THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY

QUADRATIC FIELD

The factorizations
√ √
6 = 2 · 3 = (1 + −5)(1 − −5)
show that unique factorization fails in the ring
√ √
Z[ −5] = {a + b −5 : a, b ∈ Z},
√
because 2, 3, and 1 ± −5 are irreducible and nonassociate.
These notes present a formula, due to Dirichlet, that in some sense
√ measures the
extent to which unique factorization fails in settings such as Z[ −5]. The large-
scale methodology deserves immediate note, before the reader is immersed in a long
succession of smaller attention-demanding specifics:
• algebra lets us define a group that measures the failure of unique factoriza-
tion,
• geometry shows that the group is finite and gives an algorithm to find a set
of group element representatives in any specific instance,
• and analysis yields the formula for the group’s order.
To move forward through the main storyline without bogging down, the expo-
sition quotes results from algebra and complex analysis even though elementary
arguments are possible in this context. For a more fleshed out and elementary pre-
sentation, see Tom Weston’s online notes for the 2004 Ross mathematics program,
www.math.umass.edu/~weston/oldpapers/cnf.pdf
The class number formula in general is discussed in many number theory books,
such as the books by Marcus and by Borevich and Shafarevich.

Contents

Part 1. ALGEBRA: QUADRATIC NUMBER FIELDS 2

1. Quadratic Fields and Their Integers 2
2. The Units of a Quadratic Field 4
3. The Ideals of a Quadratic Field 5
4. Unique Factorization of Ideals 7
5. The Character of a Quadratic Field 8
6. Decomposition of Rational Primes 12
7. Fractional Ideals and the Ideal Class Group 15
8. Abelian Group Structure of Ideals 16

Part 2. GEOMETRY: COMPLEX LATTICES 16

9. Complex Lattices and Homothety 17
10. Complex Multiplication 19
1
2 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD

Part 3. ANALYSIS: ZETA AND L-FUNCTIONS OF AN

IMAGINARY QUADRATIC FIELD 21
11. Summation by Parts and Dirichlet Series Convergence 21
12. The Euler–Riemann Zeta Function 23
13. The L-Function of a Quadratic Field 24
14. The Dedekind Zeta Function of a Quadratic Field 28
15. The Class Number Formula 33

Part 1. ALGEBRA: QUADRATIC NUMBER FIELDS

√
This part of these notes discusses quadratic
√ number fields (fields like Q( −5))
and their rings of integers (rings like Z[ −5]). The ideals of the ring factor uniquely
even though the elements of the ring may not. A group called the ideal class
group measures the extent to which ideals fail to correspond to ring elements, thus
measuring the extent to which unique factorization of elements fails.

1. Quadratic Fields and Their Integers

Definition 1.1. A quadratic number field is a field F (inside C) such that F
has dimension 2 as a vector space over Q. Such a field takes the form
√ √
F = Q( n) = {a + b n : a, b ∈ Q}, n ∈ Z − {0, 1} squarefree.
If n is positive then F is a real quadratic number field, and if n is negative then F
is an imaginary quadratic number field.

From now on in this writeup the symbol F denotes a quadratic number field,
and quadratic number field is freely shortened to quadratic field .
The conjugation function of F is
√ √
: F −→ F, a + b n = a − b n.
Conjugation is a ring homomorphism, meaning that
x+y =x+y and xy = x y for all x, y ∈ F .
And conjugation is an involution, meaning that
x = x for all x ∈ F .
Thus conjugation is an automorphism of F . The only other automorphism of F is
the identity map, and so the group of automorphisms of F has order 2, generated
by conjugation.
The trace function of F is the additive homomorphism
tr : F −→ Q, tr(α) = α + α.
Specifically,
√ √ √
tr(a + b n) = a + b n + a + b n = 2a.
The norm function of F is the multiplicative homomorphism
N : F × −→ Q× , N(α) = α α.
Specifically,
√ √ √
N(a + b n) = (a + b n)(a + b n) = a2 − b2 n.
 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD 3

If F is imaginary quadratic then the norm is positive on F × . We extend the norm

to N(0) = 00 = 0, and this extended norm is still multiplicative.
Because conjugation is an involution, it has no effect on trace and norm, i.e.,
tr(α) = α + α = α + α = α + α = tr(α) and similarly N(α) = N(α) for all α ∈ F .
Because each element α of F − Q has trace and norm in Q it satisfies a monic
quadratic polynomial with rational coefficients,
(X − α)(X − α) = X 2 − tr(α)X + N(α).
√
With α = a + b n this polynomial has discriminant tr(α)2 − 4N(α) = 4a2 − 4(a2 −
b2 n) = 4b2 n, and so it is irreducible because b is nonzero and n is not 0 or 1 and is
squarefree. Thus α does not satisfy a monic linear polynomial over Q. Of course,
each α ∈ Q satisfies the monic linear polynomial X − α over Q.
Definition 1.2. An element of F is an integer if its minimal monic polynomial
over Q in fact has coefficients in Z.

√ the integers of F ∩ Q (the rational integers of F ) are Z. An element α =

Thus
a + b n of F , where a, b ∈ Q, is an algebraic integer if and only if its trace 2a and
its norm a2 − b2 n are rational integers. Supposing that 2a, a2 − b2 n ∈ Z, we obtain
necessary conditions on a and b:
• The relation 4N(α) = (2a)2 − 4b2 n gives 4b2 n ∈ Z.
• If a ∈ Z then it further shows that 4b2 n ∈ 4Z, so that b ∈ Z because n is
squarefree.
/ Z then a ∈ 12 Z − Z and so 2a is an odd integer, and the relation
• If a ∈
4N(α) = (2a)2 − 4b2 n further shows that 4b2 n ≡ 1 (mod 4), so that b ∈
1 2 2
2 Z − Z and then 2b is an odd integer and so 4b = (2b) ≡ 1 (mod 4),
giving n ≡ 1 (mod 4).
• That is, the necessary conditions are that either a, b ∈ Z with no conditions
on n, or a, b ∈ 21 Z − Z and n ≡ 1 (mod 4).
Now we show that the necessary conditions on a and b are also sufficient:
• If a, b ∈ Z then certainly 2a, a2 − b2 n ∈ Z.
• If a = ao /2, b = bo /2 with ao , bo ∈ Z odd and n ≡ 1 (mod 4) then 2a =
ao ∈ Z, and because a2o − b2o n ≡ 0 (mod 4) also a2 − b2 n = (a2o − b2o n)/4 ∈ Z.
We have established
√
Proposition 1.3. The integers of the quadratic field F = Q( n) are
( √
1+ n
if n = 1 (mod 4)
OF = Z[g], g = √ 2
n if n = 2, 3 (mod 4).
The integers of F form a ring.
The minimal monic polynomial in Z[X] satisfied by the OF generator g in the
previous proposition is quadratic,
(
X 2 − X − n−1
4 if n = 1 (mod 4)
f (X) = 2
X −n if n = 2, 3 (mod 4).
Thus, as an abelian group the integer ring is in fact
OF = gZ ⊕ Z.
4 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD

The discriminant of the quadratic polynomial f (X) (the quantity b2 −4ac that goes
under the square root in the quadratic formula) is n if n = 1 (mod 4) and 4n if
n = 2, 3 (mod 4). This quantity, a structure constant of the quadratic field F , will
appear in the class number formula.
Definition 1.4. The discriminant of F is
(
n if n = 1 (mod 4)
DF =
4n if n = 2, 3 (mod 4).
The cases built into the definition of the discriminant allow it to give a uniform
description of the integers,
√
DF + DF
OF = Z[r], r= ,
2
and similarly we will see that the discriminant gives uniform descriptions of various
phenomena associated with F . The minimal polynomial of r is
f (X) = X 2 − DF X + DF (DF − 1)/4,
whose coefficients are rational integers and whose discriminant is indeed DF .
One can think of the casewise formula for the discriminant as the result of a
calculation rather than as a definition. Other definitions of the discriminant are
case-free in terms of g (where OF = gZ ⊕ Z as before), although g itself involves
cases,  
1 g 2
DF = (det )
1 g
and  
tr(1 · 1) tr(1 · g)
DF = det .
tr(g · 1) tr(g · g)

2. The Units of a Quadratic Field

Definition 2.1. A unit of F is an invertible element of the integer ring OF . The
unit group of F is the multiplicative group OF× .
Proposition 2.2. An element α of OF is a unit if and only if N(α) = ±1.
Proof. If α ∈ OF is multiplicatively invertible by β ∈ OF then
1 = N(1) = N(αβ) = N(α)N(β),
so that N(α) = ±1 because both norms are integers. Conversely, if N(α) = ±1 then
α is invertible by ±α ∈ OF because ±αα = ±N(α) = 1. 
√

If F = Q( n) is imaginary quadratic then all norms (2a + bDF )2 − b2 DF /4
are nonnegative, and inspection shows that the unit group is

{±1, ±i}
 if n = −1 √
×
OF = {±1, ±ζ3 , ±ζ3 } if n = −3 (where ζ3 = −1+2 −3 )
2

{±1} otherwise.


Further examination quickly shows that these groups are cyclic, but also we know
that any finite subgroup of the multiplicative group of any field is cyclic, so this is
 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD 5

√
only a confirmation. If F = Q( n) is real quadratic then nontrivially there exists
a so-called fundamental unit u > 1 in OF× such that the unit group is
OF× = {±un : n ∈ Z} ≈ Z/2Z × Z.
Finding the fundamental unit amounts to solving Pell’s Equation, x2 − ny 2 = 1.
Definition 2.3. The symbol w = w(F ) denotes the number of roots of unity in F ,
i.e., the number of complex numbers in F having absolute value 1. Thus

4 if F = Q(i)

√
w(F ) = 6 if F = Q( −3)

2 otherwise.


Thus w is a second structure constant of the field F , along with the discrimi-
nant DF . For imaginary quadratic fields F the number w completely describes the
unit group. For real quadratic fields the fundamental unit u is a further structure
constant necessary for a full description. The more complicated unit group struc-
ture for real quadratic fields is one reason that the class number formula is easier
in the imaginary case.

