PMATH 740: ANALYTIC NUMBER THEORY

HEESUNG YANG

1. September 14: some basics

For the sake of completeness, we are going to review some basics first.
Definition 1. For n ∈ N = {1, 2, 3, . . . }, let pn be the n-th prime. That is, p1 = 2, p2 =
3, . . . .
Theorem 1 (Euclid). There are infinitely many primes.
Proof. Suppose that there are only finitely many primes, say p1 , . . . , pn . Consider
m := p1 p2 . . . pn + 1 ≥ 2.
By the fundamental theorem of arithmetic, m is a product of primes. Thus pk | m for
some k ∈ N with 1 ≤ k ≤ n. Then pk | (m − p1 p2 . . . pn ) = 1. Then pk | 1, which is a
contradiction. 
Remark 1. There are many different ways to prove Theorem 1 – for instance, using topology.
See Furstenberg’s work (doi:10.2307/2307043).
Definition 2. For x ∈ R, let
π(x) := #{p ≤ x : p prime}.
By Theorem 1, we have π(x) → ∞ as x → ∞. Our goal is understand how “large” π(x)
is.
n
Proposition 2. For n ∈ N, we have pn ≤ 22 .
1
Proof. We prove this result by induction. For k = 1, clearly p1 = 2 ≤ 22 = 4. Suppose that
the result holds for 1 ≤ k ≤ n. We have seen in the proof of Theorem 1 that
pn+1 ≤ p1 p2 · pn + 1.
Thus by the induction hypothesis, it follows
1 2 n n+1 −2 n+1
pn+1 ≤ 22 22 · · · 22 + 1 = 22 + 1 ≤ 22 .
By induction, the result follows. 
Corollary 3. For x ∈ R with x ≥ 2, we have π(x) ≥ log log x.

Date: 2 December 2015.

1
First proof. Clearly the result holds for 2 ≤ x < 4. For x ≥ 4, let s ∈ N satisfy
5 6
22 ≤ x ≤ 22 .
s
By Proposition 2, we have ps ≤ 22 ≤ x. Thus π(x) ≥ s. By taking logarithms twice, we see
that
s+1
x < 22
⇒ log x < 2s+1 log 2
 
log x
log log 2
⇒ < s + 1.
log 2
It follows that  
log x
log log 2
π(x) ≥ s > − 1 > log log x.
log 2
The last inequality is left as an exercise (Hint: log 2 < 1 and calculus). 
Second proof. Note that for all primes p, we have
 −1
p−1 1 1
≥ ⇔ 1− ≤ 2.
p 2 p
Thus for x ≥ 2, we have
Y −1 Y   X Z x
π(x) 1 1 1 1 du
2 ≥ 1− = 1 + + 2 + ··· ≥ ≥ log x.
p≤x
p p≤x
p p n≤x
n 1 u

Thus
log log x
π(x) ≥ ≥ log log x,
log 2
as required. 
n
Conjecture (Fermat; proved to be false). The numbers of the form 22 + 1 for n ∈ N ∪ {0}
are primes.
n
Definition 3. The numbers of the form Fn := 22 + 1 are the Fermat numbers.
Remark 2. Fermat’s conjecture is false. Indeed, Fn is a prime for n = 0, 1, 2, 3, 4. However,
F5 is not, since 641 | F5 (proved by Euler in 1732). Also, it is known that F6 , . . . , F21 are
composite.
Claim 1. Fn | (Fm − 2).
n+k n
Proof. We have Fm − 2 = 22 − 1. Write x = 22 . Then
k
Fm − 2 x2 − 1 k k
= = x2 −1 − x2 −2 + · · · − 1 ∈ N.
Fn x+1
The claim now follows. 
Theorem 4 (Pólya). For m, n ∈ N with n < m, we have (Fn , Fm ) = 1.
2
Proof. Let m = n + k for some k ∈ N, and let d = (Fn , Fm ). Since d | Fn and d | Fm , by
Claim 1 we have d | (Fm − 2). Thus d | 2 so d is either 1 or 2. But since 2 - Fn , indeed d = 1,
as desired. 
Remark 3. From Theorem 4, we obtain another proof of Theorem 1 and Proposition 2.

2. September 16
Theorem 5. For x ≥ 2, we have
log x
π(x) ≥ .
2 log 2
Also, for n ≥ 1, we have pn ≤ 4n .
Proof. Let 2 = p1 , p2 , . . . , pj be the primes less than or equal to x. For n ∈ N with n ≥ x,
e
write n = n21 m with n1 ∈ N and m square-free. That is, m = pe11 · · · pj j such that ei ∈ {0, 1}
j
√
for each i = 1, 2, . . . , j. Thus, there are at most
√ 2 possible values for m. and at most x
possible values for n. Thus it follows that 2j x ≥ x, or
√
2j ≥ x. (1)
But then recall that j = π(x), so indeed we have (from (1)) that π(x) log 2 ≥ log x/2. Hence
the first claim follows.
√
Also, take x = pn . Then j = n = π(pn ). So again from (1), we have 2n ≥ pn , or
equivalently 4n ≥ pn . 
In 1896, Hadamard and de la Vallée Poussin proved independently the prime number
theorem. More precisely, they showed that (conjectured by Gauss),
π(x)
lim x = 1.
x→∞
log x

Let n ∈ N and p a prime. Then the exact power dividing n! is

X n  log p c 
bX
log n

n
= .
k≥1
pk k=1
pk

Claim 2. We have Y
pr p ≡ 0 (mod N ),
p≤2n
2n


where N = n
, where rp ∈ N ∪ {0} satisfies prp ≤ 2n < prp +1 .
Proof. Note that the exact power of p dividing (2n)! is
rp  
X 2n
k
,
k=1
p
and the exact power of p dividing n is
rp  
X n
.
k=1
pk
3
 2n


Thus the exact power of p dividing n
is
rp    
X 2n n
k
−2 k ≤ rp ,
k=1
p p
since    
2n n
− 2 ≤ 1.
pk pk
The claim now follows. 
Theorem 6. For x ≥ 2, we have
 
3 log 2 x x
< π(x) < (6 log 2) .
8 log x log x
Proof (Erdős). Consider first a lower bound for π(x). Note that the binomial coefficient
 
2n (2n)!
= ∈ N.
n (n!)2
From Claim 2, we have  
2n Y
≤ prp ≤ (2n)π(2n) .
n p≤2n
Note that
 
2n (2n)(2n − 1) · · · (n + 2)(n + 1) n+1 n+2
= = · · · · · · 2 ≥ 2n .
n 1 · 2 · ··· · n 1 2
By the above two inequalities,
2n ≤ (2n)π(2n) .
Hence,  
n log 2 log 2 2n
π(2n) ≥ = .
log(2n) 2 log(2n)
Note that x(log x)−1 is increasing for x ≥ e. For x ≥ 6, let n ∈ N satisfy 43 x < 2n ≤ x. Then
     3
log 2 2n log 2 4
x 3 log 2 x
π(x) ≥ π(2n) ≥ ≥ 3

≥ · .
2 log(2n) 2 log 4 x 8 log x
Also, we can check that the lower bound holds for 2 ≤ x ≤ 6. We now consider an upper
bound. Note that
 
Y 2n
p .
n<p≤2n
n
p < (1 + 1)2n = 22n . On the other hand, we have the following lower bound:
Q
Therefore
n<p≤2n
Y
p ≥ nπ(2n)−π(n) .
n<p≤2n

It follows that
nπ(2n)−π(n) ≤ 22n .
4
Thus,
π(2n) log n − π(n) log n < (log 2)2n.
In other words,
n
π(2n) log n − π(n) log < (log 2)2n + (log 2)π(n) < (3 log 2)n.
2
Take n = 2k . Then
π(2k+1 ) log 2k − π(2k ) log 2k−1 < (3 log 2)2k
π(2k ) log 2k−1 − π(2k−1 ) log 2k−2 < (3 log 2)2k−1
..
.
π(8) log 4 − π(4) log 2 < (3 log 2)4,
so upon adding those inequalities we see that
π(2k+1 ) log 2k < (3 log 2)(2k + 2k−1 + · · · + 4) + 2 log 2 < (3 log 2)2k+1 .
Thus
2k+1
 
k+1
π(2 ) < (3 log 2) .
log 2k
So for x ≥ 2, let k ∈ N with 2k < x ≤ 2k+1 . Then π(x) ≤ π(2k+1 ). Hence for x ≥ e,
2k+1
 k 
2 x
π(x) ≤ (3 log 2) k
≤ (6 log 2) k
≤ (6 log 2) .
log 2 log 2 log x
Also, we can check that the upper bound holds for 2 ≤ x ≤ e. 
3. September 18
In 1845, Bertrand showed that there is always a prime p in the interval [n, 2n] for n ∈ N
provided that n < 6 · 106 . He conjectured that this is always holds. Chebyshev proved this
in 1850.
p ≤ 4n .
Q
Proposition 7. For n ∈ N, we have
p≤n

Proof. We prove this result by induction. The claim holds when n = 1 and n = 2. Suppose
that the result holds for 1 ≤ k ≤ n − 1. Since for n > 2, if n is even then
Y Y
p= p.
p≤n p≤n−1

Thus we can consider only when n is odd. Write n = 2m + 1 and consider 2m+1


m
. We have
 
Y 2m + 1
p .
m+1<p≤2m+1
m

Note that 2m+1 and 2m+1 occur in the binomial expansion of (1 + 1)2m+1 and 2m+1




m m+1 m
=
2m+1

2m+1

22m+1 m
m+1
. Thus m ≤ 2 = 4 . By our induction hypothesis, we see
! !
Y Y Y
p= p p ≤ 4m+1 · 4m = 42m+1 ,
p≤2m+1 p≤m+1 m+1<p≤2m+1
5
as required. 

For α ∈ N ∪ {0}, we write pα k b to mean that pα | b but pα+1 - b.

2n
Proposition 8. If n ≥ 3 and p is a prime with 32 n < p ≤ n then p -


n
.

Proof. Since n ≥ 3, if p satisfies 2n

3
< p ≤ n, then p > 2. Thus p and 2p are the only
multiples of p with p ≤ 2n and so p k (2n)!. Since 2n
2
3
< p ≤ n, we have p k n!. Hence
2 2 2n

(2n)!
p k (n!) . The result follows upon noting that n = n!n! . 

4. September 21: Chebyshev’s theorem

Theorem 9 (Chebyshev). For n ∈ N there exists a prime p with n < p ≤ 2n.
Proof (Erdős). Note that the result holds for n = 1, 2, 3. Suppose that the result

is false
for some n ∈ N with n ≥ 4. Let p be a prime dividing 2n 2n

αp
n
and p k n
. By our
2
assumption, p ≤ n. Also, by 8, we have p ≤ 3 n. Let rp be defined in the proof of Theorem
6, i.e., prp ≤ 2n < prp +1 . We have seen in the √proof of Theorem 6 that αp ≤ rp . Thus
pαp ≤ prp ≤ 2n. If αp ≥ 2, then p2 ≤ 2n, i.e., p ≤ 2n. By Proposition 7, we have
  
√ √
  Y
2n  Y  2 2
≤ p  pαp  ≤ 4 3 n (2n)π( 2n) ≤ 4 3 n (2n) 2n .
  
n  2  2 
p≤ 3 n p≤ 3 n
αp ≤1 αp ≥2

2n


Note that n
is the largest (2n + 1) terms in the binomial expression of
     
2n 2n 2n 2n
(1 + 1) = + + ··· + .
0 1 2n
Therefore,
22n
 
2n
≥ .
n 2n + 1
Combining the above inequalities, we have
4n 2
√
≤ 4 3 n (2n) 2n ,
2n + 1
so we have
n
√ √
2n 2n+2
4 3 ≤ (2n) (2n + 1) ≤ (2n) .
Taking logarithms, we find that
n √
log 4 < ( 2n + 2) log(2n).
3
By calculus (exercise!) one can show that the above inequality is false for n ≥ 512. This
implies that the statement of the theorem holds for n ≥ 512. By checking all cases for
n < 512, we see that the result holds for all n ∈ N. 
6
 5. September 21: Möbius function and von Mangoldt function
Notation: let f and g be functions from N or R+ to R, and suppose that g maps to R+ .
Definition 4. The big-O notation f = O(g) means that there exist c, C ∈ R+ such that for
x > c we have |f (x)| ≤ Cg(x). The little-O notation f = o(g) means that
f (x)
lim = 0.
x→∞ g(x)

Finally, f (x) ∼ g(x) (i.e., “f is asymptotic to g”) means

f (x)
lim = 1.
x→∞ g(x)
R∞
Example 1. 20x = O(x), sin x = O(1), x u12 du = O(x−1 ), x = o(x2 ), sin(x) = o(log x), (logxx)2 =
  √
o logx x ; x + 1 ∼ x, x + x ∼ x. Also, recall the prime number theorem:
x
π(x) ∼ .
log x
Definition 5. The Möbius function µ is defined by
(
(−1)r if n is a product of r distinct primes and is square-free
µ(n) =
0 otherwise.

Example 2. 12 is not square-free, so µ(12) = 0. On the other hand, µ(15) = µ(3 · 5) =

(−1)2 = 1 and µ(30) = µ(2)µ(3)µ(5) = (−1)3 = −1.
Definition 6. Let n = pα1 1 · · · pαr r be the unique factorization of n into distinct prime powers.
Then N := p1 p2 · · · pr is called the radical of n.
(
P 1 if n = 1
Proposition 10. µ(d) = .
d|n 0 otherwise

Proof. If n = 1, then the result is immediate from the definition of µ(n). If n > 1, let
n = pα1 1 · · · pαr r be the unique factorization of n into distinct prime powers, and let N be the
radical of n. Since µ(d) = 0 unless d is square-free, we have
X X
µ(d) = µ(d).
d|n d|N

Note that the divisors of N are in one-to-one correspondence with the subsets of {p1
, . . . , pr }.
Thus the latter sum contains 2r summands. The number of k-element subsets is kr and the
corresponding divisor d of such a set satisfied µ(d) = (−1)k . Therefore
r  
k r
X X
µ(d) = (−1) = (1 − 1)r = 0,
k=0
k
d|N

so the claim follows. 

7
 6. September 23
Remark 4. Some analogous relations between Z and Fq [t]:
Z Fq [t]
units {±1} Fq
norm |a| =absolute value |f | = q deg f
every positive integer > 1 every monic polynomial of degree ≥ 1
unique factorization
is a product of primes is a product of monic irreducible polynomials
X Y
T deg f = (1 + T deg v + T 2 deg v + · · · )
f ∈Fq [t] v∈Fq [t]
f monic v monic irred
Y
= (1 − T deg v )−1
v∈Fq [t]
v monic irred
Y∞
−d −Nd
= (1 − T ) .
d=1
P
Proposition 11 (Möbius inversion formula). f (n) = (1 ∗ g)(n) = g(d) if and only if
d|n
P
g(n) = (µ ∗ f )(n) = µ(d)f (n/d).
d|n
P
Proof. (⇐) Suppose that g(n) = µ(d)f (n/d). Then we have
d|n
 
X XX d
g(d) = µ(e)f
e
d|n d|n e|d
 
X d
= µ(e)f (s) let s =
est=n
e
X X
= f (s) µ(e) .
s|n e| n
s
| {z }
(∗)

The claim follows upon P

noting that (∗) is 1 if n = s and 0 otherwise.
(⇒) Suppose f (n) = g(d). Then we have
d|n
X n X X
µ(d)f = µ(d) g(e)
d
d|n d|n e| n
d
X X X
= µ(d)g(e) = g(e) µ(d) = g(n). 
des=n e|n d| n
e

Definition 7. For n ∈ N, the von Mangoldt function, denoted by Λ(n) is defined by

(
log p if n = pk for some k ∈ N
Λ(n) = .
0 otherwise
8
Also for x ∈ R, we define
!
X Y
θ(x) = log p = log p
p≤x p≤x
X X
ψ(x) = log p = Λ(n).
pk ≤x n≤x
for some k ∈ N

Note that
X  log x 
ψ(x) = log p.
p≤x
log p
Also p2 ≤ x is equivalent to p ≤ x1/2 ; similarly, p3 ≤ x is equivalent to p ≤ x1/3 and so
forth. Thus we have ψ(x) = θ(x) + θ(x1/2 ) + θ(x1/3 ) + · · · . Since 2m ≤ pm ≤ x we see that
θ(x1/m ) = 0 provided that m > log x
log 2
. Thus
bX
log 2 c
log x

ψ(x) = θ(x1/k ).
k=1

Since X
θ(x) = log p ≤ x log x,
p≤x
we see that
bX
log 2 c
log x
bX
log 2 c
log x
bX
log 2 c
log x

1
θ(x1/k ) ≤ x1/k log(x1/k ) ≤ x1/2 log x = O(x1/2 (log x)2 ).
k=2 k=2 k=2
k
Therefore
ψ(x) = θ(x) + O(x1/2 (log x)2 ). (2)
But what we don’t know at this point is whether θ(x) is the dominant term, which is what
we want. We have seen in Theorem 6 that π(x) ≤ c1 logx x for some c1 > 0. Thus
X
θ(x) = log p ≤ π(x) log x ≤ c1 x.
p≤x

Combine this with (2), we see that ψ(x) ≤ c2 x for some c2 > 0. Also, we have seen in the
proof of Theorem 6 that     Y
n 2n 2n
2 ≤ and pr p
n n p≤2n
with prp ≤ 2n < prp +1 . It follows that
 X  
n 2n log 2n
n log 2 = log(2 ) ≤ log ≤ log p ≤ ψ(2n).
n p≤2n
log p
For x ≥ 2, let n ∈ N with 2n ≤ x < 2n + 2. Then we have
x−2
ψ(x) ≥ ψ(2n) ≥ n log 2 > log 2.
2
9
Thus there exists c3 > 0 such that ψ(x) > c3 x. Again, combine this with (2), we see that
θ(x) > c4 x for some c4 > 0.

