Meanders
Meanders
Meanders
Meanders
Michael La Croix
September 29, 2003
Contents
1 Introduction 1
1.1 Denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Enumerative Strategies . . . . . . . . . . . . . . . . . . . . . . 7
2 Elementary Approaches 9
2.1 Relating Open and Closed Meanders . . . . . . . . . . . . . . 9
2.2 Using Arch Congurations . . . . . . . . . . . . . . . . . . . . 10
2.2.1 Embedding Semi-Meanders in Closed Meanders . . . . 12
2.2.2 Embedding Closed Meanders in Semi-Meanders . . . . 14
2.2.3 Bounding Meandric Numbers . . . . . . . . . . . . . . 15
2.3 Filtering Meanders From Meandric Systems . . . . . . . . . . 21
2.4 Automorphisms of Meanders . . . . . . . . . . . . . . . . . . 24
2.4.1 Rigid Transformations . . . . . . . . . . . . . . . . . . 25
2.4.2 Cyclic Shifts . . . . . . . . . . . . . . . . . . . . . . . 27
2.5 Enumeration by Tree Traversal . . . . . . . . . . . . . . . . . 30
2.5.1 A Tree of Semi-Meanders . . . . . . . . . . . . . . . . 31
2.5.2 A Tree of Meanders . . . . . . . . . . . . . . . . . . . 32
3 The Symmetric Group 34
3.1 Representing Meanders As Permutations . . . . . . . . . . . . 34
3.1.1 Automorphisms of Meandric Permutations . . . . . . 36
3.2 Arch Congurations as Permutations . . . . . . . . . . . . . . 37
3.2.1 Elements of (
(2
n
)
That Are Arch Congurations . . . 38
3.2.2 Completing the Characterization . . . . . . . . . . . . 39
3.3 Expression in Terms of Characters . . . . . . . . . . . . . . . 42
4 The Matrix Model 45
4.1 Meanders as Ribbon Graphs . . . . . . . . . . . . . . . . . . . 45
4.2 Gaussian Measures . . . . . . . . . . . . . . . . . . . . . . . . 49
i
4.3 Recovering Meanders . . . . . . . . . . . . . . . . . . . . . . . 53
4.4 Another Matrix Model . . . . . . . . . . . . . . . . . . . . . . 54
5 The Temperley-Lieb Algebra 56
5.1 Strand Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.2 The Temperley-Lieb Algebra . . . . . . . . . . . . . . . . . . 57
6 Combinatorial Words 69
6.1 The Encoding . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
6.2 Irreducible Meandric Systems . . . . . . . . . . . . . . . . . . 73
6.3 Production Rules For Meanders . . . . . . . . . . . . . . . . . 74
A Tables of Numbers 76
Bibliography 79
ii
List of Figures
1.1 An open meander represented as a river and a road. . . . . . 2
1.2 Examples of open meanders . . . . . . . . . . . . . . . . . . . 3
1.3 The duality between the line and curve in an open meander . 3
1.4 Examples of closed meanders . . . . . . . . . . . . . . . . . . 4
1.5 Examples of semi-meanders . . . . . . . . . . . . . . . . . . . 5
1.6 Examples of Arch Congurations . . . . . . . . . . . . . . . . 6
1.7 The meandric congurations of order three . . . . . . . . . . 6
2.1 Closed meanders and odd order open meanders . . . . . . . . 9
2.2 Closed meanders and even order open meanders . . . . . . . . 10
2.3 The canonical form of a closed meander. . . . . . . . . . . . . 11
2.4 Semi-meanders specialize closed meanders . . . . . . . . . . . 13
2.5 Closed meanders specialize semi-meanders . . . . . . . . . . . 14
2.6 Minimal arch deletion . . . . . . . . . . . . . . . . . . . . . . 16
2.7 Deforming a meander to accomodate a missing arch . . . . . 16
2.8 Stretching an arch across the line . . . . . . . . . . . . . . . . 17
2.9 A sequence minimal arch deletions . . . . . . . . . . . . . . . 17
2.10 A sequence of meanders . . . . . . . . . . . . . . . . . . . . . 18
2.11 Meandric systems that are not formed by concatenation . . . 23
2.12 A meandric system is a collection of meanders . . . . . . . . . 24
2.13 A collection of meanders that is not a meandric system . . . 24
2.14 Rigid automorphisms of meanders . . . . . . . . . . . . . . . 25
2.15 Representative meanders of order 3 . . . . . . . . . . . . . . . 26
2.16 Fixed points of rigid automorphisms of the plane . . . . . . . 27
2.17 A cyclic shift . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.18 A complete orbit under the action of cyclic shift . . . . . . . 28
2.19 The orbit of order two under the action of cyclic shift . . . . 28
2.20 Representing an orbit under cyclic shift . . . . . . . . . . . . 29
2.21 A function on semi-meanders . . . . . . . . . . . . . . . . . . 31
2.22 The tree of semi-meanders of order 5 . . . . . . . . . . . . 32
iii
2.23 Shifting a meander to obtain a minimal arch . . . . . . . . . 33
2.24 The image of a meander . . . . . . . . . . . . . . . . . . . . . 33
3.1 The meandric permutation (1 10 9 4 3 2 5 8 7 6) . . . . . . . . 35
3.2 The arch conguration (1 10)(2 3)(4 9)(5 8)(6 7) . . . . . . . 38
3.3 The graph of = (1 10)(2 3)(4 9)(5 8)(6 7) . . . . . . . . . . . 39
4.1 Encoding a closed meander as a ribbon graph . . . . . . . . . 46
4.2 The neighbourhood of a vertex . . . . . . . . . . . . . . . . . 47
4.3 A coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.1 A strand diagram of order 6 . . . . . . . . . . . . . . . . . . . 57
5.2 A bijection between strand diagrams and arch congurations 57
5.3 The multiplicative generators of TL
n
(q) . . . . . . . . . . . . 58
5.4 A pictorial representation of (R1), (R2), and (R3) . . . . . . 59
5.5 Encoding an arch conguration as a path diagram . . . . . . 60
5.6 Decomposing a path diagram . . . . . . . . . . . . . . . . . . 60
5.7 A factorization of a strand diagram . . . . . . . . . . . . . . . 61
5.8 A strand diagram and its closure . . . . . . . . . . . . . . . . 63
5.9 tr
k
satises (T3) . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.10 A strand diagram (a), and its transpose (b) . . . . . . . . . . 65
5.11 A second encoding of arch congurations . . . . . . . . . . . . 68
6.1 An arch conguration as a Dyck word. . . . . . . . . . . . . . 70
6.2 The meander OODUUDUUDCDC . . . . . . . . . . . . . . 71
6.3 Cross sections of OODUUDUUDCDC . . . . . . . . . . . . 71
6.4 The four transitions . . . . . . . . . . . . . . . . . . . . . . . 72
6.5 Substitution rules for meandric words . . . . . . . . . . . . . 75
iv
Chapter 1
Introduction
Meanders are combinatorial objects with a topological avour, encapsulat-
ing properties of the interplay between planarity and connectedness. They
correspond to the systems formed by the intersections of two curves in the
plane, with equivalence up to homeomorphism within the plane. They arise
in other guises in polymer physics, algebraic geometry, and the study of
planar algebras, especially the Temperley-Lieb algebra. For applications of
meanders, the reader is referred to [1, 4, 12]. Associated with each meander,
is a crossing number. The focus of this essay is the problem of enumerating
inequivalent closed meanders with respect to their crossing numbers.
The problem of enumerating meanders has a long history, with interest
dating back at least to work by Poincare on dierential geometry, though
the modern study appears to have been inspired by Arnol
d in [1]. Also
considered, will be the related problem of enumerating semi-meanders. Em-
phasis will be placed on exact results, but asymptotic approximations will
also be described.
This essay presents several constructions that can be used to express
meandric numbers in terms of other combinatorial objects. The resulting
expressions are elegant, but computationally intractable to date. That ex-
pressions for meandric numbers can be derived through such diverse con-
structions, illustrates a sublte interrelation between seemingly unrelated ar-
eas of mathematics.
1
1.1 Denitions
We begin by dening the objects of study. Three varieties of meanders will
be of particular interest in this essay, the most natural to describe being
the open meander, which we introduce informally through a geographical
analogy before dening it formally.
Informally, an open meander may be represented as the conguration
formed by a river and a road. The river approaches the road from the
northwest, meanders around under an east-west road, and then continues
o to the east. Both the river and the road can be considered to be innite.
See Figure 1.1. It is from this geographical context that the name meander
is derived. The imagery of this description encapsulates several of the key
features of meanders that need to be formalized.
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
Figure 1.1: An open meander represented as a river and a road.
A meander consists of two distinguished simple planar curves; in the
case of the geographical analogy these are the river and the road. The
two curves cross a nite number of times and these crossings account for
all of the intersections between the curves. We are interested in meanders
only up to equivalence under homeomorphism within the plane and impose
an orientation on the river to prevent unwanted equivalences arising from
rotating a conguration to swap the ends of the river or road. Combining
these features, we can now dene an open meander.
Denition 1.1.1 (Open Meander). An open meander is a conguration
consisting of an oriented simple curve and a line in the plane, that cross
2
a nite number of times and intersect only transversally. Two open mean-
ders are equivalent if there is a homeomorphism of the plane that maps one
meander to the other.
A natural parameter of an open meander is its order. Given a meandric
conguration, the order of the meander is the number of crossings between
the two curves. Figure 1.2 gives three examples of open meanders and indi-
cates their orders.
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
n = 7 n = 7 n = 8
Figure 1.2: Examples of open meanders and their orders, denoted by n.
It is worth noting that in the denition of an open meander, the line and
curve play symmetric roles. Since a line and a simple curve are homeomor-
phic, and we are only interested in dening meanders up to homeomorphism,
their roles can be reversed by imposing an orientation on the line instead
of the curve. A homeomorphism of the plane can then be used to convert
this conguration to the familiar form. In light of this equivalence between
the curves, it will often be convenient to orient the line instead of the curve
when dening other varieties of meanders.
