MATHEMATICS RESEARCH DEVELOPMENTS SERIES

MATHEMATICS AND MATHEMATICAL

LOGIC: NEW RESEARCH

No part of this digital document may be reproduced, stored in a retrieval system or transmitted in any form or
by any means. The publisher has taken reasonable care in the preparation of this digital document, but makes no
expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No
liability is assumed for incidental or consequential damages in connection with or arising out of information
contained herein. This digital document is sold with the clear understanding that the publisher is not engaged in
rendering legal, medical or any other professional services.

MATHEMATICS RESEARCH DEVELOPMENTS

SERIES
Boundary Properties and Applications of the Differentiated Poisson Integral for
Different Domains
Sergo Topuria
2009. ISBN 978-1-60692-704-5
Quasi-Invariant and Pseudo-Differentiable Measures in Banach Spaces
Sergey Ludkovsky
2009. ISBN 978-1-60692-734-2
Operator Splittings and their Applications
Istvan Farago and Agnes Havasiy
2009. ISBN 978-1-60741-776-7

Measure of Non-Compactness for Integral Operators in Weighted

Lebesgue Spaces
Alexander Meskhi
2009. ISBN: 978-1-60692-886-8
Mathematics and Mathematical Logic: New Research
Peter Milosav and Irene Ercegovaca (Editors)
2010. ISBN: 978-1-60692-862-2

MATHEMATICS RESEARCH DEVELOPMENTS SERIES

MATHEMATICS AND MATHEMATICAL

LOGIC: NEW RESEARCH

PETER MILOSAV
AND

IRENE ERCEGOVACA
EDITORS

Nova Science Publishers, Inc.

New York

Copyright 2010 by Nova Science Publishers, Inc.

All rights reserved. No part of this book may be reproduced, stored in a retrieval system or
transmitted in any form or by any means: electronic, electrostatic, magnetic, tape, mechanical photocopying, recording or otherwise without the written permission of the Publisher.
For permission to use material from this book please contact us:
Telephone 631-231-7269; Fax 631-231-8175
Web Site: http://www.novapublishers.com
NOTICE TO THE READER
The Publisher has taken reasonable care in the preparation of this book, but makes no expressed or
implied warranty of any kind and assumes no responsibility for any errors or omissions. No liability is assumed for incidental or consequential damages in connection with or arising out of information contained in this book. The Publisher shall not be liable for any special, consequential, or
exemplary damages resulting, in whole or in part, from the readers use of, or reliance upon, this
material. Any parts of this book based on government reports are so indicated and copyright is
claimed for those parts to the extent applicable to compilations of such works.
Independent verification should be sought for any data, advice or recommendations contained in
this book. In addition, no responsibility is assumed by the publisher for any injury and/or damage
to persons or property arising from any methods, products, instructions, ideas or otherwise contained in this publication.
This publication is designed to provide accurate and authoritative information with regard to the
subject matter covered herein. It is sold with the clear understanding that the Publisher is not engaged in rendering legal or any other professional services. If legal or any other expert assistance
is required, the services of a competent person should be sought. FROM A DECLARATION OF
PARTICIPANTS JOINTLY ADOPTED BY A COMMITTEE OF THE AMERICAN BAR ASSOCIATION AND A COMMITTEE OF PUBLISHERS.
LIBRARY OF CONGRESS CATALOGING-IN-PUBLICATION DATA
Mathematics and mathematical logic : new research / [edited by] Peter Milosav and Irene Ercegovaca.
p. cm.
Includes bibliographical references and index.
ISBN 978-1-61470-218-4 (eBook)
1. Logic, Symbolic and mathematical. I. Milosav, Peter. II. Ercegovaca, Irene.
QA9.M3475 2009
510--dc22
2009012375

Published by Nova Science Publishers, Inc. New York

CONTENTS
Preface

vii

Chapter 1

On Numbers Nature
Dimitris Gavalas

Chapter 2

The Application of Fuzzy Linear Programming in Engineering

C. Riverol

Chapter 3

A New Consensus Scheme for Multicriteria Group Decision

Making under Linguistic Assessments
P. Bernardes, P. Ekel and R. Parreiras

67

The Mathematical Basis of Periodicity in Atomic and Molecular

Spectroscopy
K. Balasubramanian

87

Chapter 4

1
57

Chapter 5

Second Order Definability in a Model

George Weaver

115

Chapter 6

Algebraic Topics on Discrete Mathematics

Gloria Gutirrez Barranco, JavierMartnez, SalvadorMerino
and Francisco J. Rodrguez

129

Chapter 7

Mathematical models of QoS Management for Communication

Networks
Chia-Hung Wang and Hsing Luh

Chapter 8

Reversible Logic
Alexis DeVos

Chapter 9

Imaginary Cubic Oscillator and Its Square-well Approximations in

x and p Representation
Miloslav Znojil

Index

179
203

243
261

PREFACE
A computational model is a mathematical structure which is constructed in an easy way
and is useful in performing certain computations on the structure. This book discusses how
the idea of "computational model" has been formalized by connecting the domain theory with
the theory of metric spaces and related topological spaces. The nature of number is also examined, from a synthetic, holistic and interdisciplinary point-of-view, which is mostly
mathematical but also encompasses respective psychological, neuro-physiological and philosophical views. Reversible logic circuits are discussed, which are beneficial to both classical and quantum computer design. Three experimental prototypes are used to illustrate how,
in the near future, reversible computers will outperform conventional computers, in terms of
power dissipation and heat generation. In quantum mechanics and field theory, Schroedinger
equation for a single particle in one-dimensional imaginary potential represents one of the
most popular schematic non-Hermitian models. Via a solvable square-well approximation
technique, this book examines the problems in both the coordinate and momentum representations. The optimization problem is also looked at, which in traditional designs, is stated in
precise mathematical terms but in real life, are stated in vague and linguistic terms.
Chapter 1 - This chapter deals with the question regarding the essential nature of number.
It is about ideas, which commence from the Pythagoreans and through the Platonists end to
the modern era. Here the authors meet on the one hand Cantor, Gdel and their descendents
and on the other Jung and his students. The distance, between the numbers well-known concept as cardinality and means of numbering and the one of dynamical psychic factor, which is
sustained in this article, is enormous.
What really happens with numbers issue? This question concerns us in a work, which
poses more problems than those that solves. In the present text, the authors goal is to investigate the nature of number from a synthetic, holistic, interdisciplinary aspect, which mostly
uses mathematical, but also the respective psychological, neuro-physiological and philosophical view. The spread of information also, which is taken as evidence of progress and knowledge, can be viewed as an effort to compensate and counterbalance the loss of the collective
memory and the natural wisdom. For this reason the recourse to ancient knowledge and to
modern research about the collective memory, that is the unconscious and its structural elements -Jungs archetypes-, is obligatory.
The structure of this chapter goes as follows: After the Introduction and Etymology of the
Concepts Number and Archetype it encompasses four parts.

viii

Peter Milosav and Irene Ercegovaca

In Part I: Pythagoras and His Descendants on Number, the authors discuss Pythagoras
View about Number and His Vindication in Contemporary Mathematics.
In Part II: Number in Contemporary Mathematics, the authors present numbers concept
in relation to Standard Contemporary Mathematics, Logic, Category Theory, Philosophy,
Neurophysiological and Social Aspects, Fuzziness, and Non-standard View.
In Part III: Jung and the Concept of Archetype, the authors refer to subjects as Chthonic
and Celestial Mathematics, Jungs - Paulis General Hypothesis of Archetypes, Number as
Archetype, Jung and Cantor - Gdel, and Number and Jungs Psychology.
In Part IV: Number, Monoid and Archetype, the authors discuss the issues 2-categories;
Monad, Monoid, Monoidal Category, Monoid Object; The Dynamical System as Functor,
The Archetype of Number as Monoid and its Interpretation as the Set of Numbers, and A
Holistic View of Mathematics.
The chapter comes to an end with the Final View and the Epilogue.
Chapter 2 - As will be seen, the mathematical apparatus of the fuzzy theory of fuzzy sets
provides a natural basis for the theory of the possibility as well described [1]. Viewed in this
perspective, a fuzzy restriction may be interpreted as a possibility distribution with its membership function playing the role of possibility distribution function. This principle will be
described in this chapter.
Chapter 3 - This work proposes a new consensus scheme for group decision making,
which allows one to obtain a consistent collective opinion, from information provided by each
expert in terms of multigranular fuzzy estimates. It is based on a linguistic hierarchical model
with multigranular sets of linguistic terms, and the choice of the most suitable set is a prerogative of each expert. From the human viewpoint, using such model is advantageous, since
it permits each expert to utilize linguistic terms that reflect more adequately the uncertainty
level intrinsic to his evaluation. From the operational viewpoint, the advantage of using such
model lies in the fact that it allows one to express the linguistic information in a unique domain, without information losses, during the discussion process. Such consensus scheme is
applied in the analysis of a multicriteria decision problem, generated with the use of the Balanced Scorecard methodology for enterprise strategy planning. Three techniques for multicriteria analysis, based on fuzzy preference relation modeling, are considered. They permit the
evaluation, comparison, selection, prioritization, and/or ordering of alternatives with the use
of both quantitative and qualitative estimates. With the availability of different techniques, the
most appropriate one can be chosen, considering possible sources of information and its uncertainty.
Chapter 4 - This chapter applies combinatorial and group-theoretical relationships to the
study of periodicity in atomic and molecular spectroscopy. The relationship between combinatorics and both atomic and molecular energy levels must be intimate since the energy levels
arise from the combinatorics of the electronic or nuclear spin configurations or the rotational
or vibrational energy levels of molecules. Over the years the authors have done considerable
work on the use of combinatorial and group-theoretical methods for molecular spectroscopy.
The role of group theory is evident since the classification of electronic and molecular levels
has to be made according to the irreducible representations of the molecular symmetry group
of the molecule under consideration. Combinatorics plays a vital role in the enumeration of
electronic, nuclear, rotational and vibrational energy levels and wave functions. As can be
seen from other chapters in this book, the whole Periodic Table of the elements has a mathematical group-theoretical basis since the electronic shells have their origin in group theory.

Preface

ix

Indeed, this concept can even be generalized to other particles beyond electrons such as bosons or other fermions that exhibit more spin configurations than just the bi-spin orientations
of electrons.
Chapter 5 - The back and forth characterization of equivalence of interpretations for finitary (and some infinitary) second-order languages introduced in Weaver and Penev [2005] is
applied to obtain a condition necessary and sufficient for an attribute of an interpretation to be
definable in that interpretation by a second-order formula (either finitary or infinitary). This
condition is applied to obtain some "reduction" theorems for the second-order theories of
those infinite interpretations having pairing functions that are definable by two simple classes
of second-order formulas.
Chapter 6 Many applications of discrete mathematics for science, medicine, industry
and engineering are carried out using algebraic methods, such as group theory, polynomial
rings or finite fields.
This chapter is intended as a survey of the main algebraic topics used for developing
methods in ambit of combinatorial theory. Plya's enumeration method, Latin squares,
patterns design or block design are examples of this usage. Others applications can be seen in
Section 5, about foundation design, RNA patterns or octonions.
Chapter 7 - Network design and network synthesis have been the classical optimization
problems in telecommunication for a long time. In the recent past, there have been many
technological developments such as digitization of information, optical networks, Internet,
and wireless networks. These developments have led to a series of new optimization problems. In communication networks, a number of requirements can be identified. For instance,
the network operator requires a good earning capacity, and the network users require reliable
communication. The degree to which these and other requirements are fulfilled, can in many
cases be deduced from how the network resources are distributed. In contrast to the Public
Switched Telephony Network (PSTN), services offered in a data communication network
vary significantly in terms of the required bandwidth. This adds a new challenging dimension
into design of the networks, since inevitably, questions of fair medium sharing, quality guarantees, delays, etc. become crucial.
Different methods of obtaining fair resource sharing are derived and investigated, for example, Max-Min Fairness, Proportional Fairness, etc. The authors focus on fairness on backbone networks possessing certain desired fairness properties.
Chapter 8 - Reversible logic circuits are beneficial to both classical and quantum computer design. Present-day logic building-blocks (like OR gates and NAND gates) are logically
irreversible and therefore cannot be used for designing reversible computers. Thus reversible
computation needs an appropriate design methodology. In contrast to conventional digital
logic circuits, reversible logic circuits (of a same logic width w) form a mathematical group.
The reversible circuits of width w form a group isomorphic to the symmetric group S2w. Its
Young subgroups allow systematic and efficient synthesis of an arbitrary reversible circuit.
The author can choose a left coset, a right coset, or a double coset approach. The optimal design is reminiscent of the so-called banyan networks of telecommunication.
As an illustration, three experimental prototypes (in c-MOS chip technology) of reversible computing devices are presented. Special care has been taken to avoid as much as possible
the appearance of garbage bits. The examples illustrate how, in a near future, reversible com-

Peter Milosav and Irene Ercegovaca

puters will outperform conventional computers, in terms of power dissipation and heat generation.
Chapter 9 - Schrdinger equation with imaginary PT symmetric potential V (x) = i x3 is
studied using the numerical discretization methods in both the coordinate and momentum
representations. In the former case our results confirm that the model generates an innite
number of bound states with real energies. In the latter case the dierential equation is of the
third order and a square-well, solvable approximation of kinetic energy is recommended and
discussed. One finds that in the strong-coupling limit, the exact PT symmetric solutions converge to their Hermitian predecessors.

In: Mathematics and Mathematical Logic: New Research

ISBN 978-1-60692-862-2
Editors: Peter Milosav and Irene Ercegovaca
2010 Nova Science Publishers, Inc.

Chapter 1

ON NUMBERS NATURE
Dimitris Gavalas*
Varvakeios Experimental School
Athens, Greece

Where is the wisdom we have lost in knowledge?

Where is the knowledge we have lost in information?
T. S. Eliot, Choruses from the Rock, I, 15-16.

ABSTRACT
This chapter deals with the question regarding the essential nature of number. It is
about ideas, which commence from the Pythagoreans and through the Platonists end to
the modern era. Here we meet on the one hand Cantor, Gdel and their descendents and
on the other Jung and his students. The distance, between the numbers well-known concept as cardinality and means of numbering and the one of dynamical psychic factor,
which is sustained in this article, is enormous.
What really happens with numbers issue? This question concerns us in a work,
which poses more problems than those that solves. In the present text, our goal is to investigate the nature of number from a synthetic, holistic, interdisciplinary aspect, which
mostly uses mathematical, but also the respective psychological, neuro-physiological and
philosophical view. The spread of information also, which is taken as evidence of progress and knowledge, can be viewed as an effort to compensate and counterbalance the
loss of the collective memory and the natural wisdom. For this reason the recourse to ancient knowledge and to modern research about the collective memory, that is the unconscious and its structural elements -Jungs archetypes-, is obligatory.
The structure of this chapter goes as follows: After the Introduction and Etymology
of the Concepts Number and Archetype it encompasses four parts.

* Folois 6
112 56 Athens, Greece
gavdim@otenet.gr

Dimitris Gavalas
In Part I: Pythagoras and His Descendants on Number, we discuss Pythagoras View
about Number and His Vindication in Contemporary Mathematics.
In Part II: Number in Contemporary Mathematics, we present numbers concept in
relation to Standard Contemporary Mathematics, Logic, Category Theory, Philosophy,
Neurophysiological and Social Aspects, Fuzziness, and Non-standard View.
In Part III: Jung and the Concept of Archetype, we refer to subjects as Chthonic
and Celestial Mathematics, Jungs - Paulis General Hypothesis of Archetypes, Number
as Archetype, Jung and Cantor - Gdel, and Number and Jungs Psychology.
In Part IV: Number, Monoid and Archetype, we discuss the issues 2-categories; Monad, Monoid, Monoidal Category, Monoid Object; The Dynamical System as Functor,
The Archetype of Number as Monoid and its Interpretation as the Set of Numbers, and A
Holistic View of Mathematics.
The chapter comes to an end with the Final View and the Epilogue.

INTRODUCTION
From a cognitive point of view, the reason to discuss the subject of number is the fundamental changes in its mental representation. Indeed, after the failure of the effort to limit
number in a frame of Logic -that is to limit it only to its logical aspect-, nowadays inside the
cognitive realm in order to have a complete image for it, is emerged the need to see it from all
of its aspects, logical and non-logical. Gdel set aside the Logic of Frege - Russell - Hilbert
and posed the matter of uncertainty for Mathematics, using its own means. Other mathematicians also, like Weyl, talk about the abysmal character of numbers. Afterwards, the introduction of methods of Fuzzy Logic and Non-standard Analysis leads to corresponding views
about number.
In a society, and especially in a scientific community, where is considered officially that
the dominating function is the rational thinking, even though this is just a myth, a little attention is paid to the other side of things. Jung stresses that no matter how much beautiful and
perfect finds man his logic, this is just one of his psychological functions and man is covered
only by one side. From all the other sides though, man is surrounded by the irrational, the
collective, the unconscious, the chance. Therefore, next to the rational exists the irrational and
in much more proportion. When logic and the time linear sequence of events stops being applied, then our rationalism does not function anymore and in order to understand what is happening we need other functions and methods.

ETYMOLOGY OF THE NUMBER AND ARCHETYPE

In Gk. the word for number, (arithmos), is etymologized from the verb which
means: Attach, adject, seam, collect, osculate, put together, fit, piece, conjoin, unite, integrate,
and the like. So, (arithmos) means: Amplitude, deal, crowd, multitude, oodles, pack,
plurality, quantum, throng, lot, number, amount, quantity, count, enumeration, measurement,
reckoning, numeration, calculation, computation, account, sign of value, of order, of position,
of integration and so on. From (arithmos) comes the term Arithmetic.
According to another view, (arithmos) means: Best/ optimum psyche, best
movement, animate time. This view agrees entirely to Xenocrates who says that psyche is

On Numbers Nature

number which moves itself (Heinze, 1965, Fragm. 61). Aristotle explains that because psyche is both kinetic and knowing, so somebody united these two and said that psyche is number which moves itself (De an.).
In L. we have the words numero (verb), numerus (noun), from which comes the word
number (verb and noun): The sum, total, count, or aggregate of a collection of units, or the
like. Both, Gk. and L words are related to rhythm. In Mathematics and Science number is a
basic concept, and a mode of thought: Number is the basis of science.
Also, the term (archetype) means: Original pattern from which copies are
made, from L. archetypum, which comes from Gk. archetypon pattern, model, firstmoulded from arche- first + typos model, type, blow, mark of a blow. So, from the Gk.
for original pattern, archetype is a basic model from which copies are made; therefore a prototype. In general terms, the abstract idea of a class of things which represents the most typical and essential characteristics shared by the class, thus a paradigm or exemplar. Jungian
psychology speeks about pervasive idea or image from the collective unconscious.

PART I: PYTHAGORAS AND HIS DESCENDANTS ON NUMBER

1. Pythagoras View about Number
Number is the within of all things, Pythagoras.

Pythagoras considers that number lies behind all things and facts, beginning from musical
harmony and finishing to the planets orbits. This aspect leads him to the belief that everything
is number. Iamblichus (1922; Waterfield, 1988a, 1988b) writes about the Pythagorean view
on number. According to him, the Pythagorean philosophy bases mostly on the mathematical
semantics of numbers and their relationships. The study of number entails not only the material phenomena but mostly the spiritual and mental aspects, the symbolic meaning of each
separate number as well as its mystical properties. All these views about number do not limit
to the relationships of the material phenomena, instead number has mental, cosmological, and
ethical applications. This means that the Pythagorean entailment on number is not just a
mathematical science according to the modern view, but study of the hidden wisdom of number, which contains truths that are being taught on the initiatory rites of antiquity.
In Metaphysics Aristotle describes the teaching of Pythagoreans: They thought they
found in numbers, many resemblances to things which are and become; thus such and
such an attribute of numbers is justice, another is soul and mind, another is opportunity,
and so on; and again they saw in numbers the attributes and ratios of the musical scales.
Since, then, all other things seemed in their whole nature to be assimilated to numbers,
while numbers seemed to be the first things in the whole of nature, they supposed the
elements of numbers to be the elements of all things, and the whole heaven to be a musical scale and a number.
According to O Meara (1990), the Pythagorean idea that number is the key to
understanding reality inspired Neoplatonist philosophers in Late Antiquity to develop theories
in Physics and Metaphysics based on mathematical models. He examines this theme,
describing first the Pythagorean interests of Platonists in the second and third centuries and

Dimitris Gavalas

then Iamblichuss program to Pythagoreanize Platonism in the fourth century in his work On
Pythagoreanism -whose unity of conception is shown and parts of which are reconstructed for
the first time. The impact of Iamblichuss program is examined as regards Hierocles of
Alexandria and Syrianus and Proclus in Athens: Their conceptions of the figure of Pythagoras
and of Mathematics and its relation to Physics and Metaphysics are examined and compared
with those of Iamblichus. This provides insight into Iamblichuss contribution to the
evolution of Neoplatonism, to the revival of interest in Mathematics, and to the development
of a philosophy of Mathematics and a mathematizing Physics and Metaphysics.
So, for Philolaus the nature of number is gnomonical, magisterial and pedagogical: If
one focuses and reflects on number as symbol, then one can realize concepts that lie beyond
the limits of the usual human consciousness. According to Aristotle, the Pythagoreans sustain
that entities pattern upon number. Syrianus also characterizes number as eidetic, that is its
meaning is demonstrative of the kind, the quality and not only the quantity of the things and
entities.
Proclus helps us to realize the relationship between number and the rest Pythagorean philosophy. By his sayings it is clear that the mystical meaning of number is closely related to
the Pythagorean mystical revelation. This meaning is revealed to the initiates at rites, where
the relations and lows that refer to the divine aspect of number for the principles and things
are being explained to them. Many writers of the antiquity compare the Pythagorean meaning
of number with that of Orpheus who seems to be the first that introduces this mystical theory.
Syrianus says that the Pythagoreans accepted the theory about number from the Orphics.
They also extended it up to the sensual material phenomena beginning from the spiritual and
the mental aspects.
For the non-sensual that is the spiritual, mental, mystical and esoteric meaning of number, we have many fragments from the writers of antiquity. However, a complete and systematic study is not rescued nowadays except the one attributed to Iamblichus. In his work the
mathematical properties of the first ten natural numbers are developed and are referred the
many appellative names for each one of them. Both from the mathematical properties and
from the appellative names we can figure out the way and the spirit under which the Pythagoreans dealt with number. We can also compare this spirit with the modern approach and ask
ourselves, whether except from ours is there another one lost and forgotten spirit, which is
the other side of the coin that must co-exist with the current one.
Therefore, for the Pythagoreans, number is the structural element both for the material
world and for the soul and pre-existed of them and indeed the world and the soul were created
in pattern upon the number. The Pythagoreans deepen in the numbers concept, considering
that the numbers principles are all entities principles. The natural elements are inadequate for
the worlds interpretation; in contrary to the number that contains mental power. The elements consisting number are also all entities elements and the heaven is governed by
harmony and number. With Pythagoras a new multi-leveled world is created, since the sensual is reduced to the conceptual, while the conceptual in its turn is reduced to the ideal. This
view is further developed by Plato.
With the Pythagoreans, number acquires transcendental power, since by its existence derive and become thinkable the physical magnitudes. In Plato also, number tends to replace
idea. Indeed, Plato is inducted to the Hen (One) and he approaches the idea to the number,
reducing the many to the One. So, if the worlds nature is mathematical according to Pythago-

On Numbers Nature

ras and Plato, this happens in virtue of number mental power, which ensures ontological validity. Number is the highest degree of knowledge; it is knowledge itself, Plato says.

2. Pythagoras Vindication in Contemporary Mathematics

Eves (1983, Lecture 32), about Pythagoras vindication in contemporary Mathematics,
says that, if we could hear Pythagorass voice about the end of the 19th century, it would say:
I have been told you over 2000 years ago: everything depends on natural numbers. This is so
because after an extremely important line of discoveries, mathematicians showed that
Mathematics is consistent, if the system of natural numbers is consistent. Indeed, the structure
of Mathematics is a huge inverted pyramid leaning its peak on the system of natural numbers.
Analysis, mainly on its first steps, displayed many problems. Under every element in it
were deep properties of the system of the real numbers that demanded immediate understanding. Weierstrass sustained a program according to which the system of real numbers itself
should first be developed logically and afterwards Analysiss concepts to be defined according to the system of real numbers. This program is known as the Arithmetization of Analysis and was realized at the end of the 19th century. The success of this program had great impact. Since Analysis bases on the system of the real numbers we conclude that it is consistent
only if the system of the real numbers is consistent.
Euclidean Geometry, after its arithmetization by Descartes, can also base on the system
of the real numbers; therefore it is consistent only if the system of the reals is consistent. Still,
various geometries are consistent if Euclidean Geometry is consistent. The same happens also
with Algebra. All these lead to the conclusion that the majority of Mathematics is consistent if
the system of the reals is also consistent. Here lies the immense importance of the system of
the reals for the foundation of Mathematics. Finally, was adopted the view that we can start
from an initial set of axioms more basic than the one of the system of the reals and genetically, that is by stating the appropriate definitions and by applying clever techniques, to reach
the system of the reals.
This forces the consistency of the system of the reals -and therefore of the majority of
Mathematics- to depend on the consistency of this initial and more basic set of axioms. This
was succeeded and the system of the reals came along only from apposite definitions and appropriate techniques, without further assumptions, from a set of axioms for the much more
simple and basic system of the natural numbers. This success gave to the mathematicians an
important sense of security about the consistency of the majority of Mathematics. This derives from the fact that the system of the natural numbers has intuitive simplicity, and clarity
and has been vastly used for a longer period of time without presenting any internal contradictions and does not present the complexity of other mathematical systems. Mainly, it is inherent in the humans and natural, because it is an archetype.
Therefore, beginning from the naturals following various techniques we can construct the
integers and then from the integers the rational numbers. Subsequently, we can introduce the
irrationals basing on the rationales and finally we can define the reals. The final step is the
hardest, but is realized with many ways, like Dedekind cuts etc. This development is represented as follows: N Z Q R C, since from the reals we can have the complex
numbers. Hence, on the system of the naturals bases the construction of Mathematics and its
consistency. Pythagorass view that literally everything in the world depends on the naturals

Dimitris Gavalas

was vindicated and is expressed by Kronecker as follows: God has made the naturals, all the
rest is mans project. Their divine nature makes them so simple, but their deepest substance
always escapes from us.

PART II: NUMBER IN CONTEMPORARY MATHEMATICS

3. Number and Standard Contemporary Mathematics
The natural numbers are:
(i) A sequence of marks/ signs, e.g. |, ||, |||, ||||, |||||, , for use in counting or labeling.
Any other sequence of marks/ signs would do as well, provided the sequence always
provides a way to write down the next (the successor) mark/ sign. This is the case
with decimal notation 0, 1, 2, , 10, 11, 12, , 100, , 200, .
(ii) Finite sets, with equinumerous sets regarded as equal numbers, while successor
means adjoin one more element.
(iii) Cardinal-equivalence classes of finite sets.
(iv) Ordinal-equivalence classes of finite ordered sets, while successor means adjoin one
new element, to come after all the others.
(v) Finite sets of sets, linearly ordered by the membership relation as follows: 0=,
1={0}, 2={0, 1}, , ={0, 1, 2, }.
In each of these cases, the natural numbers satisfy the Peano postulates. From this multiplicity of answers to the question what is a natural number, we must conclude that there is
no answer to the question. There are various alternative concrete descriptions depending on
the sort of counting intended or on the prior assumptions. In each case, the description provides numbers which do satisfy the Peano postulates. Hence we conclude that one does not
define what a natural number is by itself. Instead, one defines the system of all natural
numbers, with successor operation s(n)=n+1. Then N is any such system which satisfies the
Peano postulates. This means that there are many such systems within Set Theory, but that
they are all isomorphic.
The postulates themselves are, by no means, unique. The Peano postulates, for example,
may be replaced by the Recursion Theorem as an axiom. Here, as in other axiomatic descriptions of mathematical objects, there are a variety of choices for lists of axioms. Number Theory, like other subjects in Mathematics, is not the study of a unique model nor yet the examination of a unique axiomatic system, is it rather a study of the form exemplified by the various models and specified in the axioms.
Perhaps the simplest aspect we have for the natural numbers is that, there is a minimum
number 0, that each number n has an immediate successor s(n) and that, if we commence with
the minimum 0 and we construct consecutively the successor of each number ad infinitum, 0,
s(0)=1, s(1)=2, s(2)=3, we enumerate all natural numbers. This aspect is expressed by
Peano axioms on the structured set <, 0, s>.
The Theorem of Existence and Uniqueness of Natural Numbers tells us that there is at
least one system of natural numbers. If there are two systems of natural numbers, then there is

On Numbers Nature

exactly one isomorphism between them that is all systems of natural numbers are isomorphic
by twos. Thus we determine a particular system <, 0, s> of natural numbers, the members of
which are called natural numbers. A choice for this system could be the following: N={,
{}, {{}}, } and s: n s(n)={n}. Another choice is ={0, 1, 2, 3, } and s(n)=n+1,
that is to accept that there is indeed the set of true natural numbers, where each of them is
not a set but a number, and the successor is not the function s: n s(n)={n} but the function
which correlates to every number n its successor n+1. The classical theory of Zermelo, allows
for instance such non-sets -as the true natural numbers- as urelements and it only insists on
the premise that the system of the natural numbers must satisfy Peanos axioms. Therefore,
we construct the natural numbers beginning with 0 or and infinitely repeating the operation
of the successor.
To summarize: The natural numbers start out from elementary operations of counting,
listing and comparing. They then develop into effective tools for calculation. The rules for
calculation are formal and can be organized as the consequences of simple systems of postulates. The consequences of these postulates include the remarkably varied and rich properties
studied in Number Theory, properties by no means apparent in the original processes of
counting and listing. (Mac Lane, 1986)
From a philosophical point of view, the common aspect of mathematicians and philosophers of science was, and up to a point still is, that it is really feasible to construct the set of
natural numbers from nothing and only with Logic. But all the relevant efforts lean against
the fallacious General Principle of Inclusion and lead to contradictions. Logic, by its nature,
seems that cannot prove the existence of anything, which means that it cannot have ontological demands. This is a critical point for the continuation of our aspects. Since it is not Logic,
and the resulting typical theories that lead to the answer about numbers existence, then which
is?

4. Number and Logic

Work in Logic has provided several developments bearing on numbers. In retrospect, the
development of the concept of number prompts the question of What number is? This can
be interpreted as the question of whether number is Platonic object, mental construction, or
nothing more than mystified numeral. Alternatively, in virtue of the plethora of kinds of numbers, we can interpret the question as asking what makes entities of certain kinds, but not others, numbers. Beyond a rather vague characterization, it seems difficult to give a general one
of number. The most fundamental numbers, the naturals, measure size or order, are subject to
distinctive operations and can be the roots of equations. Each of these central features has
played a role in generating new kinds of numbers. The result is a collection of entities which
are related by family resemblance (Wittgenstein, 1974), though the boundaries of the family
seem somewhat arbitrary and vague. It is difficult to see why, for example, complex numbers
are called numbers, but not vectors or numerical matrices; both of these share the central features of natural numbers.
This conclusion is reinforced by the work of Conways (1976). He gives a transfinite recursive construction that generalizes both constructions, the Dedekinds of the reals and the
von Neumanns of ordinals. Essentially, a number is any pair <L, R>, such that all the mem-

Dimitris Gavalas

bers of L and R are numbers, and every member of R is greater than or equal to every member of L. The and the arithmetical operations are also defined in a natural recursive manner. The construction generates virtually all the numbers, including infinitesimals, but excluding the complex numbers and if cardinals are to be identified with initial ordinals, a nonuniform definition of arithmetical operations is necessary. Moreover, the construction generates many novel numbers, for example, numbers obtained by applying the full range of realnumber operations to infinite numbers, which make no sense on the usual understanding.
Moreover, a simple generalization of the construction -dropping the ordering condition on L
and R-, produces even more number-like objects.
Just conceivably, a unifying account of number might eventually be found, but in the
meantime the emergence of new kinds of numbers seems likely. There are non-standard inconsistent models of Arithmetic which contain inconsistent numbers -natural numbers with
inconsistent properties. These have some notable applications; for example, some of them can
be shown to provide solutions for arbitrary sets of simultaneous linear equations. Just as the
existence of non-standard models of Analysis made infinitesimals legitimate, so might these
legitimize the notion of an inconsistent number.
Besides, there are three points of particular importance:
(i) The proof by Gdel in 1931 that the Peano axioms, and all other consistent axiom
systems for Arithmetic, are incomplete, in the sense that there are truths of Arithmetic that cannot be proved from the axioms -at least if the underlying Logic is firstorder (Gdels Theorems). The axioms are complete if the underlying Logic is second-order and the Induction Principle is formulated as a second-order axiom and not
just a first-order schema; but second-order Logic is not itself axiomatizable. This
raises profound questions about the nature of both numbers and our knowledge about
them.
(ii) The paradoxes surrounding transfinite numbers. The dominant view is that they are
embedded in ZF Set Theory. According to this, there is no totality of all ordinals, all
sets or other large collections, and so the question of their size does not arise. Although this account provides enough Set Theory for most Mathematics -though not
all: Category Theory appears to require large sets of just this kind-, it can hardly be
said to be conceptually adequate. Standard Logic, for example, defines the sense of a
quantifier in terms of the domain (totality) over which it ranges. It is therefore unclear what the sense of the quantifiers of ZF is, if, as it claims, there is no such totality.
(iii) Robinsons Non-standard Analysis. As was proved by Lwenheim-Skolem, firstorder theories of number have non-standard models. In particular, any theory of the
reals has such models. Robinson shows that in all of these models, there are non-zero
numbers that are smaller than any real number: Infinitesimals. Using these, he demonstrates that the reasoning of the Infinitesimal Calculus -which is much more intuitive than limit reasoning-, can be interpreted in a perfectly consistent manner. Hence,
infinitesimals have been rehabilitated as perfectly good numbers (Priest, 1998).
It is regarded that natural numbers should obey the Induction Principle, but this exhibits a
form of circularity known as impredicativity: The statement of the principle involves quantification over properties of numbers, but to understand this quantification we must assume a

On Numbers Nature

prior grasp of the number concept, which it was our intention to define. It is nowadays a
commonplace to draw a distinction between impredicative definitions and impredicative
specifications: The first are illegitimate, while the second are not. The conclusion in this case
is that the Induction Principle on its own does not provide a non-circular route to an understanding of the natural number concept. It is therefore needed an independent argument. Five
main strategies have been attempted, which are the well known: Intuitionism, Platonism,
Formalism, Logicism, and Empiricism.

5. Number and Philosophy: From Frege to Benacerraf

Frege (1884/ 1980, 1893-1903/ 1967) considered the introduction of numbers via numerically definite quantifiers. What led him to reject it was a problem facing not just this but
any strategy involving implicit definitions; such definitions do not fully determine the identity
conditions of the objects they attempt to introduce. This is known as the Julius Caesar problem because Frege posed it by asking how one could tell from the implicit definitions
whether or not Julius Caesar is a number. This question creates a problem for the strategy of
explaining the number-words substantive use in terms of their adjectival use. The problem
bites only when we attempt to move beyond elementary Arithmetic and quantify over numbers.
Another logicist strategy is to derive Arithmetic from the numerical equivalence, that is,
the contextual principle -called Humes principle- that the number of Fs is the same as the
number of Gs if and only if the Fs and the Gs can be correlated 1-1. Frege sketched a construction of the natural numbers and a proof that they satisfy Peanos axioms, assuming only
second-order Logic and the numerical equivalence. He nevertheless rejected the strategy of
basing Arithmetic on this equivalence because the Julius Caesar problem applies to it as
much as to the previous contextual strategy. Interest in the strategy has been revived by
Wright (1983), who argues that the Julius Caesar problem can be solved by appeal to an
independently plausible sortal inclusion principle to the effect that objects must be of different sorts if the content of their identity conditions is sufficiently different. The proposal Frege
settled on instead was to define the number of Fs explicitly as the class of all concepts
equinumerous with F. He showed that on this definition of number Arithmetic can be deduced
from what he took to be logical principles. However, one of his allegedly logical principles
was inconsistent, and subsequent attempts -most notably by Whitehead and Russell- to repair
Freges system have had to appeal to principles even their advocates have baulked at calling
logical.
Freges solution to the Julius Caesar problem -the underdetermination of the objects of
Arithmetic by the principles about them to which we are committed- is in any case under
threat from the opposite problem: Any non-arithmetical determination of the objects which
solves the Julius Caesar problem will give numbers extra properties which, since they do
not flow from the principles governing numbers, must be arbitrary and hence spurious. This
problem was mentioned by Dedekind (Ewald, 1996), but it has come to prominence more
recently through a much-cited paper by Benacerraf (1965). Two ways of dealing with it have
been advanced under the generic label of structuralism. Dedekinds way -that we are capable, once we have given a logical construction of one model of Peanos axioms, of abstracting
away from its particular features to gain a conception of a model without those features- ap-

10

Dimitris Gavalas

peals to a mental process (abstraction) that many have found mysterious. Benacerrafs way that Arithmetic should be seen as the study not of one particular model of Peanos axioms but
of the structure which all such models have in common- is in danger of relapsing into the
axiomatic formalism.
The well-known Russells paradox suggests to Dummett (1991) that the concept set is
what he calls indefinitely extensible. This means that any attempt to regard the objects falling under the concept as forming a definite totality leads inevitably to the realization that
there are other objects not in the totality which we are nevertheless forced to admit as falling
under the concept. Dummett holds that the presence of indefinitely extensible concepts is a
characteristic feature of Mathematics which should lead us to espouse for it the anti-realism
which his more general meaning-theoretic arguments make room for. He recommends that we
abandon the law of the excluded middle and espouse the Mathematics of Intuitionism but not
Brouwers solipsistic conception of its objects. Dummett thinks that in this way we can retain
Freges logicist insight that numbers are abstract objects truths about which embody deductive subroutines whose application to the world is validated by Logic alone. Just as Brouwers
Intuitionism has been accused of an instability which reduces it to strict Finitism, not everyone is persuaded that Dummetts position does not collapse into the ultra-intuitionism of middle-period Wittgenstein, according to which the meaning of an arithmetical generalization is
identical with its proof. On this view Goldbachs conjecture does not in the present state of
knowledge have any meaning at all. About Freges work one can see the following contemporary studies: Boolos, 1986, 1987; Heck, 1993; Maddy, 1992; Zalta, 1998, 1999.
Benacerraf (1965), on the other hand, discusses the classical problem for set-theoretic
foundations -that Arithmetic has no unique set-theoretic representation- on the line of thought
of Dedekinds (1888/ 1963), Hilberts (1900), Weyls (1927/ 1949) and Bernays (1950/
1976). This argument was influential in shaping later work. It is the inspiration of the position
known as Structuralism -the view that mathematical objects are essentially positions in structures and have no important additional internal composition or nature (Resnik, 1981, 1982;
Shapiro, 1983, 1989).
The version of questions with regard to numbers, which is discussed more often in the
modern philosophical bibliography, emanates from this work of Benacerrafs. The essay of
Benacerrafs does not aspire it explains what precisely the numbers are, but is focused on exposing the limits of various opinions on numbers, explaining what they are not. Concretely,
Benacerraf concludes that numbers are not sets, as most mathematicians believe. In the end in
any case, where he presents his own opinion for the question, he adopts the aspect that numbers do not exist at all. With this view, for which he offers some justification, avoids the difficulties of various theories for the numbers. Moreover, this view is mathematical acceptable,
after the Mathematics do not require to be the numbers certain concrete objects, but simply
some structure exists. It is possible for one to conclude that, after the numbers are a clean
mathematical affair, then the opinion that these do not exist as separate entities it should be
also philosophically acceptable. (Potter, 1998; Heijenoort, 1967).

6. Number and Neurophysiological and Social Aspects

As the reader can find in Campbell (2005), some basic questions are posed concerning
the domain of mathematical cognition: How does the mind represent number and make

On Numbers Nature

11

mathematical calculations? What underlies the cognitive development of numerical and

mathematical abilities? What factors affect the learning of numerical concepts and procedures? What are the biological bases of number knowledge? Do humans and other animals
share similar numerical representations and processes? What underlies numerical and mathematical disabilities and disorders, and what is the prognosis for rehabilitation?
These questions are the domain of mathematical cognition, the field of research concerned with the cognitive and neurological processes that underlie numerical and mathematical abilities. Mathematical cognition research intersects a wide array of disciplines including
cognitive development, neurological development, computational science, cognitive and educational psychology, animal cognition, cognitive and clinical neuropsychology, neuroscience,
and cognitive science.
Already since 1998, Ramachandran & Blakeslee wonder whether each of us has a centre
for numbers and arithmetical operations in the left angular gyrus. This region is in a way necessary for arithmetical calculations, and weirdly is not necessary for the understanding of arithmetical concepts that rule such calculations. And they continue: We dont know yet how
this arithmetic circuit works in the angular gyrus, but at least we know where to search. The
fact that we execute the arithmetical operations without special effort, allows us to conclude
that this ability is hard-wired by birth. But in reality, this ability was acquired after the introduction of basic concepts as zero and the line of numbers. It has also been sustained that
there is in the brain the line of numbers, a kind of graphical, scalar representation of numbers
and that each point of it consists of a group of neurons which point to particular mathematical
value. But are there proofs that such a line exists in the brain?
When healthy individuals are asked which of two given numbers is greater, the subjects
need more time to answer when the numbers are close than being far. Undoubtedly, there is
independent mechanism for the arithmetical operations and this mechanism bases on the angular gyrus of the left brain hemisphere. Rickard in the UCSD, using fMRI (functional magnetic resonance imaging) technique, showed that the region of arithmetical operations is not
entirely on the left angular gyrus but also a little in front. This fact does not affect the basic
assumption and it is matter of time to prove the existence of line of numbers with the aid of a
modern imaging technique. About relevant issues see also Dehaene (1997a).
Number Sense is a short-hand for our ability to quickly understand, approximate, and
manipulate numerical quantities. Dehaenes hypothesis is that number sense rests on cerebral
circuits that have evolved specifically for the purpose of representing basic arithmetic
knowledge. Four lines of evidence suggesting that number sense constitutes a domainspecific, biologically-determined ability:
(i) The presence of evolutionary precursors of arithmetic in animals;
(ii) The early emergence of arithmetic competence in infants independently of other
abilities, including language;
(iii) The existence of a homology between the animal, infant, and human adult abilities
for number processing; and
(iv) The existence of a dedicated cerebral substrate.
In adults of all cultures, lesions to the inferior parietal region can specifically impair
number sense while leaving the knowledge of other cognitive domains intact. Furthermore,
this region is demonstrably activated during number processing. Dehaene postulates that

12

Dimitris Gavalas

higher-level cultural developments in arithmetic emerge through the establishment of linkages

between this core analogical representation -the number line- and other verbal and visual
representations of number notations. The neural and cognitive organization of those
representations can explain why some mathematical concepts are intuitive, while others are so
difficult to grasp. Thus, the ultimate foundations of mathematics rest on core representations
that have been internalized in our brains through evolution (Dehaene, 2001).
In the last decade new research promotes our knowledge regarding the relationship
Mathematics-brain. Specifically, Gbel et al. (2001) sustain that in order to investigate the
hemispheric organization of a language-independent spatial representation of number magnitude in the human brain they applied focal repetitive Transcranial Magnetic Stimulation
(rTMS) to the right or left angular gyrus while subjects performed a number comparison task
with numbers between 31 and 99. Repetitive TMS over the angular gyrus disrupted performance of a visuospatial search task, and rTMS at the same site disrupted organization of the
putative number line. In some cases the pattern of disruption caused by angular gyrus rTMS
suggested that this area normally mediates a spatial representation of number. The effect of
angular gyrus rTMS on the number line task was specific. rTMS had no disruptive effect
when delivered over another parietal region, the supramarginal gyrus, in either the left or the
right hemisphere.
Besides, Dehaene (2003) claims that the recent discovery of number neurons allows for a
dissection of the neuronal implementation of number representation. Recently, some researchers demonstrate a neural correlate of Webers law, and thus resolve a classical debate in
psychophysics: The mental number line seems to be logarithmic rather than linear.
In another interesting article, Dehaene et al. (2004), refer that recent studies in human
neuroimaging, primate neurophysiology, and developmental neuropsychology indicate that
the human ability for arithmetic has a tangible cerebral substrate. The human intraparietal
sulcus (IPS) is systematically activated in all number tasks and could host a central amodal
representation of quantity. Areas of the precentral and inferior prefrontal cortex also activate
when subjects engage in mental calculation. A monkey analogue of these parieto-frontal regions has recently been identified, and a neuronal population code for number has been characterized. Finally, pathologies of this system, leading to acalculia in adults or to developmental dyscalculia in children, are beginning to be understood, thus paving the way for brainoriented intervention studies.
More specifically, regarding negative numbers and their mental representation, Fischer &
Rottmann (2005), refer that the cognitive representation of negative numbers has recently
been studied with magnitude classification and variable standard. Results suggested that negative numbers are cognitively represented on a mental number line. Conflicting with this
observation they report here that response biases for parity classification of negative numbers
are related to absolute digit magnitude only. A control experiment with magnitude classification relative to a fixed standard shows that our result does not reflect mere inattention to the
minus sign. Together, the available evidence suggests that human process negative numbers
less automatically compared to positive numbers.
The IPS is thought to play a role in mental functions, including processing symbolic numerical information. According to Cantlon et al. (2006), adult humans, infants, pre-school
children, and non-human animals appear to share a system of approximate numerical processing for non-symbolic stimuli such as arrays of dots or sequences of tones. Behavioral studies
of adult humans implicate a link between these non-symbolic numerical abilities and sym-

On Numbers Nature

13

bolic numerical processing -e.g., similar distance effects in accuracy and reaction-time for
arrays of dots and Arabic numerals. However, neuroimaging studies have remained inconclusive on the neural basis of this link. The IPS is known to respond selectively to symbolic numerical stimuli such as Arabic numerals. Recent studies, however, have arrived at conflicting
conclusions regarding the role of the IPS in processing non-symbolic, numerosity arrays in
adulthood, and very little is known about the brain basis of numerical processing early in development. Addressing the question of whether there is an early-developing neural basis for
abstract numerical processing is essential for understanding the cognitive origins of our
uniquely human capacity for math and science. Using functional Magnetic Resonance Imaging (fMRI) at 4-Tesla and an event-related fMRI adaptation paradigm, the researchers found
that adults showed a greater IPS response to visual arrays that deviated from standard stimuli
in their number of elements, than to stimuli that deviated in local element shape. These results
support previous claims that there is a neurophysiological link between non-symbolic and
symbolic numerical processing in adulthood. In parallel, they tested 4 years-old children with
the same fMRI adaptation paradigm as adults to determine whether the neural locus of nonsymbolic numerical activity in adults shows continuity in function over development. They
found that the IPS responded to numerical deviants similarly in 4 years-old children and
adults. To our knowledge, this is the first evidence that the neural locus of adult numerical
cognition takes form early in development, prior to sophisticated symbolic numerical experience. More broadly, this is also, to our knowledge, the first cognitive fMRI study to test
healthy children as young as 4 years-old, providing new insights into the neurophysiology of
human cognitive development.
Finally, according to Lakoff & Nez (2000), the only mathematical ideas that human
beings can have are ideas that the human brain allows. We know a lot about what human
ideas are like from research in Cognitive Science. Most ideas are unconscious, and that is no
less true of the mathematical ones. Abstract ideas, for the most part, arise via conceptual
metaphor -a mechanism for projecting embodied that is, sensory-motor reasoning to abstract
reasoning. They argue that conceptual metaphor plays a central, defining role in mathematical
ideas within the cognitive unconscious -from Arithmetic and Algebra to sets and Logic to
infinity in all of its forms: transfinite numbers, points at infinity, infinitesimals, and so on.
Even the real numbers, the imaginary numbers, Trigonometry, and Calculus are based on
metaphorical ideas coming out of the way we function in the everyday physical world. This
work is about mathematical ideas, about what Mathematics means -and why. The authors
believe that understanding the metaphors implicit in Mathematics will make Mathematics
make more sense. Moreover, understanding mathematical ideas and how they arise from our
bodies and brains will make it clear that the brains Mathematics is Mathematics, the only
Mathematics we know or can know.
Let us see now another aspect: For Hersh, Mathematics has existence or reality only as
part of human culture. Despite its seeming timelessness and infallibility, it is a social - cultural - historic phenomenon. The questions he posses -What are numbers? What are infinite
sets? What is the meaning and nature of Mathematics?- are answered in this framework: It is
neither physical nor mental, it is social; it is part of culture and of history. It is like law, like
religion, like all those other things which are very real, but only as part of collective human
consciousness.
To the question What kind of a thing is a number? Hersh says that we can think of two
basic answers -either it is out there some place, like material objects; or it is inside, a thought

14

Dimitris Gavalas

in somebodys mind. Philosophers have defended one or the other of those two answers,
which are both completely wrong. A number is not a thing out there; there is not any place
that it is, or any thing that it is. Neither is it just a thought, because after all, two and two is
four, whether you know it or not. Frege, made quite an issue of the fact that mathematicians
didnt know the meaning of One. What is One? Nobody could answer coherently. Of course
Frege answered, but his answer was no better, or even worse, than the previous ones; and so it
has continued to this very day, strange and incredible as it is. We know all about so much
Mathematics, but we dont know what it really is. Of course when we say, What is a number? it applies just as well to a triangle, or a circle, or a differentiable function, or a selfadjoint operator. We know a lot about it, but what is it? What kind of a thing is it?
When we say that a mathematical thing, object, entity, is completely external, independent of human thought or action, or else internal, a thought in our mind -we are not just saying
something about numbers, but about existence- that there are only two kinds of existence.
Everything is either internal or external; and given that choice, that polarity or dichotomy,
numbers do not fit -thats why it is a puzzle. The question is made difficult by a false presupposition, that there are only two kinds of things around. But if we are real, and ask what there
is around, for instance there is the traffic ticket we have to pay, there is the news on the TV,
etc. -none of these things are just thoughts in our mind, and none of them is external to human
thought or activity. They are a different kind of reality, thats the trouble. This kind of reality
has been excluded from Metaphysics and Ontology, even though it is well-known -the sciences of Anthropology and Sociology deal with it. But when we become philosophical,
somehow this third answer is overlooked or rejected.
Mathematics is neither physical nor mental, it is social, part of culture, part of history,
and it is like all those real things which are real only as part of collective human consciousness. Being part of society and culture, it is both internal and external: Internal to society and
culture as a whole, external to the individual, who has to learn it from books in school. That is
what Mathematics is. But for some Platonic mathematicians, that proposition is so outrageous
that it takes a lot of effort even to begin to consider it.
Hersh calls it humanistic philosophy of Mathematics. He uses the term humanism because it is saying that Mathematics is something human. There is no Mathematics without
people. Many people think that numbers are there whether or not any people know about
them; this is confusion. Humanistic philosophy of Mathematics faces Mathematics as part
of human culture and history. This philosophy lands Mathematics, makes it psychologically
accessible and increases the possibility for someone to learn it, because it is exactly one of the
things people do. He names his view social conceptualism too, because Mathematics consists of concepts, but not individually held concepts; socially held concepts. This view sounds
like an anthropic principle of Mathematics.
One can say There are nine planets; there were nine planets before there were any people. That means there was the number nine, before we had any people. We do see mathematical things, like small numbers, in physical reality; and that seems to contradict the idea
that numbers are social entities. We use number words in two different ways: as nouns and
adjectives. This is an important observation. We say nine apples, nine is an adjective. If it is
an objective fact that there are nine apples on the table, that is just as objective as the fact that
the apples are red, or that they are ripe, or anything else about them, that is a fact. And there is
really no special difficulty about that. Things become difficult when we switch unconsciously, and carelessly, between this real-world adjective interpretation of mathematical

On Numbers Nature

15

words like nine, and the pure abstraction. That is not really the same nine. Although there is
of course a correlation and a connection. But the number nine as an abstract object, as part of
a number system, is a human possession, a human creation; it does not exist without us. The
possible existence of collections of nine objects is a physical thing, which certainly exists
without us. The two kinds of nine are different. Like we can say a plate is round, an objective
fact, but the conception of roundness, mathematical roundness, is something else (Hersh,
1997).
According to Ernest (1995), the Philosophy of Mathematics is in the middle of a Kuhnian
revolution. For over two thousand years, Mathematics has been dominated by an absolutist
paradigm, which views it as a body of infallible and objective truth, far removed from the
affairs and values of humanity. Currently, this is being challenged by a growing number of
philosophers and mathematicians, including Lakatos, Davis and Hersh, Tymoczko. Instead,
they are affirming that Mathematics is fallible, changing and, like any other body of knowledge, the product of human inventiveness.
How Mathematics is viewed is significant on many levels, but nowhere more so than in
education and society. For if Mathematics is a body of infallible, objective knowledge then it
can bear no social responsibility. Thus, the underparticipation of sectors of the population,
such as women; the sense of cultural alienation from Mathematics felt by many groups of
students; the relationship of Mathematics to human affairs, such as the transmission of social
and political values; its role in the distribution of wealth and power; none of these issues are
relevant to Mathematics.
On the other hand, if it is acknowledged that Mathematics is a fallible social construct,
then it is a process of inquiry and coming to know, a continually expanding field of human
creation and invention, not a finished product. Such a dynamic view of Mathematics needs to
include the empowerment of learners to create their own mathematical knowledge; Mathematics can be reshaped, at least in school, to give all groups more access to its concepts, and
to the wealth and power its knowledge brings; the social contexts of the uses and practices of
Mathematics can no longer be legitimately pushed aside, the implicit values of Mathematics
need to be squarely faced. When Mathematics is seen in this way, it needs to be studied in
living contexts which are meaningful and relevant to the learners, including their languages,
cultures and everyday lives, as well as their school based experiences. This view of Mathematics provides a rationale, as well as a foundation, for multicultural and girl-friendly approaches to Mathematics. Overall, Mathematics becomes responsible for its uses and consequences in education and society.
Next Dehaene wonders What is a number? Dehaene, as a neurophysiologist studying
how the human brain wires itself to do Mathematics, answers that number is a parameter of
our physical environment which is extracted and processed by dedicated cerebral networks just like color, which is a subjective property entirely made up by brain area V4. Indeed, he
shows how animals and infants have a largely innate intuition about numerical quantities and
their properties. Recent experimental evidence suggests that: (i) The human baby is born with
innate mechanisms for individuating objects and for extracting the numerosity of small sets.
(ii) This number sense is also present in animals, and hence that it is independent of language and has a long evolutionary past. (iii) In children, numerical estimation, comparison,
counting, simple addition and subtraction all emerge spontaneously without much explicit
instruction. (iv) The inferior parietal region of both cerebral hemispheres hosts neuronal circuits dedicated to the mental manipulation of numerical quantities, and that a lesion to that

16

Dimitris Gavalas

area leads to a loss of number sense, including not knowing what is 3-1, or what number
falls between 2 and 4. This inner feeling of quantity serves as a foundation for the later construction of number through mathematical axiomatizations. Yet as a basic category of experience provided by a dedicated brain circuit, number is as undefinable as color, space,
movement, happiness, or beauty.
Dehaene agrees with Hersh that Platonism -the view that mathematical facts are abstract
and independent of human existence and knowledge- is not a tenable position. His neurobiological interpretation is that Platonism is a cognitive illusion that imposes itself upon so many
great mathematicians because with training, their brains develop a vivid, seemingly real, internal image of mathematical objects. Presumably, one can only become a mathematical genius if one has an outstanding capacity for forming vivid mental representations of abstract
mathematical concepts -mental images that soon turn into an illusion, eclipsing the human
origins of mathematical objects and endowing them with the semblance of an independent
existence.
Mathematics is indeed a product of the human mind and brain, and as such it is indeed a
very human enterprise, fallible, revisable, and highly dependent on the limits and abilities of
our cerebral equipment. Does that mean, however, that mathematics is a purely social activity? The trouble with labelling Mathematics as social or humanistic, and with comparing it
to art and religion, is that this view completely fails to capture what is so special about
Mathematics -first, its universality, and second, its effectiveness.
A mathematician can go to any place in the world and, given enough time, can convince
anyone about mathematical truths -3 is a prime number, or that the 3rd decimal of Pi is a 1 or
that Fermats Last Theorem is true. The point is universal agreement is often easily reached
about what constitutes a mathematical fact. This makes an unqualified relativistic, social,
Lakatosian, or post-modernist view of Mathematics totally untenable. Relativists notwithstanding, the value of Pi does not vary from culture to culture, nor does each culture have its
own different mathematical universality, which is one and only.
The other key difference between Mathematics and other cultural objects is its effectiveness. This was, and still is, a subject of awe and wonder for physicists like Wigner and Einstein. How is it possible that mathematics, a product of human thought that is independent of
experience, fits so excellently the objects of physical reality? Einstein asked in 1921. This is
bound to remain forever a mystery as long as you adhere to a strong relativistic position,
which asserts that Mathematics is the result of the arbitrary cultural choices of mathematical
churches. For that matter, indeed, the effectiveness of Mathematics is also not easy to explain if you believe, as Hersh seems to do, that mathematicians pursue their work for the sole
purpose of its abstract beauty.
Dehaenes solution to both of these riddles appeals to evolution -of the brain and of
Mathematics. In his opinion, mathematical objects are universal and effective, first, because
our biological brains have evolved to progressive internalize universal regularities of the external world -such as the fact that one object plus another object usually makes two objects,
and second, because our cultural mathematical constructions have also evolved to fit the
physical world. If mathematicians throughout the world converge on the same set of mathematical truths, it is because they all have a similar cerebral organization that: (i) Lets them
categorize the world into similar objects -numbers, sets, functions, projections, etc. (ii) Forces
to find over and over again the same solutions to the same problems. One can remind of the

On Numbers Nature

17

reinvention of place-value number notation in 4 different cultures widely separated in space

and time -Chinese, Babylonian, Maya and Indian.
The common structure of our brains can explain why there is trans-cultural convergence
in Mathematics despite different social perspectives. In that respect, mathematical objects,
while they are human constructions, are radically different from other cultural constructions
of Western societies such as religion, Nouveau roman, symphony orchestras etc. Pies come in
all sorts, Greek, American, or French; but hopefully therell always be only one number pi ()
(Dehaene, 1997b).
Dehaene claims that number is like color. Because we live in a world full of discrete and
movable objects, it is very useful for us to be able to extract number. This can help us to track
predators or to select the best foraging grounds, to mention only very obvious examples. This
is why evolution has endowed our brains and those of many animal species with simple numerical mechanisms. In animals, these mechanisms are very limited: they are approximate,
their representation becomes coarser for increasingly large numbers, and they involve only
the simplest arithmetic operations. Humans have also had the remarkable good fortune to develop abilities for language and for symbolic notation. This has enabled us to develop exact
mental representations for large numbers, as well as algorithms for precise calculations.
Mathematics, or at least Arithmetic and Number Theory, is a pyramid of increasingly more
abstract mental constructions based solely on (i) our ability for symbolic notation, and (ii) our
nonverbal ability to represent and understand numerical quantities. He argues that many of
the difficulties that children face when learning Mathematics and which may turn into fullblown adult innumeracy stem from the architecture of our primate brain, which has not
evolved for the purpose of doing Mathematics. It is his view that the human brain does not
work like a computer and that the physical world is not based on Mathematics -rather
Mathematics evolved to explain the physical world the way that the eye evolved to provide
sight.
What are numbers, really? If we grant that we are all born with a rudimentary number
sense that is engraved in the very architecture of our brains by evolution, then clearly numbers should be viewed as a construction of our brains. However, contrary to many social constructs such as art and religion, number and arithmetic are not arbitrary mental constructions.
Rather, they are tightly adapted to the external world. Whence this adaptation? The puzzle
about the adequacy of our mathematical constructions for the external world loses some of its
mystery when one considers two facts:
(i) The basic elements on which our mathematical constructions are based, such as
numbers, sets, space, and so on, have been rooted in the architecture of our brains by
a long evolutionary process. Evolution has incorporated in our minds/ brains structures that are essential to survival and hence to veridical perception of the external
world. At the scale we live in, number is essential because we live in a world made
of movable, denumerable objects. Things might have been very different if we lived
in a purely fluid world, or at an atomic scale, and hence we have to concur with other
mathematicians such as Poincare, Delbruck, or Hersh in thinking that other life forms
could have had Mathematics very different from our own.
(ii) Our Mathematics has seen another evolution, a much faster one: a cultural evolution.
Mathematical objects have been generated at will in the minds of mathematicians of
the past thirty centuries (this is what we call "pure mathematics"). But then they have

18

Dimitris Gavalas
been selected for their usefulness in solving real world problems, for instance in
Physics. Hence, many of our current mathematical tools are well adapted to the outside world, precisely because they were selected as a function of this fit.

Many mathematicians are Platonists. They think that the Universe is made of mathematical stuff, and that the job of mathematicians is merely to discover it. Dehaene strongly denies
this point of view. This does not mean, however, that he is a social constructivist. On the
contrary he believes that mathematical constructions transcend specific human cultures. In his
view, however, this is because all human cultures have the same brain architecture that resonates to the same mathematical tunes. The value of Pi does not change with culture. Furthermore, he is in no way denying that the external world provides a lot of structure, which
gets incorporated into our Mathematics. He only objects to calling the structure of the Universe mathematical. We develop mathematical models of the world, but these are only models, and they are never fully adequate. Planets do not move in ellipses -elliptic trajectories are
a good, but far from perfect approximation. Matter is not made of atoms, electrons, or quarks
-all these are good models, indeed, very good ones-, but ones that are bound to require revision some day. A lot of conceptual difficulties could be clarified if mathematicians and theoretical physicists paid more attention to the basic distinction between model and reality, a
concept familiar to biologists (Dehaene, 1997c).
Finally, according to Cybernetics, number is a conceptual scheme, an abstraction of the
second level from specific numbers. The abstraction procedure to recognize specific numbers
is counting. Counting is based on the ability to divide the surrounding world up into distinct
objects. This ability emerged quite far back in the course of evolution; the vertebrates appear
to have it in the same degree as humans do. The use of specific numbers is a natural integrated description complementary to the differential description by recognizing distinct objects. This ability would certainly be advantageous for higher animals in the struggle for existence. And cybernetic apparatus for counting could be very simple -incomparably simpler
than for recognition of separate objects in pictures.
Yet nature, for some reason, did not give our brain this ability. The numbers we can directly recognize are small, up to five or six at best -though it can be somewhat extended by
training. Thus the number 2 is a neuronal concept, but 20 and 200 are not. We can use them
only through counting, creating artificial representations in the material external to the brain.
The material may be, and was historically, fingers and toes, then pebbles, notches etc., and
finally sophisticated signs on paper and electronic states of computer circuitry. For theoretical
purposes the best is still the ancient-style representation where a chosen symbol, say 'I' stands
for one object. Thus 2 is 'II', and 5 is 'IIIII' (Principia Cybernetica).
From the above report, it is obvious that there is a vivid developing research concerning
the neurological and social basis of the human mathematical ability. It is also obvious that this
research has not yet arrived at definite conclusions. When this happens we will be able to explain in detail both how the human brain and society process the mathematical and especially
numerical notions and magnitudes.

On Numbers Nature

19

7. Number and Category Theory

7.1. First Definition of a Category
A category consists of a collection of objects and a collection of morphisms. Every
morphism f has a source/ domain object and a target/ co-domain object. If the source of f
is X and its target is Y, we write f: X Y. In addition, we have:
(1) Given a morphism f: X Y and a morphism g: Y Z, there is a morphism gf: X
Z, which we call the composite of f and g.
(2) Composition is associative: (fg)h = f(gh).
(3) For each object X there is morphism 1: X X, called the identity of X. For any f:
X Y we have 1X f = f 1Y = f.
We should visualize the composite of f: X Y and g: Y Z as follows:
XfY

gf
g

Z.
This is a first definition of a category.

7.2. Existence of a Natural Numbers Object

Category and Topos Theory is based only on objects and arrows/ morphisms between
them. So, for an arbitrary Topos E the Axiom of Infinity states that there is an object N of E
with arrows 0 s: 1 0 s (1) such that for any object X of E with arrows x and
f: 1 x X f X there is a unique arrow h which makes the following diagram commute:
1 0 s
h
1 x

X f

(2).
X

The object N is then called a natural numbers object (N.N.O.) for E. For such a N.N.O. N
the definition states that the diagram (1) is universal among diagrams of the form 1 x X
f X. It readily follows that N, together with the arrows 0 and s, is unique up to isomorphism. We can thus speak of the N.N.O. of a Topos E, if there is one. In Set the usual set of
all natural numbers N={0, 1, 2, } has the required universal property for a N.N.O. as in (2),
where the arrow 0 sends the one element of 1={0} to 0N, while s is the usual successor
function n n+1. Given a set X, an element xX and a function f: X X, the arrow h
uniquely provided by (2) thus satisfies h(0)=x, h(n+1)=f(h(n)) (3). In other words, h is de-

20

Dimitris Gavalas

fined from x and f by recursion or as one often says by induction. (Mac Lane, 1986; Mac
Lane & Moerdijk, 1992).

7.3. Natural Numbers Object in ETCS (Elementary Theory of the Category of Sets)
Since Dedekind raised the issue for modern Mathematics, some claim that mathematical
objects should have only relevant properties. So, natural numbers should have only arithmetic
properties. Every ZF set has non-arithmetic properties given by its particular elements. So, on
this view numbers cannot be ZF sets. But ETCS models of Arithmetic have no irrelevant
properties. So, this view could say the numbers form an ETCS set (Benacerraf, 1965).
Specifically, recursion data on a set X means an element x of X and a function f: X
X. We define a natural number object to be a set N with recursion data 0 and s: N N,
called zero and successor, such that: For any recursion data x, f on any X there is a unique
function h: N X with h(0) = x, h(sn) = f(h(n)), for all nN. Provably in ETCS there are
infinitely many different natural number objects, but all are isomorphic. None has any properties but the shared ones which follow from the definition (McLarty, 1994). Benacerraf and
McLarty show the structuralism of ETCS.

8. Number and Non-Standard View

The existence of non-standard models for mathematical structures implies from Lewenheim-Skolem Theorems. It is also known from Gdels Theorems that the intuitively clear
system of the natural numbers can not be fully characterized axiomatically. Therefore, the
axioms about N do not conceive the intuitive depth of the natural numbers; consequently is
given the chance to various intuitive aspects about N to coexist having simultaneously their
axiomatic analogue. It is generally accepted for instance that the smaller the natural numbers
are the more familiar man is with them; conversely the bigger they become the less accessible
and difficult to be understandable.
Let us consider a natural model in order to better comprehend the fuzzy observable nature
of the predicate standard. Lets assume that we have a series of cards:

We paint the first card white using a specific quantity of white color. Next, we add a tiny
bit of black color, so as we cannot discern, with our observing abilities, cards 1 and 2 in relation to color. We continue in this way until we have a completely black card and we continue
to paint in black all the following cards. Hence, the predicate standard determines the white
cards that are literally the cards which site before the completely black card, which in their
turn can mean our familiarization with the corresponding natural numbers. Therefore, when
we say standard natural we mean here white number and when we say non-standard natural we mean black number and non-accessible. We must underline here that there is not
maximum white card neither minimum black card. We can consider also that standard
natural means graduated finite white number and non-standard natural means graduated

On Numbers Nature

21

infinite black number. So the natural numbers can be presented as a tape where the white
color becomes gradually black:
standard

non-standard

4 ...

n ...
fuzzy borders

We make the following assumptions:

(i) 0 is standard natural number.
(ii) For every n, n is standard natural implies n+1 is standard.
(iii) There is a non-standard natural .
(iv) The Principle of Induction also holds: If is any logical expression, then
((0) and (n) (n+1) for every standard n), implies that formula also holds
for every standard natural (Nelson, 1977, 1988).
We have to note that (i) and (ii) indeed hold in the colorful tape: 0 is white and if a card is
white, then its successor is also white, since observably are indiscriminate. Also from (ii) is
obvious that there is no minimum non-standard that is minimum black card, neither maximum standard that is maximum white card. Hence it is obvious that standards are separated
by non-standards with fuzzy limits and therefore standards cannot be set of ZFC, but essentially are a fuzzy set or differently a class or semiset, according to Vopenkas (1979) terminology.
The local non-standard regard of N has as consequence the local non-standard regard of
many structures and particularly of R. A basic ambiguity is imported thus in Mathematics,
which is covered behind the initial not-definable predicate standard. Same observations that
are made for N are also in effect for R, that is to say the axioms of R are not capable to conceive the intuitive perception we might have for this.

9. Number and Fuzziness

According to Kosko (1994), everything is a matter of degree; the same holds for numbers, which are the hallmark of precision. The question then is: Can we fuzz these up too?
Are fuzzy numbers possible? The old answer is no; how could numbers be fuzzy? Numbers
are pure forms and they either are or are not, they belong all or none to the well-known sets
of numbers. Every whole number is either even or odd, no in between, no middle ground, no
gray cases. Mathematicians and scientists have devised Mathematics that seems to have escaped the sloppiness of everyday fuzzy sets and thoughts. Where could fuzziness hide in the
black-and-white world of Mathematics? If we consider number zero (0) we can find it at a
glance as a spike on a number line. The spike means the number zero belongs 100% to the

22

Dimitris Gavalas

set, say, ZERO and no other number belongs to it. Every number is either in the ZERO set or
out of it; all or none. In this set sense the number zero alone belongs to the set ZERO. But
what about numbers close to, almost, nearly zero? These numbers are fuzzy numbers, they
define a spectrum of numbers near zero and some belong more in the set than others. The
closer a small number to zero is, the more it belongs to the fuzzy set of small numbers. The
number 1 is closer to 0 than the number 2 is, and 2 is closer than 3 is, and so on. The number
0 belongs 100% to the set ZERO but close numbers may belong only 80% or 50% or 10%.
We might draw the fuzzy number zero as a triangle centered at the exact number 0.
If we draw the triangle narrow enough, we get back the spike of classical Mathematics.
That is another surprise: Mathematics as we know it is but a special case of fuzzy Mathematics, a special limiting case -the degenerate case of black-and-white extremes in a mathematical world of grays. We can add and subtract triangles just as we add and subtract spikes/
numbers. We can also draw the fuzzy number ZERO in infinitely many ways; each one can
draw it differently just as each one thinks differently of HOUSE, SMALL, SMART, FAIR, NICE or
CLEAN. There are as many ways to draw the fuzzy number ZERO as there are numbers.

Exact/ Non-fuzzy Number Zero (0) and Inexact/ Fuzzy Number Zero (-2, 0, 2).

What can we do with fuzzy numbers? We can reason with them; we do it all the time.
Fuzzy Logic means reasoning with fuzzy numbers and sets. The knowledge or intelligence
comes from associating fuzzy events of everyday. For now we have proved the point: Numbers are fuzzy too. We work with fuzzy numbers all the time and, if numbers are fuzzy, then
everything is and indeed is.
So, according to the above view, the consideration of a number can be double: Classical/
exact or fuzzy/ inexact. If we take the horizontal axis of the real numbers, then we can add the
vertical axis up to 1 and transform the exact representation of the number to fuzzy representation using a membership function, for example triangular, trapezoid, conoid etc.
Consequently, a fuzzy number can be represented by a triangle, the apex of which lies
exactly above the number in one unit height (1) and its basis lies on the horizontal axis. The
classical/ exact eight (8), for example, has a basis limited to a point, number 8, and therefore
is represented by a vertical arrow sited on point 8 and of one unit height. Hence, while exact 8

On Numbers Nature

23

can be considered as a unique triad of the form (8,8,8), the fuzzy 8 has the possibility for
many triangles such as: basis interval (7,9) or (6,9) according to data and then the fuzzy
numbers are correspondingly (7,8,9) and (6,8,9). The ordered triad which represents a fuzzy
number has as its centre the apex of the triangle and therefore the corresponding exact number. The first number shows the left point of the basis and the third one shows the right point.
For example, the (3,5,7) is the fuzzy number 5 with basis the interval (3,7).
In conclusion, we would say that a fuzzy number is a quantity the value of which is not
accurate, as it happens with usual numbers, but can be considered as a function with domain
the set of the real numbers and co-domain the interval of the reals between 0 and 1. To each
numerical value in the interval is attributed a particular membership degree, where 0 represents the minimum possible grade and 1 the maximum. Fuzzy numbers on many aspects represent the natural world more realistic than the usual numbers. Fuzzy numbers are used in
Statistics, Programming and Science. The concept of fuzzy number considers the fact that
natural phenomena have a degree of uncertainty. A way then to represent uncertainty is to use
fuzzy numbers and sets. In some cases, a number can be known only approximately or inexact even though its exact value is known. Such paradigms are well-known in Programming.