7. September 25
θ(x) ψ(x)
Theorem 12. π(x) ∼ log x
∼ log x
.
x
Remark 5. By Theorem 12, to prove that π(x) ∼ log x
it suffices to show that π(x) ∼ θ(x) ∼ x.

Proof of Theorem 12. We have seen that ψ(x) = θ(x) + O(x1/2 (log x)2 ). Since θ(x) > c4 x it
follows that
θ(x) ψ(x)
∼ .
log x log x
θ(x)
Thus it suffices to show that π(x) ∼ .
Note that
log x
X
θ(x) = log p‘π(x) log x.
p≤x

Thus
θ(x)
π(x) ≥ ,
log x
that is,
π(x)
θ(x)
≥ 1.
log x
Note that for any δ > 0, we have
X X
θ(x) = log p ≥ log(x1−δ ) 1 ≥ (1 − δ) log x(π(x) − π(x1−θ )).
p≤x x1−δ ≤p≤x

Thus
θ(x) + (1 − δ) log x(x1−θ ) ≥ (1 − δ)(log x)πx
θ(x)
+ x1−δ ≥ π(x)
(1 − δ) log x
1 x1−δ log x π(x) log x
+ ≥ .
1−δ θ(x) θ(x)
1
Given any ε > 0, we can choose δ > 0 so that 1−δ
< 1 + 2ε . Since θ(x) > c4 x for some c4 > 0
x1−δ log x
there exists x0 ∈ R such that for x > x0 , θ(x)
< 2ε . Thus for x > x0 , we have
π(x) log x
< 1 + ε.
θ(x)
Since
π(x) log x
1≤ < 1 + ε,
θ(x)
by choosing ε to be arbitrarily close to 0 the result follows. 
The following summation formula by Abel is useful:
10
Proposition 13 (Abel’s summation formula). Let {an }∞ n=1 be a sequence of complex num-
bers. Let f be a function from {x ∈ R : x ≥ 1} to C. For x ∈ R, we write
X
A(x) := an .
n≤x

If f has a continuous first derivative for x ≥ 1, then

X Z x
an f (n) = A(x)f (x) − A(u)f 0 (u) du.
n≤x 1

Proof. Let N = bxc. Then

X
an f (n) = A(1)f (1) + (A(2) − A(1))f (2) + · · · + (A(N ) − A(N − 1))f (N )
n≤N

= A(1)(f (1) − f (2)) + · · · + A(N − 1)(f (N − 1) − f (N )) + A(N )f (N ).

Note that for i ∈ N and u ∈ R with i ≤ u < i + 1, we have A(u) = A(i). Thus
Z i+1
A(i)(f (i) − f (i + 1)) = A(u)f 0 (u) du.
i
It follows that
X Z N
an f (n) = − A(u)f 0 (u) du + A(N )f (N ). (3)
n≤N 1

Also, for x ≥ u ≥ N , we have A(u) = A(N ). Thus

Z x
A(u)f 0 (u) du − A(x)(f (x) − f (N )) = A(x)f (x) − A(N )f (N ).
N
or equivalently Z x
0 = A(x)f (x) − A(N )f (N ) − A(u)f 0 (u) du. (4)
N
Combine (3) and (4), then we get
X X Z x
an f (n) = an f (n) = A(x)f (x) − A(u)f 0 (u) du,
n≤x n≤N 1

as required. 
Definition 8. Given x ∈ R, we denote {x} the fractional part of x. That is, {x} := x − bxc.
The Euler-Mascheroni constant (or Euler’s constant) γ is
Z ∞
{u}
γ =1− du (= 0.57721 . . . ).
1 u2
= log x + γ + O(x−1 ).
P 1
Theorem 14. n
n≤x

Proof. Take an = 1 and f (u) = u−1 . Then

X
A(x) = 1 = bxc .
n≤x
11
By Abel’s summation formula,
Z x
X1 bxc buc
= + du
n≤x
n x 1 u2
Z x
x − bxc u − {u}
= + du
x 1 u2
Z x Z x
−1 1 buc
= 1 + O(x ) + du − du
1 u 1 u2
Z ∞ Z ∞ 
−1 buc buc
= 1 + O(x ) + log x − du − du
1 u2 x u2
Z ∞
−1 buc
= log x + γ + O(x ) + du
x u2
Z ∞
−1 1
≤ log x + γ + O(x ) + du.
x u2
R∞
Note that x
1
u2
du ≤ 1
x
so the integrand is indeed O(x−1 ). The result follows. 

8. September 30
Let’s recall the Abel summation formula:
X Z x
an f (n) = A(x)f (x) − A(u)f 0 (u) du
n≤x 1

P
where A(x) = an .
n≤x

Theorem 15. We have

X Λ(n)
= log x + O(1).
n≤x
n

Proof. Let an = 1 and f (n) = log n. By Abel’s summation, we have

x
b(c u)
X Z
log n = bxc log x − du
n≤x 1 u
Z x
u − {u}
= (x − {x}) log x − du
1 u
Z x
{u}
= x log x + O(log x) − (x − 1) + du
1 u
= x log x − x + O(log x).
12
Furthermore,
∞   !
X X X x
log n = log(bxc!) = log p
n≤x p≤x k=1
pk
Xx X jxk
= log p = Λ(n)
pk n≤x
n
pk ≤x
!
X x n x o X Λ(n) X
= − Λ(n) = x +O Λ(n) .
n≤x
n n n≤x
n n≤x
P
Since Λ(n) = ψ(x) = O(x), it follows that
n≤x
X X Λ(n)
log n = x + O(x).
n≤x n≤x
n
so
X Λ(n)
x = x log x + O(x),
n≤x
n
P Λ(n)
or n
= log x + O(1), as desired. 
n≤x
X log p
Theorem 16. = log x + O(1).
p≤x
p

Proof. By Theorem 15, we have

X log p X Λ(n) X X log p
= −
p≤x
p n≤x
n m≥2 pm ≤x
pm
X X log p
= log x + O(1) − .
m≥2 pm ≤x
pm
Note that
X X log p X  1 1

m
≤ 2
+ 3 + · · · log p
m
m≥2 p ≤x
p p
p p
∞
log p X log n
X
= ≤ = O(1).
p
p(p − 1) n=2 n(n − 1)
P log p
Combining the above two inequalities, it follows p
= log x + O(1). 
p≤x

13
Theorem 17. There exists β ∈ R such that
X1
= log log x + β + O((log x)−1 ).
p≤x
p

Proof. Let (
log p
p
if n = p is a prime
an =
0 otherwise.
Write X
A(x) = an .
n≤x
By the Abel summation formula, we have
X1 Z x
A(x) A(u)
= + 2
du.
p≤x
p log x 1 u(log u)
P log p
By Theorem 16, we can write A(x) = p
= log x + γ(x), where
p≤x
X log p
γ(x) := − log x = O(1).
p≤x
p
Also, we note that A(u) = 0 for 1 ≤ u < 2. Thus
X1   Z x
1 log u + γ(u)
=1+O + du
p≤x
p log x 2 u(log u)2
Z x Z x  
1 γ(u) 1
=1+ du + 2
du + O .
2 u log u 2 u(log u) log x
It follows that
X1 Z x  
γ(u) 1
= 1 + (log log x − log log 2) + 2
du + O
p≤x
p 2 u(log u) log x
Z ∞ Z ∞  
γ(u) γ(u) 1
= log log x + (1 − log log 2) + du − du +O
2 u(log u)2 x u(log u)2 log x
| {z }
=O( log1 x )
 
1
= log log x + β + O ,
log x
R∞ γ(u)
where β = 1 − log log 2 + 2 u(log u)2
du is a constant. 
Definition 9. The constant β is called Merten’s constant.
Remark 6. One can show that there is a relationship between β and γ. More specifically,
  !
X 1 1
β=γ+ log 1 − + = 0.261497 . . . ,
p
p p
where γ is the Euler-Mascheroni constant.
14
 9. September 30: Riemann ζ-function
For s ∈ C, consider the series
∞
X 1
,
n=1
ns
which converges absolutely if Re(s) > 1.
Definition 10. For s ∈ C with Re(s) > 1, the Riemann zeta-function is defined by
∞
X 1
ζ(s) := .
n=1
ns

Remark 7. Note that

Y −1 Y  
1 1 1
1− s = 1 + s + 2s + · · · .
p
p p
p p

Since a typical term in the above product is of the form

1 1 1
ak s = ak s = s ,
pa11 s · · · pk a1
(p1 · · · pk ) n
(where n = pa11 · · · pakk ), by the fundamental theorem of arithmetic for Re(s) > 1 we have
Y −1 X ∞
1 1
1− s = s
= ζ(s).
p
p n=1
n

Q −1
1
Definition 11. The product 1− ps
is called the Euler product representation for
p
ζ(s).

10. October 2
Theorem 18. The following are true:
(1) For s ∈ C with Re(s) > 1, we have
∞
{u}
Z
s
ζ(s) = −s du.
s−1 1 us+1
Therefore it follows
lim (s − 1)ζ(s) = 1.
s→1+

(2) ζ(s) has analytic continuation to Re(s) > 0 with s 6= 1. It is analytic except for a
simple pole at s = 1.
Proof. For (a), apply Abel’s summation formula with an = 1 and f (x) = x−s . Then we have
Z x
X 1 bxc buc
s
= s
+ s s+1
du.
n≤x
n x 1 u
15
By letting x → ∞ we see that for Re(s) > 1,
Z ∞
buc
ζ(s) = 0 + s du
1 us+1
Z ∞
u − {u}
=s du
1 us+1
Z ∞ Z ∞
1 {u}
=s s
du − s du
1 u 1 us+1
Z ∞
s {u}
= −s du.
s−1 1 us+1
R∞
Therefore (s − 1)ζ(s) = s − s(s − 1) 1 u{u} s+1 du. Since {u} = O(1), the above integral
converges for Re(s) > 0. It follows that lim+ (s − 1)ζ(s) = 1.
s→1
As for part (b), we see that from the identity theorem for analytic functions, since
R ∞ {u}
1 us+1
du converges for Re(s) > 0, ζ(s) has an analytic continuation to Re(s) > 0 with
s 6= 1. 
Theorem 19. ζ(s) has no zero in the region Re(s) ≥ 1.
Proof. If Re(s) > 1 we will show in Assignment #2 that ζ(s) 6= 0. Recall that |u| < 1, we
have
∞
X un
− log(1 − u)) = .
n=1
n
Thus
Y −1 ! ∞
1 XX 1 1
log ζ(s) = log 1− s = · ns .
p
p p n=1
n p
Write s = σ + it with σ, t ∈ R. Then
p−int = e−int log p = cos(−nt log p) + i sin(−nt log p) = cos(nt log p) − i sin(nt log p).
Thus Re(p−int ) = cos(nt log p). It follows that
∞
XX p−σn cos(nt log p)
Re(log ζ(σ + it)) = .
p n=1
n
Note that for θ ∈ R,
0 ≤ 2(1 + cos θ)2 = 2 + 4 cos θ + 2 cos2 θ + 3 + 4 cos θ + cos(2θ).
Thus
∞
XX p−σn
(3 + 4 cos(nt log p) + cos(2nt log p)) ≥ 0.
p n=1
n
This implies that
Re(3 log ζ(σ) + 4 log(σ + it) + log(σ + 2it)) ≥ 0.
Particularly, for σ > 1 and t ∈ R, we have
|ζ(σ)|3 |ζ(σ + it)|4 |ζ(σ + 2it)| ≥ 1. (∗)
16
Recall that lim+ (s − 1)ζ(s) = 1. Hence
s→1

lim+ |ζ(σ)| = lim+ |(σ − 1)−1 |.

σ→1 σ→1

Suppose that 1 + it0 is a zero of ζ(s) of order m ≥ 1, i.e., when σ + it → 1 + it0 we have
ζ(σ + it) = ((σ + it) − (1 + it0 ))gσ + it)
for some function g with g(1 + it0 ) 6= 0. Since ζ(s) has a pole at s = 1, t0 6= 0. Also, by
taking t = t0 we have
lim+ |ζ(σ + it0 )| = C1
σ→1
for some constant C1 6= 0. Hence
lim |ζ(σ + it0 )| = lim+ |C1 (σ − 1)m |.
σ→1+ σ→1

Also, since 1 + 2it0 is not a pole of ζ(s) there exists C2 such that
lim |ζ(σ + 2it0 )| C2 .
σ→1+

Since m ≥ 1, we have
lim |ζ(σ)|3 · |ζ(σ + it0 )|4 · |ζ(σ + 2it0 )| = lim+ |(σ − 1)−3 · c41 (σ − 1)4m · c2 | = 0,
σ→1+ σ→1

but this contradicts (∗). Therefore ζ(s) has no zeros in Re(s) > 1. 

11. October 5
We proved last class that:
(1) the analytic continuation of ζ(s) to Re(s) > 0
(2) the non-vanishing of ζ(s) on Re(s) = 1.
These are the main ingredients to prove the prime number theorem.
Theorem 20 (Donald J. Newman). Let an ∈ C with |an | ≤ 1 for n ∈ N. Consider the series
∞
X an
,
n=1
ns
which converges to an analytic function, say F (s) for Re(s) > 1. If F (s) can be analytically
continued to Re(s) ≥ 1, then the series converges to F (s) for Re(s) ≥ 1.
Proof. Let w ∈ C with Re(w) ≥ 1. Thus F (z + w) is analytic for Re(z) ≥ 0. Choose R ≥ 1
and let δ = δ(R) > 0 so that F (z + w) is analytic on the region. Define
e := {z ∈ C : Re(z) ≥ −δ, |z| ≤ R}.
Γ
Let M denote the maximum of |F (z + w)| on Γ, e and let Γ denote the contour obtained by
following the boundary of Γ
e counterclockwise.
Let A be the part of Γ with Re(z) > 0 and B the remainder part of Γ. For N ∈ N,
consider the function  
z 1 z
F (z + w)N + ,
z R2
17
which is analytic on Γ,
e except a (possible) simple pole at z = 0. Then by Cauchy’s residue
theorem, we see
Z  
z 1 z
2πiF (w) = F (z + w)N + dz
Γ z R2
Z   Z  
z 1 z z 1 z
= F (z + w)N + dz + F (z + w)N + dz. (5)
A z R2 B z R2
We see that on A, F (z + w) is equal to its series.
Split the series as
N
X an
SN (z + w) =
n=1
nz+w
and
RN (z + w) = F (z + w) − SN (z + w).
Note that SN (z + w) is analytic for z ∈ C. Let C be the contour given by the path |z| = R,
taken in counterclockwise direction. Thus by Cauchy’s residue theorem, we have
Z  
z 1 z
2πiSN (w) = SN (z + w)N + dz.
C z R2
Note that C = A ∪ (−A) ∪ {iR, −iR}. Thus
Z   Z  
z 1 z z 1 z
2πiSN (w) = SN (z + w)N + dz + SN (z + w)N + dz.
A z R2 −A z R2
By changing variable z to −z in the above integral, we see that
Z   Z  
z 1 z −z 1 z
SN (z + w)N + dz = SN (−z + w)N + dz.
−A z R2 A z R2
Thus  
z −z 1 z
2πiSN (w) =∈A (SN (z + w)N + SN (−z + w)N ) + 2 dz.
z R
Combining this with (5), we have
Z  
z −z 1 z
2πi(F (w) − SN (w)) = (RN (z + w)N − SN (−z + w)N ) + dz
A z R2
Z  
z 1 z
+ F (z + w)N + dz
B z R2
We now need to show that SN (w) → F (w) as N → ∞. Write z = x + iy with x, y ∈ R.
Then for z ∈ A, we have |z| = R and thus
1 z 2x
+ 2 = 2.
z R R
Since |nz | = nx we have
∞ ∞ Z ∞
X 1 X 1 1 1
|RN (z + w)| ≤ ≤ ≤ du = .
n=N +1
nRe(z+w) n=N +1
nx+1 N ux+1 xN x
18
Also, we have
N N
X 1 X Nx
|SN (−z + w)| ≤ ≤ N x−1 + ux−1 du ≤ N x−1 + .
n=1
n−x+1 n=1
x
x 1 1


