-

,A Division

M&Brooks/Cole of Wadsworth.

.\dvmced lnc.

B?pks &z?oi!yare

1~ 19x6 by Wadsworth. lnc,. Belmont. California 94002. All rights reserved. No art of this rm or by an> book may be reproduced. stored in a retrieval system. or transcribed, in any f@! means -electronic. mechanical. photocopying. recording, or otherwise-without the prier wrltten permission of the publisher, Wadsworth & BrooksXole Advanced Books S Software. Monterey, California 93940. a division of Wadsworth. Inc. Printed in the Llnited States of America

10 9 s 7 6 5 4 3 2 Library or Congrcss

t
Data

Catalogiog-in-Publication

---7

Stanley. Richard Enumeratlve

P.. [date] combinatorics enumeration 51 l.62 (y. 1) problems. 86-14 1. Title.

lncludes lndcxes. 1, Comb~natorial QA 164.8.S73 19S6 ISElN 0-5~~-0bS~h-5

It is regrettable that a book, once published and on the way to starting a IXe of its own, cari no longer bear witness to the painful choices that the author had to face in the course of bis writing. There are choices that confront the writer of every book: who is the intended audience? who is to be proved wrong? who will be the most likely critic? Most of us bave indulged in the idle practice of drafting tables of contents of books we know will never sec the light of day. In some countries, some such particularly imaginative drafts bave actually been sent to press (though they may not be included among the authors list of publications). In mathcmatics. however, the burden ofchoice faced by the writer is SOheavy as to turn off a11but the most courageous. And of a11 mathematics. combinatorics is nowadays perhaps the hardest to Write on, despite an eager audience that cuts across the party lines. Shall an isolated special result be granted a section of its own? Shall a fledgling new theory with as yet sparse applications be gingerly thrust in the middle of a chapter? Shall the author yield to one of the contrary temptations of recreational math at one end, and categorical rigor at the other? or to the highly rewarding lure of the algorithm? Richard Stanley has corne through these hurdles with flying colors. It has been said that combinatorics has too many theorems, matched with very few theories; Stanleys book belies this assertion. Together with a sage choice of the most attractive theories on todays stage, he blends a variety of examples democratically chosen from topology to computer science, from algebra to complex variables. The reader will never be at a loss for an illustrative example, or for a proof that fails to meet G. H. Hardy3 criterion of pleasant surprise. His choice of exercises will at last enable us to give a satisfying reference to the colleague who knocks at our door with his combinatorial problem. %ut best ofall, Stanley has succeeded in dramatizing the subject, in a book that will engage from start to finish the attention of any mathematician who will open it at page one. Gian-Carlo Rota
V

areas, this situation arises quite frequently. As a result, 1 have made a special effort in this book to include coverage of topics from enumerative combinatorics that arise in other branches of mathematics. The exercises found at the end of each chapter play a vital role in achieving The#%!= exere&e!s (!%+y; with dif#%X&~ r&ings of 1 - to 3 -) may be attempted by students using this book as a text; the more difficult exercises are not really meant to be solved (though some readers will undoubtedly be unable to resist a real challenge), but rather serve as an entry into areas that are not directly covered by the text. 1 hope that these more difficult exercises will convince the reader of the depth and the wide applicability of enumerative combinatorics, especially in Chapter 3, where it is by no means L priori evident that partially ordered sets are more than a convenient bookkeeping device. Solutions or references to solutions are provided for almost a11 of the exercises. The method of citation and referencing is, 1 hope, largely self-explanatory. Al1 citations to references in another chapter are preceded by the relevant chapter number. For instance, [3.16] refers to reference 16 in Chapter 3. 1 have included no references to outside literature within the text itself; a11such references appear in the Notes at the end of each chapter. Each chapter bas its own list of references. while the references relevant to an exercise are given separately in the solution to that exercise. Many people have contributed in many ways to the writing of this book. Special mention must go to G.-C. Rota for introducing me to the pleasures of enumerative combinatorics and for his constant encouragement and stimulation. 1 must also mention Donald Knuth, whose superb texts on computer science inspired me to include a wide range of solved exercises with a difficulty level prescribed in advance. The following people have contributed valuable suggestions and encouragement, and 1 thank them: Ed Bender, Lou Bil]era, Anders Bjorner, Thomas Brylawski, Persi Diaconis, Dominique Foata, Adriano Garsia, Ira Gesse], Jay Goldman, Curtis Greene, Victor Klee, Pierre Leroux, and Ronald C. Mullin. In addition, the names of many whose ideas 1 have borrowed are mentioned in the Notes and Exercises. 1 am grateful to a number of typists for their lne preparation of the manuscript, including Ruby Aguirre, Louise Balzarini, Margaret Beucler, Benito Rakower, and Phyllis Ruby. Finally, thanks to John Kimmel of Wadsworth & Brooks/Cole Advanced Books & Software for his support and encouragement throughout the preparation of this book, and to Phyllis Larimore for her careful editing. For lnancial support during the writing of this book 1 wish to thank the Massachusetts Institute of Technology, the National Science Foundation, and the Guggenheim Foundation. Richard Stanley