3. The Ideals of a Quadratic Field

Definition 3.1. An ideal of OF is a subset a ⊂ OF (excluding a = {0}) that
forms an abelian group and is closed under multiplication by OF .
Thus an ideal is a particular kind of subring. The ideals of any ring are special
among subrings similarly to how the normal subgroups of any group are special
among subgroups: the quotient of the ring by an ideal again has a ring structure,
whereas the quotient of the ring by a subring in general need not.
The subset {0} ⊂ OF does form an abelian group and it is closed under mul-
tiplication by OF , so often it is considered an ideal of OF as well. However, the
zero ideal has aberrational properties, and our concern here is with the unique fac-
torization of nonzero ideals, so for brevity we exclude {0} from the discussion even
though in principle we should say nonzero ideal throughout. The zero ideal will
play a role once later in the writeup, and we will point out when that happens.
Definition 3.2. The sum of two ideals of OF is the ideal generated by the sums
of the elements
a + a0 = hx + x0 : x ∈ a, x0 ∈ a0 i,
and similarly for the product,
aa0 = hxx0 : x ∈ a, x0 ∈ a0 i.
In fact the ideal sum consists of exactly the generating element-sums, but the
ideal product consists of all finite sums of the generating element-products. Note
that the product aa0 is a subset of a and of a0 . The addition and multiplication
of ideals is commutative and associative and distributive. If the zero ideal {0}
were allowed then it would be the additive identity, and the integer ring OF is
the multiplicative identity. However, nonzero ideals do not have additive inverses.
Again, our concern here is with the multiplicative structure of ideals, so the absence
of an additive identity or additive inverses is of no concern. Before long we will
remedy the absence of multiplicative inverses by enlarging our notion of an ideal.
6 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD

Definition 3.3. Let a be an ideal of OF , and let a = {x : x ∈ a}, another ideal

of OF . The norm of a, denoted N(a), is characterized by the conditions
N(a) ∈ Z+ .
aa = N(a)OF ,
√
√ ring OF = Z[ −5] of the quadratic field F =
For example, in the integer
√
Q( −5), the ideal p = (1 + −5)OF + 2OF satisfies
√ √
pp = 6OF + 2(1 + −5)OF + 2(1 − −5)OF + 4OF = 2OF ,
showing that N(p) = 2.
The existence of the ideal norm in general is not immediate, but we will establish
it soon. Granting the ideal norm’s existence, its characterizing conditions show that
it is a multiplicative function of ideals,
N(aa0 ) = N(a)N(a0 ) for all ideals a, a0 of OF .
Indeed, N(aa0 )OF = aa0 aa0 = aaa0 a0 = N(a)OF N(a0 )OF = N(a)N(a0 )OF , giving
the result.
Continuing to grant its existence, the ideal norm lets us prove a cancellation law
for ideals of OF . Suppose that
aa0 = aa00 .
Then
N(a)a0 = aaa0 = aaa00 = N(a)a00 ,
so that by the elementwise cancellation law applied to each relation N(a)a0 = N(a)a00
with a0 ∈ a and a00 ∈ a00 ,
a0 = a00 .
Definition 3.4. An ideal is principal if it takes the form
a = xOF for some x ∈ OF .
A principal ideal is denoted by its generator in angle brackets,
hxi = xOF .
The relation between the element norm from earlier and the ideal norm just
introduced is:
For a principal ideal a = hxi, N(a) = |N(x)|.
If all ideals were principal then the theory of ideals would introduce nothing
new to the study of quadratic integer rings. We will see in the next section that
the ideals of a quadratic integer ring factor uniquely, whereas we know by example
that the elements of the ring may not. Thus the possible failure of all ideals to be
principal is related to the possible failure of unique factorization of elements.
We end the section by showing that the ideal norm exists.
Proposition 3.5. Let a be an ideal of OF . Then aa = dOF for some d ∈ Z+ .
Proof. The product aa contains elements xx = N(x) where x ∈ a is nonzero, and
these elements are nonzero rational integers. The product is closed under negation,
so it contains positive rational integers. Let d be the smallest such positive rational
integer. The ideal properties of aa show that aa ∩ Z = dZ.
 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD 7

Because the product aa is an ideal, it contains dOF . That is, we have the
containment aa ⊃ dOF , and to complete the proof we need the other containment.
It suffices to show that for any x, y ∈ a the product xy lies in dOF . The quantities
tr(xy) = xy + xy, N(x) = xx, N(y) = yy
all lie in aa ∩ Z = dZ, and so it follows that
tr(xy/d) = tr(xy)/d ∈ Z and N(xy/d) = N(x)/d · N(y)/d ∈ Z.
Thus xy/d ∈ OF , i.e., xy ∈ dOF . This completes the proof. 
The argument that the norm exists made heavy use of the particulars of the
ring OF . In fact, any number ring has an ideal norm, where a number ring is the
ring of integers in any number field , which in turn is any subfield K of C that has
finite dimension as a vector space over Q. However, a norm does not exist for a
general commutative ring.

4. Unique Factorization of Ideals

Proposition 4.1 (“To contain is to divide”). Let a and a0 be ideals of OF . Then
a0 | a ⇐⇒ a ⊂ a0 .
The implication a0 | a =⇒ a ⊂ a0 in this proposition is general, but the ideal
norm will be needed to prove that a ⊂ a0 =⇒ a0 | a. Thus, arguments that use the
latter implication are particular to rings having an ideal norm. Another way to say
this is that to contain is to divide is substantive but to divide is to contain is not.
Proof. If a0 | a then a = a0 a00 for some a00 and so a ⊂ a0 . Conversely, if a ⊂ a0 then
aa0 ⊂ a0 a0 = N(a0 )OF and thus aa0 /N(a0 ) is an ideal of OF . Because a0 · aa0 /N(a0 ) =
a, indeed a0 | a. 
Definition 4.2. An ideal a of OF that is a proper subset of OF is prime if:
For all ideals a0 , a00 of OF , a | a0 a00 =⇒ a | a0 or a | a00 .
An ideal a of OF that is a proper subset of OF is irreducible if:
For all ideals a0 , a00 of OF , a0 a00 = a =⇒ a0 = a or a00 = a.
An ideal a of OF that is a proper subset of OF is maximal if:
For all ideals a0 of OF , a ⊂ a0 =⇒ a0 = a or a0 = OF .
So
prime means doesn’t decompose as a divisor,
and
irreducible means doesn’t decompose as a product,
and
maximal means there is no bigger proper ideal.
(Maximal does not mean bigger than all other ideals.) An exercise shows that the
definition of prime ideal given here is equivalent to:
For all nonzero elements a0 , a00 of OF , a0 a00 ∈ a =⇒ a0 ∈ a or a00 ∈ a,
and furthermore, because every ideal contains 0, the word “nonzero” can be removed
from the previous display. This reproduces the usual definition of a prime ideal in
terms of elements rather than in terms of ideals.
8 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD

Proposition 4.3. Prime ideals of OF are irreducible. Irreducible ideals of OF are

maximal. Maximal ideals of OF are prime.
Proof. Consider any prime ideal a of OF . If a = a0 a00 then a | a0 a00 and so without
loss of generality a | a0 . Also, if a = a0 a00 then a0 | a. Because to divide is to contain,
a0 ⊂ a and a ⊂ a0 , and so a0 = a.
Let a be irreducible, and suppose that a ⊂ a0 . Because to contain is to divide,
a | a, i.e., a0 a00 = a for some a00 , and so a0 = a or a00 = a, but in the latter case
0

a0 = OF by the cancellation law.

Maximal ideals are prime in any commutative ring with 1, because essentially by
definition the quotient of the ring by an ideal is a field precisely when the ideal is
maximal, and the quotient is an integral domain precisely when the ideal is prime.
The reader may work out a more elementary proof here if desired. 
In the proof just given, the argument that irreducible ideals are maximal uses the
to contain is to divide principle that relies on the ideal norm, and it is the substance
of the proposition. The other two statements are basic. With the proposition
established, the proof of unique factorization is prepared.
Theorem 4.4. Any ideal a of OF factors uniquely into prime ideals.
Proof. Let a be an ideal of OF . Then a factors as a finite product of irreducibles
via a process that must terminate by induction on N(a). The fact that irreducibles
are prime makes the factorization unique, because if
a = p1 · · · ps = q1 · · · qt
where the pi are irreducible but not necessarily distinct, and similarly for the qj ,
then
ps | q1 · · · qt
so that because ps is prime we have ps | qt after reindexing if necessary, and thus,
because qt is irreducible,
ps = qt .
By cancellation,
p1 · · · ps−1 = q1 · · · qt−1 .
By induction on the norm, the two factorizations in the previous display are equal,
and hence so are the two factorizations of a. 

5. The Character of a Quadratic Field

We review the completion of the quadratic reciprocity law at 2. The four char-
acters of (Z/8Z)× are shown in the following table:
1 3 5 7 conductor
χ1 1 1 1 1 1
χ−1 1 −1 1 −1 4
χ2 1 −1 −1 1 8
χ−2 1 1 −1 −1 8
The character group is the (2, 2) abelian group, so for example χ−1 χ2 = χ−2 and
χ2i = χ1 for all i. Note that χ−1 has conductor 4 but χ2 and χ−2 are primitive
modulo 8. The characters χi are named to indicate that χ−1 (p) = (−1/p) and
 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD 9

χ2 (p) = (2/p) and χ−2 (p) = (−2/p) for odd primes p, and trivially so for χ1 as
well.
Define m
= χ2 (m), m odd,
2
with χ2 understood to be lifted from (Z/8Z)× to Z. Recall also the definition
 
m
= sgn m, m nonzero.
−1
For any odd prime p we have (p/2)(2/p) = 1 because (p/2) = (2/p). Also we
have (−1/2)(2/−1) = 1 because both (−1/2) and (2/−1) are 1. It follows that
(m/2)(2/m) = 1 for all odd integers m.
The Jacobi/Kronecker symbol (P/Q) for coprime nonzero integers now can be
defined totally multiplicatively in its numerator and denominator. Consider two
coprime nonzero integers,
P = 2a P 0 , P 0 odd
( )
.
Q = 2b Q0 , Q0 odd
Thus a, b ∈ Z≥0 with min{a, b} = 0, and P 0 , Q0 are coprime. The general Ja-
cobi/Kronecker symbol reciprocity formula is
1 if at least one of P, Q is positive
 
P 0 −1 Q0 −1
(P/Q) · (Q/P ) = (−1) 2 · 2 · .
−1 if both P, Q are negative

It follows that uniformly across the four cases of (sgn P, sgn Q),
P 0 −1 Q0 −1
· 2
(1) (P/Q) = (−1) 2 (Q/|P |), P , Q coprime and nonzero.
We will use this formula (1) several times below, rather than the boxed formula
immediately preceding it.
√
Now return to a quadratic field F = Q( n) where n 6= 0, 1 is squarefree. Recall
that the field’s discriminant is defined as
(
n if n = 1 (mod 4)
DF =
4n if n = −1, 2 (mod 4).
Definition 5.1. Let F be a quadratic field with discriminant DF . The quadratic
character of F is
 
