Software for enumerative and analytic combinatorics

Andrew MacFie

School of Mathematics and Statistics Carleton University, Ottawa, Canada

Thursday 26

th

April, 2012

Honours project (MATH 4905) Advisors: Daniel Panario and Brett Stevens

Abstract
We survey some general-purpose symbolic software packages that implement algorithms from enumerative and analytic combinatorics. Software for the following areas is covered: basic combinatorial objects, symbolic combinatorics, Plya theory, combinatorial species, and asymptotics. We describe the capabilities that the packages oer as well as some of the algorithms used, and provide links to original documentation. Most of the packages are freely downloadable from the web. Keywords: software, package, symbolic computation, combinatorics, counting, enumeration.
Note: In this project, to refer to webpages we place URL links in footnotes, and for all other types of referent we use standard endnote references.

Contents
1 Introduction 2 Enumerative and analytic combinatorics
2.1 Counting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.1 2.1.2 2.1.3 2.2

3 4
4 4 5 6 7

Exact vs. asymptotic counting . . . . . . . . . . . . . . . . . . . . . . Generating functions vs. their coecients . . . . . . . . . . . . . . . . Bijective combinatorics vs. manipulatorics . . . . . . . . . . . . . . .

Enumeration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

[email protected]
1

3 A few notes on the following packages 4 Basic combinatorial objects

4.1 Mathematical background 4.1.1 4.1.2 4.2 4.3 4.4 4.5 Orderings Gray codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7 9
9 9 9 10 10 11 13

Combinat (Maple)

Combinatorial Object Server . . . . . . . . . . . . . . . . . . . . . . . . . . . Combinatorica: basic combinatorial objects . . . . . . . . . . . . . . . . . . . Sage: basic combinatorial objects . . . . . . . . . . . . . . . . . . . . . . . .

5 Symbolic combinatorics
5.1 Mathematical background 5.1.1 5.1.2 5.1.3 5.2 5.3 5.4 Labeled classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . Unlabeled classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Multiple parameters

14
14 15 16 17 18 19 20

Combstruct

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Encyclopedia of Combinatorial Structures

Lambda-Upsilon-Omega (LUO): symbolic combinatorics

6 Plya theory
6.1 6.2 6.3 6.4 6.5 6.6 Mathematical background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A note on Plya theory packages COCO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Combinatorica: Plya theory . . . . . . . . . . . . . . . . . . . . . . . . . . . GraphEnumeration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . PermGroup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

22
22 26 26 26 28 29

7 Combinatorial species
7.1 7.2 7.3 Mathematical background 7.1.1 Weighted species . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Aldor-Combinat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sage: combinatorial species

30
30 33 35 36

8 Asymptotics
8.1 Mathematical background 8.1.1 8.1.2 8.1.3 8.1.4 8.2 8.3 8.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . Asymptotic scales and series . . . . . . . . . . . . . . . . . . . . . . . Singularity analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . Saddle-point asymptotics . . . . . . . . . . . . . . . . . . . . . . . . . Power series coecients and the radius of convergence . . . . . . . . .

38
38 38 39 39 40 40 41 42

Gdev . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Lambda-Upsilon-Omega (LUO): asymptotics . . . . . . . . . . . . . . . . . . MultiSeries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

9 Conclusions References Appendix

Maple 16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mathematica 8

43 44 47
47 47

Introduction
1
posted to his website in 1999, Doron Zeilberger challenged mathemati-

In an opinion article

cians to rethink the role of computers in mathematics. Everything that we can prove today will soon be provable, faster and better, by computers, he says, then gives the following advice: The real work of us mathematicians, from now until, roughly, fty years from now, when computers won't need us anymore, is to make the transition from human-centric math to machine-centric math as smooth and ecient as possible. . . . We could be much more useful than we are now, if, instead of proving yet another theorem, we would start teaching the computer everything we know, so that it would have a headstart. . . . Once you learned to PROGRAM (rather than just use) Maple (or, if you insist Mathematica, etc.), you should immediately get to the business of transcribing your math-knowledge into Maple. If futurist Ray Kurzweil's predictions for articial intelligence progress are to be believed, Zeilberger's suggestions will turn out to be sound. (However, utilitarians would urge us to consider the risks such technology would present.

3 4

What is certain even today is that

those who use mathematics, given the rise of computers, can ask themselves if they are failing to capitalize on a productive division of labor between man and machine. It is no 67 longer necessary to spend three years of Sundays factoring the Mersenne number 2 1, like F. N. Cole did in the 1900s [18], and neither is it necessary to use error-prone pen and paper methods to perform an ever-growing set of mathematical procedures. To illustrate this statement, this project examines symbolic computation (a.k.a. computer algebra) software packages for enumerative and algebraic combinatorics. We start, in Section 2, with an overview of the elds of enumerative and analytic combinatorics. Then we go into more detail on the scope of the project in Section 3. In Sections 48 we cover packages relating to basic combinatorial objects, symbolic combinatorics, Plya theory, combinatorial species, and asymptotics. Finally, we oer concluding observations and remarks in Section 9.

1 2 3 4

http://www.math.rutgers.edu/~zeilberg/Opinion36.html http://en.wikipedia.org/wiki/Predictions_made_by_Ray_Kurzweil http://singinst.org/research/publications http://www.existential-risk.org/

Enumerative and analytic combinatorics

Welcome to the elds of enumerative and analytic combinatorics, a.k.a. combinatorial and asymptotic enumeration! In order to completely cover what mathematicians think of when they think of these elds (and to avoid saying enumerative or combinatorial), we break up our discussion into two parts which we call

enumeration, the more algorithmic side.

2.1 Counting

counting,

the more mathematical side, and

Counting, the oldest mathematical subject [16], is enumeration in the mathematical sense: the study of the cardinalities of nite sets. addition rule [1]: The most basic principle of counting is the

Proposition 1

(Addition rule)

. If S is a set and A1 , A2 , . . . , An is a partition of S , then

|S| = |A1 | + |A2 | + + |An |.

Other similar basic ways of counting may be familiar from introductory probability, and indeed, many concepts overlap between counting and discrete probability.

Elements of the body of work on counting can be roughly categorized based on three criteria: whether they deal with exact or asymptotic results, whether they speak in terms of generating functions or their coecients, and whether they make use of bijections or manipulations.

2.1.1 Exact vs. asymptotic counting Notation 1. Boldface symbols refer to (possibly terminating) 1-based sequences, i.e. a = (a1 , a2 , . . . ).
(S, p, N ) comprising a countable set S of objects, a sequence p = (p1 , p2 , . . . ) of functions pi : S Z0 , and a set N of sequences n for which p1 (n) = {s S : p1 (s) = n1 , p2 (s) = n2 , . . . } S is nite. The problem is 1 to answer the question For n N , how many objects does the set p (n) contain? The
Generally, counting problems involve a triple denition of an

exact answer

was given by Herb Wilf: a polynomial-time algorithm that

computes the number [16].

Example 1. If Sn is a permutation, then let cj () be the number of cycles of with size j . The signature of is c() = (c1 (), c2 (), . . . ). Let S be the set of all permutations, and for s S , let p(s) be the signature of s. Let n 1 and let n = ([j = n])j0 . Then |p1 (n)| = (n1)!. The expression (n1)! immediately suggests a polynomial-time algorithm to compute |p1 (n)|.
5

http://planetmath.org/encyclopedia/Combinatorics.html

(|p (n)|)nN ,
satises [16].

Exact 1

answers

can

be

classied

by

ansatz,

meaning

the

form

of

the

sequence

which is generally determined by the simplest

reccurence relation it

Exact answers are not the end of the story, however, partly because of applications to the analysis of algorithms.

p = (p) and N = ((0), (1), (2), . . . ), an asymptotic 1 answer is a relation between f : n |p (n)| and simple functions, which holds as n . Often the simple functions come from the logarithmico-exponential class L of Hardy [17, 20], and the relation is f (n) g(n) as n , where g L. A more substantial relation
In the single-parameter case, i.e. is a full asymptotic series [10]:

Denition 1. Given a sequence of functions g with gk+1 (n) = o(gk (n)) as n for all k, and real numbers c1 , c2 , . . ., the statement
f (n) c g(n) = c1 g1 (n) + c2 g2 (n) + c3 g3 (n) +

is called an

asymptotic series

for f , and it means

f (n) = O(g1 (n)) f (n) = c1 g1 (n) + O(g2 (n)) f (n) = c1 g1 (n) + c2 g2 (n) + O(g3 (n)) f (n) = c1 g1 (n) + c2 g2 (n) + c3 g3 (n) + O(g4 (n))

. . .

(as n .)
An asymptotic answer in the multiple-parameter case is more complicated; it generally involves (possibly just some moments of ) a continuous approximation to a discrete probability distribution as one parameter approaches innity, or an asymptotic relation which holds as one or more parameters approach innity at various rates.

2.1.2 Generating functions vs. their coecients

Given a triple

function

of

(S, p, N ), let f : N Z0 be dened f (n) = |p1 (n)|. A (type u) generating S with z = (z1 , z2 , . . . ) marking p is the element of the ring Q[[z]] F (z) =
n 0

f (n)u(n)z n ,

where

exponential Example 2. Let F (z) be the ordinary generating function for words of length n on the alphabet [1..k], with zj marking the number of occurrences of j, 1 j k. We have
n 0,
and we call it an

n n z n = z1 1 z2 2 .

We call

F (z)

generating function i u(n) = 1 for all 1 generating function i u(n) = (n1 !) for all n 0. an

ordinary

F (z) = (z1 + z2 + + zk )n .
5

To dene convergence of sequences of generating functions, the norm on formal power series used in this project is dened for

f (z) = 0

as

f (z) = 2k ,
and with

where

k = max j Z0 : i [0..j], [z i ]f (z, z, . . . ) = 0 , F (z) instead of working directly , and extracting f (n) exactly from a suently simple

0 = 0. f , for a variety of reasons [13, 38] F (z) can be done in F (z)

6

It can be ecient to initially make statements about representation of

polynomial time [16].

In addition, there are very widely applicable theorems for obtaining the asymptotics of

from

[13], which involve using the power series

F (z)

to dene a complex function

analytic at the origin.

Since generating functions are heavily used in both the exact and

asymptotic worlds, Wilf says in his book [38], To omit the analytical (i.e. asymptotic) parts of [counting with generating functions] . . . is like listening to a stereo broadcast of, say, Beethoven's Ninth Symphony, using only the left audio channel. The full beauty of the subject of generating functions emerges only from tuning in on both channels: the discrete and the continuous.