Contents

Votatiou

Chupter

1
1.1 1.2 1.3 1.4

What 1s Enumerative Combinatorics?

How to Count Sets and Multisets Permutation Statistics Way

The Twelvefold Notes References

A Note about the Exercises Exercises Solutions to Exercises

Chapter

2
2.1 2.2 2.3 2.4 2.5 2.6 2.1

Sieve Methods
Inclusion-Exclusion Examples and Special Cases with Restricted Positions

Permutations Ferrers Boards F-partitions Involutions Determinants Notes References Exercises Solutions

and Unimodal

Sequences

to Exercises

C/zq~

3
3.1 3.2

Partially Ordered Sets

Basic Concepts New Posets from Old v

96 96 100

Notation

3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16

Distributive

Lattices Lattices

105 110 113 116 117 124 Lattice 126 129 131 133 135 Functions Enumeration 140 147 149 152 153 complex numbers nonnegative integers

Chains in Distributive The Incidence

Algebra of a Locally Finite Poset Formula Mobius Functions Algebras

The Mobius Inversion Techniques

for Computing

Lattices and Their Mobius The Mobius Function

of a Semimodular

Zeta Polynomials Rank-selection R-labelings Eulerian Binomial Posets Posets and Generating to Permutation

positive integers rational numbers

real numbers integers the set { 1,2,. . . ,rz), for 11 E N (SO 10] = @) for integers i 5 j, the set {i. j + 1.. . .j: greatest integer least integer >x Ut} c R, where a1 < ... < Ut delta, equal to 1 if i = j and 0 otherwise 5 Y

An Application Notes References Exercises Solutions

to Exercises

174 202 202 204 208 210 211 Diophantine Equations 221 241 260 263 264

a11used for the number of elements of the lnite set X theset {ui,...,

Chapter 4
4.1 4.2 4.3 4.4 4.5 4.6 4.1

Rational Generating Functions

Rational Further Power Series in One Variable Ramilcations

the Kronecker

equals by delnition image of the function A or linear transformation A with A

Polynomials Quasi-polynomials P-partitions Linear Homogeneous The Transfer-matrix Notes References Exercises Solutions to Exercises

kernel of the homomorphism

trace of the linear transformation

the lnite leld (unique up to isomorphism)

q elements

Method

direct sum of the vector spaces (or modules, rings, etc.) F ring of polynomials in the indeterminate in the integral domain R x with coeflcients

ring of rational functions in x with coefficients in R (R(x) is the quotient leld of R[x] when R is a field) ring of forma1 power series In>0 U~Xin x with coefficients u,, in R ring of forma1 Laurent series 1 k q~ uY, for some ne E Z, in x with coefficients un in R (R((x)) is the quotient leld of R [[x]] when R is a field) xi

275 293 296

Appendix Index

Graph Theory Terminology

CHAPTER

W_tie Combinatorics?

