You are on page 1of 38

Bol. Soc. Mat. Mex.

(2014) 20:199–236
DOI 10.1007/s40590-014-0024-8

ORIGINAL ARTICLE

On binary integral quadratic forms having isometric


groups of automorphisms

José María Montesinos Amilibia

Received: 5 June 2013 / Revised: 4 February 2014 / Accepted: 27 March 2014 /


Published online: 12 June 2014
© Sociedad Matemática Mexicana 2014

Abstract The concept of Bianchi equivalence between integral quadratic forms is


introduced. Two n-ary integral quadratic forms F and G are said to be Bianchi equiv-
alent if there is a real isometry M, from F to ±G, sending the group of automor-
phisms Aut (F) of F isomorphically onto Aut (G). An integer v > 2 is defined
for every binary, indefinite, integral quadratic form F with non-square discriminant
such that two of them are Bianchi-equivalent if and only if their v-invariants coin-
cide and both are bilateral or unilateral. The number of Z-classes inside such an
indefinite Bianchi-class is finite and a practical procedure to construct the Z-classes
with the same v-invariant is given. It is proved also that an integral quadratic form
F = (a, b, c) with a = 0, b = 0, c = 0 is bilateral if and only if the integral quadratic
form F  = (a 2 , 2b2 − ac, c2 ) represents 4b2 . The cases of definite forms an indefi-
nite forms with square discriminant are also completely studied. Finally some partial
results around the so called Eisenstein problem are given.

Keywords Integral quadratic form · Knot · Link · Hyperbolic manifold · Volume ·


Automorph · Commensurability class

Dedicated to Fico González-Acuña on his 70th birthday.

Dedicado con agradecimiento y afecto a Fico González Acuña en su septuagésimo cumpleaños.

J. M. Montesinos Amilibia (B)


Facultad de Matemáticas, Universidad Complutense, 28040 Madrid, Spain
e-mail: montesin@mat.ucm.es
200 J. M. Montesinos Amilibia

Mathematics Subject Classification 11E04 · 11E20 · 57M25 · 57M50 · 57M60

1 Introduction

The theory on integral quadratic forms has several applications to topological prob-
lems. There are applications to knot theory, to differential topology (see [16]) and to
the study of hyperbolic manifolds [12,14,20,23].
This last application starts with the work of Picard, Klein, Poincaré, Fricke and
Bianchi, among many others. These authors consider a ternary or quaternary quadratic
form F with entries in some number field L, and try to compute the group Aut(F) of
automorphs (also called group of units) of the quadratic form (or group of matrices U
with entries in the ring of integers of L such that U t FU = F).
Bianchi [1] and [2] created an algorithm to obtain Aut(F) for some particular
ternary and quaternary integral quadratic forms. He finds the subgroup of reflections
R of Aut(F) by computing the “fundamental polyhedron” (that is what we call today,
after Thurston [23], the orbifold Q = H 3 /R, where H 3 is hyperbolic 3-space). When
R has finite index in Aut(F), he then computes the finite group of symmetries of Q.
In this way, he obtains a system of generators of the group of automorphs.
The more recent work of Thurston [23] renewed the interest in these investigations,
especially for its uses in low dimensional topology. For instance, very often we are
confronted with a simple knot K in S 3 and a positive integer m and we want to
construct explicitly a discrete group G of hyperbolic isometries whose finite order
elements are generated by m-rotations, such that S 3 is the quotient space of the action
of G in hyperbolic space H 3 and K is the image of the axes of m-rotation in G. Quite
frequently G can be defined by some arithmetic method.
In the course of his laborious calculations Bianchi observed [1, Section 15] an
example of two inequivalent integral quaternary quadratic forms such that their groups
of units were related by an isometry (let us refer for the moment to this situation as a
Bianchi pair). Intrigued by this example, I investigated its scope by creating a prac-
tical algorithm computing the orbifolds associated to integral, ternary and quaternary
quadratic forms. In this way I found many partial constructions of Bianchi pairs.
As a previous approach to the problem of the complete classification of Bianchi
pairs I have solved the following related classification problem. Say that two n-ary
(indefinite) integral quadratic forms F and G are commensurable if their groups of
automorphs have finite index subgroups that are conjugated in G L(n, R) [17,18].
The corresponding classification problem is solved in any dimension in [17] via a
subset of the Conway’s excesses [9]. For binary forms the result is quite easy. Two
binary indefinite integral forms are commensurable if and only if they have the same
determinants up to squares.
In this paper the Bianchi classification of binary, integral quadratic forms is com-
pletely solved.
We shall call two n-ary integral quadratic forms, F and G, B -equivalent (resp.
B B+
B + -equivalent), denoted F ∼ G (resp. F ∼ G), if there is a real isometry M from
F to ±G such that M −1 Aut(F)M = Aut(G) (resp. M −1 Aut+ (F)M = Aut+ (G)).
Here Aut+ (F) is the subgroup of Aut(F) formed by the so called proper automorphs,
On integral quadratic forms 201

those with determinant 1. A form is bilateral (resp. unilateral) if the index of Aut+ (F)
in Aut(F) is 2 (resp. 1). The problem is to group the Z-equivalent classes of integral
quadratic forms into Bianchi classes.
We associate to each B + -class an invariant v defined as follows. Let F be a binary,
integral quadratic form F, denoted (a, b, c) in Gauss notation, with discriminant D =
b2 − ac. Assume, without lost of generality in what follows, that F is primitive, that is,
gcd(a, b, c) = 1. As we will see, the only difficult case occurs when D is a positive,
non-square integer. We distinguish two cases. If the quadratic form is of type II (a
and c are even) and there is an integral solution of v 2 = 4 + Du 2 with v odd, then
v F is the minimal integer, among all integers v > 2, satisfying v 2 = 4 + Du 2 , for
some integer u. Otherwise, v F is twice the minimal integer among all integers v > 1,
satisfying v 2 = 1 + Du 2 , for some integer u.

Theorem 1 Two binary, integral quadratic forms with positive, non-square discrim-
inants are B + -equivalent if and only if their v-invariants coincide. And they are B-
equivalent if and only if their v-invariants coincide and both are bilateral or unilateral.

To check bilaterality we give the following result. (Forms (a, b, c) with a , b or c


zero are easily disposed off):

Theorem 2 Let F = (a, b, c) be an integral quadratic form with a = 0, b = 0, c = 0.


Then F is bilateral if and only if the integral quadratic form F  = (a 2 , b2 + D, c2 )
represent 4b2 .

A consequence of these results is that each Bianchi class with positive, non-square
discriminant contains only a finite number of Z-classes. A list of the (positive) discrim-
inants of these Z-classes for values 2 < v < 33 is given. The first B-class containing
more than one discriminant is the class v = 6 which contains forms with discriminants
2 and 8. Thus the forms (1, 0, −2) and (1, 0, −8) (both bilateral) are B-equivalent.
Compare this with the result about commensurable binary forms referred to above:
they have the same determinants up to squares.

2 Quadratic forms: generalities

A general reference is [7].


Let x be the column vector with coordinates x1 , . . . , xn and F a symmetric n × n
matrix. Then, the expression x t F x is called the n-ary quadratic form with matrix F.
We simply say that F is an n-ary quadratic form.
We call F an integral form if and only if the matrix entries of F are rational integers
(i.e. if and only if it is classical integral (Gauss) or integral as a symmetric bilinear
form) and the determinant of F is nonzero.
We say that an integral form F is primitive if the great common divisor of its entries
is 1. Since we are only interested in the automorphs of quadratic forms it is sufficient
to consider in this paper classical, primitive, integral quadratic forms.
An integral quadratic form F will be called of type II if its diagonal entries are all
even; F is of type I otherwise.
202 J. M. Montesinos Amilibia

Denote by G L(n, Z) the group of n×n matrices with integer entries and determinant
±1. We shall call two n-ary integral quadratic forms F and G integrally equivalent or
Z
Z -equivalent, or say they are in the same Z-class and write F ∼ G if there is a n × n
matrix M ∈ G L(n, Z) for which

M t F M = G. (1)

The equivalence is proper if det M = +1, improper if det M = −1.


These notions can be immediately generalized to arbitrary rings R (with 1). A form
F is defined over R if the matrix F has entries from R, and two forms F and G are
R-equivalent if (1) holds for some M with entries from R and with a determinant
which is a unit of R (i.e. an element of R with an inverse in R). A matrix with these
properties is called a R-isometry from F to G. Then F and G are Z-equivalent if and
only if there is a Z-isometry M from F to G.
An n × n matrix U with real entries is a real automorph of the integral quadratic
form F if and only if U t FU = F. Then det U = ±1. The real automorph U is proper
if det U = +1, improper if det U = −1.
The set of proper real automorphs of F is a subgroup S OR (F) of index 2 of the
group OR (F) of real automorphs of F. The group OR (F) is called the real orthogonal
group of the quadratic form F. Its subgroup S OR (F) is the special real orthogonal
group of the quadratic form F.
A real automorph U with integer entries is called an automorph, proper or improper,
of F. The set of automorphs of F is the automorphism group Aut(F) of f . The subset
Aut+ (F) of proper automorphs is the proper automorphism group of F. The group
Aut+ (F) has index 1 or 2 in Aut(F). If the index is 1 we say that F is unilateral.
Otherwise we say that F is bilateral (“zweiseitig”, Dedekind; “ancept”, Gauss; “bifid”,
Legendre; “ambiguous”, Dirichlet). An integral quadratic form f is bilateral if and only
if it is properly integrally equivalent to the opposite form, represented by any U t FU ,
where U is integral with −1 determinant. Because of this, it would be proper to call
a bilateral form “achiral” (and a unilateral one “chiral”), but, following Cahen [6], I
prefer the translation of Dedekind’s “zweiseitig”, because it brings forth immediately
the idea of “bilateral symmetry”. If M is a real isometry, from F to G, the map

U → M −1 U M

defines an isomorphism from OR (F) onto OR (G).

