HISTORY OF THE THEORY OF NUMBERS
VOLUME II
QUADRATIC AND HIGHER FORMS
BY LEONARD EUGENE DICKSON
Professor of Mathematics in the University of Chicago
WITH A CHAPTER ON THE CLASS NUMBER
By G. H. Cresse
CHELSEA PUBLISHING COMPANY
NEW YORK, N.Y.
1952PREFACE.
The favorable reception accorded to the first two volumes of this history has
encouraged the author to complete promptly the present third volume which is
doubtless the most important one of the series.
By a “form” we mean a homogeneous polynomial such as f=az*+bay+cy?, all
of whose terms are of the same total degree in and y. The arithmetical theory of
forme has an important application to the problem to find all ways of expressing a
given number m in a given form f, i. e., to find all sets of integral solutions z, y of
az*+bay+oy*=m. For this application we do not consider merely the given form f,
but also the infinitude of so-called equivalent forms g which can be derived from f
by applying linear substitutions with integral coefficients of determinant unity. It
is by the consideration of all these forms g that we are able to solve completely the
proposed equation f=m. The theory needed for this purpose is called the arith-
metical theory of forms, which is the subject of the present volume. This theory
is applicable to most of the problems discussed in Volume II. The present methods
have the decided advantage over the special methods described in Volume II in that
they give at once also the solution of each of the infinitely many equations g=m.
By thus treating together whole classes of equivalent equations, the present methods
effect maximum economy of effort.
Enough has now been said to indicate that we are concerned in Volume III mainly
with general theories rather than with special problems and special theorems. The
investigations in question are largely those of leading experts and deal with the most
advanced parts of the theory of numbers. Such a large number of the important
papers are so recent tliat all previous reports and treatises (necessarily all very
incomplete) are entirely out of date.
As in the case of the first two volumes, no effort has been spared to make the list
of references wholly complete. There was now the additional burden of examining
many titles which turned out to relate to the algebraic theory of forms, rather than
the arithmetical theory.
The prefaces to the first two volumes gave a clear account of the leading results on
the subjects treated. This was possible partly on account of the elementary nature
of most of those subjects, and partly because the gist of the investigations could be
embodied in definite theorems expressed without the use of technical terms. But
in the case of Volume III, it is a question not primarily of explicit results, but chiefly
of general methods of attacking whole classes of problems, the methods being often
quite intricate and involving extensive technical terminology. Accordingly it is not
possible to present in this preface a simple summary of the contents of the volume;
perhaps the opening sentences of the preface have provided sufficient orientation in
,_ the subject. However, to each of the longer chapters is prefixed an appropriate
eo introduction and summary.
TUC unMT tinnanwwv PREFACE.
The congruencial theory of forms, presented in the final chapter, is not only of
special intrinsic interest, but has important applications in the theory of congruence
groups and in the classic theory of genera of ordinary arithmetical forms as developed
especially by Poincaré and Minkowski (Ch. XI).
The proof-sheets of each chapter were read critically by at least one specialist in
the subject of the chapter. Mrs. Mayme I. Logsdon of the University of Chicago
read the short Ch. XVII. Miss Olive C. Hazlett of Mount Holyoke College read
Chs. XVIII and XIX. All of the earlier sixteen chapters were read by EB. T. Bell
of the University of Washington, by L. J. Mordell of the University of Manchester,
England, and by A. Speiser of the University of Ziirich, each a well known expert
in the arithmetical theory of forms. Both Bell and Mordell compared many of the
reports with the original papers. Such a comparison was made independently by
the author.
G. H. Cresse devoted five years to the preparation of the report in Ch. VI on the
difficult subject of the number of classes of binary quadratic forms, which involves
many branches of pure mathematics. His historical and critical report was written
asa thesis for the doctorate at Chicago, and will prove indispensable to future
investigators in this difficult subject. Mordell read critically the manuscript of this
long Ch. VI, compared the reports with the original papers, read also the proof-sheets,
and on each occasion made numerous important suggestions. Bell, who read a portion
of the manuscript and most of the proof-sheets, made valuable suggestions resulting
in improvement in the presentation of thé reports on papers which follow the methods
of Herthite and Humbert. A. Ziwet of the University of Michigan translated for
Cresse important portions of papers in Bohemian.
