The Theory of Partitions

Cambridge Books Online © Cambridge University Press, 2010

Cambridge Books Online © Cambridge University Press, 2010
The Theory of
Partitions

George E. Andrews
Pennsylvania State University
University Park, Pennsylvania

CAMBRIDGE
UNIVERSITY PRESS

Cambridge Books Online © Cambridge University Press, 2010

 PUBLISHED BY THE PRESS SYNDICATE OF THE UNIVERSITY OF CAMBRIDGE
The Pitt Building, Trumpington Street, Cambridge CB2 1RP
CAMBRIDGE UNIVERSITY PRESS
The Edinburgh Building, Cambridge CB2 2RU, United Kingdom
40 West 20th Street, New York, NY 10011-4211, USA
10 Stamford Road, Oakleigh, Melbourne 3166, Australia
© Cambridge University Press 1984
Fhis boo]
and 1
eproduction of any part may take place wi
the written permission of Cambridge University Press.
Original hardcover edition first published 1976 by Addison-Wesley Publishing Company, Lie.
First paperback edition published 1998 by Cambridge University Press
Library of Congress Cataloging-in-Publication Data is available.

A catalog record for this book is available from the British Library.

ISBN 0-521-63766-X paperback

Transferred to digital printing 2003

Cambridge Books Online © Cambridge University Press, 2010

 To Joy
Amy, and
Katy

Cambridge Books Online © Cambridge University Press, 2010

Editor's Statement

A large body of mathematics consists of facts that can be presented and

described much like any other natural phenomenon. These facts, at times
explicitly brought out as theorems, at other times concealed within a proof,
make up most of the applications of mathematics, and are the most likely
to survive changes of style and of interest.
This ENCYCLOPEDIA will attempt to present the factual body of all
mathematics. Clarity of exposition, accessibility to the non-specialist, and a
thorough bibliography are required of each author. Volumes will appear in
no particular order, but will be organized into sections, each one comprising
a recognizable branch of present-day mathematics. Numbers of volumes and
sections will be reconsidered as times and needs change.
It is hoped that this enterprise will make mathematics more widely used
where it is needed, and more accessible in fields in which it can be applied but
where it has not yet penetrated because of insufficient information.

The theory of partitions is one of the very few branches of mathematics

that can be appreciated by anyone who is endowed with little more than a
lively interest in the subject. Its applications are found wherever discrete
objects are to be counted or classified, whether in the molecular and the
atomic studies of matter, in the theory of numbers, or in combinatorial
problems from all sources.
Professor Andrews has written the first thorough survey of this many-
sided field. The specialist will consult it for the more recondite results, the
student will be challenged by many a deceptively simple fact, and the applied
scientist may locate in it the missing identity he needs to organize his data.
Professor Turan's untimely death has left this book without a suitable
introduction. It is fitting to dedicate it to the memory of one of the masters
of number theory.
GIAN-CARLO ROTA

Cambridge Books Online © Cambridge University Press, 2010

Contents

Editor's Statement vi
Preface to Paperback Edition xiii
Preface xv

Chapter 1 The Elementary Theory of Partitions 1

1.1 Introduction 1
1.2 Infinite Product Generating Functions of One Variable 3
1.3 Graphical Representation of Partitions 6
Examples 13
Notes 14
References 14

Chapter 2 Infinite Series Generating Functions 16

2.1 Introduction 17
2.2 Elementary Series-Product Identities 17
2.3 Applications to Partitions 23
Examples 28
Notes 30
References 31

Chapter 3 Restricted Partitions and Permutations 33

3.1 Introduction 33
3.2 The Generating Function for Restricted Partitions 33
3.3 Properties of Gaussian Polynomials 35
3.4 Permutations and Gaussian Multinomial Coefficients 39
3.5 The Unimodal Property 45
Examples 49
Notes 51
References 52

Chapter 4 Compositions and Simon Newcomb's Problem 54

4.1 Introduction 54
vii

Cambridge Books Online © Cambridge University Press, 2010

 Contents

4.2 Composition of Numbers 54

4.3 Vector Compositions 57
4.4 Simon Newcomb's Problem 59
Examples 63
Notes 65
References 65

Chapter 5 The Hardy-Ramanujan-Rademacher Expansion of p(n) . . 68

5.1 Introduction 68
5.2 The Formula for p(n) 71
Examples 81
Notes 85
References 86

Chapter 6 The Asymptotics of Infinite Product Generating Functions 88

6.1 Introduction 88
6.2 Proof of Theorem 6.2 89
6.3 Applications of Theorem 6.2 97
Examples 97
Notes 100
References 101

Chapter 7 Identities of the Rogers-Ramanujan Type 103

7.1 Introduction 103
7.2 The Generating Functions 106
7.3 The Rogers-Ramanujan Identities and Gordon's Generali-
zation 109
7.4 The Gollnitz-Gordon Identities and Their Generalization 113
Examples 116
Notes 118
References 118

Chapter 8 A General Theory of Partition Identities 121

8.1 Introduction 121
8.2 Foundations 121
8.3 Partition Ideals of Order 1 124
8.4 Linked Partition Ideals 128
Examples 136
Notes 137
References 138

Cambridge Books Online © Cambridge University Press, 2010

Contents ix

Chapter 9 Sieve Methods Related to Partitions 139

9.1 Introduction 139

9.2 Inclusion-Exclusion 139
9.3 A Sieve for Successive Ranks 142
Examples 156
Notes 158
References 158

Chapter 10 Congruence Properties of Partition Functions 159

10.1 Introduction 159

10.2 Rddseth's Theorem for Binary Partitions 161
10.3 Ramanujan's Conjecture for 5" 167
Examples 175
Notes 177
References 178

Chapter 11 Higher-Dimensional Partitions 179

11.1 Introduction 179

11.2 Plane Partitions 179
11.3 The Knuth-Schensted Correspondence 184
11.4 Higher-Dimensional Partitions 189
Examples 198
Notes 199
References 200

Chapter 12 Vector or Multipartite Partitions 202

12.1 Introduction 202

12.2 Multipartite Generating Functions 202
12.3 Bell Polynomials and Formulas for Multipartite Partition
Functions 204
12.4 Restricted Bipartite Partitions 207
Examples 209
Notes 210
References 211

Cambridge Books Online © Cambridge University Press, 2010

x Contents

Chapter 13 Partitions in Combinatorics 212

13.1 Introduction 212
13.2 Partitions and Finite Vector Spaces 212
13.3 Partitions of Sets 214
13.4 The Combinatorics of Symmetric Functions 221
Examples 225
Notes 227
References 228

Chapter 14 Computations for Partitions 230

14.1 Introduction 230

14.2 Elementary Algorithms 230
14.3 Algorithms from Generating Functions 233
14.4 Computations for Higher-Dimensional Partitions . . . . 234
14.5 Brief Tables of Partition Functions 237
14.6 Table of the Plane Partition Function 237
14.7 Table of Gaussian Polynomials 240
14.8 Guide to Tables 243
Notes 243
References 244

Index for Definitions of Symbols 245

Author Index 250
Subject Index 252

Cambridge Books Online © Cambridge University Press, 2010

Cambridge Books Online © Cambridge University Press, 2010
Cambridge Books Online © Cambridge University Press, 2010
Preface to the Paperback Edition

In the past twenty years, the theory of partitions has blossomed. The object
of this book, as appropriate in this series, is to provide the fundamentals in a
form accessible to the nonspecialist, with references to the recent literature
for those who wish to pursue a particular interest. The major changes I would
have made in a total revision would have added greatly to the length of the
book. This is a task of such magnitude that it will have to wait until a number
of my other projects are completed.
In the light of this introduction, here are my comments on the chapters of
the third printing of The Theory of Partitions. Chapter 1 contains the almost
immutable basics. Chapter 2 is partially devoted to basic hypergeometric
series. As such, it is a small introduction to the wonderful world of q that has
been so beautifully chronicled by Gasper and Rahman (1990). Andrews
(1986) provides a further survey of the interactions of partitions with q. For
Chapter 3, it should be pointed out that O'Hara (1990) has shown how to
prove the unimodality of Gaussian polynomials in a purely elementary
(although not easy) manner. Chapters 3 through 5 are fairly current introduc-
tions to their topics.
The work in Chapter 6 has been greatly extended by Richard Mclntosh in a
series of papers: cf. Mclntosh (1995). The material in Chapters 7, 8, and 9 has
also been greatly extended in the past decade. Workers in partitions proper
(for example, Alladi and Gordon) have made major advances. In the 1970s
and 1980s, David Bressoud made spectacular progress in this area. For further
references and an account of his central methods, see Bressoud (1980). In
addition, the world of statistical mechanics has provided a flood of results, as
well as important questions on this topic. The fountainhead of this work is
Rodney Baxter (1982).
The introduction to Chapter 10 is reasonably up to date. However, there
have been major recent discoveries for partition function congruences; the
most notable have been found by Garavan, Kim, and Stanton (1990); Ono
(1996); and Gordon and Ono (1997).
There have also been extensive discoveries for higher-dimensional parti-
tions. Here the interested reader should consult surveys by Stanley (1986a,
1986b) and a related paper by Robbins (1991). As many of the conjectures in

Cambridge Books Online © Cambridge University Press, 2010

xiv Preface to the Paperback Edition

these papers are now theorums, these surveys are themselves becoming out of
date, but they are good leads for what has been done.
The topics in Chapters 12 and 13 might well be augmented with an account
of generalized Frobenius partitions (see Andrews [1984]).
Chapter 14 contains the appropriate computational rudiments. However,
Zeilberger (1991) has pioneered exciting new computational work related to
partitions and other combinatorial identities (cf. Zeilberger and Wilf [1990]).
The following list of references is designed only to provide the reader with
leads into the literature. It does not in any way give adequate credit to the
numerous contributions of scores of researchers during the past two decades.

References
Alladi, K. (1995). "The method of weighted words and applications to partitions," Number
Theory (S. David, ed.). Cambridge University Press, Cambridge.
Andrews, G. E. (1984). "Generalized Frobenius partitions," Mem. Amer. Math. Soc. 301.
Andrews, G. E. (1986). "^-Series: Their development and application in analysis, number
theory, combinatorics, physics and computer algebra," Mem. Amer. Math. Soc. 66.
Baxter, R. J. (1982). Exactly Solved Models in Statistical Mechanics. Academic Press, London
and New York.
Bressoud, D. M. (1980a). "Analytic and combinatorial generalizations of the Rogers-Ramanujan
identities," Mem. Amer. Math. Soc. 227.
Bressoud, D. M. (1980b). "Extension of the partition sieve," J. Number Th. 12, 76-100.
Bressoud, D. M. (1998). Proofs and Confirmations. Cambridge University Press, Cambridge.
Burge, W. H. (1981). "A correspondence between partitions related to generalizations of the
Rogers-Ramanujan identities," Discrete Math. 34, 9-15.
Garvan, E, Kim, D., and Stanton, D. (1990). "Cranks and /-cores," Invent. Math. 101, 1-17.
Gasper, G, and Rahman, M. (1990). "Basic hypergeometric series," Encycl. Math and Its
Applications 35 (G-C. Rota, ed.). Cambridge University Press, Cambridge.
Gordon, B., and Ono, K. (1997). "Divisibility properties of certain partition functions by powers
of primes," Ramunujan J. 1, 25-35.
Mclntosh, R. J. (1995). "Some asymptotic formulae for #-hypergeometric series," J. London
Math. Soc. (2) SI, 120-136.
O'Hara, K. M. (1990). "Unimodality of Gaussian coefficients: A constructive proof," J.
Combinatorial Th. A53, 29-52.
Ono, K. (1996). "On the parity of the partition function in arithmetic progressions," J. Reine
Angew. Math. 472, 1-15.
Robbins, D. P. (1991). "The story of 1, 2, 7, 42, 429, 7436 " Math. Intellig. 13, 12-19.
Stanley, R. P. (1986a). "A baker's dozen of conjectures concerning plane partitions," Lecture
Notes in Math. 1234, 285-293. Springer, Berlin.
Stanley, R. P. (1986b) "Symmetries of plane partitions," J. Comb. Th. A43, 103-113.
Zeilberger, D., and Wilf, H. S. (1990). "Rational functions certify combinatorial identities,"
J. Amer. Math. Soc. 3, 147-158.
Zeilberger, D. (1991). "The method of creative telescoping," J. Sym. Comp. 11, 195-204.

Cambridge Books Online © Cambridge University Press, 2010

Preface

Let us begin by acknowledging that the word "partition" has numerous

meanings in mathematics. Any time a division of some object into subobjects
is undertaken, the word partition is likely to pop up. For the purposes of this
book a "oartition of n" is a nonincreasing finite sequence of positive integers
whose sum is n. We shall extend this definition in Chapters 11, 12, and 13
when we consider higher-dimensional partitions, partitions of n-tuples, and
partitions of sets, respectively. Compositions or ordered partitions (merely
finite sequences of positive integers) will be considered in Chapter 4.
The theory of partitions has an interesting history. Certain special problems
in partitions certainly date back to the Middle Ages; however, the first
discoveries of any depth were made in the eighteenth century when L. Euler
proved many beautiful and significant partition theorems. Euler indeed laid
the foundations of the theory of partitions. Many of the other great math-
ematicians - Cayley, Gauss, Hardy, Jacobi, Lagrange, Legendre, Littlewood,
Rademacher, Ramanujan, Schur, and Sylvester —have contributed to the
development of the theory.
There have been almost no books devoted entirely to partitions. Generally
the combinatorial and formal power series aspects of partitions have found
a place in older books on elementary analysis (Introductio in Analysin
Infinitorum by Euler, Textbook of Algebra by Chrystal), in encyclopedic
surveys of number theory (Niedere Zahlentheorie by Bachman, Introduction
to the Theory of Numbers by Hardy and Wright), and in combinatorial
analysis books (Combinatory Analysis by MacMahon, Introduction to
Combinatorial Analysis by Riordan, Combinatorial Methods by Percus,
Advanced Combinatorics by Comtet). The asymptotic problems associated
with partitions have, on the other hand, been treated in works on analytic
or additive number theory (Introduction to the Analytic Theory of Numbers
by Ayoub, Modular Functions in Analytic Number Theory by Knopp,
Topics from the Theory of Numbers by Grosswald, Additive Zahlentheorie
by Ostmann, Topics in Analytic Number Theory by Rademacher).
If one considers the applications of partitions in various branches of
mathematics and statistics, one is struck by the interplay of combinatorial
and asymptotic methods. We have tried to organize this book so that it
adequately develops and interrelates both combinatorial and analytic methods.

Cambridge Books Online © Cambridge University Press, 2010

 Preface

Chapters 1-4 treat the elementary portions of the theory of partitions; of

primary importance here is the use of generating functions.
Chapters 5 and 6 treat the asymptotic problems. Partition identities are
dealt with in Chapters 7 through 9. Chapter 10 on partition function con-
gruences returns to the analytic aspect of partitions. Chapters 11-13 treat
several generalizations of partitions and Chapter 14 presents a brief discussion
of the computational aspect of partitions.
There are three concluding sections of each chapter: A "Notes" section
provides historical comment on the material covered; a "References" section
provides a substantial but nonexhaustive list of relevant books and papers;
and an "Examples" section provides statements of results not fully covered in
the text. Those examples that occur with an asterisk are significant advances
beyond the material presented in the text; the remainder form a reasonable
set of exercises by which the reader may determine his grasp of the subject
matter. References for the source of the examples occur in the related Notes
section.
Many of the mathematical sciences have seen applications of partitions
recently. Nonparametric statistics require restricted partitions like those in
Chapter 3. Various permutation problems in probability and statistics are
intimately linked with the Simon Newcomb problem of Chapter 4. Particle
physics uses partition asymptotics and partition identities related to the work
in Chapters 5-9. Group theory (through Young tableaux) is intimately
connected with Chapter 12, and the relationship between partitions and
combinatorial theory is explored in Chapter 13.
The material in this book has been developed over a period of years. My
first acquaintance with partitions came from thrilling lectures delivered by
my thesis adviser, the late Professor Hans Rademacher. Many of the topics
herein have been presented in graduate courses at the Pennsylvania State
University between 1964 and 1975, in seminars at MIT during the 1970-1971
academic year, at the University of Erlangen in the summer of 1975, and at
the University of Wisconsin during the 1975-1976 academic year. I owe a
great debt of gratitude to many people at these four universities. I wish to
thank specially R. Askey, K. Baclawski, B. Berndt, and L. Carlitz, who
contributed many valuable suggestions and comments during the preparation
of this book.
Finally I thank my wife, Joy, who has throughout this project been both
a help and an inspiration to me.
GEORGE E. ANDREWS

Cambridge Books Online © Cambridge University Press, 2010

CHAPTER 1

The Elementary Theory of Partitions

1.1 Introduction

In this book we shall study in depth the fundamental additive decomposition

process: the representation of positive integers by sums of other positive
integers.
DEFINITION 1.1. A partition of a positive integer n is a finite nonincreasing
sequence of positive integers kl9 k2,..., K such that £ ' = 1 kt = n. The kt are
called the parts of the partition.
Many times the partition (ku k2,..., kr) will be denoted by k, and we shall
write k h- n to denote "A is a partition of w." Sometimes it is useful to use a
notation that makes explicit the number of times that a particular integer
occurs as a part. Thus if k = (ki9 A 2 ,..., Ar) h- n, we sometimes write

where exactly/, of the ki are equal to i. Note now that Y*i>ifi* = n-

Numerous types of partition problems will concern us in this book;
however, among the most important and fundamental is the question of
enumerating various sets of partitions.
DEFINITION 1.2. The partition function p(n) is the number of partitions
of n.
Remark. Obviously p{n) = 0 when n is negative. We shall set p(0) = 1 with
the observation that the empty sequence forms the only partition of zero. The
following list presents the next six values of p(n) and tabulates the actual
partitions.

p(l) = 2: 2 = (2), 1 + 1 - (I 2 );
p(3) = 3: 3 = (3), 2 + 1 =(12), 1 + 1 + 1 = (I 3 );
p(4) = 5: 4 = (4), 3 + 1 =(13), 2 + 2 = (22),
2 + 1+ 1 = (122), 1 + 1 + 1 + 1 = (I 4 );

ENCYCLOPEDIA OF MATHEMATICS and Its Applications, Gian-Carlo Rota (ed.).

2, George E. Andrews, The Theory of Partitions
Copyright © 1976 by Addison-Wesley Publishing Company, Inc., Advanced Book Program.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical photocopying,
recording, or otherwise, without the prior permission of the publisher.

Cambridge Books Online © Cambridge University Press, 2010

 The Elementary Theory of Partitions Chap. 1.1

= 7: 5 = (5), 4 + 1 = (14), 3 + 2 = (23),

3 + 1+ 1 = (133), 2 + 2 + 1 = (122),
2 + 1+ 1 + 1 = (132), 1 + 1 + 1 + 1 + 1 = (I 5 );
K6)=ll: 6 = (6), 5 + 1 = (15), 4 + 2 = (24),
4 + 1+ 1 = (124), 3 + 3 = (32), 3 + 2 + 1 = (123),
3 + 1+ 1 + 1 = (133), 2 + 2 + 2 = (23),
2 + 2+ 1 + 1 = (1 2 2 2 ), 2 + 1 + 1 + 1 + 1 = (142),
1 + 1+ 1 + 1 + 1 + 1 = (I 6 ).
The partition function increases quite rapidly with n. For example, p(10) =
42, p(20) = 627, p(50) = 204226, p(100) = 190569292, and p(200) =
3972999029388.
Many times we are interested in problems in which our concern does not
extend to all partitions of n but only to a particular subset of the partitions
of n.
DEFINITION 1.3. Let Sf denote the set of all partitions.
DEFINITION 1.4. Let p(S, n) denote the number of partitions of n that belong
to a subset S of the set Sf of all partitions.
For example, we might consider 0 the set of all partitions with odd parts
and 3) the set of all partitions with distinct parts. Below we tabulate partitions
related to 0 and to B.
l) = l: 1 = ( 1 ) ,

3) = 2: 3= (3), 1 + 1 + 1 = (I 3 ),
4) = 2: 3+ 1= (13), 1 + 1 + 1 + 1 = (I 4 ),
5) = 3: 5= (5), 3 + 1 + 1 = (123),
1 + 1+ 1 + 1 + 1 = (I 5 ),
6) = 4: 5+ 1= (15), 3 + 3 = (32),
3 + 1+ 1 + 1 = (133),
1+ 1+ 1 + 1 + 1 + 1 = (I 6 ),
7) = 5: 7= (7), 5 + 1 + 1 = (125), 3 + 3 + 1 = (132),
3 + 1+ 1 + 1 + 1 - (143),
1+ 1+ 1 + 1 + 1 + 1 + 1 = (I 7 ).
, 1) = 1: 1= (1),
2)= 1: 2 = (2),
3) = 2: 3 = (3), 2 4• 1 = (12),
4) = 2: 4 = (4), 3 4• 1 - (13),
5) = 3: 5 = (5), 4 4• 1 = (14), 3 + 2 = (23),
6) = 4: 6 = (6), 5 4• 1 = (15), 4 + 2 = (24),
3 + 2 + 1 = (123),
7) = 5: 1 = (7), 6 4• 1 = (16), 5 + 2 = (25),
4 + 3 = (34), 4 + 2 + 1 = (124).

Cambridge Books Online © Cambridge University Press, 2010

1.2 Infinite Product Generating Functions 01 One Variable 3

We point out the rather curious fact that p((P, n) = p(S), n) for n < 7,
although there is little apparent relationship between the various partitions
listed (see Corollary 1.2).
In this chapter, we shall present two of the most elemental tools for treating
partitions: (1) infinite product generating functions; (2) graphical representa-
tion of partitions.

1,2 Infinite Product Generating Functions of One Variable

DEFINITION 1.5. The generating function/(g) for the sequence aQ, a t, a2i a3,...
is the power series f(q) = ^ o anqn.

Remark. For many of the problems we shall encounter, it suffices to

consider f(q) as a "formal power series" in q. With such an approach many
of the manipulations of series and products in what follows may be justified
almost trivially. On the other hand, much asymptotic work (see Chapter 6)
requires that the generating functions be analytic functions of the complex
variable q. In actual fact, both approaches have their special merits (recently,
E. Bender (1974) has discussed the circumstances in which we may pass from
one to the other). Generally we shall state our theorems on generating
functions with explicit convergence conditions. For the most part we shall
be dealing with absolutely convergent infinite series and infinite products;
consequently, various rearrangements of series and interchanges of summation
will be justified analytically from this simple fact.

DEFINITION 1.6. Let H be a set of positive integers. We let "//"" denote the
set of all partitions whose parts lie in H. Consequently, p("/f \ n) is the
number of partitions of n that have all their parts in H.

Thus if if o i s t h e s e t o f a11 o d d
positive integers, then "Ho" = 0.

p("//o'\ n) = p(0, n).

DEFINITION 1.7. Let H be a set of positive integers. We let "//"(< d) denote

the set of all partitions in which no part appears more than d times and each
part is in H.

Thus if N is the set of all positive integers, then p("N"(^ 1), n) = p(^, n).

THEOREM 1.1. Let H be a set of positive integers, and let

f(q) = I P("fl'\ n)q\ (1.2.1)

Cambridge Books Online © Cambridge University Press, 2010

4 The Elementary Theory of Partitions Chap. 1.2

Then for \q\ < 1

= Il(l -<lT\ (1.2.3)

neH

= n (i + «•+•••+«*•)

netf

Remark. T h e e q u i v a l e n c e o f the t w o forms for fd(q) f o l l o w s f r o m the s i m p l e

formula for the s u m o f a finite g e o m e t r i c series:

1 + x + x 2 + • • • + xr =
1 —x

Proof. We shall proceed in a formal manner to prove (1.2.3) and (1.2.4);

at the conclusion of our proof we shall sketch how to justify our steps analyt-
ically. Let us index the elements of H, so that H = {hi9h29 h3, / i 4 , . . . } . Then

neH neH

= (1 + qhi + q2hl + q3hl +•••)

x (1 + qh> + q2h2 + g3*2 + • • • )

x (1 + qh> + q2h* + q 3 * 3 + • • • )

y y y

and we observe that the exponent of q is just the partition (hlaih2a2h3a3- • •)•
Hence qN will occur in the foregoing summation once for each partition of n
into parts taken from H. Therefore

IK 1 - * T l = I PCH", ")<?".
neH nZO

The proof of (1.2.4) is identical with that of (1.2.3) except that the infinite
geometric series is replaced by the finite geometric series:

neH

) , n)q"

Cambridge Books Online © Cambridge University Press, 2010

1.2 Infinite Product Generating Functions of One Variable 5

If we are to view the foregoing procedures as operations with convergent

infinite products, then the multiplication of infinitely many series together
requires some justification. The simplest procedure is to truncate the infinite
product to n ? = i ( l ~~ Qhi)~1' This truncated product will generate those
partitions whose parts are among hu h2,..., hn. The multiplication is now
perfectly valid since only a finite number of absolutely convergent series are
involved. Now assume q is real and 0 < q < 1; then if M = hn,

Zp("H"j)qj ^ n a - q
h l
r ^ o (i - qktrl < *>•
j=0 i=l j=l

Thus the sequence of partial sums YJf=o P("H">J)QJ *s a bounded increasing

sequence and must therefore converge. On the other hand

as

Therefore

ipi"H"j)qj = n o - qhri - n o -
j=0 1=1 neH

Similar justification can be given for the proof of (1.2.4).

COROLLARY 1.2 (Euler). p((P, n) = p(@, n) for all n.

Proof. By T h e o r e m 1.1,

n^O n= 1

and

Now

1
n (1 + qn) = n = n
TT^T
Hence

Cambridge Books Online © Cambridge University Press, 2010

6 The Elementary Theory of Partitions Chap. 1.3

and since a power series expansion of a function is unique, we see that p(C, n) =
p(&, n) for ail n. •

COROLLARY 1.3 (Glaisher). Let Nd denote the set of those positive integers
not divisible by d. Then

P("N«i+r\ n) = P("N"(< d), n)

for all n.

Proof. By Theorem 1.1,

_ n _i
~~ 11 / j n\

and the result follows as before. •

There are numerous results of the type typified by Corollaries 1.2 and 1.3.
We shall run into such results again in Chapters 7 and 8, where much deeper
theorems of a similar nature will be discussed.

1.3 Graphical Representation of Partitions

Another effective elementary device for studying partitions is the graphical

representation. To each partition / is associated its graphical representation
r
&k (or Ferrers graph), which formally is the set of points with integral co-
ordinates (/,y) in the plane such that if / = ().x, /.2 A,,), then {ij)e $\
if and only if 0 ^ / ^ — n + 1, 0 ^ j ^ / J,-J + x — 1. Rather than dwell on this
formal definition, we shall, by means of a few examples, fully explain the
graphical representation.
The graphical representation of the partition 8 -f 6 + 6 + 5 + Ms

The graphical representation of t h e partition 7-h3-h3

Cambridge Books Online © Cambridge University Press, 2010

1.3 Graphical Representation of Partitions

Note that the /th row of the graphical representation of (klf k2i..., kn)
contains kt points (or dots, or nodes).
We remark that there are several equivalent ways of forming the graphical
representation. Some authors use unit squares instead of points, so that the
graphical representation of 8 + 6 + 6 + 5 + 1 becomes

Such a representation is extremely useful when we consider applications of

partitions to plane partitions or Young tableaux (see Chapter 11).
Other authors prefer the representation to be upside down (they would
say right side up); for example, in the case of 8 + 6 + 6 + 5 + 1

or

Since most of the classical texts on partitions use the first representation
shown in this section, we shall also.

1.8. If k = (kl9..., kn) is a partition, we may define a new

DEFINITION
partition k' = ( / / , . . . , kj) by choosing k{ as the number of parts of k that
are ^ /. The partition k' is called the conjugate of k.

Cambridge Books Online © Cambridge University Press, 2010

8 The Elementary Theory of Partitions Chap. 1.3

While the formal definition of conjugate is not too revealing, we may better
understand the conjugate by using graphical representation. From the
definition, we see that the conjugate of the partition 8 + 6 + 6 + 5 + 1 is
5 + 4 + 4 + 4 + 4 + 3 + 1 + 1. The graphical representation of 8 + 6 +
6 + 5 + 1 is

and the conjugate of this partition is obtained by counting the dots in successive
columns; that is, the graphical representation of the conjugate is obtained
by reflecting the graph in the main diagonal. Thus the graph of the conjugate
partition is

Notice that not only does the graphical representation provide a simple
method by which to obtain the conjugate of A, but it also shows directly that
the conjugate partition / ' is a partition of the same integer as / is; that is,
1A{ — 1).(. Furthermore, it is clear that conjugation is an involution of the
partitions of any integer, in that the conjugate of the conjugate of/, is again L
Let us now prove some theorems on partitions, using graphical representa-
tion.
THEOREM 1.4. The number of partitions of n with at most m parts equals
the number of partitions of n in which no part exceeds m.
Proof. We may set up a one-to-one correspondence between the two classes
of partitions under consideration by merely mapping each partition onto
its conjugate. The mapping is certainly one-to-one, and by considering the
graphical representation we see that under conjugation the condition "at
most m parts" is transformed into "no part exceeds m" and vice versa. •

As an example, let us consider the partitions of 6, first into at most three

parts and then into parts none of which exceeds 3. We shall list conjugates
opposite each other.

Cambridge Books Online © Cambridge University Press, 2010

1.3 Graphical Representation of Partitions

6 1 Hh 1 H
h 1 H
- 1 H
h1+ 1
5 Hh 1 2 Hh 1 h 1
H H
- 1 H
h 1
4 Hh 2 2 Hh 2 Hh 1 -
H 1
4 Hh 1 Hh 1 3 Hh 1 H
h 1 -
H 1
3 Hh 3 2 Hh 2 Hh 2
3 Hh 2 Hh 1 3 Hh 2 Hh 1
2 Hh 2 Hh 2 3 Hh 3

Theorem 1.4 is quite useful and shows how a graphical representation can
be used directly to obtain important information. More subtle uses of this
technique can be seen in the following two theorems.
THEOREM 1.5. The number of partitions of a — c into exactly b — 1 parts,
none exceeding c, equals the number of partitions of a — b into c — 1 parts,
none exceeding b.
Proof. Let us consider the graphical representation of a typical partition
of the first type mentioned in the theorem. We transform the partition as
follows: first we adjoin a new top row of c nodes; then we delete the first
column (which now has b nodes); and then we take the conjugate:
<c c c —1 ^ b

6-1- ->c- 1

We see immediately that this composite transformation provides a one-to-one

correspondence between the two types of partitions considered, and con-
sequently the theorem is established. •
As an example, let us consider the case in which a = 14, b = 5, c = 4.

4+4+1+1 - 3 + 3+ 3
4 + 3+ 2+1 -> 4 + 3+ 2
4 + 2+ 2+ 2 -> 5 + 2 + 2
3+3+3+1 - 4 + 4+1
3+3+2+2 - 5 + 3 + 1
We conclude this chapter with one of the truly remarkable achievements
of nineteenth-century American mathematics: F. Franklin's proof of Euler's
pentagonal number theorem. Franklin's accomplishment was to prove
Legendre's combinatorial interpretation of Euler's theorem. We shall actually
state the pentagonal number theorem as Corollary 1.7, and we shall show
how useful the theorem is computationally in Corollary 1.8.

Cambridge Books Online © Cambridge University Press, 2010

10 The Elementary Theory of Partitions Chap. 1.3

THEOREM 1.6. Let p e (^, n) (resp. pQ(@, n)) denote the number of partitions
of n into an even (resp. odd) number of distinct parts. Then

- l) m if n = hm(3m + 1),
otherwise ~

Proof. We shall attempt to establish a one-to-one correspondence between

the partitions enumerated by p e (^, n) and those enumerated by po(@, n).
For most integers n our attempt will be successful; however, whenever n is
one of the pentagonal numbers \m{2>m + 1), a single exceptional case will
arise.
To begin with, we note that each partition A = (A 1 ? ..., Ar) of n has a
smallest part s(k) = Ar; also, we observe that the largest part At of A =
(A1? A2,. • ., Ar) is the first of a sequence of, say, G(A) consecutive integers that
are parts of A (formally <r(A) is the largest j such that A7- = At - j + 1).
Graphically the parameters s(A) and c(A) are easily described:

X=(76432) X = (8765)

<S3> s(X) =
s(X) = 2
We transform partitions as follows.
Case 1. s(X) ^ a{k). In this event, we add one to each of the s(A) largest
parts of A and we delete the smallest part. Thus

A = (76432) - A' = (8743);

that is

Cambridge Books Online © Cambridge University Press, 2010

1.3 Graphical Representation of Partition > \\

Case 2. s(X) > o(X). In this event, we subtract one from each of the c{X)
largest parts of X and insert a new smallest part of size <T(A). Thus
X = (8743)-* (76432);
that is

The foregoing procedure in either case changes the parity of the number of
parts of the partition, and noting that exactly one case is applicable to any
partition X, we see directly that the mapping establishes a one-to-one cor-
respondence. However, there are certain partitions for which the mapping
will not work. The example X = (8765) is a case in point. Case 2 should be
applicable to it; however, the image partition is no longer one with distinct
parts. Indeed, Case 2 breaks down in precisely those cases when the partition
has r parts, a{X) = r and s(X) = r + 1, in which case the number being
partitioned is

(r + 1) + (r + 2) + • • • + 2r = |r(3r + 1).
On the other hand, Case 1 breaks down in precisely those cases when the
partition has r parts, o{X) = r and s(X) = r, in which case the number being
partitioned is

r + (r + 1) + • • • + (2r - 1) = M 3 r - 1).
Consequently, if n is not a pentagonal number, p e (^, n) = po(@, n); if
n = JK3r ± 1), pt(&, n) = p o (0, n) + ( - l) r . •
COROLLARY 1.7 (Euler's pentagonal number theorem).

n a - qn) = i + i ( - i)v m(3ffl " i] (i +

n=1 m=1

= Z ( - i)V w(3m " 1} .

Proof. Clearly
00 00
l) ^_ V (_ n«^i™(3m-1)

m=1

Cambridge Books Online © Cambridge University Press, 2010

12 The Elementary Theory of Partitions Chap. 1.3

oc

m= 1

= 1 +t(P*&>n)-Poi®>n))<lH>

by Theorem 1.6.
To complete the proof we must show that
00 00

1 + £ (pc(@, n) - po(@, n))qn = ]~J (1 — qn).

n=l n= 1

Now
oo 1 1 1

as in the proof of (1.2.4) in Theorem 1.1. Note now that each partition with
distinct parts is counted with a weight (— l)ai+ai+a> + "\ which is + 1 if the
partition has an even number of parts and — 1 if the partition has an odd
number of parts. Consequently
00 1 1 1

=
n=l a i = O fl2 0 03 = 0

and so we have the desired result. •

COROLLARY 1.8 (Euler). / / n > 0, then

p(n) - p(n - I ) - p(n - 2) + p(n - 5) + p(n - 1)

+ • • + ( - l)mp(" - irn(3m - 1))

+ ( - \)mp(n - im(3m + 1)) + • • • = 0, (1.3.2)

where we recall that p(M) = 0 for all negative M.

Proof, Let an denote the left-hand side of (1.3.2). Then clearly

I <*nqn = I P(n)qn> fl + f ( - l ) V m ( 3 m " n d +

n=0 n=0 [ m=l

Cambridge Books Online © Cambridge University Press, 2010

Examples 13

where the penultimate equation follows immediately by (1.2.3) and Corollary

1.7. Hence, an = 0 for n > 0. •

Corollary 1.8 provides an extremely efficient algorithm for computing

p(n) that we shall discuss further in Chapter 14.

Examples

1. (Subbarao) The number of partitions of n in which each part appears

two, three, or five times equals the number of partitions of n into parts
congruent to 2, 3, 6, 9, or 10 modulo 12.
2. The number of partitions of n in which only odd parts may be repeated
equals the number of partitions of n in which no part appears more than
three times.
3. The number of partitions of n in which only parts ^ 0 (mod 2m) may
be repeated equals the number of partitions of n in which no part appears
more than 2 m + 1 — 1 times.
4. (Ramanujan) The number of partitions of n with unique smallest part
and largest part at most twice the smallest part equals the number of parti-
tions of n in which the largest part is odd and the smallest part is larger than
half the largest part.
5. Let P{(r; n) denote the number of partitions of n into parts that are
either even and not congruent to 4r — 2 (mod 4r) or odd and congruent to
2r — 1 or 4r — 1 (mod 4r). Let P2(r\ n) denote the number of partitions of n
in which only even parts may be repeated and all odd parts are congruent
to 2r - 1 modulo 2/\ Then Px(r; n) = P2(r; n).
Comment on Examples 6-7. P. A. MacMahon introduced what he termed
"modular" partitions. Given the positive integers k and n, there exist (by
the Euclidean algorithm) h ^ 0 and 0 < j ^ k such that

n = kh + y.

The "modular" partitions are a modification of the Ferrers graph so that

n is represented by a row of h /c's and one j . Thus the representation of
8 + 8 + 7 + 7 + 6 + 5 + 2 to the modulus 2 is

2222
2222
2221
222 1
222
221
2
Note that the ordinary Ferrers graph is just the modular representation
with modulus 1.
6. Let Wx(r, /??, n) denote the number of partitions of n into m parts,
each larger than 1, with exactly r odd parts, each distinct. Let W2(r,m,n)
denote the number of partitions of n with 2m as largest part and exactly r

Cambridge Books Online © Cambridge University Press, 2010

14 The Elementary Theory of Partitions Chap. 1

odd parts, each distinct. Then Wx{r, m, n) = W2(r, m, n) (use modular

representations of the partitions in question with modulus 2).
7. Let P 3 (r; n) denote the number of partitions ( A ^ - • •/,) of n such that
if Af is odd, then/,. - Xi+l ^ 2r - 1 (1 < i < s, / s + 1 = 0). Then P 2 (r; n) =
Pi(r\ n) (see Exercise 5 for P 2 ( r » ")) ( u s e rnodular representation of the
partitions in question with modulus 2).
8. (Sylvester) The number of partitions of n with distinct odd parts
equals the number of partitions of n that are self-conjugate (i.e., identical
with their conjugate).
9. (MacMahon) Let Mx(n) denote the number of partitions of n into
parts, each larger than 1, such that consecutive integers do not both appear
as parts. Let M2(n) denote the number of partitions of n in which no part
appears exactly once. Then Mx{n) — M2(n).
10. (MacMahon) Let M3(N) denote the number of partitions of n into
parts not congruent to 1 or 5 modulo 6. Then M,(n) = M 3 (n) (see Example 9
for M2{n)).
11. (Euler) The absolute value of excess of the number of partitions n
with an odd number of parts over the number of those with an even number
of parts equals the number of partitions of n into distinct odd parts.

Notes

Many books on number theory or combinatorics present material of the

type chosen for this chapter, for example, Andrews (1971), Comtet (1974),
Hardy and Wright (1960), MacMahon (1916), Ostmann (1956), and Riordan
(1958). Euler's contributions are primarily found in Euler (1748). Glaisher's
theorem (Corollary 1.3) appears in Glaisher (1883). Theorem 1.5 can be
found in Sylvester's monumental paper (Sylvester, 1882-1884; see also G. W.
Starcher, 1930). Franklin's proof of the pentagonal number theorem appears
in Franklin (1881); see also Subbarao (1971a), Andrews (1972). Recurrence
relations (related to Corollary 1.8) and recent surveys are reviewed in Sec-
tions P56 and P02, respectively, of LeVeque (1974).
Example 1. Subbarao (1971b).
Example 4. Andrews (1967b).
Examples 5, 7. Andrews (1970).
Example 6. Andrews (1974), MacMahon (1923).
Examples 9, 10. Andrews (1967a).
Example 11. Sylvester (1882-1884).

References
Andrews, G. E. (1967a). "A generalization of a partition theorem of MacMahon,"
J. Combinatorial Theory 3, 100-101.
Andrews, G. E. (1967b). "Enumerative proofs of certain ^/-identities," Glasgow Math. J.
8, 33-40.
Andrews, G. E. (1970). "Note on a partition theorem."1 Glasgow Math. J. 11, 108-109.
Andrews, G. E. (1971). Number Theory. Saunders, Philadelphia.
Andrews, G. E. (1972). "Two theorems of Gauss and allied identities proved arithmet-
ically," Pacific J. Math. 41, 563-578.

Cambridge Books Online © Cambridge University Press, 2010

References 15

Andrews, G. E. (1974). "Applications of basic hypergeometric functions," S.I.A.M. Rev.

16, 441-484.
Bender, E. A. (1974). "A lifting theorem for formal power series," Proc. Amer. Math.
Soc. 42, 16-22.
Comtet, L. (1974). Advanced Combinatorics. D. Reidel, Dordrecht.
Euler, L. (1748). Introductio in analysin infinitorum, Chapter 16. Marcum-Michaelem
Bousquet, Lausannae.
Franklin, F. (1881). "Sur le developpement du produit infini (1 - .v)(l — x2)(\ - , v 3 ) . . . , "
Comptes Rendus 82, 448-450.
Glaisher, J. W. L. (1883). "A theorem in partitions," Messenger of Math. 12, 158-170.
Hardy, G. H., and Wright, E. M. (1960). An Introduction to the Theory of Numbers,
4th ed. Oxford Univ. Press, London and New York.
Hickerson, D. R. (1973). "Identities relating the number of partitions into an even and
odd number of parts," J. Combinatorial Theory A15, 351-353.
Hickerson, D. R. (1974). "A partition identity of the Euler type," Amer. Math. Monthly
81, 627-629.
Knutson, D. (1972). "A lemma on partitions," Amer. Math. Monthly 79, 1111-1112.
LeVeque, W. J. (1974). Reviews in Number Theory, Vol. 4. Amer. Math. Soc, Providence,
R.I.
MacMahon, P. A. (1916). Combinatory Analysis, Vol. 2. Cambridge Univ. Press, London
and New York (reprinted by Chelsea, New York, 1960).
MacMahon, P. A. (1923). "The theory of modular partitions," Proc. Cambridge Phil.
Soc. 21, 197-204.
Moore, E. (1974a). "Generalized Euler-type partition identities," /. Combinatorial
Theory A17, 78-83.
Moore, E. (1974b). "Partitions with parts appearing a specified number of times," Proc.
Amer. Math. Soc. 46, 205-210.
Netto, E. (1927). Lehrbuch der Kombinatorik, 2nd ed. Teubner, Leipzig (reprinted by
Chelsea, New York, 1958).
Ostmann, H. H. (1956). Additive Zahlentheorie, 2 vols. Springer, Berlin.
Rademacher, H. (1973). Topics in Analytic Number Theory. Springer, Berlin.
Riordan, J. (1958). An Introduction to Combinatorial Analysis. Wiley, New York.
Starcher, G. W. (1930). "On identities arising from solutions of ^/-difference equations
and some interpretations in number theory," Amer. J. Math. 53, 801-816.
Subbarao, M. V. (1971a). "Combinatorial proofs of some identities," Proc. Washington
State Univ. Conf. Number Theory pp. 80-91.
Subbarao, M. V. (1971b). "On a partition theorem of MacMahon-Andrews," Proc.
Amer. Math. Soc. 27, 449-450.
Sylvester, J. J. (1882-1884). "A constructive theory of partitions, arranged in three acts,
an interact and an exodion," Amer. J. Math. 5, 251-330; 6, 334-336 (or pp. 1-83 of
The Collected Mathematical Papers of J. J. Sylvester, Vol. 4. Cambridge Univ. Press,
London and New York, 1912; reprinted by Chelsea, New York, 1974).

Cambridge Books Online © Cambridge University Press, 2010

CHAPTER 2

Infinife Series Generating Functions

2.1 Introduction

In this chapter we shall consider the fruitful interaction between partition

theorems and certain elementary methods for manipulating infinite series and
products. The convergence questions that arose in Chapter 1 arise again;
as before, however, they pose few if any difficulties in our work.
We note that now we shall consider p(S, m, /?), the number of partitions
of n that lie in a set of partitions S and that have m parts. This immediately
leads to the two-variable generating function

m=0 n=0

AeS

where if X — (k^..., Ar), # ( / ) = r, G{A) = A, + • • • + kr. The double series

above converges absolutely for \z\ < \q\~l > 1. This may be easily seen by
proving (in the manner in which Theorem 1.1 is proved) that

£ £ p(y,m,«)zY = n(l -zqTX

m=0 n=0 n= 1
(
where - f is the set of all partitions.
In fact, the proof of Theorem 1.1 with slight modification may be employed
to prove that

£ £p("H",i»M!)zV = n ( l ---qnV\ (2.1.1)

m=0 n=0 neli

£ £p("H"(< d), m, n)zmq" = f [ ( l - zd+iqtd+l)n)(l - zqT1- (2-1-2)

m=0 n=0 neH

ENCYCLOPEDIA OF MATHEMATICS and Its Applications, Gian-Carlo Rota <ed.).

2, George E. Andrews, The Theory of Partitions
Copyright © 1976 by Addison-Wesley Publishing Company, Inc., Advanced Book Program.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical photocopying,
recording, or otherwise, without the prior permission of the publisher.
16

Cambridge Books Online © Cambridge University Press, 2010

2.2 Elementary Series-Product Identities 17

Other parameters besides the number of parts #(A) of a partition X will

interest us from time to time; so we shall have occasion to consider other
types of partition generating functions of several variables.
The preceding comments suggest the interest of considering infinite series
and products in two (or more) variables. In the following section, we shall
develop an elementary technique for proving many series and product
identities. We shall obtain several classical theorems of great importance,
such as Jacobi's triple product identity. As will become clear in Section 2.3,
the results of Section 2.2 are quite useful in treating partition identities. It is
possible, however, to skip Section 2.2 and read Section 2.3, referring back
only for the statements of theorems. For the reader who needs series trans-
formations to attack a partition problem, the first six examples at the end
of this chapter form a good test of the techniques used in Section 2.2.

2.2 Elementary Series-Product Identities

We begin with a theorem due to Cauchy; as we shall see, this result provides
the tool for doing everything else in this section.
THEOREM 2.1. / / \q\ < 1, \t\ < 1, then

t 4. V (1 ~ a)(l - " < ? ) ' • - d - a < ? w " 1 K ' _ fr (1 - atqn)

„=! (1 - <?)(! - q2)' ' '(I - qn) „ = <> ( 1 - tqn) ' {
' ' }

Remark. We shall try always to state our theorems with as little notational
disguise as possible. However, for the proofs, it seems only sensible to use
the following standard abbreviations

(a)m = (a; q)n = (1 - a)(l - aq)- • -(1 - aq"'1),

(«)o = 1-
We may define (a)n for all real numbers n by

(a). = (a)J(aqn)x.
The series in (2.2.1) is an example of a basic hypergeometric series. The
study of basic series (or ^-series, or Eulerian series) is an extensive branch of
analysis and we shall only touch upon it in this book. Most of the theorems
of this section may be viewed as elementary results in the theory of basic
hypergeometric series. Theorem 2.1 has become known as the "g-analog of
the binomial series/* for if we write a = qx where a is a nonnegative integer,

Cambridge Books Online © Cambridge University Press, 2010

18 Infinite Series Generating Functions Chap. 2.2

then (2.2.1) formally tends to

« la + n - 1\
i + Z( h» = (i -ty\ as g-r.
n=l\ « /
Proof. Let us consider

FU) = fl ( - ^ g = I AX (2.2.2)
where An = >ln(a, ^f). We note that the An exist since the infinite product is
uniformly convergent for fixed a and q inside |f| < 1 — 6, and therefore it
defines a function of t analytic inside \t\ < 1.
Now

(1aI)n
00
(1 - atan+1)
= (1 - fl/) n ^ ^ i — ^ = (1 - aOF(^). (2.2.3)

Clearly Ao = F(0) = 1, and by comparing coefficients of f in the extremes

of (2.2.3) we see that

or

. (1 — aqn~l) /<•* *> A\

A —. A I J 2mti\\

Iterating (2.2.4) we see that

(1 - aqn~l)(l - aqn'2)- • -(1 - a)A0

A =

Substituting this value for An into (2.2.2), we obtain the theorem. •

Euler found the two following special cases of Theorem 2.1. Each of these
identities is directly related to partitions in Example 17 at the end of this
chapter.

Cambridge Books Online © Cambridge University Press, 2010

2.2 Elementary Series-Product Identities 19

COROLLARY 2.2 (Euler). For \t\ < 1, \q\ < I,

1+ I rf"-)(1 -^iTTT-Ji—J*) = ft (1 " f«"r\ (2.2.5)

Proof. Equation (2.2.5) follows immediately by setting a = 0 in (2.2.1).

To obtain (2.2.6) we replace a by a/6 and f by bz in (2.2.1); hence for \bz\ < 1

.t-, (T-W?)" : "(T?)'"" " V

Now set b = 0, a = — 1 in (2.2.7) and we derive (2.2.6) directly. •

The following result is Heine's fundamental transformation, and it is

instrumental in proving each of the succeeding four corollaries.

COROLLARY 2.3 (Heine). For \q\ < 1, |/| < 1, \b\ < 1

aq)--(\ - aqn'l)(i - b)(\ - fe^)---(l - bqm'x)f

=n - bqm)(l -
^o (1 - e<T)(l -
atqm)

x (1 - ti/Xl ~ atq)-' (1 - at

Proof.

n»0 (<l)n(C)n (C)ao n = 0 (q)n

y y (a)S
.-o «-o (4).
{b)
2L f
(c)

. y (clb)m(t)mbm

Cambridge Books Online © Cambridge University Press, 2010

20 Infinite Series Generating Functions Chap. 2.2

COROLLARY 2.4 (Heine). / / \c\ < \ab\, \q\ < 1,

+
»?i (1 - </)(l - 4 2 )- • -(1 - qn)(\ - c)(l - cqY • -(1 - c ^ T

f f (I - cqmla)(l - cqm/b)
m = o (1 — cqm){\ — cqmjab)

Proof. By Corollary 2.3,

COROLLARY 2.5 (Bailey). / / \q\ < min(l, |6|), then

y (1
(
« (1 - a<? 2m+1 )(l + ^ + 1 ) ( 1 -aq2m + 2
lb2)

Proof. By Corollary 2.3 (interchanging a and fr)

£ (b)n\
2- — T :
g (q2lb2'q2)mam
(aqlb)x(-q/b)xm%

We remark that Corollary 2.4 is commonly referred to as the "^-analog

of Gauss's theorem," while Corollary 2.5 is the "^-analog of Kummer's
theorem."
COROLLARY 2.6. If \q\ < 1,

i+y *— -
.t*, (1 - flXl - <72)- • - d - 9"X1 - zXl - rfl)' - d - z'/"" 1 )

= fl (1 - 2 ^ m ) ' ' . (2.2.8)

Cambridge Books Online © Cambridge University Press, 2010

2.2 Elementary Series-Product Identities 21

1+ £ 2 ~-2 — = f[ (1 - qm)~l. (2.2.9)

Remark. Equation (2.2.8) is due to Cauchy, and Eq. (2.2.9) is due to Euler.

Proof. First we note that (2.2.9) is obtained from (2.2.8) by setting z = q.

In Corollary 2.4, seta = a"1,fc = /?" 1 ,c = z. Hence

. v ( a - 0(« - q)'' *(a - qn~l)(P - 0(0 - q)' • '(0 - qn~l)zn

(z) x (za/J). '

and if we set a = fi = 0 in this identity, we obtain (2.2.8).
COROLLARY 2.7. / / \q\ < 1,

. Set fe = ^"x in Corollary 2.5. Hence

Now set 0 = 0 in this identity and we obtain the desired result. •

The next result, Jacobi's triple product identity, may be viewed as a corollary
of Corollary 2.2; however, it is so important that we label it a theorem.

THEOREM 2.8. For z ^ 0, \q\ < 1,

£ z V 2 = flC 1 - ^ 2n+2 X 1 + zq2n+1)(l + z - y n +1

). (2.2.10)
n = — oo n—0

Proof. For |z| > \q\, \q\ < 1,

ft (1 + zq2n+i) = - | Z mgm
2 (by (2.2.6))
1 °°
/^,2. ^2\ L-t » *1 oo

W > q 7oc m = — oo

(since (^2m+2; q2)^ vanishes for m negative)

Cambridge Books Online © Cambridge University Press, 2010

22 Infinite Series Generating Functions Chap. 2.2

1 oo
= V
2 2
( q ; q ) o

m2 m
= ! y q z .
(q ;q )J-qiz;q2)oom=-ao
2 2

This is the desired result. Note that absolute convergence pertains everywhere
only so long as \z\ > \q\, \q\ < 1. However, the full result of the theorem
follows either by invoking analytic continuation, or by observing that the
entire argument may be carried out again with z" 1 replacing z. •

COROLLARY 2.9. For \q\ < 1,

/ |\n (2fc+l)n(n+l)/2-in/| _ (2n+l)i\

<" + 1 ) - ' ) . (2.2.11)

Proof. Replace q by qk + * and then set z = - qk+*~l in (2.2.10). This

substitution immediately yields the equality of the extremes in (2.2.11). Now

y (_ nw0<2fc+l)n(n+l)/2-in/j _ (2n+l)i\
n= 0

_ y / |\n^(2k+l)n(n+l)/2-/n _j_ y / |\n

= y / jy» (2Jk+l)n(fi+l)/2-m _^_ y / | \ n (2fc+1 )n(n+1 )/2 - in

n=0 n=- 1

00
= y / |\n (2lc+l)n(n+l)/2-/n M
n = — oo

We remark that Corollary 2.9 reduces to Corollary 1.7 when k = i = 1

once we observe that

Cambridge Books Online © Cambridge University Press, 2010

2.3 Applications to Partitions 23

n a - <?3n+3xi - <?3n+ixi - «3n+2) = n a - iny

COROLLARY 2.10 (Gauss)

Z ( - 1)V 3 = PI (I ~ ql>
:=
(2.2.12)
n "~oo HI= 1 v •" T /

oo °° (1 a^m}
A 1 (\ 2m — 1\

Proo/. By (2.2.10) with z = - 1,

where the final equation follows from (1.2.5). Next

00 00
</>+1)/2
Z^ = J I q"

where again the final equation follows from (1.2.5). •

So far this section seems filled with much mathematics and little commentary.
It has been the hope that the power of Theorem 2.1 and simple series manipula-
tion would be fully appreciated if numerous significant results followed in
rapid-fire order. The reader will have a chance to practice the techniques
involved in the many examples at the end of this chapter.

2.3 Applications to Partitions

We shall prove four theorems on partitions utilizing either the actual

results or the methods of Section 2.2. We conclude with an examination of
"Durfee squares," which allows us to obtain (2.2.9) from purely combinatorial
considerations. We begin with an interpretation of Corollary 2.9.
THEOREM 2.11. Let @(k, i) denote all those partitions with distinct parts
in which each part is congruent to 0, ± i (modulo 2fe + 1). Let pe(@(k, i)> n)
(resp. po(@(k, /), n)) denote the number of partitions of n taken from £^(/c, i)
with an even (resp. odd) number of parts. Then

Cambridge Books Online © Cambridge University Press, 2010

24 Infinite Series Generating Functions Chap. 2.3

Pt{2(k, 0, n) - i )o(^(/c, i), n)

I if n = (k \)m{m + 1) ± im,
b otherwise.

Proof. The proof is exactly like the proof of Corollary 1.7 done backwards.
Here we read the partition-theoretic result by comparing coefficients on both
sides of (2.2.11). •
THEOREM 2.12 (Sylvester). Let Ak(n) denote the number of partitions of n
into odd parts (repetitions allowed) such that exactly k different parts
occur. Let Bk(n) denote the number of partitions X = (Xly..., Xr) of n such
that the sequence (Xu.. ., Xr) is composed of exactly k noncontiguous
sequences of one or more consecutive integers. Then Ak(n) = Bk(n) for all
k and n.
Remarks. First of all, we note that this theorem of Sylvester is a refinement
of Euler's theorem (Corollary 1.2), in that p((9, n) = ^°° =0 Ak(n) and p(Q), n) =
££° =0 £fc(rt). Sylvester obtained a purely graphical proof of this result. Our
proof (due to Ramamani and Venkatachaliengar) illustrates nicely the way
in which combinatorics and formal series analysis can interact to facilitate
a proof.
The exact meaning of Ak(n) and Bk(n) should be crystal clear from the
following example: >43(14) = 7, since the relevant partitions are (1239),
(1257), (1327), (1437), (1352), (13325), (1635); and B3(14) = 7, since the
relevant partitions are (1, 3, 10), (149), (248), (158), (257), (1247), (1346).
Proof. We note that the method of proof of (1.2.3) in Theorem 1.1 may
be extended to show that

I I Ak(n)akq" = ft (1 + aq2J~l + aq2^-" + aq^2j-l) + • • •)

- q

M a-?2'1)
= ((1 ~ a) 1 ;<?2) °° - (d - a)q;q*U- ,)„ (2.3.1)

where the final equation follows from Euler's theorem (1.2.5).

It is not a simple matter to find directly the two-variable generating functions
for Bk(n). However, if we examine the conjugates of the partitions enumerated
by Bk(n), things simplify greatly. Actually the conjugate partitions X' fall

Cambridge Books Online © Cambridge University Press, 2010

2.3 Applications to Partitions 25

into two distinct classes: (1) if 1 is a part of A, then A' is a partition of n in

which the largest part is unique, every positive integer smaller than the largest
part appears, and exactly fc — 1 of these parts appear more than once;
(2) if 1 is not a part of A, then X' is a partition of n in which the largest part
is repeated, every positive integer smaller than the largest part appears, and
exactly k parts appear more than once.
Consequently

£ t Bk(n)a'q" = 1 + £ aq» "fl (qJ + aq" + aq*> + • • •)

* = 0 11 = 0 7V= 1 j=l

+ I (aq2N + aq3N + • • •) f[ (qj + aq2j + 0 ^ + • • •)

N=l j=l

= 1+ f

= ((1 - ato; « 2 ) e ( - q)a (by Corollary 2.7)

= Z I Ak(n)akqn (by (2.3.1)).

fc = 0 n = 0

Comparing coefficients of akqn in the extremes of the foregoing string of

equations, we deduce that Ak(n) = Bk(n) for all k and n. •

Now we shall consider an analytic approach to Theorem 1.5; generally,

analytic proofs require less ingenuity and provide less insight than com-
binatorial proofs.