3 Bianchi equivalence

So far all definitions are standard. Now we introduce the notion of Bianchi equivalence.
This new concept will be the main topic of this paper.
We shall call two n-ary integral quadratic forms F and G Bianchi equivalent or
B
B-equivalent, denoted F ∼ G, if there is an R-isometry M from F to ±G sending
On integral quadratic forms 203

Aut( f ) isomorphically onto Aut(g). That is M is a real n × n matrix such that

M t F M = ±G (2)

and
M −1 Aut(F)M = Aut(G). (3)
Similarly, we say that two n-ary integral quadratic forms F and G are B + -equivalent,
B+
denoted F ∼ G if we only require that

M −1 Aut+ (F)M = Aut+ (G). (4)

Proposition 1 If two n-ary integral quadratic forms F and G are B-equivalent, then
both are bilateral or unilateral.
Proof Note that if U is an improper automorph of F then M −1 U M is an improper
automorph of G.

Z B
Clearly F ∼ G implies F ∼ G, but the converse is in general false.

4 The automorphs of binary forms

Denote the binary, integral quadratic form


 
a b
F=
b c

by (a, b, c) and let D denote the discriminant b2 − ac of F. We assume in the sequel


that F is primitive, unless the opposite is stated. We will refer to these quadratic forms
as integral forms or integral binary forms.
If F = (0, ±1, 0), the group Aut+ (F) is ±
1, 1 , where
x1 , x2 denotes the diag-
onal matrix having x1 , x2 in the principal diagonal, and zeros elsewhere. The form F
is bilateral, and Aut(F) extends Aut+ (F) by the improper automorph (0, 1, 0).
Therefore, assume one of the diagonal entries (say a) of F = (a, b, c) is not zero.
Let U be an automorph of F. Since the matrix U is integral, it acts as an automorphism
√ dimensional vector space K over any field K in between Q and C. Take
of any two 2

K = Q( D). The proof of the following lemma is trivial.


Lemma 1 Let T be a 2 × 2 matrix. Then,
   
t a b A B
T T =
b c B C

implies
   
−b −c −B −C
Ad j T T = ,
a b A B
204 J. M. Montesinos Amilibia

where Ad j T and T t denote the adjoint, (det T )T −1 , and the transpose of T , respec-
tively.

Since U t FU = F and det U = ±1, Lemma 1 implies U −1 T U = (det U )T , where


 
−b −c
T = .
a b
√ √
√ of T are D and − √D and the corresponding eigenvectors are
The eigenvalues
w+ = (−b + D, a) and w− = (−b − D, a). They belong to the vector space K2 .
Next, we claim that the automorph U leaves globally invariant the union of the two
vector-lines generated by w+ and w− , and that U is proper if it leaves invariant each
vector-line, and it is improper if it permutes them.
Indeed, if U is proper, then T U = U T . Hence

T U (w+ ) = U T (w+ ) = DU (w+ ).

Hence U (w+ ) = λw+ and U (w− ) = λ−1 w− , where λ and λ−1 denote the eigenvalues
of U (in K ).
If U is improper, then T U = −U T . Hence

T U (w+ ) = −U T (w+ ) = − DU (w+ ).

Hence U (w+ ) = μw− , and, similarly, U (w− ) = τ w+ .


Next, we claim that if U is improper, then U 2 is the identity. Indeed, U 2 (w+ )
= τ μ(w+ ) and U 2 (w− ) = τ μ(w− ). Hence det U 2 = (τ μ)2 = 1. Then, τ μ = ±1
and U 2 = ±I . If U 2 = −I , then U and
 
0 −1
1 0

are conjugate in G L(2, Q), since they share the same invariant factors. This would
imply det U = 1, which is not the case, since U is improper. Hence U 2 = I .
To obtain the general form of a proper automorph U , write:
 √ √   √ √ 
−b + D −b − D λ(−b + D) λ−1 (−b − D)
U = ,
a a λa λ−1 a

that is ⎡ √ √ ⎤
b+ D+(−b+
√ D)λ2 (−b2 +D)(λ

2 −1)

U =⎣ 2 Dλ √ Dλ √
2a ⎦. (5)
√ −1) −b+ D+(b+
a(λ2 D)λ2

2 Dλ 2 Dλ
Since this matrix is integral,

a(λ2 − 1)
√ = d,
2 Dλ
On integral quadratic forms 205

where d ∈ Z. Then,
√ √
d D ± a2 + d 2 D
λ= ,
a
and, substituting in (5), we have
 −bd −cd

a +e a
U= ,
a +e
bd
d

where

2
d
e =± 1+ D .
a

Since the entries of U are integral, the numbers

2bd cd
2e, ,
a a
are integers.
The form F is primitive. Therefore, there exist integers α, β, γ such that αa + βb +
γ c = 1. Hence

2d 2d 2bd 2cd
= (αa + βb + γ c) = α(2d) + β +γ
a a a a
is an integer.
2d
Denoting this integer a by u we have
 −bu+v −cu

U= au
2 2
bu+v ,
2 2

where v = ± 4 + Du 2 is an integer such that v ≡ bu mod 2. That is [6,7,13]:

Theorem 3 Let F = (a, b, c) be a integral form with a = 0. Then the group Aut+ (F)
is the set of matrices
   
1 1 0 −b −c
U= v +u ,
2 0 1 a b

where v and u are integers such that

v 2 − Du 2 = 4
v ≡ bu mod 2
cu ≡ au ≡ 0 mod 2.
206 J. M. Montesinos Amilibia

We want to obtain the general form of an improper automorph of an integral form


F = (a, b, c). If both a and c are zero the form F possesses the improper automorph
(0, 1, 0). Then assume that one of the diagonal entries (say c) of F is not zero. Then
there is an integer m such that m(2b + mc) is not zero. Then
     
1 m 0 b 1 0 m(2b + mc) b + cm
=
0 1 b c m 1 b + cm c

is Z-equivalent to F with nonzero diagonal entries. Hence we may assume that both
diagonal entries of F are nonzero.
2 is the
To obtain the general form of an improper automorph U , remember that U√
−1
(w+ ) = μw− (U (w− ) = μ w+ ), where w+ = (−b + D, a)
identity matrix, U √
and w− = (−b − D, a). Hence
 √ √   √ √ 
−b + D μ(−b − D) μ(−b − D) −b + D
U = .
a μa μa a

That is ⎡ √ √ √ √ 2 2 ⎤
(b− D)−(b+
√ D)μ2 (b− D)2 −(b+
√ D) μ
U =⎣ 2 Dμ
2)
√2a Dμ √ ⎦. (6)
a(−1+μ
√ − (b− D)−(b+
√ D)μ2
2 Dμ 2 Dμ
Since this matrix is integral,
√ √
(b − D) − (b + D)μ2
√ = d,
2 Dμ

where d ∈ Z. Then
√ √
−d D ± ac + d 2 D
μ= √ .
b+ D

And substituting in (6), we have


 e+db

d
U= a ,
e−db
c −d

where

e = ± ac + d 2 D.

Therefore, we have proved the following


Proposition 2 Let F = (a, b, c) be a integral form with a = 0 and c = 0. Then the
set of improper automorphs of F is the set of matrices of the form
 e+db

d
U= a ,
e−db
c −d
On integral quadratic forms 207

where e and d are integers such that

e2 − Dd 2 = ac
e + db ≡ 0 mod a
e − db ≡ 0 mod c.

A better description, that I did not find in the literature, is the following.
Theorem 4 Let F = (a, b, c) be a integral form with a = 0, b = 0, c = 0. Then the
set of improper automorphs of F is the set of matrices of the form
 ax−cy 
x
U= 2b ,
y − ax−cy
2b

where (x, y) runs over all pairs of integers such that a 2 x 2 +2(b2 + D)x y+c2 y 2 = 4b2 .

Corollary 1 Let F = (a, b, c) be a integral form with a = 0, b = 0, c = 0. Then F


is bilateral if and only if the integral form F  = (a 2 , b2 + D, c2 ) represent 4b2 .

Remark 1 Every integral form F = (a, b, c) with b = 0 is bilateral. Note also that the
discriminant of F  is 4b2 D. Note also that if (a, b, c) is bilateral so are (−a, ±b, −c).

Proof To prove the Theorem, let


 e+db

d
U= a
e−db
c −d

be a matrix where e and d are integers such that

e2 − Dd 2 = ac
e + db ≡ 0 mod a
e − db ≡ 0 mod c

ax+cy
a and y =
and set x = e+db c . These (x, y) are integers such that e =
e−db
2 and
ax−cy
d = 2b are integers. Moreover e2 − Dd 2 = ac implies

a 2 x 2 + 2(b2 + D)x y + c2 y 2 = 4b2 .

Conversely, assume
 ax−cy 
x
U= 2b ,
y − ax−cy
2b

where (x, y) is a pair of integers such that

a 2 x 2 + 2(b2 + D)x y + c2 y 2 = 4b2 .


208 J. M. Montesinos Amilibia

ax+cy ax−cy
Define e = 2 and d = 2b . The condition

a 2 x 2 + 2(b2 + D)x y + c2 y 2 = 4b2

can be written as
2
ax − cy
+ x y = 1.
2b

ax−cy
This implies that the square of the rational number d = 2b is an integer. Therefore
d itself is an integer. Since

a 2 x 2 + 2(b2 + D)x y + c2 y 2 = 4b2

implies e2 − Dd 2 = ac, it follows that the square of the rational number e is an integer.
Hence e itself is an integer. Therefore,
 e+db

d
U= a ,
e−db
c −d

where e and d are integers such that

e2 − Dd 2 = ac
e + db ≡ 0 mod a
e − db ≡ 0 mod c.

This completes the proof.