PART III: JUNG AND THE CONCEPT OF ARCHETYPE

10. Chthonic and Celestial Mathematics
There is a common sense of antithesis between, on the one hand, the Superior World of
ideal mathematical objects, archetypal ideas, pure relationships, qualities and the like (Celestial Mathematics) and, on the other hand, the current inferior world of material objects,
quantity, amplitude, application, counting (Chthonic Mathematics). This is an important and
basic antithesis. This basic antithesis looks impressive and may be not so random. If we also
base on this sense of antithesis, we can say that there are some conclusions and some messages are transmitted by the nature of this situation.
Indeed, if we consider as horizontal plane the usual human empirical consciousness,
which generally conceives objects in motion into a time-spatial continuum and deals with
them and their governing laws, then this plane is crossed by a vertical plane and the only
thing we can say about this order is that it is, on the one hand, something mathematically abstract and, on the other hand, legend, sacred and mysterious.
If we face the number as revelation/ discovery and not as invention for measuring, then
because its nature is mythological it belongs to the region of divine forms and it is as archetypal as they do. But unlikely to them, it is also real, in the sense that we meet it on the
region of human experience as quantity. Therefore, it constitutes the bridge and the intersection between the natural, real, outside world and the ideal, imaginary, inner world.
This means that it participates in both worlds. It is a part of the material and a part of the
psychic. It does not only count, neither is it quantitative and static. It also denotes the quality
and it is a dynamical schema and for this reason it is an entity, which is in part revelation and
in part invention. This exact double nature of the number makes it symbolic and a vehicle of
mental and psychic processes.

24

Dimitris Gavalas

11. Jungs - Paulis General Hypothesis of Archetypes

Jung wrote that number is the key to the mystery, since it is just as much discovered as it
is invented. It is a quantity as well as a meaning. He understood number to be the most primitive element of order in the human mind and defined number psychologically as an archetype
of order which has became conscious. Von Franz (1974) undertook an exhaustive investigation of number archetype acting as dynamic organizing principle in both psyche and matter.
She published her findings in Number and Time and her work represents a significant extension of the General Hypothesis of Archetypes of Jung-Pauli.
Indeed, according to Jungs - Paulis General Hypothesis of Archetypes (Card, 1991,
1993, 1995, 1996; Cohen, 1975; Jung 1957-1977; Jung & Pauli, 1955; Meier, 1992, 2001;
Nagy, 1991; Stevens, 1982), the frames of mind and matter -psyche and nature- are complementary views of the same transcendental reality which they call unus mundus. The archetypes function as fundamental dynamical patterns, the various representations of which characterize all procedures either mental or physical. In the psychic realm the archetypes organize
the ideas and the images, while in the physical one they organize the structure and the transformations of matter and energy and also are responsible for the order. The simultaneous action of the archetypes both in the psychic and in the physical realm is responsible for the
cases of synchronistic phenomena, that is phenomena which are linked acausally, but have the
same or similar meaning for the observer (Peat, 1987).
Later on, von Franzs (1974, 1992) work on the number archetype cleared up and extended Jungs - Paulis General Hypothesis of Archetypes as follows: All mental and physical
phenomena are complementary aspects of the same unitary, transcendental reality. At the basis of all physical and mental phenomena there exist certain fundamental dynamical forms or
patterns of behavior which may be called number archetype and they are manifestations of the
later. Any specific process, physical or mental, is a particular representation of certain of this
archetype. In particular, the number archetype provides the basis for all possible symbolic
expression. Therefore, it is possible that a neutral language/ hyper-code constructed from abstract symbolic representations of the number archetype may provide highly unified, although
not unique, descriptions of all mental and physical phenomena. This view leads to the potential of arithmetization/ mathematization of the totality of psychophysical phenomena, which
are the study matter of Cognitive Science (Warner & Szubka, 1995). The parallel, also, between von Franzs search for primal archetype inherent in numbers and Chomskys search for
linguistic universals has been examined by the physicist Card (2000).
Von Franz found a quaternary of order archetypes connected immediately with the first
four numbers. The representations of these archetypes are very often dynamical in nature. She
condenses their characteristics as follows: Numbers become then typical dynamical patterns,
for which we have to say the following: One (1) comprises totality. Two (2) divides, repeats
and generates symmetries. Three (3) gives center to the symmetry and introduces linear succession. Four (4) functions as a stabilizer by returning to one and it presents four aspects creating limits. The fourth order archetype is interesting since its various representations concern
feedback and self- reference. Self- reference is a repeating subject in the development of Set
Theory and Symbolic Logic that led to Gdels Incompleteness Theorem. We have to note
here that it plays an important role in the chaotic dynamics of non-linear systems. Eventually,
the well-known mathematician and author Ian Stewart does not exclude the fact of being ar-

On Numbers Nature

25

chetypal patterns of numbers, a common set of ideas in many minds for which our intellect is
predisposed to work and thanks to them to function.
Since archetypes precondition all existence, they are manifest in the spiritual achievements of science, art and religion, as well as in the organization of organic and inorganic matter. The archetype thus provides a basis for a common understanding of data derived from all
sciences and human activities -not least because of its implications for epistemology.

12. Number as Archetype

Number is an archetype of order that has become conscious, Jung

In the platonic dialogue Menon, Socrates addresses questions to slave Menon and he
shows thus, beginning from a particular square, how is constructed another one having the
double surface. Menon recognizes that an obvious construction is not correct and afterwards
he recognizes the correct one. Then Socrates results to the following conclusion: The knowledge he has now either he acquired it some time in the past or he had it all along. If he had it
all along, he also knew all these all along, but if he acquired it some time in the past, it is not
possible he acquired it in his present life. Did someone teach him Geometry? Because he will
do the same he just did both for all Geometry and for each other discipline of knowledge. But
did anyone really teach him all these? The answer is no because Menon is an illiterate
slave. This experiment shows two things: on the one hand that there are three stages in order
for someone to reach knowledge. Firstly, knowledge is unconscious. Next, conjecture and
opinion rouse up with questions and through the dialectic approach. Finally conjecture and
opinion are transformed into knowledge through the understanding of the relation that exists
between the cognitive object and the idea/ form from which it comes. On the other hand that
we know truths we never learned externally, that is from education and experience. This
knowledge is a pattern/ model of the universal truths which we can conceive and acknowledge. Finally, there is a higher hierarchical level of the absolute knowledge and truth, fountain-head of the absolute knowledge and truth, fountain-head of the knowledge of Good (Hill,
1992).
The modern research detects the existence of innate models, universal dynamical
schemes, which are well-known as archetypes and are different from the platonic ideas/
forms. Indeed, what Plato called idea, is a model of ultimate perfection only with the positive/
bright meaning, while archetype is bipolar and realizes both the positive and the negative/
shadowy side, because it is not a pattern of perfection, but of wholeness. In addition, it is accompanied by emotional charge. Therefore a kind of knowledge based on innate principles
mostly, not acquired or resulting from the external experience, is the system of archetypes.
The study of archetypes and their structure, like the study of numbers, can lead to better understanding of the typical properties of mind. The archetypes are structures, models in the
mental realm, norms and limitations, patterns of mental and psychic behavior in the unconscious, giving the possibility to unconscious messages to be sent to the conscious mind with
the form of images.
The nature of these messages is symbolic. The archetypes correspond to a universal
grammar of mind. They predispose it towards particular kinds of universal knowledge, assuring that some issues are universal, like myths, numbers, language structures, ideas/ forms.

26

Dimitris Gavalas

The archetypes provide the potential and the circumstances for the production of ideas and
fantasy. They are not found in the mind as result of information, practice or teaching, but instead are a pan-human legacy. An archetype is not immediately known, but becomes known
only through the emergence of the archetypal image, presented spontaneously. Because the
archetypal information has a universal basis, out of the particular space-time, it is independent
on specific experience. Our coincidence with our conscious and the rational rejects, repulses
or represses archetypes and their images, but these always return with one form or another.
However, this form is always inside a frame and not completely free, i.e. it is not free of being
any form. Archetypes in general have their own special logic and structure and these special
elements illustrate the structure of mind.
The above mentioned issues lead us to think that the opposition between the Superior
World of Pure Mathematics and the Inferior World of Applied Mathematics is not absolute.
These two Worlds are seemingly incompatible, because the connecting bridge is not missing.
Between them, as common factor, stands a medium, the number, the reality of which is valid
in both Worlds, because it is exactly an archetype in its essence, in its ultimate reality. Number belongs to both Worlds real and fantastic; it is visible and invisible, quantitative and
qualitative. The fact that number participates and characterizes the nature of the medium form
and that appears as an interceder between the basic opposition and all the others, is a very
important fact. Because we immediately conclude that the symbol realizing the intermediation, the conjunction, the wholeness, the unity, the order, the identity, the synthesis, the reconciliation can be generally expressed by a mathematical fashion. The meaning of this symbol necessarily presupposes an excess which does not lead to some metaphysical hypothesis.
It is just a borderline meaning and the fact that exists something beyond this cognitive limit is
proved by the spontaneous appearance and observation of archetypes and more clearly by
number. The latter, on the one side of the limit is quantity and on the other side autonomous
entity capable of qualitative denotations expressed in a priori structures of order, regularity
and eurhythmy.
With the existence of a factor, mediating between the two phenomenally incompatible
Worlds, we know for sure today that the one World takes through this exact factor, properties
of the other World. With each others help the two Worlds interact continuously and even
more the ultimate reality bases on a common substratum, still unknown, having properties
simultaneously from the two Worlds which gives us the chance to build a new model of the
World, closer to the idea of the unus mundus. It seems that the method best fits to the nature
of order and chance is arithmetic. From ancient years people used numbers to express the
meaning of order, as well as coincidences with meaning, the ones that can be interpreted.
Also if we denude a set of objects from all of its properties and its characteristics, that is we
consider an abstract set, what finally remains is its multitude, that is its number, a fact the
denotes at least that number is a magnitude initial and irreducible.
The succession of natural numbers seems to be something more than a simple accumulation of identity monads. It includes the set of Mathematics and all that are going to be discovered in this field. That is why number is in a sense non-predictable entity. Number helps more
than anything else to the ordering of the chaos of phenomena. It is the predefined tool for the
creation of order or to become understandable some already existing, but still unknown
smooth arrangement. May be it is the most primitive element of order of the human mind.
The fact that numbers have an archetypal foundation, it makes obvious that it is not so arbitrary to define number as an archetype of order which became conscious by man. Even the

On Numbers Nature

27

intuitionist/ constructivist Brower said that Mathematics refer to becoming and not to being, a
view completely platonic. It is a spiritual action, which is in direct relation with the possibility
of conceptual construction of all mathematical entities from the natural numbers.
People believe in general that numbers were invented or were the result of thinking and
for this reason they are nothing else but concepts of quantities, the content of which was attributed by the human intellect. But it is equally possible the numbers to be discovered. In this
case they are not just concepts, but something more: autonomous entities which contain, in a
way something more than mere quantities. Contrary to the concepts, they do not base in a
hypothesis, they are un-hypothetical according to Plato, but to the property of identity, i.e.
they are themselves, something that cannot be expressed by a conscious intellectual invention.
Under these circumstances it is possible to attribute to them easily some properties which are
not yet been discovered. We tend towards the synthetic view that numbers both were invented
and discovered and therefore have an autonomy analogous to that of the archetype. Consequently, they preexisted of consciousness and that is why they determine it than being determined by it. Also, the archetype as an a priori ideal form both was invented and discovered. It
was discovered because nobody knew its unconscious, autonomous existence and was invented because its presence was the effect of analogous human psychophysical structures. It
seems though, that natural numbers have archetypal character. Some numbers and combinations of numbers relate to or act on particular archetypes. The opposite is real also, as the research of Jung, von Franz etc assure.
The set of natural numbers is an archetype of order, which is illustrated in consciousness,
helping it to place the chaos of phenomena into some sort of ordering diagram. The system of
natural numbers is a form of a priori knowledge, extremely capable of organizing the random
nature of experience. Exactly like the native language it is being learned easily, because different kinds of numerical systems are less accessible. They were not named natural numbers
without special reason. Number which is used for counting and the quantitative relations is
something more than we generally believe. It is simultaneously an entity of the same order
with the mythological elements that is why the Pythagoreans thought it sacred. But when we
use it only for practical purposes, we do not seem to realize that aspect.
The archetypal structures and numbers as such, are not static forms but dynamic elements. We perceive the special power of archetypes and numbers when we have the opportunity to appreciate the appeal, that is the emotion through the unconscious or the awe or the
fear that exert on us. They seem fateful and in this case we can think as example numerophobia, stress for Mathematics, superstitions, arithmomancy etc. It is characteristic the issue of
number 13, which is taboo for many people, it has unfavorable character and the superstitious
feels awe for this number. Archetypes are images and simultaneously emotions and we can
talk about archetypes only when these two aspects are presented simultaneously. When the
image is charged with emotional intensity it acquires mysterious, divine dimension, it becomes dynamic and it necessarily has impacts on the person. The fact that number seems to
be initial and innate concept for man, which is accompanied by emotion, is interpreted by
Gauss sayings: It is curious that all people who study seriously the science of Number Theory, are dominated by a kind of passion for it. Cantor also confesses: When I think and
study infinite, it always follows a real pleasure, in which I succumb pleasantly as I see that
the concept of the integer number is split in two concepts and we ascend to the infinite. The
one is the concept of power (cardinality) and the other the concept of counting.

28

Dimitris Gavalas

Even those contents that are considered the most certain of the consciousness are surrounded by the shadow of uncertainty and vagueness. And the most austere mathematical
concept, that we believe it contains only those elements we attribute to it and is well defined
and limited, it is something more than we believe it is. It is simply a psychic fact and as such
it is partially unknown just like number. Of course these faint and implicate differentiations
are being abolished by being considered needless, ignorable, not worth a fig or they do not
have relation to our needs and their usual applications. Even though we face exact definitions
and clear analysis, we discover sometime the most unexpected differentiations, not only with
the pure mental concept of the term, but with the value that is attributed to number as well as
with its application. The fact that these differentiations exist displays exactly what was originally sustained.
Weyl believed about natural numbers and used to say that he could not understand how
something so simple, constructed by the human mind, could contain something abysmal. He
should ask himself first, whether the human mind really constructed them. He feels that he
controls and leads completely the phenomenon, but this is not true. We claim that what happens is exactly the opposite that is numbers as archetypal dynamic processes control us at
least up to a point. Weyl by the term abysmal means that irrational factor, which is contained in all numbers, and which we cannot conceive and explain with rational terms.
It is very interesting here to see what a psychologist of the Jungian school, von Franz
(1980), exactly says about this issue:
I want to read you in detail what the well-known mathematician, Hermann Weyl, says in
his book Philosophy of Mathematics and Natural Science. You know that until about 1930 the
great and passionate occupation of most mathematicians was the discussion of the fundamentals. They hoped, as has been the fashion nowadays, to re-discuss the fundamentals of all science. But the famous German mathematician, David Hilbert, created a new construction of the
whole building of Mathematics, so to speak, and hoped that this would contain no internal
contradictions. There would be a few basic axioms on which one could build up all branches
of Mathematics: Topology, Geometry, Algebra, and so on; it was to be a big building with
solid foundations in a few axioms. That was in 1926, and Hilbert was even bold enough to
say: I think that with my theory the discussion of fundamentals has been forever removed
from Mathematics.

Then in 1931 came another very famous mathematician, Kurt Gdel, who took a
few of those basic axioms and showed that one could reach complete contradictions with
them: Starting from the same axioms, one could prove something and its complete opposite. In other words, he showed that the basic axioms contain an irrational factor which could
not be eliminated. Nowadays in Mathematics one must not say that obviously this is so-andso and that therefore that and that are also so, but: I assume that it is so-and-so, and if so then
that and that follow. The axioms must be presented as assumptions, or must be postulated,
after which a logical deduction can be made, but one cannot infer that what has been assumed
or postulated could not be contradicted or doubted as an absolute truth.
In order to make such assumptions, Mathematics is generally formulated in such terms
as: It is self-evident or It is reasonable to think -that is how mathematicians posit an
axiom nowadays, and from there they build up. From then on there is no contradiction, only
one conclusion is possible, but in it is reasonable to assume that is where the dog lies buried, as we say. Gdel showed that, and thus threw over the whole thing. Strangely enough that

On Numbers Nature

29

did not reopen the discussion of fundamentals. From then on, as Weyl says, nobody touched
that problem, they just felt awkward and scratched behind their ears and said, Don't lets
discuss fundamentals, theres nothing doing: it is reasonable to assume, we cannot go beyond
that and there the situation rests today.
Weyl, however, went through a very interesting development. At first he was very
much attracted by the physicist Werner Heisenberg. He was very much of a Pythagorean
and was attracted by the numinosity and irrationality of natural integers. Then he became fascinated by David Hilbert, and in the middle of his life had a period during
which he became more and more attracted by Hilbertian logic and dropped the problem of numbers, treating them, erroneously as I think, as simply posited quantities. He
says, for instance, that natural integers are just as though one took a stick and made a
row of marks, which one then named conventionally; there was nothing more behind
them, they were simply posited by the human mind and there was nothing mysterious
about them; it was reasonable and self-evident that one could do that. But at the end
of his life he added -only to the German edition of his book on the philosophy of Mathematics, and shortly before his death- this passage:
The beautiful hope we had of freeing the world of the discussion of fundamentals was
destroyed by Kurt Gdel in 1931 and the ultimate basis and real meaning of Mathematics are
still an open problem. Perhaps one makes Mathematics as one does music and it is just one of
mans creative activities, and though the idea of an existing completely transcendental world
is the basic principle of all formalism, each mathematical formalism has at every step the
characteristics of being incomplete (which means that every mathematical theory is consistent
in itself but is incomplete, at the borders are questions which are not self-evident, are not
clear, and are incomplete) in so far as there are always problems, even of a simple arithmetical
nature, which can be formulated in the frame of a formalism, but which cannot be decided by
deduction within the formalism itself.

That is put in a mathematicians complicated way; put simply, it means that I daresay it is
self-evident, by which I posit something irrational, because it is not self-evident. Now one
could make an uroboros movement and say: But from my deduction I can reprove my beginning. You cannot! You cannot from the deductive formalism afterwards deduce a
proof, except by a tautology, which naturally is not allowed, even in Mathematics.
We are therefore not surprised that in an isolated phenomenal existence a piece of nature surprises us by its irrationality and that one cannot analyze it completely. As we have
seen, Physics therefore projects everything which exists onto the background of possibility or
probability.

That is important because it sums up in one word what modern science does. In other
words, any fragment of phenomenal existence, let us say these spectacles, contains something
irrational which one cannot exhaust in physical analysis. Why the electrons of these millions
and millions of atoms of which my spectacles consist are in this place and not in another, I
cannot explain; therefore through Physics, when it comes to a single event in nature, there is
no completely valid explanation.
The single event is always irrational, but in Physics one proceeds by projecting this onto
the background of a possible, i.e., one makes a matrix. For instance, in these spectacles there

30

Dimitris Gavalas

are so many atoms and so many particles of them, and so on, and out of a whole group one
can make a mathematical formula in which one could even count the particles -not 1, 2, 3, 4,
5, but by projecting onto the background of what is possible. That is why these matrices are
nowadays used in engineering and so on, because one can cope with the uncountable; they
provide an instrument with which to cope with the things which cannot be counted singly.
Weyl says:
It is not surprising that any bit of nature we may choose (these spectacles or anything)
has an ultimate irrational factor which we cannot and never will explain and that we can only
describe it, as in physics, by projecting it onto the background of the possible.

But then he continues:

But it is very surprising that something which the human mind has created itself, namely
the series of whole natural integers (I told you that he has this erroneous idea that the human
mind created 1, 2, 3, 4, 5, by making dots), and which is so absolutely simple and transparent to the constructive spirit, also contains an aspect of something abysmal which we cannot grasp.

That is the confession of one of the most remarkable -because one of the most philosophically oriented- modern mathematicians, Hermann Weyl. We can naturally say that we
do not believe what he believed, namely that the natural integers simply represent the naming of posited dots, therefore to us it is not surprising that natural integers are abysmal
and beyond our grasp. He believed that, and that is why he could not understand. It is
incredible that it should be so, but it is so; in other words, because the natural integers
have something irrational (he called it abysmal) the fundamentals of Mathematics are not
solid, because the whole of Mathematics is ultimately based on the givenness of the series of
natural integers.
Now precisely because numbers are irrational and abysmal -to quote Weyl- they are a
good instrument with which to grasp something irrational. If one uses numbers to grasp
the irrational, one uses irrational means to get hold of something irrational, and that is
the basis of divination. They took those irrational, abysmal numbers which nobody has
so far understood, and tried to guess reality or their connection with reality (von Franz,
1980).
This other factor is illustrated very well on the mythological aspects of number, which
are depicted among the others to the anaglyphs of ancient peoples, like the Mayas about 700
B.C, when the numerical subdivisions of time were personalized by gods. Here characteristic
is the fact that Kant reduces the concept of number to that of time, because the numeration is
in immediate relation with some time evolution. Also, according to the known, the pyramid
with the spots presents the Pythagorean tetraktyn (quaternary number). It consists of the
most used natural numbers 1,2,3,4 which form the sum 10. 4 and 10 were for the Pythagoreans essentially sacred numbers.
The archetype, as well as number, is not just a word or concept, it is literally a part of life
itself, an undivided part of the vivid entity and this happens through emotion. While we do
not perceive the special emotional tone of the archetype, we can intellectually prove that it
has a meaning or that it has not. But the archetype starts living only when we try patiently to

On Numbers Nature

31

discover why and how it has meaning for the particular vivid person. While we consider the
archetype as a mere image or idea, the divine power of which we never felt, we will talk
without knowing for which thing we talk and the words we use are empty and of no value.
The archetype gets vivid only when we try to understand its divine existence and its relation
to the person. Only that moment do we start to understand that the word archetype means
some particular things and that everything depends on the way these are connected to us, in
other words on our stance towards it.
The field Jung considered as the most fertile for future research was that of the basic
mathematical axioms, the first mathematical intuitions, paying special attention to the idea of
the infinite series of the natural numbers and the continuum in Geometry. As Arendt (1958)
says, contemporary Mathematics becomes the science that studies the structure of the human
spirit. In this line of thought it was also discovered that our representations are ordered before we even realize it. Van der Waerden (1975), who refers many paradigms of elementary
mathematical intuitions that come from the unconscious, concluded that the unconscious is
capable not only of taking in but also of combining and judging. The judgment of the unconscious is intuitive, but when the circumstances allow it is perfectly certain.
Among the many initial mathematical intuitions or a priori ideas, the natural numbers are
the most interesting ones. These numbers are not only useful for counting or executing numerical operations, but also for centuries were the unique means man had in order to decipher
the meaning of the various divination techniques and oracles. Of such kind were astrology,
numerology, and geomancy etc, which base on the numerical calculation and were interpreted
by Jung in terms of synchronicity that is a-causal coincidence bearing meaning to the observer. Besides, the natural numbers are certainly archetypal representations, because we are
obliged to bring them in our mind in a particular way. No one for example denies that two (2)
is the only and first number which means a couple, even if one has not thought of it consciously. In other words numbers are not concepts invented consciously by man just to execute calculations. It is obvious that they are spontaneous, autonomous creations of the unconscious, just like the rest archetypal symbols. The natural numbers though, are also properties
of the external objects: We can assure and calculate that here are two stones and three trees.
When we denude the objects from their properties, what is left is the common property of
being together many of them that is a particular magnitude. But the same numbers are undoubtedly part of our psychic composition; they are abstract concepts we can study without
reference to external objects. Therefore numbers are presented to us as a concrete association
between the two fields: nature and psyche.
Therefore number is revealing: it expresses the total World and translates logically the
hidden relations of its elements, on its nature is irrational, is reduced to the wholeness, but
when it is analyzed and developed it reveals universal relations, movements and positions. In
the universe, number is greater than the natural magnitudes, because the latter are expressed
by number. Number de-materializes World and its powers, reduces the empirical thought to
pure contemplate, to noesis, it reaches the human intellect in its origins, since number belongs
neither to time nor to space, since monad, contains everything and is contained in everything.
According to the Pythagoreans, we thus reach the depths of the myth; we touch what can be
called mystery of the mysteries. It is characteristic the observation that human intellect places
everything in a finite World. Mind can assert truths, which the human intellect cannot confirm
(Gdel, undecidable propositions). Intellect is the capability of counting and numbering, mind
is the realizing of number itself.

32

Dimitris Gavalas

13. Jung and Cantor - Gdel

13.1. Cantors Continuum Hypothesis
Cantor develops Set Theory introducing the cardinal number, card A, of a set A. The cardinal numbers are linearly ordered. Cantor proves that the set R of real numbers is not denumerable; this means that card N < card R. He then raises the question Is there any cardinal
number properly between these two? The statement that there is no such is Cantors Continuum Hypothesis (CH). R is commonly called the continuum. For ZF Set Theory many sentences are independent of the axioms and so remain undecided and undecideable; one such is
the CH: There is no cardinal number between the cardinal of the natural numbers and that of
the reals. It turns out that CH can be neither proved nor disproved in ZFC: It is independent of
ZFC. The second half of this result is due to Gdel: CH is consistent with ZFC; that is its negation cannot be proved from ZFC. Subsequently Cohen establishes the other half of the independence of CH: it cannot be proved from ZFC. CH is thus an example of an interesting
statement about sets which is independent of the axioms of ZFC; similarly the Axiom of
Choice (AC) is independent of ZFC.
Cantor uses for the cardinal number of N, the symbol 0 (aleph-naught) -the countable
infinity, the infinity of the natural numbers-, the symbol 20 for the power-set of 0, and the
symbol 1 for the continuum infinity, the infinity of the real numbers. Since the cardinals are
comparable by twos, CH is expressed by the successive cardinal; then CH says, in mathematical terms, that 20 =1. This says that the continuum is the very next transfinite number
after the infinity of the natural numbers. Gdel points out just how far this is from being
proved. But nothing has been proved so far about the question what the power of the continuum is. In other words, there might be one other infinity in between, or a million, or a countably infinite number, or even an uncountably infinite number of infinities in between the natural numbers and the continuum. Cohen believes that there probably is no limit on how much
such infinity that there are and suggested the likelihood that the CH was obviously false, and
he conjectured that 20 might well turn out to be larger than any transfinite aleph. In his view,
the continuum is an incredibly rich set one produced by a bold new axiom which could never
be approached by any piecemeal process of construction (Bell, 1986; Gdel, 1947; Hofstadter, 1982; Mac Lane, 1986).
13.2. Gdels Incompleteness Theorems
Studying several views of this issue (Gdel, 1931; Nagel & Newman, 1973; Mac Lane,
1986; Uspensky, 1987), we can say that in 1931 Gdel proves two incompleteness theorems. Gdel considers a formal theory T which contains Arithmetic and shows how to construct a sentence G such that neither G nor not-G could be proved within the system T. Such a
G is then undecideable and its existence is the Gdels First Incompleteness Theorem. Gdel
also shows that we can formulate within the system T a sentence con(T) which interpreted
states that T is consistent. Then he shows this sentence could not be proved within the system T; in other words, no such system T is strong enough to establish its own consistency. It
is the Gdels Second Incompleteness Theorem.
Besides the worth of these two theorems, it is of particular significance the way which
Gdel follows to prove these theorems. The proofs depend on a coding -arithmetized symbols-, by Gdel numbers which translate statements about proofs of T to statements about

On Numbers Nature

33

numbers within T, i.e. into the language of Arithmetic. Since T contains ordinary Arithmetic,
it has constants such as 0, 0=1, 0 = 2 etc. for numerals, variables x, y, z, for numbers,
equality, a successor function and symbols for operations, as well as, the apparatus of the first
order predicate calculus. By the use of Chinese Remainder Theorem, in T we can define the
prime number and prove in T the Fundamental Theorem of Arithmetic: Every natural number is uniquely a product of primes. By explicitly enumerating formulas and proofs in T in
some code, we can replace statements about proofs by statements about the corresponding
numbers, their codes. Here is an example of a code:

0
7

11

=
13

+
17

*
19

x
23

y
29

z
31

Each formula is a finite list of symbols, so is coded by a list of numbers n1, n2, n3, :
This can be replaced by a single number m = 2n1 3n2 5n3 Conversely, give a number m, we
can factor it into primes and so read off the corresponding list of basic symbols. An analogous
procedure is followed just for the proofs in T.
The above process achieves to establish an 1-1 mapping between all the formulas and the
proofs on the one hand and a definite subset of the natural numbers -in particular the set of
prime numbers P- on the other hand. What characterizes not only Gdels work but also the
new era which initiates with it, is the arithmetizing of Meta-mathematics that is the correspondence of logical relations to pure arithmetical relations. With this arithmetizing, it is finally achieved a reduction of Meta-mathematics, as it happened before with Mathematics, to
the natural numbers. Therefore, what Gdel achieves, is to properly map all propositions of
Meta-mathematics which refer to structure properties of the logico-mathematical system in
use containing all natural numbers, to the system itself.
The method of arithmetizing Meta-mathematics that Gdel follows is like that of the
Analytic Geometry where the coordinates correspond to the geometric figures and the latter
are expressed by arithmetical relations. By Gdels work, was also made clear that the noncontradictory of a logical-mathematical system which contains all natural numbers is not expected to be proved with the means of the system itself. This fact is parallel to the descriptive
image where one who is being drawn cannot be saved by dragging own hair out of the sea.
What he needs is someone else to drag him out; the same holds for the proof of noncontradictory, where we must use proofing means out of the system. Jung uses an analogous
argument, when he sustains that the reply to the question How does the system hummer-nail
work? cannot be found into the system, but outside of it, at the hand which holds the hummer
and hits the nail. But according to Gdel himself, the problem can be solved with some appropriate philosophical realism, analogous to that of the platonic archetypal ideas. It is important to note here that when Gdel shows that no axiomatic system with enough axioms -like
that of Arithmetic- can prove the true or false of each proposition expressed inside it, he actually shows that there is inherent uncertainty in Mathematics and therefore we cannot confine
ourselves only to the study of its clear and definite characteristics.
Weyl, finally, presents the consequences of these theorems, saying that the arithmetization method leads to the aspect that the natural numbers, along with their Arithmetic, consist
such a broad field, so that any formalized theory can be mapped inside it. This astonishing
property of the number, acknowledged by Pythagoras and Plato, is being used by Gdel for

34

Dimitris Gavalas

the meta-mathematical study of a given mathematical formalism. All these show that our
view about natural numbers is still located in initial stages.
After that we have to study the interesting work of Robertson (1995), who connects
Jungs view about number with these of Cantors and Gdels. Drawing on the initial concept
advanced by Pythagoras, i.e that reality is number, and Platos later theory, i.e. that a world of
perfect ideas exists which transcends the world we live in, the author sketches the scientific
advances made from the Renaissance through the 20th century. He explains Descartes Geometry, the Calculus of Newton and Leibniz, Cantors Theory of Infinite Sets, Freuds Theory of Personality, Jungs model of the psyche, and Gdels Incompleteness Theorems. Utilizing the above mentioned background material, Robertson then explains how Jung and
Gdel both posited that a Platonic world of ideas -archetypes- exist beyond physical reality
and the inner world of the psyche. And his well reasoned conclusion is that these archetypes
are Pythagoras simple counting numbers. This is a work for anyone interested in Mathematics, Psychology, Philosophy, and archetypes.

13.3. Jung and Cantors Continuum Hypothesis

Through his study of his patients number dreams, Jung came to believe that the smaller
natural numbers are symbols in much the same sense that the people and events of our
dreams are symbols of personified collective character traits and behavioral situations. The
integers seemed to correspond to progressive stages of development within the psyche. In
brief, one corresponds to a stage of non-differentiation; two to polarity or opposition; three to
movement toward resolution; four to stability, wholeness, as in a quaternity, or a mandala,
which is most commonly four-sided. Jung went beyond this limited model, and took a brilliant leap toward generalization of these discoveries: he speculated that number itself -as expressed most basically in the small integers- was the most primitive archetype of order.
There is something peculiar, one might even say mysterious, about numbers. If a group of
objects is deprived of every single one of its properties or characteristics, there still remains,
at the end, its number, which seems to indicate that the number is something irreducible,
which helps more than anything else to bring order into the chaos of appearances. It may well
be the most primitive element of order in the human mind. We can define number psychologically as an archetype of order which has become conscious (Jung, CW8). Since the natural numbers were each true symbols of order, the implication was that the development of
Mathematics reflected the progressive development of order within the psyche. Even more
than that, Jung felt that number might be the primary archetype of order of the unus mundus
itself; i.e., the most basic building blocks of either psyche or matter are the integers.
Number is a very ancient archetype that seems to predate humanity itself. Dantzig (1954)
mentions a number of examples of animals and even insects which seem to possess a number
sense. The same does Dehaene (1999) more than forty years later. They say that animals,
insects, infants have a sense of one, two, three, and many. Interestingly, anthropologists found that the numeric systems of some African, South American, Oceanic and Australian cultures had the same limitations -e.g., the Australian aborigines only had numbers for
one through six, and many.
The archetypal quality of the smaller numbers is so ancient that it predates humanity itself, and is carried in the heritage of creatures even as primitive as insects. Because human
beings are capable of counting, we imagine that is how numbers were arrived at. But when

On Numbers Nature

35

crows can recognize one, two, three and many, few of us would argue they arrived at
these numeric relationships by counting per se. Instead there must be pattern recognition, a
primordial image, to use Jungs earlier formulation of symbol, which corresponds to the
smaller integers. In other words, we have an innate sense of what one and two and
three mean. Now, if we conceive numbers as having been discovered, and not merely invented as an instrument for counting, then on account of their mythological nature they belong to the realm of godlike human and animal figures and are just as archetypal as they are
(Jung, CW10).
As civilization developed, there was a need for ever larger numbers. This need puts a
strain on any system of separate and distinct symbols. Even among great mathematicians, it is
the rare genius for whom virtually all numbers come to possess true symbolic stature. One
such was Ramanujan, who recognized that number 1729 It is a very interesting number. It is
the smallest number expressible as the sum of two cubes in two different ways. [1729 =
123+13 = 103+93] (Hardy, 1992). For most of us, however, this archetypal pattern recognition
is unlikely to extend past the smaller counting numbers.
Since, at this stage, the recognition of number is the recognition of a primordial image or
pattern, there is as yet little if any distinction between Arithmetic and Geometry. As soon as
Arithmetic and Geometry split and go different directions, it becomes much less clear that
Jung is necessarily right in his guess that all Mathematics emerges from the smaller counting
numbers. Geometry by its various nature deals with continuous lines, figures and planes,
while Arithmetic develops out of ever grander extensions of the discrete counting numbers.
At the time when Jung was developing these ideas, he was corresponding with Pauli. Pauli,
inspired by Jung, was searching for a neutral language which could underlie both the physical and psychological worlds (Pauli, 1955; Card, 1991). Pauli recognized that the issue came
to a head when the development of Arithmetic reached the point where it was forced to deal
with infinite quantities. If, therefore, a more general concept of archetype is used today, then
it should be understood in such a way that included within it is the mathematical primal intuition which expresses itself, among other ways, in Arithmetic, in the idea of the infinite series
of integers, and in Geometry, in the idea of the continuum (Card, 1993). At that point, Jungs
speculation becomes identical with Cantors Continuum Hypothesis; i.e., are there any infinity that lie between the infinity of the integers and the infinity of the geometric continuum?
Cantors Continuum Hypothesis is the question: How many points are there on a straight
line in Euclidean space? An equivalent question is: How many different sets of integers do
there exist? This question, of course, could arise only after the concept of number had been
extended to infinite sets (Gdel, 1947).
Though Arithmetic and Geometry inhabit separate realms, it is inordinately productive
when either turns to the other for a new way of thinking about a problem. Back and forth goes
the cross-fertilization between the two fields. This is because at their extremes -dealing with
the discrete counting numbers and the geometric continuum respectively- Arithmetic and Geometry seem like very different fields. There is, however, a great fuzzy area where they overlap, since reality itself is more fuzzy than not. Similarly, though at their extremes mind and
matter seem totally distinct, the boundaries between the two are fuzzy. There is little blackand-white in the real world. Between the extremes there is a grey spectrum where everything
is matter of grade (Kosko, 1994).
The CH marks the point where the boundaries of Arithmetic and Geometry are very hazy
indeed. There is dispute that the power-set of the countable numbers is the same size as the

36

Dimitris Gavalas

continuum; the CH says that there is no infinity that lies between the two. Starting with the
small natural numbers, we build up ineluctably to the countable infinity of all integers; but
taking all the combinations of those numbers is a very different thing than merely accumulating. On the human level, we certainly understand how much more complex it is to deal with
relationships than with things. We can think of countable infinity as a natural infinity, as the
limit of what we encounter in dealing with the things of the world. The uncountable infinity
of the continuum is the limit of what we encounter dealing with the relationships in the world.
The higher infinities beyond the continuum are all power-sets of the continuum, hence relationships between relationships, etc. Though Cantor remained convinced that the CH was
true, others were less sure from the beginning. Cantors CH assumed the identifiability of two
concepts that were intrinsically different and of no comparable orders of magnitude. The two
ideas were inherently antithetical: The nature of the continuum, regarded as the collection of
all infinite sequences of rational numbers, was something totally different from the infinity of
natural numbers.
Gdel had already proved that any system at least as rich as Arithmetic contains undecidable mathematical truths; he guessed that the CH might be just such an undecidable proposition. There are -assuming the consistency of the axioms- a priori three possibilities for Cantors conjecture: It may be (i) demonstrable, (ii) disprovable, or (iii) undecidable. The third
alternative is the most likely. To seek a proof for it is, at present, perhaps the most promising
way of attacking the problem. One result along these lines has been obtained already, namely
that Cantors conjecture is not disprovable from the axioms of Set Theory, provided that these
axioms are consistent.
The result was given by Gdel by his own proof in 1940 that is, if a modified Set Theory,
which does not include the CH is consistent, then it will remain consistent if the CH is added
as an additional axiom. However, this was only half of what was needed to prove that the CH
was undecidable within Set Theory. In order to prove the other half, Gdel needed to show
that if a modified Set Theory, which does not include the CH is consistent, then it will still be
consistent if the CH is assumed to be false. Though Gdel made some progress toward a solution, he was never able to prove it, as it was simply too complex to be resolved with the
mathematical tools available at the time. Finally, in 1963, Cohen was able to prove the second
half of the problem, thus showing that within Set Theory, the CH was undecidable. So,
Gdels incompleteness proof had demonstrated that every logical system contains essentially
undecidable propositions. Now Gdel - Cohen had shown that the CH was undecidable
within Set Theory. Since by this time Mathematics and Set Theory were inseparable for most
mathematicians, this was a haunting proof.
According to Cohen (1963, 1964, and 1966) and Gdel (1958) we can present the evolution of this story as follows: Gdel had already conjectured that even if the CH was undecidable within Set Theory as then constituted, within some extended Set Theory it would
eventually be resolved in some way. The undecidability of CH from the axioms being assumed today can only mean that these axioms do not contain a complete description of the
reality.
Gdel was confident that Set Theory is an important positive step for Mathematics. He
believed that it necessarily describes a well-determined reality, perhaps the mathematical
reality which underlies the reality in which we actually exist -perhaps Jungs unus mundus; in
any case, some well-determined reality. Gdel is confident that the CH must be either true
or false within that particular reality. Since he and Cohen together proved that the CH is nei-

On Numbers Nature

37

ther true nor false, but undecidable within Set Theory as it is currently constituted, then Gdel
can only assume that is because the current set of axioms are insufficient to fully define that
reality; i.e., that there are further axioms yet to be discovered.
Whereas Gdels Incompleteness Theorems seemed to ring the death knell for the development of Mathematics, Gdel has faith that Mathematics is inexhaustible. He considered
that Set Theory is the right direction for Mathematics, and that within an extended Set Theory, the CH is resolvable, and finally that any axioms powerful enough to enable the CH to be
resolved will necessarily be powerful enough to lead Mathematics in new directions. So far,
Gdels view has not been justified. Sixty years have passed since 1947, and no extension of
Set Theory has yet been discovered in which the CH can be resolved. Perhaps it is truly an
undecidable proposition, or perhaps Mathematics is waiting for another Cantor, or Gdel.
While we wait, we have to know that both Gdel and Cohen feel that eventually the CH will
be resolved.
That is where things rest at this point. Jung guessed that the natural numbers are the primary archetype of order in the unus mundus, and that all Mathematics develops out of the
natural numbers. Cantor would have agreed with him. Jungs whole psychology developed
out of an attempt to deal with the problems presented by the fact that the world outside is
somehow contained within our minds/ brains, while at the same we and our minds/ brains are
obviously contained within the world. His Depth Psychology explored the world within, especially as it found expression in the dreams, myths, and other unconscious expressions of
human beings. At each level of the psyche, he discovered structure and order. As he went
deeper and deeper, it is only natural that eventually he arrived at a level where that structure
and order had little or no human qualities attached to it. That was the level of number as archetype. (Robertson, 1995).

13.4. Jung and Gdels Incompleteness Theorems

We can connect the works of Jung and of Gdel. Jungs view is that number itself is archetypal: It may be the most primitive element of order in the human mind. We can define
number psychologically as an archetype of order which has become conscious. Jung feels
that number might be the primary archetype of order of the unus mundus itself: i.e., the most
basic building blocks of either psyche or matter are the integers. If the Archetypal Hypothesis
that has emerged from the work of Jung - Pauli - von Franz holds that the most basic building
blocks of psyche and matter are natural numbers, then the properties of natural numbers must
be sufficient to account for the nature of the geometrical continuum. In other words, Cantors
CH must indeed be true. Here at last Jung meets Gdel.
Gdel takes up the proof of Cantors CH subsequent to his proof of the Incompleteness
Theorems, and he comes to suspect that within the axiom system of standard Set Theory, the
CH is in fact one such formally undecidable proposition, the possibility of which existence
had been established by the Incompleteness Theorems. Gdel was able to complete half of the
proof of the undecidability of the CH, and Cohen completed the remainder of the proof in
1963. If the CH is formally undecidable within standard Set Theory, then the possibility of
fully describing the properties of the continuum on the basis of infinite sequences of natural
numbers is thrown into doubt. Gdel believed that an extension to the axiom system of Set
Theory could be constructed in which the validity of the CH could be determined, but no such
extended theory has been discovered in nearly sixty years. As a consequence, the prospects

38

Dimitris Gavalas

for finding representations of the unus mundus based upon the integer nature of the number
archetype is called into question.
After all, isnt the attempt to reduce the world to the archetypes of the small natural numbers much the same as the attempt of Russell - Whitehead to reduce Mathematics to Logic,
the attempt of physicists to reduce the physical world to first atoms, then subatomic particles,
then most recently to quarks? Isnt it more likely that the world is richer than we can ever
hope to comprehend? Jung thought that there was a unitary reality -the unus mundus- that
underlay both psyche and matter, and speculated that the primary archetypes of this unitary
reality were the simple counting numbers. In this case, each number is, itself, a true symbol:
undefinable and inexhaustible -a much less reductionistic stance than the hope that all reality
can be reduced to Logic. But even so, doesnt it seem unlikely that we will ever find any lowest level to reality? The Archetypal Hypothesis is a starting point to explore the magic and
wonder of the world, not an end point to circumscribe its possibilities.
This argument raises important issues for understanding the Archetypal Hypothesis and
its implications. If the number archetype is to be understood to be completely equivalent to
the natural numbers, then the evocation of Gdels and Cohens work on the CH is valid
without qualification. The Archetypal Hypothesis would then be seen to be a recycled version
of Pythagoreanism, brought to the dust by the undecidability of the CH, in much the same
way as the original dream of the Pythagoreans foundered with the discovery of the irrational
numbers. The crucial issue, then, is whether the number archetype is simply the natural numbers. Jung does not express himself clearly on this point, and even von Franz is not altogether
consistent in her discussion of the relationship of number archetype to the natural numbers.
The problem is the failure to distinguish between the concept of the archetype-as-such
and the representations of the archetype; the images and ideas that are the specific realizations
of the archetype, each pointing toward the existence of the abstract archetype-as-such but
none being strictly equivalent to it. In the context of number archetype, this problem emerges
as the failure to distinguish between the number archetype-as-such and a specific representation of the number archetype such as the natural numbers. Thus the number archetype is not
the natural numbers; the natural numbers are only one specific representation of the number
archetype. As with specific representations of any archetype, the natural numbers are symbols
with an unlimited potential for expression, but their properties do not exhaust all of the possibilities for representation implicit in the number archetype-as-such.
The application of the undecidability of the CH to one representation of the number archetype, namely the natural numbers, does not, in fact, lead to a conundrum for the Archetypal Hypothesis. Rather, this result is in concordance with the above mentioned fact about
representations of archetypes: No representation is a complete representation. Apart from
what can be expressed by any specific representation of an archetype, there are other aspects
of the archetype that are valid but unrepresented. Here, it seems, Jungs empirical findings
resonate with Gdels Incompleteness Theorems. From this perspective, the answer to the
question about the attempt to reduce the world only to the archetypes of the small natural
numbers much the same as the attempt of Russell - Whitehead to reduce Mathematics to
Logic, is simply no.
It is an important distinguishing feature of any prospective archetypal science that it entertains no dreams of a Final Theory and has no presumptions of being a Theory of Everything in the sense that contemporary Physics presently imagines. An archetypal science
would be, by its nature, a self-reflective science that is not only aware of its own epistemo-

On Numbers Nature

39

logical capacities and limitations but actually incorporates them -e.g. archetypes- in its representations of phenomena. Given these considerations, the question naturally arises: If natural
numbers are simply one representation of the number archetype, what other representations
might there be? Finding such representations would be a prerequisite for any future development of science from an archetypal perspective. Some hints in this direction can be found in
the work of von Franz, and nowadays of Robertson and Card.
Jung and Gdel, each one in his own way, try to open our mind to the other reality,
which exists in inner and outer world. These two views of the world, external and internal,
have common basis and are expressed by a common factor, which is number. We are still
very early in our understanding of the archetypal nature of reality. Platos aspect of an ideal
world is an early attempt at describing the archetypal nature of reality, and it lacks so much.
Platos ideal world lies, like other early views of the humans, totally separate from us and our
experience. Science proceeded by exploring the outer rather, not the inner world. Mathematics plays a role in that process; without Mathematics there is no science. Gdels Incompleteness Theorems demonstrated once and for all that Mathematics is bigger than Logic; so are
science and all human endeavors. But, if Logic is insufficient, perhaps Mathematics, as is
develops out of the simple archetypal counting numbers, is enough. Is it really so surprising
that the CH proves intractable?

14. Number and Jungs Psychology

Jung (CW6) claims that in order to live in space-time and in the relations of the things of
the world, in order to communicate and to exchange information with the environment, we
need psychological functions, of which here are the four basic types: Sensation, Thinking,
Feeling and Intuition. Jung places counter-diametrically on a circle these four basic psychological functions and names the centre of the circle, where all these functions meet, with the
Greek classical term pemptusia (quintessence).
Rucker (1988) draws on Jungs view to explain why Mathematics has developed in five
ways: Number, Logic, Space, Infinity and Information. These five ways are related to the five
above-mentioned psychological functions, where here he regards Information as Quintessence; as Quintessence unifies all the psychological functions, similarly Information unifies
all the branches of Mathematics. Then he makes the following correlation with Mathematics:
Sensation Number, Thinking Logic, Feeling Space, Intuition Infinity and Quintessence Information.
The interesting, for this work, is that he explains the correlation between Sensation and
Number: Sensation informs us that something exists; this something is the result of a distinction which leads us to number. The world of Sensation and that of number is the world of
the discrete. So, distinctions among objects can be made leads to our perception of discreteness; discreteness leads, in turn, to number.
If we examine through time the basic concept of Mathematics in each era, then we observe that this is number according to the Middle Ages, space according to Renaissance,
Logic according to Industrial Revolution and infinity according to Modern Times. Nevertheless, with computers progress, we are already following the new era in Mathematics, the fifth
in a row, in which the dominant concept is the one of information. The conception of the

40

Dimitris Gavalas

world is based on the exchange of information and on communication with it. Whenever we
are learning, we are sending and receiving transmissions, generating and absorbing information; we are communicating (Gavalas, 1999).
Finally, for Jung different numbers make up various kinds of patterns. Certain archetypes
are associated with some of the smaller numbers:
1.
2.
3.
4.
5.

monad: unity
dyad: opposition
triad: thesis-antithesis-synthesis
tetrad: balance of quaternity
quintad: a step further out.

Robertson (1989) sustains that as number evolved, so did the relationship between the
ego and the Self, which is, according to Jung, the totality of the human personality:
(i) Natural numbers coincided with the period of participation mystique, when man
could only experience himself through his projections on nature.
(ii) Zero appeared at the birth of the Christian era, as the ego began to emerge as a thing
in itself. As zero gradually led to the explicit formulation of infinity, the ego unsuccessfully tried to swallow the Self.
(iii) Infinity gave way to self-reference in the twentieth century, as order began to crumble into chaos everywhere around us. The ego, unable to swallow the Self, gave way
to despair. God was dead; chaos reigned.
But even chaos was revealed as possessing a self-referential order. The relationship between ego and Self is not one in which ego rules Self or Self rules ego. Rather, the Self is at
one and the same time both the goal of the ego, the process by which the ego attains that goal.
The ego is both the Selfs expression in the world of limits and the process of evolution of the
Self.
Sensing the Self as something irrational, as an indefinable existent, to which the ego
is neither opposed nor subjected, but merely attached, and about which it revolves much
as the earth revolves around the sun -thus we come to the goal of individuation. (Jung,
CW7).
Number, as the archetype of order, is in the process of finding a new symbol with which
to clothe itself. A new order is trying to emerge from chaos. When that happens, it will correspond to a new vision of the Self, as it did when natural numbers ruled, when zero emerged,
when infinity found concrete expression.
At the end of this Part we must underline the fact that the initial work of Jung and the
other of Jung - Pauli opened a completely new way for an essential consideration of numbers
nature. The work of these pioneers was continued by von Franz; today this way is followed
mainly by Card and Robertson and also by others (Gough & Shacklett, 1993). In the next Part
IV we present our own view which is closer to Mathematics and expresses the previous aspects in a mathematical way through the contemporary Category Theory.

On Numbers Nature

41

PART IV: NUMBER, MONOID AND ARCHETYPE

15. 2-Categories; Monad, Monoid, Monoidal Category, Monoid Object; the
Dynamical System as Functor
15.1. 2-Categories
Firstly, by the term double category we mean that the set of morphisms of a category has
two different compositions which satisfy the exchange law, as we shall see. By the term 2category, which is a brevity for the term two-dimensional category, we also mean a double
category in which every identity morphism is such for both compositions. There are also ncategories.
Indeed, sets are zero-dimensional in that they only consist of objects and there is no way
to go from one object to another within a set. Nonetheless, we can go from one set to another
using a function. So the category of all sets, the Set, is one-dimensional because it has both,
objects and morphisms/ arrows/ functions between objects. In general, categories are onedimensional in this sense. But this in turn makes the collection of all categories, the Cat, into
a two-dimensional structure, a 2-category having objects, morphisms between objects/ categories which we call functors, and 2-morphisms between morphisms/ functors which we call
natural transformations. This process never stops and the collection of all n-categories is an
(n+1)-category until -categories.
In 7.1 we discourse about category, but a 2-category also consists of a collection of 2morphisms. Every 2-morphism T has a source morphism f and a target morphism g. If the
source of T is f and its target is g, we write T: f g. If T: f g, we require that f and g have
the same source and the same target; for example, f: X Y and g: X Y. We should visualize T as follows:
f
X T
Y.
g
In addition, we have:
(1) Given a 2-morphism S: f g and a 2-morphism T: g h, there is a 2-morphism ST:
f h, which we call the vertical composite of S and T.
(2) Vertical composition is associative: (ST)U = S(TU).
(3) For each morphism f there is a 2-morphism 1f: f f, called the identity of f. For
any T: f g we have 1f T = T 1g = T.
It is obvious that these are just like the previous 3 rules for a category. We draw the vertical composite of S: f g and T: g h like this:

f
S
g Y.
T
h

42

Dimitris Gavalas
We also require that we can horizontally compose 2-morphisms as follows:

f
S
g

f
T
g

Z.

So we also demand:
(1) Given morphisms f, g: X Y and f', g': Y Z and 2-morphisms S: f g and T: f'
g', there is a 2-morphism ST: ff' gg', which we call the horizontal composite
of S and T.
(2) Horizontal composition is associative: (ST)U = S(TU).
(3) The identities for vertical composition are also the identities for horizontal composition. That is, given f, g: X Y and : f g, we have 11x T = T 11y = T.
Finally, we demand the exchange law relating horizontal and vertical composition:
(ST)(S'T') = (SS')(TT'). This makes the following 2-morphism unambiguous:

f
S
g
T
h

f
S
g
T
h

Z.

We can think of it either as the result of first doing two vertical composites, and then one
horizontal composite, or as the result of first doing two horizontal composites, and then one
vertical composite.
Like the archetype of category is the category of sets and functions, that is Set, so the archetype of 2-category is the category with objects the small categories and morphisms the
functors, that is Cat, where 2-morphisms are the natural transformations. Here it also holds:
The set of all natural transformations is the set of morphisms of two different categories supplied with two different compositions which satisfy the exchange law and the identity morphism is such for both compositions. In case of Cat the objects for horizontal composition are
categories, while for the vertical ones are functors. From all the above we see 2-category can
be considered as abstract two-dimensional space where the axioms tell us how we pose together 2-morphisms in these two-dimensions. So we can pose them horizontally, that is the
one next to the other, or vertically, the one above the other. Note that all is meant by archetype of category for Set is that this is a familiar category and that, if we start from Set and
various categories of structures built using sets, e.g. groups, rings, vector and topological
spaces, we can then abstract the notion of category and thus obtain Cat. In the same sense
Cat is the archetype of 2-category and so on.

15.2. Monad, Monoid, Monoidal Category, Monoid Object

The 2-category is a framework where we can present the general concepts of multiplication or combination. Multiplication is a function : SxS S: (a, b) M(a, b) = ab, where

On Numbers Nature

43

we take two elements a and b from a set S, and multiply them to get a new one M(a,b)=ab of
the set S. We can visualize this as follows:
X

S
M

X
S

where this triangular shape takes two inputs from the two slanted edges labelled S, and spits
out one output from the horizontal edge labelled S on the bottom. It is clear from the geometry here that M is something 2-dimensional, hence, a 2-morphism, and that S is 1dimensional, hence, a morphism from X to itself. Here X, being 0-dimensional, is an object.
We can take the dual diagram like this:

which illustrates more vividly how M is the process of two copies of S getting squashed
down into one copy. This sort of diagram is called a string diagram and it is literally the
Poincare dual of the earlier picture, meaning that stuff that was k-dimensional is now drawn
as (2-k)-dimensional. The 0-dimensional object X is now the 2-dimensional background.
The essence of multiplication can be described generally in a situation where we have a
2-category with an object X in it, a morphism s: X X and a 2-morphism M: ss s. We are
often interested in situations like this where the multiplication M is associative and there is a
multiplicative unit. The first means that the composite
1s M M
M1s M
sss ss s equals sss ss s,
where 1s: s s is the identity 2-morphism from s to itself.

44

Dimitris Gavalas
The second means that there is a 2-morphism I: 1X s for which
I1s M
s = 1Xs ss s
equals 1s, and so does
1 sI M
s = s1X ss s.

If we have a 2-category with stuff in it satisfying these rules, we say we have a Monad in
that 2-category.
As an example of a Monad we can consider the original example where s is a set and M
is a function. Then the 2-category has only one object X, the morphisms of this 2-category are
sets, the composing morphisms corresponds to taking the Cartesian product of sets and the 2morphisms of this 2-category are functions between sets. In this case the Monad is formulated
as follows: The multiplicative unit 1X corresponds to the one-element set 1, s is a set, the 2morphism I: 1X s is a function from the one-element set 1 to s, which picks out a special
element of s. The 2-morphism M: ss s is the multiplication operation and is associative
while the special element of s is the multiplicative unit that is, it serves as the left and right
identity for multiplication. So we have a set with an associative multiplication and a unit for
this multiplication. That is what is called a Monoid. So a Monoid is a special sort of Monad.
Finally, a 2-category with just one object is called a Monoidal category. For example, Set is a
Monoidal category where we can multiply objects, i.e. sets, with the Cartesian product.
If we consider the 2-category with just one object, forget the object and correspond the
morphisms to objects and the 2-morphisms to morphisms, then we have got a category where
we can compose/ multiply objects, because they were secretly morphisms from X to itself. So
we do this degradation:
2-morphisms morphisms
morphisms objects
object X we forget it.
This is a very important point for our view in the following.
Because a Monoidal category is a 2-category with one object, we can talk about Monads
in any Monoidal category. These are usually called Monoid Objects, because they are like a
Monoid living in the category in question. The category Cat having (small) categories as objects and functors as morphisms becomes a Monoidal category with the Cartesian product of
categories as the way to multiply objects and a Monoid Object in this is a Monoidal category.
A Monoid Object is defined in a Monoidal category, but a Monoidal category is itself a kind
of Monoid Object.
If a category is a Monoidal category with one object X, then the set hom(X, X), of all
morphisms from X to X, is a set with an associative binary product, namely composition, and
a unit element, namely 1. Conversely, if we have a Monoid S in the traditional sense, we can
easily cook up a category with one object X and hom(X, X) = S.

On Numbers Nature

45

We can see morphisms as paths between objects, 2-morphisms as paths between paths
and so on. We can think of morphism as a process which transforms or projects an object
into another. The neutral morphism corresponds to the process of doing nothing at all. We
can compose processes h and g, do h and then g, and get the product gh. Crucially, every
process g can be undone using its inverse g-1. We tend to think of this ability to undo any
process as a key aspect of symmetry and for a full understanding of symmetry we should
really study Monoids. So, a Monoid is like a group, but the symmetries no longer need be
invertible; a category is like a Monoid, but the symmetries no longer need to be composable.
In contrast to a set, which consists of a static collection of objects, a category consists not
only of objects but also of morphisms which can be viewed as processes transforming one
object into another. Similarly, in a 2-category, the 2-morphisms can be regarded as processes
between processes, and so on.
Coming to an end: A category with one object is a Monoid and the classical definition of
a Monoid is this: A set S with an associative binary product and a unit element 1 such that a1
= 1a = a for all a in S. Monoids abound in Mathematics and they are, in a sense, the most archetypal algebraic structures.

15.3. Dynamical System as Functor

Let it be N a Monoid, so that it only has one object denoted *, the maps are natural numbers, the composition is addition (or multiplication), the identity of *, 1*, is the 0 (or the 1). A
functor, arrow between categories, Set, means that we interpret the * as a set S and that
every map n: * * in N is interpreted as an endomap of the set S, gn: S S in such a way
that g0 = 1S and gn+m = gngm (or gnm = gngm )
Example
Consider as S a set of numbers and define gn: gn(x) = n+x. The classical example of interpretation of a Monoid in Set is that the object of the Monoid * is interpreted as the set of
maps of the Monoid (natural numbers) and in this way always get a functor from the Monoid
N to the category of sets Set. Of course, there are lots of functors other than the classical one
from N to Set.
Consider a set X together with an endomap , <X, >, then we have a dynamical system.
Let interpret * as X and send each map n of N (natural number) to the composite of with
itself n times, that is n n, and 0 to the identity of X, i.e. 0 1. In this way we get a functor H from N to sets, H: N Set, which satisfies:
(i) (*) = , (ii) (n) = n, (iii) (0) = 1, (iv) H(n+m) = H(m)H(n),
because the composition of with itself is associative.
So, when we specify a set with an endomap <, >, we obtain a functorial interpretation of N in Set. This means that another name and way of being for the category of dynamical systems is SetN. Indeed, this is the category of dynamical systems of discrete-time. So, a
discrete-time dynamical system is just a functor from Monoid N to Set. In analogous way, we
can have a continuous-time dynamical system, if replace natural numbers with real numbers,
i.e. we allow in the Monoid all real numbers as maps. Then, to give a functor from the Mon-

46

Dimitris Gavalas

oid R to Set, is to give a set X with an endomap t, <X, t>, for every real number t. We must
ensure, of course, that 0= 1 and s+t= st. We can think of X as the set of states of a system which, if it is in the state x at a certain time, then t units of time later it will be in the state
t(x) (Baez, 1996; Lawvere & Schanuel, 2005).

16. The Archetype of Number as Monoid and its Interpretation as the Set
of Numbers
There is only one archetype for number, which belongs to the existential subsoil of humans. One cannot refer to his own number archetype, but only to the one and only number
archetype. This is the principle according to which Mathematics -and not only- is formulated.
Man is carrier of this archetype and contains an image or idea or model in his non-conscious
mind and, perhaps, brain. This archetype is inaccessible and unmodified, but man can capture
its manifestations. These last ones and their consequences are the only things that we know,
since at bottom we do not know the nature of the archetype itself. This point is not a paradox
if we consider that neither for energy we know something, only its manifestations and its consequences. We simply can say that all these come from the region of hyper-personal factors
which lie not on the surface of the consciousness, but on the deepest ends of man. Because
this region is much older than man and is the presupposition of his life, we call it divine in
contrary to the human. Lets not forget that the initial meaning of Mathematics and especially
of number and of geometrical figures was divine and magical. Therefore, Mathematics as
such is psychogenic that is spontaneous psychic fact, product of non-conscious procedures.
This is independent on the fact that these first divine hints are elaborated and are used consciously by the human factor
Number is one of the most important archetypes and may be the womb of all the others. It
is an archetype of order which has become conscious. Wherever exists order also exists structure, which is expressed numerically or geometrically and therefore Mathematics comes in.
Hence the archetype of number is one but the imprints are many. The question is: Which one
of these many imprints of number -that is numbers as we know and use them- are really different between them? We said before that the natural numbers produce all known numbers
and the majority of Mathematics: N Z Q R C. But as we know not all natural
numbers are of the same kind. Instead, there are the prime and the composite natural numbers
and according to the Fundamental Theorem of Arithmetic every composite is analyzed
uniquely in a product of prime numbers. Therefore, the primes are adequate to produce all
natural numbers, which produce all the others and the biggest part, if not all, of Mathematics.
If we denote with P the set of all primes, then we have: N Z Q R C. This
means that the primes are the atoms of mathematics and play an analogous role with Chemistry elements: They produce all numbers.
Since the number archetype is unique, we can consider it as the one and only object of a
Monoidal category and to represent it as . All arrows in this category are endomaps of this
unique object. There are a lot of arrows in this category all arrows are from to , since is
the unique object. As arrows, we can consider all natural numbers and therefore 0, 1 etc. are
arrows from to . Thus, we can write:

On Numbers Nature

47

...
As an arrow composition, we can choose the multiplication of natural numbers, that is the
composition of two arrows/ numbers is their product nom=nm. Because there is only one object, every pair of arrows must be composible and indeed the composition/ multiplication satisfies this convention. If 1 is identical then must 1n = n = no1 for every n. Therefore must
be 1 = 1. Such a category is called Monoidal, if it contains just one element. Next, we can
interpret this category to the well-known and most familiar category of sets Set.
If we name this Monoidal category, with composition of arrows the multiplication,
then Set: N, the unique object of is interpreted as the set of natural numbers
and that is why every arrow in -a natural number- is interpreted as an arrow from the set of
naturals to itself:
fn: N N: x fn(x) = nx, with f1 = 1N and fnfm = fnm.
All these show that the objects correspond to objects, arrows correspond to arrows, the
composition holds and the identity corresponds to identity. Therefore holds the structure of
the category. Such a transportation of one category to another, which retails the structure, is
called functor. Of course a functor must retail domain and co-domain but especially here this
happens automatically, since all arrows have the same domain and co-domain.
The conventional paradigm of interpretation of a Monoidal category to the category of
sets Set is the following: The object of the Monoidal category is interpreted as the set of arrows of the Monoidal category and in this way we always take a functor from the Monoidal
category to the category of sets Set. Therefore, the object , which is the archetype of number, is interpreted as the set of the naturals N. This means that the imprints of the archetype
are the natural numbers.
But furthermore we can consider as arrows not natural numbers but rationals and to have
in this way the inverses that is invertible arrows. And even further we can consider as arrows
all real numbers and in the same way, as above, to interpret the object/ archetype as the set of
real numbers. In this way we take all numbers, from naturals to the reals. We can interpret the
archetype of number as the set of the reals. May be it is possible to consider as 2-morphisms
the complex numbers or any other allowed procedure between numbers:

48

Dimitris Gavalas

17. A Holistic View of Mathematics

How can we, using the archetype view, integrate the partial philosophical views for
Mathematics? The following figure helps towards this direction:

Archetypal Aspect
*: Archetype-as-such
1: Archetypal Images
2: Archetypal Ideas
3: Archetypal Effect
4: Projection to the outer World.

Correspondence to Mathematics
Mathematical Archetypes, especially
natural numbers and geometrical continuum
Intuitionism
Platonism
Formalism - Logicism
Applied Mathematics, Mathematics as
Science, Empiricism.

The figure, which comes from Jungs view, and the correspondence, which I think is
natural, show that the different views on Mathematics are just the different levels of realizing/
understanding archetypes and all together belong to the same wholeness. Therefore the different views are not necessary to be considered as competitive or as contradictions but as complementary; that depends on which level of the archetype one realizes according to ones nature and consciousness.

On Numbers Nature

49

Goodman (1979, 1990 and 1991) for instance and others (Atiyah et al., 1994; Jaffe &
Quinn, 1993), conceive and deal with level 4 that is the projection to the outer world. This
shows that they have not realized their projections and are not interested in finding from
where they come and what exactly they mean. Pythagoras instead is placed in the center that
is he works out the essence of archetype. Formalists and logicians are affected by archetypes
action, but do not examine the subject further and therefore they need no interpretations; the
syntactic part suffices for them and they do not search for the meaning, which in this way
remains hidden.
In my view, the course from the centre to the circumference, faces all aspects and illuminates all elements, while simultaneously unifies all philosophical views of Mathematics. Thus
a particular mathematical issue should be firstly reached by its center, the archetype, and
tracked till its applications and only then is it complete. For this reason I insist on the archetypal view of Mathematics, because all the others have been done in a way or another and up
to a point, while the archetypal view has been abandoned.

18. Final View

We have a 2-category with just one object that is a Monoidal category in which holds:
Object Number Archetype
Morphisms Numbers
2-morphisms ?
But, as we mentioned before, we degrade this situation and we consider morphisms/
numbers as objects, 2-morphisms as morphisms/ order < and we forget the object/ archetype, transforming the 2-category into category. By restoring the natural reality and upgrading
the situation many questions arise:
(i) Which is numbers meaning as morphism/ procedure and, for example, in what is it
different procedure 3 from procedure 4? This fact forces us to move to numbers individuation, that is to the study of individual properties and differences and
not stick only to the massive encounter of their collective properties.
(ii) When we consider morphisms as numbers we introduce as composition, the operations of multiplication or sum, but when we have to face them as what really are,
that is as procedures, what will be introduced as composition of procedures?
(iii) When we degrade the 2-morphisms to simple morphisms we consider them as order
< between numbers, but when numbers upgrade to procedures, then which exactly must be the 2-morphisms that is the procedures between procedures, so as to
be simultaneously compatible with the former procedures/ numbers?
(iv) What happens with the procedures that must be considered similar, analogous, and
compatible and the like, that is qualitative procedures and what happens with the
procedures between procedures?
(v) Since numbers/ morphisms/ procedures are at the same time incoming and outcoming
arrows in reference to the unique object that is extensive and intensive and gener-

50

Dimitris Gavalas
ally both the one and the other, then what is the meaning of this double nature and
what exactly are they doing? Therefore, we pass from the either one or the other
to both the one and the other, which reminds us the dialectical synthesis, the Zen
Logic, but also the Fuzzy Logic (Kosko, 1994).
(vi) How is exactly been structuralized and formulated the object/ archetype from the arrows/ numbers which have double nature? With this procedure the object/ archetype is self-structured and self-formulated. Which exactly is the specific structure
and form that it takes?
(vii) Since we can, by the same model/ Monoidal category, produce all numbers and to
have in this way from a basic skeleton various imprints, then on the one hand we
vindicate mathematically the view about the archetype and on the other hand we
have infinite outcoming arrows, that is all numbers, therefore the maximum intension in one and the only object, that is the minimum extension and thats exactly
what Jung defines as psyche: It might be that the psyche should be understood as
unextended intensity. Intensification at the expense of extension. Psyche = highest
intensity in the smallest space. But this exactly shows that the Pythagoreans,
Plato, Xenokrates (: Psyche is automotive number), Jung are right sustaining that
number is a pre-existing element of the world, archetype, qualitative, worlds
structuring and formulating procedure, model of order etc.
(viii) The provision of the object with endomaps makes it a dynamical system. What
happens with such a dynamical system, where the object is the archetype, endomaps are numbers and there are 2-morphisms that are procedures of second grade?

If we consider the above then we reach to the conclusion that the number archetype can
be represented as a Monoidal category, where the object/ archetype is interpreted, to the archetypal category Set, as the set of all numbers. There are abeyances: How shall we define
numbers as procedures, which are the individual differences of these procedures and how
shall we define the 2-procedures as procedures between procedures?

