a

r
X
i
v
:
1
3
0
6
.
3
9
9
8
v
1


[
m
a
t
h
.
N
T
]


1
7

J
u
n

2
0
1
3
The spinor genus of the integral trace.
Guillermo Mantilla-Soler
Abstract
Let K be a number eld of degree at least 3. In this article we show that the genus
of the integral trace form of K contains only one spinor genus.
Introduction
Let K be a number eld. The rational quadratic form x tr
K/Q
(x
2
) has been extensively
studied by several authors, see for example [Ba], [BaLe], [C-P], [Ep], [Ga], [Mau], [MiRe]
and [S]. For arithmetic purposes, a ner invariant of a number eld is its integral trace form
i.e., the integral quadratic form obtained by restricting tr
K/Q
(x
2
) to the maximal order in
K. Recent applications of the integral trace form can be found in the work of Bhargava and
Shnidman (see [BhSh]) where they count cubic orders using the shape, an invariant closely
related to the integral trace form. Some other applications of the integral trace on cubic
elds can be found in [Man] and [Man2]. Given a non-degenerate integral quadratic form q,
it is of great interest to study the number of spinor classes on its genus. This is explained
by the famous result of Eichler [Eich] which says that for an indenite form q of dimension
of at least 3 the spinor genus and the isometry class coincide. In this article we analyze the
spinor genus of the integral trace form of a number eld. Our main theorem is the following:
Theorem (cf.Theorem 2.12). Let K be a non-quadratic number eld. Then, the genus of
the integral trace form of K contains only one proper spinor genus.
1 Denitions, notations and basic facts
1.1 Notation
We summarize here the most important notation used in the paper.
For a Q
p
we have that v
p
(a) is the usual p-adic valuation. We will use A to denote
the adele ring over Q, and for a eld K we will denote by G
K
the absolute Galois
group Gal(K
sep
/K).
1
Let a
1
, ..., a
n
be elements of a ring R, which in practice will be a maximal order on a
number eld or a local eld, the R-isometry class of a quadratic form a
1
x
2
1
+... +a
1
x
2
1
will be denoted by a
1
, ..., a
n
. Whenever there is a possible ambiguity in the ring of
denition of an isometry between two quadratic forms we will write

=
R
to make it
clear that the forms are considered to be over R.
The isometry class of the binary integral quadratic form 2xy over Z
2
, the hyperbolic
plane, will be denoted by H.
Most of our notation for quadratic forms is adopted from [Ca] or [O]. Unexplained
terminology is either standard or can be found in either of the aforementioned refer-
ences.
1.2 Equivalences between integral quadratic forms
We briey recall the standard notions of equivalence between integral quadratic forms. As it
is customary we use geometric language. Let V be a non-degenerate quadratic space over Q
with orthogonal (resp. special orthogonal) group O(V ) (resp. SO(V )) and let /
V
be the set
of lattices of maximal rank inside V . Let O
A
(V ) (resp. SO
A
(V )) be the adelic orthogonal
(resp. special orthogonal) group. The group O
A
(V ) acts on /
V
and the dierent notions
of equivalence between lattices are just given by orbits in /
V
, under the restricted action,
of certain subgroups of O
A
(V ). By using the usual diagonal embedding we may view O(V )
as a subgroup of O
A
(V ). The notions of class, proper class and genus can all be given in
terms of the groups O(V ), SO(V ) and SO
A
(V ), however for the spinor genus we need the
group
A
(V ) which we dene next. Let Spin(V ) be the simply connected linear algebraic
group over Q that is the universal cover of the algebraic group SO(V ). Explicitly, there is
an exact sequence of algebraic groups
1
2
Spin(V ) SO(V ) 1.
By taking G
Q
-invariance in the geometric sequence induced by the above exact sequence
(see [KMRT, Proposition 22.15]), we get a connecting homomorphism of Galois cohomology
SO(V ) H
1
(Q,
2
)

= Q

/(Q

)
2
.
The obtained homomorphism : SO(V ) Q

/(Q

)
2
is the so called spinor norm. By
considering, at every prime p, the quadratic space V
p
:= V
Q
Q
p
we have, as above, the
local spinor norm

p
: SO(V
p
) Q

p
/(Q

p
)
2
.
Putting all the local spinor norms together we have a group homomorphism
1