1.1

How to Count
The basic problem of enumerative combinatorics is that of counting the number of elements of a finite set. Usually we are given an infinite class of Inite sets Si where i ranges over some index set 1 (such as the nonnegative integers N). and we wish to Count the number j(i) of elements of each Si simultaneously.* Immediate philosophical diflculties arise. What does it mean to Count the number of elements of Si?There is no delnitive answer to this question. Only through experience does one develop an idea of what is meant by a determination of a counting function j-(i). The counting function j(i) cari be given in several standard ways: 1. The most satisfactory form of j(i) is a completely explicit closed formula involving only well-known functions, and free from summation symbols. Only in rare cases will such a formula exist. As formulas for j-(i) become more complicated, our willingness to accept them as determinations of j(i) decreases. Consider the following examples. 1.1.1 Example. For each n E N, let j(n) be the number of subsets of the set [n-j = {1,2,..., n}. Then j(n) = 2, and no one will quarrel about this being a satisfactory formula for j(n). 1.1.2 Example. Suppose n men give their n hats to a hat-check person. Let j(n) be the number of ways that the bats cari be given back to the men, each man receiving one bat, SO that no man receives his own hat. For instance, j(l) = 0, f(2) = 1, j(3) = 2. We will see in Chapter 2 that j-(n) =

n!i~f(-l)i/i!.

This formula for j(n) is not as elegant as the formula in Example 1.1.1, but for lack of a simpler answer we are willing to accept (1) as a satisfactory formula. In 1

CtYJpter1

What

1s Enumerati\e

Combinatorics?

1.1

How to Count

fact. once the derivation of (1) is understood (using the Principle of Inclusion Exclusion), every term of (1) bas an easily understood combinatorial meaning. This enables us to understand (1) intuitively, SO our willingness to accept it is enhanced. We also remark~tmat~ it follows easily from (Il that J(H) is_thee intcger to n!/e. Ths s certamy a simple explicit formula. but it bas the disadvantage of being *non-combinatorial; that is, dividing by e and rounding off to the nearest integer bas no direct combinatorial significance. 1.1.3 Example. Let ,Jrr) be the number of 11x rz matrices M of zeros and ones such that every row and column of M bas three ones. For example, J(O) = _Q1) = j(2) = 0, ,f(3) = 1. The most explicit formula known at present for f(n) is

4. Thc most useful but most diflcult to understand method for cvaluating J(i) is to give its ~~~~~~~~~ri~~~~,~~~~~~~j~~. We will not dcvelop in this chapter a rigorous abstract theory of generating functions, but will instead content ourselves with an abject that represents a counting function j(i). Usually this abject is a $NY~~LZ/ seritx The two most common types of generating functions are orditwyy~ generatit~~g /imrious and uponentiui genetxting ,fiuwtions. If 1 = ,Y%.then the ordinary gencrating function of J()r) is the forma1 power series
puver

while the exponential

generating

function of J(ri) is the formai powcr stries

where the sum is over a11(11+ 2)(/1 + l),? solutions to 2 + /1 + ;J = jr in nonnegative intcgers. This formula givcs very little insight into the behavior of J(ti). but it does allow one to compute ,f(t~) much fastcr than if only the combinatorial delnition of j(tr) were used. Hence with some reluctance we accept (2) as a determination of f(fr). Of course if someone were later to prove ~(II) = ~I(N- I)()l - 2)/6 (rather unlikely), then our enthusiasm for (2) would be considerably diminished. 1.1.4 Exampie. There are actually formulas in the literature (nameless here for evermore) for certain counting functions ~(II) whose evaluation rcquire listing a11 (or aimost ah) of thc f(tz) abjects being counted! Such a formula. is completeiy worthless. 2. A rccurrence for j(i) may be given in terms of previously calculated.f( j)+s, thereby giving a simple procedure for calculating f(i) for any desired i E I. For instance, let ,f(n) be the number of subsets of [n] that do not contain two consecutive integers. For example, for n = 4 we have the subsets @, { 1 ), {2), {3;. [4;., $31, {1,4J, {2,41, SO f(4)= 8. It is easily seen that J(a) = f(n - 1) + ,f(n - 2) for n 2 2. This makes it trivial, for example, to compute f(20). On the other hand, it cari be shown that

