod p)
k=1 k=1
p−1 p−1
(k − 1)(k − 1) ≡ 1 (mod p)
p−1 2 p−1
(k ) − 2k +1≡1 (mod p)
2 − 2k p−1 ≡ 1 (mod p)

Here, k p−1 ≡ ±1 (mod p), so we have 2 − (±2) ≡ 1 (mod p) or p = 3. Hence, proven.

Problem 6.41 (Croatia 2014). Do there exist positive integers m and n such that
m2 + n and n2 + m are squares of positive integers?
This is a problem where we use another technique, we will call it squeezing between
squares. The name is due to Dan Schwarz, who was a problem solver and creator from
Romania, and a user on Art of Problem Solving. He used the handle mavropnevma and
taught many people many things through the forum. In a solution of a similar problem,
he used the words squeezing between squares. So, out of respect we named this.
Solution. Due to symmetry, we can assume m ≥ n without loss of generality.

m2 + n ≥ m2 + m > m2 and
m2 + 2m + 1 > m2 + m so
m2 < m2 + m < (m + 1)2

This says that m2 + m resides between two squares, so it can not be a perfect square.
Therefore, no such (m, n).
Problem 6.42 (Turkey 2011). Let an+1 = a3n − 2a2n + 2 for all n ≥ 1 and a1 = 5.
Prove that if p ≡ 3 (mod 4) and p is a divisor of a2011 + 1, then p = 3.
Solution. The recursion is a cheeky one.

an+1 − 2 = a2n (an − 2)

= a2n a2n−1 (an−1 − 2)
..
.
= a2n a2n−1 · · · a2 (a1 − 2)
= 3a22 · · · a2n
an+1 + 1 = 3(a22 · · · a2n + 1)

If p|an+1 +1, then p divides 3 or p divides a22 · · · a2n +1. But we know that p ≡ 3 (mod 4)
can not divide a bisquare. Thus, p must divide 3, or p = 3.
Problem 6.43 (IMO Shortlist 2004, N2). For n ∈ N let,
n
X
f (n) = (i, n)
i=1
382

(a) Prove that f (mn) = f (m)f (n) for all m⊥n.

(b) Prove that, for all a, there is a solution to f (an) = an.

Solution. Since we are required to prove that f is multiplicative, it may be better if

f (n) can be written in terms of divisors of n. Since we are facing this
Pkind of problem
for the first time, let’s show a manipulation. It is clear that f (10) = 10
i=1 (i, 10), which
is

= (1, 10) + (2, 10) + (3, 10) + (4, 10) + (5, 10) + (6, 10)
+ (7, 10) + (8, 10) + (9, 10) + (10, 10),

and hence,

f (10) = 1 + 2 + 1 + 2 + 5 + 2 + 1 + 2 + 1 + 10
= 1 · 4 + 2 · 4 + 5 · 1 + 10 · 1.

First of all, 1, 2, 5, 10 are divisors of 10. Since (i, n) will be a divisor of n, it makes sense.
Now we just need to figure out how to determine those numbers 4, 4, 1, 1 for 1, 2, 5, 10
respectively. Let’s think about the case when divisor, d = 2. We want (i, 10) = d so we
can write i = dj and 10 = dm where (j, m) = 1. Since we are counting all such possible
j, it is simply the number of j less than or  equal to m, which are co-prime to m. In
10
other words, the coefficient of i is ϕ(m) = ϕ . Now, let’s prove it formally.
d
n
X n
X m
XX
f (n) = (i, n) = (dj, dm) = d(j, m)
i=1 i=1 d|n j=1
j⊥m
X X n Xn
= dϕ(m) = dϕ = ϕ(d)
d d
d|n d|n d|n
X ϕ(d)
=n
d
d|n

Let’s concentrate on proving the claims now.

(a) Let m⊥n.
X  mn 
f (mn) = dϕ
d
d|mn
  X m n
X mn
= ef ϕ = ef ϕ ϕ since n⊥m
d=ef
ef e f
e|m
e|m f |n
f |n
X m X n
= eϕ f
e f
e|m f |n

= f (m)f (n)
CHAPTER 6. SOLVING CHALLENGE PROBLEMS 383

We could also prove it using Dirichlet product. If f (n) = n and g(n) = ϕ(n), then
both f and g are multiplicative. So, their Dirichlet product would be multiplicative
as well.
X n
f ∗g = f (d)g
d
d|n
X n
= dϕ
d
d|n

must be multiplicative then!

(b) f (an) = an means

X  an 
dϕ = an
d
d|an

But clearly the sum on the left side is hard to deal with. It would be better if we
could have another representation, probably a closed form instead of a summation.
Let’s determine f (pi ) first.
e
X ϕ(d) X ϕ(pi )
f (pe ) = pe = pe
d pi i=0
d|pe
e
!
X pi−1 (p − 1)
= pe 1+ i
since ϕ(1) = 1
i=1
p
e
!  
X p−1 e(p − 1)
= pe 1+ e
=p 1+
i=1
p p

From the given condition, f (n) = an, so for n = pe , it implies

 
e e(p − 1)
p 1+ = ape ,
p
and so,
e(p − 1)
1+ = a.
p
e(p−1)
Here, a is integer, so p
is an integer too. Since p ⊥ p − 1, we must have p|e.
Let e = pk so that

a = 1 + k(p − 1).

If this has to be true for any a, we should assume p = 2, and we get a = 1 + k or

k = a − 1, e = 2(a − 1). This gives us an infinite solution for n = 22(a−1) .
Problem 6.44 (IMO 2006). Find all integers x and y which satisfy the equation

1 + 2x + 22x+1 = y 2 .
384

Solution. Obviously, x ≥ 0. If x = 0, then y = ±2. Suppose that x ≥ 1. Clearly, y is

an odd number, so assume that y = 2k + 1. Now:

1 + 2x + 22x+1 = (2k + 1)2 = 4k 2 + 4k + 1.

Removing 1 from both sides and then dividing by 4, one can write the above equation
as

(6.2) 2x−2 (2x+1 + 1) = k(k + 1).

We consider two cases:

1. k is even. So k + 1 is odd and (2x−2 , k + 1) = 1. Then 2x−2 |k(k + 1) reduces to
2x−2 |k. Let k = 2x−2 · t and rewrite equation ( ) to achieve

2x+1 + 1 = t(2x−2 · t + 1).

So, t is odd. If t = 1, we have 2x+1 = 2x−2 , which is absurd. Thus t ≥ 3 and

t(2x−2 · t + 1) = 2x−2 · t2 + t ≥ 9 · 2x−2 + 3 > 2x+1 + 1

and no solutions in this case.

2. k is odd. So k + 1 is even and gcd(2x−2 , k) = 1. Then 2x−2 |k(k + 1) reduces to

2x−2 |k + 1. Let k + 1 = 2x−2 · m and rewrite equation ( ) as

2x+1 + 1 = m(2x−2 · m − 1)

So m is odd. m = 1 gives no solutions, so m ≥ 3. For m = 3, we have

2x+1 + 1 = 3(2x−2 · 3 − 1) =⇒ 4 = 9 · 2x−2 − 2x+1

=⇒ 4 = 2x−2 (9 − 8)
=⇒ x = 4.

Thus for m = 3, we have k = 2x−2 · m − 1 = 11 and so y = 2k + 1 = 23, and

(x, y) = (4, 23) is a solution. If (x, y) is a solution, obviously (x, −y) is a solution
too. So (x, y) = (4, −23) is also a solution.
We will show that m ≥ 5 gives no solutions. Note that

m(2x−2 · m − 1) ≥ 5(2x−2 · 5 − 1)
= 25 · 2x−2 − 5
= (8 + 16 + 1) · 2x−2 − 5
= 2x+1 + 2x+2 + 2x−2 − 5
> 2x+1 + 1.

So m(2x−2 · m − 1) > 2x+1 + 1 for m ≥ 5 and no solutions.

Hence the solutions are: (x, y) = {(0, 2), (0, −2), (4, 23), (4, −23)}.
Chapter 7

Practice Challenge Problems

σ(n)
Problem 7.1. Show that the ratio can be arbitrarily large for infinitely many n.
n
This PDF was created
for Amazon.

Problem 7.2. Prove that for all positive integers n, there exists an n-digit prime.
Problem 7.3. There are n points on a circle with n > 10, and each point is given a
number that is equal to the average of the numbers of its two nearest neighbors. Show
that all the numbers must be equal.
Problem 7.4. Let F (n) be the nth Fibonacci number, show that for some n > 1, F (n)
ends with 2007 zeros.
Problem 7.5. Let N = .23571113... where N consists of all prime numbers concate-
nated together after the decimal. Determine if N is rational or irrational.
Problem 7.6. Prove that there exists a positive integer n such that the four leftmost
digits of the decimal representation of 2n is 2007.
Problem 7.7. The only sets of N − 1 (distinct) integers, with no non-empty subset
having its sum of elements divisible by N , are those where all integers are congruent to
a same residue modulo N , coprime with N .
Problem 7.8 (Austrian Mathematical Olympiad, 2016). Determine all composite
positive integers n with the following property: If 1 = d1 < d2 < · · · < dk are the
divisors of n then

d2 − d1 : d3 − d2 : · · · : dk − dk−1 = 1 : 2 : · · · k − 1

Problem 7.9 (Belarus 2009). Find all m, n ∈ N such that m! + n! = mn .

Problem 7.10. Integer n > 2 is given. Find the biggest integer d, for which holds, that
from any set S consisting of n integers, we can find three different (but not necessarily
disjoint) nonempty subsets, such that sum of elements of each of them is divisible by d.
Problem 7.11. Consider the setM = {1, 2, 3, ..., 2007}. Prove that in any way we
choose the subset X with 15 elements of M there exist two disjoint subsets A and B in
X such that the sum of the members of A is equal to the sum of the members of B.

385
386

Problem 7.12 (India 2014). Let n ∈ N. Show that,

jnk jnk jnk √
+ + ... + + ⌊ n⌋
1 2 n
is even.

Problem 7.13. Given 101 distinct non-negative integers less than 5050 show that one
can choose four a, b, c, d such that a + b − c − d is a multiple of 5050.

Problem 7.14 (Bulgarian Mathematical

m
n
Olympiad, 2016). Find all positive in-
tegers m and n such that 22 + 1 22 + 1 is divisible by mn.

Problem 7.15 (Slovenia 2010). Find all prime numbers p, q and r such that p > q > r
and the numbers p − q, p − r and q − r are also prime.

Problem 7.16 (Croatia Mathematical Olympiad, First Round, 2016). Deter-

mine the number of positive integers smaller than 1000000, that are also perfect squares
and give a remainder 4 when divided by 8.

Problem 7.17. Prove that among 81 natural numbers whose prime divisors are in the
set {2, 3, 5} there exist four numbers whose product is the fourth power of an integer.

Problem 7.18. We chose n + 2 numbers from set {1, 2, ...3n}. Prove that there are
always two among the chosen numbers whose difference is more than n but less than
2n.

Problem 7.19 (India 2014). Let a, b be natural numbers with ab > 2. Suppose that
the sum of their greatest common divisor and least common multiple is divisble by
a+b
a + b. Prove that the quotient is at most . When is this quotient exactly equal to
4
a+b
?
4
Problem 7.20. The integers 1, ..., n are arranged in any order. In one step any two
neighboring integers may be interchanged. Prove that the initial order can never be
reached after an odd number of steps.

Problem 7.21. A palindrome is a number or word that is the same when read forward
and backward, for example, 176671 and civic. Can the number obtained by writing the
numbers from 1 to n in order (for some n > 1) be a palindrome?

Problem 7.22 (IMO Shortlist N2, Proposed by Jorge Tipe, Peru). A positive
integer N is called balanced, if N = 1 or if N can be written as a product of an even
number of not necessarily distinct primes. Given positive integers a and b, consider the
polynomial P defined by P (x) = (x + a)(x + b).

(a) Prove that there exist distinct positive integers a and b such that all the number
P (1), P (2),. . ., P (50) are balanced.

(b) Prove that if P (n) is balanced for all positive integers n, then a = b.
CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 387

Problem 7.23. Let n be an integer. Prove that if the equation x2 + xy + y 2 = n has

a rational solution, then it also has an integer solution.

Problem 7.24 (Iran Olympiad, Thrd Round). Let p be a prime number. Prove
that, there exists integers x, y such that p = 2x2 + 3y 2 if and only if p ≡ 5, 11 (mod 24).

