Professional Documents
Culture Documents
Semester – III
Paper Code – 21MAT23DB1
ISBN :
Price :
Publisher: Maharshi Dayanand University Press
Publication Year : 2022
Contents
⚝⚝⚝
Master of Science (Mathematics)(DDE)
SECOND YEAR
Third Semester
Analytical Number Theory
Paper Code: 21MAT23DB1
Time: 03 Hours
Max Marks : 80
Course Outcomes
Students would be able to:
CO1 Know about the classical results related to prime numbers and get familiar with the irrationality of
e and Π.
CO2 Study the algebraic properties of Un and Qn.
CO3 Learn about the Waring problems and their applicability.
CO4 Learn the definition, examples and simple properties of arithmetic functions and about perfect
numbers.
CO5 Understand the representation of numbers by two or four squares.
Section - I
Distribution of primes, Fermat and Mersenne numbers, Farey series and some results concerning Farey
series, Approximation of irrational numbers by rationals, Hurwitz theorem, Irrationality of e and π..
Section - II
The arithmetic in Zn, The group Un, Primitive roots and their existence, the group Upn (p-odd) and
U2n, The group of quadratic residues Qn , Quadratic residues for prime power moduli and arbitrary
moduli, The algebraic structure of Un and Qn .
Section -III
Riemann Zeta Function ζ(s) and its convergence, Application to prime numbers, ,ζ (s) as Euler product,
Evaluation of ,ζ (2) and ,ζ (2k). Diophantine equations ax + by = c, x2+y2 = z2 and x4+y4 = z4, The
representation of number by two or four squares, Waring problem, Four square theorem, The numbers
g(k) & G(k), Lower bounds for g(k) & G(k).
Section - IV
Arithmetic functions φ (n), τ (n), σ (n) and σk (n), U(n), N(n), I(n), Definitions and examples and simple
properties, Perfect numbers, Mobius inversion formula, The Mobius function µn, The order and average
order of the function n φ(n), τ(n) and σ(n).
Note: The question paper of each course will consist of five Sections. Each of the sections I to IV will
contain two questions and the students shall be asked to attempt one question from each.
Section-V shall be compulsory and will contain eight short answer type questions without any
internal choice covering the entire syllabus.
Books Recommended:
1. G.H. Hardy and E.M. Wright, An Introduction to the Theory of Numbers.
2. D.M. Burton, Elementary Number Theory.
3. N.H. McCoy, The Theory of Number by McMillan.
4. I. Niven, I. and H.S. Zuckermann, An Introduction to the Theory of Numbers.
5. A. Gareth Jones and J Mary Jones, Elementary Number Theory, Springer Ed. 1998.
Unit– I | Primes in Certain Arithmetical Progressions
Structure
1.0 Introduction
1.1 Unit Objectives
1.2 Distribution of primes
1.3 Fermat and Mersenne numbers
1.4. Farey series and some results concerning Farey series
1.5 Irrationality of e and π
1.6 Approximation of irrational numbers by rationals
1.7 Hurwitz theorem
1.8 Check your progress
1.9 Summary
1.10 Keywords
1.11 References
1.0 Introduction
In this unit, we will study the distribution of primes, Fermat and Mersenne numbers, and approximation
of irrational numbers by rationals and Hurwitz theorem.
1.1 Unit Objectives
After going through this unit, one will be able to
• know about the classical results related to prime numbers.
• be familiar with irrationality of e and π .
• find approximation of irrational numbers by rationals.
• get aware of Hurwitz theorem.
1.2 Distribution of primes
1.2.1 Definition (Peano Axioms).
(i) 1 ∈ N .
(ii) For all natural numbers n, there exists its successor n + 1∈N .
Now, since n1 < n , therefore by induction hypothesis n1 has a prime divisor, say p, then p n1 .
Also, n1 n ⇒ p n.
Consider
= N p1 p2 ....... pr + 1.
Then, p p1 p2 ........ pr .
Then N > 1. So, N can be written as product of primes. Now N is odd, therefore N can be written as the
product of odd primes that is, N can be written as a product of primes of the form 4n + 1 and 4n + 3. But
if N were divisible by primes of the form 4n + 1 only, then N would also be of the form 4n + 1. But N is
of the form 4n + 3. So N is divisible by atleast one prime of the form 4n + 3, say p. But only primes of
the form 4n + 3 are p1 , p2 ,......., pr then p = pi for some i
Now, p N ⇒ p 4 p1 p2 ........ pr − N
⇒ p1
which is a contradiction.
1.2.8 Theorem. Number of primes of the form 6n + 5 is infinite.
Proof. If possible, suppose that number of primes of the form 6n + 5 is finite and let these
be p1 , p2 ......, pr .
Consider
= N 6 p1 p2 ...... pr − 1.
Then, N > 1. Therefore, N can be written as product of primes of the form 6n + 1 and 6n + 5. If N were
divisible by primes of the form 6n + 1 only, then N would be of the form 6n +1. But N is of the form 6n
+ 5 so N is divisible by atleast one prime of the form 6n + 5, say p. But only primes of the form 6n + 5
are p1 , p2 ,.........., pr . Therefore p = pi for some i.
1.2.9 Theorem (Without Proof). If (a, b) = 1, then every odd prime factor of a 2 + b 2 must be of the
form (4n +1).
For example, (4, 3) =1 and 42 + 32 =
25 has an odd prime factor 5 which is of the form 4n + 1.
1.2.10 Definition (Theorem). Prove that primes of the type 4n + 1 are infinite in number.
Proof. If possible, suppose that p1 , p2 ,......., pr are the only primes of the form 4n + 1.
Here, N is of the form a 2 + b 2 with (a , b) =1 . Also, N is odd so 2 does not divide N and all the prime
factors of N are odd. Therefore, by theorem 1.2.9, all the prime factors of N must be of the form 4n + 1.
Let p N , then p will be of the form 4n + 1. But the only primes of the form 4n + 1 are p1 , p2 ,....., pr .
Therefore p = pi for some i.
Now p N and p (2 p1 p2 ....... pr )2 ⇒ p1
which is a contradiction. Hence, the primes of the form 4n + 1 are infinite in number.
1.2.11 Theorem. Prove that primes of the type 8n + 5 are infinite in number.
Proof. If possible, suppose that p1 , p2 ,......., pr be the only primes of the form 8n + 5.
= ( p1 p2 ..... pr )2 + 4.
)2 4m 2 + 4m + 1
(2m + 1=
= 4m(m + 1) + 1
= 8m+ 1.
So, N is of the form 8n + 5.
Also, N is odd, therefore, by Theorem 1.2.10 every prime factor of N must be of the form 4n + 1. Now,
any number of the type 4n + 1 is either of the form 8n + 1 or 8n + 5. Thus, the prime factors of N are
either of the form 8n + 1 or 8n + 5. If every prime factor of N is of the type 8n + 1, then N would also be
of the form 8n + 1. But N is of the form 8n + 5 and so one factor of N should be of the form 8n + 5, say
p. Therefore, p = pi for some i.
Now, p N and p ( p1 p2 ..... pr )2 ⇒ p N − ( p1 p2 ......... pr )2
⇒ p/4
⇒ p≤ 4 .
But smallest prime of the form 8n + 5 is 5, so this is a contradiction and therefore, primes of the form 8n
+ 5 are infinite in number.
F4 = 65537
are all primes. Fn’s are called Fermat numbers. A Fermat number which is a prime is called a Fermat
prime. However, no Fermat primes are known beyond F4. In 1732, Euler proved that F5 is a composite
number. However, his proof was very complicated. The following proof is given by Fermat:
1.3.2 Theorem. F5 is a composite number.
Proof. Let = 7
a 2= 128, b 5.
=
Then a= b3 + 3 or a − b3 =
3.
Now, 1 + ab − b 4 = 1 + b(a − b3 ) = 1 + 3b = 16 = 24.
5
F
=5 22 + 1
( )
22
23
=2 + 1 =(28 )4 + 1
=(2 . 27 )4 + 1 =(2a)4 + 1
= (a)4 (2)4 + 1 = a 4 (1 + ab − b 4 ) + 1
= (1 + ab) a 4 + (1 − a 4b 4 )
=(1 + ab) a 4 + (1 + ab)(1 − ab)(1 + a 2 b 2 )
=(1 + ab) a 4 + (1 − ab)(1 + a 2 b 2 ) .
1.3.3 Remark. No Fermat prime number beyond F4 is obtained and research is still on. However, it is
conjectured that Fn is not prime for n > 4.
1.3.4 Theorem. All Fermat numbers are relatively prime to each other, that is, ( Fm=
, Fn ) 1 for m ≠ n .
( ) n 2k
= 22 + 1.
n k
Set 22 = x, then F
=m x2 + 1 .
k
Fm − 2 x 2 + 1 − 2
Consider, =
Fn x +1
10 The Group of Units and Quadratic Residues
k
x2 −1
=
x +1
k −1 k −1
( x2 + 1) ( x 2 − 1)
=
x +1
k −1 k −2 k −3
( x2 + 1) ( x 2 + 1) ( x 2 + 1)..........( x − 1)( x + 1)
=
x +1
k −1 k −2
= ( x2 + 1)( x 2 + 1).........( x − 1).
⇒ Fn ( Fm − 2) .
Let g . c. d . ( Fm , Fn ) = d .
⇒ d Fm , d Fn and also Fn Fm − 2
⇒ d Fm − 2
⇒ d Fm − ( Fm − 2)
⇒ d 2
⇒ d = 1 or 2.
⇒ pi ≠ p j for i ≠ j
So all pi’s are distinct primes. Thus, given any natural number n there exists at least n different primes
and so number of primes is infinite.
n
1.3.6 Corollary. pn +1 ≤ 22 + 1 =Fn where pi denotes ith prime in ascending order.
Proof. Since each Fi is divisible by a different prime and F1 < F2 < F3 < ............. < Fn , so there exist n
primes ≤ Fn. But all Fermat numbers are odd and prime 2 is less than all odd primes so there exist atleast
n + 1 primes less than or equal to Fn.
Analytical Number Theory 11
⇒ F2 ≡ 7(mod10).
= 100r 2 + 120r + 36
= 10(10r + 12r + 3) + 6
⇒ 10 ( Fk +1 − 7) .
Proof. Since a n − 1 =r is a prime, so it cannot have any factor q such that 1 < q < r.
⇒ (a − 1) a n − 1.
But a ≥ 2 , n ≥ 2. If a > 2
⇒ a − 1 > 1 is a factor of a n − 1
which is a contradiction, thus a = 2.
Again, suppose n is composite. This implies there exist p and q with 1 < p < n, 1 < q < n such that n =
pq. Now,
a n − 1 =a pq − 1 =(a q ) p − 1
( )
= (a q − 1) a q
p −1
+ (a q ) p − 2 + ........ + 1 .
Now, since a = 2, 1 < q < n , so 1 < a q − 1 < a n − 1 is a factor of a n − 1 which implies that a n − 1 is
composite, which is a contradiction. Hence, n is prime.
1.3.10 Remark. Converse of the above theorem is not true, that is, 2 p − 1 need not be a prime for all
primes p. For example, 211 − 1 is not a prime. Also,
Proof. If a is odd and a ≥ 2 , then a ≥ 3 and so a n + 1 is an even number and so it cannot be a prime
number. Therefore, for a n + 1 to be prime, a must be even.
Next, we claim that no odd prime divides n. If possible, suppose that an odd prime p divides n. Then,
n pq, 1 < q < n . So, we have
=
a n + 1 =a pq + 1 =(a q ) p + 1
⇒ a q + 1 divides a n + 1 properly.
h h′
Proof. Without loss of generality, we assume that < .
k k′
h h + h′ h′
We claim that < < (1)
k k + k′ k′
h h + h′
< ⇔ h k ′ < h′ k
k k + k′
h h′
⇔ < which is true.
k k′
h + h′ h′
In a similar way, < .
k + k′ k′
Hence, the inequalities in (1) are satisfied.
(a) If possible, suppose that k + k ′ ≤ n .
Since h ≤ k and h ′ ≤ k ′
⇒ h + h′ ≤ k + k ′
h + h′
⇒ ∈ Fn .
k + k′
h + h′ h h′
This implies a fraction of Fn lies between two consecutive fractions , of Fn.
k + k′ k k′
which is a contradiction. So, k + k ′ > n .
14 The Group of Units and Quadratic Residues
h h′
(b) If possible, suppose that k = k ′ , whereand are two consecutive fractions of Fn for some n.
k k′
0 1 h
We note that , are the only two fractions with denominator 1 satisfying 0 ≤ ≤ 1 and (h , k ) =1 .
1 1 k
1 0 1
Since n > 1, ∈ Fn for every n ≥ 2 and so and cannot be consecutive fractions in Fn for n ≥ 2.
2 1 1
Therefore, k≠1
h h′ h′
Now,
k
≠1≠
k′
=
k
[ k ≠ 1 ]
∴ h < h′ < k.
h h h + 1 h′
Now, we claim that < < ≤ . (3)
k k −1 k k
h h h + 1 h′
To prove this, we note that < and ≤ are obvious.
k k −1 k k
h h +1
So it remains to prove that < .
k −1 k
⇒ h k < hk + k − h − 1
⇒ k − h −1 > 0
⇒ k > h + 1 , which is true [(by (2)] .
h
Therefore, inequalities in (3) are proved. Thus, we have a fraction in Fn which lies between two
k −1
h h′
consecutive fraction and , which is a contradiction.
k k′
So, we can’t have k = k′ if n > 1.
h h′ h h′
1.4.3 Theorem. Let and be two successive members of Fn such that < , then
k k′ k k′
1.
h ′k − h k ′ = (1)
h h′ h
Proof. Since < so is not the last fraction of Fn
k k′ k
h
0≤ < 1. (2)
k
Analytical Number Theory 15
x h ′ k ′x − h ′y 1 x h′
Now, −= ≥ y > k ′ ∴ k ′x − h ′y ≥ 1 . (5)
y k′ yk ′ k ′y
h ′ h h ′k − hk ′ 1
Similarly, = − ≥ . (6)
k′ k kk ′ kk ′
1 h′
> . k ′ ≤ n and ∈ Fn
ky k′
16 The Group of Units and Quadratic Residues
But by (3),
x h 1
− =
y k ky
which is a contradiction.
x h′
So, we must have = .
y k′
Since y > 0, k ′ > 0 and g .c.d= h ′, k ′) 1 .
. ( x , y) 1, g .c.d . (=
Clearly,
= x0 2,
= y0 3 is a solution of (1). Choose an integer r such that
0 ≤ n − k < y0 + rk ≤ n.
or 8 < 3 + 5r ≤ 13
or 5 < 5r ≤ 10
or 1 < r ≤ 2.
x x0 + rh 2 + 2.3 8
Thus, r = 2 is the only solution. Then the required fraction
= is = = .
y y0 + rk 3 + 2.5 13
h h′ h′ h
1.4.5 Remark. < ⇒ 1− < 1− .
k k′ k′ k
h h′ h′ h
Further and are consecutive fractions of Fn so 1 − and 1 − are also consecutive fractions of
k k′ k′ k
Fn.
h h′ h h′ h ′′
1.4.6 Theorem. Let and be two consecutive terms of Fn such that < . Let be such that
k k′ k k′ k ′′
h h ′′ h ′ h ′′ h + h ′
, , be consecutive terms of Fr such that r > n then = .
k k ′′ k ′ k ′′ k + k ′
Analytical Number Theory 17
h h ′′ h h ′′
Proof. Since < are consecutive terms of Fr with < and so h ′′k − hk ′′ =
1. (1)
k k ′′ k k ′′
h ′′ h′ h ′′ h ′
Also, and are consecutive terms of Fr and < and so
k ′′ k′ k ′′ k ′
1.
h ′k ′′ − h ′′k ′ = (2)
From (1) and (2), we get
h ′′k − hk ′′ = h ′k ′′ − h ′′k ′
⇒ h ′′(k + k ′=
) k ′′(h + h ′)
h ′′ h + h ′
⇒ = .
k ′′ k + k ′
1.4.7 Remark. Now consider (h + h ′)k − (k + k ′)h= h ′k − hk =
′ 1
that is, ( h + h ′ )k − ( k + k ′ )h =1
⇒ g .c.d . (h + h ′, k + k ′ ) =1.
h ′′ h + h ′
= ⇒ h ′′ =h + h ′ and k ′′ =k + k ′ .
k ′′ k + k ′
h h′ h h′
1.4.8 Theorem. Let and are two consecutive fractions of Fn with < , then among all
k k′ k k′
h h′ h + h′
fractions between and , the fraction is the unique fraction with the smallest denominator.
k k′ k + k′
x h x h′
Proof. Let be any fraction such that < < , then
y k y k′
h′ h h′ x x h
− = − + −
k ′ k k ′ y y k
h ′y − k ′x kx − hy
= + . (1)
k ′y ky
x h′ h′ x
Since < , so − > 0 and so (h ′y − k ′x) ≥ 1 as h ′, k ′, x, y are all integers.
y k′ k′ y
Similarly, (kx − hy) ≥ 1 . Using these values in (1), we get
h′ h 1 1 k + k′
− ≥ + = . (2)
k ′ k k ′y ky k k ′ y
18 The Group of Units and Quadratic Residues
h ′ h h ′k − k ′h 1
But = − = . (3)
k′ k k k′ k k′
From (2) and (3), we get
1 k + k′
≥ ⇒ y ≥ k + k′ . (4)
k k′ k k′ y
h h′ + h h′
Since we know that, < < .
k k′ + k k′
h h′ x
So, there exists a fraction lying between and whose denominator is k + k ′. So, if is a fraction
k k′ y
h h′
with smallest denominator lying between and , we should not have y > k + k ′ , so we must have
k k′
y= k + k ′ in (4). But the equality in (4) will hold only when equality holds in (2). This implies we have
1 and
h ′y − k ′x =
1
kx − hy = ⇒ h ′y − k ′x =kx − hy
⇒ h ′y + hy = kx + k ′x
x h + h′
⇒ = .
y k + k′
a a a
1.4.9 Exercise. If Fn = 1 , 2 ,........, r , then
b1 b2 br
n
(i) r = 1 + ∑ φ( j ) .
j =1
ai 1 nr
∑ =
bi
=i 1= 2
(ii)
1 + ∑
j 1
φ( j ) .
r −1
(iii) ∑ (b
j =1
j b j + 1 ) −1 =
1 , where φ(j) is Euler function.
0 1
and we know that and are only terms in F1. So, the result is true for n = 1.
1 1
Assume that the result is true for all natural numbers < n.
Analytical Number Theory 19
h
Consider Fn. Now, Fn contains all terms of Fn − 1 + those fractions such that g.c.d. (h, n) =1 and h < n.
n
Therefore, by definition the number of terms which are extra are φ(n) .
n
= 1 + ∑ φ( j ) .
j =1
h h
(ii) We know that ∈ Fn ⇔ 1 − ∈ Fn .
k k
ai
So, we write the terms
= (i 1, 2,....., r ) in a row.
bi
a1 a2 a
, ,.........., r
b1 b2 br
a1 a a
and 1− , 1 − 2 ,...........,1 − r in second row.
b1 b2 br
ai a
As runs over terms of Fn , 1 − i must also run over terms of Fn in opposite order. Now, adding the
bi bi
two rows horizontally, we get
r
ai r
ai
∑b ∑
=
=i 1= i 1
1 − b
i i
r
ai
⇒ 2∑ =r
i =1 bi
ai 1 1 rn
∑ =
bi
=i 1= 2
r
= ⇒ ∑ φ( j ) .
2
1 +
j 1
1 0 a a
(iii) We know that the last term of Fn is and the first term is so =
that r 1=
and 1 0 .
1 1 br b1
a ar ar −1 ar −1 ar − 2 a
Now, 1 =r = − + − + ........ + 1 .
br br b r −1 br −1 br − 2 b1
20 The Group of Units and Quadratic Residues
h h′ h h′
But we know that if and are consecutive terms of Fn with < , then h ′k − hk ′ =
1.
k k′ k k′
ai ai −1 ai bi −1 − ai −1 bi 1
So, let us calculate − = = = (bi bi −1 )−1.
bi bi −1 bi bi −1 bi bi −1
Therefore,
r −1
a1 r −1 a1
1= ∑ (b j b j +1 )−1 +
= ∑ (b j b j +1 )−1 = 0 .
j 1=b1 j 1 b1
1.5 Irrationality of e and π
a
1.5.1 Definition (Irrational Numbers). A real number α is called a rational if α can be written as
b
where a, b are integers and b ≠ 0. Otherwise α is an irrational number.
a
Note. A rational number α is said to be in its standard form if α = and g.c.d.(a, b)=1.
b
Proof. Let, if possible, 2 be a rational number, then it can be written in the standard form, that is,
a
= 2 , b > 0 and (a= , b) 1.
b
Squaring both sides, we obtain
2b 2 = a 2 .