DF
χF : Z6=0 −→ C, χF (m) = .
m
Especially, χF (−1) = sgn(DF ).
√
Theorem 5.2. Let F = Q( n) be a quadratic field with discriminant DF . Then
the quadratic character χF has period |DF |.
Proof. Introduce the notation
DF = 2a δ, m = 2b µ (δ and µ odd).
As in (1) above, the Jacobi/Kronecker symbol reciprocity formula gives
 
δ−1 µ−1
· 2 m
χF (m) = (−1) 2 if gcd(m, DF ) = 1.
|DF |
10 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD

Inspect DF in terms of n for n = 1, −1 (mod 4) and n = 2, −2, (mod 8) to see that

(
1 (mod 4) if n = 1 (mod 4) or n = 2 (mod 8)
δ=
−1 (mod 4) if n = −1 (mod 4) or n = −2 (mod 8).
In the second case DF is always even so that µ = m when gcd(m, DF ) = 1. So we
δ−1 m−1
have χF (m) = (DF /m) = (−1) 2 · 2 (m/|DF |), giving
(
(m/|DF |) if n = 1 (mod 4), 2 (mod 8)
χF (m) = if gcd(m, DF ) = 1.
χ−1 (m)·(m/|DF |) if n = −1 (mod 4), −2 (mod 8)
More explicitly, because DF is n or 4n for n = 1, −1 (mod 4) and is 8 · n/2
for n = ±2 (mod 8), and because we can ignore squares in DF when evaluating
(m/|DF |),

χ1 (m) · (m/|n|)


if n= 1 (mod 4)
χ (m) · (m/|n|) if n = −1 (mod 4)
−1
χF (m) = if gcd(m, DF ) = 1.
χ2 (m) · (m/|n/2|)
 if n= 2 (mod 8)

χ−2 (m) · (m/|n/2|) if n = −2 (mod 8)


There is no need to write the trivial term χ1 (m) in the first case of this formula,
but we did so for uniformity. Because χ1 is trivial and χ−1 has conductor 4 while
χ±2 are primitive modulo 8, and because (·/N ) has period N for squarefree odd
positive N , the respective periods are

 |n|
 if n = 1 (mod 4)
4|n| if n = −1 (mod 4)

8|n/2| if n = ±2 (mod 8).


That is, χF has period |DF | in all cases. 

√
In light of Theorem 5.2 we may view the quadratic character of F = Q( n) as
a true Dirichlet character, a homomorphism
χF : (Z/DF Z)× −→ C× ,
defined by

χ1 (m) · (m/|n|)


if n = 1 (mod 4)
χ (m) · (m/|n|) if n = −1 (mod 4)
−1
χF (m + DF Z) =
χ2 (m) · (m/|n/2|)
 if n = 2 (mod 8)

χ−2 (m) · (m/|n/2|) if n = −2 (mod 8).


As usual, we extend the definition to Z/DF Z,

χF (m + DF Z) = 0 if gcd(m, |DF |) > 1.
Because χF (−1) = sgn(DF ), we have
(
1 if F is real quadratic
χF (−1 + DF Z) =
−1 if F is imaginary quadratic.
In general, a Dirichlet character that takes −1 to 1 is called even, and a Dirichlet
character that takes −1 to −1 is called odd . That is, the character of a real quadratic
field is even and the character of an imaginary quadratic field is odd.
 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD 11

For example, if F = Q(i), so that OF = Z[i] is the ring of Gaussian integers,

then the quadratic character is the odd character χF : (Z/4Z)× −→ C× given by
χF (m) = (−4/m) = (−1/m) = χ−1 (m) = (−1)(m−1)/2 .
√ √
If F = Q( 2), so that OF = Z[ 2], then the quadratic character is the even
character χF : (Z/8Z)× −→ C× given by
2
−1)/8
χF (m) = (8/m) = (2/m) = χ2 (m) = (−1)(m .
√
If F = Q( −3), so that OF = Z[ζ3 ] is the ring of Eisenstein integers, then the
quadratic character is χF : (Z/3Z)× −→ C× given by
χF (m) = (−3/m) = (m/3).
√ √
If F = Q( −5), so that OF = Z[ −5], then the quadratic character is χF :
(Z/20Z)× −→ C× given by
χF (m) = (−20/m) = (−1/m)(5/m) = (−1)(m−1)/2 (m/5).
As a remark here, though it is not necessary for the purposes of this writeup, we
state that for any nonzero integer D = 0, 1 (mod 4),
D uniquely takes the form f 2 DF where DF is a quadratic number
field discriminant or DF = 1.
The resulting character m 7→ (D/m) for nonzero integers m therefore has Q pe-
riod |DF |. To establish the statement first write D = 2e2 δ where δ = sgn(D) p pep ,
the product taken over odd prime Q divisors p of D; note that e2Q6= 1 and if e2 = 0
then δ = 1 (mod 4). Let fo = p pbep /2c and let δo = sgn(D) p:ep odd p, so that
δ = fo2 δo and δo is squarefree and δo = δ (mod 4). Now,
• for e2 ≥ 0 even and δ = 1 (mod 4) set f = 2e2 /2 fo and DF = δo ,
• for e2 ≥ 2 even and δ = 3 (mod 4) set f = 2(e2 −2)/2 fo and DF = 4δo ,
• for e2 ≥ 3 odd set f = 2(e2 −3)/2 fo and DF = 8δo = 4 · 2δo , noting that 2δo
is squarefree and 2δo = 2 (mod 4).
As for uniqueness, one can see that every step of the decomposition D = f 2 DF is
forced, including the allocation of powers of 2 between f and DF .
Finally we explain that when the discriminant is even, the quadratic character
has half-period skew periodicity. Specifically, suppose as usual that n 6= 0, 1 is
√ and suppose further that n 6= 1 (mod 4). Thus the quadratic field
squarefree,
F = Q( n) has even discriminant DF = 4n, and the quadratic character has
period |DF |. As an example, for n = −5 the values of the quadratic character
(−20/·) on 0 through 9 and then 10 through 19 are (as we have seen before in the
quadratic reciprocity writeup)
0, 1, 0, 1, 0, 0, 0, 1, 0, 1 and then 0, −1, 0, −1, 0, 0, 0, −1, 0, −1.
Two phenomena are notable here. First, the quadratic residues are concentrated
in the left half of 0 through 19. Generally, not just for n = −5, there are more
residues in the left half than in the right, but the extreme case of all the residues in
the left half is not general. The second notable phenomenon in the previous display
is that the quadratic character on the second half is the negative of the quadratic
character on the first half,
χF (r + |DF |/2) = −χ(r), 0 ≤ r < |DF |/2.
12 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD

We establish the half-period skew periodicity, to be used later in this writeup. The
decomposition
(
× (Z/4Z)× × (Z/|n|Z)× if n = 3 (mod 4),
(Z/|DF |Z) =
(Z/8Z)× × (Z/(|n|/2)Z)× if n = 2 (mod 4)
shows via the Sun Ze theorem that a Dirichlet character of full period modulo |DF |
is naturally viewed as the product of a pair of Dirichlet characters of full period
modulo 4 and |n|, or modulo 8 and |n|/2. (We leave the specifics of showing this to
the reader.) For n = 3 (mod 4), r+|DF |/2 = r+2|n| = r+2 (mod 4). The Dirichlet
character of full period modulo 4 satisfies χ(r + |DF |/2) = χ(r + 2) = −χ(r), and
every Dirichlet character modulo |n| = |DF |/4 satisfies χ(r + |DF |/2) = χ(r);
together these give χF (r + |DF |/2) = −χF (r). Similarly for n = 2 (mod 4),
r + |DF |/2 = r + 2|n| = r + 4 (mod 8), and because the two Dirichlet characters
of full period modulo 8 (these are χ±2 from the beginning of this section) satisfy
χ(r + 4) = −χ(r) it follows again that χF (r + |DF |/2) = −χF (r).

6. Decomposition of Rational Primes

Now we can see the importance of the discriminant. It is the crux of the quadratic
character, which in turn describes the decomposition of rational primes in F as
follows:
Theorem 6.1. Let p be a rational prime. The decomposition of p in OF is

pq where N(p) = N(q) = p if χF (p) = 1

pOF = p where N(p) = p2 if χF (p) = −1

 2
p where N(p) = p if χF (p) = 0.
Thus the decomposition of p in OF depends only on p (mod |DF |).
Proof. (Sketch.) Recall that
√
DF + DF
OF = Z[r] where r= ,
2
and that the polynomial of r is
DF (DF − 1)
f (X) = X 2 − DF X + , with discriminant DF .
4
Let Fp = Z/pZ and let an overbar denote reduction modulo p. There is a natural
isomorphism of quotient rings, whose details will be reviewed below,
Z[r]/pZ[r] ≈ Fp [X]/hf (X)i, ϕ(r) + pZ[r] ←→ ϕ(X) + hf (X)i,
the basic idea being that both sides are the quotient of Z[X] by p and by f (X).
The quotient rings have overrings Z[r] and Fp [X], giving a diagram is as follows:
Z[r] Fp [X]

 
Z[r]/pZ[r] o / Fp [X]/hf (X)i

We seek the prime ideals of Z[r] that contain pZ[r]. Meanwhile, Fp [X] is a Euclidean
ring, hence a PID, so working there is easy. Thus the idea of the proof is to move
 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD 13

the question from the upper left corner Z[r] of the diagram to the upper right
corner Fp [X].
A basic fact from commutative ring theory is that the prime ideals of a ring R
that contain a given ideal J correspond bijectively in the natural way with the prime
ideals of the quotient R/J; here if J is prime in R then the corresponding prime ideal
of R/J is the zero ideal, so we must admit the zero ideal into consideration. Thus,
finding the prime ideals of Z[r] that contain pZ[r], in the upper left corner of the
diagram, reduces to finding the prime ideals of Z[r]/pZ[r], in the lower left corner
of the diagram, which in turn reduces to finding the prime ideals of Fp [X]/hf (X)i,
in the lower right corner of the diagram, which reduces to finding the prime ideals
of Fp [X] that contain hf (X)i, in the upper right corner of the diagram. Because
Fp [X] is a PID this amounts to factoring f (X) in Fp [X], and the quadratic character
value χF (p) = (DF /p) describes the factorization; here the case p = 2 needs to
be checked separately, its three subcases DF = 1 mod 8, DF = 5 mod 8, DF =
0 mod 4 giving quadratic character values χF (2) = 1, −1, 0 and giving f (X) ∈ F2
with 2, 0, 1 roots. So we are done other than writing the specifics.
To do so, start from

(X − α)(X − β) if χF (p) = 1

f (X) = f (X) if χF (p) = −1
 2
(X − α) if χF (p) = 0.


