Analytic Number Theory

Dirichlet Series

Σ aₙ/n^s — generating function for arithmetic data, generalizes ζ(s) and L-functions

A Dirichlet series is an infinite series of the form Σ_{n=1}^∞ aₙ/n^s, where {aₙ} is a sequence of complex numbers and s is a complex variable. Convergence: there exists an "abscissa of convergence" σ_c such that the series converges for Re(s) > σ_c and diverges for Re(s) < σ_c (analogous to power series radius). The Riemann zeta function ζ(s) = Σ 1/n^s is the simplest case. Dirichlet L-functions L(s, χ) = Σ χ(n)/n^s for a character χ generalize ζ and prove the Dirichlet theorem on primes in arithmetic progressions (1837). The convolution of arithmetic functions (Dirichlet convolution f*g(n) = Σ_{d|n} f(d)g(n/d)) becomes ordinary multiplication of Dirichlet series — a powerful tool for proving identities like ζ(s)² = Σ d(n)/n^s where d(n) is the divisor function.

  • FormΣ aₙ/n^s
  • Abscissa of convergenceσ_c
  • ζ(s)Σ 1/n^s
  • Dirichlet L-functionΣ χ(n)/n^s
  • Dirichlet convolutionf*g
  • Dirichlet theoremprimes in AP (1837)

Watch the 60-second explainer

A condensed visual walkthrough — narrated, captioned, under a minute.

Why Dirichlet series matter

  • Analytic number theory. Dirichlet series are the central tool for studying number-theoretic functions through complex analysis. The Prime Number Theorem π(x) ~ x/ln(x) was proved (Hadamard, de la Vallée Poussin, 1896) by analyzing ζ(s) on the line Re(s) = 1 — without Dirichlet series the proof has no foothold.
  • Primes in arithmetic progressions. Dirichlet's theorem (1837): infinitely many primes ≡ a (mod q) when gcd(a,q) = 1. The proof requires Dirichlet L-functions L(s, χ) and crucially that L(1, χ) ≠ 0 for non-principal characters — a sophisticated analytic fact with no elementary proof.
  • Modular forms. Modular forms have associated L-functions (Mellin transforms of f(it) on the upper half-plane). Hecke's theory of L-functions of modular forms ties together complex analysis, modular symmetry, and number theory. The Mordell-Weil theorem and modularity theorem (Wiles, 1995) live in this language.
  • L-functions in general. Galois representations, automorphic forms, and elliptic curves all attach to Dirichlet series with Euler products. The Langlands program is a vast conjectured network of relationships among L-functions of different origin types — a unifying vision of arithmetic.
  • Möbius inversion. The identity Σ_{d|n} μ(d) = δ(n) (where δ(1) = 1, δ(n>1) = 0) makes the Möbius function the inverse of 1 under Dirichlet convolution — equivalently 1/ζ(s) = Σ μ(n)/n^s. This translates "summed over divisors" identities into clean closed forms via division of Dirichlet series.
  • Riemann Hypothesis. The most famous open problem in mathematics: all nontrivial zeros of ζ(s) lie on Re(s) = 1/2. RH would give the sharpest known bounds on prime gaps and improvements throughout analytic number theory. The framework of Dirichlet series is what lets RH be even stated.
  • Tauberian theorems. Connect asymptotic behavior of partial sums Σ_{n ≤ x} aₙ to analytic properties of the Dirichlet series F(s). Wiener's Tauberian theorem and the Wiener-Ikehara theorem are tools for converting analytic information into arithmetic information.

Common misconceptions

  • Dirichlet = Fourier. They're different. Fourier transform handles periodic or translation-invariant phenomena via integrals against e^(ixt). Dirichlet series handle multiplicative structures via series in n^(−s). They share spirit ("transform a function into a frequency-like object") but the multiplicative semigroup of natural numbers is what makes Dirichlet series the right tool for primes.
  • Always converges. A Dirichlet series only converges for Re(s) > σ_c. For ζ(s), σ_c = 1; for s = 0, ζ(0) is not the series Σ 1, but rather an analytically continued value (ζ(0) = −1/2). Mixing up the series with its analytic continuation causes endless confusion in formal manipulations.
  • Only for ζ. ζ(s) is the simplest of an enormous family of Dirichlet series and L-functions. Each character mod q gives its own L-function; each cusp form gives one; each elliptic curve gives one; each Galois representation gives one. Modern number theory is largely about these L-functions and their interrelationships.
  • Convergent everywhere series have nice analytic continuation. Convergence in a half-plane Re(s) > σ_c says nothing about behavior in Re(s) ≤ σ_c. Whether a Dirichlet series extends to a meromorphic function on all of ℂ is a separate, often deep, question. ζ extends to ℂ \ {1}; not all Dirichlet series do.
  • Coefficients aₙ recoverable from F(s). Yes, but the recovery is delicate. By Mellin inversion: aₙ = (1/(2πi)) ∫_(c−i∞)^(c+i∞) F(s) n^s ds for c large enough. The integral is over a vertical line and converges only conditionally; numerical evaluation is unstable. Theoretical use, not computational.
  • Euler product is just notation. The Euler product ζ(s) = ∏ 1/(1 − p^(−s)) is a genuine identity of analytic functions valid for Re(s) > 1, encoding the fundamental theorem of arithmetic. Logarithmic derivative gives ζ'(s)/ζ(s) = −Σ Λ(n)/n^s where Λ is the von Mangoldt function — a key tool linking ζ-zeros to prime distribution.