The solution of the Diophantine equation a 2 x 2 + 2(b2 + D)x y + c2 y 2 = 4b2 can


be found in [6], thus giving a practical procedure to recognize a bilateral form. The
problem of deciding this a priori (that is without computing the concrete solutions) is
as far as I know still open.

Example 1 The form (3, 5, −8) with discriminant D = 72 is unilateral. And so are
(3, 5, −4) of discriminant D = 37 and (a, 1, c) of discriminant 1 − ac < 0 for all
a > 2 and c = 0, 1, 2, a.

Example 2 The form (3, 5, 4) with discriminant D = 13 is bilateral: a solution of


9x 2 + 76x y + 16y 2 = 100 is (x, y) = (66, −8). Then
 
23 66
U=
−8 −23

is an improper automorph.
On integral quadratic forms 209

5 The B-equivalence of indefinite forms with non-square discriminant: the


v-invariant and the main theorem

Note that the discriminants of two B + -equivalent binary forms must have the same
sign, because they are R-isometric. We will study first the integral forms with pos-
itive discriminant (indefinite integral forms) and with non-square determinant. They
constitute the most important class.
We adopt the following definition (compare [7, p. 292]):

Definition 1 Let D be a positive integer which is not the square of an integer. Call
(vm , u m ) the fundamental solution of v 2 − Du 2 = 1 (resp. v 2 − Du 2 = 4) if vm is
the minimal integer among all integers v > 1 (resp. v > 2), satisfying v 2 = 1 + Du 2m
(resp. v 2 = 4 + Du 2m ), for some positive integer u m .

Let F = (a, b, c) be an integral form with positive non-square discriminant D F (or


D, for short). Then both, a and c, are nonzero. The group Aut+ (F) acts proper and
discontinuously on the hyperbola ax 2 + 2bx y + cy 2 = a and the quotient space of
the action is the 1-sphere S 1 . Therefore Aut+ (F) is C2 × C∞ where C2 is generated
by the automorph
−1, −1 (which permutes the two components of the hyperbola),
and C∞ is generated by the automorph
   
1 1 0 −b −c
UF = vF + uF ,
2 0 1 a b

where the pair of integers (v F , u F ) satisfy

v 2F − Du 2F = 4,
U F is integral,
v F > 2 and u F > 0,

and v F is minimal among the integral pairs (v, u) satisfying the corresponding three
above conditions (see [6,7,13]).
We introduce three non-standard definitions.

Definition 2 v F is the v-invariant of F.

Note that the value of v F depends not only on D but also on F.

Example 3 F = (a, b, c) = (1, 0, −32). D = 32. The fundamental solution of


v 2 − 32u 2 = 4 is (v, u) = (6, 1). And the minimal solution with v > 6 is (v, u) =
(34, 6). Since U F is integral, u F must be even because a is odd. Hence v F = 34.

Definition 3 We will say that F is ordinary if v F is even. If v F is odd, we say that F


is abundant.

The adjective “abundant” refers to Aut+ (F), which has more automorphs than
expected (see Theorem 9).
210 J. M. Montesinos Amilibia

Definition 4 Let D be a positive integer which is not the square of an integer. We will
say that D is of type II if there is no integral solution (v, u) of v 2 − Du 2 = 4 with v
odd. And D is of type I if there is an integral solution (v, u) with v odd.

Proposition 3 If D is of type I, then D ≡ 5 mod 8.

Proof Assume there is an integral solution (v, u) of v 2 − Du 2 = 4 with v odd. Since


v 2 ≡ 1 mod 2, D must be odd and v ≡ u ≡ 1 mod 2. This implies v 2 ≡ u 2 ≡ 1
mod 8. Hence 1 − D ≡ 4 mod 8. This completes the proof.


Remember (compare [16]) that an integral form F = (a, b, c) is of type II if its


diagonal entries a and c are both even. F is of type I otherwise. The type of a form is
a Z-invariant.

Proposition 4 If F is of type I, then, both v F and u F must be even, while if F is of


type II, then, both v F and u F must be even or odd.

Proof If F is of type I, then, both v F and u F must be even (compare Example 3)


because the matrix U F must be integral and a or c are odd. If F = (a, b, c) is of type
II, then D ≡ 1 mod 4, because b must be odd and then b2 − ac ≡ 1 mod 4. Therefore,
v 2F − Du 2F = 4 implies v F ≡ u F mod 4. Hence, both v F and u F must be even or odd.
This completes the proof.


Proposition 5 F = (a, b, c) is abundant if and only if (F, D F ) is of type (II,I).

Proof By definition, F is abundant if and only if v F is odd. Hence D F is of type I, by


definition, and F is of type II by Proposition 4.
Conversely, assume (F, D F ) is of type (II, I). Let (vm , u m ) be the fundamental
solution of v 2 − Du 2 = 4. By Proposition 3, D F ≡ 5 mod 8. This implies vm ≡ u m
mod 2 . If both (vm , u m ) were even, then all the integral solutions (v, u) of the equation
v 2 − Du 2 = 4 would be even. Indeed, suppose (vm , u m ) are both even. Now, as a and
c are both even and b is odd, we have by Theorem 3 that the proper automorphs of
(a, b, c) are in one-to-one correspondence with the integral solutions of v 2 − Du 2 = 4
with
 v−bu −cu
2 2
U=
au v+bu
2 2

corresponding to (v, u). By Theorem 3.1 on [7, p. 291] and the following remarks on
[7, p. 292], U = ±U0k , k ∈ Z, with U0 = U (vm , u m ). Now U0 reduces mod 2 to a
diagonal matrix. Therefore au/2 is even. As D F ≡ 5 mod 8, we have that ac ≡ 4
mod 8. Hence a/2 and c/2 are both odd. Therefore u is even. As (v − bu)/2 ∈ Z, we
have that v is even as well. Thus all integral solutions of v 2 − Du 2 = 4 are even. But
this is not the case because D F is of type I, by hypothesis. Hence, both vm , u m must
be odd. Moreover the matrix
   
1 1 0 −b −c
vm + um
2 0 1 a b
On integral quadratic forms 211

is integral because, being F of type II, cu m ≡ au m ≡ 0 mod 2 and vm ≡ bu m mod 2,


because b is odd. Hence v F = vm is odd. Hence F is abundant. This completes the
proof.


The proof of the next Proposition is now clear (see Proposition 4 and Example 3
and the proof of Proposition 5).

Proposition 6 Let F be an integral form with positive and non-square discriminant


D. If F is ordinary, (v F , u F ) is twice the fundamental solution of v 2 − Du 2 = 1.
While if F is abundant, (v F , u F ) is the fundamental solution of v 2 − Du 2 = 4. In this
case, (v F , u F ) are both odd.

Here is the main theorem:


Theorem 5 Let F and G be two integral binary forms with positive non-square dis-
criminants. Then:
B B+
1. F ∼ G if and only if F ∼ G and both F and G are bilateral or unilateral.
B+
2. F ∼ G if and only if v F = vG .
We will prove this theorem in the next sections.
Remark 2 In Sect. 13 we will give a practical algorithm to compute v F , based in
Proposition 6.

6 Normalizing indefinite forms with non-square discriminant up to


B + -equivalence

Proposition 7 Let F = (ga, b, gc) be an integral form with positive and non-square
discriminant and such that

gcd(a, c) = 1.

B+
Let p be a positive divisor of c. Then, F ∼ G = (gap, b, gc/ p) if F and G are of
the same type.

Proof Let M =
p, √1p . Then M t F M = G. The group Aut+ (F) is generated by

−1, −1 and U F :
   
1 1 0 −b −gc
UF = vF + uF .
2 0 1 ga b

By Lemma 1 we have
   
1 1 0 −b −gc/ p
M −1 U F M = vF + uF .
2 0 1 gap b
212 J. M. Montesinos Amilibia

Now, D F = DG , and, by hypothesis, F and G are of the same type. Then (Propo-
sition 6), vG = v F and, therefore, u G = u F . Hence, M −1
−1, −1 M =
−1, −1 ,
B+
together with M −1 U F M = UG , generate Aut+ (G). Hence F ∼ G. This completes
the proof.

Remark 3 The form F = (1, 1, −4) gives rise to G = (2, 1, −2) by trading the divisor
2 across the diagonal. But these forms are of different type. Hence Proposition 7
does not apply. In fact M −1 U F M = UG3 and therefore Aut(F) is sent by M into
a subgroup of Aut(G) of index three (see Section 10). Here D F = DG = 5 and
(v F , u F ) = (18, 8) while (vG , u G ) = (3, 1) and
     
5 16 −1 5 8 1 1
UF = ; M UF M = = UG ; UG =
3
.
4 13 8 13 1 2

Proposition 8 Let F be an integral form of type I with positive and non-square dis-
criminant. Then, F is Z-equivalent to an integral form (of type I) whose diagonal
elements are relatively prime.
Proof Set F = (a, b, c) where c is odd. Write b = b g and c = c g with c > 0 and
gcd(c , 2b ) = 1. By Dirichlet’s Theorem, there exists a prime p = c + 2b n, coprime
with a. Then
     
1 0 1 n a b + an
F = .
n 1 0 1 b + an gp + an 2

Let k = gcd(a, gp + an 2 ). Then k divides a and gp and, since gcd(a, p) = 1, k


divides g. Hence k divides a, b and c. Since F is primitive, k = 1. This concludes the
proof.

Proposition 9 Let F be an integral binary form of type II with positive and non-square
discriminant. Then, F is Z -equivalent to an integral form (of type II) whose diagonal
elements divided by 2 are relatively prime.
Proof Set F = (2a, b, 2c). Let g = gcd(b, c). Set b = b g, c = c g. Then b and
c are relatively prime. By Dirichlet’s Theorem, we can find an integer n such that
c + b n is a prime number p relatively prime to a. Then,
     
1 0 1 n 2a b + 2an
F = .
n 1 0 1 b + 2an 2(gp + an 2 )

Let k = gcd(a, gp +an 2 ). Then k divides a and gp. Since gcd(a, p) = 1, k divides
g = gcd(b, c). Hence k divides a, b and c. Since F is primitive, k = 1. This concludes
the proof.