EPILOGUE
The above discussion shows that a different view on number is mathematically and philosophically feasible. This view brings out the other side of the number, that is the dynamic
and the psychic. Indeed, number is presented as a vivid qualitative reality, which we must
reach -except all the others- and through the inner experience. Number is not just only for
use but something else too, the nature of which we must discover. It is a universal principle,
with divine existence: Because number is universal it is also divine and vice versa. The
scientific attitude and the zeitgeist, favoring the sensual things, capture and limit us mostly in
the senses. Thus they do not allow us to deal with principles and archetypes investigation.
However, the different aspect on number would alter Mathematics to that it was in the past: A
discipline which deserves to be investigated beyond utilitarianism.
The realization of this aspect has impacts not only to the scientific encounter of number,
but also to the didactic methodology. In parallel to the traditional aspect on number as a multitude of objects and a means of numbering, we must teach appropriately the aspect of the

On Numbers Nature

51

dynamic procedure and the archetypal psychic factor. Indeed, it is likely this aspect to be
closer to students nature and mans generally, than the prevalent one.
The methods of Gdel and Cohen intend to show that the Continuum Hypothesis (CH)
does not have answer inside Set Theory and its extensions. The same holds for the Axiom of
Choice as well for the Axiom of the Infinite. But the above three propositions express the
foundational principles for the sets, which are true in the classical frame, but are obviously
different from the principles the other simpler axioms express; it is understandable why the
former cannot be proven by the latter only with Logic. The Continuum Hypothesis has the
texture of a typical mathematical problem, which should be solved with some proof, but it
seems that the appropriately enlarged consciousness is missing in order to discover the congruent axioms.
A lot have been said about the independence of CH on the known axioms of Set Theory
and many have appealed it in order to sustain that there is no objective reality behind the
typical axiomatic outcomes of the discipline. But with Gdels method of numerization, the
problems of proof existence are translated into austere technical conjectures on numbers.
There are such conjectures, as CH, that are independent on all generally accepted axioms of
Mathematics, but we do not deny the objective reality of natural numbers because of it.
Such propositions are of the form: If ZFC is consistent, then T (where T is some powerful
extension of ZFC which implies the consistency of ZFC) is consistent.
But from Goedels Second Incompleteness Theorem, it is implied that such propositions
are independent on ZFC and there are examples of such propositions of which the truth is
controversial. Of course there is no lack of important problems that do not admit solution in
ZFC, but CH is the most basic and important of all these. Since it has not yet been decided
whether CH is true or not, then if we consider all set as true/ real, may be we can convince
ourselves that CH can be true or false. Then we must find a new axiom adequate enough to
answer to the relative question. And since such an axiom has not been found it remains to our
free will to accept or reject CH. The expansion of the initial limitations, as well as of our consciousness will contribute to some answers, according to my view. The Hypothesis of Archetypes and the consideration of number as archetype, contribute indeed towards this direction.
According to Jung we must discern between the archetype-as-such -being dynamically
present in every structure- and the manifested archetype -which becomes cognizable and
comes in the realm of consciousness. The manifested archetype appears as image, which its
form can vary and adjust to the environment as a way of action, procedure, stance, idea, belief, experience etc. All these, if are activated under certain circumstances, are emerged by
their non-conscious up to now situation and become visible in a way. Thus, the archetype can
manifest not just in a static form but with a dynamic procedure; in reality all typical human
manifestations of life, biological, psychic, spiritual stand on an archetypal basis.
We can correspond the archetype-as-such, its static form, to being and the manifested
archetype, its dynamic form, to becoming. Then, according to what we said above the first
correlates to Cantorian Mathematics, while the second to non-Cantorian Mathematics. Therefore, the essence of Mathematics is at Cantorian Mathematics, while its presence to nonCantorian ones. In the first case we find a clear and lucid level of reality while in the second
case a situation chaotic, fuzzy and vague. It is obvious that while mathematicians go deeper
and deeper into situations of enlarged knowledge and consciousness they will bring to light
bigger and bigger parts of the number archetype which is the basis of Mathematics, as we

52

Dimitris Gavalas

saw. So, the core of Mathematics is given to man in a natural way as archetype and is the best
for the human nature and its environment.
The continuous expanding of consciousness discovers new properties of number and of
Mathematics. Finally though, it seems that a central part of this core will never become conscious to man, but will remain in the non-conscious realm of the non-perceptual; in the opposite case man would possess the whole knowledge, something superhuman. Until then man
can study with the aid of his means the presence of archetypes and number, conceiving only
shreds of its essence, just like Cantor, Gdel and Jung did, despite Mac Lanes allegation that
all these are just a glorious fantasy. It is obvious that without the essence, the thing itself,
neither presence, nor manifestation, nor expression exists, since in order to be shown the
change of presence is needed the steadiness of the essence. The adoption of a complementarity and relativity principle, as far the absolute and local Mathematics concerns, as Bell (1986)
calls it, seems necessary if we wish a holistic encounter of the subject.

REFERENCES
Arendt, H. (1958). The Human Condition. Univ. of Chicago Press, Chicago, Ill.
Atiyah, M. et al. (1994). Responses to Jaffe, A. & Quinn, F.: Theoretical Mathematics: Towards a Cultural Synthesis of Mathematics and Theoretical Physics. Bulletin of the
AMS, 30(2).
Baez, J. (1996). This Weeks Finds in Mathematical Physics. (Week 74, 80, 89). URL: http://
math.ucr.edu/home/baez/week*.html.
Bell, J. L. (1986). From Absolute to Local Mathematics. Synthese, 69, 409-426.
Benacerraf, P. (1965). What Numbers Could Not Be. Philosophical Review, 74, 47-73.
Bernays, P. (1950/ 1976). Mathematische Existenz und Widerspruchsfreiheit. In
Abhandlungen zur Philosophie der Mathematik. Wissenschaftliche Buchgesellschaft,
Darmstadt.
Boolos, G. (1986). Saving Frege from Contradiction. Proceedings of the Aristotelian Society,
87, 137-151.
Boolos, G. (1987). The Consistency of Freges Foundations of Arithmetic. In Thomson, J. J.
(ed.), On Being and Saying. MIT Press, Cambridge, MA, pp. 3-20.
Campbell, J. I. D. (Ed.) (2005). Handbook of Mathematical Cognition. Psychology Press, An
Imprint of Taylor and Francis.
Cantlon, J. F., Brannon, E. M., Carter, E. J. & Pelphrey, K. A. (2006). Functional imaging of
numerical processing in adults and 4-y-old children. PLoS Biol, 4(5), 125.
Card, C. R. (1991). The Archetypal View of C.G. Jung and W. Pauli. Psychological Perspectives, 24, 19-33 & 25, 52-69. (Los Angeles: C. G. Jung Institute).
Card, C. R. (1993). The Archetypal Hypothesis of W. Pauli and C.G. Jung: Origins, Development and Implications. In K. V. Laurikainen and C. Montonen, eds., Proceedings of
the Symposium on the Foundations of Modern Physics - 1992 Helsinki, Finland (pp.
139-151). World Scientific Publishing Co, Singapore.
Card, C. R. (1995). A New Archetypal Science. Psychological Perspectives, 31, 62-64.

On Numbers Nature

53

Card, C. R. (1996). The Emergence of Archetypes in Present-Day Science and its Significance
for a Contemporary Philosophy of Nature. Konigshausen & Neumann, Wurzburg, Germany.
Card, C. R. (2000). The Emergence of Archetypes in Present-day Science and their Significance for a Contemporary Philosophy of Nature. In B. Goertzel, A. Coombs & M. Germine (eds.). Mind in Time. Hampton Press, Creskoll, N.J.
Cohen, E. D. (1975). C. G. Jung and the Scientific Attitude. Philosophical Library, NY.
Cohen, P. J. (1963). The Independence of the Continuum Hypothesis I. Nat. Acad. Scien.
USA, 50, 1143-1148.
Cohen, P. J. (1964). The Independence of the Continuum Hypothesis II. Nat. Acad. Scien.
USA, 51, 105-110.
Cohen, P. J. (1966). Set Theory and the Continuum Hypothesis. W. A. Benjamin, Inc, NY.
Conway, J. H. (1976). On Numbers and Games. Academic Press, London.
Dantzing, T. (1954). Number and the Language of Science. MacMillan Co, NY.
Dedekind, R. (1888/ 1963). The Nature and Meaning of Numbers. In R. Dedekind, Essays on
the Theory of Numbers (trans. W. Beman). Dover, N.Y.
Dehaene, S. (1997a). The Number Sense. Oxford University Press, NY.
Dehaene, S. (1997b).The American Pi versus the French Pi?
www.edge.org/discourse/hersh_number.html
Dehaene, S. (1997c). What Are Numbers, Really? A Cerebral Basis for Number Sense
www.edge.org/3rd_culture/dehaene/index.html
Dehaene, S. (1999). The Number Sense: How the Mind Creates Mathematics. Penguin Books,
London.
Dehaene, S. (2001). Prcis of the Number Sense. Mind & Language, 16 (1), 1636.
Dehaene, S. (2003). The Neural Basis of the WeberFechner Law: A Logarithmic Mental
Number Line. Trends in Cognitive Sciences, 7(4), 145-147.
Dehaene, S., Molko, N., Cohen, L. & Wilson, A. J. (2004). Arithmetic and the Brain. Current
Opinion in Neurobiology, 14(2), 218-224.
Dummett, M. (1991). Frege: Philosophy of Mathematics. Duckworth, London.
Ernest, P. (1995). The Philosophy of Mathematics Education. Falmer Press, London.
Eves, H. (1983). Great Moments in Mathematics: After 1650. MAA, Washington, DC.
Ewald, W. B. (ed.) (1996). From Kant to Hilbert: A Source Book in the Foundations of
Mathematics. Clarendon Press, Oxford.
Fischer, M. H. & Rottmann, J. (2005). Do negative numbers have a place on the mental number line? Psychology Science, 47(1).
Franz, M.-L. von (1974). Number and Time. Rider, London.
Franz, M.-L. von (1980). On Divination and Synchronicity: The Psychology of Meaningful
Chance. Inner City Books, Toronto.
Franz M.-L. von (1992). Psyche and Matter. Shambhala, Boston & London.
Frege, G. (1884/ 1980). The Foundations of Arithmetic: A Logico-Mathematical Enquiry into
the Concept of Number. [trans. J. L. Austin]. Blackwell, Oxford.
Frege, G. (1893-1903/ 1967). Basic Laws of Arithmetic. [trans. M. Furth]. University of California Press, Berkeley, CA.
Gavalas, D. (1999). A Four-sided View of Function. FLM, 19(2), 38-41.
Gbel, S., Walsh, V. & Rushworth, M. F. S. (2001). The Mental Number Line and the Human Angular Gyrus. NeuroImage, 14(6), 1278-1289.

54

Dimitris Gavalas

Gdel, K. (1931). ber Formal Unentscheidbare Stze der Principia Mathematica und
Verwandter Systeme. Monatshefte fr Mathematik und Physik, 38, 173-198.
Gdel, K. (1958). The Consistency of the AC and of the GCH. Princeton University Press,
Princeton, NJ.
Gdel, K. (1947). What is Cantors Continuum Problem? American Mathematical Monthly,
54(9), 515-525.
Goodman, N. D. (1979). Mathematics as an Objective Science. American Mathematical
Monthly, 86(7), 540-551.
Goodman, N. D. (1990). Mathematics as Natural Science. The Journal of Symbolic Logic,
55(1), 182-193.
Goodman, N. D. (1991). Modernazing the Philosophy of Mathematics. Synthese, 88, 119-126.
Gough, W. C. & Shacklett, R. L. (1993). Science of Connectiveness. Part I: Modelling a
Greater Unity. Subtle Energies, 4(1), 57-76.
Hardy, G. H. (1992). A Mathematicians Apology. Cambridge University Press, GB.
Heck, R. (1993). The Development of Arithmetic in Freges Grundgesetze der Arithmetik.
Journal of Symbolic Logic, 58(2), 579-600.
Heijenoort, J. van (1967). From Frege to Gdel: A Source Book in Mathematical Logic,
18791931. Harvard University Press, Cambridge, MA.
Heinze, R. (1965). Xenokrates: Darstellung der Lehre und Sammlung der Fragmente. Georg
Olms Verlagsbuchhandlung, Hildesheim.
Hersh, R. (1997). What Kind of Thing Is a Number? A Talk with Reuben Hersh.
www.edge.org/3rd_culture/hersh/hersh_p1.html
Hilbert, D. (1900). On the Number-Concept. In Foundations of Geometry. Teubner, Leipzig
and Berlin.
Hill, S. R. (1992). Concordia: The Roots of European Thought. Gerald Duckworth, London.
Hofstadter, D. R. (1982). Analogies and Metaphors to Explain Gdels Theorem. Two -Year
College Mathematics Journal, March, 98-114.
Iamblichus (1922). Theologoumena Arithmeticae. Edidit Victorius de Falco. (Bibliotheca
Scriptorum Grecorum et Romanorum Teubneriana). Teubner, Leipzig.
Jaffe, A. & Quinn, F. (1993). Theoretical Mathematics: Towards a Cultural Synthesis of
Mathematics and Theoretical Physics. Bulletin of the AMS, 29(1).
Jung, C. G. (1957-1977). Collected Works. (CW). Princeton University Press, Princeton, NJ.
(Bollingen Series XX), 20 vols.
Jung, C. G. & Pauli, W. (1955). The Interpretation of Nature and the Psyche. Pantheon
Books, NY.
Kosko, B. (1994). Fuzzy Thinking: The New Science of Fuzzy Logic. Harper Collins Publishers, London.
Lakoff, G. & Nez, R. E. (2000). Where Mathematics Comes from: How the Embodied
Mind Brings Mathematics Into Being. Basic Books.
Lawvere, F. W. & Schanuel, S. H. (2005). Conceptual Mathematics: A First Introduction to
Categories. Cambridge University Press, Cambridge, UK.
Mac Lane, S. (1986). Mathematics: Form and Function. Springer-Verlag, NY.
Mac Lane, S. & Moerdijk, I. (1992). Sheaves in Geometry and Logic: A First Introduction to
Topos Theory. Springer-Verlag, NY.
Maddy, P. (1992). Realism in Mathematics. Clarendon Press, Oxford, GB.
McLarty, C. (1994). Numbers Can Be Just What They Have To. Nos, 27, 487-98.

On Numbers Nature

55

Meier, C. A. (1992). W. Pauli und C.G. Jung: Ein Briefwechsel. Springer-Verlag, Berlin.
Meier, C. A. (ed.) (2001). Atom and Archetype: The Pauli/ Jung Letters, 1932-1958. Preface
by Beverley Zabriskie. Princeton University Press, Princeton, N.J.
Nagel, E. & Newmann, J. R. (1973). Gdels Proof. New York University Press, NY.
Nagy, M. (1991). Philosophical Issues in the Psychology of C.G. Jung. SUNY Press, Albany.
Nelson, E. (1977). Internal Set Theory: A New Approach to Nonstandard Analysis. Bulletin
of the AMS, 83(6), 1165-1198.
Nelson, E. (1988). The Syntax of Nonstandard Analysis. Annals of Pure Appl. Logic, 38(2),
123-134.
O Meara, D. J. (1990). Pythagoras Revived. Oxford Scholarship Monographs, GB.
Pauli, W. (1955). The Influence of Archetypal Ideas on the Scientific Theories of Kepler. In
C. G. Jung and Wolfgang Pauli, The Interpretation of Nature and the Psyche. Pantheon
Books, NY.
Peat, F. D. (1987). Synchronicity: The Bridge between Matter and Mind. Bantam, NY.
Potter, M. (1998). Routledge Encyclopaedia of Philosophy. Routledge, London.
Priest, G. (1998). Routledge Encyclopaedia of Philosophy. Routledge, London.
Principia Cybernetica. On line: pespmc1.vub.ac.be
Ramachandran, V. S. & Blakeslee, S. (1998). Phantoms in the Brain: Probing the Mysteries
of the Human Mind. William Morrow, NY.
Resnik, M. (1981). Mathematics as a Science of Patterns: Ontology and Reference. Nos, 15,
529-50.
Resnik, M. (1982). Mathematics as a Science of Patterns: Epistemology. Nos, 16, 95-105.
Robertson, R. (1995). Jungian Archetypes: Jung, Gdel and the History of Archetypes. Nicolas-Hay Publishers, York Beach, Maine.
Robertson, R. (1989). The Evolution of Number: The Archetype of Order.
www.secamlocal.ex.ac.uk/~mwatkins/isoc/jungianNT.htm
Rucker, R. (1988). Mind Tools: The Mathematics of Information. Penguin Books, NY.
Shapiro, S. (1983). Mathematics and Reality. Philosophy of Science, 50, 523-48.
Shapiro, S. (1989). Logic, Ontology, Mathematical Practice. Synthese, 79.
Stevens, A. (1982). Archetypes: A Natural History of the Self. Routledge and Kegan Paul,
London
Uspensky, V. A. ( 1987). Gdels Incompleteness Theorem. Mir Publishers, Moscow.
Vopenka, P. (1979). Mathematics in the Alternative Set Theory. Teubner, Leipzig.
Waerden, B. L. van der (1975). Science Awakening . Kluwer Academic Publishers B. V.
Warner, R. & Szubka, T. (Eds.) (1995). The Mind-Body Problem: A Guide to the Current
Debate. Blackwell, Cambridge, USA.
Waterfield, R. (1988a). The Theology of Arithmetic: On the Mystical, Mathematical and
Cosmological Symbolism of the First Ten Numbers. Attributed to Iamblichus. Phanes
Press.
Waterfield, R. (1988b). Emendations of [Iamblichus], Theologoumena Arithmeticae (De
Falco). The Classical Quarterly, New Series, 38(1), 215-227.
Weyl, H. (1927/ 1949). Philosophy of Mathematics and Natural Science. Princeton University Press, Princeton, NJ.
Wittgenstein, L. (1974). Philosophical Investigations. Blackwell, Oxford, GB.
Wright, C. (1983). Freges Conception of Numbers as Objects. Aberdeen University Press,
Aberdeen.

56

Dimitris Gavalas

Zalta, E. (1998). Freges Logic, Theorem and Foundations of Arithmetic. Stanford Encyclopaedia of Philosophy. plato.stanford.edu/entries/frege-logic/Q0.html.
Zalta, E. (1999). Natural Numbers and Natural Cardinals as Abstract Objects: A Partial Reconstruction of Freges Grundgesetze in Object Theory. Journal of Philosophical Logic,
28(6), 619-660.

In: Mathematics and Mathematical Logic: New Research

ISBN 978-1-60692-862-2
Editors: Peter Milosav and Irene Ercegovaca
2010 Nova Science Publishers, Inc.

Chapter 2

THE APPLICATION OF FUZZY LINEAR

PROGRAMMING IN ENGINEERING
C. Riverol*
Chemical Engineering Department, University of West Indies, Trinidad
Republic of Trinidad and Tobago

ABSTRACT
As will be seen, the mathematical apparatus of the fuzzy theory of fuzzy sets provides a natural basis for the theory of the possibility as well described [1]. Viewed in this
perspective, a fuzzy restriction may be interpreted as a possibility distribution with its
membership function playing the role of possibility distribution function. This principle
will be described in this chapter.

INTRODUCTION
In traditional designs, the optimization problem is stated in precise mathematical terms.
However, in many real world problems, the design data, objective function and constraints are
stated in vague and linguistic terms. It appears that it is more reasonable to have a transition
state from absolute permission to absolute non-permission. This implies that the constraint is
to be stated involving vague and imprecise information. In the literature several applications
of the fuzzy logic in optimization have been reported. [2, 3, 4 and 5].
A historical review about fuzzy mathematical programming points to Bellman and Zadeh
[5] in a first category treating decision-making problem under fuzzy goals and constraints for
first time. The fuzzy goals and constraints represent the flexibility of the target values of objective functions and the elasticity of constraints. From this point of view, this type of fuzzy
mathematical programming is called the flexible programming. The second category in fuzzy
mathematical programming treats ambiguous coefficients of objective functions and con*

E-mail: criverol@eng.uwi.tt

58

C. Riverol

straints but does not treat fuzzy goals and constraints. Dubois and Prade [6] treated systems of
linear equations with ambiguous coefficients suggesting the possible application to fuzzy
mathematical programming from the first time. A remarkable development was done by
Kuzmin [7]. He introduced four inequality indexes between fuzzy numbers based on the possibility theory into mathematical programming problems with fuzzy coefficients. Since the
fuzzy coefficients can be regarded as possibility distributions on coefficient values, this type
of fuzzy mathematical programming is usually called the possibilistic programming
In fact, using possibilistic optimization approach, a solution can be achieved that provides
a maximum degree of overall satisfaction [8, 9 and 10] . To determine an optimal solution,
decision problems may be formulated as a fuzzy decision model, particularly when the available data are known exactly though varying within a tolerance limit. The coefficients of some
constraints may be fuzzy numbers and the original fuzzy problem is transferred into a crisp
satisfactory model [11]. In this paper, a Fuzzy Linear Programming (FLP) model subjected to
some linear fuzzy constraints [10, 11 and 12] as well as some crisp constraints, have been
developed and transferred into a crisp model.

STRUCTURE OF THE MODEL:THEORY BACKGROUND

The methodology for fuzzy linear programming (FLP) has been widely referenced in the
literature. The approach proposed here is based on an interaction with the decision maker, the
implementer and the analyst in order to find a compromised satisfactory solution for a FLP
problem. In a decision process using FLP model, the source resource variables may be fuzzy,
instead of precisely given numbers as in crisp linear programming (CLP) model [14 and 15].
For example, machine hours, labor force, material needed and so on in a manufacturing center, are always imprecise, because of incomplete information and uncertainty in various potential suppliers and environments. Therefore, they should be considered as fuzzy resources,
and the FLP problem should be solved by using fuzzy set theory [15 and 16].
There is often confusion about fuzzy and possibilistic optimization. Fuzzy and possibilistic entities have different meanings/semantics. Fuzzy and possibility model different entities
and the associated solution methods are different. Fuzzy entities, as is well known, are sets
with non-sharp boundaries in which there is a transition between elements that belong and
elements that don't belong to the set [18]. Possibilistic entities are obtained from sets that are
classical sets (crisp) but the evidence associated with whether a particular element belongs to
the (crisp) set or not is incomplete or hard to obtain.
Very simply, fuzzy decision making selects from a set of crisp elements while possibility
selects from a set of distributions. The underlying sets associated with fuzzy decision-making
are fuzzy where one forms the decision space of crisp elements from operations (''and'' in the
case of optimization, that is, constraints) on these fuzzy sets. The underlying sets associated
with possibilistic decision-making are crisp where one forms the decision space of distributions from operations on crisp sets. Possibilistic distributions encapsulate the best estimate
about the value of an entity given the available information.
Fuzzy membership function values describe the degree to which an entity is that value. A
possibility of one means that the value of the entity has the highest possibility of being what
distribution defines. If the fuzzy membership value is one, then it is definitely the value. Thus

The Application of Fuzzy Linear Programming in Engineering

59

the nature of decision making in the presence of fuzzy/possibilistic uncertainties are quite
different in semantics and optimization procedures since fuzzy optimization optimizes over
sets of numbers and possibility optimizes over sets of distributions.
A possibilistic linear function value cannot be determined uniquely since its coefficients
are ambiguous, i.e, non-deterministic [16]. Thus, the objective, maximizing or minimizing a
possibilistic function and the constraint that a possibilistic linear function value is not greater
than a certain value do not specifically make sense. To make them clear; consider the following equations:

A ( B) = sup min( A (r ), B (r ))
N A ( B) = inf max(1 A (r ), B (r ))

(1)

where B is the membership function of the fuzzy set B. A (B ) evaluates to what extent it is possible that the possibilistic variable, , restricted by the possibility distribution A
is in the fuzzy set B. On the other hand, NA(B) evaluates to what extent it is certain that the
possibilistic variable restricted by the possibility distribution A is in the fuzzy set B.
Let be a possibilistic variable. In context to the above example, let B = (, g ] , i.e.,
B be a crisp (non-fuzzy) set of real numbers, which is not greater than g. Then we obtain the
following indexes by possibility and necessity measures defined by:

Pos ( g ) = A ((, g ]) = sup{ A (r ) r g }

(2)

Nes ( g ) = N A ((, g ]) = 1 sup{ A (r ) r > g }

(3)

Pos( g ) and Nes( g ) show the possibility and certainty degrees to what extent
is not greater than g. Those indexes using a membership triangular function as example are
depicted in Figure 1.
Similarly, letting B = [ g ,+ ) , we obtain the following two indices:

Pos ( g ) = A ([ g ,+)) = sup{ A (r ) r g }

(4)

Nes( g ) = N A ([ g ,+)) = 1 sup{ A (r ) r < g }

(5)

Pos( g ) and Nes( g ) show the possibility and certainty degrees to what extent
is not smaller than g. Those indexes are depicted in Figure 2. Thus we can get the possibility and certainty degrees to what extent a possibilistic linear function value is not greater
(smaller) than a given real number.

60

Figure 1. Possibility and necessity measures of

C. Riverol

g.

The Application of Fuzzy Linear Programming in Engineering

Figure 2. Possibility and necessity measures of

61

g.

PYROLYSIS OF ETHANE AS STUDY CASE

The ethane is pyrolyzed during the ethylene production using a furnace [17]. The feed enters the furnace convection and it is pre-heated to 833-972K by the flue gases before being
further heated in the radiation section. Dilution steam, added to inhibit coke formation, is also
preheated in the convection section before being mixed with the feed. The ethane-steam mixture is divided among parallel coils in the firebox. The mixture enters the radiation section at

62

C. Riverol

414-552 kPa where the ethane is pyrolyzed to the primary products of hydrogen, methane and
ethylene, along with a mixture of smaller amounts of C3 to C5 hydrocarbons. The effluent
leaves the furnace at 350-450 kPa and 833-972K. The furnace effluent is cooled in a transfer
line exchanger where the steam is recovery. Seven equations to describe the pyrolysis of ethane:

C2 H 6 C2 H 4 + H 2

(6)

C 2 H 4 + 2 H 2 2CH 4

(7)

C 2 H 4 0.25C 4 H 6 + 0.125C 4 H 8 + 0.125C 4 H 10 + 0.125 H 2

(8)

C 2 H 4 2C + 2 H 2

(9)

C2 H 4 C2 H 2 + H 2

(10)

C 2 H 4 + C 2 H 6 0.952C3 H 6 + 0.381C3 H 8 + 0.62 H 2

(11)

C 2 H 4 0.333C6 H 6 + H 2

(12)

For developing the model, the necessity and possibilistic measures indicated the equations (2 and (3). The model consists in a nlinear objective function and linear constraints. The
final model is as follow:

Min C i x i
n

(13)

Where C are the costs. The major costs used in this model are:

x2

Cost of reactor: CR = CRB

x4 4.99 107
450kPa

FC ( x 900)
p
3

Cost of furnace: C F = C FB

Cost of Steam: Cs = CSB x2

The Application of Fuzzy Linear Programming in Engineering

63

Where x1 is the ethane conversion (Ce=Ce0(1-x1)), x2 is the steam hydrocarbon ratio, x3 is

the inlet temperature and x4 is the inlet pressure. CRB, CFB and CSB are basis costs that can be
calculated using Guthrie correlations [17].
Subject to

Nec a i xi bi

(a )x + (a

Pos aij xi bi

(a )x + (1 )(a

ai xi bi

(14)

ai xi bi

(15)

Where is the level or degree of optimism for the satisfaction of the constraint.
Here ai and bi are triangular fuzzy numbers: 0 / ai / ai and 0 / bi / bi crisps and

[0 1]

with =0.0, 0.25, 0.50, 0.75 and 1.00 is used (these values were selected arbitrarily). Any values between [0 1] can be used. The values of a and b can be found in the Tables 1 and 2.
Table 1. ai fuzzy intervals

Minimum
Average
Maximum

Ethane Conversion
(x1)
0.55
0.59
0.62

Steam/hydrocarbon
ratio (x2)
0.45
0.52
0.67

Inlet temperature
(x3)
833
910
972

Inlet pressure
(x4)
414
490
522

Table 2. bi values
Variable
Ethane Conversion (x1)

Value
1.00

Steam//hydrocarbon ratio (x2)

0.8

Inlet temperature (x3)

Inlet pressure (x4)

980 K
455 kPa

Since the furnace has to be thermally efficient to be economical, the optimum combination of variables will tend to balance in the total cost. In Table 3 is shown the values of the
decision variables to different degree of optimism. The optimum values were obtained when
the degree of optimism is equal to 1.0, where: the conversion is equal to 0.61, the
steam/hydrocarbon ratio is equal to 0.61, the inlet temperature is 865K and the inlet pressure
is 445 kPa.

64

C. Riverol

The criterion used is based on an equilibrium approach (EA) dimensionless parameter.

Through experience in reactor design, it has been concluded that coke formation is not excessive if the approach to equilibrium of the primary decomposition of ethane is kept below 0.65.
The combination when =1.0 is realistic because EA, EA =

NC2 H 4 * N H 2 * P
N C 2 H 6 * N * 4.99 10 7

is below 0.65, the composition of the effluent gas is more according to our filed data and the
heat flow was below 14200 kJ/m2K.
It is worth noting that the result can change if another fuzzy function is used (S-shape or
Gaussian). We used triangular fuzzy function because this function adjusted satisfactory to
our data. Moreover, the objective function is independent of the average used because the
possibility and necessity measures are satisfied of the same manner.
Table 3. Composition of the effluent gas at different degrees of optimism and
operating conditions
Component/
H2
CH4
C2H2
C2H4
C2H6
C3H6
C3H8
C4
C5

0.00
35.6
5.2
0.21
33.3
23.8
1.2
0.19
0.45
1.15

0.25
34.8
5.2
0.15
32.8
23.8
0.9
0.19
0.45
1.18

0.50
35.3
4.9
0.16
33.1
24.6
0.9
0.18
0.49
1.18

0.75
34.7
4.8
0.16
32.6
23.7
0.91
0.18
0.45
1.17

1.00
35.0
5.2
0.15
33
23.6
0.91
0.19
0.44
1.13

CONCLUSION
Real world problems are not usually so easily formulated as mathematical models or
fuzzy models. Sometimes qualitative constraints and/or objectives are almost impossible to
represent in mathematical forms. The fuzzy solutions have not yet been investigated considerably. In this chapter depicts as a fuzzy programming can be useful in the formulation of
problems of any branch of the engineering, especially in relations to the roles which these
concepts play in natural languages in order to simulate the remarkable human ability to attain
imprecisely defined goals in a fuzzy environment.

REFERENCES
[1]
[2]

Zadeh L., (1978) Fuzzy sets and systems 1, 3-28

Bellman, R.E and L.A. Zadeh, (1970) Management Science 17 141-164.

The Application of Fuzzy Linear Programming in Engineering

[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]

[15]
[16]
[17]

65

Luhandjula, M. K. (1986). Fuzzy Sets and Systems 18 15-30.

Delgado, M., Verdegay, J.L. and Vila, M. A. Fuzzy Sets and Systems 29 (1989)21-29.
Jiuping, X. (2000) A Journal Of Chinese Universities 15(1) 65-72.
Dubois, D. and Prade, (1980) H.. Fuzzy Sets and Systems, Theory and Applications,
California Academic Press Inc.
Kuzmin, V. B. A . (1981) Fuzzy Sets And Systems 6 27-41.
Chanas, S. (1983) Fuzzy Sets and Systems 11 243-251.
Maleki, H. R., Tata, M and Mashinchi, M. (2000) Fuzzy Sets and Systems 109 21-33.
Ross, T, Booker J and Parkinson WJ. (2002) Fuzzy logic and probability applications.
Bridging the gap, SIAM-ASA Series, Philadelphia, Penn.
Carlsson, C., Korhonen, P.A. (1986) Fuzzy Sets and Systems 20. 7-30.
Negoita, C. V. (1981) Fuzzy Sets and Systems 6 261-269.
Sengupta, A., Pal, T. K and Chakraborty, D. (2001) Fuzzy Sets and Systems 119 129138.
Tabucanon, M.T. (1996) Multi Objective Programming for Industrial Engineers.
Mathematical Programming for Industrial Engineers: Marcel Dekker, Inc, New York
487-542.
Rommenlfanger, H. (1996) European Journal of Operational Research 92 512-527.
Inuiguchi M and Ramik J. (2000) Fuzzy set and Systems, 111 3-28.
Riverol C and MV. Pilipovik. (2007) Chemical Engineering Journal, 133, 133-137.

In: Mathematics and Mathematical Logic: New Research

ISBN 978-1-60692-862-2
Editors: Peter Milosav and Irene Ercegovaca
2010 Nova Science Publishers, Inc.

Chapter 3

A NEW CONSENSUS SCHEME FOR MULTICRITERIA

GROUP DECISION MAKING UNDER
LINGUISTIC ASSESSMENTS
P. Bernardes, P. Ekel and R. Parreiras*
Pontifcia Universidade Catlica de Minas Gerais
Av. Dom Jose Gaspar, 500, 30535-610, Belo Horizonte, MG, Brazil

ABSTRACT
This work proposes a new consensus scheme for group decision making, which allows one to obtain a consistent collective opinion, from information provided by each expert in terms of multigranular fuzzy estimates. It is based on a linguistic hierarchical
model with multigranular sets of linguistic terms, and the choice of the most suitable set
is a prerogative of each expert. From the human viewpoint, using such model is advantageous, since it permits each expert to utilize linguistic terms that reflect more adequately
the uncertainty level intrinsic to his evaluation. From the operational viewpoint, the advantage of using such model lies in the fact that it allows one to express the linguistic information in a unique domain, without information losses, during the discussion process.
Such consensus scheme is applied in the analysis of a multicriteria decision problem,
generated with the use of the Balanced Scorecard methodology for enterprise strategy
planning. Three techniques for multicriteria analysis, based on fuzzy preference relation
modeling, are considered. They permit the evaluation, comparison, selection, prioritization, and/or ordering of alternatives with the use of both quantitative and qualitative estimates. With the availability of different techniques, the most appropriate one can be chosen, considering possible sources of information and its uncertainty.

P. Bernardes: patib@pucminas.br
P. Ekel: ekel@pucminas.br
*R. Parreiras: roberta.parreiras@terra.com.br

68

P. Bernardes, P. Ekel and R. Parreiras

1. INTRODUCTION
Multiperson multicriteria decision making involves many complex and conflicting aspects intrinsic to human individuality and human nature. For instance, when a team of experts
takes part in the decision process, it is expected that their opinions disagree. Frequently, each
member of the group has different information at hand and partially shares the goals of the
other members. Thus, a satisfactory solution is the one that is the most acceptable by the
group as a whole, as a unanimously best solution almost never exists.
Besides, diverse types of uncertainty are often encountered in a group decision process.
The main uncertainty factors are related to the decision-makers (DM) role; the individual
judgments concerning the importance of each criterion and the preferences for alternatives
[1]. Fuzzy set theory might provide the needed flexibility to adequately deal with such uncertainty factors [1-3]. Particularly, in the multiperson multicriteria decision analysis, the use of
linguistic terms has been considered a realistic way to express the subjectivity and uncertainty
intrinsic to human preferences [4,5].
Each linguistic estimate is characterized by both a label (syntactic value) and a meaning
(semantic value). The label can be a word or sentence from a linguistic term set. The meaning
is a fuzzy set defined in a universe of discourse. When characterizing an object, the linguistic
term set adopted to express this information should have more or less terms, in accordance
with the uncertainty degree of the DM perception. The DM can make mistakes if he is enforced to make more precise judgments than he is capable. On the other hand, a substantial
loss of information may happen if the DM is enforced to make less precise judgments. In
other words, the granularity of uncertainty (i.e., the size of the linguistic term set utilized to
express the information) is a relevant parameter from the linguistic model and is intimately
related to the validity of using this model [4,5].
In this context, a methodology for solving multiperson multicriteria decision problem in a
fuzzy environment, with multigranular linguistic term sets, is introduced. It consists in a consensus scheme, based on a fuzzy aggregation model where the linguistic term sets are structured as a linguistic hierarchy [4]. The advantage of using such hierarchical model lies in the
fact that it permits the multigranular linguistic information to be expressed in a unified linguistic domain without loss of information [4].
The term consensus scheme is utilized here to denote a dynamic and iterative process,
where the involved experts discuss a multicriteria decision problem, under the supervision of
a moderator. All experts are invited to evaluate the alternatives, taking into account each criterion, on the basis of linguistic terms, organized as a linguistic hierarchy of linguistic terms.
The choice of the most adequate set to express the preferences is a prerogative of each DM.
At each cycle of the discussion process, the most discordant member of the group is identified, with the use of an index that reflects the discordance level among the experts, and invited to update his opinion.
The proposed consensus scheme has some positive aspects: it involves simple calculus; it
is intuitive, flexible and easy to use. Furthermore, it admits a computational component for
executing supervision functions that are usually delegated to a human moderator. In order to
demonstrate its usefulness, the consensus scheme is utilized in order to solve an enterprise
strategy planning problem, generated with the use of the Balanced Scorecard methodology
[6]. The multicriteria analysis is performed on the basis of thee different procedures described

A New Consensus Scheme for Multicriteria Group Decision Making

69

in [7,8]. These procedures have been successfully applied in diverse areas such as: power engineering [9], management [10] (including group decision problems [11]), and naval engineering [12]. It should be emphasized that the possibility to obtain different solutions on the
basis of each approach is to be considered natural. The most appropriate procedure should be
chosen, considering possible sources of information and its uncertainty.

2. MULTIPERSON MULTICRITERIA DECISION PROBLEM

In this chapter, the decision problem consists in ranking from best to worst the alternatives belonging to a set X, according to the criteria from a set C and considering the collective
opinion of a team E of experts. The following assumptions are made:

The set of alternatives X = {x1 , x2 ,..., xn } is finite, discrete and contains two or

more alternatives.
The set of criteria C = {c1 , c2 ,..., cm } (according to which the alternatives are evalu-

ated and compared) contains two or more criteria of qualitative or quantitative nature.
The team of experts involved in the decision process, E = {e1 , e2 ,..., eq } , contains

two or more qualified professionals.

The estimate provided by the yth expert for the ith alternative, considering the kth crite-

~y

rion is represented as Fk ( xi ) .

All experts are supposed to give their preferences on the basis of linguistic term sets
from the same linguistic hierarchy.

3. LINGUISTIC HIERARCHY
In the group decision context, each professional involved has his own perception of the
problem. Hence, each DM may feel more comfortable using a different linguistic term set to
express his preferences. Besides, frequently, the criteria, which are to be taken into account in
the process of decision making, are reflected by data of very different nature (for instance,
"comfort of operation", "flexibility of development", etc.). As a consequence, each DM may
prefer to utilize a different linguistic term set for each criterion. It is also common to meet, in
practice, circumstances where, when considering a single criterion, a DM chooses a linguistic
term set to evaluate one alternative and another linguistic term set (for instance, with lower
granularity) to evaluate other. In this context, the proposed consensus scheme provides sets of
linguistic terms with different sizes, organized as a linguistic hierarchy. The choice of the
most satisfactory linguistic term set is always a prerogative of each DM.
A linguistic hierarchy consists in a set of levels, being each level a linguistic term set
with different granularity from the remaining levels of the hierarchy [4]. It must be constructed on the basis of the following rules:

70

P. Bernardes, P. Ekel and R. Parreiras

the modal points of the membership functions of each level must be preserved in the
following level;
the transition between successive levels must be smooth. Each new linguistic term is
added between pairs of terms from the previous level. A graphical representation of a
linguistic hierarchy of three levels is shown in Figure 1.

Figure 1. Linguistic hierarchy of trapezoidal linguistic estimates.

The linguistic information may be represented by a 2-tuple (l,a), where l is a linguistic

label and a is the scalar value that reflects the difference of information resulted from a symbolic translation. Assume that L={l0,l1,,lg} is a linguistic term set, b [0, g ] is the result of
a symbolic aggregation operation of the indexes of a subset of labels from L, and
a [-0.5,0.5) is defined as a=b-i, where i=round(b). Then, a reflects the difference of information between b and its closest value i {0,1,..., g} . The 2-tuple (li,a) associated with b
can be obtained with the following correlation:

l , i = round(b),
(b) = i
a = b i.

(1)

The inverse function that provides the value of b associated with a 2-tuple is given by

1 (li , a) = i + a = b .

(2)

Assume that t is an index that specifies the level of the hierarchy and that n(t) is the
granularity of the linguistic term set of level t. Given a linguistic hierarchy, the transformation
t

t'

function TFt ' from a label li in level t to a label l j in level t ' can be defined as:

1 (lit , a)(n(t ' ) 1)

.
TFt t' (lit , a) =
n(t ) 1

(3)

A New Consensus Scheme for Multicriteria Group Decision Making

71

In order to exemplify the use of (3), consider the linguistic hierarchy shown in Figure 1.
3

The transformation from a label l1 in level 3 to a label l j in level 1 is carried out as:

1 (l13 ,0)(3 1)
(1 + 0)2 = (0.25) = (l 1 ,0.25) .
=
TF32 (l13 ,0) =

9 1
8

It is worth mentioning that, in this chapter, we deal with hierarchies whose linguistic term
sets have an odd value for their respective granularity and whose membership functions are
trapezoidal-shaped, symmetrical and uniformly distributed in the interval [0,10]. The central
label represents the indifference concept.

4. MULTIPERSON AGGREGATION MODEL

When considering opinions of different experts, sometimes, it is relevant to differentiate
the influence levels of the experts in determining the collective opinion. However, it is interesting to point out two types of influence. One is related to each experts role in the group,
which is determined by his knowledge and experience on the problem. For instance, the opinion of a professional who has a deeper knowledge on the problem may be considered more
relevant than the others. The most common way to include such kind of influence in a multiperson aggregation model is by assigning a different importance coefficient (weight) to each
member of the group. These weights are usually specified by a human moderator [1,10,13].
The other kind of influence is related to the experts opinions, being proportional to the concordance level between each expert and the rest of the group. Such kind of influence depends
on the way the group or the moderator prefer to deal with concordance measures: the more
concordant opinions may be emphasized (which is the attitude of our new aggregation model
and is also the most common attitude) [14-16]; equal emphasis may be given to all opinions
or more importance may be attached to the most discordant opinions [16]. In [17,18], this
second kind of influence is implemented through the use of scalar weights, which are especially chosen, in a systematic way, with the purpose of maximizing the consensus level.
The proposed multiperson aggregation model consists in a weighted sum of the individt

ual estimates. All operations are performed in terms of the parameters b y , y=1,...,q, already
translated to the same level t from the linguistic hierarchy. It may be interesting to translate
all opinions to that level t of the hierarchy, which was used by the majority of the specialists,
in order to maintain the uncertainty level intrinsic to most of the judgments.
Given the distance (which can be considered a discordance measure) between the opinions of the yth expert and of the zth expert, calculated on the basis of the parameter b:

d yz =| byt bzt | ,
each weight Cy, y {1,..., q} is calculated as follows:

(4)

72

P. Bernardes, P. Ekel and R. Parreiras

Cy =

n(t ) d yz
.
wz
z =1 z y
n(t )
q

(5)

In (5), the parameter wz is the importance weight associated with the opinion of ez, satisq

fying

w
y =1

= 1 . A human moderator must specify these weights, taking into account the

experience or knowledge of each expert: higher weights should be associated with the opinions of those expets who have deeper knowledge on the problem. As in [14,15], the influence
of ey in determining the collective estimate is higher when ey agrees with the experts whose
importance weights wz, z {1,2,..., q} z y are higher.
The parameter bc associated with the collective estimate is determined by the weighted
sum of individual estimates, on the basis of the parameter b as follows:
q

b = C z bz .
c

(6)

z =1

In (6), the normalized weight C z relative to the zth DM is calculated by

Cy =

Cy

C
z =1

It is interesting to observe in (5) that lower values of dyz correspond to higher values of
Cy. Thus, it can be said that the proposed aggregation model emphasizes the more concordant
opinions. Finally, the pair (li,a) associated with bc can be obtained by using Equation (1).

4. CONSENSUS SCHEME
Usually, in an organization, the decisions require multiple perspectives of different experts, as a unique person may not have enough knowledge on the problem to solve it alone.
Naturally, each expert may have different information at hand and a distinct subjective perception of the problem (which results in different opinions and preferences). In this way, the
group should aim at achieving a satisfactory solution, rather than the best solution, as it hardly
exists in practice. Classically, the term consensus is defined as a unanimous concordance
among all involved individuals. But, in practice, such definition is unsuitable for three main
reasons [4]:

it distinguishes only two states, the existence or nonexistence of consensus;

the chances of achieving such level of concordance are very low;
in practical situations, it is not necessary to achieve such level of concordance.

A New Consensus Scheme for Multicriteria Group Decision Making

73

In view of that, a concept of consensus level, denoted index of soft consensus, was proposed [19]. It allows evaluating not only crisp values: 1 for full and unanimous agreement and
0 for nonexistence of agreement, but also intermediate values between 0 and 1, corresponding
to partial agreement among all experts. Inspired on this key idea, several researchers have
developed consensus schemes (see [1,13,17-19], for instance).
The proposed consensus scheme has some positive aspects. It is simple, has intuitive appeal and admits a computational component for executing functions that are usually delegated
to a human moderator. Such component corresponds to a procedure that regulates the discussion process, supported by the discordance and consensus measures described next. It should
be emphasized that such measures must be computed after the fuzzy estimates given by the
experts are translated to the same level t of the hierarchy.
The discordance level between each expert ey, y {1,..., q} and the collective opinion is
given by the normalized distance:

dy =

| by b c |
n(t )

(7)

The consensus level is measured with the use of the following equation:
q

G=

y =1

n(t ) d y
q

(8)

The whole discussion process can be interrupted when any of the following conditions
are fulfilled:

An acceptable consensus level among the specialists is achieved.

The previously specified maximum number of iterations maxcycle is achieved.

The procedure that implements the guidance system for the discussion process is shown
next. It is assumed that elast is a vector utilized to store the index of the expert requested to
update his opinion, at each cycle; maxcycle stores the maximum number of cycles and must
be specified by a human moderator; minconsensus stores the minimum acceptable level of
consensus.

Guidance Procedure
Step 1) Initialize cycle=1, elast(cycle)=1 and ask the moderator to specify the value of the
input parameter minconsesus.
Step 2) Collect the opinion of each specialist.
Step 3) Identify the hierarchy level used by most of the specialists and make the needed data
translations to this level.
Step 4) Calculate the collective opinion.

74

P. Bernardes, P. Ekel and R. Parreiras

Step 5) Calculate the discordance level for each specialist and the consensus level achieved
by the group.
Step 6) If the maximum number of iterations or the minimum consensus level is achieved, go
to Step 10. If none condition is satisfied, go to Step 7.
Step 7) Identify the most discordant expert and verify, in vector elast, if he has been the most
discordant expert for the last two cycles. If it is true, repeat Step 7 for the second most
discordant expert and so on (in order to avoid to the same expert to be excessively requested).
Step 8) Add 1 to the value of variable cycle; store the index of the expert selected in Step 7 in
elast(cycle) and invite him to update his opinion.
Step 9) Collect the opinion of the selected expert and go to Step 3.
Step 10) Interrupt the procedure. The output data is the fuzzy estimate indicated by the index
c

i=round( b ).

5. FUZZY DECISION APPROACHES BASED ON FUZZY PREFERENCE

RELATION MODELING
Assume a set X of alternatives, which are to be examined following m criteria of quantitative and/or qualitative nature. The decision making problem can be represented through <X,
R> models, where X is a discrete set of alternatives and R = {R1 ,..., R m } is a vector of fuzzy
preference relations. In this case, the DM preferences, considering the kth criterion, is described by a fuzzy preference relation Rk X X , also called in the literature nonstrict
fuzzy preference relation, fuzzy binary preference relation, fuzzy weak preference relation,
and fuzzy binary relation. Its membership function Rk ( X i , X j ) : X X [0, 1] indicates
the degree to which the alternative Xi weakly dominates (or is at least as good as) Xj.
The entries of the pairwise comparison matrix Rk , k=1,,m, may be directly or indirectly defined by the experts. For instance, in [20], it is assumed that the DM directly specifies these entries. Reference [7], on the other hand, describes a natural way of constructing

these matrices, on the basis of fuzzy or linguistic estimates Fk ( X i ) , specified by the special-

ists. Given the estimates Fk ( X i ) , k=1,,m, i=1,,n, with membership functions,

[ Fk ( X i )] , the entries of Rk can be obtained as follows:

Rk ( X i , X j ) =

Rk ( X j , X i ) =

sup

min{[ Fk ( X i )], [ Fk ( X j )]} ,

(9)

sup

min{[ Fk ( X i )], [ Fk ( X j )]} ,

(10)

X i , X j X
Fk ( X i ) Fk ( X j )

X i , X j X
F k ( X j ) F k ( X i )

A New Consensus Scheme for Multicriteria Group Decision Making

75

if the kth criterion is associated with maximization. If the kth criterion is associated with
minimization, then (9) and (10) are written for Fk ( X i ) Fk ( X j ) and Fk ( X j ) Fk ( X i ) ,
respectively.
The nonstrict fuzzy preference relation Rk can be represented by a strict fuzzy preference relation Pk and an indifference relation I k : Rk = Pk I k . The strict preference relation Pk is constituted by all pairs of alternatives that satisfy the conditions: ( X i , X j ) Rk
and ( X j , X i ) Rk . If ( X i , X j ) Pk , it can be said that Xi is strictly better than Xj (or Xi
dominates Xj). The indifference relation, on the other hand, is constituted by all pairs of alternatives that simultaneously satisfy the conditions: ( X i , X j ) Rk and ( X j , X i ) Rk . If

( X i , X j ) I k , it can be said that Xi is indifferent to Xj.

With the use of the inverse relation ( Rk )

( ( X i , X j ) ( Rk )

is equivalent to

( X j , X i ) Rk ) the strict preference relation Pk can be defined as a difference of the fuzzy

1

sets: Pk = Rk \ ( Rk ) , and its membership function can be obtained as follows [21]:

Pk ( X i , X j ) = max{ Rk ( X i , X j ) Rk ( X j , X i ), 0} .

(11)

The use of (11) permits one to carry out the choice of alternatives. In particular, as
Pk ( X j , X i ) describes the set of all alternatives X i that are strictly dominated by X j , its
compliment Pk ( X j , X i ) corresponds to the set of alternatives that are not dominated by
other alternatives from X . Therefore, in order to meet the set of alternatives from X that
are not dominated by any other alternative, it suffices to obtain the intersection of all

Pk ( X j , X i ) . This intersection is the set of nondominated alternatives with the membership

function

nd
R k ( X i ) = inf {1 Pk ( X j , X i )} = 1 sup Pk ( X j , X i ) ,
X j X

(12)

X j X

which reflects the level of nondominance of each alternative X i .

Hence, a good choice for a monocriteria problem based on this model should be the alternatives providing:
nd
nd
X Rndk = { X ind X | nd
R k ( X i ) = sup R k ( X i )} .
X i X

(13)

76

P. Bernardes, P. Ekel and R. Parreiras

It is worth emphasizing that the alternatives satisfying X Rk = {X j X | Rk ( X j ) = 1}
nd

nd

nd

nd

are actually nonfuzzy nondominated and can be considered as the nonfuzzy solution for the
choice problem.
Expressions (11)-(13) may be used to solve choice or ranking problems not only with a
single criterion, but also with multiple criteria. Having at hand nonstrict preference relations
for each criterion, one possible procedure for solving multicriteria problems consists in obtaining a global relation through the intersection of these relations:

G ( X i , X j ) = min{ R1 ( X i , X j ),..., Rm ( X i , X j )} .

(14)

In this case, Equations (11)-(13) can be applied, taking into account the global relation
(14). The resulting fuzzy set of nondominated alternatives fulfills the role of a Pareto set. This
set can be contracted, in a subsequent analysis, where the importance of each relation Rk ,
k=1,,m is differentiated through their weighted aggregation:
m

T ( X i , X j ) = k Rk ( X i , X j ) .

(15)

k =1

In (15), the weights (or importance factors) of each criterion must satisfy the conditions:

k > 0 , k=1,,m;

k =1

= 1. With the construction of T ( X i , X j ) , it is possible to obtain

nd

the membership function T ( X i ) of the set of nondominated alternatives, following a procedure similar to the one described above for the global relation G ( X i , X j ) , which involves Equations (11)-(13). Finally, the nondominance level can be obtained by performing
nd

nd

the intersection between R ( X i ) and T ( X i ) :

nd
Wnd ( X i ) = min{ nd
R ( X i ), T ( X i )} ,

(16)

and the following set of fuzzy nondominated solutions can be met:

X nd = { X ind X | Wnd ( X ind ) = sup Wnd ( X i )} .

(17)

X i X

The second procedure has a lexicographic character. It consists in a step-by-step introduction of criteria for comparing alternatives. In this case, a sequence X1, X2,, Xm, such that

X X 1 ... X m , is obtained with the following expressions:

nd
Rk ( X j ) = infk 1{1 Pk ( X j , X i )} = 1 sup Pk ( X j , X i ) , k = 1,..., m ,
X i X

X i X k 1

(18)

A New Consensus Scheme for Multicriteria Group Decision Making

nd k
X k = { X ind k X k 1 | nd
) = sup nd
Rk ( X i
R k ( X i )} .

77
(19)

X j X k 1

In the third procedure, the use of (12) permits one to construct the membership functions
of the set of nondominated alternatives for each fuzzy preference relation. The membership
nd

functions Rk ( X i ) , k=1,,m play a role identical to membership functions replacing objective functions Fk ( X ) , k=1,,m in analyzing < X , M > models [7]. Therefore, in order to
obtain X

nd

, it is possible to construct:

nd ( X i ) = min nd
Rk ( X i ) .
1 k m

(20)

If it is necessary to differentiate the importance of each preference relation, it is possible

to rewrite (20) as follows:
k
nd ( X i ) = min[ nd
Rk ( X i )] .
1 k m

(21)

It should be mentioned that the use of (21) does not require the normalization of

k ,

k=1,,m as it is required to apply (17).

6. APPLICATION EXAMPLE
The enterprise's board of directors, which includes five members (e1,,e5), is to plan the
development of large projects (strategy initiatives) for the following five years. Four possible
projects (X1, X2, X3, and X4) have been marked. It is necessary to compare these projects to
select the most important of them, as well as order them from the point of view of their importance, taking into account four criteria (categories) suggested by the Balanced Scorecard
methodology (it should be noted that all of them are of the maximization type):
c1) financial perspective,
c2) the customer satisfaction,
c3) internal business process perspective,
c4) learning and growth perspective.
First, the specialists are asked to give their opinion relative to each project in terms of
fuzzy estimates, using the linguistic variables from the linguistic hierarchy shown in Figure 1.
The semantic values (trapezoidal fuzzy sets) corresponding to the labels from this hierarchy
are given in Table 1. The professionals involved are considered of the same importance, except for e1, whose opinions are judged more important. Therefore, the parameters wz are set
as: w1=0.3 and wz=0.175, z=2,,5. The parameter minconsensus is specified as 0.85 by the
moderator. It is interesting to observe that the proposed consensus scheme allows the discussion process to be divided into several sections (more specifically m n sections), being each

78

P. Bernardes, P. Ekel and R. Parreiras

section related to a specific alternative and a specific criterion. Only the sections regarding
the criterion c1 and alternative X1 are exposed here.
At the first cycle, as shown in Table 2, the consensus level is lower than the threshold parameter minconsensus specified by the moderator. In this way, as the expert e5 is the most
discordant one, he is invited to update his opinion. The data concerning the second cycle is
shown in Table 3. As it can be seen, the expert e5 made only a small change in his evaluation
and remained as the most discordant member of the group. Consequently, he is invited to review his opinion again.
Table 1. Fuzzy numbers associated with the linguistic terms of the linguistic hierarchy
Level
t=1

t=2

t=3

Granularity
3

Label

Fuzzy number
(0, 0, 0.5, 5)

1
0

l11

(0, 4.5, 5.5, 10)

l21

(5, 9.5, 10, 10)

l02

(0, 0, 0.5, 2.5)

l12

(0, 2, 3, 5)

l 22

(2.5, 4.5, 5.5, 7.5)

l32

(5, 7, 8, 10)

l 42

(7.5, 9.5, 10, 10)

l03

(0, 0, 0.5, 1.25)

l13

(0, 0.75, 1.75, 2)

l23

(1.25, 2, 3, 3.75)

l33

(2.5, 3.25, 4.25, 5)

l43

(3.75, 4.5, 5.5, 6.25)

l53

(5, 5.75, 6.75, 7.5)

l63

(6.25, 7, 8, 8.75)

l73

(7.5, 8.25, 9.25, 10)

l83

(8.75, 9.5, 10, 10)

Table 2. Fuzzy estimates given by each expert at cycle=1

e1

e2

e3

e4

e5

2
0

3
2

l13

3
4

2
3

0.16

0.18

Opinion

Discordance
Consensus level

0.27
0.79

0.05

0.40

Group
2.42

A New Consensus Scheme for Multicriteria Group Decision Making

79

Table 3. Fuzzy estimates given by each expert at cycle=2

e1

e2

e3

e4

e5

2
0

3
2

l13

3
4

3
6

0.16

0.18

Opinion

Discordance
Consensus level

0.27
0.79

0.04

Group
2.42

0.40

Table 4 presents the data collected at the third cycle of the discussion process. Although
e5 remained as the most discordant expert, as he had been already invited twice to review his
opinion, now e1 is invited (he is the second most discordant expert). It can be seen in Table 5
that, at the fourth cycle, the consensus level finally achieves a satisfactory value. Therefore,
the procedure is interrupted and the output estimate is given by (2.74) = (l33 ,0.26) .
Table 4. Fuzzy estimates given by each expert at cycle=3
e1

e2

e3

e4

e5

2
0

3
2

l13

3
4

3
5

0.15

0.19

Opinion

Discordance
Consensus level

0.26
0.82

0.03

Group
2.31

0.30

Table 5. Fuzzy estimates given by each expert at cycle=4

e1

e2

e3

e4

e5

2
1

3
2

l13

3
4

3
5

0.19

0.14

Opinion

Discordance
Consensus level

0.08
0.85

0.08

Group
2.74

0.25

Table 6. Collective evaluations of each alternative, considering each criterion

c1

c2

c3

c4

3
3

2
3

2
4

l12

l 42

l63

l32

l73

l33

l 23

l 43

l13

l53

l73

l 42

X1

X2

l32

X3
X4

The collective opinions related to each alternative, obtained with the use of the consensus
scheme, are exposed in Table 6 and illustrated in Figures 2-5. Having calculated the collective
estimates, the following comparison matrices, corresponding to the collective nonstrict fuzzy
preference relations, are obtained for each criterion, using (9) and (10):

80

P. Bernardes, P. Ekel and R. Parreiras

1
1
R1 =
1

1
0.63
1

1 0.91 1 ,
1
1
R2 =
0
1
1 1
0

0
0 1
0.91 0

1
0.45
R3 =
0

0.91

1 1

1 1
1 1 ,
1 0

1 1

1
0.45
0
1 0

1 1 0.83 ,
1 1
1
0.63 .

R4 =
1 0.45
0 1 0
1
0

1 1 1
1
1
1 1

Figure 2. Collective fuzzy estimates related to c1.

Figure 3. Collective fuzzy estimates related to c2.

A New Consensus Scheme for Multicriteria Group Decision Making

Figure 4. Collective fuzzy estimates related to c3.

Figure 5. Collective fuzzy estimates related to c4.

81

82

P. Bernardes, P. Ekel and R. Parreiras

The strict preference relations Pk, k=1,,4 can be obtained by applying (11) to each nonstrict preference relation Rk:

0
0
1
0
P1 =
1 0.09

0
0

0
0
0
0

1
0

0.37
1
, P2 =
0
1

0
0

0 0.55
0
0
P3 =
0
0

0 0.17

1 0.09
0
0
0
1
1
0
0
0.55
, P4 =
0.55
0
0
0
0

1
0
0.37
1
1

0
0
0
0

1 0.09
1
1
,
0
0

1
0
0
0
.
0

Let us consider the application of the first procedure for analyzing <X, R> models. The
aggregation of R1, R2, R3, and R4 into R, using (14), results in

1
0.45
R=
0

0
0
0
1 0.91 0.63
0 1
0

0
0
1

(22)

The strict fuzzy preference relation is derived from (22), using Equation (11):

0
0.45
P=
0

0
0
0
0 0.91 0.63
.
0
0
0

0
0
0

(23)

The nondominance level of each alternative, calculated with (12), is given by

nd
Rk

= [0.55 1 0.09 0.37] . Thus, according to the first procedure for analyzing multic-

riteria problems, the alternatives can be ranked from best to worst as X 2 ; X 1 ; X 4 ; X 3 .

The second procedure requires the ranking of criteria from the most important to the least
important one. Supposing that they can be ranked as c1 ; c2 ; c3 ; c4 , at the first step, the
nondominance

nd
R1

level

of

each

alternative

is

calculated

applying

(18)

to

P1 :

= [0 0.91 1 0] . As alternatives X1 and X4 are undistinguishable, the elements of P2

related to X1 and X4, can be analyzed as follows:

A New Consensus Scheme for Multicriteria Group Decision Making

83

0 0.09
P2 =
.
0
0
The nondominance level of X1 and X4 is given by: R2 = [1 0.91]. Therefore, the use
nd

of the second procedure provides the ranking: X 3 ; X 2 ; X 1 ; X 4 .

In the third procedure, having calculated the strict preference relations Pk, k=1,..,4, the
use of (12) provides the nondominance level of each alternative for each criterion:

nd
0.91 1 0] ,
R1 = [0

nd
0] ,
R 2 = [0.63 1 0

nd
0 0.91]
R3 = [1 0.45

and

nd
0.63 0 1] . The global nondominance level of each alternative is obtained on
R4 = [0
the basis of (20):

nd = [0 0.45 0 0] .
Hence, according to the third procedure, the alternatives X1, X3, and X4 are undistinguishable and we have X 2 ; X 1 ~ X 3 ~ X 4 .

CONCLUSION
The present work introduced a consensus scheme for analyzing decision problems, which
allows aggregating the opinions of multiple experts into a collective opinion. Each expert is
invited to evaluate the alternatives in terms of multi-granular linguistic terms that are organized as a linguistic hierarchy structure. The proposed consensus scheme has some positive
aspects: it has an intuitive appeal and involves simple calculus; it admits a computational
component for regulating the discussion process; it can be coupled to different procedures for
analyzing multicriteria problems. However, it is important to mention that loss of information
happens at the final step of the proposed consensus scheme, when the output data is generated
(more specifically, when the a parameter, which reflects the difference of information between bc and its nearest integer value, is eliminated).
In order to demonstrate the usefulness of the presented consensus scheme, three different
decision procedures based on analyzing <X, R> models are utilized for solving a hypothetical
enterprise strategy planning problem, generated with the use of the Balanced Scorecard methodology, by a group of experts.

ACKNOWLEDGEMENTS
This research is supported by the National Council for Scientific and Technological Development of Brazil (CNPq) - grant 302406/2005-0 and the State of Minas Gerais Research
Foundation (FAPEMIG) - grant TEC 00140/07.

84

P. Bernardes, P. Ekel and R. Parreiras

REFERENCES
[1]

[2]
[3]
[4]

[5]

J. Lu, G. Zhang, D. Ruan, and F. Wu, Multi-objective Group Decision Making: Methods, Software and Applications with Fuzzy Set Techniques, Imperial College Press,
London, 2007.
H. J. Zimmermann, Fuzzy Set Theory and Its Application, Kluwer Academic, Boston,
1990.
W. Pedrycz and F. Gomide, An Introduction to Fuzzy Sets: Analysis and Design, MIT
Press, Cambridge, 1998.
F. Herrera and L. Matnez, A model based on linguistic 2-tuples for dealing with multigranular hierarchical linguistic contexts in multi-expert decision-making, IEEE Transactions on Systems, Man and Cybernetics Part B: Cybernetics, vol. 31, 2001, pp. 227234.
F. Herrera and E. Herrera-Viedma, Linguistic decision analysis: steps for solving decision problems under linguistic information, Fuzzy Sets and Systems, vol. 115, 2000, pp.
67-82.

[6] R. S. Kaplan and D. Norton, The Balanced Scorecard: Translating Strategy

into Action, Harvard Business School, Boston, 1996.
[7] P. Ya. Ekel, M. R. Silva, F. Schuffner Neto, and R. M. Palhares, Fuzzy preference modeling and its application to multiobjective decision making,
Computers and Mathematics with Applications, vol. 52, 2006, pp. 179-196.
[8]

[9]

[10]

[11]

[12]

[13]

[14]
[15]

P. Ya. Ekel and F. H. Schuffner Neto, Algorithms of discrete optimization and their
application to problems with fuzzy coefficients, Information Sciences, vol. 176, 2006,
pp. 2846-2868.
P. Ya. Ekel, M. Menezes, and F. Schuffner Neto, Decision making in fuzzy environment and its application to power engineering problems, Nonlinear Analysis: Hybrid
Systems, vol. 1, 2007, pp. 527-536.
R. C. Berredo, P. Ya. Ekel, E. A. Galperin, and A. S. Sant'anna, Fuzzy preference modeling and its management applications, in Proceedings of the International Conference
on Industrial Logistics, Montevideo, 2005, pp. 41-50.
P. Bernardes, P. Ekel, J. Kotlarewski, and R. Parreiras, Fuzzy set based multicriteria
Decision making and its applications, in Progress on Nonlinear Analysis, Nova Science
Publisher, Hauppauge, 2008, pp. 247-272.
R. C. Botter and P. Ya. Ekel, Fuzzy preference relations and their naval engineering
applications, in Proceedings of the XIX Congress of the Panamerican Institute of Naval
Engineering, Guayaquil, 2005, Paper 7-8.
E. Herrera-Viedma, F. Herrera, F. Chiclana, A consensus model for multiperson decision making with different preference structures, IEEE Transactions on Systems, Man
and Cybernetics Part A: Systems and Humans, vol. 32, 2002, pp. 394-402.
H. M. Hsu and C.T. Chen, Aggregation of fuzzy opinions under group decision making,
Fuzzy Sets and Systems, vol. 79, 1996, pp. 279-285.
C. Lu, J. Lan, and Z. Wang, Aggregation of fuzzy opinions under group decisionmaking based on similarity and distance, Journal of Systems Science and Complexity,
vol. 19, 2006, pp. 63-71.

A New Consensus Scheme for Multicriteria Group Decision Making

85

[16] Z. Xu, Group decision making based on multiple types of linguistic preference relations, Information Sciences, vol. 178, 2008, pp. 452-467.
[17] H. S. Lee, Optimal consensus of fuzzy opinions under group decision making environment, Fuzzy Sets and Systems, vol. 132, 2002, pp. 303315.
[18] D. Ben-Arieh and Z. Chen, Linguistic-labels aggregation and consensus measure for
autocratic decision making using group recommendations, IEEE Transactions on Systems, Man and Cybernetics - Part A: Systems and Humans, vol. 36, 2006, pp. 558-568.
[19] J. Kacprzyk and M. Fedrizzi, A soft measure of consensus in the setting of partial
(fuzzy) preferences, European Journal of Operational Research, vol. 34, 1988, pp. 316325.
[20] J. Fodor, and M. Roubens, Fuzzy Preference Modelling and Multicriteria Decision
Support, Kluwer Publishers, Boston, 1994.
[21] S. A. Orlovski, Decision making with a fuzzy preference relation, Fuzzy Sets and Systems, vol. 1, 1978, pp. 155-167.

In: Mathematics and Mathematical Logic: New Research

ISBN 978-1-60692-862-2
Editors: Peter Milosav and Irene Ercegovaca
2010 Nova Science Publishers, Inc.

Chapter 4

THE MATHEMATICAL BASIS OF PERIODICITY IN

ATOMIC AND MOLECULAR SPECTROSCOPY

K. Balasubramanian
Center for Image Processing and Integrated Computing,
University of California Davis, Livermore, CA, USA
Chemistry and Applied Material Science Directorate, Lawrence Livermore National
Laboratory, University of California, Livermore, CA, USA
Glenn T. Seaborg Center, Lawrence Berkeley Laboratory, University of California,
Berkeley, CA, USA

INTRODUCTION
This chapter applies combinatorial and group-theoretical relationships to the study of periodicity in atomic and molecular spectroscopy. The relationship between combinatorics and
both atomic and molecular energy levels must be intimate since the energy levels arise from
the combinatorics of the electronic or nuclear spin configurations or the rotational or vibrational energy levels of molecules. Over the years we have done considerable work on the use
of combinatorial and group-theoretical methods for molecular spectroscopy [115]. The role
of group theory [140] is evident since the classification of electronic and molecular levels
has to be made according to the irreducible representations of the molecular symmetry group
of the molecule under consideration. Combinatorics plays a vital role in the enumeration of
electronic, nuclear, rotational and vibrational energy levels and wave functions. As can be
seen from other chapters in this book, the whole Periodic Table of the elements has a mathematical group-theoretical basis since the electronic shells have their origin in group theory.
Indeed, this concept can even be generalized to other particles beyond electrons such as bosons or other fermions that exhibit more spin configurations than just the bi-spin orientations
of electrons.

A version of this chapter was also published in The Mathematics of the Periodic Table, edited by D.H. Rouvray
and R.B. King published by Nova Science Publishers, Inc. It was submitted for appropriate modifications in an
effort to encourage wider dissemination of research.

88

K. Balasubramanian

It has been shown that Einsteins special theory of relativity is quite important for classifying the energy levels of very heavy atoms and molecules that contain very heavy atoms
[4148]. This is because to keep balance with the increased electrostatic attraction in heavier
nuclei having a large number of protons, the core electrons of such heavy atoms must move
with considerably faster average speeds. We have shown, for example, that the averaged
speed of the 1s electron of heavier atoms such as gold is about 60% of the speed of light.
Consequently, ordinary quantum mechanics does not hold, and one needs to invoke relativistic quantum mechanics to deal with such heavy atoms and with molecules containing very
heavy atoms. We have defined relativistic effects as the difference in the observable properties of electrons as a consequence of using the correct speed of light compared to the classical
infinite speed.
Mathematically, the introduction of relativity results in a double group symmetry owing
to the spin-orbit coupling term [41], which is a relativistic term in the Hamiltonian. This is a
natural symmetry consequence of the LS spin-orbit operator, which changes sign upon rotation by 360. Thus, the periodicity of the identity operation, which is normally envisaged as a
rotation through 360, is no longer the identity operation of the group. This is illustrated in
Figure 1 with a Mbius strip, which exemplifies the double group symmetry. As one completes a 360 rotation along the Mbius surface there is a sign change since one goes from the
inside of the surface to the outside. This requires the introduction of a new operation R in the
normal point group of a molecule that corresponds to the rotation by 360 which is not equal
to E, the identity operation. Hence we have to make use of the double group and doublevalued representations in both atomic and molecular spectroscopy.

Figure 1. A Mbius strip exemplifying the double group relativistic periodicity. The introduction of spin-orbit
coupling into the relativistic Hamiltonian changes the periodicity of the normal point group symmetry into a
double group symmetry, as rotation through 360 is not the identity operation. Note that the Mbius strip
changes sign in this operation. Generalization of this to other complex phases results in Berrys phase, where
rotation through 360 may yield exp(2i/n), thus resulting in other kinds of periodicity.

The Mathematical Basis of Periodicity in Atomic and Molecular Spectroscopy

89

The double group consists of twice the number of operations as the normal point group
but is not a simple direct product of the normal point group and another group. This is a consequence of the fact that only some of the conjugacy classes of the normal point group generate new conjugacy classes upon multiplication by the operation R. The other conjugacy
classes, which are called two-sided operations, such as the C2 rotations, double in order instead of generating new classes. This is because such operations when multiplied by R become equivalent to the operation, and thus the new operations belong to the same conjugacy
class as the corresponding old operations. This feature complicates double group theory and
the resulting periodicity of the double group. A characteristic feature is the generation of
even-dimensional double-valued representations that characterize half-integral quantum numbers. We shall discuss this in one of the ensuing sections.
In this chapter we shall consider the mathematical basis of atomic periodicity and spectroscopy with the use of group theory and combinatorics. We shall also consider the combinatorics of unitary groups and Young diagrams and their connections to the electronic spin functions. We shall also discuss molecular periodicity by considering the combinatorial basis of
molecular electronic states. We describe the double groups and the periodicity arising from
the classification of states in the double group. We expound on the combinatorics and periodicity pertinent to the rotational levels, nuclear spin functions, and rovibronic levels of molecules and give some examples.

COMBINATORIAL PERIODICITY IN MOLECULAR

ELECTRONIC AND ATOMIC SPECTROSCOPY
As might be expected, the classification of atomic states and thus the Periodic Table of
the elements are based on combinatorial and group-theoretical considerations. An interesting
related combinatorial problem has to do with the graphical unitary group approach to manyelectron configuration and correlation problems [4, 49]. The associated fermionic algebras
involving Young diagrams and the symmetric permutation group approach have been discussed previously [4, 49]. It would also be interesting to consider cases of similar enumerations for other bosons or even fermions that are more than spin 12 particles. Such cases, while
possibly not applicable to electronic systems, are applicable to nuclear spin species, and
would have considerable group-theoretical value.
There are several applications of group theory to atomic states. An early application by
Curl and Kilpatrick [16] showed that the Schur functions of the symmetric groups Sn can be
used as generating functions for atomic term symbols. The method involves replacement of
cycle index polynomial terms by the various mA and ms symbol powers for generating functions of the atomic states that transform according to the irreducible representations of the Sn
group. From this the authors were able to establish a periodic connection to the combinatorial
enumeration of atomic term symbols even for complicated cases such as those for f7 shells.
Balasubramanian [4] has established the connection between the graphical unitary group
approach for electronic configurations and the Schur function algebra of the Sn groups, which
play an important role in the Periodic Table in terms of the classification of various spin multiplets and the term symbols of electronic states. We shall consider this in some detail before
enumerating the atomic states. The electronic states arising from the many-electron configura-

90

K. Balasubramanian

tions have certain periodicities and patterns as enumerated by the Schur functions of the
symmetric groups Sn. The representation theory of the symmetric group is well known
[17, 20, 50], and we will not repeat it here.
The irreducible representations of Sn may be characterized by Young diagrams for the
various partitions of the integer n, denoted by [n]. The states of many particles (including
bosons and fermions) that possess multiple spin orientations can be represented by generalized Young Tableaus (GYTs). For example, Figure 2 shows all of the possible GYTs for the
partitions of six occupied by six particles that have three spin orientations (for example, a
spin-1 particle such as the bosonic deuterium nucleus) with the possibility that two have the
first kind of spin orientation, two have second kind and the last two particles have the third
kind. We have denoted this [122232] shape as shown in Figure 2.
1

2 3

2 3

1
3

2 2

1 1

2
3

2 3

2 2

1 1

2 3

1 1

1
3

Figure 2. Generalized Young Tableaus (GYTs) for the partition of six for a spin 1 boson (e.g., deuterium)
corresponding to the spin distribution of two particles with the first spin orientation, two with the second
orientation, and two with the third or [122232] shape.