(If 1 = p. thc positive integers, thcn these sums begin at 11= 1.) These power series are callcd formai* because we are not concerned with lctting Y takc on particular valtics. and we ignore questions of convergence and divergence. The term _Yor .Y ti! merely marks the place where ~(II) is written. If F(_~I = x,,>,, o,,.Y. we cal1 u,, the cwjficimt of _Y in F(.Y) and Write

Similarly

ke cari deal Lvith gcnerating

functions

of scveral variables.

such as

(which may be considered as ordinary in the indices 1. jrr and exponential in n), or even of infinitely many variables. In this latter case every term should involve only lnitely many of the variables. Why bother with generating functions if they are merely another way of writing a counting function? The answer is that we cari perform various natural operations on generating functions that have a combinatorial signifcance. For instance, we cari add two generating functions (in one variable) by the rule

where T = *(l + fi), T = i(l - 3). This is an explicit answer, but because it involves irrational numbers it is a matter of opinion whether it is a better answer than the recurrence f(n) = _/(a - 1) + _/(n - 2). 3. An estimate may be given for f(i). If 1 = IV, this estimate frequently takes the form of an us~~m~r&cfirrrz~/u~(~) - g(n), where g(n) is a familiar function. The notation f(n) - g(n) means that lim_ f(n)/g(n) = 1. For instance, let f(n) be the function of Example 1.1.3. It cari be shown that Rr) - e -z36-n(3n)!. For many purposes this estimate is superior to the explicit formula (2).

Similarly, we cari multiply generating

functions

according

to the rule

where Ci = xy=,, uibn_i> or

1.1

HO~

to Count

where & = L;E0 (y)~$,,-~, with (y) = ~!/tt!(tz - i)!. Note that these operations are just what we would obtain by treating generating functions as if they obeyed the ordinary laws of algebra, such as _xixj = xi+j. These operations coincide with the addition and multiplication of functions when the power series converge for ?Il@roll%iaZ~vaIues of~~~~anu iney ooey such famlia~rlaws of algbra as associativity and commutativity of addition and multiplication, distributivity of multiplication over addition, and cancellation of multiplication (i.e., if F(~)G(X) = F(x)H(x) and F(x) # 0, then G(x) = H(x)). In fact, the set of a11 forma1 power series 1 z 0 ux with complex coefficients un forms a (commutative) integral domain under the operations just delned. This integral domain is denoted by C [ [x]]. (Actually, C [[.Y]] is a very special type of integral domain. For readers with some familiarity with algebra, we remark that C[[x]] s a principal ideal domain and therefore a unique factorization domain. In fact, every ideal of C[[x]] has the form (.Y) for some n 2 0. From the viewpoint of commutative algebra, C[[x]] is a one-dimensional complete regular local ring. These general algebraic considerations will not concern us here; rather we will discuss from an elementary viewpoint the properties of C[[xJ] that will be useful to us.) Similarly, the set of forma1 power series in the rn variables xi, . . , xm (where m may be infinite) is denoted C[ [x l,. . . , .Y,,,]] and forms a unique factorization domain (though not a principal ideal domain for m 2 2). It is primarily through experience that the combinatorial signitcance of the algebraic operations of C [ [.Y]] or C [ [xi,. . . , x,,,]] is understood, as well as the problem of whether to use ordinary or exponential generating functions (or various other kinds discussed in later chapters). In Section 3.15, we will explain to some extent the combinatorial signifcance of these operations, but even then experience is indispensable. If F(x) and G(x) are elements of C[[?C]] satisfying F(x)G(x) = 1, then we (naturally) Write G(x) = F(x)-. (Here 1 is short for 1 + Ox + Ox2 + ....) It is easy to see that F(x)- exists (in which case it is unique) if and only if u0 # 0, where F(x) = 1 z ,, ux. One commonly writes symbolically u0 = F(O), even though F(x) is not considered to be a function of x. If F(O) # 0 and F(~)G(X) = H(x), then G(x) = F(x))lH(x). More generally, the operation -i satisles a11 the familiar laws of algebra, provided it is only applied to power series F(x) satisfying F(O) # 0. For instance, (F(~)G(X))- = F(x)-G(x)-, (F(x)-r)-t = F(x), and SO on. Similar results hold for C[[xt, . . . , xn J]. 1.1.5 Example. Let (~~2,,~x)(l - CYX) = ~~2,,c#, where a is a non-zero complex number. Then by the detnition of power series multiplication, cn = 1, rr=o
= 0, n 2 1.