Theorem 6 Let F be an integral form with positive and non-square discriminant. If
F is of type I, then

B+
F ∼ (1, 0, −D F ).
On integral quadratic forms 213

If F is of type II, then



B+ −D F + 1
F ∼ 2, 1, .
2

Proof Let F be an integral binary form of type I. By Proposition 8, F is Z-equivalent


to an integral form F  whose diagonal elements are relatively prime. Taking −F  , if
necessary, we obtain an integral form F  , B + -equivalent to F  , and with diagonal
elements relatively prime, and one of them (say a) positive. By Proposition 7, F  is
B + -equivalent to a form F  = (1, b, c), where c = −D F + b2 . But
     
1 0 1 0 1 b 1 b
= = F  .
b 1 0 −D F 0 1 b −D F + b2

This concludes the proof of the first part.


Now, let F be an integral form of type II. By Proposition 9, F is Z-equivalent to
an integral form F  = (2a, b, 2c) such that gcd(a, c) = 1. By Proposition 7, F  is
B + -equivalent to an integral form G = (2, b, 2e), where

D F = b2 − 4e.

Finally, this integral form G is Z-equivalent to (2, 1, −D2F +1 ). Indeed, b = 2b + 1,


and if
 
1 −b
M= ,
0 1

we have

M t G M = (2, 1, −2b2 − 2b + 2e),


−D F +1
where −2b2 − 2b + 2e = 2 . This concludes the proof.


7 Normalizing bilateral indefinite forms with non-square discriminant up to


B-equivalence

We will make use of the next proposition, due to Dirichlet (see, for instance, [6,
Chapitre XIII]).
Proposition 10 Let F be an integral, binary quadratic form. Assume F is bilateral.
Then, F is Z-equivalent to a diagonal integral form (a, 0, c) or to an integral form
(2a, a, c).
Proof Let F = (a, b, c) be bilateral. Let
 
α β
U= , α 2 + βγ = 1,
γ −α
214 J. M. Montesinos Amilibia

be an improper automorph of F. We may assume α ≥ 0. If γ = 0, then α = 1. Direct


calculation gives 2b = aβ. If γ = 0, there exist
 
λ μ
T = ,
ν ρ

such that det T = 1 and

λ α+1
= .
ν γ

For such T , the form G = T t F T has the automorph T −1 U T whose (2, 1) -entry is
zero. Therefore, we can assume that our original F = (a, b, c) satisfies 2b = aβ. Take
 ε−β

1
M= 2 ,
0 1

where ε is zero if β is even, and 1 if β is odd. Then M t F M is diagonal, if β is even,


and (a, a/2, c ) if β is odd (a even). This completes the proof.


Next, we normalize (a, 0, c) and (2a, a, c ), c odd, up to B-equivalence.

Proposition 11 The integral form F = (a, 0, c) with positive and non-square dis-
criminant, is B-equivalent to (1, 0, −D F ). Assume c odd. Then, the integral form
F = (2a, a, c) with positive and non-square discriminant, is B-equivalent to
(1, 0, −D F ).

Proof Let F = (a, 0, c). Up to B-equivalence, we may assume a positive. Moreover,


gcd(a, c) = 1, because F is primitive. By Proposition 7,

M t F M = (1, 0, −D F ) = G,

where M =
√1a , a exhibits the B + -equivalence between the integral forms F and
G. This is also a B-equivalence because M −1 V M = V , where V =
1, −1 is an
improper automorph of both F and G.
Let F = (2a, a, c), c odd. Up to B-equivalence, we may assume a positive. More-
over, gcd(2a, c) = 1, because F is primitive. By Proposition 7,

M t F M = (1, a, e) = G,

where e = 2ac and M =
√1 , 2a , is a B + -equivalence between the integral forms
2a
F and G. This is also a B-equivalence, because M −1 V M is integral, where
 
1 1
V =
0 −1
On integral quadratic forms 215

is an improper automorph of F. Finally, the integral form G = (1, a, e), such that
a 2 − e = D F , is Z-equivalent to (1, 0, −D F ). Indeed, if
 
1 −a
M= ,
0 1

we have

M t G M = (1, 0, −a 2 + e).

This completes the proof.




Similarly, we normalize F = (2a, a, 2c) up to B-equivalence:

Proposition 12 The integral form F = (2a, a, 2c) with positive and non-square
discriminant is B-equivalent to (2, 1, −D2F +1 ).

Proof Up to B-equivalence, we may assume a positive. Moreover, a is odd and


gcd(a, c) = 1, because F is primitive. By Proposition 7,

M t F M = (2, a, 2e) = G,

where e = ac and M =
√1a , a , is a B + -equivalence between the integral forms F
and G. This is also a B-equivalence, because M −1 V M is integral, where
 
1 1
V =
0 −1

is an improper automorph of F. Finally, the integral form G = (2, a, 2e), such that
a 2 − 4e = D F , is Z-equivalent to (2, 1, −D2F +1 ). Indeed, a = 2a  + 1, and if
 
1 −a 
M= ,
0 1

we have

M t G M = (2, 1, −2a 2 − 2a  + 2e),


−D F +1
where −2a 2 − 2a  + 2e = 2 . This completes the proof.


We group these results together:

Theorem 7 Let F be a bilateral integral form with positive and non-square discrim-
B
inant. If F is of type I, then F ∼ (1, 0, −D F ). If F is of type II, then

B −D F + 1
F ∼ 2, 1,
2
216 J. M. Montesinos Amilibia

8 The B-equivalence of indefinite forms with non-square discriminant:


necessary condition

The necessary part of Theorem 5 is a consequence of Lemma 1:


Proposition 13 Let F and G be two integral binary forms with positive non-square
B+
discriminants. If F ∼ G then v F = vG .
Proof Since F = (a, b, c) and G = (A, B, C) have non-square discriminants D F
B+
and DG , then a and A are nonzero. If F ∼ G there exists a real 2 × 2 matrix M such
that M t F M = ±G and such that M −1 U F M = ±UG , where
   
1 1 0 −b −c
UF = vF + uF
2 0 1 a b

and
   
1 1 0 −B −C
UG = vG + uG .
2 0 1 A B

By Lemma 1, we have
   
1 1 0 uF −B −C
M −1 U F M = vF + = ±UG .
2 0 1 det M A B

Now the matrices


   
1 0 −B −C
and
0 1 A B

are linearly independent, since A = 0. Hence the last two matrix equations imply that
v F = ±vG . As v F and vG are both positive, they coincide. This completes the proof.


From this result an Proposition 1 we obtain:
Corollary 2 Let F and G be two integral binary forms with positive non-square
B
discriminants. If F ∼ G, then v F = vG and both, F and G, are bilateral or unilateral.

9 The B-equivalence of indefinite forms with non-square discriminant:


sufficient condition

Theorem 8 Let F be an integral form with positive non-square discriminant D F . If


B+
F is ordinary, then F ∼ (1, 0, −D F ). If F is abundant, then

B+ −D F + 1
F ∼ 2, 1, .
2
On integral quadratic forms 217

Moreover, if F is bilateral, the equivalences are B-equivalences.

Proof By Theorems 6 and 7, we only need to consider the case of an ordinary integral
form F = (2, 1, −D2F +1 ) of type II. We must prove that F is B-equivalent to G =
(1, 0, −D F ). By Proposition 5, a type II form, like F, is ordinary if and only if its
discriminant is of type II. Therefore D F is of type II. That is, the fundamental solution
(vm , u m ) of v 2 − D F u 2 = 4 must satisfy vm ≡ u m ≡ 0 mod 2, since there are no odd
solutions.
Now, take
 
1 2 1
M=√ ,
2 0 1

a real matrix with det M = 1. Then M t G M = F. The group Aut+ (G) is generated
by
−1, −1 and
   
1 1 0 0 DF
UG = vG + uG ,
2 0 1 1 0

where (vG , u G ) = (vm , u m ), because this pair satisfy the two requirements of being
minimal and of making UG integral. By Lemma 1 we have
   D F −1

−1 1 1 0 −1
M UG M = vm + um 2 .
2 0 1 2 1

And this equals U F , because (vm , u m ) = (v F , u F ), since (vm , u m ) satisfy the two
requirements of being minimal and of making M −1 UG M integral. Therefore

M −1 Aut+ (G)M = Aut+ (F).

B+
Hence G ∼ F.
On the other hand, M −1 V M = W , where
 
1 0
V =
0 −1

is an improper automorph of G and


 
1 1
W =
0 −1

B
is an improper automorph of F. Therefore G ∼ F. This completes the proof.


Proposition 14 Let F = (1, 0, −D F ) and G = (1, 0, −DG ) be two integral forms


B
with positive non-square discriminants, and such that v F = vG . Then, F ∼ G.
218 J. M. Montesinos Amilibia

Proof Since F and G are both of type I (Proposition 4), then v F = vG is even.
The relations

v 2F − D F u 2F = 4,

vG
2 − D u 2 = 4,
G G

together with v F = vG , imply

D F u 2F = DG u 2G .

Define
 
1 0
M= uF .
0 uG

Then,
 
u 2F
M t F M = 1, 0, −D F = (1, 0, −DG ) = G.
u 2G

The group Aut+ (F) is generated by


−1, −1 and
   
1 1 0 0 DF
UF = vF + uF .
2 0 1 1 0

By Lemma 1 and

uF
= det M,
uG

we have
   
1 1 0 uF 0 DG
M −1 U F M = vF + .
2 0 1 det M 1 0

That is
   
1 1 0 0 DG
M −1 U F M = vG + uG = UG .
2 0 1 1 0

Therefore,

M −1 Aut+ (F)M = Aut+ (G).