Here α, β ∈ Z and α 6= β. Thus the prime ideal factorization of hf (X)i in Fp [X] is


hX − αihX − βi
 if χF (p) = 1
hf (X)i = hf (X)i if χF (p) = −1

hX − αi2 if χF (p) = 0.


The prime ideals of Fp [X]/hf (X)i are correspondingly


hX − α + hf (X)ii, hX − β + hf (X)ii if χF (p) = 1

hf (X)i (the zero ideal) if χF (p) = −1

hX − α + hf (X)ii if χF (p) = 0.


Pass the ideal generators back through the isomorphism Z[r]/pZ[r] ≈ Fp [X]/hf (X)i
given by ϕ(r) + pZ[r] ←→ ϕ(X) + hf (X)i to get that the prime ideals of Z[r]/pZ[r]
are 
hr − α + pZ[r]i, hr − β + pZ[r]i if χF (p) = 1

pZ[r] (the zero ideal) if χF (p) = −1

hr − α + pZ[r]i if χF (p) = 0.


So finally the prime ideal factorization of pZ[r] in Z[r] is


pq, p = hr − α, pi, q = hr − β, pi if χF (p) = 1

pZ[r] = pZ[r] if χF (p) = −1

 2
p , p = hr − α, pi if χF (p) = 0.
Here the ideals p and q in the first and third cases have norm p. 
As for the natural isomorphism in the proof, the idea is that we may quotient by
f (X) and p in either order because we are quotienting by them both. The general
14 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD

isomorphism that captures this idea is

(A/C) / ((B + C)/C) ≈ A/(B + C) ≈ (A/B) / ((C + B)/B).
Now start from the isomorphisms that show that the upper two rings in the proof’s
u-shaped diagram are quotients, one by f (X) and the other by p,
Z[r] ≈ Z[X]/hf (X)i, ϕ(r) ←→ ϕ(X) + hf (X)i
and
Z[X]/pZ[X] ≈ Fp [X], ϕ(X) + pZ[X] ←→ ϕ(X).
This makes the lower two rings of the diagram isomorphic because they are quotients
by both f (X) and p. More specifically, from the first of these isomorphisms, then
the general isomorphism, and then the second of these isomorphisms,



Z[r]/pZ[r] ≈ Z[X]/hf (X)i / (pZ[X] + hf (X)i)/hf (X)i



≈ Z[X]/pZ[X] / (hf (X)i + pZ[X])/pZ[X]
≈ Fp [X]/hf (X)i,
where explicitly the maps are
ϕ(r) + pZ[r] ←→ ϕ(X) + hf (X)i + (pZ[X] + hf (X)i)
←→ ϕ(X) + pZ[X] + (hf (X)i + pZ[X])
←→ ϕ(X) + hf (X)i.
Thus we have the isomorphism in the proof,
Z[r]/pZ[r] ≈ Fp [X]/hf (X)i, ϕ(r) + pZ[r] ←→ ϕ(X) + hf (X)i.
Other than the use of the quadratic character to determine the factorization of
f (X) in Fp [X], nothing in the argument just given is limited to quadratic fields.
The rest of the argument applies to any number ring of the form Z[r] where r has
polynomial f (X). That is, the decomposition of p as an ideal of OF = Z[r] is
described by the factorization of f (X) in Fp [X]: if
g
Y
f (X) = ϕi (X)ei , deg(ϕi ) = fi for each i
i=1
Pg
with i=1 ei fi = deg(f ), then
g
Y
pOF = pei i , pi = hϕi (r), pi, N(pi ) = pfi for each i
i=1
Pg
and i=1 ei fi = [F : Q]. Further we might hope for this factorization to be
determined at least sometimes by a general reciprocity law.
In situations where a number ring OF does not take the form Z[r], or where it
does but the polynomial of r is complicated, a simple polynomial may still describe
the decomposition of all but finitely many rational
√ primes p in OF . To illustrate
this, recall that our quadratic field is F = Q( n) where n ∈ Z−{0, 1} is squarefree,
that its discriminant DF is n if n = 1 (mod√ 4) and is 4n if n = 2, 3 (mod
√ 4), and
that its integer ring is OF = Z[(D
√ F + DF )/2], properly containing Z[ n] when
n = 1 (mod 4) and equal to Z[ n] when n = 2, 3 (mod 4). The polynomial
f2 (X) = X 2 − n
 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD 15

√
satisfied by n has discriminant 4n, which is 4DF if n = 1 (mod 4) and is DF if
n = 2, 3 (mod 4). The reader can check that the factorization of f2 (X) modulo p
describes the decomposition of p in OF for any odd prime p, but not necessarily
2 2
for p = 2. For example,√ the reduction X − 5 = (X − 1) modulo 2 seems to suggest
that 2 ramifies in Q( 5), but in fact 2 is inert because X 2 − 5X + 5 is irreducible
modulo 2. In general, when r lies in OF and f (X) is the polynomial of r, the
factorization of f (X) modulo p describes the decomposition of p as a product of
ideals in OF if p does not divide the index [OF : Z[r]]. A sufficient condition for
this is that p2 not divide the discriminant of f (X), and so even when we don’t know
how much bigger OF is than Z[r], we do know that the factorization describes the
decomposition
√ for all p whose squares don’t divide the discriminant. For example,
if F = Q( 3 n) where n > 1 is a cubefree integer then the factorization of X 3 − n
modulo p determines the decomposition of p in OF for all p - 3n, because X 3 − n
has discriminant −27n2 .

7. Fractional Ideals and the Ideal Class Group

Definition 7.1. A fractional ideal of F is
b = αa, α ∈ F × , a is an ideal of OF .
Sometimes ordinary ideals are called integral ideals to distinguish them from prop-
erly fractional ideals.
Any fractional ideal forms an abelian group and is closed under multiplication by
elements of OF , but a fractional ideal is not closed under multiplication by elements
of F .
Because multiplication is defined for ideals of OF , it is also defined for fractional
ideals of F ,
αa · α0 a0 = αα0 aa0 .
The multiplication of fractional ideals is commutative and associative. The integer
ring OF is the multiplicative identity. And unlike ordinary ideals, fractional ideals
are invertible. Specifically, if
b = αa
then the calculation αa · (αN(a))−1 a = OF shows that
b−1 = (αN(a))−1 a.
A fractional ideal is principal if it takes the form
b = αhxi, hxi is a principal ideal of OF .
Equivalently, b = βOF where β ∈ F × . The product of principal fractional ideals
is again principal, and the inverse of a principal fractional ideal is again principal.
Thus the principal fractional ideals form a subgroup of the multiplicative group of
fractional ideals of the quadratic field F .
Definition 7.2. The ideal class group of F is the quotient group
Cl(F ) = {fractional ideals of F }/{principal fractional ideals of F }.
The order of the ideal class group is the ideal class number, denoted h(F ).
16 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD

Thus an element of the ideal class group is an ideal class, a set of ideals,
C(b) = {αb : α ∈ F × /OF× }
and the multiplication of the ideal class group is
C(b)C(b0 ) = C(bb0 ).
We will see that the ideal class number is finite. The point here is that
All fractional ideals are principal if and only if all integral ideals are
principal, in which case nonzero elements of OF factor uniquely up
to units. Thus unique factorization of elements holds if the ideal
class group is trivial, i.e., if the ideal class number is 1.
In fact unique factorization of elements holds only if the ideal class group is trivial,
but this result is beyond our scope.
The ideal class group and the ideal class number can be constructed with refer-
ence only to integral ideals. Define two integral ideals a and a0 to be equivalent if
αa = α0 a0 for some nonzero α, α0 ∈ OF . Then the ideal class group is the set of
equivalence classes. However, the benefits of introducing fractional ideals are the
more naturally-motivated group structure of the ideal class group as a true quo-
tient group, and the greater immediacy of the fact that the class group measures
the failure of unique factorization.
The ideal class number h is another structure constant of the field F .

8. Abelian Group Structure of Ideals

The next result is preparation for the pending transition from algebra to geom-
etry in the second part of these notes.
Proposition 8.1. Let b be a fractional ideal of F . Then b takes the form
b = αZ ⊕ βZ,
where α and β are nonzero elements of F and α/β ∈
/ Q.
Proof. Because the fractional ideal takes the form b = αa where α ∈ F × and
a is an ideal of OF , it suffices to prove the result for integral ideals a. Because
aa = N(a)OF , we have
N(a)OF ⊂ a ⊂ OF .
√
Recall that
√
F = Q( n) where n ∈ Z is squarefree, and that OF = rZ ⊕ Z where
r = DF +2 DF . The previously displayed containments are
N(a)rZ ⊕ N(a)Z ⊂ a ⊂ rZ ⊕ Z.
Because the abelian group a sits between two free abelian groups of rank 2, it is
free of rank 2 as well. This point is just a matter of linear algebra over Q. 

Part 2. GEOMETRY: COMPLEX LATTICES

If the quadratic field F is imaginary then its ideals can be interpreted as lattices
in the complex plane having a special property called complex multiplication. Com-
plex geometry shows that the ideal class group of F is finite. Its order, denoted h,
is the ideal class number of F . The goal of these notes is a formula for h.
 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD 17

9. Complex Lattices and Homothety

Definition 9.1. A complex lattice is a rank-2 abelian subgroup of C,
Λ = λ1 Z ⊕ λ2 Z, λ1 , λ2 ∈ C× , λ1 /λ2 ∈
/ R.
Note that in particular, any fractional ideal of an imaginary quadratic field is
a complex lattice. (Proposition 8.1 does the bulk of the work of supporting this
observation—the only loose end is that an imaginary quadratic field contains no
irrational real numbers.)
In the previous definition, the Z-basis {λ1 , λ2 } determines the lattice Λ, but not
conversely. We adopt the convention that lattice bases are ordered and that further-
more they are always
 ordered
 so that Im(λ1 /λ2 ) > 0. Then (exercise, facilitated by
az+b
   
the formula Im cz+d = det( ac db )Im(z)/|cz + d|2 for ac db invertible with real
entries and z a complex number that isn’t real)
Proposition 9.2. The ordered pairs of nonzero complex numbers (λ1 , λ2 ) and
(λ01 , λ02 ) are bases for the same lattice Λ if and only if
 0      
λ1 a b λ1 a b
= for some ∈ SL2 (Z).
λ02 c d λ2 c d
(Here SL2 (Z) is the group of 2-by-2 matrices having integer entries and determi-
nant 1.)
It follows that if Λ is a lattice and (λ1 , λ2 ) is a basis of Λ then the area of the
parallelogram
P (λ1 , λ2 ) = {t1 λ1 + t2 λ2 : t1 , t2 ∈ [0, 1]} where (λ1 , λ2 ) is a basis of Λ
depends only Λ, not on the choice of basis. This is because if (λ1 , λ2 ) and (λ01 , λ02 )
are bases then the linear map taking λi to λ0i for i = 1, 2 preserves area because it
has determinant 1.