This formula cornes as no surprise; it is simply the formula (in a forma1 setting) for summing a geometric series. Example 1.1.5 provides a simple illustration of a general principle that, informally speakinz. states that if we have an identity involving power series that is valid when the power series are regarded as functions (SO that the variables are suffciently small complex numbers), then this identity continues to remain valid when regarded as an identity among forma1 power series, prokfed the operations involved in the formulas are well-dehned for forma1 power series. It would be unnecessarily pedantic for us to state a precise form of this principle here, since the reader should have little trouble justifying in any particular case the forma1 validity of our manipulations with power series. We will give several examples throughout this section to illustrate this contention. 1.1.6JXxample. The identity (3) is valid at the function-theorctic level (it states that eXeeX = 1) and is well-defined as a statement involving forma1 power series. Hence (3) is a valid forma1 power series identity. In other words (equating coeflcients of x#n! on both sides of (3)) we have

To justify this identity directly from (3), we may reason as follows. Both sides of (3) converge for a11.K6 :C, SO we have

But if two power series in x represent the same function J(x) in a neighborhood of 0, then these two power series must agree term-by-term, by a standard elementary result concerning power series. Hence (4) follows. 1.1.7 Example. The identity

, ,

;. (x + lY/rl! = eFo x/rI! is valid at the ,function-theoretic level (it states that eX+ = e.eX), but does not make sense as a statement involving forma1 power series. There is no formuf procedure for writing En>0 (x + ly/rr! as a member of C[[x]]. Although the expression xnzo (x + l)/n! does not make sense fir&ly, there are nevertheless certain infinite processes that cari be carried out formally in C [ [xl]. (These concepts extend straightforwardly to C [ . . . , x,J], but for simplicity we consider only C[[x]].) TO deline these processes, we need to put some additional structure on C[[x]]-namely, the notion of conuergence.

1 a - a(a-l)

Hence En>0 ax = (1 - ax)-, which cari also be written 1 anxn c ~ X 1 -ax PI20

[xr ,

From an algebraic standpoint, the delnition of convergence is inherent in the statement that C[[.Y]] is complete in a certain standard topology that cari be put on C[[X]]. However. we will assume no knowledge of topology on the part of the reader and will instead give a self-contained, elementary trgstment of con vcgeEeC If-FI(x), F2(.y), . is a sequence of forma1 power series, and if F(X) = ~20a,$ is another forma1 power series, we say by delnition that Fi(x) conuerges to F(X) as i -t x. wfitten Fi(~) -+ F(X), provided that for a11n 2 0 there is a number (j(n) such that the coeflcient of _Yin Fi(x) is an whenever i 2 J(tz). In other words, for every n the sequence [ F1 (.y), [ F* (.Y), . n n of complex numbers eventually becomes constant with value a. An equivalent definition of convergence is the following, Deline the @ree of a non-zero forma1 power series F(X) = 1 n2,, a,,.~, denoted degF(x), to be the least integer n such that un # 0. Note that deg F(~)G(X) = deg F(x) + deg G(X). Then Fi(x) converges if and only if limi_E deg(F,+,(x) - Fi(_~)) = co. We now say that an infinite sum 1. ,20f+) has the value F(X) provided that x>+, Fi(x) -+ F(+x). A similar definition is made for the infinite product nj> l Fj(.yl. TO avoid unimportant technicalities we assume that in any infinite product ~j~ 14(x), each factor q(x) satisles c(O) = 1. For instance, let e(x) = aj-x. Then for I 2 u, the coeflcient of X in E;=u Fj(x) is a,,. Hence xjZO Fj(.y) is just the power series 1 ,,20 a&. Thus we cari think of the forma1 power series xnZ0 a,,.~ as actually being the sum of its individual terms. The proofs of the following two elementary results are left to the reader. 1.1.8 Proposition. The limj_ deg e(x) = CC. inlnite series 1. ,ZQ &(x) converges if and only if 0

