Universita` degli Studi di Torino
Scuola di Scienze della Natura
Corso di Studi in Matematica
Combinatorial species and transitive relations
Tesi di laurea di secondo livello - A.A. 2012/2013
candidato: relatore:
Daniele P. Morelli Prof.ssa Lea Terracini
Combinatorial species and transitive relations
Daniele P. Morelli
Contents
Preface 4
Introduction 8
1 Preliminary Notions 9 1.1 Integer sequences and power series ...... 9 1.2 Obtaining generating function from recurrences ...... 13 1.3 Labeled and unlabeled structures ...... 15 1.4 Basic notions about categories ...... 16 1.5 Species of Structures ...... 18 1.6 Series associated to species ...... 18 1.7 Isomorphism of species ...... 21
2 Classical enumerative tools 23 2.1 Stirling numbers of the second kind ...... 23 2.2 Burnside’s lemma ...... 24 2.3 M¨obiusinversion ...... 26 2.4 Partitions ...... 26 2.5 Euler Transform ...... 29
3 Operations on species 31 3.1 Sum of species ...... 31 3.2 Product of species ...... 32 3.3 Composition of species ...... 36 3.4 Derivative of a species ...... 41 3.5 Pointing of a species ...... 43 3.6 Cartesian product of species ...... 44 3.7 Functorial composition ...... 45
4 Relation theory 47 4.1 Basic notions and results about relations ...... 47 4.2 Operations on relations ...... 49 4.3 Posets, order ideals and lattices ...... 50
2 5 Algebraic approaches to partial orders 53 5.1 Monomial orders and Gr¨obnerbases ...... 53 5.2 Toric ideals ...... 55 5.3 Hibi ideal of a poset ...... 57 5.4 Toric ideals associated to posets ...... 57 5.5 Order ideals and cartesian product of posets ...... 58 5.6 Adjoint functors in the category of posets ...... 60
6 Counting transitive relations 63 6.1 Transitivity ...... 63 6.2 Indecomposable partial orders ...... 67 6.3 Graded orders ...... 71 6.4 Finite topologies ...... 74
7 Conclusions 77 7.1 Experimental investigations ...... 77 7.2 Posets and graded posets ...... 79
Appendix 82 A1 Common species and operations ...... 82 A2 PARI/gp Routines ...... 84 A3 Tables of integer sequences ...... 89
References 96
3 Preface
The use of power series for combinatorial purposes dates back at least to the age of Euler, who gave some of the first non-trivial examples of ordinary generating functions. Some times later Laplace had the insight that alge- braic operations on power series correspond to set-theoretic manipulations on the object counted by the sequences of coefficients of such series. As al- ways happens in mathematics, several decades of studies and investigations were needed to understand how a bunch of technical tools can in fact be unified in a general setting, and a theoretical background be given to such mathematical machinery. When a scientifical discipline needs clarification on a foundational level, the most problematic aspects appear to emerge with respect to definitions rather than theorems. For instance, a good definition of combinatorial struc- ture is not easy to formulate, and a lot of philosophical questions must be answered: what is the difference between labeled and unlabeled structures? Is it possible to formulate the concept of labeled structure without refer- ring explicitly to a specific set of labels but without making the structure unlabeled? Is it possible to give a rigorous combinatorial interpretation of functional composition? Is there a general notion of isomorphism, which is independent of the particular kind of structure considered? Is there a functional equivalent of the action of making a vertex distinguishable from the others (like in rooted trees)? How does the action of the symmetric group Sn over a set with n elements induces a correspondent action over the structures of some kind built over this set? The theory of combinatorial species, introduced by Andr´eJoyal in the Eighties (see [6]), constitutes an unifying step in the foundations of combi- natorics. It provides a general setting for most well-known results in discrete mathematics, and a solid background for the understanding and the use of generating functions and of the study of labeled and unlabeled combinatorial structures. It also helps solving most of the previously mentioned questions. This theory helps us understand most previously known but somewhat obscure facts, and to give new insights in the understanding of combina- torics. We could trace a parallel with Category theory (that in fact provides the basic notions on which the theory of species is constructed), which is a well-known example of a generalizing instrument that made clear a lot of known results. The concept of the fundamental group of a topological space, for instance, would be a lot less clear if we did not know what a category is and what a functor is. In the words of Gian-Carlo Rota:1
Every discovery of a new scientific fact is a challenge to uncover the uderlying mathematical structure. This structure is not “ab-
1From [11].
4 stracted” from nature, as psychologists would have us believe. It is the basic makeup of nature, it was always there, waiting to be told and staring at us all the time. The natural laws discovered by scientists will be refined like a metal, polished like a jewel and finally stored as theorems in the archives of mathematics. Math- ematicians triumphantly point to mechanics as the example of a theory that began as an empirical science, and that eventually made its way into mathematics as a generalized geometry, geom- etry with time added. Mathematicians believe that every science will sooner or later meet the fate that befell mechanics.