As can be seen from Figure 2, the GYTs have numbers in any column in strictly ascending order while the numbers in any row must be in non-decreasing order. These tableaus represent the nuclear spin functions that transform according to the particular irreducible representation that the diagram represents. It is interesting to note that for a spin-1 particle such
GYTs can have at the most three rows and, likewise for electrons, which are spin 12 particles,
the GYTs can have at the most two rows. In general for a spin-j particle there can be at most
only 2j + 1 rows in the GYTs.

The Mathematical Basis of Periodicity in Atomic and Molecular Spectroscopy

91

The enumeration of the GYTs for the various shapes of the spin distributions is a fundamental problem that is common to electronic and nuclear structures. In the context of manyelectron spin functions, the graphical unitary group approach requires the enumeration of
Gelfand states which are the GYTs containing two rows. The results also have some interesting periodicity trends in the mathematical sense. These GYTs and the associated spin multiplets and spin states can be enumerated by polynomials called the Schur functions of the
symmetric group Sn. The Schur function corresponding to a partition of n is denoted by {}
and is defined in the following way:
1
{} = n!

(g)s1b s2 b snbn
1

(1)

gG

where (g) is value of the character for g in the group G = Sn corresponding to the irreducible representation [] of the group Sn. To illustrate this, the Schur function corresponding to
the partition 4 + 1 + 1 is given by the Schur function, {6;4,1,1}, shown below:
{6;4,1,1} =

1
120

[10s16 + 30 s14s2 + 40 s13s3 90 s12s22 120 s1s1s3

30 s23 + 40 s32 + 120 s6]

(2)

The Schur function is the generator for the GYTs, and is obtained by replacing every sk in
the Schur function or S-function by ik. The coefficient of a typical term 1a1 2a1 mam in
i

the generating function thus obtained yields the number of GYTs with the shape [1a12a1mam
]. The GYT generators are so powerful that they also enumerate the atomic states when applied to electronic spin functions which are GYTs with only two rows.
As an illustration of GYT generation, let us consider the partition 2 + 1 for three particles.
Let the particle under consideration be a spin-1 boson, which has three spin orientations that
we depict symbolically as 1, 2 and 3. The S-function in this case is given as:
1

{3;2,1} = 6 [2 s13 2 s3]

(3)

The GF of the GYTs for a spin 1 particle is thus given as:

{1,2,3;2,1} = 122 + 122 + 232 + 223 + 132+ 123 + 2 123

(4)

The above generating function thus obtained from the S-function generates all of the
GYTs shown below in Figure 3 for all possible spin distributions or shapes for the partition 2
+ 1.
The enumeration technique can be applied to GYTs of any shape belonging to any particle with any spin shape and spin distribution. The method is not restricted to just spin 1 or
spin 12 particles.

92

K. Balasubramanian

2 2

1 2

Figure 3. All possible GYTs corresponding to the partition 2 + 1 as enumerated by the S-function
{1,2,3;2,1}.

We can use the above method to generate all of the possible spin states for a manyelectron system or all of the possible atomic spectral energy levels for a given open-shell
electronic configuration. First, we illustrate the method for obtaining all of the possible electronic spin states. The GYTs for electrons may contain at most two rows since there are only
two possible distinct spin orientations for an electron ( and ) and thus there cannot be more
than two rows. Accordingly, only certain partitions are allowed for an electron. This means
that the GYTs can be formed only by the integers 1 and 2. Each spin distribution or spin
shape then contains representations that are sums of the GYTs with the appropriate shape. For
example for a system of six electrons with five spins up and one spin down there are exactly
two GYTs as shown in Figure 4.

2
Figure 4. The GYTs for six electrons with five spin ups and one spin down.

Figure 5. The many-electron spin multiplets for an even number of electrons; there are exactly N cells and at
most two rows for the spin functions.

The [1a12a1] GYTs enumerate states with a total spin quantum number
Mz = (a1 a2)/2. Consequently, once the GYTs are sorted out according to their total Mz val-

The Mathematical Basis of Periodicity in Atomic and Molecular Spectroscopy

93

ues we obtain the spin multiplets for the many-electron systems. A neat set of periodic spin
multiplets are obtained for such many-electronic systems. These are shown in Figures 5 and
6, respectively, for even and odd numbers of electrons. This important result was made possible by use of the periodic S-functions..
The method of S-functions is powerful and general in that it can be applied to more than
electrons. Thus, the same method can be applied to other particles that have integral spins
such as the deuterium nuclear spin functions or to the cases with multinomial spin distributions. In such cases the diagrams become more complex with many more rows depending on
the particles. For a spin 1 particle the diagrams have three rows at most. For a spin j particle
the diagrams will have up to (2j + 1) rows yielding an array of complex spin multiplets.
Next we demonstrate the periodic power of the S-function for enumerating the possible
electronic states of an atom [16] which are well known as atomic term symbols in atomic
spectroscopy. The method is completely analogous to generating the GYTs and manyelectron spin multiplets demonstrated above. The only difference is that we obtain a generating function for the different ML projections and spin projections and the total function must
comply with the Pauli Exclusion Principle, as electrons are fermions. The method can be applied from simple cases, e. g., main group open-shells such as p2, p3, to more complex lanthanides and actinides that have fm open-shell f-electrons (Table 1). For example, consider the
most complex half-filled 5f7 shells. The possible electronic states sorted according to the total
spin and total angular momentum in compliance with Paulis Principle are given by:
2

S(2), 2P(5), 2D(7), 2F(10), 2G(10), 2H(9), 2I(9), 2J(7), 2K(4),2M(2), 2N, 2O
S(2), 4P(2), 4D(6), 4F(5), 4G(7), 4H(5), 4I(5), 4J(3), 4K(3),4M, 4N
6 6
S, P, 6D, 6F, 6G, 6H, 6I
8
S
4

Table 1. All possible atomic term symbols for all actinides and lanthanides.
Shell
f1/f13
f2/f12
f3/f11
f4/f10

f5/f9

f6/f8

f7

States
F
1 1 1 1 3 3 3
S D G I P F H
2 2
P D(2) 2F(2) 2G(2) 2H(2) 2I 2J 2K 4S 4D 4F 4G 4I
1
S(2), 1D(4), 1F(1), 1G(4), 1H(2), 1I(3), 1J, 1K(2), 1M,
3
P(3) 3D(2) 3F(4) 3G(3) 3H(4) 3I(2) 3J(2) 3K 3L
5 5 5 5 5
S D F G I
2
P(4), 2D(5), 2F(7), 2G(6), 2H(7), 2I(5), 2J(5), 2K(3),2L(2), 2M 2N
4 4
S P(2) 4D(3) 4F(4) 4G(4) 4H(3) 4I(3) 4J(2) 4K 4L
6 6 6
P F H
1
S(4), 1P. 1D(6), 1F(4), 1G(8), 1H(4), 1I(7), 1J(3), 1K(4), 1L (2), 1M(2), 1O
3
P(6) 3D(5) 3F(9) 3G(7) 3H(9) 3I(6) 3J(6) 3K(3) 3L(3) 3M 3N
5 5 5
S P D(3) 5F(2) 5G(3) 5H(2) 5I(2) 5J 5K
7
F
2
S(2), 2P(5), 2D(7), 2F(10), 2G(10), 2H(9), 2I(9), 2J(7), 2K(5) 2L(4),2M(2),
2
N, 2O
4
S(2), 4P(2), 4D(6), 4F(5), 4G(7), 4H(5), 4I(5), 4J(3), 4K(3),4M, 4N
6 6
S, P, 6D, 6F, 6G, 6H, 6I
8
S
2

94

K. Balasubramanian

Figure 6. The many-electron spin multiplets for an odd number of electrons; there are exactly N cells and at
most two rows for the spin functions.

The mathematical aspect of periodicity in atomic states is dependent on the orbital angular momentum of the electrons and spins as exemplified by the S-function generator used
above for the generation of atomic term symbols.
Yet another aspects of periodicity involves the molecular electronic states. The electronic
configurations themselves consist of two parts, namely the spin part that was generated using
the S-functions and space types that can also be generated using multinomial generators. In
certain cases, as shown by the author, the orbital degeneracy can bring out additional symmetry. A space type can be imagined as a distribution of electrons in boxes such that a permutation of electrons within a box does not generate a new space type and the boxes themselves
can be permuted if the orbitals are degenerate. Such groups are called wreath product groups.
Balasubramanian [7] used this group theory combined with combinatorial multinomial generating functions to generate electronic space types. This can be illustrated for the benzene delocalized orbital electrons. The periodic generating function for the number of space types
of an n-orbital electronic configuration is given by:
F = (1 + w + w2)n

(5)

where the coefficient of wm gives the number of space types with m electrons distributed
among these n orbitals. For the case of benzene with six electrons distributed among six
orbitals we seek the coefficient of w6 with n = 6 in the above generating function. This is
given by:
6
6
6
1 + 3 + 2 2 2 + 4 1 1 = 141

(6)

These 141 space types of benzene enumerated here are divided into equivalence classes
of space types according to the symmetry equivalence from the wreath product groups in-

The Mathematical Basis of Periodicity in Atomic and Molecular Spectroscopy

95

duced by orbital degeneracy. As is well known, the six orbitals of benzene are divided into
1 + 2 + 2 + 1 equivalence classes of orbitals. Thus, switching of the orbitals in the second and
third set leads to equivalences, and the electrons can themselves be switched in each orbital.
The result is a direct product of wreath product groups as shown below:
S2 S2[S2] S2[S2] S2

(7)

The cycle index polynomial of the totally symmetric representation of the above group
generates the equivalence classes of the space types from the well-known Plya Theorem [1,
5260]. Consequently, the cycle index and the generating functions for the case of benzene
are as follows.
P=

F=

{ (s

1 4
8 1

+ 2 s12s2 + 3 s22 + 2 s4)

(8)

1 2
{2 (1 + w + w2)2}
28

(9)

The coefficient of w6 in the above generating function can be seen to be 58, which suggests that for benzene 141 space types are divided into 58 equivalence classes. Table 2 gives
the number of equivalence classes of the space types for the various atoms that exhibit
equivalence among the p orbitals.
Table 2. Equivalence classes of the space types for the electronic
configurations of atoms that have degenerate p orbitals.
System
He
Li
Be
B
C
N
O
F

Total No Space Types

45
156
414
882
1554
23-4
2907
3139

Equivalence Classes
17
42
86
148
223
295
349
368

In summary, we have shown that the electronic configurations of molecules and atoms
can be simplified using the mathematical periodicity of the spin functions and space types.
The former case was accomplished using the S-functions of the symmetric permutation
groups Sn while the latter case was simplified using the wreath product configuration symmetry groups.

96

K. Balasubramanian

COMBINATORIAL PERIODICITY IN MOLECULAR

AND NMR SPECTROSCOPIES
The concept of mathematical periodicity as described by the orbit structure of a permutation finds important applications in molecular and nuclear spin spectrsocopies. The orbit
structure of a permutation comprises several cycles such that in each cycle a set of nuclei is
visited followed by a return to the starting point. The cyclic structure of the permutation
(12345)(678)(9,10) is illustrated in Figure 7. The orbit structure in Figure 7 determines the
nuclear spin statistical weights of the rotational levels of molecules.
Thus, from the periodic orbit structure in Figure 7, one determines a polynomial s5s3s2,
because we have one orbit of length 5, one orbit of length 3 and one orbit of length 2. The
periodicity and the length of the period associated with each such orbit then determine a generating function for the nuclear spin statistical weights of the energy levels. This concept can
also be used in NMR and ESR spectrsocopies where the periodicity and the length of the orbits determine the NMR spin energy levels and thus the NMR spectra associated with the
molecules. Although in ordinary NMR only Zeeman-allowed transitions are observed and
thus only those transitions with changes of a single spin, multiple quantum NMR offers a
powerful tool to probe into transitions involving multiple spin quantum numbers. Thus, all
NMR interaction energy levels can be probed.

Figure 7. Periodic orbit structure for the permutation (12345)(678)(9,10).

We shall start with an application of permutational periodic structure in molecular spectroscopy. Indeed, the rotational energy levels of a molecule themselves have periodicities
based on their point groups. We illustrate this with an icosahedral cluster, namely N20 [37]
and C60 [2933] systems. Consider the highly energetic regular dodecahedral N20 cluster [37],
which exhibits icosahedral symmetry analogous to that in the fullerene C20. Since 14N is a
spin 1 particle it exhibits an interesting generating function and nuclear spin species distribu-

The Mathematical Basis of Periodicity in Atomic and Molecular Spectroscopy

97

tion. The generalized character cycle indices for all of the irreducible representations for the
N20 cluster with Ih symmetry are shown in Table 3. These were constructed using the orbit
structures of permutations as demonstrated in Figure 7.
The cycle indices for the various irreducible representations were obtained by multiplying the periodic orbit structures of each permutation by the corresponding character values.
Note that the resulting polynomials are the same for the T1g and T2g representations and likewise the T1u and T2u representations since the orbit structures multiplied by their character
values become identical owing to accidental degeneracy. We have used our generalization [1,
56, 56] of Plyas Theorem for all characters to seek generating functions for the nuclear
spin species of 14N. Note that since the 14N nucleus is a spin 1 particle, we replace every xk in
the cycle index in Table 3 by k + k + k where the symbols , and stand for 1, 0 and 1
spin projections of the spin 1 14N nucleus.
The resulting generating functions for the nuclear spin functions are shown in Table 4.
The generating functions shown in Table 4 have two parts, one consisting of coefficients and
the other of the trinomial ijk. We do not show k since k = 20 i + j) and it can thus be deduced from the values of i and j. To illustrate how the generating functions in Table 4 are obtained, let us consider the T1g or the T2g representation. From Table 3 we obtain the GCCI for
this representation and we make the substitution given by
GFx = PGx (xk k + k + k)

(10)

The above substitution yields the following expression:

1

GFT1g = 120 [3( + + )20 + 12(5 + 5 + 5)4 12(2 + 2 + 2)10 + 12(10 + 10

(11)

+ 10)2 15( + + )4(2 + 2 + 2)8]

When Equation (8) is simplified into a trinomial it has several terms with coefficients for
each term. Table 4 shows the coefficients and powers of and for the term. The power of
is simply 20 (i +j) and is thus not shown. The actual computations of the generating functions for N20 (and for C60 discussed subsequently) were carried out using computer code in
quadruple precision developed by Balasubramanian [8, 9]. It is important to employ a quadruple precision arithmetic especially for C60, as the coefficients grow astronomically and thus
any lower precision results in errors.
An interesting consequence of the periodicity is that the g and u representations differ in
some of their coefficients so significantly that one can say that there is inversion contrast in
combinatorics. For example, the coefficient of the term 965 in Table 4 for the Ag representation is 647706 while the corresponding coefficient for the Au representation is 645606.
Moreover, the first non-zero coefficient for the Au representation is for the (18,2,0) partition,
which means that at least 18 colors of one kind and two colors of another kind are needed to
induce chirality in the binomial distribution. A purely trinomial term has two chiral colorings
for the lowest order term, i. e., the (18,1,1) term in Table 4 has a coefficient of two for Au.

98

K. Balasubramanian
Table 3. The GCCIs for the dodecahedral N20 cluster.
N20
Order
Ag
Au
T1g=T2g
T1u=T2u
Gg
Gu
Hg
Hu

120
1
1
1
3
3
4
4
5
5

54
24
1
1
1
2
1
2
1
1
0
0

1236
20
1
1
0
0
1
1
1
1

210
15
1
1
1
1
0
0
1
1

210
1
1
1
3
3
4
4
5
5

Figure 8. Nuclear frequency spin spectrum for the Ag representation of N20.

102
24
1
1
1
2
1
2
1
1
0
0

263
20
1
1
0
0
1
1
1
1

1428
15
1
1
1
1
0
0
1
1

The Mathematical Basis of Periodicity in Atomic and Molecular Spectroscopy

99

The coefficients thus enumerated in Table 4 can be sorted according to their total MF values where the term l has the projection 1, m has the projection 0, and v has the projection +1.
Thus the term ijk in Table 4 represents a total nuclear spin quantum number MF of (i + k).
When these coefficients are sorted according to their total MF values, they separate into nuclear spin multiplets with MF values ranging typically from I, I+1, I+2,.0,.I2, I1, I.
Such a multiplet would represent a nuclear spin multiplet with a multiplicity of 2I + 1. In this
way for each irreducible representation the nuclear spin multiplets are separated according to
their multiplicities and the results are shown in Table 5 for N20.
Table 4. Generating functions for the dodecahedral N20 cluster.

100

K. Balasubramanian
Table 4. (Continued)

As can be seen from Table 5, the frequencies of the spin multiplets corresponding to the g
and u representations differ even for the singlet spin states. For example, the 1Ag state has a
frequency of 113035 while the 1Au state has a frequency of 112444. There is a similar difference in the triplet state and most of the spin multiplets. This means that the parity can be contrasted even in low spin nuclear states. The corresponding rovibronic levels will also be populated with appreciable differences in the populations. From the nuclear spin multiplets we can
also obtain the total nuclear spin statistical weights by the use of the Pauli Principle. Since
14
N nuclei are bosons, the overall wavefunction, which is a product of the rovibronic wavefunction and nuclear spin function, must be symmetric or must transform as the Ag irreducible
representation. The frequency of each representation is obtained by adding the product of
2S + 1 and the frequency. The results are shown as a footnote in Table 5. On the basis of this,
the frequencies shown in this footnote are themselves the nuclear spin statistical weights for
N20 (see Figure 8).

The Mathematical Basis of Periodicity in Atomic and Molecular Spectroscopy

Table 5. Nuclear spin species for the N20 cluster

101

102

K. Balasubramanian

Other irreducible representations have similar spectra comparable to that in Figure 8, except that the intensities of the peaks vary. The Ag representation is particularly important as it
gives the number of lines in multiple quantum NMR spectra. The frequencies of other irreducible representations determine the intensities of the lines in the spectra. The multiple quantum NMR spectra usually contain structural information for the (n 2) quantum as this value
exhibits dipolar couplings that contrast the structure. For the present case n 2 corresponds to
the spin multiplet 2S + 1 = 37. These multiplets have the frequencies 5, 2, 2, 7, 11, 1, 5, 5, 6,
for the Ag, T1g, T2g, Gg, Hg, Au, T1u, T2u, Gu, and Hu representations, respectively. Thus, the
dodecahedral N20 cluster exhibits interesting mathematical periodicity in spectroscopic terms.
Next we consider the C60 cluster [2933] as another example that demonstrates mathematical periodicity and its applications. The GCCIs of C60 are constructed analogously to
those of N20 discussed above. The fact that C60 has 60 vertices would of course divide the
permutation of 60 vertices into various periodic orbits. The nuclear spin species thus obtained
using the GCCIs are shown in Table 6. As seen from Table 6, the frequencies grow astronomically as expected. This is because of the combinatorial explosion of the coefficients in
the generating functions even though these functions are binomials. The binomial expansion
is due to the fact that 13C60 is comprised of 13C nuclei, which exhibit only two spin orientations, as they are spin 12 particles. The same is true of C60H60, as protons are spin 12 particles
and 12C has no nuclear spin.
Again a major contrast is that the g and u representations have different frequencies due
to the difference in the periodicity of the permutation multiplied by the character value for
these representations. This feature manifests itself as contrasting frequencies for the g and u
irreducible representations. We note that earlier work had an error in the spin statistical
weights of C60 [31] primarily owing to the arithmetical precision but this was subsequently
corrected [30, 32]. The relative differences between the g and u parities are especially significant for high-spin nuclear multiplets. For example, for the 2S + 1 = 57 spin multiplet of 13C6,
the frequencies of the Ag, T1g, T2g, Gg, Hg, Au, T1u, T2u, Gu, and Hu representations are
22, 36, 36, 58, 80, 14, 42, 42, 56, and 70, respectively. Similarly for 2S + 1 = 55 the frequency of the Ag representation is 280 while it is 260 for Au. Consequently, the contrast in the
g and u spin populations can be seen experimentally if high-spin nuclear states can be excited.
Table 6. Nuclear spin multiplets for 13C60 or C60H60.
Frequency of the irreducible representation Ag: 9607679885269312
Spin multiplets and their frequencies for Ag:
2S+1
Frequency 2S+1
1
31791575566072 3
7
150988619146706 9
13
105558807981090 15
19
31605175642230 21
25
4481735502630 27
31
298734989924 33
37
8805633300 39
43
101874363 45
49
372752 51
55
280 57
61
1

Frequency 2S+1
Frequency
89413728633564 5
13095954950748
149756091280506 11
13219208028055
76925432220000 17
5141513084676
17892025439775 23
933143835273
1980110898945 29
80345370985
101492436960 35
3139590568
2227563126 41
50512570
18110340 47
280174
41528 53
388
22 59

The Mathematical Basis of Periodicity in Atomic and Molecular Spectroscopy

Table 6. (Continued)
Frequency of the irreducible representation T1g: 28823036970926496
Spin multiplets and their frequencies for T1g :
2S+1
Frequency 2S+1
Frequency 2S+1
Frequency
1
95374646372040 3
268241251090167 5
39287856402727
7
452965902231668 9
449268197030424 11
39657626655407
13
316676363633175 15
230776308338940 17
15424535176554
19
94815530686980 21
53676052490265 23
2799431557098
25
13445194549380 27
5940332333550 29
241035603798
31
896204629630 33
304475471640 35
9418755979
37
26416344630 39
6682635360 41
151524170
43
305608974 45
54304371 47
840285
49
1114158 51
124257 53
1123
55
804 57
36 59
61
0

Frequency of the irreducible representation Gg: 38430716856193728

Spin multiplets and their frequencies for Gg:
2S+1
1
7
13
19
25
31
37
43
49
55
61

Frequency 2S+1
127166221937640 3
603954521378374 9
422235171614265 15
126420706329465 21
17926930052010 27
1194939619444 33
35221977930 39
407483337 45
1486916 51
1084 57
0

Frequency 2S+1
Frequency
357654979723731 5
52383811353475
599024288311326 11
52876834683423
307701740558940 17
20566048261230
71568077929785 23
3732575392371
7920443232495 29
321380974795
405967908600 35
12558346548
8910198522 41
202036737
72414711 47
1120460
165779 53
1512
58 59

Frequency of the irreducible representation Hg 48038396740938240

Spin multiplets and their frequencies for Hg:
2S+1
1
7
13
19
25
31
37
43
49
55
61

Frequency 2S+1
158957797411208 3
754943140441100 9
527793979532265 15
158025881932935 21
22408665535200 27
1493674601616 33
44027608785 39
509357130 45
1859568 51
1354 57
0

Frequency 2S+1
Frequency
447068708357295 5
65479766312622
748780379591832 11
66096042717787
384627172778940 17
25707561349782
89460103369560 23
4665719229588
9900554131440 29
401726346555
507460345560 35
15697937361
11137761648 41
252549365
90525051 47
1400644
207307 53
1902
80 59

103

104

K. Balasubramanian

Table 6. (Continued)
Frequency of the irreducible representation Au: 9607678793631424
Spin multiplets and their frequencies for Au:
2S+1
Frequency 2S+1
1
31791571643468 3
7
150988613640506 9
13
105558798039270 15
19
31605170531130 21
25
4481732871390 27
31
298734348764 33
37
8805495420 39
43
101861196 45
49
371694 51
55
260 57
61
0

Frequency 2S+1
Frequency
89413727296344 5
13095954114986
149756080818726 11
13219207292373
76925425313100 17
5141512318638
17892020535870 23
933143526111
1980109351620 29
80345252581
101491992360 35
3139568730
2227502850 41
50509098
18103410 47
279955
41266 53
377
14 59

Frequency of the irreducible representation T1u: 28823037990981216

Spin multiplets and their frequencies for T1u:
2S+1
Frequency 2S+1
Frequency 2S+1
Frequency
1
95374639953380 3
268241262122232 5
39287856269005
7
452965915721858 9
449268199508214 11
39657627967718
13
316676367808710 15
230776318887660 17
15424535578410
19
94815537801090 21
53676055391160 23
2799431961645
25
13445196226770 27
5940334271070 29
241035683183
31
896205406510 33
304475780520 35
9418781778
37
26416442910 39
6682705140 41
151526692
43
305623968 45
54309474 47
840531
49
1114942 51
124548 53
1132
55
826 57
42 59
61
0

Frequency of the irreducible representation Gu: 38430716784610624

Spin multiplets and their frequencies for Gu:
2S+1
1
7
13
19
25
31
37
43
49
55
61

Frequency 2S+1
127166211596396 3
603954529362364 9
422235165847980 15
126420708332460 21
17926929098160 27
1194939755164 33
35221938330 39
407485164 45
1486642 51
1086 57
0

Frequency 2S+1
Frequency
357654989418576 5
52383810383991
599024280327336 11
52876835260051
307701744200760 17
20566047897048
71568075926790 23
3732575487756
7920443622690 29
321380935775
405967772880 35
12558350508
8910208020 41
202035787
72412884 47
1120487
165808 53
1509
56 59

The Mathematical Basis of Periodicity in Atomic and Molecular Spectroscopy

105

Table 6. (Continued)
Frequency of the irreducible representation Hu: 48038395577718272
Spin multiplets and their frequencies for Hu
2S+1
1
7
13
19
25
31
37
43
49
55
61

Frequency 2S+1
158957783147612 3
754943142918890 9
527793963824370 15
158025878824830 21
22408661950230 27
1493674096176 33
44027431350 39
509345790 45
1858246 51
1336 57

Frequency 2S+1
Frequency
447068716714920 5
65479764507375
748780361146062 11
66096042558712
384627169513860 17
25707560219562
89460096462660 23
4665719015799
9900552974310 29
401726189132
507459765240 35
15697919478
11137710870 41
252544942
90516294 47
1400452
207074 53
1887
70 59
0

Table 7 shows the correlation of the rotational levels for C60 from J = 0 to 30 with the
corresponding weights only in the rotational subgroup I. Note that for purposes of comparing
with experimental results one must use the nuclear spin frequencies given in Table 6, but the
statistical weights in Table 7 in factored form yield the orders of magnitude. All levels in Table 7 are of g symmetry since the J states can correlate into g levels.
(1) The irreducible representations for J > 31 are given by q[A + 3T1 + 3T2 +4G + 5H] +
(r), where q is the quotient obtained by dividing J by 30 and r is the remainder. (r) is the
set of irreducible representations spanned by J = r listed in this Table (see text for further discussion). Note that since nuclear spin statistical weights are the same for g and u symmetries,
we do not show g or u.
(2) f = 19 215 358 678 900 736 for C60H60;
f = 706 519 304 586 988 199 183 738 259 for C60D60.
Each correlation in Table 7 was obtained using the mathematical method of subduction.
As can be seen from Table 7, we have a very interesting periodicity among rotational levels.
The correlations for the rotational levels with J > 31 have a periodic relation to the levels with
J < 30. This is another mathematical manifestation of periodicity. The relations for all J > 30
are as follows:
D(J)Ih = q(D(30)Ih A) + (D(r) Ih), q = [J/30], r = J 30[J/30]

(12)

where the function within square brackets is the greatest integer contained in the brackets and
thus q and r are quotients and remainders obtained by dividing J by 30. The term D(30) stands
for the subduced representations for J = 30 that are displayed in Table 7. To illustrate this, the
J = 195 rotational level contains the following representations:
D(195)Ih = 6(Ag + 3T1g + 3T2g + 4Gg + 5Hg) + (Ag + 2T1g + 2T2g + 2Gg + 2Hg) 7Ag +
20T1g + 20T2g + 26Gg + 32Hg

(13)

106

K. Balasubramanian
Table 7. Correlations of the rotational levels of C60: the nuclear spin statistical
weights J = 0 to 30

The above concept of the periodicity of the rotational levels of C60 is illustrated in Figure
9.
It is worthy of note that the nuclear spin statistical weights of the rotational levels vary
approximately as (2J + 1) due to large nuclear spin statistical weights.

The Mathematical Basis of Periodicity in Atomic and Molecular Spectroscopy

107

J=0
J=1

J=30
*
*
* .
*
*

J=2

.J=3
Figure 9. Periodicity of the rotational levels of buckminsterfullerene, C60.

PERIODICITY OF DOUBLE GROUPS AND ELECTRONIC STATES

The concept of the double group [17, 2427, 42, 51] is required when the normal periodicity
resulting from rotation through 360 breaks down, as demonstrated for the Mbius strip. This
happens when half-integral states are considered. For example, the rovibronic states of openshell systems with an odd number of open-shell electrons exhibit half-integral spin states due
to an odd number of open-shell electrons and thus we need a new concept of periodicity. This
is also the case when spin-orbit coupling is introduced into the Hamiltonian. This difficulty
was circumvented by Bethe through the concept of the double group. He introduced a new
operation called R that changes the sign for the rotation through 360 for half-integral states
and yet retains the same symmetry for the integral states, as shown above for the C60 integral
rotational levels. Since the periodicity and the group structure are quite different for the double group, we provide a few examples of double group character tables and correlation tables.
Most of the character tables appear in books such as those of Hamermesh [17] or
Altmann and Herzig [24] for the double groups. Balasubramanian [51] developed the character table for the icosahedral double group denoted by Ih2 that is shown in Table 8. Note that
the operation R introduces a few new conjugacy classes for the Ih2 double group while other
conjugacy classes just double in their orders. This is a consequence of the fact that certain
operations are called two-valued operations and these operations when multiplied by R become equivalent, and thus belong to the same conjugacy class. However, other operations,
such as C5 and RC5, become inequivalent, and thus belong to different conjugacy classes.
The new irreducible representations in the double group are called two-valued representations
and they are always even dimensional and correspond to half-integral representations. The
number of such representations equals the number of new conjugacy classes, as demonstrated
in Table 8. These are called E1g(1/2), Gg(3/2), Ig(5/2), E2g(7/2), with the corresponding u
representations.
The correlation table for the half-integral states of the Ih2 double group is shown in Table
9. Note that the corresponding table for the integral representations has already been discussed for C60 (Table 7). As can be seen from Table 9, the half-integral spin or rovibronic

108

K. Balasubramanian

states all correlate only into double-valued representations, which are all even dimensional.
As a result, the representation corresponding to 12 is a degenerate two-dimensional irreducible representation. The quartet state with s = 32 is also four-fold degenerate and s = 52 is
likewise the six-fold degenerate I representation in the double group. The first case which
splits into two irreducible representations is the s = 72 case. The periodicity is reduced in the
double-valued representation to half as s = 312 is related to s = 12 by periodicity. All higher
s values are obtained using a periodic relation as shown in Table 9.
Table 8. Character table for the Ih2 double group.

The Mathematical Basis of Periodicity in Atomic and Molecular Spectroscopy

Table 8. (Continued)

109

110

K. Balasubramanian

Table 9. Periodic correlation table for the half-integral states of the Ih2 double group.
s

1
3
5
7
9

2
2
2
2

2
11
2
13
2
15
2
17
2
19
2
21
2
23
2
25
2
27
2
29
2
31
2

Irreducible Representationsa
E1g(1 2)
Gg(3 2)
Ig(5 2)
E2g(7 2) + Ig(5 2)
Gg(3 2) + Ig(5 2)
E1g(1 2) + Gg(3 2) + Ig(5 2)
E1g(1 2) + Gg(3 2) + Ig (5 2) + E2g(7 2)
Gg (3 2) + 2 Ig(5 2)
Gg(3 2) + 2 Ig(5 2) + E2g(7 2)
E1g(1 2) + Gg(3 2) + 2 Ig(5 2)+ E2g(7 2)
E1g(1 2) + 2 Gg(3 2) + 2 Ig(5 2)
E1g(1 2) + 2 Gg(3 2) + 2 Ig(5 2) + E2g(7 2)
E1g(1 2) + Gg(3 2) + 3 Ig(5 2) + E2g(7 2)
2 Gg(3 2) + 3 Ig(5 2) + E2g(7 2)
E1g(1 2) + 2 Gg(3 2) + 3 Ig(5 2) + E2g(7 2)
2 E1g(1 2) + 2 Gg(3 2) + 3 Ig(5 2) + E2g(7 2)

Ds = q{E1g(1 2) + 2 Gg(3 2) + 3 Ig(5 2) + E2g(7 2)} + Ds,

2s + 1
q = 30 , s = s 15q, if s > 31 2

We have also collected the correlation tables [42] for the octahedral double group Oh2 in
Table 10, the correlation table for the Td2 in Table 11, and the correlation table for the D6h2
in Table 12. These correlation tables all demonstrate interesting mathematical periodicity for
the rotational or rovibronic levels. The octahedral integral rotational levels exhibit a period of
12 analogous to that for the tetrahedral group. However, the half-integral spin states or
rovibronic states exhibit a period of six both in the octahedral and tetrahedral double groups.
The D6h2 double group exhibits a different periodic trend as seen from Table 12. The
periodicity of six is same for both the half-valued and integral representations. Thus, the
periodicity trends exhibited by the double groups are quite interesting. These correlation
tables are quite valuable in obtaining the rovibronic levels of molecules with both an odd and
even number of electrons. It is important to obtain the overall rovibronic correlation as
opposed to individual rotational correlations owing to the fact that the total wavefunction may
become a half-integral representation, especially for systems with an odd number of electrons.
Furthermore, for molecules containing very heavy atoms spin-orbit effects become quite
significant, and thus the coupling of the spin with orbital angular momentum splits the
electronic states into spin-orbit states. The exact manner in which these states are split by
spin-orbit coupling is given by the double group correlation tables shown here.

The Mathematical Basis of Periodicity in Atomic and Molecular Spectroscopy

111

Table 10. Periodic correlation table for the half-integral states of the Oh2 double group.
Irreducible Representations in the Oh2 Groupa
A1g
T1g
Eg + T2g
A2g + T1g + T2g
A1g + Eg + T1g + T2g
Eg + 2T1g + T2 + T2g

s
0
1
2
3
4
5
6
2

A1g + A2g + Eg + T1g + 2T2g

E1g(12)
Gg(32)
E2g(52) + Gg(32)
E1g(12) + E2g(52) + Gg(32)
E1g(12) + 2 Gg(32)
E1g(12) + E2g(52) + 2 Gg(32)
E1g(12) + E2g(52) + 2 Gg(32) + terms of s but interchange E1g(12) with E2g(52)
2n {(E1g(12) + E2g(52) + 2 Gg(32)} + terms for s

2
2
7
2
9
2
11
2
6 + s
12n + s
5

Terms for other integral s values are found using the formula:
D(12n+s) = Ds + n(A1g + A2g + 2 Eg + 3 T1g + 3 T2g), s < 12.

Table 11. Periodic correlation table for the half-integral states of the Td2 group.
s
0
1
2
3
4
5
6

Irreducible Representationsa
A1
T1
E + T2
A2 + T1 + T2
A1 + E + T1 + T2
E + 2 T1 + T2
A1 + A2 + E + T1 + 2 T2

7
8
9
10
11
12
13
14
15
1
2
3
2
5
2
7
2
9
2
11
2
6 + s
2n + s

A2 + E + 2 T1 + 2 T2
A1 + 2E + 2 T1 + 2 T2
A1 + A2 + E + 3 T1 + 2 T2
A1 + A2 + 2E + 2 T1 + 3 T2
A2 + 2E + 3 T1 + 3 T2
2 A1 + A2 + 2 E + 3 T1 + 3 T2
A1 + A2 + 2 E + 4 T1 + 3 T2
A1 + A2 + 3 E + 3 T1 + 4 T1
A1 + 2 A2 + 2 E + 4 T1 + 4 T1
E1/2
G3/2
G3/2 + E5/2
E1/2 + G3/2 + E5/2
E1/2 + 2 G3/2
E1/2 + E5/2+ 2 G3/2
E1/2 + E5/2 + 2 G3/2 + terms for s, but interchange E1/2 and E5/2
2n(E1/2 + E5/2)

Other integral spin states are correlated using the formula

D(12n+s' = Ds + n(A1 + A2 + 2 E + 3T1 + 3T2), s < 12.

112

K. Balasubramanian

The concept of periodicity can be extended to cases beyond the double groups. Such
cases would involve Berrys phase where a rotation through 360 would yield a complex
number, exp(2i/n), for an integer n > 2. The symmetry exhibited by such systems could be
quite intriguing. It is hoped that this chapter will stimulate future investigations into Berrys
phase.
Table 12. Periodic correlation table for the half-integral
states of the D6h2 double group
s
0
1
2
3
4
5
6
1

2
2
5
2
7
2
9
2
11
2
3

6n + s

Irreducible Representationsa
A1g
A2g + E1g
A1g + E1g + E2g
A2g + B1g + B2g + E1g + E2g
A1g + B1g + B2g + E1g + 2 E2g
A2g + B1g + B2g + 2 E1g + 2 E2g
2 A1g + A2g + B1g + B2g + 2 E1g + 2 E2g
E1g(12)
E1g(12) + E3g(32)
E1g(12) + E2g(52) + E3g(32)
E1g(12) + 2 E2g(52) + E3g(32)
E1g(12) + 2 E2g(52) + 2 E3g(32)
2 E1g(12) + 2 E2g(52) + 2 E3g(32)
6n{E1g(12) + E2g(52) + E3g(32)} + terms of s, s < 6

Terms for other integral s values may be found by using

D(12n+s) = Ds + n(A1g + A2g + B1g + B2g + 2 E1g + 2 E2g), s < 6

ACKNOWLEDGEMENT
This research was performed under the auspices of the US Department of Energy by the
University of California, LLNL under contract number W-7405-Eng-48 while the work at UC
Davis was supported by the National Science Foundation.

REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]

K. Balasubramanian, Chem. Rev., 85, 599 (1985).

K. Balasubramanian, J. Chem. Phys., 72, 665 (1980).
K. Balasubramanian, J. Chem. Phys., 73, 3321 (1980).
K. Balasubramanian, Theor. Chim. Acta, 59, 237 (1981).
K. Balasubramanian, J. Chem. Phys., 74, 6824 (1981).
K. Balasubramanian, J. Chem. Phys., 75, 4572 (1981).
K. Balasubramanian, Int. J. Quant. Chem., 20, 1255 (1981).
K. Balasubramanian, J. Comput. Chem., 3, 69 (1982).

The Mathematical Basis of Periodicity in Atomic and Molecular Spectroscopy

[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]

[26]
[27]

[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]

113

K. Balasubramanian, J. Comput. Chem., 3, 75 (1982).

K. Balasubramanian, H. Strauss, and K. S. Pitzer, J. Mol. Spectrosc., 93, 447 (1982).
K. Balasubramanian, J. Phys. Chem., 86, 4668 (1982).
K. Balasubramanian, J. Chem. Phys, 78, 6358 (1983).
K. Balasubramanian, J. Chem. Phys, 78, 6369 (1983).
K. Balasubramanian, Group Theory of Non-rigid Molecules and its Applications, Elsevier Publishing Co., 23, 149168 (1983).
K. Balasubramanian, Theor. Chim. Acta, 78, 31 (1990).
R. F. Curl, Jr., and J. E. Kilpatrick, Amer. J. Phys. 28, 357 (1960)
M. Hamermesh, Group Theory and its Physical Applications, Addison Wesley, Reading MA, 1964
F. A. Cotton, Chemical Applications of Group Theory, Wiley Interscience, New York,
NY, 1971
P. R. Bunker, Molecular Symmetry and Spectroscopy,Academic Press, New York,
NY, 1979
D. E. Littlewood, Theory of Group Characters and Matrix Representations of Groups, ,
Oxford, New York, NY, 1958
H. Weyl, Theory of Groups and Quantum Mechanics, Dover Publications, New York,
NY, 1950
M. Tinkham, Group Theory and Quantum Mechanics, McGraw-Hill, New York, NY
1964
B. R. Judd, Operator Techniques in Atomic Spectroscopy, Princeton University Press,
Princeton,, NJ 1998
S L Altmann and P Herzig, Point-Group Theory Tables, Clarendon Press, Oxford,
1994.
P. Pyykk and H. Toivonen, Tables of Representation and Rotation Matrices for The
Relativistic Irreducible Representations of 38 Point Groups, Acta Academiae Aboensis,
Ser B, 43, 1 (1983)
H.T. Toivonen and P. Pyykk, Int. J. Quant. Chem., 11, 697 (1977)
H.T. Toivonen and P. Pyykk, Relativistic Molecular Orbitals and Representation Matrices for the Double Groups T and Th, Department of Physical Chemistry, bo
Akademi, Finland, Report No. B79 (1977), 11 pp.
K. Balasubramanian, J. Mag. Res., 91, 45 (1991).
K. Balasubramanian, Chem. Phys. Lett., 183, 292 (1991).
K. Balasubramanian, Chem. Phys. Lett., 200, 649 (1992).
W. G. Harter and T. C. Reimer, Chem. Phys. Lett., 194, 230 (1992).
W. G. Harter and T. C. Reimer, Chem. Phys. Lett., 198, 429E (1992).
W. G. Harter and D. E. Weeks, J. Chem. Phys., 90, 4727 (1989).
K. Balasubramanian, J. Chem. Phys., 95, 8273 (1991).
K. Balasubramanian and T. R. Dyke, J. Phys. Chem., 88, 4688 (1984).
K. Balasubramanian, J. Mol. Spectroscopy 157, 254 (1993).
K. Balasubramanian, Chem. Phys. Lett., 202, 271(1993).
K. Balasubramanian, J. Phys. Chem., 97, 8736 (1993)
K. Balasubramanian, Mol. Phys., 80, 655 (1993).
K. Balasubramanian, J. Chem. Phys., in press.

114

K. Balasubramanian

[41] K. Balasubramanian, Relativistic Effects in Chemistry, Part B: Applications, Wileyinterscience, New York, NY, p. 527, 1997.
[42] K. Balasubramanian, Relativistic Effects in Chemistry, Part A: Theory and Techniques,
Wiley-Interscience, New York, NY, p. 301, 1997.
[43] K. S. Pitzer, Accts. Chem. Res., 12, 271 (1979).
[44] P. Pyykk and J. P. Desclaux, Accts. Chem. Res., 12, 276 (1979).
[45] P. Pyykk, Adv. Quant. Chem., ll, 353 (1978).
[46] K. Balasubramanian, J. Phys. Chem., 93, 6585 (1989).
[47] P. Pyykk Ed. Proceedings of the Symposium on Relativistic Effects in Quantum
Chemistry; Int. J. Quantum Chem., 25 (1984).
[48] P. Pyykk, Relativistic Theory of Atoms and Molecules, Springer Verlag: Berlin and
New York, Part I 1986 Part II 1993, Part 3 2000. For comprehensive list of references
up to 2002 see http://www.csc.fi/rtam/.
[49] J. Paldus, Theoretical Chemistry: Advances and Perspectives, H. Eyring and D. J. Henderson, Eds, Academic Press, New York, NY, 1976
[50] X. Y. Liu and Balasubramanian, J. Comput. Chem., 10, 417 (1989).
[51] K. Balasubramanian, Chem. Phys. Lett., 260, 476 (1996).
[52] G. Plya, Acta Math, 68, 145 (1937)
[53] K. Balasubramanian, Theor. Chim. Acta, 53,129 (1979)
[54] G. Plya and R.C. Read, Combinatorial Enumeration of Groups, Graphs, and Chemical
Compounds, Springer, New York, NY, 1987.
[55] J. H. Redfield, Amer. J. Math. 49, 433 (1927).
[56] K. Balasubramanian, J. Math. Chem. 14, 113 (1993).
[57] D. H. Rouvray, Chem. Soc. Rev. 3, 355 (1974).
[58] D. H. Rouvray, Endeavour, 34, 28 (1975)
[59] A.T. Balaban, Chemical Applications of Graph Theory, Academic Press, New York,
NY, 1976.
[60] R. Read, in Graph Theory and Applications, Y. Alavi et al. eds., Lecture Notes in
Mathematics, 303, 243 (1972) Springer, 1972.

In: Mathematics and Mathematical Logic: New Research

ISBN 978-1-60692-862-2
c 2010 Nova Science Publishers, Inc.
Editors: Peter Milosav and Irene Ercegovaca

Chapter 5

S ECOND - ORDER D EFINABILITY

IN A M ODEL
George Weaver
Park Science Center, Bryn Mawr College
Bryn Mawr, PA, USA

Abstract
The back and forth characterization of equivalence of interpretations for finitary
(and some infinitary) second-order languages introduced in Weaver and Penev [2005]
is applied to obtain a condition necessary and sufficient for an attribute of an interpretation to be definable in that interpretation by a second-order formula (either finitary
or infinitary). This condition is applied to obtain some reduction theorems for the
second-order theories of those infinite interpretations having pairing functions that are
definable by two simple classes of second-order formulas.

Key words and phrases: second-order logic, infinitary logic, definability

1991 Mathematics Subject Classification: 03B15, 03C85

1.

Introduction

Weaver and Penev ([2005]) extended the Frasse back and forth characterization of elementary equivalence to interpretations in the standard semantics of those second-order
languages without functional variables whose non-logical vocabulary is finite and excludes
functional constants. This characterization fails when the non-logical vocabulary is infinite.
However, there is a class of infinitary second-order languages for which the characterization
holds even when the non-logical vocabulary is infinite. Weaver and Penev presented several
applications of this characterization for both classes of languages. Another application is
presented here: a necessary and sufficient condition for definability in an interpretation.
K is a set of non-logical constants, excluding functional constants. When K is finite,
LK is the (finitary) second-order language without functional variables whose non-logical

E-mail address: gweaver@brynmawr.edu

116

George Weaver

vocabulary is K. When K is infinite, L K is the infinitary second-order language without

functional variables whose non-logical vocabulary is K. As in Weaver and Penev [2005]
(page 504 and following), the set of formulas in this infinitary language includes the (finitary) atomic formulas over K and is closed under finitary quantifications and usual truth
functional combinations, and conjunctions and disjunctions of sets of formulas of bounded
quantifier rank. While the logical vocabulary of these languages includes the identity sign,
the only equations in the languages are between individual symbols (i.e. individual constants or individual variables).
An interpretation of type K is an ordered pair A=(A, fA ), where A is a nonempty set
(the domain of A) and fA is a function defined on K in the usual way. Let B be a subset of
A and (x) be a formula whose one and only free variable is the individual variable x. (x)
defines B in A (or B is defined by (x) in A ) provided for all a in A, aB iff a satisfies (x)
in A (A |= (x)[a]). B is definable in A iff there is (x), as above, such that (x) defines B
in A.
Let m be a positive integer 2, and be an m-ary relation on A. (x1...xm ) is a formula in LK whose free variables are exactly the individual variables x 1,..., xm , where these
variables are all different, and where for all i and j between 1 and m, if i <j, then the first
free occurrence of xi in (x1...xm ) precedes the first free occurrence of xj in (x1...xm ).
(x1 ...xm ) defines in A (or is defined by (x1 ...xm ) in A) provided for all a 1 ,...,am in A,
(a1 ,...,am ) iff (a1 ,...,am ) satisfies (x1 ...xm ) in A (A |= (x1...xm )[a1 ...am ]) when for all
i, between 1 and m, ai is the value of x i . is definable in A iff there is (x1 ...xm ), as above,
such that (x1...xm ) defines in A. Notice that the condition on the order of occurrence of
the variables in (x1 ...xm ) implies that (x1 ...xm ) defines exactly one relation on A.
Since second-order languages contain both set and relational variables in addition to
first-order variables, other kinds of objects are definable by second-order formulas. Let
be a subset of the power set of A and (X) be a formula whose one and only free variable
is the set variable X. (X) defines in A (or is defined by (X) in A) provided (X) is
satisfied in A by exactly those subsets of A that are members of . And, is definable in A
iff there is (X), as above, such that (X) defines in A. In the same way, if is a subset
of the power set of the m-th direct power of the domain, then is definable in A iff there
is (X), a formula whose one and only free variable is the m-ary relational variable X, and
(X) is satisfied in A by exactly those m-ary relations on A that are in .
Let n be a non-negative integer. The n-ary domain of A is defined as follows: A is the
0-ary domain of A; the power set of A is the 1-ary domain of A; and, for n 2, the power
set of An is the n-ary domain of A. A cumulative domain of A is any n-ary domain of A.
If D is a cumulative domain of A and X is a variable in L K , then X is appropriate for D
provided either X is an individual variable and D is A or X is a set variable and D is the
power set of A or X is an n-ary relational variable and D is the n-ary domain of A. Let m
be 2, D1,...,Dm be cumulative domains of A, and be a subset of D 1 ...Dm. Finally,
let (X1...Xm ) be a formula in LK such that X 1 ,...Xm are all different and are exactly the
variables occurring free in (X1 ...Xm ); for each i, between 1 and m, Xi is appropriate for
Di ; and if i<j , then the first free occurrence of X i in (X1...Xm ) precedes the first free
occurrence of Xj . (X1 ...Xm ) defines in A, is defined by (X1 ,...Xm ) in A and is
definable in A are defined as above. An attribute of A (attribute of A) is either a subset of
cumulative domain of A or a subset of a finite direct product of cumulative domains of A.

Second-order Definability in a Model

117

Given A and B interpretations of type K, A B indicates that the interpretations are

isomorphic. If C is a subset of A, then A[C] is that subinterpretation of A generated by
C. As long as C is nonempty or K contains individual constants, A[C] is an interpretation
of type K. When K does not include individual constants and C is emtpy, A[C] B[C] is
defined even though their domains are empty.
~ 0 are defined as follows. ~a is a fi~ t and ~bS~ 0R~0 ...R
~R
~2...R
Given A, B and t2, ~aS
t
2
nite(possibly empty) sequence of members of A and ~b is a finite sequence of members of
~ is a finite(possibly empty) sequence of subsets of A and S~ 0
B of the same length as ~a. S
~j is a
is a finite sequence of subsets of B of the same length. And, for each j, 2 jt, R
~0 is a finite sequence of subsets of
finite(possibly empty) sequence of subsets of Aj and R
j
B j of the same length. The first sequence is called a sequence in A, the second, a sequence
in B. Such pairs of sequences are said to be similar.
~ t is an expansion of A to an interpretation of type K 0 where K is a subset of
~R
~2...R
A~aS
0
0
K . Let k0 be the length of ~a. K includes exactly k 0 individual constants not in K and the
i-th of these new constants denotes the i-th member of ~a in this expansion. Let k 1 be the
~ K0 includes exactly k 1 unary predicate constants not in K and the i-th member
length of S.
~ is the extension in this expansion of the i-th new unary predicate constant. For each j,
of S
~j . K0 includes exactly k j j-ary predicate constants
between 2 and t, let k j be the length of R
~j is the extension in this expansion of the i-th new j-ary
not in K and the i-th member of R
0
predicate constant. Finally, K contains no predicate constant of degree strictly greater than
t that is not in K.
Let m0 ,...,mt be a sequence of non-negative integers.
~0
~ t vm ...mt B~bS~ 0R~0 ...R
~ R~2 ...R
A~aS
t
0
2
is definded as follows.
1. If for all j, 0jt, mj =0, then
~ 0 iff A~aS
~0 ...R
~ t vm ...mt B~bS~0 R
~R
~2...R
~ t[] B~bS~ 0R~0 ...R
~ t[].
~R
~2...R
A~aS
t
0
2
2
2. If there is j such that m j 6=0, then
~ 0 iff
~0 ...R
~ t vm ...mt B~bS~0 R
~R
~2...R
A~aS
t
0
2
(a) if m0 6=0, then
i. for all aA there is bB such that
~0,
~0 ...R
~ t vm 1...mt B~bbS~ 0R
~R
~2...R
A~aaS
t
0
2
ii. for all b B there is aA such that
~0;
~0 ...R
~ t vm 1...mt B~bbS~ 0R
~R
~2...R
A~aaS
t
0
2
(b) if m1 6=0, then

(1)

118

George Weaver
i. for all SA there is S0B such that
~ 0,
~ t vm m 1...mt B~bS~ 0S 0R~0 ...R
~ R
~2...R
A~aSS
0 1
2
t
ii. for all S0 B there is SA such that
~ 0;
~ t vm m 1...m B~bS~ 0S 0R~0 ...R
~ R
~2...R
A~aSS
t
t
0 1
2
(c) for all j, 2jt, if mj 6=0, then
i. for all RAj there is R0 Bj such that
~ 0 R0 ...R
~ 0,
~ t vm ...m 1...mt B~bS~0 ...R
~ R
~j R...R
A~aS...
t
0
j
j
ii. for all R0 Bj there is RAj such that
~ 0 R0 ...R
~ 0.
~ t vm ...m 1...m B~bS~0 ...R
~ R
~j R...R
A~aS...
t
t
0
j
j

Let D, D1,...,Dn , be cumulative domains of A; , a subset of D; , a subset of

D1 ...Dn ; t, an integer 2 and, m0 ...mt , a sequence of non-negative integers. is closed
under vm0 ...mt in A iff for all a and b in D, if a and Aavm0 ...mt Ab, then b . is
closed under vm0 ...mt in A iff for all a and b D1 ...Dt , if a and Aavm0 ...mt Ab,
then b . The formula is of (bound) quantifier rank (m 0 ,...,mt ) iff no relational variable
of degree strictly greater than t occurs bound in , exactly m0 individual variables occur
bound in , exactly m1 set variables occur bound in , and for all j, between 2 and t, exactly
mj j-ary relational variables occur bound in . If K contains no individual constants and
m0 =0, there are no sentences of quantifier rank (m 0 ,...,mt ). Hence, in the following it is
assumed that either K contains an individual constant or m 0 6=0. It follows from Lemma 2.2
of Weaver and Penev [2005] that any set or relation definable in A is closed under vm0 ...mt
in A where (m0 ,...,mt ) is the quantifier rank of that formula that defines that set or relation
in A. It is shown below that the sets and relations definable in A are exactly those closed
under vm0 ...mt in A, for some t2 and m0 ,...,mt non-negative integers.

2.

A Normal Form Theorem for Formulas

It was observed in Weaver and Penev [2005] (page 502) that every sentence has a logically
equivalent distributive normal form. Here similar reasoning is used to show that every
formula that is satisfied in A is equivalent in A to a formula in a similar normal form. We
proceed by recalling Lemma 2.2 of Weaver and Penev [2005] (page 502).
Lemma 2.1. Assume that A is an interpretation of type K, t2, m0 ,...,mt are non~ t is a sequence in A. Then, there is
~ R~2 ...R
negative integers and that ~aS
~R
~2...R
~ t]
[A, m0...mt, ~aS
a formula in LK of quantifier rank (m0 ,...,mt ) such that for all B of type K and all
~ 0 , a sequence in B, if the sequences are similar, then the following are equiv~0 ...R
~bS~ 0R
t
2
alent:

Second-order Definability in a Model

119

~0,
~0 ...R
~R
~2...R
~ t vm ...mt B~bS~ 0R
1. A~aS
t
0
2
~ 0 satisfies [A, m0...mt, ~aS
~R
~2...R
~ t] in B.
2. ~bS~ 0R~02 ...R
t
~R
~2...R
~ t], when
There are only finitely many formulas of the form [A, m0...mt, ~aS
K is finite. Even if K is infinite, they can be constructed in such a way that
~ 0 ] are identical iff A~aS
~0 ...R
~R
~2...R
~ t] and [B, m0...mt, ~bS~0 R
~R
~2...R
~t
[A, m0...mt, ~aS
t
2
0
0
~
~
~ R~2 ...R
~ t is empty, then [A, m0...mt, ~aS
~R
~2...R
~ t] is a senvm0 ...mt B~bS~ 0R2 ...Rt. If ~aS
tence of quantifier rank (m0 ,...,mt ). [A, m0 ...mt ] is that sentence. It follows from
Lemma 2.1 that B is a model of [A, m0 ...mt ] (B |= [A, m0 ...mt ] )iff A vm0 ...mt B.
LK [m0 ...mt ] in the set of formulas in L K that contain no more than m 0 individual variables,
no more than m1 set variables and no more than m t relational variables of degree t. A and
B are equivalent in L K iff for all , a sentence in LK , A |= iff B |= . And they are
m0 ...mt equivalent iff for all , a sentence in LK [m0 ...mt ], A |= iff B |= . It follows
from Lemma 2.1 of Weaver and Penev [2005] that if A vm0 ...mt B, then A and B are
m0 ...mt equivalent. The following is then immediate from Lemma 2.1 above.
Theorem 2.1. A and B are m0 ...mt equivalent iff A vm0 ...mt B.
It follows from Theorem 2.1 that if is a subset of a cumulative domain of A that is
definable in A, then there are m0 ,...,mt such that is closed under vm0 ...mt in A. Suppose
that (X) defines in A. Let (m0 ,...,mt ) be the quantifier rank of this formula. is closed
under vm0 ...mt in A. Suppose that a and that Aa vm0 ...mt Ab. Let be the result of
replacing every free occurrence of X in (X) by a constant of the same grammatical category not in K. This sentence is of quantifier rank (m 0 ,...,mt ) and Aa |= . By supposition,
Aa and Ab are m0 ...mt equivalent. Thus, Ab |= and A |= (X)[b]. Hence, b . Similar
reasoning establishes that if is a subset of the direct product of cumulative domains in A,
then there are m0 ,...,mt such that is closed under vm0 ...mt in A.
Let (X) be a formula of quantifier rank (m 0 ,...,mt ) that is satisfied in A.
H((X), A, m0...mt)
is
{[A, m0...mt, a]|for all a such that A |= (X)[a]}.
Notice that A |= X((X) H((X), A, m0...mt)). Suppose that A |= (X)[a]. Then
[A, m0...mt, a] is a disjunct in H((X), A, m0...mt) and a satisfies the disjunction in
A. Suppose that a satisfies the disjunction in A. Thus, by Lemma 2.1, there is b in the
appropriate cumulative domain of A such that A |= (X)[b] and Aavm0 ...mt Ab. As (X)
is of quantifier rank (m 0 ,...,mt ), A |= (X)[a].
Let D be a cumulative domain of A and a and b be members of D. a is m0 ...mt equivalent
to b in A iff Aavm0 ...mt Ab. This relation is an equivalence relation on D and the cells in
the partition on D induced by this relation consist of those members of D satisfying [A,
m0 ...mt , a] for aD. Thus, the subsets of D that are definable in A are the union of some of
the cells in the partition induced by m 0 ...mt equivalence, for appropriate m 0 ,...,mt .

120

George Weaver

Let (X1 ...Xn ) be a formula of quantifier rank (m 0 ,...,mt ) that is satisfied in A. If all
of the variables are of the same degree, or all first order variables in X 1...Xn occur before
set or relational variables, all set variables occur before all relational variables, and for all
n2, all n-ary variables occur before all variables of higher degree, then we can proceed as
above. For in these cases, if (a1,...,an ) satisfies (X1...Xn ) in A, then a1...an is a sequence
in A and [A, m0...mt, a1...an ] can be formed. Now, suppose that this condition does not
hold. Let D 1,..., Dn be cumulative domains of A. There is g, a permutation on {1, ..., n},
such that for all (a1 ,...,an )D1 ...Dn , ag(1)...ag(n) is a sequence in A.
H((X1...Xn), A, m0...mt)
is
{[A, m0...mt, ag(1)...ag(n)]|for all a1,...,an such that A |= (X1...Xn)[a1...an ]}
The reasoning above and the observation that Aa1...an vm0 ...mt
Aag(1)...ag(n) vm0 ...mt Abg(1)...bg(n) establish the following.

Ab1...bn iff

Theorem 2.2. Assume that A is an interpretation of type K, n is a non-negative integer,

and (X1 ...Xn ) is a formula in L K of quantifier rank (m0 ,...,mt ) that is satisfied in A. Then,
A |= X1 ...Xn((X1...Xn) H((X1...Xn), A, m0...mt)).
Let (a1,...an ) and (b1 ,...bn ) be members of D1 ...Dn . (a1,...an ) is m0 ...mt equivalent
to (b1 ,...bn ) in A iff Aa1...an vm0 ...mt Ab1...bn . As above this relation is an equivalence
relation on D 1 ...Dn. Further, each subset of D1 ...Dn that is definable in A is the
union of some of the cells of the partition induced by this relation.
Corollary 2.1. Assume that D and D 1,...,Dn are cumulative domains of A, D, and
that D1 ...Dn .
1. If is definable in A, then there is t 2 and non-negative integers m 0 ,...,mt such that
(a) is closed under vm0 ...mt in A;
(b) if K is finite, then there is a positive integer q and 1 ,...,q such that these
sets are cells in the partition on D induced by m 0 ,...,mt equivalence and
=1...q .
2. If is definable in A, then there is t 2 and non-negative integers m 0 ,...,mt such that
(a) is closed under vm0 ...mt in A;
(b) if K is finite then, there is a positive integer q and 1,...,q such that these sets
are cells in the partition on D 1 ...Dn induced by m0 ,...,mt equivalence and
=1...q .

Second-order Definability in a Model

3.

121

The Characterization of Definability in A

In this section it is shown that closure under vm0 ...mt is a necessary and sufficient condition
for definability in an interpretation.
Theorem 3.1. Assume that D and D 1,...,Dn are cumulative domains of A, D, and
that D1 ...Dn . Then,
1. if there is t 2 and non-negative integers m 0 ,...,mt such that is closed under
vm0 ...mt in A, then is definable in A;
2. if there is t 2 and non-negative integers m 0 ,...,mt such that is closed under
vm0 ...mt in A, then is definable in A.
Proof. Suppose that is a subset of D and that is closed under vm0 ...mt . Let =
{[A, m0 ,...,mt ,a] |a }. By construction, is finite, when K is. Without loss of
generality, X1 is the one and only variable occurring free in each member of . Let (X1 )
be the disjunction of the members of . Let aD. It suffices to show that a iff A |=
(X1 )[a]. Suppose that a . A |= [A, m0 ,...,mt ,a][a]. Thus, A |= (X1 )[a]. Suppose
that A |= (X1 )[a]. Thus, there is b such that Aa vm0 ...mt Ab. By supposition, a .
Suppose that D1 ...Dn and is closed under vm0 ...mt . Let =
{[A, m0 ,...,mt ,ag(1)...ag(n) ] |(ag(1), ..., ag(n) ) }, where g is a permutation on {1,...,n}
and ag(1)...ag(n) is a sequence in A. Let Xg(1),...,Xg(n) be the variables occurring free
in each member of . Let (Xg(1)...Xg(n) ) be the disjunction of the members of .
Let g ={(ag(1),...,ag(n) )| (a1,...,an ) }. g is closed under vm0 ...mt . By reasoning as
above, (Xg(1)...Xg(n) ) defines g in A. It suffices to show that there is (X1 ...Xn ) of
the same quantifier rank as (Xg(1)...Xg(n) ) such that for all (a1 ,...,an ) in D1 ...Dn ,
A |= (X1 ...Xn )[a1 ,...,an ] iff A |= (Xg(1)...Xg(n) )[ag(1)...ag(n) ]. For i between 1 and n,
the formula (Xi ) is defined as follows. If X i is an individual variable, (Xi ) is Xi =Xi .
If Xi is a set variable and m 0 is non-zero, then (Xi ) is y (Xi (y)Xi (y)), where y is an
individual variable occurring bound in (Xg(1)...Xg(n) ). If m0 =0, then K contains an individual constant, k, and (Xi ) is (Xi (k)Xi (k)). If Xi is a relational variable, (Xi ) is
defined analogously. (X1 ...Xn ) is (X1 )&...&(Xn )&(Xg(1)...Xg(n) ).
Even stronger results can be obtained when A is infinite. The following is essentially
Theorem 2.8 (page 504) of Weaver and Penev [2005].
Lemma 3.1. Assume that A and B are infinite interpretations of type K, t2, and
m0 ,...,mt are non-negative integers. Then, Avm0 ...mt B, if A
vmax{5,m0 +(t1)}(Pt mi )2B.
i=1

It follows from Lemma 3.1 that if A and B are infinite interpretations of type K, and
for all m0 , m1 , Avm0 m1 2 B, then A and B are equivalent in L K . Stronger versions of this
particular consequence of Lemma 3.1 are obtained below.
Corollary 3.1. Assume that A is an infinite interpretation of type K, D, D 1 ,..., Dn are
cumulative domains of A, D, and that D1 ...Dn . Then,

122

George Weaver

1. is definable in A iff there are m0 and m1 such that is closed under vm0 m1 2 in A;
and
2. is definable in A iff there are m0 and m1 such that is closed under vm0 m1 2 in A.

It follows from Corollary 3.1 and the proof of Theorem 3.1 that if or is definable
in infinite interpretation A, then there is a formula defining (or ) that contains no more
than 2 bound binary relational variables and no bound relational variables of degree greater
than 2. Note that Theorem 3.1 also yields a characterization of first-order definability in
an interpretation (closure under vm0 00) and of monadic second-order definability in an
interpretation (closure under vm0 m1 0 ).

4.

Definable Pairing Functions

A pairing function on A is an injection of A 2 into A. Only infinite interpretations have

pairng functions on their domains. Further, by Corollary 3.1, if a pairing function on the
domain of A is definable in A, there are m0 and m1 such that this function is closed under
vm0 m1 2 . In this section it is shown that if a pairing function on the domain of A is closed
under either vm0 m1 0 or vm0 m1 1 , then a stronger version of the consequence of Lemma 3.1
mentioned above can be obtained and, hence, also a stronger version of Corollary 3.1.
Let 1 and 2 be binary relations on the set A. 1+2={(a, b, c)|(c, a) 1 and (c,
b) 2}. Let be any binary function on A. Let 1 = {(a, b)| c such that (b, c)=a} and
let 2={(a, b)| c such that (c, b)=a}. It is easily verified that if is a injection on A, then
=1+2. The following are used extensively in the following.
Lemma 4.1. Assume that A and B are infinite interpretations of type K, 1 and
2 are binary relations on A, and 1, 2 are binary relations on B, 1+2 is a pairing function on A and 1 +2 is a pairing function on B. Then A12vm0 m1 2 B12 , if
A12v(m0 +2)(m1 +2)0B1 2.
Lemma 4.1 is immediate from Lemma 2.7 of Weaver and Penev [2005] (page 504). This
lemma was stated for bijections between A 2 and A. However, it is easily verified that the
reasoning for Lemma 2.7 also holds for injections of A 2 into A. In both Shapiro [1991] and
Shapiro [2001], pair function is used for injections of A 2 into A. It follows from Lemma 3.1
that infinite interpretations A and B are equivalent in L K iff for all m0 , m1 , Avm0 m1 2B.
Thus, by Lemma 4.1, when A12 and B12 satisfy the hypothesis, they are equivalent in
LK , if for all m0 , m1 , A12 vm0 m1 0 B12.
Lemma 4.2. Assume that A and B are of type K, n2, An is closed under vm0 ...mt
in A and that A vm0 +n...mt B. Then, there is Bn such that
1. is closed under vm0 ...mt in B; and
2. there is a formula (x1 ...xn ) that defines in A and in B.

Second-order Definability in a Model

123

Proof. Suppose that A vm0 +n...mt B and that An is closed under vm0 ...mt in A.
Let be {(b1,...bn )| a1,...,an such that (a1 ,...,an ) and Aa1...an vm0 ...mt Bb1 ...bn }.
Suppose that (b1,...,bn ) and that Bb1 ...bn vm0 ...mt Bd1 ...dn . By construction, there are
a1,...,an such that (a1 ,...,an ) and Aa1 ...an vm0 ...mt Bb1 ...bn . By supposition, there are
c1 ,...,cn such that Ac1 ...cn vm0 ...mt Bd1 ...dn . Therefore, Ac1...cn vm0 ...mt Aa1...an . Hence,
(c1 ,...,cn ) . By construction, (d 1 ,...,dn ) . Thus, is closed under vm0 ...mt in B.
Notice that a1...an is a sequence in A and that b1...bn is a sequence in B. is definable
in A by
{[A, m0...mt, a1...an]|(a1, ..., an) }.
Further, is definable in B by
{[B, m0...mt, b1...bn]|(b1, ..., bn) }.
[B, m0 ...mt , b1 ...bn ] and [A, m0 ...mt , a1...an ] are identical, if Aa1...an vm0 ...mt
Bb1 ...bn . Thus, by construction of , the above disjunctions are identical.
Lemma 4.3. Assume that A is infinte and that 1 and 2 are binary relations on A.
Then,
1. if 1 is closed under vm0 m1 m2 in A and 2 is closed under vn0 n1 n2 in A, then 1+2
is closed under vt0 t1 t2 in A, where for i between 0 and 2, t i is the largest of m i and
ni ;
2. if 1+2 is a pairing function on A and closed under vm0 m1 m2 in A, then both 1 and
2 are closed under vm0 +1m1 m2 in A.
Proof. Suppose that (a,b,c) 1 +2 and that Aabcvt0 t1 t2 Adeh. (c,a) 1 and (c,b)
2. By the choice of ti , Aacvm0 m1 m2 Adh, and Abcvn0 n1 n2 Aeh. Thus, Acavm0 m1 m2 Ahd,
and Acbvn0 n1 n2 Ahe. By supposition, (h,d) 1 and (h,e) 2. Hence, (d,e,h) 1+2.
Suppose that 1+2 is a pairing function on A and closed under vm0 m1 m2 in A. Suppose that (c,a) 1 and that Aca vm0 +1m1 m2 Ac0a0 . By supposition, there is b such that
(a,b,c) 1 +2. Thus, there is b0 such that Acab vm0 m1 m2 Ac0a0 b0 . Therefore, Aabc
vm0 m1 m2 Aa0b0 c0 and (a0 ,b0 ,c0 ) 1 +2. Hence, (c0 ,a0 ) 1 . Similar reasoning shows that
2 is closed under vm0 +1m1 m2 .
The following establishes one important consequence of the definability of pairing functions by formulas having fewer than two binary relational variables.
Theorem 4.1. Assume that A is an infinite interpretation of type K and that is a
pairing function on A. Then,
1. if is closed under vm0 m1 0 in A, then for all B, if for all n0, n1, A vn0 n1 0 B, then
A and B are equivalent in L K ; and
2. if is closed under vm0 m1 1 in A, then for all B, if for all n0, n1, A vn0 n1 1 B, then
A and B are equivalent in L K .