On integral quadratic forms 219

B+
Hence F ∼ G. This is also a B-equivalence, because M −1 V M = V , where
 
1 0
V =
0 −1

is an improper automorph of both F and G. This concludes the proof.

Proposition 15 Let F = (2, 1, −D2F +1 ) and G = (2, 1, −D2G +1 ) be two abundant


integral forms with positive non-square discriminants and such that v F = vG . Then
B
F ∼ G.

Proof Since F and G are both abundant, then v F = vG is odd.


The relations

v 2F − D F u 2F = 4,

vG
2 − D u 2 = 4,
G G

together with v F = vG , imply

D F u 2F = DG u 2G .

Define

uF
2 (− u G + 1)
1
1
M= uF .
0 uG

Then,

−DG + 1
M t F M = 2, 1, = G.
2

The group Aut+ (F) is generated by


−1, −1 and
   
1 1 0 −1 D F2−1
UF = vF + uF .
2 0 1 2 1

By Lemma 1 and

uF
= det M
uG

we have
   
1 1 0 uF −1 DG2−1
M −1 U F M = vF + .
2 0 1 det M 2 1
220 J. M. Montesinos Amilibia

That is
   
−1 1 1 0 −1 DG2−1
M UF M = vG + uG = UG .
2 0 1 2 1

Therefore,

M −1 Aut+ (F)M = Aut+ (G).

B+
Hence F ∼ G. This is also a B-equivalence because M −1 V M = V , where
 
1 1
V =
0 −1

is an improper automorph of both F and G. This concludes the proof.



Proposition 16 Let F and G be two integral binary forms with positive non-square
B+
discriminants. If v F = vG , then F ∼ G. If, moreover, F and G are bilateral, then
B
F ∼ G.
Proof By Theorem 8, if F and G are ordinary, then

B+
F ∼ F  = (1, 0, −D F )

and
B+
G ∼ G  = (1, 0, −DG ).

If F and G are abundant, then



B+ −D F + 1
F ∼ F  = 2, 1, ,
2

B+ −DG + 1
G ∼ G  = 2, 1, .
2

Moreover, if F and G are bilateral, these equivalences are B -equivalences. By Propo-


B B+
sition 13, v F  = vG  . By Propositions 14 and 15, F  ∼ G  . Therefore, F ∼ G, and,
B
if F and G are bilateral, F ∼ G. This concludes the proof.

This last proposition concludes the proof of the main Theorem 5.
Example 4 The B + -class corresponding to v = 15 splits in two B-classes, one bilat-
eral, represented by (−2, 1, 110), another unilateral, represented by (−14, 5, 14). Sim-
ilarly, the B + -class corresponding to v = 20 splits in two B-classes, one bilateral,
represented by (−9, 0, 11), another unilateral, represented by (−9, 3, 10). For odd
v < 15 and for even v < 20, the B-classes coincide with the B + -classes.
On integral quadratic forms 221

Now, we can solve problems such as the following

Problem 1 Given F = (a, b, c), find all the Z-classes B + - (or B-) equivalent to F.

As an example take the integral form F = (−34, 9, 74). Its discriminant is D F =


2597. The fundamental solution of v 2 − D F u 2 = 4 is (51, 1). Hence F is abundant
and v F = 51. We find the discriminants D such that the fundamental solution of
v 2 − Du 2 = 4 is (51, u). These satisfy

Du 2 = v 2F − 4 = 2597 = 53 × 72 .

The possibilities are D = 2597 and 53. We check that the fundamental solution of
v 2 −53u 2 = 4 is (51, 7). The different Z-classes of type II with discriminant D = 2597
are represented by

(2, 1, −1298), (−2, 1, 1298)

both bilateral and

(−34, 9, 74), (34, 9, −74)

both unilateral. There is just one Z-class of type II with discriminant D = 53 and it is
represented by (−2, 1, 26) and is bilateral. Hence the Z-classes B + -equivalent to F
are

(2, 1, −1298), (−2, 1, 1298), (−34, 9, 74), (34, 9, −74), (−2, 1, 26).

Since F is unilateral, there are just two Z-classes B-equivalent to F, namely

(−34, 9, 74), (34, 9, −74).

Problem 2 Given v find the Z-classes in the bilateral (resp. unilateral) B-class with
invariant v.

As an example take v = 48. The discriminants D in the B-class satisfy


 v 2
Du 2 = − 1 = 23 × 52 .
2

Hence D = 23 or 575. Here

D ≡ −1 mod 8

Hence D is of type II. Hence the forms in the B-class are ordinary of type (F, D) =
(I, II). There are two Z-classes with discriminant D = 23, both bilateral, namely
222 J. M. Montesinos Amilibia

±(1, 0, −23). And there are twelve unilateral Z-classes with discriminant 575, namely

(−22, 5, 25) (22, 5, −25) (−19, 9, 26) (19, 9, −26)


(−14, 1, 41) (14, 1, −41) (−13, 4, 43) (13, 4, −43) ,
(−11, 5, 50) (11, 5, −50) (−7, 1, 82) (7, 1, −82)

and six bilateral, namely

(−23, 0, 25) (23, 0, −25)


(−2, 1, 287) (2, 1, −287)
(−1, 0, 575) (1, 0, −575)

10 Ordinary and abundant indefinite forms with the same non-square


discriminant

Let F and G be an abundant integral form and an ordinary integral form, respectively,
with the same positive and non-square discriminants, D F = DG = D. By Theorem 8,
up to B + -equivalence, we may assume that F = (2, 1, −D+12 ), where D is of type
I, and G = (1, 0, −D). Since D is of type I, the fundamental solution (vm , u m ) of
v 2 − Du 2 = 4 must satisfy vm ≡ u m ≡ 1 mod 2 (Proposition 6).
Now, take
 
1 2 1
M=√ ,
2 0 1

a real matrix with det M = 1. Then M t G M = F.


The group Aut+ (G) is generated by
−1, −1 and
   
1 1 0 0 D
UG = vG + uG ,
2 0 1 1 0

where (vG , u G ) is twice the fundamental solution of v 2 − Du 2 = 1 (Proposition 6).


By Lemma 1 we have
   
−1 1 1 0 −1 D−1
M UG M = vG + uG 2 .
2 0 1 2 1

This is an integral matrix belonging to Aut+ (F). Hence M −1 Aut+ (G)M is a subgroup
of Aut+ (F). We want to obtain its index i.
To that end, consider
 
1 v F − u F u F D−1
UF = 2 ,
2 2u F vF + u F
On integral quadratic forms 223

which generates, together with


−1, −1 , the group Aut+ (F). Here (v F , u F ) =
(vm , u m ). Then,
 
−1 1 vm u m D
V = MU F M =
2 u m vm

generates an infinite cyclic supergroup H F of the infinite cyclic group HG ≤ Aut+ (G)
generated by UG . The index of HG in H F is i.
The third power of the generator V of H F is the integral matrix
 
1 vm (1 + Du 2m ) Du m (3 + Du 2m )
V3 = ,
2 u m (3 + Du 2m ) vm (1 + Du 2m )

while no less power gives a integral matrix (compare [19, p.119] ). Then UG = V 3 .
Hence

M −1 Aut+ (G)M

is a subgroup of Aut+ (F) of index i = 3. Note also that the equation UG = V 3


implies that

(vG , u G ) = (v F (−3 + v 2F ), u F (−1 + v 2F )).

We have proved the following theorem which explains the adjective “abundant”:

Theorem 9 Let G and F be an ordinary integral form and an abundant integral form,
respectively, with the same positive and non-square discriminants, D F = DG = D.
Then, there is a real matrix M such that M t G M = ±F, and M −1 Aut+ (G)M is a
subgroup of Aut+ (F) of index i = 3. If F and G are both bilateral, then M −1 Aut(G)M
is a subgroup of Aut(F) of index i = 3.

Remark 4 Aut+ (F) acts in level −D of the quadratic form v 2 − Du 2 . The quotient
of the action (which is free) is the 1-sphere S1 . The subgroup M −1 Aut+ (G)M ≤
Aut+ (F) of index 3 induces an ordinary 3-fold cyclic covering S1 → S1 . If F and
G are both bilateral then M −1 Aut(G)M is a subgroup of Aut(F) of index i = 3.
The corresponding orbifold covering is a 3-fold irregular dihedral covering. Both
orbifolds H 1 /Aut(G) and H 1 /Aut(F) are homeomorphic to a closed interval I of R
with silvered boundary (∂ I have isotropy dihedral of order 2) and the orbifold covering
is the result of folding H 1 /Aut(G) = I = [0, 3] along 1 and 2 and mapping it onto
H 1 /Aut(G) = [0, 1].

11 The B-equivalence of definite forms

Let F = (a, b, c) be definite, that is, F is a binary integral quadratic form with negative
discriminant D = b2 − ac. Replacing F by −F , if necessary, we may assume that
224 J. M. Montesinos Amilibia

both a and c are positive (F is positive definite). Note that F and −F are B + - and B
-equivalent.
Parametrize the set of definite positive form as follows. The equation aω2 + 2bω +
c = 0 has roots

−b ± D
.
a

Say that

−b + D
ωF =
a

is the root of F. It represents a point in the upper-half model of the hyperbolic plane
H. An element
 
α β
T =
γ δ

of G L(2, Z) acts on F and H as follows:

F → G = T t FT

αω + β
ω F → T (ω F ) := , if det T = 1
γω + δ

αω + β
ω F → T (ω F ) := , if det T = −1.
γω + δ

An easy calculation (see [5, p. 246]) shows that ωG = T −1 (ω F ).


