The Prime Number Theorem with Error Term

Jason Tang and Richard Chen

Mentor: Chengyang Shao

December 2019

1 / 81
What is the Prime Number Theorem?

Do you think it would be attractive to find an expression that could approximate

the value of the nth prime number?

2 / 81
What is the Prime Number Theorem?

Do you think it would be attractive to find an expression that could approximate

the value of the nth prime number?

Mathematicians have tried many times to find exclusively prime-generating functions:

3 / 81
What is the Prime Number Theorem?

Do you think it would be attractive to find an expression that could approximate

the value of the nth prime number?

Mathematicians have tried many times to find exclusively prime-generating functions:

x2 − 79x + 1601 works for 0, 1, . . . , 79, but fails for x = 80

4 / 81
What is the Prime Number Theorem?

Do you think it would be attractive to find an expression that could approximate

the value of the nth prime number?

Mathematicians have tried many times to find exclusively prime-generating functions:

x2 − 79x + 1601 works for 0, 1, . . . , 79, but fails for x = 80
n
22 + 1 works for 1, 2, . . . , 20, then fails at n = 21

5 / 81
What is the Prime Number Theorem?

Do you think it would be attractive to find an expression that could approximate

the value of the nth prime number?

Mathematicians have tried many times to find exclusively prime-generating functions:

x2 − 79x + 1601 works for 0, 1, . . . , 79, but fails for x = 80
n
22 + 1 works for 1, 2, . . . , 20, then fails at n = 21

On the other hand, one such approximate formula can be deduced from the Prime Number
Theorem.

6 / 81
What is the Prime Number Theorem?

Do you think it would be attractive to find an expression that could approximate

the value of the nth prime number?

Mathematicians have tried many times to find exclusively prime-generating functions:

x2 − 79x + 1601 works for 0, 1, . . . , 79, but fails for x = 80
n
22 + 1 works for 1, 2, . . . , 20, then fails at n = 21

On the other hand, one such approximate formula can be deduced from the Prime Number
Theorem.
Prime number function π(x): Equals the number of primes less than or equal to x

7 / 81
What is the Prime Number Theorem?

Do you think it would be attractive to find an expression that could approximate

the value of the nth prime number?

Mathematicians have tried many times to find exclusively prime-generating functions:

x2 − 79x + 1601 works for 0, 1, . . . , 79, but fails for x = 80
n
22 + 1 works for 1, 2, . . . , 20, then fails at n = 21

On the other hand, one such approximate formula can be deduced from the Prime Number
Theorem.
Prime number function π(x): Equals the number of primes less than or equal to x
π(x) log x
Prime Number Theorem: limx→∞ x
= 1.

8 / 81
What is the Prime Number Theorem?

Do you think it would be attractive to find an expression that could approximate

the value of the nth prime number?

Mathematicians have tried many times to find exclusively prime-generating functions:

x2 − 79x + 1601 works for 0, 1, . . . , 79, but fails for x = 80
n
22 + 1 works for 1, 2, . . . , 20, then fails at n = 21

On the other hand, one such approximate formula can be deduced from the Prime Number
Theorem.
Prime number function π(x): Equals the number of primes less than or equal to x
π(x) log x
Prime Number Theorem: limx→∞ x
= 1.

It follows that the nth prime number should be approximately of the magnitude n log n.

9 / 81
Numerical Results of the Prime Number Theorem

Figure: Ratios for π(x) and its approximations; from Wikipedia

Figure: Absolute error of the approximations of π(x); from Wikipedia

10 / 81
Historical Background

1790s: Gauss and Legendre conjectured independently from large amounts of data that as
π(a)
a → ∞, a/ log a → 1.

11 / 81
Historical Background

1790s: Gauss and Legendre conjectured independently from large amounts of data that as
π(a)
a → ∞, a/ log a → 1.
π(x) log x
1850: Chebyshev introduced the Chebyshev functions, generated bounds for x
,
concluded the limit equals 1 if it exists.

12 / 81
Historical Background

1790s: Gauss and Legendre conjectured independently from large amounts of data that as
π(a)
a → ∞, a/ log a → 1.
π(x) log x
1850: Chebyshev introduced the Chebyshev functions, generated bounds for x
,
concluded the limit equals 1 if it exists.
1859: Riemann linked π(x) with the zeros of the ζ function, obtaining the first explicit
formula – suggested a complex analysis approach.

13 / 81
Historical Background

1790s: Gauss and Legendre conjectured independently from large amounts of data that as
π(a)
a → ∞, a/ log a → 1.
π(x) log x
1850: Chebyshev introduced the Chebyshev functions, generated bounds for x
,
concluded the limit equals 1 if it exists.
1859: Riemann linked π(x) with the zeros of the ζ function, obtaining the first explicit
formula – suggested a complex analysis approach.
1896: Hadamard and de la Vallée Poussin each individually proved properties of Riemann ζ
function that completed the proof of the Prime Number Theorem.