~nx,U + $/n! does not converge in accordance with the above delnition. On the other hand, an expression like eeXml makes good sense fotmally, since it has the form F(G(x)) where F(x) = xnkOx;n! and G(x) = x,,> l x/n!.

where (i) = 2(2. - 1)...(2 - n + l)/f~!. In fact, we may regard 2 as an indeterminate and take (5) as the definition of (1 + F(.K)) as an element of C[[.Y,~_]] in (1 + F(.Y)) is a po/JXomia/ (or of C[;_l [[.Y]]; that is, the coefficient of .Y in 2). Al1 the expected properties of exponentiation are indeed valid. such as (1 + Fe))i+fl = (1 + F(.Y))(~ + F(.Y))~ (regarded as an identity in the ring C[[X,~,A~]], or in the ring C[[X]] where one takes j.. ~CC\. If F(x) = x,,kO a,,.?, deline the jtirmal deriuatire F(.Y) (also denotcd L or DF(x)) to be the forma1 power series xn2,, MZ.Y- = x,,ZO(~j + ~)u,,+~.Y.It is easy to check that a11 the familiar laws of differentiation that are well-defined formally continue to be valid for forma1 power series. In particular (F + G) = F + G (FG) = F!G + FG F(G(x)) = G(x)F(G(x)). We thus have a theory of fortna/ calchs for forma1 power series. The usefulness of this theory will become apparent in subsequent examples. We first give an example of the use of the forma1 calculus that should shed some additional light on the validity of manipulating forma1 power series as if they were actual functions of x. 1.1.11 Example. satisfying Suppose F(O) = 1, and let G(x) be the unique G(O) = 0. power series

1.1.9 Proposition. The infnite product nj2 1(1 + e(x)), where q(O) = 0, converges if and only if limj_m degq(x) = a. 0 It is essential to realize that in evaluating a convergent series xj20 4(x) (or similarly a product fl. ,> l q(x)), the coefficient of x for any given n cari be computed using only jhite processes. For if j is suflcientiy large, say j > d(n), then deg4(x) > n, SO that

G(X) = F(x)/F(x),

(6)

The latter expression involves only a finite sum. The most important combinatorial application of the notion ofconvergence is to the idea ofpower series compositon. If F(x) = JTnzOa,,~ and G(x) are forma1 power series with G(O) = 0, deline the composition F(G(x)) to be the infinite sum xn20 ~C(X). Since deg G(x) = n.deg G(x) 2 n, we see by Proposition 1.1.8 that F(G(x)) is well-defned as a forma1 power series. We also see why an expression such as elX does not make sense forrnally; namely, the inlnite series

From the function-theoretic viewpoint we cari salve (6) to obtain F(x) = exp G(x), where by definition exp G(x) = En>,, G(x)/~!. Since G(O) = 0 everything is well-defned formally, SO(6) should remain equivalent to F(x) = exp G(X) even if the power series for F(x) converges only at x = 0. How cari this assertion be justified without actually proving a combinatorial identity? Let F(x) = 1 + xn2 l ux. From (6) we cari compute explicitly G(x) = xRZ l &x, and it is quickly seen that each b is a polynomial in lnitely many of the ais. It then follows that if exp G(x) = 1 + En2 l cnx, then each c will also be a polynomial in finitely many of the ais, say c = P~(U~, a*, . . . , a,,,), where m depends on n. Now we know that F(x) = expG(x) provided 1 + x21 UX converges. If-two Taylor series convergent in some neighborhood of the origin represent the same function, then their coeflcients coincide. Hence a = ~,,(a~,u2,. . . , a,,,) provided 1 + En2 I U~X

&.*

..