2.1.3 Bijective combinatorics vs. manipulatorics

To prove two sets have equal cardinality, often one of two methods is used. First, if one has algebraic expressions for the cardinality of each set, one may perform algebraic manipulations upon them until they are syntactically equivalent. Second, one may give an explicit bijection between them. The following example demonstrates each method on the same problem:

Example 3. A partition is called odd if all its parts are odd and it is called distinct if all its parts are distinct. Let f (n) and g(n) be the number of odd and distinct partitions of size n respectively, and let us dene the ordinary generating functions F (z) = n0 f (n)z n and G(z) = n0 g(n)z n . Since
G(z) = = = (1 + z n )
n0

1 z 2n 1 zn n0
n0 (1 n0 (1 z 2n )

1 = 1 z 2n+1 n0

z 2n ) 2n+1 ) n0 (1 z

= F (z),

we have [z n ]G(z) = [z n ]F (z), and thus f (n) = g(n), for all n 0. There is also a bijective proof of this fact, due to Glaisher [16], which we sketch. The function from distinct partitions
6

http://web.mit.edu/~qchu/Public/TopicsInGF.pdf
6

to odd partitions is dened as follows: Given a distinct partition, write each of its parts as 2r s, where s is odd, and replace each by 2r copies of s. This function is invertible, with inverse computable as follows: Take an odd part a which occurs m times, and write m in base 2, i.e. m = (sk s1 )2 , then replace the m copies of a by the k parts 2s1 a, . . . , 2sk a.
Arguably, manipulations are not combinatorics, hence the name manipulatorics. and other elds are blurry. Usually, bijections are used to give exact answers, but bijections can also be used in the context of asymptotics, as described in [4], for example. Standard textbooks for the eld of counting include [6, 9, 13, 34, 35]. Indeed, here the borders between algebra, analysis, combinatorics, the analysis of algorithms,

2.2

Enumeration

Enumeration is the eld of computer science dealing with triple

algorithms

that

generate

the

elements of nite sets. Various types of enumeration problems can be proposed for a given

(S, p, N );

the most common ones are ranking and unranking, random generation,

exhaustive listing, and iteration, which we dene in that order below. 1 1 We call an of p (n) S a bijection between p (n) and the integers 1 [1..|p (n)|]. A algorithm computes this bijection and an algorithm

ordering ranking

unranking

computes its inverse. In

random generation, discrete distributions (always uniform distributions in this project)

p1 (n),

are specied on the sets

and an algorithm is required to take as input n N 1 and return a random variate drawn from p (n) according to the distribution. Unranking algorithms can be used for random generation, since an integer can be generated at random and then unranked to give a random object. An

p1 (n).

exhaustive listing

algorithm takes as input

Generally, as with random generation, if the

n N and returns a list of all elements in p1 (n) are well dened, a brute force

algorithm is trivial, and the problem lies in designing an ecient algorithm. 1 is the problem of, given n N and an object s p (n), generating the next 1 object s in a certain ordering of p (n). It is related to the problem of exhaustive listing

Iteration

since any iteration algorithm immediately leads to an exhaustive listing algorithm, and more 1 importantly, nding a particular ordering of p (n) often leads to the most ecient exhaustive listing algorithms. If one employs an iteration algorithm repeatedly, in an exhaustive listing algorithm, one aims for an ordering that takes constant amortized time for each iteration. A

Gray code

is an ordering in which sucessive objects dier in some prespecied

small way and thus is perfect for exhaustive listing through iteration.

A few notes on the following packages

In this project we focus on general-purpose software of wide interest to mathematicians, mathematics students, and perhaps those outside the eld. Many packages have been cre-

ated for solving particular counting and enumeration problems, such as Lara Pudwell's enumeration schemes packages , Donald Knuth's OBDD outside the scope of this project. There are some relatively general-purpose packages which did not make it into the project but should to be mentioned for completeness, though: The Mathematica package Omega implements MacMahon's Partition Analysis called LinDiophantus

which enumerates perfect matchings

of bipartite graphs, and many of Zeilberger's numerous Maple packages ; such packages are

10

11

, and there is a Maple analog written by Zeilberger

; an algorithm by Guo-Niu Han able to cover more general expres-

sions than Omega was implemented in Maple

12

and Mathematica

13

packages; the package

RLangGFun translates from rational generating functions to regular languages

14

, and the

regexpcount package from the INRIA Algorithms Group translates in the other direction method, an innite-matrix generalization of the transfer matrix method.

15

and Zeilberger has written a number of packages related to the umbral transfer matrix

16

This project excludes some software related to the intersection of algebraic and enumerative combinatorics, and all software related to power series summation and manipulation. General-purpose software is currently skewed towards manipulations, away from bijections. Indeed, one would not expect complicated bijections such as the proofs of Theorem 1 in [30] (balanced trees and rooted triangulations) or Theorem 4.34 in [8] (indecomposable 1342-avoiding

n-permutations

and

(0, 1)-trees

on

vertices) to be obtainable by symbolic

methods any time soon.

However, in this project, we include as many algorithms with a

bijective avor as possible. (Ultimately all computations may be considered manipulations, but we refer to a qualitative dierence in mathematical content.) Finally, we note that plenty of basic algorithms from combinatorics, as well as advanced algorithms from the symbolic computation literature, have not been implemented in a published package. (Actually, whether or not an algorithm has been implemented in a published package has a fuzzy value. For example, some have been implemented, published, but are now gone, while others have been published and are available, but are written in obscure languages that are unfamiliar or dicult to obtain compilers for. The sentence is true even in the loosest sense, however.)

7 8 9 10 11 12 13 14 15 16

http://faculty.valpo.edu/lpudwell/maple.html http://www-cs-staff.stanford.edu/~knuth/programs.html http://www.math.rutgers.edu/~zeilberg/programs.html http://www.risc.jku.at/research/combinat/software/Omega/index.php http://www.math.rutgers.edu/~zeilberg/tokhniot/LinDiophantus http://www-irma.u-strasbg.fr/~guoniu/software/omega.html http://www.risc.jku.at/research/combinat/software/GenOmega/index.php http://www.risc.jku.at/research/combinat/software/RLangGFun/index.php http://algo.inria.fr/libraries/libraries.html#regexpcount http://www.math.rutgers.edu/~zeilberg/programs.html

4
4.1

Basic combinatorial objects

Mathematical background

Algorithmically counting and enumerating the basic objects of combinatorics like graphs, trees, set partitions, integer partitions, integer compositions, subsets, and permutations is implemented in various packages and discussed in various books [23, 24, 26, 27, 37] (and many papers). As an example, [33] describes over 30 permutation generation algorithms published as of 1977. A complete comparison of the enumeration algorithms implemented in the packages in this project could be its own project. For the packages mentioned in this project that For more However, in the rest of this subsection we enumerate basic objects, we do not give full details on the algorithms used. information, see the packages' documentation.

provide some examples of two general concepts, rst mentioned in Section 2.2, which guide the discovery of enumeration algorithms: orderings and Gray codes.

4.1.1 Orderings
Many combinatorial objects can be ordered lexicographically.

Lexicographic order, a.k.a. lex

17
If

order, applies when the objects can be represented as words over an ordered alphabet.

w = w1 w2 wn and v = v1 v2 vn are words then, in lexicographic order, then w < v i w1 < v1 or there is some k [1..n 1] such that wj = vj for 1 j k and wk+1 < vk+1 .
Permutations are a clear example of a case where this order applies, and iterating through permutations in lexicographic order is an easy exercise, see [11] for a solution.

Co-lexicographic order,
Another order,

a.k.a.

co-lex order, is related: If

and

are words,

w w
reverses

in co-lexicographic order i words).

rev(w) rev(w )

in lexicographic order (where

rev

cool-lex order,

applies to binary words containing exactly

copies of

1,

which we can think of as

k -subsets

[31]. Generating the next binary word in cool-lex order

is done as follows: Find the shortest prex ending in

010

or

011,

or the entire word if no each succesive binary

such prex exists. Then cyclically shift it one position to the right. Since the shifted portion of the string consists of at most four contiguous runs of

0's

and

1's,

word can be generated by transposing only one or two pairs of bits. Thus cool-lex order for

k -subsets

is a Gray code.

4.1.2 Gray codes

Unrestricted subsets have a Gray code that is very easy to understand, called the

reected Gray code, in which, as above, we represent subsets as binary words.

to construct a Gray code code

standard

Say we want

Gn of subsets of a size-n set, and suppose we already have a Gray Gn1 of subsets of the last n 1 elements of the set. Concatenate Gn1 with a reversed copy of Gn1 with the rst element of the set added to each subset. Then all subsets dier
17

http://planetmath.org/encyclopedia/DictionaryOrder.html
9

by one from their neighbors, including the center, where the subsets are identical except for the rst element.

4.2

Combinat (Maple)

Author: Maplesoft Website:

httpXGGwwwFmplesoftFomGsupportGhelpGwpleGviewFspxcpthaomint

The combinat package, which is distributed with Maple, has routines for counting, listing, randomly generating, and ranking and unranking basic combinatorial objects such as permutations, compositions.

18

k -subsets,

unrestricted subsets, integer partitions, set partitions, and integer

Like Combinatorica and unlike Sage and the Combinatorial Object Server,

combinat does not oer a wide range of restrictions that can be placed on the objects. As mentioned in Section 5.2, most of the functionality of the combinat package is also covered by Combstruct. Most types of objects can only be enumerated in a single ordering, but unrestricted subsets (in binary word form) can be listed in Gray code order with the

gryode

function.

Example 4. We can use gryode to print all subsets of a size-3 set:

b printf@t@ ` 7FQd `6VAD op@mp@onvertD gryode@QAD inryAAA HHH HHI HII HIH IIH III IHI IHH
We note that outside the combinat package, Maple includes support for random graph generation, which is comparable to, for example, Mathematica's. For more information on Maple, see the Appendix.

4.3

Combinatorial Object Server

Author: Frank Ruskey Last modied: May 2011 Website:

httpXGGtheoryFsFuviFGosFhtml

The Combinatorial Object Server (COS) is a website that runs on the University of Victoria's domain. It has a web interface for easily specifying a set of basic combinatorial objects and viewing an exhaustive listing of all objects in the set (see Figure 1 on page 12). Objects available include permutations, derangements, involutions, subsets, set partitions, trees, necklaces, and unlabeled graphs. On each type of object, there is a set of restrictions that can be placed. Integer partitions, for example, can be restricted by largest part, and whether the parts must be odd, distinct, or odd and distinct.

k -subsets,

unrestricted

18

http://www.maplesoft.com/support/help/Maple/view.aspx?path=combinat

10

There is also a wide variety of output formats for the objects. Permutations, for example, can be printed in one line notation, cycle notation, permutation matrix form, standard Young tableau form and more. The order of output can sometimes be specied, too. Combinations, for example, can be shown in Gray code, lexicographic, co-lexicographic, cool-lex, transposition, or adjacent transposition orders.

4.4

Combinatorica: basic combinatorial objects

Authors: Sriram Pemmaraju and Steven Skiena Download: Website:

httpXGGwwwFsFuiowFeduG~srirmGgomintoriGxewgomintoriFm

Last modied: 2006

httpXGGwwwFsFsunysFeduG~skienGomintoriG

Combinatorica is a Mathematica package for discrete mathematics. In development since 1990, it includes over 450 functions in the areas of Plya theory, permutations and algebraic combinatorics, basic combinatorial objects, graph algorithms, and graph plotting. A book was written by the package authors [27], which is the denitive source of information on Combinatorica. The Combinatorica package has been included with releases of Mathematica since Mathematica version 4.2, although some of Combinatorica's functionality has recently been redone and built into the Mathematica kernel in Mathematica 8. For information on Mathematica as a programming language, see the Appendix. Combinatorica has support for counting and enumeration with permutations, and graphs.

k -subsets,

unrestricted subsets, integer partitions, integer compositions, set partitions, Young tableaus For each type of object, Combinatorica generally oers rules for counting, iteration and listing in one or two orderings, and random generation. Combinatorica does not provide as many ways to specify restrictions on the objects as COS or Sage.