Now, 2 is a prime number and 2 a 2
⇒ 2a
⇒ a = 2c.
Now, substituting these values, we get
2b 2 = 4c 2
⇒ b 2 = 2c 2
⇒ 2b
⇒ (a , b) = 2 which is a contradiction.
Hence, 2 is irrational.
a
=p where=
b > 0, (a, b) 1.
b
⇒ q / a2
⇒ q a
⇒ q gcd (a , b)
⇒ q 1 ; which is a contradiction.
Hence, p is irrational.
1.5.4 Theorem. m
n , that is, mth root of n (where n is natural number) is irrational unless n is the mth
power of an integer.
a = p s a1 and p / a1 .
∴ ( p s a1 )m .
bm n = (2)
⇒ n is mth power.
22 The Group of Units and Quadratic Residues
x m + c1 x m −1 + c2 x m − 2 + ......... + cm −1 x + cm =
0 (1)
If α is an irrational, then there is nothing to prove. If α is a rational, then we shall prove that its
denominator is unity (1).
a a
Let α = , where g.c.d.(a, b) = 1, b > 0, then satisfies (1) and so
b b
m m −1
a a a
+ c1 + ......... + cm −1 + cm =
0
b b b
⇒ b am .
If possible, suppose that b > 1 and let p be a prime factor of b, then p a m and p/a, which is a
contradiction to the fact that (a, b) = 1
⇒ b 1
= ⇒ α a.
=
Thus, α is an integer.
1.5.6 Theorem. log10 2 is irrational.
Proof. If possible, suppose that log10 2 be rational and let
a
log10=
2 , (a , =
b) 1, b > 0
b
a
⇒ 2 =(10) b
⇒ 2b = 10a. (1)
We consider different cases:
If a > 0, then 5 divides R.H.S. of (1) but 5 does not divide L.H.S. of (1), which is a contradiction. So, (1)
cannot hold if a > 0.
If a < 0, then b < 0.
Analytical Number Theory 23
−a a
Also behaves like so by the above arguments a cannot be less than zero.
−b b
1 1 1
Now, consider b ! e − 1 + + + ........ + = α.
1! 2! b !
a
Then α is an integer since e = is rational. Also, by definition of e, α > 0.
b
1 1
Also, α = b! + + ..........
(b + 1)! (b + 2)!
1 1
= + + ...........
b + 1 (b + 1)(b + 2)
1 1
< + + ................
b + 1 (b + 1)2
1
1
= b +1 = < 1. [ b > 1]
1 b
1−
b +1
Thus, 0 < α < 1, which is a contradiction because no integer lies between 0 and 1 and so e is irrational.
x n (1 − x)n
1.5.8 Lemma. Let f ( x) = , then f (0), f (1) and f i (0) , f i (1) are all integers for all i ≥ 0.
n!
1
Also, 0 < f ( x) < whenever x ∈(0,1) .
n!
1 2n
Proof. We can rewrite f ( x) as f ( x) = ∑ Ci xi .
n! i = n
Let i ≥ 1. Now, in f i ( x), for i < n , we do not have any constant term and so f i (0) = 0 for i < n. Further
f(x) is of degree 2n and so f i (0) = 0 for all i > 2n . So let n ≤ i ≤ 2n .
i!
Then f i (0) = Ci which is an integer.
n!
Therefore f i (0) is an integer for all integers i ≥ 0. Also f ( x)= f (1 − x) .
x n (1 − x)n
where f ( x) = .
n!
Then by the lemma, F(0) and F(1) both are integers.
d hx
Now ( x)) h e hx F ( x) + e hx F ′( x)
(e F=
dx
= e hx [ F ′( x) + h F ( x) ] (1)
∴ F ( x) h 2 n +1 f ( x) + f ( 2 n +1) ( x) .
F ′( x ) + h =
∴ F ( x) h 2 n + 1 f ( x) .
F ′( x ) + h =
Analytical Number Theory 25
d hx
∴ From (1) ( x)) h 2 n +1 f ( x) e hx .
(e F=
dx
Then by fundamental theorem of integral calculus,
1
1
h 2 n +1 ∫ f ( x) e = e hx F ( x)
hx
dx 0
0
= e h F (1) − F (0)
a h a
= F (1) − F (0) e = b
b
a F (1) − b F (0)
=
b
1
or b h 2 n +1 ∫ e hx f ( x) dx
= aF (1) − b F (0). (2)
0
Since F(1) and F(0) are integers so R.H.S. of (2) is an integer, then L.H.S. must also be an integer. Now,
1
we know that e hx keeps the same sign in the open interval (0, 1) and 0 < f ( x) < for all x in (0, 1).
n!
Therefore, by first mean value theorem
1 1
bh 2 n +1
0 < b h 2 n +1 ∫ e hx f ( x) dx ≤ ∫e
hx
dx
0
n! 0
b h2n 1
= e hx
n! 0
b h2n h
= (e − 1)
n!
bh 2 n h
< e
n!
ah 2 n h a
= e = b
n!
Thus, we have a positive integer lying between 0 and 1 which is a contradiction. Thus, eh cannot be
rational. Hence the result.
26 The Group of Units and Quadratic Residues
x n (1 − x)n
Define f ( x) = .
n!
∴ f (0) = f (1) = 0 and as proved in the lemma f m (0) and f m (1) are all integers for m ≥ 0.
1
Further clearly, 0 < f ( x) < ∀ x ∈(0, 1) (Prove the Lemma 1.5.8).
n!
Define a function
Then since f m (0) and f m (1) are integers for every m ≥ 0 so G(0) and G(1) are integers. Now, consider
d
(G ′( x) sin π x − π G( x) cos π x)= G ′′( x) sin π x + π G ′( x) cos π x − π G ′( x) cos π x + π 2G( x) sin π x
dx
G ′′( x) +
= π 2G( x) b n π 2 n + 2 f ( x) + (−1)n f 2 n + 2 ( x) .
∴ π 2 G ( x) b n π 2 n + 2 f ( x) .
G ′′( x) + = (2)
a
= a nπ 2 f ( x) sin π x. π 2
=
b
1
= π [ G(0) + G(1) ] .
1
+ G(1) π a
G(0)= ∫ f ( x) sin π x dx . (3)
n
∴
0
1
Now, sin πx is positive in (0, 1) and 0 < f ( x) < in (0, 1) so by first mean value theorem of integral
n!
calculus, we have
1 1
a nπ
0< a π ∫ f ( x) sin π x dx < ∫ sin π x dx
n
0
n! 0
an
=
n!
[ − cos π x ]0
1
2a n
=
n!
< 1 for n large enough.
∞
an
Since ∑ n!
n =1
is convergent series and so its nth term must tend to zero. But L.H.S. of (3) is an integer
1.6.3 Theorem. Let α be any given real number. Then for every integer t > 0, there exist integers x and y
1
such that | α x − y | < and 0 < x ≤ t .
t
Proof. Take the interval [0, 1). Divide this interval into t subintervals, that is,
1 1 2 2 3 t −1
0 , t , t , t , t , t ,.........., t , 1 .
All these subintervals are mutually disjoint. Consider the real numbers
These are t + 1 real numbers and we have only t subintervals. So, atleast one subinterval consists of
atleast two of t + 1 real numbers given in (1). So, there exist two distinct integers i and j such that
28 The Group of Units and Quadratic Residues
1
{iα } − { jα } < , 0 ≤ i < j ≤ t.
t
Now, by definition { iα } =
iα − y1 , for some y1 ∈ Z
and { jα } =
jα − y2 , for some y2 ∈ Z .
1
∴ > {iα } − { jα } = (iα − y1 ) − ( jα − y2 )
t
= (i − j )α − ( y1 − y2 =
) | ( j − i )α − ( y2 − y1 ) |
= ( j − i) α + ( y1 − y2 ) .
Set x =
j − i , y =−
y2 y1.
1
and hence |αx − y | < .
t
1.6.4 Theorem. Given real α and integer t > 0, we can find integers x and y such that
1
|αx − y | < ,=
( x , y) 1 and 0 < x ≤ t .
t
Proof. By the above theorem, we can find integers x1 and y1 such that
1
| α x1 − y1 | < and 0 < x1 ≤ t.
t
If g.c.d. ( x1 , y1 ) =1 , then we are through. Otherwise let ( x1 , y1 ) = d .
∴ =x1 dx
= and y1 d . y where g.c.d.(x, y) = 1.
1 1 1
Now, | α x −=
y| α x1 − y1 < <
d td t
Hence the theorem.
1.6.5 Corollary. Given ε > 0, however small and any real number α, there exist two integers x and y
y
such that x > 0 and α − < ε.
x
1
Proof. Since ε > 0 is given, we can choose an integer t such that t > . Now, by last theorem there exist
ε
1
x and y such that x > 0 and | α x − y | < <ε
t
Analytical Number Theory 29
y ε
⇒ α− < ≤ ε , x ≥ 1.
x x
y
⇒ α− < ε.
x
y 1
1.6.6 Theorem. Given α > 0, there exist integers x and y such that α − < 2 and
x x
g.c.d.(x, y)=1.
Proof. We know from the above theorem, that we can find integers x and y such that g.c.d. (x, y) =
1
1, 0 < x ≤ t, where t > 0 is an integer and | α x − y | < , then
t
y 1 1
α− < < 2 since x ≤ t .
x tx x
1.6.7 Theorem. Let α be any rational number. Then there exists only a finite number of pairs of integers
y 1
x, y such that x > 0,
= ( x , y) 1 and α − < 2.
x x
h
Proof. Since α is rational, so let α = , where k > 0 and g.c.d. (h, k) = 1. Then
k
h 1
0= α− < 2.
k k
Thus, there exists atleast one pair h, k satisfying the given condition.
y 1
Now, let α− < 2 s.t. x > 0 and g .c.d .( x , y) =1 (1)
x x
1 1 1
Then |αx − y | < ⇒ − < αx − y <
x x x
1 1
⇒ < y −αx < −
x x
1 1
⇒ αx − < y < αx + ,
x x
2
where y lies in an interval of length ≤ 2 and so given x, y can have atmost 3 values. Further setting
x
h
α= in (1), we get
k
30 The Group of Units and Quadratic Residues
1 h y hx − ky
2
> − = .
x k x kx
If hx − ky ≠ 0 then | hx − ky | ≥ 1 [since h, k, x, y all are integers].
1 1
Therefore, 2
> ⇒ k > x.
x kx
Also, x > 0 ⇒ 0 < x < k and so x can take atmost k − 1 values and so the pair x, y can take
atmost 3(k − 1) values.
1.6.8 Theorem. Let α be any irrational number. Then there exist infinitely many pairs x, y
y 1
satisfying α − < 2 , x > 0 , gcd ( x , y) =1 . (1)
x x
Proof. We know that there exists atleast one pair x, y satisfying (1). If possible, suppose that there be
only a finite number of pairs x, y satisfying (1) and let these pairs be
( x1 , y1 ) , ( x2 , y2 ) ......., ( xr , yr ).
1
Take t > , then there exist integers x, y such that
ε
1
x , y) 1, | α x − y | <
0 < x ≤ t , g .c.d . (= <ε.
t
y 1 1
Also, α− < < 2 (0 < x ≤ t ) .
x tx x
Therefore, this pair x, y also satisfies (1) but | α x − y | < ε and so this pair (x, y) is not equal to any pair
( xi , yi ) , which is a contradiction. Hence, the number of pairs satisfying the given condition are infinite.
Further the number of above pairs is infinite if α is irrational and finite if α is rational.
Analytical Number Theory 31
1.7 Hurwitz’s theorem
1.7.1 Lemma. If x and y are positive integers then the following two inequalities
1 1 1 1
≥ 2
+ 2
xy 5 x y
1 1 1 1
and ≥ 2
+ 2
x( x + y) 5 x ( x + y)
cannot hold simultaneously.
Proof. If possible, suppose that both inequalities hold. Then, we get
1 1 1 1
≥ 2
+ 2
xy 5 x y
1 1 1 1
and ≥ 2
+ 2
.
x( x + y) 5 x ( x + y)
From these two, we get
5 xy ≥ x 2 + y 2
5 ( x 2 + 2 xy) ≥ 3 x 2 + 2 y 2 + 2 xy
or (3 − 5 ) x 2 + 2 y 2 − 2(−1 + 5 ) xy ≤ 0.
Multiplying by 2, we get (6 − 2 5 ) x 2 + 4 y 2 − 4( 5 − 1) xy ≤ 0
2
⇒ ( 5 − 1) x − 2 y ≤ 0 .
But the square quantity cannot be less than zero.
2
⇒ ( 5 − 1) x − 2 y =0
⇒ ( 5 − 1) x − 2 y =0
2y
⇒ 5
= +1
x
1.7.2 Hurwitz’s theorem. Given any irrational number ξ, there exists infinitely many pairs (h, k) of
integers such that
h 1
ξ− < . (1)
k 5 k2
h 1 nk + h 1
Proof. Since ξ − < ⇔ (ξ + n) − < .
k 5 k2 k 5 k2
So, without loss of generality, we assume that 0 ≤ ξ < 1. Further ξ is irrational so ξ ≠ 0 . Thus, we
assume 0 < ξ < 1. Let n ∈ N. Consider Farey series of order n. Since ξ is irrational, there exist two
a c a c
consecutive Farey fractions and of order n such that <ξ< then either
b d b d
a+c a+c
ξ< or ξ> .
b+d b+d
a c a+c
We shall prove that in either case atleast one fraction out of , and satisfies (1). Suppose none
b d b+d
of these fractions satisfy (1).
To prove the theorem, we distinguish the following two cases:
a+c
Case I. ξ < .
b+d
a c a+c
Since we have supposed that none of the fraction , and satisfies (1). Therefore,
b d b+d
a 1
ξ− ≥ .
b 5 b2
a a a 1
But <ξ ⇒ ξ− =ξ − ≥ . (2)
b b b 5 b2
c c c 1
and ξ< ⇒ ξ− = −ξ ≥ . (4)
d d d 5 d2
Adding (2) and (4), we get
1 1 1 c a bc − ad 1
2 + 2 ≤ −
= = . (5)
5 b d d b bd bd
Adding (2) and (3), we get
Analytical Number Theory 33
1 1 1 a+c a
b 2 + (b + d )2 ≤ b + d − b
5
b(a + c) − a(b + d )
=
b(b + d )
bc − ad 1
= = . (6)
b(b + d ) b(b + d )
But we have proved in the Lemma 1.7.1 that not both of the inequalities
1 1 1 1
≥ 2
+ 2
xy 5 x y
1 1 1 1
and ≥ 2
+ 2
x( x + y) 5 x ( x + y)
can hold simultaneously.
a c a+c
So, (5) and (6) violate the Lemma 1.7.1 and so in this case at least one of , , must satisfy (1).
b d b+d
a+c
Case II. < ξ.
b+d
a c a+c
Since < ξ < so (2), (4) and (5) also hold. However, < ξ so
b d b+d
a+c a+c 1
ξ− ξ−
= ≥ . (7)
b+d b+d 5 (b + d )2
Adding (4) and (7), we get
1 1 1 c a+c
d 2 + (b + d )2 ≤ d − b + d
5
c(b + d ) − d (a + c)
=
d (b + d )
bc − ad 1
= = . (8)
d (b + d ) d (b + d )
a c a+c
Now, (5) and (8) violate the condition of the Lemma 1.7.1 so atleast one of , , must satisfy
b d b+d
(1) in this case also.
34 The Group of Units and Quadratic Residues
h h a c a+c
Thus, there exists atleast one fraction satisfying (1) and is either equal to or or . Since
k k b d b+d
a c
< ξ < , so
b d
h c a c a + c a + c a
ξ− < − = − + −
k d b d b + d b + d b
c a+c a+c a
≤ − + −
d b+d b+d b
1 1
= + .
d (b + d ) b(b + d )
a c
But b + d ≥ n + 1 b and d are consecutive fraction
h 1 1 2
∴ ξ−
k
≤ + ≤
d (n + 1) b(n + 1) n + 1
. [ b ≥ 1, d ≥ 1 ]
h
Now, to establish that (1) is satisfied by infinitely many rationals , suppose there are only a finite
k
h
number of satisfying (1).
k
h
Let ε min ξ −
= , where minimum ranges over the finitely many rational numbers satisfying (1).
k
Since ξ is irrational, this minimum must be > 0, that is, ε > 0. Choose a rational number n such that
2 h
n + 1 > . For this number n, as shown above there exists a rational number 1 satisfying (1) such that
ε k1
h1 2 h
ξ− ≤ < ε and so 1 must be different from finitely many rational numbers considered above,
k1 n +1 k1
h
which is a contradiction. So, there must exist infinitely many rational satisfying (1). This proves the
k
theorem.
1.7.3 Theorem. Prove that 5 occurring in the statement of Hurwitz’s theorem is best possible in the
sense that if 5 is replaced by any larger real number, say m, then there exists an irrational number ξ
such that
h 1
ξ− < (1)
k mk2
Analytical Number Theory 35
h
does not hold for infinitely many rational numbers .
k
1+ 5 1− 5
Proof. Take ξ = , then ξ > 1 and ξ= = ξ− 5.
2 2
1+ 5
We shall prove that if m is any real number with ξ = and (1) is satisfied by infinitely many
2
h
rational numbers then m ≤ 5 . So, we assume that (1) is satisfied by infinitely many rational numbers
k
h
.
k
Now, ( )
( x − ξ) x − ξ =( x − ξ)( x − ξ + 5 )
1+ 5 1+ 5
= x− x − + 5
2 2
1+ 5 1− 5
= x− x−
2 2
= x 2 − x − 1. (2)
Now, for all integers h , k ; k > 0
h h h2 h 1
−ξ −ξ + 5= 2
− − 1= 2
h 2 − hk − k 2 .
k k k k k
h
Now, since any rational number is not a root of x 2 −=
x − 1 0 so | h 2 − hk − k 2 | ≥ 1 , as
k
h, k are integers .
h h 1
Therefore, −ξ −ξ + 5 ≥ 2 . (3)
k k k
h
Since there exist infinitely many rational numbers satisfying (1), there exists a sequence
k
hi
, i = 1, 2, 3,...... of rationals satisfying (1). Then
ki
hi 1
−ξ <
ki m ki2
1
or | hi − ξ ki | < .
m ki
36 The Group of Units and Quadratic Residues
1 h h
2
≤ i −ξ . i −ξ + 5
ki ki ki
hi h
≤ −ξ . i −ξ + 5
ki ki
1 1 hi
≤ 2 2
+ 5 . satisfy (1)
m ki mki ki
Now, multiplying by m ki2 , we get
1
m≤ + 5 ⇒ m ≤ 5 for all i large enough.
mki2
1.9 Summary
In this unit, we provided the distribution of primes in different forms, Fermat and Mersenne numbers,
and construction of Farey series along with some results concerning Farey series. In the end, irrationality
of e and π , and Hurwitz theorem are discussed.
1.10 Keywords
Prime number, Irrational number, Fermat and Mersenne numbers, Farey series.
1.11 References
1. G. H. Hardy and E.M. Wright, An Introduction to the Theory of Numbers.
2. D.M. Burton, Elementary Number Theory.
3. N.H. McCoy, The Theory of Numbers by McMillan.
4. I. Niven, I and H.S. Zuckermann, An Introduction to the Theory of Numbers.
5. A. Gareth Jones and J Mary Jones, Elementary Numbers Theory, Springer Ed. 1998.
Unit– II | The Group of Units and Quadratic Residues
Structure
2.0 Introduction
2.1 Unit objectives
2.2 The arithmetic in Zn
2.3 The group U n
2.7 Quadratic residues for prime power moduli and arbitrary moduli
2.8 Check your progress
2.9 Summary
2.10 Keywords
2.11 References
2.0 Introduction
In this unit, we will study the arithmetic in Zn, the group of units U n , U pn , U 2n , the concepts of primitive
roots and their existence. In the end, quadratic residues for prime power moduli and arbitrary moduli are
also studied.
2.1 Unit objectives
After going through this unit, one will be able to
• know about the groups Zn, U n , U pn , U 2n .
Let Zn denote the set of all distinct equivalence classes modulo n, that is,
Z n = {[0],[1],...,[n − 1]} .
is a member of Zn.