PSfrag replacements
(a) (b) (c) (d)
(e)
Figure 1.3: The duality between the line and curve in an open meander
This duality is illustrated in Figure 1.3. Beginning with an open mean-
der we orient the line (a). The curve is straightened by applying a homeo-
3
morphism of the plane (b-d). Straightening the curve, curves the line. By
ignoring the nal orientation on the line, we are left with a new open mean-
der (d). By our choice in orientations, applying the transformation a second
time has the result of rotating the conguration by a half turn in the plane.
The process can be made self inverse by reecting the nal conguration
across the line.
For the purpose of enumeration a more convenient class of objects of
study is the class of closed meanders, and it is customary to refer to these
simply as meanders.
Denition 1.1.2 (Closed Meander). A closed meander is a planar con-
guration consisting of a simple closed curve and an oriented line, that cross
nitely many times and intersect only transversally. Two meanders are
equivalent if there exists a homeomorphism of the plane that maps one to
the other.
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
n = 2 n = 3 n = 4
Figure 1.4: Examples of closed meanders, and their orders, denoted by n.
The order of a closed meander is dened as the number of pairs of inter-
sections between the closed curve and the line. Since the two curves intersect
an even number of times, the order of a closed meander is an integer. Ex-
amples of closed meanders are provided in Figure 1.4.
A third variety of meanders is obtained by replacing the line in the
denition of a close meander, with a ray. If we think of the ray as a river
with a source, and think of the curve as a road, this corresponds to a road
that wraps around the source of the river.
Denition 1.1.3 (Semi-meander). A semi-meander is a planar congu-
ration consisting of a simple closed curve and a ray, that cross nitely many
4
times and intersect only transversally. Two semi-meanders are equivalent if
there exists a homeomorphism of the plane that maps one to the other.
As with open meanders, the number of intersections between the two
curves of a semi-meandric conguration may be odd, and the order of a
semi-meandric conguration is the number of intersections between the two
curves. In addition to order, semi-meanders have a second natural parameter
of enumerative importance. The winding number of a semi-meander is the
minimum, over all equivalent congurations, of the number of times the
closed curve crosses the extension of the ray to a line. Examples of semi-
meanders are provided in Figure 1.5. Notice that the middle two examples
are equivalent as semi-meanders.
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
n = 5, w = 3 n = 4, w = 0 n = 4, w = 0 n = 5, w = 1
Figure 1.5: Examples of semi-meanders, their orders, n, and their winding numbers, w.
The apparent inconsistency in the of denition of the order of a closed
meander and the order of a semi-meander is justied in Section 2.2.1, where
both classes are viewed in terms of arch congurations, which we now intro-
duce.
Denition 1.1.4 (Arch Conguration). An arch conguration is a pla-
nar conguration consisting of pairwise non-intersecting semicircular arches
lying on the same side of an oriented line, arranged such that the feet of the
arches are equally spaced along the line.
The order of an arch conguration is the number of arches it contains.
An arch conguration consisting of concentric arcs is called a rainbow con-
guration. Such congurations are used in representing semi-meanders.
Figure 1.6 illustrates some examples of arch congurations. The nal exam-
ple is the rainbow conguration of order 5.
5
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
Figure 1.6: Examples of arch congurations of order 5.
Arch congurations play an essential role in the enumerative theory of
meanders. They are used to obtain the canonical representatives of both me-
anders and semi-meanders that are used in subsequent constructions. They
also have a natural link to the Temperley-Lieb algebra, which is discussed
in Chapter 5.
The principal enumerative problem associated with meanders is to de-
termine the number of inequivalent meandric congurations of various forms
with respect to order. We introduce notation to represent these numbers for
the varieties of meanders already introduced. The numbers of inequivalent
open, closed, and semi-meanders of order n are denoted by m
n
, M
n
, and
M
n
, respectively, and are called the n-th open meandric number, (closed)
meandric number, and semi-meandric number, respectively. We let C
n
de-
note the number of arch congurations of order n.
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
Figure 1.7: The eight inequivalent meandric congurations of order three.
As an example, Figure 1.7 gives all of the non-equivalent meandric con-
gurations of order three, the lines each being understood to be oriented
from left to right. So M
3
= 8. The rst few meandric numbers are listed in
Table A.1.
6
1.2 Enumerative Strategies
To date, there is not a complete solution to the meander problem. Several
constructions can be used to rephrase it in dierent forms, but none of these
approaches has yet succeeded in making the problem more tractable.
Chapter 2 presents some elementary approaches to the problem of enu-
merating meanders. Open meanders are related to closed meanders. Mean-
ders and semi-meanders are then discussed in terms of arch congurations.
This leads to a denition of meandric and semi-meandric systems, and in
turn yields weak upper and lower bounds for the number of meanders of a
given order. Several automorphisms on the class of meanders are described
and used to derive congruences satised by the meandric numbers. The
chapter concludes with a description of an algorithm for exhaustive enumer-
ation of meanders and semi-meanders.
Each of the remaining chapters deals with a more advanced approach
to the problem, outlining the construction and mathematical tools involved,
and summarizing the conclusions that can be drawn from the approach.
Chapter 3 deals with representing meanders as permutations and then
develops some properties of the symmetric group in order to express me-
andric and semi-meandric numbers in terms of characters of the symmetric
group. These expressions provide an eective method for describing mean-
ders in purely combinatorial terms, and a bridge to the theory of symmetric
functions.
Chapter 4 presents an expression for meandric numbers in terms of ma-
trix integrals. A meandric conguration is generalized to a map with mul-
tiple roads, multiple rivers, and arbitrary genus. Counting these modied
objects with respect to number of rivers, roads, and genus is then phrased as
a problem of colouring maps. This map colouring problem is then expressed
as a matrix integral. Results about meanders can be obtained by taking
appropriate limits. To date, there are no known techniques for evaluating
the resulting integrals, but the construction rephrases the meander problem
in analytic terms.
7
Chapter 5 introduces the Temperley-Lieb algebra. The meander problem
is reduced to one of evaluating the Gram matrix of a symmetric bilinear form
on the algebra with respect to a particular basis. The chapter develops some
properties of the Temperley-Lieb algebra and relates them to the evaluation
of this form. This construction potentially opens the meander problem to
the tools of linear algebra.
Chapter 6 considers approaches to the meander problem that involve
taking cross-sections of meanders. Each intersection between the curves is
treated as a letter in a combinatorial word. This interpretation presents
the meander problem in terms of the theory of formal languages and leads
to a description of the technique that, to date, has been most eective at
exhaustively enumerating meanders.
8
Chapter 2
Elementary Approaches
2.1 Relating Open and Closed Meanders
We rst justify the exclusion of open meanders from subsequent discussion
by noting their relationship to closed meanders, which are more easily ana-
lyzed, in a large part because of their representability by arch congurations.
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
Figure 2.1: Closed meanders and odd order open meanders
Figure 2.1 illustrates the bijective correspondence between closed mean-
ders and open meanders of odd order. A unique closed meander can be
obtained from every open meander of odd order by adding an intersection
between the curves at the right of the conguration. The two free ends of
the curve are stretched around the rest of the conguration and made to
meet at this point. The process is easily reversed by breaking the closed
curve of a closed meander at its rightmost crossing with the line to obtain
9
two free ends which are extended to innity. This construction establishes
the relation:
M
n
= m
2n1
.
It is also possible to relate the number of open meanders of an even order
to closed meanders. An open meander of order 2n can be completed to a
closed meander of order n by joining the two ends of the curve at innity and
orienting the line appropriately, as illustrated in Figure 2.2. This operation
is not bijective since in the reverse operation the closed curve can be broken
along any segment incident to the innite face in the upper half-plane. Each
choice generates a dierent open meander.
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
Figure 2.2: Closed meanders and even order open meanders
We thus have the relation that the number of open meanders of order
2n is the sum, over all closed meanders of order n, of the number of pieces
of the closed curve in the upper half-plane that are incident to the innite
face.
2.2 Using Arch Congurations
Arch congurations provide a convenient medium for obtaining a canonical
representation of closed meanders. In a closed meander, the line partitions
the closed curve into two pieces, an upper piece and a lower piece, corre-
sponding to the two sides of the line when the conguration is drawn with
the line oriented from left to right. The pieces, each taken together with
the line, are referred to as the upper and lower congurations of the me-
ander. Each of these is homeomorphic to a unique arch conguration, the
existence of which requires a topological argument beyond the scope of this
10
essay. These are the canonical representations of the upper and lower con-
gurations. The canonical representation of a closed meander is the unique
meandric conguration that is homeomorphic to the original meander and
in which both the upper and lower congurations are in canonical form. PSfrag replacements
(a) (b) (c) (d)
(e)
Figure 2.3: The canonical form of a closed meander.
Figure 2.3 illustrates this representation. A closed meander (a) is split
into an upper conguration and a lower conguration given in (b). An arch
conguration is constructed homeomorphic to each piece to obtain in (c) a
canonical representation of the upper and lower congurations. Given two
arch congurations of the same order, we can superpose them by drawing
one on each side of the line and identifying their base points. Superposing
the canonical representations produces the canonical representation of the
original meander in (d).
To nd the canonical representation of a closed meander of order n,
consider the following construction. Homeomorphically deform the cong-
uration to make the points of intersection between the curve and the line
equally spaced along the line, while keeping the line xed. Deleting the in-
tersection points between the curve and the line partitions the closed curve
into 2n connected components, n on each side of the line. Replace each
of these components by a semicircular arch on the same side of the line,
with the same points of intersection with the line. The upper and lower
congurations of the resulting conguration are easily seen to both be arch
congurations of order n. Showing that the nal conguration is homeomor-
phic to the initial conguration requires repeated invocations of the Jordan
Curve Theorem.
The construction provides an injective function from classes of closed
meanders of order n to ordered pairs of arch congurations of order n, cor-
11
responding to the upper and lower arch congurations. We notice that not
every such pair of arch congurations is the image of a closed meander: in
particular, taking the same conguration for the upper congurations and
the lower congurations yields n closed curves instead of a single closed
curve. This motivates the following generalization of meanders.
Denition 2.2.1. A meandric system is the superposition of an ordered pair
of arch congurations of the same order, with the rst conguration as the
upper conguration, and the second conguration as the lower conguration.
