Math 259: Introduction to Analytic Number Theory

Download Math 259: Introduction to Analytic Number Theory

Preview text

Math 259: Introduction to Analytic Number Theory Primes in arithmetic progressions: Dirichlet characters and L-functions

Dirichlet extended Euler’s analysis from π(x) to

π(x, a mod q) := #{p ≤ x : p is a positive prime congruent to a mod q}.

We introduce his approach with the example of the distribution of primes mod 4, that is, of π(x, 1 mod 4) and π(x, 3 mod 4). The sum of these is of course π(x)−1 once x ≥ 2, and we have already obtained

∞ −1−s


s 1 π(y)y dy = log s − 1 + Os0 (1) (1 < s ≤ s0) (1)

from the Euler product for ζ(s). By omitting the factor (1 − 2−s)−1 we obtain a product formula for

(1 − 2−s)ζ(s) = 1 + 3−s + 5−s + 7−s + · · · .

If we try to estimate π(·, 1 mod 4) (or π(·, 3 mod 4)) in the same way, we are led to the sum of n−s over the integers all of whose prime factors are are congruent to 1 (or 3) mod 4, which is hard to work with. But we can analyze the difference π(x, 1 mod 4) − π(x, 3 mod 4) using an Euler product for the L-series


L(s, χ4) := 1 − 3s + 5s − 7s + − · · · = χ4(n)n−s.


Here χ4 is the function

 +1, 

χ4(n) = −1,

 


if n ≡ +1 mod 4; if n ≡ −1 mod 4; if 2|n.

This function is (strongly1) multiplicative:

χ4(mn) = χ4(m)χ4(n) (all m, n ∈ Z).


Therefore L(s, χ4) factors as did ζ(s):


L(s, χ4) =

χ(pcp )p−cps = 1 − χ(p)p−s . (3)

p prime cp=1

p prime

By comparison with the Euler product for ζ(s) we see that the manipulations

in (3) are valid for s > 1 (and in fact for s of real part > 1). Unlike ζ(s), the

function L(s, χ4) remains bounded as s→1+, because the sum

∞ n=1


may be grouped as

1Often a function f is called multiplicative when f (mn) = f (m)f (n) only for coprime m, n; see the Exercises.





1 − 3s + 5s − 7s + 9s − 11s + · · ·

in which the n-th term is O(n−(s+1)) (why?). Indeed this regrouping lets us
extend L(·, χ4) to a continuous function on (0, ∞). Moreover, each of the terms (1 − 3−s), (5−s − 7−s), (9−s − 11−s),. . . is positive, so L(s, χ4) > 0 for all s > 0, in particular for s = 1 (you probably already know that L(1, χ4) = π/4). The same analysis we used to get an estimate on the Mellin transform of π(·) from the Euler product for ζ(s) can now be used starting from (3) to obtain:2

s π(y, χ4)y−1−s dy = O(1) (1 < s ≤ 2),




π(y, χ4) := π(y, 1 mod 4) − π(y, 3 mod 4) = χ4(p).

Averaging (4) with (1), we find that

s ∞ π(y, 1 mod 4)y−1−s dy = 1 log 1 + O(1)


2 s−1

s ∞ π(y, 3 mod 4)y−1−s dy = 1 log 1 + O(1)


2 s−1

(1 < s ≤ 2), (1 < s ≤ 2).

This is consistent with π(x, ±1 mod 4) ∼ 21 x/ log x, and corroborates our expectation that there should be on the average as many primes congruent to
+1 mod 4 as −1 mod 4. Specifically, it shows that for a = ±1 the set of primes
congruent to a mod 4 has logarithmic density 1/2 in the primes. This concept
is defined as follows:

Definition. Suppose P is a set of positive integers such that verges. A subset S of P is said to have logarithmic density δ if

n∈P 1/n di-




as s→1+. Taking for P the set of primes, we see that a set S of primes has logarithmic density δ if and only if

p−s ∼ δ log 1 p∈S s − 1

as s→1+.
This notion of “logarithmic density” has the properties we would expect from a density: δ ∈ [0, 1]; a set of positive density is nonempty; if disjoint sets P1, P2
2Again, the choice of s0 > 1 does not matter, because we are concerned with the behavior near s = 1; thus we have made the traditional and convenient choice s0 = 2, rather than continue with an unspecified s0 and a distracting subscript in Os0 .