14 / 81
Historical Background

1790s: Gauss and Legendre conjectured independently from large amounts of data that as
π(a)
a → ∞, a/ log a → 1.
π(x) log x
1850: Chebyshev introduced the Chebyshev functions, generated bounds for x
,
concluded the limit equals 1 if it exists.
1859: Riemann linked π(x) with the zeros of the ζ function, obtaining the first explicit
formula – suggested a complex analysis approach.
1896: Hadamard and de la Vallée Poussin each individually proved properties of Riemann ζ
function that completed the proof of the Prime Number Theorem.
Alternate proofs were found in later years, some much simpler or more elementary.

15 / 81
Chebyshev Functions

Definition (von Mangoldt Function)

(
log p if n = pm for m ≥ 1
Λ(n) =
0 otherwise

16 / 81
Chebyshev Functions

Definition (von Mangoldt Function)

(
log p if n = pm for m ≥ 1
Λ(n) =
0 otherwise

Definition (Chebyshev Functions)

Chebyshev ϑ-function: We define X
ϑ(x) = log p,
p≤x

where the sum runs over all prime numbers less than x.
P
Chebyshev ψ-function: ψ(x) = n≤x Λ(n).

17 / 81
Chebyshev Functions

Definition (von Mangoldt Function)

(
log p if n = pm for m ≥ 1
Λ(n) =
0 otherwise

Definition (Chebyshev Functions)

Chebyshev ϑ-function: We define X
ϑ(x) = log p,
p≤x

where the sum runs over all prime numbers less than x.
P
Chebyshev ψ-function: ψ(x) = n≤x Λ(n).

We can rewrite
∞
X X X X X
ψ(x) = log p = log p = ϑ(x1/m ).
m=1 p≤x1/m m≤x p≤x1/m m≤x

18 / 81
Chebyshev Functions

Definition (von Mangoldt Function)

(
log p if n = pm for m ≥ 1
Λ(n) =
0 otherwise

Definition (Chebyshev Functions)

Chebyshev ϑ-function: We define X
ϑ(x) = log p,
p≤x

where the sum runs over all prime numbers less than x.
P
Chebyshev ψ-function: ψ(x) = n≤x Λ(n).

We can rewrite
∞
X X X X X
ψ(x) = log p = log p = ϑ(x1/m ).
m=1 p≤x1/m m≤x p≤x1/m m≤x

By Mobius Inversion, we obtain

X
ϑ(x) = µ(m)ψ(x1/m ).
m≤x
19 / 81
Relation between ϑ(x) and π(x)

P
Abel’s Summation Formula: For an arithmetic function a(n), define A(x) = n≤x a(n).
Suppose f is continuously differentiable on the interval [x, y] for 0 < x < y. Then,
X Z y
a(n)f (n) = A(y)f (y) − A(x)f (x) − A(t)f 0 (t)dt.
x<n≤y x

P
(This formula can be verified directly by expressing x<n≤y a(n)f (n) as an integral and
evaluating by parts)

20 / 81
Relation between ϑ(x) and π(x)

P
Abel’s Summation Formula: For an arithmetic function a(n), define A(x) = n≤x a(n).
Suppose f is continuously differentiable on the interval [x, y] for 0 < x < y. Then,
X Z y
a(n)f (n) = A(y)f (y) − A(x)f (x) − A(t)f 0 (t)dt.
x<n≤y x

P
(This formula can be verified directly by expressing x<n≤y a(n)f (n) as an integral and
evaluating by parts)

(
log n if n is prime 1
We take a(n) = and f (x) = log x
, then
0 otherwise

ϑ(x)
Z x ϑ(t)
π(x) = + dt.
log x 2 t log2 t

21 / 81
Relation between ϑ(x) and π(x)

P
Abel’s Summation Formula: For an arithmetic function a(n), define A(x) = n≤x a(n).
Suppose f is continuously differentiable on the interval [x, y] for 0 < x < y. Then,
X Z y
a(n)f (n) = A(y)f (y) − A(x)f (x) − A(t)f 0 (t)dt.
x<n≤y x

P
(This formula can be verified directly by expressing x<n≤y a(n)f (n) as an integral and
evaluating by parts)

(
log n if n is prime 1
We take a(n) = and f (x) = log x
, then
0 otherwise

ϑ(x)
Z x ϑ(t)
π(x) = + dt.
log x 2 t log2 t
(
1 if n is prime
Furthermore, setting a(n) = and f (x) = log x gives
0 otherwise
Z x π(t)
ϑ(x) = π(x) log x − dt.
2 t

22 / 81
Asymptotic Behaviors of the Chebyshev Functions

Theorem
(1)
π(x) log x ϑ(x)
lim = 1 ⇐⇒ lim = 1.
x→∞ x x→∞ x
(2) √
ψ(x) − ϑ(x) = O( x log2 x).

23 / 81
Asymptotic Behaviors of the Chebyshev Functions

Theorem
(1)
π(x) log x ϑ(x)
lim = 1 ⇐⇒ lim = 1.
x→∞ x x→∞ x
(2) √
ψ(x) − ϑ(x) = O( x log2 x).

We can verify these via direct calculation and using Abel’s Formula.