This may sound excessively positivist, but it is undeniable that the most important advances in mathematics are those that help us understand the “underlying mathematical structure” of some kind of discovery. Even the solution of a big problem is most often credited for the new technical ma- chinery provided and for the general understanding of a subject, rather than for supplying a mere answer to a mathematical question. On the other hand, we must admit that a good theory should in fact be able to solve new and important problems. We are confident that the theory of species will soon prove itself worth of this task, and this work would like to provide some insights that this theory allows us to make. We’d like to cite Rota again:2
It would probably be counterproductive to let it be known that behind every “genius” there lurks a beehive of research mathe- maticians who gradually built up to the “final” step in seemingly pointless research papers. And it would be fatal to let it be known that the showcase problems of mathematics are of little or no in- terest for the progress of mathematics. [...] There is a second way by which mathematics advances, one that mathematicians are also reluctant to publicize. It happens when- ever some commonsense notion that had heretofore been taken for granted is discovered to be wanting, to need clarification or definition. Such foundational advances produce substantial divi- dends, but not right away.
I like to remember how seemingly small changes in mathematical nota- tion can drastically enhance the power of our thinking. The first use of the arrow notation f : X → Y for the identification of a function dates around 1941.3 Before that date, the notation f(X) ⊆ Y was mostly used. Very few years after, the theory of categories was born.
2From the foreword of [1]. 3W. Hurewicz, On duality theorems, Bull. Am. Math. Soc., 47, 562–563 (1941), as cited in [9].
5 This is not just an historical curiosity. Category theory is the discipline of arrows, after all. Obviously, it could have been discovered even before the introduction of the arrow notation, but it didn’t, because a bunch of symbols were needed to let us keep in mind that, verily, a function is a map. The theory of species seems to be something like that: a new way to define things and refer to them. It is useful because most important com- binatorial constructions are automatically handled by the algebraic tools provided by it, and it is enlightning because it gives an elegant theoretical instrumentation that also helps us recall the true (onthological and ethimo- logical) nature of combinatorics: not simply the art of counting, but the art of combining.
6 7 Introduction
This work is intended to introduce the theory of combinatorial species and to give some ideas about how it may help to find solutions to some open problems in enumerative combinatorics. In particular, we focused on the enumeration of binary relations over a finite set, specifically transitive rela- tions. The first section is intended to review some basic facts about combina- torics and to introduce the notion of combinatorial species. Basic notions about integer sequences and power series are recalled, and standard ways to relate those sequences to power series are introduced. Fundamental re- sults about ordinary and exponential generating functions are stated, and a general notion of isomorphism is defined, leading to the dichotomy between labeled and unlabeled structures. Section two is devoted to the study of some well-known enumerative tools which will be useful in the foregoing discussion. Stirling numbers of the second kind, Burnside’s lemma and the basic theory of integer partitions (including Euler transform) are treated. Also, a brief introduction to M¨obius inversion is given. The third section is a survey of the operations that can be defined over combinatorial species. Some classical examples are introduced and we’ll state and prove very important results relating those operations to functional operations over the corresponding generating series. Section four introduces the theory of binary relations, with particular attention to the transitive ones. Some important results about posets are recalled (including a celebrated result by Garrett Birkhoff). The fifth section is a survey of some contemporary algebraic approaches to the theory of partial orders. The theory of Hibi rings is introduced, and we also report a characterisation of the associated toric rings in terms of Gr¨obnerbases. Also, some categorical results about posets are stated, and a theorem about adjoint functors is derived. Section six is about some kinds of transitive relations. Classical results are recalled and translated in the language of combinatorial species, leading to functional equations that allow us to compute or relate the number of particular families of transitive relations. Bijective proofs relating transitive relations with topological spaces over finite sets are also stated and proved. The conclusion offers some possible uses of the machinery developed before, and spots out some unsolved problems. The appendix contains several tables summarizing properties of and se- ries associated to species, some PARI/gp code, and the most important integer sequences studied in this work.
8 1 Preliminary Notions
In this section we introduce the algebraic concepts needed to develop an efficient foundation for the theory we want to construct. The main ideas about generating functions are first presented, and some basic results and operations are recalled. Then we are going to give an abstract and flexible theoretical background for the subsequent developments: category theory is known for its great generality and constitutes an ideal basis for the funda- mental notions we will introduce.
1.1 Integer sequences and power series
Let a0, a1, a2... be an integer sequence (i.e. a function N → Z). Most often such a sequence will count the number of some kind of combinatorial structures, i.e. an is the number of such structures built over a set with n elements.
