2.

S / S S~

Collected Papers of
SRINIVASA RAMANUJAN
 Collected Papers of
SRINIVASA RAMANUJAN
Edited by

G. H. HARDY
P. V. SESHU AIYAR
and
B. M. WILSON

ratfAMOUwra “’••B-jcLmRiAao

<&■

<s&*

Cambridge
AT THE UNIVERSITY PRESS
1927
 CONTENTS

PREFACE . .
NOTICE by P. V. Seshu Aiyab and R. Bamachaundba Rao
NOTICE by G. II. Hahdy.
PAPERS :
1. Some properties of Bernoulli’s numbers . . .
Journal of the Indian Mathematical Society, 3 (1911), 219-234.
2. On Question 330 of Prof. Sanjana.
Journal of the Indian Mathematical Society, 4 (1912), 59-61.
3. Note on a set of simultaneous equations.
Journal of the Indian Mathematical Society, 4 (1912), 94-96.
4. Irregular numbers.*
Journal of the Indian Mathematical Society, 0 (1913), 105-100.
5. Squaring the circle.
Journal of the Indian Mathematical Society, 5 (1913), 132.
6. Modular equations and approximations to 7r
Quarterly Journal of Mathematics, 45 (1914), 350-372.

7. On the integral J —dt.

Journal of the Indian Mathematical Society, 7 (1915), 93-90.
8. On the number of divisors of a number.
Journal of the Indian Mathematical Society, 7 (1915), 131-133.
9. On the sum of the square roots of the first n natural numbers .
Journal of the Indian Mat/iematieal Society, 7 (1915), 173-175.

10. On the product fl j^l + J.

Journal of the Indian Mathematical Society, 7 (1915), 209-211.
11. Some definite integrals.
Messenger of Mathematics, 44 (1915), 10-16.
12. Some definite integrals connected with Gauss’s sums.
Messenger of Ilatlusmatics, 44 (1915), 75-86.
13. Summation of a certain series.
Messenger of Mathematics, 44 (1915), 157-100.
14. New expressions for Riemann’s functions %(s) and 3(f)
■ Quarterly Journal of Mathematics, 46 (1915), 253-260.
 Contents

15. Highly composite numbers.. 78

Proceedings of the London Mathematical Society, 2, 14 (1915), 347-409.

16. On certain infinite series .. 129

Messenger of Mathematics, 45 (1916), 11-15.
17. Some formulae in the analytic theory of numbers
Messenger of Mathematics, 45 (1916), 81-84.

18. On certain arithmetical functions ......

Transactions of the Cambridge Philosophical Society, 22, No. 9 (1916), 159-
184.
19. A series for Euler’s constant 7. 163
Messenger of Mathematics, 46 (1917), 73-80.

20. On the expression of a number in the form ax- + by2 + cz2 + du2 .
. Proceedings of the Cambridge Philosophical Society, 19 (1917), 11-21.

21. On certain trigonometrical sums and their applications in the

theory of numbers. 179
Transactions of the dambridge Philosophical Society, 22, No. 13 (1918),
269-276.
22. Some definite integrals. 200
Proceedings of the London Mathematical Society, 2, 17 (1918), Records for

23. Some definite integrals ..202

Journal of the Indian Mathematical Society, 11 (1919), 81-87.
24. A proof of Bertrand’s postulate.208
■ Journal of the Indian Mathematical Society, 11 (1919), 181-182.
25. Some properties of p(n), the number of partitions of n . . 210
Proceedings of the Cambridge Philosophical Society, 19 (1919), 207-210.
26. Proof of certain identities in combinatory analysis . . . 214
Proceedings of the Cambridge Philosophical Society, 19 (1919), 214-216.. .
27. A class of definite integrals . . . .. . . . 216
Quarterly Journal of Mathematics, 48 (1920), 294-310.
28. Congruence properties of partitions.230
Proceedings of the London Mathematical Society, 2, 18 (1920), Records for ~
13 March 1919. .
29. Algebraic relations between certain infinite products . . . 231
Proceedings of the London Mathematical society, 2, 18 (1920), Records for
13 March 1919.
30. Congruence properties of partitions . 232
Mathematiscke Zeilschrift, 9 (1921), 147-153.
 Contents
PAPERS written in collaboration with G. H. Hardy:
31. Une formule asymptotique pour le nombre des partitions de n . 239
* Coniptes Rendus, 2 Jan. 1917.
32. Proof that almost all numbers n are composed of about log log n
prime factors. 242
Proceedings of the London Mathematical Society, 2, 16 (1917), Records for
14 Dec. 1916.
33. Asymptotic formul® in combinatory analysis .... 244
Proceedings of the London Mathematical Society, 2, 16 (1917), Records for
, 1 March 1917.
34. Asymptotic formul® for the distribution of integers of various
types. 245
Proceedings of the London Mathematical Society, 2, 16 (1917), 112-132.
35. The normal number of prime factors of a number n .
Quarterly Journal of Mathematics, 48 (1917), 76-92.
36. Asymptotic formulae in combinatory analysis .... 276
Proceedings of the Lcndon Mathematical Society, 2, 17 (1*18), 76-115.
37. On the coefficients in the expansions of certain modular functions
Proceedings of the Royal Society, A, 96 (1918), 144-155.
QUESTIONS AND SOLUTIONS. . . .
APPENDIX I: NOTES ON THE PAPERS . . . .
APPENDIX II: EURTHER EXTRACTS FROM RAMANUJAN’S LETTERS
TO G. H. HARDT . . .. 349
 CORRIGENDA

P. 37, line 14: for^ k* read^ F.

P. 45, equation (11): insert dots after the right-hand member.

P. 57, line 11: for n {1 + a2 (os+nf} read n (1 + *2/(« + »)2}-
 PREFACE

T HIS volume contains everything published by Ramanujan except a few

solutions of questions by other mathematicians printed in the. Journal
of the Indian Mathematical Society, and a certain amount of additional matter.
Its publication has been made possible by the liberality of the University of
Madras, the Royal Society, and Trinity College, Cambridge, each of which
bodies has guaranteed a proportion of the expense of printing.
The editorial comments in Appendix I do not profess to be in any way
systematic or exhaustive. We have merely put down such comments and
references to the literature as occurred to us or were suggested to us by other
mathematicians. In particular we are indebted to Prof. L. J. Mordcll for
a number of valuable suggestions.
We have also printed in Appendix II those parts of Ramanujan’s letters
from India which have not been printed before. It may seem that it would
have been more natural to incorporate these in their proper places in the
second Notice, but to do this would have expanded it unduly and destroyed
its proportion, and the letters consist so largely of an enumeration of isolated
theorems that they hardly suffer by division.
There is still a large mass of unpublished material. None of the contents
of Ramanujan’s notebooks has been printed, unless incorporated in later
papers, except that one chapter, on generalised hypergeometric series,
was analysed by Hardy* in the Proceedings of the Cambridge Philosophical
Society. This chapter is sufficient to show that, while the notebooks are
naturally unequal in quality, they contain much which should certainly
be published. It would be a very formidable task to work, through them
systematically, select particular passages, and edit these with adequate
comment, and it is impossible to print the notebooks as they stand without
further monetary assistance. The singular quality of Ramanujan’s work, and
the romance which surrounds his career, encourage us to hope that this
volume may enjoy sufficient success to make possible the publication of
another.
* G. H. Hardy, “A chapter from [Ramanujan's notebook”, Proc. Camb. Phil. Soc.,
XXI (1923), pp. 492-503.
 SRINIVASA RAMANUJAN (1887—1920)
Bv P. V. Seshu Aiyae and R, Ramachandra Rao

Srinivasa Ramanujan Aiyangar, the remarkable mathematical genius

who is the subject of this biographical sketch, was a member of a Brahmin
' family in somewhat poor circumstances in the Tanjore District of the Madras
Presidency. There is nothing specially noteworthy about his ancestry to
account for his great gifts. His father and paternal grandfather were gumastas
(petty accountants) to cloth merchants in Kumbakonam, an important town
in the Tanjore District. His mother, a woman of strong common-sense who
still survives to mourn the loss of her distinguished son, was the daughter of
a Brahmin petty official who held the position of amin (bailiff) in the Munsiff’s
cwirt at Erode in the neighbouring district of Coimbatore. For some time
after her marriage she had no children, hut her father prayed to the famous
goddess Namagiri, in the neighbouring town of Namakkal, to bless his
. daughter with children. Shortly afterwards, her eldest child, the mathe¬
matician Ramanujan, was born on the ninth day off Margasirsha in the
Samvath Sarvajit, answering to the English date of 22nd December 1887.
Ramanujan was bom in Erode, in the house of his maternal grandfather,
to which in accordance with custom his mother had gone for the birth of her •
first child. In 1892, when in his fifth year, he was, as is usual with Brahmin
boys, sent to a pial school, i.e. an indigenous elementary school conducted on
very simple lines. Two years later he was admitted into the Town High
School at Kumbakonam, in which he spent the rest of his school career.
During the first ten years of his life the only indication that he gave of
sp&cial ability was that in 1897 he stpod first amongst the successful candi¬
dates of the Tanjore District in the Primary Examination.. This success
secured for him the concession of being permitted to pay half-fees in his
school.
Even in these early days he was remarkably quiet and meditative. It is
remembered that he used to ask questions about the distances of the stars.
As he held a high place in his class his class fellows used often to go to his
house, but as he knew that his parents did not care for him to go out he used'
only to talk to them from a window which overlooked the street.
While he was in the second form he had, it appears, a great curiosity to
know the “highest truth" in Mathematics, and asked some of his friends in the
higher classes about it. It seems that some mentioned the Theorem of Pytha¬
goras as the highest truth, and that some others gave the highest place to
“Stocks and Shares”. While in the third form, when his teacher was ex-
xii Srinivasa Ramanujan

plaining to the class that any quantity divided by itself was equal to unity,
he is said to have stood up and asked if zero divided by zero also was equal
to unity. It was at about this time that he mastered the properties of the
three progressions. While in the fourth form, he took to the study of Trigo¬
nometry. He is said to have borrowed a copy of the second part of Loney’s
Trigonometry from a student of the B.A. class, who was his neighbour. This
student was struck with wonder to learn that this young lad of the fourth
form had not only finished reading the book but could do every problem in
it without any aid whatever; and not infrequently this B.A. student used to •
go to Ramanujan for the solution of difficult problems. While in the fifth
form, he obtained unaided Euler’s Theorems for the sine and the cosine and,
when he found out later that the theorems had been already proved, he kept
the paper containing the results secreted in the roofing of his house.
It was in 1903, while he was in the sixth form, on a momentous day for
Ramanujan, that a friend of his secured for him the loan of a copy of Carr’s
Synopsis of Pure Mathematic^ from the library of the local Government
College. Through the new world thus opened to him, Ramanujan went
ranging with delight. It was this book that awakened his genius. He set
himself to establish the formulae given therein. As he was without the aid of
other books, each solution was a piece of research so far as he was concerned.
He first devised some methods for constructing magic squares. Then, he
branched ob to Geometry, where he took up the squaring of the circle and
succeeded so far as to get a result for the length of the equatorial circum¬
ference of the earth which differed from the true length only by a few feet.
Finding the scope of geometry limited, he turned his attention to Algebra
and obtained several new series. Ramanujan used to say that the goddess of
Namakkal inspired him with the formulae in dreams. It is a remarkable fact
that frequently, on rising from bed, he would note down results and rapidly
verify them, though he was not always able to supply a rigorous proof. These
results were embodied in a notebook which he afterwards used to show^fo
mathematicians interested in his work.
In December 1903 he passed the Matriculation Examination of the Uni¬
versity of Madras, and in the January of the succeeding year he joined the
Junior First in Arts class, of the Government College, Kumbakonam, and won
the Subrahmanyam scholarship, which is'generally awarded for proficieneyin
English and Mathematics. By this time, he was so -much absorbed in the
study of Mathematics that in all lecture hours—whether devoted to English,
History or Physiology—he used to engage himself in some mathematical
investigation, unmindful of what was happening in the class. This excessive
devotion to Mathematics and his consequent neglect of the other subjects
resulted in his failure to secure promotion to the senior class and in the
consequent discontinuance of the scholarship. Partly owing to disappoint-,
inent and partly owing to the influence of a friend, he ran away northwards
 Srinivasa Ramanujan xiii

into the Telugu country, but returned to Kumbakonam after some wandering
and rejoined the college. As owing to his absence he failed to make sufficient
attendances to obtain his term certificate in 1905, he entered Pachaiyappa’s
College, Madras, in 1906, but falling ill returned to Kumbakonam. He
appeared as a private student for the P.A. Examination of December 1907
and failed. Afterwards he had no very definite occupation till 1909, but con¬
tinued working at Mathematics in his own way and jotting down his results
in another notebook.
In the summer of 1909 he married and wanted to settle down in life.
Belonging to a poor and humble family, with an unfortunate college career,
and without influence, he was hard put to it to secure some means of liveli¬
hood. In the hope of finding some employment he went, in 1910, to Tirukoilur,
a small sub-division town in the South Arcot District, to see Mr V. Ramaswami
Aiyar, M.A., the founder of the Indian Mathematical Society, who was then
Deputy Collector of that place, and asked him for a clerical post in a municipal
or taluq office of his division. This gentleman, being himself a mathematician
of no mean order, and finding that the results contained in Ramanujan’s note¬
book were remarkable, thought rightly that this unusual genius would be
wasted if consigned to the dull routine of a taluq office, and helped Ramanujan
on to Madras with a letter of introduction to Mr P. Vr Seshu Aiyar, now
Principal of Government College, Kumbakonam. Mr Seshu Aiyar had already
known Ramanujan while the latter was at Kumhakonam, as he was the
mathematical lecturer there while Ramanujan was in the P.A. class. Through
him Ramanujan secured for a few months an acting post in the Madras
Accountant-General’s office and, when this arrangement ceased, he lived for
a few months earning what little he could by giving private tuition. Not
satisfied with such make-shift arrangements, Mr Seshu Aiyar sent him with
a note of recommendation to Diwan Bahadur R. Ramachandra Rao, who was
then Collector at Nellore, a small town 80 miles north of Madras, and who
had 'already been introduced to Ramanujan and seen his notebook. His first
interview with Ramanujan in December 1910 is better described in his .own

“Several years ago, a nephew of mine perfectly innocent of mathematical knowledge

said to me, 1 Uncle,' 1 have a visitor who talks of mathematics; I do not understand him;
can you see if there is anything in his talk! ’ And in the plenitude of my mathematical
wisdom, I condescended to permit Ramanujan to walk into my presence. A short uncouth
figure, stout, unshaved, not overolean, with one conspicuous feature—shining eyes—walked
in with a frayed notebook under his arm. He was miserably poor. He had run away from
Kumbakonam to get leisure in Madras to pursue his studies. He never craved for any
distinction. He wanted leisure; in other words, that simple food should be provided for him
without exertion on his part and that he should be allowed to dream on.
“He opened his book and began to explain some of his discoveries. 1 saw quite at
once that {here vTas something out of the way; but my knowledge did not permit me to
judge whether he talked sense or nonsense. Suspending judgment, I asked him to come
over again, and he did. And then he had gauged my ignorance and shewed me some of
xiv Srinivasa Ramanujan
his simpler results. These transcended existing books and I had no doubt that he was a
remarkable man. Then, step by step, he led me to elliptic integrals and hypergeometrie
series'and at last his theory of divergent series not yet announced to the world converted
me. I asked him what he wanted. He said he wanted a pittance to live on so that he
might pursue his researches.” ,
Mr Ramachandra Rao sent him back to Madras, saying that it was cruel
to make an intellectual giant like Ramanujan rot in a mofussil station like
Nellore, and recommended that he should stay at Madras, undertaking to pay
his expenses for a time. After a while, other attempts to obtain for him a-
scholarship having failed and Ramanujan being unwilling to be a burden on
anybody for any length of time, he took up a small appointment on Rs 30 per
mensem in the Madras Port Trust office, on the 9th February 1912. =
He did not slacken his work at Mathematics in the meantime. His earliest
contribution to the Journal of the Indian Mathematical Society was in the
form of questions communicated by Mr Seshu Aiyar and published in
the February number of Volume III (1911). His first long article was on
“Some Properties of Bernoulli’s Numbers” and was published in the JDe-
cember number of the same volume. In 1912 he contributed two more notes
to the fourth volume of the same Journal, and also several questions for
solution.
By this time, Mr Ramachandra Rao had induced Mr Griffith of the Madras
Engineering College to take an interest in Ramanujan, and Mr Griffith spoke
to Sir Francis Spring, the Chairman of the Madras Port Trust, in which -
Ramanujan was then employed ; and from that time onwards it became easy
to secure recognition of his work. Fortunately also the then manager of the
Port Trust office was Mr S. Narayana Aiyar, M.A., a very keen and devoted
student of Mathematics. He gave every encouragement to Ramanujan and
very frequently worked with him during this period.
On the suggestion of Mr Seshu Aiyar and 'others, Ramanujan began a
correspondence with Mr G. H. Hardy, then Fellow of Trinity College, 0am- ■
bridge, on the 16th January 1913. In that letter he wrote:
“I had no University education but I have undergone the ordinary school course.
After leaving school I have been employing the spare time at my disposal to work at
Mathematics....I have made a special investigation of divergent series....Very Recently "~
I came across a tract published by you, styled Orders of. Infinity, in page 36 of which I
find a statement that no definite expression has been as yet found for the number of prime
numbers less than any given number. I have found an expression which very nearly
approximates to the real result, the error being negligible. I would request you to go
through the enclosed papers. Being poor, if you are convinced that there is anything of
value, I would like to have my theorems published I have not given, the actual investiga¬
tions nor the expressions that I get; but I have indicated the lines on'wEioh-I proceed.
Being inexperienced, I would very highly value any advice you give me....”
The papers enclosed contained the enunciations of a hundred or more mathe¬
matical theorems.
 Srinivasa Ramanujan xv

In his second letter of date 27th February 1913, he wrote:

“...I have found a friend in you who views my labours sympathetically. This is already
some encouragement to me to proceed....To preserve my brains, I want-food and this is
now my first consideration. Any sympathetic letter from you will be helpful to me here
to get a scholarship either from the University or from the Government....”
But in the meantime Mr Hardy had written to the Secretary for Indian
Students in London, saying that Ramanujan might prove to be a mathe¬
matician of the very highest class, and asking him to enquire whether some
means could not be found for getting him a Cambridge education. This
question was transmitted to the Secretary of the Students’ Advisory Com¬
mittee in Madras, who, in his turn, asked Ramanujan if he would go to
England. But since his caste prejudices were very strong, he definitely
declined to go. Upon the receipt of this unfavourable reply, the Secretary
wrote, .early in March 1913, to the Registrar of the University of Madras,
explaining the circumstances of the case.
By this time Ramanujan’s case had been brought to the notice of the
Tjniversity of Madras in another way. Early in February, Dr G. T. Walker,
F.R.S., Director-General of Observatories, Simla, and formerly Fellow of
Trinity College, Cambridge, happened to visit Madras on one of his official
tours; and Sir Francis Spring took this opportunity to .bring some of Rama¬
nujan’s work to Dr Walker’s notice. As a result, Dr Walker addressed, on the
26th February 1913, the following letter to the Registrar of the University
of Madras:
“...I have the honour to draw your attention to the case of S. Banianujan, a clerk in
the A&ounts Department of the Madras Port Trust. I have not seen him, but was
yesterday shewn some of his work in the presence of Sir Francis Spring. He is, I am
told, 22 years of age and the character of the work that I saw impressed me as comparable
in originality with that of a mathematical fellow in a Cambridge college....It was per¬
fectly deal- to me that the University would be justified in enabling S. Ramanujan for
a few years at least to spend the whole of his time on Mathemafics, without any anxiety
as*to his livelihood...."
As a result of this momentous letter and on the recommendation of the
Board of Studies in Mathematics, the University granted to Ramanujan, with
the previous approval of Government, a special scholarship of Rs75 per
mensem tenable for two< years. The Syndicate took a special interest in
getting this scholarship sanctioned, as may be seen from the following extract
from the letter of the Registrar to the Government in this connection:
“ The regulations of the University do not at present provide fot such a special scholar¬
ship. But the Syndicate assumes that Section XV of the Act of Incorporation and Section 3
of the Indian Universities Act, 1904, allow of the grant of such a scholarship, subject to
the express consent of the Governor of Fort St George in Council.”
He was accordingly relieved from his clerical post in. the Madras Port Trust
office on the 1st of May 1913, and from that time he became and remained
for the rest of his life a professional mathematician.
xvi Srinivasa Ramanujan

In accordance with the conditions of award of the scholarship, he sub¬

mitted .to the Board of Studies in Mathematics three quarterly reports on
his researches on the 5th August 1913, 7th November 1913 and 9th March
1914 respectively.
But Mr Hardy was very much disappointed at Ramanujan’s refusal to go
to Cambridge. He had been at frequent intervals writing persuasive letters
pointing out the advantages of a short stay in Cambridge, and when, early
in 1914, the University of Madras invited Mr B. H. Neville, M.A., Fellow
of Trinity College, Cambridge, to deliver a course of lectures at Madras,
Mr Hardy used this opportunity and entrusted to Mr Neville the mission of
persuading Ramanujan to give up his caste prejudices and come to Cambridge.
In the meantime, many Indian friends also had been influencing him and, by
the time Mr Neville approached him, Ramanujan himself had almost made
up his mind; but his chief difficulty was to obtain his mother’s consent. This
consent was at last got very easily in an unexpected manner. For one morn¬
ing his mother announced that she had had a dream on the previous night,
in which she saw her son seated in a big hall amidst a group of Europeans,
and that the goddess Namagiri had commanded her not to stand.in the way
of her son fulfilling his life’s purpose. This was a very agreeable surprise to
all concerned.
As soon as Ramanujan’s consent was obtained, Mr Neville sent a memo¬
randum to the authorities of the University of Madras on 28th January 1914.
The memorandum ran as follows:

estimated....
“ I see no reason to doubt that Ramanujan himself will respond fully to the stimulus
which contact with western mathematicians of the highest class will afford him. In that

University and the City of Madras will be proud to have assisted in his passage from
obscurity to fame.”
The next day, Mr R. Littlehailes, M.A., who was then Professor of Mathe¬
matics in the Presidency College, Madras, and now is the Director of Public
Instruction, Madras, wrote another long letter to the Registrar of the Univer¬
sity and made definite proposals regarding the. scholarship to he granted.
The authorities of the University readily seized the opportunity and within
a week decided, with the approval of the Government of Madras, to grant
Ramanujan a scholarship of £250.a year, tenable in England for a period of
two years, with free passage and a reasonable sum for outfit. This scholarship
was subsequently extended, up to 1st April 1919. Having arranged that the
University should forward Rs,60 per mensem out of his scholarship to his
mother at Kumbakonam, Ramanujan sailed for England on the 17th March
 Srinivasa Ramanujan xvii

1914. He reached Cambridge in April and was admitted into Trinity College,
which supplemented his scholarship by the award of an exhibition of £60.
He was now for the first time in his life in a really comfortable position
and could devote himself to his researches.without anxiety. Mr Hardy and
Mr Littlewood helped him in publishing his papers in the English periodicals
and under their guidance he developed rapidly.
On the 11th November 1915, Mr Hardy wrote to the Registrar of the
Madras University:
“ Ramanujan has been much handicapped by the war. Mr Littlewood, who would
naturally have shared his teaching with me, has been away, and one teacher is not enough
for so fertile a pupil.He is beyond question the best Indian mathematician of modern
times...He will always be rather eccentric in his choice of subjects and methods of dealing
with them.But of his extraordinary gifts there can bo no question; in some ways he is
the most remarkable mathematician I have ever known."
Mr Hardy’s official report of date 16th June 1916 to the University of
Madras was also in terms of very high praise*. Ramanujan had already pub¬
lished about a dozen papers in European journals. Everything went on well
till the spring of 1917.
About May 1917, Mr Hardy wrote that it was suspected that Ramanujan
had contracted an incurable disease. Since sea voyages were then risky on
account of submarines and since the war had depleted„India of good medical
men, it was decided that he should stay in England for some time more.
Hence he went into a nursing home at Cambridge in the early summer, and.
he was never out of bed for any length of time again. He was in sanatoria
at Wells, at Matlock and in London, and it was not until the autumn of 1918
that he shewed any decided symptom of improvement.
On the 28th February 1918, he was elected a Fellow of the Royal Society.
He was the first Indian on whom this high honour was conferred, and his
election at the early age of thirty, and on the first occasion that his name was
proposed, is a remarkable tribute to his distinguished genius. Stimulated
perhaps by this election, he resumed active work, in spite of his ill-health,
and some of his most beautiful theorems were discovered about this time. On
the 13th October of the same year, he was elected a Fellow of Trinity College,
Cambridge—a prize fellowship worth about £250 a year for six years, with
no duties or conditions. In announcing this election, Mr Hardy wrote to the
Registrar of the University of Madras, “ He will return to India with a
scientific standing and reputation such as no Indian has enjoyed before, and
I am confident that India will regard him as the treasure he is ”, and urged
the authorities of the University to make permanent provision for him in a
way which could leave him free for research. This time also the University
of Madras rose to the occasion and, in recognition of Ramanujan’s services to
the science of Mathematics, it granted him an allowance of £250 a year for
* Of. Journal of the Indian Mathematical Society, 9 (1917), pp. 30-45.
xviii Srinivasa Ramanujan

five years from 1st April 1919, the date of the expiry of his scholarship,
together with the actual expenses incurred by him in returning from England
to India and on such passages from India to Europe and back as the Syndicate
might approve of during the five years. At the suggestion of Mr Littlehailes
the University of Madras also contemplated creating a University Professor¬
ship of Mathematics and offering it to him.
By this time his health shewed some signs of improvement. Although he
shewed a tubercular tendency, the doctors said that he had never been gravely
affected. Since the climate of England was suspected of retarding his
recovery, it was decided to send him back to India. Accordingly, he left
England on 27th February 1919, landed in Bombay on 27th March and
arrived at Madras on the 2nd April. ••
When he returned he was in a precarious state of health. His friends
grew very anxious. The best medical attendance was arranged for. He
stayed three months in Madras and then spent two months in Kodumudi, a
Village on the banks of the Cauvery, not far from the place of his birth. He
was a difficult patient, always inclined to revolt against medical treatment
and after a time he declined to be treated further. On the 3rd September
he went to Kumbakonam, -and since it was reported by many medical friends
that he was getting worse, he was with great difficulty induced to come to
Madras for treatment In January 1920 and was put under the best available
medical care. Several philanthropic gentlemen assisted him during this
period, notably Mr S. Srinivasa Aiyangar, who found all his expenses, and
Eao Bahadur T. Numberumal Chetty, who gave his house free. The members
of the Syndicate of the University of Madras also made a contribution
towards his expenses in their individual capacity. But all this yas. of no
avail. He died on the 26th April 1920, at Chetput, a suburb of Madras.' He
had no children but was survived by his parents and his wife.
We must refer to Mr Hardy’s notice for an account of his mathematical
work, but we add a few words about his appearance and personality. Before
his illness he was inclined to stoutness; he was of moderate height (5 feet
5 inches); and had a big head with a large forehead and long wavy jiark
hair. His most remarkable feature was his sharp and bright dark eyes. A
fairly faithful representation of him adorns the walls of the Madras University
Library. On his return from Englaud, he- was very thin and emaciated and
had grown very pale. He looked as if racked with pain. But his intellect was
undimmed, and till about four days before he died he was engaged in work.
All his work on “mock theta functions”, of which only rough indications
survive, was done on his death bed.
Ramanujan had definite religious views. He had a special veneration
for the Namakkal goddess. Fond of the Puranas, he used to.attend popular
lectures on the Great Epics, of Ramayana and 'jjUbabharata, and to enter into
discussions with learned pundits. He believed in the existence of a Supreme
 Srinivasa Ramanujan xix

Being and in the attainment of Godhood by men by proper methods of

service and realisation of oneness with the Deity. He had settled convictions
about the problem of life and after, and even the certain approach of death
did not unsettle his faculties or spirits.
In manners he was very simple and he had absolutely no conceit. In a
letter of date 26th November 1918, i.e. after Ramanujan had been honoured
by being elected a Fellow of the Royal Society and a Fellow of Trinity,
Mr Hardy wrote: “ His natural simplicity has never been affected in the
least by his success; indeed all that is wanted is to get him to realise that
he really is a success.” He was much distressed, when he had so little money
for his own expenses, about his inability to help his poor parents; and when
hS received his scholarship, his first act was to devote a part of it to them.
Ramanujan’s simplicity and largeness of heart are further revealed in the
following letter that he sent to the Registrar of the University of Madras:
2 Colinette Road, Potnet, S.’W. IB. ’
• 11th January 1919,
To The Registrar,
University of Madras.

I beg to acknowledge the receipt of your letter of 9th December 1918, and gratefully
accept the very generous help which the University offers me.
I feel, however, that, after my return to India, which I expect to happen as soon as
arrangements can be made, the total amount of money to which I shall he entitled will be
... " ' my expenses in England have
 SRINIVASA RAMANUJAN (1887—1920)
By G. H. Hardy*
I
Srinivasa Ramanujan, who died at Kumbakonam on April 26th, 1920,
had been a member of the Society since 1917. He was not a man who
talked much about himself, and until recently I knew very little of his early
life. Two notices, by P. V. Seshu Aiyar and R. Ramachandra Rao, two of the
most devoted of Ramanujan’s Indian friends, have been published recently in
the Journal of the Indian Mathematical Society; and Sir Francis Spring has
very kindly placed at my disposal an article which appeared' in the Madras
Times of April 5th, 1919. From these sources of information I can now supply
a good many details with which I was previously unacquainted. Ramanujan
(Srinivasa Iyengar Ramanuja Iyengar, to give him for once his proper name)
was born on December 22nd, 1887, at Erode in southern India. His father
W£S an accountant (gumasta) to a cloth merchant at Kumbakonam, while his
maternal grandfather had served as amin in the Mumiff’s (or local judge’s)
Court at Erode. He first went to school at five, and was transferred before he
was seven to the Town High School at Kumbakonam, where he held a “ free
scholarship ”, and where his extraordinary powers appea~ to have been recog¬
nised immediately. “ He used ”, so writes an old schoolfellow to Mr Seshu
Aiyar, “to borrow Carr’s Synopsis of Pure Mathematics from the College
library, and delight in verifying some of the formul® given there.. ..He used to
entertain his friends with his theorems and formul®, even in those early days.....
He had an extraordinary memory and could easily repeat the complete lists
■ pf Sanscrit roots (atmanepada and parasmepada); he could give the values
of 7r, e, .... to any number of decimal places....In manners, he was
simplicity itself....”
He passed his matriculation examination to the Government College at
Kumbakonam in 1904; and secured the “Junior Subrahmanyam Scholar¬
ship”. Owing to weakness in English, he failed in his next examination and
lost his scholarship; and left Kumbakonam, first for Vizagapatam and then
for Madras. Here he presented himself for the “First Examination in Arts’
in December 1906, but failed and never tried again. For the next few years
he continued his independent work in mathematics, “jotting down his
results in two good-sized notebooks ”: I have one of these notebooks in my
possession still. In 1909 he married, and it became necessary for him to find
some permanent employment, I quote Mr Seshu Aiyar:
To this end, he went to Tirukoilur, a small sub-division town in South Arcot District,
to see Mr V. Kamaswami Aiyar, the founder of the Indian Mathematical Society, but
* Obituary notice in the Proceedings of the London Mathematical Society (2), xix (1921),
pp. xl—Iviii. The same notice was printed, with slight changes, in the Proceedings of the
Royal Society (A), xeix (1921), pp. xiii—xxix.
 xxii Srinivasa Ramanujan
Mr Aiyar, seeing his wonderful gifts, persuaded him to go to Madras. It was then after
some four years’ interval that Kamanujau met me at Madras, with his two good-sized
notebooks referred to above. I sent Ramanujan with a note of recommendation to that
true lover of Mathematics, Diwan Bahadur R. Ramaehandra Kao, who was then District
Collector at Nellore, a small town some eighty miles north of Madras. Mr Kao sent him
back to me, saying it was cruel to make an intellectual giant like Ramanujan rot at a
mofussil station like Nellore, and recommended his stay at Madras, generously under¬
taking to pay Ramanujan's expenses for a time. This was in December 1910. After
a while, other attempts to obtain for him a scholarship having failed, and Kamanujan
himself being unwilling to he a burden on anybody for any length of time, he decided to
take up a small appointment under the Madras Port Trust in 1912.
But he never slackened his work at Mathematics. His earliest contribution to the •
Journal of the Indian Mathematical Society was in the form of questions communicated by
me in. Vol. ill (1911). His first long article on “Some Properties of Bernoulli’s Numbers”
was published in the December number of the same volume. Ramanujan's methods
were so terse and novel and his presentation was so lacking in clearness and precision,
that the ordinary reader, unaccustomed to such intellectual gymnastics, could hardly
. follow him. This particular article was returned more than once by the Editor before it
took a form suitable for publication. It was during this period that he came to me one
day with some theorems on Prime Numbers, and when I referred him to Hardy’s Trffct
on Orders of Infinity, he observed that Hardy had said on p. 36 of his Tract “ the exact
order of p (a1) [defined by the equation

where rr (x) denotes the number of primes less than »], has not yet been determined ”, and
that he himself had discovered a result which gave the order of p (x). On this I suggested
that he might communicate his result to Mr Hardy, together with some more of his
results.
This passage brings me to the beginning of my own acquaintance with
Ramanujan. But before I say anything about the -letters which I received
from him, and which resulted ultimately in his journey to England, I must add
a little more about his Indian career. Dr G. T. Walker, F.R.S., Head of the
Meteorological Department, and formerly Fellow and Mathematical Lecturer
of Trinity College, Cambridge, visited Madras for some official purpose some
time in 1912 ; and Sir Francis Spring, K.C.I.E., the Chairman of the Madras
Port Authority, called his attention to Ramanujan’s work. Dr Walker was
far too good a mathematician not to recognise its quality, little as it had in
common with his own. He brought Ramanujan’s case to the notice of the
Government and the University of Madras. A research studentship, “ Rs. 75
per mensem for a period of two years ”, was awarded him; and he became,
and remained for the rest of his life, a professional mathematician.
H
Ramanujan wrote to me first on January 16th, 1913, and at fairly regular
intervals until he sailed for England in 1914. I do not believe that his letters
were entirely his own. His knowledge of English, at that stage of his life,
could scarcely have been sufficient, and there is an occasional phrase which is
hardly characteristic. Indeed I seem to remember his telling ure 'that his
 Srinivasa Ramanujan xxiii

friends had given him some assistance. However, it was the mathematics
that mattered, and that was very emphatically his.
. Madras, 16th January 1913.
Dear Sir,
I beg to introduce myself to you as a clerk in the Accounts Department of the
Port Trust Office at Madras on a salary of only £20 per annum. I am now about 23
years of age. I have had no University education but I have undergone the ordinary
school course. After leaving school I have been employing the spare time at my disposal
to work at Mathematics. I have not trodden through the conventional regular course
. which is followed in a University course, but I am striking out a new path for myself.
I have made a special investigation of divergent series in general and the results I get are
termed by the local mathematicians as “startling”.
Just as in elementary mathematics you give a meaning to a" when » is negative and
fractional to conform to the law which holds when n is a positive integer, similarly the
whole of my investigations proceed on giving a meaning to Eulerian Second Integral for ,
all values of n. My friends who have gone through the regular course of University eduea-

intggral relation is not true when n is negative. Supposing this is true only for positive
values of n and also supposing the definition «r(n) = r(» + l) to be universally true, I
have given meanings to these integrals and under the conditions I state the integral is
true for all values of n negative and fractional. My whole investigations are based upon
this and I have been developing this to a remarkable extent so much so that the local
mathematicians are not able to understand me in my higher flights.
Very recently I came across a tract published by you styled Orders of Infinity in page
36 of which I find a statement that no definite expression has been as yet found for the
number of prime numbers less than any given number. I have found au expression which
very nearly approximates to the real result, the error being negligible. I would request
you to go through the enclosed papers. Being poor, if you are convinced that there is
anything of value I would like to have my theorems published. I have not given the
actual investigations nor the expressions that I.get but I have indicated the lines on
which I proceed. Being inexperienced I would very highly value any advice you give me.
Requesting to be excused for the trouble I give you.
I remain, Dear Sir, Yours truly,
S. Ramanujan.
.. P.S. My address is S. Ramanujan, Clerk Accounts Department, Port Trust, Madras,
India.
I quote now from the “papers enclosed and from later letters*:
In page 36 it is stated that “ the number of prime numbers less than

where the precise order of p (aj has not been determined....”

I have observed that p (e2’*) is of such a nature that its value is very small when x
lies between 0 and 3 (its valuo is less than a few hundreds when x -3) and rapidly
increases when x is greater than 3....
The difference between the number of prime numbers of the form 4n -1 and which
are less than x and those of the form '4ri 4-1 less than x is infinite when x becomes
infinite....
* [See Appendix II for parts of the letters not printed here.]
b-2
XXIV Srinivasa Ramanujan

The following imples from my theorems :

(1) The numbers of the form 2»3» less than n=\ S/og21og3~ where P and ? mi
re any positive integral value including 0.
(2) Lot us take all numbers containing an odd number of dissimilarlprime divisors, v
2, 3, 5, 7, XI, 13, 17, 19, 23, 29, 30, 31, 37, 41, 42, 43, 47.

(«) numbers

W J5 + g5 + 52 + 73 + ■1 • + 3Q2 + + "' _ 2rrs'

(3) Let us take the number of divisors of natural numbers, viz.

1, 2, 2, 3, 2, 4, 2, 4, 3, 4, 2, ... (1 having 1 divisor, 2 having 2, 3 having 2, 4 having
3, 5 having 2,...).

«= »(2y -1 + log m)+i of the number of divisors of »

where y='5772166649..., the Eulerian Constant. r‘
(4) 1,2, 4, 5, 8, 9,10,13,16,17,18,... are numbers which are either themselves squares
or which can be expressed as the sum of two squares.
The number of such numbers greater than A and less than B

and 6 Uc) is very small when compared with the previous integral. E and 6 (%) have been
exactly found though complicated....
Ramanujan’s theory of primes was vitiated by his ignorance of the theory
of functions of a complex variable. It was (so to say) what the theory might
be if the Zeta-function had no complex zeros. His methods of proof depended
upon a wholesale use of divergent series. He disregarded entirely all the
difficulties which are involved in the interchange of double limit operations;
he did not distinguish, for example, between the sum of a series la,, and. the
value of the Abelian limit ' _ _
lira 2 an xn,

or that of any other limit which might be used for similar purposes by a modem
analyst. There are regions of mathematics in which the precepts of modem
rigour may be disregarded with comparative safety, but the Analytic Theory
of Numbers is not one of them, and Ramanujan’s Indian work on primes, and
on all the allied problems of the theory, was definitely wrong. That his proofs
should^ have been invalid was only to be expected. But the mistakes went
deeper than that, and many of the actual results were false. He had obtained
the dominant terms of the classical formulae, although by invalid methods;
but none of them is such a close approximation as he supposed.
This should presumably be 6 (B).
 Srinivasa Ramanuji

This may be said to have been Ramanujan’s one great failure. And yet
I am not sure that, in some ways, his failure was not more wonderful than
• any of his triumphs. Consider, for example, problem (4). The dominant term,
viz. IiB (log B)~i, in Ramanujan’s notation, was first obtained by Landau in
1908. Ramanujan had none of Landau’s weapons at his command; he had
never seen a French or German book; his knowledge even of English was in¬
sufficient to enable him to qualify for a degree. It is sufficiently marvellous
that he should have even dreamt of problems such as these, problems which
-it has taken the finest mathematicians in Europe a hundred years to soive,
and of which the solution is incomplete to the present day.

■7-.IV. Theorems on integrals. The following are a few examples :

* There is always more in one of Ramanujan’s foi-muhe than meets the eye, as anyone
who sets to work to verify those which look the easiest will soon discover. In some the
interest lies very deep, in others comparatively near the surface; hut there is not one
which is not curious and entertaining.
Srinivasa Ramanujan
Srinivasa Ramanujan
xxviii Srinivasa Ramanujan

The order of 6 M which you asked in your letter is J(j^) .

with the conditions ap = ir2...,

Srinivasa Ramanujan
 Srinivasa Ramanujc
I am willing to place unreservedly in your hands what little I have. It was on account of

my own way of arriving at the expressions I have already given....

...I ain glad to inform you that the local University has been pleased to grant me a
scholarship of £60 per annum for two years and this was at the instance of Dr Walker,
F.B.S., Head of the Meteorological Department in India, to whom my thanks are due....
I request you to convey my thanks also to Mr Littlewood, Dr Barnes, Mr Berry and others

III
It is unnecessary to repeat the story of how Ramanujan was brought to •
England. There were serious difficulties; and the credit for overcoming them
is due primarily to Prof. E. H. Neville, in whose company Ramanujan
arrived in April 1914. He had a scholarship from Madras of £250, of which
£50 was allotted to the support of his family in India, and an exhibition of
£60 from Trinity. For a man of his almost ludicrously simple tastes, this was
aja ample income; and he was able to save a good deal of money which was
.. badly wanted later. He had no duties and could do as he pleased; he wished
indeed to qualify for a Cambridge degree as a research student, but this was
a formality. He was now, for the first time in his life, in a really comfortable
position, and could devote himself to his researches without anxiety.
There was one great puzzle. Wliat was to be done in the way of teaching
him modern mathematics ? The limitations of his knowledge were as startling
as its profundity. Here was a man who could work out modular equations,
and theorems of complex multiplication, to orders unheard of, whose mastery
of continued fractions was, on the formal side at any rate, beyond that of any
mathematician in the world, who had found for himself the functional equa¬
tion of the Zeta-function, and the dominant terms of many of the most
famous problems in the analytic theory of numbers; arid he had never heard
of a doubly periodic function or of Cauchy’s theorem, and had indeed but the
vaguest idea of what a function of a complex variable was. His ideas as to
what constituted a mathematical proof were of the most shadowy description.
All his results, new or old, right or wrong, had been an-ived at by a process
of mingled argument, intuition, and induction, of which he was entirely_
unable to give any coherent account.
It was impossible to ask such a man to submit to systematic instruction,
to try to learn mathematics from the beginning once more. I was afraid too
that, if I insisted unduly on matters which Ramanujan found irksome, I might
destroy his confidence or break the spell of his inspiration. .On the other
hand there w’ere things of which it was impossible that he should remain in
ignorance. Some of his results were wrong, and in particular those which
concerned the distribution of primes, to which he attached the greatest im¬
portance. It was impossible to allow him to go through life supposing that
all the zeros of the Zeta-function were real. So I had to try to teach him,
arid in a measure I succeeded, though obviously I learnt from him much more
 Srinivasa Ramanujan xxxi

than he learnt from me. In a few years’ time he hg,d a very tolerable know¬
ledge of the theory of functions and the analytic theory of numbers. He was
never a mathematician of the modem school, and it was hardly desirable that
lie should become one; but he knew when he had proved a theorem and when
he had not. And his flow of original ideas shewed no symptom of abatement.
I should add a word here about Ramanujan’s interests outside mathe¬
matics. Like his mathematics, they shewed the strangest contrasts. He had
very little interest, I should say, in literature as such, or in art, though he
-could tell good literature from bad. On the other hand, he was a keen philo¬
sopher, of what appeared, to followers of the modern Cambridge school, a
rather nebulous kind, and an ardent politician, of a pacifist and ultra-radical
type. He adhered, with a severity most unusual in Indians' resident in
England, to the religious observances of his caste; but his religion was a
matter of observance and not of intellectual conviction, and I remember well
his telling me (much to my surprise) that all religions seemed to him more's
or (pss equally true. Alike in literature, philosophy, and mathematics, he had1'
a passion for what was unexpected, strange, and odd; he had quite a small
library of books by circle-squarers and other cranks.
It was in the spring of 1917 that Ramanujan first appeared to be unwell.
He went into a Nursing Home at Cambridge in the early summer, and was
never out of bed for any length of time again. He was in sanatoria at Wells,
at Matlock, and in London, and it was not until the autumn of 1918 that he
shewed any decided symptom of improvement. He had then resumed active
work, stimulated perhaps by his election to the Royal Society, and some of
his most beautiful theorems were discovered about this time. His election to
a Trinity Fellowship was a further encouragement; and each of those famous
societies may well congratulate themselves that they recognised his claims
before it was too late. Early in 1919 he had recovered, it seemed, sufficiently
for the voyage home to India, and the best medical opinion held out hopes of
a permanent restoration. I was rather alarmed by not hearing from him for
a considerable time; but a letter reached me in February 1920, from which
it appeared that he was still active iu research.
University op Madras.
12</i January 1920.
xxxii Srinivasa Ramanujan

Mock S-functions (of 5th oyler)

q i?4 _ 2° _,
/(?)=1+l^+(1+5)(1+e2)+(1+?) (1+?*)(1+S»)+''"

Mock 3-functions {of 1th order)

m x, i | Ql i _+ . .
He said little about liis health, and what he said was not particularly
discouraging; and I was quite unprepared for the news of his death.
IY '
Ramanujan published the following papers in Europe :
(1) “Some definite integrals”, Messenger of Mathematics, Yol. 44 (1914), pp. 10—18.

”,
(2) “Some definite integrals connected with Gauss’s sums”, ibid., pp. 75—86.
(3) “ Modular equations and approximations to jr Quarterly Journal of Mathematics
Yol. 46 (1914), pp. 350—372.
(4) “Mew expressions for Biemann's functions |(s) and 2(0”, ibid., Vol. 48 (1915),
pp. 263-260.
(5) “On certain infinite series”, Messenger of Mathematics, Vol. 46 (1916), pp. 11—15.
(6) “Summation of a certain series”, ibid., pp. 157—160.
(7) “Highly composite numbers”, Proc. London Math. Soc., Ser. 2, Yol. 14 (1915),

(8) “ Some' formula! in the analytic theory of numbers ”, Messenger of Mathematics,

Vol. 45 (1916), pp. 81—84.
(9) “ On certain arithmetical functions ”, Trans. Cambridge Phil. Soc., Vol. 22 (1916),
No. 9, pp. 159—184.
(10) “A series for Euler’s constant y”, Messenger of Mathematics, Yol. 46 (1917),
pp. 73—80.
(11) “ On the expression of a number in the form aaf + by^+cz^+dfi", Proc. Cambridge
Phil. Soc., Yol. 19 (1917), pp. 11—21.
*(12) “ Une formule asymptotique pour le nombre des partitions de n”, Gomptes Rendus,
2 Jan. 1917.
*(13) “Asymptotic formulae for the distribution of integers of various types”, Proc.
London Math. Soc., Ser. 2, Vol. 16 (1917), pp. 112—132.
*(14) “The normal number of prime factors of a number Quarterly Journal of
Mathematics, Vol. 48 (1917), pp. 76-92.
*(16) “Asymptotic formula! in Combinatory Analysis”, Proc. London Math. Soc:, Ser. 2,
Yol. 17 (1918), pp. 75-116.
*(16) “ On the coefficients in the expansions of certain modular functions", Proc. Roy.
Soc. (A), Yol. 95 (1918), pp. 144—155.
(17) “ On certain trigonometrical sums and their applications in the theory of
numbers”, Trans. Cambridge Phil. Soc., Vol. 22,No. 13 (1918), pp. 259—276.
(18) “ Some properties of p (»), the number of partitions of n ”, Proc. Cambridge Phil.
Soc., Vol. 19 (1919), pp. 207—210.
(19) “Proof of certain identities in Combinatory Analysis”, ibid., pp. 214—216.
(20) “A class of definite integrals”, Quarterly Journal of Mathematics, Yol. 48 (1920),
pp. 294—310.
(21) “Congruence properties of partitions”, Math. Zeitsehrift, Yol. 9 (1921), pp.
 Srinivasa Ramanujan xxxiii

Of these, those marked with an asterisk were written in collaboration with

me, and (21) is a posthumous extract from a much larger unpublished manu¬
script in my possession.* He also published a number of short notes in the
Recurds of Proceedings at our meetings, and in the Journal of the Indian
Mathematical Society. The complete list of these is as follows :
Records of Proceedings at Meetings.
*(22) “ Proof that almost all numbers n are composed of about log log n prime factors ",

*(23) “Asymptotic formulas in Combinatory Analysis”, 1 March 1917.

(24) “Some definite integrals” 17 Jan.1918.
(25) “ Congruence properties of partitions”, 13 March 1919.
(26) “ Algebraic relations between certain infinite products”, 13 March 1919.
Journal of the Indian Mathematical Society.
(A) Articles and Notes.
(27) “ Some properties of Bernoulli’s numbers'”, Vol. 3 (1911), pp. 219—234.
(28) i‘ On Q. 330 of Prof. Sanjana”, Vol. 4 (1912), pp. 59—61.
(29) “A set of equations”, Vol. 4 (1912), pp. 94-96.
(30) “Irregular numbers”, Vol. 5 (1913), pp. 105—106.
(31) “Squaring the circle”, Vol. 5 (1913), p. 132.
(32) “Ou the integral f* dt", Vol. 7 (1916), pp. 93-96.
(33) “ On the divisors of a number”, Vol. 7 (1915), pp. 131-133.
(34) “The sum of the square roots of the first n natural numbers”, Vol. 7 (1916),
pp. 173—176.
(35) “ On the product n [l + (a+^)2] ”, Vol. 7 (1916), pp. 209-211.
(36) “ Some definite integrals”, Vol. 11 (1919), pp. 81-87.
(37) “A proof of Bertrand’s postulate”, Vol. 11 (1919), pp. 181—182.
(38) (Communicated by S. Narayana Aiyar), Vol. 3 (1911), p. 60.
(B) Questions proposed and solved. '
Nos. 260, 261, 283, 284, 289, 294, 296, 308, 353, 358, 359, 386, 427, 441, 464, 489, 507,
524, 625, 541, 546, 671, 605, 606, 629, 642, 666, 682, 700, 723, 724, 739, 740, 753, 768, 769,
783, 786, 1070.
(C) Questions proposed but not solved.
Nos. 327, 352, 387, 441, 463, 469, 526, 584, 661, 662, 681, 699, 722, 738, 754, 755, 770,
784, 1049, and 1076.
Finally, I may mention the following writings by other authors, con¬
cerned with Ramanujan’s work.f
“ Proof of a formula of Mr Kamanujan ”, by G. H. Hardy {Messenger of Mathematics,
Vol. 44, 1915, pp. 18—21).
* All of Ramanujan's manuscripts passed through my hands, and I edited them very
carefully for publication. The earlier ones 1 rewrote completely. I had no share of any
kind in the results, except of course when I was actually a collaborator, or when explicit
 xxxiv Srinivasa Ramanujan
“ Mr S. Ramanujan’s mathematical work in England ”, by G. H. Hardy (Report to the
University of Madras, 1916, privately printed).
“On Mr Ramanujan’s empirical expansions of modular functions”, by L. J. Mordell
(Pros. Cambridge Phil. Soc., Vol. 19, 1917, pp. 117—124).
“Life sketch of Ramanujan” (editorial in the Journal of the Indian Math. Soc.,
Vol. 11, 1919, p. 122).
“ Note on the parity of the number which enumerates the partitions of a number ”, by
P. A. MacMahon (Pm. Cambridge Phil. Soc., Vol. 20, 1921, pp. 281—283).

PH° B^C.Dariingt/vl Zondon Math. Soc., Ser. 2, Vol. 19, 1921, pp. 350-372). *
“On a type of modular relation”, by L. J. Rogers (ibid., pp. 387—397).

It is plainly impossible for me, within the limits of a notice such as this,
to attempt a reasoned estimate of Kamanujan’s work. Some of it is very
intimately connected with my own, and my verdict could not be impartial;
there is much too that I am hardly competent to judge; and there is a mass
of unpublished material, in part new and in part anticipated, in part proved
"and in part only conjectured, that still awaits analysis. But it may be useful
if I state, shortly and dogmatically, what seems to me Ramanujan’s tines't,
most independent, and most characteristic work.
His most remarkable papers appear to me to be (3), (7), (9), (17), (18),
(19), and (21). The first of these is mainly Indian work, done before he came
to England; and much of it had been anticipated. But there is much that is
new, and in particular a very striking series of algebraic approximations
to tr. I may mention only the formulae
63 17 + 15^5 1 _ 1103
~ 25 7 + 15V5 1 2ttV2- 992 ’
correct to 9 and 8 places of decimals respectively.
The long memoir (7) represents work, perhaps, in a backwater of mathe¬
matics, and is somewhat overloaded with detail; but the elementary
analysis of “highly composite” numbers—numbers which have more divisors
than any preceding number—is most remarkable, and shews very clearly
Ramanujan’s extraordinary mastery over the algebra of inequalities. Papers
(9) and (17) should he read together, and in connection with Mr Mordell’s
paper mentioned above; for Mr Mordell afterwards proved a great deal that
Ramanujan conjectured. They contain, in particular, very original and im¬
portant contributions to the theory of the representation of numbers by sums
of squares. But I am inclined to think that it was in the theory of partitions,
and the allied parts of the theories of elliptic functions and continued fractions,
that Ramanujan shews at his very best. It is in papers (18), (19), and (21), ■
and in the papers of Prof. Rogers and Mr Darling that I have quoted, that
this side of his work (so far as it has been published) is to be found. It would
be difficult to find more beautiful formulae than the “ Rogers-Ramanujan ”
identities, proved in (19); hut here Ramanujan must take second place to
 Srinivasa Ramanvjan xxxv

Prof. Rogers; and, if I had to select one formula from all Ramanujans work,
I would agree with Major MacMahon in selecting a formula from (18), viz.

p{i)+p(9)x + pO-N)^+...
0 {(l-*)(l-^)(l-^)...f ’
where p (n) is the number of partitions of n.
I have often been asked whether Ramanujan had any special secret;
whether his methods differed in kind from those of other mathematicians ;
1 whether there was anything really abnormal in his mode of thought. I cannot
answer these questions with any confidence or conviction; but I do not
believe it. My belief is that all mathematicians think, at bottom, in the same
kind of way, and that Ramanujan was no exception. He had, of course, an
extraordinary memory. He could remember the idiosyncrasies of numbers in
an almost uncanny way. It was Mr Littlewood (I believe) who remarked that
“ every positive integer was one of his personal friends.” I remember once
going to see him when he was lying ill at Putney. I had ridden in taxi-cab
No. 1729, and remarked that the number (7.13.19) seemed to me rather a
dull one, and that I hoped it was not an unfavourable omen: “ No,” he replied,
“ it is a very interesting number; it is the smallest number expressible as a
sum of two cubes in two different ways.” I asked him, naturally, whether he
knew the answer to the corresponding problem for fourth powers; and he
replied, after a moment’s thought, that he could see no obvious example,
' and thought that the first such number must be very large.* His memory,
and his powers of calculation, were very unusual, but they could not reason¬
ably be called “abnormal”. If he bad to multiply two large numbers, he
multiplied them in the ordinary way; he would do it with unusual rapidity
and accuracy, but not more rapidly or more accurately than any mathematician
who is naturally quick and has the habit of computation. There is a table of
partitions at the end of our paper (15). This was, for the most part, calculated
independently by Ramanujan and Major MacMahon; and Major MacMahon
was, in general, slightly the quicker and more accurate of the two.
It was his insight into algebraical formulae, transformations of infinite
series, and so forth, that was most amazing. On this side most certainly I have
never met his equal, and I can compare him only with Euler or Jacobi. He
worked, far more than the majority of modem mathematicians, by induction
from numerical examples; all of his congruence properties of partitions, for
example, were discovered in this way. But with his memory, his patience,
and his power of calculation, he combined a power of generalisation, a feeling
for form, and a capacity for rapid modification of his hypotheses, that were
often really startling, and made him, in his own peculiar field, ivithout a rival
in his day.
* Euler gave I58t+591=1344+133'* as an example. For otter solutions see L. E.
Dickson, History of the Theory of Numbers, Vol. 2, pp. 644—647.
xxxvi Srinivasa Ramanujan

It is often said that it is much more difficult now for a mathematician to

he original than it was in the great days when the foundations of modern
analysis were laid; and no doubt in a measure it is true. Opinions may
differ as to the importance of Ramanujan’s work, the kind of standard by
which it should be judged, and the influence which it is likely to have on the
mathematics of the future. It has not the simplicity and the inevitableness
of the very greatest work; it would be greater if it were less strange. One
gift it has which no one can deny, profound and invincible originality. He
would probably have been a greater mathematician if he had been caught
and tamed a little in his youth; he would have discovered more that was
new, and that, no doubt, of greater importance. On the other hand he would
have been less of a Ramanujan, and more of a European professor, and the
loss might have been greater than the gain.
 1

v/SOME PROPERTIES OF BERNOULLI’S NUMBERS

(,Journal of the Indian Mathematical Society, nr, 1911, 219—234)

1. Let the well-known expansion of ®cota; {aide Edwards’ Differential

Calculus, § 149) be written in the form

* oot a = 1 - (2xf - (2x)‘ - (2xf -.(1)

from which we infer that may be supposed to be — 1.

2 ■ Some Properties of Bernoulli’s Numbers

2. We know (x cot xf = - ^1 + )•

Using (1) and equating the coefficients of x11 on both sides, and simplify-

£ (n +1) Bn = c, B, + c4 B, Bn.t + c„ B, +...,

the last term being cj»_i or £op, (Biny according as £» is odd or
even.....\o) .
A similar result can be obtained by equating the coefficients of x’* in the
identity
dtan« ,
—? = 1 + tan3*.
ax

— \x (cot \x + coth £«) = — £« (cot £* + i cot £ra)

by using (1). The expression may also be written

, (cos ix sm \ix + i sin \x cos £ix)
2 sin £* sin £i®
, (1 + 0 sin j* (1 + 0 - (1 - i) sin ix (1 - i)
2 cos £*(1-i)-cos£;r(l +i)

x1" ’
2! 22.6!+2tl0!
by expanding the numerator and the denominator, and simplifying by
De Moivre’s theorem.
x a?
Hence 2 (jB0 + Bt |j + B, |j + ...) = - x g-~.1..(6)
I! — 2^61+ '"
Similarly
- ix (cot ix - coth ix) = 2 (b, £ + B0 £ + B10 ^ + ...)

= isi(l-*) sin £®(l+i)-£(l + i) sin ix (1 -i)

cos£a;(l + i)-cos£®(l-i)

2.3! 2>.7! + 2'.ll!+-

a? x° ■0)
2!— 2*761 2*TTo1 ''1
 Some Properties of Bernoulli’s Numbers 3
Proceeding as in § 1 we have, if n is an even integer greater than 2,
+ + or^(-l)i'-=0, (8)

according as n or n — 2 is a multiple of 4.
Analogous results can be obtained' from tan $x ± tanh |a.
In (2), (3) and (4) there are terms, While in (5) and (8) there are |-;i
or 4 (n — 2) terms. Thus Bn can be found from only half of the previous B’s.

4. A still simpler method can be deduced from the following identities.

If 1, co, css1 be the three cube roots of unity, then
4 sin x sin xco sin xco2— - (sin %x + sin 2ioj + sin ixco1).
as may easily be verified.
By logarithmic differentiation, we have

Writing for x,
- Jas (cot \x + ® cot ^xco + a- cot ixa2) = -
and, proceeding as in § 3, we get

Multiplying both sides by - \x (co2 — co) and adding to the corresponding

sides of the previous result, we have
i cos x + ms cos xco + co cos xco2
-1* (cot $x + co- cot %xco + co cot \xco2) =
Hence, as before,
xi x'° X16
, 4! _ lol + ltf!- "■
.(10)
'x fL- '
3! 9! + 15!
Similarly
- x (cot Ja + cot \xco + cot \xm2) =
and therefore

6! 12! + 18!
...(11)
af _ af aP_ _
3! 9!+15!
4 Some Properties of Bernoulli’s Numbers

Multiplying up and equating coefficients in (9), (10) and (11) as usual,

we have,
c3S„_3-c„-B„_9+ c15S1M5-... = 0,.(12)
the last term being \n (-1)* ,’MI, $n (- 1)* f"+I,) or %n (- 1)J
Again, dividing both sides in (10) by os and differentiating, we have

of o* ocu a:' xw _
Fi~Sl+ U!-'" 4!~10l + T6l~
x‘ 0B ^ xls ' a? as” ^ xlz

Hence by (9) and (10),

Equating the coefficients of xn we have, if n> 2 and vi — 2 is a

multiple of 6,
i (n + 2) Bn = caB„-sBs + e^B^A + c„Btl_„Bls +.(13)
From (12) the B's can be calculated very quickly and (13) may prove
useful in checking the calculations. Th| number of terms is one-third of
that in (4); thus Bn is found from Ba, Bls and Bt.

5. We shall see later on how the B's can be obtained from their pro¬
perties only. But to know these properties, it will be convenient to calculate
a few B’s by substituting 3, 5, 7, 9, .... for n in succession in (12). Thus

8.=-,; Ba-iB, = -±;

-Bio Bla-llBa=-m-,Blt-^lBi B,
5 120’
Bu 5L; Bls-221Bi2 + 2^

Bn
'-T-mi
3 _nB_ i .
2 '°4 552’
 Some Properties of Bernoullis Numbers .5

and so on. Hence we have finally the following values:

B2 =g; -84=35; Bc=W B‘ = W B'°~ 661 5ll=2730:

„ 7 D 3617 „ 43867 174011 ■_834513.
Bu-q; Bu- gjjj-, 708 > 330 -■ 138 '
„ 236304091 D 8553103 „ 23749461029
■l~ 2730 ’ 6 ’ 3 870
„ 8615841276005 „ 7709321041217 D 2577687858367.
-14322-; ^-510 ; 6
„ 26315271553053477373. 2929993913841559.
• 1919190 1 6
„ 261082718496449122051 „
13530 > -• B* 00 ■

6. It will be observed* that, if n is even hut not equal to zero,

(i) B„ is a fraction and the numerator of Bnjn in its lowest terms is
a prime number, .(14)
(ii) the denominator of Bn contains each of the factors 2 and 3 once
and only once, .(15)
(iii) 2’1 (2“ — 1) Bnjn is an integer and consequently 2 (2’*— 1) £n is an
odd integer.(16)
From (16) it can easily be shewn that the denominator of 2 (2“ - 1) BJn
in its lowest terms is the greatest power of 2 which divides n\ and
consequently, if 11 is not a multiple of 4, then 4(2“ — 1) £n/n is an odd
integer.■.(17)
It follows from (14) that the numerator of Bn in its lowest terms is
divisible by the greatest measure of 11 prime to the denominator, and the
quotient is a prime number.(18)
Examples: (a) 2 and 3 are the only prime factors of 12, 24 and 36, and
they are found in the denominators of BV1, BZi and Bx and their numerators
are prime numbers.
(b) 11 is not found in the denominator of Bx, and hence its numerator is
divisible by 11; similarly, the numerators of Bx, Bu, Bx are divisible by 13,
17, 19, respectively and the quotients in all cases are prime numbers.
(c) 5 is found in the denominator of and not in that of Bx, and con¬
sequently the numerator of Bm is divisible by 5 while that of Bm is a prime
number. Thus we may say that if a prime number appearing in n is not
found in the denominator it will appear in the numerator, and vice versa.

§12'
 6 Some Properties of Bernoulli's Numbers

7. Next, let us consider the denominators.

All the denominators are divisible by 6; those of 54,58,512,... by 5; those
of B„ Bn, 5,e,... by 7; those of 5,„, 5!0, 530,... by 11; but those of 5S, 51S,
B,^, are not divisible by 9; and those of Bu, B*,... are not divisible by 15.
Hence we may infer that:
the denominator of 5„ is the continued product of prime numbers which
are the next numbers (in the natural order) to the factors of n (including
■'.unity and the number itself).(19)
As an example take the denominator of B,t. Write all the factors of 24,
viz. 1, 2, 3, 4, 6, 8, 12, 24. The next numbers to these are 2, 3, 4, 5, 7, 9,
13, 25. Strike out the composite numbers and we have the prime numbers
2, 3, 5, 7, 13. And the denominator of Blt is the product of 2, 3, 5,
7, 13, i.e. 2730.
It is unnecessary to write the odd factors of n except unity, as the next
numbers to these are even and hence composite.
The following are some further examples:

8, Again taking the fractional part of any 5 and splitting it into partial
fractions, we see that:
the fractional part of Bn = (-1)*“ (the sum of the reciprocals of the
prime factors of the denominator of Bn} - (— l)8 9 * 11*.(20)
Thus the fractional part of5M = } + J + J + 1^-l=^y;
thatof 5k = =

Indsfon + +
9. It can be inferred from (20) that:
if Q be the g.c.m. and L the L.c.M. of the denominators of Bm and Bn,
then L/G is the denominator of Bm - (-1)1 (”»-») Bn, and hence, if the
denominators of Bm and Bn are equal, then 5m - (- l)lw»-»> Bn is an
integer.(21)
 Some Properties of Bernoulli’s Numbers 7

Emmple: B,t-Bia and B&-Bib are integers, while the denominator

of Bu + B.x is 5. "
It will be observed that:
(1) if n is a multiple of 4, then the numerator of Bn — ^ in its lowest
terms is divisible by 20; but if n is not a multiple of 4 then that of -f — ^
in its lowest terms is divisible by 5; .,.(22)
(2) if n is any integer, then
2(2—!)!^, 2(2--l)2^±i, 2(2—l)|^

are Integers of the form 30p + 1.(23)

10. If a B is known to lie between certain limits, then it is possible to
find its exact value from the above properties.
Suppose we know that B& lies between 6084 and 6244; its exact value
can be found as follows.
The fractional part of B.a = ^ by (20), also Bw is divisible by 11 by (18).
And by (22) Bs, — if must be divisible by 5. To satisfy these conditions Ba
must be either 6137^ or 6192-Jfg.
But according to (18) the numerator of B& should be a prime number
after it is divided by 11; and consequently Ba must be equal to 6192^ or
i< since the numerator of 6137^ is divisible not only by 11 but also
by 7 and 17.
11. It is known (Edwards’ Differential Calculus, Ch. v, Ex. 29) that

***h>h»m.<*•>
where 2, 3, 5, ... are prime numbers.

Also !*• . <25>

For dx=J“ *n_I (e-*** + e~i™ + ...)dx

by (24). In a similar manner

and J x71-- log(l - e-2'*) dx = -

8 Some Properties of Bernoulli's Numbers

Take logarithms of both sides in (24) and write for loge?i! the well-
known expansion of logcr(?i+ 1), as in Carr’s Synopsis, viz.

(n + «log n - n + i log 2,r + ^ - ■■■

-(-i y .(27)

where 0 < 8 < 1, and where

BzpO _B^p_-ffep+2_,
(2p - 1) 2p>i-P~' (2p - 1) 2pn^~l (2p + l)(2p + 2) nv+1

= (”!?=1 log (1 - e-™) dee + i log(1 - <r“>M) dx-...

—UV Cw°g(1~g"ra)fc
_ rff>-log(l-e--*)
- J0 T(l + «f) ^

We can find the integral part of Bn, and since the fractional part can be
found, as shewn in § 8, the exact value of Bn is known. Unless the calcula¬
tion is made to depend upon the values of logue, log010, tt, .... which are
known to a great number of decimal places, we should have to find the
logarithms of certain numbers whose values are not found in the tables to
as many places of decimals as we require. Such difficulties are removed by
the method given in § 13.

12. Results (14) to (17), (20) and (21) can be obtained as follows. We

2et? T (® +1 f T (x + 2)2 + {x + 3)2 ^ (a, + 4)= + '''

111 1 1
'* 6(®’-ir) + 5(;j!5-®) + 7(*'-*) + ll(*u-«) + "'
where 5, 7, 11, 13, are prime numbers above 3. If we can prove that the
left-hand side of (28) can be expanded in ascending powers of 1 jx with
integral coefficients, then (20) and (21) are at once deduced as follows.
 Some Properties of Bernoulli’s Numbers

From (27) we have

tflogrfr + l) 11 1
d-n? (n + l)*"(n+2)2 (« + 3)7
B„ Bt & B,
'2n- ^ ( y nap+i’

• “ f n»« (/* - ~ W. +-

"Jo Vto®'+1 n!i,+:l ' ’ 7 sinh1 ira

= , f. _j£_*>_ = f__dx.
’ Jo rfe-'in' + a?) sinh27r® J„ (1 + a?) smh2 irn.ce

_ ± + ,_
Substituting the result of (29) in (28) we see that
b1 b b
x' of + a.”
_L_+_1_ + _JL_+_L_ + ...
"' 6 (a3 - x) + 5 (of - x) + 7 (a7 - ®)
where 5, 7, 11, ... are prime numbers, can be expanded in ascending powers
11 (a;'1-*)

of 1 jx with integral coefficients.

Therefore Bt - £, - B, - i + $, Be - & + f, -
■which are the coefficients of 1/ar, l/®5, l/*7, .... ■e integers.

, Writing £ + J — 1 for — ^ we get the results of (20) and (21).

Again changing n to \n in (29), and subtracting half of the result from

(29), we have
_J_L_ , _1 _ 1 (23-l)& (23-l)i?4
(n + 1)2 («+2)2 + (n + 3)"2 2)!2 n? + m3

+ + .(30)

< 6 < 1, and also, by (29),

(2*-l
( ' nw+1 J 0(1+®2
Thus we see that, if we can prove that twice the left-hand side of (30)
can be expanded in ascending powers of 1 jn with integral coefficients, then
the second part of (16) is at once proved.
10 Some Properties of Bernoulli’s Numbers

Again from (27) we have

,v ...(31)
= log7i + ^ —2^ + ^ —6^ +8„8- 2pii® ’
where 0 < 8 < 1; and also, by (25),

2pnP Jo (1 + *=)(e2^0 — 1)
from which it can easily be shewn that

-sr2*2*-1) i^+**(2'-1)sr.-2,(s,-1)ss+-

+ (-l)»2^(2*-l)f^5-..., .(32>

where 0 < $ < 1; and also, by (31),

~:1> " fo 2 (1 + a£l i ^

From the above theorem we see that, if we can prove that

can be expanded in ascending powers of 1/n with integral coefficients, then

the first part of (16) at once follows.
13. The first few digits, and the number of digits in the integral part as
well as in the numerator of Bn, can be found from the approximate formula:
login-S„ = (n + i) log,, n -1-2324743503)1 + 0-700120,
the true value being greater by about 0'0362/n when n is great.(33)
This formula is proved as follows: taking logarithms"of both sides in (24),

log, Bn = (n + £) log, n - n (1 + log, 2tr) + J log. 8tt

nearly. Multiplying both sides by log,„e or -4342944819, and reducing, we
can get the result.

14. Changing n to n — 2 in (24) and taking the ratio of the two results,

where 2, 3, 5, ... are prime numbers.