Analytic Proof of Theorem 1.5. If we let pbtC(a — c) denote the number

of the first type of partition described in Theorem 1.5, then the theorem
asserts that pbtC(a — c) = pCtb(a — b). N o w

Z E E PbMxbycqa = 1 + t xy" fl (1 + xq1 + x2q2J + *V V + • • •)

a=0 6= 0 c=0 n=l j-l

n=l

Cambridge Books Online © Cambridge University Press, 2010

26 Infinite Series Generating Functions Chap. 2.3

Therefore if
00 X 00

a=0 6= 0 c=0

x
x vnnn
= 1+
n ? 1 7^)7'
then we need only show /(.v, y) = f(y, x) to obtain the desired result:

X
' ' ' " -~n% (q)n(0)»

(q)r> (x)Jyq2)., 2 (q)nxn

(by Corollary 2.3)

Therefore

y ...^_ - y x"(

As a third example we consider a second refinement of Eulefs theorem

(Corollary 1.2) due to N. J. Fine.

THEOREM 2.13. The number of partitions of n into distinct parts with

largest part k equals the number of partitions of n into odd parts such that
2A: -h 1 equals the largest part plus twice the number of parts.

Cambridge Books Online © Cambridge University Press, 2010

2.3 Applications to Partitions 27

Proof. Combinatorial reasoning of the type used in the previous three

proofs shows that Fine's theorem is equivalent to the following assertion:

00 j+l 2J+1 oo
1 L.L
j=o(tq> q ) j + l
= tq
j=
I( 0

Now

$(q2;q2)jqJtJ

00 fJnJ 1
= tq(q2li

fJaJ oo a2jm + 2m
-A- T, ^ ^ r (by (2.2.5))

tJqJ

= tq(q2; q2)*, £ — * (by (2.2.5))

m = 0 W ; ? )m('9 )oo

_tq(q2\q2)a0 ^ (tq;q2)m0q2;q2)mq2m

00
» 1 / o o V"1 (q2;q2)Jmq2m
?o
(by Corollary 2.3)

We conclude this chapter with a look at a property of partitions called the

Durfee square, and we utilize it to provide a new proof of (2.2.9).

Combinatorial Proof of Eq. (2.2.9). To each partition X = (Xl9 X2>..., Xr)

we may assign a parameter d{X) as the number of Xj such that Ay ^ 7. Let us
see what d(X) measures in the graphical representation of A. Suppose A =
(1242572), then d(X) = 4, the graphical representation is

Cambridge Books Online © Cambridge University Press, 2010

28 Infinite Series Generating Functions Chap. 2

and as we have indicated, d(k) measures the largest square of nodes contained
in the partition /.. This square is called the Durfee square (after W. P. Durfee),
and d(k) is called the side of the Durfee square. It is clear from the graphical
representation that if A I— n and d(k) = s, then the partition k may be uniquely
written as (s5) + k' + k" where (ss) counts the nodes in the Durfee square,
k' represents the nodes below the Durfee square (and is therefore some
partition all of whose parts are ^ s), and k" represents the conjugate of
the nodes to the right of the Durfee square and so k" is also some partition
whose parts are ^ 5. In the foregoing example the partition k = (1242572)
is uniquely written as (44) -f (124) 4- (223). Since partitions with parts ^ s
are generated by

*2)"-(l-*s) (q)5
(Theorem 1.1), we see that the set of all partitions with Durfee square of
side 5 is generated by

(q)s
Therefore

Wo

Examples

1. The following generalization of Corollary 2.3 is valid for each integer

fc *» 1,
(a;qkUb)knf ) » (cjb)n(t;qk)nbn
% (q)n(at;

Cambridge Books Online © Cambridge University Press, 2010

Examples 29

oo (h\. t2n f—th\ oo fh\ tm

2
- Z .-0 - Oo

4
z2)oom=o (q2;q2Ubt;q2)m

5. (1 + a) Z 7—-r- = Z T—_w __x •

oo Q
6
- ^ 77 m= 0

7. The identity used in the proof of Theorem 2.13 is a special case of the
one in Example 4.

a
(aq)2m+l
O. 2J ("" )m\a(l) ~ Is \a )m("~ aC
l) =
^
m=0 m=0 i

9. The identity in Example 8 may be used to prove another theorem of

N.J.Fine:
U2r+i(n)= V2r+l(n)+ V2r(n)
where U2r+i(n) is the number of partitions of n with odd parts and largest
part 2r 4- 1, and where Vs(n) is the number of partitions of n into distinct
parts such that the largest part minus the number of parts equals s.
10. It is possible to prove that

V (- 1

by showing that each of the series s(x, q) in question satisfies

s(x, ^f) = 1 - x2q - q2x3s(xq, q), s(0, <?) = 1.
11. Euler's pentagonal number theorem is a corollary of Example 10.
12. Example 10 can be proved in a purely combinatorial manner. The
approach must consider not only what number n is being partitioned, but
also m, the sum of the largest part and the number of parts. The invariance
of m in the transformations of the proof of Theorem 1.6 suffices to treat
the second equation in Example 10.
In his last letter to G. H. Hardy, Ramanujan considered several families
of functions which he called "mock-theta functions." Four of the fifth-order
mock-theta functions are

Cambridge Books Online © Cambridge University Press, 2010

30 Infinite Series Generating Functions Chap. 2

oo ^n2 °°
1= Zqn\-q;q2)n,

-q)., F0{q) =
n%{q\qz)n
Exercises 13 and 14 provide several relations among these functions:

13. «K(<7) =

- ( E <-1) 1 )' E r r - 2*o(-1 2 ).

\ \q)

^r4
n = 0 V "~ z
)n + 1
S(«^)«
n=0
16. If in Theorem 2.1 we seplace 4 by q2 and then replace f by tq2 and a
by — aq, then the resulting identity may be deduced from Example 6 of
Chapter 1.
17. If t is replaced by tq in Corollary 2.2, the resulting two equations have
simple combinatorial proofs. Equation (2.2.5) may be proved by the con-
sideration of partitions of n with m parts; Eq. (2.2.6) may be proved by the
consideration of partitions of n with m distinct parts.
18. Let us say a partition X is Sylvestered if the smallest part to appear
an even number of times (0 is an even number) is even. For each set of
positive integers T, define S(T; m, n) as the number of Sylvestered partitions
of n into m parts such that no elements of T are repeated. Then

n /i V2_2fc\ oo 2n-lnn(2n-l)

£ 117(1
19. The number of non-Sylvestered partitions of n with an odd number
of parts equals the number of non-Sylvestered partitions of n with an even
number of parts.

20. Z ( - l)mS(T; m, n)qn = U (1 - q2k) I—1— - l

Notes
The origin of Theorem 2.1 is uncertain; it has been attributed to Euler
(Hardy, 1940, p. 223). Heine (1847) was the first to study systematically this
type of series, and so Theorem 2.1 is often attributed to him; however, Cauchy
proved this result in 1843 (see Cauchy, 1893, p. 45). Corollary 2.2 is truly
due to Euler (1748). Corollaries 2.3 and 2.4 are due to Heine (1847). W. N.
Bailey (1941) and J. A. Daum (1942) independently discovered Corollary 2.5.
Corollary 2.6 is due to Cauchy (1893). Corollary 2.7 appears to be due to
V. A. Lebesgue (1840). Corollary 2.8 is the celebrated Jacobi triple product
identity (Jacobi, 1829); the proof given here was found independently by

Cambridge Books Online © Cambridge University Press, 2010

References 31

Andrews (1965) and P. K. Menon (1965). Equations (2.2.12) and (2.2.13) are
generally attributed either to Jacobi (1829) or to Gauss (1866). Theorem 2.12
is an extension of Euler's theorem (Corollary 1.2) due to Sylvester (1884-
1886); the proof we give is in essence due to V. Ramamani and K.
Venkatachaliengar (1972; see also Andrews, 1974). Theorem 2.13 is an
unpublished result due to N. J. Fine (1954; see also Andrews, 1966b). The
use of Durfee squares has been studied extensively in Andrews (1971b, 1972a).
The relationship between series and product identities has not been exten-
sively treated in books. Certain elementary problems are broached in Andrews
(1971a; see also 1972b). The advanced theory of these series may be found
in Andrews (1974), Bailey (1935), Hahn (1949, 1950), and Slater (1966).
Much of the recent literature is reviewed in Section P60 of LeVeque (1974).
Examples 1-3. Andrews (1966c).
Example 4. Andrews (1966b).
Example 5. Andrews (1966a. c).
Examples 6, 7. Andrews (1966b).
Example 8. This identity is originally due to N. J. Fine (1954), and it
appears in Andrews (1966b).
Example 9. Andrews (1966b), Fine (1948).
Example 10. Rogers (1916), Andrews (1966b), Fine (1954).
Example 12. Subbarao (1971), Andrews (1972a).
Examples 13, 14. Watson (1936), Andrews (1966a).
Example 15. Carlitz (1967).
Example 16. Andrews (1974).
Example 18. Andrews (1970). In this paper, Sylvestered partitions are
called "flushed" in accordance with Sylvester's original work. Since Sylvester
was one of the all-time champion coiners of new mathematical terms, it
seems appropriate to call flushed partitions "Sylvestered."
Examples 19, 20. Andrews (1970).

References
Andrews, G. E. (1965). "A simple proof of Jacobi's triple product identity," Proc. Amer.
Math. Soc. 16, 333-334.
Andrews, G. E. (1966a). "On basic hypergeometric series, mock theta functions, and
partitions, I," Quart. J. Math. Oxford Ser. 17, 64-81.
Andrews, G. E. (1966b). "On basic hypergeometric series, mock theta functions, and
partitions, II," Quart. J. Math. Oxford Ser. 17, 132-143.
Andrews, G. E. (1966c). "^-Identities of Auluck, Carlitz and Rogers," Duke Math. J.
33, 575-582.
Andrews, G. E. (1966d). "On generalizations of Euler's partition theorem," Michigan
Math. J. 13, 491-498.
Andrews, G. E. (1970). "On a partition problem of J. J. Sylvester," J. London Math. Soc.
(2) 2, 571-576.
Andrews, G. E. (1971a). Number Theory. Saunders, Philadelphia.
Andrews, G. E. (1971b). "Generalization of the Durfee square," /. London Math. Soc.
(2) 3, 563-570.
Andrews, G. E. (1972a). "Two theorems of Gauss and allied identities proved arith-
metically," Pacific J. Math. 41, 563-578.
Andrews, G. E. (1972b). "Partition identities," Advances in Math. 9, 10-51.

Cambridge Books Online © Cambridge University Press, 2010

32 Infinite Series Generating Functions Chap. 2

Andrews, G. E. (1974). "Applications of basic hypergeometric functions,'* S.I.A.M. Rev.

16, 441-484.
Bailey, W. N. (1935). Generalized Hypergeometric Series. Cambridge Univ. Press, London
and New York (reprinted by Hafner, New York, 1964).
Bailey, W. N. (1941). "A note on certain ^-identities," Quart. J. Math. Oxford Ser. 12,
173-175.
Carlitz, L. (1967). "Problem 66-9: A pair of identities," S.I.A.M. Rev. 9, 254-256.
Cauchy, A. (1893). Oeuvres, Ser. 1, Vol. 8. Gauthier-Villars, Paris.
Daum, J. A. (1942). "The basic analog of Kummer's theorem," Bull. Amer. Math. Soc.
48, 711-713.
Euler, L. (1748). Introductio in analysin infinitorum, Chapter 16. Marcum-Michaelem
Bousquet, Lausannae.
Fine, N. J. (1948). "Some new results on partitions," Proc. Nat. Acad. Sci. USA 34,
616-618.
Fine, N. J. (1954). "Some Basic Hypergeometric Series and Applications," unpublished
monograph.
Gauss, C. F. (1866). Werke, Vol. 3. Konigliche Gesellschaft der Wissenschaften,
Gottingen.
Hahn, W. (1949). "Beitrage zur Theorie der Heineschen Reihen, Die 24 Integrale der
hypergeometrischen <7-Differenzengleichung, Das ^-Analogon der Laplace Trans-
formation," Math. Nachr. 2, 340-379.
Hahn, W. (1950). "Uber die hoheren Heineschen Reihen und eine einheitliche Theorie
der sogenannten speziellen Funktionen," Math. Nachr. 3, 257-294.
Hardy, G. H. (1940). Ramanujan. Cambridge Univ. Press, London and New York
(reprinted by Chelsea, New York).
Hardy, G. H., and Wright, E. M. (1960). An Introduction to the Theory of Numbers,
4th ed. Oxford Univ. Press, London and New York.
Heine, E. (1847). "Untersuchungen uber die Reihe . . . , " J. Reine Angew. Math. 34,
285-328.
Jacobi, C. G. J. (1829). Fundamenta nova theoriae functionum ellipticarum. Regiomonti,
fratrum Borntriiger (reprinted in Gesammelte Werke, Vol. 1, pp. 49-239; Reimer,
Berlin, 1881).
Lebesgue, V. A. (1840). "Sommation de quelques series," / . Math. Pures Appl. 5, 42-71.
LeVeque, W. J. (1974). Reviews in Number Theory, Vol. 4. Amer. Math. S o c , Providence,
R.I.
Menon, P. K. (1965). "On Ramanujan's continued fraction and related identities,"
J. London Math. Soc. 40, 49-54.
Ramamani, V., and Venkatachaliengar, K. (1972). "On a partition theorem of Sylvester,"
Michigan Math. J. 19, 137-140.
Rogers, L. J. (1916). "On two theorems of combinatory analysis and some allied
identities," Proc. London Math. Soc. (2) 16, 315-336.
Slater, L. J. (1966). Generalized Hypergeometric Functions. Cambridge Univ. Press,
London and New York.
Subbarao, M. V. (1971). "Combinatorial proofs of some identities," Proc. Washington
State Univ. Conf. Number Theory, pp. 80-91.
Sylvester, J. J. (1884-1886). "A constructive theory of partitions arranged in three acts,
an interact, and an exodion," Amer. J. Math. 5, 251-330; 6, 334-336 (or pp. 1-83
of the Collected Papers of J. J. Sylvester, Vol. 4, Cambridge Univ. Press, London
and New York, 1912; reprinted by Chelsea, New York, 1974).
Watson, G. N. (1936). "The mock theta functions (2)," Proc. London Math. Soc. (2),
42, 274-304.

Cambridge Books Online © Cambridge University Press, 2010

CHAPTER 3

Resfricfed Partitions and Permutations

3.1 Introduction

In Chapters 1 and 2, we studied various partition problems utilizing infinite

series and infinite products. In the application of partitions (such as in
statistics), we are often interested in restricted partitions, that is, partitions
in which the largest part is, say, ^ N and the number of parts is ^ M. In this
chapter, we shall examine restricted partitions. This undertaking will naturally
lead us to the Gaussian polynomials and from there we shall be led to
questions concerning permutations.

3.2 The Generating Function for Restricted Partitions

Let p(N, My n) denote the number of partitions of n into at most M parts,

each ^ N. Clearly

, M, n) = 0 if n > MAT,

p(N, M, NM) = 1.

Therefore the generating function

is a polynomial in q of degree NM.

THEOREM 3.1. For M, N ^ 0

/1 W + A/\/< N + A/ — 1 \ /i A/ + 1 \
G(N, M; q) = —

(3.2.1)

ENCYCLOPEDIA OF MATHEMATICS and Its Applications, Gian-Carlo Rota (ed.).

2, George E. Andrews, The Theory of Partitions
Copyright © 1976 by Addison-Wesley Publishing Company, Inc., Advanced Book Program.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical photocopying,
recording, or otherwise, without the prior permission of the publisher.

33

Cambridge Books Online © Cambridge University Press, 2010

34 Restricted Partitions and Permutations Chap. 3.2

Proof. Let g(N, M; q) denote the right-hand side of (3.2.1); then

,0;q) = g(0,M;q) =1 (3.2.2)

and

g(N, M\q)- g(Ny M - 1; q) = % ^ ± [(1 - qN + M) - (1 - qM)

= qMg(N- l,M;q). (3.2.3)

Note that (3.2.2) and (3.2.3) uniquely define g(N, M; q) for all nonnegative
integers M and N (a fact easily proved by a double mathematical induction
on N and M).
On the other hand,

l if ]V = M = n = 0,
(3.2.4)

since the empty partition of 0 is the only partition in which no part is positive
and is also the only partition in which the number of parts is nonpositive.
Equation (3.2.4) means that

G(N,0;q) = G(0,M;q) = 1. (3.2.5)

Furthermore, p(N, M, n) — p(N, M — 1, n) enumerates the number of

partitions of n into exactly M parts, each ^ N. We transform each of these
partitions by deleting every 1 that is a part, and subtracting 1 from each part
larger than 1. The resulting partitions of n — M have at most M parts and
each part is ^ N — 1. Since the foregoing transformation is clearly reversible,
it establishes a bijection between the partitions enumerated by p(Ny M, n) —
p(N, M — 1, n) and those enumerated by p(N — 1, M, n — M). Therefore

, M, n) - p(N, M - 1, n) = p(N - 1, M, n - M), (3.2.6)

and translating (3.2.6) into a generating function identity, we obtain

G(N, M; q) - G(N, M - 1; q) = qMG(N - 1, M; q). (3.2.7)

Thus since g(Ny M; q) and G(N, M; qr) satisfy the same initial conditions
((3.2.2) and (3.2.5) resp.) and the same defining recurrence ((3.2.3) and
(3.2.7) resp.), they must be identical. Therefore

Cambridge Books Online © Cambridge University Press, 2010

3.3 Properties of Gaussian Polynomials 35

3.3 Properties of Gaussian Polynomials

The polynomials G(N, M; q) appearing in Theorem 3.1 were first studied

by Gauss and have come to be known as Gaussian polynomials. In this
section we shall derive a number of useful formulas for these polynomials.
DEFINITION 3.1. The Gaussian polynomial [£] is defined by

if 0 ^ m
0 otherwise.

Note that

= G(N -

by Theorem 3.1. We shall, however, not make use of this fact in this section.

THEOREM 3.2. Let 0 ^ m ^ n be integers. The Gaussian polynomial [£]

is a polynomial of degree m(n — m) in q that satisfies the following relations.

•U-l! (3.3.1)

n
(3.3.2)
n —m
n - n— 1
(3.3.3)
m m - 1
n - 1 n- 1
(3.3.4)
m - 1 m

n\ n
lim (3.3.5)
m\(n — m)\ m

Proof. Equations (3.3.1) and (3.3.2) are obvious from Definition 3.1. As for
(3.3.3), we see that

n- 1
[(1 - q") - (1 - q"-m)l
m

n- 1

Cambridge Books Online © Cambridge University Press, 2010

36 Restricted Partitions and Permutations Chap. 3.3

Equation (3.3.4) follows by replacing m by n — m in (3.3.3) and applying

(3.3.2).
The fact that [£] is a polynomial in q of degree m(n — m) follows by
induction on n using (3.3.1) and (3.3.4).
Finally

- q)

n n —\ n —m + I
m m — 1 1

n\
m\(n — m)\ \m

Many times the Gaussian polynomials arise due to their relationship with
certain finite products.
THEOREM 3.3

(3.3.6)

z'. (3.3.7)

Proof. By Theorem 2.1,

(<?),

>=o

. (-

which is (3.3.6).
Again by Theorem 2.1,

-1
^ (qN)jZi

N
7j _

Cambridge Books Online © Cambridge University Press, 2010

3.3 Properties of Gaussian Polynomials 37

There are many other formulas related to the Gaussian polynomials. We

shall conclude this section with a few of the most useful ones.

THEOREM 3.4

// m = 2n,
(3.3.8)
;=o // m is odd;

+ m 4- 1"
for m, n ^ 0; (3.3.9)
m +1 | j= 0

m _(n-/c)(ft-/c) _ m
(3.3.10)
h-k

M - ml \N + m| | m + n + r (N-r)(M-r-m) _ m
m + r\\ M+ N

(3.3.11)

Proof. We begin with (3.3.8), an identity important in the evaluation of

Gaussian sums. Let/(m) denote the left-hand side of (3.3.8); then

f{m)z ,m co m

m=O
i(-iy

(-

« « ( - iy z m
^•=0 m =o (q)j(q)m

Z E^
(by (2.2.5))

- I (by (2.2.5)).

Equation (3.3.8) now follows by comparing coefficients of zm in the extremes of

the foregoing string of equations.
We may prove (3.3.9) by induction on n. If n = 0, the equation reduces
to 1 = 1. Assuming the result true for a specific n, we see that by (3.3.4)

Cambridge Books Online © Cambridge University Press, 2010

38 Restricted Partitions and Permutations Chap. 3.3

n + m + 21 \n + m + 11 fn + m + 1
9

To obtain (3.3.10) (the ^-analog of the Chu-Vandermonde summation) we

compare coefficients of zh on both sides of the identity

(z)m+n =

Instead of proving (3.3.11), we prove the more general

v (a)n(b)n(q-N)nqn _ (c/a)N(clb)N
nf0(qUc)n(abqi-»c-\ ~ (c)N(dab)N * ^ ' ^

Equation (3.3.12) was first proved by F. H. Jackson and is called the ^-analog
of Saalschutz's theorem; Eq. (3.3.12) reduces to (3.3.11) if we make the sub-
stitutions a = q~M+m9 b = qm+n+l, c = qm+1, and then simplify.
To obtain (3.3.12), we must utilize an identity that is easily deduced from
Corollary 2.3.

<by Corollar, 2.3)

£ (abtlcUbUc/b)-

(by Corollary 2.3)

(c/bUbt)x $

Cambridge Books Online © Cambridge University Press, 2010

3.4 Permutations and Gaussian Multinomial Coefficients 39

(c!b)Jbt)K
cjb)^ n% (q)n(c)n

(abt!c)v Z

Let us multiply the extremes of this string of equations by {^Jiabtjc)^ and

then compare coefficients of ts on each side. Thus

(a)n(b)n(c/ab)N.n fab\N'n (c/a)N{c/b)NaNbNc~N

) '
N N N
Multiplying both sides of (3.3.14) by (q)Na b c l(c/ab)N and simplifying,
we obtain (3.3.12). •

3.4 Permutations and Gaussian Multinomial Coefficients

The Gaussian polynomial is often called the "4-binomial coefficient" due
to (3.3.5). In this section we shall use Gaussian polynomials to study certain
types of permutation problems, and we shall be naturally led to the treat-
ment of ^-multinomial coefficients:
DEFINITION 3.2. For m , , . . . , mr ^ 0, we define the Gaussian multinomial
coefficient (or q-multinomial coefficient) by

mt (3Ai)
Note that
[ i v 1 "1 - (^r-w •
n
m ,n — m
Now when r = 2, Theorem 3.1 tells us that

{ + m2

x, m2

is the generating function for p(mu m2, n\ the number of partitions of n with
at most m2 parts, each ^ mx.
We shall now consider a different type of mathematical object with the
same enumerative function p(mly m2, n).
DEFINITION 3.3. A multiset is a set with possibly repeated elements.
To be quite correct we might define a multiset as an ordered pair (A/,/)
where M is a set a n d / is a function from M to the nonnegative integers; for
each m e M,f(m) would be called the multiplicity of m. When M is a finite

Cambridge Books Online © Cambridge University Press, 2010

40 Restricted Partitions and Permutations Chap. 3.4

set, say {ml9 m 2 , . . . , m r }, we shall write

( M , / ) = {m/ ( m i ) m 2 / ( W 2 ) - • - m / } .

Let us begin by considering permutations of multisets (a permutation of

(M, / ) is a word in which each letter belongs to M and for each m e M the
total number of appearances of m in the word is/(m)). Thus 3 2 1 2 2 3 2 1 1 2
is a permutation of the multiset {I 3 2 5 3 2 }.
DEFINITION 3.4. We let inv(m 1 , m 2 , . . . , m r ; n) denote the number of
permutations ^ 2 - • -£ m i + m 2 + ... + m r of {r*2 m2 - • •r1"'-} in which there are
exactly n pairs (£,., £y) such that i < j and £f > £,-.
THEOREM 3.5. m v ^ , m 2 ; n) = p(mu m2, n).

Proof. We shall define a bijection between the permutations enumerated

by inv(m l9 m2; n) and the partitions enumerated by p(ml9 m 2 , n).
Let us use the "box" graphical representation for the Ferrers graph of a
partition with each part ^ 1 1 and at most seven parts (here 8 + 6 + 6 + 1 + 1).

m = 11
r -> ^ _. ^

11""
m2=7<
— i )

We follow the path indicated by the dots, starting with the upper right node
and moving to the left and downward: if the path moves vertically, we write
a 2 and if horizontally we write a 1. Hence the sequence corresponding to
this graph is 1 1 1 2 1 1 2 2 1 1 1 1 1 2 2 1 2 2.
Notice that the number of Vs to the right of the first 2 tells us the largest
part of our partition; the number of Ts to the right of our second 2 tells us
the second part of our partition, and in general the number of l's to the right
of the /th 2 tells us the ith part of our partition. Clearly the above relationship
between partitions and permutations establishes a bijection between the
permutations of {lmi2m2} with n inversions and the partitions of n with at
most m2 parts, each ^ mA. Hence, mv^?^, m2\n) = p(mi> m2> w). H

Cambridge Books Online © Cambridge University Press, 2010

3.4 Permutations and Gaussian Multinomial Coefficients 41

Theorem 3.5, together with Theorem 3.1, yields the special case in which
r = 2 of the following general result:

THEOREM 3.6. For r ^ 1,

mx -f ni2 * • • -r m
(3.4.2)

Remark. This theorem is due to P. A. MacMahon.

Proof. We proceed by induction on r. When r = 1, the result in Eq. (3.4.2)

clearly reduces to I = 1. When r = 2, we see that

1 1 1 , , /?!•>

by Theorems 3.5 and 3.1.

Now in general

mr\ n) = ^ invf/?^ mr.. ,, mr;j)

7=0

•inv(/??!, r_!; « - j). (3.4.3)

To see (3.4.3), let us examine those permutations in which exactly j of the n

inverted pairs have r as first element. The easiest way to construct all such
permutations is as follows: first, take a permutation of { l m i ™ 2 + ••• + «,.-,2«-}
with j inversions; next replace each 2 by an r and then replace the m{ -f
m2 4- • • • + m r _ ! appearances of 1 by a permutation of { l m ' 2 m : • • • ( > — l) m r ~'}
with n — j inversions. Since the two choices of permutations were totally
independent, we see that the number of permutations of {l m '2 m 2 - • • rmr} with
r appearing in exactly j of the inverted pairs is just

-f r . l ; n — j).

Summing on all j ^ 0, we obtain (3.4.3).

It is now a simple matter to prove (3.4.2). It is true for /• = 1 and 2, and
assuming it true for r, we have by (3.4.3)

X inv(ml,...,mr+l\n)qn

+ • • • + mr, mr+l;j) irwim^. . ., mr\ n — j)

mr, mr+l'J)qJ £ inv(m { ,. . ., mr: k)qk

j=Q k=0

Cambridge Books Online © Cambridge University Press, 2010

42 Restricted Partitions and Permutations Chap. 3.4

+m r +

mr

and the desired result follows by mathematical induction. •

Let us now introduce a further parameter related to multiset permutations.

DEFINITION 3.5. We let ind(ml5 m 2 , . . . , mr\ n) denote the number of

permutations {^ 2 - • •fmi + ...+BIr of {l mi 2 m2 - • -rmr} for which

£ Xiid = n

+mr 1
where /(£,) = *"if£»> £i+i and #(£,.) = 0 otherwise. The sum £™= T ~ X(£»)
is called the greater index of the permutation.

Thus the greater index of 1 1 1 2 1 1 2 2 1 1 1 1 1 2 2 1 2 is 0 + 0 + 0 + 4 +

0 + 0 + 0 + 8 + 0 + 0 + 0 + 0 + 0 + 0 + 15 + 0 = 27.
We now obtain a second surprising result of MacMahon:

THEOREM 3.7. For r ^ 1

£ ind(m!, m 2 , . . . , mr\ ri)qn = X 2 r

. (3.4.4)

Proo/. We shall actually prove that

:•""'• ••'"""fr" = , „ ^ 1 , , _ • (3.4.5)

We begin by considering the right-hand side of (3.4.5). We recall from

Theorem 1.1 that (q)~tl is the generating function for partitions in which no
part exceeds m,, and by Theorem 1.4 it is also the generating function for
partitions with at most mi parts. Now each partition with at most mf (positive)
parts corresponds to a partition with exactly m{ nonnegative parts (just fill
in the necessary number of zeros). Therefore the coefficient of qN in

is the number of arrays

Cambridge Books Online © Cambridge University Press, 2010

3.4 Permutations and Gaussian Multinomial Coefficients 43

> • • • > ami

b2 >--->bm2 > 0

x2 >•••> xmr ^ 0 (3.4.6)

where l a , + Ib{ + • • • + Ix-X = N.

Now we define a one-to-one correspondence between such arrays and
ordered pairs (TT, <T) where cr is a permutation of {l mi 2 mz - • -rmr} and n is a
partition with exactly mx + m2 + • • • + mr nonnegative parts. For convenience
we let M = mt + m2 + h mr.
The correspondence is constructed thus: we take the partition n: At 4-
/1 2 + • • • + v4M and write directly below it the permutation a = £i£2'' €M:

AIA2A3- - -AM

We now form an array resembling (3.4.6) by placing each A{ (starting with

At) in the £4th row. This construction is inadequate as it stands since we have,
for example, when n = (2334) and a is a permutation of {1322}

4 3 2 2 2 4 2 2
1 2 1 1 2 ~* 3 2
4 3 2 2 2 4 2 2
—•
1 2 1 2 3 2
4 3 2 2 2 4 2 2
1 2 2 1 1 "* 3 2
However, we may make the mapping a bijection by the specification that
whenever a fall occurs in the permutation (i.e., £,- > £ J+1 ) we must also have
At > Ai+i. Examining the above three pairs (*), we find that the first pair
is the only one that fits this requirement. Clearly the mapping is now one-to-
one, since we may take any array of type (3.4.6) with the specified inequalities
and successively insert the second row into the first, the third row into the
first, and so on, always making our insertion as far to the right as possible
consistent with the nonincreasing order; this way of inserting guarantees that
the " > " will always appear whenever an element of one row appears just
to the left of one of a lower-numbered row.
For example, let us work our insertion process on

5 3 2 2 5 3 3' 3' 2 2 2' 1' 0

3' 3' 2' V - 2''i"i" "> 5 3 3 ' 3 ' 2 2 2 ' 2 " l ' l " l " 0 .
2" V 1"

Cambridge Books Online © Cambridge University Press, 2010

44 Restricted Partitions and Permutations Chap. 3.4

The appropriate permutation is 1 1 2 2 1 1 2 3 2 3 3 1 , and the mapping we

originally constructed yields

53332222 1110 _^ 5 3 2 2 0
1122 1 1 2 3 2 3 3 1 ""* 3 3 2 I
2 1 1
as desired.
To prove that we have a partition n = (A1A2m • • AM) with strict inequality
at the appropriate terms (specified by the falls in the permutation a =
Ci^2*" *CM)» w e start with an arbitrary partition

n0 = {axa2- • -aM)>
and we define

K = (a{ + </>!, C?2 + 0 2 , fl3 + 0 3 ,. . ., ClM + 0vf)

where $, is the number of falls in {<;,-<;,•+! • • 'CM)- Note t n a t 0i + ^2 + ' ' ' +
(f)M = y(^{) -1- ^(<^2) + • • • + yX£M), the greater index of o*, since on the left-
hand side the fall £; > c,-+i is counted exactly / times.
Thus our mapping actually provides a bijection between arrays (3.4.6)
whose total sum is N and ordered pairs C^0) where a is a permutation of
{l m '2 m2 - • -rmr} with greater index g(a) and n0 is a partition of N - g with at
most M parts.
Therefore

1
£ <j*. — = : ! (3.4.7)
() ^L^L (7)
But =
X«r ^^ Xn^o indfrnj, m2»- • •, nW n)qn> a n d so
(3.4.5) is established, as
desired. •

To make our correspondence clear we exhibit in full the case in which

N = 3, m, = 3 , m2 = 2.

Array (3 .4.6)
0
0 0 0 3 0 0 0 0 2 0 0 0 0 2 1 1 1 2
3 0 2 1 1 1 2

3 0 0 3 0 0 0 0 3 0 0 0 0 1 1 1 2 2
0 0 I 1 1 2 2

2 1 0 2 1 0 0 0 2 1 0 0 0 1 1 1 2 2
0 0 1 1 1 2 2

Cambridge Books Online © Cambridge University Press, 2010

3.5 The Unimodal Property 45

2 0 0 2 l 0 0 0 1 0 0 0 0 1 2 1 1 2
1 0 1 2 1 1 2

1 0 0 2 1 0 0 0 1 1 0 0 0 2 1 1 1 2
2 0 2 1 1 1 2

0 0 0 2 1 0 0 0 1 0 0 0 0 2 2 1 1 1
2 1 2 2 1 1 1

1 1 1 1 1 1 0 0 1 1 1 0 0 1 1 1 2 2
0 0 1 1 1 2 2

1 1 0 i 1 1 0 0 0 0 0 0 0 1 1 2 1 2
1 0 i 1 2 1 2

1 0 0 l 1 1 0 0 0 0 0 0 0 1 2 2 1 1
1 1 l 2 2 1 1

COROLLARY 3.8

ind(m!, m 2 ,. .., mr; n) = inv(m1, m 2 ,.. ., mr\ n).

Proof. Compare coefficients of qn in the identity

= X ind(mi9 m 2 , . . ., mr\ n)qn,

n=0

which is valid in light of Theorems 3.6 and 3.7. •

The subject matter of this section has been greatly extended by D. Foata
and R. P. Stanley. We note in passing that Foata has provided a purely
combinatorial proof of Corollary 3.8.

3.5 The Unimodal Property

In many applications, we are interested in the distribution of values of

functions like p(ml, m2; n). We shall prove a simple result that will allow
us to provide relevant information on such questions in numerous cases.

DEFINITION 3.6. A polynomial p(q) = a0 + axq + • • • + anqn is called

reciprocal if for each i, a{ = an_it equivalently qnp(q~i) = p(q).
3.7. A polynomial p(q) = a0 + aYq + • • • + anqn is called
DEFINITION
unimodal if there exists m such that

Cambridge Books Online © Cambridge University Press, 2010

46 Restricted Partitions and Permutations Chap. 3.5

a0 ^ a{ ^ a2 ^ - " ^ am^ am+l ^ am + 2 ^ • • • ^ an.

THEOREM 3.9. Let p(q) and r(q) be reciprocal, unimodal polynomials with
nonnegative coefficients; then p(q)r(q) is also a reciprocal, unimodal
polynomial with nonnegative coefficients.

Proof. Let p(q) = a0 + atq + • • • + anq\ r(q) = b0 + bxq + • • • + bmqm,

and let
s(q) = P(q)r(q) = c0 + c{q + • • • + c n + m ^ n + fn
where

with the convention that a7- = 0 if j < 0 or j > n and bj = 0 if j < 0 or

7 > wi.
Now by (3.5.1) all the c( are nonnegative. Since

we see that s(g) is reciprocal.

Finally, to prove that s(g) is unimodal, we begin by noting that aj — a;-_ t ^ 0
for all j ^ w/2 and 6 y - 6 y _ t ^ 0 for all j ^ m/2; this is because each of
p(q) and r(q) is both unimodal and reciprocal. Now

2(cy - Cj_{) = X a ^ - , . 4- X tfn-i+ifcy-n + i - i

i = — TO i = — co

i = - QO i = - oo

TO

= Z («,- - ^i-iXfcy-,- - by-n + i - i ) (since a,- = an-k)

Cambridge Books Online © Cambridge University Press, 2010

3.5 The Unimodal Property 47

(note that if (n + l)/2 is an integer, ain+l)/2 - fl<n-n/2 = 0)

nil

Hence
n/1
(3.5.2)

Since 0 < 1 ^ #i/2, we see that a,- - a , . ! ^ 0. If; - i ^ m/2, then since
n + 1 > 2/, we see that m/2 ^ 7 — / > j — M -f / — 1; hence foy_f —
6 J _ n + |._1 ^ 0 in this case. If j — i > m/2, then for 0 < 7" ^ (m + n)/2, we
see that m + n + I > 2j9 and so m/2 > m — 7 + i > j; — n + 1 — 1;
therefore in this second case bj-i — bJ-n+i-l = bm_j+i — fcy_ll+l-_1 ^ 0.
Hence, in any case, both factors of each term on the right-hand side of
(3.5.2) are nonnegative provided that 0 ^ j ^ (m + n)/2. Therefore, since
s(q) is reciprocal, we see that it is also unimodal. •
THEOREM 3.10. For all N, M,n ^ 0

p(N9 M, n) = p(M, N9 n); (3.5.3)

p(AT, M, n) = p(AT, M, NM - n); (3.5.4)

p(N, M, n) - K^» M, n - 1) ^ 0 /or 0 < n ^ M/2. (3.5.5)

Proof. By Theorem 3.1,

, A#, « V = G(N,

Equation (3.5.3) follows from the fact that

N +
\
[ M
is symmetric in N and M.
The degree of G(N, M; q) is just

+ Af + 1\ IN + 1\ /Ml

and

,-2\
'(I -

Cambridge Books Online © Cambridge University Press, 2010

48 Restricted Partitions and Permutations Chap. 3.5

(<7) .\+.\f = G(N9M;q).

Hence G(NyM;q) is a reciprocal polynomial of degree M ; therefore

(3.5.4) is true.
Now (3.5.5) is the unimodal property and unfortunately its proof (at least
with current knowledge) lies outside of the theory of partitions proper. It has
been proved through invariant theory, where it is shown that p(N, M; n) —
p(N, M; n — 1) is the number of linearly independent semi-invariants of
degree M, weight n, and extent not exceeding N (I. J. Schur, Vorlesungen
iiber Invariantentheorie, Satz 2.22, p. 76, Grundlehren der Mathematischen
Wissenschaftetu Vol. 143). To my knowledge no simple combinatorial proof
of (3.5.5) is known. •

THEOREM 3.11. For all mu m 2 , . . . , mr, n ^ 0

ind(wi 1? ..., mr\ n) = ind(m tl ,. . . , mir\ n)

= inv(m 1 ,..., m r ; n) = inv(m Jt ,. . ., mir\ n) (3.5.6)

where {il9..., ir} is a permutation of {1, 2 , . . ., r } ;

/?i 1 ,..., m r ; n) = ind(mj,..., mr\ S — n)

= inv(m 1 ,. . ., mr\ n) — inv(m 1 ,. .., m r ; 5 — n) (3.5.7)

where S = is the second elementary symmetric function of

the /?!,•;

ind(m 1 ,..., mr\ n) — i n d ( m t , . . . , mr\ n — 1)

= \nv(mi9.. ., mr\ n) — inv(m 1 ,..., m r ; n — 1) 0 (3.5.8)

rO<n^ S/2.

Proof. Since

!,. . ., mr\ n)qn = u. . ., mr\ n)qn

m{ -h * *'
lYl
2

Cambridge Books Online © Cambridge University Press, 2010

Examples 49

we see that

is symmetric in the mt (whence (3.5.6)) and is the product of unimodal reciprocal

polynomials (by Theorem 3.10) and is therefore unimodal and reciprocal by
Theorem 3.9. Equations (3.5.7) and (3.5.8) immediately follow once we
observe that the degree of

\ml + m2 4- * * * 4- mr

is just

i 4- • • • 4- mr 4- 1\ hn1 4- t\ fm2 4- l\ hnr 4-

2 ) - [2 ) - [2 ) " " " [ 2
m m
= Z i j = s- •

Examples

1. The following finite form of Jacobi's triple product identity may be

deduced from Eq. (3.3.6):

where [™]q2 is the ordinary Gaussian polynomial with q replaced by q2.

2. Jacobfs triple product identity (Theorem 2.8) may be deduced from
Example 1.
In Examples 3-9, the polynomials Hn(t) are the Rogers-Szego polynomials

HJLt) = 1 t*.
7= 0

4. Identity (3.3.8) is deduced from Example 3 in the case in which t = — 1.

5. //,(**) = (-**;**)„.

6. Hn+l(t) = (I + 0 ^ ( 0 - (1 - qn)tHn^(t).

7. Utilizing Example 6 and the recurrence formula for the Gaussian

polynomials (3.3.3), we may prove by induction on m that

Cambridge Books Online © Cambridge University Press, 2010

50 Restricted Partitions and Permutations Chap. 3

m\ \n
HJt)HHV) = I

8. From Examples 3 and 7 it follows that

(tyz)»

9. From Example 8, we may deduce that

y Hn(t)Hn(s)xn _

10. Let [ x ] denote the greatest integer function, then we may prove

n + 1 —a —j W+ I
I «'2 h = — ao

(where a = 0 or 1), by showing that the assertion is true for n — 0 and 1

and that each side satisfies the recurrence/,, = / n _ 1 + (\nfn-i-
11. The Rogers-Ramanujan identities may be deduced from Example 10,
namely, for a = 0 or 1,

12. Let Dn denote the polynomial defined in Example 10 when a = 0.

These polynomials satisfy the identities

and Dn =

13. Let t h e 4t /cth excess" o f the partition A = ( A , / 2 ' * Ar) d e n o t e A, - A k + I .

T h e generating function for partitions with j parts w h o s e first excess is at
m o s t i is

(1 -q)(\ ~q2)'(\ ~qj)

14. Example 13 may be extended to show that for any k < jy the generating
function for partitions with j parts and with £th excess at most i is
(t cl2
-g'*'Xl - l l )--(l-qi*k)qJ
(1 -q)(\ -q2)-(l -qj)

15. Theorem 3.1 is a corollary of Example 14.

16. The number of permutations of {lm»2m2* • rm-} with even greater index
equals the number with odd greater index if and only if at least two m, are odd.
17. We may define a several-variable analog of the Rogers-Szego polyno-
mials by

Cambridge Books Online © Cambridge University Press, 2010

Notes 51

The related generating function is

18. A protruded partition of H is a decreasing sequence of positive integers

Aj ^ A2 ^ • • • ^ / r > 0, together with a sequence of nonnegative integers
satisfying 0 ^ /i, ^ /, for 1 ^ / ^ r, such that !"(/, 4- AO = n- (The [ii are
the protrusions of the ordinary partition ( A ^ - • •/.,). The product

- q j - q J + l
-'"-

is the generating function for the protruded partitions with each k{ ^ m.

19. From (3.3.13) (with a = x~*, b = x~±, c = ^ 2 /(l - q), t replaced
by xzq2/(\ — q)) it follows when x -> 0 that

-q2)'"(\ -qn)(l -q-q2)'"(l -q -qn+l)

Notes

Gauss (1863, p. 16) introduced Gaussian polynomials, and he observed

the facts in Theorem 3.2. Theorem 3.1 was perhaps first stated by Sylvester
(1884-1886); however, it may be trivially deduced from (3.3.6) and (3.3.7)
identities of Cauchy (see Cauchy's Collected Works, 1893, p. 46). Equation
(3.3.8) is due to Gauss (1863) and was used by him in the treatment of
Gaussian sums; (3.3.9) is also due to Gauss (1863). Equation (3.3.10) is
actually a special case of Corollary 2.4 due to Heine (1847); actually it is
the ^-analog of the celebrated Chu-Vandermonde summation (R. Askey
1975a; see also 1975b, p. 60) and our proof parallels the standard elementary
proof of that summation. As we remarked in the text, (3.3.11) is due to
F. H. Jackson (1910); one of its most important applications was by G. N.
Watson (1929), and it has also been applied in R. P. Stanley's (1972) recent
work in combinatorics.
All the work in Section 3.4 can be found in important works by
P. A. MacMahon (1913, 1914, 1915-1916, 1916) which foreshadowed much
of the recent work in the combinatorial theory of permutations. It would
lead us afield to delve deeply into recent work; however, we should mention
the work of Foata (1965, 1968), Foata and Schutzenberger (1970), and R. P.
Stanley (1972). Interestingly enough, MacMahon's proof of Theorem 3.7 is
of principal importance for Stanley's (1972) theory of (P, co)-partitions
(Lemma 6.1 and Theorem 6.2). Applications of (P, co)-partitions to permu-
tation problems are well described by Stanley (1972, Section 25).

Cambridge Books Online © Cambridge University Press, 2010

52 Restricted Partitions and Permutations Chap. 3

The material in Section 3.5 comes from Andrews (1975).

Examples 1, 2. See Hermite (1891, pp. 155-156).
Examples 3-9. Carlitz (1956), Szego (1926), Rogers (1893a, b).
Examples 10, 11. Andrews (1970).
Example 12. Andrews (1974).
Examples 13-15 are due to F. Franklin and appear in Sylvester (1882-1884).
Example 16. MacMahon (1913, 1915-1916).
Examples 18-19. Stanley (1972, Section 24). In Section 23 of Stanley's
memoir, he considers stacks and K-partitions, and these objects also give
rise to identities related to Corollary 2.3.

References

Andrews, G. E. (1970). "A polynomial identity which implies the Rogers-Ramanujan

identities," Scripta Math. 28, 297-305.
Andrews, G. E. (1974). "Combinatorial analysis and Fibonacci numbers," Fibonacci
Quart. 12, 141-146.
Andrews, G. E. (1975). "A theorem on reciprocal polynomials with applications to
permutations and compositions," Amer. Math. Monthly 82, 830-833.
Askey, R. A. (1975a). "A note on the history of series," Math. Res. Center Tech.
Summary Rept. No. 1532, Madison, Wisconsin.
Askey, R. A. (1975b). Orthogonal Polynomials and Special Functions, No. 21. Regional
Conf. Ser. Appl. Math., SIAM, Philadelphia.
Carlitz, L. (1956). "Some polynomials related to theta functions," Ann. Math. Pura
Appl. (4) 41, 359-373.
Cauchy, A. (1893). Oeuvres, Ser. 1, Vol. 8. Gauthier-Villars, Paris.
Foata, D. (1965). "Etude algebrique de certains problemes d'Analyse Combinatoire
et du Calcul des Probability," Publ. Inst. Statist. Univ. Paris 14, 81-241.
Foata, D. (1968). "On the Netto inversion number of a sequence," Proc. Amer. Math.
Soc. 19, 236-240.
Foata, D., and Schutzenberger, M. P. (1970). Theorie geometrique des polynomes
Euleriens (Lecture Notes in Math. No. 138). Springer, New York.
Gauss, C. F. (1863). Werke, Vol. 2. Konigliche Gesellschaft der Wissenschaften,
Gottingen.
Heine, E. (1847). "Untersuchungen uber die Reihe . . . , " J. Reine Angew. Math. 34,
285-328.
Hermite, C. (1891). Oeuvres, Vol. 2. Gauthier-Villars, Paris.
Jackson, F. H. (1910). "Transformations of <7-series," Messenger of Math. 39, 145-151.
MacMahon, P. A. (1913). "The indices of permutations and the derivation therefrom of
functions of a single variable associated with the permutations of any assemblage
of objects," Amer. J. Math. 35, 281-322.
MacMahon, P. A. (1914). "The superior and inferior indices of permutations," Trans.
Cambridge Phil. Soc. 29, 55-60.
MacMahon, P. A. (1915-1916). Combinatory Analysis, 2 vols. Cambridge Univ. Press,
London and New York (reprinted by Chelsea, New York, 1960).
MacMahon, P. A. (1916). "Two applications of general theorems in combinatory
analysis: (1) to the theory of inversions of permutations; (2) to the ascertainment
of the numbers of terms in the development of a determinant which has amongst
its elements an arbitrary number of zeros," Proc. London Math. Soc. (2) 15, 314-321.

Cambridge Books Online © Cambridge University Press, 2010

References 53

Rogers, L. J. (1893a). "On a three-fold symmetry in the elements of Heine's series,"

Proc. London Math. Soc. 24, 171-179.
Rogers, L. J. (1893b). "On the expansion of certain infinite products," Proc. London
Math. Soc. 24, 337-352.
Stanley, R. P. (1972). "Ordered structures and partitions," Mem. Amer. Math. Soc.
119.
Sylvester, J. J. (1882-1884). "A constructive theory of partitions in three acts, an interact,
and an exodion," Amer. J. Math. 5, 251-330; 6, 334-336 (or pp. 1-83 of the
Collected Papers of J. J. Sylvester, Vol. 4, Cambridge Univ. Press, London and
New York, 1912; reprinted by Chelsea, New York, 1974).
Szegb, G. (1926). "Ein Beitrag zur Theorie der Thetafunktionen," S.B. Preuss. Akad.
Wiss. Phys.-Math. Kl. 242-252.
Watson, G. N. (1929). "A new proof of the Rogers-Ramanujan identities," /. London
Math. Soc. 4, 4-9.

Cambridge Books Online © Cambridge University Press, 2010

CHAPTER 4

Compositions and Simon Newcomb's

Problem

4.1 Introduction

The first three chapters treated elementary properties of partitions. Composi-

tions are merely partitions in which the order of the summands is considered.
For example, there are five partitions of 4: (4), (13), (22), (122), (I 4 ); there
are eight compositions of 4: (4), (13), (31), (22), (112), (121), (211), (1111).
We shall find that compositions of vectors (or compositions of multipartite
numbers) also have intrinsic interest and important applications, especially
to permutation problems of the Simon Newcomb type.

4.2 Compositions of Numbers

DEFINITION 4.1. We let c(m, n) denote the number of compositions of n with

exactly m parts.
THEOREM 4.1

'n-l\ (n - 1)!
c(m, n) —
\m - 1/ (m - l ) ! ( n - m ) !
First proof of Theorem 4.1. The argument used to prove Theorem 1.1
may be easily adapted to show that

- q)m
£ Ir + m - 1\
— qm L, \ )qr (by the binomial series)
r=o \ r I
'/!- 1\ «> / « -
n=m \ « - my
ENCYCLOPEDIA OF MATHEMATICS and Its Applications, Gian-Carlo Rota (ed.).
2, George E. Andrews, The Theory of Partitions
Copyright © 1976 by Addison-Wesley Publishing Company, Inc., Advanced Book Program.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical photocopying,
recording, or otherwise, without the prior permission of the publisher.

54

Cambridge Books Online © Cambridge University Press, 2010

4.2 Composition of Numbers 55

Comparing coefficients of qn in the extremes of (4.2.1), we obtain the desired

result. •

Second proof of Theorem 4.1. We introduce a graphical representation

for the compositions of n. To the composition (ala2- ' -am) of n we associate
m segments of the interval [0, w]; the first segment is of length au the second
of length a2, and so on. Thus the composition (3 2 3 1 2) of 11 is represented
as

*—•—•—x—•—x—•—•—x—x—•—»
0 I 2 3 4 5 6 7 8 9 10 II
We now observe that we may construct each of the c(m, n) compositions
of n with m parts by choosing m — 1 of the first n — 1 integers as end points
for the m segments dividing [0, ri\. Since these choices can be made in (^Z\)
ways, we see that

n- 1
m — 1

DEFINITION 4.2. We let c(N, M, n) denote the number of compositions of

n with exactly M parts, each ^ N.

Clearly c(N, M, n) = c(M, n) whenever N ^ n.

Interestingly enough, c(N, M, n) possesses symmetry properties resembling
those described in Theorem 3.10 for p(N, M, n).

THEOREM 4.2. For all N, M, n ^ 1

c(N9 M, n) = c(N, M,MN + M - n); (4.2.2)

c(N, M, n) - c(N, M, n - 1) ^ 0 for 0 < n ^ M(N + l)/2. (4.2.3)

Proof. Following the argument in the first proof of Theorem 3.1, we see
directly that

^ c(N, M, n)qn = (q + q2 + • • • + qN)M

Thus £n;*o c (^» ^ » n ) ^ " ^s a product of M unimodal, reciprocal polynomials

with nonnegative coefficients (the leading coefficient was not necessarily
assumed to be nonzero); hence, by repeated application of Theorem 3.9,
Z/ȣo c(N9 M, n)qn is a unimodal reciprocal polynomial with nonnegative
coefficients. Equations (4.2.2) and (4.2.3) follow at once from this fact. •

Cambridge Books Online © Cambridge University Press, 2010

56 Compositions and Simon Newcomb's Problem Chap. 4.2

The simplicity of Theorem 4.1 allows us to obtain sharp asymptotic

estimates for certain partition functions by relating them to composition
functions. This approach is due to H. Gupta (1942).
DEFINITION 4.3. Let pM(n) denote the number of partitions of n with
exactly M parts.
Clearly ^ = 0 pM(n) = p{n\ and p(n, M, n) - p(n, M - 1, n) = pM{n).
THEOREM 4.3 (Erdos-Lehner). As n -> oo

1 [n-V

provided that M = o(n 1/3 ).

Proof. To each partition of n with exactly M parts we may associate the
set of compositions of n made up of the same parts. There are M ! such
compositions if all the parts in the partition are distinct; otherwise there are
less. Hence
M\pM(n)>c(M,n)
or

^ l(;-1) (4.2.4)

On the other hand, if qM{n) denotes the number of partitions of n with M

parts all distinct, then

qM(n + \M{M - 1)) = pM(n). (4.2.5)

The truth of (4.2.5) may be seen by the correspondence

(A, + M - 19A2 + M -2,A3 + M- 3 4)^(4^2 4)

where Ax ^ A2 ^ • • • ^ AM and Y, ^i = n-
Now each partition enumerated by qM(n) corresponds to exactly M!
compositions under the association described in the first paragraph of this
proof. Therefore
M\qM(n)^c(M,n).
Hence
= qM(n + \M(M - 1))

~ c{M, n + \M{M - 1))

M!

(4 2 6)
M - l • ' -

Cambridge Books Online © Cambridge University Press, 2010

4.3 Vector Compositions 57

Combining (4.2.4) and (4.2.6), we see that

< PMM (n + jM(M- Q-

(n- 1\ " (n- l)!(n 4- \M{M - 3))!
M\[M- l)
- Aw - M + l)r(^ + \M(M - 1))
/"(n)r(n + \M(M - 3) + 1)
~n~M+\n + \M(M - \))M~l (provided M = o(n1'2))

= 1+

M(M -
= exp \{M - 1) log I 1 t-

\M(M - I) 2 lM2{M- I) 3
= exp{ + O .
[ I n \ n2

and this last expression approaches 1 as n -* co provided M = o(n 1/3 ). •

It is reasonably clear that numerous asymptotic questions for partition

functions may be attacked by devices similar to those utilized above.

4.3 Vector Compositions

The title of this section would have been "Compositions of Multipartite

Numbers" if it had been chosen by P. A. MacMahon; however, the title
"Vector Compositions" is more self-explanatory and reflects more recent
usage. The vectors we shall consider will just be ordered r-tuples of nonnegative
integers not all of which are zero.
DEFINITION 4.4. A partition of (a t , a 2 , . . . , ar) is a set of vectors
( j J / 0 , . . . , j8r(<)), i ^ i ^ s (order disregarded), such that £ - = 1 (jV'V • -Jrii}) =
(a l s a 2 ,.. ., ar) (here as explained earlier all vectors have nonnegative integral
coordinates not all zero). If the order of the parts is taken into account, we
call (/?! ( 1 ) ,..., /? r ( 1 ) ),..., (/V s \. • - /Vs)) a composition of (z,..., as).
DEFINITION 4.5. We let P=(y^ a 2 , . . . , a r ; m) denote the number of partitions
of (a t , a 2 , . . . , ar) with m parts, and we let c(nu a 2 ,. . ., ir\ m) denote the
number of compositions of (a t , a 2 ,. . ., ar) with m parts.
Thus P=(2, 1, 1; 2) = 5, since there are five partitions of (2, 1, 1) into two
parts: (2, 1, 0)(0, 0, 1), (2, 0, l)(0, I, 0), (2, 0, 0)(0, 1, 1), (1,1, OKI, 0, 1),
(1, 1, 1)(1, 0, 0); c(2, 1, 1; 2) = 10 since each of the five partitions produces
two compositions.

Cambridge Books Online © Cambridge University Press, 2010

58 Compositions and Simon Newcomb's Problem Chap. 4.3

DEFINITION 4.6. We let P(OLX,. .., ar) (respectively c(<xu..., ar)) denote the
total number of partitions (respectively compositions) of (ocl, a 2 , . . . , ar).
For later convenience we define c(0, 0 , . . . , 0) = \.
NotethatXm^i^=(ai>-. .,a r ;m) = P f o , . . .,a r ),
c(alf...,ar).
THEOREM 4.4

P r o o / . W e begin by observing that for m > 1,

X c ( a , , . . . , « r ; m ) r 1 " « •••*;•

not all zero

not all zero

riXi-l2')---(l-r) " l' ' (4 32)

-
Hence

I c(a,,...,a,;»!)«!««•••r,'
ai,...,a r ^0
not all zero

i+
t u 1
(l - t,Xl - ' ;
i 1
(1 - 'i)(l - '; (1 -tr)1
h+ 1
1

1 - (1 - <i)0 2 )-(l
2(1- (i-t

as desired. •

Now Eq. (4.3.2) provides a means for obtaining a reasonably simple formula
forc(a 1 ,...,a r ;m).

Cambridge Books Online © Cambridge University Press, 2010

4.4 Simon Newcomb *s Problem 59

THEOREM 4.5. For m > 0,

/ ^ £/ nlM/«i
c(a1,...,ar;m) = £ ( - 1)' 1
/a 2 + m — i — 1\ /a r 4- m — i — 1
x

Proof. By Eq. (4.3.2) (assuming c(0,..., 0; m) = 0)

X c ( a , , . . . , a,; m ) / , * ^ - • • * / • •

-1
d - / , x i -t2y- -d-o

. ( - i)'(i - /,)- m + i (i - t2ym+i- • -(i - g -

- 1 ( 7 ) ( - !>• i o f " + " ; ' " ')<,••• • - i j - + \ - '•"'

ar + m - i - 1\
u
" " ' (4.3.4)

Comparing coefficients of tl<Xlt2<X2' * -tr*r in the extremes of (4.3.4), we obtain

the desired result. •

As an example of Theorem 4.5, we note that

/3\ / 2 \ / 2 \ /2\ /2\ / 1 \ /1\ / 1 \ /0\ /0

= 3-2-2 - 2-1-1-1 + 0 0 0 = 10.