Definition 9.3. Two lattices Λ and Λ0 are homothetic if

Λ0 = cΛ for some c ∈ C× .
Homothety is clearly an equivalence relation. It preserves the geometry of any
lattice up to dilation and rotation. We now find a canonical representative of any
equivalence class of lattices under homothety.
Lemma 9.4. Let Λ be a complex lattice. Then Λ has nonzero elements of least
modulus.
Proof. First we show that the lattice point 0 is isolated, meaning that it has a C-
neighborhood containing no other lattice point. This property is preserved under
homothety, so we may assume that our lattice is Λ = τ Z ⊕ Z where Im(τ ) > 0. The
ball about 0 of radius r = min{Im(τ ), 1} contains no other lattice point.
The group structure of the lattice shows that same radius works for a similar ball
about any lattice point. Consequently any bounded subset of the lattice is finite.
The result follows. 
Now let a lattice Λ be given. After applying a homothety, we may assume that
one of its nonzero elements of least modulus is 1. Let τ ∈ Λ be an element of Λ − Z
having least modulus; we may assume that Im(τ ) > 0. Then |τ | ≥ 1, and also
18 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD

|Re(τ )| ≤ 1/2, else some τ + n (where n ∈ Z) has smaller modulus. Thus τ lies in
the fundamental domain,

D = {τ ∈ C : Im(τ ) > 0, |Re(τ )| ≤ 1/2, |τ | ≥ 1}

(see figure 1). And so every lattice is homothetic to a lattice

Λτ = τ Z ⊕ Z, τ ∈ D.

Furthermore, τ is essentially unique. One type of exception to uniqueness is easy

to find: if Re(τ ) = −1/2 then also Λ = (τ + 1)Z ⊕ Z with τ + 1 ∈ D. That is, the
two vertical sides of the fundamental domain should be identified. A second kind
of uniqueness is slightly more subtle: if |τ | = 1 then (writing “∼” for homothety)

Λτ = τ Z ⊕ Z ∼ Z ⊕ τ −1 Z = −τ −1 Z ⊕ Z.

But −τ −1 is also on the circular arc of the boundary of D, being the horizontal
reflection of τ . And so the left and right halves of the semicircular boundary arc
of D should be identified as well. Otherwise, τ is uniquely determined by the
process just described of finding it. To specify unique representatives, we may keep
only the right half of the boundary of D,
√
• Im(τ ) ≥ 3/2,
• −1/2 < Re(τ ) ≤ 1/2,
• |τ | > 1 if Re(τ ) < 0, and |τ | ≥ 1 if Re(τ ) ≥ 0.
A lattice Λτ where τ satisfies the three previous conditions is normalized.

Figure 1. The fundamental domain

Especially, we show that each ideal class in an imaginary quadratic field has
a unique normalized element. Indeed, let b be a fractional ideal of the imaginary
quadratic field F . Let α ∈ F × be a least-norm nonzero element of b. The ideal α−1 b
lies in the class of b and it contains 1 as a least-norm nonzero element. Hence it is
normalized. Any other α0 ∈ F × that is a least-norm nonzero element of b differs
multiplicatively from α by a unit, and so α−1 b is unique. The next section will
show that the normalized ideal α−1 b in the ideal class of b has a further property
that allows only finitely many ideal classes altogether.
 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD 19

10. Complex Multiplication

For any lattice Λ and any integer m we have mΛ ⊂ Λ, the lattice dilating back
into itself or collapsing to 0. But further, some lattices spiral back into themselves
under other multiplications.
Definition 10.1. Let Λ be a lattice. If
mΛ ⊂ Λ for some m ∈ C − Z
then Λ has complex multiplication (CM) by m.
The property of having complex multiplication by m is preserved by homothety,
so we may restrict our attention to lattices Λτ .
To study which such CM-values m are possible for which lattices Λτ , assume
that Λτ has CM by m. Thus
      
τ a b τ a b
m = for some ∈ M2 (Z).
1 c d 1 c d
Thus m is an eigenvalue of the matrix, and [ τ1 ] is an eigenvector. Because we are
assuming that m ∈ / Z, the condition m ∈ R is impossible, e.g., it would force m = d.
So m is an imaginary quadratic algebraic integer. Furthermore, because m = cτ +d
(with c 6= 0), the lattice basis element
m−d
τ=
c
√
lies in the same imaginary quadratic field as m. The field takes the form F = Q( n)
where n ∈ Z<0 is squarefree. Recall that the integer ring OF is
√ (
DF + DF n if n = 1 (mod 4)
OF = Z[r], r = , DF =
2 4n if n = 2, 3 (mod 4).
The key ideas connecting the previous part of this writeup to the present part are:
Fractional ideals of F are complex lattices with CM by the genera-
tor r of the integer ring OF . To find all normalized lattices Λτ ⊂ F
having complex multiplication by r is precisely to find a set of rep-
resentatives of the ideal class group of F .
Recall that normalized means that τ satisfies the three bullets at the end of the
previous section.
The condition for CM by r is
      
τ a b τ a b
r = for some ∈ M2 (Z).
1 c d 1 c d
Thus √
r−d DF − 2d + DF
τ= = , c, d ∈ Z,
c 2c
subject to four conditions on c√and d, three of them rephrasing the three bullets in
the previous section (Im(τ ) ≥ 3/2, −1/2 < Re(τ ) ≤ 1/2, |τ | > 1 if Re(τ ) < 0 and
|τ | ≥ 1 if Re(τ ) ≥ 0), and the fourth to be explained below,
p
• 0 < c ≤ |DF |/3,
• (DF − c)/2 ≤ d < (DF + c)/2,
• 4c2 < (DF − 2d)2 − DF if DF < 2d, 4c2 ≤ (DF − 2d)2 − DF if DF ≥ 2d,
• c | N(r − d).
20 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD

 
For the fourth condition, r satisfies the characteristic polynomial of ac db , i.e.,
r2 − (a + d)r + ad − bc = 0, but also r satisfies the unique minimal polynomial
relation r2 − tr(r)r + N(r) = 0; so tr(r) = a + d, which is to say r − a = −(r − d).
Thus the characteristic polynomial condition (r−a)(r−d)−bc = 0 is N(r−d) = −bc.
Conversely, if N(r−d) = −bc for some b, then let a = tr(r−d)+d; it is easy to verify
that  = a + d and N(r) = ad − bc, so r satisfies the characteristic polynomial
 tr(r)
of ac db .
The four conditions in the bullet list just above show that for a given imaginary
quadratic field F , with its discriminant DF and its integer ring generator r, there
exist only finitely many values τ = (r − d)/c that describe lattices with complex
multiplication. This shows that the class number of F is finite. The τ -values are
easy to find by hand if |DF | is small, and easy to find by algorithm in any case.
Because ideal class representatives are b = hτ, 1i where τ = (r − d)/c with c and d
satisfying the four conditions, integral representatives are a = hr − d, ci. We have
proved
Theorem 10.2. Let F be an imaginary quadratic field. The ideal class number
h(F ) is finite.
We know that there must exist at least one normalized lattice with CM by r,
corresponding to the identity element of the ideal class group. And indeed, the
lattice OF = Λr works. But there may be others.
For example, the reader can use these ideas to show that the integer ring
√ √
OF = Z[ −39+2 −39 ] of the imaginary quadratic field F = Q( −39) gives four
pairs (c, d) = (1, −20), (2, −20), (2, √
−19), (3, −21), and√so its four ideal class
√
repre-
1+ −39 1+ −39 −1+ −39
sentatives hr − d, ci are OF = h 2 , 1i, p1 = h 2 , 2i, p2 = h 2 , 2i,
√ √ √
−3+ −39 1+ −39 1+ −39
and q = h 2 , 3i. Because OF = Z[ 2 ] and the polynomial of 2
is X 2 − X + 10, the proof of Theorem 6.1 shows that 2OF = p1 p2 and 3OF = q2 .
Because the classes of p1 and p2 are distinct and each is the other’s inverse, so that
neither class is its own inverse, the ideal class group is cyclic of order 4 rather than
the product of two cyclic groups of order 2 (these are the only possibilities for a
four-element abelian group).
Our goal is a formula for h(F ) to complement the algorithm that we now have
for it. The formula requires elements of analytic number theory, to be presented
in the third part of this writeup, in addition to the algebra of the first part and
the geometry of this part. Its conceptual content beyond the ideal class number
algorithm is that it relates the algebraic structure constants of our quadratic number
field—its discriminant, descriptors of its unit group, and its ideal class number—to
an analytic datum, a special value of the quadratic L-function of the field, to be
explained. The following proposition will be cited in the course of the analysis to
follow.
Proposition 10.3. Let F be √
an imaginary quadratic field, let DF be the discrimi-
DF + DF
nant of F , and let r = 2 . Consider an integral ideal class representative
a = hr − d, ci with c and d as just above,
and let α denote the area of the parallelogram spanned by r − d and c. Then
N(a) 2
=p .
α |DF |
 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD 21

Note that the right side is independent of c and d. That is, the ideal norm
(an algebraic quantity) is the parallelogram area (a geometric quantity) times a
constant that depends only on the field F .
p
Proof. The parallelogram area is α = c |DF |/2, so we need to show that N(a) = c.
Let b = −N(r − d)/c. Because aa = N(a)OF and
aa = hN(r − d), (r − d)c, (r − d)c, c2 i = chb, r − d, r − d, ci,
it suffices to show that hb, r − d, r − d, ci contains 1. It contains the element
g = gcd(b, tr(r) − 2d, c).
To show that g = 1, note that the quantities tr(r)−2d = tr(r−d) and −bc = N(r−d)
are the coefficients of the polynomial of r − d, which has the same discriminant
DF as the polynomial of r because d is real. Also, g 2 divides both terms of this
discriminant and hence divides the discriminant altogether. In symbols,
g 2 | (tr(r) − 2d)2 + 4bc = (tr(r) − 2d)2 − 4N(r − d)
= tr(r)2 − 4tr(r)d + 4d2 − 4N(r) + 4tr(r)d − 4d2
= tr(r)2 − 4N(r) = DF2 − DF (DF − 1) = DF .
√
Recall that our quadratic field is F = Q( n) with n a squarefree negative integer.
If n = 1 (mod 4) then DF = n is squarefree, and so g = 1 and we are done. On the
other hand, if n = 2, 3 (mod 4) then DF = 4n and so g could equal 1 or 2. To show
that g = 1 in this case as well, note that the equality bc = −N(r) + tr(r)d − d2 in
the calculation just carried out is now
DF (DF − 1)
bc = − + DF d − d2 = −n(4n − 1) + 4nd − d2 = n − d2 (mod 4).
4
Thus bc 6= 0 (mod 4) because n = 2, 3 (mod 4) and d2 = 0, 1 (mod 4). So at least
one of b, c is odd, disallowing the possibility that g = gcd(b, tr(r) − 2d, c) is 2. 