A
: SO
A
(V )

p
Q

p
/(Q

p
)
2
with kernel denoted by
A
(V ) := Ker(
A
).
1
The image of
A
is in fact contained in A

/A
2
.
2
Denition 1.1. Let be a lattice in /
V
. For a subgroup G of O
A
(V ) the orbit of under
the action of G is denoted by G().
1. The class of (resp proper class) is cl() := O(V )() (resp cl
+
() := SO(V )()).
2. The genus of (resp proper genus) is gen() := O
A
(V )() (resp gen
+
() := SO
A
(V )()).
3. The spinor genus of (resp proper spinor genus) is spin() := O(V )
A
(V )() (resp
spin
+
() := SO(V )
A
(V )()).
Given M /
V
we say that and M are in the same class (resp. proper class, genus, proper
genus, spinor genus, proper spinor genus) if they belong to the same orbit by the respective
subgroup.
It is not dicult to see that gen() = gen
+
() for any lattice (see [Ca, Chapter 8
Corollary to Lemma 3.2]) so the following are immediate consequences of the denitions:
cl() spin() gen()
=
cl
+
() spin
+
() gen
+
().
The relevance of spinor equivalence, see [Eich], comes from the fact that for indenite
and of dimension bigger than 2 we have that
cl
+
() = spin
+
() and cl() = spin().
The following criterion is a very useful tool to determine when an integral lattice has
spin
+
() = spin() = gen().
Theorem 1.2. Let V be a non-degenerate quadratic space over Q, and let be an integral
lattice in V . For a prime p let
p
:= Z
p
, and let
p
be the spinor norm on the Q
p
-
quadratic space V
p
:= V
Q
Q
p
. Suppose that for all primes p we have that
2

p
(
p
) Z

p
.
Then, the genus of contains only one proper
3
spinor genus.
Proof. See [Ca, 11, Corollary pg 213].
1.3 Outline of the proof
Let K be a number eld with maximal order O
K
and let q
K
be its integral trace form i.e., the
integral quadratic form obtained by restricting x tr
K/Q
(x
2
) to O
K
. The maximal order
O
K
is an integral lattice, with integral quadratic q
K
, in the quadratic space (K, tr
K/Q
(x
2
)).
2
See [Ca, Chapter 11 1 (1.20) and (1.21)] for the denition of p(p).
3
What Cassels calls spinor genus, [Ca, Chapter 11 1], is what is usually called proper spinor genus [O,
Chapter X 102].
3
We will denote this lattice and quadratic space by and V , respectively. To obtain that
spin
+
() = gen() it is enough to show, thanks to Theorem 1.2, that for all prime p

p
(
p
) Z

p
mod (Q

p
)
2
. (1)
For primes p that are at worst tamely ramied in K there is a Jordan decomposition of
the lattice
p
that allows us to verify (1); see Theorem 2.2 for details.
To obtain (1) for wild primes we use the following result together with standard upper
bounds on the p-adic valuations of the discriminant of a number eld (see Corollary 2.5).
Theorem 1.3. Let V be a non-degenerate rational quadratic space of dimension n > 2. Let
be a full rank integral lattice of discriminant D, and suppose that p is a prime such that

p
(
p
) , Z

p
mod (Q

p
)
2
. Then,
1. n(n 1)/2 v
p
(D) if p is odd.
2. n(n 3) + 2

n+1
2

v
2
(D) if V is denite.
3. n(n 1) v
2
(D) if V is indenite.
Proof. See both Corollaries in page 214 of [Ca] and the proofs of [E-H, Theorem 4.2, Theorem
4.6].
Remark 1.4. It is worthwhile to point out that the initial step in the strategy to show (1)
is the same for every prime regardless of the ramication type. We explain this by looking
at the argument for an odd prime p. The common strategy here is to use a diagonalization