24 / 81
Asymptotic Behaviors of the Chebyshev Functions

Theorem
(1)
π(x) log x ϑ(x)
lim = 1 ⇐⇒ lim = 1.
x→∞ x x→∞ x
(2) √
ψ(x) − ϑ(x) = O( x log2 x).

We can verify these via direct calculation and using Abel’s Formula.
These relations show that the the prime number theorem can be converted to the study of
ψ(x) π(x) log x
the ψ function as if limx→∞ x = 1, then limx→∞ x
= 1.

25 / 81
Riemann Zeta Function

We now introduce the Riemann zeta function, whose distribution of zeros is connected later to
the explicit formula of the Chebyshev ψ function.

Definition (Riemann zeta function)

∞
X 1
ζ(s) = s
, s = σ + it, σ > 1.
n=1
n

26 / 81
Riemann Zeta Function

We now introduce the Riemann zeta function, whose distribution of zeros is connected later to
the explicit formula of the Chebyshev ψ function.

Definition (Riemann zeta function)

∞
X 1
ζ(s) = s
, s = σ + it, σ > 1.
n=1
n

Euler additionally found a product form in which the ζ function could be expressed. It is an
elegant rephrasing of the unique factorization property of integers:

Theorem (Euler Product)

Y 1
ζ(s) = , σ > 1.
p: prime
1 − p−s

27 / 81
Riemann Zeta Function

We now introduce the Riemann zeta function, whose distribution of zeros is connected later to
the explicit formula of the Chebyshev ψ function.

Definition (Riemann zeta function)

∞
X 1
ζ(s) = s
, s = σ + it, σ > 1.
n=1
n

Euler additionally found a product form in which the ζ function could be expressed. It is an
elegant rephrasing of the unique factorization property of integers:

Theorem (Euler Product)

Y 1
ζ(s) = , σ > 1.
p: prime
1 − p−s

Observe that since σ > 1, we may express each term in the right product as an infinite
geometric series.

28 / 81
Riemann Zeta Function

We now introduce the Riemann zeta function, whose distribution of zeros is connected later to
the explicit formula of the Chebyshev ψ function.

Definition (Riemann zeta function)

∞
X 1
ζ(s) = s
, s = σ + it, σ > 1.
n=1
n

Euler additionally found a product form in which the ζ function could be expressed. It is an
elegant rephrasing of the unique factorization property of integers:

Theorem (Euler Product)

Y 1
ζ(s) = , σ > 1.
p: prime
1 − p−s

Observe that since σ > 1, we may express each term in the right product as an infinite
geometric series.
The product converges absolutely if σ > 1 so we may use the distribution law. Each term in
ζ(s) can be expressed as a product of terms from the geometric series.

29 / 81
Logarithmic Derivative

The Logarithmic Derivative:

∞ ∞
ζ0 d Y 1 X X
−ns
X Λ(n)
(s) = log −s
= − log(p) p = − . (1)
ζ ds p: prime
1 − p p: prime n=1 n=1
ns

30 / 81
Logarithmic Derivative

The Logarithmic Derivative:

∞ ∞
ζ0 d Y 1 X X
−ns
X Λ(n)
(s) = log −s
= − log(p) p = − . (1)
ζ ds p: prime
1 − p p: prime n=1 n=1
ns

Heuristically, from this equation we can understand why the logarithmic derivative is related
to the ψ function as both can be expressed as a sum using the von Mangoldt function: ψ(x)
Λ(n)
is nothing but the partial sum of the series ∞
P
n=1 ns when s = 0.

31 / 81
Logarithmic Derivative

The Logarithmic Derivative:

∞ ∞
ζ0 d Y 1 X X
−ns
X Λ(n)
(s) = log −s
= − log(p) p = − . (1)
ζ ds p: prime
1 − p p: prime n=1 n=1
ns

Heuristically, from this equation we can understand why the logarithmic derivative is related
to the ψ function as both can be expressed as a sum using the von Mangoldt function: ψ(x)
Λ(n)
is nothing but the partial sum of the series ∞
P
n=1 ns when s = 0.
The precise relation is revealed using Perron’s formula: roughly speaking,
Z b+iT
X f (n) 1 xs
s
− F (s0 + s) ds + Error terms.
n<x
n 0 2πi b−iT s

32 / 81
Perron’s Formula

Theorem (Perron’s Formula)

f (n)
Let F (s) = ∞
P
n=1 ns be a Dirichlet series and let σa be the absolutely convergent coordinate
of F . Let s0 = σ0 + it0 , and let b be a positive number such that σ0 + b > σa . Suppose there is
a function B(σ) and increasing function H(σ) such that |f (n)| ≤ H(n) and
P∞ |f (n)|
n=1 nσ ≤ B(σ). Then for any half integer x > 2 and any T > 2,
 Z b+iT 
 X f (n) 1 xs  10xb B(σ0 + b) log x
− F (s0 + s) ds ≤ + 100 · 2b+σ0 x1−σ0 H(2x)

s0


n<x
n 2πi b−iT s  T T

We only present an outline of the proof, which consists mostly of direct computation:

33 / 81
Perron’s Formula

Theorem (Perron’s Formula)

f (n)
Let F (s) = ∞
P
n=1 ns be a Dirichlet series and let σa be the absolutely convergent coordinate
of F . Let s0 = σ0 + it0 , and let b be a positive number such that σ0 + b > σa . Suppose there is
a function B(σ) and increasing function H(σ) such that |f (n)| ≤ H(n) and
P∞ |f (n)|
n=1 nσ ≤ B(σ). Then for any half integer x > 2 and any T > 2,
 Z b+iT 
 X f (n) 1 xs  10xb B(σ0 + b) log x
− F (s0 + s) ds ≤ + 100 · 2b+σ0 x1−σ0 H(2x)

s0


n<x
n 2πi b−iT s  T T

We only present an outline of the proof, which consists mostly of direct computation:
The Dirichlet series converges absolutely and uniformly for σ0 + b > σa , so
b+iT ∞
F (s0 + s)xs 1 X f (n) b+iT  x s 1
Z Z
1
ds = ds
2πi b−iT s 2πi n=1 ns0 b−iT n s

34 / 81
Perron’s Formula

Theorem (Perron’s Formula)

f (n)
Let F (s) = ∞
P
n=1 ns be a Dirichlet series and let σa be the absolutely convergent coordinate
of F . Let s0 = σ0 + it0 , and let b be a positive number such that σ0 + b > σa . Suppose there is
a function B(σ) and increasing function H(σ) such that |f (n)| ≤ H(n) and
P∞ |f (n)|
n=1 nσ ≤ B(σ). Then for any half integer x > 2 and any T > 2,
 Z b+iT 
 X f (n) 1 xs  10xb B(σ0 + b) log x
− F (s0 + s) ds ≤ + 100 · 2b+σ0 x1−σ0 H(2x)

s0


n<x
n 2πi b−iT s  T T

We only present an outline of the proof, which consists mostly of direct computation:
The Dirichlet series converges absolutely and uniformly for σ0 + b > σa , so
b+iT ∞
F (s0 + s)xs 1 X f (n) b+iT  x s 1
Z Z
1
ds = ds
2πi b−iT s 2πi n=1 ns0 b−iT n s

We split the above sum into n < x and n > x and bound each from above; using a
rectangular contour, the integral in the sum may be evaluated and then bound.

35 / 81
Perron’s Formula

Theorem (Perron’s Formula)

f (n)
Let F (s) = ∞
P
n=1 ns be a Dirichlet series and let σa be the absolutely convergent coordinate
of F . Let s0 = σ0 + it0 , and let b be a positive number such that σ0 + b > σa . Suppose there is
a function B(σ) and increasing function H(σ) such that |f (n)| ≤ H(n) and
P∞ |f (n)|
n=1 nσ ≤ B(σ). Then for any half integer x > 2 and any T > 2,
 Z b+iT 
 X f (n) 1 xs  10xb B(σ0 + b) log x
− F (s0 + s) ds ≤ + 100 · 2b+σ0 x1−σ0 H(2x)

s0


n<x
n 2πi b−iT s  T T

We only present an outline of the proof, which consists mostly of direct computation:
The Dirichlet series converges absolutely and uniformly for σ0 + b > σa , so
b+iT ∞
F (s0 + s)xs 1 X f (n) b+iT  x s 1
Z Z
1
ds = ds
2πi b−iT s 2πi n=1 ns0 b−iT n s

We split the above sum into n < x and n > x and bound each from above; using a
rectangular contour, the integral in the sum may be evaluated and then bound.
We combine the n < x and n > x sums via the Triangle Inequality.

36 / 81
Perron’s Formula: Application to ψ(x)

We now apply Perron’s Formula to ψ(x).

0 Λ(n)
F (s) = ζζ (s) = − ∞
P
n=1 ns with σa ≤ 1
10 1
H(n) = log n, B(σ) = σ−1
, s0 = 0, b = 1 + log x

Perron’s Formula for half integer x ≥ 2 and T ≥ 2 :

b+iT ζ0 xs
Z
X 1
ψ(x) = Λ(n) = − (s) ds + R(x; T ) (2)
n<x
2πi b−iT ζ s

x log2 x
where R(x; T ) << T
.

37 / 81
Entire Functions

Definition (Entire Function and Order)

An entire function f is a function holomorphic over all of C. We define

Mf (r) = max |f (z)|.

|z|=r

λ
If there exists A, B, λ ≥ 0 such that Mf (r) ≤ AeBr , then we say that f has order ≤ λ. The
infimum of all λs such that this inequality hold is called the order of f .

38 / 81
Entire Functions

Definition (Entire Function and Order)

An entire function f is a function holomorphic over all of C. We define

Mf (r) = max |f (z)|.

|z|=r

λ
If there exists A, B, λ ≥ 0 such that Mf (r) ≤ AeBr , then we say that f has order ≤ λ. The
infimum of all λs such that this inequality hold is called the order of f .

Polynomials are entire functions of order zero.