(ii) Associativity. Let [a ],[b],[c] ∈ Z n . Then,
For if n = 10. Then, [2] ∈ Z10 . Let [x] be the inverse of [2]. Then,
2.2.4 Definition (Unit). A multiplicative inverse for a class [a ] ∈ Z n is a class [b] ∈ Z n such that [a].[b]
= [1]. A class [a ] ∈ Z n is a unit if it has a multiplicative inverse in Zn.
[a ].[b] =
[1] ⇒ ab ≡ 1(mod n) ⇒ ab = 1 + nt for some integer t.
= 1
⇒ ab − nt ⇒ g .c.d .(a, n=
) 1.
2.2.6 Example. The units in Z8 are [1], [3], [5] and [7] as 1, 3, 5 and 7 are only numbers less than 8
which are co-prime with 8. Also,
[1].[1] = [3].[3] = [5].[5] = [7].[7] = [1].
Similarly, in Z9 the units are [1], [2], [4], [5], [7] and [8]. Also,
[2].[5] = [1].
Thus, [2] and [5] are inverse of each other.
2.2.7 Remark. We know that a linear congruence ax ≡ b mod(n) has a unique solution mod(n) if gcd(a,
n) = 1. Now if n is a prime p, then gcd(a, n) = gcd(a, p) is either 1 or p; in the first case, we have a
unique solution mod(p). While in the second case (where p|a), either every x is a solution (when p|b) or
no x is a solution (when p does not divide b).
One can view this elementary result as saying that if the polynomial ax – b has degree d = 1 over Zp (that
is, if a ≠ 0 mod( p ) ), then it has atmost one root in Zp. Now, in algebra we learn that a non-trivial
polynomial of degree d, with real or complex coefficients, has atmost d distinct roots in R or C; it is
reasonable to ask whether this is also true for the number system Zp, since we have just seen that it is
true when d = 1. Our first main theorem, due to Lagrange, states that this is indeed the case.
2.2.8 Theorem. Let p be prime, and let f ( x=
) ad x d + ... + a1 x + a0 be a polynomial with integer
coefficients, where ai ≠ 0 mod( p ) for some i. Then, the congruence f ( x) ≡ 0 mod( p ) is satisfied by
atmost d congruence classes [ x] ∈ Z p .
40 Riemann Zeta Function and Diophantine Equations
Proof. We use induction on d. If d = 0 then f ( x) = a0 with p not dividing a0 , so, there are no solutions of
f(x) = 0, as required. For the inductive step, we now assume that d ≥ 1 , and that all polynomials
( x) bd −1 x d −1 + ... + b0 with some bi ≡ 0 have at most d – 1 roots [ x] ∈ Z p .
g=
If the congruence f ( x) ≡ 0 has no solutions, there is nothing left to prove, so, suppose that [a] is a
solution; thus f (a ) ≡ 0 , so, p divides f(a). Now,
d d d
f ( x) − f (a )= ∑ ai xi − ∑ ai ai =
=i 0=i 0 =i 0
∑ a (x
i
i
− a i ) . For each i = 1,…,d , we can put
2.2.9 Remark.
1. Note that this theorem allows the possibility that ad = 0 , so that f(x) has degree less than d: if so,
then by deleting ad x d , we see that there are strictly fewer than d classes [x] satisfying f ( x) ≡ 0 .
The same argument applies if we merely have ad ≡ 0 mod( p ) .
2. Even if ad ≠ 0 , f(x) may still have fewer than d roots in Z p : for instance f ( x=
) x 2 + 1 has only
one root in Z 2 , namely the class [1], and it has no roots in Z 3 .
3. The condition that ai is not congruent to 0 for some i ensures that f(x) yields a non-trivial
polynomial when we reduce it mod(p). If ai ≡ 0 for all i then all p classes [ x] ∈ Z p
satisfy f ( x) ≡ 0 , so the result will fail if d < p .
4. In the theorem, it is essential to assume that the modulus is prime: for example, the
) x 2 − 1 , of degree d = 2, has four roots in Z8 , namely the classes [1], [3], [5]
polynomial f ( x=
and [7].
A useful equivalent version of Lagrange’s Theorem is the contrapositive:
Let f ( x=
) ad x d + ... + a1 x + a0 be a polynomial with integer coefficients, and let p be prime. If f(x) has
more than d roots in Z p , then p divides each of its coefficients ai .
Analytical Number Theory 41
Lagrange’s Theorem tells us nothing new about polynomial f(x) of degree d ≥ p : there are only p
classes in Z p , so it is trivial that atmost d classes satisfy f ( x) ≡ 0 . The following result, useful in
studying polynomials of high degree, is known as Fermat’s Little Theorem though it was also known to
Leibnitz, and the first published proof was given by Euler.
Proof. The integers 1, 2,…, p – 1 form a complete set of non-zero residues (mod p).
If a ≡ 0 mod( p ) then xa ≡ ya implies x ≡ y , so that the integers a, 2a, . . ., (p - 1)a lie in distinct classes
(mod p). None of these integers is divisible by p, so they also form a complete set of non-zero residues.
It follows that a, 2a, . . ., (p - 1)a are congruent to 1, 2, . . ., (p - 1) in some order. (For instance, if p = 5
and a = 3 then multiplying the residues 1, 2, 3 , 4 by 3 we get 3, 6, 9, 12, which are respectively
congruent to 3, 1, 4, 5.) The product of these two sets of integers must therefore lie in the same class,
that is, 1 × 2 × ... × ( p − 1) ≡ a × 2a × ... × ( p − 1)a (mod p ) , or equivalently, ( p − 1)! ≡ ( p − 1)!a p −1 (mod p ) .
Since ( p − 1)! is coprime to p, divide through out by ( p − 1)! and deduce that a p −1 ≡ 1(mod p ) .
This theorem states that all the classes in Z p except [0] are roots of the polynomial x p −1 − 1 . For a
polynomial satisfied by all the classes in Z p , we simply multiply by x, to get x p − x :
Proof. If a is not congruent to 0 then by above theorem a p −1 ≡ 1(mod p ) , so multiplying each side by a
gives the result. If a ≡ 0(mod p ) then a p ≡ 0(mod p ) also, so the result is again true.
2.2.12 Remark. This corollary shows that if f(x) is any polynomial of degree d ≥ p , then by repeatedly
replacing any occurrence of x p with x, we can find a polynomial g(x) of degree less than p with the
property that f ( x) ≡ g ( x) for all integers x. In other words, when considering polynomials (mod p), it is
sufficient to restrict attention to those of degree d < p . Similarly, the coefficients can also be simplified
by reducing them (mod p).
These two results are very useful in dealing with large powers of integers.
2.2.13 Example. Let us find the least non-negative residue of 268 (mod19) . Since 19 is prime and 2 is
not divisible by 19, we have p = 19 and a = 2, so that 218 ≡ 1(mod19) . Now 68 = 18 × 3 + 14 ,
so 268= (218 )3 × 214 ≡ 13 × 214 ≡ 214 (mod19) . Since 24= 16 ≡ −3(mod19) , we can write 14 = 4 × 3 + 2 and
214 (24 )3 × 22 ≡ ( −3)3 × 22 ≡ − 27 × 4 ≡ −8 × 4 ≡ −32 ≡ 6(mod19) so that 268 ≡ 6 (mod19) .
deduce that =
2.2.14 Example. Show that a 25 − a is divisible by 30 for every integer a. By factorizing 30, we see that
it is sufficient to prove that a 25 − a is divisible by each of the prime’s p = 2, 3, and 5. Let us deal with p
= 5 first, applying above Corollary twice, we have a 25= (a 5 )5 ≡ a 5 ≡ a (mod 5),
Here, we can simply try all five classes [5] ∈ Z 5 , or else note that g ( x) ≡ ( x − 1) 2 ; either way, we find that
[x] = [1] is the only root of g ( x) ≡ 0 , so this class is the only root of f ( x) ≡ 0 .
As another application of Fermat’s Little Theorem, we prove a result known as Wilson’s Theorem,
though it was first proved by Lagrange in 1770:
2.2.16 Corollary. An integer n is prime if and only if (n − 1)! ≡ −1(mod n) .
For the converse, suppose that (n − 1)! ≡ −1(mod n) . We then have (n − 1)! ≡ −1(mod m) for any factor m
of n. If m < n then m appears as a factor of (n − 1)! . So, (n − 1)! ≡ 0(mod m) and hence −1 ≡ 0(mod m) .
This implies that m = 1, so, we conclude that n has no proper factors and is therefore prime.
2.2.17 Theorem. Let p be an odd prime. Then, the quadratic congruence x 2 + 1 ≡ 0(mod p ) has a
solution if and only if p ≡ 1(mod 4) .
Proof. Suppose that p is an odd prime, and let k = (p-1)/2. In the product
( p − 1)! =1 × 2 × ... × k × (k + 1) × ... × ( p − 2) × ( p − 1) ,
2.3 Units in Zn
2.3.1 Lemma. [a] is a unit in Z n if and only if gcd(a, n) = 1.
Proof. If [a] is a unit then ab = 1+qn for some integers b and q; any common factor of a and n would
therefore divide1, so gcd(a, n) = 1. Conversely, if gcd(a, n) = 1 then 1 = au +nv for some u and v, so, [u]
is a multiplicative inverse of [a].
2.3.2 Definition (The set Un of units in Zn). We define Un the collection of units in Zn. Thus,
U n = {[ x] : [ x] is a unit in Z n } .
Un
= {[ x] : 1 ≤ x ≤ n − 1 such that , n) 1} .
g.c.d. ( x=
2.3.3 Theorem. For each integer n ≥ 1 , the set Un forms a group under multiplication modulo n, with
identity element [1]. Also, it is abelian.
Proof. Consider=
Un {[ x] : 1 ≤ x ≤ n − 1 such that , n) 1} .
g.c.d. ( x=
[a ].[1]
= [a.1]
= [a=] [1.a=] [1].[a ]
Thus, Un is commutative.
Hence, Un is an abelian group and is called the group of units.
Note. For the simplicity of topic, we will write a ∈U n in place of [a ] ∈U n . Thus, now onwards ‘a’ will
represent the class of ‘a’, that is, [a].
×8 1 3 5 7
1 1 3 5 7
3 3 1 7 5
5 5 7 1 3
7 7 5 3 1
(iii). U10 = {1, 3, 7, 9} under multiplication modulo 10. In this group 1 and 9 are self inverses and 8 and
7 are inverse of each other.
Now, 22 4,=
= 23 8,=
24 7,=25 5,=
26 1 , so o(2) = 6 .
Again, 42 7,=
= 43 1 so =
o(4) 3 .
7 2 4,=
Again,= 73 1 so =
o(7) 3 .
Finally, 82 = 1 so o(8) = 2 .
2.3.8 Lemma. If l and m are co-prime positive integers, then 2l – 1 and 2m – 1 are co-prime.
Proof. Let n be the highest common factor of 2l – 1 and 2m – 1, i.e., g.c.d.(2l – 1, 2m – 1) = n.
Clearly, n is odd, since 2l – 1 and 2m – 1 both are odd.
So, 2 is a unit modulo n as g.c.d.(2, n) = 1. Let k be the order of the element of 2 in the group Un, so,
2k ≡ 1(mod n)
Since n divides 2l – 1, we have
2l ≡ 1(mod n) .
So, k | l .
Similarly, k | m .
Thus, k | (l , m) ⇒ k |1 ⇒ 1.
k=
2.3.16 Example. In U 5 , the element 2 has order 4: its powers are 21 ≡ 2, 22 ≡ 4, 23 ≡ 3 and
24 ≡ 1mod(5) , so k = 4 is the least positive exponent such that 2k = 1 (the identity element) in U 5 .
Similarly, the element 1 has order 1, while the elements 3 and 4 have orders 4 and 2 respectively.
2.3.17 Example. In U 8 , the elements 1, 3, 5, 7 have orders 1, 2, 2, 2 respectively.
For example, 3 and 7 are primitive roots modulo 10 but there are no primitive roots modulo 12 since
U12 is not cyclic.
2.4.2 Remark. To find the primitive roots modulo n, we have to calculate orders of elements of Un and
if an element has order φ (n) then it is a primitive root modulo n. Then, tedious work of finding the
powers of each element of Un is made easy upto some extent by the following lemma:
φ (n)
Proof. Suppose first that a is a primitive root of Un then = U n | φ (n) , where |a| and | U n | denotes
| a | |=
the orders of a and Un respectively.
⇒ a i ≠ 1 for all i s.t. 1 ≤ i < φ (n)
φ (n)
⇒ a q
≠ 1 for each prime q dividing φ (n) .
Analytical Number Theory 47
φ (n)
Conversely, suppose a q
≠ 1 for each prime q dividing φ (n) . We shall prove that a is a primitive root
mod n.
Let, if possible, ‘a’ is not a primitive root (mod n). Then, order of ‘a’ is not equal to φ (n) and by
φ ( n)
Lagrange’s theorem of groups its order, say, k must divide order φ (n) of Un, so 1.
> 1 and a k =
k
φ ( n) φ ( n) φ ( n)
If q is any prime factor of , then k divides and let = kt so that
k q q
φ (n)
a q
= a kt= (a k )=
t
1=
t
1
which is a contradiction to our hypothesis as q is also a prime factor of φ (n) . Hence ‘a’ is a primitive
root modulo n.
2.4.4 Example. By using above lemma, prove that
(i) 2 is a primitive root modulo 11 but 3 is not a primitive root modulo 11.
(ii) 3 is a primitive root modulo 17 but 2 is not a primitive root modulo 17.
(iii) 2 is a primitive root modulo 9 but 4 is not a primitive root modulo 9.
Solution. (i) Here n = 11 and φ=
(n) φ=
(11) 10 .
φ ( n) 10 10
The primes dividing φ (n) = 10 are 2 and 5 and so possible values of are and i.e., 5 and 2.
q 2 5
Now, =25 10
= and 22 4 and so 2 is a primitive root.
φ ( n) 16
The only prime dividing φ (n) = 16 is 2 and so only possible value of is = 8.
q 2
But 28 =(24 ) 2 =−
( 1) 2 =1 so 2 is not a primitive root.
2.4.5 Remark. We know that any cyclic group of order n has φ (n) generators. Using this result if Un is
a cyclic group then it must have φ ( φ (n) ) generators as order of Un is φ (n) .
48 Riemann Zeta Function and Diophantine Equations
Here, the main objective is to decide for which values of n, the group Un has a primitive root
(generator) i.e., for which values of n, the group Un is cyclic. We shall prove that Up is cyclic for every
prime p which will follow by the next theorem:
2.4.6 Theorem. If p is a prime then the group Up has φ(d) elements of order d for each d dividing p −1.
To prove the theorem, our aim is to prove that w(d ) = φ (d ) for all d.
By Lagrange’s theorem of groups, the order of each element of Up divides p −1, so the sets Ωd form a
partition of Up and therefore
∑
d / p −1
w(d )= p − 1. ......(1)
We know that ∑ φ (d ) = n
d /n
∑ φ (d )=
d / p −1
p − 1. ......(2)
∑ (φ (d ) − w(d )) =
d / p −1
0. ......(3)
We claim that w(d ) ≤ φ (d ) . If Ωd is empty then this inequality is obvious, so we assume that Ωd contain
atleast one element, say, a. By the definition of Ωd, the order of a is d and so the elements
a1 , a 2 , a 3 ,......., a d = 1 are all distinct and (a i ) d = 1 for all i.
But d is the order of b and so lower power of b than bd cannot be equal to 1 and hence j = 1. Thus, every
element b of Ωd is of the form
Analytical Number Theory 49
The number of such integers i is φ (d ) , so the number w(d ) is atmost φ (d ) i.e., w(d ) ≤ φ (d ) . So, each
summand in (3) is non-negative and their sum is zero so w(d ) = φ (d ) . Thus, Up has φ (d ) elements of
order d.
2.4.7 Theorem. If p is an odd prime then Upe is cyclic for e ≥ 1.
Proof. Theorem 2.4.6 deals with the case e = 1, so we may assume that e ≥ 2. We use the following
strategy to find a primitive root (mod pe).
(a) First, we pick a primitive root g (mod p) (possible by above theorem).
(b) Next, we show that either g or g + p is a primitive root (mod p2).
(c) Finally, we show that if h is any primitive root (mod p2), then h is a primitive root mod pe for all
e ≥ 2.
Above theorem covers step (a), giving us a primitive root g (mod p). Thus, g p −1 ≡ 1(mod p ) , but
g i ≡/ 1(mod p ) for 1 ≤ i < p − 1 . We now proceed to step (b).
Since g.c.d. (g, p) = 1, we have g.c.d. (g, p2) =1, so we can consider g as an element of U p2 . If d denotes
h p −1 = ( g + p ) p −1 = g p −1 + ( p − 1) g p − 2 p + ..... ≡ 1 − pg p − 2 (mod p 2 )
where the dots represent terms divisible by p2. Since g is co-prime to p, we have pg p − 2 ≡/ 0 (mod p 2 )
and hence h p −1 ≡/ 1 (mod p 2 ) . Thus, h does not have order p − 1 in U p2 , so, it must have order p ( p − 1)
and is therefore a primitive root. This completes step (b).
Now, we consider step (c). Let h be any primitive root (mod p2). we will show, by induction on e, that h
is a primitive root (mod pe) for all e ≥ 2. Suppose, then, that h is a primitive root (mod pe) for some e ≥
2, and let d be the order of h(mod p e +1 ) . An argument similar to that at the beginning of step (b) show
that d divides φ ( p=
e +1
) p e ( p − 1) and is divisible by φ ( p e ) p e −1 ( p − 1),
= so,
d =p e ( p − 1) or d =p e −1 ( p − 1) .
In the first case, h is a primitive root (mod p e +1 ) , as required, so it is sufficient to eliminate the second
e −1
( p −1)
case by showing that h p ≡/ 1(mod p e +1 ) .
50 Riemann Zeta Function and Diophantine Equations
p p
1 + kp e −1 + (kp e −1 ) 2 + ......
=
1 2
1
=1 + kp e + k 2 p 2 e −1 ( p − 1) + .......
2
The dots here represent terms divisible by ( p e −1 )3 and hence by p e +1 , since 3(e − 1) ≥ e + 1 for e ≥ 2, so
e −1
( p −1) 1
hp ≡ 1 + kp e + k 2 p 2 e −1 ( p − 1) (mod p e +1 )
2
Now, p is odd, so the third term k 2 p 2 e −1 ( p − 1) / 2 is also divisible by p e +1 , since 2e − 1 ≥ e + 1 for e ≥ 2.
e −1
( p −1)
Thus, hp ≡ 1 + kp e (mod p e +1 )
e −1
( p −1)
Since p does not divide k, we therefore have h p ≡/ 1(mod p e +1 ) , so step (c) is complete.
2.4.8 Remark. Notice where we need p to be odd: If p = 2 then the third term
k 2 22 e − 2 is not divisible by 2e +1 when e = 2, so the first step of the induction
k 2 p 2 e −1 ( p − 1) / 2 =
argument fails.)
2.4.9 Lemma. If n = rs where r and s are coprime and are both greater than 2, then U n is not cyclic.
Proof. Since gcd(r, s) = 1 we have φ (n) = φ (r )φ ( s ) . Since r, s > 2, both φ (r ) and φ ( s ) are even. So
φ (n) is divisible by 4. It follows that the integer e = φ (n) / 2 is divisible by both φ (r ) and φ ( s ) . If a is a
unit (mod n), then a is a unit (mod r) and also a unit (mod s), so aφ ( r ) ≡ 1(mod r ) and aφ ( s ) ≡ 1(mod s ) by
Euler’s Theorem. Since φ (r ) and φ ( s ) divide e, we therefore have a e ≡ 1(mod r ) , that is, a e ≡ 1(mod s ) .
Since r and s are coprime, this implies that a e ≡ 1(mod rs ) , that is, a e ≡ 1(mod n) . Thus, every element of
U n has order dividing e, and since e < φ (n) , this means that there is no primitive root (mod n).