Example 5. The function qrygodeusets exhaustively lists all subsets of a set in standard reected Gray code order:
snIXa qrygodeusets{ID PD QD R} yutIXa {{}D{R}D{QDR}D{Q}D{PDQ}D{PDQDR}D{PDR}D{P}D{IDP}D{IDPDR}D {IDPDQDR}D{IDPDQ}D{IDQ}D{IDQDR}D{IDR}D{I}}

One may wonder if such a Gray code is unique, and one can nd this out by rst noticing that Gray codes for the subsets of a size-n set are in bijection with Hamiltonian paths in the n-dimensional hypercube. Combinatorica includes a database of common graphs, including ryperuen, and also has the rmiltoningyle rule which replaces rmiltoningylegrphD ell with a list of all Hamiltonian cycles in grph. So to nd out if the Gray code order above is unique, one can nd the length of the list of Hamiltonian paths in the 4-dimensional hypercube:
11

Figure 1: The Combinatorial Object Server's page for set partitions.

12

snPXa vengthrmiltoningyleryperueRD ell yutPXa PTVV

It is denitely not! The number of Hamiltonian cycles in an n-dimensional hypercube is not known, even asymptotically [37].
4.5 Sage: basic combinatorial objects

httpXGGwwwFsgemthForgGdownlodFhtml Online access: httpXGGwwwFsgenForgG Website: httpXGGsgemthForgGdoGrefereneGomintGindexFhtml

Download: Sage is a free, open-source computer algebra system (CAS) rst released in 2005. Sage integrates many specialized open-source symbolic and numeric packages, such as Maxima, GAP, SciPy, and NumPy, written in various languages and allows them all to be called from a unied Python interface. In addition, it has native support for a wide and quickly expanding range of mathematical elds, including combinatorics. This section covers Sage's capabilities for counting and enumerating basic combinatorial objects; see Section 7.3 for Sage's combinatorial species capabilities. Sage uses object-oriented programming to implement a category-theoretic hierarchy of categories and objects.

19 20

Sage's support for combinatorial objects, which is part of a

migration of the MuPAD-Combinat project based on the category called (such as

21

inumertedets22 .

, which has reached end-of-life, to Sage, is Classes of basic combinatorial structures

k -subsets, unrestricted subsets, signed and unsigned integer compositions, necklaces,

integer partitions, permutations, ordered and unordered set partitions, words and subwords) all belong to the category

inumertedets

which implies that sets of objects from those

categories can be constructed which inherit at least the following methods: 1. 2. 3. 4. 5. 6.

rdinlity@A list@A

- the cardinality of the set,

- a list of all elements, - the

unrnk@nA rnk@eA first@A next@eA

nth

object in an ordering,

- the rank of the object

e,

- the rst object in the ordering, - the next object after

in an ordering,

19 20 21 22

http://www.sagemath.org/doc/reference/sage/categories/category.html http://www.sagemath.org/doc/reference/sage/categories/primer.html http://mupad-combinat.sourceforge.net/ http://www.sagemath.org/doc/reference/sage/categories/enumerated_sets.html

13

7.

rndomelement@A - an object chosen at random according to the uniform distribution.

Of course, each class of combinatorial object built in to the system may also implement many more methods. Many classes allow restrictions to be specied, but only the default ordering is available.

Example 6. The rtitions@A static method is called to construct an object representing a set of integer partitions specied by its arguments.23 For example, rtitions@RA returns all integer partitions of 4, while rtitions@RD mxprtaPA returns all partitions of 4 with maximum part size 2:
rtitions@RD mxprtaPAFrdinlity@A Q sgeX rtitions@RD mxprtaPAFlist@A PD PD PD ID ID ID ID ID I sgeX rtitions@RD mxprtaPAFrndomelement@A PDP
Sage also provides several implementations of counting functions, separate from the

inumertedets

category.

24

These include the partition-theoretic counting functions for

number of set partitions, and ordered and unordered integer partitions, and the set-theoretic counting functions for number of subsets, arrangements, derangements and permutations of a multiset.

5
5.1
Let

Symbolic combinatorics
Mathematical background
S

<2> be a set of objects, with a parameter p = (p). We dene a new set S = S S, <2> <2> <2> with parameter p = (p ), where p ((s1 , s2 )) = p(s1 ) + p(s2 ) for all s1 , s2 S . Let 1 <2> f (n) = |p (n)|, f (n) = |(p<2> )1 (n)|. Then if F (z) is the ordinary generating function

F (z) =
n0
and

f (n)z n ,

F <2> (z)

is the ordinary generating function

F <2> (z) =
n0
we have, simply,

f <2> (n)z n ,

F <2> (z) = F (z)2 .

23 24

http://sagemath.org/doc/reference/sage/combinat/partition.html http://sagemath.org/doc/reference/sage/combinat/combinat.html
14

It turns out that many other correspondences exist between the structure of a set of objects and its generating function. This project includes sections for two frameworks that develop this idea:

the theory of combinatorial species combinatorics, which is described below.

which is the focus of Section 7, and

symbolic is

The central concept of symbolic combinatorics

25

is the combinatorial class. size

Denition 2. A combinatorial class is a countable set on which a parameter called dened, such that the number of elements of any given size n 0 is nite.
If

is a combinatorial class, the size of an element

is denoted

set of elements of size

in

by

An ,

and denote its cardinality by

||. We an = |An |.

denote the

Denition 3. The

counting sequence

of a combinatorial class A is the sequence (an )n0 . unlabeled

and

There are two types of combinatorial class,

labeled.

5.1.1 Unlabeled classes The word class in this section Denition 4. The (ordinary) formal power series

refers to an unlabeled combinatorial class, which can be

thought of as a set of objects made up of nodes without unique labels (think graphs). This will become rigorous as we proceed. generating function

of a class A with z marking size is the

A(z) =
n0

an z n .

Denition 5. A (k-ary) combinatorial construction is a function that maps combinatorial classes B<1> , B<2> , . . . , B<k> to a new class A = (B<1> , B<2> , . . . , B<k> ). The combinatorial construction is admissible i the counting sequence of A only depends on the counting sequences of the arguments B<1> , B<2> , . . . , B<k> .
If a construction functions such that

is admissible, there exists a corresponding <1> if A = (B , B <2> , . . . , B <k> ), then

operator

on generating

A(z) = (B <1> (z), B <2> (z), . . . , B <k> (z)).

The basic admissible constructions for unlabeled classes are called sum, product, sequence, powerset, multiset, and cycle. The denitions and corresponding generating function operators for all of these can be found in [13]; here we only describe the rst three. 1. The of

sum
26

of two classes

and

is written

A+B

and is formed by the discriminated

union

A and B , with A + B is A(z) + B(z).

of

size inherited from the summands. The generating function

25 26

http://en.wikipedia.org/wiki/Symbolic_combinatorics http://en.wikipedia.org/wiki/Disjoint_union
15

2. The

product
B,

of two classes is written

AB AB

and is formed by the cartesian product of is

and

with size dened additively. This is the construction used above, where we

saw that the generating function for 3. Finally, the

A(z)B(z).

0.

Seq(A) is the set of all nite sequences of elements in A, with size dened additively. The generating function for Seq(A) is
For a class

sequence
A,

construction

Seq

is dened on classes with no elements of size

the value

1 + A(z) + A(z)2 + =

1 . 1 A(z)

The basic combinatorial constructions can be modied ( For example, we can x length at least Let

and dene a construction

restricted ) in a number of ways. Seqk that constructs sequences of

called the

k,

or for another example we could construct products containing an element of size

of even size from one class and an element of odd size from another, etc.

atom. Denition 6. A specication for an r-tuple of classes is a collection of r equations

be the class with a single element

be the class with a single element

0,

neutral object,

and let

of size

1,

called an

A<1> = 1 (A<1> , . . . , A<r> ) A<2> = 2 (A<1> , . . . , A<r> ) <r> A = r (A<1> , . . . , A<r> ),

where each i ( ) represents an expression built from the A's using the (possibly restricted) basic admissible constructions, as well as the classes E and Z . Example 7. Let T be the class of nonempty unlabeled plane trees, with the size of a tree being the number of nodes. Then T satises the specication
T = Z Seq(T ),

since an object in T is a single root with a sequence of subtrees. This specication implies T (z) = z(1 T (z))1 , and thus T (z) = 1 1 1 4z . 2

5.1.2 Labeled classes

An object in a labeled class is

labeled,
of size

meaning, if it has size

n,

each of its

indivisible

components is labeled with a unique integer from the set class with one neutral object element

[1..n].

A rigorous way to dene

such classes begins with a dierent denition for the elementary classes: Again, let

be the

of size

1,

labeled atom.

0,

but now let

be the class containing one labeled

Then, labeled classes can be dened by specications as

above if we dene some useful constructions on labeled classes. There is indeed a set of admissible constructions for labeled classes that is analogous to that for the unlabeled case. The sum of two classes is dened the same as for unlabeled classes, but for the others, we rst need to dene the product of two labeled objects:

16

Denition 7. Given two labeled objects, and , the labeled product is the set of all pairs ( , ) where and are relabeled versions of and such that order is perserved in the relabelings and each number in [1..|| + ||] appears as a label in either or .
This concept leads to the denition of the product, sequence, set and cycle constructions, the rst of which we dene here; the rest can be found in [13]. The product of labeled classes

and

is the set

A B=
A,B
with size dened additively.

( ),

As with the unlabeled case, the usefulness of dening a labeled class in terms of a specication comes from the fact that labeled constructions correspond to relatively simple operators on generating functions  exponenential generating functions in the labeled case.

Denition 8. The (exponential) generating fuction of a labeled class A with z marking size is the formal power series n
A(z) =
n0
For example, the generating function for

an
is

z . n!

A B

A(z)B(z).

Example 8. We dene a labeled binary tree as a labeled tree in which every internal node has two children. Let B be the labeled class representing such trees. Since an object in B is either a node with no children or a node with two children, we have
B =Z +Z B B,

which implies that B(z) = z + zB(z)2 , and thus B(z) = (1 1 4z 2 )/2z .

5.1.3 Multiple parameters

Unlabeled and labeled classes can be augmented with parameters other than size. example, if there is one more parameter that For

E = { },

-atomic

, we can redene the original elementary classes so | | = ( ) = 0, and Z = {}, where || = 1, () = 0, and dene a new class P = {}, where || = 0, () = 1. Size is the only parameter with respect to
where

which a structure is labeled or unlabeled (and the only parameter which may be marked by a variable with a factorial below it in the generating function for the class), so those words can still be used unambiguously to refer to a class with more than one parameter. All constructions for labeled and unlabeled classes discussed so far have dened size additively, e.g. the size of an object is the sum of the sizes of its components. All of these constructions can be dened on multi-parameter classes with non-size parameters dened additively, just like size. The generating function equations they correspond to are the same as the single-parameter ones, except the generating functions may be multivariate (with a dierent variable marking each parameter).