39 / 81
Entire Functions

Definition (Entire Function and Order)

An entire function f is a function holomorphic over all of C. We define

Mf (r) = max |f (z)|.

|z|=r

λ
If there exists A, B, λ ≥ 0 such that Mf (r) ≤ AeBr , then we say that f has order ≤ λ. The
infimum of all λs such that this inequality hold is called the order of f .

Polynomials are entire functions of order zero.

The exponential function is an entire function of order one. More generally,

exp (a0 + a1 z + · · · + an z n )

is an entire function of order n.

40 / 81
Entire Functions

Definition (Entire Function and Order)

An entire function f is a function holomorphic over all of C. We define

Mf (r) = max |f (z)|.

|z|=r

λ
If there exists A, B, λ ≥ 0 such that Mf (r) ≤ AeBr , then we say that f has order ≤ λ. The
infimum of all λs such that this inequality hold is called the order of f .

Polynomials are entire functions of order zero.

The exponential function is an entire function of order one. More generally,

exp (a0 + a1 z + · · · + an z n )

is an entire function of order n.

z
ee is not of finite order.

41 / 81
Entire Functions

Definition (Entire Function and Order)

An entire function f is a function holomorphic over all of C. We define

Mf (r) = max |f (z)|.

|z|=r

λ
If there exists A, B, λ ≥ 0 such that Mf (r) ≤ AeBr , then we say that f has order ≤ λ. The
infimum of all λs such that this inequality hold is called the order of f .

Polynomials are entire functions of order zero.

The exponential function is an entire function of order one. More generally,

exp (a0 + a1 z + · · · + an z n )

is an entire function of order n.

z
ee is not of finite order.

1/Γ is an entire function of order one, but the inequality 1/|Γ(z)| ≤ A + eB|z| can never hold
for any finite A, B.

42 / 81
Hadamard Factorization Theorem

We now introduce the Hadamard Factorization Theorem, which is necessary for any information
concerning the distribution of zeros of ζ(s).

Theorem (Hadamard Factorization Theorem)

Let f be an entire function of order λ and f (0) 6= 0. Let {an } be the zeros of f , and let
"  #
z 2 z p
     
z z 1 1
Ep (z; an ) = 1 − exp + + ··· +
an an 2 an p an

for p = bλc. Then, for some polynomial q of degree less than or equal to p,
∞
Y
f (z) = eq(z) Ep (z; an ),
n=1

where the infinite product converges absolutely and uniformly on compact subsets of the complex
plane to an entire function.

Proof consists of showing that the product converges and the order of q is bounded

Uses Jensen’s Inequality and Hadamard’s corollary in the bounding portion of the proof

43 / 81
Hadamard Factorization Theorem

Factorization of polynomials:
n  
Y z
a0 + a1 z + · · · + an z n = C 1− .
k=1
zk

44 / 81
Hadamard Factorization Theorem

Factorization of polynomials:
n  
Y z
a0 + a1 z + · · · + an z n = C 1− .
k=1
zk

Trigonometric function:
Y z2

sin z = z 1− .
π 2 n2

45 / 81
Hadamard Factorization Theorem

Factorization of polynomials:
n  
Y z
a0 + a1 z + · · · + an z n = C 1− .
k=1
zk

Trigonometric function:
Y z2

sin z = z 1− .
π 2 n2

The Γ function:
∞ 
1 Y z  −z/n
= zeγz 1+ e .
Γ(z) n=1
n

46 / 81
Hadamard Factorization Theorem

Factorization of polynomials:
n  
Y z
a0 + a1 z + · · · + an z n = C 1− .
k=1
zk

Trigonometric function:
Y z2

sin z = z 1− .
π 2 n2

The Γ function:
∞ 
1 Y z  −z/n
= zeγz 1+ e .
Γ(z) n=1
n

We deduce from these expressions the partial fraction developments:

∞
1 X 1
cot(z) = + 2z 2 − (kπ)2
z k=1
z

∞ 
Γ0

1 X 1 1
(z) = −γ − + − .
Γ z n=1 n z+n

47 / 81
Functional Equation of ζ

By direct calculation starting with ζ(s)Γ(s), the ζ function is analytically continued beyond
σ > 1, except for the pole at s = 1:

The functional equation for ζ function:

 πs 
ζ(s) = 2(2π)s−1 Γ(1 − s) sin ζ(1 − s). (3)
2

48 / 81
Functional Equation of ζ

By direct calculation starting with ζ(s)Γ(s), the ζ function is analytically continued beyond
σ > 1, except for the pole at s = 1:

The functional equation for ζ function:

 πs 
ζ(s) = 2(2π)s−1 Γ(1 − s) sin ζ(1 − s). (3)
2

Introduce an auxilliary function:

1 s
ξ(s) = s(s − 1)π −s/2 Γ ζ(s).
2 2

49 / 81
Functional Equation of ζ

By direct calculation starting with ζ(s)Γ(s), the ζ function is analytically continued beyond
σ > 1, except for the pole at s = 1:

The functional equation for ζ function:

 πs 
ζ(s) = 2(2π)s−1 Γ(1 − s) sin ζ(1 − s). (3)
2

Introduce an auxilliary function:

1 s
ξ(s) = s(s − 1)π −s/2 Γ ζ(s).
2 2
The ξ function has nice properties, in that the factor s(s − 1)Γ(s/2) cancels out all the trivial
zeros and the pole of ζ(s), and ξ(s) = ξ(1 − s). Therefore, ξ(s) is an entire function of order 1.