1
, ...,
n
of the Z
p
-lattice
p
and obtain (1) from information about the
i
s. In the case
of at worst tame ramication we know explicitly the values of the
i
s and from them we
get (1) (see Theorem 2.2). On the other hand, if the prime p is wild we can assume that
v
p
(
i
) ,= v
p
(
j
) for i ,= j, otherwise
p
contains an orthogonal factor
(p)
with
p
(
p
)

p
(
p
) Z

p
mod (Q

p
)
2
. In particular, we have that
v
p
(
1
...
n
) =
n

i=0
v
p
(
i
) n(n 1)/2.
This is precisely the idea behind Theorem 1.3.1, hence to obtain (1) we are only left with
the task to show that this inequality never occurs for
1
, ...,
n
.
2 The proof
As often happens, it is convenient to start with number elds with only tame ramication.
Afterwards we will proceed to the case of wild ramication. Throughout this section all
number elds are assumed to have degree bigger than 2.
4
2.1 Local integral trace at a tame prime
In this section we describe the Jordan decomposition of the integral trace form when localized
at a prime with at worst tame ramication. For details and proofs see [Man3]. Given L a
number eld and p a rational prime we denote by g
p
the number of primes in O
L
lying over
p. Furthermore,
F
p
:=
gp

i=1
f
i
where f
1
, ..., f
gp
are the inertia degrees of the prime p in L.
2.1.1 Jordan decomposition
The following result describes the Jordan decomposition of the localization of the integral
trace in terms of residue and inertia degrees. For details see [Man3].
Theorem 2.1. Let L be a degree n number eld and let q
L
be its integral trace form. Let p
be a rational prime which is not wildly ramied in L. Then,
q
L
Z
p

=
_

_
1, ...., 1,
p

. .
Fp

p 1, ..., 1,
p

. .
nFp
if p ,= 2,

1
, ....,
F
2

. .
F
2

2 (H... H)
. .
nF
2
2
otherwise.
Where
p
,
p
,
i
are some elements Z

p
that can be explicitly calculated in terms of the ram-
ication of the prime p at L.
Proof. See [EMP, Theoreme 1.3] and [Man3, Theorem 0.1].
2.1.2 Tame primes
Thanks to Theorem 2.1 we can easily deal with primes that at worst tamely ramied. Recall
that we denoted by V the quadratic space given by (K, tr
K/Q
(x
2
)), where K is number eld
of dimension at least 3, and by the integral lattice in V given by the maximal order O
K
.
Theorem 2.2. Let p be a prime, and suppose that p is either unramied or tame in K.
Then,
p
(
p
) Z

p
mod (Q

p
)
2
.
Proof. Suppose rst that p ,= 2. By Theorem 2.1 there are , Z

p
such that

p

= 1, 1, ..., 1,

p, p, ..., p
Since dim
p
3 either one of the factors in the above orthogonal decomposition has di-
mension at least 2. Hence
p
contains a two dimensional orthogonal factor
(p)
among the
5
following 1, , p, p. If p = 2 then, again by Theorem 2.1, there are
1
, ...,
m
Z

2
such
that

2

=
1
, ...,
m

2 (H... H).
Thus
2
contains an orthogonal factor
(2)
of the form 2H or
1
,
2
,
3
. The result follows
from [Ca, XI, 3, Lemma 3.7, Lemma 3.8].
2.2 Wild primes
We rst recall some classic bounds on the p-adic valuation of the discriminant of a number
eld.
Discriminant bounds. Let n be the degree of K and Let D
K
be its discriminant. Let p
be a prime and suppose that e
1
, ..., e
g
are the ramication degrees of p with corresponding
residue degrees f
1
, ..., f
g
. Let F
p
:=
g

i=1
f
i
and M
p
= maxv
p
(e
i
). The following well known
bounds on the discriminant can be found in [S2, III, 6, Remark Proposition 13.].
Theorem 2.3.
v
p
(D
K
) n F
p
+
g

i=1
e
i
f
i
v
p
(e
i
),
where equality is obtained if M
p
= 0.
Remark 2.4. Notice that M
p
= 0 is another way to say that the prime p is either unramied
or tame in K. In such case v
p
(D
K
) = n F
p
follows immediately from Theorem 2.1.
An immediate consequence of Theorem 2.3 is:
Corollary 2.5.
v
p
(D
K
) n(M
p
+ 1) F
p
.
Odd primes. In the case of wild ramication it is convenient to separate the cases p = 2
and p ,= 2. Recall that in this section we are always assuming that that n 3.
Theorem 2.6. Let p 5 be a prime. Then,
p
(
p
) Z

