Vdoc - Pub The Theory of Partitions
Vdoc - Pub The Theory of Partitions
George E. Andrews
Pennsylvania State University
University Park, Pennsylvania
CAMBRIDGE
UNIVERSITY PRESS
A catalog record for this book is available from the British Library.
Editor's Statement vi
Preface to Paperback Edition xiii
Preface xv
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
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.
1.1 Introduction
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 );
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).
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.
DEFINITION 1.5. The generating function/(g) for the sequence aQ, a t, a2i a3,...
is the power series f(q) = ^ o anqn.
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.
Thus if N is the set of all positive integers, then p("N"(^ 1), n) = p(^, n).
= n (i + «•+•••+«*•)
netf
1 + x + x 2 + • • • + xr =
1 —x
neH neH
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"
Zp("H"j)qj ^ n a - q
h l
r ^ o (i - qktrl < *>•
j=0 i=l j=l
as
Therefore
ipi"H"j)qj = n o - qhri - n o -
j=0 1=1 neH
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
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
_ n _i
~~ 11 / j n\
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.
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
or
Since most of the classical texts on partitions use the first representation
shown in this section, we shall also.
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. •
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
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.
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 ~
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
that is
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).
m=1
oc
m= 1
= 1 +t(P*&>n)-Poi®>n))<lH>
by Theorem 1.6.
To complete the proof we must show that
00 00
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
Examples
n = kh + y.
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
Notes
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.
2.1 Introduction
m=0 n=0
AeS
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
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
(«)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,
« 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)
or
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.
COROLLARY 2.3 (Heine). For \q\ < 1, |/| < 1, \b\ < 1
=n - bqm)(l -
^o (1 - e<T)(l -
atqm)
x (1 - ti/Xl ~ atq)-' (1 - at
Proof.
y y (a)S
.-o «-o (4).
{b)
2L f
(c)
. y (clb)m(t)mbm
+
»?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)
y (1
(
« (1 - a<? 2m+1 )(l + ^ + 1 ) ( 1 -aq2m + 2
lb2)
£ (b)n\
2- — T :
g (q2lb2'q2)mam
(aqlb)x(-q/b)xm%
i+y *— -
.t*, (1 - flXl - <72)- • - d - 9"X1 - zXl - rfl)' - d - z'/"" 1 )
Remark. Equation (2.2.8) is due to Cauchy, and Eq. (2.2.9) is due to Euler.
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.
ft (1 + zq2n+i) = - | Z mgm
2 (by (2.2.6))
1 °°
/^,2. ^2\ L-t » *1 oo
W > q 7oc m = — oo
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. •
y (_ nw0<2fc+l)n(n+l)/2-in/j _ (2n+l)i\
n= 0
00
= y / |\n (2lc+l)n(n+l)/2-/n M
n = — oo
Z ( - 1)V 3 = PI (I ~ ql>
:=
(2.2.12)
n "~oo HI= 1 v •" T /
oo °° (1 a^m}
A 1 (\ 2m — 1\
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.
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
- q
M a-?2'1)
= ((1 ~ a) 1 ;<?2) °° - (d - a)q;q*U- ,)„ (2.3.1)
= 1+ f
n=l
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)»
Therefore
y ...^_ - y x"(
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)a0 ^ (tq;q2)m0q2;q2)mq2m
00
» 1 / o o V"1 (q2;q2)Jmq2m
?o
(by Corollary 2.3)
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
4
z2)oom=o (q2;q2Ubt;q2)m
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
V (- 1
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) =
^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.
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
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.
3.1 Introduction
, M, n) = 0 if n > MAT,
p(N, M, NM) = 1.
/1 W + A/\/< N + A/ — 1 \ /i A/ + 1 \
G(N, M; q) = —
(3.2.1)
33
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
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
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.
•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
- 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)
(<?),
>=o
. (-
which is (3.3.6).
Again by Theorem 2.1,
-1
^ (qN)jZi
N
7j _
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)
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)).
n + m + 21 \n + m + 11 fn + m + 1
9
(z)m+n =
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.
£ (abtlcUbUc/b)-
(c/bUbt)x $
(c!b)Jbt)K
cjb)^ n% (q)n(c)n
(abt!c)v Z
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
( M , / ) = {m/ ( m i ) m 2 / ( W 2 ) - • - m / } .