have logarithmic densities δ1, δ2, then P1 ∪ P2 has logarithmic density δ1 + δ2; and if P1, P2 are sets of logarithmic densities δ1, δ2 and P1 ⊆ P2, then δ1 ≤ δ2. See the first Exercise for further information.
We can use the notion of logarithmic density to state Dirichlet’s theorem as follows:
Theorem [Dirichlet]. For any positive integer q, and any integer a coprime to q, the primes congruent to a mod q constitute a set of logarithmic density 1/ϕ(q) in the primes.
Here ϕ is the Euler phi (“totient”) function, ϕ(q) = |(Z/qZ)∗|. We have just proved the cases (q, a) = (4, ±1) of Dirichlet’s theorem. The same method disposes of (q, a) = (3, ±1), using
(1 − 3−s)ζ(s) = 1 + 2−s + 4−s + 5−s + 7−s + 8−s + · · ·



L(s, χ3) := 1 − 2s + 4s − 5s + − · · · = χ3(n)n−s,


Where χ3 is the multiplicative function defined by

 +1, 

χ3(n) = −1,

 


if n ≡ +1 mod 3; if n ≡ −1 mod 3; if 3|n.

With a tad more work we can deal with q = 8. Let χ8(n) be +1 if n ≡ ±1 mod 8, −1 if n ≡ ±3 mod 8, and 0 if n is even. This is a multiplicative function, as is
χ4χ8; the resulting L-functions


L(s, χ8) = χ8(n)n−s = 1 − 3s − 5s + 7s + − − + · · · ,



L(s, χ4χ8) =

χ4χ8(n)n−s = 1 + 3s − 5s − 7s + + − − · · ·


have Euler products for s > 1 and are positive for s > 0 (to prove this for L(s, χ8), group the terms in fours rather than pairs and use the convexity of the function n → n−s). We deduce that

χ8(p)p−s = O(1) and

χ4χ8(p)p−s = O(1)

for s ∈ (1, 2], which combined with previous results yields Dirichlet’s theorem for q = 8. Similarly we can handle q = 12, and with some more effort even q = 24.

What about q = 5? We have the “quadratic character”, which takes n to +1 or −1 if x ≡ ±1 or ±2 mod 5 (and to 0 if 5|n), but this only lets us separate quadratic from non-quadratic residues mod 5. We need a new idea to


get at the individual nonzero residue classes mod 5. (Recall that {5k + 2}

and {5k − 2} are the first cases of arithmetic progressions that we could not

prove contain infinitely many primes using variations of Euclid’s proof.) Let

χ be the multiplicative function from Z to the complex numbers which takes

n ≡ 0, 1, 2, 3, 4 mod 5 to 0, 1, i, −i, −1. Another such function is the complex

conjugate χ = χ3, while χ2 is the quadratic character and χ4 is the “trivial

character” taking n to 0 or 1 according as 5|n or not. The resulting L-functions

n χ(n)n−s, n χ(n)n−s then take complex values, but still have Euler products and extend to continuous functions on s > 0. Moreover, these functions

never vanish on s > 0; indeed their real and imaginary parts are both nonzero,

as we see by combining the real terms into (5k + 1, 5k + 4) pairs and the imag-

inary terms into (5k + 2, 5k + 3) pairs. Likewise the L-function associated to

the quadratic character χ2 has an Euler product and is positive for s > 0 by

convexity of n−s. We conclude as before that p χj(p)p−s = O(1) as s→1+ for each j = 1, 2, 3, and recover Dirichlet’s theorem for q = 5 by taking linear

combinations of these sums and





1 s−1



For general q, we proceed analogously, using linear combinations of Dirichlet characters, whose definition follows.

Definition. For a positive integer q, a Dirichlet character mod q is a function χ : Z→C that is

• q-periodic: n ≡ n mod q ⇒ χ(n) = χ(n );
• supported on the integers coprime to q and on no smaller subset of Z: (n, q) = 1 ⇔ χ(n) = 0; and
• multiplicative: χ(m)χ(n) = χ(mn) for all integers m, n.