2.4.10 Theorem. The group Un is cyclic if and only if n = 1, 2, 4, pe or 2pe where p is an odd prime.
Proof . The cases n = 1, 2 and 4 are trivial, and we have dealt with the odd prime-powers, so we may
assume that n = 2pe, where p is an odd prime. Now φ (n) = φ (2) φ (pe) = φ (pe). Therefore, there is a
primitive root g(mod pe). Then g + pe is also a primitive root (mod pe), and one of g and g + pe is odd, so
there is an odd primitive root h(mod pe). We will show that h is a primitive root (mod 2 p e ) . By its
construction, h is coprime to both 2 and pe, so h is a unit (mod 2pe). If hi ≡ 1 (mod 2pe), then certainly hi
Analytical Number Theory 51
≡ 1(mod pe); since h is a primitive root (mod pe), this implies φ (pe) divides i. Since φ (pe) = φ (2pe), this
shows that φ (2pe) divides i, so h has order φ (2pe) in U 2 pe and is therefore a primitive root.
for all odd a. We prove this by induction on e. For the lowest value e = 3, (1) says that a 2 ≡ 1(mod 8) for
2
all odd a, and this is true since if =
a 2b + 1 then a= 4b(b + 1) + 1 ≡ 1(mod 8) . If we assume (1) for some
exponent e ≥ 3, then for each odd a we have
e−2
a 2 = 1 + 2e k
for some integer k. Squaring, we get
( e + 1) − 2
a2 (1 + 2e k ) 2 =
= 1 + 2e + 1 k + 2 2 e k 2 =
1 + 2e +1 (k + 2e −1 k 2 ) ≡ 1mod(2e +1 )
which is the required form of (1) for exponent e +1. Thus, (1) is true for all integers e ≥ 3, and the proof
is complete.
2.5.2 Remark. In order to state the next result, we need some more notation. If we use the set
{ ± 1, ± 2....... ± ( p − 1) / 2 } as a reduced set of residues (mod p), then we can partition Up into two
subsets.
P ={1, 2,....., ( p − 1) / 2} ⊂ U p and N ={ − 1, − 2,....., − ( p − 1) / 2 } ⊂ U p
aP
= { ax : x ∈ P=
} { a, 2a,...., ( p − 1) a 2} ⊂ U p .
Thus, N = ( −1) P , for example.
52 Riemann Zeta Function and Diophantine Equations
2.5.3 Example. Let p = 5. We have seen that g = 2 is a primitive root (mod 5), since it has order
φ (5) = 4 as an element of U 5 . If we regard g = 2 is an element of U p2 = U 25 , then by the above
4
argument its order d in U 25 must be either p(p-1) = 20 or p – 1 = 4. Now 2= 16 ≠ 1(mod 25) , so
d ≠ 4 and hence d = 20. Thus g =2 is a primitive root (mod 25). (One can check this directly by
computing the powers of 2, 22 ,..., 220 (mod 25) , using = 210 1024 ≡ −1(mod 25) to simplify the
calculations.) Suppose instead that we had chosen g = 7; this is also a primitive root (mod 5), since
2
7 ≡ 2(mod 5) , but it is not a primitive root (mod 25): we have 7= 49 ≡ −1(mod 25) , so
7 4 ≡ 1(mod 25) and hence 7 has order 4 in U 25 . Step (b) guarantees that in this case, g + p = 12 must be a
primitive root.
n
2.5.4 Lemma. 2n+ 2 | (52 − 1) for all n ≥ 0 .
Proof. We use induction on n. The result is trivial for n = 0. Suppose it is true for some n ≥ 0 . Now,
n+1 n n n n
52 − =
1 (52 ) 2 −=
1 (52 − 1)(52 + 1) with 2n+ 2 | (52 − 1) by the induction hypothesis, and with
n n n+1
2 | (52 + 1) since 52 ≡ 1(mod 4) . Combining the powers of 2 we get 2n + 3 | (52 − 1) as required.
Proof. Let m be the order of the element 5 in U 2 . By Euler’s Theorem, m divides φ(2e ) =
e 2e −1 , so
m = 2k for some k ≤ e − 1 . Above theorem implies that U 2e has no elements of order 2e −1 , so k ≤ n − 2.
Putting n = e – 3 in above theorem we see that 2e −1 | (52 − 1) , so 52 ≠ 1(mod 2e ) and hence, k > e – 3.
e−3 e−3
Since 5 ≡ 1(mod 4) , these are all represented by integers congruent to 1 (mod4). This accounts for exactly
half of the 2e −1 elements 1, 3, 5. . . 2e − 1 of U 2 , and the other half, represented by integers congruent to
e
-1 (mod 4), must be the elements of the form −5i . This shows that every element has the form ±5i for
some i 0,1,..., 2e − 2 − 1 , as required.
=
2.6.1 Definition. An element a ∈ Un is a quadratic residue (mod n) if s2 = a for some s ∈ Un ; the set of
such quadratic residues is denoted by Qn. For small n, one can determine Qn simply by squaring all the
elements s ∈ Un.
2.6.2 Example. Q7 = {1, 2, 4} ⊂ U 7 , while Q8 = {1} ⊂ U8 .
2.6.3 Theorem. Let n = n1 ...n k where the integers n i are mutually coprime, and let f(x) be a polynomial
with integer coefficients. Suppose that for each i = 1,…,k there are N i congruence classes x ∈ Zn , such
that f (x) ≡ 0(mod n i ) . Then there are N = N1 ...N k classes x ∈ Zn such that f (x) ≡ 0(mod n) .
Proof. Since the moduli n i are mutually coprime, we have f (x) ≡ 0(mod n) if and only if f (x) ≡ 0(mod n i )
for all i. Thus, each class of solutions x ∈ Zn of f (x) ≡ 0(mod n) determines a class of solutions
Analytical Number Theory 53
x= x i ∈ Zni of f (x i ) ≡ 0(mod n i ) for each i. Conversely, if for each i, we have a class of solutions
x i ∈ Zni of f (x i ) ≡ 0(mod n i ) , then by Chinese Remainder Theorem, there is a unique class x ∈ Zn
satisfying x = x i (mod n i ) for all i, and this class satisfies f (x) ≡ 0(mod n) . Thus, there is a one-to-one
correspondence between classes x ∈ Zn satisfying f (x) ≡ 0(mod n) , and k-tuples of classes
x i ∈ Zni satisfying f (x i ) ≡ 0(mod n i ) for all i. For each I, there are N i choices for the class x i ∈ Zni , so there
are N1 ...N k such k-tuples and hence this is the number of classes x ∈ Zn satisfying f (x) ≡ 0(mod n) .
that for each odd pi there are N i = 2 classes in Zn of solutions of x 2 ≡ 1(mod n i ) whereas if pi = 2 we may
i
2k +1 if n ≡ 0(mod 8)
=N 2k −1 if n ≡ 2(mod 4)
2k otherwise
where k is the number of distinct dividing n. For instance, if n = 60 = 22.3.5 then k = 3 and there are 2k =
8 classes of solutions, namely x ≡ ±1, ±11, ±19, ±29(mod 60) .
2.6.5 Theorem. Let K denote the number of distinct primes dividing n. If a ∈ Qn, then the number N of
elements t ∈ Un such that t2 = a is given by
2k +1 if n ≡ 0(mod 8)
N = 2k −1 if n ≡ 2(mod 4)
2k if otherwise.
Proof. Let a ∈ Qn then s2 = a for some s ∈ Un . Any element t ∈ Un has the form t = sx for some unique
x ∈ Un , and we have t2 = a if and only if x2 = 1 in Un. Thus, N is the number of solutions of x2 ≡ 1 in
Un, the above example gives the required formula for N.
2.6.6 Theorem. Q n is a subgroup of U n .
54 Riemann Zeta Function and Diophantine Equations
Proof. We need to show that Q n contains the identity element of U n , and is closed under taking products
and inverses. Firstly, 1∈ Q n since 1 = 12 with 1∈ U n . If a, b ∈ Q n then a = s 2 and b = t 2 for some s, t ∈ U n , so
ab = (st) 2 with st ∈ U n , giving ab ∈ Q n . Finally, if a ∈ Q n then a = s 2 for some s ∈ U n ; since a and s are
units (mod n) they have inverses a −1 and s −1 in U n , and a −1 = (s −1 ) 2 so that a −1 ∈ Q n .
2.6.7 Theorem. Let n > 2, Suppose that there is a primitive root g(mod n); then Qn is cyclic group of
order φ(n) / 2 , generated by g2, consisting of the even powers of g.
Proof. Since n > 2, φ (n) is even. The elements a ∈ Un are the powers of gi for
a ∈ Qp.
Proof. We know that there is a primitive root g(mod pe), so with n = pe , we see that
Q pe consists of even power of g. Now, g, regarded as element of Up, is also a primitive root (mod p),
and n = p we know that Qp also consists of the even powers of g. Thus, a ∈ Qp if and only if a ∈ Q pe .
This completes the proof.
Note. For odd primes p, we can find square roots in U p for e ≥ 2 by applying the iterative method to the
e
choose k to satisfy the linear congruence q + 2rk ≡ 0(mod p) , giving s 2 ≡ a(mod pi +1 ) as required. An
element a ∈ Q p has just two square roots in U p for odd p, so these must be ±s . It follows that if we
i +1 i +1
have a square root for a in U p , then we can iterate this process to find its square roots in U p for all e. e
2.7.2 Example. Let us take a = 6 and pe = 52 . In U 5 we have a = 1 = 12, so, we can take r = 1 as a square
root (mod 5). Then r 2 =1 =6 + 5.(−1), so q = -1 and we need to solve the linear congruence
−1 + 2k ≡ 0(mod 5) . This has solution k ≡ 3(mod 5) , so we take s =+
r pi k =+ 16 , and the square
1 5.3 =
roots of 6 in Z5 are given by ±16 , or equivalently ±9(mod 52 ) . If we want the square roots of 6 in Z5 we
2 3
Analytical Number Theory 55
repeat the process: we can take r = 9 as a square root (mod 52 ) , with r 2= 81= 6 + 52.3 , so q = 3; solving
3 + 18k ≡ 0(mod 5) we have k ≡ −1 , so s =9 + 52 .(−1) =−16 , giving square roots ±16(mod 53 ) .
Proof. Parts (a) and (b) are obvious: squaring the elements U2 = {1} ⊂ Z2 and of U4 = {1, 3} ⊂ Z4 , we
see that Q2 = {1} and Q4 = {1}. For part (c) we use the theorem which states that the elements of U 2e all
have the form ± 5i for some i; squaring, we see that the quadratic residues are the even powers of 5.
Since 52 ≡ 1(mod 8), these are represented by integers a ≡ 1(mod 8). Now both the even powers of 5 and
the elements a ≡ 1(mod 8) account for exactly one quarter of the classes in U 2e ; since the first set is
contained in the second, these two sets are equal.
2.7.4 Example.
= Q8 {1},
= Q16 {1, 9},
= Q32 {1, 9,17, 25} , and so on.
Note. One can find square roots in Q 2 by adapting the iterative algorithm given earlier for odd prime-
e
square roots of a in U 2 , and these have the form t = sx, where x = ±1 or 2i ± 1 is a square root of 1.
i +1
Since a ≡ 1(mod 8), we can start with a square root r = 1 for a in U 2 , and then by iterating this process
3
2.7.5 Example. Let us find the square roots of a ≡ 17(mod 25 ) ; there exist since 17 ≡ 1(mod 8) . First, we
find a square root (mod 24). Taking r = 1 we have r 2 =12 =17 + 23.(−2) , so q = -2; taking k = 0 makes q +
4
rk = -2 even, so s =
r + 22 k =1 is a square root of 17(mod 2 ). Now we repeat this process, using r = 1 as a
square root (mod 24) to find a square root s(mod 25). We have r 2 ==1 17 + 24 .(−1) , so now q = -1; taking
k = 1 makes q + rk = 0 even, so s = r + 23k = 9 is a square root of 17(mod 25). The remaining square
roots t are found by multiplying s = 9 by -1 and by 24 ± 1 =±15 , so we have ±7, ±9 as the complete set of
square roots of 17(mod 25).
2.7.6 Theorem. Let n = n1 n2 … nk where the integers ni are mutually coprime. Then, a ∈ Qn if and only
if a ∈ Q ni for each i.
Proof. If a ∈ Qn then a ≡ s2(mod n) for some s ∈ Un. Clearly, a ≡ s2(mod ni) for each i, with s coprime
to n i , so a ∈ Q n . Conversely, if a ∈ Q ni for each i, then there exist elements si ∈ U ni such that a ≡ si2
i
56 Riemann Zeta Function and Diophantine Equations
(mod ni). By Chinese Remainder Theorem there is an element s ∈ Zn such s ≡ si(mod ni) for all
i. Then, s 2 ≡ si2 ≡ a(mod n i ) for all i, and hence s 2 ≡ a(mod n) since the moduli ni are coprime, so, a ∈ Qn.
In this unit, the groups Zn, U n , U pn , U 2n are discussed. The concepts of primitive roots and their
existence, quadratic residues for prime power moduli and arbitrary moduli are given.
2.10 Keywords: Group U n , Primitive roots, Group U pn (p-odd) and U 2n , Cyclic, Quadratic residues
2.11 References
1. G. H. Hardy and E.M. Wright, An Introduction to the Theory of Numbers.
2. D.M. Burton, Elementary Number Theory.
3. N.H. McCoy, The Theory of Numbers by McMillan.
4. I. Niven, I and H.S. Zuckermann, An Introduction to the Theory of Numbers.
5. A. Gareth Jones and J Mary Jones, Elementary Numbers Theory, Springer Ed. 1998.
Unit– III Riemann Zeta Function and
Diophantine Equations
Structure
3.0 Introduction
3.11 Summary
3.12 Keywords
3.13 References
3.0 Introduction
In this unit, we will study the Riemann Zeta function and its convergence, application of this function to
prime numbers, and representation as Euler function, Diophantine equations and Waring problem.
3.1 Unit Objectives
After going through this unit, one will be able to
This series is convergent for s > 1 (proved in next theorem) and the sum ζ ( s) is called the Riemann Zeta
function.
3.2.2 Comparison Test. We shall use the following result during the proof of the theorem:
∞ ∞
=
Letn
n 1= n 1
∑a
n and ∑b be two series of positive terms such that an ≤ bn ∀ n , then
∞ ∞
(i) ∑b
n =1
n converges ⇒ ∑a
n =1
n converges
and
∞ ∞
(ii) ∑a
n =1
n diverges ⇒ ∑b
n =1
n diverges.
∞
1
3.2.3 Theorem. The infinite series ζ ( s) = ∑ s
converges for all real s > 1 and diverges for all real s ≤ 1.
n =1 n
Case I. s > 1.
We group the terms together in blocks of length 1, 2, 4, 8,...., giving
1 1 1 1 1 1
ζ ( s ) =+
1 s
+ s + s + ...... + s + s + ...... + s + ......
2 3 4 7 8 15
1 1 1 1 2
Now, we have s
+ s ≤ s + s = s = 21− s
2 3 2 2 2
1 1 1 1 4
s
+ ...... + s ≤ s + ...... + s = s = (21 − s ) 2
4 7 4 4 4
1 1 1 1 8
s
+ ...... + s ≤ s + ...... + s = s = (21 − s )3
8 15 8 8 8
Case II. s ≤ 1.
1
The series (1) obviously diverges for s ≤ 0, since then / 0 as n → ∞ , so let us assume that s > 0. By
→
ns
grouping the terms of (1) in blocks of length 1, 1, 2, 4,...., we have :
1 1 1 1 1
ζ (s) =
1+ + + + + ...... + s + ......
2 s 3s 4 s 5 s 8
1 1
If s ≤ 1, then ≥
2 s
2
1 1 1 1 1
+ s ≥ + =
3 s
4 4 4 2
1 1 1 1 1
+ ...... + s ≥ + ...... + =
5s 8 8 8 2
1 1 1
and so on, so the series (1) diverges by comparison with the divergent series 1 + + + + ......
2 2 2
Note. Sometimes, Riemann Zeta function ζ ( s) is denoted as ξ ( s) . One should not be confused with the
notation.
3.3 Application to Prime numbers
3.3.1 Results.
∞ ∞
a. An infinite series ∑a
n =1
n is said to be absolutely convergent if ∑| a
n =1
n | converges.
b. The notion of convergence and absolute convergence are same in an infinite series of positive terms.
c. If an infinite series is absolutely convergent, then we can rearrange the terms without changing the
sum.
60 Arithmetic Functions
3.3.2 Theorem. Prove that there are infinitely many primes.
Proof. Let, if possible, there are only finitely many primes, say p1 , p2 ,......, pk . For each prime p = pi we
1
have that < 1 , so there is a convergent geometric series
p
1 1 1 1
1+ + 2 + 3 + ...... = −1 .
p p p 1− p
Now, these convergent series all consists of positive terms, so they are absolutely convergent. It follows
that we can multiply out the series in (1) and rearrange the terms.
1 1
If we take a typical term from the first series, from the second series and so on where each
p1e1 p2e2
ei ≥ 0 , then their product
1 1 1 1
...... ek = e1 e2
p1e1 p2e2 pk p1 p2 ...... pkek
will represent a typical term in the expansion of (1). By the Fundamental Theorem of Arithmetic, every
integer n ≥ 1 has a unique expression
n = p1e1 p2e2 ...... pkek (ei ≥ 0)
as a product of powers of the primes pi . Since we are assuming that these are the only primes, it must be
noted that we take ei = 0 , in case n is not divisible by a particular prime pi. This implies that each n
1
contributes exactly one term to (1), so the expansion takes the form:
n
k
1 1 ∞
1
∏ 1 + p + 2 ∑
+ ...... = .
i =1 i pi n =1 n
The right-hand side is divergent, however, we have that left hand side is finite, so this contradiction
proves that there must be infinitely many primes. This completes the proof.
3.3.3 Theorem. Let pn denotes the nth prime (in increasing order), then prove that the infinite series,
∞
1 1 1 1
∑
n =1
= + + + ......
pn 2 3 5
is divergent.
∞
1
Proof. Let, if possible, the series ∑
n =1 pn
converges to a finite sum l, then its partial sums must satisfy
N
1 1
∑
n =1 pn
−l ≤
2
for all sufficiently large N.
Analytical Number Theory 61
1 1
Thus, it follows that : ∑
n>N
≤
pn 2
for any such N.
k
∞
1
This implies that the series :
=
∑ ∑
k 1 n > N pn
......(1)
∞
1
converges by comparison with the geometric series ∑2
k =1
k
.
Let q denotes the product p1 p2 ..... pN , then it is clear that no integer of the form qr + 1 (r ≥ 1) can be
divisible by any of p1 , p2 ,....., pN . So, it must be a product of primes pn for n > N, possibly with
repetitions, say qr + 1 =pn pn ...... pn where each ni > N
1 2 k
1 1
Then, the reciprocal of each such an integer appears as a summand in the expansion
qr + 1 pn1 pn2 ..... pnk
k
1
of ∑ and hence, it appears as a summand in the expansion of (1). Since the series (1) converges,
n > N pn
∞
1
it follows that the series ∑ qr +1
r =1
which is contained within (1), also converges. But, this is a
contradiction, since in the what follows, we are going to show that this series is divergent.
∞
1 1 ∞ 1 1 ∞ 1
Consider the series,
=
∑
=
r 1 qr
=
= ∑ ∑
r +1 q r 2 r
+ q q r 1=
1 1
which is divergent and we have > for all r ≥ 1
qr + 1 qr + q
∞
1
So, by comparison test, the series ∑ qr +1
r =1
is also divergent.
∞
1
This contradiction forces us to say that the series ∑
n =1 pn
diverges and the proof is completed.
In order to prove the required result, we shall show that Pk ( s) → ζ ( s) as k → ∞. If we expand the
1
product Pk ( s) , the general term in the resulting series is where n = p1e1 p2e2 ..... pkek and each ei ≥ 0 . The
ns
Fundamental theorem of Arithmetic implies that each such n contributes just one term to Pk ( s) , so
1
Pk ( s ) = ∑n
n ∈ Ak
s
where= n : n p1e p2e ..... pke , ei ≥ 0} is the set of integers n whose
Ak {= 1 2 k
Then, we have that each n ∉ Ak is divisible by some prime p > pk and so n > pk . It follows that for s >1
1 1
| Pk ( s ) − ζ ( s ) |= ∑
n ∉ Ak ns
≤ ∑
n > pk ns
1
= ζ (s) − ∑
n ≤ pk ns
.
1
Now, we know that for s > 1, the partial sums of the series ∑n s
converge to ζ ( s) , so in particular
1
∑
n ≤ pk ns
→ ζ ( s ) as k → ∞.
3.4.3 Definition. An arithmetic function f is multiplicative if f (mn) = f (m) f (n) for all relatively prime
positive integers m and n.
(b) An arithmetic function f is completely multiplicative if f (mn) = f (m) f (n) for all positive integers m
and n.