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
Theorem 3.5, together with Theorem 3.1, yields the special case in which
r = 2 of the following general result:
mx -f ni2 * • • -r m
(3.4.2)
1 1 1 , , /?!•>
-f r . l ; n — j).
X inv(ml,...,mr+l\n)qn
+m r +
mr
£ Xiid = n
+mr 1
where /(£,) = *"if£»> £i+i and #(£,.) = 0 otherwise. The sum £™= T ~ X(£»)
is called the greater index of the permutation.
b2 >--->bm2 > 0
AIA2A3- - -AM
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
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
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. •
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
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
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.
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.
i = - QO i = - oo
TO
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
, A#, « V = G(N,
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 -
rO<n^ S/2.
Proof. Since
m{ -h * *'
lYl
2
we see that
\ml + m2 4- * * * 4- mr
is just
Examples
HJLt) = 1 t*.
7= 0
5. //,(**) = (-**;**)„.
6. Hn+l(t) = (I + 0 ^ ( 0 - (1 - qn)tHn^(t).
m\ \n
HJt)HHV) = I
(tyz)»
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
and Dn =
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)
- q j - q J + l
-'"-
Notes
References
4.1 Introduction
'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
*—•—•—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
Proof. Following the argument in the first proof of Theorem 3.1, we see
directly that
1 [n-V
^ l(;-1) (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))
(4 2 6)
M - l • ' -
= 1+
M(M -
= exp \{M - 1) log I 1 t-
\M(M - I) 2 lM2{M- I) 3
= exp{ + O .
[ I n \ n2
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
X c ( a , , . . . , « r ; m ) r 1 " « •••*;•
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).
/ ^ £/ nlM/«i
c(a1,...,ar;m) = £ ( - 1)' 1
/a 2 + m — i — 1\ /a r 4- m — i — 1
x
X c ( a , , . . . , a,; m ) / , * ^ - • • * / • •
-1
d - / , x i -t2y- -d-o
ar + m - i - 1\
u
" " ' (4.3.4)
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 <
(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
(h - n
>„-*(-i
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
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
c ( a , , . . . , a r ; w) = £ . N(a x ,. . . , a r ; n - j)
j = o\ J J
as desired. •
s=0 \ S
In 4- a r — 1 — 5
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
A - /t - 2\
= ( - \y( s I (by (3.3.9) with q = 1)
A + 1\
1. (4.4.6)
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
«r-l
c(ai,...,ar) = 2
h\ >0
r >0
+ /1,+ • • • + / i r \ /ar.,
^2
vai+i -f • • • + ar — k
*, + y
xi+l + • • * + ar — A:
9. If the polynomial
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
n \ / /i -h v
2
n — r) ' \m + n + f.i — s
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
Notes
References
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.
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.
The Hardy-Ramanujan-Rademacher
Expansion of p(n)
5.1 Introduction
C =
68
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
where
Ak(n)= X Oih,ke-2"inl"k
hmodk
(/i,fc)=l
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
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-
then
P(exp(2ni(h +
hW = - 1 (mod k\ (5.2.3)
(/- k\
- y - ) e x p ( - 7ri(i(2 - hk - h) + ^(/c - /c-1)(2/i - /i' + /i2/i')))
if /i odd,
]_^_[^]_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):
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
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
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:
2nihn
k= 1
2nn\
p(n) = exp 2
) exp ^ - —^— ]
A'
(5.2.14)
inn Inihn \
= 1
ih,k)= I
2nn\ 2nihn z
(/•.Ac) = 1
]
(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
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
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)
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
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
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
and our next task is to show that each of the four integrals / 2 , 73, 74, and 75
is negligible:
lou-plane
Figure 5.1
-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^
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)
(5.2.26)
where the constants implied in the 0 terms are absolute. Consequently, writing
2nihn\
p(n)= T_>
(M)=l
7-3/2 2nn
+o exp N -5/2
k=l
(M)=l
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 + )
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
Hence
2! 4! 6!
y2n+l
2
—E
= , - 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)* },.. '
Hence
1 Id e x p [ - 7r(2/3)*c/fc]
n{n - 1/24) U
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
(x-1/24)*
and since (5.2.33) is just (5.2.29), we see that Theorem 5.1 is established. •
Examples
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
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
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
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
h(z,s) = £ ( m z + nY
m>0
n>dm/c
s l u
• i -u - e~ du,
r{s)=
0
Res > 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/«.