 Some Properties of Bernoulli’s Numbers 11

e see that approaches very rapidly as n becomes

' greater and greater.
From the value of -rr, viz. 3-14159, 26535, 89793, 23846, 26433, 83279,
502S8, 41971, 69399.the integral part of any B can be found from the
previous B: and from the integral part the exact value can at once be
written by help of (20) as follows:

15. From the preceding theorems we know some of the properties of Bn

for all positive even values of n. As an example let us take Bm = W/D.
The fractional part of B414 is by (20). The numerator of Bui is
divisible by 37 and the quotient is a prime number by (18). Again
l°gio-B«i = 630-2433, nearly, by (33). Therefore the integral part’of BUi
contains 631 digits, the first 4 digits being 1751. Again
logi0 N = log* Bm + log10 D = 630-2433 + log10 90709710 = 638-2010
nearly. Therefore JV contains 639 digits, the first four digits being 1588.

* These integral limits are got from a rough calculation of any B from the preceding
B by the formula given in the first column.
12 Some Properties of Bernoulli’s Numbers

Again the numerator of Bm — -^ is divisible by 20; that is to say, if

[S,J is the integral part of Btii, [5«] + *} - jV = [-»«] + $?£?£& has a
numerator divisible by 20. Therefore the integral part of Biu ends with 4
and also the figure next to the last is even.
Hence N ends with 57 and also the third figure from the last is even.

16. Instead of starting with cot® as in §§ 2 and 3, we may start with

tan x or coseo x and get other similar results.
Thus:
(i) i-Bn (2" -1) = ceBn^ (2»-s -1) - cI2 S„_K (2’1~12 -1) + ...
+ or or .(36)

-- M -

= S• + C-D*(—8! or (-l)i(»+o or (_l)Hn+3)}. ..,(37)

17. The formulas obtained in §§ 1, 3, 4 may be called the one interval,

two interval and three interval formulae respectively. The p interval formulae
can be got by taking the pth roots of unity or of i according as p is odd or¬

is of unity (1, a, «2, a?, a4), and find

fa? x's a*
V5! 15! + 25! "
Then it can easily be shewn that

= <t> (2*) - <f> {2.r (a + a4)} - 0 {ix (a2 + a3)}

= 0 (2®) + $ {x (1 + V)} + 0 {* (1 - Vo)}.
Taking logarithms and differentiating both sides, we have

5(&+a4+*4+R4+-)
r,(i +

5 , (1 + «») — jg-j (1 + ««) + • ■ ■

= (c~N~) > so that «,,«„ = «,l+m + (- l)’lo

Some Properties of Bernoulli's Numbers 13
14 Some Properties of Berno ulli’s Numbers

(ii) 4{^11+B”l0!+*,IFl +

r^-mCh + 21!

4 i*2 12 ! “6 + 20 ! “10

(iii) 4 V2 |Bt J-, + 5a B, ^ + ... J

T\k -ZrMw^ —
^\Ch~12\a'a + 20\aw~'"

(iv) 4 \/2 jj?, J, + Bu ~4l + + ... J

a? «C .
9~!^-l7~!68 + 25l^~-

4! s '12 ict» + 20!ai°

 2

ON. QUESTION 330 OF PEOFESSOB SAN JANA

f n >.3. In attempting t< s for all values of n, I have arrived

it the following results :

S=- dz = i O'-1' (1 - a)-i dm

rm
(vide Williamson, Integral Calculus, p. 164).

Therefore ^ - JL _Il£±iL

Therefore log {/(p)} = log (Jit)-p log 2

here 8n = ^ + gj + ... ad inf. (vide Carr’s Synopsis, 2295).

Again, by expanding f(p) in ascending powers of p, we have

= f (</> (0) - p4> (i) +f4> (2) -fi> (3) +...},

16 On Question 330 of Professor Sanjana

1 11 1.31, 7T , . .
where jS+i + g 3^ + 174 W' + - = 2 * W
Hence (1) may be written
log + log [c/> (0) -p . <j> (1) + p2 ■ 4> (2) - ...}
= log(i7r)—jtlog2 + | (1-|)'S2-g-(1-^)lS3 + ---

= log(i7r)-p<r1+^o-2--|o-!1+ ....

where trn = l — ^ ~ ^ + .. • *

Differentiating with respect top, and equating the coefficients ofp*-1,

have
n<f> (n) = a(n -1) + (n - 2) + (n- 3) + ... to n terms.
Thus we see that

2^(1)“1 + 23!+ 2.4 52+'" 2l0g2,

^(2) = 1 + |i + ^|i + ... = ^+f(l°g2)2,

|*(8)Sl + Ji+^|i + ... = S1og2 + gaog2y+^i

= |log2+f2(log2)3 + ^3,

2. More generally, consider the series

1 a 1 , a(q-l) 1_
4» l!(4+l)”i' 2! (6 + 2)« ""

Writing <j>(n — 1) for this, and taking the identity

J_* 1 , «(«-!) 1
b+p 116 + 1+p 2! 4+2 + p

Jo r(a + 6 + p + l)

n<j> (n -1) + °v£ (n - 2) + <rs<\> (n - 3) + ... to

_ 1 1 _ __
1
bn (a + 4 4- l)n (b + l)]l (a -t + 2)n
1
 On Question 330 of Professor Sanjana 17

Exa mples: Put a — — ■&, 6 = i- Then "vve see that

11 1.311.3.51 |ra)]2
(i) 1 + 2 5 +2.4 9+ 2'.4.6 13 + ' " 4V(2tt)’
11 1.31 1.3.51 [r®|2 vr
(ii) 1 + 2 5“s+ 274, 91 + ¥7476 13* + "" 4V(2tt)4’
_ira)PK )
(iii) 1 + 2 5s + O 9» + 27476138 + " ~4 Vl^ir) (32 + 2 -J’

(iv) 1 + iI + I^I + LL«I +

1 + 2 54 + 2.4 94 z. 4.6 134 +

where, ; 1+1-1+....
 3
NOTE ON A SET OF SIMULTANEOUS EQUATIONS*

(Journal of the Indian Mathematical Society, iv, 1912, 94—96)

1. Consider the equations

x1yl + x3y3 + x3y.3+..

x1yf + x./yi + xsyf + ...
xlyf + x2yi + x,yf+...

here a-„ x,, x3, ... x„ and y,,y,.y„.

New, let us take the expression

* ' ' 1 — 8y1 +1 — By., 1 — Bye " 1 — 9y„

and expand it in'ascending powers of 6. Then we see that the exprei
equal to

But (1), when simplified, will have for n expression of the

(n- l)th degree in 8, and for its denomina i of the nth degree

Thus we may suppose that

i lff, __ _A1±_At6 + A,8‘+...+A*6~
9w-1 + Bj + B,# + B,J3 + ... + Bn8"“

id so (1 + B18 + ...)(a1+a28+...) = Ai + A28 + .

Equating the coefficients of like powers of 6, we have

A3-a3 + aiB3 + a^B,,

An = On + On-i-S, + a„B, + ... + a.B,^,

o = <hn + o*-! JS, + ... + anBn.

* for a solution, by determinants, of a similar set of equations, see Burnside and
Panton, Theory of Equations, Tol. n, p. 106, Ex. 3. [Editor, J. Indian Math. Soc.]
 Note on a set of Simultaneous Equations 19

From these Bu jB2, ... Bn c easily be found, and since A, An

depend upon these values they also be found.
Now, splitting (3) into partial fractions the form

i-qiS 1 -}.«
and comparing -with (X), w :e that

2. As an example we may solve the equations:

cc+y+z+u+v — 2,
px+ qy + rz + su + tv = 3,
psa; + g,ty + »J* + «Ju-t-£1» = 16,
p3» + 2;,2/ + r^ + s3« + t3i)= 31,
p*x + q'y + J'% + sAu + t*v = 103,
p5® + y-y + r^z + s3u + Vo = 235,
pn:c + qey + r"z + s‘u + t‘v = 674,
pyx + <fy + rz + s7u 4- tfv = 1669,
p8® + <fy + r*z + ssu + Ifv = 4526,
p'te + qay + iaz + s°u + t°v = 11595,
where as, y, z, u, v, p, q, r, s, t are the unknowns. Proceeding

1 -8p + 1 -6q + f- 8r + 1 - 6s + 1 - 6t
=2 + 30+ 16d2+31ds+103dJ+235ds+6745“+166967+45260“+115950°+_
By the method of indeterminate coefficients, this can be shewn to be

1-0-50s-
Splitting this into partial fractioi get the values of the unknowns.

p= -1

2^5 ’
8-VS.
2 V5 ’
 4
IRREGULAR NUMBERS

(,Journal of the Indian Mathematical Society, v, 1913, 105—106)

1. Let o3, as, os, a,, ... denote numbers less than unity, where the sub¬
scripts 2, 3, 5, 7, ... are the series of prime numbers. Then

+ o,.as + a7 + oa.aa.o» + a3.a3+...,.(1)

the terms being so arranged that the products obtained by multiplying the
subscripts are the series of natural numbers 2, 3, 4, 5, 6, 7,. 8, 9, —
The above result is easily got if we remember that the natural numbers
are formed by multiplying primes 'and their powers.
2. Similarly, we have
1 1 ' 1 ,
r+"cT' ~i + i ' UTa ... = 1 — Qa — Oj + Oj. <ta — as
+ aJ.aa-a7-a,.aj.a,+ o,.a,+.(2)
where the sign is negative whenever a term contains an odd number of prime
subscripts.
3. Put 02 = 1/2’*, a3 = 1/3”, as = 1/5", ... in (1), and we get

.w
where Sn denotes 1/1’*+ l/2’* + l/3“ + 1/4"+ ....
Changing n into 2?i in (3) and dividing by the original, we obtain

K)(’U.)(*4)(*GL-4-.(«
Examples: (i) (l + i) (l + i) (l + I) .., = , .,5) «

.-<«)
since S, = w=/6, Sa = ^/0O, S8 = ws/9450.

4. Subtract (2) from (1) and put a2 = 2~" ...; then

1 + I + L+I + I+A+ * _Sn,~3m
2" + 3» + 5“ + 7’‘+8» + H» + 12"+-’
where the numbers 2, 3, 5, 7, 8, ... contain an odd number of prime divisors.
 Irregular Numbers 21

...(7)

-.(8)
(n) 24 + 34 + o4 + 7*+ 84 "*
5. Again (2, 3, 5, 1, ... being the prime numbers)
(1 + O (1 + 0(1 + O (1 + O ... = 1 + «.+ a. + a«
-f a2. a3 4- ct7 4- a?. ci6 4- Mu “1 ai3 + * * - j * “(9)
where the product of the subscripts in any term is a natural number con¬
taining dissimilar prime divisors; and
.(1-0(1-0(1-0 (i-0... = i-o»-fl»-“»+tt>-a>-a"-(10)
where the signs are negative whenever the number of factors is odd.
6. Replacing as before as, a,„ as, ... by the values given in § 3 and
using (4), we deduce that

where 2, 3, 5, 6, 7, ... are the numbers containing dissimilar prime divisors.

7. Also taking half the difference between (3) and (4),

^ + ^+5»+7» + Il” + TF* + l73* + l9”+23i + 295

_1 ,_1_, .(12)

where 2, 8, 5, ... numbers containing an odd number of dissimilar prime

divisors.
Examples: « J- .(13)
217^’
15
(ii) J- .(14)
27T1'
8. Subtracting (11) from we have

e composite numbers having at least two equal prime

 j SQUARING THE CIRCLE
(Journal of the Indian Mathematical Society, v, 1913, 132)
Let PQR be a circle with centre 0, of which a diameter is PR. Bisect
PO at if and let T be the point of trisection of OR nearer R. Draw TQ
perpendicular to PR and place the chord RS = TQ.
Join PS, and draw OM and TN parallel to RS. Place a chord PK ^ PM,
and draw the tangent PL = MN. Join RL, RK and KL. Cut off R0= RH.

Note.—If the area of the circle be 140,000 square miles, then RD is

greater than the true length by about an inch.
 6
MODULAR EQUATIONS AND APPROXIMATIONS TO tt

(.Quarterly Journal of Mathematics, XLV, 1914, 350—372)

1. If we suppose that
(I + (1 + e-’**") 0 + «-"»)... - 2i G.(1)
and . = .(2)
then (?„ and <jn can always be expressed as roots of algebraical equations
when n is any rational number. For we know that
(1 + q) (1 + q>) (1 + q?)... = 2^* {kk') ' A .(3)
and (1 — 9)(1 — qs)( 1 - qs)... = 2*q*k~*k'i.(4)
Now the relation between the moduli k and l, which makes
A'_I/
11 K~ L'
where n=rjs, r and s being positive integers, is expressed by the modular
equation of the rsth degree. If we suppose that k = l’, k‘ = l, so that K — L’,
K‘ = L, then
3, = e-rfVi = e-w„)
and the corresponding value of k may be found by the solution of an alge¬
braical equation.
From (1), (2), (3), and (4) it may easily be deduced that
9m = $ gnQn,.(5)
1 l9n = giln.(6)
(,gA'f{Gn*-gn‘)=i.(7)
I shall consider only integral values of n. It follows from (7) that we
need consider only one of Gn or yn for any given value of n; and from (5)
that we may suppose n not divisible by 4. It is most convenient to con¬
sider gn when n is even, and G„ when n is odd.
2. Suppose then that n is odd. The values of Gn and g.ln are got from
the same modular equation. For example, let us take the modular equation
of the 5th degree, viz.
©•♦©•-*(—.y..... .(8.)

where 2= (1 + q) (1 + tf) (1 + 5*) ...

and 2*q*v = (1 + tf) (1 + 2“) (1 + g*)
24 Modular Equations and Approximations to n

By changing qto-q the above equation may also be written as

...»
where 2^ = (1 - g) (1 - qs) (1 - g5) • • •
and 2*q*v = (1 - g!) (1 - g“) (1 - <lx) ■ • ■ •
If we put q = e-*"5 in (8), so that u = Gi. and v = G,„ and hence u - ®, we see
that ^

Hence =

Similarly, by putting q = e~*'Ji, so that tt = £r, and v = gu, and hence

u = 1 jv, we see that , ^

Hence . ?>■ = ./) •

Similarly it can be shewn that

3. In order to obtain approximations for it we take logarithms of (1)

and (2). Thus
w= 5il°g(2ieii))

approximately, the error being nearly — eT" "n in both cases. These equations
may also be written as .
e”'B/“ = 2* (?„, —2i^r„.(H)
In those cases in which Gn'- and gna are simple quadratic surds we may use
the forms
«?»«+<?»-«*)*
instead of Qn and gn, for we have

approximately, and so
gnu + gn~ls = iei’r"n +^e~^'/n,
 Modular Equations and Approximations to tt 25
approximately, so that
= -4 log (8(ff*“ + ff,ra)}, .(12)

the error being about e-”'’1, which is of the same order as the error in
6 V»
the formulas (10). The formula (12) often leads to simpler results. Thus
the second of formulae (10) gives
= 2i?1!
or =10^2 + 8 V3.
But if we use the formula (12), or
= 2 %,,1!+</„-12)tV,
we get a simpler form, viz.
e^s/l8 = 2 ^7.

4. The values of gsn and Gn are obtained from the same equation. The
approximation by means of gm is preferable to that by Gn for the following

(a) It is more accurate. Thus the error when we use <r6s contains
a factor e-’”'65, whereas that when we use gm contains a factor
(b) For many values of n, gm is simpler in form than (?„; thus

fa=yj(H^)(^i3)|,
while

Mmmr y{y(“)vmi'
(c) For many values of n, jr2u involves quadratic surds only, even when
£?„ is a root of an equation of higher order. Thus G,a, 6a, G?A are roots of
cubic equations, G„, G7„ are those of quin tic equations, and Gn is that of a
septic equation, while gK, gx, gg,„ g™, and gm are all expressible by
quadratic surds.
5. Since (?„ and gn can be expressed as roots of algebraical equations
with rational coefficients, the same is true of (?„!J or gn!i. So let us suppose
that

But we know that

64egn3i = 1 - 24e-’"/» + 276e~2,r''" - ...,
64g^ = en "n - 24 + 276r"r''u - ...,
64a - 64bgn~*‘ +... = e’"n - 24 + 276s-"'” -.. -,
64a- 4096he-’"'" + ... = e"'"- 24 + 276s-"'“- ....
that is e”,n = (64a + 24) - (40966 + 276) s-"'” +.(13)
26 Modulcur Equations and Approximations to it

Similarly, if X = aGn~!i - bG,^ + ...,

then e"'n = (64a - 24) - (4096& + 276) e-"'" +.(14)
From (13) and (14) we can find whether e",n is very nearly an integer
for given values of n, and ascertain also the number of 9s or Os in the
decimal part. But if (?„ and gn be simple quadratic surds we may work
independently as follows. We have, for example,
g»=“J( 1 + V2).
Hence 64feM = e*"* - 24 + 276e-"22 - ...,
64<far“= 4096fr'
so that
64(ga» + ga~*) = e*"22- 24 + 4372e-"/22 + ... = 64 {(1 + V2)12 + (1 - V2)12).
Hence e”"22 = 2508951-9982....
Again e37=(6 + v'37)1,
64 Gy*1 =e"'27 + 24 + 276e-’"'2’+...,
6iG„~a = 409 Ge-’"'35 — ...,
so that
64 (<?„« + Gy21) = e""3’ + 24 + 4372<r"'" - ... = 64 {(6 + V37)6 + (6 - ^37)“}.
Hence ' e’r'/K = 199148647-999978....

Similarly, from —-^r~) >

we obtain
e"'“ - 24 + 4372e-'-

=«WT+N^)”}-
e’”,ss _ 24591257751-9!

6. I have calculated the values of Gn and gn for a large number of values

f n. Many of these results are equivalent to results given by Weber; for

q j _ 3 + V13 y _ 1 + V5

g»' = (2 + VS) (3 + V10), GV = 6 + ^37,

g& 7i + V(4 + V7) 5 + V29

,(3 + v'5)(l + V2)

S'™

©73

:^l + V5^ ^9 + V85j*

<?«
 Modular Equations and Approximations to tt 27

*
W =(2 + V5)(3 + V10),
GV = J (3 + VH) (V5 + V7) (V7 + VH) (3 + V5),
and so on. I have also many results not given by Weber. I give a com¬
plete table of new results. In Weber’s notation, &n = 2"+/(a/(— »)! and

Table I.
9m +^ = HV(1 + V2)+V(9+5v'2)},

=y{mmi lymvmi'
g*' = vV2 + V3) (7 V2 + 3 vnyi ■

<V = ft (V7 + Vll) (8 + 3 V7)}i [J )} .

G a _(2V3 + 2)Ul
“ (2 V3 — 2)1 — 1 ’

g« ={(2+V5)(^5+V6)J- (yp^) + >

g« +^iw(i+^)+da+5dm

g« +J—11^2 + ^(14 + 4^14)},

On, = \ (2 '/3 + V13)i i3^ + V (4 + V'3)!>

^*c-CT)‘{('+ra)‘+(-*W}
^ - 3^2 [(11 - 3 VH)J KWH + 3 V3 - 4)4
+ (3 Vll — 3 V3 — 4)*} — 2],
28 Modular Equations and Approximations IT

_y(3V3 + V23)(78v2 + 23V2g),

9m

«{yc-v,? vjm-
Gu .,<W3W„,f_±*)y(Hi±
^♦yew
Gu,

Gm

y<^
9m = v/{(2V2 + V7 )(^yU)[

X {\/(13+4'/22)
9m + ^; = i tV(9 + d%) + V(17 +13 \/2)),
V22)J

Gm
+ i'(?) {(1 + 37s)I + (1-37») T 1

G^, l.

)1]l
9m = V(1 + V2) (4 V2 + V33)* |v/(i+^/?3’
V(‘--s ®)}.
Gm
V(^F-1)}' .
Gm> = (5 ^3 + V71)i (p + 1i W'j

yyc-^y yp#*)}■
Gu■? = | //9 + 4V7N //11 + 4V7N)
IV l 2 J + V(-2-jf

«k/e-w yrv*)}•
<?„» ■ = (1Jr£) (2 + V3)* 1V(4 + VI0) + 15*},
 Modular Equations and Approximations to -n 29

-WW-

yymvc-^)}-
y^ psiiypnvpr’))'

, k+8,7)(?wiyi))*
x(^(«±^) + v,(12±WL8)ji

^ J (6 + V37)1 (7 pz + 2 V37)i ypi + 2 V3) + V(3 + 2 V3)),

3S3 = 2* t, where j
2i» _ p {(4 + V33) + V(H + 2 V33))
-i(l + V(ll+2V33)}-l = 0)
, , A/3+V7y.?x ,.g (2 + V7+V(7 + 4V7)| (V(3 + V7) + (6 V_7)j)
r«‘“V—2“V M 2 J V(3 +V'7)-(6 V7)iJ’

- y {(S + ',S) <T^) (^))"+ 2 ^

yyc^wm
yy^wp^8-1)}'
3»-- (2+v5> y jf^y5) a®+vxoi)}
x |^5y5 + yi01^ + ^OB + VBOB^'t

,,V(^)(i,2»+nvo,‘|yf-±r?)+y(^)},
 Modular Equations and Approximations to ir

-jy^iwLvyC00^)!

. (VH+vis)* y{(i+vz) (3-ii2) (^y-7)}

yy(^+!)V(!?1£±y^)}
*jyc'i5y7+4)+yp^^y^4)!.
- (t^) <ib+y {<*+vis) y±i?i))
„{v)(!5vovs1) + y(?1±a_vi.)}

= yj(Z + V3) (6 + V37) \j" Z'7jf (240 V7 4- 107 V37)l

yymv^)}
yye^-yt1^)}.
(^)(0 + vSs).{5±kSW!>J'
x j"[43 + 15 V7 + (8 + 3 V7) y(l0 y7)|

_l_ ^/j35 + 15^7 + (8 + 3 *J7) y(10 y7)jj

yj(3 + yil)(5 + 3 v-3) (—+/12-)j

x ^6817 + 321 V451

yyt^jv?^-8)}
X |v'/(561+g9V33] | yy9 + 99V33jj

(2 + V5) y j(3 + V7) pyS + 9V889j*

x {y(U9+y329) + y^12^+7^329^1
s {yfinsva?)+.
 Modular Equations awl Approximations to it 31

7. Hence we deduce the following approximate formulae.

Table II.
ei* "» = 2 qi, e" *«• = 2 + ^2, e* *'»> = 20 V3 + 16 V6,
gin ft St = 12 (4 + qi7)' ginr/iC = 144 (147 + 104 ^2),
_ 5 + ^29
ei”'“ = 84 + 32 V6,
V2'
= 60 V35 + 96 V14, el"/7* = 300 V3 + 208 ^6,
.1
<r= + ^+.2^g), = 800 V3 +190 ^/5i>

eW™ = 12 (323 + 40 V65), e"1^ = (2 V2 + ylO) (3 + .\/10),

«■ = viiolog t(2 V2 + V10) (S+V10)1>

TT = log [i (3 + V5) (2 + V2) {(5 + 2 vTO) + V(61 + 20 V10))],

”7i2l»s[(i:75L9)'<5''“ + u’,6)

The last five formulae are correct to 15, 16, 18, 22, and 31 places of decimals
respectively.

8. Thus we have seen how to approximate to ir by means of logarithms

of surds. I shall now shew how to obtain approximations in terms of surds
only. If

n~K~L'
_cU
ifiiy
But, by means of the modular equation connecting k and l, we can
express dlc/dl as an algebraic function of k, a function moreover in which all
coefficients which occur are algebraic numbers. Agaiq,
q = e~”K'IK, q* = e-*J
gftq-g'Xi-tfKi-g*)-
?*«(l_?m)(i_r)(i_r)... @7©.;<->
32 Modular Equations and Approximations to -rr

Differentiating this equation logarithmically, and using the formula

d<2

where d (k) denotes an algebraic function of the special class described

above. I shall use the letter A generally to denote a function of this type.
Now, if we put k = V and k' = 1 in (16), we have

- +(f)‘A <*>
The algebraic function A (Ic) of course assumes a purely numerical form
when we substitute the value of k deduced from the modular equation. But
by substituting k = V and V — l in (15) we have
nie-^Q "»)(1 -tr"*) (1 -«—**) ...
= trws* (1 _ e-^n) (1 _ e-^n) (1 _ .. .
Differentiating the above equation logarithmically we have

Now, adding (17) and (18), w

+ -)“(-) M*>)-.(19)

...) = (f)(l-2n

1 24 (e^rfn+ J + g»*i + X +

Hence, dividing (19) by (20), we have

i_ Jo4r i .
.(21)
 Modular Equations and Approximations to ir 33

where R can always "be expressed in radicals if »i is any rational number.

Hence we have

(l-R)dn.v~'
nearly, the error being about &Tre~*'‘n(w *Jn - 3).
9. We may get a still closer approximation from the following results.
It is known that

240 ,.!i (1_w^

= ^a-so4 s --i
«»»•“»_if*.
where R' can always be expressed m radicals for any rational value

...(24)
(1 -R)*Jn.
nearly, the error being about 24ir (107t pn - 31) e-5’”'”.
It will be seen that the error in (24) is much less than that’in (22), if n
is at all large.
10. In order to find R and R' the series in (16) must be calculated
in finite terms. I shall give the final results for a few values of n.

Table III.

/(*)=■»(!-24 fr-’^-fi-24!!^)'
/(2) = ^(*' + 0.

/(3) = ^(i+ « + *'*').

m -^ (3 + kl + k'V) y(i + w2+--0.

/(7) = i^x(1+H+m
34 Modular Equations and Approximations to tt

/(ll) = S;^ (2 (1 + kl + k'l') + d(ld) + d(k'V) - d(hk'U%

/(lo) = H/-[(l + Qdf + iVVfY -11 +1:1 + k'l'\l

f( 17) = -4M V{44(1 +m + lc‘-P) + lQlB,{kl + h'l'- Ids'll')

- 102 (1 - kl - k'l') (ikMUf - 192 (4,ld</tt')$\,

f(l 9) = 1(1 + kl + k'l') + q{kl) + d{kfV) - d(kk'll')},

/(23) = [11 (1 + kl + k'l')- 16 (4klc'U'f (1 + d{ld) + d(k'l')}

- 20 (4kk'll'^l
/(31) = [3 (1 + kl + k'l') + 4 [V(fcZ) + dik'V) + q(WU’)}
- 4 (Ids'll')i (1 + (ld)i + (*T)*)],

/(35)[2 y(ki) + d(k'V) - q(kk'U')}

+ (4m1 - d(H) - d(k'l’)Yl
Thus the sulh of the series (19) can be found in finite terms, when
n~ 2, 3,4, 5, from the equations in Table III. We can use the same
table to find the sum of (19) when n = 9, 25,49, but then we have also

which is got by putting k = !•' = 1//2 and n = 1 in (18).

Similarly we can find the sum of (19) when n = 21, 33, 57, 93, ..., by
combining the values of /(3) and /(7),/(3) and/(ll), and so on, obtained
from Table III.
11. The errors in (22) and (24) being about
8ire-'*'’1 (it aJn — 3), 24t (IOw - 31) e-3'*'’*,
we cannot expect a high degree of approximation for small values of n.
Thus, if we put n = 7, 9, 16, and 25 in (24), we get

j|v7 = 3-14180...,

J(l + f) = ,14l62..„ .
99 r 7 \
8o(r-T^)=31413a374-’
36 Modular Equations and Approximations to tt

(2) Let AS (Fig. 2) be a diameter of a circle whose centre is 0.

Bisect the arc AOB at C and trisect AO at T.
Join BG and cut off from it CM and MN equal ^

Join AM and AN and cut off from the latter / \

AP equal to AM. /z\L
Through P draw PQ parallel to MN and meet- Arro ‘ JE
ing AM at Q. )
Join OQ and through T draw TR, parallel to s\ /
OQ, and meeting AQ at R. \-
Draw AS perpendicular to AO and equal to 1Jig 2.
AR, and join OS.
Then the mean proportional between OS and OB will be very nearly
equal to a sixth of the circumference, the error being less than a twelfth of ar
inch when the diameter is 8000 miles long.
13. I shall conclude this paper by giving a few series for 1/w.
It is known that, when k § 1/V2,

(tJ)* ■-1+(§y my+(o)’ my+.(25)

Hence we have

- (i i1 + 8)’ mj + (o)S + ■ ■ j.(26)

Differentiating both sides in (26) logarithmically with respect to lc, we can
easily shew that
1-24 (^- + -^ + 7^■+...)
VI “S’ 1-9 1-96 I
=a - 2*0 {i+* (!)! mr+7 my(27)
But it follows from (19) that, when q = n being a rational number, the
left-hand side of (27) can be expressed in the form

where A and B are algebraic numbers expressible by surds. Combining

(25) and (27) in such a way as to eliminate the term (2K/tt)-, we are left,
with a series for 1/w. Thus, for example,

= = .(28)
16 = 5 47 ay 89 /TSy 131 (1.3.5\»
w • + 64 y 64= (.2. J + 64= 127476 + ->
.(29)
 Modular Equations and Approximations to ir 37

[5 = ,—, .(30)

here 5 \/5 —1, 4-7 \/o + 29. 89V5 + 59, ... are in arithmetical progression.

14. The ordinary modular equations express the relations which hold
between k and l when nK'jK = L'jL, or qn — Q, where
? = (r*A'7ff Q=e-„L-/L

There are corresponding theories in which q is replaced by one or other

of the functions
Is =
1.3.5. 1.3.5.7.9.11,.
if, = 1 +

if2 = 1 + ^ ** + -•'sH.'“ B + 3,.e.*.9i“ + '"

1.5.7.11.13.17
^.-i+^** + -7STir /c‘ +

From these theories we can deduce further for 1/ir, such

.,.112
£—m
...(31)

■**»Si5Gi)
15 V3
2tt
...(32)
Kri)1
5^5
2rr vdi'
+ 23 ..(33)
2.4 6.12 6.12 G25) +'
80 V85
18ttV3 2 6 6 (857

+274oF^d-:Ste)‘t--'-w
 Modular Equations ancl Approximations to n

_ 1123 22583 1 1^3 44043 1.3 1.3.5.1

= 882 8823 2 ~¥' + 8826 2.4 4s. 8s

3 43 1 1 ■ 3 , 83 1.31.3.

In all these series the first factors in each term form an arithmetical
progression; e.g. 2, 17, 32, 47.in (31), and 4, 37, 70, 103.in (32).
The first two series belong to the theory of the next two to that of q.„
and the rest to that of <?,.
The last series (44) is extremely rapidly convergent. Thus, taking only
the first term, we see that
1103

+ JSL \
'!-? 1-2" J
sd in §§ 8—13, is very closely a
perimeter of an*ellipse ■
major and the semi-min

p = 2 ira j 1 - ^ k- — —-^-2 ¥—{

 Modular Equations and Approximations to ir 39

where p is the perimeter and k the eccentricity. It can easily he seen

from (45) that
p = idle'3 |a" + .(46)
But, taking the equation
2* (1 - 2*) a - 2*) (1 - 2") • ■ ■ = (Mb')* VWO,
and differentiating both sides logarithmically with respect to k, and com¬
bining the result with (46) in such a way as to eliminate dK/dk, we can
shew that

^^[jr,(1+i',)+(*’r),{1-M(r?? + i^+-)}]- <47>

But we have shewn already that the right-hand side of (47) can be expressed
in terms of K if q — e-’"'”, where n is any rational number. It can also
be shewn that K can be expressed in terms of T-funetions if q be of the
forms e~’n, e-’r'*''2 and e~’r“'/3, where n is rational. Thus, for example, we

i = sin3>

16. The following approximations for p were obtained empirically:

p = 7T [3 (a + b) - V{(o + 36) (3a + b)} + e],.(49)
where e is about «/d2/104S576 ;

y = 7r{(a + t) + 10(a + 6) + v'(a3 + I4a6 + ia) + e}’.^

where e is about 3ai;!“/68719476736.
 7
ON THE INTEGRAL tdt

(Journal of the. Indian Mathematical Society, vir, 1915, 93—96)

1. Let <b (to) = J (tan^ dt.(1)

Then it is easy to see that
*(«) + *(-«)- 0;. ..(2)

and that <£ (x) = p — ^ - p +.(3)

provided that j ®j ^ 1.
Changing t into 1/t in (1), we obtain

.(4)

provided that the'real part of .* is positive.

The results in the following two sections can be very easily proved by
differentiating both sides with respect to x.
2. If 0 <x< \tr, then
sin 2® sin 6aj sin 10;c , , ,
-p- + + — g, - + ■•• = <t> (tan x)~x log (tan x).(5)

If, in particular, we put x •= %ir and r in (5), we obtain

p-j2+g2-^+-=72^('/2-1>+s52loff(1+'/2>+T^;
and 2<j> (1) = 3 <j> (2 - a/3) + j7rlog(2 + \/3)....(7)
If — 17r < a < ^ 7r, then

24> (tan 1) = * + g ^ + §75 ? +.(8)

If 0 < x < £7r, then

sin a; ^ ^ + sin2z ^ ^ sin 3* ^

= <f> (tan *) 4- J7r log cos x — x log sin x ; .(9)

and °osx + amx Icos^ + sird* US eo*_® _+ sin*.
1! T2 3» + 2.4 5s +"’
= 0(tana) + J7rlog(2cos®).(10)
 An Indefinite Integral 41
If - \tt<sc<\ 7r and a be any number such that

| (l-a)sina|<l, ; ^1 - cos» I < 1,

= <£(tan x) - <f> (a tan x) + x log a. (11)

3. Let R (x) and I (x) denote the real and the imaginary parts of x
respectively. Then, if - 1 < R (x) < I,

log (l - p) - 3 log (l ~ |) + 5 log (i-|)--

“4 Ift (1) - * !tan i"- (1 ~ *)}] + log taa i* (1 - »)■ -(12)

Putting x = | in (12) and using (7), we obtain

where a = ^<f> (1).(13)

Again, subtracting log (1 — x) from both sides in (12) and making x—* 1,
we obtain

(i-S"3(i-i)5(i4r(i-^-=iein.<i4)
If — 1 < I {x) < 1, then

log (l + Ji) “ 3 log (l + |) + 5 log (l + |)-...

=4 ft (1) ~ 0 - 2*tan"’ .(13>

From this and (7) we see that, if \ttx= log (2+^3), then

.<■«

where n is the same as in (13).

It follows at once from (12) and (15) that, if
- 1 < .K (/3) < 1, -1<I(«)<1,

*** .(17>
provided that cosh ^ira = sec Jirfi.
Aii Indefinite, Integral
 An Indefinite Integral

5. Remembering that
v _1 _ JX_ 5_
4 cosh irm X2 + 4ns5 32 + 4j.- + 5- + 48s
<e have
ZZ v___""£° {_ ^__§_i_
4 i ri1 cosh imas ,,^1 (i?s (Is + 4«2,i:2) n! (3* + 4»2a?)

=Zf£+___^coth^ coth£+cothi

■ That is to say, if a and 8 are real and a/3 = 7r2, then

(1) + 20 (e-») + 20 («-“) + 20 (e'3*) + ...

8 (s’1-2) 48 {l2 cosh 8 + 22 cosh 28 +

If, in particular, we put a = 8 = tt in (26), we obtain

^(1) = 48 “ 2 flV-1) “ +

2 112 (e- + e~") + 22 (e“ + «-») + 32 (es + e--3') + '" j

= •9139655942, .(27)
approximately.
 8

/on the number of divisors of a number

(.Journal of the Indian Mathematical Society, vn, 1915, 131—133)
1. If S be a divisor of A, then there is a conjugate divisor S' such that
88' = If. Thus we see that
the number of divisors from I to \/J\r is equal to the number of divisors
from \/N to iV. .(1)
From this it evidently follows that
d(N)<2M .(2)
where d (N) denotes the number of divisors of JV (including unity and the
number itself). This is only a trivial result, as all the numbers from 1 to >JN
cannot be divisors of N. So let us try to find the best possible superior limit
for d (N) by using purely elementary reasoning.
2. First let us consider the case in which all the prime divisors of N are
known. Let
N=pf>. .pf>... p„a*,
where p,, p.2, p3 ...pn are a given set of n primes. Then it is easy to see that
d (N) = (1 + of) (1 + of) (I + a,)... (1 + an). .(3)

But 1 {(1 + a0 log p, + (1 + a2) logp,, + ... + (1 + an) logp,,)

> {(1 +a,) (1 + a,)... (1 + On) logPj logp2... logp„}“, .(4)

since the arithmetic mean of unequal positive numbers is always greater than
their geometric mean. Hence

i [log ^ + logp, + ... + logpn + log A] > (logp, logp2... log pn. d (A)j».

In other words
. |-log(fhMh ...pnA)}‘
■ d(F)< --i-.(5)
logp,logpslogp3... logp,i w
for all values of N whose prime divisors are pup2,p3 ...p„.
3. Next let us consider the case in which only the number of prime
divisors of N is known. Let
N-=...pn%
where n is a given number; and let
A' = 2“>.3“*.5“* ...p°».
 On the Number of Divisors of a Number 45

where p is the natural nth prime. Then it is evident that

2T « N-, .(6)
and ./uYW.-Vi...(7)
log (2.3.5...p.i\r')j-
■But <l ^ ■* < log 2 log 3 log 5 ... log p ’ .^
bv virtue of (5). It follows from (6) to (8) that, if p be the natural «th
prime, then
log (2. 3.5 ... p . -Ao}"
^ ^ ^ < log 2 log 3 log 5 ... logp ’ .^
for all values of JY having n prime divisors.

4. Finally, let us consider the case in which nothing is know about lY.
Any integer N can be written in the form

where «»>0. Now let

.(10)
where k is any positive number. Then we have
d(N) 1 + «2 l + os 1 + a,
.(11)
A» 2'“= -" 3*“» ' 5*®* '
But from (10) we see that, if q be any prime greater than x, then

It follows from (11) and(12) that, ifp be the largest prime not exceeding

d(N) < 1+_oa 1 + «, 1 + a, 1_+ Up

jYa " 2*“* ' 3*“s ' oM‘ phap

But it is easy to shew that the ir turn value of (1 + a) 2_,i“ for the
riable a is 2—= . Hence

^ / 2»
Nh ~~ Vie log 2/ ’
where o> (®) denotes the number of primes not exceeding x. But from (10)
we have
4(j On the Number- of Divisors of a Number

Substituting this in (14), we obtain

ior2 (g108* log *1

.
But it is easy to verify that, if % St 6’05, then
2h < e (log 2)!.
From this and (15) it follows that, if x > 6'05, then
d(N) < 2llosNmoe:,') (log*)"1'1'1.
for all values of N, o> (x) being the number of primes not exceed!
5. The symbol “0” is used in the following sense :
*(«)-0 {*(»)}
means that there is a positive constant K such that
j <f> (*) |
I +(*); 5
for all sufficiently large values ot x (see Hardy, Orders of Infinity, pp.
et seq.). For example :
5x = 0(x); ix = 0(x); xsiax =0(oo); V® = 0(a:); log ® = 0 (*);
but a? ^ 0 (x); x log x 4= 0 (x).
Hence it is obvious that
• (•)-<>(•).(17)
Now, let us suppose that
log N
(log log Nf
in (16). Then v have
log x = log log N + 0 (log log log N);
log N _ log N
log® log log '•'l (log log JV)2 '}.^8>
Again

"(*>5o§’lo^=0<• loSloS*) = 0 {— ‘ "-(19)

It follows from (16), (18) and (19), that

.(!0)
for all sufficiently large values of iV.
 9
y7 ON THE SUM OF THE SQUARE ROOTS OF THE
FIRST n NATURAL NUMBERS
{Journal of the Indian Mathematical Society, vii, 1915, 173—175)
1. Let (j)\ (n) = VI + + ... 4- \hi — (0, + ^ n *Jn 4- b \hi)
— 4 2 fv'('« + i') + V()! + v+ l)}-3,
where CL is a constant such that (1) = 0. Then we see that
<4, (n) — (j>, (n +1) = — \Z(ra +1) + [f (n + 1) V(" +1) + ■} \!'(n + 1)]
— (| n <Jn + J *Jn) + J [\/ii - V(?< + l)}:i = 0.
But <#>) (1) = 0. Hence <f>l(n) = 0 for all values of ». That is to say
VI + V2 + V3 + «/4 + ■ ■ • + Vu = 0, + fra V» + i V«
+ i [(Vn + V(» + l)U3 + lv'(“ + 1)+ V(n + 2)}-,i + {V(Ji + 2)+.\/(n + 3)J_:| + ...].
.(1)
But it is known that

°1= ~ 4^{lVl + I72+3V§+ "').(2)

Putting n = 1 in (1) and using (2), we obtain
f 1 1 _1__1__ 1
**■ i<yl)° + (Vi + Wf+(V2 + •/‘if + Wi + viy + -j

= 31 (7i )a+wW + Tv^)3 + (V57 + "' I.(3)

2. Again let
<b., (n) = 1 hjl + 2 V2 - +n \!n — (C., + ftn- Vn 4-1-n \/n -r \/n)

■ - in [V (n + ») + V(n + v +
where C2 is a constant such that <£a(l) = 0. Then we have
<£3 (u) - cj&3 (n + 1) = - (n + 1) V(» + 1)
+ {|(»+ 1)! V(»+ 1) + i (n + 1) VO + l) + i V(« + l)}
- (f )t2 V'i + A n V« + i V»] + Ar (V» — V(« + UP = 0.
But <£, (1) = 0. Hence <f>3 (n) = 0. In other words
1 VI + 2 V2 + 3 V3 + » Vn= V> -f- fa-\/n-t- \ii\Jn + f Vn
+ j’o[l,/« + V(,1 + l))~5+ (V(™+1) +V(« + “)}_s+ {V(« + 2) +V(»+3))-5+---]-
.(4)
48 On the Sum of the Square Roots of

But it is known that

.(5)
C, = ~ 16ir- fcvi + 2* V2 + 3* fi +''') ' ""
It is easy to see from (4) and (5) that

27r= 1(71? + (vi+72 )•+ (72+W*+ (73 + v'47'

.(6)
=15 {<7iT5 + (Tsy+(737 + (V4)s+'' •

3. The corresponding results for higher powers are not so neat as the
previous ones. Thus for example
Is vi + 2V2 + 3V3 + •■• + «2 \fn = c, + *Jn (fn! + -|n2 + -fen)
~ A [(V« + 7(« +1)7* + {V(» +1)+ V(« + 2)]-° +...]•
+ jh [1V» + V(» +1))“’ + (V(« + 1) + V(» + 2))-’
+ jV(n + 2).+ V(« + 3))-»+ ...]; .(7)
Is VI + 2s v'2 + ■ • • + n\hi = C, + v» (4«‘ + i»> + - till)
-T^,[(Vn + V(n + 1)}“5 + {Vi™ +1) + V(n + 2)}~"+ ...]
4-tAt [[V» + V(« + !))""+ {V(™ +1) + V(« + 2))“” +(8)

The constants C3,C„... can be ascertained from the well-known result that
the constant in the approarimate summation of the series lr_1 + 21'-1 + 3r_1 + ...

2£(0 ff , I 1 , 1 ' .(9)

(2-rrY Vlr2r3r + 4r\
provided that the real part ofr is greater than 1.

4. Similarly we can shew, by induction, that

7i + ^ + ^+- + 7i = a' + 2V’l + 27T1

_ (1V» + V(«.+ 1))-* [V(n_+l) + ffn + 2y*}
M V7 («. + !)} V'{(n + lj'(« + 2)|'~ + •}.do)
The value of C„ can be determined as follows: from (10)

7T + 75 + 73 + "'‘+ V(k) ~~ 2 v'(2b) C” .(11)

as n -s- oo. Also
o (JL , 1 JL
"VV2 + V4+V6+- + /(¥«)) -2/(2n)-»C7H,V2,.(12)
 the First n Natural Numbers 49

Now subtracting (12) from (11) we see that

VI V2
That
( ...(13)
7o==_ <^1 + '/2Kvi_ V2 + V3_ V3< + '").
Wee an also shew, by induction, that

VI + V2+\/3 + ... + V« = d, + g n V« +| V» +

_in V«+V(« + l)!_s " {V(re + D + V()l + 2)}_s 1 ...(14)

24 L V{n(n+1)) V{(B+l)(n + 2)} +-"J-
The iasymptotiic expansion of VI + V2 + V3 + • • ■ + for large values of
:an be shewn to be

O + f n V” + 2 Vn + (77((li- 1920a5 + 9216)9 ...(15)

using the Eule;r-Maclaurin sum formula.

 10

ON THE PRODUCT n

i «+w\ (i+p±i^\
+ w )y n ) r _ ((a-ffl+ifr+iwan
a\a/- B\* 1 -2« t

{r (i + g) f (i + ff)]3 .
'.r(l + « + 2/3)r(l + (8 + 2a)’

_ cosh t (g + ) V3 - eos 77- (g — ff) fA s

W{d? + ccl3 + p) ".W
It follows from (1)—(4) that
<Ma./3)<M/3, “)
{r(l+«)r(l + /8)]1> fcosh ir(a + fi) \/3 cos 7r (a - /3)}
r(1 + a + 2/3)r(1 + /3 + 2a")| 2tt2(as + a/3 + 0") ‘ ’"W
But it is evident that, if a — /3 be any integer, then p (a, /3)j<p Q3, a) can
be expressed in finite terms. From this and (5) it follows that tj> (a, /3) can
be expressed in finite terms, if a — ft be any integer. That is to say

fi+Q'Hi+ tt1 + toni1 + taaJ)v-

terms if ® — 2a be a multiple of d.
 An Infinite Product 51

2. Suppose now that a = /3 in (5). We obtain

(T(l + a)}3 sinh 7t«V3 /fn

- TITT 3«j "WT.1 ’
Similarly, putting /3 = a + 1 in (5), we obtain.

{r(l+n)[3COsh7Tft + «)^3 „
_ l1 (2 + 3a) vr .W

it is easy to see that

[(1+r,)(1.|),.,][{lt3(ifJl{1 + 3(^)-}...]

r(4«) /'cosh 7r« ijZ — cos ira\ ^

-f(|-(l + «)}V 2*«WV» "I.(8)

3. It is known that, if the real part of a is positive, then

l°g r (a) = (n — i) log a — a + J log 2tr + 2 ..,(9)

From this we can shew that, if the real part of a is positive, then

4 l.j.2™ +?| + log j(l ♦ J) (l + (j (l +.|). ■. (

-■°e(,",h"tsro,’>2 /' oo)
From this and the previous section it follows that
J“ tan-1 sc' ^

:an be expressed in finite terms if a is a positive integer. Thus, for example,

/„ dx=ilo&2w ~ ~ * Ios u+.(J1)

/„ ^ZT * = 4 log 12tt - —g - i log (1 - ..(12)

 An Infinite Product

. It is also easy to see that

3 1(2 - nf + 3n* (4-u)2+3»2 (6-ii)s + 3ji2

it is clear that the left-hand side of (13) can be expressed

is any odd integer. For example.

The corresponding integral ii

n + 2_n + * n+6 1.
3 ((« + 2)2 + 3n2 (m + 4)2 + 3n2 (re + 6)2 + 3n2 •"}’ '
and so the integral on the left-hand side of (15) can be expressed in finite
terms if n is any odd integer. For example,

ech J7r ^3).(16)

 11

SOME DEFINITE INTEGEALS

{Messenger of Mathematics, slit, 1915, 10—18)

1. Consider the integral

:2 mxdx
/o„ {1 + ®=
a?ja?} {1 + M-j(a +1)5} {1 + a?/(a + 2)2}
where m and a are positive.
It can be easily proved that

{* - (D } l1 ~ fc+l) } l1 “ fc+l) } ••• l1 _ (a + Ll) }

= r(a + »-i!)r(a+«+t)[r(g)p
r(a-<)r(a + f)[r(a + «)}3 ’ ,
where n is any positive integer. Hence, by splitting
1
11 + ! 1 + «*/(“ +1)2} ■ • • {1 + &l(a + n -1 f)
into partial fractions, we see that it is equal to
2r(2a)jr(o + re)j2 f a 2a n-1 a+1
{r (ft)]2 F (n) r (2a + n) jet2 + a:2 1! » + 2tt (a + l)2+ a2
2a (2a, +1) (»-l)(n-2) a+2 )
2! (vi + 2a) (n + 2ct +1) (a + 2)2 + <c2 -}'
Multiplying both sides by cos 2mx and integrating from 0 to oo with re

cos 2mxdx
x-ja?}
/» {1 + (v-1 a J1 + x~j(ft +1)2} ... |1 + «;2/(ffi + n -1)2{
_ ^r(2a)!r(tt + n)}2 f 2ttn-l )
{r(a)]2r(»)r(2ft + n) ( lln + 2a + '"J
The limit of the right-hand side, as n —» go , is
^(2a) |e_Mra_ 2ae_,la+vill + 2a_(?£±1)e-.«+w_.. j

= ^-[TUi)seohM5

0 {1 + ofla^{l+x’:/{a+ !)■}... ■ JyV ^ pt/+y ^ sechm to.(1)

54 Some Definite Integrals

Since f1 + S)} {x + Gtti) } l1 + fers)} "•= r(o+U(r(a-i«) ■

the formula (1) is equivalent to
f j F (a + ix) |! cos 2mxdx = (a) F (a + 4) sechMm.(2)

2. In a similar manner we can prove that

■
r (1 + xW\ A + + 1)2\ A + + )2\ „„
J0 U + a?/ov VI + *5/(“ + iW VI + ®V(a + 2)V '''
2
_ *r(2a){IW fe-m_6~a~1
}r(a))2r(6 + o)r(6-a) i 1! 6 + a
2a(2a + 1) (b - a - 1)(6 - a - 2) +„
+ 2! . (6H- a)(6 + a-H 1)
where m is positive and 0 <a<b. When 0 < a< b - 4, the integral and the
series remain convergent for m = 0, and we obtain the fonnulse
' A + a?/b*\ a + x°-/(b + in a + x%b + 2f\
\1 + tf/a?) u + *V(a + If] U + xfia + 2f)
_ r(q + i)r(6)rfl-a-i)
r(a)T(b-i)r(b-a) ’
r i r(a + fo) I*
Jo r(i +15)I
, _ 4
r(q)r(q + )T (6-g-i)
r(6-i)f(6)r(i-a) ‘
.(4)

If di, a,, as, ..a„ be n positive numbers in arithmetical progression, then

f"_dx_
J0 (af + a?) (af + sc2) (o32 + a") ... (a„2 + a?)
is a particular ease of the above integral, and its value can be written down
at once. Thus, for example, by putting a — {4 and b = £4, we obtain
f" dm
J0 (x= + 11s) (ic2 + 21=) (xJ + 312) («2 + 41!) (of + 51=)
57r
= 12.13.16.17.18.22.23.24.31.32.4,1'
3. It follows at once from equation (1), by applying Fourier’s theorem
| cos ngdgj <f> (x) cos xydx = far <f>(n),

that, when a and n are positive,

f sechM x cos 2nxdx

T(a)
r (a + 4) {1 + {I + n=!(a + 1)=} {1 + nfia + 2)=} ...
 Some Definite Integrals 55

Hence the function

<f>(a)=J sechax cos nxdx (0<a<2)

satisfies the functional equation

*<8>*<2 ~ a)= 2(1 -a)(Iosh^-cos^)-

4. Let j f(x)F(nx)dx=-ifr(n),

and J <f>(x)F (ncc) dx = % (n)•

Then, if we suppose the functions f cf>, and F to be such that the order of
integration is indifferent, we have
j f(fi) X (*“0 dx= dy j /(*)'‘<f> (y) F (nxy) dx
= j <t>(y)JHny)dy.(6)
A number of curious relations between definite integrals may be deduced
from this result. We have, for example; the formulas

(7)
/; cosh irx^ 2 cosh n ..

/;’ cos 2 nxdx V3

1 + 2 cosh §rrx '”2(1 + 2 cosh 2n) ’ (8)

/: e~x‘ cos 2 nxdx = £ lire-**.

By applying the general result (6) to the integrals (7) and (S), we obtain
.(9)

H0 cosh 7tx 7Tir(l

(1 +2 cosh 2nx)
2m;) ~ /0 cc
cosh nx (1 + 2 cosh £mr) ’
n other words, if a/3 = f 7r!, then
/ f“_dx_
J D cosh ax (1 + 2 cosh irx)

Wo cosh/3®(l + 2 cosh -rrx) ’

In the same way,"from (8) and (9), we obtain
, [“ e~*’-dx f" e-^-dx
aJt> 1 + 2 cosh ax ” ^Jo l + 2cosh/3
with the condition a/3 = |ij- ; and, from (7) and (9),

.<12>
with the condition a(3 = tt*.
* Formulse equivalent to (11) and (12) were given by Hardy, Quarterly Journal,
xxxv, p. 193.
56 Some Definite Integrals

Similarly, by taking the two integrals

and J xe~x,sinnxdx = i\/irnerW,

we ean prove that, if a/3 = 7r2, then

^l1
.(13)
(jAl + WL^-1^};.
and so on.

5. Suppose inow that a, b, and n are positive, and

J <j> (a, x) nxdx = ^r (a, n). .(14)

Then, if the conditions of Fourier’s double integral theorem are satisfied, v

j ifr (6, &■) °PS nsodcc — $7r<j> (b, n). .(15)

Applying the formula (6) to (14) and (15), we obtain

i’1-Jf 0 (a, x) <j> (b, nx) dx =J ijr (b, #) ijr (a, nx) dx. .(16)

Thus, when a = b, we have the formula

Jo (mi)•f (®)^(«®)

where ijr(t) = J <j> (x) tx dx;

and, in particular, if n = 1, then

W Jo {<f> («)? dx = Jo {f (x)}’- dx.

If <f> (a, x) = ii^y^}{1 + a,/(ct+O > °)>

then, by (1), ^ (a, x) = £ Vtt F sech” \x.

Hence, by (16),

- Jo <j> (a, x) <f> (b, x) dx = r ( 2rfi) rW ^ j 0 seoh“+!!i 1® dsc 5

 Some Definite Integrals 57

J „ {1 + 3,“jar} {1 + «•/(» + 1)=} ... {1 + {1 + xfib + 1)=} • • •

_ r(a + i)r(b + i)T(a + b)
~''2'j7r r(a)r(i)r(a + 6+i; ’
a and b being positive: or
j : T (a + ix) r (6 + is) \-dx
° T(a)r(a + i)T(b)r(b + ^r(a + b)
=*w- " ' ~~ '+*)
As particular
ular cases of the above result, wew have, when b
f* _dx_
J o sinh trx {1 + a?jo?) {1 + ®s/(a + l)2} .. = 2 (1 + 2a)’

J „ sinh ™ {1 + a-.«;}{l + a2/(a + 1 )2} ... f(1 + 2a) (3 + 2a) ’

and so on. Since II {1 + xj{a + n)2} can be expressed in finite terms by
means of hyperbolic functions when 2a. is an integer, we can deduce a large
number of special formulas from the preceding results. ,
6. Another curious formula is the following. If0<7-<1, »>0, and
0 < a < 7'n~1, then
(X + arx) (1 + arsx)...
o (! + *)(!+?•*)(!+
1 n
(1 - .(19)
‘)(1-
ran integer or a is of the form r1 ■here p is a positive integer,
”, the formula reduces to

Ju (l + x)(l + rx)...(l+r”x)
-n)
_ (1 ~ r'~n) (1 ~ r*~”) •••(! — ..(20)

If n is an integer, the value of the integral is in any case

logr (1 — ?') (1 — rs)... (1 — rn~l).

My own proofs of the above results make use of a general formula, the
truth of which depends on conditions which I have not yet investigated
completely. A direct proof depending on Cauchy’s theorem will be found in
Mr Hardy’s note which follows this paper. The final formula used in
Mr Hardy’s proof can be proved as follows. Let
f{t) = "5" 4 A,t + +....
58 Some Definite Integrals

Then it is evident ti
(1 - at)f(t) = (1 - bt)f(tx).
That is
(1 - at) (A, + AJ + A,t‘ +...) = (l-bt)(A0 + Ajx + A^x? + ...).
Equating the coefficients of tn, we obtain

and At is evidently 1. Hence we have

i fa~b a(“ — b)(a — hx)
/(0 _ 1 + i 1— +1 +.(21)

7. As a particular case of (19), re have, when a — 0,

Jo (1 + x) (1 + r*)(l + rts) ... sin mr 1-r 1

When ii is an integer, the value of the integral reduces to
_ ,-*».»-« (1 _ r) (1 -1-)... (1 - r’«) log r.
When we put n = ^ in (19), we have
r_±_L±^L+^\. dx
J o 1 + 0? 1 + r’-s? 1 + r1®2
,^1- ctr-1 — ara 1-r l-r‘

If, in particular, n = J in (22), or a = 0 in (23), then

/oo (1+12
(1 + ®2) (1 + r2®2) (1 + rtr2)...

• r21 — r* 1 — r“ ’" 2(1 + r + r8 + r6 + r10 4

the nth t e denominator being r^nin~,K Thus, for es

'im ^ic2) (1 + ... 2 (1 + + e~'<" + er

= 7T* r (i) gs n J (1 + 5) {J (1 + V5)}ie-

f“_dx_
Jo (1+ of) (1 + e-^nf)(1 + e-^x1)...
= w»r(f)V5 ^2^(1 + s/5 Y {£ (1 + V5)F «-*"k;

Jo (1 + of) (1 + -001*2) (1 + -00001®2)..

2-202 002 000 200 002 000 002 ..

 12

SOME DEFINITE INTEGRALS CONNECTED

WITH GAUSS’S SUMS

(Messenger of Mathematics, XLIV, 1915, 75—85)

1. If n is real and positive, and j I (t) 1, where I (t) is the imaginary part
of t, is less than either n or I, we have
lZU20-i«i

When n = 1 the above formula reduces to

/o ^^sinM=tan{i7r(i-ti)}L ^Soos71
If t — 0, and <j> (n) = J <2®,

.
then ^ = \/ (^) ^ Q + ^ <“)■ I

Similarly, if J V3|/(<)| is less than either 1 or n, we have

Jo UfTcosh(2wa/V3) 6 ^

” Vuexp

If in (4) we suppose n = 1, we obtain

1+2 cosh (27ra;/V3)

eos 7ri* cos w®2
mG7r(l -i2)}Jo ^
60 Some Definite Integrals connected with Gauss’s Sums

and if t = 0, and

* (%) = L 1 + 2 cosh (%7rxfj3) dx’ | (6)

*(n)“/o 1 + 2 cosh (27t®/V3) dx’ J

then = (^)+^(«).|

In a similar manner -we can prove that

L ^^e^dx=-vnexHii7r(1+3}L (7)
If we put n = 1 in (7), we obtain

.I S;cos***’“tan ^ (1+*a)}/„ ssrssin "**’• • • -(8)

Now

lim J f
t~*-o E Jo tanho®

= lim i j ^'dx + lim J s‘a^tx (fc&dx

“/„ iaw=i<&+i.(®>
Hence, dividing both sides of (7) by t, and making t-*■(), we obtain the
result corresponding to (3) and (6), viz.: if

<£(«)=/]
°i r sm7nM (10>
^'sSi + Jo ¥^=idx’\

then 4, (n) - i ^g) * g) - * («), |

"f («) = ^ \/(^) £ g) + <#> (»)• J
2. I shall now shew that the integral (1) may be expressed in finite terms
for all rational values of n. Consider the integral
 Some Definite Integrals connected ivith Gams’s Sums 61

If R(a) and t are positive, we have

j,t)-*r T (-b'Cle+Dc dx
( )_77./„ ®* + (2r + l)> oa
-1 r 1)8 — (2r + I) e~
TTf 1
+2 2 ....(11)
2aeosT“ ,=0 a2 -(2r + l)2 ’
and it is easy to see that this last equation remains true when t is complex,
provided R(t)> 0 and ! I (t) j sri-a-. Thus the integral J(£) can be expressed
in finite terms for all rational values of a. Thus, for example, we have

~~n = cosh t log (2 cosh t) — t sinh £,

f" C0S'2tX „ = 2 cosh £ - (e“ tan-1 e'( + tan"1 e»),

Jo cosh was 1 +ar
>sliw«: ' '
. Now let

/w-/;

os 2te d®
j"e-”F(n)dn=J*?

Now let

/(») - l)'exp {- (Sr + 1)£ + J (2r + l)HVn}

+ . (15)
Then
_(_ iye-if+m
Jo e-™f(n)dn= ^
•o ® — i (2r + l)2wr
, //M exp{-V(2a/7r) (1 — f) £} = f" cos 2£a; da
h V \2a/ (1 + i) cosh {(i + i) \/{\iraj} Jo cosh me a + iirx- ’ ' '
of (11); and therefore

(»)-/(«)} At = 0.(17)

it is known that, if <f> (n) is continuous and

f e~m<j>(n)dn = 0,
62 Some Definite Integrals connected with Gauss’s Sums

for all positive values of u (or even only for an infinity of such values in
aritnmetical progression), then
<t> (n) = 0,
for all positive values of n. Hence
^(»)-/(»).(18)
Equating the real and imaginary parts in (13) and (15) we have
I cos 2to cog
 3. We can get many interesting results by applying the theory of Cauchy’s
reciprocal functions to the preceding results. It is known that, if

J 4>(x) cos knxdx = f{n), .(27)

then (i) i«{i^(0) + ^(a) + ^(2«) + ^(3a) + ...}

= if (0) + f(0) + f (2/3) + f (3/S) +.(27)
with the condition a/3 = 2w/i;
(ii) a V2 {<f> (a) - <f> (3a) - <f> (5a) + $ (7a) + <jf> (9a) - ...}
= + 08) - f (3/S) - * (50) + * (70) + f (90) - .... (27)
with the condition a/3 = 7r/tt;
(iii) a V3 {</> (a) - cj> (5a) - f (7a) + <j> (11a) + 4> (13a) - ...}
= * (/3) - f (5/3) - f (7/3) + f (11/9) + f (13/3) - ..., (27)
with the condition a/3 = 7J-/6&, where 1, 5, 7, 11, 13, ... are the odd natural
numbers without the multiples of 3.
64 Some Definite Integrals connected with Gauss’s Sums

There are of course corresponding results for the function

I <£(a)sin knxdx = ^{n), .(28)

a{<j>(a)-<p(2a) + <j>(5a)-...} = f'(!3)-ir(3l3)+'f(5/3)-..., ■

with the condition a/3 = tt/2&.
Thus from (23) and (27) (i) we obtain the following results. If

.<**>
then F(a,{3) = F(/3, ct) = \/(2a){J + «-’"‘ + «-J’r“ + e_!"r“+
provided that a/S = 1.

4. If, instead of starting with the integral (11), we start with the corre¬
sponding sine integral, we can shew that, when R (a) and R (t) are positive

t ” sin to, dx
J o sinh irx a* + a?
be expressed in finite terms for all rational

J o sinh \-wx 1 + a?
From (30) we can deduce that
J‘ = i - e~,t+i”n + +...

- ± exp {(Jtt + i| {,-<*+*->/» + ,-<)«+!<->/* + (32)

R(t) being positive and 11 (t)|$^7r. The right-hand side can be expressed
in finite terms for all rational values of n. Thus, for example, we have
ri — cosrf
2 irtsnda =
j„ sinhwi

■sin 2trtxdx ...(34)

2 sinh irt.

Applying the formula (28) to (33) and (34), we have, when a/3 = J,
— cos{(2r + l)°7rtt2} *■=-, y cos {(2r+1 )8 u/3g} 1
’ sinh{(2r + l)7ra}+ VP } sinh{(2r +1)tt,8} |
=2 +e-Sra + + «r“™ + K. (35)
sin{(2r+l)8™8} sin{(2r+1)8^8} I •
,~0( X) sinh{(2r+ 1)vro,} ~ ^ ,.?0 _ ^ sinh{(2r +1)^}' J
 Some Definite Integrals connected :with Gauss’s Sums 65
By successive differentiation of (32) with respect to t and n we c
the integrals

f> f
for all rational values of ?i and all positive integral values of m
example, we have
rX— dx - - [* X si‘—*5 dx - 1
Jo'"sinhwa: 8’ ,/0 sinh was. Ssr ’

^ sinh ttx ~ 167T ’

The denominators of the integrands in (25) and (36) are cosh ttx and
sinh irx. Similar integrals having the denominators of their integrands equal to
II cosh 7r«,.a: sinh ®
can be evaluated, if a, and br are rational, by splitting up the integrand into
partial fractions. •

5. The preceding formula? may be generalised. Thus it may be shewn

that, if R (a) and R (£) are positive, j 1(f) j s tt, and -1 < R (6) < 1, then
sin ttB j"° cos tx dx _ _w a-"1 sin ttB

+,?o |o= - (2r+1 — ey ~ a^Tir+TTe)’}'(38)

From (38) it can be deduced that, if n and R(t) are positive, |J(t)|^w,
and - !•< 6< 1, then

J |) COSh TTX + COS 77 6

+ ^°Xp{-ii(7r~TO)} ?! (-iy~lsinr7rde~ ■ (30)

The right-hand side can be finite terms if n and 3 rational.
In particular, when B = J, v

1 + 2 cosh (2™/V3)
q- e"i 0*V3-ifiiVM) _ .,
66 Some Definite Integrals connected with Gauss’s Sums
where 1, 2, 4, 5, ... are the natural numbers without the multiple
of 3.
As an example, when n = 1, we have
/•’ cosrfcos irtx _l-2sin{(w-3rf)/12}
/„ 1 + 2 cosh (2ir*/v'3) Scosh(wt/V3)-4 ’
...(41)
f”_sin ttx‘ cos vtx — \/3 + 2 cos{(7r- 37rf2)/12}
Io 1 + 2 cosh (2ira;j*/3) * Scosh(7rt/\/3)-4

6. The formula (32) assumes a neat and elegant form when is changed
to t + \i-rr. We have then
jmte e_,:^dx (j! >0 (>0)
tanh 7rx

...(42)

In particular, when n = 1, we have

f SSEmsin= £tanhwt^1 -cos(i1r + |

...(43)
f tanlT*'■ “n 2,r<*<*®“|tanh’rts>“(i’r + 7rf2). j

We shall now consider an important special case of (42). It can easily be

seen from (9) that the left-hand side of (42), when divided by t, tends to

/; .(44)

as t->-0. But the limit of the right-hand side of (42) divided by t.ean be
found when n is rational. Let then n = ajb, where a and b are any two
positive integers, and let

<*(«)=/“

The relation between 0(a) and 0-(ti) has been-stated already in (10'). From
(42) and (44) it can easily be deduced that, if a and b are both odd, then

+(s)--* I'»■- «•“ fr)+i-J © ’f> - 2">'“ (+■+■ I

.(45)
It can easily be seen that these satisfy the relation (10'). Similarly, when
one of a and b is odd and the other even, it can be shewn that
 13
SUMMATION OF A CERTAIN SERIES

(Mmenger of Mathematics, xuv, 1915, 157—160)

1. Let x) = 2 {V(® + n) + V(® + n+ l)]-s

=T{V(«+« + i)-'/(«+*)}*•

The object of this paper is to give a finite expression of <3? (s, 0) in terms
of Riemann f-functions, when s is an odd integer greater than 1.
Let f (s, x), \yhere x > 0, denote the function expressed by the series
xr‘ + (x+ l)"' + (a! + 2)->+
and its analytical continuations. Then
r(*,i) = ?(«), = .a)
where J (s) is the Riemann f-function;
f(*.«)-f(*,« + l) = «r*; .(2)
l» + 2* + 3s+.n + 1), ■ |
l* + 8* + 5* + ... + (an-l)'=i(l-2*)r(-*)-2,{:(-»,» + i)i "'(8)
if n is a positive integer; and

Lhn {?(., «) - lx~* + (£1 -B.fj + Bt ^-+-ff-+2* *—•

^ - * •C!±li«!±Sife±i) it±fi + ... to . terms)} - 0, ...(4)

if n is a positive integer, - (2n - 1)< s< 1, and jBs = £, = Bc = -fe,

Bt~-3*5■ • • • > are Bernoulli’s numbers.
Suppose now that
•f (<s) = 6? (- J, «) + (4® - 3) V* + <*> (3, a).
Then from (2) we see that
■f («) - 0 + 1) = 6 V® + (4® - 3) V® - (4® + 1) VO + 1)
+ {*/(* + !)- V®}s = 0;
 Summation of a certain series 69
and from (4) that i/r (x) -> 0 as x oo . It follows that -^■(x) = 0. That is
to say,
er (- h X) + 0* - 3) V® + ® (3, ®) = 0.(5)
Similarly, we can shew that
40f (- f, ®) + (163s - 20® + 5) V® + ® (5, a) = 0.(6)

2. Remembering the functional equation satisfied by f (s), viz.,

f(l -s) = 2(27r)-»r(s)f(S)cos^s, .(7)
we see from (3) and (6) that

VI + V2 + V3 + • • ■ + V» = 5^ f(t) + i® (3, »); .. .(8)

and VI + V3 + v'5 + ... + V(2n -1)

= i (2a -1)* + l.V(2» -1) + ?(l) + 3^® (3, * - *). • • -(9)

Similarly from (6), we have

1V1 + 2V2 + 3V3 +...

-1«* + i«* + j v» - f«) + A® (*.»);.(io)

and 1 VI + 3 V3 + 6 V5 + ... + (2n -1) V(2re - 1)
= H2n-l)t + i(2n-l)* + iV(2»-l)
+ 3(2^-1)?(|) + 1^4(3j,i_4).(u)
It also follows from (5) and (6) that
4- d) + *J((t + 2d) -{- V(a +■ 3d) -f-... -f \/(a + nd)

= C+^(a+nd)i + hfa + nd) + iJd$(3,n + aid)\ ...(12)

and (a + d)l + (a + 2d)- + (a + 3d)* + ... + (a + «d)*

= O' + ^ (a + ?td)- + f (a + rad)T

+ id V(o 4- nd) + Vd <£> (5, n + a/d).(13)

where G and O' are independent of m.
Putting ra = 1 in (8) and (10), we obtain

®(3, 0) = ^f(». ®(«.0) = gjf(})..(14)

70 Stimulation of a certain series

3. The preceding results may be generalised as follows. If s be an odd

integer greater than 1, then
t£> (s, x) + i {v'.n + \/(x -1)]” + £ (\/« - \A> - l))s
+ ± 2~ f (1 - is, x) + 1 2'"“ f (3 - *)

+ »(«-6)(«-7)(a-8)fr-9) 2M0 ?(5 _ ^ x)

s (s - 8) (s - 9) (s - 10) (s - 11) (s - 12) (s - 13) 9s_u

x ?(7 - i«,*)+... to [1 (s + 1)] terms = 0, .(15)

where [&■] denotes, as usual, the integral part of x. This can be proved
by induction, using the formula
{V® + V(® + l)j* + Wx- V(® + 1)}*
= (2 ± p, (2 + S~^j^ (2 V®)*-4

+ —— ——— (2 V®)S_G + ... to [1 + is] terms, .(16)

which is true for ,all positive integral values of s.