p
mod (Q

p
)
2
.
Proof. If n < 5 then p is either unramied or tame, both cases covered in Theorem 2.2.
Hence we may assume that 5 n. Under this assumption we have that n 5
n3
2
p
n3
2
.
Since p
Mp
n it follows that M
p

n3
2
, thus v
p
(D
K
) n(n 1)/2 F
p
< n(n 1)/2. The
result follows from Theorem 1.3.(1)
Theorem 2.7.
3
(
3
) Z

3
mod (Q

3
)
2
.
6
Proof. It is convenient to divide the problem in cases according to the degree n. As in the
proof of Theorem 2.6 we may assume that 3 has wild ramication in K i.e., 1 M
3
.
If 5 n 8 then M
3
= 1. Thus, M
3

n3
2
which leads to the result in the exact
same way as in the proof of Theorem 2.6.
If 9 n then n < 3
n3
2
. Since 3
M
3
n we see that M
3
<
n3
2
and we argue as before.
Suppose n = 3. Since 3 is ramied then tr
K/Q
(O
K
) 3Z. Thus, after diagonalizing
over Z
3
we get that
3

= 3
a
1
u
1
, 3
a
2
u
2
, 3
a
3
u
3
where each u
i
Z

3
and the a
i
s are
positive integers. Notice that 3 is totally ramied in K, hence by Theorem 2.3 we
have that v
3
(D
K
) 5. It follows that there are i, j with i ,= j and such that a
i
= a
j
,
otherwise v
3
(D
K
) = a
1
+a
2
+a
3
1 +2 +3 = 6. Thus
3
contains a two dimensional
orthogonal factor
(3)
of the form 3
a
, 3
a
u, where a is a positive integer and u Z

3
,
and the result follows from [Ca, XI, 3, Lemma 3.7, Lemma 3.8.].
Suppose n = 4. Since 3 is ramied then 3O
K
= B
3
1
B
2
for some prime ideals B
i
s of O
K
.
It follows that
3

= 1

3
where

3
is the integral trace lattice of a totally ramied
cubic extension of Q
3
. By the same argument we used in the case n = 3 applied to

3
we conclude that

3
has an orthogonal factor
(3)
of the form 3
a
, 3
a
u, and thus it is
also an orthogonal factor of
3
from which the result follows.
The case p = 2. Before dealing with the even prime we need an auxilliary result.
Lemma 2.8. Suppose n ,= 4. Then,
v
2
(D
K
) n(n 2)
with equality possible only in the case n = 3.
Proof. If n 6 then 2
n3
> n 2
M
2
, thus M
2
< n3. Since 2
M
2
n we have that M
2
2
if n = 5. In particular, for all n 5 we have that M
2
n 3. Thanks to Corollary 2.5
we have that v
2
(D
K
) n(n 2) F
2
< n(n 2). Now suppose n = 3. If 2 is unramied
the inequality is obvious. Otherwise 2 has ramication degrees 1, 2 and residue degrees 1, 1
which by Theorem 2.3 implies that v
2
(D
K
) 3 = n(n 2).
Proposition 2.9. Suppose that K is not a totally real quartic eld. Then,
2
(
2
) Z

2
mod
(Q

2
)
2
.
7
Proof. Since n(n 2) minn(n 3) + 2

n+1
2

, n(n 1) the result follows, in the case

n ,= 4, from Lemma 2.8 and Theorem 1.3. If n = 4, we have that M
2
2 and by Corollary
2.5 we see that v
2
(D
K
) 11. On the other hand since K is non totally real is indenite
and the result follows from Theorem 1.3.(3).
Remark 2.10. Most of the hard work in proving the main result comes when dealing with
the primes 2 and 3 for small values of n. In particular notice that at this point we can
already conclude that
gen(q
K
) = spin
+
(q
K
)
for all number elds K that are either at worst tame at 2 or that are non-quartic and totally
real. Next we deal with this, the nal case.
Theorem 2.11.
2
(
2
) Z