124

George Weaver

Proof. Assume that A is infinite and that is a pairing function on A. Suppose that
is closed under vm0 m1 0 and that for all n 0 , n1 A vn0 n1 0 B. There are 1 and 2, binary
relations on A, such that =1+2. By Lemma 4.2, both of these relations are closed under
vm0 +1m1 0 . Let B be an interpretation of type K. Suppose that for all n 0, n1 , A vn0 n1 0B.
By Lemma 4.2, there are 1 and 2, binary relations on B such that both relations are closed
under vm0 +1m1 0 in B. Further, there are 1 (x1 x2) and 2(x1 x2 ), formulas of quantifier
rank (m0 +1,m1,0), such that 1 (x1 x2) defines 1 in A and 1 in B and 2 (x1x2 ) defines 2
in A and 2 in B. There is a sentence, , in LK [(m0 +6)m10] such that C|= iff {(c,a)|
C|= 1 (x1x2 )[ca]}+ {(c,b)| C|= 2 (x1 x2)[cb]} is a pairing function on C. Thus, 1 +2 is
a pairing function on B. Hence, B is infinite and Lemma 4.1 can be applied to A12 and
B12 . It suffices to show that for all n 0, n1, that A12vn0 n1 0 B12 . Let T be the set of
~ a sequence in A, and all ~bS~ 0, a sequence in B, if ~aS
~ and
all (n0 , n1 ) such that for all ~aS,
~bS~ 0 are similar and A~aSv
~ n +(m +1)(n +m )0B~bS~ 0, then A12~aSv
~ n n 0 B12~bS~ 0. If T
0 1
0
0
1
1
includes all ordered pairs of non-negative integers, then A12 and B12 are equivalent
and so are A and B.
~ (m +1)m 0B~bS~ 0. It
Reasoning proceeds by showing that (0, 0) T. Suppose that A~aSv
0
1
0 []. By supposition, A~
~
~
~
~
~

a
S[]B
b
S~0 []. Let
b
S
suffices to show that A12~aS[]B
1 2
~ Let b and b0 be
a and a0 be members of A that are denoted by individual constants in A~aS.
0
~
~
those members of B denoted in BbS by the same individual constants. It suffices to show
that (a, a0) 1 iff (b, b0 ) 1 and that (a, a0) 2 iff (b, b0 ) 2. Suppose that (a, a0 ) 1 .
Recall that 1 (x1 ,x2) defines 1 in A and 1 in B. Further, this formula is of quantifier rank
(m0 +1,m1 ,0). (a, a0 ) satisfies the formula in A. By supposition, (b, b 0) satisfies the formula
in B. Hence, (b, b0 ) 1. The reasoning in the other direction is analogous. Similar
reasoning shows that (a, a 0) 2 iff (b, b0) 2. Therefore, (0, 0 )T.
Suppose that if n 06=0, then (n0-1, n1 )T; and that if n 1 6=0, then (n0, n1 -1)T. Reasoning
~ n +(m +1)(n +m )0
proceeds by showing that (n 0, n1)T. Suppose A~aSv
0
0
1
1
~ n n 0 B12~bS~ 0. Suppose that n 0 6=0 and that
B~bS~ 0. It suffices to show that A12~aSv
0 1
~ (n 1)+(m +1)(n +m )0
aA. By supposition, there is bB such that A~aaSv
0
0
1
1
~ (n 1)n 0B1 2~bbS~ 0. Reasoning in the same way,
B~bbS~ 0. Since (n0-1, n1 )T, A12~aaSv
0
1
~ (n 1)n 0
it can be shown that given bB there is aA such that A12~aaSv
0
1
B12~bbS~ 0. Suppose that n 1 6=0 and that SA. As above there is S0 B such that
~
~ n +(m +1)(n 1)+m 0 B~bbS~ 0S0 . (n0, n1-1)T. Hence, A12~aSSv
A~aSSv
n0 (n1 1)0
0
0
1
1
0
0
0
~
B12~bS~ S . In the same way, given S B there is SA such that A12~aSS
0
0
0
~
~
~
~
~
vn0 (n1 1)0 B12 bS S . Therefore, A12~aSvn0 n1 0 B12bS , and (n0, n1)T. Thus, T is
N2 and A and B are equivalent.
Suppose that is closed under vm0 m1 1. Reasoning proceeds as above, except that
1 (x1,x2 ) and 2 (x1,x2 ) are of quantifier rank (m0 ,m1 ,1) and (n0 , n1)T iff for all
~ and ~bS~ 0 are similar and
~ a sequence in A, and all ~bS~ 0, a sequence in B, if ~aS
~aS,
~ n n 0 B12~bS~ 0.
~ n +(m +1)(n +m )1B~bS~ 0, then A12~aSv
A~aSv
0 1
0
0
1
1
The reasoning for Theorem 4.1 together with Lemma 4.1 provide reduction results that
are used below. Assume that A and B are infinite interpretations of the same type, is a
pairing function on A and 1+ 2 is a pairing function on B. Then,
1. if is closed under vm0 m1 0 in A, then A12vn0 n1 2B1 2, if A

Second-order Definability in a Model

125

v(n0 +2+m0 +1)(n1 +2+m1 )0B; and

2. if is closed under vm0 m1 1 in A, then A12vn0 n1 2B1 2, if A
v(n0 +2+m0 +1)(n1 +2+m1 )1B.
These results are used below to establish stronger versions of Corollary 3.1.
Theorem 4.2. Assume that A is an infinite interpretation of type K, D, D 1,..., Dn are
cumulative domains of A, D, D1 ...Dn and that is a pairing function on A.
Then,
1. if is closed under vm0 m1 0 in A, then
(a) is definable in A iff there are t0 and t1 such that is closed under vt0 t1 0 in
A;
(b) is definable in A iff there are t0 and t1 such that is closed under vt0 t1 0 in A;
2. if is closed under vm0 m1 1 in A, then
(a) is definable in A iff there are t0 and t1 such that is closed under vt0 t1 1 in
A;
(b) is definable in A iff there are t0 and t1 such that is closed under vt0 t1 1 in A.
Proof. Assume that A is an infinite interpretation, is a pairing function on A closed
under vm0 m1 0 in A. Suppose that D and that is definable in A. By Corollary 3.1,
there are n0, and n1 such that is closed under vn0 n1 2 in A. It suffices to find t 0 and t1
such that is closed under vt0 t1 0 in A. Notice that is closed under vn0 n1 2 in A12 .
Let t0 =(m0 +2+n0+1) and t 1=(m1 +2+n1). Suppose that a and that Aa vt0 t1 0 Ab. By the
remarks above, Aa12 vn0 n1 2Ab12. Hence, A12 avn0 n1 2 A12b. Thus b . By
similar reasoning, if D1 ...Dn , is closed under vt0 t1 0 in A. The reasoning for case
two is analogous.
It follows from Theorem 4.1 that if A is infinite and a pairing function on A is definable
in A by a monadic formula (one of quantifier rank (n 0,n1 ,0)), then the second-order theory
of A is determined by its monadic second-order theory in the sense that if for all B, if for
all n0 , n1 Avn0 n1 0 B, then A and B are equivalent. For example, if K contains at least
three predicate constants of degree three and A is an interpretation of type K whose domain
is the set of non-negative integers and addition, multiplication and exponentiation are the
among the extensions of the members of K in A, then there is a pairing function on A that
is definable in A by a monadic formula. Other examples are given in Shapiro [1991] (page
222).
It is natural to ask whether or not if A is infinite and the second-order theory of A is
determined by its monadic second-order theory, then there is a pairing function on A that is
definable by a monadic second-order formula. The standard model of the monadic theory
of (see Gurevich [1985], page 485) provides a counterexample. It follows from Theorem
4.2 that a negative answer to the above question follows from the existence of an attribute

126

George Weaver

of A that is definable in A but not definable in A by a monadic second-order formula. Let

K consist of a single binary predicate constant. A is of type K, A is and the successor
function on is the extension of the predicate constant in K. The second-order theory of
A has a finite set of axioms all of whose models are isomorphic. None of these axioms
contains a relational variable. According to Robinson [1958], Tarski, in his lectures, raised
two problems about A: is it possible to give a monadic definition of addition in A and
is there a decision method for the monadic theory of A. Robinson observed (page 239)
that if addition is definable by a monadic formula, then the monadic theory of A is not
decidable. In 1960 B
uchi presented the desired decision method. Thus, no pairing function on is definable in A by a monadic second-order formula even though one such is
definable by a second-order formula and the second-order theory of A is determined by its
monadic second-order theory. Gurevich [1985] contains other counterexamples. Similar
results about pairing functions definable in A by a formula containing one binary relational
variable follow from Theorem 4.1. To the authors knowledge, it is open whether or not
the following is true: if the second-order theory of A is determined by the set of sentences
of quantifier rank (n 0 ,n1 ,1), for all n0 and n1, then some pairing function on A is definable
in A by a formula of quantifier rank (n 0 ,n1 ,1). It follows from Theorem 4.2 that a counterexample can be obtained by exhibiting an attribute of A that is definable in A, but not
definable in A by a formula containing exactly one bound binary relational variable and no
bound relational variables of higher degree.

5.

Conclusion

The purpose of this paper is to present additional applications of the characterization of

equivalence presented in Weaver and Penev [2005]. It should be noted that all of the results above hold for both the finitary and the infinitary languages presented in that paper.
Although the proofs are original, some of the results in Section 4 are well known in the
finitary case: parts 1 of Theorem 4.1 and Theorem 4.2. These results in the infinitary case
are, to the authors knowledge, new. Parts 2 of Theorems 4.1 and Theorem 4.2 are, to the
authors knowledge, new in both the finitary and the infinitary case. It should be noticed
that closure under vn0 00 is a necessary and sufficient for definability in both the finitary
and the infinitary first-order languages mentioned in Weaver and Penev [2005] (page 506).
Similar remarks apply to closure under vn0 n1 0 and monadic second-order languages.

References
[1] J. R. B
uchi, On a Decision Method in Restricted Second Order Arithmetic, Logic,
Methodology and the Philosophy of Science , edited by E. Nagel, P. Suppes, A. Tarski,
Stanford University Press, Stanford California (1962), 1-11.
[2] Y. Gurevich, Monadic Second-Order Theories, Model-Theoretic Logics, edited by J.
Barwise and S. Feferman, Springer-Verlag, Berlin (1985), 479-506.
[3] R. M. Robinson, Restricted Set-Theoretical Definitions in Arithmetic, Proceedings of
the American Mathematical Society , 9 (1958), 238-242.

Second-order Definability in a Model

127

[4] S. Shapiro, Foundations without Foundationalism: A Case for Second-order Logic ,

Clarendon Press, Oxford, (1991).
[5] S. Shapiro, Systems Between First-Order and Second-Order Logics, Handbook of
Philosophical logic, 2 nd Edition, Volume 1, edited by D. M. Gabby and F. Guenthner,
Kluwer Academic Press, Dordecht (2001), 131-188.
[6] G. Weaver and I. Penev, From finitary to infinitary second-order logic, The Mathematical Logic Quarterly, 51 (2005), 499-506.

In: Mathematics and Mathematical Logic: New Research

ISBN 978-1-60692-862-2
c 2010 Nova Science Publishers, Inc.
Editors: Peter Milosav and Irene Ercegovaca

Chapter 6

A LGEBRAIC T OPICS ON D ISCRETE

M ATHEMATICS *
Gloria Gutirrez Barranco, Javier Martnez,
Salvador Merino and Francisco J. Rodrguez
Departamento de Matemtica Aplicada.
E.T.S.I. Telecomunicacin. Universidad de Mlaga.
Campus de Teatinos. 29071 Mlaga. (Spain)
Abstract
Many applications of discrete mathematics for science, medicine, industry and
engineering are carried out using algebraic methods, such as group theory, polynomial
rings or finite fields.
This chapter is intended as a survey of the main algebraic topics used for developing methods in ambit of combinatorial theory. Plyas enumeration method, Latin
squares, patterns design or block design are examples of this usage. Others applications can be seen in Section 5, about foundation design, RNA patterns or octonions.

1.

Elemental Concepts

1.1.

Equivalence Relations

A binary relation on a set A is a subset R A A. If (a, b) R, we say that the element

a is related to b and we will denote aRb. Equivalence relations are binary relations that
formalize the natural process of classification of the elements in a set. Formally:
Definition 1.1. A binary relation R on a set A is an equivalence relation if it is reflexive
(aRa for all a A), symmetric (aRb implies bRa for all a, b A), and transitive (aRb
and bRc imply aRc for all a, b, c A).
Equivalence relations are usually denoted by or .
Definition 1.2. Let A be a set, be an equivalence relation on A and a A. The equivalence class of a, denoted by [a], is the set
[a] = {b A | a b}

130

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

The quotient set, denoted A/ , is the set of all the equivalence classes of A, that is, A/ =

[a] | a A .
Definition 1.3. Let A be a set. A partition of A is a family of subsets, named parts, that are
disjoint by pairs and whose union is A.
Obviously the quotient set A/ is a partition of A. Conversely, if a partition P is given
on A, we can define an equivalence relation on A as x y iff there exists a part of P which
contains both x and y. Equivalence relation and partition are thus essentially the same
notion.

1.2. Groups
The idea of operating the elements of a set arises in a natural way. Thus, for example, with
numbers it is possible to add, to multiply, etc; with sets it is possible to join, to meet, etc.
Binary operations are the keystone of the algebraic structures studied in abstract algebra 1 .
Definition 1.4. A magma is a pair (A, ) where A is a set and is a binary operation on
A, that is, a function : A A A.
Binary operations are often written using infix notation such as a b, a + b, a b or a  b
rather than by functional notation of the form f (a, b). Sometimes they are even written just
by juxtaposition: ab.
Example 1.5. The sum of natural numbers is a binary operation. Nevertheless the subtraction of natural elements is not a binary operation because for example if we consider
2, 3 N it verifies that 2 3 = 1 6 N.
The binary operations on a set A, do not have to verify any property in particular.
However, the fact that these binary operations verify certain properties is going to play a
very outstanding role.
Definition 1.6. Let be a binary operation on a set A. It is said that:
1. is associative if it satisfies that a (b c) = (a b) c for all a, b, c A.
2. is commutative if it satisfies that a b = b a for all a, b A
3. e A is an identity element or neutral element for if it satisfies that ae = ea = a
for all a A.
4. a0 A is an inverse element of a A if it satisfies that a a0 = a0 a = e being
e A the identity element.
Many binary operations of interest in the algebraic environment are commutative or
associative. Many of them also have identity elements and inverse elements.
Typical examples of associative and commutative binary operations are the addition
(+) and multiplication () of numbers. The product of matrixes and the composition of
functions are associative but not commutative. The subtraction () and the division (/) of
numbers are neither associative nor commutative.
The following result is an immediate consequence of the definition.
1

All basic results about algebra can be found in several algebra books like [5].

Algebraic Topics on Discrete Mathematics

131

Proposition 1.7. Let be a binary operation on a set A. Then:

1. If has an identity element, it is unique.
2. If is associative, and an element a A has an inverse element, this is unique.
Now we are prepared to give the definition of the well-known structures named semigroup, monoid and group:
Definition 1.8. A magma (G, ) is a semigroup if is associative. A monoid is a semigroup
that has identity element. A group is a monoid in which each element has its inverse one.
It is said that a semigroup, a monoid or a group (G, ) is commutative or abelian, if
is commutative.
It is obvious that (Z, +) is a commutative group. Nevertheless, (Z, ) is not even a
semigroup.
As a consequence of Proposition 1.7, a group has exactly one identity element and the
inverse of an element is unique. This fact allows to select symbols to denote it. So, it is
usual to use different notation, depending on the context and the group operation.
In additive groups, + denotes the operation, 0 is the identity element and a is the
inverse element of a, commonly named opposite. So we have a + (a) = 0.
In multiplicative groups, denotes the operation, 1 is the identity element and a1 is
the inverse element of a. So we have a a1 = 1. It is very common to drop the
and to use juxtaposition.
Also note that a group (G, ) is often denoted simply G when there is no ambiguity about
the operation. Omitting the symbol for the operation is generally acceptable, and leaves it
to the reader to know the context and the group operation. In addition from now on, if no
confusion arises we will use the multiplicative notation.
Definition 1.9. Let G be a group and 6= H G. If the restriction of the operation to H
is a group operation on H we say that H is a subgroup of G.
In detail, H is a subgroup of G if and only if the following conditions hold:
1. H is closed for the binary operation, that is, if a, b H then a b H. Note that the
associativity in H is an immediate consequence.
2. The identity element 1 of G belongs to H, that is, 1 H.
3. If a H then the inverse element a1 H.
As an obvious example, if we consider the group (R, +), Z is a subgroup but Q+ is not.
The following theorem provides a characterization for subgroups.
Theorem 1.10. Let G be a group and 6= H G. H is a subgroup of G if and only if for
all pair of elements a, b H it satisfies that a b1 H.

132

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

Given an element a G, and n a positive integer, we represent the power an (in

additive notation na) the operation of a by itself n times, i.e.
n times
z
}|
{
a = aa...a
n

The definition of power can be extended when n Z,

a0 = 1
If n is a negative integer, an = a1

n

The powers have the following properties:

Proposition 1.11. Given an element a of a group G and integers m and n, we have:
(an )1 = a1

n

am+n = am an

and

(am )n = amn .

Example 1.12. Given an element a G the set hai = {an | n Z} is a subgroup of G

named cyclic subgroup generated by a.
Definition 1.13. Let G be a group and H a subgroup of G. For any a G we can define
the left coset and right coset of H in G as:
aH = {a h | h H}

Ha = {h a | h H}

When the number of left cosets of H is finite, then it coincides with the number of right
cosets of H. This number is named index of H in G and is represented by the symbol
[G : H].
Definition 1.14. Let G be a finite group, |G| denotes the number of elements in G and is
named order of G. If G is not finite we say that it has infinite order. The order of an element
a in a group G is the order of the cyclic subgroup hai.
Usually, when | hai | is finite, the order of a is defined as the minor positive integer n
such that an = 1. In this case, both definitions are equivalent.
Now, we highlight an important result which is interesting for the rest of the development:
Theorem 1.15 (Lagranges Theorem). Let G be a finite group. Then
|G| = [G : H]|H|
Corollary 1.16. The order of every subgroup H of a finite group G divides the order of G.

Algebraic Topics on Discrete Mathematics

133

Quotient group. We will use a kind of groups that are known as quotient groups owing to
the support set is a quotient set. There exists a type of subgroups, named normal subgroups,
which allow to define an equivalence relation compatible with the operation and so they
induce an operation in the quotient set.
Definition 1.17. Let G be a group and H a subgroup of G. We say that H is a normal
subgroup if it satisfies that aH = Ha, for all a G.
Given a group G and a normal subgroup H of G, the relation
a H b a b1 H

(1)

is an equivalence relation, the equivalence class are the cosets and G/H denotes the quotient
set.
Lemma 1.18. Let G be a group and H a normal subgroup of G. Then (G/H, ) is a group
where is the derived operation, that is, [a] [b] = [a b] for all [a], [b] G/H.
Homomorphism of groups. Now, we are going to give the definition of homomorphism 2 ,
that is, a map from one algebraic structure to another of the same type that preserves all the
relevant properties.
Definition 1.19. Let (A, 1 ) and (B, 2 ) be groups. An homomorphism is a map f : A B
such that f (a 1 b) = f (a) 2 f (b) for all a, b A.
An homomorphism is named monomorphism, epimorphism or isomorphism if it is injective, surjective or bijective, respectively.
Definition 1.20. Let f : (A, 1 ) (B, 2 ) be a group homomorphism. We define the
kernel of f as
ker(f ) = {x A | f (x) = eB }
Theorem 1.21. Let f : (A, 1 ) (B, 2 ) be a group homomorphism. Then ker(f ) is a
normal subgroup of A and f (A) is a subgroup of B.
Usually, f (A) is named image group.
Theorem 1.22. Let f : (A, 1 ) (B, 2 ) be a group homomorphism. Then, the quotient
group A/ker(f ) and the image group f (A) are isomorphic groups.

1.3. Permutation groups

The concept of a permutation expresses the idea that distinguishable objects may be arranged in various different orders. For instance, with the numbers of the set A =
{1, 2, 3, 4, 5}, each possible order makes a list of the numbers, without repetitions. One
such permutation is: 2, 3, 1, 4, 5. So a permutation is an ordered sequence containing
each symbol from a set once and only once; neither 1, 2, 2, 4, 3 nor 2, 1, 5, 4 are permutations of the set A = {1, 2, 3, 4, 5}.
2

The word homomorphism comes from the Greek language: homo meaning same and morphos meaning
shape.

134

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

One can therefore point to the essential difference between a permutation and a set:
the elements of a permutation are arranged in a specified order. Given the set A =
{a1, a2, . . . , an } an arrangement ai1 , ai2 , . . . , ain can be seen as a bijection in A, i.e.
7 ai1
a1
a2
7 ai2
..
.

or more compact

a1
ai1

a2
ai2

...
...

an
ain

an 7 ain
In a nutshell, given a finite set A 6= , a permutation is a bijection from A onto itself. We
will denote by SA the set of all permutations of A, and by , , . . . the elements of SA .
Using that the composition of bijections is also a bijection is easy to check that SA is a
group. It is usual the multiplicative notation for this group. So
=
These groups depend only of the number of elements of A, because if |A| = |B| then SA is
isomorphic to SB . Therefore,
Definition 1.23. Given a positive integer n, the symmetric group Sn is the group of the
permutations of a set with n elements.
From now on, we will consider the group Sn defined over the set A = {1, 2, . . ., n}.




1 2 3 4 5
1 2 3 4 5
Example 1.24. Given =
and =
in S5, then
1 3 2 4 5
2 3 1 5 4
we have that:


1 2 3 4 5
=
3 2 1 5 4
Example 1.25. The element of S3 are:



1 2 3
1 =
0 =
1 2 3



1 2 3
2 =
1 =
1 3 2

1 2 3
3 1 2
1 2 3
3 2 1




2 =
3 =




1 2 3
2 3 1
1 2 3
2 1 3




and the operation of this group is described by the following table

0
1
2
1
2
3

0 1
0 1
1 2
2 0
1 2
2 3
3 1

2
2
0
1
3
1
2

1
1
3
2
0
2
1

2
2
1
3
1
0
2

3
3
2
1
2
1
0

This a typical example of non-commutative group. However, {0, 1, 2} is a subgroup

that is commutative.

Algebraic Topics on Discrete Mathematics

135

Cycles. We define a cycle of length k or k-cycle as a permutation Sn such that

there exist k elements a1 , a2, . . . , ak {1, 2, . . ., n} fulfilling (ai ) = ai+1 with i
{1, . . ., k 1} and (ak ) = a1 whereas the other elements stay invariant.
Instead writing its explicit table a cycle is usually written in the compact form =
(a1, a2, . . . , an ). In S5 , for example, = (1, 5, 2, 4) denotes the cycle


1 2 3 4 5
5 4 3 1 2

If the objects are denoted by a single letter or digit, commas are also dispensed with,
and we have a notation such as (1 5 2 4).
Any unitary cycle obviously represent the identity permutation. A cycle formed by
two elements is named transposition. Two cycles are disjoint if they move different
elements, that is, if they act over different elements. In this case, the product commutes.
Notice that in general, the product of cycles is not necessarily a cycle. For example, in
S7 the product of the cycles (2 4 6) and (3 1) is not a cycle:
(2 4 6)(3 1) =

1 2 3 4 5 6 7
1 4 3 6 5 2 7



1 2 3 4 5
3 2 1 4 5

1
=
3

6 7
6 7

2 3 4 5 6 7
4 1 6 5 2 7

Theorem 1.26. Every permutation is a cycle or a product of disjoint cycles. Moreover this
factorization is unique up to the order of the factors.
Example 1.27. In S5 we have


1 2 3 4 5 6
4 3 5 1 2 6

= (1 4)(2 3 5) = (2 3 5)(1 4)

Now we have the following result:

Theorem 1.28. Any permutation is a product of transpositions. This factorization is not
unique; however, the number of transpositions needed to factorize a given permutation is
either always even or always odd.
For instance, the permutation given in Example 1.27 can be written as


1 2 3 4 5 6
4 3 5 1 2 6

= (1 4)(2 3)(3 5) = (1 4)(2 5)(2 3)

If a permutation can be written as a product of an odd number of transpositions, it is then

named an odd permutation, otherwise it is an even permutation. Notice that the product of
two even permutations is even, the product of two odd permutations is even, and all other
products are odd.

136

2.

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

Polyas Enumeration Method

2.1. Action of a Group over a Set

Definition 2.1. Let A be a set and G a group. We will say that G acts over A if there exists
an application : G A A such that:
1. (e, c) = c, c A where e is the identity element in the group.
2. (g h, c) = (g, (h, c)), g, h G, c A.
We will represent the action of g G over a A as g(a).
Example 2.2. Let A = {1, 1, i, i} and G = Z4 , the application : GA A defined
by ([g], a) = ig a is an action of the group G over A.
G\A 1
1 i
i
[0] 1
1 i
i
[1]
i
i
1 1
[2]
1 1
i i
[3]
i i 1
1
Our objective is to establish a relation between the action of a group over a set and the
permutations of the elements of this set.
Theorem 2.3. Let A be a set, G a group that acts over A and g G. The application
g : A A defined by g (x) = g(x) is a permutation.
The permutation g defined in the previous theorem is named permutation generated
by g.
Theorem 2.4. Let A be a set, G a group that acts over A and SA the group of the permutations of A. The application : G SA given by (g) = g is a group homomorphism.
Proof. For all g, h G and for all x A we have:
((g h))(x) = (gh)(x) = (g h)(x) = g(h(x)) = g (h(x)) = ((g))(h(x)) =
= ((g))(h(x)) = (g)((h)(x)) = ((g) (h))(x)

As the application : G SA defined in the previous theorem is a group homomorphism, from Theorem 1.21 we have that ker() is a normal subgroup of G.
Using Proposition 1 we consider the equivalence relation in G defined as:
g ker() h

if and only if

g h1 ker()

Each equivalence class defined by this relation is formed by the elements of G generating
the same permutation. From Lemma 1.18 the quotient set G/ ker() is a group and from
Theorem 1.22 we have that G/ ker() and Im() are isomorphic, that is, the group of the
elements of G generating different permutations and the group of permutations generated
by the elements of G are isomorphic. For this reason, we will denote G the group of
permutations generated by the elements of G.

Algebraic Topics on Discrete Mathematics

137

2.2. Orbits
Definition 2.5. Given a set A and a group G that acts over A, we define the orbit of an
element x A as:
Ax = {y A | y = g(x) being g G}
The concept of orbit arises in a natural way if we consider the following equivalence
relation in A:
x y if and only if g G | y = g(x)
It is easy to prove that the equivalence classes defined by this relation are the orbits. As an
immediate consequence we have that the set of orbits give us a partition of the set A. The
quotient set, that is, the set of the orbits is denoted A/G.
Example 2.6. Let us consider a square whose vertexes can be colored with two different
colors. Table 1 shows the 16 different colorations that can be obtained.
Table 1. 2-colorations of a square.

C1

|
|

C2

|
|

C6

|
|

C10

|
|

C12

|
|

C16

C3

|
|

C7

|
|

C11

|
|

C13

C4

|
|

C8

C14

C5

|
|

C9

C15

A = {C1 , . . . , C16} is the set of colorations and G is the group of rotations and reflections acting over the elements of A, that is, G = {0 , 1, 2, 3, r1, r2, r3, r4} where i

138

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

with i {0, 1, 2, 3} are the rotations with amplitude 2i

4 in the clock sense and r1 , r2 , r3
and r4 are the reflections respect to the horizontal axis, the vertical axis, the secondary
diagonal and the main diagonal, respectively.
The action of G over A establishes the following orbits:
O1 = {C1 }
O4 = {C10 , C11}
O2 = {C2 , C3, C4 , C5} O5 = {C12 , C13, C14, C15}
O3 = {C6 , C7, C8 , C9} O6 = {C16 }
Definition 2.7. Given a set A and a group G that acts over A we define the set of stabilizers
of an element x A as:
Gx = {g G | g(x) = x}
So, the set of stabilizers for the coloration C2 in Example 2.6 is GC2 = {0 , r4}.
Theorem 2.8. If G is a group that acts over a set A then Gx is a subgroup of G for all
x A.
Proof. We will use the Corollary1.10. If f, g Gx then (f g 1)(x) = f (g 1(x)) =
f (x) = x and we have that f g 1 Gx .
The subgroup Gx is named stabilizing subgroup. In general, Gx is not a normal subgroup of G as the following example shows.
Example 2.9. For the coloration C2 in Example 2.6 we have that GC2 = {0 , r4}. It is
easy to prove that 1 GC2 6= GC2 1 .
Theorem 2.10. Let us consider a set A and a group G that acts over A. If y = g(x) then
Gy = gGxg 1 .
Proof. Firstly, we will see that gGxg 1 Gy . If gx Gx , then (g gx g 1 )(y) =
(g gx)(g 1(y)) = (g gx)(x) = g(gx(x)) = g(x) = y = g gx g 1 Gy .
Now we will see that Gy gGxg 1 by using that Gy gGxg 1 g 1Gy g Gx .
If gy Gy , then (g 1 gy g)(x) = (g 1 gy )(g(x)) = (g 1 gy )(y) = g 1(gy (y)) =
g 1(y) = x = g 1 gy g Gx .
Corollary 2.11. Let A be a set and G a group that acts over A. If y = g(x) then Gx and
Gy are isomorphic subgroups.
Definition 2.12. Given a set A, a group G that acts over A and two elements x, y A such
that y Ax we define the set of generators of y as:
gen(y) = {g G | g(x) = y}
Theorem 2.13. Given a set A and a group G that acts over A, if y = g(x) then |Gx | =
|gen(y)|.
Proof. We will consider the application : Gx gen(y) given by (h) = g h and we
will see that it is injective: Given h1 , h2 Gx such that (h1) = (h2), we have that,
g h1 = g h2 = g 1 (g h1 ) = g 1(g h2 ) = h1 = h2 . The application is also
surjective because h gen(y) we have that h = (g 1 h). So is a bijection and
|Gx| = |gen(y)|.

Algebraic Topics on Discrete Mathematics

139

Corollary 2.14. Given a set A and a group G that acts over A, if y = g(x) then gen(y) =
gGx.
According to the previous corollary, the left coset of Gx in G for a generator is the set
of generators.
Theorem 2.15. Let us consider a set A, a group G that acts over A and the set G/Gx of
the left cosets of Gx in G. The application : G/Gx Ax given by (gGx) = g(x) is a
bijection.
Proof. From Theorem 2.13, for any y A, we have that |gen(y)| = |Gx| and using
Lagranges Theorem 1.15, |G/Gx| = |Ax|. The map is surjective because y Ax
y = g(x) with g G and this means that y = (gGx). Hence is a bijection.
Definition 2.16. Given a set A and a group G that acts over A we define the set of invariant
elements for g G as:
Ag = {x A | g(x) = x}

2.3. Theorem of Burnside

Theorem 2.17. Given a set A and a group G that acts over A we have that:
|A/G| =

1 X g
|A |
|G|
gG

Proof. If A/G = {A1 , , Ak }, for any orbit Ai A/G we have that G =

and so |G| =

gen(y)

yAi

|gen(y)|.

yAi

If we do this for all the orbits and we have in mind that the set of orbits is a partition of
A, using Theorem 2.13 we have that:
k X
X

|gen(y)| = |A/G| |G| =

i=1 yAi

|Gx |

(2)

xA

Let us consider the set S = {(g, x) G C | g(x) = x}. Obviously,

|S| =

|Ag | =

gG

From 2 and 3 we have that |A/G| =

|Gx|

xA

1 X g
|A |.
|G|
gG

(3)

140

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

2.4. Generating Function for a Group of Permutations

The following notation will be useful in our study: if a permutation is the product of 1
unitary cycles, 2 cycles with length 2, , n cycles with length n, it will be denoted
= x1 1 x2 2 . . . xnn
and also
(1, 2, , n )
Theorem 2.18. Any permutation = x1 1 x2 2 . . . xnn , satisfies the condition:
1 + 22 + + nn = n
Using the notation than we have just presented we define the generating function for
the group of permutations G as the polynomial with variables x1 , . . . , xn :
ZG (x1 , . . ., xn ) =

x1 1 x2 2 . . . xnn

and the index of cycles as:

IG (x1 , . . ., xn ) =

1
ZG (x1, . . . , xn )
|G|

It is possible than several permutations in the group have the same structure of cycles.
The following theorem allows us to calculate the number of permutations in the group with
the same structure.
Theorem 2.19. Given a group of permutations G SA the number of permutations in the
group with the same structure of cycles than = x1 1 x2 2 xnn is:
n!
k
k=1 k k !

Qn

The following examples show the utility of knowing the structure of cycles of the permutations in a group.
Example 2.20. Let us consider the squares whose vertexes (we begin with the left upper vertex and we continue in the clock sense) are 1, 2, 3 and 4. We also consider
the group of rotations and reflections (rigid movements) acting over the set of vertexes,
G = {0, 1, 2, 3, r1, r2, r3, r4}, as they were denoted in Example 2.6. It is obvious that
every rigid movement is a permutation of the set of vertexes.
The following table shows the decomposition in cycles of the elements of the group of
permutations G.

Algebraic Topics on Discrete Mathematics

1
|
4
4
|
3
3
|
2
2
|
1
4
|
1
2
|
3
3
|
4
1
|
2

2
|
3
1
|
2
4
|
1
3
|
4
3
|
2
1
|
4
2
|
1
4
|
3

0 =

1 =

2 =

3 =

r1 =

r2 =

r3 =

r4 =










1 2 3 4
1 2 3 4
1 2 3 4
2 3 4 1
1 2 3 4
3 4 1 2
1 2 3 4
4 1 2 3
1 2 3 4
4 3 2 1
1 2 3 4
2 1 4 3
1 2 3 4
3 2 1 4
1 2 3 4
1 4 3 2










141

= (1)(2)(3)(4) = x41
= (1234) = x14
= (13)(24) = x22
= (1432) = x14

= (14)(23) = x22

= (12)(34) = x22

= (13)(2)(4) = x21 x12

= (1)(24)(3) = x21 x12

The generating function for the group of permutations G is:

ZG (x1, x2, x3, x4) = x41 + 2x21x2 + 3x22 + 2x14
Now, we can suppose that the vertexes of the square can be colored with k different
colors.
The colorations with all the vertexes in each cycle of a permutation colored whit the
same color will be invariant for such permutation. So, the number of invariant colorations
for the permutation (1, . . . , n) is k1 ++n .
The following table shows the number of invariant coloration for the permutations in
the group G.
Permutation
0
1 , 3
2, r1, r2
r3, r4

Decomposition
x41
x14
x22
x21x12

Invariant colorations
k4
k
k2
k3

By using the Theorem of Burnside(2.17) the number of non equivalent colorations for

142

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

the group of rotations and reflections is:

1
1 X g
|A | = (2k + 3k2 + 2k3 + k4 )
|G|
8
gG

It can be written using the generating function:

1
1 X g
ZG (k, k, k, k)
|A | =
|G|
|G|
gG

The decomposition in cycles also allows us to see which colorations are invariants for
k
X
cai where k is the
a permutation. The inventory of invariant coloration for a cycle xa is
i=1

number of colors and ci are the different colors. Moreover, the inventory of invariant coloration for a permutation is the product of the inventories for the cycles of this permutation.
Let us suppose k = 2, being A and B the colors.
The following table shows the inventory of invariant colorations for the permutations
of the group G.
Permutation
Inventory
4
4
0
(A + B) = A + 4A3B + 6A2B 2 + 4AB 3 + B 4
1, 3
A4 + B 4
2 , r1, r2
(A2 + B 2 )2 = A4 + 2A2 B 2 + B 4
2
2
r3, r4
(A + B) (A + B 2 ) = A4 + 2A3B + 2A2 B 2 + 2AB 3 + B 4
The coefficients of the inventory show as they are the invariant colorations. So the term
cA B b means that there exist c invariants colorations with a vertexes colored in A and b
vertexes colored in B.
Now, let us suppose k = 3, being A, B and C the colors. The following table shows the
inventory of invariant colorations for the permutations in the group G in this case.
a

Permutation
0

3
2 , r1, r2
r3, r4

Inventory
(A + B + C)4 = A4 + B 4 + C 4 +
+4(A3 B + A3 C + AB 3 + B 3 C + AC 3 + BC 3 )+
+6(A2 B 2 + A2 C 2 + B 2 C 2 ) + 12(A2BC + AB 2 C + ABC 2 )
A4 + B 4 + C 4
2
2
2
2
4
(A + B + C ) = A + B 4 + C 4 + 2(A2B 2 + A2 C 2 + B 2 C 2 )
(A + B + C)2 (A2 + B 2 + C 2 ) = A4 + B 4 + C 4 +
+2(A2B 2 + A2 C 2 + B 2 C 2 )+
3
+2(A B + A3 C + AB 3 + B 3 C + AC 3 + BC 3 )+
+2(A2BC + AB 2 C + ABC 2 )

Example 2.21. Let us consider a regular polygon with n vertexes and the action of n
rotations and n reflections. Moreover, we will suppose that the n vertexes of the polygon
can be colored with k different colors.

Algebraic Topics on Discrete Mathematics

143

The rotations that we consider have amplitude 2i

n being i {0, 1, . . ., n 1} in the
3
clock sense. This rotations are named 0 , 1, . . . , n1.
The decomposition in cycles of the rotations and the number of invariant colorations
for them are given by:
Rotation
0
i with 0 < i n2 and i divides n
i with

n
2

Decomposition Invariant colorations

xn1
kn
i
xn
ki
i

< i and (n i) divides n

xni
n

kni

ni

x1n

Another rotations

In the case of reflections we have to distinguish two situations:

1. If n is even then n2 reflections are respect to an axis joining the middle point of an
edge and the middle point of the opposite edge and the other n2 reflections are respect
to an axis joining a vertex and its opposite one.
2. If n is odd, the n reflections are respect to axis joining a vertex and the middle point
of the opposite edge.
The decomposition in cycles of the reflections and the number of invariant colorations
for them are given by:
Reflection
Respect to an axis joining

Decomposition Invariant colorations

n

the middle point of an edge

and the middle point of the
opposite one (n even)
Respect to an axis joining

x22

a vertex and the opposite edge

(n odd)
Respect to an axis joining

x11x2 2

a vertex and the opposite one

(n even)

x21x2 2

k2

n1

n+1
2

n+2
2

n2

Now we are going to calculate the generating function ZG in two cases:

1. Generating function for the group of rotations and reflections that act over a regular
pentagon.
The following table shows the decomposition in cycles of the 10 permutations in the
group. (i is the rotation with amplitude 2i
n in the clock sense, ri are reflections
3

Observe that the rotation with amplitude

in the opposite sense.

2(ni)
n

2i
n

in the clock sense is equivalent to a rotation with amplitude

144

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

respect to an axis joining a vertex and the middle point of the opposite edge)
Permutation
0
1, 2, 3, 4
r1, r2, r3, r4, r5

Decomposition
x51
x15
x11 x22

The generating function is ZG = x51 + 5x11x22 + 4x15.

2. Generating function for the group of rotations and reflections that act over a regular
hexagon.
The following table shows the decomposition in cycles of the 12 permutations in the
group. (i is the rotation with amplitude 2i
n in the clock sense; r1 , r2 and r3 are
reflections respect to an axis joining a vertex and its opposite one; r4, r5 and r6 are
reflections respect to an axis joining the middle point of an edge and the middle point
of the opposite edge)
Permutation
0
1 , 5
2 , 4
3
r1, r2, r3
r4, r5, r6

Decomposition
x61
x16
x23
x32
x21 x22
x32

The generating function is ZG = x61 + 3x21x22 + 4x32 + 2x23 + 2x16 .

Example 2.22. The ideas showed in Examples 2.20 and 2.21 can be extended to regular
solids. As is well known there exist five regular solids (with 4, 6, 8, 12 and 20 sides) in the
space. If we consider the rigid movements transforming a solid into itself then we obtain the
symmetric group for this solid. These movements are permutations of the set of vertexes.
Such permutations can be decomposed in cycles and so we can calculate the generating
function. The following table shows the number of elements (order) of the symmetric group
for the regular solids.
Number of sides Order of the symmetric group
4
24
6
48
8
48
12
120
20
120

2.5. Polyas Enumeration Method

We will consider again the square whose vertexes can be colored with two different colors.
In Example 2.6 we saw the 16 colorations and the orbits. Two colorations are in the same

Algebraic Topics on Discrete Mathematics

145

orbit if one of them can be obtained beginning with the other by means of a rotation or a
reflection. It is obvious that all the colorations in an orbit have, for each color, the same
number of vertexes. For example, all the colorations in the orbit O2 have 1 vertex colored
in a color and 3 vertexes colored in the other color. The reciprocal is not true. For example,
the colorations in both orbits O3 and O4 have 2 vertexes colored in a color and 2 vertexes
colored in the other color. The question is: How many non equivalent colorations (that is,
orbits) there exist with the same number of vertexes for each color? The following result
give us a general procedure to answer to this question.
Theorem 2.23 (Polyas Theorem). Given a group of permutations G Sn , a finite set
K = {c1, . . . , cm} and the set of functions F = {f : {1, . . . , n} K}, we consider the
equivalence relation R in F given by:
f g if and only if G | g = f
Then the number of non equivalent applications such that:
|f 1 (c1)| = k1 , |f 1(c2)| = k2, . . . , |f 1(cm )| = km
is the coefficient of ck11 ck22 . . . ckmm in the polynomial
1
ZG
|G|

m
X
i=1

ci ,

m
X
i=1

c2i , . . . ,

m
X

cni

i=1

Example 2.24. We consider the set of colorations in Example 2.6 and we name A and B
the colors. How many non equivalent colorations there exist with k1 vertexes colored in A
and k2 vertexes colored in B?
Obviously, there are only five possibilities for k1 and k2. The condition is k1 + k2 = 4,
so the cases are: (k1, k2) = (4, 0), (k1, k2) = (3, 0), (k1, k2) = (2, 2), (k1, k2) = (1, 3)
and (k1, k2) = (0, 4).
We use Polyas Theorem with G = {0 , 1, 2, 3, r1, r2, r3, r4}, K = {A, B}. The
number of non equivalent colorations with |f 1 (A)| = k1 and |f 1 (B)| = k2 is the
coefficient of Ak1 B k2 in the polynomial 4
1
[(A + B)4 + 2(A + B)2 (A2 + B 2 ) + 3(A2 + B 2 )2 + 2(A4 + B 4 )]
8
It is easy to check that:
The number of non equivalent coloration with the 4 vertexes colored in A, that is,
the coefficient os A4 is 1. Because of the symmetry, the number of non equivalent
colorations with the 4 vertexes colored in B is 1.
The number of non equivalent colorations with 3 vertexes colored in A and 1 vertex
colored in B, that is, the coefficient of A3 B es 1. Because of the symmetry, the number of non equivalent colorations with 1 vertex colored in A and 3 vertexes colored
in B is 1.
4

3x22

As we saw in Example 2.20, in this case the generating function is ZG (x1 , x2 , x3 , x4 ) = x41 + 2x21 x12 +
+ 2x14

146

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

The number of non equivalent colorations with 2 vertexes colored in A and 2 vertexes
colored in B is 2.
Example 2.25. A curious example of Polyas enumeration method appear in [11] in which
it is possible to determine the number of equivalence classes of musical objects.

3.

Finite Fields and Latin Squares

3.1. Polynomial Rings

A ring (R, +, ) is an algebraic structure consisting of a non-empty set R and two operations
addition + and multiplication which verify:
1. (R, +) is an abelian group with identity element 0.
2. (R, ) is a semigroup.
3. For all a, b, c R it satisfies:
(a) a (b + c) = (a b) + (a c)
(b) (a + b) c = (a c) + (b c)
It is usual to use the support set R to denote the ring (R, +, ) but it is important to
remember that the ring is not only the support set. The symbol is usually omitted and
multiplication is just denoted by juxtaposition a b = ab. Also the priority rules are used,
so that for example, a + bc is an abbreviation for a + (bc).
Rings that satisfy commutativity for multiplication are named commutative rings and
rings were multiplication has identity element are named unitary rings. In this case the
referred identity is expressed by 1.
Example 3.1. As a first example, the trivial ring {0} has only one element which serves
both as additive and multiplicative identity 5.
Example 3.2. (Z, +, ) is the ring of integers with the two usual operations. This is a
commutative and unitary ring. The rational, real and complex numbers form rings 6. These
are likewise commutative and unitary rings.
Example 3.3. A Gaussian integer is a complex number whose real and imaginary part are
both integers. The Gaussian integers, with ordinary addition and multiplication of complex
numbers, form a ring, usually denoted Z[i].
Every ring R satisfies the following properties:
For all a R, a0 = 0a = 0.
For all a, b R, (a)b = a(b) = (ab) and (a)(b) = ab.7
5

Some authors do not consider the trivial ring as a ring.

In fact, they are even fields (see Section 3.2.)
7
This property is usually named the rule of signs.
6

Algebraic Topics on Discrete Mathematics

147

A non-zero element a of a ring R is a left zero divisor if there exists a non-zero b such
that ab = 0. Similarly, right zero divisors are defined. That is, a non-zero element a such
that there exists a non-zero b with ba = 0.
An element that is both a left and a right zero divisor is simply named a zero divisor.
Proposition 3.4. A non-zero element in a ring R is regular or simplifiable for if and only
if it is not a zero divisor.
A ring without zero divisor elements is named integral ring and a unitary ring without zero divisor elements is named integral domain. Therefore, in an integral ring, the
simplification rule holds.
Example 3.5. An example of a zero divisor in the ring of 2 2 matrices is the matrix


1 0
0 0
because for instance

1 0
0 0



0 0
0 1

0 0
0 0

In general, in the ring of nn matrices over some integral domain, the zero divisors are
precisely the nonzero matrices with determinant zero, that is the nonzero singular matrices.
Given a unitary ring (R, +, ), the subset R of the invertible elements is closed for the
multiplication and, therefore, (R, ) is a group. This group is known as the multiplicative
group of R.
Example 3.6. The set A = {0, 1, 2, 3, 4, 5} with the two following operations is an unitary
ring.8 In this ring, 2, 3 and 4 are zero divisor and the multiplicative group is ({1, 5}, ).
+
0
1
2
3
4
5

0
0
1
2
3
4
5

1
1
2
3
4
5
0

2
2
3
4
5
0
1

3
3
4
5
0
1
2

4
4
5
0
1
2
3

5
5
0
1
2
3
4

0
1
2
3
4
5

0
0
0
0
0
0
0

1
0
1
2
3
4
5

2
0
2
4
0
2
4

3
0
3
0
3
0
3

4
0
4
2
0
4
2

5
0
5
4
3
2
1

In a ring R, a subset B is said to be a subring if B with the same operations is also

a ring. Therefore, as a consequence of Theorem 1.10 we obtain the following subrings
characterization.
Theorem 3.7. Let R be a ring. A subset 6= B R is a subring if and only if a b B
and ab B for all a, b B.
Obviously, any non-trivial ring R has two subrings: {0} and R.
8

This ring is known as the ring of residual classes module 6, usually denoted Z6 , and will be formally
introduced in Example 3.11

148

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

Example 3.8. Z, Q and R are subrings of C.

An ideal in a ring R is a subring I R such that ia, ai I for all i I and a R.
Example 3.9. For all m Z with m > 1, mZ = {mz | z Z} is a subring of Z which is
also an ideal.
Example 3.10. The subrings of C given in Example 3.8 are not ideals.
To introduce the ring known as the quotient ring, the following relation is defined: If R
is a ring and I R is an ideal, the congruence relation (module I) is defined as follows:
For all a, b R,
a I b if and only if b a I
It is easy to prove that I is an equivalence relation and the equivalence class of any a R
is
[a] = a + I = {a + i | i I}
In the quotient set R/I , the operations given by [a] + [b] = [a + b] and [a][b] = [ab] define
a ring. Usually R/I denotes this ring.
Example 3.11. Let m N with m > 1. The ring Zm is obtained as quotient ring from the
ideal mZ in the integer ring.
Given two rings (A, +, ) and (B, +, ), a ring homomorphism is an application
f : A B such that f (a + b) = f (a) + f (b) and f (ab) = f (a)f (b), for all a, b A.
Direct consequences of this definition are the following properties:
1. f (0) = 0.
2. For all a A, f (a) = f (a).
3. If A0 is a subring of A then f (A0) is a subring of B.
4. If B 0 is a subring of B then f 1 (B 0 ) is a subring of A.
5. If f is an isomorphism then f 1 is also an isomorphism.
As all ring is a group with the addition, we keep the definition of kernel. That is,
ker(f ) = f 1 (0) = {x A | f (x) = 0}
Proposition 3.12. Let (A, +, ) and (B, +, ) be rings and f : A B be an homomorphism.
1. f (A) is a subring of B.
2. ker f is an ideal of A.
Theorem 3.13. Let (A, +, ) and (B, +, ) be rings and f : A B be an homomorphism.
The rings A/ ker f and f (A) are isomorphic.

Algebraic Topics on Discrete Mathematics

149

Definition 3.14. A polynomial in a ring (R, +, ) is a formal expression:

m

p(x) = a0 + a1 x + + am x

m
X

ai xi

i=0

where the coefficients a0, a1, . . . , am are elements of R, m is a natural number and, if
m > 0, am 6= 0. We say that m is the degree of this polynomial and write (p(x)) = m.
Two polynomials are considered to be equal if the coefficients of each power of x are
equal.
Hereinafter, if no confusion arises we will write p to denote a polynomial p(x).
Polynomials in R can be added by simply adding the corresponding coefficients and
multiplied using the distributive law and the rules xa = ax, for all element a R, and
xk xl = xk+l for all natural numbers k and l.
Every polynomial p = a0 + a1 x + + am xm in a ring R define a function (not
necessarily homomorphism) f : A A with f (a) = a0 +a1 a+ +am am . The elements
that belong to f 1 (0) are named roots (or zeros) of p(x). That is, r R is a root of p if
a0 + a1 r + am rm = 0.
It is easy to prove that the set of polynomials with coefficients in a ring R with the
addition and the multiplication is also a ring. This ring is named the polynomial ring over
R and is denoted by R[x]. Formally, the addition operation is defined as follows:

max{m,n}
m
n
ai + bi , 0 i min{m, n};
X
X
X
ai xi +
bi xi =
ci xi where ci =
, m < i n;
bi

i=0
i=0
i=0
, n < i m.
ai
and the multiplication operation is defined as follows:
m
X
i=0

ai xi

n
X
i=0

bixi =

m+n
X

i=0

r+s=i

ar bs

xi

Proposition 3.15. Let R be a ring.

1. If R is commutative then R[x] is commutative.
2. If R is a unitary ring then R[x] is a unitary ring.
3. R[x] is an integral domain if and only if R is an integral domain.
Example 3.16. The polynomial ring Z6 [x] is commutative and unitary. However, it is not
a integral domain because, for example, 3x + 3 and 2x + 2 are zero divisors.
Note that both 3x + 3 and 2x + 2 are polynomials with degree 1 but, surprising, the
degree of its product is 0.
Theorem 3.17. Let R be a commutative unitary ring. It is an integral domain if and only
if, for all p, q R[x] {0},
(pq) = (p) + (q)

150

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

3.2. Finite Fields and Polynomials

A field is a non-trivial commutative unitary ring (K, +, ) in which every non-zero element
a K has inverse element. That is, (K, +) and (K {0}, )9 are abelian groups and
distributes over +. The following theorem is well-known.
Theorem 3.18.
1. Any field is an integral domain.
2. Any finite integral domain is a field.
Example 3.19. It is well-known that (Zp, +, ) is a field if an only if p is prime.
In this section we will define new fields builded in analogous mode as the Zp fields
are designed. That is, we will define the divisibility relation in a polynomial ring, the
congruence relations and the quotient ring.
From now on, we will study polynomial rings K[x] in which K is a field and therefore
K[x] is an integral domain. However, K[x] is not necessarily a field. 10 This allows us
introduce the concept of divisor. A polynomial p1 K[x] {0} is said to be divisor of
another polynomial p2 K[x] if there exists q K[x] {0} such that p1 q = p2 .
Theorem 3.20 (of division). Let K be a field and p1 , p2 K[x] with s 6= 0. There exist
unique q, r K(x) such that p2 = p1 q + r and (r) < (p1).
In this case, q is the quotient and r is named rest of the division.
Corollary 3.21. Let K be a field, a K and p K[x]. The polynomial x a is divisor of
p if and only if a is a root of p.
These results and Theorem 3.17 yield to the following theorem.
Theorem 3.22. Let K be a field and p K[x]. The number of roots of p is lower or equal
to the degree of p.
Remark 3.23. Note that the previous result is not true in every polynomial ring. For example, in Z6 [x], the polynomial 2 + 3x + x2 has four roots: 1, 2, 4 and 5.
A polynomial p K[x] is said to be prime (or unreductable) if every divisor of p is
itself or it has degree 0. In other case, we say that p is reductable.
As a direct consequence of this definition we have the following proposition.
Proposition 3.24. Let K be a field.
1. Any polynomial with degree 1 is prime.
2. A polynomial with degree 2 or 3 is reductable if and only if it has a root.
Example 3.25. The polynomial 1 + x2 is prime in R[x] but it is reductable in C[x].
9
10

Hence, K = K {0}.
For example, R[x] is not a field.

Algebraic Topics on Discrete Mathematics

151

Example 3.26. In Z2 [x],

1 + x + x3 is prime because neither 0 nor 1 are roots.
1 + x + x2 + x3 is reductable because 1 is a root. In fact, 1 + x + x2 + x3 =
(1 + x)(1 + x2 ).
Similarly to examples 3.9 and 3.11, given p K[x], the subset
pK[x] = {pq | q K[x]}
is an ideal of K[x] and, therefore, K[x]/pK[x] is also a ring. K[x]/p denote this quotient
ring.
Theorem 3.27. Let K be a field and p a prime polynomial. Then K[x]/p is a field. Moreover, if K is finite, then
|K[x]/p| = |K|(p)
Example 3.28. In R[x], 1 + x is a prime polynomial and, therefore, R[x]/(1 + x) is a
field. By Theorem 3.20, for all polynomial p there exist polynomials q and r such that
p = (1 + x)q + r and (r) < (1 + x) = 1. So,
R[x]/(1 + x) = {[a] | a R}
In consequence, R[x]/(1 + x) is isomorphic to R.
From now on, we omit the symbols [ and ] in the notation of equivalence classes.
Example 3.29. The prime polynomial 1 + x + x2 in Z2 give us the following finite field.
There exist 4 equivalence classes, because the division between 1 + x + x2 can give 4
possible rests: 0, 1, x and 1 + x. Operations in this field are shown in the following tables:
0
1
x
1+x
+
0
0
1
x
1+x
1
1
0
1+x
x
x
x
1+x
0
1
1+x 1+x
x
1
0

0
1
x
1+x

0
1
x
1+x
0
0
0
0
0
1
x
1+x
0
x
1+x
1
0 1+x
1
x

The following theorem characterize every finite field.

Theorem 3.30. For all finite field K there exist a prime number p Z+ and a prime
polynomial q Zp [x] such that Zp [x]/q is isomorphic to K.
As a consequence of the theorems 3.27 and 3.30, any finite field has pn elements, where
p is a prime number and n is a positive integer. Any two finite fields with the same number
of elements are isomorphic. Hence, we can say that there is an unique field of pn elements
named Galoiss fields of order pn . F(pn) denotes this field. 11
11

The previous theorems are consequence of the work of the French mathematician Evariste Galois (1811
1832) about the non-existence of formulae to solve general polynomials equations with degree greater to 4.

152

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

Figure 1. Drers Melancolia I and magic square under the bell.

3.3. Latin Squares

The adjective Latin comes from Leonard Euler (17071783), who used Latin characters as symbols in opposition to Greek squares in which Greek characters are used. Euler
used both kinds of squares to define Graeco-Latin squares used to construct magic squares.
Those squares are arrangements of n2 numbers in a square, such that all rows, all columns,
and both diagonals sum the same constant named magic number. More information can
be seen in [8].
Figure 1 shows one of the most famous copperplates entitled Melancolia I" made by
the painter and mathematician Albert Drer. In the right upper part of the plate, under the
bell, we can see the following magic square that could be one of the first magic square seen
by Europeans. The magic number 34 can be obtained in many different ways.
16 3 2 13
5 10 11 8
9 6 7 12
4 15 14 1
Definition 3.31. A Latin square is an n n array filled with n different symbols, usually
1, 2, . . .n, in such a way that each symbol occurs exactly once in each row and exactly once
in each column.
Here we present two examples:
1
4
2
3

2
3
1
4

3
2
4
1

4
1
3
2

a
b
d
c

b
c
a
d

d
a
c
b

c
d
b
a

Algebraic Topics on Discrete Mathematics

1 2 3
3 1 2
2 3 1

1 3 2
2 1 3
3 2 1

153

1 2 3
2 3 1
3 1 2

Figure 2. Conjugated of example 3.33

Example 3.32. For every n 2 the table of addition for Zn is a Latin square. Observe
that these tables fit to definition 3.31 when we replace all 0 by n.
A Latin square is said to be normalized if its first row is in natural order. For example,
the first Latin square above is normalized because its first row is 1, 2, 3, 4. We can normalize
any Latin square permuting the name of the symbols conveniently. For example, the second
Latin square showed above is not normalized, but permuting c and d we obtain the next
normalized Latin square:
a b c d
b d a c
c a d b
d c b a
In general, if we permute the rows, permute the columns, and permute the names of
the symbols of a Latin square, we obtain a new Latin square said to be isotopic to the first.
Isotopism is an equivalence relation, so the set of all Latin squares is divided into subsets,
named isotopy classes.
3.3.1. Orthogonal Array Representation
Every entry of a Latin square can be written as a triplet (r, c, s), where r is the row, c is
the column, and s is the symbol. Then we obtain a set of n2 triplets named the orthogonal
array representation of the square. For example, the orthogonal array representation of the
first Latin square displayed above is
T = {(1, 1, 1), (1, 2, 2), (1, 3, 3), (1, 4, 4), (2, 1, 4), (2, 2, 3), (2, 3, 2), (2, 4, 1),
(3, 1, 2), (3, 2, 1), (3, 3, 4), (3, 4, 3), (4, 1, 3), (4, 2, 4), (4, 3, 1), (4, 4, 2)}
The definition of Latin square can be written in terms of orthogonal arrays as n2 triplets
of the form (r, c, s), where 1 r, c, s n and all of the possible pairs (r, c), (r, s) and
(c, s) are different.
The orthogonal array representation shows that rows, columns and symbols play similar
roles, as will be made clear below.
If we systematically reorder the three items in each triplet, another orthogonal array
(and, thus, another Latin square) is obtained. For example, we can replace each triplet
(r, c, s) by (c, r, s) which corresponds to the transposed square (reflecting about its main
diagonal), or we could replace each triplet (r, c, s) by (c, s, r), which is a more complicated
operation. Altogether there are 6 possibilities including the identity, giving us six Latin
squares (not necessary different) named the conjugates of the original square.

154

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

Example 3.33. The normalized Latin square defined by the orthogonal array
{(1, 1, 1), (1, 2, 2), (1, 3, 3), (2, 1, 3), (2, 2, 1), (2, 3, 2), (3, 1, 2), (3, 2, 3), (3, 3, 1)}
defines a conjugates class with 3 elements that appear in figure 2
Finally, we can combine these two equivalence operations: two Latin squares are said to
be paratopic if one of them is isotopic to a conjugate of the other. This is again an equivalence relation, with the equivalence classes named species, or paratopy classes. Each
specie contains up to 6 isotopy classes.
Numbers of Latin squares. In [20] we can see the number of Latin squares (or rectangles) depending on the numbers of rows (or rows and columns). Computations have been
made until 10 10 squares and estimated for squares with n = 11, 12, 13, 14, 15.
Example 3.34. Complete Sudoku puzzles are 9 9 Latin squares, with an additional constraint: they are divided in nine 3 sub-squares and each one of them contains the nine symbols. The name Sudoku is the Japanese abbreviation of Suji wa dokushin ni kagiru,
meaning the digits must occur only once

Figure 3. An incomplete Sudoku puzzle.

3.3.2. Orthogonal Latin Squares

In 1782 Euler (see [9]) proposed the following question known as problem of the 36 officers:
How many arrangements of 36 officers of six different ranks and regiments can
be made in a 6 6 square such that each of six ranks and each of six regiments
are represented once in each row and column?
Euler conjectured that there was no solution, and introduced mutually orthogonal Latin
squares to decide the conjecture. G. Tarry proved it with an exhaustive search of all Latin
squares of order 6 in 1901, [26]. In 1984 a more elegant proof was given by Stinson in [25].
Orthogonal Latin squares solves some problems of optimization.
Example 3.35. Design a way to distribute the tests so that each one of three drivers proves
three different vehicles in the smaller time. Superposing the two following tables, we are
able to solve the problem in three turns.

Algebraic Topics on Discrete Mathematics

Drivers
1 2 3
2 3 1
3 1 2

Vehicles
1 2 3
3 1 2
2 3 1

155

Schedule
Turn 1: (1, 1) (2, 2) (3, 3)
Turn 2: (2, 3) (3, 1) (1, 2)
Turn 3: (3, 2) (1, 3) (2, 1)

Definition 3.36. Two n n Latin squares, T1 = (aij ) and T2 = (bij ), are orthogonal if
and only if the n2 ordered pairs (aij , bij ) are all different. We say that T1, T2, . . . , Tk are
mutually orthogonal if and only if any two squares Tr , Ts are orthogonal.
It is easy to prove that the orthogonality of T1 and T2 is equivalent to the orthogonality of
the normalized squares T1 and T2 . Therefore, from now on, we will talk about normalized
Latin squares.
We saw above the confirmed Eulers conjecture about there not exist mutually orthogonal 66 Latin square. Euler also conjectured that this problem could not be solved for nn
Latin squares when n 2 (mod 4). But, nevertheless, in 1960 works of Bose, Shrikhande
and Parker proved that this conjecture is false (see [23], [3]), and interest shifted to finding
how many pairs of mutually orthogonal Latin squares there are for a given n.
Theorem 3.37. For n Z+ , n > 2, the number of nn Latin squares mutually orthogonal
is at most n 1.
Proof. Let be T1, T2, . . ., Tm mutually orthogonal and normalized Latin squares. We represent Tk = (akij ). So, ak1j = j for k = 1, 2, . . .m. Now, ak21 6= 1 and different for all
k = 1, . . . , m, since, on the contrary, they would not be orthogonal. This proves than m is
smaller than n.
In general, it is difficult to construct orthogonal Latin squares. But when the order is a
power of prime number, finite fields give us a method to obtain the maximum number of
mutually orthogonal Latin squares.
Theorem 3.38. Let be n > 2 and n = pm , where p is a prime number. Then there are n 1
mutually orthogonal Latin squares.
Proof. Consider the only finite field F(n) = {c1, c2, . . . , cn } (Galoiss field), where c1 = 1
and cn = 0. For k = 1, 2, . . . , n 1, we define the square Tk = (akij ) being akij = ck ci + cj .
Then, we can prove:
1. Each Tk is a Latin square. All elements in the same row are different.
akir = akis ck ci + cr = ck ci + cs cr = cs r = s
All elements in the same column are different.
akrj = aksj ck cr + cj = ck cs + cj ck cr = ck cs
and, as ck has inverse, then cr = cs r = s.

156

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

0

2. When k 6= k0 , then Tk , Tk0 are mutually orthogonal. Suppose (akij , akij ) = (akrs , akrs)
then
ck ci + cj = ck cr + cs
ck0 ci + cj = ck0 cr + cs
and reducing, (ck ck0 )ci = (ck ck0 )cr . Hence, as ck 6= ck0 , we have ci = cr and
hence i = r. Using this with any of the previous equations, we obtain that cj = cs
and j = s.
According to the previous results we can normalize these Latin squares.
Example 3.39. For n = 5, using the Galoiss field F(5) = Z5 = {1, 2, 3, 4, 0}, and the
above theorem, we obtain the following four squares:
2
3
4
0
1

3
4
0
1
2

4
0
1
2
3

0
1
2
3
4

1
2
3
4
0

3
0
2
4
1

4
1
3
0
2

0
2
4
1
3

1
3
0
2
4

2
4
1
3
0

4
2
0
3
1

0
3
1
4
2

1
4
2
0
3

2
0
3
1
4

3
1
4
2
0

0
4
3
2
1

1
0
4
3
2

2
1
0
4
3

3
2
1
0
4

4
3
2
1
0

3
4
5
1
2

4
5
1
2
3

5
1
2
3
4

1
3
5
2
4

2
4
1
3
5

3
5
2
4
1

4
1
3
5
2

5
2
4
1
3

1
4
2
5
3

2
5
3
1
4

3
1
4
2
5

4
2
5
3
1

5
3
1
4
2

1
5
4
3
2

2
1
5
4
3

3
2
1
5
4

4
3
2
1
5

5
4
3
2
1

normalizing,
1
2
3
4
5

2
3
4
5
1

Example 3.40. With the Galoiss field F(4) = {1, x, x + 1, 0}, we obtain three mutually
orthogonal squares. The first of them is:
0
x+1
x
1
x+1
0
1
x
x
1
0
x+1
1
x
x+1
0
To construct a pair of orthogonal Latin squares of any odd order, we can use the group
Z2m+1 . Let T1 = (i + j) be the addition table for the integers modulo n = 2m + 1 and
let T2 = (2i + j), entries taken modulo n. But the number of mutually orthogonal Latin
squares can be less than n 1.
Magic squares. We are going to finish by applying some kind of orthogonal Latin
squares, those n n squares which contain all numbers from 1 to n (or equivalently, 0
to n 1) in such all rows, columns, and both diagonals. Those squares are also magic, and
by juxtaposing produce new magic squares.

Algebraic Topics on Discrete Mathematics

157

Example 3.41. The other two Latin (and magic) squares that are let to calculate in the
previous example 3.40 are
x+1
0
1
x
x
1
0
x+1
0
x+1
x
1
1
x
x+1
0

and

x
1
0
x+1
0
x+1
x
1
x+1
0
1
x
1
x
x+1
0

Replacing x by 2 and juxtaposing we obtain a square formed by pairs of numbers in

{0, 1, 2, 3}. Converted to base 10, we have
14 1 4 11
8 7 2 13
3 12 9 6
5 10 15 0
Obviously, other substitutions give other magic squares.

4.

Finite Geometry and Blocks Design

A finite geometry is any geometric system that has only a finite number of points. These
geometries were introduced by Gino Fano ([10]) at final XIX century. The usual Euclidean
geometry is not finite, because any line contains infinitely many points, in fact the same
number of points as there are in the real line R. A finite geometry can be any finite dimension, but for ours proposes we will describe only classic 2-dimensional (or plane) finite
geometries.
In this section, mainly we present two kinds of finite plane geometry: affine and projective. In an affine geometry, the normal sense of parallel lines applies. In a projective plane,
by contrast, any two lines intersect at a unique point, and so parallel lines do not exist. Both
geometries may be described by fairly simple axioms.

4.1. Finite Affine Plane

Definition 4.1. We name affine plane to a pair (P, R), where P is a nonempty set whose
elements are named points, and R is a nonempty collection of subsets of P whose elements
are named lines, such that:
AP1 Given any two distinct points, there is exactly one line that contains both points.
AP2 Given a line r and a point P , there exists exactly one line r0 containing P such that
r = r0 or r r0 = (Euclids fifth postulate).
AP3 There exists a set of four points, no three of which belong to the same line.
The last axiom ensures that the geometry is not trivial, while the first two specify the
nature of the geometry.
The first question is the possibility to build affine plane geometries with any (finite)
number of elements. Axiom AP3 indicates that is necessary at least four elements.

158

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

P1

P2

P4

P3

Figure 4. The affine plane of order 2.

Example 4.2. The simplest affine plane contains only four points; it is named the affine
plane of order 2. Since no three points
are collinear, any pair of points determines a unique
4

line, and so this plane contains 2 = 6 lines.
Above, in Figure 7, we can see a graphical representation 12.
By means of finite fields we can build affine planes. For n = pm , where p is a prime
number and m is a positive integer, we consider the Galoiss field F = F(n), and define an
affine plane AP (n) = (P, R), were the set of points is P = {(x, y) | x, y F} and the set
of lines R is defined in this way:
1. For each a F, vertical lines ra = {(a, y) | y F}, which can be represented by
equation x = a.
2. For each a, b F, lines ra,b = {(x, ax + b) | x F}, which can be represented by
equation y = ax + b.
Therefore, AP (n) contains n2 points and n2 + n lines.
Lemma 4.3. Each line r contains n points and each point is contained in n + 1 lines.
Proof. Each line contains n points is evident. Now, let (x0 , y0) be a point. This point is in
the line rx0 and also in the n lines ra,(y0ax0 ) por each a F.
Theorem 4.4. If F(n) is a field with n = pm elements, then AP (n) is an affine plane
(named affine plane of order n) with n2 points and n2 + n lines.
Proof. AP (n) verifies the three axioms:
AP1 Let P0 = (x0, y0 ), P1 = (x1, y1 ) be two different points. There are three possibilities:
1. x0 = x1 , then P0 , P1 rx0 .
2. y0 = y1 , then P0 , P1 r0,y0 .
3. x0 6= x1 and y0 6= y1 , then P0 and P1 are contained in the line defined by
equation (x x0)(x1 x0 )1 = (y y0 )(y1 y0 )1
12

Observe that the affine plane of order 2 could be seen like the complete graph K4 .

Algebraic Topics on Discrete Mathematics

159

The line that contains P0 , P1 is the unique one. Suppose P0 , P1 ra,b and P0 , P1
rc,d (other situations similar),
=
ax0 + b
=
cx0 + d
y0
=
ax1 + b
=
cx1 + d
y1
y1 y0 = a(x1 x0 ) = c(x1 x0)
Therefore a = c and hence b = d.
AP2 Lemma 4.3 stays that there are n points P1 , P2, . . . , Pn in r. If P is not in r, we
consider n lines r1, r2, . . . , rn determinate by P and Pi . But lemma 4.3 says also
that P is in n + 1 lines, then exists a line r0 different all ri and therefore r0 r = .
AP3 The four points (0, 0), (0, 1), (1, 0) and (1, 1) verify whatever three are not collinear.
That proves the theorem.
Example 4.5. Affine plane of order 2 defined by F(2) = Z2 = {0, 1} has 4 points P1 =
(0, 0), P2 = (0, 1), P1 = (1, 0) and P1 = (1, 1) and 6 lines. Figure 7 in Example 4.2
represents this plane.
We say that lines r and r0 are parallel if either r = r0 or else r r0 = . The relation
be parallel to is an equivalence relation, this takes us the following definition.
Definition 4.6 (Parallel classes). Given AP (n) of order n we can define a partition in R
of n + 1 classes which contain mutually parallel lines.
Every class is represented by any line in the class, for example, vertical lines class is
[r0].
Example 4.7. Following figure represents all parallel classes in the affine plane of order 3.
2

[r0]

[r0,0]

[r1,0]

[r2,0]

Figure 5. Parallel classes for AP (3).

4.2. Finite Projective Plane

In the beginning of this section we saw that affine geometry differs from projective geometry because there not exist parallel lines. Every pair of different lines intersect at an unique
point. A projective plane geometry can be define using following axioms.

160

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

Definition 4.8. We name projective plane to a pair (P 0 , R0), where P 0 is a nonempty set
whose elements are named points, and R0 is a nonempty collection of subsets of P 0 whose
elements are named lines, such that:
PP1 Given any two distinct points, there is exactly one line that contains both points.
PP2 The intersection of any two distinct lines contains exactly one point.
PP3 There exists a set of four points, no three of which belong to the same line.
The natural way to build a projective plane is through affine plane. We considerer
AP (n) = (P, R) and redefine this points in the projective plane like P F(n) F(n)
{0, 1} in this way:
P = {(x, y, 1) | x F, y F}
(4)
Moreover, every parallel class defines a new point at infinity in projective plane in
this way:
The vertical lines class [r0] defines the point (0, 1, 0).
The parallel class [ra,0] defines the point (1, a, 0)
Then the n2 points in P according to (4) and the n + 1 points at infinitydefine the
n2 + n + 1 points in projective 13 plane.
P 0 = P {(0, 1, 0)} {(1, a, 0) | a F}
The lines in R together the line at infinityr form the projective lines in R0 , in other
words, R0 is formed by following lines:
Line at infinity r = {(x, y, 0) P 0 }, also represented by z = 0.
Vertical lines ra = {(az, y, z) P 0 }, also represented by x = az.
The rest of lines ra,b = {(x, ax + bz, z) P 0 }, also represented by y = ax + bz.
Theorem 4.9. If F(n) is a field with n = pq elements, then there exists a projective plane
(named projective plane of order n) with n2 + n + 1 points and n2 + n + 1 lines.
Example 4.10. The smallest projective plane (of order 2), also known as Fanos plane, has
7 points and lines.
Remark 4.11. The inverse process is available. If we have a projective plane and we
delete a line and all the points incident with that line then the remaining structure is an
affine plane. The deleted line is obvious the line at infinity.
Therefore, it is well-established that both affine and projective planes of order n exist
when n is a prime power.
It is conjectured that no finite planes exist with orders that are not prime powers, although this statement has not been proved. The best result to date is the celebrated BruckRyser theorem ([4]), which was published in 1949, which states:
13

Observe that the point (0, 0, 0) never is in a projective plane.