_...

converges. Thus the two polynomials un and pJui,+,. . . ,u,,,) agree in some neighborhood of the origin of C, SO they must be equal. (It is well-known that if two complex polynomials in m variables agree in some open set of Cm, then they are identical.) Since un = p,,(~i, u2,. . . , u,,,) as polynomials, the identity #%KG, ,%Z,..,a.. a.%...-J, ,,cC+* rVwb* C?C,* 1V>. There is an alternative method for justifying the forma1 solution F(X) = exp G(x) to (6), which may appeal to topologically inclined readers. Given G(X) with G(O) = 0, deline F(x) = expG(x) and consider a map 4 : C[[x]] -+ C[[_Y]] delned by #(G(x)) = G(x) - $$. One easily vernies the following: (a) if G converges in some neighborhood of 0 then &G(x)) = 0; (b) the set 9 of a11power series G(.x) E C[[x]] that converge in some neighborhood of0 is dense in C[[S]], in the topology delned above (in fact. the set C[x] of polynomials is dense); and (c) the function 4 is continuous in the topology delned above. From this it follows that C$(G(X))= 0 for a11 G(.x)EC[[.X]] with G(O) = 0. We now present various illustrations in the manipulation of generating functions. Throughout we will be making heavy use of the principle that forma1 power series cari be treated as if they were functions. 1.1.12 Exampk. Find a simple expression for the generating function xn,0 u,$, where u,, = ~1, = 1, un = un-, + u,,_* if n 2 2. We have F(x) =

1.1.14 Example. Let P(~I) be the Mobius function of number theory; that is, ~(1) = 1, p(n) = 0 if n is divisible by the square of an integer greater than one, and p(n) = (- ly if n is the product of r distinct primes. Find a simple expression for the power series F(x) = n (1 - xn)-p(nfl. II>I First let us make sure that F(x) is well-detned by Example 1.1.10 that (8) as a forma1 power series. We have

= 1 + H(x), where deg H(x) = n. Hence by Proposition Note that (1 - .Y)-pn~n 1.1.9 the inlnite product (8) converges, SOF(x) is well-defined. Now apply log to (8). In other words, form IB$!F(x), where log(1 + x) = t
21

(- lY-x/n, for the natural logarithm. We obtain

the power series expansion log F(x) = log n

21 = & log( 1 -

(1 - x)-fl(
x)-PC

= -zi

$)log(l

- _Xfl)

= 1+

.K +

_K(F(.K)-

1) +

x2F(x).

-;lpJq).

Solving for F(x) yields F(x) = l/( 1 - .Y- x). The coeflcient of xm in the above power series is 1.1.13 Exampie. Find a simple expression for the generating function F(X) = ~,,~ou&/~!, where u. = ai = 1, un = a_i + (n - l)~_~ if n 2 2. We have F(x) = 1
Il20

;F

UX/?Z!
x +

/44, In that

where the sum is over a11 positive integers d dividing m. It is well-known

= 1+

1 UX/r7!
22

&&l)
In

= 1 + x + 1 (un-1 + (n - l)un_&c/n!.
22

0,

m =

otherwise. of this miraculous

(7) Hence logF(x) = x, SO F(x) = ex. Note that the derivation formula involved only for~~u/ manipulations. 1.1.15 Exampie. satisfying i
k=O

Let G(x) = 1 ,,2Z un_i.?/~! and H(x) = znkZ (n - l)~~-~x/n!. Then G(x) = xnkZ ~~_ix-/(n - l)! = F(x) - 1, and H(x) = xn21 u,,-~x~-/(~ - 2)! = xF(x). Hence if we differentiate (7) we obtain F(x) = 1 + (F(x) - 1) + xF(x) = (1 + x)F(x). The unique solution to this differential equation satisfying F(O) = 1 is F(x) = exp(x + 4~~). (As shown in Example 1.1.11, solving this differential equation is a purely forma1 procedure.)