Part 3. ANALYSIS: ZETA AND L-FUNCTIONS OF AN

IMAGINARY QUADRATIC FIELD
To obtain the class number formula, we encode information about the imaginary
quadratic field F in Dirichlet series, series of the form
X an
f (s) = , s ∈ C.
+
ns
n∈Z
The various Dirichlet series in question—the Euler–Riemann zeta function, the qua-
dratic L-function of F , and the Dedekind zeta function of F —have useful complex
analytic properties that combine with the number theoretic information that they
encode to give the class number formula.

11. Summation by Parts and Dirichlet Series Convergence

Let {an }n≥1 and {bn }n≥1 be complex sequences. Define
n
X
An = ak for n ≥ 0 (including A0 = 0),
k=1
so that
an = An − An−1 for n ≥ 1.
22 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD

Also define
∆bn = bn+1 − bn for n ≥ 1.
Then for any 1 ≤ m ≤ n, the summation by parts formula is
n−1
X n−1
X
ak bk = An−1 bn − Am−1 bm − Ak ∆bk .
k=m k=m
The formula is easy to verify in consequence of
ak bk = Ak bk+1 − Ak−1 bk − Ak ∆bk , k ≥ 1,
noting that the first two terms on the right side telescope when summed.
Proposition 11.1. Let {an }n≥1 be a complex sequence such that for some positive
numbers C and r,
n
X
ak ≤ Cnr for all large enough n.
k=1
Then the Dirichlet series
X an
f (s) = , s∈C
+
ns
n∈Z

is complex analytic on the open right half plane {Re(s) > r}. If furthermore {an }
is a nonnegative real sequence then f (s) converges absolutely on {Re(s) > r}.
Proof. Let
{bn } = {n−s }.
Then summation by parts gives for 1 ≤ m ≤ n,
n−1 n−1  
X ak An−1 Am−1 X 1 1
= − − Ak − s .
ks ns ms (k + 1)s k
k=m k=m
Introduce the notation
s = σ + it, σ > r,
s σ +
so that |x | = x for all x ∈ R , and estimate that
Z k+1 Z k+1
1 1 −s−1
− s = −s t dt ≤ |s| t−σ−1 dt < |s|k −σ−1 .
(k + 1)s k k k

We are given that |Ak | ≤ Ck r for all large enough k, and so the summation by
parts from a moment ago says that for all large enough 1 ≤ m ≤ n,
n−1 n−1
!
X ak 1 1 X 1
≤C + σ−r + |s| .
ks nσ−r m k σ−r+1
k=m k=m
Recall that σ > r. Let n → ∞ to see that for all large enough m ≥ 1,
∞ ∞
!
X ak 1 X 1
≤C + |s| .
ks mσ−r k σ−r+1
k=m k=m
Because σ > r, the right side goes to 0 as m → ∞. As s varies through a compact
subset K of the open right half plane {σ > r}, the right side goes to 0 at a rate that
depends only onP min{σ : σ + it ∈ K} and max{|s| : s ∈ K}, and thus the Dirichlet
series f (s) = n∈Z+ an n−s converges uniformly on K. Because the partial sums
of f (s) are analytic on {Re(s) > r}, their uniform convergence on compacta is the
 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD 23

hypothesis for a standard theorem of complex analysis that then says that f (s) is
analytic on {Re(s) > r} as well.
Now assume that an ∈ R≥0 for all n. Because f (s) converges on {Re(s) > r}, it
converges for any s = σ > r. But for general s = σ + it where σ > r we have
an an
s
= σ,
n n
And so f (s) converges absolutely because f (σ) converges. 
A slogan-encapsulation of Proposition 11.1 is
n
X X an
ak = O(nr ) =⇒ is well-behaved on {Re(s) > r}.
ns
12. The Euler–Riemann Zeta Function
Definition 12.1. The Euler–Riemann zeta function is formally
X Y
ζ(s) = n−s = (1 − p−s )−1 .
n∈Z+ p∈P

The formal equality of the sum and the product follows from the geometric series
formula and then the unique factorization of positive integers,
Y Y X X
(1 − p−s )−1 = (pep )s = n−s .
p∈P p∈P ep ≥0 n∈Z+

Proposition 12.2 (Properties of the Euler–Riemann Zeta Function). The function

ζ(s) is complex analytic on the open right half plane {Re(s) > 1}, where the formal
equality of the sum and product expressions of ζ(s) is analytically valid. The func-
tion ζ(s) extends meromorphically to the open right half plane {Re(s) > 0}, and
the extension has only a simple pole at s = 1 with residue 1. That is,
1
ζ(s) = + ψ(s), Re(s) > 0
s−1
where ψ is analytic.
Proof. The fact that ζ(s) is complex analytic on Re(s) > 1 follows from Proposi-
tion 11.1 with C = r = 1 because {ai } is the sequence {1, 1, 1, . . . }. For any bound
B > 0 we have the identity
X Y
n−s = (1 − p−s )−1 ,
n= p<B pep
Q
p<B

using the condition Re(s) > 1 to rearrange the terms because the sum converges
absolutely. As B → ∞ the sum converges to ζ(s) because it converges absolutely
and thus the order of summation is irrelevant. Consequently the product converges
to ζ(s) as well. For the last statement, compute that
Z ∞ ∞ Z n+1 ∞ Z n+1
1 X X
= t−s dt = t−s dt = ζ(s) + (t−s − n−s ) dt.
s−1 1 n=1 n n=1 n

Call the sum −ψ(s). Because for all t ∈ [n, n + 1] we have

Z t Z t
−s −s −s−1
|t − n | = |s x dx| ≤ |s| x−σ−1 dx ≤ |s|n−σ−1 (t − n) ≤ |s|n−σ−1 ,
n n
24 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD

it follows that
n+1
|s|
Z
(t−s − n−s ) dt ≤ ,
n nσ+1
and so −ψ(s) converges to an analytic function on {Re(s) > 0} by the convergence
properties of |s| n n−σ−1 .
P


13. The L-Function of a Quadratic Field

Recall the quadratic character of a quadratic field F , defined using the discrim-
inant of F and the extended Jacobi symbol,
 
DF
χF : Z+ −→ Z, χF (n) = .
n
Because the symbol DF subsumes the information formerly contained in the sym-
bol n for describing the quadratic field F , we have liberated n for the previous
display and the sequel.
Definition 13.1. Let F be a quadratic field with discriminant DF . The quadratic
L-function of F is formally
X Y
L(χF , s) = χF (n)n−s = (1 − χF (p)p−s )−1 .
n∈Z+ p∈P

The formal equality of the sum and the product follows similarly to the Euler–
Riemann zeta function because χ is multiplicative. Because the quadratic character
encodes the decomposition of rational primes in OF (Theorem 6.1), so does the
quadratic L-function.
Proposition 13.2 (Properties of the Quadratic L-Function). The quadratic L-
function L(χF , s) is complex analytic on {Re(s) > 0}. The formal equality of the
sum and product expressions of L(χF , s) is analytically valid for Re(s) > 1.
Proof. By Theorem 5.2, χF (n) depends only on n (mod |DF |)). So for any no ∈ Z+ ,
no +|DF |−1
X
χF (n) = 0,
n=no

because we are summing the nontrivial character χF over the group (Z/DF Z)× . It
follows that for all n ≥ 1,
Xn
χF (k) < C.
k=1
Now Proposition 11.1 shows that L(χF , s) is analytic on {Re(s) > 0}. The ar-
gument that the sum and the product are equal is essentially the same as for the
Euler–Riemann zeta function, requiring Re(s) > 1 for absolute convergence so that
terms can be rearranged. 

The next result evaluates L(χF , 1) as a constant factor times a roughly |DF |/2-
fold sum of values weighted by the quadratic character. The values are logarithms
of sines if F is real quadratic and they are simply 1 if F is imaginary quadratic.
The value L(χF , 1) will figure in the class number formula.
 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD 25

Proposition 13.3 (Special Value of the Quadratic L-Function). For a real qua-
dratic field F ,
2 X
L(χF , 1) = − √ χF (r) log(sin(πr/DF )).
DF
1≤r<DF /2

For an imaginary quadratic field F ,

|DF |−1
π X
(2) L(χF , 1) = − 3/2
χF (r)r,
|DF | r=1

and further this formula simplifies to

π X
(3) L(χF , 1) = p χF (r).
(2 − χ(2)) |DF | 1≤r<|D
F |/2

Before proving the proposition, we make two comments. First, the quantity
P|DF |−1
r=1 χF (r)r that arises in the imaginary quadratic case is the first so-called χF -
Bernoulli number , B1,χF , where the χF -Bernoulli numbers in general are defined
by a generating function,
|DF |−1
X χF (r)tert X tk
t
= Bk,χF .
r=1
e −1 k!
k≥0
xt
Indeed, from the definition of the Bernoulli polynomials, etet −1 = k≥0 Bk (x)tk /k!,
P
the left side of the previous display is, after reversing a double sum,
|DF |−1
X χF (r)tert X |DXF |−1
tk
= χF (r)Bk (r) ,
r=1
et − 1 r=1
k!
k≥0
P|DF |−1
so that Bk,χF = r=1 χF (r)Bk (r). In particular, because B1 (r) = r − 1/2
P P|DF |−1
and r χF (r) = 0, we have B1,χF = r=1 χF (r)r as claimed. Of course there
are χ-Bernoulli numbers for any Dirichlet character χ, not only for our particular
character χF .
Second, to get (3) from (2) in the imaginary quadratic case, introduce the nota-
tion
X X X
S= χF (r)r, S1 = χF (r)r, S0 = χF (r);
1≤r<|DF | 1≤r<|DF |/2 1≤r<|DF |/2

what needs to be√shown is that S = (|DF |/(χ(2)−2))S0 . If the imaginary quadratic

field is F = Q( n) for negative squarefree n = 1 (mod 4), so that DF = n, the
relations
 X