Therefore |SN (−z + w)| ≤ N N + x . Combining the above estimates, we have
Z  
z −z 1 z
(RN (z + w)N − SN (−z + w)N ) + dz
A z R2
Z    
1 x x 1 1 −x 2x
≤ N + N + N dz
A xN x N x R2
Z   Z  
2 1 2x 4 2x
= + dz ≤ + dz
A x N R2 A R2 N R2
 
4 2
≤ + πR (∵ x ∈ R)
R2 N R2
4π 2π
= 2+ .
R N
We now estimate the integral over B. Divide B into two parts: Re(z) = −δ and −δ <
Re(z) ≤ 0. For z ∈ B with Re(z) = −δ since |z| ≤ R we have
|z|2
 
1 z 1 z̄ z̄z 1 2
+ 2 = + ≤ 1+ 2 ≤ .
z R z z R δ R δ
Let
N
X an
SN (z + w) = ,
n=1
nz+w
where Re(w) ≥ 1. We have shown that
Z Z  
z 1 z
2πi(F (w) − SN (w)) = (∗) dz + F (z + w)N + dz. (6)
A B z R2
| {z }
≤ 4π
R
+ 2π
N

Divide B into two parts: Re(z) = −δ and −δ < Re(z) < 0. For z ∈ B, z1 + Rz2 ≤ 2δ −1 .
Since |F (z + w)| ≤ M for z ∈ B, we have
Z   Z R Z 0
z 1 z −δ 2 2x
F (z + w)N + 2 dz ≤ MN dz + 2 M N x 2 dx
B z R −R δ −δ R
Z 0
4M R 4M
≤ δ
+ 2 xN x dx
δN R

| −δ {z  }
1 1
− (ex)
log N N δ log x

4M R 4M δ
≤ δ
+ 2 . (7)
δN R log N
Combining (??) and (7) gives us
4π 2π 4M R 4M δ
|2πi(F (w) − SN (w))| ≤ + + δ
+ 2 ,
R N δN R log N
19
or equivalently
2 1 MR 2M δ
|F (w) − SN (w)| ≤
+ + δ
+ 2 .
R N δN R log N
Given ε > 0, choose R = 3ε . Then for N sufficiently large, we have |F (w) − SN (w)| < ε. It
implies that SN (w) → F (w) as N → ∞, as required. 

12. October 7
Theorem 21. Let µ be the Möbius function. Then
∞
X µ(n)
= 0.
n=1
n

Proof. Fro Re(s) > 1, we have

Y   X∞
1 1 µ(n)
= 1− s = s
.
ζ(s) p n=1
n
p prime

We have seen in Theorems 18 and 19 that the function (s − 1)ζ(s) = f (s) is analytic and
nonzero in Re(s) ≥ 1. Thus
1 s−1
= ,
ζ(s) f (s)
which is analytic in Re(s) ≥ 1. In particular, it converges at s = 1. Since ζ(s) has a (simple)
pole at s = 1, then ζ(s)−1 has a zero at s = 1. It follows that
∞
X µ(n) 1
= = 0,
n=1
n ζ(1)
as required. 
P
Theorem 22. µ(n) = o(x).
n≤x

Proof. Take an := µ(n)/n and f (u) = u. Then

X µ(n)
A(x) = = o(1)
n≤x
n
by Theorem 21. Now the theorem follows immediately from the Abel summation formula:
note that
X Z x
µ(n) = xA(x) − A(u) du = o(x). 
n≤x 1

Definition 12. We say (x1 , . . . , xn ) ∈ Rn is a lattice point if xi ∈ Z for all i = 1, 2, . . . , n.

Theorem 23. Let d(n) be the number of positive divisors of n. We have then
n n j k
X X n √
d(m) = = n log n + (2γ − 1)n + O( x),
m=1 x=1
x
where γ is the Euler-Mascheroni constant.
20
Proof. Let Dn := {(x, y) ∈ R2 : x > 0, y > 0, xy ≤ n}. Let (x, y) ∈ Dn be a lattice point.
Note that each lattice point in Dn satisfies xy = m for some m ∈ N with 1 ≤ m ≤ n. Thus
n
is the number of lattice points in Dn . Note that each fixed x ∈ {1, . . . , n}, there are nx
P  
m=1
many y with xy ≤ n. Hence
n n j k
X X n
d(m) = .
m=1 x=1
x
Divide the lattice points in Dn into three disjoint regions:
Dn,1 := {(x, y) ∈ N2 , xy ≤ n, x < y}
Dn,2 := {(x, y) ∈ N2 , xy ≤ n, x > y}
Dn,3 := {(x, y) ∈ N2 , xy ≤ n, x = y}.
√
We have |Dn,1 | = |Dn,2 |. Let (x, y) ∈ Dn,1 . Then x2 < xy ≤ n, or x < n. Also,
√ for a fixed
n
x, the number of y satisfying xy ≤ n and y > x is x − bxc. Also, |Dn,3 | = b nc. Also,
√
n j bXnc
X nk j n k  √ 
=2 − bxc + n
x=1
x x=1
x
√
b nc
X n  √ 
=2 − x + O(1) + n
x=1
x
 √ √
√

b nc (b nc + 1)
= 2n(log n + γ + O(n−1/2 )) − 2
√ 
+ O( n),
2
with the last
√ equality
√ following
√ from Theorem 14.
Since b nc = n − { n} and log(1 − r) = O(r) for 0 < r < 1 we have
√ 
√ √ √
 
√  { n}
log( n ) = log( n − { n}) = log n 1− √
n
√ 
√ √

{ n}
= log n + log 1 − √ = log n + O(n−1/2 ).
n
Combining all the estimates presented here, we get
n n j k
X X n √ √
d(m) = = 2n(log n + γ + O(n−1/2 )) − n + O( n)
m=1 x=1
x
√
= n log n + (2γ − 1)n + O( n). 

13. October 9
Proposition 24. Given a function f : R+ → C, let
X x
F (x) = f .
n≤x
n
Then X x
f (x) = µ(n)F .
n≤x
n
21
Proof. By Proposition 10, we have
 
X X x
f (x) =  µ(k) f

n≤x
n
k|n
X x
= µ(k)f
kl≤x
kl
 
X X x X x
= µ(k)  f = µ(k)F . 
k≤x x kl k≤x
k
l≤ k

x
Theorem 25 (Prime number theorem). π(x) ∼ log x
.

Proof. By Theorem 12, it suffices to prove that

X
ψ(x) = log p ∼ x.
pk ≤x

Let
X  x jxk 
F (x) := ψ − + 2γ .
n≤x
n n

By Proposition 24, we have

X x
ψ(x) − bxc + 2γ = µ(n)F ,
n≤x
n

that is,
X x
ψ(x) = x + O(1) + µ(n)F .
n≤x
n

Our goal is to show that

X x
µ(n)F = o(x).
n≤x
n

We have
X x X jxk
F (x) = ψ − + 2γ bxc . (8)
n≤x
n n≤x
n
22
Note also that
X x XX
ψ = Λ(m)
n≤x
n n≤x x
m≤ n
 
X X
= Λ(m)  1
m≤x x
n≤ m
X jxk
= Λ(m)
m≤x
m
    
X x x
= log p + 2 + ···
p≤x
p p
p prime
X
= log(bxc!) = log n.
n≤x

j k j k j k
Note that the sum xp + px2 + · · · ends at pxk where pk k x.
We have seen in the proof of Theorem 15 that
X
log n = x log x − x + O(log x),
n≤x

from which it follows

X x
ψ = x log x − x + O(log x). (9)
n≤x
n

By Theorem 23,
bxc  
X bxc
= bxc log bxc + (2γ − 1) bxc + O(x1/2 ).
n=1
n

Note that
bxc   bxc j k bxc  
X bxc X x X bxc + 1
≤ ≤ .
n=1
n n=1
n n=1
n

Therefore
bxc j k
X x
= x log x + (2γ − 1)x + O(x1/2 ). (10)
n=1
n

Combining (8), (9), and (10) gives

F (x) = (x log x − x + O(log x)) − (x log x + (2γ − 1)x + O(x1/2 )) + (2γx + O(1)) = O(x1/2 ).
23
Hence there is a constant c > 0 such that for x ≥ 1, we have |F (x)| < cx1/2 . Suppose t ∈ N
and t ≥ 2. Then
X x X x X  x 1/2
µ(n)F ≤ F ≤ c
n n n
n≤ xt n≤ xt n≤ tx

! (11)
Z x/t  
du  x 1/2 cx
≤ cx1/2 1+ 1/2
≤ cx 1/2
1+2 − 2 ≤ 2√ .
1 u t t
Observe that F is a step function. In particular, if a ∈ Z with a ≤ x < a + 1 then
F (x) = F (a). Thus
X x X X X
µ(n)F = F (1) µ(n) + F (2) µ(n) + · · · + F (t − 1) µ(n).
x n x x x x x
t
<n≤x 2
<n≤x 3
<n≤ 2 t
<n≤ t−1

So we have
X x t−1
X X
µ(n)F ≤ |F (k)| µ(n)
x n x
t
<n≤x k=1 k+1
<n≤ x
k

t−1
!
X X
≤ |F (k)| max µ(n)
2≤i≤t x x
k=1 i
<n≤ i−1

t
X X
≤ ci1/2 · max µ(n) = o(t3/2 · x).
2≤i≤t x x
i=1 i
<n≤ i−1

2c
Hence for any given ε > 0, choose t = t(ε) so that t1/2
< 2ε . Then by (11),

X x εx
µ(n)F < .
n 2
n≤ xt

εx
Now, for (some fixed) ε and t, we can choose x sufficiently large so that o(t3/2 x) ≤ 2
. Also,
since
X x ε
µ(n)F < x,
x n 2
t
<n≤x

from the two aforementioned inequalities we have

X x
µ(n)F = o(x). 
n≤x
n

14. October 14 & 16

Remark 8. In 1896, Hadamard and de la Vallée Poussin proved that
x
π(x) ∼ .
log x
24
Let
Z x ∞
1 x X k!
Li(x) := dt ∼
2 log t log x k=0 (log x)k
x x 2x
= + 2
+ + ··· .
log x (log x) (log x)3

In 1899, de la Vallée Poussin proved that as x → ∞, there exists a > 0 such that
√
π(x) = Li(x) + O(xe−a log x
).

Remark 9. The main ingredient in the proof of the prime number theorem (Theorem 25) is
that
X
µ(n) = o(x),
n≤x

which is a consequence of the analytic continuation and non-vanishing of ζ(s) at Re(s) = 1.

The Riemann hypothesis (RH) states that the ’non-trivial zeroes’ of ζ(s) all have real part
1
2
. In 1901, Helge von Koch proved that RH is true if and only if
√
π(x) = Li(x) + O( x log x).

Remark 10. We proved in Assignment #1 that

1X
Nn = µ(d)q n/d .
n
d|n

Thus we have
qn q n/k
 
Nn = +O .
n n
In other words, the “Riemann hypothesis in Fq [t]” is true.

Definition 13. For n ∈ N, let ω(n) denote the number of distinct prime factors of n, and
let Ω(n) denote the number of prime factors of n counted with multiplicity.

Example 3. If n = 23 · 35 · 52 , then ω(n) = 3 and Ω(3) = 3 + 5 + 2 = 10.

For k ∈ N and x ∈ R, define

τk (x) = #{n ≤ x : Ω(n) = k}

πk (x) = #{n ≤ x : ω(n) = Ω(n) = k},

i.e., squarefree with k prime factors. Note that π(x) = π1 (x) = τ1 (x).
1 x
Theorem 26 (Landau (1900)). For k ∈ N, πk (x) ∼ τk (x) ∼ (k−1)!
· log x
· (log log x)k−1 .
25
Proof. Define
0
X 1
Lk (x) =
p1 ···pk
p p . . . pk
≤x 1 2
0
X
Πk (x) = 1
p1 ···pk ≤x
0
X
Θk (x) = log(p1 . . . pk ),
p1 ···pk ≤x
P0
where signifies that the sum is taken over all k-tuples of primes (p1 , . . . , pk ) with p1 p2 · · · pk ≤
x. Note that different k-tuples may correspond to the same product p1 . . . pk .
For n ∈ N, let
cn = cn (k) = #{k-tuples(p1 , . . . , pk ) : p1 p2 · · · pk = n}.
P P
Then Πk (x) = cn and Θk (x) = cn log n. Note that
n≤x n≤x
(
0 if n is not a product of k primes
cn =
k! if n is squarefree and ω(n) = Ω(n) = k.
Also, 0 < cn < k! if Ω(n) = k and n is not squarefree. It follows that
k!πk (x) ≤ Πk (x) ≤ k!τk (x) (12)
For k ≥ 2, note that
τk (x) − πk (x) = #{n ≤ x : Ω(n) = k and n is not squarefree}.
Thus
0   0  
X k X k
τk (x) − πk (x) = 1≤ 1= Πk−1 (x) (13)
p1 p2 ...pk ≤x
2 p ≤x
2
1 p2 ...pk−1
pi =pj for some i6=j

By (12) and (13), to prove that

1 x
πk (x) ∼ τk (x) ∼ · (log log x)k−1 ,
(k − 1)! log x
it suffices to show that
x(log log x)k−1
Πk (x) ∼ k (†)
log x
for all k ∈ N. Let an = cn and f (u) = log u. By Abel’s summation
Z x
X Πk (u)
Θk (x) = cn log n = Πk (x) log x − du.
n≤x 1 u
Observe that
Πk (x) ≤ k!τk (x) ≤ k!x.
Thus Πk (u) = O(u). it follows that
Θk (x) = Πk (x) log x + O(x).
26
Thus to prove (†), it suffices to show that for all k ∈ N,
Θk (x) ∼ kx(log log x)k−1 .
We will prove this by induction on k. For k = 1, by the prime number theorem we
have θ1 (x) = θ(x) ∼ x (by the prime number theorem). Assume now that Θk (x) ∼
kx(log log x)k−1 . Consider Θk+1 . Note that for k ≥ 1, we have
 k !k
X 1 X1
  ≤ Lk (x) ≤ .
1/k
p p≤x
p
p≤x

By Theorem 17,
!k
X1
∼ (log log x)k
p≤x
p
and  k
X 1
  ∼ (log log x1/k )k = (log log x − log k)k ∼ (log log x)k .
1/k
p
p≤x

Thus Lk (x) ∼ (log log x)k . It follows that

Θk+1 (x) − (k + 1)x(log log x)k = Θk+1 (x) − (k + 1)xLk (x) + o(x(log log x)k ).
Note that
k log(p1 p2 . . . pk pk+1 ) = log(pk1 pk2 · · · pkk pkk+1 )
k+1
X
= log(p1 p2 · · · p̂i · · · pk pk+1 ),
i=1

where p1 p2 · · · p̂i · · · pk pk+1 denotes the product of all p1 , . . . pk+1 with pi being omitted (or
equivalently, the above sum is over all k-subsets of {1, 2, . . . , k + 1}). Thus we have
0
X
kΘk+1 (x) = k · log(p1 · · · pk+1 )
p1 ···pk+1 ≤x
0
X k+1
X
= log(p1 p2 · · · p̂i · · · pk pk+1 )
p1 ···pk+1 ≤x i=1

X 0
X
= (k + 1) log(p2 · · · pk+1 )
p1 ≤x p2 ···pk+1 ≤ px
1
 
X x
= (k + 1) Θk .
p1 ≤x
p1

Next, we put L0 (x) = 1, and note

0  
X 1 X 1 x
Lk (x) = = Lk=1 .
p1 ···pk ≤x
p 1 · · · p k p≤x
p 1 p1
27
Therefore by the above two estimates,
X 1 
x

x

x

Θk+1 (x) − (k + 1)xLk (x) = (k + 1) Θk − Lk−1
p1 ≤x
k p1 p1 p1

by the induction hypothesis. So we have

Θk (y) − kyLk−1 (y) = o(y(log log y)k−1 ).
Thus given ε > 0, there exists x0 = x0 (ε, k) such that for y ≥ x0 we have
|Θk (y) − kyLk−1 (y)| ≤ εy(log log y)k−1 .
Further, there exists c = c(ε, k) > 0 such that for y ≤ x0 ,
|Θk (y) − kyLk−1 (y)| ≤ C.
Thus for x sufficiently large (note: > x0 ⇔ p<1 xx0 ).
x
p1
 
 k−1
k+1 X x x X
|Θk+1 (x) − (k + 1)xLk (x)| ≤ ε log log + c
k x p1 p1 x
p1 < x x0
≤p1 ≤x
0
X 1
≤ 2εx(log log x)k−1 + 2cx
x p1
p1 < x
0
k
≤ 4εx(log log x) + 2cx
≤ 5εx(log log x)k .
Thus Θk+1 (x) − (k + 1)xLk (x) = o(x(log log x)k ), from which we conclude that
Θk+1 (x) ∼ (k + 1)x(log log x)k . 