Algebraic Topics on Discrete Mathematics

161

001

011

101
111

100

110

010

Figure 6. Diagram of the projective plane of order 2.

Theorem 4.12 (Bruck-Ryser). If n is a positive integer n 1 or 2 (mod 4) and n is not
equal to the sum of two integer squares, then n does not occur as the order of a finite plane.
The smallest integer that is not a prime power and not covered by the Bruck-Ryser
theorem is 10; it is of the form 4k + 2, but it is equal to the sum of squares 12 + 32. Using
sophisticated techniques and computer analysis, it has been shown that 10 is not the order of
a finite plane, [17]. The next smallest number to consider is 12, for which neither a positive
nor a negative result has been proved.

4.3. Incomplete Block Designs

Originally, statistics deals with variability and the planning of environmental or ecological
investigations, the items used to control variability are grouping (blocking) into homogeneous subgroups (blocks) for measurement of related variables and use of covariance.
There are many ways of blocking the experimental units in a comparative experiment
with v varieties (also named treatments). If the sample of experimental units is from a
homogeneous population, then no blocking is required and a completely randomized experiment design of the v varieties randomly allotted to the rv experimental units where the
replicate number (sample size) is r.
If homogeneous blocks of size v are available to accommodate all v varieties, a randomized complete block design (all v varieties in each block, not necessarily an equal number
of times) is used. In complete block design, every treatment is allocated to every block. In
other words, every combination of treatments and conditions (blocks) is tested.
For example, an agricultural experiment is aimed at finding the effect of 3 fertilizers
(A,B,C) for 5 types of soil (1...5). There are 15 plots at the disposal of the researcher - 3
plots for each type of soil. Here fertilizers (A,B,C) are treatments, types of soil (1...5)
are blocks; the design is the chosen distribution of the 3 fertilizers over the 15 available
plots. In a complete block design, every possible combination of fertilizer and soil-type is
used at least once.
In many biological and industrial experiments the number of experimental units and
their groupings into blocks are such that the experimenter cannot influence them without
sacrifice of experimental purpose or material. In this cases, the block size is constant and
exceeds the number of treatments; the cost of experimental units is high and this cost is
principally associated with blocks as a whole; and a balanced design is required. That is,
in many situations, the block size, k, is less than v and an incomplete block design (not

162

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

all varieties in each of the blocks) is used. Yates [28] first described incomplete block
experiment designs.
Example 4.13. A researcher is carrying out a study of the effectiveness of four different
skin creams for the treatment of a certain skin disease. He has eighty subjects and plans
to divide them into 4 treatment groups of twenty subjects each. Using a randomized blocks
design, the subjects are assessed and put in blocks of four according to how severe their
skin condition is; the four most severe cases are the first block, the next four most severe
cases are the second block, and so on to the twentieth block. The four members of each
block are then randomly assigned, one to each of the four treatment groups.
Definition 4.14. Let V be a set with v elements (varieties). A finite collection = {Bi }bi=1
of nonempty subsets in V is a incomplete block design, or (v, b, r, k)-design when it verifies:
1. Every Bi contains a fixed-number k of varieties, k < v.
2. Every variety x V replies r( b) times, in other words, x is contained in exactly r
different blocks Bi .
It is easy to see than vr = bk.14 If V = {vj }vj=1 , we can define the b v incidence
matrix N = (nij ) defined:

1 if vj Bi
nij =
0 other case.
Finally, it is usual to define the vv symmetric matrix = (x,y ), named concurrence
matrix, being x,y the number of blocks where the varieties x, y appear together. This
matrix balance the block design. Obvious x,x = r and, when x 6= y and x,y = is a
constant, then we say that the design is balanced.
Example 4.15. In the twentieth century, the most frequent treatment used in the strawberry
growing is the Methyl Bromide. It is an agricultural fumigant widely used because it is
a powerful agent for the control of pathogens and bad herbs. Moreover, it increases the
productivity because it reduces the effect known as stress of the land. However, it is considered as one of the causes of the hole in the ozone layer. On November of 2002, in the IV
Meeting of Montreal Protocol (Copenhagen) its progressive withdrawal was agreed, until
its total disuse in the developed countries.
One research team is studding a new treatment to replace the Methyl Bromide. Specifically, they are interested in testing its effect in 6 varieties of strawberry:: Camarosa, Andana, Chandler, Cartuno, Sophie and Tudla. For this experiment, the research team dispose
of 3 fields with 4 plots.
They use the small design of v = 6, b = 3 and k = 4 defined by the incidence matrix

1 1 1 1 0 0
N = 1 1 0 0 1 1
0 0 1 1 1 1
14

The trick for remembering this formula is using letters of the words varieties and blocks, so
varietiesblocks.

Algebraic Topics on Discrete Mathematics

163

is a (6, 3, 2, 4)-design not balanced. The concurrence matrix is

2 2 1 1 1 1
2 2 1 1 1 1

1 1 2 2 1 1

=
1 1 2 2 1 1

1 1 1 1 2 2
1 1 1 1 2 2

Definition 4.16. Let V be a set with v elements (varieties). A finite collection = {Bi }bi=1
of nonempty subsets in V is a balanced incomplete block design, or (v, b, r, k, )-design
when it verifies:
1. is a (v, b, r, k)-design.
2. Every pair x, y of varieties appears together in exactly ( b) blocks Bi .
Then the five parameters have the following relations
Theorem 4.17. Every balanced incomplete block design (v, b, r, k, )-design verifies:
1. vr = bk.
2. (v 1) = r(k 1).


Proof. 1. was commented above. To proof 2. we count the number of ones in the v2 b
matrix A = (a{x,y},i), {x, y} V , 1 i b defined

1 if {x, y} Bi
a{x,y},i =
0 other case.


Each row contains ones, then the number of ones in A is v2 . Each column contains




k
k
2 ones, then the total number of ones is b 2 . Therefore

k(k 1)
k1
v(v 1)
=b
= vr
2
2
2

This proves the theorem.

Example 4.18. When n = pm (power of prime number) then
1. The affine plane AP (n) defines a (n2 , n2 + n, n + 1, n, 1)-design. The points are the
varieties and the lines are the blocks.
2. Also the projective plane is a balanced incomplete block design. In this case there
are the same number of varieties than blocks. In other words the projective plane is
a (n2 + n + 1, n2 + n + 1, n + 1, n + 1, 1)-design.

5.

Applications

In this section we introduce some applications about different results of discrete mathematic
displayed in the previous sections.

164

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

P1

P2

P4

P3

Figure 7. The affine plane of order 2.

5.1. Discrete Modeling in Foundation Design

Developed by Gloria Gutirrez Barranco, Javier Martnez del Castillo, Salvador
Merino Crdoba, Francisco J. Rodrguez Snchez from Department of Applied Mathematics. University of Mlaga (Spain).
The study that we present arises as an application of Discrete Mathematics to the resolution of problems of circular arrangement. In particular, the original problem was the
following one: It is wanted to construct an electrical line of transmission supported by
metallic towers with their four legs anchored to the ground. The legs can be manufactured with different heights. The problem was the determination of how many foundation
drawings are necessary to lay the tower foundations according to the circumstances in the
orography.

Figure 8. Positions diagram for tower foundation.

Algebraic Topics on Discrete Mathematics

165

5.1.1. General Definitions

We begin with some concepts that we will use in the rest of the paper.
We call chain to any sequence of elements, repeated or not, placed in successive
positions.
We call length of a chain to the number of positions that it has.
We say that a chain is closed when it is considered that the successor of the last
position is the first one.
We call orbit of n elements taken from k in k to the set of circularly indistinguishable
closed chains of length k generated by elements of n different types. The cardinal of
the set of such orbits is denoted by Onk .
The problem previously exposed for the electrical transmission towers would be the
calculation of On4 . We are going to solve it using algebraic procedures.
5.1.2. Symmetric Group Solution
If we have elements of n different types we can generate nk chains of length k. We will
consider them as closed chains. The set of such chains is named C and we represent by SC
the symmetric group defined by C, this is
SC = { : C C | is one-to-one }
We can emphasize that any permutation in SC defines the set of the invariant elements
for such permutation
Inv() = {x C | (x) = x}
that is a subgroup of C.
Any group G that acts on the set C defines an equivalence relation:
x RG y exists G such that (x) = y
Thus, we can construct the quotient set C/RG
The resolution of the original problem, that is to say, the calculation of the number of
orbits, is reduced to calculate el cardinal of this quotient set, where G is the group of the
rotations that act on the nk chains in C.
The well-known Burnsides lemma establishs that the number of equivalence classes in
C/RG is given by:
1 X
|Inv()|
(5)
|G|
G

2i
radians.
k
We denote hgii the subset of G generate by the rotation gi and we denote ind(gi) the
index of such element, that is,
k
ind(gi) =
|hgii|
Concretely, G = {g0, g1, . . . , gk1}, where gi means the rotation of

166

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

as it is known ind(gi ) = gcd(k, i)

Then |Inv(gi)| = nind(gi ) , since if a chain is invariant for gi then it has ind(gi) free"
elements (these elements can be anyone of the n possible types) and the others are forced"
ones.
1
#

#+

$!

"+

*"

"

#!

$"

"

%"

%#
+!

'%

)+

*!

&#

&$

% !

)!

&"
)

& '% ($ )# *" +!

&

'$

&(

'#

(+

' +

'

(*

%(

($

'*

&*
(

')

%)

&)

G2

G3

G4

G6

Figure 9. Invariant conditions for k = 12

Using the lemma of Burnside (5) we can calculate the number of orbits and solve the
original problem.
Onk =

1 X ind(gi )
n
k
gi G

Example 5.1. Using combinatory methods we have:

On4

n
1

+4

n
2

+9

n
3

+6

n
4

(6)

By means algebraic methods, to calculate the number of different orbits for k = 4, we

have
1
On4 = (n4 + n + n2 + n)
4
that is the same value obtained using the formula (6).

5.2. Enumeration of RNA Graphs Using Graph Theory

Developed by Hin Hark Gan, Samuela Pasquali and Tamar Schlick from Howard Hughes
Medical Institute and Department of Physics. University of New York (USA).
Understanding the structural repertoire of RNA is crucial for RNA genomics research.
Yet current methods for finding novel RNAs are limited to small or known RNA families.
To expand known RNA structural motifs, we develop a two-dimensional graphical representation approach for describing and estimating the size of RNAs secondary structural
repertoire, including naturally occurring and other possible RNA motifs. These estimation
are based in tree enumeration theorems of Cayley for labeled trees and Harary-Prins for
unlabeled trees (applications of Plya method)

Algebraic Topics on Discrete Mathematics

167

Figure 10. (Left) Dual graph representations of existing RNAs. (Right) Comparing dual
(column F) and digraph (G) representations of hypothetical RNA secondary structures (E).
5.2.1. Graphical Representation of RNA Structures
Unlike graphs for chemical structures, where atoms are vertices and bonds are edges, our
RNA graphs are RNA secondary topologies where a vertex or an edge can represent multiple nucleotide bases or base pairs, which are themselves composed of multiple atoms and
bonds. To allow graphical representation of complex RNA secondary topologies, we state
below rules for defining RNA graphs and provide justifications for these rules. The rules
specify how to represent RNA loops, bulges, junctions and stems as vertices or edges in a
graph. Essentially, the tree and dual graph rules simplify RNA secondary motifs to allow
their representation as mathematical graphs; the "RNA graphs" specify the skeletal connectivity of the secondary motifs.
We use tree graphs to represent RNA trees and dual graphs to represent any RNA secondary structures, including trees and pseudoknots, since pseudoknots cannot be represented as trees. Still, the tree representation is advantageous because of its intuitive appeal
and the existence of applicable tree enumeration theorems, especially those by Cayley and
by Harary and Prins.
5.2.2. Planar Tree Graph Rules
To represent RNA trees as planar graphs, we use the following rules to assign edges and
vertices.
T1. A nucleotide bulge, hairpin loop or internal loop is considered a vertex when there
is more than one unmatched nucleotide or non-complementary base pair. The special case
of the GU wobble base pair is regarded as a complementary base pair.

168

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

Figure 11. Schematic graphical representations for three RNA secondary topologies computed using Zukers MFOLD algorithm.
T2. The 3 and 5 ends of a helical stem are considered a vertex.
T3. An RNA stem is considered an edge; we define an RNA stem to have two or more
complementary base pairs.
T4. An RNA junction is a vertex.
5.2.3. Planar Dual Graph Rules
To represent trees, pseudoknots and other RNA secondary topologies as planar graphs, we
use the following general rules.
D1. A vertex represents a double-stranded helical stem.
D2. An edge represents a single strand that may occur in segments connecting the

Algebraic Topics on Discrete Mathematics

169

secondary elements (e.g., bulges, loops, junctions and stems).

D3. No representation is required for the 3 and 5 ends.
5.2.4. Enumeration of RNA Graphs
To estimate the size of RNAs structural space and to aid in finding new RNA folds, we
now consider the enumeration of tree and dual graphs. Specifically, we seek to describe all
possible RNA topologies (NV) for a fixed number of vertices (V); the number of vertices
is a measure of RNA chain length.
Enumeration of RNA tree graphs
The number of possible RNA graphs (NV ) for a given number of vertices (V ) can be
counted using tree enumeration theorems of Cayley for labeled trees and Harary-Prins
for unlabeled trees (trees with equivalent vertices). These theorems are cornerstones in
the subarea of graph theory that deals with graphical enumeration. Labeled trees refer to
graphs with labeled vertices, as illustrated in Figure 3; graph vertices are not labeled in an
unlabeled graph. Enumeration of unlabeled trees considers the number of non-isomorphic
graphs, i.e., topologically distinct trees irrespective of the vertex identity. The labeled trees,
on the other hand, allow distinction of specific bulges, loops, junctions and ends in RNA
graphs. Both Cayley and Harary-Prins approaches are relevant to the counting of RNAs
structural repertoire.
The number of labeled trees for any V is given by the Cayley formula:
NV = V V 2

(7)

For unlabeled trees, Harary and Prins obtained the counting polynomial t(x) whose
coefficient NV is the number of distinct graphs with V vertices:
t(x) =

NV xV = T (x)

V =1

where
T (x) = x exp


1 2
T (x) t(x2 )
2

hX
1
r=1

T (xr )

(8)

(9)

The counting polynomial (equations 8 and 9) up to the first 12 terms is

t(x) = x+x2 +x3 +2x4 +3x5 +6x6 +11x7 +23x8 +47x9 +106x10 +235x11 +551x12 +. . .
(10)
In this polynomial, the coefficients of the first, second and third terms, for example,
indicate that there is only one distinct graph each for V = 1, 2 and 3; the Harary-Prins
enumeration polynomial is derived based on the Plya method. Clearly, the number of
distinct graphs (NV ) as a function of vertex number according to Cayleys formula ( 7 ) for
labeled trees grows faster than Harary-Prinss formula ( 8 ) for unlabeled trees.
Based on the counting polynomial (equation 10 ), we can estimate the number of distinct
secondary motifs for a given RNA size. Since a tree edge roughly corresponds to 20 nt, a

170

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

tree with six vertices (five edges or 100 nt) has six possible motifs, whereas an 11-vertex
(10 edges or 200 nt) tree has 235 possible motifs. As the RNA size increases from 100 to
200 nt, the number of possible motifs increases by a factor of 39, indicating the potential
of large RNAs to form many more novel secondary motifs.
Enumeration of RNA dual graphs
The enumeration of dual graphs, unlike trees, simultaneously yields tree, pseudoknot
and other possible topological motifs as defined by the dual graph rules (D1-D3). We
have heuristically enumerated all such graphs for the cases of V = 2, 3 and 4, which
correspond to 3, 8 and 30 possible dual graphs, respectively. In addition to RNA trees
(T ) and pseudoknots (P ), enumerated motifs in Figure 5.2.4. reveal graphs involving
single-edge connectors; we call such motifs bridge graphs ( B) or simply bridges. Bridges
are biologically important since they suggest existence of independent RNA submotifs and
thereby help in the modular design of RNAs. Examples of RNA bridges are box H/ACA
snoRNA, hepatitis C virus (HCV) RNA and group I intron.
The distribution of tree, pseudoknot and bridge types in V = 2, 3 and 4 motif sets is as
follows. For the V = 2 motif set, three graphs correspond to one tree, one pseudoknot and
one bridge; for V = 3 set, eight graphs correspond to two trees, three pseudoknots and three
bridges; and for V = 4 set, 30 graphs include four trees, 20 pseudoknots and 13 bridges
(thus seven graphs are both pseudoknots and bridges). These enumeration results imply
that the number of bridges (N V bridge ), trees (N V tree ), and pseudoknots (N V pseudo ) within
a given topological set follow:
N V tree N V bridge N V pseudo
The complexity and number of the dual graphs increase quickly with vertex number,
making it non-trivial to determine the number of topological possibilities for a given V .
General enumeration theorems for RNA dual graphs are not available.
5.2.5. Conclusion
We estimate the number of distinct RNA tree motifs based on the Cayley and Harary-Prins
enumeration theorems. These theorems imply that the RNA topology space is much smaller
than the sequence space, which renders our topological approach potentially effective for
finding novel RNAs. Our surveys of existing RNAs identified a number of motifs in nature
but showed that many hypothetical motifs do not exist. Since not all enumerated motifs are
probable RNAs, energetic, functional and evolutionary aspects of RNA folds must be taken
into consideration to provide better future estimates of RNAs repertoire.

5.3. Octonions and Transylvania Lottery

Developed by John Baezs Stuff from Department of Mathematics. University of California
(USA).
The quaternions, H, are a 4-dimensional algebra with basis 1, i, j, k. To describe the
product we could give a multiplication table, but it is easier to remember that:

Algebraic Topics on Discrete Mathematics

1. 1 is the multiplicative identity,

2. i, j and k are square roots of 1,

171

172

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

3. we have , ij = k, ji = k, and all identities obtained from these by cyclic permutations of (i, j, k).
We can summarize the last rule in a picture:

When we multiply two elements going clockwise around the circle we get the next one:
for example, ij = k. But when we multiply two going around counterclockwise, we get
minus the next one: for example, ji = k.
We can use the same sort of picture to remember how to multiply octonions:

This is the Fano plane, a little gadget with 7 points and 7 lines. The lines are the sides
of the triangle, its altitudes, and the circle containing all the midpoints of the sides. Each
pair of distinct points lies on a unique line. Each line contains three points, and each of
these triples has has a cyclic ordering shown by the arrows. If ei , ej and ek are cyclically
ordered in this way then
ej ei = ek
ei ej = ek
Together with these rules:
1. 1 is the multiplicative identity,
2. e1 , e2, . . ., e7 are square roots of 1

Algebraic Topics on Discrete Mathematics

173

the Fano plane completely describes the algebra structure of the octonions. Indexdoubling corresponds to rotating the picture a third of a turn. This is certainly a neat
mnemonic, but is there anything deeper lurking behind it? Yes! The Fano plane is the
projective plane over the 2-element field Z2 . In other words, it consists of lines through the
origin in the vector space Z32 . Since every such line contains a single nonzero element, we
can also think of the Fano plane as consisting of the seven nonzero elements of Z32 . If we
think of the origin in Z32 as corresponding to 1 O, we get the following picture of the
octonions:

Note that planes through the origin of this 3-dimensional vector space give subalgebras
of O isomorphic to the quaternions, lines through the origin give subalgebras isomorphic
to the complex numbers, and the origin itself gives a subalgebra isomorphic to the real
numbers.
What we really have here is a description of the octonions as a twisted group algebra.
Given any group G, the group algebra R[G] consists of all finite formal linear combinations
of elements of with real coefficients. This is an associative algebra with the product coming
from that of G . We can use any function
: G2 1
to twist this product, defining a new product
? : R[G]xR[G] R[G]
by:
g ? h = (g, h)gh
where g, h G R[G]. One can figure out an equation involving that guarantees
this new product will be associative. The category of Z32 -graded vector spaces provides a
context in which the octonions are commutative and associative. So far this idea has just
begun to be exploited.
Example 5.2. A curious application of finite geometry is how to win in Transylvanian
lottery. The Transylvanian lottery is a lottery where three numbers between 1 and 14 are
picked by the player, and three numbers are chosed randomly. The player wins if two of his

174

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

numbers are among the random ones. The problem of how many tickets the player must buy
in order to be certain of winning can be solved by the use of the Fano plane. The solution
is to buy a total of 14 tickets, comprised of two sets of seven. One set of seven is every line
of a Fano plane with the numbers 1-7, the other with 8-14, i.e.:
1 2 5, 1 3 6, 1 4 7, 2 3 7, 2 4 6, 3 4 5, 5 6 7,
8 9 12, 8 10 13, 8 11 14, 9 10 14, 9 11 13, 10 11 12, 12 13 14

Because at least two of the winning numbers must be either high (8-14) or low (1-7),
and every high and low pair is represented by exactly one ticket, you would be guaranteed
at least two correct numbers on one ticket with these 14 purchases. 21 each 26 of the time
you will have one ticket with two numbers matched. If all three winning numbers are either
high or low you would either have one ticket with all three numbers (1 each 26 chance of
this occurring), or three different tickets that each matched two (4 each 26 chance).

5.4. Discrete Mathematics with Mathematica

Based in "Computational Discrete Mathematics" by Sriram Pemmaraju and Steven Skiena
from Department of Computer Science. Universities of Iowa and New York (USA).
We recommend Mathematica programming language to solve discrete mathematic problems with computational tools. Firstly we need to load Combinatorica package:
<< DiscreteMathCombinatorica
5.4.1. Basic Combinatorial
Basic combinatorial objects such as permutations, subsets, and partitions are the raw materials from which discrete mathematics is built. Some examples of these functions are:
Permutations: Permutations[3]
Subsets: Subsets[{a,b,c,d}]

Algebraic Topics on Discrete Mathematics

175

Partitions (based in Young Tableau): Partitions[6]

Compositions: Compositions[6,4]
Ferrers Diagram: FerrersDiagrams[{1,1,5,6}]

5.4.2. Graph Theory

Graphs appear everywhere and can model almost anything, includin road networks, electronic circuits, engineering diagrams, social interactions, and even the World Wide Web.
Complete graphs: ShowGraphArray[Table[CompleteGrap h[n],{n,5,7}]]

Adjacency Matrix: (m=ToAdjacencyMatrix[g])//TableForm

Maximum Clique:
d=RandomGraph[10,.7];ShowGraph[Hi ghlight[g,{MaximumClique[d]}]];
Hamilton Cycle:
h=Partition[HamiltonCycle[g=Icosah edralGraph],2,1];
ShowGraph[Highlight[g,{c}]];

176

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

5.4.3. Plya Theory

A classic combinatorial problem is counting how many different ways necklaces can be
made out of k beads using a given set of colors. The beauty of Plyas approach to this
problem is that it yields a polynomial, the coefficients of whose terms provide answers to a
more general problems. Combinatorica provides extensive support for Plya Theory.
Cyclic is an argument to the Polya-theoretic functions ListNecklaces, NumberOfNecklace,
and NecklacePolynomial, which count or enumerate distinct necklaces. Cyclic refers to the
cyclic group acting on necklaces to make equivalent necklaces that can be obtained from
each other by rotation.
Plyas polynomial: NecklacePolynomial[6,{a,b,c},Cyclic]
List
of
Neckclace: ListNecklaces [5, {r, r, b, b, y}, Dihedral]
{r, r, b, b, y},{b, b, r, y, r},{b, r, b, r, y},{b, r, r, b, y}
Dihedral Group: DihedralGroup[4]
Symmetric Group: SymmetricGroupIndex[6, x]
x61
1
1
x3 1
1
x2 1
1
1
x6
+ x41x2 + x21 x22 + 2 + x31x3 + x1x2x3 + 3 + x21x4 + x2x4 + x1x5 +
720 48
16
48 18
6
18 8
8
5
6

Algebraic Topics on Discrete Mathematics

177

References
[1] Meenaxi Bhattacharjee, Dugald Macpherson, Rgnvaldur G. Mller, and Peter M.
Neumann, Notes on infinite permutation groups, Texts and Readings in Mathematics,
vol. 12, Hindustan Book Agency, New Delhi, 1997, , Lecture Notes in Mathematics,
1698.
[2] Mauro Biliotti and Alessandro Montinaro, Finite projective planes of order n with a
2-transitive orbit of length n3, Advances in Geometry 6 (2006), 1537.
[3] Raj Chandra Bose, S. S. Shrikhande, and E. T. Parker, Further results on the construction of mutually orthogonal latin squares and the falsity of eulers conjecture, Canad.
J. Math. 12 (1960), 189203.
[4] Richard Hubert Bruck and Herbert J. Ryser, The non-existence of certain finite projective planes, Canadian Journal Math. 1 (1949), 8893.
[5] Juan de Burgos Roman, Curso de lgebra y geometra, vol. 1, Alambra, Madrid, 1982.
[6] John D. Dixon and Brian Mortimer, Permutation groups, Graduate Texts in Mathematics, vol. 163, Springer-Verlag, New York, 1996.
[7] David S. Dummit and Richard M. Foote, Abstract algebra, Prentice Hall Inc., Englewood Cliffs, NJ, 1991.
[8] Leonard Euler, De quadratis magicis, Commentationes arithmeticae 2 (1849), 593
602.
[9] Leonard Euler, Leonardi euleri opera omina , 1, vol. 7, Tuebner, Berlin-Leipzig, 1923.
[10] Gino Fano, Sui postulati fondamentali della geometria in uno spazio lineare ad un
numero qualunque di dimensioni, Giornale di matematiche 30 (1892), 106132.
[11] Harald Fripertinger, Enumeration in musical theory, Sminaire Lotharingien de Combinatoire (Thurnau, 1991), Publ. Inst. Rech. Math. Av., vol. 476, Univ. Louis Pasteur,
Strasbourg, 1992, pp. 2942.
[12] Ralph P. Grimaldi, Discrete and combinatorial mathematics: An applied introduction ,
Addison-Wesley Longman Publishing Co., Inc., Boston, MA, USA, 1998.
[13] Hin Hark Gan, Samuela Pasquali, and Tamar Schlick, Explorin the repertoire of rna
secondary motifs using graph theory; implications for rna design, Nucleic Acids Research 31.
[14] Israel Nathan Herstein, Abstract algebra, third ed., Prentice Hall Inc., Upper Saddle
River, NJ, 1996, With a preface by Barbara Cortzen and David J. Winter.
[15] Nathan Jacobson, Basic Algebra I, New York : W.H. Freeman, 1985.
[16] Dieter Jungnickel and Marialuisa J. de Resmini, Another case of the prime power
conjecture for finite projective planes, Advances in Geometry 2 (2002), 215218.

178

Gloria Gutirrez Barranco, Javier Martnez, Salvador Merino et al.

[17] Clement Wing Hong Lam, Larry Henry Thiel, and S. Swiercz, The non-existence of
finite projective planes of order 10 , Canadian Journal Math. 41 (1989), 11171123.
[18] Serge Lang, Algebra, third ed., Graduate Texts in Mathematics, vol. 211, SpringerVerlag, New York, 2002.
[19] Dugald Macpherson and Cheryl E. Praeger, Cycle types in infinite permutation groups,
J. Algebra 175 (1995), no. 1, 212240.
[20] Brendan D. McKay and Eric Rogoyski, Latin squares of order 10,
The Electronic Journal of Combinatorics 2 (1995), no. 1,N3, 14,
http://www.combinatorics.org/Volume_2/PDFFiles/v2i1n3.pdf.
[21] Sriram Pemmaraju and Steven Skiena, Computational discrete mathematics, Cambridge University Press, Cambridge, 2003, Combinatorics and graph theory with
Mathematica.
[22] Anavai Ramesh, George Becker, and Neil V. Murray, CNF and DNF considered harmful for computing prime implicants/implicates, J. Automat. Reason. 18 (1997), no. 3,
337356.
[23] Herbert J. Ryser, Combinatorial mathematics, Carus Mathematical Monographs,
vol. 14, Mathematical Association of America, 1963.
[24] kos Seress, Permutation group algorithms, Cambridge Tracts in Mathematics, vol.
152, Cambridge University Press, Cambridge, 2003.
[25] Douglas R. Stinson, A short proof of the non-existence of a pair of orthogonal latin
squares of order six, Journal Combinatorial Theory A36 (1984), 373376.
[26] Gaston Tarry, Le probleme des 36 oficiers, Compte Rendu Ass. Franc. Pour
lavacement des Sciences 2 (1901), 170203.
[27] Eric W. Weisstein, Finite Field. From MathWorldA Wolfram Web Resource ,
http://mathworld.wolfram.com/FiniteField.html.
[28] Frank Yates, A new method of arranging variety trials involving a large number varieties, Journal of Agriculture Science 26 (1936), 424455.
*A version of this chapter was also published in Discrete Mathematics Research Progress ,
edited by Kenneth B. Moore published by Nova Science Publishers, Inc. It was submitted
for appropriate modifications in an effort to encourage wider dissemination of research.

In: Mathematics and Mathematical Logic: New Research

ISBN 978-1-60692-862-2
c 2010 Nova Science Publishers, Inc.
Editor: Peter Milosav and Irene Ercegovaca

Chapter 7

M ATHEMATICAL M ODELS OF Q O S M ANAGEMENT

FOR C OMMUNICATION N ETWORKS *
Chia-Hung Wang and Hsing Luh
Department of Mathematical Sciences, National Chengchi University,
Wen-Shan, Taipei 11623, Taiwan

1 Abstract
Network design and network synthesis have been the classical optimization problems in
telecommunication for a long time. In the recent past, there have been many technological
developments such as digitization of information, optical networks, Internet, and wireless
networks. These developments have led to a series of new optimization problems.
In communication networks, a number of requirements can be identified. For instance,
the network operator requires a good earning capacity, and the network users require reliable communication. The degree to which these and other requirements are fulfilled, can
in many cases be deduced from how the network resources are distributed. In contrast to
the Public Switched Telephony Network (PSTN), services offered in a data communication
network vary significantly in terms of the required bandwidth. This adds a new challenging
dimension into design of the networks, since inevitably, questions of fair medium sharing,
quality guarantees, delays, etc. become crucial.
Different methods of obtaining fair resource sharing are derived and investigated, for
example, Max-Min Fairness, Proportional Fairness, etc. We focus on fairness on backbone
networks possessing certain desired fairness properties.
Key Words: Resource Allocation, Communication Networks, Optimization

2 Introduction
In recent years, we have witnessed considerable accomplishments in the design and deployment of broadband communication networks. The dramatic growth of communications, as

93751502, slu@nccu.edu.tw

180

Chia-Hung Wang and Hsing Luh

arose in the 90s along with the increasing availability of Internet, has given new prominence to dimensioning of the network infrastructure facilitating transfer of information (in
form of data, speech, etc.). An Internet service can be anything from electronic mail to
video-conferencing, but is always associated with a transfer of data, i.e., data communications. Network capabilities grow at a remarkable rate. At the same time, a phenomenal
growth in data traffic and a wide range of new requirements of emerging applications call
for new mechanisms for the control and management of communication networks. The idea
of a single shared physical network that will support multiple heterogeneous applications
with different traffic characteristics and different Quality of Service (QoS) requirements, is
widely regarded as the way to meet the telecommunication challenges of the future [14].
QoS has always been the major issue for telecom providers [51]. Packet-switched networks
have been proposed to offer the QoS guarantees in integrated-services networks because
individual packets may exhibit a significant variation in network service quality.
The rules for bandwidth allocation in networks carrying elastic traffic can be defined in
several ways, meeting different overall network goals. In Internet, it is desired to avoid poor
users, meaning users possessing an unfairly small share of bandwidth. The available bandwidth can be divided between the users according to their needs. Therefore, a problem of
network dimensioning with elastic traffic requires to allocate bandwidth to maximize flows
fairly [34]. Fair resource allocation problems are concerned with the allocation of limited
bandwidth among competing activities so as to achieve the best overall performances of the
system but providing fair treatment of all the competitors (see [20], [21], [34]). We introduce the methodology that allows the decision maker to explore a set of solutions satisfying
users preferences with respect to throughput and fairness. The formulation and analysis is
carried out in a general utility-maximizing framework.
QoS routing concerns the selection of a path satisfying the QoS requirements of a connection [15], [35]. The path selection process involves the knowledge of the connections
QoS requirements and information on the availability of bandwidth. QoS routing poses
major challenges in terms of algorithmic design. On one hand, the path selection process
is a complex task, due to the need to concurrently deal with the connections QoS requirements, as well as with the global utilization of network resources [4]; on the other hand,
connection requests need to be handled promptly upon their arrival [50]. Depending on the
specifics and the number of QoS metrics involved, computation in real time required for
path selection can become prohibitively expensive as the network size grows.
Consider a fixed network topology G = (V, E), where V and E denote the set of nodes
and the set of links in the network respectively. Given the maximal possible capacity of
each link. Suppose we know that, for each link, the cost taking account of delay and the
purchasing cost of bandwidth. In this network, there arem different classes which have their
own QoS requirement. In each class, every connection is allocated the same bandwidth
and has the same QoS requirement. Suppose each connection is delivered between the
same source and destination in this (core) network. Under a limited available budget, we
want to allocate the bandwidth in order to provide each class with maximal possible QoS.
The purpose of this work is to show that a methodology that allows the decision maker to
explore a set of solutions could satisfy preferences with respect to throughput and fairness,
and choose the solution which the decision maker finds best.
The challenge in integrated-services networks is to determine general resource alloca-

Mathematical Models of QoS Management for Communication Networks

181

tion and QoS routing schemes that have the following desirable goals:
1. Links and bandwidth must be determined in a way such that the network can assure
fairness and maximize revenue.
2. Bandwidth must be allocated for each flow such that user requirements are met and
at the same time such that the predefined condition of fairness is satisfied.
3. An appropriate effective route for each flow (between two nodes) must be found.
These operational processes are involved in the efficient set-up and usage of a network.
Three main components of these processes are designed for which links to develop to meet
certain connectivity requirements, determining how much capacity to put on the links to
serve all traffic demands, and choosing which paths to use for the various traffic streams to
meet demand without violating capacity restrictions on links.
We deal with the problem of dimensioning bandwidth for elastic data applications in
packet-switched communication networks, which can be considered as a multiple-objective
optimization model. Each user is allowed to request more than one type of service, and
users satisfaction is summarized by means of their achievement functions. We focus on
allocating resources with proportional fairness and finding a routing scheme on communication networks. An approach is presented for the fair resource allocation problem and QoS
routing in networks, offering multiple services to users. The objective of the optimization
problem is to determine the amount of required bandwidth for each class to maximize the
sum of the users satisfaction.

3 Network Architecture
Internet is known to be of a best-effort nature, where each service is completed to the level of
quality that relevant transmission media (links, switches and routers) can provide. In effect,
an uncertainty is implied - a service may be satisfactorily set-up, but also, which is unfortunate, insufficiently carried out, due to time delays, connection interrupts etc. It may also
imply that particular applications work well, while others totally fail. The most common
problem, when a certain service fails, is lack of enough transmission bandwidth (measuring
the amount of data possible to transfer per second). Therefore, a high occurrence of failures
in service completion is commonly remedied by overdimensioning of network resources,
eliminating arising bottlenecks by adding capacity. The overdimensioning strategy has the
potential of solving the problem temporarily. However, regarding the fast development of
new, sophisticated Internet applications, it is reasonable to assume that the future will require even further overdimensioning. As the process continues, this strategy will result in a
network with a very large capacity, only occasionally fully used, and kept unemployed for
most of the time. Undoubtedly, this is economically unattractive.
Traffic generated in an edge network can roughly be put in either of two categories,
namely i) elastic traffic, which is a type of best-effort traffic that employs whatever resources
it is assigned, and ii) non-elastic traffic which is traffic that is typically sensitive to assigned
bandwidth and transmission delays etc. [40]. The data communications community is well
aware of the potential traffic property differences, as for instance the future Internet Protocol

182

Chia-Hung Wang and Hsing Luh

Table 1: UMTS Service Classes
Sensitivity to Sensitivity to Sensitivity to
Jitter
Delay
Packet Loss
Conversational
high
high
low
Streaming
high
high
low
Interactive
low
low
high
Background
very low
low
high
Traffic Class

(IPv6) features the so called flow-labelling function, which accommodates different QoS
requirements. Examples of elastic traffic are classical Internet applications as Web, File
Transfer Protocol (FTP), and mail. Real-time applications, as e.g. Voice over IP (VoIP) and
video-conferencing, are on the other hand typically nonelastic. Since the different types of
traffic are generated in the edge networks, they are as indicated above, multiplexed before
transmitted into the backbone network. Therefore, the aggregated backbone traffic flows
consist of both elastic and non-elastic traffic, and an appropriate denomination would be
semi-elastic traffic. This pushes forward the fact that the backbone traffic flows require a
minimal bandwidth level (constituted by the sum of the required bandwidth for the nonelastic applications), but still are able consume bandwidth exceeding this level (due to its
elastic portion).
The Universal Mobile Telecommunications System (UMTS) [41] has specified four
different traffic classes according to their QoS requirements for different applications as
Table 1 shows. The basic core network architecture for UMTS is based on GSM network
with GPRS [41]. The main function of the core network is to provide switching, routing and
transit for user traffic. A core network also contains the databases and network management
functions. For example, UMTS network services have different QoS classes for four types
of traffic:
1. Conversational class (voice, video telephony, video gaming)
2. Streaming class (multimedia, video on demand, webcast)
3. Interactive class (web browsing, network gaming, database access)
4. Background class (email, SMS, downloading)
In the communications community, a backbone network (or core network), is a network
that realizes the connections between local networks, as illustrated in Fig. 1. From the
backbone point of view, the local networks can be seen as edge networks, to which most of
the subscribers (traffic generators) are connected. Consequently, the majority of the traffic
can be regarded as generated in the local networks. Traffic from different users in a local
network is concentrated (multiplexed) by a switch which connects to more distant places,
by the backbone.
A similar process takes place on practically every level of the network hierarchy. For
instance, the local networks are connected to a regional network, which in turn is connected to a national network, and so on. As the local networks induce the traffic, the traffic

Mathematical Models of QoS Management for Communication Networks

4

e4

183

e14

1
7

e5

e15

e1

Class 1

e11

Class 1

Class 2

e2

e12

e7

e19

Class 3

Class 4

e23

e24

e18

e6

e9

e8
3

d
Class 3

e20
e16

e3

Class 2

e25

10

e21

11

e26
Class 4

e13

e22

e10
e17

Figure 1: A Simplified Network Architecture

experienced by the backbone is an aggregated version of its local network counterparts.

Considered over a sufficiently short time-period, the backbone network traffic is, compared
to an access network, predictable and stable (and of course very large) in the sense of
flow magnitudes. The hierarchical structure of network components is present in traditional telecommunication networks, as well as in modern large-scale data communication
networks. When connecting a large set of users, the hierarchical network topology arises
because of geography, cost considerations, and different communities of interest.
Traditionally, traffic flows associated with communications are modelled using the theory of stochastic processes. This approach applies very well for classical telecommunications, as the dominant service telephony is remarkably well modelled by the Poisson
process, and its variants. Unfortunately, this is rarely the case for data communications,
which is usually characterized by a significant burst, that has shown to be very difficult to
model. As the origin of traffic loading backbones is gradually changing from telephony to
computer-oriented applications, modelling by stochastic processes has to some extension
turned obsolete. Accordingly, a need for alternative models has emerged. One that seems
adequate for backbone networks is the multi-commodity flow network model.
A large portion of the traffic loading a communication network is intended for local
transmission. However, the pattern of traffic flows is in a changing process, as the cost
of long-distance transmission is decreasing, and accordingly, traffic load of the backbone is
increasing. The reduced long-distance transmission costs are mainly due to the introduction
of very efficient optical transmission techniques, along with the dramatic improvement of
the computer-controlled hardware, constituting network switches and routers.
The trend in design of large-scale networks is to rationalize the number of resource
layers that constitute the network. Presumably, this will show valid even for the next generation Internet, which is foreseen to consist merely of a packet layer and a transport layer.
Controlling distribution of network resources (bandwidth) in the IP resource layer may resolve the problems induced by repetitive overdimensioning. Instead of augmenting capacity
in avoiding the risk of a few bandwidth-exhaustive services to impoverish the network, this

184

Chia-Hung Wang and Hsing Luh

technique allows us to distribute bandwidth in a fair way. Of course, there will still be
a need to upgrade capacities from time to time, but such controlled medium-sharing will
facilitate an overall network stability and potentially enhance subscriber equity. Notably,
although new IP mechanisms supporting such control exist, very little work has been done
in this area.
Usually, it is argued that in data communications and particularly the Internet, traffic
patterns are very difficult or even impossible to predict. While this is certainly true for small,
local networks, it is a well-known fact that the larger the networks that are considered, the
more predictable and stable (in a relative sense) is the volume of the traffic. Consequently,
for a backbone network (typically very large), it might be reasonable to assume traffic flows
to be almost static, over certain periods of time. Consider for instance a transatlantic fiber
connection. The traffic on such a medium is composed of millions of individual Internet
sessions, so no single session has virtually any influence on the overall traffic behavior. On
the other hand, traffic in a small access network can be quite sensitive to a single session.

4 QoS on All-IP Networks

4.1 Quality of Service
Telecommunication is moving toward a converged network which uses a single global IP
based packet switched network to carry all types of networks to replace the traditional
separated packet switched and circuit switched networks. With the increasing commercial
deployment of wireless networks, the issue of providing multiple services (including voice,
video, data, etc.,) is becoming more and more important [41]. Thus, the concept of Quality
of Service is being widely discussed and implemented.
The Internet is based on the simple concept that datagrams with source and destination
addresses can traverse a network of IP routers independently [51]. The Internet was historically built on the concept of a dumb network, with smarts at either end (at the sender
and receiver). There is a price to pay for this simplicity, however. Routers are allowed to
discard IP datagrams en route, without notice to sender or receiver. IP relies on upper-level
transports to keep track of datagrams, and retransmit as necessary. And these reliability
mechanisms can only assure data delivery; neither IP nor its high-level protocols can ensure timely delivery or provide any guarantees about data throughput. IP provides what
is called a best effort service [23]. It can make no guarantees about when data will arrive, or how much it can deliver. This limitation has not been a problem for traditional
Internet applications like web, email, file transfer, etc. However, new applications demand
high data throughput capacity (bandwidth) and have low-latency requirements when used
in two-way communications, that is, conferencing and telephony. Public and private IP
Networks are also being used increasingly for delivery of mission-critical information that
cannot tolerate unpredictable losses. Unlike pure virtual circuit technologies, IP does not
make hard allocations of resources. This provides much more efficient use of the available
bandwidth, and it is more flexible. Typical network traffic is bursty rather than continuous.
IP is datagram-based so it uses the available bandwidth most efficiently, by sharing what is
available as needed. This also allows IP to adapt more flexibly to applications with varying
needs. However, it also leads to some unpredictability in service. The capacity to tolerate

Mathematical Models of QoS Management for Communication Networks

185

this unpredictability relates to the level of guarantee they require.

In the simplest sense, QoS means providing consistent, predictable data delivery service, in other words, satisfying customer application requirements [51]. QoS is to the ability of a network element (e.g. an application, host or router) to have some level of assurance
that its traffic and service requirements can be satisfied. To enable QoS requires the cooperation of all network layers from top-to-bottom, as well as every network element from
end-to-end. Any QoS assurances are only as good as the weakest link in the chain between
sender and receiver. QoS does not create bandwidth. QoS only manages bandwidth according to application demands and network management settings, and in that regard it cannot
provide certainty if it involves sharing. Hence, QoS with a guaranteed service level requires
bandwidth allocation to individual data streams [23]. A priority for QoS designers has been
to ensure that best-effort traffic is not starved after reservations are made. QoS-enabled
(high-priority) applications must not disable the low-priority Internet applications. Among
many challenges yet to be overcome, the QoS problem is one of the most critical problems
to be solved [14], [15], [34], [35], etc.

4.2 Resource Allocation & Routing with QoS Guarantees

Resource allocation decisions are concerned with the allocation of limited bandwidth so as
to achieve the best system performances [34] and [42]. QoS routing concerns the selection
of a path satisfying the QoS requirements of users. The path selected most likely is not
the traditional shortest path. Depending on the specifics and the number of QoS metrics
involved, computation required for path selection can become prohibitively expensive as
the network size grows [15]. The path selection process involves the knowledge of the
connections QoS requirements and characteristics on the network (,e.g., limited available
bandwidth and delay) [13] and [35].
Broadband integrated services networks are expected to support multiple and diverse
applications, with various QoS requirements. Accordingly, a key issue in the design of
broadband architectures is how to provide the bandwidth in order to meet the requirements
of each connection, and, moreover, how to meet that goal in an efficient manner.
The ability to provide end-to-end guarantees [18], such as delay, depends to a large
extent on the scheduling policy and service discipline employed in the nodes. Such disciplines are characterized by bounds on the maximal delay that any node can incur, and hence
a corresponding bound on the end-to-end delay can be derived. Such a bound provides a
valuable tool for quantifying the quality of a path in terms of its ability to meet the QoS delay requirement. The corresponding routing problem is, therefore, to identify the path that
has the best performance, according to that bound and with respect to the QoS requirement.
Typical schedulers map delay guarantees into rate guarantees, and have nodes advertise the
residual rate they have available. In particular, the Guaranteed Service class proposed for
the Internet is based on such rate-based principles.
Network design today often considers the problem of designing networks that carry
elastic traffic [18] and [19]. If the network is also used for other types of communication
that require guaranteed quality of service, the network design problem can be decomposed
into two parts: first, design the network to carry non-elastic traffic in such a way that all
demands for that communication are satisfied. Next, use the spare capacity to carry elastic

186

Chia-Hung Wang and Hsing Luh

traffic of the IP protocol. Resource allocation models may be used to solve network design
problems [34].
Network management must stay within a budget of expenses for purchasing link bandwidth. Network dimensioning with elastic traffic may be thought of as a search for such network flows that will maximize the network throughput (the sum of all flows in the network)
while staying within a budget constraint for the costs of link bandwidth. However, such a
problem of formulation would lead to the starvation of bandwidth between certain network
nodes. Looking at the problem from the user perspective, the bandwidth between different
users should be treated as fairly as possible. Whatever the users preference, it would be
expressed in terms of fairness for a certain set of criteria which depend on the individual
connection. Let us first consider providing fairness for all connections between competitive
activities. Network management must consider two goals: increasing throughput and providing fairness [34]. These two goals are clearly conflicting if the budget constraint need to
be satisfied.

5 Problem Definition
The most convenient way of representing a communication network is by the notions provided by graph theory. Specifically, a network formed by nodes and links is modelled by a
graph of vertices and edges. In the sequel, the words vertex and edge are treated equivalent
to node and link, respectively. A link is regarded as a connection between the two nodes to
which it is attached. In the general setting, a link facilitates transfer of various entities, i.e.,
it may carry flows associated with several demands simultaneously. This is in fact a key
feature for the validity of the network model, and is often referred to as multi-commodity
flow network modelling. We consider all links to be directed. Furthermore, each link is
usually assigned a capacity, limiting the amount of flow that it can carry.
A connection is an directed request for flow transfer between two nodes. Sometimes it
can be convenient to associate such a demand with its volume. The volume is then specifying the requested flow size for the associated connection. A demand for flow can be realized
in several different ways, since two nodes may be connected directly or in various ways via
other nodes. A set of links connecting two nodes is referred to as path or route between the
nodes. A path is always associated with a connection. To keep track of the communication
possibilities, each connection is assigned a routing list, in which all the potential routes for
that connection can be found.
Consider a directed network topology G = (V, E) as shown in Fig. 1, where V and
E denote the set of nodes and the set of links in the network respectively. There are m
different QoS classes of connections in the network. Let I = {1, . . . , m} be an index
set consists of m different QoS classes. The specific QoS requirements, for each class i,
include minimal bandwidth requirement bi and maximal end-to-end delay constraint Di .
We denote the total number of connections, for each class i, by Ki . Let Ji , for each class
i, be an index set consists of Ki connections, that is, Ji = {1, . . . , Ki }. All connections
are delivered between the same source o and destination d in this (core) network. Every
connection, in the same class i, is allocated the same bandwidth xi and has the same QoS
requirement.

Mathematical Models of QoS Management for Communication Networks

187

The parameters Ki are usually uncertain at the beginning of planning periods. We

solve this in the parameters by their best point estimator, i.e., expected values. Here, we
assume the expected number Ki is known for the discussion under this precomputationbased scheme.
The maximal possible link capacity is Ue on each link e E. Suppose, for each link
e, we have a mean delay de related to the links speed, propagation delay, and maximal
transfer unit. We also have the link cost e for using one unit bandwidth. In general, link
cost e is inversely proportional to mean delay de .
A connection j in each class i should be routed through some path pi,j between o
and d. Under a limited available budget B, we plan to allocate the bandwidth in order to
provide each class with maximal possible QoS and determine the optimal path from end
to end under guaranteed service. The purpose of this work is to show that a methodology
that allows the decision maker to explore a set of solutions could satisfy preferences with
fairness, and choose the solution optimally.
Definition 1 A bandwidth allocation (x1 , . . . , xm ) is feasible if the bandwidth requirements are satisfied for all classes.
A proportionally fair bandwidth allocation (rate vector) is defined in [19] as follows:
Definition 2 A feasible bandwidth allocation (x1 , . . . , xm ) is called proportionally fair if
m ), we have:
and only if, for any other feasible allocation (
x1 , . . . , x
Xx
i xi
0.
xi
iI

In other words, a proportionally fair bandwidth allocation is a feasible bandwidth allocation

such that the utility of class i is (almost) the same as that of class j, 1 i, j m.
Let fi (xi ) be a utility function of the bandwidth allocation xi for class i. We introduce
a different notion of fairness, called -proportional fairness, and we give the definition as
follows.
Definition 3 A -proportionally fair bandwidth allocation is defined as a proportionally
fair bandwidth allocation (x1 , . . . , xm ) such that
|fi (xi ) fj (xj )| , 1 i, j m,
where is a small number and fi (xi ) is a given utility function of the bandwidth allocation
xi for class i.
Note that different forms of utility functions fi (xi ) may represent different bandwidth allocation policies. In section 3, we will discuss bandwidth allocation policies associated with
different utility functions fi (xi ).
Let n(pi,j ) be the number of links along path pi,j and i be the mean packet size for
each class i, i I. When a connection j in class i (with a bandwidth xi ) is routed along
a path pi,j , the following end-to-end delay D(pi,j ) applies (Atov et al. [4], Johari and Tan
[18], Orda [35], etc.):
X
n(pi,j ) i
+
de .
(1)
xi
ep
i,j

The following definitions given by us will be used throughout this paper.

188

Chia-Hung Wang and Hsing Luh

Definition 4 A feasible path pi,j between o and d, for a connection j of class i, is defined
as a (routing) path from o to d such that D(pi,j ) Di and each link along pi,j satisfies the
capacity constraint. A connection j of class i is feasible if it has a feasible path.
Definition 5 A feasible path pi,j is called a Pareto optimal path, for a connection j in class
i, if no other feasible path is as less as pi,j with respect to two evaluation, path cost and
ene-to-end delay, and strictly less than pi,j with respect to at least one evaluation.
Definition 6 The set of all Pareto optimal paths is called the routing database P . That is,
P = {pi,j | pi,j is the Pareto optimal path from o to d, j Ji , i I}.
Definition 7 A link
Pcalled bottleneck link if the usage of bandwidth achieves its link
P e is
capacity, that is, iI jJi yi,j (e) = Ue .
Decision variables are listed as follows:
yi,j (e) the bandwidth allocated to link e E for connection j in class i;
the bandwidth allocated to each connection in classi;
xi
zi,j (e) a binary variable which determines whether the link e is chosen for
connection j in class i.
Some notations are listed below:
V
A set of nodes in the network topology
E
A set of links in the network topology
I
An index set consists of m different QoS classes
Expected number of connections in class i
Ki
An index set consists of Ki connections in class i
Ji
Bandwidth requirement for each connection in class i
bi
B
Limited available budget
Unit cost for each link e E
e
Maximal capacity of each link e
Ue
M
A sufficiently large positive number
A subset of outgoing links connected with the source o
Eo
A subset of incoming links connected with the destination d
Ed
A subset of incoming links to the node V
Ein

A subset of outgoing links from the node V

Eout

Assigned weight for each class i

wi
Aspiration level of bandwidth for each class i
ai
Reservation level of bandwidth for each class i
ri
Maximal end-to-end delay constraint for each class i
Di
Mean arrival rate of connections in class i
i
Mean packet size of connections in class i
i
Mean delay on link e
de
p
A path from the source o to the destination d
D(pi,j ) End-to-end delay along a path pi,j for connection j in class i
The bandwidth allocated to each connection in classi
xi
yi,j (e) The bandwidth allocated to link e E for connection j in class i
zi,j (e) A binary variable which determines whether the link e is chosen for
connection j in class i.

Mathematical Models of QoS Management for Communication Networks

189

Network Requirements

In designing a communication network, as in practically all problems of mathematical nature, some parameters are regarded as fixed (constant) and some are variable. The preliminary design target is to determine the variables, such that some overall objectives are
reached. Since a backbone certainly is a large and very complex network, there are several
ways of specifying a particular target, as well as variable and static parameters. Depending
on from what viewpoint the network is considered, a number of different problems can be
identified.
This paper uses the bandwidth and budget as constraints of requirements for feasible
path computations. Due to the limited budget on network planning, there exists the budget
constraint (2).
XXX
e yi,j (e) B
(2)
eE iI jJi

Because the aggregate bandwidth of all connections at any link does not exceed the capacity,
we have constraint (3).
XX
yi,j (e) Ue , e E
(3)
iI jJi

Constraints (4), (5), (6), and (7) show that every connection in the same class uses the same
bandwidth and has the same bandwidth requirement. For each e E, j = 1, . . . , Ki , and
i = 1, . . . , m,
(4)
yi,j (e) M zi,j (e) 0,
xi yi,j (e) M (1 zi,j (e)),

(5)

yi,j (e) xi M (1 zi,j (e)),

(6)

xi bi ,

(7)

and
where M is a sufficiently large number. Constraints (8), (10), and (9) express the node
conservation relations indicating that flow in equals flow out for every connectionj in class
i. For each j = 1, . . . , Ki , and i = 1, . . . , m,
X
yi,j (e) = xi ,
(8)
eEo

yi,j (e) = xi ,

(9)

eEd

and

X
eEin

yi,j (e) =

yi,j (e),

(10)

eEout

for all V \ {o, d}. Although yi,j (e) are continuous variables, constraints (8)-(9) are
flow conservation constraints. Continuous decision variables and binary variables must be
nonnegative, shown in constraints (11)-(13). For e E, j = 1, . . . , Ki , i = 1, . . . , m,
yi,j (e) 0,

(11)

190

Chia-Hung Wang and Hsing Luh

xi 0,

(12)

zi,j (e) {0, 1}.

(13)

and
Many problems in the network design context are particularly well formulated as optimization tasks. In general, an optimization task is formalized as a mathematical programme
consisting of an objective function and a set of constraints necessarily fulfilled. Our goal is
to maximize the total utility of all competing classes. Letfi be a real valued utility function
(which will be made apparent) and consider the mathematical programming
(Model 1)

max

wi fi (xi )

(14)

iI

s. t.

(2) (13)

where wi (0, 1) is the weight assigned to each class i,

P
iI wi Mi is a constant.

(15)
P

iI wi

= 1, and M =

Theorem 8 Model 1 is NP-hard.

From the optimization of these precomputation schemes, we suggest the Pareto optimal
bandwidth allocation and a routing database with end-to-end QoS guarantees, which are
described in the following sections.

7
7.1

Bandwidth Allocation Policies

Maximum Throughput Allocation

In the long run, a network operator always requires economical profit from the administrated network. For the possibility of this to be achieved, there are some obvious properties that a backbone must possess. Usually, vital parameters of the operators income are
proportional to the networks total throughput. The simplest way of defining a network allocation rule, and also the most profitable way from the network operator viewpoint, is to
maximize the total throughput of the network. This optimization task is called throughput
maximization, and is formulated as follows:
X
Ki xi
(Model 2)
max
iI

s. t.

(2) (13),

where Ki is the number of connections in class i. This is a classical optimization problem,

in most cases easily solved. To maximize throughput, it is obvious that all flows should
be routed on their shortest paths. The solution to this optimization problem is not unique.
Hence, the maximum throughput fairness principle does not give a unique bandwidth allocation for a fixed set of ongoing connections.
This is of course unrealistic, since there is always some variation in demands over time.
Anyhow, a high utilization of the network capacity is indeed considered advantageous. A

Mathematical Models of QoS Management for Communication Networks

191

requirement of this type could be classified as direct, as it is imposed directly on the network by the operator. Indirect requirements are in this notion requirements imposed on
the network via the operator by the network service customers - the subscribers. In purchasing a service, the customers want to be guaranteed a certain degree of quality. Quality
of a service in turn, depends on to what amount of bandwidth a customer is assigned, the
connection reliability, etc. The volume of bandwidth assigned to one customer is strongly
dependent on the bandwidths assigned to other customers, which is natural since, after all,
the network is a shared medium. Regulations for how the medium should be shared must
take into account aspects of fairness, operators revenue, network utilization, etc.

7.2

Arctan-Utility Maximizing Allocation

A naive way of achieving fairness is to allocate maximally possible, equal bandwidth to

each of the users. The apparent drawback is that the required equality makes it impossible
to increase certain non-blocked bandwidth allocations, despite the fact that this could be
done without affecting other already blocked allocations.Though fair, such an allocation
suffers from unnecessary ineffectiveness in terms of network total throughput. However, the
public demand for services provided by an unfair network will be very limited irrespective
the achieved throughput, which advocates that fairness should be prioritized over the total
throughput.
Many Internet applications implement a type of data transmission rate-control conforming to the TCP algorithm. Kelly [19] and Low [26] argued that this type of allocation can
be approximated by the following objective function:
X
xi
wi Ki arctan( )
(Model 3)
max
wi
iI

s. t.

(2) (13),

where wi is a positive constant.

7.3

Minimal Potential Delay Allocation

Massouli
e and Roberts ([30], [31]) proposed an allocation policy called minimum potential
delay with utility functions (16). The minimal potential delay allocation is characterized as
the following objective function:
(Model 4)

max

s. t.

(2) (13).

iI

Ki
xi

Note that when Ki = 0 and thus xi = 0, the corresponding term in the objective function
should be dropped out.

7.4

Max-Min Fair Allocation

The classical notion of fairness is the so-called Max-Min Fairness (MMF). In its original form, the MMF rule of allocating bandwidth disregards network total throughput, and

192

Chia-Hung Wang and Hsing Luh

instead emphasizes the importance of meeting the demands of the poor users. In a pure fairness sense, it could be argued that allocating bandwidth according to this rule is optimal.
This fact and simplicity/efficiency of implementation make it commonly used in various
existent networks.

7.5

Proportionally Fair Allocation

Kelly et al. [20] show that an allocation policy can be expressed in terms of a utility function, as a function of bandwidth (rate), in the sense that the desired bandwidth allocation
maximizes aggregate utility subject to constraints. Kelly et al. [20] advocated proportional
fairness characterized by log(xi ). This log utility function is strictly concave. The proportional fair bandwidth allocation is determined by the following objective function:
(Model 5)

max

Ki log(Ki xi )

iI

s. t.

(2) (13).

Determining the maximizer of this model can be done explicitly for simple networks. In this
case Kelly et al. [20] have shown that the maximizer corresponds to a proportionally fair
bandwidth allocation. Furthermore, for any other feasible rate, the aggregate weighted proportional change is nonnegative. Alternatively one can design rate control mechanisms that
converge to the associated bandwidth allocation [20]. This method of allocating bandwidth
yields a compromise between high total bandwidth allocation (total network throughput)
and fairness between users.

7.6 (w, )-Proportionally Fair Allocation

Mo and Walrand [32] characterized the class of (w, )-proportionally fair bandwidth allocation, for any given number ( > 0, 6= 1), as the following objective function:
(Model 6)

max

X
iI

s. t.

wi Ki

x1
i
1

(2) (13),

where wi is a fixed parameter. Due to the strict concavity of the function to be maximized,
this defines a unique allocation referred to as -bandwidth allocation. This allocation corresponds to the maximum throughput criterion when 0, to proportional fairness when
1, to the potential delay criterion when 2, and to max-min fairness when
[32].

7.7 -Proportionally Fair Allocation

Wang and Luh [43], [45] transformed the different QoS measurements onto a normalized
scale by using achievement functions. Depending on the specified aspiration and reservation levels, ai and ri , respectively, Wang and Luh [43], [45] proposed the proportional fair

Mathematical Models of QoS Management for Communication Networks

193

bandwidth allocation by using the achievement function ofxi constructed as follows:

(Model 7)

max

wi log i

iI

s. t.

xi
ri

(2) (13),

where wi is a fixed parameter and i = ai /ri . It is a strictly increasing function of xi ,

having value 1 if xi = ai , and value 0 if xi = ri . The use of the logarithmic function
prevents the possibility of assigning zero flow to any user, and on the other hand makes it
non-profitable to assign too much flow to the users. Note that this allocation is equivalent
to proportionally fair allocation.
In equilibrium, connections that share the same links do not necessarily equally share
the available bandwidth. Rather, their shares reflect how they value the bandwidth as expressed by their utility functions and how their use of the bandwidth implies a cost on
others. This could be a basis to provide differentiated services in terms of different bandwidth allocations. An equilibrium bandwidth allocation is usually characterized in terms of
its fairness to users. Thus given a fixed number of users and fixed network capacities, one
can typically arrange through an appropriate control mechanism to achieve an equilibrium
which represents, according to some criterion, an equitable bandwidth allocation among
users (see [20], [22], [26], [31], etc.)

Implication

We introduce the concept of majorization1 to provide the fairness. For any m-dimensional
vector x=( x1 , . . . , xm ) of reals, let x(1) . . . x(m) denote the components of x in
increasing order.
P
Pm
Pk
Definition 9 For x and t in Rm , x M t if m
i=1 s(i) =
i=1 t(i) and
i=1 x(i)
Pk
i=1 t(i) , for k = 1, . . . , m 1. When x M t then x is said to be majorized by t.
If x M t, then the allocation x is more fair than t. Next, we have the following
definition.
Definition 10 A function g : Rm R is called Schur-concave, if x M t implies g(x)
g(t).
Thus, we have the following theorem taken from [33].
Theorem 11 Let h be an arbitrary real function and defineg(x)=
then g is Schur-concave if and only if h is concave.

Pm

i=1

h(xi ) for x Rm ,

1
Multiple criteria optimization defines the dominance relation by the standard vector inequality. The theory
of majorization includes the results which allow us to express the relation of fair (equitable) dominance as a
vector inequality on the cumulative ordered outcomes (see [29] and [33]).

194

Chia-Hung Wang and Hsing Luh

Recall that the achievement function

P fi is a concave function. According to this definition, we know that the function
fi is Schur-concave. Next, we consider a generic
resource allocation problem defined as an optimization problem withm objective functions
fi (x):
max{ f (x) : x Q },
(16)
where f (x) is a vector-function that maps the decision space Rm into the criterion space
Rm , Q denotes the feasible set, and x Rm denotes the vector of decision variables.
In the following, we will introduce the concept of fairness by using the fair aggregation
function (see [19], [21], [33], [34]). Typical solution concepts for multiple criteria problems
are defined by aggregation functions g : Rm R to be maximized. Thus, (16) implies
max{g(f (x)) : x Q}

(17)

The simplest aggregation functions commonly used for the multiple criteria problem (16)
are defined as the sum of outcomes
X
fi (x),
(18)
g(f (x)) =
iI

or the worst outcome

g(f (x)) = min fi (x)
iI

(19)

An aggregation (17) is fair if it is defined by a strictly increasing and strictly Schur-concave

function g.
Definition 12 An aggregation function g satisfying all the following requirements (20), (21)
and (22), we call the corresponding problem (17) a fair aggregation of problem (16). For
all i I = {1, 2, . . . , m}
g(f1 (x), . . . , fi1 (x), fi0 (x), fi+1 (x), . . . , fm (x)) < g(f1 (x), . . . , fm (x)),

(20)

whenever fi0 (x) < fi (x). For any permutation of I,

g(f(1) (x), f(2) (x), . . . , f(m) (x)) = g(f1 (x), f2 (x), . . . , fm (x))

(21)

For any 0 <  < zi0 zi00 , we have

g(f1 (x), . . . , fi0 (x) , . . . , fi00 (x) + , . . . , fm (x)) > g(f1 (x), f2 (x), . . . , fm (x)) (22)
P
We know that
fi is a fair aggregation function according to this definition. Every
optimal solution to the fair aggregation (17) of a resource allocation problem (16) defines
some fair allocation scheme. In order to guarantee the consistency of the aggregated problem (17) with the maximization all individual objective functions in the original multiple
criteria problem, the aggregation function must be strictly increasing with respect to every
coordinate, i.e. (20). In order to guarantee the fairness of the solution concept, the aggregation function must be additionally symmetric (impartial), i.e. (21). Symmetric functions
satisfying the requirement (22), are called strictly Schur-concave functions. Next, we have
the following two theorems taken from [33].

Mathematical Models of QoS Management for Communication Networks

195

Theorem 13 For a strictly concave, increasing function fi : R R, the function

g(f (x)) =

fi (x)

(23)

iI

is a strictly monotonic and strictly Schur-concave function.

Theorem 14 For a strictly concave, increasing function fi : R R, the optimal solution
of the problem
X
fi (x) : x Q}
(24)
max{
iI

is a fair solution for resource allocation problem (16).

Blocking Probability with Predetermined Optimal Solutions

The blocking probability is an important performance measurement of network system. In

our situation here, the blocking is due to the failure of setting up the number of end-toend paths, Ki , for each class i. In this section, we study the blocking probability of an
end-to-end transmission system with predetermined optimal solutions, including optimal
bandwidth allocation, xi , and Ki optimal end-to-end paths, pi,j . At the source node o,
connections arrive at random times to enter the core network.
The predetermined number, Ki is used to denote the limit on the number of connections
in class i. A new connection in class i can not enter the source node o and is lost when all
Ki end-to-end path are busy. That is, for each class i, a connection gets dropped on its
arrival when the number of connections occupying the end-to-end paths equalsKi .
Otherwise, it will be routed through an end-to-end path pi,j with allocated bandwidth

xi predetermined by MILP. The principal quantity of interest is the blocking probability

of different QoS classes, that is, the steady-state probability that allKi end-to-end paths in
class i are busy. Our objective is to estimate these blocking probabilities.
Assume that connections in class i arrive to the source node o in accordance with independent Poisson processes at rate i , but the packet sizes have a general distribution G with
mean i . For each class i, we define i = xi /i , where xi is the optimal bandwidth allocation for each connection of class i. The average service time corresponds to the packet
transmission time and is equal to average pack size divided by bandwidth. That is,
i
1
= .
i
xi

(25)

Hence, for each class i, the service times of connections occupying the end-to-end paths
have a general distribution G with mean 1/i = i /xi . Suppose that connections occupy
the end-to-end paths in the order they arrive and that packet sizes, which need to be transmitted from o to d, are identically distributed, mutually independent, and independent of
the inter-arrival times.
Under these assumptions, we analyze this end-to-end transmission system as
M/G/Ki /Ki loss systems [7], that is, Poisson arrivals, general service, Ki end-to-end paths

196

Chia-Hung Wang and Hsing Luh

with identical bandwidth allocation xi , and no waiting space. For each class i, we can
derive the steady-state occupancy probabilities from Erlang loss system [39]. We have



i ni
i ni
i
xi
i
= Pi (0)
Pi (ni ) = Pi (0)
ni !
ni !


where ni = 1, 2, . . . , Ki , i = 1, 2, . . . , m.
P i
Solving for Pi (0) in the equation K
ni =0 Pi (ni ) = 1, we obtain that

 j 1

j 1
Ki
Ki
X
X

1
1

i
i i
=

Pi (0) =
j! i
j!
xi
j=0

(26)

j=0

and then

Pi (ni ) =

 j 1
ni X
Ki

1
i

j! i
j=0



ni X
j 1
Ki
1 i i
1 i i

ni !
xi
j!
xi
1
ni !

i
i

j=0

for ni = 1, 2, . . . , Ki . For each class i, the blocking probability is given as follows:

1
Pi (Ki ) =
Ki !

i i
xi


Ki X
j 1
Ki

i i

j!
xi

(27)

j=0

under conditions of Poisson arrival, general service time, and onlyKi end-to-end paths. Eq.
(27) is referred to as Erlangs loss formula [39]. If we denote the traffic load
i =

i i
,
Ki xi

(28)

then Eq. (27) can be rewritten

Pi (Ki ) =

1
Ki
(Ki i )Ki X
(Ki i )j
Ki !
j!
j=0

(Ki i )Ki
[exp(Ki i ) Ri (Ki )]1
Ki !

(29)

where Ri (Ki ) is the Ki th-degree Taylor remainder term of exp(Ki i ) [3]. It is valid for all
service distributions and only depends on the traffic load, i . From Taylors formula with
remainder, we have the following results.

Mathematical Models of QoS Management for Communication Networks

197

Proposition 15 There exists a real number i (0, Ki i ), such that exp(Ki i ) =

Ki
X
(Ki i )j
+ Ri (Ki ) as
j!
j=0

Ri (Ki ) =

exp(i )(Ki i )Ki +1

.
(Ki + 1)!

Moreover,
lim Ri (Ki ) = 0.

Ki

Harel [16] proved that the fraction of customers lost in the M/G/K/K system is convex
in the arrival rate, if the traffic intensity is below some and concave if the traffic intensity
is greater than . Some convexity properties of the blocking probability (29) are listed
below. These results are consistent with convexity properties showed by Harel [16].
Proposition 16 For each Ki , there exists a i such that for all i < (>)i , the blocking
probability (29) is strictly convex (concave) ini .
Proposition 17 The blocking probability (29) is strictly decreasing and strictly convex in
xi /i , provided i and Ki fixed.

10

Conclusions and Future Research

A backbone link cannot be sized in arbitrarily small portions, but are built up from fixedsized modules. Furthermore, even demand volume is usually restricted to allocation in
discrete portions. As a consequence, the resulting mathematical programmes have to be
solved over a partly integral-valued domain, rendering so-called mixed-integer programming problems (MIP problems). Large MIP problems are known to be very difficult to
handle and often require heuristics, in order to be solvable in a reasonable amount of time.
However, accounting for the requirement of modularity is a most challenging problem and
the research community of network design as well as the industry are addressing research
in this direction.
To summarize, it is our conviction that future large-scale networks, particularly the next
generation Internet, will require a large portion of sophisticated network engineering, in
order to meet the demands of service guarantees, connection reliability, cost efficiency and
structured sharing of transmission media. It is reasonable to assume that the sharing of
media will be based on fairness principles, possibly integrated in some sort of prioritization scheme, and that the network robustness will become the network design fundament.
Since the next generation Internet will be a large and indeed a very complex network of
networks (internetwork), a spirit of generality has to characterize the derivation of the new
engineering methodologies.
A future aim will also be to take the common requirement of QoS guarantees into
account. In the network optimization context such approaches are referred to as demandoriented. Our target is to make efforts in this direction, by combination of the novel techniques of network calculus along with the classical theory of optimization.