4.4 Simon Newcomb's Problem

The problem we shall treat now is one that P. A. MacMahon solved neatly
and completely in 1907. The problem and its history were succinctly described
by MacMahon in Combinatory Analysis, Vol. 1 (1915, p. 187):

The problem was suggested to the late Professor Newcomb by a game of

"patience" played with ordinary playing cards which he found to be a
recreation in the few hours that he could spare from astronomical work. It
may be stated as follows:

Cambridge Books Online © Cambridge University Press, 2010

60 Compositions and Simon New comb's Problem Chap. 4.4

A pack of cards of any specification is taken —say there are mx cards

marked 1, m2 cards 2, /?i3 cards 3, and so o n - a n d being shuffled is dealt
out on a table; so long as the cards that appear have numbers that are in
ascending order of magnitude, equality of number counted as ascending
order, they are placed together in one pack, but directly as the ascending
order is broken a fresh pack is commenced and so on until all the cards have
been dealt. The probability that there will result exactly m packs or at most
m packs is required.

We may easily translate this problem into one involving permutations of a

multiset:

DEFINITION 4.7. Let S(m{, /?i 2 ,..., mr\ n) denote the number of permuta-
tions of {lm22m2- • -rmr) with exactly n runs (a run in the permutation
Ci£ 2 ' ' ' Cm, + • - + mr i s a maximal subsequence of the form qt ^ c l + 1 ^ c i + 2 <

We shall have the answer to Simon Newcomb's problem if we can find a

closed expression for N(m{, m 2 , . . . , mr\ n).

LEMMA 4.6. Each of the following relations implies the other:

(r - n + j
tbn.j for all n>U (4.4.1)

bn = " x ( " n +
•/) ( - lVfln_y /or a// n > 1. (4.4.2)
i=o \ j J
Proof. We begin by observing that we need only show that (4.4.2) always
implies (4.4.1), since once this assertion is established the reverse implication
follows by considering bn' = {— l)nbn and an' = (— l)nan.
Now assuming (4.4.2), we see that

r - n +j\n-±-i/r - n
j

j+fc^n-l\ J
J,k>0

"-l " In - r - 1\ (r - n + h
= !«.-*(- 1)* I . . .
/. = o 7=0 \ J ) \ h - j

Cambridge Books Online © Cambridge University Press, 2010

4.4 Simon New comb's Problem 61

(h - n
>„-*(-i

(by the Chu-Vandermonde summation; i.e., (3.3.10)

with g = 1)

as desired. •

LEMMA 4.7

""l [xl 4- a 2 + • • • 4- ar - n 4- A
c ( a ! , a 2 , . . . , ar; n) = 2- I • MV(al5. . ., ar; w - ;).
y=o \ J J
(4.4.3)
Proof. The identity (4.4.3) is easily established once we provide appropriate
alternative combinatorial interpretations for

c(ccl9 a 2 , . . . , ar; w) and N(a l 9 a 2 , . . . , a r ; n).

First we see that c ( a ! , . . . , ar; H) enumerates the number of ways that ocl
balls labeled 1, a 2 balls labeled 2, . . . , ar balls labeled r may be distributed
among n labeled boxes so that no box is empty. To see this we note that the
composition

of ( a ^ . . . , ar) corresponds to the distribution that places exactly /^ 0 ) balls

labeled / in they'th box.
A very similar interpretation can be given to N ( a l s . . . , ar; n). Indeed the
jth run in a permutation may be viewed as specifying the elements to be
placed in the7th box. It is important to note that the function N(ctl9..., ar; n)
does not enumerate all possible distributions (as does c ( a 1 ? . . . , ar; n)), but
only those for which the smallest element in each box is strictly smaller than
the largest element in the preceding box. Therefore to each permutation
enumerated by N(ocl,..., ar; n — j) we may associate a unique set of distribu-
tions enumerated by c ( a t , . . . , a r ; n) as follows: First place a vertical line
between each of the n — j runs in the given permutation. There are now
a
i 4- a 2 4- • • • + ar — (n — 7' — 1) — I pairs c,^ i + 1 that have no vertical bar
between them; place a vertical bar between j of these pairs and we have a
distribution of ccl ones, a2 twos, and so on, into n labeled nonempty boxes.
Since the final set of j vertical bars may be placed in

Cambridge Books Online © Cambridge University Press, 2010

62 Compositions and Simon Newcomb's Problem Chap. 4.4

ways, we see that exactly this many of the distributions enumerated by

c(cct,.. ., a r ; n) correspond to a single permutation enumerated by
N(OL1J. . . , a r ; n — j). Hence

c ( a , , . . . , a r ; w) = £ . N(a x ,. . . , a r ; n - j)
j = o\ J J
as desired. •

THEOREM 4.8. For n > 0,

s=0 \ S

In 4- a r — 1 — 5

Proof. By applying Lemma 4.6 to Eq. (4.4.3), we see that

N(<xl9...,ar;n)
n— I /Q/ _f_ • • • _|_ Qf — tl ~\~ j \
= Z^~1)7 .' I c(a t a r ; n — j)
A \ 7 /

7=0 \ J I i=0

4- n - j - i - 1\ / a r 4- w - j - i - 1
(by Theorem 4.5)

,_, /a, -b • • • + ar - n + A In — j \
(4A5)
* S, ( J i )•
i^O^0

Now

Cambridge Books Online © Cambridge University Press, 2010

Examples 53

A - /t - 2\
= ( - \y( s I (by (3.3.9) with q = 1)

A + 1\

1. (4.4.6)

Consequently, applying (4.4.6) to (4.4.5), we obtain (4.4.4). •

Examples
1. Let Fn be the nth Fibonacci number: F o = 0, F, = 1, F n = F n . , + F rt _ 2
for n > 1. The number of compositions of n in which no Ts appear is F w _,.
2. More generally, let kFn be defined by kF0 = • • • = kFk-2 = 0, k F k _ , = 1,
and k F n = k F n _ , -f *F n _ k . The number of compositions of n in which all
parts are ^ /c is equal to kFn_x.
3. Example 2 implies that there are 2"~ * compositions of M.
4. Let ck(my n) denote the number of compositions of n into exactly m
parts, each ^ k. Then

//j - (k - l)m - 1
, n) = [
5. Examples 2 and 4 imply that

n - (k - \)m - 1
m —l

6. Let us say ( w , , . . . , /?ir) is of real dimension h if mfc i=- 0, m f c + l = 0,

. . . , / n r = 0. Then the total number of ail compositions of all vectors
(P\*- • -i Pr) W ' t h 0 ^ Pi ^ a, in which exactly hi parts have real dimension
/ is

«r-l

7. Example 6 implies that

c(ai,...,ar) = 2
h\ >0

r >0
+ /1,+ • • • + / i r \ /ar.,

^2

Cambridge Books Online © Cambridge University Press, 2010

64 Compositions and Simon Newcomb's Problem Chap. 4

8. The Long operator L, is a linear operator on polynomials in x, over the

reals:

vai+i -f • • • + ar — k

One can prove by induction on j that

*, + y
xi+l + • • * + ar — A:

9. If the polynomial

i=2

is fully expanded and then each x,fc is replaced by

the result is c ( a , , . . . , ar).

The foregoing assertion is merely that
r

c ( a , , . . . , ar) = L,L 2 • • -L^^* 1 "" 1 J~J {

i=2

We prove this through the expansion of each factor by the binomial theorem
and then invoking Example 8 to reduce the identity to Example 7.
10. By expanding (1 4- xt)j through the binomial theorem, we may deduce
the Chu-Vandermonde (Eq. (3.3.10) when q = 1) summation from Example 8.

)-(
+ i + • • • + ar - fc - .s7 \ a , + 1 + • • • + ar - k
11. The case of q = 1 of Eq. (3.3.11) can be deduced easily through the
use of Long operators. We write the desired result equivalently as

If we consider linear operators L{ and L 2 , defined by

n \ / /i -h v
2
n — r) ' \m + n + f.i — s

then Example 8 implies that

Cambridge Books Online © Cambridge University Press, 2010

References 55

In + A / // -h v + u
L,(.xr(l + x)') = and L2(y\l + v)u) = ^ ^
\n - r) \m + « + // — .s
From here it follows that

v (m\(n + ^V^ + v + r
rto \r)\n - r/\m + n +
= L t L 2 ((i + x r + m ( i + x ^ ( l ^x)" 1 )'")
/( + v
j7o\J/\m + n + n - j / \ n - j

(by Example 10).

Notes

The material in this chapter is primarily an extended and modernized

version of portions of P. A. MacMahon's (1894, 1908) work on compositions
(see also Section IV of MacMahon, 1915). Theorem 4.2 is due to Z. Star
(1976; see also Andrews, 1975a), who also obtains nice asymptotic formulas
for the c(Ny M, n). Theorem 4.3 is due to P. Erdos and J. Lehner (1941);
the proof we present is by H. Gupta (1942).
Work on vector compositions and the Simon Newcomb problem has been
extensive in recent years. The papers listed in the references by Carlitz, Dillon,
Foata, Kreweras, Roselle and Schiitzenberger provide some of the primary
references. We should also point out that the (P, o>)-partitions of R. P. Stanley
are applicable to compositions and Simon Newcomb type problems (see
Stanley, 1972, Section 25). Some of the papers reviewed in Section P80 of
LeVeque (1964) concern compositions.
Example 1. Cay ley (1876).
Examples 6-10. Long (1970), Andrews (1975b, 1976a, b).

References

Andrews, G. E. (1975a). "A theorem on reciprocal polynomials with applications to

permutations and compositions," Amer. Math. Monthly 82, 830-833.
Andrews, G. E. (1975b). **The theory of compositions, II: Simon Newcomb's problem,'*
Vtilitas Math. 7, 33-54.
Andrews, G. E. (1976a). "The theory of compositions, I: The ordered factorizations
of n and a conjecture of C. Long," Canadian Math. Bull. 18, 479-484.
Andrews, G. E. (1976b). 'The theory of compositions, III: The MacMahon formula
and the Stanton-Cowan numbers," Vtilitas Math. 9, 283-290.
Carlitz, L. (1959). "Eulerian numbers and polynomials," Math. Mag. 33, 247-260.

Cambridge Books Online © Cambridge University Press, 2010

66 Compositions and Simon New comb's Problem Chap. 4

Carlitz, L. (1960). "Eulerian numbers and polynomials of higher order," Duke Math. J.
21, 401-424.
Carlitz, L. (1964). "Extended Bernoulli and Eulerian numbers,'* Duke Math. J. 31,
667-690.
Carlitz, L. (1972a). "Enumeration of sequences by rises and falls: A refinement of the
Simon Newcomb problem," Duke Math. J. 39, 267-280.
Carlitz, L. (1972b). "Sequences, paths, and ballot numbers," Fibonacci Quart. 10, 531-
549.
Carlitz, L. (1972c). "Eulerian numbers and operators," The Theory of Arithmetic Func-
tions (A. A. Gioia and D. L. Goldsmith, eds.) (Lecture Notes in Math. No. 251).
Springer, New York.
Carlitz, L. (1973a). "Enumeration of a special class of permutations by rises," Publ.
Elek. Fak. Univ. Beogradu 451, 189-196.
Carlitz, L. (1973b). "Enumeration of up-down permutations by number of rises,"
Pacific J. Math. 45, 49-58.
Carlitz, L. (1973c). "Enumeration of up-down sequences," Discrete Math. 4, 273-286.
Carlitz, L. (1973d). "Permutations with prescribed pattern," Math. Nachr. 58, 31-53.
Carlitz, L. (1974a). "Permutations and sequences," Advances in Math. 14, 92-120.
Carlitz, L. (1974b). "Up-down and down-up partitions," Proc. Eulerian Series and
Applications Conf. Pennsylvania State Univ.
Carlitz, L., and Riordan, J. (1955). "The number of labeled two-terminal series-parallel
networks," Duke Math. J. 23, 435-446.
Carlitz, L., and Riordan, J. (1971). "Enumeration of some two-line arrays by extent,"
J. Combinatorial Theory 10, 271-283.
Carlitz, L., and Scoville, R. (1972). "Up-down sequences," Duke Math. J. 39, 583-598.
Carlitz, L., and Scoville, R. (1973). "Enumeration of rises and falls by position," Discrete
Math. 5, 45-59.
Carlitz, L., and Scoville, R. (1974). "Generalized Eulerian numbers: Combinatorial
applications," /. Reine Angew. Math. 265, 110-137.
Carlitz, L., and Scoville, R. (1975). "Enumeration of up-down permutations by upper
records," Arch. Math. {Basel)
Carlitz, L., Roselle, D., and Scoville, R. (1966). "Permutations and sequences with
repetitions by number of increases," /. Combinatorial Theory 1, 350-374.
Cayley, A. (1876). "Theorems in trigonometry and on partitions," Coll. Math. Papers
of A. Cayley 10, 16.
Dillon, J. F., and Roselle, D. (1968). "Eulerian numbers of higher order," Duke Math. J.
35, 247-256.
Dillon, J. R, and Roselle, D. (1969). "Simon Newcomb's problem," SIAM J. AppL
Math. 17, 1086-1093.
Erdos, P., and Lehner, J. (1941). "The distribution of the number of summands in the
partitions of a positive integer," Duke Math. J. 8, 335-345.
Foata, D. (1965). "Etude algebrique de certains problemes d'analyse combinatoire et
du calcul des probability, " Publ. Inst. Statist. Univ. Earis 14, 81-241.
Foata, D., and Riordan, J. (1974). "Mappings of acyclic and parking functions,"
Aequationes Math. 10, 10-22.
Foata, D., and Schiitzenberger, M.-P. (1970). Theorie geometrique des polynomes
euleriens, (Lecture Notes in Math. No. 138). Springer, New York.
Foata, D., and Schiitzenberger, M.-P. (1973). "Nombres d'Euler et permutations
alternantes," A Survey of Combinatorial Theory (J. N. Srivastava et al., eds.), pp.
173-187. North-Holland, Amsterdam.

Cambridge Books Online © Cambridge University Press, 2010

References 67

Foata, D., and Strehl, V. (1974). "Rearrangements of the symmetric group and cnu-
merative properties of the tangent and secant numbers," Math. Z. 137, 257-264.
Foulkes, H. O. (1975). "Enumeration of permutations with prescribed up-down and
inversion sequence," Discrete Math. 9, 365-374.
Fray, R. D., and Roselle, D. (1971). "Weighted lattice paths," Pacific J. Math. 37,
85-96.
Gupta, H. (1942). "On an asymptotic formula in partitions," Proc. Indian Acad. Sci.
A16, 101-102.
Gupta, H. (1955). "Partitions in general," Res. Bull. Panjab Univ. 67, 31-38.
Gupta, H. (1970). "Partitions-a survey," /. Res. Nat. Bur. Standards 74B, 1-29.
Kreweras, G. (1965). "Sur une classe de problemes de denombrement lies au treillis des
partitions des entiers," Cahiers du B.U.R.O. No. 6.
Kreweras, G. (1966). "De*nombrements de chemins minimaux a sauts imposes," C. R.
Acad. Sci. Paris 263, 1-3.
Kreweras, G. (1967). "Traitement simultane du Trobleme de Young' etdu'Problemede
Simon Newcomb*," Cahiers du B.U.R.O. No. 10.
Kreweras, G. (1969a). "Inversion des polynomes de Bell bidimensionnels et application
au denombrement des relations binaires connexes," C. R. Acad. Sci. Paris 268,
577-579.
Kreweras, G. (1969b). "Denombrement syst£matique de relations binaires externes,"
Math. Sci. Humaines 7, 5-15.
Kreweras, G. (1970). "Sur les eventails de segments," Cahiers du B.U.R.O., No. 15.
LeVeque, W. J. (1974). Reviews in Number Theory, Vol. 4. Amer. Math. Soc, Providence,
R.I.
Long, C. (1970). "On a problem in partial difference equations," Canadian Math. Bull.
13, 333-335.
MacMahon, P. A. (1894). "Memoir on the theory of the compositions of numbers,"
Philos. Trans. Roy. Soc. London A184, 835-901.
MacMahon, P. A. (1908). "Second memoir on the composition of numbers," Philos.
Trans. Roy. Soc. London A207, 65-134.
MacMahon, P. A. (1915). Combinatory Analysis, Vol. 1. Cambridge Univ. Press, London
and New York (reprinted by Chelsea, New York, 1960).
Roselle, D. (1969). "Permutations by number of rises and successions," Proc. Amer.
Math. Soc. 19, 8-16.
Stanley, R. P. (1972). "Ordered structures and partitions," Mem. Amer. Math. Soc. 119.
Star, Z. (1976). "An asymptotic formula in the theory of compositions," Aequationes
Math. 13, 279-284.

Cambridge Books Online © Cambridge University Press, 2010

CHAPTER 5

The Hardy-Ramanujan-Rademacher
Expansion of p(n)

5.1 Introduction

Having seen many elementary properties of partitions and compositions

and having seen an elementary asymptotic formula for pM{n) (Theorem 4.3),
we now move to one of the crowning achievements in the theory of partitions:
the exact formula for p(n), an achievement undertaken and mostly completed
by G. H. Hardy and S. Ramanujan and fully completed and perfected by
H. Rademacher.
The story of the Hardy and Ramanujan collaboration on this formula is
an amazing one, and is perhaps best told by J. E. Littlewood in his fascinating
review of the Collected Papers of Srinivasa Ramanujan in the Mathematical
Gazette, Vol. 14 (1929, pp. 427-428):

I must say something finally of the paper on partitions . . . written jointly

with Hardy. The number p(n) of the partitions of n increases rapidly with n,
thus:
p(200) = 3972999029388.
The authors show that p(n) is the integer nearest

where Aq(n) = Y.0)p,qe~2npnilq^ t n e s u m b e m g o v e r P*s prime to q and less than

it, (op q is a certain 24gth root of unity, v is of the order of yjn, and

C =

We may take v = 4 when n — 100. For n = 200 we may take v = 5;

five terms of the series (1) predict the correct value of p(200). We may always

ENCYCLOPEDIA OF MATHEMATICS and Its Applications, Gian-Carlo Rota (ed.).

2, George E. Andrews, The Theory of Partitions
Copyright © 1976 by Addison-Wesley Publishing Company, Inc., Advanced Book Program.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical photocopying,
recording, or otherwise, without the prior permission of the publisher.

68

Cambridge Books Online © Cambridge University Press, 2010

5.1 Introduction 59

take v = asln (or rather its integral part), where a is any positive constant
we please, provided n exceeds a value no(a) depending only on a.
The reader does not need to be told that this is a very astonishing theorem,
and he will readily believe that the methods by which it was established
involve a new and important principle, which has been found very fruitful
in other fields. The story of the theorem is a romantic one. (To do it justice
I must infringe a little the rules about collaboration. I therefore add that
Prof. Hardy confirms and permits my statements of bare fact.) One of
Ramanujan's Indian conjectures was that the first term of (1) was a very
good approximation to p(n); this was established without great difficulty.
At this stage the n — (1/24) was represented by a plain n — the distinction
is irrelevant. From this point the real attack begins. The next step in develop-
ment, not a very great one, was to treat (1) as an "asymptotic" series, of
which a fixed number of terms (e.g. v = 4) were to be taken, the error being
of the order of the next term. But from now to the very end Ramanujan
always insisted that much more was true than had been established: "there
must be a formula with error 0(1)." This was his most important contribu-
tion; it was both absolutely essential and most extraordinary. A severe
numerical test was now made, which elicited the astonishing facts about p(100)
and p(200). Then v was made a function of n; this was a very great step, and
involved new and deep function-theory methods that Ramanujan obviously
could not have discovered by himself. The complete theorem thus emerged.
But the solution of the final difficulty was probably impossible without one
more contribution from Ramanujan, this time a perfectly characteristic one.
As if its analytical difficulties were not enough, the theorem was entrenched
also behind almost impregnable defences of a purely formal kind. The form
of the function \jsq(n) is a kind of indivisible unit; among many asymptot-
ically equivalent forms it is essential to select exactly the right one. Unless
this is done at the outset, and the — 1/24 (to say nothing of the d/dn) is an
extraordinary stroke of formal genius, the complete result can never come
into the picture at all. There is, indeed, a touch of real mystery. If only we
knew there was a formula with error 0(1), we might be forced, by slow stages,
to the correct form of \pr But why was Ramanujan so certain there was one?
Theoretical insight, to be the explanation, had to be of an order hardly to
be credited. Yet it is hard to see what numerical instances could have been
available to suggest so strong a result. And unless the form of \j/q was known
already, no numerical evidence could suggest anything of the kind —there
seems no escape, at least, from the conclusion that the discovery of the correct
form was a single stroke of insight. We owe the theorem to a singularly
happy collaboration of two men, of quite unlike gifts, in which each contrib-
uted the best, most characteristic, and most fortunate work that was in him.
Ramanujan's genius did have this one opportunity worthy of it.

The formula that we shall prove is in the final form obtained by Rademacher:

THEOREM 5.1

Pin) = - L g Ak(n)ki \ f sinh((*/Jc)(M*- 1/24))*)] ,

njl A [** (x - 1/24)* I,..

Cambridge Books Online © Cambridge University Press, 2010

70 The Hardy-Ramanujan-Rademacher Expansion of p(n) Chap. 5.1

where

Ak(n)= X Oih,ke-2"inl"k
hmodk
(/i,fc)=l

with cohk a certain 24kth root of unity defined in Section 5.2.

This unbelievable identity wherein the left-hand side is the humble arithmetic
function p(n) and the right-hand side is an infinite series involving n, square
roots, complex roots of unity, and derivatives of hyperbolic functions provides
not only a theoretical formula for p(n) but also a formula which admits
relatively rapid computation.
For example, p(200) = 3972999029388, while if we compute the first eight
terms of the series in (1) we find that the result is

+ 3,972,998,993,185.896
+ 36,282.978
- 87.555
+ 5.147
4- 1.424
+ 0.071
+ 0.000
+ 0.043

3,972,999,029,388.004

which is the correct value of p(n) within 0.004. It is a simple matter to determine
explicitly the error when the infinite series is truncated, so we can determine
with certainty values of p(n) directly from (5.1.1). For small values of n we
can, of course, obtain p(n) via the recurrence given in Corollary 1.8.
Furthermore it is a simple matter to show that each term of the infinite
series in (5.1.1) is O(exp[7r(2«) l/2 //c v / 3]). Consequently the term k = 1
provides an asymptotic formula for p(n)9 and if we note that A{(n) = 1, we
see without much trouble that as n -> oo

The truth of (5.1.1) is intimately tied to the fact that

exp(7n'T/12)
=
= £ , of(t,)exp(2>r/>tt)
is actually a modular form. In order to prove (5.1.1) we must make use of
very fundamental properties of ^(T), in particular its behavior under trans-

Cambridge Books Online © Cambridge University Press, 2010

5.2 The Formula for p(n) 71

formations of the modular group: r -> (ar -f b)/(cT -h d), ad — be = 1.

A totally complete proof of (5.1.1) would now require at least two further
chapters for this book; one in which the modular group and its basic properties
were fully developed (see Knopp, 1970, Chapter 1, or LeVeque, 1956,
Chapter 1), and a second in which the fundamental transformation formulas
for r\{x) are derived (see Knopp, 1970, Chapter 3). Such a project would take
us well afield from partitions, and so at the beginning of Section 5.2 we shall
state the necessary transformation formula.

5.2 The Formula for p(n)

Instead of using the transformation for ^(T) explicitly, we shall utilize an

equivalent result that will be most suited to our purposes. Namely, we let

then

P(exp(2ni(h +

= cohtkz±exp[n(z-1 - z)/12/c]P(exp[27r/(/i' + iz" 1 )/*]); (5.2.2)

here Re z > 0, the principal branch of z* is selected, h! is a solution of the

congruence

hW = - 1 (mod k\ (5.2.3)

and coA k is a 24/cth root of unity given by

(/- k\
- y - ) e x p ( - 7ri(i(2 - hk - h) + ^(/c - /c-1)(2/i - /i' + /i2/i')))
if /i odd,

^ r - J e x p ( - 7r/(J(k - 1) + i(fc ~ k~l)(2h - h[ + /I 2 /I')))

if /codd,
(5.2.4)
with (^/fc) the Legendre-Jacobi symbol. See Knopp (1970, Chapter 3) or
Rademacher (1973, Chapter 9) for a proof of (5.2.2); there is a proof of
(5.2.2) due to B. Berndt that is sketched in Examples 6-17 at the end of this
chapter.
In passing we mention the elegant representation of Rademacher for a>hk:

cohtk = exp(7r/5(/i, k)) (5.2.5)

Cambridge Books Online © Cambridge University Press, 2010

72 The Hardy-Ramanujan-Rademacher Expansion of p(n) Chap. 5.2

where s(/?, k) is the Dedekind sum:

]_^_[^]_4). (5.2.6)
Now we come to the truly remarkable approach to the evaluation of p(n)
that was developed by Hardy and Ramanujan. Clearly Cauchy's integral
theorem implies that

x (5.2.7)

where, say, C is a circle centered on the origin and inside the unit circle:
|x| = 1. How can we evaluate this integral? By examining the generating
function P(x) = [XT=i(l ~~ x'l)~1» w e s e e t n a t e a c n Partial product
n * = 1 (1 - xH)~l has a pole of order N at x = 1, a pole of order [AT/2] at
x = — 1, poles of order [N/3] at x = exp(27r//3) and exp(47r//3), and so on.
Furthermore, we note that (5.2.2) gives us extremely good information on the
behavior of P(x) near exp(27r/7i/fc), namely, as z -* 0 (Re z > 0):

exp ^2nih - 2nz\]

^ j J ~ o>Mz* exp
Clearly then we should divide the circle of integration into segments so
that our knowledge from (5.2.8) may be most powerfully applied, depending
on which "rational point" e\p(2nih/k) we are near (if a is rational, we shall
call exp(27r/a) a rational point of the unit circlfe). Unfortunately, the rational
points are dense on the unit circle and so the preceding prescription is not
exactly helpful. To gain some idea of how to make this approach effective,
we recall that the points exp(2nih/k) where k is small seem to be the most
important of the singularities; hence instead of worrying about the dense set
of all rational points, we restrict our attention to the discrete set of those
rational points exp(27r/7i//c) with 0 < k ^ N, a fixed positive integer. Thus
we are interested in FN: the set of proper Farey fractions of order N. For
example, when N = 5,

Seeing the Farey fractions arise in this natural manner, we surmise that we
should divide our circle C of radius p into segments so that each segment
is in some sense "centered" on the successive rational points pe2nih/k where
hjk e Fs. We must now rely on the elementary properties of Fs to suggest
the optimal choice of intervais. As is well known (and easily proved), if hjk
and hijkl are successive terms in Fs, then the rational number with least

Cambridge Books Online © Cambridge University Press, 2010

5.2 The Formula for p(n) 73

denominator lying strictly between hjk and h{jkx is (h + h{)j(k + k{) and
this number is called the mediant of h/k and hxjkx. Thus using the mediants
as the end points for our intervals seems a natural dissection of C. Now if
fro/^o» J'/fc* ^1/^1 a r e three successive terms in FN, let us write

eh>k = "* ^ " - - . (5.2.9)

Then clearly

exp(-

n V f n [ /27TI/I
n + 27T/0) expl
" I
(h,k)=l
1 P l l
O^hk
(5.2.10)

All that awaits now before the application of (5.2.2) is an appropriate choice
of p; we could leave the matter indefinite and choose p as well as possible
when compelled to do so. However, for simplicity we shall make the "right"
choice immediately:

and shall content ourselves with subsequent observations on the necessity

of this choice. Hence

2nihn
k= 1

exp | ^ - ^ f A _ ik4>)\\exp(-2Kin<t»d<l>; (5.2.12)

Cambridge Books Online © Cambridge University Press, 2010

74 The Hardy-Ramanujan-Rademacher Expansion of p(n) Chap. 5.2

thus to apply (5.2.2) we must define

z = k(N~2 - /</>). (5.2.13)

Consequently, applying (5.2.2) to the integrand in (5.2.12), we find that

2nn\
p(n) = exp 2
) exp ^ - —^— ]
A'

x z' exp 7i — P exp exp(— 2nin(f))d(l).

[2k

(5.2.14)

Now as z -> 0 with Re z > 0, we see that exp[2ni(hf 4- /z"1)/^:] -• 0 rapidly.

Therefore the obvious way to evaluate (5.2.14) is to replace in the integrand
P(x) by 1 4- (P(x) - 1). Hence

inn Inihn \
= 1
ih,k)= I

2nn\ 2nihn z
(/•.Ac) = 1
]

x (pfexp[2^(/-+T' _ \)QXp( d<}> = I , + I2.

(5.2.15)

It is our natural expectation that I{ will contribute our principal estimate for
p(n) and that the contribution of I2 WM t>e negligible. We undertake our long
analysis of (5.2.15) by proving that I2 is indeed negligible: First (recalling
that z = kN~2 - /7c</>),

P exp

< |z| ! exp • (5.2.16)

Cambridge Books Online © Cambridge University Press, 2010

5.2 The Formula for p(n) 75

Now
2
N~
kN~2 - iko

From (5.2.9) it is immediate that each of Qhk and 6hk satisfies 1/2/cN
0h%k < 1/fcN, and since — 0hk ^ cp ^ O'h'tk, we see that
2 2
N~ N"
/c 2 (N 4- " 2 "" '"

Also

|z|* = (/c2N~4 + /c 2 ^ 2 ^ < (/c2N~4 + N" 2 )* ^ 2^N"^. (5.2.18)

Hence by (5.2.16), (5.2.17), and (5.2.18) we have the following estimate for I2:

(h,k)

l \1 f
x X p(m) exp - n [m - - ~\\ d<f>
Z4
m=l L \ /J J

-n\m-2- k = 1

-n\m-2-

(5.2.19)

This estimate suffices for our purposes, since N~* exp(27TMiV"2) -• 0 as

N -+ oo for fixed n.
We now handle Tj, the main term in Eq. (5.2.15). Here we shall show that
the integral is the most significant portion of a Hankel-type loop integral.
Once we establish this fact (Eq. (5.2.26)), it will then be a simple matter to
complete our evaluation of p(n).
In the integral for Ix (see (5.2.15)) we set co = iV"2 - i<j>, and thus the
integral becomes

Cambridge Books Online © Cambridge University Press, 2010

76 The Hardy-Ramanujan-Rademacher Expansion of p(n) Chap. 5.2

2
Ihk = 1 (/ceo)* e x p —— I /ceo I 4 - 2nn(a> —N ) \ i dco
J I2k\k(o I J

= exp(-2;r«JV-2)/c*r1 j (oiexpl-j^-jexpp^-^jwjdfii

i g{co)da), (5.2.20)

where

0(co) = oo* exp ]2n (n - — jco + - ^

The integrand of this last integral is single valued and analytic in the complex
co-plane cut along the entire negative real axis. Hence we may write (invoking
Cauchy's theorem):

N
(0 + ) -e -e-'V* '-"Vk

-~ • * ( ! - ! - J - i \.
2
exp(2nnN- )Ihtk

k
htk

- i - i -J) g(co) dco (5.2.21)

where the integral J ^ ^ is the loop integral along the contour & in Fig. 5.1.
We assume 0 < s < N~2 and we shall mostly be interested in what happens
as e -• 0. For brevity, we rewrite (5.2.21) as

exp(27rniV- 2 )/ M = k*rl{Lk - I, - I2 - I3 - J 4 - / 5 - I6] (5.2.22)

and our next task is to show that each of the four integrals / 2 , 73, 74, and 75
is negligible:

„>)*exp [ ^ R e ( - 1 - ^ ) ] exp [- 2n(n -

Cambridge Books Online © Cambridge University Press, 2010

5.2 The Formula for p(n) 77

lou-plane

Figure 5.1

But R e [ ( - e + to)"1] = - e/(e2 + v2) < 0, and so

-3/2
as £ -> 0.

(5.2.23)

Integral / 5 is treated in exactly the same way and (5.2.23) is also valid with
I2 replaced by / 5 . As for / 3 ,

Now

Hence

/ 3 | < (AT r^

Cambridge Books Online © Cambridge University Press, 2010

78 The Hardy-Ramanujan-Rademacher Expansion of p(n) Chap. 5.2

N'2)

(e

as e -• 0 (5.2.24)

and the final inequality in (5.2.24) is also easily seen to hold with 73 replaced
by/5.
The integrals / t and I6 are not negligible; however,

f / ni\ n / 1 ^
l
r
6
- oo
in
\J\u\ ex|
- oo
\ 2
>
cxp 24 j

(ni\
+ exp exp f- 2nln - u\du
\2) 12k2u ~2Aj
—t

2 expf - 2
00

'i'* 1\
24/
n
\2k2t
dt

= - 2iHk. (5.2.25)

Hence, letting e -• 0 in (5.2.22), we may simplify that equation to

cxp(2nnN-2)Ihk = —Lk T-5/2

(5.2.26)

where the constants implied in the 0 terms are absolute. Consequently, writing

ij/^n) = k-Lk + 2k*Hk, (5.2.27)

we see from (5.2.15), (5.2.19), and (5.2.26) that

2nihn\
p(n)= T_>

(M)=l

7-3/2 2nn
+o exp N -5/2
k=l
(M)=l

Cambridge Books Online © Cambridge University Press, 2010

5.2 The Formula for p(n) 79

MUn) + 0 [AT* e x p ^ - j ] + O(AT*). (5.2.28)

Now .4fc(rt) is precisely as it appears in (5.1.1) and the error terms here are
all -> 0 as N -* oo. Hence the final problem for us is to show that

(x-1/24)* ]
We may do this fairly easily, assuming certain classical results about Hankel's
integral. First
(0 + )
1 1f n / 1\ 1,
i i J \ 24/ J

(0 + )
[ U / l\ I f (7r/12^2co) .

(0 + )
r
(0 + )

24/ s=0 sir(s- 1/2)

where the final equation follows from Hankel's loop integral formula for the
reciprocal of the gamma function:
(0 + )

r(s - i) 2ni
— oo

Now

r(s - i ) = (s - 3/2)(S - 5/2)-

= 2-5+1TT*(2S - 3)(25 - 5)---3-l.

Hence

Cambridge Books Online © Cambridge University Press, 2010

80 The Hardy-Ramanujan-Rademacher Expansion of p(n) Chap. 5.2

2! 4! 6!
y2n+l
2
—E

and so with 7 = (7c/fc)(2/3(x - 1/24))*, we see that

= , - t , f - i 1 rf cosh((7t/fc)[f(x - 1/24)]*)}
\dx (w/fc)[Kx - 1/24)]* !,=„
= J = M cosh((7t/fc)[f(x - 1/24)]*)) 2
^72 \dx (x - 1/24)* },.. '

Finally we treat the Hk in (5.2.27) in a similar manner. We begin with a

classical evaluation of a definite integral:

exp(- c2t - ah'^'^dt = 2 exp(- c V - a2u~2)du = ^

0

Hence

exp(- c2t - a2rl)t*dt =- ^jrl \. I' (5-2-31)

o

and applying (5.2.31) to ifk we see that

1 Id e x p [ - 7r(2/3)*c/fc]
n{n - 1/24) U

Cambridge Books Online © Cambridge University Press, 2010

Examples 81

d ex
P^ < 2 / 3 )(* V 2 4 )/^) (5 2 32)
\dx (x - 1/24)*
Thus we obtain, from (5.2.27), (5.2.30), and (5.2.32), that

= /c*_ | d cosh((7r/fe)[f(x - 1/24)]*)]

ny/2\dx (x-1/24)* J,..
fe
_ * / ^ exp{- Qr/fc)B(x - l/24)]*}\
nj2 U (x - 1/24)* Jx=n
sinh((7c/fe)[Kx-l/24)]*)] 2 3

(x-1/24)*
and since (5.2.33) is just (5.2.29), we see that Theorem 5.1 is established. •

Examples

1. The number of partitions of n into at most two parts (= />("{!> 2}", n)

by Theorem 1.4) may be shown to be [n/2] + 1 by examination of the
decomposition of the generating function
1 _ 1/2 + 1/2
2 2
(1 - q){\ - q ) (1 - q) (1 - q2)

2. The number of partitions of n into at most three parts (= jp("{l> 2, 3}", n)

by Theorem 1.4) may be shown to be the nearest integer to (l/12)(n 4- 3)2
by examination of the following decomposition of the generating function:
1 1/6 1/4 1/4 1/3
(1 - q\\ - q2){\ - q*) (1 - qf (1 - q)2 (1 - q2) (1 - q3)
We remark that Examples 1 and 2 are special cases of Cayley's general
decomposition:

- q2)- - -(1 - ql) ^ 1^2^3 p 3. . • ^ 1 ! p 2 ! « • -Pi

where the summation runs over all partitions (lpi2P22P3 • • -iPi) of i. We shall
give this formula in Example 1 of Chapter 12 as a special case of results on
Bell polynomials.
3*. There are many modular functions besides rj(r) that have transformation
formulas like (5.2.2). Indeed it has been shown by Hagis and Subramanyasastri
that if H is the set of integers = ± au ± a2,. •., ± a r (mod /c), then there
exists an asymptotic series expansion for p("H", n) of the same form as that

Cambridge Books Online © Cambridge University Press, 2010

82 The Hardy-Ramanujan-Rademacher Expansion of p(n) Chap. 5

given in Theorem 5.1. In fact, the same assertion is true of /?("//"(^ 0, n)

as well. As an example of these results, we present the series for p{Q), n) =
p{(9, n) (Corollary 1.2), which was found through different techniques by
Hua, Iseki, and Hagis:

f—2ninh\ 1
/(M)exp
k= 1 hmodk k(24n + 1)* [12/c
fcodd (h,k)=l

here
00
(z/2) 2w+1
/.(*) = Z /i!(n + 1)!
and %(h, k) is the root of unity arising in the modular transformation formula
for Zn^o P(^» ")<7" precisely the way cohk arises from (5.2.2).
4*. Apart from the modular functions, there exists a class of functions
called mock-theta functions (originally studied by S. Ramanujan and G. N.
Watson) for which reasonable asymptotic series for the coefficients may be
derived. For example, let
oo q*2

= Z anq\
n=0

Then it can be shown that

an = i X ( - 0 C(k+1)/2 M 2fc l> - Kl + ( - 1)*)/C][/C(K - 1/24)]"*

k=l

where Ak(n) is the same as the exponential sum appearing after (5.1.1).
Dragonette has pointed out the possibility that the error involved in the
expression for an may generally be less than 4 in absolute value. In particular,
the series at n = 100 is - 18520.206 while a 100 = - 18520, at n = 200 the
series yields - 2660007.847 while a200 = - 2660008.
5. The proof of the result in Example 4 relies heavily on the fact that
00
(— 1\n ±n(3n+l)

= 1 + 4 Z ^ l±—H
This formula is due to Watson and is derived by techniques similar to those
used in Section 7.2.
Examples 6-17 sketch a proof of the modular transformation formula
(5.2.2) for rj{z) = eniz/l2IP(e2itiz). The following functions all play a role in

Cambridge Books Online © Cambridge University Press, 2010

Examples 83

this development:

G(z, s) = (mz + n) s
for — 7r ^ arg s < n, Im z > 0, Re s > 2;
m,n= — oo
(m,n)*(0,0)

Vz = (az + b)/(cz + d)
where a> b, c, d are integers with c > 0, and ad — be = 1;

#(z, 5) = X ( m z + ")" s
dm — en > 0

(the conditions for G(z, 5) also apply to g(z, s));

h(z,s) = £ ( m z + nY
m>0
n>dm/c

(with again the same conditions);

00

s l u
• i -u - e~ du,
r{s)=
0
Res > 0

(this is Euler's integral representation of the gamma function);

= £ n-\ Res > 1

n= I

(the Riemann zeta function);

A(z,s)= X ns-le2*imnz, Im z > 0;

.*) = du

where {x} = x — [x], with [x] the largest integer not exceeding x, and
where ^ is the following loop in the u plane oriented counterclockwise:

6. It is easy to show that A(z, 0) = (TT/Z/12) — log rj(z); we need only use
the fact that log(l - w)"1 = X ^ i wn/«.

Cambridge Books Online © Cambridge University Press, 2010

84 The Hardy-Ramanujan-Rademacher Expansion of p(n) Chap. 5

7. (cz + d)'sG(Vzys) = e~2nis £ (mz + n)'s + X

m^ 0 m>0
and dm-en > 0 or dm —en i

8. $(z, s) = e*isC(s) + ^"fc(z, 5).

9. For Re z > - d/c, Im z > 0,

s l
F(s)h(z, s) = ]T \ u e x p ( — mzu — nu) du
m>0 J
n>dm/c 0
00
oo r
= X u5"1 exp(- (tri + l)zu - (ri 4- 1 + [(m'd + d)lc])u)du
m',n' = 0 J
0
where m' = m — 1 and n' = n — \_mdjc] — 1, with [x] the greatest integer
function.
10. If in the second sum in Example 9 we replace n' by n and set m! =
pc + j — 1, 0 ^ p < co, 1 ^ ; ^ c, then it follows that
00
1
Ju"1 expJ-jzH - (l + [^])u
z,s) == j£ Ju"
r(s)h(z,s)
0

)w — nu) du

y [ «""' exp(- (cz + d)ju/c + {jd/c}u)

y=i J {1 - e x p ( - (cz + </)u)}(exp(u) - 1) "
o
/i o • u - i V f " s " 1 exp(- (cz + d)julc + {jdlc)u)
= (1 - exp(2,,s)) II )) ( 1 _ e x p ( _ ( c g + <f)u)Xexp(H)_1) du.

11. Applying Example 10 to Example 8 and hence to Example 7, we find

that for Re s > 2, Re z > - d/c, Im z > 0,
(cz + d)"T(s)C(^z, 5) = r(5)C(z, 5) - 2/ sin nsr(sX(s) + e~nisL(z, s).
12*. (The Lipschitz Summation Formula.) For Im z > 0 and Re s > 1,

13. It is possible to dissect the sum for G(z, s) as follows.

oo oo oo

G(z, s) = £ H~i + £ X ( m z ~*~ n )~ 5 + X X ( m z "*" n ) ~ s

n= — oo m<0 n=-oo m>0 n=-oo

2 + ^3

Cambridge Books Online © Cambridge University Press, 2010

Notes 85

where by Example 12
S2 = enis(- 2ni)5A(z, s)ir(s), S3 = ( - 2ni)'A(z, s)/r(s).
14. The result in Example 13 may be rewritten as
G(zy 5) = (1 + eKis)£(s) + (1 + e"' s )(- 2ni)sA(z, s)/r(s)9
which gives the analytic continuation of G(z, s) to the entire complex s plane.
15. From Example 14 and the definition of //(z, s), it is possible to rewrite
Example 11 in terms of H(z, s):
(cz + d)-sH(Vz, s) = H(z, s) - enis(2nirs(cz
+ (27ri)~T(s)(l + enis)C(s) + (2niysL(z,s).
16*. The calculus of residues may be utilized to show that

L(z, 0) = 2;
- d) I2c
where s(d9 c) is given in Eq. (5.2.6).
17. Setting s — 0 in Example 15 and applying Example 16, we may deduce
that

log r]{ Vz) = log r\{z) - ^ + £ log(cz + d) - nis(d, c) + * >

4 12c
a result equivalent to (5.2.2).

Notes
The origin of this chapter is the epoch-making paper by Hardy and
Ramanujan (1918). Our presentation follows closely Rademachefs (1937)
original exposition of this theorem. Subsequently, Rademacher (1943, 1973)
refined the circle method using what are called Ford circles. Although this
procedure is elegant and has been useful in later developments of modular
function theory, we have chosen Rademacher's original approach because
it is easily adapted to problems concerning nonmodular generating functions
(see Chapter 6). The books by Ayoub (1963), Hardy (1940), Knopp (1970),
Lehner (1964), and Rademacher (1973) present excellent accounts of the role
of modular functions in partition theory. Certain important contributions
to partition asymptotics by E. Grosswald (1958, 1960) are related to both
this chapter and Chapter 6. Reviews of the recent work extending the
Hardy-Ramanujan-Rademacher method are found in Sections P68 and P72
of LeVeque (1974).
Examples 1, 2. Cayley (1898), MacMahon (1916), Arkin (1970).
Example 3. Hagis (1962, 1963, 1964a, b, 1965a, b, c, 1966) treats many
cases when k is prime; Subramanyasastri (1972) treats arbitrary k. See also
Iseki (1959, 1960, 1961).
Example 4. Watson (1936), Dragonette (1952), Andrews (1966).
Example 5. Watson (1929, 1936).
The proof of (5.2.2) presented in Examples 6-17 is due to B. Berndt (private
communication) and is presented in much greater generality in Berndt (1973,

Cambridge Books Online © Cambridge University Press, 2010

86 The Hardy-Ramanujan-Rademacher Expansion of p(n) Chap. 5

1975). Berndt has shown this method to be applicable to numerous modular

functions and to be simpler conceptually and computationally than alter-
native methods; for these reasons this approach should have wide ramifica-
tions in partition theory.

References
Andrews, G. E. (1966). "On the theorems of Watson and Dragonette for Ramanujan's
mock theta functions," Amer. J. Math. 88, 454-490.
Arkin, J. (1970). "Researches on partitions," Duke Math. J. 38, 403-409.
Ayoub, R. (1963). An Introduction to the Analytic Theory of Numbers, American
Mathematical Soc, Providence.
Berndt, B. (1973). "Generalized Dedekind eta-functions and generalized Dedekind sums,"
Trans. Amer. Math. Soc. 178, 495-508.
Berndt, B. (1975). "Generalized Eisenstein series and modified Dedekind sums," J.
Reine Angew. Math. Ill, 182-193.
Cayley, A. (1898). "Researches in the partition of numbers," Collected Math. Papers 2,
235-249, 506-512.
Dragonette, L. A. (1952). "Some asymptotic formulae for the mock theta series of
Ramanujan," Trans. Amer. Math. Soc. 72, 474-500.
Grosswald, E. (1958). "Some theorems concerning partitions," Trans. Amer. Math.
Soc. 89. 113-128.
Grosswald, E. (1960). "Partitions into prime powers," Michigan Math. J. 7, 97-122.
Hagis, P. (1962). "A problem on partitions with a prime modulus p ^ 3," Trans. Amer.
Math. Soc. 102, 30-62.
Hagis, P. (1963). "Partitions into odd summands," Amer. J. Math. 85, 213-222.
Hagis, P. (1964a). "On a class of partitions with distinct summands," Trans. Amer.
Math. Soc. 112, 401-415.
Hagis, P. (1964b). "Partitions into odd and unequal parts," Amer. J. Math. 86, 317-324.
Hagis, P. (1965a). "A correction of some theorems on partitions," Trans. Amer. Math.
Soc. 118, 550.
Hagis, P. (1965b). "Partitions with odd summands —some comments and corrections,"
Amer. J. Math. 87, 218-220.
Hagis, P. (1965c). "On the partitions of an integer into distinct odd summands," Amer.
J. Math. 87, 867-873.
Hagis, P. (1966). "Some theorems concerning partitions into odd summands," Amer.
J. Math. 88, 664-681.
Hardy, G. H. (1940). Ramanujan. Cambridge Univ. Press, London and New York
(reprinted by Chelsea, New York).
Hardy, G. H.,and Ramanujan, S. (1918). "Asymptotic formulae in combinatory analysis,"
Proc. London Math. Soc. (2) 17, 75-115. (Also, Collected Papers of S. Ramanujan,
pp. 276-309. Cambridge Univ. Press, London and New York, 1927; reprinted by
Chelsea, New York, 1962.)
Hua, L. K. (1942). "On the number of partitions of a number into unequal parts,"
Trans. Amer. Math. Soc. 51, 194-201.
Iseki, S. (1959). "A partition function with some congruence condition," Amer. J.
Math. 81, 939-961.
Iseki, S. (1960). "On some partition functions," J. Math. Soc. Japan 12, 81-88.
Iseki, S. (1961). "Partitions in certain arithmetic progressions," Amer. J. Math. 83,
243-264.

Cambridge Books Online © Cambridge University Press, 2010

References 87

Knopp, M. I. (1970). Modular Functions in Analytic Number Theory. Markham, Chicago.

Lehmer, D. H. (1939). "On the remainders and convergence of the series for the
partition function," Trans. Amer. Math. Soc. 46, 362-373.
Lehner, J. (1964). Discontinuous Groups and Automorphic Functions. Mathematical
Surveys, No. VIII, Amer. Math. Soc, Providence.
LeVeque, W. J. (1956). Topics in Number Theory, Vol. 2. Addison-Wesley, Reading,
Mass.
LeVeque, W. J. (1974). Reviews in Number Theory, Vol. 4. Amer. Math. Soc, Providence,
R.I.
MacMahon, P. A. (1916). Combinatory Analysis, Vol. 2. Cambridge Univ. Press, London
and New York (reprinted by Chelsea, New York, 1960).
Rademacher, H. (1937). "On the partition function p(n)," Proc. London Math. Soc. (2)
43, 241-254.
Rademacher, H. (1943). "On the expansion of the partition function in a series," Ann.
of Math. 44, 416-422.
Rademacher, H. (1958). "On the Selberg formula for Ak(n)" J. Indian Math. Soc.
(N.S.) 21, 41-55.
Rademacher, H. (1973). Topics in Analytic Number Theory. Springer, Berlin.
Subramanyasastri, V. V. (1972). "Partitions with congruence conditions," /. Indian Math.
Soc. 36, 177-194.
Watson, G. N. (1936). "The final problem: an account of the mock-theta functions,"
/. London Math. Soc. 11, 55-80.
Whiteman, A. L. (1947). "A sum connected with the partition function," Bull. Amer.
Math. Soc. 53, 598-603.
Whiteman, A. L. (1956). "A sum connected with the series for the partition function,"
Pacific J. Math. 6, 159-176.

Cambridge Books Online © Cambridge University Press, 2010

CHAPTER 6

The Asymptotics of Infinife Product

Generating Functions

6.1 Introduction

As we saw in Chapter 1, many partition functions have infinite products

as the associated generating function. Numerous researchers have considered
the asymptotics of such partition functions. We shall present the general
theorem of G. Meinardus, which gives an asymptotic formula that includes
numerous partition functions. His method is also amenable to generalization
and we shall discuss some possibilities for generalization at the end of this
chapter.
The infinite product we shall consider is

= l + f r(n)q", (6.1.1)
n= 1

where q = e~x and Re r > 0 (or equivalently \q\ < 1). We restrict the an
to be nonnegative real numbers. We also consider an auxiliary Dirichlet
series:
TO

a
D(s) = I \ (s = a + it), (6.1.2)
n= 1 '1

which is assumed to converge for a > a, a positive real number. Furthermore,

we assume that D(s) possesses an analytic continuation in the region a ^
— C o (0 < C o < 1) and that in this region D(s) is analytic except for a pole
of order 1 at s = i with residue A. Finally, we assume that

D(s) = O(|r|Cl) (6.1.3)

uniformly in a ^ - Co as \t\ -• co, where C\ is a fixed positive real number.

ENCYCLOPEDIA OF MATHEMATICS and Its Applications, Gian-Carlo Rota (ed.).

2, George E. Andrews, The Theory of Partitions
Copyright © 1976 by Addison-Wesley Publishing Company, Inc., Advanced Book Program.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical photocopying,
recording, or otherwise, without the prior permission of the publisher.

Cambridge Books Online © Cambridge University Press, 2010

6.2 Proof of Theorem 6.2 89

We must finally make use of the function

a
0(T) = t n4n> 4 = *~T. (6.1-4)
n=l

Now if T = y + 27rix (x, y real), we shall also assume that for |arg T| > rc/4,

Refoto) - tfjO < - C2y-< (6.1.5)

for sufficiently small y, where e is an arbitrary fixed positive number and C 2
is a suitably chosen positive real number depending on e.
With the definitions above in mind, we state Meinardus's main theorem:

THEOREM 6.2. As n - • oo,

r(n) = CnK exp |n a / ( a + 1 Ml + ^ j [A r(a -f l)C(a + l ) ] 1 / ( a + 1 ) (1 + O(M~ K 0)

(6.1.6)

where £(s) = Xm=i m ~ 5 /s x

^e Riemann zeta function, and

C = e D ' ( 0 ) [>(l + a)]-*[yir(a + l)C(a + i ^ d - w o w ^ + i.^ (6 x 7)

1
- ^ . (6.1.8)
a

S an arbitrary real number.

The proof of this theorem relies on application of the saddle point method.
In the examples at the end of the chapter, we shall present numerous applica-
tions of Theorem 6.2.

6.2 Proof of Theorem 6.2

In applying the saddle point method, we must have information on the

behavior of / ( T ) in the half-plane Re(i) > 0 especially, near T = 0; this is
supplied by the following:

LEMMA 6.1. Under the assumptions on / ( T ) , D(S), and g(r) presented in

Section 6.1 (with T = y + 27r/x),

f{x) = exp[^r(a)C(a + l)T' a - D(0) log x + D'(0) + O ( / 0 ) ] (6.2.1)

Cambridge Books Online © Cambridge University Press, 2010

90 The Asymptotics of Infinite Product Generating Functions Chap. 6.2

uniformly in x as y -• 0, provided |arg T| ^ n/4, \x\ ^ \\ there exists a

positive e{ such that

f(y + 2nix) = O(exp[/ir(a)C(a + l)y" - C 3 )T"]) (6.2.2)

uniformly in x with yp ^ |x| ^ \, as y -+ 0, where

-|(l--H with 0<<5<f, (6.2.3)

and C3 a fixed real number.

Proof. We note that

log/(t) = - £ c
v=l

oo 4 oo

= Z T Z «v^"vfct- (6.2.4)
it=i /c v = 1

Now recall that e~~z is the Mellin transform of T(s); that is,

r o

= ^-. I
27CI J
x'T(s) ds (for Re T > 0, ff0 > 0). (6.2.5)
<T0"~ *00

Applying (6.2.5) to the exponential function in (6.2.4), we see that

1 +a+ ico

^ f ds; (6.2.6)

the interchange of summation and integration necessary to obtain (6.2.6) is

permissible due to the absolute convergence of the resulting integrated series.
Our object now is to shift the line of integration from Re T = 1 + a to
Re T = — C o . We first note that the integrand in (6.2.6) has a first-order pole
at 5 = a and a second-order pole at s = 0; the residue at s = a is clearly
T-°T(aX(a + 1)4. Around s = 0,

T"T(S)C(S + 1)Z>(5)

= (1 - s l o g t + • • - ) ( y - 7 + " • • \(~ + y + • • • W o ) 4- D'(0)s + • • •)

= -i + (D'(0)-D(0)logT)- + . . . ;
s s

Cambridge Books Online © Cambridge University Press, 2010

6.2 Proof of Theorem 6.2 91

hence, the residue of the integrand in (6.2.6) at s = 0 is D'(Q) — D(0) log r.

Hence, we see that

log/(r) = ^r(a)C(a + l)t"a - D(0)logr + D'(0)

-Co+ioo

+ 2^. I ^T(s)C(s+l)D(s)ds; (6.2.7)

the shift of the line of integration made in passing from (6.2.6) to (6.2.7)
is permissible since for |arg T| ^ n/4, we see that

|T-'| = |Tr*exp(*argT) ^ \T\~° exp(7r|*|/4)

and for a ^ C o ,

D(s) = O(|r|c0

(by assumption), while classical results on the ( and F functions assert that
c
C(s 4-1) =

and

= o(exp(-f
as t -> oo.
Finally we observe that
-Co+»oo

± \
2ni J
-Co~*oo

. 0(|T|C<>) = 0 ( / o ) , (6.2.8)

(since |arg T| ^ 7r/4 implies |2TTX| ^ y and thus |T| ^ V2 >^). Equation (6.2.1)
now follows from (6.2.7) once we estimate the integral in (6.2.7) by means
of (6.2.8).
We now turn to Eq. (6.2.2); here we consider two cases: (1) yfi < |x| <
y\2n and (2) y\2n ^ |x| < \. (One or both of these cases may be vacuous,
depending on y, but since we are interested only»in y approaching 0, we may
always assume y sufficiently small to make both nonvacuous.)

Cambridge Books Online © Cambridge University Press, 2010

92 The Asymptotics of Infinite Product Generating Functions Chap. 6.2

In case 1, we see that

27r|x| 4 n
tan|argi| = ^1 or |argt|^—,

as assumed above. Hence we may estimate (6.2.7) in the way done before, and
we determine that

|log/0> + 2TTIX)| ^ ^r(a)C(a + 1)|T|-« + C6|log >>|. (6.2.9)

The extra terms we might expect on the right-hand side of (6.2.9) have all
been accounted for by C6|log y\, since the other terms are of lower order of
magnitude (remember y is approaching 0, so that |log y\ dominates all non-
negative powers of y). Recalling that |T| 2 = y2 + 4n2x2, we see from (6.2.9)
that

log|/0> + 2xrix)|

< i4f(a)C(a 4- l)y~* + i4r(a)C(a + l)}>~a[(l + 4n2x2y~2y*/2 - 1]

+ C6|log>;|

C(a + Dy~* - Ciy-a\x\2y-2 ^ Ar(a)C(a -f l)>^"a - C 7 ^ - a + 2 ( ^ 1 ) .

(6.2.10)

The next to last inequality follows from the observation that as y - ^ 0 + ,

(1 + AY)~*12 — 1 ~ — OLAY/2 and the \logy\ is dominated by any negative
power of y as y approaches 0.
Now by (6.2.3), we see that — a 4- 2(/? — 1) = - a<5/2 ^ — e^say. Hence,
in case 1,

logl/Cy + 2nix)\ ^ Ar(xK(z + i)y'x - C3y-£l, (6.2.11)

which is equivalent to (6.2.2).

In case 2 (i.e., y/2n ^ |x| ^ |), we see from (6.1.4) and (6.2.4) that

2TTIX)| - Re(^(i)) = £ | £ ave-vkycos{2nkvx)

k=2 Kv = l

ZT t
fc=2 K v =l
(6.2.12)

since the av are all nonnegative. Now we have the hypotheses necessary to
apply (6.1.5). Hence in case 2

Cambridge Books Online © Cambridge University Press, 2010

6.2 Proof of Theorem 6.2 93

log|/0> + 2TTIX)| ^ log/(>') + Re(flf(T)) - g(y)

l)>>~a - C3>«-£1. (6.2.13)

Hence our lemma is completely proved. •

We are now prepared to obtain the asymptotic formula for r(n).

Proof of Theorem 6.2. We recall by the Cauchy integral theorem that

too + 2 n i

= \f(y + 2nix)eny+2ninx dx. (6.2.14)

-i

We wish to apply the saddle point method to the evaluation of this integral.
Since the maximum absolute value of the integrand occurs for x = 0, and
since for x = 0, Lemma 6.1 implies that the integrand is well approximated by

exp[\4r(a)C(a + l)y'* + nyl

the saddle point method suggests that we should minimize this expression;
that is, we should require that y be chosen so that

^ l)y~* + ny]} = 0.