The order of a meandric system is the order its underlying arch congu-
rations. To work with meandric systems, we use M
(k)
n
to denote the number
of k component meandric systems of order n. This is a natural generaliza-
tion of meandric numbers, in the sense that M
n
= M
(1)
n
. Values of M
(k)
n
for
small n and k are listed in Table A.4.
A sequence of polynomials is used to summarize the meandric system
numbers. The n-th meandric polynomial,
m
n
(q) =
n
k=1
M
(k)
n
q
k
, (2.1)
is the ordinary generating series of meandric systems of order n with re-
spect to number of components. Given the polynomial m
n
(q), M
n
can be
recovered as the coecient of q.
2.2.1 Embedding Semi-Meanders in Closed Meanders
Arch congurations can also be used to provide a canonical representa-
tion of semi-meanders. Consider the following construction illustrated in
Figure 2.4.
Beginning with a semi-meander (a), create a second ray with the same
origin as the rst. The new ray can be created arbitrarily close to the
original ray, such that it is distinct from the original ray but the semi-
meandric conguration consisting of the closed curve and the original ray
12
PSfrag replacements
(a) (b) (c) (d)
(e)
Figure 2.4: Semi-meanders specialize closed meanders
is equivalent to the conguration consisting of the closed curve and the
new ray (b). By homeomorphically deforming the plane, the two rays can
be separated further and aligned such that they are oriented in opposite
directions (c). Interpreting this pair of rays as a line oriented in the same
direction as the original ray, we have constructed a meandric conguration
(d).
In this presentation, the congurations swept out between the two rays
is seen to be a rainbow conguration. Moreover, beginning with a closed
meander with a rainbow conguration for its lower arch conguration, it is
possible to reverse the transformation to obtain a semi-meander by inter-
preting the line as a pair of rays originating from the center of the rainbow
conguration and deleting one of the rays. This provides a natural embed-
ding of semi-meanders in meanders and establishes the following proposition.
Proposition 2.2.2. Meandric numbers and semi-meandric numbers satisfy
the relation
M
n
M
n
. (2.2)
By representing meanders as pairs of arch congurations, we obtain a
natural representation for semi-meanders in terms of arch congurations.
Semi-meanders can be viewed as the specialization of meanders to those with
the rainbow conguration as their lower arch conguration. In addition, the
closed curve of a semi-meander of order n intersects the ray n times and
intersects the pair of rays 2n times, so the closed meander obtained by this
correspondence is also of order n. It is this relationship that motivated a
denition of order that at rst appears to be an inconsistent between the
two classes.
13
This representation also preserves winding number as an identiable pa-
rameter. When a semi-meander is presented as a closed meander with the
rainbow conguration as its lower conguration, the winding number can be
recovered as the number of arches passing over the midpoint of the upper
arch conguration.
Parallel to the denition of a meandric system, we dene a semi-meandric
system as the superposition of an arbitrary upper arch conguration with a
lower rainbow conguration.
2.2.2 Embedding Closed Meanders in Semi-Meanders
Closed meanders also have a natural embedding in semi-meanders, though
in this embedding, the order is not preserved. A semi-meander of order 2n
with winding number zero can be drawn so the ray can be extended to a
line without intersecting the closed curve. A closed meander of order n is
obtained by orienting the line consistently with the ray.
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
Figure 2.5: Closed meanders specialize semi-meanders
Figure 2.5 illustrates this interpretation: a semi-meander of order 8 with
winding number zero (left) is drawn such that the ray can be extended to
a line without intersecting the closed curve and the resulting conguration
is interpreted as a closed meander of order 4 (right) by replacing the ray by
the oriented line containing it.
In a similar manner, a closed meander can be interpreted as a semi-
meander with winding number zero by picking any point on the line that
precedes all intersections, and using it as the origin of the ray of a semi-
meandric conguration using the same closed curve. In terms of diagram-
matic representations, the construction is trivial, amounting to replacing
14
the tail of an arrow, marking the direction of a line, by a dot, indicating the
origin of a ray.
Using this interpretation, we can interpret closed meanders as a special-
ization of semi-meanders. Closed meanders are those semi-meanders with a
winding number of zero. The following proposition is an immediate conse-
quence.
Proposition 2.2.3. Meandric numbers and semi-meandric numbers satisfy
the relation
M
n
M
2n
. (2.3)
2.2.3 Bounding Meandric Numbers
We can obtain both an upper bound and a lower bound for meandric num-
bers in terms of arch congurations.
Lemma 2.2.4. The numbers M
n
and C
n
satisfy the relation
C
n
M
n
C
2
n
.
Proof. Meandric systems of order n are in bijective correspondence with
ordered pairs of arch congurations of order n, of which there are C
2
n
. So
the inequality
M
n
C
2
n
is a consequence of meanders being a subset of meandric systems of the same
order.
To establish the lower bound, and complete the proof, we show that every
arch conguration occurs as the upper conguration of some meander. We
begin with the observation that every arch conguration contains an arch
such that no arch lies between its feet. Call such an arch a minimal arch.
We employ a construction that uses one meander to generate several
closed meanders of the next higher order. Consider two arch congurations
such that the second is derived from the rst by erasing a minimal arch
15
and redrawing the resulting conguration so the feet of the arches are again
equally spaced. Figure 2.6 illustrates this derivation: beginning with (a), (b)
is obtained by deleting the minimal arch indicated with an arrow, redrawing
(b) as (c) completes the derivation.
PSfrag replacements
(a) (b) (c)
(d)
(e)
Figure 2.6: Minimal arch deletion
Starting with a meander of order n1 with the derived arch conguration
as its upper conguration, we can construct a meander of order n with the
rst arch conguration as its upper conguration. Figure 2.7 illustrates this
construction. Beginning with an arbitrary closed meander with the derived
arch conguration as its upper conguration (a), we obtain an equivalent
meander, with the arches repositioned to accommodate in situ the missing
arch (b), and locate the region of the lower conguration incident with the
feet of the missing arch (c).
PSfrag replacements
(a) (b) (c)
(d)
(e)
Figure 2.7: Deforming a meander to accomodate a missing arch
A meander of order n with the rst arch conguration as its upper con-
guration is obtained by picking any lower arch incident to this region and
stretching it to cross the line and replace the deleted arch. In general the
choice is not unique. Figure 2.8 uses an arrow to indicate the two lower
arches that can be selected, and illustrates the result of the construction in
each case.
16
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
Figure 2.8: Stretching an arch across the line
This construction can be used to produce a closed meander with any
specied arch congurations as its upper conguration. To do this, construct
a sequence of arch congurations beginning with the specied conguration
and ending with the conguration consisting of a single arch, such that
each conguration is obtained from its immediate predecessor by deleting
a minimal arch. Figure 2.9 illustrates such a sequence beginning with the
arch conguration from the previous example.
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
Figure 2.9: A sequence minimal arch deletions
The conguration with a single upper arch is uniquely completed by the
conguration with a single lower arch. Beginning with this meander use the
preceding construction to produce a meander for each arch conguration in
the sequence. Such a sequence is illustrated in Figure 2.10 with an arrow
indicating the arch that is stretched at each step.
Since this construction can be applied beginning with any arch cong-
uration to obtain a meander with the specied conguration as its upper
conguration, we conclude that, every arch conguration is the upper con-
17
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
Figure 2.10: A sequence of meanders
guration of at least one meander. This establishes the lower bound
C
n
M
n
.
Arch congurations are easily shown to be in bijective correspondence
with several other classes of combinatorial objects of known cardinality.
For completeness we provide a direct derivation of the arch conguration
numbers.
Lemma 2.2.5. The number of arch congurations of order n, C
n
is
1
n+1
_
2n
n
_
,
the n-th Catalan number.
Proof. We consider the null conguration (with no arches) to be an arch
conguration. Let / denote the class of arch congurations, and B denote
the class of arch congurations with a single arch incident with the innite
face. An arbitrary arch conguration can be decomposed as a nite sequence
of arch congurations of class B. The decomposition is reversible since a
sequence of arch congurations of class B can be concatenated to produce
an arch conguration of general type. This establishes the bijection:
/
_
i=0
B
i
which preserves the number of arches. Thus if A(x) and B(x) Q[[x]], the
ring of formal power series over Q, are the ordinary generating series for /
18
and B, respectively, then
A(x) =
i=0
B(x)
i
=
1
1 B(x)
.
But an arch conguration of general type can be obtained from one of class
B by deleting the arch incident with the innite face. Since the process
is reversible, given an arch conguration of general type we can obtain a
unique arch conguration of class B by adding an arch spanning the entire
conguration, we have the additional relation,
B(x) = xA(x),
and conclude that A(x) satises the equation
A(x) =
1
1 xA(x)
,
or equivalently
A(x) 1 = xA(x)
2
.
Letting D(x) = A(x) 1, we have
D(x) = x(1 +D(x))
2
,
and by Lagranges Implicit Function Theorem [8, Thm. 1.2.4],
D(x) =
n1
x
n
n
[
n1
](1 +)
2n
=
n1
1
n
_
2n
n 1
_
x
n
=
n1
1
n + 1
_
2n
n
_
x
n
19
is the unique solution to this equation in Q[[x]]. From this we see that
A(x) =
n0
1
n + 1
_
2n
n
_
x
n
,
and so
C
n
= [x
n
]A(x) =
1
n + 1
_
2n
n
_
,
as required.
We have an immediate corollary.
Corollary 2.2.6. The number of meandric systems of order n is
_
1
n+1
_
2n
n
_
_
2
.
We also have a rough picture of the general behaviour of the meandric
numbers. From Stirlings approximation, the asymptotic behaviour of C
n
is
known to be C
n
C
4
n
n
3/2
for a constant C. It is thus reasonable to assume
that the asymptotic behaviour of M
n
is
M
n
C
R
n
n
(2.4)
for some constants C, R, and . Indeed, it is shown in [11] that there exists
a constant A
M
such that for all A with 0 < A < A
M
, A
n
< M
n
(A
M
)
n
.