Similarly, we can shew that if s is a positive even integer, then

ij2-»{?(i-i«)-r(i-i*,«)}
+ i^£-6)2M{r(3_Ja)_f(3_i4><e;}

+ ^-«)(»-7)(«-8)(izj) 2»-M (5 - *) - r (5 - is, *)}

+ ... to [J {s + 2)] terms

= i Wa + V(« - 1)}S + 4 (V® - V(®- 1)}S -1.(17)
Now, remembering (7) and putting x = 1 in (15), we obtain

* (*, 0) = - -L ,1+.) cos ivrs (1.3.5 ... {s - 2) 7T ids)

— 3.5.7 ... (s — 4)i(s — 5) £7r3f (|s — 2)
+ 5.7.9...(s-6)£(s —7)i(s —9)i7r=?(£s-4)
— 7.9.11 ... (s —8)£(s — 9)|(s —11) £ (s —13) }tr7 J(is — 6)
+ 9.11.13 ... (« — 10)£ (s —11) £ (s — 13) £ (s —15)£ (s — 17)

x i*’ ?(£« - 8) - ... to [i(s + 1)] terms].(18)

if s is an odd integer greater than 1. Similarly, putting a=£ in (15),

we can express <E> (s, i) in terms of ^-functions, if s is an odd integer greater
 Summation of a-curtain series 71

4. It i!i also easy to shew that, if

,ws ®) + »> + *'(*+» + !)}-
/i=n \/{(^’ + ?i j (x + n + 1)} *

, Wx + ^ - 1)}* - Wx - V(s -l)!•

^ (s,
- V{®(« — l))
= __ V22'-= f(2 & a') + (*~4)t*~6)(*-"6)2-?(4-fe B);

+ (« - 6) (s - 7) (s - 8) (* - 9) (s -10) 9M# ?(6

+ ...to[J(* + l)] terms, ..(19)

provided tlbat s is a positive odd integer. For example,

|
¥(3,a) = 4V®-;i+2f(4,ffi), | .. .(20)

'F (5, x) = 16® V® -12 V® + ^ + 24?(- i, ®),j

 14
NEW EXPRESSIONS FOR E1EM ANN’S FUNCTIONS
£(«) AND S (t)
(Quarterly Journal of Mathematics, xlvi, 1915, 253—260)
1. The principal object of this paper is to prove that if the real parts of
a and /3 are positive, and a(3 = 7r, and t is real, then

.<>,
Consider the integral
J(u'> = f^T1d^
where the real part of u is positive. Since

we have + ^- 4~=fo +1~

= j J tve~,ruxS dxdy — u~-J ^ ^ dx;.(2)

and 80 -1^)^.<3>
Suppose now that s = a +it, where 0 < a < 1. Then, from (3), we have

= n-i^ a«»-« du|o (—^ - dx..(4)

. Changing u into 1 jv, we obtain

""‘J, ’■‘'*'1. "■""*(^-5-,)“'“

- f !)-i» dv f xe-™*to (-J— - jj-—) dal .
 New expressions for Niemann’s functions £(s) and B (t) 73

/„’ “K-i,f(B“> - &-.C u^l,J(mi) du -

All the inversions of the order of integration, effected in the preceding

argument, are easily justified, since every integral remains convergent when
the subject of integration is replaced by its modulus.
It follows from (4)—(7) that, if the real parts of a and /3 are positive, and
a/3 = 7ra, then
74 New expressions for Riemamis functions £ (s) and E (t)

Changing s to \ (1 + it) in (8), and writing as usual

e<i+*»■<)“ sew.
and equating the real and imaginary parts, we obtain the formula (1), and

2. We have proved (8) on the assumption that 0 < <r < 1. But it can be
shewn that the formula is true for all values of s other than integral
values.
Suppose first that - 1 < a < 0. The formula (3) is equivalent to

Using this formula as we used (3) in the previous section, we can shew
that (8) is true in the strip — 1 < cr< 0 also. In the right-hand side of (31,
the first.term in the expansion of l/(eM— 1), viz. l/(2wx), is removed, and in
that of (10) two terms are removed. By considering the corresponding
formulfe in which more and more terms in the expansion of l/(eMa!— 1), viz.
1 1 7tx ttV tt’x7 ttV
2wk~2+T “ "90 + 945 “9460 + 93555
are removed, we can shew that the formula (8) is true in the strips — 2 < cr < 1,
— 3 < a < — 2, and so on. That it is also true in the strips 1 < o- < 2,
2 < <t < 3,... is easily deduced from the functional equation ? (s) = ?(1 - s).
The formula also holds on the lines which divide the strips, except at
the special points s = k, where k is an integer. This follows at once from
the continuity of f(s) and the uniform convergence of the integrals in
question.

But the left-hand side of (11) is equal to

 o expressions for Riemann’s functions g (s) and S (f) 75

irPr1)r(:T!

«-l|l + 4aj =/3-i |l + 4/3 J -j- fej • ■•■(14)

ii this it follows that the left-hand side of (13) is an even function of n,
so the formula (13) is true for all real values of n.
1. It can easily be shewn that, if a/3 = 4tt2 and R (s), where R (s) is the
part of s, is greater than — 1, then
lOLrf)
4 cos Jtts 8sm47T5

4 cos \nts ^ 8 sin |7rs ^

im this we can shew, by arguments similar to those of §§ 1

/3 = 47r and R (s) > - 1 then
?(!-*) , a-s) giW
4 cos ^7rs s 1 t 8 sin s +1 — t
+ oi(s+i >f n_a-3L_
+0 Jo Jo tl!(* + 3-«) 3 ! (s + 7 — t)
(axyf 1 afdxdy
+ 5T(s + lT^ “ (i5«-l)(^’r»7ri)
f(l-«) C(-») fflf+i)
+ 4 cos firs s — 1 + f 8 sin |tts s + 1 + t
+ i8l(.+i) f f{_^_
P Jo Jo jl!(s + 3 + t) 3!(g + 7+«)
1 _«!*;*/
5 ! (s + 11 + f) ‘ "J (e2** - 1) (e2'» - 1)
_ /f\« 2><—>rft(»-!+«)) T ft («-l-i))
“V^J V (sTl)2-<2
■ (16)
76 Neiv expressions for Riemann’s functions g(s) and S (t)

From this we deduce that, if a/3 = 47r2 and R(s) > — 1, then

?(L-£)
7
4 COS ^ TS (S — 1)! + f
ja4(»-l)+JQK»-D)

?(-«) _L+i__ Jai(s+1) +/9H»+i)j

3 sin i-rrs (s + 1> + t‘

{aseyf s+7 ) xsdxdy

“3T~ (s + 7)- + ts+'"J (is3’*'—i) 1)

_ Q3m/)3 s+ 7 _ ) x‘dxdy
3! i> + 7)!+> "j (e2™ - 1) (d-'y - 1)
-2K.-3) r {i (s -1+»<)} r {i («-1 - it)}
1r (s+l)s+t»

.(17)

and f-(1 S) ,^ {a*<s-‘> -^ (*-«}

4 cos \tts (s - l)2 + t- 1 ^ 1
+ _£fci)-1_ |„i(s+i) _ flH*+D|
+ 8sini7rs(5 + l)2 + t!t P S

+a»c-+i >rnis._l_
Jo Jo t 1 ! (s + 3y + t*
(axi/f 1 ] x’dxdy
3 ! (s + 7)2+> + '’'J (e!«-l)(es'®-1)

_ 09 *!/)“ 1 | x?dx dy
~3 f (s + if + f 'J i)
_ 2* (s~s> r \i(s -1+it)} r a (S - i - it)\
i- (s + l)2+t2

xS^S^sin^logl).(18)

5. Proceeding as in § 3 we can shew that, if n is real, and

*■> -i: 5 (t4) s Ct) “• ■< *

 New expressions for Rienicmn’s functions f (#) and £ (t) 77

then, if R(s) > 1,

j /„ (?»—- 2r 1W r <*> «*h »(1 - *)};

.(19)
if — 1 < -R(s)<l,

J(»)-i(4*-)

if - 3 < R (s) < - l,

F( „) = X (47r)-i <-»> {Jo (—A - A + i)

x (-A - L - Aj, + l) da - r (1 + s) f(l + s) cosh « (1 + s)|;.. .(21)

and so on. If, in particular, we put s = 0 in (20), we obtain

/>(= rriri)«a<M’iTf'i‘

-W."Wn-») fc-ti-i.) *.0»)

 15

HIGHLY COMPOSITE NUMBERS

{Proceedings of the London Mathematical Society, 2, XIV, 1015, 347—409)

CONTENTS
1. Introduction and summary of results.
-5. Elementary results concerning the order of cl (-V).
2. The order'of d {N) when the prime divisors of N are known.
3. The order of d (iV) when the number of prime divisors is km
-5. The order of d(N) when nothing is known about N.
III. 6- -31. The structure of highly composite numbers.
6. Definition of a highly composite number.
7. A table of the first 102 highly composite numbers.
8. * first standard form of a highly composite number. The in
the largest prime divisor.
9. Alternative standard form of a highly composite number.
0—24. The indices of the prime divisors in the first standard form.
!5—27. The final primes in the second standard form.
28. The ratio of two consecutive highly composite numbers.
29. The form of d (Ar) for highly composite values of N.
30. The order of dd (Ar) for highly composite values of N.
31. Some apparently paradoxical forms of d !.Y) in the table.
IY. -38. Superior highly composite numbers.
32. Definition of a superior highly composite number.
33. The exact form of a superior highly composite number.
34. The number of divisors of a superior highly composite number.
35. The maximum value of d for a given value of x.
36. Consecutive superior highly composite numbers.
37. The number of superior highly composite numbers less than a
given number. A table of the first 50 superior highly com¬
posite numbers.
38. Superior highly composite numbers and the maximum order of
dIE).
Application to the determination of the order ofd{N).
The maximum order of d (AT), assuming the prime number theorem.
The maximum order of d (Y), assuming the Riemann hypothesis.
The order of the number of superior highly composite numbers
less than S.
Highly composite numbers between consecutive superior highly
 Highly Composite Numbers 79

Introduction and Summary of Results.

1. The number d (IV) of divisors of N varies with extreme irregularity
as JST tends to infinity, tending itself to infinity or remaining small according
to the form of N. In this paper I prove a large number of results which add
a good deal to our knowledge of the behaviour of d (iV).
It was proved by Dirichlet * that

dJ1} + = hgp + iy^ + o(J^t

where y is the Eulerian constant. VoronoiJ and Landau§ have shewn that
the error term maybe replaced by 0 (hT~i+‘), or indeed 0 (IV-* logiV). It
seems not unlikely that the real value of the error is of the form 0(i\r_!+t),
but this is as yet unproved. Mr Hardy has, however, shewn recently!; that

d- = logrY + 27_ 1 + 0,

is certainly false. He has also proved that

d(l) + d(2)+... + d(N-l) + $d(N)-NlogN-(2y-l)N-l

= i [J?, {4tt >} - r, {4tt V(«iV)}],

where F„ is the ordinary second solution of Bessel’s equation, and

id that the series on the right-hand side is the sum of the series

and an absolutely and uniformly convergent series.

i Crellds Journal, Yol. 126, p. 241. § GSltinyer Sadmehten, 1912.

|| Comptes Rendws, May 10, 1915. See Appendix, p. 338.
80 Highly Composite Numbers

The “ average ” order of d (iV) is thus known with considerable accuracy.

In this paper I consider, not the average order of d (N), but its maximum
order. This problem has been much less studied. It is obvious that
d(N)< 2yN
It was shewn by Wigert* that
d (N) < .(i)
for all positive values of e and all sufficiently large values of IV, and that

d(IV)> 2i5tUiw(,“') .(ii)

for an infinity of values of IV. From (i) it follows in particular that
cl (IV) < IVs
for all positive values of 8 and all sufficiently large values of 17.
Wigert proves (i) by purely elementary reasoning, but uses the “ Prime
Number Theoremf” to prove (ii). This.is, however, unnecessary, the in¬
equality (ii) being also capable of elementary proof. In § 5 I shew, by
elementary reasoning, that
’ d(N)<
for all values of N, and
d (iV) > 2 +0 sf
for an infinity of values of IV. I also shew later on that, if we assume all
known results concerning the distribution of primes, then
d(N)< 2W (log N) + 0 [l0R AT«-“''(l0|!l<>lsi'0]
for all values of IV, and
d(IV) > 2«a°g^ + Opog»--^^
for an infinity of values of N, where a is a positive constant.
I then adopt a different point o%view. I define a highly composite
number as a number whose number of divisors exceeds that of all its pre¬
decessors. Writing such a number in the form
IV = 2°*. 3““. 5“s... pa!‘,
I prove that a2 > as > aE > ... ^ap,
and that ap = 1,
for all highly composite values of IV except 4 and 36.
* Arkiv fir Maiematik, Vol. 3, No. 18.
t The theorem that
n (x) being the number of primes not e
 Highly Composite Numbers 81
I then go on to prove that the indices near the beginning form a de¬
creasing sequence in the stricter sense, i.e., that

where X is a certain function of p.

Near the end groups of equal indices may occur, and I prove that there
are actually groups of indices equal to
1, 2, 3,4,
where p. again is a certain function of p. I also prove that if
in comparison with p, then
«*1 o'gX-j^!;
6 loe 2
and that the later indices can be assigned with an error of at most unity.
I prove also that two successive highly composite numbers are asympto¬
tically equivalent, i.e., that the ratio of two consecutive such numbers tends
to unity. These are the most striking results. More precise ones will be
found in the body of the paper. These results give us a fairly accurate idea'
of the structure of a highly composite number.
I then select from the general aggregate of highly compssite numbers a
special set which I call “superior highly composite numbers.” I determine
completely the general- form of all such numbers, and I shew how a com¬
bination of the idea of a superior highly , composite number with the
assumption of the truth of the Rjemsig'W 'hypothesis concerning the roots
of the f-function leads to even more preciot^esults concerning the maximum
order of d (N). These results naturally differ from all which precede in that
they depend on the truth of a hitherto unproved hypothesis.

II.
Elementary Results concerning the Order of d (A7).
2. Let d (E) denote the number oj^divisors of N, and let
N=p^p^p^...pn^, .(1)
where pi,pi,ps, ..., pn are a given set of n primes. Then
d (N) = (1 + a,) (1 + a2) (1 +a»)... (1 + a»).(2)
From (1) we see that
(1/n) log (pap2pj ...pnN)
= (1 In) {(1 + oO logp, + (1 + a2) log p2 +... + (1 + a„) log p„}
> ((1 + a,) (1 + of) (1 + a3) ... (1 + an) logp, log p2... logp„Jln*.

Hence we have .(8)

logp, logp2logps... logp„
for all values of A7.'
82 Highly Composite Numbers

We shall now consider how near to this limit it is possible to make

d{N) by choice of the indices a,, a?, Os, .... an. Let us suppose that .

l + «m=»!°gJ,- + em 0 = 1, 2, 3.n), .(4)

where v is a large integer and - e„ < Then, from (4), it is evident

that
■0.:.(5)
Hence, by- well-known theorem due to Dirichlet* it is possible choose
values of v s large as we please and such that

f = rlogp„, 8m = emlogpm. ,..(7)

Then from (1), (4) and (7) we have
log (p,p,p,.. ,pnN) =nt+2Sm.(8)

Similarly, from (2), (4) and (7), we see that

iP] («+ *■)(« +80-(« + *»)
^ ' log pj log p3 log p3.. . log p„
_ (2Sm XSm‘ , 28,/
* P| t + St3
logp, logp2 logpa... logp„
n2|tsl-(2S,N . «>2^-(2S,„)‘
~~O ^-+ —3nV -
' ‘°g 2-’i l»g P* l°gPs ■ ■ ■ l°gPn
- logp1logp2...logp„ L[1 - i (log*rtn*SV
5 ' 1 v 8,■m-i
m - n(S
.....(9)
to choose

d (iV) = {1 _ o (log .
logp1logp.2...logp„ 6 1
where the symbol 0 has its ordinary meaning.
The following examples shew how close an approximation to d (N) may
be given by the right-hand side of (3). If
W=2’3.7!S,
then, according to (3), we have
d(IV)< 189.8-00000685...; .... ,(ii)
and as a matter of fact d(N) = 1898. Similarly, taking
W = 2“8.3s58,
* Werke, Vol. 1, p. 635.
 Highly Composite Numbers 83

we have, by (8), d (N)< 204271-000000372...; .(12)

while the actual value of d (JST) is 204271. In a similar manner, when

we have, by (3), d (iY) < 7462O00412...; .(13)

while actually d (N) = 74620. .

3. Now let us suppose that, while the number i of different prime

factors of iY remains fixed, the primes p,, as well as the indices a.„, are
allowed to vary. It is evident that considered i j a function of iY, is
greatest when the primes pv are the first n primes, saj 2, 3, 5, .... p, where
p is the nth prime. It therefore follows from (3) that
a (in < {(i/»)iog(2.3.s...i>..y)}« .(14)
' ' log 2 log 3 log 5 ... logy)
and from (10) that it is possible to choose the indices so that

4. Before we proceed to consider the most general case, in which nothing

is known about iY, we must prove certain preliminary results. Let tt(x)
denote the number of primes not exceeding x, and let
S- (*) = log 2 + log 3 + log 5 +... + logp,
and ot(«) = log2 .log3 .log5 ... logp,
where p is the largest prime not greater than a; also let <f> (t) be a function
of t such that <£' (t) is continuous between 2 and x. Then

J*7r(t)<f>'(t)dt = J + («)&+3jV (t)dt

+ 4 j cj>' (t)dt + ... +ir(as)J cf>'(t)dt

- W (3) - 4> (2)} + 2 (6) - * (3)} + 3{4>(7)-t (5)}

+ 4 (11) - * (7)} + ... + *■(«){* («) - * OOJ
= tt (x) <j> (x) — {cf> (2) + </>(3) + $ (5) + ... +<p(p)}. ...(16)
As an example let us suppose that <j> (t) = log t. Then we have

"■(*> log x-S(x)=f*^dt.(17)

Again let us suppose that <j> (t) = log log t. Then we see that

7T (*) log log x - log «r (x) = dt. .(18)

84 Highly Composite Numbers

^(*)logjjgj}-log^(«)

= - (a) log jl - J'Z® *} < - l^/2 *’

... f $M 1, 1 f^(«), *■ (a) log «-*(*) [**■(()

M^fcmgzj+bgzJ'~* —*<«)io*— J. *
-st&r.U.T-J'
It follows from (19), (20) and (21) that

"J, U(log»W. * ^ M«0

Now it is easily proved by elementary methods* that

' \log x) p Sr (a) W’

~P dt) du=J 0 {(j—^y3l ** = 0 {(W^l;

ana &(*)log®{/2 i *} ^ (x) log x u |(log *)!J wl(loga)3|

Hence we see that ^i^Ml^ = eoW(io8*)':.
n(a)

* See Landau, Handbuck, pp. 71 et seq.

 Highly Composite Numbers 85
5. We proceed to consider the case in which nothing is known about N■
Let

Then it is evident that d (N) = d (N'), and that

S- (p) «log N'« log N. .(23)
It follows from (3) that
1 (Mp) + logiV'')’* !»
d(N) = d(N')<
«(p)t *(P)
, , logff]’tfifr(p)Mp))J|*
*1 + »(P)J »(P)
_ f logoi'w e0lvm L
“t »(p)J 1 &(P) J
:e of (22) and (23). But from (17) we know that
w (p) log p-%(p) = 0 (j^) ; .
S- (p) = 7T (p) {logp + 0 (1)} = tr (p) {log Sr (p) + 0 (!)]•

«-(j>)-a(p)fT
’lloga(p)T {log&(p)laJ.
It follows from (24) and (25) that

Writing t instead of S- (p), we have

logWy"Fr+0<W .
d(N)^(l
and from (23) we have t < log W. .(27)
Now, if N is a function of t, the order of the right-hand side of (26), con¬
sidered as a function of IV, is increased when N is decreased in comparison
with t, and decreased when N is increased in comparison with t. Thus the
most unfavourable hypothesis is that W, considered as a function of t, is as
small as is compatible with the relation (27). We may therefore write
log iV fori in (26). Hence
d(N) < 2>^^+0(Wfev7’, .(28)
for all values of N*.
* If we assume Miking about v (a), we can shew that
lQSW . -logJlQKloglOBff
d(X)<2W*s OoslogW)- .

d (17) < 2 l°s 1<Js ^ +11 + “(1>I (Ff? h')’.

If we assume that irW=k^+0(i4ft’
logy msw iogw
d(N) < (logiog-V)^
80 Highly Composite Numbers

The inequality (28) has been proved by purely elementary reasoning.

We have not assumed, for example, the prime number theorem, expressed by
the relation

We can also, without assuming this theorem, shew that the right-hand
side of (28) is actually the order of d(N) for an infinity of values of JV.
Let us suppose that
JV=2.3.5.7 ...p.

Then d(JV) = =
in virtue of (25). Since log JV = ^ (p) = t, we see that
d (JV) = (ES&p;
for an infinity of values of IT. Hence the maximum order of d (N) is
2i5ifeiv+0(iogrfiilv)V.

III.
,The Structure of Highly Composite Numbers.
6. A number N may be said to be a highly composite number, if
d(N')<d(N) for all values of JV' less than JV. It is easy to see from the
definition that, if N is highly composite and d (JV') > d (JV), then there is
at least one highly composite number M, such that
N<M4N. x.(29)
If N and JV' are consecutive highly composite numbers, then d(M)^.d(N)
for all values of Mbetween N and JV'. It is obvious that
d (JV) < d (2JV) .(30)
for all values of JV. It follows from (29) and (30) that, if JV is highly
composite, then there is at least one highly composite number M such
that jV<j¥<2JV. That is to say, there is at least one highly composite
number JV, such that
x<N^2x, .(31)
if as > 1.
7. I do not know of any method for determining consecutive highly
composite numbers except by trial. The following table gives the con¬
secutive highly composite values of JV, and the corresponding values of
d (JV) and dd (N), up to d (JV) = 10080.
The numbers marked with the asterisk in the table are called superior
highly composite numbers. . Their definition and properties will be found
in §§32, 33. •
* 4 (.V) ~ * <*) means that $ (*)/* (*) ^ 1 as * * co.
88 Highly Composite Numbers

56 8640 = 2“.3“.5 3373164194400=2“. 3*. 5s.72. 11.13.17.19.23

33 9216 =210.32 4497552259200 = 2'. 3“.52.7s. 11.13.17.19.23
72 10080=2“.32. 5.7 6746328388800=2". 3*.52.72.11.13.17. 19.23

8. Now let us consider what must be the nature of N in order that N

should be a highly composite number. In the first place it must be of the

2“=. 3“3.5®s. 7®'... p1"r,,

where a* > a, & a, >...>(»,,> 1.(32)
This follows at once from the fact that
d W* ) = d (2^. 3*. 5®»... pfr ),
for all prime values of ot3, zrs,...,ia-Pl,
+ See Appendix, p. 339.
 Highly Composite Numbers 89
It follows from the definition that, if N is highly composite and
N'<P, then d(N') must be less than d(N). For example, JJTcJT’,
and so d(£N)< d (N). Hence

Kx-y>K-y.
provided that N is a multiple of 3.
It is convenient to write
«a = 0 (X>ft).(33)
Thus if N is not a multiple of 5 then as should be considered as 0.
Again, aPl must be less than or equal to 2 for all values of px. For let
P, be the prime next above pl. Then it can be shewn that P,<Pi for
all values of pC. Now, if uPl is greater than 2, let

Then N' is an integer less than N, and so d (iV') < d (N). Hence
(1 + 0 >2(^-1),
or 3 > aPl,
which contradicts our hypothesis. Hence
.(34)
for all values of p,.
Now let p[', pl, pl, P,, PI be consecutive primes in ascending order.
Then, if Pl ^ 5, aPl« must be less than or equal to 4. For, if this were not so,
we could suppose that

But it can easily be shewn that, if p1 > 5, then

(piy>Pr,
and so N' < N and d (N') < d (iV). Hence
a + v)>!(V“8).(86)
But since aPl« >5, it is evident that

* It can be proved by elementary methods that, ii+ ^ 1, there is at least one prime p
such that x<p^Sx. This result is known as Bertrand’s Postulate: for a proof, see
Landau, Handlmch, p. 89. It follows at once that Px < pl, if pt > 2; and the inequality
is obviously true when pi — 2. Some similar results used later in this and the nest section
may be proved in the same kind of way. It is for some purposes sufficient to know that
there is always a prime p such that x < p < 3x, and the proof of this is easier than that
of Bertrand’s Postulate. These inequalities are enough, for example, to shew that
log P!=log p! + 0 (1).
90 Highly Composite Numbers

which contradicts (35); therefore, if rpi >*>> then

<V<4.(36)

Now let
Pi Pi
It is easy to verify that, if 5 «19, then
PiPi>Pi"Pii
and so N'<N and d (N‘) < d (If). Hence
(1 + ^ (1 + av.)(1 + a„..) > 2anap, (2 + ap.),

or (1 + i)(1 + i)>2(1+iT^.)'
But from (36) we know that 1 + aPl» < 5. Hence

.(S”
From this it follows that aPl = 1. For, if aPl ^ 2, then

(1 + ^)(1 + ^)<2*’
in virtue of (32).' This contradicts (37). Hence, if 5 <pj< 19, then
«*-l.-(38)
Next let N' = NPNiKpiPiP").
It can easily be shewn that,' if p, > 11, then
PiPf <pipip";
and so N' < N and d (N') < d(N). Hence
(1 + aPi) (1 + aPl0 (1 + %,,») > iaPlaPl. aPL",

.•.■<”>
From this we infer that aPi must be 1. For, if aPl^2, it follows from (32)
that

which contradicts (39). Hence we see that, if p, > 11, then

a*,-l.(40)
It follows from (38) and (40) that, if p, ^ 5, then
«*, = !. (41)
But if p1 = 2 or 3, then from (34) it is clear that
«*, = ! or 2.(42)
It follows that a,,, = 1 for all highly composite numbers, except for 22,
and perhaps for certain numbers of the form 2“ . 32. In the latter case a ^ 2.
 Highly Composite Numbers 91

It is easy to shew that, if a > 3, 2“. 3! cannot be highly composite. For if

we suppose that
N’ = 2<-‘.3.5,
then it is evident that JV' < N and d (N') < d (i\r), and so
3 (1 + a) > 4 o,

Hence it is clear that a cannot have any other value except 2. Moreover we
can see by actual trial that 22 and 22.32 are highly composite. Hence
aPi = l .(43)
for all highly composite values of N save 4 and 36, when
aPl — 2.
Hereafter when we use this result it is to be understood that 4 and 36 are
exceptions.
9. It follows from (32) and (43) that H must be of the form

x2.3.5.7...p2
x 2.3.5 ...p3
...(44)
where px > p% ^p3 > ... and the number of rows is a2.
Let Pr be the prime next above pr, so that
logP, = logp, + 0 (1), . ...(45)
in virtue of Bertrand’s Postulate. Then it is evident that
“jv > ...(46)
...(47)
It is to be understood that ap, = 0, , ...(48)
in virtue of (33).
It is clear from the form of (44) that r ca exceed o2, and that
.(49)

10. Now let

where v ^ pu so that N' is an integer. Then it is evident that H' < N~ and
d(N')<d(N), and so
(1 + «.)(! + «.) » °.(l + «+ [Jjj]) .
.
* [*] denotes as usual tlie integral part of cc.
.(50)
92 Highly Composite Numbers

Since the right-hand side vanishes when v >plt we see that (50) is true for
all values of X and v *.
Again let N'=*N/iX 1 i'08*1'108^,
where [log /i/logX] < cy, so that iV' is an integer. Then it is evident that
N' < N and d (If) < d (N), and sc
iog^l) ...(51)
(l + ^)(l + o*)>(2 + o,)(o»-p
logxjy
Since the right-hand side is less than or equal to 0 when
“A<log^/logV|,
we see that (51) is true for all values of X, and p.. From (51) it evidently
follows that
(l+0 < (2 + ‘V)[^2^l)]- ...(52)

From (50) and (52) it is clear that

.<ss>
for all values of X, p and v.
Now let us oppose that v=pi and so that a„ = l and <y = 0.
Then we see that

KS]»“*I3£].<“>
for alt values of X. Thus, for example, we have
Pi = 3, l^a2«4;
p, = 5, 2<as^4;
Pi= l 2
2<a «6;
^= , ^ ^ ;11 3 05 6
and so on. It follows from (54) that, if X then
ax log X = 0 (log pi), ak log X ^ o (logp,).(55)
11. Again let
IV' = ArX['/f<1+“*+“'*) l08'l/1°s4] -l-[Vl(l+«A+»(i)IOSA/IOgrf]j
and let us assume for the moment that
cy > «/[(l + cii+ cy) log X/log p),
in order that N' may he an integer. Then AT'cIF and d (IP) < d (FT),

(1 + as*) (1 - n)> {1 + <y + [V{(1 + ai + an) log /i/log X}]}

x - [VRl + «a + log X/log /a)]}
> {“a + V{(1 + “a + a^ log /i/log X}}
x {un- Vi(l + «a + cy) log X/log /a)}.(56)
* That is to say al ime values of X and v, since X in aK is by definition prime.
 Highly Composite Numbers 93

It is evident that the right-hand side of (56) becomes negative when

a* < 4(1 + “a + <v) log A/log p],
while the left-hand side remains positive, and so the result is still true.

a„ log p - a* log A < 2 4(1 + «A + «*) log X log p),.(57)

for all values of A and p. Interchanging A and p in (57), we obtain
log X - a„ log p < 2 \7{(1 + “a + a„) log A log p}.(58)
From (57) and (58) it evidently follows that
j aK log A-a„logp|<2 4(1 + aK + af) log X log p}.(59)
for all values of A and p. It follows from this and (55) that, if A and p are
neither greater than plt then •
fix log A - a„ log p = 0 V{log Pi log (Ap)|, .(60)
and so that, if log A = o (log p{), then
a2 log 2 - a, log 3 ~ a5 log 5 ~ ... - a* log A.(61)

12. It can easily be shewn by elementary algebra that, if x, y, m and ?i

are not negative, and if
k - V | < 2 V(«m + ny + mn),
then !x/(a+ft)-v'(2/ + m)|<V(m + »0;l
\g(x+n)~ f(m + n)\<g(;y+m).} .^ ‘
From (62) and (59) it follows that
| Vl(l + “a) log A) - 4(1 + <V) log/*) i < V{log (Ap)}.(63)
and | 4(1 + ®A)log A} — x/(log(Ap)) [ < 4(1 + “*) l°gp}> .(64)
for all values of A and p. If, in particular, we put p = 2 in (63), we obtain
4(1 + l°g 2! - V{log (2A)j < 4(1 + ax) log A}
< 4(1 + «a)log2} + V{log(2A)}, ...(65)
for all values of A. Again, from (63), we have
(1 + aA) log A < (4(1 + a„) log v} + x/(log (Av)))2,
or ax log A < (1 + a,) log v + log v + 2 4(1 + a„) log v log (Av)}. ...(66)
Now let us suppose that A < p. Then, from (66), it follows that
ax log A + log p < (1 + a„) log v + log (pv) + 2 4(1 + a,) log v log (Ai/)}
« (1 + a„) log v + log (pv) + 2 4(1 + a„) log v log (py)}
= (4(1 + a„) log v\ + 4og (pv)|2, .(67)
with the condition that A < p. Similarly we can shew that
aA log A + log p > (4(1 + a„) log v} - 4°g (pv)}2, .(67')
with the condition that A^p.
 Highly Composite Numbers

U' = — 2P°s*/f,rMlog2H gC'osA/f'MinssH ... ^Uog-VKriioBri^

where irMlog/r < logX^logp,. Then it is evident that N' is an integer

less than N, and so d (iV') < d (A). Hence

(l+I) (1 + o.) (1+«.)(! +a.)-(1 + ^

r logX } ( log^ | L,_J?g?L_l-

>‘[ai + 7r(ja)log2j r ir0“)log3j "'1 M p(p) logoi’
that is , , ,,
{“* log 2 + {«* log 3 + ^T)('' • f -l0g "■+

< (l + i) (a, log 2 + log 2) (o> log 3 + log 3)... (or* log p + log p)

^ (i + i) (ct, log 2 + log p) (a, log 3 + log p)... (a* log p + log p).

In other words,

I, ^r10")!,, v«~k’8'‘} {1+Ib^!i}

[ a2log2 + log^J [ ttjlogS + logjaJ ( oMlogp + logp)

> l lVi(l + a,)logW + Vlog(^)lsJ ’ .

where v is any prime, in virtue of (67). From (68) it follows that

VRl + a,)logr} + Vlog(^)>

provided that ir (p) log p < log \ < logp,.

14. Again let

iV' = .AX,2-1_Cl0|!A/1’<'‘,ll*!H g-l-[logA/[»MlonS>] -l-[logM»<l«)logM)]
where p ^Pi and \> p. Let us assume for the moment that

a*1<®
for all values of k less than or equal to p, so that N' is an integer.
 Highly Composite Numbers

Then, by arguments similar to those of the previous section, we c,

that

£8+log^ l*
2 + Oj I (VK1 + a„) log v) - Vlog (jiv)J!J
From this it follows that

| Vl(l + «v) log v] - Vlog (H j <

provided that fazpi and /j, < X. The condition that

-as-:)”
a* log k > {log \./tt (//.)}
is unnecessary because we kno'w from (67') that

j Via + log ») - Vlog (H; < V(a« log * + log n) 4, J+ log /ij ,

when a, log k < (log X/vr (/it)},

and the last term in (72) is evidently less than the right-hand side of (71).

15. We shall consider in this and the following sections some important
deductions from the preceding formulas. Putting v = 2 in (69) and (71), we
obtain
f ^-lo,„ 1
tQ») ■l-Vlog(2/t), ...(73)
via+«=) log 2} >
Mr
provided that 7r (/t) log fi < log X < logy,, and

via+log 2l <
\2 -f d\J
provided that fi 4,p-i, and Now supposing that \=px . (73), and
X = P1in (74), we obtain
/p2IP._l0g J
Vl(l + aj log 2} > J _ vlog (2/t), . .(75)

provided that 7r (,u) log/i < log)),, and

via + °») i°g 2} Vlog (2/i), . .(76)

4 provided that p, In (75) and (76) obtain the

 96 Highly Composite Numbers

best possible inequality for a,. If p, is too small, we may abandon this result
in favour of
.(77).

which is obtained from (54) by putting X = 2.

After having obtained in this way what information we can about a3,
we may use (73) and (74) to obtain information about a*. Here also we
have to choose p so as to obtain the best possible inequality for a*. But if X
is too small we may, instead of this, use
V[(l + «2) log 2} - Vlog (2'X.) < V{(1 + a*) log X}
< VRl + «*) log 21 + Vlog (2\), .(78)
which is obtained by putting p = 2 in (63).
16. Now let us consider the order of tu. From (73) it is evident that,
if 7r (p) log p < log X < logpl: then

(I + a,) log 2 + log (2p) + 2v/{(l + a.) log 2 log(2/i)}:

But we know that for positive values of x,

(<*5T

Hence 0(1)

Again from (55) we know that a2=0(logpJ). Hence (79) may be

written as
a, log 2 + 0 V(logPi log p) + 0 (log p)

But log p=0(pah),

log A n (log X log

l A* I’
 Highly Composite Numbers 97

Again ^ >2V(logft log/“);

and so v/(logy, log „) « 0 (’0g^ + .

Hence (80) may be replaced by '

',os2>1^7I)+°(,?c?s-!!+^.
provided that it (/a) log/r < logX^logy,. Similarly, from (74), we can shew
that

provided that p^p, and p<\. Now supposing that X=y, in (81), and
X = Pi in (82), and also that
/* = 0 <J( logy, log log pi), fj.^0 V(logy, log logy,)*,

we obtain ctj log 2 ^ + 0 y^logy, log log pi),

^log 2 ^ K§f+0 loS l0«P>).

From (83) it evidently follows that "

“2 log 2 = fo|f + 0 V(log p, log log pi). (84)

And it follows from this and (60) that i{\^p, then

ak log X=+ 0 {v(log j>. log X) + V(logpi log logy,)]. (85)

Hence, if logX = o(logp,), we have

Oc log 2 ~ Os log 3 - ac log 5 ~ log P (86)

log 2 '

17. The relations (86) give us information about the order of a, when
X is sufficiently small compared to y,, in fact, when X is of the form y,%
where e-»0. Such values of X constitute but a small part of its total range
of variation, and it is clear that further formulae must be proved before we
can gain an adequate idea of the general behaviour of aFrom (81), (82)
98 Highly Composite Numbers

and (84) it follows that

log\ _ log# , n JlogXjogp. + fi}_°8Pi + ^(logPlloglog#)f>
-"^log2 1 P logx J
log ^1 +

log* > l0&£; + 0 + ^L+v(log#ioglog#)},

1 + “A' .(87)
provided that w GOlog /. < log X * log #. From this we can easily shew that
if 7T (,«) log ya < log X «log# then

Now let us suppose that

log^^VfbSS^)-
Then we can choose p. so that

• -’WflSF)}-
Now it is clear that log /a = 0 (loglog#), and so
logja_Q ^log log PYj = q |^(logPil°glog?i)j ;

, g# nJ\/(l£g#loglog&)l
and (i^r°i Togx ■■ r
From this and (88) it follows that, if

g'ojr,)'

*- .,s(2—-„-.+ „pSr|fim>U ^

*><*■»*":-»-*♦«P-^1-3P•) .
Now we shall divide the primes from 2 to# into five ranges thus

ilpiY «{K(lp)*} e^lPhp)hi

 Highly Composite Numbers 99

We shall use the inequalities (89) to specify the behaviour of aA in ranges

I and II, and the formula (85) in ranges IV and V. Eange III we shall
deal with differently, by a different choice of /x in the inequalities (88).
We can easily see that each result in the following sections gives the most
information in its particular range.
18. Kange I: log A # 0 V(log p, log logp,)*.
Let A = [(21oeA/1osj,'-1)-’],
and let (2u»v‘OS3’- -1)'1 + eA,
where — \ < eA< £, be an integer, so that
(2ioB*, _ 1)-1 = A + l .(90)
when e\ > 0, and (2loBi,I°S!’'-1)“'= A - .(91)
when ex < 0. By our supposition we have
V(logpi]oglog£i)_
log A. ~0(i).W
First let us consider the case in which
e , q fVClogPiloglogPi)]
1 logV J’
-that y(lgg^oglo_gpO=ofa).(93)

It follows from (89), (90), and (93) that, if ex > 0, then

...
Since 0 < ex < J, and and A are integers, it follows from (94) that
nx«;A, ax>A-l...(95)
Hence «x = A.(96)
Similarly from (89), (91), and (93) we see that, if eA< 0, then
a^A-ex + o(ex), |
ax»A-l-ex + o(ex).| .(97)
,Qln
Since - J < ex < 0, it follows from (97) that the inequalities (95), and there¬
fore the equation (96), still hold. Hence (96). holds whenever
.m
* We can with a little trouble replace all equations of the type /=0(<|>) which occur
by inequalities of the type |/| < K$, with definite numerical constants. This would
enable us to extend all the different ranges a little. For example, an equation true for
logX^fcOx/Clogp,)
would be replaced by an inequality true for logX > iTV(l°gPi), where E is a definite
constant, and similarly log X=o ./(log p,) would be replaced by log X < k ./(log p{).
 Highly Composite Numbers

In particular it holds whenever

* ^0(1), .(99)
Now let us consider the case in which
0 fVQogpiiogiogyOI;.(100)
1 log J
so that ex = o(l), in virtue of (92). It follows from this and (89) and (90)

milarly from (89), (91), and (100), we see that, if

a^A + o(l), 1
a^A-l + o(l).i -•

For example, let us suppose that it is required to find aA whe

We have *
(giogyiogp, _ = (2„s _ !)-i + o (j) = n-048 ... + o (1).
It is evident that A = 11 and eK j* o (1). Hence a* = 11.
19. The results in the previous section may be rewritten w
modifications, in order that the transition of a*, from one value t
may be more clearly expressed. Let

3t x + si, where — J < < J, be an integer. Then the range ol

e now considering is
/( l°gTi \
V Vloglogp,/’

a particular case of this we hav

when e„ifco(l). But if

-oM'lgr)}.
- then when e* > 0 aA = [®] or [a +1]; ..(110)
and when < 0 aA = [ffi] or [a-1].(110')
 Highly Composite Numbers

log X — 0 V(logPi log logp,), |

From (89) it follows that

a» = (21obV1°m' -1 )-i + 0 ^(lQg^logIog^)| .

(2iogViosr, _ X)-.=j + o (1).

ax log X = + 0 '/(1°S Pl l0g l0g Pl)' ''

tample we may suppose that

zi. rvangexxx. -g " = - V

logX^oOogp,)*.
Let us suppose that p. = 0 (1) in (88). Then w

= g|f+ o + Vdogp, log logiJl)}... .(114)

iogiaogx=0(logX)= /log^y

V(logPi log log Pi) = o (~|P‘) .

22. Range IV: log x=o (log pN, 1

log X ^ o (log log pi).)
102 Highly Composite Numbers

In this case it follows from (85) that

ak log X = + 0 V(log Pi log X). .(116)

As an example in this range, when we suppose that

we obtain from (116)

23. Range V: logX = 0 (log log p,)-

From (85) it follows that
ak log X = 1^| + 0 V(log?i log log#). .(117)

For example, we may suppose that

Tten ^ a-log2V(log«+0V(1OgPl)-
24 Let X' be the prime next below' X, so that X' ^ X — 1. Then it follows
from (63) that
V{(1 +av)logX'] -\/|(l +01ogX} >-^log(XX').(118)
Hence V{(1 + av)log(X- 1)}-V{(1 + «x)logX) >-V{21ogX}. ...(119)

But log(X-l)<logX-i<logx(l— i
and so (119) may he replaced by
V(1 + «»•) - V(1 + <*0 > V2.(120)
But from (54) we know that
i + 0v > ! + f!2££>l > !£££■ > l££Pi.
^ |>gvj i°g^’ iogx
From this and (120) it follows that

va+«»wa+«.)> .(iso
Now let us suppose that X2(logX)3< Jlog#. Then, from (121), we have
V(l + av)-V(l + ax)>0,
or ak->ak.(122)
From (122) it follows that, if Xa (log X)3< £log#, then
d1>as>Os>‘h> ■■■ >®x- ••• .(123)
In other words, in a large highly composite number
2»=:3a>.5“'.7
 Highly Composite Numbers 103

the indices comparatively near the beginning form a decreasing sequence in

the strict sense which forbids equality. Later on groups of equal indices will
in general occur.
To sum up, we have obtained fairly accurate information about a* for
all possible values of X. The range I is by far the most extensive, and
throughout this range a,, is known with an error never exceeding 1. The
formulfe (86) hold throughout a range which includes all the remaining
ranges II—V, and a considerable part of I as well, while we have obtained
more precise formulae for each individual range II—V.

25. Now let us consider the nature of pr. It is evident that r cannot
exceed i.e., r cannot exceed

(Toffy*+ 0 Jo§ .(124)

From (55) it evidently follows that
<br l°SPr = 0 0°g?h)> 1 .(125)
^gprfo(\ogPl): J
(1 + «p,) log^),. = 0 (log pi), ]
(1 + aPr) log 2>rfo (logpi. J .(126)

But from (46) we fa: that

a^hgpr^rlogp,., 1
{l + aPr)\ogpr*zr\ogpr,) .(127)

From (125)—(127) it follows that

rlogpr = 0 (logpi),
rlogpt fo (log pi); .(128)
and aPr = 0 (r),)
^fo(r).f .(129)

log i1 +
But, from (47), we have

1 + up J ’
Also we know that
 104 Highly Composite Numbers

Hence (131) may be replaced by

.(132)

From (130) and (132) it is evident that

.(133)

.(134)

Now supposing that i'fi = o (logp,)> 1 .(135)

rp + O (log ri, j
and dividing (134) by (133), re have

log(l + A) U icgpj

' r+W=1 + i+0 (fifr+&/

that is
u^r£{1+0(^+ioik)f-
Hence apr = aPr + l + o(^ + ^, ... .(136)

in virtue of (129). But aPr^, r — 1, and

.(137)
But we know that a^^r. Hence it is clear that

.(138)
From this and (136) it follows that
.(139)

provided that the conditions (135) are satisfied.

Now let us suppose that r = o VOogjp,). Then we can choose /a such
that r> = o(logp1) and0(1). Consequently we have

^=o(1)> i$k=o(1);
and so it follows from (138) and (139) that
.(140)
 Highly Composite Numbers 105

provided that r = o \/(logpi). From this it is clear that, if r = ov'0ogPi).

pi>p„>p.>p,>... >pr. .(HI)

In other words, in large highly composite number
2a‘. 3°s. 5“< ...pu
the indices comparatively near the end form a sequence of the type
...5...4...3...2 ...1.
Near the beginning gaps in the indices will in general occur.
Again, let us suppose that r = o(logp,), r f=o yllogp,), and p=0( 1)
in (138) and (139). Then we see that

=r+0(i^
v=r+0fe)i
provided that r.= o(logp,) and r o V(logp,)- But when r^o(logp^, we
shall use the general result, viz.,
(V = 0(r), a^olr),
.(143)
apr= 0 (r), aPr o (r),
which is true for all values of r except 1.
27. It follows from (87) and (128) that

, ,HP\

‘ .(144)
with the condition that rp = o (logp,). From this it can easily be shewn, by
arguments similar to those used in the beginning of the previous section,
that
\og(l+ljr) = ^° {l0gygM + ^ + VOogp.loglogp,)}, -(145)
provided that rp — o(Logpi).
Now let us suppose that r = o (logp,); then we can choose p such that

/•-ofP), ^0(1).
Consequently rp = o (logpd and log p — o (p), and so
106 Highly Composite Numbers

From these relations and (145) it follows that, if r=o (logp,), then
_HP'- .(146)
log(l + 1/r) log 2’
that is to say that, ifr=o (log pi), then
log_ !og^ _ l°gP> _ _ . ..(147)
log 2 log(l + J) log(l + J) ' log(l + l/r)
Again let us suppose that r— 0 \/(logpi log logp,) in (145). Then it is
possible to choose p such that
rp- OVOogpiloglogp,),) (148)
rpfo V(logp, log logp,).) .
It is evident that logp = 0 (log logp,), ancl so
0 = o V(logp, log logp,).

in virtue of (148). Hence

log aTi/r)= ‘21+0 log l0Slh)’ .(149)

provided that r = 0 V(logp, log log p,).
Now let us suppose that r = o (logpi), r^of^ogpPoglogpi), and p — 0(1),
in (145). Then it is evident that
logp, = 0 (rvXlogp, log logp,) = 0 (r),

and

Hence we see that ^ + 0 (r), .(150)

log(l + l/r) log 2 ^ '
if r = o(logp,); r^oV(logp,loglogp,).
But, .if r f o (logp,), we see from (128) that

.
logCL + lM*0^^- j
From (150) and (151) it follows that, itrfo V(logp, log logp,), then
logpr _logp, n(y
log(l + 1/ry log 2 + { .(152)
and from (149) and (152) that, if r = o(logp,), then
log Vr log Pi
log(l + l/r)~ log2’
in agreement with (147). This result will, in general, fail for the largest
possible values of r, which are of order logp,.
 Highly Composite Numbers 107

It must be remembered that a,11 the results involving », ma,y be written

in terms of IV, since p^ = 0 (log IV) and p, ^ o (log IV), and consequently
logp, = ]!ogloglV + 0(1). .(153)

28. We shall now prove that successive highly composite :numbers are
asymptoticsilly equivalent. Let to and n be any two positive integers which
are prime to each other, such that
logBrn-oflogpO-ofloglogiV); . .(154)

and let 2S= .34*.5!*...p‘s>.!. .(155)

Then it is e vident that
mn — 2i*»l. 31*«l.5l®*l... pl*t»l. .(156)
Hence SilogX = 0(logm)i) = o (logp,) = o (a* log X); ... .(157)
so that * = o(a*).

Now d\^lv) = <Z(lV)(l + r^ -). (158)

But, from (60), we know that

aK log X = a, log 2 + 0 V(log p, log X).

Hence 1-
'Tfk = 1 + + 0 {1 (SI)}
_ 8>logX
^ + a2 log 2 + »iw£via

logX
~ exp |a2 log 2 .(159)
It follows fr<jm (155), (156), (158), sind (159) that
+ S3 log 3 + ...H-SgalogfO
r)-JW«P {&«■ 05 log 2
+ 0I«,|1<*2 +■i8,|log3 + ... + !Ss,|logp //'logmnNl
log ft V \. log Pi ))

= d(N)e o.ios-2 10(l5^)‘

1 fl m
.(160)
Putting m = n +1, we see that, if
log n = o (log Pi) = 0 (log log IV),
 Highly Composite Numbers

d |^(n- =d(iV)<

1 {” 1°8 (log tog y)}

= d(W)(l + ^
;o choose n such that
n(lognf fo f(hglogJSf),

I + Oj.log.y^
Jog log N J) '

d w(i + - d(N). ..

From this and (29) it follows that, if N is a highly composite number, then the
neat highly composite number is of the form

— ws&w, .<“>
Hence the ratio of two consecutive highly composite numbers tends to unity.

0 flog «V(iog log tc)\

I (log log log *)» J'
29. Now let us consider the nature of d (N) for highly composite values
of N. From (44) we see that
d{N) = . 3'W-'<w. 4'(^'-’'<w ... (1 + a,). ...(164)
From this it follows that
d (N) = 2“*. 3“”. 5“*... .(165)
where m is the largest prime not exceeding 1 +0,; and
. a* = tr(p^) + 0(p,).(166)
It also follows that, if px, pa, ps, are a given set of primes, then a
number fi, can he found such that the equation
d(H) = p,8'. p/=. pf’..., pA • ■ ■ pA
is impossible if JV is a highly composite number and fa > jl. We may
state this roughly by saying that as N (a highly composite number) tends
to infinity, then, not merely in N itself, but also in d{N), the number of
prime factors, as well as the indices, must tend to infinity. In particular
such an equation as
d(N) = k. 2”*.(167)
where k is fixed, becomes impossible when m exceeds a certain limit depend¬
ing on k.
 Highly Composite Numbers 109
It is easily seen from (153), (154), and (165) that
w = 0 (as) = 0 (log#) = 0 (log log iV) = 0 [loglogd(IV)},)
o- + 0 (a„) = o (log#) = o (log log IV) = o [log log d (IV)}. J ''
It follows from (147) that if X = o (log#) then
log<
- ...(169)
log(l-l-) log(l-J) log(l-i) log(l-l/X)'
Similarly, from (149), it follows that if X = 0 ^(log# log log#) then

^Jf+ovdog^iogiog#).(iro)

Again, from (152), we see that if X ± o ./(log# log log#) then

fcgagHss*™.<m>
In the left-hand side we cannot write a* instead of 1 + aA, as a* may be zero
for a few values of X.
From (165) and (170) we can shew that
log d(N) = «s log 2+0 («„), log d (IV) f a, log 2 + o («„);

and so log d (IV) = cts log 2 +elog!! ‘ '

But from (163) we see that
log log d (N) = log# + 0 (log log #).
From this and (172) it follows that
logS , - /(-toKloslOK^OVn
«2 log 2 = log d (IV) — {log d (IV)}10®2 + ^ loBiosdiw) /. .

30. Now we shall consider the order of dd{N) for highly composite
values of N. It follows from (165) that
log dd (IV) = log (1 + a,) + log (1 +«,)+... + log (1 + «„). .. .(174)
Now let X, X', X", ... be consecutive primes in ascending order, and let
X = 0 v (log # log log #),
X^ o V(log#loglog#).
Then, from (174), we have
log dd (IV) = log (1 + as) + log (1 + a,) +... + log (1 + a*)
+ log(1 + «v) + log(1 + «»») + ... + log(1 + O. ...(175)
* More precisely Er~a2« But this involves the assumption that two consecutive
primes are asymptotically equivalent. This follows at once from the prime number
theorem. It appears probable that such a result cannot really be as deep as the prime
number theorem, but nobody has succeeded up to now in proving it by elementary
110 Highly Composite Numbers

But, from (170), we have

log (1 + a.) + log (1 + a») + ... + log (1 + aA)

-tK'M'1-* io-ku-b-M}
+ 0 Vfiogpi log logpi) log |(i -1) (i - i) ■ ■ ■ (* - ^)| •
.(176)
It can be shewn, without assuming the prime number theorem*, that

- log ((1 - i) (1 - i) (1 - i) - (l - J)} = log log P + 7 + 0 (j^),

.(177)
where 7 is the Eulerian constant. Hence

log {(1 - i) (1 - i) (1 - i) ... (l - |)} = 0 (log log p).

From this and (176) it follows that

log (1 + a„) + log (1 + a,) + ... + log (1 + «A)

+ 0 lV(log Pi log log pi) log log ?l}

= ~ ijflog {(1
- 4)o- -i) ••• (J
- 9}
+ 0 (V(logp, log log pi) log log logpj.(178)
Again, from (152), we see that
log (1 + ay) + log (1 + ay.) + ... + log (1 + a»)

+ 0 jv log (l - ^) + *•" log (l - i) + ... + *rlog (l - ijj

■ " c|f ‘”g ft - b) (‘" >?)■ ■• ■ (‘ - ;)} ■0 (sfej ■ • ■ ■ l-m>

1
From (175), (178), and (179) it follows that
logddW = -^flogj(l-i)(l-i)...(l-i)}

+ 0 Mlogp! log log#) log log log#}

- i|flog K1 - h) i1 ~ ^-i1 ~ i)}+0 fcSy

* See Landau, Handbuch, p. 139.
 Highly Composite Numbers 111

- '£§?■“*» M+0 (iof¥gi;)!'+0 (sp,[rf,)

Jp'./ .(18°)
in virtue of (177), (168), and (163). Heme, if N~ is a highly composite
number, then / J.
d d (N) = (log ^*,<*I0*^*V+T + °fe«u53)}. .. .(181)
31. It may be interesting^*} note that, as far as the table is constructed,

’ 7.25, 7.2’", 9, 9.2, 9.22, ...’ 9.2>»,

and so on, occur as ,v;ilues of d(N). But we know from §29 that fc.2*
cannot be the vnlttfe of d (If) for sufficiently large values of m; and so
numbers of the form k.2m which occur as the value of d$N) in the table
must disappear sooner or later when the table is extended.
Thus numbers of the form 5.2”* have begun to disappear in the table
itself. The powers of 2 disappear at any rate from 218 onwards. The least
number having 218 divisors is
2\3'’.53.7.11.13... 41.43,
while the smaller number, viz.,
2s.3J.o3.72.11.13... 41
has a larger number of divisors, viz. 135.211. The numbers of the form
7.2’“ disappear at least from 7.213 onwards. The least number having 7.213
divisors is
2°.32..5s.7.11.13 ... 31.37,
while the smaller number; viz.
2».3J.52.72.11.13... 31
has a larger number of divisors, viz. 225.2s.

IV
Superior Highly Composite Numbers
32. A number iV may be said to be a superior highly composite number
if there is a positive number e such that
d(N) d(N')
N’ >{N'Y‘ .(182)
 112 Highly Composite Numbers

for all values of N' less than N, and

d(N) d_(N)
.(183)
N- ' W'Y
for all values of N' greater than N.
All superior highly composite numbers are also highly composite. For,
if iV' < iV, it follows from (182) that

and so N is highly composite.

33. Now let us consider what must be the nature of N in order that it
should be a superior highly composite number. In the first place it must be
of the form
2'o.3“».5»*...pI“.(184)
or of the form 2.3.5.7.pt
x 2.3.5.7...pa
x 2.3.5 ... ps

i.e. it must satisfy the conditions for a highly composite number. Now let
N' = Nj\,
where A, Then from (182) it follows that ;

.(185)

Again let A" = iVX.

Then, from (183),
+ 2 + CSa

.(186)
Now supposing that -pi in (1S5) and X = P, ii (186), we obtain

.(187)
logP, log pi ’
Now let us suppose that e = 1/x. Then, from (187), we have
p1<2*<P1. ...(188)
That is, p, is the largest prime not exceeding 2X. It follows from (185) that
ax*(Vl*~ 1)-. .(189)
 Highly Composite Numbers 113
Similarly, from (186), «* >. (A1;* — l)'1 - 1.(190)
From (189) and (190) it is clear that
«A = [(V/*-l)-].(191)
Hence N is of the form
2[e1/I-ir1]. 3[a'/*-„->]. sWM-*-*) .(192)
where p1 is the largest prime not exceeding 2*.

34. Now let us suppose that X=pr in (189). Then

«*,<(Pr‘/*- I)--
But we know that r%aPr. Hence
r«(p,.>/*-1)-

or *r<(l+J)‘.(193)
Similarly by supposing that \ — Pr in (190), we see that
«pr>(-P,'/*-l)--l.
But we know that r — 1 Ss apr. Hence
r>(P,M*- l)->,

« i>r>(i+9".(194)
From (193) and (194) it is clear that pr is the largest prime not exceeding
(1 + 1 jr)x. Hence N is of the form

X2.3.5.7 ...p,
x2.3.5...jp,
x.(195)
where pj is the largest prime not greater than 2* is the largest prime
not greater than (§)*, and so on. In other words 17 is of the form
e»(21+»(i)«+»(8*+...; .(196)
and d (N) is of the form
2^.(f )'(«•.(,)-(«■.(197)
Thus to every value of x not less than 1 corresponds one, and only one,
value of N.
d(N) d_(JT)
35. Since
N'l* * (Wfl*’
for all values of N', it follov i from (196) and (197) that

.(198)
e(l/*)*(2‘) e(l/j)&(H- JllxWir
 114 Highly Composite Numbers

for all values of iV and x: and d(N) is equal to the right-hand side when
iy=ea(2*)+»(2)'+9-( Sf+-.(199)
Thus, for example, putting x = 2, 3, 4 in (198), we obtain
d(iVXV< ZN), \
d(N)^8(3NIS5)K |.(200)
d (iV) ^96 (3iV/50050)s, J
for all values of iV; and d (N) = *J(2N) when N= 25.3; d (N~) = S (HN’/Sop
when i7 = 2s. 3=. 5.7; d (N) = 96 (317/50050)1 when
JV = 2s. 3*. 3J. 7.11.13.

36. M and iV are consecutive superior highly composite numbers if

there are no superior highly composite numbers between M and N.
From (195) and (196) it is easily seen that, if M and N are any two
superior highly composite numbers, and if M > N, then M is a multiple
of N: and also that, if M and N are two consecutive superior highly com¬
posite numbers, and if M > N, then M/N is a prime number. From this it
follows that consecutive superior highly composite numbers are of the form

where ■rrl, wj, ... are primes. In order to determine 7th 7t2, ..., we
proceed as follows. Let be the smallest value of x such that [2*] is
prime, x2 the smallest value of x such that [(§)”] is prime, and so on; and
let xlt x2, ... be the numbers xi,x2, ... arranged in order of magnitude.
Then 7r„ is the prime corresponding to ®„, and
i7 = 7rJ7raw,...7r„J .(202)
if <*•„+,.

37. From the preceding results we see that the number of superior
highly composite numbers not exceeding
lS»(2*)+»(t)*+»(i)*+... .(203)
is t (2*) + 7T (f )* + ir (D* -t- . . . .
In other words if § x < a„+1 then
n = ,r(2-)+*■(*)•+ *•(#)■ +.(204)
It follows from (192) and (202) that, of the primes ir,, 7rs, ..., 77-,,, the
number of primes which are equal to a given prime w is equal to
[(w>/*- 1)-].(205)
Further, the greatest of the primes 7^, 7r2, irt, ..., irn is the largest prime
not greater than 2*, and is asymptotically equivalent to the natural nth
prime, in virtue of (204).
 Hitjlib) Composite Numbers 11
'he foil, iving table give the values of and a-„ for the first 5 0 value
that is till reaches vt y nearly 7.
_ *£S2
log 2
_a

7rj=■

a, = |5E| = 2-3219 ...

log 2
7T5 = 2
a'S=SI=2'4094-'
7r,t =3
-=«=2'7095 -

—7 ,7 „ jg =2.8073...

=2
*’“S“8’1062-.
7T, =11 .= ^=3,^4...

■*-,„ = 13 =3,004...

7Tu = 2
-=5I)=3-8017-
7T12 = 3

*V = 5 ,6=^=3-0693...

*-,,=i7 .,= ^=,0374...

*-,==io ,= =^=4,479...

*-,6 = 2
a,6= i^l) =4'4965---
7rlr = 23
" log a -4s2So-
7T1S= 7
ffii!=iS)=4'7992-
*-,,, = 29 ,lf=l2£^=4-8579...
116 Highly Composite Numbers

TS= 3
— s®-49233

ira = 31

.*, = 2
— iwi-5 1808-
.!3 = 37 ^ = !2£^7 = 5-2094.

.24 = 41 M =5-3575.
log 2

.7725=43 x = l?li® = 5-4262 .

log 2

7r,0 = 47 ^^=5-5545.
log 2

77,2=5
*‘-.S-w34s-

_ .,= 53 =5-7279.
28 log 2

.25=59 =5-8826.
log 2

.,„ = 2 *M = =5-8849,

.n = ll *3I = j£gi2 =5-9139.

log (|)

.22=61 ,2= igi =5-9307.

.12=3 .3=51=6-0256.

.„ = 67 x = =6-0660.
log 2

.25=71 ^ =6-1497,
log 2

.36 = 73 _ l°g 73 _6.189s

Xx~ log 2

.32 = 79 ,37=^=6,037.

.» = 13 Si =63259 ■
.19 = 83
S=6'3750'
 Highly Composite -Numbers 117

log 89
= 6-4757 ...
log 2

tt43=101

7T44 = 103

77.. = 107

lo£7_ = 6-7641 ...

iog(i)
log 109
7747= 109 = 6-7681...
log 2
log 113
7TjS= 113 . 6-8201...
log 2 '
17
. ^SL = 6-9875...
' log (fl
77,0= 127 .log 127 5-9886....
log 2
38. It follows from (17) and (198) that log(7(17) ^ F(x), where

.(206)
for all values of iV and or,. In order to obtain the best possible upper
limit for log d (17), we must choose x so as to make the right-hand side a
minimum.
The function F(x) is obviously continuous unless (1 + 1/r)1 =p, where
r is a positive integer and p a prime. It is easily seen to be continuous
even then, and so continuous without exception. Also

+ jM2*)log2 + 77(f)*logf + ...}

= ^ {9- (2-) + a (I)* + * (f)* + ... - log 17}. .(207)

unless (l + l/r)* = p, in virtue of (17).
118 Bujhly Composite Numbers

Thus we see that F(x) is continuous, and F'(as) exists and is continuous
except at certain isolated points. The sign of F'(x), where it exists, is
that of
* (2*)+ *($)«+ »($)* + ...- log-V.
and SK2*)+Mf)* + *(t)* + ...,
is a monotonic function. Thus F'(x) is first negative and then positive,
changing sign once only, and so Fix) has a unique minimum. Thus F(x) is

.(209)
where, x is the function of if defined by the inequalities (208). Then, from
(198), we see that
d(N)^D(F).(210)
for all values of if; and d(N) = D(F) for all superior highly composite
values of Jf. Hence D(if) is the maximum order of d(N). In other words,
d (if) will attain.-its maximum order when N is a superior highly composite

Application to the Order of d (N).

t precise result known concerning the distribution

7t (as) = Li (x) + 0 (ice_“'JIos

M®) = « +0(.w-“"los

In order to find the maximum order of d(N) we have merely to deter¬

mine the order of D(N) from the equations (208) and (209). Now, from
(208), we have
log if = a (2*) + 0 = * (2*) + 0 (2»/»);
and so & (2*) = log if + o (log if )S; .(212)
and similarly from (209) we have

w rm _ !£g^W + 0 / w Ar is.(213)

.(214)
 Highly Composite Numbers 119
It does not seem to be possible to obtain an upper limit for cl (iV) notably
more precise than 1214) without assuming results concerning the distribution
of primes which depend on hitherto unproved properties of the Riemann
((-function.
40. We shall now assume that the “Riemann hypothesis” concerning
the e-function is true, i.e., that all the complex roots of f(s) have their real
part equal to -J. Then it is known that

.(215)

where p is a complex root of J (s), and that

ir(x) = Liu-)-h Li (<Jx) -2 Li(%<•) + 0 (**)
_r V'a 2\/* 1 ^__1 Va }
'l logs,- (loga)2 log® p (log®)2 p- i(loga,')aJ ’
.(216)
is absolutely convergent when k > 1. Also it is known that

2 £-(){,/« (log a-)2}; .(217)

&(«)-»= 0{^*( log a)2}.*.(218)

(215) and (216) it is clear that

.(,) = If W + ^5^R (*) + 0 {(l0^)s} , .(219)

where .(220)

But it folk from Taylor’s theorem and (218) that

Li St (x) - Li (x) = & + 0 (log*)2, .(221)

and from (219) and (221) it follows that
.(222)

41. It follows from the functional equation satisfied by f (s), viz.,

(2*-)- T (s) ?(s) cos fa = 4?(1 - S).(223)

that ti-.)^r(^)s(i+5^

is an integral function of s whose apparent order is less than 1, and hence

r(i)f(«n{i-^}.
 I 20 Highly Composite Numbers

[where p runs through the complex roots of J (s) whose imaginary parts are
positive]. From this we can easily deduce that

* (1 + *) r (■■■ 2 S) r i(1 +0 ■= n (! + *).(224)

[where p now runs through all the roots]. Subtracting 1 from both sides,
dividing the result by s, and then making s 0, we obtain

21=1+ *(7-log4*), .(225)

where y is the Eulerian constant. Hence w e that

! p' i ^-1 p" rv ~ p (i - p) f)

— 2 \?x%^ = \/x(2 +7 — log47r). ■.(226)
It follows from (220) and (226) that
(log 4tt - 7) ,Jx ^ R (x) (log a)s $ (4 + 7 - log 4tt) ^ -(227)
It can easily be verified that
log47r — 7= 1'954,1
.(228)
4 + 7 — log 4vr = 2'046,) .
appro.
2^ + 18 (a)
. Now R(x).
(log x)"- ’
S(®) = 2
so that, considering R(x) as a function of a continuous variable, we have
re M = 1 _ 4V®+ aSjx) S'(sc)
' ^ V*(log®)! sc (log xy +(log»)J
= ^>Uo{. 1 1
0og«)* + UV« (logoi’
for all values of x for which S (x) possesses a differential coefficient,
NowTthe derived series of S(x), viz.,
S(a) = l2-,
is uniformly convergent throughout any interval of positive -values of x
which does not include any value of x of the form x=pm; and S(x) is
continuous for all values of x. It follows that
S(x1)-S(x2)=pS(x)dxy
for all positive values of x: and ®2, and that S (x) possesses a derivative
S'(x) = S(x),
 Highly Composite Numbers 121

is nut of the form Also

sv-om*
l \'x )
Hence
R (x + h) = R (x) + J 0 dt = R(.*) + 0 (~^.(229)

43. Now log X = & (2*)+ *(»)’+0(4)*

= St(2') + (l) + 0{X( ) }+0(4)*
= ^ (2-)+(§)» +0(2**/“).
Similarly logD (X) = log 2. w (2 ) +1 .<r ( )Li (')- + 0 (2»'»).
Writing X for 2*, we have
log^ = a(X) + X^®/>»^ + 0(X»), I ’ _
log D (X) = log 2. it (X) + log (|) Li + 0 (Z *). J
It follows that logX- X + 0[X“*®A<**];
and So X-logX+ 0 [(log .(231)
Again, from (230) and (231), it follows that
log X = S (X) + (log iVyo-Wi^- + o {(log X)*}; .(232)
and log D (X) = log 2. u (X) + log (f) Li (log X)10*®/10*2 + 0 {(log X)*}

= log 2 {iia (X) -R(X) + 0 [(Iog^i]}

+ log(|) £i{(logX)>°s®/“«8} + 0{(log X)*}.(233)

in virtue of (222). From (231) and (233) it evidently follows that
log D (X) = log 2 . it» (X) - log 2 . R (X) + log (I) Li {(log X^a'A*8}
{ V(logX) }
+ y{(loglogX)2j
= log 2 . Li {log X- (log X)l06<3)/i°s2 + o (log X)*}
- log 2. R {log X + 0 (log X )i<*a>A<* 2}
...(234)
ie of (231) and (232). But
Li {log X — (log X)’°s <S)/iog2 + 0 (log X) A}
f(logX)A) f(logX),810«8>A<*»i-i
^((logXJ-M^I1"2.
loglogX {log logX[ '1' u { (loglogX)2

- U (l0g N) ~ +0
R{l°g X + 0 (log X)10* a)/ioit*} = p (i0g X) + 0 {(log X)l‘°* ®/i°s2i-i}
= R (log X) + 0 (log X)*,
 ■22 Highly Composite Numbers

n virtue of (229). Hence (234) may be replaced by

log D (X) = log 2. Li (log iV) + log (|) Li {(log N Wss*)
(1 ^ )tof (Si/u*. wwn.

<£(rV) = ^f Xi{(logiV)>»e(3)

44. We can now find the order of the number of superior highly c
posite numbers not exceeding a given number N. Let S' be the smal
superior highly composite number greater than S, and let
jy' = ea(2') + »(*)'+
Then, from § 37, we know that
' .(237
so that iV' = 0(Nlog JV); and also that the number of superior highly e
posite numbers not exceeding N' is
n-*(2-)+»-(})■ + »■ (*)•+....
By arguments similar to those of the previous section we can shew that
finer
« = Li (log N) + Li (log JlT)x«<i»M* - : - R (l0S N)

45. Proceeding as in §28, we can shew that, if ST is a superior high

composite number, and m and n are any two positive integers such that [n
a divisor of X, and]
log mn = o (log log S),

.(239)
 Highly Composite Numbers 123

Fp>m this we easily shew that the next highly composite number is
of the form
.(240)

Again, let S' and S ny two consecutive superior highly composite

numbers, and let

Then it follows from § 35 that

a(X)<(*)'!xd(S), .(241)

for all values of if except S and S’. Now, if S be the nth superior highly
composite number, so that

where ,r,( is the same as in §36, we see that

cl(N)<(~jIX,‘d(S), .(241)

for all values of if except S and .S'. If if is S or .S", then the inequality
becomes an equality.
It follows from §30 that d(S) c; 2d (S'). Hence, if if*be highly com¬
posite and S' < if < S, so that d(S') < d(if) < d (8), then
id (8) < d (if) < d (8), d(S')< d (if) < 2d (S').
From this it is easy to see that the order (236) is actually attained by
d(if), whenever if is a highly composite number. Bat it may also be
attained when if is not a highly composite number. For example, if
Ar = (2.3.5...p1)x(2.3.S...p,),
where p, is the largest prime not greater than 2*, and p, the largest prime
not greater than (jj)*, it is easily seen that d(N) attains the order (236):
and if is not highly composite.