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
2 + ^3
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
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,
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.
6.1 Introduction
= 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
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,
(6.1.6)
1
- ^ . (6.1.8)
a
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.
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
^ f ds; (6.2.6)
T"T(S)C(S + 1)Z>(5)
= -i + (D'(0)-D(0)logT)- + . . . ;
s s
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
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.)
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
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)|
+ C6|log>;|
(6.2.10)
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
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
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
m
r(n) = e j f(y -f 2nix)e2ninx dx + emRly (6.2.17)
-yfi
where
-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.
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
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)
where
and
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
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
^...^ = 0(m|a)|3) =
(6.2.29)
while
/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)
(6.2.35)
2TT
I = (1 + O(m-"•>)) (6.2.36)
m(a + 1)
where
min ( ^ - (6.2.37)
I
r(«) = exp 1 + - m - (D(0) - 1 ) l o gm
- + £>'(0) (2;rm(a + 1))"
x (1 + O(m-»>)) (6.2.38)
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
2. In order to compute explicitly the r(n) in Theorem 6.2, we use the well-
known logarithmic derivative:
/V=l \d\N
nr(n) = £ r(« —
where
00
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
< Z Z avvexp
k=1u=1
a/(3t+1)
)c- a - 1 [Xa- 2 - 1 (a + l ) a + 1 r ( a + l)C(a + 1)]
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
p{"H"{^ 1), n) = 2 ( ( r - 3 ) / 2 - ^ + f l 2 + - +
Notes
References
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.
7.1 Introduction
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,
103
Hence
F(x)= f j ^ - - (7.1.5)
= 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
x
LJ ( n . Y . n \ _ V 4 a { \ - xq ){axq )00(q )n
H kti(a,x,q) -
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
we see that
Hkti(a;x;q) - Hkti_l{a\x\q)
°o Jfc/!_fcn2 + n ™ n / v ^ n + 1 \ / r , ~ l \
n=l
.-o
00 v f c
LEMMA 7.2
Proo/.
Jk.fcl x; q) - Jkti-i(a',x;q)
Proo/. By (7.2.2)
n= 0
= (^)i1 Z ( - i)
n= 0
Proof.
^ ) J / i\na2kn+2kn2-(2i-l)n
+ 2ta»-(2l-3).
f2i-2+4ni-4«\"|
(1
x \q - <? +^ ^j )
V ^ i ? )oo y / _ j\n
2 2
(4 ;€ ) o
^*
/ou
^ ' •* n=0
X (1 _
n
n= 1
n / 2 (mod 4)
i - D(mod4k)
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).
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)
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
(7.3.4)
we see that
ck0(mn n) = 0. (7.3.5)
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)
1 )
(7.3.7)
(7.3.8)
.i(O;O;<z)= 1).
To prove (7.3.9), we define
(k-l)n (k-
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 - 1 )n _(k - 1 )n 2 + (k - i)n
=Z
11
" Z
X
_ fc-1 2k-1- 1 V H_
Cit — 1 )n (k - 1 )n1 + (k + ii - 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. •
.
2 2 3
- q (1 X1 ) (i )(1 )(1 J
q22_ </6
+
f-f q
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)
*4,i(m> n) =
dktk(m, n — 2m), (7.4.2)
and for 1 < / < fc
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
easily prove that the dki(m, n) are uniquely determined by (7.4.1), (7.4.2), and
(7.4.3).
Now we consider
That Eq. (7.4.1) is also true for the ekyi(m, n) follows from
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)
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
2k — 1 if fk is even,
2k if f: i
PI
n^0,±a(mod 7)
I ^
This is established by showing that if
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
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.
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.
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
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
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
(9 ~ ® (8.2.3)
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
Finally, if we define
and
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:
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.
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
00
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
= X P(C, n)q\
= £ p(C, n)q"
Therefore
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. •
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."