Problem 7.25 (Polish Math Olympiad). Let S be a set of all positive integers
which can be represented as a2 + 5b2 for some integers a, b such that a⊥b. Let p be a
prime number such that p = 4n + 3 for some integer n. Show that if for some positive
integer k the number kp is in S, then 2p is in S as well.

Problem 7.26. Prove that the equation x3 − x + 9 = 5y 2 has no solution in integers.

Problem 7.27 (India TST). On the real number line, paint red all points that
correspond to integers of the form 81x + 100y, where x and y are positive integers.
Paint the remaining integer point blue. Find a point P on the line such that, for every
integer point T , the reflection of T with respect to P is an integer point of a different
colour than T .

Problem 7.28. Prove that, for any positive integer k, there are positive integers a, b >
1 such that
a2 + b2 − 1
k=
ab
Problem 7.29. Let a, b, and c be positive integers such that 0 ≤ a2 + b2 − abc ≤ c.
Prove that a2 + b2 − abc is a perfect square.

Problem 7.30. Let the nth Lemur set, Ln , be the set composed of all positive integers
that are equal to the sum of the squares of their first n divisors. For example, L1 = {1},
L2 = {}, and L4 = {130}.

a Find L3 , L5 , and L6 .

b Describe all n for which Ln is empty.

c Describe all n for which Ln is infinite.

d Provide a method for finding members of non-empty Lemur sets.

Problem 7.31 (IMO 2007, Problem 5). Let a and b be positive integers so that
4ab − 1 divides (4a2 − 1)2 . Prove that a = b.

Problem 7.32. Let a1 , a2 , . . . , an be positive integers such that a1 < a2 < · · · < an .
Prove that
n−1
X 1
<1
i=1
[ai , ai−1 ]

Problem 7.33. Let N = 2p1 ···pn + 1 where pi are distinct primes greater than 2 and
τ (N ) is the number of divisors of N . Maximize τ (N ).
388

Problem 7.34 (Masum Billal). Let n ≥ 3, a, d be positive integers so that a, a +

d, . . . , a + (n − 1)d are all primes. If λ(n) is the number of primes strictly less than n,
prove that,

N = 2⌊ 2 ⌋ + 1
d

λ(n)−2 −1
has at least 22 divisors.
Problem 7.35. For n ≥ 2,
Fn+m −1

Fn 2
≡1 (mod Fn+m )

Problem 7.36. Let f (x) = x3 + 17. Prove that for each natural number n ≥ 2, there
is a natural number x for which f (x) is divisible by 3n but not 3n+1 .
Problem 7.37 (Boylai). Show that every Fermat prime is of the form 6k − 1.
Problem 7.38 (Iran TST 2015). We are given three natural numbers a1 , a2 , a3 . For
n ≥ 3,

an+1 = [an−1 , an ] − [an−1 , an−2 ]

Prove that there exists an index k ≤ a3 + 4 such that ak ≤ 0.

Problem 7.39 (Bosnia Olympiad 2013, Second Day). Find all primes p and q
such that p|30q − 1 and q|30p − 1.
Problem 7.40 (IMO Shortlist 2004, N3, Proposed by Iran). f is a function with
f : N → N so that

f 2 (m) + f (n)|(m2 + n)2

Show that f (n) = n.

Problem 7.41 (Columbia 2010). Find all pairs of positive integers (m, n) such that
m2 + n2 = (m + 1)(n + 1).
Problem 7.42 (AMOC 2014, Senior Section). For which integers n ≥ 2 is it
possible to separate the numbers 1, 2, . . . , n into two sets such that the sum of the
numbers in one of the sets is equal to the product of the numbers in the other set?
Problem 7.43 (Greece). Determine all triples (p, m, n) of positive integers such that
p is a prime number and pm − 8 = n3 .
Problem 7.44 (Canadian Students Math Olympiad 2011). For a fixed positive
integer k, prove that there exist infinitely many primes p such that there is an integer
w, where w2 − 1 is not divisible by p, and the order of w modulo p is the same as the
order of w modulo pk .
Problem 7.45 (China TST 2009). Let a > b > 1 and b be an odd integer, n ∈ N. If
3n
bn |an − 1, then prove that ab > .
n
CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 389

Problem 7.46 (Kazakhstan 2015). Solve in positive integers: xy y x = (x + y)z .

Problem 7.47 (Kazakhstan 2015). Pk (n) is the product of all divisors of n that are
divisible by k (in empty case it is 1). Prove that, P1 (n) · P2 (n) · · · Pn (n) is a perfect
square.

Problem 7.48 (Russia 2000). If a perfect number greater than 6 is divisible by 3, it

is also divisible by 9. If a perfect number greater than 28 is divisible by 7, it is also
divisible by 49.

Problem 7.49 (Croatia 2014). For a positive integer n denote by s(n) the sum of all
positive divisors of n and by d(n) the number of positive divisors of n. Determine all
positive integers n such that

s(n) = n + d(n) + 1

Problem 7.50 (IMO 2003, Problem 2, N3). Find all pairs of positive integers (a, b)
such that
a2
2ab2 − b3 + 1
is a positive integer.

Problem 7.51. Show that n = ϕ(n) + τ (n) if and only if n is a prime.

Problem 7.52 (IMO Shortlist 2004, N2 Part (c)). Find all n for which f (n) = an
has a unique solution where,
n
X
f (n) = (i, n)
i=1

Problem 7.53. Let k be a positive integer. Find all positive integers n such that
3k | 2n − 1.

Problem 7.54. Let a, b be distinct real numbers such that the numbers

a − b, a2 − b2 , a3 − b3 , . . .

are all integers. Prove that a, b are both integers.

Problem 7.55 (MOSP 2001). Find all quadruples of positive integers (x, r, p, n) such
that p is a prime number, n, r > 1 and xr − 1 = pn .

Problem 7.56 (China TST 2009). Let a > b > 1 be positive integers and b be an
n
odd number, let n be a positive integer. If bn | an − 1, then show that ab > 3n .

Problem 7.57 (Romanian Junior Balkan TST 2008). Let p be a prime number,
p 6= 3, and integers a, b such that p | a + b and p2 | a3 + b3 . Prove that p2 | a + b or
p3 | a3 + b3 .
390

Problem 7.58. Let m and n be positive integers. Prove that for each odd positive
integer b there are infinitely many primes p such that pn ≡ 1 (mod bm ) implies bm−1 | n.
Problem 7.59 (IMO 1990). Determine all integers n > 1 such that
2n + 1
n2
is an integer.
Problem 7.60. Find all positive integers n such that

2n−1 + 1
.
n
is an integer.
(5p − 2p )(5q − 2q )
Problem 7.61. Find all primes p, q such that is an integer.
pq
Problem 7.62. For some natural number n let a be the greatest natural number for
which 5n − 3n is divisible by 2a . Also let b be the greatest natural number such that
2b ≤ n. Prove that a ≤ b + 3.
Problem 7.63. Determine all sets of non-negative integers x, y and z which satisfy the
equation
2x + 3y = z 2 .
Problem 7.64 (IMO ShortList 2007). Find all surjective functions f : N → N such
that for every m, n ∈ N and every prime p, the number f (m + n) is divisible by p if
and only if f (m) + f (n) is divisible by p.
Problem 7.65 (Romania TST 1994). Let n be an odd positive integer. Prove that
n
((n − 1)n + 1)2 divides n(n − 1)(n−1) +1 + n.
Problem 7.66. Find all positive integers n such that 3n − 1 is divisible by 2n .
Problem 7.67 (Romania TST 2009). Let a, n ≥ 2 be two integers, which have the
following property: there exists an integer k ≥ 2, such that n divides (a − 1)k . Prove
that n also divides an−1 + an−2 + · · · + a + 1.
5a +1
Problem 7.68. Find all the positive integers a such that 3a
is a positive integer.
Problem 7.69. Find all primes p, q such that pq|5p + 5q .
Problem 7.70. Find all primes p, q such that pq|2p + 2q .
Problem 7.71 (APMO 2016). A positive integer is called fancy if it can be expressed
in the form
2a1 + 2a2 + · · · + 2a100 ,
where a1 , a2 , · · · , a100 are non-negative integers that are not necessarily distinct. Find
the smallest positive integer n such that no multiple of n is a fancy number.
CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 391

Problem 7.72 (Argentina Intercollegiate Olympiad First Level 2016). Find

all positive integers a, b, c, and d, all less than or equal to 6, such that
a c
= + 2.
b d
Problem 7.73 (Argentina Intercollegiate Olympiad Second Level 2016). Find
all positive integers x and y which satisfy the following conditions:
1. x is a 4−digit palindromic number, and
2. y = x + 312 is a 5−digit palindromic number.
Note. A palindromic number is a number that remains the same when its digits are
reversed. For example, 16461 is a palindromic number.
Problem 7.74 (Argentina Intercollegiate Olympiad Third Level 2016). Find
a number with the following conditions:
1. it is a perfect square,
2. when 100 is added to the number, it equals a perfect square plus 1, and
3. when 100 is again added to the number, the result is a perfect square.
Problem 7.75 (Argentina Intercollegiate Olympiad Third Level 2016). Let
a1 , a2 , . . . , a15 be an arithmetic progression. If the sum of all 15 terms is twice the sum
of the first 10 terms, find ad1 , where d is the common difference of the progression.
Problem 7.76 (Austria Federal Competition for Advanced Students Final
Round 2016). Determine all composite positive integers n with the following property:
If 1 = d1 < d2 < · · · < dk = n are all the positive divisors of n, then

(d2 − d1 ) : (d3 − d2 ) : · · · : (dk − dk−1 ) = 1 : 2 · · · : (k − 1).

Problem 7.77 (Austria National Competition Final Round 2016). Let a, b,

and c be integers such that
ab ac bc
+ +
c b a
is an integer. Prove that each of the numbers
ab ac bc
, , and
c b a
is an integer.
Problem 7.78 (Austria Beginners’ Competition 2016). Determine all nonnega-
tive integers n having two distinct positive divisors with the same distance from n/3.
Problem 7.79 (Austria Regional Competition 2016). Determine all positive in-
tegers k and n satisfying the equation

k 2 − 2016 = 3n .
392

Problem 7.80 (Azerbaijan TST 2016). The set A consists of natural numbers
such that these numbers can be expressed as 2x2 + 3y 2 , where x and y are integers.
(x2 + y 2 6= 0)

1. Prove that there is no perfect square in the set A.

2. Prove that multiple of odd number of elements of the set A cannot be a perfect
square.

Problem 7.81 (Azerbaijan Junior Mathematical Olympiad 2016). Given

34! = 295232799039a041408476186096435b0000000,

in decimal representation, find the numbers a and b.

Problem 7.82 (Azerbaijan Junior Mathematical Olympiad 2016). Prove that

if for a real number a, a + a1 is integer then an + a1n is also integer for any positive integer
n.

Problem 7.83 (Azerbaijan Junior Mathematical Olympiad 2016). A quadruple

(p, a, b, c) of positive integers is called a good quadruple if

(a) p is odd prime,

(b) a, b, c are distinct,

(c) ab + 1, bc + 1, and ca + 1 are divisible by p.

a+b+c
Prove that for all good quadruple p + 2 ≤ 3
, and show the equality case.

Problem 7.84 (Balkan 2016). Find all monic polynomials f with integer coefficients
satisfying the following condition: there exists a positive integer N such that p divides
2(f (p)!) + 1 for every prime p > N for which f (p) is a positive integer.
Note: A monic polynomial has a leading coefficient equal to 1.

Problem 7.85 (Bay Area Olympiad 2016). Let A = 2k − 2 and B = 2k · A, where

k is an integer (k ≥ 2). Show that, for every integer k greater than or equal to 2,

1. A and B have the same set of distinct prime factors.

2. A + 1 and B + 1 have the same set of distinct prime factors.

Problem 7.86 (Bay Area Olympiad 2016). Find a positive integer N and a1 , a2 , · · · , aN
where ak = 1 or ak = −1, for each k = 1, 2, · · · , N, such that

a1 · 13 + a2 · 23 + a3 · 33 · · · + aN · N 3 = 20162016

or show that this is impossible.

