AN ELEMENTARY PROOF OF A THEOREM ABOUT

THE REPRESENTATION OF PRIMES BY

QUADRATIC FORMS
W. E. BRIGGS

1. Introduction. The theorem that every properly primitive binary quadratic

form is capable of representing infinitely many prime numbers was first proved
completely by H. Weber (5). The purpose of this paper is to give an elementary
proof of the case where the form is ax 2 + 2bxy + cy2, with a > 0, (a, 26, c) = 1,
and D = b2 — ac not a square. The cases where the form is ax2 + bxy + cy2
with 6 odd, and the case where the form is ax2.+ 2bxy + cy2 with D a square,
can be settled very simply once the first case is taken care of, and this is done
in a page and a half in the Weber paper. The proof follows the methods used
by Atle Selberg in his elementary proof of Dirichlet's theorem about primes
in an arithmetic progression (3).

2. Representation of numbers by quadratic forms. Some basic facts con-

cerning the representation of numbers by binary quadratic forms are now given.
The h classes of properly primitive quadratic forms of determinant D can be
taken as 0i, 02, . . • 0h, and in the case of negative determinants only the classes
which contain positive definite forms are considered. These classes considered
as elements form an Abelian group of order h under Gauss's law of composition.
A positive number m, relatively prime to 2D, is primitively representable by
forms of determinant D, if and only if D is a quadratic residue of m, and to
each root n of the congruence x2 = D (mod m) correspond one or more repre-
sentations of m by each form of the class to which mx2+2nxy+[(n2 — D)/m]y2
belongs. If p is a prime which does not divide 2D and of which D is a quadratic
residue, then to each of the two roots of x2 = D (mod p) correspond one or
more representations of p by the forms of one or two classes. These two classes
are conjugate and can be indicated as 0 and d~l. If the classes are identical, then
6 is called ambiguous. The number of representations of such an m by each
form of the class to which mx2 + 2nxy + \{n2 — D)/m]y2 belongs is equal to
the number of integral solutions of t2 — Du2 = 1. In order that this number
not be infinite when D > 0, it is required that the x and y in ax2 + 2bxy + cy2
satisfy the conditions

(2.1) y > 0, x > ^-ff^y = vy, D> 0,

Received April 6, 1953; in revised form January 5, 1954. This paper is a portion of a thesis
prepared under the direction of Professor B. W. Jones and submitted for a Ph.D. degree at
the University of Colorado.
353
354 W. E. BRIGGS

where T, U is the fundamental solution of the Pell equation. In this way there
are now w representations of m by each form of the class where
w = 1, D > 0,
(2.2) w = 4, D =-1,
w = 2, £> < - 1 .
Defining

(2.3) S,{x) = E ^ and &(*) = ^ ,

where the summation is extended over primes represented by \p =

ax2 + 2bxy + cy2, the proof will be completed by showing that Q^{x) is
greater than a positive constant for x > x0 for any \p.

3. Several preliminary lemmas.

LEMMA 1. The number of lattice points N(T), subject to restriction (2.1) if
D > 0, within ax2 + 2bxy + cy2 = T which make the form prime to 2D is

where

P = -^jY)log (r +
^V"^), D > o.
Proof. Each \2D\ by \2D\ square built up over the plane from the origin
contains \2D\(j>(2D) lattice points which make the form prime to \2D\ (1,
pp. 235-6). Let the number of these squares lying entirely within the appropri-
ate area be N'(T). Then \4zD2N'(T) — Area| is less than the area of the
squares which are cut by the perimeter, which is of the order of the length of
the perimeter. The perimeter is 0(^/T) and the area is ftT, and the result
follows.
LEMMA 2. For any D not a square

log ft _
= ii l o g x + 0 ( l ) .
V<X P
(p,2Z» = l

This was proved by Selberg in his elementary proof of the prime-number

theorem for arithmetic progressions (4).
LEMMA 3.