17

Example 9. Let A be a single-parameter unlabeled combinatorial class. One can dene a new class, B, consisting of sequences of elements of A, with size and an additional parameter such that () is the number of elements of A in , for all B . Then B = Seq(P A), and, if u marks , A(z, u) = (1 uB(z))1 .
For more details, and for information on constructions where additional parameters are not dened additively, see [13]. Given a labeled or unlabeled class random variable distribution. If

with additional parameter

one can dene a then we have the

Xn to be an u marks in

object chosen at random from

An

according to the uniform

the bivariate generating function

A(z, u),

syntactically simple relation

[z n ]A(0,1) (z, u)|u=1 . E[(Xn )] = [z n ]A(z, 1)

Higher factorial moments are obtained similarly. Techniques for obtaining limiting distributions from multivariate generating functions also exist: see [13]. However, note that these methods for obtaining probabilistic facts apply for any multivariate generating function, whether or not it was obtained with symbolic combinatorics.

5.2

Combstruct
27
and Maplesoft

Authors: INRIA Algorithms Group Website:

httpXGGwwwFmplesoftFomGsupportGhelpGwpleGviewFspxcpthaomstrut
Combstruct is a Maple package originally developed by the INRIA Algorithms Group which is now distributed with the most recently released version of Maple, Maple 16. Its functionality has changed over time, but today, it includes the capabilities of ALAS from LUO (Section 5.4) along with the ability to enumerate, both randomly and exhaustively, the objects of a given size from a specied combinatorial class. Combstruct also extends ALAS by supporting translation from multiple-parameter specications to mulivariate generating functions. Combstruct can also count, randomly and exhaustively enumerate, and iterate through a small set of built-in, predened structures. of the functionality of Maple's combinat In fact, Combstruct provides most

28

package for working with basic combinatorial

structures, but with syntax that is unied with that for working with classes of objects dened by the user by combinatorial specications. We elaborate on these areas of functionality in the rest of this section. Combstruct allows the user to create labeled or unlabeled single-parameter combinatorial specications, and augment the specications with additional parameters separately. To

27 28

http://algo.inria.fr/index.html http://www.maplesoft.com/support/help/Maple/view.aspx?path=combinat
18

specify a class, the elementary classes

and

can be used, as well as the constructions sum,

product, set, powerset, sequence, cycle and substitution. The constructions set, powerset, sequence and cycle can be restricted by an inequality or equality relation on the number of components allowed in the objects. It is possible, of course, to dene a great many types of basic combinatorial object, as well as more complicated objects, with such specications. The

gfeqns

command returns the system of equations over generating functions corre-

sponding to a well-dened single-parameter specication (see Denition 10), and attempts to return explicit expressions for the generating functions. The returns the initial values of the counting sequences of the classes.

gfseries command

gfsolve

Example 10. We create a specication intreespe for labeled binary trees, where f is the class of labeled binary trees and is the atomic class:
bintreespe Xa {fanion@D rod@D fD fAAD aetom }X

(The nion construction is the same as sum.) Then we use gfsolve to get the generation functions:
bgfsolve@intreespeD leledD zA f@zA a E@IGPAB@EICsqrt@IERBzPAAGzD @zA a z
Combstruct's

A,

and an

drw command takes a single-parameter specication, the name of a class integer n 0 and returns a object chosen at random from the set An according

to the uniform distribution. For information on the algorithms used, see the documentation. To add additional parameters to a single-parameter specication, Combstruct allows the user to use an attribute grammar (see the documentation and [25] for more information on attribute grammars; also, note that it is also possible to augment single-parameter specications in Combstruct using a more limited method based on dening and using new atomic classes as described in the Mathematical background). The cation and attribute grammar. The multidimensional sequences. The

gfeqns

command returns the

system of equations over multivariate generating functions corresponding to a given speci-

gfseries command returns the initial values of the gfmomentsolve command takes an integer k 0 and a <i> set of equations over n multivariate generating functions A (z, u), 1 i n and attempts k <i> to return explicit expressions for Du A (z, u)|u=1 , 1 i n.
A number of in-depth examples of Combstruct in action are available at the INRIA Algorithms Group's website

29

5.3

Encyclopedia of Combinatorial Structures

Authors: Frdric Chyzak, Alexis Darrasse, and Stphanie Petit Download:

httpXGGlgoFinriFfrGlirriesG5down

29

http://algo.inria.fr/libraries/autocomb/

19

Last modied: July 2000  original version

30

, 2011  online version

31

The Encyclopedia of Combinatorial Structures started out as a Maple package written by Stphanie Petit as part of the INRIA Algorithm Group's algolib and in 2009 a web interface for it was created by Alexis Darrasse and Frdric Chyzak at

frGenylopediGintroFhtml.

httpXGGlgoFinriF

The Encyclopedia is a database of counting sequences of speciable combinatorial structures. For each sequence in the database, the following elds, if available, are either computed or stored: 1. Name 2. Combinatorial specication, in Combstruct syntax 3. Initial values, obtained with Combstruct's

ount

4. Generating function, obtained with Combstruct's gfsolve 5. A linear reccurrence relation, if applicable, obtained with gfun's

diffeqtore
6. A closed-form expression, obtained with either

holexprtodiffeq and rsolve

or gfun's

Maple's

rtpolytooeff
7. Dominant asymptotic term as

computed by gdev's

equivlent

8. Description of combinatorial structure 9. References, such as entry in Sloane's Encyclopedia of Integer Sequences

32

(Note that gfun is the name of a Maple package developed by the INRIA Algorithms Group, for more information, see

33

.) It is possible to search the database by initial values of the

sequence, keywords, generating function, or closed form of the sequence.

5.4

Lambda-Upsilon-Omega (LUO): symbolic combinatorics

Authors: Bruno Salvy and Paul Zimmermann Download: Website:

httpXGGwwwFloriFfrG~zimmermGsoftwreGluoPFIFtrFgz
34

Last modied: May 1995

httpXGGlgoFinriFfrGlirriesGlirriesFhtml5luo

30 31 32 33 34

http://algo.inria.fr/salvy/index.html http://algo.inria.fr/encyclopedia/intro.html http://www.research.att.com/~njas/sequences/ http://algo.inria.fr/libraries/#gfun http://algo.inria.fr/salvy/index.html

20

LUO is a software project started in the late 1980s designed to automatically analyze algorithms. It is no longer heavily used; most of its functionality is available in Combstruct and the

equivlent

command in gdev.

35

In this project, we focus on the subset of its

capabilities related to enumerative and analytic combinatorics, omitting a discussion of its capabilities for algorithm analyis. LUO is made up of two modules: the algebraic analyzer (ALAS) and the analytic analyzer (ANANAS). ALAS takes as input a combinatorial specication, either labeled or unlabeled, and outputs a system of equations over generating functions for the classes in the specication. Then an intermediate process attempts to solve the equations explicitly; if successful, it passes the solutions to ANANAS. ANANAS then employs a routine (which became

equivlent

in gdev) on the expressions which returns asymptotic expressions for

the coecients. ALAS is described in further detail below, and ANANAS is described in Section 8.3 Ocial documentation for LUO comes in the form of a main article discussing the algorithms used, but not the code [14], a cookbook containing a summary of [14] and a selection of examples of algorithms being analyzed [15], and some notes on the code and usage of ALAS, including a reference for writing programs in the syntax that the system can analyze, and examples of ALAS in action [39]. As mentioned above, ALAS, written by Paul Zimmermann, is a tool for translating specications to systems of equations over generating functions. That is, the user supplies a list of

equations

A<1> = 1 (A<1> , . . . , A<r> ) A<2> = 2 (A<1> , . . . , A<r> ) <r> A = r (A<1> , . . . , A<r> ),

where each

i ( )

represents an expression built from the

A's

using the basic admissible

constructions, as well as the classes

and

Z.

Some restricted constructions are also allowed.

Denition 9. The valuation of a class A<i> is the minimum size of an object in A<i> built according to the specication. Denition 10. A combinatorial specication is well dened i it satises the two properties 1. each class has nite valuation, and
2. for each class A<i> and n 0, the number of objects of size n in A<i> built according to the specication is nite.
Since there is nothing preventing the user from supplying a non-well dened specication, it would be nice if the well-denedness of a specication could be checked programmaticly and indeed it can and ALAS does this. See [14] for details.

35

http://algo.inria.fr/salvy/index.html
21

Once ALAS veries that the specication is well-dened, it proceeds to generate the corresponding equations over generating functions using simple replacement rules. values of the counting sequences. The following theorem is proved in [14]: These generating function equations can then be used to (among other things) compute initial

Theorem 1. The number of arithmetic operations necessary for computing all the counting sequences associated with all classes A<i> up to size n is O(rn2 ).

6
6.1

Plya theory
Mathematical background

Plya theory is the study of counting symmetric objects, which makes use of group theory and generating functions. It was originally developed by John Redeld, then refounded by George Plya, who used it to count chemical compounds, among other objects, and after whom the central theorem (Theorem 2 on page 24) is named.

Denition 11. An action of a group H on a set X is a homomorphism : H S(X) from H to the symmetric group on X.
In this section, the groups we work with are assumed to be the images of group actions on nite sets, i.e. permutation groups. Let relation

G be such a group on a set X . g G : g(x) = y.

of is written

Dene an equivalence

on

by

xy
The set of equivalence classes of Elements of element

i

X/G
is

are called the

orbits
x

under

under

X/G. G, and

the

orbit

of an element

is the equivalence class represented by

and is denoted by

G(x).

The

stabilizer Gx

xX
of an

xX

Gx = {g G : g(x) = x}.
Our rst lemmas bring counting into the picture:

Lemma 1 (Orbit-stabilizer). If G is a permutation group on X then for all x X ,

|G(x)| = |G|/|Gx |.

Proof.

f : G X dened g g(x). Then there is a bijection between the image of f , which is G(x), and the set of left cosets of Gx , which has cardinality |G|/|Gx |, given by h(x) hGx for all h(x) G(x).
For any consider the mapping

x X,

Lemma 2 (Burnside). The number of orbits of a permutation group G on a set X is

|X/G| = 1 x(g), |G| gG

where x(g) is the number of xed points of g.

22

Proof.
x(g) =
gG gG xX

[g(x) = x] [g(x) = x]
xX gG

= =
xX

|Gx | |G| |G(x)| 1 |G(x)| 1 |A| xA |A|

AX/G

=
xX

= |G|
xX

= |G|
AX/G

= |G| = |G||X/G|

1 |A|

Denition 12. A graph automorphism of a graph G = (V, E) is a bijection : V V such that {((v1 ), (v2 ))|(v1 , v2 ) E} = E . The set of all automorphisms of G forms a group, aut(G). Example 11. Say we are given a labeled graph G, and we would like to nd the number of relabelings (bijections on the vertex set) that yield distinct graphs. Let X be set of all relabelings, and let be the induced action of aut(G) on X , i.e. if X and g aut(G), then (g)() = (g(1 ), g(2 ), . . . ). Then the number of such relabelings, which is the number of isomorphic graphs on the same vertex set, is |X/(aut(G))|. (The packages nauty36 , saucy37 and bliss38 , as well as some of the packages in this section, can compute the automorphism group of a graph. The code that nauty uses is also included in GRAPE.39 )
We now begin working up to the central theorem of Plya theory: a generalization of Lemma 2 for counting weighted colorings of an object.