198

Chia-Hung Wang and Hsing Luh

Acknowledgments.
This research was supported in part by the National Science Council, Taiwan, R.O.C., under
NSC 95-2221-E-004-007.

References
[1] Alkahtani, A. M. S., Woodward, M. E., and Al-Begain, K., Prioritised best effort routing with four quality of service metrics applying the concept of the analytic hierarchy
process, Computers & Operations Research 33, 559580 (2006).
[2] Andrade, R., Lisser, A., Maculan, N., and Plateau, G., B&B frameworks for the capacity expansion of high speed telecommunication networks under uncertainty,Annals of
Operations Research, 140, 4965 (2005)
[3] Apostol, T. M., Mathematical Analysis, 2nd ed., Addison-Wesley Publishing Company, Inc. (1974).
[4] Atov, I., Tran, H. T., and Harris, R. J., OPQR-G: Algorithm for efficient QoS partition
and routing in multiservice IP Networks, Computer Communications, 28, 19871996
(2005).
[5] Bai, Y. and Ito, M. R., Class-based packet scheduling to improve QoS for IP video,
Telecommunication Systems, 29 (1), 4760 (2005).
[6] Berger, A. W. and Kogan, Y., Dimensioning bandwidth for elastic traffic in high-speed
data networks, IEEE/ACM Transactions on Networking, 8 (5), 643654 (2000).
[7] Bertsekas, D. and Gallager, R., Data Networks, 2nd ed., Prentice Hall, New Jersey
(1992).
[8] Bertsekas, D.P., Network Optimization, Athena Scientific (1998).
[9] Correa, J. R., Schulz, A. S., and Stier-Moses, N. E., Fast, Fair, and Efficient Flows in
Networks, Operations Research, 55 (2), 215225 (2007).
[10] Djukic, P., Valaee, S., Reliable packet transmissions in multipath routed wireless networks, IEEE Transactions on Mobile Computing, 5 (5), 548559 (2006).
[11] Ferng, H. W. and Chang, J. F., The departure process of discrete-time queueing systems with Markovian type inputs, Queueing Systems, 36, 201220 (2000).
[12] Fischer, M. J., Masi, D. M. B., Gross, D., and Shortle, J. F., Loss systems with heavytailed arrivals, The Telecommunications Review, 15, 9599 (2004).
[13] Fortz, B., Rexford, J., Thorup, M., Traffic engineering with traditional IP routing
protocols, IEEE Communications Magazine, 40 (10), 118124 (2002).
[14] Gozdecki, J., Jajszczyk, A., Stankiewicz, R., Quality of service terminology in IP
networks, IEEE Communications Magazine, 41 (3), 153159 (2003).

Mathematical Models of QoS Management for Communication Networks

199

[15] Guerin, R. A., Orda, A., QoS routing in networks with inaccurate information: Theory
and algorithms, IEEE/ACM Transactions on Networking, 7 (3), 350364 (1999).
[16] Harel, A., Convexity properties of the Erlang loss formula, Operations Research, 38
(3), 499505 (1990).
[17] van Hoesel, S., Optimization in telecommunication networks, Statistica Neerlandica,
59 (2), 180205 (2005).
[18] Johari, R., Tan, D. K. H., End-to-end congestion control for the Internet: Delays and
stability, IEEE/ACM Transactions on Networking, 9 (6), 818832 (2001).
[19] Kelly, F. P., Charging and rate control for elastic traffic, European Transactions on
Telecommunications, 8, 3337 (1997).
[20] Kelly, F. P., Maulloo, A. K., Tan, D. K. H., Rate control for communication networks: Shadow prices, proportional fairness and stability, Journal of the Operational
Research Society, 49, 237252 (1998).
[21] Kelly, F. P., Fairness and stability of end-to-end congestion control,European Journal
of Control, 9, 159176 (2003).
[22] Kelly, F. P., Williams, R. J., Fluid model for a network operating under a fair
bandwidth-sharing policy, The Annals of Applied Probability, 14 (3), 10551083
(2004).
[23] Lien, Y. N., Jang, H. C., Tsai, T. C., and Luh, H., BBQ: A QoS Management Infrastructure for All-IP Networks, Communications of Institute of Information and Computing Machinery: Mobile Communications and Wireless Networks, 7 (1), 89115
(2004).
[24] Loudni, S., Boizumault, P., and David, P., On-line resources allocation for ATM networks with rerouting, Computers & Operations Research, 33, 28912917 (2006).
[25] Low, S. H., Lapsley, D. E., Optimization flow controlPart I: Basic algorithm and
convergence, IEEE/ACM Transactions on Networking, 7 (6), 861874 (1999).
[26] Low, S. H., A duality model of TCP and queue management algorithms, IEEE/ACM
Transactions on Networking, 11 (4), 525536 (2003).
[27] Luh, H. and Wang, C. H., Mathematical Models of Pareto Optimal Path Selection on
All-IP Networks, Proceedings of The First Sino-International Symposium on Probability, Statistics, and Quantitative Management, 185197 (2004).
[28] Luh, H., Wang, C. H., Proportional bandwidth allocation for unicasting in All-IP networks, Proceedings of the 2nd Sino-International Symposium on Probability, Statistics, and Quantitative Management, 111130 (2005).
[29] Marshall, A. W., Olkin, I., Inequalties: Theory of Majorization and Its Applications,
New York: Academic Press (1979).

200

Chia-Hung Wang and Hsing Luh

[30] Massouli
e , L., Roberts, J., Bandwidth sharing and admission control for elastic traffic,
Telecommunication Systems, 15, 185201 (2000).
[31] Massouli
e , L. and Roberts, J., Bandwidth sharing: objectives and algorithms,
IEEE/ACM Transactions on Networking, 10 (3), 320328 (2002).
[32] Mo, J., Walrand, J., Fair end-to-end window-based congestion control, IEEE/ACM
Transactions on Networking, 8 (5), 556567 (2000).
[33] Muller, Stoyan, A., D., Comparison Methods for Stochastic Models and Risks, Wiley,
Chichester (2002).

[34] Ogryczak, W., Sliwi

nski, T., Wierzbicki, A., Fair resource allocation schemes and network dimensioning problems, Journal of Telecommunications and Information Technology, 3, 3442 (2003).
[35] Orda, A., Routing with end-to-end QoS guarantees in broadband networks,
IEEE/ACM Transactions on Networking, 7 (3), 365374 (1999).
[36] Orda, A., Sprintson, A., Precomputation schemes for QoS routing, IEEE/ACM Transactions on Networking, 11 (4), 578-591 (2003).
[37] Ouaja, W. and Richards, B., A hybrid multicommodity routing algorithm for traffic
engineering, Networks, 43 (3), 125140 (2004).
[38] Romero, C., A general structure of achievement function for a goal programming
model, European Journal of Operational Research, 153, 675686 (2004).
[39] Ross, S. M., Stochastic Processes, New York: Wiley (1983).
[40] Stallings, W., Data and Computer Communications, Prentice Hall (2004).
[41] The UMTS Forun, Enabling UMTS/Third Generation Services and Applications,
UMTS Forun Report, 11 (2000).
[42] Thomas, P., Teneketzis, D., Mackie-Mason, J. K., A Market-Based Approach to Optimal Resource Allocation in Integrated-Services Connection-Oriented Networks, Operations Research, 50 (4), 603616 (2002).
[43] Wang, C. H., Luh, H., Network Dimensioning Problems of Applying Achievement
Functions, Lecture Notes in Operations Research, 6, 3559 (2006).
[44] Wang, C. H., Luh, H., A Two-Phase Scheme for Fair Bandwidth Allocation and QoS
Routing in All-IP Networks, Proceedings of the Third Sino-International Symposium
on Probability, Statistics, and Quantitative Management, 4963 (2006).
[45] Wang, C. H., Luh, H., A Precomputation-Based Scheme for QoS Routing and Fair
Bandwidth Allocation, The 13th annual IEEE International Conference on High Performance Computing (HiPC 2006), Lecture Notes in Computer Science, 4297, 595
606 (2006).

Mathematical Models of QoS Management for Communication Networks

201

[46] Wang, C. H., Yue, W., Luh, H., Performance Evaluation of Predetermined Bandwidth
Allocation for Heterogeneous Networks, Technical Report of IEICE, 107 (6), 3742
(2007).
[47] Wang, C. H., Luh, H., Two-Phase Modeling of QoS Routing in Communication Networks. International Workshop on Performance Modeling and Evaluation in Computer and Telecommunication Networks (PMECT07) in conjunction with IEEE ICCCN2007, Honolulu, Hawaii, USA, August 16, 2007.
[48] Wang, C. H., Luh, H., Blocking Probabilities of QoS Routing in IP Networks with
Multiple Classes, The Second Asia-Pacific Symposium on Queueing Theory and Network Applications (QTNA2007), International Conference Center, Kobe, Japan, August 1-4, 2007.
[49] Wang, C. H., Luh, H., A Fair QoS Scheme for Bandwidth Allocation by
Precomputation-based Approach, International Journal of Information and Management Sciences, 19 (3), accepted for publication (2008).
[50] Wu, H., Jia, X., He, Y., Huang, C., Bandwidth-guaranteed QoS routing of multiple
parallel paths in CDMA/TDMA ad hoc wireless networks, International Journal of
Communication Systems, 18, 803816 (2005).
[51] Xiao, X., Ni, L. M., Internet QoS: A big picture, IEEE Network, 13 (2), 818 (1999).
[52] Ye, H. Q., Qu, J., Stability of data networks: Stationary and bursty models. Operations Research, 53 (1), 107125 (2005).

* A version of this chapter was also published in Leading-Edge Applied Mathematical

Modeling Research, edited by Matas P. lvarez published by Nova Science Publishers, Inc. It
was submitted for appropriate modifications in an effort to encourage wider dissemination
of research.

In: Mathematics and Mathematical Logic: New Research

ISBN 978-1-60692-862-2
c 2010 Nova Science Publishers, Inc.
Editors: Peter Milosav and Irene Ercegovaca

Chapter 8

R EVERSIBLE L OGIC
Alexis De Vos
Department of Electronics and Information Systems, University of Ghent,
Ghent, Belgium

Abstract
Reversible logic circuits are beneficial to both classical and quantum computer design. Present-day logic building-blocks (like OR gates and NAND gates) are logically
irreversible and therefore cannot be used for designing reversible computers. Thus
reversible computation needs an appropriate design methodology. In contrast to conventional digital logic circuits, reversible logic circuits (of a same logic width w) form
a mathematical group. The reversible circuits of width w form a group isomorphic to
the symmetric group S2w . Its Young subgroups allow systematic and efficient synthesis of an arbitrary reversible circuit. We can choose either a left coset, a right coset,
or a double coset approach. The optimal design is reminiscent of the so-called banyan
networks of telecommunication.
As an illustration, three experimental prototypes (in c-MOS chip technology) of
reversible computing devices are presented. Special care has been taken to avoid as
much as possible the appearance of garbage bits. The examples illustrate how, in a
near future, reversible computers will outperform conventional computers, in terms of
power dissipation and heat generation.

1.

Introduction

Reversible computing [1] [2] is useful both in low-power classical computing [3] [4] and in
quantum computing [5]. According to the Landauer theorem, the only way to make classical digital computing lossless, is by taking care that, at each stage of the computation, no
information is lost. Indeed, each bit of information that is thrown away, causes the generation of a quantum of heat (with magnitude kT log(2), where k is the Boltzmann constant
and T is the temperature of the computer hardware). This can be easily seen, using thermodynamics. For instance, resetting a bit to 0 makes one forget about its original contents.
So, an unknown bit has become a fully specified one and thus the (macroscopic) entropy

204

Alexis De Vos

adding

computer

adding
computer

Figure 1. A logically irreversible pocket calculator.

of the machine has diminished by k log(2). Because of the second law of thermodynamics, at least an amount k log(2) of microscopic entropy has to be created simultaneously.
Therefore an amount kT log(2) of heat is released into the environment [6] [7] [8] [9]. As
stated above, reversible computing also has its uses in quantum computing, as in the latter
technology only reversible operations are possible. In the present review paper however,
we will only deal with classical circuits. Thus, if we want to avoid any heat generation in
these circuits, we have to avoid any loss of information during the computational process.
This means we have to construct a logically reversible computer.
What is a reversible computer? Figure 1 gives a counterexample. It is a small calculator, designed for one particular task: the computation of the sum of two numbers. This
computation is logically irreversible, because, if we have forgotten the input values (i.e. 3
and 1), knowledge of the output value (i.e. 4) is not sufficient to recover what the inputs
have been. Indeed, 4 could have been equally well 2 + 2 or 4 + 0 or 0 + 4 or ...
Figure 2 gives an actual example of a reversible computer. Again we have a pocket
calculator, designed for one particular task: the computation of the sum and the difference
of two numbers. This computation is reversible, because, if we have forgotten the input
values (3 and 1), knowledge of the output values (4 and 2) is sufficient to recover what the
inputs have been. Indeed, if A and B designate the two input numbers and P and Q the two
output numbers, then
P

= A+B

Q = AB
suffices to calculate backwards:
A =

1
(P + Q)
2

B =

1
(P Q) .
2

Reversible Logic

205

adding & subtracting

computer

adding & subtracting

computer

Figure 2. A logically reversible pocket calculator.

Thus, the outputs contain enough information to reconstruct the inputs. In other words: the
outputs contain the same information as the inputs.
In the present paper, we will demonstrate the application of group theory to the detailed
design of reversible classical hardware. Reversible logic circuits distinguish themselves
from arbitrary logic circuits by two properties:
the number of output bits always equals the number of input bits and
for each pair of different input words, the two corresponding output words are different.
For instance, it is clear that an AND gate (thruth table in Table 1a) is not reversible, as
it has only one output bit, but two input bits and
for three different input words (i.e. for 00, 01, and 10), the three corresponding output
words are equal.
Analogously, neither the OR gate (Table 1b), nor the NAND gate, nor the NOR gate are
reversible.
Table 2a, on the other hand, gives an example of a reversible circuit. Here, the number
of inputs equals the number of outputs, i.e. two. This number is called the width w of the
reversible circuit r. The table gives all possible input words (A, B). We see how all the
corresponding output words (P, Q) are different. Thus the four (P, Q) words are merely a
permutation of the four (A, B) words. This particular truth table may be replaced by a set
of w Boolean equations:
P (A, B) = A B
Q(A, B) = A .
We will use the following short-hand notations for the basic Boolean functions:
X = NOT X

206

Alexis De Vos

Table 1. Truth table of two irreversible logic circuits: (a) the AND gate and (b) the OR
gate.

AB

AB

00
01
10
11

0
0
0
1

00
01
10
11

0
1
1
1

(a)
XY

(b)
= X AND Y

X+Y

= X OR Y

XY

= X XOR Y ,

where XOR is an abbreviation of EXCLUSIVE OR.

In contrast to arbitrary logic circuits, reversible logic circuits form a group. For a group,
we need a set (S) as well as an operation under which each pair (x, y) of elements of the set
corresponds to a third element of the set (written xy). We are allowed to speak of a group,
if the operation is associative, i.e. if x(yz) equals (xy)z (therefore denoted simply
xyz),
if the set S contains an identity element i, i.e. if xi = ix = x for each x S, and
if for each element x of S there exists an inverse element x1 (such that xx1 =
x1 x = i).
In our case, the operation working on two circuits is the cascading of the two circuits.
Table 2b gives the truth table of the identity gate i and Table 2c gives r1 , i.e. the inverse of
circuit r. The reader will easily verify that not only the cascade rr1 , but also the cascade
r1 r equals i.
In the present paper, we will apply different techniques from group theory to the synthesis and the analysis of reversible circuits. In Sections 2 and 3, we discuss an important
group theoretical tool, i.e. the subgroup concept. Sections 4, 5, and 6 discuss cosets and
double cosets and how to use them for synthesis of a reversible logic circuit. In Section 7
we demonstrate how to change a logically irreversible gate into a reversible one. Finally, in
Section 8, we discuss the relationship with some experimental prototypes.

2.

Group Theory

All reversible circuits of the same width form a group. If we denote by w the width, then
the truth table of an arbitrary reversible circuit has 2w rows. As all output words have to be

Reversible Logic

207

Table 2. Truth table of three reversible logic circuits of width 2: (a) an arbitrary
reversible circuit r, (b) the identity gate i, and (c) the inverse r1 of r.

AB

PQ

AB

PQ

AB

PQ

00
01
10
11

00
10
11
01

00
01
10
11

00
01
10
11

00
01
10
11

00
11
01
10

(a)

(b)

(c)

different, they can merely be a repetition of the input words in a different order. In other
words: the 2w output words are a permutation of the 2w input words. There exist (2w )! ways
to permute 2w objects. Therefore there exist exactly (2w )! different reversible logic circuits
of width w. The number 2w is called the degree of the group; the number (2w )! is called the
order of the group. The group is isomorphic to a group well-known by mathematicians: the
symmetric group S2w . For further reading in the area of symmetric groups in particular, and
groups, subgroups, cosets, and double cosets in general, the reader is refered to appropriate
textbooks [10] [11].
The symmetric group has a wealth of properties. For example, it has a lot of subgroups,
of which most have been studied in detail. Some of these subgroups naturally make their
appearance in the study of reversible computing. An example is the subgroup of conservative logic circuits, studied in detail by Fredkin and Toffoli [12]. Table 3a gives an example.
In each of its rows, the output (P, Q, R, ...) contains a number of 1s equal to the number
of 1s in the corresponding input (A, B, C, ...). Table 4 gives C(w), the number of conservative logic circuits of width w. An even more important subgroup is the subgroup of linear
reversible circuits. Linear reversible circuits have been studied in detail by Patel et al. [13].
A logic circuit is linear iff each of its outputs P , Q, ... is a linear function of the inputs A,
B, ... In its turn, a linear function is defined as follows. A Boolean function f (A, B, ...)
is linear iff its ReedMuller expansion [14] only contains terms of degree 0 and terms of
degree 1. The reversible circuit of Table 3a is not linear. Indeed it can be written as a set of
three Boolean equations:
P

= B AB AC

Q = A
R = C AB AC .
Whereas the function Q(A, B, C) is linear, the function P (A, B, C) is clearly not (its
ReedMuller expansion containing two terms of second degree). Table 3b is an example of

208

Alexis De Vos

Table 3. Truth table of three reversible logic circuits of width 3: (a) a conservative
circuit, (b) a linear circuit, and (c) an exchanging circuit.
ABC

P QR

ABC

P QR

ABC

P QR

000
001
010
011
100
101
110
111

000
001
100
101
010
110
011
111

000
001
010
011
100
101
110
111

100
000
001
101
111
011
010
110

000
001
010
011
100
101
110
111

000
001
100
101
010
011
110
111

(a)

(b)

(c)

a linear circuit:
P

= 1B C

Q = A
R = AB .
Based on pioneering work by Kerntopf [15], De Vos and Storme [16] have proved that
an arbitrary Boolean function can be synthesized by a (loop-free and fanout-free) wiring
of a finite number of identical reversible gates, provided the gate is not linear. In other
words: all non-linear reversible circuits can be used as a universal building block. Thus the
linear reversible circuits constitute the weak ones. Indeed, any wiring of linear circuits
(be they reversible or not, be they identical or not) can yield only linear Boolean functions
at its outputs. The linear reversible circuits form a group isomorphic to what is called in
mathematics the affine general linear group AGL(w, 2). Its order equals 2(w+1)w/2 w!2,
where w!2 is the bifactorial of w, the q-factorial being a generalization of the ordinary
factorial w! = w!1:
w!q = 1(1 + q)(1 + q + q 2 )...(1 + q + ... + q w1 ) .
Table 4 gives the number of different linear reversible circuits. We see that (at least for
w > 2) a vast majority of the reversible circuits are non-linear and thus can act as universal
gates.
Now we descend the hierarchy of subgroups one step, by imposing that each of the
outputs equals one of the inputs. Table 3c is such a circuit:
P

= B

Q = A
R = C.

Reversible Logic

209

Table 4. The number R of different reversible circuits, the number C of different

conservative reversible circuits, the number L of different linear reversible circuits,
and the number E of different exchanging reversible circuits, as a function of the
circuit width w.
w
1
2
3
4

R
2
24
40,320
20,922,789,888,000

C
1
2
36
414,720

2
24
1,344
322,560

1
2
6
24

Such circuits are called exchangers (a.k.a. SWAP gates). They form a subgroup isomorphic
to Sw of order w!. Also this number is given in Table 4. Finally, we can impose that each
of the outputs equals the corresponding input:
P

= A

Q = B
R = C.
This results in the trivial subgroup I of order 1, merely consisting of one circuit, i.e. the
identity gate i.
We have thus constructed a chain of subgroups:
S2w AGL(w, 2) Sw I ,
with subsequent orders
(2w )! > 2(w+1)w/2 w!2 > w! > 1 ,
where we have assumed w > 1. Here, the symbol reads is proper supergroup of. For
the example w = 3, this becomes:
S8 AGL(3, 2) S3 I ,
with subsequent orders
40, 320 > 1, 344 > 6 > 1 .

3.

Control Gates

Besides using the letters A, B, C, ... for the input bits and P , Q, R, ... for the output bits,
we will also use A1 , A2, ..., Aw for the inputs and P1 , P2, ..., Pw for the outputs, because
sometimes this is more convenient.

210

Alexis De Vos

Table 5. Truth table of three reversible logic circuits of width 3: (a) an arbitrary
circuit, (b) a controlled NOT gate, and (c) a twin circuit.
A1 A2 A3

P1 P2 P3

A1 A2 A3

P1 P2 P3

A1 A2A3

P1 P2 P3

000
001
010
011
100
101
110
111

100
101
110
000
111
011
010
001

000
001
010
011
100
101
110
111

001
000
011
010
101
100
110
111

000
001
010
011
100
101
110
111

001
011
010
000
100
101
111
110

(a)

(b)

(c)

We now define a special class of reversible logic circuits, called control gates, by their
relationship between the outputs P1 , P2, ..., Pw and the inputs A1 , A2, ..., Aw. In a control
gate, we always have P1 = A1, P2 = A2 , ..., Pu = Au , where u is an integer obeying
0 < u < w. The other outputs, i.e. Pu+1 , Pu+2 , ..., and Pw , are controlled by means of
some Boolean function f of the u inputs A1, A2 , ..., Au :
If f (A1 , A2, ..., Au) = 0, then we additionally have Pu+1 = Au+1 , Pu+2 = Au+2 ,
..., Pw = Aw .
If however f (A1 , A2, ..., Au) = 1, then the values of Pu+1 , Pu+2 , ..., Pw follow
from the values of Au+1 , Au+2 , ..., Aw by the application of a reversible circuit g
of width v = w u.
Thus: if f = 0, then we apply the v-bit follower to Au+1 , Au+2 , ..., Aw, else we apply the vbit circuit g to them. We call A1 , A2, ..., Au the controlling bits and Au+1 , Au+2 , ..., Au+v
the controlled bits. Whereas w is the width, u is the controlling width and v is the controlled
width. We call f the control function and g the controlled circuit.
u
There exist 22 Boolean functions f of u binary variables. Together with the XOR
operation, they form a group isomorphic to the direct product group S2 S2 ... S2 =
u
S22 . Therefore, the control gates with a same controlled gate g form a group isomorphic to
u
u
S22 . Its order is 22 .
We also assign an icon to the control gate: Figure 3b (where u = 4 and v = 2, such that
w = 6). Note that each of the controlling bits is labelled by a small square. However, in the
special case where the controlling function f (A1 , A2, ..., Au) is an AND of some controlling bits whether inverted or not, then circular tags are used: a filled circle if the variable
is not inverted, an open circle if the variable is inverted. Figure 3c shows an example:
f (A1 , A2, A3, A4) = A1 A3 A4.
We consider two special cases in detail:

Reversible Logic

211

Figure 3. Icons of reversible circuits of width w = 6: (a) arbitrary circuit, (b) control gate
with arbitrary control function f , and (c) control gate with AND control function.

If v = 1, then only two possibilities exist for the controlled circuit g: either it is
the trivial follower or it is the inverter. We only have to consider the latter choice.
Then the controlled gate is represented by a cross: see Figure 4a. We call such
gates controlled NOTs. An example is given in Table 5b. Its controlling function is
f (A1, A2) = A1 + A2 . Note that, in the truth table of a controlled NOT, the first
two rows are either permuted or not, the second two rows are either permuted or not,
etcetera. Therefore the controlled NOTs form a subgroup isomorphic to S2 S2 ...
w1
[17]. If the gates control function f is an AND function, we call it
S2 of order 22
a TOFFOLI gate (e.g. Figure 4b with control function f = A1 A3 A4 ).

If v = 2, then 4! = 24 possibilities exist for the controlled circuit g. The most interesting case is where g is a SWAP gate. Then the controlled gate is represented by a
cross-over: see Figure 4c. We call such gates controlled SWAPs. They form a subw2
group of order 22 . If additionally the gates control function is an AND function,
we call it a FREDKIN gate (e.g. Figure 4d with control function f = A2 A3 ).

Note that the functionallity of the controlled NOT gate can be written as a set of w Boolean
equations:

P1 = A1
P2 = A2
...
Pw1 = Aw1
Pw = f (A1 , A2, ..., Aw1) Aw

(1)

212

Alexis De Vos

Figure 4. Icons of control gates: (a) arbitrary controlled NOT gate, (b) specific TOFFOLI
gate, (c) arbitrary controlled SWAP gate, and (d) specific FREDKIN gate.
and so can the functionallity of the controlled SWAP gate:
P1 = A1
P2 = A2
...
Pw2 = Aw2
Pw1 = f (A1, A2, ..., Aw2)(Aw1 Aw ) Aw1
Pw = f (A1, A2, ..., Aw2)(Aw1 Aw ) Aw .

4.

(2)

Cosets

Subgroups are at the origin of a second powerful tool in group theory: cosets. For convenience, we will, as much as possible, adopt the following style: a group is represented by a
bold-faced capital letter; its order is denoted by an upper-case letter, whereas an element of
the group is represented by a lower-case letter. E.g. a and b might be two elements among
the C elements of some group C. Already in Table 4, we used upper-case letters (i.e. R, C,
L, and E) for group orders.
If H (with order H) is a subgroup of the group G (with order G), then H partitions G
G
classes, all of the same size H. These equipartition classes are called cosets. We
into H
distinguish left cosets and right cosets.
The left coset of the element a of G is defined as all elements of G which can be written
as a cascade ba, where b is an arbitrary element of H. Such left coset forms an equipartition
class because of the following property: if c is member of the left coset of a, then a is
member of the left coset of c. Right cosets are defined in an analogous way. Note that H
itself is one of the left cosets of G, as well as one of its right cosets.
What is the reason of defining cosets? They are very handy in synthesis. Assume we
want to make an arbitrary element of the group G in hardware. Instead of solving this
problem for each of the G cases, we only synthesize the H circuits b of H and a single
G
1). If we can make each of these
representative ri of each other left coset (1 i H

Reversible Logic

213

G
H+H
1 circuits, we can make all the others by merely making a short cascade bri. If we
G
1 is much smaller than G.
cleverly choose the subgroup H, we can guarantee that H + H
G
We call the set of H + H 1 building-blocks the library for synthesizing the G circuits
of G.
The clever choice of the subgroup H of G is the challenge of the designer. He/she can
G
1)/dH = 0, what leads to
e.g. aim for minimizing the size of the library: d(H + H

H = G. Of course, seldom G will have a subgroup of order G. In most cases, G

is not even an integer number, but an irrational number instead.
Then, the designer has to

look for a subgroupwith order in the neighbourhood of G. In passing, we remark that

the condition H = G can be rewritten as

log(G)
=2.
log(H)
Ratios of logarithms of sizes will play more important parts in our story. See e.g. Appendix B.
Maslov and Dueck [18] present a method for synthesizing an arbitrary reversible circuit of width three. As a subgroup H of the group G = S8 , they propose all circuits with
output (P, Q, R) equal (0,0,0) in case of the input (A, B, C) = (0, 0, 0). This subgroup
is isomorphic to S7 . Thus the supergroup has order G = 8! = 40, 320, whereas the subgroup has order H = 7! = 5, 040. The subgroup partitions the supergroup into 8 cosets.
Interesting is the fact, that the procedure can be repeated: for designing each of the 5,040
members of S7 , Maslov and Dueck choose a subgroup of S7 . They choose all reversible
circuits where (P, Q, R) equals (0,0,0) in case (A, B, C) = (0, 0, 0) and equals (0,0,1) in
case (A, B, C) = (0, 0, 1). This is a subgroup isomorphic to S6 of order 6! = 720, which
partitions S7 into seven cosets. Etcetera. Figure 5a illustrates one step of the procedure:
the 24 elements of S4 are fabricated by means of the 6 elements of its subgroup S3 plus
the representatives of the 3 other cosets in which S4 is partitioned by S3 . Thus Maslov and
Dueck apply the following chain of subgroups:
S8 S7 S6 S5 S4 S3 S2 S1 = I ,

(3)

with subsequent orders

40, 320 > 5, 040 > 720 > 120 > 24 > 6 > 2 > 1 .

(4)

They need, for synthesizing all 40,320 members of S8 , a library of only (7+6+...+1)+1 =
29 elements (the identity gate included). For an arbitrary circuit width w, synthesis of all
(2w )! members of S2w needs a library of 22w1 2w1 + 1 elements.
Van Rentergem et al. [19] [20] also present a coset method for synthesis, however based
the following subgroup H: all circuits from G = S2w possessing the property P1 = A1. Such
circuits consist of the cascade of two control gates with u = 1 and v = w 1. See Figure 6.
If A1 = 0, then g is applied to A2 , A3 , ..., and Aw , else (thus if A1 = 1) h is applied. We
will call such circuit a twin circuit. An example is given in Table 5c. Note that, in the truth
table of a twin circuit, the upper 2w1 rows are permuted among themselves and so are
the lower 2w1 rows. Therefore the twin circuits form a subgroup isomorphic to S2w1
S2w1 . The twin circuits of width 3 form a group isomorphic to S4 S4 = S24 of order

214

Alexis De Vos
S4

S3

S4

S 2 x S2

Figure 5. The symmetric group S4 partitioned (a) as the four left cosets of S3 and (b) as the
three double cosets of S2 S2 .
Note: the dots depict the elements of S4 ; the bold-faced dots depict the elements of the subgroup and the
representatives of the (double) cosets.

Figure 6. A twin circuit of width 4: if A1 = 0 then apply g else apply h.

(4!)2 = 576. The subgroup S24 partitions its supergroup S8 into 70 cosets. Subsequently,
the members of S4 are partitioned into six cosets by use of its subgroup S22 , etcetera. Thus,
finally, Van Rentergem et al. apply the following chain of subgroups:
S8 S24 S42 S81 = I ,

(5)

40, 320 > 576 > 16 > 1 .

(6)

with subsequent orders

If we denote 2w by n, then Dueck et al. apply the subgroup Sn1 of the group Sn ,
whereas Van Rentergem et al. apply the subgroup S n2 S n2 . We note that both the group
Sn1 (which can also be written Sn1 S1 ) and the group S n2 S n2 are special cases of
Young subgroups of Sn . In general, a Young subgroup [21] [22] [23] of the symmetric group
Sn (with n an arbitrary integer number) is any subgroup isomorphic to Sn1 Sn2 ...Snk ,
with (n1 , n2, ..., nk) a partition of the number n, i.e. with n1 + n2 + ... + nk = n.
Also the conservative gates [12], mentioned in Section 2, form a Young subgroup of
w
S2 , based on the binomial partition 2w = 1 + (1w ) + (2w ) + ... + (w1
w ) + 1.

Reversible Logic

5.

215

Double Cosets

Even more powerful than cosets are double cosets. The double coset of a, element of G,
is defined as the set of all elements that can be written as b1ab2 , where both b1 and b2 are
members of the subgroup H. A surprising fact is that, in general, the double cosets, in which
G is partitioned by H, are of different sizes (ranging from H to H 2). The number of double
cosets, in which G is partitioned by H, therefore is not easy to predict. It is some number
G
G
between 1 + GH
H 2 and H . Usually, the number is much smaller than H , leading to the
(appreciated) fact that there are far fewer double cosets than there are cosets. This results
in smaller libraries for synthesis. However, there is a price to pay for such small library.
Indeed, if the chain of subgroups considered has length n, then the length of the synthesized
cascade is 2n 1 (instead of n as in the single coset synthesis).
The subgroup Sa1 partitions its supergroup Sa into only two double cosets, a small one
of size (a 1)! and a large one of size (a 1)!(a 1). Therefore, a double coset approach
using the MaslovDueck subgroup chain (3) needs only 2w library elements. However, a
w
synthesized cascade can be 22 1 gates long. Therefore this subgroup chain is not a good
choice in combination with double coset synthesis.
For the problem of synthesizing all members of S8 , Van Rentergem, De Vos and Storme
[24] have chosen the double cosets of the above mentioned subgroup obeying P1 = A1 .
They conclude that, for synthesizing all 40,320 members of S8 , they need a library of only
(4 + 2 + 1) + 1 = 8 elements. These suffice to synthesize an arbitrary member of S8 by a
cascade with length of seven or less. For an arbitrary circuit width w, synthesis of all (2w )!
members of S2w needs a library of 2w 1 elements.
Figure 5b illustrates one step of the procedure: the 24 elements of S4 are fabricated by
means of the 4 elements of its subgroup S2 S2 plus the representatives of the two other
double cosets in which S4 is partitioned by S2 S2 . Figure 7a shows how an arbitrary
member g of S16 is decomposed with the help of two members (b1 and b2) of S8 S8 and
one representative a of the double coset of g. Van Rentergem et al. have demonstrated that
it is always possible to construct a representative that is a controlled NOT gate:
P1 = f (A2, A3, ..., Aw) A1
P2 = A2
P3 = A3
...

...

Pw = Aw ,
where (in contrast to Section 3) A1 is the controlled bit (instead of Aw ). A proof is given in
Appendix A (case p = 2w1 and q = 2).
We conclude that the present synthesis for an arbitrary circuit of width w consists of the
cascade of
a first twin circuit,
a controlled NOT gate, and
a second twin circuit.

216

Alexis De Vos

Figure 7. An arbitrary circuit g (member of the group S16 ) decomposed as b1ab2 with the
help of double cosets generated (a) by its subgroup S28 and (b) by its subgroup S82 .
We illustrate the Van Rentergem procedure with an example where G = S8 and H isomorphic to S24 and thus G = 40, 320 and H = 576. We choose the truth table of Table 5a.
Figure 8a gives the result of repeated application of the procedure, until all subcircuits are
member of S2 , i.e. are equal to either the 1-bit identity gate or the 1-bit inverter. The nested
schematic can easily be translated into a chain of controlled NOTs, i.e. the conventional
way of writing down a reversible circuit: Figure 8b. This particular circuit consists of eigth
controlled NOT gates, of which seven are simply TOFFOLI gates. We now introduce a cost
function, called gate cost: we assign to each controlled NOT a unitary cost (whatever the
number of variables in the control function). The gate cost of Circuit 8b thus is 8. Note that
the present double coset approach ends up with a chain of cost of the order 4w .

6.

Double Cosets Again

Instead of applying a subgroup H isomorphic to the Young subgroup S2w1 S2w1 =

S22w1 and taking in each double coset a representative which is member of S2 S2 ...
w1
S2 = S22 , we can also work the other way around: choose a subgroup H isomorphic to the
w1
and look whether in each double coset there exists a representative
Young subgroup S22
2
which is member of S2w1 . See Figure 7b. This is indeed always possible [25] [26] [27].
Again, a proof is provided in Appendix A (case p = 2 and q = 2w1 ). Thus, we may
conclude that a synthesis of an arbitrary circuit of width w may consist of the cascade of
A

B
C

Q
R

A
B
C

P
Q
R

Figure 8. Decomposition of the example circuit of width 3 (Table 5a): (a) written as nested
controlled gates and (b) written as a chain of controlled NOTs.

Reversible Logic

217

a first controlled NOT gate,

a twin circuit, and
a second controlled NOT gate.
Note that Figure 7b is like Figure 7a inside out. We say that the two circuits are each others
dual. They are indeed based on two different Young subgroups. These subgroups are based
on two dual partitions of the number 2w (i.e. the degree of the supergroup S2w ) :
2w = 2w1 + 2w1
= 2 + 2 + ... + 2

(2w1 terms) .

Because of its importance, we give here in detail [25] [26] the synthesis algorithm using
the latter partition:

Algorithm A
In the algorithm, we use the notations like Ai (j), where the subscript i refers to the column
in the truth table, whereas the number j refers to the row. Thus these counters obey 1
i w and 1 j 2w . For finding the three parts of the decomposition, we proceed as
follows. We add to the given truth table (consisting of w input columns A and w output
columns P ) two extra sets of columns F and J. These are filled in, in three steps:
First, we fill in the w 1 columns F2 , F3 , ..., Fw by merely copying columns A2 , A3 ,
..., Aw and, analogously, fill in the w 1 columns J2, J3 , ..., Jw by merely copying
columns P2 , P3 , ..., Pw .
Then we construct a coil of 0s and 1s, starting from F1 (1) = 0.
Then we construct a second coil, starting from the non-filled-in F1 (j) with lowest j,
etcetera, until all F1 (and thus also all J1 ) are filled in.
Above, a coil consists of a finite number of windings. Here, a winding is a fourbit sequence F1 (k) = X, F1 (l) = X, J1 (l) = X, and J1 (m) = X, where l results from the condition that the string F2 (l), F3(l), ..., Fw(l) has to be equal to the
string F2 (k), F3(k), ..., Fw(k) and where m results from the condition that the string
J2 (m), J3(m), ..., Jw(m) has to be equal to the string J2 (l), J3(l), ..., Jw(l).
Although the above text might suggest that the algorithm is complicated, it is in fact
very straightforward. Figure 9 provides an illustration of this fact, by giving in detail the
synthesis of a reversible circuit of width w = 2, i.e. the circuit with two inputs (A1 and
A2 ) and two outputs (P1 = A2 and P2 = A1 , in this particular case). First, between
the columns (A1, A2) and (P1, P2) of the truth table, we insert the four empty columns
(F1 , F2) and (J1 , J2). Subsequently, columns F2 and J2 are filled in by simply copying
A2 and P2 , respectively. This step is displayed in boldface. Next comes the tricky part:
filling in the columns F1 and J1 . For this purpose, we start at F1 (1). We may set this bit
arbitrarily, but we choose to set it to 0. This starting choice is marked by a small square
in Figure 9. As a consequence, we automatically can fill in a lot of other bits in columns

218

Alexis De Vos

A1 A2

F1 F2

0
0
1
1

0
1
1
0

0
1
0
1

0
1
0
1

J1 J 2

0
1
1
0

1
1
0
0

P1 P2

0
1
0
1

1
1
0
0

Figure 9. Example of a synthesis according to the basic algorithm ( w = 2): expanded truth
table.
F1 and J1 . Indeed, as all computations need to be reversible, F1 (1) = 0 automatically
leads to F1 (3) = 1. Then we impose J1 (3) = F1 (3), i.e. J1 (3) = 1. Again, reversibility
requires that J1 (3) = 1 infers J1 (4) = 0. Etcetera, until we come back at the starting point
F1 (1). The arrows in Figure 9 show the order of filling in. Here, everything is filled in
when the travelling around is closed. So, this synthesis is finished after a single coil. This
example illustrates that during the application of the algorithm, we walk in circles, while
repeatedly assigning the bit sequence
0, 1, 1, 0, 0, 1, 1, ..., 1, 1, 0, 0, 1, 1, 0 .
In case the first coil is closed before columns J1 and F1 are completely filled in, the designer
just has to start a second coil, etcetera.
The fact that the above algorithm always comes to an end with the extended truth table
being completely filled, constitutes an additional proof that the theorem of Appendix A is
true for the special case p = 2 (and thus q = 2w1 ).
As a result, Algorithm A yields a decomposition of an arbitrary reversible circuit a (Figure 10a) into the desired cascade (Figure 10b) of a first controlled NOT gate with controlled
bit on the first wire, a twin circuit a1 leaving the first bit unaffected (P1 = A1 ), and a second
controlled NOT gate with controlled bit on the first wire. Note that circuit a1 in Figure 10b
is simpler than circuit a in Figure 10a, as a1 obeys P1 = A1.
Algorithm A can now be deepened as follows. By applying the decomposition of
a1 into three circuits, we obtain Figure 10c, where the circuit a2 is again simpler than
the circuit a1 , because it fulfils both P1 = A1 and P2 = A2 . Etcetera, until we obtain
Figure 10d, where the circuit aw1 obeys P1 = A1, P2 = A2 , ..., and Pw1 = Aw1 .
These properties reveal that aw1 is nothing but a control gate with controlled bit Aw .
Therefore, Figure 10d is equivalent to Figure 10e, such that we have decomposed a into
2w 1 controlled NOT gates. This procedure automatically leads us to Algorithm B:

Algorithm B
We add to the given truth table (consisting of w input columns A and w output columns P )
not two extra sets of columns, but 2(w 1) sets of columns. We call them A1, A2, ...,

Reversible Logic

a1

a2

a3

219

Figure 10. Decomposition of a reversible logic circuit of width w = 4 : (a) original logic
circuit, (b), (c), and (d) intermediate steps, (e) final decomposition.
Aw2 , Aw1 , P w1 , P w2 , ..., P 2 , and P 1 . Together they make 2(w 1)w new columns.
These are filled in, in the following steps:
First, we fill in all A1 columns except column A11 , by copying the w1 corresponding
A columns, and analogously, we fill in all P 1 columns except column P11 , by copying
the w 1 corresponding P columns.
Then we fill in the two columns A11 and P11 , by constructing a coil, starting from bit
A11(1), then constructing a new coil, starting at the non-filled-in A11(j) with lowest j,
etcetera, until all A11 (and thus also all P11 ) are filled in.
Then, we fill in all A2 columns except column A22 , by copying the w 1 correspond-

220

Alexis De Vos
Table 6. Expanded truth table according to Algorithm B.

A1A2 A3
0
0
0
0
1
1
1
1

0
0
1
1
0
0
1
1

0
1
0
1
0
1
0
1

A11 A12 A13

0
1
0
1
1
0
1
0

0
0
1
1
0
0
1
1

0
1
0
1
0
1
0
1

A21 A22 A23

0
1
0
1
1
0
1
0

0
0
1
1
1
0
0
1

0
1
0
1
0
1
0
1

P12 P22 P32

0
1
0
1
1
0
1
0

0
0
1
1
1
0
0
1

0
1
0
0
1
1
0
1

P11 P21 P31

0
1
0
1
1
0
1
0

0
0
1
0
1
1
1
0

0
1
0
0
1
1
0
1

P1 P2 P3
1
1
1
0
1
0
0
0

0
0
1
0
1
1
1
0

0
1
0
0
1
1
0
1

ing A1 columns, and analogously, we fill in all P 2 columns except column P22 , by
copying the w 1 corresponding P 1 columns.
Then we fill in the two columns A22 and P22 , by constructing the appropriate number
of coils, starting from bit A22 (1), until all A22 (and thus also all P22 ) are filled in.
w1
Etcetera, until finally all Aw1
w1 (and thus also all Pw1 ) are filled in. At that moment,
we have all 2w2 2w entries of the extended table.

We illustrate our deepened procedure for the example circuit in Table 5a. By applying
the above procedure, we obtain Table 6. The first step of the procedure is displayed in bold
face, whereas the second step is emphasized in italic. (The reader may verify that here
this step requires two coils, the former having three windings, the latter having only one
winding.) The third step of the algorithm is underlined.
The above procedure thus yields a decomposition of the logic circuit into five logic
circuits (one computing A1 from A, one computing A2 from A1 , ..., and one computing P
from P 1 ). All five subcircuits are automatically controlled NOT gates. By merely inspecting
Table 6, we find their subsequent control functions: f (A2 , A3) = A3 , f (A11 , A13) = A11 A13 ,
f (A21 , A22) = A21 A22 , f (P12 , P32 ) = P12 P32 , and f (P21 , P31) = P31 .
Figure 11a shows the final synthesis of Table 5a with its five controlled NOT gates.
Note that this gate cost of 5 is lower than the gate cost of 8 in Figure 8b. Noteworthy is the
automatic V-shape of the positions of the five crosses (i.e. controlled NOTs) in the figure.
When we apply the same procedure to each of the 8! = 40,320 circuits of the group S8 ,
sometimes one or more of the five control functions equals 0. This means that one or more
of the five controlled NOTs is the identity gate and thus in fact is abscent, such that there
is a total of less than five gates. We thus yield a statistical distribution of gate cost ranging
from 0 to L = 2w 1 = 5. The average gate cost turns out to be about 4.2.
We stress that a gate cost 2w 1 is very close to optimal. No synthesis method can do
better than 2w3. This is proven in Appendix B. This constitutes a significant improvement
over the method described in Section 5, as the latter yields cascade lenghts rising as 4w . As

Reversible Logic

221

A1

P1

A2

P2

A3

P3

A1

P1

A2

P2

A3

P3

Figure 11. Decomposition of the example circuit of width 3 (Table 5a), according to Algorithm B: (a) into controlled NOT gates and (b) into TOFFOLI gates.
w1

lead to (almost) optimal decompositions, we conclude that

subgroups isomorphic to S22
the controlled NOTs form a natural library for synthesis. We stress that such library is
larger than libraries with merely TOFFOLI gates. The latter are of the type of Figure 4b,
controlled by AND functions. In the synthesis approach presented here, we fully make use
of building blocks of the type of Figure 4a, controlled by arbitrary control functions.
Whereas the synthesis method of Section 5 is based on a subgroup chain like (5), i.e. of a
set of subgroups of ever diminishing size, the Algorithm B is based on a set of w subgroups
w1
all of the same order S22 . These w subgroups are conjugate to each other. This means
that one subgroup Hi can be obtained from another one Hj as follows:
Hi = c Hj c1 ,
where c is some member of the supergroup G. In our case we have that c = c1 is the
SWAP gate exchanging the wires # i and # j. Figure 12 illustrates the two approaches: the
former with ever smaller subgroups and the latter with equal-size subgroups. In order to
be efficient, the conjugate subgroups should overlap as little as possible. In our synthesis
method, they ovelap very little, as two subgroups have only one element in common, i.e.
the identity element (or w-bit follower):
Hi Hj = I .
Sometimes, for practical purposes, the library of controlled NOT gates is considered as
too large. Then a library consisting of only TOFFOLI gates is a possibility. In that case,
we may proceed as follows: each controlled NOT is decomposed into TOFFOLI gates, by
merely replacing the control function by its ReedMuller expansion. Such expansion may
contain up to 2w1 terms. From Appendix B, we see that such expansion is not optimal.
Better results can be obtained by one of the so-called ESOP expansion 1 algorithms [28]
[29]. Figure 13 gives an example: the controlled NOT gate with control function f =
A1 + A2 , i.e. an OR function. The ReedMuller expansion (Figure 13b) reads
f = A1 A2 A1 A2 ,
1

The name ESOP is derived from the expression exclusive-or sum of products. It is a variant of the SOP
or sum of producs, where sum stands for the Boolean OR function and product for AND. Thus an ESOP is
a Boolean expression which is written as a XOR of ANDs. In contrast to the ReedMuller expansion, an ESOP
expansion allows NOT functions (besides the XOR and the AND functions).

222

Alexis De Vos
S

S2
S

2
4

S2

I
I

4
S
2

S2

Figure 12. Two sets of efficient subgroups: (a) a subgroup chain and (b) a subgroup flower.
whereas one of the minimum-ESOP expansions (Figure 13c) is:
f = A1 A1 A2 .
Applying either the ReedMuller decomposition or a minimum- ESOP expansion to the
circuit of Figure 11a yields Figure 11b, with six TOFFOLI gates.

7.

Garbage Bits

In the above sections, synthesis means: finding a hardware implementation for a given
reversible truth table. However, often a synthesis job is defined by an irreversible truth
table. A famous example is the design of a full adder: see Table 7a. It has three input bits:
the augend bit A, the addend bit B, and the carry-in bit Ci ; it has two output bits: the sum

Figure 13. Decomposition of (a) a controlled NOT gate into (b) its ReedMuller expansion
and into (c) one of its minimal ESOP expansions.

Reversible Logic

223

Table 7. Full adders: (a) irreversible and (b) reversible.

ABCi

Co S

000
001
010
011
100
101
110
111

00
01
01
10
01
10
10
11
(a)

ABCi P

Co SG1G2

0000
0001
0010
0011
0100
0101
0110
0111
1000
1001
1010
1011
1100
1101
1110
1111

0000
1000
0100
1100
0101
1101
1001
0001
0111
1111
1011
0011
1010
0010
1110
0110
(b)

bit S and the carry-out bit Co . Basicly, the table gives eight additions of three numbers:
0+0+0 = 0
0+0+1 = 1
...
1+1+0 = 2
1+1+1 = 3.
The truth table surely is not reversible:
the number of output bits is not equal to the number of input bits and
various output words appear more than once in the table, the output Co S = 10 e.g.
appearing three times in Table 7a.
In order to implement this calculation in a revesible computer, we have to expand the
table, such that the original table is embedded in a large reversible one: see e.g. Table 7b.
All bits from Table 7a are repeated (in boldface) in Table 7b. The new table has two extra
output columns: G1 and G2. Those bits are called garbage bits. They are not asked for in the

224

Alexis De Vos

A
B
Ci

Co

Figure 14. Full adder with three TOFFOLI gates and one FREDKIN gate.
first place, but are added in order to guarantee that all output words are different. Because
there are now four output bits, there have to be four input bits as well. Therefore, we have
added one additional input column: bit P , called the preset. For the desired application, P
will always be put equal to 0. We note that there is no unique way to embed an irreversible
table in a larger reversible table. The embedding should thus be done carefully, in order
to minimize the resulting reversible hardware cost. Figure 14 gives the cheapest reversible
implementation of Table 7a [30]. It consists of three TOFFOLI gates, one FREDKIN gate,
and two SWAP gates. Its gate cost is 4 (as SWAPS are considered free of cost).
We note that a controlled NOT can be interpreted as a reversible embedding of the
calculation of an irreversible Boolean function. Indeed, assume that we like to calculate
the Boolean function f (A, B, C) of three Boolean variables, as defined by its truth table in
Table 8a. The table is, of course, irreversible. Indeed, it has less output columns than input
colums. Besides, the output row 0 appears not less than five times and the output row 1
appears three times.
Table 8b shows the truth table of the controlled NOT gate with control function
f (A, B, C), with controlling bits A, B, and C, and with controlled bit the extra input D.
We have the following relations between outputs and inputs: P = A, Q = B, R = C, and
S = f (A, B, C) D, in accordance with (1). The output S equals the desired function f ,
if the input D is preset to zero. The extra outputs P , Q, and R are garbage outputs. We
note that the reversible embedding (Table 8b) unfortunately has a total number of columns
which is double the number of columns in the original problem (Table 8a).
Thus, reversible digital circuits have the disadvantage to generate a lot of garbage output, not desired for the application. Because of Landauers principle, we are not allowed
to throw them away. We thus have to take them all the way through the following computational steps. Only two garbage bits in a full adder is already a matter of great concern.
Indeed, such full adder is just a building block in, say, a 32-bit adder. The latter circuit
will itself be a building block for a 32-bit multiplier. Such multiplier is used many times
in e.g. a digital filter. Many filters make up a filter bank; many filter banks make up e.g. a
speech processor. Each time we step from one architectural level to the next, the number of
garbage bits explodes. This proliferation of garbage will cause huge costs because of extra
gates and extra interconnections. Therefore, the design challenge consists of designing, at
each level of abstraction, circuits that generate as little garbage as possible. The clever Cuccaro adder [31] succeeds in avoiding one of the two garbage bits G1 and G2 in Table 7b and
thus is highly recommendable [32]. The trick consists in avoiding to calculate Co and S
simultaneously, in a single circuit. Cuccaro et al. first calculate Co and afterwards compute
S from Co , when the information of Co is not necessary anymore.

Reversible Logic

225

Table 8. Computing the Boolean function f (A, B, C): (a) irreversibly and (b)
reversibly.

ABC

000
001
010
011
100
101
110
111

0
0
0
1
1
0
0
1

(a)

ABCD

P QRS

0000
0001
0010
0011
0100
0101
0110
0111
1000
1001
1010
1011
1100
1101
1110
1111

0000
0001
0010
0011
0100
0101
0111
0110
1001
1000
1010
1011
1100
1101
1111
1110
(b)

8.

Experimental Prototypes

We can use either left cosets, or right cosets, or double cosets in the synthesis procedure;
we can choose one subgroup or another. Whatever choices we make, we obtain a procedure
for synthesizing an arbitrary circuit by cascading a small number of standard cells from a
limited library. By appropriate choice of the representatives of the (double) cosets, we can
see to it that all building blocks in the library are member of either of the two following
special subgroups:
the subgroup of controlled NOT gates and
the subgroup of controlled SWAP gates.
Among the controlled NOT gates, we note three special elements:
If f is identically zero, then Pw is always equal to Aw . Then the gate is the identity
gate i.
If f is identically one, then Pw always equals 1 Aw . Then the gate is an inverter or
NOT gate: Pw = Aw .

226

Alexis De Vos

If f (A1, A2, ..., Aw1) equals the (w 1)-bit AND function A1 A2 ...Aw1, then the
gate is a TOFFOLI gate: whenever A1 A2...Aw1 equals 0, then Pw simply equals
Aw ; but whenever A1 A2...Aw1 equals 1, then Pw equals NOT Aw .
For physical implementation, dual logic is very convenient. It means that, within the
hardware, any logic variable X is represented by two physical quantities, the first representing X itself, the other representing NOT X. Thus, e.g. the physical gate realizing logic gate
of Table 2a has four physical inputs: A, NOT A, B, and NOT B, or, in short-hand notation:
A, A, B, and B. It also has four physical outputs: P , P , Q, and Q. Such approach is common in electronics, where it is called dual-line or dual-rail electronics. Also some quantum
computers make use of dual-rail qubits [33]. As a result, half of the input pins are at logic 0
and the other half at logic 1, and analogous for the output pins. In this way, dual electronics
is physically conservative: the number of 1s at the output equals the number of 1s at the
input (i.e. equals w), even if the truth table of the reversible logic gate is not conservative.
As a result, we get the advantages of conservative logic, without having to restrict ourselves
to conservative logic.
Dual-line hardware allows very simple implementation of the inverter. It suffices to
interchange its two physical lines in order to invert a variable. In other words: in order to
hardwire the NOT gate:
output P is simply connected to input A and
output P is simply connected to input A.
Controlled NOTs are implemented as NOT gates which are controlled by switches. A first
example is the controlled NOT gate with a single controlling bit (i.e. the bit A):
P

= A

Q = AB .
These logic relationships are implemented into the physical world as follows:
output P is simply connected to input A,
output P is simply connected to input A,
output Q is connected to input B if A = 0,
but connected to B if A = 1, and
output Q is connected to input B if A = 0,
but connected to B if A = 1.
The last two implementations are shown in Figure 15a. In the figure, the arrow heads
show the position of the switches if the accompanying label is 1. A second example is a
TOFFOLI gate with two controlling bits (i.e. the bit A and the bit B):
P

= A

Q = B
R = AB C .
Its logic relationships are implemented into physical world as follows:

Reversible Logic

227

output P is simply connected to input A,

output P is simply connected to input A,
output Q is simply connected to input B,
output Q is simply connected to input B,
output R is connected to input C if either A = 0 or B = 0,
but connected to C if both A = 1 and B = 1, and
output R is connected to input C if either A = 0 or B = 0,
but connected to C if both A = 1 and B = 1.
The last two implementations are shown in Figure 15b. Note that in both Figures 15a
and 15b, switches always appear in pairs, of which one is closed whenever the other is
open and vice versa. It is clear that the above design philosophy can be extrapolated to a
controlled NOT gate with arbitrary control function f . Suffice it to wire a square circuit like
in Figures 15a and 15b, with the appropriate series and parallel connections of switches.
If we apply such hardware wiring to the controlled NOT with control function f =
A + B, we realise that direct implementation of Figure 13a needs eight switches. In contrast, Figure 13b needs sixteen switches and Figure 13c needs twelve switches. This fact
illustrates that decomposing a controlled NOT into TOFFOLIs is not necessarily advantageous.
Now that we have an implementation approach, we can realize any reversible circuit in
hardware. We will demonstrate here some examples of implementation into electronic chip.
In electronic circuits, a switch is realized by the use of a so-called transmission gate, i.e. two
MOS-transistors in parallel (one n-MOS transistor and one p-MOS transistor). Here, MOS
stands for metal-oxide-semiconductor, whereas n refers to negative type and p to positive
type. Circuits containing both n-type transistors and p-type transistors are called c-MOS
circuits, where c stands for complementary. Readers, familiar with electronics, will notice
that circuits like in Figure 15 are not of the conventional style called restoring logic, but
of a less well-known style called pass-transistor logic. The pass-transistor logic families
allow good control of leakage currents and therefore yield lower energy consumption [34].
As an example, Figure 16 shows a 4-bit reversible ripple adder [35], implemented in
full-custom 2.4 m standard c-MOS technology, consisting of sixteen TOFFOLI gates with
a total of 192 transistors. This prototype chip was fabricated in 1998. The circuit functions
equally well from left to right as it works from right to left. Conventional digital chips
do not have this property. A second example [36] (Figure 17) was fabricated in 2000, in
submicron technology: a 4-bit carry-look-ahead adder, implemented in 0.8 m standard cMOS technology, containing four TOFFFOLIs with w = 2, four controlled NOTs of width
w = 3, and one complex controlled NOT of width w = 13. It contains a total of 320
transistors.
Switches not only can decide whether an input variable is inverted or not. We can
apply switches also in order to decide whether two input variables are swapped or not. This

228

Alexis De Vos
B

_
Q

_
B

Q
C

_
R

_
C

Figure 15. Schematic for (a) a controlled NOT gate (u = 1, v = 1), (b) a controlled NOT
gate (u = 2, v = 1), and (c) a controlled SWAP gate (u = 1, v = 2).
concept leads to the implementation of the controlled SWAP gate, e.g. the FREDKIN gate:
P

= A

Q = AB AC B
R = AB AC C ,
where A is the controlling bit and B and C are the two controlled bits. The set of equations
corresponds to (2), with control function f (A) = A. Figure 15c shows the physical implementation. The reader will easily extrapolate the design philosophy to reversible logic
gates of width w = u+v, where u controlling bits decide, by means of a control function f ,
whether the v controlled bits are either subjected or not to some particular swapping and/or
inverting.
One application [32] (Figure 18) is an 8-bit Cuccaro adder, implemented in 0.35 m
standard c-MOS technology, containing 17 TOFFOLI and 16 FREDKIN gates. It contains

Reversible Logic

229

Figure 16. Microscope photograph (140 m 120 m) of a 2.4-m 4-bit reversible ripple
adder.

Figure 17. Microscope photograph (610 m 290 m) of a 0.8-m 4-bit reversible carrylook-ahead adder.
a total of 392 transistors. The prototype chip was fabricated in 2004. The reader will
observe that full-custom prototyping at a university lab follows Moores law, with a couple
of years delay with respect to industry. Indeed, many commercial chips nowadays use 0.18
or 0.13 m transistors. Some companies have entered the nanoscale era, by introducing
90 nm, 65 nm and even 45 nm products [37] to the market.
Moores law, i.e. the continuing shrinking of the transistor sizes, leads to a continuing
decrease of the energy dissipation per computational step. This heat generation Q is of the
order of magnitude of CVt2 , where Vt is the threshold voltage of the transistors and C is
the total capacitance of the capacitors in the logic gate [38]. We have C of the order of
magnitude of 0  LW
t , where L, W , and t are the length, the width, and the oxide thickness
of the transistors, whereas 0 is the permittivity of vacuum (8.85 1012 F/m) and  is the

230

Alexis De Vos

Figure 18. Microscope photograph (140 m 230 m) of a 0.35-m 8-bit reversible

Cuccaro adder.
dielectric constant of the oxide (either 3.9 for silicon oxide SiO 2 or 22 for hafnium oxide
HfO2). Table 9 gives some typical numbers. Note that a technology is named after its
value for the transistor length L. We see how Q becomes smaller and smaller, as L shrinks.
However, this dissipation in electronic circuits still is about four orders of magnitude in
excess of the Landauer quantum kT log(2), which amounts (for T = 300 K) to about 3

Reversible Logic

231

Table 9. Moores law for dimensions L, W , and t, and for threshold voltage Vt , as well
as for resulting capacitance C and heat dissipation Q.
technology
(m)

L
(m)

2.4
0.8
0.35

2.4
0.8
0.35

W
(m)

t
(nm)

Vt
(V)

2.4
2.0
0.5

42.5
15.5
7.4

0.9
0.75
0.6

C
(fF)

Q
(fJ)

46.8
3.6
0.82

38
2.0
0.30

1 femtojoule

1000

100

Q (attojoule)

10

C V_t ^2

1 attojoule

0.1

0.01
Landauer quantum
0.001

1 zeptojoule
2000

2010

2020

2030

2040

Figure 19. Adiabatic heat generation Q in future technologies.

1021 J or 3 zeptojoule.
Further shrinking of L and W and further reduction of Vt ultimately will lead to a
Q value in the neighbourhood of kT log(2). According to the International Technology
Roadmap of Semiconductors [39], Moores law will continue in the near future. Figure 19
shows how CVt2 will continue to decrease, ultimately approaching Landauers quantum
around 2034...
That day, digital electronics will have good reason to be reversible... This, however,
does not mean that the reversible MOS circuits are useless today. Indeed, as they are a reversible form of pass-transistor topology, they are particularly suited for adiabatic addressing [40], leading to substantial power saving. In subsequent gates, switches are opening
and closing, one after the other, like dominoes, transferring information from the inputs to
the outputs of the overall circuit [41]. Figure 20 shows an example of a quasi-adiabatic
experiment [42]. We see two transient signals: one of the input variables and one of the

232

Alexis De Vos

Figure 20. Oscilloscope view of a 0.35-m full adder.

resulting output bits. In practice, such procedure leads to a factor of about 10 in power
reduction [38]. The reduction of the power dissipation is even more impressive if standard
c-MOS technology is replaced by SOI (silicon-on-insulator) technology. Indeed, in the latter process, the threshold voltage Vt can be controlled better, such that low-Vt technologies
are possible [43] [44].

9.

Conclusion

We have demonstrated how the symmetric group S2w has various interesting subgroups.
Many (but not all) of them turn out to be Young subgroups. We have shown how cosets and
double cosets are particularly helpful for synthesizing arbitrary reversible circuits. E.g., we
have demonstrated the power of subgroup chains of the form S2w S2w 1 S2w 2 ...
w2
w1
w
S22
S21 for
S3 S2 S1 and the form S2w S22w1 S42w2 ... S24
single coset synthesis. A particularly efficient synthesis is obtained by the double cosets of
w1
w different but conjugate subgroups of the type S22 .
w1

Controlled NOTs belong to a subgroup isomorphic to S22 , whereas controlled SWAPs

w2
belong to a subgroup isomorphic to S22 . They form the building blocks for hardware
implementation. Silicon prototypes of adder circuits, in three different standard technologies (2.4 m, 0.8 m, and 0.35 m), have been fabricated and tested. They illustrate how
energy dissipation and heat generation in chips keep on shrinking according to Moores law,
eventually leading to computer hardware in which Landauers principle will be more than
academic...

Reversible Logic

233

Acknowledgements
The author wishes to thank his Ph.D. students Bart Desoete, Filip Beunis, and Yvan Van
Rentergem, as well as prof. Leo Storme, for valuable collaborations. He further thanks
the Invomec division of Imec v.z.w. (Leuven, Belgium) and the Eurochip and Europractice
organisations, for processing the chips at Alcatel Microelectronics (Oudenaarde, Belgium),
Austria Mikro Systeme (Unterpremstatten, Austria), and AMI Semiconductor (Oudenaarde,
Belgium).

Appendix A: A Remarkable Theorem from Combinatorics

Let n be an integer, i.e. a number of objects {1,2,...,n}. Let p be a divisor of n, i.e.
we have n = pq, with both p and q integers. We arrange the n objects into q rows, each
of p objects, the arrangement being called a Young tableau [45]. We consider an arbitrary
permutation of the n objects. During such a permutation, the q subsets exchange q 2 flows
Fij . These Fij form a matrix F with all matrix elements equal to 0, 1, 2,..., or p. The matrix
element Fij denotes the number of objects which move from row # i to row # j. There are
2q constraints
X

Fij

= p

Fij

= p,

X
i

of which 2q 1 are independent. As an example, we consider an arbitrary permutation of

the 35 objects {1,2,...,35}. Figure 21a shows the permutation a in the Young tableau
for n = 35, p = 7, and q = 5. The tableau consists of the five sets {1,2,...,7},
{8,9,...,14}, ..., and {29,30,...,35}. These sets exchange a number of objects
according to the flow matrix

F =

6
1
0
0
0

0
5
1
0
1

0
0
5
0
2

1
0
0
6
0

0
1
1
1
4

where e.g. F53 = 2 expresses that in Figure 21a two objects are mapped from the fifth row
{29,30,...,35} to the third row {15,16,...,21}. Note that the qq flow matrix F
contains incomplete information about the permutation a. The complete information on a
is given by the corresponding n n permutation matrix.
Theorem : each permutation a can be decomposed as a = h1 vh2, where both h1 and
h2 only permute objects within rows and where v only permutes objects within columns.
Figure 21a shows the permutation a, as a mapping. Figures 21b, 21c, and 21d show the
corresponding permutations h1 , v, and h2 to be performed subsequently:
The vertical permutation v (Figure 21c) is found as follows: the cycles of a are
projected vertically, yielding one or more vertical cycles.

234

Alexis De Vos

Figure 21. Decomposition of (a) an arbitrary permutation of 35 objects into (b) a first horizontal permutation, (c) a vertical permutation, and (d) a second horizontal permutation.

The horizontal permutation h1 (Figure 21b) merely consists of horizontal arrows

which map the arrow tails of vertical and oblique arrows in Figure 21a to the corresponding arrow tails of Figure 21c. Subsequently additional horizontal arrows are
added, in order to form closed horizontal cycles.
Finally, the horizontal permutation h2 (Figure 21d) simply equals v 1h1
1 a.
This theorem forms the basis of the existence of Clos networks [46], more precisely
rearrangeable (non-blocking) Clos networks [47] [48] [49]. In the past, the approach has
been applied succesfully in telephone switching systems. Nowadays, it is fruitful in internet
routing [50]. These conventional applications of the theorem are concerned with permutations of wires, i.e. with the decomposition of the members of the subgroup of exchangers
(isomorphic to Sw ). Here we apply the theorem not to the w wires, but to the 2w messages,
i.e. to the full group S2w . Whereas Figure 10b is reminiscent of Clos networks (but here
with n = 2w instead of n = w), Figure 10e is reminiscent of so-called banyan networks
(but here again with n = 2w instead of n = w). Finally, the coils/windings procedure of
Section 6 is reminiscent of the looping algorithm, presented of Hui [48].
The fact that it is always possible to construct an appropriate vertical permutation v is
a direct consequence of either Birkhoffs theorem [51] (a.k.a. the Birkhoffvon Neumann
theorem) or Halls marriage theorem or Konigs theorem or the maximum-flow theorem,
all these notorious theorems in combinatorics being equivalent [52]. As an example, one of
the formulations of the Birkhoff theorem says: any matrix of size q q and line sum p can
be decomposed as the sum of p matrices of the same size and unit line sum [53]. In other
words: any flow matrix with line sum p can be decomposed as the sum of p permutation
matrices. E.g. the above matrix F of size 5 5 with line sum 7 can be decomposed as the

Reversible Logic

235

following sum of seven permutation matrices:

F =4

1
0
0
0
0

0
1
0
0
0

0
0
1
0
0

0
0
0
1
0

0
0
0
0
1

1
0
0
0
0

0
1
0
0
0

0
0
0
0
1

0
0
0
1
0

0
0
1
0
0

1
0
0
0
0

0
0
0
0
1

0
0
1
0
0

0
0
0
1
0

0
1
0
0
0

0
1
0
0
0

0
0
1
0
0

0
0
0
0
1

1
0
0
0
0

0
0
0
1
0

It suffices to apply in each of the p columns of the Young tableau one of these p permutations, in order to obtain the desired vertical permutation v. In Figure 21c, we indeed
recognize four empty columns (corresponding to the four identity matrices in the matrix
sum) and three non-empty columns (corresponding to the remaining three matrices in the
matrix sum).
Thus, the theorem says that one permutation of n objects can be decomposed into
a product of q subpermutations, each of p objects, followed by
a product of p subpermutations, each of q objects, and finally
a second product of q subpermutations, each of p objects.
This is depicted in Figure 22. In brief: one big spaghetti is decomposed into 2q + p small
permutations. The theorem can be interpreted in terms of group theory: if we define
N as the group of all permutations of the n objects,
H as the group of all horizontal permutations of these objects, and
V as the group of all vertical permutations,
then we have that
N is isomorphic to the symmetric group Sn ,
H is isomorphic to the Young subgroup Sp Sp ...Sp = Sqp , and
V is isomorphic to the Young subgroup Sq Sq ...Sq = Spq .
Note that the Young subgroups Sqp and Spq are based on two so-called dual partitions of the
number n :
n = p + p + ... + p

(q terms) and

n = q + q + ... + q

(p terms).

Therefore these two subgroups are referred to as dual Young subgroups. The combinatorial
theorem says the following: if the group N is partitioned into double cosets by means of
the subgroup H, then we can choose in each double coset a representative which is member
of V. In how many double cosets the supergroup N is partitioned by the subgroup H is a
very difficult problem. We will call this number X(p, q). Fortunately, knowledge of the
value of X is not important; essential is the fact that (thanks to Birkhoffs theorem) each of
the X double cosets contains (at least) one member of V.
We now consider the special case where n is even. We distinguish two special subcases
[26]:

236

Alexis De Vos
p
planes

q
planes

q
planes

Figure 22. The 3-dimensional decomposition of an arbitrary permutation of 35 objects

into five horizontal subpermutations, seven vertical subpermutations, and a second five
horizontal subpermutations.
p = 2 and thus q = n/2
p = n/2 and thus q = 2.
In the latter subcase, the group Sn is partitioned by the subgroup isomorphic to S2n/2
into exactly
n
n
X( , 2) = + 1
2
2
double cosets [24]. This partitioning is applied in Section 5. In the former subcase, the
n/2
number of double cosets in which Sn is partitioned by the subgroup S2 is surprisingly
complicated [54]:
n/2
[( n2 )!]2 X 2k (n 2k)!
n
.
X(2, ) =
2
2n k=0 [( n2 k)!]2 k!

This partitioning is applied in Section 6.

Appendix B: Optimal Syntheses

In the present appendix, we follow the reasoning method introduced by Even et al. [55], by
Maslov and Dueck [18], and by Patel et al. [13]. Let us assume a set of N circuits, to be
synthesized with the help of a library of B different building blocks (a subset of the former
set, including the trivial identity gate). We build all possible cascades of length l. We call
m(l) the number of different circuits which can be synthesized by these cascades. Note that
these m circuits automatically include all circuits synthesized by a shorter cascade, because
we have included the identity building-block in the library. The question arises: which
cascade length L is necessary, in order to guarantee that these cascades contain the synthesis
of all N given circuits, i.e. to guarantee that m(L) N ? A hypothetical most efficient
library (where all different cascades yield different circuits) would realize m(l) = B l .

Reversible Logic
Therefore B L N and thus
L=

log(N )
log(B)

237

(7)

Here dxe stands for the ceiling of x, i.e. the smallest integer larger than or equal to x. We
now apply this general result to three different cases:
(1) In a first application, N is the total number of reversible circuits of width w and B
is the number of controlled NOT gates of width w. Thus:
= (2w )!

B = w (22

w1

1) + 1 .

Using the Stirling inequalities [56]

2 nn+1/2 en < n! <

1
),
2 nn+1/2 en (1 +
4n

we obtain that
w

N > 2[wlog(2)]2
and finally conclude that
L > 2w 4 .

