TheoremComplete

Dirichlet's Theorem on Primes in Arithmetic Progressions

Theorem5.1Dirichlet's Theorem

For any integers aa and qq with gcd⁑(a,q)=1\gcd(a, q) = 1, there are infinitely many primes p≑a(modq)p \equiv a \pmod{q}. Equivalently, βˆ‘p≑amodq1p=∞\sum_{\substack{p \equiv a \bmod q}} \frac{1}{p} = \infty.


Proof

Proof

The proof uses the non-vanishing of L(1,Ο‡)L(1, \chi) for Ο‡β‰ Ο‡0\chi \neq \chi_0.

Step 1: Character decomposition. By orthogonality: βˆ‘p≀xp≑amodq1p=1Ο•(q)βˆ‘Ο‡β€Šmodβ€ŠqΟ‡(a)β€Ύβˆ‘p≀xΟ‡(p)p\sum_{\substack{p \leq x \\ p \equiv a \bmod q}} \frac{1}{p} = \frac{1}{\phi(q)} \sum_{\chi \bmod q} \overline{\chi(a)} \sum_{p \leq x} \frac{\chi(p)}{p}.

Step 2: Connection to log⁑L(s,Ο‡)\log L(s, \chi). Taking logarithms of the Euler product: log⁑L(s,Ο‡)=βˆ‘pΟ‡(p)ps+O(1)\log L(s, \chi) = \sum_p \frac{\chi(p)}{p^s} + O(1) for s>1s > 1, where the O(1)O(1) accounts for prime powers.

Step 3: Principal character contribution. For Ο‡=Ο‡0\chi = \chi_0: L(s,Ο‡0)=ΞΆ(s)∏p∣q(1βˆ’pβˆ’s)L(s, \chi_0) = \zeta(s)\prod_{p|q}(1 - p^{-s}), so log⁑L(s,Ο‡0)β†’βˆž\log L(s, \chi_0) \to \infty as sβ†’1+s \to 1^+. This gives βˆ‘pΟ‡0(p)/psβ†’βˆž\sum_p \chi_0(p)/p^s \to \infty.

Step 4: Non-principal characters are bounded. For Ο‡β‰ Ο‡0\chi \neq \chi_0, if L(1,Ο‡)β‰ 0L(1, \chi) \neq 0, then log⁑L(s,Ο‡)\log L(s, \chi) remains bounded as sβ†’1+s \to 1^+, so βˆ‘pΟ‡(p)/ps=O(1)\sum_p \chi(p)/p^s = O(1).

Step 5: Non-vanishing of L(1,Ο‡)L(1, \chi). This is the crux. For complex Ο‡\chi (i.e., Ο‡β‰ Ο‡β€Ύ\chi \neq \overline{\chi}): consider F(s)=βˆΟ‡L(s,Ο‡)=βˆ‘nan/nsF(s) = \prod_\chi L(s, \chi) = \sum_{n} a_n/n^s where anβ‰₯0a_n \geq 0 (since F(s)=βˆΟ‡βˆp(1βˆ’Ο‡(p)pβˆ’s)βˆ’1=∏pβˆΟ‡(1βˆ’Ο‡(p)pβˆ’s)βˆ’1F(s) = \prod_\chi \prod_p (1 - \chi(p)p^{-s})^{-1} = \prod_p \prod_\chi (1 - \chi(p)p^{-s})^{-1}, and βˆΟ‡(1βˆ’Ο‡(p)x)=(1βˆ’xfp)Ο•(q)/fp\prod_\chi (1 - \chi(p)x) = (1 - x^{f_p})^{\phi(q)/f_p} where fpf_p is the order of pp in (Z/qZ)Γ—(\mathbb{Z}/q\mathbb{Z})^\times). Since the Dirichlet series has non-negative coefficients and a pole at s=1s = 1 (from Ο‡0\chi_0), it cannot have a zero at s=1s = 1 from another factor, as this would make F(s)F(s) analytic at s=1s = 1 with non-negative coefficients and a singularity at its abscissa of convergence -- a contradiction by Landau's theorem.

For real Ο‡\chi: a more delicate argument using the class number formula L(1,Ο‡)=2Ο€hKw∣D∣L(1, \chi) = \frac{2\pi h_K}{w\sqrt{|D|}} (for the associated quadratic field) or the product ΞΆ(s)L(s,Ο‡)β‰₯1\zeta(s)L(s,\chi) \geq 1 (both approaches) shows L(1,Ο‡)>0L(1, \chi) > 0.

Step 6: Conclusion. Combining: βˆ‘p≀xp≑amodq1p=1Ο•(q)log⁑log⁑x+O(1)\sum_{\substack{p \leq x \\ p \equiv a \bmod q}} \frac{1}{p} = \frac{1}{\phi(q)}\log\log x + O(1), proving infinitely many primes in the progression. β–‘\square

β– 

RemarkEffective vs. Ineffective Bounds

Dirichlet's original proof is effective for complex characters but ineffective for real characters (the class number argument gives no explicit lower bound for L(1,Ο‡)L(1, \chi)). Effective bounds require either assuming GRH or using Siegel's theorem L(1,Ο‡)≫qβˆ’Ξ΅L(1, \chi) \gg q^{-\varepsilon} with an ineffective constant.