Denition 13. Let g G be a permutation, and let c(g) be the signature of g. Then the cycle index of g is the formal monomial
z(g; s) = sc(g) .
36 37 38 39

http://cs.anu.edu.au/~bdm/nauty/ http://vlsicad.eecs.umich.edu/BK/SAUCY/ http://www.tcs.hut.fi/Software/bliss/ http://www.gap-system.org/Packages/grape.html

23

The cycle index of the whole permutation group G is the terminating formal power series
Z(G; s) =
Let

1 z(g; s). |G| gG

= {1 , 2 , . . . } be a countable set of colors, each of which has a non-negative 1 weight w(i ), such that w (n) is nite for all n 0. We dene the color-counting generating
function

a(t) =
n0
where

an tn ,

total weight

an = |w1 (n)|. X Let F = be the

set of functions from

of a coloring

f F

is

X ; we call these functions colorings. w(f (x)). Dene an action of G on F by xX

The

g (f (f g 1 )),
and let the coloring-counting generating function be

b(t) =
n0
where

bn tn , F
with total weight

bn

is the number of orbits of

(G)

on

n.

Lemma 3. Say |X| = m. Then the generating function for functions from X to xed by a permutation (g) with total weight marked by t is
z(g; a(t), a(t2 ), . . . , a(tm )). (g) i it is constant on the cycles of g . So a function xed by (g) is specied by, for each i, a mapping between each cycle of size i and . For each i c (g) cycle size i, these mappings have generating function a(t ) i since there are ci (g) cycles of size i and each color of weight k contributes a function of weight ki that is constant on the i-cycle. The overall generating function is thus
A function is xed by

Proof.

a(t)c1 (g) a(t2 )c2 (g) a(tm ) = z(g; a(t), a(t2 ), . . . , a(tm )).

Theorem 2
Proof.

(Plya enumeration theorem, single variable version)

. Say |X| = m, then

b(t) = Z(G; a(t), a(t2 ), . . . , a(tm )).

Immediate from Lemmas 2 and 3.

Example 12. A necklace of size n is a cycle of colored beads that can be ipped over or rotated and still be considered the same object. In this example, we count black and white necklaces of size 4 with a given number of black beads. We can use as our set X = {1, 2, 3, 4},
24

Figure 2: The six necklaces with four beads and two colors.

Figure 3: The four unlabeled graphs with

vertices.

and use as our permutation group on X the dihedral group D4 with 8 elements. We can let weight be the number of black beads, so that a(t) = 1 + t. Since D4 has cycle index
Z(D4 ; s1 , s2 , s3 , s4 ) = s4 + 2s2 s2 + 3s2 + 2s4 1 1 2 , 8

we have

b(t) = Z(D4 ; 1 + t, 1 + t2 , 1 + t3 , 1 + t4 ) = 1 + t + 2t2 + t3 + t4 .

From this generating function, we can also read o that there are 6 necklaces of size 4 with (at most) two colors of bead, as shown in Figure 2, taken from MathWorld40 . Example 13. Suppose we would like to know the number of unlabeled graphs with 3 vertices and k edges, 0 k 3, up to isomorphism. Plya's enumeration theorem can be used if we represent unlabeled graphs with 3 vertices as 2-colorings of the set E of all 3 possible edges, 2 where the color black represents an edge and the color white represents no edge. Total weight of a coloring equals the number of black edges, so our color-counting generating function is a(t) = 1 + t. The group of permutations on E that we need to quotient out is isomorphic to S3 , the symmetric group on the three vertices. The group S3 has cycle index
Z(S3 ; s1 , s2 , s3 ) = s3 + 3s1 s2 + 2s3 1 , 3!

thus

b(t) = Z(S3 ; 1 + t, 1 + t2 , 1 + t3 ) = 1 + t + t2 + t3 ,

and we see that there is exactly one graph, up to isomorphism, with 3 vertices and k edges, 0 k 3. Each of these is shown in Figure 3.
For more information on Plya theory, a more general version of Theorem 2, and many more examples, see [6, 29]

41

40 41

http://mathworld.wolfram.com/Necklace.html http://en.wikipedia.org/wiki/P%C3%B3lya_enumeration_theorem
25

6.2

A note on Plya theory packages

Capable software for the latter includes

Our goal in this section is to describe packages designed for Plya theory-type counting, rather than general computational group theory. GAP, Magma, Maple, Mathematica, and Sage. In the words of the authors of Combinatorica, Our aim in introducting permutation groups into Combinatorica is primarily for solving combinatorial enumeration problems. We make no attempt to eciently represent permutation groups or to solve many of the standard computational problems in group theory. [27]

6.3

COCO

Download:

httpXGGwwwFwinFtueFnlG~eGftpdosGmthGooGooEIFPFtrFgz

COCO is a package for doing computations with permutation groups which was designed to investigate graphs [22].

coherent congurations,

which are a certain type of edgecolored complete

COCO includes routines for, among other things, nding the automorphism

group of an edgecolored complete graph and, given a base set and a permutation group on that set, computing the induced permutation group on a set of combinatorial structures over the base set. For more information, see its documentation and [22].

6.4

Combinatorica: Plya theory

Authors: Sriram Pemmaraju and Steven Skiena Download: Website:

httpXGGwwwFsFuiowFeduG~srirmGgomintoriGxewgomintoriFm

Last modied: 2006

httpXGGwwwFsFsunysFeduG~skienGomintoriG

Combinatorica has Plya-theoretic functionality that integrates with the rest of its capabilities, such as basic combinatorial objects, a description of which, along with an overview of the Combinatorica package can be found in Section 4.4. Combinatorica has built-in rules representing the symmetric, cyclic, dihedral and alternating groups, as well as the ability to create groups from other groups or simply from a set of permutations. Combinatorica can compute the automorphism group of a graph. The rules

yrits

and

yritepresenttives

take a set and a permutation group, and

optionally how the group acts on the set (the default action being the identity map), and return the set of orbits and representatives of those orbits, respectively.

Example 14. The yritepresenttives rule can be used to list all distinct necklaces of size 4 with 2 colors. To do so, we evaluate the rule with the group hihedrlqroupR, which is D4 , and the set of words of length 4 over the letters R and B :
snIXa yritepresenttiveshihedrlqroupRD trings{D f}D R
26

yutIXa {{DDD}D{fDfDfDf}D{fDfDfD}D{fDfDD}D{fDDfD}D{fDDD}}
The rule

gyletruture gives the cycle index of a single permutation, and gylesndex

gives the cycle index of a permutation group. Combinatorica comes with special rules for cycle index of symmetric, alternating, cyclic

gylesndex. For Plya's enumeration theorem, we have the rule yritsnventory, which takes the cycle index G(Z; s) of a group G and a list (w1 , w2 , . . . , wm ) of expressions and returns 2 G(Z; m wi , m wi , . . . ). i=1 i=1
and dihedral groups which work much faster than

Example 15. Letting G = D4 and setting w1 = 1, w2 = t, we can obtain the generating function for 2-colored necklaces of size 4:
snPXa dihedrlqroupgylesndex a hihedrlqroupsndexRD xY olorinumertor a {ID t}Y snQXa yritsnventorydihedrlqroupgylesndexD xD olorinumertor yutQXa I C t C P tP C tQ C tR

It turns out that Combinatorica has a built-in rule for this type of result:
snRXa xekleolynomilRD olorinumertorD hihedrl yutRXa I C t C P tP C tQ C tR

Example 16. As our last example of Combinatorica in action, we count the number of unlabeled graphs with 3 vertices. To apply Plya's theorem, we dene the set of all 2-subsets of [1..3], representing all 3 edges: 2
snSXa set a uusetsngeQD P yutSXa {{IDP}D{IDQ}D{PDQ}}

The group of permutations to quotient out is the set containing each permutation of set obtainable by applying a permutation g S3 to each 2-set in set. Combinatorica has the rule uusetqroup to create such a group:
snT Xa group a uusetqroupymmetriqroupQD setY

We stated above that this group is isomorphic to S3 . We can prove this with the Combinatorica rule wultiplitionle, which takes a set and an operation and gives the group multiplication table of the group they form:
snU Xa wultiplitionleymmetriqroupQD ermute GG leporm

27

yutUGGleforma

I P Q R S T

P I R Q T S

Q S I T P R

R T P S I Q R T P S I Q

S Q T I R P S Q T I R P

T R S P Q I T R S P Q I

snV Xa wultiplitionlegroupD yutVGGleforma I P Q P I S Q R I R Q T S T P T S R

ermute GG leporm

In order to apply Plya's enumeration theorem, we use gylesndex to compute the cycle index of group:
snW Xa ylesndex a gylesndexgroupD x yutW Xa xIQGT C xI xPGP C xQGQ

Now yritsnventory can reproduce our result:

snIH Xa yritsnventoryylesndexDxD{IDt} yutIH Xa I C t C tP C tQ

We did not actually have to do all this to nd the total number, since Combinatorica includes the rules xumeryfqrphs and vistqrphs to count and list all nonisomorphic graphs with a given number of vertices.
snII Xa xumeryfqrphsQ yutII Xa R
6.5 GraphEnumeration

Author: Doron Zeilberger Download: Website:

httpXGGwwwFmthFrutgersFeduG~zeilergGtokhniotGqrphinumertion

Last Modied: July 20, 2010

httpXGGwwwFmthFrutgersFeduG~zeilergGmmrimGmmrimhtmlGqiFhtml

28

This Maple package implements the Plya-theoretic methods of [19] to solve various counting problems of the following type: How many nonisomorphic graphs are there with vertices,

edges, and some property

P? k -hyper-graphs
according to edges,

The types of graphs that GraphEnumeration counts include unlabeled connected graphs according to the number of edges, unlabeled regular unlabeled rooted trees, unlabeled trees, all (connected or general) unlabeled simple graphs with a given degree sequence, and all (connected or general) unlabeled multi-graphs with a given degree sequence.

6.6

PermGroup

Author: Thomas Bayer Last modied: June 21, 2004 Website:

httpXGGwwwFrisFjkuFtGreserhGomintGsoftwreGermqroupG

PermGroup is a Mathematica package for permutation groups, group actions and counting. Documentation is scant, but like Combinatorica, PermGroup has built-in rules for the most commonly used permutation groups, methods for creating custom groups, and ways to use them to solve counting problems involving symmetry. Schreier-Sims algorithm PermGroup implements the

42

for computing group orders. The examples below illustrate some

of PermGroup's functionality, for more details consult the documentation and code.