Problem 7.87 (Belgium Flanders Math Olympiad Final Round 2016). Find
the smallest positive integer n which does not divide 2016!.
CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 393

Problem 7.88 (Belgium National Olympiad Final Round 2016). Solve the
equation

22m+1 + 9 · 2m + 5 = n2

for integers m and n.

Problem 7.89 (Benelux 2016). Find the greatest positive integer N with the fol-
lowing property: there exist integers x1 , . . . , xN such that x2i − xi xj is not divisible by
1111 for any i 6= j.

Problem 7.90 (Benelux 2016). Let n be a positive integer. Suppose that its positive
divisors can be partitioned into pairs (i.e. can be split in groups of two) in such a way
that the sum of each pair is a prime number. Prove that these prime numbers are
distinct and that none of these are a divisor of n.

Problem 7.91 (Bosnia and Herzegovina TST 2016). For an infinite sequence
a1 < a2 < a3 < . . . of positive integers we say that it is nice if for every positive integer
n holds a2n = 2an . Prove the following statements:

(a) If there is given a nice sequence and prime number p > a1 , there exist some term
of the sequence which is divisible by p.

(b) For every prime number p > 2, there exist a nice sequence such that no terms of
the sequence are divisible by p.

Problem 7.92 (Bosnia and Herzegovina TST 2016). Determine the largest posi-
tive integer n which cannot be written as the sum of three numbers bigger than 1 which
are pairwise coprime.

Problem 7.93 (Bulgaria

m
Olympiad 2016). Find all positive integers m

National
n
and n such that 22 + 1 22 + 1 is divisible by mn.

Problem 7.94 (Bulgaria National Olympiad 2016). Determine whether there

exists a positive integer n < 109 such that n can be expressed as a sum of three squares
of positive integers by more than 1000 distinct ways.

Problem 7.95 (Canadian Mathematical Olympiad Qualification 2016).

(a) Find all positive integers n such that 11|(3n + 4n ).

(b) Find all positive integers n such that 31|(4n + 7n + 20n ).

Problem 7.96 (Canadian Mathematical Olympiad Qualification 2016). Deter-

mine all ordered triples of positive integers (x, y, z) such that gcd(x + y, y + z, z + x) >
gcd(x, y, z).

Problem 7.97 (Canada National Olympiad 2016). Find all polynomials P (x)
with integer coefficients such that P (P (n) + n) is a prime number for infinitely many
integers n.
394

Problem 7.98 (CCA Math Bonanza 2016). Let f (x) = x2 + x + 1. Determine the
ordered pair (p, q) of primes satisfying f (p) = f (q) + 242.

Problem 7.99 (CCA Math Bonanza 2016). Let f (x) = x2 + x + 1. Determine the
ordered pair (p, q) of primes satisfying f (p) = f (q) + 242.

Problem 7.100 (CCA Math Bonanza 2016). Compute

420
X
gcd(k, 420).
k=1

Problem 7.101 (CCA Math Bonanza 2016). Pluses and minuses are inserted in
the expression
±1 ± 2 ± 3 · · · ± 2016
such that when evaluated the result is divisible by 2017. Let there be N ways for this
to occur. Compute the remainder when N is divided by 503.

Problem 7.102 (CCA Math Bonanza 2016). What is the largest integer that must
divide n5 − 5n3 + 4n for all integers n?

Problem 7.103 (CCA Math Bonanza 2016). Determine the remainder when

2
26 · 310 · 512 − 754 262 − 1 + 310 − 506 + 512

is divided by 1001.

Problem 7.104 (CentroAmerican 2016). Find all positive integers n that have 4
digits, all of them perfect squares, and such that n is divisible by 2, 3, 5, and 7.

Problem 7.105 (CentroAmerican 2016). We say a number is irie if it can be

1
written in the form 1 + for some positive integer k. Prove that every integer n ≥ 2
k
can be written as the product of r distinct irie numbers for every integer r ≥ n − 1.

Problem 7.106 (Chile 2016). Determine all triples of positive integers (p, n, m) with
p a prime number, which satisfy the equation:

pm − n3 = 27.

Problem 7.107 (Chile 2016 ). Find all prime numbers that do not have a multiple
ending in 2015.

Problem 7.108 (Chile 2016). Find the number of different numbers of the form
i2
⌊ 2015 ⌋, where i = 1, 2, . . . , 2015.

Problem 7.109 (China Girls Mathematical Olympiad 2016). Let m and n are
relatively prime integers and m > 1, n > 1. Show that there are positive integers a, b, c
such that ma = 1 + nb c , and n and c are relatively prime.
1
Thanks to Kamal Kamrava and Behnam Sajadi for translating the problem.
CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 395

Problem 7.110 (China National Olympiad 2016). Let p be an odd prime and
a1 , a2 , ..., ap be integers. Prove that the following two conditions are equivalent:
p−1
1. There exists a polynomial P (x) with degree ≤ 2
such that P (i) ≡ ai (mod p)
for all 1 ≤ i ≤ p.
p−1
2. For any natural d ≤ 2
,
p
X
(ai+d − ai )2 ≡ 0 (mod p),
i=1

where indices are taken modulo p.

Problem 7.111 (China South East Mathematical Olympiad 2016). LetPn be a
1
positive integer and let Dn be the set of all positive divisors of n. Define f (n) = 1+d
.
d∈Dn
Prove that for any positive integer m,
m
X
f (i) < m.
i=1

Problem 7.112 (China South East Mathematical Olympiad Pn 2016). Let {an }
be a sequence consisting of positive integers such that n | i=1 ai and an ≤ (n+2016)2
2

for all n ≥ 2016. Define bn = an+1 − an . Prove that the sequence {bn } is eventually
constant.
Problem 7.113 (China South East Mathematical Olympiad 2016). Define the
sets

A = {a3 + b3 + c3 − 3abc : a, b, c ∈ N},

B = {(a + b − c)(b + c − a)(c + a − b) : a, b, c ∈ N},
P = {n : n ∈ A ∩ B, 1 ≤ n ≤ 2016}.

Find the number of elements of P .

Problem 7.114 (China TST‌2016). Let c, d ≥ 2 be positive integers. Let {an } be
the sequence satisfying a1 = c, an+1 = adn + c for n = 1, 2, . . . . Prove that for any n ≥ 2,
there exists a prime number p such that p | an and p ∤ ai for i = 1, 2, . . . , n − 1.
Problem 7.115 (China TST 2016). Set positive integer m = 2k · t, where k is a
non-negative integer, t is an odd number, and let f (m) =Qt1−k . Prove that for any
positive integer n and for any positive odd number a ≤ n, nm=1 f (m) is a multiple of
a.
Problem 7.116 (China TST 2016). Does there exist two infinite positive integer
sets S, T , such that any positive integer n can be uniquely expressed in the form

n = s1 t 1 + s2 t 2 + · · · + sk t k ,

where k is a positive integer dependent on n, s1 < s2 < · · · < sk are elements of S,

t1 , . . . , tk are elements of T ?
396

Problem 7.117 (China TST 2016). Let a, b, b′ , c, m, q be positive integers, where

m > 1, q > 1, |b − b′ | ≥ a. It is given that there exist a positive integer M such that

Sq (an + b) ≡ Sq (an + b′ ) + c (mod m)

holds for all integers n ≥ M . Prove that the above equation is true for all positive
integers n. (Here Sq (x) is the sum of digits of x taken in base q).

Problem 7.118 (China Western Mathematical Olympiad 2016). For an n-tuple

of integers, define a transformation to be:

(a1 , a2 , . . . , an−1 , an ) → (a1 + a2 , a2 + a3 , . . . , an−1 + an , an + a1 )

Find all ordered pairs of integers (n, k) with n, k ≥ 2, such that for any n-tuple of
integers (a1 , a2 , . . . , an−1 , an ), after a finite number of transformations, every element in
the of the n-tuple is a multiple of k.

Problem 7.119 (China Western Mathematical Olympiad 2016). Prove that

there exist infinitely many positive integer triples (a, b, c) such that a, b, c are pairwise
relatively prime ,and ab + c, bc + a, ca + b are pairwise relatively prime.

Problem 7.120 (Croatia First Round Competition 2016). Can the sum of
squares of three consecutive integers be divisible by 2016?

Problem 7.121 (Croatia First Round Competition 2016). Let a = 123456789

and N = a3 − 2a2 − 3a. Prove that N is a multiple of 540.

Problem 7.122 (Croatia First Round Competition 2016). Find all pairs (a, b)
of positive integers such that 1 < a, b ≤ 100 and
1 1
+
loga 10 logb 10

is a positive integer.

Problem 7.123 (Croatia First Round Competition 2016). A sequence (an ) is

given: a1 = a2 = 1, and

a22 a23 a2
an+1 = + + ··· + n for n ≥ 2.
a1 a2 an−1

Find a2016 .

Problem 7.124 (Croatia First Round Competition 2016). Let a, b, and c be

integers. If 4a + 5b − 3c is divisible by 19, prove that 6a − 2b + 5c is also divisible by 19.

Problem 7.125 (Croatia First Round Competition 2016). Determine all pairs
of positive integers (x, y) such that x2 − y! = 2016.

Problem 7.126 (Croatia Second Round Competition 2016).

CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 397

(a) Prove that there are no two positive integers such that the difference of their squares
is 987654.

(b) Prove that there are no two positive integers such that the difference of their cubes
is 987654.

Problem 7.127 (Croatia Second Round Competition 2016). How many ordered
pairs (m, k) of positive integers satisfy

20m = k(m − 15k)?

Problem 7.128 (Croatia Second Round Competition 2016). Determine all pairs
(a, b) of positive integers such that

a3 − 3b = 15,
b2 − a = 13.

Problem 7.129 (Croatia Second Round Competition 2016). Prove that, for
every positive integer n > 3, there are n different positive integers whose reciprocals
add up to 1.

Problem 7.130 (Croatia Second Round Competition 2016). Determine all pairs
(a, b) of integers such that (7a − b)2 = 2(a − 1)b2 .

Problem 7.131 (Croatia Final Round National Competition 2016). Determine

the sum
22 + 1 32 + 1 1002 + 1
+ + · · · + .
22 − 1 32 − 1 1002 − 1

Problem 7.132 (Croatia Final Round National Competition 2016). Let a, b,

and c be positive integers such that

b 1
c=a+ − .
a b
Prove that c is the square of an integer.

Problem 7.133 (Croatia Final Round National Competition 2016). Determine

all pairs (m, n) of positive integers for which exist integers a, b, and c that satisfy

a + b + c = 0 and a2 + b2 + c2 = 2m · 3n .

Problem 7.134 (Croatia Final Round National Competition 2016). Prove that
there does not exist a positive integer k such that k + 4 and k 2 + 5k + 2 are cubes of
positive integers.

Problem 7.135 (Croatia Final Round National Competition 2016). Determine

all triples (m, n, k) of positive integers such that 3m + 7n = k 2 .
398

Problem 7.136 (Croatian Mathematical Olympiad 2016). Find all pairs (p, q)
of prime numbers such that

p(p2 − p − 1) = q(2q + 3).

Problem 7.137 (Croatian TST for MEMO 2016, Sweden 2014). Find all pairs
(m, n) of positive integers such that

3 · 5m − 2 · 6n = 3.

Problem 7.138 (Croatia IMO TST 2016). Prove that for every positive integer n
there exist integers a and b such that n divides 4a2 + 9b2 − 1.
Problem 7.139 (Croatia IMO TST 2016, Bulgaria TST 2016). Let p > 109 be
a prime number such that 4p + 1 is also prime. Prove that the decimal expansion of
1
4p+1
contains all the digits 0, 1, . . . , 9.

Problem 7.140 (Denmark Georg Mohr Contest Second Round 2016). Find
all possible values of the number
a+b a+c b+c
+ + ,
c b a
a+b a+c b+c
where a, b, and c are positive integers, and , , and are also positive
c b a
integers.
Problem 7.141 (ELMO 2016). Cookie Monster says a positive integer n is crunchy
if there exist 2n real numbers x1 , x2 , . . . , x2n , not all equal, such that the sum of any
n of the xi ’s is equal to the product of the other n of the xi ’s. Help Cookie Monster
determine all crunchy integers.
Problem 7.142 (ELMO 2016). Big Bird has a polynomial P with integer coefficients
such that n divides P (2n ) for every positive integer n. Prove that Big Bird’s polynomial
must be the zero polynomial.
Problem 7.143 (Estonia IMO TST First Stage 2016). Let p be a prime. Find
all integers (not necessarily positive) a, b, and c such that

ab bc ca = p.