2
mod (Q

2
)
2
.
Proof. We may assume that n = 4 and that M
2
1, 2. If M
2
= 1 then, by Corollary
2.5, v
2
(D
K
) 7 < n(n 2) and we use the argument of Proposition 2.9. Furthermore,
if c := v
2
(D
K
) 8 the same argument applies hence we may assume that M
2
= 2 and
that 9 c. Therefore L := K
Q
Q
2
is a totally ramied quartic extension of Q
2
, with
discriminant of valuation c 9, 10, 11. There are 36 such elds (see for instance [Jo]).The
ring of integers of L is of the form O
L
= Z
2
[
L
] for some in
L
O
L
. The table below
consists of the minimal polynomial of
L
for each one of the 36 elds and their corresponding
discriminant valuation c.
c = 9 c = 10 c = 11 c = 11 c = 11
x
4
+ 6x
2
+ 2 x
4
+ 2x
2
9 x
4
+ 12x
2
+ 2 x
4
+ 6 x
4
+ 8x + 10
x
4
2x
2
+ 2 x
4
+ 2x
2
1 x
4
+ 4x
2
+ 18 x
4
+ 22 x
4
+ 8x + 6
x
4
+ 6x
2
+ 10 x
4
+ 6x
2
9 x
4
+ 12x
2
+ 18 x
4
+ 14 x
4
+ 8x + 14
x
4
+ 2x
2
+ 10 x
4
+ 6x
2
1 x
4
+ 4x
2
+ 10 x
4
+ 30 x
4
+ 8x
2
+ 8x + 22
x
4
+ 2x
2
2 x
4
6x
2
+ 3 x
4
+ 12x
2
+ 10 x
4
+ 26
x
4
2x
2
2 x
4
+ 6x
2
+ 3 x
4
+ 4x
2
+ 14 x
4
+ 10
x
4
+ 2x
2
+ 6 x
4
2x
2
+ 3 x
4
+ 4x
2
+ 6 x
4
+ 18
x
4
2x
2
+ 6 x
4
+ 2x
2
+ 3 x
4
+ 12x
2
+ 6 x
4
+ 2
Suppose that the minimal polynomial of
L
is of the form x
4
+ 2ax
2
+b.
Out of the 36 polynomials in the above table, 32 are of this form and they satisfy
exactly one of the following
(i) v
2
(a) = 0, v
2
(b) = 1,
(ii) v
2
(a) = 0, v
2
(b) = 0,
(iii) v
2
(a) = 1, v
2
(b) = 1.
(iv) a = 0, v
2
(b) = 1.
8
In particular, b/a Z

2
. Moreover, in case (ii) we have that all the values of b satisfy
that b 1 mod 4. The Gram matrix of the trace form in the basis 1,
L
,
2
L
,
3
L
is
given by
_

_
4 0 4a 0
0 4a 0 8a
2
4b
4a 0 8a
2
4b 0
0 8a
2
4b 0 16a
3
+ 12ab
_

=
4
_

_
4 0 0 0
0 4a 0 8a
2
4b
0 0 4a
2
4b 0
0 8a
2
4b 0 16a
3
+ 12ab
_

=
_

_
4 0 0 0
0 4a
2
4b 0 0
0 0 4a 8a
2
4b
0 0 8a
2
4b 16a
3
+ 12ab
_

=
_

_
4 0 0 0
0 4a
2
4b 0 0
0 0 4a 4b
0 0 4b 4ab
_

=
_

_
4 0 0 0
0 4a
2
4b 0 0
0 0 4a 0
0 0 0 4
b
a
(a
2
b)
_

_
It follows that

2

=
_

2
2
, 2
2
(a
2
b), 2
2
(a), 2
3
(
a
2
bb
2
2a
)

if v
2
(a) = 0 and v
2
(b) = 1,
_
2
2
, 2
3
(a
2
b)
2
, 2
2
(a), 2
3
(
a
2
bb
2
2a
)
_
if v
2
(a) = 0 and v
2
(b) = 0,
_
2
2
, 2
3
(a
2
b)
2
, 2
3
(
a
2
), 2
3
(
a
2
bb
2
2a
)
_
if v
2
(a) = 1 and v
2
(b) = 1,