Example 17. To obtain the generating function for 2-colored necklaces of length 4 with t marking the number of black beads, we can use the hihedrlqroupgsoly rule which returns the cycle index for the dihedral group.
snI Xa hihedrlqroupgsolyRD x yutI Xa IGR @xIP xP C xPPA C IGV @xIR C xPP C P xRA

Now we can replace xn with 1 + tn :

snP Xa 7 GF {xn Eb I C tn} yutP Xa IGR @@I C tAP @I C tPA C @I C tPAPA C IGV @@I C tAR C @I C tPAP C P @I C tRAA

Finally, we expand the polynomial and get something familar:

snQ Xa ixpndd7 yutQ Xa I C t C P tP C tQ C tR

Example 18. To obtain the generating function for unlabeled graphs with 3 vertices, we rst dene the set of all 2-subsets of [1..3], representing all 3 edges: 2
42

http://en.wikipedia.org/wiki/Schreier-Sims_algorithm

29

snR Xa grphsQ a ermqroup `iretQ yutR Xa {{IDP}D{IDQ}D{PDQ}}

We let sQ represent the symmetric group S3 using the ymmqroup rule:

snS Xa sQ a ermqroup `qenerteermqroup `ymmqroupQY

Again, the group we really need is the induced group of sQ on grphsQ. PermGroup can give this to us with the rnsformqroup rule:
snT Xa gQ a rnsformqroupsQD grphsQD iretetionY

Now we could compute the cycle index polynomial of gQ and substitute in 1 + tn as above, but PermGroup oers the olyinumertion rule to save some work. The expression olyinumertiongroupD xD list evaluates to the cycle index of group in xID xPD FFF with xi replaced with otllisti.
snU Xa ixpndolyinumertiongQD xD {ID t} yutU Xa I C t C tP C tQ

7
7.1

Combinatorial species
Mathematical background
43
is a theoretical framework

The theory of combinatorial species, the most fruitful unifying concept in enumerative combinatorics of this quarter-century according to Zeilberger, generating functions can be described. While the theory of species and symbolic combinatorics (the subject of Section 5) have a lot in common, one of the important dierences is that in the theory of species, labeled and unlabeled objects are explicitly connected through Plya theory. Martin Rubey, co-author of the Aldor-Combinat package described in the next section, said in 2008: Combstruct and MuPAD-Combinat really had usability as primary goal. As one consequence, they do not implement species, but rather combinatorial classes, that is, collections of objects with a size function. The main drawback of that method is that you cannot treat labelled and unlabelled objects uniformly, there is no such concept as an isomorphism type, which, in my opinion, is the main strength of (ordinary) species. in the language of which a great diversity of families of combinatorial objects and their

44

43 44

http://www.math.rutgers.edu/~zeilberg/khaver.html http://www.mail-archive.com/[email protected]/msg00503.html

30

Connections between symbolic combinatorics and Plya theory have been made [13]

45

, but

they are not as central to the theory as they are in species. The article [28] is self-contained and can be used as a dictionary between the theory of species and the symbolic method of Flajolet and Sedgewick. We begin with the denition of a species:

Denition 14. A species (of structures) F is a pair of functions,

1. the rst of which maps each nite set U to a nite set F [U ], and 2. the second of which maps each bijection : U V to a function
F [] : F [U ] F [V ].

The functions F [] must satisfy the following properties: 1. for each pair of bijections : U V and : V W ,
F [ ] = F [ ] F [],

2. and if IdU : U U is the identity function,

F [IdU ] = IdF [U ] .
We note that if of this action by

is a group of permutations on

U,

then the second function of on

is a

group homomorphism from

to

S(F [U ]), i.e. an action of H

F [U ].

We denote the image

F [U ; H].

Denition 15. An element s F [U ] is called an F -structure on U , or alternatively a structure of species F on U . The function F [] is called the transport of F -structures along .
s F [U ], we |F [U ]| = |F [V ]|.
If use the notation

to denote

F [](s). E

Note that

|U | = |V |

implies

Whereas in symbolic combinatorics we had the classes the most elementary species are the

empty set species E , where

{}
if

and

Z , in combinatorial species,

E[U ] =
and the

|U | = 0, otherwise,

singleton species S , where

S[U ] = {u} U = {u}, otherwise.
if

45

http://www.mathematik.uni-stuttgart.de/~riedelmo/papers/collier.pdf
31

Denition 16. Consider two F -structures s1 F [U ] and s2 F [V ]. A bijection : U V is called an isomorphism of s1 to s2 i s2 = s1 . If there is such a bijection, we write s1 s2 and we say these two structures have the same isomorphism type. The isomorphism types of a species are isomorphism classes and thus equivalence classes. An F -structure s F [U ] on a set U can be referred to as a labeled structure, whereas an isomorphism type of F -structures can be referred to unlabeled structure.
Attentive readers will be detecting a whi of Plya theory by now, which will become stronger as we dene the generating functions associated with a species.

Denition 17. The series

exponential generating function

of a species F is the formal power

F (z) =
n0

f (n)

zn , n!

where f (n) = |F [[1..n]]|. Denition 18. Denote by T (Fn ) the quotient set F [[1..n]]/F [[1..n]; Sn ] of unlabeled F structures on [1..n]. Then the isomorphism type generating function of a species F is the formal power series
F (z) = f (n)z n ,
n0

where f (n) = |T (Fn )|.

As in Lemma 2, if

is a permutation, let

x()

be the number of xed points of

Denition 19. Let Sn denote the group of permutations of [1..n]. Then the generating function of a species F is the formal power series
ZF (z) =
n0

cycle index

1 x(F [])z c() . n! S

n

Theorem 3. For any species F we have 1. F (z) = ZF (z, 0, 0, . . . )

2. F (z) = ZF (z, z 2 , z 3 , . . . ). Proof. We prove each part separately.
1. We have

ZF (z, 0, 0, . . . ) =
n0

1 x(F [])z c1 () 0c2 () 0c3 () n! S

n

=
n0

1 x(F [Id[1..n] ])z n n! f (n) zn . n!

=
n0

32

2. Now,

ZF (z, z 2 , z 3 , . . . ) =
n0

1 x(F [])z c1 () z 2c2 () z 3c3 () n! S

n

=
n0

1 x(F [])z n n! S
n

G = F [[1..n], Sn ] be the image function of F . Then we have

Let

of the action of

Sn

on

F [[1..n]]

given by the second

n0

1 x(F [])z n = n! S
n

n0

1 n! n! |G|

x(F [])z n
x(F [])G

=
n0

1 x( )z n |G| G f (n)z n ,

=
n0

where the last equality follows from Lemma 2.

Denition 20. Let F and G be two species. An isomorphism of F to G is a family of bijections bU : F [U ] G[U ] which satises the following condition: for any bijection : U V between two nite sets and any F -structure s F [U ], we must have bU (s) = bV ( s). The two species are then said to be isomorphic and we write F = G.
As with combinatorial classes from symbolic combinatorics, there is a set of operations on species which have corresponding operations on their generating functions (all three types). These operations include addition, multiplication, composition, dierentiation, pointing, cartesian product, and functorial composition, just as in symbolic combinatorics. For more information on these, and all other parts of the theory of species, see the 457-page book devoted to the subject [6]. The species analogs of combinatorial specications are systems of equations, which can be created using the operators just mentioned and the notion of equality from Denition 20.

7.1.1 Weighted species

The species equivalent of multiple parameters on combinatorial classes is weighted species, which we briey discuss. A ring

weighting function w : F [U ] R[t] maps objects in F [U ] to monomials in t over a R C, and an R[t]-weighted set is a pair (S, w), where w : S R[t] is a weighting
S.
For such a pair

function on

(S, w),

let

|S|w =

sS

w(s).

Denition 21. Let R C be a ring. An R[t]-weighted species F is a pair of functions,

33

1. the rst of which maps each nite set U to an R[t]-weighted set (F [U ], wU ) such that |F [U ]|wU converges, and 2. the second of which maps each bijection : U V to a function
F [] : (F [U ], wU ) (F [V ], wV ),

that preserves weights. The functions F [] must satisfy the following properties: 1. for each pair of bijections : U V and : V W ,
F [ ] = F [ ] F [],

2. and if IdU : U U is the identity function,

F [IdU ] = IdF [U ] .

Denition 22. Let F be a weighted species, with weight functions wn : F [[1..n]] R[t], for n 0. Then the exponential generating function of F is
Fw (z) =
n0

|F [[1..n]]|wn

zn ; n!

the isomorphsim type generating function of F is

Fw (z) =
n0

|T (Fn )|w z n ,

where T (Fn ) is dened as in Denition 18; and the cycle index generating function of F is
ZFw (z) =
n0

1 n!

| Fix(F [])|wn z c() ,

Sn

where, if Sn , Fix(F []) is the set of F -structures on [1..n] xed by F [].

Operations on weighted species can be dened similarly to the unweighted case. For more information, see [6].

Example 19. Let L1 be the species of linear orders resricted to sets of at least one element; let S be the singleton species with weight 1; and let W be the singleton species with weight q. Then the species T of ordered trees with number of internal nodes marked by q satises
T = S + W (L1 T )

where + is addition (the species analog of sum), is multiplication (the analog of product), and is composition (the analog of substitution).
Two notable packages for species, Darwin [5, 7], and Devmol [2] have been described in the literature, but no longer exist.

34

7.2

Aldor-Combinat

Authors: Ralf Hemmecke and Martin Rubey Website:

httpsXGGportlFrisFjkuFtGwemersGhemmekeGldorGomint

A project which started in 2006, Aldor-Combinat is an unnished package primarily for working with species, which is based on MuPAD-Combinat's implementation of symbolic combinatorics. It is written in the Aldor language to be used with the Axiom algebra system. details of the package, but it is designed to be read by Aldor developers. model to organize its functionality.

46

computer

The only included documentation contains many of the implementation

47

Like Sage uses for basic combinatorial objects, Aldor-Combinat uses a category-theoretic All species in Aldor-Combinat are objects in the

gomintorilpeies

category and they must implement methods returning exponen-

tial generating series, isomorphism type generating series, cycle index generating series, a listing of structures, and a listing of isomorphism types. Aldor-Combinat has the following species predened: empty set, set, singleton, linear order, cycle, permutation, and set partition (whose isomorphism types are integer partitions).

Example 20. We can compute the initial coecients of the exponential generating function of the singleton species:
v aa sntegerY i aa imptyetpeies vY gseX yrdinryqenertingeries Xa genertingeries 6 iY import from sntegerY leX vist snteger Xa oeffiient@gseD nA for n in HFFQY

This assigns ID HD HD H to le. We can compute something similar for the singleton species:
w aa sntegerY aa ingletonpeies wY gssX ixponentilqenertingeries Xa genertingeries 6 Y import from sntegerY lsX vist snteger Xa oeffiient@gssD nA for n in HFFQY

This assigns HD ID HD H to ls.

Let

be a species. Dene, for each

n 0,

the if

species F restricted to n by

Fn [U ] =
46 47

F [U ]

|U | = n, otherwise,

http://axiom-developer.org/ http://www.risc.jku.at/people/hemmecke/AldorCombinat/combinat.html

35

for all nite sets

U.

Aldor-Combinat allows species to be restricted with

estritedpeies,

which is useful for dening other species. Species may be dened implicitly or explicitly with the following operations: addition, multiplication, composition, and functorial composition.

7.3

Sage: combinatorial species

Download: Website:

httpXGGwwwFsgemthForgGdownlodFhtml httpXGGsgemthForgGdoGrefereneGomintGspeiesFhtml
For information on Sage's basic

This section covers Sage's species functionality.

combinatorial objects functionality and an overview Sage as a CAS, see Section 5.4. Sage's capabilities for working with combinatorial species, which began development around 2008, are based on those in Aldor-Combinat (covered in the previous section). project roadmap describes future plans for the project. and subset species. Also, the the restriction to

48

ghrteristipeies@nA method returns the characteristic species on n, which yields one object on sets of size n, and no objects on any other set (i.e. n
of the set species

Sage oers built-in classes representing the cycle, partition, permutation, linear-order, set,

X,

dened

X[U ] = U ).