Find the unique sequence u. = 1, ai, uZ, . . . of real numbers

U&z_~= 1

(9) side of (9) as the coeflcient

for a11 n E lV. The trick is to recognize the left-hand

10

Chapter

What

1s Enumerative

Comhinatorics?

1.1

HO~ to Count

11

(for someone suflciently versed in complex variable theory) to obtain the asymptotic formula

U,, = ( - 1y

(_+) Il

= (_ ,Y( -;)(

-;)(-;j...( n!

-y)

l 3 .5 . (2t1 2% !

1)

No other method of describing ~(II) makes it SO easy to determine these fundamental properties. Many other properties ofj(n) cari also be easily obtained from the generating function: for instance, we leave to the reader the problem of evaluating, essentially by inspection of (lO), the sum

NO~ that we have discussed thc manipulation of forma1 power series. the question arises as to the advantages of using generating functions to represent a counting function j(tt). Why, for instance. should a formula such as

be regarded as a determination of ,f(/t)? Basicaily, the answer is that there are many standard, routine techniques for extracting information from generating functions. Generating functions are frequently the most concise and eftcient way of prescnting information about their coefficients. For instance, from (10) an experienced enumerativc combinatorialist cari tel1 at a glancc the fohowing: 1. A simple recurrence for ,f(tz) cari be found by differentiation. Namely. we obtain

n;. ,/(tz).K-/yn Equating coefficients

l)! = t 1 + x)e- r+x-2, = ( 1 + .y)& of Yn! yields n 2 1. cari be obtained

j(tl).Y.tl!.

.fCtl+ 1) = _f(tz) + nf(n - l),

2. An explicit Namely, formula

for f(n)

from J?+(~~) = c&~.

Thercfore vve are ready to accept dhe generating function exp(.y + q) as a satisfactory determination of f(n). This completes our discussion of generating functions and more generally the problem of giving a satisfactory description of a counting function ,/(tr). We now turn to the question of what is the best way to pr~~~cthat a counting function bas some given description. In accordance with the Princip]e from other branches of mathematics that it is bctter to exhibit an explicit isomorphism between two abjects than merely to prove that they are isomorphic. we adopt the general principle that it is better to exhibit an cxplicit onc-to-one correspondence (bijection) between two Imite sets than merely to prove that they bave the same number of elements. A proof that shows that a certain set S bas a certain number ttt of elements by constructing an explicit bijection between S and some other set that is known to have ttl elements is called a cotnbinatorial prmf or bijectioe proof. The precise border between combinatorial and non-combinatorial proofs is rather hazy. and certain arguments that to an inexperienced enumerator will appear non-combinatorial will be recognized by a more experienced counter as C-,, icombinatorial, primarily because he or sFis aware of certain standard tech-_ .-.--niques for converting apparently non-combinatorial arguments into combinatorial ones. Such subtleties will not concern us here, and we now give some cleartut examples of the distinction between combinatorial and non-combinatorial proofs. 1.1.16 Example. Let n and k be fixed positive integers. How many sequences (XI, X2.. . . , XJ are there of subsets of the set [n] = { 1,2,. . , nl such that XI n Xl n ... n Xk = @? Let f(k, n) be this number. If we were not particuiarly inspired we could perhaps argue as follows. Suppose XI n X2 n . . . n Xk_l = T, wherelTl=~.If~=Xi-T,thenY~n...nY~_~=~and~~[n]-T.Hence there are f(k - 1,n - i) sequences (XI,. . .,Xk_l) such that XI n X2 n ... n Xk-l = T. For each such sequence, Xk cari be any of the 2-i subsets of [n] - T. As is probably familiar to most readers and will be discussed later, there are (y) = n!/i!(n - i)! i-element subsets T of [n]. Hence

SO that

3. Regarded as a function of a complex variable, exp(.x + g) is a nicelybehaved entire function, SO that standard techniques from the theory of asymptotic estimates cari be used to estimate f(n). As a first approximation, it is routine