2
2
, 2
3
(
b
2
)

2
3
H if a = 0 and v
2
(b) = 1.
Notice that in case (i), since v
2
(a
2
) < v
2
(b), we have that v
2
(a
2
b) = v
2
(a
2
) = 0 i.e.,
a
2
b Z

2
. Similarly in case (iii) it follows from v
2
(b) < v
2
(a
2
) that
(a
2
b)
2
Z

2
. In
case (ii) we also have that
(a
2
b)
2
Z

2
; this follows since in this case all the values of
b satisfy that b 1 mod 4.
If the minimal polynomial of
L
is of the form x
4
+ 8x + 2b with v
2
(b) = 0, three out
of the remaining four polynomials in the table are of this form, then similarly as above
we have that

2

=

2
2
, 2
3
(b)

2
3
H.
Finally if the minimal polynomial of
L
is x
4
+ 8x
2
+ 8x + 22 then

2

=

2
2
, 2
3
(3), 2
3
, 2
3

773
3

.
From the description above we see that
2
contains an orthogonal factor
(2)
of one of the
following forms: 2
3
H or 2
e
u
1
, 2
e
u
2
, 2
e
u
3
or 2
e
u
1
, 2
e
u
2
, 2
e+1
u
3
, where all the u

i
s are in Z

2
.
The result follows from [Ca, XI, 3, Lemma 3.8].
4
By this we mean that the quadratic forms having such Gram matrices are equivalent.
9
Theorem 2.12. Let K be a number eld of degree at least 3. Then, the genus of integral
trace form q
K
contains only one proper spinor genus, thus it only contains one spinor genus.
Proof. This follows from Theorems 2.6, 2.7, 2.11 and 1.2.
Corollary 2.13. Let K be a non totally real number eld. Then, any integral quadratic
form that is equivalent to q
K
is properly equivalent to it.
Proof. The result is clear for K quadratic, just compose with complex conjugation on K
whenever the original isometry is not proper. For higher dimensions we have
cl(q
K
)
1
spin(q
K
) gen(q
K
)
=
cl
+
(q
K
)
2
spin
+
(q
K
)
3
gen
+
(q
K
).
where contaiments 1 and 2 are equalities by the celebrated result of Eichler [Eich] and
containment 3 is an equality by Theorem 2.12. It follows that all the containments in the
above diagram are equalities, in particular we have that cl
+
(q
K
) = cl(q
K
).
3 Quadratic elds
Our interest in the spinor genus of the trace began as the search for a tool to understand
when a pair of number elds have the same integral traces. Since quadratic elds are
completely characterized by their discriminants the question about the isometry between
their traces becomes trivial. However, since all the methods applied above relied heavily on
the assumption that the number elds have dimension at least 3, it is interesting on its own
to see if Theorem 2.12 remains valid in dimension 2. The following example shows that this
is not the case.
Given a square rational matrix S the denominator of S denoted by den(S) is the smallest
positive integer such that den(S)S has integral entries. The weight of S denoted by w(S) is
the smallest positive integer such that w(S
1
)det(S
1
) is an integer whenever S
1
is any square
submatrix of S. See [Wat, Chapter 6, 1] for details.
Example 3.1. Let K be the quadratic number eld of discriminant 17. Then, its integral
trace form q
K
is equivalent to the form 2, 2, 9 which genus contains two spinor genus;
2, 2, 9 and 1, 17.
Proof. The form 2, 2, 9 is the integral trace form q
K
written in the standard ordered basis
of O
K
given by B :=
_
1,
1+

17
2
_
. The form 1, 34 corresponds to the integral lattice M
10
inside K spanned by the ordered basis B
1
:=

1 +

17
6
,
8 +

17
6

. Since gcd(3, 34) = 1

and the change of basis matrix
R =

1/3 3
1/3 0

from B
1
to B has den(R) = 3 we have that gen(O
K
) = gen(M). On the other hand
cl(O
K
) ,= cl(M) since 1 is represented by 1, 17 and not by 2, 2, 9.
The Gram matrix of O
K
in the basis B is A =