Definition 1.1 The ordinary generating function of the sequence
a0, a1, a2, a3... is the formal power series
∞ X n 2 3 anx = a0 + a1x + a2x + a3x + ... n=0
We can define the sum of two power series as follows:
X n X n X n anx + = bnx = (an + bn)x where the sums range over all non negative integers n. The product of f and g is defined according to Cauchy’s rule:
n X n X (f · g)(x) = cnx , where cn = akbn−k k=0
P n It is well known that a series f = anx has a reciprocal if and only if a0 6= 0. Another important binary operation is composition:
∞ X n (f ◦ g)(x) = an(g(x)) . n=0 A standard result says that the computation of the n-th term of the com- position f ◦ g is possible iff g(0) = 0 (or if f is a polynomial). Clearly, the series 0, 1, and x behave as neutral elements with respect to addition, product, and composition. The set Z[[x]] of formal power series is a ring when equipped with sum and product, and ◦-invertible elements form a group with respect to composition. We will denote with f −1(x) the reciprocal of f(x) (i.e. the inverse with respect to product), and f (−1) the reverse of f(x) (i.e. the inverse with respect to composition), wheneter they are defined. We now give a simple but fundamental result allowing us to invert many combinatorial identities involving power series.
Lemma 1.1 Let f(x), g(x) and h(x) be power series, such that the constant term of h(x) be zero. If f(x) = g(h(x)) then the equation g(x) = f(h(−1)(x)). holds.
Also, we state and prove the following
Lemma 1.2 Let f(x), g(x) and h(x) be power series. If the constant term of h is zero, and the constant terms of f and g are equal, then the equation
f(x) = g(h(x)) can be inverted as
h(x) = (g − c)(−1)((f − c)(x)), where c is the constant term of f and g.
Proof: Let us write g as (g − c) + c:
f(x) = ((g − c) + c) ◦ h(x).
Composition of power series is distributive on the right with respect to sum, so we get f(x) = ((g − c) ◦ h)(x) + (c ◦ h)(x). Now, the composition of a constant term c with an arbitrary series is the constant c itself, so a rearrangement of the terms yelds
f(x) − c = ((g − c) ◦ h)(x).
10 Finally we can compose to the left with the reverse series (g − c)(−1) (this is possible because the constant term of g − c is zero).
The preceding two lemmas allow us to conclude that, given an arbitrary identity between power series of the form f = g ◦ h, we can most often recover any of the three series involved by knowing the other two. Another important operation is the (formal) derivative: if
X n f(x) = anx then X n Df(x) = nan+1x . Most often we will use the xD operator, corresponding to a derivation and a multiplication by the formal indeterminate. We are now going to state some basic results about ordinary power series. A more exhaustive analysis of the following propositions can be found in [14]. We will often use the term ogf in place of ordinary generating function. The first result introduce the so-called shift operator:
∞ Theorem 1.1 Let {an}n=0 be an integer sequence, and let f be its ordinary generating function. For any integer k > 0, the series
k−1 f − a0 − a1x − ... − ak−1x xk is the ogf of the sequence {an+k}.
We now give an important interpretation of the derivative operator D:
Theorem 1.2 Let P be a polynomial and D the derivation operator. Also, ∞ let {an}n=0 be an integer sequence, and let f be its ordinary generating ∞ function. Then P (xD)f is the ogf of the sequence {P (n)an}n=0.
For instance, if P (y) = y2 + 3y + 1, then P (xD) = x2D2 + 3xD + 1 can be seen as a new operator and we have P (xD)f = x2f 0 + 3xf + 1. The ∞ previous theorem says that if f is the ogf of the sequence {an}n=0, then 2 0 ∞ x f + 3xf + 1 is the ogf of the sequence of {P (n)an}n=0. The next result is a generalisation of Cauchy’s rule for multiplication:
∞ Theorem 1.3 Let {an}n=0 be an integer sequence, and let f be its ordinary generating function. Then the series f k is the ogf of the sequence ( )∞ X an1 an2 ...ank n1+...+nk=n n=0
11 where the sum ranges over all ways to write n as a sum of k integers.
Finally we state a result giving a fast and simple way to compute the partial sums of a given sequence:
∞ Theorem 1.4 Let {an}n=0 be an integer sequence, and let f be its ordinary generating function. Then the series f/(1 − x) is the ogf of the sequence
∞ n X aj j=0 n=0 of partial sums.