The action of G L(2, Z) (that factors through P G L(2, Z) = G L(2, Z)/{±I }) on
H has the fundamental domain
 
1 2
F = x + iy ∈ C : 0 ≤ x ≤ , x + y ≥ 1 .
2
2

Therefore F is Z-equivalent to a unique G such that ωG ∈ F (G is the “forme réduite”


in [5, p. 222]). Note that F is a hyperbolic 2-orbifold. Given a point ωG ∈ F, the
preimage in G L(2, Z) of the isotropy group I (ωG ) ≤ P G L(2, Z) is the group AutG.
Therefore, since all the points ωG lying in the interior of F have trivial I (ωG ), it
follows that if G = (a, b, c), where 0 < −2b < a and c > a, then AutG = {±I }.
That is, G is unilateral and obviously all these forms are B + and B-equivalent, since
any R-isometry between two of them preserves the group {±I }.
On integral quadratic forms 225


The points ωG lying in the boundary of F, ωG = i, ωG = ξ = 21 + i 23 , all have
I (ωG ) dihedral of order two. More precisely, if G = (a, b, a), 0 < −2b < a, then
AutG is an extension of Aut+ G = ±I by the improper automorph U = (0, 1, 0). If
G = (a, 0, c), c > a, then AutG is an extension of Aut+ G = ±I by the improper
automorph V = (1, 0, −1). And if G = (2a, −a, 2c), c > a, then AutG is an
extension of Aut+ G = ±I by the improper automorph
 
1 −1
W =
0 −1

Finally all these forms G are bilateral (compare with Proposition 10 and [6, p. 294])
and obviously all are mutually B + -equivalent, since any R-isometry between two
of them preserves the group {±I }. They are also B-equivalent. Indeed, the integral
form (a, 0, c), c > a, with the improper automorph V = (1, 0, −1), is B -equivalent
to (1, 0, 2). For G = (2a, −a, 2c), c > a, we have M t G M = (a, 0, 4c − a) and
M −1 W M = V , where
 
1 1 1
M=√ .
2 0 2

For G = (a, b, a), 0 < −2b < a we have M t G M = (a −b, 0, a +b) and M −1 U M =
−V , where
 
1 1 1
M=√ .
2 −1 1

The forms represented by points of F, excepted i and ξ , are mutually B + -equivalent,


since any R-isometry between two of them preserves the group {±I }.
It remains to study the forms G represented by the corner points i and ξ .
The form G corresponding to ωG = i is (1, 0, 1). The isotropy subgroup I (i) of
P G L(2, Z) is dihedral of order 4. Its preimage in G L(2, Z) is the group AutG which
is an extension of the cyclic group Aut+ G of order 4, generated by
 
0 −1
,
1 0

by the improper automorph (1, 0, −1). The form G is therefore bilateral.


The form G corresponding to ωG = ξ is (2, −1, 2). The isotropy subgroup I (ξ ) of
P G L(2, Z) is dihedral of order 6. Its preimage in G L(2, Z) is the group AutG which
is an extension by the improper automorph
 
−1 0
,
−1 1
226 J. M. Montesinos Amilibia

of the cyclic group Aut+ G of order 6, generated by


 
0 −1
.
1 1

The form G is therefore bilateral.


We group together these facts in the following
Theorem 10 There are exactly three B + -equivalence classes of integral forms of
negative discriminant D, namely the Z-class of (1, 0, 1), with is bilateral with dis-
criminant D = −1, which is also a B-class; the Z-class of (2, −1, 2) with is bilateral
with discriminant D = −3, which is also a B-class; and the remaining Z-classes that,
together, form a unique B + -class. This class splits in two B-classes: the bilateral and
the unilateral integral forms in the class.
Given a definite form F = (a, b, c) there is an algorithm to obtain its Z-equivalent
representative G with the root ωG lying in the orbifold F (see, for instance, [5, pp.
221-226].
The content of this section is well known (see [3] or [6], for instance), and it might
be interesting to put it in the context of 2-dimensional Crystallography as exposed in
[4].
For every finite subgroup H of G L(2, Z) there is at least one definite, binary
integral quadratic form F such that H ≤ AutF. The subgroups AutF of G L(2, Z)
are distributed in Z-equivalence classes, called Z-Bravais classes in [4]. And their Q-
equivalence classes are called systems in [4]. There are four 2-dimensional systems,
which correspond to our four B-equivalence classes, and there are five Z-Bravais
classes, because the groups of automorphs of order 4 of bilateral forms, split into two
Z-classes.
The B-class represented by the roots in the interior of F is the Q-class 1/2 (holoedry
of the only system in the oblique family) containing just one Z-class, denoted 1/2/1
in [4, p. 56].
The B-class represented by the roots in the boundary of F, excepted i and ξ , is the
Q-class 2/2 (holoedry of the only system in the rectangular family) containing two Z
-classes, denoted 2/2/1 and 2/2/2 in [4, p. 57]. The Z-class 2/2/1 corresponds to the
two vertical edges of F, while the Z-class 2/2/2 corresponds to the edge of F lying
on the unit circle.
The B-class represented by the root i is the Q-class 3/2 ( holoedry of the only
system in the square family) containing just one Z-class, denoted 3/2/1 in [4, p. 58].
Finally, the B-class represented by the root ξ is the Q -class 4/4 (holoedry of the
only system in the hexagonal family) containing just one Z-class, denoted 4/4/1 in [4,
p. 59]. Compare also with the matrices given in [4, p. 261].

12 The B-equivalence of indefinite forms with square discriminant

If the discriminant D of F is the square of an integer E, the only solutions of

(v − Eu)(v + Eu) = 4
On integral quadratic forms 227

are (v, u) = (±2, 0). Therefore (Theorem 3), Aut+ (F) is C2 , generated by
−1, −1
.
The form might be bilateral or unilateral. For instance, (3, 5, −8), with discriminant
D = 72 , is unilateral and (1, 0, −4) is bilateral.

Theorem 11 Any two integral forms with positive and square discriminant are B +
-equivalent. They are B-equivalent if both are bilateral or unilateral.

Proof Any two of these forms are B + -equivalent since any R -isometry between them
preserves the group {±I }. They are also B -equivalent if both are bilateral, because
by Proposition 10 we can assume they are (a, 0, c) or (2a, a, c). We then argue as in
the proof of Theorem 10. This concludes the proof.


13 Computation of the v-invariant

Let F be an integral form with positive and non-square discriminant D. To compute v F


we distinguish two cases. If F is ordinary, (v F , u F ) is twice the fundamental solution
of v 2 − Du 2 = 1. While if F is abundant, (v F , u F ) is the fundamental solution of
v 2 − Du 2 = 4. In this case, v F and u F are both odd (Proposition 6). These
√ fundamental
solutions are computed using the continuous fraction expansion of D [6,13]. Next,
we recall the method. √
Consider the continued fraction expansion of x0 = D, extended to the n-th term
√  
D = a0 , a1 , a2 , . . . , an−1 , xn .

Here, a0 , a1 , a2 , . . . are integers ≥ 1, and x0 , x1 , x2 , . . . is called the sequence of


complete quotients . Here, the sequence x0 , x1 , x2 , . . . is obtained as follows. Take a0
to be the floor of x0 , that is the biggest integer

a0 ≤ x 0 = D.

Then,

1 b1 + D
x1 = = .
x 0 − a0 c1

Then, an−1 is the floor of xn−1 and



1 bn + D
xn := = ,
xn−1 − an−1 cn

and so on.
Define also b0 = 0, c0 = 1.
Let

Pn
, n = 1, 2, 3, . . .
Qn
228 J. M. Montesinos Amilibia

denote the n-th convergent, derived by deleting xn from [a0 , a1 , a2 , . . . , an−1 , xn ] and
define

P−1 0 P0 1
= , = .
Q −1 1 Q0 0

In matrix terms:
    
Pn Pn−1 Pn−1 Pn−2 an−1 1
= , n = 1, 2, . . . (7)
Q n Q n−1 Q n−1 Q n−2 1 0
      
Pn Pn−1 a 1 a1 1 a 1
= 0 . . . n−1 , n = 1, 2, . . . (8)
Q n Q n−1 1 0 1 0 1 0

so that  
Pn Pn−1
det = (−1)n . (9)
Q n Q n−1
Then,
xn Pn + Pn−1
x0 = . (10)
xn Q n + Q n−1
That is,

bn + D
√ cn Pn + Pn−1
D= √ .
bn + D
cn Qn + Q n−1

Equating rationals and irrationals we get

bn Pn + cn Pn−1 = D Q n
bn Q n + cn Q n−1 = Pn

that we can write as follows


    
Pn Pn−1 bn D Qn
= .
Q n Q n−1 cn Pn

This implies
    
bn Q n−1 −Pn−1 D Qn
= (−1)n .
cn −Q n Pn Pn

That is
bn = (−1)n−1 (Pn Pn−1 − D Q n Q n−1 )
, (11)
cn = (−1)n (Pn2 − D Q 2n )
and we see (−1)n cn is the value of the integral form v 2 − Du 2 when uv is the convergent
Pn
Q n . We will call c0 , c1 , c2 , . . . (resp. b0 , b1 , b2 , . . .) the c-sequence (resp. b-sequence).
On integral quadratic forms 229

The significance of the c-sequence is the following (see [21, p.53], [22, p.199],
[24]).
Proposition 17 Let u, v be relatively prime integers such that v 2 − Du 2 = c and
c2 < D. Then, uv is one of the convergents QPnn of the continued fraction expansion of

D.
A consequence of this proposition is that to obtain the fundamental solution of
v 2 − Du 2 = 1, when D > 1, we only need to obtain the appearances of the digit 1 in
the c-sequence.
A second consequence is that v 2 − Du 2 = 4, for D > 16, has a solution √ (v, u)
with v and u odd if and only if the digit 4 is a member of the c-sequence of D. (If v
and u are odd, they are relatively prime). In other words,
√ D > 16 is of type I if and
only if the digit 4 is a member of the c-sequence of D .
There are two values of D ≤ 16 for which v 2 − Du 2 = 4 might have an odd solution,
namely D = 5, 13, because exactly √ in these cases D ≡ 5 mod 8 (Proposition 3).
For D = 5 the c-sequence of 5 fails to contain the digit 4, though 5 is of type I.
Indeed the fundamental solution of v 2 − 5u 2 = 4 is (vm , √u m ) = (3, 1).
We compute (see the example below) the c-sequence of 13 and find that it contains
the digit 4 even if D = 13 < 16. Therefore D = 13 is also of type I.
√  
Example 5 D = 13. Here p = 5; 13 = 3, 1, 1, 1, 1, 6 .

