Ternary Quadratic Forms

William C. Jagy

March 19, 2010

William C. Jagy Ternary Quadratic Forms

A “cyclic” coefficient order from Brandt and Intrau
We let the sextuple
�a, b, c, d, e, f �
refer to the quadratic form

T (x, y, z) = ax 2 + by 2 + cz 2 + dyz + ezx + fxy

The Gram matrix for double the form is:

 
2a f e
 f 2b d  .
e d 2c

We define our discriminant ∆ as the absolute value of the

Brandt and Intrau [B I] discriminant, that is half the determinant
of the matrix above, so

∆ = 4abc + def − ad 2 − be2 − cf 2 .

William C. Jagy Ternary Quadratic Forms

Hsia letter 1993
John S. Hsia, in a 1993 letter to Irving Kaplansky, proved that
x 2 + y 2 + 10z 2 represents eligible numbers of the form 3m + 2,
including all 6m + 5. This is also mentioned by Byeong-Kweon
Oh [Oh] in his preprint on proofs of regularity. Here’s my proof:
n is odd and take n = 2x 2 + 2y 2 + 2yz + 3z 2 . As n ≡ 2
(mod 3), either (A) x ≡ ±z (mod 3), or (B) x �= 0 (mod 3) and
y ≡ z ≡ 0 (mod 3). In case (B), 2y 2 + 2yz + 3z 2 ≡ 0 (mod 3)
but is not 0, so a lemma very similar to those in the dissertation
of Jones [Jon] says that this can be rewritten with y , z �= 0
(mod 3). So we are back in case (A), and one of
� � � � � �
2x − 3y + z 2 2x + 3y + 4z 2 x −z 2
n= + + 10 ,
3 3 3
� � � � � �
2x − 3y − 4z 2 2x + 3y − z 2 x +z 2
n= + + 10
3 3 3
is integral.
William C. Jagy Ternary Quadratic Forms
Small Identities
These identities are in marked contrast to Schiemann’s result
and quite surprising to me. They are not apparently related to
Siegel’s averaging theorem, rarely use all forms of a genus, and
apply to all numbers. The explanation is so very simple...
Given a prime p, the two by two matrices of determinant p are
(up to multiplication on the left by SL2 Z, see Buell [Bu, pages
111-112], Milne [Mil, page 67])
� � � � � �
p 0 1 a 1 0
, , , with 1 ≤ a ≤ p − 1.
0 1 0 p 0 p

With the adjoints of these in mind, consider that a fixed pair of

integers (y , z) satisfies at least one of these p + 1 conditions:

y ≡ 0 (mod p) or y + az ≡ 0 (mod p) or z ≡ 0 (mod p),

and if any two of these are true then all p + 1 are true.
William C. Jagy Ternary Quadratic Forms
Given any positive ternary form F (x, y , z), construct
G0 (u, v , w) = F (u, pv , w),
Ga (u, v , w) = F (u, pv − aw, w) with 1 ≤ a ≤ p − 1,
Gp (u, v , w) = F (u, v , pw),
H(u, v , w) = F (u, pv , pw).
We account for the representations of any target number n by
these forms. If n is not represented by F it is not represented
by any of the Gλ or H either. When F represents n, take the
representation triples (x, y , z) one at a time. If only one of the
conditions
y ≡ 0 (mod p) or y + az ≡ 0 (mod p) or z ≡ 0 (mod p),
applies, then we increase RF and the appropriate RGλ by
exactly 1, but do not increase RH , so that the value so far of
p
�
RF − RGλ + p RH
λ=0

William C. Jagy Ternary Quadratic Forms

does not change.
If two of the conditions apply then y ≡ z ≡ 0 (mod p), all the
forms F , Gλ , H get an increase of 1 in the accounting, and once
again
�p
RF − RGλ + p RH
λ=0

is unchanged. So when we are done with all the (finite) set of

representations of n by F , we have
p
�
RF (n) − RGλ (n) + p RH (n) = 0.
λ=0

William C. Jagy Ternary Quadratic Forms

Our record for shortest identity happened when all the p + 1
forms Gλ turned out to be equivalent. Below, for any fixed
integer t ≥ 1 we get ∆F = (3t − 1)2 , and F as written is already
Schiemann reduced when t ≥ 3.