Hence we take

y = n~u(*+l\Ar(oL + l)C(a + l)) 1 / ( a + 1 \ (6.2.15)

For notational simplicity we define

m = ny = « a/(flt+1) [/ir(a -f l)C(a + l ) ] 1 / ( 2 + l ) . (6.2.16)

From (6.2.14), we see that

yfi

m
r(n) = e j f(y -f 2nix)e2ninx dx + emRly (6.2.17)
-yfi

where

Cambridge Books Online © Cambridge University Press, 2010

94 The Asymptotics of Infinite Product Generating Functions Chap. 6.2

-yfi 4.

Rx = II +
+ \\f(y + 2nix)e2ninx dx (6.2.18)
-i

and where p is defined by (6.2.3). The interval of integration is being split into
two parts in a manner related to the Hardy-Ramanujan Farey dissection
applied to the partition generating function in Chapter 5. To actually under-
stand how (6.2.3) arises we may note that we have already used explicitly in
the proof of Lemma 6.1 that — a + 2(jS — 1) should be a small negative
number.
By Lemma 6.1,

^ ]) (6.2.19)

as n -* 00 (i.e., y = m/n -• 0). Hence

exp(m)R1 = 0 Uxp M + I j m - C9meAj (6.2.20)

as n -+ 00.
Equation (6.2.20) provides an adequate estimate for the second term in
(6.2.17). We must now treat the main integral. Let us choose n ^ n2 ^ ni9
where n2 is sufficiently large that 2n(m/ny~l ^ 1 (this is quite permissible,
since /? > 1 and mjn -+ 0 as n -+ c»). Then by Lemma 6.1

K » ) - exp 4. I j m + D ' ( 0 ) ] J exp^^x)) dx + exp(m)/?1, (6.2.21)

where

_ m\
— - ) ^ ll +2ICIHJC-D(0) log(-
ViW — a
(6.2.22)

note that the choice of ni guarantees that throughout the interval of integra-
tion |x| ^ J, and the choice of n2 guarantees that arg T ^ n/4 as well.
We now make the change of variable 27ix = (m/n)co, and we obtain

l ) l o gw
- + D'(0) - \og2n •/ + exp(m) Ri9
J
(6.2.23)

Cambridge Books Online © Cambridge University Press, 2010

6.2 Proof of Theorem 6.2 95

where

/ = I exp(</>2(co)) do, (6.2.24)

and

< ^ ) = m I „ / . „ - - + to| - D(0)log(l + /co) + O(m"c o /a ) (6.2.25)

[a(l 4- /to) a J

as m -+ oo.
Our problem is now reduced to obtaining an asymptotic expression for /.
First

y l /
/= I exp|- \ co 2 |<fo + /? 2 , (6.2.26)

where

exp - m —r—-o) 2 \ [exp(03(co)) - 1] dco, (6.2.27)

with

(6.2.28)

as m -> oo. For n ^ n3 ^ n2, choose n3 sufficiently large that |co| < I
throughout the interval of integration. Hence

f 1 1 a+1
[a(l + ico)* a 2

1 v / a + j ' - 1\. i W > / ;. 1 . a + 1

X ( - l)JiJo)J + to + -——to2

^...^ = 0(m|a)|3) =

(6.2.29)

while

Cambridge Books Online © Cambridge University Press, 2010

96 The Asymptotics of Infinite Product Generating Functions Chap. 6.2

/a
log(l + ico) = - X ). (6.2.30)

Therefore, a s m - * oo

exp(</>3(co)) - 1 =
!-«/« + m -Co/«) = ) , (6.2.31)

where
:_/C0 1 35'
(6.2.32)

by (6.2.3).
Since the length of the interval of integration in (6.2.27) is O(m (1 " /?)/a ), we
see that as m -+ oo

R2 = (6.2.33)
where

(6.2.34)

As for the integral in (6.2.26), we see that

exp -co' («+!)]" exp(-zVz

(6.2.35)

Hence by (6.2.26), (6.2.33), and (6.2.35), we obtain, as m -• oo,

2TT
I = (1 + O(m-"•>)) (6.2.36)
m(a + 1)
where

min ( ^ - (6.2.37)

Finally by (6.2.17), (6.2.20), (6.2.33), and (6.2.36), we see that

I
r(«) = exp 1 + - m - (D(0) - 1 ) l o gm
- + £>'(0) (2;rm(a + 1))"

x (1 + O(m-»>)) (6.2.38)

Cambridge Books Online © Cambridge University Press, 2010

Examples 97

a s m - > 00. Using (6.2.16) to replace m by a function of n, we obtain exactly

(6.1.16). Thus Theorem 6.2 is proved. •

6.3 Applications of Theorem 6.2

Theorem 6.2 applies to numerous partition functions considered in Chapters

1 and 2. We shall see in Chapter 11, Example 6, that it also produces asymptotic
formulas for many plane partition functions.
THEOREM 6.3. (Cf. Eq. (5.1.2).)

p ( n ) ^ V

Proof. In Theorem 6.2, set an = 1 for all w; then r(n) = p(n); D(s) = ((s)
(and C'(0) = - Jlog(27r), C(0) = - 4); and g{x) = (1 - O " 1 . It is an
easy matter to check the various conditions listed in (6.1.3)—(6.1.5), and the
result above follows from Theorem 6.2 once the indicated substitutions are
made. •
THEOREM 6.4. Let Hk a denote all positive integers congruent to a modulo k.
Then for 1 < a ^ k,

In"

where

C =

and

This result follows in exactly the way Theorem 6.3 was proved. Now D(s) =
/c~5((s, a/k), where ((s, h) = ^ n > 0 (n + ^)~ s ^s t n e Hurwitz zeta function,
and g(s) = e'a\\ - e^T1- •

Examples

1. Let Hk ±a denote the set of positive integers congruent to a or — a

modulo k. Then for k prime and 1 ^ a < fc/2,
,„„ „ , csc(7ra//c) _ 3 / 4

Cambridge Books Online © Cambridge University Press, 2010

98 The Asymptotics of Infinite Product Generating Functions Chap. 6

2. In order to compute explicitly the r(n) in Theorem 6.2, we use the well-
known logarithmic derivative:

/V=l \d\N

From this it follows that

nr(n) = £ r(« —

where

3. It can be shown by partial summation that the condition ^ u ^ f a n =

At*/a + O(/a~£) implies D(s) is continuable for Re s > a — e with a first order
pole of residue A at s = a.
4. If Xi^n^r a* ^ 4f a /a f° r a11 r ^ 0, then for X > 0

00

Z mame~km ^A?ra~lr(<x + 1).

5. If it is true that the an are nonnegative and that ^ U n ^ , an ^ — for

all t ^ 0, then we can prove by essentially elementary means that

fin) < exp(rt* / ( a + 1 ) (l + l)(Ar(oc + l)C(a

To see this we note from Example 2 that

The desired inequality is trivial for n = 1, and we may use the recurrence
to prove the full inequality by induction. We sketch the procedure:
n n

«K^) < Z Z ovvc\p[c(n - /cu) a/(a+1) ]

ft= 1 v= I
kv<n

< Z Z avvexp
k=1u=1

Cambridge Books Online © Cambridge University Press, 2010

Examples 99

(by Example 4 with A = c(xkn~{a+l)~\oc + I)" 1 )

= exp(cna/i*+l))Ar(<x + O c - ' - ' a " " - 1 ^ * + l)a+1C(a + 1)

a/(3t+1)
)c- a - 1 [Xa- 2 - 1 (a + l ) a + 1 r ( a + l)C(a + 1)]

= nexp(cM a/(a+1) ) (since c = (1 + -a ) ( / i r ( a + l)C(a + l)) l / ( 3 t + 1 ) ) •

\ \ / /

Unfortunately not much is known about problems when a series rather

than a product is involved. The following examples treat some of the very
intricate researches of Meinardus on partitions with difference d between
parts.
6. If Ad(n) denotes the number of partitions of n with difference at least
d between parts, then

Fd(q) = Z Un)qn = I ?-—

n=0

7*. Using Eqs. (2.2.5) and (2.2.10), we can prove that

Fd(q)= Res\\((oq)n\q4;qd)n(- ai-'iq')^- <oqd\qd)J^

o = 0 [U> J

8*. Example 7 provides a formula for Fd(q) that can yield the behavior
of Fd(q) for \q\ near 1. Using the circle method, we can then show that

where

and ad is the positive real root of xd 4- x — 1 = 0.

9. It is possible to prove that A2 = 7r2/15 by comparing the result in
Example 1 with a = 1, k = 5. (Corollary 8.6 must also be invoked to obtain
that A2(n) = r(n)).
10*. If H is the union of a finite number of arithmetic progressions of the
i + nM}?=0(l ^ i ^ r, 1 ^ ak ^ M), thenp( t 4 //"(^ 1),/1), thenum-

Cambridge Books Online © Cambridge University Press, 2010

100 The Asymptotics of Infinite Product Generating Functions Chap. 6

ber of partitions of n into distinct parts taken from H, is monotonically

increasing for n sufficiently large if and only if g.c.d.(al9 aly..., an M) = 1.
11*. The partition function p("//"(^ 1), n) defined in Example 10 satisfies
the following asymptotic relation

p{"H"{^ 1), n) = 2 ( ( r - 3 ) / 2 - ^ + f l 2 + - +

provided g.c.d.(al9 a2,..., an M) = 1.

12*. The number pnk ( = p{n, k, n) — p(n, k — 1, ri)) of partitions of n into
exactly k parts is unimodal in k if n is sufficiently large (i.e. there exists
kt = kt(n) such that pn,k > pntk+l for /c > kl9 and />„,* < p nJk+1 for k < kt).
13*. Almost all partitions n0 of n (i.e. with the exception of o(p(n))) have
#(7r0) summands where

provided only that w(n) is nondecreasing and tends to infinity with n.

14*. If 0 < a < |, then almost all partitions of n (i.e. with the exception
of o(p(n))) contain

summands which are < n\yj6jn).

15*. Let k ^ 2, 5 > 0 be fixed. For each partition nt of n, we denote by
S( the number of different summands of TT£. Then the set of integers that are
summands common to each of the partitions nl9 n2,..., nk of n is at least
as large as ((l//c) — S) m a x ^ y ^ Sj for almost all /c-tuples (nu n2i..., nk) of
partitions of n (i.e. with the exception of o(p(n)k) /c-tuples of partitions of n).
16*. If the /c-tuples of partitions in Example 14 are restricted to those
with distinct parts then an analogous result holds with ((IIk) — S) replaced
by [(l/(/c2k log 2)) - <5], and p(n) replaced by p(^, n).

Notes

We have chosen to present the work of Meinardus. However, there have

been numerous significant papers on the subject of this chapter; in particular,
we cite those of Ingham (1941), Brigham (1950), Auluck and Haselgrove
(1952), Haselgrove and Temperley (1954), Kohlbecker (1958), Roth
and Szekeres (1954), Szekeres (1951, 1953), and Wright (1931, 1933, 1934).
Recently B. Richmond (1972, 1975a, b) has extended the important work of
Roth and Szekeres (1954); and W. Schwarz (1967, 1968, 1969) has also made
important contributions to this area. Reviews of recent work occur in Sec-
tion P72 of LeVeque (1974).
Example 1. Livingood (1945).
Example 2. The case in which ad = 1 was treated by Vahlen (1893), although
very similar results were found by Euler; ad = d was done by MacMahon
(1923).

Cambridge Books Online © Cambridge University Press, 2010

References 101

Examples 3-5 present a modification of work done by Erdos (1942) on

Pin).
Examples 6-8 appear in Meinardus (1954b).
Example 9. See Watson (1937) for a history of such problems.
Example 10. Bateman and Erdos (1956).
Example 11. Richmond (1972).
Example 12. Szekeres (1953).
Example 13. Erdos and Lehner (1941).
Example 14. Turan (1975).
Example 15. Turan (1973, 1975).
Example 16. Turan (1974, 1975).
In Examples 13-16 some of the interesting recent theorems on the statistical
aspects of partitions are listed. For an extensive account of these and related
work on the statistical aspects of group theory we refer the reader to Erdos
and Lehner (1941), Erdos and Turan (1965, 1967a, b, c, 1968, 1971a, b),
Lehmer (1972a, b, c), Richmond (1975c), Turan (1973, 1974, 1975).

References

Auluck, F. C , and Haselgrove, C. B. (1952). "On Ingham's Tanberian theorem for

partitions," Proc. Cambridge Phil. Soc. 48, 566-570.
Bateman, P., and Erdos, P. (1956). "Monotonicity of partition functions,*' Mathematika
3, 1-14.
Brigham, N. A. (1950). "A general asymptotic formula for partition functions," Proc.
Amer. Math. Soc. 1, 182-191.
Erdos, P. (1942). "On an elementary proof of some asymptotic formulas in the theory
of partitions," Ann. of Math. 43, 437-450.
Erdos, P., and Lehner, J. (1941). "The distribution of the number of summands in the
partitions of a positive integer," Duke Math. J. 8, 335-345.
Erdos, P., and Turan, P. (1965). "On some problems of a statistical group theory, I,"
Z. Wahrscheinlichkeitstheorie und Verw. Gebiete 4, 175-186.
Erdos, P., and Turan, P. (1967a). "On some problems of a statistical group theory, II,"
Ada Math. Acad. Sci. Hungar. 18, 151-163.
Erdos, P., and Turan, P. (1967b). "On some problems of a statistical group theory, III,"
Ada Math. Acad. Sci. Hungar. 18, 171-173.
Erdos, P., and Turan, P. (1967c). "Certain problems of statistical group theory,"
Magyar Tud. Akad. Mat. Fiz. Oszt. Kozl. 17, 51-57.
Erdos, P., and Turan, P. (1968). "On some problems of a statistical group theory, IV,*t
Acta Math. Acad. Sci. Hungar. 19, 413-435.
Erdos, P., and Turan, P. (1971a). "On some general problems in the theory of partitions,
I," Acta Arith. 18, 53-62.
Erdos, P., and Turan, P. (1971b). "On some problems of a statistical group theory, V,"
Period. Math. Hungar. 1, No. 1, 5-13.
Hardy, G. H., and Ramanujan, S. (1918). "Asymptotic formulae in combinatory
analysis," Proc. London Math. Soc. (2) 17, 75-115. (Also in Collected Papers of
S. Ramanujan, pp. 276-309, Cambridge Univ. Press, Cambridge, 1927; reprinted
by Chelsea, New York, 1962.)
Haselgrove, C. B., and Temperley, H. N. V. (1954).' "Asymptotic formulae in the theory
of partitions," Proc. Cambridge Phil. Soc. 50, 225-241.

Cambridge Books Online © Cambridge University Press, 2010

102 The Asymptotics of Infinite Product Generating Functions Chap. 6

Ingham, A. E. (1941). "A Tauberian theorem for partitions," Ann. of Math. (2) 42,
1075-1090.
Kohlbecker, E. E. (1958). "Weak asymptotic properties of partitions," Trans. Amer.
Math. Soc. 88, 346-365.
Lehmer, D. H. (1972a). "On reciprocally weighted partitions," Ada Arith. 21, 379-388.
Lehmer, D. H. (1972b). "Calculating moments of partitions," Proceedings of the Second
Manitoba Conference on Numerical Mathematics, Oct. 5-7, 1972.
Lehmer, D. H. (1972c). "Some structural aspects of the partitions of an integer," Pro-
ceedings of the 1972 Number Theory Conference, University of Colorado, Boulder,
pp. 122-127.
LeVeque, W. J. (1974). Reviews in Number Theory, Vol. 4, Amer. Math. Soc, Providence,
R.I.
Livingood, J. (1945). "A partition function with the prime modulus p > 3,*' Amer. J.
Math. 67, 194-208.
MacMahon, P. A. (1923). "The connexion between the sum of the squares of the divisors
and the number of partitions of a given number," Messenger of Math. 52, 113-116.
Meinardus, G. (1954a). "Asymptotische Aussagen iiber Partitionen," Math. Z. 59,
388-398.
Meinardus, G. (1954b). "Uber Partitionen mit Differenzenbedingungen," Math. Z. 61,
289-302.
Richmond, B. (1972). "On a conjecture of Andrews," Utilitas Math. 2, 3-8.
Richmond, B. (1975a). "Asymptotic relations for partitions," /. Number Theory 7,
389-405.
Richmond, B. (1975b). "A general asymptotic result for partitions," Canadian J. Math.
27, 1083-1091.
Richmond, B. (1975c). "The moments of partitions, I," Acta Arith. 26, 411-425.
Roth, K. F., and Szekeres, G. (1954). "Some asymptotic formulae in the theory of
partitions," Quart. J. Math. Oxford Ser. (2) 5, 241-259.
Schwarz, W. (1967). "C. Mahler's Partitionsproblem," J. Reine Angew. Math. 229,
182-188.
Schwarz, W. (1968). "Schwache asymptotische Eigenschaften von Partitionen," /. Reine
Angew. Math. 232, 1-16.
Schwarz, W. (1969). "Asymptotische Formeln fur Partitionen," /. Reine Angew. Math.
234, 174-178.
Szekeres, G. (1951). "An asymptotic formula in the theory of partitions, I," Quart. J.
Math. Oxford Ser. (2) 2, 85-108.
Szekeres, G. (1953). "An asymptotic formula in the theory of partitions, II," Quart.
J. Math. Oxford Ser. (2), 4, 96-111.
Turan, P. (1973). "Combinatorics, partitions, group theory," Proceedings of the Interna-
tional Colloquium on Combinatorial Theories, Rome, September 3-15, 1973.
Turan, P. (1974). "On a property of partitions," / . Number Theory 6, 405-411.
Turan, P. (1975). "On some phenomena in the theory of partitions," Journees Arith-
metiques de Bordeaux, Societe Math, de France, Asterisque 24-25, pp. 311-319.
Vahlen, K. T. (1893). "Beitrage zu einer additiven Zahlentheorie," Crelle J. 112, 1-36.
Watson, G. N. (1937). "A note on Spence's logarithmic transcendant," Quart. J. Math.
Oxford Ser. (2) 8, 39-42.
Wright, E. M. (1931). "Asymptotic partition formulae, I," Quart J. Math. Oxford Ser.
(2)2, 177-189.
Wright, E. M. (1933). "Asymptotic partition formulae, II," Proc. London Math. Soc. (2)
36, 117-141.
Wright, E. M. (1934). "Asymptotic partition formulae, III," Acta Math. 63, 143-191.

Cambridge Books Online © Cambridge University Press, 2010

CHAPTER 7

Idenfifies of the Rogers-Ramanujan

Type

7.1 Introduction

The Rogers-Ramanujan identities provide one of the most fascinating

chapters in the history of partitions. The story begins with the discovery of
the Indian genius S. Ramanujan by G. H. Hardy. In his first letter to Hardy
(dated January 16, 1913) Ramanujan stated several marvelous theorems on
continued fractions. Two are of special interest to us now:

1 _/1/5 + V5 _V5 + 1\ 2,/5

1-- j-
V5-l\^
= M/^-^^- x : V^|e/ 5
1+'

Of these (and a related formula), Hardy says, in the article "The Indian
Mathematician Ramanujan" (Amer. Math. Monthly 44 (1937), p. 144),
"[These formulas] defeated me completely. I had never seen anything in the
least like them before. A single look at them is enough to show that they could
only be written down by a mathematician of the highest class. They must be
true because, if they were not true, no one would have had the imagination
to invent them."
Before we conclude our history, it is useful to see how these two continued
fractions are related to mathematical constructs of the type previously
considered. Let us consider F(x) to be a function analytic in x at 0, F(0) = 1,

ENCYCLOPEDIA OF MATHEMATICS and Its Applications, Gian-Carlo Rota (ed.).

2, George E. Andrews, The Theory of Partitions
Copyright © 1976 by Addison-Wesley Publishing Company, Inc., Advanced Book Program.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical photocopying,
recording, or otherwise, without the prior permission of the publisher.

103

Cambridge Books Online © Cambridge University Press, 2010

104 Identities of the Rogers-Ramanujan Type Chap. 7.1

and such that

F(x) = F(xq) + xqF(xq2\ (7.1.1)

a linear second-order ^-difference equation. Then if c(x, q) = F(x)/F(xq),

we have that
Xq
c(x, q) = 1 +
c(xq, q)

and so Ramanujan's continued fraction theorems are precisely evaluations

ofc(l,<T 2 n )andc(l, - e~n).
If now we write F(x) = Y*n>o An(q)xn, a n d we substitute this series into
(7.1.1), we find, by comparing coefficients of x", that

An(q) = qnAn(q) + q2"'^-^). (7.1.3)

Hence
2n-3)

Hence

F(x)= f j ^ - - (7.1.5)

Surprisingly, F(l) and F(<?) may be evaluated in terms of infinite products

(as we shall prove later in this chapter), namely,
4 9
q q
= l +—^—+
1 - q (1 -<?)(1 - q2) (1 - <?)(1 - <?2)(1 - q3)

= U(l-q5n+lr1(i-q5n+*rl (7-1.6)

and

q2 q6 q12
H H v
F(q) = 1 + 1 - (1 - tfXl - q2) • (1 - <?)(! - q2)(l - q3)
9

= n a - q5m+2r\i - q5n+3rl. (7.1.7)

n= 0

Cambridge Books Online © Cambridge University Press, 2010

7.1 Introduction 105

From (7.1.6) and (7.1.7), we directly obtain an infinite product representa-

tion of c(l, q) = F(l)IF(q), and we may then deduce Ramanujan's continued
fraction theorems by an appeal to the theory of elliptic theta functions.
Hardy has asserted that "it would be difficult to find more beautiful
formulae than the 'Rogers-Ramanujan' identities," and he succinctly fills
in the remainder of their early history in the following passage, taken from
Ramanujan (Cambridge Univ. Press, 1940, p. 91):

They were found first in 1894 by Rogers, a mathematician of great

talent but comparatively little reputation, now remembered mainly from
Ramanujan's rediscovery of his work. Rogers was a fine analyst, whose
gifts were, on a smaller scale, not unlike Ramanujan's; but no one paid
much attention to anything he did, and the particular paper in which he
proved the formulae was quite neglected.
Ramanujan rediscovered the formulae sometime before 1913. He had then
no proof (and knew that he had none), and none of the mathematicians to
whom I communicated the formulae could find one. They are therefore
stated without proof in the second volume of MacMahon's Combinatory
Analysis.
The mystery was solved, trebly, in 1917. In that year Ramanujan, looking
through old volumes of the Proceedings of the London Mathematical
Society, came accidentally across Rogers's paper. I can remember very well
his surprise, and the admiration which he expressed for Rogers's work.
A correspondence followed in the course of which Rogers was led to a con-
siderable simplification of his original proof. About the same time I. Schur,
who was then cut off from England by the war, rediscovered the identities
again. Schur published two proofs, one of which is "combinatorial" and
quite unlike any other proof known. There are now seven published proofs,
the four referred to already, the two much simpler proofs found later by
Rogers and Ramanujan and published in the Papers, and a much later proof
by Watson based on quite different ideas. None of these proofs can be called
both "simple" and "straightforward," since the simplest are essentially ver-
ifications; and no doubt it would be unreasonable to expect a really easy
proof.

Hardy's comments about the nonexistence of a really easy proof of the

Rogers-Ramanujan identities are still true today. The approach that we shall
use here is related to one of the "verification" proofs referred to by Hardy.
On the one hand, this may seem inappropriate for this particular book;
however, we may defend our choice in observing that our approach has been
by far the most fruitful in extending our knowledge of Rogers-Ramanujan
type identities.
We shall, in fact, show that other partition identities may be treated in the
same way, and as an example, we shall prove not only the Rogers-Ramanujan
identities, but also B. Gordon's beautiful generalization of them, as well as
other identities due independently to H. Gollnitz and B. Gordon.

Cambridge Books Online © Cambridge University Press, 2010

106 Identities of the Rogers-Ramanujan Type Chap. 7.2

In every case, we find that theorems on partitions correspond to generating

function identities. The partition-theoretic interpretations of (7.1.6) and
(7.1.7) are given in Section 7.3. If the reader wishes he may proceed by accepting
Lemmas 7.2 — 7.4 and then going directly to the partition theorems in
Section 7.3.

7.2 The Generating Functions

We shall consider, for |x| < \q\~x, k l < 1>

x
LJ ( n . Y . n \ _ V 4 a { \ - xq ){axq )00(q )n
H kti(a,x,q) -

Jkti(a;x;q) = Hkti(a;xq;q) - xqaH^^^a; xq;q). (7.2.2)

We note that any value for a is admissible, even a = 0, since ^"(a" 1 ),, =
(a - l)(a - ^)- • -(a - q11"1) is merely a polynomial in a whose value at 0
i s ( - i)V(""1)/2.
Very little can be said in way of motivation for this section. Actually
extensive work in the theory of basic hypergeometric series and partition
identities shows that "well-poised" basic hypergeometric series provide the
generating functions for numerous families of partition identities. The series
in (7.2.1) (once the infinite products have been factored out: " £ OLn(Aqn)^ =
(A)*1 Y, xJiA)*1") is an example of a well-poised series.

LEMMA 7.1

Hkti(a\x\q) - H^-ifaxiq) = xl-lJk^i+l(a; x\ q). (7.2.3)

Proof. Noting that

q-in{\ - xlq2ni) - < r ( ' - 1 ) n ( l " x f - V 0 " 1 * )

= q-in(\ - qn) + X1" Y ^ - ^ l - xqn),

we see that

Hkti(a;x;q) - Hkti_l{a\x\q)
°o Jfc/!_fcn2 + n ™ n / v ^ n + 1 \ / r , ~ l \

= L (n>./Y_nx <i \i-<i)

n=O W U X ^ Joo

Cambridge Books Online © Cambridge University Press, 2010

7.2 The Generating Functions 107

n=l

.-o

00 v f c

LEMMA 7.2

- aJktk_i + 2(a; xq; q)). (7.2.4)

Proo/.

Jk.fcl x; q) - Jkti-i(a',x;q)

= (^kf,<«; xq;q) - Hk%i.t(a; xq; q))

^.^a; xq; q) - Hkti_2(a;xq; q))

In treating problems, there are numerous instances in which we require an

infinite product representation of a generating function. Such representations
are given in the following lemmas.

Cambridge Books Online © Cambridge University Press, 2010

108 Identities of the Rogers-Ramanujan Type Chap. 7.2

LEMMA 7.3. For 1 ^ i ^ /c, \q\ < 1,

JU0',Uq)= ft (l-^T1- (7.2.5)

Proo/. By (7.2.2)

n= 0

= (^)i1 Z ( - i)
n= 0

fl (l-^)"1 (by Corollary 2.9). •

n&0t±i(mod2k+ 1)

LEMMA 7.4. For 1 ^ i < /c, \q\ < 1,

^(-^" 1 ;i;^ 2 )= ft a-*")" 1 - (7.2.6)

n^0,±(

Proof.

^ ) J / i\na2kn+2kn2-(2i-l)n

+ 2ta»-(2l-3).

f2i-2+4ni-4«\"|

(1

(rt~2 in ,,2i' + 2 n i _i_ _ 1 — 2 i « + 2 n _ - 1 + 2 i + 2 n i — 2n\

x \q - <? +^ ^j )

V ^ i ? )oo y / _ j\n
2 2
(4 ;€ ) o

Cambridge Books Online © Cambridge University Press, 2010

7.3 The Rogers-Ramanujan Identities and Gordon's GeneraL£ation 109

^*

/ou
^ ' •* n=0

X (1 _

n
n= 1
n / 2 (mod 4)
i - D(mod4k)

7.3 The Rogers-Ramanujan Identities and Gordon's Generalization

In this section we shall utilize the analytic work in Section 7.2 to prove
the following theorem, which is due to B. Gordon.
THEOREM 7.5. Let Bki(n) denote the number of partitions of n of the form
(b1b2'' bs), where bj — by+*_i > 2, and at most i — 1 of the bj equal 1.
Let Aki(n) denote the number of partitions of n into parts ^ 0 , ± /(mod2fc 4- 1).
Then Akti(n) = BkJ(n) for all n.
Before we prove Theorem 7.5, it is appropriate to give center stage to its
two most celebrated corollaries, the Rogers-Ramanujan identities (stated in
terms of partitions).
COROLLARY 7.6 (The first Rogers-Ramanujan identity). The partitions of
an integer n in which the difference between any two parts is at least 2 are
equinumerous with the partitions of n into parts = 1 or 4 (modulo 5).
Proof. Take k = i = 2 in Theorem 7.5. •
COROLLARY 7.7 (The second Rogers-Ramanujan identity). The partitions
of an integer n in which each part exceeds 1 and the difference between any
two parts is at least 2 are equinumerous with the partitions of n into parts
= 2 or 3 (modulo 5).

Proof. Take k = i + 1 = 2 in Theorem 7.5. •

Proof of Theorem 7.5. Let bkJ[m, n) denote the number of partitions
(b{b2'' bm)o(n with exactly m parts such that bi > b^x, bj - bj + k_x ^ 2,
and at most i - 1 of the b} equal 1. Then for 1 ^ i ^ k

Cambridge Books Online © Cambridge University Press, 2010

110 Identities of the Rogers-Ramanujan Type Chap. 7.3

l if m = n = 0
, n) = | Q if m ^ o or n < 0 but (m, n) ^ (0, 0);

(7.3.1)

VoC" 1 , w) = 0; (7.3.2)

for 1 < / < fc

&*,,(™> n) - bjk.i-^m, w) = fcM_l+1(m - i + 1, w - m). (7.3.3)

Equations (7.3.1) and (7.3.2) are obvious once we recall that the only partition
that is either of a nonpositive number or has a nonpositive number of parts
is the empty partition of 0.
Equation (7.3.3) requires careful attention: bki(m,n) — bktl-_i(wi, n)
enumerates the number of partitions among those enumerated by bki(m, n)
that have exactly / — 1 appearances of 1. Let us transform this set of partitions
by deleting the / — 1 ones, and then subtracting 1 from each of the remaining
parts. The resulting partitions (b^- • - 6 m _ , + 1 ) have m — i + 1 parts; they
partition n — m, and the parts satisfy b/ — b'j+k_l ^ 2. Since originally 1
appeared i — 1 times and the total number appearances of ones and twos
could not exceed k — 1 (due to the difference condition), we see that originally
2 appeared at most (k — / - h i ) — 1 times, and thus after the transformation 1
appears at most k — i + 1 times. The transformation described above
establishes a one-to-one correspondence between the partitions enumerated
by bki(m,n) — 6 f c l _ 1 (m,M)andthoseenumeratedby6 f c f c _ I + 1 (m — / 4- l,n — m).
Hence (7.3.3) is established.
We now make a simple yet essential observation: the bki(m, n) (0 ^ i ^ k)
are uniquely determined by (7.3.1), (7.3.2), and (7.3.3). To see this, proceed
by a double mathematical induction first on n and then on i. Equation (7.3.1)
takes care of n ^ 0, m ^ 0, i > 0. Equation (7.3.2) handles all n when
i = 0. Equation (7.3.3) represents bki(m, n) as a two-term sum in which the
first term has a lower i index and the second a lower n index (since we can
assume m > 0).
Now let us consider

£ £ ckti(m, n)xmq\
m=0 n=0

From the fact that for 1 ^ i

we see that for 1 ^ i

Cambridge Books Online © Cambridge University Press, 2010

7.3 The Rogers-Ramanujan Identities and Gordon's Generalization 111

(7.3.4)

From the fact that

7*.o(O; X; q) = Hk,0(0; xq\ q) = 0,

we see that

ck0(mn n) = 0. (7.3.5)

Finally, by comparing coefficients of xmqn on both sides of (7.2.4) with a — 0,

we see that

cki(m, n) — ffc,i-i(w, n) = c fcfc _ l+ x(m — / -f 1, n — m). (7.3.6)

So we see that the ckti(m, n) also satisfy the system of equations (7.3.1)-
(7.3.3) that uniquely defines the bki(my n). Therefore, bkti(m, n) = ckti(m, n)
for all m and n with 0 ^ i ^ k.
Hence, since Y,m>o bkti(m, n) = Bkti(n), we see that

= Jk,i(0;l;q)

fl d-9")" 1 (by (7.2.5))

n= I
n/0,±i(mod2k+
i(mod2k+ 1)

= Z 'ViOO'T (by Theorem 1.1, Eq. (1.2.3)).

Comparing coefficients of qn in the extremes of the string of equations above,

we see that Aki(n) = Bk ti(n). •

Theorem 7.5 has an analytic counterpart, which we shall prove before

proceeding.

THEOREM 7.8. For 1 ^ i ^ fe, k ^ 2, |^| < 1

1 )
(7.3.7)

where Nj = tij + nJ+i + • • • + w t _,.

Cambridge Books Online © Cambridge University Press, 2010

112 Identities of the Rogers-Ramanujan Type Chap. 7.3

Proof. We shall prove that

A,,(0;x;4)= Z L-r j-.

(7.3.8)

We obtain Eq. (7.3.7) from (7.3.8) by setting x = 1 and invoking (7.2.5).

Equation (7.3.8) itself follows from
(fc- 1 )n (k-I)n2
Jk,i(0\x;q) = Z L Jk_Ui(0;xq2n;q), (7.3.9)

which may be seen immediately by induction on k once we observe that

•J*,*+i(0; x; q) = Jktk(O;x;q) (set i = k + 1 in (7.2.4) and recall that
A.o(0; *$;?) = Hkf0(0, ^ 2 ; ^) = 0) and that J 1§1 (0; x; ^) = 1 (since by

.i(O;O;<z)= 1).
To prove (7.3.9), we define

(k-l)n (k-

Then for 1 < i < /c,

Rkttf};q) = Rkt1Lx;0)= 1 (7.3.10)

and

^k,o(x;^) = 0. (7.3.11)

Finally

(fcl)n ( k l ) n + (fci)n
= I L (Jk-Li(0\xq2n;q)-qnJk_lti_1(0;xq2n;q))

(k-l)n (k-l)n2 + (k-i)n

= Z V; (Jk-i,i-i(0',xq2n;q)
>0 (q)

(k - 1 )n _(k - 1 )n 2 + (k - i)n

=Z
11
" Z

Cambridge Books Online © Cambridge University Press, 2010

7.4 The Gollnitz-Gordon Identities and Their Generalization 113

X
_ fc-1 2k-1- 1 V H_

Cit — 1 )n (k - 1 )n1 + (k + ii - 2 )n

k - l)n ( k - 1 )n2 + ( i k - i-2)n

|
R k . 4 - | . + l (.v^,^). (7.3.12)

Recalling that the coefficients in the expansion o(Jki(0; x; </) were uniquely
determined by (7.3.4), (7.3.5), and (7.3.6), we conclude that since Rkti(x; q)
satisfies (7.3.10), (7.3.11), and (7.3.12), and thus its coefficients must satisfy
(7.3.4), (7.3.5), and (7.3.6), therefore Rkti(x; q) = Jkti(0\ x; q) for 0 ^ i ^ k.
Thus we have (7.3.9) and with it Theorem 7.8. •

COROLLARY 7.9 (Eq. (7.1.6)).

.
2 2 3
- q (1 X1 ) (i )(1 )(1 J

Proo/. Set A: = / = 2 in Theorem 7.8.

COROLLARY 7.10 (Eq. (7.1.7)).

q22_ </6
+
f-f q

/ . Set k = 2, i = 1 in Theorem 7.8. •

7.4 The Gollnitz-Gordon Identities and Their Generalization

To appreciate the general area of partition identities such as the Rogers-

Ramanujan identities, we turn to results discovered independently in the 1960s

Cambridge Books Online © Cambridge University Press, 2010

114 Identities of the Rogers-Ramanujan Type Chap. 7.4

by H. Gollnitz and B. Gordon. Actually we shall prove a general theorem that

reduces to the Gollnitz-Gordon identities in the special cases in which
k = i -.= 2, k = i + 1 = 2.

THEOREM 7.11. Let i and k be integers with 0 < i ^ k. Let Cki{n) denote
the number of partitions of n into parts ^ 2 (mod 4) and ^ 0, + (2/ — 1)
(mod 4/c). Let Dk t(n) denote the number of partitions (blb2' • 'bs) of n in
which no odd part is repeated, bj ^ bj+l, bj — bj + k_l ^ 2 if bj odd,
bj — bj+k-Y > 2 if bj even, and at most i — 1 parts are ^ 2. Then Cki(n) =

Remark. The proof of this theorem is very similar to that of Theorem 7.5.
The role of Jkti(0\ x; q) in Theorem 7.5 is now played by Jkti{q~l\ x\ q2).
We shall, therefore, proceed somewhat rapidly through those portions of this
theorem that are familiar ground.
Proof. Let dkJ(m, n) denote the number of partitions enumerated by
Dki(n) that have exactly m parts. As in Theorem 7.5, we see that for 1 ^ / ^ k

if m = n = 0
if m ^ 0 or n ^ 0 but (m, n) = (0, 0). (7.4.1)

Next we assert that

*4,i(m> n) =
dktk(m, n — 2m), (7.4.2)
and for 1 < / < fc

dkti(m, n) - c/fc.f-^m, n) = dktk_i+l(m - i + 1, n - 2m)

4- dktk_i + 2(m - i + \,n - 2m + 1). (7.4.3)

The proof of (7.4.2) and (7.4.3) is exactly like the proof of (7.3.3). For
example, dkti(m9 n) — dkj_l(m, n) enumerates the number of partitions of
the type enumerated by dki(m, n) with the added condition that the total
number of parts each ^ 2 is exactly / — 1. Such partitions may be split into
two disjoint classes: (1) those that have 1 as a part; (2) those that do not.
We transform the partitions in class 1 by deleting the 1 and the(/ — 2)2's, and
subtracting 2 from each of the remaining parts. This transformation establishes
a one-to-one correspondence between the partitions in class 1 and those
enumerated by dkJc_i + 2(m — i + 1, n — 2m + 1). We transform the parti-
tions in class 2 by deleting the i — 1 2's and subtracting 2 from each of the
remaining parts. In this way, a one-to-one correspondence is established
between the partitions in class 2 and those enumerated by dkk.i+l(m — i + 1,
n — 2m). This establishes (7.4.3), and (7.4.2) is treated similarly. Furthermore
(just as the bkti(m, n) were uniquely determined by (7.3.1)-(7.3.3)), we may

Cambridge Books Online © Cambridge University Press, 2010

7.4 The Gollnitz-Gordon Identities and Their Generalization 115

easily prove that the dki(m, n) are uniquely determined by (7.4.1), (7.4.2), and
(7.4.3).
Now we consider

A,,<-<T1;*;<Z2)= t t ekti(m, n)xmq\

m=0 II = O

That Eq. (7.4.1) is also true for the ekyi(m, n) follows from

JU- q'\o; <?2) = JU- q~l> *; q%=0 = i,

while the truth of (7.4.2) for the eki{m, n) follows from

A . i ( - q~\ x; q2) = / / , , , ( - q-l;xq; q 2) (by (7.2.2))

= Jtjk-q~l\xq\q2) (by (7.2.3)).

Finally, we may deduce (7.4.3) for the ekti(m, n) from the fact that

Jk.ii- q-\x;q2)-Jk<i^{-q-\x;q2)

= (V)'- 1 (J tJk -, + 1 ( - q-'ixq'iq2) + q-lJkJk-l+i(- q~*; xq2; q2))

• vi-l/I2i-3 r / - 1. Y_2 . _2\

x J
+ % k,k-i+2K— q , xq , q )-
Hence, dkJ(m, n) = eki(m, n) for 1 ^ / ^ k and all m and n.
Consequently, for 1 < i < n

ft d-*")"1 (by (7.2.6))

= I Ctil(n)«" (by Theorem 1.1, Eq. (1.2.3)).

n= 0

Comparing coefficients of qn in the extremes of the string of equations above,

we see that Ckti(n) = Dki(n) for 1 ^ i'^fcand all n. •

We conclude this section by remarking that a result analogous to Theorem

7.8 exists for / M ( - q~lA\ <?2)- For example, with Nj = ni -\ + nk_x

Cambridge Books Online © Cambridge University Press, 2010

116 Identities of the Rogers-Ramanujan Type Chap. 7

f[ (1-qT1- (7.4.4)
n— 1
2(mod4)
fc- I)(mod4/c)

Examples

1. Let Sj(m, n) denote the number of partitions (/.{/.2'' *Am) of n such that
for 1 ^ / < m, /f — A l + 1 ^ 3 with strict inequality if 31/.,. and Xm ^ 7. We
can prove (similar to the proof of Theorem 7.5) that tffj(x) =Y^mtn^oSj(m,n)
x xmqn, then ~

/ i W - / 2 (x) =
fiM - / 3 (x) = xq2f2(xq3),
h(x) - Ux) = xq*Uxq\
Ux) = f^xq3).
2. From Example 1 it follows that
fx{x) = (1 + x^ + x<? 2 )/i(V) + x^3(l -
3. From Example 2, we deduce that

4. From Example 3 we may deduce Schur's partition theorem: The number

of partitions of n into parts congruent to 1 or 5 modulo 6 equals the number
of partitions of n in which the difference between all parts is at least 3 and
between multiples of 3 is at least 6.
5. The procedure described in Examples 1-4 may be extended to prove a
family of identities like Schur's theorem: Let j denote the least positive
residue of j modulo 2"; let co(j) denote the number of powers of 2 used in
the binary representation of;, and let v(j) denote the smallest power of 2
appearing in the binary representation of j (e.g., if n = 4, j = 28, then
j = 12, co(j) = 3, t>0) = 4, co(j) = 2, v(}) = 4). Let A(N) denote the num-
ber of partitions of N into parts = 1, 2" + 1, 2" + 3 , . . . , 2" + 2"' 1 - 1
(mod 2" +1 - 2). Let B(N) denote the number of partitions Uj/^,. • •» ^s) °f
N (s arbitrary) such that for 1 ^ / < s, Xt - /. f+1 ^ co(X f+1 )(2 n - 1) +
+ v(ki+l) - X,-+1. Then A(N) = B(N) for all N.
6. We may deduce Schur's theorem from Example 5 in the case in which
n = 2.
7. The case in which n = 3 of Example 5 asserts that the number of parti-
tions of N into parts = 1, 9, 11 (mod 14) equals the number of partitions
(^1^2'' *^s) (5 arbitrary) of n such that for 1 < / < s,

Cambridge Books Online © Cambridge University Press, 2010

Examples 117

7 if ;. f+1 = 1,2, 4 (mod 7),

12 if / i + l = 3 (mod 7),
10 if ;.,.+ 1 = 5, 6 (mod 7),
15 if Al + 1 = 0(mod7).

8. Let s/ka(N) denote the number of partitions of N into parts ^ 0,

± 2a (mod 4k + 2). Let ^k>a{N) denote the number of partitions ( l / l 2 / 2 3 / 3 • • •)
of N such that/! ^ 2a - 1, and

2k — 1 if fk is even,
2k if f: i

Then s/k,a(N) = &k,a(N) for all N.

We may prove this result, which resembles Theorem 7.5, by mimicking
the proof of Theorem 7.5 but using

instead of Jk ,(0; x; q).

9. The number of partitions of n into parts congruent to 2, 3, or 4 modulo 6
equals the number of partitions of n without consecutive integers, with no
ones, and with no part repeated three or more times.
The proof here considers

10. It is a simple matter to prove that for a = 1, 2, or 3

oo oo 2n2 + 2[(4-a)/2]n

PI
n^0,±a(mod 7)
I ^
This is established by showing that if

« . W = Ra(x9q) = ( - x^) w I 7-2T-2T

thcn/?3(x) - /?2(x) = x2q2Rl(xq); R2(x) - Rx(x) = xqR2(xq); R^x) =

R3(xq). From there it immediately follows that Ra(x) = J3tfl(0; x; q).
11 *. The number of partitions of n into parts congruent to 2, 5, or 11 modulo
12 equals the number of partitions (kxk2'' *AS) (5 arbitrary) of n such that
ks # 1 or 3, and for 1 ^ / < 5, / , — /,+ , ^ 6 with strict inequality for
;. l+1 = 0, 1, or 3 (mod 6).
This theorem is deduced from the following analytic identities. Define An
by

Cambridge Books Online © Cambridge University Press, 2010

118 Identities of the Rogers-Ramanujan Type Chap. 7

Let cm(n) denote the number of partitions {kx/.2' • -As) (5 arbitrary) of n such
that m ^ A1? xs ^ 1 or 3, and for 1 < / < 5, / , — / l + 1 ^ 6 with strict
inequality if Xi+ { = 0, 1 or 3 (mod 6), and let

dm(q) = I cm{n)qn.

Then

where [g]fl6 is [£] with q replaced by q6.

Notes
The material in the introduction comes primarily from Chapters 1 and 6
of Ramanujan by G. H. Hardy (1940). The first two continued fractions in
the introduction were proved by Watson (1929). The analytic identities in
Section 7.2 appear for a = 0 in Rogers (1919) and Selberg (1936) and for
a — — q~l in Andrews (1967d). Very general identities of this nature are
treated in Andrews (1968c, 1974b).
Theorem 7.5 is due to Gordon (1961), our proof is by Andrews (1966;
see also 1969d, 1975a, b). Theorem 7.8 is due to Andrews (1974c); a related
theorem involving the Alder polynomials is due to V. N. Singh (1957).
Theorem 7.11 in the case k = 2 is due to H. Gollnitz (1967) and B. Gordon
(1965); for arbitrary k the result was obtained by Andrews (1967d). Identity
(7.4.4) is also due to Andrews (1975d).
For more detailed accounts of work of this nature there are several survey
articles with extensive bibliographies: Alder (1969), Andrews (1970, 1972,
1974a), Gupta (1970). Reviews of recent work in this area occur in Section
P68 of LeVeque(1974).
Examples 1-3. Andrews (1968a).
Example 4. Andrews (1968a, 1971a, b), Schur (1926).
Examples 5-7. Andrews (1968b, 1969a, 1972).
Example 8. Andrews (1967b).
Example 9. Andrews (1967c, 1969c).
Example 10. Selberg (1936), Andrews (1968c, 1975a).
Example 11. Gollnitz (1967), Andrews (1969b).

References
Alder, H. L. (1969). "Partition identities —from Euler to the present," Amer. Math.
Monthly 76, 733-746.
Andrews, G. E. (1966). "An analytic proof of the Rogers-Ramanujan-Gordon identities,"
Amer. J. Math. 88, 844-846.
Andrews, G. E. (1967a). "On Sdiur's second partition theorem," Glasgow Math. J.
9, 127-132.
Andrews, G. E. (1967b). "Partition theorems related to the Rogers-Ramanujan iden-
tities," J. Combinatorial Theory 2, 422-430.

Cambridge Books Online © Cambridge University Press, 2010

References 119

Andrews, G. E. (1967c). "Some new partition theorems," J. Combinatorial Theory 2,

431-436.
Andrews, G. E. (1967d). "A generalization of the Gollnitz-Gordon partition theorems,"
Proc. Amer. Math. Soc. 18, 945-952.
Andrews, G. E. (1968a). "On partition functions related to Schur's second partition
theorem," Proc. Amer. Math. Soc. 18, 441-444.
Andrews, G. E. (1968b). "A new generalization of Schur's second partition theorem,"
Acta Arith. 14, 429-434.
Andrews, G. E. (1968c). "On qr-difference equations for certain well-poised basic hyper-
geometric series," Quart. J. Math. Oxford Ser. (2) 19, 433-447.
Andrews, G. E. (1969a). "A general theorem on partitions with difference conditions,"
Amer. J. Math. 91, 18-24.
Andrews, G. E. (1969b). "A partition theorem of Gollnitz and related formulae," J.
Reine Angew. Math. 236, 37-42.
Andrews, G. E. (1969c). "Some new partition theorems (II)," J. Combinatorial Theory
7, 262-263.
Andrews, G. E. (1969d). "A generalization of the classical partition theorems," Trans.
Amer. Math. Soc. 145, 205-221.
Andrews, G. E. (1970). "A polynomial identity which implies the Rogers-Ramanujan
identities," Scripta Math. 28, 297-305.
Andrews, G. E. (1971a). "The use of computers in search of identities of the Rogers-
Ramanujan type," Computers in Number Theory (A. O. L. Atkin and B. J. Birch,
eds.), pp. 377-387. Academic Press, London.
Andrews, G. E. (1971b). "On a theorem of Schur and Gleissburg," Arch. Math. (Basel)
22, 165-167.
Andrews, G. E. (1971c). Number Theory. Saunders, Philadelphia.
Andrews, G. E. (1972). "Partition identities," Advances in Math. 9, 10-51.
Andrews, G. E. (1974a). "A general theory of identities of the Rogers-Ramanujan type,"
Bull. Amer. Math. Soc. 80, 1033-1052.
Andrews, G. E. (1974b). "On the general Rogers-Ramanujan theorem," Mem. Amer.
Math. Soc. 152.
Andrews, G. E. (1974c). "An analytic generalization of the Rogers-Ramanujan identities
for odd moduli," Proc. Nat. Acad. Sci. USA 71, 4082-4085.
Andrews, G. E. (1975a). "On the Alder polynomials and a new generalization of the
Rogers-Ramanujan identities," Trans. Amer. Math. Soc. 204, 40-64.
Andrews, G. E. (1975b). "Partially ordered sets and the Rogers-Ramanujan identities,"
Aequationes Math. 12, 94-107.
Andrews, G. E. (1975c). "On Rogers-Ramanujan type identities related to the modulus
11," Proc. London Math. Soc. (3) 30, 330-346.
Andrews, G. E. (1975d). "Problems and prospects for basic hypergeometric functions,"
The Theory and Applications of Special Functions (R. Askey, ed.), pp. 191-224.
Academic Press, New York.
Cheema, M. S. (1969). "Duality in the theory of partitions," Res. Bull. Panjab Univ.
20, 201-206.
Gollnitz, H. (1967). "Partitionen mit Differenzenbedingungen," /. reine angew. Math.
225, 154-190.
Gordon, B. (1961). "A combinatorial generalization of the Rogers-Ramanujan iden-
tities," Amer. J. Math. 83, 393-399.
Gordon, B. (1965). "Some continued fractions of the Rogers-Ramanujan type," Duke
Math. J. 31, 741-748.

Cambridge Books Online © Cambridge University Press, 2010

120 Identities of the Rogers-Ramanujan Type Chap. 7

Gupta, H. (1970). "Partitions-a survey," J. Res. Nat. Bur. Standards 74B, 1-29.
Hardy, G. H. (1940). Ramanujan. Cambridge Univ. Press, London and New York;
reprinted by Chelsea, New York.
Lehner, J. (1941). "A partition function associated with the modulus five," Duke
Math. J. 8, 631-655.
LeVeque, W. J. (1974). Reviews in Number Theory, Vol. 4. Amer. Math. Soc, Providence,
R.I.
Rogers, L. J. (1894). "Second memoir on the expansion of certain infinite products,"
Proc. London Math. Soc. 25, 318-343.
Rogers, L. J. (1919). "Proof of certain identities in combinatory analysis," Proc.
Cambridge Phil. Soc. 19, 211-214.
Schur, I. J. (1917). "Ein Beitrag zur additiven Zahlentheorie und zur Theorie der
Kettenbruche," S.-B. Preuss. Akad. Wiss. Phys.-Math. Kl. pp. 302-321. (Reprinted
in I. Schur, Gesammelte Abhandlungen, Vol. 2, pp. 117-136. Springer, Berlin, 1973.)
Schur, I. J. (1926). "Zur additiven Zahlentheorie," S.-B. Preuss. Akad. Wiss. Phys.-Math.
KL pp. 488-495. (Reprinted in I. Schur, Gesammelte Abhandlungen, Vol. 3, pp. 43-50.
Springer, Berlin, 1973.)
Selberg, A. (1936). "Uber einige arithmetische Identitaten," Avhl. Norske Vid. 8, 23 pp.
Singh, V. N. (1957). "Certain generalized hypergeometric identities of the Rogers-
Ramanujan type," Pacific J. Math. 7, 1011-1014.
Watson, G. N. (1929). "Theorems stated by Ramanujan (VII): Theorems on continued
fractions," J. London Math. Soc. 4, 39-48.

Cambridge Books Online © Cambridge University Press, 2010

CHAPTER 8

A General Theory of Partition

Identities

8.1 Introduction
The results in Chapter 7 indicate that a general study of partition identities
such as the Rogers-Ramanujan identities or the Gollnitz-Gordon identities
should help to illuminate these appealing but seemingly unmotivated theorems.
In this chapter we shall undertake the foundations of this study. As will
become abundantly clear, there are very few truly satisfactory answers to the
questions that we shall examine. We shall instead have to settle for partial
answers. After presenting the fundamental structure of such problems in the
next section, we devote Section 8.3 to "partition ideals of order 1," a topic
which we can handle adequately and which suggests the type of answers
we would like for our general questions of Section 8.2. The final section of
the chapter describes a large class of partition problems wherein the related
generating function satisfies a linear homogeneous ^-difference equation with
polynomial coefficients. In some ways this final section is unsatisfactory, in
that the theory of g-difference equations has not been adequately developed
to provide answers generally to questions about partition identities and parti-
tion asymptotics; however, the theorems of Section 8.4 do suggest that
^-difference equations are indeed worthy of future research.

8.2 Foundations
We begin with a simple intuitive observation which forms the basis of our
work here. In all the partition identities considered in Chapter 7 (Theorem 7.5,
Corollaries 7.6 and 7.7, Theorem 7.11) partition functions were considered
that enumerated partitions lying in some subset C of the set of all partitions.
For example, C might be the set of all partitions where the difference between
parts is at least 2 (see Corollary 7.6), or C might be the set of all partitions
with parts congruent to 1,4,7 (mod 8) (see Theorem 7.11 when k — i = 2). The
interesting fact to note is that all the considered subsets C of partitions have

ENCYCLOPEDIA OF MATHEMATICS and Its Applications, Gian-Carlo Rota (ed.).

2, George E. Andrews, The Theory of Partitions
Copyright © 1976 by Addison-Wesley Publishing Company, Inc., Advanced Book Program.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical photocopying,
recording, or otherwise, without the prior permission of the publisher.
121

Cambridge Books Online © Cambridge University Press, 2010

122 A General Theory of Partition Identities Chap. 8.2

the property that if n is a partition in C and one or several parts are removed
from n to form a new partition n', then n' is also in C.
This observation suggests that we should define a partial ordering on
partitions by n' ^ n whenever any integer i appears at least as often in n as
in n'. This approach immediately leads us to a lattice structure on partitions:
DEFINITION 8.1. Let £f denote the set of all sequences {/,} °°= x of nonnegative
integers wherein only finitely many of t h e / , are nonzero.
For reference we note that each sequence {fi}^Ll = {/,-} in £f uniquely
corresponds to the partition ( l / l 2 / 2 3 / 3 4 / 4 - • •).

DEFINITION 8.2. We define a partial order " < " on £f by asserting that
{/,•} < {gii whenever/,- ^ gx for all i.

We note that this definition is consistent with the order structure intuitively
suggested at the beginning of this section. Furthermore, £f is actually a lattice
whose "meet" and "join" operations are given by {/J n {#J = {min(/,-, gt)},
{ft) u {9i) = {max(/, 9i)}.
DEFINITION 8.3. A subset C of Sf which has the. property that whenever
{/J e,C and {gt} ^ {/J, then necessarily also {gt} e C, is called a partition
ideal

In the notation of lattice theory the sets C would be called "semi-ideals"

or "order ideals."

DEFINITION 8.4. We define a function a which associates with each element

of 5f the number that is actually being partitioned: crd/J) = X ^ i / V -

From the lattice-theoretic point of view, a is a positive valuation on Sf

in that (7({/J n {9i}) + a({/J u {9i}) = d K / J ) + a{{gi})9 and if {/J >
{gt}9 then ( T ( { / J ) > a({g J).

DEFINITION 8.5. We say two partition ideals CA and C 2 are equivalent:

Cl - C 2 if p(Cj, n) = p(C 2 , n) for all n.

After this inundation of definitions let us slow down a little and see how
these definitions actually fit in with the results obtained in Chapter 7. First, let

0 = {{/.} 6 ^ | / i > 0 implies i odd}, (8.2.1)

^ = {{/J^I/,^ 1}. (8.2.2)

Then Euler's theorem (Corollary 1.2) asserts that

(9 ~ ® (8.2.3)

Cambridge Books Online © Cambridge University Press, 2010

8-2 Foundations 123

since p(0, n) is the number of partitions of n with odd parts and p(@, n) is
the number of partitions of n with distinct parts.
Next, let

/.+ 1 ^ k - 1 , / , ^ a - 1}, (8.2.4)

= {{/«} e ^|/ t - > 0 implies i ^ 0, ± a(mod 2k + 1)}. (8.2.5)

Then Gordon's generalization of the Rogers-Ramanujan identities is imme-

diately seen to be equivalent to the assertion that

^k,a ~ @k,a for 1 ^ a ^ k. (8.2.6)

Finally, if we define

/ , > 1 implies i c\cn,fl + f2 ^ a - 1} (8.2.7)

and

^*.a = {{/,} e ^l/f > 0 implies i ^ 2(mod 4) and

i ^ 0, ± (2a - l)(mod 4k)}, (8.2.8)

then the generalization of the Gollnitz-Gordon identities given in Theorem

7.11 asserts that

Vk,a - ®k,a for 1 ^ a ^ k. (8.2.9)

That <stfkta, @kta, ^kta, @k%a are all partition ideals is immediate from their
definitions; however, it requires a moment's calculation to see that truly
P(^k,a> n) = £fc,a(n)(andp(^fc,a> n) = Dka(n)). To aid the reader in this matter,
we observe that if (b lb2' • * b3) is a partition ofn with bj ^ bJ+iybj — bj+k-l ^ 2,
then such partitions are characterized by the fact that for any /, the total
number of appearances of i and i 4- 1 (namely, / , + fi+i) in that partition is
at most k — 1; otherwise, if bj is chosen as the first i, then bj — bj+k-l ^ 1.
Since the equivalence relation in Definition 8.5 can be used to describe the
results in Chapter 7 as well as Euler's theorem (Corollary 1.2), we are led to
the following question:

Fundamental Problem. Fully characterize the equivalence classes of

partition ideals.