50 / 81
Functional Equation of ζ

By direct calculation starting with ζ(s)Γ(s), the ζ function is analytically continued beyond
σ > 1, except for the pole at s = 1:

The functional equation for ζ function:

 πs 
ζ(s) = 2(2π)s−1 Γ(1 − s) sin ζ(1 − s). (3)
2

Introduce an auxilliary function:

1 s
ξ(s) = s(s − 1)π −s/2 Γ ζ(s).
2 2
The ξ function has nice properties, in that the factor s(s − 1)Γ(s/2) cancels out all the trivial
zeros and the pole of ζ(s), and ξ(s) = ξ(1 − s). Therefore, ξ(s) is an entire function of order 1.

Because |ξ(s)| ≤ A + eB|s| cannot hold for finite A, B, the Hadamard Factorization Theorem
implies that ξ(s) must have infinitely many zeros.

Consequently, ζ(s) must have infinitely many nontrivial zeros.

51 / 81
Zeros of the ζ Function

ζ(s) has a first order pole at s = 1 and residue 1. It is holomorphic for s 6= 1.

Figure: Zeros of the ζ function

52 / 81
Zeros of the ζ Function

ζ(s) has a first order pole at s = 1 and residue 1. It is holomorphic for s 6= 1.

ζ(s) only has zeros at negative even integers for σ < 0. Additionally, ζ(s) 6= 0 when σ > 1

Figure: Zeros of the ζ function

53 / 81
Zeros of the ζ Function

ζ(s) has a first order pole at s = 1 and residue 1. It is holomorphic for s 6= 1.

ζ(s) only has zeros at negative even integers for σ < 0. Additionally, ζ(s) 6= 0 when σ > 1

Zeros in the strip 0 ≤ σ ≤ 1 are called nontrivial zeros. They are symmetric with respect to the
real axis and the vertical line σ = 1/2. They will be denoted ρ = β + iγ.

Figure: Zeros of the ζ function

54 / 81
Zeros of the ζ Function

The Hadamard Factorization Theorem asserts that

Y s

ξ(s) = eAs+B 1− es/ρ .
ρ
ρ

55 / 81
Zeros of the ζ Function

The Hadamard Factorization Theorem asserts that

Y s

ξ(s) = eAs+B 1− es/ρ .
ρ
ρ

Consequently, the following factorization is valid:

56 / 81
Zeros of the ζ Function

The Hadamard Factorization Theorem asserts that

Y s

ξ(s) = eAs+B 1− es/ρ .
ρ
ρ

Consequently, the following factorization is valid:

∞
eA+Ds Y 
 
s  − s Y s
ζ(s) = 1+ e 2n 1− es/ρ .
s − 1 n=1 2n ρ
ρ

57 / 81
Zeros of the ζ Function

The Hadamard Factorization Theorem asserts that

Y s

ξ(s) = eAs+B 1− es/ρ .
ρ
ρ

Consequently, the following factorization is valid:

∞
eA+Ds Y 
 
s  − s Y s
ζ(s) = 1+ e 2n 1− es/ρ .
s − 1 n=1 2n ρ
ρ

Taking the logarithmic derivative gives us that

∞ 
ζ0
 X 
1 X 1 1 1 1
(s) = D − + − + + .
ζ s − 1 n=1 s + 2n 2n ρ
s−ρ ρ

58 / 81
Zeros of the ζ Function

The Hadamard Factorization Theorem asserts that

Y s

ξ(s) = eAs+B 1− es/ρ .
ρ
ρ

Consequently, the following factorization is valid:

∞
eA+Ds Y 
 
s  − s Y s
ζ(s) = 1+ e 2n 1− es/ρ .
s − 1 n=1 2n ρ
ρ

Taking the logarithmic derivative gives us that

∞ 
ζ0
 X 
1 X 1 1 1 1
(s) = D − + − + + .
ζ s − 1 n=1 s + 2n 2n ρ
s−ρ ρ

We can now bound the first sum and thus obtain the following:

ζ0 X 1   
1 1 1
(s) = − + + +O + log(|s| + 2) ,
ζ s−1 ρ
s−ρ ρ λ(s)

where λ(s) = minn≥1 |s + 2n|.