The volume has been much improved as to clearness and accuracy by the generous
aid from Bell and Mordell, whose many investigations in this field made their
cooperation most valuable. In particular, Mordell’s criticisms proved his mastery
of practically every subject in this extensive and intricate branch of mathematics.
Readers are requested to send notice to the author of errata or omissions, which
will be published later as a supplement.TABLE OF CONTENTS,
CHAPTER PAGE
Repucrion anp Equrvatence or Brvary Quapraric Forms, Representation oF
‘Inrecers ...
‘Exerierr Vauues oF 2, y IN 2?-+-Ay?
Comrosrrion oF Brnary Quapratic Forme.
Orpers anv Genera; THER Composition.
Tnnecutan Determinants:
Numer or Cuasses or Brnary Quapratic Forms wire Inrecrat Cogrricrents.
Buvary Quapratic Forms Wxose Coerrictents Are Comprex Intecers or IN-
‘TeGERS OF A Fre
‘VIII. Numser or Crasses or Brvary Quapratic
CozFFICIENTS ...
IX. Ternary Quapratic Forms
X. Quarernary Quapratic Forms
XI. Quapratic Forms mv n VARIABLES.
XII. Brvary Cusic Forms ...
XII. Cusrc Forms in Tare on More Vantastes
‘XIV. Forms or Denes n> 4.
XV. Bovary Hermrrian Forms
XVI. Hermitian Forms 1n n Vartasces anp Taem Consucate:
XVI. Brurear Forms, Marrices, Linear Susstrrvtions .
XXVIII. Representation sy Pouynomtars Moputo p
Analytic Representation of Substitutions,
Integers Modulo p..
Polynomials Representing Only Numbers of Prescribed Nature
XIX. Conorvencta, Txeory or Forms .
Modular Invariants and Covariant
Reduction of Modular Forms to Canonical Types
Formal Modular Invariants and Covariants
ds<2He n
Avutuor Inpex
Supsecr InvexCHAPTER I.
REDUCTION AND EQUIVALENCE OF BINARY QUADRATIC FORMS,
REPRESENTATION OF INTEGERS.
Inrropucrion.
‘There will be given in Chapters II-VIII reports of the literature on the following
topics of the arithmetical theory of binary quadratic forms: Explicit values of z,
y in 2*+Ay*=g, composition, orders and genera, irregular determinants, number
of classes of forms with integral coefficients, forms whose coefficients are complex
integers or integers of a field, and their class number. The present chapter deals
with the remaining, miscellaneous topics, the nature of which will be clear from the
following brief summary.
Euler was the first to publish (in 1761, 1763) proofs of the facts that every prime
of the form 6n+1 can be represented by 2*-+3y?, and every prime 8n-+1 by a*-+2y?.
These and a few similar theorems had been merely stated by Fermat in 1654, In
1778, Lagrange proved many such facts by means of his general theory of reduc-
tion and equivalence of binary quadratic forms. In 1798, Legendre simplified and
extended Lagrange’s methods and tables, being aided largely by the reciprocity law
for quadratic residues (although his proof of it was not quite complete).
In 1801, Gauss introduced many new concepts and extended the theory in various
directions. His work has continued to occupy the central position in the literature,
although many of his methods have since been materially simplified by Dirichlet,
and to a less extent by Arndt and Mertens.
In 1851, Hermite developed his fundamental method of continual reduction.
Closely related to it is the geometrical theory introduced by Smith in 1876 and
applied by him to elliptic modular functions, later simplified by Hurwitz in 1894, by
Rein in 1890, 1896, and by Humbert in 1916, 1917. A like goal was reached by
Dedekind in 1877 and Hurwitz in 1881, both by means of the equivalence of com-
plex numbers.
Selling gave in 1874 important methods of reducing both definite and indefinite
forms. In 1880, Poincaré gave extensions of the methods of representing numbers
a+bVD by points, and forms by lattices. In 1881 and 1905, he constructed trans-
cendental arithmetic invariants. Kronecker in 1883 and Stouff in 1889 studied
reduction and equivalence under special types of substitutions. The investigation
by Markoff in 1879 of the upper limits of the minima of forms was resumed by
Schur and Frobenius in 1913 and Humbert in 1916. There are many further
investigations of a special or miscellaneous character.2 Hisrory or tHE THrory or Numpens. (Cur. I
Dzrinitions anp Notations.