This problem is phrased intentionally in a vague manner. A minimally

satisfactory answer would be some sort of readily usable algorithm for
determining whether two partition ideals are equivalent. We shall, unfor-

Cambridge Books Online © Cambridge University Press, 2010

124 A General Theory of Partition Identities Chap. 8.3

tunately, be unable to scratch the surface of this problem; however, we shall

consider a refinement in the next section, and for this simpler problem we
shall obtain an adequate solution (Theorem 8.4).

8.3 Partition Ideals of Order 1

The last section was based on the observation that a certain order of
partitions is inherent in all the partition problems treated in Chapter 7.
Next we note a certain "local" property involved in these problems: to
determine if {/,} ss/ka (see (8.2.5)) we need only examine the sequence one
term at a time, at each stage checking the condition "/, > 0 implies i =
0, ± a (mod 2k + 1)." However, to determine if {/,} e & k a we must check
consecutive pairs /,-,/,+1 against the condition "/,. +/,-+i < fc — 1" after
having checked the initial condition fl^a — l. The following definition
makes precise the notion of how narrowly an examination can be focused on
{/J e ^ t o determine whether {/J e C.
DEFINITION 8.6. We shall say that a partition ideal C has order k if k is the
least positive integer such that whenever {/,} £ C, then there exists m such
that {//} $ C where

fi for i = m, m + 1 , . . . , m + k — 1,
otherwise.

For the partition ideals considered in Section 8.2 we observe immediately

that 0, 9, s/kai and Vk>a are all of order 1, while ^ kffl is of order 2 and ®k%a
is of order 3. Inasmuch as Euler's theorem (Corollary 1.2) asserts the equiv-
alence of two partition ideals of order 1, it is reasonable to ask the following:
Problem. Fully characterize the equivalence classes of partition ideals of
order 1.
We shall complete this section by treating this problem in quite a satisfactory
manner.
THEOREM 8.1. Let Cbea partition ideal. Then C is of order 1 if and only if
there exists du d2, d3i... such that

^di for alii} (8.3.1)

where each dt is a nonnegative integer or + oo.

Remark. Once Theorem 8.1 is proved we note that each di is determined
by dj = sup {/ , }eC /,,
Proof. We let C denote the set of partitions on the right-hand side of
(8.3.1). Clearly any such set is a partition ideal of order 1 since it is clearly a

Cambridge Books Online © Cambridge University Press, 2010

8.3 Partition Ideals of Order 1 125

partition ideal and if {/J £ C, then there exists j such that fi > dj for some
j and hence {/,'} £ C where// = 0 if i # j a n d / / = fy
On the other hand, suppose C is of order 1; then define dj = sup{/4}€C/y.
With this definition of C it is clear that C c C. Conversely, if {/J £ C, then
there exists j such that {//} £ C where / / = 0 if / # y and / / = /,-. I claim
that fj > dj9 for if not, by the very definition of dj we see that there must
exist {h(} e C such that/,- ^ hj ^ dj and thus {//} ^ {/ij e C, which would
imply that {//} e C, an impossibility. Thus since fj > djy we see that {/,} £ C,
and therefore C = C, as desired. •

Our next result shows that the partition ideals of order 1 have a large
amount of algebraic structure:
THEOREM 8.2. The ideals of &* are just the partition ideals of order 1.
Proof. Recall the meet and join operations in Sf\ {/J n {gt} =
and {/J u {g(} = {max(/,, #,)}, and recall that an ideal S of ^ is merely
a partition ideal that is closed under the join operation.
Let us therefore assume that S is an ideal of Sf. Thus 5 is a partition ideal.
Define dj = sup^,^/,-, and let

S = { { / J 6 se\ft < dt for all /}.

Clearly S c 5. If now {/J e S, we may define {//-0} by ftU) = 0 if i ^ 7,

//^) = / y and we note that {/J = {/,.(1)} u {/, (2) } u • • • u {//">} where N is
chosen so that f{ = 0 for all i > N. Next we note that for each j there exists
{ht} e S such that fj ^ hj ^ c/y by the definition of the dj. Consequently,
{fiU)} < {*i} e S and so {/,w>} e 5 for each j . Therefore, {/J = {/,(1)} u
{// 2) } u • • • u {/f(N)} is in S since S is an ideal of &. Hence 5 = S and so by
Theorem 8.1, S is a partition ideal of order 1.
The reverse implication is trivial. For if C is a partition ideal of order 1,
then by Theorem 8.1,
C = {{/,} e sr\ft < df for all 1}.
Hence, if {/J e C and {^J e C, then so is {/,} u {g(} = {maxC/i, gt)}9 and so
C is an ideal of £?. •

Our next result provides a very usable representation of the generating

function for P(C, n) whenever C is a partition ideal of order 1.
THEOREM 8.3. Let C be a partition ideal of order 1 with dj = sup {/j}€C /y.
Then

Cambridge Books Online © Cambridge University Press, 2010

126 A General Theory of Partition Identities Chap. 8.3

Proof. By Theorem 8.1, p(C, n) is precisely the number of partitions of n

in which each integer i appears as a part at most dt times. Hence (as in the
proof of Theorem 1.1)

, n)qn = n C1 + <{' + 12J + ' ' ' +

d^< 00

00

_no+ ,' + <« + ,"

d^= 00

oo ^1 ~j{d,+ l)\ -r.

=n (1 - qJ) f.\ d-qJ)

dj< oo dj= ao

n °° M /7^di+1h

This last result allows us to provide a relatively simple test for determining
the equivalence of two partition ideals of order 1, and in this way we have
what might be termed a "minimal" solution to the problem posed at the
beginning of this section.
THEOREM 8.4. Let C and C be partition ideals of order 1 with dj = sup {/l}eC /y
and d/ = s\xp{fi)eCfj. Then C ~ C if and only if the two sequences of
positive integers {j(dj + 1)}JL t and {j(d/ + 1)}JL t are merely reorderings
dj< ao dj<oo
of each other.
Remark. This theorem may be rephrased in terms of multisets (a topic
treated in Section 3.4): A multiset M is a pair ( S , / ) where 5 is a set a n d / is a
function from 5 into the positive integers. For each a e S, f(a) is termed the
"multiplicity of a in Af." Now each of the sequences {j(dj 4- l)}J°=i and
dj< 00
{j(d/ + 1)}JL t uniquely defines a corresponding multiset D (resp. £>') where,
d;<oo
say, D = (A, S) with A the set of positive integers of the form j(dj 4- 1) for
some; and S(r) the multiplicity function which counts the number of appear-
ances of r in the sequence {j(dj + l)}JL t . Theorem 8.4 merely asserts that
d;<oo
C - C if and only if D = D'.
Proof. If the two sequences in question are merely reorderings of each other,
then by Theorem 8.3

Cambridge Books Online © Cambridge University Press, 2010

8.3 Partition Ideals of Order 1 127

= X P(C, n)q\

Hence p(C, n) = p(C\ n) for all n and so C - C.

Suppose the two sequences in question are not merely reorderings of each
other. Let S(r) (resp. S'(r)) denote the number of times r appears in
{j(dj + 1)}J°=1 (resp. {j(d/ + 1)}J°=1 ), and let h be the least integer such
dj< oo dj'< co
that 5(h) ^ 8(W). Now let us suppose C - C . Hence

= £ p(C, n)q"

Therefore

and this is impossible since the coefficient of qh on the left-hand side is

— S(h), which does not equal — 6\h), the coefficient of qh on the right-hand
side. This contradiction shows that C * C. •

A myriad of partition theorems follow easily from Theorem 8.4. We

record a few next.
COROLLARY 8.5. Let Px{n) denote the number of partitions of n in which
each integer i appears at most i — 1 times. Let P2(n) denote the number of
partitions of n into nonsquares. Then P^n) = P2(n)for all n.

Proof. Let Cx = {{/,} e * U < * - 1} and C 2 = {{/,} e ^\ft > 0 - i

not a perfect square}. Then the assertion P{(n) = P2(n) is just C\ ~ C 2 . For
Cx the associated {j(dj + l)}j°=1 is {j(j - 1 + 1)}J°=1 = {j2}?=l, while for
C 2 it is also {;2(0 + 1)}^ =1 = {J°2},°°=1. Hence Ci - C 2 by Theorem 8.4. •

8.6. Let Ml and M2 be two sets of positive integers. Let 2Aft

COROLLARY
denote the doubles of elements of M{ (i.e., 2M{ — {j\(j/2) e Mx}). Then the
number of partitions of n into distinct parts taken from M{ always equals

Cambridge Books Online © Cambridge University Press, 2010

128 A General Theory of Partition Identities Chap. 8.4

the number of partitions of n into parts taken from M2 if and only if

2Ml c Mx and M2 = M t - 2Mt.
Proof. Let Cx = {{/,} e 9f\fi ^ 1 for all i,f{ = 1 implies i e M t } ; and let
C
2 = {{/»} e &\fi > 0 implies ft e M 2 }. Then we are to show that Cx ~ C2
if and only if2Ml £ M1 and M2=Ml- 2MV The sequence {;(</; + 1)};°=1
dj< 00

associated with C\ consists of 0" • 2}JL t together with {j}JLl9 while that
associated with C 2 consists of {j}JL x , and these two multisets are identical
if and only if 2Mt n M t c = 0 (otherwise a second-order multiplicity would
occur in C t 's multiset while only first-order multiplicities occur in C 2 's) and
2M! u M! c = M2C. Hence by Theorem 8.4 C t - C 2 if and only if 2MV c M t
and M2 = Mt - 2MV •
COROLLARY 8.7. Every integer is uniquely the sum of distinct powers of 2.
Proof. Let M t = {1, 2, 4, 8, 16, 32,...} and let M 2 = {1}. Then 2Mt =
{2, 4, 8, 16,...} c M x , and M2 = Mx - 2MX. Hence by Corollary 8.6 the
number of partitions of n into distinct parts taken from Mt equals the number
of partitions of n into parts taken from M 2 , which is exactly 1, since there is
obviously a unique partition of n that uses only ones. •

8.4 Linked Partition Ideais

The last section provided a nice guide for us concerning how we would
like to treat the entire question of equivalence classes of partition ideals.
Unfortunately the power of a result like Theorem 8.3 is unavailable to us when
the partition ideal C is not of order 1. In this section we shall at least show
that a wide class of partition ideals (linked partition ideals) have two-variable
generating functions that are the solutions of finite linear homogeneous
^-difference equations with polynomial coefficients. The class of linked
partition ideals includes both the &kk and @kk of Chapter 7, as well as most
of the partition ideals that appear in known partition identities. It includes
partition ideals C of order 1 if and only if the sequence of local suprema
{^i)T=i IS Periodic modulo k for some k and dt ^ oo for any i; however,
partition ideals of order 1 have been adequately treated in Section 8.3, so that
the important applications of linked partition ideals occur when the ideal in
question is of order 2 or more.
In this section we shall consider a two-variable generating function for each
partition ideal C:
Mx;q)= X x»'qIfti
{fi)eC

Cambridge Books Online © Cambridge University Press, 2010

8.4 Linked Partition Ideals 129

where the exponent on x is the number of parts of the partition ([fl2f23f3- • •)

and the exponent of q is the number being partitioned. Note that

fdUq) = Z qa{{fi)) = f P(C, n)q\ (8.4.2)

{fi}eC n=0

Furthermore, it is not difficult to show (by the means used in Theorem 1.1)
that

n= 1

a product absolutely convergent for \q\ < 1, |x| < \q\~l. Consequently, since
the number of partitions of n with m parts in an arbitrary partition ideal C
clearly does not exceed the total of such in y , we see by the comparison test
(applying (8.4.3) to (8.4.1)) that the series in (8.4.1) is absolutely convergent
for |*| < 1, |JC| < |<*r l .
Q-difference equations (like (7.1.1) and (7.2.4)) were very important in
Chapter 7. Now we must clear away the specific nature of the problems to
determine general qualities of partition ideals that allow ^-difference equations
to arise. The first important property is the existence of a "modulus."

DEFINITION 8.7. Let C be a partition ideal; we define C(m) = {{/,} e C\fx =

Thus C(m) is the set of partitions in C that have all their parts larger than m.
(1)
DEFINITION 8.8. We denote by 0 the bijection </>: Sf -+ y given by
4>({fi)) = {Qi) where
fO if i = 1,
9i
l/i-i if / > 1.
Intuitively the function c/> merely adds 1 to each part of the partition, since

i);
i=2 i=l

thus where initially i appeared/, times now i + 1 appears jx times.

DEFINITION 8.9. We shall say that a partition ideal C has modulus m if mis
a positive integer such that </>mC = C(m). A modulus is not necessarily
unique; indeed, any integral multiple of a modulus for C is also one.

Many partition ideals have no modulus (e.g., the C{ and C2 considered

in Section 8.3 in the proof of Corollary 8.5). The partition ideal C in (8.2.1)
has modulus 2; £? in (8.2.2) has modulus 1; ^ka in (8.2.4) has modulus 1;

Cambridge Books Online © Cambridge University Press, 2010

130 A General Theory of Partition Identities Chap. 8.4

s/ka in (8.2.5) has modulus 2k + 1; @ktO in (8.2.7) has modulus 2, and ^ktU
in (8.2.8) has modulus 4/c.
The relationship between the modulus and ^-difference equations is made
explicit by the following:
LEMMA 8.8. / / the partition ideal C has modulus m, then fc(xqm; q) =
fcfnix;q).
Proof.
fc(xqm;q)= Y

ifi)eC

= Y x#({/aV({/<>)

DEFINITION 8.10. We define, for each partition ideal C of modulus m,

Lc = {{ft} e C|/j = 0 for i > m}.

Thus Z c is just a subpartition ideal of C consisting of all partitions in C
whose parts do not exceed m. Note that Lc depends on which modulus m is
chosen.
DEFINITION 8.11. For each {/J and {gt} in S?9 define {/J © {# J = {/, + ^J.
LEMMA 8.9. / / C is a partition ideal of modulus m, then for each n e C
there exists a unique sequence nu n29 n3, n4,. . . of elements of Lc such that

n = nt ® (<t>mn2) © (02m7i3) © {^mn4) ©• • • . (8.4.4)

Proof. We note that since C has modulus m, (j)jmC = C°'m) for each 7 =
1, 2, 3 , . . . . (This is because </>: C -• C(m) is a bijection and consequently
</>C(m) must contain precisely the elements {/,} of C with fl — f2 = • • • =
/ 2 m = 0; that is, <£mC(m) = C (2m) and so 0 2m C = (/>m0mC = <j)mC{m) = C (2m) ;
the proof for arbitrary j follows similar lines.)
Now for any n = {/J, we have uniquely

n = {/!,/ 2 ,...,/ m , 0 , 0 , 0 , . . . } + { 0 , 0 , . . . , 0 , / m + 1 ,

fm + 2,. • 'Jim, 0, 0,. . .} + {0, 0,. . ., 0 , / 2 m + 1 ,

/ 2 m + 2»- • ->/3m> 0 , 0 , . . . } + " '

/>m7T2) e (</>2m7r3) e - • •

Cambridge Books Online © Cambridge University Press, 2010

8.4 Linked Partition Ideals 131

where nj = {g^Li and

_ \fjm+i for / = 1 , 2 , . . . , m,
9i
~ [0 for i > m.

To conclude our lemma, we must show that when C has modulus m and
n e C, then each KJ G LC. Clearly, (j)u~l)mnj ^neC; hence </>(y~ I)m7r> e C.
u l)m (O 1)m)
But 4) - C = C " . Therefore since <j> - nj e c{{i~Um) = ^ " 1 ) m C ,
iJ l)m

we see that 7Ty e C, and so 7ry e L c . •

DEFINITION 8.12. We say that C is a linked partition ideal if
(i) C has a modulus, say m;
(ii) the L c corresponding to ra is a finite set;
(iii) for each na e L c there corresponds a subset ^ c C O of Lc (called the
linking set of na) and a positive integer l(na) (called the span of na) such that
for any n e V, 7r e C if and only if the representation (8.4.4) has the property
that for each;, nJ+l,nJ + 2,. .., nj + l{nj)_ { = {0,0,0,0,. . . } and nj + l(nj)e£>c{nj).

Since the definition of a linked partition ideal is somewhat complicated,

let us consider some examples. From (8.2.4) let us consider

/ f + 1 < fc - 1}.

Here ^ has modulus 1, and if we let nt = {/, 0, 0, 0 , . . . } , then Ls =

{7r0, ni, 7r 2 ,..., TTfc-i}, /(7r,) = 1 for 0 ^ / ^ k — I and
^ ( T T . ) = {n0, 7rt rrfc_,_x} for 0 < / < fc - 1.

From (8.2.2), we see that Q has modulus 1, and if n{ = {/, 0, 0, 0,. . . } ,

then L = {n0, 7^}, l(n0) = /(TT!) = I and

From (8.2.7), we see that Qk%k = @* has modulus 2, and if TT, = {0, /, 0 , 0 , . . . } ,
ti = {1, /, 0 , 0 , . . . } , then L ^ = {n0, n{,..., nk_ly \j/0, i//^. .., ij/k_2}J(ni) =
Wd = / ( ^ - i ) = 1 for i = 0, 1 , . . . , k - 2 , ^ 0 A , . ) = tfu.{Ki) = {^o, * ! , . . . ,

If the linking set ^cC71) = l^o) an<^ l(n) = r » t n e n there actually is a second
choice of span and linking set for 7r, namely, ^c{n) = ^-c an( ^ K^) = r 4- 1;
generally the first choice is best in order to keep the spans as small as possible.
We remark that dSka for 1 ^ a < k and i/k a for 1 ^ a < k are not linked
partition ideals; however, they are importantly related to J#kk and £/ktk,
respectively, as we now make clear.
DEFINITION 8.13. If C is a linked partition ideal and n e L c , define Cn as the
subset of C consisting of all n in C for which the representation (8.4.4) has
7T, = n.

Cambridge Books Online © Cambridge University Press, 2010

132 A General Theory of Partition Identities Chap. 8.4

Thus for 1 ^ i < k

As we shall see, the Cn are very important in the main theorem on linked
partition ideals (Theorem 8.11). Crucial to the proof of Theorem 8.11 will
be the following result, which is an adaptation to ^-difference equations of
an algorithm of F. J.Murray and K. S. Miller (1954) for the corresponding
problem in differential equations.

LEMMA 8.10. Suppose we have a system of q-difference equations

t x), l^j^n, (8.4.5),

valid for \q\ < 1, |JC| < l^l"1, where the )'j(x) are analytic in x and q inside
this domain and the pjk(x) are rational functions in x and q. Then there
exists r ^ n such that

£ pMy,(xqkm) = 0, (8.4.6)
*=0

where the pk(x) are rational functions of x and q and pr(x) = 1.

Proof. Throughout this proof we shall be replacing the rational functions
pjk(x) by new rational functions, say p*k(x); rather than constantly change
notation (and wind up with p*k**\x)), w e shall write the new functions as
pjk(x) also. The reader must be warned, then, that the pjk(x) in one line may
not be the same as the pjk(x) in the next line.
We begin by considering (8.4.5)j.

yMqm) = PnWyM + (PuWyiW + • • • + Pi*(x)yn(x)).

Either p12(*)>. •., />«„(*) a r e aU identically 0 or not. If they are, we have
(8.4.6) with r = 1. If they are not identically 0, we may (through a suitable
renumbering of the y2(x)>- •., yn(x)) assume then that pl2(x) is not identically 0.
Define

vv2(*) = Pi2(x)y2(x) + • • • + pln(x)yn(x) (8.4.7)

and now we may eliminate y2(x) from our system of equations via

(w2(x) - pl3(x)y3(x) - • • • - pin(x)yn(x))lpl2{x). (8.4.8)

Cambridge Books Online © Cambridge University Press, 2010

8.4 Linked Partition Ideals 133

Our new system is

w 2 (x)

=Z

fc = 2

=1*21 \fc=2

x (w2(x) - p n (x)y 3 (x) PiwW^n(x)), (8.4.9)!

and following our initially stated convention we write

+ • * * + P2nW^(x). (8.4.9)2

After substituting (8.4.8) into (8.4.5)y for 3 ^ j < n, we see that the resulting
equations are again of the same form, namely

+ Z Py*W^*). 3 ^ 7 ^ M. (8.4.9)y

The foregoing procedure can be repeated; this time we examine in (8.4.9)2

w 3 « = P2*(x)y3(x) + • • • + P2fl(x)>;n(x). (8.4.10)

If the p2i{x),..., p2nW a r e a ^ identically zero, then (8.4.9)x and (8.4.9)2 imply

) - Pn(xqm)yi(xqm) = p21(x)>;1(x) + p22(x)(yi(xqm)

which is (8.4.6) with r = 2, and we would be done. If not all the p 2 3 (x),..., p2n(x)
are identically zero, then the elimination )>3(x) (in the same manner in which
y2(x) was eliminated) yields

Cambridge Books Online © Cambridge University Press, 2010

134 A General Theory of Partition Identities Chap. 8.4

w2(xqm) = p2i(*).Vi(*) + P22(x)w2(x) + w3(x),

w3{xqm) = p31(x)yl(x) + p 32 Ww 2 (x)

n
+ Z P3*(*).Hfc(x),

x) + P;2(x)w2(x)

We may repeat this process until either all of the y2(x), y3(x),..., yn(x) have
been eliminated or until one of the resulting

has PJJ+I(X)9. .., pjn(x) all identically zero. In any event, there exists r ^ n
such that

yi(xqm) = PnWy^x) + w2(x),

w2(x) + w3(x),

The first of these equations may be used to eliminate vv2(x) from the system.
Hence

yi(xqm) = p2l(x)yi(x) + p22(x)(yi(xqm)

- Pn(x)yi(x)) + w3(x);

W 4 (X),

+ Pr3(x)w3(x) + • • • + prr(x)wr(x).

The next step produces

Cambridge Books Online © Cambridge University Press, 2010

8.4 Linked Partition Ideals 135

3m
yi(xq ) + 0L(x)yi(xq2m) + P(x)yi(xqm) + y(x)^(x) = w4(x),

and so on. Finally the elimination of vvr(x) yields (8.4.6) as desired. •

THEOREM 8.11. Let C be a linked partition ideal, then fc(x; q) =
Z(/<}ec xIfiqIfii satisfies a finite linear homogeneous q-difference equation
with coefficients that are polynomials in x and q.
Proof. We begin by numbering the elements of L c , say Lc = {no,nl,K2,. • .,7**},
and since for all C, {0, 0, 0 , . . . } e L c , we make the convention that n0 =
{0, 0, 0 , . . . } . We should quickly remark that the 7T, here are not necessarily
the same as the n-^ used in treating the examples considered prior to this
theorem.
Let us define for 0 < / < fc

//,(*) = //,(*, q) = I x* ( 7 V ( 7 r ) . (8.4.11)

Now since n0 = {0, 0, 0 , . . . } , it follows immediately that Cno = C(m) = (j)mC.

Hence by Lemma 8.8

H0(x) = X jc*(«y<*>
7teC
«o

= X x # ( V ( n ) =fc(xqm',q)- (8.4.12)

Therefore we need only prove that H0(x) satisfies a linear homogeneous

^-difference equation with polynomial coefficients, since then (8.4.12) will
automatically imply the same for/ c (x; q).
Our first goal in this proof is to show that the Ht(x) satisfy a system of
linear homogeneous ^-difference equations. To this end we note that if
n e Cni, then by part (iii) of Definition 8.11

n = nt ® n0 ® nQ ©• • • © n0 ® 0 /(7r ' ) 'V

l(ni)- 1 times

where

with S£c{n^ = {njl9 nj2, nh>..., nJr}. Obviously the C ^ , C ^ , . . . , C ^ are

disjoint sets of partitions. Hence for 0 ^ i ^ k

H((x) = X x # ( J V ( 7 °

n'eCn u
• ' l

Cambridge Books Online © Cambridge University Press, 2010

136 A General Theory of Partition Identities Chap. 8

i) y y x#(n) /
h=l n'eCn,

= x#(nt)qo(m) £ HJh(xqKit<)m). (8.4.13)

To conclude our theorem we need only show that our system of ^-difference
equations can be reduced to one higher-order equation in H0(x)9 and this
requires the utilization of Lemma 8.10.
Define

hu(x) = / / , ( x - V j m ) , 0 ^ i ^ K 0 < j < /(TC,). (8.4.14)

The system (8.4.13) is easily seen to be equivalent to the following system of

Q = /(TT0) 4- l(nt) + • • • + l(nk) equations:

hij(xqm) = hiJ+i(x)9 0 ^ i ^ k9 0 ^ j < /(TT,) - 1,

*!.K«,)-i(*O = x-0int)q9in>)'mlini)0iKt) t K.oM- ( 8 - 4 - 15 )

Thus by Lemma 8.10, hoo(x) = H^x'1) satisfies

^ f c ( x ) / / o ( x - V 1 = 0;
fc = 0

hence

0 = £ pk(x-lq-nm)H0(xqW-k)my, (8.4.16)
k=0

furthermore, we may assume that the pk(x~lq~Qm) are polynomials in x and

q since we may clear denominators by multiplication of (8.4.16) by a suitably
chosen polynomial. By the remark following Eq. (8.4.12) we see that the
establishment (8.4.16) is sufficient to prove Theorem 8.11. •

Examples

1. For each k ^ 2, every integer has a unique representation to the base

k\ that is, every integer is uniquely representable as a sum of powers of k
where no part is repeated more than k — 1 times.
2. A perfect partition of H is a partition n = (Xu..., Xs) such that for
each m ^ n there is exactly one n' ^ n such that n' is a partition of m. The
number of perfect partitions of n is the same as the number of ordered
factorizations of ft + 1. To see this, we remark that all perfect partitions of n
may be written

Cambridge Books Online © Cambridge University Press, 2010

Notes 137

hence

* "9k ~
3. MacMahon, who introduced perfect partitions, generalized the idea to
a partition of infinity, by which he meant the formal expression
1 3 1
** = ( I * " V ' " ^)' " " ' '(9x92" '9k-i)°*-1- 0
defined for each sequence {guT=i of integers each larger than 1. From
Example 2 it is obvious that for each integer n there is a unique n with n < n^
such that 7i is a partition of n. A moment's reflection shows that if Jl is the
equivalence class of ideals of order 1 that contains {{/,}!/, = 0 if / > 1},
then
${J(\ n) = 1 for all n,
and each "partition of infinity" is essentially a partition ideal in Jt.
4. The constant sequences {#,-},* i with gt = k ^ 2 produce the partitions
of infinity (or elements of J() that occur in Example 1.
5. It is a simple matter to show that every integer is uniquely representable
as a sum of nonconsecutive Fibonacci numbers (the Fibonacci numbers are
1, 2, 3, 5, 8, 13, 2 1 , . . . , un{= un_{ + u n _ 2 ),. • •)• This shows that not every
element of Jt corresponds to a partition of infinity.
6. Let $(r\ b{, b2,..., bm\m) denote the set of all partitions UY/*2''''*s)
(s arbitrary) such that /,_,. — lx ^ b} whenever r < / ^ 5 and A, = j ( m o d m ) .
Then Q\r\ bx,..., bm\ m) is a linked partition ideal.
7. We let £* denote the set of all partitions U ^ ' ' *^s) (s arbitrary) such
that /.,_! - /.,. ^ 9, (1 < i ^ s) if kt odd, ;.f_2 - kt ^ 5, (2 < / ^ 5) if
kt = 2 (mod 4), and / , _ 2 - ^-« > 0 (2 < / ^ 5) always. Then S is a linked
partition ideal. A modulus of & is 4. The set L# has nine elements:
7ro = { 0 , 0 , 0 , 0 , 0 , . . . } , TTJ = {1,0,0,0,0,...}, TT2 = { 0 , 0 , 1 , 0 , 0 , . . . } ,
7T3 = { 0 , 1 , 1 , 0 , 0 , . . . } , 7T4= {0,1,0,1,0,...}, 7T5 = { 0 , 2 , 0 , 0 , 0 , . . . } ,
7T6= { 0 , 1 , 0 , 0 , 0 , . . . } , 7T7 = {0,0,0, 1,0,...}, 7T8 = { 0 , 0 , 0 , 2 , 0 , . . . } .
The spans of n{, n2, n3 are 2, while the remaining 7r, have span 1. The linking
sets a r e ^ ( T T 0 ) = L*\ 2£nx) = Ls - {n{}; J?#(n2) = {TT7, KS} =
) = {n2, 7T7, 7T8} =

Notes
The material presented here originally appeared in Andrews (1972, 1974,
1975). The presentation of linked partition ideals is expanded and improved
over that in Andrews (1974) and was first exposed in this form in lectures
at the University of Erlangen in 1975. The pair of sets (M x , M2) in Corollary

Cambridge Books Online © Cambridge University Press, 2010

138 A General Theory of Partition Identities Chap. 8

8.6 is called an Euler pair; this result was originally given in Andrews (1969b),
and was extended by Subbarao (1971). Reviews of work related to the material
in this chapter occur in LeVeque (1974), Section P68.
Example 1. This is just the ancient radix representation theorem (see
Andrews, 1969a, b for this treatment).
Examples 2-4. MacMahon (1886, 1891, 1923).
Example 5. This is Zeckendorf s theorem. See Daykin (1960) for a more
extensive treatment.

References
Andrews, G. E. (1969a). "On radix representation and the Euclidean algorithm," Amer.
Math. Monthly 76, 66-68.
Andrews, G. E. (1969b). "Two theorems of Euler and a general partition theorem,"
Proc. Amer. Math. Soc. 20, 499-502.
Andrews, G. E. (1971). Number Theory. Saunders, Philadelphia.
Andrews, G. E. (1972). "Partition identities," Advances in Math. 9, 10-51.
Andrews, G. E. (1974). "A general theory of identities of the Rogers-Ramanujan type,"
Bull. Amer. Math. Soc. 80, 1033-1052.
Andrews, G. E. (1975). "Problems and prospects for basic hypergeometric functions,"
Theory and Application of Special Functions. (R. Askey, ed.), pp. 191-224. Academic
Press, New York.
Daykin, D. E. (1960). "Representations of natural numbers as sums of generalized
Fibonacci numbers," /. London Math. Soc. 35, 143-160.
LeVeque, W. J. (1974). Reviews in Number Theory, Vol. 4. Amer. Math. Soc, Providence,
R.I.
MacMahon, P. A. (1886). "Certain special partitions of numbers," Quart. J. Math.
OxfordSer.il, 367-373.
MacMahon, P. A. (1891). "The theory of perfect partitions and the compositions of
multipartite numbers," Messenger of Math. 20, 103-119.
MacMahon, P. A. (1915). Combinatory Analysis, Vol. 1. Cambridge Univ. Press, London
and New York (reprinted by Chelsea, New York, 1960).
MacMahon, P. A. (1923). "The partitions of infinity with some arithmetic and algebraic
consequences," Proc Cambridge Phil. Soc. 21, 642-650.
Miller, K. S., and Murray, F. J. (1954). Existence Theorems for Ordinary Differential
Equations. New York Univ. Press, New York.
Subbarao, M. V. (1971). "Partition theorems for Euler pairs," Proc. Amer. Math. Soc.
28, 330-336.

Cambridge Books Online © Cambridge University Press, 2010

CHAPTER 9

Sieve Methods Related to Partitions

9.1 Introduction

Up to this point we have established identities between partition functions

either by exhibiting a bijection between the sets of partitions enumerated
(e.g., Theorems 1.4-1.6) or by proving that the related generating functions
were identical (e.g., Theorems 7.5 and 7.11). In this chapter we shall illustrate
how various types of inclusion-exclusion arguments, or sieves, may be used
to prove many results of interest in the theory of partitions. In Section 9.2,
we shall prove Euler's theorem (Corollary 1.2) and an identity of Sylvester
using the inclusion-exclusion principle. In Section 9.3, we study an unusual
sieve related to Atkin's successive ranks of a partition, and we shall prove a
new partition identity (Theorem 9.12) related to the Aki(n) in Theorem 7.5.
This last sieve introduces a number of tantalizing unanswered questions about
partitions.

9.2 Inclusion-Exclusion
We begin with a reexamination of Corollary 1.2 that will yield a purely
combinatorial treatment of Euler's theorem.
THEOREM 9.1 (see Corollary 1.2). p(@, n) = p(@, n)for all n.
Proof. Recall the main argument in the proof of Corollary 1.2
00 00 00 § 1 y"¥ • " \ 00 1 00

£ p(^, n)qn = Y\ (* + <ln) = 17 "" n

=
II 2^T = Z ^
The key to our proof lies in a combinatorial interpretation of
(1 - q2n)/(l - qn). Namely, we write

E(n) = Z ( - l) r

where the summation is over all representations of n of the form

ENCYCLOPEDIA OF MATHEMATICS and Its Applications, Gian-Carlo Rota (ed.).
2, Geoige E. Andrews, The Theory of Partitions
Copyright © 1976 by Addison-Wesley Publishing Company, Inc., Advanced Book Program.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical photocopying,
recording, or otherwise, without the prior permission of the publisher.

139

Cambridge Books Online © Cambridge University Press, 2010

140 Sieve Methods Related to Partitions Chap. 9.2

n = fl1+'" + flr+fe1+-" + fcJ, a1>-->ar9 b^'-^b,, (9.2.1)

and each at is even. Now each ordinary partition n = (CXC2'' Ct) may be
rearranged to yield several representations (9.2.1). In fact, if m is the number
of different even C,, then the contribution of n to E{n) is precisely

m\ m\ m lm\ fl if m = 0,
y }
2 \m [(1 - l) m = 0 if m > 0.

Consequently n contributes 1 if all parts are odd and 0 otherwise. Therefore

£(n) = p(0, n).

Next we note that since each ax is even in (9.2.1), we may write at = 2a,, and
thus our representation is equivalent to

n = aj -f-***+(xr + a1 + * * * + <xr -+•foj+ * * * -h fc5,

« ! > • • • > arf bx>-->bv (9.2.2)

Now each ordinary partition n = (CXC2' • Ct) may be rearranged to yield

several representations (9.2.2). In fact, if m is the number of parts of n that
occur with repetition, then the contribution of n to E{n) is precisely

if m = 0,
- l) m = 0 if m > 0.

Consequently, n contributes 1 if all parts are distinct and 0 otherwise. Hence

E(n) = p(®, n\

and thus p(0, n) = p(^, n) ( = £(n)), as desired. •

The foregoing argument may be easily extended to numerous other results,

such as Corollary 8.6. Our next result utilizes a truncated inclusion-exclusion
argument.

THEOREM 9.2 (Sylvester)

oo (-l^W^Dd.^2^1) ^

(«)(X4" +I )

Remark. Theorem 9.2 is easy to establish from the results in Chapter 7.

In fact, the left-hand side of (9.2.3) is just J l f l (0; x; q) and by Lemma 7.2,
A,i(0; *; <l) = A,i(0; *<?; <?)• Hence

A,i(0; x; q) = J l t l (0; xqN; q) - J l f l (0; 0; q) = 1 as AT - oo.

Cambridge Books Online © Cambridge University Press, 2010

9.2 Inclusion-Exclusion 141

Proof. We shall examine the left-hand side of (9.2.3) as the generating

function for
C(M,N)= X a.

and our object will be to determine the weighting factor <xn.

The left-hand side of (9.2.3) is

y (— i)nxna • - •
+I
it)
i oo
(9).
(«"+I))«
( ) (x«" n±n(n+l)
(9.2.4)

and the approach used in the combinatorial proof of Eq. (2.2.9) at the end
of Chapter 2 shows that xnqn2'q*n(n+i)l(q)n is the generating function for all
partitions with n parts, all distinct and each > n. Hence

generates representations of integers N as follows:

N = at + a2 + • • • + an + bi + • • • + bs (9.2.5)

where ax > a2 > * * * > an > n, bx ^ • • • ^ bs > n with s arbitrary. Now each
ordinary partition n whose smallest part is m and that contains t different
integers can be broken into representation (9.2.5) in numerous ways merely
by selecting a subset of size n of the t possible choices for the a{. Hence, the
( - l)n factor in the first sum in (9.2.4) means that the total contribution of n
to the first sum is

:
The second sum in (9.2.4) is very much the same as the first except now the
representations of N generated are

N = at + a2 + * • • + an + (n + 1) + bt + • • • + bs (9.2.6)

where at > a2 > * * * > an > n 4- 1, bt ^ • • • ^ bs > n with s arbitrary.

In this instance, each ordinary partition n whose smallest part is m and
that contains t different integers (some or all of these t integers may occur in
n with repetition) can be broken into several representations (9.2.6); namely,
provided that m = n + 1, exactly

Cambridge Books Online © Cambridge University Press, 2010

142 Sieve Methods Related to Partitions Chap. 9.3

m - 1

choices may be made for the at. Hence n contributes only to the term of the
second sum in (9.2.4) with n = m — 1 and its contribution to this term is

t- 1

Hence the total contribution of a partition n with smallest part m is

Hence for
H
C(M,N)= X «*
n

<r(n) = N

we see that the weighting factor a,, is zero if n has a smallest part m. Thus the
only partition that can have a positive weighting factor is the empty partition
of zero whose weighting factor is clearly 1. Therefore

£ ( - l) n xV" ( 3 ' l + 1 ) (l - xq2"^'

n
)oo

= C(0,0)=l. •

9.3 A Sieve for Successive Ranks

As we know, any partition may be represented by a Ferrers graph. For

example, ( 7 7 5 3 3 1 1 1 ) has the representation

In 1944, F. J. Dyson defined the rank of a partition as the largest part minus
the number of parts. Thus the rank of the preceding partition is 7 — 8 = — 1.

Cambridge Books Online © Cambridge University Press, 2010

9.3 A Sieve for Successive Ranks 143

Later, Atkin defined the successive ranks of a partition. One can view the
graphical representation of a partition as a set of nested right angles of nodes.
The partition ( 7 7 5 3 3 1 1 1 ) may be viewed as three such right angles

If 7i is a partition, we define r^n) as the number of nodes in the horizontal

part of the /th right angle in the graph of n minus the number of nodes in the
vertical part of the /th right angle. Thus if n is (7 7 5 3 3 1 1 1), then r^n) =
7 - 8 = - 1, r2(n) = 6 - 4 = 2, r3(n) = 3 - 3 = 0.
Our interest lies in the "oscillation" of the ranks between certain positive
and negative bounds.

DEFINITION 9.1. Let h be the largest integer for which there exists a sequence
Ji <Ji <"'< Jh such that rj^n) > 2k - i - 1, rJ2(n) ^ - (i - 1), rJ3(n) >
2k - i - l,rJ4(n) ^ — (i — 1), and so on. We define h to be the (/c, /)-positive
oscillation of n.
DEFINITION 9.2. Let g be the largest integer for which there exists a sequence
Ji <J2 <"'< Jg such that rjv(n) ^ - (i - 1), rJ2(n) > 2k - i - 1, rh(n) ^
- (i - 1), rj4(n) > 2k - i - 1, and so on. We define g to be the (/c, /)-
negative oscillation of 7r.

If n denotes the partition ( 7 7 5 3 3 1 1 1 ) , then the (1, l)-positive oscillation

of 7u is 2, the (2, l)-negative oscillation of n is 1.
Our interest centers upon partition functions related to these oscillations.

DEFINITION 9.3. Let pkJ(a, b; \i\ N) (resp. mki(a, b; n\ N)) denote the
number of partitions of N with at most b parts, with largest part at most a,
and with (/c, i)-positive (resp. (k, /)-negative) oscillation at least \i.

To these partition functions we associate the related generating functions

Pkti(a9b\fi\q) = I; N)q\ (9.3.1)

i

Mkti(a, b\n\q) = (9.3.2)

We next shall derive recurrences for these functions that will allow us
to identify them with expressions involving Gaussian polynomials. We start

Cambridge Books Online © Cambridge University Press, 2010

144 Sieve Methods Related to Partitions Chap. 9.3

by noting the obvious close relationship between negative and positive

oscillation.

LEMMA 9.3. The (/c, i)-positive oscillation of n is either 1 larger or 1 smaller

than the (/c, i)-negative oscillation of n.

Proof. Let j i < j 2 < • • • < j h be the sequence related to the (/c, /)-positive
oscillation as given in Definition 9.1. Either there does exist aj0 < j x such
that rjo(n) ^ — (i — 1) or there does not. In the first case, we see that the
(fc, /)-negative oscillation is 1 larger and in the second case it is 1 smaller. •

LEMMA 9.4. The following recurrences hold for \i ^ 1:

mki(a9b;fi;N) - mkti(a - \,b;n;N)

- mkti(a, 6 - 1; fi; N) + mki(a - 1, b - \\f*;N)

jmk>i(a - 1 , 6 - l ; j i ; A T - a - b+ 1) if a - b > - (i - 1),

U > - Ub- Up- l ; N - a - & + l) if fl-6^-(i-l);

(9.3.3)

mkti{a - i , 6 - l ; ^ - l ; J V - a - 6 + l) i/ a - 6 > 2ik - i - 1,

, I < 0 - 1 , 6 - l ; | i ; A T - a - 6 + 1) i/ a - 6 < 2fe - i - 1.

(9.3.4)

Proof. Let us start by examining the left-hand side of (9.3.3). The expression
mkti(a, b; \i\ N) — mk ((a — 1, b\ fx; N) denotes the number of partitions of
N with at most 6 parts, with (/c, /)-negative oscillation at least //, and with
largest part exactly equal to a. Therefore, the expression

(mkti(a, b;n\N) - mkti(a - 1, b;/i; N)) - (mM(a, 6 - l;/x;N)

- m*> - 1,6- l;;

(which is the left-hand side of (9.3.3)) denotes the number of partitions of N

with exactly 6 parts, with largest part exactly a, and with (/c, i)-negative
oscillation at least \i.
We now transform these partitions that are enumerated by the left-hand
side of (9.3.3) by deleting the largest part (namely a) and subtracting 1 from
each of the 6 - 1 remaining parts. In terms of the graphical representation,
we have removed the outer right angle of nodes from each partition. The
transformed panitions are partitions of N — a — fc+1 into at most 6 — 1
parts with largest part at most a — 1.

Cambridge Books Online © Cambridge University Press, 2010

9.3 A Sieve for Successive Ranks 145

Suppose first that a — b > — (i — 1); then the removal of the outer right
angle of nodes from the graphical representation has no effect on the (/c, i)-
negative oscillation, which is still equal to fi. Consequently our transformed
partition is of the type enumerated by mki(a — 1, b — 1; \x\ N — a — b + 1).
Since the foregoing procedure is clearly reversible, we see that if a — b >
- (i - 1), then

mk,i(a,b;n',N) - mk>i(a - l,6;/*;n) - mkti(a, b - l;fi;N)

+ m * > - 1 , 6 - Ufi;N)

= m k J ( a - 1 , 6 - l ; n ; N - a - b + 1).

Therefore the top half of (9.3.3) is established.

Now suppose that a — b ^ — (/ — 1); in this case, the removal of the
outer right angle of nodes from the graphical representation may affect the
(k, /)-negative oscillation. In fact, the resulting partition is now characterized
by the fact that it has (/c, /)-positive oscillation at least \i — 1. Thus the
transformed partitions in this case are of the type enumerated by pkti(a — 1,
b — 1; fi — 1; N — a — 6 + 1). Again the transformation used is reversible.
Therefore if a — 6 ^ — (i — 1), then

mkti(a,b; n;N) - mkti(a - l , 6 ; ^ ; i V ) - mkti(a,b- l;/x;iV)

+ mkti(a - 1 , 6 - l;jx;iV)

= pkti(a- 1 , 6 - l ; / x - l ; i V - a - 6 + 1).

Therefore the bottom half of (9.3.3) is established.

We omit the proof of (9.3.4) since it perfectly parallels the proof of (9.3.3)
with the roles of (/c, f)-positive and (/c, /)-negative oscillation interchanged. •

It is now a simple matter to translate Lemma 9.4 into relations among the
generating functions.

COROLLARY 9.5. For each /J. ^ 1,

Mkti(a, 6; ft; q) - Mkti(a - 1, 6; \i\ q)

- Mkti(a, b-Up;q) + Mkti(a - 1, 6 - 1; JI; q)

k^^Ub^U^q) if a-6>-(i-l),
a

~q [Pkti(a - 1 , 6 - l ; / z - l;q) i f a - 6 < - (i - 1);

(9.3.5)

Cambridge Books Online © Cambridge University Press, 2010

146 Sieve Methods Related to Partitions Chap. 9.3

Pkfi(a,b;n;q) - PkJ(a - l,b;n;q)

-Pkti(a,b- 1;M;4) + P * > - 1 , 6 - U/i;q)

kM-1'b-Uii-Uq) if a — fe > 21c — i — 1,

> -l,b-l;ii;q) if a - b ^ 2k - i - 1.
(9.3.6)
N
Proof. Comparing coefficients of g in these identities, we can establish
these results directly from Lemma 9.4. •

LEMMA 9.6. The following relations hold:

PUa, b; 0; q) = M k > , 6; 0; q) = T * , (9.3.7)

and for \i ^ 1

P M (0, 6; /z; «) = P f c > , 0; /i; ^) = Afkfl(0, fc; /z; ^f)

= M f c > , 0 ; / i ; ^ ) = 0, (9.3.8)

w/iere [fla6] is ^/ie Gaussian polynomial discussed in Section 3.2.

Proof. For (9.3.7), we observe that all partitions have (fc, i)-positive and
(fc, i)-negative oscillation at least 0. Hence PkJ(a, b; 0; q) and M M (a, 6; 0; (?)
are each the generating function for partitions with largest part at most a
and with at most b parts. By Theorem 3.1, this generating function is [fl^fc];
thus (9.3.7) is established.
As for (9.3.8), we observe that only the empty partition of 0 has either no
parts or no largest part. Since the empty partition of 0 has 0 for all of its
(fc, i)-oscillations, we see that the generating functions appearing in (9.3.8)
must all be identically 0. •

LEMMA 9.7.

+ B _ n U + B-11 U + B - 2
H q
B-X \B - X- l\ \ B - X - l

= q q \
q 9
[ B - X - l

Proof. By (3.3.3), we see that

B] U + B - l ] IA + B - 1 ] [A + B - 2
\ \ B - X \ \ \

Cambridge Books Online © Cambridge University Press, 2010

9.3 A Sieve for Successive Ranks 147

B-X-2

LEMMA 9.8. For each integer r > 0, let

f(A,r;X,Y,L) = qL(l~q^ ^+ }

Then for each integer r ^ 1,

f(A, r; X, Y, L) - f(A, r - 1; X, Y, L) - f(A - 1, r + 1; X, Y, L)
+ f(A- l,r;X,Y,L)

[ A + Y
Proof.
f{A, r; X, Y,L) - f(A, r-l;X,Y,L)

A+Y
A
y L + A + Y + j+2 + (r-j-2)(A + Y+l) ^ ^
q
h [A+Y-I
Also, by repeated application of (3.3.4), we see that

A+Y
r-3
nj(A+Y) + L + 2A + 2r+r \ *""
A+ Y- 1
[2/4 + XI

r-3 ["2,4 -J- J^

=
y=O^ [ A + Y- 1
2,4 + XI
x + y

Cambridge Books Online © Cambridge University Press, 2010

148 Sieve Methods Related to Partitions Chap. 9.3

r-3 f>A i y i ;
_ y aL + A + Y + j+2 + lr-j-2HA+Y+l) r" 1 T
^ ^ 7
;=o [A+Y-l
2/l + X
, ff
q
[A+ Y

Therefore, combining these two results, we derive that

f(A,r;X,Y,L)-f(A,r- I; A, Y, L) - q \ + X + r- 2
\2A

+ f{A - l,r;X, F,L)

jt0 I A + Y- 2

+ q
[A+Y-]

+ f(A - 1, r; X, Y, L) (by repeated application of (3.3.4))

2A
q
~'jh ' \ A+Y-2

^_ j +f(A _ u r. x> Y L)
1) + X
A-l + Y

+ (1 - qA + Y) [ A - 1+ Y- 1

= f(A- l,r+l;X,Y,L). •

We are now prepared to prove the main theorem on generating functions

that will be essential to our sieve.

THEOREM 9.9. / / b = a or a - 1 and k > i > 0, then

Mki(a, b;2n\q) —

Cambridge Books Online © Cambridge University Press, 2010

9.3 A Sieve for Successive Ranks 149

-«•.„
a + b
•
[b - {2k -M)
(9.3. 10)

Pkti(a, b\ 2n\ q) = (9.3. 11)

[b + {2k +

(9.3. 12)
a - {2k + l)n

Proof. We first note that the recurrence relations (9.3.5) and (9.3.6) together
with the initial conditions (9.3.7) and (9.3.8) uniquely determine Mki(ay b\fi;q)
and Pkti(a< b\ JX\ q). We now define new functions Mki(a,b;fi;q) and
P*t(ch b\ n; q) in terms of Gaussian polynomials. Our object is to identify
these new functions with the generating functions being considered.

a + b
(9.3.13)
b - (2/c +

for a - b ^ - (/ - 1);

Af ;.,{<!, 6 ; 2 j l - l ; ^ ) =
- (2/c

(9.3.14)

fora - fc ^ - (i - 0 ;

a + b
(9.3.15)
a -{2k +

for a — b < 2/c — i;

a + b
(9.3.16)
a - (2/c + l)

for a —fe^ 2fc — /;

Af»*,{ti - l,

=
\ _

2a + j - 3

Cambridge Books Online © Cambridge University Press, 2010

150 Sieve Methods Related to Partitions Chap. 9.3

(1_q(r-i)(.+ i-i-(2»+i)^)r 2a + i - 3 + j

for r ^ — 1, a ^ 1, // > 1;

= (2M

2a + i - 3
X
[a - (2/c + 1)JI + 2fc - 1
r-l

•_3

for r ^ - 1, a S* 1, /i ^ 1;

_ _,.„,... .,M-(2)t-2i+l))V

2b + 2k - i - 2
* [b + 2k - i - 1 - (2k +
r-2

( 1 - q(r->- » ) ( • + » - • - < " + »W) [ 2b + 2 k - i - 2 + j

(9.3.19)

for r ^ 0, 6 ^ 1, \i ^ 1;

P?,,<6 + 2fc - i - 1 + r , 6 - l ;

_ UM-l)((2fc+ D M - Q

26 + 2fc - i - 2
X
[6 + 2A: - 1 - (2Jfc +

Cambridge Books Online © Cambridge University Press, 2010

9.3 A Sieve for Successive Ranks 151

r-2

7=0
j— l)(b + 2k —( 2 k + 1)M)\
2b + 2/c - i - 2
x(1~ (9.3.20)
_ b + 2fc — ( 2 k + l)fi
i- 6 -h 2A: ~ 2 - ( 2 / c

for r ^ 0, b>\. 1.
Finally

(9.3.21)

We first observe that the initial condition (9.3.7) is prescribed in (9.3.21).

Furthermore, if either a or b is set equal to zero in M*,(#, b\ /i; q) or
P* -(a, b; /i; g), with JI ^ 1, then inspection of the appropriate defining
equation among (9.3.13)-<9.3.20) shows that the resulting function is identically
zero. Hence the initial conditions prescribed in our Lemma 9.6 are satisfied
by M*f.(a, b\ /i; q) and P*§j(fl, b;n;q).
We now shall examine the recurrence relations (9.3.5) and (9.3.6). The
top line of (9.3.5) and the bottom line of (9.3.6) follow in each case by applying
Lemma 9.7 to the appropriate expressions among (9.3.13)—(9.3.16). Finally,
in the notation of Lemma 9.8, we see that

Af k V(0- l,a + i - 1 + r;2fi;q)

= f(a9r + l ; i - 1, i - 2 - (2fc + l)/i, (4fc + 2)/x2 + (2fc - 2/

(9.3.22)

for a > 1, r > - 1, /x > 0;

= / ( f l , r + 1; i — 3,2ik — 1 - (2fc + l)/i9(2/i — lX(2fc + l)/i - 2/c 4- i - 1))

(9.3.23)

for a ^ 1, r ^ - 1, \i > 0;

- i - 1 - (2fe 4- / - (2fc - 2/
(9.3.24)

for 1, r > 0, /x > 0;

Cambridge Books Online © Cambridge University Press, 2010

152 Sieve Methods Related to Partitions Chap. 9.3

P* J(b + 2/c — / — 1 + r, b — 1, 2fi — \\q)

= /(ft, r; 2/c - / - 2, 2/c - 1 - (2/c + l)/i, (2/x - l)((2/c + l)/x - /)),

(9.3.25)

for ft > 1, r ^ 0, JI > 0.

It is now simply a matter of inspection to verify that Lemma 9.8 implies that
the bottom line of (9.3.5) and the top line of (9.3.6) hold for Mki(a, b\ \i\ q)
and P*.(fl, b\ii\q) if \i > 0.
Thus we have observed that M*ti(a,b; \i\ q) and P* t(a, b; ft; q) satisfy
(9.3.5), (9.3.6), (9.3.7), and (9.3.8); and since these four equations uniquely
define Mki(a, ft; n; q) and Pkti(a9 ft; \i\ q), we see that

P k J ( a , b\n\q) = P*ti(a, b\ii\q) (9.3.26)

and

Mki(a, ft; [i\ q) = M*.(a, ft; \i\ q). (9.3.27)

Equation (9.3.9) now follows from (9.3.13) and (9.3.27); (9.3.10)follows from
(9.3.14) and (9.3.27); (9.3.11) follows from (9.3.15) and (9.3.25); finally,
(9.3.12) follows from (9.3.16) and (9.3.26). •

DEFINITION 9.4. Let pktfji\ N) (resp. mki(fi\ N) denote the number of

partitions of N with (/c, i)-p°sitive (resp. (/c, i')-negative) oscillation at least /*.

DEFINITION 9.5. Let

THEOREM 9.10. The following relations hold for \q\ < 1.

li > 0; (9.3.28)

0; (9.3.29)

-, fi>0; (9.3.30)

—, fi>0. (9.3.31)
(«).

Cambridge Books Online © Cambridge University Press, 2010

9.3 A Sieve for Successive Ranks 153

Proof, First we observe that for \q\ < 1,

\Pk g{/i; q) — Pki{a, a; ji; q)\ ^ £ p(n)|</|n -• 0 as a -+ oo

and
oo
\Mubi',q)-Mu(a,a',ti;q)\< ^ p(n)\q\n-> 0 as a -> oo

where p(n) is the ordinary partition function. Finally

x Via + y] a*

Consequently (9.3.28) follows from (9.3.9), (9.3.29) from (9.3.10), (9.3.30)

from (9.3.11), and (9.3.31) from (9.3.12). •

Now we introduce the partition function that arises from our sieve technique.
DEFINITION9.6. Let Qkt£n) denote the number of partitions n of n such
that — (i — 2) < rj(n) ^ 2fc — i — I for each of the successive ranks of n.
The following lemma is the inclusion-exclusion aspect of our sieve.
LEMMA 9.11. For each integer n ^ 0,
oo on

Proof. First we remark that Qkti(n) counts the set of all partitions of n
that have 0 as (/c, /)-positivc oscillation and 0 as (fc, /)-negative oscillation.
On the other hand, the right-hand side of (9.3.32) is a weighted count of
the partitions of n. First suppose n is a partition of n with 0 as (k, /)-positive
and (A:, /)-negative oscillation. Then n is counted once by pkti(0\ n) and not
at all by each of mk ^i; n) and pkti(fi\ n) for each /i > 0. Next suppose that
7i is a partition of n with (/c, />positivc oscillation r > 0. By Lemma 9.3,
the {k, i)-negative oscillation of n is cither r — 1 orr + I; if r — 1, then the
weighted count for n is
r-1 r r- I r-1

if r + 1, then the weighted count for n is

Finally, if the (k - /)-positive oscillation of n is 0 and the (/c, /)-negative

oscillation is 1, then the weighted count of n is 1 - 1 = 0.

Cambridge Books Online © Cambridge University Press, 2010

154 Sieve Methods Related to Partitions Chap. 9.3

Thus we see that the right-hand expression in (9.3.32) counts once each
partition of n with 0 as (k, i)-positive and (/c, i')-negative oscillation, while it
counts 0 for each of the other partitions of n. Consequently the right-hand
expression in (9.3.32) is just equal to Qkj(n). •
THEOREM 9.12. Recall that Akt{n) denotes the number of partitions of n
into parts that are not congruent to 0, ± i (mod 2fc + 1). Then for 0 <
i ^ k and for each n ^ 0

Proof. By Lemma 9.11, we see that

00

E&»9"
00 00 00 00 00

= E iU°; W + E (- i)" E »»*.«0«;")<T + E (- ^ E P*>; «)«"

oo oo

= E Af*.^; g) - f Mt>j(2|» - 1; q) + f i**.^;«) - Z P*.X2M ~ 1:«)

- Z ^(2"
00

1
l)l
J
= (^)oo I! (-
n = — oo
ao
i
= (^); n(i-
(by Jacobi's identity, Theorem 2.8)

o-o-1
00

= n
= E ^,i(")9"-

Cambridge Books Online © Cambridge University Press, 2010

9.3 A Sieve for Successive Ranks 155

Hence comparing coefficients of qn in the identities above, we see that

for each n ^ 0.

It is very easy to show that Q2%2(n) = B2f2(n) ar

*d Q2,i(n) — B2ti(n),
then Corollaries 7.6 and 7.7 (the Rogers-Ramanujan identities) follow
directly from Theorem 9.12.
To prove that Q2t2(n) — B2t2(n), l et u s consider a partition n of the type
enumerated by B22{n), say n = (CyC2 • • • Ct), where Cjf — Ci+ x ^ 2. We form
a graphical representation of n as follows. The ith part of n is represented as
the ith right angle in a graphical representation where if C, = 2s + 1, the
angle is

s + 1

s + 1

and if Cx = 2s, the angle is

s 4- 1

Since Cj? — C i + 1 ^ 2, these angles do indeed form the successive angles of a

graphical representation of a partition, and clearly the successive ranks are
either zero or one. Thus a partition of the type enumerated by Q2t2(n) is
produced. The foregoing procedure clearly establishes a one-to-one cor-
respondence between the two types of partitions, and therefore B2 2(n) =
&»•
To prove that B 2 i l («) = Q2,i(n) we follow the same procedure except that
now for parts Ct — 2s + 1, the angle is

s+ 2

Cambridge Books Online © Cambridge University Press, 2010

156 Sieve Methods Related to Partitions Chap. 9

The argument proceeds as before with the only difference being that now an
angle with only one node in it cannot arise, and this is precisely the difference
between B2t2(n) and £ 2>1 (n). Thus B2tl(ri) = Q 2 ,i( n )-
There are several questions of interest that arise naturally from this work.