J2 ~ = w log*+ 0(1).
P<x P
(p,2Z»=l
 QUADRATIC FORMS REPRESENTING PRIMES 355

Here X / means a summation over all representations, subject to restriction

(2.1) if D > 0, by a representative system of one form from each properly
primitive class of determinant D, and where w has the meaning (2.2).
Proof. This follows from Lemma 2 since each prime p with (p, 2D) = 1 and
(D\p) = 1 has 2w representations in all by the classes 6P and #p_1. If 6P is an
ambiguous class, then there are 2w representations by the single class 6P = dp'1.
LEMMA 4.
^ log r+ V 1 log r+1 x
2r + 1+ OQofx).
(P.2D)=1
(D\p) = l

Proof. This follows from Lemma 2 by partial summation.

LEMMA 5.
' logr+1ft log r+1 x -„
Z
P<X P r -\- 1
r N

(P,2D)=1

Proof. This follows from Lemma 4 and the proof of Lemma 3.

4. Proof of the theorem. Next consider

n=4>
(n,2D)=l

where con = contZ = £A d , X<z = \z,z = n(d) log2 -;, ^ is a properly primitive
a
d\n m

form of determinant D, and n = \p means n is represented by \f/ and that each

representation is counted with the usual restriction (2.1). But
log2x n = 1,
log £ log (x2/p), n = pa, a > 1,
a
2 log £ log g, n = p gf, ap > 1,
0 for all other n>
where p and q denote prime numbers.
Therefore, where p and q do not divide 2D,
H un = X) log p log (x2/p) + J2 logp log q + <9(log2x)
n<x pa^z paqP<x
(n,2D)=l a/S>l

= 12 (2 log x log £ - log2/*) + £ log £ log (x2/p)

P<x pa<x
V^i> P<*=4,
a>l

+ 12 logp log q+ ^ logp log q + 0(log 2 x).

PQ<x p<*q%<x
VQ=V> p<*qP=ij/
aj9>l
356 W. E. BRIGGS

From this follows, using the arguments of Selberg (3) with appropriate
changes in the indices of summation, that

(4.1) E ^= E Wp+ Z logplogq + 0(x).

tt<£ pKx pq<^x
n=4/ P=*l/ PQ=}I/
in, 2D)-
(n,2D)=l (p,2D) = l (Pff,2D)=l

On the other hand,

n<lx dKx d \n
n=4 (d,2D)=l n<x
(n,2D)=l n=ty
(n,2D)=l

The second sum on the right is the number of multiples of d which are rela-
tively prime to 2D, less than or equal to x, and represented by \p. This is RS
where R is the number of representations of d by the forms of determinant D>
and S is the number of numbers relatively prime to 2D and less than or equal
to x/d which are represented by 0<fty if 6d represents d. From (2, p. 144)
R = wJ2 (D\d)
ô\d
and from Lemma 1,
»(|2P|)
s-*œ?i>î+oivm.
since ft does not depend on any particular class but only on D. Therefore

RS = { W £ {D\6)}{p' | + 0( V " ^ ) } . ff = --^-^ P,

and

Z «- = «tf'* E ^ £ Pl«) + V^ o( £ -%Lz PI*)) •

TC<Z d<z ^ old \ d<x - \ / /J &\d '
n=$ (d,2D)=l (d,2D)=l V ^
(n.2D)=l
Next the error term is estimated. Let

where l?d is the number of representations of d by classes of forms of determin-

ant D. By Lemma 1, N(T) is independent of the class and since Rd is
N(d) — N(d — 1) summed over the h classes, it follows that

\E\ < h2^ —— ^ log - + O(log x).

A simple calculation shows that E = 0{s/~x), and therefore

(4.2) E <*. = w/5'* E T £ (^l 5 ) + ° ( x ) '

TO<X d<x & ô\d
n=rp (d,2D)=l
(n,2D)=l
 QUADRATIC FORMS REPRESENTING PRIMES 357

where ft' and w depend only on D, so that (4.2) holds for a \p in any properly
primitive class.
Comparing (4.1) and (4.2) and summing over the h properly primitive
classes, there results
(4.3)
X log2p + Z) log ft log q = T\ 2 logV + S log P log q [ + 0(tf) .
P=4> pq=ty (p,2£>)=l (pq,2D)=l
(p,2Z>)=l (p f f .2JD)=l ^