 χF (r)r + χ(|DF | − r)(|DF | − r)

 1≤r<|D |/2
F
S= X



 χF (2r)2r + χ(|DF | − 2r)(|DF | − 2r)

1≤r<|DF |/2

and the fact that χF is odd and quadratic give S = 2S1 − |DF |S0 and χ(2)S =
4S1 − |DF |S0 . √These two relations give S = (|DF |/(χ(2) − 2))S0 as desired. If
instead F = Q( n) for negative squarefree n = 2, 3 (mod 4), so that DF = −4n and
26 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD

now χF (2) = 0, then recall from the end of section 5 that χF (r+|DF |/2) = −χF (r).
Consequently,
X

S= χF (r)r + χF (r + |DF |/2)(r + |DF |/2) = (−|DF |/2)S0 ,
1≤r<|DF |/2

and again we have S = (|DF |/(χ(2) − 2))S0 as desired. Now we proceed to the
proof of Proposition 13.3, needing to establish only (2) in the imaginary quadratic
case.

Proof. Recall that χF has period |DF |. Compute that for Re(s) > 1 (so that we
may rearrange the terms),
|DF |−1
X X X
−s
L(χF , s) = χF (n)n = χF (t) n−s .
n∈Z+ t=0 n∈Z +
n=t (|DF |)

The inner sum is

(
X
−s
X
−s 1 if n = t (mod |DF |)
n = an (t)n where an (t) =
0 6 t (mod |DF |),
if n =
n∈Z+ n∈Z+
n=t (|DF |)

and the casewise coefficient has a uniform description as a character sum,

|DF |−1
1 X (t−n)r
an (t) = ζ , where ζ|DF | = e2πi/|DF | .
|DF | r=0 |DF |

Thus we have for Re(s) > 1,

|DF |−1 |DF |−1
X X 1 X (t−n)r
L(χF , s) = χF (t) ζ n−s
t=0
|DF | r=0 |DF |
n∈Z+
|DF |−1 |DF |−1
1 X X
rt
X
−nr −s
= χF (t)ζ|D F|
ζ|DF|
n .
|DF | r=0 t=0 + n∈Z

Let τr (χF ) and τ (χF ) respectively denote the variant Gauss sum of χF that has
appeared in the calculation and the basic Gauss sum of χF ,
|DF |−1
X
rt
τr (χF ) = χF (t)ζ|D F|
, τ (χF ) = τ1 (χF ).
t=0

The rth variant Gauss sum is the character value at r times the basic Gauss sum,
τr (χF ) = χF (r)τ (χF ).
When gcd(r, |DF |) = 1 this equality follows from a quick substitution, but when
gcd(r, |DF |) > 1 the equality (which says in this case that τr (χF ) = 0; in particular
τ0 (χ) = 0 and so there is no need to sum over r = 0) relies on the fact that χF
has no period smaller than |DF |. See the handout on continuations and functional
equations for the argument. Furthermore, as shown at the end of the ninth online
lecture for this course, if we set δ = 0 for an even quadratic character, such as arises
from a real quadratic field, and if we set δ = 1 for an odd quadratic character, such
 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD 27

as arises from an imaginary quadratic field, then the basic Gauss sum of a quadratic
character is
τ (χF ) = iδ |DF |1/2 .
Returning to our computed value, no longer bothering to sum over r = 0,
|DF |−1
1 X X
−nr −s
L(χF , s) = τr (χ) ζ|DF|
n
|DF | r=1 + n∈Z
|DF |−1
τ (χF ) X X
−nr −s
= χF (r) ζ|DF|
n
|DF | r=1 + n∈Z
|DF |−1
iδ X X
−nr −s
= χF (r) ζ|DF|
n ,
|DF |1/2 r=1 + n∈Z
+
let s → 1 to get
|DF |−1
iδ X
−r
L(χF , 1) = 1/2
χF (r) log(1 − ζ|DF|
)−1 .
|DF | r=1
Let S denote the sum in the previous display,
|DF |−1
X
−r
(4) S=− χF (r) log(1 − ζ|DF|
).
r=1
A little algebraic manipulation, or a small exercise in geometry, gives the polar
decomposition
−r
1 − ζ|DF|
= 2 sin(πr/|DF |)ei(π/2−πr/|DF |) .
r −r
Also 1 − ζ|DF|
is the complex conjugate of 1 − ζ|DF|
. Thus, from the general formula
iθ
log(re ) = log(r) + iθ,
∓r
log(1 − ζ|D F|
) = log(2 sin(πr/|DF |)) ± i(π/2 − πr/|DF |).
Consequently,
(
−r r 2 log(2 sin(πr/|DF |)) for “+”
log(1 − ζ|DF|
) ± log(1 − ζ|DF|
) =
2i(π/2 − πr/|DF |) for “−”.
If F is real quadratic then χF is even and DF is positive, and so substituting
DF − r for r in (4) gives that also (repeating (4) in the next display and also giving
a second expression for S)
F −1
DX F −1
DX
−r r
S=− χF (r) log(1 − ζDF
) =− χF (r) log(1 − ζD F
).
r=1 r=1
Add the values of S shown in the previous display, making reference to the penul-
timate display, to get
F −1
DX
S=− χF (r) log(2 sin(πr/DF )).
r=1
P
We may drop the 2 from the input to the logarithm because r χF (r) = 0.
And so in the real quadratic case, we get the claimed result upon multiplying S
28 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD

√
by 1/ DF and then using the symmetry of the sine function about π/2 and the
value sin(π/2) = 1,
F −1
DX
1
L(χF , 1) = − √ χF (r) log(sin(πr/DF ))
DF r=1
2 X
= −√ χF (r) log(sin(πr/DF )).
DF 1≤r<D /2
F

If F is imaginary quadratic then still we have

(
−r r 2 log(2 sin(πr/|DF |)) for “+”
log(1 − ζ|DF | ) ± log(1 − ζ|DF | ) =
2i(π/2 − πr/|DF |) for “−”,

but now χF is odd, and so substituting |DF |−r for r in (4) gives that also (repeating
(4) in the next display and also giving a second expression for S)
|DF |−1 |DF |−1
X X
−r r
S=− χF (r) log(1 − ζ|DF|
) = χF (r) log(1 − ζ|DF|
).
r=1 r=1

Add the values of S shown in the previous display, making reference to the penul-
timate display, to get
|DF |−1 |DF |−1
X πi X
S= χF (r)i(πr/|DF | − π/2) = χF (r)r.
r=1
|DF | r=1

And so in the imaginary quadratic case, we get the claimed result upon multiply-
ing S by i/|DF |1/2 ,
|DF |−1
π X
L(χF , 1) = − 3/2
χF (r)r.
|DF | r=1

14. The Dedekind Zeta Function of a Quadratic Field

The symbol a denotes an integral ideal throughout this section, as compared to
a nonintegral fractional ideal.
Definition 14.1. Let F be a quadratic field. The Dedekind zeta function of F is
formally (summing over ideals of OF and multiplying over irreducible ideals of OF )
X Y
ζF (s) = N(a)−s = (1 − N(p)−s )−1 .
a p

The formal equality of the sum and the product follows similarly to the Euler–
Riemann zeta function, this time because integral ideals factor uniquely and because
the norm is multiplicative.
The Dedekind zeta function rearranges as a Dirichlet series,
X an
ζF (s) = , where an = #{a : N(a) = n}.
+
ns
n∈Z
 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD 29

Figure 2. Parallelogram-tessellation approximations of a complex

ball; inner, outer balls of the approximations

To analyze ζF (s) using summation by parts, we need to introduce

n
X
An = ak = #{a : N(a) ≤ n}, n ≥ 1,
k=1

and estimate An . To carry out the estimate, we will define for each of the finitely
many ideal classes C of F
An (C) = #{a ∈ C : N(a) ≤ n}, n ≥ 1,
P
so that An = C An (C). The problem of estimating each An (C) can be reduced
to an estimation problem in the principal class. The principal class estimation
problem is a matter of estimating the number of lattice points in a disk. Thus the
following lemma will provide the key result that we need.
Lemma 14.2. Let Λ be a complex lattice, and let α denote the area of any of its
fundamental parallelograms,
P (λ1 , λ2 ) = {t1 λ1 + t2 λ2 : t1 , t2 ∈ [0, 1]} where (λ1 , λ2 ) is a basis of Λ.
As noted just after Proposition 9.2, α is well defined. For any r > 0 let Br denote
the closed complex ball of radius r. Then for some positive constant C,
πr2
#((Λ − 0) ∩ Br ) − ≤ Cr for all r ≥ 1.
α

Proof. The geometric objects in this proof are shown in figure 2. The thick circle
is the boundary of Br . Both #(Λ ∩ Br ) and πr2 /α lie between the number of
light parallelograms and the number of parallelograms altogether. The light par-
allelograms have more area than the inner circle, all the parallelograms less area
30 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD

than the outer circle. Consecutive circle radii differ by the longer diagonal of the
parallelograms.
Fix a fundamental parallelogram P , and for any λ ∈ C let Pλ denote the λ-
translate of P . For any r ≥ 0 let
n1 (r) = #{λ ∈ Λ : Pλ ⊂ Br },
n2 (r) = #{λ ∈ Λ : Pλ ∩ Br 6= ∅}.
Then
n1 (r) ≤ #(Λ ∩ Br ) ≤ n2 (r).
Let δ > 0 be the length of the longer diagonal of P . Then for any r ≥ δ,
π(r − δ)2 ≤ n1 (r)α ≤ πr2 ≤ n2 (r)α ≤ π(r + δ)2 ,
and dividing by α gives
π(r − δ)2 πr2 π(r + δ)2
≤ n1 (r) ≤ ≤ n2 (r) ≤ .
α α α
Thus #(Λ ∩ Br ) and πr2 /α both lie in [π(r − δ)2 /α, π(r + δ)2 /α]. Consequently
the absolute value of their difference is at most the interval length,
πr2
 
4πδ
#(Λ ∩ Br ) − ≤ r for all r ≥ δ.
α α
The function f (r) = |#(Λ ∩ Br ) − πr2 /α|/r is bounded on [1, δ], and so in fact
πr2
#(Λ ∩ Br ) − ≤ Cr for all r ≥ 1.
α
Finally, excluding 0 from Λ ∩ Br changes the left side by at most r because r ≥ 1.
The result follows. 