15. October 19
Theorem 27. The following statements hold:
X
ω(n) = x log log x + βx + o(x)
n≤x
X
Ω(n) = x log log x + β̃x + o(x),
n≤x

1
P
where β is Merten’s constant and β̃ = β + p(p−1)
.
p
P P
Proof. Let S(x) := ω(n) and T (x) := Ω(n). By Theorem 17, we have
n≤x n≤x

XX X x X1
S(x) = 1= =x + O(π(x))
n≤x p|n p≤x
p p≤x
p
= x(log log x + β + o(1)) + O(π(x)) = x log log x + βx + o(x).
28
Note that  
X x   X x X 
T (x) − S(x) = = + O 1 .
pm pm

pm ≤x pm ≤x pm ≥x
m≥2 m≥2 m≥2
m m
Note√also that 2 ≤ p ≤ x and thus m ≤ log x/ log 2. Also, p2 ≤ pm ≤ x implies that
p ≤ x. Thus
X 1
T (x) − S(x) = x + O(x1/2 log x)
pm ≤x
pm
m≥2
 
 
X 1 1 X 1 
1/2
= x 2
+ 3
+ · · · − m  + O(x log x).
p
p p pm >x
p
m≥2

Note also that

X 1 X 1 1

m
≤ 2
+ 3 + · · · = O(x−1 ) = o(1).
pm >x
p n≥x
n n
m≥2
Therefore !
X 1
T (x) − S(x) = x + o(1) + O(x1/2 log x).
p
p(p − 1)
By our estimate of S(x) we have
!
X 1
T (x) = x log log x + β+ x + o(x),
p
p(p − 1)
| {z }
β̃

as required. 
Definition 14. Let A ⊆ N. For n ∈ N, let
A(n) := A ∩ {1, 2, . . . , n}.
¯
Then the upper asymptotic density of A, denotes d(A), is defined by

¯ |A(n)|
d(A) := lim sup .
n→∞ n
Similarly, we define d(A) the low asymptotic density of A by
|A(n)|
d(A) = lim inf .
nn→∞
¯
If d(A) ¯
= d(A), we say A has an asymptotic density, and we say d(A) = d(A) = d(A).
¯
Example 4. If P is the set of primes, then d(A) = d(A)) = 0. If A := {n ∈ N, 5 | n},
¯ 1
then d(A) = d(A) = 5 . If B = {n ∈ N, n not of the form k 2 + 1 for some k ∈ N}, then
¯
d(B) = d(B) = 1.
29
Example 5. Let D = {a ∈ N : (2k)! < a < (2k + 1)! for some k ∈ N}. Then for n = (2k + 1)!
any a satisfying (2k)! < a < (2k + 1)! will be counted. Thus
D((2k + 1)!) (2k + 1)! − (2k)! 2k
1≥ ≥ = .
(2k + 1)! (2k + 1)! 2k + 1
Thus as k → ∞ we have
D((2k + 1)!)
→ 1.
(2k + 1)!
¯
Hence d(D) = 1. On the other hand, if n = (2k)!, then we only count a up to (2k − 1)!.
Thus
D((2k)!) (2k − 1)! 1
0≤ ≤ = .
(2k)! (2k)! 2k
Hence as k → ∞ we have as k → ∞
D((2k)!)
→ 0.
(2k)!
So d(D) = 0. Therefore D has no asymptotic density.
Definition 15. Let f (n) and F (n) be functions from N to R+ . We say that f (n) has normal
order F (n) if for any ε > 0 the set
A(ε) = {n ∈ N : (1 − ε)F (n) < f (n) < (1 + ε)F (n)}
has the property that d(A(ε)) = 1.
Definition 16. Let f (n) and F (n) be functions from N to R+ . We say that f (n) has average
order F (n) if
Xn n
X
f (i) ∼ F (i).
i=1 i=1

Example 6. Let (
1 if n 6= k! for all k ∈ N
f (n) =
n if n = k!.
Then f has normal order 1, but not average order 1. Similarly, if
(
2 if n ≡ 1 (mod 2)
g(n) =
0 if n ≡ 0 (mod 2),
then g has average order 1 but does not have normal order 1. The third example is
(
log n + (log n)1/2 if n ≡ 1 (mod 2)
h(n) =
log n − (log n)1/2 if n ≡ 0 (mod 2).
h(n) has both normal and average order log n.
P
From Theorem 27, we see that ω(n) and Ω(n) have average log log n. Note that log log n ∼
n≤x
x log log x). We will prove next class that they have normal order log log n.
30
 16. October 21
We proved last time that
X
ω(n) = x log log x + O(x).
n≤x

Also note that

X X X X
log log n = log log n + log log n = O(x1/2 log log x) + log log n.
n≤x n≤x1/2 x1/2 <n≤x x1/2 <n≤x

Note that
X X
log log n ≥ (log log x − log 2) 1 = x log log x + O(x1/2 log log x)
x1/2 <n≤x x1/2 <n≤x
X X
log log n ≤ log log x 1 = x log log x + O(x1/2 log log x).
x1/2 <n≤x x1/2 <n≤x

Thus X
log log n = x log log x + O(x1/2 log log x).
n≤x

Thus the average order of ω(n) is log log n.

Theorem 28. We have
X
(ω(n) − log log x)2 = O(x log log x).
n≤x

Proof. Note that

X X X X
(ω(n) − log log x)2 = ω 2 (n) − 2 log log x ω(n) + (log log x)2 1 .
n≤x n≤x n≤x n≤x
| {z }
x(log log x)2 +O((log log x)2 )

By Theorem 27, we have

X
2 log log x ω(n) = 2x(log log x)2 + O(x log log x).
n≤x

We now consider the sum of ω 2 (n). W ehave

 
  
X X X X X X X 
ω 2 (n) =  1  1 = 
 1 + 1

n≤x n≤x p|n q|n n≤x pq|n p|n
p6=q
XX Xx
X
= 1+ ω(n) = + O(x log log x)
pq≤x n≤x n≤x pq≤x
pq
p6=q pq|n p6=q
X 1
=x + O(x) + O(x log log x).
pq≤x
pq
p6=q
31
Also we have X 1 X 1 X 1 X 1
= − 2
= + O(1).
pq≤x
pq pq≤x
pq 2
p pq≤x
pq
p ≤x
p6=q
Observe that  2 !2
X 1 X 1 X1
  ≤ ≤ .
1/2
p pq≤x
pq p≤x
p
p≤x
By Theorem 17, we have
!2
X1
= (log log x)2 + O(log log x)
p≤x
p
and
 2
X 1
  = (log log x1/2 + O(1))2 = (log log x − log 2 + O(1))2 = (log log x)2 + O(log log x).
1/2
p
p≤x

Therefore X 1
= (log log x)2 + O(log log x).
pq≤x
pq
Combining the above estimates gives
X
ω 2 (n) = x(log log x)2 + O(x log log x).
n≤x

It follows that
X X X X
(ω(n) − log log x)2 = ω 2 (n) − 2 log log x ω(n) + (log log x)2 1
n≤x n≤x n≤x n≤x

= x(log log x) + O(x log log x) − 2x(log log x) + x(log log x)2
2 2

= O(x log log x). 

Corollary 29. Let δ > 0. Then
#{n ≤ x : |ω(n) − log log n| > (log log n)1/2+δ }
is o(x). Thus the normal order of ω(n) is log log n.
Proof. The number of n ≤ x1/2 is o(x). Also, for x1/2 < n ≤ x we have log log x ≥ log log n >
log log x − log 2. Thus to prove the corollary it suffices to show that
E(x) := #{n ≤ x : |ω(n) − log log x| > (log log x)1/2+δ } = o(x).
By Theorem 28, we have
X
E(x) · (log log x)1+2δ ≤ (ω(n) − log log x)2 = O(x log log x).
n≤x

So it follows that  
x log log x
E(x) = O = o(x),
(log log x)1+2δ
as required. 
32
Corollary 30. The normal order of Ω(n) is log log n.
Proof. By Theorem 27, we have
X
(Ω(n) − ω(n)) = O(x).
n≤x

Thus #{n ≤ x : Ω(n) − ω(n) > (log log n)1/2+δ }. Then the results follows from Corollary
29. 
Remark 11. Since the average
P order of ω(n) is log log n, which is asymptotic to log log x for
almost all n, we can view (ω(n) − log log x)2 as the square of the standard deviation of
n≤x
ω(n). On Assignment #3 we shall prove that actually
X
(ω(n) − log log x)2 ∼ x log log x.
n≤x
√
Thus the standard variation of ω(n) is about log log n.
Definition 17. Let Z γ
1 t2
G(γ) := √ e− 2 dt.
2π −∞
Then G(γ) is called the Gaussian normal distribution.
Remark 12. In 1934, Erdős and Kac proved that
 
1 ω(n) − log log n
# n≤x: √ ≤ γ = G(γ).
x log log n
17. October 23
Recall that the normal order of ω(n) and Ω(n) is log log n. Let d(n) be the number of
positive divisors of n. If n = pa11 · · · par r with a1 , . . . , ar ∈ N and p1 , . . . , pr distinct primes
then ω(n) = r and Ω(n) = a1 + · · · + ar and d(n) = (a1 + 1) · · · (ar + 1).
Theorem 31. For ε > 0, define the set
S(ε) := {n ∈ N, 2(1−ε) log log n < d(n) < 2(1+ε) log log n }.
Then S(ε) has asymptotic density 1.
Proof. Note that for a ∈ N, we have 2 ≤ 1 + a ≤ 2a . Thus we have
2ω(n) ≤ d(n) ≤ 2Ω(n) .
Then the result follows from Corollaries 29 and 30. 
Remark 13. Recall that X
d(n) ∼ x log x.
n≤x
Thus the average order of d(n) is log n. However, by the above theorem, for almost all n,
d(n) satisfies
(log n)log 2−ε < d(n) < (log n)log 2+ε
for any ε > 0.
33
 18. October 23: Quadratic reciprocity
Definition 18. For n ∈ N, the Euler-totient function is defined by
ϕ(n) = #{1 ≤ m ≤ n : (m, n) = 1}.
Theorem 32 (Euler’s theorem). Let a ∈ N and (a, n) = 1. Then
aϕ(n) ≡ 1 (mod n).
Proof. Let c1 , c2 , . . . , cϕ(n) be a reduced residue system mod n. Since (a, n) = 1, the set
{ac1 , . . . , acϕ(n) } is also a reduced residue system mod n. Thus
c1 c2 · · · cϕ(n) ≡ (ac1 )(ac2 ) · · · (acϕ(n) ) (mod n)
ϕ(n)
c1 c2 · · · cϕ(n) ≡ a (c1 c2 · · · cϕ(n) ) (mod n)
ϕ(n)
a ≡1 (mod n). 
Corollary 33 (Fermat’s little theorem). Let p be a prime. Then for any a ∈ Z with p - a we
have ap−1 ≡ 1 (mod p).
Theorem 34 (Wilson’s theorem). Let p be a prime. Then
(p − 1)! ≡ −1 (mod p).
Proof. Consider the polynomial xp−1 − 1 in (Z/pZ)[x]. By Corollary 33, 1, 2, . . . , p − 1 are
its roots.Thus in (Z/pZ)[x] we have
xp−1 − 1 = (x − 1)(x − 2) · · · (x − (p − 1)).
Consider the constant coefficients on both sides; we see that −1 ≡ (−1)(−2) · · · (−(p − 1)) =
(−1)p−1 (p − 1)! (mod p). If p = 2 then the result holds since −1 ≡ 1 (mod 2). Otherwise if
p is odd then (−1)p−1 = 1 so the result follows. 

  19. Let p be a prime and let a ∈ Z with (a, p) = 1. We define the Legendre
Definition
symbol ap by the rule
  (
a 1 if x2 ≡ a (mod p) has a solution
=
p −1 otherwise.
 
If ap = 1 then a is a quadratic residue mod p. Otherwise, then a is a quadratic non-residue
mod p.
Theorem 35 (Euler’s criterion). Let p be an odd prime and let a ∈ Z with (a, p) = 1. Then
 
p−1 a
a 2 ≡ (mod p).
p
Proof. The congruence x2 ≡ 1 (mod p) has at most two solutions mod p. Two cases:
 
(1) Suppose that there is a solution, say b. Then ap = 1. Since b2 ≡ a (mod p), we
have p−1
a 2 ≡ bp−1 ≡ 1 (mod p).
p−1
 
Thus a 2 ≡ 1 ≡ ap (mod p).
34
  
(2) Suppose that there is no solution. In this case, ap = −1. Since (a, p) = 1, for each
fixed r ∈ (Z/pZ)∗ , there exists a unique s ∈ (Z/pZ)∗ such that rs ≡ a (mod p).
Since x2 ≡ a (mod p) has no solution, we see that r 6= s. Split elements in (Z/pZ)∗
into p−1
2
pairs (r, s) with r 6= s and rs ≡ a (mod p). Thus
p−1
(p − 1)! ≡ a 2 (mod p).
p−1
 
But Theorem 34 says a 2 ≡ −1 ≡ ap (mod p), so the claim follows. 

19. October 26
Theorem 36. Let p be an odd prime and let a, b, ∈ Z. Then
    
a b ab
= .
p p p
Also,
 
−1
= (−1)p(p−1)/2 .
p
Note
  that if a ∈ Z and p | a, then we extend the definition of the Legendre symbol by letting
a
p
= 0.

Proof. The statement holds if p | ab (i.e., p | a or p | b). Thus we may assume p - a and p - b. By
Euler’s criterion we have
    
ab p−1 p−1 p−1 a b
= (ab) 2 ≡ a 2 b 2 ≡ (mod p).
p p p
     
Since ab p
, ap , pb ∈ {−1, 1} and p is an odd prime, the result follows. Again, by Euler’s
criterion we have  
−1 p−1
≡ (−1) 2 (mod p).
p
 
Since −1 p
∈ {−1, 1} and p is an odd prime, it follows that
 
−1 p−1 p(p−1)
= (−1) 2 = (−1) 2 ,
p
as required. 
Theorem 37 (Gauss’s lemma). Let p be an odd
p−1

prime and a ∈ Z with (a, p) = 1. Let µ
be the number of integers from a, 2a, . . . , 2 a whose residues mod p of least absolute
value (i.e., in − p−1
2
, p−1
2
) are negative. Then
 
a
= (−1)µ .
p
Example 7. If p = 5 and a = 2 then we have {2, 4}, or equivalent {2, −1}. Therefore µ = −1.
35
Proof. We first replace the integers a, 2a, 3a, . . . , p−1
2
a by their residues of least absolute value.
Denote the negative ones by −s1 , −s2 , . . . , −sµ and the positive ones by r1 , . . . , r p−1 −µ . Since
2
1 ≤ ri , sj ≤ p−1 2
, no two ri ’s are equal and no two sj ’s are equal. We further claim that
ri 6= sj for all i, j. To see this, note that if m1 a ≡ ri (mod p) and m2 a ≡ −sj (mod p) with
ri = sj , then (m1 + m2 )a ≡ 0 (mod p). Since (a, p) = 1, it follows that p | m1 + m2 . But
this is a contradiction since 1 ≤ m1 , m2 ≤ p−1 2
. Since ri 6= rj and si 6= sj and ri 6= sj for all
i 6= j, we see that s1 , . . . , sµ , r1 , . . . , r p−1 −µ is a rearrangement of 1, 2, . . . , p−1
2
. Thus
2
   
p−1 p−1
a · 2a · · · · · a ≡ 1 · 2 · ······ · (−1)µ (mod p)
2 2
p−1
⇔ a 2 ≡ (−1)µ (mod p)
 
a
⇔ = (−1)µ ,
p
 
by Euler’s criterion (and also that (−1) , ap ∈ {±1} and p is odd).
µ


Corollary 38. If p is an odd prime, then

 
2 p2 −1
= (−1) 8 .
p
Proof. Consider the set {2, 4, 6, . . . , p−1
2
· 2}. Note that
p−1 p−1
2r ≤
⇔r≤ .
2 4
Thus the number of integers on the set whose residues of least absolute value is negative is
equal to
 
p−1 p−1
µ= − .
2 4
Note that if p ≡ 
±1(mod 8) then µ is even; if p ≡ ±3 (mod 8) then µ is odd. By Gauss’s
lemma, we have p2 = (−1)µ . Thus w is a quadratic residue if p ≡ ±1 (mod 8) and is a
quadratic non-residue if p ≡ ±3 (mod 8). 