2 1
1 9

, and the Gram matrix of M in the

basis B
1
is B =

1 0
0 17

. Since R
t
BR = A and w(R) = 3 we have, by [Wat, Theorem 64],
that spin(O
K
) = spin(M) if and only if there is an integer k and a rational matrix S such
that
SBS
t
= B and k
2
3w(S) 1 mod 34. (2)
Suppose such a matrix S exists. We may assume that det(S) = 1 since otherwise we can
replace S by TS, where T =

1 0
0 1

. If the matrix S is given by

a b
c d

with ad bc = 1
then the equation SB = B(S
t
)
1
leads to a = d, c = 17b. Thus, d
2
+ 17b
2
= 1. Using the
point (b, d) = (0, 1) one nds that the rational points of this conic are of the form
b =
2mn
17m
2
+n
2
; d =
17m
2
n
2
17m
2
+n
2
for some m and n relatively prime integers. Hence S can be written as
S(m, n) :=
1
17m
2
+n
2

17m
2
n
2
2mn
34mn 17m
2
n
2

for some m, n.
Since gcd(m, n) = 1 and for all m, n
1
integers S(m, 17n
1
) = S(n
1
, m) we may assume
that 17 n. If n is even there is n
1
relatively prime to 17m such that
gcd(17m
2
2n
2
1
, 17m
2
+ 2n
2
1
) = gcd(17m
2
, 2n
2
1
) = 1.
Therefore w(S) = 17m
2
+ 2n
2
1
. If n is odd and m is even we have that
gcd(17m
2
n
2
, 17m
2
+n
2
) = gcd(17m
2
, n
2
) = 1
thus w(S) = 17m
2
+n
2
. Finally if both m and n are odd
gcd

17m
2
n
2
2
,
17m
2
+n
2
2

= gcd(17m
2
, n
2
) = 1
thus w(S) =
17m
2
+n
2
2
. We have shown that for any rational matrix S such that SBS
t
= B
there are integers m, n such that either w(S) = 17m
2
+ n
2
or w(S) = 17m
2
+ 2n
2
or
11
2w(S) = 17m
2
+n
2
. Therefore, thanks to (2), if O
K
and M were in the same spinor genus
there would exists an integer d such that
either 3d
2
1 mod 17 or 6d
2
1 mod 17 or 3d
2
2 mod 17.
Since

3
17

6
17

= 1 we conclude that spin(O

K
) ,= spin(M).
Remark 3.2. The above example is minimal in the sense that for any quadratic eld K with
discriminant D with [D[ < 17 one has that spin(q
K
) = gen(q
K
).
It would be interesting to see for which quadratic elds K the conclusion of Theorem
2.12 is valid. A simple way to construct quadratic elds that satisfy Theorem 2.12 comes
from Gauss genus theory. Lets say for simplicity that the discriminant D of K is square
free. If the narrow class group Cl
+
(D) of the quadratic eld Q(

D) has exponent at
most 2 then for every primitive binary quadratic form q = a, b, c, with b
2
4ac = 4D, we
have that cl(q) = gen(q) so in particular spin(q) = gen(q). Examples of this are given by the
quadratic elds Q(