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
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({/<>)
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 ,
/>m7T2) e (</>2m7r3) e - • •
_ \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
/ f + 1 < fc - 1}.
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.
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.
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
and now we may eliminate y2(x) from our system of equations via
w 2 (x)
=Z
fc = 2
=1*21 \fc=2
+ • * * + 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
If the p2i{x),..., p2nW a r e a ^ identically zero, then (8.4.9)x and (8.4.9)2 imply
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
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
w2(x) + w3(x),
The first of these equations may be used to eliminate vv2(x) from the system.
Hence
- Pn(x)yi(x)) + w3(x);
W 4 (X),
+ Pr3(x)w3(x) + • • • + prr(x)wr(x).
3m
yi(xq ) + 0L(x)yi(xq2m) + P(x)yi(xqm) + y(x)^(x) = w4(x),
H0(x) = X jc*(«y<*>
7teC
«o
= X x # ( V ( n ) =fc(xqm',q)- (8.4.12)
where
H((x) = X x # ( J V ( 7 °
n'eCn u
• ' l
i) y y x#(n) /
h=l n'eCn,
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
^ 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
Examples
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
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.
9.1 Introduction
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
E(n) = Z ( - l) r
139
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.
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
if m = 0,
- l) m = 0 if m > 0.
E(n) = p(®, n\
(«)(X4" +I )
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
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)
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 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
= C(0,0)=l. •
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.
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
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.
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.
We next shall derive recurrences for these functions that will allow us
to identify them with expressions involving Gaussian polynomials. We start
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. •
(9.3.3)
(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
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
+ m * > - 1 , 6 - Ufi;N)
= m k J ( a - 1 , 6 - l ; n ; N - a - b + 1).
+ mkti(a - 1 , 6 - l;jx;iV)
= pkti(a- 1 , 6 - l ; / x - l ; i V - a - 6 + 1).
It is now a simple matter to translate Lemma 9.4 into relations among the
generating functions.
k^^Ub^U^q) if a-6>-(i-l),
a
and for \i ^ 1
= M f c > , 0 ; / i ; ^ ) = 0, (9.3.8)
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
B] U + B - l ] IA + B - 1 ] [A + B - 2
\ \ B - X \ \ \
B-X-2
f(A,r;X,Y,L) = qL(l~q^ ^+ }
[ 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 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
f(A,r;X,Y,L)-f(A,r- I; A, Y, L) - q \ + X + r- 2
\2A
jt0 I A + Y- 2
+ q
[A+Y-]
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). •
Mki(a, b;2n\q) —
-«•.„
a + b
•
[b - {2k -M)
(9.3. 10)
(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 +
a + b
(9.3.16)
a - (2/c + l)
Af»*,{ti - l,
=
\ _
2a + j - 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
(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 +
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)
(9.3.22)
(9.3.23)
for a ^ 1, r ^ - 1, \i > 0;
- i - 1 - (2fe 4- / - (2fc - 2/
(9.3.24)
(9.3.25)
and
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). •
li > 0; (9.3.28)
0; (9.3.29)
-, fi>0; (9.3.30)
—, fi>0. (9.3.31)
(«).
x Via + y] a*
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
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
E&»9"
00 00 00 00 00
- 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"-
for each n ^ 0.
s + 1
s + 1
s 4- 1
s+ 2
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 2. Are there sieves that can be used in studying some or all of the
partition functions found in Chapters 7 and 8?
Examples
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
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 :1 2 )* y -r—,• y fl
, - -—• (see Example 17, Chapter 2)
( U t ( )
n=0
=
, - ? . ( ~ 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.
Notes
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.
10.1 Introduction
159
4V = 5n ( 7 r A,^. dO.1.3)
however,
24-243 = I(mod7 3 ).
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).
since
m if r = 0(mod m\
O if r ^ 0(mod m).
n=Q n=0
(10.2.2)
First we have
3
i r
l
n I 1 [ y (1 + '^x1 + |2 V) 2 | _ A
1 °° 1
_ g*(3 - g*) Qo 1
Hence
where
In general, we have
Therefore
j ^ dO.2.5)
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)
7=0 11 - 4 ; - - -
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
co(m + 1) = co(m) = 1;
= cx(m) + m2m~l.