20. October 28
Theorem 39 (Law of quadratic reciprocity). If p, q are distinct odd primes, then
  
p q p−1 q−1
= (−1) 2 · 2 .
q p
13−1 17−1
13
= (−1) 2 · 2 13 17
= 17 4





Example 8. 17 13
= 13
= 1.
Example 9. We claim that 5 is a quadratic residue mod p if and only if p ≡ ±1 (mod 10).
Indeed, note that
 
5 5−1 p−1
· 2
p p
= (−1) 2 = ,
p 5 5
36
and that
(
p 1 if p ≡ ±1 (mod 10)
=
5 −1 if p ≡ ±3 (mod 10).
   
Proof. By Gauss’s lemma, we have pq = (−1)µ and pq = (−1)ν , where µ is the number of
integers from {p, 2p, · · · , q−1


2
p} whose residue mod

q of least absolute value is negative and
p−1
ν is the number of integers from {q, 2q, . . . , 2 q} whose residue mod p of least absolute
value is negative. Thus to prove the claim, it suffices to show that
  
p−1 q−1
µ+ν ≡ (mod 2).
2 2
We claim that

Claim 3. For x ∈ Z with 1 ≤ x ≤ q−1 2

there exists a unique y ∈ Z such that − 2q < xp − qy <
q
2
. Furthermore, y ≥ 0 (note that xp − qy is the residue mod q of the least absolute value of
xp).

Proof of Claim 3. Note that

q q xp 1 xp 1
− < xp − yq < ⇔ − − < −y < − + . (?)
2 2 q 2 q 2
Thus y is uniquely
determined. Also, we note that if y < 0 then xp − yq ≥ q. Since
xp − yq ∈ − 2q , 2q we see that y ≥ 0. 
q−1
Note that if y = 0 there is no contribution from xp−yq to µ since xp > 0. Also, if x = 2
,
then from (?), we have
q−1

 
xp 1 2
p 1 p q−1 1 p+1
y< + = + = + < ,
q 2 q 2 2 q 2 2

since y ∈ Z, y ≤ p−1 2
. Thus the number µ corresponds to the number of combinations
of x and y from the sequences (A) 1, 2, . . . , q−1
2
and (B) 1, 2, . . . , p−1
2
respectively such that
−q q
2
< xp−yq < 0 or equivalently 0 < yq −xp < 2 . Similarly, ν is the number of combinations
of x and y from the sequences (A) and (B) respectively, for which − p2 < yq − xp < 0. So for
any other pairs (x, y) with x from (A) and y from (B), either yq − xp < − p2 or yq − xp > 2q .
Let ρ be the number of pairs (x, y) for which yq − xp < − p2 and λ be the number of pairs
(x, y) for which yq − xp > 2q . Then
  
q−1 p−1
= ν + µ + ρ + λ.
2 2
As x and y run through (A) and (B) respectively,
q+1 p+1
x0 = − x and y 0 = −y
2 2
37
run through (A) and (B) respectively but in reserve order. Note that yq − xp < − p2 if and
only if
   
0 0 p+1 q+1
yq−xp= −y q− −x p
2 2
q−p q
= − (yq − xp) > .
2 2
Then ρ = λ. It follows that
  
q−1 p−1
= µ + ν + 2λ ≡ µ + ν (mod 2). 
2 2

21. October 30
Example 10. The equation x4 − 17y 4 = 2w2 has no integral solution. Suppose otherwise,
that is there exist x, y, w ∈ Z such that x4 − 17y 4 = 2w2 . Without loss of generality, we can
assume (x, y) = 1. Thus x and w are coprime. Note that if p is an oddprime which divides
w, since x4 ≡ 17y 4 (mod p), i.e., 17 ≡ (x2 y −2 )2 (mod p) we have 17 p
= 1. By the law of
quadratic reciprocity,  
p p−1 17−1 17
· 2
= (−1) 2 = 1.
17 p
Thus an odd prime p dividing w is a quadratic residue mod 17. Also by Corollary 38 we
have  
2 172 −1
= (−1) 8 = 1,
17
so 2 is a quadratic residue mod 17. By the above arguments, we see that any prime (either 2
or p) dividing w is a quadratic residue mod 17. Therefore we have w ≡ t2 (mod 17) for some
t ∈ Z. Note that 17 - w and that 17 - t. Now, since x4 − 17y 4 = 22p , it follows that xf ≡ 2t4
(mod 17). Thus 2 ≡ x4 t−4 (mod 17), that is there exists r ∈ Z such that 2 ≡ r4 ≡ 17, which
is a contradiction. One can generalize it to x4 − ay 4 = bw2 , i.e., determining with what kind
of a and b would this will work.
Example 11. Is the congruence 3x2 + 7x − 42 ≡ 0 (mod 391) solvable? First, multiply
both sides by 12 to get 36x2 + 84x − 516 ≡ 0 (mod 391), or (6x + 7)2 ≡ 565 (mod 391).
Thus it suffices to consider y 2 ≡ 174 (mod 391). Note that 391 = 17 · 23. We see that
y 2 ≡ 174 (mod 17) ⇔ y 2 ≡ 4 (mod 17) which has a solution. Also, note that y 2 ≡ 174
(mod 391) ⇔ y 2 ≡ 13 (mod 23). By the law of quadratic reciprocity, we have
           
13 23−1 13−1
· 2 23 10 2 5 2 3
= (−1) 2 = = = = (−1)(−1) = 1.
23 13 13 13 13 13 5
Thus y 2 ≡ 13 (mod 23) has a solution. Since y 2 ≡ 13 (mod 17) and y 2 ≡ 13 (mod 23), by
the Chinese remainder theorem, y 2 ≡ 174 (mod 391) has a solution.
713


Example 12. What is 1009 ?
Note that 713 = 23 · 31. Then by Theorem 36, we have
    
713 23 31
= .
1009 1009 1009
38
By the law of quadratic reciprocity, it follows
         
23 23−1 1009−1
· 1009 1009 20 5
= (−1) 2 2 = = =
1009 23 23 23 23
   
5−1 23−1 23 3
= (−1) 2 · 2 = = −1.
5 5
Similar argument yields
    
31 2 7
= = 1 · (−1) = −1.
1009 17 17
Hence  
713
= (−1)(−1) = 1.
1009
Remark 14. In the above calculation, we are given the fact that 713 = 23 · 31. However, it is
notalways easy to find the prime factorization of an integer a. Yet, it is possible to evaluate

a
p
without knowing the prime factorization of a. The idea is to ‘flip’ the Legendre symbol
p


to a even when a is not a prime.
Definition 20. Let a ∈ Z and n ∈ Z be odd. If n = pα1 1 · · · pαr r , we define the Jacobi symbol
to be
a k  αi
Y a
:= .
n i=1
p i

Theorem 40 (Generalized law of quadratic reciprocity). Let a, b ∈ N be odd. Then

  (
−1 1 if b ≡ 1 (mod 4)
(1) =
b −1 if b ≡ 3 (mod 4).
  (
2 1 if b ≡ 1 or 7 (mod 8)
(2) =
b −1 if b ≡ 3 or 5 (mod 8).
(

b
a
a
if a ≡ 1 (mod 4) or b ≡ 1 (mod 4)
(3) = b


b − a if a ≡ b ≡ 3 (mod 4).
Proof. Exercise! 
713


Example 13. Now compute 1009 using the generalized law of quadratic reciprocity. Since
713 ≡ 1 (mod 4) and 713 ≡ 1 (mod 8), we have
       3   
713 1009 296 2 · 37 37
= = = = .
1009 713 713 713 713
Now, since 37 ≡ 1 (mod 4) and 37 ≡ 5 (mod 8), we have
          
37 713 10 2 5 5
= = = =− .
713 37 37 37 37 37
Since 5 ≡ 1 (mod 4), we have
     
5 37 2
= = = −1.
37 5 5
39
  
713
Therefore = 1.
1009
22. November 2: primitive roots
We recall the Euclidean algorithm: for a, b ∈ Z, there exist x, y ∈ Z such that ax + by =
(a, b).
Theorem 41 (Chinese remainder theorem). Let m1 , . . . , mt ∈ Z with (mi , mj ) = 1 for all
i 6= j, and let m = m1 · · · mt . Let b1 , . . . , bt ∈ Z. Then the simultaneous congruences
x ≡ b1 (mod m1 )
..
.
x ≡ bt (mod mt )

has a unique solution modulo m.

Proof. Let ni = m/mi (1 ≤ i ≤ t). Then (mi , ni ) = 1. Then there exist ri , si ∈ Z such that
ri ni + si mi = 1 (1 ≤ i ≤ t). Let ei = ri ni . Then ei ≡ 0 (mod mj ) and ei ≡ 1 (mod mj ) for
i 6= j. Consider now
Xt
x0 = bi ei .
i=1
Then x0 ≡ bi (mod mi ) for each 1 ≤ i ≤ t. That is, it is a solution of the simultaneous
congruences. To prove the uniqueness of x0 , suppose that x1 ≡ bi (mod mi ) for all 1 ≤ i ≤ t.
Then mi | (x1 − x0 ) (1 ≤ i ≤ t). Since (mi , mj ) = 1 for i 6= j and m = m1 . . . mt , we have
m | (x1 − x0 ), i.e., x1 ≡ x0 (mod m). 
For n ∈ Z, let (Z/nZ)∗ denote the invertible elements in Z/nZ. That is, they are the
congruence classes (r + nZ) for which there exists (s + nZ) with (r + nZ)(s + nZ) = 1 + nZ.
This is equivalent to saying that (r, n) = 1.
Theorem 42. Let m1 , . . . , mt ∈ N with (mi , mj ) = 1 for i 6= j, and let m = m1 m2 . . . mt .
Then
Z/mZ ∼= Z/m1 Z × · · · × Z/mt Z
as rings. Also,
(Z/mZ)∗ ∼ = (Z/m1 Z)∗ × · · · × (Z/mt Z)∗
as groups.
Proof. Let ψ : Z → Z/m1 Z × · · · × Z/mk Z defined by ψ(n) 7→ (n + m1 Z, . . . , n + mt Z). It is
a straightforward verification that ψ is a ring homomorphism. Note that ψ is surjective and
ker ψ = mZ, by the Chinese remainder theorem. Thus by the first isomorphism theorem for
rings, the first claim follows. Now let λ : (Z/mZ)∗ → (Z/m1 Z)∗ × · · · × (Z/mt Z)∗ be defined
by λ(r + mZ) = (r + m1 Z, . . . , r + mt Z). Note that (r, m) = 1 if and only if (r, mi ) = 1
for all 1 ≤ i ≤ t. Thus the map is well-defined. It is straightforward to verify λ is a group
homomorphism. It is also bijective by the Chinese remainder theorem. 
Corollary 43. Let m1 , . . . , mt ∈ N with (mi , mj ) = 1 for i 6= j, and let m = m1 m2 . . . mt .
Then ϕ(m) = ϕ(m1 )ϕ(m2 ) · · · ϕ(mt ).
40
Proof. Note that ϕ(m) = |(Z/mZ)∗ | and ϕ(m1 ) · · · ϕ(mt ) = |(Z/m1 Z)∗ | · · · |(Z/mt Z)∗ | =
|(Z/m1 Z)∗ × · · · × (Z/mt Z)∗ |. So now the result follows from Theorem 42. 
Corollary 44. Let m = pa11 · · · pat t , where p1 . . . pt are distinct primes and a1 , . . . , at ∈ N.
Then
t  
Y 1
ϕ(m) = m 1− .
i=1
p i

ai ai ai ai −1
Proof.
 Take mi = pi , where 1 ≤ i ≤ t in Corollary 43. Note that ϕ(pi ) = pi − pi =
ai 1
pi 1 − pi . So it follows that
t     t  
Y 1 1 Y 1
ϕ(m) = ϕ(pai i ) = pa11 · · · pat t 1− ··· 1 − =m 1− . 
i=1
p1 pt i=1
p i

Proposition 45. Let p be a prime. If d | (p−1), then xd ≡ 1 (mod p) has exactly d solutions
mod p.
Proof. Write p − 1 = dk with k ∈ N. Then
xp−1 − 1 (xd )k − 1
= = (xd )k−1 + · · · + xd + 1 = g(x) ∈ (Z/pZ)[x].
xd − 1 xd − 1
By Fermat’s little theorem, xp−1 − 1 has p − 1 distinct roots in Z/pZ. This (xd − 1)g(x)
factors into linear factors in (Z/pZ)[x] and the result follows. 
Theorem 46. (Z/pZ)∗ is a cyclic group.
Proof. For each divisor d of p − 1, let λ(d) denote the number of elements of (Z/pZ)∗ of order
d. By Prop 45, there are exactly d elements whose order divides d. Thus
X
d= λ(c).
c|d

By the Möbius inversion formula, we have

X µ(c) X µ(c) Y 1

λ(d) = d=d =d 1− = ϕ(d).
c c p
c|d c|d p|d

Thus there are ϕ(p − 1) elements of (Z/pZ)∗ of order p − 1. In particular, (Z/pZ)∗ is

cyclic. 
Remark 15. For a general n ∈ N, the group (Z/nZ)∗ is not always cyclic. For example,
(Z/8Z)∗ = {1, 3, 5, 7}, but none of the elements have order 4.

23. November 4
Definition 21. Let n ∈ N and a ∈ Z. We say a is a primitive root modulo n if a + nZ
generates (Z/nZ)∗ .
Example 14. w is a primitive root mod 5 but is not a primitive root module 7, since 23 ≡ 1
(mod 7).
41
Remark 16. We have seen in the proof of Theorem 46 that for a prime p, (Z/pZ)∗ is cyclic.
Thus there exists a primitive root mod p. In fact, we see from the proof of the theorem,
there are ϕ(p − 1) primitive root modulo p. Note that if a ∈ N is a square, then it is a
quadratic residue mod p. Thus a is not a primitive root mod p.
Conjecture (Artin’s primitive root conjecture). If a ∈ N is not a perfect square, then a isa
primitive root mod p for infinitely many primes p.
Remark 17. The conjecture still remain open, but some progress has been made. In 1967,
Hooley proved that the conjecture is true under the assumption of the generalized Riemann
hypothesis (GRH). In 1980’s, using sieve theory, Gupta, K. Murty, R. Murty, and Heath-
Brown showed unconditionally that given any non-square a, b, c ∈ N, then at least one of
them is a primitive root mod p for infinitely many primes p. For example, one of 2, 3, 5 is a
primitive root mod p for infinitely many primes p. However, the result is not constructive,
and thus we do not know which one satisfies the condition.
Proposition 47. Let p be a prime and l ∈ N. If a ≡ b (mod pl ), then ap ≡ bp ≡ pl+1 .
Proof. Write a = b + cpl for some c ∈ Z. Then
     
p l p p p p−1 l p p−2 l 2 p
a = (b + cp ) = b + b cp + b (cp ) + · · · + (cpl )p .
1 2 p
Since pl+1 | p1 bp−1 cpl and pl+1 | pil for 2 ≤ i ≤ p, it follows that ap ≡ bp (mod pl+1 ).



Proposition 48. Let p be an odd prime and l ∈ N with l ≥ 2. Then for a ∈ Z, we have
l−2
(1 + ap)p ≡ 1 + apl−1 (mod pl ).
Proof. We prove the result by induction on l. The result is immediate for l = 2. Suppose
that the result holds for some l ∈ N with l ≥ 2. And we prove it for l + 1. By Proposition
47 and the induction hypothesis, we have
   
pl−1 pl−2 p l−1 p p l−1 p
(1+ap) ≡ ((1+ap) ) ≡ (1+ap ) ≡ 1+ ap +· · ·+ (apl−1 )p (mod modpl+1 ).
1 p
Note that for l ≥ 2 and k ≥ 3, we have
2(l − 1) + 1 ≤ 3(l − 1) ≤ k(l − 1).
It follows that
p2(l−1)+1 | (apl−1 )k
for k = 3, . . . , p. Also, we note that
 
p p(p − 1) l−1 2 p − 1 2(l−1)+1
(apl−1 )2 = (ap ) = ap .
2 2 2
p−1
Since 2
∈ Z as p is odd, it follows that
 
2(l−1)+1 p
p (apl−1 )2 .
2
42
Note that 2(l − 1) + 1 ≥ l + 1 for l ≥ 2. Thus pl+1 | p2(l−1)+1 . Thus we have
     
pl−1 p l−1 p l−1 2 p
(1 + ap) ≡1+ ap + (ap ) + · · · + (apl−1 )p
1 2 p
 
p
≡1+ apl−1 ≡ 1 + apl (mod pl+1 ).
1
By induction, the result follows. 