Question 1. Can a direct graphical proof be found to show that BkJn) =

Qkta(ri) in general?

Question 2. Are there sieves that can be used in studying some or all of the
partition functions found in Chapters 7 and 8?

Question 3. We can directly deduce from Theorem 9.10 that

mkti(2n; n) = p(n - fi((4k + 2)\i + 2k - 2i + 1)),

m M (2/x - 1; n) = p{n - (2fi - l)((2fc + l)fi - 2k + i - 1)),

pkti(2fi; n) = p(n - M(4fe + 2)/i - 2fc + 2i - 1)),

Pkti(2fi -l;n) = p(n- (2/x - l)((2fe + l)/x - 0).

Are there direct combinatorial proofs of these relations?

Examples

1. The inclusion-exclusion method used in the proof of Theorem 9.1 may

be employed to provide combinatorial interpretations and proofs of many
elementary infinite product identities, such as

*=l(l - q ) _ lL,= l ( l - q ) TT n _ nx-l

2. Let se(n) (resp. eo(n)) denote the number of partitions of n into distinct
nonnegative parts with smallest part even and an even (resp. odd) number of
even parts. Then

{
0
1
if n is not a square,
if n is a square.
For example, when n = 9 the five partitions enumerated by ee(9) are 8 + 1 + 0,
7 + 2 + 0, 6 + 3 + 0, 5 + 4 + 0, 4 + 3 + 2, while the six partitions enu-
merated by eo(9) are 9 + 0, 7 + 2, 6 + 2 + 1 + 0, 5 + 4, 5 + 3 + 1 + 0,
4 + 3 + 2 + 0.
The related generating function identity is

11 = 0 m=0

Cambridge Books Online © Cambridge University Press, 2010

Examples 157

This analytic assertion is quite easily proved (e.g., in Corollary 2.3, replace q
by q2, then set c = — </, a = 0. / = </2, and let b -• 0). Actually we can
combinatorially prove each step in the following analytic argument:

n=O

= (<?2;</2)«, V q2n . - L . . . ( - ^ • * ' ; ^ ) x (use Example 1)

B=o (9 ; r ) »

= (</ 2 :1 2 )* y -r—,• y fl
, - -—• (see Example 17, Chapter 2)
( U t ( )

(<?2; « 2 ) . I ~~ r -' -T- (see Example 17, Chapter 2)

= ^ ^m2 (use Example I).

m=O

3. The work in Example 2 can be generalized by means of the identity

n=0

4. The identities (here [x] is the greatest integer function)

=
, - ? . ( ~ DV""*"-™ [ [|(n + , _ 5A)] + flj (a = 0 or 1),
and

"+1 l.i
(n + 1 - 3A)]J
are both deducible from Theorem 9.9. The sieve used in Theorem 9.12 is
now applied directly to the generating functions in Theorem 9.9. The case
in which a + fc = n + 1, it = 2, i = 2 - a essentially yields the top result;
the case in which d + fc = n + l,fc = i = l yields the second result imme-
diately.

Cambridge Books Online © Cambridge University Press, 2010

158 Sieve Methods Related to Partitions Chap. 9

Notes

In the late nineteenth and early twentieth centuries, K. T. Vahlen (1893)

and L. von Schrutka (1916, 1917) undertook an extensive study of partitions
from the point of view taken in Section 9.2. The theorems we have chosen
come from Andrews (1969, 1975).
The work in Section 9.3 also comes from Andrews (1971, 1972a). Such
results were discovered in an effort to obtain a combinatorial interpretation
of Schuf s (1917) analytic proof of the Rogers-Ramanujan-Schur identities.
Reviews of recent work related to this area may be found in Section P68
of LeVeque(1974).
Example 1. Vahlen (1893), Andrews (1969).
Example 2. Andrews (1972b), Stenger (1973).
Example 4. Schur (1917), Andrews (1970, 1971, 1972a).
Question 1. W.H. Burge has found such a proof. See Discrete Math., 34 (1981),
9-15.
Question 3. D.M. Bressoud has given a combinatorial account of these identities
as well as having observed that Theorem 9.12 is also true when k is half an odd
integer. See J. Number Th., 12 (1980), 87-100. Indeed the proof given above
works in this case as well.

References
Andrews, G. E. (1969). "On a calculus of partition functions,'* Pacific J. Math. 31,
555-562.
Andrews, G. E. (1970). "A polynomial identity which implies the Rogers-Ramanujan
identities,'* Scripta Math. 28, 297-305.
Andrews, G. E. (1971). "Sieves for theorems of Euler, Rogers and Ramanujan," The
Theory of Arithmetic Functions (A. A. Gioia and D. L. Goldsmith, eds.) (Lecture
Notes in Math. No. 251), pp. 1-20. Springer, Berlin.
Andrews, G. E. (1972a). "Sieves in the theory of partitions," Amer. J. Math. 94, 1214-
1230.
Andrews, G. E. (1972b). "Problem 5865/* Amer. Math. Monthly 79, 668.
Andrews, G. E. (1975). "Partially ordered sets and the Rogers-Ramanujan identities,"
Aequationes Math. 12, 94-107.
Atkin, A. O. L. (1966). "A note on ranks and conjugacy of partitions," Quart. J.
Math. Oxford Ser. (2) 17, 335-338.
Dyson, F. J. (1944). "Some guesses in the theory of partitions," Eureka (Cambridge)
8, 10-15.
LeVeque, W. J. (1974). Reviews in Number Theory, Vol. 4. Amer. Math. Soc, Providence,
R.I.
Schrutka, L. von (1916). "Zur Systematik der additiven Zahlentheorie," J. Reine Angew.
Math. 146, 245-254.
Schrutka, L. von (1917). "Zur additiven Zahlentheorie,** S.-B. Kaiserl. Akad. Wiss.
Wien, Abt. lla 125, 1081-1163.
Schur, I. J. (1917). "Ein Beitrag zur additiven Zahlentheorie und zur Theoric der Kettcn-
bruchc," S.-B. Preuss. Akad. Wiss. Phys.-Math. Kl. pp. 302-321. (Reprinted in
I. Schur, Gesammelte Abhandlungen, Vol. 2, pp. 117-136. Springer, Berlin, 1973.)
Stenger, A. (1973). "Solution to Problem 5865,** Amer. Math. Monthly 80, 1148.
Vahlen, K. T. (1893). "Beitragc zu cincr additiven Zahlentheorie,** J. Reine Angew. Math.
112, 1-36.

Cambridge Books Online © Cambridge University Press, 2010

CHAPTER 10

Congruence Properties of Partition

Functions

10.1 Introduction

In Chapter 5 we observed what remarkable fruits arose from the Hardy-

Ramanujan collaboration. Out of this effort also arose Ramanujan's amazing
discoveries about divisibility properties of p(n). The following words of
Ramanujan describe his famous conjecture on this topic:

A recent paper by Mr. Hardy and myself contains a table, calculated by

Major MacMahon, of the values of p(n)9 the number of unrestricted parti-
tions of w, for all values of n from 1 to 200. On studying the numbers in
this table I observed a number of curious congruence properties, apparently
satisfied by p(n). Thus

(1) K4), p(9), p(14), p(19), . . . = 0 (mod 5),

(2) J*5), p(12), p(19), P(26), . . . s 0 (mod 7),
(3) P(6), p(17), P(28), K39), . . . = 0 (mod 11),
(4) P(24), P(49), P(74), K99), . . . = 0 (mod 25),
(5) K19), P(54), P(89), K 1 2 4 ) , . . . = 0 (mod 35),
(6) P(47), K96), P(145), K 1 9 4 ) , . . . = 0 (mod 49),
(7) JK39), P(94), P(149), s 0 (mod 55),
(8) K61), p(138), = 0 (mod 77),
(9) p(H6), . . . ss 0 (mod 121),
(10) K99), . . . = 0 (mod 125).
From these data I conjectured the truth of the following theorem:
If 5 = 5fl7fcllc and 24A = 1 (mod <5), then
pOl), p(X + <5), p(k + 25),... = 0 (mod S).
This theorem is supported by all the available evidence; but I have not
yet been able to find a general proof.
Ramanujan then presents in this same paper proofs of the congruences

p(5n + 4) = 0(mod 5), (10.1.1)

ENCYCLOPEDIA OF MATHEMATICS and Its Applications, Gian-Cark) Rota (ed.).

2, George E. Andrews, The Theory of Partitions
Copyright © 1976 by Addison-Wesley Publishing Company, Inc., Advanced Book Program.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical photocopying,
recording, or otherwise, without the prior permission of the publisher.

159

Cambridge Books Online © Cambridge University Press, 2010

160 Congruence Properties of Partition Functions Chap. 10.1

p(7n + 5) = 0(mod7), (10.1.2)

and he sketches proofs of the results:

p(25n + 24) = 0 (mod 25) and p(49n + 47) = 0 (mod 49).

In connection with these congruences, Ramanujan proves a number of

identities related to the partition function, such as

4V = 5n ( 7 r A,^. dO.1.3)

a result considered by G. H. Hardy to be an example of Ramanujan's best

work.
The consideration of Ramanujan's full conjecture on the congruence
properties of p(n) produced a number of partial results, such as those of
Ramanujan described earlier. In the 1930s, S. Chowla noticed (from the
examination of an extended table of values of p(n) due to H. Gupta) that

p(243) = 133978259344888 ^ 0 (mod 7 3 );

however,

24-243 = I(mod7 3 ).

Hence Ramanujan's conjecture as it stands is false. G. N. Watson showed in

1938 that a modified version of the conjecture was true for all powers of 5
and 7, and in 1967 (48 years after Ramanujan originally published his con-
jecture) A. O. L. Atkin proved the full modified conjecture:

If S = 5alb\\c and 24/1 = 1 ( m o d 8\ then p{X) = 0 ( m o d sa7iib + 2)/2]

\lc).

In this chapter, our object is to obtain an understanding of the analytic

methods that have been most effective in attacking such problems. Some form
of the Hecke operator almost always appears in the analytic considerations,
and in Section 10.2, we shall see that such operators suffice to yield
Rodsetrf s proof of the Churchhouse conjecture on binary partitions.
As for Ramanujan's conjecture, we shall content ourselves with the powers
of 5, and we shall present a simplified version of Watson's original proof due
to Atkin. This proof has received an excellent presentation in M. I. Knopp's
book, Modular Functions in Analytic Number Theory (1970); our version
differs primarily in the fact that it is less complete. In particular, properties of
p(n) that rely primarily on the fact that ^ o p(n)e2ninx = enixn\n{i)yl
(where ^(T) is a modular form) would require extensive preliminary work,
which would be much more appropriate in a book on analytic number theory

Cambridge Books Online © Cambridge University Press, 2010

10.2 Rodseth's Theorem for Binary Partitions 161

or elliptic function theory (Knopp's book cited above, e.g.). Thus in Chapter 5
we assumed the truth of (5.2.2), which is a disguised form of the relation
describing the behavior of t](x) under the action of the modular group. In
this chapter we shall assume (10.3.1), the modular equation of fifth degree.
After obtaining Ramanujan's conjecture for powers of 5, we shall conclude
this chapter with a brief discussion of the Dyson conjectures (proved by
Atkin and Swinnerton-Dyer in the 1950s) which provide combinatorial
interpretations of (10.1.1) and (10.1.2).

10.2 Rodseth's Theorem for Binary Partitions

We begin with some elementary remarks about certain operators closely
related to the Hecke operators: These operators Um are defined on all func-
tions f(q) = Y,n= - oc an(in rneromorphic around q = 0 and are given by

Thus if p = e2ni/m, then

UJ(q) = f(pm~ (10.2.1)

since
m if r = 0(mod m\
O if r ^ 0(mod m).

Clearly Um is a linear operator.

The type of partitions we shall be considering in this section are binary
partitions, that is, partitions in which the parts are all powers of 2. Thus the
six binary partitions of 7 are (124), (134), (123), (1322), (I 7 ), (215).
We denote by b(n) the number of binary partitions of M, and we let

n=Q n=0

Next we define a sequence of functions for m ^ 1,

*m(q) = (U2m+l - U2m-l)B(q)

(10.2.2)

First we have

Cambridge Books Online © Cambridge University Press, 2010

162 Congruence Properties of Partition Functions Chap. 10.2

i(<Z4) = ( t V - U2°)B(q*) = £ (b(4n) - b{nW

n=0

3
i r

l
n I 1 [ y (1 + '^x1 + |2 V) 2 | _ A

1 °° 1

_ g*(3 - g*) Qo 1

Hence

^"I(^) = "" 3 G(«) (10.2.3)

where

G(^f) = f l _ 2Wx = (1 - q)B(q).

In general, we have

^m+lG? 2 ) = Ul^miq2) = K^mC^f) + ^mC ~ ^)) (10.2.4)

(by (10.2.1)). Thus

Therefore

j ^ dO.2.5)

Cambridge Books Online © Cambridge University Press, 2010

10.2 Rddseth's Theorem for Binary Partitions 163

We now derive similar formulas for ^3(q) and

2
q) + &2(- q))

Thus

This implies

^ 64qG(q)
(1 - qY
Next, applying U2 to (10.2.7), we find that

76&q2G(q) lO24q2G(q)
24 25

Thus

or

(10.2.8)

It is now possible to prove Theorem 10.1, which generalizes (10.2.5),

(10.2.7), and (10.2.8).

THEOREM 10.1. There exist integers c/jn) such that for m ^ 2

m 2
" 2^ 2

Cambridge Books Online © Cambridge University Press, 2010

164 Congruence Properties of Partition Functions Chap. 10.2

furthermore, co(m) = 1, ^Ic^m) if m ^ 4, 8|c2(m), 16fc2(m), and 22j\Cj(m),

for 3 ^ j' ^ m — 2.
Proo/. From (10.2.5), (10.2.7), and (10.2.8) we see that Theorem 10.1 is
true for m — 2, 3, and 4. We now proceed by mathematical induction on m,
and we assume m ^ 5.
By applying U2 to the equation stated in the theorem (with q replaced by q2)
we find
(m+l\
m-2 9V 2

7=0 11 - 4 ; - - -

Now we define integers (5k(m + 2) for 0 ^ k < [(m + l)/2] by

]
Sk(m + 2)(1 - ^ 2 ) /( ;

this equation truly defines integers 5k(m + 2), since the left-hand side of the
equation is an odd polynomial. Furthermore, we see that

do(m + 2) = 2 m + 1

and

= - m2m~l.

Thus
/m+2>
m-2 ?V 2 >

k=2
m-k- 1

Thus we may derive expressions for the Cj(m + 1):

co(m + 1) = co(m) = 1;

c^m 4- 1) = cx(m) + m2 m " 1 c 0 (m)

= cx(m) + m2m~l.

Cambridge Books Online © Cambridge University Press, 2010

10.2 Rodseth's Theorem for Binary Partitions 165

Now since 16|c1(m) and m > 4, we see that I^lc^m + 1).

c2(m + 1) = c2(m) + m2m"1c1(m) + <52(m + 2)2mc0(m - 1).

Since 8|c2(m), and m ^ 5, we see that 8|c2(m + 1). Also since 16-Tc2(m),
we see that 16f c2(m + 1), because all of the remaining terms on the right-hand
side of the equation above are divisible by 16.
Now for 2 < ; ^ m - 1

Cj(m + 1) = Cj(m) + (m + 2)2 m ~ 1 c,_ 1 (m)

ch(m - k + 1).

By hypothesis, 2 2y |c/m) and 22y|2m"~ ! c y _ ^m). Thus to prove that 22j\cj(m + 1),
we need only establish that

m 4- 1\ /m - fc + 2
2 )"( 2

when k + h = j , [(m + l)/2] ^ k ^ 2, A ^ 0.

m + 1\ /m - k + 2\ Ik + 1

^ km - K^2 + fc) - 2 + 2/
= fc(m - Kfc + 1)) - 2 + 2)

since 2 ^ k ^ [(m + l)/2]; thus 2 2 i |c/m + 1), and Theorem 10.1 is

established. •

The theorem above allows us to prove certain congruences for b(4n) — b(n)
originally conjectured by R. F. Churchhouse and proved independently by
O. Rodseth and H. Gupta.

THEOREM 10.2. If k ^ 1 and t = 1 (mod 2), then

b(22k+2t) - (b2kt) = 0 (mod 2 3 * +2 ), (10.2.9)

b(22k+lt) - b(22k~1t) = 0(mod 23k); (10.2.10)

furthermore, (10.2.9) and (10.2.10) are best possible in that no higher power
of 2 divides b(4n) — b(n).

Cambridge Books Online © Cambridge University Press, 2010

166 Congruence Properties of Partition Functions Chap. 10.2

Proof. We shall show that

2-3k'2^2k+i(q) and 2-3k^2k(q)

have integral coefficients and that the coefficients of the odd powers of q
are themselves odd. This is equivalent to Theorem 10.2.
If k = 1, then (10.2.5) implies

t
(I-*)* (I-*) 3 (l-«)2 A
Also by (10.2.6),

Thus Theorem 2 is established for k = 1.

Assume that Theorem 10.2 is true for all integers less than k. Then by
Theorem 10.1

By Theorem 10.1, 16|c1(2/c), 8|c2(2/c), and 22J\cj(2k), also for k > 1,

(2fc2+1)-3*>0.

Hence from the induction hypothesis we see that 2~3k^2k((l) nas

integral
coefficients. Furthermore since 16-rc2(2/c),

(mod 2).

Now by Theorem 10.1,

2--3k
3k 22
~ c2(2k

qG(q)
\2k+3

Cambridge Books Online © Cambridge University Press, 2010

10.3 Ramanujan's Conjecture for 5 n 157

Again by Theorem 10.1, 16|c1(2fc + 1), S\c2(2k + 1), 22J\cj(2k), also for
k> 1,

Hence from the induction hypothesis we see that 2~3k~2&r2k+1(#) nas

integral
coefficients. Furthermore, since 16ic2(2k + 1),

^ f ^ + 1 (mod 2).
i=o
Thus we have Theorem 10.2. •

10.3 Ramanujan's Conjecture for 5"

Our first task is to translate Ramanujan's conjecture into terms amenable

to the use of generating functions. In particular, we shall obtain a general
family of identities of which (10.1.3) is the simplest case. Besides the Hecke
operator U5, we also require the modular equation of fifth degree, which we
state in the form

qs(q"iq")l_ q$(q"; q ) l L 2 q(q Iq & ,

s 55
;q ))i
(q ;q

For a proof of this result we refer the reader to Knopp (1970, p. 119, Eq. (2.5)).
To simplify notation, we define

*, (10.3.2)

and

(10.3.3)
(q)l
T h u s w e m a y rewrite (10.3.1) as

<W#)5 = #(4 5 ){5 2 </>(<z) 4 + 52<f>(q)3 + 3-5<t>(q)2 4- 5<t>(q) + 1 } . (10.3.4)

In the f o l l o w i n g t h e o r e m s it is useful to have

Sr = 5U5<t>r(q) = Y 0 r (^ 1 / 5 p l ) (10.3.5)

Cambridge Books Online © Cambridge University Press, 2010

168 Congruence Properties of Partition Functions Chap. 10.3

where p = e2ni/5. Also we require the 5-adic valuation:

DEFINITION 10.1. If r is a rational number, then by the fundamental theorem
of arithmetic r may be uniquely written as ± Piaip2a2''' Pr°r where the pt
are prime numbers and the a, are integers (positive, negative, or zero). The
5-adic valuation v(r) is defined to be the exponent of 5 in this prime decomposi-
tion of r.
LEMMA 10.3. For each positive integer r, S r is a polynomial in g(q) of the
form

where 5 divides each arj, also v(arj) ^ [(5/ — r 4- l)/2], and arj = 0 unless
[(r 4- 4)/5] ^ j ^ r, where [x] /s //ie greatest integer function.
Proof. We begin by fixing q and considering the fifth-degree polynomial
equation in M:

w5 - 0(4)(5 2 " 4 4- 5 2 M 3 4- 3-5M 2 4- 5M + 1) = 0. (10.3.6)

From (10.3.4) we see that u = (f>(qxl5) is one root; in fact, u = (f)(ql/5pJ),

j = 0, 1, 2, 3, 4 (where p = e2nt/5), are roots of (10.3.6) since the replacement
of q by pjq in (10.3.4) only alters (f>(q). Furthermore, since 1 is the coefficient of
q in (j)(q) (by (10.3.2)), we see that the <j){qll5pJ) are all different functions of
q and are indeed distinct as complex numbers for q sufficiently close to zero,
since (f){q) ~ q as q -> 0. Hence we have determined five distinct roots of
(10.3.6) when q is a sufficiently small complex number, and so

Y\(u - ^ ' V ' ) ) = u5 - g(q)(52u4 4- 5 2 M 3 4- 3-5M 2 4- 5w 4- 1). (10.3.7)

By analytic continuation (10.3.7) is valid for all q with \q\ < 1.

Now let us recall Newton's formula that connects the sums of like powers
of the roots of a polynomial with the coefficients of the polynomial. Namely, if
n 1 n 2 n
f(x) = x - tf!*"" 4- a2x ~ - • • • 4- ( - l) anf

= (x - rx)(x - r2)-"(x - rn)

and if

5 = r , m 4- r2m 4- • • * 4- r n m ,
tn l z ' n *

then for 1 < j < n

2 -••• + (- iy-v,*! + (- o v = °.

Cambridge Books Online © Cambridge University Press, 2010

10.3 Ramanujan's Conjecture for 5 n 169

and for j > n

sj - <*iSj-i + a2Sj-2 - • • • + ( - lYajSj-n = 0.

Applying Newton's formula to /(x), we find that since Sr is the sum of the
rth powers of the roots of the polynomial in (10.3.6),

Si = 52g(q), (10.3.8)!

s2 = 5V(«) + 2-52g(q), (10.3.8)2

s3 = S6g\q) + 3-5+g2(q) + 9-5g(q), (10.3.8)3

s* = 5*9\q) + 4.5V(«) + 22-5V(«) + 4- 5$Gj), (10.3.8)*

Ss = 51 V(<Z) + 5-5 V ( « ) + 40-5533(^)

+ 20-5 3 (10.3.8),

and for r > 5

4 + flf(«)Sr.5. (10.3.8)r

We may now easily check all the assertions of Theorem 10.3 directly; the
only one that requires any real care is the inequality for v(arj), which is directly
verified by induction on r. For 1 ^ r < 5, we need only examine the explicit
form of the identity for Sr. For r > 5, assume v(asj) ^ 1(5j — s + l)/5] for
s < r and then examine (10.3.8)r:

arj = 52ar-1J.l + 52ar.2J.l + 15ar.3J.x + 5a r « 4J _ 1 + flr-5J-i.

Therefore

v(arj) ^ min(v(ar-lJ_l) + 2, u(a r _ 2J _ 1 ) + 2,

5// -- r + ll [5/ - r + 2l [5/ - r + ll

5 J' [ 5 J ' I 2 J '
5/ - r + 2

5j - r + 1
5

and Lemma 10.3 is established.

Cambridge Books Online © Cambridge University Press, 2010

170 Congruence Properties of Partition Functions Chap. 10.3

We are now prepared for the proof of Ramanujan's conjecture for powers
of 5. Let us restate the conjecture for powers of 5 only:
If 24A = 1 (mod 5fl), then p(X) = 0 (mod 5").
We begin by remarking that the smallest positive integral solution of
24A = 1 (mod 5a) is

^ ( 1 9 - 5 2 r " 1 + 1) if a = 2r- 1,
c* = \
i-(23-5 2 '+l) if a =

Consequently we wish to show that

p(5am + c a ) = 0 (mod 5")

for each m ^ 0.
In light of our work in Section 10.2, it would be natural to guess that our
interest should center on

£ p(5'm + ca)qm = U5 £ p{5''lm + ca)qm;

m=0 m=0

however, as we shall see in the utilization of the modular equation, we must

instead consider
oo
^2n-l(4) ~ (<Z5» 45)oo Z PiS2"'1™ + C2n-l)<lm+l (10.3.9)
m= 0

and

L2n(q) = (q)oo £ P(52nm + c2n)qm+l. (10.3.10)

m= 0

The following theorem provides us with strong information concerning the

coefficients of Ln(q), and Ramanujan's conjecture for powers of 5 becomes a
simple corollary.

THEOREM 10.4. Each Ln(q) is a polynomial in g(q) (see (10.3.3) for the
definition of g(q)) with integral coefficients. Furthermore, if we write this
polynomial as Ln(q) = £,°°=o bMg\q\ then

bn0 = 0, (10.3.11)

v(bni) = n, (10.3.12)

>*bj^n for all s ^ 1. (10.3.13)

Cambridge Books Online © Cambridge University Press, 2010

10.3 Ramanujan's Conjecture for 5" 171

Proof. The proof proceeds by mathematical induction on M, much in the

manner of the proof of Theorem 10.1. The primary complication arises from
the peculiarities of the two-line definition of Ln(q) (Eqs. (10.3.9) and (10.3.10)
are forced on us by the varying form of the solutions to the congruence
24A = 1 (mod 5n)). Also, we are forced to prove more than (10.3.13), so we
replace (10.3.13) by

v(bns) >n + [i(10s - 9 + ( - 1)")]. (10.3.13)'

When n = 1,

(since UJM5)fi{q) = Mq)UJ2(q))

-is,
= Sg(q) (10.3.14)

(by (10.3.8),). Hence, bl0 = O,fc,, = 5, bu = Ofors > 1, and Eqs. (10.3.11),
(10.3.12), and (10.3.13)' are all clearly valid when n = 1.
Let us now assume that our theorem is valid for every integer n < 2r — 1.
We must show that this implies the result is valid for n = 2r and 2r + 1. By
(10.3.9)

L2r(q) = ( « ) . I P(52rm + c2p)<zm+l

= («). f P(52rm + 4-5 2 '" 1 + c,,.,)^* 1

Cambridge Books Online © Cambridge University Press, 2010

172 Congruence Properties of Partition Functions Chap. 10.3

s=0

s=0

1
T- Z ^65j9j(q) (by Theorem 10.3)
•> 7 = 1

f=-oo s=0

,, (10.3.15)
f= -00

where

*2r,t = iZ*2r-1.^6,.r + .. (10.3.16)

s=0

We may now deduce the truth of (10.3.11), (10.3.12), and (10.3.13)' for
b2rft from (10.3.16) using Theorem 10.3 (for properties of the a6st+s) and
the induction hypothesis (for properties of the b2r-itS). In particular, if
t < 0, then t + s ^ 5, and so by Theorem 10.3 a6Stt+s = 0 for s > 0 since

> 5 ^ t + S.

By the induction hypothesis fc2r-i,o = 0- Hence we see that 62r,f = 0 if

f ^ 0, confirming (10.3.11) when n = 2r.
Furthermore, since L2r-i(q) is (by hypothesis) a polynomial in g(q), we
see that b2r-i,s = 0 for 5 ^ so(r). Also by Theorem 10.3, a6st+s = 0 unless
r ^ 5s. Hence, if t ^ 5so(r), then one of the factors of each term in (10.3.16)
is zero; so b2rt = 0 for t sufficiently large. Therefore L2r(q) is a polynomial
in g(q). Next
00
b fo
2r.l = i Z 2r-1,5*^65,5+1
5= 0

Cambridge Books Online © Cambridge University Press, 2010

10.3 Ramanujan's Conjecture for 5 n \ 73

= 6 2r _ 1 , 1 -5-63

(by evaluation of a6 2 from Eq. (10.3.8)6). Hence

by the induction hypothesis, and (10.3.12) is valid for n = 2r.

Finally for t ^ 1, by (10.3.16), Theorem 10.3, and (10.3.13)', with n
2 r - 1,

min r_ ltJ

55-5 5* - s + 1
= — 1 -f min - 1+

-5+ 1
^ - at$=l

- 4
2r

which establishes (10.3.13), with n = 2r.

To conclude our proof by induction we must now prove our assertions
for n = 2r + 1 using the induction hypotheses and the case in which n = 2r.
Since the steps resemble those above, we omit most of the details; we derive
the counterpart of (10.3.16), and we leave to the reader the necessary deduc-
tions from this result.

Llt+M) = (<?'; <?5)oo I P(5 2r+1 m + c 2 r + 1 ),T + 1

m=0

= 0? 5 ; <75)oo Z K 5 2 r ^ + 3 - 5 2 ' + c2r)qm+l

m=O

)f
m= 0

Cambridge Books Online © Cambridge University Press, 2010

174 Congruence Properties of Partition Functions Chap. 10.3

= V5 £ b2rtS<t>(q)g°(q)
s= 0

= £ b2r,,U5(<l>(q)g%q))

jq) J | , a « * + » V ^ (by Theorem

co ao

= E ff'(9)Zi*2,y»6.+u (io.3.i7)

whence

2rs<>6,+ l t + s - (10.3.18)
s=0

The remainder of the proof follows just as in the treatment of (10.3.16). •

COROLLARY 10.5 (Ramanujan's conjecture for powers of 5). / / 24A =

1 (mod 5"), then

p(X) = 0 (mod 5").

Proof. By the remarks preceding Theorem 10.4, we see that we must show
that

p(5am 4- O = 0(mod5 fl )

for each m ^ 0. If a = 2n — 1, then

M(fS- dO.3.19)
s=l KH i H Jao

Now gs(q)l(q5l q5)^ is always a power series in g with integral coefficients,

and by Theorem 10.4 5 2 " " 1 | 6 2 I I - I ^ f° r each s. Hence the coefficient of

Cambridge Books Online © Cambridge University Press, 2010

Examples 175

<f+ 1
on the right-hand side of (10.3.19) is divisible by 52""1; therefore

The same argument handles the case in which a = 2n; the only change is
the replacement of (q5; q5)^ by (q)^. •

COROLLARY 10.6.

)1

(q)Z>
Proof. This is just a restatement of Eq. (10.3.14). •
A natural question to ask at this stage is: Are there any combinatorial
interpretations for these congruences? Actually the two congruences (10.1.1)
and (10.1.2) have combinatorial interpretations; however, none of the other
congruences have known interpretations. The following theorem (conjectured
by F. J. Dyson and proved by A. O. L. Atkin and H. P. F. Swinnerton-Dyer)
provides the known interpretations:

THEOREM 10.7. Let R(k, a; n) denote the number of partitions of n whose

first rank (see Section 9.3) is congruent to a modulo k. Then

, 0; 5m + 4) = K(5, 1; 5m 4- 4) = R(5, 2; 5m + 4) = R(5, 3; 5m 4- 4)

= 11(5,4; 5m 4- 4);
, 0; 7m + 5) = R(l, 1; 7m + 5) = K(7, 2; 7m + 5) = R(7, 3; 7m + 5)
= K(7, 4; 7m + 5) = R(7, 5; 7m + 5) = K(7, 6; 7m + 5).

The only known proof of this would require too much space for its presenta-
tion here. We should point out that it is primarily an analytic proof which
relies heavily on the properties of modular functions. No combinatorial proof
of Theorem 10.7 is known.

Examples

In Examples 1-6, we consider problems for the number b(n; m) of parti-

tions of n into powers of m. The functions we shall consider are

*(m; q) = t Km; n)qn = f} (1 - qmT\

Cambridge Books Online © Cambridge University Press, 2010

176 Congruence Properties of Partition Functions Chap. 10

r(m; q) = (1 - q)<P(m; q) = 4>(m; qm),

00

note
y r + 1 ( r n ; * ) = UmYr(m\q) for r > 0.

qF(m; q)
(i /j^

3.

/m\ / /m\ 2 /m
3
4. ^j ^ ^ j ^
3

5. There exist integers ^,(r) (== ^,(^5 0 ) such that

}
G(m;q)q
i=0 U "~ 4)

where ao(r) = 1, mla^r) if m is odd, |w|fli(r) if m is even, and m2J~2\aj(r)

for 2 ^7"^ r - 1.
6. For each m, we let /z = m if m is odd and n = Jm if m is even. Then
b(m; mr+ln) - 6(m; mr«) = 0 (mod / ) .
In Examples 7-13, the simplest known proof that 5|p(5« + 4) is sketched.
7. It is a simple matter to deduce from Jacobfs triple product identity
(Theorem 2.8) that

m= 0

8. For — oo < r < oo, s ^ 0, the expression 1 -f |K3r + 1) + As(s +

is divisible by 5 if and only if r = 4 (mod 5) and 5 = 2 (mod 5).
9. From Examples 7 and 8 and the trivial identity

it follows that the coefficient of g5m + 5 in q(q)^ is divisible by 5.

Cambridge Books Online © Cambridge University Press, 2010

Notes 177

10. We may define f(q) = g(q) (mod 5) if an = bn (mod 5) for all n where
/(tf) = Zn^o ancLn anc* g(q) = X"^o bnqn. It is easy to show, using the binomial
series, that

11. From Example 10, we see that

12. Examples 9 and 11 imply that the coefficient of q5m+5 in (q(q5; ^5)00)/(<?)00
is divisible by 5.
13. From Example 12 it follows that 5|p(5m + 4) for all m ^ 0.

Notes
As was mentioned in Section 10.1, the most readable and complete recent
work on the divisibility properties of p(n) is the book of M. I. Knopp (1970).
Ramanujan's contributions are primarily contained in Paper 25 of his Col-
lected Papers (1927). The paper by Atkin (1967) contains a full account of
the IT case; G. N. Watson (1938) was the first person to treat fully the
cases 5" and T.
Atkin (1969), and others (see Lehner, 1969) have greatly extended the study
of arithmetic properties of modular forms; for example, Atkin (1969) has also
shown that
p(206839n + 2623) = 0 (mod 17).
The material on binary partitions is due to Rodseth (1970) and Gupta
(1971); our treatment is patterned along the lines of Andrews (1971). Theorem
10.2 was originally conjectured by Churchhouse (1969).
Kolberg (1957, 1960) has devised elegant elementary techniques for proving
some of the Ramanujan congruences, such as (10.1.1) and (10.1.2); Rademacher
(1973) gives a nice exposition of some of Kolberg's work.
Of Corollary 10.6 (due to Ramanujan), G. H. Hardy (Ramanujan, 1927,
p. xxxv) wrote: "It would be difficult to find more beautiful formulae than
the 'Rogers-Ramanujan' identities...; but here Ramanujan must take second
place to Prof. Rogers; and, if I had to select one formula from all Ramanujan's
work, I would agree with Major MacMahon in selecting..., viz.
• { . _5{(l-x5)(l-x10)(l-x1V-}5

where p(n) is the number of partitions of n."

Reviews of recent work on congruences for partition functions occur in
Section P76 of LeVeque (1974).
Examples 1-6. Churchhouse (1969), Rodseth (1970), Andrews (1971),
Gupta (1972).
Examples 7-13. Paper 25 of Ramanujan (1927).

Cambridge Books Online © Cambridge University Press, 2010

178 Congruence Properties of Partition Functions Chap. 10

References
Andrews, G. E. (1971). "Congruence properties of the m-ary partition function,"
/. Number Theory 3, 104-110.
Atkin, A. O. L. (1967). "Proof of a conjecture of Ramanujan," Glasgow Math. J. 8,
14-32.
Atkin, A. O. L. (1969). "Congruence Hecke operators,** Proc. Symp. Pure Math. 12,
33-40.
Atkin, A. O. L. and Swinnerton-Dyer, H. P. F. (1953). "Some properties of parti-
tions,'* Proc. London Math. Soc. (3) 4, 84-106.
Churchhouse, R. F. (1969). "Congruence properties of the binary partition function,'*
Proc. Cambridge Phil. Soc. 66, 371-376.
Dyson, F. J. (1944). "Some guesses in the theory of partitions," Eureka (Cambridge)
8, 10-15.
Gupta, H. (1971). "Proof of the Churchhouse conjecture concerning binary partitions,"
Proc. Cambridge Phil. Soc. 70, 53-56.
Gupta, H. (1972). "On m-ary partitions," Proc. Cambridge Phil. Soc. 71, 343-345.
Knopp, M. I. (1970). Modular Functions in Analytic Number Theory. Markham, Chicago.
Kolberg, O. (1957). "Some identities involving the partition function,'* Math. Scand.
5, 77-92.
Kolberg, O. (1960). "Congruences involving the partition function for the moduli 17,
19, and 23,** Univ. Bergen Arbok Naturvit. Rekke 1959, No. 15, 10 pp.
Lehner, J. (1969). Lectures on Modular Forms (Appl. Math. Ser. No. 61). Nat. Bur.
Standards, Washington, D.C.
LeVeque, W. J. (1974). Reviews in Number Theory, Vol. 4. Amer. Math. Soc, Providence,
R.I.
Rademacher, H. (1973). Topics in Analytic Number Theory. Springer, Berlin.
Ramanujan, S. (1927). Collected Papers of S. Ramanujan. Cambridge Univ. Press, London
and New York (reprinted by Chelsea, New York).
Rodseth, O. (1970). "Some arithmetical properties of m-ary partitions," Proc. Cambridge
Phil. Soc. 68, 447-453.
Watson, G. N. (1938). "Ramanujans Vermutung iiber Zerfallungsanzahlen," /. reine
angew. Math. 179, 97-128.

Cambridge Books Online © Cambridge University Press, 2010

CHAPTER 11

Higher-Dimensional Parfifions

11.1 Introduction

The previous chapters have (with the exception of the multipartite com-
>sitions treated in Chapter 4) treated partitions as a linear array whose sum
positions
is prescribed:

n = nx + n2 + • • • + ns = £ ni9
In this chapter we shall look at higher-dimensional partitions, that is,
arrays whose sum is n:

whenever it < j u i2 < j 2 , . • •> K ^ h (all nhi2 ~ir nonnegative integers).

Surprisingly, there is much of interest when the dimension is 1 or 2, and
very little when the dimension exceeds 2. In Section 11.2, we shall follow
L. Carlitz's proof of MacMahon's fundamental theorem utilizing determinant
representations of the appropriate generating functions; the recurrence
method of attack used here also arises in Chapter 12 when we consider vector
partitions.
In Section 11.3, we study a fundamental algorithm of Knuth that provides
the method of proof for numerous theorems concerning plane partitions.
In Section 11.4, we consider questions concerning higher-dimensional
partitions. Unfortunately, the most important results in this section are
negative. In particular, the conjectured form for the generating function for
higher-dimensional partitions is shown to be invalid.

11.2 Plane Partitions

From the comments in the preceding section we see that the plane partitions
of n are two-dimensional arrays of nonnegative integers in the first quadrant

ENCYCLOPEDIA OF MATHEMATICS and Its Applications, Gian-Cark) Rota (ed.).

2, George E. Andrews, The Theory of Partitions
Copyright © 1976 by Addison-Wesley Publishing Company, Inc., Advanced Book Program.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical photocopying,
recording, or otherwise, without the prior permission of the publisher.

179

Cambridge Books Online © Cambridge University Press, 2010

180 Higher-Dimensional Partitions Chap. 11.2

subject to a "nonincreasing" condition along rows and columns. For example,

there are six plane partitions of 3:

0 0 0 ... 0 0 0 0 0 0
0 0 0 ... 0 0 0 1 0 0
3 0 0 ... 2 1 0 2 0 0
0 0 0 0 0 0 1 0 0
0 0 0 1 0 0 1 0 0
1 1 1 1 1 0 1 0 0

zerOS and write

1
1 1 1
3, 21, 2, HI, 11, 1

We remark that many times plane partitions are represented in the fourth
quadrant rather than the first; thus the plane partitions of 3 are written

3, 21, 2, 111, 11, 1

1 1 1
1
We shall follow the latter custom to keep our results consistent with the
current literature; however, we point out that the use of the first quadrant
simplifies notation when investigating higher-dimensional partitions.
DEFINITION 11.1. Let nr(nu n2,..., nk; q) denote the generating function
for plane partitions with at most r columns, at most k rows, and with n{
the first entry in the ith row.
Our work on nr(nl9 n2,..., nk; q) will require an extensive use of Gaussian
polynomials and we refer the reader to Section 3.3 for the appropriate back-
ground material.
We start by noting that the nr{nu..., nk\ q) are completely determined by
the following recurrence and initial condition:

7 T r + ^ M j , ^ 2 , • • • , ^k, ^7/ ^ q La L-t ''' Lu

mk-0 mk->-mk mi-m2 ^ ^

nl(nl,n2i.. .,nk\q) = qni+n*+-+n*. (11.2.2)

Equation (11.2.2) is obvious, and Eq. (11.2.1) is easily seen by examining

the array left after the initial column (nl9 n2,- • •» nt) n a s t>een removed from
an (r + l)-column array.

Cambridge Books Online © Cambridge University Press, 2010

11.2 Plane Partitions 181

We now observe that nr(nl9..., nk; q) is easily computed using (11.2.1),

(11.2.2). For example
m n

0 n\ =nti

mi=o - q

-<72 l-q
1 - qm+1)(l - qm\
(1 - q)(l - q>)
^ l\n + 11 fm + 11 m + 11\

n + 11 [m + 1

= qn+m
• J «[ (11.2.3)
n + 11 Tm +
0 1
We may now apply (11.2.3) to the right-hand side of (11.2.1) with r = 2;
using (3.3.9) we may easily deduce that
m + 2

n3(n,m;q) = qn
m •[•
n + 2
3
m + 2
1
From here we immediately conjecture that
n + r - 1 m + r —1
r - 1
7rr(n, m;q) = qn
n + r — 1 m + r - 1
r - 2 r - 1
and we easily prove this by induction on r using (3.3.9).
After a few calculations with k = 3, we can make a reasonable guess
concerning the representation of nr(nl9 n2,-. ., «*; g) as a determinant:
THEOREM 11.1

n
£nl9 n 2 , . . . , Hfc; ^ ) = qni +
'"+tt'<det

(11.2.4)

Cambridge Books Online © Cambridge University Press, 2010

182 Higher-Dimensional Partitions Chap. 11.2

Proof. When r = 1, the determinant is upper triangular with ones down

the main diagonal. Hence (11.2.4) reduces to the trivial (11.2.2) when r = 1.
Assuming the result for a particular r, we proceed using (11.2.1):

q-»i »kjrr+!(«!,..., nk\q)

x det(«*«-»«-'-'> [ ^ ^ ^ J ) . (11.2.5)
We may now sum the inner sum, using (3.3.9). The result transforms the
right-hand side by leaving a (k — l)-fold summation and replacing the ith
entry in the first column of the determinant in the summand by

Now the unaltered ith entry of the second column is

r - iI + 1

so if we multiply the second column by q~l and add the result to our first
column, we obtain as ith entry for the transformed first column

We now sum with respect to m 2 , then we multiply the third column by q"1
and add the result to our new second column, whose ith entry becomes

At the 7th step, we sum with respect to my, we then multiply the (; + l)st
column by q~J and add the result to our7th column. The resulting jth column
has as ith entry

Hence, after all A: summations are completed, we have

^J) (11.2.6)

Cambridge Books Online © Cambridge University Press, 2010

11.2 Plane Partitions 183

If we multiply the /th row of the determinant in (11.2.6) by ql" l and divide the
7th column by qJ~l (and as a result do not alter the value of the determinant),
we obtain

n r
i + \\
. . . (11.2.7)
r - i +7J/

which is (11.2.4) with r replaced by r + 1. Hence our theorem is established

by mathematical induction. •

DEFINITION 11.2. Let pktr(m, n) denote the number of plane partitions of m

with at most r columns, at most k rows, and with each entry ^ n, and let

y^ VkAmi n)qm-
m=0

We observe immediately that

= q-knnr+l(n,. ..,n;q) (11.2.8)

(by (11.2.1)). Hence

= det( 9 «'-^'->-» f ) 1) . (11.2.9)

\ [r i h J}/
It is possible to represent the polynomial ^fc,r(n; q) as a simple quotient of
products of expressions (</),. This is easy to do by direct computation for
small /c, and the results lead us to conjecture:

THEOREM 11.2

Proof. We proceed by using another ingenious device of L. Carlitz. Let

H'(fc.r) = det f^* 1 " PI

Hence by the rule for multiplication of determinants

nk%,(n\q)W(k%r) =

where

Cambridge Books Online © Cambridge University Press, 2010

184 Higher-Dimensional Partitions Chap. 11.3

s= 0 n+ i- s

= q
[ n+ i
(by (3.3.10)). Therefore

,r) = ( - ^

> r) (n. 2 .n)

(^)n(^)n+r-*(^)« + fc-l
where neither H^(/c, r) (which is upper triangular with no zeros on the main
diagonal) nor C(k, r) depends on n. We may therefore determine C(k, r)/ W(k, r)
by setting n = 0 in (11.2.11). Thus

q)(q)'-(q) (n 2 12)
=
) (q)r(q)r+l"(q)r+k-i
and therefore

as desired. •

Theorem 11.2 directly implies MacMahon's famous formulas for the

generating function of /c-rowed plane partitions nktO0(co; q):

COROLLARY 11.3. For \q\ < 1,

Z pk.Jm* oo)qm = f [ ( l - ^)-««»(*^ ; (11.2.14)

m=0

, 00)9" = ft (1 - 9;)"7'- (11.2.15)

m= 0

Proo/. Equation (11.2.14) follows from Theorem 11.2 by letting r and

n -> oo. Equation (11.2.15) follows from (11.2.14) by letting k -• oo in
(11.2.14). •

11.3 The Knuth-Schensted Correspondence

In the 1960s a correspondence between certain matrices and plane parti-

tions was developed by Schensted and later extended by Knuth. This

Cambridge Books Online © Cambridge University Press, 2010

11.3 The Knuth-Schensted Correspondence 185

combinatorial mapping allowed a great simplification in the deduction of

many known restricted plane partition generating functions. Furthermore,
many new generating functions can be treated using this elegant procedure.

THEOREM 11.4. There is a one-to-one correspondence between

(i) the set ofkxk matrices A with nonnegative integer entries;

(ii) the set of all lexicographically ordered sequences of ordered pairs
of integers, each ^ k;
(iii) the set of ordered pairs (n, n') of plane partitions in which there is
strict decrease along columns, each entry does not exceed k, and the cor-
responding rows of n and nf are of the same length.
Furthermore the desired correspondence is such that the ith column sum
(resp. ith row sum) of a matrix A from (i) equals the number of times i
appears in the corresponding plane partition n (resp. nr) in (iii).

Proof. The correspondence between (i) and (ii) is simple; to each sequence
described in (ii) associate the incidence matrix A = (a 0 ), that is, atj is the
number of times the pair (i,j) (written for our purposes as j) occurs in the
corresponding sequence of (ii). For example

2 2 2 1 1 1 1 1 1 1
(1L31)
3 2 1 3 3 2 2 1 1 1

We now produce the ordered pair of plane partitions (n, n')\ the entries in
n arise from the second elements (or bottom row) in the sequence of ordered
pairs; those in n' arise from the first elements (or top row). Our procedure
is an algorithm that describes how to build up n and n' step by step from the
sequence of pairs in (ii).
The second elements j from each term j of our sequence are to be inserted
in their appropriate place in the top row of n (i.e., so that nonincrease along
rows is maintained). If the appropriate place of insertion for; already contains
a smaller integer j l9 then; takes the place ofjt and the smaller integer ; x is
inserted in the second row. If the appropriate place in the second row for ;\
is taken by; 2 , then; 2 is inserted in the third row, and so on. Once the sequence
of bumpings caused by the insertion of; is concluded, then i (the first element
of the pair j) is inserted into nr in such a way that the corresponding rows of
n and n' are always of the same length.
To obtain a feel for the mechanics of the foregoing procedure, we produce
step by step the (TT, n') corresponding to (11.3.1) (the circled number is the
new insertion, the underlined numbers are ones "bumped" by the new
insertion).

Cambridge Books Online © Cambridge University Press, 2010

186 Higher-Dimensional Partitions Chap. 11.3

3 © 3®
3 2® 3 3 (3)

3 2 2 (D 3 3 3 ®
3 2 2 2 © 3 3 3 3 (3)
3 2 2 2 I (£ 3 3 3 3 3®
3 ~ 2 2 I I 3 3 3 3 3 3
2

3 3 2 2 I 3 3 3 3 3 3
2 I

3 3 2 2 2 1 3 3 3 3 3 3 (2)
2 I 2 2

3 3 2 2 11 3 3 3 3 3 3 2
2 2 2 2

3 3 3 2 II 3 3 3 3 3 3 2
2 2 1 22©
I I

3 3 3 3 2 (2) I 3 3 3 3 3 3 2
2 2 2 1 2 2 I ©
I I

3 3 3 3 22(2) 3 3 3 3 3 3 2
2 2 2 11 2 2 I I ©
I I

3 3 3 3 2 2 2 © 3 3 3 3 3 3 2 ®
2 2 2 11 2 2 I II
I I

3 3 3 3 2 2 2 I© 3 3 3 3 3 3 2 1 ©
2 2 2 11 2 2 I II
I I

3 3 3 3 2 2 2 1I 3 3 3 3 3 3 2 1 I
2 2 2 11 2 2 I II
I I

Cambridge Books Online © Cambridge University Press, 2010

11.3 The Knuth-Schensted Correspondence 187

Inspection of the foregoing procedure shows immediately that the TT and TT'
have corresponding rows of equal length and the correspondence between
the appearances of / in n (resp. TT') and the /th column (resp. row) sum is
clear. Furthermore (using mathematical induction on the number of steps
completed in the procedure), we see that bumping can only result in entries
moving downward and not to the right, and thus strict decrease along col-
umns is maintained in TT at each step. Since insertions into ri are successively
made from a nonincreasing sequence, we see that nonincrease must hold
along rows and columns; moreover, strict decrease must obtain in the col-
umns of ri since if ji j 2 * r e t w o successive terms in our sequence, the number
of bumpings induced by jx is greater than or equal to the number induced by
j 2 (hence, insertions of / move successively to the right).
The main question, however, is: Does the foregoing procedure establish
a one-to-one correspondence between the elements defined in (ii) and those
defined in (iii)? Let us show that our procedure is uniquely reversible and
consequently does establish a one-to-one correspondence. First we point out
that the remark made at the end of the preceding paragraph allows us to pick
out the last insertion made in n' (it is the rightmost appearance of the smallest
integer appearing). If this integer, say /, occurs on the first row of n', then
the corresponding entry in n is the last insertion in n. If / occurs on some
other row, then the corresponding entry in n was the end of a bumping chain
and we can clearly reverse the bumping process to determine what entry was
inserted in n. Thus the procedure yields a bijection, and Theorem 11.4 is
established. •

THEOREM 11.5. / / the matrix A in (i) of Theorem 11.4 corresponds to

(TT, 7:') in (iii), then the transposed matrix AT corresponds to (n\ n).

Proof. Let us define the class of j in the sequence of pairs corresponding

to A as the position in the first row of n where./ is initially inserted. Returning
to our example (11.3.1), we may list the classes of each pair directly below
the pair as follows:

3 3 3 3 3 3 2 2 2 1 1 1 1 1 1 1
3 2 2 2 1 1 3 2 1 3 3 2 2 1 1 1
class 1 2 3 4 5 6 2 5 7 3 4 6 7 8 9 10

Now the entries in the first row of n are made up in order of the last entries
in each class: namely, 3 3 3 3 2 2 2 1 1 1 .
The important thing about class membership is that it is possible to define
it in a manner that does not depend on the lexicographic ordering of the
sequence. Namely, j is in class t if and only if t is the largest subset of pairs

Cambridge Books Online © Cambridge University Press, 2010

188 Higher-Dimensional Partitions Chap. 11.3

such that
J1J2'' 'Jt-iJ '1 ^ h ^ ' * * ^ h-i ^ '•

Thus from this last observation we see that the class of j in our sequence is
the same as that of { in the sequence of reversed ordered pairs, which clearly
corresponds to AT. Hence the first row of the first plane partition corresponding
to AT is the first row of n' and the first row of the second plane partition
corresponding to AT is the first row of n.
Deleting from the ordered pairs sequence the entries that produce the first
rows of the plane partitions, we may follow exactly the same procedure to
establish the interchange of the second rows, and so on. Thus we see that
(n\ n) corresponds to AT. •

COROLLARY 11.6. The correspondence described in Theorem 11.4 produces

a one-to-one correspondence between symmetric matrices A of nonnegative
integers and plane partitions n with strict decrease along columns. Further-
more, the number of appearances of i in n equals the ith row sum (or ith
column sum) of A.

Proof. We invoke Theorem 11.5, noting that A is symmetric if and only

if A = AT. Hence, A <-• (n, n). •

This corollary has many interesting applications; we present one of the

most striking next.

THEOREM 11.7. Let S be a set of positive integers. The number of plane

partitions of n with strict decrease along columns and with each summand
in S is the coefficient of qn in

ieS iJeS

Proof. By Corollary 11.6, the desired generating function is

z n <?•""
A iJZl

where the sum is over all symmetric matrices A of nonnegative integers subject
to the condition that au = 0 if / ^ S or j' $ S. Hence

/ I I Q -
A i,j>l

Cambridge Books Online © Cambridge University Press, 2010

11.4 Higher-Dimensional Partitions 189

ieS

In the examples there is discussed the fact that certain two-variable generating
functions arise in a very natural way relative to plane partitions through
Corollary 11.6. For example, the coefficient of amqn in

is the number of plane partitions of n for which the sum of the diagonal
parts is m.

11.4 Higher-Dimensional Partitions

P. A. MacMahon at one time conjectured that if Mk(n) denotes the number

of /c-dimensional partitions of n (see the definition in Section 11.1), then

f Mk(n)qn = f [ ( l - q*)~( k~1 \ (11.4.1)

n=0 i= 1

The conjecture is certainly true for k = 1 and 2. MacMahon eventually came

to doubt the truth of (11.4.1) in general; however, its falsehood for all k > 2
was only established in the late 1960s. The following theorem provides a
simple (though laborious) computational method for disproving MacMahon's
conjecture; this method is easily extended to any problem involving k-
dimensional partitions of small integers.
THEOREM 11.8. Let

oo _(*+«-2\

n=0 i=l

Then

= Mk(0) = 1, (11.4.3)
= Mk(l) = 1, (11.4.4)
/i*(2)= Mk(2) = fc+ 1, (11.4.5)

(11.4.6)

Cambridge Books Online © Cambridge University Press, 2010

190 Higher-Dimensional Partitions Chap. 11.4

lk\ lk\
-Mk(4)= 1 + 4*+ 4 , + , , (11.4.7)

lk\ lk\ lk\

Hk(5) = Mk(5)= 1 + 6 * + 11 L + 7 L+ J . (11.4.8)

M,(6) + Q + Q = 1 + 10* + 27 Q + 29 Q + 12 Q +
(11.4.9)
Proof. We begin by treating //*(n):

K
x (1 - q*) 3 ; (1 - q5) (
* \ i - q6) ( s
'• • •
2 3 A 5 6
= (1 + q + q + q + q + q + q + • • •)

1 + q + (k + l)q2 + 11 + 2k + ( 2 jj q3

4fe + 4 (

6/c

+ 11 + 10* + 27 ( j + 2

+ ••-, (11.4.10)

Cambridge Books Online © Cambridge University Press, 2010

11.4 Higher-Dimensional Partitions 191

Table 11.1

Total choices
Ordinary partition of 6 Arrangements in higher-dimensional space for placing
arrangement

6 6 at origin 1

-axis

5+1

jr. -axis

4+ 2

r*J -

4
4+ 1 + 1 '

-axis

^ j r , -axis

^rX. -axis
3+ 3 3^"

^rX. -axis

3+2+1 3^"

^ ^ jr.-axis

3 + 1 + 1+ 1

Cambridge Books Online © Cambridge University Press, 2010

192 Higher-Dimensional Partitions Chap. 11.4

Table 11.1 (continued)

Total choices
Ordinary partition of 6 Arrangements in higher-dimensional space for placing
arrangement

^^x.-axis

-axis

. -axis

2 4-2 + 2

rx. -axis

^ * x. -axis

2+2+1+1

-axis

-axis

Cambridge Books Online © Cambridge University Press, 2010

11.4 Higher-Dimensional Partitions 193

Table 11.1 (continued)

Total choices
Ordinary partition of 6 Arrangements in higher-dimensional space for placing
arrangement

rx. -axis

V^-axis

rx. -axis

jr-axis

2+ 1 + 1 + 1 + 1 2'

, x, -axis

-axis

wx. -axis

jr £ -axis

x. -axis

-axis

Cambridge Books Online © Cambridge University Press, 2010

194 Higher-Dimensional Partitions Chap. 11.4

Table 11.1 (continued)

Total choices
Ordinary partition of 6 Arrangements in higher-dimensional space for placing
arrangement

, x, -axis

-axis

2C
\ s^x

axis

^x,-axi

1+1+1+1+1+1+1
^.-ax.s

Cambridge Books Online © Cambridge University Press, 2010

11.4 Higher-Dimensional Partitions 195

Table 11.1 (continued)

Total choices
Ordinary partition of 6 Arrangements in higher-dimensional space for placing
arrangement

.,, x. -axis

,x. -axis

-axis
' xh - a x i s

-axis

-axis
\
xh -axis

xf - a x i s

\\Sr,o
I V - axis
^-oxis
C)

-axis

Cambridge Books Online © Cambridge University Press, 2010

196 Higher-Dimensional Partitions Chap. 11.4

Table 11.1 (continued)

Total choices
Ordinary partition of 6 Arrangements in higher-dimensional space for placing
arrangement

, x. -axis

-axis

-axis

x. -axis

i ^ '
xn -axis

,,-oxis

-axis
1

r. -axis

-axis
A'
1^— 1 " ^ -axis

^ ' x. -axis

A', -axis

'x:^1 -axis

-axis

Cambridge Books Online © Cambridge University Press, 2010

11.4 Higher-Dimensional Partitions 197

and we may now determine /.ik(n) for 0 ^ n ^ 6 by comparing coefficients

in the extremes of (11.4.10).
To compute Mk(n) we observe that each /c-dimensional partition is made
up of an ordinary (one-dimensional) partition of n appropriately distributed
in the first quadrant of /c-dimensional space. Thus to evaluate Mk(2) we see
that the ordinary partition 2 can be placed in /c-dimensional space in only
one way; 1 + 1, on the other hand, can be distributed in k ways: one 1 is
placed at the origin and the second is placed on any of the /c-coordinate
axes. Hence, Mk{2) = 1 + k. The remaining six expressions for Mk(n) are
derived in the same way. Since Mk(6) is crucial for Corollary 11.9, we evaluate
it using Table 11.1, which is self-explanatory. To obtain all possible /c-dimen-
sional partitions of 6, we add up all the entries in the right-hand column:

lk\ /k\ (k\ lk

Mk(6) = 1 + 10/c + 27 I 1 + 2 8 ( 1 + 11 ( j +
which is the result asserted in (11.4.9). •
COROLLARY 11.9. Equation (11.4.1) is valid for k = 1 or 2 and false for
all k > 2.
Proof. The case in which fc = 1 is a special case of Theorem 1.1. The case
in which k = 2 is Eq. (11.2.15). Equation (11.4.1) is equivalent to the assertion
that iik(n) = Mk(n) for all «; however
k k
l \j + ^l \j > 0
A*k(6) - Mk(6) = ^ for k > 3.