Worked example

The Riemann zeta function ζ(s) = Σ 1/n^s converges absolutely for Re(s) > 1. Its Euler product: for Re(s) > 1, ζ(s) = ∏_p (1 − p^(−s))⁻¹. Verify for s = 2: ζ(2) = π²/6. The first few terms 1 + 1/4 + 1/9 + 1/16 + 1/25 + … converge slowly to π²/6 ≈ 1.6449.

The Möbius function μ(n) is defined: μ(1) = 1, μ(n) = (−1)^k if n = p₁p₂…p_k is squarefree with k distinct primes, μ(n) = 0 if n has a squared prime factor. Its Dirichlet series is 1/ζ(s) = Σ μ(n)/n^s for Re(s) > 1. Proof: 1/(1 − p^(−s)) becomes 1 − p^(−s) when reciprocated, and expanding ∏_p (1 − p^(−s)) = Σ μ(n)/n^s by including/excluding squarefree subsets of primes.

The divisor function d(n) = number of positive divisors of n. Then Σ d(n)/n^s = ζ(s)². Proof: ζ(s)² = (Σ 1/d^s)(Σ 1/e^s) = Σ_{d,e ≥ 1} 1/(de)^s. Group by n = de: coefficient of 1/n^s is the number of (d, e) pairs with de = n — exactly d(n). Hence ζ(s)² = Σ d(n)/n^s. Convergence for Re(s) > 1.

Dirichlet L-function example: take χ the unique non-principal character mod 4 (χ(1) = 1, χ(3) = −1, χ(even) = 0). Then L(s, χ) = 1 − 1/3^s + 1/5^s − 1/7^s + 1/9^s − …. This converges for Re(s) > 0 (alternating series). Special value: L(1, χ) = 1 − 1/3 + 1/5 − 1/7 + … = π/4 (Leibniz's formula). The non-vanishing L(1, χ) ≠ 0 is exactly what Dirichlet needed to prove there are infinitely many primes ≡ 1 mod 4 and infinitely many primes ≡ 3 mod 4.

Frequently asked questions

What does the abscissa of convergence mean?

For a Dirichlet series Σ aₙ/n^s, there exists a real number σ_c (possibly ±∞) called the abscissa of convergence such that the series converges for Re(s) > σ_c and diverges for Re(s) < σ_c. Behavior on the line Re(s) = σ_c is delicate. This is the Dirichlet-series analog of radius of convergence for power series, but unlike power series the convergence region is a half-plane rather than a disk. There's also an abscissa of absolute convergence σ_a ≥ σ_c, where the series converges absolutely. For ζ(s): σ_c = σ_a = 1.

How does Dirichlet convolution become product of series?

Define f * g (Dirichlet convolution) by (f * g)(n) = Σ_{d|n} f(d) g(n/d), where the sum runs over divisors of n. Then if F(s) = Σ f(n)/n^s and G(s) = Σ g(n)/n^s, the product F(s)G(s) = Σ (f*g)(n)/n^s. Proof: expand (Σ f(d)/d^s)(Σ g(e)/e^s) = Σ_{d,e} f(d)g(e)/(de)^s, then group by n = de. Examples: 1 * 1 = d(n) (divisor count), so ζ(s)² = Σ d(n)/n^s. The constant function 1 has Dirichlet series ζ(s); the Möbius function μ has 1/ζ(s); convolution 1 * μ = δ (identity for convolution) gives ζ(s) · 1/ζ(s) = 1.

What is a Dirichlet character?

A Dirichlet character mod q is a completely multiplicative function χ: ℤ → ℂ that is periodic with period q (so χ(n+q) = χ(n)) and vanishes on integers sharing a factor with q (gcd(n, q) > 1 implies χ(n) = 0). It's induced by a homomorphism (ℤ/qℤ)* → ℂ* extended by zero. The principal character χ₀ is 1 on units mod q, 0 elsewhere. There are exactly φ(q) characters mod q (where φ is Euler's totient). Characters distinguish residue classes — they're the Fourier basis on (ℤ/qℤ)*.

What does the Dirichlet theorem on arithmetic progressions say?

Dirichlet's theorem (1837): for any coprime integers a, q with gcd(a, q) = 1, the arithmetic progression a, a+q, a+2q, a+3q, … contains infinitely many primes. Equivalently: there are infinitely many primes p ≡ a (mod q). Stronger: each residue class mod q with gcd(a,q)=1 receives a 1/φ(q) share of primes (Dirichlet density). The proof uses non-vanishing of L(1, χ) for non-principal characters — combined with the analytic continuation of Dirichlet L-functions, this isolates the contribution of primes in any chosen residue class.

How does ζ(s) factor as the Euler product?

Euler product: for Re(s) > 1, ζ(s) = ∏_{p prime} 1/(1 − p^(−s)). Proof: expand each factor as a geometric series 1 + p^(−s) + p^(−2s) + …; multiplying over all primes and using unique prime factorization, every n ≥ 1 appears exactly once with weight n^(−s). The product encodes the multiplicative structure: ζ is built from primes. L-functions have analogous Euler products: L(s, χ) = ∏_p 1/(1 − χ(p) p^(−s)). Euler products are why Dirichlet series are useful for studying primes — they translate prime questions into analytic ones.

Why are L-functions central to modern number theory?

L-functions package arithmetic data — primes, characters, modular forms, elliptic curves, Galois representations — into analytic objects with Euler products and functional equations. Their analytic properties encode deep arithmetic facts: zeros of ζ control prime distribution (Riemann Hypothesis), L-function values at integers relate to special arithmetic invariants (Birch-Swinnerton-Dyer conjecture for elliptic curves), Langlands program proposes a vast web of correspondences among L-functions of different origins. Most major open problems in number theory — RH, BSD, Tate, Langlands — are ultimately statements about L-functions.