Lagrange considered the general binary quadratic form
(1) ax? + bay + cy?
with integral coefficients. Gauss restricted attention to
(2) az? +-2bry+cy*,
in which the middle coefficient is even, and called b’—ac the determinant of (2).
The form (1) is not only more general than (2), but has the special advantage of
being more suitable for Dedekind’s important correspondence between classes of forms
(1) and certain sets of algebraic numbers determined by a root of af*+bé+c=0.
The discriminant d=b*—4ac of this equation is called the discriminant' of (1),
and plays an important réle in the correspondence mentioned. Except when the
contrary is expressly stated, the notation (a, b, c) will be used for (2), and not
for (1).
If there exist integers , y for which
(3) ax? + bay+cy?=m,
the form (1) is said to represent m. According as x and y are relatively prime or
not, the representation (2, y) is said to be proper or improper.
If to the form f defined by (1) we apply a substitution
(4) waar’ + By’, y=ya'+8y’,
with integral coefficients of determinant A=a3— fy 70, we obtain
(8) pa'e"4$)'r'y!+e'y",
where
(8) a’=aa*+bay+cy?, oe” = a8" + BBS+ 08?
B/= 2008 +b (ad+ By) +2cy8.
The discriminant d’ of 7’ is equal to A’d. Thus d’=d if and only if
latter is a necessary and sufficient condition that the inverse of the substitution (4)
shall have integral coefficients. When this is the case, the forms f and f’ evidently
represent the same numbers and are said to be properly or improperly equivalent,
according as A=+1 or A=—1.
Fermat? stated that he had a solution of a problem he had proposed to Frenicle de
Bessy: To find in how many ways a given number is the sum of the two legs of a
right triangle. In his reply, Frenicle® stated that every prime of the form 8n:1 is
the sum of the two legs of a right triangle, and every number which is the sum of
the legs of a primitive right triangle is of the form 8n-+1. Every product of primes
of the forms 8n-+1 is the difference between the legs of an infinitude of primitive
1 This is in accord ae Dedekind, Kronecker, Weber, and others. But in the Encyclopédie
des sc. math., t. I, v. 3, p. 101, p. 182, 4ac—b? is called the discriminant of (1), and
ac—b? that of (2).
2 Oeuvres, 2, 1894, 221, 226 (§ 11); letters to Mersenne and Frenicle, June 15, 1641.
5 Oeuvres de Fermat, 2, 1894, 231, "985; letters to Fermat, Aug. 2, Sept. 6, 1641.Cuap. I] Binary Quapratic Forms. 3
right triangles. [‘The sum of the legs 2*—y?, 2ay is (v-+y)?—2y? and their differ-
ence is (1—y)?—2y? or 2y*—(z—y)?.]
Fermat* stated that every prime 8n-+1 or 8n+3 is of the form 2*+2y’, that every
prime 3n+1 is of the form 2°+3y?, and conjectured that the product of two primes
ending in 3 or 7, and each having the form 4n+-3, is of the form 2*+5y*. Fermat®
stated that no prime 3n—1 is of the form 2*+3y’.
L, Euler* stated that, if x and y are relatively prime, 227+? has no prime divisor
other than 2 and primes of the forms 8n+1 and 8n+3; if 8n+1 or 8n+3 is a
prime it is expressible in the form 227+ y? in one and but one way. Every odd prime
divisor of 22—y? is of the form 81.
Euler’ stated many special empirical theorems on the representability of primes
by 2?Ny?, where 2 and y are relatively prime (in connection with empirical
theorems on the linear forms of the prime divisors of z*+ Ny’, to be quoted under
the quadratic reciprocity law in Vol. IV). Every prime 8n++1 can be expressed in
the form 2?—2y? in infinitely many ways. Every prime 12n-1 can be expressed
in either of the forms 2*—3y*, 32*—y? in infinitely many ways. Every prime
20n+1 or 20n+9, and the double of any prime 20n+3 or 20n+7 is of the form
zt+5y?, Any prime 14n+k (k=1, 9 or 11) is of the form 2*+7y*. Similar state-
ments are made for 27+ Ny?, N=11, 13, 17, 19. Any prime 24n+1 or 24n+7 is of
the form «?+6y?; any prime 24n-+5 or 24n+11 is of the form 3z*+2y*; analogous
results were stated for N=2-5, 2-7, 3-5, 3-7, 5-7.