Example 21. We assign an empty set species to i:

sgeX i a speiesFimptyetpeies@A

We list all structures of on the empty set and the set {1, 2}:
sgeX {} sgeX iFstrutures@AFlist@A iFstrutures@IDPAFlist@A

We nd the rst four coecients of the exponential generating function of :

sgeX iFgenertingseries@AFoeffiients@RA ID HD HD H

Since the empty set species is isomorphic to the characteristic species on 0, we get identical output using ghrteristipeies@HA:
sgeX sgeX {} sgeX
48

gH a speiesFghrteristipeies@HA gHFstrutures@AFlist@A gHFstrutures@IDPAFlist@A

http://trac.sagemath.org/sage_trac/ticket/10662
36

 sgeX gHFgenertingseries@AFoeffiients@RA ID HD HD H

The characteristic species on 1 is isomorphic to the singleton species:

sgeX sgeX I sgeX sgeX HD ID gI a speiesFghrteristipeies@IA gIFstrutures@IAFlist@A gIFstrutures@IDPAFlist@A gIFgenertingseries@AFoeffiients@RA HD H

The operations on species of addition, multiplication, composition, and functorial composition are supported. Species can be dened explicitly or implicitly (define). Each species object When species are created, a weight can be specied, as well as a restriction on the size of the set.

fFstrutures@setA is set, fFisotypes@setA is the set of equivalence classes of f structures (isomorphism types) on the set set, yleindexseries@A is the cycle index series, genertingseries@A is the exponential series, and isotypegenertingseries@A
implements a number of useful methods: the set of

f-structures

on the set

is the isomorphism type series.

Example 22. We wish to count unlabeled ordered trees by total number of nodes and number of internal nodes. To achieve this, we begin by assigning a weight of 1 to the leaves and q to internal nodes, which are each singleton species:
sgeX sgeX sgeX q a 9q9Fgen@A lef a speiesFingletonpeies@A internlnode a speiesFingletonpeies@weightaqA

Now we dene a species representing the trees, dened as lef C internlnodeBv@A, where v is the species of linear orders restricted to sets of size 1 or greater:
sgeX sgeX sgeX v a speiesFvineryrderpeies@minaIA a speiesFgomintorilpeies@A Fdefine@lef C internlnodeBv@AA

All that remains, since the trees are unlabeled, is to compute the coecients of the isomorphism type generating function:
sgeX Fisotypegenertingseries@AFoeffiients@TA HD ID qD qP C qD qQ C QBqP C qD qR C TBqQ C TBqP C q
37

Further examples may be found in a Sage species demo by Mike Hansen, one of the developers.

49

8
8.1

Asymptotics
Mathematical background

8.1.1 Asymptotic scales and series Denition 23. Let P be an open set and let L be a limit point of P . Let SP be a set of functions from P to C. Then the set SP is an asymptotic scale at L i for every pair of dierent functions 1 , 2 SP , the limit of 1 (x)/2 (x) as x L is either 0 or +. Example 23. If P = Z0 {}, L = , then an asymptotic scale is SP = {n 1, n n, n n2 , . . . }.
The following denition, which comes from [32] (as does the previous denition), generalizes Denition 1 for terminating series:

Denition 24. Let SP be an asymptotic scale at L and let f be a complex valued function on P . Then f is said to admit a terminating asymptotic series at L i there exists a sequence (a1 , . . . , an ) of complex numbers and a sequence (1 , . . . , n ) of elements of SP such that
i

i {1, . . . , n}, SP :

i (x) = o((x)) = f (x)

j=1

aj j (x) = o((x)) as x L.
k

z Example 24. With P, L, and SL dened as in Example 23, let fk (n) = [z n ] (1z)k be the number of integer compositions of n with k parts. Then a terminating asymptotic series for k(k1) 1 fk (n) at L consists of the complex numbers (k1)! , 2(k1)! and functions (n nk1 , n nk2 ) since

fk (n) =

nk1 k(k 1)nk2 + o(nk2 ). (k 1)! 2(k 1)!

Various CASes oer support for generating and working with asymptotic series expansions, including Maple, Mathematica, Maxima, and MATLAB's Symbolic Math Toolbox (the current product containing the code of MuPAD) of implementations of special interest.

50

; in this section we include a couple

49 50

http://sage.math.washington.edu/home/mhansen/CombinatorialSpeciesDemo.html http://www.mathworks.com/help/toolbox/mupad/stdlib/asympt.html

38

8.1.2 Singularity analysis

Singularity analysis is a method due to Flajolet and Odlyzko which shows how to compute asymptotic expressions for the coecients of some classes of generating functions [12]. In order to apply the theorem, we forget about power series and treat generating functions as elements of C (C; 0) instead of Q[[z]]. In its simplest form, the method can be expressed as the following theorem:

formal

Theorem 4 (Singularity analysis). Fix the range of Arg to be [, ). Let F C (C; 0) be a function analytic in a domain
D = z : |z| s1 , | Arg(z s)| > , 2

where s, s1 > s, and are three positive real numbers. Assume that, with (u) = u log u and {0, 1, 2, . . . }, we have /
F (z) 1 1 z/s

as z s in D.
(n) . n()

Then, as n , the Maclaurin coecients of F satisfy

[z n ]F (z) sn

Generalizations are discussed at length in [13].

8.1.3 Saddle-point asymptotics

The saddle-point method is another way of computing asympotic expressions for the coefcients of generating functions. In [21], W.K. Hayman dened a set of functions called the

2b(r(n)) where r(n) is the smallest positive root of r(n)F (r(n))/F (r(n)) = n and b(r) = rDr (rF (r)).
Useful information on which functions are H-admissible is given by the following theorem (also from [21]):

H-admissible functions for which the following theorem holds: Theorem 5. If F is a function dened around the origin by F (z) = H-admissible, then F (r(n)) f (n) , as n n
r(n)

n0

f (n)z n and F is

Theorem 6. The following are three properties of H-admissible functions. 1. Let and 1 be positive real numbers, and let 2 and 3 be real numbers. Then F dened by
F (z) = exp 1 (1 z) 1 1 log z 1z
2

2 log z

1 1 log z 1z

is H-admissible.
39

2. If F and G are H-admissible, and P is a polynomial function with real coecients and positive leading coecient, then exp(F ), F +G, F +P, P (F ), and P F are H-admissible. 3. If P is a polynomial function such that P (z) cannot be written as P (Q(z k )) for any polynomial function Q and k > 1, then exp(P ) is H-admissible.
The saddle-point method are discussed in further detail in [13].

8.1.4 Power series coecients and the radius of convergence Notation 2 (from [13]). Given a sequence (a0 , a1 , a2 , . . . ) of rational numbers and r [0, ], we write
an rn

i where 1/ = 0, 1/0 = .

lim sup |an |1/n = r,

n0

Proposition 2. Given a sequence (a0 , a1 , a2 , . . . ) of rational numbers and r [0, ], we have an r i an rn c(n), where lim sup |c(n)|1/n = 1. In other words fn r i 1/r is the radius of convergence of n n0 an z . Denition 25. Given a function F with Maclaurin series expansion F (z) = n0 f (n)z n having radius of convergence r, with a nite set of singularities at distance r from the origin, the dominant directions of F is the set where
[, ) and F is singular at rei .

8.2

Gdev

Author: Bruny Salvy Download: Webpage:

httpXGGlgoFinriFfrGlirriesG5down httpXGGlgoFinriFfrGlirriesG5gdev
51

Last modied: March 2003

Gdev is a package for computing asymptotic expressions for functions and generating function coecients, which, except for the

equivlent function (described below), has been

52
. For more information about the

superseeded by the MultiSeries package (see Section 8.4).

capabilities of gdev see an article written by Bruno Salvy [32].

51 52

http://algo.inria.fr/salvy/index.html http://algo.inria.fr/salvy/index.html

40

Many packages provide routines for computing Taylor expansions, but the capabilities of gdev's function

gdev

go well beyond that particular asymptotic scale. The function

gdev

takes a function given by an explicit expression and, automatically choosing the asymptotic scale to use, nds an asymptotic expansion around a point, which ends with a big O term. The function

equivlent,

originally in ANANAS from LUO, extracts an asymptotic See the next section for more information on the

expression for the coecients of an explictly given generating function using singularity analysis or the saddle-point method. underlying algorithm.

8.3

Lambda-Upsilon-Omega (LUO): asymptotics

Authors: Bruno Salvy and Paul Zimmermann Download: Website: This

httpXGGwwwFloriFfrG~zimmermGsoftwreGluoPFIFtrFgz

Last modied: May 1995

httpXGGlgoFinriFfrGlirriesGlirriesFhtml5luo
section covers ANANAS, the component of LUO for extracting asymptotic

information from generating functions. To nd out about LUO's other component, ALAS, and how ANANAS ts in to the LUO system, see Section 5.4. Once ALAS produces a system of equations for Bruno Salvy. Not all explicit expressions can be handled by ANANAS, however  in particular, not all expressions produced by ALAS.

A<1> (z), . . . , A<r> (z),

an attempt to

solve it is made. If an explicit expression is obtained, it is passed to ANANAS, written by

1 , E(F ) = exp(F ). 1F ANANAS can handle some functions not in E , but for the rest of this section, we only consider ANANAS's behavior on functions in E , since the documentation focuses on this Q(F ) = 1 , 1F L(F ) = log
case. The set

Denition 26. The set E of functions is the set of generating functions corresponding to labeled classes dened by well-dened explicit specications. That is, the functions containing the monomial functions z 1 and z z , and closed under the operations {+, , Q, L, E}, where

is subdivided into three disjoint sets

E = EAL Eentire Eother ,

where

logarithmic

Eentire

is the set of entire functions and

(AL) growth, meaning that

F EAL

EAL
i

is the set of functions with

algebraic-

has radius of convergence

0<r<

and there exist

R, k Z0 F (z)

such that, as

1 1 logk (1 z/r) 1 z/r

41

z r .

ANANAS determines which subset of some functions from as complete as for algorithm used

a function is a member of and proceeds accord-

ingly. For the details of the steps involved in this choice, see [14]. ANANAS is able to handle

Eother ,

Eentire and Eother using the saddle-point method, but the coverage is not EAL , where singularity analysis is very eective. We briey describe the for the class EAL ; for information on what ANANAS does with Eentire and EAL ,
ANANAS performs the algorithm Equivalent:

see [14].

Given a function in

Algorithm 1 Equivalent Require: F is an expression corresponding to a function in EAL

1. Determine the radius of convergence 2. Find the dominant directions

of

if F

of

has a single dominant direction

then
F
around

F,

3.1 Obtain an algebraic-logarithmic asymptotic expansion of

else

3.2 Apply singularity analysis 4.1 Apply a singularity-analysis style result for functions with more than one dominant

singularity

end if
Step

turns out to be simple; it takes the form of the Radius algorithm:

Algorithm 2 Radius Require: F is an expression corresponding to a function in EAL if F is a polynomial then else if F
then return

exp(G)
or

matches is

then then
G g(x) = 1 G2 G2 then Radius on G1 and G2
or

else if F else if F end if

return the result of Radius on

Q(G)

L(G)

return the smallest real positive root of matches

G1 + G2

return the minimum of

For more details on the rest of Equivalent, see [14].