To such a character is associated the Dirichlet L-series

L(s, χ) := ∞ χ(n)n−s = 1

(s > 1).


1 − χ(p)p−s



Examples: The trivial character χ0 mod q is defined by χ(n) = 1 if (n, q) = 1 and χ(n) = 0 otherwise. Its associated L-series is

L(s, χ0) = (1 − p−s) · ζ(s).



If l is prime then the Legendre symbol (·/l), defined by (n/l) = 0, 1, −1 according as n is zero, a nonzero square, or not a square mod l, is a character mod l. If χ is a Dirichlet character mod q then so is its complex conjugate χ (defined of course by χ(n) = χ(n) ), with L(s, χ) = L(s, χ) for s > 1. If χ, χ are characters mod q, q then χχ is a character mod lcm(q, q ). In particular, we have:
Lemma: For each q, the characters mod q constitute a group under pointwise multiplication, with identity χ0 and inverse χ−1 = χ.


What is this group? A Dirichlet character mod q is just a homomorphism from (Z/qZ)∗ to the unit circle, extended by zero to a function on Z/qZ and lifted to Z. Therefore the group of such characters is the Pontrjagin dual of (Z/qZ)∗. Pontrjagin duality for finite abelian groups like (Z/qZ)∗ is easy, since it is equivalent to the theory of the discrete Fourier transform. We next recall the basic facts.
For any finite abelian group G, let Gˆ be its Pontrjagin dual, defined as the group of homomorphisms from G to the unit circle in C. Then the dual of G × H is Gˆ × Hˆ , and the dual of Z/mZ is a cyclic group of order m. Since any finite abelian group is a product of cyclic groups, it follows that Gˆ is isomorphic with G. This isomorphism is not in general canonical,3 but there is a canonical isomorphism from G to the dual of Gˆ (the second dual of G), namely the map taking any g ∈ G to the homomorphism χ → χ(g). That this is an isomorphism can be checked directly for cyclic groups, and then deduced for any finite abelian G because all such G are direct sums of cyclic groups.
The characters of G are orthogonal :

χ (g) χ (g) = |G|, if χ1 = χ2;



0, if χ1 = χ2.


In particular, they are linearly independent, and since there are |G| of them, they form a basis for the vector space of complex-valued functions on G. The decomposition of an arbitrary such function f : G→C as a linear combination of characters is achieved by the inverse Fourier transform:

f = fχχ,


1 fχ := |G| χ(g)f (g).

In particular, the characteristic function of any g0 ∈ G is |G|−1 χ χ(g0)χ.
What does all this tell us about Dirichlet L-functions and the distribution of primes mod q? First, that if we define π(·, χ) by

π(x, χ) :=

χ(a)π(x, a mod q) = χ(p)

a mod q

then, for all a coprime to q,


π(x, a mod q) =

χ(a)π(x, χ).

ϕ(q) χ mod q

Second, that

s π(y, χ)y−1−s dy = χ(p)p−s = log L(s, χ) + O(1)




3For instance, if G is cyclic of order 5, there can be no canonical nondegenerate pairing ·, · : G × G→C∗, because such a pairing would have to be invariant under Aut(G) = (Z/5)∗, but g2, g2 = g, g 4 = g, g .


for 1 < s ≤ 2. This is again obtained by taking logarithms in the Euler product (5). The Euler product shows that L(s, χ) = 0 for s > 1; if χ is complex, “log L(s, χ)” means the branch of the logarithm that approaches 0 as s→∞.
For the behavior of L(s, χ) near s = 1, we have:
Lemma. i) If χ = χ0 then log L(s, χ) = log(1/(s − 1)) + O(1) as s→1+. ii) For nontrivial χ, the sum defining L(s, χ) converges for s > 0 and defines a continuous function on the positive reals.
Proof : (i) follows from (6), together with our estimate on ζ(s) for s→1+. As for (ii), as a special case of character orthogonality we have a mod q χ(a) = 0, so Sχ(x) := 0
χ(n) ns =


y−s dSχ(y) = Sχ(y)y−s + s



y−1−sSχ(y) dy

χ M −s + N −s,

which for fixed s > 0 tends to zero as M, N →∞. Thus the sum ∞ n=1 χ(n)n−s converges. Moreover, for any s0 > 0, the convergence is uniform in s ≥ s0. Hence ∞ n=1 χ(n)n−s is the uniform limit of continuous functions Nn=1 χ(n)n−s, and is therefore a continuous function on (0, ∞), as claimed.