0 1 2 3 4 5 6 7 8 9 10
a 3 1 1 1 1 6 1 1 1 1 6
P 1 3 4 7 11 18 119 137 256 393 649
Q 0 1 1 2 3 5 33 38 71 109 180
b 0 3 1 2 1 3 3 1 2 1 3
c 1 4 3 3 4 1 4 3 3 4 1

Here

c5 = 182 − 13 × 52 = (−1)5

and

c10 = 6492 − 13 × 1802 = (−1)10 .

Therefore, if F = (1, 0, −13) (an ordinary form with discriminant D = 13), then
v F = 2 × P10 = 2 × 649. If F = (2, 1, −6) (an even form with discriminant
D = 13), then v F = P4 = 11, because

c1 = 32 − 13 × 12 = 4(−1)1

and

c4 = 112 − 13 × 32 = 4(−1)4 .

Therefore, F = (2, 1, −6) is abundant, because D = 13 is of type I.


230 J. M. Montesinos Amilibia

Proposition 18 A positive and non-square discriminant D is of type I if and only if


either D = 5;√or D > 5, D ≡ 5 mod 8, and the number 4 is an element of the
c-sequence of D.
To compute practically the c-sequence we proceed as follows.
Write (11) as follows
 t     
Pn Pn−1 1 0 Pn Pn−1 −cn bn
= (−1)n−1 . (12)
Q n Q n−1 0 −D Q n Q n−1 bn cn−1

(Thus, we obtain a sequence of integral quadratic forms Z -equivalent to


1, −D . By
(9) this Z-equivalence is proper if and only if n is even.)
Now we can compute the c-sequence by recurrence. Start with b0 = 0, c0 = 1.
Taking determinants in (12) we get

cn cn−1 = D − bn2 . (13)

Using (7) in (12) we obtain


     
an 1 −cn bn an 1 cn+1 −bn+1
= , (14)
1 0 bn cn−1 1 0 −bn+1 −cn

which implies
bn+1 = cn an − bn (15)
and
cn+1 = −cn an2 + 2bn an + cn−1 . (16)
The √
c-sequence inherits the periodicity properties of the continued fraction expan-
sion of D. It is well known that√there is a positive integer p (the period) such that √the
continued fraction expansion of D is periodic. In fact it turns out that x p = a0 + D.
Hence
√  √   
D = a0 , a1 , a2 , . . . , a p−1 , a0 + D = a0 , a1 , a2 , . . . , a p−1 , 2a0

has the periodic part a1 , a2 , . . . , an−1 , 2a0 . Then x p+1 = x1 , x p+2 = x2 , etc. This
implies that the c-sequence is periodic:

c0 = c p , c1 = c p+1 , . . . (17)

and
b1 = b p+1 , b2 = b p+2 , . . . (18)
The continued fraction expansion of
√  
D = a0 , a1 , a2 , . . . , a p−1 , 2a0

is periodic and symmetric (see for instance [6]). That is


On integral quadratic forms 231

a1 = a p−1 , a2 = a p−2 , . . . .

Using this, we will see that the c-sequence is symmetric .


Using (14) with n = p , (17) and (18) and a p = 2a0 , we obtain:
       
2a0 1 −c p b p 2a0 1 c p+1 −b p+1 c1 −b1
= = . (19)
1 0 bp c p−1 1 0 −b p+1 −c p −b1 −c0

That is,
     
−c p b p −c0 −b1 + 2a0 c0 −c0 a0
= = ,
bp c p−1 −b1 + 2a0 c0 c1 + 4(a0 b1 − a02 c0 ) a0 c1

or
c p = c0 = 1,
c p−1 = c1 , (20)
b p = a0 = b1 .
Using again (14), with n = p − 1, a p−1 = a1 , and (20), we obtain
       
a1 1 −c p−1 b p−1 a1 1 cp −b p c −b1
= = 0 . (21)
1 0 b p−1 c p−2 1 0 −b p −c p−1 −b1 −c1

That is,
       
−c p−1 b p−1 0 1 c0 −b1 0 1 −c1 b2
= = , (22)
b p−1 c p−2 1 −a1 −b1 −c1 1 −a1 b2 c2

where the last equality of (22) holds because it is the result of inverting (14) with
n = 1. Thus

c p−1 = c1 ,
c p−2 = c2 ,
b p−1 = b2 .

Proceeding in this way we see that the c-sequence is symmetric. That is

c p = c0 = 1, c p−1 = c1 , c p−2 = c2 , . . . .

Similarly we also see that the b-sequence is symmetric:

b p = b1 = a0 , b p−1 = b2 , . . .

We will call the periodic part of the c-sequence, namely c0 , c1 , c2 , . . . , c2 , c1 , c0 ,


the Pell sequence (compare [10]).
232 J. M. Montesinos Amilibia

14 Listing B-classes

We have seen that the v-invariant classifies integral indefinite binary forms, with non-
square discriminant, up to B + -equivalence. This, together with the bilaterality of the
form classifies it, up to B -equivalence.
The v-invariant is a positive integer ≥3. An interesting problem is to list, given
a value of v, all the Z-classes with the same v -invariant, bilateral and unilateral. It
would be enough to list the discriminants corresponding to the Z-classes inside a B +
-class, because the problem of listing the Z-classes with a given discriminant is solved
[6,7,13], and the problem of checking bilaterality is the content of Corollary 1 and
Proposition 10.
This can be done as follows.
Case 1. v odd. Here Du 2 = v 2 − 4.
Start with v = 3. Then, Du 2 = 5, u = 1, D = 5. Find the Z -classes of forms of
type II with discriminant D = 5 and separate the bilateral from the unilateral. There
is only one class represented by (2, 1, −2) and it is bilateral.
Continuing like this we find:
v = 5, Du 2 = 21, u = 1, D = 21.
v = 7, Du 2 = 45, u = 1, D = 45; u = 3, D = 5. We discard this last discriminant
D = 5 because it has been obtained already (for v = 3). Here the solution (7, 3) of
Du 2 = v 2 − 4 is not minimal.
We see that it is very easy to find the list of discriminants in the B + -class defined
by a given v.
Of course all these discriminants so obtained are of type I.
Case 2. v even. Here v = 2w and Du 2 = w 2 − 1.
Start with v = 4. Then, w = 2, Du 2 = 3, u = 1, D = 3. Find the Z-classes of
forms (of type I) with discriminant D = 3 and separate the bilateral from the unilateral.
There is only one class represented by (1, 0, −3) and it is bilateral.
Continuing like this we find:
v = 6, w = 3, Du 2 = 8, u = 1, D = 8; u = 2, D = 2. The forms (1, 0, −2) and
(1, 0, −8) have different discriminants but they are B -equivalent. This is the simplest
example of this interesting occurrence.
In the course of these calculations, it helps to remember that the discriminant of an
indefinite integral form of type II is congruent with 1 or 5 mod 8. In view of these
calculations and of Theorem 13 below we have:
Theorem 12 For each value of the invariant v ≥ 3 there is a non-empty B + -class of
integral, binary quadratic forms, with positive non-square discriminants, all realizing
v. Each such B + -class contains only finitely many Z-classes.
Here are the discriminants for the first values of v.
Case 1. v odd.
v| 3 5 7 9 11 13 15 17 19
D| 5 21 45 77 (13; 117) 165 221 285 357

21 23 25 27 29 31
437 525 (69; 621) (29, 725) (93; 837) 957
On integral quadratic forms 233

Case 2. v even.
v 4 6 8 10 12 14 16 18 20
D 3 (2; 8) 15 (6; 24) 35 (12; 48) (7; 63) (5; 20; 80) (11; 99)

22 24 26 28 30 32
(30; 120) 143 (42; 168) 195 (14; 56; 224) 255

15 Some results on Eisenstein problem

Checking a priori if a given form (a, b, c) is ordinary or abundant reduces to the


classical problem, stated by Eisenstein in 1844, that given a positive integer D ≡ 5
mod 8, which is not a square, find a priori (i.e. without finding the solutions) if the
Pell’s equation v 2 − Du 2 = 4 can be solved in (necessarily odd) relatively prime
integers v and u [11]. The interesting point, in the statement of this problem, is of
course the proviso a priori. If, given √ D ≡ 5 mod 8, we could calculate an upper bound
on the length of the period of D, then Eisenstein problem would be solved √ by just
checking in an a priori known number of steps if the Pell’s sequence of D contains
or not the digit 4 (compare [8]). This, as far as I know being not the case, I will next
give some results that can be immediately deduced. We also call the attention of the
reader to the methods of [15], in which the author is interested in a similar problem.
From the proof of Theorem 9 it is not difficult to prove √(see [8]) the following
properties of the Pell’s sequence c0 , c1 , c2 , . . . , c2 , c1 , c0 of D, when D ≡ 5 mod 8.
The length l(D) of the Pell’s sequence will be the number of members √ of the sequence
c0 , c1 , c2 , . . . , c2 , c1 . It coincides with the length of the period of D.