F = �3, t, t, 1 − t, 2, 2�

G = �3, t, 3t − 1, 0, 0, 2�
H = �3, 4t − 1, 4t − 1, 2 − 4t, 2, 2�
Here the prime is p = 2, and

RF (n) − 3 RG (n) + 2 RH (n) = 0.

The multiplier 3 reflects the fact that this G is really

G = G0 = G1 = G2 .

William C. Jagy Ternary Quadratic Forms

If we do the analogous thing with SL3 Z and involve all three
variables, we do get H = p2 F as desired but some of the
p2 + p + 1 forms Gλ will also be imprimitive with a coefficient
GCD of p. These discriminants will be ∆F , p4 ∆F , p6 ∆F . As
an example, with p = 3 and the sum of three squares, I got

F = �1, 1, 1, 0, 0, 0�
G1 = �1, 9, 9, 0, 0, 0�
G2 = �2, 5, 9, 0, 0, 2�
G3 = �3, 6, 6, 6, 0, 0�
H = �9, 9, 9, 0, 0, 0�
The coefficient of RH will be p2 + p, and we find

RF (n) − 3 RG1 (n) − 6 RG2 (n) − 4 RG3 (n) + 12 RH (n) = 0.

We could do this with quaternary F and get primitive forms

again.
William C. Jagy Ternary Quadratic Forms
Glossary of the S-genus

This is from [B J]. Start with an odd positive squarefree

number S = p1 p2 · · · pr with the pj distinct primes. It is known
that there are exactly 2r genera of binary quadratic forms of the
same discriminant as x 2 + 2Sy 2 , this discriminant being −8S
with most authors. From each genus, take a form
g(x, y) = ax 2 + bxy + cy 2 and create the ternary form
h(x, y, z) = (ax 2 + bxy + cy 2 ) + 2Sz 2 . Now expand each such
h to a genus of ternary forms. There are again 2r of these, call
them TGi with 1 ≤ i ≤ 2r . For each one, throw in a convenient
factor of 16 and define
� 16
Mi := 16 Mass(TGi ) = .
|Aut(f )|
f ∈TGi

where the sum is over the equivalence classes of forms in the

genus, hence finite.

William C. Jagy Ternary Quadratic Forms

The collection of genera exhibits a mass of its own,
� 16
M(S-genus) := = M1 + . . . + M2r = S.
|Aut(f )|
f ∈S -genus

As usual, consider a weighted average of representations of

some number by a genus TGi as
� 16 Rf (n)
Wi (n) := ,
|Aut(f )|
f ∈TGi

and while we are at it define

H(x, y, z) = x 2 + y 2 + z 2 .

The next dramatic bit is that, for any integer n ≡ 1, 2 (mod 4),
2r
�
3 Wi (n) = RH (n).
i=1

William C. Jagy Ternary Quadratic Forms

Take a number ni with gcd(ni , 2S) = 1 and with ni represented
by some form in TGi . Define ε(i, 1) = 1, for any p|S define
ε(i, p) = (−ni |p), and for that matter define ε(i, w) = (−ni |w)
for any w|S. I should point out that ε(i, p) = 1 if the forms in TGi
are isotropic at p, and ε(i, p) = −1 if the forms in TGi are
anisotropic at p. This next one is Alex’s favorite:
r
� pj + ε(i, pj )
Mi = ,
2
j=1

For n ≡ 1, 2 (mod 4) and with w|S and w|n,

2r
�
3 ε(i, w) Wi (n) = w RH (n/w 2 ),
i=1

where both sides are zero unless w 2 |n.