∞
3.4.4 Theorem. (a) If f is multiplicative and ∑
n =1
f ( n) is absolutely convergent then prove that
∞
∑
n =1
f ( n) = ∏ (1 + f ( p) + f ( p
p
2
) + .....) where p ranges over all primes.
∞
(b) If f is completely multiplicative and ∑
n =1
f ( n) is absolutely convergent, then prove that
∏ (1 + f ( p) + ( f ( p) ) )
∞
1
∑ + ..... = ∏ where p ranges over all primes.
2
f ( n) =
n =1 p p 1 − f ( p)
Proof. (a) Let p1 , p2 ,...., pk be the first k primes. Then consider the products
k
Pk = ∏ (1 + f ( p ) + f ( p
i =1
i
2
i ) + ......) .
Analytical Number Theory 63
The general term in the expansion of Pk is:
n whose prime factors are among p1 , p2 ,....., pk . Then, we have that each n ∉ Ak is divisible by some prime
p > pk and so n > pk .
∞
Now, consider Pk − ∑ f (n) = ∑ f ( n) ≤ ∑ | f ( n) | ≤ ∑ | f ( n) | .
n =1 n ∉ Ak n ∉ Ak n > pk
∞ ∞
Since it is given that ∑ f ( n)
n =1
is absolutely convergent so ∑ | f ( n) |
n =1
is convergent, so that
∞
∑
n > pk
| f (n) | → 0 as k → ∞ . It follows that Pk → ∑
n =1
f (n) as k → ∞ . This completes the proof.
∑
n =1
f ( n) = ∏ p
(1 + f ( p ) + f ( p 2 ) + ......)
= ∏ (1 + f ( p) + ( f ( p))
p
2
+ ......)
1
=∏ .
p 1 − f ( p )
convergent.
∞
(b) Give the Euler product representation of a complete multiplicative function f, given that ∑ f ( n)
n =1
is
absolutely convergent.
∞
f ( n)
3.4.6 Corollary. Suppose that ∑
n =1 ns
converges absolutely, then
∞
f ( n) f ( p) f ( p2 )
(i) If f is multiplicative, ∑ n s
= ∏ 1 + s +
p p p 2s
+ ...... .
n =1
∞
f ( n) 1
(ii) If f is completely multiplicative, ∑n =1 ns
=∏
p 1 − f ( p) p − s
.
64 Arithmetic Functions
Proof. It is clear that if f (n) is multiplicative (completely multiplicative), then f (n). n − s is also
multiplicative (completely multiplicative), so proof is completed if we replace f (n) by f (n). n − s in the
above theorem.
∞
µ ( n) 1
3.4.7 Theorem. Prove that, for s > 1 ∑
n =1 ns
=
ζ (s)
where µ (n) is Mobius function.
1
= ∏ (1 − p
p
−s
)=
ζ (s)
.
(ii) The infinite product expansion of sin z (taken from the analysis) is given by
∞
z2
sin z z
= ∏ 1 − 2 2
.
n =1 nπ
eθ + e −θ
(iii) (a) coth θ = (b) coth θ = i cot iθ
eθ − e −θ
π2
3.5.2 Theorem. Prove that ζ (2) = .
6
We shall expand this product and find the coefficient of z3 in the expansion, we have :
z2 z2 z3 z2
sin z =z 1 − 2 1 − 2 2 1 − 2 2 1 − 2 2 ......
π 2 π 3 π 4 π
Analytical Number Theory 65
z3 z2 z2 z2
= z− 1 − 1 − 1 − ......
π 2 22 π 2 32 π 2 42 π 2
z3 z3 z5 z2 z2
= z − 2 2 − 2 + 2 4 1 − 2 2 1 − 2 2 ......
2π π 2 π 3π 4 π
z3 z3 z5 z3 z5 z5 z7 z2
= z − 2 2 − 2 2 + 2 2 4 − 2 + 2 4 + 2 4 − 2 2 6 1 − 2 2 ......
3π 2π 2 .3 π π 3π 2π 2 3 π 4 π
Continuing like this, we observe that the coefficient of z3 in the expansion of sin z is
1 1 1 ∞
1
−∑ 2 2
− 2 + 2 2 + 2 2 + ...... = .
π π 2 π 3 n =1 n π
Hence proved.
(−1) k −1 22 k −1 π 2 k B2 k
3.5.3 Theorem. Prove that ζ (2k ) = where Bm’s are called Bernoulli’s number and are
2k !
1 r −1 r 1 if r =1 r
given by the equations ∑ Bm = 0 if
r! m=0 m r >1
. Here, denotes the Binomial coefficients.
m
∞
z 2 k +1 ∞
z 2 k −1
=
∑
= 2=
n
k 0=
2k + 2
π 2k + 2
2 ∑ 2k 2k
k 1 n π
1 ∞ ∞
z 2 k −1
so that (1) becomes cot z= − 2∑ ∑ 2 k 2 k
z =k 1=n 1 n π
1 ∞
z 2 k −1
= − 2∑ ζ (2k ) 2 k . ......(2)
z k =1 π
We shall find another expansion of cot z now. Consider the exponential series
t2 t3
et = 1 + t + + + ......
2! 3!
et − 1 t t2
⇒ =1 + + + ......
t 2! 3!
The reciprocal of this has a Taylor series expansion which can be written in the form
−1
t t t2 ∞
Bm m
= 1 + + + ...... =∑ t ......(3)
e −1 2! 3!
t
m=0 m!
for certain constants B0 , B1 ,......, known as Bernoulli numbers and will be evaluated in the last part of the
theorem.
t −
t
t t et + 1 t e 2 + e 2
Now, we can write = =
− 1 − 1
et − 1 2 et − 1 2 2t −
t
e −e
2
t t t it
= coth
= − 1 i cot − 1 .
2 2 2 2
it
Putting z = , we get
2
m
z ∞
Bm 2 z
z cot z − =∑
i m = 0 m! i
m
∞
Bm 2 m −1
⇒ cot z =− i + ∑
m=0
z
m! i
. ......(4)
(−1) k −1 22 k −1 π 2 k B2 k
This implies that ζ (2k ) = .
(2k )!
∞
Bm m ∞ t n
t= ∑ t ∑ [Binomial expansion of et ] ......(5)
=m m! n 1 n!
0=
π 4 B4 2π 6 B6
⇒ ζ (2) ==
π 2 B2 , ζ (4) − , ζ (6) = and so on.
3 45
B0 + 2 B1 =
0
0 and
B0 + 3B1 + 3B2 = so on.
π2
so that we get ζ (2) =
6
π4
ζ (4) = ,
90
π6
ζ (6) = and so on.
945
3.6 Diophantine Equations. A Diophantine equation is an equation in more than one variable and with
integral coefficients such as ax + by
= c , x 3 + y=
3
z 3 , z 2 + y=
2
z2 .
Similarly, if y = 0 then x2 = z2 ⇒ x= ± z
and if z = 0 then x2 + y 2 =
0 ⇒ x= 0= y
Again if ( x, y, z ) is a solution of (1), then (dx, dy, dz ) is also a solution of (1) for all d. Without loss of
generality we assume g.c.d. ( x, y, z ) =1.
⇒ d 2 / x2 , d 2 / y 2 ⇒ d 2 / x2 + y 2
⇒ d 2 / z2 ⇒ d/z
⇒ g.c.d.( x, y, z ) ≥ d > 1
Similarly if g.c.d.( x, z )= d > 1 , then g.c.d. ( x, y, z ) ≥ d > 1 and same holds if g.c.d.(y, z) > 1
Since g.c.d. (x, y) = 1, so both of x and y cannot be even. Let both x and y be odd, then
x 2 ≡ 1(mod 8) , y 2 ≡ 1(mod 8)
Analytical Number Theory 69
⇒ x 2 + y 2 ≡ 2(mod 8)
⇒ z 2 ≡ 2(mod 8)
But there is no integer z with z 2 ≡ 2(mod 8) . So if ( x, y, z ) satisfies (1) and g.c.d. (x, y) = 1, then one of x
and y must be odd and other must be even.
Without loss of generality we assume that x is even and y is odd, since if ( x, y, z ) is a solution of (1),
( y , x, z ) is also a solution of (1).
where a > b > 0 and a and b are of opposite parity with g.c.d. (a, b) =1.
Proof. Suppose ( x, y, z ) is given by (2) where a and b satisfy given condition. Then we shall show that
x, y, z are positive solutions of (1). Clearly x > 0, y > 0, z > 0 since a > b > 0 .
Setting = y a 2 − b 2 , we get
x 2ab, =
x 2 + y 2 = (2ab) 2 + (a 2 − b 2 ) 2 = (a 2 + b 2 ) 2 = z 2 .
To prove g.c.d. ( x, y, z ) = 1 , it is enough to prove that g.c.d. (y, z) = 1, where y and z are given by (2)
Let d= g.c.d. (a 2 − b 2 , a 2 + b 2 ) .
Then d a 2 − b 2 and d a 2 + b 2 ⇒ d / (a 2 − b 2 ) ± (a 2 + b 2 )
⇒ d / 2 g.c.d.(a 2 , b 2 )
⇒ d /2 ⇒ d =1 or d = 2.
Since a and b are of opposite parity, both of a 2 − b 2 and a 2 + b 2 are odd.
⇒ d≠2 ⇒ d = 1.
Conversely, let ( x, y, z ) be any positive primitive solution of (1). Then, we know that
70 Arithmetic Functions
g.c.d.( x, y ) 1,=
= g.c.d.( x, z ) 1 .
g.c.d.( y, z ) 1,=
From (1),
x2 = z 2 − y 2 = ( z + y) ( z − y) ...(3)
z+ y z−y
Now, we claim that g.c.d. , = d= 1 .
2 2
z+ y z− y z+ y z− y
Now, d , d ⇒ d ±
2 2 2 2
⇒ d / z and d / y ⇒ d / g.c.d.( y, z ) = 1 ⇒ d = 1.
x
Since x is even ⇒ is an integer. So, from equation (4), we see that product of two coprime natural
2
z+ y z−y
numbers and is the square of an integer.
2 2
z+ y z−y
⇒ Both of and are square of integers.
2 2
z+ y z−y
Let
= 2
a= and b2 where a > 0, b > 0 . ...(5)
2 2
2 2
⇒ g.c.d.(a, b) = 1
Substituting these values in (1) and noting that x > 0, we get x = 2ab.
Since y is odd, so from (6), we get a and b are of opposite parity. Hence the theorem.
3.6.3 Example. Find all the solutions of =
x 2 + y 2 z 2 where 0 < z < 30 ......(1)
Solution. First we assume x > 0, y > 0, z > 0, g.c.d. ( x, y ) ≡ 1 and 2/x. Then we know that all the
solutions of (1) are given by
Analytical Number Theory 71
2ab, y =
x= a 2 − b2 , z =
a 2 + b2 .......(2)
∴ =x 20,
= y 21,
= z 29 .
Solution. Without loss of generality, we assume x > 0, y > 0, z > 0, g.c.d. (x, y) = 1 and 2/x.
We know x = 2ab, y =a 2 − b 2 , z =a 2 + b 2 where a > b > 0
Setting = y a 2 − b 2 , we get =
x 2ab, = xy 2ab(a 2 − b 2 ) ......(3)
Since a and b are of opposite parity, one of a and b must be even and other must be odd.
Therefore, xy ≡ 0(mod 4) ......(4)
72 Arithmetic Functions
If 3/a or 3/b then from (3), xy ≡ 0(mod 3) .
a 2 ≡ 1 ≡ b 2 (mod 3) ⇒ a 2 − b 2 ≡ 0(mod 3) .
From (6) and (8), we get xyz ≡ 0(mod 60) in this case.
a 4 ≡ 1 ≡ b 4 (mod 5) ⇒ a 4 − b 4 ≡ 0(mod 5)
and in this case also, from (6) and (7), xyz ≡ 0(mod 60) . This proves (i).
Now, y a 2 − b2
x 2ab, =
=
xy ( x 2 − y=
2
) 2ab(a 2 − b 2 ) (2ab) 2 − (a 2 − b 2 ) 2
= 2ab(a 2 − b 2 ) (4a 2b 2 − a 4 − b 4 + 2a 2b 2 )
= 2ab(a 2 − b 2 ) (6a 2 b 2 − a 4 − b 4 )
= 2ab(a 2 − b 2 ) (7 a 2 b 2 − a 2 b 2 − a 4 − b 4 )
≡ 2ab(a 2 − b 2 ) (− a 2 b 2 − a 4 − b 4 ) (mod 7)
≡ − 2ab(a 2 − b 2 ) (a 4 + b 4 + a 2 b 2 ) (mod 7)
≡ − 2ab(a 6 − b 6 ) (mod 7)
b 6 ≡ 1 ≡ a 6 (mod 7)
3.6.6 Theorem. x 4 + y 4 =
u 2 has no non-trivial solution.
Then by assumption, S ≠ φ . So by law of well ordering S has a least element. Let u0 be the least element
of S. Then there exists x0 ∈ N , y0 ∈ N such that
x04 + y04 =
u02 ...(2)
Let g.c.d. ( x0 , y0 , u0 )= d ≥ 1 .
⇒ d 4 x04 + y04
⇒ d 4 / u02
⇒ d 2 / u0 .
4 4 2
x y u0 x0 y0 u0
Then 0 + 0 = d 2 , that is, d , d , d 2 satisfy equation (1) and so
d
d
u0 u0
∈S ⇒ ≥ u0 [ u0 is least element is S ]
d2 d2
⇒ 1≥ d2 ⇒ d=1
∴ g.c.d. ( x0 , y0 , u0 ) = 1 .
74 Arithmetic Functions
Then x0, y0 cannot be both even, since then u0 is also even and g.c.d. ( x0 , y0 , u0 ) ≥ 2 . But also, x0 and y0
cannot be both odd, since in that case
Then by last theorem these exist positive integers a and b such that
x02 =
2ab, y02 =−
a 2 b 2 , u0 =+
a 2 b2 ...(3)
1 and a and b are of opposite parity. If possible, let ‘a’ be even and b be odd.
where a > b > 0, g.c.d.(a, b) =
Then from (3), y02= a 2 − b 2 ≡ −1(mod 4)
But there does not exist any integer n such that n 2 ≡ −1(mod 4) .
Since g.c.d.
= (a, b) 1=
and b 2c ⇒ g.c.d.(a, c) = 1 [ a is odd ]
Now (4) gives that square of an integer is equal to product of two positive integers where both are
relatively coprime. So, both a and c must be square of integers.
Let a
= 2
f= and c g 2 . Since g.c.d. (a, b) = 1
⇒ g.c.d.( f 2 , 2 g 2 ) = 1 .
⇒ y02 + 4 g 4 =
f4 ...(5)
But g.c.d. ( f 2 , 2 g 2 ) = 1
⇒ g.c.d. ( y0 , 2 g 2 ) = 1 .
Further a and b are of opposite parity so from (3), y0 must be odd. Now (5) can be written as
2
( f 2 ) 2 where g.c.d. (2 g 2 , y0 ) = 1 , y0 is odd, 2g is even.
( y0 ) 2 + (2 g 2 ) 2 =
Analytical Number Theory 75
Then by last theorem, there exist integers r and s such that
2g 2 = r 2 − s2 ,
2rs , y0 = f2 =
r 2 + s2 ...(6)
But g.c.d. (r, s) = 1, so we have product of relatively prime integers is the square of an integer.
⇒ r and s must themselves be squares.
Let r = v2 and s = w2, where v > 0, w > 0.
Now, from (6), f 2 = r 2 + s 2 = (v 2 ) 2 + ( w2 )=
2
v 4 + w4 .
⇒ f ≥ u0 .
3.7.1 Theorem. Let n be a natural number of the form 4k + 3 , then n cannot be written as sum of two
squares.
Proof. If possible, let =
n x2 + y 2 .
⇒ n = x 2 + y 2 ≡ 0 or 1 or 2(mod 4) .
3.7.2 Theorem. Let = n x 2 + y 2 , then primes of the form 4k +3 can occur in the prime factorization of n
to an even degree only. In other words, if a prime of the form 4k +3 occurs to an odd degree in the prime
factorization of a natural number n then n cannot be written as a sum of squares of two natural numbers.
Proof. Let p be a prime of the type 4k + 3 and let n = p 2 k +1 . n1 , where k ≥ 0 and g.c.d. (n1 , p) = 1 .
Let =
n p 2 k +1 n=
1 x2 + y 2 .
Then, x 2 + y 2 ≡ 0 (mod p) .
Now, g.c.d.(p, x) = 1
⇒ ∃ an integer q such that xq ≡ 1(mod p)
76 Arithmetic Functions
Now, x 2 + y 2 ≡ 0(mod p ) .
Let x
= px
= 1 and y py1 .
⇒ x12 + y12 =
p 2 k −1 . n1
Then n1n2 =
(a 2 + b 2 ) (c 2 + d 2 ) = (ac + bd ) 2 + (ad − bc) 2 .
This set contains n 2 > p elements. So, by Pigeon hole principle, there are atleast two members of S
which must be congruent modulo p. Let these be ax1 − y1 and ax2 − y2 , where either x1 ≠ x2 or y1 ≠ y2 . Then
we can write
ax1 − y1 ≡ ax2 − y2 (mod p )
⇒ a ( x1 − x2 ) ≡ y1 − y2 (mod p ) .
ax ≡ y (mod p ) .
Similarly, if x0 = 0, then we can obtain y0 = 0. But we have assumed above that either x1 ≠ x2 or y1 ≠ y2 ,
that is, either x0 ≠ 0 or y0 ≠ 0 . Hence both x0 and y0 must be non-zero. Hence 0 < | x0 | < p and
0 < | y0 | < p .
Proof of theorem. We write n as n = p1 p2 ...... pk m12 where p1 , p2 ,...., pk are primes of the type 4k + 1 or 2
and m1 is a product of primes of the type 4k + 3.
Now, first we prove that each prime p of the form 4k +1 can be written as a sum of two squares. We
know that −1 is a quadratic residue for the primes of the form 4k +1. So, the quadratic congruence
x 2 ≡ −1(mod p ) has a solution, say, a such that g.c.d. (a, p ) = 1 and so a 2 ≡ −1(mod p ) .
Now, since g.c.d. (a, p) = 1 , so the linear congruence ax ≡ y (mod p) has solution ( x0 , y0 ) such that
0 < | x0 | < p and 0 < | y0 | < p .
Now, a 2 ≡ −1(mod p )
⇒ 0 < mp < 2 p
⇒ 0 < m < 2.
But m is an integer ⇒ m =1
∴ p x02 + y02 .
=
Hence each pi in the representation of n can be written as a sum of two squares and so by lemma (1),
p1 p2 ...... pk can also be written as sum of the two squares. Let p1 p2 ...... p=
k A2 + B 2 .
Now, =
p x2 + y 2 ⇒ x 2 + y 2 ≡ 0(mod p )
⇒ x 2 ≡ − y 2 (mod p )
⇒ x 2 ≡ h 2 y 2 (mod p ) [ h 2 ≡ −1(mod p) ]
⇒ x ≡ hy (mod p )
or x ≡ − hy (mod p ) [ p is prime ]
By changing the sign of y if necessary, we can assume x ≡ hy (mod p) .
Now, p 2 = ( x 2 + y 2 ) ( X 2 + Y 2 ) = ( xX + yY ) 2 + ( xY − yX ) 2 ......(1)
The only representation of (2) as sum of two squares are 1 =(± 1)2 + 0 or 1 = 0 + (± 1)2 .
or 0
xY − yX = ......(4)
Case I : xY − yX =
0
⇒ xY = yX ......(5)
Now, p x2 + y 2
= ⇒ g.c.d.( x, y ) = 1
and p X 2 +Y2
= ⇒ g.c.d.( X , Y ) = 1 .
But x2 + y 2 = X 2 + Y 2 = p
⇒ y2 = Y 2 ⇒ y= ±Y .
Now, xX = − yY
⇒ x / − yY ⇒ x / −Y
⇒ x= ±Y
and Y / − xX ⇒ Y /−x.
Similarly, y = ± X .
80 Arithmetic Functions
3.7.6 Four square theorem. Every natural number n can be written as a sum of four squares.
Proof. If n = 1, then 1 = 12 + 02 + 02 + 02 .
be the prime factorization of n. If every prime p can be written as a sum of four squares then the above
theorem will follow from (1), if we able to prove following two lemmas:
Lemma 1. Product of two numbers which can be written as a sum of four squares is also representable
as sum of four squares.
