The Prime Number Theorem and Dirichlet's Theorem

Prime Number Theorem: π(x) ~ x/log x asymptotically; probability a random integer near x is prime ≈ 1/log x
Proved 1896 by Hadamard and de la Vallée Poussin using complex analysis; conjectured by Gauss and Legendre ~1800
Dirichlet's Theorem (1837): primes equidistribute among invertible residue classes modulo any integer in arithmetic progressions
Launched analytic number theory by resolving discrete arithmetic problems via continuous analysis and zeta functions
Riemann zeta function and Dirichlet L-functions connect prime distribution to complex analysis; Riemann Hypothesis remains open

The Prime Number Theorem and Dirichlet's Theorem

Overview

The distribution of prime numbers among the integers represents one of the deepest mysteries in mathematics, transforming from empirical observation to rigorous theory through the nineteenth century. The Prime Number Theorem (PNT) and Dirichlet's Theorem on Primes in Arithmetic Progressions together establish the fundamental asymptotic laws governing how primes are distributed, both globally and within arithmetic structures. While individual primes appear chaotic, these theorems reveal profound regularity in their collective behavior.

The Prime Number Theorem asserts that the number of primes up to $x$ , denoted $π (x)$ , is asymptotic to $x / lo g x$ , meaning the "probability" that a random integer near $x$ is prime is approximately $1/ lo g x$ . This result, conjectured by Gauss and Legendre around 1800 based on numerical evidence, remained unproved until 1896 when Hadamard and de la Vallée Poussin independently established it using complex analysis. Dirichlet's earlier theorem (1837) showed that primes are equidistributed among the invertible residue classes modulo any integer, establishing arithmetic progressions as the natural setting for understanding prime distribution.

These theorems inaugurated analytic number theory, demonstrating that questions about discrete arithmetic objects could be resolved through continuous analysis. The interplay between the Riemann zeta function, Dirichlet L-functions, and prime distribution remains one of the most fertile areas in mathematics, with the Riemann Hypothesis representing the most famous open problem concerning the finer structure of prime distribution.

Primes, though individually unpredictable, obey precise statistical laws: globally, their density near $x$ is $1/ lo g x$ , and locally, they distribute equally among all allowable arithmetic progressions. The bridge between primes and analysis is built through generating functions that encode arithmetic data in their analytic properties.

The Prime Number Theorem and Dirichlet's Theorem:

(Prime Number Theorem) Let $π (x) = # {p \leq x : p prime}$ denote the prime counting function. Then

π (x) \sim \frac{x}{lo g x} as x \to \infty,

meaning

lim_{x \to \infty} \frac{π ( x )}{x / l o g x} = 1

(Dirichlet's Theorem) Let $a, q \in Z$ with $g cd (a, q) = 1$ . Then the arithmetic progression ${a + n q : n \in N}$ contains infinitely many primes. More precisely, let $π (x; q, a) = # {p \leq x : p prime, p \equiv a (mod q)}$ . Then

π (x; q, a) \sim \frac{1}{ϕ ( q )} \cdot \frac{x}{lo g x} as x \to \infty,

where

ϕ (q)

is the Euler totient function.

Core Theory

Main Result

Theorem [1] (Prime Number Theorem — Analytic Form): The Riemann zeta function

ζ (s) = \sum_{n = 1}^{\infty} n^{- s}

for

ℜ (s) > 1

extends to a meromorphic function on

C

with a simple pole at

s = 1

, and satisfies

ζ (s) \neq = 0

for

ℜ (s) = 1

. This implies

π (x) \sim x / lo g x

. Theorem [2] (Dirichlet's Theorem): For

g cd (a, q) = 1

, let

χ

be a Dirichlet character modulo

q

, and define the Dirichlet L-function

L (s, χ) = n = 1 \sum \infty \frac{χ ( n )}{n ^{s}}, ℜ (s) > 1.

Then

L (1, χ) \neq = 0

for all non-principal characters

χ

, which implies infinitely many primes in each residue class coprime to

q

. Proof of Theorem [1] (Sketch of classical approach): Step 1: Euler product and logarithmic connection. For

ℜ (s) > 1

, the Euler product formula gives

ζ (s) = p prime \prod \frac{1}{1 - p ^{- s}} .

Taking logarithms,

lo g ζ (s) = - p \sum lo g (1 - p^{- s}) = p \sum (p^{- s} + O (p^{- 2 s})) = p \sum p^{- s} + O (1) .