VI.
Special Forms of if.
46. In §§33—38 we have indirectly solved the following problem: to
find the relations which must hold between x„ &'2, a3, ... in order that

may be a maximum, when it is given that

»(*.) + *(«*) + »(*.) + •■•
is a fixed number. The relations which we obtained are

log SC, log X„ log T.J

 124 Highly Composite Numbers

This suggests the following more general problem. If X is an integer of

theform *.«***«*»♦**»-, .(242)

where c,, &>, c3, ... are any given positive integers, it is required to find the
nature of N, that is to say the relations which hold between x.2t x.a, ...,
when d (X) is of maximum order. From (242) we see that

d(W) = ( l + c,)'1*-' (- .(242')

'(
If we define the “superior” numbers of the class (242) by the inequalities
d(X) d(X')
X’~ * {NJ ’
for all values of X' less than N, and
dJN) d(X')
' X- > (NJ ’
for all values of Ar' greater than N, X and A" in the two inequalities being
of the form (242), and proceed as in § 33, we can shew that

d(X)<X „(i +clYt( ., ...(243)

for all values of x, and for all values of Ar of the form (242). From this we
can shew, by arguments similar to those of §38, that X must be of the

.(244)
. and d (X) of the form

^ ...
(1+«,)''«'■
...(244')
From (244) and (244') w order of d (X), a sin §43.

47. We shall now consider the order of d(X) for special forms
of X. The simplest case is that in which X is of the forr

logjV=*(p),
d(X) = 2'<».

.(245)
e d (X) is exactly a power of 2, and this naturally suggests the
 Highly Composite Numbers 125

question: what is the maximum order of d(N) when d(X) is exactly a

power of 2 ?
It is evident that, if d(X) is a power of 2, the indices of the prime
divisors of X cannot be any other numbers except 1, 3, 7, 15, 31, and
so in order that d (X) should be of maximum order, X must be of the form

and d(X) of the form 2’r,!t*l+',Ial+’'te>1+-.

It follows from §46 that, in order that d(X) should be of maximum order,
X must be of the form
(246)
and d(X) of the form 2*^+'<'»+*'^+'fcA+---. (247)
Hence the maximum order of d (X) can easily be shewn to be

.(248)
It is easily seen from (246) that the least number having 2n divisors is
2.3.4.5.7.9 . II. 13.16.17.19.23.25.29 ... to n factors, .. .(249)
where 2, 3, 4, 5, 7, ... are the natural primes, their squares, fourth powers,
and so on, arranged according to order of magnitude.

48. We have seen that the last indices of the prime divisors of X must
be 1, if d (X) is of maximum order. Now we shall consider the maximum
order of d (X) when the indices of the prime divisors of X are never less
than an integer ti. In the first place, in order that d(X) should be of
maximum order, X must be of the form

'and d (X) of the form

e-sHifr
It follows from § 46 that X must be of the form
8.»K1+»^+*{(*±g«}+»{gg)«}+„-.
..(250)
and d (X) of the form

..(251)

Then, by arguments similar to those of § 43, we can shew that the maximum
order of d (X) is
(ft + |) £i l(l/») log *')+♦ (X .(252)
 Highly Composite Numbers

* «> - 1
- 4a is
\ 10B(»+2) j
N)’mi' 1

If n ^ 3, it is easy to verify that

log(n + i)
and so (252) reduces to
(n + d^W.U^vi-^I'W^.vho®} _ .(253)

provided that n ^ 3.

49. Let us next consider the maximum order of d (N) -when iV is a

perfect nth power. In order that d(JSr) should be of maximum order, N
must be of the form
enS. (aii) to) +>i> (a*) + ...
and d (Ar) of the form
- (>+-)■'-(T^rd-rs)"”--
It follows from § 46 that iV must he of the form
»% «1+’l>'r!+»* { (tS) V1* ((!+’*) >-.(254)
and d (N) of the form
(^.)-«S)'((l±|r)-((Sn.(!M|
Hence we can shew that the maximum order of d (iV) is
(» + .. (256)
provided that ?! > 1.

60. Let l(N) denote the least common, multiple of the first N natural
numbers. Then it can easily be shewn that
l(jV) = attw-v/ioss]. SPos-v/iogs]. 5[iogX7iog5j pi .(257) '
where p is the largest prime not greater than N. From this we can shew

l (N) = eMy> + M''*>+M-Vb+M.vb + ...; .(258)

and so d[l{N)}-*W(iy'0'x>{iyurt).(259)
From (258) and (259) we can shew that, if iV is of the form l (M), then
d(N) = 2L‘V°zX)+*uV.(260)
 Highly Composite Numbers 127

,.v. U«(S)V(|-«^)A4%(§) V(1<#-V) _sll(,xN)+0l V(M^LI

log2 i^ioK3'+'lo32 "(log log IV)4 J‘(k*A, + e' ((log log IV ):1J'
It follows from (258) that
Z(iV)=e-v+0W-V( l°e-v)S|; .(261)
and from (253) that d (l(iV)| = 2^C-V)+oc-*vio«.v».(262)
51. Finally, i re shall consider the number of divisors of 1Y!. It is easily
;en that
iV! = 4i.3R*.i ...(263)
where p is the largest prime not greater than iY, and

It is evident that the primes greater than £17 and not exceeding 17 appear
once in iY!, the primes greater than £17 and not, exceeding £17 appear
twice, and so on up to those greater than iY/[V17] and not exceeding
Y/([\/iY] — 1), appearing [YiY] — 1 times*. The indices of the smaller
primes cannot be specified so simply. Hence it is clear that
(264)
where w is the largest prime not greater than pN, and

From (264) we see that

<* (IV I)-*!*) (£)'<»»’> (!)'«*> ... to [ViV]-! factors

= 2»W(j)*IW(j)*OT.., to [ylV]-l factors

x goooviog-v).(265)

Since Li IN) = +0 jr.J .

' ' log 17 (.(log N f)

we see that d(iV!) = G^+0Wvr=}.(266)

where C=( 1 + l)1 (1 + J)i(l 4- J)‘ (1 + £)* ....
From this we can easily deduce that, if N is of the form Ml, then
lop -V 2 log .VIog log log lY , (• log A’
d (JV) = £f(log log IV),+ (ioinrlog-V)- ^tuogiog.vpJ s .(267)
where G is the same constant as in (266).
* Strictly speaking, this is true only when 4.
128 Highly Composite Numbers

52. It is interesting in this connection to shew how, by considering

numbers of certain special forms, we can obtain lower limits for the maxi¬
mum orders of the iterated functions eld (n) and ddd (n). By supposing that .
N = 2a_1.33-1... pv~\

:an shew that dd{n) > .... .(268)

in infinity of values of n. By supposing that
N = 22“!~1. S3"'-1... p1‘‘p
ro „ . ■,
A l>gxj ’
;an shew that ddd(n) > (logn)lc«10gi°si° .(269)
.n infinity of values of n.
 16
ON CERTAIN INFINITE SERIES

t cosh {(1 - i) -J{at)} cosh {(1 + i) »J(fit)}'

Lerch f, we obtain
F («)=/(«) .(6)
130 On certain Infinite Series
On certain Infinite Series
132 On certain Infinite Series

As particular cases of (16) we have

1 + cos <>/«) + 2 i^s (itr/a) + 3 cos (9tt/(z) +

= /;^^*+«V(ia)(^+^1+^ + ...), ...(17)

if a is a positive even integer:

1 cos (w'a) 2cos(4ir/n) 3 cos (9 v/a)
8t+ e2' —1 + e‘*~l 1

if a is a positive odd integer; and *

1 cos(2^/a) , 2 cos (8rr/o) 3cos(187r/a)
8tt + e2' —1 + e" — 1 e” — 1

if a is a positive odd integer.

It may be interesting to note that different functions dealt with in this
paper have the same asymptotic expansion for small values of n. For example,
the two different functions
1 cos n 2 cos 4n 3 cos 9n
^ + ^zr1+'^TT + >i3i +•••

and j ^°S ^ dec

have the same asymptotic expansion, viz.

.(20)
24 - 1008 + 6336 ~ 17280'
* This series (in spite of the appearance of the first few terms) diverges for all values
 17
SOME FORMULAE IN THE ANALYTIC THEORY
OF NUMBERS*
(Messenger of Mathematic*, xiv, 1916, 81—84)
I have found the following formuliB incidentally in the course of other
investigations. None of them seem to be of particular importance, nor does
their proof involve the use of any new ideas, but some of them are so curious
that they seem to be worth printing. I denote by d (x) the number of
divisors of .r, if x is an integer, and zero otherwise, and by £(s) the Riemann
Zeta-function.

(A) gjg - I-** CD + 2-* (2) + 8-* <*> + •• •> .(!)

- s-‘* <3> + *** (5) - • - .<»)

(B) d5 (1) + d2 (2)+ d5 (3)+ ... + (£(«) »

= An (log lif + Bn (logn)2 + On log n + Dn + 0 (n* + E)f,.. .(3)
, , 1 ,, 127-3 36
here A = —, B = -f (2),
is Euler’s constant, 0, D more complicated constants, and e any positive

(o *(?)+* ©+* ©+- - {* ©♦<©+<©+

.(4)
2a-^(«) = !rWls>W.(5)
where tji(s) is absolutely convergent for R(s)> and in particular

?»^0“?{p'1o«(i^)}“^(*)^W.
P) d{l) d (2) ^ d (3) ^ d (n)

((log™)* (log n)* "+ (logif=i + 0 <(logoi’

^=in{v(p-p) log(^)}
... A, are more complicated constants.
* [See Appendix, p. 339.]
+ If we assume the Riemann hypothesis, the error term here is of the form 0 (a*+*).
; Mr Hardy has pointed out to me that this formula has been given already by'
Liouville, Journal de Matkimatiqv.es, Ser. 2, Vol. ii (1857), p. 393.
134 Some Formulce in the Analytic Theory of Numbers

More generally
<J*a) + d*(2) + d'(3) + ...+d*(«)
-n{At(lognf~' +4,(log«r* + ... + ^) + 0 («»+•)* ...(8)
if 2* is an integer, and
d«(l) + <2"(2) + d“(3) + ... + d*(n)

= » f1' (log nf-' + A, (log »rs + ••• + + 0 [(log n^]}'' • -(9)

if 2* is not an integer, the it’s being constants.
(E) d(l)d(2)d(3)...d(«0 = 2’l|locloB,‘+';i+*w, .(10)
where (7 = 7 + 2 jlog2 f 1 + ~j — -j (2-1' + 3^ + 5~' + ...).
Here 2, 3, o, ... are the primes and

^} = ^n+(4y(7 + 7^1) + (^y(7 + 7l + 72~1) + '''

+ 0^ (7 + 7. + 7. + - + 7«-1) + 0 jai,g‘,}4 .

where £(l + i) = j + 7 — 7i« + 72s2 - 73s3 + ...

or r! yr = Lire |(log l)r + i (log 2)r + ... + 1 (log vf - (log k)^1 j.

(F) d (uv) = (n) d g) d Q = 2/t (S) d (|) d (|),.(11)

where & is a common factor of u and v, and
1 _S /*• (»)
1 «s '
(G) If D,{n) = d(v) + d(2v) + ... + d (nv),
we have Dfn) = 2^(8) <Z (0 A Q), .(12)
a divisor of v, and
A (n) = a (0) n (log n + 27 -1) + /S (v) n + A„ (n),.(13)
?»(») rus) g/3(») pwra+«)
7 »* ?(l + «)’ 1 v- ?*(l + s) ’
A„ (n) = 0 (n* log m)f.
(H) tZ (0 + c) + d (2« + e) + d (3t) + c) +... + d {nv + c)
= «„ (0) n (log n + 27 - 1) + & (®) nA„,„ (n),.. .(14)
;«,(») = f(8)<r_(|c|)
7 if ?(1 + s) ’
v=_?(*)»-.(i«i) irw . r_(i+*).
7 V*. ?(l + «) (?(*) r(l + s) cr^dcDf’
4 Assuming the Riemami hypothesis.
+ It seems not unlikely that A„ (a) is of the form 0 («^+e), Mr Hardy has recently
shewn that Aj(?0 is not of the form o {(n log w)* log log n}. The same is true in this case
 Some Formula' in the Analytic Theory of Numbers 135
er„(n) being the sum of the 5th powers of the divisors of n and <r/(«) the
derivative of trs(n) with respect to s, and
Ac,c (n) = 0 (n* logfl)*.
(I) The formulas (1) and (2) are special cases of
fO) 0
?(2 s-a-b)
= 1-* <r„ (1) <rt (1) + 2-* <7„ (2) ah (2) + 3- <r„ (3) <r6 (3) + ...;.. .(15)
V O) -g (s - a) v(s-b)v (s-a-b)

= 1- «r. (1) n (1) - 3- <r. (3) (3) + 5- (5) <r, (5) -.(16)
It is possible to find an approximate formula for the general sum
<7„ (1) ab (1) + <r„ (2) <?„ (2) + ... + <r„ (n) ab (n).(17)
The general formula is complicated. The most interesting cases are a = 0,
ft = 0, when the formula is (3); a = 0, 6 = 1, when it is
(18)
72f(S)1"8 'T“
c-y-t+m
and the order of E («) is the Si is that of Aj (n); and a =

J»*f(3) + m»..(19)
where E(n) = 0 (n2 (log n)5}, E(n) + o (n*lcign).
(J) If s > 0, then
<r, (1) a. (2) <r, (3) <r. (4)... <r. (») = Sc" {n !)«.(20)
where 1 > 8 > (1 - 2-) (1 - 3—) (1 - 5-")... (1 - «•-»),
m is the greatest prime not exceeding n, and

(K) If (i + ? + + f + 2” + .. .)2 = i + Sr («) q",

so that f(.),(.)-fr(»)»-,

fchen (i+~^K(L)= 1"r’(1) + 2"s,'a (2) + 3_S,J!(3) +.(21)

r= (1) + ra (2) + >■= (3) +... + r= (n) = ~ (log n + C) + 0 (ni+e),... (22)

where C = 4y — 1 + J log 2 — log w + 4 log T (J) — ~ (2).

These formula are analogous to (1) and (3).
, * It is very likely that the order of Au, c (n) is the same as that of A! (»).
 18
ON CERTAIN ARITHMETICAL FUNCTIONS

(Transitions of the Cambridge Philosophical Societ,/, xxn, No. 9, 1918, 109—184)

1. Let as(n) denote the sum of the sth powers of the divisors of n
(including 1 and n), and let
<r»(0) = i?(-s).
where f (s) is the Riemann Zeta-function. Further let
5,,„ 00 = OV (0) a, 00 + <rr (1) <r. (n -1) + ... + <r, (n) <rs (0 ). .. .(1)
In this paper I prove that
V m r(r + l)r(» + l)C(r + l)r(« + l) ,
' T(r + s + 2) f(r + s+"2) r+*+1v"

- + + na-r+s-i (n) + O {nSCH^+i)}, (2)

whenever r and s are positive odd integers. I also prove that there is no
error term on the right-hand side of (2) in the following nine cases: r = 1,
s = l; r = 1, s = 3; r = 1, s = 5; r = 1, s = 7; r = 1, s = 11; r = 3, s = 3; r = 3,
s = 5; r = 3, s = 9; r = 5, 5 = 7. That is to say 2,v,(n) has a finite expression
in terms of ov+s+1 (n) and 0-^^ (n) in these nine cases; but for other values
of r and s it involves other arithmetical functions as well.
It appears probable, from the empirical results I obtain in §§ IS—23,
that the error term on the right-hand side of (2) is of the form
0{wKr+*+i+'*>}, .(3)
where e is any positive number, and not of the form
0 {jliCr+s+D}.(4)
But all I can prove rigorously is (i) that the error is of the form
0 {nSO’+s+D}
in all cases, (ii) that it is of the form
0 {«iO-+*+i)} .(5)
if r + s is of the form 6m, (iii) that it is of the form
0{?i3fr+s+Hj .(6)
if r + s is of the form 6m + 4, and (iv) that it is not of the form
(U
 On certain Arithmetical Functions 13 7

It follows from (2) that, if r and s are positive odd integers, then
* r(')• +1) r(,? +1) f(>• +1) f(* +1)
*'■'(>° ~ --r(r + ,+2)- ■ Wr (n).(8)
It seems very likely that (8) is true for all positive values of r and s, but
this I am at present unable to prove.

2. If 2r,» (w)/«’r+«+i (») tends to a limit, then the limit must be

r(r + l)r(s + l) f(r+ !)£(.;+ 1)

■» oV+*+i(«) »-►» <rr+t+1 (I) + <r,+.+i(2)+ ... + <rr+rH («)

~ tt” <r,.+s+1 (0) + cr,.,SH (1)a + <rwi (2) *=+...

, = i?(-r) + g + g.. ...(9)

1 - aa'
Now it is known that, if r > 0, then
S' r(r + l)f(r + I) .(10)
(l-*)r+1
as x —»1*. Hence we obtain the result stated.
3. It is easy to see that
£r,(l) + <r,.(2) + aI.(3)+...+ov00

where ut = lr + 2r + 3'+...
From this it is easy to deduce that
♦H'
<rr(l) + ov(2 )+... + «rr(n). (n)t

r(r + l)r(a+l)
Ml) («-If + ov(2) (n - 2)» +... 4- *r{n-1) F
T(r + s + 2)
provided r > 0, s > 0. Now
<r,(n)>n‘,
and a„ (n) < ns (1~> + 2~‘ + 3~" +...) = n*?(s).
From these inequalities and (1) it follows that

.(12)
* Knopp, Dissertation (Berlin, 1907), p. 34.
+ (10) follows from this as an immediate corollary.
138 On curtain. Arithmetical Functions

*>_+ 1) J>+i) r(r+ n r(r + l)r(« + i)

r (r + s + 2) + F (r + s + 2)
On the other hand n-1'-*-1 o-,.+s+I («) oscillates between 1 and f(?- + s + l),
assuming values as near as we please to either of these limits. The formula (8)
shews that the actual limits of indetermination of S,,s(n) are
r(r+l)r(s + l)f(r + l)?(.v + l)
r (r + s + 2) f (r + s + 2)
r(r + 1) r (s + 1) g(r + !)?(«+!) f(r + « + 1) n4A
r(r + s+2) f(r + s + 2) ' ” '^
Naturally
f (r +1) < t(r+»((' + }) < t(r + l)((s + l)Ur + , + l)
^ +1;< f(r + s + 2) < ?(>• + * +2)

= (i cot \ey+c0+o, cos e + o2 cos 20+cos ze+...,

.where (7„ is independent of 6. Then we have

21(1 - (I - (X -
 On certain. Arithmetical Functions 139

+ f{^(1_costf,+i'S(1~cos2<’)+ T=i (1 - ■
.(IT)
Similarly, using the equation
cot516 (1 - cos n6) = (2)i - 1) + 4 (n - 1) cos 6 + 4 (n - 2) cos 20 + ...
+ 4 cos (n -1)0 + cos n8,
140 On certain Arithmetical Functions

where 1, 2, 4, 5, ... are the natural numbers without the multiples of 3; and

where 1, 2, 3, 5, ... are the natural numbers without the multiples of 4.

5. It follows from (18) that

where 8,. is the same as in (9). Equating the coefficients of 8'1 in both, sides
in (21), we obtain
(H-M*L±5) « - nets 8 + »c,S, S,+«C,,8-S,
12 (u + 1)(n + 2) ''

“C,=

. is an even integer greater than 2.

Let us now suppose that
<S>r,,(x) = ‘Y Tm’V
io that <J>r,s(a;) = <I\r(x),
2s a? 3V ,

V ' (T - T (1 - x‘f ^ (1 - off T -

Further let

Q= 240S,-l + 240(j^ + i^ + ~>+...), 1

R = - 5048, = 1-504 + • • •) j
* If x=(/2, then in the notation of elliptic functions

,5=^4 = (^)4(1_F+i4)i

^=2’5- = (t)V + *2> (1 -4F).

 On certain Arithmetical Functions 141
Then putting u = 4, 6, 8,... in (22) we obtain the results contained in the

5. 1-2S44
6. 691+ m20%,,, ■» = 441 ^ + 250R1.
7. 1-240,., aUc) = Q-R.
S. 3617 4- 16320*., i?, • j = 1617 ^ + 2000QJt-.
9. 43867-28728*., r ix) = 38367$W + 5500/P.
10. 174611 + 13200*,, lft121250^.
11. 77683 - 552*,, U'j = 57183£4£+ 20500§#:.
12. 236364091 +131040* ^. = 49679091 (/' + 176400000^2+10285000/^.
13. 657931-24*,. a*.0*-; = 392031 <fli + 265000^.#.
14. 3392780147+6960*,, ^(x) = 489693897'/ + 2507636250^^2 + 395450000#/tt
15. 1723168255201- 171864*,, (.v) = 815806500201^ + 881340705000^
+ 26021050000/i;i.
16. 7709321041217 +32640*,,,;tl (.f) = 764412173217^ + 5323905468000^'/^
+ 1621003400000^.
In general if(-s) + 4>0jS(,r:) = SAr'Mi„QmI2'‘, .(26)
where ATm>„ is a constant and to and n are positive integers (including zero)
satisfying the equation
4w + 6n = s + l.
This is easily proved by induction, using (22).

6. Again from (17) we have

/ JL +1 Si _ & Sj + & & _ j

=i+- S <J>i'2(a;)+fi *>.* (*> - ?j ■ (*)+•• •

+K£*-£*4.*-4 ^
Equating the coefficients of 0" in both sides in (27) we obtain

2 (uTf) S',+1 ~ *.« <•) = "W- + Wh + nCAS,i-s + • • • + -CLA-A,

m .
if n is a positive even integer. From this we deduce the results contained in
Table II.
142 On certain Arithmetical Functions

Table II.
1. 288*i 2 (.r) = <2-P2.
2. 720*!, 4 (,i:)=PQ-R.
3. 1008
4. 720*,, s(.r) = Q(PQ-It,.
5. 1584*lll[J(.i-,; = 3<22 + 2/(2-5P^.
0. 85020*,,(.() = F(441«S+200R!) - 691 (fit.
7. ill*, u(*)=<2 W + 4/2»-7P«A).
In general <S>ht(x) = '5,Ki,m,nPlQmR’‘, .(29)
where 2 and 21 + 4m + 6» = s + 2. This is easily proved by induction,
using (28).

■£~ 24 *-«—iP--
040*. *

~ = — 504$,, # (»■) = -—^ -

s and that r + s is even. Then

.(31)
and («) is a polynomial in Q and R. Also «
dP dQ dR
dx ’ dx' dm
are polynomials in P, Q and R. Hence (®) is a polynomial in P, Q and R.
Thus we deduce the results contained in Table III.

Table III.

2. 1728*0 5 (x)=Pi6-2PA+§2.
3. 1728*2,r(A')=2Pg2-P2a-<2A
4. 8640*2, s (*) = SR1®2 - 18P§.S+ 5 +iRK
5. 1728*2, „ (*) = 6P<? - 5P2QR+4 PR3 - 6 t^R.
6. 6912*3,4 (is) = 6 P2^ - 8PP -t- 3Q3 - P4.
7. 3456*3,li(.r)=i»§-3P2A+3Pl32-§A.
8. 5184*3,8 (*) = 6P2<22 - 2P3£ - ePQR +1?3+RK
9. 20736*4,5 (a;)=15P§2 - 20R’-R + 10P2f) -4QR-I*.
10. 41472*a r (*) = 7 (P>Q - 4P*R+eP*Q* - iPQR) + 3<?» + 4It?.
 On certain Arithmetical Functions 143

, (x) = 1Ki, „ P‘ QmR",

it exceed the smaller of r and s anc
21 + + 6n =
The results contained in these three tables are of course really results in
the theory of elliptic functions. For example Q and R are substantially the
invariants g. and g„ and the formulae of Table I are equivalent to the
formulas which express the coefficients in the series

in terms of g~ and g

8. In what follows we shall require to know the form of more

precisely than is shewn by the formula (29).
We have f> + (a-) = 2ICm nQ*R*.(33)
where .s is an odd integer greater than 1 and 4m + 6» = s + 1. Also

.c A (Q*R*) = (f +PQ™R" - (| «"-> R',+1 +1 • (34)

Differentiating (33) and using (34) we obtain

4>m+1 (x) = * (s + 1) P (K(- «) + (*)} + 2Km,nQmR'\ ■ . -(35)
where s is an odd integer greater than 1 and 4m + 6a = s + 3. But when
s = 1 we have
.(36)
9. Suppose now that
FrA*) - «?(-»■) +■**,(«)}{*?(-«) +
_ f (1 -r) + f (1 -_£) _ F (?•+ 1) r (a + X) £(i-+ !)£(» + !)
r+s " " f(»- + « + S) f(r + s + 2)
x{if(-r-»- II + ^hwW) (37)
Then it follows from (33), (35) and (36) that, if r and s are positive odd
integers,
Fr,Az) = 2Km,nQmR<‘.(38)
where 4m + 6n = r + s + 2.
But it is easy to see, from the functional equation satisfied by f (s), viz.
(2w)-« r («)?(«) cos Jtts = J?(l - «).(39)
that Fr.iW = 0.(40)
Hence <23 - Ba is a factor of the right-hand side in (38), that is to say
Pr.AA = (<?-m SKm,nQmR’‘, .(41)
where 4m + 6a = r + s - 10.
 144 On certain Arithmetical Functions

10. It is easy to deduce from (30) that

®|Uog(<?-iP) = P.(42)
But is is obvious that

P = * g log [a {(1 - «) (1 - ®!) (1 - P) • • •}“] ; .(43)

' and the coefficient of x in Qs - JR1 is 1728. Hence
O' - iP = 1728a- {(1 - a) (1 - a2) (1 - ar>)...}“.(44)
But it is known that
{(1 -a)(l-«2)(1 -®”)(1 —a')...};l= 1— Sx+5x"-7x‘+ 9*"-.... (45)
Hence Q*-JP- 1728a (1 - 3® + 5«*-7«*+...)•.(46)
The coefficient of a""1 in 1 — 3® + 5as — ... is numerically less than \/(8v),
and the coefficient of xv in Q: — R- is therefore numerically less than that of

1728® ydiv) (1 + * + x? + + ... )}8-

But x (1 + a + + *+ ...)* - ^ +.(47)

and the coefficient of x" in the right-’hand side is positive and less than

Hence the coefficient of x" in Q“ - R- is of the form

fo 0(O = 0(,/).
That is to say Q3 - R* = 2 0 (v‘) x".(48)
Differentiating (48) and using (42) we obtain
P (Q3-£"-) = 2 0(v8)*v.(49)
Differentiating this again with respect to & we have
A (P- - Q) (Q“ -R-) + BQ (Q> - P2) = 2 0 (»■) x\
where A and B are constants. But
P--<2 = -288*,1(«) —288{^ + ^+...}, .

and the coefficient of xv in the right-hand side is a constant multiple of •

rcr, (r). Hence
(P2 — Q)(Q2 — R1) = 2 0va1 (v) xv 2 0 (v~‘) a"
= SO !y) {a, (1) + Oi (2) + ... + (*)} = 2 0 (r>°) ,
and so Q(Q1 - P2) = 2 0 (t-10)a?.(50)
Differentiating this again with respect to x and using arguments similar
to those used above, we deduce
.B(!?-.H>) = 20 (*»)«’. .(51)
 On certain Arithmetical Functions 145

Suppose now that m and n are any two positive integers including zero,
and that m + n is not zero. Then
QmEn (<? - m = Q (<? - m Qm~‘ r°-
= 20 (V»)r'{S0 (v3)»'}“-■ {2 0(vs)«”}“
= 20 (vw) a:8 2 0 (v4m-s) x' 2 0 (r8’*-1) xv
= 2O(i-4“+ra+0)is'',
if m is not zero. Similarly we can shew that
<2“-R» (Q* -R?) = R (Q3 - R-) Q™R*r'
'
if n is not zero. Therefore in any case
(Q3 - R’) QmR" = 20 (r4"-™+8) a-.(52)

11. Now let r and s be any two positive odd integers including zero. Then,
when r + s is equal to 2, 4, 6,8 or 12, there are no values of m and n satisfying

in (41); consequently in these eases ’

*V,.(*) = 0.(53)
When r + s = 10, m and n must both be zero, and this result does not
apply; but it follows from (41) and (48) that
Fr,, (ffl) = 10 (v7) x”.(54)
And when r + s ^ 14 it follows from (52) that
/ri!W=S0(»-f-)*-.(55)
Equating the coefficients of x" in both sides in (53), (54) and (55) we
obtain
s- r(r + l)r(» + l)f(r + l)f(. + l)
’\S{) ~ r (r + . + 2) f (r + 7+2T ' *i(»)
| jT(l-r) + (r(l-g)^
.(#) + *,.(»), (56)
Fr,s(n) = 0, r + s = 2, 4, 6, 8, 12;
-®r,8(n) = 0(O, r + « = 10;
■®r,.(») = 0 (»’+»-*), r + s^U.
h (») is of order nr+s+1, it follows that in all
rp- + 1)T(» + 1) ijr + 1) ?(s + l)
Sr.aW' »(»)• -(57)
r(r + i + 2) f(r + s + 2) 0
146 On certain Arithmetical Functions

The following table gives the values of 2r,s(n) when r + s = 2, 4, 6, 8, 12.

Table IV.

7<rs(tt)-10w3(tt)
2.

— lOcr? (w) - 2i n<r&(n)

" 252
llo-sin)
5040 '
llcrs(n)~SOncr7(n)

12. In this connection it may be interesting to note that

<r,(l)<r1(n) + »1(8)»,(»-l) + cr1(6)<r,(»-2)+...
+ oi(2n +1) <7S(0) = ^<7,(2n +1). ...(58)
This formula may he deduced from the identity
i’* , sw ^ 5w- ■ _„/ * .a®2 4. , \

which can be proved by means of the theory of elliptic functions or by

elementary methods.

13. More precise results concerning the order of ETit (n) can be deduced
from the theory of elliptic functions. Let
x = f.
Then we have
Q = P{q){l-(kk')*} |
R - f* (q) (k'°- - ¥) {1 + J (Mf)2} [ , .(60)
= ^(?){H-i(M')2}V{l-(2^')2}]
where <j>(q) = l + 2q + 2qi + 2q’+....
But. if f(q) = S* (1 - 9) (1 - <f) (1 - q>) ...,
 On certain'Arithmetical Functions 147

then we know that

2'7V;) = k '-k'« <j>(q) I

.(61)
2 J/(9=) = (kk'fs <j>(q) I
2 lf{q‘) = 1%-<H?)J
It follows from (41), (60) and (61) that, if )• + s is of the fori:m 4m + 2, hut
not equal to 2 or to 6, then
r .........(62)
7=,,,(- 2)’
and if r + s is of the form 4m, but not equal to 4, 8 or 12, then
(63)

where Kn depends on r and s only. Hence it is easy to see that in all cases
FT,s(<f) can be expressed as

where a, b, c, d, e, h, Ic are zero or positive integers such that

a + b + c 4- 2 (d + e) = [f- (r + s + 2)],
/» + l- = 2(r + s + 2)-3[f(r+s + 2)],
and [x] denotes as usual the greatest integer in x. But

/‘(g) = g?-3sf+5^-75J + ...

fltf) “ 22*-5gm4-7e5-iis« +

ere 1, 2, 4, 5, ... are the natural numbers without the multiples of 3,

1 1, 5, 7, 11, ... are the natural odd numbers without the multiples of 3.
Hence it is easy to see that

lot of higher order than the coefficient of qln in

F ir) ft (q*) ft (qb {<j> (q&) <t> (q«)Y {<f> (ft) $ (ft)f ft' (qft) ft (2*),
148 On certain Arithmetical Functions

or the coefficient of ?4S,i in

<j>a+d (qs) <t>H,l+h (?) F (?8) <f>e (?'“) 4>e (?12) (?3)-
But the coefficient of qv in cf>- (?-) cannot exceed that of q“ in <j>'(q), since
^(?) + ^(-?) = 2^(?!)i ...(66)
and it is evident that the coefficient of ?“ in <£ (?“) cannot exceed that of ?-
in <j>(qO- Hence it follows that
n-i»0-+«+2» ^rs(re)
is not of higher order than the coefficient of q,m in
r(q)V(9s)<l>C( ?)>
where A, B, 0 are zero or positive integers such that
A + B+C=2(r + s + 2)-2[$(r + s + 2)\,
and 0 is 0 or 1.
Now, if r + s > 14, we have
A + B + OSs 12,
and so ^1 + jB>11.
Therefore one a-t least of A and B is greater than 5. But
^(?) = 20(,’) ?’*..(67)
Hence it is easily deduced that
V (?) $£ (?3) tfif) = 20 {»*W+-S+01-1} g». (68)
It follows that
^r,s(n)=0K+*-l»(H-s-i>j}; . (69)
if r + s> 14. We have already shewn in § 11 that, if r + s = 10, then
^r,s(«) = 0(?i7). (70)
This agrees with (69). Thus we see that in all cases
-®r,s(«)=0{«S(’-+s+1)}; . (71)
and that, if r + s is of the form 6m, then
ErAn)= 0{n»<’-+s+»}, ... (72)
and if of the form Gm + 4, then
#>•,.(») = 0{n«’,+*+*>}.. (73)
14. I shall now prove that the order of Er<s (n) is not less than that of
nl(r+s>_ in order to prove this result I shall follow the method used by
Messrs Hardy and Littlewood in their paper “ Some problems of Diophantine
approximation” (II)f.
 On certain Arithmetical Functions 149

Let
_ c + dr
where
~ U + br'

Also let

Then we have <a (v, r) = V 7^ ( V, T), .(74)

where a is an eighth root of unity and
(?. r) - 2 sin «v. ji 5 (1 - j») (1 - 2c/’* cos 2ir» + qut)- ■■■(15)

From (75) we have

logS-,(v, T)=log(2sinwr) + ilog5-2 ® (7®)
It follows from (74) and (76) that
log sin w + 1 log v + ■] log q + log a - X ?

= log sin irV+ilogT + i log - mbvV - X • (77)

Equating the coefficients of tis_!_1 on the two sides of (77), we obtain

provided that s is an odd integer greater than 1. If, in particular, we put s = I

and s = 5 in (78) we obtain

(.+w{i-504(iM-,+ £t+i51t + ...)}

4-m(££*£S'*ffS>+~)}-(80)
It follows from (38), (79) and (80) that
(a + bT)r+MFTtt(cp) = Fr,, (q'").(81)
It can easily be seen from (56) and (37) that

Fr:l(v) = X,Er, S(«)®B. .(82)

150 On certain Arithmetical Functions

Hence (a + brf^ XEr,, (n)g“ = 2 Er,,(n)cf*. .(83)

It is important to observe that

g(l-*■) + ?(!-*)

rfr +1)J> + 1_) f(r+I)f(a+l) .(84)

r (r + s + 2) Z(r + s + 2)
if r + s is not equal 2, 4, 6, 8 or 12. This is easily proved by the help of
the equation (39).
15. Now let
r = u + iy, t = e-"*' (« > 0, y > 0, 0 < t < 1),
so that q = e”'”-’1' = ie"“;
and let us suppose that p„lqn is a convergent to

~~ (h + a, + a, ’" ’
so that Vn=P«-iq» ~P«qn-1 = ± 1-
Further, let suppose that
a = Pn, b — — q,i,
c = VnP,i-u d=-yaqn-lf
ad — bc = tin = 1.
Furthermore, let y = l/(?*2'ii+i),
where q'n+i = “Vn2n + ?»-i,
and a'„+1 is the complete quotient corresponding to an+1.
Then we have
| a + It I = I pn - qnu - iq„y | = — . .(85)

and jq |=

where X =1 (T) = X (^) = 11 ^ 6t)}

.;-(86)
and 1(3’) is the imaginary part of T. It follows from (83), (85) and (86) that

: 1l Er,, (n) 9-1 = 1 f *,,.(*) e'2" |

> i e-M-\ Br,.{ 2) i «-*-1 ^,.(8) I

.(81)
 On certain Arithmetical 'Functions 151
We can choose a number X„, depending only on r and s, such that
:«,,.(!); «-** >2 (2) : e-4'A + i E„ (3) | <r**» + ...}
for X ^ X0. Let us suppose X0 > 10. Let us also suppose that the continued
fraction for u satisfies the condition
4X0qn > r/,w > 2X„y„ .(88)
for an infinity of values of n. Then

| Zgr,. («) F! > i \E,,. (1) | > K (j'„+1)'+s+!. (89)

where K depends on r and s only. Also

ffnS'wi = J/y.

>7y = J {log (i;«)|> vo^lj ■

It follows that, if !f is an irrational number such that the condition (88)
is satisfied for an infinity of values of n, then

tEr,A»)q'-n j>ir(i-1)-*^**).,.(90)

for an infinity of values of t tending to unit}7.

But if we had Er<t(n) = o
then we should have

jI^r,s(«)f»j=0{(i-ri(,-+*+2)},

which contradicts (90). It follows that the error term in 2r,a(n) is not of
the form
.(91)

The arithmetical function r(n).

16. We have seen that Er<s{ri) = 0,
if r + s is equal to 2,4, 6, 8 or 12. In these cases 2,,s (n) has a finite expression
in terms of o-,+s+1(m) and o-,+s_,(n). In other cases 2r,s(n) involves other
arithmetical functions as well. The simplest of these is the function r («)
defined by
I t (») a- = a {(1 - x) (1 - a?) (1 - o?)...}».(92)

These eases arise when r + s has one of the values 10, 14, 16, 18, 20
or 24.
152 On certain Arithmetical Functions

Suppose that r + s has one of these values. Then

1728 2 £,,,(»)

is, by (11) and (82), equal to the corresponding one of the functions
1, Q, R, Qr-, QR, QJR.
In other words

tSr,s(n)** = K,sm-(n)*n{' + !’l,'+S_U l -14 '

.(93)
We thus deduce the formulae
£r,,(#) = £f,,(l)T(n), .(94)
if r + s = 10; and
°V+s-n (0) Er,s (n) = Et,.( 1) {cT+s_n (0) r (n)
+ <r,+s_u (1) r (n - 1) + ... + fftH. (» - 1) t (1)>, ■ • -(95)
if r + s is equal to 14, 16, 18, 20 or 24. It follows from (94) and (95) that,
if »• + .? = r' + s', then
Er,s (n) Er%s. (1) -- Er,. (1) Er%i (»).(96)
and in general Er.s(m) E?,.- («) = Er,t (n) Et-j (m), .(97)
when r + s has one of the values in question. The different cases in which
?• + s has the same value are therefore not fundamentally distinct.

17. The values of t (n) may be calculated as follows: differentiating (92)

logarithmically with respect to x, we obtain

I n r (n) = P | t (») &... .(98)

Equating the coefficients of xn in both sides in (98), we have

r (»)'= TITn (!) T(*■- !)■+ (2)-T(»■- 2).+ ... + or,(n - 1).r (1)}.
.(99)
If, instead of starting with (92), we start with

lT(n)*" = i(l — 3x + 5x? - 7ic“+ ...)“,

we can shew that
(n — 1) r (u) — 3 (m —10) t (»i — 1) + 5 (« — 28) t (to — 3) — 7 (» — 55) t (» - 6)
+ ... to [*{1 + V(8» - 7)}] terms = 0.(100)
where the rth term of the sequence 0, 1, 3, 6, ... is %r(r — 1), and the
rth term of the sequence 1, 10, 28, 55, ... is 1 + fr(r — 1). We thus obtain
the values of r (n) in the following table.
 On certain Arithmetical Functions 153

Table V.

16 +987136
17 -6905934
18 +2727432
19 +10661420

23 +18643272
24 +21288960
10 • -115920
11 i +534612 26 +13865712
27 - 73279080
13 : -577738 28 +24647168
14 ! +401856 29 +128406630
15 | +1217160 30 -29211840

18. Let us consider more particularly the case in which r + $ = 10. The
order of !?,.,»(«) is then the same as that of t(n). The determination of this
order is a problem interesting in itself. We have proved that EriS(n), and
therefore t(«), is of the form 0 («’) and not of the form*o(n5). There is
reason for supposing that t (n) is of the form 0(nh? +€) and not of the form
o (n3^). For it appears that

...(101)
1 - rJp)p-T+pa-
This assertion is equivalent to the assertion that, if
n=pla'p,**iha*...pr\
wherepbp,., ... p, are the prime divisors of n, then
W-YT(}i)=Mn(l + al)^sin.(1. + cu)6P! _sin(1.+

where cos 9P = Jp ^ t (p).

It would follow that, if n and n' are prime to each other, we must have
t {mi') = t (n) t (»').(103)
Let us suppose that (102) is true, and also that (as appears to be highly
probable)
{2t(p)}2«p1!, .(104)
so that 8P is real. Then it follows from (102) that
1
«-^|T(n)|«(l+a )(l+os)...(l + o,),
that is to say I r (re) | < n’2’’ d (n), .(105)
whbre d («) denotes the number of divisors of n.
154 On certain Arithmetical Functions

Now let us suppose that n =pa, so that

^ ^ _ sin (1 + a) 9P

Then we can choose a as large as we please and such that

|5£^r|>i.

Hence .(106)
for an infinity of values of n.

19. It should be observed that precisely similar questions arise with

regard to the arithmetical function yjr (n) defined by
2 + (n) =/“* (afi)f* (af‘)... f“' (**),.(107)

where /(*) = ** (1 — *)(1 — a?) (1 — a?)...,

the a’s and c’s are integers, the latter being positive,
^(alCl + o2c2+... + ctrcr)

is equal to 0 or 7, and l j + ... + ,

where l is the least common multiple of c,, cs, ... cr, is equal to 0 or to a
divisor of 24.
The arithmetical functions x(n\ P(n), x*(n\ H(n) and ®(n), studied by
Dr Glaisher in the Quarterly Journal, Vols. xxxvi—xxxvi [I, are of this
type. Thus
tx (»)«“=/•(«*),

SP («)«(*-/• («?)/*(»*),

1 (n) *» =/* (a)/2 (a2)/' (so1),

2D(»)®’‘=/12(iC2),

2 © (r») xn =/“ (a)/“ (*2).

20. The results (101) and (104) may be written as

2 = Er>, (1) n -s- * , .(108)

1 n‘ J, l-2cpp l+pr+’+l-’-t v ’
ep>^pr*’*\
2cpEr,s(l) = KAp)-
 On certain Arithmetical Functions 155

It seems probable that the result (108) is true not only for r + s = 10 but
also when r +$ is equal to 14, 16, 18, 20 or 24, and that

.<109>
for all values of n, and * I > ni 1'+s+11 .(HO)

for an infinity of values of n. If this be so, then

Sr,.(n) = 0 {?ii«+*+■+•>}. Er,.(n) * o {ni <>•+«+»}.(Ill)
And it seems very likely that these equations hold generally, whenever
r and s are positive odd integers.

21. It is of some interest to see what confirmation of these conjectures

can be found from a study of the coefficients in the expansion of

* {(1 - (1 - W*) (1 - *»/“) ...}• = 1 (n) x'\

where a is a divisor of 24. When a = 1 and a = 3 we know the actual value

of ijra (»). For we have

+ + .(112)

where 1, 5, 7, 11,... are the natural odd numbers without the multiples
of 3; and

Sir,(n)a” = - 3a>2 + 6a55 - 7ar +.(113)

The corresponding Dirichlet’s series are

1 _1_, ...(114)
r »• (i+6-)a+7-^(1-n-)(i-i8-*)...’ v ;
where 5, 7,11,13,... are the primes greater than 3, those of the form 12n + 5
having the plus sign and those of the form 12/1 + 1 the minus sign; and
_l_ ...(115)
T »• (l + 8‘-)(l-5>-)(l + 7—)(1 + 11—)...’ ' ;
where 3, 5, 7, 11,... are the odd primes, those of the form in — 1 having the
plus sign and those of the form in + 1 the minus sign.
It is easy to see that
.(H6)
for all values of n, and
I’M”)!-1, |f3(/i)i = V/i. .(117)
for an infinity of values of n.
156 On certain Arithmetical Functions

The next simplest case is that in which a = 2. In this ease it appears that

i±?i!L) = nin2, .(118)

where II,- ^ (1 _ 7_*,} (1 _ n ■<i + 17-)..'.1

5, 7, 11, ... being the primes of the forms 12m- 1 and 12n + 5, those of the
form 12» + 5 having the plus sign and the rest the minus sign; and

111 = ’
13, 37, 61, ... being the primes of the form 12m +1, those of the form
m- + (6m — 3)2 having the plus sign and those of the form m- + (6n)“ the minus

This is equivalent to the assertion that if

n = (5*<. 7"’. 11““. 17“»..,)! 13““. 37““. 61““. 73““
where ap is zero or a positive integer, then
■ft («)=(—1 (1 +a,3) (1 + 0 (1+«„,)...... .(119)
where 5, 13, 17, 29,... are the primes of the form 4m+ 1, excluding those of
the form in- + (6n)2; and that otherwise
ijr2 (n) = 0.(120)
It follows that | i/r2 (n) | ^ d (m) .(121)
for all values of n, and | ^ (m) |»1 .(122)
for an infinity of values of n. These results are easily proved to be actually

22. I have investigated also the eases in which « has one of the values
4, 6, 8 or 12. Thus for example, when a — 6, I find

l1M^ = n,n3*.(123)

where “ (1 _ IP-) (1 -7—) (1 - 11-“)... ’

3, 7, 11, ... being the primes of the form 4m — 1; and

112=(i - 2c7.6-*+"6“-“) (i -To,;. i3-s+l-F^y:;, ’

5, 13,17, ... being the primes of'the form 4m + l, and cp - it- - {2vf, where u
and v are the unique pair of positive integers for which p = v? + (2v)s. This
is equivalent to the assertion that if
n = (3“». 7“>. 11““ .. ,)s. 5“». 13““. 17““ ...,
is Dr Glaisher’s X (m).
 On certain Arithmetical Functions 157

then .(124)

where tan 18P = ^ (0< 8r<ir),

and that othenvi.se yjre(n) = 0. From these results it would follow that
; i^s (n) ] net(re) .(125)
for all values of re, and ;1K(n) ,^ 11 .(126)
for an infinity of values of re. What can actually be proved to be true is that
] i|r0(re)j < 2red(re)
for all values of re, and : ijr,. (re) | > re
for an infinity of values of re.

23. In the case in which a = 4 I find that, if

re = (5B>. 11“». 17“"... f. 7"’. 13®“. 19®“....
where 5, 11, 17,... are the primes of the form 6m- 1 and 7, 13, 19,... are
those of the form 6m +1, then
+«(«) _ , «°(1 + Sin(l+OA 7
V» sin07 ' sin8a "'U '

where tan 8P = (0 <6P< w),

and u and v are the unique pair of positive integers for which p= 3re2+(l ± 3n)!;
and that i/r4 (re) = 0 for other values.
In the case in which a = 8 I find that, if
n = (2“=. 5««. 11®“ .. ,)2. 7“’. 13““. 19““ ...,
where 2, 5, 11,... are the primes of the form 3m-1 and 7, 13, 19,... are
those of the form 6m +1, then
f»W. sin3(l+g7)g„ sin3(1 +

where 8P is the same as in (127); and that ijrs (re) = 0 for other values.
The case in which a= 12 will be considered in § 28.
In short, such evidence as I have been able to find, while not conclusive,
points to the truth of the results conjectured in § 18.
24. Analysis similar to that of the preceding sections may be applied to
some interesting arithmetical functions of a different kind. Let

<j,‘(q)=l + 2tr,(n)qn.(129)
where $ (g) = 1 + 2g + 2y1 + 2g“ + ...,
158 On certain Arithmetical Functions

so that )•« (n) is the number of representations of n as the sum of s squares.

Further let

...(130)

«, (») r=5 (irf + rrj +it|+• • •) ’ • ••(131)

when s is a multiple of 4;

(j.-uasfc.ws— (^+fiy+K?+ -)• -<m>

when s + 2 is a multiple of 4;

when s - 1 is a multiple of 4 ;
♦•(g-EHEJ--)-**
B.1S, (.)!>- * (j^ t rff* ITfr+ •■•)

-!(R-££+£?-4 ■■llS4)
■ when s +1 is a multiple of 4. In these formula
&-*. *-A. A-A. £-A.
are Bernoulli’s numbers, and
£, = 1, Es = 1, £s = 5, Z?7=61, E,—1385, ...
are Euler’s numbers. Then Z&(n) is in all cases an arithmetical function
depending on the real divisors of n; thus, for example, when s + 2 is a
multiple of 4, we have
(2s - 1) BA (n) - s {ow (n) - 2s<rM (In)}.(135)
where <rs (x) should be considered as equal to zero if x is not an integer.
Now let r„ (n) = S„, (n) + e2a (n).(136)
Then I can prove (see § 26) that
Ssi («) = 0 .(137)
if s = 1, 2, 3, 4; and that
eM(n) = 0(»s-'-i .(138)
for all positive integral values of s. But it is easy to see that, if 3, then
Hn*-1 < «* (n) < Kn‘~\ .(139)
where H and K are positive constants. It follows that
n»(n)~ 8»(») ..(140)
for all positive integral values of s.
 On certain Arithmetical Functions 159

It appears probable, from the empirical results I obtain at the end of this
paper, that ^

for all positive integral values of s; and that

*,(n) + o{nK-»} .(142)
if s>5. But all that I can actually prove is that
-<%,(«) = 0(n-- HS«1).(143)
if s >9 ; and that ew(n) =f o (re**-1) .(144)
if s*5.
25. Let /a.(2) = ^«a.(H)2,* = S{ra,(n)-Ss>(«)}g’‘. (145)

Then it can be shewn by the theory of elliptic functions that

u (s)=r (?); <a s(s_i) *n w. (146)
that is to say that

Mq) rm . (147)
where <j>(q) and f(q) are the same as in §13. We thus obtain the results
contained in the following table. ,
Table VI.
1. />(?)=o, Mg)-o, Mg)~o, Mg)-o.
2. 5/10(?) = 16^-®, /12(?) = 8/l2(?2).
3. 6I/14 (?) = 728/4 (— ?)/l0 (?3), 17/„(?) = 256/»(-2)/s (?-).
4. 13S5/18 (?)=24416/‘3 (-?)/» (?2) - 286 •

5. 3L/20 (?) = 616/» (- ?)/< (?=) -128 .

6. 50521/B (?) = 1103272/»(- ?)/2 (?«) - 821888 ■

7. 691/s, (?) = 16576/“ (- ?) - 32768/“ (?2).
It follows from the last formula of Table VI that
WM»> =(- l)’^^ (n) - 512t (in).(148)
where t (n) is the same as in § 16, and t (x) should be considered as equal to
zero if x is not an integer.
Results equivalent to 1, 2, 3, 4 of Table VI ivere given by Dr Glaisher in
the Quarterly Journal, Vo], xxxviii. The arithmetical functions called by him
3&(»). D(n), Tf(?i), ©(»), U(n)
are the coefficients of qn in

/"(s’). /‘(<2)/8(<f)> ^(-?)/6(2!).

 160 On certain Arithmetical Functions

He gave reduction formulae for these functions and observed how the functions
which I call e,„(m), en(n) and «„(«) can be defined by means of the complex
divisors of n. It "is very likely that t(«) is also capable of such a definition.
26. Now let us consider the order of e2s («). It is easy to see from (147)
that /.* (q) can be expressed in the form

2if (- 9)Y F {£$)}'^

where a, b, c, k, h are zero or positive integers, such that
a + 6+ c = [fs], A + & = 2s — 3 [fs].
Proceeding as in § 13 we can easily shew that

cannot be of higher order than the coefficient of g““ in

<t>A(q)<l>B(.f)4>c(l2). .(!50)
where 0 is 0 or 1 and A + B + 0 = 2s-2 [|s].
Now, if s > 5, A + B+C^i; and so A + B »3. Hence one at least of A
and B is greater than 1. But we know that
, ^ P(g) = X0 Mf
It follows that tire coefficient of qnn in (150) is of order not exceeding
ni(A+B+a-i+<.
Thus e!S(n) = O0is—1M+-) .(151)
for all positive integral values of s.
27. When o 9 we can obtain a slightly more precise result.
Ifs> 16 we have A+B+C^ 12; and so A + B11. Hence one at least
of A and B is greater than 5. But
*6(<Z) = SOW 9*-
It follows that the coefficient of g*“ in (150) is of order not exceeding
^U+B+Cl-^
or that eu(n)~0(n‘-'-W% . .(152)
if s ^ 16. We can easily shew that (152) is true when 9 ^ s < 16 considering
all the cases separately, using the identities ■

\F(f)Y
r-{-q) \r-(-<!)) ’

and proceeding as in the previous two sections.

 On certain ArithmeUcal Functions 161

The argument of §§14—15 may also be applied to the function e^(n).

We find that
«.<»)•+*<»»-)..(153)
I leave the proof to the reader.

28. There is reason to suppose that

.
*,(*)±0 {***-»} )
if s 3= 5. I find, for example, that

..ass)

Where ' ".= (T--'3^^)(T^^■

3, 7, 11, ... being the primes of the form 4n-l, and

112 = (l-2c,.5-' + 5*--”)(1-2o13.13-«+ 13j'“) ... ’

5, 13, 17, ... being the primes of the form 4« +1, and
cp = v?-(ivf,
where n and v are the unique pair of positive integers satisfying the equation
v?+ (4 v? =p‘.
The equation (155) is equivalent to the assertion that, if
n = (3“=. 7“’. 11““ ...)5.2“=. 5“*. 13a»
where a,, is zero or a positive integer, then
»■(”> = (_ i)«,8iii-(l±£?)-j? (156)
n!e10 (1) ( ' sin 4 0S ’ sin40ls

where tan Sp = - (0 <BV< Jw),

■u and v being integers satisfying the equation «! + ir = p; and (n) = 0

otherwise. If this is true then we should have

.<is7>
for all values of n, and | j 3tns .(158)
for an infinity of values of n. In this case we can prove that, if n is the
square of a prime of the form 4m - 1, then .
162 On certain Arithmetical Functions

Similarly I find that

| «■= W : (159)
Sl2(1)p il + 2cp.1r‘+ps--‘)’.
p being an odd prime and c ^p‘. From this it would follow that

»'<*(») (160)
e,s(l)|
for all values of n, and [ ‘ > ni (161)
ei2(l)l
for an infinity of values of n.
Finally I find that
(162)
7 n* 1 + 2*-' j, Vl + 2cp. p~8 + p'-’-’J.
p being an odd prime and cr-^p7. From this it would follow that
(163;

for all values of n, and «■« (») j (164)

for an infinity of values of n.
In the ease in which 2s = 24 we have
(«) = (- lf-^oOr (») - 512r (in).
I have already stated the reasons for supposing that
\T(n)\^n^d(n)
for all values of n, and | t (n) \
for an infinity of values of n.
 19
A SERIES FOR EULER’S CONSTANT y

1. In a paper recently published in this Journal (Yol. xuv, pp. 1—10),

Dr Glaisher proves a number of formulse of the type

where Sn = I-" + 2 “ + 8"“ + 4_,‘ + ...,

and conjectures the existence of a general formula
7 = X,-(/-+ 1) (r + 2) ... (2r)
( s, s 5 1
(3 (r + 3)(r + 4)... (2r + 2) + 5 (r + o') (r + 6) ... (2r + 4) + '"f ’
where X,. is a rational number. I propose now to prove the general formula
of which Dr Glaisher’s are particular cases: this formula is itself a particular
case of still more general formula;.

2. Let r and t be any two positive numbers. Then

£(l _ xy-r log r(I-a) dx=£*<- (1 -»)<"■ log r(*) dx

= J x‘~‘ (1 — x)r~l log T(1 + x) dx — J a;1-1 (1 — x)r~‘ log x dx.(1)

But J1 X'-1 (1 - ®)‘-> log r (1 - x) dx

=£ «•-- (1 - xr {yx + s, | + Ss | + .. .j dx
r(i+r)r«) r(2+r)r(t) s, r(8 + r)r(Q&
r(l+r + 07 r(2 + r + «) 2 + r(3-fr + «) 3+.^
Similarly J af-1 (1 - xj~l log T(l+x)dx
r(l + t)r(r) r(2 + t)V(r)Se r(3 + t)V(r)Sa
r(l+r + t) T(2+r + t) 2 T(3+r + t) 3'
And also

L ('X - ^r_i log 0xdx=<rJ

r(r)T(4) [F (t) rp- + f)j
r(r+o |r(«) r(r+i)|
164 A Series for Euler’s Constant y

It follows from (1)—(3) that, if r and t are positive, then

r r()- + 1)_ „ rQ + l)(r+ 2~)
l(r + <)7 + 2(r + t)>-M + l) Z(r+ t){r+t + l){r+ t+2)‘ 3
A_l _ t(t +1) „ t(t + 1) (t + 2)_ _
+ 1 (>•+'«) 7 2(r + t)(j- + < + i),s 3(r+«)(r + t + l)(f + < + 2) !

.(4)

Now, interchanging r and t in (4), and taking the sum and the difference of
the two results, we see that, if r and t are positive, then
r+t r(r+l)(r + 2) + <(« + l)(« + 2)

-I],* .(5)

2 (r+ «)(»• + <+ 11 2

r ()' + 1) (r + 2) (r + 3) — i (t + 1) (t + 2) (t + 3) „
4(r + t)(y + i + l)(r + f + 2)(r + t+3) *“
.(6)
The right-hand sides of (5) and (6) can be expressed in finite terms if r
and t are rational. If, in particular, r and t are integers, then

and | ^ — (2® = ^l+J + J + J+ ... + ^

— ^1+{ + j + J +

3. Let us now suppose that t = r in (5). Then it is clear that

(r + !)(?• + 2) (r + l)(r + 2)(r + 8)(r + 4)
7+3(2»-+l)(2r + 2) s+5 (2r +1) (2r + 2)(2r + 3) (2r + 4) ”+"‘

.<’>
if r > 0. If we suppose, in (7), that r is an integer, we obtain the formula
conjectured by Dr Glaisher, the value of Xr being
f1 l + , , , , , , 1
 A Series for Euler's Constant y 165

Again, dividing both sides in (6) by r— t and making t-*-r, we see that, if
;• > 0, then
J1±L_ (l - 1 V<5,
2 (2r + 1) \r ' r+lJ "
(r+ !)(/•+2) (r + 3) /I 1 1 1 \ ,
+ i'(2;-+i)(2r+2)(2r + 3) Vr /• + ! r+2 r + 3) J

= ~ /„ l-l dX = r= + C+lf + Fb2)a + V+V1 +.(8)

Thus for example we have

j5=a+i)^ + (i+m+i)^+(1+4+4+i+i+i)o + ""

4. If we start with the integral

f^o - *r* log r(i-1) da,

and proceed as in § 2, we can shew that, if r and t are positive, then

r _>•(»•+1) r(r + !)(?• + 2;_
1 (r •!•1)' ' +2(r+«)(r + «+l) 1 3(r + t)(?- + t+l)fr + t + 2) =
t _ Ut+1) t(t + l)(t + 2) _
1 (?• + «) 1 + 2(r + i)(r + t + l) 2 3(r + t)(i- + « + l)(r + t + 2) J

.o)

From (9) we ca easily deduce that, if r and t are positive, then

r (r + -1) + f (t +1) , r(r+lHr + 2Hr + 3) + «(< + l)(t + 2)ft+3)„,
2(r + t)(r + t + l)‘ : 4(r + <)(r + t + l)(r + « + 2)(r + t + 3) ^ +
xr~i + rf-i __ %xr+t-l
-»/: .(10)

+ 2) —t(,s + l)(t + 2)
3 (r +1) (r + i +1) (r +1 + 2) °s+'"

= .--(U)
As particular case of (10) and (11), we have
(r + 1) (r + 2)(r + 3) 1 + a”-1
1o4 + 2(WtT)^ *4(2r + l)(2j- + 2) (2r + 3)‘ dx\
1 +x
.(12)
 A Series for Euler’s Constant y

(r + l)(r+2) (l _J_ _1_\ g.

3 (2?- +1) (2r + 2)irr+l
2) r + 2| 3
(r+I)(r + 2)(r + 3)(r + 4)
+ 5 (2r + 1) (2r + 2) (2r + 3) (2r + 4)

x (; + TTl + r ?2 + + TTl)

r^Cr + l)2 (r + 2)2^.

provided that r > 9. Thus for example we have

1 = ,0g 2 2 2^3
+ ( + O + 677 + ''') ’

To = Hs + rpr (1 + i + i) + ^7. (! + 4 + i + i + i) +

5. The preceding results may he generalised as follows. Let f(s, x)

denote the function represented by the series
x^ + (x+ 1)~" + (x + 2)-* + (x + 3)-> + ... (® > 0)
and its analytical continuations, so that f(s, 1)= f(s) and f (s,|) = (2s- 1) f(s),
f (s) being the Riemann f-function. Then

/o ^ ^ 1 ~ ^ = f0 ^ - ?(*■ *) ^
= «t_I (1 - a)1'-1 ?(s, 1 + x)dx + J^ xl~‘~l (1 - x)r~l dx, .(14)

provided that r and t are positive. But we know that, if | x [ < 1, then

?(«, l-«) = ?W + -]-!?(s + l)® + S-^^-)?(a + 2);r2+...; ...(15)

and that J‘(1 -*Y~'dx = .(16)

provided that t > s. It follows from (14)—(16) that, if r and t are positive
and t > s, then
'0 + 1) r (r + 1)
{?(s>+n^?<s+1>- ?0+2)+...J
2! (r + t)(r + t+1)
sQ + 1) t(t + 1)
2! (r +1) (r + t + 1)
... r(r + t)r(t-s)
.(17)
r(t)r(r-»'+t) •
 A Series for Euler's Constant y

As particular cases of (17), we have

_1 r(r + o jrq-a) r(>--a)|

2 T(r — a + f) ( F(t) + F(r) )’
s(s + l)r(r + l)-f(t + l)
?(s + 2)
[r + t)(r + t +
s(s + l)(s + 2)(s + 3) r(r + l)(r + 2)(r+3)-t(t+l)(t+2)(t+3) ^
-?(s + 4)+...
4! (r + i) (r +1 + 1) (r + t + 2) (;• +1 + 3)
= 1 _C(r + i)_ (r(i-8) r(r-s)}
...(19)
2 r(r-i + ()t r(«) T(r) }’
provided that r and t are positive and greater than s. From (18) and (19)
we deduce that, if r is positive and greater than s, then
£ jr y. , y. , »(» + l)(« + 2) r (r + 1) (r + 2)_
1! 2r^ + 3! n *t (s + 3) + ..
2?- (2r + 1) (2r + 2)
_ 1 T (2r) r (r - s)
...(20)
~ 2 F (r) F (2r — s) ’

-
s(s + ll(s + 2)(s+3) r(r+l)(r+2)(r+3)
4! 2r(2r + l)(2-r + 2)(2r + 3)
(1 ,
vt-2+,-ts)«<+4>+-
_1 r(2r)_r(r_-s) fi (l-s8),
-tdw. .
2 F (?•) r (2r — s) J o

6. If we start with the integral

J**«(i-*y»r(«,i-D<fa,
nd proceed as in 8 5, we can shew that, if r and f a

s(s + l) r(r +1)

?.<») + £; r,?.(s+l) + -AV,-- (r ■+•1) (?■-)-1 + 1) 1
s(s + 1) t(t + 1)
(, + i)(r + i + l)^i
_ r(r + t)r(t-«)
r(i)r(r-.+'o.
where (s) is the function represented by the series
1-«_2-» + 3-s-4-®+ ...
168 A Series for Euler’s Constant y

and its analytical continuations. From (22) we deduce that, if r and * ai

positive and greater than s, then

+ IK1W+
(I + W + '(S-+1)r(r
2! + 1)+^
(r+t){r +t
1 r ()■ + *-;> jr(i-s) r(r-s)}
...(23)
— 2 r (r — s + t)t r(<) + r(r) i’

and f,^|6(' + l)
. s (s + 1) (s + 2) r(r+ 1) (r + 2) — t(t + l)(t + 2) „
+ ...
3! (7’ 4-1) {v +1 + 1) 0’ 4- t 4- 2)
1 r (»■ +1) j;r(t-s) r(r-.s))
....(24)
2 r(r-s + tj 11 r(t) f(r) i
As particular cases of (23) and (24), vve have
1 r (2r) F (r — .s)
t («) + s + r(r+1) r/6.+2) ...(25)
f,W+ 2! 2r(2r+l)?l( +2) + -"2 T (r)T (2)— s) ’

and D2r^(*+1)
s (s + 1) (s,+2) r(r +1) (r 4-2) *
3! 2r(2r + l)(2r + 2) '
1 r (2r) r (r — s)
2 F (r) T (2r - s) J
. ....(26)

provided that r is positive and greater th an s.

 20
ON THE EXPRESSION OF A NUMBER IN THE FORM
oaf + by" + car + (to5
(Proceedings of the Cambridge Philosophical Society, xix, 1917, 11—21)
T. It is well known that all positive integers can be expressed as the
sum of four squares. This naturally suggests the question: For what positive
integral values of a, h, c, d can-all positive integers be expressed in the form
acd1 + by- + cz- + du" ? .(U)
I prove in this paper that there are only 55 sets of values of «, b, c, d for
which this is true.
The more general problem of finding all sets of values of a, b, c, d, for
which all integers with a finite number of exceptions can be expressed in the
form (IT), is much more difficult and interesting. I have considered only
very special cases of this problem, with two variables instead of four; namely,
the cases in which (IT) has one of the special forms
a(x- + y" + z=) + btd, .a.(1'2)
and a + f) + b (z- + u*).(1'3)
These two cases are comparatively easy to discuss. In this paper I give
the discussion of (1-2) only, reserving that of (T8) for another paper.
2. Let us begin with the first problem. We can suppose, without loss of
generality, that
a*b*oid.(21)
If a > 1, then 1 cannot be expressed in the form (IT); and so
a-1.(2-2)
If b > 2, then 2 is an exception; and so
.(S-8)
We have therefore only to consider the two cases in which (IT) has one or
other of the forms
x- + y2 + cz- 4- du\ x- + 2y- + cz- +■ du".
In the first case, if c > 3, then 3 is an exception; and so
I<c-s8.(2-81)
In the second case, if c > 5, then 5 is an exception ; and so
2<e«5.,.(2-32)
We can now distinguish 7 possible cases.
(2‘41) a? + f + z- + du".
If d > 7, 7 is an exception; and so
1 <d«7. .(2-411)
170 On certain Quaternary Forms
(242) tf+y' + W + du*.
If d > 14, 14 is an exception ; and so
2*d<14. ....(2-421)
(243) a? + f + 3«* + du*.
If d > 6, 6 is an exception: and so
3 < d < 6. ....(2-431)

If d > 7, 7 is an exception; and so

2<d<;7. ....(2-441)
1
(2-45) x + 2i/! + Zz" + du?.
If d > 10, 10 is an exception ; and so
3 < d < 10. ....(2-451)
(2-46) + if + iz- + du3.
If d > 14, 14 is an exception; and so
4<d«14. ....(2-461)
(2-47) x- + iy- + 52a + du-.
Ifd>10, lO'is an exception; and so
5«d<10. ....(2-471)
We have thus eliminated all possible sets of values of a, l c, d, except
the following 55:
1, 1 1, 1 1, 2, 3, 5 1, 2, 4, 8

1, 1 2, 2 1, % o’, 5 l’, 1, 1 9
2, 2 1, 1, 1, 6 1, 2, 3, 9
1, 3 1, 1, 2, 6 1, i, 4, 9
2, 3 1, 2, 2, 6 1, 2, 5, 9
1, 5 2, 3 1, 1, 3, 6 1,1, 2, 10
1, 1 3, 3 1, 2, 3, 6 1, 2, 3, 10
1, 5 3, 3 1, 2, 4, 6 1, 2, 4, 10
1, 1 1, 4 1, 2, 5, 6 1, 2, 5, 10
1, ] 2, 4 1, 1, 1, 7 1, 1, 2, 11
2, 4 1, 1, 2, 7 1, 2, 4, 11
1, 1 3, 4 1, 2, 2, 7 1, 1, 2, 12
I, 3, 4 1, 2, 3, 7 1, 2, 4, 12
1, 4, 4 • 1, 2, 4, 7 1, 1, 2, IS
1, 5 1, 2, 5, 7 1, 2, 4, 13
2, 5 1, 1, 2, 8 1, 1, 2, 14
1,
l’, s’ 5
 On certain Quaternary Forms 171

Of these 55 forms, the 12 forms

1, 1, 1, 2 1, 1, 2, 4 1, 2, 4, 8
1, 1, 2, 2 1, 2, 2, 4 1, 1, 3, 3
1, 2, 2, 2 1, 2, 4, 4 1, 2, 3, 6
1, 1, 1, 4 1, 1, 2, S 1, 2, 5, 10
have been already considered by Liouville and Pepin*.
3. I shall now prove that all integers can be expressed in each of the
55 forms. In order to prove this we shall consider the seven cases (2'41)—
(2'47) of the previous section separately. We shall require the following
results concerning ternary quadratic arithmetical forms.
The necessary and sufficient condition that a number cannot be expressed

is that it should be of the form

4*(8/4+ 7), (\ = 0, 1, 2, .... /4 = -).(3dl)
Similarly the necessary and sufficient conditic number cannot be
expressed in the forms
®2+ f + ^. .(3-2)
x- + y- + 3a5,.... .(3-3)
x1 + 2f + 2e",.... .(3-4)
x- + 2y- + 3a2, .... .(.3-5)
3s + 2y2 + 4x2, .... .(3-6)
x* + 2f + oz-,.... .(3-7)
are that it should be of the forms
4* (16/4 +14), .(3-21)
9‘( !-V+ 0).(3-31)
4*( 8/4+ 7), .(3-41)
4* (16/4 + 10).(3-51)
4* (16/4+ 14).(3-61)
25* (25/4 + 10) or 25* (25/4 + 15)f.(3’71)
The result concerning af + t^ + a2 is due to Cauchy: for a proof see
Landau, Handbuch der Lekre von der Verteilung der Primsahlen, p. 550.
The other results can be proved in 'an analogous manner. The form
Xs+ y‘‘ + 2z- has 'been considered by Lebesgue, and the form a? + j/! + 3z!
by Dirichlet. For references see Baehmann, Zahlentheorie, Vol. IT, p. 149.
* There are a large number of short notes by Liouville in Vols. v—vrn of the second
series of his Journal: See also Pepin, ibid., Ser. 4, Vol. VI, pp. 1—67. The object of the
work of Liouville and Pepin is rather different from mine, viz. to determine, in a number

t Results (3-11)—(3-71) may tempt us to suppose that there are similar simple results
for the form a^+bf+cz1, whatever are the values of a, b, c. It appears, however, that
172 On certain Quaternary Forma

4. We proceed to consider the seven cases (2'41)—(2'47). In the first

case we have to shew that any number N can be expressed in the form
N = as2 + f + ? + dur, .(41)
d being any integer between 1 and 7 inclusive.
If iV is not of the form 4*(8/i + 7), we can satisfy (4-1) with u = 0. We
may therefore suppose that iV = 4* (8/t + 7).
First, suppose that d has one of the values 1, 2, 4, 5, 6. Take u = 2*.
Then the number
iY-du2 = 4*(8Ju + 7-d)
is plainly not of the form 4* (S/a + 7), and is therefore expressible in the
form a? + y- + z2.
Next, let d = 3. If y. = 0, take u = 2*. Then
iV-du2 = 4i+1.
If n ^ 1, take u = 2A+1. Then
I7-d(i2 = 4i(8/r-5).
In neither of these cases is N — du- of the form 4*(8,u. + 7), and therefore in
either case it ca^ be expressed in the form x' + y2 + a2.
in most oases there are no such simple results. For instance, the numbers which are not
ot the form .r2 + 2y2+ 10z2 are those belonging to one or other of the four classes
25*(8,x+7), 25* (25/i4-5), 25* (26,1+16), 25*(25p+20).
Here some of the number's of the first class belong also to one of the next three classes.
Again, the even numbers which are not of the form x% +y2+ 10s2 are the numbers
4* (16,1+6),
while the odd numbers that are not of that form, viz.
3, 7, 21, 31, 33, 43, 67, 79, 87, 133, 217, 219, 223, 263, 307, 391, ...
do not seem to obey any simple law.

x'-+ y2+ 4s2,

xl t /+5z2,
*“+ y2 + 6s2,
*2+ y‘+8s2,
**+2/+6s8,
tf+W + Sz*.