Problem 7.144 (Estonia IMO TST First Stage 2016). Prove that for every
positive integer n ≥ 3,
√ √ 3 √
2 · 3 · 4 . . . n−1 n > n.

Problem 7.145 (Estonia IMO TST Second Stage 2016). Find all positive integers
n such that


n2 + 11n − 4 · n! + 33 · 13n + 4

is a perfect square.
CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 399

Problem 7.146 (Estonia National Olympiad Tenth Grade 2016). Find all pairs
of integers (a, b) which satisfy

3(a2 + b2 ) − 7(a + b) = −4.

Problem 7.147 (Estonia National Olympiad Eleventh Grade 2016). Find the
greatest positive integer n for which 32016 − 1 is divisible by 2n .
Problem 7.148 (Estonia National Olympiad Eleventh Grade 2016). Let n be
a positive integer. Let δ(n) be the number of positive divisors of n and let σ(n) be their
sum. Prove that
(δ(n))2
σ(n) > .
2
Problem 7.149 (Estonia Regional Olympiad Tenth Grade 2016). Does the
equation

x2 + y 2 + z 2 + w2 = 3 + xy + yz + zw

has a solution in which x, y, z, and w are different integers?

Problem 7.150 (Estonia Regional Olympiad Twelfth Grade 2016). Determine
7
whether the logarithm of 6 in base 10 is larger or smaller than .
9
Problem 7.151 (Estonia Regional Olympiad Twelfth Grade 2016). Find the
largest positive integer n so that one can select n primes p1 , p2 , . . . , pn (not necessarily
distinct) such that

p 1 , p1 + p 2 , . . . , p 1 + p 2 + · · · + p n

are all primes.

Problem 7.152 (European Girls’ Mathematical Olympiad 2016). Let S be
the set of all positive integers n such that n4 has a divisor in the range n2 + 1, n2 +
2, . . . , n2 + 2n. Prove that there are infinitely many elements of S of each of the forms
7m, 7m + 1, 7m + 2, 7m + 5, 7m + 6 and no elements of S of the form 7m + 3 and 7m + 4,
where m is an integer.
Problem 7.153 (European Mathematical Cup Seniors 2016). A = {a, b, c} is
a set containing three positive integers. Prove that we can find a set B ⊂ A, say
B = {x, y}, such that for all odd positive integers m and n,

10 | xm y n − xn y m .

Problem 7.154 (European Mathematical Cup Juniors 2016). Let d(n) denote
the number of positive divisors of n. For a positive integer n we define f (n) as

f (n) = d(k1 ) + d(k2 ) + · · · + d(km ),

where 1 = k1 < k2 < · · · < km = n are all divisors of the number n. We call an integer
n > 1 almost perfect if f (n) = n. Find all almos perfect numbers.
400

Problem 7.155 (Finland MAOL Competition 2016). Let n be a positive integer.

Find all pairs (x, y) of positive integers such that

(4a − b)(4b − a) = 1770n .

Problem 7.156 (Germany National Olympiad First Round Ninth/Tenth Grade,

2016).
(A) Prove that there exists an integer a > 1 such that the number


82 · a8 − a4

is divisible by the product of three consecutive positive integers each of which has
at least two digits.

(B) Determine the smallest prime number a with at least two digits such that the
number


82 · a8 − a4

is divisible by the product of three consecutive positive integers each of which has
at least two digits.

(C) Determine the smallest integer a > 1 such that the number


82 · a8 − a2

is divisible by the product of three consecutive positive integers each of which has
at least two digits.
Problem 7.157 (Germany National Olympiad First Round Eleventh/Twelfth
Grade, 2016). Consider the following system of equations:

2(z − 1) − x = 55,
4xy − 8z = 12,
a(y + z) = 11.

Find two largest real values for a for which there are positive integers x, y, and z that
satisfy the system of equations. In each of these solutions, determine xyz.
Problem 7.158 (Germany National Olympiad First Round Eleventh/Twelfth
Grade, 2016). Find all pairs (a, b) of positive integers for which (a+1)(b+1) is divisible
by ab.
Problem 7.159 (Germany National Olympiad Second Round Tenth Grade,
2016). For each of the following cases, determine whether there exist prime numbers
x, y, and z such that the given equality holds
(a) y = z 2 − x2 .

(b) x2 + y = z 4 .
CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 401

(c) x2 + y 3 = z 4 .
Problem 7.160 (Germany National Olympiad Second Round Eleventh/Twelfth
Grade, 2016). The sequence x1 , x2 , x3 , . . . is defined as x1 = 1 and

xk+1 = xk + yk for k = 1, 2, 3, . . .

where yk is the last digit of decimal representation of xk . Prove that the sequence
x1 , x2 , x3 , . . . contains all powers of 4. That is, for every positive integer n, there exists
some natural k for which xk = 4n .
Problem 7.161 (Germany National Olympiad Third Round Eleventh/Twelfth
Grade, 2016). Find all positive integers a and b which satisfy
 
ab + 1
= 2ab(a + b).
2
Problem 7.162 (Germany National Olympiad Third Round Eleventh/Twelfth
Grade, 2016). Let m and n be two positive integers. Prove that for every positive
integer k, the following statements are equivalent:
1. n + m is a divisor of n2 + km2 .
2. n + m is a divisor of k + 1.
Problem 7.163 (Germany National Olympiad Fourth Round Ninth Grade,
2016). Find all triples (a, b, c) of integers which satisfy

a3 + b3 = c3 + 1,
b2 − a2 = a + b,
2a3 − 6a = c3 − 4a2 .

Problem 7.164 (Germany National Olympiad Fourth Round Tenth Grade,

2016 ). A sequence of positive integers a1 , a2 , a3 , . . . is defined as follows: a1 is a 3
digit number and ak+1 (for k ≥ 1) is obtained by

ak+1 = ak + 2 · Q(ak ),

where Q(ak ) is the sum of digits of ak when represented in decimal system. For instance,
if one takes a1 = 358 as the initial term, the sequence would be

a1 = 358,
a2 = 358 + 2 · 16 = 390,
a3 = 390 + 2 · 12 = 414,
a4 = 414 + 2 · 9 = 432,
..
.

Prove that no matter what we choose as the starting number of the sequence,
2
Thanks to Arian Saffarzadeh for translating the problem.
402

(a) the sequence will not contain 2015.

(b) the sequence will not contain 2016.
Problem 7.165 (Germany National Olympiad Fourth Round Eleventh Grade,
2016). Find all positive integers m and n with m ≤ 2n which satisfy
   2
2n m
m· = .
n 2
Problem 7.166 (Germany TST 2016). The positive integers a1 , a2 , . . . , an are
aligned clockwise in a circular line with n ≥ 5. Let a0 = an and an+1 = a1 . For
each i ∈ {1, 2, . . . , n} the quotient
ai−1 + ai+1
qi =
a1
is an integer. Prove
2n ≤ q1 + q2 + · · · + qn < 3n.
Problem 7.167 (Germany TST 2016, Taiwan TST First Round 2016). Deter-
mine all positive integers M such that the sequence a0 , a1 , a2 , · · · defined by
1
a0 = M + and ak+1 = ak ⌊ak ⌋ for k = 0, 1, 2, · · ·
2
contains at least one integer term.
Problem 7.168 (Greece 2016). Find all triplets of nonnegative integers (x, y, z) and
x ≤ y such that
x2 + y 2 = 3 · 2016z + 77.
Problem 7.169 (Greece TST 2016). Given is the sequence (an )n≥0 which is defined
as follows:a0 = 3 and an+1 − an = n(an − 1) , ∀n ≥ 0. Determine all positive integers
m such that gcd(m, an ) = 1 , ∀n ≥ 0.
Problem 7.170 (Harvard-MIT Math Tournament 2016). Denote by N the pos-
itive integers. Let f : N → N be a function such that, for any w, x, y, z ∈ N,

f (f (f (z)))f (wxf (yf (z))) = z 2 f (xf (y))f (w).

Show that f (n!) ≥ n! for every positive integer n.

Problem 7.171 (Hong Kong (China) Mathematicalr r 2016). Find all
r Olympiad
2015 2015 2015
integral ordered triples (x, y, z) such that + + are positive
x+y y+z x+z
integers.
Problem 7.172 (Hong Kong Preliminary Selection Contest 2016). Find the
remainder when 1 3
.
..
1715
19
is divided by 100.
CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 403

Problem 7.173 (Hong Kong Preliminary Selection Contest 2016). Let k be an

integer. If the equation
kx2 + (4k − 2)x + (4k − 7) = 0
has an integral root, find the sum of all possible values of k.
Problem 7.174 (Hong Kong Preliminary Selection Contest 2016). Let n be a
positive integer. If the two numbers (n + 1)(2n + 15) and n(n + 5) have exactly the
same prime factors, find the greatest possible value of n.
Problem 7.175 (Hong Kong Preliminary Selection Contest 2016). An arith-
metic sequence with 10 terms has common difference d > 0. If the absolute value of
each term is a prime number, find the smallest possible value of d.
2
Problem 7.176 (Hong Kong Preliminary Selection Contest 2016). Let a1 = 3
and
r
an 24an + 9 9
an+1 = + −
4 256 48
for all integers n ≥ 1. Find the value of
a1 + a2 + a3 + . . .
Problem 7.177 (Hong Kong TST 2016). Find all natural numbers n such that n,
n2 + 10, n2 − 2, n3 + 6, and n5 + 36 are all prime numbers.
Problem 7.178 (Hong Kong TST 2016). Find all triples (m, p, q) such that
2 m p2 + 1 = q 7 ,
where p and q are primes and m is a positive integer.
Problem 7.179 (Hong Kong TST 2016). Find all prime numbers p and q such that
p2 |q 3 + 1 and q 2 |p6 − 1.
Problem 7.180 (Hong Kong TST 2016). Let p be a prime number greater than 5.
Suppose there is an integer k satisfying that k 2 + 5 is divisible by p. Prove that there
are positive integers m and n such that p2 = m2 + 5n2 .
Problem 7.181 (IberoAmerican 2016). Find all prime numbers p, q, r, k such that
pq + qr + rp = 12k + 1.
Problem 7.182 (IberoAmerican 2016). Let k be a positive integer and a1 , a2 , . . . , ak
digits. Prove that there exists a positive integer n such that the last 2k digits of 2n are,
in the following order, a1 , a2 , . . . , ak , b1 , b2 , . . . , bk , for certain digits b1 , b2 , . . . , bk .
Problem 7.183 (IMO Shortlist 2015, India TST 2016, Taiwan TST Sec-
ond Round 2016, Croatian Mathematical Olympiad 2016, Switzerland TST
2016). Let m and n be positive integers such that m > n. Define
m+k
xk =
n+k
for k = 1, 2, . . . , n + 1. Prove that if all the numbers x1 , x2 , . . . , xn+1 are integers, then
x1 x2 . . . xn+1 − 1 is divisible by an odd prime.
404

Problem 7.184 (IMO 2016). A set of postive integers is called fragrant if it contains
at least two elements and each of its elements has a prime factor in common with at
least one of the other elements. Let P (n) = n2 + n + 1. What is the least possible
positive integer value of b such that there exists a non-negative integer a for which the
set
{P (a + 1), P (a + 2), . . . , P (a + b)}
is fragrant?
Problem 7.185 (India IMO Training Camp 2016). Given that n is a natural
number such that the leftmost digits in the decimal representations of 2n and 3n are
the same, find all possible values of the leftmost digit.
Problem 7.186 (India IMO Practice Test 2016). We say a natural number n is
perfect if the sum of all the positive divisors of n is equal to 2n. For example, 6 is
perfect since its positive divisors 1, 2, 3, 6 add up to 12 = 2 × 6. Show that an odd
perfect number has at least 3 distinct prime divisors.
Problem 7.187 (India TST 2016). Let n be a natural number. We define sequences
hai i and hbi i of integers as follows. We let a0 = 1 and b0 = n. For i > 0, we let

(2ai−1 + 1, bi−1 − ai−1 − 1) , if ai−1 < bi−1 ,

(ai , bi ) = (ai−1 − bi−1 − 1, 2bi−1 + 1) , if ai−1 > bi−1 ,


(ai−1 , bi−1 ) , if ai−1 = bi−1 .