8.4

MultiSeries

Authors: Bruno Salvy and Maplesoft Download: Website:

httpXGGlgoFinriFfrGlirriesG5down

42

httpXGGwwwFmplesoftFomGsupportGhelpGwpleGviewFspxcpthawultieries
Originally developed by Bruno Salvy at the INRIA Algorithms Group and now built into Maple releases, the MultiSeries Maple package provides functionality for computing asymptotic series expansions that superseeds that provided in gdev, the subject of Section 8.2. MultiSeries contains the function special cases of

multiseries that cover the same functionality of core Maple functions: sympt, series, limit, and tylor. According to the Maple documentation, the simplest use of the package is by overriding the standard sympt, series, limit using with. The
corresponding MultiSeries functions are often more powerful than the default ones, but require more computational time. (Presumably The the variable

multiseries,

the sucessor to gdev's

gdev,

and four

multiseries function can take as arguments an expression expr, the limit point , x in the expression tending to , and the truncation order n. It returns, like gdev, an asymptotic expansion of expr as x tends to ending in a big O term, in terms of
an asymptotic scale automatically chosen. It is also possible to specify some extra options such as choosing the path of approach, and there is support for choosing the asymptotic scale used.

tylor

can be used in the same way.)

53

Conclusions
First, recall that we started this project with the claim that [it is no longer] necessary

A number of kinds of conclusions can be made from this project.

to use error-prone pen and paper methods to perform an ever-growing set of mathematical procedures. Indeed, the packages in this project cover a lot of ground, and much of the progress is recent, but there is massive potential for more work both with mathematical algorithms that have been implemented and those that have not. Note that there are few packages available for any one CAS/programming language. The average mathematician cannot be expected to be familiar with more than one system, if any. There is therefore a need for projects to be ported to systems without those capabilities, preferably the big three CASes: Maple, Mathematica and Sage. Next, future package-writers can note that not all of the packages in this project come with sucient documentation for all uses. documentation should be available: To be maximally helpful, two forms of one for mere users, and one for developers or other

people interested in the implementation details. Users want to know what the package can do and how to do those things. Developers wishing to port or extend the software want to know the algorithms and code used.

53

http://www.maplesoft.com/support/help/Maple/view.aspx?path=MultiSeries%2fmultiseries

43

Lastly, as casual observations, we can say that learning how to use a package can be a good way to learn a mathematical concept or increase one's understanding, and also that these packages are for the most part not huge collaborative eorts  often there is only one author, sometimes two. Writing symbolic packages is not trivial, since it requires skills from multiple disciplines, namely mathematics, computer science, and software engineering. However, it is clear that the future of the eld is bright, and the global project to replace pen and paper, and ultimately more and more of the problem solving, theorem proving and question answering now done only by human mathematicians will continue to progress and advance signicantly in the years to come.

References
[1] T. Andreescu and Z. Feng.

gies.

A path to combinatorics for undergraduates: counting strateSm. Lothar. Combin., 49:Art. B49z, 34, 2002/04.
Computing the molecular expansion of species

Birkhuser, Boston, 2004.

[2] P. Auger, G. Labelle, and P. Leroux. with the Maple package Devmol. [3] F. Baader and T. Nipkow. Cambridge, 1998.

Term rewriting and all that.

Cambridge University Press,

[4] E.A. Bender and D. Zeilberger. Some asymptotic bijections.

A, 38(1):9698, 1985.

J. Combin. Theory Ser. Lecture Notes

[5] F. Bergeron and G. Cartier. Darwin: Computer algebra and enumerative combinatorics.

in Computer Science, pages 393394. Springer Berlin / Heidelberg, 1988.

[6] F. Bergeron, G. Labelle, and P. Leroux.

In Robert Cori and Martin Wirsing, editors,

STACS 88,

volume 294 of

Combinatorial species and tree-like structures, volume 67 of Encyclopedia of Mathematics and its Applications. Cambridge University

Press, Cambridge, 1998. Translated from the 1994 French original by Margaret Readdy, With a foreword by Gian-Carlo Rota. [7] F. Bergeron and C. Pichet. Darwin, a system designed for the study of enumerative

combinatorics. In Bob Caviness, editor,

in Computer Science, pages 444445. Springer Berlin / Heidelberg, 1985. Combinatorics of permutations. Advanced combinatorics.
(Boca Raton). Chapman & Hall/CRC, Boca Raton, FL, 2004. Richard Stanley.

EUROCAL '85,

volume 204 of

Lecture Notes

[8] M. Bna.

Discrete Mathematics and its Applications With a foreword by

[9] L. Comtet.

D. Reidel Publishing Co., Dordrecht, enlarged

edition, 1974. The art of nite and innite expansions.

44

[10] N.G. de Bruijn.

Asymptotic methods in analysis.

Dover Publications Inc., New York,

third edition, 1981. [11] E.W. Dijkstra. tion. [12] P. Flajolet and A. Odlyzko. Singularity analysis of generating functions.

A discipline of programming.

Prentice-Hall Inc., Englewood Clis, N.J.,

1976. With a foreword by C.A.R. Hoare, Prentice-Hall Series in Automatic Computa-

Discrete Math., 3(2):216240, 1990.

Cambridge, 2009.

SIAM J.

[13] P. Flajolet and R. Sedgewick.

Analytic combinatorics.

Cambridge University Press,

[14] P. Flajolet, P. Zimmermann, and B. Salvy.

Lambda-Upsilon-Omega :

an assistant

algorithms analyzer. Rapport de recherche RR-0876, INRIA, 1988. [15] P. Flajolet, P. Zimmermann, and B. Salvy. Lambda-Upsilon-Omega the 1989 cookbook. Rapport de recherche RR-1073, INRIA, 1989. [16] T. Gowers, J. Barrow-Green, and I. Leader, editors.

ematics.

The Princeton companion to mathAddison-Wesley

Princeton University Press, Princeton, NJ, 2008.

[17] R.L. Graham, D.E. Knuth, and O. Patashnik. science.

Concrete mathematics.

Publishing Company, Reading, MA, second edition, 1994. A foundation for computer

[18] N.T. Gridgeman. The search for perfect numbers. 1963. [19] F. Harary and E.M. Palmer. [20] G.H. Hardy.

The New Scientist,

18(334):8588,

Graphical enumeration.

Academic Press, New York, 1973. Hafner

Orders of innity. The Innitrcalcl of Paul du Bois-Reymond.

Publishing Co., New York, 1971.

Reprint of the 1910 edition, Cambridge Tracts in

Mathematics and Mathematical Physics, No. 12. [21] W.K. Hayman. A generalisation of Stirling's formula. 95, 1956. [22] M. Klin, S. Reichard, and A. Woldar. algebra experimentation. In pages 67112. Springer, Berlin, 2009. [23] D.E. Knuth. Siamese combinatorial objects via computer

J. Reine Angew. Math., 196:67

Algorithmic algebraic combinatorics and Grbner bases,

The art of computer programming. Vol. 4A.

Upper Saddle River, New

Jersey: Addison-Wesley, 2011. Combinatorial Algorithms, Part 1. [24] D.L. Kreher and D.R. Stinson.

and search.

Combinatorial algorithms: generation, enumeration,

CRC Press, 1999.

45

[25] M. Mishna.

Math., 30(1-2):189207, 2003.

dale, AZ, 2001).

Attribute grammars and automatic complexity analysis.

Adv. in Appl.

Formal power series and algebraic combinatorics (Scotts-

[26] A. Nijenhuis and H.S. Wilf.

Combinatorial algorithms.

Academic Press Inc. [Harcourt For computers and

Brace Jovanovich Publishers], New York, second edition, 1978. calculators, Computer Science and Applied Mathematics. [27] S. Pemmaraju and S. Skiena.

Computational discrete mathematics.

Cambridge Univer-

sity Press, Cambridge, 2003. Combinatorics and graph theory with Mathematica. [28] C. Pivoteau, B. Salvy, and M. Soria. Algorithms for combinatorial systems. Technical report, arXiv, 2011. 58 pages. [29] G. Plya and R. C. Read.

compounds.

Combinatorial enumeration of groups, graphs, and chemical Algo-

Springer-Verlag, New York, 1987. Plya's contribution translated from the

German by Dorothee Aeppli. [30] D. Poulalhon and G. Schaeer. Optimal coding and sampling of triangulations.

rithmica, 46(3-4):505527, 2006.

[31] F. Ruskey and A. Williams. Generating combinations by prex shifts. In Berlin, 2005. [32] B. Salvy. Examples of automatic asymptotic expansions. 1991. [33] R. Sedgewick. Permutation generation methods. [34] R.P. Stanley.

combinatorics, volume 3595 of Lecture Notes in Comput. Sci., pages 570576. Springer, SIGSAM Bull., 25:417, April

Computing and

Comput. Surveys, 9(2):137164, 1977.

Enumerative combinatorics. Vol. 1, volume 49 of Cambridge Studies in Advanced Mathematics. Cambridge University Press, Cambridge, 1997. With a foreword Enumerative combinatorics. Vol. 2, volume 62 of Cambridge Studies in Advanced Mathematics. Cambridge University Press, Cambridge, 1999. With a foreword The Mathematica Guidebook: Programming with CD-ROM.
Springer-Verlag

by Gian-Carlo Rota, Corrected reprint of the 1986 original. [35] R.P. Stanley.

by Gian-Carlo Rota and Appendix 1 by Sergey Fomin. [36] M. Trott.

New York, Inc., Secaucus, NJ, USA, 1st edition, 2002. [37] H.S. Wilf.

Combinatorial algorithms: an update, volume 55 of CBMS-NSF Regional Conference Series in Applied Mathematics. Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA, 1989.

[38] H.S. Wilf.

generatingfunctionology. A K Peters Ltd., Wellesley, MA, third edition, 2006.

[39] P. Zimmermann. ALAS : Un systeme d'analyse algebrique. Rapport de recherche RR0968, INRIA, 1989.

46

Appendix
Maple 16
Maple is a modern commercial CAS whose development began in 1980 by the Symbolic Computation Group at the University of Waterloo, and, since 1988, has been developed and sold by Waterloo Maple Inc., a.k.a. Maplesoft. The most recent version as of time of writing is Maple 16, released in 2012.

Mathematica 8
Mathematica, another modern commercial CAS, is developed by Wolfram Research. It began as a project called Symbolic Manipulation Program created by Stephen Wolfram around 1979 at Caltech, and then after Wolfram Research was founded in 1987, was renamed Mathematica. Version 1.0 of Mathematica was released in 1988; the most recent release was Mathematica 8, in 2010. Mathematica is a Turing complete

term rewriting system,

which means that all MathWhen an expression is

ematica programs compute by manipulating expressions (terms).

evaluated

by Mathematica, it searches through a sequence of stored pattern-replacement When a rule's pattern matches a

rules (in which form Mathematica's mathematical routines exist) looking for a match between a pattern and a subexpression of the expression. is evaluated. subexpression, that subexpression is replaced according to the rule, and the new expression For more on term rewriting systems, see [3], and for more on Mathematica programming, see its documentation and [36].

47