From (7) we see that the crucial question is whether L(1, χ) is nonzero for χ = χ0: the right-hand side is O(1) if L(1, χ) = 0 but ≤ − log(1/(s − 1)) + O(1) if L(1, χ) = 0 (since L(s, χ) is differentiable at s = 1). Our experience with small q, and our expectation that the primes should not favor one congruence class in (Z/qZ)∗ over another, both suggest that L(1, χ) will not vanish. This is true, and can be checked in any given case by a finite computation; but our methods thus far do not let us prove it in general (try doing it for χ = (·/67) or (·/163)!). For the time being, then, we can obtain only a conditional result:

Proposition. Assume that L(1, χ) = 0 for all nontrivial characters χ mod q. Then Dirichlet’s theorem holds for all arithmetic progressions mod q.

Proof : For each a ∈ (Z/qZ)∗, multiply (7) by χ(a), and average over χ to obtain

p−s = 1



χ(a) log L(s, χ) + O(1) =


+ O(1)

p≡a mod q

ϕ(q) χ

ϕ(q) s − 1











ζ(s) + O(1)



1 s−1

+ O(1)



sum, while the other terms remain bounded by hypothesis. Thus the primes

congruent to a mod q have logarithmic density 1/ϕ(q), as claimed.

In fact the nonvanishing of L(1, χ) was proved by Dirichlet, who thus established his celebrated theorem on primes in arithmetic progressions. At least three

4We require that M, N ∈/ Z to avoid the distraction of whether the Riemann-Stieltjes integral RMN y−s dSχ(y) includes the terms with n = M or n = N in the sum.


proofs are now known. These three proofs all start with the product of the L-functions associated to all ϕ(q) Dirichlet characters mod q:


L(s, χ) = 

(1 − χ(p)p−s) .

χ mod q

p χ mod q

The inner product can be evaluated with the following cyclotomic identity: Let G be a finite abelian group and g ∈ G an element of order m. Then

(1 − χ(g)z) = (1 − zm)|G|/m



hold identically for all z.
The identity is an easy consequence of the factorization of 1 − zm together with the fact that any character of a subgroup H ⊆ G extends in [G : H] ways to a character of G (in our case H will be the cyclic subgroup generated by g).
Let mp, then, be the multiplicative order of p mod q (for all but the finitely many primes p dividing q). Then we get

L(s, χ) = (1 − p−mps)−ϕ(q)/mp .


χ mod q


The left-hand side contains the factor L(s, χ0), which is C/(s − 1) + O(1) as s→1+ for some C > 0 [in fact C = ϕ(q)/q]. Since the remaining factors are differentiable at s = 1, if any of them were to vanish there the product would remain bounded as s→1+. So we must show that this cannot happen.
Dirichlet’s original approach was to observe that (9) is, up to a few factors 1 − p−ms with p|q, the “zeta function of the cyclotomic number field Q(e2πi/q)”. He then proved that the zeta function ζK(s) of any number field K is ∼ C/(s−1) as s→1+ for some positive constant C (and gave an exact formula for C, which includes the class number of K and is thus called the “Dirichlet class number formula”). That is undoubtedly the best way to go about it — but it requires more algebraic number theory than I want to assume here. Fortunately there are at least two ad-hoc simplifications available.
The first is that we need only worry about real characters. If L(1, χ) = 0 then also L(1, χ) = 0. Hence if χ = χ but L(1, χ) = 0 then there are at least two factors in the left-hand side of (9) that vanish at s = 1; since they are differentiable there, the product would not only be bounded as s→1+, but approach zero there — which is impossible because the right-hand side is > 1 for all s > 1. But if χ is a real character then L(s, χ0)L(s, χ) is (again within a few factors 1 − n−s of) the L-function of a quadratic number field. Developing the algebraic number theory of quadratic number fields takes considerably less work than is needed for the full Dirichlet class number formula, and if we only want to get unboundedness as s→1+ it is even easier — for instance, if χ(−1) = −1