Proof. Let n1 = a 2 + b 2 + c 2 + d 2 and n2 = x 2 + y 2 + z 2 + u 2 .
Then, n1n2 = (a 2 + b 2 + c 2 + d 2 ) ( x 2 + y 2 + z 2 + u 2 )
p − 1
Proof. Consider the sets S= 2
1+ x : x =
0, 1, 2,....,
2
2 p −1
and − y : y =
T = 0, 1, 2,...., .
2
p +1
Then, clearly, each of S and T contains elements. We claim that elements of S are mutually
2
incongruent (mod p). If possible, let x1 , x2 be two elements of S such that x1 ≠ x2 and
p −1
1 + x12 ≡ 1 + x22 (mod p ) where 0 ≤ x1 , x2 ≤
2
⇒ p / ( x22 − x12 )
⇒ p / ( x2 − x1 ) ( x2 + x1 ) .
So, either of (1) is not possible, so our assumption is wrong. Hence, elements of S are mutually
incongruent (mod p).
Similarly, we can prove that elements of T are incongruent (mod p).
Analytical Number Theory 81
Now, consider the set S ∪ T . Clearly, S ∪ T contains ( p + 1) distinct elements, so atleast two of them
must be congruent (mod p). But elements of S (as well as elements of T) are mutually incongruent (mod
p), so there must exist an element of S which is congruent to an element of T (mod p), that is, there exist
p −1 p −1
integers x, y, 0 ≤ x ≤ , 0≤ y≤ such that
2 2
1 + x 2 ≡ − y 2 (mod p ) ⇒ 1 + x 2 + y 2 ≡ 0(mod p ) .
2 2
Proof of the Theorem. We have to show that each prime p is representable as sums of four squares. If p
= 2, then 2 = 12 + 12 + 02 + 02 and result is true.
Now, let p be any odd prime, then there exist integers a, b, c, d , m such that
a 2 + b2 + c2 + d 2 =
mp where 1 ≤ m < p .
or 0 (mod m) .
a 2 + b2 + c2 + d 2 =
m m
where − ≤ x, y , z , u ≤
2 2
then x 2 + y 2 + z 2 + u 2 ≡ a 2 + b 2 + c 2 + d 2 ≡ 0(mod m)
⇒ r≤m
82 Arithmetic Functions
So, we have that 0≤r≤m
⇒ x= y= z= u= 0
⇒ m 2 / mp [Using (2)]
⇒ m/ p ⇒=m 1=
or m p
But m < p ⇒ m = 1 and hence by (2), p is representable as a sum of four squares and this proves
the theorem.
On the other hand, if r ≠ 0 , then 1 ≤ r ≤ m . We shall show that m = 1. On the contrary, we assume that
m ≠ 1 , so that 1 < m < p .
Let r = m, then x2 + y 2 + z 2 + u 2 =
m2 [By (4)] ...(5)
m m m
Since − ≤ x, y , z , u ≤ , so (5) is possible only if x= y= z= u= , so that we have by (3) that
2 2 2
m m m m
a= (mod m) , b = (mod m) , c ≡ (mod m), d ≡ (mod m)
2 2 2 2
m m
a
= + a1m , b
= + a2 m
2 2
m m
c
= + a3 m , d= + a4 m
2 2
Thus, we have mp = a 2 + b 2 + c 2 + d 2
= 0(mod m 2 )
cx + dy − az − bu ≡ ca + db − ac − bd ≡ 0(mod m)
dx − cy + bz − au ≡ da − cb + bc − ad ≡ 0(mod m)
where the expression in the R.H.S. of (7) are integers, so we can write rp as a sum of four squares. But
1 ≤ r < m and this contradicts to the minimality of m. Hence, our assumption that 1 < m < p is wrong ⇒
m=1. This completes the proof.
3.7.7 Remark. The above theorem is a very famous one given by Lagrange and is termed as
“Lagrange’s four square theorem”.
3.8 Waring Problem. Waring problem deals with the representation of a natural number as a sum of
fixed number of squares or cubes or fourth powers and so on. In 1770, Waring stated (without proof)
that a natural number can be written as a sum of four squares, nine cubes, nineteen biquadratics and 37
fifth powers and so on.
In 1909, Hilbert proved that given any natural number n and k ≥ 2, there exists a fixed number s
(depending on k) such that n can be written as a sum of s kth powers. In the context of Waring problem,
we define the following terms:
3.9 The numbers g(k) & G(k)
g(k) : g(k) is defined as the smallest number such that every natural number can be written as a sum of
g(k) kth powers. Clearly g(k) depends only on k and not on the natural numbers to be represented as a
sum.
G(k) : G(k) is defined as the smallest natural number such that every natural number, except a finite
natural numbers, can be written as a sum of G(k) kth powers.
⇒ g (2) = 4
84 Arithmetic Functions
3.9.3 Theorem. Prove that G(2) = 4.
Proof. We know G (2) ≤ g (2) =
4. ......(1)
By definition, G (k ) is the smallest natural number that all the natural numbers (except a finite number)
can be written as a sum of G(k) kth powers. So to prove G(2) = 4, it is enough to prove that an infinite
number of natural numbers cannot be written as a sum of three squares.
For this we shall prove that no natural number n of the form 8k+7 can be written as a sum of three
squares.
Let n = a 2 + b2 + c2 . ......(2)
b2 ≡ 0 or 4(mod 8)
c 2 ≡ 0 or 4(mod 8)
∴ n = a 2 + b 2 + c 2 ≡ 0 or 4(mod 8)
(b) Two of the numbers a, b, c are even and one is odd. To be specific, let a and b be even and c be odd,
then
a 2 ≡ 0 or 4(mod 8)
b2 ≡ 0 or 4(mod 8)
c 2 ≡ 1 (mod 8)
∴ n = a 2 + b 2 + c 2 ≡ 1 or 5(mod 8)
b 2 ≡ 1 (mod 8)
c 2 ≡ 1 (mod 8)
∴ n = a 2 + b 2 + c 2 ≡ 2 or 6(mod 8)
b 2 ≡ 1 (mod 8)
c 2 ≡ 1 (mod 8)
Analytical Number Theory 85
∴ n = a 2 + b 2 + c 2 ≡ 3 (mod 8)
3.9.4 Remark. It is clear from the proof that if n ≡ 0(mod 4) and n = a 2 + b 2 + c 2 then a, b, c are all even.
3.9.5 Example. Prove that no number of the form 4m (8k + 7) , m ≥ 0, k ≥ 0 can be written as a sum of three
squares.
Solution. First, we prove no number of the form 8k+7 can be written as a sum of three squares. We shall
prove the result by induction on m. If m = 0, then = n 8k + 7 and we have proved that no natural number
of this form can be written as a sum of three squares. So assume no number of the form
4m −1 (8k + 7), (m ≥ 1) can be written as a sum of three squares.
Then, n ≡ 0(mod 4) .
If possible, let ∃ integers a, b, c such that n = a 2 + b 2 + c 2 , then by the remark made earlier a, b, c must be
all even.
2 2 2
n a 2 b2 c2 a b c
∴ = + + = + + ,
4 4 4 4 2 2 2
n n
that is, can also be written as a sum of three squares. But
= 4m −1 (8k + 7) , which contradicts to the
4 4
assumption that no number of the form 4m −1 (8k + 7) can be written as a sum of three squares. This
completes the exercise.
3 k
3 k 3 k 3 k k 3
k
Solution. Let q = − 1 ≤ , ∴ N ≤ 2 − 1 = 3 − 1 ⇒ N < 3 .
k k
2 2
2 2
N 2k q − 1 , then by definition, N < 3k .
Let =
So, if we want to represent N as a sum of kth powers, the 3k cannot occur in this representation. Further,
N < 2k q . So, for the most economical representation of N as a sum of kth powers, we must take q − 1
powers of 2k in its representation.
∴ N= 2k q −=
1 2k q − 1 + 2k − 2 k
86 Arithmetic Functions
= 2k (q − 1) + 2k − 1
= (q − 1) 2k + (2k − 1) . 1k .
Thus we need exactly q − 1 + 2k − 1 =q + 2k − 2 kth powers to represent N as a sum of kth powers in the most
economical representation.
3 k
∴ − 2 + 2 k − 2 .
g (k ) ≥ q + 2= k
2
3.9.7 Theorem. G (k ) ≥ k + 1 .
Proof. To prove the theorem, we first prove a lemma.
Lemma. For every integer b ≥ 1 and r ≥ 1
b
(a + 1) (a + 2).......(a + r − 1) (b + 1) (b + 2)......(b + r )
∑
a=0 (r − 1)!
=
(r )!
......(1)
Proof. We shall prove the lemma by induction on b. However first we note that (n!) divides the product
of n consecutive integers. So, fractions appearing on both sides of (1) are integers.
Take b = 1, then L.H.S. of (1) for b =1, is equal to
(r − 1)!+ (r )! (r − 1)! r! (r + 1)!
= + =1 + r = .
(r − 1)! (r − 1)! (r − 1)! (r )!
So, assume that (1) holds for (b −1), where b ≥ 2 and we shall prove it for b. Now, L.H.S. of (1) is equal
to
b −1
b
(a + 1) (a + 2)......(a + r − 1) (a + 1) (a + 2)......(a + r − 1) (b + 1) (b + 2)......(b + r − 1)
∑
= ∑
(r − 1)!
a 0=a 0 (r − 1)!
+
(r − 1)!
(b + 1) (b + 2)......(b + r − 1) (b + r )
= = R.H.S. of (1)
(r !)
is solvable.
By interchanging x1 , x2 ,...., xk if necessary, we assume
Analytical Number Theory 87
1
0 ≤ x1 ≤ x2 ≤ ...... ≤ xk and xk ≤ N k . ......(2)
Since n ≤ N, then to every solution of (1), we must have a solution of (2), so that
A( N ) ≤ B( N ) ......(3)
1
N k
xk x4 x3
=
=
∑ ∑
xk 0 =
xk −1 0
......
=x3 0=x2 0
∑ ∑ (x 2 + 1)
1
N k
xk x4
( x3 + 1) ( x3 + 2)
B( N ) =
=
∑ ∑
xk 0=
xk −1 0 =x3 0
...... ∑
(2)!
.
1
N k
( xk + 1) ( xk + 2)...... xk + k −1
B( N ) = ∑
xk = 0 (k −1)!
1k 1k 1
N + 1 N + 2 ...... N k + k
= ......(4)
(k )!
Now, if possible, let G (k ) ≤ k , so that all but a finite number can be written as a sum of k kth powers, so
there exists a finite number C such that
A( N ) ≥ N − C
But, we have A( N ) ≤ B( N )
N
N, B( N ) ∼ . Thus, it follows that from (5), we have for sufficiently large N,
k!
N
N ≤ , a contradiction for k ≥ 2, hence we must have :
k!
G (k ) ≥ k + 1 .
Thus, if we consider the number of the form 16m + 15, then any such number require atleast 15
biquadrates. It follows that G (4) ≥ 15 .
From (1), it follows that if 16n is the sum of 15 or fewer biquadrates, then each biquadrate must be a
multiple of 16. Hence, we can write :
15 15 15
16n
= 4
i
=i 1=i 1
∑
= x ∑ (2 yi ) 4 so that n = ∑ yi4
i =1
Hence, if 16n is the sum of 15 or fewer biquadrates, so is n. But we observe that 31 is not the sum of 15
or fewer biquadrates. In fact, the most economical representation contains 16 biquadrates given by
= 24 + 15 . 14 .
31
x 2 ≡ 0(mod 2θ + 2 ) .
θ
⇒
θ θ
x 2 = (1 + 2m) 2 ≡ 1 + 2θ +1 m + 2θ +1 (2θ −1) m 2
≡ 1 + 2θ +1 m + 22θ +1 m 2 − 2θ +1 m 2
≡ 1 + 2θ +1 m − 2θ +1 m 2
≡ 1 − 2θ +1 m(m − 1) ≡ 1(mod 2θ + 2 )
Analytical Number Theory 89
Now, let n be any odd number and suppose that 2θ + 2 n is written as a sum of 2θ +1 −1 or fewer kth powers
where k = 2.
2θ + 2 . n = x1k + x2k + ...... + x2kθ + 2 −1
then from (2), we get that each xi must be even and hence divisible by 2k . Hence, we obtain that
2k −θ − 2 / n which implies that n is even, a contradiction. Hence, we must have
G (2θ ) ≥ 2θ + 2 for θ ≥ 2 .
Let n be a natural number such that p / n i.e., ( p, n) = 1 and suppose that pθ +1 . n is the sum of pθ +1 − 1 or
fewer kth powers, that is,
pθ +1 . n = x1k + x2k + ...... + x kpθ +1 −1
then each xi must be divisible by p and hence each factor on R.H.S. must be divisible by pk which
implies that p k / pθ +1 . n , a contradiction, since k > θ +1 and (p, n) = 1
that is, G ( pθ ( p − 1) ) ≥ pθ +1 .
1 θ
Proof. Let
= k p ( p −1) , then we have
2
90 Arithmetic Functions
1 θ
θ + 1 < pθ ≤ p ( p −1) =k (except in the trivial case, p = 3, θ = 0 and k =1).
2
Since p > 2, so, p cannot divide both x k + 1 and x k − 1 and so one of x k − 1 and x k + 1 is divisible by pθ +1 .
Thus, we have obtained that x k ≡ 0, 1 or − 1(mod pθ +1 )
1 1 θ +1
It follows that number of the form pθ +1 m ± ( pθ +1 − 1) requires atleast ( p − 1) kth powers.
2 2
1 θ +1
⇒ G (k ) ≥ ( p − 1) and the proof is completed.
2
2. Check whether 2x + 3y =
c is Diophantine equation or not.
3.11 Summary
In this unit, we studied about Riemann Zeta function, Diophantine equations and the numbers g(k)
and G(k).
3.12 Keywords
Riemann zeta function, Bernoulli’s numbers, Euler product, Diophantine equation.
3.13 References
1. G. H. Hardy and E.M. Wright, An Introduction to the Theory of Numbers.
2. D.M. Burton, Elementary Number Theory.
3. N.H. McCoy, The Theory of Numbers by McMillan.
4. I. Niven, I and H.S. Zuckermann, An Introduction to the Theory of Numbers.
5. A. Gareth Jones and J Mary Jones, Elementary Numbers Theory, Springer Ed. 1998.
Unit– IV | Arithmetic Functions
Structure
4.0 Introduction
4.1 Unit objectives
4.2 Arithmetic functions
4.3 Perfect numbers
4.4 Mobius inversion formula
4.5 A general principle and its applications
4.6 The order and the average order of the arithmetic functions
4.7 Check your progress
4.8 Summary
4.9 Keywords
4.10 References
4.0 Introduction
In this unit, the concept of arithmetic functions, their definitions, examples, and simple properties, their
order, and average order are discussed. Also, perfect numbers and Mobius inversion formula are
derived.
4.1 Unit objectives
After going through this unit, one will be able to
• know about arithmetic functions and their properties.
• familiar with perfect numbers.
• derive Mobius inversion formula.
• compute the order and average order of arithmetic functions.
4.2 Arithmetic Functions
4.2.1 Arithmetic Function. A function f defined for all natural numbers n is called an arithmetic
function and is denoted by f (n) .
f (n=
1 . n2 ) f (n1 ) . f (n2 ) ∀ n1 , n2 ∈ N .
92 Arithmetic Functions
For example,
= 2
f (n) n= and f (n) 1 .
1 if n =1
µ ( n) = 0 if n is not square free .
(−1) r if n= p1 p2 ...... pr where pi 's are distinct primes
4.2.4 Definition. A number is said to be square free if it is not divisible by square of prime number, that
is, in power prime factorization each prime must have power unity (i.e., no prime is repeated).
4.2.5 Theorem. The mobius function µ(n) is a multiplicative function.
Proof. Let m, n ∈ N such that (m, n) = 1.
µ (mn) = (−1) r + s
=(−1) r (−1) s
= µ ( m) . µ ( n) .
(1)
g= (1) 1 .
f=
If either
= or n2 1 then clearly, g (n1 . n2 ) = g (n1 ) g (n2 ) .
n1 1=
By definition, we have
g (n1 . n2 )
= ∑ f (d )
=
d / n1n2
∑
d1 / n1
f (d1 . d 2 )
d 2 / n2
= ∑
d1 / n1
f (d1 ) . f (d 2 ) [ f is multiplicative]
d 2 / n2
= ∑ f (d1 ) ∑ f (d 2 )
d1 / n1 d2 / n2
= g (n1 ) . g (n2 ) .
Since µ is multiplicative function of n, so, g(n) is also multiplicative function (using Theorem 4.2.7).
If n =1, then by definition g= (1) 1 and we are through.
(1) µ=
(n) g=
Now, let n > 1 and suppose n = p1α p2α ...... prα be the prime factorization of n.
1 2 r
Then, g (n) g=
= (
p1α1 p2α 2 ........ prα r )
g ( p1α1 ) . g ( p2α 2 )........ g ( prα r ) [ g is multiplicative]
Now to prove the result, we shall prove that g ( pα ) = 0 for all primes and α ≥ 1 .
Now, by definition g ( pα ) = ∑ µ (d )
d / pα
94 Arithmetic Functions
= µ (1) + µ ( p ) + µ ( p 2 ) + ...... + µ ( pα )
= 1 − 1 + 0 + ....... + 0 = 0.
Now, divisors of n are of the form p1β p2β ...... prβ with 0 ≤ βi ≤ α i . 1 2 r
So, while considering the divisors of n, we leave out those divisors which are divisible by a square.
Therefore, the only divisors to be considered are
1, p1 , p2 ,....., pr , p1 p2 , p1 p3 ,......., p1 pr , p1 p2 p3 ,...... and so on upto p1 p2 ....... pr .
r
Then g (n) = µ (1) +
=i 1
∑ µ( p ) + ∑ i
1≤ i , j ≤ r
µ ( pi p j ) + ...... + µ ( p1 p2 ........ pr )
i≠ j
= 1 − r C1 + r C2 − r C3 + ........ + (−1) r r Cr
=(1 − 1) r =0 .
4.2.9 Exercise. Let n > 1 and let n have r distinct prime divisors, then ∑ µ (d ) = 2r .
d /n
∑
d /n
µ (d ) = µ (1) +
=i 1
∑ µ ( pi ) + ∑
1 ≤ i, j ≤ r
µ ( pi p j ) + ...... + µ ( p1 p2 ....... pr )
i≠ j
1 + r C1 + r C2 + ........ + r Cr
=
=(1 + 1) r =2r .
1
4.2.11 Theorem. If p is a prime and k > 0, then φ ( p k ) =p k − p k −1 =p k 1 − .
p
There are exactly p k −1 integers between 1 and p k which are divisible by p namely
p, 2 p, 3 p,....., ( p k −1 ) p .
Thus, the set {1, 2,....., p k } contains exactly p k −1 integers which are not relatively prime to p k and so by
1
definition of φ (n) , φ ( p=
k
) pk − =
p k −1 p k (1 − ) .
p
4.2.12 Theorem. ∑ φ (d ) = n ∀ n ∈N .
d /n
So let n > 1.
Let n = p1α p2α ........ prα be the prime factorization of n. Then
1 2 r
f ( pα ) = ∑ φ (d )
d / pα
= φ (1) + φ ( p ) + φ ( p 2 ) + ........ + φ ( pα )
=1 + ( p − 1) + ( p 2 − p ) + ........ + ( pα − pα −1 )
= pα .
Hence, ∑ φ (d ) = n .
d /n
96 Arithmetic Functions
n n
4.2.13 Theorem. Prove
= that φ (n) ∑
= µ (d ) ∑ d µ d .
d /n d d /n
n
Proof. It is clear that above two expressions on the R.H.S. are equal. For this, change d to since d/n
d
n
⇒ n.
d
Now, if n = 1, then =
φ (1) µ
= (1) 1 and we are through.
So, let n > 1 and let n = p1α p2α ........ prα , where pi’s are distinct primes.
1 2 r
1 1 1
Then we know that φ (n) =n 1 − 1 − ........ 1 −
p1 p2 pr
r
1 1 1
=n 1 − ∑ + ∑ − ∑ + ........
=i 1 p 1 ≤ i, j ≤ r pi p j 1 ≤ i , j , k ≤ r pi p j pk
i
i ≠ j i≠ j≠k
Therefore,
r
µ ( pi ) µ ( pi p j )
n 1+ ∑
φ ( n) = + ∑ + ........