Step 2: Connection to prime counting. Define the Chebyshev function

ψ (x) = \sum_{p^{k} \leq x} lo g p

. By Perron's formula and Contour Integration,

ψ (x) = \frac{1}{2 π i} \int_{c - i \infty}^{c + i \infty} (- \frac{ζ ^{'} ( s )}{ζ ( s )}) \frac{x ^{s}}{s} d s

for suitable

c > 1

. The PNT is equivalent to

ψ (x) \sim x

. Step 3: Analytic continuation. The functional equation

π^{- s /2} Γ (s /2) ζ (s) = π^{- (1 - s) /2} Γ ((1 - s) /2) ζ (1 - s)

shows

ζ (s)

extends meromorphically to

C

with only a simple pole at

s = 1

(residue 1). Step 4: Zero-free region. The critical step: showing

ζ (s) \neq = 0

ℜ (s) = 1

. For

σ > 1

, the inequality

ζ (σ)^{3} ∣ ζ (σ + i t) ∣^{4} ∣ ζ (σ + 2 i t) ∣ \geq 1

follows from

3 + 4 cos θ + cos 2 θ \geq 0

. As

σ \to 1^{+}

, if

ζ (1 + i t) = 0

, the left side would tend to 0, contradiction. Step 5: Shifting the contour. The zero-free region on

ℜ (s) = 1

allows shifting the integration contour in Step 2 to

ℜ (s) = 1

, picking up the residue at

s = 1

(which is

x

) and showing error terms are

o (x)

. Thus

ψ (x) \sim x

, which implies

π (x) \sim x / lo g x

by partial summation. Proof of Theorem [2] (Sketch): Step 1: Orthogonality of characters. The Dirichlet characters modulo

q

form an orthogonal basis for class functions on

(Z / q Z)^{*}

. By orthogonality,

χ mod q \sum \overline{χ (a)} L (s, χ) = n \equiv a (mod q) \sum \frac{1}{n ^{s}} .

Step 2: Showing $L (1, χ) \neq = 0$ . For the principal character

χ_{0}

, we have

L (s, χ_{0}) = ζ (s) \prod_{p ∣ q} (1 - p^{- s})

, which has a pole at

s = 1

. For non-principal

χ

If $χ^{2} = χ_{0}$ (quadratic), use class number formula connections
In general, if $L (1, χ) = 0$ , analyzing $\prod_{χ} L (s, χ)$ near $s = 1$ via the product formula leads to contradictions with growth estimates

Step 3: Conclusion. Since

L (1, χ_{0})

has a pole and

L (1, χ) \neq = 0

for non-principal

χ

, the orthogonality sum shows infinitely many primes

p \equiv a (mod q)

, with the asymptotic following from refined analysis. □ Common error: The PNT does NOT say

π (x) = x / lo g x

, only that the ratio tends to 1. The better approximation is the logarithmic integral

li (x) = \int_{2}^{x} \frac{d t}{l o g t}

, which satisfies

li (x) \sim x / lo g x

but

li (x) - π (x) = O (x e^{- c l o g x})

. Also, Dirichlet's theorem requires

g cd (a, q) = 1

; there are obviously no primes

p \equiv 0 (mod q)

for

q > 1

except possibly

p = q

itself. Peter Gustav Lejeune Dirichlet (1837) — Proved the first deep result on primes in arithmetic progressions using the analytic properties of L-functions he introduced, creating the field of analytic number theory.

Examples and Applications

1

Computing π(100) and comparing with PNT approximation

The primes up to 100 are: 2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47, 53, 59, 61, 67, 71, 73, 79, 83, 89, 97.

Thus $π (100) = 25$ .

The PNT approximation gives:

\frac{100}{lo g 100} = \frac{100}{ln 100} = \frac{100}{4.605...} \approx 21.71

The relative error is $\frac{∣25 - 21.71∣}{25} \approx 13.2%$ .

The better approximation via the logarithmic integral:

li (100) = \int_{2}^{100} \frac{d t}{lo g t} \approx 30.12

This overshoots but is known to be systematically higher than $π (x)$ for small $x$ (though this pattern eventually reverses, conjecturally around $x \approx 1 0^{316}$ by Skewes' number).

For larger values: $π (1 0^{6}) = 78498$ while $1 0^{6} / lo g (1 0^{6}) \approx 72382$ (error ~8%), showing convergence of the asymptotic.

2

Primes in arithmetic progressions modulo 4

Consider $q = 4$ . The invertible residue classes are $a \in {1, 3}$ with $ϕ (4) = 2$ .

Dirichlet's theorem predicts equal distribution:

π (x; 4, 1) \sim π (x; 4, 3) \sim \frac{1}{2} \cdot \frac{x}{lo g x}

For $x = 100$ :

Primes $\equiv 1 (mod 4)$ : 5, 13, 17, 29, 37, 41, 53, 61, 73, 89, 97 — total: 11
Primes $\equiv 3 (mod 4)$ : 3, 7, 11, 19, 23, 31, 43, 47, 59, 67, 71, 79, 83 — total: 13

Prediction: $\frac{1}{2} \cdot 21.71 \approx 10.85$ for each class.