From (2.2) and (2.3), we can bound the semi-meandric numbers by M
n
M
n
M
2n
and can conclude that they also exhibit exponential growth. It
is conjectured that
M
n
C
R
n
n
(2.4
)
for some constants C, R, and . For reasons stemming from interpreta-
tions of meanders and semi-meanders in statistical mechanics, it is further
conjectured, in [5], that R = R
2
.
20
2.3 Filtering Meanders From Meandric
Systems
By considering a meandric system as a pair of arch congurations of the
same order we have determined that the number of such systems is C
2
n
.
We now give an expression, as described in [11], for the generating series for
meandric systems with respect to twice their order,
B(x) =
n0
C
2
n
x
2n
.
Proposition 2.3.1. The generating series B(x) has the expression,
B(x) =
1
4x
2
_
1 +
1
2
_
2
0
_
1 8xcos + 16x
2
d
_
. (2.5)
Proof. We work in C((x)), the eld of formal Laurent series over C, and
make use of a result due to Parseval, see for example [11].
Lemma (Parseval 1805). If f(x) =
n0
a
n
x
n
is analytic in a neighbour-
hood of zero in C, then F(x) =
n0
(a
n
x
n
)
2
is analytic in a neighbourhood
of zero and representable as
F(x) = Res
=0
f(x)f(
1
x)
.
By considering the natural embedding of Q[[x]] in C((x)), we can view
A(x), the generating series for arch congurations as the function,
A(x) =
n0
C
n
x
n
=
n0
1
n + 1
_
2n
n
_
x
n
=
1
1 4x
2x
.
which is analytic on the disc D = x C: [x[ <
1
4
. So, by the lemma, with
21
A and B taking the roles of f and F, respectively,
B(x) = Res
=0
A(x)A(
1
x)
= Res
=0
1
1 4x
2x
1
1 4
1
x
2
1
x
=
1
4x
2
Res
=0
1
(1
1 4x)(1
_
1 4
1
x)
=
1
4x
2
_
Res
=0
1
(1
1 4x) Res
=0
1
(
_
1 4
1
x)
+Res
=0
1
(
_
1 4x 4
1
x + 16x
2
)
_
=
1
4x
2
_
1 +
1
2
_
2
0
_
1 8xcos + 16x
2
d
_
.
The rst two residues are seen to be 0 and 1, respectively, while the nal
residue is obtained by integrating around the perimeter of the unit circle
with replaced by cos +i sin .
Given the relative ease with which we obtain a solution to the problem of
enumerating systems of meanders, and that a system of meanders is a multi-
component meander, one might hope to be able to exploit this knowledge to
obtain an exact solution to the number of single component, or connected
meanders. Indeed, under the appropriate combinatorial conditions, there is
a simple relation between the generating series for connected objects and the
generating series for the set of objects constructed from a nite collection of
connected objects. We consider two classes of objects for which the number
of connected objects can be obtained from the number of general objects,
and show that meandric systems do not fall into either of these classes.
For the rst class of objects, a general object is obtained from connected
objects by Cartesian product, and can be decomposed into a sequence of
connected objects. For this class, the ordinary generating series for general
objects G(x) and the ordinary generating series for connected objects C(x)
satisfy the relation
G(x) =
1
1 C(x)
. (2.6)
22
Meanders and meandric systems are not in this class of objects. Any
sequence of meanders can be concatenated along the line to form a meandric
system, but not every meandric system takes this form. Two closed curves
may be nested as in Figure 2.11 (a), intertwined as in Figure 2.11 (b), or
both as in Figure 2.11 (c), for example.
PSfrag replacements
(a) (b) (c)
(d)
(e)
Figure 2.11: Meandric systems that are not formed by concatenation
For the second class of objects, a general object is obtained from con-
nected objects by -product [8, Def. 3.2.9], and can be decomposed into a
collection of connected objects on disjoint sets of labels. In this case, G(x),
the exponential generating series for the general objects, and C(x), the ex-
ponential generating series for the connected objects, satisfy the relation
G(x) = exp(C(x)). (2.7)
Considering the base points as the labelled objects, every component of a
meandric system is a closed meander on a subset of the labels: the meandric
system Figure 2.12 (a) is the superposition of the meander Figure 2.12 (b) on
the second and forth pairs of intersections, with the meander Figure 2.12 (c)
on the rst and third pairs of intersections, for example. Meandric systems
do not, however, fall into this second class of objects.
The superposition of two copies of the meander in Figure 2.13 (a) using
the same partition as in Figure 2.12 (a) produces a conguration, Figure 2.13
(b), that is not a meandric system; the upper conguration is not planar.
The allowable congurations for one component are dependent on the form
of the other components, and this violates the conditions under which (2.7)
holds.
23
PSfrag replacements
(a) (b) (c)
(d)
(e)
Figure 2.12: A meandric system is a collection of meanders
PSfrag replacements
(a) (b)
(c)
(d)
(e)
Figure 2.13: A collection of meanders that is not a meandric system
There does not appear to be an elementary combinatorial description
of the operation of constructing a meandric system from meanders. As a
result, there is no known way to obtain the meandric numbers for meandric
system numbers. In particular, the generating series for meanders cannot
be related to the generating series for meandric systems by either (2.6) or
(2.7).
2.4 Automorphisms of Meanders
Automorphisms on the class of meanders can be used constructively in al-
gorithms for constructing meanders of a given order. They can also be used
to simplify an exhaustive enumeration of meanders by eliminating redun-
dancy. Given an automorphism on the class of meanders, it is possible to
encode the entire class by a list that includes only a single element of each
orbit under the action of the automorphism. If the automorphism is easily
computable, then analysis on the entire class can be carried out by iterating
through the list of orbit representatives and expanding each orbit in turn.
For certain automorphisms, it is possible to derive congruences satised by
the meandric numbers by considering the size of orbits under the action of
the automorphism.
24
Two classes of automorphisms are particularly useful: those dened by
rigid transformations of the plane, and those dened by cyclic shifts of the
intersection points. These are now considered in greater detail.
2.4.1 Rigid Transformations
The most obvious class of automorphisms consists of rigid transformations
of the plane. In particular, the closed curve of a meander can be reected in
() or along the line (), or reected through a point ( = = )
i.e. rotated through 180
and experi-
mentally determine values for the constants C, R, and . This approach
requires the knowledge of a large number of terms. The most eective way
to obtain these numbers has been to list all the meanders of the required
form and count them.
One method for enumerating a class of objects is to construct a forest,
such that the vertices at depth n are exactly the objects of order n. We
consider only the case where there is a single object of order one, and the
forest is a tree. In order to construct such a tree, we need only have a method
for identifying incidence. This can be accomplished by dening a function
mapping objects of order n > 1 to objects of order n 1. By means of such
a function, a tree can be formed by taking the unique object of order one
as the root, and dening the children of a given object to be its preimages
under the function.
If the preimages are easily computable, then it is possible to enumerate
all objects of a given order by using standard tree traversing algorithms.
This traversal can be done with an amount of memory polynomial in the
desired order, and in time proportional to the number of objects of that
order.
30
2.5.1 A Tree of Semi-Meanders
To construct such a function for the class of semi-meanders, we consider
semi-meanders as closed meanders with a lower rainbow conguration. In
this presentation, dene the image of a semi-meander to be the semi-meander
obtained by breaking the outer arch of the lower conguration, contracting
the two free ends across the line, and rejoining them to create a new arch
in the upper conguration. This is illustrated in Figure 2.21.
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
Figure 2.21: A function on semi-meanders
This function satises the condition of having an easily computable
preimage. A semi-meander has one preimage for each arch of the upper
conguration that is incident with the innite face. For a given arch, the
corresponding preimage is obtained by breaking the arch, stretching the two
free ends, one around each end of the conguration, and rejoining them to
create a new outer arch in the lower conguration. Since every semi-meander
of order 2 has at least 2 upper arches incident to the innite face, we have
the immediate consequence that, for n 2,
M
n
2M
n+1
.
Figure 2.22 illustrates the tree of semi-meanders obtained using this
function. Notice that the left half of this tree is the mirror image of the right
half. Mirroring a semi-meander in this presentation corresponds to mirror-
ing a semi-meander in the standard presentation across the line. Using this
tree, it is thus possible, by mirroring every semi-meander, to enumerate all
semi-meanders while traversing only the branch corresponding one of the
semi-meanders of order 3.
31
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
Figure 2.22: The tree of semi-meanders of order 5
This method can be extended to enumerate all systems of semi-meanders.
In this case, each semi-meandric system has, as its children, all of its preim-
ages under the function and the system obtained by adding a circle sur-
rounding the original system.
2.5.2 A Tree of Meanders
A similar construction can be used to generate all closed meanders. To dene
the incidence function, we use the observation that every closed meander has
at least one minimal arch in its upper conguration.
To nd the image of a closed meander, apply cyclic shifts, if necessary,
until a minimal arch appears as the leftmost arch of the upper conguration.
For all four meanders in Figure 2.23, the rightmost meander is obtained by
this step. The image of the original meander is obtained by pushing this
minimal arch across the line. Figure 2.24 illustrates this nal step.
As with semi-meanders, the preimage is easily computed. A meander
has one primary child per arch of the lower conguration incident with the
innite face. For a given arch, the corresponding child is found by pulling the
arch across the line to the left of the existing conguration. The remaining
32
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
Figure 2.23: Shifting a meander to obtain a minimal arch
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
Figure 2.24: The image of a meander
children are found by applying reverse cyclic shifts to the primary children
until the upper conguration has a single arch incident with the innite
face, a subsequent shift would result in a dierent minimal arch. Since every
meander has at least one child, we conclude that the sequence of meandric
numbers is monotonically increasing.
33
Chapter 3
The Symmetric Group
Meanders have a natural representation as permutations. This chapter uses
this representation to present a combinatorial construction for meanders.
The meandric and semi-meandric numbers can then be expressed in terms
of characters of the symmetric group. Throughout this chapter, elements of
S
n
are respresented by their disjoint cycles representation.
3.1 Representing Meanders As Permutations
Given a closed meander of order n, we represent it such that the line is
horizontal and oriented from left to right, the closed curve is oriented such
that at the rst intersection between the curves it is directed from bottom
to top. The intersections between the two curves have a natural labelling by
the integers 1, 2, . . . , 2n, dened by their ordering along the line. These
labels also have a cyclic order dened by their order along the closed curve.