=i 1 pi 1 ≤ i, j ≤ r pi p j
i≠ j
µ (d ) n
∑
= n=
d /n d
∑
d /n
µ (d ) .
d
4.2.14 Divisor Function. Let n be a natural number. We define the divisor function d(n) or τ (n) as
d ( n) = ∑ 1 = number of divisors of n.
d /n
Clearly,
= d ( p ) 2 for every prime p and d ( p n )= n + 1 .
d (1) 1,=
4.2.15 Theorem. Prove that d(n) is a multiplicative function. Also find a formula for d(n).
Proof. The function f (n) ≡ 1 is a multiplicative function and so the function
d ( n)
= ∑
= f (d ) ∑ 1
d /n d /n
is a multiplicative function.
To find the formula. If n = 1, then clearly d (1) = 1
So, let n > 1 and let n = p1α p2α ........ prα be the prime decomposition of n.
1 2 r
Analytical Number Theory 97
Now d (n) = d ( p1α p2α ........ prα )
1 2 r
Now d ( pα ) = ∑ 1
d / pα
r
⇒ d ( n)
= ∏ (α
i =1
i + 1)
4.2.16 Sum Function. Let n ≥ 1 be any natural number. We define sum function (σ n ) by
σ n σ=
= ( n) ∑ d = Sum of all positive divisors of n.
d /n
Further, σ (n) > n ∀ n > 1 since there are atleast two divisors of n namely 1 and n.
4.2.17 Theorem. Prove that σ (n) is a multiplicative function of n. Also find a formula for σ (n) .
Proof. Since f(n) = n is a multiplicative function of n, so
∑
d /n
(d )
f= ∑
= d
d /n
σ ( n)
Now, let n > 1 and let n = p1α p2α ........ prα be the prime power decomposition of n.
1 2 r
σ ( pα ) = ∑ d =1 + p + p 2 + ........ + pα
d / pα
pα + 1 − 1
= .
p −1
α +1
r
pi i −1
So, by (1) , σ ( n) = ∏
i =1 pi − 1
98 Arithmetic Functions
∞
4.2.18 Example. Evaluate ∑ µ (n !) .
n =1
∞
Solution. ∑
n =1
µ (n !) = u (1!) + µ (2!) + µ (3!) + ........ + µ (n !) + ...... ..
= 1 − 1 + 1 + 0 + 0 + ........ + 0 + ........
= 1.
4.2.19 Example. Prove that µ (n) µ (n + 1) µ (n + 2) µ (n + 3) =∀
0 n ≥ 1.
Solution. Since n, n + 1, n + 2, n + 3 are four consecutive integers and so atleast one of them is divisible by
4 and consequently µ of that number is equal to zero and so
0.
µ (n) µ (n + 1) µ (n + 2) µ (n + 3) =
Since µ=
(n) 1, 0, − 1, therefore (1) is possible only when
µ (n) = µ (n + 1) = µ (n + 2) = 1 .
For this, n, n + 1, n + 2 must be square free and each one of them must be product of even number of
primes.
Now, 33, 34, 35 satisfy above conditions. Hence, the least value of n is 33.
4.2.21 Example. Show that if f is a multiplicative function of n then g (n) = ∑ µ (d ) f (d ) is a
d /n
= ψ (n1 ) ψ (n2 ) .
∴ g ( n) = 0 .
∴ (1) 1 .
(1) µ (1) f=
(n) g=
g=
So let n > 1 and let n = p1α p2α ........ prα be the prime decomposition of n, where pi’s are distinct primes.
1 2 r
g ( pα ) = ∑ µ (d ) f (d )
d / pα
= 1 + (−1) f ( p ) + 0 + ........ + 0
= 1 − f ( p) .
Thus, we have
r
g ( n)
= ∏ (1 − f ( p ))
i =1
i
or ∏ (1 − f ( p)) , n > 1 .
g ( n) =
p/n
⇒ ∑
d /n
µ (d ) σ=
(d ) ∏p/n
(1 − σ ( p ))
=∏ (1 − ( p + 1) ) =∏ (− p ) =−
( 1) r ∏ p.
p/n p/n p/n
⇒ ∑
d /n
µ (d ) φ (d=
) ∏ (1 − φ ( p=
p/n
)) ∏ (1 − ( p −1)=
) ∏ (2 − p) .
p/n p/n
100 Arithmetic Functions
4.2.25 Remark. Now, we lead to an important theorem showing that φ (n) is a multiplicative function.
For this, we first state these results.
(i) Given integers a, b, c, we have that g.c.d. (a, bc) = 1 iff g.c.d. (a, b) = 1 and g.c.d. (a, c) = 1.
(iii) If S
= {0, 1, 2,........, n −1} , then no two elements of this set are congruent modulo n.
4.2.26 Theorem. Euler’s φ function, φ (n) is a multiplicative function, that is,
φ (mn) = φ (m) φ (n) where g.c.d.(m, n) = 1.
Proof. By definition, we know that φ (mn) is the number of natural numbers less than or equal to mn
which are coprime to mn. Equivalently, we can say that φ (mn) is the number of natural numbers less
than or equal to mn which are coprime to both m and n [using Remark 4.2.25(i)].
If m = 1 or n = 1, then the theorem is proved, so let m > 1 and n >1. Let us arrange first mn natural
numbers in n rows and m columns as :
1 2 3 ... m
m+1 m+2 m+3 ... 2m
2m +1 2m + 2 2m + 3 ... 3m
(n − 1) m + 1 (n − 1) m + 2 (n − 1)m + 3 ... nm
In this arrangement, we have to count the numbers which are coprime to both m and n. We observe that
in the rth column, the numbers are of the form qm + r (0 ≤ q ≤ n − 1) and since
g .c.d .(r , m) , it follows that the numbers in the rth column are coprime to m iff r itself
g .c.d .(qm + r , m) =
is coprime to m. But in this arrangement number of such columns is φ (m) , that is, only φ (m) columns
contain integers coprime to m.
Now, we claim that in a particular column, there are φ (n) numbers which are coprime to n. Consider rth
column. The entries in the rth column are.
r , m + r , 2m + r , ....., (n − 1) m + r .
These are n numbers and we show that no two of these congruent (mod n). Let, if possible,
km + r ≡ jm + r (mod n) where 0 ≤ k < j < n
⇒ km ≡ jm(mod n)
Combining all the above discussion, we conclude that in this arrangement, the numbers which are
coprime to both m and n, that is, to mn are φ (m) φ (n) . Thus, we have obtained
Now, let n > 1 be any natural number with prime factorization given by
n = p1k1 p2k2 ........ prkr
(This step is possible due to the last theorem and the fact that ( pik , p kj ) = 1 ) i j
1 1 1
p1k1 1 − p2k2 1 − ........ prkr
= 1 −
p1 p2 p r
1 1 1
⇒ φ ( n) =n 1 − 1 − ...... 1 − .
p1 p2 pr
This is very important result used to find the Euler’s φ-function for any natural number n. A second
proof is based on induction which we are proving in next theorem. A third proof will be given later in
this unit.
4.2.28 Theorem. If the integer n > 1 has the prime factorization, n = p1k p2k ...... prk then 1 2 r
φ ( n) =
( p1k − p1k −1 ) ( p2k − p2k
1 1 2 2 −1
)........( prkr − prkr −1 )
1 1 1
= n 1 − 1 − ........ 1 − .
p1 p2 pr
Proof. We shall prove the theorem by induction on r, the number of distinct prime factors of n. Since we
have proved earliar that
φ ( p k=
) p k − p k −1
Now, we have φ ( p1k1 p2k2 ... piki pik+i+11 ) = φ ( p1k1 )φ ( p2k2 )...φ ( pik+i+11 ).
[By using the fact that φ is a multiplicative function and that g.c.d. ( p1k p2k ........ pik , pik+1 ) = 1] 1 2 i i +1
This shows that result is true for r = i + 1 and hence by principle of mathematical induction, the result is
true for all natural numbers.
4.2.29 Example. Let us calculate the value φ (360) . The prime power decomposition of 360 is
360 = 23 . 32 . 5 .
1 2 4
= 360
= . . . 96 .
2 3 5
which is an even integer. If n does not happen to be a power of 2, then it is divisible by an odd prime p.
We may therefore write.
n = p k . m where k ≥ 1 and g.c.d. ( p k , m) = 1 .
and if n is even, then let k be the highest power of 2 in n, that is, n = 2k . m where g.c.d.(2, m) = 1.
Then φ (2n) φ=
= (2k +1 m) φ (2k +1 ) φ ( m)
Analytical Number Theory 103
1
=2 k + 1 1 − φ ( m) =2 k φ ( m) =
2 . 2 k − 1 φ ( m)
2
k 1 k −1
= 2 . φ (2k ) φ (m) φ (2 ) = 2 1 − 2 = 2
k
. φ (2k . m) 2 . φ (n)
= 2=
σ (28) = 1 + 2 + 4 + 7 + 14 + 28 = 56 .
= (2n +1 − 1) . 2n +1
= 2. 2n (2n +1 − 1) = 2m
⇒ m is a perfect number.
4.3.3 Remark. All the known perfect numbers are even. We don’t know any odd perfect number and
neither it has been proved that all perfect numbers must be even.
4.3.4 Theorem. Every even perfect number must be of the form 2n (2n +1 −1) , where 2n +1 −1 is a prime
number.
Proof. To prove the theorem, we first prove a lemma.
Lemma 1. Let σ (m=
) m + l where 1 ≤ l < m and l/m. Then l = 1 and m is a prime number.
Proof. If possible, let l > 1, then since l/m and 1 < l < m, so m has atleast three divisors 1, l, m.
∴ σ (m) =l + m + 1 , which contradicts the hypothesis σ (m) = l + m .
So, m can have only two divisors 1 and m and so m must be a prime number.
Proof of the theorem. Let k be a given even perfect number. Then k is of the form
k = 2n . m where n ≥ 1 and m is odd ......(1)
104 Arithmetic Functions
[we cannot have k = 2n , since in that case σ (2n=
) 2n +1 − 1 ≠ 2. 2n [ σ (m) ≠ 2m ] which contradicts the
fact that k is perfect.]
Now, let σ (m=
) m + l where l ≥ 1.
Also, σ (k ) = σ (2n . m)
= (2n +1 − 1) (l + m) . ......(3)
⇒ 1
2n += m 2n +1 m − l − m + 2n +1 l
⇒ =m l (2n +1 − 1) ......(4)
⇒ l/m
Also, l ≥ 1 and l < m. So by the lemma, l = 1 and m is a prime number. Setting l =1 in (4), we get
= m (2n +1 − 1) . Hence, from (1),
= k 2n (2n +1 − 1) .
(i) Let =
n 3m − 1 . Then n ≡ −1 or 2(mod 3).
So, n cannot be a perfect square since k 2 ≡ 0 or 1(mod 3) for any natural number k., that is, every perfect
square is ≡ 0 or 1(mod 3).
n n
Since d / n ⇔ n so d ≠ . [ n is not perfect square]
d d
Also, 3 / n [ n = 3m −1]
n
∴ We write σ (n)= σ (3m − 1) = ∑ d +
d /n d
n
d ≠ , 1≤ d < n
d
d2 + n
σ ( n) = ∑ ......(1)
n
d /n d
d≠ , 1≤ d < n
d
⇒ 3 / d2
⇒ d 2 ≡ 1(mod 3)
Also, n ≡ −1(mod 3)
n
⇒ d≠ [ n is not perfect square]
d
n
So, we write σ (n=
) σ (8m − 1)= ∑ d +
d /n d
n
d≠ , 1≤ d < n
d
d2 + n
or σ (8m − 1) = ∑ ......(3)
n
d /n d
d ≠ , 1≤ d < n
d
Adding, d 2 + n ≡ 0 (mod 8)
d2 + n
Dividing by d , ≡ 0(mod 8) 8 / d ∴ (8, d ) =
1 ......(4)
d
106 Arithmetic Functions
From (3) and (4), we have ) σ (8m − 1) ≡ 0(mod 8) which proves (ii).
σ (n=
(i) ∑ d (m) . This function gives us the number of divisors of all divisors of n.
m/n
=1 + 2 + 3 + 4 = 10.
(ii) ∑d 3
( m) = ∑ (d (m)) . This function gives the sum of cubes of divisors of all divisors of n.
3
m/n m/n
m /8
= 13 + 23 + 33 + 43 = 100 .
We give a theorem showing that square of first function is equal to second function.
2
4.3.6 Example. ∑
m/n
d 3 ( m) = ∑ d ( m)
m/n
.
2
Solution.
= Let f (n) ∑
= d 3 ( m) , g ( n) ∑ d ( m) .
m/ n m/ n
d 3 (m) = ( d (m) ) is also a multiplicative function and so f (n) = ∑ is also a multiplicative function.
3
d 3 ( m)
m/n
2
Further since d(n) is multiplicative, g (n) = ∑ d (m) is also multiplicative function.
m/n
To prove f (n) = g (n) , it is sufficient to prove that f ( pα ) = g ( pα ) for every prime p and α ≥ 1.
Now, f ( pα ) = ∑ α
d 3 ( m)
m/ p
f ( pα ) = d 3 (1) + d 3 ( p ) + d 3 ( p 2 ) + ........ + d 3 ( pα )
= 13 + 23 + 33 + ........ + (α + 1)3
Analytical Number Theory 107
2
(α + 1) (α + 2)
=
2
= [1 + 2 + 3 + ........ + (α + 1) ]
2
2
= d (1) + d ( p ) + ........ + d ( pα )
2
= = ∑
m / pα
d ( m)
g ( pα )
⇒ f ( n) = g ( n) .
divisor.
Solution. ∑ (−1)
d /n
d
φ (d )= ∑
d /n
φ (d ) − ∑
d /n
φ (d )
d → Even d → odd
= ∑
d /n
φ (d ) − ∑d /n
φ (d ) − ∑d /n
φ (d ) + ∑d /n
φ (d )
d → even d → odd d → odd d → odd
∑ φ (d ) − 2
=
d /n
∑d /n
n − 2 ∑ φ (d )
φ (d ) =
d /m
d → odd
= n − 2m .
Let n > 1 and n = 2r p1α p2α ........ pkα , where r ≥ 0 and p1 , p2 ,........, pk are distinct odd primes
1 2 k
Now, 1 + 2 + 3 + ...... + 2=
r
(2r +1 − 1) is odd ∀ r ≥ 0 .
However, if α i is odd for some i (1 ≤ i ≤ k ) then 1 + pi + pi2 + ........ + piα is even (since number of terms in the i
When each α i is even then p1α p2α ........ pkα = m 2 for some integer m. If r is also even then n = m 2 and if r
1 2 k
d
n or 0 according as n is odd or n is even.
φ =
n n
∴ ∑ (−1)
d /n
d −1
φ =∑ φ = ∑ φ (d ) =n .
d /n d d /n
d
Let n be even. Let n = 2r . p1α p2α ........ pkα be the prime power decomposition of n. Then, we have
1 2 k
n n/2
= ∑ φ −
d
∑ φ
d /n
d → odd
d1
n d1
2
n
= φ (2r ) ∑ φ ( pβ 1
1
p2β2 ........ pkβk ) −
2
n
(2r − 2r −1 )
= ∑ φ ( p1β1 p2β2 ........ pkβk ) −
β β β α α α 2
( p1 1 p2 2 .... pk k ) ( p1 1 p2 2 .... pk k )
n n n
= 2r −1 p1α1 ........ pkα k − = − = 0.
2 2 2
n n
Then f (n)
= ∑
= F (d ) µ ∑ µ (d ) F . Also prove its converse.
d /n d d d /n
n n n
Proof. Clearly, ∑ F (d ) µ d = ∑
d /n d /n
µ (d ) F since d / n ⇒
d d
n.
n
So let us prove f (n) = ∑ µ (d ) F .
d /n d
By definition, F n = ∑ f (c )
d c
n
d
Analytical Number Theory 109
n
∴ ∑ µ (d ) F = ∑ µ (d ) ∑ f (c)
d /n d d /n c n
d
n n
= ∑c/n
f (c ) ∑ n
µ (d ) d / n and c d
iff c / n and d
c
d
c
= f ( n) ∑ µ (d ) + ∑ f (c) ∑ µ (d )
......(1)
d /1 c/n d n
c<n c
1 if n =1
But we have ∑ µ (d ) = .
d /n 0 if n >1
So, we have ∑=
µ (d )
n
0 if c < n ......(2)
d
c
and ∑=
µ (d )
n
1=
if c n ......(3)
d
c
n
Conversely. Let f (n) = ∑ µ (d ) F .
d /n d
d
Now, ∑d /n
f (d ) = ∑ ∑ µ (δ ) F
d /n δ /d δ
[By definition of f(n)]
d
= ∑∑
δ
d /n /d
µ (δ ) F
δ
d d
= ∑∑ µ F (γ )
γ Set = γ
δ
γ
d /n /d
= ∑ F (γ ) ∑ µ ( β )
γ/n β n
γ
∑ µ (β ) 0 if n > γ
=
β n
= F ( n) . γ
= 1= if n γ
110 Arithmetic Functions
n n
then φ (n) ∑
4.4.2 Corollary. If n ≥ 1, = = d µ ∑ µ (d ) d .
d /n d d /n
⇒ N ( n) = ∑ φ (d )
d /n
n n
=φ ( n) ∑
= µ N (d ) ∑ µ d
d /nd d d /n
......(2)
n n
Proof. Given that ∑τ (d ) µ d = ∑ µ (d ) τ d
d /n d /n
n n
So by inversion formula
= u ( n) ∑
= µ τ (d ) ∑ µ (d ) τ but u (n) = 1 for each n ∈ N .
d /n d d /n d
n n
Hence, ∑
= µ τ (d ) ∑
= µ (d ) τ 1 .
d /n d d /n d
(ii) We know that N (n) = n ∀ n ∈ N , identity over N.
Therefore, by definition
σ
= ( n) ∑
= d ∑ N (d ) [ N (d ) = d ]
d /n d /n
n n
Hence, ∑
= µ σ (d ) ∑
= µ (d ) σ n .
d /n d d /n d
Analytical Number Theory 111
µ (d ) φ ( n)
4.4.4 Example. Prove that ∑
d /n d
=
n
.
Since µ(n) is a multiplicative function of n, therefore f(n) is also a multiplicative function of n. Let
n = p1α p2α ........ pkα be the prime power factorization of n. Then
1 2 k
Now, let us compute f ( pα ) , where p is any prime and α ≥ 1. The only divisors of pα are
1, p, p 2 ,........, pα .
µ (d )
∴ f ( pα ) = ∑ d
d / pα
µ ( p) µ ( p2 ) µ ( pα )
= µ (1) + + + ........ +
p p2 pα
1
=1 − + 0........ + 0
p
p −1
=
p
p − 1 pα − 1 pα − pα −1 φ ( pα )
= = . α −1 =
p p pα pα
k
φ ( piα ) i
φ ( p1α ) φ ( p2α )
1 2
φ ( pkα )k
= α1
. α2
........ αk
p1 p2 pk
= α1 α2 αk [ φ is multiplicative ]
p1 p2 ........ pk
φ ( n)
= .
n
n
nφ (n)
4.4.5 Theorem. Prove that ∑ i =1
i=
2
.
( i , n ) =1
n
Proof. Let S = ∑ i =1
i = a1 + a2 + ........ + aφ ( n ) where 1 ≤ ai ≤ n and g.c.d. (ai , n) =
1
( i , n ) =1
112 Arithmetic Functions
∴ S = (n − a1 ) + (n − a2 ) + ........ + (n − aφ ( n ) )
= nφ (n) − S
nφ (n)
∴ 2 S = n φ ( n) ⇒ S = .
2
Proof. It is enough to prove (1) since (2) can be obtained by subtracting (1) from N.
Let ‘O’ be any one of these N objects. Then ‘O’ contributes 1 to the N of (1).
Let ‘O’ possesses exactly k of these properties. If k = 0, then ‘O’ does not contribute to any of the term
∑ Nα , ∑ Nαβ , ∑ Nαβγ ,........ and so on.
Therefore ‘O’ contributes 1 to (1) in this particular case.
Now let k ≥ 1 and finite. The ‘O’ contributes 1 to term N. ‘O’ contributes 1 to exactly k C1 of the terms in
∑ Nα .
‘O’ contributes 1 to exactly k C2 of the terms in ∑ Nαβ and so on.