then the right-hand side of (9) is dominated by the zeta function of a binary quadratic form, which is easily seen to be 1/(s−1). However, even this easier proof is beyond the scope of what I want to assume or fully develop in this class.
Fortunately there is a way to circumvent any ζK beyond K = Q, using the fact that the right-hand side of (9) also dominates the series ζ(ϕ(q) · s), which diverges not at s = 1 but at s = 1/ϕ(q). Since this s is still positive, we can still get a proof of L(1, χ) = 0 from it, but only by appealing to the magic of complex analysis. We thus defer the proof until we have considered ζ(s) and more generally L(s, χ) as functions of a complex variable s, which we shall have to do anyway to obtain the Prime Number Theorem and results on the density (not just logarithmic density) of primes in arithmetic progressions.
Pontrjagin duals. The general setting for Pontrjagin duality is a locally compact abelian topological group G. The Pontrjagin dual of such G is the group Gˆ of continuous homomorphisms from G to the unit circle in C. (When G is finite, or more generally when G is discrete, continuity is automatic.) It is still true, but much harder to prove, that Gˆ is itself locally compact and that the natural map from G to the Pontrjagin dual of Gˆ is an isomorphism. The two most familiar examples of a pair of Pontrjagin-dual groups are G = Gˆ = R and {G, Gˆ} = {R/Z, Z}, where in both cases y ∈ Gˆ corresponds to the homomorphism x → e2πixy. Two further examples are G = Gˆ = Qp (the p-adic rationals, for some prime p), and {G, Gˆ} = {Qp/Zp, Zp}, using the homomorphisms x → exp(2πixy) where xy is any rational number of the form pαz (α, z ∈ Z) that is congruent to xy mod 1.
As in the discrete case, Pontrjagin duality is used to write functions f : G→C as linear combinations of elements of Gˆ, though in general one must replace finite sums by integrals with respect to Haar measure and place some integrability conditions on f , which leads to a much subtler and richer theory. For G = Gˆ = R and {G, Gˆ} = {R/Z, Z}, this theory specializes to the study of the Fourier transform and Fourier series respectively. We shall use Fourier transforms and series several times in the sequel. Fourier analysis on p-adic numbers will not concern us here, but is an important tool in Tate’s generalization of properties of ζ(s) and L(s, χ) to zeta and L functions of number fields.
Zeta functions of number fields. Let K be any number field (finite algebraic extension of Q), and OK its ring of algebraic integers. The “zeta function” ζK (s) is I |I|−s, where I ranges over nonzero ideals of OK and |I| = [OK : I] is the norm of I. For instance, ζ(s) = ζQ(s), and if K = Q[i] then ζK (s) = 41 (m2 + n2)−s, the sum extending over all (m, n) ∈ Z2 other than (0, 0). The relation between the product (9) and the zeta function of Q(e2πi/q) can be made more precise: if we replace each χ by its underlying primitive character (see the Exercises), the product is exactly the zeta function of that cyclotomic number field. Similarly, for any quadratic field K there is a primitive Dirichlet character χ such that ζK(s) = ζ(s)L(s, χ). These are the prototypical examples

of the factorization of a zeta function as a product of Artin L-functions; the fact that the “Artin L-functions” for 1-dimensional representations of Gal(Q/Q) are Dirichlet series is a prototype for class field theory. Dirichlet’s theorem in turn generalizes to the Cˇ ebotarev density theorem. These theorems all require more algebraic machinery than the results we shall obtain using only the Riemann zeta function and Dirichlet L-functions, but much the same analytic methods. Therefore we shall not develop them further in Math 259.
Concerning density:
1. If P is an infinite set of integers, the (natural) density of any subset S ⊆ P is

lim #{n ∈ S : n < x}/#{n ∈ P : n < x},

if the limit exists. Check that this satisfies the same properties we noted for
the logarithmic density (density of subsets, disjoint unions, etc.). Show that if
n∈P 1/n diverges and S ⊂ P has density δ in P then it also has logarithmic density δ in P . (Use partial summation to write n∈S n−s as an integral involving #{n ∈ S : n < x}.) If P is the set of natural numbers and Sd (d = 1, 2, . . . , 9) is the subset consisting of integers whose first decimal digit is d, show that Sd has logarithmic density log10(1 + d1 ) in P but no natural density. Does every set of natural numbers have a logarithmic density?