William C. Jagy Ternary Quadratic Forms

Owing to an abundance of orthogonality results, we get the pair
�
ε(i, w)(S/w) = 2r Mi ,
w|S

2 r
�
ε(i, w)Mi = (S/w).
i=1

Our final inversion requires caution, it is necessary to have S|n

to justify naive summation over all w|S to use orthogonality, but
if S|n and n ≡ 1, 2 (mod 4) we do get
�
w · ε(k , w) · RH (n/w 2 ) = 3 · 2r · Wk (n).
w|S

So, if n ≡ 1, 2 (mod 4) and p � n for all p|S, then only w = 1

gives a nonzero contribution and all the Wk (n) are equal.

William C. Jagy Ternary Quadratic Forms

Forms representing multiples of forms

Let M stand for the lattice �a2 , b2 , c2 , d2 , e2 , f2 � and let B stand

for the relevant Gram matrix. Then let L stand for the lattice
�a1 , b1 , c1 , d1 , e1 , f1 � and let A stand for the relevant Gram
matrix. Now, in terminology taught me by Wai Kiu (Billy) Chan,
we say that M represents L(n) if there is an integral matrix P
with transpose P � that satisfies

P � BP = nA.

And �a2 , b2 , c2 , d2 , e2 , f2 � represents n times every integer that

is represented by �a1 , b1 , c1 , d1 , e1 , f1 �.

William C. Jagy Ternary Quadratic Forms

I ran a computer for three people-years looking for all the spinor
regular forms, based on a recipe in Chan and Earnest [C E]. I
probably found them all, list on Sloane’s website.
I noticed when the smallest spinor exceptional integer was
especially large. This happened with the genus of
�1, 1, 16n, 0, 0, 0� where n = u 2 + v 2 in integers but n itself was
squarefree. I had found a smallest spinor exception as large as
∆/64.
J. S. Hsia confirmed for me that the genus in question had
exactly two spinor genera and that n was a spinor exception.
He mentioned that the methods were in [E H H]. He also
pointed me to his proof that, if there are any spinor exceptions,
there is one that divides 2∆, this is Theorem 2 in [Hsia].

William C. Jagy Ternary Quadratic Forms

I was unable to get “verbatim” to work in Beamer. Oh, well, for
n = 5, the good (regular) spinor genus consists of two classes,

�1, 1, 80, 0, 0, 0�

and
�4, 5, 5, 0, 0, 4�,
both of which represent 5.
The irregular spinor genus also consists of two classes,

� 5m2
�1, 9, 9, 2, 0, 0� =

and
� 5m2 .
�2, 2, 21, 2, 2, 0� =

William C. Jagy Ternary Quadratic Forms

The surprise for Prof. Hsia was that I found that every class in
the good spinor genus containing �1, 1, 16n, 0, 0, 0� also
represented n, where n = u 2 + v 2 but is squarefree, and we will
stick with n odd as well.
First, the spinor exceptions are nm2 , where all prime factors of
m, if any, are 1 (mod 4).
Everything in the good spinor genus represents G(n) where
G = �1, 1, 16, 0, 0, 0�, while everything in the bad spinor genus
represents B (n) where B = �2, 2, 5, 2, 2, 0�. The proof uses
results of Duke and Schulze-Pillot [D S-P] and one,
unpublished, by Billy Chan.
So we have the cheapest possible test for splitting spinor
genera, a form in the genus is in the same spinor genus as
�1, 1, 16n, 0, 0, 0� if and only if it represents the single number n.