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
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
m + 1\ /m - k + 2\ Ik + 1
^ km - K^2 + fc) - 2 + 2/
= fc(m - Kfc + 1)) - 2 + 2)
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.
furthermore, (10.2.9) and (10.2.10) are best possible in that no higher power
of 2 divides b(4n) — b(n).
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),
(2fc2+1)-3*>0.
(mod 2).
2--3k
3k 22
~ c2(2k
qG(q)
\2k+3
Again by Theorem 10.1, 16|c1(2fc + 1), S\c2(2k + 1), 22J\cj(2k), also for
k> 1,
^ f ^ + 1 (mod 2).
i=o
Thus we have Theorem 10.2. •
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
Sr = 5U5<t>r(q) = Y 0 r (^ 1 / 5 p l ) (10.3.5)
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:
and if
5 = r , m 4- r2m 4- • • * 4- r n m ,
tn l z ' n *
2 -••• + (- iy-v,*! + (- o v = °.
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)!
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:
Therefore
5j - r + 1
5
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 =
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
and
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)
When n = 1,
-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)
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
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.
= 6 2r _ 1 , 1 -5-63
min r_ ltJ
55-5 5* - s + 1
= — 1 -f min - 1+
-5+ 1
^ - at$=l
- 4
2r
m=O
)f
m= 0
= V5 £ b2rtS<t>(q)g°(q)
s= 0
= £ b2r,,U5(<l>(q)g%q))
co ao
= E ff'(9)Zi*2,y»6.+u (io.3.i7)
whence
2rs<>6,+ l t + s - (10.3.18)
s=0
Proof. By the remarks preceding Theorem 10.4, we see that we must show
that
p(5am 4- O = 0(mod5 fl )
M(fS- dO.3.19)
s=l KH i H Jao
<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:
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
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
m= 0
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
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
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.
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:
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
179
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
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
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)
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
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
^J) (11.2.6)
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/
y^ VkAmi n)qm-
m=0
THEOREM 11.2
nk%,(n\q)W(k%r) =
where
s= 0 n+ i- s
= q
[ n+ i
(by (3.3.10)). Therefore
,r) = ( - ^
q)(q)'-(q) (n 2 12)
=
) (q)r(q)r+l"(q)r+k-i
and therefore
as desired. •
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).
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
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. •
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
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. •
ieS iJeS
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
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.
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)
lk\ lk\
-Mk(4)= 1 + 4*+ 4 , + , , (11.4.7)
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)
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
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
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
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
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
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
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
T : ( x l J x 2 , . . . , x k + l ) - (xk + l , x l 9 x 2 , . . . , x k )
Examples
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,
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.
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
202
(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:
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
h(t) = f(g(t)).
If we denote
= h = f
3/20201
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)
= t ^T" (12.3.5)
n\
The desired expression is
log # ( M ) = I = - ^ . (12.3.6)
G(M) = 1 4- f ^ w y . (12.3.11)
THEOREM T2.3
(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
and
Y"1 V"1 U X^ l
X2 2
' ' ' Xr '
0 0
.i"!
m = l 'W
Z ~ ( m - \)%{m). (12.3.16)
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
= - I ^ O n - 1)!W4 (12.3.17)
m=i m\
and this yields (12.3.13). •
+ j? r (l) 2 )
(12.3.18)
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.
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)
where we have merely classified each partition with largest part max(r, 5) ^
Hence if
00
r=0
00 n
r,s=0
r,s = 0 n = max(r,s)
n—r r^s
f
n=s
f
r= 0 n=r
F^uxy). (12.4.4)
(1 - u X l -xu)(i-yu)
-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 1 1 - xyu2
1 — xu j 1 — yu i
(1 — xw)(l — yu)
is a special case of
1 — X 1 X 2 * * 'XrUr
( 1 — M X i X l — M X 2 ) ' ' ( 1 — UXr)
(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).
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.
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.
\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.