24. November 6
Proposition 49. If p is an odd prime, l ∈ N with l ≥ 2 and a ∈ Z with (a, p) = 1, then
1 + ap has order pl−1 in (Z/pl Z)∗ .
Proof. Note that the group (Z/pl Z)∗ is of order pl − pl−1 = pl−1 (p − 1). By Proposition 48,
we have
l−2
(1 + ap)p ≡ 1 + apl−1 (mod pl ).
l−2
Since (a, p) = 1, we have (1 + ap)p 6≡ 1 (mod pl ). Then by Proposition 48 again,
l−1
(1 + ap)p ≡ 1 + apl (mod pl+1 ).
l−1
Thus (1 + ap)p ≡ 1 (mod pl ). It follows that (1 + ap) has order pl−1 in (Z/pl Z)∗ . 
Theorem 50. Let p be an odd prime and l ∈ N with l ≥ 2. Then (Z/pl Z)∗ is a cyclic group.
Proof. By Theorem 46, there exists a primitive root modulo p. Note that
     
p−1 p−1 p − 1 p−2 p − 1 p−3 2 p − 1 p−1
(g + p) =g + g p+ g p + ··· + p .
1 2 p−1
If we assume that g p−1 ≡ 1 (mod p2 ) then
 
p−1 p − 1 p−2
(g + p) ≡1+ g p (mod p2 ).
1
Since p - (p − 1) and (g, p) = 1 we see that
(g + p)p−1 6≡ 1 (mod p2 ).
Thus at least one of g p−1 and (g + p)p−1 is not congruent to 1(mod p2 ). Without loss of
generality, we may assume that g p−1 6≡ 1 (mod p2 ).
We claim that if g p−1 6≡ 1 (mod p2 ), then g is a primitive root modulo p. We note that
once we prove this claim, we are home free. Time to prove this claim. Suppose that g has
order m in (Z/pl Z)∗ . Since |(Z/pl Z)∗ | = pl − pl−1 = pl−1 (p − 1), we have m - pl−1 (p − 1).
Write m = dps where d | (p − 1) and 0 ≤ s ≤ l − 1. By Fermat’s little theorem, we have
s s
g p ≡ g (mod p). Thus g p ≡ g (mod p). Thus, g p ≡ g (mod p) for s ∈ N. Since g m ≡ 1
(mod pl ) and thus g m ≡ 1 (mod p), we have
s
g d ≡ (g p )d ≡ g m ≡ 1 (mod p).
Since g is a primitive root mod p, we have (p − 1) | d. Thus d = (p − 1). Since g p−1 ≡ 1
(mod p) and g p−1 6≡ 1 (mod p2 ), there exists a ∈ Z with (a, p) = 1 such that g p−1 ≡ 1 + ap
(mod p2 ). By Proposition (49), 1 + ap has order pl−1 in (Z/pl Z)∗ . Thus g has order (p − 1)pl ,
which implies that (Z/pl Z)∗ is cyclic. 
43
 25. November 9
Theorem 51. Let l ∈ N.
(1) If l = 1, 2 then (Z/2l Z)∗ is cyclic.
(2) For l ≥ 3, (Z/2l Z)∗ ∼
= Z/2Z × Z/2l−2 Z. In particular, we have
(Z/2l Z)∗ = {(−1)a 5b + 2l Z : a ∈ {0, 1}, b ∈ {0, 1, . . . , 2l−2 − 1}}.
Proof. It is straightforward to verify that (Z/2Z)∗ and (Z/4Z)∗ are cyclic. Thus we focus on
the second part.
l−3
Claim 4 (Claim 1). For l ≥ 3, 52 ≡ 1 + 2l−1 (mod 2l ).
Proof of Claim 1. We prove the claim by induction on l. For l = 3, we have 5 ≡ 1 + 22
(mod 23 ) as required. Suppose that the above congruence holds for some l ∈ N with l ≥ 3
l−3
and we prove it for l + 1. Write 52 ≡ 1 + 2l−1 + k2l for some k ∈ Z. It follows that
l−2
52 = (1 + 2l−1 + k2l )2
= 1 + (2l−1 )2 + (k2l )2 + 2 · 2l−1 + 2 · k2l + 2 · 2l−1 k2l
= 1 + 2l + k2l+1 + 22l−2 + k22l + k 2 22l .
Note that 2(l − 1) ≥ l + 1 for l ≥ 3. Thus we have
l−2
52 ≡ 1 + 2l (mod 2l+1 ).
l−3
By induction, the claim holds. From the above proof, we see that 52 6≡ 1 (mod 2l ) and
l−2
52 ≡ 1 (mod 2l ). Thus 5 has order 2l−2 in (Z/2l Z)∗ . 
Claim 5 (Claim 2). For l ≥ 3, the numbers
(−1)a 5b with a ∈ {0, 1} and b ∈ {0, 1, . . . , 2l−2 − 1}
are distinct modulo 2l .
Proof of Claim 2. Suppose that (−1)a1 5b1 ≡ (−1)a2 5b2 (mod 2l ) with 0 ≤ ai ≤ 1 and 0 ≤
bi < 2l−2(i = 1, 2). Then (−1)a1 5b1 ≡ (−1)a2 5b2 (mod 4). Since 5 ≡ 1 (mod 4), we see that
(−1)a1 ≡ (−1)a2 (mod 4). Thus a1 = a2 . We now have 5b1 ≡ 5b2 (mod 2l ). Since 5 has
order 2l−2 in (Z/2l Z)∗ and 0 ≤ bi ≤ 2l−2 , it follows that b1 = b2 . 
Since (Z/2l Z)∗ = {(−1)a 5b : a ∈ {0, 1}, b ∈ {0, 1, . . . , 2b−2 − 1}} it follows that
(Z/2l Z)∗ ∼
= (Z/2Z) × Z/2l−2 Z.


26. November 9
Theorem 52. The group (Z/nZ)∗ is cyclic (i.e. it has a primitive root) if and only if
n = 1, 2, 4, pl , 2pl with p being an odd prime and l ∈ N.
Proof. Let n = 2l0 pl11 · · · plrr where l0 ∈ N ∪ {0} and l1 , . . . , lr ∈ N, and pi distinct odd primes.
Then by Theorem 42,
r
∗ ∼
Y
(Z/nZ) = (Z/2 Z) × (Z/plii Z)∗ .
l0 ∗

i=1
44
By Theorem 46, (Z/plii Z)∗ is cyclic for 1 ≤ i ≤ r. By Theorem 51, (Z/2l0 Z)∗ is cyclic for
0 ≤ l0 ≤ 2 and is isomorphic to Z/2Z × Z/2l0 −2 Z for l ≥ 3. Thus, the order of any element
of (Z/nZ)∗ is a divisor of λ(n) := lcm(b, ϕ(pl11 ), . . . , ϕ(plrr )), where
(
ϕ(2l0 ) if 0 ≤ l0 ≤ 2
b= 1 l0
2
ϕ(2 ) if l0 ≥ 3.
Note that 2 | ϕ(plii ) for all 1 ≤ i ≤ r. It thus follows that
λ(n) < ϕ(2l0 )ϕ(pl11 ) · · · ϕ(plrr )
except in the cases n = 1, 2, 4, pl , 2pl . Since (Z/nZ)∗ is cyclic if and only if λ(n) =
ϕ(2l0 )ϕ(pl11 ) · · · ϕ(plrr ), the result follows. 
Definition 22. The number
λ(n) = lcm(b, ϕ(pl11 ), ϕ(pl22 ), . . . , ϕ(plrr ))
is called the universal exponent of n.
Theorem 53. For n ∈ N, let λ(n) be the universal exponent. Then for any a ∈ Z with
(a, n) = 1 we have
aλ(n) ≡ 1 (mod n).
Remark 18. Euler’s theorem states that for any a ∈ Z with (a, n) = 1, then we have
aϕ(n) ≡ 1 (mod n).
The above theorem gives a strengthening of Euler’s theorem.
Remark 19. Given a prime p, one can ask for an upper bound for the smallest positive integer
√
b which is a primitive root mod p. Hua proved that b < 2ω(p−1)+1 p.
Theorem 54. If p is a prime of the form 4q + 1 with q an odd prime, then 2 is a primitive
root mod p.
Proof. Let m be the order of 2 mod p. By Fermat’s little theorem, m | (p − 1) and thus m | 4q.
It follows that m = 1, 2, 4, 2q, 4q. Since p is a prime of the form 4q + 1 with q an odd prime,
we have p = 13 or p > 20. Thus m 6= 1, 2, 4. Also, by Euler’s criterion we have
 
2q p−1 2
2 ≡2 2 ≡ (mod p).
p
On the other hand, by Corollary 38, since q is odd we have
 
2 p2 −1 (4q+1)2 −1 2
= (−1) 8 = (−1) 8 = (−1)2q +q = −1.
p
Thus 22q ≡ −1 (mod p). That is, m 6= q, 2q. It follows that m = 4q = p − 1, i.e., 2 is a
primitive root mod p. 
Let k, l ∈ N with (k, l) = 1. Dirichlet’s theorem states that there are infinitely many
primes p with p ≡ l (mod k). To prove this theorem, we will introduce later the notion of
L functions. However, for many pairs (k, l), we can prove Dirichlet’s theorem by elementary
means. For example on Assignment #1, we show that there are infinitely many primes p
with p ≡ 5 (mod 6).
45
 27. November 11 & 13
Theorem 55. Let n ∈ N. There are infinitely many primes p with p ≡ 1 (mod n).
Proof. This proof is due to Birkhoff and Vandiver (1904). Let a ∈ N with a > 2 and
ζn = e2πi/n . Consider Φn (a), the n-th cyclotomic polynomial evaluated at a, i.e.,
Yn
Φn (a) = (a − ζnj ).
j=1
(j,n)=1

We recall that Φn (x) ∈ Z[x] and xn − 1 =

Q
Φd (x). We claim that:
d|n

Claim 6. If p is a prime dividing Φn (a) then p | n or p ≡ 1 (mod n).

Proof of the claim. Note that p | (an − 1) and thus p - a. Two cases:
(1) If p - (ad − 1) for all d | n with d 6= n. Then the order of a mod p is n. By Fermat’s
little theorem, n | (p − 1) and p ≡ 1 (mod n).
n −1
(2) Suppose that p | (ad − 1) for some d | n with d 6= n. Note that Φn (x)| xxd −1 (in Z[x]).
an −1
Since p | Φn (a), it follows that p | ad −1 .
We have  
n d n n d n/d
a = (1 + (a − 1) = 1 + (a − 1) +
d (ad − 1)2 + · · · + .
d 2
Thus
an − 1
 
n n/d
= + (ad − 1) + · · · .
ad − 1 d 2
n −1
Since p | aad −1 and p | (ad − 1), we conclude that p | nd . Thus we have p | n. This completes the
proof of the claim. 
We are now ready to prove the theorem. Suppose that there are only finitely many primes
p1 , . . . , pr such that pj ≡ 1 (mod n) for all 1 ≤ j ≤ r. Write Φn (x) = xϕ(n) + · · · + ±1.
Consider then Φn (np1 p2 · · · pr m). We see that (Φ(np1 . . . pr m), n) = 1. Also, since pj -
Φn (np1 . . . pr m)(1 ≤ j ≤ r). Letting m → ∞, we see that for m sufficiently large we have
Φn (np1 . . . pr m) ≥ 2. Thus it has a prime divisor p, which is not equal to p1 , . . . , pr . By the
claim we have either p ≡ 1 (mod n) or p | n. Since (Φn (np1 . . . pr m), n) = 1, we have p - n.
Thus p ≡ 1 (mod n). However p ∈ / {p1 , . . . , pr }, and this leads to a contradiction. 
Definition 23. Let G be a finite abelian group. A character of G is a homomorphism
χ : G → C∗ . The set of characters of G forms a group under the operation
(χ1 · χ2 )(g) := χ1 (g)χ2 (g).
This group is called the dual group of G and is denoted by G b := {χ : G → C∗ homomorphism}.
The identity of Gb is the principal character χ0 , where χ0 (g) = 1 for all g ∈ G. Note that if
|G| = n then g n = e (the identity element) for all g ∈ G. It follows that (χ(g))n = 1 and
thus χ(g) is an n-th root of unity.
Theorem 56. Let G be a finite abelian group. Then
(1) |G| = |G|
b
(2) G ∼
=G b
46
 (3) We have
(
X |G| if g = e
χ(g) =
0 otherwise.
χ∈Gb

and (
X |G| if χ = χ0
χ(g) =
g∈G
0 otherwise.

Proof. Suppose that |G| = n. Since G is a finite abelian group, we have

G∼
= Z/n1 Z × · · · × Z/nr Z.
h
Thus there exist g1 , . . . , gr ∈ G such that gj j = e(1 ≤ j ≤ r) and every element g ∈ G has
a unique representation in the form g = g1a1 · · · grar with 0 ≤ aj ≤ hj (1 ≤ j ≤ r). Note that
any character χ is determined by its action on g1 , . . . , gr . Since (χ(gj ))hj = 1, we see that
χ(gj ) is an hj -th root of unity. Thus there are at most h1 . . . hr characters. On the other
hand, if wj is a hj -th root of unity, we can define χ(gj ) = wj for (1 ≤ j ≤ r and extend it
multiplicatively to all elements of G. Thus there are at least h1 . . . hj characters. It follows
therefore that |G|b = |G|.
For the second part, let χj be the character defined by χj (gj ) = e2πi/hj and χj (gk ) = 1 for
j 6= k. Define ϕ : G → G b by

ϕ(g1a1 · · · grar ) = χa11 · · · χar r .

One can check that ϕ is a group homomorphism. Also, since
χa11 · · · χar r (gj ) = e2πiaj /hj ,
we see that χa11 · · · χar r = χ0 if and only if aj = hj for all 1 ≤ j ≤ r. And this corresponds to
g1h1 · · · grhr = e, the identity of G. Thus ϕ is injective. Finally, since G is finite and |G|
b = |G|,
we see that ϕ is surjective also. Hence G b∼ = G as desired.
For the last part, we start by letting
X
S(g) = χ(g).
χ∈G
b

If g = e then χ(e) = 1 for all χ ∈ G.

b Thus

S(e) = |G|
b = |G|.

We now assume that g 6= e. By (2), there exists a character χ1 ∈ G b such that χ1 (G) 6= −1.
Also, since Gb∼= G, if χ ∈ G −1
b with χ 6= χ0 then there exists χ ∈ G b such that χχ−1 = χ0 . In
particular, if χ runs through all the elements of G,
b so does χ1 χ. Thus we have
X X X
S(g) = χ(g) = (χ1 χ)(g) = χ1 (g) χ(g) = χ1 (g)S(g).
χ∈G
b χ∈G
b χ∈G
b

Since (1 − χ1 (g) 6= 0, it follows that S(g) = 0 as required. 