4*(8,i+7) or .8,1+3,
4* (8,1+3),
9*(9/i+3),
4* (16^ + 14), 16,1 + 6, or 4,1+3,
(8/i+5),
4*(8/i+7) or 8/1 + 5.
 On certain Quaternary Forms 173

Finally, let d = 7. If ft is equal to 0, 1, or 2, take u = 2\ Then N-dv?

is equal to 0, 2.4l+1, or If ft 5= 3, take u = 2*+1. Then
IV-dt i"'=4A('SM-21).
Therefore in either case IV — da- can be expressed in the form *» + y- + i*.
Thus in all cases lY is expressible in the form (4'T). Similarly ive can
dispose of the remaining cases, with the help of the results stated in § 3.
Thus in discussing (2-42) we use the theorem that every number not of the
form (3'21) can be expressed in the form (3'2). The proofs differ only in
detail, and it is not worth while to state them at length.
5. We have seen that all integers without any exception can be expressed
in the form
m{s? + f + ^) + na?, .(51)

We shall now consider the values of m and a for which all integers with
a finite number of exceptions can be expressed in the form (5T).
In the first place m must be 1 or 2. For, if m > 2, we can choose an
integer v so that
nu- ¥ r(mod m)
for all values of u. Then •,

where ft is any positive integer, is not an integer; and so m/t + v can

certainly not be expressed in the form (o'l).
We have therefore only to consider the two cases in which m is 1 or 2.
First let us consider the form
+ y°- + z- + nu-.(5-2)
I shall shew that, when n has any of the values
1, 4, 9, 17, 25, 36, 68, 100,.(5'2I)
or is of any of the forms
41-+ 2, ik + 3, 81-+ 5, 16& + 12, 32/c + 20.(5-22)
then all integers save a finite number, and in fact all integers from 4»
onwards at any rate, can be expressed in the form (5’2); but that for the
remaining values of ?i there is an infinity of integers which cannot be ex¬
pressed in the form required.
In proving the first result we need obviously only consider numbers of
the form 4A (8p + 7) greater than n, since otherwise we may take u = 0. The
numbers of this form less than n are plainly among the exceptions.
6. I shall consider the various cases which may arise in order of
simplicity.
(6-1) n = 0 (mod 8).
There are an infinity of exceptions. For suppose that
W=8/r + 7.
174 On certain Quaternary Forms

Then the number N — nu- = 7 (mod 8)

cannot be expressed in the form +1/2 + za.
(6'2) n = 2 (mod 4).
There is only a finite number of exceptions. In proving this we may
suppose that S = 4*(8/i + 7). Takeu = l. Then the number
S-mr-=^(S/M + 7)-n
is congruent to 1, 2, 5, or 6 to modulus 8, and so can be expressed in the
form x- + if + z".
Hence the only numbers which cannot be expressed in the form (5'2) in
this case are the numbers of the form 4A (8/m + 7) not exceeding n.
(6-3) ?i = 5 (mod 8).
There is only a finite number of exceptions. We may suppose again that
_Y = 4-' (S/x + 7). First, let X =(= 1. Take n = 1. Then
iV- nu- = 41 (8/m + 7) - n = 2 or 3 (mod 8).
If X = 1 we cannot take u = 1, since
F - n = 7 (mod 8);
so we take u = 2. Then
N - mi- = 4*(8/t + 7) - 4n = 8 (mod 32).
In either of these eases S — nu2 is of the form x- 4- y- + z“.
Hence the only numbers which cannot be expressed in the form (5'2) are
those of the form 4* (8/i + 7) not exceeding n, and those of the form 4 (8/i + 7)
lying between n and 4n.
(6'4) n = 3 (mod 4).
There is only a finite number of exceptions. Take
iV= 4* (8/m + 7).
If XJt 1, take u = 1. Then
S-nn?=l or 5 (mod 8).
If X = 0, take u = 2. Then
S - mi* s 3 (mod 8).
In either case the proof is completed as before.
In order to determine precisely which are the exceptional numbers,
we must consider more particularly the numbers between n and 4n for
which X = 0. For these u must be 1, and
S — nu3 = 0 (mod 4).
But the numbers which are multiples of 4 and which cannot be expressed in
the form x2 + f + z2 are the numbers
4”(8r + 7), (* = 1, 2, 3,...,r = 0, 1, 2,3,...).
The exceptions required are therefore those of the numbers
« + 4«(8i/+7) .'..(6-41)
which lie between n and 4n and are of the form
.(6'42).
 On certain Quaternary Forms 175

Xmv in order that ((>'41) may be of the form (642), tc must be 1 if n is of

the form 81- + 3, and k may have any of the values 2, 3, 4,... if n is of the form
8/„• + 7. Thus the only numbers which cannot be expressed in the form (5'2), in
this case, are those of the form 4X (8/2 + 7 J less than n and those of the form
M+4'(8>;+7), („ = 0, 1,2,3,...),
lying between n and 4n, where « = 1 if n is of the form 8k + 3, and k > 1 if
n is of the form 8k + 7.
(6'5) n = 1 (mod 8).
In this ease we have to prove that
(i) if n > 33, there is an infinity of integers which cannot be expressed
in the form (5'2);
(ii) if n is 1, 9, 17, or 25, there is only a finite number of exceptions.
In order to prove (i) suppose that 17 = 7.4\ Then obviously u cannot
be zero. But if u is not zero u" is always of the form 4“ (8v +1). Hence
rV-m,= = 7.4*-n.4'(8„+l).
Since a ^ 33, X must be greater than or equal to k + 2, to ensure that the
right-hand side shall not be negative. Hence
j\r_,lM! = 4«(si- + 7),
where ls = 14.44-*"8 — nv —£ (n + 7)
is an integer; and so 17 - nu" is not of the form x- + y5 + 22.
In order to prove (ii) we may suppose, as usual, that
17=4* (8/2+ 7).
If X = 0, take u = l. Then
17 - nv? = 8/t + 7 — n = 6 (mod 8).
If X ^ 1, take « = 2X~l. Then
17-m<s = 4*-i(8/i; + 3),
where k = 4 <ji+ 1) — & (n + 7).
In either case the proof may be completed as before. Thus the only numbers
which cannot be expressed in the form (5'2), in this case, are those of the
form 8/2 + 7 not exceeding n. In other words, there is no exception when
« = 1; 7 is the only exception when ?i = 9; 7 and 15 are the only exceptions
when n = 17 ; 7, 15 and 23 are the only exceptions when n = 25.
(6-6) n = 4 (mod 32).
By arguments similar to those used in (6’5), we can shew that
(i) if n h 132, there is an infinity of integers which cannot be expressed
in the form (5'2);
(ii) if n is equal to 4, 36, 68, or 100, there is only a finite number of
exceptions, namely the numbers of the form 4x(8/a + 7) not
exceeding n.
176 On certain Quaternary Forms

(6-7) n = 20 (mod 32).

By arguments similar to those used in (6*3), we can shew that the only
numbers which cannot be expressed in the form (5"2) are those of the form
4i(8Ja + 7) not exceeding n, and those of the form 42(Sfi + 7) lying between
n and 4n.
(6-8) n = 12 (mod 16).
By arguments similar to those used in (6'4), we can shew that the out¬
numbers which cannot be expressed in the form (5'2) are those of the form
4A (8n + 7) less than n, and those of the form
■n + 4-(8i/ + 7), (j> = 0, 1,2,3,...),
lying between n and in, where k = 2 if n is of the form 4 (8/c + 3) and k > 2
if n is of the form 4 (8& + 7).
We have thus completed the discussion of the form (5'2), and determined
the' exceptional values of N precisely whenever they are finite in number.

7. We shall proceed to consider the form

2(x* + f + s*) + nu°-.(7-1)
In the first place n must be odd; otherwise the odd numbers cannot be
expressed in this form. Suppose then that n is odd. I shall shew that all
integers save a finite number can he expressed in the form (7-1); and that
the numbers which cannot be so expressed are
(i) the odd numbers less than n,
(ii) the numbers of the form 4* (16fi +14) less than 4n,
(iii) the numbers of the form n .+ (16yu. + 14) greater than n and less
than 9n,
(iv) the numbers of the form
cn + 4" (I6v + 14), (v'= 0, 1, 2, 3,...),
greater than 9n and less than 25re, where c = 1 if n = 1 (mod 4),
c = 9 if m = 3 (mod 4), * = 2 if rfsl (mod 16), and k>2 if
n- = 9 (mod 16).
First, let us suppose N even. Then, since u is odd and N is even, it is
clear that u must be even. Suppose then that
u = 2», W = 21H
We have to shew that M can he expressed in the form
n? + y"- + z2+2nv*.(7'2)
Since 2n = 2 (mod 4), it follows from (6’2) that all integers except those
-which are less than 2n and of the form 4* (8/i + 7) can be expressed in the
 On certain Quaternary Forms 177
form (7-2). Hence the only even integers which cannot be expressed in the
form (7-1) are those of the form 4*(16n + 14) less than in.
This completes the discussion of the case in which N is even. If N
is odd the discussion is more difficult. In the first place, all odd numbers
less than n are plainly among the exceptions. Secondly, since n and N are
both odd, u must also be odd. We can therefore suppose that
N=n+2M, a2 = 1 + 8A,
where A is an integer of the form p(/fc + l), so that A may assume the
values 0, 1, 3, 6,.... And we have to consider whether n + MI can be
expressed in the form
2 (a?+ y2+ *») + «( 1 + 8A),
or if in the form a* + f- + z"- + 4nA.(7'3)
If M is not of the form 4* (8/a + 7), we can take A = 0. If it is of this
form, and less than in, it is plainly an exception. These numbers give rise
to the exceptions specified in (iii) of section 7. We may therefore suppose
that M is of the form 4* (8/j. + 7) and greater than 4».

8. In order to complete the discussion, we must consider the three cases

in which n = 1 (mod 8), n = 5 (mod 8), and n = 3 (mod 4) separately.
(8'1) »sl (mod 8).
If X is equal to 0, 1, or 2, take A = 1. Then
i/-4«A = 4A(8/r + 7)-4«
is of one of the forms
8*+ 3, 4 (8v + 3), 4 (8k + 6).
If X 5* 3 we cannot take A = 1, since M — 4nA assumes the form 4 (8» + 7);
so we take A = 3. Then
—4nA=4i(8/i+ 7)-12n
is of the form 4(8v + 5). In either of these cases M — 4kA is of the form
a? + y- + z-. Hence the only values of if, other than those already specified,
which cannot be expressed in the form (7'3), are those of the form
4*(8k + 7), (m=0, 1, 2, .... *>2),
lying between 4n and 12m. In other words, the only numbers greater than
9n which cannot be expressed in the form (7T), in this case, are the numbers
of the form
n + 4' (8z/+ 7), (i/ = 0, 1, 2,...,*>2),
lying between 9m and 25n.
(8'2) n = 5 (mod 8).
If A+2, take A = l. Then
if — 4nA = 4A (8/x + 7) — 4k
is of one of the forms
8n + 3, 4 (8m+ 2), 4(8k + 3).
178 On certain Quaternary Forms

If X = 2, we cannot take A = 1, since M- 4ftA assumes the form 4 (8v + 7);

so we take A — 3. Then
U — 4«A = 4* (8/r + 7) — 12«
is of the form 4(8j/ + 5). In either of these cases M — 4uA is of the form
x- + y’- + z-. Hence the only values of M, other than those already specified,
which cannot be expressed in the form (7*3), are those of the form 16 (8/r + 7)
lying between 4n and 12n. In other words, the only numbers greater than
9n which cannot be expressed in the form (7'1), in this case, are the numbers
of the form n + 43 (16/r + 14) lying between 9a and 25«.
(8-3) n = 3 (mod 4).
If X 1, take A = 1. Then
M — 4reA = 4i- (8^i + 7) — 4»,
is of one of the forms 8v + 3, 4 (4r + 1).
If X = 1, take A = 3. Then
if - 4nA = 4 (8^ + 7)-12)i
is of the form 4 (4r + 2). In either of these eases M - 4«A is of the form
+ y-- +
This completes the proof that there is only a finite number of exceptions.
In order to determine what they are in this case, we have to consider the
values of M, between 4n and 12n, for which A = 1 and
M - 4«A = 4 (8fj. + 7 - «) = 0 (mod 16).
But the numbers which are multiples of 16 and which cannot be expressed
in the form x- + y° + z- are the numbers
4“ (Sr + 7), (« = 2, 3, 4, ..., v = 0, 1,2, ...).
The exceptional values of M required are therefore those of the numbers
4»i + 4“(8v + 7) .•.(8*31)
which lie between 4n and 12n and are of the form
4 (8ju + 7)...(8-32)
But in order that (8*31) may be of the form (8'32), k must be 2 if n is of the
form 8k + 3, and k may have any of the values 3, 4, 5,... if n is of the form
8k + 7. It follows that the only numbers greater than 9n which cannot be
expressed in the form (7*1), in this case, are the numbers of the form
9n + 4“ (16v +14), (v = 0, 1, 2, ...),
lying between 9n and 25n, where k = 2 if n is of the form 8k + 3, and * > 2 if
n is of the form 8k + 7.
This completes the proof of the results stated in section 7.
 21
ON CERTAIN TRIGONOMETRICAL SUMS AND THEIR
APPLICATIONS IN THE THEORY OF NUMBERS
(Transactions of ike Cambridge Philosophical Society, 5x11, No. 13, 1918, 259—276)
1. The trigonometrical sums with which this paper is concerned are of
the type
es(n) = 2cos—

where \ is prime to s and not greater than s. It is plain that

where a is a primitive root of the equation

These sums are obviously of very great interest, and 3, few of their
properties have been discussed already*. But, so far as I know, they have
never been considered from the point of view which I adopt in this paper;
and I believe that all the results which it contains are new.
My principal object is to obtain expressions for a variety of well-known
arithmetical functions of n in the form of a series

» fo,(n) cs(n) c3(»)

where a (n) is the sum of the divisors of n. I give two distinct methods for
the proof of this and a large variety of similar formulae. The majority of my
formulae are “ elementary ” in the technical sense of the word—they can (that
is to say) be proved by a combination of processes involving only finite
algebra and simple general theorems concerning infinite series. There are
however some which are of a "deeper” character, and can only be proved by
means of theorems which seem to depend essentially on the theory of analytic
functions. A typical formula of this class is
Ci (n) + ic2 (n) + Jc, («)+... = 0,
a formula which depends upon, and is indeed substantially equivalent to,
the “ Prime Number Theorem ” of Hadamard and de la Vallee Poussin.
* Sec, e.g., Diriehlet-Dedekind, Vorlesungen tiler Zuklentheorie, ed. 4, Supplement vu,
pp. 360—370.
180 On certain Trigonometrical Sums
Many of my formul® are intimately connected with those of my previous
paper “On certain arithmetical functions”, published in 1916 in these Trans¬
actions*. They are also connected (in a manner pointed out in § 15) with a
joint paper by Mr Hardy and myself, “Asymptotic Formulae in Combinatory
Analysis”, in course of publication in the Proceedings of the London Mathc-
■matical Society^.
2. Let F(u, v) be any function of u and v, and let
(2-1) D(,.)=S F(S,S).

L(l) = l'(l,l); D(2) = F(1,2) + F(2,1);

D(3) = F(1,3) + F(3,1); D (4) = F(l, 4) + F(2,2) + F(4,1);
D(o) = F(l,5) + F(o, 1); D(6) = F(1,6) + F(2,3) + F(3,2)+F(6,1);.
It is clear that D (u) may also be expressed in the form
(2-2) D(n) = SF(S’,S).

Suppose now that

(2-3) ^ ,,(»)=SeoS^,
so that 7}s (n) — s if s is a divisor of u and vs(n) = 0 otherwise. Then
(2-4) D(n) = ilvfn)F(n,lf,
where t is any number not less than n. Now let

(2-5) cs(n) = 2 cos — ,

where X, is prime to s and does not exceed s; e.g.

<H(«) = 1; c.(n) = cosntr; cs(n) = 2cosgiiir;
c4()i) = 2cos-J-n-7r; c„ (n) = 2 cos gnir + 2 cos gror;
ca (m) = 2 cos Jnir; c7 (n.) = 2 cos fiwr + 2 cos |mr + 2 cos fmr;
Cs (n) = 2 cos Imr + 2 cos fair; c„ (») = 2 cos giwr + 2 cos grew + 2 cos §«ir;
cM(n) = 2 cos Jn7r + 2cos|»7r; .
It follows from (2'3) and (2-5) that
(2-6) !,„(») = 2 cs(n),

where S is a divisor of s; and hence§ that

(2-7) c,(«) = 2
 On certain Trigonometrical Sums 181
where 8 is a divisor of s, 88' = s, and

t(s) being the Riemann Zeta-function. In particular

<h(n) = vi(n); c.(») = v~.(n)-Vi(n)', e3(n) = v.(n)-vi(n)>
oi(n) = V>(«)-%(»); *(»)-„.(»)-*(»); .
But from (2-3) we know that v&(n) = 0 if 8 is not a divisor of n; and so
we can suppose that, in (2'7), 8 is a common divisor of n and s. It follows

jcg (»)|<; 28,

where 8 is a divisor of n; so that
(2-9) e„(n) = 0(1)
if n is fixed and v x . Since
Vs (n) = >)«(» + «); c, (a) = c„ (n + s),
the values of e,(n) for » = 1, 2,3,... can be shewn conveniently by writing
Cl(n) = I; a,(») = —~1,1; c„(«) = -W72;
c4(n) = 0f-X0fi2l o,(n) = ^l^'l,-l,-l,4;
Cs 00 = I, -1, -TTi; o7 («) = -1,' -T“- lT^X- 1,-T6;
c, (n) = 6,0,0,-4,070, M; c9 (n) = 0,0, STM,~S76, <16;
Cl0(n)=W;i,-l,-4,-l,l,-l,l,4: .
the meaning of the third formula, for example, being that c„(l) = — 1,
cs(2) = —1, c.(3) = 2, and that these values are then repeated periodically.
It is plain that we have also
(2-91) cy(n) = 0(1),
when v is fixed and n —> co.

3. Substituting (2'6) in (2'4), and collecting the coefficients of c,(n)>

c.(n), e.(n), ..., we find that

Cn)
D („) = * (n) 2 \ F („, 2) + c3 (») S l F (2v, £) + c3 (») | ± F (&, £) + ...,

where t is any number not less than n. If we use (2-2) instead of (2-1) we
obtain another expression, viz.
(S'S)
D(») = ClWl lif(1l,v) + c2(«)!^(|;,2v) + c304i;l’g;,3,) + ...,

where t is any number not less than n.

182
Suppose now that
Fl (it, v) = F(ti, v) log u, F, (u, v) = F(u,v) log v.
Then we have
D (n) log n = 2 F (8, 8') log n = $F(8, S') log (SS')

where S is a divisor of n and SS' = n.

Now for S Ft (S, S') we shall write the expression corresponding to (31)
and for S F„ (8, S') the expression corresponding to (3‘2). Then we have

(3-3) D(u)logn = c1wi'0^-f’(.',^)+o2(n)| ^Ffa, £)

+ (»)| ^ *(s-. £) + ■•■ + «•(»)f^^(;■ »)

where r and t are any two numbers not less than n. If, in particular,
F(u,v) = F(v, u), then (3'3) reduces to
(3-4) i D (n) log n = c, (n) 2 f(v,^

+ c, (n) ! F fa, i) + 03 («) I H (3r, i) + ...,

where t is any number not less than n.

4. We may also write D(n) in the form

(4-1) D («) = tF(8, S') +2F(.S', S),

where S is a divisor of «, S8' = n, and u, v are any two positive numbers such
that uv = n, it being understood that, if u and v are both integral, a term
F(u,v) is to be subtracted from the right-hand side. Hence (with the same
conventions)
D{n) = ifa{n)Ffaf)fa\Vu{n)Ffav).

Applying to this formula transformations similar to those of § 3, we obtain

(4-2) 1
D («) = Cl (n) 1 F (v, ?) + c2 (n)% ± F fa, £) + ...
+ c1(«)2iHg1r) + C2(,!)i|Aj’g-,2r) + ...>

where v and v are positive numbers such that uv = n. If u and v are integers
then a term F{u, v) should be subtracted from the right-hand side.
 On certain Trigonometrical Sums 183

If we suppose that ()< Htgl then (4’2) reduces to (32), and if Ocu^l it
reduces to (3-1). Another particular case of interest is that in which u = v.

(*3) D (») = c, (n) 2 i jF fj + F g, „)}

+ c5(«)i2 i (Zv, +F[~, 2ir)| + ....

If » is a perfect square then F(-Jn, gn) should be subtracted from the right-
hand side.

5. We shall now consider some special forms of these general equations.

Suppose that F(u, v) = if, so that D («) is the sum <rs (n) of the sth powers of
the divisors of n. Then from (3-1) and (3-2) we have

(5-1) = Cl (») 2 ^ + c3 (») i + CS (») 2 + ...,

(5-2) <r, (n) = c, («) 2 v*~l + c. (n) | (2i/)"-> + c3 (n) 2 (3r)s-‘ +

where t is any number not less than n: from (3 3)

(5-3) o~s (n) log n = c, (n) 2 ws~' log » + c2 («) 2 (2e)8-‘ log 2v + ...

+ »s |c1(7i)2^^ + c2(»)2^&^ + ...|,

where r and t are any two numbers not less than n: and from (4;2)

(5-4) ors (n) = <h(n) 2 v*-1 + c» (n) ^ (2r)s_1 + o, (m) 2 (3i/)M +...

+ «* je, (n) 2 ^ + o2 (») S (1~f-+; + c3 (n) *2 ~^ + ...},

where uv = n. If u and v are integers then u‘ should be subtracted from the

right-hand side.
Let d(n) = o'0{ri) denote the number of divisors of n and o{n) = o'1(n) the
sum of the divisors of n. Then from (5'1)—(5-4) we obtain

(5-5) d(n) = eI(?i)2i + ca(n)2 ^+<%(»)I ^+...,

(5-6) o(n) = o1{n)[t] + c,(n)[if] + cSr)m + -.

(5-7) id («) log n = ct(n) 2 + c2(n) 2 + c3 (n) k + ... ,
(5-8)
d(») = Cl(n) jl I + | Ij+c.Cn) jl‘I+^J + cs(n){2 i+*2 + ,
184 On certain Trigonometrical Sums

where t&n ami uv = n. If u and v are integers then 1 should be subtracted

from the right-hand side of (5'8). Putting u =v= \/n in (5*8) we obtain

(5-9) Id (n) - c, (n) |l- + cs (•») ‘s i + c, (»)V

unless n is a perfect square, when \ should be subtracted from the right-
hand side. It may be interesting to note that, if we replace the left-hand
side in (5-9) bv
[i+ #(»)].
then the formula is true without exception.

So far our work has been based on elementary formal transformations,

no questions of convergence have arisen. We shall now consider the
tion (5-1) more carefully. Let us suppose that s > 0. Then

The number of terms i: the ! of (5-1) is [t].

that cv (n) = 0(1) asti-s

-'=r(*+i)2^+o

Making £->oo , we obtain

(61) <r„. (n) = n» f(i + 1) j-p+r + -p+r + -jji+i + —| >
if s > 0. Similarly, if we make t -*■ oo in (5'3), we obtain
r Jr
<rs(n)logn = Ci(n)2 v‘~l logv + c2(n) 2 (2v)*-1 log 2v + ...

+ ^^01)2-^

7 (for1
It follows from this and (61) that
(6-2)
<r. (n) + l°g »j = <h (») 2 ^ log v + c2 (n) 2 (2i/)s-' log 2v + ...

■here s> 0 and t^n. Putting s= 1 in (61) and (6'2) we obtain

(6-3) ff(w)=^,l|^)+^)+^(!!) + ..l
 On certain Trigonometrical Sums

+ Cj (n) [f] log 1 + c., (n) [£«] log 2 + ...

+ C; (b) log \t\ ! + c, (n) log [£ i] ! 4-...,

7. Since
(7-1)
rite (6'1) in the form
= o, (n) c20?) c, (n)
(7'2)
£■(« +1) 1,+1 2‘« 3*» +-’
0. This result has been proved by purely elementary n’
But in order to know whether the right-hand side of (7-2) is eonver
not for values of s less than or equal to zero we require the help of tl
which have only been established by transcendental methods.
Now the right-hand side of (7’2) is an ordinary Dirichlet’s series f<
1

The first factor is a finite Dirichlet’s series and so an absolutely c

Dirichlet’s series. It follows that the right-hand side of (7'2) is c
whenever the Dirichlet’s series for 1/J(s + 1), viz.
(7-3)
invergent. But it * that the s> :s (7-£ convergent when
' and that
It follows from this that
(7-4) c1(») + Jc5(B) + ,ics(B) + ... = 0.
Nothing is known about the convergence of (7’3) when — i < s < 0. But
with the assumption of the truth of the hitherto unproved Riemann hypo¬
thesis it has been provedf that (7’3) is convergent when s>-£. With
this assumption we see that (7’2) is true when s > — 4-. In other words, if
— J < s < then
(7-5) ■ <7,(n) = ?(1 - *) |^r+ Cj~ + + •■■}

= B«?(1+S)- CM. 2I+S + 31+fi + •■•[

8. It is known i that all the series obtained from (7’3) by term-by-term

differentiation with respect to s’ are convergent when s = 0; and it is obvious
* Landau, Ecmdbiwh, p. 591. + LitUewood, Comptes Bendm, 29 Jau. 1912.
t Landau, Eandbueh, p. 594.
186 On certain Trigonometrical Sums

that the derivatives of tr_*fn) with respect to s are all finite Diriehlet’s
series and so absolutely convergent. It follows that all the derivatives of the
right-hand side of (7-2) are convergent when * = 0; and so we can equate.the
coefficients of like powers of s from the two sides of (7‘2). Now

(8-1) ?-(F+i) = s">'sS+--’

where y is Euler’s constant. And
«-_*(«) = :£ S-»==Sl-sSlogS + ...,

where S is a divisor of n. But

2 log 8 = 2 log S' = i 2 log (S3') = \ d (n) log n,
where S3' = n. Hence
(8-2) o-_s(n) = d(re)-Jsd(j!)logm + ....
Now equating the coefficients of s and s* from the two sides of (7'2), and
using (8'1) and (8'2), we obtain
(8-3) c, (n) log 1 + ic, (n) log 2 + jc, (n) log 3 + ... = - d (n),
(84) c1(«)(logl)3+ics(n)(log2)=+|c3()i)(log3)!+...=-d(»)(2-y+logn).
9. I shall now find an expression of the same kind for <j> (n), the number
of numbers prime to and not exceeding n. Let plt ps, ps, ... be the prime
divisors of n, and let
(91) ^(n) = n*(l-pr0(l-ft-0(l-|<rs)-.
so that (^j (n) = <f> (n). Suppose that
F(u, v) — /j,(u) vs.

Hence, from (31), we have

But i' easily be shewn that

(94) S /^(nv) =_p(n)_
? ?(S)(l-prs)(l-iPa-8)(I-l)s-s)-’
where pu a3, ... are the prime divisors of n. In other words

(9-5)
, *8 («)?(«)'
 10. I shall now consider an application of the main formulae to the
problem of the number of representations of a number as the sum of
2, 4, 6, 8, ... squares. We shall require the following preliminary results.
(1) Let

(10-1) + +

We shall choose
F(u,v) = e*-', « = 1 (mod 2),
F(u, v) = - if-\ u = 2 (mod 4),
F(u,v) = ( 28-l)re—, « = 0 (mod 4).
Then from (ST) we can shew, by arguments similar to those used in
§ 6, that
(10-11) D (re) = «*-' (l-8 + S-* + 5~s + ...) {I'8c, (re) + 2~8 c4 (re) + 3~s c3(re)
+ 4_* ca (re) + 5~“ c, (re) + 6-8 ols (re) + 7~s c; (re) + S_s cm (re) + ... ]
if a > 1.
(2) Let
(10-2) 2 D (re) *’• = X, = ~ + ....

F (re, v) = d8-1, re = 1 (mod 2),

F (u, re) = re8-1, re = 2 (mod 4),
F(u, re) = (1 — 28) re*-1, re 0 (mod 4).
Then we obtain as before
(10-21) D (re) = re8- (l-8 + 3~8 + 5- + ...) {T-* c, (re) - 2- c, (re) + 3~8 c3 (re)
- 4-* Cg (re) + 5-* c„ (re) - 6“8 Cj. (re) + 7-* c, (re) - S“8 cls (re) + ...
(3) Let

(10-3) 2D(re)®>‘ = Xa = ^—^+^C + ^^+....

v ' w 3 1 + a2 1 + ar1 1 + #°
188 On certain Trigonometrical Sums

We shall choose
F (m, v) = 0, u a 0 (mod 2),
F (a,») = ®*-‘, «e1 (mod 4),
F(u, v) = -v>~\ a = 3 (mod 4).
Then we obtain as before
(10-31)
D (») = ^ (1- - 3- + 5- - ...) {I- c, (n) - 3-» («) + 5- c5 (n)
(4) We shall also require a similar formula for the function D(n)
defined by
(10-4) = +

The formula required is not a direct consequence of the preceding

analysis, but if, instead of starting with the function

we start with the function

s,- ( h) = S (— 1 )=|A_I) si n ,

where X is prime to r and does not exceed r, and proceed as in §§ 2—3, we

can shew that
(10-41)
D (n) = *«- (l-» - 3-* + 5- - ...) {1-* s4(n) + 2- s8 (a) + 3- (n) +...].
It should be observed that there is a correspondence between cr (n) and
the ordinary f-function on the one hand and sr (n) and the function

on the other. It is possible to define an infinity of systems of trigono¬

metrical sums such as c,.(«), sr(n), each corresponding to one of the general
class of “£-functions*” of which f(s) and r](s) are the simplest members.
We have shewn that (10-31) and (10-41) are true when s > 1. But if we
assume that the Dirichlet’s series for 1/??(«) is convergent when s=l, a
result which is precisely of the same depth as the prime number theorem
and has only been established by transcendental methods, then w-e can shew
by arguments similar to those of § 7 that (10-31) and (10-41) are true when

that if s is
= (l + 2a + i
= SM («) +

t Transactions of the Cambridge 1kilosophical .

[No. 18 of this voirume; seeiapartic, lr §§ 24-28,
 On certain Trigonometrical Sums 18ft

where e2S(«.) = 0 when s = l, 2, 3 or 4 and is of lower order* than 8a(n) in

all cases; that if s is a multiple of 4 then

(11-1) (1- +3-' + 5- + ...)28s. (») = —— Z,;

if s is of the form 4k+ 2 then

(ii-2) (1- + s-'+5- +...) SM*0 «n - ;

if s is of the form 4k +1 then

(11-3) (1-* - 3~® + 5-‘ - ...) t K (n) at' = (-~y, (X, + 2-* X,),

except when s = 1; and if s is of the form 4k + 3 then

(11-4) (1~® - 3-' + 5" - ...) 2 SM (re) at' = (Z, - 2-* X,),

X„ Xu, X3, Xt being the same as in § 10.

In the case in which s = 1 it is well known that

=4G+«2+r+^+i +®6 + "')'

It follows from § 10 that, if s is a multiple of 4 then

(11-11) (re) = ~~~ [1- Cl (re) + 2~* c, (a) + 3- c3 (re) + 4- ca (re)

+ B~‘ c„ (re) + 6~“ cB (n) + 7-' c, (re) + 8~® c„ (n) + ...};
if s is of the form 4,1c + 2 then

(11-21) (re) = (1-* o, (n) - 2~® Cj (re) + 3-‘ c3(«)- 4- c8 (n)

+ 5"s c5 (re) - 6_® cI2 (re) + 7"s c7 (re) - 8~“ c,8 (re) + ...);
if s is of the form 4>Jc +1 then

(11-31) 8* (re) =y^-, {l-s Ci (») + 2"s a. («) - 3- c3 (re) + 4- * (re)

+ 5-‘ c, (re) + 6-“ s12 (re) - 7~“ e, (re) + 8~* s10 (n) + ... J,
except when s = 1; and if s is of the form 47c + 3 then

(11-41) &* (re) = (I-® <h («) - 2-® s4 (re) - 3~® c3 (re) - 4~* cs (re)
+ 5“® Cs (re) - 6-s sls (re) - 7-® e, (re) - 8-® sle (re) + ...).

* For a more precise result concerning the order of (n) see § 15.
190 On certain Trigonometrical Sums

From (11-5) and the remarks at the end of the previous section, it follows
that
(11-51) r,(n) = Ss (n) = tr {cx(n) - Jc, (n) + ics(n) - ...}
- ir lK(") + i«.(") + *«» (»)+-}.
but this is of course not such an elementary result as the preceding ones.
We can combine all the formulae (11-11)—(11-41) in one by writing

(11-6) S»(n) =^^y7(l_sc1(m) + 2-8o4(»i) + 3-sc3(n) + 4-®c8(n)

+ 5-' c5 (n) + 6-® Ojjj («) + 7-' c, (n) + 8-* c16 (n) + ...),
where s is an integer greater than 1 and
Cr (n) = cr (n) cos Jits (r - 1) - *, («) sin (r -1).
12. We can obtain analogous results concerning the number of repre¬
sentations of a number as the sum of 2, 4, 6, 8, ... triangular numbers.
Equation (147) of my former paper* is equivalent to

(12-1) (l-2® + 2®‘-2®'+...)«

i / OTi«»s!(i-D /“(*)
a constant and
/(*) = (!-«)(!-*■)(!-*)....
Suppose now that £ = «"■, ®'= «"*'/*.
Then w-e know that
(12-2) V« (1 - 2® + 2®* - 2®» + ...) = 2®1* (1 + x' + ^ + x'° + ...),
(12-3) «)«(*/(*) = *'A/W, V«®*/(*s) = *'A/0O-
Finally 1 + 28* (n)(-®)n can be expressed in powers of x by using the
formulae:
(12-4) «■{*?(!-2,) + ^+|^+j^+...].

-<-W{lr(i-2.)+;ei+|5i+|=i+..i,
where a/3 = w2 and s is an integer greater than 1; and
{1 28 928 *>28 *

=(-/3)®V(2«{i,(-2S) + £I--^I + -^-...},

where a/S = ir>, s is any positive integer, and y(s) is the function represented
by the series l_s —3-s + 5-* — ... and its analytical continuations.
* Loc. cit., p. 181 [p. 159 of this volume].
 On certain Trigonometrical Sums
It follows from all these formulae that, if s is a positive integer and
(12-6) (l+» + x? + x‘ + ...f = %>•'„(n)xf = 28'„(n)of + 2«'*(»)»’*,

then =

where ICn is a constant, and/(®) is the same as in (12-1);

(12-61)
(h-rrY
(1 -' + 3-* + + ...) 2 S'.,s (n) of- =
if s is a multiple of 4:
(12-62)
(l-! + 3-. + 5-S+...)28W«),^pg,-,(g + |^ + f^-l
if s is of the form ik + 2;
(12-63)
-1-^1
...)2S'!s(n),» = |^^
1 +.r“ 1 + .«
| 1« 'o-i | 5»-a* ^

if s is of the form ik +1 (except when s = 1); and

(12-64)
Jl’-'xi i 3*-1 a* 5S~'xi
(1-*— 3-* + 5-
+ l + ^ + i+a# +
3*_‘ 5‘-'xi
1 - xi 1-
if s is of the form 4k + 3. In the which s - 1 we hi

(12-65) SV(n)**-«r*(j^ + -
xi [ J

= *-!(-

It is easy to see that the principal results proved about eM (») in my former
paper are also true of e'!s (n), and in particular that
<’.(»)-0
when s — 1, 2, 3 or 4, and r'^ (n) ~ S'M («)
for all values of s.
13. It follows from (12-62) that, if s is of the form 4k+ 2, then
(l-s + 3~* + 5-s +...) 8'js (n)
192 On certain Trigonometrical Sums

is the coefficient of xn in
2(177)* , fl'-'ab 2s'1* 3*—as5' \

Similarly from (12-63) and (12-64) it follows that, if s is an odd- integer

greater than 1, then (l"* — 3"* + 5“*-...) S'2s (n) is the coefficient of x" in
4(i7r)s , /l*-1 a* 2s-’ x1 3*-1 \
<13'2) (J^)!ar!(r^i+ 1 + a +l + ief )'
Now by applying our main formula: to (12-61), (13'1) and (13’2) we obtain:
(13-3)
S'qs (n) = (n + is)*-111-* c, (71 + is) + 3- cs (n + is) + o- cs (n + is) + ... J

if s is a multiple of 4;
(13-4) S'* (n) = (n + is)*- {1- c, (2» + ^s) + 3- cs (2>i + is)
+ 5-*e„(2tt + is)+ ...}
odd number; and

(13-5) S'M(«); (i7r)‘- (n, + is)*- {l-* c, (4?i + s) - 3— c3 (4n + s)

' (s — 1)•
+ 5-*c5(4n + s)-..
if s is an odd number greater than 1.
Since the coefficient of x" in (1 + « + s? + .. .)2 is that of a4”— in
(* + « + *. + „.).,
it follows from (11-51) that

(13-6) rt («) = 8,' (n) = ^ {c. (4n +1) - fa (4n +1) + fa (4n +1) -

This result however depends on the fact that the Dirichlet’s ser is for 1/t) (s)
is convergent when s = 1.

14. The preceding formulae for <rs (?i), (»), S'^ (n) may be ; rrived at by
another method. We understand by

(14-1)
in (mrjk)

the limit of
k sin (xTrjk)
when x n. It is easy to see that, if n and k are positive integers, and k
odd, then (14-1) is equal to 1 if k is a divisor of n and to 0 otherwise.
When k is even we have (with similar conventions)

(14-2)
A tan (rwr/fc) — ^ •
 On certain Trigonometrical Sums 19!
according as k is a divisor of a or not. It follows that

Similarly from the definitions of (n) and (n) we find that

(14-4) |1- + (- 3)-s + 5_s + (- 7)- + ... J 5. („) - ^ {l- (”)

+ ^ (sin (J'imt+ is7r)) + 3 (sin Qmr + stt)) + 4 (sin (\mr + fair)) + ' "J
if s is an integer greater than 1;

„„) ».(.)-4{{£3 ■-t(££)

(14-6)
(l-r + 3-S + 5-M i° (» + *»)*■

ini(« + i*K
if s is a multiple of 4;
(14-7)
<.-.+«-.+s-.+...>r„c,>- (.- (S||S|^)
, sin (2n +Js)7r \ ,f sin (2n + |s) w \
+ 3 U4(2n,is)J '5 "^sin4(2»i + 4«)7r/
if s is tivice a odd number;
(14-8)
(l-*_3-s + 5 8 — ■ •) 8'k (») = /sin(4n + a)*-\
(n + is)“_1 {l_*
\sin (4n +s) it)
o_»/sin(4n + s)7r\ . ( sin(4n + s)7r \
3 Uj(4» + .)J+6 U'ai(4»+V)W ■
•n odd number greater than 1; and
, / sin (4n. + l)ir \
“ \sin 4 (in. +1) ir)
, { sm(4.n+l)-rr ^
rVsin|(4>
In all these equations the series on the right hand e finite Dirichlet’s
series and therefore absolutely convergent.
194 On certain Trigonometrical Sums

But the series (14'3) is (as is easily shewn by actual multiplication) the
product of the two series
l-‘c1(n) + 2-sc.2(n)+ ...
and n*-1(l-s + 2-‘ + 3-s+ ...).
We thus obtain an alternative proof of the formulae (7*5). Similarly taking
the previous expression of £»,(«), viz. the right-hand side of (11*6), and
multiplying it by the series
l- + (-3)- + 5- + (-7)-+...
we can shew that the product is actually the right-hand side of (144). The
formulae for S'w (re) can be disposed of similarly.
15. The formulae which I have found are closely connected with a method
used for another purpose by Mr Hardy and myself *. The function
(151) (1 + 2® + + ...)» = 2 r!s («)
has every point of the unit circle as a singular point. If x approaches a
“ rational point ” exp (- 2piri/q) On the circle, the function behaves roughly
like

(152) ' (-(2^7rt/s)-iog®[»’

where aPiq = l, 0, or —1 according as q is of the form 4/c + l, 4& + 2 or
4k + 3, while if q is of the form 4lc then av<q = - 2i or 2i according as p is of
the form ik + 1 or 4ft + 3.
Following the argument of our paper referred to, we can construct simple
functions of x which are regular except at one point of the circle of conver¬
gence, and there behave in a manner very similar to that of the function
(15*1); for example at the point exp(— ipirilq) such a function is

(15*3)
(S-1)1
The method which we used, with particular reference to the function

was to approximate to the coefficients by means of a sum of a large number

of the coefficients of these auxiliary functions. This method leads, in the
present problem, to formulae of the type
r2t(n) = S!s(n) + 0(ni°),
the first term on the right-hand side presenting itself precisely in the form
of the series (11-11) etc.
* “Asymptotic formulae in Combinatory Analysis," Proc. London Math.^Soc., Ser. 2,
Vol. xvir, 1918, pp. 75—115 [Ho. 36 of this volume].
 On certain Trigonometrical Sums 195

It is a very interesting problem to determine in such cases whether the

approximate formula gives an exact representation of such an arithmetical
function. The results proved here shew that, in the case of ru (n), this is in
general not so. The formula represents not r.,s («) but (except when s = 1) its
dominant term &■#(»), which is equal to r2s(n) only when s = l, 2, 3, or 4.
When s = 1 the formula gives 28. («.)*.
16. We shall now consider the sum
<16-1) <r,(l) + <r,(2) + ... + *,(«)•
Suppose that

.{(■2n + r)TT\/r}
r-<“> - i * ('5J£^7“I -') ■ '* * 15 sin (t-\/r)
where h. is prime to r and does not exceed r, so that
rr(») = cr(l) + c,(-2) + ... + c,(n)
and Ur(n)=T,.(n) + i<t>(r),
where <j> (n) is the same as in § 9. Since c,. (n) = 0 (1) T —> CO , it follows
that
(16-21) 2V(n) = 0(l), Vr(n) = 0 (r),
as r x . It follows from (7-5) that, if s > 0, then
(16-3)
T,(n) T,(n) , T, (n)
«■-.(!) + + o--.(") = f(* + l)|»+-^

g ■*>(»)_ £(*)
T vs+1 f(s + l)
if s > 1, (16'3) can be written as
• (16-31) <r_,(l) + <r^(2) + ... +<r_(n)

if s> 1. Similarly from (8'3), (8-4) and (ITol) we obtain

(16-4) d(l)+d(2) + ...+c!(n)
= -J3’a(n)log2-jr, (n)log 3 - i Tt (n) log 4 - ...,
(16-5) d (1) log 1 + d (2) log 2 + ... + d («) log n
= i T, (n) [2y log 2 - (log 2)=) + $ («) |2y log 3 - (log 3)=J +
(16-6) r, (1) + ra (2) + ... + («) = cr (« -$T,(«) + ±T, («) -\T7(n) + ...).
* The method is also applicable to the problem of the representation of a number by
the sum of an odd number of squares, and gives an exact result when the number of
squares is 3, 5, or 7. See G. H. Hardy, “ On the representation of a number as the sum of
any number of squares, and in particular of five or seven,” Proc. London Math. Soo. (Records
of proceedings at meetings, March 1918). A fuller account of this paper will appear shortly
in the Proceedings of the National Academy of Sciences (Washington, D.C.) [loc. cit.,
Vol. iv, 191*8, 189—193].
196 On certain Trigonometrical Sums

Suppose now that

T„. (») = 2 (l* cos 2-^ + 2* cos ^ + . - + n*cos ,

where X is prime to r and does not exceed r, so that

Tr,.(n) = Is cr (1) + 2‘ cr (2) + ... + n* cr (n).
Then it follows from (7’5) that
(16-7) <r,(l) + <r,(2) + ... + <r<(n)

if s > 0. Putting s = 1 in (16-3) and (16-7), we find that

(16-8) (n — 1) (1) + (n — 2) a_, (2) + ... + (n - n) (n)
_t . *(») | ».(«) | »«(») ,

n2 (TrnXjr) _
»r(n) = *S
n2 (irX/r)
X being prime to r and not exceeding r.
It has been proved by Wigert*, by less elementary methods, that the
left-hand side of (16'8) is equal to

(16-9) ^ ri1 -\n (y - 1 + log 2?itt) - ^2 /i {4ir V(wi)},

lvhere J"a is the ordinary Bessel's function.

17. We shall now find a relation between the functions (16'1) and (16'3)
which enables us to determine the behaviour of the former for large values
of n. It is easily shewn that this function is equal to

(17-1) 2* + 3s + .
+[9')+iv{3_[vn]l/s-
Now l' + 2*+...+*' = ?(-s)+^

for all values of s, it being understood that

«-., + g±i£
denotes 7 + log (k + j) when s = — 1. Let

&]-
1 (p. 140).
 On certain Trigonometrical Sums 197

Then we have

2
l* + * + ... + [£] -r(-*)+ ^-i 4 (3 + £-(;) +o(^z i)

and v“ = nv‘~' - £rs + e„ jA

It follows from these equations and (17T) that

(17*2) crs (1) + <rs(2)+ ... +<r,(n)= 2 |f (-«) + ) + Wt_1

+ e„ gj + - (0t + e) + + 0 gg)} .

Changing s to -s in (17-2) we have

(17-21)
It* W-. (1) + <T-S (2) + ... + <r_ (»)J = ^2 {„■ ?(,) + 5^ + g)** + e„v°

+e.g)'_(Vn+e)rg*+0(^)1.
It follows that
(17-3)
»s (»■-. (1) + <r-s(2) + ... + o-_s («)) - [a-, (1) + crs (2) + ... + <rs (n))

= ^ |n* ?(s) -?(-«) + g) + nv*-1 + (0i + e) »•

-w-oer+ofs'^))-
Suppose that s > 0. Then, si from 1 to t, it is obvious
that
1/*+1

01
0-
Also £ {»* f«- f(-s)l = (0t-4 +e) |n* ?(s) - ?(-s)J;

| rfs ©>+1w+s> - iri +e)-s+0 (»is)i

2 j—- nv*-' = f (1 — s) + ^ (0r + e)‘ + 0 (»0);

2^ (0i + e) ** = (0i + e) f (- s) + -1- 0 («*“);

^2 (01 + e) g) = n‘ (0i + e) f (s) + gg (0i + 02_! + 0 (?0);

198 On certain Trigonometrical Sums

and J0(^) = 0(4

(17-4) m = ■»**(«< 2), m = nlog n (s = 2), m = n‘~'(s>2).

It follows that the right-hand side of (17'3) is equal to

+?L^.t£i!r£iy^i±£)a:s+ o («).
But Wn + er'- rf+'Un+jri = 2(mj (1+t) + Q {nis).

n(s/n +e)‘ — n* {\Jn + ef~s _ _ B+,, + q

It follows that
(17-5)
<r, (1) + <r, (2) + ... + <r, (n) = *« [<r_ (1) + <r_, (2) + ... + <r_ («)]

- f (1 + «) + *»* ((a) - ^ r (1 - «) + 0 (m)*

if s > 0, m being the same as in (17'4). If s = 1, (17'5) reduces to
(17-6) (n - 1) cr., (1) + (» - 2) <r_, (2) + ... + (n - n) (a)

= »2 - i” (7 ~1 + l°g 2nir) + 0 (V»)t-

From (16-2) and (17-5) it follows that

(17-7) <r, (1) + «r.(2) + ... + »■. («) = ~ S (1 + s) + in* ?(»)

+ sJTlf(1 " s) + »" ?d + •> j|£r + + %3T+ •••} + 0 (»).

for all positive values of s. If s> 1, the right-hand side can be written as
(17-8)

^r(l-S)+n*?(l + S){1^ + H-C|fe) + ^i + %^+-..} + 0(m).

Putting s= 1 in (17'7) we obtain

(17-9) <r, (1) + ffl (2) + ... + ffl (n) = — n* + \n (y - 1 + log 2n.Tr)

* [See Appendix, p. 343].

t This result has been proved by Landau. See his report on Wigert’s memoir in the
Outtingieche gelekrte Anzdgen, 1915, pp. 377—414 (p. 402). Landau has also, by a more
transcendental method, replaced 0 (,/») by 0 («i) (loc. eft. p. 414).
 On certain Trigonometrical Sums 199

Additional note to § 7 (May 1, 1918).

From (7'2) it follows that

f~ {1-«rw(l) + 2-V1_r(2) + ...} = 1-1 nr'cm (1) + 2-2 mr'cm(2) + ...,

C(«)£(r + »-l) _ 11 am (») ,

from which we deduce

?(*)^(8) S'- = ^ + c-^l + + ...,

8 being a divisor of m and S' its conjugate. The series on the right-hand
side is convergent for s> 0 (except when m = 1, when it reduces to the
ordinary series for f(s)).
When s = 1, m> 1, we have to replace the left-hand side by its limit as
s-»-l. We find that
(18) c„(l) + Jc,„(2) + ^c„, (3) + ... = — A (m),
A (m) being the well-known arithmetical function which is equal to logp if
m is a power of a prime p and to zero otherwise.
 22
SOME DEFINITE INTEGRALS
t Proceedings of the London Mathematical Society, 2, xvii, 1918, Records for li Jan. 1918)
Typical formulae are:
m [” —_gin(?-«)( (or 0)
(1) J-,rc«+«)r(^-*) r(a + /s-i) (
° r(«-Hs) . z7n (2 sin jiv y—1 ..
,xr(0+®) - r(/3-a)
(8) J r (a + x) r (/9 - x) en!x Ax

Here n is real, n = 2/ctt 4 i\r (0 <; JV < 2tt) in (2), and n = (2* - 1) t + N
(0 ^ N < 2tt) in (3). In (1) the zero value is to be taken if j n ] 3s it, the non¬
zero value otherwise. In (2) a must be complex: the zero value is to be
taken if n and 31(a) have the same sign, the positive sign if and H(«)<0,
and the negative sign if n < 0 and 3T (a) > 0. In (3) a and /3 must both be
complex; and e„ (?) is 0, 1, or -1 according as (i) rr — n and J (?) have the
same sign, (ii) « ^ it and 3T (?) < 0, (iii) n > it and 3f (?) > 0.
The convergence conditions are, in general, (1) 3&(a+/8) >1,(2) 3&.(a-0)< 0,
(3) fe(a+/3)< 1. But there are certain special cases in which a more stringent
condition is required.
A formula of a different character, deduced from (1), is

fx eniX(lx ~ «** ^lJ.+! {V(2fl cos in)} (or 0).

Here H= +
the zero value is to be taken if |n|>w, the non-zero value othenvise; and
the condition of convergence is, in general, that
*(« + £)>-!■
The formulae include a large number of interesting special cases, such as
” dx 2“+^
-.f(a + a)r(^-a:)T(a + jS-l)’
I _"if_-( i)*2*-'”
x (a?-V) (a?y > (2fc)!’

[” /«+«(X)JrlS_a,()i)da' = J'aW(2X).
 Some Definite Integrals 201

The formula

/_, r <«+«) r (is - .r) r (7+r (S - »)

r(n + /3 + 7+ S — 3)
r(a+/8-i)r(^+7-i)r(7 + s-i)r(s+a-i)'
may also be mentioned: it holds, in general, if
&(a + /9 + 7 + 8) >3.
A fuller account of these formulas will be published in the Quarterly
Journal of Mathematics*.
 23
SOME DEFINITE INTEGRALS
(,Journal of the Indian Mathematical Society, XL, 1919, 81-87)
I have shewn elsewhere* that the definite integrals

in be evaluated in finite terms if w is any rational multiple of i.

In this paper I shall shew, by a much simpler method, that these integrals
m be evaluated not only for these values but also for many other values of

, coshTrfm^.
\/w Jo cosh™
where to stands for 1/ui.
It follows that

W “ -jr„ e (*"')• • •

<j>w(t + w) =

x [ytoe^+2fo jo
,-ilr(t + w)»/w

* Messenger of Mat/iematics, Vol. 44, 1915, pp. 75—85 [No. 12 of this volume].
 Some Definite Integrals 203

From (1), (2) and (3) we easily find t]

It is easy to see that

<#>»(*') = Aj! f»W = 2^; <f>w(w) = ie"1x117

0=( +-V1,r“’
’ \2 ^ _ 27

+.(«-±0 = i ± s-W'1”; *.(*»>+ +•(*»>-*•

From (1) and (5) we deduce that

^j=r(!+!o)-/ii>^ (j _|_ 4. e^(‘~w^lw <f>w(t- w) - e^1.(1)

Similarly from (4) and (6) we obtain

ei- <‘+»)> {J - fw (t + w)} = eir (t' v,)> {i + irv (t - w)>.(S)
It is easy to deduce from (5) that if n is a positive integer, then
^(t) + (-l)’*«^(t + 2ni)
= + t0 n teraJ . /9)

Similarly from (6) we have

fw(t)-fw(t±2ni)
= z± L-i*-(<+<)> -i*(t+8«M® + e-M‘+K>> + _ t0 B termsl
204 Some Definite Integrals

Again from (7) we have

,r («) + (-1)”+1 eiTlt + 2"'1’>2/”' <t>« (t + Znw)

= e^‘+v)'la - eilr<,+3“» + e^{,+~ow)ilm - ... to n terms; ...(H)
and from (8)
e^1* ,i + ^ + 1)Ii+ici»-('+-2'aO> {J +^w(« + 2nw))

^(«+4»w« + t0 n terras. ...(12)

Now, combining (9) and (11), we deduce that, if m and ?i are positive
integers and s = t + 2mw ± 2ni, then
<j>w (s) + (- l),"l+’1 "+1' e-i™|s+l! 4>,c (t)

= e - *’”> | eir'(s - “r~,w - eir (s “ 3‘°r',,c + eir(s~ 6b)i/” - ... to m termsj

+ Jw e
xj,-M<*f)»/»_e-W«*»0a/- + a-M«*«0*/-_...to „ terms(.

.(13)
Similarly, combining (10) and (12), we obtain
i~ (s) + (- 1 _ fw («)}

=a' - 2il,» - eiTts' 4w» + ei,r(s ‘ 6,")> - ... to m termsl

+ (_ 1 jMn+m+i _L &-irrm (s+t)

x {e+ e-^<I±3i»+ ...to « terms},

.(14)
where s and t have the same relation as in (13).
Suppose now that s = t in (13) and (14). Then we see that, if w = w/m,
then
(pa (f) (1 + (- l)(m+1) <»+» e-™1}
= \e W‘-)>_eiv(1-M> + ^0-51r)>_ _ t0 m t)

+ (-^ jt0 B termsJ

.(15)
 Some Definite Integrals 205

.(16)
where \hu should be taken as

•*ve>
In (15) and (16) there is no loss of generality in supposing that one of the
two numbers m and ti is odd.
Now equating the real and imaginary parts in (15), we deduce that, if m
and n are positive integers of which one is odd, then

= [cosh {(1 - n) f} cos (irm/4n) - cosh {(3 - n) t} cos (dirmjin) + ... to n terms]

= - [cosh {(1 - n) t} sin (wm/4n) - cosh {(3 - n) t} sin (dmnjin)

+ cosh {(5 — n) t} sin (2o7r/4n) - ... to n terms]

V©M(*-=M*(i-=*S

.(18)
Equating the real and imaginary parts in (16), we can find similar
expressions for the integrals
206 Some Definite Integrals

From these formula) we can evaluate a number of definite integrals,

such as
r cos 2trtx ,, 1 + V2 sin irt2
Jo cosh irx C°S X 2 ^/2 cosh 7r< ’

Again supposing that s = -t in (13), we deduce that if t = mw + ni, where

and n are positive integers of which one at least is odd, then

This formula is not true when both m and n are even.

If t = mw + ni, where m and n are both even, then

<K» CO + (- l)“+iml <1+in| (0)

This is easily obtained by putting t = 0 and then changing s to t in (13).

Similarly from (14) we deduce that if t = mw ± ni, where m and n are both
even, or both odd, or m is even and n is odd, then
 Some Definite IntegraU

if t = raw ± ni, where m is odd and n is even, then

i- (t) + {(- 1)'+1l»+5' e~i’ <t>w (0)
_ - iafjw toft* , . . ,

J -ir(tcii)*/«
+ ... to fn terms). ...(22)

is obtained by putting i = » in (14). A number of definite integrals

is the following can be evaluated with the help of the above formulas:

g-nt+ijj’jfo, —-
 24

A PROOF OF BERTRAND’S POSTULATE

(Journal of the Indian Mathematical Society, XI, 1919, 181—182)

1. Landau in his Eandbuch, pp. 89—92, gives a proof of a theorem the

truth of which was conjectured by Bertrand: namely that there is at least
one prime jj'such that ®<ps;2®, if ®>1. Landau’s proof is substantially
the same as that given by Tschebyschef. The following is a much simpler one.
Let v (of) denote the sum of the logarithms of all the primes not exceeding
x and let
+ («) = » («) + »(«*)+>'(«») +.(!)
log [x]\=f(x) + f(%x) + ir(ix)+ ..., .(2)
where [®] denotes as usual the greatest integer in ®.
From (1) we have
f (®) - 2* ( V®) = v O) - v (si) + v (si) -.(3)
and from (2)
log [s]! - 2 log [Js] !- + («)-* (is) + + (*«)-.(4)
Now remembering that v (s) and i|r (at) are steadily increasing functions, we
find from (3) and (4) that
^(®)-2t(v'®)«i'(®)<f (*•);.(5)
and f (at) - f (4®) < log [®]! - 2 log [is] k -f (*) - f (4®) + -f (4*)- • - (6)
But it is easy to see that
log r (at) - 2 log r (4®+4)«log [®] i - 2 log [4®] \
log r (®+1) - 2 log r (i® + i). ...(7)
Now using Stirling’s approximation we deduce from (7) that
log M! - 2 log [|®]! < i x, if ® > .(8)
and log [®]! - 2 log [4®]! > §®, if ® > 300.(9)
It follows from (6), (8) and (9) that
-f (®) - \jr (4®) < |®, if ® > 0; .(10)
and (x) ~ 'h(4*) + 'h(4*) > 1 ®. if »>300.(11)
 A Proof of Bertrand’s Postulate 209
Now changing x to \xt \x, J«,... in (10) and adding up all the results,
we obtain
*(*)<§*, if *>0.(12)
Again we have
f(x) -■f(ix)+i/rQx)^p(x) + 2ifr(^/x)-v (i x) + -f Qx)
<v(x)-v(\x) + \x + Zd*, ~.{rz)
in virtue of (5) and (12).
It follows from (11) and (13) that
v(x)-v(lx)>\x-3dx, if *>300.(14)
But it is obvious that — 3 \Jx > 0, if x ^ 324.
Hence v(2*)- v (x) >0, if *^162.(15)
In other words there is at least one prime between x and 2x if x > 162.
Thus Bertrand’s Postulate is proved for all values of x, not less than 162: and,
by actual verification, we find that it is true for smaller values.
2. Let tr (x) denote the number of primes not exceeding x. Then, since
ir (x) - w (\x) is the number of primes between x and J®, and v(x)-v(\x) is
the sum of logarithms of primes between x and \x, it is obvious that
v{x)-v (\x) « {tt (*) - -x (**)} log®.(16)
for all values of*. It follows from (14) and (16) that

w (*) - w {\x) > -L (j* - 3 V*), if * > 300.(17)

From this we easily deduce that

7T (*) - 7T (ix) > 1, 2, 3, 4,5,..., if * » 2, 11, 17, 29, 41,...... .(18)
respectively.
 25

SOME PROPERTIES OF p [n), TPIE NUMBER OF

PARTITIONS OF n*

(.Proceedings of the Cambridge Philosophical Society, xrx, 1919, 207—210)

1. A recent paper by Mr Hardy and myselff contains a table, calculated
by Major MaeMahon, of the values of p(n), the number of unrestricted
partitions of n, for all values of n from 1 to 200. On studying the numbers
in this table I observed a number of curious congruence properties, appa¬
rently satisfied by p (n). Thus
(1) P (4), 2* 9), A (14), it (19), ... = 0 (mod 5).
(2) 2*5), 2*12). 2*19). P (26), ... = 0 (mod 7),
(3) p( 6), 2*17), P (28), it (39), ... = 0 (mod 11),
(4) 2*24), p (49), it (74), it (99), ... = 0 (mod 25), N
(5) vm. P (54), it (89), it (124), ... = 0 (mod 35),
(6) 2*47), P (96), p(145). P (194), ... = 0 (mod 49),
(U Pm, 2*94), p (149), = 0 (mod 55),
(8) 2*61), p (138), = 0 (mod 77),
(9) 2*116). = 0 (mod 121),
(10) P (99), • = 0 (mod 125).
From these data I conjectured the truth of the following theorem:
If S = 5«7>11« and 24A = 1 (mod S), then
p(X), p(\ + S), p(\ + 2&), ... =0(mod 8).
This theorem is supported by all the available evidence; but I have not
yet been able to find a general proof.
I have, however, found quite simple proofs of the theorems expressed
by (1) and (2), viz.
(1) • p (om + 4) = 0 (mod 5)
and (2) p (7m + 5) = 0 (mod 7).
From these
(5) p (35m + 19) = 0 (mod 35)
* [See also Ramanujan’s posthumous paper “ Congruence properties of partitions ” in
the Jlath. Zeitsehrift, No. 30 of this volume.]
v t G.I. Hardy and S. Ramanujan, “ Asymptotic formula in Combinatory Analysis,”
Proc. London Math. Soc., Ser. 2, Vol. xvn, 1918, pp. 75—115 (Table IV, pp. 114—115)
[No. 36 of this volume].
 Some Properties of Partitions 211
follows at once as a corollary. These proofs I give in §2 and §3. I can also

(4) P (25n + 24) = 0 (mod 25)

and (6) P (49n + 47) s 0 (mod 49),
but only in a more recondite way, which I sketch in §4.

2. Proof of {1). We have

(11) x ((1 — *) (1 — a2) (1 — xs) ...}4
= « (1 — 3a+ 5iS3 - 7a5 + ...) (1 — a; — «2 + a5+ ...)
= 2 (-1)“+» (2p + 1) cc'+i"
the summation extending from p = 0 to p = «> and from v=~x 1011=00.
Now if
1 + ir- (p + 1) + (3v +1) = 0 (mod 5),
then 8 + 4*p(p + 1) + 4j/ (Sv +1) = 0 (mod 5),
and therefore
(12) (2^ + 1)“ + 2 (1/ + l)2 = 0 (mod 5).
But (2p + l)2 is congruent to 0, 1, or 4, and 2 {v + l)2 to 0, 2, or 3. Hence it
follows from (12) that 2p + 1 and v + 1 are both multiples of 5. That is to
say, the coefficient of in (11) is a multiple of 5.
Again, all the coefficients in (1 - *)~s are multiples of 0, except those of
1, a?, .... which are congruent to 1: that is to say

dip r2pmod 5)-

” (modS)-
Thus all the coefficients in
(1-^)(1-^)(1-^)...
{(l-aKl-s2) (1 -a?)...)”
(except the first) ai mltiples of 5. Hence the coefficient of a“ in

(l-«)d-^)d-«>)
a multiple of 5. And hence, finally, the coefficient of a™

(1 — as) (1 — a?) (1 —so*) ...

is a multiple of 5; which proves (1).
212 Some Properties of Partitions

3. Proof of (2). The proof of (2) is very similar. We have

(13) a:2 {(I — aj) (1 — rc=) (1 — *3)

= 2 (- 1 r+v (2/i+ 1) (2j/ + 1) s*+1m

the summation now extending from 0 to co for both /i and v. If
2 + i/i (/i+ 1) + iv(v + 1) s 0 (mod 7),
then 16 + 4/i (/i + 1) + 4n(n + 1) = 0 (mod 7),
(2/i + 1)= + (2i< +1)2 = 0 (mod 7),
and 2/t+l and 2v + 1 are both divisible by 7. Thus the coefficient of x'n in
(13) is divisible by 49.
Again, all the coefficients in
(1-®')(1-®“)(1-*")...
{(l-®)(l-ir')(l-®»)...}'
(except the first) are multiples of 7. Hence (arguing as in § 2) we see that
the coefficient of x'n in

is a multiple off7; which proves (2). As I have already pointed out, (5) is a
corollary.
4. The proofs of (4) and (6) are more intricate, and in order to give them
I have to consider a much more difficult problem, viz. that of expressing
p(X) + ppc + 8)® + p(\+2S)*2+...
in terms of Theta-functions, in such a manner as to exhibit explicitly the
common factors of the coefficients, if such common factors exist. I shall
content myself with sketching the method of proof, reserving any detailed
discussion of it for another paper.
It can be shewn that
(1-^(1-^) (!-*»)■■■ _
(14)
(l-«*)(l-®»)(l-«*)... t1-
- 3*? + (t* + 2xp) + x% (2|-2 - xP) + x* (3f- + xp) + 5®*
p*-\\x-a?p ’
where ^... (1-«) (1-^) (1-«■) (1

the indices of the powers of x, in both numerator and denominator of f,

forming two arithmetical progressions with common difference 5. It follows

(15) (1 - s») (1 - s>°) (1 - a“)... {p (4) +p (9) * +p (14) + ...}

f->-Ux-x*p'
 Some Properties of Partitions
Again, if in (14) we substitute cox-, or'xK and <b'®=, where a
or xi, and multiply the resulting five equations, we obtain
f(l - a-) (1 _ *») (1 _ «-) 1
1 ' 1 (i - X)a-»’) (i -x>)...}
From (15) and (16) we deduce
(17) p(4)+p(9)* + p(14)ir!+...
= {(l-af)fl-Q(I -x«) p
{(i-«)(i-^)(r-^,';..p>
■oni which it appears directly that p (ora + 4) is divisible by 5.
The corresponding formula involving 7 is
(18) p(o)+p(12)x+p(19)x'- + ...

Kl-®5)-}'
-*7) (1 - zu) (1 - a?1) ■
' {(!-*)(!-
which shews that p (7m + 5) is divisible by 7.
From (16) it follows that
p(4,)x + p(9)z’ + p(U)x? +
5 {(1-*•)(!-*“) (1- •••}*
(1-*)(1-»»)(1-«»)■.
(1 -x) (1 -x?) (1 - a?) ... {(1 -X) (1 - a?) (1 - s?) ...}s'
As the coefficient of xm on the right-hand side is a multiple of 5, it follows
that p (25m + 24) is divisible by 25.
Similarly
p(B) x+p (12) + p (19)«3 + ...
t {(1— s?) (1 — ®“) (1 —x?')...}!
(1-^(1-«»)■■■
= x(l-Sx + 5x3-1af+...) ^
■;{(l-a')(l-a?)...F
{(l-^)(l-Q.-}t.

from which it follows that p (49m + 47) is divisible by 49.

Another proof of (1) and (2) has been found by Mr H. B. C. Darling, to whom my
conjecture had been communicated by Major MacMahon. This proof will also be pub¬
lished in these Proceedings [see Appendix, p. 343]. I have since found proofs of (3), (7),
and (8).
 26
PROOF OF CERTAIN IDENTITIES IN COMBINATORY
ANALYSIS

{Proceedings of the Cambridge Philosophical Society ^

’.q){l-x<f) ... (1 -«g»-‘)

1
“1"(1" *?!) ~q + (1"xt) RR --.(I)
If we write 1 -xp~ 1 - g' + g" (1 -mg"),
every term in (1) is split up into two parts. Associating the second part of
each term with the first part of the succeeding term, we obtain

G (x) = (1 - x?f) - as!gs (1 - n?qB) \~fg

...(2)

Now consider H (x) = - & (xq).(3)

Substituting for the first term from (2) and for the second term from (1),
we obtain ,
H (x) = xq - (d ~ 2) + *24 d ~ ®92)}

+(f-Sa-f) «x - «*>+^ - *9s)}

Associating, as before, the second part of each term with the first part of the
succeeding term, we obtain
H (x) = xq (1 - *2=) jl - x"q‘ (1 - xq') -L-

+ tc*g17 (1 - *Y)
(1-gKl-S*) .
-aVMl-atfl (1-^3)(I -xq')
3 (1 X^' (T — g) (1 — g!) (1 — g3) ”
— xq(l — xq*) G (xq%
* [See the Appendix, p. 344, for ■
 Identities in Combinatory Analysis 215

If now we write K (x) = r——r,

(1 — xq) G (xq)
we obtain, from (3) and (4),

.<«>
In particular we have
J_ JL __2l_ = 1 .(l-g)g(?).
1 + 1+1 + ... if'(l) G (1) .w
or J_JL g’ _l-g-9J + g7 + ?11-...
1 + 1 + 1 + ... 1 -qa-q' + qs + q1'- ....1 '
This equation may also be written in the form
JLJL JH_(I - q) (1 -<f) (I -<?6) (i -0°) q- gn)...
1 + 1+1 + ... (l-52)(l-r/)(l-?')(l-(/)(l-2.»)...- -W

If we write F(x)=T-- ^ ^ 7-
(l-xq)(l-xq-)(l-a;f)...’
then (4) becomes F(x) = F (xq) + xqF (xq-), ,
from which it readily follows that

F (x) = 1 + -3L +__+ _ _ , ,fl,

' 1-2 (1 -?) (1~22) (1 ~?) (1 — 22)(1 —
In particular we have
l ■ g I t_+ gw
• 1-3 (l-?)(l-95) (1 — ?) (1 — q") (1 — 93) •••
l-q--qa+q0 + qu-...
(i — ?) (i — q~) (1 — 2s) - • -

= (l-2) (l-2‘)(l-r/)(l-2»)(l-?“)-’ .(10)

and 1+ 'f +_t_+ „_(i-g)G(q)_
i-9 (i-2)(i-?3) (i — 9)(i-?2)(i — qs)■■■

(l-2)(l-9=)(l-28)...
.(11)
(1-22) (I-23) (1-27) (1-2*) (1-?'“)•••
 27
A CLASS OF DEFINITE INTEGRALS

(quarterly Journal of Mathematics, xlyiii, 1020, 294—310)

1. It is well known that

(l-D =^

if R (m) > - 1. It follows from this and Fourier’s Theorem that, if n is any
real number except ± ir and R(a+fi)>l, or if u = ±tt and R(a + jS)>2, then

(1'2) /.„ fca+^ros-^jdx=(r(«+V-i) e4"‘^a)or 0>

according as \ n\<ir or |n| In particular we have
f" * 2“+?”2
(1’21) J r{a + x)T(R-x) ~ f (« + £-l)
if U(a + /3) > 1; and

(1’22) io r(a +.a) T(a — a) = r^TT^T)

if R (a) > J. If a is an integer n + 1, (1’22) reduces to

(1'23) /„ ®{1 - <^/l=)}{! ~ OV22)}... {1 - 07«2)} & = 2 "(2^ ’

Again, if m is a positive integer, we have

+ 2 cos 2™ 4- 2 cos 4? ; (m +1) terms

2 cos ™ + 2 cos 3™ + ... to \in terms,
according is odd or even. It follows from this and (T2) that, if R(a) >1,
f*smm7ra; dx 2!“-3
o sin™ F(et + a)r (a — — T (2a — 1) or U’
according as m is odd or even. Hence, if m and n are positive integers, we

(1-24)
f_sip™^_or 0
)0 x{l-(a?/!2)}(1 - (®!/22)}• • • {1 - (®>2)} " 2 (2») 1 U’
according as m is odd or even. From this we easily deduce that, if l, m, and n
are positive integers,
(1"25)
[“ (sin ,
0,
i 0 « {1 - wm {1 - (rf/2*)} («■/»•)} :
 A Class of Definite Integrals

cording as m is odd or even. It follows that

® {i - m) {i - m) •■•{!- ~5
cording as m is odd or even. Similarly we can shew that
(sin rmrz)-1
(1-27) dm = 0
/o
for all positive integral values of l, m, and n.
In this connection it is interesting to note the following results:
(i) If R (a +13) > 1 and B (7 + 8) > 1, then
ft (2 cos 7ra)y+s-s e1'
a- r(« + /3-i) r ck
-i V(a + x)r(f3-x)

— dx.