William C. Jagy Ternary Quadratic Forms

Switching to �1, 16n, 16n, 0, 0, 0� with same n’s.
This time, for n = 5, the good spinor genus consists of

�1, 80, 80, 0, 0, 0�

and
� q2,
�16, 20, 25, 20, 0, 0� =
� q2
both of which represent 25. However �16, 20, 25, 20, 0, 0� =
for prime q ≡ 3 (mod 4).
The irregular spinor genus consists of

� m2
�4, 41, 41, 2, 4, 4� =

and
� m2 .
�9, 9, 80, 0, 0, 2� =

William C. Jagy Ternary Quadratic Forms

With n odd and squarefree but n = u 2 + v 2 , Watson’s
transformation [Wat] with parameter 64n shows that the genus
of �1, 16n, 16n, 0, 0, 0� also splits into exactly two spinor genera.
It is elementary to show that the only form in the genus of
�1, 16n, 16n, 0, 0, 0� that is allowed to represent 1 is the form
displayed. So 1 is a spinor exception, as are Jones-Pall m2 .
Now, no form in the spinor genus of �1, 16n, 16n, 0, 0, 0� can
primitively represent any q 2 with prime q ≡ 3 (mod 4). So no
form in that spinor genus, other than �1, 16n, 16n, 0, 0, 0� itself,
can represent any such q 2 at all!
But every form in this spinor genus represents n2 . Everything in
2
the good spinor genus represents G(n ) where
G = �1, 16, 16, 0, 0, 0�, while everything in the bad spinor genus
2
represents B (n ) where B = �4, 9, 9, 2, 4, 4�. Finally, at least one
form in the bad spinor genus represents each q 2 , primitively of
course, while at least one form in the good spinor genus
primitively represents each p2 for prime p ≡ 1 (mod 4).

William C. Jagy Ternary Quadratic Forms

The Involutions

I noticed that, with odd n ≤ 157, for either the genus of

�1, 1, 16n, 0, 0, 0� or of �1, 16n, 16n, 0, 0, 0�, the two spinor
genera have the same number of classes. Meanwhile, take
some L in either genus, I found that there is exactly one M in
the same genus such that L represents M (4) and M represents
L(4) but L �= M, while L and M are never in the same spinor
genus. So we have an involution that interchanges the spinor
genera, checked for n ≤ 157. I think this is wonderful.
A very similar thing happens for the �1, 4, 225n, 0, 0, 1� and
�2, 2, 225n, 0, 0, 1� examples, where if n is not divisible by 3 or
by 5, we get one involution where L represents M (9) , then a
second involution where L represents M (25) . These were
checked for n ≤ 141.

William C. Jagy Ternary Quadratic Forms

Whimsical question

We know from Ono and Soundararajan [O S] that

x 2 + y 2 + 10z 2 represents all non-squarefree (“squareful?”)
numbers permitted to its genus. Let us denote
L = �1, 1, 10, 0, 0, 0� and M = �2, 2, 3, 2, 0, 0�. Is it possible that
2
L represents M (p ) for all primes p? It does for p ≤ 37. I am not
claiming any significance for this, it is just what occurred to me
at the time.

William C. Jagy Ternary Quadratic Forms

The references are listed after this frame. A few of them, along
with some personal communications I mentioned, are available
at
http://zakuski.math.utsa.edu/∼kap/forms.html

William C. Jagy Ternary Quadratic Forms

[B J] A. Berkovich and W. C. Jagy.
Ternary quadratic forms, modular equations and certain
positivity conjectures.
In K. Alladi, J. Klauder, and C. R. Rao, editors, The legacy
of Alladi Ramakrishnan in the mathematical sciences.
Springer, 2010, to appear.
[B I] H. Brandt and O. Intrau.
Tabelle reduzierten positiver ternärer quadratischer
Formen.
Abh. der Sächsischen Akad. der Wissenschaften zu
Leipzig, Math.-Naturw., 45(4), 1958.
This article is said (Zentralblatt) to have taken up the entire
issue, total 261 pages, and been available for sale
separately, almost as though it were a book. I’ve seen
excerpts only.

William C. Jagy Ternary Quadratic Forms

[Bu] D. A. Buell.
Binary Quadratic Forms: Classical Theory and Modern
Computations.
Springer-Verlag, 1989.
[C E] W. K. Chan and A. G. Earnest.
Discriminant bounds for spinor regular ternary quadratic
lattices.
Journal of the London Mathematical Society, 69:545–561,
2004.
[D S-P] W. Duke and R. Schulze-Pillot.
Representations of integers by positive ternary quadratic
forms and equidistribution of lattice points on ellipsoids.
Inventiones Mathematicae, 99:49–57, 1990.