The counts (11 and 13) are close and roughly equal, as predicted. The equidistribution becomes more pronounced as $x$ increases. This connects to the fact that the non-principal character modulo 4 is $χ (n) = {1 - 1 n \equiv 1 (mod 4) n \equiv 3 (mod 4)$ , and $L (1, χ) = 1 - 1/3 + 1/5 - 1/7 + ... = π /4 \neq = 0$ .

Why gcd condition is necessary in Dirichlet's theorem

Consider the arithmetic progression $2, 6, 10, 14, 18, 22, ...$ (i.e., $a = 2, q = 4$ ). Here $g cd (2, 4) = 2 \neq = 1$ .

This progression contains only one prime: $p = 2$ itself. All other terms are composite (even numbers $> 2$ ).

This shows the coprimality condition $g cd (a, q) = 1$ cannot be dropped. The underlying reason is group-theoretic: characters are defined on the unit group $(Z / q Z)^{*}$ , and orthogonality relations only apply to invertible classes. Elements sharing a common factor with $q$ don't participate in the multiplicative structure that L-functions detect.

More generally, if $d = g cd (a, q) > 1$ , then every element of $a + q Z$ is divisible by $d$ , so at most one prime (namely $p = d$ if it divides $a$ ) can occur.

The sieve perspective: Imagine "sieving" the integers, removing multiples of each prime in turn (Sieve of Eratosthenes). After removing multiples of all primes up to

x

, approximately

x \prod_{p \leq x} (1 - 1/ p)

numbers remain. By Mertens' theorems, this product behaves like

C / lo g x

for some constant

C

, explaining the

x / lo g x

density.

For arithmetic progressions, imagine the integers arranged in columns by residue class modulo $q$ . Sieving acts independently on each column (except for primes dividing $q$ ), so primes distribute equally among the $ϕ (q)$ allowable columns. The L-function machinery makes this heuristic precise through harmonic analysis on the finite abelian group $(Z / q Z)^{*}$ .

Probabilistic model: If we randomly assign each integer

n

to be "prime" with probability

1/ lo g n

, independently, we'd expect about

\int_{2}^{x} \frac{d t}{l o g t} \sim x / lo g x

primes up to

x

. The PNT says the actual deterministic distribution matches this random model asymptotically—a deep instance of pseudorandomness in number theory.

Historical Development

The empirical investigation of prime distribution began with Gauss and Legendre around 1792-1798. Gauss, through extensive hand computation, conjectured that $π (x) \sim li (x)$ , recorded in his unpublished notes. Legendre proposed the form $x / (A lo g x + B)$ based on his tables.

Dirichlet proved his theorem on primes in arithmetic progressions in 1837, published in "Beweis des Satzes, dass jede unbegrenzte arithmetische Progression...". This work introduced Dirichlet characters, Dirichlet L-functions, and established the paradigm of using analytic properties (specifically non-vanishing at

s = 1

) to deduce arithmetic results. His techniques relied heavily on Fourier analysis on finite abelian groups. Riemann's revolutionary 1859 paper "Über die Anzahl der Primzahlen unter einer gegebenen Größe" connected

π (x)

explicitly to zeros of

ζ (s)

via the explicit formula

π (x) = li (x) - ρ \sum li (x^{ρ}) + lower order terms

where

ρ

ranges over non-trivial zeros of

ζ (s)

. He conjectured these zeros all have real part

1/2

(Riemann Hypothesis).

The first complete proof of PNT came in 1896 independently from Jacques Hadamard ("Sur la distribution des zéros de la fonction $ζ (s)$ ...") and Charles-Jean de la Vallée Poussin ("Recherches analytiques sur la théorie des nombres premiers"), both using complex analysis to establish the zero-free region $ℜ (s) = 1$ . Their methods were refined by Landau in his systematic treatise.

Erdős and Selberg shocked the mathematical world in 1949 by discovering elementary proofs (avoiding complex analysis), though these are technically intricate. The analytic approach remains standard for its power and generalizations.

Modern developments include effective versions (explicit error bounds), the explicit formula, connections to random matrix theory (zeros behaving like eigenvalues of random Hermitian matrices), and vast generalizations to automorphic L-functions and Langlands program.

Connections to Other Areas

Random Matrix Theory and the Montgomery-Odlyzko Law: In the 1970s, Hugh Montgomery discovered that the pair correlation of zeros of

ζ (s)

(when rescaled appropriately on the critical line) matches the pair correlation of eigenvalues of random matrices from the Gaussian Unitary Ensemble (GUE). Andrew Odlyzko's extensive numerical computations in the 1980s confirmed this with stunning precision. This connection suggests deep relationships between prime number distribution and Quantum Chaos, since GUE eigenvalues model energy levels of quantum systems with chaotic classical limits. The correspondence remains mysterious but has led to using random matrix theory to predict statistics of zeros and primes, with applications to cryptography and models in quantum physics.

The Prime Number Theorem and Dirichlet's Theorem