This cyclic order, when interpreted as a a full cycle in S
2n
, is referred to
as a meandric permutation. There is a one-to-one correspondence between
meanders and meandric permutations. Figure 3.1 shows the meandric per-
mutation (1 10 9 4 3 2 5 8 7 6).
Interest in meandric permutations predates modern interest in the enu-
merative theory of meanders. They were discussed for instance by Rosen-
34
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
1 2 3 4 5 6 7 8 9 10
Figure 3.1: The meandric permutation (1 10 9 4 3 2 5 8 7 6)
stiehl in [12] as planar permutations. These permutations occur in the anal-
ysis of geographical data, and have the property that they can be sorted in
linear time [12] . Enumerative information about meanders can be used to
bound the performance of algorithms dealing with this sorting.
It is an immediate consequence of the denition, that every meandric
permutation is in the conjugacy class (
(2n)
of S
2n
, where n is the order
of the corresponding closed meander. A natural question is how to deter-
mine whether a given permutation in S
2n
of cycle type (2n) is a meandric
permutation. We have the following necessary condition.
Proposition 3.1.1. If S
2n
is a meandric permutation, then
2
is a
permutation of cycle type [n
2
], where one cycle is on the odd symbols and
the other is on the even symbols.
Proof. We consider the upper and lower arch congurations of the meander
corresponding to . Every arch spans an integral number of smaller arches,
each of which has an even number of base points. Thus every arch has one
foot on an even symbol and the other on an odd symbol.
Following the closed curve involves steps that alternate between upper
and lower arches, so as a consequence, every meandric permutations alter-
nates between even and odd symbols. The result follows.
This condition does not characterize meandric permutations. Consider
the permutation (1 4 3 6 5 2), with square (1 3 5)(2 4 6) which is not a
35
meandric permutation. Attempting to draw the meander corresponding to
(1 4 3 6 5 2), one nds that the closed curve must be self intersecting in the
upper half plane.
The diculty in identifying meandric permutations is a consequence of
the fact that planarity is a property of the upper and lower congurations
rather than the meander as a whole. It is possible to recover the upper and
lower congurations from a meandric permutation, as will be described later
in this chapter, but there is no apparent way to do so algebraically.
3.1.1 Automorphisms of Meandric Permutations
The automorphisms described in Section 2.4 all have natural interpretations
in terms of meandric permutations.
Reecting a meander in the line reverses the orientation of the closed
curve. The corresponding meandric permutation is thus obtained by read-
ing the meandric permutation of the original meander in the reverse order.
This is simply the inverse of the permutation obtained from the original
meander. We conclude that the inverse of a meandric permutation is a
meandric permutation.
To interpret reection along the line and cyclic shifts in terms of permu-
tations, we introduce two permutations,
=
n
= (1 2n)(2 2n1) (n n+1) and
=
n
= (1 2 . . . 2n),
(3.1)
corresponding respectively to the actions of reection along the line and
right cyclic shift on the labels.
Reection along the line reverses the order of the labels and reverses the
orientation of the closed curve. Similarly, a cyclic shift, shifts the labels and
reverses the orientation of the curve. So, for a meandric permutation,
the images of under reection along the line and right cyclic shift are,
respectively,
1
1
and
1
1
. Since rotation of the closed curve is the
36
composition of reection across and along the line, the image of a rotation
is
1
. The following proposition is a consequence of reection and cyclic
shift being automorphisms.
Proposition 3.1.2. If S
2n
is a meandric permutation, then so are
1
,
1
, and
1
; the class of meandric permutations is closed under
inverse, conjugation by , and conjugation by .
That the automorphisms from Section 2.4 are easily accessible in this
encoding, is a testament to how natural the encoding is. It also suggests that
the encoding is unlikely to make any additional structure more accessible.
3.2 Arch Congurations as Permutations
There is a second natural encoding of meanders in terms of the symmet-
ric group. In order to deal with planarity, we use this alternate encoding.
Instead of directly encoding meanders as permutations, we encode each me-
ander as an ordered pair of arch congurations. Each arch conguration is
then represented as a permutation. Using this encoding, it is possible to
describe compactly which pairs of permutations correspond to meanders.
The permutation representation of arch congurations is a natural par-
allel to the representation of meanders as permutations. Each arch is a
transposition on its endpoints, and the entire conguration is the product
of these disjoint transpositions. An arch conguration of order n is the
product of n disjoint transpositions in S
2n
. As with meanders, the base
points of the arches are labelled according to the orientation of the line.
Figure 3.2 gives the arch conguration corresponding to the permutation
(1 10)(2 3)(4 9)(5 8)(6 7).
There are two steps in identifying which pairs of permutations in the
conjugacy class (
(2
n
)
of S
2n
correspond to closed meanders. We must rst
identify which elements of (
(2
n
)
correspond to arch congurations. Then,
given two such elements, we must determine whether the meandric system
obtained by interpreting one as an upper arch conguration and the other
37
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
1 2 3 4 5 6 7 8 9 10
Figure 3.2: The arch conguration (1 10)(2 3)(4 9)(5 8)(6 7)
as a lower arch conguration is connected. These steps can be carried out
algebraically. We deal with each step in turn.
3.2.1 Elements of C
(2
n
)
That Are Arch Congurations
Identifying arch congurations among elements of (
(2
n
)
is accomplished by
interpreting them permutations as graphs, a class of objects for which pla-
narity is more easily described. A graph corresponding to (
(2
n
)
is
constructed on 2n vertices labelled 1, 2, . . . , 2n by creating an edge be-
tween every pair of vertices adjacent in the cyclic order (1 2 . . . 2n), and
adding an edge between every pair of vertices whose labels are in the same
transposition of .
Such a graph is seen to have 3n edges, n corresponding to transpositions,
and 2n corresponding to consecutive pairs of vertices. For the permutation
to be an arch conguration, its graph must be planar. Since the graph is
planar and has 2n vertices it has n+2 faces, by Eulers theorem. Figure 3.3
shows the graph of the arch conguration from Figure 3.2 together with its
face boundaries.
We consider the cycles that bound the faces of the graph of an arch con-
guration. A single face, corresponding to the lower half-plane, is bounded
entirely by edges dened by the cyclic order of the labels, while the bound-
ary cycles of the remaining n + 1 faces are seen to alternate between edges
dened by transpositions and edges dened by the cyclic order of the base
38
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
1
2
3
4
5
6
7
8
9
10
Figure 3.3: The graph of = (1 10)(2 3)(4 9)(5 8)(6 7) with its face boundaries
points. These are the cycles of . Continuing the above example, =
(1 3 9)(2)(4 8)(5 7)(6)(10) which consists of 6 cycles.
Letting () denote the number of cycles in the disjoint cycle represen-
tation of the permutations , we have the following characterization of the
permutations that correspond to arch congurations.
Lemma 3.2.1. The permutations in S
2n
that correspond to arch congu-
rations form the set
(
(2
n
)
: () = n + 1.
Proof. If (
(2
n
)
corresponds to an arch conguration, then by the above
observation consists of n+1 cycles. Conversely, if does not correspond
to an arch conguration, then at least two arches cross and consists of
at most n cycles.
As expected, this class of permutations is closed under conjugation by
and , operations that correspond respectively to reection along the
line and cyclic shift of the base points. We also note, that for to be a
permutation with n+1 cycles on 2n elements, at least two cycles have length
1. These cycles correspond to the minimal arches of the arch conguration.
3.2.2 Completing the Characterization
Pairs of arch congurations of order n that form meandric systems with a
single connected component can also be characterized algebraically. There
39
is a single closed curve if and only if it is possible to walk through all the
labels using a closed path alternating between arches from the upper cong-
uration and arches from the lower conguration. Such a walk is a meandric
permutation, so by taking it two steps at a time, by Proposition 3.1.1, we
obtain two cycles of length n. We obtain the following characterization of
which meandric systems are meanders.
Lemma 3.2.2. If (
1
,
2
) is an ordered pair of transposition representations
of arch congurations of order n, then the meandric system for which
1
represents the upper conguration and
2
represents the lower conguration
is a meander if and only if
1
2
(
(n
2
)
.
Proof. The cycles of
1
2
correspond to walks through the labels taken
two steps at a time along paths that alternate between arches described
by transpositions of
1
and arches described by transpositions of
2
. So if
(
1
,
2
) corresponds to a meander, then by the above observation each cycle
of this type is of length n, and
1
2
(
(n
2
)
.
Conversely, if
1
and
2
are both transposition representations of arch
congurations of order n, then every transposition of each contains an odd
label and an even label. So if
1
2
(
(n
2
)
, then each cycle consists only of
labels with the same parity. A walk through all the labels can thus be formed
by interleaving the two cycles and so (
1
,
2
) corresponds to a meander.
Noting that (
1
,
2
) represents a meander only if
1
2
is the square of
the corresponding meandric permutation, we can see how to extract the
upper and lower congurations from a meandric permutation. Consecutive
pairs of symbols in the meandric permutation dene the arches of the upper
and lower congurations. The upper arch conguration is read by taking
pairs of symbols starting with 1, while the lower conguration is read by
taking symbols starting at either of the labels adjacent to 1.
Example 3.2.3. For the meandric permutation = (1 10 9 4 3 2 5 8 7 6) of
order 5, the upper conguration is
1
= (1 10)(9 4)(3 2)(5 8)(7 6), while the
40
lower conguration is
2
= (10 9)(4 3)(2 5)(8 7)(6 1). Verifying the proper-
ties,
1
= (1 3 9)(2)(4 8)(5 7)(6)(10) and
2
= (1 5)(2 4)(3)(6 8 10)(7)(8).
So (
1
) = (
2
) = 6, and
2
=
1
2
= (1 9 3 5 7)(2 8 6 10 4) (
(n
2
)
.
We can now complete our combinatorial characterization of meanders
with the following immediate consequence of Lemma 3.2.1 and Lemma 3.2.2.