By partial summation one gets

y* i°g2£ i y i°gftlog g _ 1 j y*' i°gV , y' log ^ log g 1

P<X P Pq<x pq h\ p^x p pQ^x pq
P=4 pq=4 \(p,2D)=l (pq,2D)=l
V
(p,2D)=l (pq,2D) = l
+ O(logx).
Since each pq in the summation on the right above appears éw times, it
follows that
y/ log p log q = 4 w y* log ft y log g
P<?<£ ftg p<x P q<x/p q
(pq,2D)=l (p,2D)=l (q,2D)=l
(D\p)=l (Z»lff)=l

This is easily evaluated by using Lemmas 2 and 4 giving w/2 log2x + 0(log x).
Therefore from this and Lemma 5,

! 5 S
(4.4) Z ^ + Z ^ = flog 2 x + 0(logx).
p<x P pq<x Pq h
p=*j/ pq=ty
(p,2D)=l (pq,2D)=l
By partial summation from (4.3) results

(p,2Z>)=l (pff,2i))=l

>H z><* ft pfl<z ftg

^(p,2Z»=l (p<z,2Z»=l J
y / l o g ft l o g g j = y' log'ft log g + y' log ft lQg2g
pq<x Pq pq<x Pq Pq<x Pq
(pq,2D) = l (pq,2D)*=l (p<z,2Z>)»=l

But each of the two symmetric terms on the right above can be written as

4W£ ^ Z ista,
p<z ft <?<£//> q
(p,2D)=l (<7,2Z»=1
(2>lp)=l (Z>l f f )=l

and by Lemma 4 this equals \w log3 x + 0(log 2 x). Using this and Lemma 5
results in
358 W. E. BRIGGS

3 .

(4-5) Z ^ +Z ^ ^ l o g ^ = | l o g 3 x + 0(log 2 x).

p<x ft pq<x ftg Oft
p=^ pq=t
(p,2D)=l (pff,2D)=l
Next,
y log p log2g = y logftj y log2g y log2g]
P<Z<Z ftg p<x P 1 <z<z/p g q<x_/p g
pq=ip (p,2D) = l q=x(,p q=tp-x
(pq,2D)=l (Dlp)=l V( ff ,2Z>)=l (q,2D)=l

where p is represented by \J/P and ^ p - 1 and \[/p{pp = ^ ~ ^ ~ v

1
v
1
= \p. The above
expression is equal to
log ft
P<x P
? W|- Z ^ ^ + 0(logx)|
(2>,2D)=1 " ft çr<x/p g^
(Dlp) = l qr=\{/p
(qr,2D)=l

by (4.4) (which holds for a ^ from any class), and where r denotes a prime
number. Expanding and simplifying by Lemma 4 gives the last expression
equal to
w
i 3 v^ logftlog g log r * rin 2 x
^rlogx- 22 JL +O(logs).
o/& M K i ftg^
(p«7-,2Z))=l
Therefore from (4.5),
(4.6) S ^ = 2 S !2gJLl2gji2g£ + 0(log2x).
2?<a; ft pqrKx PQX
p=fy pqr=4>
(p,2D) = l (pqr,2D)=l

From (4.4), we have

2
(4.7) Z ^ < ? l o g x + 0(logx),
p<x ft «
P=t
(P,2D)=1
which yields by partial summation

(4.8) Z ^ < C l o g x + 0(loglogx).

P<x ft rl
p=x[/
(p,2D)=l
Next,
lQ
y logftlog g ^ y y gftlQ g g + 2 V lQ
gft v logq
8 x 3
PQ<x ftg p<a;i/ ç<a;i/3 pÇ[ x / <p<x ft ç<a;/p g
PQ=^ PQ=4> (p,2D) = l q=xb
(pfff2D)=l (Pff,27»=l (Dlp) = l ff=^p-t
(ff,2D) = l

Applying (4.8) and Lemma 4, we find

logftlog g ^ v 1 V* logftlog g , 8ai, 2 - /, i v
E n t

>,„ < 2 , L / T - ^ - + QT log x + O(log x log log x .