59 / 81
Formula for ζ 0 /ζ

By the equation from the previous slide, we have an estimate for the sum over all the zeros.
Therefore, we obtain the following:

Theorem (Asymptotic Formula for ζ 0 /ζ)

The following asymptotic formula holds for any s ∈ C:

ζ0 X  1   
1 1 1
(s) = − + + +O + log(|s| + 2) .
ζ s−1 s−ρ ρ λ(s)
ρ:|γ−t|≤1

60 / 81
Formula for ζ 0 /ζ

By the equation from the previous slide, we have an estimate for the sum over all the zeros.
Therefore, we obtain the following:

Theorem (Asymptotic Formula for ζ 0 /ζ)

The following asymptotic formula holds for any s ∈ C:

ζ0 X  1   
1 1 1
(s) = − + + +O + log(|s| + 2) .
ζ s−1 s−ρ ρ λ(s)
ρ:|γ−t|≤1

Note that we have simplified the formula for ζ 0 /ζ greatly – previously, the sum was over all zeros
of ζ(s), whereas now it is only over zeros close to t.

61 / 81
Formula for ζ 0 /ζ

By the equation from the previous slide, we have an estimate for the sum over all the zeros.
Therefore, we obtain the following:

Theorem (Asymptotic Formula for ζ 0 /ζ)

The following asymptotic formula holds for any s ∈ C:

ζ0 X  1   
1 1 1
(s) = − + + +O + log(|s| + 2) .
ζ s−1 s−ρ ρ λ(s)
ρ:|γ−t|≤1

Note that we have simplified the formula for ζ 0 /ζ greatly – previously, the sum was over all zeros
of ζ(s), whereas now it is only over zeros close to t.
We now introduce an important corollary.

Corollary
ζ0
For every T ≥ 2, there exists T 0 ∈ [T, T + 1] such that ζ
(σ + iT 0 ) << log2 |σ + iT | for every
σ ∈ R.

62 / 81
Zero Free Region

Theorem (Zero-free region of ζ(s); due to de la Vallée Poussin)

(1) ζ(1 + it) 6= 0 for any real number t;
A
(2) There is a constant A > 0 such that ζ(s) is zero-free for σ ≥ 1 − log t
,t ≥ 2, shown as the
shaded region in the following figure.

Figure: Zero-free region of ζ(s)

63 / 81
Zero Free Region

Let s = σ + it with σ > 1, ρ = β + iγ. Then,

∞
ζ0 X X log p
(s) = − (cos (nt log p) − i sin (nt log p)).
ζ p n=1
pnσ

64 / 81
Zero Free Region

Let s = σ + it with σ > 1, ρ = β + iγ. Then,

∞
ζ0 X X log p
(s) = − (cos (nt log p) − i sin (nt log p)).
ζ p n=1
pnσ

We utilize the identity 3 + 4 cos θ + cos 2θ = 2(1 + cos θ)2 ≥ 0. Letting nt log p = θ, and taking
real part,

ζ0 ζ0 ζ0
− 3Re (σ) − 4Re (σ + it) − Re (σ + 2it)
ζ ζ ζ
∞
XX log p
= (3 + 4 cos (nt log p) + cos (2nt log p)) ≥ 0.
p n=1
pnσ

65 / 81
Zero Free Region

Let s = σ + it with σ > 1, ρ = β + iγ. Then,

∞
ζ0 X X log p
(s) = − (cos (nt log p) − i sin (nt log p)).
ζ p n=1
pnσ

We utilize the identity 3 + 4 cos θ + cos 2θ = 2(1 + cos θ)2 ≥ 0. Letting nt log p = θ, and taking
real part,

ζ0 ζ0 ζ0
− 3Re (σ) − 4Re (σ + it) − Re (σ + 2it)
ζ ζ ζ
∞
XX log p
= (3 + 4 cos (nt log p) + cos (2nt log p)) ≥ 0.
p n=1
pnσ

If we let σ → 1+ , by using the asymptotic formula attained previously we obtain that for any
non-trivial zero ρ = β + iγ,

A
1−β ≥ .
log γ

66 / 81
Explicit Formula for ψ(x)
We start with our expression derived from Perron’s Formula:
b+iT ζ0 xs x log2 x
Z
X 1
ψ(x) = Λ(n) = − (s) ds + O( ).
n<x
2πi b−iT ζ s T

67 / 81
Explicit Formula for ψ(x)
We start with our expression derived from Perron’s Formula:
b+iT ζ0 xs x log2 x
Z
X 1
ψ(x) = Λ(n) = − (s) ds + O( ).
n<x
2πi b−iT ζ s T

68 / 81
Explicit Formula for ψ(x)
We start with our expression derived from Perron’s Formula:
b+iT ζ0 xs x log2 x
Z
X 1
ψ(x) = Λ(n) = − (s) ds + O( ).
n<x
2πi b−iT ζ s T

By the residue theorem,

ζ0
b+iT xs
Z
1
−
(s) ds
2πi b−iT ζ s
(K−1)/2
ζ0 xs ζ0 xρ x−2n
Z Z Z 
1 X X
=− + + − (s) ds + x − (0) + + .
2πi c2 c3 c4 ζ s ζ ρ n=1
2n
ρ:|γ|≤T

69 / 81
Explicit Formula for ψ(x)

We just obtained from our contour:

ζ0
b+iT xs
Z
1
−
(s) ds
2πi b−iT ζ s
k
ζ0 xs ζ0 xρ x−2n
Z Z Z 
1 X X
=− + + − (s) ds + x − (0) + + .
2πi c2 c3 c4 ζ s ζ ρ n=1
2n
ρ:|γ|≤T

70 / 81
Explicit Formula for ψ(x)

We just obtained from our contour:

ζ0
b+iT xs
Z
1
−
(s) ds
2πi b−iT ζ s
k
ζ0 xs ζ0 xρ x−2n
Z Z Z 
1 X X
=− + + − (s) ds + x − (0) + + .
2πi c2 c3 c4 ζ s ζ ρ n=1
2n
ρ:|γ|≤T

ζ0
From a corollary above, we can always find a T 0 ∈ [T, T + 1] that has ζ
(σ + iT 0 ) = O(log2 T ).