Euler* stated that every prime 8nt1 is represented by 24—u? and by u*—20?.
A like result holds for 2 and any square, and hence for any product of primes 8n-+ 1,
2 and a square, since
(2a? —B) (2y?—8*) = (2ay-+f8)?~2(By+08)*, — a*—2y* = 2(a+-y)*—(e+2y)*.
Chr. Goldbach® stated erroneously that any prime 4n+1 is of the form da?+y?
if d is any divisor of n. Euler'® remarked that this is probably true if a, y are
permitted to be rational, but not always for integers. Thus 8954112?+y? in
integers, but
sv=11(3)'+(3)° =u($ y+ @y’.
At least for d=1, 2, 3, a representation by fractions implies one by integers. Euler
and Goldbach** discussed special methods of expressing a prime 4dk-+1 in the form
da? +y?, where x, y are rational.
* Oeuvres, 2, 1894, 313, 403-5; letters to Pascal, Sept., 1654, and to K. Digby, June, 1658
(French transl, of the second letter in Oeuvres, 3, 1896, 315-7). Wallis, ieee 2, 1693,
® Oeuvres, 2, 1894, 431; letter to Carcavi, Aug., 1659.
* Correspondance Mathématique et Physique (ed., P. H. Fuss), St. Pétersbourg, 1, 1843, 146,
149; letter to Goldbach, Aug. 28, 1742, Also, Euler’, Theorems 4, 5, 42. Cf. this His
tory, Vol. II, p. 260, Euler.
Comm. ‘Acad. Pelion, 14, (1744-6) 1751, 151-181; Comm. Arith. Coll, St. Pétersbourg, 1,
1849, 35-49; Opera ers Omnia, qQ),
* Novi Comm. Kea | Eetrop, 2 ;, (1749) iia, 8: Comm. Arith. Coll., 1, 1849, 69-70; Opera
Omnia, (1), II, 236-
Serpe, Mat Math. th. Piva, (e ci, Fuss), 1, 1843, 602, March 12, 1753.
11 [bid., 610, as 616, tr, 621, 625.4 History or THE THEory or NuMBERS. [Cuar. I
Buler? proved that every prime divisor of 22+y?, where x and y are relatively
prime, is itself of that form. First, if N=2a*+ 0? is divisible by a prime P=2p*+q",
the quotient is of that form. Foy, P divides a*?P—p*N=a"q?—b*p? and hence
divides one of its factors. Let agtbp=mP. Then a=mq+(2mp+b)p/q. Hence
(2mp+b) /q is an integer, say =n. Then
bang+2mp, a=mqenp, N=P(2m*+n*).
‘The last identity shows also that the product of several numbers of the form 2a?+b?
is itself of that form. It now follows that if 2a?+b? is divisible by a number not of
that form, the quotient is not a prime 227+? nor the product of primes of that form.
To prove the main theorem, suppose that 227+? (a and y relatively prime) has
a prime divisor A’ not of that form. We may eet
a=mA'+a, y=nA’+b, OSaS}4’, 0SbS}4.
Hence A’ divides A=2a*+b*<3A”. The quotient and hence A has a prime factor
BY’ not of the form 2r7+s?, Also B’<3A’. In case A>%B”, we obtain as before a
number B=2c?+ d? divisible by B’, where c and d are relatively prime, and B<3B”;
in the contrary case, we take B=A. Similarly, B has a prime divisor 0’<4B’,
where C’ is not of the form 227+y’ but is a divisor of a number C=2e?+f?<3C",
e and f relatively prime. Continuing thus we get smaller and smaller numbers
22*-+w* (z prime to w), divisible by numbers not of that form. But the small num-
bers 2z?+-w* have all their divisors of that form.
Buler noted (§ 46) that the method is not applicable to ma?-+y? if $(m-+1)>1,
while for m=8, 2 divides 327+? for s=y=1 but is not of that form, so that the
method does not apply immediately to 32*+-y?.
Euler** later noted the modification of the preceding method which proves that
every prime divisor 2 of 32*+y? (2, y relatively prime) is itself of that form.