47
 P
Let T (χ) := χ(g). Then if χ = χ0 then χ0 (g) = 1 for all g ∈ G so T (χ0 ) = |G|. If
g∈G
χ 6= χ0 then there exists g1 ∈ G such that χ(g1 ) 6= 1. Thus T (χ) = 0, since
X X X
T (χ) = χ(g) = χ(g1 g) = χ(g1 ) χ(g) = χ(g1 )T (χ).
g∈G g∈G g∈G

Let k ∈ N with k ≥ 2. Let χ be a character on (Z/kZ)∗ . We extend the definition of χ to

Z, also denoted by χ, by putting
(
χ(a + kZ) if (a, k) = 1
χ(a) =
0 otherwise.
Definition 24. We call such χ defined above a character mod k.
Theorem 57. Let χ be a character mod k.
(1) If (n, k) = 1 then χ(n) is a ϕ(k)-th root of unity.
(2) The function χ is completely multiplicative. That is, χ(mn) = χ(m)χ(n) for all
m, n ∈ Z.
(3) χ is periodic modulo k, that is, χ(n + k) = χ(n) for all n ∈ Z.
(4) We have that
(
X ϕ(k) if n ≡ 1 (mod k)
χ(n) =
0 otherwise.
χ char mod k

and (
k
X ϕ(k) if χ = χ0
χ(n) =
n=1
0 otherwise.
(5) Let χ denote the conjugate character to χ, i.e., χ(n) = χ(n) for all n ∈ Z. Let χ0 be
a character mod k. Then for (m, k) = 1 we have
(
X ϕ(k) if n ≡ m (mod k)
χ(n)χ(m) =
0 otherwise.
χ char mod k

and (
k
X ϕ(k) if χ0 = χ
χ(n)χ0 (n) =
n=1
0 otherwise.
Proof. The first four parts follow either from definition or Theorem 56. So we focus on (5)
only. Note that χ(m)χ(m) = 1 = χ(m)χ(m−1 ), where m−1 is the multiplicative inverse of
m modulo k. Thus χ(m) = χ(m−1 ). It follows that
X X X
χ(n)χ(m) = χ(n)χ(m−1 ) = χ(nm−1 ).
χ char mod k χ χ

By Theorem 56(3), the last sum is ϕ(k) if and only if nm−1 ≡ 1 (mod k), or equivalently
n ≡ m (mod k); and 0 otherwise.
Also, we note that if χ0 = χ, then χχ0 = χ0 . Otherwise, χχ0 is a non-principal character.
Thus the second result also follows, again from Theorem 56(3). 
48
 We now describe the group of characters mod k. By multiplicity, it is enough to discuss
the characters mod pl for a prime p.
(1) Assume first that p is an odd prime and let g be a primitive root mod pl . For n ∈ Z
with (n, p) = 1, there exists a unique ν ∈ Z with 1 ≤ ν ≤ ϕ(pl ) such that n ≡ g ν
(mod pl ). For d ∈ Z with 1 ≤ d ≤ ϕ(pl ), we define the character χd (n) by
 
d 2πidν
χ (n) = exp .
ϕ(pl )
We get in this way ϕ(pl ) different characters mod pl , and this gives the complete list
of characters mod pl .
(2) Consider characters mod 2l . If l = 1 then we only have the principal character. If
l = 2, then we have the principal chracter and the character χ4 which is defined by

1
 if n ≡ 1 (mod 4)
χ4 (n) = −1 if n ≡ 3 (mod 4)

0 otherwise.
If l ≥ 3, then (Z/2l Z)∗ is not cyclic. However, we have seen in Theorem 51 that for
each n ∈ Z with (2, n) = 1, i.e., n + 2l Z ∈ (Z/2l Z)∗ , there exists a unique integer
pair (a, b) with 0 ≤ a ≤ 1 and 0 ≤ b ≤ 2l−2 such that n ≡ (−1)a 5b (mod 2l ). Thus
for d ∈ Z with 1 ≤ d ≤ ϕ(2l )
(
2πida 2πidb


exp + if n ≡ 1 (mod 2)
χd (n) = 2 2l−2

0 otherwise.

We get in this way ϕ(2l ) different characters mod 2l and this gives the complete list
of characters mod 2l .

28. November 16: L-functions and Dirichlet’s theorem

Let k ∈ N with k ≥ 2 and χ be a character mod k. For Re(s) > 1, define
∞
X χ(n)
L(s, χ) := .
n=1
ns
Let χ0 be the principal character.
Theorem 58. The following hold:
(1) If χ 6= χ0 then L(s, χ) has an analytic continuation to Re(s) > 0.
(2) If χ = χ0 then L(s, χ0 ) has an analytic continuation to Re(s) > 0 with s 6= 1. At
s = 1, L(s, χ0 ) has a simple pole with residue ϕ(k)
k
.
P
Proof. Let A(x) := χ(n) and
n≤x
(
1 if χ = χ0
E(χ) :=
0 otherwise.
49
Also, notice that (
k
X ϕ(k) if χ = χ0
χ(n) =
n=1
0 otherwise.
By Theorem 57(4), we have
( 
x
ϕ(k) + T (x) if χ = χ0
A(x) =  xk 
k
· 0 + T (x) otherwise,
with |T (x)| ≤ ϕ(k).
It follows that
ϕ(k)
A(x) = E(χ) x + R(x),
k
where |R(x)| ≤ 2ϕ(k). Let f (n) = n−s . By Abel’s summation, we have
X χ(n) Z x
A(x) A(u)
s
= s +s s+1
du
n≤x
n x 1 u
x Z x
us+1

ϕ(k) 1 R(x) ϕ(k) R(u)
= E(χ) · s−1 + s + sE(χ) − +s s+1
du
k x x k s−1 1 1 u
Z x
ϕ(k) 1−s s 1−s R(x) R(u)
= E(χ) (x + (x − 1)) + s + s s+1
du. (14)
k 1−s x 1 u
Now we prove each claim. As for (1), if χ 6= χ0 then we have E(χ) = 0. We see from (14)
that Z x
X χ(n) R(x) R(u)
s
= s
+s s+1
du.
n≤x
n x 1 u

By letting x → ∞, since |R(x)| ≤ 2ϕ(k) for Re(s) > 0 we have

Z ∞
R(u)
L(s, χ) = s du.
1 us+1
Since the integral converges for Re(s) > 0, it follows that L(s, χ) has an analytic continuation
to Re(s) > 0.
We move on to (2). If χ = χ0 then E(χ) = 1. Thus by (14) we have
X χ(n)   Z x
ϕ(k) 1−s s 1−s R(x) R(u)
= x + (x − 1) + s + s s+1
du.
n≤x
n k 1 − s x 1 u
Consider Re(s) > 1. By letting x → ∞, we have
Z ∞
ϕ(k) s R(u)
L(s, χ0 ) = · +s du.
k s−1 1 us+1
Since the integral converges for Re(s) > 0, the function L(s, χ0 ) has an analytic continuation
to Re(s) > 0 except at a simple pole at s = 1 with residue ϕ(k)k
. 
Definition 25. Let {λn }∞ n=1 be a strictly increasing sequence of positive real numbers. For
z ∈ C, a Dirichlet series attached to {λn }∞n=1 is a series of the form
X
an e−λn z
n≥1
50
where {an }n≥1 is a sequence of complex numbers.

an e−λn z converges for z = z0 then it converges

P
Theorem 59. If the Dirichlet series
n≥1
uniformly for Re(z − z0 ) > 0 and |arg(z − z0 )| < a with a < π2 .
P
Proof. Without loss of generality, we may assume z0 = 0. Since an converges, for any
n≥1
ε > 0 there exists N = N (ε) ∈ N such that if l, m > N then
m
X
|an | < ε.
n=l

m
P
Let Al,m = an . By taking the convention that Al,l−1 = 0, we have
n=l

m
X m
X
−λn z
an e = (Al,n − Al,n−1 )e−λn z
n=l n=l
m−1
X
= Al,n (e−λn z + e−λn+1 z ) + Al,m e−λm z .
n=l

Thus for Re(z) ≥ 0, we have

m m−1
!
X X
−λn z −λn z −λn+1 z
an e ≤ε |e −e |+1
n=l n=l
R λn+1
Note that e−λn z − e−λn+1 z = z λn
e−tz dt. Also, for z = x + iy with x, y ∈ R we have
|e−tz | = e−tx . Thus we have
Z λn+1
−λn z −λn+1 z
|e −e | ≤ |z| e−tx dt
λn
|z| −λn x
≤ (e − e−λn+1 x ).
x
Therefore we have
m  
X
−λn z |z| −λl x −λm x
an e ≤ε (e −e )+1 .
n=l
x

Note that for |arg(z)| < α, we have |z|

x
< c for some c = c(α). Also, we have |e−λl x −e−λm x | ≤
2. It follows that
Xm
an e−λn z < (2c + 1)ε.
n=l

Therefore the Dirichlet series converges for Re(z) ≥ 0 and |arg(z)| ≤ α, as desired. 
51
 29. November 18 & 20
Theorem 60. If
∞
X
f (z) = an e−λn z
n=1
be a Dirichlet series with an ∈ R and an ≥ 0 for all n ∈ N. Suppose that the series converges
for Re(z) > σP 0 with σ0 ∈ R. Suppose also that f (z) can be analytically continued in a
−s −λn z
P
neighbourhood n of σ0 . Then there exists a real number ε > 0 such that an e for
n≥1
Re(z) > σ0 − ε.
Proof. Without loss of generality, we may assume that σ0 = 0. Since f (z) is analytic in a
neighbourhood of 0, by Theorem 59, it is analytic for Re(z) > 0. Since f (z) is analytic for
Re(z) > 0, and is also analytic in a neighbourhood of 0, there exists ε > 0 such that f is
analytic in |z − 1| ≤ 1 + ε. Note that for Re(z) > 0,
∞
X
f (m)
(z) = an (−λn )m e−λn z .
n=1

This implies that f (m) (1) = an (−λn )m e−λn . Thus the Taylor series expansion of f (z)
P
around 1 in |z − 1| ≤ 1 + ε is of the form
∞
X f (m) (1)
(z − 1)m .
m=0
m!
We now consider f (z) at the point z = −ε. We have
 
∞ ∞
X X (−1 − ε)m
f (−ε) =  an (−λn )m e−λn 
m=0
m!
n=1)
∞ ∞
!
X X
−λn (1 + ε)m
= an λ m
n e .
m=0 n=1
m!
Since an ≥ 0 and all other terms are positive, we can switch the order of summation and
obtain
∞ ∞
!
m m
X X (λ n ) (1 + ε)
f (−ε) = an e−λn
n=1 m=0
m!
∞
X ∞
X ∞
X
−λn λn (1+ε) λn ε
= an e ·e = an e = an e(−λn )(−ε) .
n=1 n=1 n=1

an e−λn z converges to f (z) at z = −ε. By Theorem 59 it converges to f (z)

P
Thus the series
for Re(z) > −ε. 
Theorem 61. For k ∈ N with k ≥ 2, let χ be a character mod k. Then
(1) L(s, χ) is non-zero for Re(s) > 1
(2) If χ 6= χ0 , then L(1, χ) is non-zero.
52
Proof. (1) Note that L(s, χ) converges absolutely for Re(s) > 1. Since χ is completely
multiplicative, L(s, χ) has a Euler product representation for Re(s) > 1 which is
Y −1
χ(p)
L(s, χ) = 1− s
p
p

for Re(s) > 1. Since

X χ(p)

p
ps
converges for Re(s) > 1, it follows that L(s, χ) is non-zero for Re(s) > 1.
(2) We recall that for |u| < 1, we have
∞
X un
− log(1 − u) = .
n=1
n

Thus for Re(s) > 1 we have

Y −1 !
χ(p)
log L(s, χ) = log 1− s
p
p
 
X χ(p)
= − log 1 − s
p
p
∞
XX χ(ps )
= .
p n=1
npns

Let l ∈ Z with (l, k) = 1. By summing over all characters mod k we have

∞
X XX 1 X
χ(l) log L(s, χ) = χ(l)χ(pn )
p n=1
npns
χ char mod k χ char mod k
∞
X X 1
= ϕ(k) ,
n=1 pn ≡l
npns
(mod k)

by Theorem 57(5). By taking l = 1 and exponentiating both sides, we have

 
∞
Y X X 1 
L(s, χ) = exp ϕ(k) .
n=1 n
npns
χ char mod k p ≡1 (mod k)

Thus if s ∈ R with s > 1, then we have

Y
L(s, χ) ≥ 1.
χ char mod k

We now split into cases depending on if χ is a real character or not.

(1) Suppose that L(1, χ) = 0 where χ is a non-real character. Since L(s, χ) = L(s, χ) for
s ∈ R with s > 1, we have L(1, χ) = L(1, χ) = 0.
53
 We also recall that for s ∈ R with s > 1, we have
Y
L(s, χ) > 1. (15)
χ char mod k

We have seen in Theorem 58 that L(s, χ0 ) has a simple pole at s = 1 and L(s, χ)
does not have a pole at s = 1 for any χ 6= χ0 . Thus as s → 1+ on the real line, we
have
Y
L(s, χ) = O((s − 1)−1 (s − 1)2 = O(s − 1),
χ char mod k

which contradicts (15). Thus L(1, χ) 6= 0 for χ a non-real character.

(2) Now suppose that L(1, χ) = 0 with χ a real character. For Re(s) > 1, define

ζ(s)L(s, χ)
g(s) := .
ζ(2s)

Consider the Euler product representation of g(s) for Re(s) > 1 we have
Y 1 − p−2s Y 1 + p−s
g(s) = =
p
(1 − p−s )(1 − χ(p)p−s ) p
1 − χ(p)p−s
∞
! ∞
!
Y X Y X
= (1 + p−s ) χ(pl )p−ls = 1+ (χ(pl−1 ) + χ(pl ))p−ls
p l=0 p l=1
∞
!
Y X
= 1+ b(pl )p−ls ,
p l=1

where b(pl ) = χ(pl−1 ) + χ(pl ) (l ≥ 1). Since χ is a real character, we have χ(p) ∈
{0, ±1}. Since χ is multiplicative, we have
(
0 if χ(p) = 0 or − 1
b(pl ) = χ(pl−1 ) + χ(pl ) =
2 if χ(p) = 1.

In all cases, we have b(pl ) ≥ 1 for all l ≥ 1. Thus

∞
X an
g(s) = 1 + , (16)
n=2
ns

with an ∈ R≥0 for all n ≥ 2. Since the zero of L(1, χ) eliminates the pole of ζ(s) at
s =! and since ζ(2s) is non-zero and analytic for Re(s) > 12 , it follows that g(s) has
an analytic continuation to Re(s) > 12 . By Theorem 60, we conclude that the series
+
defining g converges to g for Re(s) > 12 . As
s → 21 on the real axis, since ζ(2s) has
a pole at s = 12 we see that g(s) = O s − 21 . But this contradicts (16) as g(s) ≥ 1
for Re(s) > 21 . Thus L(1, χ) 6= 0 for χ real characters. 
54
 30. November 23
Theorem 62. Let l, k ∈ Z with k ≥ 2 and (l, k) = 1. Then the series
X
p−1
p≡l (mod k)

diverges. This implies that there are infinitely many primes p with p ≡ l (mod k).
Remark 20. For x ∈ R, let
π(x; k, l) = #{p ≤ x : p is a prime and p ≡ l (mod k)}.
Then using similar method used by Newman for his proof of the prime number theroem, one
can prove that
1 x
π(x; k, l) ∼ · .
ϕ(k) log x
This was proved by Valleé-Poussin. In the case when k is “small”, the Siegel-Walfisz theorem
gives a refinement of the above result. More precisely, define
X
ψ(x; k, l) = Λ(n)
n≤x
n≡l (mod k)

If k ≤ (log x)N for some N ∈ N, then

x
ψ(x; k, l) = + O(x exp(−CN (log x)1/2 )),
ϕ(k)
where the constant CN depends on N .
Proof. We have seen in the proof of Theorem 61 that
∞
1 X X X 1
χ(l) log L(s, χ) = . (17)
ϕ(k) n=1 pn ≡l (mod k)
npns
χ char mod k

We recall that (
1 if χ = χ0
E(χ) =
0 6 χ0 .
if χ =
As s → 1+ on the real axis, by Theorems 58 and 61, we see that (s − 1)E(χ) L(s, χ) tends to
a finite non-zero limit. Thus E(χ) log(s − 1) + log L(s, χ) tends to a limit. It follows that as
s → 1+ on the real axis, we have
log L(s, χ) = −E(χ) log(s − 1) + O(1).
Thus we have
1 X 1 1 X
χ(l) log L(s, χ) = log L(s, χ0 ) + χ(l) log L(s, χ)
ϕ(k) ϕ(k) ϕ(k)
χ char mod k χ char mod k
χ6=χ0

= − log(s − 1) + O(1).
55
Combining this with (17) we have
∞
X X 1 1
= − log(s − 1) + O(1).
n=1 pn ≡l (mod k)
npns ϕ(k)

Thus
∞
X
−s
X X 1 1
p + ns
=− log(s − 1) + O(1).
n=2 pn ≡l (mod k)
np ϕ(k)
p≡l (mod k)

Note that for Re(s) ≥ 1 and s ∈ R,

∞ ∞
X X 1 1X X 1
≤
n=2 pn ≡l (mod k)
npns 2 n=2 pns
pn ≡l (mod k)
∞  
1X 1 1
= + + ···
2 m=2 m2s m3s
∞  
1X 1 1
≤
2 m=2 m2s 1 − m1s
∞ ∞
X 1 X 1 π2
≤ 2s
≤ 2
≤ .
m=2
m m=2
m 6

Thus
X 1 1
s
=− log(s − 1) + O(1).
p ϕ(k)
p≡l (mod k)