While not every set with a logarithmic density has a natural density, we shall see that the primes congruent to a mod q do have natural density 1/ϕ(q) in the primes. As for the sets Sd, their logarithmic densities account for “Benford’s Law”, the observation that in many naturally occurring “random numbers” the initial digit d occurs with frequency log10(1 + d1 ), rather than 1/9 as one might expect.
Concerning Euler products:

2. One may associate to any sequence (a1, a2, a3, . . .) of complex numbers a

Dirichlet series L(s) =

∞ n=1








half-plane s > s0 if an ns0−1. Show that L(s) has an Euler product

L(s) =

∞ apc pcs

if and only if amn = aman for any m, n such that gcd(m, n) = 1. (Such functions n → an are called “multiplicative”. Note that necessarily a1 = 1 if {an} is multiplicative.)

3. Let f (s) be the sum of n−s over squarefree positive integers n. Express f (s)
in terms of the zeta function, and evaluate f (2). Given k, what is the coefficient of n−s in the Dirichlet series for ζ(s − k), or ζ(s)ζ(s − k)?

4. Find a1, a2, . . . such that

p p−s =

∞ n=1












(and a computer package that knows about ζ(2n) and high-precision arithmetic)


to calculate that 1 p2 = 0.45224742004106549850654336483224793417323 . . .

Note that this is much greater accuracy than we could reasonably expect to reach by

summing the series directly. We shall see that this trick can be adapted to efficiently



f (p)







Concerning Pontrjagin duality:

5. Show that to any homomorphism α : H→G between finite abelian groups there is a canonically associated homomorphism αˆ : Gˆ→Hˆ in the opposite
direction between their Pontrjagin duals. Check that α is the dual of αˆ (under the canonical identification of G and H with the duals of Gˆ, Hˆ ), and that if β
is a homomorphism from G to a finite abelian group K then the dual of the composite homomorphism β ◦ α : H→K is αˆ ◦ βˆ. Prove that im(α) = ker(β) if and only if im(βˆ) = ker(αˆ).

In particular, if H→G is an injection, it follows (by taking β to be the quotient map G→G/α(H)) that the restriction map αˆ : Gˆ→Hˆ is a surjection; this was used to prove
the cyclotomic identity (8). An adherent of the categorical imperative would sum-
marize this exercise, together with the easy observations that ibd = id (when G = H) and ˆ0 = 0, by saying that Pontrjagin duality is an “exact contravariant functor on the
category of finite abelian groups”.

Concerning Dirichlet characters:

6. Show that the integers q modulo which all the Dirichlet characters are real
(take on only the values 0, ±1) are precisely 24 and its factors. Show that every
real Dirichlet character is of the form χ0ψ l∈S(·/l), where χ0 is the trivial character, ψ = χ44 χ88 for some 4, 8 ∈ {0, 1}, and S is a (possibly empty) finite set S of odd primes.

7. Let χ0 be the trivial character mod q, and let q1 be some factor of q. For any character χ1 mod q1 there is a character χ mod q defined by χ = χ0χ1. Express L(s, χ) in terms of L(s, χ1). Conclude that L(1, χ) = 0 if and only if L(1, χ1) = 0.
8. A character mod q that cannot be obtained in this way from any character
mod a proper factor q1|q (a factor other than q itself) is called primitive. Show that any Dirichlet character χ comes from a unique primitive character χ1. [The modulus of this χ1 is called the conductor of χ.] Show that the number of primitive characters mod n is n p|n αp, where αp = ((p − 1)/p)2 if p2|n and (p − 2)/p if p n. NB there are no primitive characters mod n when 2 n.

The notation pf n means that pf divides n “exactly”; that is, pf |n but pf+1 does not divide n. Equivalently, the p-valuation of n is f .

9. Deduce the fact that for any q there is at most one nontrivial character χ mod q such that L(1, χ) = 0, as a consequence of (7) together with the fact that π(x, a mod q) ≥ 0 for all x, a, q. [In the final analysis, this is not much different


Preparing to load PDF file. please wait...

0 of 0
Math 259: Introduction to Analytic Number Theory