1. Only the extreme members of the Pell’s sequence c0 , c1 , c2 , . . . , c2 , c1 , c0 of D
equal 1. √
2. The Pell’s sequence c0 , c1 , c2 , . . . , c2 , c1 , c0 of D contains the digit 4 at most
twice (and symmetrically disposed). √
3. If the Pell’s sequence c0 , c1 , c2 , . . . , c2 , c1 , c0 of D contains the digit 4 and l(D)
is odd (resp. even) then each digit 4 occupies an even (resp. odd) position counting
from the nearest end.

For instance the Pell’s sequence of 13 has odd length l(13) = 5, and the digit 4 is
in the position 2 from each end

1, 4, 3, 3, 4, 1.

The Pell’s sequence of 45 has even length l(13) = 6, and the digit 4 is in the
position 3 from each end

1, 9, 4, 5, 4, 9, 1.

Corollary 3 If D ≡ 5 mod 8 and the length l(D) of the Pell’s sequence

c0 , c1 , c2 , . . . , c2 , c1 , c0
234 J. M. Montesinos Amilibia


of D is 1, 2 or 4, then, it fails to contain the digit 4.

Theorem 13 Let D a positive integer that is not a square. Then



1. l(D) = 1 if and only if D = n 2 + 1, n ≥ 1. Then, D = [n, 2n]. The Pell’s
sequence is 1, 1. √
2. l(D) = 2 if and only if D = n 2 + t, where n ≥ 1 and 1 < t|2n. Then, D =
[n, 2n/t, 2n] and the Pell’s sequence is 1, t, 1. It might contain the digit 4, but not
if D ≡ 5 mod 8.
3. l(D) = 3 if and only if D = n 2 + t,√where n = (4vw + 1)v + w, t = 4vw + 1, and
v > 0, w > 0 are integers. Then, D = [n, 2v, 2v, 2n] and the Pell’s sequence
is 1, t, t, 1. Therefore the Pell’s sequence does not contain the digit 4.

Proof Set D = n 2 + t, where 1 ≤ t ≤ 2n. Note that, then, [ D] = n, because
n 2 < D < (n + 1)2 . Thus, to compute c1 , we consider

1 1 n+ D
√ √ =√ = .
D − [ D] D−n t

Therefore, b1 = n, c1 = t. Then, l(D) = 1 if and only if c0 = c1 = t = 1, and this


proves part 1. From (13) and (15 ), we have

D − b22
c2 = = 1 − a1 (ta1 − 2n).
c1

Since a1 = 0, it follows that c2 = 1 if and only if ta1 = 2n. This completes the proof
of part 2.
Assume now that l(D) = 3. This is equivalent to have
√    √ 
D = n, a1 , a1 , 2n = n, a1 , a1 , n + D , where a1 < 2n

An easy calculation shows that the last equation is equivalent to have

t (a12 + 1) = 2na1 + 1, where t > 1. (23)

Since t (a12 + 1) ≡ 1 mod 2, we deduce that t = 2u + 1 with u > 0 , a1 = 2v with


v > 0, n = 2uv + v + u/2v. Therefore u = 2vw with w > 0. Hence

n = (4vw + 1)v + w, with v > 0, w > 0,


t = 4vw + 1,
a1 = 2v.

Conversely, if D = n 2 +t, where n = (4vw+1)v+w, t = 4vw+1, and v > 0, w > 0


are integers, then (23) holds. This proves part 3 and completes the proof. 

On integral quadratic forms 235

The first values of D ≡ 5 mod 8 with l(D) = 1 are

5, 37, 101, 197, 325.

The first values of D ≡ 5 mod 8 with l(D) = 2 are

333, 405, 1173, 1605.

The first values of D ≡ 5 mod 8 with l(D) = 3 are

269, 1325, 1613, 3181, 4933, 5837.

Another corollary of the above three properties of the Pell’s sequence is the following:

Corollary 4 Assume the Pell’s sequence c0 , c1 , c2 , . . . , c2 , c1 , c0 of D contains the
digit 4 and D ≡ 5 mod 8. If l(D) is 5 then the digit 4 appears in positions 2 from
each end of the Pell’s sequence. If l(D) is 6 or 8 then the digit 4 appears in positions
3 from each end of the Pell’s sequence.
This corollary prompts one to obtain the numbers D such that 4 appears in place 2,
3, etc. in the Pell’s sequence. I think it is hopeless trying to understand Eisenstein
problem from this perspective. However, here are some obvious results.

Theorem 14 The Pell’s sequence c0 , c1 , c2 , . . . , c2 , c1 , c0 of D starts with c1 = 4
if and only if D = n 2 + 4, n ≥ 2. And it starts with c1 = 1, c2 = 4 if and only if
D = n 2 − 4, n ≥ 5 or D = (9n − 4)n, n ≥ 2.
Proof The first part is proved as in Theorem 13, part 1. From the proof of Theorem 13,
part 2, we have

c2 = 1 − a1 (ta1 − 2n) = 4.

Therefore the Pell’s sequence of D starts with c1 = 1, c2 = 4 if and only if
D = n 2 + t, where t does not divide 2n and a1 (2n − ta1 ) = 3. Therefore, either
a1 = 1 and 2n − ta1 = 3 or a1 = 3 and 2n − ta1 = 1. In both cases t is odd, that is
t = 2u + 1. In the first case

t = 2u + 1, n = 2 + u, D = u 2 + 6u + 5

and u ≥ 2 since, exactly for u = 1, t divides 2n. Hence, setting v = u + 3, we have


D = v 2 − 4, v ≥ 5. In the second case,

t = 2u + 1, n = 2 + 3u, u ≥ 1, D = 9u 2 + 14u + 5.

Hence, setting u = v − 1, we have

D = 9v 2 − 4v, v ≥ 2.

This completes the proof.



236 J. M. Montesinos Amilibia

The first values of D = n 2 − 4, n ≥ 5 are

21, 32, 45, 60, 77, 96, 117, 140.

The first values of D = (9n − 4)n, n ≥ 2 are

28, 69, 128, 205, 300, 413, 544, 693.

Acknowledgments I am very grateful to the referee for his careful reading of the manuscript and for his
suggestions that have enhanced the paper greatly.

References

1. Bianchi, L.: Complemento alle ricerche sulle forme quaternarie quadratiche e sui gruppi poliedrici.
Annali di Matematica Pura e applicata 2(23), 1–44 (1894)
2. Bianchi, L.: Ricerche sulle forme quaternarie quadratiche e sui gruppi poliedrici. Annali di Matematica
Pura e applicata 2(21), 237–288 (1893)
3. Bianchi, L.: Lezioni sulla teoria delle funzioni di variable complessa e delle funzioni ellittiche. Enrico
Spoerri, Pisa (1901)
4. Brown, H., Bülow, R., Neubüser, J., Wondratschek, H., Zassenhaus, H.: Crystallographic groups of
four-dimensional space. Wiley, New York (1978)
5. Cahen, E.: Éléments de la Théorie des Nombres. Gauthier-Villards, Paris (1900)
6. Cahen, E.: Théorie des Nombres, Tome second: le second degré binaire. Hermann, Paris (1924)
7. Cassels, J.W.S.: Rational quadratic forms. Academic Press, New York (1978)
8. Cayley, A.: Note sur l ’équation x 2 − Dy 2 = ±4, D ≡ 5 mod 8. Journal für die Reine und andgewandte
Mathematik 9, 369–371 (1826)
9. Conway, J.H.: The sensual (quadratic) form. Carus Math. Monographs No. 26. MAA (1977) (assisted
by Francis Y. C. Fung)
10. Degen, C.F.: Cannon Pellianus sive tabulla simplicissimam aequationis celebratissimae y 2 = ax 2 + 1
solutiones pro singulis numeri dati valoribus ab 1 usque ad 1000 in numeris rationalibus iisdemque
integris exhibens, Havniae (Copenhagen) (1817)
11. Eisenstein, G.: Aufgaben. Journal für die Reine und andgewandte Mathematik 27, 86 (1844)
12. Elstrodt, J., Grunewald, F., Mennicke, J.: Groups Acting on Hyperbolic Space. Springer Monographs
in Mathematics. Springer, New York (1998)
13. Jones, B.W.: The Arithmetic Theory of Quadratic Forms, MAA. Carus Mathematical Monographs,
No. 10 (1950)
14. Maclachlan, C., Reid, A.W.: The Arithmetic of Hyperbolic 3-Manifolds, Graduate Texts in Mathemat-
ics, vol. 219. Springer, New York (2003)
15. Mc Laughlin, J.: Multi-variable polynomial solutions to Pell’s equation and fundamental units in real
quadratic fields. Pac. J. Math. 210, 335–349 (2003)
16. Milnor, J., Husemoller, D.: Symmetric bilinear forms, Ergebnisse der Math. No. 73. Springer, Berlin
(1973)
17. Montesinos-Amilibia, J.M.: On integral quadratic forms having commensurable groups of automor-
phisms. Hiroshima Math. J. 43, 371–411 (2013)
18. Montesinos-Amilibia, J.M.: Addendum to “On integral quadratic forms having commensurable groups
of automorphisms”
19. Narkiewicz, W.: Elementary and Analytic Theory of Algebraic Numbers, 2nd edn. Springer, Berlin
(1990)
20. Ratcliffe, J.G.: Foundations of Hyperbolic Manifolds, 2nd edn. Graduate Texts in Mathematics, vol.
149. Springer Verlag, New York (2006)
21. Scharlau, W., Opolka, H.: From Fermat to Minkowski. Springer, Berlin (1985)
22. Smith, H.J.S.: Report on the Theory of Numbers. Chelsea Pub, New York (1965)
23. Thurston, W.P.: The Geogmetry and Topology of Three-manifolds. Princeton University Lectures
(1976–1977)
24. Wezel, J.L.: Annales Acad. Leodiensis, Liège 2, 24–30 (1821)

You might also like