Recall that we are interested in the Dedekind zeta function of the imaginary
quadratic field F , whose Dirichlet series is
X an
ζF (s) = , where an = #{a : N(a) = n}.
+
ns
n∈Z

As we did for the Euler–Riemann zeta function and for the quadratic field L-
function, we want to estimate the absolute values of the sums
n
X
An = ak = #{a : N(a) ≤ n}.
k=1

Because the an are nonnegative, so are the An . We √ show that An is roughly a

certain constant multiple of n, within a multiple of n. The constant contains the
ideal class number.
Proposition 14.3. Let F be an imaginary quadratic field. Let DF denote the
discriminant of F , let w denote the number of roots of unity in F , and let h denote
the ideal class number of F . Then

2πhn √
An − p < C n, n ≥ 1.
w |DF |
 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD 31

Proof. Let C be any ideal class of F , and let ao ∈ C −1 be any integral ideal in the
inverse class of C. Then the map
b 7−→ ao b
is a bijection of the fractional ideals of F . In particular, it restricts to a bijection
between two sets of integral ideals,
∼
{a ∈ C : N(a) ≤ n} −→ {principal a0 : ao | a0 and N(a0 ) ≤ n N(ao )}.
Equivalently, because to contain is to divide and because the ideal norm is the
absolute value of the element norm, which is the square of the element absolute
value,
∼ p
{a ∈ C : N(a) ≤ n} −→ {hxi ⊂ ao : x 6= 0, |x| ≤ n N(ao )}.
As in the discussion leading into Lemma 14.2, define
An (C) = #{a ∈ C : N(a) ≤ n}, n ≥ 1.
Because associate elements generate the same ideal, and because all units of OF
are roots of unity because F is imaginary quadratic, the previous set bijection gives
 
# (ao − 0) ∩ B√n N(a )
o
(5) An (C) = .
w
Now specifically take ao = hr − d, ci as in Proposition 10.3, and let αo denote the
p of the parallelogram spanned by c and r − d. By (5) and by the relation
area
2/ |DF | = N(ao )/αo from Proposition 10.3, and then by Lemma 14.2,

2πn 1   πn N(a ) √
# (ao − 0) ∩ B√n N(a ) −
o
An (C) − p = < C n.
w |DF | w o αo
The constant C in the previous display depends on the ideal class C. Finally,
because X
An = An (C), n ≥ 1,
C∈Cl(F )

sum over ideal classes and use the triangle inequality to get
2πhn √
An − p ≤ C n,
w |DF |
where now the constant C is independent of ideal classes. 

Proposition 14.4 (Properties of the Dedekind Zeta Function). Let F be an imag-

inary quadratic field. The Dedekind zeta function ζF (s) is analytic on {Re(s) > 1},
where the formal equality of the sum and product expressions of ζF (s) is analyt-
ically valid. Furthermore, the Dedekind zeta function of F is the product of the
Euler–Riemann zeta function and the quadratic L-function of F .
ζF (s) = ζ(s)L(χF , s), Re(s) > 1.
The function ζF (s) extends meromorphically to the open right half plane {s > 0},
and the extension has only a simple pole at s = 1 with residue L(χF , 1). That is,
L(χF , 1)
ζF (s) = + ψ(s), Re(s) > 0
s−1
32 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD

where ψ is analytic. Thus

lim (s − 1)ζF (s) = L(χF , 1).

s→1

Proof. Recall that An ≥ 0. Compute that by Proposition 14.3,

2πhn 2πhn √
An − p ≤ An − p ≤ C n,
w |DF | w |DF |

so that An ≤ Cn. The analyticity of ζF (s) on {Re(s) > 1} follows from Proposi-
tion 11.1.
For the equality of the sum and product expressions of ζF (s), recall yet again
that the terms of the sum rearrange as the Dirichlet series
X an
ζF (s) = , where an = #{a : N(a) = n}.
+
ns
n∈Z

By the last statement of Proposition 11.1, the Dirichlet series converges absolutely
on {Re(s) > 1}. Hence so does its rearrangement a N(a)−s , and now an argument
P
similar to the argument for theQ Euler–Riemann zeta function shows the equality of
this last sum and the product p (1 − N(p)−s )−1 on {Re(s) > 1}.
As for the factorization of ζF (s), because
 Y Y 
−s −1


 ζ F (s) = (1 − N(p) ) 


p p|pOF
 
Y , Re(s) > 1,
−s −1 −s −1 



 ζ(s) L(χ F , s) = (1 − p ) (1 − χ F (p)p ) 


p

it suffices to show that for each rational prime p,

Y
(1 − N(p)−s ) = (1 − p−s )(1 − χF (p)p−s ).
p|pOF

But by Theorem 6.1, the decomposition of a rational prime in OF is


pq where N(p) = N(q) = p if χF (p) = 1

pOF = p where N(p) = p2 if χF (p) = −1

 2
p where N(p) = p if χF (p) = 0,

and so
• if χF (p) = 1 then both sides are (1 − p−s )2 ,
• if χF (p) = −1 then both sides are 1 − p−2s ,
• and if χF (p) = 0 then both sides are 1 − p−s .
Finally, the meromorphic continuation of ζF (s) follows from the properties of ζ(s)
and of L(χF , s) because ζF (s) = ζ(s)L(χF , s) for Re(s) > 1. 

Thus the equality ζF (s) = ζ(s)L(χF , s) is an analytic encoding of the arithmetic

of OF .
 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD 33

15. The Class Number Formula

We have not yet used the full strength of Proposition 14.3. Recall its statement
that if
Xn
an = #{a : N(a) = n} and An = ak , n ≥ 1
k=1
then
2πhn √
An − p < C n, n ≥ 1.
w |DF |
So far we have used this only to show that An is O(n). To use the estimate in the
previous display incisively, let
2πh
ãn = an − p , n ≥ 1,
w |DF |
so that the partial sums of the ãn are

en = An − 2πhn
A p , n ≥ 1.
w |DF |
Thus the estimate is |Aen | ≤ C √n for n ≥ 1, with the power of n now 1/2 rather
than 1, and so Proposition 11.1 says that the Dirichlet series
X ãn 2πh
f (s) = s
= ζF (s) − p ζ(s)
+
n w |DF |
n∈Z

is analytic on {Re(s) > 1/2}. In particular it is analytic at s = 1. Because

2πh
f (s) = ζF (s) − p ζ(s) is analytic at s = 1,
w |DF |
and because
L(χF , 1) 1
ζF (s) ∼ and ζ(s) ∼ ,
s−1 s−1
it follows that
2πh
L(χF , 1) = p .
w |DF |
That is,
The tight estimate of Proposition 14.3 shows that the ideal class
number of F appears in the residue of the Dedekind zeta func-
tion ζF (s) at s = 1. By Proposition 14.4, the residue is L(χF , 1),
for which Proposition 13.3 gives a formula.
In sum,
Theorem 15.1 (Dirichlet Class Number Formula for Imaginary Quadratic Fields).
Let F be an imaginary quadratic field. Let DF denote the discriminant of F , let w
denote the number of roots of unity in F , and let h denote the ideal class number
of F . Let L(χF , s) be the quadratic L-function of F . Then
2πh
p = L(χF , 1).
w |DF |
34 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD

Conceptually the boxed formula is best left as it is, with the field structure
constants h and w and DF on one side and with the analytic quantity L(χF , 1) on
the other. However, because we haveP
formula (3) (page 25) from Proposition 13.3,
π√
stating that L(χ, 1) = 1≤r<|DF |/2 χF (r), the class number is
(2−χ(2)) |DF |

w/2 X
h= χF (r).
2 − χ(2)
1≤r<|DF |/2
√
As an example, let F = Q( −5), so that DF = −20. We have seen that the
corresponding quadratic character is
(
× 1 if t = 1, 3, 7, 9
χF : (Z/20Z) −→ {±1}, χF (t) =
−1 if t = 11, 13, 17, 19,
and so the sum in the boxed formula for h is 4 and the formua gives
1
h= 4,
2−0
which is to say, √
The class number of Q( −5) is 2.
√
As a second example, let F = Q( −39). The quadratic character on (Z/39Z)×
is (·/3)(·/13), giving
(
1 if t = 1, 2, 4, 5, 8, 10, 11, 16 (and 20, 22, 25, 32)
χF (t) =
−1 if t = 7, 14, 17, 19 (and 23, 28, 29, 31, 34, 35, 37, 38)
(as noted earlier in this writeup, the squares are concentrated more in the left half),
and so the ideal class number is
1
h= (8 − 4),
2−1
which is to say, √
The class number of Q( −39) is 4.
In fact, we have already seen on page 20 that the ideal class group is cyclic of
order 4.
√
As a third example, let F = Q( −163), noting that 163 is prime. The quadratic
character is   
−163 r 
χF (r) = = .
r 163
One readily checks that 2 is a generator modulo 163 by using fast modular exponen-
tiation or a machine to compute that 281 = −1 (mod 163) and 254 = 104 (mod 163).
Thus the squares modulo 163 are the even powers of 2 reduced modulo 163, and
similarly for the nonsquares. In particular χF (2) = −1, and one can compute by
P81
hand or by machine that r=1 χF (r) = 3. So the class number formula gives
1
h= 3,
2+1
which is say, √
The class number of Q( −163) is 1.
 THE IDEAL CLASS NUMBER FORMULA FOR AN IMAGINARY QUADRATIC FIELD 35

It is for reasons related to the class number being 1 that the number
√
eπ 163/3
= 640320.0000000006 . . .
is so nearly an integer.

Recall that for a real quadratic number field F , there is a unique smallest fun-
damental unit u > 1 in the unit group OF× . We end by stating that in this case the
class number theorem is similar to Theorem 15.1 but incorporates the fundamental
unit,
2 log(u)h
√ = L(χF , 1).
DF
Much of the argument is similar to the work in this writeup. The main difference
is that the counterpart of Proposition 14.3 requires a different approach.