PQ<X ftg zKW 3 K Ï 1 / 3 ftg vn
PQ=^ PQ.=4>
(pq,2D)=l (pq,2D)=l
 QUADRATIC FORMS REPRESENTING PRIMES 359

Therefore from (4.4),

ZHÈ>fhl°ZX- £ Z 3
! 9 £
1 3
^ + OClog*logIog*);
P<X P vn p<W K 1 / PQ.
p=\p pq=4>
(p,2D)=l (pq,2D) = l
that is, for x > x0,

p<2; £ 10A p,«<z»/« P q.

p=i> p=\f/p
(p,2Z))=l q=^a
(p,2Z» = l
(q,2D)=l

where the latter sum is taken over primes p and g with \l/p\[/q = yp. Recalling
(2.3), this can be written as
logx S*(x) > 7^rlog 2 x — ^ Se(x*) 6V(x*),

where the sum is taken over all pairs of classes 0, 6' such that \j/ belongs to the
class 66'.
Division of both sides by log2 x — 9 log2 x* yields

(4.9) Qt(x) >~-lY, Q,(**) Qf{x>), x > xo.

By (4.6)

log 3 .&(x)>2 E *X*1%*^ + OQofx),

pqr<xi/> Pqr
pqr=^
(pqr,2D) = l

&(X) >
27 M ?L, I^T*"175 J?/. "f~\\\^x:m JJ,. ~T7
(p.2D) = l J \> (q,2D)=l

X JÎ
1 log x
T/3 A^
r <^/ 3
r=0"
r 4+°(s^
(r,2D)=l
or
(4.10) Qt(x)>~ Z &(**) ^(**)&»(**)+ O ( ; T M .
Then (4.8) gives

At this point the characters of classes of forms of determinant D are intro-

duced where the character x of a class 6, % (0), is obtained as an Abelian group
character from the group which the classes form under composition. In general
they are divided into three categories: The principal character xo, with
Xo(0) = 1 for all 0, real non-principal characters, with x(0) = ± 1 for all 0
360 W. E. BRIGGS

and which exist if and only if h is even since each character is an h-t\i root of
unity, and the non-real characters. These characters are needed only when h(D)
is even.

LEMMA 6. If h is even and % is any real non-principal character, then

L
Z °-^ = flog,+ 0(1).
(P,2D) = 1

Here the summation is extended over all primes which are representable by
forms of classes of determinant D and for which classes x(0) = 1-
Proof. By considering all possible products of Gauss's generic characters,
one gets that for any real non-principal character, there exists a factor D\ of D
such that x(0) = (Di\m)j where m is any number prime to 2D which is repre-
sentable by forms of the class 0 (5, pp. 311-312). Therefore x(0p) = (Di\p)
and since if p is represented by a form of determinant D, (D\p) = 1, the sum
takes the form

P<x P
(p,2D)=l

However,
logp
V -£-£ - V + V
. P<x P p<x P p<x
(DDi lp) = l (Di l p ) = - l (D l p ) = l

since each desired prime appears twice in the sums of the right member. Since
by Lemma 2

E i ^ = Z ^ - E ^ - i i o g « + o<D,
p<x f? 2?<x f? P<X P
(DI lp)=-l (Z>! \p)=l
it follows, again by Lemma 2, that
W = i ( i - i + i ) l o g i c + 0(1) = | l o g * + 0 ( l ) .
LEMMA 7. Suppose h(D) is even and there is a set of different classes of properly
primitive forms of determinant D, 6\, 02, . . . , 6k, and that k > \h, and that for
each real character x for forms of determinant D, there is a 6 in the set with
x(0) = 1. Let xp be a properly primitive form of determinant D, and suppose that
there is a d and a 0', not necessarily different, belonging to the set, such that \j/
belongs to the class 6df. Then there is a triple of classes belonging to the set, 6, 6', 6",
such that 6 9f 6" = ip under composition.
Proof. The proof follows from the proof of Selberg's lemma (3, Lemma 2)
by replacing primitive residue classes by properly primitive classes of quad-
ratic forms.
 QUADRATIC FORMS REPRESENTING PRIMES 361

LEMMA 8. If h(D) is odd, and there is a set of different classes of properly

primitive forms of determinant D, 0i, 02, . . . , 0k, with k > h/2, and if ad and a 0',
not necessarily different, such that 00' = \p, belong to this set, then there exists a
triple of classes belonging to the set, 0, 0', 0", such that 0 0' 0" = $.