Given that ak = bk for some natural number k, prove that n + 3 is a power of two.
Problem 7.188 (India TST 2016). Let N denote the set of all natural numbers.
Show that there exists two nonempty subsets A and B of N such that
1. A ∩ B = {1};

2. every number in N can be expressed as the product of a number in A and a

number in B;

3. each prime number is a divisor of some number in A and also some number in B;

4. one of the sets A and B has the following property: if the numbers in this set are
written as x1 < x2 < x3 < · · · , then for any given positive integer M there exists
k ∈ N such that xk+1 − xk ≥ M ;

5. Each set has infinitely many composite numbers.

Problem 7.189 (India National Olympiad 2016). Let N denote the set of natural
numbers. Define a function T : N → N by T (2k) = k and T (2k + 1) = 2k + 2. We write
T 2 (n) = T (T (n)) and in general T k (n) = T k−1 (T (n)) for any k > 1.
(i) Show that for each n ∈ N, there exists k such that T k (n) = 1.

(ii) For k ∈ N, let ck denote the number of elements in the set {n : T k (n) = 1}. Prove
that ck+2 = ck+1 + ck , for k ≥ 1.
CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 405

Problem 7.190 (India National Olympiad 2016). Consider a non-constant arith-

metic progression a1 , a2 , · · · , an , · · · . Suppose there exist relatively prime positive in-
tegers p > 1 and q > 1 such that a21 , a2p+1 and a2q+1 are also the terms of the same
arithmetic progression. Prove that the terms of the arithmetic progression are all inte-
gers.

Problem 7.191 (Iran Third Round National Olympiad 2016). Let F be a subset
of the set of positive integers with at least two elements and P be a polynomial with
integer coefficients such that for any two elements of F like a and b, the following two
conditions hold

(i) a + b ∈ F , and

(ii) gcd(P (a), P (b)) = 1.

Prove that P (x) is a constant polynomial.

Problem 7.192 (Iran Third Round National Olympiad 2016). Let P be a poly-
nomial with integer coefficients. We say P is good if there exist infinitely many prime
numbers q such that the set

X = {P (n) mod q : n ∈ N}
q+1
has at least 2
members. Prove that the polynomial x3 + x is good.

Problem 7.193 (Iran Third Round National Olympiad 2016). Let m be a posi-
tive integer. The positive integer a is called a golden residue modulo m if gcd(a, m) = 1
and xx ≡ a (mod m) has a solution for x. Given a positive integer n, suppose that a
n
is a golden residue modulo nn . Show that a is also a golden residue modulo nn .

Problem 7.194 (Iran Third Round National Olympiad 2016). Let p, q be prime
numbers (q is odd). Prove that there exists an integer x such that

q|(x + 1)p − xp

if and only if
q≡1 (mod p).

Problem 7.195 (Iran Third Round National Olympiad 2016). We call a function
g special if g(x) = af (x) (for all x) where a is a positive integer and f is polynomial
with integer coefficients such that f (n) > 0 for all positive integers n.
A function is called an exponential polynomial if it is obtained from the product or
2
sum of special functions. For instance, 2x 3x +x−1 + 52x is an exponential polynomial.
Prove that there does not exist a non-zero exponential polynomial f (x) and a non-
constant polynomial P (x) with integer coefficients such that

P (n)|f (n)

for all positive integers n.

406

Problem 7.196 (Iran Third Round National Olympiad 2016). A sequence P =

{an }∞
n=1 is called a permutation of natural numbers if for any natural number m, there
exists a unique natural number n such that an = m.
We also define Sk (P ) as Sk (P ) = a1 + a2 + · · · + ak (the sum of the first k elements
of the sequence).
Prove that there exists infinitely many distinct permutations of natural numbers
like P1 , P2 , . . . such that
∀k, ∀i < j : Sk (Pi )|Sk (Pj ).

Problem 7.197 (Iran TST 2016). Let p 6= 13 be a prime number of the form 8k + 5
such that 39 is a quadratic non-residue modulo p. Prove that the equation

x41 + x42 + x43 + x44 ≡ 0 (mod p)

has a solution in integers such that p ∤ x1 x2 x3 x4 .

Problem 7.198 (Italy National Olympiad 2016). Determine all pairs of positive
integers (a, n) with a ≥ n ≥ 2 for which (a + 1)n + a − 1 is a power of 2.

Problem 7.199 (Japan Mathematical Olympiad Preliminary 2016). For 1 ≤

n ≤ 2016, how many integers n satisfying the condition: the reminder divided by 20 is
smaller than the one divided by 16.

Problem 7.200 (Japan Mathematical Olympiad Preliminary 2016). Determine

the number of pairs (a, b) of integers such that 1 ≤ a, b ≤ 2015, a is divisible by b + 1,
and 2016 − a is divisible by b.

Problem 7.201 (Japan Mathematical Olympiad Finals 2016). Let p be an odd

prime number. For positive integer k satisfying 1 ≤ k ≤ p − 1, the number of divisors
of kp + 1 between k and p exclusive is ak . Find the value of a1 + a2 + . . . + ap−1 .

Problem 7.202 (Junior Balkan Mathematical Olympiad 2016). Find all triplets
of integers (a, b, c) such that the number

(a − b)(b − c)(c − a)
N= +2
2
is a power of 2016.

Problem 7.203 (Korea Summer Program Practice Test 2016). A infinite se-
quence {an }n≥0 of real numbers satisfy an ≥ n2 . Suppose that for each i, j ≥ 0 there
exist k, l with (i, j) 6= (k, l), l−k = j−i, and al −ak = aj −ai . Prove that an ≥ (n+2016)2
for some n.

Problem 7.204 (Korea Summer Program Practice Test 2016). A finite set S
of positive integers is given. Show that there is a positive integer N dependent only on
S, such that any x1 , . . . , xm ∈ S whose sum is a multiple of N , can be partitioned into
groups each of whose sum is exactly N . (The numbers x1 , . . . , xm need not be distinct.)
CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 407

Problem 7.205 (Korea Winter Program Practice Test 2016). p(x) is an irre-
ducible polynomial with integer coefficients, and q is a fixed prime number. Let an be a
number of solutions of the equation p(x) ≡ 0 mod q n . Prove that we can find M such
that {an }n≥M is constant.
Problem 7.206 (Korea Winter Program Practice Test 2016). Find all {an }n≥0
that satisfies the following conditions.
1. an ∈ Z,

2. a0 = 0, a1 = 1,

3. For infinitely many m, am = m, and

4. For every n ≥ 2, {2ai − ai−1 |i = 1, 2, 3, · · · , n} ≡ {0, 1, 2, · · · , n − 1} mod n.

Problem 7.207 (Korea Winter Program Practice Test 2016). Find all positive
integers a, b, m, and n such that

a2 + b2 = m2 − n2 , and
ab = 2mn.

Problem 7.208 (Korea Winter Program Practice Test 2016). Find all pairs of
positive integers (n, t) such that 6n + 1 = n2 t, and (n, 29 × 197) = 1.
Problem 7.209 (Korea National Olympiad Final Round 2016). Prove that for
all rationals x, y, x − x1 + y − y1 = 4 is not true.
n
Problem 7.210 (Kosovo TST 2016). Show that for any n ≥ 2, the number 22 + 1
ends with 7.
Problem 7.211 (Latvia National Olympiad 2016).
1. Given positive integers x and y such that xy 2 is a perfect cube, prove that x2 y is
also a perfect cube.

2. Given that x and y are positive integers such that xy 10 is perfect 33rd power of a
positive integer, prove that x10 y is also a perfect 33rd power.

3. Given that x and y are positive integers such that xy 433 is a perfect 2016-power
of a positive integer, prove that x433 y is also a perfect 2016-power.

4. Given that x, y and z are positive integers such that x3 y 5 z 6 is a perfect 7th power
of a positive integer, show that also x5 y 6 z 3 is a perfect 7th power.
Problem 7.212 (Latvia National Olympiad 2016). Prove that among any 18
consecutive positive 3 digit numbers, there is at least one that is divisible by the sum
of its digits.
Problem 7.213 (Latvia National Olympiad 2016). Two functions are defined by
equations: f (a) = a2 + 3a + 2 and g(b, c) = b2 − b + 3c2 + 3c. Prove that for any positive
integer a there exist positive integers b and c such that f (a) = g(b, c).
408

Problem 7.214 (Macedonian National Olympiad 2016). Solve the equation in

the set of natural numbers 1 + xz + y z = lcm(xz , y z ).

Problem 7.215 (Macedonian National Olympiad 2016). Solve the equation in

the set of natural numbers xyz + yzt + xzt + xyt = xyzt + 3.

Problem 7.216 (Macedonian Junior Mathematical Olympiad 2016). Solve the

equation

x41 + x42 + · · · + x414 = 20163 − 1

in the set of integers.

Problem 7.217 (Macedonian Junior Mathematical Olympiad 2016). Solve the

equation

x + y 2 + (gcd(x, y))2 = xy · gcd(x, y)

in the set of positive integers.

Problem 7.218 (Mediterranean Mathematics Olympiad 2016). Determine all

integers n ≥ 1 for which the number n8 + n6 + n4 + 4 is prime.

Problem 7.219 (Middle European Mathematical Olympiad 2016). Find all

f : N → N such that f (a) + f (b) divides 2(a + b − 1) for all a, b ∈ N.

Problem 7.220 (Middle European Mathematical Olympiad 2016). A positive

integer n is Mozart if the decimal representation of the sequence 1, 2, . . . , n contains
each digit an even number of times. Prove that:

1. All Mozart numbers are even.

2. There are infinitely many Mozart numbers.

Problem 7.221 (Middle European Mathematical Olympiad 2016). For a pos-

itive integer n, the equation a2 + b2 + c2 + n = abc is given in the positive integers.
Prove that:

1. There does not exist a solution (a, b, c) for n = 2017.

2. For n = 2016, a is divisible by 3 for all solutions (a, b, c).

3. There are infinitely many solutions (a, b, c) for n = 2016.

Problem 7.222 (Netherlands TST 2016). Find all positive integers k for which
the equation:
lcm(m, n) − gcd(m, n) = k(m − n)
has no solution in integers positive (m, n) with m 6= n.
CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 409

Problem 7.223 (Nordic Mathematical Competition 2016). Determine all se-

quences (an )2016
n=1 of non-negative integers such that all sequence elements are less than
or equal to 2016 and

i + j | iai + jaj

for all i, j ∈ {1, 2, . . . , 2016}.

Problem 7.224 (Norway Niels Henrik Abel Mathematics Competition Final

Round 2016).

(a) Find all positive integers a, b, c, and d with a ≤ b and c ≤ d such that

a + b = cd, c + d = ab.

(b) Find all non-negative integers x, y, and z such that

x3 + 2y 3 + 4z 3 = 9!.

Problem 7.225 (Pan-African Mathematical Olympiad 2016). For any positive

integer n, we define the integer P (n) by

P (n) = n(n + 1)(2n + 1)(3n + 1) . . . (16n + 1)

Find the greatest common divisor of the integers P (1), P (2), P (3), . . . , P (2016).

Problem 7.226 (Philippine Mathematical Olympiad Area Stage 2016). Let

a, b, and c be three consecutive even numbers such that a > b > c. What is the value
of a2 + b2 + c2 − ab − bc − ac?

Problem 7.227 (Philippine Mathematical Olympiad Area Stage 2016). Find

the sum of all the prime factors of 27,000,001.

Problem 7.228 (Philippine Mathematical Olympiad Area Stage 2016). Find

the largest number N so that
N
X 1 1
< .
n=5
n(n − 2) 4

Problem 7.229 (Philippine Mathematical Olympiad Area Stage 2016). Let

sn be the sum of the digits of a natural number n. Find the smallest value of snn if n is
a four-digit number.

Problem 7.230 (Philippine Mathematical Olympiad Area Stage 2016). The

6 digit number 739ABC is divisible by 7, 8, and 9. What values can A, B, and C take?

Problem 7.231 (Polish Mathematical Olympiad 2016). Let p be a certain prime

number. Find all non-negative integers n for which polynomial P (x) = x4 −2(n+p)x2 +
(n − p)2 may be rewritten as product of two quadratic polynomials P1 , P2 ∈ Z[X].
410

Problem 7.232 (Polish Mathematical Olympiad 2016). Let k, n be odd positive

integers greater than 1. Prove that if there a exists natural number a such that k|2a +
1, n|2a − 1, then there is no natural number b satisfying k|2b − 1, n|2b + 1.
Problem 7.233 (Polish Mathematical Olympiad 2016). There are given two
positive real number a < b. Show that there exist positive integers p, q, r, s satisfying
following conditions:
p r
1. a < q
< s
< b.