Hence (11.4.1) is false for k > 2. •

Given the disappointing nature of the foregoing results, we can well expect
disappointment on other questions concerning higher-dimensional partitions.
Observing that M2(3) = 6 = 3-2, M2(6) = 48 = 3 x 16, M2(9) = 282 =
3 x 94, M2(12) = 1479 = 3 x 493, M2(15) = 6879 = 3 x 2293, we might
be tempted to conjecture that 3|M2(3n) for all n. Our hopes are dashed,
however, by the following:
THEOREM 11.10. If k + \ is a prime, then

Mk((k + \)n) = 0(mod/c + 1 ) , 1 ^ n < (k + l ) \ (11.4.11)

Mk((k + l) fc+1 ) = 1 (mod k + 1). (11.4.12)

Proof. We may represent each /c-dimensional partition by a (k -f 1)-

dimensional graphical representation or Ferrers graph (merely place a column
of njlj2...jk nodes on the line parallel to the xk + l axis passing through the

Cambridge Books Online © Cambridge University Press, 2010

198 Higher-Dimensional Partitions Chap. 11

*!,^,. • . ,^,0). The group G of/c + 1 transformations T°, 7"1, JH2,. . ., Tk

where

T : ( x l J x 2 , . . . , x k + l ) - (xk + l , x l 9 x 2 , . . . , x k )

always either produces from a given Ferrers graph k + 1 different Ferrers

graphs or else leaves the initial Ferrers graph fixed. This is because G is a
cyclic group of prime order and so any nonidentity in G generates the full
group. If a Ferrers graph is left fixed by G, then since the only fixed points
under any Tl e G are points on the diagonal (x, x , . . . , x), we see that the
first Ferrers graph that partitions a multiple of k + 1 and is invariant under
G is just the (k + l)-dimensional hypercube with k + 1 nodes on each edge.
Therefore, the /e-dimensional partitions of (k + \)n can be divided into
disjoint sets of (k 4- 1) elements each (namely, the orbits of G) whenever
0 < n < (k 4- \)k. Thus (11.4.11) is valid.
On the other hand, the only (k + l)-dimensional Ferrers graph of (k + l)k + 1
that has A; + 1 nodes on the diagonal is the hypercube just described. Since
all the other ^-dimensional partitions of (k + l) fc+1 may be separated into
sets of k + 1 elements each, we see that

Mk((k 4- l)fc+1 ) = 1 (mod Jfc + 1). •

Examples

1. An examination of the structure of the successive ranks of a partition

(see Section 9.3) shows that there is a one-to-one correspondence between
each ordinary partition n = (A l s ..., /fc) of n and ordered pairs of partitions
{nu n2) where T^ = (kx\k2\. . .,/./),n2 = ( V \ . •., V M i = V> £ - * ='-i">
j] (// + A/') = /i, /.,' — k{' is the /th successive rank of n.
2. The correspondence in Example 1 may be utilized to establish a one-
to-one correspondence between plane partitions n of n with at most r rows
and each part ^ m and ordered pairs (n\ n") of plane partitions, each with
strict decrease along columns such that each part of n' is at most m and each
part of n" is at most r.
3. Example 2 and Theorem 11.4 may be used to prove Corollary 11.3 by
the use of the technique that appears in Theorem 11.7.
4. The Ferrers graphs of plane partitions (a concept introduced in Section
11.4) may be used to define six conjugates of a plane partition n (these arise
from the 3! possible permutations of the coordinate axes related to the three-
dimensional Ferrers graph of n). We consider one of these conjugations,
say T, for which TK is the plane partition in which each row of TK is the con-
jugate (in the ordinary partition sense) of the corresponding row of K. The
trace of a plane partition K is the sum £ nu of the diagonal entries in K;
the conjugate trace of K is the number of entries n^ of K such that nu ^ /.
The trace of K and the conjugate trace of TK are identical.
5. The proof in Example 3 may be extended and combined with Example 4
to show that the coefficient of zmqn in

Cambridge Books Online © Cambridge University Press, 2010

Notes 199

is the number of plane partitions of n whose trace is m.

6. The asymptotic work of Chapter 6, namely, Theorem 6.2, is applicable
to many of the plane partition generating functions. In particular, the number
of plane partitions of n, Pa^ooin, oo), satisfies
Pco.coOi, °°) ~ (C(3)2~ n ) 1 / 3 6 n- 2 5 / 3 6 exp(3-2- 2 / 3 C(3) 1 / 3 « 2 / 3 + 2c)
where

The relevant functions for Theorem 6.2 are D(s) = ((s — 1), g(x) =

Similar results can be obtained for numerous special cases of Theorem 11.7.
7*. Let us consider those r-dimensional partitions whose nonzero entries
n
ili2-ir o c c u r precisely at the points (il9 i2,..., ir) that define the graphical
representation of a fixed (r — l)-dimensional partition n of N. Let
am(su..., sv) denote the number of r-dimensional partitions of m with strict
decrease holding in Eq. (11.1.1) whenever any of / S| < j S l 9 iS2 < j S 2 , . . . , iSv <jSv
holds, where {sl9s29.. .9sv} is a fixed subset of { 1 , 2 , . . . , r } . Let {tl9t29.. .,/r-J
be the complement of {sl9 s2,..., sv} in {1, 2 , . . . , r}. If

and
B0(q) = Y.am(tut29..

then

B0(q) = ( - l)p(7t)^p
where p(n) is the number of (r — l)-dimensional partitions whose graphical
representation is contained within the graphical representation of n.

Notes
This chapter has only briefly touched a very active area of research. The
topic of higher-dimensional partitions, as such, originated with MacMahon
(1916). However, Young tableaux (which are essentially equivalent to plane
partitions with strict decrease along columns) were originated earlier by
Rev. Alfred Young in his work on invariant theory. Since then, Young
tableaux have played an important role in the representation theory of the
symmetric group (see Rutherford, 1948); they also occur in algebraic geometry
(see Lascoux, 1974a, b), and in many combinatorial problems (Kreweras,

Cambridge Books Online © Cambridge University Press, 2010

200 Higher-Dimensional Partitions Chap. 11

1965, 1967). Stanley (1971) presents an extensive and readable account of

current work. The material in Section 11.2 is from Carlitz (1967); his tech-
niques were extended by Knuth and Bender (1972) to other plane partition
problems. Hodge and Pedoe (1952) considered the case in which q = 1 for
a problem in algebraic geometry. The material in Section 11.3 was initially
considered by Robinson (1938) and Schensted (1961) (see also Schiitzenberger,
1963); we have presented Knuth's (1970) extension of the Schensted construc-
tion. The application to plane partitions is given by Knuth and Bender (1972).
Recent research on plane partitions was surely inspired by the pathbreaking
papers of B. Gordon and L. Houten (1968) and B. Gordon (1971).
Theorem 11.8 and its corollary were given by Atkin et al. (1967). Tietze
(1940, 1941) presented all the formulas for Mk(j), 1 ^ 7 ^ 5 ; however, he
was apparently unaware of MacMahon's conjecture. Wright (1968) has devel-
oped another technique for the disproof of many instances of MacMahon's
conjecture. Theorem 11.10 is due to Andrews (1971); the technique arises
from Wright (1968). The possibility of extending Rogers-Ramanujan type
identities to plane partitions was considered briefly by Chaundy (1931);
however, Gordon (1962) points out that Chaundy's derivation is in error.
Reviews of recent work related to the material in this chapter occur in
LeVeque (1974), Section P64.
Examples 1-3. Knuth and Bender (1972).
Examples 4-5. Stanley (1973).
Example 6. Wright (1931).
Example 7. Stanley (1972); Stanley's work on (P, co)-partitions allows him
to deduce the result in this example easily from a general reciprocity theorem.

References

Andrews, G. E. (1971). "On a conjecture of Guinand for the plane partition function,"
Proc. Edinburgh Math. Soc. (2) 17, 275-276.
Atkin, A. O. L., Bratley, P., MacDonald, I. G., and McKay, J. K. S. (1967). "Some
computations for ///-dimensional partitions," Proc. Cambridge Phil. Soc. 63, 1097-
1100.
Carlitz, L. (1967). "Rectangular arrays and plane partitions," Acta Arith. 13, 22-47.
Chaundy, T. W. (1931). "Partition-generating functions," Quart. J. Math. Oxford Ser.
2, 234-240.
Gordon, B. (1962). "Two new representations of the partition function," Proc. Amer.
Math. Soc. 13, 869-873.
Gordon, B. (1971). "Notes on plane partitions. V," J. Combinatorial Theory B l l ,
157-168.
Gordon, B. and Houten, L. (1968). "Notes on plane partitions. I, II," / . Combinatorial
Theory 4, 72-80, 81-99.
Hodge, W. V. D., and Pedoe, D. (1952). Methods of Algebraic Geometry, Vol. 2.
Cambridge Univ. Press, London and New York.
Knuth, D. E. (1970). "Permutations matrices and generalized Young tableaux," Pacific
J. Math. 34, 709-727.
Knuth, D. E., and Bender, E. A. (1972). "Enumeration of plane partitions," / . Com-
binatorial Theory 13, 40-54.

Cambridge Books Online © Cambridge University Press, 2010

References 201

Kreweras, G. (1965). "Sur une classe de problemes de denombrement lies au treillis

des partitions des entiers," Cahiers du B.U.R.O. No. 6, Paris.
Kreweras, G. (1967). "Traitment simultane du 'Probleme de Young' et 'Probleme
Simon Newcomb'," Cahiers du B.U.R.O. No. 10, Paris.
Lascoux, A. (1974a). "Polyndmes symetriques et coefficients d'intersection de cycles de
Schubert," C. R. Acad. Sci. Paris 279, 201-204.
Lascoux, A. (1974b). "Tableaux de Young et fonctions de Schur Littlewood," Seminaire
Delange-Pisot-Poitou No. 4, 1-7.
LeVeque, W. J. (1974). Reviews in Number Theory, Vol. 4. Amer. Math. Soc, Providence,
R.I.
MacMahon, P. A. (1916). Combinatory Analysis, Vol. 2. Cambridge Univ. Press, London
and New York (reprinted by Chelsea, New York, 1960).
Robinson, G. deB. (1938). "On the representations of the symmetric group," Amer. J.
Math. 60, 745-760.
Rutherford, D. E. (1948). Substitutional Analysis. Edinburgh Univ. Press, Edinburgh
(reprinted by Hafner, New York, 1968).
Schensted, C. (1961). "Longest increasing and decreasing subsequences," Canadian J.
Math. 13, 179-191.
Schiitzenberger, M.-P. (1963). "Quelques remarques sur une construction de Schensted,"
Math. Scand. 12, 117-128.
Stanley, R. P. (1971). "Theory and application of plane partitions, I, II," Studies in
Appl. Math. 50, 167-188, 259-279.
Stanley, R. P. (1972). "Ordered structures and partitions," Mem. Amer. Math. Soc.
119.
Stanley, R. P. (1973). "The conjugate trace and trace of a plane partition," /. Com-
binatorial Theory A14, 53-65.
Tietze, H. (1940-1941). "Systeme von Partitionen und Gitterpunktfiguren, I-IX," S.-B.
Math.-Natur. Abt. Bayer. Akad. Wiss. 23-54, 69-131, 133-145, 147-166; (1940);
1-37, 39-55, 165-170, 171-186, 187-191 (1941).
Tietze, H. (1941). "Uber die Anzahl komprimierter Gitterpunktmengen von gegebener
Punktezahl," Math. Z. 47, 352-356.
Wright, E. M. (1931). "Asymptotic partition formulae, I: Plane partitions," Quart. J.
Math. Oxford Ser. 2, 177-189.
Wright, E. M. (1968). "Rotatable partitions," /. London Math. Soc. 43, 501-505.

Cambridge Books Online © Cambridge University Press, 2010

CHAPTER 12

Vector or Multipartite Partitions

12.1 Introduction
As we saw in Chapter 4, problems often arise concerning the additive
decomposition of nonzero vectors with nonnegative integral coordinates
(also called multipartite numbers). Our work in Chapter 4 dealt with applica-
tions of "compositions" of multipartite numbers. Now we shall consider
partitions of multipartite numbers. As usual, compositions take order of
summands into account; partitions do not. The elementary theory of multi-
partite partitions resembles greatly the elementary theory of ordinary partitions
presented in Chapter 1 related to infinite products. Unfortunately there is
not known any truly simple type of infinite series that is as useful in treating
multipartite partitions as the basic hypergeometric series (introduced in
Chapter 2) are in treating ordinary partitions. Thus we only know of com-
paratively simple identities related to multipartite partition functions (see
Section 12.2), and we do not have any device like Euler's pentagonal number
theorem (Corollary 1.7) for the rapid calculation of any multipartite partition
functions. The most useful theorems for actual evaluation of multipartite
partition functions are presented in Theorem 12.3.
L. Carlitz and others have considered problems of restricted multipartite
partitions. In particular, they look at /c-partite partitions

(«!,..., nk) = £ (m u , m2p..., mkj)

subject to the "decreasing" condition

m i n ( m i y , . . . , mkJ) ^ max(mj J + 1 , . . . , mkJ+l).

We shall examine such questions in Section 12.4.

12.2 Multipartite Generating Functions

We recall from Section 4.3 that by P(n) = P(nl9 n 2 , . . . , nr) we denote the
number of partitions of the "r-partite" or "multipartite" number (n t , n2i..., nr)

ENCYCLOPEDIA OF MATHEMATICS and Its AppUcations, Gian-Cark) Rota (ed.).

2, George E. Andrews, The Theory of Partitions
Copyright © 1976 by Addison-Wesley Publishing Company, Inc., Advanced Book Program.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical photocopying,
recording, or otherwise, without the prior permission of the publisher.

202

Cambridge Books Online © Cambridge University Press, 2010

12.2 Multipartite Generating Functions 203

(i.e., an ordered r-tuple of nonnegative integers not all zero). That is, P(n) is
the number of distinct representations of n as a sum of multipartite numbers:
n = g") + § (2) + • • • + £ s ) (12.2.1)
(i) i+1
subject to lexicographic ordering g ^ g( > of the parts:

provided f / ° > £ / l + 1 ) where; is the least integer such that £ / ° # £ / l + 1).

If the number of parts in the partition is restricted to be at most j , we write

For simplicity in the treatment of the related generating functions we define

Q(n) as the number of partitions of n into distinct parts where (0, 0 , . . . , 0)
may be a part, and Q(n;j) is the number of such partitions withy parts. If we
desire that (0) be excluded as a part, we shall write Q+(n) or Q+(n;j). The
treatment in Chapter 1 extends directly to the study of P(n) and Q(n). In fact,
obvious alterations of Theorem 1.1 allow us to prove that
2
£ P(n)xr- 'X* = ft (1 - W ' * 'XrnT\ (12.2.2)
fii,...,nr^O m^0"nr>0,
not all zero

£ Q(n)x 1 ---x r -'= [] (1 + x1"'x2"J--xr"0; (12.2.3)

here for absolute convergence we need only require that |JC,| < 1 for 1 <
i ^ r.
It is now fairly obvious that all properties of ordinary (or unipartite)
partitions that depend solely on infinite products can be extended without
much effort to properties of multipartite numbers. For example, we have
Cheema's extension of Euler's theorem:
THEOREM 12.1. For every n, Q+(n) equals (P(n), the number of partitions
of n (see (12.2.1)) in which each part (^Z 0 ,..., ^r(l)) has at least one odd
component.
Proof. As in the proof of Corollary 1.2,

l 2 X r
~ r )

()()
at least one n< odd

hence, Q+(n) = (9(n) for every (n).

Cambridge Books Online © Cambridge University Press, 2010

204 Vector or Multipartite Partitions Chap. 12.3

In surveying our work on ordinary partitions we see that the properties of

partition ideals of order 1 presented in Section 8.3 were almost entirely
related to the fact that the relevant generating functions were infinite products.
We can, therefore, develop a corresponding theory for multipartite partitions.
For example, we can obtain Subbarao's generalization of Euler pairs for
multipartite partitions:
THEOREM 12.2. Let St and S2 be sets of positive integers. The number of
partitions of n into parts %(i) = (^ ( l ) ,..., £r(l)) in which £ / ° e St for all i
and 1 ^ j ' ^ r and where no part repeats more than t — 1 times always
equals the number of partitions of n into parts £ (l) with £jil)eSl9 1^
j ^ r, and some £jo(i) e S2 for all i if and only iftSi c Si and S2 = Si — tSt.

12.3 Bell Polynomials and Formulas for Multipartite Partition

Functions

To obtain useful formulas for computation, we shall utilize the well-known

Bell polynomials. Although this procedure will not produce any result even
approximating Corollary 1.7, it does produce effective computational for-
mulas, especially since there exist extensive tables of the Bell polynomials.
The Bell polynomials (first extensively studied by E. T. Bell) arise as an
aid to the task of taking the nth derivative of a composite function. Namely,
we hope to find a formula for the nth derivative of

h(t) = f(g(t)).
If we denote

= h = f

df "' dgn Jni

dtn

then we see that

3/20201

It is a simple matter to establish by mathematical induction that

K =/ia/»l(01»--,0n) + flZnligi*- >0n) +'"

+ fn<Xnn(9l,..,9n) (12.3.1)
where 0Lni(gu..., grn) is a h o m o g e n e o u s p o l y n o m i a l o f degree i in g l 9 . . . , gn.

Cambridge Books Online © Cambridge University Press, 2010

12.3 Bell Polynomials and Formulas for Multipartite Partition Functions 205

In light of this last observation, we see that the study of hn may be reduced
to the study of the Bell polynomials:

+ '-+*m(gi9...9gm). (12.3.2)

Note that Yn is a polynomial in n variables and the fact that g{ was originally
an /th derivative is not necessary in the consideration of (12.3.2). Finally, we
note that the choice of the/(f) as e* in (12.3.1) produces the simple formula

^ (12.3.3)

This formula is important for two reasons: First, it provides a nice recurrence
for the Bell polynomials (we now write D = d/dt):

(12.3.4)

Second, we obtain from (12.3.4) a concise expression for the generating

function of the Bell polynomials:

= t ^T" (12.3.5)
n\
The desired expression is

log # ( M ) = I = - ^ . (12.3.6)

To verify (12.3.6) we need only differentiate with respect to u and observe

that a comparison of the coefficients of un in the resulting equation produces an
identity equivalent to (12.3.4).
If we exponentiate (12.3.6) and expand the infinite product of exponential
functions on the resulting right-hand side, we obtain the following explicit
formula for the Bell polynomials:

Cambridge Books Online © Cambridge University Press, 2010

206 Vector or Multipartite Partitions Chap. 12.3

The Bell polynomials are useful in many problems in combinatorics; we

shall restrict our consideration to their application in multipartite partition
problems.
Let

< ? / * „ . . . , xr) = &j=l+ X P*(n;j)xr- 'Vr; (12.3.8)

(n)>0

J2/*!,...,*,) = J , = 1 + X Q(n;j)xl"*---n,"-9 (12.3.9)

(n)>0

F(u) = 1 + X ^ , t / ' ; (12.3.10)

G(M) = 1 4- f ^ w y . (12.3.11)

THEOREM T2.3

^ y = r/0!&(!), 1 !/Jr(2), 2!]8r(3), 3!/f r (4),..., (7 - IVAUWU

(12.3.12)

( - i y j y = r , < - 0!j?r(l), - l!/»r(2), - 2!/U3),..., - (; - l)\Pr(j))/jU

(12.3.13)
l 1 1
where ^(m) = UUi( ~ ^i ")" -
Proo/. The arguments establishing Eqs. (12.2.2) and (12.2.3) are easily
refined to show that

F(u) = n ( l - uxlnix2n2"'Xr"ryl (12.3.14)

and

G(u) = n (1 + ux^xt**- - -xrn'). (12.3.15)

O
Therefore

Y"1 V"1 U X^ l
X2 2
' ' ' Xr '

0 0
.i"!

m = l 'W

Z ~ ( m - \)%{m). (12.3.16)

Cambridge Books Online © Cambridge University Press, 2010

12.4 Restricted Bipartite Partitions 207

Equation (12.3.14) now follows from the fact that (12.3.16) is a special case
of (12.3.6), which is equivalent to (12.3.5). Next

l o g G ( - II) = - log F(u)

= - I ^ O n - 1)!W4 (12.3.17)
m=i m\
and this yields (12.3.13). •

To illustrate the usefulness of Theorem 12.3, we calculate &2: By (12.3.12)

+ j? r (l) 2 )

(12.3.18)

From here it is a reasonably simple matter to compute the actual series

expansion.

12.4 Restricted Bipartite Partitions

To conclude this chapter we shall consider certain restricted partitions of

bipartite numbers for which considerations similar to those used in Section 11.2
seem to work surprisingly well.
We shall look at n(n9 m), the number of partitions of (n, m) into "steadily
decreasing" parts, that is, partitions of the form

(n, m) = (nl9 mt) + (n 2 , m2) + • • • + (nr, mr) (12.4.1)

where min(ni9 mf) ^ max(nl + 1, m i + 1 ), i < i < r. Our main result (due to
L. Carlitz) is Corollary 12.5, which is a partition identity different from the
simple extensions of partition ideals of order 1 considered in Section 12.2.

THEOREM 12.4. For \x\ < 1, \y\ < 1,

)"Hl-*""Vr 1 (l-* 2 VT 1 . (12.4.2)

Cambridge Books Online © Cambridge University Press, 2010

208 Vector or Multipartite Partitions Chap. 12.4

Proof. Let n(a, b\n, m) denote the number of partitions of (n, m) of the
type (12.4.1) subject to the additional restriction min(a, b) ^ max(nl5 mx), and
write
7r fl
tab = Z ( ' ^lr> s)xrys.

Then clearly
min(n,m)

£,*,= I ^/^ (12.4.3)

r,s=0

where we have merely classified each partition with largest part max(r, 5) ^

Hence if
00

r=0

then by (12.4.3) (note that {Hm = £nn if m > n, £nm = ^mm if n ^ m)

00 n

r,s=0

r,s = 0 n = max(r,s)

n—r r^s
f
n=s
f
r= 0 n=r

+ (1 - yu) lf,(xyu)rZrr - ZixyuYZA

r=0 r=0 J

F^uxy). (12.4.4)
(1 - u X l -xu)(i-yu)

Iteration of (12.4.4) yields directly

(1 _ X2n+ly2n+lu2j

" - xnynu)(i - x" +1 /«Xl - xnyn+lu)

Cambridge Books Online © Cambridge University Press, 2010

Examples 209

Finally (by Abel's lemma):

00

£ n(n, m)xnym = lim irr

n,m = 0 r-»oo

= lim (1 - M)F 1 (M)

-n ~y)(l - xnyn~l)
1
JA (1 - x y )(\ - X"" VX1 " ^ V 1 )
2n 2n

Since the infinite product in (12.4.2) is clearly the generating function for
nx(n, m), the number of bipartite partitions of (n, m) in which all parts are
of the forms (2a, 2a) or (a — 1, a) or (a, a — 1), we obtain the following
result immediately.
COROLLARY 12.5. For all n and m, n(n, m) = n^n, m). •

Examples
1. The formula of Cay ley (referred to after Example 2 in Chapter 5)

(1 - q)(i - q2)- • -(1 - q') ^ I'W' - - i»>Pl\p2\-- -Pil

1

where the summation runs over all partitions (P»2 P2 3 P3 * • •) of i, is an imme-

diate corollary of Theorem 12.3.
2. The summatory maximum of n{,n2,...,nr(writtensmax(n1,n2,.. .,« r ))is
defined by smax(« 1 ,n 2 ,.. .,nr) = nY 4- n2 + • * * + nr — (r— I)min(n 1 ,n 2 ,.. .,nr).
The identity smax(n!, M 2 ,. .., nr) — m2i\(nl, n2,..., nr) is valid in general
only if r = 1 of 2.
3. The proof of Theorem 12.4 may be extended to prove that the number
of partitions of ( n ^ . . . , nr) in which the minimum coordinate of each part
is at least as large as the summatory maximum of the next part equals the
number of partitions of (nu..., nr) in which each part has one of the 2r — 1
forms (a + 1, a , . . . , a),(a, a + 1, a , . . . , a ) , . . . , ( a , a , . . . , a, a + l),(ar + 2,
ar + 2 , . . . , ar 4- 2), (ar + 3, ar + 3 , . . . , ar + 3 ) , . . . , (ar + r, ar + r , . . . ,
ar + r) (a ^ 0). The proof relies on the crucial observation that

1 1 1 - xyu2
1 — xu j 1 — yu i
(1 — xw)(l — yu)
is a special case of

Cambridge Books Online © Cambridge University Press, 2010

210 Vector or Multipartite Partitions Chap. 12

- 1+ - UXX)' • ' ( I — UXr)

1 — X 1 X 2 * * 'XrUr
( 1 — M X i X l — M X 2 ) ' ' ( 1 — UXr)

4. From (12.3.18) it follows easily that the number of partitions of

(n 1 n 2 - • -nr) into at most two parts is {_i(nt + i)(n2 + 1)* • (nr + 1) + i ] .
We note that such formulas for any P(n;;) are directly computable from
Theorem 12.3.
5. Application of Eq. (12.3.6) to Example 5 of Chapter 11 shows that the
generating function for plane partitions with trace n is
_^y, 0\q V.q2 2lq> (n -
n\
6. From Example 5 we see that the generating function for plane partitions
with trace 2 is

(l-^l-S2)2*
7. The number of partitions of (nl9..., nr) is the same as the number of
factorizations (order discounted) of the integer N = plrtlp2n2'' 'Prnr where
the Pi are primes. Thus there are four factorizations of 12: 12, 6 • 2,4 • 3, 3 • 2 • 2;
and there are four partitions of (2, 1): (2, 1), (1,1) + (1,0), (2,0) + (0,1),
(1,0) + (1,0) + (0,1).
8*. It is not difficult to show that

} l9 x2,...9xr)
is a polynomial in xu x 2 , . . . , xr; this follows from Theorem 12.3. Somewhat
surprising is B. Gordon's theorem that the coefficients of this polynomial
are all nonnegative.

Notes
MacMahon (1915-1916, 1917) was the first person to investigate multi-
partitite partitions in detail. An extensive account of recent work is given by
Cheema and Motzkin (1971). Theorem 12.1 is due to Cheema (1964) and
Theorem 12.2 is due to Subbarao (1971). The Bell polynomials (studied
extensively by Bell, 1934) are treated in greater detail by Riordan (1958) and
Comtet (1974); Eq. (12.3.7) is known as Faa di Bruno's formula. N. J. Fine
has done some very interesting work connected with sums like (12.3.7).
Theorem 12.3 goes back to MacMahon (1917); it has also been rediscovered
by Wright (1956). Theorem 12.4 is by Carlitz (1963a, b); related work is
done by Carlitz and Roselle (1966), Roselle (1966a, b), and Andrews (1976).
Reviews related to the material in this chapter are found in Section P64 of
LeVeque (1974).

Cambridge Books Online © Cambridge University Press, 2010

References 211

Example 1. MacMahon (1915-1916).

Examples 2, 3. Andrews (1976).
Example 8. Gordon (1963).
With regard to Example 8, we must mention the work of L. Solomon
(1977). He interprets (12.3.10) and (12.3.11) as Poincare series of certain
graded vector spaces associated with the symmetric group Sn on n letters,
and he is able to show that the polynomial in Example 8 is also a certain
Poincare series. This interpretation of these polynomials immediately implies
Gordon's theorem. Numerous other interesting relations between multi-
partitite partitions and group theory are explored in Solomon's paper.

References
Andrews, G. E. (1976). "An extension of Carlitz's bipartition identity," to appear.
Bell, E. T. (1934). "Exponential polynomials/' Ann. of Math. 35, 258-277.
Carlitz, L. (1963a). "Some generating functions," Duke Math. J. 30, 191-201.
Carlitz, L. (1963b). "A problem in partitions," Duke Math. J. 30, 203-213.
Carlitz, L., and Roselle, D. P. (1966). "Restricted bipartite partitions," Pacific J. Math.
19, 221-228.
Cheema, M. S. (1964). "Vector partitions and combinatorial identities," Math. Comp.
18, 414-420.
Cheema, M. S., and Motzkin, T. S. (1971). "Multipartitions and multipermutations,"
Proc. Symp. Pure Math. 19, 39-70.
Comtet, L. (1974). Advanced Combinatorics. D. Reidel, Dordrecht.
Fine, N. J. (1959). "Sums over partitions," Rep. Inst. Theory of Numbers (Boulder)
pp. 86-94.
Gordon, B. (1963). "Two theorems on multipartite partitions," /. London Math. Soc.
38, 459-464.
LeVeque, W. J. (1974). Reviews in Number Theory, Vol. 4. Amer. Math. Soc, Providence,
R.I.
MacMahon, P. A. (1915-1916). Combinatory Analysis, Vols. 1 and 2. Cambridge Univ.
Press, London and New York (reprinted by Chelsea, New York, 1960).
MacMahon, P. A. (1917). "Memoir on the theory of partitions of numbers, VII,"
Philos. Trans. Roy. Soc. London A217, 81-113.
Riordan, J. (1958). An Introduction to Combinatorial Analysis. Wiley, New York.
Robertson, M. M. (1962). "Partitions of large multipartites," Amer. J. Math. 84, 16-34.
Roselle, D. P. (1966a). "Generalized Eulerian functions and a problem in partitions,"
Duke Math. J. 33, 293-304.
Roselle, D. P. (1966b). "Restricted ^-partite partitions," Math. Nachr. 32, 139-148.
Roselle, D. P. (1974). "Coefficients associated with the expansion of certain products,"
Proc. Amer. Math. Soc. 45, 144-150.
Solomon, L. (1977). "Partition identities related to symmetric polynomials in several
sets of variables," to appear.
Subbarao, M. V. (1971). "Partition theorems for Euler pairs," Proc. Amer. Math. Soc.
28, 330-336.
Wright, E. M. (1956). "Partitions of multipartite numbers," Proc. Amer. Math. Soc. 7,
880-890.

Cambridge Books Online © Cambridge University Press, 2010

CHAPTER 13

Parfifions in Combinaforics

13.1 Introduction

The subject matter for this chapter could well fill a book by itself. Our
justification for an abbreviated treatment lies in the fact that our expressed
goal was to treat the term "partitions" primarily as "partitions of numbers."
Not surprisingly, partitions of numbers often are closely related to partition
problems in combinatorics. For this reason, we shall look at a few topics
where partitions of numbers play an important role: finite vector spaces,
partitions of sets, and symmetric functions. The first topic mentioned has
been studied extensively in recent years; we shall present a simple but fun-
damental result of Knuth that relates partitions to the combinatorics of both
finite vector spaces and finite sets. Partitions of sets have been neatly related
to symmetric functions recently by P. Doubilet and we shall introduce this
work in Sections 13.3 and 13.4.

13.2 Partitions and Finite Vector Spaces

We recall from Chapter 3 (Theorem 3.1) that the Gaussian polynomial

\N + M ]

is the generating function for p(N, M, n), the number of partitions of n into
at most M parts each not exceeding N. Surprisingly (at first), these polynomials
arise in the study of finite vector spaces:

THEOREM 13.1. Let Vn(q) denote the vector space of dimension n over the
finite field GF(q) of q ( = a prime power) elements. Then there are [JJJ sub-
spaces of Vn(q) of dimension m.

Remark. We present two proofs: The first is fast and elementary. The second
shows explicitly the relationship between partitions and finite vector spaces.

ENCYCLOPEDIA OF MATHEMATICS and Its Applications, Gian-Carlo Rota (ed.).

2, George E. Andrews, The Theory of Partitions
Copyright © 1976 by Addison-Wesley Publishing Company, Inc., Advanced Book Program.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical photocopying,
recording, or otherwise, without the prior permission of the publisher.

212

Cambridge Books Online © Cambridge University Press, 2010

13.2 Partitions and Finite Vector Spaces 213

First proof of Theorem 13.1. To determine the w-dimensional subspaces

°f Vn(Q)> w e first determine all possible m-tuples ( v h v 2 , . . . , v m ) o f m linear-
ly independent vectors. Such m-tuples may be chosen as follows: vj may be
any of the q" - 1 nonzero elements of Vn(q); v2 may be any of the qn - q
vectors lying outside the subspace spanned by v,; v3 may be any of the
qn - q2 vectors lying outside the subspace spanned by {v,, v 2 }, and so on.
Hence the m-tuple (v,, v 2 , . . . , vm) may be chosen in
(qn - \)(q» - q ) ( q » - q * y - - ( q » -q»>-i)

ways.
Now each such m-tuple (v,, v 2 ,..., vm) spans an m-dimensional subspace
of Vn(q); however, several may span the same subspace. In fact, precisely

(qm- l)(qm-q)'"(qm -qm~l)

span the same subspace since this number is the number of ways of choosing
a subset of m linearly independent elements from Vm(.q). Hence the number
of m-dimensional subspaces of Vn(q) is

(<?"- l)(qn-q)-(qn-qm-t)
m
(q -

- <D

m
Second proof of Theorem 13.1. We choose a fixed basis for Vn(q), say
uuu2,.. .,un. Then we know from linear algebra that each subspace of
dimension m has a canonical basis vt, v2,..., vm given by

C
»i = t UUJ = (C.-i, Ci29..., Cin) (13.2.1)

where Ciri = 1, Cu = 0 for j > rh Csn = 0 for s < i, for 1 ^ i ^ m,

n ^ rl > r2 > • • • > rk ^ 1. For clarity let us consider n = 9, m = 5, r t = 8,
r
2 = 7, r3 = 5, r 4 = 3, r5 = 2:

Ux = ( C 1 1 > 0 , 0 > C 1 4 > 0 , C 1 6 , 0,1,0),

U2 = ( C 2 1 , 0 , 0 , C 2 4 > 0 , C 2 6 , 1,0,0),

U3 = ( C 3 1 , 0 , 0 , C 3 4 , 1 , 0 , 0 , 0 , 0 ) ,

Cambridge Books Online © Cambridge University Press, 2010

214 Partitions in Combinatorics Chap. 13.3

l/ 4 = (C 4 1 ,0, 1,0,0,0,0,0,0),

U5 = (C 51 , 1 , 0 , 0 , 0 , 0 , 0 , 0 , 0 ) .

Now the positions of the undetermined Cu in the foregoing array correspond

to the graphical representation or Ferrers graph

of the partition 3 + 3 + 2 + 1 + 1 of 10. Thus since each undetermined

Cu may be chosen in q ways, there are q10 subspaces of V9(q) with this form
of canonical basis. Indeed we see that to each partition of r with at most m
parts and largest part at most n — m we may by the foregoing procedure
produce qr different subspaces of dimension m. Therefore the total number
of subspaces of dimension m is

by Theorem 3.1. •

There are many other theorems relating Gaussian polynomials to finite

vector spaces; we mention a few in the examples.

13.3 Partitions of Sets

Let us begin with the simple question: How many ways can a set of n elements
be split up into a set of disjoint subsets, that is, how many "partitions" are
there of n = { 1 , 2 , . . . , «}?
The answer to this question is called the nth Bell number Bn (after Eric
Temple Bell). Table 13.1 gives the first few Bell numbers and the related
partitions.

Table 13.1

n Partitions of {1, 2 , . . . , n) Bn

0 <f> 1
1 {1} 1
2 {1,2};{1},{2} 2
3 { 1 , 2, 3 } ; {1, 2 } , { 3 } ; { 1 , 3}, { 2 } ; {2, 3}, { 1 } ; {1}, {2}, {3} 5

Cambridge Books Online © Cambridge University Press, 2010

13.3 Partitions of Sets 215

We remark that if n = {Ply / ? 2 , . . . , f$m} where each /?f c {1, 2 , . . . , n} = n,

then 7r is a partition of n provided (J7=i /?,- = n and /?,- n j87- = <f> (the empty
set) whenever / # 7. The /?,- are called the blocks of TT. Notice that there is a
unique partition (in the sense of Chapter 1) of the number n associated with
each 7T, namely, the partition X(n) whose parts are |/?J, |/? 2 |, | / J 3 | , . . . , |/?J
where 1)5,1 is the number of elements of the block /?,..
For simplicity of notation we make the following conventions concerning
the partition A = (nln2"-nj = (l r i 2 r 2 3 r 3 - • •) of the number n:

k\ = nl\n2\n3\'-nm\ = 1 !r»2!ri3 !r3 • • •,

W = r1!r2!r3!-",
Sign A = ( - l)'2+2r3 + 3r 4 +- = ( _ iy~rl-r2-r3-•- = ^_ 1 ^(A)-#(A)

THEOREM 13.2. For each partition A of H given by (1 A| 2 A2 3 A3 - • mAm) there

are exactly

n! n\

partitions 7i of n for which X(n) = A.

Proo/. We begin with the standard interpretation of the multinomial

coefficient, namely, that there are

—r^r i=

ordered 7-tuples of subsets (Su...,Sj) that are mutually disjoint, whose

union is {1, 2 , . . . , n}, and where |S,| = n,. To see this we note that the set St
can be chosen in („") ways and the remaining) — 1 sets can then be chosen in

n —

ways. Consequently

nl9 n2,..., rij/ \nt/ \n2

_ ln\tn

Cambridge Books Online © Cambridge University Press, 2010

216 Partitions in Combinatorics Chap. 13.3

n\/n - ni
nj\ n2
n\
(13.3.2)

Hence if there are Xx singleton subsets, X2 two-element subsets, and so on,

then the number of ordered^! -f X2 + • • • + Am)-tuples of subsets partitioning
n is

(13.3.3)

Several of these "ordered" partitions may correspond to the same partition

of n; in fact, the one-element subsets may be permuted Xt\ ways, the two-
element subsets X2! ways, and so on; consequently, the number of partitions
n of n with X(n) = X is just the number in (13.3.3) divided by X^\X2\- • -Am!,
and this is precisely the assertion of this theorem. •

THEOREM 13.3. The following formulas hold for the Bell numbers:

Bn = Yn(l,\,...,l) (13.3.5)

(where Yn(gu g2,..., gn) is the Bell polynomial defined in (12.3.2));

n!
Proof. We shall prove (13.3.4) using a simple combinatorial argument. The
other identities follow directly from properties of the Bell polynomials
derived in Chapter 12.
There are Bn+l partitions of n + 1 = {1, 2 , . . . , n -h 1}. Now n + 1 lies
in a block of size k + 1 where 0 ^ k ^ n, and there are clearly (£) choices
for this block. Once this block is chosen, the remaining set of n — k numbers
may be partitioned in Bn_k ways. Hence, summing over all admissible k, we
see that

which is (13.3.4).
Equation (13.3.4) together with Bo = 1 uniquely determines the Bell
numbers. By (12.3.4) we see that Yn(l, ! , . . . , ! ) satisfies the same recurrence,

Cambridge Books Online © Cambridge University Press, 2010

13.3 Partitions of Sets 217

and since r o ( l , 1 , . . . , 1) = 1, it follows that Bn = Yn(l, 1 , . . . , 1) for all n.

Hence (13.3.5) is valid.
Finally by (12.3.6),

^ B^X_ ^ /nVi,i,...,

which is (13.3.6). •

For our applications to symmetric functions in Section 13.4 we must briefly

consider the lattice Ln of partitions of the set {1, 2 , . . . , n) = n. Our ultimate
object is to obtain Theorem 13.9, which provides an explicit computational
formula for the Mobius function of this lattice.
First we define a partial order on Ln by saying nl ^ n2 whenever each
block of 7i2 is contained in a block of nv. It is a simple matter to show that
under this partial order Ln forms a lattice. The union (or join) ni v n2 is the
coarsest common refinement of nl and n2, namely, i and j are in the same
block of nx v n2 if and only if they are in the same block in nl and in the same
block in n2. The intersection (or meet) n^ A n2 is, on the other hand, defined
by the condition that i and j are in the same block if and only if there exists
a chain i = i0, ii9 i2y..., ir = j such that for 0 < m < r, im and fm+1 are
in the same block of nl or n2.
The incidence algebra is the algebra of real-valued functions of two
variables whose domain is Ln x Ln where addition is standard addition of
functions, and multiplication is defined by the convolution:

/ fn\(-m rr \ \~* f(rr rr\n( yr rr \ (\ \ % H\

The unit of this algebra is clearly the Kronecker S-function: 5(nlf n2) = 1 if
nl = n2 and S(nly n2) = 0 if nl ^ n2. Another fundamental function is the
zeta function: t,(nu n2) = 1 if nx ^ 7r2, (,(nu n2) = 0 if nt ^ n2.
The following two results are phrased for the lattice of partitions Ln\
however, they are valid in a much more general setting where Ln is replaced
by any locally finite partially ordered set.

LEMMA 13.4. The zeta function C(^i, ^2) possesses an inverse in the incidence
algebra (to be called the Mobius function n(nl9 n2)).

Proof We proceed by induction on the number of elements in the segment

Cambridge Books Online © Cambridge University Press, 2010

218 Partitions in Combinatorics Chap. 13.3

If the segment has only one element, then clearly nt = n2 and ^i(nli 7rt) = 1.
If fi(nl9 n2) has now been defined whenever [n^ 7r2] has less than k elements,
we see that if C^i» ^i\ has k elements, then

fi(nu n2) = — YJ M^I, TT),

and so fi(nl9 n2) is uniquely determined. •

THEOREM 13.5 (Mobius inversion). Letf(n) and g(n) be real-valued functions

with domain Ln, Then

g(*o) = Z /(*) for all n0 e Ln (13.3.8)

if and only if

/(^o) = Z 9(nMn> TTO) for all n0 e Ln, (13.3.9)

Furthermore

9(no) = Z /C71) for all noeLn (13.3.10)

if and only if

for all noeLn, (13.3.11)

Remark, Other names for the Mobius inversion process are "the generalized
inclusion-exclusion principle" and "generalized sieve methods." When the
lattice under consideration is not Ln but the lattice of subsets of a finite set
ordered by "inclusion," the Mobius inversion process is precisely classical
inclusion-exclusion.

Proof, Each of the four implications asserted in the theorem is proved in the
same manner. We shall therefore only prove that (13.3.8) implies (13.3.9).
Assuming (13.3.8), we see that

£ g(n)n(n9 n0) = £ £ f{n')ii(n, n0)

= Z /(*') Z l*(n> no)

it' < «o it' < it < wo

(«W, «o) = /(«o)-

Cambridge Books Online © Cambridge University Press, 2010

13.3 Partitions of Sets 219

Our next result provides the necessary groundwork for the computation
of /x(0, /) where 0 is the minimal element of Ln (0 = [{1, 2 , . . . , n}] and / is
the maximal element of Ln (/ = [{1}, { 2 } , . . . , {AI}]).
DEFINITION 13.1. Let (f> be a function from n to a set C. We form a partition
n by putting i and j in the same block if and only if (p(i) = <t>(J); n is called
the kernel of </>, and we write Ker </> = n.
LEMMA 13.6. Let n e Lni and let C be a set with X elements. Letf(n) denote
the number of functions <f> from n to C with Ker (f> = n, then

f(n) = X(X - 1)- • -(X - vOr) + 1) = (X)y(n) (13.3.12)

where v(n) is the number of blocks of n.

Proof. The elements of the first block may map onto any one of the X
elements of C; the elements of the second block may map onto any one of
the remaining X — 1 elements of C, and so on. Hence, the total number of
functions admissible is

x(x - i)(x - 2)- • -(X - v(;r) + i) = (x\iK). m

COROLLARY 13.7. For each n0 e Ln

X 0 = Xv<*°>; (13.3.13)

X X'<*V(;r, 7t0) = (X)Hxo). (13.3.14)

Proof. Equation (13.3.14) follows from (13.3.13) by Theorem 13.5. To see

(13.3.13) we count in two ways the number of functions from n to C that
remain constant on the blocks of n0. Clearly there are Xv{no) such functions.
On the other hand, we may split these functions up according to their kernel n.
Now the only admissible kernels n are those partitions which are "refined" by
7r0, that is, those n such that n ^ n0. Hence, by Lemma 13.6, the number
of functions from {1, 2 , . . . , n} to C that remain constant on blocks of n is
(X)v(n)\ therefore
£ (X)viK) = *'<*<>>,
x<xo

which establishes (13.3.13) and completes our proof. •

We now introduce the well-known Stirling numbers of the first kind

s(n, k), and Stirling numbers of the second kind S(n, k):

= £ s(n,k)Xk; (13.3.15)

Cambridge Books Online © Cambridge University Press, 2010

220 Partitions in Combinatorics Chap. 13.3

n
Yn V <\(n lrVY\ (X'X 1 1 f\\
j\. — ? kj^fl, /Cy^A jj^. \LD.J.L\J)
fc = 0

THEOREM 13.8.

X 1 = S(n f k); (13.3.17)

v(n) = k

£ n(nj) = s(n,/c); (13.3.18)

v(«) = fc

^(0, /) = ( - l)n-\n - 1)!. (13.3.19)

Proof. From (13.3.13) with n0 = /, and thus v(7r0) = n, we see that

X" = I (X)v(It) = I (X), / I A, (!3-3-20)

*eLn k=0 I 7teLn I
\v(tc) = fc /

but the S(n, k) are clearly uniquely determined by (13.3.16) and so since
(13.3.20) is of precisely the same form, we conclude that (13.3.17) is valid.
Equation (13.3.18) follows from (13.3.14) and (13.3.15) in exactly the same
way.
Finally, since there is only one single block partition of n = {1, 2 , . . . , n)
(namely, 0), we see that by (13.3.18)

MO, /) = 5(n, 1)

= coefficient of X in X(X - 1)- • -(X - n 4- 1)

= ( - iy-l(n- 1)!,

which is (13.3.19). •

THEOREM 13.9. Suppose nu n2 e Ln, nv ^ n2. We define the partition

Kn\> n2) = (l ri 2 r2 3 r3 - • -mrm), where r, is the number of blocks of nt made
up of exactly i blocks of n2. Then
m

ti*u *i) = ( - l)r'<>0 * ft 0 " 0! r '. (13.3.21)

Proof. Inspection of the segment [rc^ 7r2] reveals that it is isomorphic

to the direct product of rl copies of Ll% r2 copies of L 2 , . . . , and rm copies
of Lm. It is not difficult to see that the Mobius function on the direct product
of partially ordered sets is just the product of the Mobius functions. Hence
we obtain (13.3.21). •

Cambridge Books Online © Cambridge University Press, 2010

13.4 The Combinatorics of Symmetric Functions 221

13.4 The Combinatorics of Symmetric Functions

If we refine the considerations of Section 13.3 concerning functions whose

domain is a set of n elements, we are immediately led to the elementary theory
of symmetric functions. This approach to symmetric functions not only
provides us with a simple proof of the fundamental theorem of symmetric
functions (Theorem 13.12), but, more importantly, yields quite computable
formulas for the coefficients in many classical symmetric function identities
(see Theorems 13.11 and 13.12).
We now consider the set F of functions from n = {1, 2, 3 , . . . , n} to a set
C that is countably infinite:

The elements of C are to be viewed as the indeterminates for the ring of

polynomials with countably many variables over the reals.

DEFINITION 13.2. For e a c h / e F, the generating function y(f) is defined by

den i=l

and if T c F, we define

Our initial work concerns three special subsets of F:

n}, (13.4.1)

7r}, (13.4.2)

s/n = {/e F|Ker/ v n = /}, (13.4.3)

and we let kn = y{Xn\ sn = y(SfK), an = y(s/n).
Notice that the subscript n on each of kn, sn, and an refers to a partition of
the set n. The following definition concerns subscripts that are partitions of
integers.

DEFINITION 13.3. For each integer n and each partition X = (A1^2^3* * *^m)
of n we define

= sJLlsX2'-sXm.

Cambridge Books Online © Cambridge University Press, 2010

222 Partitions in Combinatorics Chap. 13.4

The next theorem clarifies the relationship between functions whose sub-
script is a partition of n and those whose subscript is a partition of an integer.
THEOREM 13.10. For each partition n ofn,

K = |A(7r)|/cA(ff); (13.4.4)

Sit = Sx(n)l (13.4.5)

an = Kn)\aXin). (13.4.6)

Proof. We write

First we examine kn: I f / e kn, then Ker/ = n, and so

Furthermore it is clear that rx \r2lr3\- • • = |A(TT)| elements of kn have exactly

this generating function. Consequently

Next we treat Sfn\ From (13.4.2) we see that fe£fnif and only if its kernel
is less refined than 7r, that is, if and only if/ is constant on blocks of n. Hence
if the blocks of n are Bl9 B2, B 3 , . . . , Bm, then

m
=
n s i B i i = su*

Cambridge Books Online © Cambridge University Press, 2010

13.4 The Combinatorics of Symmetric Functions 223

Finally we consider s/n: For each fes/n, we have K e r / v n = /; this

means that no two numbers in the same block of/can be in the same block
of 7i as well. The latter formulation is equivalent to saying that/is one-to-one
on the blocks of n. Therefore if the blocks of n are Bt, B2, B3,..., Bm, then

= I y(f\BMf\B2)'"y(f\Bj

-n/ i */)
i= 1 I onc-to-onc
\ functions
\ f.Bi-C

N o w e a c h y ( / ) in this last expression is o f the form xj{xJ2' • • x / | B | | , and e a c h

such term arises from \Bk\\ such functions. H e n c e

1=1

= Kn)\aX(ny •
Our next result will be seen to imply easily the fundamental theorem on
symmetric functions (Theorem 13.12).
THEOREM 13.11. For each partition n of n,

k. = £ n{o, n)sa, (13.4.7)

K = Z £ ^ Z M<T, TK. (13.4.8)

Proo/. We begin by observing from (13.4.1), (13.4.2), and (13.4.3) that

*« = £ * . ; (13A9)

aK= Z *-• (13.4.10)

<r:<r v « = /

Equation (13.4.9) can be easily transformed via Mobius inversion (Eq.

(13.3.8)):

which is (13.4.7).

Cambridge Books Online © Cambridge University Press, 2010

224 Partitions in Combinatorics Chap. 13.4

The second portion of this theorem is somewhat harder, requiring a double

application of Mobius inversion: By (13.4.10)

<*n = Z S(a v n
* 7)fc«r

(
a \x > a v n

Hence by Mobius inversion

[i(n, /) Z ^ = Z M^» ^)at,

or

which is permissible since /x(7r, /) ^ 0. Applying Mobius inversion to (13.4.11),

we see that

which is (13.4.8). •

THEOREM 13.12 (Fundamental theorem on symmetric functions). Every

polynomial symmetric function in the variables xt, x2, * 3 , . . . is actually
a polynomial in au a2, a3,... .

Proof. Let p(x1? x 2 , x 3 , . . . ) be a polynomial symmetric in the x,. Then

p(x1,x2,x3,...)

= C
2- «i«2 •n m ' C (ni,n2 n m)

(where each K is some partition of some n (n < iVt + • • • + Nm) with m

blocks of size n{> n2,..., «m)

Cambridge Books Online © Cambridge University Press, 2010

Examples 225

, 7T) v

n 2L /'(

and since the aMx) are products of the ah we have the desired result. •

Examples

1. If we consider a vector space Vn(q) over GF(q) of dimension /i, and

count linear transformations / of Vn(q) into a space ./* over GF(q) of X
elements subject to the condition j(Vn(q)) n Jf = (0), where J (with Z
elements) is a subspace of ./\ then it is easy to show that the number of such
transformations is
(X - Z)(X - Zq)"(X - Zqn~{).

2. If in Example 1 we require further that the linear transformations / must

be such that / ( V n ( q ) ) n ¥ is of dimension k, where Z c F c f with Z , Y, and
X elements respectively, then the number of such / is

(Y - - Zq)--(Y- Zqk~l)(X - Y)(X - Y q ) - - ( X - Yq"

3. Examples 1 and 2 imply that

f] (X - Zq{) = t U - Yqh).

4. Example 3 may also be deduced from Corollary 2.4.

5. The linear operator L on polynomials in X over the complex numbers
given by

U(X - \)(X -

satisfies

L(Xn) =

6. Let yjq) be an m-dimcnsional subspace of Vm + n(q). By counting the

/i-dimcnsional subspaccs of Vm + n(q) that intersect Vm{q) in a k-dimensional
space, we may prove that

m k{m-h + k) _
m 4- n
h - it h

7. F o r a n y s e q u e n c e o f i n t e g e r s al% (/,,. . ., an s u c h that a ^ . x - a{ = 0 o r

for 1 ^ / < n% it i s true t h a t

Cambridge Books Online © Cambridge University Press, 2010

226 Partitions in Combinatorics Chap. 13

k\[ n- k
8. F r o m ( 1 3 . 3 . 1 6 ) it is easy to d e d u c e that

S(n + 1, k) = S(ny k - 1) + fcS(w, k).

9. F r o m ( 1 3 . 3 . 1 6 ) and the trivial identity

Xn+l = X(i 4- ( X - l)) n ,

it is easy to s h o w that

y=0

10. F r o m E x a m p l e 8, it f o l l o w s that if Sk(t) = Y,T=o S(w + fc, /c)f\ then

1 - kt

Hence

1
(1 - 0 ( 1 - 2/)- • • ( ! - f c t )

11. For e a c h partition / o f n, we consider ax = ^t-n ^. F r o m T h e o r e m

13.11 w e can show that CMx = C ^ .
12. For every pair o f positive integers n and /c, each partition o f n\k into
parts each not e x c e e d i n g n is actually a s u m o f A: partitions o f n\. For e x a m p l e ;
there are 37 partitions o f 18 = 3!3 into parts each ^ 3 :

(36) = (32) + (32) + (32) (29) = (23) + (23) + (23)

( 1 2 3 s ) = (123) + ( 3 2 ) + ( 3 2 ) ( 1 3 2 3 3 3 ) = ( 1 3 3 ) + ( 2 3 ) -f ( 3 2 )

( I 3 3 5 ) = (133) + (32) + (32) (152233) = (133) + (1222) + (32)

(2334) = (23) + (32) + (32) ( 1 7 2 3 3 ) = ( I 6 ) + (123) -h ( 3 2 )

(122234) = (1222) + (32) + (32) (1933) = (I6) + (133) + (32)

(14234) = (142) + (32) + (32) (2632) = (23) + (23) + (32)

(1634) = (I6) + (32) + (32) (122532) = (1222) + (23) + (32)

(12433) = (123) + (23) + (32) ( 1 4 2 4 3 2 ) = ( 1 4 2 ) + ( 2 3 ) -f ( 3 2 )

(182232) = ( I 6 ) + (1222) + (32) ( I 6 2 3 3 2 ) = ( l f t ) 4- ( 2 3 ) + ( 3 2 )

( 1 1 O 2 3 2 ) = ( I 6 ) -f ( 1 4 2 ) -h ( 3 2 ) ( 1 2 2 8 ) = ( 1 2 2 2 ) 4- ( 2 3 ) + ( 2 3 )

( l l 2 3 2 ) = ( l b ) 4- ( I 6 ) + ( 3 2 ) ( 1 4 2 7 ) = ( 1 4 2 ) 4- ( 2 3 ) 4- ( 2 J )

Cambridge Books Online © Cambridge University Press, 2010

Notes 227

( 1 2 7 3 ) = ( 1 2 3 ) 4- ( 2 3 ) 4- ( 2 3 ) (1626) = (I6) + (23) 4 (23)

( i 3 2 6 3 ) = ( 1 3 3 ) 4- ( 2 3 ) 4- ( 2 3 ) (1825) = (I6) + (1222) 4 (23)
( 1 5 2 5 3 ) = ( 1 3 3 ) + ( 1 2 2 ) 4- ( 2 3 ) ( 1 1 O 2 4 ) = ( I 6 ) 4- ( 1 4 2 ) -h ( 2 3 )
( 1 7 2 4 3 ) - ( I 6 ) + ( 1 2 3 ) 4- ( 2 3 ) ( 1 I 2 2 3 ) = ( I 6 ) 4- ( I 6 ) 4- ( 2 3 )
( 1 9 2 3 3 ) = ( I 6 ) 4- ( 1 3 3 ) 4- ( 2 3 ) ( l i 4 2 2 ) = ( I 6 ) + ( I 6 ) 4- ( 1 2 2 2 )
( l l l 2 2 3 ) = ( I 6 ) 4- ( 1 4 2 ) 4- ( 1 2 3 ) ( 1 I 6 2 ) = ( I 6 ) 4- ( I 6 ) + ( 1 4 2 )
( 1 1 3 2 3 ) = ( I 6 ) 4- ( I 6 ) 4- ( 1 2 3 ) ( I 1 8 ) = ( I 6 ) 4- ( I 6 ) 4- ( I 6 )
( 1 1 5 3 ) = ( I 6 ) 4 ( I 6 ) 4- ( 1 3 3 )

Notes

The material in Section 13.2 was taken from D. E. Knuth (1971). The
relationship between {/-series and finite vector spaces has an extensive lit-
erature; the beginnings of this subject may be traced to Jordan (1870),
Landsberg (1893), and Dickson (1901). Among the recent works in this area
we mention Carlitz (1954), Carlitz and Hodges (1955a, b), Hodges (1964,
1965), Fulton (1969), Rota and Goldman (1969, 1970), and Andrews (1971).
Reviews of recent work in this area are given in Section T35 of LeVeque
(1974).
The literature on the partitions of finite sets is extensive; see Rota (1964a).
We have taken most of our work from Rota (1964a), Rota and Frucht (1965),
and Doubilet (1972).
Apart from the examples we have considered there are many other applica-
tions of partitions in pure and applied mathematics. Some of the applications
to particle physics are given in Temperley (1952) and Bohr and Kalckar
(1937). Partitions as applied to the representation theory of the symmetric
group have already been mentioned in Chapter 11; see also Littlewood (1950),
Robinson (1961), and Rutherford (1947).
The partitions of n also possess a lattice structure relative to the partial
ordering: (/,/ 2 * ' ''-*) ^ 0-\i>2'' */./) when X;=i /V ^ Z)=i V ^ or e a c n '•
1 ^ / ^ s. This lattice has been extensively treated by Brylawski (1973) and
has been applied by Snapper (1971) to problems in group theory.
Attention should also be drawn to Stanley (1972), who has extended the
partition analysis of MacMahon (1916) to numerous combinatorial problems
through the use of(P, (o)-partitions.
Examples 1, 2. Rota and Goldman (1970).
Example 5. Rota and Goldman (1969).
Examples 6, 7. Bender (1971). Example 6 is the {/-analog of the Chu-
Vandermonde summation and is a special case of Corollary 2.4. Example 7
is an extremely surprising and unusual generalization o( Example 6.
Example if. Doubilet\\972).
Example 12. Knutson (1972).