The proof is the same as the proof of the preceding lemma for k > \h.
Now it can be shown that

Assume that for some large x

1
(4.12) &<*) < m h •
By Lemma 3,

foT^ P** 1 + ° M = - + ° ( s b ) •
Ç <?•<**> =log
But
^ , iv . 2w . ~ flog log x \
log:
for all 0 by (4.11).
Therefore there are at least the greatest integer in (h + l ) / 2 classes 0 with

From (4.9),
>
Q*^ TÔh ~ 9 ,JÇ Çe(xi) ^ ^ X > Xo
'
and (4.12),
Ç (x)<
' ïéï-
Therefore

£j^)Q*&)>*(m-m)>à'
Therefore there exists at least one pair of classes 0, 0' with 00' = \f/ such that

g,(**) &.(**) > J^J

or
1. b
CM**) > „..,.../.,,». > - ^ — r — ; y <4-u>

-^ i o n 7 2 » # ^ ^0>
30M + 15A « 130A-
362 W. E. BRIGGS

and likewise

(M**) > Y ^ 2 , X > Xo.

5. Completion of proof for h(D) even. By Lemma 6

X) -~— = ~ log x + 0(1) > - log x, x > x0,

P=0P

or

X X —T~ > n l°g x> x

> XQ>
e v<x p y
x(d)=i p=e
or
E r - i / , E -J^ = Z CM**) > £, x>*„
e log x p<o;i/3 p e y
x(0)=i ?=0 x(0)=i
Therefore there exists at least one 0 with
Q,( > >
**> à m?
and with x(^) = 1 for each real non-principal character x-
Thus there is a set of different classes 0i, 02, . . . 0k, with & > A/2, such that
for i = 1 , 2 , . . . &,

and such that for each real character x there is a 0* with xifii) — 1» a n d finally
such that there exist classes 0, 0', with 00' = \f/. Therefore by Lemma 7, there
exist classes 0, 0', 0", belonging to the set with 00'0" = ^. Then by (4.10)

CM*) >| <M*J) CM*4) e.»(«») - o ( ^ ) > ^ ^

for x > Xo, which completes the proof for h(D) even.

6. Completion of proof for h(D) odd. Again there is a set of different classes
0i, 02, . . . 0*, with A > A/2 such that for i = 1, 2, . . . k,
1
(M*1) > 130A2
a n d such t h a t there exist classes 0, 0' in t h e set with 00' = ^. Therefore b y
L e m m a 8 there exists a triple of classes 0, 0', 0", belonging t o t h e set with
00'0" = $. T h e n again b y (4.10)

&(*) > ^ e*(^) &*(**) &-(**) - o ( j ^ - ) > (130)V

for x > Xo which completes the proof of the theorem.

 QUADRATIC FORMS REPRESENTING PRIMES 363

REFERENCES

1. P. G. L. Dirichlet, Vorlesungen iiber Zahlentheorie (Brauschweig, 1879).

2. E. Landau, Vorlesungen iiber Zahlentheorie, vol. 1 (New York, 1950).
3. A. Selberg, An elementary proof of Dirichlet''s theorem about primes in an arithmetic pro-
gression, Annals of Math., 50 (1949), 297-304.
4. , An elementary proof of the prime-number theorem for arithmetic progressions, Can. J.
Math., 2 (1950), 66-78.
5. H. Weber, Beweis des Satzes, dass jede eigentlich primitive quadratische Form unendlich viele
Primzahlen darzustellen fahig ist, Math. Annalen, 20 (1882), 301-329.

University of Colorado