William C. Jagy Ternary Quadratic Forms

[E H H] A. G. Earnest, J. S. Hsia, and D. C. Hung.
Primitive representations by spinor genera of ternary
quadratic forms.
Journal of the London Mathematical Society, 50:222–230,
1994.
[Ha 03] J. Hanke.
On a local-global principle for integral quadratic forms.
Unpublished preprint, 2003.
[Ha 04] J. Hanke.
Some recent results about (ternary) quadratic forms.
In : Hershy Kisilevsky and Eyal Z. Goren, editors, Number
Theory, pages 147–164. American Mathematical Society
and Centre de Recherches Mathématiques, 2004.
Series: CRM Proceedings & Lecture Notes, volume 36.

William C. Jagy Ternary Quadratic Forms

[Hsia] J. S. Hsia.
Regular positive ternary quadratic forms.
Mathematika, 28:231–238, 1981.
[Jon] B. W. Jones.
Representation by positive ternary quadratic forms.
PhD thesis, University of Chicago, 1928.
[J P] B. W. Jones and G. Pall.
Regular and semi-regular positive ternary quadratic forms.
Acta Mathematica, 70:165–191, 1939.

William C. Jagy Ternary Quadratic Forms

[Mil] James S. Milne.
Modular functions and modular forms (v1.20), 2009.
Available at www.jmilne.org/math/.
[Oh] B.-K. Oh.
Regular positive ternary quadratic forms.
Preprint, 2008.
[O S] K. Ono and K. Soundararajan.
Ramanujan’s ternary quadratic form.
Inventiones Mathematicae, 130:415–454, 1997.
[Schiem] A. Schiemann.
Ternary positive definite quadratic forms are determined by
their theta series.
Mathematische Annalen, 308:507–517, 1997.

William C. Jagy Ternary Quadratic Forms

[S-P 00] R. Schulze-Pillot.
Exceptional integers for genera of integral ternary positive
definite quadratic forms.
Duke Mathematical Journal, 102:351–357, 2000.
[S-P 04] R. Schulze-Pillot.
Representation by integral quadratic forms - a survey.
In Ricardo Baeza, John S. Hsia, Bill Jacob, and Alexander
Prestel, editors, Algebraic and Arithmetic Theory of
Quadratic Forms, pages 303–321. American Mathematical
Society, 2004.
Series: Contemporary Mathematics, volume 344.
[Tim] V. N. Timofeev.
On positive quadratic forms representing the same
numbers. (Russian).
Uspekhi Matematicheskikh Nauk, 18:191–193, 1963.

William C. Jagy Ternary Quadratic Forms

[Wat] G. L. Watson.
Transformations of a quadratic form which do not increase
the class-number.
Proceedings of the London Mathematical Society,
12:577–587, 1962.

William C. Jagy Ternary Quadratic Forms

Strengthening Schiemann

In 1963 V. N. Timofeev [Tim] published a short piece in Uspekhi

displaying pairs of positive ternary forms that represented the
same numbers, in the same sense that x 2 + xy + y 2 and
x 2 + 3y 2 represent the same numbers.
Alexander Schiemann [Schiem] showed that the theta series
determines the positive ternary form up to equivalence.
In a 1997 email to Schiemann, Irving Kaplansky conjectured
that when two positive ternaries represent the same numbers,
ignoring multiplicity, at least one of these three holds:
(I) both are regular
(II) one is equivalent to �s, t, t, t, 0, 0� and the other to
�s, t, 3t, 0, 0, 0� .
(III) one is equivalent to �t, t, t, s, s, s� , the other to
�t, 2t − s, 2t + s, 0, 2s, 0� .
Timofeev took s even in pattern (III), that is not necessary.

William C. Jagy Ternary Quadratic Forms