This is easily proved by writing

J4 (2 cos Trzy+t-* e<’z

insteadof r(g'+^ro8-g)
in the left-hand side of (1'3).
(ii) If m and n are integers of which one is odd and the other e\
m>0 and R (a + /3) > 2, then
(1-4)
(cos 7rx)m e‘‘ +I (cos Trx)m e‘'"x ,
(a + ft — 2)
r(ct + x)-r(ff-x) r (a — I + a) r (,8 — a) *
where f is any real number. This is proved as follows: Suppose that the
left-hand side, minus the right-hand side, is/(f). Then
+ ff-2) (cos 7r£)’» . .
fr« + f>ros-f)
(- 1 )m+n (cos 7rjf)”* e"
r(u + ?)r(^-f-i) Tr(a_i + ?ir(/9-f) •
Hence/(f) is a constant which is easily seen to he zero.

2. Before proceeding further I shall give a few general rules for

generalising the results in the previous and the following sections. If
 218 A Class of Definite Integrals

where r is zero or a positive integer, and the e’s, f, t], ? and F(x) are all
arbitrary, then it is easy to see that
(2d) /w (f + 1) = + {/,<?)- (? + ert.)/r(f + 1)},
provided the necessary convergence conditions are satisfied.
Similarly if
/-• (?) = /’’(* + <u) (* + <*) ...(* + <?,) T (? + *) P(«) diK,

(2-2) /,+I (?) = ± {/„(?+ 1) - (? + **)/,(?)}.

Thus we see that, if/0(?) is known,_/(•(?) can be easily determined.
Suppose now that P(.-c) is a polynomial of the rth degree and N any
integer greater than or equal to r. Let D, E and A denote the usual operators
so that
E = l + A = sD.

«»-/,'rfiy--
■ (2'3)J^^i^* = !^^(±A)”P{-^±(l-?+iv)},

as is easily seen by replacing P (x) by

(1 + AP-1±5)«*-'P{+(1-?)}.
Similarly, using the equation
P (x) = (1 + P (+ ?),
we find that, if
/a)=Jja±^)E(x)dx,

(2-4) £ T (? ± x) P (*) F(fi) dx = |'^pL) (± A)> P {- iv + (? + J „)}.

As an illustration let us apply (2-3) to (1'2). We find that if n is any

real number except +tt, andP(a+/3)>l+r, or if ti=±tt and P(n+/3)>2+r,
and N is any integer greater than or equal to r, where r is the degree of the
polynomial P (x), then

^v\T(u + ^-v-l)e
according as ]?i| > ir or \n\ < v, kv being either e**1' A"P (1 — a) or
e-Hn•■ (_ A)’P 08
It is immaterial which value of h, we take.
 A Class of Definite Integrals

If w r (?„+*)’
where - £> is a positive integer, then it is well known that
A. p M -.-(&-»■ X (6+f)
("6) () (?.-f.-«')ir(g, + a + v)-
This affords a very good example for the previous formula?.
It is easy to see that (1-2) can be restated as

(2-7)
L,.V(a+ x) ?{$-«:)

(2'71)
Lrf. + rjros-xj^ifwi

rrwr(«+^„)!
according as ;n| > 5 7r (n being of course real). But
>+l"|a;P( 0)
/-«r(a+ffi)r( -*) 0 dx=j.„
r(«+»)r(/ -^) ““ 3
Hence, if the conditions stated for (2-5) are satisfied, ’
fx P OK'”* , n
(*8) J (a + &) r (ft — sc)dx ~°
(2-81)
J_»r(a + .r)r(/S-a!) frwr(«+/s-r)
| e^-QP(l3-v)

3. We shall now consider an important extension of (T2). Let \x] denote

the greatest integer not exceeding x, so that (e.g.) [- 5^] = - 6. Let us agree
further that

■e real and tf> (z) is a function that can be expanded

IcU-
when \z\ = 1, we have

u
,ia!i=lfe+
220 A Class of Definite Integrals

where the summation is bounded by

provided that either

(i) 7r + n and 7r — n are not multiples of s, and B (a + fi) > 1,
or (ii) rr + n or rr - n is a multiple of s, and B (a + /3) > 2,
or (iii) QiT-.ni!* =0 and C'_^+„i/s = 0, whenever one or the other or both of
the suffixes of 0 happen to be integral, and B (a + /3) > 1,
or (iv) w + n and rr — n are multiples of s, a + /3 is an integer greater than 1,
and CV-m/s = e2'™
The formula (3T) is easily obtained by substituting the series for $ and
integrating term-by-term, using (1‘2).
It should be remembered that (3d) is not trrn if n 0:
though the integral may be convergent. In such c s, generally, the integral
cannot be evaluated in finite terms.
The following integrals can be evaluated at on in finite terms, with the
helpof(S'l): •
f“ _ dx_
(3-11) )-x(peimx + qe,n:e) T(a + x) T(/3- x) ’
where m and n •e real and \p\J='ql;

(3-12)

where n and real and to is a positive integer;

(3'13)
L T(a + x)T{fi-sc) ’
where m is real and

(3T4)
r(a + ®)r(/3-«)
For instance, the value of (3'14) is

2(f
where Jv (®) is the ordinary Bessel function of the vth order, and {{q+p)/(q -p)}il'
should be interpreted so that the first term in the expansion of
{(?+P)l(q ~ i))}1" J. W(q- - ?’)}
is {i(p + q)YM-
 A Class of Definite Integrals

5 _e;..*.u
.D\« + !c)T{13~x)
= C {2cw(t»-W-[(9r + «)/ar]))^

provided that either

(i) n ,n odd multiple of tt, and R (a + /9) > 1,
• (ii) * n odd multiple of jr, and R (a + 0) > 2,
'• (iii) n n odd multiple of nr, Oy-^/K=0, C_(„+„i;w=0, and jR(a+/3)>l,
■ (i'O multiple of 7r, a + /3 is an integer greater than 1, and
Cw_„,>=e3,’"‘CU+
Thus we see that the value of each of the integrals (312)—(314), when
s = 2ir, reduces to a single term.
The next section will be devoted to the application of (3’2) in evaluating
some special integrals. ,
4. Suppose that a is not real and
<£(*) = 1 + +...(/(«)< 0),
and <£(*) = - e*’-*-1 - -...(/(«)> o),
so that <fi (z) is convergent when \z| = 1. Then it is easy to see that

vjr~4^k^!^ **■
It follows from (3'2) that if a is not real, n real, and
(i) n is neither 0 nor any multiple of 27r, and R (a — ft) < 0,
or (ii) n has the same sign as 1(a) and R(a- ft) < 0,
or (iii) n is 0, or a multiple of 2w, having the sign opposite to that of I (a),
and R(a- ft)< — 1,
or (iv) n is not 0 and a — /S is a negative integer, then

<«>

the zero value being taken when n and I (a) have the si ae sign, the plus
sign when n ^ 0 and / (a) < 0, and the minus sign when n t ) and I (a) > 0.
 A Class of Definite Integrals

As particular < ;cs of (4'1) we have

(4-11)

if a is not real and R (a - ft) < -1. If a is not real, n real, and (i) r is
positive integer, or (ii) r = 0 and n 10, then

or

the different values being selected as in (4T).

Similarly we can shew that if a and ft are not real and n is real, and
(i) n is not an odd multiple of it and R (a + ft) < 1,
or (ii) n is an odd multiple of it which has either the sign of I (ft) or the
sign opposite to that of I (a), and R (a + ft) < 0,
or (iii) « is an odd multiple of w which has neither the sign of I (ft) nor the
sign opposite to that of I (a), and R (a + ft) < 1, then
(4-2)
f_J(a + x)T(ft- x) e»'*dx - **• «■-*

X (nn(ft) exp (ITT (a + ft) [(tt + «)/2’r]l - Vn(- a) exp {- iir (a+ft) [(tt + n)j
where (f) is equal to 0 when v — n and I(f) have the same sign, to 1 when
n ^ 7r and I (f) < 0, and to - 1 when n ^ it and I (g) > 0.
It should be remembered that for real values of n
12 cos }n| = 2 cos Qn-ir [(* + »)/2tt]).
It follows, in particular, that if a and ft are not real, and R (a + ft) < 1,

(4-21) J r (a + x) r (ft - x) dx = 0 or + j7iT (a + ft),

the zero value being chosen when I (a) and I (ft) have different signs, the
plus sign when I (a)' and I (ft) are both negative, and the minus sign when
I (a) and I (ft) are both positive.
The following results can either be deduced from (1-5) or be proved
independently in the same way as (1-4).
If n is zero or any multiple of 27t, f is real, a any number except the real
numbers less than or equal to - £ and ft is any number such that R(ft-a) > 1,
then
“ r(g + a> gn,i2. [t+
er(ft + x)e d“-.l{
 A Class of Definite Integrals 223

If n is any odd multiple of ir, a and f are the same as in (4'3), and /3 is
not real and If (a + /?) < 0, then

(44) (a + /3)J T(a + x)T($-x)ei"Idx

= J Tl r (a + x) V (0 + 1 - x) e^dx.

5. We now proceed to consider an application of (T2) to some other

functions. Suppose that U,{x), VB (®) and Ws(x) are many-valued functions
of x defined by
_ ?/„.** v.2ri*+-'
Us (ie) “ r(i+s)+ F(i +7+ 0 + r (1 +7+Te)+''' ’

Fs (a?) = rjh s)+ fo+T+O + r(T7JT2e)+'' ’ ’

W‘ (,c) = F(I+7)+ ro+7+7)+ F(T+7+ 2<o+'''' ■’

where If (e) > 0, the tis, v’s, and w’s are any numbers connected by the relation
to be found by equating the coefficients of the various powers of k in the
equation
Wv + wjc + wjir + ... = {ua + u,k + u.,k- + ...) (v„ + vjc + v'k-+ ,
and the series XJS {x), Vs (*) and Ws (x) are convergent at least for the values
of s and x that appear in the equation (5’2).
The functions U, V, and W are many valued. If j xjy \ = 1 and [ arg {xjy) j < ir,
then one value of arg (* + y) is given by the equation
(o-l) arg x + arg y = 2 arg (x + y).
If we choose arg® and argy arbitrarily, and agree that
®s+"' = exp {(« 4- ne) (log J x | + i arg x)},
and that and (x + yY+li< are to be interpreted similarly, that value of
arg (x + y) being chosen which is given by (5'1), then a definite branch of IF
is associated with any arbitrary pair of branches of U and V.
If a, /3, x, y are any numbers such that | xjy j = 1, and If (a + /3) > 0 when
j arg {xjy) | = 7r and If (a + ft) > — 1 otherwise, then

(5'2) J_a or Wa^(x + y),

according as [ arg {xjy) | > 7r or < w, whatever be the branches of Z7 (x) and
V(y), provided that the corresponding branch of W (x + y) is fixed in accord¬
ance with the convention explained above. This is proved as follows. Suppose
that
x = teiin, y = te~Vn,
* These series need not, of course, be convergent for any value of k.
224 A Class of Definite Integrals

where t is arbitrary and n is any real number. Then the integral becomes

If we expand the integrand in powers of t, and integrate term by term

with the help of (1-2), and then make use of the relations between the us, v's,
and tti’s, the result will be
F„+P"(2f cos H) = F.+„ (x + y),
or zero, according to the conditions stated with regard to (5'2).
In particular, if R (u + /3) > - 1, we have

(5-21) U.+t 0) TVi0) d£ = F„+, (2x).

Suppose now that G, {p, x) is a many-valued function of x defined by

a, „ ao-
’xskp.x) r(s + 1)
_ P (S+2)+iP(y2!+ 1> _£!?_
iir r(S+3) _
Then it follows from (5'2) that if a, /3,x, y are any numbers such that ja;/y| = l,
and .R (a + /9) > 0 when | arg {xjy) i = ir and R{ a + /9) > — 1 otherwise, we have
(5-3) ^
&+{(?> «)<?,,_«(?> y){xjy)£d% = 0 or G,+f(p + q, x + y),

according as j arg (xjy) j > ir or j arg {xjy) | < ir, whatever be the branches of
G(p, x) and G(q, y), provided the branch of G(p + q, x + y) is chosen
according to our former conventions. If, in particular, R (a + /3) > — 1, then

(5-31) J“_ ff„+i (p, x) Giut (?, •) dt = GW (*> + ?, 2«).

It may be interesting to note that the right-hand sides of (5-3) and (5'31)
are of the form (p + q, z), which reduces to

r(« + /3 + l)
when p = — q, becoming independent of p or q.
The ordinary Bessel’s functions are particular cases of the function
Gs(p, x). Hence we have the following particular results. If n is real, and
R (a + /3) > 0 when n = + ir and R(u + fi)> — 1 otherwise, then
(5-4)

°r {^1%^ ejin(8-«) Ja+/) [^/(2 cos \n + y°-eVn)}\

 A Class of Definite Integrals

541) = 0 or (2*

r or < 7T. If R(a, + fi) >- 1, then

Ig+i (a) Jn-iiy) _ Ja+tWlZa?

(5’^3) J ^</«+{ («) (ie)rff - </«+£ (2®).

6. We shall now consider some special cases of i

.« r (a + a) r os - ®) r (7 + ix) r (s -

rs from (1-2) that (6'1) is equal tc

exp- {i (0 - «) (Jit - Iz) + i (8 - 7) z} dz

-K<><K. -*»<**■•
If the intervals do not overlap, the value of (6'1) is zero. It is easy to s
that if
(612) |B|>,r(l + jJj),
the intervals do not overlap; and that, if they do overlap and l> 0, then

It should also be observed that, though (611) may not be convergent for all
the values of a, yS, 7 and 8 for which (61) is convergent, yet we may evaluate
(6'11) when it is convergent, and so obtain a formula for (Gl) which may be
extended, by the theory of analytic continuation, to all values of the para¬
meters for which the integral converges.
226 A Class of Definite Integrals

From (6-12) we see that, if l and n are any real numbers such that
j!7r(! f / ), then
(6-2)
. r (a+ x) r o - r (7 + fa) r («- fa)dx °’
provided that (i) R (a + 8 + 7 + S) > 2
|n| >7r(l + !<j),
(ii) R(a + f3 + 7 + S) > 3
|n[ = ir(l + |i|).
7. Suppose now that 1 = 1 and n = 0; then (6'11) reduces to

7rr(a + /3-l)r>/ + S-l)/_w (2 00S z)*' dz,

which is easily evaluated by the help of (11). Hence

(7'x) J_„ r(« + »)ro8-*)r(7+*)r(5-*)

T (a+ 8 + 7 + 8-3)
r(«+/8-i)ro8 + 7-i)r(7 + s-i)r(8 + a-i)’
provided that (i) R(a + 8 + 7 + 8) > 3, or (ii) 2 (a — 7) and 2 (8 — 8) are odd
integers and It (a + 8 + 7 + 8) >2.
It should be noted that the formula fails when a + /3+y + 8 = 3 and
2 (a —7) and 2(8 — S) are odd integers. The value of the integral in this
case is sometimes 1/27T and sometimes —1/2-rr. The value to be selected
may be fixed as follows. It is easy to see that, in this ease, one and only one
of the numbers a + j3 — 1, 8 + 7-I, 7+8-1, and S + a-1 will be an
integer less than or equal to zero. If a + 8 — 1 or (8 + 7 — 1 happens to be
such a number, then the value of the integral is +1 /2vr, according as
2 (8 — 8) s + I (mod 4). But if 7+8 — 1 or 8 + a-1 happens to be such a
number, the value of the integral is ±1/2?r, according as 2(8 —8)= + l
(mod 4).
As particular cases of (7T), we have
” dx T (2a + 2(8-3)
(Ml) /-.{r(«+*)ro8-*)}*_{r(«+/B-i)r
provided that R (a + 8) > §,
7-12) 1“_—_:_
' ; Jo r(a + a)r(«-ffi)r((8 + a)r(/3-®)
r (2a + 2,8 - 3)
2F (2a - 1) T (28 - 1) {r (a + 8 -1)}’ ’
provided that (i) 8(a + 8)>|, or (ii) 2 (a-8) is an odd integer and
R (a + 8) > 1- If 2 (a + 8) = 3 and 2 (a — 8) is an odd integer, then the
value of the integral (712), when a ^ 1, is +1 /27t, according as 2 (a — ft) = + 1
(mod 4), and when a < 1 it is + 1/2tt, according as 2 (a - 8) = + 1 (mod 4).
 A Class of Definite Integrals 227
Putting a =/9 in (7-11) or in (7-12), we obtain
,7.13) r __r(4*-3)
( ; J0 {r(a + «)r(a-®)}*~2{r(2a-iy
if I? (a) > f. Suppose again that i = 1, « = u, and a + S = /3 +' y. Then (6-11)
reduces to

^r(«+^-i)r(7 + s-i>/„ (2sm*)a+',~!(2cos*)1,+a“2dz-

Hence we see that

(7'2) /-„r(a + a')r(/3-*)r(7 + ®)r(S-a!)

2T {-H« + £)} r {*(y+ S)}T («+s -1) -

ifa+S=/3 + 7 and R (a + ft + 7 + S) > 2. In particular

(7'21) /-.{!'(«+«) r (/s - .*)}* ^ = 2r>+13 -1) [r & («+ /j)}]*1

if R(a + @)>1, and

(7'22) /, {r (« + «) r (« - *)}■dx = 4f (2a -1) {r wr

if E(a)>i.

8. It follows from (6'2), (7‘1), and (7-2) that, if

<£ W = 2 c,„ z!", ir (z) = 2 c.,v+1

he series being convergent when | $ | = 1, then

,o.,x r tie™)
{ ’ J-.a r'(a + <r)r(^-a) r(7 + ii!)r(S-a-)“
onr(a + ff + 7 +
r(a + /3-i)r(/3+7-i)r(7 + 8-i)r(2+a-i)'
provided that (i) _B (a + /3 + 7 + 8) > 3 or (ii) iJ (a + /3 + 7 + S) > 2 and
c, e“ t3-*-*) + c-a e-" »!+8' = 2c0 cos tr(/3- S),
-i* |a+yj + c_2e»(«+r) — 2c„C0S IT (a-7);
and that

(S-2) £- ■f 0"*) -r (fa

. r(«+<)roe-a)r(7 + ir)r(S-a)u"

= 2rin5+3)}'T[iV+ 8)IT(«+ *“!)’

provided that a + $ = /3 + 7 and R(a + /3 + 7 + <S) > 2.
If a + 8 — /3 — 7 is an integer other than zero, it is possible to evaluate
the integrals (7'2) and (S'2) in finite terms, but not as a single term.
228 A Class of Definite Integrals

The following integrals can he evaluated as a single term, with the help
of (8-1) and (8*2), whenever they are convergent:
r r (8+ «)«*» d
(&3) J_„F(« + ir)r(/5-!r)r(7+»)to-
where (i) n is an odd multiple of tt, or (ii) n is lultiple of 7r and
a + l=/3 + 7 + S;
r(7 + tt)r(S±«)
(8'4) r(a+ir)rO±a;)
where (i) n is , multiple of it, or (ii) n is an odd multiple of 7r and
a+8 = /S + 7;
r (>3 + a;) r (7 - re) r (g + re) ^
(8-5)
r (a + x)
where (i) n is an odd multiple of it, or (ii) n is an even multiple of -tt and
a + /J + 7= 1 + 8;

(8-6) I r (a + ,r) 1" {(3 — xi) r (y + a) r (& — x) einx dm,

where (i) n is an even multiple of it, or (ii) n is an odd multiple of 7r and

a + 8 — ft + 7. Thus, for instance, if 8 is not real, a + l = /3+7 + S, and
It (a + + 7 - 8) > 1, then
ri? + *) __ dx
„) n3-i)f(yti)
Tre±ii*t s-y>
= T (a- 8) T {H« + £)[ T {£ (7 - 8 + 1)} ’
according as I (8) is positive or negative; and if 7 and real and
R (a + /3 - 7 - 8) > 1, then
r(7 + s)r(s + a)
■0
(8-41)
/: r(a + a0r(/3 + »)
W T(a + 0-y-8-l)
:!:sin^(7-S)^(a-7)^(a-8)^(/3-7)^(/S-8)■
the zero value being taken when I (7) and I (8) have the same sign, the
plus sign when I(7) > 0 and I (8) < 0, and the minus sign when I(y) < 0
and / (8) > 0.

9, The following results are easily obtained with the help of (6T1). If
2 (a —/3) = 7 — 8 and R(a + /8 + 7+ 8) > 3, then

(0-1) j _ „ T (a + as) T (/3 — a;) T (7 + 2x) T (8 — 2x)

2.+fi+r+t-, is-., p (g + j3 + 7 + 8 - g)>
Vwr{J (a + ft)} r(7 + S — I)T (2a + 8 - 2) '
 A Class of Definite Integrals 229

If R(a + /3) >|, then

(9'n) /o r (a+x) r («- X) r oe + 2*) r (,3 - 2*)dx

2^s-»r(«+,8-4)
■JttT (a) r (2/3 -1) r (2a + /3 - 2) ‘
If a + /3 + 7 + 8 = 4, then

(9'12) /.„ r (o+*•) r (/3 - «) r (7+L) r (8 - 2*)^

“ 2r (7 + 8 -1) r (2a + S - 2) T (2/3 + 7 - 2)'
If 2 (a - fi) = 7 - 8 + k, where k is ± 1 or + 2, then
f® sin 7r (2x + a - /3)
(9'2) j _» r (« + a-) r (/3 - a) r (7 + 2a) r (s - 2a)&
= ± g-irT (/S + 7-a+J)r(2a+8-2)’
provided that .R (a + $ + 7 + S) > 2.
If 3(a-£) = 7-S + ife, where i is + 1 or +2, then
«,.ov r sin 7r (2.r + a — ff)_
j } J . . r (a + a) r (/3 - a) r (7 + 3a') r (S - 3a) “
3j«+«-i r (2a — /S + 8 - 2)
“±47rr(7 + S-l)r(3a+S-3)’
provided that (i) R (a + R + 7 + S) > 3, or (ii) /3 + 7 - 2a is integral and
U(a + £ + 7 + S)>2.
In (9'2) and (9'3) the plus or minus sign on the right-hand side is to be
taken according as k is positive or negative. If k is an integer other than
+1 or +2, the integrals in (9‘2) and (9'3) can still be evaluated in finite
terms, but in a less simple form.
10. In this connection it may be interesting to note that, if n is an even
multiple of tr, and a + jS + 7 + B = 4, then
(10-1)

■/, -1 I'f/J ,il iv-i, ■, r (i - ,,"''

for all real values of £. The proof of this is the same as that of (T4).
Finally I may mention the formula
(10-2) f (x) (x) Jy+i (x) J^t (x) d£ = (J*)**+7+*
„T— (-l^)»-1{r(a + g + 7 + 8+2v-iy_
rwrca+zs+y+s+^r^+zs+^r^^+^r^+s+^ixs+a+v)’
which holds if (i) R(a + 0 + 7 + S)> — 1, or (ii) 2(a — 7) and 2(3-8) are
odd integers and R (a + fl + 7 + 8) > - 2.
 28
CONGRUENCE PROPERTIES OE PARTITIONS

(.Proceedings of the London Mathematical Society, 2, firm, 1020, Records for 13 March 1910)
In a paper published recently in the Proceedings of the Cambridge Philo¬
sophical Society*, I proved a number of arithmetical properties of p(n), the
number of unrestricted partitions of n, and in particular that
p (5n + 4) = 0 (mod 5),
and pOsi + 5) = 0 (mod 7).
Alternative proofs of these two theorems were found afterwards by Mr H. B. C.
Darlingt-
I have since found another method which enables me to prove all these
properties and a variety of others, of which the most striking is
p(lln + 6) = 0 (mod 11).
There are also corresponding properties in which the moduli are powers of
5, 7, or 11; thus
p (25n + 24) = 0 (mod 25),
p (49a +19), p (49n + S3), p(49» + 40), p(49;i+47) = 0 (mod 49),
p(121n +116) = 0 (mod 121).
It appears that there are no equally simple properties for any moduli in¬
volving primes other than these three.
The function t (n) defined by the equation

l t (») cfl = « {(1 - x) (1 - <*) (1 - .*■’)...}»,

also possesses very remarkable arithmetical properties. Thus

t (5n) = 0 (mod 5),
t (7n), T(7n + 3), r(7m + 5), r(7n + 6) = 0 (mod 7),
while t (23n+ r) = 0 (mod 23),
if v is any one of the numbers
5, 7, 10, 11, 14, 15, 17, 19, 20, 21, 22.
* Yol. xxx, 1919, pp. 207—210 [No. 25 of this volume : see also No. 30].
t Ibid., pp. 217, 218.
 29
ALGEBRAIC RELATIONS BETWEEN CERTAIN
INFINITE PRODUCTS
{Proceedings of the London Mathematical Society, 2, xvm, 1920, Records for 13 March 1919)
It was proved by Prof. L. J. Rogers * that

G <*> = 1 + + (1 _*) (1 _ a.) + (1 (1 — :i--) (1 - *») +-

1 1
X (1-a*)(1-«*K1

and E(x) = 1 + f=-- + ^ _ x) (1 _ ^ + (1 _ x) (1 _ ^ (i + ■-

1 w 1
(1 - a*) (1 - <*) (1 - *»)... X (1 - a*) (1 - «■) (1 - *»)
Simpler proofs were afterwards found by Prof. Rogers and myself f.
I have now found an algebraic relation between G(x) and II (x), viz.:
H (x) {G (a)}11 — x-G (x) {H{x)Y —1 + 11*10 (x) H (*)}'.
Another noteworthy formula is
H (x) G (x") - x?G (x) H (a11) = 1.
Each of these formulas is the simplest of a large class.

* Froc. London Math. Soe., Ser. 1, Tol. xxv, 1894, pp. 318—343.
t Froo. Camb. Phil. Soe., Vol. six, 1919, pp. 211—216. A short account of the history
of the theorems is given by Mr Hardy in a note attached to this paper. [For Ramanujan’s
proofs see No. 26 of this volume: those of Rogers, and the note by Hardy referred to, are
reproduced in the notes on No. 26 in the Appendix.]
 30
CONGRUENCE PROPERTIES OF PARTITIONS
(MatkcmatiscM Zeitschnft, IX, 1921, 147—153)
[Extracted from the manuscripts of the author by 0. H. Hardy*]
1. Let
(1-11) P-1-24 (j-^ + i^+CT+■•■)■

(i-i2) «=1+240(rr^+rS+S+-)’

<™>
(1-2) /(«)-(l-«)(l
Then it is veil known that
(1-3)
/ (CC) = 1 —CG + + + 2 (— l)n ($in -f (3?l+1)),

(1-4) <?-£*= 1728</X®))a4-

* Srinivasa Ramanujan, Fellow of Trinity College, Cambridge, and of the Royal Society
of London, died in India on 26 April, 1920, aged 32. The manuscript from which this
note is derived is a sequel to a short memoir “Some properties of p(n)> the number of
partitions of n” Proceedings of the Cambridge Philosophical Society, Vol. xrx (1919), 207—
210 [No. 25 of this volume]. In this memoir Ramanujan proves that
p(nn +4) = 0 (mod 5)
and yj(7n+5) = 0 (mod 7),
and states without proof a number of further congruences to moduli of the form 5n 7b 1 lc,
of which the most striking is
p(ll?i + 6) = 0 (mod 11).
Here new proofs are given of the first two congruences, and the first published proof of

The manuscript contains a large number of further results. It is very incomplete, and
will require very careful editing before it can be published in full. I have taken from it
the three simplest and most striking results, as a short but characteristic example of the
work of a man who was beyond question one of the most remarkable mathematicians of

I have adhered to Ramanujan’s notation, and followed his manuscript as closely as I

can. A few insertions of my own are marked by brackets. The most substantial of these
is in § 5, where Ramanujan’s manuscript omits the proof of (5-4). Whether I have recon¬
structed his argument correctly I cannot say.
The references given in the footnotes to “Ramanujan” are to bis memoir “On certain
arithmetical functions/' Transactions of the Cambridge Philosophical Society, Yol. xxii,
No. 9 (1916), 159-184 [No. 18 of this volume].
 Congruence Properties of Partitions 233
Further, let
(1-51) *r,s(a;) = 2 2 m’n»a>»'»= 2 n,o-_f(«)an.

where <rt (n) is the am of the &th powers of the divisors of n : that
2'a? 3W
(1-52) 3>m(*) = A
s+r^+r^+---'
aDd in particular
(1-53) P=l—24*0,,(a:), <2 = l + 240*0,.,O), jR=1-504*0,s(*).
Then [it may he deduced from the theory of the elliptic modular functions,
and has been shewn by the author in a direct and elementary manner*, that,
when r + s is odd, and r< s, *,,s(») is expressible as a polynomial in P, Q,
and JR, in the form
*, .,(»)= Sii,m,nP,QmR\
where l — 1 & Min (r, s), 21 + 4»t + 6n = )■ +s +1.
In particular fl
(1-61) Q2 = 1 + 480*0,. (x) = 1 +480 + ...),

(1-62) QR = 1 - 264*o, = (®) = 1 -264 + ..,),

(1-63) 441Q2 + 250P2 = 691 + 65520*,,„ (a)

= 691 + 65520 +

(Ill) Q —P2 = 288*0,2 (*),

(1-72) PQ-R = 720*,,4 (*),
(1-73) Q! — PR = 1008*i,6 (a),
(T74) Q(PQ-R) = 720*,, a (®),
(l'Sl) ' 3PQ -2R- P> = 1728*2,3 (*),
(1-82) P2Q-2PP + Q2 = 1728*2,5(»,
(1-83) 2PQ2 — P2JJ — QR = 1728*2,7 (*),
(1-91) 6P2Q — 8PP + SQ- — P‘= 6912*,, , (x),
(1-92) P3Q - 3P’-R + 3PQ- -QR= 3456*,,, O),
(1-93) 15PQ2 - 20P!R + 10PSQ - 4QR - P‘ = 20736*,, „ (4
* Ramanujan, p. 165 [pp. 142—143].
t Ramanujan, pp. 163—165 [pp. 141—142] (Tables I to III). Ramanujan carried the
calculation of formula! of this kind to considerable lengths, the last formula; of Table I being
7 709 321 041 217 + 32 6404>M, (*) = 764 412 173 -217(J8
+ 5 323 905 468 000^+1 621 003 400 OOOQ’-W.
It is worth while to quote one such formula; for it is impossible to understand Ramanujan
234 Congruence Properties of Partitions

Modulus 5.
2. We denote generally by J an integral power-series in x whose coefficients
are integers. It is obvious from (1'12) that
0 = 1 + 5/.
Also »5-«s0 (mod 5), and so, from (I'll) and (1'13),
a—j? + 5/,
Hence Q= - = Q (1 + 5Jf - (P + bJf = 0 - P2 + 5/
Using (1A), (1-71), and (1-51), we obtain

(2-1) 1728® (/(x))-' = 288 I no-, (n) + 5/

Also (l-®)2a = l-®K + 5/,

^ (f(x)r=f(x-->) + 5J,

(2-2) (/(*))-=£g +5/

But ^ j^-) = l + P(l)»+P(2)*2 + ..-,

and therefore, by (21) and (2-2),

(2-3)
1728^M(1+?(1)* + J(2)^+...)
= 1728® = 1728®(/(*))JJ + 5/ = 288 2 «o-, (n)xn + 5/

Multiplying by 2, rejecting multiples of 5, and replacing /(®a) by its

expansion given by (1'3), we obtain
(®-/r=«-^ + ^+...)(l+p(l)® + ^(2)^+...)

= 2 no, (n) xn + o J.

(2-4) p (fJ -1) -p (» - 26) ~p (n - 51) +p (a -126) +p (n -176)

—p (n — 301) — ... =no, (a) (mod 5),
the numbers 1, 26, 51,... being the numbers of the forms
z£-n('3n — 1) +1, (3n + 1) +1,
or, what is the same thing, of the forms
i (5n — l) (Ion - 2), \ (on +1) (157! + 2).
In particular it follows from (2'3) that
(2-5) p(5m-l) = 0 (mod 5).
 Congruence Properties of Partitions 235

Modulus 7.
3. It is obvious from (1*13) that
R = 1 + 7/.
Also ii7 — ji = 0 (mod 7), and so, from (I'll) and (1'61),
Q°-=P+7P.
Hence (Qs - R3)3 = (PQ -1 + 7J)= = P3Q* - 2PQ +1 + 7 J
= P3-2PQ + R + 7J.
But, from (1-72) and (1-81),

P3 - 2PQ + R = 144 S (on cr3 (n) -12»V, (b)) a"

= l(n^(n)-na2(n))X'' + 7J.
And therefore
(3-1) (Q3-RJ=l(,r-a;(n)-na,(n))ic’>+ 7 J.

Again (by the same argument which led to (2’2)) we have,

(3-2) (/(a0r = ^ + 7/.

Combining (3'1) and (3'2), we obtain

(3-3) = # (/(*))• + 7J = 1728!a;2 (/(*))“ + 7 J

= (Q3-R3f + 1J

= ii(«2<r1(n.)-!!lr3 («))•» + 7/.

From (3-3) it follows (just as (2-4) and (2-5) followed from (2 3)) that
(3-4) p (n - 2)-p (n - 51) - p (n- 100) +p (n-2i7) + p (n - 345)
—p (n — 590) -... = na<r, (n) - ncr3 (n) (mod 7),
the numbers 2, 51, 100,... being those of the forms
h (7n - 1) (21;i - 4), £ (7n +1) (21» + 4);
and that
(3-5) p (7m - 2) = 0 (mod 7).

Modulus 11.
*4. It is obvious from (1'62) that
(4-1) QK = 1 4- 11/.
236 Congruence Properties of Partitions

Also nu — n = 0 (mod 11), and so, from (I'll) and (1'63),

(4-2) £•• - 3BT- = 441Q3 + 250IP + 11/

= 691 + 65520 (jr^+f^S + ■••) + 11/

= — 2 + 48 (j^~ + + ... ^ +11/

= — 2P + 11/
It is easily deduced that
(4-3)
3
(Q3 - B?y = (Q - 3R1)1 - Q (<? - 3P=)3 - R (Q3 - 3 A3)3 + 6QP +11/
= Ps - 3P3Q - 4P1R + 6QR +11/
[For
(<23 - 3 IP)S - <2 (<? - 3IP)3-P(Q3- 3 JR1)"- + 6 QR
= (Q3 - 3P3)3 - Q3P= (Q> - 3S?f - Q3PJ (Q3 - 3P3)3 + 6Q«P“ + 11/
= Q'5 - 1QQ1-R- + 9SQ»IP - 285Q6P» + 423Q3P» - 243P‘° + 11/
= (<23-/.=)= +11/
by (4’1), and (AS) then follows from (4-2).]
Again, [if we multiply (1-74), (1'83), (1-92), and (1’93) by - 1, 3, -4, and
— 1, and add, we obtain, on rejecting multiples of 11,]
P= - 3P3Q - 4P'-R + 6QR = - 5<£>m + 3<fc ? + 3$s,6 - <!><,,+ 11/;
and from this and (4’3) follows
(4-4)
(Q3 - sr-f = - Z(5n<r,(») - 3»V, («) - 3nV, (») + nVj (»))*” +11/

But (by the same argument which led to (2’2) and (3'2)) we have
(4-5) (/(*))-=«+ 11/

From (4-4) and (4'5)

s5 (/(®))l2° +11/ = 1728V (/(«))“ + 11/

= (Q3-IP)3 + 11/
= - l^noy(«) - 3nV, (») - 3»VS (n) + rf<r, (a)) a" + 11/
It now follows as before that
(4'6)
p(n-5)-p(n- 126)-p (n - 247) +p (n - 610) +p (n - 852)
-p(n —1457) — u4<rj (n) + 3n3<r3 (») + 3?i!<rs (n) *
— on <r,(n) (mod 11),
 Congruence Properties of Partitions 237
5, 126, 247, ... being the numbers of the forms
i (11» - 2) (83» - 5), i (11,, + 2) (33„ + 5);
and in particular that
(4-7) p (11m- 5)= 0 (mod 11).

5. If we are only concerned to prove (4'7), it is not necessary to assume

quite so much.
Let us write S- for the operation x ^. Then* we have

(5-11) &P = ^(P2 -Q),

(5T2) $Q = i (PQ - R),
(5-13) %R=% (PR-Qf
From these equations we deduce [by straightforward calculation
864SHP = Ps - lOP’Q - 1SPQ= + 20 P'-R + 4 QR,
72 = 5 P*Q + 15PQ2 - 15P!P - 5 QR,
24— 14PQ! + tP’-R + 'iQR:
The left-hand side of each of these equations is of the form
xdJ

Multiplying by 1, 8, and 2, adding, and rejecting multiples of 11, we find

(5-2) Ps — 3P!Q + 2P’R=s:^+ IIP,

We have also, by (5T1),

eP1R-6QR = na:R~.

But, differentiating (4'2), and using (4T), we obtain

72*2?^ = 36*P (- 3O’-^ + 6R^) + UJ

= - 108*Q + 216*P2^ +11/

(5-3) 6P1K — 6QR = x~ + llj.

* Ramanujan, p. 165 [p. 142].

238 Congruence Properties of Partitions

From (5'2) and (5'3) we deduce

P5- 3P*Q-4.P*P + 6QR = * ~+ 11/,
and from (4'3)]
(5-4) (<?-i^ = *g + ll/.

Finally, from (4'5) and (5'4),

*"fjQ = *5(/W)!“ + H/= (Q* -P=)5 +11/

dJ

As the coefficient of *um on the right-hand side is a multiple of 11, (4'7)

follows immediately.
 MEMOIRS WRITTEN IN COLLABORATION WITH
G. H. HARDY (31—37)

31*
UNE FORMULE ASYMPTOTIQUE POUR LE NOMBRE
DES PARTITIONS DE n

(Compta Rendm, 2 Jan. 1917)

1. Les divers problemes de la theorie de la partition des nombres out et6
etudi^s surtout par les mathematiciens anglais, Cayley, Sylvester et Mac-
maliont, qui les ont abordes d’un point de vue purement algebrique. Ces
auteurs n’y ont fait aucune application des methodes de la theorie des
fonctions, de sorte qu’on ne trouve pas, dans la theorie en question, de
formules asymptotiques, telles qu’on en rencontre, par example, dans la
theorie des nombres premiers. II nous semble done que It's rdsultats que
nous allons faire connaitre peuvent presenter quelque nouveautA

2. Nous nous sommes occup6s surtoufc de la fonction p (n), nombre des

partitions de n. On a

-(i~=f(n)^ (!*!<1)-
Nous avons pense d’abord a faire usage de quelque thdoreme de caractere
Tauberien: on designe ainsi les theoremes reciproques du theoreme classique
d’Abel et de ses generalisations. A cette categorie appartient l’enonc6
suivant: •
Soit g(x) =.2anx’1 une serie de puissances a coefficients positifs, telle qu’on

logff
quand x tend vers un par des valours positives. Alors on a
log sn = log (a, + «! + ... + a„) ~ 2 \!{An),
quand n tend vers I'infini^.
* [For proofs of the theorems enunciated in this note see No. 36 of this volume.]
+ Voir le grand traite Combinatory Analysis de M. P. A. Macmahon (Cambridge,
1915-16).
qui doit parattre dans un autre l-ccueil. [The paper referred to is No. 34 of this volume ;
see, in particular, pp. 252—258.]
240 Une Formide Asymptotiqiie

En posant (j (x) = (1 - te)f (sc), on a

•(1)
oil e tend vers zero avec 1/n.
3. Pour pousser [’approximation plus loin, il faut recourir au theoreme
de Cauchy. Des formules

avec un chemin d’integration convenable infcerieur an cercle de rayon un, efe

= vS) \! (log i)exp (ero^))^ H (~ .(2)

(fournie par la theorie de la transformation lineaire des fonctions elliptiques),
nous avons tire, en premier lieu, la formule vraiment asymptotique

P(»0~PW = i^3^'/(*n).(3)
On a
p (10) = 42, p (20) = 627, p (50) = 204 226, p (80) = 15 796 476;
P (10) = 48, P (20) = 692, P (50) = 217 590, P (80) = 16 606 781.
Les valeurs correspoudantes de P (n) :p (n) sont
1-145: 1-104; 1-065; 1-051:

«V2 7*.t V(»-A) I

En faisant usage des formules sommatoires que d^montre M. E. Lindelof
dans son beau livre Le calcul des residus, on trouve aisement que F(x) (on
parle, il va sans dire, de la branehe principale) a pour seul point singulier
le point 3 = 1, et que la fonotion

(l0^) Me
est reguliere pour 3 = 1. On est conduit naturellement b, appliquer le
theoreme de Cauchy a la fonction / (®) — P(a:),-et l’on trouve
1 d e"Vl$ <»-*)!
 Une Formule Asymj)totiqne 241

oi 1; designe un nombre quelconque superieur a tt/VC. L’approximation, pour

des valeurs assez grandes de n, est tres bonne: on trouve, en effet,
p (Cl) = 1 121 505, p (62) = 1300156, p (63) = 1 505 499;
Q (61) = 1 121 539, Q (62) = 1 300121, Q (63) = 1 505 536.
La valeur approximative est, pour les valeurs.. suffisainment grandes de n,
alternativement en exces et en defaut.
6. On peut pousser ces calculs beauooup plus loin. On forme des func¬
tions, analogues a F {w), qui presentent, pour les valeurs

des singularity d’un type tres analogue a cedes que presente f(x). On
soustrait alors de f(x) une somme d'un nombre fini convenable de ces fonc-
tions. On trouve ainsi, par exemple,
(-1)" d
2W2 dn V(n-A)

ou k designe un nombre quelconque plus grand que Jirv'f • ’

 32
PROOF THAT ALMOST ALL NUMBERS n ARE
COMPOSED OF ABOUT log log n PRIME FACTORS
(.Proceedings of the London Mathematical Society 2, xvi, 1917, Seconds for 14 Dec. 1916)
A number n is described in popular language as a round number if it
is composed of a considerable number of comparatively small factors: thus
1200 = 2J. 3 . o2 would generally be said to be a round number. It is a matter
of common experience that round numbers are exceedingly rare. The fact
may be verified by anybody who will make a practice of factorising numbers,
such as the numbers of taxi-cabs or railway carriages, which are presented to
his attention in moments of leisure. The object of this paper* is to provide
the mathematical explanation of this phenomenon.
Let 7t, (x) denote the number of numbers which do not exceed x and are
formed of exactly v prime factors. There is an ambiguity in this definition,
for we may count multiple factors multiply or not. But the results are sub¬
stantially the same on either interpretation.
It has been proved by Landau f that

as x -*■ cc , for every fixed value of v. It is moreover obvious that

[*]-*,(*)+*■,(*) + w, (*) + ..., .(2)

and * " EP l1 + l0g l0g a + (1°J liF'-r + -}.(3>

Landau’s result shews that there is a certain correspondence between the
terms of the series (2) and (3). The correspondence is far from exact. The
first series is finite, for it is obvious that ir„ (x) = 0 if v > (log *)/(log 2); and
the second is infinite. But it is reasonable to anticipate a correspondence
accurate enough to throw considerable light on the distribution of the
numbers less than x in respect of the number of their prime factors.
The greatest term of the series (3) occurs when v is about log log x. And
if we consider the block of terms for which
log log x — <j> (x) V(Iog log x)<v< log log x + <f> (x) ^ (log log x), ...(4)
* The paper has been published in the Quarterly Journal of Mathematics, Vol. xlviii,
pp. 76-92 [No. 35 of this volume].
t See Handbuch der J,ehre von der Verteiluny dor Primiaklen, pp. 203—213.
 The Prime Factors of a Number 243

where <j> (x) is any function of x which tends to infinity with x, we find
without difficulty that it is these terms which contribute almost all the sum
of the series: the ratio of their sum to that of the remaining terms tends to
infinity with os.
In this paper we shew that the same conclusion holds for the series (2).
Let us consider all numbers n which do not exceed x, and denote by xp the
number of them which possess a property P (n, x): this property may be a
function of both n and x, or of one variable only. If then xP;x->-l when
x, we say that almost all numbers less than os possess the property P.
And if P is a function of « only, we say simply that almost all numbers
possess the property P. This being so, we prove the following theorems.
1. Almost all numbers n less than x are formed, of more than
log log x <f> (x) V(log log x)
and less than log log x + </> (at) f (log log x)
prime factors.
2. Almost all numbers n are formed of more than
log log n - ^ (a) V(log log n)
and less than log log n + <j> (ft) ./(log log n)
prime factors.
In these theorems is any function of x (or n) which tends to infinity
with its argument: and either theorem is true in whichever manner the
factors of n are counted. The only serious difficulty in the proofs lies in
replacing Landau’s asymptotic relations (1) by inequalities valid for all values
of v and x.
Since log log n tends to infinity with extreme slowness, the theorems are
fully sufficient to explain the observations which suggested them.
 33 .

ASYMPTOTIC FORMULA IN COMBINATORY

ANALYSIS
(.Proceedings of the London Mathematical Society, 2, xvi, 1917, Records for 1 March 1917)
A preliminary account of some of the contents of this paper* appeared in
the Oomptes fyendus of January 2nd, 1917. The paper contains a full dis¬
cussion and proof of the results there stated. The asymptotic formula for
p (n), the number of unrestricted partitions of n, of which only the first three
terms were given, is completed; and it is shewn -that, by taking a number of
terms of order s/n, the exact value of p(?i) can be obtained for all sufficiently
large values of n. Some account is also given of actual or possible applications
of the method used to other problems in Combinatory Analysis or the Analytic
Theory of Numbers.
[No. 36 of this volume.]
 34
ASYMPTOTIC FORMULAE FOR THE DISTRIBUTION
OF INTEGERS OF VARIOUS TYPES*
{Proceedings of the London Mathematical Society, 2, xvi, 1917, 112—132)
1. Statement of the problem.
l'l. We denote by q a number of the form
(1-11) 2®*3“*5 a‘...par,
where 2, 3, 5, ..., p are primes and
(Mil) aa^a,>as> — >aP',
and by Q (x) the number of such numbers which do not exceed x: and our
problem is that of determining the order of Q {x). We prove that

(M2) <3W = exP[!l + 0(l))5v/(j||^)], '

that is to say that to every positive e corresponds an x„ = (e), such that

(M21)

(S -o y tfe) <io®«w < (5+e) y(bfc) •

for x > x„. The function Q (x) is thus of higher order than any power of log x,
but of lower order than any power of x.
The interest of the problem is threefold. In the first place the result
itself, and the method by which it is obtained, are curious and interesting in
themselves. Secondly, the method of proof is one which, as we shew at the
end of the paper, may be applied to a whole class of problems in the analytic
theory of numbers: it enables us, for example, to find asymptotic formulae .
for the number of partitions of n into positive integers, or into different
positive integers, or into primes. Finally, the class of numbers q includes as
a sub-class the “highly composite” numbers recently studied by Mr Ramanujan
in an elaborate memoir in these Proceedingsf. The problem of determining,
with any precision, the number H (x) of highly composite numbers not
exceeding x appears to be one of. extreme difficulty. Mr Ramanujan has
proved, by elementary methods, that the order of H (x) is at any rate greater
* This paper was originally communicated under the title “A problem in the Analytic
Theory of Numbers.”
t Ramanujan, “ Highly Composite Numbers,” Proc. London Math. Soc., Ser. 2, Vol. xrv
1915, pp. 347.-409 [No. 15 of this volume].
246 Distribution of Integers of Various Types

than that of log a* : hut it is still uncertain whether or no the order of H (®)
is greater than that of any power of log x. In order to apply transcendental
methods to this problem, it would be necessary to study the properties of the
function

where h is a highly composite number, and we have not been able to make
any progress in this direction. It is therefore very desirable to study the
distribution of wider classes of numbers which include the highly composite
numbers and possess some at any rate of their characteristic properties. The
simplest and most natural such class is that of the numbers q; and here
progress is comparatively easy, since the function

, (1-13) «(»)-2p

possesses a product expression analogous to Euler’s product expression for

CM. viz.

where ln = 2.3.5 ... is the product of the first n primes.

We have not been able to apply to this problem the methods, depending
on the theory of functions of a complex variable, by which the Prime Number
Theorem was proved. The function ©(s) has the line cr = 0t as a line of
essential singularities, and we are not able to obtain sufficiently accurate
information concerning the nature of these singularities. But it is easy
enough to determine the behaviour of (@ (s) as a function of the real variable s;
and it proves sufficient for our purpose to determine an asymptotic formula
for © (s) when s -*■ 0, and then to apply a “ Tauberian ” theorem similar to
those proved by Messrs Hardy and Littlewood in a series of papers published
in these Proceedings and elsewhere^

This “ Tauberian ” theorem is in itself of considerable interest as being

(so far as we are aware) the first such theorem which deals with functions or
sequences tending to infinity more rapidly than any power of the variable.

* As great as that of

t We write as usual s=.

J See, in particular, Hardy and Littlewood, “Tauberian theorems concerning power
series and Dirichlet’s series whose coefficients are positive,” Proe. London Math. Soc.,
Ser. 2, Vol. xm, 1914, pp. 174—191; and “Some theorems concerning Diriohlet's series,’’
Messenger of Mathematics, Vol. xun, 1914, pp. 134-147.
 Distribution of Integers of Various Types 247

2. Elementary results.
21. Let us consider, before proceeding further, ivhat information con¬
cerning the order of Q {x) can be obtained by purely elementary methods.
Let
(2-11) Z„ = 2.3.5...p„ = e»<M
where ^ iso) is Tschebyschef’s function
M®) = Sjogp-
The class of numbers q is plainly identical with the class of numbers of the

(2T2) khl2h---l,S
where > 0, i2 > 0, .... bn > 0.
Now every b can be expressed in one and only one way in the form
(2-13) bi = Ci.m2m+ + ... +Ci,o,
where every c is equal to zero or to unity. We have therefore

(2-i4) ? = n (y-0 ' j = n 5^,/*= nr/, ^

say, where
(2-141) ...1^,1.
Let r denote, generally, a number of the form
(2-15) r=
where every c is zero or unity: and R (*) the number of such numbers which
do not exceed x. If q x, we have
r0^x, r2‘<®, ....
The number of possible values of r„, in the formula (2T4), cannot therefore
exceed R (x); the number of possible values of n cannot exceed R (xi); and
so on* The total number of values of q can therefore not exceed
(2-16) S (x) = R (x) R (xi) R [xi)... R (arw),
where vr is the largest number such that
(2-161) x>2**.
Thus
(2-17) Q(x)^S(x).

22. We denote by/and g the largest numbers such that

(2-211) lf^x,
(2-212) kk--lg<x-
248 Distribution of Integers of Van ous Types

It is known* (and may be proved by elementary methods) that constants

A and B exist, such that
(2-221) ^{x)^Ax (a >2),

(2-222) p„^Bnlogn (»»1).

We have therefore eAvl^x,
/log/=0(loga),
(2-23) log/= 0 (log log a);

and 1 & (2>,) < log x, ip, = 0 (log ®),

lvlogv = 0(log®), gHogg = 0 (log«),

(2-24)

But it is e sy to obtain an upper bound for B (x) i n terms of f and g. The

number ol numbers lu k.not exceeding *, is not greater than /; the
number ol products, not exceeding a, of pairs of s uch numbers, is a fortiori
not greate than if(f-1); and so on. Thus

^)</+/(471,+/(/-\Y/'=3 + ...,
where the summation need be extended to g terms only, since

A fortiori, we have
= evi°g(i+/>.
i(*)<l+/+{ + -+J<(1 +/)‘
Thus
(2-25) B (x) =.eOtei»s/) = sOMiog*iogi
by (2-23) a nd (2-24). Finally, since
log </* log log V® < i log x log °g®,
it follows from (2'16) and (2'17) that

(2-26) Q (a)=exp [^0 |(7 + i ... + jjs) V(log ® log log®)|J

2-3. A lower bound for Q (a-) may be found as ■ollows. If ^ is defined as

in 2-2, we rave
llL.,.lg<x<l,lz..Agl(}+l
* See Landau, Handbuch, pp. 79, £ 3, 214.
 Distribution of Integers of Various Types 249

It follows from the analysis of 2’2 that

lg+l = e%(l,s+i> = e0&l0srt

and k k ■ ■ ■ l, = exp If & (p,)l = eOto=ioM).

Thus %<eom°s<,).
which is only possible if # is greater than a constant positive multiple of

Now the numbers l„ 4, ..., lg can be combined in 20 different ways, and

each such combination gives a number q not greater than x. Thus

(MI) «<*»»> «=n.{V(r.?iii#

where K is a positive constant. From (2’26) and (2‘31) it follows that there
are positive constants K and L such that

(2-32) Kj< log Q («) < L V(log * log log a-).

The inequalities (2'32) give a fairly accurate idea as to the order of

magnitude of Q (x). But they are much less precise than \he inequalities
(T121). To obtain these requires the use of less elementary methods.

3. The behaviour o/©(s) when's-* 0 by positive values.

31. From the fact, already used in 21, that the class of numbers q is
identical with the class of numbers of the form (212), it follows at once that

(i-14) aw-xl-nfj-y.
Both series and product are absolutely convergent for a > 0, and
(311) log® (a) = 4> (s) + i<j, (2s) + (3s) + ...,

(3111) <!>(s) = hn~\

We have also
(312)
$(«) =
250 Distribution of Integers of Various Types

3'2. Lemma.—If x > 1, s > 0, then

(3'21)
Write Xs = e“: then we have to prove that

(3-22) 11 1
42 12 (e“ —l)2 u2
for all positive values oi >r (writing w for %u) that

<3'23> i-5<S®5<i .
for all positive values of w. But it is easy to prove that the function

is a steadily decreasing function of w, and that its limit when w-a-0

and this establishes the truth of the lemma.
3'3. We have therefore

(3-31) f (s) = ~ - 0, (s) = -j~ - 0. M + 0(1).

1 [ 1 s=) dse . . 1 e~’*m dx

From the second of these inequalities, and (2'221), it follows that

1 dsc e-*As
0 .(»)< (logic)2 x slog2 J2 logx

= s-l42-<S£& + 0(1);
and so that
(3'32) 4>(s)>A J”j^ck+0( 1).

On the other hand there is a positive constant B, such that

»(•)<& (*>2)*.

slog2 ^/2 logx^a

fs ^ dm+ 0(1).
* Landau, Eandbuch, loc. cit.
 Distribution of Integers of Various Types 251

3'4 Lemma.—If E is any positive number, then

when 0.
Given any positive number e, we can choose f and X, so that

jeEe~B*di, < e, f“E<rs*cte < e.

N„,,

= jl (») +f («) +.;3 («) +74 («),

say. And we have

0 < i* W < = 0 K* log (1/*)} = o (1),

0 < f (s) < 2 j*E<rBudu < 2e,

js («) = [ Ee~Budu + o (1),

0 <y'j (s)<J Ee~Budu < e;

11 - s log ^ J (s) j = J ^ Ee~Budu - j, (s) - j2 (a) -7, («) - ji («) J

< ne + 0 (1) < 6e,

for all sufficiently small values of s,
3'5. From (3'32), (3'33), and the lemma just proved, it follows that
(3-51) ^(s) = 24-s^__i__.

From this formula we can deduce an asymptotic formula for log <© (s). We
choose N so that

(3-53) log<©(s)=S-<i(«s)= S + 1 + 2 + 2
. n JVOXl/Vs l/VsSn^l/s 1 /SOI
= <1>, (s) + $2 (s) + $3 (s) + $4 (s),

1 + 0(1)
'slog (1/s) 7«A
252 Distribution of Integers of Various Types

In the second place

l«s log (1/ns) j ’

log (1/ns) >Jiog (1/s),
< 1/Vs. It follows that a constant K exists such that

Slog (1/S) slog (1/s)'

n $1 (s), and a constant L exists such that

Thus

(3'543) —
for all sufficiently small values of s.
Finally, in 5>4 (s) we 1 ns > 1, and a constant M e;
<j, (ns) < M2~
Thus
(3-544) <J>/(s)< M ~<sM^S 2_ns < = 0 (1).

From (3’53), (3'541)—(3 544), and (3’52) it follows that

<3'55>

0
+ lvslog(l/s)l + 0(1)’
where |p|<6, |p'!<Ke.
Thus
(3-56)
10S®(S)~6TOT)’
a(s) = exp ["{1 + 0(1)}—^-

4. A Tauberian theorem.
4‘1. The passage from (3’57) to (1-12) depends upon a theorem of the
' Tauberian ” type.
Theorem A. Suppose that
(1) X^O, Xa>Xn_1, X„-»-co;
(2) Viw^i;
(3) o„>0;
 Distribution of Integers of Various Types 253

(4) ^>0, «> 0;

(5) 2a„e-V is convergent for s > 0;
(6) /(«) = 2a»e-*«» = exp |jl + o (1)} A s— jlog (i) j
when 8-f 0. Then

where B= «-'(•+*) (1 + ajH-ts/iwoU

when n-> so.
We are given that
(4-11) (l-S)jls-(log^ * < log/(s) <(1 + 8) 4 s-* (log -
for every positive S and all sufficiently small values of s; a

(4-12)
(1 - e) BK‘/°+al (logX„)-«»+*' < log An
< (1 + v) (log?L,l)_Wl,+“l.
for every positive e and all sufficiently large values of n.
In the argument which follows we shall be dealing with three variables,
8, s, and n (or m), the two latter variables being connected by an equation or
by inequalities, and with an auxiliary parameter £. We shall use the letter y,
without a suffix, to denote generally a function of S, s, and n (or m)*, which
is not the same in different formulas, but in all cases tends to zero when S
and s tend to zero and n (or m) to infinity; so that, given any positive e, we

0 < i VI < «,
for 0<8<B„, 0<s<s„, ?i>«0.
We shall use the symbol ys to denote a function of f only which tends to
zero with £ so that
0<|Ufl<«.
if J is small enough. It is to be understood that the choice of a f to satisfy
certain conditions is in all cases prior to that of 8, s, and n (or m). Finally,
we use the letters H,K,... to denote positive numbers independent of these
variables and of J.
The second of the inequalities (4'12) is very easily proved. For
(4-131) . Jdne-V<ffl1e-V+a2e-iss + ... + a„e-V

</(*)< “P |(l + s)-^*-‘(log*j,

(4-132) 4„<exp%,

(4-1321) x = (l + S)zls-(log|)"\xn«.
254 Distribution of Integers of Various Types

We can choose a value of s, corresponding to every large value of n, such

that
(4-14) (1 - 8)4 as"1- (log-J) ? < X„ < (1 4- 8) A as-1-" (log i) \

From these inequalities we deduce, by an elementary process of approximation,

(4'151) (1 - V) (4 «)-/«+•' X,,1/'1-' (log ^)S/‘1+ ’

< i < (1 + V) a)-1/‘1+*>X,1'l,+“l (log ly'"**',

(4-152) ^ log X„ < log i < log X„,

(4-153) (1 - v) ~-a V"1,'(1+‘l (log

< s < (i + V) jf~aXn-,/,1+“' (log K)~^,

(4-154) %<(!+»/) B\n*K'+°' (log X„)-^/'1+“'.

We have therefore
(4-16) log An < (1 + e) BX„“/l1+“> (logX„)-W(>+“>/
for every positive e and all sufficiently large values of n*.

4-2. We have
(4-21) - /(s) =2a,e-A*‘ = Sd„(e-V-rhu*)

= sl An f*”+V“d»=sI"a (<s) e~**dx,

where <SA (x) is the discontinuous function defined by

&t{x)—An (Xn ^ x < Xn-!.1) -J-,
so that, by (416),
(4-22) log &t (a) < (1 + e) S^/o+ai (log a,)-(i/ff+.)
for every positive e and all sufficiently large values of x. We have therefore

(4-23) exp j(l - 8) As- (log < s f*£4 (*) e-^dx

<exp|(l + 8)2L«-(log^ *}
for every positive 8 and all sufficiently small values of s.
* We use the second inequality (4-12) in the proof of the first. It would be sufficient
for our purpose to begin by proving a result cruder than (4-16), with any constant K on
the right-hand side instead of (1 + e) B. But it is equally easy to obtain the more precise
inequality. Compare the argument in the second of the two papers by Hardy and Little-
wood quoted on p. 114 [p. 246 of this volume] (pp. 143 el seg.).
+ Compare Hardy and Kiesz, “ The General Theory of Diricblet’s Series,” Cambridge
Tracts m Mathematics, No. 18, 1915, p. 24.
 Distribution of Integers of Various Types 255

We define \x, a steadily increasing and continuous function of the con¬

tinuous variable x, by the equation
= (*- n) (U - K) (n^x^n + l).
We can then choose m so that

(4-24) j= A-mVfl+“> (log >Os/il+“’ •

We shall now shew that the limits of the integral in (4'23) may be replaced
by (1 - f) and (1 + f) ~Km, where f is an arbitrary positive number less
than unity.
We write
(4-25)

= 1+1+1 +1+1,
where H is a constant, in any case greater than 1, and large enough to satisfy
certain further conditions which will appear in a moment; and we proceed to
shew that «/„ J„ JA, and Js are negligible in comparison with the exponentials
which occur in (4'23), and so in comparison with Jt.
4'3. The integrals and J, are easily disposed of. In the first place we

(4-31) J, = s jyUs4(x)e->Hx<M(^)

< exp {(1 + 8) B gp)^’ (l°g Jf)~',/'l+a’} -

by (4'22)*. It will be found, by a straightforward calculation, that this

(4-32) exp j(l 4- v) A- (1 + a) s- (log j) ,

and is therefore certainly negligible if H is sufficiently large.

Thus Ji is negligible. To prove that J, is negligible we prove first that
sx > 4£a“/i1+«i (log*)-W1+“,>
if x >H\m and H is large enough f. It follows that

Js = sj M(x) e-‘xdx <sj exp {(1 + S) (log *)-^/i1+“i -sa.-) dx

<sj^ e-i‘xdx = i,
and is therefore negligible.
* With S in the place of e.
f We suppress the details of the calculation, which is quite straightforward.
256 Distribution of Integers of Various Types

44. The integrals Ja and J., may be discu ssed in practically7 the same
way, and \re may confine ourselv<is to the latter.
We ha ve
(4-41) rd(x) e~‘xdx<s
:C'Le*dx’
(4-411) f = (1 + 8) i (loga:)-*/
The maximum of the function occurs for x = *v„, where

(4-42) t?ic0‘/iI+“)(log:
h{l+v)h
From this equation, and (4'24), iti plainly resultsi that
(443) (1 — ij)) < «„ < (1 + v)
and that < falls (when S and s are small enoiugh) between (1 -- f) X,a and
(1 + f) x»
Let us write x = xa+ %
in Jf Th en
+ M-*W + *(i + w (log xJ-m+%

where x„< xl<x and a fortiori

(1 -£)\m<xl<
It follows that

(4-44) log*.)"*' ,|1+“'-2(logX,„)-W|1+

On the other hand, an easy calculation shews tb>at

(4-45) {l-flA s-* ^log ^1 '< + W<(H-r;)4s-“(logi) ^

Thus
(4-46) /1<exp|(l + ij)^.s-
•wn '(logU-*/'^>)^
X/(t_, |A,„eXpl
< exp |(l + r;)4s- •(iogj) ,/t^}

< exp |(1 + r)-N,n^(iogJ)

1-
Since ?is independent of 8 and is, this inequality shews that Jt is negligible;
and a similar argument may be iapplied to Jt.
 Distribution of Integers of Various Types 257

4'5. We may therefore replace the inequalities (4'23) by

(4-51) exp j(l - 8) (log *j

<sli + }•
Since (») is a steadily increasing function of x, it follows that
(4-521)
exp ((1 -8) As-° (logij + e~’x^

(4-522) ^ n+|
exp ((1 + S) A s- (log |> {(1 - f) X*,} f-’*dx;

°r (4-531)
(«***-»- «-**,») ^ ((1 - f) xj < exp |(1 + S) ^ (logsj,
(4-532)
(e<*m - e-i**,) s4 ((1 + K) \») > exp |(1 - S) A *- (log + Xm al.

If we substitute for s, in terms of \m, in the right-hand sides*of (4'531) and

(4-532), we obtain expressions of the form
exp {(1 + 7}) B\m‘la+a> (logXj-?/11-*1).
On the other hand
is of the form exp (log Xm)^114*’).
We hare thus
(4-541)
A !(1 - ?) 5t».} < exp {(1 + V{ + v) # V/|1+<0 (log
(4-542)
A j(l + ?) Xm) > exp ((I - ^ - 7i) BXnfw (log Xm)-W»+“)),
Now let v be any number such that
(4-55) (l-?)Xra«X,<(l+ ?)*».
Since XnjX^-7-1, it is clear that all numbers n from a certain point onwards
will fall among the numbers v. It follows from (4-541) and (4-542) that
(4-56)
exp {(1 -V{-v) (1 -vs) (logXj-'V'H*)} < A (\„)
< exp ((1 + vs + v) (1 + vs) Ww (log \„)-*7B+“>l;
and therefore that, given e, we can choose first f and then ?i„ so that
(4-57) exp ((1 - e) Bxjl^ (logX»)-^'"+“'( < A (K)
< exp ((1 + e) BXn“/'I+") (log X„)-?/tl+°>},
for n >n0. This completes the proof of the theorem.
 258 Distribution of Integers of Various Types

4-6. There is of course a corresponding “Abelian” theorem, which we

content ourselves with enunciating. This theorem is naturally not limited
by the restriction that the coefficients an are positive.
Theorem B. Suppose that
(1) \^0, Xn>X,M, X„-*ao;
(2)
(3) A > 0, 0 < a < 1;
(4) An = a, + as + ... + an = exp [{1 + o (1)) A V (log X.H].
when n~f-x . Then the series Sa„e“Ans is convergent for s > 0, and
/(s)=2a„e-V = expj^{l+o(l)}5s-V«-) ^log^j 'J ,
where B = A«-/<:-*> (1 - a)i+Wn-«)],
when s-> 0.
The proof of this theorem, which is naturally easier than that of the
correlative Tauberian theorem, should present no difficulty to anyone who
has followed the analysis which precedes.
4'7. The simplest and most interesting cases of Theorems A and B are
those in which

It is then convenient to write x for e~‘. We thus obtain

Theorem C. If A > 0, 0 < a < 1, and
log A„ = log (a, + a, + ... + a„) ~ An1,
then the series Sanis* is convergent for \ x | < 1, and
log/(*) = l"g (2an®’‘) ~ B (1 -
where B = (1 - a) A'/n~“>,
when x—>*1 by real values.
If the coefficients are positive the converse inference is also correct. That is
to say, if
A > 0, a > 0,
and log/ (x) ~ A (1 — a)-*,
then ' logA„~.fin‘‘/f'+<*>,
where B = (1 + a) ora/|1+** A1/114*1.

5. Application to our problem, and to other problems in the

Theory of Numbers.
51. We proved in 3 that
 Distribution of Integers of Various Types 259

In Theorem A take
= log n, A=~, « = 1, £ = 1.

Then all the conditions of the theorem are satisfied. And A„ is Q (n), the
number of numbers q not exceeding n. We have therefore

(i-ni

(s«)
52. The method which we have followed in solving this problem is one
capable of many other interesting applications.
Suppose, for example, that Rr (n) is the number of ways in which n can
be represented as the sum of any number of rth powers of positive integers*.
We shall prove that
(5-21) log Rr (n) ~ (r + 1) {£ T g + l) ?g +

In particular, if P (n) (n) is the number of partitions of n, then

(5-22) log P(n)~vr ^/(y)-

We need only sketch the proof, which is in principle similar to the main
proof of this paper. We have

(5-23) /(») - 2 [Rr (*) - Rr (# -1)) «— = 5

It is obvious that Rr (n) increases with n and that all the coefficients in
/ (s) are positive. Again,

(5-24) log/(a) = | log = 2 (<r!"r + ^r +...)

(5-241) <j>(s) = 2e-‘'’r.

* Thus 28 = 33+l3=3.2a + 4.13=2.23 + 12.13= 23 + 20.13= 28.13:

and 2?3 (28)=5.
The order of the powers is supposed to be indifferent, so that (e.g.) 33+l3
not reckoned as separate representations,
t Rt (0) is to be interpreted as zero.
 Distribution of Integers of Various Types

But
(5-25) *M~rg + l)r^,
when s-s-O; and we can deduce, by an argument similar to that of 3'5, that

(5-26) log/M~r(i + l)r(i + l)ar->*

We now obtain (5'21) by an application of Theorem A, taking

K = n, « =i /3 = 0, A = r^ + l)?0 + l).

In a similar manner we can shew that, if S (n) is the number of»partitions

of n into different positive integers, so that
(n) e~»‘ = (1 + er>) (1 + <r“) (1 + <r») + ...
1
-(!-«-) (l-r*)(l-«r-)...*

(5-27) logS(n)~7r^/^;

that if Tr (■n) ip the number of representations of n as the sum of rth powers

of primes, then
(5-2S)
log 1\ (n) ~ (r + 1) |r g + 2) f g + l)Jr/,’'+1’ BV*+» (log*i)-’7<--«>;
and, in particular, that if T (n) = T, (n) is the number of partitic of n into
primes, then

Finally, w can shew that if r and s are positive integers, . >0, and
0 ^ b $ 1, and
{(1 + ax) (1 + ax?) (1 + ax?).. .jr
(5-291) 2 £(11) a’1.
{(1 — 6a;) (1-fcc2) (1 -bst?) ...}s’

(5-292) log <j>{n)~~2 VM,

where
(5-2921) 0 = r jai°g + t) dt-sj6-°g.~^dt.