Cambridge Books Online © Cambridge University Press, 2010

228 Partitions in Combinatorics Chap. 13

References
Andrews, G. E. (1971). "On the foundations of combinatorial theory, V: Eulerian
differential operators," Studies in Appl. Math. 50, 345-375.
Auluck, F. C. (1951). "On some new types of partitions associated with generalized
Ferrars graphs," Proc. Cambridge Phil. Soc. 47, 679-686.
Auluck, F. C , and Kothari, D. S. (1946). "Statistical mechanics and the partition of
numbers," Proc. Cambridge Phil. Soc. 42, 272-277.
Bender, E. A. (1971). "A generalized ^-binomial Vandermonde convolution," Discrete
Math. 1, 115-119.
Bohr, N., and Kalckar, F. (1937). "On the transmutation of atomic nuclei by impact
of material particles," Kgl. Danske Vidensk Selskab Mat. Fys. Medd. 14, No. 10.
Brylawski, T. (1973). "The lattice of integer partitions," Discrete Math. 6, 201-219.
Carlitz, L. (1954). "Representations by quadratic forms in a finite field," Duke Math. J.
21, 123-137.
Carlitz, L., and Hodges, J. H. (1955a). "Representations by Hermitian forms in a finite
field," Duke Math. J. 22, 393-405.
Carlitz, L., and Hodges, J. H. (1955b). "Distribution of bordered symmetric, skew and
hermitian matrices in a finite field," J. reine angew. Math. 195, 192-201.
Dickson, L. E. (1901). Linear Groups with an Exposition of the Galois Field Theory.
Teubner, Leipzig (reprinted by Dover, New York, 1958).
Doubilet, P. (1972). "On the foundations of combinatorial theory, VII: Symmetric
functions through the theory of distribution and occupancy," Studies in Appl. Math.
51, 377-396.
Fulton, J. D. (1969). "Symmetric involutory matrices over finite fields and modular
rings of integers," Duke Math. J. 36, 401-407.
Hodges, J. H. (1964). "Simultaneous pairs of linear and quadratic matrix equations
over a finite field," Math. Z. 84, 38-44.
Hodges, J. H. (1965). "A symmetric matrix equation over a finite field," Math. Nachr.
30, 221-228.
Jordan, C. (1870). Traite dessubstitutions et des equationsalgebrique. Gauthier-Villars, Paris.
Knuth, D. E. (1971). "Subspaces, subsets, and partitions," /. Combinatorial Theory
A-10, 178-180.
Knutson, D. (1972). "A lemma on partitions," Amer. Math. Monthly 79, 1111-1112.
Landsberg, G. (1893). "Ober eine Anzahlbestimmung und eine damit zusammen-
hangende Reihe," /. reine angew. Math. Ill, 87-88.
LeVeque, W. J. (1974). Reviews in Number Theory, Vol. 5. Amer. Math. Soc, Providence,
R.I.
Littlewood, D. E. (1950). The Theory of Group Characters and Matrix Representations
of Groups, 2nd ed. Oxford Univ. Press, London and New York.
MacMahon, P. A. (1916). Combinatory Analysis, Vol. 2. Cambridge Univ. Press, London
and New York (reprinted by Chelsea, New York, 1960).
Robinson, G. de B. (1961). Representation Theory of the Symmetric Group. Univ. of
Toronto Press, Toronto.
Rota, G.-C. (1964a). "The number of partitions of a set," Amer. Math. Monthly 71,
498-504.
Rota, G.-C. (1964b). "On the foundations of combinatorial theory, I: Theory of Mobius
functions," Z. Wahrscheinlichkeitstheorie und verw. Gebiete 2, 340-368.
Rota, G.-C, and Frucht, R. (1965). "Polinomios de Bell y particiones de conjuntos
finitos," Scientia: Rev. Ci. Tecnica {Chile) No. 126, 5-10.

Cambridge Books Online © Cambridge University Press, 2010

References 229

Rota, G . - C , and Goldman, J. (1969). "The number of subspaces of a vector space,"

Recent Progress in Combinatorics (W. Tutte, ed.), pp. 75-83. Academic Press,
New York.
Rota, G . - C , and Goldman, J. (1970). "On the foundations of combinatorial theory, IV:
Finite vector spaces and Eulerian generating functions," Studies in Appl. Math.
49, 239-258.
Rutherford, D. E. (1947). Substitutional Analysis. Edinburgh Univ. Press, Edinburgh
(reprinted by Hafner, New York, 1968).
Snapper, E. (1971). "Group characters and nonnegative integral matrices," / . Algebra
19, 520-535.
Stanley, R. P. (1972). "Ordered structures and partitions," Mem. Amer. Math. Soc. 119.
Temperley, H. N. V. (1952). "Statistical mechanics and the partition of numbers, II:
The form of crystal surfaces," Proc. Cambridge Phil. Soc. 48, 683-697.
Wright, E. M. (1968, 1971, 1972). "Stacks, I, II, III," Quart. J. Math. 19, 313-320;
22, 107-116; 23, 153-158.

Cambridge Books Online © Cambridge University Press, 2010

CHAPTER 14

Computations for Partitions

14.1 Introduction
In most of the applied and many of the theoretical aspects of partitions we
are interested in actually enumerating or perhaps completely exhibiting a
set of partitions of n subject to certain conditions. There are certain elementary
algorithms to apply that generally do the job if n is small; these we discuss
in the next section.
The parity of p(n) has been of interest over the years, and we describe in
Section 14.3 the most effective algorithm known for determining p(n) modulo
2, as well as other algorithms that derive from generating functions.
Computations for higher-dimensional partitions are considered in Section
14.4; here we include the work of D. E. Knuth, which is the only quite general
simplification known. Sections 14.5-14.7 give reasonably short tables related
to well-known partition functions. In Section 14.8, we present a guide to more
extensive tables.

14.2 Elementary Algorithms

The simplest way to enumerate all the partitions of n is to write them in
lexicographic order. To proceed from (Als A2,. . ., As) to the next partition in
this order we note that
if As > 1, the next partition is (A1? A 2 ,.. ., / 5 - i , As — 1, 1); (14.2.1)
if As_r = c > 1 but A s _ r+1 = As_r + 2 = • • • = A, = 1,
then the next partition is obtained by replacing
A s . r , A s _ r + 1 , . . . , A, by (c - 1), (c - 1),..., (c - 1), d
where 0 < d ^ c — 1, and the number a of appearances
of (c — 1) is so chosen that a(c — 1) + d = c + r = As_r
+ A s _ r+1 + • • • + Ar (14.2.2)
The following computer program in FORTRAN IV utilizes the foregoing
algorithm to compute all the partitions of each integer not exceeding 20:

ENCYCLOPEDIA OF MATHEMATICS and Its Applications, Gian-Carlo Rota (ed.).

2, George E. Andrews, The Theory of Partitions
Copyright © 1976 by Addison-Wesley Publishing Company, Inc., Advanced Book Program.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical photocopying,
recording, or otherwise, without the prior permission of the publisher.

230

Cambridge Books Online © Cambridge University Press, 2010

 Elementary Algorithms 231

DIMENSION IP(20)
DO 80 N = 1,20
WRITE (6,5) N
5 FORMAT (110)

DO 10 L = 2,20
10 IP(L) = 0
15 CONTINUE
WRITE (6,20) IP(1), IP(2), . . . , IP(20)
COMMENT IP(J) is the Jth part of the partition of N just computed.
20 FORMAT (2013)
J=l
25 CONTINUE
IF (IP(J)-1)35,35,30
30 J = J+1
GO TO 25
35 CONTINUE
IF (J-1)75,75,40
40 M= J-2
K= N
IF (M) 45,55,45
45 DO50L=l,M
50 K = K-IP(L)
55 IQ = IP(J-1)

60 D O 6 5 L Q = 1 , J Q
65 IP(J-2 + LQ) = I Q - l
IP(J-1+JQ) = K - J Q * ( I Q - 1 )
MQ = J + JQ
D O 7 0 N Q = MQ,N
70 IP(NQ) = 0
GO TO 15
75 CONTINUE
80 CONTINUE
STOP
END

Cambridge Books Online © Cambridge University Press, 2010

232 Computations for Partitions Chap. 14.2

The printout begins with

1 0 0

2 0 0 ... 0
1 1 0 ... 0

3
3 0 0 ... 0
2 1 0 ... o
1 1 1 ... o
It is a simple matter to put in a subroutine that will examine whether each
partition (A l 9 ..., Xs) = (IP(1), IP(2),..., IP(S)) fulfills various prescribed
conditions.
There is also a related algorithm due to Hindenburg that may be restricted
to the partitions of n into exactly m parts. The order now is reverse lexico-
graphic: the parts are written in ascending order and (Al"-Am) appears
before (Ax' • • mAm') in the list if for some./: At = A± > A2 = A2'9. •., Aj_i =
AJ_VAJ < A/.
The algorithm begins with the first element in this ordering: ( l , l , . . . , l , n —
m + 1), and it passes from (Al9..., Am) to the next partition as follows:

find the largest./ such that Am - Aj ^ 2; (14.2.3)

replace Ap AJ+i,..., Am by (Aj + 1, Ai + 1 , . . . , Aj + 1, Am')

where Am' is so chosen that the number being partitioned
is still n. For example, when m = 5, n = 12 (14.2.4)

1 1L 1 1 8
1 1L 1 2 7
1 JI 1 3 6
1 1L 1 4 5
1 1I 2 2 6
1 1[ 2 3 5
1 1i 2 4 4
1 1 3 3 4
i :I 2 2 5
I :I 2 3 4
I :I 3 3 3
2 :I 2 2 4
2 2 2 3 3
Again we have an easily programmable algorithm.

Cambridge Books Online © Cambridge University Press, 2010

14.3 Algorithms from Generating Functions 233

14.3 Algorithms from Generating Functions

One value of our work on generating functions and partition identities lies
in the fact that the time of computation for certain partition functions may
be greatly decreased through algorithms derived from these theoretical
considerations.
For example, Corollary 1.8: p(n) — p(n — 1) — p(n — 2) -h p(n — 5) +
p(n — 7) = 0 , provides us with all values of p(n) for n ^ N after on
the order of %(6n3)* operations. This is quite efficient compared to the
algorithm that arises from Example 2 in Chapter 6, where the r(n) in
FL^i (1 ~~ <ln)~an = Zn^o r{n)qn may be computed from the recurrence

nr(n) = £ r(n - hj)jap

which requires on the order of n2 log n operations.

Exact formulas such as those in Chapter 5 also provide useful means for
computation.
The partition identities of Chapters 7-9 are also useful in this regard. For
example, in Theorem 8.5 we see that Aki(n) can be computed in on the order
of cn3/2 (c a constant that depends on k) steps from

/ j fik iV/*/*f — ~, 7 ^^ v *• / H v •*• — H ) '

On the other hand, if we compute Bki(n) from

X bkti(my n) = Bkti(n),

bk,i(m, n) = fefc.f-^m, n) + bktk_i+l(m - i + 1, n - m),

fl if m = n = 0,
if m < 0 or n ^ 0 but m2 + n2 ^ 0,

then we must use essentially dn2 steps (d a constant that depends on k).
In general, when we undertake computations of a particular partition
function, consideration of and transformations of the related generating
function should always be carefully examined in view of the possible time-
saving that can be effected.
We close this section with the following theorem, due to MacMahon,
which provides an even better algorithm than Corollary 1.8 for computing
the parity of p(n):

Cambridge Books Online © Cambridge University Press, 2010

234 Computations for Partitions Chap. 14.4

THEOREM 14.1

p(4n) = p(n) 4- p(n - 7) + p(n - 9) + • • • + p(n - a,) + • • • (mod 2),

p(4n + 1) = p(n) 4- p(n - 5) + p(n - 11) 4- • • • 4- p(n - /?,) 4- • • • (mod 2),

p(4n + 3) = p(n) 4- p(n - 3) 4- p(n - 13) 4- • • • 4- p(« - ? , - ) + • • • (mod 2),

p(4n + 6) = p(n) 4- p(n - 1) + p(n - 15) + • • • 4- p(n - <5,) 4- • • • (mod 2),

where the right-hand series are terminated just before the arguments become
negative and a, = 1(81 + 1), fit = I(8I + 3), y,. = I(8I + 5), 54 = I(8I + 7).
Proo/.

+ ^)"1 ( m od 2)

-q2""1) (Eq. (1.2.5))

+ q*"-^ + q4--3) (mod 2)

n=l

= ft(i - ^ T ' f k l - «4")(i + «*"-'Xi + <74n~3)

n=l n=l

= ( I P(n)q4n) f q in(B+ u (by Theorem 2.8).

The recurrences stated in Theorem 14.1 are now easily obtained by comparing
coefficients of q*n, q4rn+l, qAn+3 and q4n + e in the foregoing congruence. •

14.4 Computations for Higher-Dimensional Partitions

As long as we are treating problems in which the related generating function

is of some well-known form, for example, an infinite product, then it is
possible to proceed as we have indicated in Section 14.3. Unfortunately, such
good fortune is not available when we consider partitions of dimension d
with d > 2, as we have seen in Chapter 11. Thus, computation of even
three-dimensional partitions becomes a cumbersome problem. The only
useful tool known is a device of MacMahon that is closely related to our
proof of Theorem 3.7. MacMahon's technique was seen by D. E. Knuth
to apply to a quite general setting, and we shall present his generalization.

Cambridge Books Online © Cambridge University Press, 2010

14.4 Computations for Higher-Dimensional Partitions 235

We shall consider a set P partially ordered by © and an order reversing

map a* from P to the set of nonnegative integers; that is, if x © v in P, then
co(x) ^ a)(y).

DEFINITION 14.1. We shall call the order reversing map to a labeling if

additionally only finitely many x have co(x) > 0.
The following lemma isolates those elements of P with positive labels; this
immediately reduces our consideration from all of P (which is generally
infinite) to a finite subset of P.
LEMMA 14.2. Let the partial ordering © be extended to a linear ordering
^ of P. There is a bijection between the labelings of P and pairs of sequences

A?, ^ n2 ^ • * • ^ nm (positive integers),

x,, x 2 xm (elements of P), (14.4.1)

subject to the conditions that (i) x e P and x © Xj implies x = x{for some

i < j \ (ii) xt > xi+l implies n( > ni+l,for 1 ^ / < m.
Proof. The correspondence is obvious once we exhibit the appropriate
construction: Write down all x with a given largest label in increasing order
relative to ^ ; then write the x with the next largest label, again in increasing
order relative to ^ ; and so on. Once all positive labels are exhausted, write
the label of each x listed directly above that x. Inspection shows that this
procedure produces the desired sequences and is uniquely reversible. •
DEFINITION 14.2. We shall say that a labeling co of P is a P-partition
of n if Y,xeP °KX) — n-
Now by Lemma 14.2, every P-partition corresponds uniquely to a
pair of sequences

nx ^ n2 ^ H 3 ^ • * • (an infinite sequence of nonnegative integers),

JC,, x 2 , . . . , xm (distinct elements of P), (14.4.2)

where m ^ 0; (i) x e P and x ^ Xj implies x = x t for some i < j , (ii) x, > x / + ,

implies nL > ni+i for 1 ^ / < m\ (iii) if m > 0, then nm > / ? w + 1 ; (iv) if
m > 0, then there exists x e P such that x < x m and x # x, for 1 ^ / ^ /H.
Finite sequences {x,} satisfying (i) and (iv) are called topological sequences.
To see this correspondence, we merely attach infinitely many zeros to the
sequence nt and we delete x m if it is the least element of P — {xt,..., x,n_t}.
For example, let P be the set of points in the plane with nonnegative
coordinates and with the partial ordering (ij) © ( / ' , / ) if i ^ i" and j ^ /'.
This partial ordering may be extended to a linear ordering through the

Cambridge Books Online © Cambridge University Press, 2010

236 Computations for Partitions Chap. 14.4

lexicographic order (i,j) ^ (/",/) if / < i' or / = /" and j ^ / . In this case,
P-partitions as characterized in (14.4.1) correspond precisely to plane
partitions (written upside down). Thus the plane partition

1
12
431
5332

corresponds to the (P, a>)-partition

5 4 3 3 3 2 2 1 1 1
(0,0) (0,1) (1,0) (1,1) (2,0) (1,2) (3,0) (2,1) (0,2) (0,3)

The corresponding sequence (14.4.2) is

5 4 3 3 3 2 2 11100...
(0,0) (0,1) (1,0) (1,1) (2,0) (1,2) (3,0) (2,1)

Clearly there is a one-to-one correspondence between the (14.4.1) sequences

and the (14.4.2) sequences.
DEFINITION 14.3. The index of a topological sequence xl9 x2,..., xm is the
sum of ally such that Xj > xj+l with the convention that xm > xm+l.
Thus, returning to our example, we see that the topological sequence

(0,0) ( 0 , 1 ) ( 1 , 0 ) ( 1 , 1 ) (2,0) (1,2) (3,0) ( 2 , 1 )

has index 5 + 7 + 8 = 20.

Given a topological sequence x t , x 2 , . . . , x m , every sequence nx ^ n2 ^ • • *
of nonnegative integers that satisfies conditions (ii) and (iii) listed after
(14.4.2) corresponds uniquely to an infinite sequence p{ ^ p2 ^ p 3 ^ * * *
obtained by subtracting 1 from each of nl9..., nj for each j such that Xj > xj+l
orj = m. Returning to our example, we see that the sequence px ^ p2 > p3 ^ * * *
in this case is
2,1,0,0,0,0,0,0,0,0,... .

The remarkable fact is that the sum of the px plus the index equals the sum
of the Hi and this is the only relationship between the three sequences
{x{,..., x m }, {«,}, {pi). These observations establish the following result.
THEOREM 14.3. Let P be an infinite partially ordered set. Then there is a
one-to-one between correspondence P-partitions of n and ordered pairs
({x,}, {pi}) where {.xj is a topological sequence and {p,} is a partition of
n — /c, with k the index of {*,}. •

Cambridge Books Online © Cambridge University Press, 2010

14.6 Table of the Plane Partition Function 237

COROLLARY 14.4. Let P be an infinite partially ordered set; let s(n) be

the number of P-partitions of «, and let t(k) be the number of topological
sequences of P having index k. Then

Generally topological sequences are easier to find (and index) than arbitrary
(P, co)-partitions. For example, we may construct topological sequences of
points in the first quadrant as follows: First choose a set of, say, m points
that correspond to the graphical representation of some partition. Then
number these points using the first m integers, so that there is strict increase
along rows and columns, making sure that m is not in the first column. If, say,

is chosen, then the corresponding topological sequences are

3 2
124 and 134,
meaning xl = (0, 0), x2 = (1, 0), x 3 = (0, 1), x 4 = (2, 0), and x t = (0, 0),
x2 = (0, 1), x 3 = (1, 0), x 4 = (2, 0), respectively. In this instance our topolog-
ical sequences are essentially standard Young tableaux (to obtain standard
Young tableaux, reverse the sequence), with the special requirement that m
not be in the first column.

14.5 Brief Tables of Partition Functions

Table 14.1 lists p(n)y the number of partitions of n\ p(O, n)(= p(^, n) the
number of partitions of n with odd parts; A12(ri)(= B2,2(H)X t n e number
of partitions of n into parts congruent to 1 or 4 modulo 5, and A2l{n)
( = £ 2 , I ( H ) ) , t n e nu mber of partitions of n into parts congruent to 2 or 3
modulo 5.

14.6 Table of the Plane Partition Function

Table 14.2, for M2(n), the number of plane partitions of n, was easily
computed from the simple recurrence (Chapter 6, Example 2)

nM2(n) = t M2(n - j)a2(j)

where G2(J) *S t n e sum

° f t n e squares of the divisors of j .

Cambridge Books Online © Cambridge University Press, 2010

238 Computations for Partitions Chap. 14.6

Table 14.1

n P(n) POP.-) A12{n) ^ »

1 1 1 1 0
2 2 1 1 1
3 3 2 1 1
4 5 2 2 1
5 7 3 2 1

6 11 4 3 2
7 15 5 3 2
8 22 6 4 3
9 30 8 5 3
10 42 10 6 4

11 56 12 7 4
12 77 15 9 6
13 101 18 10 6
14 135 22 12 8
15 176 27 14 9

16 231 32 17 11
17 297 38 19 12
18 385 46 23 15
19 490 54 26 16
20 627 64 31 20

21 792 76 35 22
22 1002 89 41 26
23 1255 104 46 29
24 1575 122 54 35
25 1958 142 61 38

26 2436 165 70 45
27 3010 192 79 50
28 3718 222 91 58
29 4565 256 102 64
30 5604 296 117 75

31 6842 340 131 82

32 8349 390 149 95
33 10143 448 167 105
34 12310 512 189 120
35 14883 585 211 133
36 17977 668 239 152
37 21637 760 266 167
38 26015 864 299 190
39 31185 982 333 210
40 37338 1113 374 237

Cambridge Books Online © Cambridge University Press, 2010

14.6 Table of the Plane Partition Function 239

Table 14.1 (continued)

n Pin) P(Q. n) A2t2(n) A2tl(n)

41 44583 1260 415 261

42 53174 1426 465 295
43 63261 1610 515 324
44 75175 1816 575 364
45 89134 2048 637 401

46 105558 2304 709 448

47 124754 2590 783 493
48 147273 2910 871 551
49 173525 3264 961 604
50 204226 3658 1065 673

51 239943 4097 1174 739

52 281589 4582 1299 820
53 329931 5120 1429 899
54 386155 5718 1579 997
55 451276 6378 1735 1091

56 526823 7108 1913 1207

57 614154 7917 2100 1321
58 715220 8808 2311 1457
59 831820 9792 2533 1593
60 966467 10880 2785 1756

61 1121505 12076 3049 1916

62 1300156 13394 3345 2108
63 1505499 14848 3659 2301
64 1741630 16444 4010 2525
65 2012558 18200 4380 2753

66 2323520 20132 4794 3019

67 2679689 22250 5231 3287
68 3087735 24576 5717 3599
69 3554345 27130 6233 3917
70 4087968 29927 6804 4281

71 4697205 32992 7409 4655

72 5392783 36352 8080 5084
73 6185689 40026 8790 5521
74 7089500 44046 9573 6021
75 8118264 48446 10406 6537

76 9289091 53250 11322 7118

77 10619863 58499 12294 7721
78 12132164 64234 13363 8401
79 13848650 70488 14498 9103
80 15796476 77312 15742 9894

Cambridge Books Online © Cambridge University Press, 2010

240 Computations for Partitions Chap. 14.7

Table 14.1 (continued)

n Pin) p{&, n) A2>l(n)

81 18004327 84756 17066 10715

82 20506255 92864 18512 11631
83 23338469 101698 20050 12587
84 26543660 111322 21732 13653
85 30167357 121792 23519 14761

86 34262962 133184 25466 15995

87 38887673 145578 27540 17285
88 44108109 159046 29796 18710
89 49995925 173682 32196 20203
90 56634173 189586 34806 21854

91 64112359 206848 37582 23579

92 72533807 225585 40594 25483
93 82010177 245920 43802 27480
94 92669720 267968 47276 29671
95 104651419 291874 50974 31975

96 118114304 317788 54979 34502

97 133230930 345856 59239 37153
98 150198136 376256 63843 40058
99 169229875 409174 68747 43114
100 190569292 444793 74040 46447

14.7 Table of Gaussian Polynomials

Since these polynomials

have been so important throughout our work, we include self-explanatory

Table 14.3, complete for n ^ 12. We note that since

n
n —r
we need only consider 0 < r ^ w/2. Furthermore

J = 1 + q + q2 +
+
hence we consider only 2 < r ^ n/2.

Cambridge Books Online © Cambridge University Press, 2010

14.7 Table of Gaussian Polynomials 241

Table 14.2

n M2{n) n M2{n) n M2(n)

1 1 41 409383981 81 23498 9042219523

2 3 42 593001267 82 31738 2398602028
3 6 43 856667495 83 42817 1324714100
4 13 44 1234363833 84 57697 8362008262
5 24 45 1774079109 85 77663 3557947931

6 48 46 2543535902 86 104422 6582040722

7 86 47 3537993036 87 140249 8445554353
8 160 48 5191304973 88 188168 0993051045
9 282 49 7391026522 89 252192 5777635221
10 500 50 1 0499640707 90 337650 8618954817

11 859 51 14883573114 91 451605 3506209319

12 1479 52 21053676445 92 603409 5681424573
13 2485 53 2 9720561230 93 805440 1761691018
14 4167 54 4 1871334614 94 1074059 4259633282
15 6879 55 5 8874385349 95 1430879 4779412286

16 11297 56 8 2623976486 96 1904421 9737399823

17 18334 57 11 5737404664 97 2532294 5742768655
18 29601 58 16 1825846160 98 3364043 3065994170
19 47330 59 22 5863047430 99 4464887 4867815365
20 75278 60 31 4689799781 100 5920606 6030052023

21 118794 61 43 7699333376
22 186475 62 60 7771804065
23 290783 63 84 2541287719
24 451194 64 116 6117605448
25 696033 65 161 1415838202

26 1068745 66 222 3312543970

27 1632658 67 306 2906627106
28 2483234 68 421 3276093961
29 3759612 69 578 7232662336
30 5668963 70 793 7771067795

31 8512309 71 1087 2114256046

32 12733429 72 1487 0591703377
33 18974973 73 2031 1959869076
34 28175955 74 2770 7354427729
35 41691046 75 3774 5732428153
36 61484961 76 5135 4666391773
37 90379784 77 6978 1543686979
38 132441995 78 9470 1959299374
39 193487501 79 12836 4196860773
40 281846923 80 17378 1688194937

Cambridge Books Online © Cambridge University Press, 2010

242 Computations for Partitions Chap. 14.7

Table 14.3

1 a a2 a3 a* a5 a6 a1 8 9 10
a11 a12 a13 a14" a15 a16 a11 a18
r

4
2

1 1 2 2 2 1 1

61
1 1 2 2 3 2 2 1 1

61
1 1 2 3 3 3 3 2 1 1

1 1 2 2 3 3 3 2 2 1 1

71
1 1 2 3 4 4 5 4 4 3 2 1 1

81
1 1 2 2 3 3 4 3 3 , 2 2 1 1

8
1
1 1 2 3 4 5 6 6 6 6 5 4 3 2 1 1

81
1 1 2 3 5 5 7 7 8 7 7 5 5 3 2 1 1

9
1
1 1 2 2 3 3 4 4 4 3 3 2 2 1 1

1 1 2 3 4 5 7 7 8 8 8 7 7 5 4 3 2 1 1

[91
1 1 2 3 5 6 8 9 11 11 12 11 11 9 8 6 5 3 2
4
10
1 1 2 2 3 3 4 4 5 4 4 3 3 2 2 1 1
2
10
1 1 2 3 4 5 7 8 9 10 10 10 10 9 8 7 5 4 3
3
10
1 1 2 3 5 6 9 10 13 14 16 16 18 16 16 14 13 10 9
4
10
1 1 2 3 5 7 9 11 14 16 18 19 20 20 19 18 16 14 11

Cambridge Books Online © Cambridge University Press, 2010

Notes 243

Table 14.3 (continued)

[1 i q a1 g3 / <f <f *7 ,8 , V V V V V V V V V 8
11]
1 1 2 2 3 3 4 4 5 5 5 4 4 3 3 2 2 1 1

H
l
1 1 2 3 4 5 7 8 10 11 12 12 13 12 12 11 10 8 7
H
l
1 1 2 3 5 6 9 11 14 16 19 20 23 23 24 23 23 20 19

1 1 2 3 5 7 10 12 16 19 23 25 29 30 32 32 32 30 29

1 1 2 2 3 3 4 4 5 5 6 5 5 4 4 3 3 2 2

1 1 2 3 4 5 7 8 ' 10 12 13 14 15 15 15 15 14 13 12

1 1 2 3 5 6 9 11 15 17 21 23 27 28 31 31 33 31 31

12]
1 1 2 3 5 7 10 13 17 21 26 30 35 39 43 46 48 49 49

1 1 2 3 5 7 11 13 18 22 28 32 39 42 48 51 55 55 58

14.8 Guide to Tables

Gupta, Gwyther, and Miller (1958) present the most extensive readily
available table of partitions. This table lists p(n) for n ^ 600, and extensive
tables are given for the number of partitions of n into at most m parts. Their
work greatly extends a previous table by Gupta (1939). Barton, David, and
Kendall (1966) present short tables of partitions and bipartitions. N. J. A.
Sloane (1973) lists numerous partition sequences in his fascinating Handbook
of Integer Sequences. The papers by Andrews, Cheema, Churchhouse, and
Burnell and Houten in the book Computers in Number Theory (Academic
Press, New York, 1971) describe some recent uses of computers in partition
problems.

Notes
The algorithm of Hindenburg in Section 14.2 is taken from Dickson's
History of the Theory of Numbers (1920, Vol. 2, p. 106); the entire Chapter 3

Cambridge Books Online © Cambridge University Press, 2010

244 Computations for Partitions Chap. 14

of that book presents a detailed early history of partitions. The history from
1940 to 1972 may be found in LeVeque's (1974) Reviews in Number Theory,
Chapter P, Vol. 4, while in Vol. 6, Section Z30, reviews of various tables of
partitions may be found.
Theorem 14.1 is from MacMahon (1921); MacMahon's work was greatly
extended by Parkin and Shanks (1967). The material in Section 14.4 is taken
from Knuth (1970); Knuth includes a computer program for finding the
topological sequences related to solid partitions. The P-partition concept has
been generalized extensively by Stanley (1972) to(P, co)-partitions, and he
has shown that numerous partition and permutation problems can be treated
through the use of this powerful method.
The excellent book by Nijenhuis and Wilf (1975) presents a number of
algorithms for the computation of compositions and partitions; Chapters 5
and 6 of their text treat compositions while Chapters 9 and 10 are related
to partitions.

References
Andrews, G. E. (1971). "The use of computers in search of identities of the Rogers-
Ramanujan type," Computers in Number Theory (A. O. L. Atkin and B. J. Birch,
eds.), pp. 377-387. Academic Press, New York.
Barton, D. E., David, F. N., and Kendall, M. G.(1966). Symmetric Functions and Allied
Tables. Cambridge Univ. Press, London and New York.
Burnell, D., and Houten, L. (1971). "Multiplanar partitions," Computers in Number
Theory (A. O. L. Atkin and B. J. Birch, eds.), pp. 401-404. Academic Press, New
York.
Cheema, M. S. (1971). "Computers in the theory of partitions,'* Computers in Number
Theory (A. O. L. Atkin and B. J. Birch, eds.), pp. 389-395. Academic Press, New
York.
Churchhouse, R. F. (1971). "Binary partitions," Computers in Number Theory (A. O. L.
Atkin and B. J. Birch, eds.), pp. 397-400. Academic Press, New York.
Dickson, L. E. (1920). History of the Theory of Numbers, Vol. 2. Carnegie Institution
of Washington, Washington, D.C. (reprinted by Chelsea, New York, 1952, 1966).
Gupta, H. (1939). Tables of Partitions. Indian Math. S o c , Madras.
Gupta, H., Gwyther, A. E., and Miller, J. C. P. (1958). Tables of Partitions (Roy. Soc.
Math. Tables, Vol. 4).
Knuth, D. E. (1970). "A note on partitions," Math. Comp. 24, 955-961.
LeVeque, W. J. (1974). Reviews in Number Theory, Vols. 4 and 6. Amer. Math. S o c ,
Providence, R.I.
MacMahon, P. A. (1921). "Note on the parity of the number which enumerates the
partitions of a number," Proc. Cambridge Phil. Soc. 20, 281-283.
Nijenhuis, A., and Wilf, H. S. (1975). Combinatorial Algorithms. Academic Press, New
York.
Parkin, T. R., and Shanks, D. (1967). "On the distribution of parity in the partition
function," Math. Comp. 21, 466-480.
Sloane, N. J. A. (1973). A Handbook of Integer Sequences. Academic Press, New York.
Stanley, R. P. (1972). "Ordered structures and partitions," Mem. Amer. Math. Soc.
119.

Cambridge Books Online © Cambridge University Press, 2010

Index for Definitions of Symbols

(§) 71 b(m;n) 175

A 99 b(n) 161
d B
» 214
£(<?) 161
*<<'> l?6 *(«) 205
'^ f j 123 # o to) 199
-dka(N) 117 Cj 170

^ ( f l ) (Sylvester partition c (a 1 ,. .. , a ) 58
r
function) 24
c a
Ak(n) (sum of roots of v p • • • »Gr,/n) 57
unity) 70 ^ 99
>!..(«) 109 *
*' 7 Y(A,A:) 82
x
a . 99
Y tt/) 42
ax (\ a partition of a number) 221 *
C (m) 163
*„ 221 f
I23
(«)„ to-factorial) 17 **•«
/o \ 1QQ Ct (W, /i) 111

! 14
aw (ir a partition of a set) 221 ^ / ^
199 ck(m,n) 63
(w)
17 C 129
17
/) 141
C(W W) 54
^(1.5)83 '
jS^/ii) 206 c(^,M,«) 55
<m 19-5 ® 2,122
ka
A
3 (TV) H7 ^to) 99
bk.(m,n) 109 S O ^ . ^ ) 217
5. </i) 109 D. 98
Bk(n) 24 3. 123

245

Cambridge Books Online © Cambridge University Press, 2010

246 Index for Definitions of Symbols

.i) 23 83
dki(m,n) 114 7(70 221
Dki(n) 114 GF(q) 212
G(N,M\q) 33
8k(m) 164
G(g) (related to binary
/>„ 50 partitions) 162
g(q) (modular function) 167
dots 7
g"(r) (auxiliary generating
® (r;bl,b2,...,bm;m) 137 function) 89
Z)(5) 88 G(u) (multipartite partition
£ 137 generating function) 206
eki(m,n) 115 G(z,s) 83
g(z,s) 83
£(/i) 140
"H" 3
6,(A2) 156
eo(n) 156
F 221 ) 135
AA,nX, Y,L) 147 //,.(*) 135
/ c (*;*) 128
/ / , 78
Fd{q) 99
97
/Z.(JC) 116 ^.«
97
F ^ (the Farey series) 72 **.±a
Hki(a\x\q) 106
9m{q) 161
#„(') 49
Fn 63
Fj(w) 208
fs{z\q) 16 H(z,s) 83
h(z,s) 83
F(w) 206 / 219
/^fe) 30 Ih k 76
FQ{q) 30 ind(m t , . . . ,inr\n) 42
G 198 inv(w?j,. . . ,mrn) 40
7 (Euler's constant) 90
/j(z) 82
7(/) 221
r(m\q) 176 Jk/a;x:q) 106

Cambridge Books Online © Cambridge University Press, 2010

Index for Definitions of Symbols 247

Ker0 219
k (X a partition of a number) 221 j t l « ) 189
k (n a partition of a set) 221
aiTTpTT^ 217
IXl 215
n (as multipartite number) 202
X! 215
n (as the set of the first n
X' 7 integers) 214
Xhn 1
X(TT) 2 1 5
X(7r lf 7r 2 ) 2 2 0

2c{*a) 131 60
Z, c 130
215
Li (Long operator) 64

Z,^ (loop integral) 76 nodes 7

L^ (lattice of partitions) 217 v(j) (least power of 2 in / ) 116
v(n) 219
Ln(q) 170
v(r) (5 -adic valuation) 168
/(*,) 131 f 2,122
L(zfs) 83 w
/,.* 71

^137
co(/) (binary representation) 116
M.(n) (MacMahon restricted
partitions) 14 CJ(JC) (labeling) 235
Mk(n) (A:-dimensional ^'(n) 203
partitions) 189
P(av...,ar) 58
m 143
>
/ =(a1,...,c^.;m) 57
143
P A > c ( « - c ) 25
Ml^bwq) 149
Pe(9,n) 10
mkiUi\N) 152
pe(®(k,i),n) 23
MuAu\Q) 152
0 129
39 <I>(w;^) 175
«(«) 167
4>0(q) 30
M, 96 0, (*) 94

Cambridge Books Online © Cambridge University Press, 2010

248 Index for Definitions of Symbols

95 p(it) 199
71
0 3 ( w ) 95
**(«) 78
n(a,b/n,m) 208
*(m;<7) 176
*.(«;«) 183
*' r *0(<7) 30
n, (n, m) 209 "
1
... P(*«) 2
f v f
l 2 Z1/ /»2(«) 127
J , - ^ ^ , . . . ^ 206
217
^ . ( W ) ,53

^ ' ( « ) 56
nr(nl,n2,.. . , « k\q) 180
/>« (n;/) 203
D i
!2(n) 203
p i
i [r, n) 13 G(n;/) 203
. , * , ) 206 Q+(n) 203

Pk,ti(a9b\yL\N) 143 Q + (n;/) 203

p (a,b\ii\q) 143 ^ (*) 117
k. ti
n. {a,b\ii\q)
ti 149 r . ^ ) 143
p
k, tiUi\ff) 152 R{k,a;n) 175
p
k, i^q) 152 ^.z^1^ 112

88
p
k, r(m,n) 183
/?, 94
4
© 130
95
2
P\> i) 1 set
^ ( °^ Partitions) 2
Pi} l) 202
PQ V,M,n) 33 ^ ( s e t °^ Partitions as "frequency"
sequences) 122
n
P f <?) n} 10
0 #(X) 16
Po (9(*.0,/i) 23 s(hJ() 72
P
X W 127 a ({/;}) (sum of parts of
partition) 16, 122
(P. cj)-partition 51,235 a (X) (Euler's pentagonal number
PM («) 56 theorem) 10

Cambridge Books Online © Cambridge University Press, 2010

Index for Definitions of Symbols 249

o2(j) 237 6^2 + (n) 29

sign X 2 1 5 ^fe) 212

Sk(t) 226 Vs(n) 29
s(X) 10 Kz 83
sK (X a partition of a V,(rfm,fi) 13
number) 221
s m a x ( M p « 2 , . . . ,nr) 209 W(k,r) 183
[x] 83
5 221
n {*} 83
s(n,k) 219 {X)n (falling factorial) 2 1 9
S(ntk) 219 2 0 8
$nh
sn (IT a partition of a set) 221 ^o h
Y (&vg2 g ) 205
Sr 167
0 219
S(T\m,n) 30
r 198 partition lattice) 217
f/m 161 Us) 83
f (s, h) (Hurwitz zeta
w 13/ function) 97

Cambridge Books Online © Cambridge University Press, 2010

Author Index
Numbers set in italics designate pages on which complete literature citations are given.

Alder, H. L., / 1 8 Fine, N. J., 26, 29, 31,32, 210, 277

Andrews, G. E., 74, 15, 31, 32, 52, 65, Foata, D., 45, 51,52, 6 5 , 6 6 , 6 7
85, 86, 118, 119, 137, 138, 158, 111, Foulkes, H. O., 67
178, 200, 210, 211, 227, 228, 243,244 Franklin, F., 9, 14,75, 52
Arkin,J., 8 5 , 5 6 Fray, R. D., 67
Askey,R., 5 1 , 5 2 Frucht, R., 227,225
Atkin, A. O. L., 139, 143, 158, 160, 161, Fulton, J.D., 227,225
175,177,7 75,200
Auluck, F. C , 100 J01,228 Gauss, C. F., 23, 31,32, 35, 51, 52
Ayoub, R.,xiii, 85,56 Glaisher.J.W. L., 6 , 9 , 14,75
Goldman, J., 227,229
Bachman, ?.,xiii Gollnitz, H., 105, 113, 114, 118, 779,
Bailey, W.N., 20,30, 31, 32 121,123
Barton, D. E., 243,244 Gordon, B., 105, 109, 113, 114, 118, 779,
Bateman, P. T., 101 121,123,200, 210,277
Bell, E. T., 204, 205, 206, 210, 211, 214 Grosswald,*///, 85,56
Bender, E., 3, 75, 200, 221, 228 Gupta, H., 56, 65, 67, 118,720, 160, 165,
Berndt, B., 7 1 , 8 5 , 5 6 177,7 75, 243,244
Bohr, N., 227, 225 Gwyther, A. E., 243, 244
Bratley, P., 200
Brigham,N. A., 100,707 Hagis, P., 81, 82, 85, 56
Brylawski, T., 227,225 Hahn,W., 31, J2
Burnell,D., 243,244 Hardy, G. H., xiii, 14, 75, 30, 32, 68, 69,
72, 85, 86, 94, 101, 103, 105, 118, 720,
Carlitz, L., 31, 32, 52, 65, 66, 179, 183, 159,160,177
200, 202, 207, 210, 277, 227,228 Haselgrove, C. B., 100,707
Cauchy, A., 17, 21, 30,32, 51,52 Heine, E., 19,20, 30,32, 5 1 , 5 2
Cayley, A., 65, 66, 85,56, 209 HermitcC, 52
Chaundy,T.W.,200 Hickerson, D. R., 75
Cheema, M. S., 779, 203, 210, 277, 243, Hodge, W. V. D., 200
244 Hodges,J. H., 227,228
Chowla,S., 160 Houten,L., 200, 243,244
Chrystal,G.,xm Hua, L. K.,82,56
Churchhouse, R. F., 160, 165, 177, 178,
243,244
Ingham, A. E., 100,702
Comtet, L.,xiii, 14, 75. 210,277
Iseki,S., 82, 85,56, 87

Daum,J. A., 30,32 Jackson, F.H., 3 8 , 5 1 , 5 2

David, F.N., 243,244 Jacobi, C. G. J., 17, 21, 30, 31,32, 49
Daykin, D. E., 138 Jordan, C , 227,225
Dickson, L. E., 227,228, 243, 244
Dillon, J.F., 65, 66 Kalckar,F., 227,225
Doubilet,P., 212, 227,228 Kendall, M.G., 243,244
Dragonette, L., 8 2 , 8 5 , 5 6 Knopp, M. I., xiii, 71, 85, 87, 160, 167,
Dyson, F. J., 142, 755, 161, 175, 7 78 111,178
Knuth, D., 179, 184, 200, 212, 227, 228,
Erdos, P., 56, 65, 66, 101 230,234,244
Euler, L., xiii, 5, 9 , 1 1 , 1 4 , 75, 19, 21, 24, Knutson,D.,75, 221,228
26, 29, 30, 31,32, 123,139,203 Kohlbecker, E. E., 100,702

250

Cambridge Books Online © Cambridge University Press, 2010

Author Index 251

Kolberg,O., 111,178 Robinson, G. deB., 200, 201, 227, 228

Kothari, D. S., 228 R6dseth,O., 160, 161, 165, X11J7S
Kreweras,G., 65,6 7, 199,201 Rogers, L. J., 31, 32, 49, 50, 52, 53, 103,
105, 109, 113, 118,726;, 121, 123, 155,
Landsberg, G., 221,228 158, 177,200
Lascoux, A., 199, 201 Roselle, D., 65, 66, 67, 210, 277
Lebesgue, V. A., 30, J2 Rota,G. - C , 227,225.229
Lehmer, D. H., 86, 101,702 Roth,K.F., 100.702
Lehncr, J., 56, 65, 66, 85, 86, 101, 120, Rutherford, D. E., 199, 201, 227, 229
111,178
LeVeque, W. J., 14, 15, 3 1 , 32, 65, 67, Schensted, C , 184,200, 201
71, 87, 100, 102, 118, 120, 138, 158, Schrutka, L. von, 158
111, 178, 200, 201, 210,277, 221,228, Schur, I. J., 48, 105,116, 118, 720, 158
244 Schutzenberger, M. P., 51, 52, 65, 66,
Littlewood,D. E., 227, 228 200,201
Littlewood, J. E., 68 Schwarz,W., 100,102
Livingood, J., 100, 102 Scoville, R., 66
Long,C, 65, 6 7 Selberg,A.,118,720
Shanks, D., 244
MacDonald, I. G., 200 Singh, V.N., 118,720
MacMahon, P. A., xiii, 13, 14, 75, 4 1 , 4 2 , Slater, L.J., 31,32
51, 52, 57, 59, 65, 67, 85, 87, 100, Sloane, N. J. A., 243,244
102, 105, 137, 138, 159,177,179, 184, Snapper, E., 227, 229
189, 199, 200,201, 210,277, 227, 228, Solomon, L.,277
233,234,244 Stanley, R. P., 45, 51, 52,53, 65, 67, 200,
McKay, J. K.S.,200 201,221,229,244
Meinardus, G., 88, 89, 99,100, 101, 102 Star, Z., 65, 67
Menon,P. K . , 3 1 , 5 2 Starcher, G. W., 14,75
Miller, J.C. P., 243,244 Stenger, A., 158
Miller, K. S., 132,138 Strehl,V.,67
Moore, E., 75 Subbararo, M. V., 13, 14,75, 31,32, 138,
Motzkin, T. S., 210,277 204,210,277
Murray, M., 132,138 Subramanyasastri, V. V., 81, 85, 87
Swinnerton-Dyer, H. P. F., 161,175,178
Netto,E.,75 Sylvester, J. J., 14, 75, 24, 31,32, 51, 52,
Nijenhuis, A., 244 53,139,140
Szego,G.,49,50,52,5J
Ostmann, H. H., xiii, 14, 15 Szekeres,G., 100, 101,702

Parkin, T. R., 244 Temperley, H. N. V., 100, 707, 227, 229

Pedoe, D., 200 Tietze,H., 200,207
Percus, J. K.yxiii Turin, P., 707, 702

Vahlen, K. T., 100, 702, 755

Rademacher, H., xiii, 15, 68, 69, 71, 85, Venkatachaliengar, K., 24,31,32
87, 111,178
Ramamani,V.,24,31,i2 Watson, G. N., 31,32, 51, 53, 82, 85,37,
Ramanujan, S., 50, 68, 69, 72, 85, 86, 94, 101,702, 105, 118,720, 160, 111,178
101, 103, 105, 1 0 9 , 1 1 3 , 1 2 1 , 1 2 3 , 1 5 5 , Whiteman, A., 87
158, 1 5 9 , 1 6 0 , 1 6 1 , 1 6 7 , 1 7 0 , 1 7 4 , 177, Wilf, H. S., 244
773,200 Wright, E. M., xiii, 14, 75, 32, 100, 702.
Richmond, B., 100,101,102 200,207,210,277,229
Riordan, J.,JC//J, 14,75, 66, 210,277
Robertson, M. M., 211 Young, A., 199

Cambridge Books Online © Cambridge University Press, 2010

Subject Index

Abel's lemma, 209 Equivalence, see Partition

Alder polynomials, 118-119 Erlangen
Algebraic geometry, 199-200 University of, 137
rj-function, 70-71
Basic hypergeometric series, 17, 202, 227 Euler pair, 138
Heine's fundamental transformation, 19 Eulerian series, see Basic hypergeometric
well-poised, 106 series
Bell numbers, 214, 216 Euler's partition theorem, 24, 26, 122 —
Bell polynomials, 204-206, 210, 216 123,139,203
Binary partitions, see Partition Euler's pentagonal number theorem, see
Binary representation, 116 Pentagonal number theorem
Binomial series, 54, 177
Binomial series, see also ^-analog Faa di Bruno's formula, 205, 210
Bipartite number, 207 Factorizations
Blocks, 2 1 5 - 2 1 7 , 2 1 9 - 2 2 0 , 222-223 unordered, 210
Bumping, 185, 187 Fall, 43
Farey dissection, 72, 94
Canonical basis. 213-214 Farey fractions, 72
Cauchy's integral theorem, 72, 93 Ferrers graph, see Partition
Cheema's theorem, 203 Fibonacci numbers, 63, 137
Churchhouse conjecture, 160,165, 177 Ford circles, 85
Chu-Vandermonde summation, 51, 61, 64, Formal power series, 3
227 FORTRAN IV, 230
Chu-Vandermonde summation, see also Fractional part, 83
^-analog Fundamental partition ideal problem, 123
Circle method, 85
Class, 187-188 Gamma function
Combinatorics, 199, 212ff Euler's integral, 83
Composition, 5 4 - 5 8 , 63, 65 Gaussian multinomial coefficient, 39
algorithms, 244 Gaussian polynomial, 3 3 , 3 5 - 3 9 , 4 9 , 146,
asymptotic formulas, 65, 68 180,212,214,240
graphical representation, 55 Gaussian sum, 37
multipartite number, 54, 57, 179, 202 Generating function, 221
vector, 5 4 , 5 7 , 6 5 algorithms from, 233
Continued fraction, see Raman ujan's con- Bell polynomials, 205
tinued fraction determinants, 179
finite product, 42
Dedekind's rj-function, see ^-function infinite product, 3, 88, 203-204, 209
Dedekind sum, 72 multipartite partitions, 203
Derivative nonmodular, 85
composite function, 204 plane partition, 185, 199, 210
Differential equations, 132 recurrences, 179-180
Distributions, 6 1 - 6 2 two-variable, 16, 128
Dots, 7 Generating function, see also Partition
Durfee square, 23, 2 7 - 2 8 Geometric series
side of, 28 finite, 4
Elliptic modular function, see Modular infinite, 4
function Gollnitz-Gordon identities, 113-114, 123
Elliptic theta functions, see Theta func- generalization, 113-114, 123
tions Gollnitz 's theorem , 1 1 7 - 1 1 8

252

Cambridge Books Online © Cambridge University Press, 2010

Subject Index 253

Gordon's multipartite partition theorem, Matrices, 1 8 4 - 1 8 5 , 1 8 7 - 1 8 8

210-211 Maximal element of L ,219
Gordon's theorem, see Rogers-Ramanujan Mediant, 73
identities Meet, 122, 217
Greater index, 42 Mellin transform, 90
Greatest integer function, 83, 157 Minimal element of L 219
Group, 198 Mobius function, 217, 220
cyclic, 198 Mobius inversion, 218, 223-224
Group theory, 211,227 Mock-theta functions, 2 9 - 3 0 , 82
statistical aspects, 101 fifth-order, 2 9 - 3 0
Hankel's loop integral, 75, 79 Modular equation, 161, 167, 170
Hecke operator, 1 6 0 - 1 6 1 , 167 Modular form, 70, 160, 177
Hindenburg algorithm, 232, 243 Modular function, 8 1 - 8 2 , 8 5 - 8 6 , 160,
Homogeneous polynomial, 204 175
Hurwitz zeta function, 97 Modular function, see also Modular forms
Hypercube Modular group, 71, 161
(k + 1)-dimensional, 198 Modular transformation, 71, 82
Modulus, 129-131
Ideals, 125 Multinomial coefficient, 215
Ideals, see also Partition Multipartite numbers, 57, 202-203
Incidence algebra, 217 Multipartite number, see also Partition
Inclusion-exclusion, 139, 153, 156, 218 Multiset, 39,126
generalized, 218 multiplicity in, 39
truncated, 140 permutation, 4 0 - 4 2 , 5 1 , 6 0 - 6 1
Index, 236-237 Murray-Miller algorithm, 132
Intersection, 217
Invariant theory, 199 Newton's formula, 168-169
Inversion, 4 0 - 4 1
Order ideal, see Partition
Jacobi's triple product identity, see Oscillation, see (k, i)-positive oscillation
Triple product identity and {k, /)-negative oscillation
Join, 122, 217
Parity of p(n), 233
Kernel, 219 Partial summation, 98
(k, /)-negative oscillation, 143-145, Partially ordered set, 220, 235-237
152-154 locally finite, 217
(k, /)-positive oscillation, 143-145, 152 — infinite, 237
154 Partition, 1, 204, 212, 222, 226, 243
Knuth-Schensted correspondence, 179,184- algorithms, 230, 244
185,187,200 analysis, 227
Kronecker 6 -function, 217 applications of, 33, 227
asymptotic formulas, 56, 6 8 - 7 0 , 8 8 -
Labeling, 235 89,93,97
Legendre-Jacobi symbol, 71 binary, 160-161,177
Lexicographic order, 185, 187, 203, 236 bipartite numbers, 207-209, 243
reverse, 232 Cayley's decomposition, 81
Linearly independent vectors, 213 computations for, 220ff
Linear transformation, 225 congruences, 159-160
Linked partition ideals, see Partition conjugate, 7 - 9 , 24,28
Lipschitz summation formula, 84 Ferrers graph, 6, 13,40, 142, 214
Logarithmic derivative, 98 flushed,31
Long operator, 64 from set//, 3, 13, 16, 100

Cambridge Books Online © Cambridge University Press, 2010

254 Subject Index

functions. 1 - 2 , 6 8 , 8 8 . 9 7 , 1 5 9 - 1 6 1 . 2 3 7 Partition, see also Composition, Generat-

generating function, 3 - 4 , 94, 125, 139 ing function. Plane partitions, Sets
graphical representation, 6 - 9 , 4 0 , 142 — Patience, 59
145,214 Pentagonal number. 10
higher dimensional, 1 7 9 - 1 8 0 , 189, 197, Pentagonal number Theorem, 9, 11, 14, 29,
199, 2 3 0 . 2 3 4 . 244 202
Ferrers graph, 197-198 Permutations, 244
graphical representation, 197, 199 see also Multiset
history, 244 Plane partitions. 97, 1 7 9 - 1 8 0 , 1 8 3 - 1 8 5 ,
ideal equivalence, 1 2 2 - 1 2 3 , 128 188, 1 9 8 - 2 0 0 , 2 3 6 - 2 3 7
ideals, 121 - 1 2 5 . 129 asymptotics, 199
ideals of order 1, 1 2 1 , 1 2 4 - 1 2 5 , 204, conjugate, 198
207 Ferrars graph, 198
identities, 17, 18, 118, 121, 128, 2 0 2 , 2 0 7 , MacMahon's theorem, 179
233 with strict column decrease, 188
in combinatorics, 212ff Plane partition, see also Generating func-
in^o powers of m, 1 75 - 176 tion, Partition
A:-dimensional. 189, 197 -199 Poincare series, 211
k-th excess, 50 Protrusions. 51
lattice structure, 122
linked ideals. 128, 131, 135, 137
linking set, 131 ?-analog
modular, 13, 14 binomial series, 17
multipartite number, 5 7 - 5 8 , 2 0 2 - 2 0 4 , Chu-Vandermonde summation, 38, 51
206,210-211 Gauss's theorem, 20
distinct parts, 203 Kummer's theorem, 20
odd parts, 203 Saalschutz's theorem, 38
of infinity, 137 ^/-binomial coefficient, see Gaussian poly-
order of ideal, 124 nomial
P, 235-237, 244 ^-difference equations, 104, 121, 1 2 8 -
partial ordering, 122 130, 132, 1 3 5 - 1 3 6
parts, 1 ^-multinomial coefficient, see Gaussian
perfect, 1 3 6 - 1 3 7 multinomial coefficient
(P, cu), 51, 65, 200, 227, 244 ^-series, see Basic hypergeometric series
protruded, 51
restricted. 3 3 , 210, 243
restricted bipartite, 207 - 2 0 9 Radix representation theorem, 138
self-conjugate, 14 Ramanujan's congruences, see Raman-
sieve methods. 139, 142, 148, 153. 156 ujan's conjecture
span, 131 Ramanujan's conjecture, 160, 167, 170.
statistical aspects, 101 174,177
three dimensional. 234 Ramanujan's continued fraction, 103 -105
successive rank, 139, 143, 153, 198 Rank, 142, 175
Sylvestered, 3 0 - 3 1 Rational point, 72
tables of. 2 3 7 - 2 4 4 Real dimension, 63
r-,52 Reciprocal polynomial. 45 - 48, 55
vector, 57 58 Riemann zeta function, 83
with congruence conditions, 1 3 - 1 4 , Rbdseth's theorem, 161, 165
116-118 Rogers-Ramanujan identities. 50, 103,
with difference conditions, 1 3 - 1 4 , 99, 1 0 5 , 1 0 9 , 1 1 3 , 1 2 3 , 1 5 5 , 158, 1 7 7 , 2 0 0
116-118 general theorem, 1 1 8 - 1 1 9
with distinct parts. 2, 5 -6, 123, 156 Gordon's generalization. 109, 118, 123,
with odd parts. 2. 5 6, 1 3 - 1 4 . 123, 237 156
with parts - +2 (mod 5). 109, 237 Rogers-Ramanujan-Schur identities, see
with parts + 1 (mod 5), 109. 237 Rogers-Ramanujan identities

Cambridge Books Online © Cambridge University Press, 2010

Subject Index 255

Rogers-Szego polynomials, 50 Successive rank, see Partition

generating function, 4 9 - 5 1 Summatory maximum, 209
r-partite number, see multipartite number Sum of squares of divisors, 237
Run, 6 0 - 6 1 Symmetric functions, 212, 217, 221
fundamental theorem, 221, 223-224
Saddle point method, 89, 93 Symmetric group, 211
Schur's partition theorem, 116 representation theory, 199
Segment, 217-218, 220
Semi-ideal, see Partition Theta functions, 105
Semi-invariant, 48 Topological sequence, 235-237, 244
Sets Trace, 198,210
inclusion, 218 conjugate, 198
lattice of subsets, 218 Triple product identity, 17, 21, 30, 49,
partition lattice, 217 108-109, 176
partition refinement, 219 Two-dimensional arrays, 179
partitions, 212, 214-217, 2 2 0 - 2 2 2 ,
227 Unimodal distribution, 100
partial order, 217 Unimodal polynomial, 4 5 - 4 8 , 55
Sieve methods, 139ff Union, 217
generalized, 218
Sieve methods, see also Partition Valuation
Simon Newcomb's problem, 54, 5 9 - 6 0 , S-adic, 168
65 positive, 122
Solid partitions, see Partition Vector spaces
Span, see Partition finite, 212, 225, 227
Stacks, 52 graded,211
Steadily decreasing, 207
Stirling numbers Young tableaux, 199, 237
first kind, 2 1 9 - 2 2 0
second kind, 2 1 9 - 2 2 0 ZeckendorPs theorem, 138
Subspaces, 2 1 3 - 2 1 4 , 2 2 5 Zeta function
m-dimensional, 213 of a lattice, 217

Cambridge Books Online © Cambridge University Press, 2010