3), Q(

5) and Q(

13). There exist only nitely many positive Ds

with such properties; they are called idoneal numbers (see [Cox, Chapter 1 2 C]). As far as
we know, for general D, it is not known whether or not there are innitely many quadratic
elds of discriminant D such that Cl
+
(D)[2] = Cl
+
(D). We nish with the following
related question.
Question 3.3. Are there innitely many K, quadratic elds, with spin(q
K
) = gen(q
K
)?
Acknowledgements
I would like to thank Eva Bayer-Fluckiger for many helpful conversations around the topic
of this paper, and for her thorough and helpful comments on an earlier version of it.
References
[Ba] E. Bayer-Fluckiger Galois Cohomology and the Trace form, Jahresber. Deutsch. Math.-
Verein 96-2 (1994), 35-55.
[BaLe] E. Bayer-Fluckiger, H.W. Lenstra forms in odd degree extensions and self-dual nor-
mal bases, Amer. J. Math. 112 (1990), 359-373.
[BhSh] M. Bhargava, A. Shnidman On the number of cubic orders of bounded discriminant
having automorphism group C
3
, and related problems, preprint (2012).
[Ca] J.W. S. Cassels, Rational quadratic forms, Dover publications, Inc., Mineola, NY,
(2008).
12
[Cox] D. A. Cox, Primes of the form x
2
+ ny
2
: Fermat, Class Field Theory, and Complex
Multiplication, John Wiley & Sons, Inc. (1989).
[C-P] P.E. Conner, R. Perlis, A survey of trace forms of algebraic number elds, World
Scientic, Singapore, 1984.
[C-S] J.H. Conway, N.J.A. Sloane, Sphere packings, lattices and groups, Third edition.,
Springer-Verlag New York, Inc. (1999).
[C-Y] P.E. Conner, N. Yui, The additive characters of the Witt ring of an algebraic number
eld, Can. J. Math., Vol. XL, No. 3 (1988), 546-588.
[Eich] M. Eichler, Quadratische Formen und Orthogonal Gruppen, Springer-Verlag, 1952
[15, 16].
[E-H] A.G. Ernst, J.S. Hsia, Spinor norms of local integral rotations II, Pacic Journal of
Mathematics, Vol. 61,No.1,1975.
[EMP] B. Erez, J. Morales, R. Perlis Sur le Genre de la form trace, Seminaire de Theorie
des Nombres de Bordeaux. 19871988 (Talence, 19871988), Exp.No. 18, 15 pp., Univ.
Bordeaux I, Talence,
[Ep] M. Epkenhans, On trace forms of algebraic number elds., Arch. Math. (Basel) 60
(1993), no. 6, 527529
[Ga] V.P. Gallagher, Local trace forms., Linear and Multilinear Alg, 7 (1979), 167-174.
[Jo] J. Jones, Number Fields Database, http://hobbes.la.asu.edu/NFDB/
[KMRT] M.A. Knus, A. Merkurjev, M. Rosr, J.P. Tignol, The Book of Involutions., Collo-
quium Publications, American Mathematical Society, Vol 44 (1998).
[Kr] M. Kr uskemper, Algebraic number eld extensions with prescribed trace form., Journal
of Number Theory, 40 (1992), no. 1, 120124.
[Man] G. Mantilla-Soler, Integral trace forms associated to cubic extensions., Algebra Num-
ber Theory, 4-6 (2010), 681-699.
[Man2] G. Mantilla-Soler, A space of weight one modular forms attached to totally real cubic
number elds., preprint (2012)
[Man3] G. Mantilla-Soler, A note on the genus of the integral trace., Authors webpage.
[Mau] D. Maurer, The Trace-Form of an algebraic number eld., Journal of Number Theory,
5 (1973), 379-384.
[MiRe] J. Minac, Z. Reischtein, Trace forms of Galois extensions in the presence of a fourth
root of unity., Int. Math. Res. Not. 2004, no. 8, 389410.
13
[O] O.T.Omeara. Introduction to quadratic forms, Die Grundlehren der mathematischen
Wissenschaften, Vol. 117. Springer, Berlin; Academic Press, New York; 1963. xi+342
pp
[Pi] E.J. Pickett Explicit construction of self-dual integral normal bases for the square-root
of the inverse dierent., J. Number Theory 129 (2009), 17731785.
[S] J.P. Serre, L Invariant de Witt de la forme Tr(x
2
), Comm. Math. Helv. 27 (1984),
651.
[S2] J.P. Serre, Local elds, Graduate Texts in Mathematics, 67. Springer-Verlag, New York-
Berlin, 1979. viii+241 pp.
[Ta] O. Taussky, The discriminant matrix of a number eld, J. London. Math. Soc. 43
(1968), 152-154.
[Wat] G.L. Watson, Integral quadratic forms, Cambridge Tracts in Mathematics and Math-
ematical Physics, No. 51 Cambridge at the University Press, (1970).
Guillermo Mantilla-Soler

Ecole Polytechnique Federale de Lausanne,

EPFL-FSB-MATHGEOM-CSAG,
Station 8, 1015 Lausanne,
Switzerland.
guillermo.mantilla@ep.ch
14