71 / 81
Explicit Formula for ψ(x)

We just obtained from our contour:

ζ0
b+iT xs
Z
1
−
(s) ds
2πi b−iT ζ s
k
ζ0 xs ζ0 xρ x−2n
Z Z Z 
1 X X
=− + + − (s) ds + x − (0) + + .
2πi c2 c3 c4 ζ s ζ ρ n=1
2n
ρ:|γ|≤T

0
From a corollary above, we can always find a T 0 ∈ [T, T + 1] that has ζζ (σ + iT 0 ) = O(log2 T ).
Thus, we can bound the integrals from the previous expression, arriving at the following explicit
formula.

Explicit formula of prime numbers:

ζ0 xρ x log2 T x log2 x
   
X 1 1
ψ(x) = x − (0) + + log 1 − 2 + O +
ζ ρ 2 x T log x T
ρ:|γ|≤T

72 / 81
Proof of the Prime Number Theorem

We focus on estimating
 
ρ
 X 

 x .
ρ:|γ|≤T ρ 
 

73 / 81
Proof of the Prime Number Theorem

We focus on estimating
 
ρ
 X 

 x .
ρ:|γ|≤T ρ 
 

Because of our restrictions on the zero-free region,

 
xρ 
 X 
 − Alog
log x X 1
 ≤ xe
 T

ρ:|γ|≤T ρ |ρ|
 ρ:|γ|≤T
[T ]+1
− Alog
log x X X N (k + 1) − N (k)
<< xe T

k=1 ρ:k<|γ|≤k+1
k

[T ]+1
− Alog
log x X log k
<< xe T

k=1
k
− Alog
log x
<< xe T log2 T.

74 / 81
Proof of the Prime Number Theorem

√
Take T = e log x + O(1). We can thus adjust our explicit formula for ψ(x) to be
√
ψ(x) = x + O(xe−c log x
).

Hence √
ϑ(x) = x + O(xe−c log x
),

75 / 81
Proof of the Prime Number Theorem

√
Take T = e log x + O(1). We can thus adjust our explicit formula for ψ(x) to be
√
ψ(x) = x + O(xe−c log x
).

Hence √
ϑ(x) = x + O(xe−c log x
),

Finally,

ϑ(y)
Z x ϑ(y)
π(x) = + dy
log x 2 y log2 y
√
O(xe−c log x )
Z x
x 1
= + 2
dy +
log x 2 log y log x
√
Z x
O(ye −c log y )
+ dy
2 y log2 y
√
= Li(x) + O(xe−c log x
).

76 / 81
Applications

Expression for π(x)

There is a constant c such that
√
π(x) = Li(x) + O(xe−c log x
).

77 / 81
Applications

Expression for π(x)

There is a constant c such that
√
π(x) = Li(x) + O(xe−c log x
).

This version of Prime Number Theorem clearly indicates that Li(x) approximates π(x) much
better than x/ log x. In fact, for any fixed n,
n  
x X k!x x
Li(x) = + + O .
log x k=2 logk x logn+1 x

78 / 81
Applications

Expression for π(x)

There is a constant c such that
√
π(x) = Li(x) + O(xe−c log x
).

This version of Prime Number Theorem clearly indicates that Li(x) approximates π(x) much
better than x/ log x. In fact, for any fixed n,
n  
x X k!x x
Li(x) = + + O .
log x k=2 logk x logn+1 x

Additionally, the explicit formula for ψ(x) suggests that the distribution of zeros of ζ(s) is
equivalent to the distribution of prime numbers.

79 / 81
Applications

Expression for π(x)

There is a constant c such that
√
π(x) = Li(x) + O(xe−c log x
).

This version of Prime Number Theorem clearly indicates that Li(x) approximates π(x) much
better than x/ log x. In fact, for any fixed n,
n  
x X k!x x
Li(x) = + + O .
log x k=2 logk x logn+1 x

Additionally, the explicit formula for ψ(x) suggests that the distribution of zeros of ζ(s) is
equivalent to the distribution of prime numbers.

Riemann’s hypothesis asserts that the nontrivial zeros are always on the line Re(s) = 1/2. If this
is true, it follows easily from the explicit formula that
√
π(x) = Li(x) + O( x log2 x),

the optimal result on distribution of prime numbers.

80 / 81
Acknowledgements

Our mentor, Chengyang Shao

MIT PRIMES

Our parents

81 / 81