212
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
(<?"- 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)
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 ) ,
l/ 4 = (C 4 1 ,0, 1,0,0,0,0,0,0),
U5 = (C 51 , 1 , 0 , 0 , 0 , 0 , 0 , 0 , 0 ) .
by Theorem 3.1. •
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
W = r1!r2!r3!-",
Sign A = ( - l)'2+2r3 + 3r 4 +- = ( _ iy~rl-r2-r3-•- = ^_ 1 ^(A)-#(A)
n! n\
—r^r i=
n —
ways. Consequently
_ ln\tn
n\/n - ni
nj\ n2
n\
(13.3.2)
(13.3.3)
THEOREM 13.3. The following formulas hold for the Bell numbers:
Bn = Yn(l,\,...,l) (13.3.5)
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,
^ B^X_ ^ /nVi,i,...,
which is (13.3.6). •
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)).
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
if and only if
Furthermore
if and only if
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
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
X 0 = Xv<*°>; (13.3.13)
= £ s(n,k)Xk; (13.3.15)
n
Yn V <\(n lrVY\ (X'X 1 1 f\\
j\. — ? kj^fl, /Cy^A jj^. \LD.J.L\J)
fc = 0
THEOREM 13.8.
but the S(n, k) are clearly uniquely determined by (13.3.16) and so since
(13.3.20) is of precisely the same form, we conclude that (13.3.17) is valid.
Equation (13.3.18) follows from (13.3.14) and (13.3.15) in exactly the same
way.
Finally, since there is only one single block partition of n = {1, 2 , . . . , n)
(namely, 0), we see that by (13.3.18)
MO, /) = 5(n, 1)
= ( - iy-l(n- 1)!,
which is (13.3.19). •
den i=l
and if T c F, we define
n}, (13.4.1)
7r}, (13.4.2)
DEFINITION 13.3. For each integer n and each partition X = (A1^2^3* * *^m)
of n we define
= sJLlsX2'-sXm.
The next theorem clarifies the relationship between functions whose sub-
script is a partition of n and those whose subscript is a partition of an integer.
THEOREM 13.10. For each partition n ofn,
K = |A(7r)|/cA(ff); (13.4.4)
an = Kn)\aXin). (13.4.6)
Proof. We write
Next we treat Sfn\ From (13.4.2) we see that fe£fnif and only if its kernel
is less refined than 7r, that is, if and only if/ is constant on blocks of n. Hence
if the blocks of n are Bl9 B2, B 3 , . . . , Bm, then
m
=
n s i B i i = su*
= I y(f\BMf\B2)'"y(f\Bj
-n/ i */)
i= 1 I onc-to-onc
\ functions
\ f.Bi-C
1=1
= Kn)\aX(ny •
Our next result will be seen to imply easily the fundamental theorem on
symmetric functions (Theorem 13.12).
THEOREM 13.11. For each partition n of n,
*« = £ * . ; (13A9)
which is (13.4.7).
<*n = Z S(a v n
* 7)fc«r
(
a \x > a v n
or
which is (13.4.8). •
p(x1,x2,x3,...)
= C
2- «i«2 •n m ' C (ni,n2 n m)
, 7T) v
n 2L /'(
and since the aMx) are products of the ah we have the desired result. •
Examples
f] (X - Zq{) = t U - Yqh).
U(X - \)(X -
satisfies
L(Xn) =
m k{m-h + k) _
m 4- n
h - it h
k\[ n- k
8. F r o m ( 1 3 . 3 . 1 6 ) it is easy to d e d u c e that
it is easy to s h o w that
y=0
1 - kt
Hence
1
(1 - 0 ( 1 - 2/)- • • ( ! - f c t )
( 1 2 3 s ) = (123) + ( 3 2 ) + ( 3 2 ) ( 1 3 2 3 3 3 ) = ( 1 3 3 ) + ( 2 3 ) -f ( 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 )
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).
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.
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.
230
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
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:
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.
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
X bkti(my n) = Bkti(n),
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):
THEOREM 14.1
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)
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. •
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
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)
5 4 3 3 3 2 2 11100...
(0,0) (0,1) (1,0) (1,1) (2,0) (1,2) (3,0) (2,1)
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 {*,}. •
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,
Table 14.1
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
n
n —r
we need only consider 0 < r ^ w/2. Furthermore
J = 1 + q + q2 +
+
hence we consider only 2 < r ^ n/2.
Table 14.2
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
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
[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
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
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.
^ ( 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
.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
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
^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
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
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
250
252