Corollary 3.2.4. The class of meanders of order n is in bijective corre-
spondence with the set
(
1
,
2
) (
(2
n
)
(
(2
n
)
: (
1
) = (
2
) = n + 1,
1
2
(
(n
2
)
.
Taking semi-meanders as the restriction of meanders to those with the
rainbow conguration as a lower conguration, and noting that the rainbow
conguration has the permutation representation , we obtain Corollary 3.2.5
as a specialization of Corollary 3.2.4 and achieve an algebraic characteriza-
tion of semi-meanders.
Corollary 3.2.5. The class of semi-meanders of order n is in bijective
correspondence with the set
(
(2
n
)
: () = n + 1, (
(n
2
)
.
The bijections described in Corollary 3.2.4 and Corollary 3.2.5 give us
combinatorial expressions for meandric and semi-meandric numbers. Letting
(
(n+1)
denote the set of all conjugacy classes of S
2n
with n + 1 cycles, we
obtain the expressions:
M
n
=
1
,
2
C
(n+1)
1
,
2
C
(2
n
)
[
1
],
1
[
2
],
2
[
1
2
],(n
2
)
(3.2)
M
n
=
C
(n+1)
C
(2
n
)
[],
[],(n
2
)
, (3.3)
where
a,b
is one if a and b are the same conjugacy class and zero otherwise.
In this form, it looks as though it might be possible to determine the mean-
dric numbers by working entirely within the center of the symmetric group,
41
since, as a function of , the expression for M
n
is constant on the conjugacy
class (2
n
). The same is not possible for the semi-meandric numbers, since
the expression for M
n
involves specic elements of two conjugacy classes.
3.3 Expression in Terms of Characters
For the remainder of the chapter, we will assume a familiarity with char-
acter theory. The reader is referred to [2] for a treatment of this subject.
Characters of the symmetric group are class functions, and can be used to
obtain an alternative expression for (3.2) and (3.3) that avoids the use of
the -function.
The construction of characters involves representation theory and is of
little interest to the present discussion. For our purposes, it suces to note
that the irreducible characters of a nite group G are functions
(i)
: G C.
They are class functions, that is they are constant on conjugacy classes of G,
and are naturally indexed by the conjugacy classes of the group. In addition,
for g in G and a character, (g) = (g
1
).
We introduce some notation for a nite group G with k conjugacy classes
(
1
, (
2
, . . . , (
k
of sizes h
(i)
= [C
i
[, i = 1, 2, . . . , k. We let
(i)
be the irreducible
character indexed by conjugacy class (
i
and let
(i)
j
denote the evaluation
of
(i)
at any g (
j
. The irreducible characters (properly characters of
irreducible representations) satisfy two orthogonality relations:
(1)
1
[G[
k
i=1
h
(i)
(p)
i
(q)
i
=
p,q
for 1 p, q k
(2)
k
i=1
(i)
p
(i)
q
=
[G[
h
(p)
p,q
for 1 p, q k.
We use the second orthogonality relation for the group G = S
2n
to
obtain an expression for meandric numbers and semi-meandric numbers in
terms of characters of irreducible representations of the symmetric group.
42
Theorem 3.3.1. If
(1)
,
(2)
, . . . ,
(k)
are the characters of the irreducible
representations of the group S
2n
, and and are as in (3.1), then
M
n
=
1
,
2
C
(n+1)
[(2
n
)[
2
[
1
[ [
2
[ [(n
2
)[
((2n)!)
5
1
,
2
S
2n
k
f,g,h,i,j=1
(f)
(
1
)
(f)
(2
n
)
(g)
(
2
)
(g)
(2
n
)
(h)
(
1
)
(h)
(
1
)
(i)
(
2
)
(i)
(
2
)
(j)
(
1
2
)
(j)
(n
2
)
M
n
=
C
(n+1)
[(2
n
)[ [[ [(n
2
)[
((2n)!)
3
S
2n
k
h,i,j=1
(h)
()
(h)
(2
n
)
(i)
()
(i)
()
(j)
()
(j)
(n
2
)
Proof. For S
2n
the conjugacy classes and characters are naturally indexed
by partitions of 2n. For v n, the size of the conjugacy class (1
v
1
2
v2
)
with v
i
cycles of length i for each i is
[(1
v
1
2
v2
)[ =
(2n)!
i
i
v
i
v
i
!
.
In addition, for every g in S
2n
, g and g
1
are in the same conjugacy class.
Thus for any character ,
(g) = (g
1
) = (g)
so is real valued and in the case of S
2n
we have the relation
[],[]
=
[[][
(2n)!
k
i=1
(i)
()
(i)
() (3.4)
Replacing the sum over (
(2
n
)
by a sum over S
2n
and substituting (3.4) into
(3.2) and (3.3) we obtain the desired expression.
Theorem 3.3.1 provides an explicit way to compute meandric and semi-
meandric numbers but is of little practical use. To determine, for instance,
that there is a unique semi-meander of order 2 would involve summing over
43
6000 terms, since for S
4
there are two conjugacy classes with 3 cycles, 24
elements, and 5 conjugacy classes.
In fact the expression in terms of characters is a direct translation of the
algebraic characterization and does not immediately provide any additional
insight into the structure of the problem. It preserves the parts of the com-
binatorial construction of semi-meanders as identiable pieces. Considering
the expression for M
n
, for example, we begin with a sum over all permuta-
tions in S
2n
. Summing over h restricts the sum to products of n disjoint
transpositions, while summing over i further restricts the sum to include
only those elements of (
(2
n
)
that correspond to arch congurations. Taking
the nal sum over j leaves only semi-meandric systems that are connected.
The utility of the expression comes from the fact that it provides a bridge
from combinatorics to representation theory. It may be possible to expand
the characters in terms of symmetric functions, or even nd an alternate
expansion for the -function, although, to date, no signicant advances have
been made in this direction.
44
Chapter 4
The Matrix Model
The method of matrix integrals can be used to transform the meander prob-
lem into an analytic question. By selecting an appropriate Gaussian measure
on the space of Hermitian matrices, the average value of a matrix expression
can be discretised so that the non-vanishing terms correspond to a class of
decorated ribbon graphs from which meanders can be isolated. The con-
struction used is from [11].
4.1 Meanders as Ribbon Graphs
We begin with an informal denition of a ribbon graph.
Denition 4.1.1. A ribbon graph is an object obtained from a graph, by
embedding it in a locally orientable surface, thickening the edges into ribbons,
and deleting the faces. The embedding imposes a cyclic ordering of the edges
around each vertex.
A ribbon graph preserves the face boundaries of its embedding. Con-
sequently, the original graph and embedding can be recovered by stitching
an open disc along each face boundary, and contracting each ribbon into an
edge.
45
By considering the one-point compactication of the plane, a meander
has a natural interpretation as a graph embedded in the sphere: the ver-
tices are intersections between the curves, and the edges are derived from
segments of the two curves, with an additional edge joining the extreme
vertices along the line. These edges are decorated with two colours, to dis-
tinguish those associated with the closed curve from those associated with
the line. The encoding is made reversible by rooting the graph. The rst
vertex along the line is the root vertex, and the edge joining this vertex to the
second vertex along the line is the root edge. We produce a decorated ribbon
graph by thickening the edges, and preserving their colouring. Figure 4.1
illustrates this encoding. For the purpose of enumeration, we dene a
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
Figure 4.1: Encoding a closed meander as a ribbon graph
class of ribbon graphs that generalizes the ribbon graphs that correspond to
meanders.
Denition 4.1.2. The class is the class of oriented 4-regular ribbon
graphs on labelled vertices, with edges divided into two classes, such that
around every vertex the edges alternate between the two classes. For each
vertex, one of the edges of the second class is designated as up.
The graph induced on a ribbon graph of class by taking only the
edges of one class is 2-regular. Thus the two classes of edges induce two
collections of disjoint cycles. We have the following lemma for identifying
meanders within the class .
Lemma 4.1.3. Meanders of order n are in 4n to (2n)! 2
2n
correspondence
with graphs in on 2n vertices that are genus zero and have exactly two
cycles induced by the partitioning of the edges, one of each class.
46
Proof. Beginning with a ribbon graph we obtain an unlabelled graph by
discarding the labels and orientations on the vertices. This results in a
(2n)! 2
2n
to one correspondence. A meander is obtained from such an
unlabelled graph by designating a root vertex and picking one of the edges
of the second class incident to it as the root edge. Picking the root vertex
and edge accounts for the additional factor of 2n 2 = 4n.
Guided by Lemma 4.1.3, we will use
m
to denote the elements of
with m vertices, and consider the generating series
Z(s, q, N) =
1
N
2
m1
(1)
m
m!
s
m
N
m
GRm
N
p(G)
q
r(G)
, (4.1)
where p(G), and r(G) denote, respectively, the number of faces of G, and the
number of cycles induced by the edge colouring, in G. By further decorating
the elements of , assigning each face a label from the set 1, 2, . . . , N, and
assigning each of the cycles induced by the partitioning of the edges a label
from the set 1, 2, . . . , q, we obtain the term N
p(G)
q
r(G)
as the number of
distinct decorations of the map G.
Associated with each vertex are the labels of the four faces with which
it is incident, labelled i
1
, i
2
, i
3
, and i
4
in cyclic order, and two labels on the
edges, k for the edges of the rst class, l for the edges of the second class.
Figure 4.2 gives such a labelled neighbourhood.
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
i
1
i
2
i
3
i
4
k
l
Figure 4.2: The neighbourhood of a vertex
We assign to each half-edge a variable carrying its labelling information,
and assign to each vertex the product of these variables. In clockwise order
47
starting with the upper branch, the branches of Figure 4.2 are assigned the
variables h
(k)
i
1
i
2
, g
(l)
i
2
i
3
, h
(k)
i
3
i
4
, and g
(l)
i
4
i
1
, and the vertex is assigned the monomial
h
(k)
i
1
i
2
g
(l)
i
2
i
3
h
(k)
i
3
i
4
g
(l)
i
4
i
1
,
where the hs and gs are triply indexed variables.
Proposition 4.1.4. If, for 1 k q, we let H
k
denote the N N ma-
trix
_
h
(k)
ij
_
and G
k
denote the N N matrix
_
g
(k)
ij
_
, then the monomials
associated with all possible vertex neighbourhoods are enumerated by
tr
q
k,l=1
(H
k
G
l
)
2
.