We now have a prime divisor A’ 342 of A=3a*+b?|a|, write o=2.+uys, y=yx. Then b:=b+2ap will be numerically
0. Then 44C=B*4+K>0. By (7), |AC|SB* Hence
KS 8B, According as X is even or odd, B may take the even or odd integral values
numerically S VK/3. For each B, A and C are the factors, neither numerically
<|B, of 3(K+B*).
(ii) Let d>0. Then |AC| S B* requires that AC be negative. Thus dS 5B?.
According as d is even or odd, B may take the even or odd integral values numerically
Sv4d/s. For each B, A and C are factors, neither numerically <|B|, of {(B?—d).
Hence for # given discriminant not a perfect square, there is a finite number of
forms (7). In view also of Theorem I, the latter are spoken of as the forms of the
divisors of f=Pt?+Qtu+Ru®. These forms of the divisors depend therefore only
upon the discriminant d of f. This follows also from 4Pf=z*—du?, where
z=2Pt+Qu, whence the divisors of f are divisors of 2?~du?.
For #+au*, a>0, we have P=1, Q=0, R=a, d=—K=~—4a. Hence B is even
and will be replaced by 2B. Since AC is positive, the form represents positive num-
bers only when A and B are positive. If C [el we wnite gm, y=ts+ na, and obtain
=b-+2cu, which ‘can be made numerically 8.
Lagrange investigated (§ 22) the equivalence of (8) with another reduced form
A’c"-+... of the same determinant —a under the substitution
2=Ma’+Ny’, y=ma’+ny/.
As shown in the Introduction above, Mn—Nm=-1. Let |M|>|N|, since the
argument with M and N interchanged applies when |N|>|M|. We can determine
an integer p 5 2 such that |f’|<|N| in M=pN-+M’. Determine m’ by m=pn+m’.
For these values of Mf and m, we get B’=n0’+B”, where
BY =AM'N+B(M’n+Nm’) +Cm/n,
By hypothesis, C’ S 2|B’|, whence |B’|C’. In terms of y”=p2’+y’, our substitution becomes
2=M’s'+Ny”, y=m’2’+ny”, which has the determinant +1 and replaces (8) by
Alc" +2B"2/y"+C’y". Thus 4”0’—B’"=a>0, whence A” >|B”|, since |B”
>O’. Since |N|> ||, we may write N=y’M’ +N’, where p’ is an integer S 2, and
[W’|<|M’|. Writing also n=p’m’+n’, 2’ =p’y"” +2’, we see that our substitution
becomes = Ma" + N’y”, y=m’a” +n’y”, of determinant +1. As before it replaces
(8) by a form Ae’ 2B"a"y" +0"y!" with B’=p’A” +B’, whence |B’”|>
|A”|, |[B’’|<|@”|. The series of decreasing integers |M|, |N|, |M’l, |V’|, -.-
terminates with zero. Assuming that W’=0 for example, we readily conclude from
the above relations that the two reduced forms are identical. But Lagrange failed
to treat the case in which M and W are numerically equal (and hence equal to =+1) ;
in this case the series M, NV, M’, ... does not terminate with zero, but contains only
terms +1. Contrary to his conclusion, s=2’—y’, y=y’ transforms Az?+Acy+Cy?
into Az*—Azy+Cy*. The only other exceptional case is that in which C=A
(Gauss,** Art. 172).
Similarly, every divisor of #—au? (a>0, t, u relatively prime) can be represented
properly by one of the forms (some of which may be equivalent)
(9) 1 Aa*4+2Bry=Cy, CSAS2\BI,
of determinant B?+AC=c. Thus |B|SVa/5. For a=2, there are two such forms
a?—2y* and —2*+2y?, and the former is transformed into the latter by s=2/-+2y’,
y=a'+y’. Hence every divisor of t—2u? is of the form «*—2y?.
Lagrange (§§ 28, 24) gave the following method of testing the equivalence of
forms (9) having a given positive determinant a. For this purpose he used the
notation
(10) ry? +2qyy’—ry”, — 2IglSr>0, %IqISr'>0, rr’ tq?=a.
This form is transformed by y=m’y’+y” into
fary™ +2 yyy", vaqtr'm’, ert gaa
If possible, select an integer m’>0O such that 2lo’! exceeds neither r’ nor r’’: then