Therefore our decomposition in Section 6 (with l = 2w 1) is close to optimal.

(2) If we keep N = (2w )!, but we choose for B is the number of controlled SWAP gates
of width w, then B is somewhat smaller:
B=

w(w 1) 2w2
(2
1) + 1 ,
2

leading to
L > 4w 10 .
This proofs that the synthesis method of Section 6 cannot work with 2w1 controlled SWAP
gates. The reader is kindly invited to verify that this (negative) result is not in conflict with
the theorem of Appendix A for n equal to a multiple of 4 and p = 4 and q = n4 .
(3) If we apply the general result (7) to the decomposition of controlled NOT gates into
TOFFOLI gates, we have
N

= 22

w1

B = 3w1
and thus obtain
L>

2w
.
2 log2(3)w

Therefore, the ReedMuller decomposition, needing a cascade of up to

is not optimal.

2w
2

building blocks,

238

Alexis De Vos

References
[1] Markov, I. An introduction to reversible circuits. Proceedings of the 12 th International Workshop on Logic and Synthesis , Laguna Beach (May 2003), pp. 318 - 319.
[2] Frank, M. Introduction to reversible computing: motivation, progress, and challenges.
Proceedings of the 2005 Computing Frontiers Conference , Ischia (May 2005), pp.
385 - 390.
[3] De Vos, A. Lossless computing. Proceedings of the I.E.E.E. Workshop on Signal Processing, Poznan (October 2003), pp. 7 - 14.
[4] Hayes, B. Reverse engineering. American Scientist 94, 107 - 111 (MarchApril 2006).
[5] Feynman, R. Quantum mechanical computers. Optics News 11, 11 - 20 (1985).
[6] Landauer, R. Irreversibility and heat generation in the computational process. I.B.M.
Journal of Research and Development 5, 183 - 191 (1961).
[7] Keyes, R. and Landauer, R. Minimal energy dissipation in logic. I.B.M. Journal of
Research and Development 14, 153 - 157 (1970).
[8] Bennett, C. Logical reversibility of computation. I.B.M. Journal of Research and Development 17, 525 - 532 (1973).
[9] Bennett, C. and Landauer, R. The fundamental physical limits of computation. Scientific American 253, 38 - 46 (July 1985).
[10] Scott, W. Group theory. New York: Dover Publications (1964).
[11] Hall, P. The theory of groups. Providence: AMS Chelsea Publishing (1968).
[12] Fredkin, E. and Toffoli, T. Conservative logic. International Journal of Theoretical
Physics 21, 219 - 253 (1982).
[13] Patel, K., Markov, I. and Hayes, J. Optimal synthesis of linear reversible circuits.
Proceedings of the 13 th International Workshop on Logic and Synthesis , Temecula
(June 2004), pp. 470 - 477.
[14] Wang, L. and Almaini, A. Optimisation of ReedMuller PLA implementations. I.E.E.
Proceedings Circuits, Devices and Systems 149, 119 - 128 (2002).
[15] Kerntopf, P. On universality of binary reversible logic gates. Proceedings of the 5 th
Workshop on Boolean Problems , Freiberg (September 2002), pp. 47 - 52.
[16] De Vos, A. and Storme, L. r-Universal reversible logic gates. Journal of Physics A:
Mathematical and General 37, 5815 - 5824 (2004).
[17] De Vos, A., Raa, B. and Storme, L. Generating the group of reversible logic gates.
Journal of Physics A: Mathematical and General 35, 7063 - 7078 (2002).

Reversible Logic

239

[18] Maslov, D. and Dueck, G. Reversible cascades with minimal garbage. I.E.E.E. Transactions on Computer-Aided Design of Integrated Circuits and Systems 23, 1497 - 1509
(2004).
[19] Van Rentergem, Y., De Vos, A. and De Keyser, K. Using group theory in reversible
computing. Proceedings of the I.E.E.E. World Congress on Computational Intelligence, Vancouver (July 2006), pp. 8566 - 8573.
[20] Van Rentergem, Y., De Vos, A. and De Keyser, K. Six synthesis methods for reversible
logic. Open Systems & Information Dynamics 14, 91 - 116 (2007).
[21] Kerber, A. Representations of permutation groups I. Lecture Notes in Mathematics
240, Berlin: Springer Verlag (1970), pp. 17 - 23.
[22] James, G. and Kerber, A. The representation theory of the symmetric group. Encyclopedia of Mathematics and its Applications 16, 15 - 33 (1981).
[23] Jones, A. A combinatorial approach to the double cosets of the symmetric group with
respect to Young subgroups. European Journal of Combinatorics 17, 647 - 655 (1996).
[24] Van Rentergem, Y., De Vos, A. and Storme, L. Implementing an arbitrary reversible
logic gate. Journal of Physics A: Mathematical and General 38, 3555 - 3577 (2005).
[25] Van Rentergem, Y. and De Vos, A. Synthesis and optimization of reversible circuits.
Proceedings of the ReedMuller 2007 Workshop , Oslo (May 2007), pp. 67 - 75.
[26] De Vos, A. and Van Rentergem, Y. Young subgroups for reversible computers. Advances in Mathematics of Communications . 2, 183 - 200 (2008).
[27] De Vos, A. and Van Rentergem, Y. Networks for reversible logic. Proceedings of the
8 th International Workshop on Boolean Problems , Freiberg (September 2008).
[28] Gaidukov, A. Algorithm to derive minimum ESOP for 6-variable function. Proceedings of the 5 th International Workshop on Boolean Problems , Freiberg (September
2002), pp. 141 - 148.
[29] Pogosyan, G., Rosenberg, I. and Takada, S. Building minimum ESOPs through redundancy elimination. Proceedings of the 6 th International Workshop on Boolean
Problems, Freiberg (September 2004), pp. 201 - 206.
[30] Van Rentergem, Y. and De Vos, A. Optimal design of a reversible full adder. International Journal of Unconventional Computing 1, 339 - 355 (2005).
[31] Cuccaro, S., Draper, T., Kutin, S. and Moulton, D. A new quantum ripple-carry addition circuit. Proceedings of the 8 th Workshop on Quantum Information Processing ,
Cambridge (June 2005), arXiv:quant-ph/0410184.v1.
[32] Skoneczny, M., Van Rentergem, Y. and De Vos, A. Reversible Fourier transform chip.
Proceedings of the 15 th International Conference on Mixed Design of Integrated
Circuits and Systems, Poznan (June 2008).

240

Alexis De Vos

[33] Chuang, I. and Yamamoto, Y. The dual-rail quantum bit and quantum error correction.
Proceedings of the 4 th Workshop on Physics and Computation , Boston (November
1996), pp. 82 - 91.
[34] Singh M., Giacomotto C., Zeydel B. and Oklobdzija V. Logic style comparison for
ultra low power operation in 65 nm technology. Proceedings of the 17 th International
PATMOS Workshop, Goteborg (September 2007), pp. 181 - 190.
[35] Desoete, B., De Vos, A., Sibinski, M. and Widerski, T. Feynmans reversible logic
gates, implemented in silicon. Proceedings of the 6 th International Conference on
Mixed Design of Integrated Circuits and Systems , Krakow (June 1999), pp. 497 - 502.
[36] Desoete, B. and De Vos, A. A reversible carry-look-ahead adder using control gates.
Integration, the V.L.S.I. Journal 33, 89 - 104 (2002).
[37] Bohr, M., Chau, R., Ghani, T. and Mistry, K. The high-k solution. I.E.E.E. Spectrum
44, 23 - 29 (October 2007).
[38] De Vos, A. and Van Rentergem, Y. Energy dissipation in reversible logic addressed by
a ramp voltage. Proceedings of the 15 th International PATMOS Workshop , Leuven
(September 2005), pp. 207 - 216.
[39] Zeitzoff, P. and Chung, J. A perspective from the 2003 ITRS. I.E.E.E. Circuits &
Systems Magazine 21, 4 - 15 (2005).
[40] Patra, P. and Fussell, D. On efficient adiabatic design of MOS circuits. Proceedings of
the 4 th Workshop on Physics and Computation , Boston (November 1996), pp. 260 269.
[41] Alioto, M. and Palumbo, G. Analysis and comparison on full adder block in submicron
technology. I.E.E.E. Transactions on Very Large Scale Integration Systems 10, 806 823 (2002).
[42] Van Rentergem, Y. and De Vos, A. Reversible full adders applying Fredkin gates.
Proceedings of the 12 th International Conference on Mixed Design of Integrated
Circuits and Systems, Krakow (June 2005), pp. 179 - 184.
[43] Belleville, M. and Faynot, O. Low-power SOI design. Proceedings of the 11 th International PATMOS Workshop , Yverdon (September 2001), pp. 8.1.1 - 8.1.10.
[44] Nagaya, M. Fully-depleted type SOI device enabling an ultra low-power solar radio
wristwatch. O.K.I. Technical Review 70, 48 - 51 (2003).
[45] Yong, A. What is ... a Young tableau? Notices of the A.M.S. 54, 240 - 241 (2007).
[46] Clos, C. A study of non-blocking switching networks. Bell Systems Technical Journal
32, 406 - 424 (1953).
[47] Hwang, F. Control algorithms for rearrangeable Clos networks. I.E.E.E. Transactions
on Communications 31, 952 - 954 (1983).

Reversible Logic

241

[48] Hui, J. Switching and traffic theory for integrated broadband networks . Boston:
Kluwer Academic Publishers (1990), pp. 53 - 138.
[49] Chao, J., Jing, Z. and Liew, S. Matching algorithms for three-stage bufferless Clos
network switches. I.E.E.E. Communications Magazine 41, 46 - 54 (2003).
[50] Jajszczyk, A. Nonblocking, repackable, and rearrangeable Clos networks: fifty years
of the theory evolution. I.E.E.E. Communications Magazine 41, 28 - 33 (2003).
[51] Bhatia, R. Matrix analysis. New York: Springer (1997), pp. 28 - 56.
[52] Borgersen, R. Equivalence of seven major theorems in combinatorics. http://
home.cc.umanitoba.ca/umborger/Presentations/GS-05R-1.pdf (2004).
[53] de Werra, D. Path coloring in bipartite graphs. European Journal of Operational Research 164, 575 - 584 (2005).
[54] Comtet, L. Advanced combinatorics. Dordrecht: Reidel Publishing Company (1974),
pp. 124 - 125.
[55] Even, S., Kohavi, I. and Paz, A. On minimal modulo 2 sums of products for switching
functions. I.E.E.E. Transactions on Electronic Computers 16, 671 - 674 (1967).
[56] Courant, R. Differential and integral calculus . London: Blackie & Son Limited
(1970), p. 361.

In: Mathematics and Mathematical Logic: New Research

ISBN 978-1-60692-862-2
Editors: Peter Milosav and Irene Ercegovaca
2010 Nova Science Publishers, Inc.

Chapter 9

IMAGINARY CUBIC OSCILLATOR

AND ITS SQUARE-WELL APPROXIMATIONS
IN X AND P REPRESENTATION
Miloslav Znojil
Theory Group, Nuclear Physics Institute ASCR,, Czech Republic

ABSTRACT
Schrdinger equation with imaginary PT symmetric potential V (x) = i x3 is studied using the
numerical discretization methods in both the coordinate and momentum representations. In the
former case our results conrm that the model generates an infinite number of bound states with
real energies. In the latter case the differential equation is of the third order and a square-well,
solvable approximation of kinetic energy is recommended and discussed. One finds that in the
strong-coupling limit, the exact PT symmetric solutions converge to their Hermitian predecessors.

PACS 03.65.Ge

ACKNOWLEDGEMENTS
Work supported by the GACR grant Nr. 202/07/1307, by the MSMT \Doppler Institute"
project Nr. LC06002 and by the Institutional Research Plan AV0Z10480505.

e-mail: znojil@ujf.cas.cz

244

Miloslav Znojil

Introduction

An interest in imaginary cubic anharmonic oscillators dates back to their perturbation analysis by Caliceti et al [1]. The simplified homework example with the mere
two-term non-Hermitian Hamiltonian
HBZ = p2 + i x3
has been proposed by D. Bessis and J. Zinn-Justin who had in mind its possible
applicability in the context of statistical physics [2]. The example has been revitalized by C. Bender et al due to its possible methodical relevance in relativistic
quantum field theory [3]. They emphasized that the apparent reality of the spectrum
of energies EBZ is quite puzzling.
The conjecture of a full compatibility of similar Hamiltonians with the postulates
of quantum mechanics [4] opened many new and interesting questions. The current
Hermiticity of Hamiltonians was replaced by a weaker condition of their commutativity with a product PT of the spatial parity P and of the complex conjugation T .
The latter factor is to be understood as a one-dimensional version of the operator of
time reversal.
The discussion between C. Bender and A. Mezincescu [5] pointed out that one
of the key problems of the new studies lies in the ambiguity of the spectrum which
depends quite crucially on our choice of the boundary conditions which can be, in
general, complexified [6]. The fragile character of the reality of the energies has been
confirmed by the WKB and perturbative studies [7, 8] and by the quasi-exact and
exact models [9] where the admissible unavoided level crossings [10] prove sometimes
followed by the spontaneous breakdown of PT symmetry [11].
In such a context we propose here an extremely elementary approach to similar
models replacing interactions which admit just a numerical treatment (typically,
V (x) = ix3 ) by their exactly solvable square-well analogues.

Models in coordinate representation

In a search for analogies between the solvable and unsolvable models in one dimension, all of the possible forms of a confining well are often being approximated by
the ordinary real and symmetric square well
(

(SQW )

(x) =

S 2 , x (, ) (, ),
0, x (, ).

(1)

In this spirit one can also replace the antisymmetric and imaginary homework potential VBZ (x) = i x3 by its elementary square-well analogue

i T , x (, ),

x (, ),
V (ISQW ) (x) = 0,

+i T 2 , x (, ).

Imaginary Cubic Oscillator and Its...

245

Schrodinger equation which appears in such a setting,

"

h
d

+ V (ISQW ) (x) (x) = E(x)

2m dx2

(2)

will be complemented by the standard L2 (lR) boundary conditions

() = 0.

(3)

The well known PT symmetric normalization convention will be employed, with a

free real parameter G in the unbroken PT -symmetry requirements [12]
(0) = 1,

x (0) = i G.

(4)

Putting h
= 2m = 1 and using the ansatz
(

(x) =

cos k x + B sin k x,
x (0, ),
k 2 = E,
(L + i N ) exp( x), x (, ), 2 = i T 2 k 2 ,

(5)

we shall guarantee the full compatibility of such a convention with the symmetry
requirements (4) by the choice of the purely imaginary constant B = i G/k in wave
functions (5).

Matching conditions at x =

Let us split = p + i q in its real and imaginary part with a fixed sign, p, q 0. This
gives p2 + k 2 = q 2 and 2pq = T 2 . These rules are easily re-parameterized in terms of
a single variable ,
p = q cos ,

k = q sin ,

q=

T
,
2 cos

(0, /2).

(6)

The standard matching at the point of discontinuity is immediate,

cos k + B sin k = (L + i N ) exp( ),

sin k + B cos k = (L + i N ) exp( ).

k
After we abbreviate /k = tan , we get an elementary complex condition of
matching of logarithmic derivatives at x = ,
G = i k tan(k + ).

(7)

The real part defines our first unknown parameter, G = G(). Due to our normalization conventions, the imaginary part of the right-hand-side expression must vanish,
Re[tan(k +)] = 0. An elementary re-arrangement of such an equation acquires the
form of an elementary quadratic algebraic equation for X = tan k. Its two explicit
solutions read
pq
p+q
X2 =
X1 =
,
(8)
k
k
3

246

Miloslav Znojil

or, after all the insertions,

"

"

(+)
T sin (+)
,
tan
= tan
2
2 cos (+)
"

(9)

"

T sin ()
()

tan
.
(10)
= tan
2
2 cos ()
In implicit manner these equations specify the two respective infinite series of appropriately bounded real roots = n() (0, /2).

Energies

For (0, /2) the left-hand-side arguments [. . .] in eqs. (9) and (10) run from zero
to infinity. Their tangens functions oscillate infinitely many times from minus infinity
to plus infinity. Within the same interval, the limited variation of the argument
makes both the eligible right-hand side functions monotonic, very smooth and
bounded, tan[( (+) )/2] (1, ) and tan[() /2] (0, 1). A priori this indicates
that our roots k = k(n() ) will all lie within well determined intervals,

kn(+)

1
1
,
n + ,n +
4
2

n = 0, 1, . . . ,

3
m + ,m + 1
m = 0, 1, . . . .
4
After such an approximate localization of the roots, an unexpected additional merit
of our parametrization (6) manifests itself in an unambiguous removal of the tangens
operators from both eqs. (9) and (10). This gives the following two relations,
()
km

kn(+) = n +

1 n(+)

,
2
4

()
km
=m+1

()
m
,
4

n() =

2n()
(0, 1).

After an elementary change of notation with n(+) = 2n and n() = 2n+1 , we may
finally combine the latter two rules in the single secular equation

sin
N
2

2N + 2 N
=

4T

2 cos

N
2

N = 0, 1, . . . ,

(11)

In a graphical interpretation this equation represents an intersection of a tangens-like

curve with the infinite family of parallel lines. This is illustrated in Figure 1. The
equation generates, therefore, an infinite number of the real roots N (0, 1) at all
the non-negative integers N = 0, 1, . . ..

Imaginary Cubic Oscillator and Its...

247

2k
6

0.1

0.2

0.3

0.4

0.5

Figure 1: Graphical solution of eq. (11) (y = N /2, T = 1).

Wave functions in the weak coupling regime

Equation (7) in combination with eqs. (9) and (10) determines the real parameter
G = G() =

k2
qp

(12)

responsible for the behavior of the wave functions near the origin [remember that
B = iG/k in eq. (5)]. For its analysis let us introduce an auxiliary linear function
of and N ,
q

2N + 2 N
R(N , N ) =

4T

N + 1/2 N + 1
,
2T
2T

Our secular eq. (11) can be then read as an algebraic quadratic equation with the
unique positive solution,

cos

N
2

R(N , N ) +

R2 (N , N ) + 1

(13)

This is an amended implicit definition of sequence N . As long as the right hand

side expression is very smooth and never exceeds one, the latter formula re-verifies
that the root N is always real and bounded as required.
In the domain of the large and almost constant R 1 (i.e., for small square-well
heights T or at the higher excitations), our new secular equation (13) gives a better
picture of our bound-state parameters N = 1 N which all lie very close to one.
The estimate
1
1

N = arcsin

+ ...
2
2
2R 48 R3
R+ R +1
5

248

Miloslav Znojil

represents a quickly convergent iterative algorithm for the efficient numerical evaluation of the roots N . One can conclude that in a way compatible with our a
priori expectations, the value of p = pN = Re q/2R lies very close to zero. As a
consequence, the asymptotic decrease of our wave functions remains slow. We have
q = qN = Im k so that, asymptotically, our wave functions very much resemble
free waves exp(ikx). In the light of eq. (12) we have (x) exp(ikx) near the
origin.

Wave functions in the strong coupling regime

For the models with a very small R (i.e., for the low-lying excitations in a deep well
with T 1) we get an alternative estimate

N = arcsin
4

i
1h
1
1

R
1 + R2 1 R R2 + . . . .
2
2
4
4

In the limit R 0 the present spectrum of energies moves towards (and precisely
coincides with) the well known levels of the infinitely deep Hermitian square well
of the same width I = (, ) (cf. eq. (1) with S ). The complex-rotation
transition from the Hermitian well V (SQW ) (x) of eq. (1) (with S 1) to its present
non-Hermitian PT symmetric alternative V (ISQW ) (x) of eq. (2) (with T 1) proves
amazingly smooth.
Wave functions exhibit a similar tendency. In outer region, they are proportional
to exp(px) and decay very quickly since p = O(R1/2 ). Parameter G() becomes
strongly superscript-dependent,
G(+) =

k2
= O(R3/2 ),
q+p

G() = (q + p) = O(R1/2 ).

In the interior domain of x (, ) the wave functions with superscripts (+) and
()
become dominated by their spatially even and odd components cos kx and sin kx,
respectively. The superscript mimics (or at least keeps the trace of) the quantum
number of the slightly broken spatial parity P.
We can summarize that our present PT symmetric model is quite robust. Independently of the coupling T the spectrum is unbounded from above and remains
constrained by inequalities
(N + 1)2
(N + 1/2)2
EN
.
(14)
4
4
The analogy between our exactly solvable square-well model and the standard or
paradigmatic PT symmetric Hamiltonian HBZ appears closer than expected.

Transition to the momentum representation

Let us turn our attention to one-dimensional harmonic oscillator H (HO) = p2 + x2

which is exactly solvable and which appears in virtually any textbook on quantum

Imaginary Cubic Oscillator and Its...

249

mechanics. In PT symmetric quantum mechanics a similar guiding role can be and

has been attributed to the non-Hermitian cubic Hamiltonian H (CO) = p2 + i x3 of
Bender et al [4]. We have seen that a straightforward numerical and semi-classical
analysis of the related Schrodinger equation
H (CO) |n i = En(CO) |n i,

n = 0, 1, . . .

(15)

supports a highly plausible conjecture that the spectrum of energies is real, discrete and bounded below. The conjectured absence of its imaginary components is
indicated by the Hilbert-Schmidt analysis [5] and by the perturbation calculations
in both the weak-coupling regime [13] or in its strong-coupling, purely numerically
generated re-arrangement [12].
Certainly, the problem deserves a change of the traditional perspective. Let us,
therefore, move now to its momentum representation. This would give the momentum as a mere number, p (, ) while the coordinate x becomes represented by
the differential operator x = i p . Equation (15) then acquires a new form containing
the purely real differential Schrodinger operator H (CO) E of third order,
"

d3
+ p2 (p) = E (p) ,
dp3

(p) L2 (IR).

This gives an unusual formulation of our bound-state problem where the quadratic
pdependence of the kinetic term T (p) = p2 does not seem to make the equation
any easier to solve. For this reason we shall drastically simplify the kinetic term and
deduce some consequences.

Piecewise constant approximate kinetic energy

In a way proposed by Pr
ufer [14] many wave functions can be visualized as certain
deformations of solutions which correspond to a locally constant potential, (x)
c1 sin[%(x)] + c2 cos[%(x)]. In the standard quantum mechanics such a trick found
immediate applications in numerical computations [15] while it still admits an easy
interpretation via some traditional Sturm Liouvillean oscillation theorems [16, 17].
Using this idea as a methodical guide let us now replace the kinetic energy operator
T (p) = p2 by the most elementary square well of a finite depth Z > 0,
T (p) =

Z,

p (, 1),
p (1, 1),
p (1, ).

0,
Z,

In the bounded range of energies E < Z this splits our toy model in the two separate
differential equations,
"

d3
8 3 (p) = 0,
3
dp

p (1, 1),

250

Miloslav Znojil
"

d3
+ 8 3 (p) = 0,
dp3

p (, 1)

(1, ).

The two auxiliary parameters = (E) > 0 and = (E) > 0 are defined in
such a way that Z = E + 8 3 > E = 8 3 . They appear in the three independent
(exponential) solutions of our equation. Their general superpositions are complex
but they may be given the real, trigonometric form. Near the origin we have

0 (p) = d e2 p + f e p cos(
p + ),
p (1, 1),

= 3
where the symbols d, f and stand for the three undetermined constant parameters.
In the right and left asymptotic regions we obtain the similar formulae. After we
omit their exponentially growing and normalization-violating unphysical components
we get the one-parametric family
+ (p) = g e2 p ,

p (1, ).

The two-parametric left-barrier counterpart of this formula reads

(p) = c e p cos( p + ),

p (, 1),

3 .

At the right discontinuity p = 1 we have to guarantee the continuity of (p), p (p)

and p2 (p). This is equivalent to the three matching conditions
d e2 + f e cos(
+ ) = g e2 ,
+ )] = 2 g e2 ,
2d e2 f e [cos(
+ ) + 3sin(
42 d e2 22 f e [cos(
+ ) 3 sin(
+ )] = 4 2 g e2 .

(16)

The weighted sum of these equations re-scales and interrelates the unknown coefficients d and g in terms of a new energy parameter t = t(E) = (E)/(E) > 0 or
rather R = R(E) = (1 t + t2 )1/2 > 0,
d e2 D(E) =

G(E)
,
3 R2 (E)

G(E) g e2 .

After we eliminate G from the last two equations (16) which are linear in d and f
we obtain the elementary formula which defines the shift = (E),

3
.
tan(
+ ) =
2/t 1
The trigonometric factors become fixed up to their common sign = 1,

3
t(E) R(E).
cos(
+ ) = [1 t(E)/2] R(E),
sin(
+ ) =
2
The same sign enters our last decoupled definition
f e F (E) =

2[t(E) + 1] G(E)
.
3 R(E)
8

Imaginary Cubic Oscillator and Its...

251

Up to an overall normalization (say, g = 1) in our wave functions matched at p = 1,

all the free parameters become specified as functions of the energy.
At the second, p = 1 discontinuity we have to satisfy three matching conditions
as well. In terms of abbreviations
L1 = d e2 + f e cos(
+
),
2

3 sin(
L2 = 2d e
f e [cos(
+ ) +
+ )],
L3 = 42 d e2 22 f e [cos(
+ ) 3 sin(
+ )]
they read
L1 (, d, f, ) = c e cos(+ ),
L2 (, d, f, ) = c e [cos( + )
3 sin( + ),
L3 (, d, f, ) = 2 2 c e [cos( + ) + 3 sin( + ).
They determine the values of c = c(E) and = (E). Their properly weighted sum
gives

1 t + t2
3 F (E) sin(
+ ) + (2t 1) F (E) cos(
+ ) + 2
D(E) e6 = 0 .
t+1

After its slight re-arrangement we arrive at the amazingly transparent relation

2

(1 4 t + t ) cos(2 3) +

3 (1 t ) sin(2 3) =

1 t + t2 3
e
1+t

It should be read as an implicit definition of physical energies E.

10

y 5

0.01

0.02

0.03

0.04

0.05

-5

-10

Figure 2: Graphical solution of equation (17) at Z = 1/1000.

!2

(17)

252

Miloslav Znojil

The Zdependence of the spectrum

It is quite instructive to search for the physical energies numerically. Starting from
the very-shallow-well extreme in eq. (17) we find the two clearly distinguished energy
roots. The qualitative features of the graph of secular determinant remain unchanged
in a broad interval of the inverse strengths 1/Z. Its shape is sampled in Figure 2 at
Z = 103 . We have tested that even the approximate height 5 of its left plateau
stays virtually unchanged between Z = 105 and Z = 103 . Within the same interval
of the shallowest wells the left zero grows from the value 0.00280 till 0.0241. Beyond
the broad, downwards-oriented peak one finds the second, right zero moving from
the value 0.01047 (found very close to the instantaneous threshold 0.01077) up the
value 0.1056 (not very far from its threshold 0.1077, either) within the same interval
of 1/Z.

0.0004

y
0.0002

0
0.6235

0.624

0.6245

0.625

0.6255

-0.0002

-0.0004

Figure 3: Local maximum giving the new doublet of roots at Z = 5.3005.

A new feature emerges around Z = 101 (with the left zero at 0.0445 and with
the right zero 0.222 still quite close to the threshold 0.232) and Z = 1 (with the left
zero at 0.072 and with the right zero 0.446, not that close already to the threshold
1/2) in the left half of the picture. The plateau develops a local, safely negative
maximum.
In the subsequent domain of Z > 1 we have to switch our attention back to the
right half of our graph. Immediately before the coupling reaches the integer value of
Z = 5, the end of the curve returns to the negative half-plane near the maximal (i.e.,
threshold) energy. This means that there emerges the third energy level there. The
total number of bound states grows to N = 3 (cf. the leftmost items in our Table 1).

10

Imaginary Cubic Oscillator and Its...

253

Table 1.
Number of levels N and its changes with growing Z.
N

3
1

4
1

5
1

6
2

7
1

9
2

10
1

Beyond Z = 5, our attention has to return quickly to the left half of the picture
where the very slow growth of the local maximum creates a new quality at last. The
top of the local bump touches and crosses the horizontal axis at Z 5.3003 and
E 0.6244. At Z = 5.3005 a new doublet of energies is formed in a way illustrated
in Figure 3. The number of levels jumps to N = 5.

1.2

1.3

1.4

1.5

1.6

-1

-2
y
-3

-4

-5

Figure 4: The local maximum not giving the doublet of roots at Z = 35.
A smooth deformation of the graph takes place when the value of Z grows on.
During this evolution we discover that our (originally broad), downward-oriented
peak shrinks quite quickly and moves comparatively slowly to the right. It gets close
to the rightmost and, to its bad luck, slightly more slowly moving zero number five.
The magnified picture of the resulting collision is displayed here in Figure 4. At
Z = 35 it shows that in the threshold region of the energies,
the wavy motion of the threshold end of our graph still did not manage to
reach the zero axis;
the downwards-oriented peak has already left the positive part (and moved to
the negative part) of the curve in question.
As a consequence, the number of levels drops, quite unexpectedly, down to 3 again
(cf. Table 1).

11

254

Miloslav Znojil

10

y 5

0.5

1.5

2.5

-5

-10

Figure 5: Typical Z 1 graph of eq. (17) (Z = 190).

In the vicinity of Z = 40 the new, rightmost energy root emerges at last. Up to
Z = 100 and beyond, the number of levels stays equal to 4. Then it increases to 5,
due to the emergence of the next threshold zero. Only after that, the slowly moving
downward peak reaches the domain of the fourth zero. At almost exactly Z = 190
its left (and temporarily negative) local maximum reaches the zero value again (cf.
Figure 5). At this moment the number of states jumps up by two to seven. The
magnified graphical proof is offered by Figure 5 at Z = 200.

y0.5

2.26

2.28

2.3

2.32

2.34

-0.5

-1

Figure 6: Quasi-degeneracy of the doublet of roots at Z = 200.

The latter Figure illustrates nicely the rapid shrinking of our peak with Z. The
numerical detection of its position becomes more and more difficult. Although this
position plays a crucial role in the practical determination of the number of levels at

12

Imaginary Cubic Oscillator and Its...

255

a given Z, we must be very careful in distinguishing the subgraph of Figure 6 (with

three zeros) from a simple straight line with the single zero.
The pattern is deceitful and the standard software which searches for roots has
to be used with due care. Vice versa, the above analysis enables us to take into
account all the specific features of the Zdependence of the graph in eq. (17). We
get a regular pattern summarized in Table 1 and exhibiting a certain regularity of
the Zdependence of the number of levels N = N (Z).

10

Energies in the square-well approximation

The main consequence of the presence of the above-mentioned narrow peak is an

unusual irregularity observed in the emergence of the new levels in the deeper wells.
We may conclude that this irregularity is not an artifact of the computation method.
The energy formula (17) for our square-well toy model is exact and the seemingly
unpredictable emergence of its roots just reflects the fact that our Schrodinger equation is of the third order. In particular, there exists no symmetry/antisymmetry with
respect to the parity p p etc.

10

y 5

-5

-10

Figure 7: Numerical invisibility of the narrow peak and of the new threshold root at
Z = 1200.
Methodical consequences of our analysis are a bit discouraging. Firstly, the very
symbolic-manipulation derivation of our present formulae proved unexpectedly complicated even in comparison with the multiple standard square wells in textbooks.
Thats why we did not move to any further piece-wise constant approximations
of T (p).
Secondly, even our use of the most elementary solvable example revealed quite
clearly a very real danger of the possible loss of certain levels. For an illustration
let us imagine that our numerical study would have been started in the deep-well
domain, i.e., at the large Z. It is quite easy to generate the graphs of eq. (17) there.

13

256

Miloslav Znojil

In the standard and routine finite-precision computer arithmetics one discovers that
the results are very smooth and look virtually the same, say, in the interval of
Z (1000, 1200).
Let us pick up, for definiteness, the larger sample Z = 1200. We get a picture
(cf. our last Figure 7) which is regular and, deceptively, indicates that N (1200) = 7.
Unfortunately, the correct answer (appearing at the right end of our Table 1) is
N (1200) = 10. Its derivation requires the use of a significantly enhanced precision.
Otherwise, whenever we use just the standard 14 digits and Figure 7, we would have
missed as much as three (i.e., cca 30 % of all) energy levels.
In the light of our preceding considerations, an easy explanation of the latter
numerical paradox lies in the presence of the narrow peak. A priori, it is hardly
predictable of course. It is necessary to spot it by brute force. One finds that at
Z = 1000, this anomalous peak still lives safely below the sixth energy level. The
related number of levels is reliably confirmed as equal to seven, indeed.
In between Z = 1000 and Z = 1200, it is necessary to work in an enhanced
precision arithmetics. One finds that the upper, threshold end of the curve crosses
the horizontal axis only slightly above Z = 1100. Due to the very steep slope of the
curve in this region, this crossing is not visible even at z = 1200 in Figure 7.
One has to trace the narrow peak carefully. It overtakes the sixth energy level
at Z 1190 (and E = 4.217), in an arrangement resembling our Figure 5 above.
Thus, one concludes, finally, that the new, almost degenerate pair of the energy levels
emerges immediately beyond this point.

11

Wave functions and their zeros

A marginal merit of our use of the square-well-shaped T (p) lies in the availability
of the explicit wave functions. For the lack of space we have to omit illustrative
pictures, mentioning just a few of their most characteristic features.
In the first step we notice that in the rightmost interval of p the absence of any
nodal zero in the wave function is in fact very similar to the usual Sturm Liouville
behavior. Less expectedly, at the exact energy value one encounters an infinity of
the nodal zeros in the leftmost subinterval of p (, 1). In this domain we are
fortunate in studying the exactly solvable case. The very presence of this infinite
left set of nodal zeros is extremely sensitive to the numerical level of precision we
use. Indeed, the errors are proportional to the unphysical (unphys.) (p) exp(2 p)
which is growing rapidly at p 1.
After the smallest deviation of the energy E from its absolutely precise boundstate value even the non-numerical and absolutely precise wave functions will be
dominated by the growing asymptotics (unphys.) (p) exp(2 p) near the left infinity. The change of sign of the asymptotics is a reliable source of information about
the fact that the energy crossed it physical value. This observation survives in the
shooting numerical algorithms [18] as well as in the rigorously proved versions of the
method of Hill determinants [19]). In this context our present numerical experiments
could be perceived as opening a number of new questions. Some of them emerge in

14

Imaginary Cubic Oscillator and Its...

257

ufer-type althe purely numerical context of an appropriate generalization of the Pr

gorithms. Especially in the vicinity of the correct physical energies they could lead
to reliable and robust right-to-left shooting numerical recipes.

12

Outlook

In the xrepresentation of our problem our main emphasis has been put on the
exact solvability of its replacement by the purely imaginary square well model. New
light has been thrown on some properties of wave functions. One can expect that the
further detailed study of the PT symmetric square wells will give new answers to the
puzzles concerning the irregular behavior of the nodal zeros in the complex plane as
formulated in ref. [20]. Our present study indicates that some complexified versions
of the Sturm Liouville oscillation theorems should be developed for the study of zeros
of the separate real and imaginary parts of PT symmetric wave functions.
After the standard Fourier-transformation transition to the prepresentation of
our imaginary cubic oscillator the underlying eigenvalue problem can be seen from a
different perspective. Its Hamiltonian is being replaced by a real differential expression. On a suitable Hilbert space this specifies the Hamiltonian operator with the
numerical range (and, hence, spectrum) which is, obviously, real. This complements
the extensive discussion of this topic in [5]. Among several immediate constructive
consequences of the latter observation we underlined the consistency of the approximations imposed directly upon the kinetic term T (p).
As long as the behavior of wave-function asymptotics at large |p| 1 differs in
the left and right infinity, several new qualitative aspects of the problem emerge and
became clarified by our schematic piece-wise constant approximation of T (p). At
small p 0 the emergence and motion of the nodal zeros can be interpreted in a
graphical manner explaining some features of the N (Z) dependence. In particular,
the puzzling loss of its monotonicity seems confirmed by our solvable model.
The use of the momentum representation proved able to throw a new light on
the counterintuitive bound states in PT symmetric quantum mechanics. The emergence/disappearance of our quasi-degenerate doublets should be emphasized as, perhaps, analogous to the unavoided level crossings in harmonic oscillators [10] and/or
to the anomalous doubling of levels in the models of Natanzon type [21]. Similar
irregularities in the spectra could be, perhaps, attributed to a peculiar combination
of the analyticity and non-Hermiticity in PT symmetric systems.
In a brief summary of our numerical observations let us point out the regularity
of the Zdependence of the number N (Z) of the bound states. This indicates that
one should search for an improved application of Sturm-Liouville theory in complex
domain [22]. The possibility of deduction of new oscillation-type theorems exists,
first of all, in the middle interval of p (1, 1) where, for a continuously growing
energy parameter E, a steady right-ward movement of the nodal zeros competes with
the exponential terms which are varying slowly.

15

258

Miloslav Znojil

References
[1] Caliceti E, Graffi S and Maioli M 1980 Commun. Math. Phys. 75 51
[2] D. Bessis 1992 private communication
[3] Bender C M and Milton K A 1997 Phys. Rev. D 55 R3255 and 1998 Phys. Rev.
D 57 3595 and 1999 J. Phys. A: Math. Gen. 32 L87
[4] Bender C M and Boettcher S 1998 Phys. Rev. Lett. 80 5243;
Bender C M, Boettcher S and Meisinger P N 1999 J. Math. Phys. 40 2201;
Znojil M and Tater M 2001 J. Phys. A: Math. Gen. 34 1793
[5] Mezincescu G A 2000 J. Phys. A: Math. Gen. 33 4911;
Bender C M and Wang Q 2001 J. Phys. A: Math. Gen. 34 3325;
Mezincescu G A 2001 J. Phys. A: Math. Gen. 34 3329
[6] Bender C M and Turbiner A 1993 Phys. Lett. A 173 442;
Buslaev V and Grecchi V 1993 J. Phys. A: Math. Gen. 26 5541;
Fernandez F, Guardiola R, Ros J and Znojil M 1999 J. Phys. A: Math. Gen. 32
3105
[7] Alvarez G 1995 J. Phys. A: Math. Gen. 27 4589
[8] Delabaere F and Pham F 1998 Phys. Lett. A 250 25;
Delabaere F and Trinh D T 2000 J. Phys. A: Math. Gen. 33 8771
[9] Bender C M and Boettcher S 1998 J. Phys. A: Math. Gen. 31 L273;
Cannata F, Junker G and Trost J 1998 Phys. Lett. A 246 219;
Bagchi B, Cannata F and Quesne C 2000 Phys. Lett. A 269 79
Znojil M 2000 J. Phys. A: Math. Gen. 33 L61 and 4203 and 4561 and 6825;
Levai G and Znojil M 2000 J. Phys. A: Math. Gen. 33 7165
[10] Znojil M 1999 Phys. Lett. A. 259 220 and 264 108
[11] Khare A and Mandel B P 2000 Phys. Lett. A 272 53;
Bender C M, Berry M, Meisinger P N, Savage V M and Simsek M 2001 J. Phys.
A: Math. Gen. 34 L31
[12] Fernandez F, Guardiola R, Ros J and Znojil M 1998 J. Phys. A: Math. Gen. 31
10105
[13] Bender C M and Weniger E J 2001 J. Math. Phys. 42 2167
[14] Pr
ufer H 1926 Math. Ann. 95 499

16

Imaginary Cubic Oscillator and Its...

259

I, Havlcek M. and Horejs J 1981 Phys. Lett. A 82 64

[15] Ulehla
[16] Fl
ugge S 1971 Practical quantum mechanics I (Berlin: Springer), p. 153
[17] Ince E L 1956 Ordinary differential equations (New York: Dover), p. 223
[18] Killingbeck J P, Gordon N A and Witwit M R M 1995 Phys. Lett. A 206 279;
Znojil M 1997 Phys. Lett. A 230 283
[19] Hautot A 1986 Phys. Rev. D 33 437;
Znojil M 1992 J. Math. Phys. 33 213
[20] Bender C M, Boettcher S and Savage V M 2000 J. Math. Phys. 41 6381
[21] Znojil M, Levai G, Roy P and Roychoudhury R 2001 Phys. Lett. A 290 249
[22] Hille E 1969 Lectures on Ordinary Differential Equations (Reading: AddisonWesley)

17

INDEX
A
academic, 232
accounting, 197
accuracy, 13
achievement, 181, 192, 193, 194, 200
ACM, 198, 199, 200
adaptation, 13, 17
adiabatic, 231, 240
adult, 11, 13, 17
adulthood, 13
adults, 12, 13, 53
Ag, 97, 98, 100, 102, 105
agent, 162
aggregation, 68, 70, 71, 72, 76, 82, 85, 194
agricultural, 161, 162
aid, 11, 18, 52, 169
air, 256
algorithm, 168, 191, 199, 200, 217, 218, 220,
234, 248
alienation, 15
alternative, 6, 36, 69, 74, 75, 78, 79, 82, 83, 183,
248
alternatives, viii, 67, 68, 69, 74, 75, 76, 77, 82, 83
ambiguity, 21, 131, 244
amplitude, 23, 138, 143, 144
AMS, 52, 55, 238
angular momentum, 93, 94, 110
animal cognition, 11
animals, 11, 15, 17, 18, 34
anomalous, 256, 257
anthropic principle, 14
antithesis, 23, 40
appendix, 236
application, 10, 23, 28, 38, 58, 82, 84, 89, 96,
115, 136, 138, 139, 148, 164, 173, 185, 205,
210, 216, 218, 224, 228, 237, 257

archetype, 3, 5, 24, 25, 26, 27, 31, 34, 35, 37, 38,
39, 41, 42, 46, 47, 48, 49, 50, 52
argument, 9, 10, 33, 38, 176, 246
Aristotelian, 53
Aristotle, 3, 4
arithmetic, 11, 12, 17, 20, 26, 97
Asia, 201
aspiration, 192
assumptions, 5, 6, 21, 28, 29, 69, 195
asymptotic, 248, 250
asymptotically, 248
asymptotics, 256, 257
Athens, 1, 4
Atiyah, 49, 52
ATM, 199
atom, 93
atoms, 18, 30, 38, 47, 88, 95, 110, 167
Austria, 233
autonomy, 27
availability, viii, 67, 180, 256

B
Balanced Scorecard, viii, 67, 68, 77, 83, 84
bandwidth, ix, 179, 180, 181, 182, 183, 184, 185,
186, 187, 188, 189, 190, 191, 192, 193, 195,
196, 198, 199
bandwidth allocation, 180, 185, 187, 190, 191,
192, 193, 195, 196, 199
banks, 224
barrier, 250
base pair, 167, 168
behavior, 24, 26, 184, 247, 256, 257
Belgium, 203, 233
bell, 152
benzene, 94, 95
binomial distribution, 97
bipolar, 25
birth, 11, 40

262

Index

blocks, ix, 37, 161, 162, 163, 203, 213

Bohr, 240
Boltzmann constant, 203
bonds, 167
borderline, 26
Borel, 49, 52
Bose, 155, 177
boson, 90, 91
bosons, ix, 87, 89, 90, 100
Boston, 54, 84, 85, 177, 240, 241
bottleneck, 188
bottlenecks, 181
boundary conditions, 244, 245
bounds, 185
brain, 11, 12, 13, 15, 16, 17, 18, 46
Brazil, 67, 83
breakdown, 244
broadband, 179, 185, 200, 241
Broadband, 185
Bromide, 162
browsing, 182
building blocks, 34, 37, 221, 225, 232, 236, 237

C
calculus, 33, 68, 83, 197, 241
capacitance, 229, 231
carrier, 46
category a, 41, 43, 44, 47
cation, 181, 187, 190, 195
cerebral hemisphere, 16
chaos, 27, 34, 40, 41
chemical structures, 167
children, 12, 13, 16, 17, 53
chiral, 97
chirality, 97
classes, ix, 6, 89, 94, 95, 107, 115, 130, 137, 146,
147, 151, 153, 154, 159, 165, 180, 182, 186,
187, 188, 190, 195, 212
classical, vii, ix, 7, 10, 12, 22, 23, 39, 45, 51, 58,
88, 179, 182, 183, 190, 191, 197, 203, 204,
205, 249
classification, viii, 12, 87, 89, 129
closure, 121, 122, 126
codes, 33
coding, 33
cognition, 11, 13
cognitive development, 11, 13
cognitive domains, 12
cognitive science, 11
coil, 217, 218, 219
coke, 61, 64
coke formation, 61, 64
collective unconscious, 3

colors, 97, 137, 141, 142, 144, 145, 176

combinatorics, viii, 87, 89, 97, 178, 234, 241
commodity, 183, 186
communication, ix, 40, 179, 180, 181, 183, 185,
186, 189, 199, 258
communities, 183
community, 2, 181, 182, 197
commutativity, 146
comparison task, 12
compatibility, 244, 245
competence, 11
complementarity, 52
complex numbers, 5, 7, 8, 48, 146, 173
complexity, 5, 170
compliance, 93
components, 181, 183, 193, 248, 249, 250
composites, 42
composition, 10, 31, 41, 42, 43, 45, 46, 47, 50,
64, 130, 134
computation, ix, 2, 180, 185, 203, 204, 238, 255
computing, ix, 178, 203, 204, 207, 220, 238, 239
conception, 4, 10, 15, 40
concordance, 39, 71, 72
confession, 30
configuration, 89, 92, 94, 95
conflict, 237
confusion, 14, 58, 131, 149
Congress, 84, 239
congruence, 148, 150
conjecture, 10, 25, 36, 154, 155, 177, 244, 249
conjugation, 244
connectivity, 181
consciousness, 4, 14, 23, 27, 28, 46, 49, 51, 52
consensus, viii, 67, 68, 69, 71, 72, 73, 74, 77, 78,
79, 83, 84, 85
conservation, 189
constraints, 57, 58, 62, 65, 189, 190, 192, 233
construction, 5, 7, 8, 9, 10, 16, 17, 25, 27, 28, 32,
76, 121, 123
constructivist, 18, 27
consumption, 227
continuity, 13, 250
control, 12, 28, 161, 162, 180, 184, 191, 192,
193, 199, 200, 210, 211, 212, 213, 216, 218,
220, 221, 224, 227, 228, 240
convection, 61
convergence, 17, 199
conversion, 63
convex, 197
conviction, 197
Copenhagen, 162
correlation, 15, 39, 40, 70, 89, 105, 107, 110,
111, 112

Index
correlations, 63, 105, 110
costs, 62, 63, 183, 186, 224
counterbalance, vii, 1
coupling, x, 88, 107, 110, 243, 247, 248, 249, 252
cross-fertilization, 36
culture, 13, 14, 16, 18, 53, 54
customers, 191, 197
Cybernetics, 18, 84, 85
cycles, 73, 74, 96, 135, 140, 142, 143, 144, 233,
234
Czech Republic, 243

D
danger, 10, 255
data communication, ix, 179, 181, 183, 184
database, 182, 188, 190
death, 29, 37
decay, 248
decision making, viii, 58, 59, 67, 68, 69, 74, 84,
85
decisions, 72, 185
decomposition, 64, 140, 142, 143, 144, 217, 218,
219, 220, 222, 234, 236, 237
deduction, 29, 257
definition, 8, 9, 19, 20, 45, 72, 126, 130, 131,
132, 133, 148, 150, 153, 159, 187, 193, 194,
247, 250, 251
deformation, 253
degenerate, 22, 94, 95, 108, 256, 257
degradation, 44
delivery, 184, 185
Department of Energy, 112
derivatives, 245
designers, 185
detection, 254
developed countries, 162
deviation, 256
dichotomy, 14
dielectric constant, 230
differentiation, 34
digit magnitude, 12
digitization, ix, 179
discipline, 25, 51, 185
discontinuity, 245, 250, 251
discordance, 68, 71, 73, 74
discourse, 41, 53, 68
discreteness, 40
discretization, x, 243
distribution, viii, 15, 57, 58, 59, 90, 91, 92, 94,
97, 161, 170, 183, 195, 220
distribution function, viii, 57
division, 130, 150, 151, 233
dominance, 193

263

Doppler, 243
dream, 38
duality, 199
dust, 38
dynamical system, 46, 50
dyscalculia, 12

E
earth, 40
ecological, 161
education, 54
educational psychology, 11
effluent, 62, 64
ego, 40
elasticity, 57
electron, 88, 89, 90, 91, 92, 93, 94
electronic circuits, 227, 230
electronic systems, 89, 93
electrons, ix, 18, 30, 87, 88, 90, 92, 93, 94, 95,
107, 110
email, 182, 184
emotion, 27, 31
emotional, 25, 27, 31
emotions, 27
Empiricism, 9, 49
empowerment, 15
energy, viii, x, 24, 46, 87, 88, 92, 96, 227, 229,
232, 238, 243, 249, 250, 251, 252, 254, 255,
256, 257
energy consumption, 227
enterprise, viii, 16, 67, 68, 77, 83
entropy, 203, 204
environment, 39, 52, 65, 68, 84, 85, 130, 204
epistemological, 39
epistemology, 25
equality, 33, 191
equilibrium, 64, 193
equity, 184
ESR, 96
estimating, 166
estimator, 187
ethane, 61, 63, 64
ethylene, 61
Euclidean space, 35
Europeans, 152
evolution, 4, 12, 16, 17, 18, 31, 37, 40, 241, 253
evolutionary process, 17
expansions, 222

F
factorial, 208
failure, 2, 38, 195

264

Index

fairness, ix, 179, 180, 181, 186, 187, 190, 191,

192, 193, 194, 197, 199
family, 7, 130, 246, 250
fear, 27
feedback, 25
Fermat, 16
fermions, ix, 87, 89, 90, 93
fertilizer, 161
fertilizers, 161
Feynman, 238, 240
fiber, 184
field theory, vii
filters, 224
Finland, 53, 113
flexibility, 57, 68, 69
flow, 9, 64, 181, 182, 183, 186, 189, 193, 199,
233, 234
flue gas, 61
fluid, 17
Fourier, 239, 257
free will, 51
frequency, 98, 100, 102
Freud, 34
fullerene, 96
function values, 58
functional magnetic resonance imaging (fMRI),
11, 13
future, 112
fuzzy logic, 57
Fuzzy set theory, 68
fuzzy sets, viii, 22, 57, 58, 65, 75, 77

G
garbage, ix, 203, 223, 224, 239
gas, 64
Gaussian, 64, 146
generalization, 8, 10, 34, 97, 208, 257
generation, vii, x, 89, 91, 94, 197, 203, 204, 229,
231, 232, 238
generators, 91, 94, 138, 139, 182
genomics, 166
geography, 183
Germany, 53
goals, 57, 65, 68, 180, 181, 186
God, 6, 40
gold, 88
Goodman, 54
government, iv
GPRS, 182
graph, 158, 167, 169, 170, 177, 178, 186, 252,
253, 254, 255
Greece, 1
grouping, 161

groups, 15, 43, 89, 94, 95, 96, 107, 110, 112, 131,
133, 134, 150, 162, 177, 178, 207, 238, 239
growth, 77, 179, 180, 253
GSM, 182
Guaranteed Service, 185
guidance, 73
gyrus, 11, 12
Gyrus, 54

H
H2, 64
hafnium, 230
Hamiltonian, 88, 107, 244, 248, 249, 257
happiness, 16
harm, 3, 4, 178
harmony, 3, 4
Harvard, 54, 84
Hawaii, 201
heat, vii, x, 64, 203, 204, 229, 231, 232, 238
height, 23, 252
Heisenberg, 29
hemisphere, 11
hepatitis, 170
hepatitis C, 170
herbs, 162
heterogeneous, 180
Hilbert, 2, 10, 28, 29, 54, 249, 257
Hilbert space, 257
holistic, vii, 1, 52
homework, 244
homology, 11
homomorphism, 133, 136, 148, 149
host, 12, 185
human, viii, 4, 11, 12, 13, 14, 15, 16, 17, 18, 23,
24, 25, 26, 27, 28, 29, 30, 31, 32, 34, 35, 36,
37, 39, 40, 46, 52, 65, 67, 68, 71, 72, 73
human animal, 13
human brain, 12, 13, 15, 17, 18
human experience, 23
human nature, 52, 68
humanism, 14
humanity, 15, 34, 35
humans, 5, 11, 13, 18, 39, 46
hybrid, 200
hydro, 62
hydrocarbon, 63
hydrocarbons, 62
hydrogen, 62
hypothesis, 11, 26, 27, 122

I
icosahedral, 96, 107

Index
identity, 9, 19, 26, 27, 41, 42, 44, 45, 46, 47, 88,
116, 130, 131, 135, 136, 146, 153, 169, 171,
172, 206, 207, 209, 213, 216, 220, 221, 225,
235, 236
illusion, 16
images, 16, 24, 26, 27, 38
imaging, 11, 53
implementation, 12, 192, 222, 224, 226, 227,
228, 232, 238
inattention, 12
incidence, 162
inclusion, 9
income, 190
independence, 32, 51
Indian, 17
indices, 59, 97
individual differences, 50
individuality, 68
induction, 20
industrial, 161
industry, ix, 129, 197, 229
ineffectiveness, 191
inequality, 58, 193
infants, 11, 13, 15, 35
inferior parietal region, 12, 16
infinite, ix, 8, 13, 21, 28, 31, 32, 35, 36, 38, 50,
88, 115, 116, 119, 121, 122, 123, 124, 125,
132, 177, 178, 243, 246, 256
Information System, 203
infrastructure, 180
injections, 122
inorganic, 25
insects, 34, 35
insight, 4, 10
inspiration, 10
instability, 10
instruction, 16
integration, 2
intellect, 25, 27, 32
intelligence, 22
interaction, 58, 96
interactions, 175, 244
interdisciplinary, vii, 1
Internet, ix, 179, 180, 181, 182, 183, 184, 185,
191, 197, 199, 201, 234
Internet Protocol, 181
interval, 23, 71, 246, 252, 256, 257
intervention, 12
intrinsic, viii, 67, 68, 71
intron, 170
intuition, 15, 35
invariants, 142
inventiveness, 15

265

inventories, 142
inversion, 97
Investigations, 56
IP, 182, 183, 184, 186, 198, 199, 200, 201
IPv6, 182
irrationality, 29
isomorphism, 7, 19, 133, 148
Israel, 177

J
Jaffe, 49, 52, 55
Japan, 201
Japanese, 154
joining, 143, 144
judgment, 31
Jung, vii, viii, 1, 2, 23, 24, 25, 27, 31, 32, 33, 34,
35, 37, 38, 39, 40, 41, 49, 50, 52, 53, 55
Jungian, 3, 28, 55
justice, 3
justification, 10

K
Kant, 30, 54
kernel, 133, 148
kinetic energy, x, 249
King, 87
Kobe, 201

L
labeling, 6
labor, 58
labor force, 58
land, 162
language, 11, 12, 16, 17, 24, 26, 27, 33, 35, 115,
116, 133, 174
latency, 184
Latin squares, ix, 152, 153, 154, 155, 156, 178
law, 10, 12, 14, 41, 42, 149, 204, 229, 231, 232
laws, 23
leakage, 227
learners, 15
learning, 11, 17, 40, 77
Leibniz, 34
lesions, 12
life forms, 18
light, 88
likelihood, 32
limitation, 184
limitations, 26, 35, 39, 51
linear, 2, 8, 12, 24, 58, 59, 62, 173, 207, 208,
209, 238, 247, 250
linear function, 59, 207, 247

266

Index

linear programming, 58
linear systems, 25
linguistic, vii, viii, 24, 57, 67, 68, 69, 70, 71, 74,
77, 78, 83, 84, 85
linguistic information, viii, 67, 68, 70, 84
links, 180, 181, 186, 187, 188, 193
loading, 183
localization, 246
location, 190
locus, 13
logical deduction, 29
London, 53, 54, 55, 56, 84, 241
Los Angeles, 53
losses, viii, 67, 184
low power, 240
lying, 248

M
magma, 130, 131
magnetic, 11
magnetic resonance imaging, 11, 13
Maine, 55
management, 69, 84, 180, 182, 185, 186, 199
manipulation, 16, 255
manufacturing, 58
mapping, 33, 233
market, 229
Markov, 238
Markovian, 198
marriage, 234
mathematical disabilities, 11
mathematical knowledge, 15
mathematical programming, 57, 190
mathematicians, 2, 5, 7, 10, 14, 15, 16, 17, 18,
28, 29, 30, 35, 36, 52, 207
mathematics, ix, 12, 16, 18, 33, 47, 129, 174,
177, 178, 208
matrices, 113
matrix, 30, 74, 147, 162, 163, 233, 234, 235
Max-Min Fairness, ix
meanings, 58
measurement, 2, 161, 195
measures, 59, 60, 61, 62, 64, 71, 73
media, 181, 197
medicine, ix, 129
membership, viii, 6, 22, 23, 57, 58, 59, 70, 71,
74, 75, 76, 77
memory, vii, 1
mental image, 16
mental power, 4, 5
mental representation, 2, 12, 16, 17
messages, 23, 26, 234
metaphor, 13

metaphors, 13
methane, 62
metric spaces, vii
Middle Ages, 40
Mind-Body, 56
MIP, 197
MIT, 53, 84
modeling, viii, 67, 84
models, v, vii, 3, 6, 8, 10, 18, 20, 25, 65, 74, 77,
82, 83, 126, 179, 183, 186, 201, 244, 248, 257
modules, 197
molecules, viii, 87, 88, 89, 95, 96, 110
momentum, vii, x, 93, 94, 110, 243, 248, 249,
257
MOS, ix, 203, 227, 228, 231, 232, 240
Moscow, 56
motion, 23, 253, 257
motivation, 238
movement, 2, 16, 29, 34, 140, 257
multicultural, 15
multimedia, 182
multiplication, 43, 44, 45, 47, 50, 89, 125, 130,
146, 147, 149, 170
multiplicity, 6, 99
multiplier, 224
music, 29
mutations, 140

N
National Science Foundation, 112
natural, vii, viii, 1, 4, 5, 6, 7, 8, 9, 18, 19, 20, 21,
23, 27, 28, 29, 30, 31, 32, 33, 34, 36, 37, 38,
39, 41, 42, 45, 46, 47, 48, 49, 51, 52, 57, 65,
69, 74, 88, 125, 129, 130, 137, 149, 153, 160,
191, 221
network, ix, 179, 180, 181, 182, 183, 184, 185,
186, 188, 189, 190, 191, 192, 193, 195, 197,
199, 241
neurobiological, 16
neuroimaging, 12, 13
neuronal circuits, 16
neurons, 11, 12
neurophysiology, 12, 13
neuropsychology, 11, 12
neuroscience, 11
New Jersey, 198
New Science, 55
New York, 55, 65, 113, 114, 166, 174, 177, 178,
199, 200, 238, 241, 259
Newton, 34
next generation, 197
Ni, 201
NMR, 96, 102

Index
nodes, 180, 181, 185, 186, 188
non-human, 13
non-uniform, 8
nonverbal, 17
normal, 88, 89, 107, 118, 133, 136, 138, 157
normalization, 77, 245, 250, 251
norms, 26
NSC, 198
nuclear, viii, 87, 89, 90, 91, 93, 96, 97, 99, 100,
102, 105, 106
nuclei, 88, 96, 100, 102
nucleus, 90, 97
number-words, 9
numerical computations, 249

O
objective reality, 51
observations, 21, 257
obsolete, 183
operator, ix, 14, 88, 179, 190, 191, 244, 249, 257
opposition, 26, 34, 40, 152
optical, ix, 179, 183
optical transmission, 183
optimism, 63, 64
optimization, vii, ix, 57, 58, 59, 84, 154, 179,
181, 190, 193, 194, 197, 239
orbit, 88, 96, 97, 107, 110, 137, 139, 145, 165,
177
organic, 25
orientation, 90
orthogonal Latin squares, 155, 156
orthogonality, 155
oscillation, 249, 257
oscillator, 243, 248, 257
oxide, 227, 229, 230
oxide thickness, 229
ozone, 162

P
Pacific, 201
packets, 180
pairing, ix, 115, 122, 123, 124, 125, 126
paradox, 10, 46, 256
parameter, 15, 64, 68, 71, 72, 73, 77, 78, 83, 192,
193, 245, 247, 250, 257
Pareto, 76, 188, 190, 199
Pareto optimal, 188, 190
Parkinson, 65
particles, ix, 30, 38, 87, 89, 90, 91, 93, 102
partition, 90, 91, 92, 97, 119, 120, 130, 137, 139,
159, 198, 214, 217
pathogens, 162

267

patients, 34
pattern recognition, 35
pedagogical, 4
perception, 17, 21, 40, 68, 69, 72
periodic, 89, 93, 94, 96, 97, 102, 105, 108, 110
Periodic Table, viii, 87, 89
periodicity, viii, 87, 88, 89, 91, 94, 95, 96, 97,
102, 105, 106, 107, 108, 110, 112
permit, viii, 67
permittivity, 229
personality, 40
perturbation, 249
Philadelphia, 65
philosophers, 3, 7, 15
philosophical, vii, 1, 7, 10, 14, 33, 48, 49
philosophy, 3, 4, 14, 29, 227, 228
physical environment, 15
physical world, 13, 17, 38, 226
physicists, 16, 18, 38
physics, 30, 244
physiological, vii, 1
planar, 167, 168
planets, 3, 14
planning, viii, 67, 68, 83, 161, 187, 189
Plato, 4, 25, 27, 34, 39, 50
play, 13, 47, 65, 77, 89, 130, 153, 213
pleasure, 28
plurality, 2
Poisson, 183, 195, 196
polarity, 14, 34
polynomial, ix, 89, 95, 96, 129, 140, 145, 149,
150, 151, 169, 176
polynomials, 91, 97, 149, 151
poor, 180, 192
population, 12, 15, 161
power, vii, x, 4, 15, 27, 31, 32, 36, 69, 84, 93, 97,
116, 132, 149, 155, 160, 161, 163, 177, 203,
231, 232, 240
powers, 32, 89, 97, 132, 160
predators, 17
predicate, 20, 21, 33, 117, 125, 126
pre-existing, 50
preference, viii, 67, 74, 75, 76, 77, 79, 82, 83, 84,
85, 186
prefrontal cortex, 12
press, 113
pressure, 63
prices, 199
primary products, 62
primate, 12, 17
private, 184, 258
probability, 29, 65, 195, 196, 197
probe, 96

268

Index

production, 26, 61
productivity, 162
profit, 190
prognosis, 11
program, 4, 5, 197
programming, 57, 58, 65, 174, 200
proliferation, 224
propagation, 187
properties, 88
property, 15, 20, 27, 31, 34, 130, 146, 181, 212,
213, 227
Proportional Fairness, ix, 179
proposition, 14, 34, 36, 37, 38, 150
protocols, 184, 186, 198
protons, 88, 102
prototype, 3, 227, 229
prototyping, 229
pseudo, 170
psyche, 2, 24, 31, 34, 37, 38, 50
psychic process, 24
psychological functions, 2, 39
Psychological Perspective, 53
psychologist, 28
psychology, 3, 37
psychophysics, 12
PT, x, 243, 244, 245, 248, 249, 257
public, 191
Public Switched Telephony Network PSTN), ix,
179
pyrolysis, 62

Q
QoS, v, 179, 180, 181, 182, 183, 184, 185, 186,
187, 188, 189, 190, 191, 192, 193, 195, 197,
198, 199, 200, 201
quality of service, 185, 198
quantum, vii, ix, 2, 88, 89, 92, 96, 99, 102, 203,
204, 226, 230, 231, 239, 240, 244, 248, 249,
257, 259
quantum computing, 203, 204
quantum field theory, 244
quantum mechanics, vii, 88, 244, 249, 257, 259
quarks, 18, 38
qubits, 226
Quinn, 49, 52, 55

R
radiation, 61
radio, 240
rail, 226, 240
random, 23, 27, 161, 174, 195
range, 8, 180, 249, 257

reading, 207
real numbers, 5, 13, 22, 23, 32, 46, 48, 59
real time, 180
realism, 10, 33
reality, 3, 11, 13, 14, 16, 18, 24, 26, 30, 34, 36,
37, 38, 39, 49, 51, 52, 244
reasoning, 8, 13, 22, 118, 119, 120, 121, 122,
123, 124, 125, 236
recalling, 118
recognition, 18, 35
reconciliation, 26
recovery, 62
recursion, 20
reflection, 145
regional, 182
regular, 96, 142, 143, 144, 147, 255, 256
rehabilitation, 11
relationship, viii, 4, 12, 15, 38, 40, 87, 206, 210
relationships, viii, 3, 23, 35, 36, 87, 226
relativity, 52, 88
relevance, 244
reliability, 184, 191, 197
religion, 14, 16, 17, 25
Renaissance, 34, 40
repetitions, 133
research, 112
Research and Development, 238
resolution, 34, 165
resource allocation, 180, 181, 194, 195, 200
resources, ix, 58, 179, 180, 181, 183, 184, 199
retail, 47
returns, 252
revenue, 181, 191
rhythm, 3
right hemisphere, 12
rings, ix, 43, 129, 146, 148, 150
risk, 183
RNA, ix, 129, 166, 167, 168, 169, 170
robustness, 197
rotations, 89, 137, 138, 140, 142, 143, 144, 165
routing, 180, 181, 182, 185, 186, 188, 190, 198,
199, 200, 201, 234

S
sacred, 23, 27, 31
sample, 161, 256
satisfaction, 58, 63, 77, 181
scalar, 11, 70, 71
scheduling, 185, 198
schema, 8, 24
school, 13, 14, 15, 28
Schrdinger equation, x, 243
scientific community, 2

Index
search, 11, 12, 24, 49, 154, 177, 186, 241, 244,
252, 257
searches, 255
searching, 35
secular, 246, 247, 252
security, 5
Self, 25, 40, 41, 56
semantic, 68, 77
semantics, 3, 58, 59, 115
semiconductor, 227
semigroup, 131, 146
sentences, 32, 118, 126
series, ix, 20, 30, 31, 35, 179, 227, 246
service quality, 180
services, ix, 179, 180, 181, 182, 183, 184, 185,
191, 193
set theory, 58, 68
shape, 13, 43, 64, 90, 91, 92, 133, 220, 252
shaping, 10
Shapiro, 56
shares, 68, 193
sharing, ix, 179, 184, 185, 197, 199, 200
Shell, 93
sign, 2, 6, 12, 88, 107, 116, 203, 245, 250, 256
signals, 231
signs, 6, 18, 146
silicon, 230, 232, 240
similarity, 84
Singapore, 53
singular, 147
skeleton, 50
skin, 162
SMS, 182
social construct, 15, 17, 18
social context, 15
social responsibility, 15
Socrates, 25
software, 255
SOI, 232, 240
soil, 161
solar, 240
sounds, 14
South America, 35
space-time, 26, 39
Spain, 129, 164
spare capacity, 185
spatial, 12, 244, 248
special theory of relativity, 88
species, 17, 89, 96, 97, 101, 102, 154
spectroscopy, viii, 87, 88, 89, 93, 96
spectrum, 22, 36, 98, 244, 248, 249, 252, 257
speculation, 35
speech, 180, 224

269

speed, 88, 187, 198

speed of light, 88
spin, viii, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96,
97, 98, 99, 100, 101, 102, 105, 106, 107, 110,
111
spin-1, 90, 91
spiritual, 3, 4, 25, 27, 52
stability, 34, 184, 199
stabilizers, 138
stages, 25, 34
standard model, 8, 20, 125
standards, 21
starvation, 186
statistics, 161
stochastic, 183
Stochastic, 200
stochastic processes, 183
strain, 35
strategies, 9
streams, 181, 185
stress, 27, 162, 220, 221
structuralism, 10, 20
structuring, 50
students, vii, 1, 15, 233
subgroups, ix, 131, 133, 138, 161, 203, 207, 208,
209, 213, 214, 215, 217, 221, 222, 225, 232,
235, 239
subjective, 15, 72
subjectivity, 68
subrings, 147, 148
subscribers, 182, 191
substitution, 97
subtraction, 16, 130
superstitious, 27
supervision, 68
suppliers, 58
surprise, 22
survival, 17
switching, 95, 182, 234, 240, 241
symbolic, 3, 13, 17, 24, 26, 35, 70, 255
symbols, 31, 33, 34, 35, 38, 89, 93, 94, 97, 116,
131, 151, 152, 153, 250
symmetry, viii, 24, 45, 87, 88, 94, 95, 96, 105,
107, 112, 145, 244, 245, 255
syntactic, 49, 68
synthesis, ix, 26, 40, 50, 179, 206, 212, 213, 215,
216, 217, 218, 220, 221, 222, 225, 232, 236,
237, 238, 239
systems, 5, 6, 7, 8, 27, 35, 46, 58, 65, 89, 93, 96,
107, 110, 112, 195, 198, 234, 257

T
Taiwan, 179, 198

270
tangible, 12
TCP, 191, 199
teaching, 3, 26
technological developments, ix
technology, ix, 203, 204, 227, 228, 230, 231, 232,
240
telecommunication, ix, 179, 180, 183, 198, 199,
203
telecommunication networks, 183, 198, 199
telephone, 234
telephony, 182, 183, 184
temperature, 63, 203
Tesla, 13
tetrad, 40
textbooks, 207, 255
Theory of Everything, 39
thinking, 2, 18, 27, 36
third order, x, 243, 249, 255
Thomson, 53
threat, 9
threshold, 78, 229, 231, 232, 252, 253, 254, 255,
256
tics, 199
time, ix, 2, 4, 5, 11, 13, 16, 17, 22, 23, 25, 30, 32,
35, 36, 40, 46, 50, 57, 154, 174, 179, 180, 181,
182, 183, 184, 190, 195, 196, 197, 198, 224,
244
tolerance, 58
topological, vii, 43, 170
topology, 170, 180, 183, 186, 188, 231
Topos, 19, 55
total utility, 190
traction, 130
traffic, 14, 180, 181, 182, 183, 184, 185, 186,
196, 197, 198, 199, 200, 241
traffic flow, 182, 183, 184
training, 16, 18
traits, 34
trans, 17, 31, 53, 54, 195
transfer, 62, 180, 181, 184, 186, 187
transformation, 70, 71, 257
transformations, 24, 41, 42
transistor, 227, 229, 230, 231
transistors, 227, 229
transition, 57, 58, 70, 248, 257
transitions, 96
translation, 70
transmission, 15, 164, 165, 181, 183, 191, 195,
197, 227
transparent, 30, 251
transport, 183
transportation, 47
Transylvania, 170

Index
trees, 31, 166, 167, 168, 169, 170
Trinidad and Tobago, 57
two-dimensional, 41, 43, 108
two-dimensional space, 43

U
UMTS, 182, 200
uncertainty, viii, 2, 23, 28, 34, 58, 67, 68, 69, 71,
181, 198
uniform, 8
universal grammar, 26
universality, 16, 238
universe, 31, 68
unpredictability, 184, 185
utilitarianism, 51

V
vacuum, 229
validity, 5, 38, 68, 186
values, 15, 57, 58, 63, 72, 73, 77, 93, 97, 99, 108,
111, 112, 187, 204, 210, 251
variability, 161
variables, 33, 58, 63, 77, 115, 116, 118, 119, 120,
121, 122, 123, 126, 140, 161, 188, 189, 194,
210, 216, 224, 227, 231
variation, 180, 190, 246
vector, 43, 73, 74, 173, 187, 193, 194
vehicles, 154
vertebrates, 18
virus, 170
visible, 26, 52, 256
vision, 41
visuospatial, 12
vocabulary, 115, 116
voice, 5, 182, 184
voids, 10
VoIP, 182

W
wealth, 15, 207
web, 182, 184
wells, 252, 255, 257
West Indies, 57
Western societies, 17
winning, 174
wireless, ix, 179, 184, 198, 201
wireless networks, ix, 184, 201
wires, 15, 221, 234
wisdom, vii, 1, 3
withdrawal, 162
women, 15
wood, 177

Index
World Wide Web, 175

271

Z
Y

yield, 88, 105, 112, 150, 208, 220, 227, 236

zeitgeist, 51
Zen, 50