2. p2 + q 2 = r2 + s2 .
Problem 7.234 (Romania Danube Mathematical Competition 2016). Deter-
mine all positive integers n such that all positive integers less than or equal to n and
prime to n are pairwise coprime.
Problem 7.235 (Romania Danube Mathematical Competition 2016). Given
an integer n ≥ 2, determine the numbers that can be written in the form
k
X
ai−1 ai ,
i=2

where k is an integer greater than or equal to 2, and a1 , a2 , . . . , ak are positive integers

that add up to n.
Problem 7.236 (Romania Imar Mathematical Competition 2016). Determine
all positive integers expressible, for every integer n ≥ 3, in the form
(a1 + 1)(a2 + 1) . . . (an + 1) − 1
,
a1 a2 . . . an
where a1 , a2 , . . . , an are pairwise distinct positive integers.
Problem 7.237 (Romanian Masters in Mathematics 2016). A cubic sequence is
a sequence of integers given by an = n3 + bn2 + cn + d, where b, c and d are integer
constants and n ranges over all integers, including negative integers.
(a) Show that there exists a cubic sequence such that the only terms of the sequence
which are squares of integers are a2015 and a2016 .
(b) Determine the possible values of a2015 · a2016 for a cubic sequence satisfying the
condition in part (a).
Problem 7.238 (Romanian Mathematical Olympiad District Round Grade
5, 2016). Find all three-digit numbers which decrease 13 times when the tens’ digit is
suppressed.
Problem 7.239 (Romanian Mathematical Olympiad District Round Grade
5, 2016). If A and B are positive integers, then AB will denote the number obtained
by writing, in order, the digits of B after the digits of A. For instance, if A = 193 and
B = 2016, then AB = 1932016. Prove that there are infinitely many perfect squares of
the form AB in each of the following situations:
CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 411

(a) A and B are perfect squares;

(b) A and B are perfect cubes;
(c) A is a perfect cube and B is a perfect square;
(d) A is a perfect square and B is a perfect cube.
Problem 7.240 (Romanian Mathematical Olympiad District Round Grade
6, 2016). The positive integers m and n are such that m2016 + m + n2 is divisible by
mn.
(a) Give an example of such m and n, with m > n.
(b) Prove that m is a perfect square.
Problem 7.241 (Romanian Mathematical Olympiad District Round Grade
7, 2016). Find all pairs of positive integers (x, y) such that
√ √ √
x + y = x + y + xy.
Problem 7.242 (Romanian Mathematical Olympiad District Round Grade
7, 2016). Let
 
M = x1 + 2x2 + 3x3 + · · · + 2015x2015 : x1 , x2 , . . . , x2015 ∈ {−2, 3} .

Prove that 2015 ∈ M but 2016 6∈ M .

Problem 7.243 (Romanian Mathematical Olympiad District Round Grade
8, 2016). For each positive integer n denote xn the number of the positive integers
with n digits, divisible by 4, formed with digits 2, 0, 1, or 6.
(a) Compute x1 , x2 , x3 , and x4 ;
(b) Find n so that
     
x2 x3 xn+1
1+ + + ··· + = 2016.
x1 x2 xn

Problem 7.244 (Romanian Mathematical Olympiad District Round Grade

8, 2016).
(a) Prove that, for every integer k, the equation x3 − 24x + k = 0 has at most one
integer solution.
(b) Prove that the equation x3 + 24x − 2016 = 0 has exactly one integer solution.
Problem 7.245 (Romanian Mathematical Olympiad District Round Grade
9, 2016). Let a and n be positive integers such that
(q )
√ √
n+ n = a ,

where {·} denotes the fractional part. Prove that 4a + 1 is a perfect square.
412

Problem 7.246 (Romanian Mathematical Olympiad District Round Grade

9, 2016). Let a ≥ 2 be an integer. Prove that the following statements are equivalent:

(a) One can find positive integers b and c such that a2 = b2 + c2 .

(b) One can find a positive integer d such that the equations x2 − ax + d = 0 and
x2 − ax − d = 0 have integer roots.

Problem 7.247 (Romanian Mathematical Olympiad Final Round Grade 5,

2016). Two positive integers x and y are such that

2010 x 2011
< < .
2011 y 2012

Find the smallest possible value of the sum x + y.

Problem 7.248 (Romanian Mathematical Olympiad Final Round Grade 5,

2016). Find all the positive integers a, b, and c with the property a + b + c = abc.

Problem 7.249 (Romanian Mathematical Olympiad Final Round Grade 6,

2016). We will call a positive integer exquisite if it is a multiple of the number of its
divisors (for instance, 12 is exquisite because it has 6 divisors and 12 is a multiple of
6).

(a) Find the largest exquisite two digit number.

(b) Prove that no exquisite number has its last digit 3.

Problem 7.250 (Romanian Mathematical Olympiad Final Round Grade 6,

2016). Find all positive integers a and b so that a+1
b
and b+2
a
are simultaneously positive
integers.

Problem 7.251 (Romanian Mathematical Olympiad Final Round Grade 6,

2016). Let a and b be positive integers so that there exists a prime number p with the
property [a, a + p] = [b, b + p]. Prove that a = b. Here, [x, y] denotes the least common
multiple of x and y.

Problem 7.252 (Romanian Mathematical Olympiad Final Round Grade 7,

2016). Find all non-negative integers n such that
q
√ √
n+3+ n+ n+3

is an integer.

Problem 7.253 (Romanian Mathematical Olympiad Final Round Grade 7,

2016). Find all the positive integers p with the property that the sum of the first p
positive integers is a four-digit positive integer whose decomposition into prime factors
is of the form 2m 3n (m + n), where m and n are non-negative integers.
CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 413

Problem 7.254 (Romanian Mathematical Olympiad Final Round Grade 8,

2016). Let n be a non-negative integer. We will say that the non-negative integers
x1 , x2 , . . . , xn have property (P ) if
x1 x2 . . . xn = x1 + 2x2 + · · · + nxn .
(a) Show that for every non-negative integer n, there exists n positive integers with
property (P ).
(b) Find all integers n ≥ 2 so that there exists n positive integers x1 , x2 , . . . , xn with
x< x2 < · · · < xn , having property (P ).
Problem 7.255 (Romanian Mathematical Olympiad Final Round Grade 9,
2016).
(a) Prove that 7 cannot be written as a sum of squares of three rational numbers.
(b) Let a be a rational number that can be written as a sum of squares of three rational
numbers. Prove that am can be written as a sum of squares of three rational
numbers, for any positive integer m.
Problem 7.256 (Romanian National Mathematical Olympiad Small Juniors
Shortlist, 2016). Find all non-negative integers n so that n2 − 4n + 2, n2 − 3n + 13
and n2 − 6n + 19 are simultaneously primes.
Problem 7.257 (Romanian National Mathematical Olympiad Small Juniors
Shortlist, 2016). We will call a number good if it is a positive integer with at least
two digits and by removing one of its digits we get a number which is equal to the sum
of its initial digits (for instance, 109 is good: remove 9 to get 10 = 1 + 0 + 9).
(a) Find the smallest good number.
(b) Find how many numbers are good.
Problem 7.258 (Romanian National Mathematical Olympiad Small Juniors
Shortlist, 2016). Four positive integers a, b, c, and d are not divisible by 5 and the
sum of their squares is divisible by 5. Prove that
N = (a2 + b2 )(a2 + c2 )(a2 + d2 )(b2 + c2 )(b2 + d2 )(c2 + d2 )
is divisible by 625.
Problem 7.259 (Romanian National Mathematical Olympiad Small Juniors
Shortlist, 2016). For a positive integer n denote d(n) the number of its positive
divisors and s(n) their sum. It is known that n + d(n) = s(n) + 1, m + d(m) = s(m) + 1,
and nm + d(nm) + 2016 = s(nm). Find n and m.
Problem 7.260 (Romanian National Mathematical Olympiad Small Juniors
Shortlist, 2016). Prove that there are no positive integers of the form
n = aa . . . a} +5a,
| {z k>1
k times

divisible by 2016.
414

Problem 7.261 (Romanian National Mathematical Olympiad Small Juniors

Shortlist, 2016). Find the smallest positive integer of the form
n = aa . . . a} +a(a − 2)2 ,
| {z k>1
k times

divisible by 2016.
Problem 7.262 (Romanian National Mathematical Olympiad Small Juniors
Shortlist, 2016). A positive integer k will be called of type n (n 6= k) if n can be
obtained by adding to k the sum or the product of the digits of k.
(a) Show that there are at least two numbers of type 2016.
(b) Find all numbers of type 216.
Problem 7.263 (Romanian National Mathematical Olympiad Juniors Short-
list, 2016).
(a) Prove that 2n + 3n + 5n + 8n is not a perfect square for any positive integer n.
(b) Find all positive integers n so that
1 n + 4 n + 6 n + 7 n = 2n + 3 n + 5 n + 8 n .

Problem 7.264 (Romanian National Mathematical Olympiad Juniors Short-

list, 2016).
(a) Find all perfect squares of the form aabcc.
(b) Let n be a given positive integer. Prove that there exists a perfect square of the
form
· · · }c.
aab |cc {z
2n times

Problem 7.265 (Romanian National Mathematical Olympiad Juniors Short-

list, 2016). Prove that 2n2 + 27n + 91 is a perfect square for infinitely many positive
integers n.
Problem 7.266 (Romanian National Mathematical Olympiad Seniors Short-
list, 2016). Let p be a prime number and n1 , n2 , . . . , nk ∈ {1, 2, . . . , p − 1} be positive
integers. Show that the equation
xn1 1 + xn2 2 + · · · + xnk k = xpk+1
has infinitely many positive integer solutions.
Problem 7.267 (Romanian National Mathematical Olympiad Seniors Short-
list, 2016). Let n ≥ 4 be a positive integer and define An = {1, 2, . . . , n − 1}. Find
the number of solutions in the set An × An × An × An of the system
(
x + z = 2y,
y + t = 2z.
CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 415

Problem 7.268 (Romanian Stars of Mathematics Junior Level 2016). Show

that there are positive odd integers m1 < m2 < . . . and positive integers n1 < n2 < . . .
such that mk and nk are relatively prime, and mkk − 2n4k is a perfect square for each
index k.

Problem 7.269 (Romanian Stars of Mathematics Junior Level 2016). Given

an integer n ≥
√ 3 and a permutation a1 , a2 , . . . , an of the first n positive integers, show
that at least n distinct residue classes modulo n occur in the list

a1 , a1 + a2 , . . . , a1 + a2 + · · · + an .

Problem 7.270 (Romanian Stars of Mathematics Senior Level 2016). Let n

be a positive integer and let a1 , a2 , . . . , an be n positive integers. Show that
n √ n2
X ak X1
< .
k=1
1 + a1 + a2 + · · · + ak k=1
k

Problem 7.271 (Romania TST for Junior Balkan Mathematical Olympiad

2016). Let M be the set of natural numbers k for which there exists a natural number
n such that
3n ≡ k (mod n).
Prove that M has infinitely many elements.

Problem 7.272 (Romania TST for Junior Balkan Mathematical Olympiad

2016). Let n be an integer greater than 2 and consider the set

A = {2n − 1, 3n − 1, . . . , (n − 1)n − 1}.

Given that n does not divide any element of A, prove that n is a square-free number.
Does it necessarily follow that n is a prime?

Problem 7.273 (Romania TST for Junior Balkan Mathematical Olympiad

2016). Let n be a positive integer and consider the system
(
x2 + ny 2 = z 2
S(n) : ,
nx2 + y 2 = t2

where x, y, z, and t are naturals. If

• M1 = {n ∈ N : system S(n) has infinitely many solutions}, and

• M1 = {n ∈ N : system S(n) has no solutions},

prove that

(a) 7 ∈ M1 and 10 ∈ M2 .

(b) sets M1 and M2 are infinite.