Proof. The sum of the monomials associated with all possible labellings is
q
k,l=1
N
i
1
,i
2
,i
3
,i
4
=1
h
(k)
i
1
i
2
g
(l)
i
2
i
3
h
(k)
i
3
i
4
g
(l)
i
4
i
1
= tr
q
k,l=1
H
k
G
l
H
k
G
l
Our strategy is to enumerate all possible decorated ribbon graphs of
class on m vertices by picking out all the graphs with a given m-tuple of
vertex neighbourhoods, for all possible m-tuples. Given a collection of vertex
neighbourhoods, we can describe a ribbon graph by specifying a coupling
(a complete matching) of the branches of the vertex neighbourhoods into
the 2m pairs of branches that are connected to form edges. Such a coupling
must be consistent with the labelling, and not introduce any twists into the
ribbons, as depicted in Figure 4.3.
Thus, to enumerate all valid maps for a given m-tuple of vertex neigh-
bourhoods, we need to determine the number of consistent branch couplings
for that m-tuple. In terms of the monomials, a coupling is consistent if every
variable h
(k)
ij
is paired with h
(k)
ji
and every variable g
(k)
ij
is paired with g
(k)
ji
.
To deal with these pairings, we use some results from probability theory
about Gaussian measures.
48
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
i
1
i
1
i
2
i
2
k
k
Figure 4.3: A coupling
4.2 Gaussian Measures
If B is a positive denite matrix giving a quadratic form on R
n
, then the
measure
d(x) = (2)
n/2
(det B)
1/2
exp
_
1
2
x
T
Bx
_
dx,
is the Gaussian measure on R
n
associated with B, where x is an n component
column vector, and dx is the Lebesgue measure on R
n
. The measure d is
a probability measure, that is, the measure of the entire space is 1. For a
function f we use the notation
f) =
_
R
n
f(x)d(x)
to denote the average value of f with respect to the measure d. We note
that, since d(x) = d(x) for all x in R
n
, the average value of any odd
degree monomial with respect to this measure is zero. The average value of
a degree two monomial is given by the following lemma.
Lemma 4.2.1. If x
1
, x
2
, . . . , x
n
are the coordinate functions on R
n
, with
respect to the measure d associated with the matrix B, then
x
i
x
j
) = a
ij
,
for all 1 i, j n, where A = (a
ij
) = B
1
.
49
A second lemma allows us to calculate the average value of an arbitrary
monomial.
Lemma 4.2.2 (Wicks Formula). If f
1
, f
2
, . . . , f
2m
are linear functions
of x, not necessarily distinct, then
f
1
f
2
f
2m
) =
f
p
1
f
q
1
) f
p
2
f
q
2
) f
pm
f
qm
) ,
where the sum is taken over the (2m 1)!! couplings of 1, 2, . . . , 2m into
pairs p
i
, q
i
. Such a partition is referred to as a Wick coupling.
By the linearity of the integral operator, ) is a linear operator, so these
lemmas are sucient for evaluating the average value of any polynomial.
We now consider a specic Gaussian measure on the space H
N
of NN
Hermitian matrices. Matrices in this space can be coordinatized by N
2
real
numbers: for the matrix H = (h
ij
), we use the coordinates x
ij
= Re h
ij
for
i j, and y
ij
= Imh
ij
for i < j. In these coordinates, H
N
is isomorphic to
R
N
2
. Note that for H H
N
,
tr H
2
=
N
i,j=1
h
ij
h
ji
=
N
i,j=1
h
ij
h
ij
=
N
i=1
x
2
ii
+ 2
i<j
(x
2
ij
+y
2
ij
)
is a positive denite quadratic form on H
N
. With respect to the coordi-
nates x
ij
and y
ij
, the matrix for the quadratic form is diagonal, with N 1s
corresponding to the coordinates x
ii
, and N
2
N 2s corresponding to the
coordinates x
ij
and y
ij
with i ,= j. It thus has determinant 2
N
2
N
. We can
construct the associated Gaussian measure d(H) on H
N
as
d(H) = (2)
N
2
/2
2
(N
2
N)/2
exp
_
1
2
tr H
2
_
dv(H),
where
dv(H) =
N
i=1
dx
ii
1i<jN
dx
ij
dy
ij
is the usual Lebesgue measure on the space. We view the matrix entries h
ij
as linear functions of the coordinates x
ij
and y
ij
, and obtain the following.
50
Lemma 4.2.3. With respect to the measure d associated with the quadratic
form tr H
2
,
h
ij
h
kl
) =
i,l
j,k
.
Proof. The inverse of the matrix of the quadratic form is diagonal with N
1s and N
2
N
1
2
s. So by Lemma 4.2.1 we have,
h
ii
h
ii
) =
x
2
ii
_
= 1
h
ji
h
ij
) = h
ij
h
ji
) =
h
ij
h
ij
_
=
x
2
ij
+y
2
ij
_
=
1
2
+
1
2
= 1 when i < j.
In the remaining cases, when (i, j) ,= (l, k), we get no contribution from
diagonal terms and the average is zero.
To enumerate decorated elements of we extend the measure on H
N
to a
measure on the space (H
N
)
2q
of 2q-tuples (H
1
, H
2
, . . . , H
q
, G
1
, G
2
, . . . , G
q
)
of N N Hermitian matrices. As a corollary to Lemma 4.2.3 we have the
following.
Corollary 4.2.4. In the space (H
N
)
2q
where H
k
=
_
h
(k)
ij
_
and G
k
=
_
g
(k)
ij
_
for 1 k q, the following relations hold with respect to the Gaussian
measure described above:
_
h
(k
1
)
i
1
j
1
h
(k
2
)
i
2
j
2
_
=
k
1
,k
2
i
1
,j
2
i
2
,j
1
,
_
g
(k
1
)
i
1
j
1
g
(k
2
)
i
2
j
2
_
=
k
1
,k
2
i
1
,j
2
i
2
,j
1
, and
_
h
(k
1
)
i
1
j
1
g
(k
2
)
i
2
j
2
_
= 0.
Interpreting the variables as labels on branches of vertex neighbour-
hoods, as in Section 4.1, we conclude that for branches labelled a and b,
ab) =
_
1 if the branches can be paired consistently
0 if the branches cannot be paired.
Armed with Wicks Formula, we can give a combinatorial interpretation to
51
the expression
_
_
tr
q
k,l=1
H
k
G
l
H
k
G
l
_
m
_
.
By the linearity of ), this expression is a sum over all Wick couplings of
the variables involved in all possible m-tuples of vertex neighbourhoods.
By the above observation, a coupling contributes 1 to the sum if it can be
interpreted as a decorated element of and 0 otherwise. We conclude that
_
_
tr
q
k,l=1
H
k
G
l
H
k
G
l
_
m
_
=
GRm
N
p(G)
q
r(G)
. (4.2)
By substituting this expression into (4.1), we obtain the expression
Z(s, q, N) =
1
N
2
m0
(1)
m
m!
s
m
N
m
_
_
tr
q
k,l=1
H
k
G
l
H
k
G
l
_
m
_
=
1
N
2
_
exp
_
s
N
tr
q
k,l=1
H
k
G
l
H
k
G
l
_
_
=
1
N
2
_
(H
N
)
2q
exp
_
s
N
tr
q
k,l=1
H
k
G
l
H
k
G
l
_
q
k,l=1
d(H
k
)d(G
l
).
(4.3)
Notice, that in this expression, N and q are not indeterminates. The expres-
sion holds only positive integers N and q. Using the expression to evaluate
Z(s, q, N) at dierent values of q or N requires evaluating dierent integrals,
but,
P
m
(q, N) =
GRm
N
p(G)
q
r(G)
(4.4)
is a polynomial of bounded degree in q and N, and can, in principle, be
determined by interpolation.
52
4.3 Recovering Meanders
Expression (4.3) describes a generating series for with respect to number of
vertices, faces, and induced cycles. A generating series for elements of with
exactly two induced cycles can be obtained from Z(s, q, N) by considering
the coecient of q
2
,
Y (s, N) = [q
2
]Z(s, q, N).
In Y (s, N) a graph G with m vertices and p faces is weighted by a factor of
(1)
m
s
m
/(m! N
m+2p
).
We further specialize to planar graphs by considering the Euler charac-
teristic of the graphs under consideration. Since a 4-regular graph on m ver-
tices has 2m edges, such a graph has Euler characteristic m2m+p = pm.
Now a graph of genus g has Euler characteristic 2 2g, so m+ 2 p = 2g.
Thus we obtain
Y
0
(s) = lim
N
Y (s, N)
as the generating series for genus zero elements of with exactly two in-
duced cycles. In Y
0
(s) a graph G with m vertices is weighted by a factor of
1
m!
(1)
m
s
m
. This yields the main theorem for the chapter.
Theorem 4.3.1. The meandric numbers can be obtained from Y
0
(s) by the
relation
2
2n
4n
M
n
= [s
2n
]Y
0
(s).
Proof. This is an immediate consequence of the preceding observations and
Lemma 4.1.3.
Corollary 4.3.2. Where M
n
(s) is the ordinary generating series for mean-
ders with respect to their order
2sY
0
(s) = M(4s
2
).
The construction discussed in this chapter does not provide a nal so-
lution to the meander problem. In particular, a complete solution would
53
require a method to evaluate the integral in (4.3), a task for which little
progress has been made to date. Several techniques exist for evaluating
matrix integrals, but none of them are directly applicable in this case.
In general, matrix integrals are evaluated by a change of coordinates,
eectively diagonalizing the matrices involved and working over R
n
. The
diculty in this case comes from the fact that the matrices H
k
and G
l
are
not simultaneously diagonalizable. Unlike the expression tr H
4
, which is
used in enumerating 4-regular maps with no added structure, the expres-
sion tr H
k
G
l
H
k
G
l
is not invariant under unitary transformations on H
k
.
The obvious change of coordinates H (U, ), where U is unitary, is
diagonal, and H = UU
1
, is not applicable.