We now turn to another kind of generating function: the exponential one. The only difference is the presence of a factorial term n! appearing at the denominator of the n-th term:
Definition 1.2 The exponential generating function of the sequence
a0, a1, a2, a3, ... is the formal power series
∞ X an a2 a3 xn = a + a x + x2 + x3 + ... n! 0 1 2 6 n=0
We will also write egf in place of exponential generating function, for the sake of brevity. Also, we’ll write egf(an) to refer to the egf of the sequence ∞ {an}n=0. The calculus of egfs presents some differences with respect to the ordi- nary case. The sum of two such series is defined term-by-term as in the ordinary case: (f + g)(x) = f(x) + g(x). The product of two exponential series presents an interesting property. If f = egf(an), g = egf(bn), then ∞ k ! ∞ j k+j X akx X bjx X akbsx f · g = · = , k! j! k!j! k=0 j=0 k,j≥0 that is, X X akbj xn k!j! n≥0 k+j=n
12 and the coefficient of xn/n! in f · g is given by
X n!akbj . k!j! k+j=n
Then, in the exponential case, the Cauchy product becomes
∞ ! ∞ ! ∞ n ! X xn X xn X X n xn a · b = a b n n! n n! k k n−k n! n=0 n=0 n=0 k=0 We now turn to the basic properties of the egfs. We begin with the shift operator:
∞ Theorem 1.5 Let D be the derivative operator. Also, let {an}n=0 be an integer sequence and f its egf. Then, for any integer k > 0, the series Dkf is the egf of the sequence {an+k}.
The next results is exactly the same as for ogfs:
Theorem 1.6 Let P be a polynomial and D the derivation operator. Let ∞ {an}n=0 be an integer sequence and f its egf. Then P (xD)f is the egf of the ∞ sequence {P (n)an}n=0.
We end this section with a very useful notation we will use in the fol- lowing:
P∞ n Definition 1.3 Given an ordinary series F (x) = n=0 fnx , we will often use the notation [xn]F (x) to indicate the coeffcient of xn in the expansion n of F (i.e. fn). Similarly, we may use the notation [x /n!]F (x) to indicate the coefficient of xn/n! if f is an exponential generating function.
1.2 Obtaining generating function from recurrences Using generating functions can be very useful when trying to understand the behaviour of some integer sequence defined from a recurrence relation. We now give a couple of simple examples which are very illustrative, as the technique used is very general and can be applied in a great amount of situations. A general treatment of this subject can be found in Wilf’s classical book [14]. Some insights about this kind of problems are also trated in [3]. The following section will also make clear the meaning of some aforementioned propositions. Suppose to have a sequence of integer numbers a(n)(n ≥ 0) satisfying the recurrence a(n) = a(n − 1) + a(n − 2). (1.1)
13 This is a very celebrated recurrence that, equipped with suitable initial conditions, gives rise to Fibonacci’s sequence. The initial conditions are a(0) = 0, a(1) = 1. We also assume a(n) = 0 if n < 0. From these informations we can obtain a generating function for the Fibonacci numbers, and an approximate formula (which in fact can be made exact by rounding) for computing them. Note that the recurrence (1.1) of Fibonacci numbers is true for every integer n with the exception of n = 1. Still, we can write
a(n) = a(n − 1) + a(n − 2) + [n = 1], where in the last summand we used Iverson’s notation: [n = 1] is 1 if n = 1 and 0 otherwise. Now it is easy to see that the equation is true for all positive values of n. Now, by multiplying by a formal indeterminate xn and summing over all non-negative values of n, we get
∞ ∞ ∞ ∞ X X X X a(n)xn = a(n − 1)xn + a(n − 2)xn + [n = 1]xn i=1 i=1 i=1 i=1 Now we can see that the first member is nothing but the ogf of the integer sequence we are studying. Let us call it F (x). The other summands can be written in terms of F (x) by means of proposition 1.1. We get:
F (x) = xF (x) + x2F (x) + x, and, solving for F , x F (x) = . 1 − x − x2 which is the celebrated generating function for Fibonacci numbers. Another pretty example is about binary rooted trees, which are an im- portant kind of structure very often used by computer scientists. A possible recurrent definition would be the following:
Definition 1.4 Binary rooted trees are defined recursively as follows. The empty set is a binary rooted tree. If r is a singleton, and B1 and B2 are binary rooted trees, then the ordered triple (r, B1,B2) is a binary rooted tree.
B1 and B2 are called left and right subtrees, respectively. The theory of species we want to present allows us to translate this definition into an equation that is satisfied by the species B of binary rooted trees. In particular, we have that if B(x) is the ogf of binary rooted trees, then the following equation holds:
B(x) = 1 + xB2(x),
14 from which we can easily get √ 1 − 1 − 4x B(x) = . 2x Expanding this function (e.g. by using generalized Newton binomial for- mula) one gets
∞ 2 3 4 5 X n B(x) = 1 + x + 2x + 5x + 14x + 42x + ... = cnx n=0