416

Problem 7.274 (Romania TST 2016). Let n be a positive integer and let a1 , a2 , . . . , an
be pairwise distinct positive integers. Show that
n
X 1
< 4,
k=1
[a1 , a2 , . . . , ak ]

where [a1 , a2 , . . . , ak ] is the least common multiple of the integers a1 , a2 , . . . , ak .

Problem 7.275 (Romania TST 2016). Determine the integers k ≥ 2 for which the
sequence
 
2n
(mod k), n = 0, 1, 2, . . . ,
n
is eventually periodic.

Problem
7.276 (Romania TST 2016). Given positive integers k and m, show that
m and nk are coprime for infinitely many integers n ≥ k.

Problem 7.277 (Romania TST 2016). Prove that:

(a) If (an )n≥1 is a strictly increasing sequence of positive integers such that (a2n−1 +
a2n )/an is constant as n runs through all positive integers, then this constant is an
integer greater than or equal to 4; and

(b) Given an integer N ≥ 4, there exists a strictly increasing sequence (an )n≥1 of
positive integers such that (a2n−1 + a2n )/an = N for all indices n.

Problem 7.278 (Romaina TST 2016). Given a positive integer k and an integer
a ≡ 3 (mod 8), show that am + a + 2 is divisible by 2k for some positive integer m.

Problem 7.279 (Romaina TST 2016). Given√a positive integer n, show that for no
set of integers modulo n, whose size exceeds 1 + n + 4, is it possible that the pairwise
sums of unordered pairs be all distinct.
⌊q/p⌋
P
Problem 7.280 (Romania TST 2016). Given a prime p, prove that the sum k p−1
k=1
is not divisible by q for all but finitely many primes q.

Problem 7.281 (Romania TST 2016). Determine the positive integers expressible
x2 +y
in the form xy+1 , for at least two pairs (x, y) of positive integers.

Problem 7.282 (All-Russian Olympiads 2016, Grade 11). Let n be a positive

integer and let k0 , k1 , . . . , k2n be nonzero integers such that k0 + k1 + · · · + k2n 6= 0.
Is it always possible to a permutation (a0 , a1 , . . . , a2n ) of (k0 , k1 , . . . , k2n ) so that the
equation

a2n x2n + a2n−1 x2n−1 + · · · + a0 = 0

has not integer roots?

CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 417

Problem 7.283 (San Diego Math Olympiad 2016). Let a, b, c, d be four integers.
Prove that
(b − a) (c − a) (d − a) (d − c) (d − b) (c − b)
is divisible by 12.
Problem 7.284 (San Diego Math Olympiad 2016). Quadratic equation x2 + ax +
b + 1 = 0 have 2 positive integer roots, for integers a, b. Show that a2 + b2 is not a
prime.
Problem 7.285 (Saudi Arabia Preselection Test 2016). Let p be a given prime.
For each prime r, we define
(prp − 1)(p − 1)
F (r) = .
(pr − 1)(pp − 1)
1. Show that F (r) is a positive integer for any prime r 6= p.
2. Show that F (r) and F (s) are coprime for any primes r and s such that r 6= p, s 6= p
and r 6= s.
3. Fix a prime r 6= p. Show that there is a prime divisor q of F (r) such that p|q − 1
but p2 ∤ q − 1.
Problem 7.286 (Saudi Arabia Preselection Test 2016). Let u and v be positive
rational numbers with u 6= v. Assume that there are infinitely many positive integers
n with the property that un − v n is an integer. Prove that u and v are integers.
Problem 7.287 (Saudi Arabia Preselection Test 2016). Let a and b be two
positive integers such that
b + 1 | a2 + 1 and a + 1 | b2 + 1
Prove that both a and b are odd.
Problem 7.288 (Saudi Arabia Preselection Test 2016).
1. Prove that there are infinitely many positive integers n such that there exists a
permutation of 1, 2, 3, . . . , n with the property that the difference between any
two adjacent numbers is equal to either 2015 or 2016.
2. Let k be a positive integer. Is the statement in part 1 still true if we replace the
numbers 2015 and 2016 by k and k + 2016, respectively?
Problem 7.289 (Saudi Arabia Preselection Test 2016). Let n be a given positive
integer. Prove that there are infinitely many pairs of positive integers (a, b) with a, b > n
such that
2015
Y
(a + i) | b(b + 2016),
i=1
2015
Y
(a + i) ∤ b,
i=1
2015
Y
(a + i) ∤ (b + 2016).
i=1
418

Problem 7.290 (Saudi Arabia TST for Gulf Mathematical Olympiad 2016).
Find all positive integer n such that there exists a permutation (a1 , a2 , . . . , an ) of
(1, 2, 3, . . . , n) satisfying the condition:

a1 + a2 + · · · + ak is divisible by k for each k = 1, 2, 3, . . . , n.

Problem 7.291 (Saudi Arabia TST for Balkan Mathematical Olympiad 2016).
Show that there are infinitely many positive integers n such that n has at least two
prime divisors and 20n + 16n is divisible by n2 .
Problem 7.292 (Saudi Arabia TST for Balkan Mathematical Olympiad 2016).
Let m and n be odd integers such that (n2 − 1) is divisible by m2 + 1 − n2 . Prove that
|m2 + 1 − n2 | is a perfect square.
Problem 7.293 (Saudi Arabia TST for Balkan Mathematical Olympiad 2016).
Let a > b > c > d be positive integers such that

a2 + ac − c2 = b2 + bd − d2 .

Prove that ab + cd is a composite number.

Problem 7.294 (Saudi Arabia TST for Balkan Mathematical Olympiad 2016).
For any positive integer n, show that there exists a positive integer m such that n divides
2016m + m.
Problem 7.295 (Saudi Arabia TST for Balkan Mathematical Olympiad 2016).
Let d be a positive integer. Show that for every integer S, there exist a positive integer
n and a sequence a1 , a2 , . . . , an ∈ {−1, 1} such that

S = a1 (1 + d)2 + a2 (1 + 2d)2 + · · · + an (1 + nd)2 .

Problem 7.296 (Saudi Arabia TST for Balkan Mathematical Olympiad 2016).
Let p and q be given primes and the sequence (pn )n≥1 defined recursively as follows:
p1 = p, p2 = q, and pn+2 is the largest prime divisor of the number (pn + pn+1 + 2016)
for all n ≥ 1. Prove that this sequence is bounded. That is, there exists a positive real
number M such that an < M for all positive integers n.
Problem 7.297 (Saudi Arabia IMO TST 2016). Let n ≥ 3 be an integer and let
x1 , x2 , . . . , xn be n distinct integers. Prove that

(x1 − x2 )2 + (x2 − x3 )2 + · · · + (xn − x1 )2 ≥ 4n − 6.

Problem 7.298 (Saudi Arabia IMO TST 2016). Let k be a positive integer. Prove
that there exist integers x and y, neither of which divisible by 7, such that

x2 + 6y 2 = 7k .

Problem 7.299 (Saudi Arabia IMO TST 2016). Define the sequence a1 , a2 , . . . as
follows: a1 = 1, and for every n ≥ 2, an = n − 2 if an−1 = 0 and an = an−1 − 1,
otherwise. Find the number of 1 ≤ k ≤ 2016 such that there are non-negative integers
r and s and a positive integer n satisfying k = r + s and an+r = an + s.
CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 419

Problem 7.300 (Saudi Arabia IMO TST 2016). Let a be a positive integer. Find
all prime numbers p with the following property: there exist exactly p ordered pairs of
integers (x, y), with 0 ≤ x, y ≤ p − 1, such that p divides y 2 − x3 − a2 x.

Problem 7.301 (Saudi Arabia IMO TST 2016). Find the number of permuta-
tions (a1 , a2 , . . . , a2016 ) of the first 2016 positive integers satisfying the following two
conditions:

1. ai+1 − ai ≤ 1 for all i = 1, 2, . . . , 2015, and

2. There are exactly two indices i < j with 1 ≤ i < j ≤ 2016 such that ai = i and
aj = j.

Problem 7.302 (Saudi Arabia IMO TST 2016). Call a positive integer N ≥ 2
special if for every k such that 2 ≤ k ≤ N , N can be expressed as a sum of k positive
integers that are relatively prime to N (although not necessarily relatively prime to
each other). Find all special positive integers.

Problem 7.303 (Serbia Additional TST 2016). Let w(x) be largest odd divisor of
x. Let a, b be natural numbers such that (a, b) = 1 and a + w(b + 1) and b + w(a + 1)
are powers of two. Prove that a + 1 and b + 1 are powers of two.

Problem 7.304 (Serbia National Olympiad 2016). Let n > 1 be an integer. Prove
that there exist m > nn such that
nm − mn
m+n
is a positive integer.

Problem 7.305 (Serbia National Olympiad 2016). Let a1 , a2 , . . . , a22016 be positive

integers not bigger than 2016. We know that for each n ≤ 22016 , a1 a2 . . . an + 1 is a
perfect square. Prove that for some i , ai = 1.

Problem 7.306 (Serbia TST for Junior Balkan Mathematical Olympiad 2016).
Find minimal number of divisors that can number |2016m − 36n | have,where m and n
are natural numbers.

Problem 7.307 (Slovakia Domestic Category B Mathematical Olympiad 2016).

Let k, l, and m be positive integers such that

k + m + klm 2051
= .
lm + 1 44
Find all possible values for klm.

Problem 7.308 (Slovakia Domestic Category B Mathematical Olympiad 2016).

A positive integer has the property that the number of its even divisors is 3 more than
the number of its odd divisors. What is the ratio of sum of all even divisors over the
sum of all odd divisors of this number? Find all possible answers.
420

Problem 7.309 (Slovakia Domestic Category C Mathematical Olympiad 2016).

Find all possible values for the product pqr, where p, q, and r are primes satisfying

p2 − (q + r)2 = 637.

Problem 7.310 (Slovakia School Round Category C Mathematical Olympiad

2016). Find all four digit numbers abcd such that

abcd = 20 · ab + 16 · cd.

Problem 7.311 (Slovakia Regional Round Category B Mathematical Olympiad

2016). Determine all positive integers k, l, and m such that
3l + 1 lm + 1
= .
3kl + k + 3 5lm + m + 5
Problem 7.312 (Slovakia Regional Round Category C Mathematical Olympiad
2016). Find the least possible value of

3x2 − 12xy + y 4 ,

where x and y are non-negative integers.

Problem 7.313 (Slovakia National Round Category A Mathematical Olympiad
2016). Let p > 3 be a prime. Determine the number of all 6−tuples (a, b, c, d, e, f ) of
positive integers with sum 3p such that
a+b b+c c+d d+e e+f
, , , ,
c+d d+e e+f f +a a+b
are all integers.
Problem 7.314 (Slovakia TST 2016). Let n be a positive integer and let Sn be the
set of all positive divisors of n (including 1 and n). Prove that the rightmost digit of
more than half of the elements of Sn is 3.
Problem 7.315 (Slovakia TST 2016). Find all odd integers M for which the se-
quence a0 , a1 , a2 , . . . defined by a0 = 12 (2M + 1) and ak+1 = ak ⌊ak ⌋ for k = 0, 1, 2, . . .
contains at least one integer.
Problem 7.316 (South Africa National Olympiad 2016). Let k and m be integers
with 1 < k < m. For a positive integer i, let Li be the least common multiple of
1, 2, . . . , i. Prove that k is a divisor of
   
m m−k
Li · −
i i
for all i ≥ 1.
Problem 7.317 (Slovenia National Math Olympiad First Grade 2016). Find
all relatively prime integers x and y that solve the equation

4x3 + y 3 = 3xy 2 .
CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 421

Problem 7.318 (Slovenia National Math Olympiad Fourth Grade 2016). Find
all integers a, b, c, and d that solve the equation

a2 + b2 + c2 = d + 13,
d
a + 2b + 3c = + 13.
2
Problem 7.319 (Slovenia IMO TST 2016). Let

N = 215 · 2015.

How many divisors of N 2 are strictly smaller than N and do not divide N ?
Problem 7.320 (Slovenia IMO TST 2016, Philippine 2015). Prove that for all
positive integers n ≥ 2,
r r r
1 1 3 2 n n − 1 n2
+ + + ··· + < .
2 2 3 n n+1
Problem 7.321 (Slovenia IMO TST 2016, Romania JBMO TST 2015). Find
all positive integers a, b, c, and d such that

4a · 5b − 3c · 11d = 1.