4.4 Another Matrix Model
The integral from (4.3) is not unique in its utility in enumerating meanders.
It is possible that some similar integral may carry the same enumerative
information but be more susceptible to evaluation.
In particular, it is worth mentioning the matrix integral used by Di Fran-
cesco, Golinelli, and Guitter, in [5] and [7]. They consider a dierent Gaus-
sian measure, one normalized such that h
ij
h
kl
) =
1
N
i,l
j,k
, and consider an
expression with an extra parameter: instead of working over (H
N
)
2q
, they
dene an integral over (H
N
)
q
1
+q
2
, separately controlling the number of H
and G matrices.
The result is an expression that records separately the number of cycles
of each class. A restriction to connected graphs is obtained by taking a
logarithm. Interpreting the resulting expression as a polynomial in q
2
, and
taking the limit as q
2
approaches zero, further restricts the expression to
graphs with only a single induced cycle of the second class. The genus zero
case is recovered as before. This approach has the advantage that it has the
potential to be used for determining the number of meandric systems with
respect to both order and number of components, since the number of cycles
of the second class is recorded.
54
To date, the integral obtained by Di Francesco et al. has been no more
susceptible to existing evaluation techniques than the integral from (4.3),
but, in [7], Di Francesco et al. note that it is suggestive of a two-dimensional
conformal eld theory, and using this similarity, and universality principles
from statistical mechanics, they conjecture the exact values
=
29
12
_
29 +
5
_
and = 1 +
11
24
_
29 +
5
_
in the asymptotic approximations to M
n
and M
n
(2.4) and (2.4
). Numerical
estimates of and , obtained by Jensen and Guttmann in [10] through the
method of dierential approximants, disagree with the conjectured values.
55
Chapter 5
The Temperley-Lieb Algebra
In this chapter, we shift attention to meandric systems (recall Denition 2.2.1).
Upper and lower arch congurations are encoded as strand diagrams, which
are in turn interpreted as elements of the Temperley-Lieb algebra. Under
this encoding, the number of components of a meandric system can be re-
covered by evaluating a bilinear form on its upper and lower congurations.
Choosing an appropriate basis for the algebra, allows us to write the mean-
dric polynomials in terms of the Gram matrix of this bilinear form.
5.1 Strand Diagrams
Before dening the Temperley-Lieb algebra, we introduce strand diagrams,
which can be used to pictorially represent the algebra. Strand diagrams
provide a combinatorial presentation of the algebra and a vehicle for linking
the algebra to the meander problem.
Denition 5.1.1. A strand diagram of order n, is a conguration of n
pairwise non-intersecting curves, called strands, in an open disc, that con-
nect 2n endpoints on the boundary of the disc. These points are labelled
cyclically by 1, 2, . . . , 2n. Two strand diagrams are equivalent if there is
a homeomorphism from one to the other that respects the labelling of the
endpoints.
56
We represent strand diagrams by rectangles, with the labels 1, 2, . . . , n
running down the left side of the rectangle, and the labels n+1, n+2, . . . , 2n
running up the right side. Figure 5.3 gives a strand diagram of order 6.
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
1
2
3
4
5
6 7
8
9
10
11
12
Figure 5.1: A strand diagram of order 6
The connection between strand diagrams and meandric systems comes
from the following lemma, which is illustrated in Figure 5.2.
Lemma 5.1.2. There is a natural bijection between arch congurations of
order n and strand diagrams of order n.
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
1
1
2
2
3
3
4
4
5
5
6 6 7 7 8
8
9
9
10
10
11
11
12
12
)
A proof that there is a unique Markov trace on TL
k
(q), up to a multi-
plicative factor, is given in [3, Appendix A]. As a consequence, a Markov
trace is completely determined by any non-zero evaluation. To give a com-
binatorial interpretation to this trace, we introduce the closure of a strand
diagram.
62
Denition 5.2.7. The closure of a strand diagram S, of order n, is obtained
from S by placing the diagram on a cylinder, and identifying the endpoint i
with the endpoint ni+1 for each i in the range 1 i n. Pictorially, this
is represented by connecting the endpoints around the outside of the disc, in
a planar fashion. Figure 5.8 gives a strand diagram and its closure.
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
1
2
3
4
5
6 7
8
9
10
11
12
fB
1
tr
n
f. (5.3)
Proof. By Lemma 5.1.2, we take the sum to be over all arch congurations
of order n. An arch conguration a with corresponding strand diagram S
64
contributes the term tr
n
(S) = q
#
loop
(S)
. Since the closure of S and the
semi-meandric system with a as an upper conguration are obtained by
connecting the same endpoints, the number of components in the semi-
meandric system with a as an upper conguration is #
loop
(S). Thus an
arch conguration contributes the term q precisely if the corresponding semi-
meandric system is connected.
In order to make a similar statement relating the meandric numbers to
the trace function, we dene the transpose f
t
of f TL
n
(q).
Denition 5.2.13. The transpose function
t
: TL
n
(q) TL
n
(q) is dened
on the multiplicative basis by
e
i
t
= e
i
and extended to the whole algebra through
(ef)
t
= f
t
e
t
and (e +f)
t
= e
t
+f
t
for all e, f in TL
n
(q) and all , in C(q).
In terms of strand diagrams, the action of
t
is to reect the diagram in
the vertical axis and to swap the labels i and 2ni+1 for each i in the range
1 i n. Figure 5.10 gives a strand diagram (a) and its transpose (b).
PSfrag replacements
(a) (b)
(c)
(d)
(e)
1 1
2 2
3 3
4 4
5 5
6 6 7 7
8 8
9 9
10 10
11 11
12 12
Figure 5.10: A strand diagram (a), and its transpose (b)
Using
t
we can dene a symmetric bilinear form
, )
n
: TL
n
(q) TL
n
(q) C(q) (5.4)
(e, f) tr
n
(ef
t
) (5.5)
65
with the property that if e and f are the representation of arch congurations
a and b, then e, f) = q
c(a,b)
, where c(a, b) is the number of components in
the meandric system with a as its upper conguration and b as its lower
conguration.
This last claim is veried by noting that the closure of ef
t
is obtained
by identifying the endpoint i in e with the endpoint i in f for every i in the
range 1 i 2n. This is precisely the identication involved in creating
the meandric system with a as its upper conguration and b as its lower
conguration. So the number of components in (a, b) is #
loop
(ef
t
), and we
obtain the expression,
m
n
(q) =
n
k=1
M
(k)
n
q
k
=
e,fB
1
e, f)
n
, (5.6)
for the n-th meandric polynomial, since, by Lemma 5.1.2, the sum can be
taken to be over all meandric systems of order n. We summarize the form
, )
n
with its Gram matrix, M
n
(q), a C
n
C
n
matrix such that
_
M
n
(q)
_
ij
= a
i
, a
j
)
n
,
where B
1
is the ordered basis (a
1
, a
2
, . . . , a
Cn
), and obtain the following
theorem.
Theorem 5.2.14. The meandric polynomials can be expressed by
m
n
(q) =
Cn
i,j=1
_
M
n
(q)
_
ij
= tr(M
n
(q) J
n
) = u
n
t
M
n
(q)u
n
, and (5.7)
m
n
(q
2
) = tr
_
M
n
(q)
2
_
, (5.8)
where u
n
is the C
n
dimensional column vector of 1s and J
n
is the C
n
C
n
matrix of 1s.
Proof. The expression (5.7) is obtained by rewriting (5.6) in terms of M
n
(q).
Since M
n
(q) is symmetric, and every entry is a monic monomial, every term
66
in tr (M
n
(q)J
n
) appears squared in tr
_
M
n
(q)
2
_
, so (5.8) follows from (5.7).
Since M
n
can be recovered from these expressions through
M
n
= [q] m
n
(q) =
_
q
2
m
n
(q
2
),
we have reduced the meander problem to the problem of determining the
Gram matrix of , )
n
on TL
n
(q) with respect to the basis B
1
. As with the
techniques discussed in previous chapters this is a computationally dicult
task.
Example 5.2.15. With respect to the basis B
1
= e
1
, e
2
e
1
, e
1
e
2
, e
2
, 1, the
Gram matrix of , )
3
is
M
3
(q) =
_
_
_
_
_
_
_
_
q
3
q
2
q
2
q q
2
q
2
q
3
q q
2
q
q
2
q q
3
q
2
q
q q
2
q
2
q
3
q
2
q
2
q q q
2
q
3
_
_
_
_
_
_
_
_
.
Using this matrix, we see that m
3
(q) = tr(M
3
(q)J
3
) = 8q +12q
2
+5q
3
, and
recover M
3
= [q]m
3
(q) = 8.
Despite the apparent computational diculty, this approach is promis-
ing. The Gram matrix is richly structured and opens the meander problem
to the tools of linear algebra. In [6], Di Francesco, et al. construct a second
basis B
2
for TL
n
(q) with respect to which the Gram matrix is diagonal, and
order the elements of B
1
and B
2
such that the change of coordinate matrix
from B
1
to B
2
is triangular. In principle, the meander problem has been
reduced to expressing B
2
in terms of B
1
, but in practice, only the diagonal
entries of the change of coordinate matrix have been determined. This is
sucient to calculate the determinant of M
n
(q) but does not provide a nal
solution to the problem.
67
PSfrag replacements
(a)
(b)
(c)
(d)
(e)
1
1
2
2
3
3
4
4
5
5
6
6
7
7
8
8 9
10
11
12
13
14
15
16
2
2n
denotes a generic irreducible word, and there is one pro-
duction rule for every irreducible word. By showing that every meandric
system has a unique derivation using these rules, they are able to show that
the ordinary generating series for meandric systems with respect to order,
B(x), and the ordinary generating series for irreducible meandric systems
with respect to order, N(x), satisfy the functional equation
B(x) = N(xB
2
(x)). (6.3)
Using analytic techniques, Lando and Zvonkin combine (6.3) and (2.5)
73
to conclude that the radius of convergence of N is
_
4
_
2
.
This leads to their main result, that meandric numbers satisfy the inequality,
M
n
<
_
4
_
2n
, (6.4)
for all suciently large n values of n.
6.3 Production Rules For Meanders
We do not yet have a collection of production rules for the language L