1
As s → 1+ on the real axis the quantity − ϕ(k) p−1
P
log(s − 1) → ∞. It follows that
p≡l (mod k)
diverges. 

31. November 23: Waring’s problem

In 1770, Edward Waring asserted without proof that every natural number is a sum of at
most 4 squares, 9 cubes, 19 biquadrates and so on. Waring’s problem states that: for k ∈ N
with k ≥ 2, there exists a number s = s(k) such that every natural number is a sum of at
most s k-th powers of natural numbers, i.e., n = xk1 + · · · + xks with xi ∈ N ∪ {0} (1 ≤ i ≤ s).
Let g(k) denote the least s such that the above statement holds. Then Waring’s problem
states that g(k) < ∞. In 1770, Lagrange proved that g(2) = 4. By 1909, only known
cases were k = 2, 3, 4, 5, 6, 7, 8, 10. In 1909, by a combinatorial method, Hilbert proved that
g(k) < ∞ for every k ≥ 2. By the work of Vinogradov, we now have an almost complete
solutions to g(k).
Consider the integer
$  %
k
k 3
n=2 − 1 < 3k .
2
56
 j
k 
3 k
The most efficient representation for n is to use 2
− 1 ) many 2k and
$  % !
k
3
n = 2k − 1 + 1k (2k − 1).
2

Thus we obtain a result of Euler that

$  %
k
k 3
g(k) ≥ 2 + − 2.
2

Indeed, the equality holds for all but finitely many k. In fact, the equality holds when
 k $  %
k
3 3
2k { }+ ≤ 2k .
2 2

In 1957, Mahler showed that the above inequality holds for all but finitely many k.
Let G(k) be the least s such that for n sufficiently large we can write n = xk1 + · · · + xks .
In the following, we will establish g(2) = 4. Observe that as x runs over Z/8Z we have
x ≡ 0, 1, 4 (mod 8). Since x21 + x22 + x23 6≡ 7 (mod 8) we see that g(2) ≥ 4.
2

32. November 25
Recall that for k ∈ N with k ≥ 2, let g(k) denote the least integer s such that for all n ∈ N
we have
n = xk1 + xk2 + · · · + xks
with xi ∈ N ∪ {0}. Our goal is to prove that g(2) = 4.
Theorem 63. If p is an odd prime then there exists integers x, y such that 1 + x2 + y 2 = mp
where m ∈ Z with 1 ≤ m ≤ p − 1.
Proof. Consider the sets
 
2 p−1
S1 = x + pZ : x ∈ Z, 0 ≤ x ≤ .
2
and  
2 p−1
S2 = −1 − y + pZ : y ∈ Z, 0 ≤ y ≤ .
2
Note that x21 ≡ x22 (mod p) if and only if x1 ≡ ±x2 (mod p). Since 0 ≤ x ≤ p−1 2
, all elements
p+1
in S1 are distinct, and so are S2 . Since |S1 | = |S2 | = 2 , we have S1 ∩ S2 6= ∅. Thus there
exist x, y ∈ Z with 0 ≤ x, y ≤ p−1 2
such that x2 ≡ −1 − y 2 (mod p), or 1 + x2 + y 2 ≡ 0
2 2
(mod p). Thus 1 + x2 + y 2 = mp for some m ∈ Z. We also have 0 < m < 1+xp+y ≤
2 2
1+( p−1
2 )
+( p−1
2 )
p
< p. 

Theorem 64 (Lagrange). We have g(2) = 4. That is, every natural number can be written
as a sum of at most four squares.
57
Proof. We have the Lagrange identity
(x21 + x22 + x23 + x24 )(y12 + y22 + y32 + y42 ) = (x1 y1 + x2 y2 + x3 y3 + x4 y4 )2
+ (x1 y2 − x2 y1 + x3 y4 − x4 y3 )2
+ (x1 y3 − x3 y1 + x4 y2 − x2 y4 )2
+ (x1 y4 − x4 y1 + x2 y3 − x3 y2 )2 .
We see that the product of two members which are representable as a sum of four squares
is also representable as a sum of four squares. Thus it suffices to prove that every prime
can be written as a sum of four squares. Note that 2 = 12 + 12 + 02 + 02 . Let p be an odd
prime. By Theorem 63, there exist x1 , x2 , x3 , x4 ∈ Z such that x21 + x22 + x23 + x24 = mp with
1 ≤ m ≤ p − 1. Let m0 be the smallest natural number such that m0 p is a sum of four
squares. It remains to show that m0 = 1. Suppose that m0 is even. Note that
X
(x1 + x2 + x3 + x4 )2 = x21 + x22 + x23 + x24 + 2 xi xj .
1≤i<j≤4

Since x21 + x22 + x23 + x24

= m0 p is even, we see that x1 + x2 + x3 + x4 is even also. Thus either
x1 , x2 , x3 , x4 are all even, all odd, or only two of them are even (without loss of generality,
let’s say x1 and x2 are even). In all cases, we see that
x1 + x2 , x1 − x2 , x3 + x4 , x3 − x4
are all even. So it follows that
2  2  2  2
x2 + x22 + x23 + x24

x1 + x2 x1 − x2 x3 + x4 x3 − x4 m0
+ + + = 1 = p.
2 2 2 2 2 2
But this contradicts the minimality of m0 . Hence m0 is odd. Suppose now that m0 > 1.
Since x21 + x22 + x23 + x24 = m0 p and 1 ≤ m0 ≤ p − 1, not all of x1 , x2 , x3 , x4 are divisible by
m0 , for otherwise we would have m20 | m0 p, which is impossible since this would imply m0 | p.
Thus there exist b1 , b2 , b3 , b4 ∈ Z such that yi = xi − bi m0 and |yi | < m20 (1 ≤ i ≤ 4) and not

2
all the yi ’s are zero. Then 0 < y12 + y22 + y32 + y42 < 4 m20 = m20 and y12 + y22 + y32 + y42 ≡ 0
(mod m0 ). Thus there exist m1 ∈ N with m1 < m0 such that
y12 + y22 + y32 + y42 = m0 m1 .
We recall that x21 + x22 + x23 + x24 = m0 p. Multiply the above two equalities together. By the
Lagrange identity, there exist z1 , z2 , z3 , z4 ∈ Z such that
z12 + z22 + z32 + z42 = m20 m1 p,
where z1 = x1 y1 + x2 y2 + x3 y3 + x4 y4 and etc. Since
4
X 4
X
z1 = xi (xi − bi m0 ) = x2i + m0 K
i=1 i=1

for some K ∈ Z. Since x21

+ + + x22 x23 x24
= m0 p we have m0 | z1 . Similarly, we see that
z2 , z3 , z4 are all divisible by m0 . Let ti = zi /m0 (1 ≤ i ≤ 4). Then
t21 + t22 + t23 + t24 = m1 p
and 1 ≤ m1 < m0 , which contradicts the minimality of m0 . Thus m0 = 1 for all odd primes
p. Thus we see that all primes are representable as a sum of four squares. 
58
Theorem 65. g(4) ≤ 53.
Proof. We have the identity
6(a2 + b2 + c2 + d2 )2 = (a + b)4 + (a − b)4 + (c + d)4 + (c − d)4
+ (a + c)4 + (a − c)4 + (b + d)4 + (b − d)4
+ (a + d)4 + (a − d)4 + (b + c)4 + (b − c)4 .
Combining the above identity with Theorem 64, we see that every integer of the form 6x2
can be expressed as a sum of 12 fourth powers. Note that every natural number can be
written in the form 6k + r with k ∈ N ∪ {0} and 0 ≤ r ≤ 5. Then by Theorem 64 we can
write k as a sum of four squares, say k = x21 + x22 + x23 + x24 . Then 6k = 6x21 + 6x22 + 6x23 + 6x24 .
Since each term in the above sum is a sum of 12 fourth powers, 6k can be expressed as a
sum of 48 fourth powers. Finally, we note that r = 14 + · · · + 14 (r times). Since 0 ≤ r ≤ 5,
it follows that 6k + r is a sum of 53 fourth powers as needed. 

33. November 27: Hardy-Littlewood circle method

We recall that the number g(k) is determined by “small” numbers of special form. Thus
a more interesting question is to estimate G(k), defined to be the least integer s = s(k)
such that every sufficiently large integer is the sum of at most k-powers of natural numbers.
Clearly, we have G(k) ≤ g(k). Also, a conjecture states that G(k) = max{k+1, Γ0 (k)} where
Γ0 (k) is the least integers s such that for every prime p and m ∈ N, we have n = xk1 + · · · + xks
has a soultion in mod pm where (x1 , p) = 1. For large k, Wooley proved in 1992 that
G(k) ≤ k log k + O(k log log k).
One can consider a more refined question: for fixed k ∈ N with k ≥ 2, let
Rs (n) = Rs,k (n) = #{n : xk1 + · · · + xks , xi ∈ N, 1 ≤ i ≤ s}.
Note that if the above equality holds, then xi ≤ n1/k . Also the sum xk1 + · · · + xks ranges
from s to sn. Thus we expect that Rs (n) is of size
(n1/k )s · (sn − s)−1  ns/k−1 .
Note that (n1/k )s denotes the choices for x1 , . . . , xs and (sn − s)−1 the probability that their
sum is n. That is, we expect Rs (n) ∼ C(s, k; n)ns/k−1 for some appropriate constant C > 0.
Let G(k)
e be the least integer s = s(k) such that the above asymptotic formula holds for
every sufficiently large integer n. Note that for G(k), we only need Rs (n) > 0. Thus we
have G(k) ≤ G(k).
e To estimate Rs (n), we apply the exponential function. For α ∈ R, let
2πiα
e(α) = e . We have
e(α)e(β) = e(α + β).
Moreover, for h ∈ Z we have the following orthogonal relation
Z 1 (
1 if h = 0
e(αh) dα =
0 0 if h ∈ Z \ {0}.
Define p = n1/k and X
f (α) = e(αxk ).
1≤x≤p
59
It follows that
Z 1 !s
Z 1 X
s
f (α) e(−nα) dα = e(αxk ) e(−nα) dα
0 0 x≤p
X XZ 1
= ··· e(αxk1 ) · · · e(αxks )e(−nα) dα
x1 ≤p xs ≤p 0

X XZ 1
= ··· e(α(xk1 + · · · + xks − n)) dα = Rs (n).
x1 ≤p xs ≤p | 0 {z }
(∗)

Note that (∗) is 1 if n = xk1 + · · · + xks and 0 otherwise. Note that as α runs between 0 and
1, e(α) runs through the unit circl. This is why we call this approach the circle method.
34. November 30
Define G(k)
e the least integer s = s(k) such that the expected asymptotic formula holds
for every sufficiently large n.
Conjecture. G(k)
e = max{k +1, Γ0 (n)} where Γ0 (k) is the least integer s such that for every
prime p and m ∈ N, we have
n = xk1 + · · · + xks
has a solution in mod pm with (x1 , p) = 1.
For α ∈ R, let e(α) := e2πiα . Let p = n1/k and f (α) = e(αxk ). We have seen
P
1≤x≤p
Z 1
Rs (n) = f (α)s e(−nα) dα.
0
Idea is to divide [0, 1) into two parts: major arc M and minor arc m, where M contains
α ∈ [0, 1) that are “close” to a rational number of “small” denominators and m = [0, 1) \ M.
Consider α = aq with (a, q) = 1. Write x = yq + r with 1 ≤ r ≤ q. We have
  q
a X a  X X 
a

k k
e = e x = e (yq + r)
q 1≤xleqp
q r=1 1−r p−r
q
q
≤y≤ q
q k q
ark
   
X X ar pX
= e ∼ e .
r=1
q q r=1 q
1−r
q
≤y≤ p−r
q

We need q ≤ p for the ∼ part to be true. To extend the above estimate to α[0, 1) that is
close enough to aq , we note that
  X    X  
a −k a −k a k
f + cp = e + cp k
x = e x e(cp−k xk ).
q x≤p
q x≤p
q
   
To approximate f aq + cp−k by f aq , we need e(cp−k xk ) to be “close” to 1. Since p−k xk ≤
1, it suffices to choose to be “small”. This motivates the following definition of the major
arcs.
60
Definition 26. Let δ ∈ R with 0 < δ < 51 , and let a, q ∈ N ∪ {0}. We define the major arcs
to be
[
M := M(q, a)
0≤a<q≤pδ
(a,q)=1

where
 
a δ−k
M(q, a) = α ∈ [0, 1) : α − ≤p .
q
The remaining portion m = [0, 1) \ M is the minor arcs. One can show that
Z
s
f (α)s e(−nα) dα ∼ C(s, k, n)k k −1 .
M

A trivial bound for f (α) is

X
|f (α)| = e(αxk ) ≤ p.
x≤p

Suppose that we can show sup |f (α)|  p1−ν where ν = ν(k) > 0. Then it follows that
α∈m

Z  s Z
s
f (α) e(−nα) dα ≤ sup |f (α)| 1 dα  (p1−ν )s .
m α∈m m

Here, we want to get  ps−k−λ for some λ > 0. For this, it suffices to have sν > k + λ. Thus
s ν
(p1−ν )s  ps−k−ν  n k −1− k . Now we need to show sν > k + λ. Write s = 2r + 1. We see
that
Z Z 1
s
f (α) dα  sup |f (α)| |f (α)|2r dα
m α∈m 0
= sup |f (α)|#{xk1 + · · · + xkr = y1k + · · · + yrk , xi , yi ∈ p}
α∈m
1−ν
∼p · p2r−k .

We need to find r sufficiently large, namely r ∼ k 2 .

35. December 2
Let a, q ∈ N ∪ {0}. Suppose that 0 < a < q and (a, q) = 1. For α ∈ M(q, a) ⊆ M we have
q ≤ pδ (0 < δ < 15 ) and
p
α− ≤ pδ−k .
q
61
 a
Write α = q
+ β. Then
X
f (α) = e(αxn )
1≤x≤p
q   
X X a k
= e + β (qy + r)
r=1
q
1−r
q
≤y≤ p−r
q
q   X
X ark
= e e(β(qy + r)k ).
r=1
q 1−r p−r
q
≤y≤ q

Since e( ) is smooth, we have

Z p−r
X q
k
e(β(qy + r) ) ∼ e(β(zq + r)k ) dz
1−r
1−r
q
≤y≤ p−r
q
q

Z p−r
q
∼ e(β(zq + r)k ) dz
− rq
Z p
1
∼ e(βγ k ) dγ,
q 0
where γ := zq + r and dγ = q dz.
q   Rp
e ark and v(β) = 0 e(βγ k ) dγ. Then one can show that for
P
Define S(q, a) := q
r=1
α ∈ M(q, a) ⊆ M,  
1 a
f (α) = S(q, a)v α − + p2δ .
q q
Since f (α) ∼ q −1 S(q, a) we have
Z q−1 Z  
s
X X
s nα
f (α) e(−nα) dα = f (α) e − e(−nβ) dβ
q
1≤q≤pδ a=0 |
M α− aq |=|β|≤pδ−k

q−1  Z
X X
−1 s na
∼ (q S(q, a)) e − v(β)s e(−nβ) dβ.
δ a=0
q |β|≤pr−k
1≤q≤p
(a,q)=1

For Q > 0, define

q−1  
X X
−1 na s
Ss (n, Q) = (q S(q, a)) e −
1≤q≤Q a=0
q
(a,q)=1
and
Q
X
Js (n, Q) = v(β)e(−nβ) dβ.
−Q
Then one can showZ that
f (α)s e(−nα) dα = Ss (n, pδ )Js (n, pδ−k ) + O(ps−k−u )
M
62
for some u > 0.
Define the singular series
q−1
∞ X  
X
−1 na e
Ss (n) = (q S(q, a)) −
q=1 a=0
q
(a,q)=1

and the singular integral Z ∞

Js (n) = v(β)s e(−nβ) dβ.
−∞
k
One can show that for s > 2 there exists w > 0 so that
Z
f (α)s e(−nα) dα = Ss (n)Js (n) + O(ps−k−w ).
M
One can also show that for s ≥ 2 we have

s
Γ 1 + k1
Js (n) =
s ,
Γ ks n k −1
where Γ is the gamma function defined by
Z ∞
Γ(x) := tx−1 e−t dt.
0
It remains to show that 1  Ss (n)  1. One can show that
Y
Ss (n) = σ(p),
p prime

where ∞
X
σ(p) := A(pk , n)
n=0
and
q−1  
X
−1 na s
A(q, n) = (q S(q, a)) e − .
a=0
q
(a,q)=1

Indeed, σ(p) corresponds to the p-adic solutions of xk1 + · · · + xks = n. More precisely,
if we let Mn (q) = #{mk1 + · · · + Msk ≡ n (mod q), 1 ≤ mi ≤ q}, then we have σ(p) =
lim ph(1−s) Mn (ph ) > 0. It follows therefore that
h→∞

s
Γ 1 + k1
Z
s s
s
f (α) e(−nα) dα = Ss (n) n k −1 + O(n k −1−w )
M Γ(s/k)
for some w > 0.
Department of Pure Mathematics, University of Waterloo, 200 University Avenue West,
Waterloo, ON, Canada N2L 3G1
E-mail address: hsyang@uwaterloo.ca

63