So, the total contribution of ‘O’ to (1) is
1 − k C1 + k C2 − k C3 + ... =(1 − 1) k =0 . [Binomial expansion]
1 1 1
φ ( n) =n 1 − 1 − ........ 1 − .
p1 p2 pk
Analytical Number Theory 113
Proof. To find out φ (n) , by definition, we have to find the natural numbers less than equal to n which are
not divisible by any of p1 , p2 ,........, pk . Let Nα (i = 1, 2,......, k ) be the number of natural numbers less than
i
equal to n divisible by pi , (1 ≤ i ≤ k ) .
n
In fact, the natural numbers less than equal to n divisible by pi are pi , 2 pi ,........, pi .
pi
n
Similarly, Nα α = and so on.
i j
pi p j
∴ By general principle, the number of natural numbers less than equal to n and not divisible by any of
p1 , p2 ,........, pk is
n n
n −∑
φ ( n) = + ∑ − ........
pi pi p j
1 1
=n 1 − ∑ + ∑ − ........
pi pi p j
1 1 1
=n 1 − 1 − ....... 1 −
p1 p2 pk
k
1
= n ∏ 1 − .
i =1 pk
4.5.2.2 Example. The sum of squares of natural numbers which are less than equal to n and relatively
1 2 1
coprime to n is n φ (n) + n(1 − p1 ) (1 − p2 )........(1 − pk ) , n > 1 ; where p1 , p2 ,........, pk are the only distinct
3 6
divisors of n.
Solution. To find the required sum, we show that the sum of squares of numbers less than equal to n and
not coprime to n, that is, sum of squares of numbers less than equal to n which are divisible by atleast
one pi.
By general principle, this sum is
( pi p j ) 2 + (2 pi p j ) 2 + ....... + n pi p j
k
∑ 2 2
( pi ) + (2 pi ) + ....... + pi − ∑
+....
i 1 pi 1 ≤ i, j ≤ k
i≠j
pi p j
n
2
....+ (−1) k −1 2 2
( p1 p2 ....... pk ) + (2 p1 p2 ....... pk ) + ....... + p1 p2 ....... pk ......(2)
p1 p2 ....... pk
Now, let d be any divisor of n. Then the sum of squares of natural numbers less than equal to n and
divisible by d is
2
n n
2
1 nn 2n
= d2 . + 1 + 1
6 d d d
n3 n 2 nd
= + + . ......(3)
3d 2 6
n3 k 1 1 (−1) k −1 n 2 k k
k −1
∑ ∑ ∑ ∑
= − + ....... + + 1 − 1 + ....... + ( −1)
3 = i 1 pi 1 ≤ i , j ≤ k pi p j p1 p2 ....... pk 2 = i 1 1≤ i , j ≤ k
i≠ j i≠ j
n k ......(4)
∑ ∑ i j k −1
+ p − p p + ....... + ( −1) p p ....... p
6= i 1
i
1≤ i , j ≤ k
1 2 k
i≠ j
But the sum of squares of all natural numbers less than equal to n is
1 n3 n 2 n
n(n + 1) (2n + 1) = + + ......(5)
6 3 2 6
Now, (5) − (4) gives sum of squares of natural numbers less than equal to n and coprime to n.
2
n3 k
1 1 (−1) k + n 1 − 1 +
k
k
(5) − (4) = 1− ∑ + ∑ − ....... + ∑ 1≤ i∑ 1 − ...... + ( −1)
3 =i 1 pi 1≤ i , j ≤ k pi p j p1 p2 ....... pk 2 =i 1 , j≤k
i≠ j i≠ j
Analytical Number Theory 115
n k
+ 1 − ∑ pi + ∑ pi p j − ....... + (−1) p1 p2 ....... pk
k
6 =i 1 1≤ i , j ≤ k
i≠ j
n3 k 1 n2 n k
∏ 1 − +
3 i 1= pi
=
2
(1 − 1) k + ∏ (1 − pi )
6 i 1
=
1 2 n k
= n φ (n) + ∏ (1 − pi ) .
3 6 i =1
Hence proved.
4.5.2.3 Example. Find the sum of cubes of natural numbers less than equal to n and relatively coprime
to n, where n > 1.
Solution. Let x be any natural number less than equal to n and coprime to n. Also, we know that
g .c.d .(n, x) =
1 ⇔ g .c.d .(n − x, n) =
1.
Let S =∑=x3 ∑ (n − x) 3
= ∑ (n 3
− 3n 2 x + 3nx 2 − x 3 )
=∑ n3 − ∑ 3n2 x + ∑ 3nx 2 − ∑ x3
⇒ 2 S =∑ n3 − 3∑ n 2 x + 3∑ nx 2
=n3 ∑1 − 3n 2 ∑ x + 3n ∑ x 2
n 1 n
= n3φ (n) − 3n 2 φ (n) + 3n n 2φ (n) + Π (1 − pi )
2 3 6
n 1
= n3φ (n) − 3n 2 φ (n) + n3φ (n) + n 2 Π (1 − pi )
2 2
1 3 1
= n φ (n) + n 2 Π (1 − pi )
2 2
n3 n2
⇒ =S φ (n) + Π (1 − pi ) .
4 4
4.6 The order and the average order of the arithmetic functions
Here, we interpret the notations O, o and ∼. These symbols are defined as follows:
Let x be a continuous variable which tends to infinity or to zero or to some other limiting value and let
φ ( x) and f ( x) be two functions of x, then we have that
(i) f = O(φ ) means that | f | < Aφ for some constant A which is independent of x and works for all values
of x.
116 Arithmetic Functions
f
(ii) f = o(φ ) means that, →0 .
φ
f
(iii) f ∼ φ means that, → 1.
φ
x = O( x 2 ) , x = o( x 2 ) , x +1 ∼ x .
(b) As x → 0, we have
= x 2 O( x=
), x 2 o( x), sin x ∼ x, 1 + x ∼ 1
4.6.1 Result. (i) ‘o’ is stronger than ‘O’, that is, f = o(φ ) implies that f = O(φ ) .
(ii) So far, we have defined like this. f = O(1) , but we have not defined O(1) in isolation, so it is
convenient to make our notations more elastic. We say that O(φ) denotes an unspecified f such that
f = O(φ ) . We can, then, write, for example:
O(1) + O(1) =
O(1)
∑ O(1) = O(n)
v =1
meaning by this that the sum of n terms, each numerically less than a constant, is numerically less than a
constant multiple of n.
(iv) There is another phrase which it is convenient to define here. Suppose that P is a possible property
of a positive integer and P(x) denotes the number of numbers less than x which possess the property P.
Then, if P( x) ∼ x as x → ∞
i.e., if the number of numbers less than x which do not possess the property is o(x), then we say that
almost all numbers possess the property.
Now, we lead to find order of some arithmetic functions which we have studied earlier in this
unit like d(n) and φ (n) . Also, we shall use all the results related to these, proved earlier.
4.6.2 Theorem. Prove that d (n) = O(nδ ) for all positive δ. In other words, d (n) < nδ . A for some constant
A and such A exists for all δ but is independent of n.
r
Proof. We know that if n = p1a p2a ........ pra then=
1 2
d ( n)
r
∏
i =1
(ai + 1) .
Analytical Number Theory 117
d ( n) r
ai + 1 (ai + 1) ai + 1
⇒ =
nδ
∏
=
(p ) δ
i =1
∏ ai
1 piai δ
∏ 1 piaiδ
......(1)
i
pi < 2δ pi ≥ 2δ
1
Now, for p ≥ 2δ i.e., pδ ≥ 2 , we have
a +1 a +1 a +1
= ≤ a ≤ 1 for all a
p aδ ( pδ ) a 2
a +1 1 a a
Thus, we have. = + aδ ≤ 1 +
p aδ p aδ p aδ log 2
1 1
1+
= ≤ exp.
δ log 2 δ log 2
1
2δ
< exp = A (say)
δ log 2
4.6.3 Definition. Let f (n) and g (n) be two arithmetic functions of n, then we say that f (n) is of the
average order of g (n) if
f (1) + f (2) + f (3) + ........ + f (n) ∼ g (1) + g (2) + ........ + g (n)
4.6.4 Results. (i) A point (x, y) is said to be a Lattice point if x and y are integers.
1 1 1 1
(ii) Sterling formula. 1 + + + ...... + = log n + γ + O where γ is called sterling constant.
2 3 n n
n +1 n +1
= ∫ 1
log
= t dt t log t − t 1
(n + 1) log (n + 1) − n
= (n + 1) log(n + 1) − n ∼ n log n nlim = 1
→∞ n log n
Now, consider the lattice, whose vertices (x, y) are the points in xy-
Y
plane with integral coordinates. Denote by D, the region in the
upper right-hand corner contained between the rectangular axes xy = n → rectangular
and the rectangular hyperbola, xy = n leaving out the coordinate
axes. D
n
Secondly, the number of lattice points in this region with x coordinate equal to 1 will be , the
1
n
number of lattice points in this region with x-coordinate equal to 2 will be and so on.
2
Hence, the total number of lattice points in this region will be integral part of
n n n
1 + 2 + ........ + n .
n n
∴ d (1) + d (2) + ........ + d (n) =[ n ] + + ........ +
2
n
n n n
=n + + O(1) + + O(1) + ........ + + O(1)
2 3 n
1 1 1
= n 1 + + + ........ + + O(n)
2 3 n
1
= n log n + γ + O + O (n) [Sterling formula]
n
Analytical Number Theory 119
= n log n + nγ + O(1) + O(n)
= n log n + O(n)
∼ n log n
This implies that d (1) + d (2) + ........ + d (n) ∼ n log n ∼ log1 + log 2 + ...... + log n .
So, by definition of average order, d(n) is of average order of log n.
This completes the proof.
4.6.6 Result. We know that σ(n) denotes the sum of divisors of n. We, then, give this result.
n n
=i 1=x 1
∑ σ (i) = ∑ ∑ n
y. ......(1)
1≤ y ≤
x
= (1) + (1 + 2) + (1 + 3) + (1 + 2 + 4) + (1 + 5) + (1 + 2 + 3 + 6) .
6
Now, R.H.S. = ∑ ∑
x =1 6
y
1≤ y≤
x
= ∑6
y+ ∑ 6
y+ ∑ 6
y+ ∑ 6
y+ ∑ 6
y+ ∑ 6
y
1≤ y≤ 1≤ y≤ 1≤ y≤ 1≤ y≤ 1≤ y≤ 1≤ y≤
1 2 3 4 5 6
= (1 + 2 + 3 + 4 + 5 + 6) + (1 + 2 + 3) + (1 + 2) + 1 + 1 + 1
=1 + (1 + 2) + (1 + 3) + (1 + 2 + 4) + (1 + 5) + (1 + 2 + 3 + 6) .
1 2
4.6.7 Theorem. The average order of σ(n) is π n . More precisely.
6
1 2 2
σ (1) + σ (2) + ........ + σ (n=
) π n + O(n log n) .
12
Proof. We have.
n n
=i 1=x 1
∑ σ (i) = ∑ ∑ n
y
1≤ y ≤
x
1 n
n n n(n + 1)
=
2
∑ x + 1 Sum of n natural numbers = 2
x =1 x
1 n n n
= ∑ + O(1) + O(1)
2 x =1 x 2
120 Arithmetic Functions
1 n
n2 n
=
2
∑ 2 + O + O(1)
x
x =1 x
1 2 n 1 n 1
n ∑ 2 + O n ∑ + O ( n) .
=
=2 x
x 1= x 1 n
n
1 ∞
1 1
But we have
=
2
x 1=
∑
= ∑
x 1 x x2
+ O
n
π2 1
= + O
6 n
n
1 1
and ∑=x x =1
log n + O .
n
n
1 2 π2 1 1
Hence, (1) becomes. ∑ σ (i) = n
2 6
+ O + O n log n + O
n n
+ 0(n)
i =1
1 2 2
= π n + O(n) + O(n log n) + O(n)
12
1 2 2
= π n + O ( n log n )
12
σ ( n) φ ( n)
4.6.9 Theorem. There exists a constant A s.t. A < < 1 for all n > 1 .
n2
pα + 1 − 1 pα +1 (1 − p − α −1 )
=σ ( n) ∏
=
p/n p −1
∏
p/n p (1 − p −1 )
pα (1 − p − α −1 ) 1 − p − α −1
= ∏
=
p/n (1 − p −1 )
n ∏
p / n 1− p
−1
.
∴ ) φ (n) n 2 ∏ (1 − p − α −1 )
σ (n=
p/n
σ ( n) φ ( n)
⇒ =∏ (1 − p − α −1 ) < 1 .
n2 p/n
1 1
≥ ∏ 1 − 2 ≥ ∏ 1 − 2
p/n p p p
∞
1
≥ ∏ 1 − 2 .
k =1 k
1 ∞
1
But we know that the infinite series ∑ k2
converges, so by above result, the product ∏ 1 − k
k =1
2
is
convergent.
σ ( n) φ ( n)
So, there exists a constant A such that A < < 1.
n2
6n
4.6.10 Theorem. Prove that the average order of φ (n) is . In fact,
π2
3n 2
Φ (n) = φ (1) + φ (2) + ........ + φ (n) = + O ( n(log n) ) .
π2
µ (d ) n
Proof. We have= ∑
φ (n) n= ∑=
µ (d )
d /n d d /n d
∑
dd ′ = n
d ′ µ (d )
n
so that, we obtain Φ (n) = φ (1) + φ (2) + ........ + φ (n) = ∑ φ ( m)
m =1
n
=
=
∑
= ∑ d ′ µ (d ) ∑
dd ′ m
m 1= dd ′ ≤ n
d ′ µ (d )
[n / d ] n n
n
1 n
= ∑
= µ (d ) ∑ d ′
2
∑ µ (d ) + 1
d d
=d 1=d′ 1 = d 1
1 n n2 n
= ∑
2 d =1
µ (d ) 2 + O
d
d
n 2 n µ (d ) ∞ 1
= ∑ d2
2 d 1=
+ O n∑
= d 1 d
n 2 ∞ µ (d ) 1
= ∑ 2
+ O + O ( n log n )
2 d =1 d n
n2 ∞
µ (d )
=
2
∑
d =1 d2
+ O(n) + O ( n log n ) .
∞
µ (d ) 6 3n 2
Now, we use the result that ∑
d =1 d2
converges to
π2
so that we obtain Φ ( n ) =
π2
+ O ( n log n )
3n 2
4.6.11 Theorem. The number of terms in the Farey Series of order n is approximately .
π2
3n 2
Thus, we get that the number of terms in Farey series of order n is approximately for large n.
π2
6
4.6.12 Theorem. The probability that the two given integers should be coprime to each other is .
π2
p p
Consider the fraction . The total number of fractions with all such that 1 ≤ p ≤ q ≤ n is
q q
n
1 1
∑ =i
i =1 2
n(n + 1) ∼ n 2 .
2
p
On the other hand, we have proved that number of fractions where 1 ≤ p ≤ q ≤ n and g .c.d .( p, q ) =
1
q
3n 2 3n 2 / π 2 6
is for large n. Thus, required probability = = 2 .
π2 1 2 π
n
2
4.6.13 Example. If p is a prime such that p and p + 2 are both primes then φ ( p + 2)= φ ( p) + 2 .
= ( p − 1) + 2
= φ ( p) + 2 .
4.6.14 Example. Let p and 2p +1 be both odd primes (For example, 3, 7) and let n = 4p, then
φ (n + 2)= φ (n) + 2 .
= φ ( 2(2 p + 1) )
= φ (2) φ (2 p + 1)
⇒ φ (n + 2)= φ (2 p + 1)
Analytical Number Theory 123
= 2p. [ 2 p + 1 is prime ]
Again, φ (n) +=
2 φ (4 p ) + 2
= φ (4) φ ( p ) + 2 [ (4, p) = 1 ]
= 2( p − 1) + 2 [ p is prime]
= 2p
This follows that φ (n + 2)= φ (n) + 2 .
4.6.15 Definition. A natural number is said to be of Euclid type or a Euclid number if it is of the form
2k −1 (2k − 1) where 2k − 1 is a prime and k > 1 is an integer.
k
1 1 1 1
Solution. (i) By definition Pk ( s)= ∏ 1 + s + 2 s + ........ =
pi pi
∑ s
n
, that is, Pk ( s) = ∑ ns
i =1 n ∈ Xk n∈ X k
X k {=
where = n : n p1e p2e ........ pke
1 2 k
} be the set of those integers which are written as the product of
p1 , p2 , p3 ,........, pk and their powers where pk is the largest prime in our consideration.
1 1 1
Now, Pk ( s )
= ∑
n ∈ Xk
≥
n s n ≤ pk
∑ ns
1 1
Then trivially,=
Pk (1) ∑
n ∈ Xk n
≥ ∑
n ≤ pk n
......(1)
1 1 ∞
1
Consider the summation ∑
n ≤ pk n
then klim
→∞
∑ =
n ≤ pk n
∑ 1 n
n=
− ζ (1)
1
⇒ lim
k →∞
∑
n ≤ pk n
→∞
(ii) Now, let n be any positive integer, then n = p1e p2e ........ pke where ei ≥ 0 and p1 , p2 ,...... pk are arbitrary
1 2 k
distinct primes.
k
1
Then φ ( n) n ∏ 1 −
=
i =1 pi
φ ( n) k
1 k
1 − = ∏ ( (1 − pi ) )
−1 −1 −1
n
= ∏ pi i 1
=i
1=
−1
k −1 k 1
= ∏ ( (1− pi−1 )−1 ) = ∏ 1 −
pi−1
i =1 i =1
1
= ( Pk (1))
= −1
→ 0 as k → ∞
Pk (1)
φ ( n)
⇒ → 0 as k → ∞ (arbitrary small).
n
φ ( n)
Thus for a given ε > 0, we can find n such that <ε.
n
4.6.19 Theorem. (Statement only) Let f, g, h be arithmetic functions such that h= f ∗ g , that is,
h( n) = ∑
de = n
f ( d ) g (e) . Prove that, H ( s) = F (r ) G ( s) , where
∞
f ( n) ∞
g ( n) ∞
h( n)
=F (s)
=
s
n 1=
s
n 1= n 1
∑=
n
, G ( s) ∑
=
n
, H ( s) ∑
ns
be the Dirichlet series of f (n) , g (n), h(n) . Also, it is given that, F ( s) and G ( s) both converge absolutely.
∞
φ ( n) ζ ( s − 1)
4.6.20 Example. Prove that ∑=
n
n =1
s
ζ (s)
∀ s>2.
Solution.=
Let f φ=
and g u where u (n) = 1 ∀ n ∈ N then
∞
g ( n) ∞
u ( n) ∞
1
G (s)
=
=
s
n 1=
s
n 1= n 1
∑=
n
∑=
n
∑
=
n s
ζ (s)
∞
φ ( n)
⇒ ∑n =1 ns
.ζ (=
s ) ζ ( s −1)
∞
φ (n) ζ ( s − 1)
⇒ ∑n =1 ns
=
ζ (s)
for all s > 2.
∞
τ ( n)
that ∑ s [ζ ( s)] for all s > 1 .
4.6.21 Example. Show=
2
n =1 n
∞
τ (k )
∑ = [ζ ( s ) ] .
2
⇒
k =1 ks
∞
σ k ( n)
4.6.22 Example. Express ∑
n =1 ns
in terms of Riemann Zeta function where σ k (n) = ∑ d k .
d /n
n
We note that ∑
d /n
N k (d ) . u =
d
∑
d /n
N k (d ) . 1
∑ [ N (d )] ∑
k
= = dk
d /n d /n
= σ k ( n) for any n.
126 Arithmetic Functions
⇒ σk
Nk ∗ u = ......(1)
[ N=
( n) ]
k
∞
N k ( n) ∞ ∞
nk ∞
1
Now F ( s)
=
=n 1
∑ = s
n=n 1
∑ s
n=n 1=
∑
=
n s ∑
n 1 n
s−k
= ζ (s − k )
Now, above theorem is valid for s > max(k + 1, 1) so by above theorem with N k ∗ u =σ k , we have
∞
N k ( n) ∞ u ( n) ∞
( N k ∗ u ) (k ) ∞
σ k (k )
F (s) G (s)
=
=n 1
∑ =
=
n s
.∑ s
n
n 1=
∑=
k 1 k =k 1 k s
s ∑
∞
N k ( n) ∞ u ( n) ∞
σ (k )
⇒
=n 1
∑ = n s
.∑
n
n 1=
s
=∑ ks
k 1 k
∞
σ k (k )
⇒ ∑
ζ (s − k ) ζ (s) =
k =1 ks
∞
σ k (k )
⇒ ∑
k =1 ks
ζ ( s − k ) ζ ( s ) for s > max(k + 1, 1) .
=