In particular, if a = 1, b = 1, and r = s, we have

(5-293) 2$ (n) xn = (1 — 2x + 2s? +;...)-r»
(5-294) log cj> (n) ~ it V(™).
[Added March 2&th, 1917.—Since this paper was written M. Q. Valiron
(“Sur la croissance du module maximum des series entieres,” Bulletin de la
 Distribution of Integers of Various Types 261

Societe matMmatique de France, Vol. xliv, 1916, pp. 45—64) has published a
number of very interesting theorems concerning power-series which are more
or less directly related to ours. M. Valiron considers power-series only, and
his point of view is different from ours, in some respects more restricted and
in others more general.
He proves in particular that the necessary and sufficient conditions that

where M (r) is the maximum modulus of f{x) = 2 «„«“ for \ x | = r, are that

log|«,|<(l + e) (l+«MV»«g)*,M*'
for n > ?j0 (e), and
log [ an! > (1 - ep) (1 + «) 4-/M*)

for n = np {p = 1, 2, 3, ...), where np+fnr-*-1 and e?-»-0 as p-*-x,.

M. Valiron refers to previous, but less general or less precise, results
given by Borel (Leqons sur les series d termes positifs, 1902, Ch. v) and by
Wiman (“Uber dem Zusammenhang zwischen dem Maximal-betrage einer
analytischen Funktion und dem grossten Gliede der zugehorigen Taylor’schen
Reihe,” Acta Mathematica, Vol. xxxvii, 1914, pp. 305—326). We may add a
reference to Le Roy, “Valeurs asymptotiques de certaines series procddant
suivant les puissances entires et positives d’une variable reelle,” Bulletin des
sciences matMmatiques, Ser. 2, Vol. xxiv, 1900, pp. 245—268.
We have more recently obtained results concerning P (»), the number of
partitions of n, far more precise than (5'22). A preliminary account of these
researches has appeared, under the title “ Une formule asymptotique pour le
nombre des partitions de n,” in the Gomptes Rendus of January 2nd, 1917*;
and a fuller account has been presented to the Society. See Records of
Proceedings at Meetings, March 1st, 1917 f.]
 35
THE NORMAL NUMBER OF PRIME FACTORS
OF A NUMBER n
(Quarterly Journal of Mathematics, xlviii, 1917, 76—92)
I.
Statement of the problem.
11. The problem with which we are concerned in this paper may be
stated roughly as follows: What is the normal degree of compositeness of a
number n ? We shall prove a number of theorems the general result of which
is to shew that it is, as a rule, composed of about log log n factors.
These statements are vague, and we must define our problem more pre¬
cisely before we proceed further.
T2. There are two ways in which it is natural to measure the " degree of
compositeness ” of n, viz.
(1) by its number of divisors,
(2) by its number of prime factors.
In this paper we adopt the second point of view. A distinction arises
according as multiple factors are or are not counted multiply. We shall
denote the number of different prime factors by/(«), and the total number of .
prime factors by F(n), so that (e.g.)
/(23325) = 3, F (2S325) = 6.
With regard to these functions (or any other arithmetical functions of n)
four questions naturally suggest themselves.
(1) In the first place we may ask what is the minimum order of the
function considered. We wish to determine an elementary function of n,
with as low a rate of increase as possible, such that (e.g.)
/(»)<*(»)
for an infinity of values of n. This question is, for the functions now under
consideration, trivial; for it is plain that
/(«)-*(») = 1
when n is a prime.
(2) Secondly, we may ask what is the maximum order of the function.
This question also is trivial for F (n): we have
 Normal member of Prime Factors of a number n 263

whenever n is of the form 2,!; and there is no <j> (m) of slower increase which
satisfies the conditions. The answer is less obvious in the case off(n): but,
supposing n to be .the product of the first h primes, we can shew (by purely
elementary reasoning) that, if e is any positive number, we have

for all sufficiently large values of n, and

for an infinity of values; so that the maximum order off(n) is

log"
log log n
It is worth mentioning, for the sake of comparison, that the minimum
order of d (n), the number of divisors of n. is 2. while the maximum order
lies between , l • ■ j •

for every positive value of e*.

(8) Thirdly, we may ask what is the average order of the function. It is
well known—to return for a moment to the theory of d (n)—that
(1-21) d (1) + d (2) + ... + d (n) ~ n log n.
This result, indeed, is almost trivial, and far more is known; it is known in
fact that
(1-22) d (1) + d (2) + ... + d (n) = n log n+ (2y - 1) n + 0 (n* log n),
this result being one of the deepest in the analytic theory of numbers. It
would be natural, then, to say that the average order of d (n) is log n.
A similar problem presents itself for /(») and F (n)\ and it is easy to
shew that, in this sense, the average order of f(n) and F(n) is log log n. In
fact it inay be shewn, by purely elementary methods, that
(1*23) / (1) +/ (2) +... +/ (n) = n log log n + An + O (j^j ,

(1-24) .F (1) +1? (2) + ... + If (») = n log log n + Bn+ 0 (j—^) ,
where A and B are certain constants.
This problem, however, we shall dismiss for the present, as results still
more precise than (1*23) and (124) can be found by transcendental methods.
(4) Fourthly, we may ask what is the normal order of the function.
This phrase requires a little more explanation.
* Wigort, “Sur l’ordre de grandeur du nombre des diviseurs d’un entier,” Arkiv far
matmatik, Vol. in (1907), No. 18, pp. 1- 9. SeeBamanujan, “Highly Composite Numbers,”
Proc. London Math. jSoc., Ser. 2, Vol. xiv (1916), pp. 347—409 [No. 15 of this volume], for
more precise* results.
264 Normal number of Prime Factors of a number n

Suppose that Ff (x) is the number of numbers, not exceeding x, which

possess a certain property P. This property may be a function of re only, or
of x only, or of both n and x: we shall be concerned only with cases in which
it is a function of one variable alone. And suppose further that
(1-25) N(x)~x
when x —► oo. Then we shall say, if P is a function of n only, that almost all
numbers possess the property, and, if P is a function of x only, that almost
all numbers less than x possess the property. Thus, to take a trivial example,
almost all numbers are composite.
If then g (n) is an arithmetical function of n, and (re) an elementary
increasing function, and if, for every positive e, we have
(1-26) (1 - c) <Mn) <</(«)< (I+ e)<M»)
for almost all values of n, we shall say that the normal order of g (n) is
<£(«)•
It is in no way necessary that a function should possess either a deter¬
minate average order or a determinate normal order, or that one should be
determinate when the other is, or that, if both are determinate, they should
be the same. But we shall find that each of the functions / (n) and F (re.) has
both the average order and the normal order log log re.
The definitions just given may be modified so as to apply only to numbers
of a special class. If Q (x) is the number of numbers of the class not exceed¬
ing x, and JST (x) the number of these numbers which possess the property P,

(1-27) N(x)~Q{x),
then we shall say that almost all of the numbers (or almost all of the
numbers less than a) possess the property.

II.
“ Quadratfrei ” numbers.
2'1. It is most convenient to begin by confining our attention to
quadratfrei numbers, numbers, that is to say, which contain no prime factor
raised to a power higher than the first. It is well known that the number
Q (x) of such numbers, not exceeding x, satisfies the relation
(2-11)

We shall denote by tr„ (x) the number of numbers which are products of
just v different prime factors and do not exceed x, so that (x) = it (x) is
the number of primes not exceeding x, and

Sir, («)=«(«).
* Landau, Handbuch, p. 581.
 Normal number of Prime Factors of a number n 265

It has been shewn by Landau * that

(2-12)

for every fixed value of v. In what follows we shall require, not an asymp¬
totic equality, but an inequality satisfied for all values of v and x.
22. Lemma A. There are absolute constants G and K such that
(2-21) 7r„+lW<j^(l°glp + 6')“

for >< = 0,1,2,..., ®>2.

The inequality (2'21) is certainly true when v - 0, whatever the value of
G. It is known (and may be proved by elementary methods f) that
(2-221) 2 i<loglogx + B

(2-222) 2 1-Sp< H log x,

where B and H are constants. We shall prove by induction that (2'2I) is
true if
(2-223) C>B + H.
Consider the numbers which do not exceed x and are comprised in the

2 .p, ,p2...p„,
3 .p^p^.-p,,
5 .Pl.p2...py,

P-Pi.pt-P„
where p^pt, ..., p„ are, in each row, different primes arranged in ascending
order of magnitude, and where p, > 2 in the first row, p, > 3 in the second,
5 in the third, and so on. It is plain that The total number of
numbers in the table is plainly not greater than

If now <»,, (Ba, ... are primes and

the number <b,cs.,... co^ will occur at least v times in the table,
row in which the first figure is to,, once in that in which it is <oa
that in which it is to,. We have therefore
(2-23)

Landau, ffandbuch, pp. 203 et seq.

In applying (2-21) we assume that xjp ^ 2. If xjp < 2, .(*/*>) is
266 Normal number of Prime Factors of a number n

s true when v is replaced by v -1, v

^{loglog g)+c}«

. ‘logP (i + }2I2 _1_ | 21ogp

(2-25) .-L-
log * — logp log® (log®)2 i log® log X (log®)2 ’
since log p < J log ®; and
(2-26) 2 1_.. 1 ^ I ^ 2 2 logp
' 1,‘SxP log (®/p) " log Xp,Ssp (logX}‘ p„! s ,
< log log X + B H < log log X + 0
log® +log® log®
Substituting in (2-24), we obtain (2-21).

2'3. We can now prove one of our main theorems.

Theorem A. Suppose that <f> is a function of x which tends steadily to
infinity with x. Fhen
(2'31) log log®- cj> </(logIog®) < f(n) < loglog® + <£y'(l°gl°g;c)
for almost all qnadratfrei numbers n less than x.
It is plainly enough to prove that
(2-321) &= % tr„+1(®) = o(®),

(2-322) 8,~ 2 w„+I (®) = o(®)f.

v>llX+4,J[.llx)
It will be sufficient to consider one of the two sums Sls S.2, say the latter;
the discussion of S2 proceeds on the same lines and is a little simpler. We
have
(2-33) - x> 2 (Ux + CY
log® *>7**+**/{W») v•
Write . log log®+ <7=!.
Then the condition that
r > log log ® + ^(log log ®),
where tp is some function of ® which tends steadily to infinity with ®, is
plainly equivalent to the condition that

* It may be well to observe explicitly that log log 2 + 0 is positive,

t We shall sometimes write lx, llx, ..., instead of log®, log log®, ..., in order to
shorten our formulae.
 Normal number of Pr ime Factors of a number n 267

fhere i/r is some function of £ which tends steadily to infinity with f; and
o what we have to prove is that
(2-34) S= £ = o(et).
We choose a positive number S so small that
$ S2 S3
W 0 + .0 + 4T6 + -<-

t
S-~S' + S",
(2-36) S={+Wf<S<(1+,
say. In the first place we have
(2-371) 1+ p
+1 (vi+ 1) (i'i + 2)
_1
<Plh 4. — + -1_ + l-l+It
h! C 1 + 8^ (1+8)^ ■■■[ S vxV
(1 +
where v, is the si llest integer greater than (1 + 8) f. It follows that
(2-372) ' S" < e»ilI°st-Iog-,+l)
8
K
avr
: absolute constants and
A = (l+B)log(l + S)-8.
And si
(1 + 8) log (1 + 8) - 8 > (1 + 8) (5 — 4 S-) — S = J S= (1 — S) =
say, we obtain
(2-373)
a vr
where t) is positive: so that
(2-374) S" = o(a«).
In S', we write v = ? + /+ so that ^ Vf < /i < Sf. Then

|®81» frerrsi
< exp {(? + /*) log f - (f + /a) log (f + 1 + /i}

= |exp{(? + ^)(1-| + ^-..

* Since the exponent ia equal to

Knf

'(2-35).
268 Normal number of Prime Factors of a number n

Thus

(2*382) S'-°(mkrmy
-»{AC,
= 0-(etj e~“’ duj = o (ef),

since -vjr =o. From (2 36), (2-374), and (2-382) follows the truth of (2-34),
and so that of (2'322). As (2-321) may he proved in the same manner, the
proof of Theorem A is completed.
2'4. Theorem A'. If $ is a function of re which tends steadily to infinity
with n, then almost all quadratfrei numbers have between
log log n - <f> /(log log re) and log log n + $ /(log log «)
prime factors.
This theorem is a simple corollary of Theorem A. Consider the numbers
not exceeding x. We may plainly neglect numbers less than /*; so that
| log « ^ log re «log as,
(2-41) , log log x - log 2 ^ log log re< log log x.
Given <f>, we can determine a function -<|r (x) such that tends steadily to
infinity with x and
(2-42) *(*)< l'MV’4
Then - ^M<i#(re) (fx^n^x).
In the first place
4> V(l°g l°g re) > f -/(log log x)
if log log re > J log log®,
which is certainly true, in virtue of (2-41), for sufficiently large values of x.
Thus
(2-43) log log n-4> /(log log re) < log log x - -f /(log log x).
In the second place the corresponding inequality
(244) log log re. + 0 /(log log re) > log log a: + i|r / (log log x)
is certainly true if
<j> /(log log re) > -f /(log log x) + log 2;
Knd this also is true, in virtue of (2’41) and (2'42), for sufficiently large
values of x. From (2-43), (2-44), and Theorem A, Theorem A' follows at

As a corollary we have
Theorem A". The normal order of the number of prime factors of a
quadratfrei number is log log re.
* In this sum the values of p. are not, in general, integral: £+/i is integral.
 Normal number of Prime Factors of a number n 269

III.
The normal order off (n).
31. So far we have confined our attention to numbers which have no
repeated factors. When we remove this restriction, the functions /(«) and
F (n) have to be distinguished from one another.
We shall denote by (*) the number of numbers, not exceeding x, for

WhlCh /(»)=»■
It is obvious that ra-„ (x) > ir„ (x).
We require an inequality for vr„ (x) similar to that for irv [x) given by
Lemma A.
Lemma B. There are absolute constants D and L such that
Lx (log log i -Df
(3-11) i (*) < log* v\
for >' = 0,1,2. *>2.
It is plain that
*> (®) = "■ (*) + * W*) -Mr «) + ■■■ = 0 .
The inequality is therefore true for v =0, whatever the value of D. We
shall prove by induction that it is true in general if
(3T2) D > B + II + /,
where B and S have the same values as in 2-2, and

(3-13) X (s + 1) (2~s + 3- + 5-' + ...).

Consider the numbers which do not exceed x and are comprised in the

3“ . pi11. p2°3 - - - pTv,

P°. ...pA,
where pu pit ...,p„ satisfy the same conditions as in the table of 2-2. It is
plain that P-^fx if a—l,P^$x if a = 2, and so on, so that the total
number of numbers in the table does not exceed
2

If]

the number of the tables

270 Normal number of Prime Factors of a number n

which correspond to different values of a. We have therefore

(3-14) + +

Nowjlet us suppose that (3-11) is true when v — 1 is substituted for v.

Then it is plain* that
(3'15)
• (log log x + DY 1 [ 2 1 , ? 1 ,
'+1 U v! 1,4 * V log (*/p) s’s * f log («/i>2)
Now
log log x + B + H
(3-16)
log®
as we have already seen in 2'2. Also, if ps+I <®, we have

(317) 2 ■■■ ■ ■■ ~ f < f— (2- + 3-’ + 5- + ...),

„-««*p!log(®/p‘) log»v
if s 2. Hence ,

2 ^ 1 log log® + ,B + .H'+ / log log® + -P

» „■««*?>* log log® log®
From (3T5) and (3T8) the truth of (3-11) follows immediately.

3'2. We can now argue with sr„ (®) as we argued with it, (®) in 2'3 and
the paragraphs which follow; and we may, without further preface, state the
following theorems.
Theorem B. If ip is a function of ® which tends steadily to infinity to x,
then
log log x - <j> V(log log ®) </ (n) < log log x + 4> V(log log x)
for almost all numbers n less than x.
Theorem B\ If cf> is a function of n which tends steadily to infinity with
n, then almost all numbers have between
log log »- v^loglogn) and log log n+V(log log n)
different prime factors.
Theorem B". The normal order of the number of different prime factors
of a number is log log n.
i p. 61. [Foe p. 266.]
 Normal number of Prime Factors of a number n 271

IT.
The normal order of F (n),
41. We lave now to consider the corresponding theorems for F{n), the
number of prime factors of n when multiple factors are counted multiply.
These theorems are slightly more difficult. The additional difficulty occurs,
however, only in the first stage of the argument, which requires the proof of
some inequality analogous to those given by Lemmas A and B.
We denote by II„ (as)
the number of numbers, not exceeding sc, for which
F(n) = v.
It would be natural to expect an inequality of the same form as those of
Lemmas A and B, though naturally with different values of the constants.
It is easy to see, however, that no such inequality can possibly be true.
For, if v is greater than a constant multiple of logic, the function
x (log log x+ oy
log® v1 *
js—as may be seen at once by a simple approximation based upon Stirling’s
theorem—exceedingly small; and II„+: (x), being an integer, cannot be small
unless it is zero. Thus such an inequality as is suggested would shew that
F (n) cannot be of order as high as log x for any n less than a>; and this is
false, as we can see by taking
® = 2* + l, » = 2*. F(n)=j^p.

The inequality required must therefore be of a less simple character.

42. Lemma C. Suppose that K and G ham the same meaning as in
Lemma A, and that
(4-21)
Then

(4-22) ■

x |(loglog® + ff)1' | x (loglog®TO)1'-1 [ v (loglogg + C)"-2 | ^ j ,

the series being continued to the term V.

It is plainly sufficient to prove this inequality when \ = T%. In what
follows we shall suppose that X has this particular value.
272 Normal number of Prime Factors of a number n

We require a preliminary inequality analogous to (2'23) and (314).

Consider the numbers which do not exceed x and are comprised in the
table
2“ .Pl
3“ • pi • Pi---p,+i-a,

P'-Pl .pS...p.+l-.,
where now pu p,, ..., p„+are primes, arranged in ascending order of
magnitude but not necessarily different, and where p„+1_„5:2 in the first
row, 3 in the second, and so forth. Arguing as in 2'2 and 3*1, we
now find
(4-23)

;,n. (D (?)
\.in« (?)+
■■■■
This inequality differs formally from (314) in that the suffixes of the func¬
tions on the right-hand side sink by one from term to term.
We can now prove (4-22) by a process of induction similar in principle to
that of 2-2 and 31. Let us suppose that the inequality is true when v is
replaced by v — >, and let us write
log log&- + 0= f,
as in 2’3. Substituting in (4*23), and observing that

-e obtain
(4-24) IL+1(a)<-
xp\og{xjp) ((>-1)1
1 (
ep2 log (a/p5) ( (v-2)! (n — 3)!
_ 1 1_ £’-‘' ,
S-l**!og(*/P,)l(*'-8)l'r (n — 4)!+
seen already,
^ 1 < log log x + G _
(4-251)
s'«*plog(®/p) logic
ad
1 s+1
(4-252) (2-
y»««aP*log(a/f>*) log X '
if s ^ 2. It is moreover easy to prove that
(4-2531) (s + 1) (2- + 3-* + 5- +...)< 2\‘ = 2 (&)•
if s = 2, and
(4-2532) (s + 1) (2-* + 3-* + 6- + ...)< Xs = (■*)»
 Normal number of Prime Factors of a number n 273
if s>2*. From (4-24)—(4-2532) it follows that

,^1,4111
*log* \(v-2)r („-8)I + -J
i x* y ( , r- )
"log* U--3)!+ (v —4) ! ”)

"When we collect together the various terms on the right-hand side which
involve the same powers of f, it will be found that the coefficient of p-r is
exactly
Kx Xp
log x\v-p)V
except when p = 1, when it
l-i}
, v)\oSx(v-l)\:
We thus obtain (4-22).
4 3. We may now argue substantially as in 2'3. We have to shew, for
example, that ’
(431) &= n„+1 (x) =o (x);
and this is equivalent to proving that
(4'32) £=^2^11,^ (*) = <, (it),
where i/r is any function of x which tends to infinity with x.
We choose S so that

log (1+|) <tt.

(4-34) 2 nK+1(®)+ 2 n,+1(»)

say. In S', we have

(4-351)
<(»-!) , <K1x?
P log® vV
where
K
(4-352) Fl = ,
-*(! + «)’
irectly for s = 2 and »=3, and th
274 Normal number of Prime Factors of a number n

and by means of this inequality we can shew, just as in 2-3, that

(4-353) S'= o(a).
In discussing 8" we must use (4-22) in a different manner. We have
(4-381)
n,+1 (e) <

(4-362)

(4-363) *7=1- (1 A) + (! + «) (1/M = (1 + S) log $ - i > 0,

by (4-33). Hence
(4-364) S" = o(4
From (4-34), (4-353), and (4-364), it follows that 8 = o (x).

4 4. We hq.ve therefore the following theorems.

Theorem C. The result of Theorem B remains true when F (n) is sub¬
stituted forf(n).
Theorem O'. The result of Theorem B' remains true when the word
“ different ” is omitted.
Theorem C". The normal order of the total number of prime factors of
a number is log log n.

V.
The normal order of d (n).
5T. It is natural to ask whether similar theorems cannot be proved with
regard to some of the other standard arithmetical functions of n, such as
d (n).
If n=
we have d (n) = (1 + a,) (1 + a,)... (1 + a,).
Since 2 < 1 + a < 2“
if a >1, we obtain at once
2v<d(m)s2°i+‘ts+-.-+‘v,

(5-11) V™ <d(n)% 2F<nK

 Normal number of Prime Factors of a number n 275
From and Theorems B' and C', we obtain at once
Theorem D'. The inequalities
(512) 2lo5,oe’*'*’/(’<«l05”>< d (n) < 21"e1°G’*+*-/<iogiog>o
where tj> is any Junction of n which tends to infinity with n, are satisfied for
almost all numbers n.
The inequalities (512) are of a much less precise type than (1-26): we
cannot say that the normal order of d (n) is a10*11’^. We can however say
that (to put it roughly) the normal order of d (n) is about
210Blosn = (l0g n)los2 _ (]0g ,))•»...
It should be observed that this order is far removed from log n, the average
order of d (n). The explanation of this apparent paradox is simple. The
majority of numbers have about (log n)l°*! divisors. But those which have an
abnormal number may have a very much larger number indeed: the excess of
the maximum order over the normal order is so great that, when we compute
the average order, it is the numbers with an abnormal number of divisors
which dominate the calculation. The maximum order of the number of
prime factors is not large enough to give rise to a similar phenomenon.
 36
ASYMPTOTIC FORMULAE IN COMBINATORY
ANALYSIS*
(.Proceedings of the London Mathematical Society, 2, xvn, 1918, 75—115)
1. Introduction and summary of results.
Tl. The present paper is the outcome of an attempt to apply to the
principal problems of the theory of partitions the methods, depending upon
the theory of analytic functions, which have proved so fruitful in the theory
of the distribution of primes and allied branches of the analytic theory of
 Asymptotic Formula) in Combinatory Analysis 277

been unable to discover in the literature of the subject any allusion whatever
to the question of the order of magnitude of p (n).
1-2. The function p (n) may, of course, be expressed in the form of an
integral
(i'2i) P(B)“ifr£raio*
by means of Cauchy’s theorem, the path T enclosing the origin and lying
entirely inside the unit circle. The idea which dominates this paper is that
of obtaining asymptotic formula?, for p (n) by a detailed study of the integral
(1-21). This idea is an extremely obvious one; it is the idea which has
dominated nine-tenths of modern research in the analytic theory of numbers:
and it may seem very strange that it should never have been applied to this
particular problem before. Of this there are no doubt two explanations. The
first is that the theory of partitions has received its most important develop¬
ments, since its foundation by Euler, at the hands of a series of mathematicians
whose interests have lain primarily in algebra. The second and more funda¬
mental reason is to be found in the extreme complexity of the behaviour of
the generating function f(x) near a point of the unit circle.
It is instructive to contrast this problem with the corresphnding problems
which arise for the arithmetical functions ir{n), ^(n), i|/-(n), p(n), d(n), ...
which have their genesis in Riemann’s Zeta-function and the functions allied

Sylvester, and Glaisher: we may refer in particular to J. J. Sylvester, “On a discovery in

the theory of partitions,” Quarterly Journal, Vol. I, 1857, pp. 81—85, and “On the parti¬
tion of numbers,” ibid., pp. 141—152 (Sylvester’s Works, Vol. II, pp. 86—89 and 90-99) ;
J. W. L. Glaisher, “On the number of partitions of a number into a given number of
parts," Quarterly Journal, Vol. Xi, 1909, pp. 67—143 ; “ Formula for partitions into given
elements, derived from Sylvester’s Theorem,” ibid., pp. 275—348; “Formula for the
number of partitions of a number into the elements 1, 2, 3, ..., n up to a = 9,” ibid.,
Vol. xli, 1910, pp. 94—112; and further references will be found in Macllahon, loo. cit.,
pp. 59—71, and E. Netto, Lehrbuch der Combinatorik, 1901, pp. 146—158. Thus, for
example, the coefficient of as" in

is * jD3(TO)=^(« + 3)2_^ + ^(-_i)n+^cos2-^;

as is easily found by separating the function into partial fractions. This function may
also be expressed in the forms
A <» + 3)*+<* eosfr*)*- (§8m*ir»)’
l + [*»(»+S)l {-h(n+m,
.where [»] and {«} denote the greatest integer contained in si and the integer nearest to n.
These formula; do, of course, furnish incidentally asymptotic formulas for the functions in
question. But they are, from this point of view, of a very trivial character: the interest
278 Asymptotic Formulae in Combinatory Analysis

to it. In the latter problems we are dealing with functions defined by

Dirichlet’s series. The study of such functions presents difficulties far more
fundamental than any which confront us in the theory of the modular
functions. These difficulties, however, relate to the distribution of the zeros
of the functions and their general behaviour at infinity: no difficulties what¬
ever are occasioned by the crude singularities of the functions in the finite
part of the plane. The single finite singularity of ?(s), for example, the pole
at s=l, is a singularity of the simplest possible character. It is this pole
which gives rise to the dominant terms in the asymptotic formulas for the
arithmetical functions associated with f (s). To prove such a formula rigorously
is often exceedingly difficult; to determine precisely the order of the error
which it involves is in many cases a problem which still defies the utmost
resources of analysis. But to write down the dominant terms involves, as
a rule, no difficulty more formidable than that of deforming a path of integra¬
tion over a pole of the subject of integration and calculating the. corresponding

In the theory of partitions, on the other hand, we are dealing with

functions which do not exist at all outside the unit circle. Every point of
the circle is an essential singularity of the function, and no part of the contour
of integration can be deformed in such a manner as to make its contribution
obviously negligible. Every element of the contour requires special study;
and there is no obvious method of writing down a “ dominant term.”
The difficulties of the problem appear then, at first sight, to be very serious.
We possess, however, in the formulae of the theory of the linear transformation
of the elliptic functions, an extremely powerful analytical weapon by means
of which we can study the behaviour of /(x) near any assigned point of the
unit circle*. It is to an appropriate use of these formulae that the accuracy
of our final results, an accuracy which will, we think, be found to be quite
startling, is due.
T3. It is very important, in dealing with such a problem as this, to
distinguish clearly the various stages to which we can progress by arguments
of a progressively “ deeper ” and less elementary character. The earlier results
are naturally (so far as the particular problem is concerned) superseded by
the later. But the more elementary methods are likely to be applicable to
other problems in which the more subtle analysis is impracticable.
We have attacked this particular problem by a. considerable number of.
different methods,-and cannot profess to have reached any very precise con¬
clusions as to the possibilities of each. A detailed comparison of the results
* See G. H. Hardy and J. E. Littlewood, “Some problems of Diophantine approxima¬
tion (II: The trigonometrical series associated with the elliptic Theta-functions),” Acta
Mathematical Vol. XXXVII, 1914, pp. 193—238, for applications of the formulae to different
but not unrelated problems.
 Asymptotic Formula in Combinatory Analysis 279

to which they lead would moreover expand this paper to a quite unreasonable
length. But we have thought it worth while to include a short account of
two of them. The first is quite elementary; it depends only on Euler’s
identity
(1'31) j
(l-*)(l-0 (1-^ = 1 + (T^)*+ (T1 ^(1
—an identity capable of wide generalisation—and on elementary algebraical
reasoning. By these means we shew, in section 2, that
(1-32) eAJn < p (n) < eBJn,
where A and B are positive constants, for all sufficiently large values of Ji¬
lt follows that
(1-33) A *Jn < logp (n) < B yffi ;
and the next question which arises is the question whether a constant G exists
such that
(1-34) log_p(ji)~Cy>t.
We prove that this is so in section 3. Our proof is still, in a sense, “ ele¬
mentary.” It does not appeal to the theory of analytic functions, depending
only on a general arithmetic theorem concerning infinite series; but this
theorem is of the difficult and delicate type which Messrs Hardy and Little-
wood have called “ Tauberian.” The actual theorem required was proved by
us in a paper recently printed in these Proceedings*. It shews that
(1-35)
in other words that
(1-36) p (n) = exp
VflF
where e is small when n is large. This method is one of very wide application.
It may be used, for example, to prove that, if p<!| (n) denotes the number of
partitions of n into perfect s-th powers, then
log;p« (n) ~ (s + 1) g T (l + i) ?(l +
It isfcertainly possible to obtain, by means of arguments of this general
character, information about p (u) more precise than that furnished by the
formula (T36). And it is equally possible to prove (I'36) by reasoning of a
more elementary, though more special, character: we have a proof, for example,
based on the identity
np(;>i)= %a-(v)p(n-v),
* G. H. Hardy and S. Ramanujan, “Asymptotic formulas for the distribution of
integers of various types,” Proc. London Math. Soc., Ser. 2, Vol. xvi, 1917, pp. 112—132
[No. 34 of this volume].
280 Asymptotic Formula in Combinatory Analysis

where er (») is the sum of the divisors of v, and a process of induction. But
we are at present unable to obtain, by any method which does not depend
upon Cauchy’s theorem, a result as precise as that which we state in the next
paragraph, a result, that is to say, which is “ vraiment asymptotique.”
1-4, Our next step was to replace (1'36) by the much more precise
formula

The proof of this formula appears to necessitate the use of much more
powerful machinery, Cauchy’s integral (121) and the functional relation

^ ^ (l0*;) exP
where
(W3) * — exp | log-(i/-}|.
This formula is merely a statement in a different notation o
between h (t) and h (T), where

It is interesting to observe the correspondence between (141) and the

results of numerical computation. Numerical data furnished to us by Major
MacMahon gave the following results: we denote the right-hand side of

It will be observed that the progress of ■ajp towards its limit unity is not
very rapid, and that — p is always positive and appears to tend rapidly to
infinity.
Tannery and Molk, loc. oit., p. 265 (Table XLV, 5).
 Asymptotic Formulae in Combinatory Analysis 281

T5. In order to obtain more satisfactory results it is necessary to con¬

struct some auxiliary function F(x) which is regular at all points of the unit
circle save x = 1, and has there a singularity of a type as near as possible to
that of the singularity of f(x). We may then hope to obtain a much more
precise approximation by applying Cauchy’s theorem to/—F instead of to/.
For although every point of the circle is a singular point of / x = l is, to put
it roughly, much the heaviest singularity. When a-a-l by real values, f(x)
tends to infinity like an exponential

exp{e(rb)}’
when *=
p and q being co-prime integers, and r-*-l, \f{x) \ tends to infinity like an
exponential

while, if * = re29",
where 8 is irrational, | f(x) | can become infinite at most like an exponential
of the type

The function required is

' (1-61)
where
, , . d fcoshCA„-l)
(1-52)
• —t*
(1-53) G = 2vr/V6 = W(f), = /(» - A).
This function may be transformed into an integral by means of a general
formula given by Lindelofif; and it is then easy to prove that the “ principal
branch ” of F(x) is regular all over the plane except at « = 1 + ; and that
F(x)-X(x),
* The statements concerning the “rational” points are corollaries of the formula? of
the transformation theory, and proofs of them are contained in the body of the paper.
The proposition concerning “irrational” points may be proved by arguments similar to
those used by Hardy and Littlewood in their memoir already quoted. It is not needed for
our present purpose. As a matter of fact it is generally true that fix) 0 when 6 is
irrational, and very nearly as rapidly as $(l-r). It is in reality owing to this that our
final method is so successful.
+ E. Liudelfif, Le calmil des r&idns et ses applications d la the'orie des foliations
(Gauthier-Villars, Collection Borel, 1905), p. 111.
1 We speak, of course, of the principal branch of the function, viz. that represented by
the series (1-51) when x is small. The other branches are singular at the origin.
282 Asymptotic Formulae in Combinatory Analysis

where

(1'64) x7(,ogShiskiwi-*]
is regular for x=l. If we compare (1 -42) and (T54), and observe that /(«')
tends to unity with extreme rapidity when sc tends to 1 along any regular
path which does not touch the circle of convergence, we can see at once the
very close similarity between the behaviour of / and F inside the unit circle
and in the neighbourhood of x — 1.
It should be observed that the term -1 in (1-52) and (1-54) is—so far as
our present assertions are concerned—otiose: all that we have said remains
true if it is omitted ; the resemblance between the singularities of f and F
becomes indeed even closer. The term is inserted merely in order to facilitate
some of oar preliminary analysis, and will prove to be without influence on
the final result.
Applying Cauchy’s theorem to f—F, we obtain

where D is any number greater than

1"6. The formula (1'55) is an asymptotic formula of a type far more precise
than that of (1'41). The error term is, however, of an exponential type, and
may be expected ultimately to increase with very great rapidity. It was
therefore with considerable surprise that we found what exceedingly good
results the formula gives for fairly large values of n. For » = 61, 62, 63 it

1121538-972, 1300121-359, 1505535-606,

while the correct values are
1121505, 1300156, 1505499.
The errors 33-972, -34-641, 36 606
are relatively very small, and alternate in sign.

The next step is naturally to direct our attention to the singular point of
f(x) next in importance after that at ®=1, viz., that at x— — 1; and to
subtract from f(x) a second auxiliary function, related to this point as F(x)
is to x- 1. No new difficulty of principle is involved, and we find that
. , Id /ecM (- 1)» d
(1-61) ■0(e™“),
 Asymptotic Formulce in Combinatory Analysis 283

where D is now any number greater than }C. It now becomes obvious why
our earlier approximation gave errors alternately of excess and of defect.
It is obvious that this process may be repeated indefinitely. The
singularities next in importance are those at x = e!jn and z = e&"1; the next
those at x—i and»=—i) and so on. The next two terms in the approximate
formula are found to be

As we proceed further, the complexity of the calculations increases. The

auxiliary function associated with the point x = e’**'■!* involves a certain
24}-th root of unity, connected with the linear transformation which must
be used in order to elucidate the behaviour of f(x) near the point; and the
explicit expression of this root in terms of p and q, though known, is some¬
what complex. But it is plain that, by taking a sufficient number of terms,
we can find a formula in which the error is
0(W),
where v is a fixed but arbitrarily large integer.

17. A final question remains. We have still the resource of making

v a function of n, that is to say of making the number of terms in our
approximate formula itself a function of n. In this way we may reasonably
hope, at any rate, to find a formula in which the error is of order less than
that of any exponential of the type e™; of the order of a power of n, for
example, or even bounded.
When, however, we proceeded 'to test this hypothesis by means of the
numerical data most kindly provided for us by Major MacMahon, we found
a correspondence between the real -and the approximate values of such
astonishing accuracy as to lead us to hope for even more. Taking n = 100,
we found that the first six terms of our formula gave
190568944-783
+ 348-872

+ •685
+ •318
-■064
190569291-996,
while ^,(100) = 190569292;
284 Asymptotic Formulae in Combinatory Analysis

so that the error after six terms is only -004. We then proceeded to calculate
p (200), and found
3, 972, 998, 993, 180-896
+ 36, 282-978

+ 5-147
+ 1-424
+ 0-071
+ o-ooo*
_ + 0-043
3, "972, 999~ 029, 388-004,
and Major MacMahon’s subsequent calculations shewed that p (200) is, in fact,
3, 972, 999, 029, 388.
These results suggest very forcibly that it is possible to obtain a formula for
p (n), which not only exhibits its order of magnitude aud structure, but may¬
be used to calculate its exact value for any value of n. That this' is in fact so
is shewn by the following- theorem.

Statement of the main theorem.

where G and Xn are defined by the equations (1-53), for all positive integral
values of q; that p is a positive integer less than and prime to q; that a>Vli is
a 24q-th root of unity, defined when p is odd by the formula
(1-721)

= exp[-{i (2-P2-P) + t’i(s-^) (2p-p' +pp')])rri\^ ,

and when q is odd by the formula

(1-722)
= (=-) exP [- {i (2 -1) + A (2 “) (2p-P +fp')\ .
where (ajb) is the symbol of Legendre and Jacobif, and p' is any positive
integer such that 1 + pp' is divisible by q; that
(1-73) A,{n) = ZnPi,e-**«l*;

and that a is any positive constant, and v the integral part of a>Jn.
* This term vanishes identically.
t See Tannery and Molk, loc. cit., pp. 104—106, for a complete set of rules for the cal¬
culation of the value of (ajb), which is, of course, always 1 or -1. When both p and q
are odd it is indifferent which formula is adopted.
 Asymptotic Formula in Combinatory Analysis 285

Then
(1-74) p(n)=tAt<l>t + 0[n-L),

so that p (n) is, for all sufficiently large values of n, the integer nearest to

(1-75) iAt<j)t.

It should be observed that all the numbers Aq are real. A table of Aq

from q = 1 to q= 18 is given at the end of the paper (Table II).
The proof of this theorem is given in section 5; section 4 being devoted
to a number of preliminary lemmas. The proof is naturally somewhat
intricate; and we trust that we have arranged it in such a form as to be
readily intelligible. In section 6 we draw attention to one or two questions
which our theorem, in spite of its apparent completeness, still leaves open.
In section 7 we indicate some other problems in combinatory analysis and
the analytic theory of numbers to which our method may be applied; and
we conclude by giving some functional and numerical tables: for the latter
we are indebted to Major MacMahon and Mr H. B. C. Darling. To Major
MacMahon in particular we owe many thanks for the amount of trouble he
has taken over very tedious calculations. It is certain thtvt, without the
encouragement given by the results of these calculations, we should never
have attempted to prove theoretical results at all comparable in precision
with those which we have enunciated.

2. Elementary proof that eAJn<p (n) < eBJn for sufficiently

LARGE VALUES OF 11.
21. In this section we give the elementary proof of the inequalities
(T32). We prove, in fact, rather more, viz., that positive constants H and K
exist such that
(211) ' — («)<—

for n > 1*. We shall use in our proof only Euler’s formula (131) and a
debased form of Stirling’s theorem, easily demonstrable by quite elementary
methods : the proposition that

lies between two positive constants for all positive integral values of n.

* Somewhat inferior inequalities, of the typo

2A^<p(n)<ns^’l\
may be proved by entirety elementary reasoning; by reasoning, that is to say, which
depends only on the arithmetical definition of p(n) and on elementary finite algebra, and
does not presuppose the notion of a limit or the definitions of the logarithmic or expo-
286 Asymptotic Formula? in Combinatory Analysis

2'2. The proof of the first of the two inequalities is slightly the simpler.
It is obvious that if

W*“ = (!-«)
so that pr (n) is the number of partitions of n into parts not'exceeding r, then
(2'21) pr (n) = (?!) + (n - r) + pr.t (n - 2r) + ....
We shall use this equation to prove, by induction, that

(2-22)

It is obvious that (2'22) is true for r = 1. Assuming it to be true for r — s,

and using (2'21), we obtain

ps+l («) +(»-»- I)8"1 + (» - 2s - 2ri + ...}

■ « fn»-(n-a-I)» (n - s -1)* - - 2s - 2)* 1

s (a +1) + *(« + l) J
»8 (« + l)n«
(s + 1) (s!)! {(s +1) !}2'
This proves (2 22). Now p (n) is obviously not less than pr (n), whatever the
value of r. Take r = [V»] : then
. I\/»] «[J“1 H .
P ('«) » PlJn] (»)S ^e * >

by a simple application of the degenerate form of Stirling’s theorem men¬

tioned above.
2'3. The proof of the second inequality depends upon Euler’s identity.
If we write

k = (l-^..(l-«?)2’

(2-31) qT (n) = gy_, (n) + 22,_, (n-r) + 32r_i (n-2r) + ...,

(2-32) p (n)=q, (n-l) + q, (n-4) + 2,(n-9) + ....

We shall first prove by induction that

<2'33> yw
This is obviously, true for r = 1. Assuming it to be true for r = s, and using
(2-31), we obtain

9s+1 (n) ^ (2s — 1)! (a l)a + + 2 (n + s - 1)M_1

. +3 (« + «“-2s-2)“->+
 Asymptotic Formula? in Combinatory Analysis 287
Now m (m — 1) a"*~s 6! <; (a + b)m - 2am + (a - 4)“
if m is a positive integer, and a, b, and a-b are positive, while if a - 4$0,
and to is odd, the term (a - 4)“ may be omitted. In this inequality write
m = 2s + l, a = n + ss — ks -k (k = 0,1/2,...), 6 = s + l,
and sum with respect to k. We find that
(2s + 1) 2s (s -f IT {(« + s’?*"1 + 2 (« + s’ - s -1)«~ + ...}«(rc + s’ + s +1)”+’;

. (,,.+S’+s+ir-’-H {)t+(,+i)y+.
.2s+I' " (2s +1) 2s (s + 1)’ (2s -1)! (s !)> * (2s +1)! {(s +1)!}’'
Hence (2'33) is true generally.
It follows from (2'32) that
?(.,=fl(.-i)+6(»-4)+...a^;Fo.,
But, using the degenerate form of Stirling’s theorem once more, we find
without difficulty that
_1_2"K
(2r-l)!(r!)’< 4H >
where K is a constant. Hence ,
p (») < 8K t < 8K£ ^s’

This is the second of the inequalities (2T1).

3. Application of a Tauberian theorem to the determination

of the constant C.
31. The value of the constant
C = lim Io gpM
s/n
is most naturally determined by the u:lse of the following theorem.
V g M = %a„xn is a power-series iwith positive coefficients, and

log
when a->-1, then
log sn = log (a, + a, + ... + a») ~ 2 V(-4n)
when n-t- oo .
. This theorem is a special case* of Theorem C in our paper already
referred to.
Now suppose that
g(*) = (i-*)f(*) = 2{pM-P(n-i)}^(1^)(iJV(1_^ ,

* Loc. cit., p. 129 (with a= 1) [p. 258 of tliis volume].

288 Asymptotic Formulce in Combinatory Analysis
Then an=p(n)-p(n-l)
is plainly positive. And
(311)
log y (.x) = S log =S \ ~ |i ,

when x -*■ 1 *. Hence

(3-12) logp(») = a0 + aa + ... + a„~OVn,
whore C' = 2tt;V6 = ir \/(|). as in (1-53).
3'2. There is no doubt that it is possible, by ‘'Tauberian” arguments, to
prove a good deal more about p (n) than is asserted by (312). The functional
equation satisfied by/(r) shews, for example, that

a relation far more precise than (311). From this relation, and the fact that
the coefficients in y (,r) are positive, it is certainly possible to deduce more
than (312). But it hardly seems likely that arguments of this character will
lead us to a proof of (111). It would be exceedingly interesting to know
exactly how fan-- they will carry us, since the method is comparatively ele¬
mentary, and has a much wider range of application than the more powerful
methods employed later in this paper. We must, however, reserve the dis¬
cussion of this question for some future occasion.

4. Lemmas preliminary to the proof of the main theorem.

41. We proceed now to the proof of our main theorem. The proof is
somewhat intricate, and depends on a number of subsidiary theorems which
we shall state as lemmas.
Lemmas concerning Farey's series.
4'21. The Farey’s series of order m is the aggregate of irreducible rational
fractions
plq (0 -ip^q^m),
* This is a special case of much more general theorems: see K. Knopp, “Grenzwerte
von Reiken bei der Anniikemng an die lionvQvgemgi'tiii2e,v Inaugural-Dissertation, Berlin,
 Asymptotic Formula; in Combinatory Analysis 289

arranged in ascending order of magnitude. Thus

t. h b b i, h b i b b h 11. h i, t, i h i
is the Farey’s series of order 7.
Lemma 4-21. If plq, p jq' are two successive terms of a Farexjs series, then
(4'211) p'q-p<f = l.
This is, of course, a well-known theorem, first observed by Farey and
first proved by Cauchy*. The following exceedingly simple proof is due to
Hurwitzf.
The result is plainly true when m = 1. Let us suppose it true for m = le;
and let plq, p'jq be two consecutive terms in the series of order k.
Suppose now that p"/q" is a term of the series of order k +1 which falls
between plq and p'jq'. Let
p"q-pf' = X> 0, pY~p‘V = P>0.
Solving these equations for p", q", and observing that p'q ~pq' = 1, we obtain
p" = pp + Xp, q" = pq+Xq'.
Consider now the aggregate of fractions
(pp + Xpj/(pq + Xqj,
where X and p are positive integers without common factoL All of these
fractions lie between p/q and p'/q ; and all are in their lowest terms, since
a factor common to numerator and denominator would divide
\=q(pp + Xp') — p (pq + Xg),
and p =p {pq + Xqj - q (pp +Xpj.
Each of them first makes its appearance in the Farey’s series of order
pq + Xq', and the first of them to make its appearance must be that for
which X = 1, p = 1. Hence
p"=p + p', q" = q + q,
p"q-pq" =p'q"-p"q'=l.
The lemma is consequently proved by induction.
Lemma 4-22. Suppose that p/q is a term of the Farey s series of order m,
andp'jq", p'jf the adjacent terms on the left and the right: and let jVtq denote
the interval
p_*_ £+—5_t
2 (7(9 + 9")' 9 9(2 + 9')
* J. Farey, “On a curious property of vulgar fractious,” Phil. Mag., Sur. 1, Vol, xlvii,
1816, pp. 385, 386; A. L. Cauchy, “Demonstration d’un thdoreme curicux sur les nom-
bres,” Exercises de mathdmatiques, Vol. r, 1826, pp. 114—116. Cauchy’s proof was first
published in the Bulletin de la Societe Philomatique in 1816.
t A. Hurrvite, “Ueberdie angoniiherte Darstellung der Zahlou duroh rationale Briiohe,”
Math. Ann., Vol. xt,iv, J894, pp. 417—436.
| When plq is 0/1 or 1/1, only the part of this interval inside (0, 1) is to he taken;
thus^j is 0, l/(m+l) and/,., is l-l/(m4-l), 1.
B.C.F. 19
290 Asymptotic Formula, in Combinatory Analysis

Then (i) the intervals jPt} exactly fill up the continuum (0,1), and (ii) the length
of each of the parts into which jP}, is divided by pjq * is greater than lj2mq
and less than IJmq.
(i) Since
1_ , 1 _ 1 _v'q~pq' _P' P
2(2 + 2') ?'(?' + ?) 24 22' ?' ?’
the intervals just fill up the continuum.
(ii) Since neither q nor q' exceeds m, and one at least must he less than m,

1 1
q(q + q)>2mq'
Also 2 + 2' >m> sinoe otherwise {p+pjj(q + q) would be a term in the series
between pjq and pjq'. Hence
1 J_
2(2 + 2') m'
Standard dissection of a circle.
423. The following mode of dissection of a circle, based upon Lemma
4-22, is of fundamental importance for our analysis.
Suppose that the circle is defined by
x = Re^> (O<0<1).
Construct the Farey’s series of order m, and the corresponding intervals jp>q.
When these intervals are' considered as intervals of variation of 8, and the
two extreme intervals, which correspond to abutting arcs on the circle, are
regarded as constituting a single interval the circle is divided into a
number of arcs
k*.
where q ranges from 1 to ni and p through the numbers not exceeding and
prime to q-\. We call this dissection of the circle the dissection Hm.

Lemmas from the theory of the linear transformation of the elliptic

modular junctions.
4'3. Lemma 4-31. Suppose that q is a positive integer; that p is a positive
integer not exceeding and prime to q; that p' is a positive integer such that
1 +pp' is divisible by q; that mM is defined by the formulas. (1-721) or (F722);
that
( 2tt2 2prri\ , f 2rr 2 p'm\

* See the preceding footnote [footnote J of p. 289].

+ ^5 = 0 occurring with q = l only.
 Asymptotic Formulae, in Combinatory Analysis 291
where the real part of z is positive; and that

f{x) = (l-x)(l-a?) (1
Then
This lemma is merely a restatement in a different notation of well-known
formulae in the transformation theory.
Suppose, for example, that p is odd. If we take
a = p, b = -q, c = ii—, d = — p',
so that ad —be- 1; and write
a; = 9« = e«"> a,' = Qs = e»*i t

so that r = £ + -, T = ^ + -;
2 9 2 1Z
then we can easily verify that

Also, in the notation of Tannery and Molk, we have

and the formula for the linear transformation of h (r) is

h (T) = exp [ft (a -1) - * [a (b - c) + bd (a* -1)]} wt] y'(a + br) h (r),
where >f(a + br) has its real part positive *. A little elementary algebra will
shew the equivalence of this result and-ours.
The other formula for mPtq may be verified similarly, but in this case we

a = -p, b = q, e=- d = p’.

We have included in the Appendix (Table I) a short table of some values

of aPti, or rather of (log a>Pig)jTri.
Lemma 4-32. The function f(x) satisfies the equation
(4-321)

•where
/(*) - «*. 7 [l (£) } 4 «P 1

(4-322)

^-”P{-^iog%*-)+2£r\-
* Tannery and Molk, loo. citpp. 113, 267.
19—2
292 Asymptotic Formula in Combinatory Analysis

This is an immediate corollary from Lemma 4'31, since

If we observe that
f{x'M) = l + p(l)x'lhq+...,
.we see that, if x tends to along a radius vector, or indeed any regular
path which does not touch the circle of convergence, the difference

/ (*) - * J [l % ty} 4 -P
tends to zero with great rapidity. It is on this fact that our analysis is based.

Lemmas concerning the auxiliary function Fa («).

441. The auxiliary function Fa (x) is defined by the equation

Fa(x) = Xfa(n)x’',

, , , , d cosh a\n — 1
where (») ^-,

Xn = f(n — 4f), a >0.

Lemma 4-41. Suppose that a cut is made along the segment (I, co) in the
plane of x. Then Fa (x) is regular at all points inside the region thus defined.
This lemma is an immediate corollary of a general theorem proved by
Lindelof on pp. 109 et seq. of his Galcul des residue*.
™, 4. . . . d cosh a f(z - Jf) - 1
The function 1]'a(e)=j-n-fr-
Y ' dz V(*-A)
satisfies the conditions imposed upon it by Lindelof, if the number which he
calls a is greater than ; and

(4-411) F. («) = |“_+“ <f> (a) dz,

if. x = reie, 0<8< 2rr, x? = exp{z (\ogr + id)}.

4'42. Lemma 4'42. Suppose that D is the region defined by the in¬
equalities
— 7t < — 80< 8 < < 7r, r0 < r, 0 < r0 < 1,

* Lindelof gives references to Meliin and Le Koy, who had previously established the
theorem in less general forms.
 Asymptotic Formula in Combinatory Analysis 293

and that log (1/a) has its principal value, so that Log(l/a-) is one-valued, and
its square root two-vahiei, in D. Further, let

*• (*) = Vk log (1/.)} [exp {^'m} ~ l] ,

that value of the square root being chosen which is positive when 0<x<l.
Then
(*>-%.(*)
is regular inside D *.
We observe first that, when 8 has a fixed value between 0 and 2ir, the
Integral on the right-hand side of (4-411) is uniformly convergent for
Hence we may take a — fa in (4-411). We thus obtain

Fa (x) = ixis J _ Jfcm - 2rt fa (fa + H) dt + fit* + fa(fa-it) dt,

where the f(it) and a/(-it) which occur in (fa + it) and (fa - it) are to
be interpreted as e^'ft and e~iniyt respectively. We write this in the form

(4-421) Fa (a,) = x„ (x) + B* 1 f“ (& + »*) dt

+fa (fa - it) h

= Xa (x) + ©! (x) + ©2 (*),

say, where (*) = fit*j“ x“ ajra (fa + it) dt.

Now, since | xu | = e~et, \ x~u | = e“,
the functions ® are i-egular throughout the angle of Lemma 4-4‘2. And

where X — i log ^, g — afi.

The form of this integral may be calculated by supposing X and g positive,

when we obtain

J —~Jdw=2xJ e_to2 (cosh gw — 1) dw

= f(Xfa(e^-l).

(4-422) Xa (x) = ffa log (1/®)} a* [exp - 1J = Xa (*),

and the proof of the lemma is completed.
* Both Fa (*) and Xa (x) are two-valued in I). The value of Fa (x) contemplated is
naturally that represented by the power-series.
294 Asymptotic Formulce in Combinatory Analysis

Lemmas 441 and 442 shew that « = 1 is the sole finite singularity of
the principal branch of Fa (x).

4'43. Lemma 4-43. Suppose that P, 61, and A are positive constants,
0, being less than tt. Then
\Fa(x)\<K = K(F,61,A),
for 0<r^P, — 0< a < 4.
We use K generally to denote a positive number independent of x and
of a. We may employ the formula (4411). It is plain that

I ± (cosha \/(z — j-() — I KeKJl<l

I’M*) ^1 \'(*-if) J |
where r, is the imaginary part of z. Hence

1 Fa (x) j < K j‘ dr, < K.

4’44. Lemma 444. Let c be a circle whose centre is x=-l, and whose
radius S is less than unity. Then
\Fa(x)-Xa(x)\<Ka\
if x lies in c and 0 < a ^ A, K — K (S, A) being as before independent of x and
of a.
If we refer back to (4421) and (4422), we see that it is sufficient to prove
that 0
| ®! (x) | < Ka2, |, ®2 (») | < Kd?;
and we may plainly confine ourselves to the first of these inequalities. We

1) dt.
ft j
Rejecting the extraneous factor, which is plainly without importance, and
integrating by parts, we obtain
@w=J”4)(i)cosha^)-l^

where <t> (t) ■

e - h*‘ + 2,r« _ 2 (g - jVirt + 2irt_ JJS '
Now I e I < \rr and | | < Ke^. It follows that
|«(*))<JSC«-*;
 Asymptotic Formulce in Combinatory Analysis 295

and |®Mi< e-^ \ sinh2 |a </(»«) j dt

<K Jo V ‘C°Sh “ ~ C°S a dt

= K (ea'-l‘” - e-«s/s-) < Ka\

5. Proof of the main theorem.

51. We write
(511)

where G=tt \/|, *j),3 = ae-2*”"/?; and

(512) $ (*) =/(®) -tS,F.Aq (*),

where the summation applies to all values of p not exceeding q and prime
to q, and to all values of q such that
(513) l<5<.»=[a,/'»],
a being positive and independent of n. If then
(514) FPiq(x) = tcM,nx’‘,
we have
(515) p(n)-t%cM,n = -^-.

where T is a circle whose centre is the origin and whose radius R is less
than unity. We take ^
(516) R = l~n’
where fi also is positive and independent of n.
Our object is to shew that the integral on the right-hand side of (515)
is of the form 0 (n-*); the constant implied in the 0 will of course be a
function of a and fi. It is to be understood throughout that 0’s are used in
this sense; 0 (1), for instance, stands for a function of x, n, p, q, a, and ft (or
of some only of these variables) which is less in absolute value than a number
K=K(a, fi) independent of x, n, p, and q.
We divide up the circle P, by means of the dissection 3* of 123, into
arcs fj,i5 each associated with a point Be®”"/*; and we denote by rjpiq the arc
of T complementary to gPiq. This being so, we have

(517) JT%gd*-sJt m-JtsMto-sf'

 Asymptotic Formulas m Combinatory Analysis 297

, Hence (5-221) holds for all yPig.

iV,5 = 0(Ur"V?),
~ 0 (-Hr” 2 g-) = 0 (ifi Rrn) *■

It is plain that, if we substitute y for then write so again for y, an

finally substitute for xc/q its explicit expression as an elementary functioi
given in Lemma 442, we obtain
(5-2321)
f
j",.t-OWq) {E(x)-l}J^OK^x-n-&dx=Oyq)J,
298 Asymptotic Fonmdai in Combinatory Analysis
say, where
(5-23211) P^^}.
and the path of integration is now a line related to i = l as is to
x= ; the line defined by x = reie, where R^r^Ru and 6 is fixed and
(by Lemma 422) lies between Ijiqv and l/qv.
Integrating J by parts, we find
(52322)
(n - J,) J- - [{*» - lj yflogj)

- £ Ji# (a) -1} (log ® - ” - ti dx

+ ~ \E (x) (log ~'5 a ■- ■" " H *= = Ji + -h + -h,

5-233. In estimating J), J.3t and J„ we must bear the following facts in

(1) Since \&\>R, it follows from (52312) that |a 1== 0 (1) throughout
the range of integration.
(2) Since 1 — R = fS/n and 1/2qv < 8 < 1 jqv, where = [a \/»-], we have

logS)=°fri)’
when r = R, and i_^_ = O(?0i)J

throughout the range of integration.

(3) Since
"' ' P 6 f [{log (l/r)Js + ’
E (x) is less than n absolute value when r> 1. And, the part of the
path for which r < t is of the form

eXp0(^) = eXp0(1) = 0(1)'

It is accordingly of the form 0 (1) throughout the range of integration.
5234. Thus we have, first
(5-2341)
J^O (1) 0 (1) 0 (RT) + 0(1)0 (q-in-i) 0(1) = 0 (q-in-i),
secondly

(52342) J,= 0(1)0 (qiA) - 0 (qi n~ i),

 Asymptotic Formula? in Combinatory Analysis 299

and thirdly

(5-2343) J, = 0 (<?-) 0(1)0 (9V) £' ^ = 0 (q-in~l).

From (5-2341), (5-2342), (52343), and (5-2322), we obtain

J=0(q~in-i) + 0(qhi-i)-
and, from (5 2321), j"p,q = 0 (»-*) + 0 (3n -*).
Summing, we obtain
(5-2344) 2y"„? = 0 (tf* § > q) + 0 (n~* S 3=)
= 0 (n-t) + 0 (n-i) = 0 (jri).
5-235. From (52311), (5-2313), and (5-2344), we obtain
(52351) = ri);
and in exactly the same way we can prove
(5-2352) 1j\q=0(n~i).
And from (5-212), (5-224), (5-2351), and (5-2352), we obtain, finally,
(52353) sjy,s = 0(n-»). ,

Proof that XJp,q= 0 («-i).

5'31. We turn now to the discussion of

(5-311)

-jt

say, where ft,,, (a) = (JL log J-) a,* ,

M = Xp,} (®) + ft.,? (*) = ft>,9 W (Ojid).

E (a) being defined as in (5-23211).
Discussion ofZJ‘Plt and
5-32. The discussion of these two sums is, after the analysis which
precedes, a simple matter. The arc |j,,3 is less than a constant multiple of
1/q *Jn; and x~n - 0 (1) on %Piq. Also
I (®) — 1= 0 (?“-)>
by (5222); and
(5-321) N/(log^) = 0

K,il = -B = i- (£/»). Iam %?! < I/?"-

300 Asymptotic Formula) in Combinatory Analysis

Hence = 0 (q-hi~*),

(5-322) ,|W)9-S)“°(”'1);
and J\q - 0 (q-'n-i),
(5-323) 2/V, = 0(«“3 2 1) = 0 (it-*).
2 < 0 (.to)

ofZJ\q.
5-33. From (4-321) and (5-2221), we have
(5-331)
fix) - xM (x)=«A, yy (y log )| r^Eu,,.,

where O (a) =/ («) - 1 = XI

if 111 < 1, and + .

l°g(l/*3>,5)
4tt"- log jl/R) 1
= exP -„T
^|[loga/fJ)p+^J’
where 8 is the iplitude of Also

!■ {[Jog (TO + *1 = 0 [q> g + y)} = 0 g) ,

while log (1/jR) is greater than a constant multiple of 1 jn. There is therefore
a positive number 3, less than unity and independent of n and of q, such that

and we may write fl (scp ,) = 0 (j ic'ft , |),

We have therefore
E (xv, a) n (x'p, q) = 0 (j xp, „ j -*) 0 (| s'y, s |) = 0 ([ ^ , j«) = 0(1);
and so, by (5-321),
/(«) ~ fei («) - 0 (\/?) 0 (y/j logg- j) 0 (1) = 0 (n~t).
And hence, as the length of fj,>5 is of the form 0 (Ijqfn), we obtain
J'P,q = 0(q-'n-i),
(5-332) XFPiq=0(n~i 2^1 ) = 0 («-*).

5-34. From (5-311), (5-322), (5-323), and (5-332), we obtain

(5-341) 2/riJ = 0(n-i).
 Asymptotic Formulce in Combinatory Analysis 301
Completion of the proof.
54. From (5T5), (5T7), (5-2353), and (5-341), wo obtain
(5-41) p{n)~%lcfi,Iin=0(n-\).

But

All that remains, in order to complete the proof of the theorem, is to shew
that
cosh (C\njq) — 1
may be replaced by \
and in order to prove this it is only necessary to shew that
v » d ^cV'«_cosh(CX„/?)+l
q<o^»)q'dii \n U{ h
On differentiating we find that the sum is of the form

Thus the theorem is proved.

6. Additional remarks on the theorem.

6‘1. The theorem which we have proved gives information about p («)
which is in some ways extraordinarily exact. We are for this reason the more
anxious to point out explicitly two respects in which the results of our
analysis are incomplete.
6-21. We have proved that

where the summation extends over the values of q specified in the theorem,
for every fixed value of a; that is to say that, when a is given, a number
K - K (a) can be found such that
\p(n)-tAq<f>q\< Kn~i
for every value of n. It follows that
(6-211) p(n) = {tAq4,,},
where {®} denotes the integer nearest to a, for »!>)!„, where >i„ = na(a) is
a certain function of a.
The question remains whether we can, by an appropriate choice of a,
secure the truth of (6'211) for all values of n, and not merely for all sufficiently
large values. Our opinion is that this is possible, and that it could be proved
to be possible without any fundamental change in our analysis. Such a proof
302 Asymptotic Formula2 in Combinatory Analysis

would however involve a very careful revision of our argument. It would be

necessary to replace all formulas involving O's by inequalities, containing only
numbers expressed explicitly as functions of the various parameters employed.
This process would certainly add very considerably to the length and the
complexity of our argument. It is, as it stands, sufficient to prove what is,
from our point of view, of the greatest interest; and wo have not thought it
worth while to elaborate it further.
6'22. The second point of incompleteness of our results is of much greater
interest and importance. We have not proved either that the series

is convergent, or that, if it is convergent, it represents p (n). Nor does it seem

likely that our method is one intrinsically capable of proving these results,
if they are true—a point on which we are not prepared to express any definite
opinion.
It should be observed in this connection that we have not even discovered
anything definite concerning the order of magnitude of Aq for large values
of q. We can prove nothing better than the absolutely trivial equation
Aq = 0 (q). On the other hand we cannot assert that Aq is, for an infinity of
values of q, effectively of an order as great as q, or indeed even that it does
not tend to zero (though of course this is most unlikely).
6'3. Our formula directs us, if we wish to obtain the exact value of p («)
for a large value of n, to take a number of terms of order >Jn. The numerical
data suggest that a considerably smaller number of terms will be equally
effective; and it is easy to see that this conjecture is correct.

Let us write £ = W(f)-.«* /.-[g],

and let us suppose that a < 2. Then

2 2 0 (<f) 0 Q 0 (<P■>*!*) = 0 (^ 2

=0 yxec'lnlxds^j,

since y’qePJnli decreases steadily throughout the range of summation*.

Writing y?i/y for x, we obtain

0 (n-i /^Ve eCydy) = 0 jn-* } = 0 {»-* (log

= 0(logn)~- = o(l).
equal to 9,C-Jn
 Asymptotic Formula in Combinatory Analysis 303

It follows that it is enough, when n is sufficiently large, to take

log n
terms of the series. It is probably also necessary to take a number of terms
of order s/nj(logn); but it is not possible to prove this rigorously without
a more exact knowledge of the properties of Aq than we possess.
6'4. We add a word on certain simple approximate formula; for logp(«)
found empirically by Major MaeMahon and by ourselves. Major MacMahon
found that if
(641) log»j>(nW(« + 4)-ail,
then is approximately equal to 2 within the limits of his table of values
of p (n) (Table IV). This suggested to us that we should endeavour to find
more accurate formulae of the same type. The most striking that we have

(6 42) . ' log,, p (n) = J# {V(« +10) - a,,};

the mode of variation of an is shewn in Table III.
In this connection it is interesting to observe that the function

(which ultimately tends to infinity with exponential rapidity) is equal to '973

for n = 30000000000.

7. Further applications of the method.

71. We shall conclude with a few remarks concerning actual or possible
applications of our method to other problems in Combinatory Analysis or the
Analytic Theory of Numbers.
The class of problems in which the method gives the most striking results
may be defined as follows. Suppose that q (n) is the coefficient of xn in the
expansion of F(x), where F (x) is a function of the form

(711) U(±«r)Ym±?)}‘

f (*) being the function considered in this paper, the a’s, b’s, a’s, and /3’s being
positive integers, and the number of factors in numerator and denominator
being finite; and suppose that | F (x) | tends exponentially to infinity when x
tends in an appropriate manner to some or all of the points Then our
method may be applied in its full power to the asymptotic study of q (n), and
yields results very similar to those which we have found concerning p (n).
, OW
'”/(■*)/(?)’
ie eliminated if this i
304 Asymptotic Formula• in Combinatory Analysis

Thus, if

F r
<*> =/& (l+x) (1+(1+*3) - ° (1 (1 ■
so that q(n) is the number of partitions of n into orld parts, or into unequal
parts®, we find that
4 00 = ^|/o[*Vv'{H»+A)}]

+ y'2 cos (f nir - J- v) ^ [J tV V{i (» + yj)}] + —

The error after [a \/n] terms is of the form 0 (1). We are not in a position to
assert that the exact value of q (n) can always he obtained from the formula
(though this is probable); but the error is certainly hounded.

If F (x) - - (1 + *) (1 + *■) (1 + «•) • ■ •,

so that q (») is the number of partitions of n into parts which are both odd
and unequal, then

+ 2 cos (| n-x - §■*■) ^ J, [*«r ^ (» - *)}] + ■...

The error is again bounded (and probably tends to zero).

if ^=78? !* + 2a‘-2r;» + ...’

q (n) has no very simple arithi iterpretation; but the series ii
is the direct reciprocal of a s iple 3--funetion, of particular ir
In this ceise we find
. . Id «"'» V3
rd» \/tt 2ttC°
The error herd is (as in the partition problem) of order 0 (n~i), and the exact
value can always be found from the formula.
7'2. The method may also be applied to products of the form (711) which
have (to put the matter roughly) no exponential infinities. In such cases the
approximation is of a much less exact character. On the other hand the
problems of this character are of even greater arithmetical interest.
The standard problem of this category is that of the representation of a
number as the sum of s squares, s being any positive integer odd or evenf.
We must reserve the application of our method to this problem for another
occasion; but we can indicate the character of our main result as follows.
* Of. MacMakon, loc. cii,, p. 11. tVe give at the end of the paper a table (Table V) of
the values of q («) up to « = 100. This table was calculated by Mr Darling.
t As is well known, the arithmetical difficulties of the problem are much greater when
s is odd.
 Asymptotic Formula} in Combinatory Analysis 305
If ?'s(n) is the number of representatii

F (%) - Srs (n) xn = (1 + 2.* + 2afi + ...)* = {/Ml* JJ{p)r{JA*QT

'{/(-«•)}“ {/(*=)}“ '
We find that

(T'21) rs(,!) = rTP)"!i_12

where cv is a function of q and of n of the le general type as the function
Aq of this paper. The series
^ (7-22) 2
is absolutely convergent for sufficiently large values of s, and the summation
in (7-21) may be regarded indifferently as extended over all values of q or
only over a range 1 ^q^as/u. It should be observed that the scries (7'22)
is quite different in form from any of the infinite series which arc already
known to occur in connection with this problem.
73. There is also a wide range of problems to which our methods are
partly applicable. Suppose, for example, that
F (x) = Sp* («) x« = (i _ x) (1 _ ^ (r_-^y-. ■ ^
so that p1 In) is the number of partitions of n into squares. Then F (x) is not an
elliptic modular function; it possesses no general transformation theory: and
the full force of our method cannot be applied. We can still, however, apply
some of our preliminary methods. Thus the “Tauberian” argument shews that
. logj)*(n)~2-*3^{r(t)}l»*.
And although there is no general transformation theory, there is a formula which
enables us to specify the nature of the singularity at x = 1. This formula is

X n {1 - COS SW(«/*) +
By the use of this formula, in conjunction with Cauchy’s theorem, it is
certainly possible to obtain much more precise information about pl(n), and
in particular the formula
p? (n) ~ 3'4 (4n)' * {£(§)}* iSri <>li.
The corresponding formula for-p'(»), the number of partitions of n into
perfect s-th powers, is
306 Asymptotic Fornmlce in Combinatory Analysis
The series (7-21) may be written in the form

where is always one of the five numbers 0, e*", — e*", — e'f". When

_2__nl»-i{X-5» + 2cos(£nir-£«r)2-a* + 2cos(§mr-ktnr) 3-*"+ ...}.

It has been proved by Ramanujan that the series gives an exact repre¬
sentation of vs (n) when s = 4, 6, 8; and by Hardy that this is also true when
s = 3, 5, 7. See Ramanujan, “ On certain trigonometrical sums and their
applications in the Theory of Numbers”; Hardy, “On the expression of a
number as the sum of any number of squares, and in particular of five or

Table I

-7/if
- 29/32
27/32

* [Ramanujan’s paper referred to is Ko. 21 of this volume. That of Hardy was

published, in the first instance, in Pros. National Acad, of Sciences (Washington), Yol. nr,
1918, pp. 189—193, and later (in fuller form and with a slightly different title) in Trans.
American Math. Soc., Vol. xxi, 1920, pp. 25u—284.]
 Asymptotic Formulce in Combinatory Analysis 307

Table II: Ar

A, =2 cos (S'-,
J4 =2cos(^«n-~-^r).
A5 = 2 cos(fftTT - i?r) + 2 cos fair.

Ar = 2 COS (fair - -^tt) + 2 cos (fair - r) + 2 cos ($«ir + fair).

ig =2 COS (£?i7T -^(jTr) + 2 COS (^Wff —
An -2 COS (fair — J£jt) + 2 COS (fair - :iW) + 2 COS ($71 jr + r,W).