Problem 7.322 (Spain National Olympiad 2016). Two real number sequences are
guiven, one arithmetic (an )n∈N and another geometric sequence (gn )n∈N none of them
constant. Those sequences verifies a1 = g1 6= 0, a2 = g2 and a10 = g3 . Find with proof
that, for every positive integer p, there is a positive integer m, such that gp = am .
Problem 7.323 (Spain National Olympiad 2016). Given a positive prime number
p. Prove that there exist a positive integer α such that p|α(α − 1) + 3, if and only if
there exist a positive integer β such that p|β(β − 1) + 25.
Problem 7.324 (Spain National Olympiad 2016). Let m be a positive integer
and a and b be distinct positive integers strictly greater than m2 and strictly less than
m2 + m. Find all integers d such that m2 < d < m2 + m and d divides ab.
Problem 7.325 (Switzerland Preliminary Round 2016). Determine all natural
numbers n such that for all positive divisors d of n,

d + 1 | n + 1.

Problem 7.326 (Switzerland Final Round 2016). Find all positive integers n for
which primes p and q exist such that

p(p + 1) + q(q + 1) = n(n + 1)

Problem 7.327 (Switzerland Final Round 2016). Let an be a sequence of positive

integers defined by a1 = m and an = a2n−1 − 1 for n = 2, 3, 4, . . . . A pair (ak , al ) is
called interesting if
422

(i) 0 < l − k < 2016, and

(ii) ak divides al .
Prove that there exists a positive integer m such that the sequence an contains no
interesting pair.
Problem 7.328 (Switzerland TST 2016). Let n be a positive integer. We call a
pair of natural numbers incompatible if their greatest common divisor is equal to 1.
Find the minimum value of incompatible pairs when one divides the set {1, 2, . . . , 2n}
into n pairs.
Problem 7.329 (Switzerland TST 2016). Let n be a positive integer. Show that
n
77 + 1 has at least 2n + 3 prime divisors (not necessarily distinct).
Problem 7.330 (Switzerland TST 2016). Find all positive integers n such that
X
d2 = 5(n + 1).
d|n
1≤d≤n

Problem 7.331 (Syria Central Round First Stage 2016). A positive integer
n ≥ 2 is called special if n2 can be written as sum of n consecutive positive integers
(for instance, 3 is special since 32 = 2 + 3 + 4).
(i) Prove that 2016 is not special.

(ii) Prove that the product of two special numbers is also special.
Problem 7.332 (Syria Central Round Second Stage 2016). Find all integers a
and b such that a3 − b2 = 2.
Problem 7.333 (Syria TST 2016). Find all positive integers m and n such that
1 1 3
+ = .
m n 2014
Problem 7.334 (Taiwan TST First Round 2016). Find all ordered pairs (a, b) of
positive integers that satisfy a > b and the equation (a − b)ab = ab ba .
Problem 7.335 (Taiwan TST Second Round 2016). Let a and b be positive
integers such that a! + b! divides a!b!. Prove that 3a ≥ 2b + 2.
Problem 7.336 (Taiwan TST Second Round 2016). Let hFn i be the Fibonacci
sequence, that is, F0 = 0, F1 = 1, and Fn+2 = Fn+1 + Fn holds for all nonnegative
integers n. Find all pairs (a, b) of positive integers with a < b such that Fn − 2nan is
divisible by b for all positive integers n.
Problem 7.337 (Taiwan TST Third Round 2016). Let n be a positive integer.
Find the number of odd coefficients of the polynomial (x2 − x + 1)n .
Problem 7.338 (Taiwan TST Third Round 2016). Let k be a positive integer. A
sequence a0 , a1 , . . . , an (n > 0) of positive integers satisfies the following conditions:
CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 423

(i) a0 = an = 1;
(ii) 2 ≤ ai ≤ k for each k = 1, 2, . . . , n − 1
(iii) For each j = 2, 3, . . . , k, the number j appears ϕ(j) times in the sequence a0 , a1 , . . . , an
(ϕ(j) is the number of positive integers that do not exceed j and are coprime to
j);
(iv) For any i = 1, 2, . . . , n − 1, gcd(ai−1 , ai ) = 1 = gcd(ai , ai+1 ), and ai divides
ai−1 + ai+1
There is another sequence b0 , b1 , . . . , bn of integers such that
bi+1 bi
>
ai+1 ai
for all i = 0, 1, . . . , n − 1. Find the minimum value for bn − b0 .
Problem 7.339 (Taiwan TST Third Round 2016). Let f (x) be the polynomial
with integer coefficients (f (x) is not constant) such that

(x3 + 4x2 + 4x + 3)f (x) = (x3 − 2x2 + 2x − 1)f (x + 1)

Prove that for each positive integer n ≥ 8, f (n) has at least five distinct prime divisors.
Problem 7.340 (Turkey TST for European Girls’ Mathematical Olympiad
2016). Prove that for every square-free integer n > 1, there exists a prime number p
and an integer m satisfying

p | n and n | p2 + p · mp .

Problem 7.341 (Turkey TST for Junior Balkan Mathematical Olympiad

2016). Let n be a positive integer, p and q be prime numbers such that

pq | np + 2 and n + 2 | np + q p .

Prove that there exists a positive integer m satisfying q | 4m · n + 2.

Problem 7.342 (Turkey TST for Junior Balkan Mathematical Olympiad
2016). Find all pairs (p, q) of prime numbers satisfying

p3 + 7q = q 9 + 5p2 + 18p.

Problem 7.343 (Turkey TST 2016). p is a prime. Let Kp be the set of all polyno-
mials with coefficients from the set {0, 1, . . . , p−1} and degree less than p. Assume that
for all pairs of polynomials P, Q ∈ Kp such that P (Q(n)) ≡ n (mod p) for all integers
n, the degrees of P and Q are equal. Determine all primes p with this condition.
Problem 7.344 (Turkmenistan Regional Olympiad 2016). Find all distinct prime
numbers p, q, r, s such that
1 1 1 1 1
1− − − − = .
p q r s pqrs
424

Problem 7.345 (Tuymaada Senior League 2016). For each positive integer k
determine the number of solutions of the equation

8k = x3 + y 3 + z 3 − 3xyz

in non-negative integers x, y, and z such that 0 ≤ x ≤ y ≤ z.

Problem 7.346 (Tuymaada Senior League 2016). The ratio of prime numbers p
and q does not exceed 2 (p 6= q). Prove that there are two consecutive positive integers
such that the largest prime divisor of one of them is p and that of the other is q.

Problem 7.347 (Tuymaada Junior League 2016). Is there a positive integer N >
1020 such that all its decimal digits are odd, the numbers of digits 1, 3, 5, 7, 9 in its
decimal representation are equal, and it is divisible by each 20−digit number obtained
from it by deleting digits? (Neither deleted nor remaining digits must be consecutive.)

Problem 7.348 (Ukraine TST for UMO 2016). Find all numbers n such, that in
[1; 1000] there exists exactly 10 numbers with digit sum equal to n.

Problem 7.349 (Ukraine TST for UMO 2016). Number 125 is written as the sum
of several pairwise distinct and relatively prime numbers, greater than 1. What is the
maximal possible number of terms in this sum?

Problem 7.350 (Ukraine TST for UMO 2016). Given prime number p and differ-
2 2
ent natural numbers m, n such that p2 = m +n
2
. Prove that 2p − m − n is either square
or doubled square of an integer number.

Problem 7.351 (Ukraine TST for UMO 2016). Solve the equation n(n2 + 19) =
m(m2 − 10) in positive integers.

Problem 7.352 (USA AIME 2016). For −1 < r < 1, let S(r) denote the sum of
the geometric series
12 + 12r + 12r2 + 12r3 + · · · .
Let a between −1 and 1 satisfy S(a)S(−a) = 2016. Find S(a) + S(−a).

Problem 7.353 (USA AIME 2016). For a permutation p = (a1 , a2 , . . . , a9 ) of the

digits 1, 2, . . . , 9, let s(p) denote the sum of the three 3-digit numbers a1 a2 a3 , a4 a5 a6 ,
and a7 a8 a9 . Let m be the minimum value of s(p) subject to the condition that the units
digit of s(p) is 0. Let n denote the number of permutations p with s(p) = m. Find
|m − n|.

Problem 7.354 (USA AIME 2016). A strictly increasing sequence of positive in-
tegers a1 , a2 , a3 , . . . has the property that for every positive integer k, the subsequence
a2k−1 , a2k , a2k+1 is geometric and the subsequence a2k , a2k+1 , a2k+2 is arithmetic. Sup-
pose that a13 = 2016. Find a1 .

Problem 7.355 (USA AIME 2016). Find the least positive integer m such that
m2 − m + 11 is a product of at least four not necessarily distinct primes.
CHAPTER 7. PRACTICE CHALLENGE PROBLEMS 425

Problem 7.356 (USA AIME 2016). Let x, y and z be real numbers satisfying the
system

log2 (xyz − 3 + log5 x) = 5,

log3 (xyz − 3 + log5 y) = 4,
log4 (xyz − 3 + log5 z) = 4.

Find the value of | log5 x| + | log5 y| + | log5 z|.

Problem 7.357 (USA AIME 2016). For polynomial P (x) = 1 − 13 x + 16 x2 , define
50
X
3 5 7 9
Q(x) = P (x)P (x )P (x )P (x )P (x ) = ai xi .
i=0

Then,
50
X m
|ai | = ,
i=0
n
where m and n are relatively prime positive integers. Find m + n.
Problem 7.358 (USA AIME 2016). Find the number of sets {a, b, c} of three dis-
tinct positive integers with the property that the product of a, b, and c is equal to the
product of 11, 21, 31, 41, 51, and 61.
Problem 7.359 (USA AIME 2016). The sequences of positive integers 1, a2 , a3 , . . .
and 1, b2 , b3 , . . . are an increasing arithmetic sequence and an increasing geometric se-
quence, respectively. Let cn = an + bn . There is an integer k such that ck−1 = 100 and
ck+1 = 1000. Find ck .
Problem 7.360 (USA AIME 2016). For positive integers N and k, define N to be
k-nice if there exists a positive integer a such that ak has exactly N positive divisors.
Find the number of positive integers less than 1000 that are neither 7-nice nor 8-nice.
Problem 7.361 (USAJMO 2016). Prove that there exists a positive integer n < 106
such that 5n has six consecutive zeros in its decimal representation.
Problem 7.362 (USAMO 2016). Prove that for any positive integer k,
k−1
2
Y j!
(k )! ·
j=0
(j + k)!

is an integer.
Problem 7.363 (USAMO 2016).
(a) Prove that if n is an odd perfect number then n has the following form

n = ps m 2

where p is prime has form 4k +1, s is positive integers has form 4h+1, and m ∈ Z+ ,
m is not divisible by p.
426
n(n+1)
(b) Find all n ∈ Z+ , n > 1 such that n − 1 and 2
is perfect number.

Problem 7.364 (USA TSTST 2016). Decide whether or not there exists a noncon-
stant polynomial Q(x) with integer coefficients with the following property: for every
positive integer n > 2, the numbers

Q(0), Q(1), Q(2), . . . , Q(n − 1)

produce at most 0.499n distinct residues when taken modulo n.

Problem 7.365 (USA TSTST 2016). Suppose that n and k are positive integers
such that
1 = ϕ(ϕ(. . . ϕ( n) . . . )).
| {z }
k times

Prove that n ≤ 3k .
√
Problem√7.366 (USA TST 2016). Let 3 = 1.b1 b2 b3 . . .(2) be the binary represen-
tation of 3. Prove that for any positive integer n, at least one of the digits bn , bn+1 ,
. . . , b2n equals 1.

Problem 7.367 (Venezuela Final Round Fourth Year 2106). Find all pairs of
prime numbers (p, q), with p < q, such that the numbers p + 2q, 2p + q and p + q − 22
are also primes.

Problem 7.368 (Zhautykov Olympiad 2016). a1 , a2 , ..., a100 are permutation of

1, 2, ..., 100. S1 = a1 , S2 = a1 + a2 , ..., S100 = a1 + a2 + ... + a100 Find the maximum
number of perfect squares from Si .

Problem 7.369 (Zhautykov Olympiad 2016). We call a positive integer q a con-

venient denominator for a real number α if
 
α −  < 1
p
 
 q  10q

for some integer p. Prove that if two irrational numbers α and β have the same set of
convenient denominators then either α + β or α − β is an integer. This PDF was created
for Amazon.