= 2 COS (^nir — £§ ir) 4- 2 COS (T\n7T - fvrir) + 2 COS r\?i7r - sV?) + 2 cos (*j\nn- -t^-tt)
4- 2 COS (^tt-TT + ^7r).
-4,2=2 COS (fair - f|-tt) + 2 COS (fair + -Mr).
{fair - i^Tr) 4- 2 COS (yfynr + ^tt)
+ 2 cos ? 0/1+2 cos (-[sHir + ^jw).
Au~2cos(fair-\fa) + 2cos(fair-^n ( r-fV).
A .5 = 2 COS (jfifWjr - SV0 + 2 cos (ftwr - Vgrr + 2cos(^„r+*,r).
-<w-2 cos a»T + if«)+2cos (9»«- + IW+8 ( + e*p+&»).
-4,-=2 cos (W»W +#*) + 2 cos (£»* - tV)+2 cos ) + 2cosyV»*
+ 2 COS QS-lrr - frir) + 2 COS (}?7Wr - fi. -nr) - 2 CUB - -fair) + 2 COS (\$nir + ^tt).
-1,8 = 2 COS ({ -nr + tf*) + 2 cos (|-nr - *-) + (
It may be observed that
is 1, 2 (mod 5)), = 0 (nal, 3, 4 (mod 7)),
i10 = 0(wsl, 2 (mod 5)), = 0 (nst, 2, 3, 5, 7 (mod 11)),
is0(na2, 3, 5, 7, 9, 10 (mod 13)), =0(71 = 1, 3, 4 (mod 7)),
5=0(71=0 (mod 2)), =0 (71 = 1, 3, 4, 6, 7, 9, 13, 14 (mod 17));
while 4x, A2, Aa, A4, xlfl, J8, Aa, Av2, -4,ft never vanish.

Table III: logwj)(») = ^{V(» + 10)-a»}.

3000
308 Asymptotic Formula in Combinatory Analysis
Asymptotic Formulas in Comlinatory Analysis 309

Slillllillllllllil!!-
 37
ON THE COEFFICIENTS IN THE EXPANSIONS
OF CERTAIN MODULAR FUNCTIONS
(Proceedings of the Royal Society, A, xcv, 1919, 144—155)

1. A very large proportion of the most interesting arithmetical functions

—of the functions, for example, which occur in the theory of partitions, the
theory of the divisors of numbers, or the theory of the representation of
numbers by sums of squares—occur as the coefficients in the expansions of
elliptic modular functions in powers of the variable q = e"T. All of these
functions have a restricted region of existence, the unit circle | q j = 1 being
a “natural boundary or line of essential singularities. The most important
of them, such as the functions*
(l-i) foM-A=</-«!- if) (i-T) ..r.
(1-2) (0) = 1 + 2g + 2<2J + 2r/+ ....

(1-3) ' 12g)V;=l + 240 (^t + i|2l1 + ...),

(!•*) 216 (!)" cf$ — 1 504 + *SL +

are regular inside the unit circle; and many, such as the functions (IT) and
(T2), have the additional property of having no zeros inside the circle, so that .
their reciprocals are also regular.
In a series of recent papers! we Have applied a new method to the study
of these arithmetical functions. Our aim has been to express them as series
which exhibit explicitly their order of magnitude, and the genesis of their
irregular variations as n increases. Wo find, for example, for p («), the riamber
* -We follow, in general, the notation of Tannery and Hoik's Elements do la thlorie
des fonctions elliptiques. Tannery and Hoik, however, write 16<? in place of the more

+ (1) G. H. Hardy and S. Ramanujan, “Uue formule asymptotique pour le t)ombre

des partitions de n,” Comptes Rendus, January 2, 1917 [No. 31 of this volume];
(2) G. H. Hardy and S. Ramanujan, “Asymptotic Formula; in Combinatory Analysis,”
Proc. London Math. Soe., Ser. 2, Vol. xvn, 1918, pp. 75—115 [No. 36 of this volume];
(3) S. Ramanujan, “On Certain Trigonometrical Sums’and their Applications in the
Theory of Numbers,” Trans. Camb. PhiZ. Soe., Vol. jam, 1918, pp. 259-276 [No. 21 of this
volume]; (4) G. H. Hardy, “On the Expression of a Number as tbe Sum of any Number
of Squares, and in Particular of Five or Seven,” Proc. National Acad, of Sciences, Vol. iv,
1918, pp. 189—193: [and G. H. Hardy, “On the expression of a number as tbe sum of
any number of squares, and in particular of five,” Trans. American Math. Soe., Vol. xxi,
1920, pp. 255-284],
 On certain Modular Functions 311
of unrestricted partitions of n, and for -r6 (n), the number of representations
of 7i as the sum of an even number s of squares, the series

1 ± . hll! ±
2tr \/'2 dn \ \n J 2tt dn V A.„ t

nH-i [i-fc + 2 cos (J«7T - i&ir) 2-3* + 2 cos (J«tt - isir) 3“3S +

and our methods enable us to write down similar formulae for a very large
variety of other arithmetical functions.
The study of series such as (1-5) and (16) raises a number of interesting
problems, some of which appear to be exceedingly difficult. The first purpose
for which they are intended is that of obtaining approximations to the
functions with which they are associated. Sometimes they give also an exact
representation of the functions, and sometimes they do noN Thus the sum
of the series (T6) is equal to r, («) if s is 4, 6, or 8, but not in any other case.
The series (To) enables us, by stopping after an appropriate number of terms,
to find approximations to p (n) of quite startling accuracy; thus six terms of
the series give p (200) = 3972999029388, a "number of 13 figures, with an error
of 0004. But we have never been able to prove that the sum of the series is
p (n) exactly, nor even that it is convergent.
There is one class of series, of the same general character as (To) or (T6),
which lends itself to comparatively simple treatment. These series arise
when the generating modular function f[q) or <j> (t) satisfies an equation

<j> (t) = (a + br)njf> »

where n is a positive integer, and behaves, inside the unit circle, like a
rational function; that is to say, possesses no singularities but poles. The
simplest examples of such functions are the reciprocals of the functions (T3)
and (T4). The coefficients in their expansions are integral, but possess other¬
wise no particular arithmetical interest. The results, however, are very
remarkable from the point of view of approximation : and it is, in any case,
well worth while, in view of the many arithmetical applications of this type
of series, to study in detail any example in which the results can be obtained
by comparatively simple analysis.
We begin by proving a general theorem (Theorem 1) concerning the
expression of a modular function with poles as a series of partial fractions.
312 On certain Modular Functions

This series is (as appears in Theorem 2) a “ Poincare’s series ”: what our

theorem asserts is, in effect, that the sum of a certain Poincare’s series is the
only function which satisfies certain conditions. It would, no doubt, be
possible to obtain this result as a corollary from propositions in the general
theory of automorphie functions; but we thought it best to give an inde¬
pendent proof, which is interesting in itself and demands no knowledge of
this theory.
2. Theorem 1. Suppose that
(2-1) /(?)=/(e™) = <M t)
is regular for q — 0, has no singularities save poles within the unit circle, and
satisfies the functional equation

(2-2) *(T) = (a + &T)«*(£^) = (a + M"*(?>.

n being a positive integer and a, b, c, d any integers such that ad —be- 1. Then
(2-3) /(9) = XB,
where R is a residue of f{x)!(q — x)
at a pole of f(x), if | q i < 1; while if ) q | > 1 the sum of the series on the right-
hand side of (2-J) is zero.
The proof requires certain geometrical preliminaries.
3. The half-plane I (t) > 0, which corresponds to the inside of the unit
circle in the plane of g, is divided up, by the substitutions of the modular
group, into a series of triangles whose sides are arcs of circles and whose
angles are \ir, J-7T, and 0*. One of these, which is called the fundamental
polygon (P) f, has its vertices at the points p, p2, and i oo , where p = eiri, and
its sides are parts of the unit circle j t [ = 1 and the lines R (t) = + J.
Suppose that Pm is the “ Farcy’s series ” of order m, that is to say the
aggregate of the rational fractions between 0 and 1, whose denominators are
not greater than m, arranged in order of magnitude^, and that h'lk' and hjlc,
where 0 < h'lk' < hjk< 1, are two adjacent terms in the series. We shall
consider what regions in the T-plane correspond to P in the P-plane, when

^ T~T=£- W TJ^bV
Both of these substitutions belong to the modular group, since hk' — h'k = l.
The Doints too. 4. -4. in the T-nlane corresDond to hlk. Ih + ‘2h')l(le + 2k').
314 On certain Modular Functions

the length of Em is
 On certain Modular Functions

We now consider the integral

(4-1) '

where ) q \ < 1 and the contour of integration is HM*. By Cauchy’s Theorem,

the integral is equal to

where R is a residue of f(x)j(q-!c) at a pole of f(x) inside H„,f. To prove

our theorem, then, we have merely to shew that the integral (41) tends to

Let a,' and ta, he the left- arid right-hand parts of o>, and f/, and £ the
corresponding arcs of Hm. The length of w, is, as we have seen, less than
ior/W, and that of £j than 1-ir-jkk'. Further, we have, on f,,

Thus the contribution of J, to the integral is numerically less than

C/(lck'n+v), where G is independent of m; and the whole integral (41) is
numerically less than

where the summation extends to all pairs of adjacent terms of Fm.

When V is fixed and m>v, the number of terms of Fm whose denominators
are less than v is a function of v only, say N (v). If h/k is one of these, and
h'jk' is adjacent to it, k + b'>vi\, and so h' > m-v. Thus the terms of (42)
in which either k or k' is less than v contribute less than 8CH(v)j{m — v).
The remaining terms contribute less than
40,, 1 _i0
v» ~ W u" ’
Hence the sum (4‘3) is less than
6CN{v) W

and it is plain that, by choice of first v and then m, this may be made as small
as we please. Thus (41) tends to zero and the theorem is proved. It should
be observed that 2 R must, for the present at any rate, be interpreted as
meaning the limit of the sum of terms corresponding to poles inside we
have not established the absolute convergence of the series.
* Strictly speaking, f(x) is not defined at the points where Hm meets the unit circle,
and we should integrate round a path just inside and proceed to the limit. The point
is trivial, as f{x), in'virtue of the functional equation, tends to zero when we approach
a cusp of from inside.
t We suppose m large enough to ensure that :c=g lies inside JTm.
{ See our paper (2), loo. oil. [p. 290],
316 On certain Modular Functions

We supposed that no pole of <f> (t) lies on the boundary of P. This restric¬
tion, however, is in no way essential; if it is not satisfied, we have only to
select our “fundamental polygon” somewhat differently. The theorem is
consequently true independently of any such restriction.
So far we have supposed [y|<l. It is plain that, if jgr|> 1, the same
reasoning proves that
(4-3) ZR = 0.
5. Suppose in particular that <p (t) has one pole only, and that a simple
pole at r = a, with residue A. The complete system of poles is then given by

(5-1) «. T = a = (ad-bc^l).

If a and b are fixed, and (c, d) is one pair of solutions of ad-bo= 1, the com¬
plete system of solutions is (c +ma, d + mb), where m is an integer. To each
pair (a, b) correspond an infinity of poles in the plane of t; but these poles
correspond to two different poles only in the plane of q, viz.,
(5-2) q = + q = ± e«'a,
the positive and negative signs corresponding to even and odd values of m
respectively. It,is to be observed, moreover, that different pairs (a, b) may
give rise to the same pole q.
The residue of <f> (t) for t = a is, in virtue of the functional equation (2-2),

(ct + ba)n+2'
and the residue of f(q) for q = q is
A fdq\ TriAq
(a + ba)n+1 \dr)r=B, (a + ba)n+'‘'
Thus the sum of the terms of our series which cc
(5-2) is
iriA r q q \ _ '2mA
(a + ba)n+1 - q q + qJ ~ (a + ba)n+21
We thus obtain :
Theorem 2. If <f> (t) has one pole only in P, viz., a simple pole a.
with residue A, and \ q \ < 1, then

w " (a + ba)n+1 cf - q2 ’
fc + da\ .

c, d being any pair of solutions of ad — bc = 1, and the summation extending

over all pairs a, b, which give nse to distinct values of q. If [ q ] > 1, the sum
of the series on the right-hand side of (5'3) is zero.
 On certain Modular Functions 317
If <fi (t) has several poles in P, f (q), of course, will be the sum of a number
of series such as (53). Incidentally, we may observe that it now appears that
the series in question are absolutely convergent.
6. As an example, we select the function

/(?) “ - VI1 -504! i-^)

say, where x = q-. It is evident that pn is always an integer; the values of
the first 13 coefficients are
# = 1, pi = 504, #, = 270648, # = 144912096,
# = 77599626552, # = 41553943041744, # = 22251789971649504,
# = 11915647845248387520, p, = 6380729991419236488504,
# = 3416827666558895485479576,
#0 = 1829682703808504464920468048,
pn =979779820147442370107345764512,
= 524663917940510191509934144603104;
that pu is a number of 33 figures.
By means of the formulte*
9> = fr W (1 + W (1 - 2i-),

«. ■-% = (OH ~ (0)F,

we find that J (1 + fc) (1 - |7c2) (1 - 2lc-).

• The value of n is 6. The poles of f(q) correspond to the values of t which

make K = ki equal to — 1, 2, or £. It is easily verified^ that these values are
given by the general formula

(6-2) q = exp *i) - exp vri- ^p) •

The value of a is it. In order to determine A we observe that

q 1(1—2a)s (l -yr
* All the formulie which we "quote are given in Tannery and Molk’s Tables; see in
particular Tables XXXVI (3), LXXI (3), XCVI, CX (3).
t A full account of the problem of finding r when k is given will be found in Tannery
and Molk, loo. cit., Vri. in, oh. 7 (“ On donne £-> 0u g,, m; trouver t on *,, s>3 ”).
I It will be observed that in this case a is on the boundary of P; see the concluding
remarks of § 4. As it happens, r = i lies on that edge of P (the circular edge) which was
not used in the eoustruction of so that our analysis is applicable as it stands.
 318 On certain Modular Functions

The series in curly brackets is the function called by Ramanujan* <J>1|0 andf
1008(1\, = Q‘-PR,

where P= , Q = 12j'. ^) , E = 216</3^^ .

Here. R — 0, so that
1008*,,, = Q1 = 1 + 480®0,:t = 1 + 430 + *31 + ... j .

Hence we find that A = i/trC, 2iriA = — 2/G,

(6-3) c=1 + 4SOfj5rri + plri + ---)-

Another expression for C is

(6-4) G = 144©3,

(6-41) (ir d6 {ftW

'J0 V(l~Jsin20) 4V-7T '
We have still to consider more closely the values of a and b, over which
the summation is effected. Let us fix h, and suppose that (a, b) is a pair of
positive solutions of the equation a? + = 1c. This pair gives rise to a system
of eight solutions, viz.,
{±a, ±6), (±b,±a).
But it is obvious that, if we change the signs of both a and b, we do not
affect the aggregate of values of a. Thus we need only consider the four pairs
(a, b), (a, - b), {b, a), (b, - a).
If a or b is zero, or if a = b, these four pairs reduce to two.

It is easily verified that, if (a, b) leads to the pair of poles

, (ac + bd . nr \
S-±q-±exp(7fTjr«-—

then (a, — b) and (b, a) each lead to q = ± q, where q-is the conjugate of q.
Thus, in general (a, b) and the solutions derived from it lead to four distinct
poles, viz., ± q and ± q. These four reduce to two in two cases, when q is
real, so that q = q, and when q is purely imaginary, so that q = - q. It is

* S. Ramanujan, “On Certain Arithmetical Functions,” Trans. 'Camb. Phil. Soc., Vol.
xxn, pp. 159—184 (p. 163) [No. 18 of this volume, p. 140].
t Ramanujan, loc. at., p. 164 [p. 142].
1 Ramanujan, loc. cit., p. 163 [p. 141}
 On certain Modular Functions 319

easy to see that the first ease can occur only when k = 1, and the second

If Ic = 1 we take a = 1, b = 0, c = 0, q = q = <T'. If/,- = 2 we

d — 1; and q = ■— q = ieri”. The corresponding terms
re a = 1, 6 = 1, c = 0, d—

If k is greater than 2, and is the sum of two coprime squares a? and b", it
gives rise to terms

1 -(?/'q? + l-(?/q)2'
There is, of course, a similar pair of terms corresponding to every other
distinct representation of k as a sum of coprime squares. Thus finally we
obtain the following result:

Theorem 3. If-

/(«) = -l/(l-504 2 j^L) = Ipnr,

and | q \ < 1, then

( l _ i_ l_ii.
+ " l(a + bif 1 — (q/qj2 + (a - 6ij8 X - (?/q)2) ’

where (7= 1 + 480 (^-j + + ■■■) = 16 {r (})}■«'

q=exP (SS «) =exp GSf ^ - ^2) ’

and 4 is the conjugate of q. The summation applies to every pair of coprime
positive numbers a and b, such that le= a? + b‘^5, such pairs, howevec, only
being counted as distinct if they correspond to independent representations of
Ic as a sum of squares. If | q | > 1, then the sum of the seines on the right-hand
side of (6’5) is zero.
* When a and b are given, we can always choose c and d so that |ac+M|$£(a2+&2).
If q is real, we have ad-be = 1 and «c+M=0 simultaneously: whence
(«3+&2) (c2 -(- o/2) = 1.

’ ad—bc= 1, 2|ac+M|=a!+i2,
■ whence (c2+c/2)2=(|<zc+Mj-c2-</2)2+l.
This is possible only if c2+cP = 1 and |ac + M| = l, whence a!+i>2=2.
320 On certain Modular Functions

7. It follows that

(7-D iOPll = ^+ +

say. Here X. is the sum of two coprime squares, so that

where a2 is 0 or 1 and 5, 13, 17, ... are the primes of the form 4k+ 1 ; and
the first few values of cA (») are
*(») = !. W = (-D". °r> in)= 2 cos (|>i7r + 8 are tan 2),
0,. (li) = 2 cos (*jwr - 8 arc tan 2), e„ (») = 2 cos (*§n,r + 8 arc tan 5).
The approximations to the coefficients given by the formula (7T) are
exceedingly remarkable. Dividing by \C, and taking n= 0,1, 2, 3, 6, and 12,
we find the following results:
(0) 0-944 (1) 505-361 (2) 270616406
+ 0-059 -1-365 + 31-585
+ 0004 + 0-009
pa = 1000 Pl = 504000 ' p.2 = 270648-000
(3) 144912827002 (6) 22251789962592450-237
- 730-900 + 9057051-688
-0101 + 2081
-0001 - 0 006
pa= 144912096-000 p6 = 22251789971649504000
(12)
524663917940510190119197271938395-329
+1390736872662028-140
+ 2680-418
+ 0130
-0-014
_-0-003
= 524663917940510191509934144603104-000
An alternative expression for C is *
<7=96.r*A'{(1-a-)(1-r*1
by means of which 0 may be calculated with great accuracy*. To five places
we have 2/(7 = 0-94373, which is very nearly equal to 352/373 = 0-94370.

* Gauss, Warte, Vol. nr, pp. 418—419, gives the values of various powers of e~* to a
large number of figures.
 On certain Modular Functions 321
It is easy to see directly that pn lies between the coefficients of .*'■ in tilt-
expansions of
1 1-7-5®
(1 - 535®) (1 + 31®) ’ (1 - 535-5®) (1 + 24®) ’
and so that
(535)”+1 - (- 31)n+1 , 352 (535-5)" + 21 (- 24)"
566 " 373

The function Q(») = «=■>'/*-

has very remarkable properties. It is an integral function of a, whose maxi¬

mum modulus is less than a constant multiple of e2"’1*1. It is equal to pn, an
integer, when x = n> a positive integer; and to zero when x*= — n. But we
must reserve the discussion of these peculiarities for some other occasion.
 QUESTIONS AND SOLUTIONS

(published by Srinivasa Ramanujan in the Journal of the Indian

Mathematical Society) -
Question 260 (in, 4.3 ; communicated by P. V. Seshu Aiyar):
Shew, without using calculus, that
^log2 = l+p=-j + p^ + i^—+ ad inf.
[Solution by K. J. Sanj&ua and S. Narayanan, in, 86—87.]
Question 261 (tit, 43; communicated by P. V. Seshu. Aiyar):
Shew that
(«) (1+l;.)(1 + ^)(1+^)- =^oSh(W3),
(b) ^_2^1-i)(l-^...=LcoSh(|wN/3);
and prove from first principles that (b) = \ (a).
r [Solution by N, B. Fendse, m, 124—125.]
Question 283 (nr, 89, and iv, 106):
Shew that it is possible to solve the equations
.v+y + z=a, px±qy+rz=b,
p-x + q\y+r2z=c, p^x+q^y+i-^z^d,
phv + fy+r^e, p^ + fy+r^f,
where a-, y, z, p, q, r are the unknowns. Solve the above when «=2, 6=3, c = 4, d=8,
e=12, and/=32.
[Solutions (1) by M. K. Sudaniva Aiyar, and (2) by Sankara Aiyar, in, 198—200. See
also “Note on a Set of Simultaneous Equations," Journal of the Indian. Mathematical
Society, IV (1912), 94—96, and pp. 18—19 of this volume.]
Question 284 (m, 89):

SolTC 1>-
Solution by Srinivasa Ramanujan, iv, 183.
Let x=a + l3 + y, y^a/3 + fly + ya,

(a+8+y)6-6=5 {(a+S+yy-(o3+fly + ya)}{(a+^ + y)(aS+^+ya)-aSyU

(a3+3y+yaf-9=»{(a3+3y3-yar-a3y(a+S+y)}{(a+S+y)(aSf^y+ya)aSy-a^y}.
Simplifying, • a'+^'+y^e,
(affi« + (/3yy + (ye)*=9,
(^y)==l.
 Questions and Solutions
Thus wo see that ar\ y5 are the roots of the equation

But it is easily seen that the roots of this equation arc

4 cos3 20\ 4 cos2 40\ 4 cos2 SO".

««(2cos20’)*, 0=(2 cos 40")>', y=(2cos80’)s,

*-(2 cos 20’)*+(2 cos 40”)® + (2 cos 80’)*,
y=(J sec20”)*+(|seo 40“)* + (isec80’)*.
All the values of x are
a+,8 + y, a + |3p+yp4, a+£p2 + yp3, ap+8p + yp3,
ap+3p2+yp2, np2+/3p4+yp4, ap3+/3p3+yp4,
where p is art imaginary fifth root of unity. Similarly for y.
Question 289 (iii, 90):
Find the value of (i) ^/[l + -2^/{l + 3vi(l + ...)}],
(ii) 46+2 47+ 3 48+ ...)}].

Solution by Srinivasa Ramanujan, IV, 226.

(i) »(» + 2) = »41+(« + l) (a+ 3)}.
Let n(« + 2)=/(it);

/(n)-»41 +/(« + l)}

= »^{l + (n + l)41+/(K+2))}
= n 41 + (« +1) VU ■+ (»+2) 41 +/(» + 3))}]

that is, «(*+2) = » 41 + (» +1) 41 + (a + 2) 41 ■+ (»+3) 4 + •

Putting n = l, we have
41+241+3 41 + ...))] = 3.
(ii) In a similar manner,
»(n + 3) - # 4»+5 + (»+1) (a + 4)}.
Supposing f{n) = n (a+ 3),

/(»)-n4n+6 + /(« + l)}

= n 4» + 5 + (11 +1) 4» + 8 +/ (,l + 2)>]

t (n + 3)=n 4« + S + (» +1) 4* + 6 + (■» + 2) 4>l + 7 +.

46 +2 47 +3 48+4 49 + ...))}]= 4.

Question 294 (tn, 128):

Shew that [if x is a positive integer]

where 0 lies between i and

324 Questions and Solutions

Therefore, when x = x,
2/=Lim ttx)4-1 - j (/-^(i^+terms containing x~^, ar1, ete.)tfe|
= l-5 = Ji
since, by Stirling’s theorem,

Alsoy=J when ® = 0.
Though the limits of y are known to be £ and -1 when ar is 0 and co respectively, yet it
is difficult to prove that y lies between \ and J.
The expansion of y in ascending powers of x~l (which can he found from the above

This result is important as it helps to express a* between rational limits.

Question 295 (in, 128): ,

If 0/9= rr, shew that
 Questions and Solutions 325

,(h

[See also “Some definite integrals,” Messenger of Mathematics, :v (1915), 10-18, and
pp. 53—58 of this volume, equation (12).]
Question 308 (hi, 1G8):
Shew that
(i) jJ 0 cot0\ogsin8d0~J(log2)2,

(ii) J I tai^“ d;c= ~ log2.

[Solution by Iv. J. Sanjana,' r, 248.]
Question 327 (hi, 209):
Shew that Euler’s constant, namely [the limit of]
1 + 2 + J + ... + --logw
when n is infinite, is equal to

the first term in the «th group being

2
{i<3» + 3 )}3-£<3“ + 3)-
[See “ A. series for Euler’s constant y,n Messenger of Mathematics, xlvi (1917), 73—80,
and pp. 163—168 of this volume.]

Shew that
C) yz rrTT!T-=WS(s+',6)!^W5+1)]'*T"

Question 353 (iv, 4(5):

If n is any positive odd integer, shew that
326 Questions and Solutions

and hence prove that

i,r=(coshS+cos£) 1-KC0SbS+C0SS) , + }(C0sll^+CO3S)

for all odd values of n.
[.Solution by A. C. L. Wilkinson, vm, 106—110 ; also Remarks by S. D. S. C ho via, xvi,
119—120.]
Question SS8 (iv, 78):
If n is a multiple of 4, excluding zero, shew that
l—1 sech (i ir) - 3"-1 sech (f v) + S»-> sech (£ *•) - ... = 0.
[Solution by M. Bhimasena Bao, vu, 99—101.]

Question 359 (iv, 78):

If sill(.■» + ,;/) = 2sin (i(x-y)), sin (y+z) = 2sin (J(y-6),
prove that (Isin.c cos zn + (-) cos x sin zp = (sin %)T'h
and verify the result when
sin %x = (v/5 - -2)3 (4 + v/15)!, sin 2y = Jo - 2, sin 2s=(v/5 - 2)3 (4 - v/15)2.
[Solution by T. E. Srinivasa Iyer, xv, 114—117.]

Question 386 (nf, 120):

Shew that

jo (1 +.*>) (1 + n‘.v‘) (1 +»**=) (1 + n°x‘)... = 2(i+ir+i> + »» + a>»+...) ’

where the indices in the denominator are the sums of the natural numbers.
[Solution by N. Durai Bajan, vu, 143—144. See also “Some definite integrals,”
Messenger of Mathematics, XLIV (1916), 10—18, and pp. 63—58 of this volume, equation (24).]
Question 387 (iv, 120):

Shew that + lsirz\ + -“24“

Question 427 (IVJ23S):
Express . (Axt-kBxy+Cy1) (Ap^+Byg+Cq1)
in the form Aui+Buv+ Co2; and hence shew that, if
(2.v2+3isy+5ys)(2p2+3« + 59'2) = 2Ks+3i»+5tia,
then one set of the values of u and v is
u^i(x+y)(p + q)-2xp, v=2qy-(x+y)(p + q).
[Solutions by (1) “Zero,” and (2) S. Narayanan, x, 320—jgl.]
Question 441 (v, 39):
Shew that
(6a2—4a&4-4&2)3=(3a2+Sab - 662)3 + (4a2 - 4a& + 6&2)3 + (5a2 - Sab - 362)3,
and find other quadratic expressions satisfying similar relations.
[Solution by S. Narayanan, vi, 226.]
 Questions and Solutions 327
Question 463 (v, 120):

' If =
then /„ «-5I + * (?) \f(%) ■
Find <fi (n), and hence shew that
0(0)-*, 0(| w)=i, 0(«-)-?.^?,
0(2”-) = 7s. 0(k)=O-
[See “Some definite integrals connected with Gauss's sums," Jrevenger of Xu thematic*,
xuv (1915), 75—85, and pp. 59—67 of this volume.]
Question 464 (v, 120):
2"- 7 is a perfect square for the values 3, 4, 5, 7, 15 of n. Find other values.
[Solution by K. J. Sanjana and T. P. Tvivedi, v, 227—228.]
Question 469 (v, 159):
The number l + »! is a perfect square for the values 4, 5, 7 of n. Find other values.
[Comment by M. Ehimasena Ran, xv, 97.]
Question 489 (v, 200) :
Shew that '
(1 pe-v"5)(1 + «-»"'«)(1 L±y(?f 2 e-*vVSS.
[Solution by A. C. L. Wilkinson, VII, 104. See also “ Modular equations and approxima¬
tions to Quarterly Journal of Mathematics, xtv (1914), 350—372, and pp. 23-39 of
this volume.]
Question 607 (v, 240):
Solve completely s2—y + a} y2=z+a, z-=x + ci;
and hence shew that
(a) J[8 - V{8 + */(8 -...)}]■= 1 + 3 v/3 sin 20°,
(fi) v/[ll-2^/{ll4-2v'(ll-...)}] = l+4sinlOaJ
'* (c) s/[23 - 2 ^{23 + 2 v/(23 + 2 v/23 -...)}] = 1 +4 sin 20°.
Solution by Srinivasa Ramanujan, vr, 74—77.
We have .r=- a=(y* - af - <*={(*» - af- af - a,
or *» - 4M>a + 2x> (3a« - a) - 4x°- (a* - a?) - a- + a* - 2a3 + «z - a = 0.(1)
Evidently y and a also are roots of this equation. But * may be written as
J[a+y)=J{a + J(a+*)}=J[a+J{a+J{a+X)}]
-*/[»+^{aW(« + ...)}W(«+*)-
Hence we have ^ - ,r - a=0; .(2)
therefore a?-x-a nfhst bo a factor of the expression in (1). Now, dividing (1) by
**-*-«, we have
s*+M‘+a*{l-%a)+s?(\-‘la)+x'‘(\-Sa+‘Sa?)+x{l-1a + dl)
+ (l-a+2a2-a3)=,o.(3)
328 Questions and Solutions
Let a, ft y, a', ft, y‘ be the roots of this equation. Then, since y and s are also roots, we

p2=u+y, ji’2=a + y,
y2=a + a, y'2 = a + o'.
Let a+/3+y = w, a' + {3' +y =v;
then we see that a2+/32+y3=3« + ?<,
thatis a/3+/3y+ya=^(«2-M-3a).
Again we have cr/3=«/3+/32, ya? = ay+fry,

y3a=«a+a3, 0/^a/S+tf;
adding up all the six results we have
2a2 (/$ + y) = %a(a+{i + y)+a?+P+yi+a.3+py+ya-,
i. e. (a+/3 + y) (aj3-f&y+ya) - 3a|3y = 2a (a + /3+y) + (a2 +£2 +y‘2) + (a/9 -f /9y + ya),

Hence we have 6a&y=v?-2u--'7au-tt- 3a.

Similarly we have a'+/3' + y'=v, a’fi+&'y+y'a' = b (y2- y-3a),

Hence the sextic in (3) is identical with

{x3-x2u + barfyfi - u - 3a) - £ (u3 - 2u- - 7au ~~u~ 3«)}
{x*-;cH + (tf3-v-3a)- J(v3-2i>2 - 7av -v- 3a)}.(4)
Now, equating the coefficients of & and x5 in (3) and (4), we have

«s+t>* - 2 («*+«*)-(7a+I) (m + i>) —6a+3{«»(it+v)-2MV — 3a (w+v)} = 6(2a-l)

Substituting for u + v in the above result, we have

hence - i {l ■+ V(4a - 7)}, v = - *{1 “ a'(4« - 7)}.(5)

These values of u and v, when substituted in (4), reduce the sextic equation (3) to the
two cubics:
a»+i«*{l+%/(4»-7)}-^{2a+lW(4«-7)>+i{a-8+oN/(4a-7)}=0, ^
^+ir3{l-V(4a-7)}-^{2a+l-V(4«-7)}+i{a-2-av/(4a-7)}===0,
which can he solved by the usual methods.
In the numerical examples proposed, the combinations of the signs plus and minus
may be determined by proceeding as follows:
1+ 2 V3 sin 20°=V(l + 4 J3 sin 20° +12 sin2 20°)
= sj{7+4 V3 sin 20° - 6 cos 40°)
=V(7+4 V3 sin 20° - 4 ,/3 cos 30° cos 40°)
= ^(7 +4 */3 sin 20°-2 JZ cos 70°-2 JZ cos 10°)
= n'(7+2 >J3 cos 70° - 2 s/3 cos 10°)
=V(7 —4^/3sin30“ sin 40°)
— */{8 -*(1-4-2 s/3 sin 40°)}
=V[S-V{8 + (2V3sin80e-l)}]
- 48 ■- J{8 + V(8 - (1 + 2 V3 sin 20”))}].
 Questions and Solutions 329
In a similar manner we have
1 +4 sin 10°=411-2(1 +4 sin SO*)}
= 411-2 411 + 2 (4 sin 70” - 1)}]
=N'[11-2411+ 2411-2(1+48^ 107;;];
and also 1+4 43 sin 20” =423 - 2 (4 4! sin 80° - l)j-
= 7[23 - 2 4 [23 + 2 (1 + 4 43 si n 40”)}]
• = 423 - 2 423 + 2 423 + 2 (1 +4 43 sin 20=));].
Question 524 (vi, 30):
Shew that (i) 4cosiv)+4cosiw)+=/(cosS^)=4i(5-34r)},
(ii) “/(cos %w) + sj(cos tv)+4(cos g4=44 (3 41 - 6)}.
[Solution by N. Sankara Aiyar, vi, 100—191.]
Question 525 (vi, 39):
Shew how to find the square roots of surds of the form + 4#, and hence prove that
(i) 4^-(/4)=4(=/2+“/20-4/25),
(ii) 4418 - zm)=j (v/98 - zm -1).
[Solution by N. Sankara Aiyar, vi, 191—192.]
Question 526 (vi, 39):
If n is positive shew that

» ? ,1 +1 + (a + 2)S + 4 + 3)“ + (it + 4)* + (« + S')*'+''';

and find approximately the difference when n is great. Hence shew that

1001 + lot® + 10033 + 1004* + 1005*+'''

is less than y^qq % aPPros*niately 10-440.
Question 541 (vi, 79):
Prove that

1+i.3 + 1.3.D'
[Solutions, with remarks, by K. B. Madhava, vm, 17—20.]
Question 546 (vi, 80):

<*> (r i)+*G ■- i)+K Q~?}+ - "l0S('2W3)’

(h) i-|+|^-- -K-i{iog(i+VW.
[Solution by M. Bhimaseua Rao, vn, 107—109; also K. J. Sanjana, “Notes on Questions
546, 572, 573, 683, 585, 606,” VII, 136—141.]
Question 571 (vi, 160):
If }va =log tan {^tt (1 +6)}

[Solution by J. C. Swaminarayan, R. J. Poeock, A^Narasinga Rao and K. B. Madhava,

VII, 32. See also “On the Integral j* ttt° — dt,” Journal of the Indian Mathematical
Society, vn (1915), 93—96, and pp. 40—43 of this volume, equation (17).]
330 Questions and Solutions
Question 584 (vr, 19!

Question 605 (vr, 239):

Question 606 (vi, 239):

Shew that

{Solution by X. Sankara Aiyar, vir, 192; also the note by K. J. Sanjana quoted with
reference to Question 546.]
Question 629 (vii, 40):

{Solutions by (1) K. B. Madhava, (2) N. Durai Eajan. and (3) M. Bhimasena Bao,
Tin, 25—30.]
Question 642 (vii, 80):
Shew that
(i) J.(1+HK... + 2lrl_)2^ = i^, ,
 Questions and Solutions 831
Question 661 (vn, 119):
Solve in integers
and deduce the following:
63- 53— 3s = 2« 83+ 63 + l3= 3%
' 123- 103+ l3= 3Bj 463 — 37a —33= 6->,
1743 + 1333 — 4o3 = 14e, U883 — 5093 - 33 = 34,:.
[Solutions by N. B. Mitra, xm, 15—17. Additional solution and remarks hy X. B.
Mitra, xiv, 73—77.]
Question 662 (vn, 119—120):
Let AB be a diameter and BO be a chord of a circle ABV. Bisect the minor arc BO at
with A and B as centres and AM and BN as radii, cutting each other at B and >$", and
cutting the given circle again at the points W and N' respectively. Join AN and BM
intersecting at R, and also join AN' and BM‘ intersecting at R'. Through B draw a
tangent to the given circle, meeting AM and AM' produced at Q aud (/ respectively.
Produce AN and M'B to meet at P, and also produce AN' and MB to meet at P\ Shew
that the eight points P, Q, R, S, 6", R\ Q1, P' are cyclic, and that the circle passing
through these eight points is orthogonal to the given circle ABC.
Question 666 (vn, 120):
Solve in positive rational numbers
Foresample: x = 4, y=2; #=3§, y=2£.
[SoMcm by J. C. Swaminarayan and R. Vythynathaswamy, vm, 31.]
Question 681 (vn, 160):
Solve in integers x3+y*+zi= 1,
and deduce the following:
63+ 83= 93-1, 93 + 103= 123 + 1, 1353+ 1383= 1723-1,
7913 + 8123=10103 — 1, 111613+114683 = 142583 + 1, 656013 + 674023 = 838023+1.
[Pcivtial solution by N. B. Mitra, xm, 17. See also N. B. Mitra, xxv, 7*?—77 (*6—77).]
Question 682 (vn, 160):
SJaew how to find the cube roots of surds of the form A + tyB; and deduce that
4/(3/2-D=^-)-^(§)+^a).
[Solution by “Zero,” x,325.]
Question 699 (vn, 199);
Shew that the roots of the equations
(i) — 1=0,
(ii) ^+^-^-^-.*+1=0
can be expressed in terms of radicals.
Question 700 (vn, 199):

(a+6 + l)(|) +&+ +(« + & +5) } + -to?lterms-

[.Solutions by (1) K. E. Kama Aiyar, (-2) K. Appukuttau Erady, Till, 162.]
Questions and Solutions
 Questions and Solutions
Question 753 (vm, 80):
if
where [*] denotes the greatest integer ir ew that

[Solutions by (1) N. Durairajan, (2) K. B. Madhava, X, 395—397.]

Question 754 (vm, 80):
Shew that wl T (1 +*)«(M+'to*+*+r/*,
where £ lies Between T()i; and PG for all positive values of x.
[Partial solution by K. B. iladhava, xn, 101. See also remarks by E. H. Neville and
0. ICrishnamachari, xm, 151.]
Question 755 (vm, 80):
Let p he the perimeter and e the eccentricity of au ellipse whose centre is C, and let
CA and CB be a semi-major and a semi-minor axis. From CA cut off C'Q ecpial to CB,
and also produce AC to 1‘ making CP equal to CB. From A draw .Id’ perpendicular to
CA (in the direction of CB). From Q draw QM making with QA an angle equal to 0
(which is to be determined) and meeting .1 d' at J/. Join Pitf and draw PS making with
PM an angle equal to half of the angle A PM, and meeting .Id'at d". With P as centre
and PA as radius describe a circle, cutting PS at K, and meeting PB produced at L.
Then, if

trace the changes in 0 when e varies from 0 to 1. In particular, shew that 0 = 30° when

about 30° 44J' when e is about 0-9589.

Question 768 (vnr, 119;:
If ,Kr) = (a! + 2)/(.rS+.r+l),
shew that (i) (**) + ...=(log*)-‘+(l-*)->
for all positive values of x; and that
(h) (**)+i * (**)+&+<**)+...-a
for all negative values of x.
r [Solution, with correction, by K. J. Sanjana and N. Durairajan, VIII, 227.]
Question 769 (vm, 120):

l0g 2 (2 log 2 log 4 + 3 log 3 lo.

1
+ 21og2 31og 3t4 log 4 51og5 + " = log 2 *
[Solution, with correction, by K. B. Madhava, M. K. Kewalramat N. Durairajan, and
S. V. Venkatachala Aiyar, IX, 120—121.]
Question 770 (vm, 120):
If 8„ denote the number of divisors of n (e.g., 8, = 1, 82=2, S3=2,,
(i) 81-JS3+i85-|8; + J89-...

(ii) 8I-p2+J83-i84 + 485-...

334 Questions and Solutions

Question 785 (via, 159—160):

Shew that [3{(«*+*»)»-a){(o.3+i-')!-irf = (« + bf -(a*-ab+ b'Q.
This is analogous to
[2 {(a3 + &») i - a) {(a2 + V-f - b}f=a + b - (a2 + lfi)K
[Solution by lv. K. Ranganatlm Aiyar, E. D. Karve, G. A. Ilamtekar, L. N. Datta
and L. N. Subramauyam, vm, 232.]
Question 1049 (xt, 120):
... sin nxdx ./() A)
Shew that (l) j -j—z—o“— = -j—o'” >

Question 1070 (xi, 160):

[Solution by S. D. Chowla, N. B. Mitra, and S. V. Venkataraya Sastri, xrr, 122—1-23.]

Question 1076 (xi, 199): '
Shew that (i) - (7^20-19)J=4/|-^l;
(ii) =
 APPENDIX I: NOTES ON THE PAPERS

1.

P-M*
P. 4, 1. 4. (12) is true if n is odd and greater than I, and the last terms are given in the

P. 8, 31. 7—9. The proof is invalid as it stands, since the series in the integrand is
divergent when x£n. The same remark applies to p. 9,11. 6—8.
P. 9,1. 2 from below. The coefficients, if integral, are necessarily odd.
P. 10, (32). Apply (31) to the well-known formula

where if- (2) is the derivative of log r

P. 11,1. 2 from below. N and D are the numerator and denominator of Bui.
The paper is interesting as a specimen of Ramanujan's earliest manner, hut the
principal results are well known and the proofs are incomplete. Thus (20) of § 8 is simply
the theorem of von Staudt and Clausen, and (16), which is a consequence of (20), was
known to Euler: see P. Bachmann, Kiedere Zahlentheorie, vol. 11, pp. 4.3—51 and p. 54;
or N. Nielsen, Traite eUmentaire des nombres de Bernoulli, pp. 244—245 and 258.
Ramanujan states eight theorems, viz. (14)—(21), embodying arithmetical properties
of the Bn. Of these, proofs are indicated for three, viz. (16), (20), and (21); but the
theorems on which these proofs would depend, viz. (28) and the corresponding propositions
about the series (30) and (32), are never proved. Two other theorems, (17) and (18), are
stated to be corollaries of (16) and (14); and (14), (15), and (19) are stated merely as
conjectures.
4.
All the results of this note are naturally familiar in the elements of the analytic theory
of numbers: see Landau’s Handbueh der Lehre von der Yerleilung der Primza/dcn, especially
pp. 507 et seq., dealing with Mobius’ function p (n).
6.
This paper is of special interest because it embodies so much of Ramanujan’s early
Indian work. It was rewritten in England, the references to Weber inserted, and many
results given by Weber deleted, but all the work was done quite independently.
Professor L. J. Mordell writes :
“ It would be extremely interesting to know if and how much Ramanujan is indebted
to other writers. Some results are easily accessible in English books. See, for example,
Greenhill’s Elliptic Functions, chap, x, in particular pp. 330—339, for results and references
concerning complex multiplication; and Cayley’s Elliptic Functions, chap. vn. It is of
course possible that Jhese books were known to Ramanujan.
11 As Ramanujan says, his functions Gn and gn are x>ractically Weber’s/{V(-«)} and
fi{s!{-n)}. The properties of f(w) and fx (w) are given in Weber’s Lehrbuck der Algebra,
vol. m (1908), § 34, pp. 112—116. The modular equations, which are merely the algebraic
336 Appendix I

the other roots of th/iodiilar equation *»/(«•) are expressible very simply hy means of
a finite number of the values /(—^), "'here a, b, c, d are integers such that ad-bc=n.
If we put f(nw)=f(v>) in the equation it splits up into factors, the roots of each factor
being of the form f£ (a), where now D is a, function of n and a is a root, with positive
imaginary part, of one of the system of non-equivalent quadratic forms of determinant -■’I).
Each of these factors in turn splits up info simpler factors, whose roots are now restricted
to belong to a genus of quadratic forms of determinant -D. All this is developed in
chapters xvii—xx of Weber’s book. It appears from Ramanujan's 5; 4 that he was not
aware that the character of the surds for G?» and depends on the number of classes of
quadratic forms.
“Ramanujan’s method of approximating to ir by means of equations

where m is nearly an integer, is due to Hermite. See Ch. Hermite, ‘ Sur la theorie des
equations modulaires’ [Comptes Rendus, 16 May, 13 June, 20 June, 4 July, 18 July, and
25 July, 1859 (20 June), and CEuvres, vol. II, pp. 38—82 (pp. 60—61)]; L. Kronecker,
Berliner Monatsberichte, 1863, pp. 340—345; and H. J. S. Smith, ‘ Report on the theory
of numbers ’ [Report of the British Association, 1865, pp. 3-22—375, and Collected Papers,
vol. I, pp. 289—358 (p. 357)]. Hermite and Kronecker give some of Ramanujan’s approxi¬
mations, or results equivalent to them. On the other hand Ramanujan’s method of
obtaining purely algebraical approximations appears to be new.
“It is unfortunate that Ramanujan has not developed in detail the corresponding
theories referred to in § 14. His IC1 is

which is reducible to an elliptic integral. The result in § 15, that K can be expressed
in terms of r-functions when q is of the forms e~*n, e'”’**'2, e~'^n,,3 and n is rational, also
seems to be new. It may be deduced from the equation
2* (1 - f) (1 - q>)... = (2W)i •
For the left-hand side is Weber’s tj (w) (Weber, l.c., p. 85), and there is an algebraic
equation connecting r) (mi) and o (nut) when n is an integer. There is also an algebraic
equation connecting kp and its transformed value, and so one connecting K and its trans¬
formed valuo. Finally, K is known in terms of r-funetions when q is e~”, e~7r‘12, or e~”/3:
for these classic results see Whittaker and Watson, Modem Analysis (ed. 3, 1920), pp. 524—
526.

theory of the algebraic equations connecting A (wq, k>2) and A (nwu wf), see their Elliptisehe
Modulfunktionen, vol. n, pp. 62-82. See also pp. 117—159 for the theory of the modular
equation and for further references.”

7.
P. 40, (4). If R (x) > 0, we have

= log x + <l> 0 + ^lirn {log A - 0 (A)}.

 Notes on the Papers 337
If we write A for the last limit, then
0 (.*) = log * + <#> (1) + A,
and A may be determined by putting ,r= 1.
P. 40, (6). This equation should read

■ T5 + p-6S“^+95+*"“^(^2"l)+472lQ8(1+*/3)*
P. 42, § 4. Since -1 < t(x) < 1 in (15), the a’ of this section must in the first instance
lie (when real) between — £ and The results must then be extended by analytic
continuation, in which some care is necessary with the'many-valued functions.
P. 43, (26). The real parts of a and 0 must be positive.
The transcendent considered in this note is closely connected with the integrals

Jlog(l±f)f, J.vcotsAv, /logons^-,...,

whose properties have been investigated by a number of writers. See, for example
i’s Calcul integral, §§ 270 et seq.; Lobatsehewsky’s memoirs “Imagmiire
I!?!

metrie” and “Anwendung der imaginaren Geometrie auf einige Integrate” (German
ion by H. Liebmann, Leipzig, 1904); and L. J. Rogers, “ On function sum theorems
lected with the series Sn-***", Proc. London Math. Soe. (2), vol. iv (1906), pp. 169-189.
It appears from Ramanujan’s notebooks that he had found the values of

for the special values §, 2, $(^5-1),..., given by Bertrand, and Rogers’ result that
L (a-) +L(g)-L (ay) - L -L
is an elementary function, without knowledge of the work of these writers.

This is merely a preliminary account of a small part of No. 15.

9.
P. 48,1.17. The constant is of course the constant of the Euler-Maclaurin sum-formula,
which is in this case f (1 - r).
n.
P. 57, bottom. The reference is to G. H. Hardy, “ Proof of a formula of Mr Ramanujan
Messenger of Math., vol. xliv (1915), pp. 18—21. The proof given by Hardy depends on
the application of Cauchy’s Theorem to the integral

wher F{z) = {\+*) (1+6*) (1 + 6%)....

P 8, formulse after (24). See No. 6, text and comments in Appendix. The special
cases e naturally more difficult than the general formula.
3. p. 22
338 Appendix I

12.
P. 61, bottom. The theorem referred to is due to Lerch. See M. Lerch, “Sur un point
de In theorie des fonctions generatrices d’Abel ”, Acta Math., vol. Jtxvir (1903), pp. 339—351;
E. W. Hobson, “The fundamental lemma of the calculus of variations”, Proc. London
Math. Soc. (2), vol. xi (1911), pp. 17—28. Lerch proves that if f{x) is continuous and
/(»><to-0
for a=a+nfi (vi = 0, 1, 2,...), then f(x)=0. A substantially equivalent theorem is that
jA<‘f(tc)dx = 0 (»“0, 1, 2,...)
implies f(x) = 0. Hobson gives more general forms of this result. Ramanujan requires
only the special case of Lerch’s theorem in which the integral vanishes for all (sufficiently
large) positive values of a; this indeed was one of his favourite weapons.
P. 63, § 3. See A. L. Cauchy, Exerctces de mathematiques, vol. n (1827), pp. 141—156.
A great deal of calculation is suppressed in this paper, and the verification of par¬
ticular formulas i.s sometimes laborious. More general results were found independently by
L. J. Mordell, “The value of the definite integral j dt”, Quarterly Journal of
Math., vol. xlviii (1920), pp. 329-342.
14.
Ramanujan does not attempt to use his identities to obtain approximations to f (s), hut
the idea of splitting up £(«) into two parts is the same as that which underlies Hardy and
Littlewood’s “ approximate functional equation ”. See G. H. Hardy and J. E. Littlewood,
11 The zeros of Biemann’s Zeta-function on the critical line ”, Math. Zeitschrift, vol. X (] 921),
pp. 283—317; and “ The approximate functional equation in the theory of the Zeta-
function, with applications to the divisor problems of Dirichlet and Piltz ”, Proc. London
Math. Soc. (2), vol. xxi (1923), pp. 39-74.
This paper was followed in the Quarterly Journal by a short note by Hardy which we
ono repro uce. jg

P. 79,1. 6 from below. The identity attributed to Hardy was first proved by Vnronol:
seeG. H. Hardy, “On Diriohlet’s divisor problem”, Proc. London Math. Soc. (2),apl. xv
(1916), pp. 1—25 (p. 21). The result referred to immediately above, viz. that (in the
notation of 17) At (») is of order not lower than mi, was proved independently and almost
simultaneously by Hardy [f.c. supra, and “Sur le problkme des diviseurs de Dirichlet”,
Comptes Rendus, 10 May 1915] and Landau [“tlber die Gitterpunkte in einem Kreise
(Zweite Mitteilung)", Gottinger Nachrichten, 1915, pp. 161—171: see also “tfber die
Heckesche Ptmktionalgleichnng”, ibid., 1917, pp. 102—111 (footnote 3 to p. 102)]. For
further developments, see the notes to 17.
P. 80,1. 5. See also E. Landau, Sandbuoh der Lehre von der ferteilung der PrimeahUn,
§ 60, pp. 219—222.
P. 85,1. 8 from below. The argument beginning “How, if if is a function of t,..." is
correct in principle but expressed in a puzzling manner. Write ..

d (17) £ (l +-7^) ,
where A is constant, and allow t to vary iu a range (T, log E), where T is sufficiently large
but fixed. It is then easily verified that the right-hand side is greatest when t=log N.
340 Appendix I
P. 134, footnote t. See G. H. Hard}', “ On Dirichlet’s divisor problem”, Proc. London
Math. Soc. (2), vol. sv (1916), pp. 1—23. We take this opportunity of observing that (as
was pointed out to us by Professor E. Landau) Hardy states there rather more than is
justilied. His argument proves that A, (n) is sometimes greater than K (n log»)i log log n,
and so justifies Ramanujan's statement, but it fails to prove that Aj («) is sometimes less
than - K (n log n)k log log n. For further results in this direction see G. Szego und A.
Walfisz, “tlber das Piltzselie Teilerproblem in algebraisohen Zahlkorpern”, Math.
Zeilschrift, vol. xxvi (1927), pp. 138—136 and 467—486.

18 .
Many of the results conjectured by Ramanujan in the latter part of this paper wore
afterwards proved by Mordell: see L. J. Mordell, “On Air Ramanujan’s empirical
expansions of modular functions”, Proc. Cumb. Phil. Soc., vol. sis (1919), pp. 117—124.
In particular Mordell proves Ramanujan’s results (101), (103), (118), (123), (127), (128),
and (159), and indicates bow, in a similar way, proofs of (155) and (162) may be obtained.
A considerable number of problems remain open for future research. Thus Ramanujan’s

UJe0Ur |rW|S»^{«)
(p. 163, equation (105)) is atiil unproved; it is uot even known that r(») = 0 (?s‘V + e) for
every positive e, though Hardy has proved that r (u) = 0 (a°) and
{r(l)^+!T(-2)P + ... + {r(».)p=0(tt>=).
Similar questions remain open concerning the functions (>i).
Other interesting problems suggested by Ramanujans work are those of finding
asymptotic formulas for
<i(l)d(«-l)+d(2)d(n-2) + ...+tf(«-l)tf(l)
and 2 d(»)rf(» + i), 2 r(n)r(n + k),
%<x »S.r
where r(n)=r2(n).
It has been proved by A. E. Ingham [“Some asymptotic formulas in the theory of
numbers ”, Journal London Math. Soe., vol. n (1927), pp. 202—208] that
^2 d («) d («+k) - ^ * (log .r^s g+0 (* log a),

and Jsrf(,)d(»-v)=^»(2i2(lOgi)S+0(»lOg»sJ)i
more generally, iDgham obtains asymptotic formulas for
Jv.Wv.ts + i), ^<r,{y)a,{n-r)
valid for all real positive values of r and s.
Among other recent memoirs dealing with similar questions we may mention
L. J. Mordell, “On the representations of numbers as a sum of 2r squares”, Quarterly
Journal of Math., vol. xlviii (1920), pp. 93-104;
L. J. Mordell, “ On the representations of a number as a sum of an odd number of
squares”, Trans. Canib. Phil. Soc., vol. xxn (1923), pp. 361—372;
G. H. Hardy, “ On the expression of a number as the sum of any number of squares,
and in particular of five or seven”, Proc. National Acad, of Sciences, vol. iv (1918),
pp. 189—193;
G. H. Hardy, “On the representation of a number as the sum of any number of
squares, and in particular of five", Trans. American Math. Soc., vol. xxr (1920), pp. 255—284;
’G. H. Hardy, “Note on Ramanujau’s arithmetical function t(m)”, Proa. Canib. Phil.
Soc., vol. xxni (1927), pp. 675—6S0;
 Notes on the Papers 341
G. K. Stanley, “On the representation of a number as the sum of seven squares =,
Journal London Math. Soc., vol. II (1927), pp. 91—96;
E. Landau, “Uber Gitterpunkte in mehrdimensionalen Ellipsoidon”, Math. Zeitschrift,
vol. XXI (1.024), pp. 126—132, and vol. xxiv (1926), pp. 299-310;
A. Walfisz, “Uber Gitterpunkte in mehrdimctisionalen EHipsoiden’', Math. Zeitschrift.
vol. xix (1924), pp. 300—307, and vol. xxvi (1927), pp. 106-124 ;
H. Peterssou, “Uber die Anzahl der Gitterpunkte in mehrdiniensionalen Ellipsoideu”,
Hamburg Math. Abhandhmgen, vol. V (1926), pp. 116—150.
These memoirs give further references to the literature of the subject.

P. 170. It has been pointed out by L. E. Dickson that Ramanujan overlooks the fact
that (1, 2, 5, 5) will not represent 15 : see his note in the Bulletin aimer. Math. Soc. referred
to below. This reduces the 55 forms to 54.
P. 171,1. 6. The forms 1, 1, 3, 3 and 1, 2, 3, 6 were considered also bv H. .1. S. Smith,
Collecteana Mathematica in memoriaa Motmmci Chelini (lhtsl), pp. 117—143 (Collected
Math. Papers, vol. II, pp. 287—311, particularly pp. 309—311!.
Pp. 171—172. A number of the results concerning ternary forms stated bv Ramanujan
seem to be new. In this connection Prof. Mordell writes: "suppose that
f(*, V, »)=«af!+4y+c*-+2#*+20f^ + 2/uy
is a properly primitive positive ternary form, so Hint a, b, c, t, g, h have no common factor
and a, b, c are not all even. This form lias two arithmetical invariants for linear transfor¬
mations of determinant 1, namely the invariants 12, A dehned by -

(1) ■ b f =a*A,

(2) a is the highest common factor of the coefficients of the adjoint form
F(x, y, z) = {be -f*) a? + ... + 2 (gh - af) ys + ....
There are only a finite number of classes of forms with given invariants a, A.
Consider the simplest case, in which a and A are both odd. If/ is auy form of a given
class, <0 is any prime factor of a, and & auy prime factor of a, the symbols of quadratic

(3)
reciprocity

hav^onstant values for all x, y, z which make/ prime to and F to 5. In fact, given F,
we can find a $ equivalent to/, with adjoint <&, such that
<f> = a.22 -J-/3Q;/2 -b yQ. A22, 4> = ftyQ,£xl + yaky- + a/322,
where a, ft, y are odd and afty= 1, the congruences being to modulus F: see P. Bachmann,
Die Arithmetik der quadratkehen For men, vol. I (Leipzig, 1S93), p. 54. Hence, if « is any
prime factor of Q, and iV=&>,

ffl-(f)-G)-
and is independent of x, y, z.
Hence the classes may be subdivided into genera corresponding to the values of the
symbols (3). Given a genus, we can calculate

M£K£>(~i)Mn+1>li+l>’
and it may be shewu that
^(-IjKV+DinF+H;
 342 Appendix I
If B=-l for a particular genus, we deduce from this that f(x, y, z) can take odd
values congruent to A, 3A, oA (mod 8), but not those congruent to 7A. If e.g.

then A = Q = 1, so that there is only one form in the genus, and E= - 1. Hence f takes
odd values congruent to 1, 3, 0 (mod 8), but not those congruent to 7. Similarly xi+yi + hz1
takes odd values congruent to 1,5, 7, but not those congruent to 3. "When B=\, the form

It is thus a simple matter to write down the conditions that a given odd number m,
which must of course give the symbols their appropriate values, should be represent¬
able by a given genus of forms; and when there is only one form in the genus, the problem
of representation of an odd number by that form is solved. The results for even numbers

aP+if+z2 and x2+y2+5P ,a divisible by 4 only if x, y, z are all even, so that the
forms cannot assume the values la($n + 7) or 4“(8ii + 3) respectively; but the residues
2 and 6 (mod 8) cannot be disposed of so simply.
The foundations of the theory were laid by Eisenstein [Journal fur Math., 35 (1847),
117—136 ; or Math. Abhandlungen, 177—206], and it is likely enough that he knew most

will represent all odd numbers, and that he will not delay to discuss even numbers also. I
may add that some of Ramanujan's results are easily deduced from the fundamental result
about the form (3T). Thus m = x2+y2 +2z2 gives
2»i=(*+3-)*+(x - y)2+(2z?=P + ,2 + T,
say, and f, r? have the same parity. If both are even, m is even and
i ™=(W+(iir-+A
which is possible unless m = 2.4’'(8/i+7). If both are odd, m is odd. In this case 2<n is
representable by three squares, say
2m=A'*+ Y* + Z\

m = {i (JT+ E)P+'U (A- A)}2+2 (i£)2=*!+/+2r2.

In a recent note in the Bulletin of the American Math. Soc. [vol. xxxiii (1927),
pp. 63—70] L. E. Dickson gives proofs of all Ramanujan’s results about ternary forms,
and others. To show, for example, that any integer not of the form 25*(25p±i5) is
representable by fix, y, z)=x2+2y2 + 5z2, he uses the fact that the classes of forms with
O!A = 10 are given by this form and three others. He writes down a particular ternary
form with S12A = 10 and first coefficient n and shews that it must be equivalent to f(x, y, z)
and so that n is representable by f(x, y, z). A continuation of this note will appear in
the Annals of Mathematics. There are two other recent papers by Dickson relevant to
Ramanujan’s work, viz. (1) ‘Quadratic forms which represent all integers’, Proc. Nat.
Acad, of Sciences, vol. xil (1926), pp. 756—757, and (2) 1 Quaternary quadratic forms
representing all integers’, American Journal of Math., vol. xlix (1927), pp. 39—66.”
See also H. D. Kloosterman: (1) “ Over het splitsen van gehule positive getallon in een
som van kwadraten”, Dissertation, Groningen, 1924; (2) “ On the representation of numbers
in the form ax2 + by2 + cz2 + dt2", Proc. London Math. Soc. (2), vol. xxv'l926), pp. 143-173;
(3) “tJber Gitterpunkte in vierdimensionalen EUipsoiden”, Math. Zeitschrift, vol. xxiv
(1926), pp. 519—529; (4) “On the representation of numbers in the form ajp+btf+czt+dt2”,
Acta Math., vol. xlix (1927), pp. 407—464.
 24.
n 208,1. 5. V (*) is what is usually denoted by 3 (x).

25.
Alternative proofs of (1) and (2), and a proof of (3), appear in No. 30. The results (5),
(7), and (8) are corollaries. The manuscript from which No. 30 is extracted also contains
alternative proofs of (4), (6), and a proof of (9), which we hope may ultimately he published.
For other proofs of (1) and (2) see
H. B. C. Darling: (1) “On Mr Ramanujan’s congruence properties of p(n)”> Proc.
Camb. Phil Socvol. xix (1919), pp. 217—218; (2) “Proofs of certain identities and con¬
gruences enunciated by S. Ramanujan”, Proc. London Math. Soc. (2), vol. xix (1921),
pp. 350—372; -* '
L. J. Rogers, “On a type of modular relation”, Proc. London Math. Soc. (2), vol. xix
(1921), pp. 387-397;
L. J. Mordell, “Note on certain modular relations considered by Messrs Ramanujan,
Darling, and Rogers”, Proc. London Math. Soc. (2), vol. xx (1922), pp. 408—416.
 344 Appendix I

26.
This p/ipnr appeared in the Proceedings of the Cambridge Philosophical Society
accompanied by an explanatory note by Hardy and a new proof of the identities by
Rogers, to whom the identities are originally due. Both these notes are reprinted below.

1. By G. II. Hardy. •
The identities in question are those numbered (10) and (11) in each of the two following

1 + l-g' + (l-?)(l-?2) + (l-?)(l-?'i)(l-53) + -

=(T^yrr-1) (i - f) (i - j») (i - ?“) (i - <n... ■ -(1)

i ,jl._£_+_£2_+
+l-y + (i-?)(1 -£) (1 -?) (1 -£) (I -£)
=~£) (l-</) (.i ~q-) (l- f) (l - 212) (i - f1)-' -('2J .
On the left-hand sides the indices of the powers of q in the numerators are n- and n (»+1),
while in the products on the right-hand sides the indices of the powers of q form two
arithmetical progressions with difference 0.
The formulae (1) and (2) were first discovered by Prof. Rogers, and are contained in
a paper published by him in 1894* In this paper they appear as corollaries of a series of
. general theorems, and, possibly for this reason, they seem to have escaped notice, in spite
of their obvious interest and beauty. They were rediscovered nearly 20 years later by
Mr Ramanujan, who communicated them to me in a letter from India in February 1913.
Mr Ramanujan had then no proof of the formulas, which he had found by a process of
induction, I communicated them in turn to Major MacMalion and to Prof. O. Perron
of Tubingen; but none of us was able to suggest a proof; and they appear, unproved, in
ch. m, vol. II, 1916, of Major MacMahon’s Combinatoiy Analysing:
Since 1916 three further proofs have been published, one by Prof. Rogers \ and two by
Prof. I. Schur of Strassburg, who appears to have rediscovered the formula) once raore§.
The proofs which follow are very much simpler than any published hitherto. The first
is extracted from a letter written by Prof. Rogers to Major MacMahoti in October 1917 ;
the second from a letter written by Mr Ramanujan to me in April of this year. They are
in principle the same, though the details di£fer||. It seemed to me most desirable tha^.the
simplest and most elegant proofs of such very beautiful formula should be made public
without delay, and I have therefore obtained the consent of the authors to their insertion
here.
* L. J. Rogers, “Second memoir on the expansion of certain infinite products”, Proc.
London Math. Soc. (1), vol. xxv (1894), pp. 318—343 (§ 5, pp. 328—329, formulae (1) and (2)).
+ ^ 33, 35. . _
% L. J. Rogers, “ On two theorems of Combinatory Analysis and some allied identities”,
Proc. London Math. Soc. (2), vol. xvi (1917), pp. 315—336 (pp. 315—317).
§ I. Schur, “Bin Beitrag zur additiven Zalilentheorie und zur Theorie der Ketten-
briiche”, Berliner Sitzungsberickte, 1917, No. 23, pp. 301—321. [Prpf. Schur’s work was
entirely independent; but it was naturally impossible to communicate with him at the

{| I have altered the notation of Mr Ramanujan’s letter so as to agree with that of

Prof. Rogers.
346 Appendix 1

and, when x- ,= 1+JL.+_t_+

I ?3 (1 - 2°)+?u (1 - ?10) - ■ ■ ■ ■
(l-jfl-ysq-yW (1 —y10) —...
(1-})(1-S3)(l-S3)

= (1 -f){l-P)(l-q")(l-qu)..: •"(11)

P. 224. An alternative proof of (o*4) may be found in Watson’s Bessel Functions,

For a proof that p (lln + Sj-O (mod 11) see No. 30. The manuscript from which
No. 30 was extracted contains a proof that r (5n) = 0 (mod 5); others were given by.
Darling (2) and Mordell in the papers referred to under 25.

29.
Proofs of the two formula} stated here were given later by Rogers and Mordell: see the
papers referred to under 25.
30.
The footnote by Hardy explains the history of this paper. It is impossible to include
more of it without very elaborate editorial work, hut we hope that the manuscript may
ultimately be published in full. In connection with the questiou of the parity of p (n), see
P. A. MaeMahon, “Note on the parity of the number which enumerates the partitions of
a number”, Proc. Camb. Phil. Soc., vul. xx, 1921, pp. 281—283, and “The parity of p (n),
the number of partitions of n, when n s 1000”, Journal London Math. Soc., vol. i, 1926,
pp. 2*24—225.
31.
A number of misprints in the original have been corrected.

In connection with this paper, and the more elementary parts of No. 36, the reader

K. Knopp, “ Asymptotischo Formeln der additiven Zahlentheorie”, Schriften der

Eonigsberger Gelekrten Gesellsehaft, Naturwiss. Elasse, vol. II, 1925, pp. 45-74 ;
K. Knopp and I. Schur, “ Elementarer Beweis einiger asymptotisoher Formeln der
additiven Zahlentheorie”, Math. Zeitschrift, vol. xxrv (1926), pp. 559—574.
Knopp and Sohur give simpler proofs of a number of results deduced here from the
general Tauherian theorems. They base the results on the following general theorem: if
P (x) = 2p„x« is a power-series with positive increasing coefficients, and
P(.r) = exp (JJa-tJNW*),
when S(x)='S.snx1l~1, sn^ian2 (a>0),
then \ogpn~2j(<m).
 348 Appendix I

36.
P. 287, § 3-1. Sec also the memoirs of Knopp and Schur referred to under 34.
Mr T. Timyaraghavan has proved that, if cn SO, /(.».’) — 2cnxn, and
.r) ~ A N'(I -x) exp ,

then C,n-<to + c1 + ... + en=0 j»"*exp 77^/(y)} ’

and CiX^pf
fora constant B; and that these results are substantially the best of their kind.

Questions and solutions.

We have made no systematic attempt to verify the questions of which no solutions
are given, but we add a. note on Question 289, p. 323. Mr T.^Vijayaraghavan has
pointed out that in order to justify Ramanujan's formal processes one has only to observe
that if
a„ > 0, Tn=N/[</! + J{Oi+v/(«3 + ...+«,/«»)'(],
then a necessary and sufficient condition for the existence of the limit lim Tn is that

For, if the condition is satisfied, there is a constant S such that

n«V[fi2+Vf^4+V(^s+...+V-ff2")}]<a's/[i+d,{i+v'(i+-)}]=^i>=ia’(i+.s/5);
and the sufficiency of the condition follows, since Tn increases steadily with n. On the
other hand, if

then T.„ZHn, so that the condition is also necessary. For the evaluation of P, and a less

site am der Analysis, voL I, pp. 29—30 and 184.

In the first of Hatnanujan’s examples,

2>. — 2 4 8
log«» log 2 log 3 log 4
+ + +->
^

so that the formula is convergent. The second example differs only trivially.
 APPENDIX II. FURTHER EXTRACTS FROM
RAMANUJAN'S LETTERS TO G. H. HARDY
We give here the theorems stated by Ramanujan in his letters from India, apart from
those already quoted in Hardy’s notice, and those given on one sheet of the first letter
which has been lost. The gaps in the numeration correspond to the formulse printed in
the notice. Thus the formula) (20), (i) and (v), missing on p. 353, will be found on p. xxix,
and what is printed under (21) on p, 353 is a continuation of what appears under the same
number on p. xxix.

I. I have found a function which exactly represents the number of prime numbers less
than “exactly ” in the sense that the difference between the function and the actual
number of primes is generally 0 or some small finite value even when x becomes infinite.
Jirti
854 Appendix II

12 Jan. 1920.
[This letter was written under difficulties and is in places very obscure. Ramanujan
however makes it clear that what he means by a “mock ^-function'5 is a function, defined
by a ^-series convergent for j q | < 1, for which we can calculate asymptotic formula),
when q tends toa “rational point” e2l'wlls) of the same degree of precision as 'those furnished,
for the ordinary 5-functions, by the theory of linear transformation. Thus he asserts, fa-
example, that if

= 1 +{T+if + (1 +qf(\ +?s)3+

and q-e~L-*-\ by positive values, then

His list of functions is]

Mock ^-functions.

^({)=1+I+?+(i+}!jii+rt+-’
_+.!_t_+

x(?)=1 + + • ■• ■• >

^ (?)-/(?) = 3^
[/■(#) being the function defined above].
Mock ^-functions (of 5th order).

-^>=1 + if0 r+i + 7T

a+?)(i+</) ’
<t> (?) = !+? (l+?) + ?4 (1 + 2) (l + 23)+29(l+2) (1 + 23)(1 + ?s) + ...,
f te) = 2+28(l + j) + 2°(l + ?)(l+Js) + ?10(l + ?)(l + ?2)(l + 23) + ...

(1-ff4) (l-2s)(l-?°)

1-2 (1-2!)(1-23)'T(1-23)(1-24)(1-s5) + -

^(j)=1 + rr2 + (l-2)(l-23) + '"’

<#>(-2) + x(2)=2-P(2),
/(-2) + 2-f(22)-2 = 0(-22) + ^(-2)

(1 ~2) (1- 24) (1 -2°) (1- 2°) ■■■

+ (q)-F(f) + l=q- 1 +22 + 2° + 212 + '
(1 —28) (1 -212) (1 — 228)
 Further extracts from Ramanujan’s Letters
Mock 9-functions (of 5th order).

^l+q + (l+q)(l+<?f (l+q)(l + f){l+qsr-'

<t>(q) = q + qi(l + q) + f(l+q)(l+q°) + ...,

\ f(q) = l + q(l + q) + fCl + q)(l + q1) + i‘(l+q)(l + q1)(l + q3) + ...

x{q)=rfr+ (iz^y{r-f)+ (i-/>+

2 22
•P( ):=r^+(i-! ) (i-ri)+(i-?)(i-20) (i-r)+"'
satisfy similar relations.
Mock ^-functions {of 1th order).